28 void zip_1x8_aligned(const std::uint8_t* source, std::int32_t count,  in zip_1x8_aligned()
29                      std::int32_t stride, std::uint8_t* destination,  in zip_1x8_aligned()
30                      std::int32_t multiplicative_offset,  in zip_1x8_aligned()
31                      std::int32_t additive_offset) {  in zip_1x8_aligned()
62 void zip_1x8_1_aligned(const std::uint8_t* source, std::int32_t count,  in zip_1x8_1_aligned()
63                        std::int32_t stride, std::uint8_t* destination,  in zip_1x8_1_aligned()
64                        std::int32_t multiplicative_offset,  in zip_1x8_1_aligned()
65                        std::int32_t additive_offset) {  in zip_1x8_1_aligned()
103 void zip_1x8_2_aligned(const std::uint8_t* source, std::int32_t count,  in zip_1x8_2_aligned()
104                        std::int32_t stride, std::uint8_t* destination,  in zip_1x8_2_aligned()
105                        std::int32_t multiplicative_offset,  in zip_1x8_2_aligned()
106                        std::int32_t additive_offset) {  in zip_1x8_2_aligned()
144 void zip_1x8_3_aligned(const std::uint8_t* source, std::int32_t count,  in zip_1x8_3_aligned()
145                        std::int32_t stride, std::uint8_t* destination,  in zip_1x8_3_aligned()
146                        std::int32_t multiplicative_offset,  in zip_1x8_3_aligned()
147                        std::int32_t additive_offset) {  in zip_1x8_3_aligned()
186 void zip_1x8_4_aligned(const std::uint8_t* source, std::int32_t count,  in zip_1x8_4_aligned()
187                        std::int32_t stride, std::uint8_t* destination,  in zip_1x8_4_aligned()
188                        std::int32_t multiplicative_offset,  in zip_1x8_4_aligned()
189                        std::int32_t additive_offset) {  in zip_1x8_4_aligned()
227 void zip_1x8_5_aligned(const std::uint8_t* source, std::int32_t count,  in zip_1x8_5_aligned()
228                        std::int32_t stride, std::uint8_t* destination,  in zip_1x8_5_aligned()
229                        std::int32_t multiplicative_offset,  in zip_1x8_5_aligned()
230                        std::int32_t additive_offset) {  in zip_1x8_5_aligned()
269 void zip_1x8_6_aligned(const std::uint8_t* source, std::int32_t count,  in zip_1x8_6_aligned()
270                        std::int32_t stride, std::uint8_t* destination,  in zip_1x8_6_aligned()
271                        std::int32_t multiplicative_offset,  in zip_1x8_6_aligned()
272                        std::int32_t additive_offset) {  in zip_1x8_6_aligned()
311 void zip_1x8_7_aligned(const std::uint8_t* source, std::int32_t count,  in zip_1x8_7_aligned()
312                        std::int32_t stride, std::uint8_t* destination,  in zip_1x8_7_aligned()
313                        std::int32_t multiplicative_offset,  in zip_1x8_7_aligned()
314                        std::int32_t additive_offset) {  in zip_1x8_7_aligned()
354 void zip_2x8_aligned(const std::uint8_t* source, std::int32_t count,  in zip_2x8_aligned()
355                      std::int32_t stride, std::uint8_t* destination,  in zip_2x8_aligned()
356                      std::int32_t multiplicative_offset,  in zip_2x8_aligned()
357                      std::int32_t additive_offset) {  in zip_2x8_aligned()
395 void zip_2x8_1_aligned(const std::uint8_t* source, std::int32_t count,  in zip_2x8_1_aligned()
396                        std::int32_t stride, std::uint8_t* destination,  in zip_2x8_1_aligned()
397                        std::int32_t multiplicative_offset,  in zip_2x8_1_aligned()
398                        std::int32_t additive_offset) {  in zip_2x8_1_aligned()
446 void zip_2x8_2_aligned(const std::uint8_t* source, std::int32_t count,  in zip_2x8_2_aligned()
447                        std::int32_t stride, std::uint8_t* destination,  in zip_2x8_2_aligned()
448                        std::int32_t multiplicative_offset,  in zip_2x8_2_aligned()
449                        std::int32_t additive_offset) {  in zip_2x8_2_aligned()
497 void zip_2x8_3_aligned(const std::uint8_t* source, std::int32_t count,  in zip_2x8_3_aligned()
498                        std::int32_t stride, std::uint8_t* destination,  in zip_2x8_3_aligned()
499                        std::int32_t multiplicative_offset,  in zip_2x8_3_aligned()
500                        std::int32_t additive_offset) {  in zip_2x8_3_aligned()
550 void zip_2x8_4_aligned(const std::uint8_t* source, std::int32_t count,  in zip_2x8_4_aligned()
551                        std::int32_t stride, std::uint8_t* destination,  in zip_2x8_4_aligned()
552                        std::int32_t multiplicative_offset,  in zip_2x8_4_aligned()
553                        std::int32_t additive_offset) {  in zip_2x8_4_aligned()
601 void zip_2x8_5_aligned(const std::uint8_t* source, std::int32_t count,  in zip_2x8_5_aligned()
602                        std::int32_t stride, std::uint8_t* destination,  in zip_2x8_5_aligned()
603                        std::int32_t multiplicative_offset,  in zip_2x8_5_aligned()
604                        std::int32_t additive_offset) {  in zip_2x8_5_aligned()
654 void zip_2x8_6_aligned(const std::uint8_t* source, std::int32_t count,  in zip_2x8_6_aligned()
655                        std::int32_t stride, std::uint8_t* destination,  in zip_2x8_6_aligned()
656                        std::int32_t multiplicative_offset,  in zip_2x8_6_aligned()
657                        std::int32_t additive_offset) {  in zip_2x8_6_aligned()
707 void zip_2x8_7_aligned(const std::uint8_t* source, std::int32_t count,  in zip_2x8_7_aligned()
708                        std::int32_t stride, std::uint8_t* destination,  in zip_2x8_7_aligned()
709                        std::int32_t multiplicative_offset,  in zip_2x8_7_aligned()
710                        std::int32_t additive_offset) {  in zip_2x8_7_aligned()
762 void zip_3x8_aligned(const std::uint8_t* source, std::int32_t count,  in zip_3x8_aligned()
763                      std::int32_t stride, std::uint8_t* destination,  in zip_3x8_aligned()
764                      std::int32_t multiplicative_offset,  in zip_3x8_aligned()
765                      std::int32_t additive_offset) {  in zip_3x8_aligned()
811 void zip_3x8_1_aligned(const std::uint8_t* source, std::int32_t count,  in zip_3x8_1_aligned()
812                        std::int32_t stride, std::uint8_t* destination,  in zip_3x8_1_aligned()
813                        std::int32_t multiplicative_offset,  in zip_3x8_1_aligned()
814                        std::int32_t additive_offset) {  in zip_3x8_1_aligned()
873 void zip_3x8_2_aligned(const std::uint8_t* source, std::int32_t count,  in zip_3x8_2_aligned()
874                        std::int32_t stride, std::uint8_t* destination,  in zip_3x8_2_aligned()
875                        std::int32_t multiplicative_offset,  in zip_3x8_2_aligned()
876                        std::int32_t additive_offset) {  in zip_3x8_2_aligned()
935 void zip_3x8_3_aligned(const std::uint8_t* source, std::int32_t count,  in zip_3x8_3_aligned()
936                        std::int32_t stride, std::uint8_t* destination,  in zip_3x8_3_aligned()
937                        std::int32_t multiplicative_offset,  in zip_3x8_3_aligned()
938                        std::int32_t additive_offset) {  in zip_3x8_3_aligned()
1000 void zip_3x8_4_aligned(const std::uint8_t* source, std::int32_t count,  in zip_3x8_4_aligned()
1001                        std::int32_t stride, std::uint8_t* destination,  in zip_3x8_4_aligned()
1002                        std::int32_t multiplicative_offset,  in zip_3x8_4_aligned()
1003                        std::int32_t additive_offset) {  in zip_3x8_4_aligned()
1062 void zip_3x8_5_aligned(const std::uint8_t* source, std::int32_t count,  in zip_3x8_5_aligned()
1063                        std::int32_t stride, std::uint8_t* destination,  in zip_3x8_5_aligned()
1064                        std::int32_t multiplicative_offset,  in zip_3x8_5_aligned()
1065                        std::int32_t additive_offset) {  in zip_3x8_5_aligned()
1127 void zip_3x8_6_aligned(const std::uint8_t* source, std::int32_t count,  in zip_3x8_6_aligned()
1128                        std::int32_t stride, std::uint8_t* destination,  in zip_3x8_6_aligned()
1129                        std::int32_t multiplicative_offset,  in zip_3x8_6_aligned()
1130                        std::int32_t additive_offset) {  in zip_3x8_6_aligned()
1192 void zip_3x8_7_aligned(const std::uint8_t* source, std::int32_t count,  in zip_3x8_7_aligned()
1193                        std::int32_t stride, std::uint8_t* destination,  in zip_3x8_7_aligned()
1194                        std::int32_t multiplicative_offset,  in zip_3x8_7_aligned()
1195                        std::int32_t additive_offset) {  in zip_3x8_7_aligned()
1260 void zip_1x8(const std::uint8_t* source, std::int32_t count,  in zip_1x8()
1261              std::int32_t stride, std::uint8_t* destination,  in zip_1x8()
1262              std::int32_t multiplicative_offset, std::int32_t additive_offset) {  in zip_1x8()
1293 void zip_1x8_1(const std::uint8_t* source, std::int32_t count,  in zip_1x8_1()
1294                std::int32_t stride, std::uint8_t* destination,  in zip_1x8_1()
1295                std::int32_t multiplicative_offset,  in zip_1x8_1()
1296                std::int32_t additive_offset) {  in zip_1x8_1()
1334 void zip_1x8_2(const std::uint8_t* source, std::int32_t count,  in zip_1x8_2()
1335                std::int32_t stride, std::uint8_t* destination,  in zip_1x8_2()
1336                std::int32_t multiplicative_offset,  in zip_1x8_2()
1337                std::int32_t additive_offset) {  in zip_1x8_2()
1375 void zip_1x8_3(const std::uint8_t* source, std::int32_t count,  in zip_1x8_3()
1376                std::int32_t stride, std::uint8_t* destination,  in zip_1x8_3()
1377                std::int32_t multiplicative_offset,  in zip_1x8_3()
1378                std::int32_t additive_offset) {  in zip_1x8_3()
1417 void zip_1x8_4(const std::uint8_t* source, std::int32_t count,  in zip_1x8_4()
1418                std::int32_t stride, std::uint8_t* destination,  in zip_1x8_4()
1419                std::int32_t multiplicative_offset,  in zip_1x8_4()
1420                std::int32_t additive_offset) {  in zip_1x8_4()
1458 void zip_1x8_5(const std::uint8_t* source, std::int32_t count,  in zip_1x8_5()
1459                std::int32_t stride, std::uint8_t* destination,  in zip_1x8_5()
1460                std::int32_t multiplicative_offset,  in zip_1x8_5()
1461                std::int32_t additive_offset) {  in zip_1x8_5()
1500 void zip_1x8_6(const std::uint8_t* source, std::int32_t count,  in zip_1x8_6()
1501                std::int32_t stride, std::uint8_t* destination,  in zip_1x8_6()
1502                std::int32_t multiplicative_offset,  in zip_1x8_6()
1503                std::int32_t additive_offset) {  in zip_1x8_6()
1542 void zip_1x8_7(const std::uint8_t* source, std::int32_t count,  in zip_1x8_7()
1543                std::int32_t stride, std::uint8_t* destination,  in zip_1x8_7()
1544                std::int32_t multiplicative_offset,  in zip_1x8_7()
1545                std::int32_t additive_offset) {  in zip_1x8_7()
1585 void zip_2x8(const std::uint8_t* source, std::int32_t count,  in zip_2x8()
1586              std::int32_t stride, std::uint8_t* destination,  in zip_2x8()
1587              std::int32_t multiplicative_offset, std::int32_t additive_offset) {  in zip_2x8()
1625 void zip_2x8_1(const std::uint8_t* source, std::int32_t count,  in zip_2x8_1()
1626                std::int32_t stride, std::uint8_t* destination,  in zip_2x8_1()
1627                std::int32_t multiplicative_offset,  in zip_2x8_1()
1628                std::int32_t additive_offset) {  in zip_2x8_1()
1676 void zip_2x8_2(const std::uint8_t* source, std::int32_t count,  in zip_2x8_2()
1677                std::int32_t stride, std::uint8_t* destination,  in zip_2x8_2()
1678                std::int32_t multiplicative_offset,  in zip_2x8_2()
1679                std::int32_t additive_offset) {  in zip_2x8_2()
1727 void zip_2x8_3(const std::uint8_t* source, std::int32_t count,  in zip_2x8_3()
1728                std::int32_t stride, std::uint8_t* destination,  in zip_2x8_3()
1729                std::int32_t multiplicative_offset,  in zip_2x8_3()
1730                std::int32_t additive_offset) {  in zip_2x8_3()
1780 void zip_2x8_4(const std::uint8_t* source, std::int32_t count,  in zip_2x8_4()
1781                std::int32_t stride, std::uint8_t* destination,  in zip_2x8_4()
1782                std::int32_t multiplicative_offset,  in zip_2x8_4()
1783                std::int32_t additive_offset) {  in zip_2x8_4()
1831 void zip_2x8_5(const std::uint8_t* source, std::int32_t count,  in zip_2x8_5()
1832                std::int32_t stride, std::uint8_t* destination,  in zip_2x8_5()
1833                std::int32_t multiplicative_offset,  in zip_2x8_5()
1834                std::int32_t additive_offset) {  in zip_2x8_5()
1884 void zip_2x8_6(const std::uint8_t* source, std::int32_t count,  in zip_2x8_6()
1885                std::int32_t stride, std::uint8_t* destination,  in zip_2x8_6()
1886                std::int32_t multiplicative_offset,  in zip_2x8_6()
1887                std::int32_t additive_offset) {  in zip_2x8_6()
1937 void zip_2x8_7(const std::uint8_t* source, std::int32_t count,  in zip_2x8_7()
1938                std::int32_t stride, std::uint8_t* destination,  in zip_2x8_7()
1939                std::int32_t multiplicative_offset,  in zip_2x8_7()
1940                std::int32_t additive_offset) {  in zip_2x8_7()
1992 void zip_3x8(const std::uint8_t* source, std::int32_t count,  in zip_3x8()
1993              std::int32_t stride, std::uint8_t* destination,  in zip_3x8()
1994              std::int32_t multiplicative_offset, std::int32_t additive_offset) {  in zip_3x8()
2040 void zip_3x8_1(const std::uint8_t* source, std::int32_t count,  in zip_3x8_1()
2041                std::int32_t stride, std::uint8_t* destination,  in zip_3x8_1()
2042                std::int32_t multiplicative_offset,  in zip_3x8_1()
2043                std::int32_t additive_offset) {  in zip_3x8_1()
2102 void zip_3x8_2(const std::uint8_t* source, std::int32_t count,  in zip_3x8_2()
2103                std::int32_t stride, std::uint8_t* destination,  in zip_3x8_2()
2104                std::int32_t multiplicative_offset,  in zip_3x8_2()
2105                std::int32_t additive_offset) {  in zip_3x8_2()
2164 void zip_3x8_3(const std::uint8_t* source, std::int32_t count,  in zip_3x8_3()
2165                std::int32_t stride, std::uint8_t* destination,  in zip_3x8_3()
2166                std::int32_t multiplicative_offset,  in zip_3x8_3()
2167                std::int32_t additive_offset) {  in zip_3x8_3()
2229 void zip_3x8_4(const std::uint8_t* source, std::int32_t count,  in zip_3x8_4()
2230                std::int32_t stride, std::uint8_t* destination,  in zip_3x8_4()
2231                std::int32_t multiplicative_offset,  in zip_3x8_4()
2232                std::int32_t additive_offset) {  in zip_3x8_4()
2291 void zip_3x8_5(const std::uint8_t* source, std::int32_t count,  in zip_3x8_5()
2292                std::int32_t stride, std::uint8_t* destination,  in zip_3x8_5()
2293                std::int32_t multiplicative_offset,  in zip_3x8_5()
2294                std::int32_t additive_offset) {  in zip_3x8_5()
2356 void zip_3x8_6(const std::uint8_t* source, std::int32_t count,  in zip_3x8_6()
2357                std::int32_t stride, std::uint8_t* destination,  in zip_3x8_6()
2358                std::int32_t multiplicative_offset,  in zip_3x8_6()
2359                std::int32_t additive_offset) {  in zip_3x8_6()
2421 void zip_3x8_7(const std::uint8_t* source, std::int32_t count,  in zip_3x8_7()
2422                std::int32_t stride, std::uint8_t* destination,  in zip_3x8_7()
2423                std::int32_t multiplicative_offset,  in zip_3x8_7()
2424                std::int32_t additive_offset) {  in zip_3x8_7()
2491                                      std::int32_t count, std::int32_t* result,  in mul_1x8_1x8_int32_rhsadd()
2492                                      std::int32_t result_stride) {  in mul_1x8_1x8_int32_rhsadd()
2536                                      std::int32_t count, std::int32_t* result,  in mul_1x8_2x8_int32_rhsadd()
2537                                      std::int32_t result_stride) {  in mul_1x8_2x8_int32_rhsadd()
2586                                      std::int32_t count, std::int32_t* result,  in mul_1x8_3x8_int32_rhsadd()
2587                                      std::int32_t result_stride) {  in mul_1x8_3x8_int32_rhsadd()
2646                                      std::int32_t count, std::int32_t* result,  in mul_2x8_1x8_int32_rhsadd()
2647                                      std::int32_t result_stride) {  in mul_2x8_1x8_int32_rhsadd()
2699                                      std::int32_t count, std::int32_t* result,  in mul_2x8_2x8_int32_rhsadd()
2700                                      std::int32_t result_stride) {  in mul_2x8_2x8_int32_rhsadd()
2761                                      std::int32_t count, std::int32_t* result,  in mul_2x8_3x8_int32_rhsadd()
2762                                      std::int32_t result_stride) {  in mul_2x8_3x8_int32_rhsadd()
2840                                      std::int32_t count, std::int32_t* result,  in mul_3x8_1x8_int32_rhsadd()
2841                                      std::int32_t result_stride) {  in mul_3x8_1x8_int32_rhsadd()
2900                                      std::int32_t count, std::int32_t* result,  in mul_3x8_2x8_int32_rhsadd()
2901                                      std::int32_t result_stride) {  in mul_3x8_2x8_int32_rhsadd()
2973                                      std::int32_t count, std::int32_t* result,  in mul_3x8_3x8_int32_rhsadd()
2974                                      std::int32_t result_stride) {  in mul_3x8_3x8_int32_rhsadd()
3071                                             std::int32_t count,  in mul_1x8_1x8_int32_lhsadd_rhsadd()
3072                                             std::int32_t* result,  in mul_1x8_1x8_int32_lhsadd_rhsadd()
3073                                             std::int32_t result_stride) {  in mul_1x8_1x8_int32_lhsadd_rhsadd()
3122                                             std::int32_t count,  in mul_1x8_2x8_int32_lhsadd_rhsadd()
3123                                             std::int32_t* result,  in mul_1x8_2x8_int32_lhsadd_rhsadd()
3124                                             std::int32_t result_stride) {  in mul_1x8_2x8_int32_lhsadd_rhsadd()
3178                                             std::int32_t count,  in mul_1x8_3x8_int32_lhsadd_rhsadd()
3179                                             std::int32_t* result,  in mul_1x8_3x8_int32_lhsadd_rhsadd()
3180                                             std::int32_t result_stride) {  in mul_1x8_3x8_int32_lhsadd_rhsadd()
3244                                             std::int32_t count,  in mul_2x8_1x8_int32_lhsadd_rhsadd()
3245                                             std::int32_t* result,  in mul_2x8_1x8_int32_lhsadd_rhsadd()
3246                                             std::int32_t result_stride) {  in mul_2x8_1x8_int32_lhsadd_rhsadd()
3305                                             std::int32_t count,  in mul_2x8_2x8_int32_lhsadd_rhsadd()
3306                                             std::int32_t* result,  in mul_2x8_2x8_int32_lhsadd_rhsadd()
3307                                             std::int32_t result_stride) {  in mul_2x8_2x8_int32_lhsadd_rhsadd()
3375                                             std::int32_t count,  in mul_2x8_3x8_int32_lhsadd_rhsadd()
3376                                             std::int32_t* result,  in mul_2x8_3x8_int32_lhsadd_rhsadd()
3377                                             std::int32_t result_stride) {  in mul_2x8_3x8_int32_lhsadd_rhsadd()
3463                                             std::int32_t count,  in mul_3x8_1x8_int32_lhsadd_rhsadd()
3464                                             std::int32_t* result,  in mul_3x8_1x8_int32_lhsadd_rhsadd()
3465                                             std::int32_t result_stride) {  in mul_3x8_1x8_int32_lhsadd_rhsadd()
3533                                             std::int32_t count,  in mul_3x8_2x8_int32_lhsadd_rhsadd()
3534                                             std::int32_t* result,  in mul_3x8_2x8_int32_lhsadd_rhsadd()
3535                                             std::int32_t result_stride) {  in mul_3x8_2x8_int32_lhsadd_rhsadd()
3617                                             std::int32_t count,  in mul_3x8_3x8_int32_lhsadd_rhsadd()
3618                                             std::int32_t* result,  in mul_3x8_3x8_int32_lhsadd_rhsadd()
3619                                             std::int32_t result_stride) {  in mul_3x8_3x8_int32_lhsadd_rhsadd()
3725                                             std::int32_t count, float* result,  in mul_1x8_1x8_float_lhsadd_rhsadd()
3726                                             std::int32_t result_stride,  in mul_1x8_1x8_float_lhsadd_rhsadd()
3782                                             std::int32_t count, float* result,  in mul_1x8_2x8_float_lhsadd_rhsadd()
3783                                             std::int32_t result_stride,  in mul_1x8_2x8_float_lhsadd_rhsadd()
3844                                             std::int32_t count, float* result,  in mul_1x8_3x8_float_lhsadd_rhsadd()
3845                                             std::int32_t result_stride,  in mul_1x8_3x8_float_lhsadd_rhsadd()
3916                                             std::int32_t count, float* result,  in mul_2x8_1x8_float_lhsadd_rhsadd()
3917                                             std::int32_t result_stride,  in mul_2x8_1x8_float_lhsadd_rhsadd()
3985                                             std::int32_t count, float* result,  in mul_2x8_2x8_float_lhsadd_rhsadd()
3986                                             std::int32_t result_stride,  in mul_2x8_2x8_float_lhsadd_rhsadd()
4063                                             std::int32_t count, float* result,  in mul_2x8_3x8_float_lhsadd_rhsadd()
4064                                             std::int32_t result_stride,  in mul_2x8_3x8_float_lhsadd_rhsadd()
4159                                             std::int32_t count, float* result,  in mul_3x8_1x8_float_lhsadd_rhsadd()
4160                                             std::int32_t result_stride,  in mul_3x8_1x8_float_lhsadd_rhsadd()
4239                                             std::int32_t count, float* result,  in mul_3x8_2x8_float_lhsadd_rhsadd()
4240                                             std::int32_t result_stride,  in mul_3x8_2x8_float_lhsadd_rhsadd()
4333                                             std::int32_t count, float* result,  in mul_3x8_3x8_float_lhsadd_rhsadd()
4334                                             std::int32_t result_stride,  in mul_3x8_3x8_float_lhsadd_rhsadd()
4449 void qnt_1x8_aligned(const std::int32_t* source, std::int32_t count,  in qnt_1x8_aligned()
4450                      std::int32_t stride, const std::int32_t* offsets,  in qnt_1x8_aligned()
4451                      std::uint8_t* destination, std::int32_t destination_stride,  in qnt_1x8_aligned()
4452                      std::int32_t multiplicative_offset,  in qnt_1x8_aligned()
4453                      std::int32_t rounding_offset, std::int32_t shift) {  in qnt_1x8_aligned()
4489 void qnt_1x8_1_aligned(const std::int32_t* source, std::int32_t count,  in qnt_1x8_1_aligned()
4490                        std::int32_t stride, const std::int32_t* offsets,  in qnt_1x8_1_aligned()
4492                        std::int32_t destination_stride,  in qnt_1x8_1_aligned()
4493                        std::int32_t multiplicative_offset,  in qnt_1x8_1_aligned()
4494                        std::int32_t rounding_offset, std::int32_t shift) {  in qnt_1x8_1_aligned()
4541 void qnt_1x8_2_aligned(const std::int32_t* source, std::int32_t count,  in qnt_1x8_2_aligned()
4542                        std::int32_t stride, const std::int32_t* offsets,  in qnt_1x8_2_aligned()
4544                        std::int32_t destination_stride,  in qnt_1x8_2_aligned()
4545                        std::int32_t multiplicative_offset,  in qnt_1x8_2_aligned()
4546                        std::int32_t rounding_offset, std::int32_t shift) {  in qnt_1x8_2_aligned()
4593 void qnt_1x8_3_aligned(const std::int32_t* source, std::int32_t count,  in qnt_1x8_3_aligned()
4594                        std::int32_t stride, const std::int32_t* offsets,  in qnt_1x8_3_aligned()
4596                        std::int32_t destination_stride,  in qnt_1x8_3_aligned()
4597                        std::int32_t multiplicative_offset,  in qnt_1x8_3_aligned()
4598                        std::int32_t rounding_offset, std::int32_t shift) {  in qnt_1x8_3_aligned()
4647 void qnt_1x8_4_aligned(const std::int32_t* source, std::int32_t count,  in qnt_1x8_4_aligned()
4648                        std::int32_t stride, const std::int32_t* offsets,  in qnt_1x8_4_aligned()
4650                        std::int32_t destination_stride,  in qnt_1x8_4_aligned()
4651                        std::int32_t multiplicative_offset,  in qnt_1x8_4_aligned()
4652                        std::int32_t rounding_offset, std::int32_t shift) {  in qnt_1x8_4_aligned()
4699 void qnt_1x8_5_aligned(const std::int32_t* source, std::int32_t count,  in qnt_1x8_5_aligned()
4700                        std::int32_t stride, const std::int32_t* offsets,  in qnt_1x8_5_aligned()
4702                        std::int32_t destination_stride,  in qnt_1x8_5_aligned()
4703                        std::int32_t multiplicative_offset,  in qnt_1x8_5_aligned()
4704                        std::int32_t rounding_offset, std::int32_t shift) {  in qnt_1x8_5_aligned()
4758 void qnt_1x8_6_aligned(const std::int32_t* source, std::int32_t count,  in qnt_1x8_6_aligned()
4759                        std::int32_t stride, const std::int32_t* offsets,  in qnt_1x8_6_aligned()
4761                        std::int32_t destination_stride,  in qnt_1x8_6_aligned()
4762                        std::int32_t multiplicative_offset,  in qnt_1x8_6_aligned()
4763                        std::int32_t rounding_offset, std::int32_t shift) {  in qnt_1x8_6_aligned()
4816 void qnt_1x8_7_aligned(const std::int32_t* source, std::int32_t count,  in qnt_1x8_7_aligned()
4817                        std::int32_t stride, const std::int32_t* offsets,  in qnt_1x8_7_aligned()
4819                        std::int32_t destination_stride,  in qnt_1x8_7_aligned()
4820                        std::int32_t multiplicative_offset,  in qnt_1x8_7_aligned()
4821                        std::int32_t rounding_offset, std::int32_t shift) {  in qnt_1x8_7_aligned()
4876 void qnt_2x8_aligned(const std::int32_t* source, std::int32_t count,  in qnt_2x8_aligned()
4877                      std::int32_t stride, const std::int32_t* offsets,  in qnt_2x8_aligned()
4878                      std::uint8_t* destination, std::int32_t destination_stride,  in qnt_2x8_aligned()
4879                      std::int32_t multiplicative_offset,  in qnt_2x8_aligned()
4880                      std::int32_t rounding_offset, std::int32_t shift) {  in qnt_2x8_aligned()
4934 void qnt_2x8_1_aligned(const std::int32_t* source, std::int32_t count,  in qnt_2x8_1_aligned()
4935                        std::int32_t stride, const std::int32_t* offsets,  in qnt_2x8_1_aligned()
4937                        std::int32_t destination_stride,  in qnt_2x8_1_aligned()
4938                        std::int32_t multiplicative_offset,  in qnt_2x8_1_aligned()
4939                        std::int32_t rounding_offset, std::int32_t shift) {  in qnt_2x8_1_aligned()
5012 void qnt_2x8_2_aligned(const std::int32_t* source, std::int32_t count,  in qnt_2x8_2_aligned()
5013                        std::int32_t stride, const std::int32_t* offsets,  in qnt_2x8_2_aligned()
5015                        std::int32_t destination_stride,  in qnt_2x8_2_aligned()
5016                        std::int32_t multiplicative_offset,  in qnt_2x8_2_aligned()
5017                        std::int32_t rounding_offset, std::int32_t shift) {  in qnt_2x8_2_aligned()
5090 void qnt_2x8_3_aligned(const std::int32_t* source, std::int32_t count,  in qnt_2x8_3_aligned()
5091                        std::int32_t stride, const std::int32_t* offsets,  in qnt_2x8_3_aligned()
5093                        std::int32_t destination_stride,  in qnt_2x8_3_aligned()
5094                        std::int32_t multiplicative_offset,  in qnt_2x8_3_aligned()
5095                        std::int32_t rounding_offset, std::int32_t shift) {  in qnt_2x8_3_aligned()
5172 void qnt_2x8_4_aligned(const std::int32_t* source, std::int32_t count,  in qnt_2x8_4_aligned()
5173                        std::int32_t stride, const std::int32_t* offsets,  in qnt_2x8_4_aligned()
5175                        std::int32_t destination_stride,  in qnt_2x8_4_aligned()
5176                        std::int32_t multiplicative_offset,  in qnt_2x8_4_aligned()
5177                        std::int32_t rounding_offset, std::int32_t shift) {  in qnt_2x8_4_aligned()
5250 void qnt_2x8_5_aligned(const std::int32_t* source, std::int32_t count,  in qnt_2x8_5_aligned()
5251                        std::int32_t stride, const std::int32_t* offsets,  in qnt_2x8_5_aligned()
5253                        std::int32_t destination_stride,  in qnt_2x8_5_aligned()
5254                        std::int32_t multiplicative_offset,  in qnt_2x8_5_aligned()
5255                        std::int32_t rounding_offset, std::int32_t shift) {  in qnt_2x8_5_aligned()
5342 void qnt_2x8_6_aligned(const std::int32_t* source, std::int32_t count,  in qnt_2x8_6_aligned()
5343                        std::int32_t stride, const std::int32_t* offsets,  in qnt_2x8_6_aligned()
5345                        std::int32_t destination_stride,  in qnt_2x8_6_aligned()
5346                        std::int32_t multiplicative_offset,  in qnt_2x8_6_aligned()
5347                        std::int32_t rounding_offset, std::int32_t shift) {  in qnt_2x8_6_aligned()
5432 void qnt_2x8_7_aligned(const std::int32_t* source, std::int32_t count,  in qnt_2x8_7_aligned()
5433                        std::int32_t stride, const std::int32_t* offsets,  in qnt_2x8_7_aligned()
5435                        std::int32_t destination_stride,  in qnt_2x8_7_aligned()
5436                        std::int32_t multiplicative_offset,  in qnt_2x8_7_aligned()
5437                        std::int32_t rounding_offset, std::int32_t shift) {  in qnt_2x8_7_aligned()
5526 void qnt_3x8_aligned(const std::int32_t* source, std::int32_t count,  in qnt_3x8_aligned()
5527                      std::int32_t stride, const std::int32_t* offsets,  in qnt_3x8_aligned()
5528                      std::uint8_t* destination, std::int32_t destination_stride,  in qnt_3x8_aligned()
5529                      std::int32_t multiplicative_offset,  in qnt_3x8_aligned()
5530                      std::int32_t rounding_offset, std::int32_t shift) {  in qnt_3x8_aligned()
5602 void qnt_3x8_1_aligned(const std::int32_t* source, std::int32_t count,  in qnt_3x8_1_aligned()
5603                        std::int32_t stride, const std::int32_t* offsets,  in qnt_3x8_1_aligned()
5605                        std::int32_t destination_stride,  in qnt_3x8_1_aligned()
5606                        std::int32_t multiplicative_offset,  in qnt_3x8_1_aligned()
5607                        std::int32_t rounding_offset, std::int32_t shift) {  in qnt_3x8_1_aligned()
5706 void qnt_3x8_2_aligned(const std::int32_t* source, std::int32_t count,  in qnt_3x8_2_aligned()
5707                        std::int32_t stride, const std::int32_t* offsets,  in qnt_3x8_2_aligned()
5709                        std::int32_t destination_stride,  in qnt_3x8_2_aligned()
5710                        std::int32_t multiplicative_offset,  in qnt_3x8_2_aligned()
5711                        std::int32_t rounding_offset, std::int32_t shift) {  in qnt_3x8_2_aligned()
5810 void qnt_3x8_3_aligned(const std::int32_t* source, std::int32_t count,  in qnt_3x8_3_aligned()
5811                        std::int32_t stride, const std::int32_t* offsets,  in qnt_3x8_3_aligned()
5813                        std::int32_t destination_stride,  in qnt_3x8_3_aligned()
5814                        std::int32_t multiplicative_offset,  in qnt_3x8_3_aligned()
5815                        std::int32_t rounding_offset, std::int32_t shift) {  in qnt_3x8_3_aligned()
5920 void qnt_3x8_4_aligned(const std::int32_t* source, std::int32_t count,  in qnt_3x8_4_aligned()
5921                        std::int32_t stride, const std::int32_t* offsets,  in qnt_3x8_4_aligned()
5923                        std::int32_t destination_stride,  in qnt_3x8_4_aligned()
5924                        std::int32_t multiplicative_offset,  in qnt_3x8_4_aligned()
5925                        std::int32_t rounding_offset, std::int32_t shift) {  in qnt_3x8_4_aligned()
6024 void qnt_3x8_5_aligned(const std::int32_t* source, std::int32_t count,  in qnt_3x8_5_aligned()
6025                        std::int32_t stride, const std::int32_t* offsets,  in qnt_3x8_5_aligned()
6027                        std::int32_t destination_stride,  in qnt_3x8_5_aligned()
6028                        std::int32_t multiplicative_offset,  in qnt_3x8_5_aligned()
6029                        std::int32_t rounding_offset, std::int32_t shift) {  in qnt_3x8_5_aligned()
6149 void qnt_3x8_6_aligned(const std::int32_t* source, std::int32_t count,  in qnt_3x8_6_aligned()
6150                        std::int32_t stride, const std::int32_t* offsets,  in qnt_3x8_6_aligned()
6152                        std::int32_t destination_stride,  in qnt_3x8_6_aligned()
6153                        std::int32_t multiplicative_offset,  in qnt_3x8_6_aligned()
6154                        std::int32_t rounding_offset, std::int32_t shift) {  in qnt_3x8_6_aligned()
6271 void qnt_3x8_7_aligned(const std::int32_t* source, std::int32_t count,  in qnt_3x8_7_aligned()
6272                        std::int32_t stride, const std::int32_t* offsets,  in qnt_3x8_7_aligned()
6274                        std::int32_t destination_stride,  in qnt_3x8_7_aligned()
6275                        std::int32_t multiplicative_offset,  in qnt_3x8_7_aligned()
6276                        std::int32_t rounding_offset, std::int32_t shift) {  in qnt_3x8_7_aligned()
6399 void qnt_1x8(const std::int32_t* source, std::int32_t count,  in qnt_1x8()
6400              std::int32_t stride, const std::int32_t* offsets,  in qnt_1x8()
6401              std::uint8_t* destination, std::int32_t destination_stride,  in qnt_1x8()
6402              std::int32_t multiplicative_offset, std::int32_t rounding_offset,  in qnt_1x8()
6403              std::int32_t shift) {  in qnt_1x8()
6439 void qnt_1x8_1(const std::int32_t* source, std::int32_t count,  in qnt_1x8_1()
6440                std::int32_t stride, const std::int32_t* offsets,  in qnt_1x8_1()
6441                std::uint8_t* destination, std::int32_t destination_stride,  in qnt_1x8_1()
6442                std::int32_t multiplicative_offset, std::int32_t rounding_offset,  in qnt_1x8_1()
6443                std::int32_t shift) {  in qnt_1x8_1()
6490 void qnt_1x8_2(const std::int32_t* source, std::int32_t count,  in qnt_1x8_2()
6491                std::int32_t stride, const std::int32_t* offsets,  in qnt_1x8_2()
6492                std::uint8_t* destination, std::int32_t destination_stride,  in qnt_1x8_2()
6493                std::int32_t multiplicative_offset, std::int32_t rounding_offset,  in qnt_1x8_2()
6494                std::int32_t shift) {  in qnt_1x8_2()
6541 void qnt_1x8_3(const std::int32_t* source, std::int32_t count,  in qnt_1x8_3()
6542                std::int32_t stride, const std::int32_t* offsets,  in qnt_1x8_3()
6543                std::uint8_t* destination, std::int32_t destination_stride,  in qnt_1x8_3()
6544                std::int32_t multiplicative_offset, std::int32_t rounding_offset,  in qnt_1x8_3()
6545                std::int32_t shift) {  in qnt_1x8_3()
6594 void qnt_1x8_4(const std::int32_t* source, std::int32_t count,  in qnt_1x8_4()
6595                std::int32_t stride, const std::int32_t* offsets,  in qnt_1x8_4()
6596                std::uint8_t* destination, std::int32_t destination_stride,  in qnt_1x8_4()
6597                std::int32_t multiplicative_offset, std::int32_t rounding_offset,  in qnt_1x8_4()
6598                std::int32_t shift) {  in qnt_1x8_4()
6645 void qnt_1x8_5(const std::int32_t* source, std::int32_t count,  in qnt_1x8_5()
6646                std::int32_t stride, const std::int32_t* offsets,  in qnt_1x8_5()
6647                std::uint8_t* destination, std::int32_t destination_stride,  in qnt_1x8_5()
6648                std::int32_t multiplicative_offset, std::int32_t rounding_offset,  in qnt_1x8_5()
6649                std::int32_t shift) {  in qnt_1x8_5()
6703 void qnt_1x8_6(const std::int32_t* source, std::int32_t count,  in qnt_1x8_6()
6704                std::int32_t stride, const std::int32_t* offsets,  in qnt_1x8_6()
6705                std::uint8_t* destination, std::int32_t destination_stride,  in qnt_1x8_6()
6706                std::int32_t multiplicative_offset, std::int32_t rounding_offset,  in qnt_1x8_6()
6707                std::int32_t shift) {  in qnt_1x8_6()
6760 void qnt_1x8_7(const std::int32_t* source, std::int32_t count,  in qnt_1x8_7()
6761                std::int32_t stride, const std::int32_t* offsets,  in qnt_1x8_7()
6762                std::uint8_t* destination, std::int32_t destination_stride,  in qnt_1x8_7()
6763                std::int32_t multiplicative_offset, std::int32_t rounding_offset,  in qnt_1x8_7()
6764                std::int32_t shift) {  in qnt_1x8_7()
6819 void qnt_2x8(const std::int32_t* source, std::int32_t count,  in qnt_2x8()
6820              std::int32_t stride, const std::int32_t* offsets,  in qnt_2x8()
6821              std::uint8_t* destination, std::int32_t destination_stride,  in qnt_2x8()
6822              std::int32_t multiplicative_offset, std::int32_t rounding_offset,  in qnt_2x8()
6823              std::int32_t shift) {  in qnt_2x8()
6877 void qnt_2x8_1(const std::int32_t* source, std::int32_t count,  in qnt_2x8_1()
6878                std::int32_t stride, const std::int32_t* offsets,  in qnt_2x8_1()
6879                std::uint8_t* destination, std::int32_t destination_stride,  in qnt_2x8_1()
6880                std::int32_t multiplicative_offset, std::int32_t rounding_offset,  in qnt_2x8_1()
6881                std::int32_t shift) {  in qnt_2x8_1()
6954 void qnt_2x8_2(const std::int32_t* source, std::int32_t count,  in qnt_2x8_2()
6955                std::int32_t stride, const std::int32_t* offsets,  in qnt_2x8_2()
6956                std::uint8_t* destination, std::int32_t destination_stride,  in qnt_2x8_2()
6957                std::int32_t multiplicative_offset, std::int32_t rounding_offset,  in qnt_2x8_2()
6958                std::int32_t shift) {  in qnt_2x8_2()
7031 void qnt_2x8_3(const std::int32_t* source, std::int32_t count,  in qnt_2x8_3()
7032                std::int32_t stride, const std::int32_t* offsets,  in qnt_2x8_3()
7033                std::uint8_t* destination, std::int32_t destination_stride,  in qnt_2x8_3()
7034                std::int32_t multiplicative_offset, std::int32_t rounding_offset,  in qnt_2x8_3()
7035                std::int32_t shift) {  in qnt_2x8_3()
7112 void qnt_2x8_4(const std::int32_t* source, std::int32_t count,  in qnt_2x8_4()
7113                std::int32_t stride, const std::int32_t* offsets,  in qnt_2x8_4()
7114                std::uint8_t* destination, std::int32_t destination_stride,  in qnt_2x8_4()
7115                std::int32_t multiplicative_offset, std::int32_t rounding_offset,  in qnt_2x8_4()
7116                std::int32_t shift) {  in qnt_2x8_4()
7189 void qnt_2x8_5(const std::int32_t* source, std::int32_t count,  in qnt_2x8_5()
7190                std::int32_t stride, const std::int32_t* offsets,  in qnt_2x8_5()
7191                std::uint8_t* destination, std::int32_t destination_stride,  in qnt_2x8_5()
7192                std::int32_t multiplicative_offset, std::int32_t rounding_offset,  in qnt_2x8_5()
7193                std::int32_t shift) {  in qnt_2x8_5()
7280 void qnt_2x8_6(const std::int32_t* source, std::int32_t count,  in qnt_2x8_6()
7281                std::int32_t stride, const std::int32_t* offsets,  in qnt_2x8_6()
7282                std::uint8_t* destination, std::int32_t destination_stride,  in qnt_2x8_6()
7283                std::int32_t multiplicative_offset, std::int32_t rounding_offset,  in qnt_2x8_6()
7284                std::int32_t shift) {  in qnt_2x8_6()
7369 void qnt_2x8_7(const std::int32_t* source, std::int32_t count,  in qnt_2x8_7()
7370                std::int32_t stride, const std::int32_t* offsets,  in qnt_2x8_7()
7371                std::uint8_t* destination, std::int32_t destination_stride,  in qnt_2x8_7()
7372                std::int32_t multiplicative_offset, std::int32_t rounding_offset,  in qnt_2x8_7()
7373                std::int32_t shift) {  in qnt_2x8_7()
7462 void qnt_3x8(const std::int32_t* source, std::int32_t count,  in qnt_3x8()
7463              std::int32_t stride, const std::int32_t* offsets,  in qnt_3x8()
7464              std::uint8_t* destination, std::int32_t destination_stride,  in qnt_3x8()
7465              std::int32_t multiplicative_offset, std::int32_t rounding_offset,  in qnt_3x8()
7466              std::int32_t shift) {  in qnt_3x8()
7538 void qnt_3x8_1(const std::int32_t* source, std::int32_t count,  in qnt_3x8_1()
7539                std::int32_t stride, const std::int32_t* offsets,  in qnt_3x8_1()
7540                std::uint8_t* destination, std::int32_t destination_stride,  in qnt_3x8_1()
7541                std::int32_t multiplicative_offset, std::int32_t rounding_offset,  in qnt_3x8_1()
7542                std::int32_t shift) {  in qnt_3x8_1()
7641 void qnt_3x8_2(const std::int32_t* source, std::int32_t count,  in qnt_3x8_2()
7642                std::int32_t stride, const std::int32_t* offsets,  in qnt_3x8_2()
7643                std::uint8_t* destination, std::int32_t destination_stride,  in qnt_3x8_2()
7644                std::int32_t multiplicative_offset, std::int32_t rounding_offset,  in qnt_3x8_2()
7645                std::int32_t shift) {  in qnt_3x8_2()
7744 void qnt_3x8_3(const std::int32_t* source, std::int32_t count,  in qnt_3x8_3()
7745                std::int32_t stride, const std::int32_t* offsets,  in qnt_3x8_3()
7746                std::uint8_t* destination, std::int32_t destination_stride,  in qnt_3x8_3()
7747                std::int32_t multiplicative_offset, std::int32_t rounding_offset,  in qnt_3x8_3()
7748                std::int32_t shift) {  in qnt_3x8_3()
7853 void qnt_3x8_4(const std::int32_t* source, std::int32_t count,  in qnt_3x8_4()
7854                std::int32_t stride, const std::int32_t* offsets,  in qnt_3x8_4()
7855                std::uint8_t* destination, std::int32_t destination_stride,  in qnt_3x8_4()
7856                std::int32_t multiplicative_offset, std::int32_t rounding_offset,  in qnt_3x8_4()
7857                std::int32_t shift) {  in qnt_3x8_4()
7956 void qnt_3x8_5(const std::int32_t* source, std::int32_t count,  in qnt_3x8_5()
7957                std::int32_t stride, const std::int32_t* offsets,  in qnt_3x8_5()
7958                std::uint8_t* destination, std::int32_t destination_stride,  in qnt_3x8_5()
7959                std::int32_t multiplicative_offset, std::int32_t rounding_offset,  in qnt_3x8_5()
7960                std::int32_t shift) {  in qnt_3x8_5()
8080 void qnt_3x8_6(const std::int32_t* source, std::int32_t count,  in qnt_3x8_6()
8081                std::int32_t stride, const std::int32_t* offsets,  in qnt_3x8_6()
8082                std::uint8_t* destination, std::int32_t destination_stride,  in qnt_3x8_6()
8083                std::int32_t multiplicative_offset, std::int32_t rounding_offset,  in qnt_3x8_6()
8084                std::int32_t shift) {  in qnt_3x8_6()
8201 void qnt_3x8_7(const std::int32_t* source, std::int32_t count,  in qnt_3x8_7()
8202                std::int32_t stride, const std::int32_t* offsets,  in qnt_3x8_7()
8203                std::uint8_t* destination, std::int32_t destination_stride,  in qnt_3x8_7()
8204                std::int32_t multiplicative_offset, std::int32_t rounding_offset,  in qnt_3x8_7()
8205                std::int32_t shift) {  in qnt_3x8_7()
8328 void multi_qnt_1x8_aligned(const std::int32_t* source, std::int32_t count,  in multi_qnt_1x8_aligned()
8329                            std::int32_t stride, const std::int32_t* offsets,  in multi_qnt_1x8_aligned()
8331                            std::int32_t destination_stride,  in multi_qnt_1x8_aligned()
8332                            std::int32_t multiplicative_offset,  in multi_qnt_1x8_aligned()
8333                            std::int32_t rounding_offset, std::int32_t shift) {  in multi_qnt_1x8_aligned()
8378 void multi_qnt_2x8_aligned(const std::int32_t* source, std::int32_t count,  in multi_qnt_2x8_aligned()
8379                            std::int32_t stride, const std::int32_t* offsets,  in multi_qnt_2x8_aligned()
8381                            std::int32_t destination_stride,  in multi_qnt_2x8_aligned()
8382                            std::int32_t multiplicative_offset,  in multi_qnt_2x8_aligned()
8383                            std::int32_t rounding_offset, std::int32_t shift) {  in multi_qnt_2x8_aligned()
8428 void multi_qnt_3x8_aligned(const std::int32_t* source, std::int32_t count,  in multi_qnt_3x8_aligned()
8429                            std::int32_t stride, const std::int32_t* offsets,  in multi_qnt_3x8_aligned()
8431                            std::int32_t destination_stride,  in multi_qnt_3x8_aligned()
8432                            std::int32_t multiplicative_offset,  in multi_qnt_3x8_aligned()
8433                            std::int32_t rounding_offset, std::int32_t shift) {  in multi_qnt_3x8_aligned()
8478 void multi_qnt_1x8(const std::int32_t* source, std::int32_t count,  in multi_qnt_1x8()
8479                    std::int32_t stride, const std::int32_t* offsets,  in multi_qnt_1x8()
8480                    std::uint8_t* destination, std::int32_t destination_stride,  in multi_qnt_1x8()
8481                    std::int32_t multiplicative_offset,  in multi_qnt_1x8()
8482                    std::int32_t rounding_offset, std::int32_t shift) {  in multi_qnt_1x8()
8519 void multi_qnt_2x8(const std::int32_t* source, std::int32_t count,  in multi_qnt_2x8()
8520                    std::int32_t stride, const std::int32_t* offsets,  in multi_qnt_2x8()
8521                    std::uint8_t* destination, std::int32_t destination_stride,  in multi_qnt_2x8()
8522                    std::int32_t multiplicative_offset,  in multi_qnt_2x8()
8523                    std::int32_t rounding_offset, std::int32_t shift) {  in multi_qnt_2x8()
8560 void multi_qnt_3x8(const std::int32_t* source, std::int32_t count,  in multi_qnt_3x8()
8561                    std::int32_t stride, const std::int32_t* offsets,  in multi_qnt_3x8()
8562                    std::uint8_t* destination, std::int32_t destination_stride,  in multi_qnt_3x8()
8563                    std::int32_t multiplicative_offset,  in multi_qnt_3x8()
8564                    std::int32_t rounding_offset, std::int32_t shift) {  in multi_qnt_3x8()
8602                            const std::uint8_t* rhs, std::int32_t m,  in gemm_q8_0_0_0_aligned()
8603                            std::int32_t n, std::int32_t k,  in gemm_q8_0_0_0_aligned()
8604                            std::int32_t lhs_offset, std::int32_t rhs_offset,  in gemm_q8_0_0_0_aligned()
8605                            std::int32_t result_offset,  in gemm_q8_0_0_0_aligned()
8606                            std::int32_t multiplicative_offset,  in gemm_q8_0_0_0_aligned()
8607                            std::int32_t shift, std::uint8_t* result,  in gemm_q8_0_0_0_aligned()
8608                            std::int32_t result_stride) {  in gemm_q8_0_0_0_aligned()
8609   const std::int32_t row_chunks = m / 3;  in gemm_q8_0_0_0_aligned()
8610   const std::int32_t col_chunks = n / 3;  in gemm_q8_0_0_0_aligned()
8611   const std::int32_t padded_k = ((k + 7) / 8) * 8;  in gemm_q8_0_0_0_aligned()
8612   const std::int32_t chunk_size = k * 3;  in gemm_q8_0_0_0_aligned()
8613   const std::int32_t zipped_chunk_size = (padded_k + 16) * 3;  in gemm_q8_0_0_0_aligned()
8614   const std::int32_t zipped_rhs_size = (padded_k + 16) * n;  in gemm_q8_0_0_0_aligned()
8618   std::int32_t* zipped_lhs_3_offsets =  in gemm_q8_0_0_0_aligned()
8619       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 3);  in gemm_q8_0_0_0_aligned()
8622   const std::int32_t result_chunk_stride = result_stride * 3;  in gemm_q8_0_0_0_aligned()
8624   const std::int32_t const_offset = lhs_offset * rhs_offset * k + result_offset;  in gemm_q8_0_0_0_aligned()
8625   const std::int32_t rounding_offset = (1 << (shift - 1));  in gemm_q8_0_0_0_aligned()
8626   std::int32_t* temp_result = reinterpret_cast<std::int32_t*>(  in gemm_q8_0_0_0_aligned()
8629   std::int32_t* mul_result_chunk = temp_result;  in gemm_q8_0_0_0_aligned()
8630   const std::int32_t mul_result_chunk_stride_bytes = ((n * 4 + 7) / 8) * 8;  in gemm_q8_0_0_0_aligned()
8657                            const std::uint8_t* rhs, std::int32_t m,  in gemm_q8_0_0_1_aligned()
8658                            std::int32_t n, std::int32_t k,  in gemm_q8_0_0_1_aligned()
8659                            std::int32_t lhs_offset, std::int32_t rhs_offset,  in gemm_q8_0_0_1_aligned()
8660                            std::int32_t result_offset,  in gemm_q8_0_0_1_aligned()
8661                            std::int32_t multiplicative_offset,  in gemm_q8_0_0_1_aligned()
8662                            std::int32_t shift, std::uint8_t* result,  in gemm_q8_0_0_1_aligned()
8663                            std::int32_t result_stride) {  in gemm_q8_0_0_1_aligned()
8664   const std::int32_t row_chunks = m / 3;  in gemm_q8_0_0_1_aligned()
8665   const std::int32_t col_chunks = n / 3;  in gemm_q8_0_0_1_aligned()
8666   const std::int32_t padded_k = ((k + 7) / 8) * 8;  in gemm_q8_0_0_1_aligned()
8667   const std::int32_t chunk_size = k * 3;  in gemm_q8_0_0_1_aligned()
8668   const std::int32_t zipped_chunk_size = (padded_k + 16) * 3;  in gemm_q8_0_0_1_aligned()
8669   const std::int32_t zipped_rhs_size = (padded_k + 16) * n;  in gemm_q8_0_0_1_aligned()
8673   std::int32_t* zipped_lhs_3_offsets =  in gemm_q8_0_0_1_aligned()
8674       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 3);  in gemm_q8_0_0_1_aligned()
8677   const std::int32_t result_chunk_stride = result_stride * 3;  in gemm_q8_0_0_1_aligned()
8679   const std::int32_t const_offset = lhs_offset * rhs_offset * k + result_offset;  in gemm_q8_0_0_1_aligned()
8680   const std::int32_t rounding_offset = (1 << (shift - 1));  in gemm_q8_0_0_1_aligned()
8681   std::int32_t* temp_result = reinterpret_cast<std::int32_t*>(  in gemm_q8_0_0_1_aligned()
8684   std::int32_t* mul_result_chunk = temp_result;  in gemm_q8_0_0_1_aligned()
8685   const std::int32_t mul_result_chunk_stride_bytes = ((n * 4 + 7) / 8) * 8;  in gemm_q8_0_0_1_aligned()
8712                            const std::uint8_t* rhs, std::int32_t m,  in gemm_q8_0_0_2_aligned()
8713                            std::int32_t n, std::int32_t k,  in gemm_q8_0_0_2_aligned()
8714                            std::int32_t lhs_offset, std::int32_t rhs_offset,  in gemm_q8_0_0_2_aligned()
8715                            std::int32_t result_offset,  in gemm_q8_0_0_2_aligned()
8716                            std::int32_t multiplicative_offset,  in gemm_q8_0_0_2_aligned()
8717                            std::int32_t shift, std::uint8_t* result,  in gemm_q8_0_0_2_aligned()
8718                            std::int32_t result_stride) {  in gemm_q8_0_0_2_aligned()
8719   const std::int32_t row_chunks = m / 3;  in gemm_q8_0_0_2_aligned()
8720   const std::int32_t col_chunks = n / 3;  in gemm_q8_0_0_2_aligned()
8721   const std::int32_t padded_k = ((k + 7) / 8) * 8;  in gemm_q8_0_0_2_aligned()
8722   const std::int32_t chunk_size = k * 3;  in gemm_q8_0_0_2_aligned()
8723   const std::int32_t zipped_chunk_size = (padded_k + 16) * 3;  in gemm_q8_0_0_2_aligned()
8724   const std::int32_t zipped_rhs_size = (padded_k + 16) * n;  in gemm_q8_0_0_2_aligned()
8728   std::int32_t* zipped_lhs_3_offsets =  in gemm_q8_0_0_2_aligned()
8729       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 3);  in gemm_q8_0_0_2_aligned()
8732   const std::int32_t result_chunk_stride = result_stride * 3;  in gemm_q8_0_0_2_aligned()
8734   const std::int32_t const_offset = lhs_offset * rhs_offset * k + result_offset;  in gemm_q8_0_0_2_aligned()
8735   const std::int32_t rounding_offset = (1 << (shift - 1));  in gemm_q8_0_0_2_aligned()
8736   std::int32_t* temp_result = reinterpret_cast<std::int32_t*>(  in gemm_q8_0_0_2_aligned()
8739   std::int32_t* mul_result_chunk = temp_result;  in gemm_q8_0_0_2_aligned()
8740   const std::int32_t mul_result_chunk_stride_bytes = ((n * 4 + 7) / 8) * 8;  in gemm_q8_0_0_2_aligned()
8767                            const std::uint8_t* rhs, std::int32_t m,  in gemm_q8_0_0_3_aligned()
8768                            std::int32_t n, std::int32_t k,  in gemm_q8_0_0_3_aligned()
8769                            std::int32_t lhs_offset, std::int32_t rhs_offset,  in gemm_q8_0_0_3_aligned()
8770                            std::int32_t result_offset,  in gemm_q8_0_0_3_aligned()
8771                            std::int32_t multiplicative_offset,  in gemm_q8_0_0_3_aligned()
8772                            std::int32_t shift, std::uint8_t* result,  in gemm_q8_0_0_3_aligned()
8773                            std::int32_t result_stride) {  in gemm_q8_0_0_3_aligned()
8774   const std::int32_t row_chunks = m / 3;  in gemm_q8_0_0_3_aligned()
8775   const std::int32_t col_chunks = n / 3;  in gemm_q8_0_0_3_aligned()
8776   const std::int32_t padded_k = ((k + 7) / 8) * 8;  in gemm_q8_0_0_3_aligned()
8777   const std::int32_t chunk_size = k * 3;  in gemm_q8_0_0_3_aligned()
8778   const std::int32_t zipped_chunk_size = (padded_k + 16) * 3;  in gemm_q8_0_0_3_aligned()
8779   const std::int32_t zipped_rhs_size = (padded_k + 16) * n;  in gemm_q8_0_0_3_aligned()
8783   std::int32_t* zipped_lhs_3_offsets =  in gemm_q8_0_0_3_aligned()
8784       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 3);  in gemm_q8_0_0_3_aligned()
8787   const std::int32_t result_chunk_stride = result_stride * 3;  in gemm_q8_0_0_3_aligned()
8789   const std::int32_t const_offset = lhs_offset * rhs_offset * k + result_offset;  in gemm_q8_0_0_3_aligned()
8790   const std::int32_t rounding_offset = (1 << (shift - 1));  in gemm_q8_0_0_3_aligned()
8791   std::int32_t* temp_result = reinterpret_cast<std::int32_t*>(  in gemm_q8_0_0_3_aligned()
8794   std::int32_t* mul_result_chunk = temp_result;  in gemm_q8_0_0_3_aligned()
8795   const std::int32_t mul_result_chunk_stride_bytes = ((n * 4 + 7) / 8) * 8;  in gemm_q8_0_0_3_aligned()
8822                            const std::uint8_t* rhs, std::int32_t m,  in gemm_q8_0_0_4_aligned()
8823                            std::int32_t n, std::int32_t k,  in gemm_q8_0_0_4_aligned()
8824                            std::int32_t lhs_offset, std::int32_t rhs_offset,  in gemm_q8_0_0_4_aligned()
8825                            std::int32_t result_offset,  in gemm_q8_0_0_4_aligned()
8826                            std::int32_t multiplicative_offset,  in gemm_q8_0_0_4_aligned()
8827                            std::int32_t shift, std::uint8_t* result,  in gemm_q8_0_0_4_aligned()
8828                            std::int32_t result_stride) {  in gemm_q8_0_0_4_aligned()
8829   const std::int32_t row_chunks = m / 3;  in gemm_q8_0_0_4_aligned()
8830   const std::int32_t col_chunks = n / 3;  in gemm_q8_0_0_4_aligned()
8831   const std::int32_t padded_k = ((k + 7) / 8) * 8;  in gemm_q8_0_0_4_aligned()
8832   const std::int32_t chunk_size = k * 3;  in gemm_q8_0_0_4_aligned()
8833   const std::int32_t zipped_chunk_size = (padded_k + 16) * 3;  in gemm_q8_0_0_4_aligned()
8834   const std::int32_t zipped_rhs_size = (padded_k + 16) * n;  in gemm_q8_0_0_4_aligned()
8838   std::int32_t* zipped_lhs_3_offsets =  in gemm_q8_0_0_4_aligned()
8839       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 3);  in gemm_q8_0_0_4_aligned()
8842   const std::int32_t result_chunk_stride = result_stride * 3;  in gemm_q8_0_0_4_aligned()
8844   const std::int32_t const_offset = lhs_offset * rhs_offset * k + result_offset;  in gemm_q8_0_0_4_aligned()
8845   const std::int32_t rounding_offset = (1 << (shift - 1));  in gemm_q8_0_0_4_aligned()
8846   std::int32_t* temp_result = reinterpret_cast<std::int32_t*>(  in gemm_q8_0_0_4_aligned()
8849   std::int32_t* mul_result_chunk = temp_result;  in gemm_q8_0_0_4_aligned()
8850   const std::int32_t mul_result_chunk_stride_bytes = ((n * 4 + 7) / 8) * 8;  in gemm_q8_0_0_4_aligned()
8877                            const std::uint8_t* rhs, std::int32_t m,  in gemm_q8_0_0_5_aligned()
8878                            std::int32_t n, std::int32_t k,  in gemm_q8_0_0_5_aligned()
8879                            std::int32_t lhs_offset, std::int32_t rhs_offset,  in gemm_q8_0_0_5_aligned()
8880                            std::int32_t result_offset,  in gemm_q8_0_0_5_aligned()
8881                            std::int32_t multiplicative_offset,  in gemm_q8_0_0_5_aligned()
8882                            std::int32_t shift, std::uint8_t* result,  in gemm_q8_0_0_5_aligned()
8883                            std::int32_t result_stride) {  in gemm_q8_0_0_5_aligned()
8884   const std::int32_t row_chunks = m / 3;  in gemm_q8_0_0_5_aligned()
8885   const std::int32_t col_chunks = n / 3;  in gemm_q8_0_0_5_aligned()
8886   const std::int32_t padded_k = ((k + 7) / 8) * 8;  in gemm_q8_0_0_5_aligned()
8887   const std::int32_t chunk_size = k * 3;  in gemm_q8_0_0_5_aligned()
8888   const std::int32_t zipped_chunk_size = (padded_k + 16) * 3;  in gemm_q8_0_0_5_aligned()
8889   const std::int32_t zipped_rhs_size = (padded_k + 16) * n;  in gemm_q8_0_0_5_aligned()
8893   std::int32_t* zipped_lhs_3_offsets =  in gemm_q8_0_0_5_aligned()
8894       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 3);  in gemm_q8_0_0_5_aligned()
8897   const std::int32_t result_chunk_stride = result_stride * 3;  in gemm_q8_0_0_5_aligned()
8899   const std::int32_t const_offset = lhs_offset * rhs_offset * k + result_offset;  in gemm_q8_0_0_5_aligned()
8900   const std::int32_t rounding_offset = (1 << (shift - 1));  in gemm_q8_0_0_5_aligned()
8901   std::int32_t* temp_result = reinterpret_cast<std::int32_t*>(  in gemm_q8_0_0_5_aligned()
8904   std::int32_t* mul_result_chunk = temp_result;  in gemm_q8_0_0_5_aligned()
8905   const std::int32_t mul_result_chunk_stride_bytes = ((n * 4 + 7) / 8) * 8;  in gemm_q8_0_0_5_aligned()
8932                            const std::uint8_t* rhs, std::int32_t m,  in gemm_q8_0_0_6_aligned()
8933                            std::int32_t n, std::int32_t k,  in gemm_q8_0_0_6_aligned()
8934                            std::int32_t lhs_offset, std::int32_t rhs_offset,  in gemm_q8_0_0_6_aligned()
8935                            std::int32_t result_offset,  in gemm_q8_0_0_6_aligned()
8936                            std::int32_t multiplicative_offset,  in gemm_q8_0_0_6_aligned()
8937                            std::int32_t shift, std::uint8_t* result,  in gemm_q8_0_0_6_aligned()
8938                            std::int32_t result_stride) {  in gemm_q8_0_0_6_aligned()
8939   const std::int32_t row_chunks = m / 3;  in gemm_q8_0_0_6_aligned()
8940   const std::int32_t col_chunks = n / 3;  in gemm_q8_0_0_6_aligned()
8941   const std::int32_t padded_k = ((k + 7) / 8) * 8;  in gemm_q8_0_0_6_aligned()
8942   const std::int32_t chunk_size = k * 3;  in gemm_q8_0_0_6_aligned()
8943   const std::int32_t zipped_chunk_size = (padded_k + 16) * 3;  in gemm_q8_0_0_6_aligned()
8944   const std::int32_t zipped_rhs_size = (padded_k + 16) * n;  in gemm_q8_0_0_6_aligned()
8948   std::int32_t* zipped_lhs_3_offsets =  in gemm_q8_0_0_6_aligned()
8949       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 3);  in gemm_q8_0_0_6_aligned()
8952   const std::int32_t result_chunk_stride = result_stride * 3;  in gemm_q8_0_0_6_aligned()
8954   const std::int32_t const_offset = lhs_offset * rhs_offset * k + result_offset;  in gemm_q8_0_0_6_aligned()
8955   const std::int32_t rounding_offset = (1 << (shift - 1));  in gemm_q8_0_0_6_aligned()
8956   std::int32_t* temp_result = reinterpret_cast<std::int32_t*>(  in gemm_q8_0_0_6_aligned()
8959   std::int32_t* mul_result_chunk = temp_result;  in gemm_q8_0_0_6_aligned()
8960   const std::int32_t mul_result_chunk_stride_bytes = ((n * 4 + 7) / 8) * 8;  in gemm_q8_0_0_6_aligned()
8987                            const std::uint8_t* rhs, std::int32_t m,  in gemm_q8_0_0_7_aligned()
8988                            std::int32_t n, std::int32_t k,  in gemm_q8_0_0_7_aligned()
8989                            std::int32_t lhs_offset, std::int32_t rhs_offset,  in gemm_q8_0_0_7_aligned()
8990                            std::int32_t result_offset,  in gemm_q8_0_0_7_aligned()
8991                            std::int32_t multiplicative_offset,  in gemm_q8_0_0_7_aligned()
8992                            std::int32_t shift, std::uint8_t* result,  in gemm_q8_0_0_7_aligned()
8993                            std::int32_t result_stride) {  in gemm_q8_0_0_7_aligned()
8994   const std::int32_t row_chunks = m / 3;  in gemm_q8_0_0_7_aligned()
8995   const std::int32_t col_chunks = n / 3;  in gemm_q8_0_0_7_aligned()
8996   const std::int32_t padded_k = ((k + 7) / 8) * 8;  in gemm_q8_0_0_7_aligned()
8997   const std::int32_t chunk_size = k * 3;  in gemm_q8_0_0_7_aligned()
8998   const std::int32_t zipped_chunk_size = (padded_k + 16) * 3;  in gemm_q8_0_0_7_aligned()
8999   const std::int32_t zipped_rhs_size = (padded_k + 16) * n;  in gemm_q8_0_0_7_aligned()
9003   std::int32_t* zipped_lhs_3_offsets =  in gemm_q8_0_0_7_aligned()
9004       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 3);  in gemm_q8_0_0_7_aligned()
9007   const std::int32_t result_chunk_stride = result_stride * 3;  in gemm_q8_0_0_7_aligned()
9009   const std::int32_t const_offset = lhs_offset * rhs_offset * k + result_offset;  in gemm_q8_0_0_7_aligned()
9010   const std::int32_t rounding_offset = (1 << (shift - 1));  in gemm_q8_0_0_7_aligned()
9011   std::int32_t* temp_result = reinterpret_cast<std::int32_t*>(  in gemm_q8_0_0_7_aligned()
9014   std::int32_t* mul_result_chunk = temp_result;  in gemm_q8_0_0_7_aligned()
9015   const std::int32_t mul_result_chunk_stride_bytes = ((n * 4 + 7) / 8) * 8;  in gemm_q8_0_0_7_aligned()
9042                            const std::uint8_t* rhs, std::int32_t m,  in gemm_q8_0_1_0_aligned()
9043                            std::int32_t n, std::int32_t k,  in gemm_q8_0_1_0_aligned()
9044                            std::int32_t lhs_offset, std::int32_t rhs_offset,  in gemm_q8_0_1_0_aligned()
9045                            std::int32_t result_offset,  in gemm_q8_0_1_0_aligned()
9046                            std::int32_t multiplicative_offset,  in gemm_q8_0_1_0_aligned()
9047                            std::int32_t shift, std::uint8_t* result,  in gemm_q8_0_1_0_aligned()
9048                            std::int32_t result_stride) {  in gemm_q8_0_1_0_aligned()
9049   const std::int32_t row_chunks = m / 3;  in gemm_q8_0_1_0_aligned()
9050   const std::int32_t col_chunks = n / 3;  in gemm_q8_0_1_0_aligned()
9051   const std::int32_t padded_k = ((k + 7) / 8) * 8;  in gemm_q8_0_1_0_aligned()
9052   const std::int32_t chunk_size = k * 3;  in gemm_q8_0_1_0_aligned()
9053   const std::int32_t zipped_chunk_size = (padded_k + 16) * 3;  in gemm_q8_0_1_0_aligned()
9054   const std::int32_t zipped_rhs_size = (padded_k + 16) * n;  in gemm_q8_0_1_0_aligned()
9058   std::int32_t* zipped_lhs_3_offsets =  in gemm_q8_0_1_0_aligned()
9059       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 3);  in gemm_q8_0_1_0_aligned()
9062   const std::int32_t result_chunk_stride = result_stride * 3;  in gemm_q8_0_1_0_aligned()
9064   const std::int32_t const_offset = lhs_offset * rhs_offset * k + result_offset;  in gemm_q8_0_1_0_aligned()
9065   const std::int32_t rounding_offset = (1 << (shift - 1));  in gemm_q8_0_1_0_aligned()
9066   std::int32_t* temp_result = reinterpret_cast<std::int32_t*>(  in gemm_q8_0_1_0_aligned()
9069   std::int32_t* mul_result_chunk = temp_result;  in gemm_q8_0_1_0_aligned()
9070   const std::int32_t mul_result_chunk_stride_bytes = ((n * 4 + 7) / 8) * 8;  in gemm_q8_0_1_0_aligned()
9100                            const std::uint8_t* rhs, std::int32_t m,  in gemm_q8_0_1_1_aligned()
9101                            std::int32_t n, std::int32_t k,  in gemm_q8_0_1_1_aligned()
9102                            std::int32_t lhs_offset, std::int32_t rhs_offset,  in gemm_q8_0_1_1_aligned()
9103                            std::int32_t result_offset,  in gemm_q8_0_1_1_aligned()
9104                            std::int32_t multiplicative_offset,  in gemm_q8_0_1_1_aligned()
9105                            std::int32_t shift, std::uint8_t* result,  in gemm_q8_0_1_1_aligned()
9106                            std::int32_t result_stride) {  in gemm_q8_0_1_1_aligned()
9107   const std::int32_t row_chunks = m / 3;  in gemm_q8_0_1_1_aligned()
9108   const std::int32_t col_chunks = n / 3;  in gemm_q8_0_1_1_aligned()
9109   const std::int32_t padded_k = ((k + 7) / 8) * 8;  in gemm_q8_0_1_1_aligned()
9110   const std::int32_t chunk_size = k * 3;  in gemm_q8_0_1_1_aligned()
9111   const std::int32_t zipped_chunk_size = (padded_k + 16) * 3;  in gemm_q8_0_1_1_aligned()
9112   const std::int32_t zipped_rhs_size = (padded_k + 16) * n;  in gemm_q8_0_1_1_aligned()
9116   std::int32_t* zipped_lhs_3_offsets =  in gemm_q8_0_1_1_aligned()
9117       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 3);  in gemm_q8_0_1_1_aligned()
9120   const std::int32_t result_chunk_stride = result_stride * 3;  in gemm_q8_0_1_1_aligned()
9122   const std::int32_t const_offset = lhs_offset * rhs_offset * k + result_offset;  in gemm_q8_0_1_1_aligned()
9123   const std::int32_t rounding_offset = (1 << (shift - 1));  in gemm_q8_0_1_1_aligned()
9124   std::int32_t* temp_result = reinterpret_cast<std::int32_t*>(  in gemm_q8_0_1_1_aligned()
9127   std::int32_t* mul_result_chunk = temp_result;  in gemm_q8_0_1_1_aligned()
9128   const std::int32_t mul_result_chunk_stride_bytes = ((n * 4 + 7) / 8) * 8;  in gemm_q8_0_1_1_aligned()
9158                            const std::uint8_t* rhs, std::int32_t m,  in gemm_q8_0_1_2_aligned()
9159                            std::int32_t n, std::int32_t k,  in gemm_q8_0_1_2_aligned()
9160                            std::int32_t lhs_offset, std::int32_t rhs_offset,  in gemm_q8_0_1_2_aligned()
9161                            std::int32_t result_offset,  in gemm_q8_0_1_2_aligned()
9162                            std::int32_t multiplicative_offset,  in gemm_q8_0_1_2_aligned()
9163                            std::int32_t shift, std::uint8_t* result,  in gemm_q8_0_1_2_aligned()
9164                            std::int32_t result_stride) {  in gemm_q8_0_1_2_aligned()
9165   const std::int32_t row_chunks = m / 3;  in gemm_q8_0_1_2_aligned()
9166   const std::int32_t col_chunks = n / 3;  in gemm_q8_0_1_2_aligned()
9167   const std::int32_t padded_k = ((k + 7) / 8) * 8;  in gemm_q8_0_1_2_aligned()
9168   const std::int32_t chunk_size = k * 3;  in gemm_q8_0_1_2_aligned()
9169   const std::int32_t zipped_chunk_size = (padded_k + 16) * 3;  in gemm_q8_0_1_2_aligned()
9170   const std::int32_t zipped_rhs_size = (padded_k + 16) * n;  in gemm_q8_0_1_2_aligned()
9174   std::int32_t* zipped_lhs_3_offsets =  in gemm_q8_0_1_2_aligned()
9175       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 3);  in gemm_q8_0_1_2_aligned()
9178   const std::int32_t result_chunk_stride = result_stride * 3;  in gemm_q8_0_1_2_aligned()
9180   const std::int32_t const_offset = lhs_offset * rhs_offset * k + result_offset;  in gemm_q8_0_1_2_aligned()
9181   const std::int32_t rounding_offset = (1 << (shift - 1));  in gemm_q8_0_1_2_aligned()
9182   std::int32_t* temp_result = reinterpret_cast<std::int32_t*>(  in gemm_q8_0_1_2_aligned()
9185   std::int32_t* mul_result_chunk = temp_result;  in gemm_q8_0_1_2_aligned()
9186   const std::int32_t mul_result_chunk_stride_bytes = ((n * 4 + 7) / 8) * 8;  in gemm_q8_0_1_2_aligned()
9216                            const std::uint8_t* rhs, std::int32_t m,  in gemm_q8_0_1_3_aligned()
9217                            std::int32_t n, std::int32_t k,  in gemm_q8_0_1_3_aligned()
9218                            std::int32_t lhs_offset, std::int32_t rhs_offset,  in gemm_q8_0_1_3_aligned()
9219                            std::int32_t result_offset,  in gemm_q8_0_1_3_aligned()
9220                            std::int32_t multiplicative_offset,  in gemm_q8_0_1_3_aligned()
9221                            std::int32_t shift, std::uint8_t* result,  in gemm_q8_0_1_3_aligned()
9222                            std::int32_t result_stride) {  in gemm_q8_0_1_3_aligned()
9223   const std::int32_t row_chunks = m / 3;  in gemm_q8_0_1_3_aligned()
9224   const std::int32_t col_chunks = n / 3;  in gemm_q8_0_1_3_aligned()
9225   const std::int32_t padded_k = ((k + 7) / 8) * 8;  in gemm_q8_0_1_3_aligned()
9226   const std::int32_t chunk_size = k * 3;  in gemm_q8_0_1_3_aligned()
9227   const std::int32_t zipped_chunk_size = (padded_k + 16) * 3;  in gemm_q8_0_1_3_aligned()
9228   const std::int32_t zipped_rhs_size = (padded_k + 16) * n;  in gemm_q8_0_1_3_aligned()
9232   std::int32_t* zipped_lhs_3_offsets =  in gemm_q8_0_1_3_aligned()
9233       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 3);  in gemm_q8_0_1_3_aligned()
9236   const std::int32_t result_chunk_stride = result_stride * 3;  in gemm_q8_0_1_3_aligned()
9238   const std::int32_t const_offset = lhs_offset * rhs_offset * k + result_offset;  in gemm_q8_0_1_3_aligned()
9239   const std::int32_t rounding_offset = (1 << (shift - 1));  in gemm_q8_0_1_3_aligned()
9240   std::int32_t* temp_result = reinterpret_cast<std::int32_t*>(  in gemm_q8_0_1_3_aligned()
9243   std::int32_t* mul_result_chunk = temp_result;  in gemm_q8_0_1_3_aligned()
9244   const std::int32_t mul_result_chunk_stride_bytes = ((n * 4 + 7) / 8) * 8;  in gemm_q8_0_1_3_aligned()
9274                            const std::uint8_t* rhs, std::int32_t m,  in gemm_q8_0_1_4_aligned()
9275                            std::int32_t n, std::int32_t k,  in gemm_q8_0_1_4_aligned()
9276                            std::int32_t lhs_offset, std::int32_t rhs_offset,  in gemm_q8_0_1_4_aligned()
9277                            std::int32_t result_offset,  in gemm_q8_0_1_4_aligned()
9278                            std::int32_t multiplicative_offset,  in gemm_q8_0_1_4_aligned()
9279                            std::int32_t shift, std::uint8_t* result,  in gemm_q8_0_1_4_aligned()
9280                            std::int32_t result_stride) {  in gemm_q8_0_1_4_aligned()
9281   const std::int32_t row_chunks = m / 3;  in gemm_q8_0_1_4_aligned()
9282   const std::int32_t col_chunks = n / 3;  in gemm_q8_0_1_4_aligned()
9283   const std::int32_t padded_k = ((k + 7) / 8) * 8;  in gemm_q8_0_1_4_aligned()
9284   const std::int32_t chunk_size = k * 3;  in gemm_q8_0_1_4_aligned()
9285   const std::int32_t zipped_chunk_size = (padded_k + 16) * 3;  in gemm_q8_0_1_4_aligned()
9286   const std::int32_t zipped_rhs_size = (padded_k + 16) * n;  in gemm_q8_0_1_4_aligned()
9290   std::int32_t* zipped_lhs_3_offsets =  in gemm_q8_0_1_4_aligned()
9291       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 3);  in gemm_q8_0_1_4_aligned()
9294   const std::int32_t result_chunk_stride = result_stride * 3;  in gemm_q8_0_1_4_aligned()
9296   const std::int32_t const_offset = lhs_offset * rhs_offset * k + result_offset;  in gemm_q8_0_1_4_aligned()
9297   const std::int32_t rounding_offset = (1 << (shift - 1));  in gemm_q8_0_1_4_aligned()
9298   std::int32_t* temp_result = reinterpret_cast<std::int32_t*>(  in gemm_q8_0_1_4_aligned()
9301   std::int32_t* mul_result_chunk = temp_result;  in gemm_q8_0_1_4_aligned()
9302   const std::int32_t mul_result_chunk_stride_bytes = ((n * 4 + 7) / 8) * 8;  in gemm_q8_0_1_4_aligned()
9332                            const std::uint8_t* rhs, std::int32_t m,  in gemm_q8_0_1_5_aligned()
9333                            std::int32_t n, std::int32_t k,  in gemm_q8_0_1_5_aligned()
9334                            std::int32_t lhs_offset, std::int32_t rhs_offset,  in gemm_q8_0_1_5_aligned()
9335                            std::int32_t result_offset,  in gemm_q8_0_1_5_aligned()
9336                            std::int32_t multiplicative_offset,  in gemm_q8_0_1_5_aligned()
9337                            std::int32_t shift, std::uint8_t* result,  in gemm_q8_0_1_5_aligned()
9338                            std::int32_t result_stride) {  in gemm_q8_0_1_5_aligned()
9339   const std::int32_t row_chunks = m / 3;  in gemm_q8_0_1_5_aligned()
9340   const std::int32_t col_chunks = n / 3;  in gemm_q8_0_1_5_aligned()
9341   const std::int32_t padded_k = ((k + 7) / 8) * 8;  in gemm_q8_0_1_5_aligned()
9342   const std::int32_t chunk_size = k * 3;  in gemm_q8_0_1_5_aligned()
9343   const std::int32_t zipped_chunk_size = (padded_k + 16) * 3;  in gemm_q8_0_1_5_aligned()
9344   const std::int32_t zipped_rhs_size = (padded_k + 16) * n;  in gemm_q8_0_1_5_aligned()
9348   std::int32_t* zipped_lhs_3_offsets =  in gemm_q8_0_1_5_aligned()
9349       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 3);  in gemm_q8_0_1_5_aligned()
9352   const std::int32_t result_chunk_stride = result_stride * 3;  in gemm_q8_0_1_5_aligned()
9354   const std::int32_t const_offset = lhs_offset * rhs_offset * k + result_offset;  in gemm_q8_0_1_5_aligned()
9355   const std::int32_t rounding_offset = (1 << (shift - 1));  in gemm_q8_0_1_5_aligned()
9356   std::int32_t* temp_result = reinterpret_cast<std::int32_t*>(  in gemm_q8_0_1_5_aligned()
9359   std::int32_t* mul_result_chunk = temp_result;  in gemm_q8_0_1_5_aligned()
9360   const std::int32_t mul_result_chunk_stride_bytes = ((n * 4 + 7) / 8) * 8;  in gemm_q8_0_1_5_aligned()
9390                            const std::uint8_t* rhs, std::int32_t m,  in gemm_q8_0_1_6_aligned()
9391                            std::int32_t n, std::int32_t k,  in gemm_q8_0_1_6_aligned()
9392                            std::int32_t lhs_offset, std::int32_t rhs_offset,  in gemm_q8_0_1_6_aligned()
9393                            std::int32_t result_offset,  in gemm_q8_0_1_6_aligned()
9394                            std::int32_t multiplicative_offset,  in gemm_q8_0_1_6_aligned()
9395                            std::int32_t shift, std::uint8_t* result,  in gemm_q8_0_1_6_aligned()
9396                            std::int32_t result_stride) {  in gemm_q8_0_1_6_aligned()
9397   const std::int32_t row_chunks = m / 3;  in gemm_q8_0_1_6_aligned()
9398   const std::int32_t col_chunks = n / 3;  in gemm_q8_0_1_6_aligned()
9399   const std::int32_t padded_k = ((k + 7) / 8) * 8;  in gemm_q8_0_1_6_aligned()
9400   const std::int32_t chunk_size = k * 3;  in gemm_q8_0_1_6_aligned()
9401   const std::int32_t zipped_chunk_size = (padded_k + 16) * 3;  in gemm_q8_0_1_6_aligned()
9402   const std::int32_t zipped_rhs_size = (padded_k + 16) * n;  in gemm_q8_0_1_6_aligned()
9406   std::int32_t* zipped_lhs_3_offsets =  in gemm_q8_0_1_6_aligned()
9407       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 3);  in gemm_q8_0_1_6_aligned()
9410   const std::int32_t result_chunk_stride = result_stride * 3;  in gemm_q8_0_1_6_aligned()
9412   const std::int32_t const_offset = lhs_offset * rhs_offset * k + result_offset;  in gemm_q8_0_1_6_aligned()
9413   const std::int32_t rounding_offset = (1 << (shift - 1));  in gemm_q8_0_1_6_aligned()
9414   std::int32_t* temp_result = reinterpret_cast<std::int32_t*>(  in gemm_q8_0_1_6_aligned()
9417   std::int32_t* mul_result_chunk = temp_result;  in gemm_q8_0_1_6_aligned()
9418   const std::int32_t mul_result_chunk_stride_bytes = ((n * 4 + 7) / 8) * 8;  in gemm_q8_0_1_6_aligned()
9448                            const std::uint8_t* rhs, std::int32_t m,  in gemm_q8_0_1_7_aligned()
9449                            std::int32_t n, std::int32_t k,  in gemm_q8_0_1_7_aligned()
9450                            std::int32_t lhs_offset, std::int32_t rhs_offset,  in gemm_q8_0_1_7_aligned()
9451                            std::int32_t result_offset,  in gemm_q8_0_1_7_aligned()
9452                            std::int32_t multiplicative_offset,  in gemm_q8_0_1_7_aligned()
9453                            std::int32_t shift, std::uint8_t* result,  in gemm_q8_0_1_7_aligned()
9454                            std::int32_t result_stride) {  in gemm_q8_0_1_7_aligned()
9455   const std::int32_t row_chunks = m / 3;  in gemm_q8_0_1_7_aligned()
9456   const std::int32_t col_chunks = n / 3;  in gemm_q8_0_1_7_aligned()
9457   const std::int32_t padded_k = ((k + 7) / 8) * 8;  in gemm_q8_0_1_7_aligned()
9458   const std::int32_t chunk_size = k * 3;  in gemm_q8_0_1_7_aligned()
9459   const std::int32_t zipped_chunk_size = (padded_k + 16) * 3;  in gemm_q8_0_1_7_aligned()
9460   const std::int32_t zipped_rhs_size = (padded_k + 16) * n;  in gemm_q8_0_1_7_aligned()
9464   std::int32_t* zipped_lhs_3_offsets =  in gemm_q8_0_1_7_aligned()
9465       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 3);  in gemm_q8_0_1_7_aligned()
9468   const std::int32_t result_chunk_stride = result_stride * 3;  in gemm_q8_0_1_7_aligned()
9470   const std::int32_t const_offset = lhs_offset * rhs_offset * k + result_offset;  in gemm_q8_0_1_7_aligned()
9471   const std::int32_t rounding_offset = (1 << (shift - 1));  in gemm_q8_0_1_7_aligned()
9472   std::int32_t* temp_result = reinterpret_cast<std::int32_t*>(  in gemm_q8_0_1_7_aligned()
9475   std::int32_t* mul_result_chunk = temp_result;  in gemm_q8_0_1_7_aligned()
9476   const std::int32_t mul_result_chunk_stride_bytes = ((n * 4 + 7) / 8) * 8;  in gemm_q8_0_1_7_aligned()
9506                            const std::uint8_t* rhs, std::int32_t m,  in gemm_q8_0_2_0_aligned()
9507                            std::int32_t n, std::int32_t k,  in gemm_q8_0_2_0_aligned()
9508                            std::int32_t lhs_offset, std::int32_t rhs_offset,  in gemm_q8_0_2_0_aligned()
9509                            std::int32_t result_offset,  in gemm_q8_0_2_0_aligned()
9510                            std::int32_t multiplicative_offset,  in gemm_q8_0_2_0_aligned()
9511                            std::int32_t shift, std::uint8_t* result,  in gemm_q8_0_2_0_aligned()
9512                            std::int32_t result_stride) {  in gemm_q8_0_2_0_aligned()
9513   const std::int32_t row_chunks = m / 3;  in gemm_q8_0_2_0_aligned()
9514   const std::int32_t col_chunks = n / 3;  in gemm_q8_0_2_0_aligned()
9515   const std::int32_t padded_k = ((k + 7) / 8) * 8;  in gemm_q8_0_2_0_aligned()
9516   const std::int32_t chunk_size = k * 3;  in gemm_q8_0_2_0_aligned()
9517   const std::int32_t zipped_chunk_size = (padded_k + 16) * 3;  in gemm_q8_0_2_0_aligned()
9518   const std::int32_t zipped_rhs_size = (padded_k + 16) * n;  in gemm_q8_0_2_0_aligned()
9522   std::int32_t* zipped_lhs_3_offsets =  in gemm_q8_0_2_0_aligned()
9523       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 3);  in gemm_q8_0_2_0_aligned()
9526   const std::int32_t result_chunk_stride = result_stride * 3;  in gemm_q8_0_2_0_aligned()
9528   const std::int32_t const_offset = lhs_offset * rhs_offset * k + result_offset;  in gemm_q8_0_2_0_aligned()
9529   const std::int32_t rounding_offset = (1 << (shift - 1));  in gemm_q8_0_2_0_aligned()
9530   std::int32_t* temp_result = reinterpret_cast<std::int32_t*>(  in gemm_q8_0_2_0_aligned()
9533   std::int32_t* mul_result_chunk = temp_result;  in gemm_q8_0_2_0_aligned()
9534   const std::int32_t mul_result_chunk_stride_bytes = ((n * 4 + 7) / 8) * 8;  in gemm_q8_0_2_0_aligned()
9564                            const std::uint8_t* rhs, std::int32_t m,  in gemm_q8_0_2_1_aligned()
9565                            std::int32_t n, std::int32_t k,  in gemm_q8_0_2_1_aligned()
9566                            std::int32_t lhs_offset, std::int32_t rhs_offset,  in gemm_q8_0_2_1_aligned()
9567                            std::int32_t result_offset,  in gemm_q8_0_2_1_aligned()
9568                            std::int32_t multiplicative_offset,  in gemm_q8_0_2_1_aligned()
9569                            std::int32_t shift, std::uint8_t* result,  in gemm_q8_0_2_1_aligned()
9570                            std::int32_t result_stride) {  in gemm_q8_0_2_1_aligned()
9571   const std::int32_t row_chunks = m / 3;  in gemm_q8_0_2_1_aligned()
9572   const std::int32_t col_chunks = n / 3;  in gemm_q8_0_2_1_aligned()
9573   const std::int32_t padded_k = ((k + 7) / 8) * 8;  in gemm_q8_0_2_1_aligned()
9574   const std::int32_t chunk_size = k * 3;  in gemm_q8_0_2_1_aligned()
9575   const std::int32_t zipped_chunk_size = (padded_k + 16) * 3;  in gemm_q8_0_2_1_aligned()
9576   const std::int32_t zipped_rhs_size = (padded_k + 16) * n;  in gemm_q8_0_2_1_aligned()
9580   std::int32_t* zipped_lhs_3_offsets =  in gemm_q8_0_2_1_aligned()
9581       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 3);  in gemm_q8_0_2_1_aligned()
9584   const std::int32_t result_chunk_stride = result_stride * 3;  in gemm_q8_0_2_1_aligned()
9586   const std::int32_t const_offset = lhs_offset * rhs_offset * k + result_offset;  in gemm_q8_0_2_1_aligned()
9587   const std::int32_t rounding_offset = (1 << (shift - 1));  in gemm_q8_0_2_1_aligned()
9588   std::int32_t* temp_result = reinterpret_cast<std::int32_t*>(  in gemm_q8_0_2_1_aligned()
9591   std::int32_t* mul_result_chunk = temp_result;  in gemm_q8_0_2_1_aligned()
9592   const std::int32_t mul_result_chunk_stride_bytes = ((n * 4 + 7) / 8) * 8;  in gemm_q8_0_2_1_aligned()
9622                            const std::uint8_t* rhs, std::int32_t m,  in gemm_q8_0_2_2_aligned()
9623                            std::int32_t n, std::int32_t k,  in gemm_q8_0_2_2_aligned()
9624                            std::int32_t lhs_offset, std::int32_t rhs_offset,  in gemm_q8_0_2_2_aligned()
9625                            std::int32_t result_offset,  in gemm_q8_0_2_2_aligned()
9626                            std::int32_t multiplicative_offset,  in gemm_q8_0_2_2_aligned()
9627                            std::int32_t shift, std::uint8_t* result,  in gemm_q8_0_2_2_aligned()
9628                            std::int32_t result_stride) {  in gemm_q8_0_2_2_aligned()
9629   const std::int32_t row_chunks = m / 3;  in gemm_q8_0_2_2_aligned()
9630   const std::int32_t col_chunks = n / 3;  in gemm_q8_0_2_2_aligned()
9631   const std::int32_t padded_k = ((k + 7) / 8) * 8;  in gemm_q8_0_2_2_aligned()
9632   const std::int32_t chunk_size = k * 3;  in gemm_q8_0_2_2_aligned()
9633   const std::int32_t zipped_chunk_size = (padded_k + 16) * 3;  in gemm_q8_0_2_2_aligned()
9634   const std::int32_t zipped_rhs_size = (padded_k + 16) * n;  in gemm_q8_0_2_2_aligned()
9638   std::int32_t* zipped_lhs_3_offsets =  in gemm_q8_0_2_2_aligned()
9639       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 3);  in gemm_q8_0_2_2_aligned()
9642   const std::int32_t result_chunk_stride = result_stride * 3;  in gemm_q8_0_2_2_aligned()
9644   const std::int32_t const_offset = lhs_offset * rhs_offset * k + result_offset;  in gemm_q8_0_2_2_aligned()
9645   const std::int32_t rounding_offset = (1 << (shift - 1));  in gemm_q8_0_2_2_aligned()
9646   std::int32_t* temp_result = reinterpret_cast<std::int32_t*>(  in gemm_q8_0_2_2_aligned()
9649   std::int32_t* mul_result_chunk = temp_result;  in gemm_q8_0_2_2_aligned()
9650   const std::int32_t mul_result_chunk_stride_bytes = ((n * 4 + 7) / 8) * 8;  in gemm_q8_0_2_2_aligned()
9680                            const std::uint8_t* rhs, std::int32_t m,  in gemm_q8_0_2_3_aligned()
9681                            std::int32_t n, std::int32_t k,  in gemm_q8_0_2_3_aligned()
9682                            std::int32_t lhs_offset, std::int32_t rhs_offset,  in gemm_q8_0_2_3_aligned()
9683                            std::int32_t result_offset,  in gemm_q8_0_2_3_aligned()
9684                            std::int32_t multiplicative_offset,  in gemm_q8_0_2_3_aligned()
9685                            std::int32_t shift, std::uint8_t* result,  in gemm_q8_0_2_3_aligned()
9686                            std::int32_t result_stride) {  in gemm_q8_0_2_3_aligned()
9687   const std::int32_t row_chunks = m / 3;  in gemm_q8_0_2_3_aligned()
9688   const std::int32_t col_chunks = n / 3;  in gemm_q8_0_2_3_aligned()
9689   const std::int32_t padded_k = ((k + 7) / 8) * 8;  in gemm_q8_0_2_3_aligned()
9690   const std::int32_t chunk_size = k * 3;  in gemm_q8_0_2_3_aligned()
9691   const std::int32_t zipped_chunk_size = (padded_k + 16) * 3;  in gemm_q8_0_2_3_aligned()
9692   const std::int32_t zipped_rhs_size = (padded_k + 16) * n;  in gemm_q8_0_2_3_aligned()
9696   std::int32_t* zipped_lhs_3_offsets =  in gemm_q8_0_2_3_aligned()
9697       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 3);  in gemm_q8_0_2_3_aligned()
9700   const std::int32_t result_chunk_stride = result_stride * 3;  in gemm_q8_0_2_3_aligned()
9702   const std::int32_t const_offset = lhs_offset * rhs_offset * k + result_offset;  in gemm_q8_0_2_3_aligned()
9703   const std::int32_t rounding_offset = (1 << (shift - 1));  in gemm_q8_0_2_3_aligned()
9704   std::int32_t* temp_result = reinterpret_cast<std::int32_t*>(  in gemm_q8_0_2_3_aligned()
9707   std::int32_t* mul_result_chunk = temp_result;  in gemm_q8_0_2_3_aligned()
9708   const std::int32_t mul_result_chunk_stride_bytes = ((n * 4 + 7) / 8) * 8;  in gemm_q8_0_2_3_aligned()
9738                            const std::uint8_t* rhs, std::int32_t m,  in gemm_q8_0_2_4_aligned()
9739                            std::int32_t n, std::int32_t k,  in gemm_q8_0_2_4_aligned()
9740                            std::int32_t lhs_offset, std::int32_t rhs_offset,  in gemm_q8_0_2_4_aligned()
9741                            std::int32_t result_offset,  in gemm_q8_0_2_4_aligned()
9742                            std::int32_t multiplicative_offset,  in gemm_q8_0_2_4_aligned()
9743                            std::int32_t shift, std::uint8_t* result,  in gemm_q8_0_2_4_aligned()
9744                            std::int32_t result_stride) {  in gemm_q8_0_2_4_aligned()
9745   const std::int32_t row_chunks = m / 3;  in gemm_q8_0_2_4_aligned()
9746   const std::int32_t col_chunks = n / 3;  in gemm_q8_0_2_4_aligned()
9747   const std::int32_t padded_k = ((k + 7) / 8) * 8;  in gemm_q8_0_2_4_aligned()
9748   const std::int32_t chunk_size = k * 3;  in gemm_q8_0_2_4_aligned()
9749   const std::int32_t zipped_chunk_size = (padded_k + 16) * 3;  in gemm_q8_0_2_4_aligned()
9750   const std::int32_t zipped_rhs_size = (padded_k + 16) * n;  in gemm_q8_0_2_4_aligned()
9754   std::int32_t* zipped_lhs_3_offsets =  in gemm_q8_0_2_4_aligned()
9755       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 3);  in gemm_q8_0_2_4_aligned()
9758   const std::int32_t result_chunk_stride = result_stride * 3;  in gemm_q8_0_2_4_aligned()
9760   const std::int32_t const_offset = lhs_offset * rhs_offset * k + result_offset;  in gemm_q8_0_2_4_aligned()
9761   const std::int32_t rounding_offset = (1 << (shift - 1));  in gemm_q8_0_2_4_aligned()
9762   std::int32_t* temp_result = reinterpret_cast<std::int32_t*>(  in gemm_q8_0_2_4_aligned()
9765   std::int32_t* mul_result_chunk = temp_result;  in gemm_q8_0_2_4_aligned()
9766   const std::int32_t mul_result_chunk_stride_bytes = ((n * 4 + 7) / 8) * 8;  in gemm_q8_0_2_4_aligned()
9796                            const std::uint8_t* rhs, std::int32_t m,  in gemm_q8_0_2_5_aligned()
9797                            std::int32_t n, std::int32_t k,  in gemm_q8_0_2_5_aligned()
9798                            std::int32_t lhs_offset, std::int32_t rhs_offset,  in gemm_q8_0_2_5_aligned()
9799                            std::int32_t result_offset,  in gemm_q8_0_2_5_aligned()
9800                            std::int32_t multiplicative_offset,  in gemm_q8_0_2_5_aligned()
9801                            std::int32_t shift, std::uint8_t* result,  in gemm_q8_0_2_5_aligned()
9802                            std::int32_t result_stride) {  in gemm_q8_0_2_5_aligned()
9803   const std::int32_t row_chunks = m / 3;  in gemm_q8_0_2_5_aligned()
9804   const std::int32_t col_chunks = n / 3;  in gemm_q8_0_2_5_aligned()
9805   const std::int32_t padded_k = ((k + 7) / 8) * 8;  in gemm_q8_0_2_5_aligned()
9806   const std::int32_t chunk_size = k * 3;  in gemm_q8_0_2_5_aligned()
9807   const std::int32_t zipped_chunk_size = (padded_k + 16) * 3;  in gemm_q8_0_2_5_aligned()
9808   const std::int32_t zipped_rhs_size = (padded_k + 16) * n;  in gemm_q8_0_2_5_aligned()
9812   std::int32_t* zipped_lhs_3_offsets =  in gemm_q8_0_2_5_aligned()
9813       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 3);  in gemm_q8_0_2_5_aligned()
9816   const std::int32_t result_chunk_stride = result_stride * 3;  in gemm_q8_0_2_5_aligned()
9818   const std::int32_t const_offset = lhs_offset * rhs_offset * k + result_offset;  in gemm_q8_0_2_5_aligned()
9819   const std::int32_t rounding_offset = (1 << (shift - 1));  in gemm_q8_0_2_5_aligned()
9820   std::int32_t* temp_result = reinterpret_cast<std::int32_t*>(  in gemm_q8_0_2_5_aligned()
9823   std::int32_t* mul_result_chunk = temp_result;  in gemm_q8_0_2_5_aligned()
9824   const std::int32_t mul_result_chunk_stride_bytes = ((n * 4 + 7) / 8) * 8;  in gemm_q8_0_2_5_aligned()
9854                            const std::uint8_t* rhs, std::int32_t m,  in gemm_q8_0_2_6_aligned()
9855                            std::int32_t n, std::int32_t k,  in gemm_q8_0_2_6_aligned()
9856                            std::int32_t lhs_offset, std::int32_t rhs_offset,  in gemm_q8_0_2_6_aligned()
9857                            std::int32_t result_offset,  in gemm_q8_0_2_6_aligned()
9858                            std::int32_t multiplicative_offset,  in gemm_q8_0_2_6_aligned()
9859                            std::int32_t shift, std::uint8_t* result,  in gemm_q8_0_2_6_aligned()
9860                            std::int32_t result_stride) {  in gemm_q8_0_2_6_aligned()
9861   const std::int32_t row_chunks = m / 3;  in gemm_q8_0_2_6_aligned()
9862   const std::int32_t col_chunks = n / 3;  in gemm_q8_0_2_6_aligned()
9863   const std::int32_t padded_k = ((k + 7) / 8) * 8;  in gemm_q8_0_2_6_aligned()
9864   const std::int32_t chunk_size = k * 3;  in gemm_q8_0_2_6_aligned()
9865   const std::int32_t zipped_chunk_size = (padded_k + 16) * 3;  in gemm_q8_0_2_6_aligned()
9866   const std::int32_t zipped_rhs_size = (padded_k + 16) * n;  in gemm_q8_0_2_6_aligned()
9870   std::int32_t* zipped_lhs_3_offsets =  in gemm_q8_0_2_6_aligned()
9871       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 3);  in gemm_q8_0_2_6_aligned()
9874   const std::int32_t result_chunk_stride = result_stride * 3;  in gemm_q8_0_2_6_aligned()
9876   const std::int32_t const_offset = lhs_offset * rhs_offset * k + result_offset;  in gemm_q8_0_2_6_aligned()
9877   const std::int32_t rounding_offset = (1 << (shift - 1));  in gemm_q8_0_2_6_aligned()
9878   std::int32_t* temp_result = reinterpret_cast<std::int32_t*>(  in gemm_q8_0_2_6_aligned()
9881   std::int32_t* mul_result_chunk = temp_result;  in gemm_q8_0_2_6_aligned()
9882   const std::int32_t mul_result_chunk_stride_bytes = ((n * 4 + 7) / 8) * 8;  in gemm_q8_0_2_6_aligned()
9912                            const std::uint8_t* rhs, std::int32_t m,  in gemm_q8_0_2_7_aligned()
9913                            std::int32_t n, std::int32_t k,  in gemm_q8_0_2_7_aligned()
9914                            std::int32_t lhs_offset, std::int32_t rhs_offset,  in gemm_q8_0_2_7_aligned()
9915                            std::int32_t result_offset,  in gemm_q8_0_2_7_aligned()
9916                            std::int32_t multiplicative_offset,  in gemm_q8_0_2_7_aligned()
9917                            std::int32_t shift, std::uint8_t* result,  in gemm_q8_0_2_7_aligned()
9918                            std::int32_t result_stride) {  in gemm_q8_0_2_7_aligned()
9919   const std::int32_t row_chunks = m / 3;  in gemm_q8_0_2_7_aligned()
9920   const std::int32_t col_chunks = n / 3;  in gemm_q8_0_2_7_aligned()
9921   const std::int32_t padded_k = ((k + 7) / 8) * 8;  in gemm_q8_0_2_7_aligned()
9922   const std::int32_t chunk_size = k * 3;  in gemm_q8_0_2_7_aligned()
9923   const std::int32_t zipped_chunk_size = (padded_k + 16) * 3;  in gemm_q8_0_2_7_aligned()
9924   const std::int32_t zipped_rhs_size = (padded_k + 16) * n;  in gemm_q8_0_2_7_aligned()
9928   std::int32_t* zipped_lhs_3_offsets =  in gemm_q8_0_2_7_aligned()
9929       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 3);  in gemm_q8_0_2_7_aligned()
9932   const std::int32_t result_chunk_stride = result_stride * 3;  in gemm_q8_0_2_7_aligned()
9934   const std::int32_t const_offset = lhs_offset * rhs_offset * k + result_offset;  in gemm_q8_0_2_7_aligned()
9935   const std::int32_t rounding_offset = (1 << (shift - 1));  in gemm_q8_0_2_7_aligned()
9936   std::int32_t* temp_result = reinterpret_cast<std::int32_t*>(  in gemm_q8_0_2_7_aligned()
9939   std::int32_t* mul_result_chunk = temp_result;  in gemm_q8_0_2_7_aligned()
9940   const std::int32_t mul_result_chunk_stride_bytes = ((n * 4 + 7) / 8) * 8;  in gemm_q8_0_2_7_aligned()
9970                            const std::uint8_t* rhs, std::int32_t m,  in gemm_q8_1_0_0_aligned()
