tor-browser

The Tor Browser
git clone https://git.dasho.dev/tor-browser.git
Log | Files | Refs | README | LICENSE

msa_macro.h (64181B)


      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 "src/dsp/dsp.h"
     18 
     19 #if defined(WEBP_USE_MSA)
     20 
     21 #include <stdint.h>
     22 #include <msa.h>
     23 
     24 #if defined(__clang__)
     25  #define CLANG_BUILD
     26 #endif
     27 
     28 #ifdef CLANG_BUILD
     29  #define ALPHAVAL  (-1)
     30  #define ADDVI_H(a, b)  __msa_addvi_h((v8i16)a, b)
     31  #define ADDVI_W(a, b)  __msa_addvi_w((v4i32)a, b)
     32  #define SRAI_B(a, b)  __msa_srai_b((v16i8)a, b)
     33  #define SRAI_H(a, b)  __msa_srai_h((v8i16)a, b)
     34  #define SRAI_W(a, b)  __msa_srai_w((v4i32)a, b)
     35  #define SRLI_H(a, b)  __msa_srli_h((v8i16)a, b)
     36  #define SLLI_B(a, b)  __msa_slli_b((v4i32)a, b)
     37  #define ANDI_B(a, b)  __msa_andi_b((v16u8)a, b)
     38  #define ORI_B(a, b)   __msa_ori_b((v16u8)a, b)
     39 #else
     40  #define ALPHAVAL  (0xff)
     41  #define ADDVI_H(a, b)  (a + b)
     42  #define ADDVI_W(a, b)  (a + b)
     43  #define SRAI_B(a, b)  (a >> b)
     44  #define SRAI_H(a, b)  (a >> b)
     45  #define SRAI_W(a, b)  (a >> b)
     46  #define SRLI_H(a, b)  (a << b)
     47  #define SLLI_B(a, b)  (a << b)
     48  #define ANDI_B(a, b)  (a & b)
     49  #define ORI_B(a, b)   (a | b)
     50 #endif
     51 
     52 #define LD_B(RTYPE, psrc) *((RTYPE*)(psrc))
     53 #define LD_UB(...) LD_B(v16u8, __VA_ARGS__)
     54 #define LD_SB(...) LD_B(v16i8, __VA_ARGS__)
     55 
     56 #define LD_H(RTYPE, psrc) *((RTYPE*)(psrc))
     57 #define LD_UH(...) LD_H(v8u16, __VA_ARGS__)
     58 #define LD_SH(...) LD_H(v8i16, __VA_ARGS__)
     59 
     60 #define LD_W(RTYPE, psrc) *((RTYPE*)(psrc))
     61 #define LD_UW(...) LD_W(v4u32, __VA_ARGS__)
     62 #define LD_SW(...) LD_W(v4i32, __VA_ARGS__)
     63 
     64 #define ST_B(RTYPE, in, pdst) *((RTYPE*)(pdst)) = in
     65 #define ST_UB(...) ST_B(v16u8, __VA_ARGS__)
     66 #define ST_SB(...) ST_B(v16i8, __VA_ARGS__)
     67 
     68 #define ST_H(RTYPE, in, pdst) *((RTYPE*)(pdst)) = in
     69 #define ST_UH(...) ST_H(v8u16, __VA_ARGS__)
     70 #define ST_SH(...) ST_H(v8i16, __VA_ARGS__)
     71 
     72 #define ST_W(RTYPE, in, pdst) *((RTYPE*)(pdst)) = in
     73 #define ST_UW(...) ST_W(v4u32, __VA_ARGS__)
     74 #define ST_SW(...) ST_W(v4i32, __VA_ARGS__)
     75 
     76 #define MSA_LOAD_FUNC(TYPE, INSTR, FUNC_NAME)               \
     77  static inline TYPE FUNC_NAME(const void* const psrc) {    \
     78    const uint8_t* const psrc_m = (const uint8_t*)psrc;     \
     79    TYPE val_m;                                             \
     80    __asm__ volatile("" #INSTR " %[val_m], %[psrc_m]  \n\t" \
     81                     : [val_m] "=r"(val_m)                  \
     82                     : [psrc_m] "m"(*psrc_m));              \
     83    return val_m;                                           \
     84  }
     85 
     86 #define MSA_LOAD(psrc, FUNC_NAME)  FUNC_NAME(psrc)
     87 
     88 #define MSA_STORE_FUNC(TYPE, INSTR, FUNC_NAME)                 \
     89  static inline void FUNC_NAME(TYPE val, void* const pdst) {   \
     90    uint8_t* const pdst_m = (uint8_t*)pdst;                    \
     91    TYPE val_m = val;                                          \
     92    __asm__ volatile(" " #INSTR "  %[val_m],  %[pdst_m]  \n\t" \
     93                     : [pdst_m] "=m"(*pdst_m)                  \
     94                     : [val_m] "r"(val_m));                    \
     95  }
     96 
     97 #define MSA_STORE(val, pdst, FUNC_NAME)  FUNC_NAME(val, pdst)
     98 
     99 #if (__mips_isa_rev >= 6)
    100  MSA_LOAD_FUNC(uint16_t, lh, msa_lh);
    101  #define LH(psrc)  MSA_LOAD(psrc, msa_lh)
    102  MSA_LOAD_FUNC(uint32_t, lw, msa_lw);
    103  #define LW(psrc)  MSA_LOAD(psrc, msa_lw)
    104  #if (__mips == 64)
    105    MSA_LOAD_FUNC(uint64_t, ld, msa_ld);
    106    #define LD(psrc)  MSA_LOAD(psrc, msa_ld)
    107  #else  // !(__mips == 64)
    108    #define LD(psrc)  ((((uint64_t)MSA_LOAD(psrc + 4, msa_lw)) << 32) | \
    109                       MSA_LOAD(psrc, msa_lw))
    110  #endif  // (__mips == 64)
    111 
    112  MSA_STORE_FUNC(uint16_t, sh, msa_sh);
    113  #define SH(val, pdst)  MSA_STORE(val, pdst, msa_sh)
    114  MSA_STORE_FUNC(uint32_t, sw, msa_sw);
    115  #define SW(val, pdst)  MSA_STORE(val, pdst, msa_sw)
    116  MSA_STORE_FUNC(uint64_t, sd, msa_sd);
    117  #define SD(val, pdst)  MSA_STORE(val, pdst, msa_sd)
    118 #else  // !(__mips_isa_rev >= 6)
    119  MSA_LOAD_FUNC(uint16_t, ulh, msa_ulh);
    120  #define LH(psrc)  MSA_LOAD(psrc, msa_ulh)
    121  MSA_LOAD_FUNC(uint32_t, ulw, msa_ulw);
    122  #define LW(psrc)  MSA_LOAD(psrc, msa_ulw)
    123  #if (__mips == 64)
    124    MSA_LOAD_FUNC(uint64_t, uld, msa_uld);
    125    #define LD(psrc)  MSA_LOAD(psrc, msa_uld)
    126  #else  // !(__mips == 64)
    127    #define LD(psrc)  ((((uint64_t)MSA_LOAD(psrc + 4, msa_ulw)) << 32) | \
    128                        MSA_LOAD(psrc, msa_ulw))
    129  #endif  // (__mips == 64)
    130 
    131  MSA_STORE_FUNC(uint16_t, ush, msa_ush);
    132  #define SH(val, pdst)  MSA_STORE(val, pdst, msa_ush)
    133  MSA_STORE_FUNC(uint32_t, usw, msa_usw);
    134  #define SW(val, pdst)  MSA_STORE(val, pdst, msa_usw)
    135  #define SD(val, pdst) do {                                               \
    136    uint8_t* const pdst_sd_m = (uint8_t*)(pdst);                           \
    137    const uint32_t val0_m = (uint32_t)(val & 0x00000000FFFFFFFF);          \
    138    const uint32_t val1_m = (uint32_t)((val >> 32) & 0x00000000FFFFFFFF);  \
    139    SW(val0_m, pdst_sd_m);                                                 \
    140    SW(val1_m, pdst_sd_m + 4);                                             \
    141  } while (0)
    142 #endif  // (__mips_isa_rev >= 6)
    143 
    144 /* Description : Load 4 words with stride
    145 * Arguments   : Inputs  - psrc, stride
    146 *               Outputs - out0, out1, out2, out3
    147 * Details     : Load word in 'out0' from (psrc)
    148 *               Load word in 'out1' from (psrc + stride)
    149 *               Load word in 'out2' from (psrc + 2 * stride)
    150 *               Load word in 'out3' from (psrc + 3 * stride)
    151 */
    152 #define LW4(psrc, stride, out0, out1, out2, out3) do {  \
