• Home
  • History
  • Annotate
  • Line#
  • Scopes#
  • Navigate#
  • Raw
  • Download
1  // Copyright 2016 Google Inc. All Rights Reserved.
2  //
3  // Use of this source code is governed by a BSD-style license
4  // that can be found in the COPYING file in the root of the source
5  // tree. An additional intellectual property rights grant can be found
6  // in the file PATENTS. All contributing project authors may
7  // be found in the AUTHORS file in the root of the source tree.
8  // -----------------------------------------------------------------------------
9  //
10  // MSA common macros
11  //
12  // Author(s):  Prashant Patil   (prashant.patil@imgtec.com)
13  
14  #ifndef WEBP_DSP_MSA_MACRO_H_
15  #define WEBP_DSP_MSA_MACRO_H_
16  
17  #include <stdint.h>
18  #include <msa.h>
19  
20  #if defined(__clang__)
21    #define CLANG_BUILD
22  #endif
23  
24  #ifdef CLANG_BUILD
25    #define ADDVI_H(a, b)  __msa_addvi_h((v8i16)a, b)
26    #define ADDVI_W(a, b)  __msa_addvi_w((v4i32)a, b)
27    #define SRAI_B(a, b)  __msa_srai_b((v16i8)a, b)
28    #define SRAI_H(a, b)  __msa_srai_h((v8i16)a, b)
29    #define SRAI_W(a, b)  __msa_srai_w((v4i32)a, b)
30    #define SRLI_H(a, b)  __msa_srli_h((v8i16)a, b)
31    #define SLLI_B(a, b)  __msa_slli_b((v4i32)a, b)
32    #define ANDI_B(a, b)  __msa_andi_b((v16u8)a, b)
33    #define ORI_B(a, b)   __msa_ori_b((v16u8)a, b)
34  #else
35    #define ADDVI_H(a, b)  (a + b)
36    #define ADDVI_W(a, b)  (a + b)
37    #define SRAI_B(a, b)  (a >> b)
38    #define SRAI_H(a, b)  (a >> b)
39    #define SRAI_W(a, b)  (a >> b)
40    #define SRLI_H(a, b)  (a << b)
41    #define SLLI_B(a, b)  (a << b)
42    #define ANDI_B(a, b)  (a & b)
43    #define ORI_B(a, b)   (a | b)
44  #endif
45  
46  #define LD_B(RTYPE, psrc) *((RTYPE*)(psrc))
47  #define LD_UB(...) LD_B(v16u8, __VA_ARGS__)
48  #define LD_SB(...) LD_B(v16i8, __VA_ARGS__)
49  
50  #define LD_H(RTYPE, psrc) *((RTYPE*)(psrc))
51  #define LD_UH(...) LD_H(v8u16, __VA_ARGS__)
52  #define LD_SH(...) LD_H(v8i16, __VA_ARGS__)
53  
54  #define LD_W(RTYPE, psrc) *((RTYPE*)(psrc))
55  #define LD_UW(...) LD_W(v4u32, __VA_ARGS__)
56  #define LD_SW(...) LD_W(v4i32, __VA_ARGS__)
57  
58  #define ST_B(RTYPE, in, pdst) *((RTYPE*)(pdst)) = in
59  #define ST_UB(...) ST_B(v16u8, __VA_ARGS__)
60  #define ST_SB(...) ST_B(v16i8, __VA_ARGS__)
61  
62  #define ST_H(RTYPE, in, pdst) *((RTYPE*)(pdst)) = in
63  #define ST_UH(...) ST_H(v8u16, __VA_ARGS__)
64  #define ST_SH(...) ST_H(v8i16, __VA_ARGS__)
65  
66  #define ST_W(RTYPE, in, pdst) *((RTYPE*)(pdst)) = in
67  #define ST_UW(...) ST_W(v4u32, __VA_ARGS__)
68  #define ST_SW(...) ST_W(v4i32, __VA_ARGS__)
69  
70  #define MSA_LOAD_FUNC(TYPE, INSTR, FUNC_NAME)             \
71    static inline TYPE FUNC_NAME(const void* const psrc) {  \
72      const uint8_t* const psrc_m = (const uint8_t*)psrc;   \
73      TYPE val_m;                                           \
74      asm volatile (                                        \
75        "" #INSTR " %[val_m], %[psrc_m]  \n\t"              \
76        : [val_m] "=r" (val_m)                              \
77        : [psrc_m] "m" (*psrc_m));                          \
78      return val_m;                                         \
79    }
80  
81  #define MSA_LOAD(psrc, FUNC_NAME)  FUNC_NAME(psrc)
82  
83  #define MSA_STORE_FUNC(TYPE, INSTR, FUNC_NAME)               \
84    static inline void FUNC_NAME(TYPE val, void* const pdst) { \
85      uint8_t* const pdst_m = (uint8_t*)pdst;                  \
86      TYPE val_m = val;                                        \
87      asm volatile (                                           \
88        " " #INSTR "  %[val_m],  %[pdst_m]  \n\t"              \
89        : [pdst_m] "=m" (*pdst_m)                              \
90        : [val_m] "r" (val_m));                                \
91    }
92  
93  #define MSA_STORE(val, pdst, FUNC_NAME)  FUNC_NAME(val, pdst)
94  
95  #if (__mips_isa_rev >= 6)
96    MSA_LOAD_FUNC(uint16_t, lh, msa_lh);
97    #define LH(psrc)  MSA_LOAD(psrc, msa_lh)
98    MSA_LOAD_FUNC(uint32_t, lw, msa_lw);
99    #define LW(psrc)  MSA_LOAD(psrc, msa_lw)
100    #if (__mips == 64)
101      MSA_LOAD_FUNC(uint64_t, ld, msa_ld);
102      #define LD(psrc)  MSA_LOAD(psrc, msa_ld)
103    #else  // !(__mips == 64)
104      #define LD(psrc)  ((((uint64_t)MSA_LOAD(psrc + 4, msa_lw)) << 32) | \
105                         MSA_LOAD(psrc, msa_lw))
106    #endif  // (__mips == 64)
107  
108    MSA_STORE_FUNC(uint16_t, sh, msa_sh);
109    #define SH(val, pdst)  MSA_STORE(val, pdst, msa_sh)
110    MSA_STORE_FUNC(uint32_t, sw, msa_sw);
111    #define SW(val, pdst)  MSA_STORE(val, pdst, msa_sw)
112    MSA_STORE_FUNC(uint64_t, sd, msa_sd);
113    #define SD(val, pdst)  MSA_STORE(val, pdst, msa_sd)
114  #else  // !(__mips_isa_rev >= 6)
115    MSA_LOAD_FUNC(uint16_t, ulh, msa_ulh);
116    #define LH(psrc)  MSA_LOAD(psrc, msa_ulh)
117    MSA_LOAD_FUNC(uint32_t, ulw, msa_ulw);
118    #define LW(psrc)  MSA_LOAD(psrc, msa_ulw)
119    #if (__mips == 64)
120      MSA_LOAD_FUNC(uint64_t, uld, msa_uld);
121      #define LD(psrc)  MSA_LOAD(psrc, msa_uld)
122    #else  // !(__mips == 64)
123      #define LD(psrc)  ((((uint64_t)MSA_LOAD(psrc + 4, msa_ulw)) << 32) | \
124                          MSA_LOAD(psrc, msa_ulw))
125    #endif  // (__mips == 64)
126  
127    MSA_STORE_FUNC(uint16_t, ush, msa_ush);
128    #define SH(val, pdst)  MSA_STORE(val, pdst, msa_ush)
129    MSA_STORE_FUNC(uint32_t, usw, msa_usw);
130    #define SW(val, pdst)  MSA_STORE(val, pdst, msa_usw)
131    #define SD(val, pdst) do {                                               \
132      uint8_t* const pdst_sd_m = (uint8_t*)(pdst);                           \
133      const uint32_t val0_m = (uint32_t)(val & 0x00000000FFFFFFFF);          \
134      const uint32_t val1_m = (uint32_t)((val >> 32) & 0x00000000FFFFFFFF);  \
135      SW(val0_m, pdst_sd_m);                                                 \
136      SW(val1_m, pdst_sd_m + 4);                                             \
137    } while (0)
138  #endif  // (__mips_isa_rev >= 6)
139  
140  /* Description : Load 4 words with stride
141   * Arguments   : Inputs  - psrc, stride
142   *               Outputs - out0, out1, out2, out3
143   * Details     : Load word in 'out0' from (psrc)
144   *               Load word in 'out1' from (psrc + stride)
145   *               Load word in 'out2' from (psrc + 2 * stride)
146   *               Load word in 'out3' from (psrc + 3 * stride)
147   */
148  #define LW4(psrc, stride, out0, out1, out2, out3) do {  \
149    const uint8_t* ptmp = (const uint8_t*)psrc;           \
150    out0 = LW(ptmp);                                      \
151    ptmp += stride;                                       \
152    out1 = LW(ptmp);                                      \
153    ptmp += stride;                                       \
154    out2 = LW(ptmp);                                      \
155    ptmp += stride;                                       \
156    out3 = LW(ptmp);                                      \
157  } while (0)
158  
159  /* Description : Store words with stride
160   * Arguments   : Inputs - in0, in1, in2, in3, pdst, stride
161   * Details     : Store word from 'in0' to (pdst)
162   *               Store word from 'in1' to (pdst + stride)
163   *               Store word from 'in2' to (pdst + 2 * stride)
164   *               Store word from 'in3' to (pdst + 3 * stride)
165   */
166  #define SW4(in0, in1, in2, in3, pdst, stride) do {  \
167    uint8_t* ptmp = (uint8_t*)pdst;                   \
168    SW(in0, ptmp);                                    \
169    ptmp += stride;                                   \
170    SW(in1, ptmp);                                    \
171    ptmp += stride;                                   \
172    SW(in2, ptmp);                                    \
173    ptmp += stride;                                   \
174    SW(in3, ptmp);                                    \
175  } while (0)
176  
177  #define SW3(in0, in1, in2, pdst, stride) do {  \
178    uint8_t* ptmp = (uint8_t*)pdst;              \
179    SW(in0, ptmp);                               \
180    ptmp += stride;                              \
181    SW(in1, ptmp);                               \
182    ptmp += stride;                              \
183    SW(in2, ptmp);                               \
184  } while (0)
185  
186  #define SW2(in0, in1, pdst, stride) do {  \
187    uint8_t* ptmp = (uint8_t*)pdst;         \
188    SW(in0, ptmp);                          \
189    ptmp += stride;                         \