9971                            std::int32_t n, std::int32_t k,  in gemm_q8_1_0_0_aligned()
9972                            std::int32_t lhs_offset, std::int32_t rhs_offset,  in gemm_q8_1_0_0_aligned()
9973                            std::int32_t result_offset,  in gemm_q8_1_0_0_aligned()
9974                            std::int32_t multiplicative_offset,  in gemm_q8_1_0_0_aligned()
9975                            std::int32_t shift, std::uint8_t* result,  in gemm_q8_1_0_0_aligned()
9976                            std::int32_t result_stride) {  in gemm_q8_1_0_0_aligned()
9977   const std::int32_t row_chunks = m / 3;  in gemm_q8_1_0_0_aligned()
9978   const std::int32_t col_chunks = n / 3;  in gemm_q8_1_0_0_aligned()
9979   const std::int32_t padded_k = ((k + 7) / 8) * 8;  in gemm_q8_1_0_0_aligned()
9980   const std::int32_t chunk_size = k * 3;  in gemm_q8_1_0_0_aligned()
9981   const std::int32_t zipped_chunk_size = (padded_k + 16) * 3;  in gemm_q8_1_0_0_aligned()
9982   const std::int32_t zipped_rhs_size = (padded_k + 16) * n;  in gemm_q8_1_0_0_aligned()
9986   std::int32_t* zipped_lhs_3_offsets =  in gemm_q8_1_0_0_aligned()
9987       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 3);  in gemm_q8_1_0_0_aligned()
9988   std::int32_t* zipped_lhs_1_offsets =  in gemm_q8_1_0_0_aligned()
9989       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 1);  in gemm_q8_1_0_0_aligned()
9992   const std::int32_t result_chunk_stride = result_stride * 3;  in gemm_q8_1_0_0_aligned()
9994   const std::int32_t const_offset = lhs_offset * rhs_offset * k + result_offset;  in gemm_q8_1_0_0_aligned()
9995   const std::int32_t rounding_offset = (1 << (shift - 1));  in gemm_q8_1_0_0_aligned()
9996   std::int32_t* temp_result = reinterpret_cast<std::int32_t*>(  in gemm_q8_1_0_0_aligned()
9999   std::int32_t* mul_result_chunk = temp_result;  in gemm_q8_1_0_0_aligned()
10000   const std::int32_t mul_result_chunk_stride_bytes = ((n * 4 + 7) / 8) * 8;  in gemm_q8_1_0_0_aligned()
10040                            const std::uint8_t* rhs, std::int32_t m,  in gemm_q8_1_0_1_aligned()
10041                            std::int32_t n, std::int32_t k,  in gemm_q8_1_0_1_aligned()
10042                            std::int32_t lhs_offset, std::int32_t rhs_offset,  in gemm_q8_1_0_1_aligned()
10043                            std::int32_t result_offset,  in gemm_q8_1_0_1_aligned()
10044                            std::int32_t multiplicative_offset,  in gemm_q8_1_0_1_aligned()
10045                            std::int32_t shift, std::uint8_t* result,  in gemm_q8_1_0_1_aligned()
10046                            std::int32_t result_stride) {  in gemm_q8_1_0_1_aligned()
10047   const std::int32_t row_chunks = m / 3;  in gemm_q8_1_0_1_aligned()
10048   const std::int32_t col_chunks = n / 3;  in gemm_q8_1_0_1_aligned()
10049   const std::int32_t padded_k = ((k + 7) / 8) * 8;  in gemm_q8_1_0_1_aligned()
10050   const std::int32_t chunk_size = k * 3;  in gemm_q8_1_0_1_aligned()
10051   const std::int32_t zipped_chunk_size = (padded_k + 16) * 3;  in gemm_q8_1_0_1_aligned()
10052   const std::int32_t zipped_rhs_size = (padded_k + 16) * n;  in gemm_q8_1_0_1_aligned()
10056   std::int32_t* zipped_lhs_3_offsets =  in gemm_q8_1_0_1_aligned()
10057       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 3);  in gemm_q8_1_0_1_aligned()
10058   std::int32_t* zipped_lhs_1_offsets =  in gemm_q8_1_0_1_aligned()
10059       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 1);  in gemm_q8_1_0_1_aligned()
10062   const std::int32_t result_chunk_stride = result_stride * 3;  in gemm_q8_1_0_1_aligned()
10064   const std::int32_t const_offset = lhs_offset * rhs_offset * k + result_offset;  in gemm_q8_1_0_1_aligned()
10065   const std::int32_t rounding_offset = (1 << (shift - 1));  in gemm_q8_1_0_1_aligned()
10066   std::int32_t* temp_result = reinterpret_cast<std::int32_t*>(  in gemm_q8_1_0_1_aligned()
10069   std::int32_t* mul_result_chunk = temp_result;  in gemm_q8_1_0_1_aligned()
10070   const std::int32_t mul_result_chunk_stride_bytes = ((n * 4 + 7) / 8) * 8;  in gemm_q8_1_0_1_aligned()
10110                            const std::uint8_t* rhs, std::int32_t m,  in gemm_q8_1_0_2_aligned()
10111                            std::int32_t n, std::int32_t k,  in gemm_q8_1_0_2_aligned()
10112                            std::int32_t lhs_offset, std::int32_t rhs_offset,  in gemm_q8_1_0_2_aligned()
10113                            std::int32_t result_offset,  in gemm_q8_1_0_2_aligned()
10114                            std::int32_t multiplicative_offset,  in gemm_q8_1_0_2_aligned()
10115                            std::int32_t shift, std::uint8_t* result,  in gemm_q8_1_0_2_aligned()
10116                            std::int32_t result_stride) {  in gemm_q8_1_0_2_aligned()
10117   const std::int32_t row_chunks = m / 3;  in gemm_q8_1_0_2_aligned()
10118   const std::int32_t col_chunks = n / 3;  in gemm_q8_1_0_2_aligned()
10119   const std::int32_t padded_k = ((k + 7) / 8) * 8;  in gemm_q8_1_0_2_aligned()
10120   const std::int32_t chunk_size = k * 3;  in gemm_q8_1_0_2_aligned()
10121   const std::int32_t zipped_chunk_size = (padded_k + 16) * 3;  in gemm_q8_1_0_2_aligned()
10122   const std::int32_t zipped_rhs_size = (padded_k + 16) * n;  in gemm_q8_1_0_2_aligned()
10126   std::int32_t* zipped_lhs_3_offsets =  in gemm_q8_1_0_2_aligned()
10127       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 3);  in gemm_q8_1_0_2_aligned()
10128   std::int32_t* zipped_lhs_1_offsets =  in gemm_q8_1_0_2_aligned()
10129       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 1);  in gemm_q8_1_0_2_aligned()
10132   const std::int32_t result_chunk_stride = result_stride * 3;  in gemm_q8_1_0_2_aligned()
10134   const std::int32_t const_offset = lhs_offset * rhs_offset * k + result_offset;  in gemm_q8_1_0_2_aligned()
10135   const std::int32_t rounding_offset = (1 << (shift - 1));  in gemm_q8_1_0_2_aligned()
10136   std::int32_t* temp_result = reinterpret_cast<std::int32_t*>(  in gemm_q8_1_0_2_aligned()
10139   std::int32_t* mul_result_chunk = temp_result;  in gemm_q8_1_0_2_aligned()
10140   const std::int32_t mul_result_chunk_stride_bytes = ((n * 4 + 7) / 8) * 8;  in gemm_q8_1_0_2_aligned()
10180                            const std::uint8_t* rhs, std::int32_t m,  in gemm_q8_1_0_3_aligned()
10181                            std::int32_t n, std::int32_t k,  in gemm_q8_1_0_3_aligned()
10182                            std::int32_t lhs_offset, std::int32_t rhs_offset,  in gemm_q8_1_0_3_aligned()
10183                            std::int32_t result_offset,  in gemm_q8_1_0_3_aligned()
10184                            std::int32_t multiplicative_offset,  in gemm_q8_1_0_3_aligned()
10185                            std::int32_t shift, std::uint8_t* result,  in gemm_q8_1_0_3_aligned()
10186                            std::int32_t result_stride) {  in gemm_q8_1_0_3_aligned()
10187   const std::int32_t row_chunks = m / 3;  in gemm_q8_1_0_3_aligned()
10188   const std::int32_t col_chunks = n / 3;  in gemm_q8_1_0_3_aligned()
10189   const std::int32_t padded_k = ((k + 7) / 8) * 8;  in gemm_q8_1_0_3_aligned()
10190   const std::int32_t chunk_size = k * 3;  in gemm_q8_1_0_3_aligned()
10191   const std::int32_t zipped_chunk_size = (padded_k + 16) * 3;  in gemm_q8_1_0_3_aligned()
10192   const std::int32_t zipped_rhs_size = (padded_k + 16) * n;  in gemm_q8_1_0_3_aligned()
10196   std::int32_t* zipped_lhs_3_offsets =  in gemm_q8_1_0_3_aligned()
10197       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 3);  in gemm_q8_1_0_3_aligned()
10198   std::int32_t* zipped_lhs_1_offsets =  in gemm_q8_1_0_3_aligned()
10199       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 1);  in gemm_q8_1_0_3_aligned()
10202   const std::int32_t result_chunk_stride = result_stride * 3;  in gemm_q8_1_0_3_aligned()
10204   const std::int32_t const_offset = lhs_offset * rhs_offset * k + result_offset;  in gemm_q8_1_0_3_aligned()
10205   const std::int32_t rounding_offset = (1 << (shift - 1));  in gemm_q8_1_0_3_aligned()
10206   std::int32_t* temp_result = reinterpret_cast<std::int32_t*>(  in gemm_q8_1_0_3_aligned()
10209   std::int32_t* mul_result_chunk = temp_result;  in gemm_q8_1_0_3_aligned()
10210   const std::int32_t mul_result_chunk_stride_bytes = ((n * 4 + 7) / 8) * 8;  in gemm_q8_1_0_3_aligned()
10250                            const std::uint8_t* rhs, std::int32_t m,  in gemm_q8_1_0_4_aligned()
10251                            std::int32_t n, std::int32_t k,  in gemm_q8_1_0_4_aligned()
10252                            std::int32_t lhs_offset, std::int32_t rhs_offset,  in gemm_q8_1_0_4_aligned()
10253                            std::int32_t result_offset,  in gemm_q8_1_0_4_aligned()
10254                            std::int32_t multiplicative_offset,  in gemm_q8_1_0_4_aligned()
10255                            std::int32_t shift, std::uint8_t* result,  in gemm_q8_1_0_4_aligned()
10256                            std::int32_t result_stride) {  in gemm_q8_1_0_4_aligned()
10257   const std::int32_t row_chunks = m / 3;  in gemm_q8_1_0_4_aligned()
10258   const std::int32_t col_chunks = n / 3;  in gemm_q8_1_0_4_aligned()
10259   const std::int32_t padded_k = ((k + 7) / 8) * 8;  in gemm_q8_1_0_4_aligned()
10260   const std::int32_t chunk_size = k * 3;  in gemm_q8_1_0_4_aligned()
10261   const std::int32_t zipped_chunk_size = (padded_k + 16) * 3;  in gemm_q8_1_0_4_aligned()
10262   const std::int32_t zipped_rhs_size = (padded_k + 16) * n;  in gemm_q8_1_0_4_aligned()
10266   std::int32_t* zipped_lhs_3_offsets =  in gemm_q8_1_0_4_aligned()
10267       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 3);  in gemm_q8_1_0_4_aligned()
10268   std::int32_t* zipped_lhs_1_offsets =  in gemm_q8_1_0_4_aligned()
10269       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 1);  in gemm_q8_1_0_4_aligned()
10272   const std::int32_t result_chunk_stride = result_stride * 3;  in gemm_q8_1_0_4_aligned()
10274   const std::int32_t const_offset = lhs_offset * rhs_offset * k + result_offset;  in gemm_q8_1_0_4_aligned()
10275   const std::int32_t rounding_offset = (1 << (shift - 1));  in gemm_q8_1_0_4_aligned()
10276   std::int32_t* temp_result = reinterpret_cast<std::int32_t*>(  in gemm_q8_1_0_4_aligned()
10279   std::int32_t* mul_result_chunk = temp_result;  in gemm_q8_1_0_4_aligned()
10280   const std::int32_t mul_result_chunk_stride_bytes = ((n * 4 + 7) / 8) * 8;  in gemm_q8_1_0_4_aligned()
10320                            const std::uint8_t* rhs, std::int32_t m,  in gemm_q8_1_0_5_aligned()
10321                            std::int32_t n, std::int32_t k,  in gemm_q8_1_0_5_aligned()
10322                            std::int32_t lhs_offset, std::int32_t rhs_offset,  in gemm_q8_1_0_5_aligned()
10323                            std::int32_t result_offset,  in gemm_q8_1_0_5_aligned()
10324                            std::int32_t multiplicative_offset,  in gemm_q8_1_0_5_aligned()
10325                            std::int32_t shift, std::uint8_t* result,  in gemm_q8_1_0_5_aligned()
10326                            std::int32_t result_stride) {  in gemm_q8_1_0_5_aligned()
10327   const std::int32_t row_chunks = m / 3;  in gemm_q8_1_0_5_aligned()
10328   const std::int32_t col_chunks = n / 3;  in gemm_q8_1_0_5_aligned()
10329   const std::int32_t padded_k = ((k + 7) / 8) * 8;  in gemm_q8_1_0_5_aligned()
10330   const std::int32_t chunk_size = k * 3;  in gemm_q8_1_0_5_aligned()
10331   const std::int32_t zipped_chunk_size = (padded_k + 16) * 3;  in gemm_q8_1_0_5_aligned()
10332   const std::int32_t zipped_rhs_size = (padded_k + 16) * n;  in gemm_q8_1_0_5_aligned()
10336   std::int32_t* zipped_lhs_3_offsets =  in gemm_q8_1_0_5_aligned()
10337       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 3);  in gemm_q8_1_0_5_aligned()
10338   std::int32_t* zipped_lhs_1_offsets =  in gemm_q8_1_0_5_aligned()
10339       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 1);  in gemm_q8_1_0_5_aligned()
10342   const std::int32_t result_chunk_stride = result_stride * 3;  in gemm_q8_1_0_5_aligned()
10344   const std::int32_t const_offset = lhs_offset * rhs_offset * k + result_offset;  in gemm_q8_1_0_5_aligned()
10345   const std::int32_t rounding_offset = (1 << (shift - 1));  in gemm_q8_1_0_5_aligned()
10346   std::int32_t* temp_result = reinterpret_cast<std::int32_t*>(  in gemm_q8_1_0_5_aligned()
10349   std::int32_t* mul_result_chunk = temp_result;  in gemm_q8_1_0_5_aligned()
10350   const std::int32_t mul_result_chunk_stride_bytes = ((n * 4 + 7) / 8) * 8;  in gemm_q8_1_0_5_aligned()
10390                            const std::uint8_t* rhs, std::int32_t m,  in gemm_q8_1_0_6_aligned()
10391                            std::int32_t n, std::int32_t k,  in gemm_q8_1_0_6_aligned()
10392                            std::int32_t lhs_offset, std::int32_t rhs_offset,  in gemm_q8_1_0_6_aligned()
10393                            std::int32_t result_offset,  in gemm_q8_1_0_6_aligned()
10394                            std::int32_t multiplicative_offset,  in gemm_q8_1_0_6_aligned()
10395                            std::int32_t shift, std::uint8_t* result,  in gemm_q8_1_0_6_aligned()
10396                            std::int32_t result_stride) {  in gemm_q8_1_0_6_aligned()
10397   const std::int32_t row_chunks = m / 3;  in gemm_q8_1_0_6_aligned()
10398   const std::int32_t col_chunks = n / 3;  in gemm_q8_1_0_6_aligned()
10399   const std::int32_t padded_k = ((k + 7) / 8) * 8;  in gemm_q8_1_0_6_aligned()
10400   const std::int32_t chunk_size = k * 3;  in gemm_q8_1_0_6_aligned()
10401   const std::int32_t zipped_chunk_size = (padded_k + 16) * 3;  in gemm_q8_1_0_6_aligned()
10402   const std::int32_t zipped_rhs_size = (padded_k + 16) * n;  in gemm_q8_1_0_6_aligned()
10406   std::int32_t* zipped_lhs_3_offsets =  in gemm_q8_1_0_6_aligned()
10407       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 3);  in gemm_q8_1_0_6_aligned()
10408   std::int32_t* zipped_lhs_1_offsets =  in gemm_q8_1_0_6_aligned()
10409       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 1);  in gemm_q8_1_0_6_aligned()
10412   const std::int32_t result_chunk_stride = result_stride * 3;  in gemm_q8_1_0_6_aligned()
10414   const std::int32_t const_offset = lhs_offset * rhs_offset * k + result_offset;  in gemm_q8_1_0_6_aligned()
10415   const std::int32_t rounding_offset = (1 << (shift - 1));  in gemm_q8_1_0_6_aligned()
10416   std::int32_t* temp_result = reinterpret_cast<std::int32_t*>(  in gemm_q8_1_0_6_aligned()
10419   std::int32_t* mul_result_chunk = temp_result;  in gemm_q8_1_0_6_aligned()
10420   const std::int32_t mul_result_chunk_stride_bytes = ((n * 4 + 7) / 8) * 8;  in gemm_q8_1_0_6_aligned()
10460                            const std::uint8_t* rhs, std::int32_t m,  in gemm_q8_1_0_7_aligned()
10461                            std::int32_t n, std::int32_t k,  in gemm_q8_1_0_7_aligned()
10462                            std::int32_t lhs_offset, std::int32_t rhs_offset,  in gemm_q8_1_0_7_aligned()
10463                            std::int32_t result_offset,  in gemm_q8_1_0_7_aligned()
10464                            std::int32_t multiplicative_offset,  in gemm_q8_1_0_7_aligned()
10465                            std::int32_t shift, std::uint8_t* result,  in gemm_q8_1_0_7_aligned()
10466                            std::int32_t result_stride) {  in gemm_q8_1_0_7_aligned()
10467   const std::int32_t row_chunks = m / 3;  in gemm_q8_1_0_7_aligned()
10468   const std::int32_t col_chunks = n / 3;  in gemm_q8_1_0_7_aligned()
10469   const std::int32_t padded_k = ((k + 7) / 8) * 8;  in gemm_q8_1_0_7_aligned()
10470   const std::int32_t chunk_size = k * 3;  in gemm_q8_1_0_7_aligned()
10471   const std::int32_t zipped_chunk_size = (padded_k + 16) * 3;  in gemm_q8_1_0_7_aligned()
10472   const std::int32_t zipped_rhs_size = (padded_k + 16) * n;  in gemm_q8_1_0_7_aligned()
10476   std::int32_t* zipped_lhs_3_offsets =  in gemm_q8_1_0_7_aligned()
10477       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 3);  in gemm_q8_1_0_7_aligned()
10478   std::int32_t* zipped_lhs_1_offsets =  in gemm_q8_1_0_7_aligned()
10479       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 1);  in gemm_q8_1_0_7_aligned()
10482   const std::int32_t result_chunk_stride = result_stride * 3;  in gemm_q8_1_0_7_aligned()
10484   const std::int32_t const_offset = lhs_offset * rhs_offset * k + result_offset;  in gemm_q8_1_0_7_aligned()
10485   const std::int32_t rounding_offset = (1 << (shift - 1));  in gemm_q8_1_0_7_aligned()
10486   std::int32_t* temp_result = reinterpret_cast<std::int32_t*>(  in gemm_q8_1_0_7_aligned()
10489   std::int32_t* mul_result_chunk = temp_result;  in gemm_q8_1_0_7_aligned()
10490   const std::int32_t mul_result_chunk_stride_bytes = ((n * 4 + 7) / 8) * 8;  in gemm_q8_1_0_7_aligned()
10530                            const std::uint8_t* rhs, std::int32_t m,  in gemm_q8_1_1_0_aligned()
10531                            std::int32_t n, std::int32_t k,  in gemm_q8_1_1_0_aligned()
10532                            std::int32_t lhs_offset, std::int32_t rhs_offset,  in gemm_q8_1_1_0_aligned()
10533                            std::int32_t result_offset,  in gemm_q8_1_1_0_aligned()
10534                            std::int32_t multiplicative_offset,  in gemm_q8_1_1_0_aligned()
10535                            std::int32_t shift, std::uint8_t* result,  in gemm_q8_1_1_0_aligned()
10536                            std::int32_t result_stride) {  in gemm_q8_1_1_0_aligned()
10537   const std::int32_t row_chunks = m / 3;  in gemm_q8_1_1_0_aligned()
10538   const std::int32_t col_chunks = n / 3;  in gemm_q8_1_1_0_aligned()
10539   const std::int32_t padded_k = ((k + 7) / 8) * 8;  in gemm_q8_1_1_0_aligned()
10540   const std::int32_t chunk_size = k * 3;  in gemm_q8_1_1_0_aligned()
10541   const std::int32_t zipped_chunk_size = (padded_k + 16) * 3;  in gemm_q8_1_1_0_aligned()
10542   const std::int32_t zipped_rhs_size = (padded_k + 16) * n;  in gemm_q8_1_1_0_aligned()
10546   std::int32_t* zipped_lhs_3_offsets =  in gemm_q8_1_1_0_aligned()
10547       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 3);  in gemm_q8_1_1_0_aligned()
10548   std::int32_t* zipped_lhs_1_offsets =  in gemm_q8_1_1_0_aligned()
10549       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 1);  in gemm_q8_1_1_0_aligned()
10552   const std::int32_t result_chunk_stride = result_stride * 3;  in gemm_q8_1_1_0_aligned()
10554   const std::int32_t const_offset = lhs_offset * rhs_offset * k + result_offset;  in gemm_q8_1_1_0_aligned()
10555   const std::int32_t rounding_offset = (1 << (shift - 1));  in gemm_q8_1_1_0_aligned()
10556   std::int32_t* temp_result = reinterpret_cast<std::int32_t*>(  in gemm_q8_1_1_0_aligned()
10559   std::int32_t* mul_result_chunk = temp_result;  in gemm_q8_1_1_0_aligned()
10560   const std::int32_t mul_result_chunk_stride_bytes = ((n * 4 + 7) / 8) * 8;  in gemm_q8_1_1_0_aligned()
10605                            const std::uint8_t* rhs, std::int32_t m,  in gemm_q8_1_1_1_aligned()
10606                            std::int32_t n, std::int32_t k,  in gemm_q8_1_1_1_aligned()
10607                            std::int32_t lhs_offset, std::int32_t rhs_offset,  in gemm_q8_1_1_1_aligned()
10608                            std::int32_t result_offset,  in gemm_q8_1_1_1_aligned()
10609                            std::int32_t multiplicative_offset,  in gemm_q8_1_1_1_aligned()
10610                            std::int32_t shift, std::uint8_t* result,  in gemm_q8_1_1_1_aligned()
10611                            std::int32_t result_stride) {  in gemm_q8_1_1_1_aligned()
10612   const std::int32_t row_chunks = m / 3;  in gemm_q8_1_1_1_aligned()
10613   const std::int32_t col_chunks = n / 3;  in gemm_q8_1_1_1_aligned()
10614   const std::int32_t padded_k = ((k + 7) / 8) * 8;  in gemm_q8_1_1_1_aligned()
10615   const std::int32_t chunk_size = k * 3;  in gemm_q8_1_1_1_aligned()
10616   const std::int32_t zipped_chunk_size = (padded_k + 16) * 3;  in gemm_q8_1_1_1_aligned()
10617   const std::int32_t zipped_rhs_size = (padded_k + 16) * n;  in gemm_q8_1_1_1_aligned()
10621   std::int32_t* zipped_lhs_3_offsets =  in gemm_q8_1_1_1_aligned()
10622       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 3);  in gemm_q8_1_1_1_aligned()
10623   std::int32_t* zipped_lhs_1_offsets =  in gemm_q8_1_1_1_aligned()
10624       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 1);  in gemm_q8_1_1_1_aligned()
10627   const std::int32_t result_chunk_stride = result_stride * 3;  in gemm_q8_1_1_1_aligned()
10629   const std::int32_t const_offset = lhs_offset * rhs_offset * k + result_offset;  in gemm_q8_1_1_1_aligned()
10630   const std::int32_t rounding_offset = (1 << (shift - 1));  in gemm_q8_1_1_1_aligned()
10631   std::int32_t* temp_result = reinterpret_cast<std::int32_t*>(  in gemm_q8_1_1_1_aligned()
10634   std::int32_t* mul_result_chunk = temp_result;  in gemm_q8_1_1_1_aligned()
10635   const std::int32_t mul_result_chunk_stride_bytes = ((n * 4 + 7) / 8) * 8;  in gemm_q8_1_1_1_aligned()
10680                            const std::uint8_t* rhs, std::int32_t m,  in gemm_q8_1_1_2_aligned()
10681                            std::int32_t n, std::int32_t k,  in gemm_q8_1_1_2_aligned()
10682                            std::int32_t lhs_offset, std::int32_t rhs_offset,  in gemm_q8_1_1_2_aligned()
10683                            std::int32_t result_offset,  in gemm_q8_1_1_2_aligned()
10684                            std::int32_t multiplicative_offset,  in gemm_q8_1_1_2_aligned()
10685                            std::int32_t shift, std::uint8_t* result,  in gemm_q8_1_1_2_aligned()
10686                            std::int32_t result_stride) {  in gemm_q8_1_1_2_aligned()
10687   const std::int32_t row_chunks = m / 3;  in gemm_q8_1_1_2_aligned()
10688   const std::int32_t col_chunks = n / 3;  in gemm_q8_1_1_2_aligned()
10689   const std::int32_t padded_k = ((k + 7) / 8) * 8;  in gemm_q8_1_1_2_aligned()
10690   const std::int32_t chunk_size = k * 3;  in gemm_q8_1_1_2_aligned()
10691   const std::int32_t zipped_chunk_size = (padded_k + 16) * 3;  in gemm_q8_1_1_2_aligned()
10692   const std::int32_t zipped_rhs_size = (padded_k + 16) * n;  in gemm_q8_1_1_2_aligned()
10696   std::int32_t* zipped_lhs_3_offsets =  in gemm_q8_1_1_2_aligned()
10697       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 3);  in gemm_q8_1_1_2_aligned()
10698   std::int32_t* zipped_lhs_1_offsets =  in gemm_q8_1_1_2_aligned()
10699       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 1);  in gemm_q8_1_1_2_aligned()
10702   const std::int32_t result_chunk_stride = result_stride * 3;  in gemm_q8_1_1_2_aligned()
10704   const std::int32_t const_offset = lhs_offset * rhs_offset * k + result_offset;  in gemm_q8_1_1_2_aligned()
10705   const std::int32_t rounding_offset = (1 << (shift - 1));  in gemm_q8_1_1_2_aligned()
10706   std::int32_t* temp_result = reinterpret_cast<std::int32_t*>(  in gemm_q8_1_1_2_aligned()
10709   std::int32_t* mul_result_chunk = temp_result;  in gemm_q8_1_1_2_aligned()
10710   const std::int32_t mul_result_chunk_stride_bytes = ((n * 4 + 7) / 8) * 8;  in gemm_q8_1_1_2_aligned()
10755                            const std::uint8_t* rhs, std::int32_t m,  in gemm_q8_1_1_3_aligned()
10756                            std::int32_t n, std::int32_t k,  in gemm_q8_1_1_3_aligned()
10757                            std::int32_t lhs_offset, std::int32_t rhs_offset,  in gemm_q8_1_1_3_aligned()
10758                            std::int32_t result_offset,  in gemm_q8_1_1_3_aligned()
10759                            std::int32_t multiplicative_offset,  in gemm_q8_1_1_3_aligned()
10760                            std::int32_t shift, std::uint8_t* result,  in gemm_q8_1_1_3_aligned()
10761                            std::int32_t result_stride) {  in gemm_q8_1_1_3_aligned()
10762   const std::int32_t row_chunks = m / 3;  in gemm_q8_1_1_3_aligned()
10763   const std::int32_t col_chunks = n / 3;  in gemm_q8_1_1_3_aligned()
10764   const std::int32_t padded_k = ((k + 7) / 8) * 8;  in gemm_q8_1_1_3_aligned()
10765   const std::int32_t chunk_size = k * 3;  in gemm_q8_1_1_3_aligned()
10766   const std::int32_t zipped_chunk_size = (padded_k + 16) * 3;  in gemm_q8_1_1_3_aligned()
10767   const std::int32_t zipped_rhs_size = (padded_k + 16) * n;  in gemm_q8_1_1_3_aligned()
10771   std::int32_t* zipped_lhs_3_offsets =  in gemm_q8_1_1_3_aligned()
10772       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 3);  in gemm_q8_1_1_3_aligned()
10773   std::int32_t* zipped_lhs_1_offsets =  in gemm_q8_1_1_3_aligned()
10774       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 1);  in gemm_q8_1_1_3_aligned()
10777   const std::int32_t result_chunk_stride = result_stride * 3;  in gemm_q8_1_1_3_aligned()
10779   const std::int32_t const_offset = lhs_offset * rhs_offset * k + result_offset;  in gemm_q8_1_1_3_aligned()
10780   const std::int32_t rounding_offset = (1 << (shift - 1));  in gemm_q8_1_1_3_aligned()
10781   std::int32_t* temp_result = reinterpret_cast<std::int32_t*>(  in gemm_q8_1_1_3_aligned()
10784   std::int32_t* mul_result_chunk = temp_result;  in gemm_q8_1_1_3_aligned()
10785   const std::int32_t mul_result_chunk_stride_bytes = ((n * 4 + 7) / 8) * 8;  in gemm_q8_1_1_3_aligned()
10830                            const std::uint8_t* rhs, std::int32_t m,  in gemm_q8_1_1_4_aligned()
10831                            std::int32_t n, std::int32_t k,  in gemm_q8_1_1_4_aligned()
10832                            std::int32_t lhs_offset, std::int32_t rhs_offset,  in gemm_q8_1_1_4_aligned()
10833                            std::int32_t result_offset,  in gemm_q8_1_1_4_aligned()
10834                            std::int32_t multiplicative_offset,  in gemm_q8_1_1_4_aligned()
10835                            std::int32_t shift, std::uint8_t* result,  in gemm_q8_1_1_4_aligned()
10836                            std::int32_t result_stride) {  in gemm_q8_1_1_4_aligned()
10837   const std::int32_t row_chunks = m / 3;  in gemm_q8_1_1_4_aligned()
10838   const std::int32_t col_chunks = n / 3;  in gemm_q8_1_1_4_aligned()
10839   const std::int32_t padded_k = ((k + 7) / 8) * 8;  in gemm_q8_1_1_4_aligned()
10840   const std::int32_t chunk_size = k * 3;  in gemm_q8_1_1_4_aligned()
10841   const std::int32_t zipped_chunk_size = (padded_k + 16) * 3;  in gemm_q8_1_1_4_aligned()
10842   const std::int32_t zipped_rhs_size = (padded_k + 16) * n;  in gemm_q8_1_1_4_aligned()
10846   std::int32_t* zipped_lhs_3_offsets =  in gemm_q8_1_1_4_aligned()
10847       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 3);  in gemm_q8_1_1_4_aligned()
10848   std::int32_t* zipped_lhs_1_offsets =  in gemm_q8_1_1_4_aligned()
10849       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 1);  in gemm_q8_1_1_4_aligned()
10852   const std::int32_t result_chunk_stride = result_stride * 3;  in gemm_q8_1_1_4_aligned()
10854   const std::int32_t const_offset = lhs_offset * rhs_offset * k + result_offset;  in gemm_q8_1_1_4_aligned()
10855   const std::int32_t rounding_offset = (1 << (shift - 1));  in gemm_q8_1_1_4_aligned()
10856   std::int32_t* temp_result = reinterpret_cast<std::int32_t*>(  in gemm_q8_1_1_4_aligned()
10859   std::int32_t* mul_result_chunk = temp_result;  in gemm_q8_1_1_4_aligned()
10860   const std::int32_t mul_result_chunk_stride_bytes = ((n * 4 + 7) / 8) * 8;  in gemm_q8_1_1_4_aligned()
10905                            const std::uint8_t* rhs, std::int32_t m,  in gemm_q8_1_1_5_aligned()
10906                            std::int32_t n, std::int32_t k,  in gemm_q8_1_1_5_aligned()
10907                            std::int32_t lhs_offset, std::int32_t rhs_offset,  in gemm_q8_1_1_5_aligned()
10908                            std::int32_t result_offset,  in gemm_q8_1_1_5_aligned()
10909                            std::int32_t multiplicative_offset,  in gemm_q8_1_1_5_aligned()
10910                            std::int32_t shift, std::uint8_t* result,  in gemm_q8_1_1_5_aligned()
10911                            std::int32_t result_stride) {  in gemm_q8_1_1_5_aligned()
10912   const std::int32_t row_chunks = m / 3;  in gemm_q8_1_1_5_aligned()
10913   const std::int32_t col_chunks = n / 3;  in gemm_q8_1_1_5_aligned()
10914   const std::int32_t padded_k = ((k + 7) / 8) * 8;  in gemm_q8_1_1_5_aligned()
10915   const std::int32_t chunk_size = k * 3;  in gemm_q8_1_1_5_aligned()
10916   const std::int32_t zipped_chunk_size = (padded_k + 16) * 3;  in gemm_q8_1_1_5_aligned()
10917   const std::int32_t zipped_rhs_size = (padded_k + 16) * n;  in gemm_q8_1_1_5_aligned()
10921   std::int32_t* zipped_lhs_3_offsets =  in gemm_q8_1_1_5_aligned()
10922       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 3);  in gemm_q8_1_1_5_aligned()
10923   std::int32_t* zipped_lhs_1_offsets =  in gemm_q8_1_1_5_aligned()
10924       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 1);  in gemm_q8_1_1_5_aligned()
10927   const std::int32_t result_chunk_stride = result_stride * 3;  in gemm_q8_1_1_5_aligned()
10929   const std::int32_t const_offset = lhs_offset * rhs_offset * k + result_offset;  in gemm_q8_1_1_5_aligned()
10930   const std::int32_t rounding_offset = (1 << (shift - 1));  in gemm_q8_1_1_5_aligned()
10931   std::int32_t* temp_result = reinterpret_cast<std::int32_t*>(  in gemm_q8_1_1_5_aligned()
10934   std::int32_t* mul_result_chunk = temp_result;  in gemm_q8_1_1_5_aligned()
10935   const std::int32_t mul_result_chunk_stride_bytes = ((n * 4 + 7) / 8) * 8;  in gemm_q8_1_1_5_aligned()
10980                            const std::uint8_t* rhs, std::int32_t m,  in gemm_q8_1_1_6_aligned()
10981                            std::int32_t n, std::int32_t k,  in gemm_q8_1_1_6_aligned()
10982                            std::int32_t lhs_offset, std::int32_t rhs_offset,  in gemm_q8_1_1_6_aligned()
10983                            std::int32_t result_offset,  in gemm_q8_1_1_6_aligned()
10984                            std::int32_t multiplicative_offset,  in gemm_q8_1_1_6_aligned()
10985                            std::int32_t shift, std::uint8_t* result,  in gemm_q8_1_1_6_aligned()
10986                            std::int32_t result_stride) {  in gemm_q8_1_1_6_aligned()
10987   const std::int32_t row_chunks = m / 3;  in gemm_q8_1_1_6_aligned()
10988   const std::int32_t col_chunks = n / 3;  in gemm_q8_1_1_6_aligned()
10989   const std::int32_t padded_k = ((k + 7) / 8) * 8;  in gemm_q8_1_1_6_aligned()
10990   const std::int32_t chunk_size = k * 3;  in gemm_q8_1_1_6_aligned()
10991   const std::int32_t zipped_chunk_size = (padded_k + 16) * 3;  in gemm_q8_1_1_6_aligned()
10992   const std::int32_t zipped_rhs_size = (padded_k + 16) * n;  in gemm_q8_1_1_6_aligned()
10996   std::int32_t* zipped_lhs_3_offsets =  in gemm_q8_1_1_6_aligned()
10997       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 3);  in gemm_q8_1_1_6_aligned()
10998   std::int32_t* zipped_lhs_1_offsets =  in gemm_q8_1_1_6_aligned()
10999       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 1);  in gemm_q8_1_1_6_aligned()
11002   const std::int32_t result_chunk_stride = result_stride * 3;  in gemm_q8_1_1_6_aligned()
11004   const std::int32_t const_offset = lhs_offset * rhs_offset * k + result_offset;  in gemm_q8_1_1_6_aligned()
11005   const std::int32_t rounding_offset = (1 << (shift - 1));  in gemm_q8_1_1_6_aligned()
11006   std::int32_t* temp_result = reinterpret_cast<std::int32_t*>(  in gemm_q8_1_1_6_aligned()
11009   std::int32_t* mul_result_chunk = temp_result;  in gemm_q8_1_1_6_aligned()
11010   const std::int32_t mul_result_chunk_stride_bytes = ((n * 4 + 7) / 8) * 8;  in gemm_q8_1_1_6_aligned()
11055                            const std::uint8_t* rhs, std::int32_t m,  in gemm_q8_1_1_7_aligned()
11056                            std::int32_t n, std::int32_t k,  in gemm_q8_1_1_7_aligned()
11057                            std::int32_t lhs_offset, std::int32_t rhs_offset,  in gemm_q8_1_1_7_aligned()
11058                            std::int32_t result_offset,  in gemm_q8_1_1_7_aligned()
11059                            std::int32_t multiplicative_offset,  in gemm_q8_1_1_7_aligned()
11060                            std::int32_t shift, std::uint8_t* result,  in gemm_q8_1_1_7_aligned()
11061                            std::int32_t result_stride) {  in gemm_q8_1_1_7_aligned()
11062   const std::int32_t row_chunks = m / 3;  in gemm_q8_1_1_7_aligned()
11063   const std::int32_t col_chunks = n / 3;  in gemm_q8_1_1_7_aligned()
11064   const std::int32_t padded_k = ((k + 7) / 8) * 8;  in gemm_q8_1_1_7_aligned()
11065   const std::int32_t chunk_size = k * 3;  in gemm_q8_1_1_7_aligned()
11066   const std::int32_t zipped_chunk_size = (padded_k + 16) * 3;  in gemm_q8_1_1_7_aligned()
11067   const std::int32_t zipped_rhs_size = (padded_k + 16) * n;  in gemm_q8_1_1_7_aligned()
11071   std::int32_t* zipped_lhs_3_offsets =  in gemm_q8_1_1_7_aligned()
11072       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 3);  in gemm_q8_1_1_7_aligned()
11073   std::int32_t* zipped_lhs_1_offsets =  in gemm_q8_1_1_7_aligned()
11074       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 1);  in gemm_q8_1_1_7_aligned()
11077   const std::int32_t result_chunk_stride = result_stride * 3;  in gemm_q8_1_1_7_aligned()
11079   const std::int32_t const_offset = lhs_offset * rhs_offset * k + result_offset;  in gemm_q8_1_1_7_aligned()
11080   const std::int32_t rounding_offset = (1 << (shift - 1));  in gemm_q8_1_1_7_aligned()
11081   std::int32_t* temp_result = reinterpret_cast<std::int32_t*>(  in gemm_q8_1_1_7_aligned()
11084   std::int32_t* mul_result_chunk = temp_result;  in gemm_q8_1_1_7_aligned()
11085   const std::int32_t mul_result_chunk_stride_bytes = ((n * 4 + 7) / 8) * 8;  in gemm_q8_1_1_7_aligned()
11130                            const std::uint8_t* rhs, std::int32_t m,  in gemm_q8_1_2_0_aligned()
11131                            std::int32_t n, std::int32_t k,  in gemm_q8_1_2_0_aligned()
11132                            std::int32_t lhs_offset, std::int32_t rhs_offset,  in gemm_q8_1_2_0_aligned()
11133                            std::int32_t result_offset,  in gemm_q8_1_2_0_aligned()
11134                            std::int32_t multiplicative_offset,  in gemm_q8_1_2_0_aligned()
11135                            std::int32_t shift, std::uint8_t* result,  in gemm_q8_1_2_0_aligned()
11136                            std::int32_t result_stride) {  in gemm_q8_1_2_0_aligned()
11137   const std::int32_t row_chunks = m / 3;  in gemm_q8_1_2_0_aligned()
11138   const std::int32_t col_chunks = n / 3;  in gemm_q8_1_2_0_aligned()
11139   const std::int32_t padded_k = ((k + 7) / 8) * 8;  in gemm_q8_1_2_0_aligned()
11140   const std::int32_t chunk_size = k * 3;  in gemm_q8_1_2_0_aligned()
11141   const std::int32_t zipped_chunk_size = (padded_k + 16) * 3;  in gemm_q8_1_2_0_aligned()
11142   const std::int32_t zipped_rhs_size = (padded_k + 16) * n;  in gemm_q8_1_2_0_aligned()
11146   std::int32_t* zipped_lhs_3_offsets =  in gemm_q8_1_2_0_aligned()
11147       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 3);  in gemm_q8_1_2_0_aligned()
11148   std::int32_t* zipped_lhs_1_offsets =  in gemm_q8_1_2_0_aligned()
11149       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 1);  in gemm_q8_1_2_0_aligned()
11152   const std::int32_t result_chunk_stride = result_stride * 3;  in gemm_q8_1_2_0_aligned()
11154   const std::int32_t const_offset = lhs_offset * rhs_offset * k + result_offset;  in gemm_q8_1_2_0_aligned()
11155   const std::int32_t rounding_offset = (1 << (shift - 1));  in gemm_q8_1_2_0_aligned()
11156   std::int32_t* temp_result = reinterpret_cast<std::int32_t*>(  in gemm_q8_1_2_0_aligned()
11159   std::int32_t* mul_result_chunk = temp_result;  in gemm_q8_1_2_0_aligned()
11160   const std::int32_t mul_result_chunk_stride_bytes = ((n * 4 + 7) / 8) * 8;  in gemm_q8_1_2_0_aligned()
11205                            const std::uint8_t* rhs, std::int32_t m,  in gemm_q8_1_2_1_aligned()
11206                            std::int32_t n, std::int32_t k,  in gemm_q8_1_2_1_aligned()
11207                            std::int32_t lhs_offset, std::int32_t rhs_offset,  in gemm_q8_1_2_1_aligned()
11208                            std::int32_t result_offset,  in gemm_q8_1_2_1_aligned()
11209                            std::int32_t multiplicative_offset,  in gemm_q8_1_2_1_aligned()
11210                            std::int32_t shift, std::uint8_t* result,  in gemm_q8_1_2_1_aligned()
11211                            std::int32_t result_stride) {  in gemm_q8_1_2_1_aligned()
11212   const std::int32_t row_chunks = m / 3;  in gemm_q8_1_2_1_aligned()
11213   const std::int32_t col_chunks = n / 3;  in gemm_q8_1_2_1_aligned()
11214   const std::int32_t padded_k = ((k + 7) / 8) * 8;  in gemm_q8_1_2_1_aligned()
11215   const std::int32_t chunk_size = k * 3;  in gemm_q8_1_2_1_aligned()
11216   const std::int32_t zipped_chunk_size = (padded_k + 16) * 3;  in gemm_q8_1_2_1_aligned()
11217   const std::int32_t zipped_rhs_size = (padded_k + 16) * n;  in gemm_q8_1_2_1_aligned()
11221   std::int32_t* zipped_lhs_3_offsets =  in gemm_q8_1_2_1_aligned()
11222       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 3);  in gemm_q8_1_2_1_aligned()
11223   std::int32_t* zipped_lhs_1_offsets =  in gemm_q8_1_2_1_aligned()
11224       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 1);  in gemm_q8_1_2_1_aligned()
11227   const std::int32_t result_chunk_stride = result_stride * 3;  in gemm_q8_1_2_1_aligned()
11229   const std::int32_t const_offset = lhs_offset * rhs_offset * k + result_offset;  in gemm_q8_1_2_1_aligned()
11230   const std::int32_t rounding_offset = (1 << (shift - 1));  in gemm_q8_1_2_1_aligned()
11231   std::int32_t* temp_result = reinterpret_cast<std::int32_t*>(  in gemm_q8_1_2_1_aligned()
11234   std::int32_t* mul_result_chunk = temp_result;  in gemm_q8_1_2_1_aligned()
11235   const std::int32_t mul_result_chunk_stride_bytes = ((n * 4 + 7) / 8) * 8;  in gemm_q8_1_2_1_aligned()
11280                            const std::uint8_t* rhs, std::int32_t m,  in gemm_q8_1_2_2_aligned()
11281                            std::int32_t n, std::int32_t k,  in gemm_q8_1_2_2_aligned()
11282                            std::int32_t lhs_offset, std::int32_t rhs_offset,  in gemm_q8_1_2_2_aligned()
11283                            std::int32_t result_offset,  in gemm_q8_1_2_2_aligned()
11284                            std::int32_t multiplicative_offset,  in gemm_q8_1_2_2_aligned()
11285                            std::int32_t shift, std::uint8_t* result,  in gemm_q8_1_2_2_aligned()
11286                            std::int32_t result_stride) {  in gemm_q8_1_2_2_aligned()
11287   const std::int32_t row_chunks = m / 3;  in gemm_q8_1_2_2_aligned()
11288   const std::int32_t col_chunks = n / 3;  in gemm_q8_1_2_2_aligned()
11289   const std::int32_t padded_k = ((k + 7) / 8) * 8;  in gemm_q8_1_2_2_aligned()
11290   const std::int32_t chunk_size = k * 3;  in gemm_q8_1_2_2_aligned()
11291   const std::int32_t zipped_chunk_size = (padded_k + 16) * 3;  in gemm_q8_1_2_2_aligned()
11292   const std::int32_t zipped_rhs_size = (padded_k + 16) * n;  in gemm_q8_1_2_2_aligned()
11296   std::int32_t* zipped_lhs_3_offsets =  in gemm_q8_1_2_2_aligned()
11297       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 3);  in gemm_q8_1_2_2_aligned()
11298   std::int32_t* zipped_lhs_1_offsets =  in gemm_q8_1_2_2_aligned()
11299       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 1);  in gemm_q8_1_2_2_aligned()
11302   const std::int32_t result_chunk_stride = result_stride * 3;  in gemm_q8_1_2_2_aligned()
11304   const std::int32_t const_offset = lhs_offset * rhs_offset * k + result_offset;  in gemm_q8_1_2_2_aligned()
11305   const std::int32_t rounding_offset = (1 << (shift - 1));  in gemm_q8_1_2_2_aligned()
11306   std::int32_t* temp_result = reinterpret_cast<std::int32_t*>(  in gemm_q8_1_2_2_aligned()
11309   std::int32_t* mul_result_chunk = temp_result;  in gemm_q8_1_2_2_aligned()
11310   const std::int32_t mul_result_chunk_stride_bytes = ((n * 4 + 7) / 8) * 8;  in gemm_q8_1_2_2_aligned()
11355                            const std::uint8_t* rhs, std::int32_t m,  in gemm_q8_1_2_3_aligned()
11356                            std::int32_t n, std::int32_t k,  in gemm_q8_1_2_3_aligned()
11357                            std::int32_t lhs_offset, std::int32_t rhs_offset,  in gemm_q8_1_2_3_aligned()
11358                            std::int32_t result_offset,  in gemm_q8_1_2_3_aligned()
11359                            std::int32_t multiplicative_offset,  in gemm_q8_1_2_3_aligned()
11360                            std::int32_t shift, std::uint8_t* result,  in gemm_q8_1_2_3_aligned()
11361                            std::int32_t result_stride) {  in gemm_q8_1_2_3_aligned()
11362   const std::int32_t row_chunks = m / 3;  in gemm_q8_1_2_3_aligned()
11363   const std::int32_t col_chunks = n / 3;  in gemm_q8_1_2_3_aligned()
11364   const std::int32_t padded_k = ((k + 7) / 8) * 8;  in gemm_q8_1_2_3_aligned()
11365   const std::int32_t chunk_size = k * 3;  in gemm_q8_1_2_3_aligned()
11366   const std::int32_t zipped_chunk_size = (padded_k + 16) * 3;  in gemm_q8_1_2_3_aligned()
11367   const std::int32_t zipped_rhs_size = (padded_k + 16) * n;  in gemm_q8_1_2_3_aligned()
11371   std::int32_t* zipped_lhs_3_offsets =  in gemm_q8_1_2_3_aligned()
11372       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 3);  in gemm_q8_1_2_3_aligned()
11373   std::int32_t* zipped_lhs_1_offsets =  in gemm_q8_1_2_3_aligned()
11374       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 1);  in gemm_q8_1_2_3_aligned()
11377   const std::int32_t result_chunk_stride = result_stride * 3;  in gemm_q8_1_2_3_aligned()
11379   const std::int32_t const_offset = lhs_offset * rhs_offset * k + result_offset;  in gemm_q8_1_2_3_aligned()
11380   const std::int32_t rounding_offset = (1 << (shift - 1));  in gemm_q8_1_2_3_aligned()
11381   std::int32_t* temp_result = reinterpret_cast<std::int32_t*>(  in gemm_q8_1_2_3_aligned()
11384   std::int32_t* mul_result_chunk = temp_result;  in gemm_q8_1_2_3_aligned()
11385   const std::int32_t mul_result_chunk_stride_bytes = ((n * 4 + 7) / 8) * 8;  in gemm_q8_1_2_3_aligned()
11430                            const std::uint8_t* rhs, std::int32_t m,  in gemm_q8_1_2_4_aligned()
11431                            std::int32_t n, std::int32_t k,  in gemm_q8_1_2_4_aligned()
11432                            std::int32_t lhs_offset, std::int32_t rhs_offset,  in gemm_q8_1_2_4_aligned()
11433                            std::int32_t result_offset,  in gemm_q8_1_2_4_aligned()
11434                            std::int32_t multiplicative_offset,  in gemm_q8_1_2_4_aligned()
11435                            std::int32_t shift, std::uint8_t* result,  in gemm_q8_1_2_4_aligned()
11436                            std::int32_t result_stride) {  in gemm_q8_1_2_4_aligned()
11437   const std::int32_t row_chunks = m / 3;  in gemm_q8_1_2_4_aligned()
11438   const std::int32_t col_chunks = n / 3;  in gemm_q8_1_2_4_aligned()
11439   const std::int32_t padded_k = ((k + 7) / 8) * 8;  in gemm_q8_1_2_4_aligned()
11440   const std::int32_t chunk_size = k * 3;  in gemm_q8_1_2_4_aligned()
11441   const std::int32_t zipped_chunk_size = (padded_k + 16) * 3;  in gemm_q8_1_2_4_aligned()
11442   const std::int32_t zipped_rhs_size = (padded_k + 16) * n;  in gemm_q8_1_2_4_aligned()
11446   std::int32_t* zipped_lhs_3_offsets =  in gemm_q8_1_2_4_aligned()
11447       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 3);  in gemm_q8_1_2_4_aligned()
11448   std::int32_t* zipped_lhs_1_offsets =  in gemm_q8_1_2_4_aligned()
11449       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 1);  in gemm_q8_1_2_4_aligned()
11452   const std::int32_t result_chunk_stride = result_stride * 3;  in gemm_q8_1_2_4_aligned()
11454   const std::int32_t const_offset = lhs_offset * rhs_offset * k + result_offset;  in gemm_q8_1_2_4_aligned()
11455   const std::int32_t rounding_offset = (1 << (shift - 1));  in gemm_q8_1_2_4_aligned()
11456   std::int32_t* temp_result = reinterpret_cast<std::int32_t*>(  in gemm_q8_1_2_4_aligned()
11459   std::int32_t* mul_result_chunk = temp_result;  in gemm_q8_1_2_4_aligned()
11460   const std::int32_t mul_result_chunk_stride_bytes = ((n * 4 + 7) / 8) * 8;  in gemm_q8_1_2_4_aligned()
11505                            const std::uint8_t* rhs, std::int32_t m,  in gemm_q8_1_2_5_aligned()
11506                            std::int32_t n, std::int32_t k,  in gemm_q8_1_2_5_aligned()
11507                            std::int32_t lhs_offset, std::int32_t rhs_offset,  in gemm_q8_1_2_5_aligned()
11508                            std::int32_t result_offset,  in gemm_q8_1_2_5_aligned()
11509                            std::int32_t multiplicative_offset,  in gemm_q8_1_2_5_aligned()
11510                            std::int32_t shift, std::uint8_t* result,  in gemm_q8_1_2_5_aligned()
11511                            std::int32_t result_stride) {  in gemm_q8_1_2_5_aligned()
11512   const std::int32_t row_chunks = m / 3;  in gemm_q8_1_2_5_aligned()
11513   const std::int32_t col_chunks = n / 3;  in gemm_q8_1_2_5_aligned()
11514   const std::int32_t padded_k = ((k + 7) / 8) * 8;  in gemm_q8_1_2_5_aligned()
11515   const std::int32_t chunk_size = k * 3;  in gemm_q8_1_2_5_aligned()
11516   const std::int32_t zipped_chunk_size = (padded_k + 16) * 3;  in gemm_q8_1_2_5_aligned()
11517   const std::int32_t zipped_rhs_size = (padded_k + 16) * n;  in gemm_q8_1_2_5_aligned()
11521   std::int32_t* zipped_lhs_3_offsets =  in gemm_q8_1_2_5_aligned()
11522       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 3);  in gemm_q8_1_2_5_aligned()
11523   std::int32_t* zipped_lhs_1_offsets =  in gemm_q8_1_2_5_aligned()
11524       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 1);  in gemm_q8_1_2_5_aligned()
11527   const std::int32_t result_chunk_stride = result_stride * 3;  in gemm_q8_1_2_5_aligned()
11529   const std::int32_t const_offset = lhs_offset * rhs_offset * k + result_offset;  in gemm_q8_1_2_5_aligned()
11530   const std::int32_t rounding_offset = (1 << (shift - 1));  in gemm_q8_1_2_5_aligned()
11531   std::int32_t* temp_result = reinterpret_cast<std::int32_t*>(  in gemm_q8_1_2_5_aligned()
11534   std::int32_t* mul_result_chunk = temp_result;  in gemm_q8_1_2_5_aligned()
11535   const std::int32_t mul_result_chunk_stride_bytes = ((n * 4 + 7) / 8) * 8;  in gemm_q8_1_2_5_aligned()
11580                            const std::uint8_t* rhs, std::int32_t m,  in gemm_q8_1_2_6_aligned()
11581                            std::int32_t n, std::int32_t k,  in gemm_q8_1_2_6_aligned()
11582                            std::int32_t lhs_offset, std::int32_t rhs_offset,  in gemm_q8_1_2_6_aligned()
11583                            std::int32_t result_offset,  in gemm_q8_1_2_6_aligned()
11584                            std::int32_t multiplicative_offset,  in gemm_q8_1_2_6_aligned()
11585                            std::int32_t shift, std::uint8_t* result,  in gemm_q8_1_2_6_aligned()
11586                            std::int32_t result_stride) {  in gemm_q8_1_2_6_aligned()
11587   const std::int32_t row_chunks = m / 3;  in gemm_q8_1_2_6_aligned()
11588   const std::int32_t col_chunks = n / 3;  in gemm_q8_1_2_6_aligned()
11589   const std::int32_t padded_k = ((k + 7) / 8) * 8;  in gemm_q8_1_2_6_aligned()
11590   const std::int32_t chunk_size = k * 3;  in gemm_q8_1_2_6_aligned()
11591   const std::int32_t zipped_chunk_size = (padded_k + 16) * 3;  in gemm_q8_1_2_6_aligned()
11592   const std::int32_t zipped_rhs_size = (padded_k + 16) * n;  in gemm_q8_1_2_6_aligned()
11596   std::int32_t* zipped_lhs_3_offsets =  in gemm_q8_1_2_6_aligned()
11597       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 3);  in gemm_q8_1_2_6_aligned()
11598   std::int32_t* zipped_lhs_1_offsets =  in gemm_q8_1_2_6_aligned()
11599       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 1);  in gemm_q8_1_2_6_aligned()
11602   const std::int32_t result_chunk_stride = result_stride * 3;  in gemm_q8_1_2_6_aligned()
11604   const std::int32_t const_offset = lhs_offset * rhs_offset * k + result_offset;  in gemm_q8_1_2_6_aligned()
11605   const std::int32_t rounding_offset = (1 << (shift - 1));  in gemm_q8_1_2_6_aligned()
11606   std::int32_t* temp_result = reinterpret_cast<std::int32_t*>(  in gemm_q8_1_2_6_aligned()
11609   std::int32_t* mul_result_chunk = temp_result;  in gemm_q8_1_2_6_aligned()
11610   const std::int32_t mul_result_chunk_stride_bytes = ((n * 4 + 7) / 8) * 8;  in gemm_q8_1_2_6_aligned()
11655                            const std::uint8_t* rhs, std::int32_t m,  in gemm_q8_1_2_7_aligned()
11656                            std::int32_t n, std::int32_t k,  in gemm_q8_1_2_7_aligned()
11657                            std::int32_t lhs_offset, std::int32_t rhs_offset,  in gemm_q8_1_2_7_aligned()
11658                            std::int32_t result_offset,  in gemm_q8_1_2_7_aligned()
11659                            std::int32_t multiplicative_offset,  in gemm_q8_1_2_7_aligned()
11660                            std::int32_t shift, std::uint8_t* result,  in gemm_q8_1_2_7_aligned()
11661                            std::int32_t result_stride) {  in gemm_q8_1_2_7_aligned()
11662   const std::int32_t row_chunks = m / 3;  in gemm_q8_1_2_7_aligned()
11663   const std::int32_t col_chunks = n / 3;  in gemm_q8_1_2_7_aligned()
11664   const std::int32_t padded_k = ((k + 7) / 8) * 8;  in gemm_q8_1_2_7_aligned()
11665   const std::int32_t chunk_size = k * 3;  in gemm_q8_1_2_7_aligned()
11666   const std::int32_t zipped_chunk_size = (padded_k + 16) * 3;  in gemm_q8_1_2_7_aligned()
11667   const std::int32_t zipped_rhs_size = (padded_k + 16) * n;  in gemm_q8_1_2_7_aligned()
11671   std::int32_t* zipped_lhs_3_offsets =  in gemm_q8_1_2_7_aligned()
11672       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 3);  in gemm_q8_1_2_7_aligned()
11673   std::int32_t* zipped_lhs_1_offsets =  in gemm_q8_1_2_7_aligned()
11674       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 1);  in gemm_q8_1_2_7_aligned()
11677   const std::int32_t result_chunk_stride = result_stride * 3;  in gemm_q8_1_2_7_aligned()
11679   const std::int32_t const_offset = lhs_offset * rhs_offset * k + result_offset;  in gemm_q8_1_2_7_aligned()
11680   const std::int32_t rounding_offset = (1 << (shift - 1));  in gemm_q8_1_2_7_aligned()
11681   std::int32_t* temp_result = reinterpret_cast<std::int32_t*>(  in gemm_q8_1_2_7_aligned()
11684   std::int32_t* mul_result_chunk = temp_result;  in gemm_q8_1_2_7_aligned()
11685   const std::int32_t mul_result_chunk_stride_bytes = ((n * 4 + 7) / 8) * 8;  in gemm_q8_1_2_7_aligned()
11730                            const std::uint8_t* rhs, std::int32_t m,  in gemm_q8_2_0_0_aligned()
11731                            std::int32_t n, std::int32_t k,  in gemm_q8_2_0_0_aligned()
11732                            std::int32_t lhs_offset, std::int32_t rhs_offset,  in gemm_q8_2_0_0_aligned()
11733                            std::int32_t result_offset,  in gemm_q8_2_0_0_aligned()
11734                            std::int32_t multiplicative_offset,  in gemm_q8_2_0_0_aligned()
11735                            std::int32_t shift, std::uint8_t* result,  in gemm_q8_2_0_0_aligned()
11736                            std::int32_t result_stride) {  in gemm_q8_2_0_0_aligned()
11737   const std::int32_t row_chunks = m / 3;  in gemm_q8_2_0_0_aligned()
11738   const std::int32_t col_chunks = n / 3;  in gemm_q8_2_0_0_aligned()
11739   const std::int32_t padded_k = ((k + 7) / 8) * 8;  in gemm_q8_2_0_0_aligned()
11740   const std::int32_t chunk_size = k * 3;  in gemm_q8_2_0_0_aligned()
11741   const std::int32_t zipped_chunk_size = (padded_k + 16) * 3;  in gemm_q8_2_0_0_aligned()
11742   const std::int32_t zipped_rhs_size = (padded_k + 16) * n;  in gemm_q8_2_0_0_aligned()
11746   std::int32_t* zipped_lhs_3_offsets =  in gemm_q8_2_0_0_aligned()
11747       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 3);  in gemm_q8_2_0_0_aligned()
11748   std::int32_t* zipped_lhs_2_offsets =  in gemm_q8_2_0_0_aligned()
11749       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 2);  in gemm_q8_2_0_0_aligned()
11752   const std::int32_t result_chunk_stride = result_stride * 3;  in gemm_q8_2_0_0_aligned()
11754   const std::int32_t const_offset = lhs_offset * rhs_offset * k + result_offset;  in gemm_q8_2_0_0_aligned()
11755   const std::int32_t rounding_offset = (1 << (shift - 1));  in gemm_q8_2_0_0_aligned()
11756   std::int32_t* temp_result = reinterpret_cast<std::int32_t*>(  in gemm_q8_2_0_0_aligned()
11759   std::int32_t* mul_result_chunk = temp_result;  in gemm_q8_2_0_0_aligned()
11760   const std::int32_t mul_result_chunk_stride_bytes = ((n * 4 + 7) / 8) * 8;  in gemm_q8_2_0_0_aligned()
11800                            const std::uint8_t* rhs, std::int32_t m,  in gemm_q8_2_0_1_aligned()
11801                            std::int32_t n, std::int32_t k,  in gemm_q8_2_0_1_aligned()
11802                            std::int32_t lhs_offset, std::int32_t rhs_offset,  in gemm_q8_2_0_1_aligned()
11803                            std::int32_t result_offset,  in gemm_q8_2_0_1_aligned()
11804                            std::int32_t multiplicative_offset,  in gemm_q8_2_0_1_aligned()
11805                            std::int32_t shift, std::uint8_t* result,  in gemm_q8_2_0_1_aligned()
11806                            std::int32_t result_stride) {  in gemm_q8_2_0_1_aligned()
11807   const std::int32_t row_chunks = m / 3;  in gemm_q8_2_0_1_aligned()
11808   const std::int32_t col_chunks = n / 3;  in gemm_q8_2_0_1_aligned()
11809   const std::int32_t padded_k = ((k + 7) / 8) * 8;  in gemm_q8_2_0_1_aligned()
11810   const std::int32_t chunk_size = k * 3;  in gemm_q8_2_0_1_aligned()
11811   const std::int32_t zipped_chunk_size = (padded_k + 16) * 3;  in gemm_q8_2_0_1_aligned()
11812   const std::int32_t zipped_rhs_size = (padded_k + 16) * n;  in gemm_q8_2_0_1_aligned()
11816   std::int32_t* zipped_lhs_3_offsets =  in gemm_q8_2_0_1_aligned()
11817       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 3);  in gemm_q8_2_0_1_aligned()
11818   std::int32_t* zipped_lhs_2_offsets =  in gemm_q8_2_0_1_aligned()
11819       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 2);  in gemm_q8_2_0_1_aligned()
11822   const std::int32_t result_chunk_stride = result_stride * 3;  in gemm_q8_2_0_1_aligned()
11824   const std::int32_t const_offset = lhs_offset * rhs_offset * k + result_offset;  in gemm_q8_2_0_1_aligned()
11825   const std::int32_t rounding_offset = (1 << (shift - 1));  in gemm_q8_2_0_1_aligned()
11826   std::int32_t* temp_result = reinterpret_cast<std::int32_t*>(  in gemm_q8_2_0_1_aligned()
11829   std::int32_t* mul_result_chunk = temp_result;  in gemm_q8_2_0_1_aligned()
11830   const std::int32_t mul_result_chunk_stride_bytes = ((n * 4 + 7) / 8) * 8;  in gemm_q8_2_0_1_aligned()
11870                            const std::uint8_t* rhs, std::int32_t m,  in gemm_q8_2_0_2_aligned()
11871                            std::int32_t n, std::int32_t k,  in gemm_q8_2_0_2_aligned()
11872                            std::int32_t lhs_offset, std::int32_t rhs_offset,  in gemm_q8_2_0_2_aligned()
11873                            std::int32_t result_offset,  in gemm_q8_2_0_2_aligned()
11874                            std::int32_t multiplicative_offset,  in gemm_q8_2_0_2_aligned()
11875                            std::int32_t shift, std::uint8_t* result,  in gemm_q8_2_0_2_aligned()
11876                            std::int32_t result_stride) {  in gemm_q8_2_0_2_aligned()
11877   const std::int32_t row_chunks = m / 3;  in gemm_q8_2_0_2_aligned()
11878   const std::int32_t col_chunks = n / 3;  in gemm_q8_2_0_2_aligned()
11879   const std::int32_t padded_k = ((k + 7) / 8) * 8;  in gemm_q8_2_0_2_aligned()
11880   const std::int32_t chunk_size = k * 3;  in gemm_q8_2_0_2_aligned()
11881   const std::int32_t zipped_chunk_size = (padded_k + 16) * 3;  in gemm_q8_2_0_2_aligned()
11882   const std::int32_t zipped_rhs_size = (padded_k + 16) * n;  in gemm_q8_2_0_2_aligned()
11886   std::int32_t* zipped_lhs_3_offsets =  in gemm_q8_2_0_2_aligned()
11887       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 3);  in gemm_q8_2_0_2_aligned()
11888   std::int32_t* zipped_lhs_2_offsets =  in gemm_q8_2_0_2_aligned()
11889       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 2);  in gemm_q8_2_0_2_aligned()
11892   const std::int32_t result_chunk_stride = result_stride * 3;  in gemm_q8_2_0_2_aligned()
11894   const std::int32_t const_offset = lhs_offset * rhs_offset * k + result_offset;  in gemm_q8_2_0_2_aligned()
11895   const std::int32_t rounding_offset = (1 << (shift - 1));  in gemm_q8_2_0_2_aligned()
11896   std::int32_t* temp_result = reinterpret_cast<std::int32_t*>(  in gemm_q8_2_0_2_aligned()
11899   std::int32_t* mul_result_chunk = temp_result;  in gemm_q8_2_0_2_aligned()
11900   const std::int32_t mul_result_chunk_stride_bytes = ((n * 4 + 7) / 8) * 8;  in gemm_q8_2_0_2_aligned()
11940                            const std::uint8_t* rhs, std::int32_t m,  in gemm_q8_2_0_3_aligned()
11941                            std::int32_t n, std::int32_t k,  in gemm_q8_2_0_3_aligned()
11942                            std::int32_t lhs_offset, std::int32_t rhs_offset,  in gemm_q8_2_0_3_aligned()
11943                            std::int32_t result_offset,  in gemm_q8_2_0_3_aligned()
11944                            std::int32_t multiplicative_offset,  in gemm_q8_2_0_3_aligned()
11945                            std::int32_t shift, std::uint8_t* result,  in gemm_q8_2_0_3_aligned()
11946                            std::int32_t result_stride) {  in gemm_q8_2_0_3_aligned()
11947   const std::int32_t row_chunks = m / 3;  in gemm_q8_2_0_3_aligned()
11948   const std::int32_t col_chunks = n / 3;  in gemm_q8_2_0_3_aligned()
11949   const std::int32_t padded_k = ((k + 7) / 8) * 8;  in gemm_q8_2_0_3_aligned()
11950   const std::int32_t chunk_size = k * 3;  in gemm_q8_2_0_3_aligned()
11951   const std::int32_t zipped_chunk_size = (padded_k + 16) * 3;  in gemm_q8_2_0_3_aligned()
11952   const std::int32_t zipped_rhs_size = (padded_k + 16) * n;  in gemm_q8_2_0_3_aligned()
11956   std::int32_t* zipped_lhs_3_offsets =  in gemm_q8_2_0_3_aligned()
11957       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 3);  in gemm_q8_2_0_3_aligned()
11958   std::int32_t* zipped_lhs_2_offsets =  in gemm_q8_2_0_3_aligned()
11959       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 2);  in gemm_q8_2_0_3_aligned()
11962   const std::int32_t result_chunk_stride = result_stride * 3;  in gemm_q8_2_0_3_aligned()
11964   const std::int32_t const_offset = lhs_offset * rhs_offset * k + result_offset;  in gemm_q8_2_0_3_aligned()
11965   const std::int32_t rounding_offset = (1 << (shift - 1));  in gemm_q8_2_0_3_aligned()
11966   std::int32_t* temp_result = reinterpret_cast<std::int32_t*>(  in gemm_q8_2_0_3_aligned()
11969   std::int32_t* mul_result_chunk = temp_result;  in gemm_q8_2_0_3_aligned()
11970   const std::int32_t mul_result_chunk_stride_bytes = ((n * 4 + 7) / 8) * 8;  in gemm_q8_2_0_3_aligned()
12010                            const std::uint8_t* rhs, std::int32_t m,  in gemm_q8_2_0_4_aligned()
12011                            std::int32_t n, std::int32_t k,  in gemm_q8_2_0_4_aligned()
12012                            std::int32_t lhs_offset, std::int32_t rhs_offset,  in gemm_q8_2_0_4_aligned()
12013                            std::int32_t result_offset,  in gemm_q8_2_0_4_aligned()
12014                            std::int32_t multiplicative_offset,  in gemm_q8_2_0_4_aligned()
12015                            std::int32_t shift, std::uint8_t* result,  in gemm_q8_2_0_4_aligned()
12016                            std::int32_t result_stride) {  in gemm_q8_2_0_4_aligned()
12017   const std::int32_t row_chunks = m / 3;  in gemm_q8_2_0_4_aligned()
12018   const std::int32_t col_chunks = n / 3;  in gemm_q8_2_0_4_aligned()
12019   const std::int32_t padded_k = ((k + 7) / 8) * 8;  in gemm_q8_2_0_4_aligned()
12020   const std::int32_t chunk_size = k * 3;  in gemm_q8_2_0_4_aligned()
12021   const std::int32_t zipped_chunk_size = (padded_k + 16) * 3;  in gemm_q8_2_0_4_aligned()
12022   const std::int32_t zipped_rhs_size = (padded_k + 16) * n;  in gemm_q8_2_0_4_aligned()
12026   std::int32_t* zipped_lhs_3_offsets =  in gemm_q8_2_0_4_aligned()
12027       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 3);  in gemm_q8_2_0_4_aligned()
12028   std::int32_t* zipped_lhs_2_offsets =  in gemm_q8_2_0_4_aligned()
12029       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 2);  in gemm_q8_2_0_4_aligned()
12032   const std::int32_t result_chunk_stride = result_stride * 3;  in gemm_q8_2_0_4_aligned()
12034   const std::int32_t const_offset = lhs_offset * rhs_offset * k + result_offset;  in gemm_q8_2_0_4_aligned()
12035   const std::int32_t rounding_offset = (1 << (shift - 1));  in gemm_q8_2_0_4_aligned()
12036   std::int32_t* temp_result = reinterpret_cast<std::int32_t*>(  in gemm_q8_2_0_4_aligned()
12039   std::int32_t* mul_result_chunk = temp_result;  in gemm_q8_2_0_4_aligned()
12040   const std::int32_t mul_result_chunk_stride_bytes = ((n * 4 + 7) / 8) * 8;  in gemm_q8_2_0_4_aligned()
12080                            const std::uint8_t* rhs, std::int32_t m,  in gemm_q8_2_0_5_aligned()
12081                            std::int32_t n, std::int32_t k,  in gemm_q8_2_0_5_aligned()
12082                            std::int32_t lhs_offset, std::int32_t rhs_offset,  in gemm_q8_2_0_5_aligned()
12083                            std::int32_t result_offset,  in gemm_q8_2_0_5_aligned()
12084                            std::int32_t multiplicative_offset,  in gemm_q8_2_0_5_aligned()
12085                            std::int32_t shift, std::uint8_t* result,  in gemm_q8_2_0_5_aligned()
12086                            std::int32_t result_stride) {  in gemm_q8_2_0_5_aligned()
12087   const std::int32_t row_chunks = m / 3;  in gemm_q8_2_0_5_aligned()
12088   const std::int32_t col_chunks = n / 3;  in gemm_q8_2_0_5_aligned()
12089   const std::int32_t padded_k = ((k + 7) / 8) * 8;  in gemm_q8_2_0_5_aligned()
12090   const std::int32_t chunk_size = k * 3;  in gemm_q8_2_0_5_aligned()
12091   const std::int32_t zipped_chunk_size = (padded_k + 16) * 3;  in gemm_q8_2_0_5_aligned()
12092   const std::int32_t zipped_rhs_size = (padded_k + 16) * n;  in gemm_q8_2_0_5_aligned()
12096   std::int32_t* zipped_lhs_3_offsets =  in gemm_q8_2_0_5_aligned()
12097       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 3);  in gemm_q8_2_0_5_aligned()
12098   std::int32_t* zipped_lhs_2_offsets =  in gemm_q8_2_0_5_aligned()
12099       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 2);  in gemm_q8_2_0_5_aligned()
12102   const std::int32_t result_chunk_stride = result_stride * 3;  in gemm_q8_2_0_5_aligned()
12104   const std::int32_t const_offset = lhs_offset * rhs_offset * k + result_offset;  in gemm_q8_2_0_5_aligned()
12105   const std::int32_t rounding_offset = (1 << (shift - 1));  in gemm_q8_2_0_5_aligned()
12106   std::int32_t* temp_result = reinterpret_cast<std::int32_t*>(  in gemm_q8_2_0_5_aligned()
12109   std::int32_t* mul_result_chunk = temp_result;  in gemm_q8_2_0_5_aligned()
12110   const std::int32_t mul_result_chunk_stride_bytes = ((n * 4 + 7) / 8) * 8;  in gemm_q8_2_0_5_aligned()
12150                            const std::uint8_t* rhs, std::int32_t m,  in gemm_q8_2_0_6_aligned()
12151                            std::int32_t n, std::int32_t k,  in gemm_q8_2_0_6_aligned()
12152                            std::int32_t lhs_offset, std::int32_t rhs_offset,  in gemm_q8_2_0_6_aligned()
12153                            std::int32_t result_offset,  in gemm_q8_2_0_6_aligned()
12154                            std::int32_t multiplicative_offset,  in gemm_q8_2_0_6_aligned()
12155                            std::int32_t shift, std::uint8_t* result,  in gemm_q8_2_0_6_aligned()
12156                            std::int32_t result_stride) {  in gemm_q8_2_0_6_aligned()
12157   const std::int32_t row_chunks = m / 3;  in gemm_q8_2_0_6_aligned()
12158   const std::int32_t col_chunks = n / 3;  in gemm_q8_2_0_6_aligned()
12159   const std::int32_t padded_k = ((k + 7) / 8) * 8;  in gemm_q8_2_0_6_aligned()
12160   const std::int32_t chunk_size = k * 3;  in gemm_q8_2_0_6_aligned()
12161   const std::int32_t zipped_chunk_size = (padded_k + 16) * 3;  in gemm_q8_2_0_6_aligned()
12162   const std::int32_t zipped_rhs_size = (padded_k + 16) * n;  in gemm_q8_2_0_6_aligned()
12166   std::int32_t* zipped_lhs_3_offsets =  in gemm_q8_2_0_6_aligned()
12167       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 3);  in gemm_q8_2_0_6_aligned()
12168   std::int32_t* zipped_lhs_2_offsets =  in gemm_q8_2_0_6_aligned()
12169       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 2);  in gemm_q8_2_0_6_aligned()
12172   const std::int32_t result_chunk_stride = result_stride * 3;  in gemm_q8_2_0_6_aligned()
12174   const std::int32_t const_offset = lhs_offset * rhs_offset * k + result_offset;  in gemm_q8_2_0_6_aligned()
12175   const std::int32_t rounding_offset = (1 << (shift - 1));  in gemm_q8_2_0_6_aligned()
12176   std::int32_t* temp_result = reinterpret_cast<std::int32_t*>(  in gemm_q8_2_0_6_aligned()
12179   std::int32_t* mul_result_chunk = temp_result;  in gemm_q8_2_0_6_aligned()
12180   const std::int32_t mul_result_chunk_stride_bytes = ((n * 4 + 7) / 8) * 8;  in gemm_q8_2_0_6_aligned()
12220                            const std::uint8_t* rhs, std::int32_t m,  in gemm_q8_2_0_7_aligned()
12221                            std::int32_t n, std::int32_t k,  in gemm_q8_2_0_7_aligned()
12222                            std::int32_t lhs_offset, std::int32_t rhs_offset,  in gemm_q8_2_0_7_aligned()
12223                            std::int32_t result_offset,  in gemm_q8_2_0_7_aligned()
12224                            std::int32_t multiplicative_offset,  in gemm_q8_2_0_7_aligned()
12225                            std::int32_t shift, std::uint8_t* result,  in gemm_q8_2_0_7_aligned()
12226                            std::int32_t result_stride) {  in gemm_q8_2_0_7_aligned()
12227   const std::int32_t row_chunks = m / 3;  in gemm_q8_2_0_7_aligned()
12228   const std::int32_t col_chunks = n / 3;  in gemm_q8_2_0_7_aligned()
12229   const std::int32_t padded_k = ((k + 7) / 8) * 8;  in gemm_q8_2_0_7_aligned()
12230   const std::int32_t chunk_size = k * 3;  in gemm_q8_2_0_7_aligned()
12231   const std::int32_t zipped_chunk_size = (padded_k + 16) * 3;  in gemm_q8_2_0_7_aligned()
12232   const std::int32_t zipped_rhs_size = (padded_k + 16) * n;  in gemm_q8_2_0_7_aligned()
12236   std::int32_t* zipped_lhs_3_offsets =  in gemm_q8_2_0_7_aligned()
12237       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 3);  in gemm_q8_2_0_7_aligned()
12238   std::int32_t* zipped_lhs_2_offsets =  in gemm_q8_2_0_7_aligned()
12239       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 2);  in gemm_q8_2_0_7_aligned()
12242   const std::int32_t result_chunk_stride = result_stride * 3;  in gemm_q8_2_0_7_aligned()
12244   const std::int32_t const_offset = lhs_offset * rhs_offset * k + result_offset;  in gemm_q8_2_0_7_aligned()
12245   const std::int32_t rounding_offset = (1 << (shift - 1));  in gemm_q8_2_0_7_aligned()
12246   std::int32_t* temp_result = reinterpret_cast<std::int32_t*>(  in gemm_q8_2_0_7_aligned()
12249   std::int32_t* mul_result_chunk = temp_result;  in gemm_q8_2_0_7_aligned()
12250   const std::int32_t mul_result_chunk_stride_bytes = ((n * 4 + 7) / 8) * 8;  in gemm_q8_2_0_7_aligned()
12290                            const std::uint8_t* rhs, std::int32_t m,  in gemm_q8_2_1_0_aligned()
12291                            std::int32_t n, std::int32_t k,  in gemm_q8_2_1_0_aligned()
12292                            std::int32_t lhs_offset, std::int32_t rhs_offset,  in gemm_q8_2_1_0_aligned()
12293                            std::int32_t result_offset,  in gemm_q8_2_1_0_aligned()
12294                            std::int32_t multiplicative_offset,  in gemm_q8_2_1_0_aligned()
12295                            std::int32_t shift, std::uint8_t* result,  in gemm_q8_2_1_0_aligned()
12296                            std::int32_t result_stride) {  in gemm_q8_2_1_0_aligned()
12297   const std::int32_t row_chunks = m / 3;  in gemm_q8_2_1_0_aligned()
12298   const std::int32_t col_chunks = n / 3;  in gemm_q8_2_1_0_aligned()
12299   const std::int32_t padded_k = ((k + 7) / 8) * 8;  in gemm_q8_2_1_0_aligned()
12300   const std::int32_t chunk_size = k * 3;  in gemm_q8_2_1_0_aligned()
12301   const std::int32_t zipped_chunk_size = (padded_k + 16) * 3;  in gemm_q8_2_1_0_aligned()
12302   const std::int32_t zipped_rhs_size = (padded_k + 16) * n;  in gemm_q8_2_1_0_aligned()
12306   std::int32_t* zipped_lhs_3_offsets =  in gemm_q8_2_1_0_aligned()
12307       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 3);  in gemm_q8_2_1_0_aligned()
12308   std::int32_t* zipped_lhs_2_offsets =  in gemm_q8_2_1_0_aligned()
12309       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 2);  in gemm_q8_2_1_0_aligned()
12312   const std::int32_t result_chunk_stride = result_stride * 3;  in gemm_q8_2_1_0_aligned()
12314   const std::int32_t const_offset = lhs_offset * rhs_offset * k + result_offset;  in gemm_q8_2_1_0_aligned()
12315   const std::int32_t rounding_offset = (1 << (shift - 1));  in gemm_q8_2_1_0_aligned()
12316   std::int32_t* temp_result = reinterpret_cast<std::int32_t*>(  in gemm_q8_2_1_0_aligned()
12319   std::int32_t* mul_result_chunk = temp_result;  in gemm_q8_2_1_0_aligned()
12320   const std::int32_t mul_result_chunk_stride_bytes = ((n * 4 + 7) / 8) * 8;  in gemm_q8_2_1_0_aligned()
12365                            const std::uint8_t* rhs, std::int32_t m,  in gemm_q8_2_1_1_aligned()
12366                            std::int32_t n, std::int32_t k,  in gemm_q8_2_1_1_aligned()
12367                            std::int32_t lhs_offset, std::int32_t rhs_offset,  in gemm_q8_2_1_1_aligned()
12368                            std::int32_t result_offset,  in gemm_q8_2_1_1_aligned()
12369                            std::int32_t multiplicative_offset,  in gemm_q8_2_1_1_aligned()
12370                            std::int32_t shift, std::uint8_t* result,  in gemm_q8_2_1_1_aligned()
12371                            std::int32_t result_stride) {  in gemm_q8_2_1_1_aligned()
12372   const std::int32_t row_chunks = m / 3;  in gemm_q8_2_1_1_aligned()
12373   const std::int32_t col_chunks = n / 3;  in gemm_q8_2_1_1_aligned()
12374   const std::int32_t padded_k = ((k + 7) / 8) * 8;  in gemm_q8_2_1_1_aligned()
12375   const std::int32_t chunk_size = k * 3;  in gemm_q8_2_1_1_aligned()
12376   const std::int32_t zipped_chunk_size = (padded_k + 16) * 3;  in gemm_q8_2_1_1_aligned()
12377   const std::int32_t zipped_rhs_size = (padded_k + 16) * n;  in gemm_q8_2_1_1_aligned()
12381   std::int32_t* zipped_lhs_3_offsets =  in gemm_q8_2_1_1_aligned()
12382       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 3);  in gemm_q8_2_1_1_aligned()
12383   std::int32_t* zipped_lhs_2_offsets =  in gemm_q8_2_1_1_aligned()
12384       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 2);  in gemm_q8_2_1_1_aligned()
12387   const std::int32_t result_chunk_stride = result_stride * 3;  in gemm_q8_2_1_1_aligned()
12389   const std::int32_t const_offset = lhs_offset * rhs_offset * k + result_offset;  in gemm_q8_2_1_1_aligned()
12390   const std::int32_t rounding_offset = (1 << (shift - 1));  in gemm_q8_2_1_1_aligned()
12391   std::int32_t* temp_result = reinterpret_cast<std::int32_t*>(  in gemm_q8_2_1_1_aligned()
12394   std::int32_t* mul_result_chunk = temp_result;  in gemm_q8_2_1_1_aligned()
12395   const std::int32_t mul_result_chunk_stride_bytes = ((n * 4 + 7) / 8) * 8;  in gemm_q8_2_1_1_aligned()
12440                            const std::uint8_t* rhs, std::int32_t m,  in gemm_q8_2_1_2_aligned()
12441                            std::int32_t n, std::int32_t k,  in gemm_q8_2_1_2_aligned()
12442                            std::int32_t lhs_offset, std::int32_t rhs_offset,  in gemm_q8_2_1_2_aligned()
12443                            std::int32_t result_offset,  in gemm_q8_2_1_2_aligned()
12444                            std::int32_t multiplicative_offset,  in gemm_q8_2_1_2_aligned()
12445                            std::int32_t shift, std::uint8_t* result,  in gemm_q8_2_1_2_aligned()
12446                            std::int32_t result_stride) {  in gemm_q8_2_1_2_aligned()
12447   const std::int32_t row_chunks = m / 3;  in gemm_q8_2_1_2_aligned()
12448   const std::int32_t col_chunks = n / 3;  in gemm_q8_2_1_2_aligned()
12449   const std::int32_t padded_k = ((k + 7) / 8) * 8;  in gemm_q8_2_1_2_aligned()
12450   const std::int32_t chunk_size = k * 3;  in gemm_q8_2_1_2_aligned()
12451   const std::int32_t zipped_chunk_size = (padded_k + 16) * 3;  in gemm_q8_2_1_2_aligned()
12452   const std::int32_t zipped_rhs_size = (padded_k + 16) * n;  in gemm_q8_2_1_2_aligned()
12456   std::int32_t* zipped_lhs_3_offsets =  in gemm_q8_2_1_2_aligned()
12457       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 3);  in gemm_q8_2_1_2_aligned()
12458   std::int32_t* zipped_lhs_2_offsets =  in gemm_q8_2_1_2_aligned()
12459       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 2);  in gemm_q8_2_1_2_aligned()
12462   const std::int32_t result_chunk_stride = result_stride * 3;  in gemm_q8_2_1_2_aligned()
12464   const std::int32_t const_offset = lhs_offset * rhs_offset * k + result_offset;  in gemm_q8_2_1_2_aligned()
12465   const std::int32_t rounding_offset = (1 << (shift - 1));  in gemm_q8_2_1_2_aligned()
12466   std::int32_t* temp_result = reinterpret_cast<std::int32_t*>(  in gemm_q8_2_1_2_aligned()
12469   std::int32_t* mul_result_chunk = temp_result;  in gemm_q8_2_1_2_aligned()
12470   const std::int32_t mul_result_chunk_stride_bytes = ((n * 4 + 7) / 8) * 8;  in gemm_q8_2_1_2_aligned()
12515                            const std::uint8_t* rhs, std::int32_t m,  in gemm_q8_2_1_3_aligned()
12516                            std::int32_t n, std::int32_t k,  in gemm_q8_2_1_3_aligned()
12517                            std::int32_t lhs_offset, std::int32_t rhs_offset,  in gemm_q8_2_1_3_aligned()
12518                            std::int32_t result_offset,  in gemm_q8_2_1_3_aligned()
12519                            std::int32_t multiplicative_offset,  in gemm_q8_2_1_3_aligned()
12520                            std::int32_t shift, std::uint8_t* result,  in gemm_q8_2_1_3_aligned()
12521                            std::int32_t result_stride) {  in gemm_q8_2_1_3_aligned()
12522   const std::int32_t row_chunks = m / 3;  in gemm_q8_2_1_3_aligned()
12523   const std::int32_t col_chunks = n / 3;  in gemm_q8_2_1_3_aligned()
12524   const std::int32_t padded_k = ((k + 7) / 8) * 8;  in gemm_q8_2_1_3_aligned()
12525   const std::int32_t chunk_size = k * 3;  in gemm_q8_2_1_3_aligned()
12526   const std::int32_t zipped_chunk_size = (padded_k + 16) * 3;  in gemm_q8_2_1_3_aligned()
12527   const std::int32_t zipped_rhs_size = (padded_k + 16) * n;  in gemm_q8_2_1_3_aligned()
12531   std::int32_t* zipped_lhs_3_offsets =  in gemm_q8_2_1_3_aligned()
12532       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 3);  in gemm_q8_2_1_3_aligned()
12533   std::int32_t* zipped_lhs_2_offsets =  in gemm_q8_2_1_3_aligned()
12534       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 2);  in gemm_q8_2_1_3_aligned()
12537   const std::int32_t result_chunk_stride = result_stride * 3;  in gemm_q8_2_1_3_aligned()
12539   const std::int32_t const_offset = lhs_offset * rhs_offset * k + result_offset;  in gemm_q8_2_1_3_aligned()
12540   const std::int32_t rounding_offset = (1 << (shift - 1));  in gemm_q8_2_1_3_aligned()
12541   std::int32_t* temp_result = reinterpret_cast<std::int32_t*>(  in gemm_q8_2_1_3_aligned()
12544   std::int32_t* mul_result_chunk = temp_result;  in gemm_q8_2_1_3_aligned()
12545   const std::int32_t mul_result_chunk_stride_bytes = ((n * 4 + 7) / 8) * 8;  in gemm_q8_2_1_3_aligned()
12590                            const std::uint8_t* rhs, std::int32_t m,  in gemm_q8_2_1_4_aligned()
12591                            std::int32_t n, std::int32_t k,  in gemm_q8_2_1_4_aligned()
12592                            std::int32_t lhs_offset, std::int32_t rhs_offset,  in gemm_q8_2_1_4_aligned()
12593                            std::int32_t result_offset,  in gemm_q8_2_1_4_aligned()
12594                            std::int32_t multiplicative_offset,  in gemm_q8_2_1_4_aligned()
12595                            std::int32_t shift, std::uint8_t* result,  in gemm_q8_2_1_4_aligned()
12596                            std::int32_t result_stride) {  in gemm_q8_2_1_4_aligned()
12597   const std::int32_t row_chunks = m / 3;  in gemm_q8_2_1_4_aligned()
12598   const std::int32_t col_chunks = n / 3;  in gemm_q8_2_1_4_aligned()
12599   const std::int32_t padded_k = ((k + 7) / 8) * 8;  in gemm_q8_2_1_4_aligned()
12600   const std::int32_t chunk_size = k * 3;  in gemm_q8_2_1_4_aligned()
12601   const std::int32_t zipped_chunk_size = (padded_k + 16) * 3;  in gemm_q8_2_1_4_aligned()
12602   const std::int32_t zipped_rhs_size = (padded_k + 16) * n;  in gemm_q8_2_1_4_aligned()
12606   std::int32_t* zipped_lhs_3_offsets =  in gemm_q8_2_1_4_aligned()
12607       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 3);  in gemm_q8_2_1_4_aligned()
12608   std::int32_t* zipped_lhs_2_offsets =  in gemm_q8_2_1_4_aligned()
12609       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 2);  in gemm_q8_2_1_4_aligned()
12612   const std::int32_t result_chunk_stride = result_stride * 3;  in gemm_q8_2_1_4_aligned()
12614   const std::int32_t const_offset = lhs_offset * rhs_offset * k + result_offset;  in gemm_q8_2_1_4_aligned()
12615   const std::int32_t rounding_offset = (1 << (shift - 1));  in gemm_q8_2_1_4_aligned()
12616   std::int32_t* temp_result = reinterpret_cast<std::int32_t*>(  in gemm_q8_2_1_4_aligned()
12619   std::int32_t* mul_result_chunk = temp_result;  in gemm_q8_2_1_4_aligned()
12620   const std::int32_t mul_result_chunk_stride_bytes = ((n * 4 + 7) / 8) * 8;  in gemm_q8_2_1_4_aligned()
12665                            const std::uint8_t* rhs, std::int32_t m,  in gemm_q8_2_1_5_aligned()
12666                            std::int32_t n, std::int32_t k,  in gemm_q8_2_1_5_aligned()
12667                            std::int32_t lhs_offset, std::int32_t rhs_offset,  in gemm_q8_2_1_5_aligned()
12668                            std::int32_t result_offset,  in gemm_q8_2_1_5_aligned()
12669                            std::int32_t multiplicative_offset,  in gemm_q8_2_1_5_aligned()
12670                            std::int32_t shift, std::uint8_t* result,  in gemm_q8_2_1_5_aligned()
12671                            std::int32_t result_stride) {  in gemm_q8_2_1_5_aligned()
12672   const std::int32_t row_chunks = m / 3;  in gemm_q8_2_1_5_aligned()
12673   const std::int32_t col_chunks = n / 3;  in gemm_q8_2_1_5_aligned()
12674   const std::int32_t padded_k = ((k + 7) / 8) * 8;  in gemm_q8_2_1_5_aligned()
12675   const std::int32_t chunk_size = k * 3;  in gemm_q8_2_1_5_aligned()
12676   const std::int32_t zipped_chunk_size = (padded_k + 16) * 3;  in gemm_q8_2_1_5_aligned()
12677   const std::int32_t zipped_rhs_size = (padded_k + 16) * n;  in gemm_q8_2_1_5_aligned()
12681   std::int32_t* zipped_lhs_3_offsets =  in gemm_q8_2_1_5_aligned()
12682       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 3);  in gemm_q8_2_1_5_aligned()
12683   std::int32_t* zipped_lhs_2_offsets =  in gemm_q8_2_1_5_aligned()
12684       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 2);  in gemm_q8_2_1_5_aligned()
12687   const std::int32_t result_chunk_stride = result_stride * 3;  in gemm_q8_2_1_5_aligned()
12689   const std::int32_t const_offset = lhs_offset * rhs_offset * k + result_offset;  in gemm_q8_2_1_5_aligned()
12690   const std::int32_t rounding_offset = (1 << (shift - 1));  in gemm_q8_2_1_5_aligned()
12691   std::int32_t* temp_result = reinterpret_cast<std::int32_t*>(  in gemm_q8_2_1_5_aligned()
12694   std::int32_t* mul_result_chunk = temp_result;  in gemm_q8_2_1_5_aligned()
12695   const std::int32_t mul_result_chunk_stride_bytes = ((n * 4 + 7) / 8) * 8;  in gemm_q8_2_1_5_aligned()
12740                            const std::uint8_t* rhs, std::int32_t m,  in gemm_q8_2_1_6_aligned()
12741                            std::int32_t n, std::int32_t k,  in gemm_q8_2_1_6_aligned()
12742                            std::int32_t lhs_offset, std::int32_t rhs_offset,  in gemm_q8_2_1_6_aligned()
12743                            std::int32_t result_offset,  in gemm_q8_2_1_6_aligned()
12744                            std::int32_t multiplicative_offset,  in gemm_q8_2_1_6_aligned()
12745                            std::int32_t shift, std::uint8_t* result,  in gemm_q8_2_1_6_aligned()
12746                            std::int32_t result_stride) {  in gemm_q8_2_1_6_aligned()
12747   const std::int32_t row_chunks = m / 3;  in gemm_q8_2_1_6_aligned()
12748   const std::int32_t col_chunks = n / 3;  in gemm_q8_2_1_6_aligned()
12749   const std::int32_t padded_k = ((k + 7) / 8) * 8;  in gemm_q8_2_1_6_aligned()
12750   const std::int32_t chunk_size = k * 3;  in gemm_q8_2_1_6_aligned()
12751   const std::int32_t zipped_chunk_size = (padded_k + 16) * 3;  in gemm_q8_2_1_6_aligned()
12752   const std::int32_t zipped_rhs_size = (padded_k + 16) * n;  in gemm_q8_2_1_6_aligned()
12756   std::int32_t* zipped_lhs_3_offsets =  in gemm_q8_2_1_6_aligned()
12757       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 3);  in gemm_q8_2_1_6_aligned()
12758   std::int32_t* zipped_lhs_2_offsets =  in gemm_q8_2_1_6_aligned()
12759       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 2);  in gemm_q8_2_1_6_aligned()
12762   const std::int32_t result_chunk_stride = result_stride * 3;  in gemm_q8_2_1_6_aligned()
12764   const std::int32_t const_offset = lhs_offset * rhs_offset * k + result_offset;  in gemm_q8_2_1_6_aligned()
12765   const std::int32_t rounding_offset = (1 << (shift - 1));  in gemm_q8_2_1_6_aligned()
12766   std::int32_t* temp_result = reinterpret_cast<std::int32_t*>(  in gemm_q8_2_1_6_aligned()
12769   std::int32_t* mul_result_chunk = temp_result;  in gemm_q8_2_1_6_aligned()
12770   const std::int32_t mul_result_chunk_stride_bytes = ((n * 4 + 7) / 8) * 8;  in gemm_q8_2_1_6_aligned()
12815                            const std::uint8_t* rhs, std::int32_t m,  in gemm_q8_2_1_7_aligned()
12816                            std::int32_t n, std::int32_t k,  in gemm_q8_2_1_7_aligned()
12817                            std::int32_t lhs_offset, std::int32_t rhs_offset,  in gemm_q8_2_1_7_aligned()
12818                            std::int32_t result_offset,  in gemm_q8_2_1_7_aligned()
12819                            std::int32_t multiplicative_offset,  in gemm_q8_2_1_7_aligned()
12820                            std::int32_t shift, std::uint8_t* result,  in gemm_q8_2_1_7_aligned()
12821                            std::int32_t result_stride) {  in gemm_q8_2_1_7_aligned()
12822   const std::int32_t row_chunks = m / 3;  in gemm_q8_2_1_7_aligned()
12823   const std::int32_t col_chunks = n / 3;  in gemm_q8_2_1_7_aligned()
12824   const std::int32_t padded_k = ((k + 7) / 8) * 8;  in gemm_q8_2_1_7_aligned()
12825   const std::int32_t chunk_size = k * 3;  in gemm_q8_2_1_7_aligned()
12826   const std::int32_t zipped_chunk_size = (padded_k + 16) * 3;  in gemm_q8_2_1_7_aligned()
12827   const std::int32_t zipped_rhs_size = (padded_k + 16) * n;  in gemm_q8_2_1_7_aligned()
12831   std::int32_t* zipped_lhs_3_offsets =  in gemm_q8_2_1_7_aligned()
12832       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 3);  in gemm_q8_2_1_7_aligned()
12833   std::int32_t* zipped_lhs_2_offsets =  in gemm_q8_2_1_7_aligned()
12834       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 2);  in gemm_q8_2_1_7_aligned()
12837   const std::int32_t result_chunk_stride = result_stride * 3;  in gemm_q8_2_1_7_aligned()
12839   const std::int32_t const_offset = lhs_offset * rhs_offset * k + result_offset;  in gemm_q8_2_1_7_aligned()
12840   const std::int32_t rounding_offset = (1 << (shift - 1));  in gemm_q8_2_1_7_aligned()
12841   std::int32_t* temp_result = reinterpret_cast<std::int32_t*>(  in gemm_q8_2_1_7_aligned()
12844   std::int32_t* mul_result_chunk = temp_result;  in gemm_q8_2_1_7_aligned()
12845   const std::int32_t mul_result_chunk_stride_bytes = ((n * 4 + 7) / 8) * 8;  in gemm_q8_2_1_7_aligned()
12890                            const std::uint8_t* rhs, std::int32_t m,  in gemm_q8_2_2_0_aligned()
12891                            std::int32_t n, std::int32_t k,  in gemm_q8_2_2_0_aligned()
12892                            std::int32_t lhs_offset, std::int32_t rhs_offset,  in gemm_q8_2_2_0_aligned()
12893                            std::int32_t result_offset,  in gemm_q8_2_2_0_aligned()
12894                            std::int32_t multiplicative_offset,  in gemm_q8_2_2_0_aligned()
12895                            std::int32_t shift, std::uint8_t* result,  in gemm_q8_2_2_0_aligned()
12896                            std::int32_t result_stride) {  in gemm_q8_2_2_0_aligned()
12897   const std::int32_t row_chunks = m / 3;  in gemm_q8_2_2_0_aligned()
12898   const std::int32_t col_chunks = n / 3;  in gemm_q8_2_2_0_aligned()
12899   const std::int32_t padded_k = ((k + 7) / 8) * 8;  in gemm_q8_2_2_0_aligned()
12900   const std::int32_t chunk_size = k * 3;  in gemm_q8_2_2_0_aligned()
12901   const std::int32_t zipped_chunk_size = (padded_k + 16) * 3;  in gemm_q8_2_2_0_aligned()
12902   const std::int32_t zipped_rhs_size = (padded_k + 16) * n;  in gemm_q8_2_2_0_aligned()
12906   std::int32_t* zipped_lhs_3_offsets =  in gemm_q8_2_2_0_aligned()
12907       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 3);  in gemm_q8_2_2_0_aligned()
12908   std::int32_t* zipped_lhs_2_offsets =  in gemm_q8_2_2_0_aligned()
12909       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 2);  in gemm_q8_2_2_0_aligned()
12912   const std::int32_t result_chunk_stride = result_stride * 3;  in gemm_q8_2_2_0_aligned()
12914   const std::int32_t const_offset = lhs_offset * rhs_offset * k + result_offset;  in gemm_q8_2_2_0_aligned()
12915   const std::int32_t rounding_offset = (1 << (shift - 1));  in gemm_q8_2_2_0_aligned()
12916   std::int32_t* temp_result = reinterpret_cast<std::int32_t*>(  in gemm_q8_2_2_0_aligned()
12919   std::int32_t* mul_result_chunk = temp_result;  in gemm_q8_2_2_0_aligned()
12920   const std::int32_t mul_result_chunk_stride_bytes = ((n * 4 + 7) / 8) * 8;  in gemm_q8_2_2_0_aligned()
12965                            const std::uint8_t* rhs, std::int32_t m,  in gemm_q8_2_2_1_aligned()
12966                            std::int32_t n, std::int32_t k,  in gemm_q8_2_2_1_aligned()
12967                            std::int32_t lhs_offset, std::int32_t rhs_offset,  in gemm_q8_2_2_1_aligned()
12968                            std::int32_t result_offset,  in gemm_q8_2_2_1_aligned()
12969                            std::int32_t multiplicative_offset,  in gemm_q8_2_2_1_aligned()
12970                            std::int32_t shift, std::uint8_t* result,  in gemm_q8_2_2_1_aligned()
12971                            std::int32_t result_stride) {  in gemm_q8_2_2_1_aligned()
12972   const std::int32_t row_chunks = m / 3;  in gemm_q8_2_2_1_aligned()
12973   const std::int32_t col_chunks = n / 3;  in gemm_q8_2_2_1_aligned()
12974   const std::int32_t padded_k = ((k + 7) / 8) * 8;  in gemm_q8_2_2_1_aligned()
12975   const std::int32_t chunk_size = k * 3;  in gemm_q8_2_2_1_aligned()
12976   const std::int32_t zipped_chunk_size = (padded_k + 16) * 3;  in gemm_q8_2_2_1_aligned()
12977   const std::int32_t zipped_rhs_size = (padded_k + 16) * n;  in gemm_q8_2_2_1_aligned()
12981   std::int32_t* zipped_lhs_3_offsets =  in gemm_q8_2_2_1_aligned()
12982       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 3);  in gemm_q8_2_2_1_aligned()
12983   std::int32_t* zipped_lhs_2_offsets =  in gemm_q8_2_2_1_aligned()
12984       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 2);  in gemm_q8_2_2_1_aligned()
12987   const std::int32_t result_chunk_stride = result_stride * 3;  in gemm_q8_2_2_1_aligned()
12989   const std::int32_t const_offset = lhs_offset * rhs_offset * k + result_offset;  in gemm_q8_2_2_1_aligned()
12990   const std::int32_t rounding_offset = (1 << (shift - 1));  in gemm_q8_2_2_1_aligned()
12991   std::int32_t* temp_result = reinterpret_cast<std::int32_t*>(  in gemm_q8_2_2_1_aligned()
12994   std::int32_t* mul_result_chunk = temp_result;  in gemm_q8_2_2_1_aligned()
12995   const std::int32_t mul_result_chunk_stride_bytes = ((n * 4 + 7) / 8) * 8;  in gemm_q8_2_2_1_aligned()
13040                            const std::uint8_t* rhs, std::int32_t m,  in gemm_q8_2_2_2_aligned()
13041                            std::int32_t n, std::int32_t k,  in gemm_q8_2_2_2_aligned()
13042                            std::int32_t lhs_offset, std::int32_t rhs_offset,  in gemm_q8_2_2_2_aligned()
13043                            std::int32_t result_offset,  in gemm_q8_2_2_2_aligned()
13044                            std::int32_t multiplicative_offset,  in gemm_q8_2_2_2_aligned()
13045                            std::int32_t shift, std::uint8_t* result,  in gemm_q8_2_2_2_aligned()
13046                            std::int32_t result_stride) {  in gemm_q8_2_2_2_aligned()
13047   const std::int32_t row_chunks = m / 3;  in gemm_q8_2_2_2_aligned()
13048   const std::int32_t col_chunks = n / 3;  in gemm_q8_2_2_2_aligned()
13049   const std::int32_t padded_k = ((k + 7) / 8) * 8;  in gemm_q8_2_2_2_aligned()
13050   const std::int32_t chunk_size = k * 3;  in gemm_q8_2_2_2_aligned()
13051   const std::int32_t zipped_chunk_size = (padded_k + 16) * 3;  in gemm_q8_2_2_2_aligned()
13052   const std::int32_t zipped_rhs_size = (padded_k + 16) * n;  in gemm_q8_2_2_2_aligned()
13056   std::int32_t* zipped_lhs_3_offsets =  in gemm_q8_2_2_2_aligned()
13057       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 3);  in gemm_q8_2_2_2_aligned()
13058   std::int32_t* zipped_lhs_2_offsets =  in gemm_q8_2_2_2_aligned()
13059       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 2);  in gemm_q8_2_2_2_aligned()
13062   const std::int32_t result_chunk_stride = result_stride * 3;  in gemm_q8_2_2_2_aligned()
13064   const std::int32_t const_offset = lhs_offset * rhs_offset * k + result_offset;  in gemm_q8_2_2_2_aligned()
13065   const std::int32_t rounding_offset = (1 << (shift - 1));  in gemm_q8_2_2_2_aligned()
13066   std::int32_t* temp_result = reinterpret_cast<std::int32_t*>(  in gemm_q8_2_2_2_aligned()
13069   std::int32_t* mul_result_chunk = temp_result;  in gemm_q8_2_2_2_aligned()
13070   const std::int32_t mul_result_chunk_stride_bytes = ((n * 4 + 7) / 8) * 8;  in gemm_q8_2_2_2_aligned()
13115                            const std::uint8_t* rhs, std::int32_t m,  in gemm_q8_2_2_3_aligned()
13116                            std::int32_t n, std::int32_t k,  in gemm_q8_2_2_3_aligned()
13117                            std::int32_t lhs_offset, std::int32_t rhs_offset,  in gemm_q8_2_2_3_aligned()
13118                            std::int32_t result_offset,  in gemm_q8_2_2_3_aligned()
13119                            std::int32_t multiplicative_offset,  in gemm_q8_2_2_3_aligned()
13120                            std::int32_t shift, std::uint8_t* result,  in gemm_q8_2_2_3_aligned()
13121                            std::int32_t result_stride) {  in gemm_q8_2_2_3_aligned()
13122   const std::int32_t row_chunks = m / 3;  in gemm_q8_2_2_3_aligned()
13123   const std::int32_t col_chunks = n / 3;  in gemm_q8_2_2_3_aligned()
13124   const std::int32_t padded_k = ((k + 7) / 8) * 8;  in gemm_q8_2_2_3_aligned()
13125   const std::int32_t chunk_size = k * 3;  in gemm_q8_2_2_3_aligned()
13126   const std::int32_t zipped_chunk_size = (padded_k + 16) * 3;  in gemm_q8_2_2_3_aligned()
13127   const std::int32_t zipped_rhs_size = (padded_k + 16) * n;  in gemm_q8_2_2_3_aligned()
13131   std::int32_t* zipped_lhs_3_offsets =  in gemm_q8_2_2_3_aligned()
13132       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 3);  in gemm_q8_2_2_3_aligned()
13133   std::int32_t* zipped_lhs_2_offsets =  in gemm_q8_2_2_3_aligned()
13134       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 2);  in gemm_q8_2_2_3_aligned()
13137   const std::int32_t result_chunk_stride = result_stride * 3;  in gemm_q8_2_2_3_aligned()
13139   const std::int32_t const_offset = lhs_offset * rhs_offset * k + result_offset;  in gemm_q8_2_2_3_aligned()
13140   const std::int32_t rounding_offset = (1 << (shift - 1));  in gemm_q8_2_2_3_aligned()
13141   std::int32_t* temp_result = reinterpret_cast<std::int32_t*>(  in gemm_q8_2_2_3_aligned()
13144   std::int32_t* mul_result_chunk = temp_result;  in gemm_q8_2_2_3_aligned()
13145   const std::int32_t mul_result_chunk_stride_bytes = ((n * 4 + 7) / 8) * 8;  in gemm_q8_2_2_3_aligned()
13190                            const std::uint8_t* rhs, std::int32_t m,  in gemm_q8_2_2_4_aligned()
13191                            std::int32_t n, std::int32_t k,  in gemm_q8_2_2_4_aligned()
13192                            std::int32_t lhs_offset, std::int32_t rhs_offset,  in gemm_q8_2_2_4_aligned()
13193                            std::int32_t result_offset,  in gemm_q8_2_2_4_aligned()
13194                            std::int32_t multiplicative_offset,  in gemm_q8_2_2_4_aligned()
13195                            std::int32_t shift, std::uint8_t* result,  in gemm_q8_2_2_4_aligned()
13196                            std::int32_t result_stride) {  in gemm_q8_2_2_4_aligned()
13197   const std::int32_t row_chunks = m / 3;  in gemm_q8_2_2_4_aligned()
13198   const std::int32_t col_chunks = n / 3;  in gemm_q8_2_2_4_aligned()
13199   const std::int32_t padded_k = ((k + 7) / 8) * 8;  in gemm_q8_2_2_4_aligned()