    153  const uint8_t* ptmp = (const uint8_t*)psrc;           \
    154  out0 = LW(ptmp);                                      \
    155  ptmp += stride;                                       \
    156  out1 = LW(ptmp);                                      \
    157  ptmp += stride;                                       \
    158  out2 = LW(ptmp);                                      \
    159  ptmp += stride;                                       \
    160  out3 = LW(ptmp);                                      \
    161 } while (0)
    162 
    163 /* Description : Store words with stride
    164 * Arguments   : Inputs - in0, in1, in2, in3, pdst, stride
    165 * Details     : Store word from 'in0' to (pdst)
    166 *               Store word from 'in1' to (pdst + stride)
    167 *               Store word from 'in2' to (pdst + 2 * stride)
    168 *               Store word from 'in3' to (pdst + 3 * stride)
    169 */
    170 #define SW4(in0, in1, in2, in3, pdst, stride) do {  \
    171  uint8_t* ptmp = (uint8_t*)pdst;                   \
    172  SW(in0, ptmp);                                    \
    173  ptmp += stride;                                   \
    174  SW(in1, ptmp);                                    \
    175  ptmp += stride;                                   \
    176  SW(in2, ptmp);                                    \
    177  ptmp += stride;                                   \
    178  SW(in3, ptmp);                                    \
    179 } while (0)
    180 
    181 #define SW3(in0, in1, in2, pdst, stride) do {  \
    182  uint8_t* ptmp = (uint8_t*)pdst;              \
    183  SW(in0, ptmp);                               \
    184  ptmp += stride;                              \
    185  SW(in1, ptmp);                               \
    186  ptmp += stride;                              \
    187  SW(in2, ptmp);                               \
    188 } while (0)
    189 
    190 #define SW2(in0, in1, pdst, stride) do {  \
    191  uint8_t* ptmp = (uint8_t*)pdst;         \
    192  SW(in0, ptmp);                          \
    193  ptmp += stride;                         \
    194  SW(in1, ptmp);                          \
    195 } while (0)
    196 
    197 /* Description : Store 4 double words with stride
    198 * Arguments   : Inputs - in0, in1, in2, in3, pdst, stride
    199 * Details     : Store double word from 'in0' to (pdst)
    200 *               Store double word from 'in1' to (pdst + stride)
    201 *               Store double word from 'in2' to (pdst + 2 * stride)
    202 *               Store double word from 'in3' to (pdst + 3 * stride)
    203 */
    204 #define SD4(in0, in1, in2, in3, pdst, stride) do {  \
    205  uint8_t* ptmp = (uint8_t*)pdst;                   \
    206  SD(in0, ptmp);                                    \
    207  ptmp += stride;                                   \
    208  SD(in1, ptmp);                                    \
    209  ptmp += stride;                                   \
    210  SD(in2, ptmp);                                    \
    211  ptmp += stride;                                   \
    212  SD(in3, ptmp);                                    \
    213 } while (0)
    214 
    215 /* Description : Load vectors with 16 byte elements with stride
    216 * Arguments   : Inputs  - psrc, stride
    217 *               Outputs - out0, out1
    218 *               Return Type - as per RTYPE
    219 * Details     : Load 16 byte elements in 'out0' from (psrc)
    220 *               Load 16 byte elements in 'out1' from (psrc + stride)
    221 */
    222 #define LD_B2(RTYPE, psrc, stride, out0, out1) do {  \
    223  out0 = LD_B(RTYPE, psrc);                          \
    224  out1 = LD_B(RTYPE, psrc + stride);                 \
    225 } while (0)
    226 #define LD_UB2(...) LD_B2(v16u8, __VA_ARGS__)
    227 #define LD_SB2(...) LD_B2(v16i8, __VA_ARGS__)
    228 
    229 #define LD_B3(RTYPE, psrc, stride, out0, out1, out2) do {  \
    230  LD_B2(RTYPE, psrc, stride, out0, out1);                  \
    231  out2 = LD_B(RTYPE, psrc + 2 * stride);                   \
    232 } while (0)
    233 #define LD_UB3(...) LD_B3(v16u8, __VA_ARGS__)
    234 #define LD_SB3(...) LD_B3(v16i8, __VA_ARGS__)
    235 
    236 #define LD_B4(RTYPE, psrc, stride, out0, out1, out2, out3) do {  \
    237  LD_B2(RTYPE, psrc, stride, out0, out1);                        \
    238  LD_B2(RTYPE, psrc + 2 * stride , stride, out2, out3);          \
    239 } while (0)
    240 #define LD_UB4(...) LD_B4(v16u8, __VA_ARGS__)
    241 #define LD_SB4(...) LD_B4(v16i8, __VA_ARGS__)
    242 
    243 #define LD_B8(RTYPE, psrc, stride,                                  \
    244              out0, out1, out2, out3, out4, out5, out6, out7) do {  \
    245  LD_B4(RTYPE, psrc, stride, out0, out1, out2, out3);               \
    246  LD_B4(RTYPE, psrc + 4 * stride, stride, out4, out5, out6, out7);  \
    247 } while (0)
    248 #define LD_UB8(...) LD_B8(v16u8, __VA_ARGS__)
    249 #define LD_SB8(...) LD_B8(v16i8, __VA_ARGS__)
    250 
    251 /* Description : Load vectors with 8 halfword elements with stride
    252 * Arguments   : Inputs  - psrc, stride
    253 *               Outputs - out0, out1
    254 * Details     : Load 8 halfword elements in 'out0' from (psrc)
    255 *               Load 8 halfword elements in 'out1' from (psrc + stride)
    256 */
    257 #define LD_H2(RTYPE, psrc, stride, out0, out1) do {  \
    258  out0 = LD_H(RTYPE, psrc);                          \
    259  out1 = LD_H(RTYPE, psrc + stride);                 \
    260 } while (0)
    261 #define LD_UH2(...) LD_H2(v8u16, __VA_ARGS__)
    262 #define LD_SH2(...) LD_H2(v8i16, __VA_ARGS__)
    263 
    264 /* Description : Load vectors with 4 word elements with stride
    265 * Arguments   : Inputs  - psrc, stride
    266 *               Outputs - out0, out1, out2, out3
    267 * Details     : Load 4 word elements in 'out0' from (psrc + 0 * stride)
    268 *               Load 4 word elements in 'out1' from (psrc + 1 * stride)
    269 *               Load 4 word elements in 'out2' from (psrc + 2 * stride)
    270 *               Load 4 word elements in 'out3' from (psrc + 3 * stride)
    271 */
    272 #define LD_W2(RTYPE, psrc, stride, out0, out1) do {  \
    273  out0 = LD_W(RTYPE, psrc);                          \
    274  out1 = LD_W(RTYPE, psrc + stride);                 \