190    SW(in1, ptmp);                          \
191  } while (0)
192  
193  /* Description : Store 4 double words with stride
194   * Arguments   : Inputs - in0, in1, in2, in3, pdst, stride
195   * Details     : Store double word from 'in0' to (pdst)
196   *               Store double word from 'in1' to (pdst + stride)
197   *               Store double word from 'in2' to (pdst + 2 * stride)
198   *               Store double word from 'in3' to (pdst + 3 * stride)
199   */
200  #define SD4(in0, in1, in2, in3, pdst, stride) do {  \
201    uint8_t* ptmp = (uint8_t*)pdst;                   \
202    SD(in0, ptmp);                                    \
203    ptmp += stride;                                   \
204    SD(in1, ptmp);                                    \
205    ptmp += stride;                                   \
206    SD(in2, ptmp);                                    \
207    ptmp += stride;                                   \
208    SD(in3, ptmp);                                    \
209  } while (0)
210  
211  /* Description : Load vectors with 16 byte elements with stride
212   * Arguments   : Inputs  - psrc, stride
213   *               Outputs - out0, out1
214   *               Return Type - as per RTYPE
215   * Details     : Load 16 byte elements in 'out0' from (psrc)
216   *               Load 16 byte elements in 'out1' from (psrc + stride)
217   */
218  #define LD_B2(RTYPE, psrc, stride, out0, out1) do {  \
219    out0 = LD_B(RTYPE, psrc);                          \
220    out1 = LD_B(RTYPE, psrc + stride);                 \
221  } while (0)
222  #define LD_UB2(...) LD_B2(v16u8, __VA_ARGS__)
223  #define LD_SB2(...) LD_B2(v16i8, __VA_ARGS__)
224  
225  #define LD_B3(RTYPE, psrc, stride, out0, out1, out2) do {  \
226    LD_B2(RTYPE, psrc, stride, out0, out1);                  \
227    out2 = LD_B(RTYPE, psrc + 2 * stride);                   \
228  } while (0)
229  #define LD_UB3(...) LD_B3(v16u8, __VA_ARGS__)
230  #define LD_SB3(...) LD_B3(v16i8, __VA_ARGS__)
231  
232  #define LD_B4(RTYPE, psrc, stride, out0, out1, out2, out3) do {  \
233    LD_B2(RTYPE, psrc, stride, out0, out1);                        \
234    LD_B2(RTYPE, psrc + 2 * stride , stride, out2, out3);          \
235  } while (0)
236  #define LD_UB4(...) LD_B4(v16u8, __VA_ARGS__)
237  #define LD_SB4(...) LD_B4(v16i8, __VA_ARGS__)
238  
239  #define LD_B8(RTYPE, psrc, stride,                                  \
240                out0, out1, out2, out3, out4, out5, out6, out7) do {  \
241    LD_B4(RTYPE, psrc, stride, out0, out1, out2, out3);               \
242    LD_B4(RTYPE, psrc + 4 * stride, stride, out4, out5, out6, out7);  \
243  } while (0)
244  #define LD_UB8(...) LD_B8(v16u8, __VA_ARGS__)
245  #define LD_SB8(...) LD_B8(v16i8, __VA_ARGS__)
246  
247  /* Description : Load vectors with 8 halfword elements with stride
248   * Arguments   : Inputs  - psrc, stride
249   *               Outputs - out0, out1
250   * Details     : Load 8 halfword elements in 'out0' from (psrc)
251   *               Load 8 halfword elements in 'out1' from (psrc + stride)
252   */
253  #define LD_H2(RTYPE, psrc, stride, out0, out1) do {  \
254    out0 = LD_H(RTYPE, psrc);                          \
255    out1 = LD_H(RTYPE, psrc + stride);                 \
256  } while (0)
257  #define LD_UH2(...) LD_H2(v8u16, __VA_ARGS__)
258  #define LD_SH2(...) LD_H2(v8i16, __VA_ARGS__)
259  
260  /* Description : Load vectors with 4 word elements with stride
261   * Arguments   : Inputs  - psrc, stride
262   *               Outputs - out0, out1, out2, out3
263   * Details     : Load 4 word elements in 'out0' from (psrc + 0 * stride)
264   *               Load 4 word elements in 'out1' from (psrc + 1 * stride)
265   *               Load 4 word elements in 'out2' from (psrc + 2 * stride)
266   *               Load 4 word elements in 'out3' from (psrc + 3 * stride)
267   */
268  #define LD_W2(RTYPE, psrc, stride, out0, out1) do {  \
269    out0 = LD_W(RTYPE, psrc);                          \
270    out1 = LD_W(RTYPE, psrc + stride);                 \
271  } while (0)
272  #define LD_UW2(...) LD_W2(v4u32, __VA_ARGS__)
273  #define LD_SW2(...) LD_W2(v4i32, __VA_ARGS__)
274  
275  #define LD_W3(RTYPE, psrc, stride, out0, out1, out2) do {  \
276    LD_W2(RTYPE, psrc, stride, out0, out1);                  \
277    out2 = LD_W(RTYPE, psrc + 2 * stride);                   \
278  } while (0)
279  #define LD_UW3(...) LD_W3(v4u32, __VA_ARGS__)
280  #define LD_SW3(...) LD_W3(v4i32, __VA_ARGS__)
281  
282  #define LD_W4(RTYPE, psrc, stride, out0, out1, out2, out3) do {  \
283    LD_W2(RTYPE, psrc, stride, out0, out1);                        \
284    LD_W2(RTYPE, psrc + 2 * stride, stride, out2, out3);           \
285  } while (0)
286  #define LD_UW4(...) LD_W4(v4u32, __VA_ARGS__)
287  #define LD_SW4(...) LD_W4(v4i32, __VA_ARGS__)
288  
289  /* Description : Store vectors of 16 byte elements with stride
290   * Arguments   : Inputs - in0, in1, pdst, stride
291   * Details     : Store 16 byte elements from 'in0' to (pdst)
292   *               Store 16 byte elements from 'in1' to (pdst + stride)
293   */
294  #define ST_B2(RTYPE, in0, in1, pdst, stride) do {  \
295    ST_B(RTYPE, in0, pdst);                          \
296    ST_B(RTYPE, in1, pdst + stride);                 \
297  } while (0)
298  #define ST_UB2(...) ST_B2(v16u8, __VA_ARGS__)
299  #define ST_SB2(...) ST_B2(v16i8, __VA_ARGS__)
300  
301  #define ST_B4(RTYPE, in0, in1, in2, in3, pdst, stride) do {  \
302    ST_B2(RTYPE, in0, in1, pdst, stride);                      \
303    ST_B2(RTYPE, in2, in3, pdst + 2 * stride, stride);         \
304  } while (0)
305  #define ST_UB4(...) ST_B4(v16u8, __VA_ARGS__)
306  #define ST_SB4(...) ST_B4(v16i8, __VA_ARGS__)
307  
308  #define ST_B8(RTYPE, in0, in1, in2, in3, in4, in5, in6, in7,    \
309                pdst, stride) do {                                \
310    ST_B4(RTYPE, in0, in1, in2, in3, pdst, stride);               \
311    ST_B4(RTYPE, in4, in5, in6, in7, pdst + 4 * stride, stride);  \
312  } while (0)
313  #define ST_UB8(...) ST_B8(v16u8, __VA_ARGS__)
314  
315  /* Description : Store vectors of 4 word elements with stride
316   * Arguments   : Inputs - in0, in1, in2, in3, pdst, stride
317   * Details     : Store 4 word elements from 'in0' to (pdst + 0 * stride)
318   *               Store 4 word elements from 'in1' to (pdst + 1 * stride)
319   *               Store 4 word elements from 'in2' to (pdst + 2 * stride)
320   *               Store 4 word elements from 'in3' to (pdst + 3 * stride)
321   */
322  #define ST_W2(RTYPE, in0, in1, pdst, stride) do {  \
323    ST_W(RTYPE, in0, pdst);                          \
324    ST_W(RTYPE, in1, pdst + stride);                 \
325  } while (0)
326  #define ST_UW2(...) ST_W2(v4u32, __VA_ARGS__)
327  #define ST_SW2(...) ST_W2(v4i32, __VA_ARGS__)
328  
329  #define ST_W3(RTYPE, in0, in1, in2, pdst, stride) do {  \
330    ST_W2(RTYPE, in0, in1, pdst, stride);                 \
331    ST_W(RTYPE, in2, pdst + 2 * stride);                  \
332  } while (0)
333  #define ST_UW3(...) ST_W3(v4u32, __VA_ARGS__)
334  #define ST_SW3(...) ST_W3(v4i32, __VA_ARGS__)
335  
336  #define ST_W4(RTYPE, in0, in1, in2, in3, pdst, stride) do {  \
337    ST_W2(RTYPE, in0, in1, pdst, stride);                      \
338    ST_W2(RTYPE, in2, in3, pdst + 2 * stride, stride);         \
339  } while (0)
340  #define ST_UW4(...) ST_W4(v4u32, __VA_ARGS__)
341  #define ST_SW4(...) ST_W4(v4i32, __VA_ARGS__)
342  
343  /* Description : Store vectors of 8 halfword elements with stride
344   * Arguments   : Inputs - in0, in1, pdst, stride
345   * Details     : Store 8 halfword elements from 'in0' to (pdst)
346   *               Store 8 halfword elements from 'in1' to (pdst + stride)
347   */
348  #define ST_H2(RTYPE, in0, in1, pdst, stride) do {  \
349    ST_H(RTYPE, in0, pdst);                          \
350    ST_H(RTYPE, in1, pdst + stride);                 \
351  } while (0)
352  #define ST_UH2(...) ST_H2(v8u16, __VA_ARGS__)
353  #define ST_SH2(...) ST_H2(v8i16, __VA_ARGS__)
354  
355  /* Description : Store 2x4 byte block to destination memory from input vector
356   * Arguments   : Inputs - in, stidx, pdst, stride
357   * Details     : Index 'stidx' halfword element from 'in' vector is copied to
358   *               the GP register and stored to (pdst)
359   *               Index 'stidx+1' halfword element from 'in' vector is copied to
360   *               the GP register and stored to (pdst + stride)
361   *               Index 'stidx+2' halfword element from 'in' vector is copied to