13200   const std::int32_t chunk_size = k * 3;  in gemm_q8_2_2_4_aligned()
13201   const std::int32_t zipped_chunk_size = (padded_k + 16) * 3;  in gemm_q8_2_2_4_aligned()
13202   const std::int32_t zipped_rhs_size = (padded_k + 16) * n;  in gemm_q8_2_2_4_aligned()
13206   std::int32_t* zipped_lhs_3_offsets =  in gemm_q8_2_2_4_aligned()
13207       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 3);  in gemm_q8_2_2_4_aligned()
13208   std::int32_t* zipped_lhs_2_offsets =  in gemm_q8_2_2_4_aligned()
13209       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 2);  in gemm_q8_2_2_4_aligned()
13212   const std::int32_t result_chunk_stride = result_stride * 3;  in gemm_q8_2_2_4_aligned()
13214   const std::int32_t const_offset = lhs_offset * rhs_offset * k + result_offset;  in gemm_q8_2_2_4_aligned()
13215   const std::int32_t rounding_offset = (1 << (shift - 1));  in gemm_q8_2_2_4_aligned()
13216   std::int32_t* temp_result = reinterpret_cast<std::int32_t*>(  in gemm_q8_2_2_4_aligned()
13219   std::int32_t* mul_result_chunk = temp_result;  in gemm_q8_2_2_4_aligned()
13220   const std::int32_t mul_result_chunk_stride_bytes = ((n * 4 + 7) / 8) * 8;  in gemm_q8_2_2_4_aligned()
13265                            const std::uint8_t* rhs, std::int32_t m,  in gemm_q8_2_2_5_aligned()
13266                            std::int32_t n, std::int32_t k,  in gemm_q8_2_2_5_aligned()
13267                            std::int32_t lhs_offset, std::int32_t rhs_offset,  in gemm_q8_2_2_5_aligned()
13268                            std::int32_t result_offset,  in gemm_q8_2_2_5_aligned()
13269                            std::int32_t multiplicative_offset,  in gemm_q8_2_2_5_aligned()
13270                            std::int32_t shift, std::uint8_t* result,  in gemm_q8_2_2_5_aligned()
13271                            std::int32_t result_stride) {  in gemm_q8_2_2_5_aligned()
13272   const std::int32_t row_chunks = m / 3;  in gemm_q8_2_2_5_aligned()
13273   const std::int32_t col_chunks = n / 3;  in gemm_q8_2_2_5_aligned()
13274   const std::int32_t padded_k = ((k + 7) / 8) * 8;  in gemm_q8_2_2_5_aligned()
13275   const std::int32_t chunk_size = k * 3;  in gemm_q8_2_2_5_aligned()
13276   const std::int32_t zipped_chunk_size = (padded_k + 16) * 3;  in gemm_q8_2_2_5_aligned()
13277   const std::int32_t zipped_rhs_size = (padded_k + 16) * n;  in gemm_q8_2_2_5_aligned()
13281   std::int32_t* zipped_lhs_3_offsets =  in gemm_q8_2_2_5_aligned()
13282       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 3);  in gemm_q8_2_2_5_aligned()
13283   std::int32_t* zipped_lhs_2_offsets =  in gemm_q8_2_2_5_aligned()
13284       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 2);  in gemm_q8_2_2_5_aligned()
13287   const std::int32_t result_chunk_stride = result_stride * 3;  in gemm_q8_2_2_5_aligned()
13289   const std::int32_t const_offset = lhs_offset * rhs_offset * k + result_offset;  in gemm_q8_2_2_5_aligned()
13290   const std::int32_t rounding_offset = (1 << (shift - 1));  in gemm_q8_2_2_5_aligned()
13291   std::int32_t* temp_result = reinterpret_cast<std::int32_t*>(  in gemm_q8_2_2_5_aligned()
13294   std::int32_t* mul_result_chunk = temp_result;  in gemm_q8_2_2_5_aligned()
13295   const std::int32_t mul_result_chunk_stride_bytes = ((n * 4 + 7) / 8) * 8;  in gemm_q8_2_2_5_aligned()
13340                            const std::uint8_t* rhs, std::int32_t m,  in gemm_q8_2_2_6_aligned()
13341                            std::int32_t n, std::int32_t k,  in gemm_q8_2_2_6_aligned()
13342                            std::int32_t lhs_offset, std::int32_t rhs_offset,  in gemm_q8_2_2_6_aligned()
13343                            std::int32_t result_offset,  in gemm_q8_2_2_6_aligned()
13344                            std::int32_t multiplicative_offset,  in gemm_q8_2_2_6_aligned()
13345                            std::int32_t shift, std::uint8_t* result,  in gemm_q8_2_2_6_aligned()
13346                            std::int32_t result_stride) {  in gemm_q8_2_2_6_aligned()
13347   const std::int32_t row_chunks = m / 3;  in gemm_q8_2_2_6_aligned()
13348   const std::int32_t col_chunks = n / 3;  in gemm_q8_2_2_6_aligned()
13349   const std::int32_t padded_k = ((k + 7) / 8) * 8;  in gemm_q8_2_2_6_aligned()
13350   const std::int32_t chunk_size = k * 3;  in gemm_q8_2_2_6_aligned()
13351   const std::int32_t zipped_chunk_size = (padded_k + 16) * 3;  in gemm_q8_2_2_6_aligned()
13352   const std::int32_t zipped_rhs_size = (padded_k + 16) * n;  in gemm_q8_2_2_6_aligned()
13356   std::int32_t* zipped_lhs_3_offsets =  in gemm_q8_2_2_6_aligned()
13357       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 3);  in gemm_q8_2_2_6_aligned()
13358   std::int32_t* zipped_lhs_2_offsets =  in gemm_q8_2_2_6_aligned()
13359       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 2);  in gemm_q8_2_2_6_aligned()
13362   const std::int32_t result_chunk_stride = result_stride * 3;  in gemm_q8_2_2_6_aligned()
13364   const std::int32_t const_offset = lhs_offset * rhs_offset * k + result_offset;  in gemm_q8_2_2_6_aligned()
13365   const std::int32_t rounding_offset = (1 << (shift - 1));  in gemm_q8_2_2_6_aligned()
13366   std::int32_t* temp_result = reinterpret_cast<std::int32_t*>(  in gemm_q8_2_2_6_aligned()
13369   std::int32_t* mul_result_chunk = temp_result;  in gemm_q8_2_2_6_aligned()
13370   const std::int32_t mul_result_chunk_stride_bytes = ((n * 4 + 7) / 8) * 8;  in gemm_q8_2_2_6_aligned()
13415                            const std::uint8_t* rhs, std::int32_t m,  in gemm_q8_2_2_7_aligned()
13416                            std::int32_t n, std::int32_t k,  in gemm_q8_2_2_7_aligned()
13417                            std::int32_t lhs_offset, std::int32_t rhs_offset,  in gemm_q8_2_2_7_aligned()
13418                            std::int32_t result_offset,  in gemm_q8_2_2_7_aligned()
13419                            std::int32_t multiplicative_offset,  in gemm_q8_2_2_7_aligned()
13420                            std::int32_t shift, std::uint8_t* result,  in gemm_q8_2_2_7_aligned()
13421                            std::int32_t result_stride) {  in gemm_q8_2_2_7_aligned()
13422   const std::int32_t row_chunks = m / 3;  in gemm_q8_2_2_7_aligned()
13423   const std::int32_t col_chunks = n / 3;  in gemm_q8_2_2_7_aligned()
13424   const std::int32_t padded_k = ((k + 7) / 8) * 8;  in gemm_q8_2_2_7_aligned()
13425   const std::int32_t chunk_size = k * 3;  in gemm_q8_2_2_7_aligned()
13426   const std::int32_t zipped_chunk_size = (padded_k + 16) * 3;  in gemm_q8_2_2_7_aligned()
13427   const std::int32_t zipped_rhs_size = (padded_k + 16) * n;  in gemm_q8_2_2_7_aligned()
13431   std::int32_t* zipped_lhs_3_offsets =  in gemm_q8_2_2_7_aligned()
13432       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 3);  in gemm_q8_2_2_7_aligned()
13433   std::int32_t* zipped_lhs_2_offsets =  in gemm_q8_2_2_7_aligned()
13434       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 2);  in gemm_q8_2_2_7_aligned()
13437   const std::int32_t result_chunk_stride = result_stride * 3;  in gemm_q8_2_2_7_aligned()
13439   const std::int32_t const_offset = lhs_offset * rhs_offset * k + result_offset;  in gemm_q8_2_2_7_aligned()
13440   const std::int32_t rounding_offset = (1 << (shift - 1));  in gemm_q8_2_2_7_aligned()
13441   std::int32_t* temp_result = reinterpret_cast<std::int32_t*>(  in gemm_q8_2_2_7_aligned()
13444   std::int32_t* mul_result_chunk = temp_result;  in gemm_q8_2_2_7_aligned()
13445   const std::int32_t mul_result_chunk_stride_bytes = ((n * 4 + 7) / 8) * 8;  in gemm_q8_2_2_7_aligned()
13490                    const std::uint8_t* rhs, std::int32_t m, std::int32_t n,  in gemm_q8_0_0_0()
13491                    std::int32_t k, std::int32_t lhs_offset,  in gemm_q8_0_0_0()
13492                    std::int32_t rhs_offset, std::int32_t result_offset,  in gemm_q8_0_0_0()
13493                    std::int32_t multiplicative_offset, std::int32_t shift,  in gemm_q8_0_0_0()
13494                    std::uint8_t* result, std::int32_t result_stride) {  in gemm_q8_0_0_0()
13495   const std::int32_t row_chunks = m / 3;  in gemm_q8_0_0_0()
13496   const std::int32_t col_chunks = n / 3;  in gemm_q8_0_0_0()
13497   const std::int32_t padded_k = ((k + 7) / 8) * 8;  in gemm_q8_0_0_0()
13498   const std::int32_t chunk_size = k * 3;  in gemm_q8_0_0_0()
13499   const std::int32_t zipped_chunk_size = (padded_k + 16) * 3;  in gemm_q8_0_0_0()
13500   const std::int32_t zipped_rhs_size = (padded_k + 16) * n;  in gemm_q8_0_0_0()
13504   std::int32_t* zipped_lhs_3_offsets =  in gemm_q8_0_0_0()
13505       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 3);  in gemm_q8_0_0_0()
13508   const std::int32_t result_chunk_stride = result_stride * 3;  in gemm_q8_0_0_0()
13510   const std::int32_t const_offset = lhs_offset * rhs_offset * k + result_offset;  in gemm_q8_0_0_0()
13511   const std::int32_t rounding_offset = (1 << (shift - 1));  in gemm_q8_0_0_0()
13512   std::int32_t* temp_result = reinterpret_cast<std::int32_t*>(  in gemm_q8_0_0_0()
13515   std::int32_t* mul_result_chunk = temp_result;  in gemm_q8_0_0_0()
13516   const std::int32_t mul_result_chunk_stride_bytes = ((n * 4 + 7) / 8) * 8;  in gemm_q8_0_0_0()
13543                    const std::uint8_t* rhs, std::int32_t m, std::int32_t n,  in gemm_q8_0_0_1()
13544                    std::int32_t k, std::int32_t lhs_offset,  in gemm_q8_0_0_1()
13545                    std::int32_t rhs_offset, std::int32_t result_offset,  in gemm_q8_0_0_1()
13546                    std::int32_t multiplicative_offset, std::int32_t shift,  in gemm_q8_0_0_1()
13547                    std::uint8_t* result, std::int32_t result_stride) {  in gemm_q8_0_0_1()
13548   const std::int32_t row_chunks = m / 3;  in gemm_q8_0_0_1()
13549   const std::int32_t col_chunks = n / 3;  in gemm_q8_0_0_1()
13550   const std::int32_t padded_k = ((k + 7) / 8) * 8;  in gemm_q8_0_0_1()
13551   const std::int32_t chunk_size = k * 3;  in gemm_q8_0_0_1()
13552   const std::int32_t zipped_chunk_size = (padded_k + 16) * 3;  in gemm_q8_0_0_1()
13553   const std::int32_t zipped_rhs_size = (padded_k + 16) * n;  in gemm_q8_0_0_1()
13557   std::int32_t* zipped_lhs_3_offsets =  in gemm_q8_0_0_1()
13558       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 3);  in gemm_q8_0_0_1()
13561   const std::int32_t result_chunk_stride = result_stride * 3;  in gemm_q8_0_0_1()
13563   const std::int32_t const_offset = lhs_offset * rhs_offset * k + result_offset;  in gemm_q8_0_0_1()
13564   const std::int32_t rounding_offset = (1 << (shift - 1));  in gemm_q8_0_0_1()
13565   std::int32_t* temp_result = reinterpret_cast<std::int32_t*>(  in gemm_q8_0_0_1()
13568   std::int32_t* mul_result_chunk = temp_result;  in gemm_q8_0_0_1()
13569   const std::int32_t mul_result_chunk_stride_bytes = ((n * 4 + 7) / 8) * 8;  in gemm_q8_0_0_1()
13596                    const std::uint8_t* rhs, std::int32_t m, std::int32_t n,  in gemm_q8_0_0_2()
13597                    std::int32_t k, std::int32_t lhs_offset,  in gemm_q8_0_0_2()
13598                    std::int32_t rhs_offset, std::int32_t result_offset,  in gemm_q8_0_0_2()
13599                    std::int32_t multiplicative_offset, std::int32_t shift,  in gemm_q8_0_0_2()
13600                    std::uint8_t* result, std::int32_t result_stride) {  in gemm_q8_0_0_2()
13601   const std::int32_t row_chunks = m / 3;  in gemm_q8_0_0_2()
13602   const std::int32_t col_chunks = n / 3;  in gemm_q8_0_0_2()
13603   const std::int32_t padded_k = ((k + 7) / 8) * 8;  in gemm_q8_0_0_2()
13604   const std::int32_t chunk_size = k * 3;  in gemm_q8_0_0_2()
13605   const std::int32_t zipped_chunk_size = (padded_k + 16) * 3;  in gemm_q8_0_0_2()
13606   const std::int32_t zipped_rhs_size = (padded_k + 16) * n;  in gemm_q8_0_0_2()
13610   std::int32_t* zipped_lhs_3_offsets =  in gemm_q8_0_0_2()
13611       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 3);  in gemm_q8_0_0_2()
13614   const std::int32_t result_chunk_stride = result_stride * 3;  in gemm_q8_0_0_2()
13616   const std::int32_t const_offset = lhs_offset * rhs_offset * k + result_offset;  in gemm_q8_0_0_2()
13617   const std::int32_t rounding_offset = (1 << (shift - 1));  in gemm_q8_0_0_2()
13618   std::int32_t* temp_result = reinterpret_cast<std::int32_t*>(  in gemm_q8_0_0_2()
13621   std::int32_t* mul_result_chunk = temp_result;  in gemm_q8_0_0_2()
13622   const std::int32_t mul_result_chunk_stride_bytes = ((n * 4 + 7) / 8) * 8;  in gemm_q8_0_0_2()
13649                    const std::uint8_t* rhs, std::int32_t m, std::int32_t n,  in gemm_q8_0_0_3()
13650                    std::int32_t k, std::int32_t lhs_offset,  in gemm_q8_0_0_3()
13651                    std::int32_t rhs_offset, std::int32_t result_offset,  in gemm_q8_0_0_3()
13652                    std::int32_t multiplicative_offset, std::int32_t shift,  in gemm_q8_0_0_3()
13653                    std::uint8_t* result, std::int32_t result_stride) {  in gemm_q8_0_0_3()
13654   const std::int32_t row_chunks = m / 3;  in gemm_q8_0_0_3()
13655   const std::int32_t col_chunks = n / 3;  in gemm_q8_0_0_3()
13656   const std::int32_t padded_k = ((k + 7) / 8) * 8;  in gemm_q8_0_0_3()
13657   const std::int32_t chunk_size = k * 3;  in gemm_q8_0_0_3()
13658   const std::int32_t zipped_chunk_size = (padded_k + 16) * 3;  in gemm_q8_0_0_3()
13659   const std::int32_t zipped_rhs_size = (padded_k + 16) * n;  in gemm_q8_0_0_3()
13663   std::int32_t* zipped_lhs_3_offsets =  in gemm_q8_0_0_3()
13664       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 3);  in gemm_q8_0_0_3()
13667   const std::int32_t result_chunk_stride = result_stride * 3;  in gemm_q8_0_0_3()
13669   const std::int32_t const_offset = lhs_offset * rhs_offset * k + result_offset;  in gemm_q8_0_0_3()
13670   const std::int32_t rounding_offset = (1 << (shift - 1));  in gemm_q8_0_0_3()
13671   std::int32_t* temp_result = reinterpret_cast<std::int32_t*>(  in gemm_q8_0_0_3()
13674   std::int32_t* mul_result_chunk = temp_result;  in gemm_q8_0_0_3()
13675   const std::int32_t mul_result_chunk_stride_bytes = ((n * 4 + 7) / 8) * 8;  in gemm_q8_0_0_3()
13702                    const std::uint8_t* rhs, std::int32_t m, std::int32_t n,  in gemm_q8_0_0_4()
13703                    std::int32_t k, std::int32_t lhs_offset,  in gemm_q8_0_0_4()
13704                    std::int32_t rhs_offset, std::int32_t result_offset,  in gemm_q8_0_0_4()
13705                    std::int32_t multiplicative_offset, std::int32_t shift,  in gemm_q8_0_0_4()
13706                    std::uint8_t* result, std::int32_t result_stride) {  in gemm_q8_0_0_4()
13707   const std::int32_t row_chunks = m / 3;  in gemm_q8_0_0_4()
13708   const std::int32_t col_chunks = n / 3;  in gemm_q8_0_0_4()
13709   const std::int32_t padded_k = ((k + 7) / 8) * 8;  in gemm_q8_0_0_4()
13710   const std::int32_t chunk_size = k * 3;  in gemm_q8_0_0_4()
13711   const std::int32_t zipped_chunk_size = (padded_k + 16) * 3;  in gemm_q8_0_0_4()
13712   const std::int32_t zipped_rhs_size = (padded_k + 16) * n;  in gemm_q8_0_0_4()
13716   std::int32_t* zipped_lhs_3_offsets =  in gemm_q8_0_0_4()
13717       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 3);  in gemm_q8_0_0_4()
13720   const std::int32_t result_chunk_stride = result_stride * 3;  in gemm_q8_0_0_4()
13722   const std::int32_t const_offset = lhs_offset * rhs_offset * k + result_offset;  in gemm_q8_0_0_4()
13723   const std::int32_t rounding_offset = (1 << (shift - 1));  in gemm_q8_0_0_4()
13724   std::int32_t* temp_result = reinterpret_cast<std::int32_t*>(  in gemm_q8_0_0_4()
13727   std::int32_t* mul_result_chunk = temp_result;  in gemm_q8_0_0_4()
13728   const std::int32_t mul_result_chunk_stride_bytes = ((n * 4 + 7) / 8) * 8;  in gemm_q8_0_0_4()
13755                    const std::uint8_t* rhs, std::int32_t m, std::int32_t n,  in gemm_q8_0_0_5()
13756                    std::int32_t k, std::int32_t lhs_offset,  in gemm_q8_0_0_5()
13757                    std::int32_t rhs_offset, std::int32_t result_offset,  in gemm_q8_0_0_5()
13758                    std::int32_t multiplicative_offset, std::int32_t shift,  in gemm_q8_0_0_5()
13759                    std::uint8_t* result, std::int32_t result_stride) {  in gemm_q8_0_0_5()
13760   const std::int32_t row_chunks = m / 3;  in gemm_q8_0_0_5()
13761   const std::int32_t col_chunks = n / 3;  in gemm_q8_0_0_5()
13762   const std::int32_t padded_k = ((k + 7) / 8) * 8;  in gemm_q8_0_0_5()
13763   const std::int32_t chunk_size = k * 3;  in gemm_q8_0_0_5()
13764   const std::int32_t zipped_chunk_size = (padded_k + 16) * 3;  in gemm_q8_0_0_5()
13765   const std::int32_t zipped_rhs_size = (padded_k + 16) * n;  in gemm_q8_0_0_5()
13769   std::int32_t* zipped_lhs_3_offsets =  in gemm_q8_0_0_5()
13770       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 3);  in gemm_q8_0_0_5()
13773   const std::int32_t result_chunk_stride = result_stride * 3;  in gemm_q8_0_0_5()
13775   const std::int32_t const_offset = lhs_offset * rhs_offset * k + result_offset;  in gemm_q8_0_0_5()
13776   const std::int32_t rounding_offset = (1 << (shift - 1));  in gemm_q8_0_0_5()
13777   std::int32_t* temp_result = reinterpret_cast<std::int32_t*>(  in gemm_q8_0_0_5()
13780   std::int32_t* mul_result_chunk = temp_result;  in gemm_q8_0_0_5()
13781   const std::int32_t mul_result_chunk_stride_bytes = ((n * 4 + 7) / 8) * 8;  in gemm_q8_0_0_5()
13808                    const std::uint8_t* rhs, std::int32_t m, std::int32_t n,  in gemm_q8_0_0_6()
13809                    std::int32_t k, std::int32_t lhs_offset,  in gemm_q8_0_0_6()
13810                    std::int32_t rhs_offset, std::int32_t result_offset,  in gemm_q8_0_0_6()
13811                    std::int32_t multiplicative_offset, std::int32_t shift,  in gemm_q8_0_0_6()
13812                    std::uint8_t* result, std::int32_t result_stride) {  in gemm_q8_0_0_6()
13813   const std::int32_t row_chunks = m / 3;  in gemm_q8_0_0_6()
13814   const std::int32_t col_chunks = n / 3;  in gemm_q8_0_0_6()
13815   const std::int32_t padded_k = ((k + 7) / 8) * 8;  in gemm_q8_0_0_6()
13816   const std::int32_t chunk_size = k * 3;  in gemm_q8_0_0_6()
13817   const std::int32_t zipped_chunk_size = (padded_k + 16) * 3;  in gemm_q8_0_0_6()
13818   const std::int32_t zipped_rhs_size = (padded_k + 16) * n;  in gemm_q8_0_0_6()
13822   std::int32_t* zipped_lhs_3_offsets =  in gemm_q8_0_0_6()
13823       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 3);  in gemm_q8_0_0_6()
13826   const std::int32_t result_chunk_stride = result_stride * 3;  in gemm_q8_0_0_6()
13828   const std::int32_t const_offset = lhs_offset * rhs_offset * k + result_offset;  in gemm_q8_0_0_6()
13829   const std::int32_t rounding_offset = (1 << (shift - 1));  in gemm_q8_0_0_6()
13830   std::int32_t* temp_result = reinterpret_cast<std::int32_t*>(  in gemm_q8_0_0_6()
13833   std::int32_t* mul_result_chunk = temp_result;  in gemm_q8_0_0_6()
13834   const std::int32_t mul_result_chunk_stride_bytes = ((n * 4 + 7) / 8) * 8;  in gemm_q8_0_0_6()
13861                    const std::uint8_t* rhs, std::int32_t m, std::int32_t n,  in gemm_q8_0_0_7()
13862                    std::int32_t k, std::int32_t lhs_offset,  in gemm_q8_0_0_7()
13863                    std::int32_t rhs_offset, std::int32_t result_offset,  in gemm_q8_0_0_7()
13864                    std::int32_t multiplicative_offset, std::int32_t shift,  in gemm_q8_0_0_7()
13865                    std::uint8_t* result, std::int32_t result_stride) {  in gemm_q8_0_0_7()
13866   const std::int32_t row_chunks = m / 3;  in gemm_q8_0_0_7()
13867   const std::int32_t col_chunks = n / 3;  in gemm_q8_0_0_7()
13868   const std::int32_t padded_k = ((k + 7) / 8) * 8;  in gemm_q8_0_0_7()
13869   const std::int32_t chunk_size = k * 3;  in gemm_q8_0_0_7()
13870   const std::int32_t zipped_chunk_size = (padded_k + 16) * 3;  in gemm_q8_0_0_7()
13871   const std::int32_t zipped_rhs_size = (padded_k + 16) * n;  in gemm_q8_0_0_7()
13875   std::int32_t* zipped_lhs_3_offsets =  in gemm_q8_0_0_7()
13876       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 3);  in gemm_q8_0_0_7()
13879   const std::int32_t result_chunk_stride = result_stride * 3;  in gemm_q8_0_0_7()
13881   const std::int32_t const_offset = lhs_offset * rhs_offset * k + result_offset;  in gemm_q8_0_0_7()
13882   const std::int32_t rounding_offset = (1 << (shift - 1));  in gemm_q8_0_0_7()
13883   std::int32_t* temp_result = reinterpret_cast<std::int32_t*>(  in gemm_q8_0_0_7()
13886   std::int32_t* mul_result_chunk = temp_result;  in gemm_q8_0_0_7()
13887   const std::int32_t mul_result_chunk_stride_bytes = ((n * 4 + 7) / 8) * 8;  in gemm_q8_0_0_7()
13914                    const std::uint8_t* rhs, std::int32_t m, std::int32_t n,  in gemm_q8_0_1_0()
13915                    std::int32_t k, std::int32_t lhs_offset,  in gemm_q8_0_1_0()
13916                    std::int32_t rhs_offset, std::int32_t result_offset,  in gemm_q8_0_1_0()
13917                    std::int32_t multiplicative_offset, std::int32_t shift,  in gemm_q8_0_1_0()
13918                    std::uint8_t* result, std::int32_t result_stride) {  in gemm_q8_0_1_0()
13919   const std::int32_t row_chunks = m / 3;  in gemm_q8_0_1_0()
13920   const std::int32_t col_chunks = n / 3;  in gemm_q8_0_1_0()
13921   const std::int32_t padded_k = ((k + 7) / 8) * 8;  in gemm_q8_0_1_0()
13922   const std::int32_t chunk_size = k * 3;  in gemm_q8_0_1_0()
13923   const std::int32_t zipped_chunk_size = (padded_k + 16) * 3;  in gemm_q8_0_1_0()
13924   const std::int32_t zipped_rhs_size = (padded_k + 16) * n;  in gemm_q8_0_1_0()
13928   std::int32_t* zipped_lhs_3_offsets =  in gemm_q8_0_1_0()
13929       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 3);  in gemm_q8_0_1_0()
13932   const std::int32_t result_chunk_stride = result_stride * 3;  in gemm_q8_0_1_0()
13934   const std::int32_t const_offset = lhs_offset * rhs_offset * k + result_offset;  in gemm_q8_0_1_0()
13935   const std::int32_t rounding_offset = (1 << (shift - 1));  in gemm_q8_0_1_0()
13936   std::int32_t* temp_result = reinterpret_cast<std::int32_t*>(  in gemm_q8_0_1_0()
13939   std::int32_t* mul_result_chunk = temp_result;  in gemm_q8_0_1_0()
13940   const std::int32_t mul_result_chunk_stride_bytes = ((n * 4 + 7) / 8) * 8;  in gemm_q8_0_1_0()
13970                    const std::uint8_t* rhs, std::int32_t m, std::int32_t n,  in gemm_q8_0_1_1()
13971                    std::int32_t k, std::int32_t lhs_offset,  in gemm_q8_0_1_1()
13972                    std::int32_t rhs_offset, std::int32_t result_offset,  in gemm_q8_0_1_1()
13973                    std::int32_t multiplicative_offset, std::int32_t shift,  in gemm_q8_0_1_1()
13974                    std::uint8_t* result, std::int32_t result_stride) {  in gemm_q8_0_1_1()
13975   const std::int32_t row_chunks = m / 3;  in gemm_q8_0_1_1()
13976   const std::int32_t col_chunks = n / 3;  in gemm_q8_0_1_1()
13977   const std::int32_t padded_k = ((k + 7) / 8) * 8;  in gemm_q8_0_1_1()
13978   const std::int32_t chunk_size = k * 3;  in gemm_q8_0_1_1()
13979   const std::int32_t zipped_chunk_size = (padded_k + 16) * 3;  in gemm_q8_0_1_1()
13980   const std::int32_t zipped_rhs_size = (padded_k + 16) * n;  in gemm_q8_0_1_1()
13984   std::int32_t* zipped_lhs_3_offsets =  in gemm_q8_0_1_1()
13985       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 3);  in gemm_q8_0_1_1()
13988   const std::int32_t result_chunk_stride = result_stride * 3;  in gemm_q8_0_1_1()
13990   const std::int32_t const_offset = lhs_offset * rhs_offset * k + result_offset;  in gemm_q8_0_1_1()
13991   const std::int32_t rounding_offset = (1 << (shift - 1));  in gemm_q8_0_1_1()
13992   std::int32_t* temp_result = reinterpret_cast<std::int32_t*>(  in gemm_q8_0_1_1()
13995   std::int32_t* mul_result_chunk = temp_result;  in gemm_q8_0_1_1()
13996   const std::int32_t mul_result_chunk_stride_bytes = ((n * 4 + 7) / 8) * 8;  in gemm_q8_0_1_1()
14026                    const std::uint8_t* rhs, std::int32_t m, std::int32_t n,  in gemm_q8_0_1_2()
14027                    std::int32_t k, std::int32_t lhs_offset,  in gemm_q8_0_1_2()
14028                    std::int32_t rhs_offset, std::int32_t result_offset,  in gemm_q8_0_1_2()
14029                    std::int32_t multiplicative_offset, std::int32_t shift,  in gemm_q8_0_1_2()
14030                    std::uint8_t* result, std::int32_t result_stride) {  in gemm_q8_0_1_2()
14031   const std::int32_t row_chunks = m / 3;  in gemm_q8_0_1_2()
14032   const std::int32_t col_chunks = n / 3;  in gemm_q8_0_1_2()
14033   const std::int32_t padded_k = ((k + 7) / 8) * 8;  in gemm_q8_0_1_2()
14034   const std::int32_t chunk_size = k * 3;  in gemm_q8_0_1_2()
14035   const std::int32_t zipped_chunk_size = (padded_k + 16) * 3;  in gemm_q8_0_1_2()
14036   const std::int32_t zipped_rhs_size = (padded_k + 16) * n;  in gemm_q8_0_1_2()
14040   std::int32_t* zipped_lhs_3_offsets =  in gemm_q8_0_1_2()
14041       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 3);  in gemm_q8_0_1_2()
14044   const std::int32_t result_chunk_stride = result_stride * 3;  in gemm_q8_0_1_2()
14046   const std::int32_t const_offset = lhs_offset * rhs_offset * k + result_offset;  in gemm_q8_0_1_2()
14047   const std::int32_t rounding_offset = (1 << (shift - 1));  in gemm_q8_0_1_2()
14048   std::int32_t* temp_result = reinterpret_cast<std::int32_t*>(  in gemm_q8_0_1_2()
14051   std::int32_t* mul_result_chunk = temp_result;  in gemm_q8_0_1_2()
14052   const std::int32_t mul_result_chunk_stride_bytes = ((n * 4 + 7) / 8) * 8;  in gemm_q8_0_1_2()
14082                    const std::uint8_t* rhs, std::int32_t m, std::int32_t n,  in gemm_q8_0_1_3()
14083                    std::int32_t k, std::int32_t lhs_offset,  in gemm_q8_0_1_3()
14084                    std::int32_t rhs_offset, std::int32_t result_offset,  in gemm_q8_0_1_3()
14085                    std::int32_t multiplicative_offset, std::int32_t shift,  in gemm_q8_0_1_3()
14086                    std::uint8_t* result, std::int32_t result_stride) {  in gemm_q8_0_1_3()
14087   const std::int32_t row_chunks = m / 3;  in gemm_q8_0_1_3()
14088   const std::int32_t col_chunks = n / 3;  in gemm_q8_0_1_3()
14089   const std::int32_t padded_k = ((k + 7) / 8) * 8;  in gemm_q8_0_1_3()
14090   const std::int32_t chunk_size = k * 3;  in gemm_q8_0_1_3()
14091   const std::int32_t zipped_chunk_size = (padded_k + 16) * 3;  in gemm_q8_0_1_3()
14092   const std::int32_t zipped_rhs_size = (padded_k + 16) * n;  in gemm_q8_0_1_3()
14096   std::int32_t* zipped_lhs_3_offsets =  in gemm_q8_0_1_3()
14097       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 3);  in gemm_q8_0_1_3()
14100   const std::int32_t result_chunk_stride = result_stride * 3;  in gemm_q8_0_1_3()
14102   const std::int32_t const_offset = lhs_offset * rhs_offset * k + result_offset;  in gemm_q8_0_1_3()
14103   const std::int32_t rounding_offset = (1 << (shift - 1));  in gemm_q8_0_1_3()
14104   std::int32_t* temp_result = reinterpret_cast<std::int32_t*>(  in gemm_q8_0_1_3()
14107   std::int32_t* mul_result_chunk = temp_result;  in gemm_q8_0_1_3()
14108   const std::int32_t mul_result_chunk_stride_bytes = ((n * 4 + 7) / 8) * 8;  in gemm_q8_0_1_3()
14138                    const std::uint8_t* rhs, std::int32_t m, std::int32_t n,  in gemm_q8_0_1_4()
14139                    std::int32_t k, std::int32_t lhs_offset,  in gemm_q8_0_1_4()
14140                    std::int32_t rhs_offset, std::int32_t result_offset,  in gemm_q8_0_1_4()
14141                    std::int32_t multiplicative_offset, std::int32_t shift,  in gemm_q8_0_1_4()
14142                    std::uint8_t* result, std::int32_t result_stride) {  in gemm_q8_0_1_4()
14143   const std::int32_t row_chunks = m / 3;  in gemm_q8_0_1_4()
14144   const std::int32_t col_chunks = n / 3;  in gemm_q8_0_1_4()
14145   const std::int32_t padded_k = ((k + 7) / 8) * 8;  in gemm_q8_0_1_4()
14146   const std::int32_t chunk_size = k * 3;  in gemm_q8_0_1_4()
14147   const std::int32_t zipped_chunk_size = (padded_k + 16) * 3;  in gemm_q8_0_1_4()
14148   const std::int32_t zipped_rhs_size = (padded_k + 16) * n;  in gemm_q8_0_1_4()
14152   std::int32_t* zipped_lhs_3_offsets =  in gemm_q8_0_1_4()
14153       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 3);  in gemm_q8_0_1_4()
14156   const std::int32_t result_chunk_stride = result_stride * 3;  in gemm_q8_0_1_4()
14158   const std::int32_t const_offset = lhs_offset * rhs_offset * k + result_offset;  in gemm_q8_0_1_4()
14159   const std::int32_t rounding_offset = (1 << (shift - 1));  in gemm_q8_0_1_4()
14160   std::int32_t* temp_result = reinterpret_cast<std::int32_t*>(  in gemm_q8_0_1_4()
14163   std::int32_t* mul_result_chunk = temp_result;  in gemm_q8_0_1_4()
14164   const std::int32_t mul_result_chunk_stride_bytes = ((n * 4 + 7) / 8) * 8;  in gemm_q8_0_1_4()
14194                    const std::uint8_t* rhs, std::int32_t m, std::int32_t n,  in gemm_q8_0_1_5()
14195                    std::int32_t k, std::int32_t lhs_offset,  in gemm_q8_0_1_5()
14196                    std::int32_t rhs_offset, std::int32_t result_offset,  in gemm_q8_0_1_5()
14197                    std::int32_t multiplicative_offset, std::int32_t shift,  in gemm_q8_0_1_5()
14198                    std::uint8_t* result, std::int32_t result_stride) {  in gemm_q8_0_1_5()
14199   const std::int32_t row_chunks = m / 3;  in gemm_q8_0_1_5()
14200   const std::int32_t col_chunks = n / 3;  in gemm_q8_0_1_5()
14201   const std::int32_t padded_k = ((k + 7) / 8) * 8;  in gemm_q8_0_1_5()
14202   const std::int32_t chunk_size = k * 3;  in gemm_q8_0_1_5()
14203   const std::int32_t zipped_chunk_size = (padded_k + 16) * 3;  in gemm_q8_0_1_5()
14204   const std::int32_t zipped_rhs_size = (padded_k + 16) * n;  in gemm_q8_0_1_5()
14208   std::int32_t* zipped_lhs_3_offsets =  in gemm_q8_0_1_5()
14209       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 3);  in gemm_q8_0_1_5()
14212   const std::int32_t result_chunk_stride = result_stride * 3;  in gemm_q8_0_1_5()
14214   const std::int32_t const_offset = lhs_offset * rhs_offset * k + result_offset;  in gemm_q8_0_1_5()
14215   const std::int32_t rounding_offset = (1 << (shift - 1));  in gemm_q8_0_1_5()
14216   std::int32_t* temp_result = reinterpret_cast<std::int32_t*>(  in gemm_q8_0_1_5()
14219   std::int32_t* mul_result_chunk = temp_result;  in gemm_q8_0_1_5()
14220   const std::int32_t mul_result_chunk_stride_bytes = ((n * 4 + 7) / 8) * 8;  in gemm_q8_0_1_5()
14250                    const std::uint8_t* rhs, std::int32_t m, std::int32_t n,  in gemm_q8_0_1_6()
14251                    std::int32_t k, std::int32_t lhs_offset,  in gemm_q8_0_1_6()
14252                    std::int32_t rhs_offset, std::int32_t result_offset,  in gemm_q8_0_1_6()
14253                    std::int32_t multiplicative_offset, std::int32_t shift,  in gemm_q8_0_1_6()
14254                    std::uint8_t* result, std::int32_t result_stride) {  in gemm_q8_0_1_6()
14255   const std::int32_t row_chunks = m / 3;  in gemm_q8_0_1_6()
14256   const std::int32_t col_chunks = n / 3;  in gemm_q8_0_1_6()
14257   const std::int32_t padded_k = ((k + 7) / 8) * 8;  in gemm_q8_0_1_6()
14258   const std::int32_t chunk_size = k * 3;  in gemm_q8_0_1_6()
14259   const std::int32_t zipped_chunk_size = (padded_k + 16) * 3;  in gemm_q8_0_1_6()
14260   const std::int32_t zipped_rhs_size = (padded_k + 16) * n;  in gemm_q8_0_1_6()
14264   std::int32_t* zipped_lhs_3_offsets =  in gemm_q8_0_1_6()
14265       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 3);  in gemm_q8_0_1_6()
14268   const std::int32_t result_chunk_stride = result_stride * 3;  in gemm_q8_0_1_6()
14270   const std::int32_t const_offset = lhs_offset * rhs_offset * k + result_offset;  in gemm_q8_0_1_6()
14271   const std::int32_t rounding_offset = (1 << (shift - 1));  in gemm_q8_0_1_6()
14272   std::int32_t* temp_result = reinterpret_cast<std::int32_t*>(  in gemm_q8_0_1_6()
14275   std::int32_t* mul_result_chunk = temp_result;  in gemm_q8_0_1_6()
14276   const std::int32_t mul_result_chunk_stride_bytes = ((n * 4 + 7) / 8) * 8;  in gemm_q8_0_1_6()
14306                    const std::uint8_t* rhs, std::int32_t m, std::int32_t n,  in gemm_q8_0_1_7()
14307                    std::int32_t k, std::int32_t lhs_offset,  in gemm_q8_0_1_7()
14308                    std::int32_t rhs_offset, std::int32_t result_offset,  in gemm_q8_0_1_7()
14309                    std::int32_t multiplicative_offset, std::int32_t shift,  in gemm_q8_0_1_7()
14310                    std::uint8_t* result, std::int32_t result_stride) {  in gemm_q8_0_1_7()
14311   const std::int32_t row_chunks = m / 3;  in gemm_q8_0_1_7()
14312   const std::int32_t col_chunks = n / 3;  in gemm_q8_0_1_7()
14313   const std::int32_t padded_k = ((k + 7) / 8) * 8;  in gemm_q8_0_1_7()
14314   const std::int32_t chunk_size = k * 3;  in gemm_q8_0_1_7()
14315   const std::int32_t zipped_chunk_size = (padded_k + 16) * 3;  in gemm_q8_0_1_7()
14316   const std::int32_t zipped_rhs_size = (padded_k + 16) * n;  in gemm_q8_0_1_7()
14320   std::int32_t* zipped_lhs_3_offsets =  in gemm_q8_0_1_7()
14321       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 3);  in gemm_q8_0_1_7()
14324   const std::int32_t result_chunk_stride = result_stride * 3;  in gemm_q8_0_1_7()
14326   const std::int32_t const_offset = lhs_offset * rhs_offset * k + result_offset;  in gemm_q8_0_1_7()
14327   const std::int32_t rounding_offset = (1 << (shift - 1));  in gemm_q8_0_1_7()
14328   std::int32_t* temp_result = reinterpret_cast<std::int32_t*>(  in gemm_q8_0_1_7()
14331   std::int32_t* mul_result_chunk = temp_result;  in gemm_q8_0_1_7()
14332   const std::int32_t mul_result_chunk_stride_bytes = ((n * 4 + 7) / 8) * 8;  in gemm_q8_0_1_7()
14362                    const std::uint8_t* rhs, std::int32_t m, std::int32_t n,  in gemm_q8_0_2_0()
14363                    std::int32_t k, std::int32_t lhs_offset,  in gemm_q8_0_2_0()
14364                    std::int32_t rhs_offset, std::int32_t result_offset,  in gemm_q8_0_2_0()
14365                    std::int32_t multiplicative_offset, std::int32_t shift,  in gemm_q8_0_2_0()
14366                    std::uint8_t* result, std::int32_t result_stride) {  in gemm_q8_0_2_0()
14367   const std::int32_t row_chunks = m / 3;  in gemm_q8_0_2_0()
14368   const std::int32_t col_chunks = n / 3;  in gemm_q8_0_2_0()
14369   const std::int32_t padded_k = ((k + 7) / 8) * 8;  in gemm_q8_0_2_0()
14370   const std::int32_t chunk_size = k * 3;  in gemm_q8_0_2_0()
14371   const std::int32_t zipped_chunk_size = (padded_k + 16) * 3;  in gemm_q8_0_2_0()
14372   const std::int32_t zipped_rhs_size = (padded_k + 16) * n;  in gemm_q8_0_2_0()
14376   std::int32_t* zipped_lhs_3_offsets =  in gemm_q8_0_2_0()
14377       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 3);  in gemm_q8_0_2_0()
14380   const std::int32_t result_chunk_stride = result_stride * 3;  in gemm_q8_0_2_0()
14382   const std::int32_t const_offset = lhs_offset * rhs_offset * k + result_offset;  in gemm_q8_0_2_0()
14383   const std::int32_t rounding_offset = (1 << (shift - 1));  in gemm_q8_0_2_0()
14384   std::int32_t* temp_result = reinterpret_cast<std::int32_t*>(  in gemm_q8_0_2_0()
14387   std::int32_t* mul_result_chunk = temp_result;  in gemm_q8_0_2_0()
14388   const std::int32_t mul_result_chunk_stride_bytes = ((n * 4 + 7) / 8) * 8;  in gemm_q8_0_2_0()
14418                    const std::uint8_t* rhs, std::int32_t m, std::int32_t n,  in gemm_q8_0_2_1()
14419                    std::int32_t k, std::int32_t lhs_offset,  in gemm_q8_0_2_1()
14420                    std::int32_t rhs_offset, std::int32_t result_offset,  in gemm_q8_0_2_1()
14421                    std::int32_t multiplicative_offset, std::int32_t shift,  in gemm_q8_0_2_1()
14422                    std::uint8_t* result, std::int32_t result_stride) {  in gemm_q8_0_2_1()
14423   const std::int32_t row_chunks = m / 3;  in gemm_q8_0_2_1()
14424   const std::int32_t col_chunks = n / 3;  in gemm_q8_0_2_1()
14425   const std::int32_t padded_k = ((k + 7) / 8) * 8;  in gemm_q8_0_2_1()
14426   const std::int32_t chunk_size = k * 3;  in gemm_q8_0_2_1()
14427   const std::int32_t zipped_chunk_size = (padded_k + 16) * 3;  in gemm_q8_0_2_1()
14428   const std::int32_t zipped_rhs_size = (padded_k + 16) * n;  in gemm_q8_0_2_1()
14432   std::int32_t* zipped_lhs_3_offsets =  in gemm_q8_0_2_1()
14433       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 3);  in gemm_q8_0_2_1()
14436   const std::int32_t result_chunk_stride = result_stride * 3;  in gemm_q8_0_2_1()
14438   const std::int32_t const_offset = lhs_offset * rhs_offset * k + result_offset;  in gemm_q8_0_2_1()
14439   const std::int32_t rounding_offset = (1 << (shift - 1));  in gemm_q8_0_2_1()
14440   std::int32_t* temp_result = reinterpret_cast<std::int32_t*>(  in gemm_q8_0_2_1()
14443   std::int32_t* mul_result_chunk = temp_result;  in gemm_q8_0_2_1()
14444   const std::int32_t mul_result_chunk_stride_bytes = ((n * 4 + 7) / 8) * 8;  in gemm_q8_0_2_1()
14474                    const std::uint8_t* rhs, std::int32_t m, std::int32_t n,  in gemm_q8_0_2_2()
14475                    std::int32_t k, std::int32_t lhs_offset,  in gemm_q8_0_2_2()
14476                    std::int32_t rhs_offset, std::int32_t result_offset,  in gemm_q8_0_2_2()
14477                    std::int32_t multiplicative_offset, std::int32_t shift,  in gemm_q8_0_2_2()
14478                    std::uint8_t* result, std::int32_t result_stride) {  in gemm_q8_0_2_2()
14479   const std::int32_t row_chunks = m / 3;  in gemm_q8_0_2_2()
14480   const std::int32_t col_chunks = n / 3;  in gemm_q8_0_2_2()
14481   const std::int32_t padded_k = ((k + 7) / 8) * 8;  in gemm_q8_0_2_2()
14482   const std::int32_t chunk_size = k * 3;  in gemm_q8_0_2_2()
14483   const std::int32_t zipped_chunk_size = (padded_k + 16) * 3;  in gemm_q8_0_2_2()
14484   const std::int32_t zipped_rhs_size = (padded_k + 16) * n;  in gemm_q8_0_2_2()
14488   std::int32_t* zipped_lhs_3_offsets =  in gemm_q8_0_2_2()
14489       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 3);  in gemm_q8_0_2_2()
14492   const std::int32_t result_chunk_stride = result_stride * 3;  in gemm_q8_0_2_2()
14494   const std::int32_t const_offset = lhs_offset * rhs_offset * k + result_offset;  in gemm_q8_0_2_2()
14495   const std::int32_t rounding_offset = (1 << (shift - 1));  in gemm_q8_0_2_2()
14496   std::int32_t* temp_result = reinterpret_cast<std::int32_t*>(  in gemm_q8_0_2_2()
14499   std::int32_t* mul_result_chunk = temp_result;  in gemm_q8_0_2_2()
14500   const std::int32_t mul_result_chunk_stride_bytes = ((n * 4 + 7) / 8) * 8;  in gemm_q8_0_2_2()
14530                    const std::uint8_t* rhs, std::int32_t m, std::int32_t n,  in gemm_q8_0_2_3()
14531                    std::int32_t k, std::int32_t lhs_offset,  in gemm_q8_0_2_3()
14532                    std::int32_t rhs_offset, std::int32_t result_offset,  in gemm_q8_0_2_3()
14533                    std::int32_t multiplicative_offset, std::int32_t shift,  in gemm_q8_0_2_3()
14534                    std::uint8_t* result, std::int32_t result_stride) {  in gemm_q8_0_2_3()
14535   const std::int32_t row_chunks = m / 3;  in gemm_q8_0_2_3()
14536   const std::int32_t col_chunks = n / 3;  in gemm_q8_0_2_3()
14537   const std::int32_t padded_k = ((k + 7) / 8) * 8;  in gemm_q8_0_2_3()
14538   const std::int32_t chunk_size = k * 3;  in gemm_q8_0_2_3()
14539   const std::int32_t zipped_chunk_size = (padded_k + 16) * 3;  in gemm_q8_0_2_3()
14540   const std::int32_t zipped_rhs_size = (padded_k + 16) * n;  in gemm_q8_0_2_3()
14544   std::int32_t* zipped_lhs_3_offsets =  in gemm_q8_0_2_3()
14545       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 3);  in gemm_q8_0_2_3()
14548   const std::int32_t result_chunk_stride = result_stride * 3;  in gemm_q8_0_2_3()
14550   const std::int32_t const_offset = lhs_offset * rhs_offset * k + result_offset;  in gemm_q8_0_2_3()
14551   const std::int32_t rounding_offset = (1 << (shift - 1));  in gemm_q8_0_2_3()
14552   std::int32_t* temp_result = reinterpret_cast<std::int32_t*>(  in gemm_q8_0_2_3()
14555   std::int32_t* mul_result_chunk = temp_result;  in gemm_q8_0_2_3()
14556   const std::int32_t mul_result_chunk_stride_bytes = ((n * 4 + 7) / 8) * 8;  in gemm_q8_0_2_3()
14586                    const std::uint8_t* rhs, std::int32_t m, std::int32_t n,  in gemm_q8_0_2_4()
14587                    std::int32_t k, std::int32_t lhs_offset,  in gemm_q8_0_2_4()
14588                    std::int32_t rhs_offset, std::int32_t result_offset,  in gemm_q8_0_2_4()
14589                    std::int32_t multiplicative_offset, std::int32_t shift,  in gemm_q8_0_2_4()
14590                    std::uint8_t* result, std::int32_t result_stride) {  in gemm_q8_0_2_4()
14591   const std::int32_t row_chunks = m / 3;  in gemm_q8_0_2_4()
14592   const std::int32_t col_chunks = n / 3;  in gemm_q8_0_2_4()
14593   const std::int32_t padded_k = ((k + 7) / 8) * 8;  in gemm_q8_0_2_4()
14594   const std::int32_t chunk_size = k * 3;  in gemm_q8_0_2_4()
14595   const std::int32_t zipped_chunk_size = (padded_k + 16) * 3;  in gemm_q8_0_2_4()
14596   const std::int32_t zipped_rhs_size = (padded_k + 16) * n;  in gemm_q8_0_2_4()
14600   std::int32_t* zipped_lhs_3_offsets =  in gemm_q8_0_2_4()
14601       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 3);  in gemm_q8_0_2_4()
14604   const std::int32_t result_chunk_stride = result_stride * 3;  in gemm_q8_0_2_4()
14606   const std::int32_t const_offset = lhs_offset * rhs_offset * k + result_offset;  in gemm_q8_0_2_4()
14607   const std::int32_t rounding_offset = (1 << (shift - 1));  in gemm_q8_0_2_4()
14608   std::int32_t* temp_result = reinterpret_cast<std::int32_t*>(  in gemm_q8_0_2_4()
14611   std::int32_t* mul_result_chunk = temp_result;  in gemm_q8_0_2_4()
14612   const std::int32_t mul_result_chunk_stride_bytes = ((n * 4 + 7) / 8) * 8;  in gemm_q8_0_2_4()
14642                    const std::uint8_t* rhs, std::int32_t m, std::int32_t n,  in gemm_q8_0_2_5()
14643                    std::int32_t k, std::int32_t lhs_offset,  in gemm_q8_0_2_5()
14644                    std::int32_t rhs_offset, std::int32_t result_offset,  in gemm_q8_0_2_5()
14645                    std::int32_t multiplicative_offset, std::int32_t shift,  in gemm_q8_0_2_5()
14646                    std::uint8_t* result, std::int32_t result_stride) {  in gemm_q8_0_2_5()
14647   const std::int32_t row_chunks = m / 3;  in gemm_q8_0_2_5()
14648   const std::int32_t col_chunks = n / 3;  in gemm_q8_0_2_5()
14649   const std::int32_t padded_k = ((k + 7) / 8) * 8;  in gemm_q8_0_2_5()
14650   const std::int32_t chunk_size = k * 3;  in gemm_q8_0_2_5()
14651   const std::int32_t zipped_chunk_size = (padded_k + 16) * 3;  in gemm_q8_0_2_5()
14652   const std::int32_t zipped_rhs_size = (padded_k + 16) * n;  in gemm_q8_0_2_5()
14656   std::int32_t* zipped_lhs_3_offsets =  in gemm_q8_0_2_5()
14657       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 3);  in gemm_q8_0_2_5()
14660   const std::int32_t result_chunk_stride = result_stride * 3;  in gemm_q8_0_2_5()
14662   const std::int32_t const_offset = lhs_offset * rhs_offset * k + result_offset;  in gemm_q8_0_2_5()
14663   const std::int32_t rounding_offset = (1 << (shift - 1));  in gemm_q8_0_2_5()
14664   std::int32_t* temp_result = reinterpret_cast<std::int32_t*>(  in gemm_q8_0_2_5()
14667   std::int32_t* mul_result_chunk = temp_result;  in gemm_q8_0_2_5()
14668   const std::int32_t mul_result_chunk_stride_bytes = ((n * 4 + 7) / 8) * 8;  in gemm_q8_0_2_5()
14698                    const std::uint8_t* rhs, std::int32_t m, std::int32_t n,  in gemm_q8_0_2_6()
14699                    std::int32_t k, std::int32_t lhs_offset,  in gemm_q8_0_2_6()
14700                    std::int32_t rhs_offset, std::int32_t result_offset,  in gemm_q8_0_2_6()
14701                    std::int32_t multiplicative_offset, std::int32_t shift,  in gemm_q8_0_2_6()
14702                    std::uint8_t* result, std::int32_t result_stride) {  in gemm_q8_0_2_6()
14703   const std::int32_t row_chunks = m / 3;  in gemm_q8_0_2_6()
14704   const std::int32_t col_chunks = n / 3;  in gemm_q8_0_2_6()
14705   const std::int32_t padded_k = ((k + 7) / 8) * 8;  in gemm_q8_0_2_6()
14706   const std::int32_t chunk_size = k * 3;  in gemm_q8_0_2_6()
14707   const std::int32_t zipped_chunk_size = (padded_k + 16) * 3;  in gemm_q8_0_2_6()
14708   const std::int32_t zipped_rhs_size = (padded_k + 16) * n;  in gemm_q8_0_2_6()
14712   std::int32_t* zipped_lhs_3_offsets =  in gemm_q8_0_2_6()
14713       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 3);  in gemm_q8_0_2_6()
14716   const std::int32_t result_chunk_stride = result_stride * 3;  in gemm_q8_0_2_6()
14718   const std::int32_t const_offset = lhs_offset * rhs_offset * k + result_offset;  in gemm_q8_0_2_6()
14719   const std::int32_t rounding_offset = (1 << (shift - 1));  in gemm_q8_0_2_6()
14720   std::int32_t* temp_result = reinterpret_cast<std::int32_t*>(  in gemm_q8_0_2_6()
14723   std::int32_t* mul_result_chunk = temp_result;  in gemm_q8_0_2_6()
14724   const std::int32_t mul_result_chunk_stride_bytes = ((n * 4 + 7) / 8) * 8;  in gemm_q8_0_2_6()
14754                    const std::uint8_t* rhs, std::int32_t m, std::int32_t n,  in gemm_q8_0_2_7()
14755                    std::int32_t k, std::int32_t lhs_offset,  in gemm_q8_0_2_7()
14756                    std::int32_t rhs_offset, std::int32_t result_offset,  in gemm_q8_0_2_7()
14757                    std::int32_t multiplicative_offset, std::int32_t shift,  in gemm_q8_0_2_7()
14758                    std::uint8_t* result, std::int32_t result_stride) {  in gemm_q8_0_2_7()
14759   const std::int32_t row_chunks = m / 3;  in gemm_q8_0_2_7()
14760   const std::int32_t col_chunks = n / 3;  in gemm_q8_0_2_7()
14761   const std::int32_t padded_k = ((k + 7) / 8) * 8;  in gemm_q8_0_2_7()
14762   const std::int32_t chunk_size = k * 3;  in gemm_q8_0_2_7()
14763   const std::int32_t zipped_chunk_size = (padded_k + 16) * 3;  in gemm_q8_0_2_7()
14764   const std::int32_t zipped_rhs_size = (padded_k + 16) * n;  in gemm_q8_0_2_7()
14768   std::int32_t* zipped_lhs_3_offsets =  in gemm_q8_0_2_7()
14769       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 3);  in gemm_q8_0_2_7()
14772   const std::int32_t result_chunk_stride = result_stride * 3;  in gemm_q8_0_2_7()
14774   const std::int32_t const_offset = lhs_offset * rhs_offset * k + result_offset;  in gemm_q8_0_2_7()
14775   const std::int32_t rounding_offset = (1 << (shift - 1));  in gemm_q8_0_2_7()
14776   std::int32_t* temp_result = reinterpret_cast<std::int32_t*>(  in gemm_q8_0_2_7()
14779   std::int32_t* mul_result_chunk = temp_result;  in gemm_q8_0_2_7()
14780   const std::int32_t mul_result_chunk_stride_bytes = ((n * 4 + 7) / 8) * 8;  in gemm_q8_0_2_7()
14810                    const std::uint8_t* rhs, std::int32_t m, std::int32_t n,  in gemm_q8_1_0_0()
14811                    std::int32_t k, std::int32_t lhs_offset,  in gemm_q8_1_0_0()
14812                    std::int32_t rhs_offset, std::int32_t result_offset,  in gemm_q8_1_0_0()
14813                    std::int32_t multiplicative_offset, std::int32_t shift,  in gemm_q8_1_0_0()
14814                    std::uint8_t* result, std::int32_t result_stride) {  in gemm_q8_1_0_0()
14815   const std::int32_t row_chunks = m / 3;  in gemm_q8_1_0_0()
14816   const std::int32_t col_chunks = n / 3;  in gemm_q8_1_0_0()
14817   const std::int32_t padded_k = ((k + 7) / 8) * 8;  in gemm_q8_1_0_0()
14818   const std::int32_t chunk_size = k * 3;  in gemm_q8_1_0_0()
14819   const std::int32_t zipped_chunk_size = (padded_k + 16) * 3;  in gemm_q8_1_0_0()
14820   const std::int32_t zipped_rhs_size = (padded_k + 16) * n;  in gemm_q8_1_0_0()
14824   std::int32_t* zipped_lhs_3_offsets =  in gemm_q8_1_0_0()
14825       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 3);  in gemm_q8_1_0_0()
14826   std::int32_t* zipped_lhs_1_offsets =  in gemm_q8_1_0_0()
14827       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 1);  in gemm_q8_1_0_0()
14830   const std::int32_t result_chunk_stride = result_stride * 3;  in gemm_q8_1_0_0()
14832   const std::int32_t const_offset = lhs_offset * rhs_offset * k + result_offset;  in gemm_q8_1_0_0()
14833   const std::int32_t rounding_offset = (1 << (shift - 1));  in gemm_q8_1_0_0()
14834   std::int32_t* temp_result = reinterpret_cast<std::int32_t*>(  in gemm_q8_1_0_0()
14837   std::int32_t* mul_result_chunk = temp_result;  in gemm_q8_1_0_0()
14838   const std::int32_t mul_result_chunk_stride_bytes = ((n * 4 + 7) / 8) * 8;  in gemm_q8_1_0_0()
14878                    const std::uint8_t* rhs, std::int32_t m, std::int32_t n,  in gemm_q8_1_0_1()
14879                    std::int32_t k, std::int32_t lhs_offset,  in gemm_q8_1_0_1()
14880                    std::int32_t rhs_offset, std::int32_t result_offset,  in gemm_q8_1_0_1()
14881                    std::int32_t multiplicative_offset, std::int32_t shift,  in gemm_q8_1_0_1()
14882                    std::uint8_t* result, std::int32_t result_stride) {  in gemm_q8_1_0_1()
14883   const std::int32_t row_chunks = m / 3;  in gemm_q8_1_0_1()
14884   const std::int32_t col_chunks = n / 3;  in gemm_q8_1_0_1()
14885   const std::int32_t padded_k = ((k + 7) / 8) * 8;  in gemm_q8_1_0_1()
14886   const std::int32_t chunk_size = k * 3;  in gemm_q8_1_0_1()
14887   const std::int32_t zipped_chunk_size = (padded_k + 16) * 3;  in gemm_q8_1_0_1()
14888   const std::int32_t zipped_rhs_size = (padded_k + 16) * n;  in gemm_q8_1_0_1()
14892   std::int32_t* zipped_lhs_3_offsets =  in gemm_q8_1_0_1()
14893       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 3);  in gemm_q8_1_0_1()
14894   std::int32_t* zipped_lhs_1_offsets =  in gemm_q8_1_0_1()
14895       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 1);  in gemm_q8_1_0_1()
14898   const std::int32_t result_chunk_stride = result_stride * 3;  in gemm_q8_1_0_1()
14900   const std::int32_t const_offset = lhs_offset * rhs_offset * k + result_offset;  in gemm_q8_1_0_1()
14901   const std::int32_t rounding_offset = (1 << (shift - 1));  in gemm_q8_1_0_1()
14902   std::int32_t* temp_result = reinterpret_cast<std::int32_t*>(  in gemm_q8_1_0_1()
14905   std::int32_t* mul_result_chunk = temp_result;  in gemm_q8_1_0_1()
14906   const std::int32_t mul_result_chunk_stride_bytes = ((n * 4 + 7) / 8) * 8;  in gemm_q8_1_0_1()
14946                    const std::uint8_t* rhs, std::int32_t m, std::int32_t n,  in gemm_q8_1_0_2()
14947                    std::int32_t k, std::int32_t lhs_offset,  in gemm_q8_1_0_2()
14948                    std::int32_t rhs_offset, std::int32_t result_offset,  in gemm_q8_1_0_2()
14949                    std::int32_t multiplicative_offset, std::int32_t shift,  in gemm_q8_1_0_2()
14950                    std::uint8_t* result, std::int32_t result_stride) {  in gemm_q8_1_0_2()
14951   const std::int32_t row_chunks = m / 3;  in gemm_q8_1_0_2()
14952   const std::int32_t col_chunks = n / 3;  in gemm_q8_1_0_2()
14953   const std::int32_t padded_k = ((k + 7) / 8) * 8;  in gemm_q8_1_0_2()
14954   const std::int32_t chunk_size = k * 3;  in gemm_q8_1_0_2()
14955   const std::int32_t zipped_chunk_size = (padded_k + 16) * 3;  in gemm_q8_1_0_2()
14956   const std::int32_t zipped_rhs_size = (padded_k + 16) * n;  in gemm_q8_1_0_2()
14960   std::int32_t* zipped_lhs_3_offsets =  in gemm_q8_1_0_2()
14961       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 3);  in gemm_q8_1_0_2()
14962   std::int32_t* zipped_lhs_1_offsets =  in gemm_q8_1_0_2()
14963       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 1);  in gemm_q8_1_0_2()
14966   const std::int32_t result_chunk_stride = result_stride * 3;  in gemm_q8_1_0_2()
14968   const std::int32_t const_offset = lhs_offset * rhs_offset * k + result_offset;  in gemm_q8_1_0_2()
14969   const std::int32_t rounding_offset = (1 << (shift - 1));  in gemm_q8_1_0_2()
14970   std::int32_t* temp_result = reinterpret_cast<std::int32_t*>(  in gemm_q8_1_0_2()
14973   std::int32_t* mul_result_chunk = temp_result;  in gemm_q8_1_0_2()
14974   const std::int32_t mul_result_chunk_stride_bytes = ((n * 4 + 7) / 8) * 8;  in gemm_q8_1_0_2()
15014                    const std::uint8_t* rhs, std::int32_t m, std::int32_t n,  in gemm_q8_1_0_3()
15015                    std::int32_t k, std::int32_t lhs_offset,  in gemm_q8_1_0_3()
15016                    std::int32_t rhs_offset, std::int32_t result_offset,  in gemm_q8_1_0_3()
15017                    std::int32_t multiplicative_offset, std::int32_t shift,  in gemm_q8_1_0_3()
15018                    std::uint8_t* result, std::int32_t result_stride) {  in gemm_q8_1_0_3()
15019   const std::int32_t row_chunks = m / 3;  in gemm_q8_1_0_3()
15020   const std::int32_t col_chunks = n / 3;  in gemm_q8_1_0_3()
15021   const std::int32_t padded_k = ((k + 7) / 8) * 8;  in gemm_q8_1_0_3()
15022   const std::int32_t chunk_size = k * 3;  in gemm_q8_1_0_3()
15023   const std::int32_t zipped_chunk_size = (padded_k + 16) * 3;  in gemm_q8_1_0_3()
15024   const std::int32_t zipped_rhs_size = (padded_k + 16) * n;  in gemm_q8_1_0_3()
15028   std::int32_t* zipped_lhs_3_offsets =  in gemm_q8_1_0_3()
15029       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 3);  in gemm_q8_1_0_3()
15030   std::int32_t* zipped_lhs_1_offsets =  in gemm_q8_1_0_3()
15031       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 1);  in gemm_q8_1_0_3()
15034   const std::int32_t result_chunk_stride = result_stride * 3;  in gemm_q8_1_0_3()
15036   const std::int32_t const_offset = lhs_offset * rhs_offset * k + result_offset;  in gemm_q8_1_0_3()
15037   const std::int32_t rounding_offset = (1 << (shift - 1));  in gemm_q8_1_0_3()
15038   std::int32_t* temp_result = reinterpret_cast<std::int32_t*>(  in gemm_q8_1_0_3()
15041   std::int32_t* mul_result_chunk = temp_result;  in gemm_q8_1_0_3()
15042   const std::int32_t mul_result_chunk_stride_bytes = ((n * 4 + 7) / 8) * 8;  in gemm_q8_1_0_3()
15082                    const std::uint8_t* rhs, std::int32_t m, std::int32_t n,  in gemm_q8_1_0_4()
15083                    std::int32_t k, std::int32_t lhs_offset,  in gemm_q8_1_0_4()
15084                    std::int32_t rhs_offset, std::int32_t result_offset,  in gemm_q8_1_0_4()
15085                    std::int32_t multiplicative_offset, std::int32_t shift,  in gemm_q8_1_0_4()
15086                    std::uint8_t* result, std::int32_t result_stride) {  in gemm_q8_1_0_4()
15087   const std::int32_t row_chunks = m / 3;  in gemm_q8_1_0_4()
15088   const std::int32_t col_chunks = n / 3;  in gemm_q8_1_0_4()
15089   const std::int32_t padded_k = ((k + 7) / 8) * 8;  in gemm_q8_1_0_4()
15090   const std::int32_t chunk_size = k * 3;  in gemm_q8_1_0_4()
15091   const std::int32_t zipped_chunk_size = (padded_k + 16) * 3;  in gemm_q8_1_0_4()
15092   const std::int32_t zipped_rhs_size = (padded_k + 16) * n;  in gemm_q8_1_0_4()
15096   std::int32_t* zipped_lhs_3_offsets =  in gemm_q8_1_0_4()
15097       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 3);  in gemm_q8_1_0_4()
15098   std::int32_t* zipped_lhs_1_offsets =  in gemm_q8_1_0_4()
15099       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 1);  in gemm_q8_1_0_4()
15102   const std::int32_t result_chunk_stride = result_stride * 3;  in gemm_q8_1_0_4()
15104   const std::int32_t const_offset = lhs_offset * rhs_offset * k + result_offset;  in gemm_q8_1_0_4()
15105   const std::int32_t rounding_offset = (1 << (shift - 1));  in gemm_q8_1_0_4()
15106   std::int32_t* temp_result = reinterpret_cast<std::int32_t*>(  in gemm_q8_1_0_4()
15109   std::int32_t* mul_result_chunk = temp_result;  in gemm_q8_1_0_4()
15110   const std::int32_t mul_result_chunk_stride_bytes = ((n * 4 + 7) / 8) * 8;  in gemm_q8_1_0_4()
15150                    const std::uint8_t* rhs, std::int32_t m, std::int32_t n,  in gemm_q8_1_0_5()
15151                    std::int32_t k, std::int32_t lhs_offset,  in gemm_q8_1_0_5()
15152                    std::int32_t rhs_offset, std::int32_t result_offset,  in gemm_q8_1_0_5()
15153                    std::int32_t multiplicative_offset, std::int32_t shift,  in gemm_q8_1_0_5()
15154                    std::uint8_t* result, std::int32_t result_stride) {  in gemm_q8_1_0_5()
15155   const std::int32_t row_chunks = m / 3;  in gemm_q8_1_0_5()
15156   const std::int32_t col_chunks = n / 3;  in gemm_q8_1_0_5()
15157   const std::int32_t padded_k = ((k + 7) / 8) * 8;  in gemm_q8_1_0_5()
15158   const std::int32_t chunk_size = k * 3;  in gemm_q8_1_0_5()
15159   const std::int32_t zipped_chunk_size = (padded_k + 16) * 3;  in gemm_q8_1_0_5()
15160   const std::int32_t zipped_rhs_size = (padded_k + 16) * n;  in gemm_q8_1_0_5()
15164   std::int32_t* zipped_lhs_3_offsets =  in gemm_q8_1_0_5()
15165       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 3);  in gemm_q8_1_0_5()
15166   std::int32_t* zipped_lhs_1_offsets =  in gemm_q8_1_0_5()
15167       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 1);  in gemm_q8_1_0_5()
15170   const std::int32_t result_chunk_stride = result_stride * 3;  in gemm_q8_1_0_5()
15172   const std::int32_t const_offset = lhs_offset * rhs_offset * k + result_offset;  in gemm_q8_1_0_5()
15173   const std::int32_t rounding_offset = (1 << (shift - 1));  in gemm_q8_1_0_5()
15174   std::int32_t* temp_result = reinterpret_cast<std::int32_t*>(  in gemm_q8_1_0_5()
15177   std::int32_t* mul_result_chunk = temp_result;  in gemm_q8_1_0_5()
15178   const std::int32_t mul_result_chunk_stride_bytes = ((n * 4 + 7) / 8) * 8;  in gemm_q8_1_0_5()
15218                    const std::uint8_t* rhs, std::int32_t m, std::int32_t n,  in gemm_q8_1_0_6()
15219                    std::int32_t k, std::int32_t lhs_offset,  in gemm_q8_1_0_6()
15220                    std::int32_t rhs_offset, std::int32_t result_offset,  in gemm_q8_1_0_6()
15221                    std::int32_t multiplicative_offset, std::int32_t shift,  in gemm_q8_1_0_6()
15222                    std::uint8_t* result, std::int32_t result_stride) {  in gemm_q8_1_0_6()
15223   const std::int32_t row_chunks = m / 3;  in gemm_q8_1_0_6()
15224   const std::int32_t col_chunks = n / 3;  in gemm_q8_1_0_6()
15225   const std::int32_t padded_k = ((k + 7) / 8) * 8;  in gemm_q8_1_0_6()
15226   const std::int32_t chunk_size = k * 3;  in gemm_q8_1_0_6()
15227   const std::int32_t zipped_chunk_size = (padded_k + 16) * 3;  in gemm_q8_1_0_6()
15228   const std::int32_t zipped_rhs_size = (padded_k + 16) * n;  in gemm_q8_1_0_6()
15232   std::int32_t* zipped_lhs_3_offsets =  in gemm_q8_1_0_6()
15233       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 3);  in gemm_q8_1_0_6()
15234   std::int32_t* zipped_lhs_1_offsets =  in gemm_q8_1_0_6()
15235       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 1);  in gemm_q8_1_0_6()
15238   const std::int32_t result_chunk_stride = result_stride * 3;  in gemm_q8_1_0_6()
15240   const std::int32_t const_offset = lhs_offset * rhs_offset * k + result_offset;  in gemm_q8_1_0_6()
15241   const std::int32_t rounding_offset = (1 << (shift - 1));  in gemm_q8_1_0_6()
15242   std::int32_t* temp_result = reinterpret_cast<std::int32_t*>(  in gemm_q8_1_0_6()
15245   std::int32_t* mul_result_chunk = temp_result;  in gemm_q8_1_0_6()
15246   const std::int32_t mul_result_chunk_stride_bytes = ((n * 4 + 7) / 8) * 8;  in gemm_q8_1_0_6()
15286                    const std::uint8_t* rhs, std::int32_t m, std::int32_t n,  in gemm_q8_1_0_7()
15287                    std::int32_t k, std::int32_t lhs_offset,  in gemm_q8_1_0_7()
15288                    std::int32_t rhs_offset, std::int32_t result_offset,  in gemm_q8_1_0_7()
15289                    std::int32_t multiplicative_offset, std::int32_t shift,  in gemm_q8_1_0_7()
15290                    std::uint8_t* result, std::int32_t result_stride) {  in gemm_q8_1_0_7()
15291   const std::int32_t row_chunks = m / 3;  in gemm_q8_1_0_7()
15292   const std::int32_t col_chunks = n / 3;  in gemm_q8_1_0_7()
15293   const std::int32_t padded_k = ((k + 7) / 8) * 8;  in gemm_q8_1_0_7()
15294   const std::int32_t chunk_size = k * 3;  in gemm_q8_1_0_7()
15295   const std::int32_t zipped_chunk_size = (padded_k + 16) * 3;  in gemm_q8_1_0_7()
15296   const std::int32_t zipped_rhs_size = (padded_k + 16) * n;  in gemm_q8_1_0_7()
15300   std::int32_t* zipped_lhs_3_offsets =  in gemm_q8_1_0_7()
15301       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 3);  in gemm_q8_1_0_7()
15302   std::int32_t* zipped_lhs_1_offsets =  in gemm_q8_1_0_7()
15303       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 1);  in gemm_q8_1_0_7()
15306   const std::int32_t result_chunk_stride = result_stride * 3;  in gemm_q8_1_0_7()
15308   const std::int32_t const_offset = lhs_offset * rhs_offset * k + result_offset;  in gemm_q8_1_0_7()
15309   const std::int32_t rounding_offset = (1 << (shift - 1));  in gemm_q8_1_0_7()
15310   std::int32_t* temp_result = reinterpret_cast<std::int32_t*>(  in gemm_q8_1_0_7()
15313   std::int32_t* mul_result_chunk = temp_result;  in gemm_q8_1_0_7()
15314   const std::int32_t mul_result_chunk_stride_bytes = ((n * 4 + 7) / 8) * 8;  in gemm_q8_1_0_7()
15354                    const std::uint8_t* rhs, std::int32_t m, std::int32_t n,  in gemm_q8_1_1_0()
15355                    std::int32_t k, std::int32_t lhs_offset,  in gemm_q8_1_1_0()
15356                    std::int32_t rhs_offset, std::int32_t result_offset,  in gemm_q8_1_1_0()
15357                    std::int32_t multiplicative_offset, std::int32_t shift,  in gemm_q8_1_1_0()
15358                    std::uint8_t* result, std::int32_t result_stride) {  in gemm_q8_1_1_0()
15359   const std::int32_t row_chunks = m / 3;  in gemm_q8_1_1_0()
15360   const std::int32_t col_chunks = n / 3;  in gemm_q8_1_1_0()
15361   const std::int32_t padded_k = ((k + 7) / 8) * 8;  in gemm_q8_1_1_0()
15362   const std::int32_t chunk_size = k * 3;  in gemm_q8_1_1_0()
15363   const std::int32_t zipped_chunk_size = (padded_k + 16) * 3;  in gemm_q8_1_1_0()
15364   const std::int32_t zipped_rhs_size = (padded_k + 16) * n;  in gemm_q8_1_1_0()
15368   std::int32_t* zipped_lhs_3_offsets =  in gemm_q8_1_1_0()
15369       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 3);  in gemm_q8_1_1_0()
15370   std::int32_t* zipped_lhs_1_offsets =  in gemm_q8_1_1_0()
15371       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 1);  in gemm_q8_1_1_0()
15374   const std::int32_t result_chunk_stride = result_stride * 3;  in gemm_q8_1_1_0()
15376   const std::int32_t const_offset = lhs_offset * rhs_offset * k + result_offset;  in gemm_q8_1_1_0()
15377   const std::int32_t rounding_offset = (1 << (shift - 1));  in gemm_q8_1_1_0()
15378   std::int32_t* temp_result = reinterpret_cast<std::int32_t*>(  in gemm_q8_1_1_0()
15381   std::int32_t* mul_result_chunk = temp_result;  in gemm_q8_1_1_0()
15382   const std::int32_t mul_result_chunk_stride_bytes = ((n * 4 + 7) / 8) * 8;  in gemm_q8_1_1_0()
15427                    const std::uint8_t* rhs, std::int32_t m, std::int32_t n,  in gemm_q8_1_1_1()
15428                    std::int32_t k, std::int32_t lhs_offset,  in gemm_q8_1_1_1()
15429                    std::int32_t rhs_offset, std::int32_t result_offset,  in gemm_q8_1_1_1()
15430                    std::int32_t multiplicative_offset, std::int32_t shift,  in gemm_q8_1_1_1()
15431                    std::uint8_t* result, std::int32_t result_stride) {  in gemm_q8_1_1_1()
15432   const std::int32_t row_chunks = m / 3;  in gemm_q8_1_1_1()
15433   const std::int32_t col_chunks = n / 3;  in gemm_q8_1_1_1()
15434   const std::int32_t padded_k = ((k + 7) / 8) * 8;  in gemm_q8_1_1_1()
15435   const std::int32_t chunk_size = k * 3;  in gemm_q8_1_1_1()
15436   const std::int32_t zipped_chunk_size = (padded_k + 16) * 3;  in gemm_q8_1_1_1()
15437   const std::int32_t zipped_rhs_size = (padded_k + 16) * n;  in gemm_q8_1_1_1()
15441   std::int32_t* zipped_lhs_3_offsets =  in gemm_q8_1_1_1()
15442       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 3);  in gemm_q8_1_1_1()
15443   std::int32_t* zipped_lhs_1_offsets =  in gemm_q8_1_1_1()
15444       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 1);  in gemm_q8_1_1_1()
15447   const std::int32_t result_chunk_stride = result_stride * 3;  in gemm_q8_1_1_1()
15449   const std::int32_t const_offset = lhs_offset * rhs_offset * k + result_offset;  in gemm_q8_1_1_1()
15450   const std::int32_t rounding_offset = (1 << (shift - 1));  in gemm_q8_1_1_1()
15451   std::int32_t* temp_result = reinterpret_cast<std::int32_t*>(  in gemm_q8_1_1_1()
15454   std::int32_t* mul_result_chunk = temp_result;  in gemm_q8_1_1_1()
15455   const std::int32_t mul_result_chunk_stride_bytes = ((n * 4 + 7) / 8) * 8;  in gemm_q8_1_1_1()
15500                    const std::uint8_t* rhs, std::int32_t m, std::int32_t n,  in gemm_q8_1_1_2()
15501                    std::int32_t k, std::int32_t lhs_offset,  in gemm_q8_1_1_2()
15502                    std::int32_t rhs_offset, std::int32_t result_offset,  in gemm_q8_1_1_2()
15503                    std::int32_t multiplicative_offset, std::int32_t shift,  in gemm_q8_1_1_2()
15504                    std::uint8_t* result, std::int32_t result_stride) {  in gemm_q8_1_1_2()
15505   const std::int32_t row_chunks = m / 3;  in gemm_q8_1_1_2()
15506   const std::int32_t col_chunks = n / 3;  in gemm_q8_1_1_2()
15507   const std::int32_t padded_k = ((k + 7) / 8) * 8;  in gemm_q8_1_1_2()
15508   const std::int32_t chunk_size = k * 3;  in gemm_q8_1_1_2()
15509   const std::int32_t zipped_chunk_size = (padded_k + 16) * 3;  in gemm_q8_1_1_2()
15510   const std::int32_t zipped_rhs_size = (padded_k + 16) * n;  in gemm_q8_1_1_2()
15514   std::int32_t* zipped_lhs_3_offsets =  in gemm_q8_1_1_2()
15515       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 3);  in gemm_q8_1_1_2()
15516   std::int32_t* zipped_lhs_1_offsets =  in gemm_q8_1_1_2()
15517       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 1);  in gemm_q8_1_1_2()
15520   const std::int32_t result_chunk_stride = result_stride * 3;  in gemm_q8_1_1_2()
15522   const std::int32_t const_offset = lhs_offset * rhs_offset * k + result_offset;  in gemm_q8_1_1_2()
15523   const std::int32_t rounding_offset = (1 << (shift - 1));  in gemm_q8_1_1_2()
15524   std::int32_t* temp_result = reinterpret_cast<std::int32_t*>(  in gemm_q8_1_1_2()
15527   std::int32_t* mul_result_chunk = temp_result;  in gemm_q8_1_1_2()
15528   const std::int32_t mul_result_chunk_stride_bytes = ((n * 4 + 7) / 8) * 8;  in gemm_q8_1_1_2()
15573                    const std::uint8_t* rhs, std::int32_t m, std::int32_t n,  in gemm_q8_1_1_3()
15574                    std::int32_t k, std::int32_t lhs_offset,  in gemm_q8_1_1_3()
15575                    std::int32_t rhs_offset, std::int32_t result_offset,  in gemm_q8_1_1_3()
15576                    std::int32_t multiplicative_offset, std::int32_t shift,  in gemm_q8_1_1_3()
15577                    std::uint8_t* result, std::int32_t result_stride) {  in gemm_q8_1_1_3()
15578   const std::int32_t row_chunks = m / 3;  in gemm_q8_1_1_3()
15579   const std::int32_t col_chunks = n / 3;  in gemm_q8_1_1_3()
15580   const std::int32_t padded_k = ((k + 7) / 8) * 8;  in gemm_q8_1_1_3()
15581   const std::int32_t chunk_size = k * 3;  in gemm_q8_1_1_3()
15582   const std::int32_t zipped_chunk_size = (padded_k + 16) * 3;  in gemm_q8_1_1_3()
15583   const std::int32_t zipped_rhs_size = (padded_k + 16) * n;  in gemm_q8_1_1_3()
15587   std::int32_t* zipped_lhs_3_offsets =  in gemm_q8_1_1_3()
15588       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 3);  in gemm_q8_1_1_3()
15589   std::int32_t* zipped_lhs_1_offsets =  in gemm_q8_1_1_3()
15590       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 1);  in gemm_q8_1_1_3()
15593   const std::int32_t result_chunk_stride = result_stride * 3;  in gemm_q8_1_1_3()
15595   const std::int32_t const_offset = lhs_offset * rhs_offset * k + result_offset;  in gemm_q8_1_1_3()
15596   const std::int32_t rounding_offset = (1 << (shift - 1));  in gemm_q8_1_1_3()
15597   std::int32_t* temp_result = reinterpret_cast<std::int32_t*>(  in gemm_q8_1_1_3()
15600   std::int32_t* mul_result_chunk = temp_result;  in gemm_q8_1_1_3()
15601   const std::int32_t mul_result_chunk_stride_bytes = ((n * 4 + 7) / 8) * 8;  in gemm_q8_1_1_3()
15646                    const std::uint8_t* rhs, std::int32_t m, std::int32_t n,  in gemm_q8_1_1_4()
15647                    std::int32_t k, std::int32_t lhs_offset,  in gemm_q8_1_1_4()
15648                    std::int32_t rhs_offset, std::int32_t result_offset,  in gemm_q8_1_1_4()
15649                    std::int32_t multiplicative_offset, std::int32_t shift,  in gemm_q8_1_1_4()
15650                    std::uint8_t* result, std::int32_t result_stride) {  in gemm_q8_1_1_4()
15651   const std::int32_t row_chunks = m / 3;  in gemm_q8_1_1_4()
15652   const std::int32_t col_chunks = n / 3;  in gemm_q8_1_1_4()
15653   const std::int32_t padded_k = ((k + 7) / 8) * 8;  in gemm_q8_1_1_4()
15654   const std::int32_t chunk_size = k * 3;  in gemm_q8_1_1_4()
15655   const std::int32_t zipped_chunk_size = (padded_k + 16) * 3;  in gemm_q8_1_1_4()
15656   const std::int32_t zipped_rhs_size = (padded_k + 16) * n;  in gemm_q8_1_1_4()
15660   std::int32_t* zipped_lhs_3_offsets =  in gemm_q8_1_1_4()
15661       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 3);  in gemm_q8_1_1_4()
15662   std::int32_t* zipped_lhs_1_offsets =  in gemm_q8_1_1_4()
15663       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 1);  in gemm_q8_1_1_4()
15666   const std::int32_t result_chunk_stride = result_stride * 3;  in gemm_q8_1_1_4()
15668   const std::int32_t const_offset = lhs_offset * rhs_offset * k + result_offset;  in gemm_q8_1_1_4()
15669   const std::int32_t rounding_offset = (1 << (shift - 1));  in gemm_q8_1_1_4()
15670   std::int32_t* temp_result = reinterpret_cast<std::int32_t*>(  in gemm_q8_1_1_4()
15673   std::int32_t* mul_result_chunk = temp_result;  in gemm_q8_1_1_4()
15674   const std::int32_t mul_result_chunk_stride_bytes = ((n * 4 + 7) / 8) * 8;  in gemm_q8_1_1_4()
15719                    const std::uint8_t* rhs, std::int32_t m, std::int32_t n,  in gemm_q8_1_1_5()
15720                    std::int32_t k, std::int32_t lhs_offset,  in gemm_q8_1_1_5()
15721                    std::int32_t rhs_offset, std::int32_t result_offset,  in gemm_q8_1_1_5()
15722                    std::int32_t multiplicative_offset, std::int32_t shift,  in gemm_q8_1_1_5()
15723                    std::uint8_t* result, std::int32_t result_stride) {  in gemm_q8_1_1_5()
15724   const std::int32_t row_chunks = m / 3;  in gemm_q8_1_1_5()
15725   const std::int32_t col_chunks = n / 3;  in gemm_q8_1_1_5()
15726   const std::int32_t padded_k = ((k + 7) / 8) * 8;  in gemm_q8_1_1_5()
15727   const std::int32_t chunk_size = k * 3;  in gemm_q8_1_1_5()
15728   const std::int32_t zipped_chunk_size = (padded_k + 16) * 3;  in gemm_q8_1_1_5()
15729   const std::int32_t zipped_rhs_size = (padded_k + 16) * n;  in gemm_q8_1_1_5()
15733   std::int32_t* zipped_lhs_3_offsets =  in gemm_q8_1_1_5()
15734       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 3);  in gemm_q8_1_1_5()
15735   std::int32_t* zipped_lhs_1_offsets =  in gemm_q8_1_1_5()
15736       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 1);  in gemm_q8_1_1_5()
15739   const std::int32_t result_chunk_stride = result_stride * 3;  in gemm_q8_1_1_5()
15741   const std::int32_t const_offset = lhs_offset * rhs_offset * k + result_offset;  in gemm_q8_1_1_5()
15742   const std::int32_t rounding_offset = (1 << (shift - 1));  in gemm_q8_1_1_5()
15743   std::int32_t* temp_result = reinterpret_cast<std::int32_t*>(  in gemm_q8_1_1_5()
15746   std::int32_t* mul_result_chunk = temp_result;  in gemm_q8_1_1_5()
15747   const std::int32_t mul_result_chunk_stride_bytes = ((n * 4 + 7) / 8) * 8;  in gemm_q8_1_1_5()
15792                    const std::uint8_t* rhs, std::int32_t m, std::int32_t n,  in gemm_q8_1_1_6()
15793                    std::int32_t k, std::int32_t lhs_offset,  in gemm_q8_1_1_6()
15794                    std::int32_t rhs_offset, std::int32_t result_offset,  in gemm_q8_1_1_6()
15795                    std::int32_t multiplicative_offset, std::int32_t shift,  in gemm_q8_1_1_6()
15796                    std::uint8_t* result, std::int32_t result_stride) {  in gemm_q8_1_1_6()
15797   const std::int32_t row_chunks = m / 3;  in gemm_q8_1_1_6()
15798   const std::int32_t col_chunks = n / 3;  in gemm_q8_1_1_6()
15799   const std::int32_t padded_k = ((k + 7) / 8) * 8;  in gemm_q8_1_1_6()
15800   const std::int32_t chunk_size = k * 3;  in gemm_q8_1_1_6()
15801   const std::int32_t zipped_chunk_size = (padded_k + 16) * 3;  in gemm_q8_1_1_6()
15802   const std::int32_t zipped_rhs_size = (padded_k + 16) * n;  in gemm_q8_1_1_6()
15806   std::int32_t* zipped_lhs_3_offsets =  in gemm_q8_1_1_6()
15807       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 3);  in gemm_q8_1_1_6()
15808   std::int32_t* zipped_lhs_1_offsets =  in gemm_q8_1_1_6()
15809       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 1);  in gemm_q8_1_1_6()
15812   const std::int32_t result_chunk_stride = result_stride * 3;  in gemm_q8_1_1_6()
15814   const std::int32_t const_offset = lhs_offset * rhs_offset * k + result_offset;  in gemm_q8_1_1_6()
15815   const std::int32_t rounding_offset = (1 << (shift - 1));  in gemm_q8_1_1_6()
15816   std::int32_t* temp_result = reinterpret_cast<std::int32_t*>(  in gemm_q8_1_1_6()
15819   std::int32_t* mul_result_chunk = temp_result;  in gemm_q8_1_1_6()
15820   const std::int32_t mul_result_chunk_stride_bytes = ((n * 4 + 7) / 8) * 8;  in gemm_q8_1_1_6()
15865                    const std::uint8_t* rhs, std::int32_t m, std::int32_t n,  in gemm_q8_1_1_7()
15866                    std::int32_t k, std::int32_t lhs_offset,  in gemm_q8_1_1_7()
15867                    std::int32_t rhs_offset, std::int32_t result_offset,  in gemm_q8_1_1_7()
15868                    std::int32_t multiplicative_offset, std::int32_t shift,  in gemm_q8_1_1_7()
15869                    std::uint8_t* result, std::int32_t result_stride) {  in gemm_q8_1_1_7()
15870   const std::int32_t row_chunks = m / 3;  in gemm_q8_1_1_7()
15871   const std::int32_t col_chunks = n / 3;  in gemm_q8_1_1_7()
15872   const std::int32_t padded_k = ((k + 7) / 8) * 8;  in gemm_q8_1_1_7()
15873   const std::int32_t chunk_size = k * 3;  in gemm_q8_1_1_7()
15874   const std::int32_t zipped_chunk_size = (padded_k + 16) * 3;  in gemm_q8_1_1_7()
15875   const std::int32_t zipped_rhs_size = (padded_k + 16) * n;  in gemm_q8_1_1_7()
15879   std::int32_t* zipped_lhs_3_offsets =  in gemm_q8_1_1_7()
15880       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 3);  in gemm_q8_1_1_7()
15881   std::int32_t* zipped_lhs_1_offsets =  in gemm_q8_1_1_7()
15882       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 1);  in gemm_q8_1_1_7()
15885   const std::int32_t result_chunk_stride = result_stride * 3;  in gemm_q8_1_1_7()
15887   const std::int32_t const_offset = lhs_offset * rhs_offset * k + result_offset;  in gemm_q8_1_1_7()
15888   const std::int32_t rounding_offset = (1 << (shift - 1));  in gemm_q8_1_1_7()
15889   std::int32_t* temp_result = reinterpret_cast<std::int32_t*>(  in gemm_q8_1_1_7()
15892   std::int32_t* mul_result_chunk = temp_result;  in gemm_q8_1_1_7()
15893   const std::int32_t mul_result_chunk_stride_bytes = ((n * 4 + 7) / 8) * 8;  in gemm_q8_1_1_7()
15938                    const std::uint8_t* rhs, std::int32_t m, std::int32_t n,  in gemm_q8_1_2_0()
15939                    std::int32_t k, std::int32_t lhs_offset,  in gemm_q8_1_2_0()
15940                    std::int32_t rhs_offset, std::int32_t result_offset,  in gemm_q8_1_2_0()
15941                    std::int32_t multiplicative_offset, std::int32_t shift,  in gemm_q8_1_2_0()
15942                    std::uint8_t* result, std::int32_t result_stride) {  in gemm_q8_1_2_0()
15943   const std::int32_t row_chunks = m / 3;  in gemm_q8_1_2_0()
15944   const std::int32_t col_chunks = n / 3;  in gemm_q8_1_2_0()
15945   const std::int32_t padded_k = ((k + 7) / 8) * 8;  in gemm_q8_1_2_0()
15946   const std::int32_t chunk_size = k * 3;  in gemm_q8_1_2_0()
15947   const std::int32_t zipped_chunk_size = (padded_k + 16) * 3;  in gemm_q8_1_2_0()
15948   const std::int32_t zipped_rhs_size = (padded_k + 16) * n;  in gemm_q8_1_2_0()
15952   std::int32_t* zipped_lhs_3_offsets =  in gemm_q8_1_2_0()
15953       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 3);  in gemm_q8_1_2_0()
15954   std::int32_t* zipped_lhs_1_offsets =  in gemm_q8_1_2_0()
15955       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 1);  in gemm_q8_1_2_0()
15958   const std::int32_t result_chunk_stride = result_stride * 3;  in gemm_q8_1_2_0()
15960   const std::int32_t const_offset = lhs_offset * rhs_offset * k + result_offset;  in gemm_q8_1_2_0()
15961   const std::int32_t rounding_offset = (1 << (shift - 1));  in gemm_q8_1_2_0()
15962   std::int32_t* temp_result = reinterpret_cast<std::int32_t*>(  in gemm_q8_1_2_0()
15965   std::int32_t* mul_result_chunk = temp_result;  in gemm_q8_1_2_0()
15966   const std::int32_t mul_result_chunk_stride_bytes = ((n * 4 + 7) / 8) * 8;  in gemm_q8_1_2_0()
16011                    const std::uint8_t* rhs, std::int32_t m, std::int32_t n,  in gemm_q8_1_2_1()
16012                    std::int32_t k, std::int32_t lhs_offset,  in gemm_q8_1_2_1()
16013                    std::int32_t rhs_offset, std::int32_t result_offset,  in gemm_q8_1_2_1()
16014                    std::int32_t multiplicative_offset, std::int32_t shift,  in gemm_q8_1_2_1()
16015                    std::uint8_t* result, std::int32_t result_stride) {  in gemm_q8_1_2_1()
16016   const std::int32_t row_chunks = m / 3;  in gemm_q8_1_2_1()
16017   const std::int32_t col_chunks = n / 3;  in gemm_q8_1_2_1()
16018   const std::int32_t padded_k = ((k + 7) / 8) * 8;  in gemm_q8_1_2_1()
16019   const std::int32_t chunk_size = k * 3;  in gemm_q8_1_2_1()
16020   const std::int32_t zipped_chunk_size = (padded_k + 16) * 3;  in gemm_q8_1_2_1()
16021   const std::int32_t zipped_rhs_size = (padded_k + 16) * n;  in gemm_q8_1_2_1()
16025   std::int32_t* zipped_lhs_3_offsets =  in gemm_q8_1_2_1()
16026       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 3);  in gemm_q8_1_2_1()
16027   std::int32_t* zipped_lhs_1_offsets =  in gemm_q8_1_2_1()
16028       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 1);  in gemm_q8_1_2_1()
16031   const std::int32_t result_chunk_stride = result_stride * 3;  in gemm_q8_1_2_1()
16033   const std::int32_t const_offset = lhs_offset * rhs_offset * k + result_offset;  in gemm_q8_1_2_1()
16034   const std::int32_t rounding_offset = (1 << (shift - 1));  in gemm_q8_1_2_1()
16035   std::int32_t* temp_result = reinterpret_cast<std::int32_t*>(  in gemm_q8_1_2_1()
16038   std::int32_t* mul_result_chunk = temp_result;  in gemm_q8_1_2_1()
16039   const std::int32_t mul_result_chunk_stride_bytes = ((n * 4 + 7) / 8) * 8;  in gemm_q8_1_2_1()
16084                    const std::uint8_t* rhs, std::int32_t m, std::int32_t n,  in gemm_q8_1_2_2()
16085                    std::int32_t k, std::int32_t lhs_offset,  in gemm_q8_1_2_2()
16086                    std::int32_t rhs_offset, std::int32_t result_offset,  in gemm_q8_1_2_2()
16087                    std::int32_t multiplicative_offset, std::int32_t shift,  in gemm_q8_1_2_2()
16088                    std::uint8_t* result, std::int32_t result_stride) {  in gemm_q8_1_2_2()
16089   const std::int32_t row_chunks = m / 3;  in gemm_q8_1_2_2()
16090   const std::int32_t col_chunks = n / 3;  in gemm_q8_1_2_2()
16091   const std::int32_t padded_k = ((k + 7) / 8) * 8;  in gemm_q8_1_2_2()
16092   const std::int32_t chunk_size = k * 3;  in gemm_q8_1_2_2()
16093   const std::int32_t zipped_chunk_size = (padded_k + 16) * 3;  in gemm_q8_1_2_2()
16094   const std::int32_t zipped_rhs_size = (padded_k + 16) * n;  in gemm_q8_1_2_2()
16098   std::int32_t* zipped_lhs_3_offsets =  in gemm_q8_1_2_2()
16099       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 3);  in gemm_q8_1_2_2()
16100   std::int32_t* zipped_lhs_1_offsets =  in gemm_q8_1_2_2()
16101       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 1);  in gemm_q8_1_2_2()
16104   const std::int32_t result_chunk_stride = result_stride * 3;  in gemm_q8_1_2_2()
16106   const std::int32_t const_offset = lhs_offset * rhs_offset * k + result_offset;  in gemm_q8_1_2_2()
16107   const std::int32_t rounding_offset = (1 << (shift - 1));  in gemm_q8_1_2_2()
16108   std::int32_t* temp_result = reinterpret_cast<std::int32_t*>(  in gemm_q8_1_2_2()
16111   std::int32_t* mul_result_chunk = temp_result;  in gemm_q8_1_2_2()
16112   const std::int32_t mul_result_chunk_stride_bytes = ((n * 4 + 7) / 8) * 8;  in gemm_q8_1_2_2()
16157                    const std::uint8_t* rhs, std::int32_t m, std::int32_t n,  in gemm_q8_1_2_3()
16158                    std::int32_t k, std::int32_t lhs_offset,  in gemm_q8_1_2_3()
16159                    std::int32_t rhs_offset, std::int32_t result_offset,  in gemm_q8_1_2_3()
16160                    std::int32_t multiplicative_offset, std::int32_t shift,  in gemm_q8_1_2_3()
16161                    std::uint8_t* result, std::int32_t result_stride) {  in gemm_q8_1_2_3()
16162   const std::int32_t row_chunks = m / 3;  in gemm_q8_1_2_3()
16163   const std::int32_t col_chunks = n / 3;  in gemm_q8_1_2_3()
16164   const std::int32_t padded_k = ((k + 7) / 8) * 8;  in gemm_q8_1_2_3()
16165   const std::int32_t chunk_size = k * 3;  in gemm_q8_1_2_3()
16166   const std::int32_t zipped_chunk_size = (padded_k + 16) * 3;  in gemm_q8_1_2_3()
16167   const std::int32_t zipped_rhs_size = (padded_k + 16) * n;  in gemm_q8_1_2_3()
16171   std::int32_t* zipped_lhs_3_offsets =  in gemm_q8_1_2_3()
16172       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 3);  in gemm_q8_1_2_3()
16173   std::int32_t* zipped_lhs_1_offsets =  in gemm_q8_1_2_3()
16174       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 1);  in gemm_q8_1_2_3()
16177   const std::int32_t result_chunk_stride = result_stride * 3;  in gemm_q8_1_2_3()
16179   const std::int32_t const_offset = lhs_offset * rhs_offset * k + result_offset;  in gemm_q8_1_2_3()
16180   const std::int32_t rounding_offset = (1 << (shift - 1));  in gemm_q8_1_2_3()
16181   std::int32_t* temp_result = reinterpret_cast<std::int32_t*>(  in gemm_q8_1_2_3()
16184   std::int32_t* mul_result_chunk = temp_result;  in gemm_q8_1_2_3()
16185   const std::int32_t mul_result_chunk_stride_bytes = ((n * 4 + 7) / 8) * 8;  in gemm_q8_1_2_3()
16230                    const std::uint8_t* rhs, std::int32_t m, std::int32_t n,  in gemm_q8_1_2_4()
16231                    std::int32_t k, std::int32_t lhs_offset,  in gemm_q8_1_2_4()
16232                    std::int32_t rhs_offset, std::int32_t result_offset,  in gemm_q8_1_2_4()
16233                    std::int32_t multiplicative_offset, std::int32_t shift,  in gemm_q8_1_2_4()
16234                    std::uint8_t* result, std::int32_t result_stride) {  in gemm_q8_1_2_4()
16235   const std::int32_t row_chunks = m / 3;  in gemm_q8_1_2_4()
16236   const std::int32_t col_chunks = n / 3;  in gemm_q8_1_2_4()
16237   const std::int32_t padded_k = ((k + 7) / 8) * 8;  in gemm_q8_1_2_4()
16238   const std::int32_t chunk_size = k * 3;  in gemm_q8_1_2_4()
16239   const std::int32_t zipped_chunk_size = (padded_k + 16) * 3;  in gemm_q8_1_2_4()
16240   const std::int32_t zipped_rhs_size = (padded_k + 16) * n;  in gemm_q8_1_2_4()
16244   std::int32_t* zipped_lhs_3_offsets =  in gemm_q8_1_2_4()
16245       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 3);  in gemm_q8_1_2_4()
16246   std::int32_t* zipped_lhs_1_offsets =  in gemm_q8_1_2_4()
16247       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 1);  in gemm_q8_1_2_4()
16250   const std::int32_t result_chunk_stride = result_stride * 3;  in gemm_q8_1_2_4()
16252   const std::int32_t const_offset = lhs_offset * rhs_offset * k + result_offset;  in gemm_q8_1_2_4()
16253   const std::int32_t rounding_offset = (1 << (shift - 1));  in gemm_q8_1_2_4()
16254   std::int32_t* temp_result = reinterpret_cast<std::int32_t*>(  in gemm_q8_1_2_4()
16257   std::int32_t* mul_result_chunk = temp_result;  in gemm_q8_1_2_4()
16258   const std::int32_t mul_result_chunk_stride_bytes = ((n * 4 + 7) / 8) * 8;  in gemm_q8_1_2_4()
16303                    const std::uint8_t* rhs, std::int32_t m, std::int32_t n,  in gemm_q8_1_2_5()
16304                    std::int32_t k, std::int32_t lhs_offset,  in gemm_q8_1_2_5()
16305                    std::int32_t rhs_offset, std::int32_t result_offset,  in gemm_q8_1_2_5()
16306                    std::int32_t multiplicative_offset, std::int32_t shift,  in gemm_q8_1_2_5()
16307                    std::uint8_t* result, std::int32_t result_stride) {  in gemm_q8_1_2_5()
16308   const std::int32_t row_chunks = m / 3;  in gemm_q8_1_2_5()
16309   const std::int32_t col_chunks = n / 3;  in gemm_q8_1_2_5()
16310   const std::int32_t padded_k = ((k + 7) / 8) * 8;  in gemm_q8_1_2_5()
16311   const std::int32_t chunk_size = k * 3;  in gemm_q8_1_2_5()
16312   const std::int32_t zipped_chunk_size = (padded_k + 16) * 3;  in gemm_q8_1_2_5()
16313   const std::int32_t zipped_rhs_size = (padded_k + 16) * n;  in gemm_q8_1_2_5()
16317   std::int32_t* zipped_lhs_3_offsets =  in gemm_q8_1_2_5()
16318       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 3);  in gemm_q8_1_2_5()
16319   std::int32_t* zipped_lhs_1_offsets =  in gemm_q8_1_2_5()
16320       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 1);  in gemm_q8_1_2_5()
16323   const std::int32_t result_chunk_stride = result_stride * 3;  in gemm_q8_1_2_5()
16325   const std::int32_t const_offset = lhs_offset * rhs_offset * k + result_offset;  in gemm_q8_1_2_5()
16326   const std::int32_t rounding_offset = (1 << (shift - 1));  in gemm_q8_1_2_5()
16327   std::int32_t* temp_result = reinterpret_cast<std::int32_t*>(  in gemm_q8_1_2_5()
16330   std::int32_t* mul_result_chunk = temp_result;  in gemm_q8_1_2_5()
16331   const std::int32_t mul_result_chunk_stride_bytes = ((n * 4 + 7) / 8) * 8;  in gemm_q8_1_2_5()
16376                    const std::uint8_t* rhs, std::int32_t m, std::int32_t n,  in gemm_q8_1_2_6()
16377                    std::int32_t k, std::int32_t lhs_offset,  in gemm_q8_1_2_6()
16378                    std::int32_t rhs_offset, std::int32_t result_offset,  in gemm_q8_1_2_6()
16379                    std::int32_t multiplicative_offset, std::int32_t shift,  in gemm_q8_1_2_6()
16380                    std::uint8_t* result, std::int32_t result_stride) {  in gemm_q8_1_2_6()
16381   const std::int32_t row_chunks = m / 3;  in gemm_q8_1_2_6()
16382   const std::int32_t col_chunks = n / 3;  in gemm_q8_1_2_6()
16383   const std::int32_t padded_k = ((k + 7) / 8) * 8;  in gemm_q8_1_2_6()
16384   const std::int32_t chunk_size = k * 3;  in gemm_q8_1_2_6()
16385   const std::int32_t zipped_chunk_size = (padded_k + 16) * 3;  in gemm_q8_1_2_6()
16386   const std::int32_t zipped_rhs_size = (padded_k + 16) * n;  in gemm_q8_1_2_6()
16390   std::int32_t* zipped_lhs_3_offsets =  in gemm_q8_1_2_6()
16391       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 3);  in gemm_q8_1_2_6()
16392   std::int32_t* zipped_lhs_1_offsets =  in gemm_q8_1_2_6()
16393       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 1);  in gemm_q8_1_2_6()
16396   const std::int32_t result_chunk_stride = result_stride * 3;  in gemm_q8_1_2_6()
16398   const std::int32_t const_offset = lhs_offset * rhs_offset * k + result_offset;  in gemm_q8_1_2_6()
16399   const std::int32_t rounding_offset = (1 << (shift - 1));  in gemm_q8_1_2_6()
16400   std::int32_t* temp_result = reinterpret_cast<std::int32_t*>(  in gemm_q8_1_2_6()
16403   std::int32_t* mul_result_chunk = temp_result;  in gemm_q8_1_2_6()
16404   const std::int32_t mul_result_chunk_stride_bytes = ((n * 4 + 7) / 8) * 8;  in gemm_q8_1_2_6()
16449                    const std::uint8_t* rhs, std::int32_t m, std::int32_t n,  in gemm_q8_1_2_7()
16450                    std::int32_t k, std::int32_t lhs_offset,  in gemm_q8_1_2_7()
16451                    std::int32_t rhs_offset, std::int32_t result_offset,  in gemm_q8_1_2_7()
16452                    std::int32_t multiplicative_offset, std::int32_t shift,  in gemm_q8_1_2_7()
16453                    std::uint8_t* result, std::int32_t result_stride) {  in gemm_q8_1_2_7()
16454   const std::int32_t row_chunks = m / 3;  in gemm_q8_1_2_7()
16455   const std::int32_t col_chunks = n / 3;  in gemm_q8_1_2_7()
16456   const std::int32_t padded_k = ((k + 7) / 8) * 8;  in gemm_q8_1_2_7()
16457   const std::int32_t chunk_size = k * 3;  in gemm_q8_1_2_7()
16458   const std::int32_t zipped_chunk_size = (padded_k + 16) * 3;  in gemm_q8_1_2_7()
16459   const std::int32_t zipped_rhs_size = (padded_k + 16) * n;  in gemm_q8_1_2_7()
16463   std::int32_t* zipped_lhs_3_offsets =  in gemm_q8_1_2_7()
16464       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 3);  in gemm_q8_1_2_7()
16465   std::int32_t* zipped_lhs_1_offsets =  in gemm_q8_1_2_7()
16466       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 1);  in gemm_q8_1_2_7()
16469   const std::int32_t result_chunk_stride = result_stride * 3;  in gemm_q8_1_2_7()
16471   const std::int32_t const_offset = lhs_offset * rhs_offset * k + result_offset;  in gemm_q8_1_2_7()
16472   const std::int32_t rounding_offset = (1 << (shift - 1));  in gemm_q8_1_2_7()
16473   std::int32_t* temp_result = reinterpret_cast<std::int32_t*>(  in gemm_q8_1_2_7()
16476   std::int32_t* mul_result_chunk = temp_result;  in gemm_q8_1_2_7()
16477   const std::int32_t mul_result_chunk_stride_bytes = ((n * 4 + 7) / 8) * 8;  in gemm_q8_1_2_7()
16522                    const std::uint8_t* rhs, std::int32_t m, std::int32_t n,  in gemm_q8_2_0_0()
16523                    std::int32_t k, std::int32_t lhs_offset,  in gemm_q8_2_0_0()
16524                    std::int32_t rhs_offset, std::int32_t result_offset,  in gemm_q8_2_0_0()
16525                    std::int32_t multiplicative_offset, std::int32_t shift,  in gemm_q8_2_0_0()
16526                    std::uint8_t* result, std::int32_t result_stride) {  in gemm_q8_2_0_0()
16527   const std::int32_t row_chunks = m / 3;  in gemm_q8_2_0_0()
16528   const std::int32_t col_chunks = n / 3;  in gemm_q8_2_0_0()
16529   const std::int32_t padded_k = ((k + 7) / 8) * 8;  in gemm_q8_2_0_0()
16530   const std::int32_t chunk_size = k * 3;  in gemm_q8_2_0_0()
16531   const std::int32_t zipped_chunk_size = (padded_k + 16) * 3;  in gemm_q8_2_0_0()
16532   const std::int32_t zipped_rhs_size = (padded_k + 16) * n;  in gemm_q8_2_0_0()
16536   std::int32_t* zipped_lhs_3_offsets =  in gemm_q8_2_0_0()
16537       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 3);  in gemm_q8_2_0_0()
16538   std::int32_t* zipped_lhs_2_offsets =  in gemm_q8_2_0_0()
16539       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 2);  in gemm_q8_2_0_0()
16542   const std::int32_t result_chunk_stride = result_stride * 3;  in gemm_q8_2_0_0()
16544   const std::int32_t const_offset = lhs_offset * rhs_offset * k + result_offset;  in gemm_q8_2_0_0()
16545   const std::int32_t rounding_offset = (1 << (shift - 1));  in gemm_q8_2_0_0()
16546   std::int32_t* temp_result = reinterpret_cast<std::int32_t*>(  in gemm_q8_2_0_0()
16549   std::int32_t* mul_result_chunk = temp_result;  in gemm_q8_2_0_0()
16550   const std::int32_t mul_result_chunk_stride_bytes = ((n * 4 + 7) / 8) * 8;  in gemm_q8_2_0_0()
16590                    const std::uint8_t* rhs, std::int32_t m, std::int32_t n,  in gemm_q8_2_0_1()
16591                    std::int32_t k, std::int32_t lhs_offset,  in gemm_q8_2_0_1()
16592                    std::int32_t rhs_offset, std::int32_t result_offset,  in gemm_q8_2_0_1()
16593                    std::int32_t multiplicative_offset, std::int32_t shift,  in gemm_q8_2_0_1()
16594                    std::uint8_t* result, std::int32_t result_stride) {  in gemm_q8_2_0_1()
16595   const std::int32_t row_chunks = m / 3;  in gemm_q8_2_0_1()
16596   const std::int32_t col_chunks = n / 3;  in gemm_q8_2_0_1()
16597   const std::int32_t padded_k = ((k + 7) / 8) * 8;  in gemm_q8_2_0_1()
16598   const std::int32_t chunk_size = k * 3;  in gemm_q8_2_0_1()
16599   const std::int32_t zipped_chunk_size = (padded_k + 16) * 3;  in gemm_q8_2_0_1()
16600   const std::int32_t zipped_rhs_size = (padded_k + 16) * n;  in gemm_q8_2_0_1()
16604   std::int32_t* zipped_lhs_3_offsets =  in gemm_q8_2_0_1()
16605       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 3);  in gemm_q8_2_0_1()
16606   std::int32_t* zipped_lhs_2_offsets =  in gemm_q8_2_0_1()
16607       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 2);  in gemm_q8_2_0_1()
16610   const std::int32_t result_chunk_stride = result_stride * 3;  in gemm_q8_2_0_1()
16612   const std::int32_t const_offset = lhs_offset * rhs_offset * k + result_offset;  in gemm_q8_2_0_1()
16613   const std::int32_t rounding_offset = (1 << (shift - 1));  in gemm_q8_2_0_1()
16614   std::int32_t* temp_result = reinterpret_cast<std::int32_t*>(  in gemm_q8_2_0_1()
16617   std::int32_t* mul_result_chunk = temp_result;  in gemm_q8_2_0_1()
16618   const std::int32_t mul_result_chunk_stride_bytes = ((n * 4 + 7) / 8) * 8;  in gemm_q8_2_0_1()
16658                    const std::uint8_t* rhs, std::int32_t m, std::int32_t n,  in gemm_q8_2_0_2()
16659                    std::int32_t k, std::int32_t lhs_offset,  in gemm_q8_2_0_2()
16660                    std::int32_t rhs_offset, std::int32_t result_offset,  in gemm_q8_2_0_2()
16661                    std::int32_t multiplicative_offset, std::int32_t shift,  in gemm_q8_2_0_2()
16662                    std::uint8_t* result, std::int32_t result_stride) {  in gemm_q8_2_0_2()
16663   const std::int32_t row_chunks = m / 3;  in gemm_q8_2_0_2()
16664   const std::int32_t col_chunks = n / 3;  in gemm_q8_2_0_2()
16665   const std::int32_t padded_k = ((k + 7) / 8) * 8;  in gemm_q8_2_0_2()
16666   const std::int32_t chunk_size = k * 3;  in gemm_q8_2_0_2()
