1 /*
2 * Copyright (c) 2015 The WebM project authors. All Rights Reserved.
3 *
4 * Use of this source code is governed by a BSD-style license
5 * that can be found in the LICENSE file in the root of the source
6 * tree. An additional intellectual property rights grant can be found
7 * in the file PATENTS. All contributing project authors may
8 * be found in the AUTHORS file in the root of the source tree.
9 */
10
11 #include "./vpx_dsp_rtcd.h"
12 #include "vpx_dsp/mips/inv_txfm_msa.h"
13
vpx_idct16_1d_rows_msa(const int16_t * input,int16_t * output)14 void vpx_idct16_1d_rows_msa(const int16_t *input, int16_t *output) {
15 v8i16 loc0, loc1, loc2, loc3;
16 v8i16 reg0, reg2, reg4, reg6, reg8, reg10, reg12, reg14;
17 v8i16 reg3, reg13, reg11, reg5, reg7, reg9, reg1, reg15;
18 v8i16 tmp5, tmp6, tmp7;
19
20 LD_SH8(input, 16, reg0, reg1, reg2, reg3, reg4, reg5, reg6, reg7);
21 input += 8;
22 LD_SH8(input, 16, reg8, reg9, reg10, reg11, reg12, reg13, reg14, reg15);
23
24 TRANSPOSE8x8_SH_SH(reg0, reg1, reg2, reg3, reg4, reg5, reg6, reg7, reg0, reg1,
25 reg2, reg3, reg4, reg5, reg6, reg7);
26 TRANSPOSE8x8_SH_SH(reg8, reg9, reg10, reg11, reg12, reg13, reg14, reg15, reg8,
27 reg9, reg10, reg11, reg12, reg13, reg14, reg15);
28 DOTP_CONST_PAIR(reg2, reg14, cospi_28_64, cospi_4_64, reg2, reg14);
29 DOTP_CONST_PAIR(reg10, reg6, cospi_12_64, cospi_20_64, reg10, reg6);
30 BUTTERFLY_4(reg2, reg14, reg6, reg10, loc0, loc1, reg14, reg2);
31 DOTP_CONST_PAIR(reg14, reg2, cospi_16_64, cospi_16_64, loc2, loc3);
32 DOTP_CONST_PAIR(reg0, reg8, cospi_16_64, cospi_16_64, reg0, reg8);
33 DOTP_CONST_PAIR(reg4, reg12, cospi_24_64, cospi_8_64, reg4, reg12);
34 BUTTERFLY_4(reg8, reg0, reg4, reg12, reg2, reg6, reg10, reg14);
35 SUB4(reg2, loc1, reg14, loc0, reg6, loc3, reg10, loc2, reg0, reg12, reg4,
36 reg8);
37 ADD4(reg2, loc1, reg14, loc0, reg6, loc3, reg10, loc2, reg2, reg14, reg6,
38 reg10);
39
40 /* stage 2 */
41 DOTP_CONST_PAIR(reg1, reg15, cospi_30_64, cospi_2_64, reg1, reg15);
42 DOTP_CONST_PAIR(reg9, reg7, cospi_14_64, cospi_18_64, loc2, loc3);
43
44 reg9 = reg1 - loc2;
45 reg1 = reg1 + loc2;
46 reg7 = reg15 - loc3;
47 reg15 = reg15 + loc3;
48
49 DOTP_CONST_PAIR(reg5, reg11, cospi_22_64, cospi_10_64, reg5, reg11);
50 DOTP_CONST_PAIR(reg13, reg3, cospi_6_64, cospi_26_64, loc0, loc1);
51 BUTTERFLY_4(loc0, loc1, reg11, reg5, reg13, reg3, reg11, reg5);
52
53 loc1 = reg15 + reg3;
54 reg3 = reg15 - reg3;
55 loc2 = reg2 + loc1;
56 reg15 = reg2 - loc1;
57
58 loc1 = reg1 + reg13;
59 reg13 = reg1 - reg13;
60 loc0 = reg0 + loc1;
61 loc1 = reg0 - loc1;
62 tmp6 = loc0;
63 tmp7 = loc1;
64 reg0 = loc2;
65