    275 } while (0)
    276 #define LD_UW2(...) LD_W2(v4u32, __VA_ARGS__)
    277 #define LD_SW2(...) LD_W2(v4i32, __VA_ARGS__)
    278 
    279 #define LD_W3(RTYPE, psrc, stride, out0, out1, out2) do {  \
    280  LD_W2(RTYPE, psrc, stride, out0, out1);                  \
    281  out2 = LD_W(RTYPE, psrc + 2 * stride);                   \
    282 } while (0)
    283 #define LD_UW3(...) LD_W3(v4u32, __VA_ARGS__)
    284 #define LD_SW3(...) LD_W3(v4i32, __VA_ARGS__)
    285 
    286 #define LD_W4(RTYPE, psrc, stride, out0, out1, out2, out3) do {  \
    287  LD_W2(RTYPE, psrc, stride, out0, out1);                        \
    288  LD_W2(RTYPE, psrc + 2 * stride, stride, out2, out3);           \
    289 } while (0)
    290 #define LD_UW4(...) LD_W4(v4u32, __VA_ARGS__)
    291 #define LD_SW4(...) LD_W4(v4i32, __VA_ARGS__)
    292 
    293 /* Description : Store vectors of 16 byte elements with stride
    294 * Arguments   : Inputs - in0, in1, pdst, stride
    295 * Details     : Store 16 byte elements from 'in0' to (pdst)
    296 *               Store 16 byte elements from 'in1' to (pdst + stride)
    297 */
    298 #define ST_B2(RTYPE, in0, in1, pdst, stride) do {  \
    299  ST_B(RTYPE, in0, pdst);                          \
    300  ST_B(RTYPE, in1, pdst + stride);                 \
    301 } while (0)
    302 #define ST_UB2(...) ST_B2(v16u8, __VA_ARGS__)
    303 #define ST_SB2(...) ST_B2(v16i8, __VA_ARGS__)
    304 
    305 #define ST_B4(RTYPE, in0, in1, in2, in3, pdst, stride) do {  \
    306  ST_B2(RTYPE, in0, in1, pdst, stride);                      \
    307  ST_B2(RTYPE, in2, in3, pdst + 2 * stride, stride);         \
    308 } while (0)
    309 #define ST_UB4(...) ST_B4(v16u8, __VA_ARGS__)
    310 #define ST_SB4(...) ST_B4(v16i8, __VA_ARGS__)
    311 
    312 #define ST_B8(RTYPE, in0, in1, in2, in3, in4, in5, in6, in7,    \
    313              pdst, stride) do {                                \
    314  ST_B4(RTYPE, in0, in1, in2, in3, pdst, stride);               \
    315  ST_B4(RTYPE, in4, in5, in6, in7, pdst + 4 * stride, stride);  \
    316 } while (0)
    317 #define ST_UB8(...) ST_B8(v16u8, __VA_ARGS__)
    318 
    319 /* Description : Store vectors of 4 word elements with stride
    320 * Arguments   : Inputs - in0, in1, in2, in3, pdst, stride
    321 * Details     : Store 4 word elements from 'in0' to (pdst + 0 * stride)
    322 *               Store 4 word elements from 'in1' to (pdst + 1 * stride)
    323 *               Store 4 word elements from 'in2' to (pdst + 2 * stride)
    324 *               Store 4 word elements from 'in3' to (pdst + 3 * stride)
    325 */
    326 #define ST_W2(RTYPE, in0, in1, pdst, stride) do {  \
    327  ST_W(RTYPE, in0, pdst);                          \
    328  ST_W(RTYPE, in1, pdst + stride);                 \
    329 } while (0)
    330 #define ST_UW2(...) ST_W2(v4u32, __VA_ARGS__)
    331 #define ST_SW2(...) ST_W2(v4i32, __VA_ARGS__)
    332 
    333 #define ST_W3(RTYPE, in0, in1, in2, pdst, stride) do {  \
    334  ST_W2(RTYPE, in0, in1, pdst, stride);                 \
    335  ST_W(RTYPE, in2, pdst + 2 * stride);                  \
    336 } while (0)
    337 #define ST_UW3(...) ST_W3(v4u32, __VA_ARGS__)
    338 #define ST_SW3(...) ST_W3(v4i32, __VA_ARGS__)
    339 
    340 #define ST_W4(RTYPE, in0, in1, in2, in3, pdst, stride) do {  \
    341  ST_W2(RTYPE, in0, in1, pdst, stride);                      \
    342  ST_W2(RTYPE, in2, in3, pdst + 2 * stride, stride);         \
    343 } while (0)
    344 #define ST_UW4(...) ST_W4(v4u32, __VA_ARGS__)
    345 #define ST_SW4(...) ST_W4(v4i32, __VA_ARGS__)
    346 
    347 /* Description : Store vectors of 8 halfword elements with stride
    348 * Arguments   : Inputs - in0, in1, pdst, stride
    349 * Details     : Store 8 halfword elements from 'in0' to (pdst)
    350 *               Store 8 halfword elements from 'in1' to (pdst + stride)
    351 */
    352 #define ST_H2(RTYPE, in0, in1, pdst, stride) do {  \
    353  ST_H(RTYPE, in0, pdst);                          \
    354  ST_H(RTYPE, in1, pdst + stride);                 \
    355 } while (0)
    356 #define ST_UH2(...) ST_H2(v8u16, __VA_ARGS__)
    357 #define ST_SH2(...) ST_H2(v8i16, __VA_ARGS__)
    358 
    359 /* Description : Store 2x4 byte block to destination memory from input vector
    360 * Arguments   : Inputs - in, stidx, pdst, stride
    361 * Details     : Index 'stidx' halfword element from 'in' vector is copied to
    362 *               the GP register and stored to (pdst)
    363 *               Index 'stidx+1' halfword element from 'in' vector is copied to
    364 *               the GP register and stored to (pdst + stride)
    365 *               Index 'stidx+2' halfword element from 'in' vector is copied to
    366 *               the GP register and stored to (pdst + 2 * stride)
    367 *               Index 'stidx+3' halfword element from 'in' vector is copied to
    368 *               the GP register and stored to (pdst + 3 * stride)
    369 */
    370 #define ST2x4_UB(in, stidx, pdst, stride) do {                   \
    371  uint8_t* pblk_2x4_m = (uint8_t*)pdst;                          \
    372  const uint16_t out0_m = __msa_copy_s_h((v8i16)in, stidx);      \
    373  const uint16_t out1_m = __msa_copy_s_h((v8i16)in, stidx + 1);  \
    374  const uint16_t out2_m = __msa_copy_s_h((v8i16)in, stidx + 2);  \
    375  const uint16_t out3_m = __msa_copy_s_h((v8i16)in, stidx + 3);  \
    376  SH(out0_m, pblk_2x4_m);                                        \
    377  pblk_2x4_m += stride;                                          \
    378  SH(out1_m, pblk_2x4_m);                                        \
    379  pblk_2x4_m += stride;                                          \
    380  SH(out2_m, pblk_2x4_m);                                        \
    381  pblk_2x4_m += stride;                                          \
    382  SH(out3_m, pblk_2x4_m);                                        \
    383 } while (0)
    384 
    385 /* Description : Store 4x4 byte block to destination memory from input vector
    386 * Arguments   : Inputs - in0, in1, pdst, stride
    387 * Details     : 'Idx0' word element from input vector 'in0' is copied to the
    388 *               GP register and stored to (pdst)
    389 *               'Idx1' word element from input vector 'in0' is copied to the