16667   const std::int32_t zipped_chunk_size = (padded_k + 16) * 3;  in gemm_q8_2_0_2()
16668   const std::int32_t zipped_rhs_size = (padded_k + 16) * n;  in gemm_q8_2_0_2()
16672   std::int32_t* zipped_lhs_3_offsets =  in gemm_q8_2_0_2()
16673       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 3);  in gemm_q8_2_0_2()
16674   std::int32_t* zipped_lhs_2_offsets =  in gemm_q8_2_0_2()
16675       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 2);  in gemm_q8_2_0_2()
16678   const std::int32_t result_chunk_stride = result_stride * 3;  in gemm_q8_2_0_2()
16680   const std::int32_t const_offset = lhs_offset * rhs_offset * k + result_offset;  in gemm_q8_2_0_2()
16681   const std::int32_t rounding_offset = (1 << (shift - 1));  in gemm_q8_2_0_2()
16682   std::int32_t* temp_result = reinterpret_cast<std::int32_t*>(  in gemm_q8_2_0_2()
16685   std::int32_t* mul_result_chunk = temp_result;  in gemm_q8_2_0_2()
16686   const std::int32_t mul_result_chunk_stride_bytes = ((n * 4 + 7) / 8) * 8;  in gemm_q8_2_0_2()
16726                    const std::uint8_t* rhs, std::int32_t m, std::int32_t n,  in gemm_q8_2_0_3()
16727                    std::int32_t k, std::int32_t lhs_offset,  in gemm_q8_2_0_3()
16728                    std::int32_t rhs_offset, std::int32_t result_offset,  in gemm_q8_2_0_3()
16729                    std::int32_t multiplicative_offset, std::int32_t shift,  in gemm_q8_2_0_3()
16730                    std::uint8_t* result, std::int32_t result_stride) {  in gemm_q8_2_0_3()
16731   const std::int32_t row_chunks = m / 3;  in gemm_q8_2_0_3()
16732   const std::int32_t col_chunks = n / 3;  in gemm_q8_2_0_3()
16733   const std::int32_t padded_k = ((k + 7) / 8) * 8;  in gemm_q8_2_0_3()
16734   const std::int32_t chunk_size = k * 3;  in gemm_q8_2_0_3()
16735   const std::int32_t zipped_chunk_size = (padded_k + 16) * 3;  in gemm_q8_2_0_3()
16736   const std::int32_t zipped_rhs_size = (padded_k + 16) * n;  in gemm_q8_2_0_3()
16740   std::int32_t* zipped_lhs_3_offsets =  in gemm_q8_2_0_3()
16741       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 3);  in gemm_q8_2_0_3()
16742   std::int32_t* zipped_lhs_2_offsets =  in gemm_q8_2_0_3()
16743       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 2);  in gemm_q8_2_0_3()
16746   const std::int32_t result_chunk_stride = result_stride * 3;  in gemm_q8_2_0_3()
16748   const std::int32_t const_offset = lhs_offset * rhs_offset * k + result_offset;  in gemm_q8_2_0_3()
16749   const std::int32_t rounding_offset = (1 << (shift - 1));  in gemm_q8_2_0_3()
16750   std::int32_t* temp_result = reinterpret_cast<std::int32_t*>(  in gemm_q8_2_0_3()
16753   std::int32_t* mul_result_chunk = temp_result;  in gemm_q8_2_0_3()
16754   const std::int32_t mul_result_chunk_stride_bytes = ((n * 4 + 7) / 8) * 8;  in gemm_q8_2_0_3()
16794                    const std::uint8_t* rhs, std::int32_t m, std::int32_t n,  in gemm_q8_2_0_4()
16795                    std::int32_t k, std::int32_t lhs_offset,  in gemm_q8_2_0_4()
16796                    std::int32_t rhs_offset, std::int32_t result_offset,  in gemm_q8_2_0_4()
16797                    std::int32_t multiplicative_offset, std::int32_t shift,  in gemm_q8_2_0_4()
16798                    std::uint8_t* result, std::int32_t result_stride) {  in gemm_q8_2_0_4()
16799   const std::int32_t row_chunks = m / 3;  in gemm_q8_2_0_4()
16800   const std::int32_t col_chunks = n / 3;  in gemm_q8_2_0_4()
16801   const std::int32_t padded_k = ((k + 7) / 8) * 8;  in gemm_q8_2_0_4()
16802   const std::int32_t chunk_size = k * 3;  in gemm_q8_2_0_4()
16803   const std::int32_t zipped_chunk_size = (padded_k + 16) * 3;  in gemm_q8_2_0_4()
16804   const std::int32_t zipped_rhs_size = (padded_k + 16) * n;  in gemm_q8_2_0_4()
16808   std::int32_t* zipped_lhs_3_offsets =  in gemm_q8_2_0_4()
16809       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 3);  in gemm_q8_2_0_4()
16810   std::int32_t* zipped_lhs_2_offsets =  in gemm_q8_2_0_4()
16811       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 2);  in gemm_q8_2_0_4()
16814   const std::int32_t result_chunk_stride = result_stride * 3;  in gemm_q8_2_0_4()
16816   const std::int32_t const_offset = lhs_offset * rhs_offset * k + result_offset;  in gemm_q8_2_0_4()
16817   const std::int32_t rounding_offset = (1 << (shift - 1));  in gemm_q8_2_0_4()
16818   std::int32_t* temp_result = reinterpret_cast<std::int32_t*>(  in gemm_q8_2_0_4()
16821   std::int32_t* mul_result_chunk = temp_result;  in gemm_q8_2_0_4()
16822   const std::int32_t mul_result_chunk_stride_bytes = ((n * 4 + 7) / 8) * 8;  in gemm_q8_2_0_4()
16862                    const std::uint8_t* rhs, std::int32_t m, std::int32_t n,  in gemm_q8_2_0_5()
16863                    std::int32_t k, std::int32_t lhs_offset,  in gemm_q8_2_0_5()
16864                    std::int32_t rhs_offset, std::int32_t result_offset,  in gemm_q8_2_0_5()
16865                    std::int32_t multiplicative_offset, std::int32_t shift,  in gemm_q8_2_0_5()
16866                    std::uint8_t* result, std::int32_t result_stride) {  in gemm_q8_2_0_5()
16867   const std::int32_t row_chunks = m / 3;  in gemm_q8_2_0_5()
16868   const std::int32_t col_chunks = n / 3;  in gemm_q8_2_0_5()
16869   const std::int32_t padded_k = ((k + 7) / 8) * 8;  in gemm_q8_2_0_5()
16870   const std::int32_t chunk_size = k * 3;  in gemm_q8_2_0_5()
16871   const std::int32_t zipped_chunk_size = (padded_k + 16) * 3;  in gemm_q8_2_0_5()
16872   const std::int32_t zipped_rhs_size = (padded_k + 16) * n;  in gemm_q8_2_0_5()
16876   std::int32_t* zipped_lhs_3_offsets =  in gemm_q8_2_0_5()
16877       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 3);  in gemm_q8_2_0_5()
16878   std::int32_t* zipped_lhs_2_offsets =  in gemm_q8_2_0_5()
16879       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 2);  in gemm_q8_2_0_5()
16882   const std::int32_t result_chunk_stride = result_stride * 3;  in gemm_q8_2_0_5()
16884   const std::int32_t const_offset = lhs_offset * rhs_offset * k + result_offset;  in gemm_q8_2_0_5()
16885   const std::int32_t rounding_offset = (1 << (shift - 1));  in gemm_q8_2_0_5()
16886   std::int32_t* temp_result = reinterpret_cast<std::int32_t*>(  in gemm_q8_2_0_5()
16889   std::int32_t* mul_result_chunk = temp_result;  in gemm_q8_2_0_5()
16890   const std::int32_t mul_result_chunk_stride_bytes = ((n * 4 + 7) / 8) * 8;  in gemm_q8_2_0_5()
16930                    const std::uint8_t* rhs, std::int32_t m, std::int32_t n,  in gemm_q8_2_0_6()
16931                    std::int32_t k, std::int32_t lhs_offset,  in gemm_q8_2_0_6()
16932                    std::int32_t rhs_offset, std::int32_t result_offset,  in gemm_q8_2_0_6()
16933                    std::int32_t multiplicative_offset, std::int32_t shift,  in gemm_q8_2_0_6()
16934                    std::uint8_t* result, std::int32_t result_stride) {  in gemm_q8_2_0_6()
16935   const std::int32_t row_chunks = m / 3;  in gemm_q8_2_0_6()
16936   const std::int32_t col_chunks = n / 3;  in gemm_q8_2_0_6()
16937   const std::int32_t padded_k = ((k + 7) / 8) * 8;  in gemm_q8_2_0_6()
16938   const std::int32_t chunk_size = k * 3;  in gemm_q8_2_0_6()
16939   const std::int32_t zipped_chunk_size = (padded_k + 16) * 3;  in gemm_q8_2_0_6()
16940   const std::int32_t zipped_rhs_size = (padded_k + 16) * n;  in gemm_q8_2_0_6()
16944   std::int32_t* zipped_lhs_3_offsets =  in gemm_q8_2_0_6()
16945       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 3);  in gemm_q8_2_0_6()
16946   std::int32_t* zipped_lhs_2_offsets =  in gemm_q8_2_0_6()
16947       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 2);  in gemm_q8_2_0_6()
16950   const std::int32_t result_chunk_stride = result_stride * 3;  in gemm_q8_2_0_6()
16952   const std::int32_t const_offset = lhs_offset * rhs_offset * k + result_offset;  in gemm_q8_2_0_6()
16953   const std::int32_t rounding_offset = (1 << (shift - 1));  in gemm_q8_2_0_6()
16954   std::int32_t* temp_result = reinterpret_cast<std::int32_t*>(  in gemm_q8_2_0_6()
16957   std::int32_t* mul_result_chunk = temp_result;  in gemm_q8_2_0_6()
16958   const std::int32_t mul_result_chunk_stride_bytes = ((n * 4 + 7) / 8) * 8;  in gemm_q8_2_0_6()
16998                    const std::uint8_t* rhs, std::int32_t m, std::int32_t n,  in gemm_q8_2_0_7()
16999                    std::int32_t k, std::int32_t lhs_offset,  in gemm_q8_2_0_7()
17000                    std::int32_t rhs_offset, std::int32_t result_offset,  in gemm_q8_2_0_7()
17001                    std::int32_t multiplicative_offset, std::int32_t shift,  in gemm_q8_2_0_7()
17002                    std::uint8_t* result, std::int32_t result_stride) {  in gemm_q8_2_0_7()
17003   const std::int32_t row_chunks = m / 3;  in gemm_q8_2_0_7()
17004   const std::int32_t col_chunks = n / 3;  in gemm_q8_2_0_7()
17005   const std::int32_t padded_k = ((k + 7) / 8) * 8;  in gemm_q8_2_0_7()
17006   const std::int32_t chunk_size = k * 3;  in gemm_q8_2_0_7()
17007   const std::int32_t zipped_chunk_size = (padded_k + 16) * 3;  in gemm_q8_2_0_7()
17008   const std::int32_t zipped_rhs_size = (padded_k + 16) * n;  in gemm_q8_2_0_7()
17012   std::int32_t* zipped_lhs_3_offsets =  in gemm_q8_2_0_7()
17013       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 3);  in gemm_q8_2_0_7()
17014   std::int32_t* zipped_lhs_2_offsets =  in gemm_q8_2_0_7()
17015       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 2);  in gemm_q8_2_0_7()
17018   const std::int32_t result_chunk_stride = result_stride * 3;  in gemm_q8_2_0_7()
17020   const std::int32_t const_offset = lhs_offset * rhs_offset * k + result_offset;  in gemm_q8_2_0_7()
17021   const std::int32_t rounding_offset = (1 << (shift - 1));  in gemm_q8_2_0_7()
17022   std::int32_t* temp_result = reinterpret_cast<std::int32_t*>(  in gemm_q8_2_0_7()
17025   std::int32_t* mul_result_chunk = temp_result;  in gemm_q8_2_0_7()
17026   const std::int32_t mul_result_chunk_stride_bytes = ((n * 4 + 7) / 8) * 8;  in gemm_q8_2_0_7()
17066                    const std::uint8_t* rhs, std::int32_t m, std::int32_t n,  in gemm_q8_2_1_0()
17067                    std::int32_t k, std::int32_t lhs_offset,  in gemm_q8_2_1_0()
17068                    std::int32_t rhs_offset, std::int32_t result_offset,  in gemm_q8_2_1_0()
17069                    std::int32_t multiplicative_offset, std::int32_t shift,  in gemm_q8_2_1_0()
17070                    std::uint8_t* result, std::int32_t result_stride) {  in gemm_q8_2_1_0()
17071   const std::int32_t row_chunks = m / 3;  in gemm_q8_2_1_0()
17072   const std::int32_t col_chunks = n / 3;  in gemm_q8_2_1_0()
17073   const std::int32_t padded_k = ((k + 7) / 8) * 8;  in gemm_q8_2_1_0()
17074   const std::int32_t chunk_size = k * 3;  in gemm_q8_2_1_0()
17075   const std::int32_t zipped_chunk_size = (padded_k + 16) * 3;  in gemm_q8_2_1_0()
17076   const std::int32_t zipped_rhs_size = (padded_k + 16) * n;  in gemm_q8_2_1_0()
17080   std::int32_t* zipped_lhs_3_offsets =  in gemm_q8_2_1_0()
17081       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 3);  in gemm_q8_2_1_0()
17082   std::int32_t* zipped_lhs_2_offsets =  in gemm_q8_2_1_0()
17083       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 2);  in gemm_q8_2_1_0()
17086   const std::int32_t result_chunk_stride = result_stride * 3;  in gemm_q8_2_1_0()
17088   const std::int32_t const_offset = lhs_offset * rhs_offset * k + result_offset;  in gemm_q8_2_1_0()
17089   const std::int32_t rounding_offset = (1 << (shift - 1));  in gemm_q8_2_1_0()
17090   std::int32_t* temp_result = reinterpret_cast<std::int32_t*>(  in gemm_q8_2_1_0()
17093   std::int32_t* mul_result_chunk = temp_result;  in gemm_q8_2_1_0()
17094   const std::int32_t mul_result_chunk_stride_bytes = ((n * 4 + 7) / 8) * 8;  in gemm_q8_2_1_0()
17139                    const std::uint8_t* rhs, std::int32_t m, std::int32_t n,  in gemm_q8_2_1_1()
17140                    std::int32_t k, std::int32_t lhs_offset,  in gemm_q8_2_1_1()
17141                    std::int32_t rhs_offset, std::int32_t result_offset,  in gemm_q8_2_1_1()
17142                    std::int32_t multiplicative_offset, std::int32_t shift,  in gemm_q8_2_1_1()
17143                    std::uint8_t* result, std::int32_t result_stride) {  in gemm_q8_2_1_1()
17144   const std::int32_t row_chunks = m / 3;  in gemm_q8_2_1_1()
17145   const std::int32_t col_chunks = n / 3;  in gemm_q8_2_1_1()
17146   const std::int32_t padded_k = ((k + 7) / 8) * 8;  in gemm_q8_2_1_1()
17147   const std::int32_t chunk_size = k * 3;  in gemm_q8_2_1_1()
17148   const std::int32_t zipped_chunk_size = (padded_k + 16) * 3;  in gemm_q8_2_1_1()
17149   const std::int32_t zipped_rhs_size = (padded_k + 16) * n;  in gemm_q8_2_1_1()
17153   std::int32_t* zipped_lhs_3_offsets =  in gemm_q8_2_1_1()
17154       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 3);  in gemm_q8_2_1_1()
17155   std::int32_t* zipped_lhs_2_offsets =  in gemm_q8_2_1_1()
17156       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 2);  in gemm_q8_2_1_1()
17159   const std::int32_t result_chunk_stride = result_stride * 3;  in gemm_q8_2_1_1()
17161   const std::int32_t const_offset = lhs_offset * rhs_offset * k + result_offset;  in gemm_q8_2_1_1()
17162   const std::int32_t rounding_offset = (1 << (shift - 1));  in gemm_q8_2_1_1()
17163   std::int32_t* temp_result = reinterpret_cast<std::int32_t*>(  in gemm_q8_2_1_1()
17166   std::int32_t* mul_result_chunk = temp_result;  in gemm_q8_2_1_1()
17167   const std::int32_t mul_result_chunk_stride_bytes = ((n * 4 + 7) / 8) * 8;  in gemm_q8_2_1_1()
17212                    const std::uint8_t* rhs, std::int32_t m, std::int32_t n,  in gemm_q8_2_1_2()
17213                    std::int32_t k, std::int32_t lhs_offset,  in gemm_q8_2_1_2()
17214                    std::int32_t rhs_offset, std::int32_t result_offset,  in gemm_q8_2_1_2()
17215                    std::int32_t multiplicative_offset, std::int32_t shift,  in gemm_q8_2_1_2()
17216                    std::uint8_t* result, std::int32_t result_stride) {  in gemm_q8_2_1_2()
17217   const std::int32_t row_chunks = m / 3;  in gemm_q8_2_1_2()
17218   const std::int32_t col_chunks = n / 3;  in gemm_q8_2_1_2()
17219   const std::int32_t padded_k = ((k + 7) / 8) * 8;  in gemm_q8_2_1_2()
17220   const std::int32_t chunk_size = k * 3;  in gemm_q8_2_1_2()
17221   const std::int32_t zipped_chunk_size = (padded_k + 16) * 3;  in gemm_q8_2_1_2()
17222   const std::int32_t zipped_rhs_size = (padded_k + 16) * n;  in gemm_q8_2_1_2()
17226   std::int32_t* zipped_lhs_3_offsets =  in gemm_q8_2_1_2()
17227       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 3);  in gemm_q8_2_1_2()
17228   std::int32_t* zipped_lhs_2_offsets =  in gemm_q8_2_1_2()
17229       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 2);  in gemm_q8_2_1_2()
17232   const std::int32_t result_chunk_stride = result_stride * 3;  in gemm_q8_2_1_2()
17234   const std::int32_t const_offset = lhs_offset * rhs_offset * k + result_offset;  in gemm_q8_2_1_2()
17235   const std::int32_t rounding_offset = (1 << (shift - 1));  in gemm_q8_2_1_2()
17236   std::int32_t* temp_result = reinterpret_cast<std::int32_t*>(  in gemm_q8_2_1_2()
17239   std::int32_t* mul_result_chunk = temp_result;  in gemm_q8_2_1_2()
17240   const std::int32_t mul_result_chunk_stride_bytes = ((n * 4 + 7) / 8) * 8;  in gemm_q8_2_1_2()
17285                    const std::uint8_t* rhs, std::int32_t m, std::int32_t n,  in gemm_q8_2_1_3()
17286                    std::int32_t k, std::int32_t lhs_offset,  in gemm_q8_2_1_3()
17287                    std::int32_t rhs_offset, std::int32_t result_offset,  in gemm_q8_2_1_3()
17288                    std::int32_t multiplicative_offset, std::int32_t shift,  in gemm_q8_2_1_3()
17289                    std::uint8_t* result, std::int32_t result_stride) {  in gemm_q8_2_1_3()
17290   const std::int32_t row_chunks = m / 3;  in gemm_q8_2_1_3()
17291   const std::int32_t col_chunks = n / 3;  in gemm_q8_2_1_3()
17292   const std::int32_t padded_k = ((k + 7) / 8) * 8;  in gemm_q8_2_1_3()
17293   const std::int32_t chunk_size = k * 3;  in gemm_q8_2_1_3()
17294   const std::int32_t zipped_chunk_size = (padded_k + 16) * 3;  in gemm_q8_2_1_3()
17295   const std::int32_t zipped_rhs_size = (padded_k + 16) * n;  in gemm_q8_2_1_3()
17299   std::int32_t* zipped_lhs_3_offsets =  in gemm_q8_2_1_3()
17300       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 3);  in gemm_q8_2_1_3()
17301   std::int32_t* zipped_lhs_2_offsets =  in gemm_q8_2_1_3()
17302       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 2);  in gemm_q8_2_1_3()
17305   const std::int32_t result_chunk_stride = result_stride * 3;  in gemm_q8_2_1_3()
17307   const std::int32_t const_offset = lhs_offset * rhs_offset * k + result_offset;  in gemm_q8_2_1_3()
17308   const std::int32_t rounding_offset = (1 << (shift - 1));  in gemm_q8_2_1_3()
17309   std::int32_t* temp_result = reinterpret_cast<std::int32_t*>(  in gemm_q8_2_1_3()
17312   std::int32_t* mul_result_chunk = temp_result;  in gemm_q8_2_1_3()
17313   const std::int32_t mul_result_chunk_stride_bytes = ((n * 4 + 7) / 8) * 8;  in gemm_q8_2_1_3()
17358                    const std::uint8_t* rhs, std::int32_t m, std::int32_t n,  in gemm_q8_2_1_4()
17359                    std::int32_t k, std::int32_t lhs_offset,  in gemm_q8_2_1_4()
17360                    std::int32_t rhs_offset, std::int32_t result_offset,  in gemm_q8_2_1_4()
17361                    std::int32_t multiplicative_offset, std::int32_t shift,  in gemm_q8_2_1_4()
17362                    std::uint8_t* result, std::int32_t result_stride) {  in gemm_q8_2_1_4()
17363   const std::int32_t row_chunks = m / 3;  in gemm_q8_2_1_4()
17364   const std::int32_t col_chunks = n / 3;  in gemm_q8_2_1_4()
17365   const std::int32_t padded_k = ((k + 7) / 8) * 8;  in gemm_q8_2_1_4()
17366   const std::int32_t chunk_size = k * 3;  in gemm_q8_2_1_4()
17367   const std::int32_t zipped_chunk_size = (padded_k + 16) * 3;  in gemm_q8_2_1_4()
17368   const std::int32_t zipped_rhs_size = (padded_k + 16) * n;  in gemm_q8_2_1_4()
17372   std::int32_t* zipped_lhs_3_offsets =  in gemm_q8_2_1_4()
17373       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 3);  in gemm_q8_2_1_4()
17374   std::int32_t* zipped_lhs_2_offsets =  in gemm_q8_2_1_4()
17375       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 2);  in gemm_q8_2_1_4()
17378   const std::int32_t result_chunk_stride = result_stride * 3;  in gemm_q8_2_1_4()
17380   const std::int32_t const_offset = lhs_offset * rhs_offset * k + result_offset;  in gemm_q8_2_1_4()
17381   const std::int32_t rounding_offset = (1 << (shift - 1));  in gemm_q8_2_1_4()
17382   std::int32_t* temp_result = reinterpret_cast<std::int32_t*>(  in gemm_q8_2_1_4()
17385   std::int32_t* mul_result_chunk = temp_result;  in gemm_q8_2_1_4()
17386   const std::int32_t mul_result_chunk_stride_bytes = ((n * 4 + 7) / 8) * 8;  in gemm_q8_2_1_4()
17431                    const std::uint8_t* rhs, std::int32_t m, std::int32_t n,  in gemm_q8_2_1_5()
17432                    std::int32_t k, std::int32_t lhs_offset,  in gemm_q8_2_1_5()
17433                    std::int32_t rhs_offset, std::int32_t result_offset,  in gemm_q8_2_1_5()
17434                    std::int32_t multiplicative_offset, std::int32_t shift,  in gemm_q8_2_1_5()
17435                    std::uint8_t* result, std::int32_t result_stride) {  in gemm_q8_2_1_5()
17436   const std::int32_t row_chunks = m / 3;  in gemm_q8_2_1_5()
17437   const std::int32_t col_chunks = n / 3;  in gemm_q8_2_1_5()
17438   const std::int32_t padded_k = ((k + 7) / 8) * 8;  in gemm_q8_2_1_5()
17439   const std::int32_t chunk_size = k * 3;  in gemm_q8_2_1_5()
17440   const std::int32_t zipped_chunk_size = (padded_k + 16) * 3;  in gemm_q8_2_1_5()
17441   const std::int32_t zipped_rhs_size = (padded_k + 16) * n;  in gemm_q8_2_1_5()
17445   std::int32_t* zipped_lhs_3_offsets =  in gemm_q8_2_1_5()
17446       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 3);  in gemm_q8_2_1_5()
17447   std::int32_t* zipped_lhs_2_offsets =  in gemm_q8_2_1_5()
17448       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 2);  in gemm_q8_2_1_5()
17451   const std::int32_t result_chunk_stride = result_stride * 3;  in gemm_q8_2_1_5()
17453   const std::int32_t const_offset = lhs_offset * rhs_offset * k + result_offset;  in gemm_q8_2_1_5()
17454   const std::int32_t rounding_offset = (1 << (shift - 1));  in gemm_q8_2_1_5()
17455   std::int32_t* temp_result = reinterpret_cast<std::int32_t*>(  in gemm_q8_2_1_5()
17458   std::int32_t* mul_result_chunk = temp_result;  in gemm_q8_2_1_5()
17459   const std::int32_t mul_result_chunk_stride_bytes = ((n * 4 + 7) / 8) * 8;  in gemm_q8_2_1_5()
17504                    const std::uint8_t* rhs, std::int32_t m, std::int32_t n,  in gemm_q8_2_1_6()
17505                    std::int32_t k, std::int32_t lhs_offset,  in gemm_q8_2_1_6()
17506                    std::int32_t rhs_offset, std::int32_t result_offset,  in gemm_q8_2_1_6()
17507                    std::int32_t multiplicative_offset, std::int32_t shift,  in gemm_q8_2_1_6()
17508                    std::uint8_t* result, std::int32_t result_stride) {  in gemm_q8_2_1_6()
17509   const std::int32_t row_chunks = m / 3;  in gemm_q8_2_1_6()
17510   const std::int32_t col_chunks = n / 3;  in gemm_q8_2_1_6()
17511   const std::int32_t padded_k = ((k + 7) / 8) * 8;  in gemm_q8_2_1_6()
17512   const std::int32_t chunk_size = k * 3;  in gemm_q8_2_1_6()
17513   const std::int32_t zipped_chunk_size = (padded_k + 16) * 3;  in gemm_q8_2_1_6()
17514   const std::int32_t zipped_rhs_size = (padded_k + 16) * n;  in gemm_q8_2_1_6()
17518   std::int32_t* zipped_lhs_3_offsets =  in gemm_q8_2_1_6()
17519       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 3);  in gemm_q8_2_1_6()
17520   std::int32_t* zipped_lhs_2_offsets =  in gemm_q8_2_1_6()
17521       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 2);  in gemm_q8_2_1_6()
17524   const std::int32_t result_chunk_stride = result_stride * 3;  in gemm_q8_2_1_6()
17526   const std::int32_t const_offset = lhs_offset * rhs_offset * k + result_offset;  in gemm_q8_2_1_6()
17527   const std::int32_t rounding_offset = (1 << (shift - 1));  in gemm_q8_2_1_6()
17528   std::int32_t* temp_result = reinterpret_cast<std::int32_t*>(  in gemm_q8_2_1_6()
17531   std::int32_t* mul_result_chunk = temp_result;  in gemm_q8_2_1_6()
17532   const std::int32_t mul_result_chunk_stride_bytes = ((n * 4 + 7) / 8) * 8;  in gemm_q8_2_1_6()
17577                    const std::uint8_t* rhs, std::int32_t m, std::int32_t n,  in gemm_q8_2_1_7()
17578                    std::int32_t k, std::int32_t lhs_offset,  in gemm_q8_2_1_7()
17579                    std::int32_t rhs_offset, std::int32_t result_offset,  in gemm_q8_2_1_7()
17580                    std::int32_t multiplicative_offset, std::int32_t shift,  in gemm_q8_2_1_7()
17581                    std::uint8_t* result, std::int32_t result_stride) {  in gemm_q8_2_1_7()
17582   const std::int32_t row_chunks = m / 3;  in gemm_q8_2_1_7()
17583   const std::int32_t col_chunks = n / 3;  in gemm_q8_2_1_7()
17584   const std::int32_t padded_k = ((k + 7) / 8) * 8;  in gemm_q8_2_1_7()
17585   const std::int32_t chunk_size = k * 3;  in gemm_q8_2_1_7()
17586   const std::int32_t zipped_chunk_size = (padded_k + 16) * 3;  in gemm_q8_2_1_7()
17587   const std::int32_t zipped_rhs_size = (padded_k + 16) * n;  in gemm_q8_2_1_7()
17591   std::int32_t* zipped_lhs_3_offsets =  in gemm_q8_2_1_7()
17592       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 3);  in gemm_q8_2_1_7()
17593   std::int32_t* zipped_lhs_2_offsets =  in gemm_q8_2_1_7()
17594       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 2);  in gemm_q8_2_1_7()
17597   const std::int32_t result_chunk_stride = result_stride * 3;  in gemm_q8_2_1_7()
17599   const std::int32_t const_offset = lhs_offset * rhs_offset * k + result_offset;  in gemm_q8_2_1_7()
17600   const std::int32_t rounding_offset = (1 << (shift - 1));  in gemm_q8_2_1_7()
17601   std::int32_t* temp_result = reinterpret_cast<std::int32_t*>(  in gemm_q8_2_1_7()
17604   std::int32_t* mul_result_chunk = temp_result;  in gemm_q8_2_1_7()
17605   const std::int32_t mul_result_chunk_stride_bytes = ((n * 4 + 7) / 8) * 8;  in gemm_q8_2_1_7()
17650                    const std::uint8_t* rhs, std::int32_t m, std::int32_t n,  in gemm_q8_2_2_0()
17651                    std::int32_t k, std::int32_t lhs_offset,  in gemm_q8_2_2_0()
17652                    std::int32_t rhs_offset, std::int32_t result_offset,  in gemm_q8_2_2_0()
17653                    std::int32_t multiplicative_offset, std::int32_t shift,  in gemm_q8_2_2_0()
17654                    std::uint8_t* result, std::int32_t result_stride) {  in gemm_q8_2_2_0()
17655   const std::int32_t row_chunks = m / 3;  in gemm_q8_2_2_0()
17656   const std::int32_t col_chunks = n / 3;  in gemm_q8_2_2_0()
17657   const std::int32_t padded_k = ((k + 7) / 8) * 8;  in gemm_q8_2_2_0()
17658   const std::int32_t chunk_size = k * 3;  in gemm_q8_2_2_0()
17659   const std::int32_t zipped_chunk_size = (padded_k + 16) * 3;  in gemm_q8_2_2_0()
17660   const std::int32_t zipped_rhs_size = (padded_k + 16) * n;  in gemm_q8_2_2_0()
17664   std::int32_t* zipped_lhs_3_offsets =  in gemm_q8_2_2_0()
17665       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 3);  in gemm_q8_2_2_0()
17666   std::int32_t* zipped_lhs_2_offsets =  in gemm_q8_2_2_0()
17667       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 2);  in gemm_q8_2_2_0()
17670   const std::int32_t result_chunk_stride = result_stride * 3;  in gemm_q8_2_2_0()
17672   const std::int32_t const_offset = lhs_offset * rhs_offset * k + result_offset;  in gemm_q8_2_2_0()
17673   const std::int32_t rounding_offset = (1 << (shift - 1));  in gemm_q8_2_2_0()
17674   std::int32_t* temp_result = reinterpret_cast<std::int32_t*>(  in gemm_q8_2_2_0()
17677   std::int32_t* mul_result_chunk = temp_result;  in gemm_q8_2_2_0()
17678   const std::int32_t mul_result_chunk_stride_bytes = ((n * 4 + 7) / 8) * 8;  in gemm_q8_2_2_0()
17723                    const std::uint8_t* rhs, std::int32_t m, std::int32_t n,  in gemm_q8_2_2_1()
17724                    std::int32_t k, std::int32_t lhs_offset,  in gemm_q8_2_2_1()
17725                    std::int32_t rhs_offset, std::int32_t result_offset,  in gemm_q8_2_2_1()
17726                    std::int32_t multiplicative_offset, std::int32_t shift,  in gemm_q8_2_2_1()
17727                    std::uint8_t* result, std::int32_t result_stride) {  in gemm_q8_2_2_1()
17728   const std::int32_t row_chunks = m / 3;  in gemm_q8_2_2_1()
17729   const std::int32_t col_chunks = n / 3;  in gemm_q8_2_2_1()
17730   const std::int32_t padded_k = ((k + 7) / 8) * 8;  in gemm_q8_2_2_1()
17731   const std::int32_t chunk_size = k * 3;  in gemm_q8_2_2_1()
17732   const std::int32_t zipped_chunk_size = (padded_k + 16) * 3;  in gemm_q8_2_2_1()
17733   const std::int32_t zipped_rhs_size = (padded_k + 16) * n;  in gemm_q8_2_2_1()
17737   std::int32_t* zipped_lhs_3_offsets =  in gemm_q8_2_2_1()
17738       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 3);  in gemm_q8_2_2_1()
17739   std::int32_t* zipped_lhs_2_offsets =  in gemm_q8_2_2_1()
17740       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 2);  in gemm_q8_2_2_1()
17743   const std::int32_t result_chunk_stride = result_stride * 3;  in gemm_q8_2_2_1()
17745   const std::int32_t const_offset = lhs_offset * rhs_offset * k + result_offset;  in gemm_q8_2_2_1()
17746   const std::int32_t rounding_offset = (1 << (shift - 1));  in gemm_q8_2_2_1()
17747   std::int32_t* temp_result = reinterpret_cast<std::int32_t*>(  in gemm_q8_2_2_1()
17750   std::int32_t* mul_result_chunk = temp_result;  in gemm_q8_2_2_1()
17751   const std::int32_t mul_result_chunk_stride_bytes = ((n * 4 + 7) / 8) * 8;  in gemm_q8_2_2_1()
17796                    const std::uint8_t* rhs, std::int32_t m, std::int32_t n,  in gemm_q8_2_2_2()
17797                    std::int32_t k, std::int32_t lhs_offset,  in gemm_q8_2_2_2()
17798                    std::int32_t rhs_offset, std::int32_t result_offset,  in gemm_q8_2_2_2()
17799                    std::int32_t multiplicative_offset, std::int32_t shift,  in gemm_q8_2_2_2()
17800                    std::uint8_t* result, std::int32_t result_stride) {  in gemm_q8_2_2_2()
17801   const std::int32_t row_chunks = m / 3;  in gemm_q8_2_2_2()
17802   const std::int32_t col_chunks = n / 3;  in gemm_q8_2_2_2()
17803   const std::int32_t padded_k = ((k + 7) / 8) * 8;  in gemm_q8_2_2_2()
17804   const std::int32_t chunk_size = k * 3;  in gemm_q8_2_2_2()
17805   const std::int32_t zipped_chunk_size = (padded_k + 16) * 3;  in gemm_q8_2_2_2()
17806   const std::int32_t zipped_rhs_size = (padded_k + 16) * n;  in gemm_q8_2_2_2()
17810   std::int32_t* zipped_lhs_3_offsets =  in gemm_q8_2_2_2()
17811       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 3);  in gemm_q8_2_2_2()
17812   std::int32_t* zipped_lhs_2_offsets =  in gemm_q8_2_2_2()
17813       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 2);  in gemm_q8_2_2_2()
17816   const std::int32_t result_chunk_stride = result_stride * 3;  in gemm_q8_2_2_2()
17818   const std::int32_t const_offset = lhs_offset * rhs_offset * k + result_offset;  in gemm_q8_2_2_2()
17819   const std::int32_t rounding_offset = (1 << (shift - 1));  in gemm_q8_2_2_2()
17820   std::int32_t* temp_result = reinterpret_cast<std::int32_t*>(  in gemm_q8_2_2_2()
17823   std::int32_t* mul_result_chunk = temp_result;  in gemm_q8_2_2_2()
17824   const std::int32_t mul_result_chunk_stride_bytes = ((n * 4 + 7) / 8) * 8;  in gemm_q8_2_2_2()
17869                    const std::uint8_t* rhs, std::int32_t m, std::int32_t n,  in gemm_q8_2_2_3()
17870                    std::int32_t k, std::int32_t lhs_offset,  in gemm_q8_2_2_3()
17871                    std::int32_t rhs_offset, std::int32_t result_offset,  in gemm_q8_2_2_3()
17872                    std::int32_t multiplicative_offset, std::int32_t shift,  in gemm_q8_2_2_3()
17873                    std::uint8_t* result, std::int32_t result_stride) {  in gemm_q8_2_2_3()
17874   const std::int32_t row_chunks = m / 3;  in gemm_q8_2_2_3()
17875   const std::int32_t col_chunks = n / 3;  in gemm_q8_2_2_3()
17876   const std::int32_t padded_k = ((k + 7) / 8) * 8;  in gemm_q8_2_2_3()
17877   const std::int32_t chunk_size = k * 3;  in gemm_q8_2_2_3()
17878   const std::int32_t zipped_chunk_size = (padded_k + 16) * 3;  in gemm_q8_2_2_3()
17879   const std::int32_t zipped_rhs_size = (padded_k + 16) * n;  in gemm_q8_2_2_3()
17883   std::int32_t* zipped_lhs_3_offsets =  in gemm_q8_2_2_3()
17884       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 3);  in gemm_q8_2_2_3()
17885   std::int32_t* zipped_lhs_2_offsets =  in gemm_q8_2_2_3()
17886       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 2);  in gemm_q8_2_2_3()
17889   const std::int32_t result_chunk_stride = result_stride * 3;  in gemm_q8_2_2_3()
17891   const std::int32_t const_offset = lhs_offset * rhs_offset * k + result_offset;  in gemm_q8_2_2_3()
17892   const std::int32_t rounding_offset = (1 << (shift - 1));  in gemm_q8_2_2_3()
17893   std::int32_t* temp_result = reinterpret_cast<std::int32_t*>(  in gemm_q8_2_2_3()
17896   std::int32_t* mul_result_chunk = temp_result;  in gemm_q8_2_2_3()
17897   const std::int32_t mul_result_chunk_stride_bytes = ((n * 4 + 7) / 8) * 8;  in gemm_q8_2_2_3()
17942                    const std::uint8_t* rhs, std::int32_t m, std::int32_t n,  in gemm_q8_2_2_4()
17943                    std::int32_t k, std::int32_t lhs_offset,  in gemm_q8_2_2_4()
17944                    std::int32_t rhs_offset, std::int32_t result_offset,  in gemm_q8_2_2_4()
17945                    std::int32_t multiplicative_offset, std::int32_t shift,  in gemm_q8_2_2_4()
17946                    std::uint8_t* result, std::int32_t result_stride) {  in gemm_q8_2_2_4()
17947   const std::int32_t row_chunks = m / 3;  in gemm_q8_2_2_4()
17948   const std::int32_t col_chunks = n / 3;  in gemm_q8_2_2_4()
17949   const std::int32_t padded_k = ((k + 7) / 8) * 8;  in gemm_q8_2_2_4()
17950   const std::int32_t chunk_size = k * 3;  in gemm_q8_2_2_4()
17951   const std::int32_t zipped_chunk_size = (padded_k + 16) * 3;  in gemm_q8_2_2_4()
17952   const std::int32_t zipped_rhs_size = (padded_k + 16) * n;  in gemm_q8_2_2_4()
17956   std::int32_t* zipped_lhs_3_offsets =  in gemm_q8_2_2_4()
17957       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 3);  in gemm_q8_2_2_4()
17958   std::int32_t* zipped_lhs_2_offsets =  in gemm_q8_2_2_4()
17959       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 2);  in gemm_q8_2_2_4()
17962   const std::int32_t result_chunk_stride = result_stride * 3;  in gemm_q8_2_2_4()
17964   const std::int32_t const_offset = lhs_offset * rhs_offset * k + result_offset;  in gemm_q8_2_2_4()
17965   const std::int32_t rounding_offset = (1 << (shift - 1));  in gemm_q8_2_2_4()
17966   std::int32_t* temp_result = reinterpret_cast<std::int32_t*>(  in gemm_q8_2_2_4()
17969   std::int32_t* mul_result_chunk = temp_result;  in gemm_q8_2_2_4()
17970   const std::int32_t mul_result_chunk_stride_bytes = ((n * 4 + 7) / 8) * 8;  in gemm_q8_2_2_4()
18015                    const std::uint8_t* rhs, std::int32_t m, std::int32_t n,  in gemm_q8_2_2_5()
18016                    std::int32_t k, std::int32_t lhs_offset,  in gemm_q8_2_2_5()
18017                    std::int32_t rhs_offset, std::int32_t result_offset,  in gemm_q8_2_2_5()
18018                    std::int32_t multiplicative_offset, std::int32_t shift,  in gemm_q8_2_2_5()
18019                    std::uint8_t* result, std::int32_t result_stride) {  in gemm_q8_2_2_5()
18020   const std::int32_t row_chunks = m / 3;  in gemm_q8_2_2_5()
18021   const std::int32_t col_chunks = n / 3;  in gemm_q8_2_2_5()
18022   const std::int32_t padded_k = ((k + 7) / 8) * 8;  in gemm_q8_2_2_5()
18023   const std::int32_t chunk_size = k * 3;  in gemm_q8_2_2_5()
18024   const std::int32_t zipped_chunk_size = (padded_k + 16) * 3;  in gemm_q8_2_2_5()
18025   const std::int32_t zipped_rhs_size = (padded_k + 16) * n;  in gemm_q8_2_2_5()
18029   std::int32_t* zipped_lhs_3_offsets =  in gemm_q8_2_2_5()
18030       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 3);  in gemm_q8_2_2_5()
18031   std::int32_t* zipped_lhs_2_offsets =  in gemm_q8_2_2_5()
18032       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 2);  in gemm_q8_2_2_5()
18035   const std::int32_t result_chunk_stride = result_stride * 3;  in gemm_q8_2_2_5()
18037   const std::int32_t const_offset = lhs_offset * rhs_offset * k + result_offset;  in gemm_q8_2_2_5()
18038   const std::int32_t rounding_offset = (1 << (shift - 1));  in gemm_q8_2_2_5()
18039   std::int32_t* temp_result = reinterpret_cast<std::int32_t*>(  in gemm_q8_2_2_5()
18042   std::int32_t* mul_result_chunk = temp_result;  in gemm_q8_2_2_5()
18043   const std::int32_t mul_result_chunk_stride_bytes = ((n * 4 + 7) / 8) * 8;  in gemm_q8_2_2_5()
18088                    const std::uint8_t* rhs, std::int32_t m, std::int32_t n,  in gemm_q8_2_2_6()
18089                    std::int32_t k, std::int32_t lhs_offset,  in gemm_q8_2_2_6()
18090                    std::int32_t rhs_offset, std::int32_t result_offset,  in gemm_q8_2_2_6()
18091                    std::int32_t multiplicative_offset, std::int32_t shift,  in gemm_q8_2_2_6()
18092                    std::uint8_t* result, std::int32_t result_stride) {  in gemm_q8_2_2_6()
18093   const std::int32_t row_chunks = m / 3;  in gemm_q8_2_2_6()
18094   const std::int32_t col_chunks = n / 3;  in gemm_q8_2_2_6()
18095   const std::int32_t padded_k = ((k + 7) / 8) * 8;  in gemm_q8_2_2_6()
18096   const std::int32_t chunk_size = k * 3;  in gemm_q8_2_2_6()
18097   const std::int32_t zipped_chunk_size = (padded_k + 16) * 3;  in gemm_q8_2_2_6()
18098   const std::int32_t zipped_rhs_size = (padded_k + 16) * n;  in gemm_q8_2_2_6()
18102   std::int32_t* zipped_lhs_3_offsets =  in gemm_q8_2_2_6()
18103       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 3);  in gemm_q8_2_2_6()
18104   std::int32_t* zipped_lhs_2_offsets =  in gemm_q8_2_2_6()
18105       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 2);  in gemm_q8_2_2_6()
18108   const std::int32_t result_chunk_stride = result_stride * 3;  in gemm_q8_2_2_6()
18110   const std::int32_t const_offset = lhs_offset * rhs_offset * k + result_offset;  in gemm_q8_2_2_6()
18111   const std::int32_t rounding_offset = (1 << (shift - 1));  in gemm_q8_2_2_6()
18112   std::int32_t* temp_result = reinterpret_cast<std::int32_t*>(  in gemm_q8_2_2_6()
18115   std::int32_t* mul_result_chunk = temp_result;  in gemm_q8_2_2_6()
18116   const std::int32_t mul_result_chunk_stride_bytes = ((n * 4 + 7) / 8) * 8;  in gemm_q8_2_2_6()
18161                    const std::uint8_t* rhs, std::int32_t m, std::int32_t n,  in gemm_q8_2_2_7()
18162                    std::int32_t k, std::int32_t lhs_offset,  in gemm_q8_2_2_7()
18163                    std::int32_t rhs_offset, std::int32_t result_offset,  in gemm_q8_2_2_7()
18164                    std::int32_t multiplicative_offset, std::int32_t shift,  in gemm_q8_2_2_7()
18165                    std::uint8_t* result, std::int32_t result_stride) {  in gemm_q8_2_2_7()
18166   const std::int32_t row_chunks = m / 3;  in gemm_q8_2_2_7()
18167   const std::int32_t col_chunks = n / 3;  in gemm_q8_2_2_7()
18168   const std::int32_t padded_k = ((k + 7) / 8) * 8;  in gemm_q8_2_2_7()
18169   const std::int32_t chunk_size = k * 3;  in gemm_q8_2_2_7()
18170   const std::int32_t zipped_chunk_size = (padded_k + 16) * 3;  in gemm_q8_2_2_7()
18171   const std::int32_t zipped_rhs_size = (padded_k + 16) * n;  in gemm_q8_2_2_7()
18175   std::int32_t* zipped_lhs_3_offsets =  in gemm_q8_2_2_7()
18176       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 3);  in gemm_q8_2_2_7()
18177   std::int32_t* zipped_lhs_2_offsets =  in gemm_q8_2_2_7()
18178       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 2);  in gemm_q8_2_2_7()
18181   const std::int32_t result_chunk_stride = result_stride * 3;  in gemm_q8_2_2_7()
18183   const std::int32_t const_offset = lhs_offset * rhs_offset * k + result_offset;  in gemm_q8_2_2_7()
18184   const std::int32_t rounding_offset = (1 << (shift - 1));  in gemm_q8_2_2_7()
18185   std::int32_t* temp_result = reinterpret_cast<std::int32_t*>(  in gemm_q8_2_2_7()
18188   std::int32_t* mul_result_chunk = temp_result;  in gemm_q8_2_2_7()
18189   const std::int32_t mul_result_chunk_stride_bytes = ((n * 4 + 7) / 8) * 8;  in gemm_q8_2_2_7()
18234                             const std::uint8_t* rhs, std::int32_t m,  in gemm_i32_0_0_0_aligned()
18235                             std::int32_t n, std::int32_t k,  in gemm_i32_0_0_0_aligned()
18236                             std::int32_t lhs_offset, std::int32_t rhs_offset,  in gemm_i32_0_0_0_aligned()
18237                             std::int32_t* result, std::int32_t result_stride) {  in gemm_i32_0_0_0_aligned()
18238   const std::int32_t row_chunks = m / 3;  in gemm_i32_0_0_0_aligned()
18239   const std::int32_t col_chunks = n / 3;  in gemm_i32_0_0_0_aligned()
18240   const std::int32_t padded_k = ((k + 7) / 8) * 8;  in gemm_i32_0_0_0_aligned()
18241   const std::int32_t chunk_size = k * 3;  in gemm_i32_0_0_0_aligned()
18242   const std::int32_t zipped_chunk_size = (padded_k + 16) * 3;  in gemm_i32_0_0_0_aligned()
18243   const std::int32_t zipped_rhs_size = (padded_k + 16) * n;  in gemm_i32_0_0_0_aligned()
18247   std::int32_t* zipped_lhs_3_offsets =  in gemm_i32_0_0_0_aligned()
18248       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 3);  in gemm_i32_0_0_0_aligned()
18251   const std::int32_t result_chunk_stride = result_stride * 3;  in gemm_i32_0_0_0_aligned()
18253   const std::int32_t const_offset = lhs_offset * rhs_offset * k;  in gemm_i32_0_0_0_aligned()
18254   std::int32_t* result_chunk = result;  in gemm_i32_0_0_0_aligned()
18255   std::int32_t* mul_result_chunk = result;  in gemm_i32_0_0_0_aligned()
18256   const std::int32_t mul_result_chunk_stride_bytes = result_stride * 4;  in gemm_i32_0_0_0_aligned()
18281                             const std::uint8_t* rhs, std::int32_t m,  in gemm_i32_0_0_1_aligned()
18282                             std::int32_t n, std::int32_t k,  in gemm_i32_0_0_1_aligned()
18283                             std::int32_t lhs_offset, std::int32_t rhs_offset,  in gemm_i32_0_0_1_aligned()
18284                             std::int32_t* result, std::int32_t result_stride) {  in gemm_i32_0_0_1_aligned()
18285   const std::int32_t row_chunks = m / 3;  in gemm_i32_0_0_1_aligned()
18286   const std::int32_t col_chunks = n / 3;  in gemm_i32_0_0_1_aligned()
18287   const std::int32_t padded_k = ((k + 7) / 8) * 8;  in gemm_i32_0_0_1_aligned()
18288   const std::int32_t chunk_size = k * 3;  in gemm_i32_0_0_1_aligned()
18289   const std::int32_t zipped_chunk_size = (padded_k + 16) * 3;  in gemm_i32_0_0_1_aligned()
18290   const std::int32_t zipped_rhs_size = (padded_k + 16) * n;  in gemm_i32_0_0_1_aligned()
18294   std::int32_t* zipped_lhs_3_offsets =  in gemm_i32_0_0_1_aligned()
18295       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 3);  in gemm_i32_0_0_1_aligned()
18298   const std::int32_t result_chunk_stride = result_stride * 3;  in gemm_i32_0_0_1_aligned()
18300   const std::int32_t const_offset = lhs_offset * rhs_offset * k;  in gemm_i32_0_0_1_aligned()
18301   std::int32_t* result_chunk = result;  in gemm_i32_0_0_1_aligned()
18302   std::int32_t* mul_result_chunk = result;  in gemm_i32_0_0_1_aligned()
18303   const std::int32_t mul_result_chunk_stride_bytes = result_stride * 4;  in gemm_i32_0_0_1_aligned()
18328                             const std::uint8_t* rhs, std::int32_t m,  in gemm_i32_0_0_2_aligned()
18329                             std::int32_t n, std::int32_t k,  in gemm_i32_0_0_2_aligned()
18330                             std::int32_t lhs_offset, std::int32_t rhs_offset,  in gemm_i32_0_0_2_aligned()
18331                             std::int32_t* result, std::int32_t result_stride) {  in gemm_i32_0_0_2_aligned()
18332   const std::int32_t row_chunks = m / 3;  in gemm_i32_0_0_2_aligned()
18333   const std::int32_t col_chunks = n / 3;  in gemm_i32_0_0_2_aligned()
18334   const std::int32_t padded_k = ((k + 7) / 8) * 8;  in gemm_i32_0_0_2_aligned()
18335   const std::int32_t chunk_size = k * 3;  in gemm_i32_0_0_2_aligned()
18336   const std::int32_t zipped_chunk_size = (padded_k + 16) * 3;  in gemm_i32_0_0_2_aligned()
18337   const std::int32_t zipped_rhs_size = (padded_k + 16) * n;  in gemm_i32_0_0_2_aligned()
18341   std::int32_t* zipped_lhs_3_offsets =  in gemm_i32_0_0_2_aligned()
18342       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 3);  in gemm_i32_0_0_2_aligned()
18345   const std::int32_t result_chunk_stride = result_stride * 3;  in gemm_i32_0_0_2_aligned()
18347   const std::int32_t const_offset = lhs_offset * rhs_offset * k;  in gemm_i32_0_0_2_aligned()
18348   std::int32_t* result_chunk = result;  in gemm_i32_0_0_2_aligned()
18349   std::int32_t* mul_result_chunk = result;  in gemm_i32_0_0_2_aligned()
18350   const std::int32_t mul_result_chunk_stride_bytes = result_stride * 4;  in gemm_i32_0_0_2_aligned()
18375                             const std::uint8_t* rhs, std::int32_t m,  in gemm_i32_0_0_3_aligned()
18376                             std::int32_t n, std::int32_t k,  in gemm_i32_0_0_3_aligned()
18377                             std::int32_t lhs_offset, std::int32_t rhs_offset,  in gemm_i32_0_0_3_aligned()
18378                             std::int32_t* result, std::int32_t result_stride) {  in gemm_i32_0_0_3_aligned()
18379   const std::int32_t row_chunks = m / 3;  in gemm_i32_0_0_3_aligned()
18380   const std::int32_t col_chunks = n / 3;  in gemm_i32_0_0_3_aligned()
18381   const std::int32_t padded_k = ((k + 7) / 8) * 8;  in gemm_i32_0_0_3_aligned()
18382   const std::int32_t chunk_size = k * 3;  in gemm_i32_0_0_3_aligned()
18383   const std::int32_t zipped_chunk_size = (padded_k + 16) * 3;  in gemm_i32_0_0_3_aligned()
18384   const std::int32_t zipped_rhs_size = (padded_k + 16) * n;  in gemm_i32_0_0_3_aligned()
18388   std::int32_t* zipped_lhs_3_offsets =  in gemm_i32_0_0_3_aligned()
18389       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 3);  in gemm_i32_0_0_3_aligned()
18392   const std::int32_t result_chunk_stride = result_stride * 3;  in gemm_i32_0_0_3_aligned()
18394   const std::int32_t const_offset = lhs_offset * rhs_offset * k;  in gemm_i32_0_0_3_aligned()
18395   std::int32_t* result_chunk = result;  in gemm_i32_0_0_3_aligned()
18396   std::int32_t* mul_result_chunk = result;  in gemm_i32_0_0_3_aligned()
18397   const std::int32_t mul_result_chunk_stride_bytes = result_stride * 4;  in gemm_i32_0_0_3_aligned()
18422                             const std::uint8_t* rhs, std::int32_t m,  in gemm_i32_0_0_4_aligned()
18423                             std::int32_t n, std::int32_t k,  in gemm_i32_0_0_4_aligned()
18424                             std::int32_t lhs_offset, std::int32_t rhs_offset,  in gemm_i32_0_0_4_aligned()
18425                             std::int32_t* result, std::int32_t result_stride) {  in gemm_i32_0_0_4_aligned()
18426   const std::int32_t row_chunks = m / 3;  in gemm_i32_0_0_4_aligned()
18427   const std::int32_t col_chunks = n / 3;  in gemm_i32_0_0_4_aligned()
18428   const std::int32_t padded_k = ((k + 7) / 8) * 8;  in gemm_i32_0_0_4_aligned()
18429   const std::int32_t chunk_size = k * 3;  in gemm_i32_0_0_4_aligned()
18430   const std::int32_t zipped_chunk_size = (padded_k + 16) * 3;  in gemm_i32_0_0_4_aligned()
18431   const std::int32_t zipped_rhs_size = (padded_k + 16) * n;  in gemm_i32_0_0_4_aligned()
18435   std::int32_t* zipped_lhs_3_offsets =  in gemm_i32_0_0_4_aligned()
18436       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 3);  in gemm_i32_0_0_4_aligned()
18439   const std::int32_t result_chunk_stride = result_stride * 3;  in gemm_i32_0_0_4_aligned()
18441   const std::int32_t const_offset = lhs_offset * rhs_offset * k;  in gemm_i32_0_0_4_aligned()
18442   std::int32_t* result_chunk = result;  in gemm_i32_0_0_4_aligned()
18443   std::int32_t* mul_result_chunk = result;  in gemm_i32_0_0_4_aligned()
18444   const std::int32_t mul_result_chunk_stride_bytes = result_stride * 4;  in gemm_i32_0_0_4_aligned()
18469                             const std::uint8_t* rhs, std::int32_t m,  in gemm_i32_0_0_5_aligned()
18470                             std::int32_t n, std::int32_t k,  in gemm_i32_0_0_5_aligned()
18471                             std::int32_t lhs_offset, std::int32_t rhs_offset,  in gemm_i32_0_0_5_aligned()
18472                             std::int32_t* result, std::int32_t result_stride) {  in gemm_i32_0_0_5_aligned()
18473   const std::int32_t row_chunks = m / 3;  in gemm_i32_0_0_5_aligned()
18474   const std::int32_t col_chunks = n / 3;  in gemm_i32_0_0_5_aligned()
18475   const std::int32_t padded_k = ((k + 7) / 8) * 8;  in gemm_i32_0_0_5_aligned()
18476   const std::int32_t chunk_size = k * 3;  in gemm_i32_0_0_5_aligned()
18477   const std::int32_t zipped_chunk_size = (padded_k + 16) * 3;  in gemm_i32_0_0_5_aligned()
18478   const std::int32_t zipped_rhs_size = (padded_k + 16) * n;  in gemm_i32_0_0_5_aligned()
18482   std::int32_t* zipped_lhs_3_offsets =  in gemm_i32_0_0_5_aligned()
18483       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 3);  in gemm_i32_0_0_5_aligned()
18486   const std::int32_t result_chunk_stride = result_stride * 3;  in gemm_i32_0_0_5_aligned()
18488   const std::int32_t const_offset = lhs_offset * rhs_offset * k;  in gemm_i32_0_0_5_aligned()
18489   std::int32_t* result_chunk = result;  in gemm_i32_0_0_5_aligned()
18490   std::int32_t* mul_result_chunk = result;  in gemm_i32_0_0_5_aligned()
18491   const std::int32_t mul_result_chunk_stride_bytes = result_stride * 4;  in gemm_i32_0_0_5_aligned()
18516                             const std::uint8_t* rhs, std::int32_t m,  in gemm_i32_0_0_6_aligned()
18517                             std::int32_t n, std::int32_t k,  in gemm_i32_0_0_6_aligned()
18518                             std::int32_t lhs_offset, std::int32_t rhs_offset,  in gemm_i32_0_0_6_aligned()
18519                             std::int32_t* result, std::int32_t result_stride) {  in gemm_i32_0_0_6_aligned()
18520   const std::int32_t row_chunks = m / 3;  in gemm_i32_0_0_6_aligned()
18521   const std::int32_t col_chunks = n / 3;  in gemm_i32_0_0_6_aligned()
18522   const std::int32_t padded_k = ((k + 7) / 8) * 8;  in gemm_i32_0_0_6_aligned()
18523   const std::int32_t chunk_size = k * 3;  in gemm_i32_0_0_6_aligned()
18524   const std::int32_t zipped_chunk_size = (padded_k + 16) * 3;  in gemm_i32_0_0_6_aligned()
18525   const std::int32_t zipped_rhs_size = (padded_k + 16) * n;  in gemm_i32_0_0_6_aligned()
18529   std::int32_t* zipped_lhs_3_offsets =  in gemm_i32_0_0_6_aligned()
18530       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 3);  in gemm_i32_0_0_6_aligned()
18533   const std::int32_t result_chunk_stride = result_stride * 3;  in gemm_i32_0_0_6_aligned()
18535   const std::int32_t const_offset = lhs_offset * rhs_offset * k;  in gemm_i32_0_0_6_aligned()
18536   std::int32_t* result_chunk = result;  in gemm_i32_0_0_6_aligned()
18537   std::int32_t* mul_result_chunk = result;  in gemm_i32_0_0_6_aligned()
18538   const std::int32_t mul_result_chunk_stride_bytes = result_stride * 4;  in gemm_i32_0_0_6_aligned()
18563                             const std::uint8_t* rhs, std::int32_t m,  in gemm_i32_0_0_7_aligned()
18564                             std::int32_t n, std::int32_t k,  in gemm_i32_0_0_7_aligned()
18565                             std::int32_t lhs_offset, std::int32_t rhs_offset,  in gemm_i32_0_0_7_aligned()
18566                             std::int32_t* result, std::int32_t result_stride) {  in gemm_i32_0_0_7_aligned()
18567   const std::int32_t row_chunks = m / 3;  in gemm_i32_0_0_7_aligned()
18568   const std::int32_t col_chunks = n / 3;  in gemm_i32_0_0_7_aligned()
18569   const std::int32_t padded_k = ((k + 7) / 8) * 8;  in gemm_i32_0_0_7_aligned()
18570   const std::int32_t chunk_size = k * 3;  in gemm_i32_0_0_7_aligned()
18571   const std::int32_t zipped_chunk_size = (padded_k + 16) * 3;  in gemm_i32_0_0_7_aligned()
18572   const std::int32_t zipped_rhs_size = (padded_k + 16) * n;  in gemm_i32_0_0_7_aligned()
18576   std::int32_t* zipped_lhs_3_offsets =  in gemm_i32_0_0_7_aligned()
18577       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 3);  in gemm_i32_0_0_7_aligned()
18580   const std::int32_t result_chunk_stride = result_stride * 3;  in gemm_i32_0_0_7_aligned()
18582   const std::int32_t const_offset = lhs_offset * rhs_offset * k;  in gemm_i32_0_0_7_aligned()
18583   std::int32_t* result_chunk = result;  in gemm_i32_0_0_7_aligned()
18584   std::int32_t* mul_result_chunk = result;  in gemm_i32_0_0_7_aligned()
18585   const std::int32_t mul_result_chunk_stride_bytes = result_stride * 4;  in gemm_i32_0_0_7_aligned()
18610                             const std::uint8_t* rhs, std::int32_t m,  in gemm_i32_0_1_0_aligned()
18611                             std::int32_t n, std::int32_t k,  in gemm_i32_0_1_0_aligned()
18612                             std::int32_t lhs_offset, std::int32_t rhs_offset,  in gemm_i32_0_1_0_aligned()
18613                             std::int32_t* result, std::int32_t result_stride) {  in gemm_i32_0_1_0_aligned()
18614   const std::int32_t row_chunks = m / 3;  in gemm_i32_0_1_0_aligned()
18615   const std::int32_t col_chunks = n / 3;  in gemm_i32_0_1_0_aligned()
18616   const std::int32_t padded_k = ((k + 7) / 8) * 8;  in gemm_i32_0_1_0_aligned()
18617   const std::int32_t chunk_size = k * 3;  in gemm_i32_0_1_0_aligned()
18618   const std::int32_t zipped_chunk_size = (padded_k + 16) * 3;  in gemm_i32_0_1_0_aligned()
18619   const std::int32_t zipped_rhs_size = (padded_k + 16) * n;  in gemm_i32_0_1_0_aligned()
18623   std::int32_t* zipped_lhs_3_offsets =  in gemm_i32_0_1_0_aligned()
18624       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 3);  in gemm_i32_0_1_0_aligned()
18627   const std::int32_t result_chunk_stride = result_stride * 3;  in gemm_i32_0_1_0_aligned()
18629   const std::int32_t const_offset = lhs_offset * rhs_offset * k;  in gemm_i32_0_1_0_aligned()
18630   std::int32_t* result_chunk = result;  in gemm_i32_0_1_0_aligned()
18631   std::int32_t* mul_result_chunk = result;  in gemm_i32_0_1_0_aligned()
18632   const std::int32_t mul_result_chunk_stride_bytes = result_stride * 4;  in gemm_i32_0_1_0_aligned()
18661                             const std::uint8_t* rhs, std::int32_t m,  in gemm_i32_0_1_1_aligned()
18662                             std::int32_t n, std::int32_t k,  in gemm_i32_0_1_1_aligned()
18663                             std::int32_t lhs_offset, std::int32_t rhs_offset,  in gemm_i32_0_1_1_aligned()
18664                             std::int32_t* result, std::int32_t result_stride) {  in gemm_i32_0_1_1_aligned()
18665   const std::int32_t row_chunks = m / 3;  in gemm_i32_0_1_1_aligned()
18666   const std::int32_t col_chunks = n / 3;  in gemm_i32_0_1_1_aligned()
18667   const std::int32_t padded_k = ((k + 7) / 8) * 8;  in gemm_i32_0_1_1_aligned()
18668   const std::int32_t chunk_size = k * 3;  in gemm_i32_0_1_1_aligned()
18669   const std::int32_t zipped_chunk_size = (padded_k + 16) * 3;  in gemm_i32_0_1_1_aligned()
18670   const std::int32_t zipped_rhs_size = (padded_k + 16) * n;  in gemm_i32_0_1_1_aligned()
18674   std::int32_t* zipped_lhs_3_offsets =  in gemm_i32_0_1_1_aligned()
18675       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 3);  in gemm_i32_0_1_1_aligned()
18678   const std::int32_t result_chunk_stride = result_stride * 3;  in gemm_i32_0_1_1_aligned()
18680   const std::int32_t const_offset = lhs_offset * rhs_offset * k;  in gemm_i32_0_1_1_aligned()
18681   std::int32_t* result_chunk = result;  in gemm_i32_0_1_1_aligned()
18682   std::int32_t* mul_result_chunk = result;  in gemm_i32_0_1_1_aligned()
18683   const std::int32_t mul_result_chunk_stride_bytes = result_stride * 4;  in gemm_i32_0_1_1_aligned()
18712                             const std::uint8_t* rhs, std::int32_t m,  in gemm_i32_0_1_2_aligned()
18713                             std::int32_t n, std::int32_t k,  in gemm_i32_0_1_2_aligned()
18714                             std::int32_t lhs_offset, std::int32_t rhs_offset,  in gemm_i32_0_1_2_aligned()
18715                             std::int32_t* result, std::int32_t result_stride) {  in gemm_i32_0_1_2_aligned()
18716   const std::int32_t row_chunks = m / 3;  in gemm_i32_0_1_2_aligned()
18717   const std::int32_t col_chunks = n / 3;  in gemm_i32_0_1_2_aligned()
18718   const std::int32_t padded_k = ((k + 7) / 8) * 8;  in gemm_i32_0_1_2_aligned()
18719   const std::int32_t chunk_size = k * 3;  in gemm_i32_0_1_2_aligned()
18720   const std::int32_t zipped_chunk_size = (padded_k + 16) * 3;  in gemm_i32_0_1_2_aligned()
18721   const std::int32_t zipped_rhs_size = (padded_k + 16) * n;  in gemm_i32_0_1_2_aligned()
18725   std::int32_t* zipped_lhs_3_offsets =  in gemm_i32_0_1_2_aligned()
18726       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 3);  in gemm_i32_0_1_2_aligned()
18729   const std::int32_t result_chunk_stride = result_stride * 3;  in gemm_i32_0_1_2_aligned()
18731   const std::int32_t const_offset = lhs_offset * rhs_offset * k;  in gemm_i32_0_1_2_aligned()
18732   std::int32_t* result_chunk = result;  in gemm_i32_0_1_2_aligned()
18733   std::int32_t* mul_result_chunk = result;  in gemm_i32_0_1_2_aligned()
18734   const std::int32_t mul_result_chunk_stride_bytes = result_stride * 4;  in gemm_i32_0_1_2_aligned()
18763                             const std::uint8_t* rhs, std::int32_t m,  in gemm_i32_0_1_3_aligned()
18764                             std::int32_t n, std::int32_t k,  in gemm_i32_0_1_3_aligned()
18765                             std::int32_t lhs_offset, std::int32_t rhs_offset,  in gemm_i32_0_1_3_aligned()
18766                             std::int32_t* result, std::int32_t result_stride) {  in gemm_i32_0_1_3_aligned()
18767   const std::int32_t row_chunks = m / 3;  in gemm_i32_0_1_3_aligned()
18768   const std::int32_t col_chunks = n / 3;  in gemm_i32_0_1_3_aligned()
18769   const std::int32_t padded_k = ((k + 7) / 8) * 8;  in gemm_i32_0_1_3_aligned()
18770   const std::int32_t chunk_size = k * 3;  in gemm_i32_0_1_3_aligned()
18771   const std::int32_t zipped_chunk_size = (padded_k + 16) * 3;  in gemm_i32_0_1_3_aligned()
18772   const std::int32_t zipped_rhs_size = (padded_k + 16) * n;  in gemm_i32_0_1_3_aligned()
18776   std::int32_t* zipped_lhs_3_offsets =  in gemm_i32_0_1_3_aligned()
18777       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 3);  in gemm_i32_0_1_3_aligned()
18780   const std::int32_t result_chunk_stride = result_stride * 3;  in gemm_i32_0_1_3_aligned()
18782   const std::int32_t const_offset = lhs_offset * rhs_offset * k;  in gemm_i32_0_1_3_aligned()
18783   std::int32_t* result_chunk = result;  in gemm_i32_0_1_3_aligned()
18784   std::int32_t* mul_result_chunk = result;  in gemm_i32_0_1_3_aligned()
18785   const std::int32_t mul_result_chunk_stride_bytes = result_stride * 4;  in gemm_i32_0_1_3_aligned()
18814                             const std::uint8_t* rhs, std::int32_t m,  in gemm_i32_0_1_4_aligned()
18815                             std::int32_t n, std::int32_t k,  in gemm_i32_0_1_4_aligned()
18816                             std::int32_t lhs_offset, std::int32_t rhs_offset,  in gemm_i32_0_1_4_aligned()
18817                             std::int32_t* result, std::int32_t result_stride) {  in gemm_i32_0_1_4_aligned()
18818   const std::int32_t row_chunks = m / 3;  in gemm_i32_0_1_4_aligned()
18819   const std::int32_t col_chunks = n / 3;  in gemm_i32_0_1_4_aligned()
18820   const std::int32_t padded_k = ((k + 7) / 8) * 8;  in gemm_i32_0_1_4_aligned()
18821   const std::int32_t chunk_size = k * 3;  in gemm_i32_0_1_4_aligned()
18822   const std::int32_t zipped_chunk_size = (padded_k + 16) * 3;  in gemm_i32_0_1_4_aligned()
18823   const std::int32_t zipped_rhs_size = (padded_k + 16) * n;  in gemm_i32_0_1_4_aligned()
18827   std::int32_t* zipped_lhs_3_offsets =  in gemm_i32_0_1_4_aligned()
18828       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 3);  in gemm_i32_0_1_4_aligned()
18831   const std::int32_t result_chunk_stride = result_stride * 3;  in gemm_i32_0_1_4_aligned()
18833   const std::int32_t const_offset = lhs_offset * rhs_offset * k;  in gemm_i32_0_1_4_aligned()
18834   std::int32_t* result_chunk = result;  in gemm_i32_0_1_4_aligned()
18835   std::int32_t* mul_result_chunk = result;  in gemm_i32_0_1_4_aligned()
18836   const std::int32_t mul_result_chunk_stride_bytes = result_stride * 4;  in gemm_i32_0_1_4_aligned()
18865                             const std::uint8_t* rhs, std::int32_t m,  in gemm_i32_0_1_5_aligned()
18866                             std::int32_t n, std::int32_t k,  in gemm_i32_0_1_5_aligned()
18867                             std::int32_t lhs_offset, std::int32_t rhs_offset,  in gemm_i32_0_1_5_aligned()
18868                             std::int32_t* result, std::int32_t result_stride) {  in gemm_i32_0_1_5_aligned()
18869   const std::int32_t row_chunks = m / 3;  in gemm_i32_0_1_5_aligned()
18870   const std::int32_t col_chunks = n / 3;  in gemm_i32_0_1_5_aligned()
18871   const std::int32_t padded_k = ((k + 7) / 8) * 8;  in gemm_i32_0_1_5_aligned()
18872   const std::int32_t chunk_size = k * 3;  in gemm_i32_0_1_5_aligned()
18873   const std::int32_t zipped_chunk_size = (padded_k + 16) * 3;  in gemm_i32_0_1_5_aligned()
18874   const std::int32_t zipped_rhs_size = (padded_k + 16) * n;  in gemm_i32_0_1_5_aligned()
18878   std::int32_t* zipped_lhs_3_offsets =  in gemm_i32_0_1_5_aligned()
18879       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 3);  in gemm_i32_0_1_5_aligned()
18882   const std::int32_t result_chunk_stride = result_stride * 3;  in gemm_i32_0_1_5_aligned()
18884   const std::int32_t const_offset = lhs_offset * rhs_offset * k;  in gemm_i32_0_1_5_aligned()
18885   std::int32_t* result_chunk = result;  in gemm_i32_0_1_5_aligned()
18886   std::int32_t* mul_result_chunk = result;  in gemm_i32_0_1_5_aligned()
18887   const std::int32_t mul_result_chunk_stride_bytes = result_stride * 4;  in gemm_i32_0_1_5_aligned()
18916                             const std::uint8_t* rhs, std::int32_t m,  in gemm_i32_0_1_6_aligned()
18917                             std::int32_t n, std::int32_t k,  in gemm_i32_0_1_6_aligned()
18918                             std::int32_t lhs_offset, std::int32_t rhs_offset,  in gemm_i32_0_1_6_aligned()
18919                             std::int32_t* result, std::int32_t result_stride) {  in gemm_i32_0_1_6_aligned()
18920   const std::int32_t row_chunks = m / 3;  in gemm_i32_0_1_6_aligned()
18921   const std::int32_t col_chunks = n / 3;  in gemm_i32_0_1_6_aligned()
18922   const std::int32_t padded_k = ((k + 7) / 8) * 8;  in gemm_i32_0_1_6_aligned()
18923   const std::int32_t chunk_size = k * 3;  in gemm_i32_0_1_6_aligned()
18924   const std::int32_t zipped_chunk_size = (padded_k + 16) * 3;  in gemm_i32_0_1_6_aligned()
18925   const std::int32_t zipped_rhs_size = (padded_k + 16) * n;  in gemm_i32_0_1_6_aligned()
18929   std::int32_t* zipped_lhs_3_offsets =  in gemm_i32_0_1_6_aligned()
18930       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 3);  in gemm_i32_0_1_6_aligned()
18933   const std::int32_t result_chunk_stride = result_stride * 3;  in gemm_i32_0_1_6_aligned()
18935   const std::int32_t const_offset = lhs_offset * rhs_offset * k;  in gemm_i32_0_1_6_aligned()
18936   std::int32_t* result_chunk = result;  in gemm_i32_0_1_6_aligned()
18937   std::int32_t* mul_result_chunk = result;  in gemm_i32_0_1_6_aligned()
18938   const std::int32_t mul_result_chunk_stride_bytes = result_stride * 4;  in gemm_i32_0_1_6_aligned()
18967                             const std::uint8_t* rhs, std::int32_t m,  in gemm_i32_0_1_7_aligned()
18968                             std::int32_t n, std::int32_t k,  in gemm_i32_0_1_7_aligned()
18969                             std::int32_t lhs_offset, std::int32_t rhs_offset,  in gemm_i32_0_1_7_aligned()
18970                             std::int32_t* result, std::int32_t result_stride) {  in gemm_i32_0_1_7_aligned()
18971   const std::int32_t row_chunks = m / 3;  in gemm_i32_0_1_7_aligned()
18972   const std::int32_t col_chunks = n / 3;  in gemm_i32_0_1_7_aligned()
18973   const std::int32_t padded_k = ((k + 7) / 8) * 8;  in gemm_i32_0_1_7_aligned()
18974   const std::int32_t chunk_size = k * 3;  in gemm_i32_0_1_7_aligned()
18975   const std::int32_t zipped_chunk_size = (padded_k + 16) * 3;  in gemm_i32_0_1_7_aligned()
18976   const std::int32_t zipped_rhs_size = (padded_k + 16) * n;  in gemm_i32_0_1_7_aligned()
18980   std::int32_t* zipped_lhs_3_offsets =  in gemm_i32_0_1_7_aligned()
18981       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 3);  in gemm_i32_0_1_7_aligned()
18984   const std::int32_t result_chunk_stride = result_stride * 3;  in gemm_i32_0_1_7_aligned()
18986   const std::int32_t const_offset = lhs_offset * rhs_offset * k;  in gemm_i32_0_1_7_aligned()
18987   std::int32_t* result_chunk = result;  in gemm_i32_0_1_7_aligned()
18988   std::int32_t* mul_result_chunk = result;  in gemm_i32_0_1_7_aligned()
18989   const std::int32_t mul_result_chunk_stride_bytes = result_stride * 4;  in gemm_i32_0_1_7_aligned()
19018                             const std::uint8_t* rhs, std::int32_t m,  in gemm_i32_0_2_0_aligned()
19019                             std::int32_t n, std::int32_t k,  in gemm_i32_0_2_0_aligned()
19020                             std::int32_t lhs_offset, std::int32_t rhs_offset,  in gemm_i32_0_2_0_aligned()
19021                             std::int32_t* result, std::int32_t result_stride) {  in gemm_i32_0_2_0_aligned()
19022   const std::int32_t row_chunks = m / 3;  in gemm_i32_0_2_0_aligned()
19023   const std::int32_t col_chunks = n / 3;  in gemm_i32_0_2_0_aligned()
19024   const std::int32_t padded_k = ((k + 7) / 8) * 8;  in gemm_i32_0_2_0_aligned()
19025   const std::int32_t chunk_size = k * 3;  in gemm_i32_0_2_0_aligned()
19026   const std::int32_t zipped_chunk_size = (padded_k + 16) * 3;  in gemm_i32_0_2_0_aligned()
19027   const std::int32_t zipped_rhs_size = (padded_k + 16) * n;  in gemm_i32_0_2_0_aligned()
19031   std::int32_t* zipped_lhs_3_offsets =  in gemm_i32_0_2_0_aligned()
19032       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 3);  in gemm_i32_0_2_0_aligned()
19035   const std::int32_t result_chunk_stride = result_stride * 3;  in gemm_i32_0_2_0_aligned()
19037   const std::int32_t const_offset = lhs_offset * rhs_offset * k;  in gemm_i32_0_2_0_aligned()
19038   std::int32_t* result_chunk = result;  in gemm_i32_0_2_0_aligned()
19039   std::int32_t* mul_result_chunk = result;  in gemm_i32_0_2_0_aligned()
19040   const std::int32_t mul_result_chunk_stride_bytes = result_stride * 4;  in gemm_i32_0_2_0_aligned()
19069                             const std::uint8_t* rhs, std::int32_t m,  in gemm_i32_0_2_1_aligned()
19070                             std::int32_t n, std::int32_t k,  in gemm_i32_0_2_1_aligned()
19071                             std::int32_t lhs_offset, std::int32_t rhs_offset,  in gemm_i32_0_2_1_aligned()
19072                             std::int32_t* result, std::int32_t result_stride) {  in gemm_i32_0_2_1_aligned()
19073   const std::int32_t row_chunks = m / 3;  in gemm_i32_0_2_1_aligned()
19074   const std::int32_t col_chunks = n / 3;  in gemm_i32_0_2_1_aligned()
19075   const std::int32_t padded_k = ((k + 7) / 8) * 8;  in gemm_i32_0_2_1_aligned()
19076   const std::int32_t chunk_size = k * 3;  in gemm_i32_0_2_1_aligned()
19077   const std::int32_t zipped_chunk_size = (padded_k + 16) * 3;  in gemm_i32_0_2_1_aligned()
19078   const std::int32_t zipped_rhs_size = (padded_k + 16) * n;  in gemm_i32_0_2_1_aligned()
19082   std::int32_t* zipped_lhs_3_offsets =  in gemm_i32_0_2_1_aligned()
19083       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 3);  in gemm_i32_0_2_1_aligned()
19086   const std::int32_t result_chunk_stride = result_stride * 3;  in gemm_i32_0_2_1_aligned()
19088   const std::int32_t const_offset = lhs_offset * rhs_offset * k;  in gemm_i32_0_2_1_aligned()
19089   std::int32_t* result_chunk = result;  in gemm_i32_0_2_1_aligned()
19090   std::int32_t* mul_result_chunk = result;  in gemm_i32_0_2_1_aligned()
19091   const std::int32_t mul_result_chunk_stride_bytes = result_stride * 4;  in gemm_i32_0_2_1_aligned()
19120                             const std::uint8_t* rhs, std::int32_t m,  in gemm_i32_0_2_2_aligned()
19121                             std::int32_t n, std::int32_t k,  in gemm_i32_0_2_2_aligned()
19122                             std::int32_t lhs_offset, std::int32_t rhs_offset,  in gemm_i32_0_2_2_aligned()
19123                             std::int32_t* result, std::int32_t result_stride) {  in gemm_i32_0_2_2_aligned()
19124   const std::int32_t row_chunks = m / 3;  in gemm_i32_0_2_2_aligned()
19125   const std::int32_t col_chunks = n / 3;  in gemm_i32_0_2_2_aligned()
19126   const std::int32_t padded_k = ((k + 7) / 8) * 8;  in gemm_i32_0_2_2_aligned()
19127   const std::int32_t chunk_size = k * 3;  in gemm_i32_0_2_2_aligned()
19128   const std::int32_t zipped_chunk_size = (padded_k + 16) * 3;  in gemm_i32_0_2_2_aligned()
19129   const std::int32_t zipped_rhs_size = (padded_k + 16) * n;  in gemm_i32_0_2_2_aligned()
19133   std::int32_t* zipped_lhs_3_offsets =  in gemm_i32_0_2_2_aligned()
19134       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 3);  in gemm_i32_0_2_2_aligned()
19137   const std::int32_t result_chunk_stride = result_stride * 3;  in gemm_i32_0_2_2_aligned()
19139   const std::int32_t const_offset = lhs_offset * rhs_offset * k;  in gemm_i32_0_2_2_aligned()
19140   std::int32_t* result_chunk = result;  in gemm_i32_0_2_2_aligned()
19141   std::int32_t* mul_result_chunk = result;  in gemm_i32_0_2_2_aligned()
19142   const std::int32_t mul_result_chunk_stride_bytes = result_stride * 4;  in gemm_i32_0_2_2_aligned()
19171                             const std::uint8_t* rhs, std::int32_t m,  in gemm_i32_0_2_3_aligned()
19172                             std::int32_t n, std::int32_t k,  in gemm_i32_0_2_3_aligned()
19173                             std::int32_t lhs_offset, std::int32_t rhs_offset,  in gemm_i32_0_2_3_aligned()
19174                             std::int32_t* result, std::int32_t result_stride) {  in gemm_i32_0_2_3_aligned()
19175   const std::int32_t row_chunks = m / 3;  in gemm_i32_0_2_3_aligned()
19176   const std::int32_t col_chunks = n / 3;  in gemm_i32_0_2_3_aligned()
19177   const std::int32_t padded_k = ((k + 7) / 8) * 8;  in gemm_i32_0_2_3_aligned()
19178   const std::int32_t chunk_size = k * 3;  in gemm_i32_0_2_3_aligned()
19179   const std::int32_t zipped_chunk_size = (padded_k + 16) * 3;  in gemm_i32_0_2_3_aligned()
19180   const std::int32_t zipped_rhs_size = (padded_k + 16) * n;  in gemm_i32_0_2_3_aligned()
19184   std::int32_t* zipped_lhs_3_offsets =  in gemm_i32_0_2_3_aligned()
19185       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 3);  in gemm_i32_0_2_3_aligned()
19188   const std::int32_t result_chunk_stride = result_stride * 3;  in gemm_i32_0_2_3_aligned()
19190   const std::int32_t const_offset = lhs_offset * rhs_offset * k;  in gemm_i32_0_2_3_aligned()
19191   std::int32_t* result_chunk = result;  in gemm_i32_0_2_3_aligned()
19192   std::int32_t* mul_result_chunk = result;  in gemm_i32_0_2_3_aligned()
19193   const std::int32_t mul_result_chunk_stride_bytes = result_stride * 4;  in gemm_i32_0_2_3_aligned()
19222                             const std::uint8_t* rhs, std::int32_t m,  in gemm_i32_0_2_4_aligned()
19223                             std::int32_t n, std::int32_t k,  in gemm_i32_0_2_4_aligned()
19224                             std::int32_t lhs_offset, std::int32_t rhs_offset,  in gemm_i32_0_2_4_aligned()
19225                             std::int32_t* result, std::int32_t result_stride) {  in gemm_i32_0_2_4_aligned()
19226   const std::int32_t row_chunks = m / 3;  in gemm_i32_0_2_4_aligned()
19227   const std::int32_t col_chunks = n / 3;  in gemm_i32_0_2_4_aligned()
19228   const std::int32_t padded_k = ((k + 7) / 8) * 8;  in gemm_i32_0_2_4_aligned()
19229   const std::int32_t chunk_size = k * 3;  in gemm_i32_0_2_4_aligned()
19230   const std::int32_t zipped_chunk_size = (padded_k + 16) * 3;  in gemm_i32_0_2_4_aligned()
19231   const std::int32_t zipped_rhs_size = (padded_k + 16) * n;  in gemm_i32_0_2_4_aligned()
19235   std::int32_t* zipped_lhs_3_offsets =  in gemm_i32_0_2_4_aligned()
19236       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 3);  in gemm_i32_0_2_4_aligned()
19239   const std::int32_t result_chunk_stride = result_stride * 3;  in gemm_i32_0_2_4_aligned()
19241   const std::int32_t const_offset = lhs_offset * rhs_offset * k;  in gemm_i32_0_2_4_aligned()
19242   std::int32_t* result_chunk = result;  in gemm_i32_0_2_4_aligned()
19243   std::int32_t* mul_result_chunk = result;  in gemm_i32_0_2_4_aligned()
19244   const std::int32_t mul_result_chunk_stride_bytes = result_stride * 4;  in gemm_i32_0_2_4_aligned()
19273                             const std::uint8_t* rhs, std::int32_t m,  in gemm_i32_0_2_5_aligned()
19274                             std::int32_t n, std::int32_t k,  in gemm_i32_0_2_5_aligned()
19275                             std::int32_t lhs_offset, std::int32_t rhs_offset,  in gemm_i32_0_2_5_aligned()
19276                             std::int32_t* result, std::int32_t result_stride) {  in gemm_i32_0_2_5_aligned()
19277   const std::int32_t row_chunks = m / 3;  in gemm_i32_0_2_5_aligned()
19278   const std::int32_t col_chunks = n / 3;  in gemm_i32_0_2_5_aligned()
19279   const std::int32_t padded_k = ((k + 7) / 8) * 8;  in gemm_i32_0_2_5_aligned()
19280   const std::int32_t chunk_size = k * 3;  in gemm_i32_0_2_5_aligned()
19281   const std::int32_t zipped_chunk_size = (padded_k + 16) * 3;  in gemm_i32_0_2_5_aligned()
19282   const std::int32_t zipped_rhs_size = (padded_k + 16) * n;  in gemm_i32_0_2_5_aligned()
19286   std::int32_t* zipped_lhs_3_offsets =  in gemm_i32_0_2_5_aligned()
19287       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 3);  in gemm_i32_0_2_5_aligned()
19290   const std::int32_t result_chunk_stride = result_stride * 3;  in gemm_i32_0_2_5_aligned()
19292   const std::int32_t const_offset = lhs_offset * rhs_offset * k;  in gemm_i32_0_2_5_aligned()
19293   std::int32_t* result_chunk = result;  in gemm_i32_0_2_5_aligned()
19294   std::int32_t* mul_result_chunk = result;  in gemm_i32_0_2_5_aligned()
19295   const std::int32_t mul_result_chunk_stride_bytes = result_stride * 4;  in gemm_i32_0_2_5_aligned()
19324                             const std::uint8_t* rhs, std::int32_t m,  in gemm_i32_0_2_6_aligned()
19325                             std::int32_t n, std::int32_t k,  in gemm_i32_0_2_6_aligned()
19326                             std::int32_t lhs_offset, std::int32_t rhs_offset,  in gemm_i32_0_2_6_aligned()
19327                             std::int32_t* result, std::int32_t result_stride) {  in gemm_i32_0_2_6_aligned()
19328   const std::int32_t row_chunks = m / 3;  in gemm_i32_0_2_6_aligned()
19329   const std::int32_t col_chunks = n / 3;  in gemm_i32_0_2_6_aligned()
19330   const std::int32_t padded_k = ((k + 7) / 8) * 8;  in gemm_i32_0_2_6_aligned()
19331   const std::int32_t chunk_size = k * 3;  in gemm_i32_0_2_6_aligned()
19332   const std::int32_t zipped_chunk_size = (padded_k + 16) * 3;  in gemm_i32_0_2_6_aligned()
19333   const std::int32_t zipped_rhs_size = (padded_k + 16) * n;  in gemm_i32_0_2_6_aligned()
19337   std::int32_t* zipped_lhs_3_offsets =  in gemm_i32_0_2_6_aligned()
19338       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 3);  in gemm_i32_0_2_6_aligned()
19341   const std::int32_t result_chunk_stride = result_stride * 3;  in gemm_i32_0_2_6_aligned()
19343   const std::int32_t const_offset = lhs_offset * rhs_offset * k;  in gemm_i32_0_2_6_aligned()
19344   std::int32_t* result_chunk = result;  in gemm_i32_0_2_6_aligned()
19345   std::int32_t* mul_result_chunk = result;  in gemm_i32_0_2_6_aligned()
19346   const std::int32_t mul_result_chunk_stride_bytes = result_stride * 4;  in gemm_i32_0_2_6_aligned()
19375                             const std::uint8_t* rhs, std::int32_t m,  in gemm_i32_0_2_7_aligned()
19376                             std::int32_t n, std::int32_t k,  in gemm_i32_0_2_7_aligned()
19377                             std::int32_t lhs_offset, std::int32_t rhs_offset,  in gemm_i32_0_2_7_aligned()
19378                             std::int32_t* result, std::int32_t result_stride) {  in gemm_i32_0_2_7_aligned()
19379   const std::int32_t row_chunks = m / 3;  in gemm_i32_0_2_7_aligned()
19380   const std::int32_t col_chunks = n / 3;  in gemm_i32_0_2_7_aligned()
19381   const std::int32_t padded_k = ((k + 7) / 8) * 8;  in gemm_i32_0_2_7_aligned()
19382   const std::int32_t chunk_size = k * 3;  in gemm_i32_0_2_7_aligned()
19383   const std::int32_t zipped_chunk_size = (padded_k + 16) * 3;  in gemm_i32_0_2_7_aligned()
19384   const std::int32_t zipped_rhs_size = (padded_k + 16) * n;  in gemm_i32_0_2_7_aligned()
19388   std::int32_t* zipped_lhs_3_offsets =  in gemm_i32_0_2_7_aligned()
19389       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 3);  in gemm_i32_0_2_7_aligned()
19392   const std::int32_t result_chunk_stride = result_stride * 3;  in gemm_i32_0_2_7_aligned()
19394   const std::int32_t const_offset = lhs_offset * rhs_offset * k;  in gemm_i32_0_2_7_aligned()
19395   std::int32_t* result_chunk = result;  in gemm_i32_0_2_7_aligned()
19396   std::int32_t* mul_result_chunk = result;  in gemm_i32_0_2_7_aligned()
19397   const std::int32_t mul_result_chunk_stride_bytes = result_stride * 4;  in gemm_i32_0_2_7_aligned()
19426                             const std::uint8_t* rhs, std::int32_t m,  in gemm_i32_1_0_0_aligned()
19427                             std::int32_t n, std::int32_t k,  in gemm_i32_1_0_0_aligned()
19428                             std::int32_t lhs_offset, std::int32_t rhs_offset,  in gemm_i32_1_0_0_aligned()
19429                             std::int32_t* result, std::int32_t result_stride) {  in gemm_i32_1_0_0_aligned()
19430   const std::int32_t row_chunks = m / 3;  in gemm_i32_1_0_0_aligned()
19431   const std::int32_t col_chunks = n / 3;  in gemm_i32_1_0_0_aligned()
19432   const std::int32_t padded_k = ((k + 7) / 8) * 8;  in gemm_i32_1_0_0_aligned()
19433   const std::int32_t chunk_size = k * 3;  in gemm_i32_1_0_0_aligned()
19434   const std::int32_t zipped_chunk_size = (padded_k + 16) * 3;  in gemm_i32_1_0_0_aligned()
19435   const std::int32_t zipped_rhs_size = (padded_k + 16) * n;  in gemm_i32_1_0_0_aligned()
19439   std::int32_t* zipped_lhs_3_offsets =  in gemm_i32_1_0_0_aligned()
19440       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 3);  in gemm_i32_1_0_0_aligned()
19441   std::int32_t* zipped_lhs_1_offsets =  in gemm_i32_1_0_0_aligned()
19442       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 1);  in gemm_i32_1_0_0_aligned()
19445   const std::int32_t result_chunk_stride = result_stride * 3;  in gemm_i32_1_0_0_aligned()
19447   const std::int32_t const_offset = lhs_offset * rhs_offset * k;  in gemm_i32_1_0_0_aligned()
19448   std::int32_t* result_chunk = result;  in gemm_i32_1_0_0_aligned()
19449   std::int32_t* mul_result_chunk = result;  in gemm_i32_1_0_0_aligned()
19450   const std::int32_t mul_result_chunk_stride_bytes = result_stride * 4;  in gemm_i32_1_0_0_aligned()
19486                             const std::uint8_t* rhs, std::int32_t m,  in gemm_i32_1_0_1_aligned()
19487                             std::int32_t n, std::int32_t k,  in gemm_i32_1_0_1_aligned()
19488                             std::int32_t lhs_offset, std::int32_t rhs_offset,  in gemm_i32_1_0_1_aligned()
19489                             std::int32_t* result, std::int32_t result_stride) {  in gemm_i32_1_0_1_aligned()
19490   const std::int32_t row_chunks = m / 3;  in gemm_i32_1_0_1_aligned()
19491   const std::int32_t col_chunks = n / 3;  in gemm_i32_1_0_1_aligned()
19492   const std::int32_t padded_k = ((k + 7) / 8) * 8;  in gemm_i32_1_0_1_aligned()
19493   const std::int32_t chunk_size = k * 3;  in gemm_i32_1_0_1_aligned()
19494   const std::int32_t zipped_chunk_size = (padded_k + 16) * 3;  in gemm_i32_1_0_1_aligned()
19495   const std::int32_t zipped_rhs_size = (padded_k + 16) * n;  in gemm_i32_1_0_1_aligned()
19499   std::int32_t* zipped_lhs_3_offsets =  in gemm_i32_1_0_1_aligned()
19500       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 3);  in gemm_i32_1_0_1_aligned()
19501   std::int32_t* zipped_lhs_1_offsets =  in gemm_i32_1_0_1_aligned()
19502       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 1);  in gemm_i32_1_0_1_aligned()
19505   const std::int32_t result_chunk_stride = result_stride * 3;  in gemm_i32_1_0_1_aligned()
19507   const std::int32_t const_offset = lhs_offset * rhs_offset * k;  in gemm_i32_1_0_1_aligned()
19508   std::int32_t* result_chunk = result;  in gemm_i32_1_0_1_aligned()
19509   std::int32_t* mul_result_chunk = result;  in gemm_i32_1_0_1_aligned()
19510   const std::int32_t mul_result_chunk_stride_bytes = result_stride * 4;  in gemm_i32_1_0_1_aligned()
19546                             const std::uint8_t* rhs, std::int32_t m,  in gemm_i32_1_0_2_aligned()
19547                             std::int32_t n, std::int32_t k,  in gemm_i32_1_0_2_aligned()
19548                             std::int32_t lhs_offset, std::int32_t rhs_offset,  in gemm_i32_1_0_2_aligned()
19549                             std::int32_t* result, std::int32_t result_stride) {  in gemm_i32_1_0_2_aligned()
19550   const std::int32_t row_chunks = m / 3;  in gemm_i32_1_0_2_aligned()
19551   const std::int32_t col_chunks = n / 3;  in gemm_i32_1_0_2_aligned()
19552   const std::int32_t padded_k = ((k + 7) / 8) * 8;  in gemm_i32_1_0_2_aligned()
19553   const std::int32_t chunk_size = k * 3;  in gemm_i32_1_0_2_aligned()
19554   const std::int32_t zipped_chunk_size = (padded_k + 16) * 3;  in gemm_i32_1_0_2_aligned()
19555   const std::int32_t zipped_rhs_size = (padded_k + 16) * n;  in gemm_i32_1_0_2_aligned()
19559   std::int32_t* zipped_lhs_3_offsets =  in gemm_i32_1_0_2_aligned()
19560       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 3);  in gemm_i32_1_0_2_aligned()
19561   std::int32_t* zipped_lhs_1_offsets =  in gemm_i32_1_0_2_aligned()
19562       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 1);  in gemm_i32_1_0_2_aligned()
19565   const std::int32_t result_chunk_stride = result_stride * 3;  in gemm_i32_1_0_2_aligned()
19567   const std::int32_t const_offset = lhs_offset * rhs_offset * k;  in gemm_i32_1_0_2_aligned()
19568   std::int32_t* result_chunk = result;  in gemm_i32_1_0_2_aligned()
19569   std::int32_t* mul_result_chunk = result;  in gemm_i32_1_0_2_aligned()
19570   const std::int32_t mul_result_chunk_stride_bytes = result_stride * 4;  in gemm_i32_1_0_2_aligned()
19606                             const std::uint8_t* rhs, std::int32_t m,  in gemm_i32_1_0_3_aligned()
19607                             std::int32_t n, std::int32_t k,  in gemm_i32_1_0_3_aligned()
19608                             std::int32_t lhs_offset, std::int32_t rhs_offset,  in gemm_i32_1_0_3_aligned()
19609                             std::int32_t* result, std::int32_t result_stride) {  in gemm_i32_1_0_3_aligned()
19610   const std::int32_t row_chunks = m / 3;  in gemm_i32_1_0_3_aligned()
19611   const std::int32_t col_chunks = n / 3;  in gemm_i32_1_0_3_aligned()
19612   const std::int32_t padded_k = ((k + 7) / 8) * 8;  in gemm_i32_1_0_3_aligned()
19613   const std::int32_t chunk_size = k * 3;  in gemm_i32_1_0_3_aligned()
19614   const std::int32_t zipped_chunk_size = (padded_k + 16) * 3;  in gemm_i32_1_0_3_aligned()
19615   const std::int32_t zipped_rhs_size = (padded_k + 16) * n;  in gemm_i32_1_0_3_aligned()
19619   std::int32_t* zipped_lhs_3_offsets =  in gemm_i32_1_0_3_aligned()
19620       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 3);  in gemm_i32_1_0_3_aligned()
19621   std::int32_t* zipped_lhs_1_offsets =  in gemm_i32_1_0_3_aligned()
19622       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 1);  in gemm_i32_1_0_3_aligned()
19625   const std::int32_t result_chunk_stride = result_stride * 3;  in gemm_i32_1_0_3_aligned()
19627   const std::int32_t const_offset = lhs_offset * rhs_offset * k;  in gemm_i32_1_0_3_aligned()
19628   std::int32_t* result_chunk = result;  in gemm_i32_1_0_3_aligned()
19629   std::int32_t* mul_result_chunk = result;  in gemm_i32_1_0_3_aligned()
19630   const std::int32_t mul_result_chunk_stride_bytes = result_stride * 4;  in gemm_i32_1_0_3_aligned()
19666                             const std::uint8_t* rhs, std::int32_t m,  in gemm_i32_1_0_4_aligned()
19667                             std::int32_t n, std::int32_t k,  in gemm_i32_1_0_4_aligned()
19668                             std::int32_t lhs_offset, std::int32_t rhs_offset,  in gemm_i32_1_0_4_aligned()
19669                             std::int32_t* result, std::int32_t result_stride) {  in gemm_i32_1_0_4_aligned()
19670   const std::int32_t row_chunks = m / 3;  in gemm_i32_1_0_4_aligned()
19671   const std::int32_t col_chunks = n / 3;  in gemm_i32_1_0_4_aligned()
19672   const std::int32_t padded_k = ((k + 7) / 8) * 8;  in gemm_i32_1_0_4_aligned()
19673   const std::int32_t chunk_size = k * 3;  in gemm_i32_1_0_4_aligned()
19674   const std::int32_t zipped_chunk_size = (padded_k + 16) * 3;  in gemm_i32_1_0_4_aligned()
19675   const std::int32_t zipped_rhs_size = (padded_k + 16) * n;  in gemm_i32_1_0_4_aligned()
19679   std::int32_t* zipped_lhs_3_offsets =  in gemm_i32_1_0_4_aligned()
19680       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 3);  in gemm_i32_1_0_4_aligned()
19681   std::int32_t* zipped_lhs_1_offsets =  in gemm_i32_1_0_4_aligned()
19682       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 1);  in gemm_i32_1_0_4_aligned()
19685   const std::int32_t result_chunk_stride = result_stride * 3;  in gemm_i32_1_0_4_aligned()
19687   const std::int32_t const_offset = lhs_offset * rhs_offset * k;  in gemm_i32_1_0_4_aligned()
19688   std::int32_t* result_chunk = result;  in gemm_i32_1_0_4_aligned()
19689   std::int32_t* mul_result_chunk = result;  in gemm_i32_1_0_4_aligned()
19690   const std::int32_t mul_result_chunk_stride_bytes = result_stride * 4;  in gemm_i32_1_0_4_aligned()
19726                             const std::uint8_t* rhs, std::int32_t m,  in gemm_i32_1_0_5_aligned()
19727                             std::int32_t n, std::int32_t k,  in gemm_i32_1_0_5_aligned()
19728                             std::int32_t lhs_offset, std::int32_t rhs_offset,  in gemm_i32_1_0_5_aligned()
19729                             std::int32_t* result, std::int32_t result_stride) {  in gemm_i32_1_0_5_aligned()
19730   const std::int32_t row_chunks = m / 3;  in gemm_i32_1_0_5_aligned()
19731   const std::int32_t col_chunks = n / 3;  in gemm_i32_1_0_5_aligned()
19732   const std::int32_t padded_k = ((k + 7) / 8) * 8;  in gemm_i32_1_0_5_aligned()
19733   const std::int32_t chunk_size = k * 3;  in gemm_i32_1_0_5_aligned()
19734   const std::int32_t zipped_chunk_size = (padded_k + 16) * 3;  in gemm_i32_1_0_5_aligned()
19735   const std::int32_t zipped_rhs_size = (padded_k + 16) * n;  in gemm_i32_1_0_5_aligned()
19739   std::int32_t* zipped_lhs_3_offsets =  in gemm_i32_1_0_5_aligned()
19740       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 3);  in gemm_i32_1_0_5_aligned()
19741   std::int32_t* zipped_lhs_1_offsets =  in gemm_i32_1_0_5_aligned()
19742       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 1);  in gemm_i32_1_0_5_aligned()
19745   const std::int32_t result_chunk_stride = result_stride * 3;  in gemm_i32_1_0_5_aligned()
19747   const std::int32_t const_offset = lhs_offset * rhs_offset * k;  in gemm_i32_1_0_5_aligned()
19748   std::int32_t* result_chunk = result;  in gemm_i32_1_0_5_aligned()
19749   std::int32_t* mul_result_chunk = result;  in gemm_i32_1_0_5_aligned()
19750   const std::int32_t mul_result_chunk_stride_bytes = result_stride * 4;  in gemm_i32_1_0_5_aligned()
19786                             const std::uint8_t* rhs, std::int32_t m,  in gemm_i32_1_0_6_aligned()
19787                             std::int32_t n, std::int32_t k,  in gemm_i32_1_0_6_aligned()
19788                             std::int32_t lhs_offset, std::int32_t rhs_offset,  in gemm_i32_1_0_6_aligned()
19789                             std::int32_t* result, std::int32_t result_stride) {  in gemm_i32_1_0_6_aligned()
19790   const std::int32_t row_chunks = m / 3;  in gemm_i32_1_0_6_aligned()
19791   const std::int32_t col_chunks = n / 3;  in gemm_i32_1_0_6_aligned()
19792   const std::int32_t padded_k = ((k + 7) / 8) * 8;  in gemm_i32_1_0_6_aligned()
19793   const std::int32_t chunk_size = k * 3;  in gemm_i32_1_0_6_aligned()
19794   const std::int32_t zipped_chunk_size = (padded_k + 16) * 3;  in gemm_i32_1_0_6_aligned()
19795   const std::int32_t zipped_rhs_size = (padded_k + 16) * n;  in gemm_i32_1_0_6_aligned()
19799   std::int32_t* zipped_lhs_3_offsets =  in gemm_i32_1_0_6_aligned()
19800       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 3);  in gemm_i32_1_0_6_aligned()
19801   std::int32_t* zipped_lhs_1_offsets =  in gemm_i32_1_0_6_aligned()
19802       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 1);  in gemm_i32_1_0_6_aligned()
19805   const std::int32_t result_chunk_stride = result_stride * 3;  in gemm_i32_1_0_6_aligned()
19807   const std::int32_t const_offset = lhs_offset * rhs_offset * k;  in gemm_i32_1_0_6_aligned()
19808   std::int32_t* result_chunk = result;  in gemm_i32_1_0_6_aligned()
19809   std::int32_t* mul_result_chunk = result;  in gemm_i32_1_0_6_aligned()
19810   const std::int32_t mul_result_chunk_stride_bytes = result_stride * 4;  in gemm_i32_1_0_6_aligned()
19846                             const std::uint8_t* rhs, std::int32_t m,  in gemm_i32_1_0_7_aligned()
19847                             std::int32_t n, std::int32_t k,  in gemm_i32_1_0_7_aligned()
19848                             std::int32_t lhs_offset, std::int32_t rhs_offset,  in gemm_i32_1_0_7_aligned()
19849                             std::int32_t* result, std::int32_t result_stride) {  in gemm_i32_1_0_7_aligned()
19850   const std::int32_t row_chunks = m / 3;  in gemm_i32_1_0_7_aligned()
19851   const std::int32_t col_chunks = n / 3;  in gemm_i32_1_0_7_aligned()
19852   const std::int32_t padded_k = ((k + 7) / 8) * 8;  in gemm_i32_1_0_7_aligned()
19853   const std::int32_t chunk_size = k * 3;  in gemm_i32_1_0_7_aligned()
19854   const std::int32_t zipped_chunk_size = (padded_k + 16) * 3;  in gemm_i32_1_0_7_aligned()
19855   const std::int32_t zipped_rhs_size = (padded_k + 16) * n;  in gemm_i32_1_0_7_aligned()
19859   std::int32_t* zipped_lhs_3_offsets =  in gemm_i32_1_0_7_aligned()
19860       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 3);  in gemm_i32_1_0_7_aligned()
19861   std::int32_t* zipped_lhs_1_offsets =  in gemm_i32_1_0_7_aligned()
19862       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 1);  in gemm_i32_1_0_7_aligned()
19865   const std::int32_t result_chunk_stride = result_stride * 3;  in gemm_i32_1_0_7_aligned()
19867   const std::int32_t const_offset = lhs_offset * rhs_offset * k;  in gemm_i32_1_0_7_aligned()
19868   std::int32_t* result_chunk = result;  in gemm_i32_1_0_7_aligned()
19869   std::int32_t* mul_result_chunk = result;  in gemm_i32_1_0_7_aligned()
19870   const std::int32_t mul_result_chunk_stride_bytes = result_stride * 4;  in gemm_i32_1_0_7_aligned()
19906                             const std::uint8_t* rhs, std::int32_t m,  in gemm_i32_1_1_0_aligned()
19907                             std::int32_t n, std::int32_t k,  in gemm_i32_1_1_0_aligned()
19908                             std::int32_t lhs_offset, std::int32_t rhs_offset,  in gemm_i32_1_1_0_aligned()
19909                             std::int32_t* result, std::int32_t result_stride) {  in gemm_i32_1_1_0_aligned()
19910   const std::int32_t row_chunks = m / 3;  in gemm_i32_1_1_0_aligned()
19911   const std::int32_t col_chunks = n / 3;  in gemm_i32_1_1_0_aligned()
19912   const std::int32_t padded_k = ((k + 7) / 8) * 8;  in gemm_i32_1_1_0_aligned()
19913   const std::int32_t chunk_size = k * 3;  in gemm_i32_1_1_0_aligned()
19914   const std::int32_t zipped_chunk_size = (padded_k + 16) * 3;  in gemm_i32_1_1_0_aligned()
19915   const std::int32_t zipped_rhs_size = (padded_k + 16) * n;  in gemm_i32_1_1_0_aligned()
19919   std::int32_t* zipped_lhs_3_offsets =  in gemm_i32_1_1_0_aligned()
19920       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 3);  in gemm_i32_1_1_0_aligned()
19921   std::int32_t* zipped_lhs_1_offsets =  in gemm_i32_1_1_0_aligned()
19922       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 1);  in gemm_i32_1_1_0_aligned()
19925   const std::int32_t result_chunk_stride = result_stride * 3;  in gemm_i32_1_1_0_aligned()
19927   const std::int32_t const_offset = lhs_offset * rhs_offset * k;  in gemm_i32_1_1_0_aligned()
19928   std::int32_t* result_chunk = result;  in gemm_i32_1_1_0_aligned()
19929   std::int32_t* mul_result_chunk = result;  in gemm_i32_1_1_0_aligned()
19930   const std::int32_t mul_result_chunk_stride_bytes = result_stride * 4;  in gemm_i32_1_1_0_aligned()
19973                             const std::uint8_t* rhs, std::int32_t m,  in gemm_i32_1_1_1_aligned()
19974                             std::int32_t n, std::int32_t k,  in gemm_i32_1_1_1_aligned()
19975                             std::int32_t lhs_offset, std::int32_t rhs_offset,  in gemm_i32_1_1_1_aligned()
19976                             std::int32_t* result, std::int32_t result_stride) {  in gemm_i32_1_1_1_aligned()
19977   const std::int32_t row_chunks = m / 3;  in gemm_i32_1_1_1_aligned()
19978   const std::int32_t col_chunks = n / 3;  in gemm_i32_1_1_1_aligned()
19979   const std::int32_t padded_k = ((k + 7) / 8) * 8;  in gemm_i32_1_1_1_aligned()
19980   const std::int32_t chunk_size = k * 3;  in gemm_i32_1_1_1_aligned()
19981   const std::int32_t zipped_chunk_size = (padded_k + 16) * 3;  in gemm_i32_1_1_1_aligned()
19982   const std::int32_t zipped_rhs_size = (padded_k + 16) * n;  in gemm_i32_1_1_1_aligned()
19986   std::int32_t* zipped_lhs_3_offsets =  in gemm_i32_1_1_1_aligned()
19987       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 3);  in gemm_i32_1_1_1_aligned()
19988   std::int32_t* zipped_lhs_1_offsets =  in gemm_i32_1_1_1_aligned()
19989       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 1);  in gemm_i32_1_1_1_aligned()
19992   const std::int32_t result_chunk_stride = result_stride * 3;  in gemm_i32_1_1_1_aligned()
19994   const std::int32_t const_offset = lhs_offset * rhs_offset * k;  in gemm_i32_1_1_1_aligned()
19995   std::int32_t* result_chunk = result;  in gemm_i32_1_1_1_aligned()
19996   std::int32_t* mul_result_chunk = result;  in gemm_i32_1_1_1_aligned()
19997   const std::int32_t mul_result_chunk_stride_bytes = result_stride * 4;  in gemm_i32_1_1_1_aligned()
20040                             const std::uint8_t* rhs, std::int32_t m,  in gemm_i32_1_1_2_aligned()
20041                             std::int32_t n, std::int32_t k,  in gemm_i32_1_1_2_aligned()
20042                             std::int32_t lhs_offset, std::int32_t rhs_offset,  in gemm_i32_1_1_2_aligned()
20043                             std::int32_t* result, std::int32_t result_stride) {  in gemm_i32_1_1_2_aligned()
20044   const std::int32_t row_chunks = m / 3;  in gemm_i32_1_1_2_aligned()
20045   const std::int32_t col_chunks = n / 3;  in gemm_i32_1_1_2_aligned()
20046   const std::int32_t padded_k = ((k + 7) / 8) * 8;  in gemm_i32_1_1_2_aligned()