66 DOTP_CONST_PAIR(reg7, reg9, cospi_24_64, cospi_8_64, reg7, reg9);
67 DOTP_CONST_PAIR((-reg5), (-reg11), cospi_8_64, cospi_24_64, reg5, reg11);
68
69 loc0 = reg9 + reg5;
70 reg5 = reg9 - reg5;
71 reg2 = reg6 + loc0;
72 reg1 = reg6 - loc0;
73
74 loc0 = reg7 + reg11;
75 reg11 = reg7 - reg11;
76 loc1 = reg4 + loc0;
77 loc2 = reg4 - loc0;
78 tmp5 = loc1;
79
80 DOTP_CONST_PAIR(reg5, reg11, cospi_16_64, cospi_16_64, reg5, reg11);
81 BUTTERFLY_4(reg8, reg10, reg11, reg5, loc0, reg4, reg9, loc1);
82
83 reg10 = loc0;
84 reg11 = loc1;
85
86 DOTP_CONST_PAIR(reg3, reg13, cospi_16_64, cospi_16_64, reg3, reg13);
87 BUTTERFLY_4(reg12, reg14, reg13, reg3, reg8, reg6, reg7, reg5);
88
89 reg13 = loc2;
90
91 /* Transpose and store the output */
92 reg12 = tmp5;
93 reg14 = tmp6;
94 reg3 = tmp7;
95
96 /* transpose block */
97 TRANSPOSE8x8_SH_SH(reg0, reg2, reg4, reg6, reg8, reg10, reg12, reg14, reg0,
98 reg2, reg4, reg6, reg8, reg10, reg12, reg14);
99 ST_SH8(reg0, reg2, reg4, reg6, reg8, reg10, reg12, reg14, output, 16);
100
101 /* transpose block */
102 TRANSPOSE8x8_SH_SH(reg3, reg13, reg11, reg5, reg7, reg9, reg1, reg15, reg3,
103 reg13, reg11, reg5, reg7, reg9, reg1, reg15);
104 ST_SH8(reg3, reg13, reg11, reg5, reg7, reg9, reg1, reg15, (output + 8), 16);
105 }
106
vpx_idct16_1d_columns_addblk_msa(int16_t * input,uint8_t * dst,int32_t dst_stride)107 void vpx_idct16_1d_columns_addblk_msa(int16_t *input, uint8_t *dst,
108 int32_t dst_stride) {
109 v8i16 loc0, loc1, loc2, loc3;
110 v8i16 reg0, reg2, reg4, reg6, reg8, reg10, reg12, reg14;
111 v8i16 reg3, reg13, reg11, reg5, reg7, reg9, reg1, reg15;
112 v8i16 tmp5, tmp6, tmp7;
113
114 /* load up 8x8 */
115 LD_SH8(input, 16, reg0, reg1, reg2, reg3, reg4, reg5, reg6, reg7);
116 input += 8 * 16;
117 /* load bottom 8x8 */
118 LD_SH8(input, 16, reg8, reg9, reg10, reg11, reg12, reg13, reg14, reg15);
119
120 DOTP_CONST_PAIR(reg2, reg14, cospi_28_64, cospi_4_64, reg2, reg14);
121 DOTP_CONST_PAIR(reg10, reg6, cospi_12_64, cospi_20_64, reg10, reg6);
122 BUTTERFLY_4(reg2, reg14, reg6, reg10, loc0, loc1, reg14, reg2);
123 DOTP_CONST_PAIR(reg14, reg2, cospi_16_64, cospi_16_64, loc2, loc3);
124 DOTP_CONST_PAIR(reg0, reg8, cospi_16_64, cospi_16_64, reg0, reg8);
125 DOTP_CONST_PAIR(reg4, reg12, cospi_24_64, cospi_8_64, reg4, reg12);
126 BUTTERFLY_4(reg8, reg0, reg4, reg12, reg2, reg6, reg10, reg14);
127
128 reg0 = reg2 - loc1;
129 reg2 = reg2 + loc1;
130 reg12 = reg14 - loc0;
131 reg14 = reg14 + loc0;
132 reg4 = reg6 - loc3;
133 reg6 = reg6 + loc3;
134 reg8 = reg10 - loc2;
135 reg10 = reg10 + loc2;
136
137 /* stage 2 */
138 DOTP_CONST_PAIR(reg1, reg15, cospi_30_64, cospi_2_64, reg1, reg15);