362   *               the GP register and stored to (pdst + 2 * stride)
363   *               Index 'stidx+3' halfword element from 'in' vector is copied to
364   *               the GP register and stored to (pdst + 3 * stride)
365   */
366  #define ST2x4_UB(in, stidx, pdst, stride) do {                   \
367    uint8_t* pblk_2x4_m = (uint8_t*)pdst;                          \
368    const uint16_t out0_m = __msa_copy_s_h((v8i16)in, stidx);      \
369    const uint16_t out1_m = __msa_copy_s_h((v8i16)in, stidx + 1);  \
370    const uint16_t out2_m = __msa_copy_s_h((v8i16)in, stidx + 2);  \
371    const uint16_t out3_m = __msa_copy_s_h((v8i16)in, stidx + 3);  \
372    SH(out0_m, pblk_2x4_m);                                        \
373    pblk_2x4_m += stride;                                          \
374    SH(out1_m, pblk_2x4_m);                                        \
375    pblk_2x4_m += stride;                                          \
376    SH(out2_m, pblk_2x4_m);                                        \
377    pblk_2x4_m += stride;                                          \
378    SH(out3_m, pblk_2x4_m);                                        \
379  } while (0)
380  
381  /* Description : Store 4x4 byte block to destination memory from input vector
382   * Arguments   : Inputs - in0, in1, pdst, stride
383   * Details     : 'Idx0' word element from input vector 'in0' is copied to the
384   *               GP register and stored to (pdst)
385   *               'Idx1' word element from input vector 'in0' is copied to the
386   *               GP register and stored to (pdst + stride)
387   *               'Idx2' word element from input vector 'in0' is copied to the
388   *               GP register and stored to (pdst + 2 * stride)
389   *               'Idx3' word element from input vector 'in0' is copied to the
390   *               GP register and stored to (pdst + 3 * stride)
391   */
392  #define ST4x4_UB(in0, in1, idx0, idx1, idx2, idx3, pdst, stride) do {  \
393    uint8_t* const pblk_4x4_m = (uint8_t*)pdst;                          \
394    const uint32_t out0_m = __msa_copy_s_w((v4i32)in0, idx0);            \
395    const uint32_t out1_m = __msa_copy_s_w((v4i32)in0, idx1);            \
396    const uint32_t out2_m = __msa_copy_s_w((v4i32)in1, idx2);            \
397    const uint32_t out3_m = __msa_copy_s_w((v4i32)in1, idx3);            \
398    SW4(out0_m, out1_m, out2_m, out3_m, pblk_4x4_m, stride);             \
399  } while (0)
400  
401  #define ST4x8_UB(in0, in1, pdst, stride) do {                     \
402    uint8_t* const pblk_4x8 = (uint8_t*)pdst;                       \
403    ST4x4_UB(in0, in0, 0, 1, 2, 3, pblk_4x8, stride);               \
404    ST4x4_UB(in1, in1, 0, 1, 2, 3, pblk_4x8 + 4 * stride, stride);  \
405  } while (0)
406  
407  /* Description : Immediate number of elements to slide
408   * Arguments   : Inputs  - in0, in1, slide_val
409   *               Outputs - out
410   *               Return Type - as per RTYPE
411   * Details     : Byte elements from 'in1' vector are slid into 'in0' by
412   *               value specified in the 'slide_val'
413   */
414  #define SLDI_B(RTYPE, in0, in1, slide_val)                      \
415          (RTYPE)__msa_sldi_b((v16i8)in0, (v16i8)in1, slide_val)  \
416  
417  #define SLDI_UB(...) SLDI_B(v16u8, __VA_ARGS__)
418  #define SLDI_SB(...) SLDI_B(v16i8, __VA_ARGS__)
419  #define SLDI_SH(...) SLDI_B(v8i16, __VA_ARGS__)
420  
421  /* Description : Shuffle byte vector elements as per mask vector
422   * Arguments   : Inputs  - in0, in1, in2, in3, mask0, mask1
423   *               Outputs - out0, out1
424   *               Return Type - as per RTYPE
425   * Details     : Byte elements from 'in0' & 'in1' are copied selectively to
426   *               'out0' as per control vector 'mask0'
427   */
428  #define VSHF_B(RTYPE, in0, in1, mask)                              \
429          (RTYPE)__msa_vshf_b((v16i8)mask, (v16i8)in1, (v16i8)in0)
430  
431  #define VSHF_UB(...) VSHF_B(v16u8, __VA_ARGS__)
432  #define VSHF_SB(...) VSHF_B(v16i8, __VA_ARGS__)
433  #define VSHF_UH(...) VSHF_B(v8u16, __VA_ARGS__)
434  #define VSHF_SH(...) VSHF_B(v8i16, __VA_ARGS__)
435  
436  #define VSHF_B2(RTYPE, in0, in1, in2, in3, mask0, mask1, out0, out1) do {  \
437    out0 = VSHF_B(RTYPE, in0, in1, mask0);                                   \
438    out1 = VSHF_B(RTYPE, in2, in3, mask1);                                   \
439  } while (0)
440  #define VSHF_B2_UB(...) VSHF_B2(v16u8, __VA_ARGS__)
441  #define VSHF_B2_SB(...) VSHF_B2(v16i8, __VA_ARGS__)
442  #define VSHF_B2_UH(...) VSHF_B2(v8u16, __VA_ARGS__)
443  #define VSHF_B2_SH(...) VSHF_B2(v8i16, __VA_ARGS__)
444  
445  /* Description : Shuffle halfword vector elements as per mask vector
446   * Arguments   : Inputs  - in0, in1, in2, in3, mask0, mask1
447   *               Outputs - out0, out1
448   *               Return Type - as per RTYPE
449   * Details     : halfword elements from 'in0' & 'in1' are copied selectively to
450   *               'out0' as per control vector 'mask0'
451   */
452  #define VSHF_H2(RTYPE, in0, in1, in2, in3, mask0, mask1, out0, out1) do {  \
453    out0 = (RTYPE)__msa_vshf_h((v8i16)mask0, (v8i16)in1, (v8i16)in0);        \
454    out1 = (RTYPE)__msa_vshf_h((v8i16)mask1, (v8i16)in3, (v8i16)in2);        \
455  } while (0)
456  #define VSHF_H2_UH(...) VSHF_H2(v8u16, __VA_ARGS__)
457  #define VSHF_H2_SH(...) VSHF_H2(v8i16, __VA_ARGS__)
458  
459  /* Description : Dot product of byte vector elements
460   * Arguments   : Inputs  - mult0, mult1, cnst0, cnst1
461   *               Outputs - out0, out1
462   *               Return Type - as per RTYPE
463   * Details     : Signed byte elements from 'mult0' are multiplied with
464   *               signed byte elements from 'cnst0' producing a result
465   *               twice the size of input i.e. signed halfword.
466   *               The multiplication result of adjacent odd-even elements
467   *               are added together and written to the 'out0' vector
468  */
469  #define DOTP_SB2(RTYPE, mult0, mult1, cnst0, cnst1, out0, out1) do {  \
470    out0 = (RTYPE)__msa_dotp_s_h((v16i8)mult0, (v16i8)cnst0);           \
471    out1 = (RTYPE)__msa_dotp_s_h((v16i8)mult1, (v16i8)cnst1);           \
472  } while (0)
473  #define DOTP_SB2_SH(...) DOTP_SB2(v8i16, __VA_ARGS__)
474  
475  /* Description : Dot product of halfword vector elements
476   * Arguments   : Inputs  - mult0, mult1, cnst0, cnst1
477   *               Outputs - out0, out1
478   *               Return Type - as per RTYPE
479   * Details     : Signed halfword elements from 'mult0' are multiplied with
480   *               signed halfword elements from 'cnst0' producing a result
481   *               twice the size of input i.e. signed word.
482   *               The multiplication result of adjacent odd-even elements
483   *               are added together and written to the 'out0' vector
484   */
485  #define DOTP_SH2(RTYPE, mult0, mult1, cnst0, cnst1, out0, out1) do {  \
486    out0 = (RTYPE)__msa_dotp_s_w((v8i16)mult0, (v8i16)cnst0);           \
487    out1 = (RTYPE)__msa_dotp_s_w((v8i16)mult1, (v8i16)cnst1);           \
488  } while (0)
489  #define DOTP_SH2_SW(...) DOTP_SH2(v4i32, __VA_ARGS__)
490  
491  /* Description : Dot product of unsigned word vector elements
492   * Arguments   : Inputs  - mult0, mult1, cnst0, cnst1
493   *               Outputs - out0, out1
494   *               Return Type - as per RTYPE
495   * Details     : Unsigned word elements from 'mult0' are multiplied with
496   *               unsigned word elements from 'cnst0' producing a result
497   *               twice the size of input i.e. unsigned double word.
498   *               The multiplication result of adjacent odd-even elements
499   *               are added together and written to the 'out0' vector
500   */
501  #define DOTP_UW2(RTYPE, mult0, mult1, cnst0, cnst1, out0, out1) do {  \
502    out0 = (RTYPE)__msa_dotp_u_d((v4u32)mult0, (v4u32)cnst0);           \
503    out1 = (RTYPE)__msa_dotp_u_d((v4u32)mult1, (v4u32)cnst1);           \
504  } while (0)
505  #define DOTP_UW2_UD(...) DOTP_UW2(v2u64, __VA_ARGS__)
506  
507  /* Description : Dot product & addition of halfword vector elements
508   * Arguments   : Inputs  - mult0, mult1, cnst0, cnst1
509   *               Outputs - out0, out1
510   *               Return Type - as per RTYPE
511   * Details     : Signed halfword elements from 'mult0' are multiplied with
512   *               signed halfword elements from 'cnst0' producing a result
513   *               twice the size of input i.e. signed word.