    390 *               GP register and stored to (pdst + stride)
    391 *               'Idx2' word element from input vector 'in0' is copied to the
    392 *               GP register and stored to (pdst + 2 * stride)
    393 *               'Idx3' word element from input vector 'in0' is copied to the
    394 *               GP register and stored to (pdst + 3 * stride)
    395 */
    396 #define ST4x4_UB(in0, in1, idx0, idx1, idx2, idx3, pdst, stride) do {  \
    397  uint8_t* const pblk_4x4_m = (uint8_t*)pdst;                          \
    398  const uint32_t out0_m = __msa_copy_s_w((v4i32)in0, idx0);            \
    399  const uint32_t out1_m = __msa_copy_s_w((v4i32)in0, idx1);            \
    400  const uint32_t out2_m = __msa_copy_s_w((v4i32)in1, idx2);            \
    401  const uint32_t out3_m = __msa_copy_s_w((v4i32)in1, idx3);            \
    402  SW4(out0_m, out1_m, out2_m, out3_m, pblk_4x4_m, stride);             \
    403 } while (0)
    404 
    405 #define ST4x8_UB(in0, in1, pdst, stride) do {                     \
    406  uint8_t* const pblk_4x8 = (uint8_t*)pdst;                       \
    407  ST4x4_UB(in0, in0, 0, 1, 2, 3, pblk_4x8, stride);               \
    408  ST4x4_UB(in1, in1, 0, 1, 2, 3, pblk_4x8 + 4 * stride, stride);  \
    409 } while (0)
    410 
    411 /* Description : Immediate number of elements to slide
    412 * Arguments   : Inputs  - in0, in1, slide_val
    413 *               Outputs - out
    414 *               Return Type - as per RTYPE
    415 * Details     : Byte elements from 'in1' vector are slid into 'in0' by
    416 *               value specified in the 'slide_val'
    417 */
    418 #define SLDI_B(RTYPE, in0, in1, slide_val)                      \
    419        (RTYPE)__msa_sldi_b((v16i8)in0, (v16i8)in1, slide_val)  \
    420 
    421 #define SLDI_UB(...) SLDI_B(v16u8, __VA_ARGS__)
    422 #define SLDI_SB(...) SLDI_B(v16i8, __VA_ARGS__)
    423 #define SLDI_SH(...) SLDI_B(v8i16, __VA_ARGS__)
    424 
    425 /* Description : Shuffle byte vector elements as per mask vector
    426 * Arguments   : Inputs  - in0, in1, in2, in3, mask0, mask1
    427 *               Outputs - out0, out1
    428 *               Return Type - as per RTYPE
    429 * Details     : Byte elements from 'in0' & 'in1' are copied selectively to
    430 *               'out0' as per control vector 'mask0'
    431 */
    432 #define VSHF_B(RTYPE, in0, in1, mask)                              \
    433        (RTYPE)__msa_vshf_b((v16i8)mask, (v16i8)in1, (v16i8)in0)
    434 
    435 #define VSHF_UB(...) VSHF_B(v16u8, __VA_ARGS__)
    436 #define VSHF_SB(...) VSHF_B(v16i8, __VA_ARGS__)
    437 #define VSHF_UH(...) VSHF_B(v8u16, __VA_ARGS__)
    438 #define VSHF_SH(...) VSHF_B(v8i16, __VA_ARGS__)
    439 
    440 #define VSHF_B2(RTYPE, in0, in1, in2, in3, mask0, mask1, out0, out1) do {  \
    441  out0 = VSHF_B(RTYPE, in0, in1, mask0);                                   \
    442  out1 = VSHF_B(RTYPE, in2, in3, mask1);                                   \
    443 } while (0)
    444 #define VSHF_B2_UB(...) VSHF_B2(v16u8, __VA_ARGS__)
    445 #define VSHF_B2_SB(...) VSHF_B2(v16i8, __VA_ARGS__)
    446 #define VSHF_B2_UH(...) VSHF_B2(v8u16, __VA_ARGS__)
    447 #define VSHF_B2_SH(...) VSHF_B2(v8i16, __VA_ARGS__)
    448 
    449 /* Description : Shuffle halfword vector elements as per mask vector
    450 * Arguments   : Inputs  - in0, in1, in2, in3, mask0, mask1
    451 *               Outputs - out0, out1
    452 *               Return Type - as per RTYPE
    453 * Details     : halfword elements from 'in0' & 'in1' are copied selectively to
    454 *               'out0' as per control vector 'mask0'
    455 */
    456 #define VSHF_H2(RTYPE, in0, in1, in2, in3, mask0, mask1, out0, out1) do {  \
    457  out0 = (RTYPE)__msa_vshf_h((v8i16)mask0, (v8i16)in1, (v8i16)in0);        \
    458  out1 = (RTYPE)__msa_vshf_h((v8i16)mask1, (v8i16)in3, (v8i16)in2);        \
    459 } while (0)
    460 #define VSHF_H2_UH(...) VSHF_H2(v8u16, __VA_ARGS__)
    461 #define VSHF_H2_SH(...) VSHF_H2(v8i16, __VA_ARGS__)
    462 
    463 /* Description : Dot product of byte vector elements
    464 * Arguments   : Inputs  - mult0, mult1, cnst0, cnst1
    465 *               Outputs - out0, out1
    466 *               Return Type - as per RTYPE
    467 * Details     : Signed byte elements from 'mult0' are multiplied with
    468 *               signed byte elements from 'cnst0' producing a result
    469 *               twice the size of input i.e. signed halfword.
    470 *               The multiplication result of adjacent odd-even elements
    471 *               are added together and written to the 'out0' vector
    472 */
    473 #define DOTP_SB2(RTYPE, mult0, mult1, cnst0, cnst1, out0, out1) do {  \
    474  out0 = (RTYPE)__msa_dotp_s_h((v16i8)mult0, (v16i8)cnst0);           \
    475  out1 = (RTYPE)__msa_dotp_s_h((v16i8)mult1, (v16i8)cnst1);           \
    476 } while (0)
    477 #define DOTP_SB2_SH(...) DOTP_SB2(v8i16, __VA_ARGS__)
    478 
    479 /* Description : Dot product of halfword vector elements
    480 * Arguments   : Inputs  - mult0, mult1, cnst0, cnst1
    481 *               Outputs - out0, out1
    482 *               Return Type - as per RTYPE
    483 * Details     : Signed halfword elements from 'mult0' are multiplied with
    484 *               signed halfword elements from 'cnst0' producing a result
    485 *               twice the size of input i.e. signed word.
    486 *               The multiplication result of adjacent odd-even elements
    487 *               are added together and written to the 'out0' vector
    488 */
    489 #define DOTP_SH2(RTYPE, mult0, mult1, cnst0, cnst1, out0, out1) do {  \
    490  out0 = (RTYPE)__msa_dotp_s_w((v8i16)mult0, (v8i16)cnst0);           \
    491  out1 = (RTYPE)__msa_dotp_s_w((v8i16)mult1, (v8i16)cnst1);           \
    492 } while (0)
    493 #define DOTP_SH2_SW(...) DOTP_SH2(v4i32, __VA_ARGS__)
    494 
    495 /* Description : Dot product of unsigned word vector elements
    496 * Arguments   : Inputs  - mult0, mult1, cnst0, cnst1
    497 *               Outputs - out0, out1
    498 *               Return Type - as per RTYPE
    499 * Details     : Unsigned word elements from 'mult0' are multiplied with
    500 *               unsigned word elements from 'cnst0' producing a result
    501 *               twice the size of input i.e. unsigned double word.