139 DOTP_CONST_PAIR(reg9, reg7, cospi_14_64, cospi_18_64, loc2, loc3);
140
141 reg9 = reg1 - loc2;
142 reg1 = reg1 + loc2;
143 reg7 = reg15 - loc3;
144 reg15 = reg15 + loc3;
145
146 DOTP_CONST_PAIR(reg5, reg11, cospi_22_64, cospi_10_64, reg5, reg11);
147 DOTP_CONST_PAIR(reg13, reg3, cospi_6_64, cospi_26_64, loc0, loc1);
148 BUTTERFLY_4(loc0, loc1, reg11, reg5, reg13, reg3, reg11, reg5);
149
150 loc1 = reg15 + reg3;
151 reg3 = reg15 - reg3;
152 loc2 = reg2 + loc1;
153 reg15 = reg2 - loc1;
154
155 loc1 = reg1 + reg13;
156 reg13 = reg1 - reg13;
157 loc0 = reg0 + loc1;
158 loc1 = reg0 - loc1;
159 tmp6 = loc0;
160 tmp7 = loc1;
161 reg0 = loc2;
162
163 DOTP_CONST_PAIR(reg7, reg9, cospi_24_64, cospi_8_64, reg7, reg9);
164 DOTP_CONST_PAIR((-reg5), (-reg11), cospi_8_64, cospi_24_64, reg5, reg11);
165
166 loc0 = reg9 + reg5;
167 reg5 = reg9 - reg5;
168 reg2 = reg6 + loc0;
169 reg1 = reg6 - loc0;
170
171 loc0 = reg7 + reg11;
172 reg11 = reg7 - reg11;
173 loc1 = reg4 + loc0;
174 loc2 = reg4 - loc0;
175 tmp5 = loc1;
176
177 DOTP_CONST_PAIR(reg5, reg11, cospi_16_64, cospi_16_64, reg5, reg11);
178 BUTTERFLY_4(reg8, reg10, reg11, reg5, loc0, reg4, reg9, loc1);
179
180 reg10 = loc0;
181 reg11 = loc1;
182
183 DOTP_CONST_PAIR(reg3, reg13, cospi_16_64, cospi_16_64, reg3, reg13);
184 BUTTERFLY_4(reg12, reg14, reg13, reg3, reg8, reg6, reg7, reg5);
185 reg13 = loc2;
186
187 /* Transpose and store the output */
188 reg12 = tmp5;
189 reg14 = tmp6;
190 reg3 = tmp7;
191
192 SRARI_H4_SH(reg0, reg2, reg4, reg6, 6);
193 VP9_ADDBLK_ST8x4_UB(dst, dst_stride, reg0, reg2, reg4, reg6);
194 dst += (4 * dst_stride);
195 SRARI_H4_SH(reg8, reg10, reg12, reg14, 6);
196 VP9_ADDBLK_ST8x4_UB(dst, dst_stride, reg8, reg10, reg12, reg14);
197 dst += (4 * dst_stride);
198 SRARI_H4_SH(reg3, reg13, reg11, reg5, 6);
199 VP9_ADDBLK_ST8x4_UB(dst, dst_stride, reg3, reg13, reg11, reg5);
200 dst += (4 * dst_stride);
201 SRARI_H4_SH(reg7, reg9, reg1, reg15, 6);
202 VP9_ADDBLK_ST8x4_UB(dst, dst_stride, reg7, reg9, reg1, reg15);
203 }
204
vpx_idct16x16_256_add_msa(const int16_t * input,uint8_t * dst,int32_t dst_stride)205 void vpx_idct16x16_256_add_msa(const int16_t *input, uint8_t *dst,
206 int32_t dst_stride) {
207 int32_t i;
208 DECLARE_ALIGNED(32, int16_t, out_arr[16 * 16]);
209 int16_t *out = out_arr;
210
211 /* transform rows */
212 for (i = 0; i < 2; ++i) {
213 /* process 16 * 8 block */
214 vpx_idct16_1d_rows_msa((input + (i << 7)), (out + (i << 7)));
215 }
216
217 /* transform columns */
218 for (i = 0; i < 2; ++i) {
219 /* process 8 * 16 block */
220 vpx_idct16_1d_columns_addblk_msa((out + (i << 3)), (dst + (i << 3)),
221 dst_stride);
222 }
223 }
224