514   *               The multiplication result of adjacent odd-even elements
515   *               are added to the 'out0' vector
516   */
517  #define DPADD_SH2(RTYPE, mult0, mult1, cnst0, cnst1, out0, out1) do {      \
518    out0 = (RTYPE)__msa_dpadd_s_w((v4i32)out0, (v8i16)mult0, (v8i16)cnst0);  \
519    out1 = (RTYPE)__msa_dpadd_s_w((v4i32)out1, (v8i16)mult1, (v8i16)cnst1);  \
520  } while (0)
521  #define DPADD_SH2_SW(...) DPADD_SH2(v4i32, __VA_ARGS__)
522  
523  /* Description : Clips all signed halfword elements of input vector
524   *               between 0 & 255
525   * Arguments   : Input/output  - val
526   *               Return Type - signed halfword
527   */
528  #define CLIP_SH_0_255(val) do {                   \
529    const v8i16 max_m = __msa_ldi_h(255);           \
530    val = __msa_maxi_s_h((v8i16)val, 0);            \
531    val = __msa_min_s_h(max_m, (v8i16)val);         \
532  } while (0)
533  
534  #define CLIP_SH2_0_255(in0, in1) do {  \
535    CLIP_SH_0_255(in0);                  \
536    CLIP_SH_0_255(in1);                  \
537  } while (0)
538  
539  #define CLIP_SH4_0_255(in0, in1, in2, in3) do {  \
540    CLIP_SH2_0_255(in0, in1);                      \
541    CLIP_SH2_0_255(in2, in3);                      \
542  } while (0)
543  
544  /* Description : Clips all unsigned halfword elements of input vector
545   *               between 0 & 255
546   * Arguments   : Input  - in
547   *               Output - out_m
548   *               Return Type - unsigned halfword
549   */
550  #define CLIP_UH_0_255(in) do {                    \
551    const v8u16 max_m = (v8u16)__msa_ldi_h(255);    \
552    in = __msa_maxi_u_h((v8u16) in, 0);             \
553    in = __msa_min_u_h((v8u16) max_m, (v8u16) in);  \
554  } while (0)
555  
556  #define CLIP_UH2_0_255(in0, in1) do {  \
557    CLIP_UH_0_255(in0);                  \
558    CLIP_UH_0_255(in1);                  \
559  } while (0)
560  
561  /* Description : Clips all signed word elements of input vector
562   *               between 0 & 255
563   * Arguments   : Input/output  - val
564   *               Return Type - signed word
565   */
566  #define CLIP_SW_0_255(val) do {                   \
567    const v4i32 max_m = __msa_ldi_w(255);           \
568    val = __msa_maxi_s_w((v4i32)val, 0);            \
569    val = __msa_min_s_w(max_m, (v4i32)val);         \
570  } while (0)
571  
572  #define CLIP_SW4_0_255(in0, in1, in2, in3) do {   \
573    CLIP_SW_0_255(in0);                             \
574    CLIP_SW_0_255(in1);                             \
575    CLIP_SW_0_255(in2);                             \
576    CLIP_SW_0_255(in3);                             \
577  } while (0)
578  
579  /* Description : Horizontal addition of 4 signed word elements of input vector
580   * Arguments   : Input  - in       (signed word vector)
581   *               Output - sum_m    (i32 sum)
582   *               Return Type - signed word (GP)
583   * Details     : 4 signed word elements of 'in' vector are added together and
584   *               the resulting integer sum is returned
585   */
func_hadd_sw_s32(v4i32 in)586  static WEBP_INLINE int32_t func_hadd_sw_s32(v4i32 in) {
587    const v2i64 res0_m = __msa_hadd_s_d((v4i32)in, (v4i32)in);
588    const v2i64 res1_m = __msa_splati_d(res0_m, 1);
589    const v2i64 out = res0_m + res1_m;
590    int32_t sum_m = __msa_copy_s_w((v4i32)out, 0);
591    return sum_m;
592  }
593  #define HADD_SW_S32(in) func_hadd_sw_s32(in)
594  
595  /* Description : Horizontal addition of 8 signed halfword elements
596   * Arguments   : Input  - in       (signed halfword vector)
597   *               Output - sum_m    (s32 sum)
598   *               Return Type - signed word
599   * Details     : 8 signed halfword elements of input vector are added
600   *               together and the resulting integer sum is returned
601   */
func_hadd_sh_s32(v8i16 in)602  static WEBP_INLINE int32_t func_hadd_sh_s32(v8i16 in) {
603    const v4i32 res = __msa_hadd_s_w(in, in);
604    const v2i64 res0 = __msa_hadd_s_d(res, res);
605    const v2i64 res1 = __msa_splati_d(res0, 1);
606    const v2i64 res2 = res0 + res1;
607    const int32_t sum_m = __msa_copy_s_w((v4i32)res2, 0);
608    return sum_m;
609  }
610  #define HADD_SH_S32(in) func_hadd_sh_s32(in)
611  
612  /* Description : Horizontal addition of 8 unsigned halfword elements
613   * Arguments   : Input  - in       (unsigned halfword vector)
614   *               Output - sum_m    (u32 sum)
615   *               Return Type - unsigned word
616   * Details     : 8 unsigned halfword elements of input vector are added
617   *               together and the resulting integer sum is returned
618   */
func_hadd_uh_u32(v8u16 in)619  static WEBP_INLINE uint32_t func_hadd_uh_u32(v8u16 in) {
620    uint32_t sum_m;
621    const v4u32 res_m = __msa_hadd_u_w(in, in);
622    v2u64 res0_m = __msa_hadd_u_d(res_m, res_m);
623    v2u64 res1_m = (v2u64)__msa_splati_d((v2i64)res0_m, 1);
624    res0_m = res0_m + res1_m;
625    sum_m = __msa_copy_s_w((v4i32)res0_m, 0);
626    return sum_m;
627  }
628  #define HADD_UH_U32(in) func_hadd_uh_u32(in)
629  
630  /* Description : Horizontal addition of signed half word vector elements
631     Arguments   : Inputs  - in0, in1
632                   Outputs - out0, out1
633                   Return Type - as per RTYPE
634     Details     : Each signed odd half word element from 'in0' is added to
635                   even signed half word element from 'in0' (pairwise) and the
636                   halfword result is written in 'out0'
637  */
638  #define HADD_SH2(RTYPE, in0, in1, out0, out1) do {       \
639    out0 = (RTYPE)__msa_hadd_s_w((v8i16)in0, (v8i16)in0);  \
640    out1 = (RTYPE)__msa_hadd_s_w((v8i16)in1, (v8i16)in1);  \
641  } while (0)
642  #define HADD_SH2_SW(...) HADD_SH2(v4i32, __VA_ARGS__)
643  
644  #define HADD_SH4(RTYPE, in0, in1, in2, in3, out0, out1, out2, out3) do {  \
645    HADD_SH2(RTYPE, in0, in1, out0, out1);                                  \
646    HADD_SH2(RTYPE, in2, in3, out2, out3);                                  \
647  } while (0)
648  #define HADD_SH4_SW(...) HADD_SH4(v4i32, __VA_ARGS__)
649  
650  /* Description : Horizontal subtraction of unsigned byte vector elements
651   * Arguments   : Inputs  - in0, in1
652   *               Outputs - out0, out1
653   *               Return Type - as per RTYPE
654   * Details     : Each unsigned odd byte element from 'in0' is subtracted from
655   *               even unsigned byte element from 'in0' (pairwise) and the
656   *               halfword result is written to 'out0'
657   */
658  #define HSUB_UB2(RTYPE, in0, in1, out0, out1) do {       \
659    out0 = (RTYPE)__msa_hsub_u_h((v16u8)in0, (v16u8)in0);  \
660    out1 = (RTYPE)__msa_hsub_u_h((v16u8)in1, (v16u8)in1);  \
661  } while (0)
662  #define HSUB_UB2_UH(...) HSUB_UB2(v8u16, __VA_ARGS__)
663  #define HSUB_UB2_SH(...) HSUB_UB2(v8i16, __VA_ARGS__)
664  #define HSUB_UB2_SW(...) HSUB_UB2(v4i32, __VA_ARGS__)
665  
666  /* Description : Set element n input vector to GPR value
667   * Arguments   : Inputs - in0, in1, in2, in3
668   *               Output - out
669   *               Return Type - as per RTYPE
670   * Details     : Set element 0 in vector 'out' to value specified in 'in0'
671   */
672  #define INSERT_W2(RTYPE, in0, in1, out) do {        \
673    out = (RTYPE)__msa_insert_w((v4i32)out, 0, in0);  \
674    out = (RTYPE)__msa_insert_w((v4i32)out, 1, in1);  \
675  } while (0)
676  #define INSERT_W2_UB(...) INSERT_W2(v16u8, __VA_ARGS__)
677  #define INSERT_W2_SB(...) INSERT_W2(v16i8, __VA_ARGS__)
678  
679  #define INSERT_W4(RTYPE, in0, in1, in2, in3, out) do {  \
680    out = (RTYPE)__msa_insert_w((v4i32)out, 0, in0);      \
681    out = (RTYPE)__msa_insert_w((v4i32)out, 1, in1);      \
682    out = (RTYPE)__msa_insert_w((v4i32)out, 2, in2);      \
683    out = (RTYPE)__msa_insert_w((v4i32)out, 3, in3);      \
684  } while (0)
685  #define INSERT_W4_UB(...) INSERT_W4(v16u8, __VA_ARGS__)
686  #define INSERT_W4_SB(...) INSERT_W4(v16i8, __VA_ARGS__)
687  #define INSERT_W4_SW(...) INSERT_W4(v4i32, __VA_ARGS__)
688  
689  /* Description : Set element n of double word input vector to GPR value
690   * Arguments   : Inputs - in0, in1
691   *               Output - out
692   *               Return Type - as per RTYPE
693   * Details     : Set element 0 in vector 'out' to GPR value specified in 'in0'
694   *               Set element 1 in vector 'out' to GPR value specified in 'in1'
695   */
696  #define INSERT_D2(RTYPE, in0, in1, out) do {        \
697    out = (RTYPE)__msa_insert_d((v2i64)out, 0, in0);  \
698    out = (RTYPE)__msa_insert_d((v2i64)out, 1, in1);  \
699  } while (0)
700  #define INSERT_D2_UB(...) INSERT_D2(v16u8, __VA_ARGS__)
701  #define INSERT_D2_SB(...) INSERT_D2(v16i8, __VA_ARGS__)
702  
703  /* Description : Interleave even byte elements from vectors
704   * Arguments   : Inputs  - in0, in1, in2, in3
705   *               Outputs - out0, out1
706   *               Return Type - as per RTYPE
707   * Details     : Even byte elements of 'in0' and 'in1' are interleaved
708   *               and written to 'out0'
709   */