    502 *               The multiplication result of adjacent odd-even elements
    503 *               are added together and written to the 'out0' vector
    504 */
    505 #define DOTP_UW2(RTYPE, mult0, mult1, cnst0, cnst1, out0, out1) do {  \
    506  out0 = (RTYPE)__msa_dotp_u_d((v4u32)mult0, (v4u32)cnst0);           \
    507  out1 = (RTYPE)__msa_dotp_u_d((v4u32)mult1, (v4u32)cnst1);           \
    508 } while (0)
    509 #define DOTP_UW2_UD(...) DOTP_UW2(v2u64, __VA_ARGS__)
    510 
    511 /* Description : Dot product & addition of halfword vector elements
    512 * Arguments   : Inputs  - mult0, mult1, cnst0, cnst1
    513 *               Outputs - out0, out1
    514 *               Return Type - as per RTYPE
    515 * Details     : Signed halfword elements from 'mult0' are multiplied with
    516 *               signed halfword elements from 'cnst0' producing a result
    517 *               twice the size of input i.e. signed word.
    518 *               The multiplication result of adjacent odd-even elements
    519 *               are added to the 'out0' vector
    520 */
    521 #define DPADD_SH2(RTYPE, mult0, mult1, cnst0, cnst1, out0, out1) do {      \
    522  out0 = (RTYPE)__msa_dpadd_s_w((v4i32)out0, (v8i16)mult0, (v8i16)cnst0);  \
    523  out1 = (RTYPE)__msa_dpadd_s_w((v4i32)out1, (v8i16)mult1, (v8i16)cnst1);  \
    524 } while (0)
    525 #define DPADD_SH2_SW(...) DPADD_SH2(v4i32, __VA_ARGS__)
    526 
    527 /* Description : Clips all signed halfword elements of input vector
    528 *               between 0 & 255
    529 * Arguments   : Input/output  - val
    530 *               Return Type - signed halfword
    531 */
    532 #define CLIP_SH_0_255(val) do {                   \
    533  const v8i16 max_m = __msa_ldi_h(255);           \
    534  val = __msa_maxi_s_h((v8i16)val, 0);            \
    535  val = __msa_min_s_h(max_m, (v8i16)val);         \
    536 } while (0)
    537 
    538 #define CLIP_SH2_0_255(in0, in1) do {  \
    539  CLIP_SH_0_255(in0);                  \
    540  CLIP_SH_0_255(in1);                  \
    541 } while (0)
    542 
    543 #define CLIP_SH4_0_255(in0, in1, in2, in3) do {  \
    544  CLIP_SH2_0_255(in0, in1);                      \
    545  CLIP_SH2_0_255(in2, in3);                      \
    546 } while (0)
    547 
    548 /* Description : Clips all unsigned halfword elements of input vector
    549 *               between 0 & 255
    550 * Arguments   : Input  - in
    551 *               Output - out_m
    552 *               Return Type - unsigned halfword
    553 */
    554 #define CLIP_UH_0_255(in) do {                    \
    555  const v8u16 max_m = (v8u16)__msa_ldi_h(255);    \
    556  in = __msa_maxi_u_h((v8u16) in, 0);             \
    557  in = __msa_min_u_h((v8u16) max_m, (v8u16) in);  \
    558 } while (0)
    559 
    560 #define CLIP_UH2_0_255(in0, in1) do {  \
    561  CLIP_UH_0_255(in0);                  \
    562  CLIP_UH_0_255(in1);                  \
    563 } while (0)
    564 
    565 /* Description : Clips all signed word elements of input vector
    566 *               between 0 & 255
    567 * Arguments   : Input/output  - val
    568 *               Return Type - signed word
    569 */
    570 #define CLIP_SW_0_255(val) do {                   \
    571  const v4i32 max_m = __msa_ldi_w(255);           \
    572  val = __msa_maxi_s_w((v4i32)val, 0);            \
    573  val = __msa_min_s_w(max_m, (v4i32)val);         \
    574 } while (0)
    575 
    576 #define CLIP_SW4_0_255(in0, in1, in2, in3) do {   \
    577  CLIP_SW_0_255(in0);                             \
    578  CLIP_SW_0_255(in1);                             \
    579  CLIP_SW_0_255(in2);                             \
    580  CLIP_SW_0_255(in3);                             \
    581 } while (0)
    582 
    583 /* Description : Horizontal addition of 4 signed word elements of input vector
    584 * Arguments   : Input  - in       (signed word vector)
    585 *               Output - sum_m    (i32 sum)
    586 *               Return Type - signed word (GP)
    587 * Details     : 4 signed word elements of 'in' vector are added together and
    588 *               the resulting integer sum is returned
    589 */
    590 static WEBP_INLINE int32_t func_hadd_sw_s32(v4i32 in) {
    591  const v2i64 res0_m = __msa_hadd_s_d((v4i32)in, (v4i32)in);
    592  const v2i64 res1_m = __msa_splati_d(res0_m, 1);
    593  const v2i64 out = res0_m + res1_m;
    594  int32_t sum_m = __msa_copy_s_w((v4i32)out, 0);
    595  return sum_m;
    596 }
    597 #define HADD_SW_S32(in) func_hadd_sw_s32(in)
    598 
    599 /* Description : Horizontal addition of 8 signed halfword elements
    600 * Arguments   : Input  - in       (signed halfword vector)
    601 *               Output - sum_m    (s32 sum)
    602 *               Return Type - signed word
    603 * Details     : 8 signed halfword elements of input vector are added
    604 *               together and the resulting integer sum is returned
    605 */
    606 static WEBP_INLINE int32_t func_hadd_sh_s32(v8i16 in) {
    607  const v4i32 res = __msa_hadd_s_w(in, in);
    608  const v2i64 res0 = __msa_hadd_s_d(res, res);
    609  const v2i64 res1 = __msa_splati_d(res0, 1);
    610  const v2i64 res2 = res0 + res1;
    611  const int32_t sum_m = __msa_copy_s_w((v4i32)res2, 0);
    612  return sum_m;
    613 }
    614 #define HADD_SH_S32(in) func_hadd_sh_s32(in)
    615 
    616 /* Description : Horizontal addition of 8 unsigned halfword elements
    617 * Arguments   : Input  - in       (unsigned halfword vector)
    618 *               Output - sum_m    (u32 sum)
    619 *               Return Type - unsigned word
    620 * Details     : 8 unsigned halfword elements of input vector are added
    621 *               together and the resulting integer sum is returned
    622 */
    623 static WEBP_INLINE uint32_t func_hadd_uh_u32(v8u16 in) {
    624  uint32_t sum_m;
    625  const v4u32 res_m = __msa_hadd_u_w(in, in);
    626  v2u64 res0_m = __msa_hadd_u_d(res_m, res_m);
    627  v2u64 res1_m = (v2u64)__msa_splati_d((v2i64)res0_m, 1);
    628  res0_m = res0_m + res1_m;
    629  sum_m = __msa_copy_s_w((v4i32)res0_m, 0);