vpx_idct16x16_10_add_msa(const int16_t * input,uint8_t * dst,int32_t dst_stride)225 void vpx_idct16x16_10_add_msa(const int16_t *input, uint8_t *dst,
226 int32_t dst_stride) {
227 uint8_t i;
228 DECLARE_ALIGNED(32, int16_t, out_arr[16 * 16]);
229 int16_t *out = out_arr;
230
231 /* process 16 * 8 block */
232 vpx_idct16_1d_rows_msa(input, out);
233
234 /* short case just considers top 4 rows as valid output */
235 out += 4 * 16;
236 for (i = 12; i--;) {
237 __asm__ __volatile__(
238 "sw $zero, 0(%[out]) \n\t"
239 "sw $zero, 4(%[out]) \n\t"
240 "sw $zero, 8(%[out]) \n\t"
241 "sw $zero, 12(%[out]) \n\t"
242 "sw $zero, 16(%[out]) \n\t"
243 "sw $zero, 20(%[out]) \n\t"
244 "sw $zero, 24(%[out]) \n\t"
245 "sw $zero, 28(%[out]) \n\t"
246
247 :
248 : [out] "r"(out));
249
250 out += 16;
251 }
252
253 out = out_arr;
254
255 /* transform columns */
256 for (i = 0; i < 2; ++i) {
257 /* process 8 * 16 block */
258 vpx_idct16_1d_columns_addblk_msa((out + (i << 3)), (dst + (i << 3)),
259 dst_stride);
260 }
261 }
262
vpx_idct16x16_1_add_msa(const int16_t * input,uint8_t * dst,int32_t dst_stride)263 void vpx_idct16x16_1_add_msa(const int16_t *input, uint8_t *dst,
264 int32_t dst_stride) {
265 uint8_t i;
266 int16_t out;
267 v8i16 vec, res0, res1, res2, res3, res4, res5, res6, res7;
268 v16u8 dst0, dst1, dst2, dst3, tmp0, tmp1, tmp2, tmp3;
269
270 out = ROUND_POWER_OF_TWO((input[0] * cospi_16_64), DCT_CONST_BITS);
271 out = ROUND_POWER_OF_TWO((out * cospi_16_64), DCT_CONST_BITS);
272 out = ROUND_POWER_OF_TWO(out, 6);
273
274 vec = __msa_fill_h(out);
275
276 for (i = 4; i--;) {
277 LD_UB4(dst, dst_stride, dst0, dst1, dst2, dst3);
278 UNPCK_UB_SH(dst0, res0, res4);
279 UNPCK_UB_SH(dst1, res1, res5);
280 UNPCK_UB_SH(dst2, res2, res6);
281 UNPCK_UB_SH(dst3, res3, res7);
282 ADD4(res0, vec, res1, vec, res2, vec, res3, vec, res0, res1, res2, res3);
283 ADD4(res4, vec, res5, vec, res6, vec, res7, vec, res4, res5, res6, res7);
284 CLIP_SH4_0_255(res0, res1, res2, res3);
285 CLIP_SH4_0_255(res4, res5, res6, res7);
286 PCKEV_B4_UB(res4, res0, res5, res1, res6, res2, res7, res3, tmp0, tmp1,
287 tmp2, tmp3);
288 ST_UB4(tmp0, tmp1, tmp2, tmp3, dst, dst_stride);
289 dst += (4 * dst_stride);
290 }
291 }
292
vpx_iadst16_1d_rows_msa(const int16_t * input,int16_t * output)293 void vpx_iadst16_1d_rows_msa(const int16_t *input, int16_t *output) {
294 v8i16 r0, r1, r2, r3, r4, r5, r6, r7, r8, r9, r10, r11, r12, r13, r14, r15;
295 v8i16 l0, l1, l2, l3, l4, l5, l6, l7, l8, l9, l10, l11, l12, l13, l14, l15;
296
297 /* load input data */
298 LD_SH16(input, 8, l0, l8, l1, l9, l2, l10, l3, l11, l4, l12, l5, l13, l6, l14,
299 l7, l15);