710  #define ILVEV_B2(RTYPE, in0, in1, in2, in3, out0, out1) do {  \
711    out0 = (RTYPE)__msa_ilvev_b((v16i8)in1, (v16i8)in0);        \
712    out1 = (RTYPE)__msa_ilvev_b((v16i8)in3, (v16i8)in2);        \
713  } while (0)
714  #define ILVEV_B2_UB(...) ILVEV_B2(v16u8, __VA_ARGS__)
715  #define ILVEV_B2_SB(...) ILVEV_B2(v16i8, __VA_ARGS__)
716  #define ILVEV_B2_UH(...) ILVEV_B2(v8u16, __VA_ARGS__)
717  #define ILVEV_B2_SH(...) ILVEV_B2(v8i16, __VA_ARGS__)
718  #define ILVEV_B2_SD(...) ILVEV_B2(v2i64, __VA_ARGS__)
719  
720  /* Description : Interleave odd byte elements from vectors
721   * Arguments   : Inputs  - in0, in1, in2, in3
722   *               Outputs - out0, out1
723   *               Return Type - as per RTYPE
724   * Details     : Odd byte elements of 'in0' and 'in1' are interleaved
725   *               and written to 'out0'
726   */
727  #define ILVOD_B2(RTYPE, in0, in1, in2, in3, out0, out1) do {  \
728    out0 = (RTYPE)__msa_ilvod_b((v16i8)in1, (v16i8)in0);        \
729    out1 = (RTYPE)__msa_ilvod_b((v16i8)in3, (v16i8)in2);        \
730  } while (0)
731  #define ILVOD_B2_UB(...) ILVOD_B2(v16u8, __VA_ARGS__)
732  #define ILVOD_B2_SB(...) ILVOD_B2(v16i8, __VA_ARGS__)
733  #define ILVOD_B2_UH(...) ILVOD_B2(v8u16, __VA_ARGS__)
734  #define ILVOD_B2_SH(...) ILVOD_B2(v8i16, __VA_ARGS__)
735  #define ILVOD_B2_SD(...) ILVOD_B2(v2i64, __VA_ARGS__)
736  
737  /* Description : Interleave even halfword elements from vectors
738   * Arguments   : Inputs  - in0, in1, in2, in3
739   *               Outputs - out0, out1
740   *               Return Type - as per RTYPE
741   * Details     : Even halfword elements of 'in0' and 'in1' are interleaved
742   *               and written to 'out0'
743   */
744  #define ILVEV_H2(RTYPE, in0, in1, in2, in3, out0, out1) do {  \
745    out0 = (RTYPE)__msa_ilvev_h((v8i16)in1, (v8i16)in0);        \
746    out1 = (RTYPE)__msa_ilvev_h((v8i16)in3, (v8i16)in2);        \
747  } while (0)
748  #define ILVEV_H2_UB(...) ILVEV_H2(v16u8, __VA_ARGS__)
749  #define ILVEV_H2_UH(...) ILVEV_H2(v8u16, __VA_ARGS__)
750  #define ILVEV_H2_SH(...) ILVEV_H2(v8i16, __VA_ARGS__)
751  #define ILVEV_H2_SW(...) ILVEV_H2(v4i32, __VA_ARGS__)
752  
753  /* Description : Interleave odd halfword elements from vectors
754   * Arguments   : Inputs  - in0, in1, in2, in3
755   *               Outputs - out0, out1
756   *               Return Type - as per RTYPE
757   * Details     : Odd halfword elements of 'in0' and 'in1' are interleaved
758   *               and written to 'out0'
759   */
760  #define ILVOD_H2(RTYPE, in0, in1, in2, in3, out0, out1) do {  \
761    out0 = (RTYPE)__msa_ilvod_h((v8i16)in1, (v8i16)in0);        \
762    out1 = (RTYPE)__msa_ilvod_h((v8i16)in3, (v8i16)in2);        \
763  } while (0)
764  #define ILVOD_H2_UB(...) ILVOD_H2(v16u8, __VA_ARGS__)
765  #define ILVOD_H2_UH(...) ILVOD_H2(v8u16, __VA_ARGS__)
766  #define ILVOD_H2_SH(...) ILVOD_H2(v8i16, __VA_ARGS__)
767  #define ILVOD_H2_SW(...) ILVOD_H2(v4i32, __VA_ARGS__)
768  
769  /* Description : Interleave even word elements from vectors
770   * Arguments   : Inputs  - in0, in1, in2, in3
771   *               Outputs - out0, out1
772   *               Return Type - as per RTYPE
773   * Details     : Even word elements of 'in0' and 'in1' are interleaved
774   *               and written to 'out0'
775   */
776  #define ILVEV_W2(RTYPE, in0, in1, in2, in3, out0, out1) do {  \
777    out0 = (RTYPE)__msa_ilvev_w((v4i32)in1, (v4i32)in0);        \
778    out1 = (RTYPE)__msa_ilvev_w((v4i32)in3, (v4i32)in2);        \
779  } while (0)
780  #define ILVEV_W2_UB(...) ILVEV_W2(v16u8, __VA_ARGS__)
781  #define ILVEV_W2_SB(...) ILVEV_W2(v16i8, __VA_ARGS__)
782  #define ILVEV_W2_UH(...) ILVEV_W2(v8u16, __VA_ARGS__)
783  #define ILVEV_W2_SD(...) ILVEV_W2(v2i64, __VA_ARGS__)
784  
785  /* Description : Interleave even-odd word elements from vectors
786   * Arguments   : Inputs  - in0, in1, in2, in3
787   *               Outputs - out0, out1
788   *               Return Type - as per RTYPE
789   * Details     : Even word elements of 'in0' and 'in1' are interleaved
790   *               and written to 'out0'
791   *               Odd word elements of 'in2' and 'in3' are interleaved
792   *               and written to 'out1'
793   */
794  #define ILVEVOD_W2(RTYPE, in0, in1, in2, in3, out0, out1) do {  \
795    out0 = (RTYPE)__msa_ilvev_w((v4i32)in1, (v4i32)in0);          \
796    out1 = (RTYPE)__msa_ilvod_w((v4i32)in3, (v4i32)in2);          \
797  } while (0)
798  #define ILVEVOD_W2_UB(...) ILVEVOD_W2(v16u8, __VA_ARGS__)
799  #define ILVEVOD_W2_UH(...) ILVEVOD_W2(v8u16, __VA_ARGS__)
800  #define ILVEVOD_W2_SH(...) ILVEVOD_W2(v8i16, __VA_ARGS__)
801  #define ILVEVOD_W2_SW(...) ILVEVOD_W2(v4i32, __VA_ARGS__)
802  
803  /* Description : Interleave even-odd half-word elements from vectors
804   * Arguments   : Inputs  - in0, in1, in2, in3
805   *               Outputs - out0, out1
806   *               Return Type - as per RTYPE
807   * Details     : Even half-word elements of 'in0' and 'in1' are interleaved
808   *               and written to 'out0'
809   *               Odd half-word elements of 'in2' and 'in3' are interleaved
810   *               and written to 'out1'
811   */
812  #define ILVEVOD_H2(RTYPE, in0, in1, in2, in3, out0, out1) do {  \
813    out0 = (RTYPE)__msa_ilvev_h((v8i16)in1, (v8i16)in0);          \
814    out1 = (RTYPE)__msa_ilvod_h((v8i16)in3, (v8i16)in2);          \
815  } while (0)
816  #define ILVEVOD_H2_UB(...) ILVEVOD_H2(v16u8, __VA_ARGS__)
817  #define ILVEVOD_H2_UH(...) ILVEVOD_H2(v8u16, __VA_ARGS__)
818  #define ILVEVOD_H2_SH(...) ILVEVOD_H2(v8i16, __VA_ARGS__)
819  #define ILVEVOD_H2_SW(...) ILVEVOD_H2(v4i32, __VA_ARGS__)
820  
821  /* Description : Interleave even double word elements from vectors
822   * Arguments   : Inputs  - in0, in1, in2, in3
823   *               Outputs - out0, out1
824   *               Return Type - as per RTYPE
825   * Details     : Even double word elements of 'in0' and 'in1' are interleaved
826   *               and written to 'out0'
827   */
828  #define ILVEV_D2(RTYPE, in0, in1, in2, in3, out0, out1) do {  \
829    out0 = (RTYPE)__msa_ilvev_d((v2i64)in1, (v2i64)in0);        \
830    out1 = (RTYPE)__msa_ilvev_d((v2i64)in3, (v2i64)in2);        \
831  } while (0)
832  #define ILVEV_D2_UB(...) ILVEV_D2(v16u8, __VA_ARGS__)
833  #define ILVEV_D2_SB(...) ILVEV_D2(v16i8, __VA_ARGS__)
834  #define ILVEV_D2_SW(...) ILVEV_D2(v4i32, __VA_ARGS__)
835  #define ILVEV_D2_SD(...) ILVEV_D2(v2i64, __VA_ARGS__)
836  
837  /* Description : Interleave left half of byte elements from vectors
838   * Arguments   : Inputs  - in0, in1, in2, in3
839   *               Outputs - out0, out1
840   *               Return Type - as per RTYPE
841   * Details     : Left half of byte elements of 'in0' and 'in1' are interleaved
842   *               and written to 'out0'.
843   */
844  #define ILVL_B2(RTYPE, in0, in1, in2, in3, out0, out1) do {  \
845    out0 = (RTYPE)__msa_ilvl_b((v16i8)in0, (v16i8)in1);        \
846    out1 = (RTYPE)__msa_ilvl_b((v16i8)in2, (v16i8)in3);        \
847  } while (0)
848  #define ILVL_B2_UB(...) ILVL_B2(v16u8, __VA_ARGS__)
849  #define ILVL_B2_SB(...) ILVL_B2(v16i8, __VA_ARGS__)
850  #define ILVL_B2_UH(...) ILVL_B2(v8u16, __VA_ARGS__)
851  #define ILVL_B2_SH(...) ILVL_B2(v8i16, __VA_ARGS__)
852  #define ILVL_B2_SW(...) ILVL_B2(v4i32, __VA_ARGS__)
853  
854  /* Description : Interleave right half of byte elements from vectors
855   * Arguments   : Inputs  - in0, in1, in2, in3
856   *               Outputs - out0, out1
857   *               Return Type - as per RTYPE
858   * Details     : Right half of byte elements of 'in0' and 'in1' are interleaved
859   *               and written to out0.
860   */
861  #define ILVR_B2(RTYPE, in0, in1, in2, in3, out0, out1) do {  \
862    out0 = (RTYPE)__msa_ilvr_b((v16i8)in0, (v16i8)in1);        \
863    out1 = (RTYPE)__msa_ilvr_b((v16i8)in2, (v16i8)in3);        \
864  } while (0)
865  #define ILVR_B2_UB(...) ILVR_B2(v16u8, __VA_ARGS__)
866  #define ILVR_B2_SB(...) ILVR_B2(v16i8, __VA_ARGS__)
867  #define ILVR_B2_UH(...) ILVR_B2(v8u16, __VA_ARGS__)
868  #define ILVR_B2_SH(...) ILVR_B2(v8i16, __VA_ARGS__)
869  #define ILVR_B2_SW(...) ILVR_B2(v4i32, __VA_ARGS__)
870  
871  #define ILVR_B4(RTYPE, in0, in1, in2, in3, in4, in5, in6, in7,  \
872                  out0, out1, out2, out3) do {                    \
873    ILVR_B2(RTYPE, in0, in1, in2, in3, out0, out1);               \
874    ILVR_B2(RTYPE, in4, in5, in6, in7, out2, out3);               \
875  } while (0)
876  #define ILVR_B4_UB(...) ILVR_B4(v16u8, __VA_ARGS__)
877  #define ILVR_B4_SB(...) ILVR_B4(v16i8, __VA_ARGS__)
878  #define ILVR_B4_UH(...) ILVR_B4(v8u16, __VA_ARGS__)
879  #define ILVR_B4_SH(...) ILVR_B4(v8i16, __VA_ARGS__)
880  #define ILVR_B4_SW(...) ILVR_B4(v4i32, __VA_ARGS__)
881  
882  /* Description : Interleave right half of halfword elements from vectors
883   * Arguments   : Inputs  - in0, in1, in2, in3
884   *               Outputs - out0, out1
885   *               Return Type - as per RTYPE
886   * Details     : Right half of halfword elements of 'in0' and 'in1' are
887   *               interleaved and written to 'out0'.