    630  return sum_m;
    631 }
    632 #define HADD_UH_U32(in) func_hadd_uh_u32(in)
    633 
    634 /* Description : Horizontal addition of signed half word vector elements
    635   Arguments   : Inputs  - in0, in1
    636                 Outputs - out0, out1
    637                 Return Type - as per RTYPE
    638   Details     : Each signed odd half word element from 'in0' is added to
    639                 even signed half word element from 'in0' (pairwise) and the
    640                 halfword result is written in 'out0'
    641 */
    642 #define HADD_SH2(RTYPE, in0, in1, out0, out1) do {       \
    643  out0 = (RTYPE)__msa_hadd_s_w((v8i16)in0, (v8i16)in0);  \
    644  out1 = (RTYPE)__msa_hadd_s_w((v8i16)in1, (v8i16)in1);  \
    645 } while (0)
    646 #define HADD_SH2_SW(...) HADD_SH2(v4i32, __VA_ARGS__)
    647 
    648 #define HADD_SH4(RTYPE, in0, in1, in2, in3, out0, out1, out2, out3) do {  \
    649  HADD_SH2(RTYPE, in0, in1, out0, out1);                                  \
    650  HADD_SH2(RTYPE, in2, in3, out2, out3);                                  \
    651 } while (0)
    652 #define HADD_SH4_SW(...) HADD_SH4(v4i32, __VA_ARGS__)
    653 
    654 /* Description : Horizontal subtraction of unsigned byte vector elements
    655 * Arguments   : Inputs  - in0, in1
    656 *               Outputs - out0, out1
    657 *               Return Type - as per RTYPE
    658 * Details     : Each unsigned odd byte element from 'in0' is subtracted from
    659 *               even unsigned byte element from 'in0' (pairwise) and the
    660 *               halfword result is written to 'out0'
    661 */
    662 #define HSUB_UB2(RTYPE, in0, in1, out0, out1) do {       \
    663  out0 = (RTYPE)__msa_hsub_u_h((v16u8)in0, (v16u8)in0);  \
    664  out1 = (RTYPE)__msa_hsub_u_h((v16u8)in1, (v16u8)in1);  \
    665 } while (0)
    666 #define HSUB_UB2_UH(...) HSUB_UB2(v8u16, __VA_ARGS__)
    667 #define HSUB_UB2_SH(...) HSUB_UB2(v8i16, __VA_ARGS__)
    668 #define HSUB_UB2_SW(...) HSUB_UB2(v4i32, __VA_ARGS__)
    669 
    670 /* Description : Set element n input vector to GPR value
    671 * Arguments   : Inputs - in0, in1, in2, in3
    672 *               Output - out
    673 *               Return Type - as per RTYPE
    674 * Details     : Set element 0 in vector 'out' to value specified in 'in0'
    675 */
    676 #define INSERT_W2(RTYPE, in0, in1, out) do {        \
    677  out = (RTYPE)__msa_insert_w((v4i32)out, 0, in0);  \
    678  out = (RTYPE)__msa_insert_w((v4i32)out, 1, in1);  \
    679 } while (0)
    680 #define INSERT_W2_UB(...) INSERT_W2(v16u8, __VA_ARGS__)
    681 #define INSERT_W2_SB(...) INSERT_W2(v16i8, __VA_ARGS__)
    682 
    683 #define INSERT_W4(RTYPE, in0, in1, in2, in3, out) do {  \
    684  out = (RTYPE)__msa_insert_w((v4i32)out, 0, in0);      \
    685  out = (RTYPE)__msa_insert_w((v4i32)out, 1, in1);      \
    686  out = (RTYPE)__msa_insert_w((v4i32)out, 2, in2);      \
    687  out = (RTYPE)__msa_insert_w((v4i32)out, 3, in3);      \
    688 } while (0)
    689 #define INSERT_W4_UB(...) INSERT_W4(v16u8, __VA_ARGS__)
    690 #define INSERT_W4_SB(...) INSERT_W4(v16i8, __VA_ARGS__)
    691 #define INSERT_W4_SW(...) INSERT_W4(v4i32, __VA_ARGS__)
    692 
    693 /* Description : Set element n of double word input vector to GPR value
    694 * Arguments   : Inputs - in0, in1
    695 *               Output - out
    696 *               Return Type - as per RTYPE
    697 * Details     : Set element 0 in vector 'out' to GPR value specified in 'in0'
    698 *               Set element 1 in vector 'out' to GPR value specified in 'in1'
    699 */
    700 #define INSERT_D2(RTYPE, in0, in1, out) do {        \
    701  out = (RTYPE)__msa_insert_d((v2i64)out, 0, in0);  \
    702  out = (RTYPE)__msa_insert_d((v2i64)out, 1, in1);  \
    703 } while (0)
    704 #define INSERT_D2_UB(...) INSERT_D2(v16u8, __VA_ARGS__)
    705 #define INSERT_D2_SB(...) INSERT_D2(v16i8, __VA_ARGS__)
    706 
    707 /* Description : Interleave even byte elements from vectors
    708 * Arguments   : Inputs  - in0, in1, in2, in3
    709 *               Outputs - out0, out1
    710 *               Return Type - as per RTYPE
    711 * Details     : Even byte elements of 'in0' and 'in1' are interleaved
    712 *               and written to 'out0'
    713 */
    714 #define ILVEV_B2(RTYPE, in0, in1, in2, in3, out0, out1) do {  \
    715  out0 = (RTYPE)__msa_ilvev_b((v16i8)in1, (v16i8)in0);        \
    716  out1 = (RTYPE)__msa_ilvev_b((v16i8)in3, (v16i8)in2);        \
    717 } while (0)
    718 #define ILVEV_B2_UB(...) ILVEV_B2(v16u8, __VA_ARGS__)
    719 #define ILVEV_B2_SB(...) ILVEV_B2(v16i8, __VA_ARGS__)
    720 #define ILVEV_B2_UH(...) ILVEV_B2(v8u16, __VA_ARGS__)
    721 #define ILVEV_B2_SH(...) ILVEV_B2(v8i16, __VA_ARGS__)
    722 #define ILVEV_B2_SD(...) ILVEV_B2(v2i64, __VA_ARGS__)
    723 
    724 /* Description : Interleave odd byte elements from vectors
    725 * Arguments   : Inputs  - in0, in1, in2, in3
    726 *               Outputs - out0, out1
    727 *               Return Type - as per RTYPE
    728 * Details     : Odd byte elements of 'in0' and 'in1' are interleaved
    729 *               and written to 'out0'
    730 */
    731 #define ILVOD_B2(RTYPE, in0, in1, in2, in3, out0, out1) do {  \
    732  out0 = (RTYPE)__msa_ilvod_b((v16i8)in1, (v16i8)in0);        \
    733  out1 = (RTYPE)__msa_ilvod_b((v16i8)in3, (v16i8)in2);        \
    734 } while (0)
    735 #define ILVOD_B2_UB(...) ILVOD_B2(v16u8, __VA_ARGS__)
    736 #define ILVOD_B2_SB(...) ILVOD_B2(v16i8, __VA_ARGS__)
    737 #define ILVOD_B2_UH(...) ILVOD_B2(v8u16, __VA_ARGS__)
    738 #define ILVOD_B2_SH(...) ILVOD_B2(v8i16, __VA_ARGS__)
    739 #define ILVOD_B2_SD(...) ILVOD_B2(v2i64, __VA_ARGS__)
    740 
    741 /* Description : Interleave even halfword elements from vectors
    742 * Arguments   : Inputs  - in0, in1, in2, in3
    743 *               Outputs - out0, out1
    744 *               Return Type - as per RTYPE
    745 * Details     : Even halfword elements of 'in0' and 'in1' are interleaved
    746 *               and written to 'out0'
    747 */
    748 #define ILVEV_H2(RTYPE, in0, in1, in2, in3, out0, out1) do {  \