300 TRANSPOSE8x8_SH_SH(l0, l1, l2, l3, l4, l5, l6, l7, l0, l1, l2, l3, l4, l5, l6,
301 l7);
302 TRANSPOSE8x8_SH_SH(l8, l9, l10, l11, l12, l13, l14, l15, l8, l9, l10, l11,
303 l12, l13, l14, l15);
304
305 /* ADST in horizontal */
306 VP9_IADST8x16_1D(l0, l1, l2, l3, l4, l5, l6, l7, l8, l9, l10, l11, l12, l13,
307 l14, l15, r0, r1, r2, r3, r4, r5, r6, r7, r8, r9, r10, r11,
308 r12, r13, r14, r15);
309
310 l1 = -r8;
311 l3 = -r4;
312 l13 = -r13;
313 l15 = -r1;
314
315 TRANSPOSE8x8_SH_SH(r0, l1, r12, l3, r6, r14, r10, r2, l0, l1, l2, l3, l4, l5,
316 l6, l7);
317 ST_SH8(l0, l1, l2, l3, l4, l5, l6, l7, output, 16);
318 TRANSPOSE8x8_SH_SH(r3, r11, r15, r7, r5, l13, r9, l15, l8, l9, l10, l11, l12,
319 l13, l14, l15);
320 ST_SH8(l8, l9, l10, l11, l12, l13, l14, l15, (output + 8), 16);
321 }
322
vpx_iadst16_1d_columns_addblk_msa(int16_t * input,uint8_t * dst,int32_t dst_stride)323 void vpx_iadst16_1d_columns_addblk_msa(int16_t *input, uint8_t *dst,
324 int32_t dst_stride) {
325 v8i16 v0, v2, v4, v6, k0, k1, k2, k3;
326 v8i16 r0, r1, r2, r3, r4, r5, r6, r7, r8, r9, r10, r11, r12, r13, r14, r15;
327 v8i16 out0, out1, out2, out3, out4, out5, out6, out7;
328 v8i16 out8, out9, out10, out11, out12, out13, out14, out15;
329 v8i16 g0, g1, g2, g3, g4, g5, g6, g7, g8, g9, g10, g11, g12, g13, g14, g15;
330 v8i16 h0, h1, h2, h3, h4, h5, h6, h7, h8, h9, h10, h11;
331 v8i16 res0, res1, res2, res3, res4, res5, res6, res7;
332 v8i16 res8, res9, res10, res11, res12, res13, res14, res15;
333 v16u8 dst0, dst1, dst2, dst3, dst4, dst5, dst6, dst7;
334 v16u8 dst8, dst9, dst10, dst11, dst12, dst13, dst14, dst15;
335 v16i8 zero = { 0 };
336
337 r0 = LD_SH(input + 0 * 16);
338 r3 = LD_SH(input + 3 * 16);
339 r4 = LD_SH(input + 4 * 16);
340 r7 = LD_SH(input + 7 * 16);
341 r8 = LD_SH(input + 8 * 16);
342 r11 = LD_SH(input + 11 * 16);
343 r12 = LD_SH(input + 12 * 16);
344 r15 = LD_SH(input + 15 * 16);
345
346 /* stage 1 */
347 k0 = VP9_SET_COSPI_PAIR(cospi_1_64, cospi_31_64);
348 k1 = VP9_SET_COSPI_PAIR(cospi_31_64, -cospi_1_64);
349 k2 = VP9_SET_COSPI_PAIR(cospi_17_64, cospi_15_64);
350 k3 = VP9_SET_COSPI_PAIR(cospi_15_64, -cospi_17_64);
351 MADD_BF(r15, r0, r7, r8, k0, k1, k2, k3, g0, g1, g2, g3);
352 k0 = VP9_SET_COSPI_PAIR(cospi_9_64, cospi_23_64);
353 k1 = VP9_SET_COSPI_PAIR(cospi_23_64, -cospi_9_64);
354 k2 = VP9_SET_COSPI_PAIR(cospi_25_64, cospi_7_64);
355 k3 = VP9_SET_COSPI_PAIR(cospi_7_64, -cospi_25_64);
356 MADD_BF(r11, r4, r3, r12, k0, k1, k2, k3, g8, g9, g10, g11);
357 BUTTERFLY_4(g0, g2, g10, g8, h8, h9, v2, v0);
358 k0 = VP9_SET_COSPI_PAIR(cospi_4_64, cospi_28_64);
359 k1 = VP9_SET_COSPI_PAIR(cospi_28_64, -cospi_4_64);