20047   const std::int32_t chunk_size = k * 3;  in gemm_i32_1_1_2_aligned()
20048   const std::int32_t zipped_chunk_size = (padded_k + 16) * 3;  in gemm_i32_1_1_2_aligned()
20049   const std::int32_t zipped_rhs_size = (padded_k + 16) * n;  in gemm_i32_1_1_2_aligned()
20053   std::int32_t* zipped_lhs_3_offsets =  in gemm_i32_1_1_2_aligned()
20054       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 3);  in gemm_i32_1_1_2_aligned()
20055   std::int32_t* zipped_lhs_1_offsets =  in gemm_i32_1_1_2_aligned()
20056       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 1);  in gemm_i32_1_1_2_aligned()
20059   const std::int32_t result_chunk_stride = result_stride * 3;  in gemm_i32_1_1_2_aligned()
20061   const std::int32_t const_offset = lhs_offset * rhs_offset * k;  in gemm_i32_1_1_2_aligned()
20062   std::int32_t* result_chunk = result;  in gemm_i32_1_1_2_aligned()
20063   std::int32_t* mul_result_chunk = result;  in gemm_i32_1_1_2_aligned()
20064   const std::int32_t mul_result_chunk_stride_bytes = result_stride * 4;  in gemm_i32_1_1_2_aligned()
20107                             const std::uint8_t* rhs, std::int32_t m,  in gemm_i32_1_1_3_aligned()
20108                             std::int32_t n, std::int32_t k,  in gemm_i32_1_1_3_aligned()
20109                             std::int32_t lhs_offset, std::int32_t rhs_offset,  in gemm_i32_1_1_3_aligned()
20110                             std::int32_t* result, std::int32_t result_stride) {  in gemm_i32_1_1_3_aligned()
20111   const std::int32_t row_chunks = m / 3;  in gemm_i32_1_1_3_aligned()
20112   const std::int32_t col_chunks = n / 3;  in gemm_i32_1_1_3_aligned()
20113   const std::int32_t padded_k = ((k + 7) / 8) * 8;  in gemm_i32_1_1_3_aligned()
20114   const std::int32_t chunk_size = k * 3;  in gemm_i32_1_1_3_aligned()
20115   const std::int32_t zipped_chunk_size = (padded_k + 16) * 3;  in gemm_i32_1_1_3_aligned()
20116   const std::int32_t zipped_rhs_size = (padded_k + 16) * n;  in gemm_i32_1_1_3_aligned()
20120   std::int32_t* zipped_lhs_3_offsets =  in gemm_i32_1_1_3_aligned()
20121       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 3);  in gemm_i32_1_1_3_aligned()
20122   std::int32_t* zipped_lhs_1_offsets =  in gemm_i32_1_1_3_aligned()
20123       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 1);  in gemm_i32_1_1_3_aligned()
20126   const std::int32_t result_chunk_stride = result_stride * 3;  in gemm_i32_1_1_3_aligned()
20128   const std::int32_t const_offset = lhs_offset * rhs_offset * k;  in gemm_i32_1_1_3_aligned()
20129   std::int32_t* result_chunk = result;  in gemm_i32_1_1_3_aligned()
20130   std::int32_t* mul_result_chunk = result;  in gemm_i32_1_1_3_aligned()
20131   const std::int32_t mul_result_chunk_stride_bytes = result_stride * 4;  in gemm_i32_1_1_3_aligned()
20174                             const std::uint8_t* rhs, std::int32_t m,  in gemm_i32_1_1_4_aligned()
20175                             std::int32_t n, std::int32_t k,  in gemm_i32_1_1_4_aligned()
20176                             std::int32_t lhs_offset, std::int32_t rhs_offset,  in gemm_i32_1_1_4_aligned()
20177                             std::int32_t* result, std::int32_t result_stride) {  in gemm_i32_1_1_4_aligned()
20178   const std::int32_t row_chunks = m / 3;  in gemm_i32_1_1_4_aligned()
20179   const std::int32_t col_chunks = n / 3;  in gemm_i32_1_1_4_aligned()
20180   const std::int32_t padded_k = ((k + 7) / 8) * 8;  in gemm_i32_1_1_4_aligned()
20181   const std::int32_t chunk_size = k * 3;  in gemm_i32_1_1_4_aligned()
20182   const std::int32_t zipped_chunk_size = (padded_k + 16) * 3;  in gemm_i32_1_1_4_aligned()
20183   const std::int32_t zipped_rhs_size = (padded_k + 16) * n;  in gemm_i32_1_1_4_aligned()
20187   std::int32_t* zipped_lhs_3_offsets =  in gemm_i32_1_1_4_aligned()
20188       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 3);  in gemm_i32_1_1_4_aligned()
20189   std::int32_t* zipped_lhs_1_offsets =  in gemm_i32_1_1_4_aligned()
20190       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 1);  in gemm_i32_1_1_4_aligned()
20193   const std::int32_t result_chunk_stride = result_stride * 3;  in gemm_i32_1_1_4_aligned()
20195   const std::int32_t const_offset = lhs_offset * rhs_offset * k;  in gemm_i32_1_1_4_aligned()
20196   std::int32_t* result_chunk = result;  in gemm_i32_1_1_4_aligned()
20197   std::int32_t* mul_result_chunk = result;  in gemm_i32_1_1_4_aligned()
20198   const std::int32_t mul_result_chunk_stride_bytes = result_stride * 4;  in gemm_i32_1_1_4_aligned()
20241                             const std::uint8_t* rhs, std::int32_t m,  in gemm_i32_1_1_5_aligned()
20242                             std::int32_t n, std::int32_t k,  in gemm_i32_1_1_5_aligned()
20243                             std::int32_t lhs_offset, std::int32_t rhs_offset,  in gemm_i32_1_1_5_aligned()
20244                             std::int32_t* result, std::int32_t result_stride) {  in gemm_i32_1_1_5_aligned()
20245   const std::int32_t row_chunks = m / 3;  in gemm_i32_1_1_5_aligned()
20246   const std::int32_t col_chunks = n / 3;  in gemm_i32_1_1_5_aligned()
20247   const std::int32_t padded_k = ((k + 7) / 8) * 8;  in gemm_i32_1_1_5_aligned()
20248   const std::int32_t chunk_size = k * 3;  in gemm_i32_1_1_5_aligned()
20249   const std::int32_t zipped_chunk_size = (padded_k + 16) * 3;  in gemm_i32_1_1_5_aligned()
20250   const std::int32_t zipped_rhs_size = (padded_k + 16) * n;  in gemm_i32_1_1_5_aligned()
20254   std::int32_t* zipped_lhs_3_offsets =  in gemm_i32_1_1_5_aligned()
20255       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 3);  in gemm_i32_1_1_5_aligned()
20256   std::int32_t* zipped_lhs_1_offsets =  in gemm_i32_1_1_5_aligned()
20257       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 1);  in gemm_i32_1_1_5_aligned()
20260   const std::int32_t result_chunk_stride = result_stride * 3;  in gemm_i32_1_1_5_aligned()
20262   const std::int32_t const_offset = lhs_offset * rhs_offset * k;  in gemm_i32_1_1_5_aligned()
20263   std::int32_t* result_chunk = result;  in gemm_i32_1_1_5_aligned()
20264   std::int32_t* mul_result_chunk = result;  in gemm_i32_1_1_5_aligned()
20265   const std::int32_t mul_result_chunk_stride_bytes = result_stride * 4;  in gemm_i32_1_1_5_aligned()
20308                             const std::uint8_t* rhs, std::int32_t m,  in gemm_i32_1_1_6_aligned()
20309                             std::int32_t n, std::int32_t k,  in gemm_i32_1_1_6_aligned()
20310                             std::int32_t lhs_offset, std::int32_t rhs_offset,  in gemm_i32_1_1_6_aligned()
20311                             std::int32_t* result, std::int32_t result_stride) {  in gemm_i32_1_1_6_aligned()
20312   const std::int32_t row_chunks = m / 3;  in gemm_i32_1_1_6_aligned()
20313   const std::int32_t col_chunks = n / 3;  in gemm_i32_1_1_6_aligned()
20314   const std::int32_t padded_k = ((k + 7) / 8) * 8;  in gemm_i32_1_1_6_aligned()
20315   const std::int32_t chunk_size = k * 3;  in gemm_i32_1_1_6_aligned()
20316   const std::int32_t zipped_chunk_size = (padded_k + 16) * 3;  in gemm_i32_1_1_6_aligned()
20317   const std::int32_t zipped_rhs_size = (padded_k + 16) * n;  in gemm_i32_1_1_6_aligned()
20321   std::int32_t* zipped_lhs_3_offsets =  in gemm_i32_1_1_6_aligned()
20322       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 3);  in gemm_i32_1_1_6_aligned()
20323   std::int32_t* zipped_lhs_1_offsets =  in gemm_i32_1_1_6_aligned()
20324       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 1);  in gemm_i32_1_1_6_aligned()
20327   const std::int32_t result_chunk_stride = result_stride * 3;  in gemm_i32_1_1_6_aligned()
20329   const std::int32_t const_offset = lhs_offset * rhs_offset * k;  in gemm_i32_1_1_6_aligned()
20330   std::int32_t* result_chunk = result;  in gemm_i32_1_1_6_aligned()
20331   std::int32_t* mul_result_chunk = result;  in gemm_i32_1_1_6_aligned()
20332   const std::int32_t mul_result_chunk_stride_bytes = result_stride * 4;  in gemm_i32_1_1_6_aligned()
20375                             const std::uint8_t* rhs, std::int32_t m,  in gemm_i32_1_1_7_aligned()
20376                             std::int32_t n, std::int32_t k,  in gemm_i32_1_1_7_aligned()
20377                             std::int32_t lhs_offset, std::int32_t rhs_offset,  in gemm_i32_1_1_7_aligned()
20378                             std::int32_t* result, std::int32_t result_stride) {  in gemm_i32_1_1_7_aligned()
20379   const std::int32_t row_chunks = m / 3;  in gemm_i32_1_1_7_aligned()
20380   const std::int32_t col_chunks = n / 3;  in gemm_i32_1_1_7_aligned()
20381   const std::int32_t padded_k = ((k + 7) / 8) * 8;  in gemm_i32_1_1_7_aligned()
20382   const std::int32_t chunk_size = k * 3;  in gemm_i32_1_1_7_aligned()
20383   const std::int32_t zipped_chunk_size = (padded_k + 16) * 3;  in gemm_i32_1_1_7_aligned()
20384   const std::int32_t zipped_rhs_size = (padded_k + 16) * n;  in gemm_i32_1_1_7_aligned()
20388   std::int32_t* zipped_lhs_3_offsets =  in gemm_i32_1_1_7_aligned()
20389       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 3);  in gemm_i32_1_1_7_aligned()
20390   std::int32_t* zipped_lhs_1_offsets =  in gemm_i32_1_1_7_aligned()
20391       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 1);  in gemm_i32_1_1_7_aligned()
20394   const std::int32_t result_chunk_stride = result_stride * 3;  in gemm_i32_1_1_7_aligned()
20396   const std::int32_t const_offset = lhs_offset * rhs_offset * k;  in gemm_i32_1_1_7_aligned()
20397   std::int32_t* result_chunk = result;  in gemm_i32_1_1_7_aligned()
20398   std::int32_t* mul_result_chunk = result;  in gemm_i32_1_1_7_aligned()
20399   const std::int32_t mul_result_chunk_stride_bytes = result_stride * 4;  in gemm_i32_1_1_7_aligned()
20442                             const std::uint8_t* rhs, std::int32_t m,  in gemm_i32_1_2_0_aligned()
20443                             std::int32_t n, std::int32_t k,  in gemm_i32_1_2_0_aligned()
20444                             std::int32_t lhs_offset, std::int32_t rhs_offset,  in gemm_i32_1_2_0_aligned()
20445                             std::int32_t* result, std::int32_t result_stride) {  in gemm_i32_1_2_0_aligned()
20446   const std::int32_t row_chunks = m / 3;  in gemm_i32_1_2_0_aligned()
20447   const std::int32_t col_chunks = n / 3;  in gemm_i32_1_2_0_aligned()
20448   const std::int32_t padded_k = ((k + 7) / 8) * 8;  in gemm_i32_1_2_0_aligned()
20449   const std::int32_t chunk_size = k * 3;  in gemm_i32_1_2_0_aligned()
20450   const std::int32_t zipped_chunk_size = (padded_k + 16) * 3;  in gemm_i32_1_2_0_aligned()
20451   const std::int32_t zipped_rhs_size = (padded_k + 16) * n;  in gemm_i32_1_2_0_aligned()
20455   std::int32_t* zipped_lhs_3_offsets =  in gemm_i32_1_2_0_aligned()
20456       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 3);  in gemm_i32_1_2_0_aligned()
20457   std::int32_t* zipped_lhs_1_offsets =  in gemm_i32_1_2_0_aligned()
20458       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 1);  in gemm_i32_1_2_0_aligned()
20461   const std::int32_t result_chunk_stride = result_stride * 3;  in gemm_i32_1_2_0_aligned()
20463   const std::int32_t const_offset = lhs_offset * rhs_offset * k;  in gemm_i32_1_2_0_aligned()
20464   std::int32_t* result_chunk = result;  in gemm_i32_1_2_0_aligned()
20465   std::int32_t* mul_result_chunk = result;  in gemm_i32_1_2_0_aligned()
20466   const std::int32_t mul_result_chunk_stride_bytes = result_stride * 4;  in gemm_i32_1_2_0_aligned()
20509                             const std::uint8_t* rhs, std::int32_t m,  in gemm_i32_1_2_1_aligned()
20510                             std::int32_t n, std::int32_t k,  in gemm_i32_1_2_1_aligned()
20511                             std::int32_t lhs_offset, std::int32_t rhs_offset,  in gemm_i32_1_2_1_aligned()
20512                             std::int32_t* result, std::int32_t result_stride) {  in gemm_i32_1_2_1_aligned()
20513   const std::int32_t row_chunks = m / 3;  in gemm_i32_1_2_1_aligned()
20514   const std::int32_t col_chunks = n / 3;  in gemm_i32_1_2_1_aligned()
20515   const std::int32_t padded_k = ((k + 7) / 8) * 8;  in gemm_i32_1_2_1_aligned()
20516   const std::int32_t chunk_size = k * 3;  in gemm_i32_1_2_1_aligned()
20517   const std::int32_t zipped_chunk_size = (padded_k + 16) * 3;  in gemm_i32_1_2_1_aligned()
20518   const std::int32_t zipped_rhs_size = (padded_k + 16) * n;  in gemm_i32_1_2_1_aligned()
20522   std::int32_t* zipped_lhs_3_offsets =  in gemm_i32_1_2_1_aligned()
20523       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 3);  in gemm_i32_1_2_1_aligned()
20524   std::int32_t* zipped_lhs_1_offsets =  in gemm_i32_1_2_1_aligned()
20525       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 1);  in gemm_i32_1_2_1_aligned()
20528   const std::int32_t result_chunk_stride = result_stride * 3;  in gemm_i32_1_2_1_aligned()
20530   const std::int32_t const_offset = lhs_offset * rhs_offset * k;  in gemm_i32_1_2_1_aligned()
20531   std::int32_t* result_chunk = result;  in gemm_i32_1_2_1_aligned()
20532   std::int32_t* mul_result_chunk = result;  in gemm_i32_1_2_1_aligned()
20533   const std::int32_t mul_result_chunk_stride_bytes = result_stride * 4;  in gemm_i32_1_2_1_aligned()
20576                             const std::uint8_t* rhs, std::int32_t m,  in gemm_i32_1_2_2_aligned()
20577                             std::int32_t n, std::int32_t k,  in gemm_i32_1_2_2_aligned()
20578                             std::int32_t lhs_offset, std::int32_t rhs_offset,  in gemm_i32_1_2_2_aligned()
20579                             std::int32_t* result, std::int32_t result_stride) {  in gemm_i32_1_2_2_aligned()
20580   const std::int32_t row_chunks = m / 3;  in gemm_i32_1_2_2_aligned()
20581   const std::int32_t col_chunks = n / 3;  in gemm_i32_1_2_2_aligned()
20582   const std::int32_t padded_k = ((k + 7) / 8) * 8;  in gemm_i32_1_2_2_aligned()
20583   const std::int32_t chunk_size = k * 3;  in gemm_i32_1_2_2_aligned()
20584   const std::int32_t zipped_chunk_size = (padded_k + 16) * 3;  in gemm_i32_1_2_2_aligned()
20585   const std::int32_t zipped_rhs_size = (padded_k + 16) * n;  in gemm_i32_1_2_2_aligned()
20589   std::int32_t* zipped_lhs_3_offsets =  in gemm_i32_1_2_2_aligned()
20590       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 3);  in gemm_i32_1_2_2_aligned()
20591   std::int32_t* zipped_lhs_1_offsets =  in gemm_i32_1_2_2_aligned()
20592       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 1);  in gemm_i32_1_2_2_aligned()
20595   const std::int32_t result_chunk_stride = result_stride * 3;  in gemm_i32_1_2_2_aligned()
20597   const std::int32_t const_offset = lhs_offset * rhs_offset * k;  in gemm_i32_1_2_2_aligned()
20598   std::int32_t* result_chunk = result;  in gemm_i32_1_2_2_aligned()
20599   std::int32_t* mul_result_chunk = result;  in gemm_i32_1_2_2_aligned()
20600   const std::int32_t mul_result_chunk_stride_bytes = result_stride * 4;  in gemm_i32_1_2_2_aligned()
20643                             const std::uint8_t* rhs, std::int32_t m,  in gemm_i32_1_2_3_aligned()
20644                             std::int32_t n, std::int32_t k,  in gemm_i32_1_2_3_aligned()
20645                             std::int32_t lhs_offset, std::int32_t rhs_offset,  in gemm_i32_1_2_3_aligned()
20646                             std::int32_t* result, std::int32_t result_stride) {  in gemm_i32_1_2_3_aligned()
20647   const std::int32_t row_chunks = m / 3;  in gemm_i32_1_2_3_aligned()
20648   const std::int32_t col_chunks = n / 3;  in gemm_i32_1_2_3_aligned()
20649   const std::int32_t padded_k = ((k + 7) / 8) * 8;  in gemm_i32_1_2_3_aligned()
20650   const std::int32_t chunk_size = k * 3;  in gemm_i32_1_2_3_aligned()
20651   const std::int32_t zipped_chunk_size = (padded_k + 16) * 3;  in gemm_i32_1_2_3_aligned()
20652   const std::int32_t zipped_rhs_size = (padded_k + 16) * n;  in gemm_i32_1_2_3_aligned()
20656   std::int32_t* zipped_lhs_3_offsets =  in gemm_i32_1_2_3_aligned()
20657       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 3);  in gemm_i32_1_2_3_aligned()
20658   std::int32_t* zipped_lhs_1_offsets =  in gemm_i32_1_2_3_aligned()
20659       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 1);  in gemm_i32_1_2_3_aligned()
20662   const std::int32_t result_chunk_stride = result_stride * 3;  in gemm_i32_1_2_3_aligned()
20664   const std::int32_t const_offset = lhs_offset * rhs_offset * k;  in gemm_i32_1_2_3_aligned()
20665   std::int32_t* result_chunk = result;  in gemm_i32_1_2_3_aligned()
20666   std::int32_t* mul_result_chunk = result;  in gemm_i32_1_2_3_aligned()
20667   const std::int32_t mul_result_chunk_stride_bytes = result_stride * 4;  in gemm_i32_1_2_3_aligned()
20710                             const std::uint8_t* rhs, std::int32_t m,  in gemm_i32_1_2_4_aligned()
20711                             std::int32_t n, std::int32_t k,  in gemm_i32_1_2_4_aligned()
20712                             std::int32_t lhs_offset, std::int32_t rhs_offset,  in gemm_i32_1_2_4_aligned()
20713                             std::int32_t* result, std::int32_t result_stride) {  in gemm_i32_1_2_4_aligned()
20714   const std::int32_t row_chunks = m / 3;  in gemm_i32_1_2_4_aligned()
20715   const std::int32_t col_chunks = n / 3;  in gemm_i32_1_2_4_aligned()
20716   const std::int32_t padded_k = ((k + 7) / 8) * 8;  in gemm_i32_1_2_4_aligned()
20717   const std::int32_t chunk_size = k * 3;  in gemm_i32_1_2_4_aligned()
20718   const std::int32_t zipped_chunk_size = (padded_k + 16) * 3;  in gemm_i32_1_2_4_aligned()
20719   const std::int32_t zipped_rhs_size = (padded_k + 16) * n;  in gemm_i32_1_2_4_aligned()
20723   std::int32_t* zipped_lhs_3_offsets =  in gemm_i32_1_2_4_aligned()
20724       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 3);  in gemm_i32_1_2_4_aligned()
20725   std::int32_t* zipped_lhs_1_offsets =  in gemm_i32_1_2_4_aligned()
20726       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 1);  in gemm_i32_1_2_4_aligned()
20729   const std::int32_t result_chunk_stride = result_stride * 3;  in gemm_i32_1_2_4_aligned()
20731   const std::int32_t const_offset = lhs_offset * rhs_offset * k;  in gemm_i32_1_2_4_aligned()
20732   std::int32_t* result_chunk = result;  in gemm_i32_1_2_4_aligned()
20733   std::int32_t* mul_result_chunk = result;  in gemm_i32_1_2_4_aligned()
20734   const std::int32_t mul_result_chunk_stride_bytes = result_stride * 4;  in gemm_i32_1_2_4_aligned()
20777                             const std::uint8_t* rhs, std::int32_t m,  in gemm_i32_1_2_5_aligned()
20778                             std::int32_t n, std::int32_t k,  in gemm_i32_1_2_5_aligned()
20779                             std::int32_t lhs_offset, std::int32_t rhs_offset,  in gemm_i32_1_2_5_aligned()
20780                             std::int32_t* result, std::int32_t result_stride) {  in gemm_i32_1_2_5_aligned()
20781   const std::int32_t row_chunks = m / 3;  in gemm_i32_1_2_5_aligned()
20782   const std::int32_t col_chunks = n / 3;  in gemm_i32_1_2_5_aligned()
20783   const std::int32_t padded_k = ((k + 7) / 8) * 8;  in gemm_i32_1_2_5_aligned()
20784   const std::int32_t chunk_size = k * 3;  in gemm_i32_1_2_5_aligned()
20785   const std::int32_t zipped_chunk_size = (padded_k + 16) * 3;  in gemm_i32_1_2_5_aligned()
20786   const std::int32_t zipped_rhs_size = (padded_k + 16) * n;  in gemm_i32_1_2_5_aligned()
20790   std::int32_t* zipped_lhs_3_offsets =  in gemm_i32_1_2_5_aligned()
20791       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 3);  in gemm_i32_1_2_5_aligned()
20792   std::int32_t* zipped_lhs_1_offsets =  in gemm_i32_1_2_5_aligned()
20793       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 1);  in gemm_i32_1_2_5_aligned()
20796   const std::int32_t result_chunk_stride = result_stride * 3;  in gemm_i32_1_2_5_aligned()
20798   const std::int32_t const_offset = lhs_offset * rhs_offset * k;  in gemm_i32_1_2_5_aligned()
20799   std::int32_t* result_chunk = result;  in gemm_i32_1_2_5_aligned()
20800   std::int32_t* mul_result_chunk = result;  in gemm_i32_1_2_5_aligned()
20801   const std::int32_t mul_result_chunk_stride_bytes = result_stride * 4;  in gemm_i32_1_2_5_aligned()
20844                             const std::uint8_t* rhs, std::int32_t m,  in gemm_i32_1_2_6_aligned()
20845                             std::int32_t n, std::int32_t k,  in gemm_i32_1_2_6_aligned()
20846                             std::int32_t lhs_offset, std::int32_t rhs_offset,  in gemm_i32_1_2_6_aligned()
20847                             std::int32_t* result, std::int32_t result_stride) {  in gemm_i32_1_2_6_aligned()
20848   const std::int32_t row_chunks = m / 3;  in gemm_i32_1_2_6_aligned()
20849   const std::int32_t col_chunks = n / 3;  in gemm_i32_1_2_6_aligned()
20850   const std::int32_t padded_k = ((k + 7) / 8) * 8;  in gemm_i32_1_2_6_aligned()
20851   const std::int32_t chunk_size = k * 3;  in gemm_i32_1_2_6_aligned()
20852   const std::int32_t zipped_chunk_size = (padded_k + 16) * 3;  in gemm_i32_1_2_6_aligned()
20853   const std::int32_t zipped_rhs_size = (padded_k + 16) * n;  in gemm_i32_1_2_6_aligned()
20857   std::int32_t* zipped_lhs_3_offsets =  in gemm_i32_1_2_6_aligned()
20858       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 3);  in gemm_i32_1_2_6_aligned()
20859   std::int32_t* zipped_lhs_1_offsets =  in gemm_i32_1_2_6_aligned()
20860       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 1);  in gemm_i32_1_2_6_aligned()
20863   const std::int32_t result_chunk_stride = result_stride * 3;  in gemm_i32_1_2_6_aligned()
20865   const std::int32_t const_offset = lhs_offset * rhs_offset * k;  in gemm_i32_1_2_6_aligned()
20866   std::int32_t* result_chunk = result;  in gemm_i32_1_2_6_aligned()
20867   std::int32_t* mul_result_chunk = result;  in gemm_i32_1_2_6_aligned()
20868   const std::int32_t mul_result_chunk_stride_bytes = result_stride * 4;  in gemm_i32_1_2_6_aligned()
20911                             const std::uint8_t* rhs, std::int32_t m,  in gemm_i32_1_2_7_aligned()
20912                             std::int32_t n, std::int32_t k,  in gemm_i32_1_2_7_aligned()
20913                             std::int32_t lhs_offset, std::int32_t rhs_offset,  in gemm_i32_1_2_7_aligned()
20914                             std::int32_t* result, std::int32_t result_stride) {  in gemm_i32_1_2_7_aligned()
20915   const std::int32_t row_chunks = m / 3;  in gemm_i32_1_2_7_aligned()
20916   const std::int32_t col_chunks = n / 3;  in gemm_i32_1_2_7_aligned()
20917   const std::int32_t padded_k = ((k + 7) / 8) * 8;  in gemm_i32_1_2_7_aligned()
20918   const std::int32_t chunk_size = k * 3;  in gemm_i32_1_2_7_aligned()
20919   const std::int32_t zipped_chunk_size = (padded_k + 16) * 3;  in gemm_i32_1_2_7_aligned()
20920   const std::int32_t zipped_rhs_size = (padded_k + 16) * n;  in gemm_i32_1_2_7_aligned()
20924   std::int32_t* zipped_lhs_3_offsets =  in gemm_i32_1_2_7_aligned()
20925       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 3);  in gemm_i32_1_2_7_aligned()
20926   std::int32_t* zipped_lhs_1_offsets =  in gemm_i32_1_2_7_aligned()
20927       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 1);  in gemm_i32_1_2_7_aligned()
20930   const std::int32_t result_chunk_stride = result_stride * 3;  in gemm_i32_1_2_7_aligned()
20932   const std::int32_t const_offset = lhs_offset * rhs_offset * k;  in gemm_i32_1_2_7_aligned()
20933   std::int32_t* result_chunk = result;  in gemm_i32_1_2_7_aligned()
20934   std::int32_t* mul_result_chunk = result;  in gemm_i32_1_2_7_aligned()
20935   const std::int32_t mul_result_chunk_stride_bytes = result_stride * 4;  in gemm_i32_1_2_7_aligned()
20978                             const std::uint8_t* rhs, std::int32_t m,  in gemm_i32_2_0_0_aligned()
20979                             std::int32_t n, std::int32_t k,  in gemm_i32_2_0_0_aligned()
20980                             std::int32_t lhs_offset, std::int32_t rhs_offset,  in gemm_i32_2_0_0_aligned()
20981                             std::int32_t* result, std::int32_t result_stride) {  in gemm_i32_2_0_0_aligned()
20982   const std::int32_t row_chunks = m / 3;  in gemm_i32_2_0_0_aligned()
20983   const std::int32_t col_chunks = n / 3;  in gemm_i32_2_0_0_aligned()
20984   const std::int32_t padded_k = ((k + 7) / 8) * 8;  in gemm_i32_2_0_0_aligned()
20985   const std::int32_t chunk_size = k * 3;  in gemm_i32_2_0_0_aligned()
20986   const std::int32_t zipped_chunk_size = (padded_k + 16) * 3;  in gemm_i32_2_0_0_aligned()
20987   const std::int32_t zipped_rhs_size = (padded_k + 16) * n;  in gemm_i32_2_0_0_aligned()
20991   std::int32_t* zipped_lhs_3_offsets =  in gemm_i32_2_0_0_aligned()
20992       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 3);  in gemm_i32_2_0_0_aligned()
20993   std::int32_t* zipped_lhs_2_offsets =  in gemm_i32_2_0_0_aligned()
20994       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 2);  in gemm_i32_2_0_0_aligned()
20997   const std::int32_t result_chunk_stride = result_stride * 3;  in gemm_i32_2_0_0_aligned()
20999   const std::int32_t const_offset = lhs_offset * rhs_offset * k;  in gemm_i32_2_0_0_aligned()
21000   std::int32_t* result_chunk = result;  in gemm_i32_2_0_0_aligned()
21001   std::int32_t* mul_result_chunk = result;  in gemm_i32_2_0_0_aligned()
21002   const std::int32_t mul_result_chunk_stride_bytes = result_stride * 4;  in gemm_i32_2_0_0_aligned()
21038                             const std::uint8_t* rhs, std::int32_t m,  in gemm_i32_2_0_1_aligned()
21039                             std::int32_t n, std::int32_t k,  in gemm_i32_2_0_1_aligned()
21040                             std::int32_t lhs_offset, std::int32_t rhs_offset,  in gemm_i32_2_0_1_aligned()
21041                             std::int32_t* result, std::int32_t result_stride) {  in gemm_i32_2_0_1_aligned()
21042   const std::int32_t row_chunks = m / 3;  in gemm_i32_2_0_1_aligned()
21043   const std::int32_t col_chunks = n / 3;  in gemm_i32_2_0_1_aligned()
21044   const std::int32_t padded_k = ((k + 7) / 8) * 8;  in gemm_i32_2_0_1_aligned()
21045   const std::int32_t chunk_size = k * 3;  in gemm_i32_2_0_1_aligned()
21046   const std::int32_t zipped_chunk_size = (padded_k + 16) * 3;  in gemm_i32_2_0_1_aligned()
21047   const std::int32_t zipped_rhs_size = (padded_k + 16) * n;  in gemm_i32_2_0_1_aligned()
21051   std::int32_t* zipped_lhs_3_offsets =  in gemm_i32_2_0_1_aligned()
21052       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 3);  in gemm_i32_2_0_1_aligned()
21053   std::int32_t* zipped_lhs_2_offsets =  in gemm_i32_2_0_1_aligned()
21054       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 2);  in gemm_i32_2_0_1_aligned()
21057   const std::int32_t result_chunk_stride = result_stride * 3;  in gemm_i32_2_0_1_aligned()
21059   const std::int32_t const_offset = lhs_offset * rhs_offset * k;  in gemm_i32_2_0_1_aligned()
21060   std::int32_t* result_chunk = result;  in gemm_i32_2_0_1_aligned()
21061   std::int32_t* mul_result_chunk = result;  in gemm_i32_2_0_1_aligned()
21062   const std::int32_t mul_result_chunk_stride_bytes = result_stride * 4;  in gemm_i32_2_0_1_aligned()
21098                             const std::uint8_t* rhs, std::int32_t m,  in gemm_i32_2_0_2_aligned()
21099                             std::int32_t n, std::int32_t k,  in gemm_i32_2_0_2_aligned()
21100                             std::int32_t lhs_offset, std::int32_t rhs_offset,  in gemm_i32_2_0_2_aligned()
21101                             std::int32_t* result, std::int32_t result_stride) {  in gemm_i32_2_0_2_aligned()
21102   const std::int32_t row_chunks = m / 3;  in gemm_i32_2_0_2_aligned()
21103   const std::int32_t col_chunks = n / 3;  in gemm_i32_2_0_2_aligned()
21104   const std::int32_t padded_k = ((k + 7) / 8) * 8;  in gemm_i32_2_0_2_aligned()
21105   const std::int32_t chunk_size = k * 3;  in gemm_i32_2_0_2_aligned()
21106   const std::int32_t zipped_chunk_size = (padded_k + 16) * 3;  in gemm_i32_2_0_2_aligned()
21107   const std::int32_t zipped_rhs_size = (padded_k + 16) * n;  in gemm_i32_2_0_2_aligned()
21111   std::int32_t* zipped_lhs_3_offsets =  in gemm_i32_2_0_2_aligned()
21112       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 3);  in gemm_i32_2_0_2_aligned()
21113   std::int32_t* zipped_lhs_2_offsets =  in gemm_i32_2_0_2_aligned()
21114       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 2);  in gemm_i32_2_0_2_aligned()
21117   const std::int32_t result_chunk_stride = result_stride * 3;  in gemm_i32_2_0_2_aligned()
21119   const std::int32_t const_offset = lhs_offset * rhs_offset * k;  in gemm_i32_2_0_2_aligned()
21120   std::int32_t* result_chunk = result;  in gemm_i32_2_0_2_aligned()
21121   std::int32_t* mul_result_chunk = result;  in gemm_i32_2_0_2_aligned()
21122   const std::int32_t mul_result_chunk_stride_bytes = result_stride * 4;  in gemm_i32_2_0_2_aligned()
21158                             const std::uint8_t* rhs, std::int32_t m,  in gemm_i32_2_0_3_aligned()
21159                             std::int32_t n, std::int32_t k,  in gemm_i32_2_0_3_aligned()
21160                             std::int32_t lhs_offset, std::int32_t rhs_offset,  in gemm_i32_2_0_3_aligned()
21161                             std::int32_t* result, std::int32_t result_stride) {  in gemm_i32_2_0_3_aligned()
21162   const std::int32_t row_chunks = m / 3;  in gemm_i32_2_0_3_aligned()
21163   const std::int32_t col_chunks = n / 3;  in gemm_i32_2_0_3_aligned()
21164   const std::int32_t padded_k = ((k + 7) / 8) * 8;  in gemm_i32_2_0_3_aligned()
21165   const std::int32_t chunk_size = k * 3;  in gemm_i32_2_0_3_aligned()
21166   const std::int32_t zipped_chunk_size = (padded_k + 16) * 3;  in gemm_i32_2_0_3_aligned()
21167   const std::int32_t zipped_rhs_size = (padded_k + 16) * n;  in gemm_i32_2_0_3_aligned()
21171   std::int32_t* zipped_lhs_3_offsets =  in gemm_i32_2_0_3_aligned()
21172       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 3);  in gemm_i32_2_0_3_aligned()
21173   std::int32_t* zipped_lhs_2_offsets =  in gemm_i32_2_0_3_aligned()
21174       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 2);  in gemm_i32_2_0_3_aligned()
21177   const std::int32_t result_chunk_stride = result_stride * 3;  in gemm_i32_2_0_3_aligned()
21179   const std::int32_t const_offset = lhs_offset * rhs_offset * k;  in gemm_i32_2_0_3_aligned()
21180   std::int32_t* result_chunk = result;  in gemm_i32_2_0_3_aligned()
21181   std::int32_t* mul_result_chunk = result;  in gemm_i32_2_0_3_aligned()
21182   const std::int32_t mul_result_chunk_stride_bytes = result_stride * 4;  in gemm_i32_2_0_3_aligned()
21218                             const std::uint8_t* rhs, std::int32_t m,  in gemm_i32_2_0_4_aligned()
21219                             std::int32_t n, std::int32_t k,  in gemm_i32_2_0_4_aligned()
21220                             std::int32_t lhs_offset, std::int32_t rhs_offset,  in gemm_i32_2_0_4_aligned()
21221                             std::int32_t* result, std::int32_t result_stride) {  in gemm_i32_2_0_4_aligned()
21222   const std::int32_t row_chunks = m / 3;  in gemm_i32_2_0_4_aligned()
21223   const std::int32_t col_chunks = n / 3;  in gemm_i32_2_0_4_aligned()
21224   const std::int32_t padded_k = ((k + 7) / 8) * 8;  in gemm_i32_2_0_4_aligned()
21225   const std::int32_t chunk_size = k * 3;  in gemm_i32_2_0_4_aligned()
21226   const std::int32_t zipped_chunk_size = (padded_k + 16) * 3;  in gemm_i32_2_0_4_aligned()
21227   const std::int32_t zipped_rhs_size = (padded_k + 16) * n;  in gemm_i32_2_0_4_aligned()
21231   std::int32_t* zipped_lhs_3_offsets =  in gemm_i32_2_0_4_aligned()
21232       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 3);  in gemm_i32_2_0_4_aligned()
21233   std::int32_t* zipped_lhs_2_offsets =  in gemm_i32_2_0_4_aligned()
21234       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 2);  in gemm_i32_2_0_4_aligned()
21237   const std::int32_t result_chunk_stride = result_stride * 3;  in gemm_i32_2_0_4_aligned()
21239   const std::int32_t const_offset = lhs_offset * rhs_offset * k;  in gemm_i32_2_0_4_aligned()
21240   std::int32_t* result_chunk = result;  in gemm_i32_2_0_4_aligned()
21241   std::int32_t* mul_result_chunk = result;  in gemm_i32_2_0_4_aligned()
21242   const std::int32_t mul_result_chunk_stride_bytes = result_stride * 4;  in gemm_i32_2_0_4_aligned()
21278                             const std::uint8_t* rhs, std::int32_t m,  in gemm_i32_2_0_5_aligned()
21279                             std::int32_t n, std::int32_t k,  in gemm_i32_2_0_5_aligned()
21280                             std::int32_t lhs_offset, std::int32_t rhs_offset,  in gemm_i32_2_0_5_aligned()
21281                             std::int32_t* result, std::int32_t result_stride) {  in gemm_i32_2_0_5_aligned()
21282   const std::int32_t row_chunks = m / 3;  in gemm_i32_2_0_5_aligned()
21283   const std::int32_t col_chunks = n / 3;  in gemm_i32_2_0_5_aligned()
21284   const std::int32_t padded_k = ((k + 7) / 8) * 8;  in gemm_i32_2_0_5_aligned()
21285   const std::int32_t chunk_size = k * 3;  in gemm_i32_2_0_5_aligned()
21286   const std::int32_t zipped_chunk_size = (padded_k + 16) * 3;  in gemm_i32_2_0_5_aligned()
21287   const std::int32_t zipped_rhs_size = (padded_k + 16) * n;  in gemm_i32_2_0_5_aligned()
21291   std::int32_t* zipped_lhs_3_offsets =  in gemm_i32_2_0_5_aligned()
21292       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 3);  in gemm_i32_2_0_5_aligned()
21293   std::int32_t* zipped_lhs_2_offsets =  in gemm_i32_2_0_5_aligned()
21294       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 2);  in gemm_i32_2_0_5_aligned()
21297   const std::int32_t result_chunk_stride = result_stride * 3;  in gemm_i32_2_0_5_aligned()
21299   const std::int32_t const_offset = lhs_offset * rhs_offset * k;  in gemm_i32_2_0_5_aligned()
21300   std::int32_t* result_chunk = result;  in gemm_i32_2_0_5_aligned()
21301   std::int32_t* mul_result_chunk = result;  in gemm_i32_2_0_5_aligned()
21302   const std::int32_t mul_result_chunk_stride_bytes = result_stride * 4;  in gemm_i32_2_0_5_aligned()
21338                             const std::uint8_t* rhs, std::int32_t m,  in gemm_i32_2_0_6_aligned()
21339                             std::int32_t n, std::int32_t k,  in gemm_i32_2_0_6_aligned()
21340                             std::int32_t lhs_offset, std::int32_t rhs_offset,  in gemm_i32_2_0_6_aligned()
21341                             std::int32_t* result, std::int32_t result_stride) {  in gemm_i32_2_0_6_aligned()
21342   const std::int32_t row_chunks = m / 3;  in gemm_i32_2_0_6_aligned()
21343   const std::int32_t col_chunks = n / 3;  in gemm_i32_2_0_6_aligned()
21344   const std::int32_t padded_k = ((k + 7) / 8) * 8;  in gemm_i32_2_0_6_aligned()
21345   const std::int32_t chunk_size = k * 3;  in gemm_i32_2_0_6_aligned()
21346   const std::int32_t zipped_chunk_size = (padded_k + 16) * 3;  in gemm_i32_2_0_6_aligned()
21347   const std::int32_t zipped_rhs_size = (padded_k + 16) * n;  in gemm_i32_2_0_6_aligned()
21351   std::int32_t* zipped_lhs_3_offsets =  in gemm_i32_2_0_6_aligned()
21352       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 3);  in gemm_i32_2_0_6_aligned()
21353   std::int32_t* zipped_lhs_2_offsets =  in gemm_i32_2_0_6_aligned()
21354       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 2);  in gemm_i32_2_0_6_aligned()
21357   const std::int32_t result_chunk_stride = result_stride * 3;  in gemm_i32_2_0_6_aligned()
21359   const std::int32_t const_offset = lhs_offset * rhs_offset * k;  in gemm_i32_2_0_6_aligned()
21360   std::int32_t* result_chunk = result;  in gemm_i32_2_0_6_aligned()
21361   std::int32_t* mul_result_chunk = result;  in gemm_i32_2_0_6_aligned()
21362   const std::int32_t mul_result_chunk_stride_bytes = result_stride * 4;  in gemm_i32_2_0_6_aligned()
21398                             const std::uint8_t* rhs, std::int32_t m,  in gemm_i32_2_0_7_aligned()
21399                             std::int32_t n, std::int32_t k,  in gemm_i32_2_0_7_aligned()
21400                             std::int32_t lhs_offset, std::int32_t rhs_offset,  in gemm_i32_2_0_7_aligned()
21401                             std::int32_t* result, std::int32_t result_stride) {  in gemm_i32_2_0_7_aligned()
21402   const std::int32_t row_chunks = m / 3;  in gemm_i32_2_0_7_aligned()
21403   const std::int32_t col_chunks = n / 3;  in gemm_i32_2_0_7_aligned()
21404   const std::int32_t padded_k = ((k + 7) / 8) * 8;  in gemm_i32_2_0_7_aligned()
21405   const std::int32_t chunk_size = k * 3;  in gemm_i32_2_0_7_aligned()
21406   const std::int32_t zipped_chunk_size = (padded_k + 16) * 3;  in gemm_i32_2_0_7_aligned()
21407   const std::int32_t zipped_rhs_size = (padded_k + 16) * n;  in gemm_i32_2_0_7_aligned()
21411   std::int32_t* zipped_lhs_3_offsets =  in gemm_i32_2_0_7_aligned()
21412       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 3);  in gemm_i32_2_0_7_aligned()
21413   std::int32_t* zipped_lhs_2_offsets =  in gemm_i32_2_0_7_aligned()
21414       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 2);  in gemm_i32_2_0_7_aligned()
21417   const std::int32_t result_chunk_stride = result_stride * 3;  in gemm_i32_2_0_7_aligned()
21419   const std::int32_t const_offset = lhs_offset * rhs_offset * k;  in gemm_i32_2_0_7_aligned()
21420   std::int32_t* result_chunk = result;  in gemm_i32_2_0_7_aligned()
21421   std::int32_t* mul_result_chunk = result;  in gemm_i32_2_0_7_aligned()
21422   const std::int32_t mul_result_chunk_stride_bytes = result_stride * 4;  in gemm_i32_2_0_7_aligned()
21458                             const std::uint8_t* rhs, std::int32_t m,  in gemm_i32_2_1_0_aligned()
21459                             std::int32_t n, std::int32_t k,  in gemm_i32_2_1_0_aligned()
21460                             std::int32_t lhs_offset, std::int32_t rhs_offset,  in gemm_i32_2_1_0_aligned()
21461                             std::int32_t* result, std::int32_t result_stride) {  in gemm_i32_2_1_0_aligned()
21462   const std::int32_t row_chunks = m / 3;  in gemm_i32_2_1_0_aligned()
21463   const std::int32_t col_chunks = n / 3;  in gemm_i32_2_1_0_aligned()
21464   const std::int32_t padded_k = ((k + 7) / 8) * 8;  in gemm_i32_2_1_0_aligned()
21465   const std::int32_t chunk_size = k * 3;  in gemm_i32_2_1_0_aligned()
21466   const std::int32_t zipped_chunk_size = (padded_k + 16) * 3;  in gemm_i32_2_1_0_aligned()
21467   const std::int32_t zipped_rhs_size = (padded_k + 16) * n;  in gemm_i32_2_1_0_aligned()
21471   std::int32_t* zipped_lhs_3_offsets =  in gemm_i32_2_1_0_aligned()
21472       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 3);  in gemm_i32_2_1_0_aligned()
21473   std::int32_t* zipped_lhs_2_offsets =  in gemm_i32_2_1_0_aligned()
21474       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 2);  in gemm_i32_2_1_0_aligned()
21477   const std::int32_t result_chunk_stride = result_stride * 3;  in gemm_i32_2_1_0_aligned()
21479   const std::int32_t const_offset = lhs_offset * rhs_offset * k;  in gemm_i32_2_1_0_aligned()
21480   std::int32_t* result_chunk = result;  in gemm_i32_2_1_0_aligned()
21481   std::int32_t* mul_result_chunk = result;  in gemm_i32_2_1_0_aligned()
21482   const std::int32_t mul_result_chunk_stride_bytes = result_stride * 4;  in gemm_i32_2_1_0_aligned()
21525                             const std::uint8_t* rhs, std::int32_t m,  in gemm_i32_2_1_1_aligned()
21526                             std::int32_t n, std::int32_t k,  in gemm_i32_2_1_1_aligned()
21527                             std::int32_t lhs_offset, std::int32_t rhs_offset,  in gemm_i32_2_1_1_aligned()
21528                             std::int32_t* result, std::int32_t result_stride) {  in gemm_i32_2_1_1_aligned()
21529   const std::int32_t row_chunks = m / 3;  in gemm_i32_2_1_1_aligned()
21530   const std::int32_t col_chunks = n / 3;  in gemm_i32_2_1_1_aligned()
21531   const std::int32_t padded_k = ((k + 7) / 8) * 8;  in gemm_i32_2_1_1_aligned()
21532   const std::int32_t chunk_size = k * 3;  in gemm_i32_2_1_1_aligned()
21533   const std::int32_t zipped_chunk_size = (padded_k + 16) * 3;  in gemm_i32_2_1_1_aligned()
21534   const std::int32_t zipped_rhs_size = (padded_k + 16) * n;  in gemm_i32_2_1_1_aligned()
21538   std::int32_t* zipped_lhs_3_offsets =  in gemm_i32_2_1_1_aligned()
21539       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 3);  in gemm_i32_2_1_1_aligned()
21540   std::int32_t* zipped_lhs_2_offsets =  in gemm_i32_2_1_1_aligned()
21541       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 2);  in gemm_i32_2_1_1_aligned()
21544   const std::int32_t result_chunk_stride = result_stride * 3;  in gemm_i32_2_1_1_aligned()
21546   const std::int32_t const_offset = lhs_offset * rhs_offset * k;  in gemm_i32_2_1_1_aligned()
21547   std::int32_t* result_chunk = result;  in gemm_i32_2_1_1_aligned()
21548   std::int32_t* mul_result_chunk = result;  in gemm_i32_2_1_1_aligned()
21549   const std::int32_t mul_result_chunk_stride_bytes = result_stride * 4;  in gemm_i32_2_1_1_aligned()
21592                             const std::uint8_t* rhs, std::int32_t m,  in gemm_i32_2_1_2_aligned()
21593                             std::int32_t n, std::int32_t k,  in gemm_i32_2_1_2_aligned()
21594                             std::int32_t lhs_offset, std::int32_t rhs_offset,  in gemm_i32_2_1_2_aligned()
21595                             std::int32_t* result, std::int32_t result_stride) {  in gemm_i32_2_1_2_aligned()
21596   const std::int32_t row_chunks = m / 3;  in gemm_i32_2_1_2_aligned()
21597   const std::int32_t col_chunks = n / 3;  in gemm_i32_2_1_2_aligned()
21598   const std::int32_t padded_k = ((k + 7) / 8) * 8;  in gemm_i32_2_1_2_aligned()
21599   const std::int32_t chunk_size = k * 3;  in gemm_i32_2_1_2_aligned()
21600   const std::int32_t zipped_chunk_size = (padded_k + 16) * 3;  in gemm_i32_2_1_2_aligned()
21601   const std::int32_t zipped_rhs_size = (padded_k + 16) * n;  in gemm_i32_2_1_2_aligned()
21605   std::int32_t* zipped_lhs_3_offsets =  in gemm_i32_2_1_2_aligned()
21606       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 3);  in gemm_i32_2_1_2_aligned()
21607   std::int32_t* zipped_lhs_2_offsets =  in gemm_i32_2_1_2_aligned()
21608       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 2);  in gemm_i32_2_1_2_aligned()
21611   const std::int32_t result_chunk_stride = result_stride * 3;  in gemm_i32_2_1_2_aligned()
21613   const std::int32_t const_offset = lhs_offset * rhs_offset * k;  in gemm_i32_2_1_2_aligned()
21614   std::int32_t* result_chunk = result;  in gemm_i32_2_1_2_aligned()
21615   std::int32_t* mul_result_chunk = result;  in gemm_i32_2_1_2_aligned()
21616   const std::int32_t mul_result_chunk_stride_bytes = result_stride * 4;  in gemm_i32_2_1_2_aligned()
21659                             const std::uint8_t* rhs, std::int32_t m,  in gemm_i32_2_1_3_aligned()
21660                             std::int32_t n, std::int32_t k,  in gemm_i32_2_1_3_aligned()
21661                             std::int32_t lhs_offset, std::int32_t rhs_offset,  in gemm_i32_2_1_3_aligned()
21662                             std::int32_t* result, std::int32_t result_stride) {  in gemm_i32_2_1_3_aligned()
21663   const std::int32_t row_chunks = m / 3;  in gemm_i32_2_1_3_aligned()
21664   const std::int32_t col_chunks = n / 3;  in gemm_i32_2_1_3_aligned()
21665   const std::int32_t padded_k = ((k + 7) / 8) * 8;  in gemm_i32_2_1_3_aligned()
21666   const std::int32_t chunk_size = k * 3;  in gemm_i32_2_1_3_aligned()
21667   const std::int32_t zipped_chunk_size = (padded_k + 16) * 3;  in gemm_i32_2_1_3_aligned()
21668   const std::int32_t zipped_rhs_size = (padded_k + 16) * n;  in gemm_i32_2_1_3_aligned()
21672   std::int32_t* zipped_lhs_3_offsets =  in gemm_i32_2_1_3_aligned()
21673       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 3);  in gemm_i32_2_1_3_aligned()
21674   std::int32_t* zipped_lhs_2_offsets =  in gemm_i32_2_1_3_aligned()
21675       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 2);  in gemm_i32_2_1_3_aligned()
21678   const std::int32_t result_chunk_stride = result_stride * 3;  in gemm_i32_2_1_3_aligned()
21680   const std::int32_t const_offset = lhs_offset * rhs_offset * k;  in gemm_i32_2_1_3_aligned()
21681   std::int32_t* result_chunk = result;  in gemm_i32_2_1_3_aligned()
21682   std::int32_t* mul_result_chunk = result;  in gemm_i32_2_1_3_aligned()
21683   const std::int32_t mul_result_chunk_stride_bytes = result_stride * 4;  in gemm_i32_2_1_3_aligned()
21726                             const std::uint8_t* rhs, std::int32_t m,  in gemm_i32_2_1_4_aligned()
21727                             std::int32_t n, std::int32_t k,  in gemm_i32_2_1_4_aligned()
21728                             std::int32_t lhs_offset, std::int32_t rhs_offset,  in gemm_i32_2_1_4_aligned()
21729                             std::int32_t* result, std::int32_t result_stride) {  in gemm_i32_2_1_4_aligned()
21730   const std::int32_t row_chunks = m / 3;  in gemm_i32_2_1_4_aligned()
21731   const std::int32_t col_chunks = n / 3;  in gemm_i32_2_1_4_aligned()
21732   const std::int32_t padded_k = ((k + 7) / 8) * 8;  in gemm_i32_2_1_4_aligned()
21733   const std::int32_t chunk_size = k * 3;  in gemm_i32_2_1_4_aligned()
21734   const std::int32_t zipped_chunk_size = (padded_k + 16) * 3;  in gemm_i32_2_1_4_aligned()
21735   const std::int32_t zipped_rhs_size = (padded_k + 16) * n;  in gemm_i32_2_1_4_aligned()
21739   std::int32_t* zipped_lhs_3_offsets =  in gemm_i32_2_1_4_aligned()
21740       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 3);  in gemm_i32_2_1_4_aligned()
21741   std::int32_t* zipped_lhs_2_offsets =  in gemm_i32_2_1_4_aligned()
21742       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 2);  in gemm_i32_2_1_4_aligned()
21745   const std::int32_t result_chunk_stride = result_stride * 3;  in gemm_i32_2_1_4_aligned()
21747   const std::int32_t const_offset = lhs_offset * rhs_offset * k;  in gemm_i32_2_1_4_aligned()
21748   std::int32_t* result_chunk = result;  in gemm_i32_2_1_4_aligned()
21749   std::int32_t* mul_result_chunk = result;  in gemm_i32_2_1_4_aligned()
21750   const std::int32_t mul_result_chunk_stride_bytes = result_stride * 4;  in gemm_i32_2_1_4_aligned()
21793                             const std::uint8_t* rhs, std::int32_t m,  in gemm_i32_2_1_5_aligned()
21794                             std::int32_t n, std::int32_t k,  in gemm_i32_2_1_5_aligned()
21795                             std::int32_t lhs_offset, std::int32_t rhs_offset,  in gemm_i32_2_1_5_aligned()
21796                             std::int32_t* result, std::int32_t result_stride) {  in gemm_i32_2_1_5_aligned()
21797   const std::int32_t row_chunks = m / 3;  in gemm_i32_2_1_5_aligned()
21798   const std::int32_t col_chunks = n / 3;  in gemm_i32_2_1_5_aligned()
21799   const std::int32_t padded_k = ((k + 7) / 8) * 8;  in gemm_i32_2_1_5_aligned()
21800   const std::int32_t chunk_size = k * 3;  in gemm_i32_2_1_5_aligned()
21801   const std::int32_t zipped_chunk_size = (padded_k + 16) * 3;  in gemm_i32_2_1_5_aligned()
21802   const std::int32_t zipped_rhs_size = (padded_k + 16) * n;  in gemm_i32_2_1_5_aligned()
21806   std::int32_t* zipped_lhs_3_offsets =  in gemm_i32_2_1_5_aligned()
21807       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 3);  in gemm_i32_2_1_5_aligned()
21808   std::int32_t* zipped_lhs_2_offsets =  in gemm_i32_2_1_5_aligned()
21809       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 2);  in gemm_i32_2_1_5_aligned()
21812   const std::int32_t result_chunk_stride = result_stride * 3;  in gemm_i32_2_1_5_aligned()
21814   const std::int32_t const_offset = lhs_offset * rhs_offset * k;  in gemm_i32_2_1_5_aligned()
21815   std::int32_t* result_chunk = result;  in gemm_i32_2_1_5_aligned()
21816   std::int32_t* mul_result_chunk = result;  in gemm_i32_2_1_5_aligned()
21817   const std::int32_t mul_result_chunk_stride_bytes = result_stride * 4;  in gemm_i32_2_1_5_aligned()
21860                             const std::uint8_t* rhs, std::int32_t m,  in gemm_i32_2_1_6_aligned()
21861                             std::int32_t n, std::int32_t k,  in gemm_i32_2_1_6_aligned()
21862                             std::int32_t lhs_offset, std::int32_t rhs_offset,  in gemm_i32_2_1_6_aligned()
21863                             std::int32_t* result, std::int32_t result_stride) {  in gemm_i32_2_1_6_aligned()
21864   const std::int32_t row_chunks = m / 3;  in gemm_i32_2_1_6_aligned()
21865   const std::int32_t col_chunks = n / 3;  in gemm_i32_2_1_6_aligned()
21866   const std::int32_t padded_k = ((k + 7) / 8) * 8;  in gemm_i32_2_1_6_aligned()
21867   const std::int32_t chunk_size = k * 3;  in gemm_i32_2_1_6_aligned()
21868   const std::int32_t zipped_chunk_size = (padded_k + 16) * 3;  in gemm_i32_2_1_6_aligned()
21869   const std::int32_t zipped_rhs_size = (padded_k + 16) * n;  in gemm_i32_2_1_6_aligned()
21873   std::int32_t* zipped_lhs_3_offsets =  in gemm_i32_2_1_6_aligned()
21874       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 3);  in gemm_i32_2_1_6_aligned()
21875   std::int32_t* zipped_lhs_2_offsets =  in gemm_i32_2_1_6_aligned()
21876       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 2);  in gemm_i32_2_1_6_aligned()
21879   const std::int32_t result_chunk_stride = result_stride * 3;  in gemm_i32_2_1_6_aligned()
21881   const std::int32_t const_offset = lhs_offset * rhs_offset * k;  in gemm_i32_2_1_6_aligned()
21882   std::int32_t* result_chunk = result;  in gemm_i32_2_1_6_aligned()
21883   std::int32_t* mul_result_chunk = result;  in gemm_i32_2_1_6_aligned()
21884   const std::int32_t mul_result_chunk_stride_bytes = result_stride * 4;  in gemm_i32_2_1_6_aligned()
21927                             const std::uint8_t* rhs, std::int32_t m,  in gemm_i32_2_1_7_aligned()
21928                             std::int32_t n, std::int32_t k,  in gemm_i32_2_1_7_aligned()
21929                             std::int32_t lhs_offset, std::int32_t rhs_offset,  in gemm_i32_2_1_7_aligned()
21930                             std::int32_t* result, std::int32_t result_stride) {  in gemm_i32_2_1_7_aligned()
21931   const std::int32_t row_chunks = m / 3;  in gemm_i32_2_1_7_aligned()
21932   const std::int32_t col_chunks = n / 3;  in gemm_i32_2_1_7_aligned()
21933   const std::int32_t padded_k = ((k + 7) / 8) * 8;  in gemm_i32_2_1_7_aligned()
21934   const std::int32_t chunk_size = k * 3;  in gemm_i32_2_1_7_aligned()
21935   const std::int32_t zipped_chunk_size = (padded_k + 16) * 3;  in gemm_i32_2_1_7_aligned()
21936   const std::int32_t zipped_rhs_size = (padded_k + 16) * n;  in gemm_i32_2_1_7_aligned()
21940   std::int32_t* zipped_lhs_3_offsets =  in gemm_i32_2_1_7_aligned()
21941       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 3);  in gemm_i32_2_1_7_aligned()
21942   std::int32_t* zipped_lhs_2_offsets =  in gemm_i32_2_1_7_aligned()
21943       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 2);  in gemm_i32_2_1_7_aligned()
21946   const std::int32_t result_chunk_stride = result_stride * 3;  in gemm_i32_2_1_7_aligned()
21948   const std::int32_t const_offset = lhs_offset * rhs_offset * k;  in gemm_i32_2_1_7_aligned()
21949   std::int32_t* result_chunk = result;  in gemm_i32_2_1_7_aligned()
21950   std::int32_t* mul_result_chunk = result;  in gemm_i32_2_1_7_aligned()
21951   const std::int32_t mul_result_chunk_stride_bytes = result_stride * 4;  in gemm_i32_2_1_7_aligned()
21994                             const std::uint8_t* rhs, std::int32_t m,  in gemm_i32_2_2_0_aligned()
21995                             std::int32_t n, std::int32_t k,  in gemm_i32_2_2_0_aligned()
21996                             std::int32_t lhs_offset, std::int32_t rhs_offset,  in gemm_i32_2_2_0_aligned()
21997                             std::int32_t* result, std::int32_t result_stride) {  in gemm_i32_2_2_0_aligned()
21998   const std::int32_t row_chunks = m / 3;  in gemm_i32_2_2_0_aligned()
21999   const std::int32_t col_chunks = n / 3;  in gemm_i32_2_2_0_aligned()
22000   const std::int32_t padded_k = ((k + 7) / 8) * 8;  in gemm_i32_2_2_0_aligned()
22001   const std::int32_t chunk_size = k * 3;  in gemm_i32_2_2_0_aligned()
22002   const std::int32_t zipped_chunk_size = (padded_k + 16) * 3;  in gemm_i32_2_2_0_aligned()
22003   const std::int32_t zipped_rhs_size = (padded_k + 16) * n;  in gemm_i32_2_2_0_aligned()
22007   std::int32_t* zipped_lhs_3_offsets =  in gemm_i32_2_2_0_aligned()
22008       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 3);  in gemm_i32_2_2_0_aligned()
22009   std::int32_t* zipped_lhs_2_offsets =  in gemm_i32_2_2_0_aligned()
22010       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 2);  in gemm_i32_2_2_0_aligned()
22013   const std::int32_t result_chunk_stride = result_stride * 3;  in gemm_i32_2_2_0_aligned()
22015   const std::int32_t const_offset = lhs_offset * rhs_offset * k;  in gemm_i32_2_2_0_aligned()
22016   std::int32_t* result_chunk = result;  in gemm_i32_2_2_0_aligned()
22017   std::int32_t* mul_result_chunk = result;  in gemm_i32_2_2_0_aligned()
22018   const std::int32_t mul_result_chunk_stride_bytes = result_stride * 4;  in gemm_i32_2_2_0_aligned()
22061                             const std::uint8_t* rhs, std::int32_t m,  in gemm_i32_2_2_1_aligned()
22062                             std::int32_t n, std::int32_t k,  in gemm_i32_2_2_1_aligned()
22063                             std::int32_t lhs_offset, std::int32_t rhs_offset,  in gemm_i32_2_2_1_aligned()
22064                             std::int32_t* result, std::int32_t result_stride) {  in gemm_i32_2_2_1_aligned()
22065   const std::int32_t row_chunks = m / 3;  in gemm_i32_2_2_1_aligned()
22066   const std::int32_t col_chunks = n / 3;  in gemm_i32_2_2_1_aligned()
22067   const std::int32_t padded_k = ((k + 7) / 8) * 8;  in gemm_i32_2_2_1_aligned()
22068   const std::int32_t chunk_size = k * 3;  in gemm_i32_2_2_1_aligned()
22069   const std::int32_t zipped_chunk_size = (padded_k + 16) * 3;  in gemm_i32_2_2_1_aligned()
22070   const std::int32_t zipped_rhs_size = (padded_k + 16) * n;  in gemm_i32_2_2_1_aligned()
22074   std::int32_t* zipped_lhs_3_offsets =  in gemm_i32_2_2_1_aligned()
22075       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 3);  in gemm_i32_2_2_1_aligned()
22076   std::int32_t* zipped_lhs_2_offsets =  in gemm_i32_2_2_1_aligned()
22077       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 2);  in gemm_i32_2_2_1_aligned()
22080   const std::int32_t result_chunk_stride = result_stride * 3;  in gemm_i32_2_2_1_aligned()
22082   const std::int32_t const_offset = lhs_offset * rhs_offset * k;  in gemm_i32_2_2_1_aligned()
22083   std::int32_t* result_chunk = result;  in gemm_i32_2_2_1_aligned()
22084   std::int32_t* mul_result_chunk = result;  in gemm_i32_2_2_1_aligned()
22085   const std::int32_t mul_result_chunk_stride_bytes = result_stride * 4;  in gemm_i32_2_2_1_aligned()
22128                             const std::uint8_t* rhs, std::int32_t m,  in gemm_i32_2_2_2_aligned()
22129                             std::int32_t n, std::int32_t k,  in gemm_i32_2_2_2_aligned()
22130                             std::int32_t lhs_offset, std::int32_t rhs_offset,  in gemm_i32_2_2_2_aligned()
22131                             std::int32_t* result, std::int32_t result_stride) {  in gemm_i32_2_2_2_aligned()
22132   const std::int32_t row_chunks = m / 3;  in gemm_i32_2_2_2_aligned()
22133   const std::int32_t col_chunks = n / 3;  in gemm_i32_2_2_2_aligned()
22134   const std::int32_t padded_k = ((k + 7) / 8) * 8;  in gemm_i32_2_2_2_aligned()
22135   const std::int32_t chunk_size = k * 3;  in gemm_i32_2_2_2_aligned()
22136   const std::int32_t zipped_chunk_size = (padded_k + 16) * 3;  in gemm_i32_2_2_2_aligned()
22137   const std::int32_t zipped_rhs_size = (padded_k + 16) * n;  in gemm_i32_2_2_2_aligned()
22141   std::int32_t* zipped_lhs_3_offsets =  in gemm_i32_2_2_2_aligned()
22142       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 3);  in gemm_i32_2_2_2_aligned()
22143   std::int32_t* zipped_lhs_2_offsets =  in gemm_i32_2_2_2_aligned()
22144       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 2);  in gemm_i32_2_2_2_aligned()
22147   const std::int32_t result_chunk_stride = result_stride * 3;  in gemm_i32_2_2_2_aligned()
22149   const std::int32_t const_offset = lhs_offset * rhs_offset * k;  in gemm_i32_2_2_2_aligned()
22150   std::int32_t* result_chunk = result;  in gemm_i32_2_2_2_aligned()
22151   std::int32_t* mul_result_chunk = result;  in gemm_i32_2_2_2_aligned()
22152   const std::int32_t mul_result_chunk_stride_bytes = result_stride * 4;  in gemm_i32_2_2_2_aligned()
22195                             const std::uint8_t* rhs, std::int32_t m,  in gemm_i32_2_2_3_aligned()
22196                             std::int32_t n, std::int32_t k,  in gemm_i32_2_2_3_aligned()
22197                             std::int32_t lhs_offset, std::int32_t rhs_offset,  in gemm_i32_2_2_3_aligned()
22198                             std::int32_t* result, std::int32_t result_stride) {  in gemm_i32_2_2_3_aligned()
22199   const std::int32_t row_chunks = m / 3;  in gemm_i32_2_2_3_aligned()
22200   const std::int32_t col_chunks = n / 3;  in gemm_i32_2_2_3_aligned()
22201   const std::int32_t padded_k = ((k + 7) / 8) * 8;  in gemm_i32_2_2_3_aligned()
22202   const std::int32_t chunk_size = k * 3;  in gemm_i32_2_2_3_aligned()
22203   const std::int32_t zipped_chunk_size = (padded_k + 16) * 3;  in gemm_i32_2_2_3_aligned()
22204   const std::int32_t zipped_rhs_size = (padded_k + 16) * n;  in gemm_i32_2_2_3_aligned()
22208   std::int32_t* zipped_lhs_3_offsets =  in gemm_i32_2_2_3_aligned()
22209       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 3);  in gemm_i32_2_2_3_aligned()
22210   std::int32_t* zipped_lhs_2_offsets =  in gemm_i32_2_2_3_aligned()
22211       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 2);  in gemm_i32_2_2_3_aligned()
22214   const std::int32_t result_chunk_stride = result_stride * 3;  in gemm_i32_2_2_3_aligned()
22216   const std::int32_t const_offset = lhs_offset * rhs_offset * k;  in gemm_i32_2_2_3_aligned()
22217   std::int32_t* result_chunk = result;  in gemm_i32_2_2_3_aligned()
22218   std::int32_t* mul_result_chunk = result;  in gemm_i32_2_2_3_aligned()
22219   const std::int32_t mul_result_chunk_stride_bytes = result_stride * 4;  in gemm_i32_2_2_3_aligned()
22262                             const std::uint8_t* rhs, std::int32_t m,  in gemm_i32_2_2_4_aligned()
22263                             std::int32_t n, std::int32_t k,  in gemm_i32_2_2_4_aligned()
22264                             std::int32_t lhs_offset, std::int32_t rhs_offset,  in gemm_i32_2_2_4_aligned()
22265                             std::int32_t* result, std::int32_t result_stride) {  in gemm_i32_2_2_4_aligned()
22266   const std::int32_t row_chunks = m / 3;  in gemm_i32_2_2_4_aligned()
22267   const std::int32_t col_chunks = n / 3;  in gemm_i32_2_2_4_aligned()
22268   const std::int32_t padded_k = ((k + 7) / 8) * 8;  in gemm_i32_2_2_4_aligned()
22269   const std::int32_t chunk_size = k * 3;  in gemm_i32_2_2_4_aligned()
22270   const std::int32_t zipped_chunk_size = (padded_k + 16) * 3;  in gemm_i32_2_2_4_aligned()
22271   const std::int32_t zipped_rhs_size = (padded_k + 16) * n;  in gemm_i32_2_2_4_aligned()
22275   std::int32_t* zipped_lhs_3_offsets =  in gemm_i32_2_2_4_aligned()
22276       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 3);  in gemm_i32_2_2_4_aligned()
22277   std::int32_t* zipped_lhs_2_offsets =  in gemm_i32_2_2_4_aligned()
22278       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 2);  in gemm_i32_2_2_4_aligned()
22281   const std::int32_t result_chunk_stride = result_stride * 3;  in gemm_i32_2_2_4_aligned()
22283   const std::int32_t const_offset = lhs_offset * rhs_offset * k;  in gemm_i32_2_2_4_aligned()
22284   std::int32_t* result_chunk = result;  in gemm_i32_2_2_4_aligned()
22285   std::int32_t* mul_result_chunk = result;  in gemm_i32_2_2_4_aligned()
22286   const std::int32_t mul_result_chunk_stride_bytes = result_stride * 4;  in gemm_i32_2_2_4_aligned()
22329                             const std::uint8_t* rhs, std::int32_t m,  in gemm_i32_2_2_5_aligned()
22330                             std::int32_t n, std::int32_t k,  in gemm_i32_2_2_5_aligned()
22331                             std::int32_t lhs_offset, std::int32_t rhs_offset,  in gemm_i32_2_2_5_aligned()
22332                             std::int32_t* result, std::int32_t result_stride) {  in gemm_i32_2_2_5_aligned()
22333   const std::int32_t row_chunks = m / 3;  in gemm_i32_2_2_5_aligned()
22334   const std::int32_t col_chunks = n / 3;  in gemm_i32_2_2_5_aligned()
22335   const std::int32_t padded_k = ((k + 7) / 8) * 8;  in gemm_i32_2_2_5_aligned()
22336   const std::int32_t chunk_size = k * 3;  in gemm_i32_2_2_5_aligned()
22337   const std::int32_t zipped_chunk_size = (padded_k + 16) * 3;  in gemm_i32_2_2_5_aligned()
22338   const std::int32_t zipped_rhs_size = (padded_k + 16) * n;  in gemm_i32_2_2_5_aligned()
22342   std::int32_t* zipped_lhs_3_offsets =  in gemm_i32_2_2_5_aligned()
22343       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 3);  in gemm_i32_2_2_5_aligned()
22344   std::int32_t* zipped_lhs_2_offsets =  in gemm_i32_2_2_5_aligned()
22345       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 2);  in gemm_i32_2_2_5_aligned()
22348   const std::int32_t result_chunk_stride = result_stride * 3;  in gemm_i32_2_2_5_aligned()
22350   const std::int32_t const_offset = lhs_offset * rhs_offset * k;  in gemm_i32_2_2_5_aligned()
22351   std::int32_t* result_chunk = result;  in gemm_i32_2_2_5_aligned()
22352   std::int32_t* mul_result_chunk = result;  in gemm_i32_2_2_5_aligned()
22353   const std::int32_t mul_result_chunk_stride_bytes = result_stride * 4;  in gemm_i32_2_2_5_aligned()
22396                             const std::uint8_t* rhs, std::int32_t m,  in gemm_i32_2_2_6_aligned()
22397                             std::int32_t n, std::int32_t k,  in gemm_i32_2_2_6_aligned()
22398                             std::int32_t lhs_offset, std::int32_t rhs_offset,  in gemm_i32_2_2_6_aligned()
22399                             std::int32_t* result, std::int32_t result_stride) {  in gemm_i32_2_2_6_aligned()
22400   const std::int32_t row_chunks = m / 3;  in gemm_i32_2_2_6_aligned()
22401   const std::int32_t col_chunks = n / 3;  in gemm_i32_2_2_6_aligned()
22402   const std::int32_t padded_k = ((k + 7) / 8) * 8;  in gemm_i32_2_2_6_aligned()
22403   const std::int32_t chunk_size = k * 3;  in gemm_i32_2_2_6_aligned()
22404   const std::int32_t zipped_chunk_size = (padded_k + 16) * 3;  in gemm_i32_2_2_6_aligned()
22405   const std::int32_t zipped_rhs_size = (padded_k + 16) * n;  in gemm_i32_2_2_6_aligned()
22409   std::int32_t* zipped_lhs_3_offsets =  in gemm_i32_2_2_6_aligned()
22410       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 3);  in gemm_i32_2_2_6_aligned()
22411   std::int32_t* zipped_lhs_2_offsets =  in gemm_i32_2_2_6_aligned()
22412       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 2);  in gemm_i32_2_2_6_aligned()
22415   const std::int32_t result_chunk_stride = result_stride * 3;  in gemm_i32_2_2_6_aligned()
22417   const std::int32_t const_offset = lhs_offset * rhs_offset * k;  in gemm_i32_2_2_6_aligned()
22418   std::int32_t* result_chunk = result;  in gemm_i32_2_2_6_aligned()
22419   std::int32_t* mul_result_chunk = result;  in gemm_i32_2_2_6_aligned()
22420   const std::int32_t mul_result_chunk_stride_bytes = result_stride * 4;  in gemm_i32_2_2_6_aligned()
22463                             const std::uint8_t* rhs, std::int32_t m,  in gemm_i32_2_2_7_aligned()
22464                             std::int32_t n, std::int32_t k,  in gemm_i32_2_2_7_aligned()
22465                             std::int32_t lhs_offset, std::int32_t rhs_offset,  in gemm_i32_2_2_7_aligned()
22466                             std::int32_t* result, std::int32_t result_stride) {  in gemm_i32_2_2_7_aligned()
22467   const std::int32_t row_chunks = m / 3;  in gemm_i32_2_2_7_aligned()
22468   const std::int32_t col_chunks = n / 3;  in gemm_i32_2_2_7_aligned()
22469   const std::int32_t padded_k = ((k + 7) / 8) * 8;  in gemm_i32_2_2_7_aligned()
22470   const std::int32_t chunk_size = k * 3;  in gemm_i32_2_2_7_aligned()
22471   const std::int32_t zipped_chunk_size = (padded_k + 16) * 3;  in gemm_i32_2_2_7_aligned()
22472   const std::int32_t zipped_rhs_size = (padded_k + 16) * n;  in gemm_i32_2_2_7_aligned()
22476   std::int32_t* zipped_lhs_3_offsets =  in gemm_i32_2_2_7_aligned()
22477       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 3);  in gemm_i32_2_2_7_aligned()
22478   std::int32_t* zipped_lhs_2_offsets =  in gemm_i32_2_2_7_aligned()
22479       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 2);  in gemm_i32_2_2_7_aligned()
22482   const std::int32_t result_chunk_stride = result_stride * 3;  in gemm_i32_2_2_7_aligned()
22484   const std::int32_t const_offset = lhs_offset * rhs_offset * k;  in gemm_i32_2_2_7_aligned()
22485   std::int32_t* result_chunk = result;  in gemm_i32_2_2_7_aligned()
22486   std::int32_t* mul_result_chunk = result;  in gemm_i32_2_2_7_aligned()
22487   const std::int32_t mul_result_chunk_stride_bytes = result_stride * 4;  in gemm_i32_2_2_7_aligned()
22530                     const std::uint8_t* rhs, std::int32_t m, std::int32_t n,  in gemm_i32_0_0_0()
22531                     std::int32_t k, std::int32_t lhs_offset,  in gemm_i32_0_0_0()
22532                     std::int32_t rhs_offset, std::int32_t* result,  in gemm_i32_0_0_0()
22533                     std::int32_t result_stride) {  in gemm_i32_0_0_0()
22534   const std::int32_t row_chunks = m / 3;  in gemm_i32_0_0_0()
22535   const std::int32_t col_chunks = n / 3;  in gemm_i32_0_0_0()
22536   const std::int32_t padded_k = ((k + 7) / 8) * 8;  in gemm_i32_0_0_0()
22537   const std::int32_t chunk_size = k * 3;  in gemm_i32_0_0_0()
22538   const std::int32_t zipped_chunk_size = (padded_k + 16) * 3;  in gemm_i32_0_0_0()
22539   const std::int32_t zipped_rhs_size = (padded_k + 16) * n;  in gemm_i32_0_0_0()
22543   std::int32_t* zipped_lhs_3_offsets =  in gemm_i32_0_0_0()
22544       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 3);  in gemm_i32_0_0_0()
22547   const std::int32_t result_chunk_stride = result_stride * 3;  in gemm_i32_0_0_0()
22549   const std::int32_t const_offset = lhs_offset * rhs_offset * k;  in gemm_i32_0_0_0()
22550   std::int32_t* result_chunk = result;  in gemm_i32_0_0_0()
22551   std::int32_t* mul_result_chunk = result;  in gemm_i32_0_0_0()
22552   const std::int32_t mul_result_chunk_stride_bytes = result_stride * 4;  in gemm_i32_0_0_0()
22577                     const std::uint8_t* rhs, std::int32_t m, std::int32_t n,  in gemm_i32_0_0_1()
22578                     std::int32_t k, std::int32_t lhs_offset,  in gemm_i32_0_0_1()
22579                     std::int32_t rhs_offset, std::int32_t* result,  in gemm_i32_0_0_1()
22580                     std::int32_t result_stride) {  in gemm_i32_0_0_1()
22581   const std::int32_t row_chunks = m / 3;  in gemm_i32_0_0_1()
22582   const std::int32_t col_chunks = n / 3;  in gemm_i32_0_0_1()
22583   const std::int32_t padded_k = ((k + 7) / 8) * 8;  in gemm_i32_0_0_1()
22584   const std::int32_t chunk_size = k * 3;  in gemm_i32_0_0_1()
22585   const std::int32_t zipped_chunk_size = (padded_k + 16) * 3;  in gemm_i32_0_0_1()
22586   const std::int32_t zipped_rhs_size = (padded_k + 16) * n;  in gemm_i32_0_0_1()
22590   std::int32_t* zipped_lhs_3_offsets =  in gemm_i32_0_0_1()
22591       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 3);  in gemm_i32_0_0_1()
22594   const std::int32_t result_chunk_stride = result_stride * 3;  in gemm_i32_0_0_1()
22596   const std::int32_t const_offset = lhs_offset * rhs_offset * k;  in gemm_i32_0_0_1()
22597   std::int32_t* result_chunk = result;  in gemm_i32_0_0_1()
22598   std::int32_t* mul_result_chunk = result;  in gemm_i32_0_0_1()
22599   const std::int32_t mul_result_chunk_stride_bytes = result_stride * 4;  in gemm_i32_0_0_1()
22624                     const std::uint8_t* rhs, std::int32_t m, std::int32_t n,  in gemm_i32_0_0_2()
22625                     std::int32_t k, std::int32_t lhs_offset,  in gemm_i32_0_0_2()
22626                     std::int32_t rhs_offset, std::int32_t* result,  in gemm_i32_0_0_2()
22627                     std::int32_t result_stride) {  in gemm_i32_0_0_2()
22628   const std::int32_t row_chunks = m / 3;  in gemm_i32_0_0_2()
22629   const std::int32_t col_chunks = n / 3;  in gemm_i32_0_0_2()
22630   const std::int32_t padded_k = ((k + 7) / 8) * 8;  in gemm_i32_0_0_2()
22631   const std::int32_t chunk_size = k * 3;  in gemm_i32_0_0_2()
22632   const std::int32_t zipped_chunk_size = (padded_k + 16) * 3;  in gemm_i32_0_0_2()
22633   const std::int32_t zipped_rhs_size = (padded_k + 16) * n;  in gemm_i32_0_0_2()
22637   std::int32_t* zipped_lhs_3_offsets =  in gemm_i32_0_0_2()
22638       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 3);  in gemm_i32_0_0_2()
22641   const std::int32_t result_chunk_stride = result_stride * 3;  in gemm_i32_0_0_2()
22643   const std::int32_t const_offset = lhs_offset * rhs_offset * k;  in gemm_i32_0_0_2()
22644   std::int32_t* result_chunk = result;  in gemm_i32_0_0_2()
22645   std::int32_t* mul_result_chunk = result;  in gemm_i32_0_0_2()
22646   const std::int32_t mul_result_chunk_stride_bytes = result_stride * 4;  in gemm_i32_0_0_2()
22671                     const std::uint8_t* rhs, std::int32_t m, std::int32_t n,  in gemm_i32_0_0_3()
22672                     std::int32_t k, std::int32_t lhs_offset,  in gemm_i32_0_0_3()
22673                     std::int32_t rhs_offset, std::int32_t* result,  in gemm_i32_0_0_3()
22674                     std::int32_t result_stride) {  in gemm_i32_0_0_3()
22675   const std::int32_t row_chunks = m / 3;  in gemm_i32_0_0_3()
22676   const std::int32_t col_chunks = n / 3;  in gemm_i32_0_0_3()
22677   const std::int32_t padded_k = ((k + 7) / 8) * 8;  in gemm_i32_0_0_3()
22678   const std::int32_t chunk_size = k * 3;  in gemm_i32_0_0_3()
22679   const std::int32_t zipped_chunk_size = (padded_k + 16) * 3;  in gemm_i32_0_0_3()
22680   const std::int32_t zipped_rhs_size = (padded_k + 16) * n;  in gemm_i32_0_0_3()
22684   std::int32_t* zipped_lhs_3_offsets =  in gemm_i32_0_0_3()
22685       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 3);  in gemm_i32_0_0_3()
22688   const std::int32_t result_chunk_stride = result_stride * 3;  in gemm_i32_0_0_3()
22690   const std::int32_t const_offset = lhs_offset * rhs_offset * k;  in gemm_i32_0_0_3()
22691   std::int32_t* result_chunk = result;  in gemm_i32_0_0_3()
22692   std::int32_t* mul_result_chunk = result;  in gemm_i32_0_0_3()
22693   const std::int32_t mul_result_chunk_stride_bytes = result_stride * 4;  in gemm_i32_0_0_3()
22718                     const std::uint8_t* rhs, std::int32_t m, std::int32_t n,  in gemm_i32_0_0_4()
22719                     std::int32_t k, std::int32_t lhs_offset,  in gemm_i32_0_0_4()
22720                     std::int32_t rhs_offset, std::int32_t* result,  in gemm_i32_0_0_4()
22721                     std::int32_t result_stride) {  in gemm_i32_0_0_4()
22722   const std::int32_t row_chunks = m / 3;  in gemm_i32_0_0_4()
22723   const std::int32_t col_chunks = n / 3;  in gemm_i32_0_0_4()
22724   const std::int32_t padded_k = ((k + 7) / 8) * 8;  in gemm_i32_0_0_4()
22725   const std::int32_t chunk_size = k * 3;  in gemm_i32_0_0_4()
22726   const std::int32_t zipped_chunk_size = (padded_k + 16) * 3;  in gemm_i32_0_0_4()
22727   const std::int32_t zipped_rhs_size = (padded_k + 16) * n;  in gemm_i32_0_0_4()
22731   std::int32_t* zipped_lhs_3_offsets =  in gemm_i32_0_0_4()
22732       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 3);  in gemm_i32_0_0_4()
22735   const std::int32_t result_chunk_stride = result_stride * 3;  in gemm_i32_0_0_4()
22737   const std::int32_t const_offset = lhs_offset * rhs_offset * k;  in gemm_i32_0_0_4()
22738   std::int32_t* result_chunk = result;  in gemm_i32_0_0_4()
22739   std::int32_t* mul_result_chunk = result;  in gemm_i32_0_0_4()
22740   const std::int32_t mul_result_chunk_stride_bytes = result_stride * 4;  in gemm_i32_0_0_4()
22765                     const std::uint8_t* rhs, std::int32_t m, std::int32_t n,  in gemm_i32_0_0_5()
22766                     std::int32_t k, std::int32_t lhs_offset,  in gemm_i32_0_0_5()
22767                     std::int32_t rhs_offset, std::int32_t* result,  in gemm_i32_0_0_5()
22768                     std::int32_t result_stride) {  in gemm_i32_0_0_5()
22769   const std::int32_t row_chunks = m / 3;  in gemm_i32_0_0_5()
22770   const std::int32_t col_chunks = n / 3;  in gemm_i32_0_0_5()
22771   const std::int32_t padded_k = ((k + 7) / 8) * 8;  in gemm_i32_0_0_5()
22772   const std::int32_t chunk_size = k * 3;  in gemm_i32_0_0_5()
22773   const std::int32_t zipped_chunk_size = (padded_k + 16) * 3;  in gemm_i32_0_0_5()
22774   const std::int32_t zipped_rhs_size = (padded_k + 16) * n;  in gemm_i32_0_0_5()
22778   std::int32_t* zipped_lhs_3_offsets =  in gemm_i32_0_0_5()
22779       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 3);  in gemm_i32_0_0_5()
22782   const std::int32_t result_chunk_stride = result_stride * 3;  in gemm_i32_0_0_5()
22784   const std::int32_t const_offset = lhs_offset * rhs_offset * k;  in gemm_i32_0_0_5()
22785   std::int32_t* result_chunk = result;  in gemm_i32_0_0_5()
22786   std::int32_t* mul_result_chunk = result;  in gemm_i32_0_0_5()
22787   const std::int32_t mul_result_chunk_stride_bytes = result_stride * 4;  in gemm_i32_0_0_5()
22812                     const std::uint8_t* rhs, std::int32_t m, std::int32_t n,  in gemm_i32_0_0_6()
22813                     std::int32_t k, std::int32_t lhs_offset,  in gemm_i32_0_0_6()
22814                     std::int32_t rhs_offset, std::int32_t* result,  in gemm_i32_0_0_6()
22815                     std::int32_t result_stride) {  in gemm_i32_0_0_6()
22816   const std::int32_t row_chunks = m / 3;  in gemm_i32_0_0_6()
22817   const std::int32_t col_chunks = n / 3;  in gemm_i32_0_0_6()
22818   const std::int32_t padded_k = ((k + 7) / 8) * 8;  in gemm_i32_0_0_6()
22819   const std::int32_t chunk_size = k * 3;  in gemm_i32_0_0_6()
22820   const std::int32_t zipped_chunk_size = (padded_k + 16) * 3;  in gemm_i32_0_0_6()
22821   const std::int32_t zipped_rhs_size = (padded_k + 16) * n;  in gemm_i32_0_0_6()
22825   std::int32_t* zipped_lhs_3_offsets =  in gemm_i32_0_0_6()
22826       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 3);  in gemm_i32_0_0_6()
22829   const std::int32_t result_chunk_stride = result_stride * 3;  in gemm_i32_0_0_6()
22831   const std::int32_t const_offset = lhs_offset * rhs_offset * k;  in gemm_i32_0_0_6()
22832   std::int32_t* result_chunk = result;  in gemm_i32_0_0_6()
22833   std::int32_t* mul_result_chunk = result;  in gemm_i32_0_0_6()
22834   const std::int32_t mul_result_chunk_stride_bytes = result_stride * 4;  in gemm_i32_0_0_6()
22859                     const std::uint8_t* rhs, std::int32_t m, std::int32_t n,  in gemm_i32_0_0_7()
22860                     std::int32_t k, std::int32_t lhs_offset,  in gemm_i32_0_0_7()
22861                     std::int32_t rhs_offset, std::int32_t* result,  in gemm_i32_0_0_7()
22862                     std::int32_t result_stride) {  in gemm_i32_0_0_7()
22863   const std::int32_t row_chunks = m / 3;  in gemm_i32_0_0_7()
22864   const std::int32_t col_chunks = n / 3;  in gemm_i32_0_0_7()
22865   const std::int32_t padded_k = ((k + 7) / 8) * 8;  in gemm_i32_0_0_7()
22866   const std::int32_t chunk_size = k * 3;  in gemm_i32_0_0_7()
22867   const std::int32_t zipped_chunk_size = (padded_k + 16) * 3;  in gemm_i32_0_0_7()
22868   const std::int32_t zipped_rhs_size = (padded_k + 16) * n;  in gemm_i32_0_0_7()
22872   std::int32_t* zipped_lhs_3_offsets =  in gemm_i32_0_0_7()
22873       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 3);  in gemm_i32_0_0_7()
22876   const std::int32_t result_chunk_stride = result_stride * 3;  in gemm_i32_0_0_7()
22878   const std::int32_t const_offset = lhs_offset * rhs_offset * k;  in gemm_i32_0_0_7()
22879   std::int32_t* result_chunk = result;  in gemm_i32_0_0_7()
22880   std::int32_t* mul_result_chunk = result;  in gemm_i32_0_0_7()
22881   const std::int32_t mul_result_chunk_stride_bytes = result_stride * 4;  in gemm_i32_0_0_7()
22906                     const std::uint8_t* rhs, std::int32_t m, std::int32_t n,  in gemm_i32_0_1_0()
22907                     std::int32_t k, std::int32_t lhs_offset,  in gemm_i32_0_1_0()
22908                     std::int32_t rhs_offset, std::int32_t* result,  in gemm_i32_0_1_0()
22909                     std::int32_t result_stride) {  in gemm_i32_0_1_0()
22910   const std::int32_t row_chunks = m / 3;  in gemm_i32_0_1_0()
22911   const std::int32_t col_chunks = n / 3;  in gemm_i32_0_1_0()
22912   const std::int32_t padded_k = ((k + 7) / 8) * 8;  in gemm_i32_0_1_0()
22913   const std::int32_t chunk_size = k * 3;  in gemm_i32_0_1_0()
22914   const std::int32_t zipped_chunk_size = (padded_k + 16) * 3;  in gemm_i32_0_1_0()
22915   const std::int32_t zipped_rhs_size = (padded_k + 16) * n;  in gemm_i32_0_1_0()
22919   std::int32_t* zipped_lhs_3_offsets =  in gemm_i32_0_1_0()
22920       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 3);  in gemm_i32_0_1_0()
22923   const std::int32_t result_chunk_stride = result_stride * 3;  in gemm_i32_0_1_0()
22925   const std::int32_t const_offset = lhs_offset * rhs_offset * k;  in gemm_i32_0_1_0()
22926   std::int32_t* result_chunk = result;  in gemm_i32_0_1_0()
22927   std::int32_t* mul_result_chunk = result;  in gemm_i32_0_1_0()
22928   const std::int32_t mul_result_chunk_stride_bytes = result_stride * 4;  in gemm_i32_0_1_0()
22957                     const std::uint8_t* rhs, std::int32_t m, std::int32_t n,  in gemm_i32_0_1_1()
22958                     std::int32_t k, std::int32_t lhs_offset,  in gemm_i32_0_1_1()
22959                     std::int32_t rhs_offset, std::int32_t* result,  in gemm_i32_0_1_1()
22960                     std::int32_t result_stride) {  in gemm_i32_0_1_1()
22961   const std::int32_t row_chunks = m / 3;  in gemm_i32_0_1_1()
22962   const std::int32_t col_chunks = n / 3;  in gemm_i32_0_1_1()
22963   const std::int32_t padded_k = ((k + 7) / 8) * 8;  in gemm_i32_0_1_1()
22964   const std::int32_t chunk_size = k * 3;  in gemm_i32_0_1_1()
22965   const std::int32_t zipped_chunk_size = (padded_k + 16) * 3;  in gemm_i32_0_1_1()
22966   const std::int32_t zipped_rhs_size = (padded_k + 16) * n;  in gemm_i32_0_1_1()
22970   std::int32_t* zipped_lhs_3_offsets =  in gemm_i32_0_1_1()
22971       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 3);  in gemm_i32_0_1_1()
22974   const std::int32_t result_chunk_stride = result_stride * 3;  in gemm_i32_0_1_1()
22976   const std::int32_t const_offset = lhs_offset * rhs_offset * k;  in gemm_i32_0_1_1()
22977   std::int32_t* result_chunk = result;  in gemm_i32_0_1_1()
22978   std::int32_t* mul_result_chunk = result;  in gemm_i32_0_1_1()
22979   const std::int32_t mul_result_chunk_stride_bytes = result_stride * 4;  in gemm_i32_0_1_1()
23008                     const std::uint8_t* rhs, std::int32_t m, std::int32_t n,  in gemm_i32_0_1_2()
23009                     std::int32_t k, std::int32_t lhs_offset,  in gemm_i32_0_1_2()
23010                     std::int32_t rhs_offset, std::int32_t* result,  in gemm_i32_0_1_2()
23011                     std::int32_t result_stride) {  in gemm_i32_0_1_2()
23012   const std::int32_t row_chunks = m / 3;  in gemm_i32_0_1_2()
23013   const std::int32_t col_chunks = n / 3;  in gemm_i32_0_1_2()
23014   const std::int32_t padded_k = ((k + 7) / 8) * 8;  in gemm_i32_0_1_2()
23015   const std::int32_t chunk_size = k * 3;  in gemm_i32_0_1_2()
23016   const std::int32_t zipped_chunk_size = (padded_k + 16) * 3;  in gemm_i32_0_1_2()
23017   const std::int32_t zipped_rhs_size = (padded_k + 16) * n;  in gemm_i32_0_1_2()
23021   std::int32_t* zipped_lhs_3_offsets =  in gemm_i32_0_1_2()
23022       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 3);  in gemm_i32_0_1_2()
23025   const std::int32_t result_chunk_stride = result_stride * 3;  in gemm_i32_0_1_2()
23027   const std::int32_t const_offset = lhs_offset * rhs_offset * k;  in gemm_i32_0_1_2()
23028   std::int32_t* result_chunk = result;  in gemm_i32_0_1_2()
23029   std::int32_t* mul_result_chunk = result;  in gemm_i32_0_1_2()
23030   const std::int32_t mul_result_chunk_stride_bytes = result_stride * 4;  in gemm_i32_0_1_2()
23059                     const std::uint8_t* rhs, std::int32_t m, std::int32_t n,  in gemm_i32_0_1_3()
23060                     std::int32_t k, std::int32_t lhs_offset,  in gemm_i32_0_1_3()
23061                     std::int32_t rhs_offset, std::int32_t* result,  in gemm_i32_0_1_3()
23062                     std::int32_t result_stride) {  in gemm_i32_0_1_3()
23063   const std::int32_t row_chunks = m / 3;  in gemm_i32_0_1_3()
23064   const std::int32_t col_chunks = n / 3;  in gemm_i32_0_1_3()
23065   const std::int32_t padded_k = ((k + 7) / 8) * 8;  in gemm_i32_0_1_3()
23066   const std::int32_t chunk_size = k * 3;  in gemm_i32_0_1_3()
23067   const std::int32_t zipped_chunk_size = (padded_k + 16) * 3;  in gemm_i32_0_1_3()
23068   const std::int32_t zipped_rhs_size = (padded_k + 16) * n;  in gemm_i32_0_1_3()
23072   std::int32_t* zipped_lhs_3_offsets =  in gemm_i32_0_1_3()
23073       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 3);  in gemm_i32_0_1_3()
23076   const std::int32_t result_chunk_stride = result_stride * 3;  in gemm_i32_0_1_3()
23078   const std::int32_t const_offset = lhs_offset * rhs_offset * k;  in gemm_i32_0_1_3()
23079   std::int32_t* result_chunk = result;  in gemm_i32_0_1_3()
23080   std::int32_t* mul_result_chunk = result;  in gemm_i32_0_1_3()
23081   const std::int32_t mul_result_chunk_stride_bytes = result_stride * 4;  in gemm_i32_0_1_3()
23110                     const std::uint8_t* rhs, std::int32_t m, std::int32_t n,  in gemm_i32_0_1_4()
23111                     std::int32_t k, std::int32_t lhs_offset,  in gemm_i32_0_1_4()
23112                     std::int32_t rhs_offset, std::int32_t* result,  in gemm_i32_0_1_4()
23113                     std::int32_t result_stride) {  in gemm_i32_0_1_4()
23114   const std::int32_t row_chunks = m / 3;  in gemm_i32_0_1_4()
23115   const std::int32_t col_chunks = n / 3;  in gemm_i32_0_1_4()
23116   const std::int32_t padded_k = ((k + 7) / 8) * 8;  in gemm_i32_0_1_4()
23117   const std::int32_t chunk_size = k * 3;  in gemm_i32_0_1_4()
23118   const std::int32_t zipped_chunk_size = (padded_k + 16) * 3;  in gemm_i32_0_1_4()
23119   const std::int32_t zipped_rhs_size = (padded_k + 16) * n;  in gemm_i32_0_1_4()
23123   std::int32_t* zipped_lhs_3_offsets =  in gemm_i32_0_1_4()
23124       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 3);  in gemm_i32_0_1_4()
23127   const std::int32_t result_chunk_stride = result_stride * 3;  in gemm_i32_0_1_4()
23129   const std::int32_t const_offset = lhs_offset * rhs_offset * k;  in gemm_i32_0_1_4()
23130   std::int32_t* result_chunk = result;  in gemm_i32_0_1_4()
23131   std::int32_t* mul_result_chunk = result;  in gemm_i32_0_1_4()
23132   const std::int32_t mul_result_chunk_stride_bytes = result_stride * 4;  in gemm_i32_0_1_4()
23161                     const std::uint8_t* rhs, std::int32_t m, std::int32_t n,  in gemm_i32_0_1_5()
23162                     std::int32_t k, std::int32_t lhs_offset,  in gemm_i32_0_1_5()
23163                     std::int32_t rhs_offset, std::int32_t* result,  in gemm_i32_0_1_5()
23164                     std::int32_t result_stride) {  in gemm_i32_0_1_5()
23165   const std::int32_t row_chunks = m / 3;  in gemm_i32_0_1_5()
23166   const std::int32_t col_chunks = n / 3;  in gemm_i32_0_1_5()
23167   const std::int32_t padded_k = ((k + 7) / 8) * 8;  in gemm_i32_0_1_5()
23168   const std::int32_t chunk_size = k * 3;  in gemm_i32_0_1_5()
23169   const std::int32_t zipped_chunk_size = (padded_k + 16) * 3;  in gemm_i32_0_1_5()
23170   const std::int32_t zipped_rhs_size = (padded_k + 16) * n;  in gemm_i32_0_1_5()
23174   std::int32_t* zipped_lhs_3_offsets =  in gemm_i32_0_1_5()
23175       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 3);  in gemm_i32_0_1_5()
23178   const std::int32_t result_chunk_stride = result_stride * 3;  in gemm_i32_0_1_5()
23180   const std::int32_t const_offset = lhs_offset * rhs_offset * k;  in gemm_i32_0_1_5()
23181   std::int32_t* result_chunk = result;  in gemm_i32_0_1_5()
23182   std::int32_t* mul_result_chunk = result;  in gemm_i32_0_1_5()
23183   const std::int32_t mul_result_chunk_stride_bytes = result_stride * 4;  in gemm_i32_0_1_5()
23212                     const std::uint8_t* rhs, std::int32_t m, std::int32_t n,  in gemm_i32_0_1_6()
23213                     std::int32_t k, std::int32_t lhs_offset,  in gemm_i32_0_1_6()
23214                     std::int32_t rhs_offset, std::int32_t* result,  in gemm_i32_0_1_6()
23215                     std::int32_t result_stride) {  in gemm_i32_0_1_6()
23216   const std::int32_t row_chunks = m / 3;  in gemm_i32_0_1_6()
23217   const std::int32_t col_chunks = n / 3;  in gemm_i32_0_1_6()
23218   const std::int32_t padded_k = ((k + 7) / 8) * 8;  in gemm_i32_0_1_6()
23219   const std::int32_t chunk_size = k * 3;  in gemm_i32_0_1_6()
23220   const std::int32_t zipped_chunk_size = (padded_k + 16) * 3;  in gemm_i32_0_1_6()
23221   const std::int32_t zipped_rhs_size = (padded_k + 16) * n;  in gemm_i32_0_1_6()
23225   std::int32_t* zipped_lhs_3_offsets =  in gemm_i32_0_1_6()
23226       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 3);  in gemm_i32_0_1_6()
23229   const std::int32_t result_chunk_stride = result_stride * 3;  in gemm_i32_0_1_6()
23231   const std::int32_t const_offset = lhs_offset * rhs_offset * k;  in gemm_i32_0_1_6()
23232   std::int32_t* result_chunk = result;  in gemm_i32_0_1_6()
23233   std::int32_t* mul_result_chunk = result;  in gemm_i32_0_1_6()
23234   const std::int32_t mul_result_chunk_stride_bytes = result_stride * 4;  in gemm_i32_0_1_6()
23263                     const std::uint8_t* rhs, std::int32_t m, std::int32_t n,  in gemm_i32_0_1_7()
23264                     std::int32_t k, std::int32_t lhs_offset,  in gemm_i32_0_1_7()
23265                     std::int32_t rhs_offset, std::int32_t* result,  in gemm_i32_0_1_7()
23266                     std::int32_t result_stride) {  in gemm_i32_0_1_7()
23267   const std::int32_t row_chunks = m / 3;  in gemm_i32_0_1_7()
23268   const std::int32_t col_chunks = n / 3;  in gemm_i32_0_1_7()
23269   const std::int32_t padded_k = ((k + 7) / 8) * 8;  in gemm_i32_0_1_7()
23270   const std::int32_t chunk_size = k * 3;  in gemm_i32_0_1_7()
23271   const std::int32_t zipped_chunk_size = (padded_k + 16) * 3;  in gemm_i32_0_1_7()
23272   const std::int32_t zipped_rhs_size = (padded_k + 16) * n;  in gemm_i32_0_1_7()
23276   std::int32_t* zipped_lhs_3_offsets =  in gemm_i32_0_1_7()
23277       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 3);  in gemm_i32_0_1_7()
23280   const std::int32_t result_chunk_stride = result_stride * 3;  in gemm_i32_0_1_7()
23282   const std::int32_t const_offset = lhs_offset * rhs_offset * k;  in gemm_i32_0_1_7()
23283   std::int32_t* result_chunk = result;  in gemm_i32_0_1_7()
23284   std::int32_t* mul_result_chunk = result;  in gemm_i32_0_1_7()
23285   const std::int32_t mul_result_chunk_stride_bytes = result_stride * 4;  in gemm_i32_0_1_7()
23314                     const std::uint8_t* rhs, std::int32_t m, std::int32_t n,  in gemm_i32_0_2_0()
23315                     std::int32_t k, std::int32_t lhs_offset,  in gemm_i32_0_2_0()
23316                     std::int32_t rhs_offset, std::int32_t* result,  in gemm_i32_0_2_0()
23317                     std::int32_t result_stride) {  in gemm_i32_0_2_0()
23318   const std::int32_t row_chunks = m / 3;  in gemm_i32_0_2_0()
23319   const std::int32_t col_chunks = n / 3;  in gemm_i32_0_2_0()
23320   const std::int32_t padded_k = ((k + 7) / 8) * 8;  in gemm_i32_0_2_0()
23321   const std::int32_t chunk_size = k * 3;  in gemm_i32_0_2_0()
23322   const std::int32_t zipped_chunk_size = (padded_k + 16) * 3;  in gemm_i32_0_2_0()
23323   const std::int32_t zipped_rhs_size = (padded_k + 16) * n;  in gemm_i32_0_2_0()
23327   std::int32_t* zipped_lhs_3_offsets =  in gemm_i32_0_2_0()
23328       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 3);  in gemm_i32_0_2_0()
23331   const std::int32_t result_chunk_stride = result_stride * 3;  in gemm_i32_0_2_0()
23333   const std::int32_t const_offset = lhs_offset * rhs_offset * k;  in gemm_i32_0_2_0()
23334   std::int32_t* result_chunk = result;  in gemm_i32_0_2_0()
23335   std::int32_t* mul_result_chunk = result;  in gemm_i32_0_2_0()
23336   const std::int32_t mul_result_chunk_stride_bytes = result_stride * 4;  in gemm_i32_0_2_0()
23365                     const std::uint8_t* rhs, std::int32_t m, std::int32_t n,  in gemm_i32_0_2_1()
23366                     std::int32_t k, std::int32_t lhs_offset,  in gemm_i32_0_2_1()
23367                     std::int32_t rhs_offset, std::int32_t* result,  in gemm_i32_0_2_1()
23368                     std::int32_t result_stride) {  in gemm_i32_0_2_1()
23369   const std::int32_t row_chunks = m / 3;  in gemm_i32_0_2_1()
23370   const std::int32_t col_chunks = n / 3;  in gemm_i32_0_2_1()
23371   const std::int32_t padded_k = ((k + 7) / 8) * 8;  in gemm_i32_0_2_1()
23372   const std::int32_t chunk_size = k * 3;  in gemm_i32_0_2_1()
23373   const std::int32_t zipped_chunk_size = (padded_k + 16) * 3;  in gemm_i32_0_2_1()
23374   const std::int32_t zipped_rhs_size = (padded_k + 16) * n;  in gemm_i32_0_2_1()
23378   std::int32_t* zipped_lhs_3_offsets =  in gemm_i32_0_2_1()
23379       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 3);  in gemm_i32_0_2_1()
23382   const std::int32_t result_chunk_stride = result_stride * 3;  in gemm_i32_0_2_1()
23384   const std::int32_t const_offset = lhs_offset * rhs_offset * k;  in gemm_i32_0_2_1()
23385   std::int32_t* result_chunk = result;  in gemm_i32_0_2_1()
23386   std::int32_t* mul_result_chunk = result;  in gemm_i32_0_2_1()
23387   const std::int32_t mul_result_chunk_stride_bytes = result_stride * 4;  in gemm_i32_0_2_1()
23416                     const std::uint8_t* rhs, std::int32_t m, std::int32_t n,  in gemm_i32_0_2_2()
23417                     std::int32_t k, std::int32_t lhs_offset,  in gemm_i32_0_2_2()
23418                     std::int32_t rhs_offset, std::int32_t* result,  in gemm_i32_0_2_2()
23419                     std::int32_t result_stride) {  in gemm_i32_0_2_2()
23420   const std::int32_t row_chunks = m / 3;  in gemm_i32_0_2_2()
23421   const std::int32_t col_chunks = n / 3;  in gemm_i32_0_2_2()
23422   const std::int32_t padded_k = ((k + 7) / 8) * 8;  in gemm_i32_0_2_2()
23423   const std::int32_t chunk_size = k * 3;  in gemm_i32_0_2_2()
23424   const std::int32_t zipped_chunk_size = (padded_k + 16) * 3;  in gemm_i32_0_2_2()
23425   const std::int32_t zipped_rhs_size = (padded_k + 16) * n;  in gemm_i32_0_2_2()
23429   std::int32_t* zipped_lhs_3_offsets =  in gemm_i32_0_2_2()
23430       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 3);  in gemm_i32_0_2_2()
23433   const std::int32_t result_chunk_stride = result_stride * 3;  in gemm_i32_0_2_2()
23435   const std::int32_t const_offset = lhs_offset * rhs_offset * k;  in gemm_i32_0_2_2()
23436   std::int32_t* result_chunk = result;  in gemm_i32_0_2_2()
23437   std::int32_t* mul_result_chunk = result;  in gemm_i32_0_2_2()
23438   const std::int32_t mul_result_chunk_stride_bytes = result_stride * 4;  in gemm_i32_0_2_2()
23467                     const std::uint8_t* rhs, std::int32_t m, std::int32_t n,  in gemm_i32_0_2_3()
23468                     std::int32_t k, std::int32_t lhs_offset,  in gemm_i32_0_2_3()
23469                     std::int32_t rhs_offset, std::int32_t* result,  in gemm_i32_0_2_3()
23470                     std::int32_t result_stride) {  in gemm_i32_0_2_3()
23471   const std::int32_t row_chunks = m / 3;  in gemm_i32_0_2_3()
23472   const std::int32_t col_chunks = n / 3;  in gemm_i32_0_2_3()
23473   const std::int32_t padded_k = ((k + 7) / 8) * 8;  in gemm_i32_0_2_3()
23474   const std::int32_t chunk_size = k * 3;  in gemm_i32_0_2_3()
23475   const std::int32_t zipped_chunk_size = (padded_k + 16) * 3;  in gemm_i32_0_2_3()
23476   const std::int32_t zipped_rhs_size = (padded_k + 16) * n;  in gemm_i32_0_2_3()
23480   std::int32_t* zipped_lhs_3_offsets =  in gemm_i32_0_2_3()
23481       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 3);  in gemm_i32_0_2_3()
23484   const std::int32_t result_chunk_stride = result_stride * 3;  in gemm_i32_0_2_3()
23486   const std::int32_t const_offset = lhs_offset * rhs_offset * k;  in gemm_i32_0_2_3()
23487   std::int32_t* result_chunk = result;  in gemm_i32_0_2_3()
23488   std::int32_t* mul_result_chunk = result;  in gemm_i32_0_2_3()
23489   const std::int32_t mul_result_chunk_stride_bytes = result_stride * 4;  in gemm_i32_0_2_3()
23518                     const std::uint8_t* rhs, std::int32_t m, std::int32_t n,  in gemm_i32_0_2_4()
23519                     std::int32_t k, std::int32_t lhs_offset,  in gemm_i32_0_2_4()
23520                     std::int32_t rhs_offset, std::int32_t* result,  in gemm_i32_0_2_4()
23521                     std::int32_t result_stride) {  in gemm_i32_0_2_4()
23522   const std::int32_t row_chunks = m / 3;  in gemm_i32_0_2_4()
23523   const std::int32_t col_chunks = n / 3;  in gemm_i32_0_2_4()
23524   const std::int32_t padded_k = ((k + 7) / 8) * 8;  in gemm_i32_0_2_4()
23525   const std::int32_t chunk_size = k * 3;  in gemm_i32_0_2_4()
23526   const std::int32_t zipped_chunk_size = (padded_k + 16) * 3;  in gemm_i32_0_2_4()
23527   const std::int32_t zipped_rhs_size = (padded_k + 16) * n;  in gemm_i32_0_2_4()
23531   std::int32_t* zipped_lhs_3_offsets =  in gemm_i32_0_2_4()
23532       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 3);  in gemm_i32_0_2_4()
23535   const std::int32_t result_chunk_stride = result_stride * 3;  in gemm_i32_0_2_4()