888   */
889  #define ILVR_H2(RTYPE, in0, in1, in2, in3, out0, out1) do {  \
890    out0 = (RTYPE)__msa_ilvr_h((v8i16)in0, (v8i16)in1);        \
891    out1 = (RTYPE)__msa_ilvr_h((v8i16)in2, (v8i16)in3);        \
892  } while (0)
893  #define ILVR_H2_UB(...) ILVR_H2(v16u8, __VA_ARGS__)
894  #define ILVR_H2_SH(...) ILVR_H2(v8i16, __VA_ARGS__)
895  #define ILVR_H2_SW(...) ILVR_H2(v4i32, __VA_ARGS__)
896  
897  #define ILVR_H4(RTYPE, in0, in1, in2, in3, in4, in5, in6, in7,  \
898                  out0, out1, out2, out3) do {                    \
899    ILVR_H2(RTYPE, in0, in1, in2, in3, out0, out1);               \
900    ILVR_H2(RTYPE, in4, in5, in6, in7, out2, out3);               \
901  } while (0)
902  #define ILVR_H4_UB(...) ILVR_H4(v16u8, __VA_ARGS__)
903  #define ILVR_H4_SH(...) ILVR_H4(v8i16, __VA_ARGS__)
904  #define ILVR_H4_SW(...) ILVR_H4(v4i32, __VA_ARGS__)
905  
906  /* Description : Interleave right half of double word elements from vectors
907   * Arguments   : Inputs  - in0, in1, in2, in3
908   *               Outputs - out0, out1
909   *               Return Type - as per RTYPE
910   * Details     : Right half of double word elements of 'in0' and 'in1' are
911   *               interleaved and written to 'out0'.
912   */
913  #define ILVR_D2(RTYPE, in0, in1, in2, in3, out0, out1) do {  \
914    out0 = (RTYPE)__msa_ilvr_d((v2i64)in0, (v2i64)in1);        \
915    out1 = (RTYPE)__msa_ilvr_d((v2i64)in2, (v2i64)in3);        \
916  } while (0)
917  #define ILVR_D2_UB(...) ILVR_D2(v16u8, __VA_ARGS__)
918  #define ILVR_D2_SB(...) ILVR_D2(v16i8, __VA_ARGS__)
919  #define ILVR_D2_SH(...) ILVR_D2(v8i16, __VA_ARGS__)
920  
921  #define ILVR_D4(RTYPE, in0, in1, in2, in3, in4, in5, in6, in7,  \
922                  out0, out1, out2, out3) do {                    \
923    ILVR_D2(RTYPE, in0, in1, in2, in3, out0, out1);               \
924    ILVR_D2(RTYPE, in4, in5, in6, in7, out2, out3);               \
925  } while (0)
926  #define ILVR_D4_SB(...) ILVR_D4(v16i8, __VA_ARGS__)
927  #define ILVR_D4_UB(...) ILVR_D4(v16u8, __VA_ARGS__)
928  
929  /* Description : Interleave both left and right half of input vectors
930   * Arguments   : Inputs  - in0, in1
931   *               Outputs - out0, out1
932   *               Return Type - as per RTYPE
933   * Details     : Right half of byte elements from 'in0' and 'in1' are
934   *               interleaved and written to 'out0'
935   */
936  #define ILVRL_B2(RTYPE, in0, in1, out0, out1) do {     \
937    out0 = (RTYPE)__msa_ilvr_b((v16i8)in0, (v16i8)in1);  \
938    out1 = (RTYPE)__msa_ilvl_b((v16i8)in0, (v16i8)in1);  \
939  } while (0)
940  #define ILVRL_B2_UB(...) ILVRL_B2(v16u8, __VA_ARGS__)
941  #define ILVRL_B2_SB(...) ILVRL_B2(v16i8, __VA_ARGS__)
942  #define ILVRL_B2_UH(...) ILVRL_B2(v8u16, __VA_ARGS__)
943  #define ILVRL_B2_SH(...) ILVRL_B2(v8i16, __VA_ARGS__)
944  #define ILVRL_B2_SW(...) ILVRL_B2(v4i32, __VA_ARGS__)
945  
946  #define ILVRL_H2(RTYPE, in0, in1, out0, out1) do {     \
947    out0 = (RTYPE)__msa_ilvr_h((v8i16)in0, (v8i16)in1);  \
948    out1 = (RTYPE)__msa_ilvl_h((v8i16)in0, (v8i16)in1);  \
949  } while (0)
950  #define ILVRL_H2_UB(...) ILVRL_H2(v16u8, __VA_ARGS__)
951  #define ILVRL_H2_SB(...) ILVRL_H2(v16i8, __VA_ARGS__)
952  #define ILVRL_H2_SH(...) ILVRL_H2(v8i16, __VA_ARGS__)
953  #define ILVRL_H2_SW(...) ILVRL_H2(v4i32, __VA_ARGS__)
954  #define ILVRL_H2_UW(...) ILVRL_H2(v4u32, __VA_ARGS__)
955  
956  #define ILVRL_W2(RTYPE, in0, in1, out0, out1) do {     \
957    out0 = (RTYPE)__msa_ilvr_w((v4i32)in0, (v4i32)in1);  \
958    out1 = (RTYPE)__msa_ilvl_w((v4i32)in0, (v4i32)in1);  \
959  } while (0)
960  #define ILVRL_W2_UB(...) ILVRL_W2(v16u8, __VA_ARGS__)
961  #define ILVRL_W2_SH(...) ILVRL_W2(v8i16, __VA_ARGS__)
962  #define ILVRL_W2_SW(...) ILVRL_W2(v4i32, __VA_ARGS__)
963  #define ILVRL_W2_UW(...) ILVRL_W2(v4u32, __VA_ARGS__)
964  
965  /* Description : Pack even byte elements of vector pairs
966   *  Arguments   : Inputs  - in0, in1, in2, in3
967   *                Outputs - out0, out1
968   *                Return Type - as per RTYPE
969   *  Details     : Even byte elements of 'in0' are copied to the left half of
970   *                'out0' & even byte elements of 'in1' are copied to the right
971   *                half of 'out0'.
972   */
973  #define PCKEV_B2(RTYPE, in0, in1, in2, in3, out0, out1) do {  \
974    out0 = (RTYPE)__msa_pckev_b((v16i8)in0, (v16i8)in1);        \
975    out1 = (RTYPE)__msa_pckev_b((v16i8)in2, (v16i8)in3);        \
976  } while (0)
977  #define PCKEV_B2_SB(...) PCKEV_B2(v16i8, __VA_ARGS__)
978  #define PCKEV_B2_UB(...) PCKEV_B2(v16u8, __VA_ARGS__)
979  #define PCKEV_B2_SH(...) PCKEV_B2(v8i16, __VA_ARGS__)
980  #define PCKEV_B2_SW(...) PCKEV_B2(v4i32, __VA_ARGS__)
981  
982  #define PCKEV_B4(RTYPE, in0, in1, in2, in3, in4, in5, in6, in7,  \
983                   out0, out1, out2, out3) do {                    \
984    PCKEV_B2(RTYPE, in0, in1, in2, in3, out0, out1);               \
985    PCKEV_B2(RTYPE, in4, in5, in6, in7, out2, out3);               \
986  } while (0)
987  #define PCKEV_B4_SB(...) PCKEV_B4(v16i8, __VA_ARGS__)
988  #define PCKEV_B4_UB(...) PCKEV_B4(v16u8, __VA_ARGS__)
989  #define PCKEV_B4_SH(...) PCKEV_B4(v8i16, __VA_ARGS__)
990  #define PCKEV_B4_SW(...) PCKEV_B4(v4i32, __VA_ARGS__)
991  
992  /* Description : Pack even halfword elements of vector pairs
993   * Arguments   : Inputs  - in0, in1, in2, in3
994   *               Outputs - out0, out1
995   *               Return Type - as per RTYPE
996   * Details     : Even halfword elements of 'in0' are copied to the left half of
997   *               'out0' & even halfword elements of 'in1' are copied to the
998   *               right half of 'out0'.
999   */
1000  #define PCKEV_H2(RTYPE, in0, in1, in2, in3, out0, out1) do {  \
1001    out0 = (RTYPE)__msa_pckev_h((v8i16)in0, (v8i16)in1);        \
1002    out1 = (RTYPE)__msa_pckev_h((v8i16)in2, (v8i16)in3);        \
1003  } while (0)
1004  #define PCKEV_H2_UH(...) PCKEV_H2(v8u16, __VA_ARGS__)
1005  #define PCKEV_H2_SH(...) PCKEV_H2(v8i16, __VA_ARGS__)
1006  #define PCKEV_H2_SW(...) PCKEV_H2(v4i32, __VA_ARGS__)
1007  #define PCKEV_H2_UW(...) PCKEV_H2(v4u32, __VA_ARGS__)
1008  
1009  /* Description : Pack even word elements of vector pairs
1010   * Arguments   : Inputs  - in0, in1, in2, in3
1011   *               Outputs - out0, out1
1012   *               Return Type - as per RTYPE
1013   * Details     : Even word elements of 'in0' are copied to the left half of
1014   *               'out0' & even word elements of 'in1' are copied to the
1015   *               right half of 'out0'.
1016   */
1017  #define PCKEV_W2(RTYPE, in0, in1, in2, in3, out0, out1) do {  \
1018    out0 = (RTYPE)__msa_pckev_w((v4i32)in0, (v4i32)in1);        \
1019    out1 = (RTYPE)__msa_pckev_w((v4i32)in2, (v4i32)in3);        \
1020  } while (0)
1021  #define PCKEV_W2_UH(...) PCKEV_W2(v8u16, __VA_ARGS__)
1022  #define PCKEV_W2_SH(...) PCKEV_W2(v8i16, __VA_ARGS__)
1023  #define PCKEV_W2_SW(...) PCKEV_W2(v4i32, __VA_ARGS__)
1024  #define PCKEV_W2_UW(...) PCKEV_W2(v4u32, __VA_ARGS__)
1025  
1026  /* Description : Pack odd halfword elements of vector pairs
1027   * Arguments   : Inputs  - in0, in1, in2, in3
1028   *               Outputs - out0, out1
1029   *               Return Type - as per RTYPE
1030   * Details     : Odd halfword elements of 'in0' are copied to the left half of
1031   *               'out0' & odd halfword elements of 'in1' are copied to the
1032   *               right half of 'out0'.
1033   */
1034  #define PCKOD_H2(RTYPE, in0, in1, in2, in3, out0, out1) do {  \
1035    out0 = (RTYPE)__msa_pckod_h((v8i16)in0, (v8i16)in1);        \
1036    out1 = (RTYPE)__msa_pckod_h((v8i16)in2, (v8i16)in3);        \
1037  } while (0)
1038  #define PCKOD_H2_UH(...) PCKOD_H2(v8u16, __VA_ARGS__)
1039  #define PCKOD_H2_SH(...) PCKOD_H2(v8i16, __VA_ARGS__)
1040  #define PCKOD_H2_SW(...) PCKOD_H2(v4i32, __VA_ARGS__)
1041  #define PCKOD_H2_UW(...) PCKOD_H2(v4u32, __VA_ARGS__)
1042  
1043  /* Description : Arithmetic immediate shift right all elements of word vector
1044   * Arguments   : Inputs  - in0, in1, shift
1045   *               Outputs - in place operation
1046   *               Return Type - as per input vector RTYPE
1047   * Details     : Each element of vector 'in0' is right shifted by 'shift' and
1048   *               the result is written in-place. 'shift' is a GP variable.