360 k2 = VP9_SET_COSPI_PAIR(-cospi_28_64, cospi_4_64);
361 MADD_BF(g1, g3, g9, g11, k0, k1, k2, k0, h0, h1, h2, h3);
362
363 r1 = LD_SH(input + 1 * 16);
364 r2 = LD_SH(input + 2 * 16);
365 r5 = LD_SH(input + 5 * 16);
366 r6 = LD_SH(input + 6 * 16);
367 r9 = LD_SH(input + 9 * 16);
368 r10 = LD_SH(input + 10 * 16);
369 r13 = LD_SH(input + 13 * 16);
370 r14 = LD_SH(input + 14 * 16);
371
372 k0 = VP9_SET_COSPI_PAIR(cospi_5_64, cospi_27_64);
373 k1 = VP9_SET_COSPI_PAIR(cospi_27_64, -cospi_5_64);
374 k2 = VP9_SET_COSPI_PAIR(cospi_21_64, cospi_11_64);
375 k3 = VP9_SET_COSPI_PAIR(cospi_11_64, -cospi_21_64);
376 MADD_BF(r13, r2, r5, r10, k0, k1, k2, k3, g4, g5, g6, g7);
377 k0 = VP9_SET_COSPI_PAIR(cospi_13_64, cospi_19_64);
378 k1 = VP9_SET_COSPI_PAIR(cospi_19_64, -cospi_13_64);
379 k2 = VP9_SET_COSPI_PAIR(cospi_29_64, cospi_3_64);
380 k3 = VP9_SET_COSPI_PAIR(cospi_3_64, -cospi_29_64);
381 MADD_BF(r9, r6, r1, r14, k0, k1, k2, k3, g12, g13, g14, g15);
382 BUTTERFLY_4(g4, g6, g14, g12, h10, h11, v6, v4);
383 BUTTERFLY_4(h8, h9, h11, h10, out0, out1, h11, h10);
384 out1 = -out1;
385 SRARI_H2_SH(out0, out1, 6);
386 dst0 = LD_UB(dst + 0 * dst_stride);
387 dst1 = LD_UB(dst + 15 * dst_stride);
388 ILVR_B2_SH(zero, dst0, zero, dst1, res0, res1);
389 ADD2(res0, out0, res1, out1, res0, res1);
390 CLIP_SH2_0_255(res0, res1);
391 PCKEV_B2_SH(res0, res0, res1, res1, res0, res1);
392 ST8x1_UB(res0, dst);
393 ST8x1_UB(res1, dst + 15 * dst_stride);
394
395 k0 = VP9_SET_COSPI_PAIR(cospi_12_64, cospi_20_64);
396 k1 = VP9_SET_COSPI_PAIR(-cospi_20_64, cospi_12_64);
397 k2 = VP9_SET_COSPI_PAIR(cospi_20_64, -cospi_12_64);
398 MADD_BF(g7, g5, g15, g13, k0, k1, k2, k0, h4, h5, h6, h7);
399 BUTTERFLY_4(h0, h2, h6, h4, out8, out9, out11, out10);
400 out8 = -out8;
401
402 SRARI_H2_SH(out8, out9, 6);
403 dst8 = LD_UB(dst + 1 * dst_stride);
404 dst9 = LD_UB(dst + 14 * dst_stride);
405 ILVR_B2_SH(zero, dst8, zero, dst9, res8, res9);
406 ADD2(res8, out8, res9, out9, res8, res9);
407 CLIP_SH2_0_255(res8, res9);
408 PCKEV_B2_SH(res8, res8, res9, res9, res8, res9);
409 ST8x1_UB(res8, dst + dst_stride);
410 ST8x1_UB(res9, dst + 14 * dst_stride);
411
412 k0 = VP9_SET_COSPI_PAIR(cospi_8_64, cospi_24_64);
413 k1 = VP9_SET_COSPI_PAIR(cospi_24_64, -cospi_8_64);
414 k2 = VP9_SET_COSPI_PAIR(-cospi_24_64, cospi_8_64);
415 MADD_BF(v0, v2, v4, v6, k0, k1, k2, k0, out4, out6, out5, out7);
416 out4 = -out4;
417 SRARI_H2_SH(out4, out5, 6);
418 dst4 = LD_UB(dst + 3 * dst_stride);
419 dst5 = LD_UB(dst + 12 * dst_stride);
420 ILVR_B2_SH(zero, dst4, zero, dst5, res4, res5);
421 ADD2(res4, out4, res5, out5, res4, res5);
422 CLIP_SH2_0_255(res4, res5);