    749  out0 = (RTYPE)__msa_ilvev_h((v8i16)in1, (v8i16)in0);        \
    750  out1 = (RTYPE)__msa_ilvev_h((v8i16)in3, (v8i16)in2);        \
    751 } while (0)
    752 #define ILVEV_H2_UB(...) ILVEV_H2(v16u8, __VA_ARGS__)
    753 #define ILVEV_H2_UH(...) ILVEV_H2(v8u16, __VA_ARGS__)
    754 #define ILVEV_H2_SH(...) ILVEV_H2(v8i16, __VA_ARGS__)
    755 #define ILVEV_H2_SW(...) ILVEV_H2(v4i32, __VA_ARGS__)
    756 
    757 /* Description : Interleave odd halfword elements from vectors
    758 * Arguments   : Inputs  - in0, in1, in2, in3
    759 *               Outputs - out0, out1
    760 *               Return Type - as per RTYPE
    761 * Details     : Odd halfword elements of 'in0' and 'in1' are interleaved
    762 *               and written to 'out0'
    763 */
    764 #define ILVOD_H2(RTYPE, in0, in1, in2, in3, out0, out1) do {  \
    765  out0 = (RTYPE)__msa_ilvod_h((v8i16)in1, (v8i16)in0);        \
    766  out1 = (RTYPE)__msa_ilvod_h((v8i16)in3, (v8i16)in2);        \
    767 } while (0)
    768 #define ILVOD_H2_UB(...) ILVOD_H2(v16u8, __VA_ARGS__)
    769 #define ILVOD_H2_UH(...) ILVOD_H2(v8u16, __VA_ARGS__)
    770 #define ILVOD_H2_SH(...) ILVOD_H2(v8i16, __VA_ARGS__)
    771 #define ILVOD_H2_SW(...) ILVOD_H2(v4i32, __VA_ARGS__)
    772 
    773 /* Description : Interleave even word elements from vectors
    774 * Arguments   : Inputs  - in0, in1, in2, in3
    775 *               Outputs - out0, out1
    776 *               Return Type - as per RTYPE
    777 * Details     : Even word elements of 'in0' and 'in1' are interleaved
    778 *               and written to 'out0'
    779 */
    780 #define ILVEV_W2(RTYPE, in0, in1, in2, in3, out0, out1) do {  \
    781  out0 = (RTYPE)__msa_ilvev_w((v4i32)in1, (v4i32)in0);        \
    782  out1 = (RTYPE)__msa_ilvev_w((v4i32)in3, (v4i32)in2);        \
    783 } while (0)
    784 #define ILVEV_W2_UB(...) ILVEV_W2(v16u8, __VA_ARGS__)
    785 #define ILVEV_W2_SB(...) ILVEV_W2(v16i8, __VA_ARGS__)
    786 #define ILVEV_W2_UH(...) ILVEV_W2(v8u16, __VA_ARGS__)
    787 #define ILVEV_W2_SD(...) ILVEV_W2(v2i64, __VA_ARGS__)
    788 
    789 /* Description : Interleave even-odd word elements from vectors
    790 * Arguments   : Inputs  - in0, in1, in2, in3
    791 *               Outputs - out0, out1
    792 *               Return Type - as per RTYPE
    793 * Details     : Even word elements of 'in0' and 'in1' are interleaved
    794 *               and written to 'out0'
    795 *               Odd word elements of 'in2' and 'in3' are interleaved
    796 *               and written to 'out1'
    797 */
    798 #define ILVEVOD_W2(RTYPE, in0, in1, in2, in3, out0, out1) do {  \
    799  out0 = (RTYPE)__msa_ilvev_w((v4i32)in1, (v4i32)in0);          \
    800  out1 = (RTYPE)__msa_ilvod_w((v4i32)in3, (v4i32)in2);          \
    801 } while (0)
    802 #define ILVEVOD_W2_UB(...) ILVEVOD_W2(v16u8, __VA_ARGS__)
    803 #define ILVEVOD_W2_UH(...) ILVEVOD_W2(v8u16, __VA_ARGS__)
    804 #define ILVEVOD_W2_SH(...) ILVEVOD_W2(v8i16, __VA_ARGS__)
    805 #define ILVEVOD_W2_SW(...) ILVEVOD_W2(v4i32, __VA_ARGS__)
    806 
    807 /* Description : Interleave even-odd half-word elements from vectors
    808 * Arguments   : Inputs  - in0, in1, in2, in3
    809 *               Outputs - out0, out1
    810 *               Return Type - as per RTYPE
    811 * Details     : Even half-word elements of 'in0' and 'in1' are interleaved
    812 *               and written to 'out0'
    813 *               Odd half-word elements of 'in2' and 'in3' are interleaved
    814 *               and written to 'out1'
    815 */
    816 #define ILVEVOD_H2(RTYPE, in0, in1, in2, in3, out0, out1) do {  \
    817  out0 = (RTYPE)__msa_ilvev_h((v8i16)in1, (v8i16)in0);          \
    818  out1 = (RTYPE)__msa_ilvod_h((v8i16)in3, (v8i16)in2);          \
    819 } while (0)
    820 #define ILVEVOD_H2_UB(...) ILVEVOD_H2(v16u8, __VA_ARGS__)
    821 #define ILVEVOD_H2_UH(...) ILVEVOD_H2(v8u16, __VA_ARGS__)
    822 #define ILVEVOD_H2_SH(...) ILVEVOD_H2(v8i16, __VA_ARGS__)
    823 #define ILVEVOD_H2_SW(...) ILVEVOD_H2(v4i32, __VA_ARGS__)
    824 
    825 /* Description : Interleave even double word elements from vectors
    826 * Arguments   : Inputs  - in0, in1, in2, in3
    827 *               Outputs - out0, out1
    828 *               Return Type - as per RTYPE
    829 * Details     : Even double word elements of 'in0' and 'in1' are interleaved
    830 *               and written to 'out0'
    831 */
    832 #define ILVEV_D2(RTYPE, in0, in1, in2, in3, out0, out1) do {  \
    833  out0 = (RTYPE)__msa_ilvev_d((v2i64)in1, (v2i64)in0);        \
    834  out1 = (RTYPE)__msa_ilvev_d((v2i64)in3, (v2i64)in2);        \
    835 } while (0)
    836 #define ILVEV_D2_UB(...) ILVEV_D2(v16u8, __VA_ARGS__)
    837 #define ILVEV_D2_SB(...) ILVEV_D2(v16i8, __VA_ARGS__)
    838 #define ILVEV_D2_SW(...) ILVEV_D2(v4i32, __VA_ARGS__)
    839 #define ILVEV_D2_SD(...) ILVEV_D2(v2i64, __VA_ARGS__)
    840 
    841 /* Description : Interleave left half of byte elements from vectors
    842 * Arguments   : Inputs  - in0, in1, in2, in3
    843 *               Outputs - out0, out1
    844 *               Return Type - as per RTYPE
    845 * Details     : Left half of byte elements of 'in0' and 'in1' are interleaved
    846 *               and written to 'out0'.
    847 */
    848 #define ILVL_B2(RTYPE, in0, in1, in2, in3, out0, out1) do {  \
    849  out0 = (RTYPE)__msa_ilvl_b((v16i8)in0, (v16i8)in1);        \
    850  out1 = (RTYPE)__msa_ilvl_b((v16i8)in2, (v16i8)in3);        \
    851 } while (0)
    852 #define ILVL_B2_UB(...) ILVL_B2(v16u8, __VA_ARGS__)
    853 #define ILVL_B2_SB(...) ILVL_B2(v16i8, __VA_ARGS__)
    854 #define ILVL_B2_UH(...) ILVL_B2(v8u16, __VA_ARGS__)
    855 #define ILVL_B2_SH(...) ILVL_B2(v8i16, __VA_ARGS__)
    856 #define ILVL_B2_SW(...) ILVL_B2(v4i32, __VA_ARGS__)
    857 
    858 /* Description : Interleave right half of byte elements from vectors
    859 * Arguments   : Inputs  - in0, in1, in2, in3
    860 *               Outputs - out0, out1
    861 *               Return Type - as per RTYPE
    862 * Details     : Right half of byte elements of 'in0' and 'in1' are interleaved
    863 *               and written to out0.