1049   */
1050  #define SRAI_W2(RTYPE, in0, in1, shift_val) do {  \
1051    in0 = (RTYPE)SRAI_W(in0, shift_val);            \
1052    in1 = (RTYPE)SRAI_W(in1, shift_val);            \
1053  } while (0)
1054  #define SRAI_W2_SW(...) SRAI_W2(v4i32, __VA_ARGS__)
1055  #define SRAI_W2_UW(...) SRAI_W2(v4u32, __VA_ARGS__)
1056  
1057  #define SRAI_W4(RTYPE, in0, in1, in2, in3, shift_val) do {  \
1058    SRAI_W2(RTYPE, in0, in1, shift_val);                      \
1059    SRAI_W2(RTYPE, in2, in3, shift_val);                      \
1060  } while (0)
1061  #define SRAI_W4_SW(...) SRAI_W4(v4i32, __VA_ARGS__)
1062  #define SRAI_W4_UW(...) SRAI_W4(v4u32, __VA_ARGS__)
1063  
1064  /* Description : Arithmetic shift right all elements of half-word vector
1065   * Arguments   : Inputs  - in0, in1, shift
1066   *               Outputs - in place operation
1067   *               Return Type - as per input vector RTYPE
1068   * Details     : Each element of vector 'in0' is right shifted by 'shift' and
1069   *               the result is written in-place. 'shift' is a GP variable.
1070   */
1071  #define SRAI_H2(RTYPE, in0, in1, shift_val) do {  \
1072    in0 = (RTYPE)SRAI_H(in0, shift_val);            \
1073    in1 = (RTYPE)SRAI_H(in1, shift_val);            \
1074  } while (0)
1075  #define SRAI_H2_SH(...) SRAI_H2(v8i16, __VA_ARGS__)
1076  #define SRAI_H2_UH(...) SRAI_H2(v8u16, __VA_ARGS__)
1077  
1078  /* Description : Arithmetic rounded shift right all elements of word vector
1079   * Arguments   : Inputs  - in0, in1, shift
1080   *               Outputs - in place operation
1081   *               Return Type - as per input vector RTYPE
1082   * Details     : Each element of vector 'in0' is right shifted by 'shift' and
1083   *               the result is written in-place. 'shift' is a GP variable.
1084   */
1085  #define SRARI_W2(RTYPE, in0, in1, shift) do {     \
1086    in0 = (RTYPE)__msa_srari_w((v4i32)in0, shift);  \
1087    in1 = (RTYPE)__msa_srari_w((v4i32)in1, shift);  \
1088  } while (0)
1089  #define SRARI_W2_SW(...) SRARI_W2(v4i32, __VA_ARGS__)
1090  
1091  #define SRARI_W4(RTYPE, in0, in1, in2, in3, shift) do {  \
1092    SRARI_W2(RTYPE, in0, in1, shift);                      \
1093    SRARI_W2(RTYPE, in2, in3, shift);                      \
1094  } while (0)
1095  #define SRARI_W4_SH(...) SRARI_W4(v8i16, __VA_ARGS__)
1096  #define SRARI_W4_UW(...) SRARI_W4(v4u32, __VA_ARGS__)
1097  #define SRARI_W4_SW(...) SRARI_W4(v4i32, __VA_ARGS__)
1098  
1099  /* Description : Shift right arithmetic rounded double words
1100   * Arguments   : Inputs  - in0, in1, shift
1101   *               Outputs - in place operation
1102   *               Return Type - as per RTYPE
1103   * Details     : Each element of vector 'in0' is shifted right arithmetically by
1104   *               the number of bits in the corresponding element in the vector
1105   *               'shift'. The last discarded bit is added to shifted value for
1106   *               rounding and the result is written in-place.
1107   *               'shift' is a vector.
1108   */
1109  #define SRAR_D2(RTYPE, in0, in1, shift) do {            \
1110    in0 = (RTYPE)__msa_srar_d((v2i64)in0, (v2i64)shift);  \
1111    in1 = (RTYPE)__msa_srar_d((v2i64)in1, (v2i64)shift);  \
1112  } while (0)
1113  #define SRAR_D2_SW(...) SRAR_D2(v4i32, __VA_ARGS__)
1114  #define SRAR_D2_SD(...) SRAR_D2(v2i64, __VA_ARGS__)
1115  #define SRAR_D2_UD(...) SRAR_D2(v2u64, __VA_ARGS__)
1116  
1117  #define SRAR_D4(RTYPE, in0, in1, in2, in3, shift) do {  \
1118    SRAR_D2(RTYPE, in0, in1, shift);                      \
1119    SRAR_D2(RTYPE, in2, in3, shift);                      \
1120  } while (0)
1121  #define SRAR_D4_SD(...) SRAR_D4(v2i64, __VA_ARGS__)
1122  #define SRAR_D4_UD(...) SRAR_D4(v2u64, __VA_ARGS__)
1123  
1124  /* Description : Addition of 2 pairs of half-word vectors
1125   * Arguments   : Inputs  - in0, in1, in2, in3
1126   *               Outputs - out0, out1
1127   * Details     : Each element in 'in0' is added to 'in1' and result is written
1128   *               to 'out0'.
1129   */
1130  #define ADDVI_H2(RTYPE, in0, in1, in2, in3, out0, out1) do {  \
1131    out0 = (RTYPE)ADDVI_H(in0, in1);                            \
1132    out1 = (RTYPE)ADDVI_H(in2, in3);                            \
1133  } while (0)
1134  #define ADDVI_H2_SH(...) ADDVI_H2(v8i16, __VA_ARGS__)
1135  #define ADDVI_H2_UH(...) ADDVI_H2(v8u16, __VA_ARGS__)
1136  
1137  /* Description : Addition of 2 pairs of word vectors
1138   * Arguments   : Inputs  - in0, in1, in2, in3
1139   *               Outputs - out0, out1
1140   * Details     : Each element in 'in0' is added to 'in1' and result is written
1141   *               to 'out0'.
1142   */
1143  #define ADDVI_W2(RTYPE, in0, in1, in2, in3, out0, out1) do {  \
1144    out0 = (RTYPE)ADDVI_W(in0, in1);                            \
1145    out1 = (RTYPE)ADDVI_W(in2, in3);                            \
1146  } while (0)
1147  #define ADDVI_W2_SW(...) ADDVI_W2(v4i32, __VA_ARGS__)
1148  
1149  /* Description : Fill 2 pairs of word vectors with GP registers
1150   * Arguments   : Inputs  - in0, in1
1151   *               Outputs - out0, out1
1152   * Details     : GP register in0 is replicated in each word element of out0
1153   *               GP register in1 is replicated in each word element of out1
1154   */
1155  #define FILL_W2(RTYPE, in0, in1, out0, out1) do {  \
1156    out0 = (RTYPE)__msa_fill_w(in0);                 \
1157    out1 = (RTYPE)__msa_fill_w(in1);                 \
1158  } while (0)
1159  #define FILL_W2_SW(...) FILL_W2(v4i32, __VA_ARGS__)
1160  
1161  /* Description : Addition of 2 pairs of vectors
1162   * Arguments   : Inputs  - in0, in1, in2, in3
1163   *               Outputs - out0, out1
1164   * Details     : Each element in 'in0' is added to 'in1' and result is written
1165   *               to 'out0'.
1166   */
1167  #define ADD2(in0, in1, in2, in3, out0, out1) do {  \
1168    out0 = in0 + in1;                                \
1169    out1 = in2 + in3;                                \
1170  } while (0)
1171  
1172  #define ADD4(in0, in1, in2, in3, in4, in5, in6, in7,  \
1173               out0, out1, out2, out3) do {             \
1174    ADD2(in0, in1, in2, in3, out0, out1);               \
1175    ADD2(in4, in5, in6, in7, out2, out3);               \
1176  } while (0)
1177  
1178  /* Description : Subtraction of 2 pairs of vectors
1179   * Arguments   : Inputs  - in0, in1, in2, in3
1180   *               Outputs - out0, out1
1181   * Details     : Each element in 'in1' is subtracted from 'in0' and result is
1182   *               written to 'out0'.
1183   */
1184  #define SUB2(in0, in1, in2, in3, out0, out1) do {  \
1185    out0 = in0 - in1;                                \
1186    out1 = in2 - in3;                                \
1187  } while (0)
1188  
1189  #define SUB3(in0, in1, in2, in3, in4, in5, out0, out1, out2) do {  \
1190    out0 = in0 - in1;                                                \
1191    out1 = in2 - in3;                                                \
1192    out2 = in4 - in5;                                                \
1193  } while (0)
1194  
1195  #define SUB4(in0, in1, in2, in3, in4, in5, in6, in7,  \
1196               out0, out1, out2, out3) do {             \
1197    out0 = in0 - in1;                                   \
1198    out1 = in2 - in3;                                   \
1199    out2 = in4 - in5;                                   \
1200    out3 = in6 - in7;                                   \
1201  } while (0)
1202  
1203  /* Description : Addition - Subtraction of input vectors
1204   * Arguments   : Inputs  - in0, in1
1205   *               Outputs - out0, out1
1206   * Details     : Each element in 'in1' is added to 'in0' and result is
1207   *               written to 'out0'.
1208   *               Each element in 'in1' is subtracted from 'in0' and result is
1209   *               written to 'out1'.