423 PCKEV_B2_SH(res4, res4, res5, res5, res4, res5);
424 ST8x1_UB(res4, dst + 3 * dst_stride);
425 ST8x1_UB(res5, dst + 12 * dst_stride);
426
427 MADD_BF(h1, h3, h5, h7, k0, k1, k2, k0, out12, out14, out13, out15);
428 out13 = -out13;
429 SRARI_H2_SH(out12, out13, 6);
430 dst12 = LD_UB(dst + 2 * dst_stride);
431 dst13 = LD_UB(dst + 13 * dst_stride);
432 ILVR_B2_SH(zero, dst12, zero, dst13, res12, res13);
433 ADD2(res12, out12, res13, out13, res12, res13);
434 CLIP_SH2_0_255(res12, res13);
435 PCKEV_B2_SH(res12, res12, res13, res13, res12, res13);
436 ST8x1_UB(res12, dst + 2 * dst_stride);
437 ST8x1_UB(res13, dst + 13 * dst_stride);
438
439 k0 = VP9_SET_COSPI_PAIR(cospi_16_64, cospi_16_64);
440 k3 = VP9_SET_COSPI_PAIR(-cospi_16_64, cospi_16_64);
441 MADD_SHORT(out6, out7, k0, k3, out6, out7);
442 SRARI_H2_SH(out6, out7, 6);
443 dst6 = LD_UB(dst + 4 * dst_stride);
444 dst7 = LD_UB(dst + 11 * dst_stride);
445 ILVR_B2_SH(zero, dst6, zero, dst7, res6, res7);
446 ADD2(res6, out6, res7, out7, res6, res7);
447 CLIP_SH2_0_255(res6, res7);
448 PCKEV_B2_SH(res6, res6, res7, res7, res6, res7);
449 ST8x1_UB(res6, dst + 4 * dst_stride);
450 ST8x1_UB(res7, dst + 11 * dst_stride);
451
452 MADD_SHORT(out10, out11, k0, k3, out10, out11);
453 SRARI_H2_SH(out10, out11, 6);
454 dst10 = LD_UB(dst + 6 * dst_stride);
455 dst11 = LD_UB(dst + 9 * dst_stride);
456 ILVR_B2_SH(zero, dst10, zero, dst11, res10, res11);
457 ADD2(res10, out10, res11, out11, res10, res11);
458 CLIP_SH2_0_255(res10, res11);
459 PCKEV_B2_SH(res10, res10, res11, res11, res10, res11);
460 ST8x1_UB(res10, dst + 6 * dst_stride);
461 ST8x1_UB(res11, dst + 9 * dst_stride);
462
463 k1 = VP9_SET_COSPI_PAIR(-cospi_16_64, -cospi_16_64);
464 k2 = VP9_SET_COSPI_PAIR(cospi_16_64, -cospi_16_64);
465 MADD_SHORT(h10, h11, k1, k2, out2, out3);
466 SRARI_H2_SH(out2, out3, 6);
467 dst2 = LD_UB(dst + 7 * dst_stride);
468 dst3 = LD_UB(dst + 8 * dst_stride);
469 ILVR_B2_SH(zero, dst2, zero, dst3, res2, res3);
470 ADD2(res2, out2, res3, out3, res2, res3);
471 CLIP_SH2_0_255(res2, res3);
472 PCKEV_B2_SH(res2, res2, res3, res3, res2, res3);
473 ST8x1_UB(res2, dst + 7 * dst_stride);
474 ST8x1_UB(res3, dst + 8 * dst_stride);
475
476 MADD_SHORT(out14, out15, k1, k2, out14, out15);
477 SRARI_H2_SH(out14, out15, 6);
478 dst14 = LD_UB(dst + 5 * dst_stride);
479 dst15 = LD_UB(dst + 10 * dst_stride);
480 ILVR_B2_SH(zero, dst14, zero, dst15, res14, res15);
481 ADD2(res14, out14, res15, out15, res14, res15);
482 CLIP_SH2_0_255(res14, res15);
483 PCKEV_B2_SH(res14, res14, res15, res15, res14, res15);
484 ST8x1_UB(res14, dst + 5 * dst_stride);
485 ST8x1_UB(res15, dst + 10 * dst_stride);
486 }
487