23537   const std::int32_t const_offset = lhs_offset * rhs_offset * k;  in gemm_i32_0_2_4()
23538   std::int32_t* result_chunk = result;  in gemm_i32_0_2_4()
23539   std::int32_t* mul_result_chunk = result;  in gemm_i32_0_2_4()
23540   const std::int32_t mul_result_chunk_stride_bytes = result_stride * 4;  in gemm_i32_0_2_4()
23569                     const std::uint8_t* rhs, std::int32_t m, std::int32_t n,  in gemm_i32_0_2_5()
23570                     std::int32_t k, std::int32_t lhs_offset,  in gemm_i32_0_2_5()
23571                     std::int32_t rhs_offset, std::int32_t* result,  in gemm_i32_0_2_5()
23572                     std::int32_t result_stride) {  in gemm_i32_0_2_5()
23573   const std::int32_t row_chunks = m / 3;  in gemm_i32_0_2_5()
23574   const std::int32_t col_chunks = n / 3;  in gemm_i32_0_2_5()
23575   const std::int32_t padded_k = ((k + 7) / 8) * 8;  in gemm_i32_0_2_5()
23576   const std::int32_t chunk_size = k * 3;  in gemm_i32_0_2_5()
23577   const std::int32_t zipped_chunk_size = (padded_k + 16) * 3;  in gemm_i32_0_2_5()
23578   const std::int32_t zipped_rhs_size = (padded_k + 16) * n;  in gemm_i32_0_2_5()
23582   std::int32_t* zipped_lhs_3_offsets =  in gemm_i32_0_2_5()
23583       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 3);  in gemm_i32_0_2_5()
23586   const std::int32_t result_chunk_stride = result_stride * 3;  in gemm_i32_0_2_5()
23588   const std::int32_t const_offset = lhs_offset * rhs_offset * k;  in gemm_i32_0_2_5()
23589   std::int32_t* result_chunk = result;  in gemm_i32_0_2_5()
23590   std::int32_t* mul_result_chunk = result;  in gemm_i32_0_2_5()
23591   const std::int32_t mul_result_chunk_stride_bytes = result_stride * 4;  in gemm_i32_0_2_5()
23620                     const std::uint8_t* rhs, std::int32_t m, std::int32_t n,  in gemm_i32_0_2_6()
23621                     std::int32_t k, std::int32_t lhs_offset,  in gemm_i32_0_2_6()
23622                     std::int32_t rhs_offset, std::int32_t* result,  in gemm_i32_0_2_6()
23623                     std::int32_t result_stride) {  in gemm_i32_0_2_6()
23624   const std::int32_t row_chunks = m / 3;  in gemm_i32_0_2_6()
23625   const std::int32_t col_chunks = n / 3;  in gemm_i32_0_2_6()
23626   const std::int32_t padded_k = ((k + 7) / 8) * 8;  in gemm_i32_0_2_6()
23627   const std::int32_t chunk_size = k * 3;  in gemm_i32_0_2_6()
23628   const std::int32_t zipped_chunk_size = (padded_k + 16) * 3;  in gemm_i32_0_2_6()
23629   const std::int32_t zipped_rhs_size = (padded_k + 16) * n;  in gemm_i32_0_2_6()
23633   std::int32_t* zipped_lhs_3_offsets =  in gemm_i32_0_2_6()
23634       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 3);  in gemm_i32_0_2_6()
23637   const std::int32_t result_chunk_stride = result_stride * 3;  in gemm_i32_0_2_6()
23639   const std::int32_t const_offset = lhs_offset * rhs_offset * k;  in gemm_i32_0_2_6()
23640   std::int32_t* result_chunk = result;  in gemm_i32_0_2_6()
23641   std::int32_t* mul_result_chunk = result;  in gemm_i32_0_2_6()
23642   const std::int32_t mul_result_chunk_stride_bytes = result_stride * 4;  in gemm_i32_0_2_6()
23671                     const std::uint8_t* rhs, std::int32_t m, std::int32_t n,  in gemm_i32_0_2_7()
23672                     std::int32_t k, std::int32_t lhs_offset,  in gemm_i32_0_2_7()
23673                     std::int32_t rhs_offset, std::int32_t* result,  in gemm_i32_0_2_7()
23674                     std::int32_t result_stride) {  in gemm_i32_0_2_7()
23675   const std::int32_t row_chunks = m / 3;  in gemm_i32_0_2_7()
23676   const std::int32_t col_chunks = n / 3;  in gemm_i32_0_2_7()
23677   const std::int32_t padded_k = ((k + 7) / 8) * 8;  in gemm_i32_0_2_7()
23678   const std::int32_t chunk_size = k * 3;  in gemm_i32_0_2_7()
23679   const std::int32_t zipped_chunk_size = (padded_k + 16) * 3;  in gemm_i32_0_2_7()
23680   const std::int32_t zipped_rhs_size = (padded_k + 16) * n;  in gemm_i32_0_2_7()
23684   std::int32_t* zipped_lhs_3_offsets =  in gemm_i32_0_2_7()
23685       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 3);  in gemm_i32_0_2_7()
23688   const std::int32_t result_chunk_stride = result_stride * 3;  in gemm_i32_0_2_7()
23690   const std::int32_t const_offset = lhs_offset * rhs_offset * k;  in gemm_i32_0_2_7()
23691   std::int32_t* result_chunk = result;  in gemm_i32_0_2_7()
23692   std::int32_t* mul_result_chunk = result;  in gemm_i32_0_2_7()
23693   const std::int32_t mul_result_chunk_stride_bytes = result_stride * 4;  in gemm_i32_0_2_7()
23722                     const std::uint8_t* rhs, std::int32_t m, std::int32_t n,  in gemm_i32_1_0_0()
23723                     std::int32_t k, std::int32_t lhs_offset,  in gemm_i32_1_0_0()
23724                     std::int32_t rhs_offset, std::int32_t* result,  in gemm_i32_1_0_0()
23725                     std::int32_t result_stride) {  in gemm_i32_1_0_0()
23726   const std::int32_t row_chunks = m / 3;  in gemm_i32_1_0_0()
23727   const std::int32_t col_chunks = n / 3;  in gemm_i32_1_0_0()
23728   const std::int32_t padded_k = ((k + 7) / 8) * 8;  in gemm_i32_1_0_0()
23729   const std::int32_t chunk_size = k * 3;  in gemm_i32_1_0_0()
23730   const std::int32_t zipped_chunk_size = (padded_k + 16) * 3;  in gemm_i32_1_0_0()
23731   const std::int32_t zipped_rhs_size = (padded_k + 16) * n;  in gemm_i32_1_0_0()
23735   std::int32_t* zipped_lhs_3_offsets =  in gemm_i32_1_0_0()
23736       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 3);  in gemm_i32_1_0_0()
23737   std::int32_t* zipped_lhs_1_offsets =  in gemm_i32_1_0_0()
23738       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 1);  in gemm_i32_1_0_0()
23741   const std::int32_t result_chunk_stride = result_stride * 3;  in gemm_i32_1_0_0()
23743   const std::int32_t const_offset = lhs_offset * rhs_offset * k;  in gemm_i32_1_0_0()
23744   std::int32_t* result_chunk = result;  in gemm_i32_1_0_0()
23745   std::int32_t* mul_result_chunk = result;  in gemm_i32_1_0_0()
23746   const std::int32_t mul_result_chunk_stride_bytes = result_stride * 4;  in gemm_i32_1_0_0()
23782                     const std::uint8_t* rhs, std::int32_t m, std::int32_t n,  in gemm_i32_1_0_1()
23783                     std::int32_t k, std::int32_t lhs_offset,  in gemm_i32_1_0_1()
23784                     std::int32_t rhs_offset, std::int32_t* result,  in gemm_i32_1_0_1()
23785                     std::int32_t result_stride) {  in gemm_i32_1_0_1()
23786   const std::int32_t row_chunks = m / 3;  in gemm_i32_1_0_1()
23787   const std::int32_t col_chunks = n / 3;  in gemm_i32_1_0_1()
23788   const std::int32_t padded_k = ((k + 7) / 8) * 8;  in gemm_i32_1_0_1()
23789   const std::int32_t chunk_size = k * 3;  in gemm_i32_1_0_1()
23790   const std::int32_t zipped_chunk_size = (padded_k + 16) * 3;  in gemm_i32_1_0_1()
23791   const std::int32_t zipped_rhs_size = (padded_k + 16) * n;  in gemm_i32_1_0_1()
23795   std::int32_t* zipped_lhs_3_offsets =  in gemm_i32_1_0_1()
23796       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 3);  in gemm_i32_1_0_1()
23797   std::int32_t* zipped_lhs_1_offsets =  in gemm_i32_1_0_1()
23798       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 1);  in gemm_i32_1_0_1()
23801   const std::int32_t result_chunk_stride = result_stride * 3;  in gemm_i32_1_0_1()
23803   const std::int32_t const_offset = lhs_offset * rhs_offset * k;  in gemm_i32_1_0_1()
23804   std::int32_t* result_chunk = result;  in gemm_i32_1_0_1()
23805   std::int32_t* mul_result_chunk = result;  in gemm_i32_1_0_1()
23806   const std::int32_t mul_result_chunk_stride_bytes = result_stride * 4;  in gemm_i32_1_0_1()
23842                     const std::uint8_t* rhs, std::int32_t m, std::int32_t n,  in gemm_i32_1_0_2()
23843                     std::int32_t k, std::int32_t lhs_offset,  in gemm_i32_1_0_2()
23844                     std::int32_t rhs_offset, std::int32_t* result,  in gemm_i32_1_0_2()
23845                     std::int32_t result_stride) {  in gemm_i32_1_0_2()
23846   const std::int32_t row_chunks = m / 3;  in gemm_i32_1_0_2()
23847   const std::int32_t col_chunks = n / 3;  in gemm_i32_1_0_2()
23848   const std::int32_t padded_k = ((k + 7) / 8) * 8;  in gemm_i32_1_0_2()
23849   const std::int32_t chunk_size = k * 3;  in gemm_i32_1_0_2()
23850   const std::int32_t zipped_chunk_size = (padded_k + 16) * 3;  in gemm_i32_1_0_2()
23851   const std::int32_t zipped_rhs_size = (padded_k + 16) * n;  in gemm_i32_1_0_2()
23855   std::int32_t* zipped_lhs_3_offsets =  in gemm_i32_1_0_2()
23856       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 3);  in gemm_i32_1_0_2()
23857   std::int32_t* zipped_lhs_1_offsets =  in gemm_i32_1_0_2()
23858       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 1);  in gemm_i32_1_0_2()
23861   const std::int32_t result_chunk_stride = result_stride * 3;  in gemm_i32_1_0_2()
23863   const std::int32_t const_offset = lhs_offset * rhs_offset * k;  in gemm_i32_1_0_2()
23864   std::int32_t* result_chunk = result;  in gemm_i32_1_0_2()
23865   std::int32_t* mul_result_chunk = result;  in gemm_i32_1_0_2()
23866   const std::int32_t mul_result_chunk_stride_bytes = result_stride * 4;  in gemm_i32_1_0_2()
23902                     const std::uint8_t* rhs, std::int32_t m, std::int32_t n,  in gemm_i32_1_0_3()
23903                     std::int32_t k, std::int32_t lhs_offset,  in gemm_i32_1_0_3()
23904                     std::int32_t rhs_offset, std::int32_t* result,  in gemm_i32_1_0_3()
23905                     std::int32_t result_stride) {  in gemm_i32_1_0_3()
23906   const std::int32_t row_chunks = m / 3;  in gemm_i32_1_0_3()
23907   const std::int32_t col_chunks = n / 3;  in gemm_i32_1_0_3()
23908   const std::int32_t padded_k = ((k + 7) / 8) * 8;  in gemm_i32_1_0_3()
23909   const std::int32_t chunk_size = k * 3;  in gemm_i32_1_0_3()
23910   const std::int32_t zipped_chunk_size = (padded_k + 16) * 3;  in gemm_i32_1_0_3()
23911   const std::int32_t zipped_rhs_size = (padded_k + 16) * n;  in gemm_i32_1_0_3()
23915   std::int32_t* zipped_lhs_3_offsets =  in gemm_i32_1_0_3()
23916       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 3);  in gemm_i32_1_0_3()
23917   std::int32_t* zipped_lhs_1_offsets =  in gemm_i32_1_0_3()
23918       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 1);  in gemm_i32_1_0_3()
23921   const std::int32_t result_chunk_stride = result_stride * 3;  in gemm_i32_1_0_3()
23923   const std::int32_t const_offset = lhs_offset * rhs_offset * k;  in gemm_i32_1_0_3()
23924   std::int32_t* result_chunk = result;  in gemm_i32_1_0_3()
23925   std::int32_t* mul_result_chunk = result;  in gemm_i32_1_0_3()
23926   const std::int32_t mul_result_chunk_stride_bytes = result_stride * 4;  in gemm_i32_1_0_3()
23962                     const std::uint8_t* rhs, std::int32_t m, std::int32_t n,  in gemm_i32_1_0_4()
23963                     std::int32_t k, std::int32_t lhs_offset,  in gemm_i32_1_0_4()
23964                     std::int32_t rhs_offset, std::int32_t* result,  in gemm_i32_1_0_4()
23965                     std::int32_t result_stride) {  in gemm_i32_1_0_4()
23966   const std::int32_t row_chunks = m / 3;  in gemm_i32_1_0_4()
23967   const std::int32_t col_chunks = n / 3;  in gemm_i32_1_0_4()
23968   const std::int32_t padded_k = ((k + 7) / 8) * 8;  in gemm_i32_1_0_4()
23969   const std::int32_t chunk_size = k * 3;  in gemm_i32_1_0_4()
23970   const std::int32_t zipped_chunk_size = (padded_k + 16) * 3;  in gemm_i32_1_0_4()
23971   const std::int32_t zipped_rhs_size = (padded_k + 16) * n;  in gemm_i32_1_0_4()
23975   std::int32_t* zipped_lhs_3_offsets =  in gemm_i32_1_0_4()
23976       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 3);  in gemm_i32_1_0_4()
23977   std::int32_t* zipped_lhs_1_offsets =  in gemm_i32_1_0_4()
23978       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 1);  in gemm_i32_1_0_4()
23981   const std::int32_t result_chunk_stride = result_stride * 3;  in gemm_i32_1_0_4()
23983   const std::int32_t const_offset = lhs_offset * rhs_offset * k;  in gemm_i32_1_0_4()
23984   std::int32_t* result_chunk = result;  in gemm_i32_1_0_4()
23985   std::int32_t* mul_result_chunk = result;  in gemm_i32_1_0_4()
23986   const std::int32_t mul_result_chunk_stride_bytes = result_stride * 4;  in gemm_i32_1_0_4()
24022                     const std::uint8_t* rhs, std::int32_t m, std::int32_t n,  in gemm_i32_1_0_5()
24023                     std::int32_t k, std::int32_t lhs_offset,  in gemm_i32_1_0_5()
24024                     std::int32_t rhs_offset, std::int32_t* result,  in gemm_i32_1_0_5()
24025                     std::int32_t result_stride) {  in gemm_i32_1_0_5()
24026   const std::int32_t row_chunks = m / 3;  in gemm_i32_1_0_5()
24027   const std::int32_t col_chunks = n / 3;  in gemm_i32_1_0_5()
24028   const std::int32_t padded_k = ((k + 7) / 8) * 8;  in gemm_i32_1_0_5()
24029   const std::int32_t chunk_size = k * 3;  in gemm_i32_1_0_5()
24030   const std::int32_t zipped_chunk_size = (padded_k + 16) * 3;  in gemm_i32_1_0_5()
24031   const std::int32_t zipped_rhs_size = (padded_k + 16) * n;  in gemm_i32_1_0_5()
24035   std::int32_t* zipped_lhs_3_offsets =  in gemm_i32_1_0_5()
24036       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 3);  in gemm_i32_1_0_5()
24037   std::int32_t* zipped_lhs_1_offsets =  in gemm_i32_1_0_5()
24038       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 1);  in gemm_i32_1_0_5()
24041   const std::int32_t result_chunk_stride = result_stride * 3;  in gemm_i32_1_0_5()
24043   const std::int32_t const_offset = lhs_offset * rhs_offset * k;  in gemm_i32_1_0_5()
24044   std::int32_t* result_chunk = result;  in gemm_i32_1_0_5()
24045   std::int32_t* mul_result_chunk = result;  in gemm_i32_1_0_5()
24046   const std::int32_t mul_result_chunk_stride_bytes = result_stride * 4;  in gemm_i32_1_0_5()
24082                     const std::uint8_t* rhs, std::int32_t m, std::int32_t n,  in gemm_i32_1_0_6()
24083                     std::int32_t k, std::int32_t lhs_offset,  in gemm_i32_1_0_6()
24084                     std::int32_t rhs_offset, std::int32_t* result,  in gemm_i32_1_0_6()
24085                     std::int32_t result_stride) {  in gemm_i32_1_0_6()
24086   const std::int32_t row_chunks = m / 3;  in gemm_i32_1_0_6()
24087   const std::int32_t col_chunks = n / 3;  in gemm_i32_1_0_6()
24088   const std::int32_t padded_k = ((k + 7) / 8) * 8;  in gemm_i32_1_0_6()
24089   const std::int32_t chunk_size = k * 3;  in gemm_i32_1_0_6()
24090   const std::int32_t zipped_chunk_size = (padded_k + 16) * 3;  in gemm_i32_1_0_6()
24091   const std::int32_t zipped_rhs_size = (padded_k + 16) * n;  in gemm_i32_1_0_6()
24095   std::int32_t* zipped_lhs_3_offsets =  in gemm_i32_1_0_6()
24096       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 3);  in gemm_i32_1_0_6()
24097   std::int32_t* zipped_lhs_1_offsets =  in gemm_i32_1_0_6()
24098       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 1);  in gemm_i32_1_0_6()
24101   const std::int32_t result_chunk_stride = result_stride * 3;  in gemm_i32_1_0_6()
24103   const std::int32_t const_offset = lhs_offset * rhs_offset * k;  in gemm_i32_1_0_6()
24104   std::int32_t* result_chunk = result;  in gemm_i32_1_0_6()
24105   std::int32_t* mul_result_chunk = result;  in gemm_i32_1_0_6()
24106   const std::int32_t mul_result_chunk_stride_bytes = result_stride * 4;  in gemm_i32_1_0_6()
24142                     const std::uint8_t* rhs, std::int32_t m, std::int32_t n,  in gemm_i32_1_0_7()
24143                     std::int32_t k, std::int32_t lhs_offset,  in gemm_i32_1_0_7()
24144                     std::int32_t rhs_offset, std::int32_t* result,  in gemm_i32_1_0_7()
24145                     std::int32_t result_stride) {  in gemm_i32_1_0_7()
24146   const std::int32_t row_chunks = m / 3;  in gemm_i32_1_0_7()
24147   const std::int32_t col_chunks = n / 3;  in gemm_i32_1_0_7()
24148   const std::int32_t padded_k = ((k + 7) / 8) * 8;  in gemm_i32_1_0_7()
24149   const std::int32_t chunk_size = k * 3;  in gemm_i32_1_0_7()
24150   const std::int32_t zipped_chunk_size = (padded_k + 16) * 3;  in gemm_i32_1_0_7()
24151   const std::int32_t zipped_rhs_size = (padded_k + 16) * n;  in gemm_i32_1_0_7()
24155   std::int32_t* zipped_lhs_3_offsets =  in gemm_i32_1_0_7()
24156       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 3);  in gemm_i32_1_0_7()
24157   std::int32_t* zipped_lhs_1_offsets =  in gemm_i32_1_0_7()
24158       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 1);  in gemm_i32_1_0_7()
24161   const std::int32_t result_chunk_stride = result_stride * 3;  in gemm_i32_1_0_7()
24163   const std::int32_t const_offset = lhs_offset * rhs_offset * k;  in gemm_i32_1_0_7()
24164   std::int32_t* result_chunk = result;  in gemm_i32_1_0_7()
24165   std::int32_t* mul_result_chunk = result;  in gemm_i32_1_0_7()
24166   const std::int32_t mul_result_chunk_stride_bytes = result_stride * 4;  in gemm_i32_1_0_7()
24202                     const std::uint8_t* rhs, std::int32_t m, std::int32_t n,  in gemm_i32_1_1_0()
24203                     std::int32_t k, std::int32_t lhs_offset,  in gemm_i32_1_1_0()
24204                     std::int32_t rhs_offset, std::int32_t* result,  in gemm_i32_1_1_0()
24205                     std::int32_t result_stride) {  in gemm_i32_1_1_0()
24206   const std::int32_t row_chunks = m / 3;  in gemm_i32_1_1_0()
24207   const std::int32_t col_chunks = n / 3;  in gemm_i32_1_1_0()
24208   const std::int32_t padded_k = ((k + 7) / 8) * 8;  in gemm_i32_1_1_0()
24209   const std::int32_t chunk_size = k * 3;  in gemm_i32_1_1_0()
24210   const std::int32_t zipped_chunk_size = (padded_k + 16) * 3;  in gemm_i32_1_1_0()
24211   const std::int32_t zipped_rhs_size = (padded_k + 16) * n;  in gemm_i32_1_1_0()
24215   std::int32_t* zipped_lhs_3_offsets =  in gemm_i32_1_1_0()
24216       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 3);  in gemm_i32_1_1_0()
24217   std::int32_t* zipped_lhs_1_offsets =  in gemm_i32_1_1_0()
24218       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 1);  in gemm_i32_1_1_0()
24221   const std::int32_t result_chunk_stride = result_stride * 3;  in gemm_i32_1_1_0()
24223   const std::int32_t const_offset = lhs_offset * rhs_offset * k;  in gemm_i32_1_1_0()
24224   std::int32_t* result_chunk = result;  in gemm_i32_1_1_0()
24225   std::int32_t* mul_result_chunk = result;  in gemm_i32_1_1_0()
24226   const std::int32_t mul_result_chunk_stride_bytes = result_stride * 4;  in gemm_i32_1_1_0()
24269                     const std::uint8_t* rhs, std::int32_t m, std::int32_t n,  in gemm_i32_1_1_1()
24270                     std::int32_t k, std::int32_t lhs_offset,  in gemm_i32_1_1_1()
24271                     std::int32_t rhs_offset, std::int32_t* result,  in gemm_i32_1_1_1()
24272                     std::int32_t result_stride) {  in gemm_i32_1_1_1()
24273   const std::int32_t row_chunks = m / 3;  in gemm_i32_1_1_1()
24274   const std::int32_t col_chunks = n / 3;  in gemm_i32_1_1_1()
24275   const std::int32_t padded_k = ((k + 7) / 8) * 8;  in gemm_i32_1_1_1()
24276   const std::int32_t chunk_size = k * 3;  in gemm_i32_1_1_1()
24277   const std::int32_t zipped_chunk_size = (padded_k + 16) * 3;  in gemm_i32_1_1_1()
24278   const std::int32_t zipped_rhs_size = (padded_k + 16) * n;  in gemm_i32_1_1_1()
24282   std::int32_t* zipped_lhs_3_offsets =  in gemm_i32_1_1_1()
24283       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 3);  in gemm_i32_1_1_1()
24284   std::int32_t* zipped_lhs_1_offsets =  in gemm_i32_1_1_1()
24285       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 1);  in gemm_i32_1_1_1()
24288   const std::int32_t result_chunk_stride = result_stride * 3;  in gemm_i32_1_1_1()
24290   const std::int32_t const_offset = lhs_offset * rhs_offset * k;  in gemm_i32_1_1_1()
24291   std::int32_t* result_chunk = result;  in gemm_i32_1_1_1()
24292   std::int32_t* mul_result_chunk = result;  in gemm_i32_1_1_1()
24293   const std::int32_t mul_result_chunk_stride_bytes = result_stride * 4;  in gemm_i32_1_1_1()
24336                     const std::uint8_t* rhs, std::int32_t m, std::int32_t n,  in gemm_i32_1_1_2()
24337                     std::int32_t k, std::int32_t lhs_offset,  in gemm_i32_1_1_2()
24338                     std::int32_t rhs_offset, std::int32_t* result,  in gemm_i32_1_1_2()
24339                     std::int32_t result_stride) {  in gemm_i32_1_1_2()
24340   const std::int32_t row_chunks = m / 3;  in gemm_i32_1_1_2()
24341   const std::int32_t col_chunks = n / 3;  in gemm_i32_1_1_2()
24342   const std::int32_t padded_k = ((k + 7) / 8) * 8;  in gemm_i32_1_1_2()
24343   const std::int32_t chunk_size = k * 3;  in gemm_i32_1_1_2()
24344   const std::int32_t zipped_chunk_size = (padded_k + 16) * 3;  in gemm_i32_1_1_2()
24345   const std::int32_t zipped_rhs_size = (padded_k + 16) * n;  in gemm_i32_1_1_2()
24349   std::int32_t* zipped_lhs_3_offsets =  in gemm_i32_1_1_2()
24350       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 3);  in gemm_i32_1_1_2()
24351   std::int32_t* zipped_lhs_1_offsets =  in gemm_i32_1_1_2()
24352       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 1);  in gemm_i32_1_1_2()
24355   const std::int32_t result_chunk_stride = result_stride * 3;  in gemm_i32_1_1_2()
24357   const std::int32_t const_offset = lhs_offset * rhs_offset * k;  in gemm_i32_1_1_2()
24358   std::int32_t* result_chunk = result;  in gemm_i32_1_1_2()
24359   std::int32_t* mul_result_chunk = result;  in gemm_i32_1_1_2()
24360   const std::int32_t mul_result_chunk_stride_bytes = result_stride * 4;  in gemm_i32_1_1_2()
24403                     const std::uint8_t* rhs, std::int32_t m, std::int32_t n,  in gemm_i32_1_1_3()
24404                     std::int32_t k, std::int32_t lhs_offset,  in gemm_i32_1_1_3()
24405                     std::int32_t rhs_offset, std::int32_t* result,  in gemm_i32_1_1_3()
24406                     std::int32_t result_stride) {  in gemm_i32_1_1_3()
24407   const std::int32_t row_chunks = m / 3;  in gemm_i32_1_1_3()
24408   const std::int32_t col_chunks = n / 3;  in gemm_i32_1_1_3()
24409   const std::int32_t padded_k = ((k + 7) / 8) * 8;  in gemm_i32_1_1_3()
24410   const std::int32_t chunk_size = k * 3;  in gemm_i32_1_1_3()
24411   const std::int32_t zipped_chunk_size = (padded_k + 16) * 3;  in gemm_i32_1_1_3()
24412   const std::int32_t zipped_rhs_size = (padded_k + 16) * n;  in gemm_i32_1_1_3()
24416   std::int32_t* zipped_lhs_3_offsets =  in gemm_i32_1_1_3()
24417       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 3);  in gemm_i32_1_1_3()
24418   std::int32_t* zipped_lhs_1_offsets =  in gemm_i32_1_1_3()
24419       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 1);  in gemm_i32_1_1_3()
24422   const std::int32_t result_chunk_stride = result_stride * 3;  in gemm_i32_1_1_3()
24424   const std::int32_t const_offset = lhs_offset * rhs_offset * k;  in gemm_i32_1_1_3()
24425   std::int32_t* result_chunk = result;  in gemm_i32_1_1_3()
24426   std::int32_t* mul_result_chunk = result;  in gemm_i32_1_1_3()
24427   const std::int32_t mul_result_chunk_stride_bytes = result_stride * 4;  in gemm_i32_1_1_3()
24470                     const std::uint8_t* rhs, std::int32_t m, std::int32_t n,  in gemm_i32_1_1_4()
24471                     std::int32_t k, std::int32_t lhs_offset,  in gemm_i32_1_1_4()
24472                     std::int32_t rhs_offset, std::int32_t* result,  in gemm_i32_1_1_4()
24473                     std::int32_t result_stride) {  in gemm_i32_1_1_4()
24474   const std::int32_t row_chunks = m / 3;  in gemm_i32_1_1_4()
24475   const std::int32_t col_chunks = n / 3;  in gemm_i32_1_1_4()
24476   const std::int32_t padded_k = ((k + 7) / 8) * 8;  in gemm_i32_1_1_4()
24477   const std::int32_t chunk_size = k * 3;  in gemm_i32_1_1_4()
24478   const std::int32_t zipped_chunk_size = (padded_k + 16) * 3;  in gemm_i32_1_1_4()
24479   const std::int32_t zipped_rhs_size = (padded_k + 16) * n;  in gemm_i32_1_1_4()
24483   std::int32_t* zipped_lhs_3_offsets =  in gemm_i32_1_1_4()
24484       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 3);  in gemm_i32_1_1_4()
24485   std::int32_t* zipped_lhs_1_offsets =  in gemm_i32_1_1_4()
24486       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 1);  in gemm_i32_1_1_4()
24489   const std::int32_t result_chunk_stride = result_stride * 3;  in gemm_i32_1_1_4()
24491   const std::int32_t const_offset = lhs_offset * rhs_offset * k;  in gemm_i32_1_1_4()
24492   std::int32_t* result_chunk = result;  in gemm_i32_1_1_4()
24493   std::int32_t* mul_result_chunk = result;  in gemm_i32_1_1_4()
24494   const std::int32_t mul_result_chunk_stride_bytes = result_stride * 4;  in gemm_i32_1_1_4()
24537                     const std::uint8_t* rhs, std::int32_t m, std::int32_t n,  in gemm_i32_1_1_5()
24538                     std::int32_t k, std::int32_t lhs_offset,  in gemm_i32_1_1_5()
24539                     std::int32_t rhs_offset, std::int32_t* result,  in gemm_i32_1_1_5()
24540                     std::int32_t result_stride) {  in gemm_i32_1_1_5()
24541   const std::int32_t row_chunks = m / 3;  in gemm_i32_1_1_5()
24542   const std::int32_t col_chunks = n / 3;  in gemm_i32_1_1_5()
24543   const std::int32_t padded_k = ((k + 7) / 8) * 8;  in gemm_i32_1_1_5()
24544   const std::int32_t chunk_size = k * 3;  in gemm_i32_1_1_5()
24545   const std::int32_t zipped_chunk_size = (padded_k + 16) * 3;  in gemm_i32_1_1_5()
24546   const std::int32_t zipped_rhs_size = (padded_k + 16) * n;  in gemm_i32_1_1_5()
24550   std::int32_t* zipped_lhs_3_offsets =  in gemm_i32_1_1_5()
24551       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 3);  in gemm_i32_1_1_5()
24552   std::int32_t* zipped_lhs_1_offsets =  in gemm_i32_1_1_5()
24553       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 1);  in gemm_i32_1_1_5()
24556   const std::int32_t result_chunk_stride = result_stride * 3;  in gemm_i32_1_1_5()
24558   const std::int32_t const_offset = lhs_offset * rhs_offset * k;  in gemm_i32_1_1_5()
24559   std::int32_t* result_chunk = result;  in gemm_i32_1_1_5()
24560   std::int32_t* mul_result_chunk = result;  in gemm_i32_1_1_5()
24561   const std::int32_t mul_result_chunk_stride_bytes = result_stride * 4;  in gemm_i32_1_1_5()
24604                     const std::uint8_t* rhs, std::int32_t m, std::int32_t n,  in gemm_i32_1_1_6()
24605                     std::int32_t k, std::int32_t lhs_offset,  in gemm_i32_1_1_6()
24606                     std::int32_t rhs_offset, std::int32_t* result,  in gemm_i32_1_1_6()
24607                     std::int32_t result_stride) {  in gemm_i32_1_1_6()
24608   const std::int32_t row_chunks = m / 3;  in gemm_i32_1_1_6()
24609   const std::int32_t col_chunks = n / 3;  in gemm_i32_1_1_6()
24610   const std::int32_t padded_k = ((k + 7) / 8) * 8;  in gemm_i32_1_1_6()
24611   const std::int32_t chunk_size = k * 3;  in gemm_i32_1_1_6()
24612   const std::int32_t zipped_chunk_size = (padded_k + 16) * 3;  in gemm_i32_1_1_6()
24613   const std::int32_t zipped_rhs_size = (padded_k + 16) * n;  in gemm_i32_1_1_6()
24617   std::int32_t* zipped_lhs_3_offsets =  in gemm_i32_1_1_6()
24618       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 3);  in gemm_i32_1_1_6()
24619   std::int32_t* zipped_lhs_1_offsets =  in gemm_i32_1_1_6()
24620       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 1);  in gemm_i32_1_1_6()
24623   const std::int32_t result_chunk_stride = result_stride * 3;  in gemm_i32_1_1_6()
24625   const std::int32_t const_offset = lhs_offset * rhs_offset * k;  in gemm_i32_1_1_6()
24626   std::int32_t* result_chunk = result;  in gemm_i32_1_1_6()
24627   std::int32_t* mul_result_chunk = result;  in gemm_i32_1_1_6()
24628   const std::int32_t mul_result_chunk_stride_bytes = result_stride * 4;  in gemm_i32_1_1_6()
24671                     const std::uint8_t* rhs, std::int32_t m, std::int32_t n,  in gemm_i32_1_1_7()
24672                     std::int32_t k, std::int32_t lhs_offset,  in gemm_i32_1_1_7()
24673                     std::int32_t rhs_offset, std::int32_t* result,  in gemm_i32_1_1_7()
24674                     std::int32_t result_stride) {  in gemm_i32_1_1_7()
24675   const std::int32_t row_chunks = m / 3;  in gemm_i32_1_1_7()
24676   const std::int32_t col_chunks = n / 3;  in gemm_i32_1_1_7()
24677   const std::int32_t padded_k = ((k + 7) / 8) * 8;  in gemm_i32_1_1_7()
24678   const std::int32_t chunk_size = k * 3;  in gemm_i32_1_1_7()
24679   const std::int32_t zipped_chunk_size = (padded_k + 16) * 3;  in gemm_i32_1_1_7()
24680   const std::int32_t zipped_rhs_size = (padded_k + 16) * n;  in gemm_i32_1_1_7()
24684   std::int32_t* zipped_lhs_3_offsets =  in gemm_i32_1_1_7()
24685       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 3);  in gemm_i32_1_1_7()
24686   std::int32_t* zipped_lhs_1_offsets =  in gemm_i32_1_1_7()
24687       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 1);  in gemm_i32_1_1_7()
24690   const std::int32_t result_chunk_stride = result_stride * 3;  in gemm_i32_1_1_7()
24692   const std::int32_t const_offset = lhs_offset * rhs_offset * k;  in gemm_i32_1_1_7()
24693   std::int32_t* result_chunk = result;  in gemm_i32_1_1_7()
24694   std::int32_t* mul_result_chunk = result;  in gemm_i32_1_1_7()
24695   const std::int32_t mul_result_chunk_stride_bytes = result_stride * 4;  in gemm_i32_1_1_7()
24738                     const std::uint8_t* rhs, std::int32_t m, std::int32_t n,  in gemm_i32_1_2_0()
24739                     std::int32_t k, std::int32_t lhs_offset,  in gemm_i32_1_2_0()
24740                     std::int32_t rhs_offset, std::int32_t* result,  in gemm_i32_1_2_0()
24741                     std::int32_t result_stride) {  in gemm_i32_1_2_0()
24742   const std::int32_t row_chunks = m / 3;  in gemm_i32_1_2_0()
24743   const std::int32_t col_chunks = n / 3;  in gemm_i32_1_2_0()
24744   const std::int32_t padded_k = ((k + 7) / 8) * 8;  in gemm_i32_1_2_0()
24745   const std::int32_t chunk_size = k * 3;  in gemm_i32_1_2_0()
24746   const std::int32_t zipped_chunk_size = (padded_k + 16) * 3;  in gemm_i32_1_2_0()
24747   const std::int32_t zipped_rhs_size = (padded_k + 16) * n;  in gemm_i32_1_2_0()
24751   std::int32_t* zipped_lhs_3_offsets =  in gemm_i32_1_2_0()
24752       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 3);  in gemm_i32_1_2_0()
24753   std::int32_t* zipped_lhs_1_offsets =  in gemm_i32_1_2_0()
24754       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 1);  in gemm_i32_1_2_0()
24757   const std::int32_t result_chunk_stride = result_stride * 3;  in gemm_i32_1_2_0()
24759   const std::int32_t const_offset = lhs_offset * rhs_offset * k;  in gemm_i32_1_2_0()
24760   std::int32_t* result_chunk = result;  in gemm_i32_1_2_0()
24761   std::int32_t* mul_result_chunk = result;  in gemm_i32_1_2_0()
24762   const std::int32_t mul_result_chunk_stride_bytes = result_stride * 4;  in gemm_i32_1_2_0()
24805                     const std::uint8_t* rhs, std::int32_t m, std::int32_t n,  in gemm_i32_1_2_1()
24806                     std::int32_t k, std::int32_t lhs_offset,  in gemm_i32_1_2_1()
24807                     std::int32_t rhs_offset, std::int32_t* result,  in gemm_i32_1_2_1()
24808                     std::int32_t result_stride) {  in gemm_i32_1_2_1()
24809   const std::int32_t row_chunks = m / 3;  in gemm_i32_1_2_1()
24810   const std::int32_t col_chunks = n / 3;  in gemm_i32_1_2_1()
24811   const std::int32_t padded_k = ((k + 7) / 8) * 8;  in gemm_i32_1_2_1()
24812   const std::int32_t chunk_size = k * 3;  in gemm_i32_1_2_1()
24813   const std::int32_t zipped_chunk_size = (padded_k + 16) * 3;  in gemm_i32_1_2_1()
24814   const std::int32_t zipped_rhs_size = (padded_k + 16) * n;  in gemm_i32_1_2_1()
24818   std::int32_t* zipped_lhs_3_offsets =  in gemm_i32_1_2_1()
24819       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 3);  in gemm_i32_1_2_1()
24820   std::int32_t* zipped_lhs_1_offsets =  in gemm_i32_1_2_1()
24821       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 1);  in gemm_i32_1_2_1()
24824   const std::int32_t result_chunk_stride = result_stride * 3;  in gemm_i32_1_2_1()
24826   const std::int32_t const_offset = lhs_offset * rhs_offset * k;  in gemm_i32_1_2_1()
24827   std::int32_t* result_chunk = result;  in gemm_i32_1_2_1()
24828   std::int32_t* mul_result_chunk = result;  in gemm_i32_1_2_1()
24829   const std::int32_t mul_result_chunk_stride_bytes = result_stride * 4;  in gemm_i32_1_2_1()
24872                     const std::uint8_t* rhs, std::int32_t m, std::int32_t n,  in gemm_i32_1_2_2()
24873                     std::int32_t k, std::int32_t lhs_offset,  in gemm_i32_1_2_2()
24874                     std::int32_t rhs_offset, std::int32_t* result,  in gemm_i32_1_2_2()
24875                     std::int32_t result_stride) {  in gemm_i32_1_2_2()
24876   const std::int32_t row_chunks = m / 3;  in gemm_i32_1_2_2()
24877   const std::int32_t col_chunks = n / 3;  in gemm_i32_1_2_2()
24878   const std::int32_t padded_k = ((k + 7) / 8) * 8;  in gemm_i32_1_2_2()
24879   const std::int32_t chunk_size = k * 3;  in gemm_i32_1_2_2()
24880   const std::int32_t zipped_chunk_size = (padded_k + 16) * 3;  in gemm_i32_1_2_2()
24881   const std::int32_t zipped_rhs_size = (padded_k + 16) * n;  in gemm_i32_1_2_2()
24885   std::int32_t* zipped_lhs_3_offsets =  in gemm_i32_1_2_2()
24886       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 3);  in gemm_i32_1_2_2()
24887   std::int32_t* zipped_lhs_1_offsets =  in gemm_i32_1_2_2()
24888       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 1);  in gemm_i32_1_2_2()
24891   const std::int32_t result_chunk_stride = result_stride * 3;  in gemm_i32_1_2_2()
24893   const std::int32_t const_offset = lhs_offset * rhs_offset * k;  in gemm_i32_1_2_2()
24894   std::int32_t* result_chunk = result;  in gemm_i32_1_2_2()
24895   std::int32_t* mul_result_chunk = result;  in gemm_i32_1_2_2()
24896   const std::int32_t mul_result_chunk_stride_bytes = result_stride * 4;  in gemm_i32_1_2_2()
24939                     const std::uint8_t* rhs, std::int32_t m, std::int32_t n,  in gemm_i32_1_2_3()
24940                     std::int32_t k, std::int32_t lhs_offset,  in gemm_i32_1_2_3()
24941                     std::int32_t rhs_offset, std::int32_t* result,  in gemm_i32_1_2_3()
24942                     std::int32_t result_stride) {  in gemm_i32_1_2_3()
24943   const std::int32_t row_chunks = m / 3;  in gemm_i32_1_2_3()
24944   const std::int32_t col_chunks = n / 3;  in gemm_i32_1_2_3()
24945   const std::int32_t padded_k = ((k + 7) / 8) * 8;  in gemm_i32_1_2_3()
24946   const std::int32_t chunk_size = k * 3;  in gemm_i32_1_2_3()
24947   const std::int32_t zipped_chunk_size = (padded_k + 16) * 3;  in gemm_i32_1_2_3()
24948   const std::int32_t zipped_rhs_size = (padded_k + 16) * n;  in gemm_i32_1_2_3()
24952   std::int32_t* zipped_lhs_3_offsets =  in gemm_i32_1_2_3()
24953       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 3);  in gemm_i32_1_2_3()
24954   std::int32_t* zipped_lhs_1_offsets =  in gemm_i32_1_2_3()
24955       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 1);  in gemm_i32_1_2_3()
24958   const std::int32_t result_chunk_stride = result_stride * 3;  in gemm_i32_1_2_3()
24960   const std::int32_t const_offset = lhs_offset * rhs_offset * k;  in gemm_i32_1_2_3()
24961   std::int32_t* result_chunk = result;  in gemm_i32_1_2_3()
24962   std::int32_t* mul_result_chunk = result;  in gemm_i32_1_2_3()
24963   const std::int32_t mul_result_chunk_stride_bytes = result_stride * 4;  in gemm_i32_1_2_3()
25006                     const std::uint8_t* rhs, std::int32_t m, std::int32_t n,  in gemm_i32_1_2_4()
25007                     std::int32_t k, std::int32_t lhs_offset,  in gemm_i32_1_2_4()
25008                     std::int32_t rhs_offset, std::int32_t* result,  in gemm_i32_1_2_4()
25009                     std::int32_t result_stride) {  in gemm_i32_1_2_4()
25010   const std::int32_t row_chunks = m / 3;  in gemm_i32_1_2_4()
25011   const std::int32_t col_chunks = n / 3;  in gemm_i32_1_2_4()
25012   const std::int32_t padded_k = ((k + 7) / 8) * 8;  in gemm_i32_1_2_4()
25013   const std::int32_t chunk_size = k * 3;  in gemm_i32_1_2_4()
25014   const std::int32_t zipped_chunk_size = (padded_k + 16) * 3;  in gemm_i32_1_2_4()
25015   const std::int32_t zipped_rhs_size = (padded_k + 16) * n;  in gemm_i32_1_2_4()
25019   std::int32_t* zipped_lhs_3_offsets =  in gemm_i32_1_2_4()
25020       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 3);  in gemm_i32_1_2_4()
25021   std::int32_t* zipped_lhs_1_offsets =  in gemm_i32_1_2_4()
25022       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 1);  in gemm_i32_1_2_4()
25025   const std::int32_t result_chunk_stride = result_stride * 3;  in gemm_i32_1_2_4()
25027   const std::int32_t const_offset = lhs_offset * rhs_offset * k;  in gemm_i32_1_2_4()
25028   std::int32_t* result_chunk = result;  in gemm_i32_1_2_4()
25029   std::int32_t* mul_result_chunk = result;  in gemm_i32_1_2_4()
25030   const std::int32_t mul_result_chunk_stride_bytes = result_stride * 4;  in gemm_i32_1_2_4()
25073                     const std::uint8_t* rhs, std::int32_t m, std::int32_t n,  in gemm_i32_1_2_5()
25074                     std::int32_t k, std::int32_t lhs_offset,  in gemm_i32_1_2_5()
25075                     std::int32_t rhs_offset, std::int32_t* result,  in gemm_i32_1_2_5()
25076                     std::int32_t result_stride) {  in gemm_i32_1_2_5()
25077   const std::int32_t row_chunks = m / 3;  in gemm_i32_1_2_5()
25078   const std::int32_t col_chunks = n / 3;  in gemm_i32_1_2_5()
25079   const std::int32_t padded_k = ((k + 7) / 8) * 8;  in gemm_i32_1_2_5()
25080   const std::int32_t chunk_size = k * 3;  in gemm_i32_1_2_5()
25081   const std::int32_t zipped_chunk_size = (padded_k + 16) * 3;  in gemm_i32_1_2_5()
25082   const std::int32_t zipped_rhs_size = (padded_k + 16) * n;  in gemm_i32_1_2_5()
25086   std::int32_t* zipped_lhs_3_offsets =  in gemm_i32_1_2_5()
25087       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 3);  in gemm_i32_1_2_5()
25088   std::int32_t* zipped_lhs_1_offsets =  in gemm_i32_1_2_5()
25089       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 1);  in gemm_i32_1_2_5()
25092   const std::int32_t result_chunk_stride = result_stride * 3;  in gemm_i32_1_2_5()
25094   const std::int32_t const_offset = lhs_offset * rhs_offset * k;  in gemm_i32_1_2_5()
25095   std::int32_t* result_chunk = result;  in gemm_i32_1_2_5()
25096   std::int32_t* mul_result_chunk = result;  in gemm_i32_1_2_5()
25097   const std::int32_t mul_result_chunk_stride_bytes = result_stride * 4;  in gemm_i32_1_2_5()
25140                     const std::uint8_t* rhs, std::int32_t m, std::int32_t n,  in gemm_i32_1_2_6()
25141                     std::int32_t k, std::int32_t lhs_offset,  in gemm_i32_1_2_6()
25142                     std::int32_t rhs_offset, std::int32_t* result,  in gemm_i32_1_2_6()
25143                     std::int32_t result_stride) {  in gemm_i32_1_2_6()
25144   const std::int32_t row_chunks = m / 3;  in gemm_i32_1_2_6()
25145   const std::int32_t col_chunks = n / 3;  in gemm_i32_1_2_6()
25146   const std::int32_t padded_k = ((k + 7) / 8) * 8;  in gemm_i32_1_2_6()
25147   const std::int32_t chunk_size = k * 3;  in gemm_i32_1_2_6()
25148   const std::int32_t zipped_chunk_size = (padded_k + 16) * 3;  in gemm_i32_1_2_6()
25149   const std::int32_t zipped_rhs_size = (padded_k + 16) * n;  in gemm_i32_1_2_6()
25153   std::int32_t* zipped_lhs_3_offsets =  in gemm_i32_1_2_6()
25154       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 3);  in gemm_i32_1_2_6()
25155   std::int32_t* zipped_lhs_1_offsets =  in gemm_i32_1_2_6()
25156       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 1);  in gemm_i32_1_2_6()
25159   const std::int32_t result_chunk_stride = result_stride * 3;  in gemm_i32_1_2_6()
25161   const std::int32_t const_offset = lhs_offset * rhs_offset * k;  in gemm_i32_1_2_6()
25162   std::int32_t* result_chunk = result;  in gemm_i32_1_2_6()
25163   std::int32_t* mul_result_chunk = result;  in gemm_i32_1_2_6()
25164   const std::int32_t mul_result_chunk_stride_bytes = result_stride * 4;  in gemm_i32_1_2_6()
25207                     const std::uint8_t* rhs, std::int32_t m, std::int32_t n,  in gemm_i32_1_2_7()
25208                     std::int32_t k, std::int32_t lhs_offset,  in gemm_i32_1_2_7()
25209                     std::int32_t rhs_offset, std::int32_t* result,  in gemm_i32_1_2_7()
25210                     std::int32_t result_stride) {  in gemm_i32_1_2_7()
25211   const std::int32_t row_chunks = m / 3;  in gemm_i32_1_2_7()
25212   const std::int32_t col_chunks = n / 3;  in gemm_i32_1_2_7()
25213   const std::int32_t padded_k = ((k + 7) / 8) * 8;  in gemm_i32_1_2_7()
25214   const std::int32_t chunk_size = k * 3;  in gemm_i32_1_2_7()
25215   const std::int32_t zipped_chunk_size = (padded_k + 16) * 3;  in gemm_i32_1_2_7()
25216   const std::int32_t zipped_rhs_size = (padded_k + 16) * n;  in gemm_i32_1_2_7()
25220   std::int32_t* zipped_lhs_3_offsets =  in gemm_i32_1_2_7()
25221       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 3);  in gemm_i32_1_2_7()
25222   std::int32_t* zipped_lhs_1_offsets =  in gemm_i32_1_2_7()
25223       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 1);  in gemm_i32_1_2_7()
25226   const std::int32_t result_chunk_stride = result_stride * 3;  in gemm_i32_1_2_7()
25228   const std::int32_t const_offset = lhs_offset * rhs_offset * k;  in gemm_i32_1_2_7()
25229   std::int32_t* result_chunk = result;  in gemm_i32_1_2_7()
25230   std::int32_t* mul_result_chunk = result;  in gemm_i32_1_2_7()
25231   const std::int32_t mul_result_chunk_stride_bytes = result_stride * 4;  in gemm_i32_1_2_7()
25274                     const std::uint8_t* rhs, std::int32_t m, std::int32_t n,  in gemm_i32_2_0_0()
25275                     std::int32_t k, std::int32_t lhs_offset,  in gemm_i32_2_0_0()
25276                     std::int32_t rhs_offset, std::int32_t* result,  in gemm_i32_2_0_0()
25277                     std::int32_t result_stride) {  in gemm_i32_2_0_0()
25278   const std::int32_t row_chunks = m / 3;  in gemm_i32_2_0_0()
25279   const std::int32_t col_chunks = n / 3;  in gemm_i32_2_0_0()
25280   const std::int32_t padded_k = ((k + 7) / 8) * 8;  in gemm_i32_2_0_0()
25281   const std::int32_t chunk_size = k * 3;  in gemm_i32_2_0_0()
25282   const std::int32_t zipped_chunk_size = (padded_k + 16) * 3;  in gemm_i32_2_0_0()
25283   const std::int32_t zipped_rhs_size = (padded_k + 16) * n;  in gemm_i32_2_0_0()
25287   std::int32_t* zipped_lhs_3_offsets =  in gemm_i32_2_0_0()
25288       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 3);  in gemm_i32_2_0_0()
25289   std::int32_t* zipped_lhs_2_offsets =  in gemm_i32_2_0_0()
25290       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 2);  in gemm_i32_2_0_0()
25293   const std::int32_t result_chunk_stride = result_stride * 3;  in gemm_i32_2_0_0()
25295   const std::int32_t const_offset = lhs_offset * rhs_offset * k;  in gemm_i32_2_0_0()
25296   std::int32_t* result_chunk = result;  in gemm_i32_2_0_0()
25297   std::int32_t* mul_result_chunk = result;  in gemm_i32_2_0_0()
25298   const std::int32_t mul_result_chunk_stride_bytes = result_stride * 4;  in gemm_i32_2_0_0()
25334                     const std::uint8_t* rhs, std::int32_t m, std::int32_t n,  in gemm_i32_2_0_1()
25335                     std::int32_t k, std::int32_t lhs_offset,  in gemm_i32_2_0_1()
25336                     std::int32_t rhs_offset, std::int32_t* result,  in gemm_i32_2_0_1()
25337                     std::int32_t result_stride) {  in gemm_i32_2_0_1()
25338   const std::int32_t row_chunks = m / 3;  in gemm_i32_2_0_1()
25339   const std::int32_t col_chunks = n / 3;  in gemm_i32_2_0_1()
25340   const std::int32_t padded_k = ((k + 7) / 8) * 8;  in gemm_i32_2_0_1()
25341   const std::int32_t chunk_size = k * 3;  in gemm_i32_2_0_1()
25342   const std::int32_t zipped_chunk_size = (padded_k + 16) * 3;  in gemm_i32_2_0_1()
25343   const std::int32_t zipped_rhs_size = (padded_k + 16) * n;  in gemm_i32_2_0_1()
25347   std::int32_t* zipped_lhs_3_offsets =  in gemm_i32_2_0_1()
25348       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 3);  in gemm_i32_2_0_1()
25349   std::int32_t* zipped_lhs_2_offsets =  in gemm_i32_2_0_1()
25350       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 2);  in gemm_i32_2_0_1()
25353   const std::int32_t result_chunk_stride = result_stride * 3;  in gemm_i32_2_0_1()
25355   const std::int32_t const_offset = lhs_offset * rhs_offset * k;  in gemm_i32_2_0_1()
25356   std::int32_t* result_chunk = result;  in gemm_i32_2_0_1()
25357   std::int32_t* mul_result_chunk = result;  in gemm_i32_2_0_1()
25358   const std::int32_t mul_result_chunk_stride_bytes = result_stride * 4;  in gemm_i32_2_0_1()
25394                     const std::uint8_t* rhs, std::int32_t m, std::int32_t n,  in gemm_i32_2_0_2()
25395                     std::int32_t k, std::int32_t lhs_offset,  in gemm_i32_2_0_2()
25396                     std::int32_t rhs_offset, std::int32_t* result,  in gemm_i32_2_0_2()
25397                     std::int32_t result_stride) {  in gemm_i32_2_0_2()
25398   const std::int32_t row_chunks = m / 3;  in gemm_i32_2_0_2()
25399   const std::int32_t col_chunks = n / 3;  in gemm_i32_2_0_2()
25400   const std::int32_t padded_k = ((k + 7) / 8) * 8;  in gemm_i32_2_0_2()
25401   const std::int32_t chunk_size = k * 3;  in gemm_i32_2_0_2()
25402   const std::int32_t zipped_chunk_size = (padded_k + 16) * 3;  in gemm_i32_2_0_2()
25403   const std::int32_t zipped_rhs_size = (padded_k + 16) * n;  in gemm_i32_2_0_2()
25407   std::int32_t* zipped_lhs_3_offsets =  in gemm_i32_2_0_2()
25408       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 3);  in gemm_i32_2_0_2()
25409   std::int32_t* zipped_lhs_2_offsets =  in gemm_i32_2_0_2()
25410       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 2);  in gemm_i32_2_0_2()
25413   const std::int32_t result_chunk_stride = result_stride * 3;  in gemm_i32_2_0_2()
25415   const std::int32_t const_offset = lhs_offset * rhs_offset * k;  in gemm_i32_2_0_2()
25416   std::int32_t* result_chunk = result;  in gemm_i32_2_0_2()
25417   std::int32_t* mul_result_chunk = result;  in gemm_i32_2_0_2()
25418   const std::int32_t mul_result_chunk_stride_bytes = result_stride * 4;  in gemm_i32_2_0_2()
25454                     const std::uint8_t* rhs, std::int32_t m, std::int32_t n,  in gemm_i32_2_0_3()
25455                     std::int32_t k, std::int32_t lhs_offset,  in gemm_i32_2_0_3()
25456                     std::int32_t rhs_offset, std::int32_t* result,  in gemm_i32_2_0_3()
25457                     std::int32_t result_stride) {  in gemm_i32_2_0_3()
25458   const std::int32_t row_chunks = m / 3;  in gemm_i32_2_0_3()
25459   const std::int32_t col_chunks = n / 3;  in gemm_i32_2_0_3()
25460   const std::int32_t padded_k = ((k + 7) / 8) * 8;  in gemm_i32_2_0_3()
25461   const std::int32_t chunk_size = k * 3;  in gemm_i32_2_0_3()
25462   const std::int32_t zipped_chunk_size = (padded_k + 16) * 3;  in gemm_i32_2_0_3()
25463   const std::int32_t zipped_rhs_size = (padded_k + 16) * n;  in gemm_i32_2_0_3()
25467   std::int32_t* zipped_lhs_3_offsets =  in gemm_i32_2_0_3()
25468       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 3);  in gemm_i32_2_0_3()
25469   std::int32_t* zipped_lhs_2_offsets =  in gemm_i32_2_0_3()
25470       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 2);  in gemm_i32_2_0_3()
25473   const std::int32_t result_chunk_stride = result_stride * 3;  in gemm_i32_2_0_3()
25475   const std::int32_t const_offset = lhs_offset * rhs_offset * k;  in gemm_i32_2_0_3()
25476   std::int32_t* result_chunk = result;  in gemm_i32_2_0_3()
25477   std::int32_t* mul_result_chunk = result;  in gemm_i32_2_0_3()
25478   const std::int32_t mul_result_chunk_stride_bytes = result_stride * 4;  in gemm_i32_2_0_3()
25514                     const std::uint8_t* rhs, std::int32_t m, std::int32_t n,  in gemm_i32_2_0_4()
25515                     std::int32_t k, std::int32_t lhs_offset,  in gemm_i32_2_0_4()
25516                     std::int32_t rhs_offset, std::int32_t* result,  in gemm_i32_2_0_4()
25517                     std::int32_t result_stride) {  in gemm_i32_2_0_4()
25518   const std::int32_t row_chunks = m / 3;  in gemm_i32_2_0_4()
25519   const std::int32_t col_chunks = n / 3;  in gemm_i32_2_0_4()
25520   const std::int32_t padded_k = ((k + 7) / 8) * 8;  in gemm_i32_2_0_4()
25521   const std::int32_t chunk_size = k * 3;  in gemm_i32_2_0_4()
25522   const std::int32_t zipped_chunk_size = (padded_k + 16) * 3;  in gemm_i32_2_0_4()
25523   const std::int32_t zipped_rhs_size = (padded_k + 16) * n;  in gemm_i32_2_0_4()
25527   std::int32_t* zipped_lhs_3_offsets =  in gemm_i32_2_0_4()
25528       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 3);  in gemm_i32_2_0_4()
25529   std::int32_t* zipped_lhs_2_offsets =  in gemm_i32_2_0_4()
25530       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 2);  in gemm_i32_2_0_4()
25533   const std::int32_t result_chunk_stride = result_stride * 3;  in gemm_i32_2_0_4()
25535   const std::int32_t const_offset = lhs_offset * rhs_offset * k;  in gemm_i32_2_0_4()
25536   std::int32_t* result_chunk = result;  in gemm_i32_2_0_4()
25537   std::int32_t* mul_result_chunk = result;  in gemm_i32_2_0_4()
25538   const std::int32_t mul_result_chunk_stride_bytes = result_stride * 4;  in gemm_i32_2_0_4()
25574                     const std::uint8_t* rhs, std::int32_t m, std::int32_t n,  in gemm_i32_2_0_5()
25575                     std::int32_t k, std::int32_t lhs_offset,  in gemm_i32_2_0_5()
25576                     std::int32_t rhs_offset, std::int32_t* result,  in gemm_i32_2_0_5()
25577                     std::int32_t result_stride) {  in gemm_i32_2_0_5()
25578   const std::int32_t row_chunks = m / 3;  in gemm_i32_2_0_5()
25579   const std::int32_t col_chunks = n / 3;  in gemm_i32_2_0_5()
25580   const std::int32_t padded_k = ((k + 7) / 8) * 8;  in gemm_i32_2_0_5()
25581   const std::int32_t chunk_size = k * 3;  in gemm_i32_2_0_5()
25582   const std::int32_t zipped_chunk_size = (padded_k + 16) * 3;  in gemm_i32_2_0_5()
25583   const std::int32_t zipped_rhs_size = (padded_k + 16) * n;  in gemm_i32_2_0_5()
25587   std::int32_t* zipped_lhs_3_offsets =  in gemm_i32_2_0_5()
25588       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 3);  in gemm_i32_2_0_5()
25589   std::int32_t* zipped_lhs_2_offsets =  in gemm_i32_2_0_5()
25590       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 2);  in gemm_i32_2_0_5()
25593   const std::int32_t result_chunk_stride = result_stride * 3;  in gemm_i32_2_0_5()
25595   const std::int32_t const_offset = lhs_offset * rhs_offset * k;  in gemm_i32_2_0_5()
25596   std::int32_t* result_chunk = result;  in gemm_i32_2_0_5()
25597   std::int32_t* mul_result_chunk = result;  in gemm_i32_2_0_5()
25598   const std::int32_t mul_result_chunk_stride_bytes = result_stride * 4;  in gemm_i32_2_0_5()
25634                     const std::uint8_t* rhs, std::int32_t m, std::int32_t n,  in gemm_i32_2_0_6()
25635                     std::int32_t k, std::int32_t lhs_offset,  in gemm_i32_2_0_6()
25636                     std::int32_t rhs_offset, std::int32_t* result,  in gemm_i32_2_0_6()
25637                     std::int32_t result_stride) {  in gemm_i32_2_0_6()
25638   const std::int32_t row_chunks = m / 3;  in gemm_i32_2_0_6()
25639   const std::int32_t col_chunks = n / 3;  in gemm_i32_2_0_6()
25640   const std::int32_t padded_k = ((k + 7) / 8) * 8;  in gemm_i32_2_0_6()
25641   const std::int32_t chunk_size = k * 3;  in gemm_i32_2_0_6()
25642   const std::int32_t zipped_chunk_size = (padded_k + 16) * 3;  in gemm_i32_2_0_6()
25643   const std::int32_t zipped_rhs_size = (padded_k + 16) * n;  in gemm_i32_2_0_6()
25647   std::int32_t* zipped_lhs_3_offsets =  in gemm_i32_2_0_6()
25648       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 3);  in gemm_i32_2_0_6()
25649   std::int32_t* zipped_lhs_2_offsets =  in gemm_i32_2_0_6()
25650       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 2);  in gemm_i32_2_0_6()
25653   const std::int32_t result_chunk_stride = result_stride * 3;  in gemm_i32_2_0_6()
25655   const std::int32_t const_offset = lhs_offset * rhs_offset * k;  in gemm_i32_2_0_6()
25656   std::int32_t* result_chunk = result;  in gemm_i32_2_0_6()
25657   std::int32_t* mul_result_chunk = result;  in gemm_i32_2_0_6()
25658   const std::int32_t mul_result_chunk_stride_bytes = result_stride * 4;  in gemm_i32_2_0_6()
25694                     const std::uint8_t* rhs, std::int32_t m, std::int32_t n,  in gemm_i32_2_0_7()
25695                     std::int32_t k, std::int32_t lhs_offset,  in gemm_i32_2_0_7()
25696                     std::int32_t rhs_offset, std::int32_t* result,  in gemm_i32_2_0_7()
25697                     std::int32_t result_stride) {  in gemm_i32_2_0_7()
25698   const std::int32_t row_chunks = m / 3;  in gemm_i32_2_0_7()
25699   const std::int32_t col_chunks = n / 3;  in gemm_i32_2_0_7()
25700   const std::int32_t padded_k = ((k + 7) / 8) * 8;  in gemm_i32_2_0_7()
25701   const std::int32_t chunk_size = k * 3;  in gemm_i32_2_0_7()
25702   const std::int32_t zipped_chunk_size = (padded_k + 16) * 3;  in gemm_i32_2_0_7()
25703   const std::int32_t zipped_rhs_size = (padded_k + 16) * n;  in gemm_i32_2_0_7()
25707   std::int32_t* zipped_lhs_3_offsets =  in gemm_i32_2_0_7()
25708       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 3);  in gemm_i32_2_0_7()
25709   std::int32_t* zipped_lhs_2_offsets =  in gemm_i32_2_0_7()
25710       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 2);  in gemm_i32_2_0_7()
25713   const std::int32_t result_chunk_stride = result_stride * 3;  in gemm_i32_2_0_7()
25715   const std::int32_t const_offset = lhs_offset * rhs_offset * k;  in gemm_i32_2_0_7()
25716   std::int32_t* result_chunk = result;  in gemm_i32_2_0_7()
25717   std::int32_t* mul_result_chunk = result;  in gemm_i32_2_0_7()
25718   const std::int32_t mul_result_chunk_stride_bytes = result_stride * 4;  in gemm_i32_2_0_7()
25754                     const std::uint8_t* rhs, std::int32_t m, std::int32_t n,  in gemm_i32_2_1_0()
25755                     std::int32_t k, std::int32_t lhs_offset,  in gemm_i32_2_1_0()
25756                     std::int32_t rhs_offset, std::int32_t* result,  in gemm_i32_2_1_0()
25757                     std::int32_t result_stride) {  in gemm_i32_2_1_0()
25758   const std::int32_t row_chunks = m / 3;  in gemm_i32_2_1_0()
25759   const std::int32_t col_chunks = n / 3;  in gemm_i32_2_1_0()
25760   const std::int32_t padded_k = ((k + 7) / 8) * 8;  in gemm_i32_2_1_0()
25761   const std::int32_t chunk_size = k * 3;  in gemm_i32_2_1_0()
25762   const std::int32_t zipped_chunk_size = (padded_k + 16) * 3;  in gemm_i32_2_1_0()
25763   const std::int32_t zipped_rhs_size = (padded_k + 16) * n;  in gemm_i32_2_1_0()
25767   std::int32_t* zipped_lhs_3_offsets =  in gemm_i32_2_1_0()
25768       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 3);  in gemm_i32_2_1_0()
25769   std::int32_t* zipped_lhs_2_offsets =  in gemm_i32_2_1_0()
25770       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 2);  in gemm_i32_2_1_0()
25773   const std::int32_t result_chunk_stride = result_stride * 3;  in gemm_i32_2_1_0()
25775   const std::int32_t const_offset = lhs_offset * rhs_offset * k;  in gemm_i32_2_1_0()
25776   std::int32_t* result_chunk = result;  in gemm_i32_2_1_0()
25777   std::int32_t* mul_result_chunk = result;  in gemm_i32_2_1_0()
25778   const std::int32_t mul_result_chunk_stride_bytes = result_stride * 4;  in gemm_i32_2_1_0()
25821                     const std::uint8_t* rhs, std::int32_t m, std::int32_t n,  in gemm_i32_2_1_1()
25822                     std::int32_t k, std::int32_t lhs_offset,  in gemm_i32_2_1_1()
25823                     std::int32_t rhs_offset, std::int32_t* result,  in gemm_i32_2_1_1()
25824                     std::int32_t result_stride) {  in gemm_i32_2_1_1()
25825   const std::int32_t row_chunks = m / 3;  in gemm_i32_2_1_1()
25826   const std::int32_t col_chunks = n / 3;  in gemm_i32_2_1_1()
25827   const std::int32_t padded_k = ((k + 7) / 8) * 8;  in gemm_i32_2_1_1()
25828   const std::int32_t chunk_size = k * 3;  in gemm_i32_2_1_1()
25829   const std::int32_t zipped_chunk_size = (padded_k + 16) * 3;  in gemm_i32_2_1_1()
25830   const std::int32_t zipped_rhs_size = (padded_k + 16) * n;  in gemm_i32_2_1_1()
25834   std::int32_t* zipped_lhs_3_offsets =  in gemm_i32_2_1_1()
25835       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 3);  in gemm_i32_2_1_1()
25836   std::int32_t* zipped_lhs_2_offsets =  in gemm_i32_2_1_1()
25837       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 2);  in gemm_i32_2_1_1()
25840   const std::int32_t result_chunk_stride = result_stride * 3;  in gemm_i32_2_1_1()
25842   const std::int32_t const_offset = lhs_offset * rhs_offset * k;  in gemm_i32_2_1_1()
25843   std::int32_t* result_chunk = result;  in gemm_i32_2_1_1()
25844   std::int32_t* mul_result_chunk = result;  in gemm_i32_2_1_1()
25845   const std::int32_t mul_result_chunk_stride_bytes = result_stride * 4;  in gemm_i32_2_1_1()
25888                     const std::uint8_t* rhs, std::int32_t m, std::int32_t n,  in gemm_i32_2_1_2()
25889                     std::int32_t k, std::int32_t lhs_offset,  in gemm_i32_2_1_2()
25890                     std::int32_t rhs_offset, std::int32_t* result,  in gemm_i32_2_1_2()
25891                     std::int32_t result_stride) {  in gemm_i32_2_1_2()
25892   const std::int32_t row_chunks = m / 3;  in gemm_i32_2_1_2()
25893   const std::int32_t col_chunks = n / 3;  in gemm_i32_2_1_2()
25894   const std::int32_t padded_k = ((k + 7) / 8) * 8;  in gemm_i32_2_1_2()
25895   const std::int32_t chunk_size = k * 3;  in gemm_i32_2_1_2()
25896   const std::int32_t zipped_chunk_size = (padded_k + 16) * 3;  in gemm_i32_2_1_2()
25897   const std::int32_t zipped_rhs_size = (padded_k + 16) * n;  in gemm_i32_2_1_2()
25901   std::int32_t* zipped_lhs_3_offsets =  in gemm_i32_2_1_2()
25902       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 3);  in gemm_i32_2_1_2()
25903   std::int32_t* zipped_lhs_2_offsets =  in gemm_i32_2_1_2()
25904       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 2);  in gemm_i32_2_1_2()
25907   const std::int32_t result_chunk_stride = result_stride * 3;  in gemm_i32_2_1_2()
25909   const std::int32_t const_offset = lhs_offset * rhs_offset * k;  in gemm_i32_2_1_2()
25910   std::int32_t* result_chunk = result;  in gemm_i32_2_1_2()
25911   std::int32_t* mul_result_chunk = result;  in gemm_i32_2_1_2()
25912   const std::int32_t mul_result_chunk_stride_bytes = result_stride * 4;  in gemm_i32_2_1_2()
25955                     const std::uint8_t* rhs, std::int32_t m, std::int32_t n,  in gemm_i32_2_1_3()
25956                     std::int32_t k, std::int32_t lhs_offset,  in gemm_i32_2_1_3()
25957                     std::int32_t rhs_offset, std::int32_t* result,  in gemm_i32_2_1_3()
25958                     std::int32_t result_stride) {  in gemm_i32_2_1_3()
25959   const std::int32_t row_chunks = m / 3;  in gemm_i32_2_1_3()
25960   const std::int32_t col_chunks = n / 3;  in gemm_i32_2_1_3()
25961   const std::int32_t padded_k = ((k + 7) / 8) * 8;  in gemm_i32_2_1_3()
25962   const std::int32_t chunk_size = k * 3;  in gemm_i32_2_1_3()
25963   const std::int32_t zipped_chunk_size = (padded_k + 16) * 3;  in gemm_i32_2_1_3()
25964   const std::int32_t zipped_rhs_size = (padded_k + 16) * n;  in gemm_i32_2_1_3()
25968   std::int32_t* zipped_lhs_3_offsets =  in gemm_i32_2_1_3()
25969       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 3);  in gemm_i32_2_1_3()
25970   std::int32_t* zipped_lhs_2_offsets =  in gemm_i32_2_1_3()
25971       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 2);  in gemm_i32_2_1_3()
25974   const std::int32_t result_chunk_stride = result_stride * 3;  in gemm_i32_2_1_3()
25976   const std::int32_t const_offset = lhs_offset * rhs_offset * k;  in gemm_i32_2_1_3()
25977   std::int32_t* result_chunk = result;  in gemm_i32_2_1_3()
25978   std::int32_t* mul_result_chunk = result;  in gemm_i32_2_1_3()
25979   const std::int32_t mul_result_chunk_stride_bytes = result_stride * 4;  in gemm_i32_2_1_3()
26022                     const std::uint8_t* rhs, std::int32_t m, std::int32_t n,  in gemm_i32_2_1_4()
26023                     std::int32_t k, std::int32_t lhs_offset,  in gemm_i32_2_1_4()
26024                     std::int32_t rhs_offset, std::int32_t* result,  in gemm_i32_2_1_4()
26025                     std::int32_t result_stride) {  in gemm_i32_2_1_4()
26026   const std::int32_t row_chunks = m / 3;  in gemm_i32_2_1_4()
26027   const std::int32_t col_chunks = n / 3;  in gemm_i32_2_1_4()
26028   const std::int32_t padded_k = ((k + 7) / 8) * 8;  in gemm_i32_2_1_4()
26029   const std::int32_t chunk_size = k * 3;  in gemm_i32_2_1_4()
26030   const std::int32_t zipped_chunk_size = (padded_k + 16) * 3;  in gemm_i32_2_1_4()
26031   const std::int32_t zipped_rhs_size = (padded_k + 16) * n;  in gemm_i32_2_1_4()
26035   std::int32_t* zipped_lhs_3_offsets =  in gemm_i32_2_1_4()
26036       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 3);  in gemm_i32_2_1_4()
26037   std::int32_t* zipped_lhs_2_offsets =  in gemm_i32_2_1_4()
26038       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 2);  in gemm_i32_2_1_4()
26041   const std::int32_t result_chunk_stride = result_stride * 3;  in gemm_i32_2_1_4()
26043   const std::int32_t const_offset = lhs_offset * rhs_offset * k;  in gemm_i32_2_1_4()
26044   std::int32_t* result_chunk = result;  in gemm_i32_2_1_4()
26045   std::int32_t* mul_result_chunk = result;  in gemm_i32_2_1_4()
26046   const std::int32_t mul_result_chunk_stride_bytes = result_stride * 4;  in gemm_i32_2_1_4()
26089                     const std::uint8_t* rhs, std::int32_t m, std::int32_t n,  in gemm_i32_2_1_5()
26090                     std::int32_t k, std::int32_t lhs_offset,  in gemm_i32_2_1_5()
26091                     std::int32_t rhs_offset, std::int32_t* result,  in gemm_i32_2_1_5()
26092                     std::int32_t result_stride) {  in gemm_i32_2_1_5()
26093   const std::int32_t row_chunks = m / 3;  in gemm_i32_2_1_5()
26094   const std::int32_t col_chunks = n / 3;  in gemm_i32_2_1_5()
26095   const std::int32_t padded_k = ((k + 7) / 8) * 8;  in gemm_i32_2_1_5()
26096   const std::int32_t chunk_size = k * 3;  in gemm_i32_2_1_5()
26097   const std::int32_t zipped_chunk_size = (padded_k + 16) * 3;  in gemm_i32_2_1_5()
26098   const std::int32_t zipped_rhs_size = (padded_k + 16) * n;  in gemm_i32_2_1_5()
26102   std::int32_t* zipped_lhs_3_offsets =  in gemm_i32_2_1_5()
26103       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 3);  in gemm_i32_2_1_5()
26104   std::int32_t* zipped_lhs_2_offsets =  in gemm_i32_2_1_5()
26105       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 2);  in gemm_i32_2_1_5()
26108   const std::int32_t result_chunk_stride = result_stride * 3;  in gemm_i32_2_1_5()
26110   const std::int32_t const_offset = lhs_offset * rhs_offset * k;  in gemm_i32_2_1_5()
26111   std::int32_t* result_chunk = result;  in gemm_i32_2_1_5()
26112   std::int32_t* mul_result_chunk = result;  in gemm_i32_2_1_5()
26113   const std::int32_t mul_result_chunk_stride_bytes = result_stride * 4;  in gemm_i32_2_1_5()
26156                     const std::uint8_t* rhs, std::int32_t m, std::int32_t n,  in gemm_i32_2_1_6()
26157                     std::int32_t k, std::int32_t lhs_offset,  in gemm_i32_2_1_6()
26158                     std::int32_t rhs_offset, std::int32_t* result,  in gemm_i32_2_1_6()
26159                     std::int32_t result_stride) {  in gemm_i32_2_1_6()
26160   const std::int32_t row_chunks = m / 3;  in gemm_i32_2_1_6()
26161   const std::int32_t col_chunks = n / 3;  in gemm_i32_2_1_6()
26162   const std::int32_t padded_k = ((k + 7) / 8) * 8;  in gemm_i32_2_1_6()
26163   const std::int32_t chunk_size = k * 3;  in gemm_i32_2_1_6()
26164   const std::int32_t zipped_chunk_size = (padded_k + 16) * 3;  in gemm_i32_2_1_6()
26165   const std::int32_t zipped_rhs_size = (padded_k + 16) * n;  in gemm_i32_2_1_6()
26169   std::int32_t* zipped_lhs_3_offsets =  in gemm_i32_2_1_6()
26170       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 3);  in gemm_i32_2_1_6()
26171   std::int32_t* zipped_lhs_2_offsets =  in gemm_i32_2_1_6()
26172       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 2);  in gemm_i32_2_1_6()
26175   const std::int32_t result_chunk_stride = result_stride * 3;  in gemm_i32_2_1_6()
26177   const std::int32_t const_offset = lhs_offset * rhs_offset * k;  in gemm_i32_2_1_6()
26178   std::int32_t* result_chunk = result;  in gemm_i32_2_1_6()
26179   std::int32_t* mul_result_chunk = result;  in gemm_i32_2_1_6()
26180   const std::int32_t mul_result_chunk_stride_bytes = result_stride * 4;  in gemm_i32_2_1_6()
26223                     const std::uint8_t* rhs, std::int32_t m, std::int32_t n,  in gemm_i32_2_1_7()
26224                     std::int32_t k, std::int32_t lhs_offset,  in gemm_i32_2_1_7()
26225                     std::int32_t rhs_offset, std::int32_t* result,  in gemm_i32_2_1_7()
26226                     std::int32_t result_stride) {  in gemm_i32_2_1_7()
26227   const std::int32_t row_chunks = m / 3;  in gemm_i32_2_1_7()
26228   const std::int32_t col_chunks = n / 3;  in gemm_i32_2_1_7()
26229   const std::int32_t padded_k = ((k + 7) / 8) * 8;  in gemm_i32_2_1_7()
26230   const std::int32_t chunk_size = k * 3;  in gemm_i32_2_1_7()
26231   const std::int32_t zipped_chunk_size = (padded_k + 16) * 3;  in gemm_i32_2_1_7()
26232   const std::int32_t zipped_rhs_size = (padded_k + 16) * n;  in gemm_i32_2_1_7()
26236   std::int32_t* zipped_lhs_3_offsets =  in gemm_i32_2_1_7()
26237       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 3);  in gemm_i32_2_1_7()
26238   std::int32_t* zipped_lhs_2_offsets =  in gemm_i32_2_1_7()
26239       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 2);  in gemm_i32_2_1_7()
26242   const std::int32_t result_chunk_stride = result_stride * 3;  in gemm_i32_2_1_7()
26244   const std::int32_t const_offset = lhs_offset * rhs_offset * k;  in gemm_i32_2_1_7()
26245   std::int32_t* result_chunk = result;  in gemm_i32_2_1_7()
26246   std::int32_t* mul_result_chunk = result;  in gemm_i32_2_1_7()
26247   const std::int32_t mul_result_chunk_stride_bytes = result_stride * 4;  in gemm_i32_2_1_7()
26290                     const std::uint8_t* rhs, std::int32_t m, std::int32_t n,  in gemm_i32_2_2_0()
26291                     std::int32_t k, std::int32_t lhs_offset,  in gemm_i32_2_2_0()
26292                     std::int32_t rhs_offset, std::int32_t* result,  in gemm_i32_2_2_0()
26293                     std::int32_t result_stride) {  in gemm_i32_2_2_0()
26294   const std::int32_t row_chunks = m / 3;  in gemm_i32_2_2_0()
26295   const std::int32_t col_chunks = n / 3;  in gemm_i32_2_2_0()
26296   const std::int32_t padded_k = ((k + 7) / 8) * 8;  in gemm_i32_2_2_0()
26297   const std::int32_t chunk_size = k * 3;  in gemm_i32_2_2_0()
26298   const std::int32_t zipped_chunk_size = (padded_k + 16) * 3;  in gemm_i32_2_2_0()
26299   const std::int32_t zipped_rhs_size = (padded_k + 16) * n;  in gemm_i32_2_2_0()
26303   std::int32_t* zipped_lhs_3_offsets =  in gemm_i32_2_2_0()
26304       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 3);  in gemm_i32_2_2_0()
26305   std::int32_t* zipped_lhs_2_offsets =  in gemm_i32_2_2_0()
26306       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 2);  in gemm_i32_2_2_0()
26309   const std::int32_t result_chunk_stride = result_stride * 3;  in gemm_i32_2_2_0()
26311   const std::int32_t const_offset = lhs_offset * rhs_offset * k;  in gemm_i32_2_2_0()
26312   std::int32_t* result_chunk = result;  in gemm_i32_2_2_0()
26313   std::int32_t* mul_result_chunk = result;  in gemm_i32_2_2_0()
26314   const std::int32_t mul_result_chunk_stride_bytes = result_stride * 4;  in gemm_i32_2_2_0()
26357                     const std::uint8_t* rhs, std::int32_t m, std::int32_t n,  in gemm_i32_2_2_1()
26358                     std::int32_t k, std::int32_t lhs_offset,  in gemm_i32_2_2_1()
26359                     std::int32_t rhs_offset, std::int32_t* result,  in gemm_i32_2_2_1()
26360                     std::int32_t result_stride) {  in gemm_i32_2_2_1()
26361   const std::int32_t row_chunks = m / 3;  in gemm_i32_2_2_1()
26362   const std::int32_t col_chunks = n / 3;  in gemm_i32_2_2_1()
26363   const std::int32_t padded_k = ((k + 7) / 8) * 8;  in gemm_i32_2_2_1()
26364   const std::int32_t chunk_size = k * 3;  in gemm_i32_2_2_1()
26365   const std::int32_t zipped_chunk_size = (padded_k + 16) * 3;  in gemm_i32_2_2_1()
26366   const std::int32_t zipped_rhs_size = (padded_k + 16) * n;  in gemm_i32_2_2_1()
26370   std::int32_t* zipped_lhs_3_offsets =  in gemm_i32_2_2_1()
26371       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 3);  in gemm_i32_2_2_1()
26372   std::int32_t* zipped_lhs_2_offsets =  in gemm_i32_2_2_1()
26373       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 2);  in gemm_i32_2_2_1()
26376   const std::int32_t result_chunk_stride = result_stride * 3;  in gemm_i32_2_2_1()
26378   const std::int32_t const_offset = lhs_offset * rhs_offset * k;  in gemm_i32_2_2_1()
26379   std::int32_t* result_chunk = result;  in gemm_i32_2_2_1()
26380   std::int32_t* mul_result_chunk = result;  in gemm_i32_2_2_1()
26381   const std::int32_t mul_result_chunk_stride_bytes = result_stride * 4;  in gemm_i32_2_2_1()
26424                     const std::uint8_t* rhs, std::int32_t m, std::int32_t n,  in gemm_i32_2_2_2()
26425                     std::int32_t k, std::int32_t lhs_offset,  in gemm_i32_2_2_2()
26426                     std::int32_t rhs_offset, std::int32_t* result,  in gemm_i32_2_2_2()
26427                     std::int32_t result_stride) {  in gemm_i32_2_2_2()
26428   const std::int32_t row_chunks = m / 3;  in gemm_i32_2_2_2()
26429   const std::int32_t col_chunks = n / 3;  in gemm_i32_2_2_2()
26430   const std::int32_t padded_k = ((k + 7) / 8) * 8;  in gemm_i32_2_2_2()
26431   const std::int32_t chunk_size = k * 3;  in gemm_i32_2_2_2()
26432   const std::int32_t zipped_chunk_size = (padded_k + 16) * 3;  in gemm_i32_2_2_2()
26433   const std::int32_t zipped_rhs_size = (padded_k + 16) * n;  in gemm_i32_2_2_2()
26437   std::int32_t* zipped_lhs_3_offsets =  in gemm_i32_2_2_2()
26438       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 3);  in gemm_i32_2_2_2()
26439   std::int32_t* zipped_lhs_2_offsets =  in gemm_i32_2_2_2()
26440       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 2);  in gemm_i32_2_2_2()
26443   const std::int32_t result_chunk_stride = result_stride * 3;  in gemm_i32_2_2_2()
26445   const std::int32_t const_offset = lhs_offset * rhs_offset * k;  in gemm_i32_2_2_2()
26446   std::int32_t* result_chunk = result;  in gemm_i32_2_2_2()
26447   std::int32_t* mul_result_chunk = result;  in gemm_i32_2_2_2()
26448   const std::int32_t mul_result_chunk_stride_bytes = result_stride * 4;  in gemm_i32_2_2_2()
26491                     const std::uint8_t* rhs, std::int32_t m, std::int32_t n,  in gemm_i32_2_2_3()
26492                     std::int32_t k, std::int32_t lhs_offset,  in gemm_i32_2_2_3()
26493                     std::int32_t rhs_offset, std::int32_t* result,  in gemm_i32_2_2_3()
26494                     std::int32_t result_stride) {  in gemm_i32_2_2_3()
26495   const std::int32_t row_chunks = m / 3;  in gemm_i32_2_2_3()
26496   const std::int32_t col_chunks = n / 3;  in gemm_i32_2_2_3()
26497   const std::int32_t padded_k = ((k + 7) / 8) * 8;  in gemm_i32_2_2_3()
26498   const std::int32_t chunk_size = k * 3;  in gemm_i32_2_2_3()
26499   const std::int32_t zipped_chunk_size = (padded_k + 16) * 3;  in gemm_i32_2_2_3()
26500   const std::int32_t zipped_rhs_size = (padded_k + 16) * n;  in gemm_i32_2_2_3()
26504   std::int32_t* zipped_lhs_3_offsets =  in gemm_i32_2_2_3()
26505       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 3);  in gemm_i32_2_2_3()
26506   std::int32_t* zipped_lhs_2_offsets =  in gemm_i32_2_2_3()
26507       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 2);  in gemm_i32_2_2_3()
26510   const std::int32_t result_chunk_stride = result_stride * 3;  in gemm_i32_2_2_3()
26512   const std::int32_t const_offset = lhs_offset * rhs_offset * k;  in gemm_i32_2_2_3()
26513   std::int32_t* result_chunk = result;  in gemm_i32_2_2_3()
26514   std::int32_t* mul_result_chunk = result;  in gemm_i32_2_2_3()
26515   const std::int32_t mul_result_chunk_stride_bytes = result_stride * 4;  in gemm_i32_2_2_3()
26558                     const std::uint8_t* rhs, std::int32_t m, std::int32_t n,  in gemm_i32_2_2_4()
26559                     std::int32_t k, std::int32_t lhs_offset,  in gemm_i32_2_2_4()
26560                     std::int32_t rhs_offset, std::int32_t* result,  in gemm_i32_2_2_4()
26561                     std::int32_t result_stride) {  in gemm_i32_2_2_4()
26562   const std::int32_t row_chunks = m / 3;  in gemm_i32_2_2_4()
26563   const std::int32_t col_chunks = n / 3;  in gemm_i32_2_2_4()
26564   const std::int32_t padded_k = ((k + 7) / 8) * 8;  in gemm_i32_2_2_4()
26565   const std::int32_t chunk_size = k * 3;  in gemm_i32_2_2_4()
26566   const std::int32_t zipped_chunk_size = (padded_k + 16) * 3;  in gemm_i32_2_2_4()
26567   const std::int32_t zipped_rhs_size = (padded_k + 16) * n;  in gemm_i32_2_2_4()
26571   std::int32_t* zipped_lhs_3_offsets =  in gemm_i32_2_2_4()
26572       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 3);  in gemm_i32_2_2_4()
26573   std::int32_t* zipped_lhs_2_offsets =  in gemm_i32_2_2_4()
26574       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 2);  in gemm_i32_2_2_4()
26577   const std::int32_t result_chunk_stride = result_stride * 3;  in gemm_i32_2_2_4()
26579   const std::int32_t const_offset = lhs_offset * rhs_offset * k;  in gemm_i32_2_2_4()
26580   std::int32_t* result_chunk = result;  in gemm_i32_2_2_4()
26581   std::int32_t* mul_result_chunk = result;  in gemm_i32_2_2_4()
26582   const std::int32_t mul_result_chunk_stride_bytes = result_stride * 4;  in gemm_i32_2_2_4()
26625                     const std::uint8_t* rhs, std::int32_t m, std::int32_t n,  in gemm_i32_2_2_5()
26626                     std::int32_t k, std::int32_t lhs_offset,  in gemm_i32_2_2_5()
26627                     std::int32_t rhs_offset, std::int32_t* result,  in gemm_i32_2_2_5()
26628                     std::int32_t result_stride) {  in gemm_i32_2_2_5()
26629   const std::int32_t row_chunks = m / 3;  in gemm_i32_2_2_5()
26630   const std::int32_t col_chunks = n / 3;  in gemm_i32_2_2_5()
26631   const std::int32_t padded_k = ((k + 7) / 8) * 8;  in gemm_i32_2_2_5()
26632   const std::int32_t chunk_size = k * 3;  in gemm_i32_2_2_5()
26633   const std::int32_t zipped_chunk_size = (padded_k + 16) * 3;  in gemm_i32_2_2_5()
26634   const std::int32_t zipped_rhs_size = (padded_k + 16) * n;  in gemm_i32_2_2_5()
26638   std::int32_t* zipped_lhs_3_offsets =  in gemm_i32_2_2_5()
26639       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 3);  in gemm_i32_2_2_5()
26640   std::int32_t* zipped_lhs_2_offsets =  in gemm_i32_2_2_5()
26641       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 2);  in gemm_i32_2_2_5()
26644   const std::int32_t result_chunk_stride = result_stride * 3;  in gemm_i32_2_2_5()
26646   const std::int32_t const_offset = lhs_offset * rhs_offset * k;  in gemm_i32_2_2_5()
26647   std::int32_t* result_chunk = result;  in gemm_i32_2_2_5()
26648   std::int32_t* mul_result_chunk = result;  in gemm_i32_2_2_5()
26649   const std::int32_t mul_result_chunk_stride_bytes = result_stride * 4;  in gemm_i32_2_2_5()
26692                     const std::uint8_t* rhs, std::int32_t m, std::int32_t n,  in gemm_i32_2_2_6()
26693                     std::int32_t k, std::int32_t lhs_offset,  in gemm_i32_2_2_6()
26694                     std::int32_t rhs_offset, std::int32_t* result,  in gemm_i32_2_2_6()
26695                     std::int32_t result_stride) {  in gemm_i32_2_2_6()
26696   const std::int32_t row_chunks = m / 3;  in gemm_i32_2_2_6()
26697   const std::int32_t col_chunks = n / 3;  in gemm_i32_2_2_6()
26698   const std::int32_t padded_k = ((k + 7) / 8) * 8;  in gemm_i32_2_2_6()
26699   const std::int32_t chunk_size = k * 3;  in gemm_i32_2_2_6()
26700   const std::int32_t zipped_chunk_size = (padded_k + 16) * 3;  in gemm_i32_2_2_6()
26701   const std::int32_t zipped_rhs_size = (padded_k + 16) * n;  in gemm_i32_2_2_6()
26705   std::int32_t* zipped_lhs_3_offsets =  in gemm_i32_2_2_6()
26706       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 3);  in gemm_i32_2_2_6()
26707   std::int32_t* zipped_lhs_2_offsets =  in gemm_i32_2_2_6()
26708       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 2);  in gemm_i32_2_2_6()
26711   const std::int32_t result_chunk_stride = result_stride * 3;  in gemm_i32_2_2_6()
26713   const std::int32_t const_offset = lhs_offset * rhs_offset * k;  in gemm_i32_2_2_6()
26714   std::int32_t* result_chunk = result;  in gemm_i32_2_2_6()
26715   std::int32_t* mul_result_chunk = result;  in gemm_i32_2_2_6()
26716   const std::int32_t mul_result_chunk_stride_bytes = result_stride * 4;  in gemm_i32_2_2_6()
26759                     const std::uint8_t* rhs, std::int32_t m, std::int32_t n,  in gemm_i32_2_2_7()
26760                     std::int32_t k, std::int32_t lhs_offset,  in gemm_i32_2_2_7()
26761                     std::int32_t rhs_offset, std::int32_t* result,  in gemm_i32_2_2_7()
26762                     std::int32_t result_stride) {  in gemm_i32_2_2_7()
26763   const std::int32_t row_chunks = m / 3;  in gemm_i32_2_2_7()
26764   const std::int32_t col_chunks = n / 3;  in gemm_i32_2_2_7()
26765   const std::int32_t padded_k = ((k + 7) / 8) * 8;  in gemm_i32_2_2_7()
26766   const std::int32_t chunk_size = k * 3;  in gemm_i32_2_2_7()
26767   const std::int32_t zipped_chunk_size = (padded_k + 16) * 3;  in gemm_i32_2_2_7()
26768   const std::int32_t zipped_rhs_size = (padded_k + 16) * n;  in gemm_i32_2_2_7()
26772   std::int32_t* zipped_lhs_3_offsets =  in gemm_i32_2_2_7()
26773       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 3);  in gemm_i32_2_2_7()
26774   std::int32_t* zipped_lhs_2_offsets =  in gemm_i32_2_2_7()
26775       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 2);  in gemm_i32_2_2_7()
26778   const std::int32_t result_chunk_stride = result_stride * 3;  in gemm_i32_2_2_7()
26780   const std::int32_t const_offset = lhs_offset * rhs_offset * k;  in gemm_i32_2_2_7()
26781   std::int32_t* result_chunk = result;  in gemm_i32_2_2_7()
26782   std::int32_t* mul_result_chunk = result;  in gemm_i32_2_2_7()
26783   const std::int32_t mul_result_chunk_stride_bytes = result_stride * 4;  in gemm_i32_2_2_7()
26826                           const std::uint8_t* rhs, std::int32_t m,  in gemm_f_0_0_0_aligned()
26827                           std::int32_t n, std::int32_t k,  in gemm_f_0_0_0_aligned()
26828                           std::int32_t lhs_offset, std::int32_t rhs_offset,  in gemm_f_0_0_0_aligned()
26830                           std::int32_t result_stride) {  in gemm_f_0_0_0_aligned()
26831   const std::int32_t row_chunks = m / 3;  in gemm_f_0_0_0_aligned()
26832   const std::int32_t col_chunks = n / 3;  in gemm_f_0_0_0_aligned()
26833   const std::int32_t padded_k = ((k + 7) / 8) * 8;  in gemm_f_0_0_0_aligned()
26834   const std::int32_t chunk_size = k * 3;  in gemm_f_0_0_0_aligned()
26835   const std::int32_t zipped_chunk_size = (padded_k + 16) * 3;  in gemm_f_0_0_0_aligned()
26836   const std::int32_t zipped_rhs_size = (padded_k + 16) * n;  in gemm_f_0_0_0_aligned()
26840   std::int32_t* zipped_lhs_3_offsets =  in gemm_f_0_0_0_aligned()
26841       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 3);  in gemm_f_0_0_0_aligned()
26844   const std::int32_t result_chunk_stride = result_stride * 3;  in gemm_f_0_0_0_aligned()
26846   const std::int32_t const_offset = lhs_offset * rhs_offset * k;  in gemm_f_0_0_0_aligned()
26849   const std::int32_t mul_result_chunk_stride_bytes = result_stride * 4;  in gemm_f_0_0_0_aligned()
26874                           const std::uint8_t* rhs, std::int32_t m,  in gemm_f_0_0_1_aligned()
26875                           std::int32_t n, std::int32_t k,  in gemm_f_0_0_1_aligned()
26876                           std::int32_t lhs_offset, std::int32_t rhs_offset,  in gemm_f_0_0_1_aligned()
26878                           std::int32_t result_stride) {  in gemm_f_0_0_1_aligned()
26879   const std::int32_t row_chunks = m / 3;  in gemm_f_0_0_1_aligned()
26880   const std::int32_t col_chunks = n / 3;  in gemm_f_0_0_1_aligned()
26881   const std::int32_t padded_k = ((k + 7) / 8) * 8;  in gemm_f_0_0_1_aligned()
26882   const std::int32_t chunk_size = k * 3;  in gemm_f_0_0_1_aligned()
26883   const std::int32_t zipped_chunk_size = (padded_k + 16) * 3;  in gemm_f_0_0_1_aligned()
26884   const std::int32_t zipped_rhs_size = (padded_k + 16) * n;  in gemm_f_0_0_1_aligned()
26888   std::int32_t* zipped_lhs_3_offsets =  in gemm_f_0_0_1_aligned()
26889       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 3);  in gemm_f_0_0_1_aligned()
26892   const std::int32_t result_chunk_stride = result_stride * 3;  in gemm_f_0_0_1_aligned()
26894   const std::int32_t const_offset = lhs_offset * rhs_offset * k;  in gemm_f_0_0_1_aligned()
26897   const std::int32_t mul_result_chunk_stride_bytes = result_stride * 4;  in gemm_f_0_0_1_aligned()
26922                           const std::uint8_t* rhs, std::int32_t m,  in gemm_f_0_0_2_aligned()
26923                           std::int32_t n, std::int32_t k,  in gemm_f_0_0_2_aligned()
26924                           std::int32_t lhs_offset, std::int32_t rhs_offset,  in gemm_f_0_0_2_aligned()
26926                           std::int32_t result_stride) {  in gemm_f_0_0_2_aligned()
26927   const std::int32_t row_chunks = m / 3;  in gemm_f_0_0_2_aligned()
26928   const std::int32_t col_chunks = n / 3;  in gemm_f_0_0_2_aligned()
26929   const std::int32_t padded_k = ((k + 7) / 8) * 8;  in gemm_f_0_0_2_aligned()
26930   const std::int32_t chunk_size = k * 3;  in gemm_f_0_0_2_aligned()
26931   const std::int32_t zipped_chunk_size = (padded_k + 16) * 3;  in gemm_f_0_0_2_aligned()
26932   const std::int32_t zipped_rhs_size = (padded_k + 16) * n;  in gemm_f_0_0_2_aligned()
26936   std::int32_t* zipped_lhs_3_offsets =  in gemm_f_0_0_2_aligned()
26937       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 3);  in gemm_f_0_0_2_aligned()
26940   const std::int32_t result_chunk_stride = result_stride * 3;  in gemm_f_0_0_2_aligned()
26942   const std::int32_t const_offset = lhs_offset * rhs_offset * k;  in gemm_f_0_0_2_aligned()
26945   const std::int32_t mul_result_chunk_stride_bytes = result_stride * 4;  in gemm_f_0_0_2_aligned()
26970                           const std::uint8_t* rhs, std::int32_t m,  in gemm_f_0_0_3_aligned()
26971                           std::int32_t n, std::int32_t k,  in gemm_f_0_0_3_aligned()
26972                           std::int32_t lhs_offset, std::int32_t rhs_offset,  in gemm_f_0_0_3_aligned()
26974                           std::int32_t result_stride) {  in gemm_f_0_0_3_aligned()
26975   const std::int32_t row_chunks = m / 3;  in gemm_f_0_0_3_aligned()
26976   const std::int32_t col_chunks = n / 3;  in gemm_f_0_0_3_aligned()
26977   const std::int32_t padded_k = ((k + 7) / 8) * 8;  in gemm_f_0_0_3_aligned()
26978   const std::int32_t chunk_size = k * 3;  in gemm_f_0_0_3_aligned()
26979   const std::int32_t zipped_chunk_size = (padded_k + 16) * 3;  in gemm_f_0_0_3_aligned()
26980   const std::int32_t zipped_rhs_size = (padded_k + 16) * n;  in gemm_f_0_0_3_aligned()
26984   std::int32_t* zipped_lhs_3_offsets =  in gemm_f_0_0_3_aligned()
26985       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 3);  in gemm_f_0_0_3_aligned()
26988   const std::int32_t result_chunk_stride = result_stride * 3;  in gemm_f_0_0_3_aligned()
26990   const std::int32_t const_offset = lhs_offset * rhs_offset * k;  in gemm_f_0_0_3_aligned()
26993   const std::int32_t mul_result_chunk_stride_bytes = result_stride * 4;  in gemm_f_0_0_3_aligned()
27018                           const std::uint8_t* rhs, std::int32_t m,  in gemm_f_0_0_4_aligned()
27019                           std::int32_t n, std::int32_t k,  in gemm_f_0_0_4_aligned()
27020                           std::int32_t lhs_offset, std::int32_t rhs_offset,  in gemm_f_0_0_4_aligned()
27022                           std::int32_t result_stride) {  in gemm_f_0_0_4_aligned()
27023   const std::int32_t row_chunks = m / 3;  in gemm_f_0_0_4_aligned()
27024   const std::int32_t col_chunks = n / 3;  in gemm_f_0_0_4_aligned()
27025   const std::int32_t padded_k = ((k + 7) / 8) * 8;  in gemm_f_0_0_4_aligned()
27026   const std::int32_t chunk_size = k * 3;  in gemm_f_0_0_4_aligned()
27027   const std::int32_t zipped_chunk_size = (padded_k + 16) * 3;  in gemm_f_0_0_4_aligned()
27028   const std::int32_t zipped_rhs_size = (padded_k + 16) * n;  in gemm_f_0_0_4_aligned()
27032   std::int32_t* zipped_lhs_3_offsets =  in gemm_f_0_0_4_aligned()
27033       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 3);  in gemm_f_0_0_4_aligned()
27036   const std::int32_t result_chunk_stride = result_stride * 3;  in gemm_f_0_0_4_aligned()
27038   const std::int32_t const_offset = lhs_offset * rhs_offset * k;  in gemm_f_0_0_4_aligned()
27041   const std::int32_t mul_result_chunk_stride_bytes = result_stride * 4;  in gemm_f_0_0_4_aligned()
27066                           const std::uint8_t* rhs, std::int32_t m,  in gemm_f_0_0_5_aligned()
27067                           std::int32_t n, std::int32_t k,  in gemm_f_0_0_5_aligned()
27068                           std::int32_t lhs_offset, std::int32_t rhs_offset,  in gemm_f_0_0_5_aligned()
27070                           std::int32_t result_stride) {  in gemm_f_0_0_5_aligned()
27071   const std::int32_t row_chunks = m / 3;  in gemm_f_0_0_5_aligned()
27072   const std::int32_t col_chunks = n / 3;  in gemm_f_0_0_5_aligned()
27073   const std::int32_t padded_k = ((k + 7) / 8) * 8;  in gemm_f_0_0_5_aligned()
27074   const std::int32_t chunk_size = k * 3;  in gemm_f_0_0_5_aligned()
27075   const std::int32_t zipped_chunk_size = (padded_k + 16) * 3;  in gemm_f_0_0_5_aligned()
27076   const std::int32_t zipped_rhs_size = (padded_k + 16) * n;  in gemm_f_0_0_5_aligned()
27080   std::int32_t* zipped_lhs_3_offsets =  in gemm_f_0_0_5_aligned()
27081       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 3);  in gemm_f_0_0_5_aligned()
27084   const std::int32_t result_chunk_stride = result_stride * 3;  in gemm_f_0_0_5_aligned()
27086   const std::int32_t const_offset = lhs_offset * rhs_offset * k;  in gemm_f_0_0_5_aligned()
27089   const std::int32_t mul_result_chunk_stride_bytes = result_stride * 4;  in gemm_f_0_0_5_aligned()
27114                           const std::uint8_t* rhs, std::int32_t m,  in gemm_f_0_0_6_aligned()
27115                           std::int32_t n, std::int32_t k,  in gemm_f_0_0_6_aligned()
27116                           std::int32_t lhs_offset, std::int32_t rhs_offset,  in gemm_f_0_0_6_aligned()
27118                           std::int32_t result_stride) {  in gemm_f_0_0_6_aligned()
27119   const std::int32_t row_chunks = m / 3;  in gemm_f_0_0_6_aligned()
27120   const std::int32_t col_chunks = n / 3;  in gemm_f_0_0_6_aligned()
27121   const std::int32_t padded_k = ((k + 7) / 8) * 8;  in gemm_f_0_0_6_aligned()
27122   const std::int32_t chunk_size = k * 3;  in gemm_f_0_0_6_aligned()
27123   const std::int32_t zipped_chunk_size = (padded_k + 16) * 3;  in gemm_f_0_0_6_aligned()
27124   const std::int32_t zipped_rhs_size = (padded_k + 16) * n;  in gemm_f_0_0_6_aligned()
27128   std::int32_t* zipped_lhs_3_offsets =  in gemm_f_0_0_6_aligned()
27129       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 3);  in gemm_f_0_0_6_aligned()
27132   const std::int32_t result_chunk_stride = result_stride * 3;  in gemm_f_0_0_6_aligned()
27134   const std::int32_t const_offset = lhs_offset * rhs_offset * k;  in gemm_f_0_0_6_aligned()
27137   const std::int32_t mul_result_chunk_stride_bytes = result_stride * 4;  in gemm_f_0_0_6_aligned()
27162                           const std::uint8_t* rhs, std::int32_t m,  in gemm_f_0_0_7_aligned()
27163                           std::int32_t n, std::int32_t k,  in gemm_f_0_0_7_aligned()
27164                           std::int32_t lhs_offset, std::int32_t rhs_offset,  in gemm_f_0_0_7_aligned()
27166                           std::int32_t result_stride) {  in gemm_f_0_0_7_aligned()
27167   const std::int32_t row_chunks = m / 3;  in gemm_f_0_0_7_aligned()
27168   const std::int32_t col_chunks = n / 3;  in gemm_f_0_0_7_aligned()
27169   const std::int32_t padded_k = ((k + 7) / 8) * 8;  in gemm_f_0_0_7_aligned()
27170   const std::int32_t chunk_size = k * 3;  in gemm_f_0_0_7_aligned()
27171   const std::int32_t zipped_chunk_size = (padded_k + 16) * 3;  in gemm_f_0_0_7_aligned()
27172   const std::int32_t zipped_rhs_size = (padded_k + 16) * n;  in gemm_f_0_0_7_aligned()
27176   std::int32_t* zipped_lhs_3_offsets =  in gemm_f_0_0_7_aligned()
27177       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 3);  in gemm_f_0_0_7_aligned()
27180   const std::int32_t result_chunk_stride = result_stride * 3;  in gemm_f_0_0_7_aligned()
27182   const std::int32_t const_offset = lhs_offset * rhs_offset * k;  in gemm_f_0_0_7_aligned()
27185   const std::int32_t mul_result_chunk_stride_bytes = result_stride * 4;  in gemm_f_0_0_7_aligned()
27210                           const std::uint8_t* rhs, std::int32_t m,  in gemm_f_0_1_0_aligned()
27211                           std::int32_t n, std::int32_t k,  in gemm_f_0_1_0_aligned()
27212                           std::int32_t lhs_offset, std::int32_t rhs_offset,  in gemm_f_0_1_0_aligned()
27214                           std::int32_t result_stride) {  in gemm_f_0_1_0_aligned()
27215   const std::int32_t row_chunks = m / 3;  in gemm_f_0_1_0_aligned()
27216   const std::int32_t col_chunks = n / 3;  in gemm_f_0_1_0_aligned()
27217   const std::int32_t padded_k = ((k + 7) / 8) * 8;  in gemm_f_0_1_0_aligned()
27218   const std::int32_t chunk_size = k * 3;  in gemm_f_0_1_0_aligned()
27219   const std::int32_t zipped_chunk_size = (padded_k + 16) * 3;  in gemm_f_0_1_0_aligned()
27220   const std::int32_t zipped_rhs_size = (padded_k + 16) * n;  in gemm_f_0_1_0_aligned()
27224   std::int32_t* zipped_lhs_3_offsets =  in gemm_f_0_1_0_aligned()
27225       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 3);  in gemm_f_0_1_0_aligned()
27228   const std::int32_t result_chunk_stride = result_stride * 3;  in gemm_f_0_1_0_aligned()
27230   const std::int32_t const_offset = lhs_offset * rhs_offset * k;  in gemm_f_0_1_0_aligned()
27233   const std::int32_t mul_result_chunk_stride_bytes = result_stride * 4;  in gemm_f_0_1_0_aligned()
27262                           const std::uint8_t* rhs, std::int32_t m,  in gemm_f_0_1_1_aligned()
27263                           std::int32_t n, std::int32_t k,  in gemm_f_0_1_1_aligned()
27264                           std::int32_t lhs_offset, std::int32_t rhs_offset,  in gemm_f_0_1_1_aligned()
27266                           std::int32_t result_stride) {  in gemm_f_0_1_1_aligned()
27267   const std::int32_t row_chunks = m / 3;  in gemm_f_0_1_1_aligned()
27268   const std::int32_t col_chunks = n / 3;  in gemm_f_0_1_1_aligned()
27269   const std::int32_t padded_k = ((k + 7) / 8) * 8;  in gemm_f_0_1_1_aligned()
27270   const std::int32_t chunk_size = k * 3;  in gemm_f_0_1_1_aligned()
27271   const std::int32_t zipped_chunk_size = (padded_k + 16) * 3;  in gemm_f_0_1_1_aligned()
27272   const std::int32_t zipped_rhs_size = (padded_k + 16) * n;  in gemm_f_0_1_1_aligned()
27276   std::int32_t* zipped_lhs_3_offsets =  in gemm_f_0_1_1_aligned()
27277       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 3);  in gemm_f_0_1_1_aligned()
27280   const std::int32_t result_chunk_stride = result_stride * 3;  in gemm_f_0_1_1_aligned()
27282   const std::int32_t const_offset = lhs_offset * rhs_offset * k;  in gemm_f_0_1_1_aligned()
27285   const std::int32_t mul_result_chunk_stride_bytes = result_stride * 4;  in gemm_f_0_1_1_aligned()
27314                           const std::uint8_t* rhs, std::int32_t m,  in gemm_f_0_1_2_aligned()
27315                           std::int32_t n, std::int32_t k,  in gemm_f_0_1_2_aligned()
27316                           std::int32_t lhs_offset, std::int32_t rhs_offset,  in gemm_f_0_1_2_aligned()
27318                           std::int32_t result_stride) {  in gemm_f_0_1_2_aligned()
27319   const std::int32_t row_chunks = m / 3;  in gemm_f_0_1_2_aligned()
27320   const std::int32_t col_chunks = n / 3;  in gemm_f_0_1_2_aligned()
27321   const std::int32_t padded_k = ((k + 7) / 8) * 8;  in gemm_f_0_1_2_aligned()
27322   const std::int32_t chunk_size = k * 3;  in gemm_f_0_1_2_aligned()
27323   const std::int32_t zipped_chunk_size = (padded_k + 16) * 3;  in gemm_f_0_1_2_aligned()
27324   const std::int32_t zipped_rhs_size = (padded_k + 16) * n;  in gemm_f_0_1_2_aligned()
27328   std::int32_t* zipped_lhs_3_offsets =  in gemm_f_0_1_2_aligned()
27329       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 3);  in gemm_f_0_1_2_aligned()
27332   const std::int32_t result_chunk_stride = result_stride * 3;  in gemm_f_0_1_2_aligned()
27334   const std::int32_t const_offset = lhs_offset * rhs_offset * k;  in gemm_f_0_1_2_aligned()
27337   const std::int32_t mul_result_chunk_stride_bytes = result_stride * 4;  in gemm_f_0_1_2_aligned()
27366                           const std::uint8_t* rhs, std::int32_t m,  in gemm_f_0_1_3_aligned()
27367                           std::int32_t n, std::int32_t k,  in gemm_f_0_1_3_aligned()
27368                           std::int32_t lhs_offset, std::int32_t rhs_offset,  in gemm_f_0_1_3_aligned()