1210   */
1211  #define ADDSUB2(in0, in1, out0, out1) do {  \
1212    out0 = in0 + in1;                         \
1213    out1 = in0 - in1;                         \
1214  } while (0)
1215  
1216  /* Description : Multiplication of pairs of vectors
1217   * Arguments   : Inputs  - in0, in1, in2, in3
1218   *               Outputs - out0, out1
1219   * Details     : Each element from 'in0' is multiplied with elements from 'in1'
1220   *               and the result is written to 'out0'
1221   */
1222  #define MUL2(in0, in1, in2, in3, out0, out1) do {  \
1223    out0 = in0 * in1;                                \
1224    out1 = in2 * in3;                                \
1225  } while (0)
1226  
1227  #define MUL4(in0, in1, in2, in3, in4, in5, in6, in7,  \
1228               out0, out1, out2, out3) do {             \
1229    MUL2(in0, in1, in2, in3, out0, out1);               \
1230    MUL2(in4, in5, in6, in7, out2, out3);               \
1231  } while (0)
1232  
1233  /* Description : Sign extend halfword elements from right half of the vector
1234   * Arguments   : Input  - in    (halfword vector)
1235   *               Output - out   (sign extended word vector)
1236   *               Return Type - signed word
1237   * Details     : Sign bit of halfword elements from input vector 'in' is
1238   *               extracted and interleaved with same vector 'in0' to generate
1239   *               4 word elements keeping sign intact
1240   */
1241  #define UNPCK_R_SH_SW(in, out) do {                   \
1242    const v8i16 sign_m = __msa_clti_s_h((v8i16)in, 0);  \
1243    out = (v4i32)__msa_ilvr_h(sign_m, (v8i16)in);       \
1244  } while (0)
1245  
1246  /* Description : Sign extend halfword elements from input vector and return
1247   *               the result in pair of vectors
1248   * Arguments   : Input   - in            (halfword vector)
1249   *               Outputs - out0, out1   (sign extended word vectors)
1250   *               Return Type - signed word
1251   * Details     : Sign bit of halfword elements from input vector 'in' is
1252   *               extracted and interleaved right with same vector 'in0' to
1253   *               generate 4 signed word elements in 'out0'
1254   *               Then interleaved left with same vector 'in0' to
1255   *               generate 4 signed word elements in 'out1'
1256   */
1257  #define UNPCK_SH_SW(in, out0, out1) do {              \
1258    const v8i16 tmp_m = __msa_clti_s_h((v8i16)in, 0);   \
1259    ILVRL_H2_SW(tmp_m, in, out0, out1);                 \
1260  } while (0)
1261  
1262  /* Description : Butterfly of 4 input vectors
1263   * Arguments   : Inputs  - in0, in1, in2, in3
1264   *               Outputs - out0, out1, out2, out3
1265   * Details     : Butterfly operation
1266   */
1267  #define BUTTERFLY_4(in0, in1, in2, in3, out0, out1, out2, out3) do {  \
1268    out0 = in0 + in3;                                                   \
1269    out1 = in1 + in2;                                                   \
1270    out2 = in1 - in2;                                                   \
1271    out3 = in0 - in3;                                                   \
1272  } while (0)
1273  
1274  /* Description : Transpose 16x4 block into 4x16 with byte elements in vectors
1275   * Arguments   : Inputs  - in0, in1, in2, in3, in4, in5, in6, in7,
1276   *                         in8, in9, in10, in11, in12, in13, in14, in15
1277   *               Outputs - out0, out1, out2, out3
1278   *               Return Type - unsigned byte
1279   */
1280  #define TRANSPOSE16x4_UB_UB(in0, in1, in2, in3, in4, in5, in6, in7,        \
1281                              in8, in9, in10, in11, in12, in13, in14, in15,  \
1282                              out0, out1, out2, out3) do {                   \
1283    v2i64 tmp0_m, tmp1_m, tmp2_m, tmp3_m, tmp4_m, tmp5_m;                    \
1284    ILVEV_W2_SD(in0, in4, in8, in12, tmp2_m, tmp3_m);                        \
1285    ILVEV_W2_SD(in1, in5, in9, in13, tmp0_m, tmp1_m);                        \
1286    ILVEV_D2_UB(tmp2_m, tmp3_m, tmp0_m, tmp1_m, out1, out3);                 \
1287    ILVEV_W2_SD(in2, in6, in10, in14, tmp4_m, tmp5_m);                       \
1288    ILVEV_W2_SD(in3, in7, in11, in15, tmp0_m, tmp1_m);                       \
1289    ILVEV_D2_SD(tmp4_m, tmp5_m, tmp0_m, tmp1_m, tmp2_m, tmp3_m);             \
1290    ILVEV_B2_SD(out1, out3, tmp2_m, tmp3_m, tmp0_m, tmp1_m);                 \
1291    ILVEVOD_H2_UB(tmp0_m, tmp1_m, tmp0_m, tmp1_m, out0, out2);               \
1292    ILVOD_B2_SD(out1, out3, tmp2_m, tmp3_m, tmp0_m, tmp1_m);                 \
1293    ILVEVOD_H2_UB(tmp0_m, tmp1_m, tmp0_m, tmp1_m, out1, out3);               \
1294  } while (0)
1295  
1296  /* Description : Transpose 16x8 block into 8x16 with byte elements in vectors
1297   * Arguments   : Inputs  - in0, in1, in2, in3, in4, in5, in6, in7,
1298   *                         in8, in9, in10, in11, in12, in13, in14, in15
1299   *               Outputs - out0, out1, out2, out3, out4, out5, out6, out7
1300   *               Return Type - unsigned byte
1301   */
1302  #define TRANSPOSE16x8_UB_UB(in0, in1, in2, in3, in4, in5, in6, in7,        \
1303                              in8, in9, in10, in11, in12, in13, in14, in15,  \
1304                              out0, out1, out2, out3, out4, out5,            \
1305                              out6, out7) do {                               \
1306    v8i16 tmp0_m, tmp1_m, tmp4_m, tmp5_m, tmp6_m, tmp7_m;                    \
1307    v4i32 tmp2_m, tmp3_m;                                                    \
1308    ILVEV_D2_UB(in0, in8, in1, in9, out7, out6);                             \
1309    ILVEV_D2_UB(in2, in10, in3, in11, out5, out4);                           \
1310    ILVEV_D2_UB(in4, in12, in5, in13, out3, out2);                           \
1311    ILVEV_D2_UB(in6, in14, in7, in15, out1, out0);                           \
1312    ILVEV_B2_SH(out7, out6, out5, out4, tmp0_m, tmp1_m);                     \
1313    ILVOD_B2_SH(out7, out6, out5, out4, tmp4_m, tmp5_m);                     \
1314    ILVEV_B2_UB(out3, out2, out1, out0, out5, out7);                         \
1315    ILVOD_B2_SH(out3, out2, out1, out0, tmp6_m, tmp7_m);                     \
1316    ILVEV_H2_SW(tmp0_m, tmp1_m, out5, out7, tmp2_m, tmp3_m);                 \
1317    ILVEVOD_W2_UB(tmp2_m, tmp3_m, tmp2_m, tmp3_m, out0, out4);               \
1318    ILVOD_H2_SW(tmp0_m, tmp1_m, out5, out7, tmp2_m, tmp3_m);                 \
1319    ILVEVOD_W2_UB(tmp2_m, tmp3_m, tmp2_m, tmp3_m, out2, out6);               \
1320    ILVEV_H2_SW(tmp4_m, tmp5_m, tmp6_m, tmp7_m, tmp2_m, tmp3_m);             \
1321    ILVEVOD_W2_UB(tmp2_m, tmp3_m, tmp2_m, tmp3_m, out1, out5);               \
1322    ILVOD_H2_SW(tmp4_m, tmp5_m, tmp6_m, tmp7_m, tmp2_m, tmp3_m);             \
1323    ILVEVOD_W2_UB(tmp2_m, tmp3_m, tmp2_m, tmp3_m, out3, out7);               \
1324  } while (0)
1325  
1326  /* Description : Transpose 4x4 block with word elements in vectors
1327   * Arguments   : Inputs  - in0, in1, in2, in3
1328   *                Outputs - out0, out1, out2, out3
1329   *                Return Type - as per RTYPE
1330   */
1331  #define TRANSPOSE4x4_W(RTYPE, in0, in1, in2, in3,                            \
1332                         out0, out1, out2, out3) do {                          \
1333    v4i32 s0_m, s1_m, s2_m, s3_m;                                              \
1334    ILVRL_W2_SW(in1, in0, s0_m, s1_m);                                         \
1335    ILVRL_W2_SW(in3, in2, s2_m, s3_m);                                         \
1336    out0 = (RTYPE)__msa_ilvr_d((v2i64)s2_m, (v2i64)s0_m);                      \
1337    out1 = (RTYPE)__msa_ilvl_d((v2i64)s2_m, (v2i64)s0_m);                      \
1338    out2 = (RTYPE)__msa_ilvr_d((v2i64)s3_m, (v2i64)s1_m);                      \
1339    out3 = (RTYPE)__msa_ilvl_d((v2i64)s3_m, (v2i64)s1_m);                      \
1340  } while (0)
1341  #define TRANSPOSE4x4_SW_SW(...) TRANSPOSE4x4_W(v4i32, __VA_ARGS__)
1342  
1343  /* Description : Add block 4x4
1344   * Arguments   : Inputs - in0, in1, in2, in3, pdst, stride
1345   * Details     : Least significant 4 bytes from each input vector are added to
1346   *               the destination bytes, clipped between 0-255 and stored.
1347   */
1348  #define ADDBLK_ST4x4_UB(in0, in1, in2, in3, pdst, stride) do {  \
1349    uint32_t src0_m, src1_m, src2_m, src3_m;                      \
1350    v8i16 inp0_m, inp1_m, res0_m, res1_m;                         \
1351    v16i8 dst0_m = { 0 };                                         \
1352    v16i8 dst1_m = { 0 };                                         \
1353    const v16i8 zero_m = { 0 };                                   \
1354    ILVR_D2_SH(in1, in0, in3, in2, inp0_m, inp1_m);               \
1355    LW4(pdst, stride, src0_m, src1_m, src2_m, src3_m);            \
1356    INSERT_W2_SB(src0_m, src1_m, dst0_m);                         \
1357    INSERT_W2_SB(src2_m, src3_m, dst1_m);                         \
1358    ILVR_B2_SH(zero_m, dst0_m, zero_m, dst1_m, res0_m, res1_m);   \
1359    ADD2(res0_m, inp0_m, res1_m, inp1_m, res0_m, res1_m);         \
1360    CLIP_SH2_0_255(res0_m, res1_m);                               \
1361    PCKEV_B2_SB(res0_m, res0_m, res1_m, res1_m, dst0_m, dst1_m);  \
1362    ST4x4_UB(dst0_m, dst1_m, 0, 1, 0, 1, pdst, stride);           \
1363  } while (0)
1364  
1365  /* Description : Pack even byte elements, extract 0 & 2 index words from pair
1366   *               of results and store 4 words in destination memory as per
1367   *               stride
1368   * Arguments   : Inputs - in0, in1, in2, in3, pdst, stride
1369   */
1370  #define PCKEV_ST4x4_UB(in0, in1, in2, in3, pdst, stride) do {  \
1371    v16i8 tmp0_m, tmp1_m;                                        \
1372    PCKEV_B2_SB(in1, in0, in3, in2, tmp0_m, tmp1_m);             \
1373    ST4x4_UB(tmp0_m, tmp1_m, 0, 2, 0, 2, pdst, stride);          \
1374  } while (0)
1375  
1376  /* Description : average with rounding (in0 + in1 + 1) / 2.
1377   * Arguments   : Inputs  - in0, in1, in2, in3,
1378   *               Outputs - out0, out1
1379   *               Return Type - as per RTYPE
1380   * Details     : Each unsigned byte element from 'in0' vector is added with
1381   *               each unsigned byte element from 'in1' vector. Then the average
1382   *               with rounding is calculated and written to 'out0'
1383   */
1384  #define AVER_UB2(RTYPE, in0, in1, in2, in3, out0, out1) do {  \
1385    out0 = (RTYPE)__msa_aver_u_b((v16u8)in0, (v16u8)in1);       \
1386    out1 = (RTYPE)__msa_aver_u_b((v16u8)in2, (v16u8)in3);       \
1387  } while (0)
1388  #define AVER_UB2_UB(...) AVER_UB2(v16u8, __VA_ARGS__)
1389  
1390  #endif  /* WEBP_DSP_MSA_MACRO_H_ */
1391