27370                           std::int32_t result_stride) {  in gemm_f_0_1_3_aligned()
27371   const std::int32_t row_chunks = m / 3;  in gemm_f_0_1_3_aligned()
27372   const std::int32_t col_chunks = n / 3;  in gemm_f_0_1_3_aligned()
27373   const std::int32_t padded_k = ((k + 7) / 8) * 8;  in gemm_f_0_1_3_aligned()
27374   const std::int32_t chunk_size = k * 3;  in gemm_f_0_1_3_aligned()
27375   const std::int32_t zipped_chunk_size = (padded_k + 16) * 3;  in gemm_f_0_1_3_aligned()
27376   const std::int32_t zipped_rhs_size = (padded_k + 16) * n;  in gemm_f_0_1_3_aligned()
27380   std::int32_t* zipped_lhs_3_offsets =  in gemm_f_0_1_3_aligned()
27381       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 3);  in gemm_f_0_1_3_aligned()
27384   const std::int32_t result_chunk_stride = result_stride * 3;  in gemm_f_0_1_3_aligned()
27386   const std::int32_t const_offset = lhs_offset * rhs_offset * k;  in gemm_f_0_1_3_aligned()
27389   const std::int32_t mul_result_chunk_stride_bytes = result_stride * 4;  in gemm_f_0_1_3_aligned()
27418                           const std::uint8_t* rhs, std::int32_t m,  in gemm_f_0_1_4_aligned()
27419                           std::int32_t n, std::int32_t k,  in gemm_f_0_1_4_aligned()
27420                           std::int32_t lhs_offset, std::int32_t rhs_offset,  in gemm_f_0_1_4_aligned()
27422                           std::int32_t result_stride) {  in gemm_f_0_1_4_aligned()
27423   const std::int32_t row_chunks = m / 3;  in gemm_f_0_1_4_aligned()
27424   const std::int32_t col_chunks = n / 3;  in gemm_f_0_1_4_aligned()
27425   const std::int32_t padded_k = ((k + 7) / 8) * 8;  in gemm_f_0_1_4_aligned()
27426   const std::int32_t chunk_size = k * 3;  in gemm_f_0_1_4_aligned()
27427   const std::int32_t zipped_chunk_size = (padded_k + 16) * 3;  in gemm_f_0_1_4_aligned()
27428   const std::int32_t zipped_rhs_size = (padded_k + 16) * n;  in gemm_f_0_1_4_aligned()
27432   std::int32_t* zipped_lhs_3_offsets =  in gemm_f_0_1_4_aligned()
27433       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 3);  in gemm_f_0_1_4_aligned()
27436   const std::int32_t result_chunk_stride = result_stride * 3;  in gemm_f_0_1_4_aligned()
27438   const std::int32_t const_offset = lhs_offset * rhs_offset * k;  in gemm_f_0_1_4_aligned()
27441   const std::int32_t mul_result_chunk_stride_bytes = result_stride * 4;  in gemm_f_0_1_4_aligned()
27470                           const std::uint8_t* rhs, std::int32_t m,  in gemm_f_0_1_5_aligned()
27471                           std::int32_t n, std::int32_t k,  in gemm_f_0_1_5_aligned()
27472                           std::int32_t lhs_offset, std::int32_t rhs_offset,  in gemm_f_0_1_5_aligned()
27474                           std::int32_t result_stride) {  in gemm_f_0_1_5_aligned()
27475   const std::int32_t row_chunks = m / 3;  in gemm_f_0_1_5_aligned()
27476   const std::int32_t col_chunks = n / 3;  in gemm_f_0_1_5_aligned()
27477   const std::int32_t padded_k = ((k + 7) / 8) * 8;  in gemm_f_0_1_5_aligned()
27478   const std::int32_t chunk_size = k * 3;  in gemm_f_0_1_5_aligned()
27479   const std::int32_t zipped_chunk_size = (padded_k + 16) * 3;  in gemm_f_0_1_5_aligned()
27480   const std::int32_t zipped_rhs_size = (padded_k + 16) * n;  in gemm_f_0_1_5_aligned()
27484   std::int32_t* zipped_lhs_3_offsets =  in gemm_f_0_1_5_aligned()
27485       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 3);  in gemm_f_0_1_5_aligned()
27488   const std::int32_t result_chunk_stride = result_stride * 3;  in gemm_f_0_1_5_aligned()
27490   const std::int32_t const_offset = lhs_offset * rhs_offset * k;  in gemm_f_0_1_5_aligned()
27493   const std::int32_t mul_result_chunk_stride_bytes = result_stride * 4;  in gemm_f_0_1_5_aligned()
27522                           const std::uint8_t* rhs, std::int32_t m,  in gemm_f_0_1_6_aligned()
27523                           std::int32_t n, std::int32_t k,  in gemm_f_0_1_6_aligned()
27524                           std::int32_t lhs_offset, std::int32_t rhs_offset,  in gemm_f_0_1_6_aligned()
27526                           std::int32_t result_stride) {  in gemm_f_0_1_6_aligned()
27527   const std::int32_t row_chunks = m / 3;  in gemm_f_0_1_6_aligned()
27528   const std::int32_t col_chunks = n / 3;  in gemm_f_0_1_6_aligned()
27529   const std::int32_t padded_k = ((k + 7) / 8) * 8;  in gemm_f_0_1_6_aligned()
27530   const std::int32_t chunk_size = k * 3;  in gemm_f_0_1_6_aligned()
27531   const std::int32_t zipped_chunk_size = (padded_k + 16) * 3;  in gemm_f_0_1_6_aligned()
27532   const std::int32_t zipped_rhs_size = (padded_k + 16) * n;  in gemm_f_0_1_6_aligned()
27536   std::int32_t* zipped_lhs_3_offsets =  in gemm_f_0_1_6_aligned()
27537       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 3);  in gemm_f_0_1_6_aligned()
27540   const std::int32_t result_chunk_stride = result_stride * 3;  in gemm_f_0_1_6_aligned()
27542   const std::int32_t const_offset = lhs_offset * rhs_offset * k;  in gemm_f_0_1_6_aligned()
27545   const std::int32_t mul_result_chunk_stride_bytes = result_stride * 4;  in gemm_f_0_1_6_aligned()
27574                           const std::uint8_t* rhs, std::int32_t m,  in gemm_f_0_1_7_aligned()
27575                           std::int32_t n, std::int32_t k,  in gemm_f_0_1_7_aligned()
27576                           std::int32_t lhs_offset, std::int32_t rhs_offset,  in gemm_f_0_1_7_aligned()
27578                           std::int32_t result_stride) {  in gemm_f_0_1_7_aligned()
27579   const std::int32_t row_chunks = m / 3;  in gemm_f_0_1_7_aligned()
27580   const std::int32_t col_chunks = n / 3;  in gemm_f_0_1_7_aligned()
27581   const std::int32_t padded_k = ((k + 7) / 8) * 8;  in gemm_f_0_1_7_aligned()
27582   const std::int32_t chunk_size = k * 3;  in gemm_f_0_1_7_aligned()
27583   const std::int32_t zipped_chunk_size = (padded_k + 16) * 3;  in gemm_f_0_1_7_aligned()
27584   const std::int32_t zipped_rhs_size = (padded_k + 16) * n;  in gemm_f_0_1_7_aligned()
27588   std::int32_t* zipped_lhs_3_offsets =  in gemm_f_0_1_7_aligned()
27589       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 3);  in gemm_f_0_1_7_aligned()
27592   const std::int32_t result_chunk_stride = result_stride * 3;  in gemm_f_0_1_7_aligned()
27594   const std::int32_t const_offset = lhs_offset * rhs_offset * k;  in gemm_f_0_1_7_aligned()
27597   const std::int32_t mul_result_chunk_stride_bytes = result_stride * 4;  in gemm_f_0_1_7_aligned()
27626                           const std::uint8_t* rhs, std::int32_t m,  in gemm_f_0_2_0_aligned()
27627                           std::int32_t n, std::int32_t k,  in gemm_f_0_2_0_aligned()
27628                           std::int32_t lhs_offset, std::int32_t rhs_offset,  in gemm_f_0_2_0_aligned()
27630                           std::int32_t result_stride) {  in gemm_f_0_2_0_aligned()
27631   const std::int32_t row_chunks = m / 3;  in gemm_f_0_2_0_aligned()
27632   const std::int32_t col_chunks = n / 3;  in gemm_f_0_2_0_aligned()
27633   const std::int32_t padded_k = ((k + 7) / 8) * 8;  in gemm_f_0_2_0_aligned()
27634   const std::int32_t chunk_size = k * 3;  in gemm_f_0_2_0_aligned()
27635   const std::int32_t zipped_chunk_size = (padded_k + 16) * 3;  in gemm_f_0_2_0_aligned()
27636   const std::int32_t zipped_rhs_size = (padded_k + 16) * n;  in gemm_f_0_2_0_aligned()
27640   std::int32_t* zipped_lhs_3_offsets =  in gemm_f_0_2_0_aligned()
27641       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 3);  in gemm_f_0_2_0_aligned()
27644   const std::int32_t result_chunk_stride = result_stride * 3;  in gemm_f_0_2_0_aligned()
27646   const std::int32_t const_offset = lhs_offset * rhs_offset * k;  in gemm_f_0_2_0_aligned()
27649   const std::int32_t mul_result_chunk_stride_bytes = result_stride * 4;  in gemm_f_0_2_0_aligned()
27678                           const std::uint8_t* rhs, std::int32_t m,  in gemm_f_0_2_1_aligned()
27679                           std::int32_t n, std::int32_t k,  in gemm_f_0_2_1_aligned()
27680                           std::int32_t lhs_offset, std::int32_t rhs_offset,  in gemm_f_0_2_1_aligned()
27682                           std::int32_t result_stride) {  in gemm_f_0_2_1_aligned()
27683   const std::int32_t row_chunks = m / 3;  in gemm_f_0_2_1_aligned()
27684   const std::int32_t col_chunks = n / 3;  in gemm_f_0_2_1_aligned()
27685   const std::int32_t padded_k = ((k + 7) / 8) * 8;  in gemm_f_0_2_1_aligned()
27686   const std::int32_t chunk_size = k * 3;  in gemm_f_0_2_1_aligned()
27687   const std::int32_t zipped_chunk_size = (padded_k + 16) * 3;  in gemm_f_0_2_1_aligned()
27688   const std::int32_t zipped_rhs_size = (padded_k + 16) * n;  in gemm_f_0_2_1_aligned()
27692   std::int32_t* zipped_lhs_3_offsets =  in gemm_f_0_2_1_aligned()
27693       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 3);  in gemm_f_0_2_1_aligned()
27696   const std::int32_t result_chunk_stride = result_stride * 3;  in gemm_f_0_2_1_aligned()
27698   const std::int32_t const_offset = lhs_offset * rhs_offset * k;  in gemm_f_0_2_1_aligned()
27701   const std::int32_t mul_result_chunk_stride_bytes = result_stride * 4;  in gemm_f_0_2_1_aligned()
27730                           const std::uint8_t* rhs, std::int32_t m,  in gemm_f_0_2_2_aligned()
27731                           std::int32_t n, std::int32_t k,  in gemm_f_0_2_2_aligned()
27732                           std::int32_t lhs_offset, std::int32_t rhs_offset,  in gemm_f_0_2_2_aligned()
27734                           std::int32_t result_stride) {  in gemm_f_0_2_2_aligned()
27735   const std::int32_t row_chunks = m / 3;  in gemm_f_0_2_2_aligned()
27736   const std::int32_t col_chunks = n / 3;  in gemm_f_0_2_2_aligned()
27737   const std::int32_t padded_k = ((k + 7) / 8) * 8;  in gemm_f_0_2_2_aligned()
27738   const std::int32_t chunk_size = k * 3;  in gemm_f_0_2_2_aligned()
27739   const std::int32_t zipped_chunk_size = (padded_k + 16) * 3;  in gemm_f_0_2_2_aligned()
27740   const std::int32_t zipped_rhs_size = (padded_k + 16) * n;  in gemm_f_0_2_2_aligned()
27744   std::int32_t* zipped_lhs_3_offsets =  in gemm_f_0_2_2_aligned()
27745       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 3);  in gemm_f_0_2_2_aligned()
27748   const std::int32_t result_chunk_stride = result_stride * 3;  in gemm_f_0_2_2_aligned()
27750   const std::int32_t const_offset = lhs_offset * rhs_offset * k;  in gemm_f_0_2_2_aligned()
27753   const std::int32_t mul_result_chunk_stride_bytes = result_stride * 4;  in gemm_f_0_2_2_aligned()
27782                           const std::uint8_t* rhs, std::int32_t m,  in gemm_f_0_2_3_aligned()
27783                           std::int32_t n, std::int32_t k,  in gemm_f_0_2_3_aligned()
27784                           std::int32_t lhs_offset, std::int32_t rhs_offset,  in gemm_f_0_2_3_aligned()
27786                           std::int32_t result_stride) {  in gemm_f_0_2_3_aligned()
27787   const std::int32_t row_chunks = m / 3;  in gemm_f_0_2_3_aligned()
27788   const std::int32_t col_chunks = n / 3;  in gemm_f_0_2_3_aligned()
27789   const std::int32_t padded_k = ((k + 7) / 8) * 8;  in gemm_f_0_2_3_aligned()
27790   const std::int32_t chunk_size = k * 3;  in gemm_f_0_2_3_aligned()
27791   const std::int32_t zipped_chunk_size = (padded_k + 16) * 3;  in gemm_f_0_2_3_aligned()
27792   const std::int32_t zipped_rhs_size = (padded_k + 16) * n;  in gemm_f_0_2_3_aligned()
27796   std::int32_t* zipped_lhs_3_offsets =  in gemm_f_0_2_3_aligned()
27797       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 3);  in gemm_f_0_2_3_aligned()
27800   const std::int32_t result_chunk_stride = result_stride * 3;  in gemm_f_0_2_3_aligned()
27802   const std::int32_t const_offset = lhs_offset * rhs_offset * k;  in gemm_f_0_2_3_aligned()
27805   const std::int32_t mul_result_chunk_stride_bytes = result_stride * 4;  in gemm_f_0_2_3_aligned()
27834                           const std::uint8_t* rhs, std::int32_t m,  in gemm_f_0_2_4_aligned()
27835                           std::int32_t n, std::int32_t k,  in gemm_f_0_2_4_aligned()
27836                           std::int32_t lhs_offset, std::int32_t rhs_offset,  in gemm_f_0_2_4_aligned()
27838                           std::int32_t result_stride) {  in gemm_f_0_2_4_aligned()
27839   const std::int32_t row_chunks = m / 3;  in gemm_f_0_2_4_aligned()
27840   const std::int32_t col_chunks = n / 3;  in gemm_f_0_2_4_aligned()
27841   const std::int32_t padded_k = ((k + 7) / 8) * 8;  in gemm_f_0_2_4_aligned()
27842   const std::int32_t chunk_size = k * 3;  in gemm_f_0_2_4_aligned()
27843   const std::int32_t zipped_chunk_size = (padded_k + 16) * 3;  in gemm_f_0_2_4_aligned()
27844   const std::int32_t zipped_rhs_size = (padded_k + 16) * n;  in gemm_f_0_2_4_aligned()
27848   std::int32_t* zipped_lhs_3_offsets =  in gemm_f_0_2_4_aligned()
27849       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 3);  in gemm_f_0_2_4_aligned()
27852   const std::int32_t result_chunk_stride = result_stride * 3;  in gemm_f_0_2_4_aligned()
27854   const std::int32_t const_offset = lhs_offset * rhs_offset * k;  in gemm_f_0_2_4_aligned()
27857   const std::int32_t mul_result_chunk_stride_bytes = result_stride * 4;  in gemm_f_0_2_4_aligned()
27886                           const std::uint8_t* rhs, std::int32_t m,  in gemm_f_0_2_5_aligned()
27887                           std::int32_t n, std::int32_t k,  in gemm_f_0_2_5_aligned()
27888                           std::int32_t lhs_offset, std::int32_t rhs_offset,  in gemm_f_0_2_5_aligned()
27890                           std::int32_t result_stride) {  in gemm_f_0_2_5_aligned()
27891   const std::int32_t row_chunks = m / 3;  in gemm_f_0_2_5_aligned()
27892   const std::int32_t col_chunks = n / 3;  in gemm_f_0_2_5_aligned()
27893   const std::int32_t padded_k = ((k + 7) / 8) * 8;  in gemm_f_0_2_5_aligned()
27894   const std::int32_t chunk_size = k * 3;  in gemm_f_0_2_5_aligned()
27895   const std::int32_t zipped_chunk_size = (padded_k + 16) * 3;  in gemm_f_0_2_5_aligned()
27896   const std::int32_t zipped_rhs_size = (padded_k + 16) * n;  in gemm_f_0_2_5_aligned()
27900   std::int32_t* zipped_lhs_3_offsets =  in gemm_f_0_2_5_aligned()
27901       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 3);  in gemm_f_0_2_5_aligned()
27904   const std::int32_t result_chunk_stride = result_stride * 3;  in gemm_f_0_2_5_aligned()
27906   const std::int32_t const_offset = lhs_offset * rhs_offset * k;  in gemm_f_0_2_5_aligned()
27909   const std::int32_t mul_result_chunk_stride_bytes = result_stride * 4;  in gemm_f_0_2_5_aligned()
27938                           const std::uint8_t* rhs, std::int32_t m,  in gemm_f_0_2_6_aligned()
27939                           std::int32_t n, std::int32_t k,  in gemm_f_0_2_6_aligned()
27940                           std::int32_t lhs_offset, std::int32_t rhs_offset,  in gemm_f_0_2_6_aligned()
27942                           std::int32_t result_stride) {  in gemm_f_0_2_6_aligned()
27943   const std::int32_t row_chunks = m / 3;  in gemm_f_0_2_6_aligned()
27944   const std::int32_t col_chunks = n / 3;  in gemm_f_0_2_6_aligned()
27945   const std::int32_t padded_k = ((k + 7) / 8) * 8;  in gemm_f_0_2_6_aligned()
27946   const std::int32_t chunk_size = k * 3;  in gemm_f_0_2_6_aligned()
27947   const std::int32_t zipped_chunk_size = (padded_k + 16) * 3;  in gemm_f_0_2_6_aligned()
27948   const std::int32_t zipped_rhs_size = (padded_k + 16) * n;  in gemm_f_0_2_6_aligned()
27952   std::int32_t* zipped_lhs_3_offsets =  in gemm_f_0_2_6_aligned()
27953       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 3);  in gemm_f_0_2_6_aligned()
27956   const std::int32_t result_chunk_stride = result_stride * 3;  in gemm_f_0_2_6_aligned()
27958   const std::int32_t const_offset = lhs_offset * rhs_offset * k;  in gemm_f_0_2_6_aligned()
27961   const std::int32_t mul_result_chunk_stride_bytes = result_stride * 4;  in gemm_f_0_2_6_aligned()
27990                           const std::uint8_t* rhs, std::int32_t m,  in gemm_f_0_2_7_aligned()
27991                           std::int32_t n, std::int32_t k,  in gemm_f_0_2_7_aligned()
27992                           std::int32_t lhs_offset, std::int32_t rhs_offset,  in gemm_f_0_2_7_aligned()
27994                           std::int32_t result_stride) {  in gemm_f_0_2_7_aligned()
27995   const std::int32_t row_chunks = m / 3;  in gemm_f_0_2_7_aligned()
27996   const std::int32_t col_chunks = n / 3;  in gemm_f_0_2_7_aligned()
27997   const std::int32_t padded_k = ((k + 7) / 8) * 8;  in gemm_f_0_2_7_aligned()
27998   const std::int32_t chunk_size = k * 3;  in gemm_f_0_2_7_aligned()
27999   const std::int32_t zipped_chunk_size = (padded_k + 16) * 3;  in gemm_f_0_2_7_aligned()
28000   const std::int32_t zipped_rhs_size = (padded_k + 16) * n;  in gemm_f_0_2_7_aligned()
28004   std::int32_t* zipped_lhs_3_offsets =  in gemm_f_0_2_7_aligned()
28005       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 3);  in gemm_f_0_2_7_aligned()
28008   const std::int32_t result_chunk_stride = result_stride * 3;  in gemm_f_0_2_7_aligned()
28010   const std::int32_t const_offset = lhs_offset * rhs_offset * k;  in gemm_f_0_2_7_aligned()
28013   const std::int32_t mul_result_chunk_stride_bytes = result_stride * 4;  in gemm_f_0_2_7_aligned()
28042                           const std::uint8_t* rhs, std::int32_t m,  in gemm_f_1_0_0_aligned()
28043                           std::int32_t n, std::int32_t k,  in gemm_f_1_0_0_aligned()
28044                           std::int32_t lhs_offset, std::int32_t rhs_offset,  in gemm_f_1_0_0_aligned()
28046                           std::int32_t result_stride) {  in gemm_f_1_0_0_aligned()
28047   const std::int32_t row_chunks = m / 3;  in gemm_f_1_0_0_aligned()
28048   const std::int32_t col_chunks = n / 3;  in gemm_f_1_0_0_aligned()
28049   const std::int32_t padded_k = ((k + 7) / 8) * 8;  in gemm_f_1_0_0_aligned()
28050   const std::int32_t chunk_size = k * 3;  in gemm_f_1_0_0_aligned()
28051   const std::int32_t zipped_chunk_size = (padded_k + 16) * 3;  in gemm_f_1_0_0_aligned()
28052   const std::int32_t zipped_rhs_size = (padded_k + 16) * n;  in gemm_f_1_0_0_aligned()
28056   std::int32_t* zipped_lhs_3_offsets =  in gemm_f_1_0_0_aligned()
28057       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 3);  in gemm_f_1_0_0_aligned()
28058   std::int32_t* zipped_lhs_1_offsets =  in gemm_f_1_0_0_aligned()
28059       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 1);  in gemm_f_1_0_0_aligned()
28062   const std::int32_t result_chunk_stride = result_stride * 3;  in gemm_f_1_0_0_aligned()
28064   const std::int32_t const_offset = lhs_offset * rhs_offset * k;  in gemm_f_1_0_0_aligned()
28067   const std::int32_t mul_result_chunk_stride_bytes = result_stride * 4;  in gemm_f_1_0_0_aligned()
28103                           const std::uint8_t* rhs, std::int32_t m,  in gemm_f_1_0_1_aligned()
28104                           std::int32_t n, std::int32_t k,  in gemm_f_1_0_1_aligned()
28105                           std::int32_t lhs_offset, std::int32_t rhs_offset,  in gemm_f_1_0_1_aligned()
28107                           std::int32_t result_stride) {  in gemm_f_1_0_1_aligned()
28108   const std::int32_t row_chunks = m / 3;  in gemm_f_1_0_1_aligned()
28109   const std::int32_t col_chunks = n / 3;  in gemm_f_1_0_1_aligned()
28110   const std::int32_t padded_k = ((k + 7) / 8) * 8;  in gemm_f_1_0_1_aligned()
28111   const std::int32_t chunk_size = k * 3;  in gemm_f_1_0_1_aligned()
28112   const std::int32_t zipped_chunk_size = (padded_k + 16) * 3;  in gemm_f_1_0_1_aligned()
28113   const std::int32_t zipped_rhs_size = (padded_k + 16) * n;  in gemm_f_1_0_1_aligned()
28117   std::int32_t* zipped_lhs_3_offsets =  in gemm_f_1_0_1_aligned()
28118       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 3);  in gemm_f_1_0_1_aligned()
28119   std::int32_t* zipped_lhs_1_offsets =  in gemm_f_1_0_1_aligned()
28120       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 1);  in gemm_f_1_0_1_aligned()
28123   const std::int32_t result_chunk_stride = result_stride * 3;  in gemm_f_1_0_1_aligned()
28125   const std::int32_t const_offset = lhs_offset * rhs_offset * k;  in gemm_f_1_0_1_aligned()
28128   const std::int32_t mul_result_chunk_stride_bytes = result_stride * 4;  in gemm_f_1_0_1_aligned()
28164                           const std::uint8_t* rhs, std::int32_t m,  in gemm_f_1_0_2_aligned()
28165                           std::int32_t n, std::int32_t k,  in gemm_f_1_0_2_aligned()
28166                           std::int32_t lhs_offset, std::int32_t rhs_offset,  in gemm_f_1_0_2_aligned()
28168                           std::int32_t result_stride) {  in gemm_f_1_0_2_aligned()
28169   const std::int32_t row_chunks = m / 3;  in gemm_f_1_0_2_aligned()
28170   const std::int32_t col_chunks = n / 3;  in gemm_f_1_0_2_aligned()
28171   const std::int32_t padded_k = ((k + 7) / 8) * 8;  in gemm_f_1_0_2_aligned()
28172   const std::int32_t chunk_size = k * 3;  in gemm_f_1_0_2_aligned()
28173   const std::int32_t zipped_chunk_size = (padded_k + 16) * 3;  in gemm_f_1_0_2_aligned()
28174   const std::int32_t zipped_rhs_size = (padded_k + 16) * n;  in gemm_f_1_0_2_aligned()
28178   std::int32_t* zipped_lhs_3_offsets =  in gemm_f_1_0_2_aligned()
28179       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 3);  in gemm_f_1_0_2_aligned()
28180   std::int32_t* zipped_lhs_1_offsets =  in gemm_f_1_0_2_aligned()
28181       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 1);  in gemm_f_1_0_2_aligned()
28184   const std::int32_t result_chunk_stride = result_stride * 3;  in gemm_f_1_0_2_aligned()
28186   const std::int32_t const_offset = lhs_offset * rhs_offset * k;  in gemm_f_1_0_2_aligned()
28189   const std::int32_t mul_result_chunk_stride_bytes = result_stride * 4;  in gemm_f_1_0_2_aligned()
28225                           const std::uint8_t* rhs, std::int32_t m,  in gemm_f_1_0_3_aligned()
28226                           std::int32_t n, std::int32_t k,  in gemm_f_1_0_3_aligned()
28227                           std::int32_t lhs_offset, std::int32_t rhs_offset,  in gemm_f_1_0_3_aligned()
28229                           std::int32_t result_stride) {  in gemm_f_1_0_3_aligned()
28230   const std::int32_t row_chunks = m / 3;  in gemm_f_1_0_3_aligned()
28231   const std::int32_t col_chunks = n / 3;  in gemm_f_1_0_3_aligned()
28232   const std::int32_t padded_k = ((k + 7) / 8) * 8;  in gemm_f_1_0_3_aligned()
28233   const std::int32_t chunk_size = k * 3;  in gemm_f_1_0_3_aligned()
28234   const std::int32_t zipped_chunk_size = (padded_k + 16) * 3;  in gemm_f_1_0_3_aligned()
28235   const std::int32_t zipped_rhs_size = (padded_k + 16) * n;  in gemm_f_1_0_3_aligned()
28239   std::int32_t* zipped_lhs_3_offsets =  in gemm_f_1_0_3_aligned()
28240       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 3);  in gemm_f_1_0_3_aligned()
28241   std::int32_t* zipped_lhs_1_offsets =  in gemm_f_1_0_3_aligned()
28242       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 1);  in gemm_f_1_0_3_aligned()
28245   const std::int32_t result_chunk_stride = result_stride * 3;  in gemm_f_1_0_3_aligned()
28247   const std::int32_t const_offset = lhs_offset * rhs_offset * k;  in gemm_f_1_0_3_aligned()
28250   const std::int32_t mul_result_chunk_stride_bytes = result_stride * 4;  in gemm_f_1_0_3_aligned()
28286                           const std::uint8_t* rhs, std::int32_t m,  in gemm_f_1_0_4_aligned()
28287                           std::int32_t n, std::int32_t k,  in gemm_f_1_0_4_aligned()
28288                           std::int32_t lhs_offset, std::int32_t rhs_offset,  in gemm_f_1_0_4_aligned()
28290                           std::int32_t result_stride) {  in gemm_f_1_0_4_aligned()
28291   const std::int32_t row_chunks = m / 3;  in gemm_f_1_0_4_aligned()
28292   const std::int32_t col_chunks = n / 3;  in gemm_f_1_0_4_aligned()
28293   const std::int32_t padded_k = ((k + 7) / 8) * 8;  in gemm_f_1_0_4_aligned()
28294   const std::int32_t chunk_size = k * 3;  in gemm_f_1_0_4_aligned()
28295   const std::int32_t zipped_chunk_size = (padded_k + 16) * 3;  in gemm_f_1_0_4_aligned()
28296   const std::int32_t zipped_rhs_size = (padded_k + 16) * n;  in gemm_f_1_0_4_aligned()
28300   std::int32_t* zipped_lhs_3_offsets =  in gemm_f_1_0_4_aligned()
28301       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 3);  in gemm_f_1_0_4_aligned()
28302   std::int32_t* zipped_lhs_1_offsets =  in gemm_f_1_0_4_aligned()
28303       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 1);  in gemm_f_1_0_4_aligned()
28306   const std::int32_t result_chunk_stride = result_stride * 3;  in gemm_f_1_0_4_aligned()
28308   const std::int32_t const_offset = lhs_offset * rhs_offset * k;  in gemm_f_1_0_4_aligned()
28311   const std::int32_t mul_result_chunk_stride_bytes = result_stride * 4;  in gemm_f_1_0_4_aligned()
28347                           const std::uint8_t* rhs, std::int32_t m,  in gemm_f_1_0_5_aligned()
28348                           std::int32_t n, std::int32_t k,  in gemm_f_1_0_5_aligned()
28349                           std::int32_t lhs_offset, std::int32_t rhs_offset,  in gemm_f_1_0_5_aligned()
28351                           std::int32_t result_stride) {  in gemm_f_1_0_5_aligned()
28352   const std::int32_t row_chunks = m / 3;  in gemm_f_1_0_5_aligned()
28353   const std::int32_t col_chunks = n / 3;  in gemm_f_1_0_5_aligned()
28354   const std::int32_t padded_k = ((k + 7) / 8) * 8;  in gemm_f_1_0_5_aligned()
28355   const std::int32_t chunk_size = k * 3;  in gemm_f_1_0_5_aligned()
28356   const std::int32_t zipped_chunk_size = (padded_k + 16) * 3;  in gemm_f_1_0_5_aligned()
28357   const std::int32_t zipped_rhs_size = (padded_k + 16) * n;  in gemm_f_1_0_5_aligned()
28361   std::int32_t* zipped_lhs_3_offsets =  in gemm_f_1_0_5_aligned()
28362       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 3);  in gemm_f_1_0_5_aligned()
28363   std::int32_t* zipped_lhs_1_offsets =  in gemm_f_1_0_5_aligned()
28364       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 1);  in gemm_f_1_0_5_aligned()
28367   const std::int32_t result_chunk_stride = result_stride * 3;  in gemm_f_1_0_5_aligned()
28369   const std::int32_t const_offset = lhs_offset * rhs_offset * k;  in gemm_f_1_0_5_aligned()
28372   const std::int32_t mul_result_chunk_stride_bytes = result_stride * 4;  in gemm_f_1_0_5_aligned()
28408                           const std::uint8_t* rhs, std::int32_t m,  in gemm_f_1_0_6_aligned()
28409                           std::int32_t n, std::int32_t k,  in gemm_f_1_0_6_aligned()
28410                           std::int32_t lhs_offset, std::int32_t rhs_offset,  in gemm_f_1_0_6_aligned()
28412                           std::int32_t result_stride) {  in gemm_f_1_0_6_aligned()
28413   const std::int32_t row_chunks = m / 3;  in gemm_f_1_0_6_aligned()
28414   const std::int32_t col_chunks = n / 3;  in gemm_f_1_0_6_aligned()
28415   const std::int32_t padded_k = ((k + 7) / 8) * 8;  in gemm_f_1_0_6_aligned()
28416   const std::int32_t chunk_size = k * 3;  in gemm_f_1_0_6_aligned()
28417   const std::int32_t zipped_chunk_size = (padded_k + 16) * 3;  in gemm_f_1_0_6_aligned()
28418   const std::int32_t zipped_rhs_size = (padded_k + 16) * n;  in gemm_f_1_0_6_aligned()
28422   std::int32_t* zipped_lhs_3_offsets =  in gemm_f_1_0_6_aligned()
28423       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 3);  in gemm_f_1_0_6_aligned()
28424   std::int32_t* zipped_lhs_1_offsets =  in gemm_f_1_0_6_aligned()
28425       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 1);  in gemm_f_1_0_6_aligned()
28428   const std::int32_t result_chunk_stride = result_stride * 3;  in gemm_f_1_0_6_aligned()
28430   const std::int32_t const_offset = lhs_offset * rhs_offset * k;  in gemm_f_1_0_6_aligned()
28433   const std::int32_t mul_result_chunk_stride_bytes = result_stride * 4;  in gemm_f_1_0_6_aligned()
28469                           const std::uint8_t* rhs, std::int32_t m,  in gemm_f_1_0_7_aligned()
28470                           std::int32_t n, std::int32_t k,  in gemm_f_1_0_7_aligned()
28471                           std::int32_t lhs_offset, std::int32_t rhs_offset,  in gemm_f_1_0_7_aligned()
28473                           std::int32_t result_stride) {  in gemm_f_1_0_7_aligned()
28474   const std::int32_t row_chunks = m / 3;  in gemm_f_1_0_7_aligned()
28475   const std::int32_t col_chunks = n / 3;  in gemm_f_1_0_7_aligned()
28476   const std::int32_t padded_k = ((k + 7) / 8) * 8;  in gemm_f_1_0_7_aligned()
28477   const std::int32_t chunk_size = k * 3;  in gemm_f_1_0_7_aligned()
28478   const std::int32_t zipped_chunk_size = (padded_k + 16) * 3;  in gemm_f_1_0_7_aligned()
28479   const std::int32_t zipped_rhs_size = (padded_k + 16) * n;  in gemm_f_1_0_7_aligned()
28483   std::int32_t* zipped_lhs_3_offsets =  in gemm_f_1_0_7_aligned()
28484       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 3);  in gemm_f_1_0_7_aligned()
28485   std::int32_t* zipped_lhs_1_offsets =  in gemm_f_1_0_7_aligned()
28486       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 1);  in gemm_f_1_0_7_aligned()
28489   const std::int32_t result_chunk_stride = result_stride * 3;  in gemm_f_1_0_7_aligned()
28491   const std::int32_t const_offset = lhs_offset * rhs_offset * k;  in gemm_f_1_0_7_aligned()
28494   const std::int32_t mul_result_chunk_stride_bytes = result_stride * 4;  in gemm_f_1_0_7_aligned()
28530                           const std::uint8_t* rhs, std::int32_t m,  in gemm_f_1_1_0_aligned()
28531                           std::int32_t n, std::int32_t k,  in gemm_f_1_1_0_aligned()
28532                           std::int32_t lhs_offset, std::int32_t rhs_offset,  in gemm_f_1_1_0_aligned()
28534                           std::int32_t result_stride) {  in gemm_f_1_1_0_aligned()
28535   const std::int32_t row_chunks = m / 3;  in gemm_f_1_1_0_aligned()
28536   const std::int32_t col_chunks = n / 3;  in gemm_f_1_1_0_aligned()
28537   const std::int32_t padded_k = ((k + 7) / 8) * 8;  in gemm_f_1_1_0_aligned()
28538   const std::int32_t chunk_size = k * 3;  in gemm_f_1_1_0_aligned()
28539   const std::int32_t zipped_chunk_size = (padded_k + 16) * 3;  in gemm_f_1_1_0_aligned()
28540   const std::int32_t zipped_rhs_size = (padded_k + 16) * n;  in gemm_f_1_1_0_aligned()
28544   std::int32_t* zipped_lhs_3_offsets =  in gemm_f_1_1_0_aligned()
28545       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 3);  in gemm_f_1_1_0_aligned()
28546   std::int32_t* zipped_lhs_1_offsets =  in gemm_f_1_1_0_aligned()
28547       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 1);  in gemm_f_1_1_0_aligned()
28550   const std::int32_t result_chunk_stride = result_stride * 3;  in gemm_f_1_1_0_aligned()
28552   const std::int32_t const_offset = lhs_offset * rhs_offset * k;  in gemm_f_1_1_0_aligned()
28555   const std::int32_t mul_result_chunk_stride_bytes = result_stride * 4;  in gemm_f_1_1_0_aligned()
28598                           const std::uint8_t* rhs, std::int32_t m,  in gemm_f_1_1_1_aligned()
28599                           std::int32_t n, std::int32_t k,  in gemm_f_1_1_1_aligned()
28600                           std::int32_t lhs_offset, std::int32_t rhs_offset,  in gemm_f_1_1_1_aligned()
28602                           std::int32_t result_stride) {  in gemm_f_1_1_1_aligned()
28603   const std::int32_t row_chunks = m / 3;  in gemm_f_1_1_1_aligned()
28604   const std::int32_t col_chunks = n / 3;  in gemm_f_1_1_1_aligned()
28605   const std::int32_t padded_k = ((k + 7) / 8) * 8;  in gemm_f_1_1_1_aligned()
28606   const std::int32_t chunk_size = k * 3;  in gemm_f_1_1_1_aligned()
28607   const std::int32_t zipped_chunk_size = (padded_k + 16) * 3;  in gemm_f_1_1_1_aligned()
28608   const std::int32_t zipped_rhs_size = (padded_k + 16) * n;  in gemm_f_1_1_1_aligned()
28612   std::int32_t* zipped_lhs_3_offsets =  in gemm_f_1_1_1_aligned()
28613       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 3);  in gemm_f_1_1_1_aligned()
28614   std::int32_t* zipped_lhs_1_offsets =  in gemm_f_1_1_1_aligned()
28615       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 1);  in gemm_f_1_1_1_aligned()
28618   const std::int32_t result_chunk_stride = result_stride * 3;  in gemm_f_1_1_1_aligned()
28620   const std::int32_t const_offset = lhs_offset * rhs_offset * k;  in gemm_f_1_1_1_aligned()
28623   const std::int32_t mul_result_chunk_stride_bytes = result_stride * 4;  in gemm_f_1_1_1_aligned()
28666                           const std::uint8_t* rhs, std::int32_t m,  in gemm_f_1_1_2_aligned()
28667                           std::int32_t n, std::int32_t k,  in gemm_f_1_1_2_aligned()
28668                           std::int32_t lhs_offset, std::int32_t rhs_offset,  in gemm_f_1_1_2_aligned()
28670                           std::int32_t result_stride) {  in gemm_f_1_1_2_aligned()
28671   const std::int32_t row_chunks = m / 3;  in gemm_f_1_1_2_aligned()
28672   const std::int32_t col_chunks = n / 3;  in gemm_f_1_1_2_aligned()
28673   const std::int32_t padded_k = ((k + 7) / 8) * 8;  in gemm_f_1_1_2_aligned()
28674   const std::int32_t chunk_size = k * 3;  in gemm_f_1_1_2_aligned()
28675   const std::int32_t zipped_chunk_size = (padded_k + 16) * 3;  in gemm_f_1_1_2_aligned()
28676   const std::int32_t zipped_rhs_size = (padded_k + 16) * n;  in gemm_f_1_1_2_aligned()
28680   std::int32_t* zipped_lhs_3_offsets =  in gemm_f_1_1_2_aligned()
28681       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 3);  in gemm_f_1_1_2_aligned()
28682   std::int32_t* zipped_lhs_1_offsets =  in gemm_f_1_1_2_aligned()
28683       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 1);  in gemm_f_1_1_2_aligned()
28686   const std::int32_t result_chunk_stride = result_stride * 3;  in gemm_f_1_1_2_aligned()
28688   const std::int32_t const_offset = lhs_offset * rhs_offset * k;  in gemm_f_1_1_2_aligned()
28691   const std::int32_t mul_result_chunk_stride_bytes = result_stride * 4;  in gemm_f_1_1_2_aligned()
28734                           const std::uint8_t* rhs, std::int32_t m,  in gemm_f_1_1_3_aligned()
28735                           std::int32_t n, std::int32_t k,  in gemm_f_1_1_3_aligned()
28736                           std::int32_t lhs_offset, std::int32_t rhs_offset,  in gemm_f_1_1_3_aligned()
28738                           std::int32_t result_stride) {  in gemm_f_1_1_3_aligned()
28739   const std::int32_t row_chunks = m / 3;  in gemm_f_1_1_3_aligned()
28740   const std::int32_t col_chunks = n / 3;  in gemm_f_1_1_3_aligned()
28741   const std::int32_t padded_k = ((k + 7) / 8) * 8;  in gemm_f_1_1_3_aligned()
28742   const std::int32_t chunk_size = k * 3;  in gemm_f_1_1_3_aligned()
28743   const std::int32_t zipped_chunk_size = (padded_k + 16) * 3;  in gemm_f_1_1_3_aligned()
28744   const std::int32_t zipped_rhs_size = (padded_k + 16) * n;  in gemm_f_1_1_3_aligned()
28748   std::int32_t* zipped_lhs_3_offsets =  in gemm_f_1_1_3_aligned()
28749       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 3);  in gemm_f_1_1_3_aligned()
28750   std::int32_t* zipped_lhs_1_offsets =  in gemm_f_1_1_3_aligned()
28751       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 1);  in gemm_f_1_1_3_aligned()
28754   const std::int32_t result_chunk_stride = result_stride * 3;  in gemm_f_1_1_3_aligned()
28756   const std::int32_t const_offset = lhs_offset * rhs_offset * k;  in gemm_f_1_1_3_aligned()
28759   const std::int32_t mul_result_chunk_stride_bytes = result_stride * 4;  in gemm_f_1_1_3_aligned()
28802                           const std::uint8_t* rhs, std::int32_t m,  in gemm_f_1_1_4_aligned()
28803                           std::int32_t n, std::int32_t k,  in gemm_f_1_1_4_aligned()
28804                           std::int32_t lhs_offset, std::int32_t rhs_offset,  in gemm_f_1_1_4_aligned()
28806                           std::int32_t result_stride) {  in gemm_f_1_1_4_aligned()
28807   const std::int32_t row_chunks = m / 3;  in gemm_f_1_1_4_aligned()
28808   const std::int32_t col_chunks = n / 3;  in gemm_f_1_1_4_aligned()
28809   const std::int32_t padded_k = ((k + 7) / 8) * 8;  in gemm_f_1_1_4_aligned()
28810   const std::int32_t chunk_size = k * 3;  in gemm_f_1_1_4_aligned()
28811   const std::int32_t zipped_chunk_size = (padded_k + 16) * 3;  in gemm_f_1_1_4_aligned()
28812   const std::int32_t zipped_rhs_size = (padded_k + 16) * n;  in gemm_f_1_1_4_aligned()
28816   std::int32_t* zipped_lhs_3_offsets =  in gemm_f_1_1_4_aligned()
28817       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 3);  in gemm_f_1_1_4_aligned()
28818   std::int32_t* zipped_lhs_1_offsets =  in gemm_f_1_1_4_aligned()
28819       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 1);  in gemm_f_1_1_4_aligned()
28822   const std::int32_t result_chunk_stride = result_stride * 3;  in gemm_f_1_1_4_aligned()
28824   const std::int32_t const_offset = lhs_offset * rhs_offset * k;  in gemm_f_1_1_4_aligned()
28827   const std::int32_t mul_result_chunk_stride_bytes = result_stride * 4;  in gemm_f_1_1_4_aligned()
28870                           const std::uint8_t* rhs, std::int32_t m,  in gemm_f_1_1_5_aligned()
28871                           std::int32_t n, std::int32_t k,  in gemm_f_1_1_5_aligned()
28872                           std::int32_t lhs_offset, std::int32_t rhs_offset,  in gemm_f_1_1_5_aligned()
28874                           std::int32_t result_stride) {  in gemm_f_1_1_5_aligned()
28875   const std::int32_t row_chunks = m / 3;  in gemm_f_1_1_5_aligned()
28876   const std::int32_t col_chunks = n / 3;  in gemm_f_1_1_5_aligned()
28877   const std::int32_t padded_k = ((k + 7) / 8) * 8;  in gemm_f_1_1_5_aligned()
28878   const std::int32_t chunk_size = k * 3;  in gemm_f_1_1_5_aligned()
28879   const std::int32_t zipped_chunk_size = (padded_k + 16) * 3;  in gemm_f_1_1_5_aligned()
28880   const std::int32_t zipped_rhs_size = (padded_k + 16) * n;  in gemm_f_1_1_5_aligned()
28884   std::int32_t* zipped_lhs_3_offsets =  in gemm_f_1_1_5_aligned()
28885       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 3);  in gemm_f_1_1_5_aligned()
28886   std::int32_t* zipped_lhs_1_offsets =  in gemm_f_1_1_5_aligned()
28887       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 1);  in gemm_f_1_1_5_aligned()
28890   const std::int32_t result_chunk_stride = result_stride * 3;  in gemm_f_1_1_5_aligned()
28892   const std::int32_t const_offset = lhs_offset * rhs_offset * k;  in gemm_f_1_1_5_aligned()
28895   const std::int32_t mul_result_chunk_stride_bytes = result_stride * 4;  in gemm_f_1_1_5_aligned()
28938                           const std::uint8_t* rhs, std::int32_t m,  in gemm_f_1_1_6_aligned()
28939                           std::int32_t n, std::int32_t k,  in gemm_f_1_1_6_aligned()
28940                           std::int32_t lhs_offset, std::int32_t rhs_offset,  in gemm_f_1_1_6_aligned()
28942                           std::int32_t result_stride) {  in gemm_f_1_1_6_aligned()
28943   const std::int32_t row_chunks = m / 3;  in gemm_f_1_1_6_aligned()
28944   const std::int32_t col_chunks = n / 3;  in gemm_f_1_1_6_aligned()
28945   const std::int32_t padded_k = ((k + 7) / 8) * 8;  in gemm_f_1_1_6_aligned()
28946   const std::int32_t chunk_size = k * 3;  in gemm_f_1_1_6_aligned()
28947   const std::int32_t zipped_chunk_size = (padded_k + 16) * 3;  in gemm_f_1_1_6_aligned()
28948   const std::int32_t zipped_rhs_size = (padded_k + 16) * n;  in gemm_f_1_1_6_aligned()
28952   std::int32_t* zipped_lhs_3_offsets =  in gemm_f_1_1_6_aligned()
28953       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 3);  in gemm_f_1_1_6_aligned()
28954   std::int32_t* zipped_lhs_1_offsets =  in gemm_f_1_1_6_aligned()
28955       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 1);  in gemm_f_1_1_6_aligned()
28958   const std::int32_t result_chunk_stride = result_stride * 3;  in gemm_f_1_1_6_aligned()
28960   const std::int32_t const_offset = lhs_offset * rhs_offset * k;  in gemm_f_1_1_6_aligned()
28963   const std::int32_t mul_result_chunk_stride_bytes = result_stride * 4;  in gemm_f_1_1_6_aligned()
29006                           const std::uint8_t* rhs, std::int32_t m,  in gemm_f_1_1_7_aligned()
29007                           std::int32_t n, std::int32_t k,  in gemm_f_1_1_7_aligned()
29008                           std::int32_t lhs_offset, std::int32_t rhs_offset,  in gemm_f_1_1_7_aligned()
29010                           std::int32_t result_stride) {  in gemm_f_1_1_7_aligned()
29011   const std::int32_t row_chunks = m / 3;  in gemm_f_1_1_7_aligned()
29012   const std::int32_t col_chunks = n / 3;  in gemm_f_1_1_7_aligned()
29013   const std::int32_t padded_k = ((k + 7) / 8) * 8;  in gemm_f_1_1_7_aligned()
29014   const std::int32_t chunk_size = k * 3;  in gemm_f_1_1_7_aligned()
29015   const std::int32_t zipped_chunk_size = (padded_k + 16) * 3;  in gemm_f_1_1_7_aligned()
29016   const std::int32_t zipped_rhs_size = (padded_k + 16) * n;  in gemm_f_1_1_7_aligned()
29020   std::int32_t* zipped_lhs_3_offsets =  in gemm_f_1_1_7_aligned()
29021       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 3);  in gemm_f_1_1_7_aligned()
29022   std::int32_t* zipped_lhs_1_offsets =  in gemm_f_1_1_7_aligned()
29023       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 1);  in gemm_f_1_1_7_aligned()
29026   const std::int32_t result_chunk_stride = result_stride * 3;  in gemm_f_1_1_7_aligned()
29028   const std::int32_t const_offset = lhs_offset * rhs_offset * k;  in gemm_f_1_1_7_aligned()
29031   const std::int32_t mul_result_chunk_stride_bytes = result_stride * 4;  in gemm_f_1_1_7_aligned()
29074                           const std::uint8_t* rhs, std::int32_t m,  in gemm_f_1_2_0_aligned()
29075                           std::int32_t n, std::int32_t k,  in gemm_f_1_2_0_aligned()
29076                           std::int32_t lhs_offset, std::int32_t rhs_offset,  in gemm_f_1_2_0_aligned()
29078                           std::int32_t result_stride) {  in gemm_f_1_2_0_aligned()
29079   const std::int32_t row_chunks = m / 3;  in gemm_f_1_2_0_aligned()
29080   const std::int32_t col_chunks = n / 3;  in gemm_f_1_2_0_aligned()
29081   const std::int32_t padded_k = ((k + 7) / 8) * 8;  in gemm_f_1_2_0_aligned()
29082   const std::int32_t chunk_size = k * 3;  in gemm_f_1_2_0_aligned()
29083   const std::int32_t zipped_chunk_size = (padded_k + 16) * 3;  in gemm_f_1_2_0_aligned()
29084   const std::int32_t zipped_rhs_size = (padded_k + 16) * n;  in gemm_f_1_2_0_aligned()
29088   std::int32_t* zipped_lhs_3_offsets =  in gemm_f_1_2_0_aligned()
29089       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 3);  in gemm_f_1_2_0_aligned()
29090   std::int32_t* zipped_lhs_1_offsets =  in gemm_f_1_2_0_aligned()
29091       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 1);  in gemm_f_1_2_0_aligned()
29094   const std::int32_t result_chunk_stride = result_stride * 3;  in gemm_f_1_2_0_aligned()
29096   const std::int32_t const_offset = lhs_offset * rhs_offset * k;  in gemm_f_1_2_0_aligned()
29099   const std::int32_t mul_result_chunk_stride_bytes = result_stride * 4;  in gemm_f_1_2_0_aligned()
29142                           const std::uint8_t* rhs, std::int32_t m,  in gemm_f_1_2_1_aligned()
29143                           std::int32_t n, std::int32_t k,  in gemm_f_1_2_1_aligned()
29144                           std::int32_t lhs_offset, std::int32_t rhs_offset,  in gemm_f_1_2_1_aligned()
29146                           std::int32_t result_stride) {  in gemm_f_1_2_1_aligned()
29147   const std::int32_t row_chunks = m / 3;  in gemm_f_1_2_1_aligned()
29148   const std::int32_t col_chunks = n / 3;  in gemm_f_1_2_1_aligned()
29149   const std::int32_t padded_k = ((k + 7) / 8) * 8;  in gemm_f_1_2_1_aligned()
29150   const std::int32_t chunk_size = k * 3;  in gemm_f_1_2_1_aligned()
29151   const std::int32_t zipped_chunk_size = (padded_k + 16) * 3;  in gemm_f_1_2_1_aligned()
29152   const std::int32_t zipped_rhs_size = (padded_k + 16) * n;  in gemm_f_1_2_1_aligned()
29156   std::int32_t* zipped_lhs_3_offsets =  in gemm_f_1_2_1_aligned()
29157       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 3);  in gemm_f_1_2_1_aligned()
29158   std::int32_t* zipped_lhs_1_offsets =  in gemm_f_1_2_1_aligned()
29159       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 1);  in gemm_f_1_2_1_aligned()
29162   const std::int32_t result_chunk_stride = result_stride * 3;  in gemm_f_1_2_1_aligned()
29164   const std::int32_t const_offset = lhs_offset * rhs_offset * k;  in gemm_f_1_2_1_aligned()
29167   const std::int32_t mul_result_chunk_stride_bytes = result_stride * 4;  in gemm_f_1_2_1_aligned()
29210                           const std::uint8_t* rhs, std::int32_t m,  in gemm_f_1_2_2_aligned()
29211                           std::int32_t n, std::int32_t k,  in gemm_f_1_2_2_aligned()
29212                           std::int32_t lhs_offset, std::int32_t rhs_offset,  in gemm_f_1_2_2_aligned()
29214                           std::int32_t result_stride) {  in gemm_f_1_2_2_aligned()
29215   const std::int32_t row_chunks = m / 3;  in gemm_f_1_2_2_aligned()
29216   const std::int32_t col_chunks = n / 3;  in gemm_f_1_2_2_aligned()
29217   const std::int32_t padded_k = ((k + 7) / 8) * 8;  in gemm_f_1_2_2_aligned()
29218   const std::int32_t chunk_size = k * 3;  in gemm_f_1_2_2_aligned()
29219   const std::int32_t zipped_chunk_size = (padded_k + 16) * 3;  in gemm_f_1_2_2_aligned()
29220   const std::int32_t zipped_rhs_size = (padded_k + 16) * n;  in gemm_f_1_2_2_aligned()
29224   std::int32_t* zipped_lhs_3_offsets =  in gemm_f_1_2_2_aligned()
29225       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 3);  in gemm_f_1_2_2_aligned()
29226   std::int32_t* zipped_lhs_1_offsets =  in gemm_f_1_2_2_aligned()
29227       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 1);  in gemm_f_1_2_2_aligned()
29230   const std::int32_t result_chunk_stride = result_stride * 3;  in gemm_f_1_2_2_aligned()
29232   const std::int32_t const_offset = lhs_offset * rhs_offset * k;  in gemm_f_1_2_2_aligned()
29235   const std::int32_t mul_result_chunk_stride_bytes = result_stride * 4;  in gemm_f_1_2_2_aligned()
29278                           const std::uint8_t* rhs, std::int32_t m,  in gemm_f_1_2_3_aligned()
29279                           std::int32_t n, std::int32_t k,  in gemm_f_1_2_3_aligned()
29280                           std::int32_t lhs_offset, std::int32_t rhs_offset,  in gemm_f_1_2_3_aligned()
29282                           std::int32_t result_stride) {  in gemm_f_1_2_3_aligned()
29283   const std::int32_t row_chunks = m / 3;  in gemm_f_1_2_3_aligned()
29284   const std::int32_t col_chunks = n / 3;  in gemm_f_1_2_3_aligned()
29285   const std::int32_t padded_k = ((k + 7) / 8) * 8;  in gemm_f_1_2_3_aligned()
29286   const std::int32_t chunk_size = k * 3;  in gemm_f_1_2_3_aligned()
29287   const std::int32_t zipped_chunk_size = (padded_k + 16) * 3;  in gemm_f_1_2_3_aligned()
29288   const std::int32_t zipped_rhs_size = (padded_k + 16) * n;  in gemm_f_1_2_3_aligned()
29292   std::int32_t* zipped_lhs_3_offsets =  in gemm_f_1_2_3_aligned()
29293       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 3);  in gemm_f_1_2_3_aligned()
29294   std::int32_t* zipped_lhs_1_offsets =  in gemm_f_1_2_3_aligned()
29295       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 1);  in gemm_f_1_2_3_aligned()
29298   const std::int32_t result_chunk_stride = result_stride * 3;  in gemm_f_1_2_3_aligned()
29300   const std::int32_t const_offset = lhs_offset * rhs_offset * k;  in gemm_f_1_2_3_aligned()
29303   const std::int32_t mul_result_chunk_stride_bytes = result_stride * 4;  in gemm_f_1_2_3_aligned()
29346                           const std::uint8_t* rhs, std::int32_t m,  in gemm_f_1_2_4_aligned()
29347                           std::int32_t n, std::int32_t k,  in gemm_f_1_2_4_aligned()
29348                           std::int32_t lhs_offset, std::int32_t rhs_offset,  in gemm_f_1_2_4_aligned()
29350                           std::int32_t result_stride) {  in gemm_f_1_2_4_aligned()
29351   const std::int32_t row_chunks = m / 3;  in gemm_f_1_2_4_aligned()
29352   const std::int32_t col_chunks = n / 3;  in gemm_f_1_2_4_aligned()
29353   const std::int32_t padded_k = ((k + 7) / 8) * 8;  in gemm_f_1_2_4_aligned()
29354   const std::int32_t chunk_size = k * 3;  in gemm_f_1_2_4_aligned()
29355   const std::int32_t zipped_chunk_size = (padded_k + 16) * 3;  in gemm_f_1_2_4_aligned()
29356   const std::int32_t zipped_rhs_size = (padded_k + 16) * n;  in gemm_f_1_2_4_aligned()
29360   std::int32_t* zipped_lhs_3_offsets =  in gemm_f_1_2_4_aligned()
29361       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 3);  in gemm_f_1_2_4_aligned()
29362   std::int32_t* zipped_lhs_1_offsets =  in gemm_f_1_2_4_aligned()
29363       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 1);  in gemm_f_1_2_4_aligned()
29366   const std::int32_t result_chunk_stride = result_stride * 3;  in gemm_f_1_2_4_aligned()
29368   const std::int32_t const_offset = lhs_offset * rhs_offset * k;  in gemm_f_1_2_4_aligned()
29371   const std::int32_t mul_result_chunk_stride_bytes = result_stride * 4;  in gemm_f_1_2_4_aligned()
29414                           const std::uint8_t* rhs, std::int32_t m,  in gemm_f_1_2_5_aligned()
29415                           std::int32_t n, std::int32_t k,  in gemm_f_1_2_5_aligned()
29416                           std::int32_t lhs_offset, std::int32_t rhs_offset,  in gemm_f_1_2_5_aligned()
29418                           std::int32_t result_stride) {  in gemm_f_1_2_5_aligned()
29419   const std::int32_t row_chunks = m / 3;  in gemm_f_1_2_5_aligned()
29420   const std::int32_t col_chunks = n / 3;  in gemm_f_1_2_5_aligned()
29421   const std::int32_t padded_k = ((k + 7) / 8) * 8;  in gemm_f_1_2_5_aligned()
29422   const std::int32_t chunk_size = k * 3;  in gemm_f_1_2_5_aligned()
29423   const std::int32_t zipped_chunk_size = (padded_k + 16) * 3;  in gemm_f_1_2_5_aligned()
29424   const std::int32_t zipped_rhs_size = (padded_k + 16) * n;  in gemm_f_1_2_5_aligned()
29428   std::int32_t* zipped_lhs_3_offsets =  in gemm_f_1_2_5_aligned()
29429       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 3);  in gemm_f_1_2_5_aligned()
29430   std::int32_t* zipped_lhs_1_offsets =  in gemm_f_1_2_5_aligned()
29431       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 1);  in gemm_f_1_2_5_aligned()
29434   const std::int32_t result_chunk_stride = result_stride * 3;  in gemm_f_1_2_5_aligned()
29436   const std::int32_t const_offset = lhs_offset * rhs_offset * k;  in gemm_f_1_2_5_aligned()
29439   const std::int32_t mul_result_chunk_stride_bytes = result_stride * 4;  in gemm_f_1_2_5_aligned()
29482                           const std::uint8_t* rhs, std::int32_t m,  in gemm_f_1_2_6_aligned()
29483                           std::int32_t n, std::int32_t k,  in gemm_f_1_2_6_aligned()
29484                           std::int32_t lhs_offset, std::int32_t rhs_offset,  in gemm_f_1_2_6_aligned()
29486                           std::int32_t result_stride) {  in gemm_f_1_2_6_aligned()
29487   const std::int32_t row_chunks = m / 3;  in gemm_f_1_2_6_aligned()
29488   const std::int32_t col_chunks = n / 3;  in gemm_f_1_2_6_aligned()
29489   const std::int32_t padded_k = ((k + 7) / 8) * 8;  in gemm_f_1_2_6_aligned()
29490   const std::int32_t chunk_size = k * 3;  in gemm_f_1_2_6_aligned()
29491   const std::int32_t zipped_chunk_size = (padded_k + 16) * 3;  in gemm_f_1_2_6_aligned()
29492   const std::int32_t zipped_rhs_size = (padded_k + 16) * n;  in gemm_f_1_2_6_aligned()
29496   std::int32_t* zipped_lhs_3_offsets =  in gemm_f_1_2_6_aligned()
29497       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 3);  in gemm_f_1_2_6_aligned()
29498   std::int32_t* zipped_lhs_1_offsets =  in gemm_f_1_2_6_aligned()
29499       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 1);  in gemm_f_1_2_6_aligned()
29502   const std::int32_t result_chunk_stride = result_stride * 3;  in gemm_f_1_2_6_aligned()
29504   const std::int32_t const_offset = lhs_offset * rhs_offset * k;  in gemm_f_1_2_6_aligned()
29507   const std::int32_t mul_result_chunk_stride_bytes = result_stride * 4;  in gemm_f_1_2_6_aligned()
29550                           const std::uint8_t* rhs, std::int32_t m,  in gemm_f_1_2_7_aligned()
29551                           std::int32_t n, std::int32_t k,  in gemm_f_1_2_7_aligned()
29552                           std::int32_t lhs_offset, std::int32_t rhs_offset,  in gemm_f_1_2_7_aligned()
29554                           std::int32_t result_stride) {  in gemm_f_1_2_7_aligned()
29555   const std::int32_t row_chunks = m / 3;  in gemm_f_1_2_7_aligned()
29556   const std::int32_t col_chunks = n / 3;  in gemm_f_1_2_7_aligned()
29557   const std::int32_t padded_k = ((k + 7) / 8) * 8;  in gemm_f_1_2_7_aligned()
29558   const std::int32_t chunk_size = k * 3;  in gemm_f_1_2_7_aligned()
29559   const std::int32_t zipped_chunk_size = (padded_k + 16) * 3;  in gemm_f_1_2_7_aligned()
29560   const std::int32_t zipped_rhs_size = (padded_k + 16) * n;  in gemm_f_1_2_7_aligned()
29564   std::int32_t* zipped_lhs_3_offsets =  in gemm_f_1_2_7_aligned()
29565       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 3);  in gemm_f_1_2_7_aligned()
29566   std::int32_t* zipped_lhs_1_offsets =  in gemm_f_1_2_7_aligned()
29567       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 1);  in gemm_f_1_2_7_aligned()
29570   const std::int32_t result_chunk_stride = result_stride * 3;  in gemm_f_1_2_7_aligned()
29572   const std::int32_t const_offset = lhs_offset * rhs_offset * k;  in gemm_f_1_2_7_aligned()
29575   const std::int32_t mul_result_chunk_stride_bytes = result_stride * 4;  in gemm_f_1_2_7_aligned()
29618                           const std::uint8_t* rhs, std::int32_t m,  in gemm_f_2_0_0_aligned()
29619                           std::int32_t n, std::int32_t k,  in gemm_f_2_0_0_aligned()
29620                           std::int32_t lhs_offset, std::int32_t rhs_offset,  in gemm_f_2_0_0_aligned()
29622                           std::int32_t result_stride) {  in gemm_f_2_0_0_aligned()
29623   const std::int32_t row_chunks = m / 3;  in gemm_f_2_0_0_aligned()
29624   const std::int32_t col_chunks = n / 3;  in gemm_f_2_0_0_aligned()
29625   const std::int32_t padded_k = ((k + 7) / 8) * 8;  in gemm_f_2_0_0_aligned()
29626   const std::int32_t chunk_size = k * 3;  in gemm_f_2_0_0_aligned()
29627   const std::int32_t zipped_chunk_size = (padded_k + 16) * 3;  in gemm_f_2_0_0_aligned()
29628   const std::int32_t zipped_rhs_size = (padded_k + 16) * n;  in gemm_f_2_0_0_aligned()
29632   std::int32_t* zipped_lhs_3_offsets =  in gemm_f_2_0_0_aligned()
29633       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 3);  in gemm_f_2_0_0_aligned()
29634   std::int32_t* zipped_lhs_2_offsets =  in gemm_f_2_0_0_aligned()
29635       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 2);  in gemm_f_2_0_0_aligned()
29638   const std::int32_t result_chunk_stride = result_stride * 3;  in gemm_f_2_0_0_aligned()
29640   const std::int32_t const_offset = lhs_offset * rhs_offset * k;  in gemm_f_2_0_0_aligned()
29643   const std::int32_t mul_result_chunk_stride_bytes = result_stride * 4;  in gemm_f_2_0_0_aligned()
29679                           const std::uint8_t* rhs, std::int32_t m,  in gemm_f_2_0_1_aligned()
29680                           std::int32_t n, std::int32_t k,  in gemm_f_2_0_1_aligned()
29681                           std::int32_t lhs_offset, std::int32_t rhs_offset,  in gemm_f_2_0_1_aligned()
29683                           std::int32_t result_stride) {  in gemm_f_2_0_1_aligned()
29684   const std::int32_t row_chunks = m / 3;  in gemm_f_2_0_1_aligned()
29685   const std::int32_t col_chunks = n / 3;  in gemm_f_2_0_1_aligned()
29686   const std::int32_t padded_k = ((k + 7) / 8) * 8;  in gemm_f_2_0_1_aligned()
29687   const std::int32_t chunk_size = k * 3;  in gemm_f_2_0_1_aligned()
29688   const std::int32_t zipped_chunk_size = (padded_k + 16) * 3;  in gemm_f_2_0_1_aligned()
29689   const std::int32_t zipped_rhs_size = (padded_k + 16) * n;  in gemm_f_2_0_1_aligned()
29693   std::int32_t* zipped_lhs_3_offsets =  in gemm_f_2_0_1_aligned()
29694       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 3);  in gemm_f_2_0_1_aligned()
29695   std::int32_t* zipped_lhs_2_offsets =  in gemm_f_2_0_1_aligned()
29696       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 2);  in gemm_f_2_0_1_aligned()
29699   const std::int32_t result_chunk_stride = result_stride * 3;  in gemm_f_2_0_1_aligned()
29701   const std::int32_t const_offset = lhs_offset * rhs_offset * k;  in gemm_f_2_0_1_aligned()
29704   const std::int32_t mul_result_chunk_stride_bytes = result_stride * 4;  in gemm_f_2_0_1_aligned()
29740                           const std::uint8_t* rhs, std::int32_t m,  in gemm_f_2_0_2_aligned()
29741                           std::int32_t n, std::int32_t k,  in gemm_f_2_0_2_aligned()
29742                           std::int32_t lhs_offset, std::int32_t rhs_offset,  in gemm_f_2_0_2_aligned()
29744                           std::int32_t result_stride) {  in gemm_f_2_0_2_aligned()
29745   const std::int32_t row_chunks = m / 3;  in gemm_f_2_0_2_aligned()
29746   const std::int32_t col_chunks = n / 3;  in gemm_f_2_0_2_aligned()
29747   const std::int32_t padded_k = ((k + 7) / 8) * 8;  in gemm_f_2_0_2_aligned()
29748   const std::int32_t chunk_size = k * 3;  in gemm_f_2_0_2_aligned()
29749   const std::int32_t zipped_chunk_size = (padded_k + 16) * 3;  in gemm_f_2_0_2_aligned()
29750   const std::int32_t zipped_rhs_size = (padded_k + 16) * n;  in gemm_f_2_0_2_aligned()
29754   std::int32_t* zipped_lhs_3_offsets =  in gemm_f_2_0_2_aligned()
29755       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 3);  in gemm_f_2_0_2_aligned()
29756   std::int32_t* zipped_lhs_2_offsets =  in gemm_f_2_0_2_aligned()
29757       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 2);  in gemm_f_2_0_2_aligned()
29760   const std::int32_t result_chunk_stride = result_stride * 3;  in gemm_f_2_0_2_aligned()
29762   const std::int32_t const_offset = lhs_offset * rhs_offset * k;  in gemm_f_2_0_2_aligned()
29765   const std::int32_t mul_result_chunk_stride_bytes = result_stride * 4;  in gemm_f_2_0_2_aligned()
29801                           const std::uint8_t* rhs, std::int32_t m,  in gemm_f_2_0_3_aligned()
29802                           std::int32_t n, std::int32_t k,  in gemm_f_2_0_3_aligned()
29803                           std::int32_t lhs_offset, std::int32_t rhs_offset,  in gemm_f_2_0_3_aligned()
29805                           std::int32_t result_stride) {  in gemm_f_2_0_3_aligned()
29806   const std::int32_t row_chunks = m / 3;  in gemm_f_2_0_3_aligned()
29807   const std::int32_t col_chunks = n / 3;  in gemm_f_2_0_3_aligned()
29808   const std::int32_t padded_k = ((k + 7) / 8) * 8;  in gemm_f_2_0_3_aligned()
29809   const std::int32_t chunk_size = k * 3;  in gemm_f_2_0_3_aligned()
29810   const std::int32_t zipped_chunk_size = (padded_k + 16) * 3;  in gemm_f_2_0_3_aligned()
29811   const std::int32_t zipped_rhs_size = (padded_k + 16) * n;  in gemm_f_2_0_3_aligned()
29815   std::int32_t* zipped_lhs_3_offsets =  in gemm_f_2_0_3_aligned()
29816       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 3);  in gemm_f_2_0_3_aligned()
29817   std::int32_t* zipped_lhs_2_offsets =  in gemm_f_2_0_3_aligned()
29818       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 2);  in gemm_f_2_0_3_aligned()
29821   const std::int32_t result_chunk_stride = result_stride * 3;  in gemm_f_2_0_3_aligned()
29823   const std::int32_t const_offset = lhs_offset * rhs_offset * k;  in gemm_f_2_0_3_aligned()
29826   const std::int32_t mul_result_chunk_stride_bytes = result_stride * 4;  in gemm_f_2_0_3_aligned()
29862                           const std::uint8_t* rhs, std::int32_t m,  in gemm_f_2_0_4_aligned()
29863                           std::int32_t n, std::int32_t k,  in gemm_f_2_0_4_aligned()
29864                           std::int32_t lhs_offset, std::int32_t rhs_offset,  in gemm_f_2_0_4_aligned()
29866                           std::int32_t result_stride) {  in gemm_f_2_0_4_aligned()
29867   const std::int32_t row_chunks = m / 3;  in gemm_f_2_0_4_aligned()
29868   const std::int32_t col_chunks = n / 3;  in gemm_f_2_0_4_aligned()
29869   const std::int32_t padded_k = ((k + 7) / 8) * 8;  in gemm_f_2_0_4_aligned()
29870   const std::int32_t chunk_size = k * 3;  in gemm_f_2_0_4_aligned()
29871   const std::int32_t zipped_chunk_size = (padded_k + 16) * 3;  in gemm_f_2_0_4_aligned()
29872   const std::int32_t zipped_rhs_size = (padded_k + 16) * n;  in gemm_f_2_0_4_aligned()
29876   std::int32_t* zipped_lhs_3_offsets =  in gemm_f_2_0_4_aligned()
29877       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 3);  in gemm_f_2_0_4_aligned()
29878   std::int32_t* zipped_lhs_2_offsets =  in gemm_f_2_0_4_aligned()
29879       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 2);  in gemm_f_2_0_4_aligned()
29882   const std::int32_t result_chunk_stride = result_stride * 3;  in gemm_f_2_0_4_aligned()
29884   const std::int32_t const_offset = lhs_offset * rhs_offset * k;  in gemm_f_2_0_4_aligned()
29887   const std::int32_t mul_result_chunk_stride_bytes = result_stride * 4;  in gemm_f_2_0_4_aligned()
29923                           const std::uint8_t* rhs, std::int32_t m,  in gemm_f_2_0_5_aligned()
29924                           std::int32_t n, std::int32_t k,  in gemm_f_2_0_5_aligned()
29925                           std::int32_t lhs_offset, std::int32_t rhs_offset,  in gemm_f_2_0_5_aligned()
29927                           std::int32_t result_stride) {  in gemm_f_2_0_5_aligned()
29928   const std::int32_t row_chunks = m / 3;  in gemm_f_2_0_5_aligned()
29929   const std::int32_t col_chunks = n / 3;  in gemm_f_2_0_5_aligned()
29930   const std::int32_t padded_k = ((k + 7) / 8) * 8;  in gemm_f_2_0_5_aligned()
29931   const std::int32_t chunk_size = k * 3;  in gemm_f_2_0_5_aligned()
29932   const std::int32_t zipped_chunk_size = (padded_k + 16) * 3;  in gemm_f_2_0_5_aligned()
29933   const std::int32_t zipped_rhs_size = (padded_k + 16) * n;  in gemm_f_2_0_5_aligned()
29937   std::int32_t* zipped_lhs_3_offsets =  in gemm_f_2_0_5_aligned()
29938       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 3);  in gemm_f_2_0_5_aligned()
29939   std::int32_t* zipped_lhs_2_offsets =  in gemm_f_2_0_5_aligned()
29940       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 2);  in gemm_f_2_0_5_aligned()
29943   const std::int32_t result_chunk_stride = result_stride * 3;  in gemm_f_2_0_5_aligned()
29945   const std::int32_t const_offset = lhs_offset * rhs_offset * k;  in gemm_f_2_0_5_aligned()
29948   const std::int32_t mul_result_chunk_stride_bytes = result_stride * 4;  in gemm_f_2_0_5_aligned()
29984                           const std::uint8_t* rhs, std::int32_t m,  in gemm_f_2_0_6_aligned()
29985                           std::int32_t n, std::int32_t k,  in gemm_f_2_0_6_aligned()
29986                           std::int32_t lhs_offset, std::int32_t rhs_offset,  in gemm_f_2_0_6_aligned()
29988                           std::int32_t result_stride) {  in gemm_f_2_0_6_aligned()
29989   const std::int32_t row_chunks = m / 3;  in gemm_f_2_0_6_aligned()
29990   const std::int32_t col_chunks = n / 3;  in gemm_f_2_0_6_aligned()
29991   const std::int32_t padded_k = ((k + 7) / 8) * 8;  in gemm_f_2_0_6_aligned()
29992   const std::int32_t chunk_size = k * 3;  in gemm_f_2_0_6_aligned()
29993   const std::int32_t zipped_chunk_size = (padded_k + 16) * 3;  in gemm_f_2_0_6_aligned()
29994   const std::int32_t zipped_rhs_size = (padded_k + 16) * n;  in gemm_f_2_0_6_aligned()
29998   std::int32_t* zipped_lhs_3_offsets =  in gemm_f_2_0_6_aligned()
29999       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 3);  in gemm_f_2_0_6_aligned()
30000   std::int32_t* zipped_lhs_2_offsets =  in gemm_f_2_0_6_aligned()
30001       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 2);  in gemm_f_2_0_6_aligned()
30004   const std::int32_t result_chunk_stride = result_stride * 3;  in gemm_f_2_0_6_aligned()
30006   const std::int32_t const_offset = lhs_offset * rhs_offset * k;  in gemm_f_2_0_6_aligned()
30009   const std::int32_t mul_result_chunk_stride_bytes = result_stride * 4;  in gemm_f_2_0_6_aligned()
30045                           const std::uint8_t* rhs, std::int32_t m,  in gemm_f_2_0_7_aligned()
30046                           std::int32_t n, std::int32_t k,  in gemm_f_2_0_7_aligned()
30047                           std::int32_t lhs_offset, std::int32_t rhs_offset,  in gemm_f_2_0_7_aligned()
30049                           std::int32_t result_stride) {  in gemm_f_2_0_7_aligned()
30050   const std::int32_t row_chunks = m / 3;  in gemm_f_2_0_7_aligned()
30051   const std::int32_t col_chunks = n / 3;  in gemm_f_2_0_7_aligned()
30052   const std::int32_t padded_k = ((k + 7) / 8) * 8;  in gemm_f_2_0_7_aligned()
30053   const std::int32_t chunk_size = k * 3;  in gemm_f_2_0_7_aligned()
30054   const std::int32_t zipped_chunk_size = (padded_k + 16) * 3;  in gemm_f_2_0_7_aligned()
30055   const std::int32_t zipped_rhs_size = (padded_k + 16) * n;  in gemm_f_2_0_7_aligned()
30059   std::int32_t* zipped_lhs_3_offsets =  in gemm_f_2_0_7_aligned()
30060       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 3);  in gemm_f_2_0_7_aligned()
30061   std::int32_t* zipped_lhs_2_offsets =  in gemm_f_2_0_7_aligned()
30062       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 2);  in gemm_f_2_0_7_aligned()
30065   const std::int32_t result_chunk_stride = result_stride * 3;  in gemm_f_2_0_7_aligned()
30067   const std::int32_t const_offset = lhs_offset * rhs_offset * k;  in gemm_f_2_0_7_aligned()
30070   const std::int32_t mul_result_chunk_stride_bytes = result_stride * 4;  in gemm_f_2_0_7_aligned()
30106                           const std::uint8_t* rhs, std::int32_t m,  in gemm_f_2_1_0_aligned()
30107                           std::int32_t n, std::int32_t k,  in gemm_f_2_1_0_aligned()
30108                           std::int32_t lhs_offset, std::int32_t rhs_offset,  in gemm_f_2_1_0_aligned()
30110                           std::int32_t result_stride) {  in gemm_f_2_1_0_aligned()
30111   const std::int32_t row_chunks = m / 3;  in gemm_f_2_1_0_aligned()
30112   const std::int32_t col_chunks = n / 3;  in gemm_f_2_1_0_aligned()
30113   const std::int32_t padded_k = ((k + 7) / 8) * 8;  in gemm_f_2_1_0_aligned()
30114   const std::int32_t chunk_size = k * 3;  in gemm_f_2_1_0_aligned()
30115   const std::int32_t zipped_chunk_size = (padded_k + 16) * 3;  in gemm_f_2_1_0_aligned()
30116   const std::int32_t zipped_rhs_size = (padded_k + 16) * n;  in gemm_f_2_1_0_aligned()
30120   std::int32_t* zipped_lhs_3_offsets =  in gemm_f_2_1_0_aligned()
30121       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 3);  in gemm_f_2_1_0_aligned()
30122   std::int32_t* zipped_lhs_2_offsets =  in gemm_f_2_1_0_aligned()
30123       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 2);  in gemm_f_2_1_0_aligned()
30126   const std::int32_t result_chunk_stride = result_stride * 3;  in gemm_f_2_1_0_aligned()
30128   const std::int32_t const_offset = lhs_offset * rhs_offset * k;  in gemm_f_2_1_0_aligned()
30131   const std::int32_t mul_result_chunk_stride_bytes = result_stride * 4;  in gemm_f_2_1_0_aligned()
30174                           const std::uint8_t* rhs, std::int32_t m,  in gemm_f_2_1_1_aligned()
30175                           std::int32_t n, std::int32_t k,  in gemm_f_2_1_1_aligned()
30176                           std::int32_t lhs_offset, std::int32_t rhs_offset,  in gemm_f_2_1_1_aligned()
30178                           std::int32_t result_stride) {  in gemm_f_2_1_1_aligned()
30179   const std::int32_t row_chunks = m / 3;  in gemm_f_2_1_1_aligned()
30180   const std::int32_t col_chunks = n / 3;  in gemm_f_2_1_1_aligned()
30181   const std::int32_t padded_k = ((k + 7) / 8) * 8;  in gemm_f_2_1_1_aligned()
30182   const std::int32_t chunk_size = k * 3;  in gemm_f_2_1_1_aligned()
30183   const std::int32_t zipped_chunk_size = (padded_k + 16) * 3;  in gemm_f_2_1_1_aligned()
30184   const std::int32_t zipped_rhs_size = (padded_k + 16) * n;  in gemm_f_2_1_1_aligned()
30188   std::int32_t* zipped_lhs_3_offsets =  in gemm_f_2_1_1_aligned()
30189       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 3);  in gemm_f_2_1_1_aligned()
30190   std::int32_t* zipped_lhs_2_offsets =  in gemm_f_2_1_1_aligned()
30191       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 2);  in gemm_f_2_1_1_aligned()
30194   const std::int32_t result_chunk_stride = result_stride * 3;  in gemm_f_2_1_1_aligned()
30196   const std::int32_t const_offset = lhs_offset * rhs_offset * k;  in gemm_f_2_1_1_aligned()
30199   const std::int32_t mul_result_chunk_stride_bytes = result_stride * 4;  in gemm_f_2_1_1_aligned()
30242                           const std::uint8_t* rhs, std::int32_t m,  in gemm_f_2_1_2_aligned()
30243                           std::int32_t n, std::int32_t k,  in gemm_f_2_1_2_aligned()
30244                           std::int32_t lhs_offset, std::int32_t rhs_offset,  in gemm_f_2_1_2_aligned()
30246                           std::int32_t result_stride) {  in gemm_f_2_1_2_aligned()
30247   const std::int32_t row_chunks = m / 3;  in gemm_f_2_1_2_aligned()
30248   const std::int32_t col_chunks = n / 3;  in gemm_f_2_1_2_aligned()
30249   const std::int32_t padded_k = ((k + 7) / 8) * 8;  in gemm_f_2_1_2_aligned()
30250   const std::int32_t chunk_size = k * 3;  in gemm_f_2_1_2_aligned()
30251   const std::int32_t zipped_chunk_size = (padded_k + 16) * 3;  in gemm_f_2_1_2_aligned()
30252   const std::int32_t zipped_rhs_size = (padded_k + 16) * n;  in gemm_f_2_1_2_aligned()
30256   std::int32_t* zipped_lhs_3_offsets =  in gemm_f_2_1_2_aligned()
30257       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 3);  in gemm_f_2_1_2_aligned()
30258   std::int32_t* zipped_lhs_2_offsets =  in gemm_f_2_1_2_aligned()
30259       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 2);  in gemm_f_2_1_2_aligned()
30262   const std::int32_t result_chunk_stride = result_stride * 3;  in gemm_f_2_1_2_aligned()
30264   const std::int32_t const_offset = lhs_offset * rhs_offset * k;  in gemm_f_2_1_2_aligned()
30267   const std::int32_t mul_result_chunk_stride_bytes = result_stride * 4;  in gemm_f_2_1_2_aligned()
30310                           const std::uint8_t* rhs, std::int32_t m,  in gemm_f_2_1_3_aligned()
30311                           std::int32_t n, std::int32_t k,  in gemm_f_2_1_3_aligned()
30312                           std::int32_t lhs_offset, std::int32_t rhs_offset,  in gemm_f_2_1_3_aligned()
30314                           std::int32_t result_stride) {  in gemm_f_2_1_3_aligned()
30315   const std::int32_t row_chunks = m / 3;  in gemm_f_2_1_3_aligned()
30316   const std::int32_t col_chunks = n / 3;  in gemm_f_2_1_3_aligned()
30317   const std::int32_t padded_k = ((k + 7) / 8) * 8;  in gemm_f_2_1_3_aligned()
30318   const std::int32_t chunk_size = k * 3;  in gemm_f_2_1_3_aligned()
30319   const std::int32_t zipped_chunk_size = (padded_k + 16) * 3;  in gemm_f_2_1_3_aligned()
30320   const std::int32_t zipped_rhs_size = (padded_k + 16) * n;  in gemm_f_2_1_3_aligned()
30324   std::int32_t* zipped_lhs_3_offsets =  in gemm_f_2_1_3_aligned()
30325       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 3);  in gemm_f_2_1_3_aligned()
30326   std::int32_t* zipped_lhs_2_offsets =  in gemm_f_2_1_3_aligned()
30327       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 2);  in gemm_f_2_1_3_aligned()
30330   const std::int32_t result_chunk_stride = result_stride * 3;  in gemm_f_2_1_3_aligned()
30332   const std::int32_t const_offset = lhs_offset * rhs_offset * k;  in gemm_f_2_1_3_aligned()
30335   const std::int32_t mul_result_chunk_stride_bytes = result_stride * 4;  in gemm_f_2_1_3_aligned()
30378                           const std::uint8_t* rhs, std::int32_t m,  in gemm_f_2_1_4_aligned()
30379                           std::int32_t n, std::int32_t k,  in gemm_f_2_1_4_aligned()
30380                           std::int32_t lhs_offset, std::int32_t rhs_offset,  in gemm_f_2_1_4_aligned()
30382                           std::int32_t result_stride) {  in gemm_f_2_1_4_aligned()
30383   const std::int32_t row_chunks = m / 3;  in gemm_f_2_1_4_aligned()
30384   const std::int32_t col_chunks = n / 3;  in gemm_f_2_1_4_aligned()
30385   const std::int32_t padded_k = ((k + 7) / 8) * 8;  in gemm_f_2_1_4_aligned()
30386   const std::int32_t chunk_size = k * 3;  in gemm_f_2_1_4_aligned()
30387   const std::int32_t zipped_chunk_size = (padded_k + 16) * 3;  in gemm_f_2_1_4_aligned()
30388   const std::int32_t zipped_rhs_size = (padded_k + 16) * n;  in gemm_f_2_1_4_aligned()
30392   std::int32_t* zipped_lhs_3_offsets =  in gemm_f_2_1_4_aligned()
30393       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 3);  in gemm_f_2_1_4_aligned()
30394   std::int32_t* zipped_lhs_2_offsets =  in gemm_f_2_1_4_aligned()
30395       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 2);  in gemm_f_2_1_4_aligned()
30398   const std::int32_t result_chunk_stride = result_stride * 3;  in gemm_f_2_1_4_aligned()
30400   const std::int32_t const_offset = lhs_offset * rhs_offset * k;  in gemm_f_2_1_4_aligned()
30403   const std::int32_t mul_result_chunk_stride_bytes = result_stride * 4;  in gemm_f_2_1_4_aligned()
30446                           const std::uint8_t* rhs, std::int32_t m,  in gemm_f_2_1_5_aligned()
30447                           std::int32_t n, std::int32_t k,  in gemm_f_2_1_5_aligned()
30448                           std::int32_t lhs_offset, std::int32_t rhs_offset,  in gemm_f_2_1_5_aligned()
30450                           std::int32_t result_stride) {  in gemm_f_2_1_5_aligned()
30451   const std::int32_t row_chunks = m / 3;  in gemm_f_2_1_5_aligned()
30452   const std::int32_t col_chunks = n / 3;  in gemm_f_2_1_5_aligned()
30453   const std::int32_t padded_k = ((k + 7) / 8) * 8;  in gemm_f_2_1_5_aligned()
30454   const std::int32_t chunk_size = k * 3;  in gemm_f_2_1_5_aligned()
30455   const std::int32_t zipped_chunk_size = (padded_k + 16) * 3;  in gemm_f_2_1_5_aligned()
30456   const std::int32_t zipped_rhs_size = (padded_k + 16) * n;  in gemm_f_2_1_5_aligned()
30460   std::int32_t* zipped_lhs_3_offsets =  in gemm_f_2_1_5_aligned()
30461       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 3);  in gemm_f_2_1_5_aligned()
30462   std::int32_t* zipped_lhs_2_offsets =  in gemm_f_2_1_5_aligned()
30463       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 2);  in gemm_f_2_1_5_aligned()
30466   const std::int32_t result_chunk_stride = result_stride * 3;  in gemm_f_2_1_5_aligned()
30468   const std::int32_t const_offset = lhs_offset * rhs_offset * k;  in gemm_f_2_1_5_aligned()
30471   const std::int32_t mul_result_chunk_stride_bytes = result_stride * 4;  in gemm_f_2_1_5_aligned()
30514                           const std::uint8_t* rhs, std::int32_t m,  in gemm_f_2_1_6_aligned()
30515                           std::int32_t n, std::int32_t k,  in gemm_f_2_1_6_aligned()
30516                           std::int32_t lhs_offset, std::int32_t rhs_offset,  in gemm_f_2_1_6_aligned()
30518                           std::int32_t result_stride) {  in gemm_f_2_1_6_aligned()
30519   const std::int32_t row_chunks = m / 3;  in gemm_f_2_1_6_aligned()
30520   const std::int32_t col_chunks = n / 3;  in gemm_f_2_1_6_aligned()
30521   const std::int32_t padded_k = ((k + 7) / 8) * 8;  in gemm_f_2_1_6_aligned()
30522   const std::int32_t chunk_size = k * 3;  in gemm_f_2_1_6_aligned()
30523   const std::int32_t zipped_chunk_size = (padded_k + 16) * 3;  in gemm_f_2_1_6_aligned()
30524   const std::int32_t zipped_rhs_size = (padded_k + 16) * n;  in gemm_f_2_1_6_aligned()
30528   std::int32_t* zipped_lhs_3_offsets =  in gemm_f_2_1_6_aligned()
30529       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 3);  in gemm_f_2_1_6_aligned()
30530   std::int32_t* zipped_lhs_2_offsets =  in gemm_f_2_1_6_aligned()
30531       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 2);  in gemm_f_2_1_6_aligned()
30534   const std::int32_t result_chunk_stride = result_stride * 3;  in gemm_f_2_1_6_aligned()
30536   const std::int32_t const_offset = lhs_offset * rhs_offset * k;  in gemm_f_2_1_6_aligned()
30539   const std::int32_t mul_result_chunk_stride_bytes = result_stride * 4;  in gemm_f_2_1_6_aligned()
30582                           const std::uint8_t* rhs, std::int32_t m,  in gemm_f_2_1_7_aligned()
30583                           std::int32_t n, std::int32_t k,  in gemm_f_2_1_7_aligned()
30584                           std::int32_t lhs_offset, std::int32_t rhs_offset,  in gemm_f_2_1_7_aligned()
30586                           std::int32_t result_stride) {  in gemm_f_2_1_7_aligned()
30587   const std::int32_t row_chunks = m / 3;  in gemm_f_2_1_7_aligned()
30588   const std::int32_t col_chunks = n / 3;  in gemm_f_2_1_7_aligned()
30589   const std::int32_t padded_k = ((k + 7) / 8) * 8;  in gemm_f_2_1_7_aligned()
30590   const std::int32_t chunk_size = k * 3;  in gemm_f_2_1_7_aligned()
30591   const std::int32_t zipped_chunk_size = (padded_k + 16) * 3;  in gemm_f_2_1_7_aligned()
30592   const std::int32_t zipped_rhs_size = (padded_k + 16) * n;  in gemm_f_2_1_7_aligned()
30596   std::int32_t* zipped_lhs_3_offsets =  in gemm_f_2_1_7_aligned()
30597       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 3);  in gemm_f_2_1_7_aligned()
30598   std::int32_t* zipped_lhs_2_offsets =  in gemm_f_2_1_7_aligned()
30599       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 2);  in gemm_f_2_1_7_aligned()
30602   const std::int32_t result_chunk_stride = result_stride * 3;  in gemm_f_2_1_7_aligned()
30604   const std::int32_t const_offset = lhs_offset * rhs_offset * k;  in gemm_f_2_1_7_aligned()
30607   const std::int32_t mul_result_chunk_stride_bytes = result_stride * 4;  in gemm_f_2_1_7_aligned()
30650                           const std::uint8_t* rhs, std::int32_t m,  in gemm_f_2_2_0_aligned()
30651                           std::int32_t n, std::int32_t k,  in gemm_f_2_2_0_aligned()
30652                           std::int32_t lhs_offset, std::int32_t rhs_offset,  in gemm_f_2_2_0_aligned()
30654                           std::int32_t result_stride) {  in gemm_f_2_2_0_aligned()
30655   const std::int32_t row_chunks = m / 3;  in gemm_f_2_2_0_aligned()
30656   const std::int32_t col_chunks = n / 3;  in gemm_f_2_2_0_aligned()
30657   const std::int32_t padded_k = ((k + 7) / 8) * 8;  in gemm_f_2_2_0_aligned()
30658   const std::int32_t chunk_size = k * 3;  in gemm_f_2_2_0_aligned()
30659   const std::int32_t zipped_chunk_size = (padded_k + 16) * 3;  in gemm_f_2_2_0_aligned()
30660   const std::int32_t zipped_rhs_size = (padded_k + 16) * n;  in gemm_f_2_2_0_aligned()
30664   std::int32_t* zipped_lhs_3_offsets =  in gemm_f_2_2_0_aligned()
30665       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 3);  in gemm_f_2_2_0_aligned()
30666   std::int32_t* zipped_lhs_2_offsets =  in gemm_f_2_2_0_aligned()
30667       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 2);  in gemm_f_2_2_0_aligned()
30670   const std::int32_t result_chunk_stride = result_stride * 3;  in gemm_f_2_2_0_aligned()
30672   const std::int32_t const_offset = lhs_offset * rhs_offset * k;  in gemm_f_2_2_0_aligned()
30675   const std::int32_t mul_result_chunk_stride_bytes = result_stride * 4;  in gemm_f_2_2_0_aligned()
30718                           const std::uint8_t* rhs, std::int32_t m,  in gemm_f_2_2_1_aligned()
30719                           std::int32_t n, std::int32_t k,  in gemm_f_2_2_1_aligned()
30720                           std::int32_t lhs_offset, std::int32_t rhs_offset,  in gemm_f_2_2_1_aligned()
30722                           std::int32_t result_stride) {  in gemm_f_2_2_1_aligned()
30723   const std::int32_t row_chunks = m / 3;  in gemm_f_2_2_1_aligned()
30724   const std::int32_t col_chunks = n / 3;  in gemm_f_2_2_1_aligned()
30725   const std::int32_t padded_k = ((k + 7) / 8) * 8;  in gemm_f_2_2_1_aligned()
30726   const std::int32_t chunk_size = k * 3;  in gemm_f_2_2_1_aligned()
30727   const std::int32_t zipped_chunk_size = (padded_k + 16) * 3;  in gemm_f_2_2_1_aligned()
30728   const std::int32_t zipped_rhs_size = (padded_k + 16) * n;  in gemm_f_2_2_1_aligned()
30732   std::int32_t* zipped_lhs_3_offsets =  in gemm_f_2_2_1_aligned()
30733       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 3);  in gemm_f_2_2_1_aligned()
30734   std::int32_t* zipped_lhs_2_offsets =  in gemm_f_2_2_1_aligned()
30735       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 2);  in gemm_f_2_2_1_aligned()
30738   const std::int32_t result_chunk_stride = result_stride * 3;  in gemm_f_2_2_1_aligned()
30740   const std::int32_t const_offset = lhs_offset * rhs_offset * k;  in gemm_f_2_2_1_aligned()
30743   const std::int32_t mul_result_chunk_stride_bytes = result_stride * 4;  in gemm_f_2_2_1_aligned()
30786                           const std::uint8_t* rhs, std::int32_t m,  in gemm_f_2_2_2_aligned()
30787                           std::int32_t n, std::int32_t k,  in gemm_f_2_2_2_aligned()
30788                           std::int32_t lhs_offset, std::int32_t rhs_offset,  in gemm_f_2_2_2_aligned()
30790                           std::int32_t result_stride) {  in gemm_f_2_2_2_aligned()
30791   const std::int32_t row_chunks = m / 3;  in gemm_f_2_2_2_aligned()
30792   const std::int32_t col_chunks = n / 3;  in gemm_f_2_2_2_aligned()
30793   const std::int32_t padded_k = ((k + 7) / 8) * 8;  in gemm_f_2_2_2_aligned()
30794   const std::int32_t chunk_size = k * 3;  in gemm_f_2_2_2_aligned()
30795   const std::int32_t zipped_chunk_size = (padded_k + 16) * 3;  in gemm_f_2_2_2_aligned()
30796   const std::int32_t zipped_rhs_size = (padded_k + 16) * n;  in gemm_f_2_2_2_aligned()
30800   std::int32_t* zipped_lhs_3_offsets =  in gemm_f_2_2_2_aligned()
30801       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 3);  in gemm_f_2_2_2_aligned()
30802   std::int32_t* zipped_lhs_2_offsets =  in gemm_f_2_2_2_aligned()
30803       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 2);  in gemm_f_2_2_2_aligned()
30806   const std::int32_t result_chunk_stride = result_stride * 3;  in gemm_f_2_2_2_aligned()
30808   const std::int32_t const_offset = lhs_offset * rhs_offset * k;  in gemm_f_2_2_2_aligned()
30811   const std::int32_t mul_result_chunk_stride_bytes = result_stride * 4;  in gemm_f_2_2_2_aligned()
30854                           const std::uint8_t* rhs, std::int32_t m,  in gemm_f_2_2_3_aligned()
30855                           std::int32_t n, std::int32_t k,  in gemm_f_2_2_3_aligned()
30856                           std::int32_t lhs_offset, std::int32_t rhs_offset,  in gemm_f_2_2_3_aligned()
30858                           std::int32_t result_stride) {  in gemm_f_2_2_3_aligned()
30859   const std::int32_t row_chunks = m / 3;  in gemm_f_2_2_3_aligned()
30860   const std::int32_t col_chunks = n / 3;  in gemm_f_2_2_3_aligned()
30861   const std::int32_t padded_k = ((k + 7) / 8) * 8;  in gemm_f_2_2_3_aligned()
30862   const std::int32_t chunk_size = k * 3;  in gemm_f_2_2_3_aligned()
30863   const std::int32_t zipped_chunk_size = (padded_k + 16) * 3;  in gemm_f_2_2_3_aligned()
30864   const std::int32_t zipped_rhs_size = (padded_k + 16) * n;  in gemm_f_2_2_3_aligned()
30868   std::int32_t* zipped_lhs_3_offsets =  in gemm_f_2_2_3_aligned()
30869       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 3);  in gemm_f_2_2_3_aligned()
30870   std::int32_t* zipped_lhs_2_offsets =  in gemm_f_2_2_3_aligned()
30871       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 2);  in gemm_f_2_2_3_aligned()
30874   const std::int32_t result_chunk_stride = result_stride * 3;  in gemm_f_2_2_3_aligned()
30876   const std::int32_t const_offset = lhs_offset * rhs_offset * k;  in gemm_f_2_2_3_aligned()
30879   const std::int32_t mul_result_chunk_stride_bytes = result_stride * 4;  in gemm_f_2_2_3_aligned()
30922                           const std::uint8_t* rhs, std::int32_t m,  in gemm_f_2_2_4_aligned()
30923                           std::int32_t n, std::int32_t k,  in gemm_f_2_2_4_aligned()
30924                           std::int32_t lhs_offset, std::int32_t rhs_offset,  in gemm_f_2_2_4_aligned()
30926                           std::int32_t result_stride) {  in gemm_f_2_2_4_aligned()
30927   const std::int32_t row_chunks = m / 3;  in gemm_f_2_2_4_aligned()
30928   const std::int32_t col_chunks = n / 3;  in gemm_f_2_2_4_aligned()
30929   const std::int32_t padded_k = ((k + 7) / 8) * 8;  in gemm_f_2_2_4_aligned()
30930   const std::int32_t chunk_size = k * 3;  in gemm_f_2_2_4_aligned()
30931   const std::int32_t zipped_chunk_size = (padded_k + 16) * 3;  in gemm_f_2_2_4_aligned()
30932   const std::int32_t zipped_rhs_size = (padded_k + 16) * n;  in gemm_f_2_2_4_aligned()
30936   std::int32_t* zipped_lhs_3_offsets =  in gemm_f_2_2_4_aligned()
30937       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 3);  in gemm_f_2_2_4_aligned()
30938   std::int32_t* zipped_lhs_2_offsets =  in gemm_f_2_2_4_aligned()
30939       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 2);  in gemm_f_2_2_4_aligned()
30942   const std::int32_t result_chunk_stride = result_stride * 3;  in gemm_f_2_2_4_aligned()
30944   const std::int32_t const_offset = lhs_offset * rhs_offset * k;  in gemm_f_2_2_4_aligned()
30947   const std::int32_t mul_result_chunk_stride_bytes = result_stride * 4;  in gemm_f_2_2_4_aligned()
30990                           const std::uint8_t* rhs, std::int32_t m,  in gemm_f_2_2_5_aligned()
30991                           std::int32_t n, std::int32_t k,  in gemm_f_2_2_5_aligned()
30992                           std::int32_t lhs_offset, std::int32_t rhs_offset,  in gemm_f_2_2_5_aligned()
30994                           std::int32_t result_stride) {  in gemm_f_2_2_5_aligned()
30995   const std::int32_t row_chunks = m / 3;  in gemm_f_2_2_5_aligned()
30996   const std::int32_t col_chunks = n / 3;  in gemm_f_2_2_5_aligned()
30997   const std::int32_t padded_k = ((k + 7) / 8) * 8;  in gemm_f_2_2_5_aligned()
30998   const std::int32_t chunk_size = k * 3;  in gemm_f_2_2_5_aligned()
30999   const std::int32_t zipped_chunk_size = (padded_k + 16) * 3;  in gemm_f_2_2_5_aligned()
31000   const std::int32_t zipped_rhs_size = (padded_k + 16) * n;  in gemm_f_2_2_5_aligned()
31004   std::int32_t* zipped_lhs_3_offsets =  in gemm_f_2_2_5_aligned()
31005       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 3);  in gemm_f_2_2_5_aligned()
31006   std::int32_t* zipped_lhs_2_offsets =  in gemm_f_2_2_5_aligned()
31007       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 2);  in gemm_f_2_2_5_aligned()
31010   const std::int32_t result_chunk_stride = result_stride * 3;  in gemm_f_2_2_5_aligned()
31012   const std::int32_t const_offset = lhs_offset * rhs_offset * k;  in gemm_f_2_2_5_aligned()
31015   const std::int32_t mul_result_chunk_stride_bytes = result_stride * 4;  in gemm_f_2_2_5_aligned()
31058                           const std::uint8_t* rhs, std::int32_t m,  in gemm_f_2_2_6_aligned()
31059                           std::int32_t n, std::int32_t k,  in gemm_f_2_2_6_aligned()
31060                           std::int32_t lhs_offset, std::int32_t rhs_offset,  in gemm_f_2_2_6_aligned()
31062                           std::int32_t result_stride) {  in gemm_f_2_2_6_aligned()
31063   const std::int32_t row_chunks = m / 3;  in gemm_f_2_2_6_aligned()
31064   const std::int32_t col_chunks = n / 3;  in gemm_f_2_2_6_aligned()
31065   const std::int32_t padded_k = ((k + 7) / 8) * 8;  in gemm_f_2_2_6_aligned()
31066   const std::int32_t chunk_size = k * 3;  in gemm_f_2_2_6_aligned()
31067   const std::int32_t zipped_chunk_size = (padded_k + 16) * 3;  in gemm_f_2_2_6_aligned()
31068   const std::int32_t zipped_rhs_size = (padded_k + 16) * n;  in gemm_f_2_2_6_aligned()
31072   std::int32_t* zipped_lhs_3_offsets =  in gemm_f_2_2_6_aligned()
31073       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 3);  in gemm_f_2_2_6_aligned()
31074   std::int32_t* zipped_lhs_2_offsets =  in gemm_f_2_2_6_aligned()
31075       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 2);  in gemm_f_2_2_6_aligned()
31078   const std::int32_t result_chunk_stride = result_stride * 3;  in gemm_f_2_2_6_aligned()
31080   const std::int32_t const_offset = lhs_offset * rhs_offset * k;  in gemm_f_2_2_6_aligned()
31083   const std::int32_t mul_result_chunk_stride_bytes = result_stride * 4;  in gemm_f_2_2_6_aligned()
31126                           const std::uint8_t* rhs, std::int32_t m,  in gemm_f_2_2_7_aligned()
31127                           std::int32_t n, std::int32_t k,  in gemm_f_2_2_7_aligned()
31128                           std::int32_t lhs_offset, std::int32_t rhs_offset,  in gemm_f_2_2_7_aligned()
31130                           std::int32_t result_stride) {  in gemm_f_2_2_7_aligned()
31131   const std::int32_t row_chunks = m / 3;  in gemm_f_2_2_7_aligned()
31132   const std::int32_t col_chunks = n / 3;  in gemm_f_2_2_7_aligned()
31133   const std::int32_t padded_k = ((k + 7) / 8) * 8;  in gemm_f_2_2_7_aligned()
31134   const std::int32_t chunk_size = k * 3;  in gemm_f_2_2_7_aligned()
31135   const std::int32_t zipped_chunk_size = (padded_k + 16) * 3;  in gemm_f_2_2_7_aligned()
31136   const std::int32_t zipped_rhs_size = (padded_k + 16) * n;  in gemm_f_2_2_7_aligned()
31140   std::int32_t* zipped_lhs_3_offsets =  in gemm_f_2_2_7_aligned()
31141       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 3);  in gemm_f_2_2_7_aligned()
31142   std::int32_t* zipped_lhs_2_offsets =  in gemm_f_2_2_7_aligned()
31143       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 2);  in gemm_f_2_2_7_aligned()
31146   const std::int32_t result_chunk_stride = result_stride * 3;  in gemm_f_2_2_7_aligned()
31148   const std::int32_t const_offset = lhs_offset * rhs_offset * k;  in gemm_f_2_2_7_aligned()
31151   const std::int32_t mul_result_chunk_stride_bytes = result_stride * 4;  in gemm_f_2_2_7_aligned()
31194                   const std::uint8_t* rhs, std::int32_t m, std::int32_t n,  in gemm_f_0_0_0()
31195                   std::int32_t k, std::int32_t lhs_offset,  in gemm_f_0_0_0()
31196                   std::int32_t rhs_offset, float result_scale, float* result,  in gemm_f_0_0_0()
31197                   std::int32_t result_stride) {  in gemm_f_0_0_0()
31198   const std::int32_t row_chunks = m / 3;  in gemm_f_0_0_0()
31199   const std::int32_t col_chunks = n / 3;  in gemm_f_0_0_0()
31200   const std::int32_t padded_k = ((k + 7) / 8) * 8;  in gemm_f_0_0_0()
31201   const std::int32_t chunk_size = k * 3;  in gemm_f_0_0_0()
31202   const std::int32_t zipped_chunk_size = (padded_k + 16) * 3;  in gemm_f_0_0_0()
31203   const std::int32_t zipped_rhs_size = (padded_k + 16) * n;  in gemm_f_0_0_0()
31207   std::int32_t* zipped_lhs_3_offsets =  in gemm_f_0_0_0()
31208       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 3);  in gemm_f_0_0_0()
31211   const std::int32_t result_chunk_stride = result_stride * 3;  in gemm_f_0_0_0()
31213   const std::int32_t const_offset = lhs_offset * rhs_offset * k;  in gemm_f_0_0_0()
31216   const std::int32_t mul_result_chunk_stride_bytes = result_stride * 4;  in gemm_f_0_0_0()
31241                   const std::uint8_t* rhs, std::int32_t m, std::int32_t n,  in gemm_f_0_0_1()
31242                   std::int32_t k, std::int32_t lhs_offset,  in gemm_f_0_0_1()
31243                   std::int32_t rhs_offset, float result_scale, float* result,  in gemm_f_0_0_1()
31244                   std::int32_t result_stride) {  in gemm_f_0_0_1()
31245   const std::int32_t row_chunks = m / 3;  in gemm_f_0_0_1()
31246   const std::int32_t col_chunks = n / 3;  in gemm_f_0_0_1()
31247   const std::int32_t padded_k = ((k + 7) / 8) * 8;  in gemm_f_0_0_1()
31248   const std::int32_t chunk_size = k * 3;  in gemm_f_0_0_1()
31249   const std::int32_t zipped_chunk_size = (padded_k + 16) * 3;  in gemm_f_0_0_1()
31250   const std::int32_t zipped_rhs_size = (padded_k + 16) * n;  in gemm_f_0_0_1()
31254   std::int32_t* zipped_lhs_3_offsets =  in gemm_f_0_0_1()
31255       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 3);  in gemm_f_0_0_1()
31258   const std::int32_t result_chunk_stride = result_stride * 3;  in gemm_f_0_0_1()
31260   const std::int32_t const_offset = lhs_offset * rhs_offset * k;  in gemm_f_0_0_1()
31263   const std::int32_t mul_result_chunk_stride_bytes = result_stride * 4;  in gemm_f_0_0_1()
31288                   const std::uint8_t* rhs, std::int32_t m, std::int32_t n,  in gemm_f_0_0_2()
31289                   std::int32_t k, std::int32_t lhs_offset,  in gemm_f_0_0_2()
31290                   std::int32_t rhs_offset, float result_scale, float* result,  in gemm_f_0_0_2()
31291                   std::int32_t result_stride) {  in gemm_f_0_0_2()
31292   const std::int32_t row_chunks = m / 3;  in gemm_f_0_0_2()
31293   const std::int32_t col_chunks = n / 3;  in gemm_f_0_0_2()
31294   const std::int32_t padded_k = ((k + 7) / 8) * 8;  in gemm_f_0_0_2()
31295   const std::int32_t chunk_size = k * 3;  in gemm_f_0_0_2()
31296   const std::int32_t zipped_chunk_size = (padded_k + 16) * 3;  in gemm_f_0_0_2()
31297   const std::int32_t zipped_rhs_size = (padded_k + 16) * n;  in gemm_f_0_0_2()
31301   std::int32_t* zipped_lhs_3_offsets =  in gemm_f_0_0_2()
31302       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 3);  in gemm_f_0_0_2()
31305   const std::int32_t result_chunk_stride = result_stride * 3;  in gemm_f_0_0_2()
31307   const std::int32_t const_offset = lhs_offset * rhs_offset * k;  in gemm_f_0_0_2()
31310   const std::int32_t mul_result_chunk_stride_bytes = result_stride * 4;  in gemm_f_0_0_2()
31335                   const std::uint8_t* rhs, std::int32_t m, std::int32_t n,  in gemm_f_0_0_3()
31336                   std::int32_t k, std::int32_t lhs_offset,  in gemm_f_0_0_3()
31337                   std::int32_t rhs_offset, float result_scale, float* result,  in gemm_f_0_0_3()
31338                   std::int32_t result_stride) {  in gemm_f_0_0_3()
31339   const std::int32_t row_chunks = m / 3;  in gemm_f_0_0_3()
31340   const std::int32_t col_chunks = n / 3;  in gemm_f_0_0_3()
31341   const std::int32_t padded_k = ((k + 7) / 8) * 8;  in gemm_f_0_0_3()
31342   const std::int32_t chunk_size = k * 3;  in gemm_f_0_0_3()
31343   const std::int32_t zipped_chunk_size = (padded_k + 16) * 3;  in gemm_f_0_0_3()
31344   const std::int32_t zipped_rhs_size = (padded_k + 16) * n;  in gemm_f_0_0_3()
31348   std::int32_t* zipped_lhs_3_offsets =  in gemm_f_0_0_3()
31349       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 3);  in gemm_f_0_0_3()
31352   const std::int32_t result_chunk_stride = result_stride * 3;  in gemm_f_0_0_3()
31354   const std::int32_t const_offset = lhs_offset * rhs_offset * k;  in gemm_f_0_0_3()
31357   const std::int32_t mul_result_chunk_stride_bytes = result_stride * 4;  in gemm_f_0_0_3()
31382                   const std::uint8_t* rhs, std::int32_t m, std::int32_t n,  in gemm_f_0_0_4()
31383                   std::int32_t k, std::int32_t lhs_offset,  in gemm_f_0_0_4()
31384                   std::int32_t rhs_offset, float result_scale, float* result,  in gemm_f_0_0_4()
31385                   std::int32_t result_stride) {  in gemm_f_0_0_4()
31386   const std::int32_t row_chunks = m / 3;  in gemm_f_0_0_4()
31387   const std::int32_t col_chunks = n / 3;  in gemm_f_0_0_4()
31388   const std::int32_t padded_k = ((k + 7) / 8) * 8;  in gemm_f_0_0_4()
31389   const std::int32_t chunk_size = k * 3;  in gemm_f_0_0_4()
31390   const std::int32_t zipped_chunk_size = (padded_k + 16) * 3;  in gemm_f_0_0_4()
31391   const std::int32_t zipped_rhs_size = (padded_k + 16) * n;  in gemm_f_0_0_4()
31395   std::int32_t* zipped_lhs_3_offsets =  in gemm_f_0_0_4()
31396       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 3);  in gemm_f_0_0_4()
31399   const std::int32_t result_chunk_stride = result_stride * 3;  in gemm_f_0_0_4()
31401   const std::int32_t const_offset = lhs_offset * rhs_offset * k;  in gemm_f_0_0_4()
31404   const std::int32_t mul_result_chunk_stride_bytes = result_stride * 4;  in gemm_f_0_0_4()
31429                   const std::uint8_t* rhs, std::int32_t m, std::int32_t n,  in gemm_f_0_0_5()
31430                   std::int32_t k, std::int32_t lhs_offset,  in gemm_f_0_0_5()
31431                   std::int32_t rhs_offset, float result_scale, float* result,  in gemm_f_0_0_5()
31432                   std::int32_t result_stride) {  in gemm_f_0_0_5()
31433   const std::int32_t row_chunks = m / 3;  in gemm_f_0_0_5()
31434   const std::int32_t col_chunks = n / 3;  in gemm_f_0_0_5()
31435   const std::int32_t padded_k = ((k + 7) / 8) * 8;  in gemm_f_0_0_5()
31436   const std::int32_t chunk_size = k * 3;  in gemm_f_0_0_5()
31437   const std::int32_t zipped_chunk_size = (padded_k + 16) * 3;  in gemm_f_0_0_5()
31438   const std::int32_t zipped_rhs_size = (padded_k + 16) * n;  in gemm_f_0_0_5()
31442   std::int32_t* zipped_lhs_3_offsets =  in gemm_f_0_0_5()
31443       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 3);  in gemm_f_0_0_5()
31446   const std::int32_t result_chunk_stride = result_stride * 3;  in gemm_f_0_0_5()
31448   const std::int32_t const_offset = lhs_offset * rhs_offset * k;  in gemm_f_0_0_5()
31451   const std::int32_t mul_result_chunk_stride_bytes = result_stride * 4;  in gemm_f_0_0_5()
31476                   const std::uint8_t* rhs, std::int32_t m, std::int32_t n,  in gemm_f_0_0_6()
31477                   std::int32_t k, std::int32_t lhs_offset,  in gemm_f_0_0_6()
31478                   std::int32_t rhs_offset, float result_scale, float* result,  in gemm_f_0_0_6()
31479                   std::int32_t result_stride) {  in gemm_f_0_0_6()
31480   const std::int32_t row_chunks = m / 3;  in gemm_f_0_0_6()
31481   const std::int32_t col_chunks = n / 3;  in gemm_f_0_0_6()
31482   const std::int32_t padded_k = ((k + 7) / 8) * 8;  in gemm_f_0_0_6()
31483   const std::int32_t chunk_size = k * 3;  in gemm_f_0_0_6()
31484   const std::int32_t zipped_chunk_size = (padded_k + 16) * 3;  in gemm_f_0_0_6()
31485   const std::int32_t zipped_rhs_size = (padded_k + 16) * n;  in gemm_f_0_0_6()
31489   std::int32_t* zipped_lhs_3_offsets =  in gemm_f_0_0_6()
31490       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 3);  in gemm_f_0_0_6()
31493   const std::int32_t result_chunk_stride = result_stride * 3;  in gemm_f_0_0_6()
31495   const std::int32_t const_offset = lhs_offset * rhs_offset * k;  in gemm_f_0_0_6()
31498   const std::int32_t mul_result_chunk_stride_bytes = result_stride * 4;  in gemm_f_0_0_6()
31523                   const std::uint8_t* rhs, std::int32_t m, std::int32_t n,  in gemm_f_0_0_7()
31524                   std::int32_t k, std::int32_t lhs_offset,  in gemm_f_0_0_7()
31525                   std::int32_t rhs_offset, float result_scale, float* result,  in gemm_f_0_0_7()
31526                   std::int32_t result_stride) {  in gemm_f_0_0_7()
31527   const std::int32_t row_chunks = m / 3;  in gemm_f_0_0_7()
31528   const std::int32_t col_chunks = n / 3;  in gemm_f_0_0_7()
31529   const std::int32_t padded_k = ((k + 7) / 8) * 8;  in gemm_f_0_0_7()
31530   const std::int32_t chunk_size = k * 3;  in gemm_f_0_0_7()
31531   const std::int32_t zipped_chunk_size = (padded_k + 16) * 3;  in gemm_f_0_0_7()
31532   const std::int32_t zipped_rhs_size = (padded_k + 16) * n;  in gemm_f_0_0_7()
31536   std::int32_t* zipped_lhs_3_offsets =  in gemm_f_0_0_7()
31537       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 3);  in gemm_f_0_0_7()
31540   const std::int32_t result_chunk_stride = result_stride * 3;  in gemm_f_0_0_7()
31542   const std::int32_t const_offset = lhs_offset * rhs_offset * k;  in gemm_f_0_0_7()
31545   const std::int32_t mul_result_chunk_stride_bytes = result_stride * 4;  in gemm_f_0_0_7()
31570                   const std::uint8_t* rhs, std::int32_t m, std::int32_t n,  in gemm_f_0_1_0()
31571                   std::int32_t k, std::int32_t lhs_offset,  in gemm_f_0_1_0()
31572                   std::int32_t rhs_offset, float result_scale, float* result,  in gemm_f_0_1_0()
31573                   std::int32_t result_stride) {  in gemm_f_0_1_0()
31574   const std::int32_t row_chunks = m / 3;  in gemm_f_0_1_0()
31575   const std::int32_t col_chunks = n / 3;  in gemm_f_0_1_0()
31576   const std::int32_t padded_k = ((k + 7) / 8) * 8;  in gemm_f_0_1_0()
31577   const std::int32_t chunk_size = k * 3;  in gemm_f_0_1_0()
31578   const std::int32_t zipped_chunk_size = (padded_k + 16) * 3;  in gemm_f_0_1_0()
31579   const std::int32_t zipped_rhs_size = (padded_k + 16) * n;  in gemm_f_0_1_0()
31583   std::int32_t* zipped_lhs_3_offsets =  in gemm_f_0_1_0()
31584       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 3);  in gemm_f_0_1_0()
31587   const std::int32_t result_chunk_stride = result_stride * 3;  in gemm_f_0_1_0()
31589   const std::int32_t const_offset = lhs_offset * rhs_offset * k;  in gemm_f_0_1_0()
31592   const std::int32_t mul_result_chunk_stride_bytes = result_stride * 4;  in gemm_f_0_1_0()
31621                   const std::uint8_t* rhs, std::int32_t m, std::int32_t n,  in gemm_f_0_1_1()
31622                   std::int32_t k, std::int32_t lhs_offset,  in gemm_f_0_1_1()
31623                   std::int32_t rhs_offset, float result_scale, float* result,  in gemm_f_0_1_1()
31624                   std::int32_t result_stride) {  in gemm_f_0_1_1()
31625   const std::int32_t row_chunks = m / 3;  in gemm_f_0_1_1()
31626   const std::int32_t col_chunks = n / 3;  in gemm_f_0_1_1()
31627   const std::int32_t padded_k = ((k + 7) / 8) * 8;  in gemm_f_0_1_1()
31628   const std::int32_t chunk_size = k * 3;  in gemm_f_0_1_1()
31629   const std::int32_t zipped_chunk_size = (padded_k + 16) * 3;  in gemm_f_0_1_1()
31630   const std::int32_t zipped_rhs_size = (padded_k + 16) * n;  in gemm_f_0_1_1()
31634   std::int32_t* zipped_lhs_3_offsets =  in gemm_f_0_1_1()
31635       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 3);  in gemm_f_0_1_1()
31638   const std::int32_t result_chunk_stride = result_stride * 3;  in gemm_f_0_1_1()
31640   const std::int32_t const_offset = lhs_offset * rhs_offset * k;  in gemm_f_0_1_1()
31643   const std::int32_t mul_result_chunk_stride_bytes = result_stride * 4;  in gemm_f_0_1_1()
31672                   const std::uint8_t* rhs, std::int32_t m, std::int32_t n,  in gemm_f_0_1_2()
31673                   std::int32_t k, std::int32_t lhs_offset,  in gemm_f_0_1_2()
31674                   std::int32_t rhs_offset, float result_scale, float* result,  in gemm_f_0_1_2()
31675                   std::int32_t result_stride) {  in gemm_f_0_1_2()
31676   const std::int32_t row_chunks = m / 3;  in gemm_f_0_1_2()
31677   const std::int32_t col_chunks = n / 3;  in gemm_f_0_1_2()
31678   const std::int32_t padded_k = ((k + 7) / 8) * 8;  in gemm_f_0_1_2()
31679   const std::int32_t chunk_size = k * 3;  in gemm_f_0_1_2()
31680   const std::int32_t zipped_chunk_size = (padded_k + 16) * 3;  in gemm_f_0_1_2()
31681   const std::int32_t zipped_rhs_size = (padded_k + 16) * n;  in gemm_f_0_1_2()
31685   std::int32_t* zipped_lhs_3_offsets =  in gemm_f_0_1_2()
31686       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 3);  in gemm_f_0_1_2()
31689   const std::int32_t result_chunk_stride = result_stride * 3;  in gemm_f_0_1_2()
31691   const std::int32_t const_offset = lhs_offset * rhs_offset * k;  in gemm_f_0_1_2()
31694   const std::int32_t mul_result_chunk_stride_bytes = result_stride * 4;  in gemm_f_0_1_2()
31723                   const std::uint8_t* rhs, std::int32_t m, std::int32_t n,  in gemm_f_0_1_3()
31724                   std::int32_t k, std::int32_t lhs_offset,  in gemm_f_0_1_3()
31725                   std::int32_t rhs_offset, float result_scale, float* result,  in gemm_f_0_1_3()
31726                   std::int32_t result_stride) {  in gemm_f_0_1_3()
31727   const std::int32_t row_chunks = m / 3;  in gemm_f_0_1_3()
31728   const std::int32_t col_chunks = n / 3;  in gemm_f_0_1_3()
31729   const std::int32_t padded_k = ((k + 7) / 8) * 8;  in gemm_f_0_1_3()
31730   const std::int32_t chunk_size = k * 3;  in gemm_f_0_1_3()
31731   const std::int32_t zipped_chunk_size = (padded_k + 16) * 3;  in gemm_f_0_1_3()
31732   const std::int32_t zipped_rhs_size = (padded_k + 16) * n;  in gemm_f_0_1_3()
31736   std::int32_t* zipped_lhs_3_offsets =  in gemm_f_0_1_3()
31737       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 3);  in gemm_f_0_1_3()
31740   const std::int32_t result_chunk_stride = result_stride * 3;  in gemm_f_0_1_3()
31742   const std::int32_t const_offset = lhs_offset * rhs_offset * k;  in gemm_f_0_1_3()
31745   const std::int32_t mul_result_chunk_stride_bytes = result_stride * 4;  in gemm_f_0_1_3()
31774                   const std::uint8_t* rhs, std::int32_t m, std::int32_t n,  in gemm_f_0_1_4()
31775                   std::int32_t k, std::int32_t lhs_offset,  in gemm_f_0_1_4()
31776                   std::int32_t rhs_offset, float result_scale, float* result,  in gemm_f_0_1_4()
31777                   std::int32_t result_stride) {  in gemm_f_0_1_4()
31778   const std::int32_t row_chunks = m / 3;  in gemm_f_0_1_4()
31779   const std::int32_t col_chunks = n / 3;  in gemm_f_0_1_4()
31780   const std::int32_t padded_k = ((k + 7) / 8) * 8;  in gemm_f_0_1_4()
31781   const std::int32_t chunk_size = k * 3;  in gemm_f_0_1_4()
31782   const std::int32_t zipped_chunk_size = (padded_k + 16) * 3;  in gemm_f_0_1_4()
31783   const std::int32_t zipped_rhs_size = (padded_k + 16) * n;  in gemm_f_0_1_4()
31787   std::int32_t* zipped_lhs_3_offsets =  in gemm_f_0_1_4()
31788       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 3);  in gemm_f_0_1_4()
31791   const std::int32_t result_chunk_stride = result_stride * 3;  in gemm_f_0_1_4()
31793   const std::int32_t const_offset = lhs_offset * rhs_offset * k;  in gemm_f_0_1_4()
31796   const std::int32_t mul_result_chunk_stride_bytes = result_stride * 4;  in gemm_f_0_1_4()
31825                   const std::uint8_t* rhs, std::int32_t m, std::int32_t n,  in gemm_f_0_1_5()
31826                   std::int32_t k, std::int32_t lhs_offset,  in gemm_f_0_1_5()
31827                   std::int32_t rhs_offset, float result_scale, float* result,  in gemm_f_0_1_5()
31828                   std::int32_t result_stride) {  in gemm_f_0_1_5()
31829   const std::int32_t row_chunks = m / 3;  in gemm_f_0_1_5()
31830   const std::int32_t col_chunks = n / 3;  in gemm_f_0_1_5()
31831   const std::int32_t padded_k = ((k + 7) / 8) * 8;  in gemm_f_0_1_5()
31832   const std::int32_t chunk_size = k * 3;  in gemm_f_0_1_5()
31833   const std::int32_t zipped_chunk_size = (padded_k + 16) * 3;  in gemm_f_0_1_5()
31834   const std::int32_t zipped_rhs_size = (padded_k + 16) * n;  in gemm_f_0_1_5()
31838   std::int32_t* zipped_lhs_3_offsets =  in gemm_f_0_1_5()
31839       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 3);  in gemm_f_0_1_5()
31842   const std::int32_t result_chunk_stride = result_stride * 3;  in gemm_f_0_1_5()
31844   const std::int32_t const_offset = lhs_offset * rhs_offset * k;  in gemm_f_0_1_5()
31847   const std::int32_t mul_result_chunk_stride_bytes = result_stride * 4;  in gemm_f_0_1_5()
31876                   const std::uint8_t* rhs, std::int32_t m, std::int32_t n,  in gemm_f_0_1_6()
31877                   std::int32_t k, std::int32_t lhs_offset,  in gemm_f_0_1_6()
31878                   std::int32_t rhs_offset, float result_scale, float* result,  in gemm_f_0_1_6()
31879                   std::int32_t result_stride) {  in gemm_f_0_1_6()
31880   const std::int32_t row_chunks = m / 3;  in gemm_f_0_1_6()
31881   const std::int32_t col_chunks = n / 3;  in gemm_f_0_1_6()
31882   const std::int32_t padded_k = ((k + 7) / 8) * 8;  in gemm_f_0_1_6()
31883   const std::int32_t chunk_size = k * 3;  in gemm_f_0_1_6()
31884   const std::int32_t zipped_chunk_size = (padded_k + 16) * 3;  in gemm_f_0_1_6()
31885   const std::int32_t zipped_rhs_size = (padded_k + 16) * n;  in gemm_f_0_1_6()
31889   std::int32_t* zipped_lhs_3_offsets =  in gemm_f_0_1_6()
31890       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 3);  in gemm_f_0_1_6()
31893   const std::int32_t result_chunk_stride = result_stride * 3;  in gemm_f_0_1_6()
31895   const std::int32_t const_offset = lhs_offset * rhs_offset * k;  in gemm_f_0_1_6()
31898   const std::int32_t mul_result_chunk_stride_bytes = result_stride * 4;  in gemm_f_0_1_6()
31927                   const std::uint8_t* rhs, std::int32_t m, std::int32_t n,  in gemm_f_0_1_7()
31928                   std::int32_t k, std::int32_t lhs_offset,  in gemm_f_0_1_7()
31929                   std::int32_t rhs_offset, float result_scale, float* result,  in gemm_f_0_1_7()
31930                   std::int32_t result_stride) {  in gemm_f_0_1_7()
31931   const std::int32_t row_chunks = m / 3;  in gemm_f_0_1_7()
31932   const std::int32_t col_chunks = n / 3;  in gemm_f_0_1_7()
31933   const std::int32_t padded_k = ((k + 7) / 8) * 8;  in gemm_f_0_1_7()
31934   const std::int32_t chunk_size = k * 3;  in gemm_f_0_1_7()
31935   const std::int32_t zipped_chunk_size = (padded_k + 16) * 3;  in gemm_f_0_1_7()
31936   const std::int32_t zipped_rhs_size = (padded_k + 16) * n;  in gemm_f_0_1_7()
31940   std::int32_t* zipped_lhs_3_offsets =  in gemm_f_0_1_7()
31941       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 3);  in gemm_f_0_1_7()
31944   const std::int32_t result_chunk_stride = result_stride * 3;  in gemm_f_0_1_7()
31946   const std::int32_t const_offset = lhs_offset * rhs_offset * k;  in gemm_f_0_1_7()
31949   const std::int32_t mul_result_chunk_stride_bytes = result_stride * 4;  in gemm_f_0_1_7()
31978                   const std::uint8_t* rhs, std::int32_t m, std::int32_t n,  in gemm_f_0_2_0()
31979                   std::int32_t k, std::int32_t lhs_offset,  in gemm_f_0_2_0()
31980                   std::int32_t rhs_offset, float result_scale, float* result,  in gemm_f_0_2_0()
31981                   std::int32_t result_stride) {  in gemm_f_0_2_0()
31982   const std::int32_t row_chunks = m / 3;  in gemm_f_0_2_0()
31983   const std::int32_t col_chunks = n / 3;  in gemm_f_0_2_0()
31984   const std::int32_t padded_k = ((k + 7) / 8) * 8;  in gemm_f_0_2_0()
31985   const std::int32_t chunk_size = k * 3;  in gemm_f_0_2_0()
31986   const std::int32_t zipped_chunk_size = (padded_k + 16) * 3;  in gemm_f_0_2_0()
31987   const std::int32_t zipped_rhs_size = (padded_k + 16) * n;  in gemm_f_0_2_0()
31991   std::int32_t* zipped_lhs_3_offsets =  in gemm_f_0_2_0()
31992       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 3);  in gemm_f_0_2_0()
31995   const std::int32_t result_chunk_stride = result_stride * 3;  in gemm_f_0_2_0()
31997   const std::int32_t const_offset = lhs_offset * rhs_offset * k;  in gemm_f_0_2_0()
32000   const std::int32_t mul_result_chunk_stride_bytes = result_stride * 4;  in gemm_f_0_2_0()
32029                   const std::uint8_t* rhs, std::int32_t m, std::int32_t n,  in gemm_f_0_2_1()
32030                   std::int32_t k, std::int32_t lhs_offset,  in gemm_f_0_2_1()
32031                   std::int32_t rhs_offset, float result_scale, float* result,  in gemm_f_0_2_1()
32032                   std::int32_t result_stride) {  in gemm_f_0_2_1()
32033   const std::int32_t row_chunks = m / 3;  in gemm_f_0_2_1()
32034   const std::int32_t col_chunks = n / 3;  in gemm_f_0_2_1()
32035   const std::int32_t padded_k = ((k + 7) / 8) * 8;  in gemm_f_0_2_1()
32036   const std::int32_t chunk_size = k * 3;  in gemm_f_0_2_1()
32037   const std::int32_t zipped_chunk_size = (padded_k + 16) * 3;  in gemm_f_0_2_1()
32038   const std::int32_t zipped_rhs_size = (padded_k + 16) * n;  in gemm_f_0_2_1()
32042   std::int32_t* zipped_lhs_3_offsets =  in gemm_f_0_2_1()
32043       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 3);  in gemm_f_0_2_1()
32046   const std::int32_t result_chunk_stride = result_stride * 3;  in gemm_f_0_2_1()
32048   const std::int32_t const_offset = lhs_offset * rhs_offset * k;  in gemm_f_0_2_1()
32051   const std::int32_t mul_result_chunk_stride_bytes = result_stride * 4;  in gemm_f_0_2_1()
32080                   const std::uint8_t* rhs, std::int32_t m, std::int32_t n,  in gemm_f_0_2_2()
32081                   std::int32_t k, std::int32_t lhs_offset,  in gemm_f_0_2_2()
32082                   std::int32_t rhs_offset, float result_scale, float* result,  in gemm_f_0_2_2()
32083                   std::int32_t result_stride) {  in gemm_f_0_2_2()
32084   const std::int32_t row_chunks = m / 3;  in gemm_f_0_2_2()
32085   const std::int32_t col_chunks = n / 3;  in gemm_f_0_2_2()
32086   const std::int32_t padded_k = ((k + 7) / 8) * 8;  in gemm_f_0_2_2()
32087   const std::int32_t chunk_size = k * 3;  in gemm_f_0_2_2()
32088   const std::int32_t zipped_chunk_size = (padded_k + 16) * 3;  in gemm_f_0_2_2()
32089   const std::int32_t zipped_rhs_size = (padded_k + 16) * n;  in gemm_f_0_2_2()
32093   std::int32_t* zipped_lhs_3_offsets =  in gemm_f_0_2_2()
32094       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 3);  in gemm_f_0_2_2()
32097   const std::int32_t result_chunk_stride = result_stride * 3;  in gemm_f_0_2_2()
32099   const std::int32_t const_offset = lhs_offset * rhs_offset * k;  in gemm_f_0_2_2()
32102   const std::int32_t mul_result_chunk_stride_bytes = result_stride * 4;  in gemm_f_0_2_2()
32131                   const std::uint8_t* rhs, std::int32_t m, std::int32_t n,  in gemm_f_0_2_3()
32132                   std::int32_t k, std::int32_t lhs_offset,  in gemm_f_0_2_3()
32133                   std::int32_t rhs_offset, float result_scale, float* result,  in gemm_f_0_2_3()
32134                   std::int32_t result_stride) {  in gemm_f_0_2_3()
32135   const std::int32_t row_chunks = m / 3;  in gemm_f_0_2_3()
32136   const std::int32_t col_chunks = n / 3;  in gemm_f_0_2_3()
32137   const std::int32_t padded_k = ((k + 7) / 8) * 8;  in gemm_f_0_2_3()
32138   const std::int32_t chunk_size = k * 3;  in gemm_f_0_2_3()
32139   const std::int32_t zipped_chunk_size = (padded_k + 16) * 3;  in gemm_f_0_2_3()
32140   const std::int32_t zipped_rhs_size = (padded_k + 16) * n;  in gemm_f_0_2_3()
32144   std::int32_t* zipped_lhs_3_offsets =  in gemm_f_0_2_3()
32145       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 3);  in gemm_f_0_2_3()
32148   const std::int32_t result_chunk_stride = result_stride * 3;  in gemm_f_0_2_3()
32150   const std::int32_t const_offset = lhs_offset * rhs_offset * k;  in gemm_f_0_2_3()
32153   const std::int32_t mul_result_chunk_stride_bytes = result_stride * 4;  in gemm_f_0_2_3()
32182                   const std::uint8_t* rhs, std::int32_t m, std::int32_t n,  in gemm_f_0_2_4()
32183                   std::int32_t k, std::int32_t lhs_offset,  in gemm_f_0_2_4()
32184                   std::int32_t rhs_offset, float result_scale, float* result,  in gemm_f_0_2_4()
32185                   std::int32_t result_stride) {  in gemm_f_0_2_4()
32186   const std::int32_t row_chunks = m / 3;  in gemm_f_0_2_4()
32187   const std::int32_t col_chunks = n / 3;  in gemm_f_0_2_4()
32188   const std::int32_t padded_k = ((k + 7) / 8) * 8;  in gemm_f_0_2_4()
32189   const std::int32_t chunk_size = k * 3;  in gemm_f_0_2_4()
32190   const std::int32_t zipped_chunk_size = (padded_k + 16) * 3;  in gemm_f_0_2_4()
32191   const std::int32_t zipped_rhs_size = (padded_k + 16) * n;  in gemm_f_0_2_4()
32195   std::int32_t* zipped_lhs_3_offsets =  in gemm_f_0_2_4()
32196       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 3);  in gemm_f_0_2_4()
32199   const std::int32_t result_chunk_stride = result_stride * 3;  in gemm_f_0_2_4()
32201   const std::int32_t const_offset = lhs_offset * rhs_offset * k;  in gemm_f_0_2_4()
32204   const std::int32_t mul_result_chunk_stride_bytes = result_stride * 4;  in gemm_f_0_2_4()
32233                   const std::uint8_t* rhs, std::int32_t m, std::int32_t n,  in gemm_f_0_2_5()
32234                   std::int32_t k, std::int32_t lhs_offset,  in gemm_f_0_2_5()
32235                   std::int32_t rhs_offset, float result_scale, float* result,  in gemm_f_0_2_5()
32236                   std::int32_t result_stride) {  in gemm_f_0_2_5()
32237   const std::int32_t row_chunks = m / 3;  in gemm_f_0_2_5()
32238   const std::int32_t col_chunks = n / 3;  in gemm_f_0_2_5()
32239   const std::int32_t padded_k = ((k + 7) / 8) * 8;  in gemm_f_0_2_5()
32240   const std::int32_t chunk_size = k * 3;  in gemm_f_0_2_5()
32241   const std::int32_t zipped_chunk_size = (padded_k + 16) * 3;  in gemm_f_0_2_5()
32242   const std::int32_t zipped_rhs_size = (padded_k + 16) * n;  in gemm_f_0_2_5()
32246   std::int32_t* zipped_lhs_3_offsets =  in gemm_f_0_2_5()
32247       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 3);  in gemm_f_0_2_5()
32250   const std::int32_t result_chunk_stride = result_stride * 3;  in gemm_f_0_2_5()
32252   const std::int32_t const_offset = lhs_offset * rhs_offset * k;  in gemm_f_0_2_5()
32255   const std::int32_t mul_result_chunk_stride_bytes = result_stride * 4;  in gemm_f_0_2_5()
32284                   const std::uint8_t* rhs, std::int32_t m, std::int32_t n,  in gemm_f_0_2_6()
32285                   std::int32_t k, std::int32_t lhs_offset,  in gemm_f_0_2_6()
32286                   std::int32_t rhs_offset, float result_scale, float* result,  in gemm_f_0_2_6()
32287                   std::int32_t result_stride) {  in gemm_f_0_2_6()
32288   const std::int32_t row_chunks = m / 3;  in gemm_f_0_2_6()
32289   const std::int32_t col_chunks = n / 3;  in gemm_f_0_2_6()
32290   const std::int32_t padded_k = ((k + 7) / 8) * 8;  in gemm_f_0_2_6()
32291   const std::int32_t chunk_size = k * 3;  in gemm_f_0_2_6()
32292   const std::int32_t zipped_chunk_size = (padded_k + 16) * 3;  in gemm_f_0_2_6()
32293   const std::int32_t zipped_rhs_size = (padded_k + 16) * n;  in gemm_f_0_2_6()
32297   std::int32_t* zipped_lhs_3_offsets =  in gemm_f_0_2_6()
32298       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 3);  in gemm_f_0_2_6()
32301   const std::int32_t result_chunk_stride = result_stride * 3;  in gemm_f_0_2_6()
32303   const std::int32_t const_offset = lhs_offset * rhs_offset * k;  in gemm_f_0_2_6()
32306   const std::int32_t mul_result_chunk_stride_bytes = result_stride * 4;  in gemm_f_0_2_6()
32335                   const std::uint8_t* rhs, std::int32_t m, std::int32_t n,  in gemm_f_0_2_7()
32336                   std::int32_t k, std::int32_t lhs_offset,  in gemm_f_0_2_7()
32337                   std::int32_t rhs_offset, float result_scale, float* result,  in gemm_f_0_2_7()
32338                   std::int32_t result_stride) {  in gemm_f_0_2_7()
32339   const std::int32_t row_chunks = m / 3;  in gemm_f_0_2_7()
32340   const std::int32_t col_chunks = n / 3;  in gemm_f_0_2_7()
32341   const std::int32_t padded_k = ((k + 7) / 8) * 8;  in gemm_f_0_2_7()
32342   const std::int32_t chunk_size = k * 3;  in gemm_f_0_2_7()
32343   const std::int32_t zipped_chunk_size = (padded_k + 16) * 3;  in gemm_f_0_2_7()
32344   const std::int32_t zipped_rhs_size = (padded_k + 16) * n;  in gemm_f_0_2_7()
32348   std::int32_t* zipped_lhs_3_offsets =  in gemm_f_0_2_7()
32349       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 3);  in gemm_f_0_2_7()
32352   const std::int32_t result_chunk_stride = result_stride * 3;  in gemm_f_0_2_7()
32354   const std::int32_t const_offset = lhs_offset * rhs_offset * k;  in gemm_f_0_2_7()
32357   const std::int32_t mul_result_chunk_stride_bytes = result_stride * 4;  in gemm_f_0_2_7()
32386                   const std::uint8_t* rhs, std::int32_t m, std::int32_t n,  in gemm_f_1_0_0()
32387                   std::int32_t k, std::int32_t lhs_offset,  in gemm_f_1_0_0()
32388                   std::int32_t rhs_offset, float result_scale, float* result,  in gemm_f_1_0_0()
32389                   std::int32_t result_stride) {  in gemm_f_1_0_0()
32390   const std::int32_t row_chunks = m / 3;  in gemm_f_1_0_0()
32391   const std::int32_t col_chunks = n / 3;  in gemm_f_1_0_0()
32392   const std::int32_t padded_k = ((k + 7) / 8) * 8;  in gemm_f_1_0_0()
32393   const std::int32_t chunk_size = k * 3;  in gemm_f_1_0_0()
32394   const std::int32_t zipped_chunk_size = (padded_k + 16) * 3;  in gemm_f_1_0_0()
32395   const std::int32_t zipped_rhs_size = (padded_k + 16) * n;  in gemm_f_1_0_0()
32399   std::int32_t* zipped_lhs_3_offsets =  in gemm_f_1_0_0()
32400       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 3);  in gemm_f_1_0_0()
32401   std::int32_t* zipped_lhs_1_offsets =  in gemm_f_1_0_0()
32402       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 1);  in gemm_f_1_0_0()
32405   const std::int32_t result_chunk_stride = result_stride * 3;  in gemm_f_1_0_0()
32407   const std::int32_t const_offset = lhs_offset * rhs_offset * k;  in gemm_f_1_0_0()
32410   const std::int32_t mul_result_chunk_stride_bytes = result_stride * 4;  in gemm_f_1_0_0()
32446                   const std::uint8_t* rhs, std::int32_t m, std::int32_t n,  in gemm_f_1_0_1()
32447                   std::int32_t k, std::int32_t lhs_offset,  in gemm_f_1_0_1()
32448                   std::int32_t rhs_offset, float result_scale, float* result,  in gemm_f_1_0_1()
32449                   std::int32_t result_stride) {  in gemm_f_1_0_1()
32450   const std::int32_t row_chunks = m / 3;  in gemm_f_1_0_1()
32451   const std::int32_t col_chunks = n / 3;  in gemm_f_1_0_1()
32452   const std::int32_t padded_k = ((k + 7) / 8) * 8;  in gemm_f_1_0_1()
32453   const std::int32_t chunk_size = k * 3;  in gemm_f_1_0_1()
32454   const std::int32_t zipped_chunk_size = (padded_k + 16) * 3;  in gemm_f_1_0_1()
32455   const std::int32_t zipped_rhs_size = (padded_k + 16) * n;  in gemm_f_1_0_1()
32459   std::int32_t* zipped_lhs_3_offsets =  in gemm_f_1_0_1()
32460       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 3);  in gemm_f_1_0_1()
32461   std::int32_t* zipped_lhs_1_offsets =  in gemm_f_1_0_1()
32462       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 1);  in gemm_f_1_0_1()
32465   const std::int32_t result_chunk_stride = result_stride * 3;  in gemm_f_1_0_1()
32467   const std::int32_t const_offset = lhs_offset * rhs_offset * k;  in gemm_f_1_0_1()
32470   const std::int32_t mul_result_chunk_stride_bytes = result_stride * 4;  in gemm_f_1_0_1()
32506                   const std::uint8_t* rhs, std::int32_t m, std::int32_t n,  in gemm_f_1_0_2()
32507                   std::int32_t k, std::int32_t lhs_offset,  in gemm_f_1_0_2()
32508                   std::int32_t rhs_offset, float result_scale, float* result,  in gemm_f_1_0_2()
32509                   std::int32_t result_stride) {  in gemm_f_1_0_2()
32510   const std::int32_t row_chunks = m / 3;  in gemm_f_1_0_2()
32511   const std::int32_t col_chunks = n / 3;  in gemm_f_1_0_2()
32512   const std::int32_t padded_k = ((k + 7) / 8) * 8;  in gemm_f_1_0_2()
32513   const std::int32_t chunk_size = k * 3;  in gemm_f_1_0_2()
32514   const std::int32_t zipped_chunk_size = (padded_k + 16) * 3;  in gemm_f_1_0_2()
32515   const std::int32_t zipped_rhs_size = (padded_k + 16) * n;  in gemm_f_1_0_2()
32519   std::int32_t* zipped_lhs_3_offsets =  in gemm_f_1_0_2()
32520       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 3);  in gemm_f_1_0_2()
32521   std::int32_t* zipped_lhs_1_offsets =  in gemm_f_1_0_2()
32522       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 1);  in gemm_f_1_0_2()
32525   const std::int32_t result_chunk_stride = result_stride * 3;  in gemm_f_1_0_2()
32527   const std::int32_t const_offset = lhs_offset * rhs_offset * k;  in gemm_f_1_0_2()
32530   const std::int32_t mul_result_chunk_stride_bytes = result_stride * 4;  in gemm_f_1_0_2()
32566                   const std::uint8_t* rhs, std::int32_t m, std::int32_t n,  in gemm_f_1_0_3()
32567                   std::int32_t k, std::int32_t lhs_offset,  in gemm_f_1_0_3()
32568                   std::int32_t rhs_offset, float result_scale, float* result,  in gemm_f_1_0_3()
32569                   std::int32_t result_stride) {  in gemm_f_1_0_3()
32570   const std::int32_t row_chunks = m / 3;  in gemm_f_1_0_3()
32571   const std::int32_t col_chunks = n / 3;  in gemm_f_1_0_3()
32572   const std::int32_t padded_k = ((k + 7) / 8) * 8;  in gemm_f_1_0_3()
32573   const std::int32_t chunk_size = k * 3;  in gemm_f_1_0_3()
32574   const std::int32_t zipped_chunk_size = (padded_k + 16) * 3;  in gemm_f_1_0_3()
32575   const std::int32_t zipped_rhs_size = (padded_k + 16) * n;  in gemm_f_1_0_3()
32579   std::int32_t* zipped_lhs_3_offsets =  in gemm_f_1_0_3()
32580       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 3);  in gemm_f_1_0_3()
32581   std::int32_t* zipped_lhs_1_offsets =  in gemm_f_1_0_3()
32582       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 1);  in gemm_f_1_0_3()
32585   const std::int32_t result_chunk_stride = result_stride * 3;  in gemm_f_1_0_3()
32587   const std::int32_t const_offset = lhs_offset * rhs_offset * k;  in gemm_f_1_0_3()
32590   const std::int32_t mul_result_chunk_stride_bytes = result_stride * 4;  in gemm_f_1_0_3()
32626                   const std::uint8_t* rhs, std::int32_t m, std::int32_t n,  in gemm_f_1_0_4()
32627                   std::int32_t k, std::int32_t lhs_offset,  in gemm_f_1_0_4()
32628                   std::int32_t rhs_offset, float result_scale, float* result,  in gemm_f_1_0_4()
32629                   std::int32_t result_stride) {  in gemm_f_1_0_4()
32630   const std::int32_t row_chunks = m / 3;  in gemm_f_1_0_4()
32631   const std::int32_t col_chunks = n / 3;  in gemm_f_1_0_4()
32632   const std::int32_t padded_k = ((k + 7) / 8) * 8;  in gemm_f_1_0_4()
32633   const std::int32_t chunk_size = k * 3;  in gemm_f_1_0_4()
32634   const std::int32_t zipped_chunk_size = (padded_k + 16) * 3;  in gemm_f_1_0_4()
32635   const std::int32_t zipped_rhs_size = (padded_k + 16) * n;  in gemm_f_1_0_4()
32639   std::int32_t* zipped_lhs_3_offsets =  in gemm_f_1_0_4()
32640       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 3);  in gemm_f_1_0_4()
32641   std::int32_t* zipped_lhs_1_offsets =  in gemm_f_1_0_4()
32642       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 1);  in gemm_f_1_0_4()
32645   const std::int32_t result_chunk_stride = result_stride * 3;  in gemm_f_1_0_4()
32647   const std::int32_t const_offset = lhs_offset * rhs_offset * k;  in gemm_f_1_0_4()
32650   const std::int32_t mul_result_chunk_stride_bytes = result_stride * 4;  in gemm_f_1_0_4()
32686                   const std::uint8_t* rhs, std::int32_t m, std::int32_t n,  in gemm_f_1_0_5()
32687                   std::int32_t k, std::int32_t lhs_offset,  in gemm_f_1_0_5()
32688                   std::int32_t rhs_offset, float result_scale, float* result,  in gemm_f_1_0_5()
32689                   std::int32_t result_stride) {  in gemm_f_1_0_5()
32690   const std::int32_t row_chunks = m / 3;  in gemm_f_1_0_5()
32691   const std::int32_t col_chunks = n / 3;  in gemm_f_1_0_5()
32692   const std::int32_t padded_k = ((k + 7) / 8) * 8;  in gemm_f_1_0_5()
32693   const std::int32_t chunk_size = k * 3;  in gemm_f_1_0_5()
32694   const std::int32_t zipped_chunk_size = (padded_k + 16) * 3;  in gemm_f_1_0_5()
32695   const std::int32_t zipped_rhs_size = (padded_k + 16) * n;  in gemm_f_1_0_5()
32699   std::int32_t* zipped_lhs_3_offsets =  in gemm_f_1_0_5()
32700       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 3);  in gemm_f_1_0_5()
32701   std::int32_t* zipped_lhs_1_offsets =  in gemm_f_1_0_5()
32702       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 1);  in gemm_f_1_0_5()
32705   const std::int32_t result_chunk_stride = result_stride * 3;  in gemm_f_1_0_5()
32707   const std::int32_t const_offset = lhs_offset * rhs_offset * k;  in gemm_f_1_0_5()
32710   const std::int32_t mul_result_chunk_stride_bytes = result_stride * 4;  in gemm_f_1_0_5()
32746                   const std::uint8_t* rhs, std::int32_t m, std::int32_t n,  in gemm_f_1_0_6()
32747                   std::int32_t k, std::int32_t lhs_offset,  in gemm_f_1_0_6()
32748                   std::int32_t rhs_offset, float result_scale, float* result,  in gemm_f_1_0_6()
32749                   std::int32_t result_stride) {  in gemm_f_1_0_6()
32750   const std::int32_t row_chunks = m / 3;  in gemm_f_1_0_6()
32751   const std::int32_t col_chunks = n / 3;  in gemm_f_1_0_6()
32752   const std::int32_t padded_k = ((k + 7) / 8) * 8;  in gemm_f_1_0_6()
32753   const std::int32_t chunk_size = k * 3;  in gemm_f_1_0_6()
32754   const std::int32_t zipped_chunk_size = (padded_k + 16) * 3;  in gemm_f_1_0_6()
32755   const std::int32_t zipped_rhs_size = (padded_k + 16) * n;  in gemm_f_1_0_6()
32759   std::int32_t* zipped_lhs_3_offsets =  in gemm_f_1_0_6()
32760       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 3);  in gemm_f_1_0_6()
32761   std::int32_t* zipped_lhs_1_offsets =  in gemm_f_1_0_6()
32762       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 1);  in gemm_f_1_0_6()
32765   const std::int32_t result_chunk_stride = result_stride * 3;  in gemm_f_1_0_6()
32767   const std::int32_t const_offset = lhs_offset * rhs_offset * k;  in gemm_f_1_0_6()
32770   const std::int32_t mul_result_chunk_stride_bytes = result_stride * 4;  in gemm_f_1_0_6()
32806                   const std::uint8_t* rhs, std::int32_t m, std::int32_t n,  in gemm_f_1_0_7()
32807                   std::int32_t k, std::int32_t lhs_offset,  in gemm_f_1_0_7()
32808                   std::int32_t rhs_offset, float result_scale, float* result,  in gemm_f_1_0_7()
32809                   std::int32_t result_stride) {  in gemm_f_1_0_7()
32810   const std::int32_t row_chunks = m / 3;  in gemm_f_1_0_7()
32811   const std::int32_t col_chunks = n / 3;  in gemm_f_1_0_7()
32812   const std::int32_t padded_k = ((k + 7) / 8) * 8;  in gemm_f_1_0_7()
32813   const std::int32_t chunk_size = k * 3;  in gemm_f_1_0_7()
32814   const std::int32_t zipped_chunk_size = (padded_k + 16) * 3;  in gemm_f_1_0_7()
32815   const std::int32_t zipped_rhs_size = (padded_k + 16) * n;  in gemm_f_1_0_7()
32819   std::int32_t* zipped_lhs_3_offsets =  in gemm_f_1_0_7()
32820       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 3);  in gemm_f_1_0_7()
32821   std::int32_t* zipped_lhs_1_offsets =  in gemm_f_1_0_7()
32822       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 1);  in gemm_f_1_0_7()
32825   const std::int32_t result_chunk_stride = result_stride * 3;  in gemm_f_1_0_7()
32827   const std::int32_t const_offset = lhs_offset * rhs_offset * k;  in gemm_f_1_0_7()
32830   const std::int32_t mul_result_chunk_stride_bytes = result_stride * 4;  in gemm_f_1_0_7()
32866                   const std::uint8_t* rhs, std::int32_t m, std::int32_t n,  in gemm_f_1_1_0()
32867                   std::int32_t k, std::int32_t lhs_offset,  in gemm_f_1_1_0()
32868                   std::int32_t rhs_offset, float result_scale, float* result,  in gemm_f_1_1_0()
32869                   std::int32_t result_stride) {  in gemm_f_1_1_0()
32870   const std::int32_t row_chunks = m / 3;  in gemm_f_1_1_0()
32871   const std::int32_t col_chunks = n / 3;  in gemm_f_1_1_0()
32872   const std::int32_t padded_k = ((k + 7) / 8) * 8;  in gemm_f_1_1_0()
32873   const std::int32_t chunk_size = k * 3;  in gemm_f_1_1_0()
32874   const std::int32_t zipped_chunk_size = (padded_k + 16) * 3;  in gemm_f_1_1_0()
32875   const std::int32_t zipped_rhs_size = (padded_k + 16) * n;  in gemm_f_1_1_0()
32879   std::int32_t* zipped_lhs_3_offsets =  in gemm_f_1_1_0()
32880       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 3);  in gemm_f_1_1_0()
32881   std::int32_t* zipped_lhs_1_offsets =  in gemm_f_1_1_0()
32882       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 1);  in gemm_f_1_1_0()
32885   const std::int32_t result_chunk_stride = result_stride * 3;  in gemm_f_1_1_0()
32887   const std::int32_t const_offset = lhs_offset * rhs_offset * k;  in gemm_f_1_1_0()
32890   const std::int32_t mul_result_chunk_stride_bytes = result_stride * 4;  in gemm_f_1_1_0()
32933                   const std::uint8_t* rhs, std::int32_t m, std::int32_t n,  in gemm_f_1_1_1()
32934                   std::int32_t k, std::int32_t lhs_offset,  in gemm_f_1_1_1()
32935                   std::int32_t rhs_offset, float result_scale, float* result,  in gemm_f_1_1_1()
32936                   std::int32_t result_stride) {  in gemm_f_1_1_1()
32937   const std::int32_t row_chunks = m / 3;  in gemm_f_1_1_1()
32938   const std::int32_t col_chunks = n / 3;  in gemm_f_1_1_1()
32939   const std::int32_t padded_k = ((k + 7) / 8) * 8;  in gemm_f_1_1_1()
32940   const std::int32_t chunk_size = k * 3;  in gemm_f_1_1_1()
32941   const std::int32_t zipped_chunk_size = (padded_k + 16) * 3;  in gemm_f_1_1_1()
32942   const std::int32_t zipped_rhs_size = (padded_k + 16) * n;  in gemm_f_1_1_1()
32946   std::int32_t* zipped_lhs_3_offsets =  in gemm_f_1_1_1()
32947       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 3);  in gemm_f_1_1_1()
32948   std::int32_t* zipped_lhs_1_offsets =  in gemm_f_1_1_1()
32949       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 1);  in gemm_f_1_1_1()
32952   const std::int32_t result_chunk_stride = result_stride * 3;  in gemm_f_1_1_1()
32954   const std::int32_t const_offset = lhs_offset * rhs_offset * k;  in gemm_f_1_1_1()
32957   const std::int32_t mul_result_chunk_stride_bytes = result_stride * 4;  in gemm_f_1_1_1()
33000                   const std::uint8_t* rhs, std::int32_t m, std::int32_t n,  in gemm_f_1_1_2()
33001                   std::int32_t k, std::int32_t lhs_offset,  in gemm_f_1_1_2()
33002                   std::int32_t rhs_offset, float result_scale, float* result,  in gemm_f_1_1_2()
33003                   std::int32_t result_stride) {  in gemm_f_1_1_2()
33004   const std::int32_t row_chunks = m / 3;  in gemm_f_1_1_2()
33005   const std::int32_t col_chunks = n / 3;  in gemm_f_1_1_2()
33006   const std::int32_t padded_k = ((k + 7) / 8) * 8;  in gemm_f_1_1_2()
33007   const std::int32_t chunk_size = k * 3;  in gemm_f_1_1_2()
33008   const std::int32_t zipped_chunk_size = (padded_k + 16) * 3;  in gemm_f_1_1_2()
33009   const std::int32_t zipped_rhs_size = (padded_k + 16) * n;  in gemm_f_1_1_2()
33013   std::int32_t* zipped_lhs_3_offsets =  in gemm_f_1_1_2()
33014       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 3);  in gemm_f_1_1_2()
33015   std::int32_t* zipped_lhs_1_offsets =  in gemm_f_1_1_2()
33016       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 1);  in gemm_f_1_1_2()
33019   const std::int32_t result_chunk_stride = result_stride * 3;  in gemm_f_1_1_2()
33021   const std::int32_t const_offset = lhs_offset * rhs_offset * k;  in gemm_f_1_1_2()
33024   const std::int32_t mul_result_chunk_stride_bytes = result_stride * 4;  in gemm_f_1_1_2()
33067                   const std::uint8_t* rhs, std::int32_t m, std::int32_t n,  in gemm_f_1_1_3()
33068                   std::int32_t k, std::int32_t lhs_offset,  in gemm_f_1_1_3()
33069                   std::int32_t rhs_offset, float result_scale, float* result,  in gemm_f_1_1_3()
33070                   std::int32_t result_stride) {  in gemm_f_1_1_3()
33071   const std::int32_t row_chunks = m / 3;  in gemm_f_1_1_3()
33072   const std::int32_t col_chunks = n / 3;  in gemm_f_1_1_3()
33073   const std::int32_t padded_k = ((k + 7) / 8) * 8;  in gemm_f_1_1_3()
33074   const std::int32_t chunk_size = k * 3;  in gemm_f_1_1_3()
33075   const std::int32_t zipped_chunk_size = (padded_k + 16) * 3;  in gemm_f_1_1_3()
33076   const std::int32_t zipped_rhs_size = (padded_k + 16) * n;  in gemm_f_1_1_3()
33080   std::int32_t* zipped_lhs_3_offsets =  in gemm_f_1_1_3()
33081       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 3);  in gemm_f_1_1_3()
33082   std::int32_t* zipped_lhs_1_offsets =  in gemm_f_1_1_3()
33083       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 1);  in gemm_f_1_1_3()
33086   const std::int32_t result_chunk_stride = result_stride * 3;  in gemm_f_1_1_3()
33088   const std::int32_t const_offset = lhs_offset * rhs_offset * k;  in gemm_f_1_1_3()
33091   const std::int32_t mul_result_chunk_stride_bytes = result_stride * 4;  in gemm_f_1_1_3()
33134                   const std::uint8_t* rhs, std::int32_t m, std::int32_t n,  in gemm_f_1_1_4()
33135                   std::int32_t k, std::int32_t lhs_offset,  in gemm_f_1_1_4()
33136                   std::int32_t rhs_offset, float result_scale, float* result,  in gemm_f_1_1_4()
33137                   std::int32_t result_stride) {  in gemm_f_1_1_4()
33138   const std::int32_t row_chunks = m / 3;  in gemm_f_1_1_4()
33139   const std::int32_t col_chunks = n / 3;  in gemm_f_1_1_4()
33140   const std::int32_t padded_k = ((k + 7) / 8) * 8;  in gemm_f_1_1_4()
33141   const std::int32_t chunk_size = k * 3;  in gemm_f_1_1_4()
33142   const std::int32_t zipped_chunk_size = (padded_k + 16) * 3;  in gemm_f_1_1_4()
33143   const std::int32_t zipped_rhs_size = (padded_k + 16) * n;  in gemm_f_1_1_4()
33147   std::int32_t* zipped_lhs_3_offsets =  in gemm_f_1_1_4()
33148       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 3);  in gemm_f_1_1_4()
33149   std::int32_t* zipped_lhs_1_offsets =  in gemm_f_1_1_4()
33150       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 1);  in gemm_f_1_1_4()
33153   const std::int32_t result_chunk_stride = result_stride * 3;  in gemm_f_1_1_4()
33155   const std::int32_t const_offset = lhs_offset * rhs_offset * k;  in gemm_f_1_1_4()
33158   const std::int32_t mul_result_chunk_stride_bytes = result_stride * 4;  in gemm_f_1_1_4()
33201                   const std::uint8_t* rhs, std::int32_t m, std::int32_t n,  in gemm_f_1_1_5()
33202                   std::int32_t k, std::int32_t lhs_offset,  in gemm_f_1_1_5()
33203                   std::int32_t rhs_offset, float result_scale, float* result,  in gemm_f_1_1_5()
33204                   std::int32_t result_stride) {  in gemm_f_1_1_5()
33205   const std::int32_t row_chunks = m / 3;  in gemm_f_1_1_5()
33206   const std::int32_t col_chunks = n / 3;  in gemm_f_1_1_5()
33207   const std::int32_t padded_k = ((k + 7) / 8) * 8;  in gemm_f_1_1_5()
33208   const std::int32_t chunk_size = k * 3;  in gemm_f_1_1_5()
33209   const std::int32_t zipped_chunk_size = (padded_k + 16) * 3;  in gemm_f_1_1_5()
33210   const std::int32_t zipped_rhs_size = (padded_k + 16) * n;  in gemm_f_1_1_5()
33214   std::int32_t* zipped_lhs_3_offsets =  in gemm_f_1_1_5()
33215       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 3);  in gemm_f_1_1_5()
33216   std::int32_t* zipped_lhs_1_offsets =  in gemm_f_1_1_5()
33217       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 1);  in gemm_f_1_1_5()
33220   const std::int32_t result_chunk_stride = result_stride * 3;  in gemm_f_1_1_5()
33222   const std::int32_t const_offset = lhs_offset * rhs_offset * k;  in gemm_f_1_1_5()
33225   const std::int32_t mul_result_chunk_stride_bytes = result_stride * 4;  in gemm_f_1_1_5()
33268                   const std::uint8_t* rhs, std::int32_t m, std::int32_t n,  in gemm_f_1_1_6()
33269                   std::int32_t k, std::int32_t lhs_offset,  in gemm_f_1_1_6()
33270                   std::int32_t rhs_offset, float result_scale, float* result,  in gemm_f_1_1_6()
33271                   std::int32_t result_stride) {  in gemm_f_1_1_6()
33272   const std::int32_t row_chunks = m / 3;  in gemm_f_1_1_6()
33273   const std::int32_t col_chunks = n / 3;  in gemm_f_1_1_6()
33274   const std::int32_t padded_k = ((k + 7) / 8) * 8;  in gemm_f_1_1_6()
33275   const std::int32_t chunk_size = k * 3;  in gemm_f_1_1_6()
33276   const std::int32_t zipped_chunk_size = (padded_k + 16) * 3;  in gemm_f_1_1_6()
33277   const std::int32_t zipped_rhs_size = (padded_k + 16) * n;  in gemm_f_1_1_6()
33281   std::int32_t* zipped_lhs_3_offsets =  in gemm_f_1_1_6()
33282       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 3);  in gemm_f_1_1_6()
33283   std::int32_t* zipped_lhs_1_offsets =  in gemm_f_1_1_6()
33284       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 1);  in gemm_f_1_1_6()
33287   const std::int32_t result_chunk_stride = result_stride * 3;  in gemm_f_1_1_6()
33289   const std::int32_t const_offset = lhs_offset * rhs_offset * k;  in gemm_f_1_1_6()
33292   const std::int32_t mul_result_chunk_stride_bytes = result_stride * 4;  in gemm_f_1_1_6()
33335                   const std::uint8_t* rhs, std::int32_t m, std::int32_t n,  in gemm_f_1_1_7()
33336                   std::int32_t k, std::int32_t lhs_offset,  in gemm_f_1_1_7()
33337                   std::int32_t rhs_offset, float result_scale, float* result,  in gemm_f_1_1_7()
33338                   std::int32_t result_stride) {  in gemm_f_1_1_7()
33339   const std::int32_t row_chunks = m / 3;  in gemm_f_1_1_7()
33340   const std::int32_t col_chunks = n / 3;  in gemm_f_1_1_7()
33341   const std::int32_t padded_k = ((k + 7) / 8) * 8;  in gemm_f_1_1_7()
33342   const std::int32_t chunk_size = k * 3;  in gemm_f_1_1_7()
33343   const std::int32_t zipped_chunk_size = (padded_k + 16) * 3;  in gemm_f_1_1_7()
33344   const std::int32_t zipped_rhs_size = (padded_k + 16) * n;  in gemm_f_1_1_7()
33348   std::int32_t* zipped_lhs_3_offsets =  in gemm_f_1_1_7()
33349       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 3);  in gemm_f_1_1_7()
33350   std::int32_t* zipped_lhs_1_offsets =  in gemm_f_1_1_7()
33351       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 1);  in gemm_f_1_1_7()
33354   const std::int32_t result_chunk_stride = result_stride * 3;  in gemm_f_1_1_7()
33356   const std::int32_t const_offset = lhs_offset * rhs_offset * k;  in gemm_f_1_1_7()
33359   const std::int32_t mul_result_chunk_stride_bytes = result_stride * 4;  in gemm_f_1_1_7()
33402                   const std::uint8_t* rhs, std::int32_t m, std::int32_t n,  in gemm_f_1_2_0()
33403                   std::int32_t k, std::int32_t lhs_offset,  in gemm_f_1_2_0()
33404                   std::int32_t rhs_offset, float result_scale, float* result,  in gemm_f_1_2_0()
33405                   std::int32_t result_stride) {  in gemm_f_1_2_0()
33406   const std::int32_t row_chunks = m / 3;  in gemm_f_1_2_0()
33407   const std::int32_t col_chunks = n / 3;  in gemm_f_1_2_0()
33408   const std::int32_t padded_k = ((k + 7) / 8) * 8;  in gemm_f_1_2_0()
33409   const std::int32_t chunk_size = k * 3;  in gemm_f_1_2_0()
33410   const std::int32_t zipped_chunk_size = (padded_k + 16) * 3;  in gemm_f_1_2_0()
33411   const std::int32_t zipped_rhs_size = (padded_k + 16) * n;  in gemm_f_1_2_0()
33415   std::int32_t* zipped_lhs_3_offsets =  in gemm_f_1_2_0()
33416       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 3);  in gemm_f_1_2_0()
33417   std::int32_t* zipped_lhs_1_offsets =  in gemm_f_1_2_0()
33418       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 1);  in gemm_f_1_2_0()
33421   const std::int32_t result_chunk_stride = result_stride * 3;  in gemm_f_1_2_0()
33423   const std::int32_t const_offset = lhs_offset * rhs_offset * k;  in gemm_f_1_2_0()
33426   const std::int32_t mul_result_chunk_stride_bytes = result_stride * 4;  in gemm_f_1_2_0()
33469                   const std::uint8_t* rhs, std::int32_t m, std::int32_t n,  in gemm_f_1_2_1()
33470                   std::int32_t k, std::int32_t lhs_offset,  in gemm_f_1_2_1()
33471                   std::int32_t rhs_offset, float result_scale, float* result,  in gemm_f_1_2_1()
33472                   std::int32_t result_stride) {  in gemm_f_1_2_1()
33473   const std::int32_t row_chunks = m / 3;  in gemm_f_1_2_1()
33474   const std::int32_t col_chunks = n / 3;  in gemm_f_1_2_1()
33475   const std::int32_t padded_k = ((k + 7) / 8) * 8;  in gemm_f_1_2_1()
33476   const std::int32_t chunk_size = k * 3;  in gemm_f_1_2_1()
33477   const std::int32_t zipped_chunk_size = (padded_k + 16) * 3;  in gemm_f_1_2_1()
33478   const std::int32_t zipped_rhs_size = (padded_k + 16) * n;  in gemm_f_1_2_1()
33482   std::int32_t* zipped_lhs_3_offsets =  in gemm_f_1_2_1()
33483       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 3);  in gemm_f_1_2_1()
33484   std::int32_t* zipped_lhs_1_offsets =  in gemm_f_1_2_1()
33485       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 1);  in gemm_f_1_2_1()
33488   const std::int32_t result_chunk_stride = result_stride * 3;  in gemm_f_1_2_1()
33490   const std::int32_t const_offset = lhs_offset * rhs_offset * k;  in gemm_f_1_2_1()
33493   const std::int32_t mul_result_chunk_stride_bytes = result_stride * 4;  in gemm_f_1_2_1()
33536                   const std::uint8_t* rhs, std::int32_t m, std::int32_t n,  in gemm_f_1_2_2()
33537                   std::int32_t k, std::int32_t lhs_offset,  in gemm_f_1_2_2()
33538                   std::int32_t rhs_offset, float result_scale, float* result,  in gemm_f_1_2_2()
33539                   std::int32_t result_stride) {  in gemm_f_1_2_2()
33540   const std::int32_t row_chunks = m / 3;  in gemm_f_1_2_2()
33541   const std::int32_t col_chunks = n / 3;  in gemm_f_1_2_2()
33542   const std::int32_t padded_k = ((k + 7) / 8) * 8;  in gemm_f_1_2_2()
33543   const std::int32_t chunk_size = k * 3;  in gemm_f_1_2_2()
33544   const std::int32_t zipped_chunk_size = (padded_k + 16) * 3;  in gemm_f_1_2_2()
33545   const std::int32_t zipped_rhs_size = (padded_k + 16) * n;  in gemm_f_1_2_2()
33549   std::int32_t* zipped_lhs_3_offsets =  in gemm_f_1_2_2()
33550       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 3);  in gemm_f_1_2_2()
33551   std::int32_t* zipped_lhs_1_offsets =  in gemm_f_1_2_2()
33552       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 1);  in gemm_f_1_2_2()
33555   const std::int32_t result_chunk_stride = result_stride * 3;  in gemm_f_1_2_2()
33557   const std::int32_t const_offset = lhs_offset * rhs_offset * k;  in gemm_f_1_2_2()
33560   const std::int32_t mul_result_chunk_stride_bytes = result_stride * 4;  in gemm_f_1_2_2()
33603                   const std::uint8_t* rhs, std::int32_t m, std::int32_t n,  in gemm_f_1_2_3()
33604                   std::int32_t k, std::int32_t lhs_offset,  in gemm_f_1_2_3()
33605                   std::int32_t rhs_offset, float result_scale, float* result,  in gemm_f_1_2_3()
33606                   std::int32_t result_stride) {  in gemm_f_1_2_3()
33607   const std::int32_t row_chunks = m / 3;  in gemm_f_1_2_3()
33608   const std::int32_t col_chunks = n / 3;  in gemm_f_1_2_3()
33609   const std::int32_t padded_k = ((k + 7) / 8) * 8;  in gemm_f_1_2_3()
33610   const std::int32_t chunk_size = k * 3;  in gemm_f_1_2_3()
33611   const std::int32_t zipped_chunk_size = (padded_k + 16) * 3;  in gemm_f_1_2_3()
33612   const std::int32_t zipped_rhs_size = (padded_k + 16) * n;  in gemm_f_1_2_3()
33616   std::int32_t* zipped_lhs_3_offsets =  in gemm_f_1_2_3()
33617       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 3);  in gemm_f_1_2_3()
33618   std::int32_t* zipped_lhs_1_offsets =  in gemm_f_1_2_3()
33619       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 1);  in gemm_f_1_2_3()
33622   const std::int32_t result_chunk_stride = result_stride * 3;  in gemm_f_1_2_3()
33624   const std::int32_t const_offset = lhs_offset * rhs_offset * k;  in gemm_f_1_2_3()
33627   const std::int32_t mul_result_chunk_stride_bytes = result_stride * 4;  in gemm_f_1_2_3()
33670                   const std::uint8_t* rhs, std::int32_t m, std::int32_t n,  in gemm_f_1_2_4()
33671                   std::int32_t k, std::int32_t lhs_offset,  in gemm_f_1_2_4()
33672                   std::int32_t rhs_offset, float result_scale, float* result,  in gemm_f_1_2_4()
33673                   std::int32_t result_stride) {  in gemm_f_1_2_4()
33674   const std::int32_t row_chunks = m / 3;  in gemm_f_1_2_4()
33675   const std::int32_t col_chunks = n / 3;  in gemm_f_1_2_4()
33676   const std::int32_t padded_k = ((k + 7) / 8) * 8;  in gemm_f_1_2_4()
33677   const std::int32_t chunk_size = k * 3;  in gemm_f_1_2_4()
33678   const std::int32_t zipped_chunk_size = (padded_k + 16) * 3;  in gemm_f_1_2_4()
33679   const std::int32_t zipped_rhs_size = (padded_k + 16) * n;  in gemm_f_1_2_4()
33683   std::int32_t* zipped_lhs_3_offsets =  in gemm_f_1_2_4()
33684       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 3);  in gemm_f_1_2_4()
33685   std::int32_t* zipped_lhs_1_offsets =  in gemm_f_1_2_4()
33686       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 1);  in gemm_f_1_2_4()
33689   const std::int32_t result_chunk_stride = result_stride * 3;  in gemm_f_1_2_4()
33691   const std::int32_t const_offset = lhs_offset * rhs_offset * k;  in gemm_f_1_2_4()
33694   const std::int32_t mul_result_chunk_stride_bytes = result_stride * 4;  in gemm_f_1_2_4()
33737                   const std::uint8_t* rhs, std::int32_t m, std::int32_t n,  in gemm_f_1_2_5()
33738                   std::int32_t k, std::int32_t lhs_offset,  in gemm_f_1_2_5()
33739                   std::int32_t rhs_offset, float result_scale, float* result,  in gemm_f_1_2_5()
33740                   std::int32_t result_stride) {  in gemm_f_1_2_5()
33741   const std::int32_t row_chunks = m / 3;  in gemm_f_1_2_5()
33742   const std::int32_t col_chunks = n / 3;  in gemm_f_1_2_5()
33743   const std::int32_t padded_k = ((k + 7) / 8) * 8;  in gemm_f_1_2_5()
33744   const std::int32_t chunk_size = k * 3;  in gemm_f_1_2_5()
33745   const std::int32_t zipped_chunk_size = (padded_k + 16) * 3;  in gemm_f_1_2_5()
33746   const std::int32_t zipped_rhs_size = (padded_k + 16) * n;  in gemm_f_1_2_5()
33750   std::int32_t* zipped_lhs_3_offsets =  in gemm_f_1_2_5()
33751       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 3);  in gemm_f_1_2_5()
33752   std::int32_t* zipped_lhs_1_offsets =  in gemm_f_1_2_5()
33753       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 1);  in gemm_f_1_2_5()
33756   const std::int32_t result_chunk_stride = result_stride * 3;  in gemm_f_1_2_5()
33758   const std::int32_t const_offset = lhs_offset * rhs_offset * k;  in gemm_f_1_2_5()
33761   const std::int32_t mul_result_chunk_stride_bytes = result_stride * 4;  in gemm_f_1_2_5()
33804                   const std::uint8_t* rhs, std::int32_t m, std::int32_t n,  in gemm_f_1_2_6()
33805                   std::int32_t k, std::int32_t lhs_offset,  in gemm_f_1_2_6()
33806                   std::int32_t rhs_offset, float result_scale, float* result,  in gemm_f_1_2_6()
33807                   std::int32_t result_stride) {  in gemm_f_1_2_6()
33808   const std::int32_t row_chunks = m / 3;  in gemm_f_1_2_6()
33809   const std::int32_t col_chunks = n / 3;  in gemm_f_1_2_6()
33810   const std::int32_t padded_k = ((k + 7) / 8) * 8;  in gemm_f_1_2_6()
33811   const std::int32_t chunk_size = k * 3;  in gemm_f_1_2_6()
33812   const std::int32_t zipped_chunk_size = (padded_k + 16) * 3;  in gemm_f_1_2_6()
33813   const std::int32_t zipped_rhs_size = (padded_k + 16) * n;  in gemm_f_1_2_6()
33817   std::int32_t* zipped_lhs_3_offsets =  in gemm_f_1_2_6()
33818       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 3);  in gemm_f_1_2_6()
33819   std::int32_t* zipped_lhs_1_offsets =  in gemm_f_1_2_6()
33820       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 1);  in gemm_f_1_2_6()
33823   const std::int32_t result_chunk_stride = result_stride * 3;  in gemm_f_1_2_6()
33825   const std::int32_t const_offset = lhs_offset * rhs_offset * k;  in gemm_f_1_2_6()
33828   const std::int32_t mul_result_chunk_stride_bytes = result_stride * 4;  in gemm_f_1_2_6()
33871                   const std::uint8_t* rhs, std::int32_t m, std::int32_t n,  in gemm_f_1_2_7()
33872                   std::int32_t k, std::int32_t lhs_offset,  in gemm_f_1_2_7()
33873                   std::int32_t rhs_offset, float result_scale, float* result,  in gemm_f_1_2_7()
33874                   std::int32_t result_stride) {  in gemm_f_1_2_7()
33875   const std::int32_t row_chunks = m / 3;  in gemm_f_1_2_7()
33876   const std::int32_t col_chunks = n / 3;  in gemm_f_1_2_7()
33877   const std::int32_t padded_k = ((k + 7) / 8) * 8;  in gemm_f_1_2_7()
33878   const std::int32_t chunk_size = k * 3;  in gemm_f_1_2_7()
33879   const std::int32_t zipped_chunk_size = (padded_k + 16) * 3;  in gemm_f_1_2_7()
33880   const std::int32_t zipped_rhs_size = (padded_k + 16) * n;  in gemm_f_1_2_7()
33884   std::int32_t* zipped_lhs_3_offsets =  in gemm_f_1_2_7()
33885       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 3);  in gemm_f_1_2_7()
33886   std::int32_t* zipped_lhs_1_offsets =  in gemm_f_1_2_7()
33887       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 1);  in gemm_f_1_2_7()
33890   const std::int32_t result_chunk_stride = result_stride * 3;  in gemm_f_1_2_7()
33892   const std::int32_t const_offset = lhs_offset * rhs_offset * k;  in gemm_f_1_2_7()
33895   const std::int32_t mul_result_chunk_stride_bytes = result_stride * 4;  in gemm_f_1_2_7()
33938                   const std::uint8_t* rhs, std::int32_t m, std::int32_t n,  in gemm_f_2_0_0()
33939                   std::int32_t k, std::int32_t lhs_offset,  in gemm_f_2_0_0()
33940                   std::int32_t rhs_offset, float result_scale, float* result,  in gemm_f_2_0_0()
33941                   std::int32_t result_stride) {  in gemm_f_2_0_0()
33942   const std::int32_t row_chunks = m / 3;  in gemm_f_2_0_0()
33943   const std::int32_t col_chunks = n / 3;  in gemm_f_2_0_0()
33944   const std::int32_t padded_k = ((k + 7) / 8) * 8;  in gemm_f_2_0_0()
33945   const std::int32_t chunk_size = k * 3;  in gemm_f_2_0_0()
33946   const std::int32_t zipped_chunk_size = (padded_k + 16) * 3;  in gemm_f_2_0_0()
33947   const std::int32_t zipped_rhs_size = (padded_k + 16) * n;  in gemm_f_2_0_0()
33951   std::int32_t* zipped_lhs_3_offsets =  in gemm_f_2_0_0()
33952       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 3);  in gemm_f_2_0_0()
33953   std::int32_t* zipped_lhs_2_offsets =  in gemm_f_2_0_0()
33954       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 2);  in gemm_f_2_0_0()
33957   const std::int32_t result_chunk_stride = result_stride * 3;  in gemm_f_2_0_0()
33959   const std::int32_t const_offset = lhs_offset * rhs_offset * k;  in gemm_f_2_0_0()
33962   const std::int32_t mul_result_chunk_stride_bytes = result_stride * 4;  in gemm_f_2_0_0()
33998                   const std::uint8_t* rhs, std::int32_t m, std::int32_t n,  in gemm_f_2_0_1()
33999                   std::int32_t k, std::int32_t lhs_offset,  in gemm_f_2_0_1()
34000                   std::int32_t rhs_offset, float result_scale, float* result,  in gemm_f_2_0_1()
34001                   std::int32_t result_stride) {  in gemm_f_2_0_1()
34002   const std::int32_t row_chunks = m / 3;  in gemm_f_2_0_1()
34003   const std::int32_t col_chunks = n / 3;  in gemm_f_2_0_1()
34004   const std::int32_t padded_k = ((k + 7) / 8) * 8;  in gemm_f_2_0_1()
34005   const std::int32_t chunk_size = k * 3;  in gemm_f_2_0_1()
34006   const std::int32_t zipped_chunk_size = (padded_k + 16) * 3;  in gemm_f_2_0_1()
34007   const std::int32_t zipped_rhs_size = (padded_k + 16) * n;  in gemm_f_2_0_1()
34011   std::int32_t* zipped_lhs_3_offsets =  in gemm_f_2_0_1()
34012       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 3);  in gemm_f_2_0_1()
34013   std::int32_t* zipped_lhs_2_offsets =  in gemm_f_2_0_1()
34014       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 2);  in gemm_f_2_0_1()
34017   const std::int32_t result_chunk_stride = result_stride * 3;  in gemm_f_2_0_1()
34019   const std::int32_t const_offset = lhs_offset * rhs_offset * k;  in gemm_f_2_0_1()
34022   const std::int32_t mul_result_chunk_stride_bytes = result_stride * 4;  in gemm_f_2_0_1()
34058                   const std::uint8_t* rhs, std::int32_t m, std::int32_t n,  in gemm_f_2_0_2()
34059                   std::int32_t k, std::int32_t lhs_offset,  in gemm_f_2_0_2()
34060                   std::int32_t rhs_offset, float result_scale, float* result,  in gemm_f_2_0_2()
34061                   std::int32_t result_stride) {  in gemm_f_2_0_2()
34062   const std::int32_t row_chunks = m / 3;  in gemm_f_2_0_2()
34063   const std::int32_t col_chunks = n / 3;  in gemm_f_2_0_2()
34064   const std::int32_t padded_k = ((k + 7) / 8) * 8;  in gemm_f_2_0_2()
34065   const std::int32_t chunk_size = k * 3;  in gemm_f_2_0_2()
34066   const std::int32_t zipped_chunk_size = (padded_k + 16) * 3;  in gemm_f_2_0_2()
34067   const std::int32_t zipped_rhs_size = (padded_k + 16) * n;  in gemm_f_2_0_2()
34071   std::int32_t* zipped_lhs_3_offsets =  in gemm_f_2_0_2()
34072       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 3);  in gemm_f_2_0_2()
34073   std::int32_t* zipped_lhs_2_offsets =  in gemm_f_2_0_2()
34074       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 2);  in gemm_f_2_0_2()
34077   const std::int32_t result_chunk_stride = result_stride * 3;  in gemm_f_2_0_2()
34079   const std::int32_t const_offset = lhs_offset * rhs_offset * k;  in gemm_f_2_0_2()
34082   const std::int32_t mul_result_chunk_stride_bytes = result_stride * 4;  in gemm_f_2_0_2()
34118                   const std::uint8_t* rhs, std::int32_t m, std::int32_t n,  in gemm_f_2_0_3()
34119                   std::int32_t k, std::int32_t lhs_offset,  in gemm_f_2_0_3()
34120                   std::int32_t rhs_offset, float result_scale, float* result,  in gemm_f_2_0_3()
34121                   std::int32_t result_stride) {  in gemm_f_2_0_3()
34122   const std::int32_t row_chunks = m / 3;  in gemm_f_2_0_3()
34123   const std::int32_t col_chunks = n / 3;  in gemm_f_2_0_3()
34124   const std::int32_t padded_k = ((k + 7) / 8) * 8;  in gemm_f_2_0_3()
34125   const std::int32_t chunk_size = k * 3;  in gemm_f_2_0_3()
34126   const std::int32_t zipped_chunk_size = (padded_k + 16) * 3;  in gemm_f_2_0_3()
34127   const std::int32_t zipped_rhs_size = (padded_k + 16) * n;  in gemm_f_2_0_3()
34131   std::int32_t* zipped_lhs_3_offsets =  in gemm_f_2_0_3()
34132       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 3);  in gemm_f_2_0_3()
34133   std::int32_t* zipped_lhs_2_offsets =  in gemm_f_2_0_3()
34134       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 2);  in gemm_f_2_0_3()
34137   const std::int32_t result_chunk_stride = result_stride * 3;  in gemm_f_2_0_3()
34139   const std::int32_t const_offset = lhs_offset * rhs_offset * k;  in gemm_f_2_0_3()
34142   const std::int32_t mul_result_chunk_stride_bytes = result_stride * 4;  in gemm_f_2_0_3()
34178                   const std::uint8_t* rhs, std::int32_t m, std::int32_t n,  in gemm_f_2_0_4()
34179                   std::int32_t k, std::int32_t lhs_offset,  in gemm_f_2_0_4()
34180                   std::int32_t rhs_offset, float result_scale, float* result,  in gemm_f_2_0_4()
34181                   std::int32_t result_stride) {  in gemm_f_2_0_4()
34182   const std::int32_t row_chunks = m / 3;  in gemm_f_2_0_4()
34183   const std::int32_t col_chunks = n / 3;  in gemm_f_2_0_4()
34184   const std::int32_t padded_k = ((k + 7) / 8) * 8;  in gemm_f_2_0_4()
34185   const std::int32_t chunk_size = k * 3;  in gemm_f_2_0_4()
34186   const std::int32_t zipped_chunk_size = (padded_k + 16) * 3;  in gemm_f_2_0_4()
34187   const std::int32_t zipped_rhs_size = (padded_k + 16) * n;  in gemm_f_2_0_4()
34191   std::int32_t* zipped_lhs_3_offsets =  in gemm_f_2_0_4()
34192       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 3);  in gemm_f_2_0_4()
34193   std::int32_t* zipped_lhs_2_offsets =  in gemm_f_2_0_4()
34194       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 2);  in gemm_f_2_0_4()
34197   const std::int32_t result_chunk_stride = result_stride * 3;  in gemm_f_2_0_4()
34199   const std::int32_t const_offset = lhs_offset * rhs_offset * k;  in gemm_f_2_0_4()
34202   const std::int32_t mul_result_chunk_stride_bytes = result_stride * 4;  in gemm_f_2_0_4()
34238                   const std::uint8_t* rhs, std::int32_t m, std::int32_t n,  in gemm_f_2_0_5()
34239                   std::int32_t k, std::int32_t lhs_offset,  in gemm_f_2_0_5()
34240                   std::int32_t rhs_offset, float result_scale, float* result,  in gemm_f_2_0_5()
34241                   std::int32_t result_stride) {  in gemm_f_2_0_5()
34242   const std::int32_t row_chunks = m / 3;  in gemm_f_2_0_5()
34243   const std::int32_t col_chunks = n / 3;  in gemm_f_2_0_5()
34244   const std::int32_t padded_k = ((k + 7) / 8) * 8;  in gemm_f_2_0_5()
34245   const std::int32_t chunk_size = k * 3;  in gemm_f_2_0_5()
34246   const std::int32_t zipped_chunk_size = (padded_k + 16) * 3;  in gemm_f_2_0_5()
34247   const std::int32_t zipped_rhs_size = (padded_k + 16) * n;  in gemm_f_2_0_5()
34251   std::int32_t* zipped_lhs_3_offsets =  in gemm_f_2_0_5()
34252       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 3);  in gemm_f_2_0_5()
34253   std::int32_t* zipped_lhs_2_offsets =  in gemm_f_2_0_5()
34254       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 2);  in gemm_f_2_0_5()
34257   const std::int32_t result_chunk_stride = result_stride * 3;  in gemm_f_2_0_5()
34259   const std::int32_t const_offset = lhs_offset * rhs_offset * k;  in gemm_f_2_0_5()
34262   const std::int32_t mul_result_chunk_stride_bytes = result_stride * 4;  in gemm_f_2_0_5()
34298                   const std::uint8_t* rhs, std::int32_t m, std::int32_t n,  in gemm_f_2_0_6()
34299                   std::int32_t k, std::int32_t lhs_offset,  in gemm_f_2_0_6()
34300                   std::int32_t rhs_offset, float result_scale, float* result,  in gemm_f_2_0_6()
34301                   std::int32_t result_stride) {  in gemm_f_2_0_6()
34302   const std::int32_t row_chunks = m / 3;  in gemm_f_2_0_6()
34303   const std::int32_t col_chunks = n / 3;  in gemm_f_2_0_6()
34304   const std::int32_t padded_k = ((k + 7) / 8) * 8;  in gemm_f_2_0_6()
34305   const std::int32_t chunk_size = k * 3;  in gemm_f_2_0_6()
34306   const std::int32_t zipped_chunk_size = (padded_k + 16) * 3;  in gemm_f_2_0_6()
34307   const std::int32_t zipped_rhs_size = (padded_k + 16) * n;  in gemm_f_2_0_6()
34311   std::int32_t* zipped_lhs_3_offsets =  in gemm_f_2_0_6()
34312       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 3);  in gemm_f_2_0_6()
34313   std::int32_t* zipped_lhs_2_offsets =  in gemm_f_2_0_6()
34314       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 2);  in gemm_f_2_0_6()
34317   const std::int32_t result_chunk_stride = result_stride * 3;  in gemm_f_2_0_6()
34319   const std::int32_t const_offset = lhs_offset * rhs_offset * k;  in gemm_f_2_0_6()
34322   const std::int32_t mul_result_chunk_stride_bytes = result_stride * 4;  in gemm_f_2_0_6()
34358                   const std::uint8_t* rhs, std::int32_t m, std::int32_t n,  in gemm_f_2_0_7()
34359                   std::int32_t k, std::int32_t lhs_offset,  in gemm_f_2_0_7()
34360                   std::int32_t rhs_offset, float result_scale, float* result,  in gemm_f_2_0_7()
34361                   std::int32_t result_stride) {  in gemm_f_2_0_7()
34362   const std::int32_t row_chunks = m / 3;  in gemm_f_2_0_7()
34363   const std::int32_t col_chunks = n / 3;  in gemm_f_2_0_7()
34364   const std::int32_t padded_k = ((k + 7) / 8) * 8;  in gemm_f_2_0_7()
34365   const std::int32_t chunk_size = k * 3;  in gemm_f_2_0_7()
34366   const std::int32_t zipped_chunk_size = (padded_k + 16) * 3;  in gemm_f_2_0_7()
34367   const std::int32_t zipped_rhs_size = (padded_k + 16) * n;  in gemm_f_2_0_7()
34371   std::int32_t* zipped_lhs_3_offsets =  in gemm_f_2_0_7()
34372       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 3);  in gemm_f_2_0_7()
34373   std::int32_t* zipped_lhs_2_offsets =  in gemm_f_2_0_7()
34374       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 2);  in gemm_f_2_0_7()
34377   const std::int32_t result_chunk_stride = result_stride * 3;  in gemm_f_2_0_7()
34379   const std::int32_t const_offset = lhs_offset * rhs_offset * k;  in gemm_f_2_0_7()
34382   const std::int32_t mul_result_chunk_stride_bytes = result_stride * 4;  in gemm_f_2_0_7()
34418                   const std::uint8_t* rhs, std::int32_t m, std::int32_t n,  in gemm_f_2_1_0()
34419                   std::int32_t k, std::int32_t lhs_offset,  in gemm_f_2_1_0()
34420                   std::int32_t rhs_offset, float result_scale, float* result,  in gemm_f_2_1_0()
34421                   std::int32_t result_stride) {  in gemm_f_2_1_0()
34422   const std::int32_t row_chunks = m / 3;  in gemm_f_2_1_0()
34423   const std::int32_t col_chunks = n / 3;  in gemm_f_2_1_0()
34424   const std::int32_t padded_k = ((k + 7) / 8) * 8;  in gemm_f_2_1_0()
34425   const std::int32_t chunk_size = k * 3;  in gemm_f_2_1_0()
34426   const std::int32_t zipped_chunk_size = (padded_k + 16) * 3;  in gemm_f_2_1_0()
34427   const std::int32_t zipped_rhs_size = (padded_k + 16) * n;  in gemm_f_2_1_0()
34431   std::int32_t* zipped_lhs_3_offsets =  in gemm_f_2_1_0()
34432       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 3);  in gemm_f_2_1_0()
34433   std::int32_t* zipped_lhs_2_offsets =  in gemm_f_2_1_0()
34434       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 2);  in gemm_f_2_1_0()
34437   const std::int32_t result_chunk_stride = result_stride * 3;  in gemm_f_2_1_0()
34439   const std::int32_t const_offset = lhs_offset * rhs_offset * k;  in gemm_f_2_1_0()
34442   const std::int32_t mul_result_chunk_stride_bytes = result_stride * 4;  in gemm_f_2_1_0()
34485                   const std::uint8_t* rhs, std::int32_t m, std::int32_t n,  in gemm_f_2_1_1()
34486                   std::int32_t k, std::int32_t lhs_offset,  in gemm_f_2_1_1()
34487                   std::int32_t rhs_offset, float result_scale, float* result,  in gemm_f_2_1_1()
34488                   std::int32_t result_stride) {  in gemm_f_2_1_1()
34489   const std::int32_t row_chunks = m / 3;  in gemm_f_2_1_1()
34490   const std::int32_t col_chunks = n / 3;  in gemm_f_2_1_1()
34491   const std::int32_t padded_k = ((k + 7) / 8) * 8;  in gemm_f_2_1_1()
34492   const std::int32_t chunk_size = k * 3;  in gemm_f_2_1_1()
34493   const std::int32_t zipped_chunk_size = (padded_k + 16) * 3;  in gemm_f_2_1_1()
34494   const std::int32_t zipped_rhs_size = (padded_k + 16) * n;  in gemm_f_2_1_1()
34498   std::int32_t* zipped_lhs_3_offsets =  in gemm_f_2_1_1()
34499       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 3);  in gemm_f_2_1_1()
34500   std::int32_t* zipped_lhs_2_offsets =  in gemm_f_2_1_1()
34501       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 2);  in gemm_f_2_1_1()
34504   const std::int32_t result_chunk_stride = result_stride * 3;  in gemm_f_2_1_1()
34506   const std::int32_t const_offset = lhs_offset * rhs_offset * k;  in gemm_f_2_1_1()
34509   const std::int32_t mul_result_chunk_stride_bytes = result_stride * 4;  in gemm_f_2_1_1()
34552                   const std::uint8_t* rhs, std::int32_t m, std::int32_t n,  in gemm_f_2_1_2()
34553                   std::int32_t k, std::int32_t lhs_offset,  in gemm_f_2_1_2()
34554                   std::int32_t rhs_offset, float result_scale, float* result,  in gemm_f_2_1_2()
34555                   std::int32_t result_stride) {  in gemm_f_2_1_2()
34556   const std::int32_t row_chunks = m / 3;  in gemm_f_2_1_2()
34557   const std::int32_t col_chunks = n / 3;  in gemm_f_2_1_2()
34558   const std::int32_t padded_k = ((k + 7) / 8) * 8;  in gemm_f_2_1_2()
34559   const std::int32_t chunk_size = k * 3;  in gemm_f_2_1_2()
34560   const std::int32_t zipped_chunk_size = (padded_k + 16) * 3;  in gemm_f_2_1_2()
34561   const std::int32_t zipped_rhs_size = (padded_k + 16) * n;  in gemm_f_2_1_2()
34565   std::int32_t* zipped_lhs_3_offsets =  in gemm_f_2_1_2()
34566       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 3);  in gemm_f_2_1_2()
34567   std::int32_t* zipped_lhs_2_offsets =  in gemm_f_2_1_2()
34568       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 2);  in gemm_f_2_1_2()
34571   const std::int32_t result_chunk_stride = result_stride * 3;  in gemm_f_2_1_2()
34573   const std::int32_t const_offset = lhs_offset * rhs_offset * k;  in gemm_f_2_1_2()
34576   const std::int32_t mul_result_chunk_stride_bytes = result_stride * 4;  in gemm_f_2_1_2()
34619                   const std::uint8_t* rhs, std::int32_t m, std::int32_t n,  in gemm_f_2_1_3()
34620                   std::int32_t k, std::int32_t lhs_offset,  in gemm_f_2_1_3()
34621                   std::int32_t rhs_offset, float result_scale, float* result,  in gemm_f_2_1_3()
34622                   std::int32_t result_stride) {  in gemm_f_2_1_3()
34623   const std::int32_t row_chunks = m / 3;  in gemm_f_2_1_3()
34624   const std::int32_t col_chunks = n / 3;  in gemm_f_2_1_3()
34625   const std::int32_t padded_k = ((k + 7) / 8) * 8;  in gemm_f_2_1_3()
34626   const std::int32_t chunk_size = k * 3;  in gemm_f_2_1_3()
34627   const std::int32_t zipped_chunk_size = (padded_k + 16) * 3;  in gemm_f_2_1_3()
34628   const std::int32_t zipped_rhs_size = (padded_k + 16) * n;  in gemm_f_2_1_3()
34632   std::int32_t* zipped_lhs_3_offsets =  in gemm_f_2_1_3()
34633       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 3);  in gemm_f_2_1_3()
34634   std::int32_t* zipped_lhs_2_offsets =  in gemm_f_2_1_3()
34635       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 2);  in gemm_f_2_1_3()
34638   const std::int32_t result_chunk_stride = result_stride * 3;  in gemm_f_2_1_3()
34640   const std::int32_t const_offset = lhs_offset * rhs_offset * k;  in gemm_f_2_1_3()
34643   const std::int32_t mul_result_chunk_stride_bytes = result_stride * 4;  in gemm_f_2_1_3()
34686                   const std::uint8_t* rhs, std::int32_t m, std::int32_t n,  in gemm_f_2_1_4()
34687                   std::int32_t k, std::int32_t lhs_offset,  in gemm_f_2_1_4()
34688                   std::int32_t rhs_offset, float result_scale, float* result,  in gemm_f_2_1_4()
34689                   std::int32_t result_stride) {  in gemm_f_2_1_4()
34690   const std::int32_t row_chunks = m / 3;  in gemm_f_2_1_4()
34691   const std::int32_t col_chunks = n / 3;  in gemm_f_2_1_4()
34692   const std::int32_t padded_k = ((k + 7) / 8) * 8;  in gemm_f_2_1_4()
34693   const std::int32_t chunk_size = k * 3;  in gemm_f_2_1_4()
34694   const std::int32_t zipped_chunk_size = (padded_k + 16) * 3;  in gemm_f_2_1_4()
34695   const std::int32_t zipped_rhs_size = (padded_k + 16) * n;  in gemm_f_2_1_4()
34699   std::int32_t* zipped_lhs_3_offsets =  in gemm_f_2_1_4()
34700       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 3);  in gemm_f_2_1_4()
34701   std::int32_t* zipped_lhs_2_offsets =  in gemm_f_2_1_4()
34702       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 2);  in gemm_f_2_1_4()
34705   const std::int32_t result_chunk_stride = result_stride * 3;  in gemm_f_2_1_4()
34707   const std::int32_t const_offset = lhs_offset * rhs_offset * k;  in gemm_f_2_1_4()
34710   const std::int32_t mul_result_chunk_stride_bytes = result_stride * 4;  in gemm_f_2_1_4()
34753                   const std::uint8_t* rhs, std::int32_t m, std::int32_t n,  in gemm_f_2_1_5()
34754                   std::int32_t k, std::int32_t lhs_offset,  in gemm_f_2_1_5()
34755                   std::int32_t rhs_offset, float result_scale, float* result,  in gemm_f_2_1_5()
34756                   std::int32_t result_stride) {  in gemm_f_2_1_5()
34757   const std::int32_t row_chunks = m / 3;  in gemm_f_2_1_5()
34758   const std::int32_t col_chunks = n / 3;  in gemm_f_2_1_5()
34759   const std::int32_t padded_k = ((k + 7) / 8) * 8;  in gemm_f_2_1_5()
34760   const std::int32_t chunk_size = k * 3;  in gemm_f_2_1_5()
34761   const std::int32_t zipped_chunk_size = (padded_k + 16) * 3;  in gemm_f_2_1_5()
34762   const std::int32_t zipped_rhs_size = (padded_k + 16) * n;  in gemm_f_2_1_5()
34766   std::int32_t* zipped_lhs_3_offsets =  in gemm_f_2_1_5()
34767       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 3);  in gemm_f_2_1_5()
34768   std::int32_t* zipped_lhs_2_offsets =  in gemm_f_2_1_5()
34769       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 2);  in gemm_f_2_1_5()
34772   const std::int32_t result_chunk_stride = result_stride * 3;  in gemm_f_2_1_5()
34774   const std::int32_t const_offset = lhs_offset * rhs_offset * k;  in gemm_f_2_1_5()
34777   const std::int32_t mul_result_chunk_stride_bytes = result_stride * 4;  in gemm_f_2_1_5()
34820                   const std::uint8_t* rhs, std::int32_t m, std::int32_t n,  in gemm_f_2_1_6()
34821                   std::int32_t k, std::int32_t lhs_offset,  in gemm_f_2_1_6()
34822                   std::int32_t rhs_offset, float result_scale, float* result,  in gemm_f_2_1_6()
34823                   std::int32_t result_stride) {  in gemm_f_2_1_6()
34824   const std::int32_t row_chunks = m / 3;  in gemm_f_2_1_6()
34825   const std::int32_t col_chunks = n / 3;  in gemm_f_2_1_6()
34826   const std::int32_t padded_k = ((k + 7) / 8) * 8;  in gemm_f_2_1_6()
34827   const std::int32_t chunk_size = k * 3;  in gemm_f_2_1_6()
34828   const std::int32_t zipped_chunk_size = (padded_k + 16) * 3;  in gemm_f_2_1_6()
34829   const std::int32_t zipped_rhs_size = (padded_k + 16) * n;  in gemm_f_2_1_6()
34833   std::int32_t* zipped_lhs_3_offsets =  in gemm_f_2_1_6()
34834       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 3);  in gemm_f_2_1_6()
34835   std::int32_t* zipped_lhs_2_offsets =  in gemm_f_2_1_6()
34836       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 2);  in gemm_f_2_1_6()
34839   const std::int32_t result_chunk_stride = result_stride * 3;  in gemm_f_2_1_6()
34841   const std::int32_t const_offset = lhs_offset * rhs_offset * k;  in gemm_f_2_1_6()
34844   const std::int32_t mul_result_chunk_stride_bytes = result_stride * 4;  in gemm_f_2_1_6()
34887                   const std::uint8_t* rhs, std::int32_t m, std::int32_t n,  in gemm_f_2_1_7()
34888                   std::int32_t k, std::int32_t lhs_offset,  in gemm_f_2_1_7()
34889                   std::int32_t rhs_offset, float result_scale, float* result,  in gemm_f_2_1_7()
34890                   std::int32_t result_stride) {  in gemm_f_2_1_7()
34891   const std::int32_t row_chunks = m / 3;  in gemm_f_2_1_7()
34892   const std::int32_t col_chunks = n / 3;  in gemm_f_2_1_7()
34893   const std::int32_t padded_k = ((k + 7) / 8) * 8;  in gemm_f_2_1_7()
34894   const std::int32_t chunk_size = k * 3;  in gemm_f_2_1_7()
34895   const std::int32_t zipped_chunk_size = (padded_k + 16) * 3;  in gemm_f_2_1_7()
34896   const std::int32_t zipped_rhs_size = (padded_k + 16) * n;  in gemm_f_2_1_7()
34900   std::int32_t* zipped_lhs_3_offsets =  in gemm_f_2_1_7()
34901       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 3);  in gemm_f_2_1_7()
34902   std::int32_t* zipped_lhs_2_offsets =  in gemm_f_2_1_7()
34903       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 2);  in gemm_f_2_1_7()
34906   const std::int32_t result_chunk_stride = result_stride * 3;  in gemm_f_2_1_7()
34908   const std::int32_t const_offset = lhs_offset * rhs_offset * k;  in gemm_f_2_1_7()
34911   const std::int32_t mul_result_chunk_stride_bytes = result_stride * 4;  in gemm_f_2_1_7()
34954                   const std::uint8_t* rhs, std::int32_t m, std::int32_t n,  in gemm_f_2_2_0()
34955                   std::int32_t k, std::int32_t lhs_offset,  in gemm_f_2_2_0()
34956                   std::int32_t rhs_offset, float result_scale, float* result,  in gemm_f_2_2_0()
34957                   std::int32_t result_stride) {  in gemm_f_2_2_0()
34958   const std::int32_t row_chunks = m / 3;  in gemm_f_2_2_0()
34959   const std::int32_t col_chunks = n / 3;  in gemm_f_2_2_0()
34960   const std::int32_t padded_k = ((k + 7) / 8) * 8;  in gemm_f_2_2_0()
34961   const std::int32_t chunk_size = k * 3;  in gemm_f_2_2_0()
34962   const std::int32_t zipped_chunk_size = (padded_k + 16) * 3;  in gemm_f_2_2_0()
34963   const std::int32_t zipped_rhs_size = (padded_k + 16) * n;  in gemm_f_2_2_0()
34967   std::int32_t* zipped_lhs_3_offsets =  in gemm_f_2_2_0()
34968       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 3);  in gemm_f_2_2_0()
34969   std::int32_t* zipped_lhs_2_offsets =  in gemm_f_2_2_0()
34970       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 2);  in gemm_f_2_2_0()
34973   const std::int32_t result_chunk_stride = result_stride * 3;  in gemm_f_2_2_0()
34975   const std::int32_t const_offset = lhs_offset * rhs_offset * k;  in gemm_f_2_2_0()
34978   const std::int32_t mul_result_chunk_stride_bytes = result_stride * 4;  in gemm_f_2_2_0()
35021                   const std::uint8_t* rhs, std::int32_t m, std::int32_t n,  in gemm_f_2_2_1()
35022                   std::int32_t k, std::int32_t lhs_offset,  in gemm_f_2_2_1()
35023                   std::int32_t rhs_offset, float result_scale, float* result,  in gemm_f_2_2_1()
35024                   std::int32_t result_stride) {  in gemm_f_2_2_1()
35025   const std::int32_t row_chunks = m / 3;  in gemm_f_2_2_1()
35026   const std::int32_t col_chunks = n / 3;  in gemm_f_2_2_1()
35027   const std::int32_t padded_k = ((k + 7) / 8) * 8;  in gemm_f_2_2_1()
35028   const std::int32_t chunk_size = k * 3;  in gemm_f_2_2_1()
35029   const std::int32_t zipped_chunk_size = (padded_k + 16) * 3;  in gemm_f_2_2_1()
35030   const std::int32_t zipped_rhs_size = (padded_k + 16) * n;  in gemm_f_2_2_1()
35034   std::int32_t* zipped_lhs_3_offsets =  in gemm_f_2_2_1()
35035       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 3);  in gemm_f_2_2_1()
35036   std::int32_t* zipped_lhs_2_offsets =  in gemm_f_2_2_1()
35037       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 2);  in gemm_f_2_2_1()
35040   const std::int32_t result_chunk_stride = result_stride * 3;  in gemm_f_2_2_1()
35042   const std::int32_t const_offset = lhs_offset * rhs_offset * k;  in gemm_f_2_2_1()
35045   const std::int32_t mul_result_chunk_stride_bytes = result_stride * 4;  in gemm_f_2_2_1()
35088                   const std::uint8_t* rhs, std::int32_t m, std::int32_t n,  in gemm_f_2_2_2()
35089                   std::int32_t k, std::int32_t lhs_offset,  in gemm_f_2_2_2()
35090                   std::int32_t rhs_offset, float result_scale, float* result,  in gemm_f_2_2_2()
35091                   std::int32_t result_stride) {  in gemm_f_2_2_2()
35092   const std::int32_t row_chunks = m / 3;  in gemm_f_2_2_2()
35093   const std::int32_t col_chunks = n / 3;  in gemm_f_2_2_2()
35094   const std::int32_t padded_k = ((k + 7) / 8) * 8;  in gemm_f_2_2_2()
35095   const std::int32_t chunk_size = k * 3;  in gemm_f_2_2_2()
35096   const std::int32_t zipped_chunk_size = (padded_k + 16) * 3;  in gemm_f_2_2_2()
35097   const std::int32_t zipped_rhs_size = (padded_k + 16) * n;  in gemm_f_2_2_2()
35101   std::int32_t* zipped_lhs_3_offsets =  in gemm_f_2_2_2()
35102       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 3);  in gemm_f_2_2_2()
35103   std::int32_t* zipped_lhs_2_offsets =  in gemm_f_2_2_2()
35104       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 2);  in gemm_f_2_2_2()
35107   const std::int32_t result_chunk_stride = result_stride * 3;  in gemm_f_2_2_2()
35109   const std::int32_t const_offset = lhs_offset * rhs_offset * k;  in gemm_f_2_2_2()
35112   const std::int32_t mul_result_chunk_stride_bytes = result_stride * 4;  in gemm_f_2_2_2()
35155                   const std::uint8_t* rhs, std::int32_t m, std::int32_t n,  in gemm_f_2_2_3()
35156                   std::int32_t k, std::int32_t lhs_offset,  in gemm_f_2_2_3()
35157                   std::int32_t rhs_offset, float result_scale, float* result,  in gemm_f_2_2_3()
35158                   std::int32_t result_stride) {  in gemm_f_2_2_3()
35159   const std::int32_t row_chunks = m / 3;  in gemm_f_2_2_3()
35160   const std::int32_t col_chunks = n / 3;  in gemm_f_2_2_3()
35161   const std::int32_t padded_k = ((k + 7) / 8) * 8;  in gemm_f_2_2_3()
35162   const std::int32_t chunk_size = k * 3;  in gemm_f_2_2_3()
35163   const std::int32_t zipped_chunk_size = (padded_k + 16) * 3;  in gemm_f_2_2_3()
35164   const std::int32_t zipped_rhs_size = (padded_k + 16) * n;  in gemm_f_2_2_3()
35168   std::int32_t* zipped_lhs_3_offsets =  in gemm_f_2_2_3()
35169       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 3);  in gemm_f_2_2_3()
35170   std::int32_t* zipped_lhs_2_offsets =  in gemm_f_2_2_3()
35171       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 2);  in gemm_f_2_2_3()
35174   const std::int32_t result_chunk_stride = result_stride * 3;  in gemm_f_2_2_3()
35176   const std::int32_t const_offset = lhs_offset * rhs_offset * k;  in gemm_f_2_2_3()
35179   const std::int32_t mul_result_chunk_stride_bytes = result_stride * 4;  in gemm_f_2_2_3()
35222                   const std::uint8_t* rhs, std::int32_t m, std::int32_t n,  in gemm_f_2_2_4()
35223                   std::int32_t k, std::int32_t lhs_offset,  in gemm_f_2_2_4()
35224                   std::int32_t rhs_offset, float result_scale, float* result,  in gemm_f_2_2_4()
35225                   std::int32_t result_stride) {  in gemm_f_2_2_4()
35226   const std::int32_t row_chunks = m / 3;  in gemm_f_2_2_4()
35227   const std::int32_t col_chunks = n / 3;  in gemm_f_2_2_4()
35228   const std::int32_t padded_k = ((k + 7) / 8) * 8;  in gemm_f_2_2_4()
35229   const std::int32_t chunk_size = k * 3;  in gemm_f_2_2_4()
35230   const std::int32_t zipped_chunk_size = (padded_k + 16) * 3;  in gemm_f_2_2_4()
35231   const std::int32_t zipped_rhs_size = (padded_k + 16) * n;  in gemm_f_2_2_4()
35235   std::int32_t* zipped_lhs_3_offsets =  in gemm_f_2_2_4()
35236       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 3);  in gemm_f_2_2_4()
35237   std::int32_t* zipped_lhs_2_offsets =  in gemm_f_2_2_4()
35238       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 2);  in gemm_f_2_2_4()
35241   const std::int32_t result_chunk_stride = result_stride * 3;  in gemm_f_2_2_4()
35243   const std::int32_t const_offset = lhs_offset * rhs_offset * k;  in gemm_f_2_2_4()
35246   const std::int32_t mul_result_chunk_stride_bytes = result_stride * 4;  in gemm_f_2_2_4()
35289                   const std::uint8_t* rhs, std::int32_t m, std::int32_t n,  in gemm_f_2_2_5()
35290                   std::int32_t k, std::int32_t lhs_offset,  in gemm_f_2_2_5()
35291                   std::int32_t rhs_offset, float result_scale, float* result,  in gemm_f_2_2_5()
35292                   std::int32_t result_stride) {  in gemm_f_2_2_5()
35293   const std::int32_t row_chunks = m / 3;  in gemm_f_2_2_5()
35294   const std::int32_t col_chunks = n / 3;  in gemm_f_2_2_5()
35295   const std::int32_t padded_k = ((k + 7) / 8) * 8;  in gemm_f_2_2_5()
35296   const std::int32_t chunk_size = k * 3;  in gemm_f_2_2_5()
35297   const std::int32_t zipped_chunk_size = (padded_k + 16) * 3;  in gemm_f_2_2_5()
35298   const std::int32_t zipped_rhs_size = (padded_k + 16) * n;  in gemm_f_2_2_5()
35302   std::int32_t* zipped_lhs_3_offsets =  in gemm_f_2_2_5()
35303       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 3);  in gemm_f_2_2_5()
35304   std::int32_t* zipped_lhs_2_offsets =  in gemm_f_2_2_5()
35305       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 2);  in gemm_f_2_2_5()
35308   const std::int32_t result_chunk_stride = result_stride * 3;  in gemm_f_2_2_5()
35310   const std::int32_t const_offset = lhs_offset * rhs_offset * k;  in gemm_f_2_2_5()
35313   const std::int32_t mul_result_chunk_stride_bytes = result_stride * 4;  in gemm_f_2_2_5()
35356                   const std::uint8_t* rhs, std::int32_t m, std::int32_t n,  in gemm_f_2_2_6()
35357                   std::int32_t k, std::int32_t lhs_offset,  in gemm_f_2_2_6()
35358                   std::int32_t rhs_offset, float result_scale, float* result,  in gemm_f_2_2_6()
35359                   std::int32_t result_stride) {  in gemm_f_2_2_6()
35360   const std::int32_t row_chunks = m / 3;  in gemm_f_2_2_6()
35361   const std::int32_t col_chunks = n / 3;  in gemm_f_2_2_6()
35362   const std::int32_t padded_k = ((k + 7) / 8) * 8;  in gemm_f_2_2_6()
35363   const std::int32_t chunk_size = k * 3;  in gemm_f_2_2_6()
35364   const std::int32_t zipped_chunk_size = (padded_k + 16) * 3;  in gemm_f_2_2_6()
35365   const std::int32_t zipped_rhs_size = (padded_k + 16) * n;  in gemm_f_2_2_6()
35369   std::int32_t* zipped_lhs_3_offsets =  in gemm_f_2_2_6()
35370       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 3);  in gemm_f_2_2_6()
35371   std::int32_t* zipped_lhs_2_offsets =  in gemm_f_2_2_6()
35372       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 2);  in gemm_f_2_2_6()
35375   const std::int32_t result_chunk_stride = result_stride * 3;  in gemm_f_2_2_6()
35377   const std::int32_t const_offset = lhs_offset * rhs_offset * k;  in gemm_f_2_2_6()
35380   const std::int32_t mul_result_chunk_stride_bytes = result_stride * 4;  in gemm_f_2_2_6()
35423                   const std::uint8_t* rhs, std::int32_t m, std::int32_t n,  in gemm_f_2_2_7()
35424                   std::int32_t k, std::int32_t lhs_offset,  in gemm_f_2_2_7()
35425                   std::int32_t rhs_offset, float result_scale, float* result,  in gemm_f_2_2_7()
35426                   std::int32_t result_stride) {  in gemm_f_2_2_7()
35427   const std::int32_t row_chunks = m / 3;  in gemm_f_2_2_7()
35428   const std::int32_t col_chunks = n / 3;  in gemm_f_2_2_7()
35429   const std::int32_t padded_k = ((k + 7) / 8) * 8;  in gemm_f_2_2_7()
35430   const std::int32_t chunk_size = k * 3;  in gemm_f_2_2_7()
35431   const std::int32_t zipped_chunk_size = (padded_k + 16) * 3;  in gemm_f_2_2_7()
35432   const std::int32_t zipped_rhs_size = (padded_k + 16) * n;  in gemm_f_2_2_7()
35436   std::int32_t* zipped_lhs_3_offsets =  in gemm_f_2_2_7()
35437       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 3);  in gemm_f_2_2_7()
35438   std::int32_t* zipped_lhs_2_offsets =  in gemm_f_2_2_7()
35439       reinterpret_cast<std::int32_t*>(zipped_lhs + padded_k * 2);  in gemm_f_2_2_7()
35442   const std::int32_t result_chunk_stride = result_stride * 3;  in gemm_f_2_2_7()
35444   const std::int32_t const_offset = lhs_offset * rhs_offset * k;  in gemm_f_2_2_7()
35447   const std::int32_t mul_result_chunk_stride_bytes = result_stride * 4;  in gemm_f_2_2_7()
35492                      const std::uint8_t* rhs, std::int32_t m, std::int32_t n,  in gemm_q8_strided()
35493                      std::int32_t k, std::int32_t lhs_offset,  in gemm_q8_strided()
35494                      std::int32_t rhs_offset, std::int32_t result_offset,  in gemm_q8_strided()
35495                      std::int32_t multiplicative_offset, std::int32_t shift,  in gemm_q8_strided()
35496                      std::uint8_t* result, std::int32_t result_stride) {  in gemm_q8_strided()
36475                       const std::uint8_t* rhs, std::int32_t m, std::int32_t n,  in gemm_i32_strided()
36476                       std::int32_t k, std::int32_t lhs_offset,  in gemm_i32_strided()
36477                       std::int32_t rhs_offset, std::int32_t* result,  in gemm_i32_strided()
36478                       std::int32_t result_stride) {  in gemm_i32_strided()
37237                     const std::uint8_t* rhs, std::int32_t m, std::int32_t n,  in gemm_f_strided()
37238                     std::int32_t k, std::int32_t lhs_offset,  in gemm_f_strided()
37239                     std::int32_t rhs_offset, float result_scale, float* result,  in gemm_f_strided()
37240                     std::int32_t result_stride) {  in gemm_f_strided()
38071              const std::uint8_t* rhs, std::int32_t m, std::int32_t n,  in gemm_q8()
38072              std::int32_t k, std::int32_t lhs_offset, std::int32_t rhs_offset,  in gemm_q8()
38073              std::int32_t result_offset, std::int32_t multiplicative_offset,  in gemm_q8()
38074              std::int32_t shift, std::uint8_t* result) {  in gemm_q8()
38080               const std::uint8_t* rhs, std::int32_t m, std::int32_t n,  in gemm_i32()
38081               std::int32_t k, std::int32_t lhs_offset, std::int32_t rhs_offset,  in gemm_i32()
38082               std::int32_t* result) {  in gemm_i32()
38088             const std::uint8_t* rhs, std::int32_t m, std::int32_t n,  in gemm_f()
38089             std::int32_t k, std::int32_t lhs_offset, std::int32_t rhs_offset,  in gemm_f()