    864 */
    865 #define ILVR_B2(RTYPE, in0, in1, in2, in3, out0, out1) do {  \
    866  out0 = (RTYPE)__msa_ilvr_b((v16i8)in0, (v16i8)in1);        \
    867  out1 = (RTYPE)__msa_ilvr_b((v16i8)in2, (v16i8)in3);        \
    868 } while (0)
    869 #define ILVR_B2_UB(...) ILVR_B2(v16u8, __VA_ARGS__)
    870 #define ILVR_B2_SB(...) ILVR_B2(v16i8, __VA_ARGS__)
    871 #define ILVR_B2_UH(...) ILVR_B2(v8u16, __VA_ARGS__)
    872 #define ILVR_B2_SH(...) ILVR_B2(v8i16, __VA_ARGS__)
    873 #define ILVR_B2_SW(...) ILVR_B2(v4i32, __VA_ARGS__)
    874 
    875 #define ILVR_B4(RTYPE, in0, in1, in2, in3, in4, in5, in6, in7,  \
    876                out0, out1, out2, out3) do {                    \
    877  ILVR_B2(RTYPE, in0, in1, in2, in3, out0, out1);               \
    878  ILVR_B2(RTYPE, in4, in5, in6, in7, out2, out3);               \
    879 } while (0)
    880 #define ILVR_B4_UB(...) ILVR_B4(v16u8, __VA_ARGS__)
    881 #define ILVR_B4_SB(...) ILVR_B4(v16i8, __VA_ARGS__)
    882 #define ILVR_B4_UH(...) ILVR_B4(v8u16, __VA_ARGS__)
    883 #define ILVR_B4_SH(...) ILVR_B4(v8i16, __VA_ARGS__)
    884 #define ILVR_B4_SW(...) ILVR_B4(v4i32, __VA_ARGS__)
    885 
    886 /* Description : Interleave right half of halfword elements from vectors
    887 * Arguments   : Inputs  - in0, in1, in2, in3
    888 *               Outputs - out0, out1
    889 *               Return Type - as per RTYPE
    890 * Details     : Right half of halfword elements of 'in0' and 'in1' are
    891 *               interleaved and written to 'out0'.
    892 */
    893 #define ILVR_H2(RTYPE, in0, in1, in2, in3, out0, out1) do {  \
    894  out0 = (RTYPE)__msa_ilvr_h((v8i16)in0, (v8i16)in1);        \
    895  out1 = (RTYPE)__msa_ilvr_h((v8i16)in2, (v8i16)in3);        \
    896 } while (0)
    897 #define ILVR_H2_UB(...) ILVR_H2(v16u8, __VA_ARGS__)
    898 #define ILVR_H2_SH(...) ILVR_H2(v8i16, __VA_ARGS__)
    899 #define ILVR_H2_SW(...) ILVR_H2(v4i32, __VA_ARGS__)
    900 
    901 #define ILVR_H4(RTYPE, in0, in1, in2, in3, in4, in5, in6, in7,  \
    902                out0, out1, out2, out3) do {                    \
    903  ILVR_H2(RTYPE, in0, in1, in2, in3, out0, out1);               \
    904  ILVR_H2(RTYPE, in4, in5, in6, in7, out2, out3);               \
    905 } while (0)
    906 #define ILVR_H4_UB(...) ILVR_H4(v16u8, __VA_ARGS__)
    907 #define ILVR_H4_SH(...) ILVR_H4(v8i16, __VA_ARGS__)
    908 #define ILVR_H4_SW(...) ILVR_H4(v4i32, __VA_ARGS__)
    909 
    910 /* Description : Interleave right half of double word elements from vectors
    911 * Arguments   : Inputs  - in0, in1, in2, in3
    912 *               Outputs - out0, out1
    913 *               Return Type - as per RTYPE
    914 * Details     : Right half of double word elements of 'in0' and 'in1' are
    915 *               interleaved and written to 'out0'.
    916 */
    917 #define ILVR_D2(RTYPE, in0, in1, in2, in3, out0, out1) do {  \
    918  out0 = (RTYPE)__msa_ilvr_d((v2i64)in0, (v2i64)in1);        \
    919  out1 = (RTYPE)__msa_ilvr_d((v2i64)in2, (v2i64)in3);        \
    920 } while (0)
    921 #define ILVR_D2_UB(...) ILVR_D2(v16u8, __VA_ARGS__)
    922 #define ILVR_D2_SB(...) ILVR_D2(v16i8, __VA_ARGS__)
    923 #define ILVR_D2_SH(...) ILVR_D2(v8i16, __VA_ARGS__)
    924 
    925 #define ILVR_D4(RTYPE, in0, in1, in2, in3, in4, in5, in6, in7,  \
    926                out0, out1, out2, out3) do {                    \
    927  ILVR_D2(RTYPE, in0, in1, in2, in3, out0, out1);               \
    928  ILVR_D2(RTYPE, in4, in5, in6, in7, out2, out3);               \
    929 } while (0)
    930 #define ILVR_D4_SB(...) ILVR_D4(v16i8, __VA_ARGS__)
    931 #define ILVR_D4_UB(...) ILVR_D4(v16u8, __VA_ARGS__)
    932 
    933 /* Description : Interleave both left and right half of input vectors
    934 * Arguments   : Inputs  - in0, in1
    935 *               Outputs - out0, out1
    936 *               Return Type - as per RTYPE
    937 * Details     : Right half of byte elements from 'in0' and 'in1' are
    938 *               interleaved and written to 'out0'
    939 */
    940 #define ILVRL_B2(RTYPE, in0, in1, out0, out1) do {     \
    941  out0 = (RTYPE)__msa_ilvr_b((v16i8)in0, (v16i8)in1);  \
    942  out1 = (RTYPE)__msa_ilvl_b((v16i8)in0, (v16i8)in1);  \
    943 } while (0)
    944 #define ILVRL_B2_UB(...) ILVRL_B2(v16u8, __VA_ARGS__)
    945 #define ILVRL_B2_SB(...) ILVRL_B2(v16i8, __VA_ARGS__)
    946 #define ILVRL_B2_UH(...) ILVRL_B2(v8u16, __VA_ARGS__)
    947 #define ILVRL_B2_SH(...) ILVRL_B2(v8i16, __VA_ARGS__)
    948 #define ILVRL_B2_SW(...) ILVRL_B2(v4i32, __VA_ARGS__)
    949 
    950 #define ILVRL_H2(RTYPE, in0, in1, out0, out1) do {     \
    951  out0 = (RTYPE)__msa_ilvr_h((v8i16)in0, (v8i16)in1);  \
    952  out1 = (RTYPE)__msa_ilvl_h((v8i16)in0, (v8i16)in1);  \
    953 } while (0)
    954 #define ILVRL_H2_UB(...) ILVRL_H2(v16u8, __VA_ARGS__)
    955 #define ILVRL_H2_SB(...) ILVRL_H2(v16i8, __VA_ARGS__)
    956 #define ILVRL_H2_SH(...) ILVRL_H2(v8i16, __VA_ARGS__)
    957 #define ILVRL_H2_SW(...) ILVRL_H2(v4i32, __VA_ARGS__)
    958 #define ILVRL_H2_UW(...) ILVRL_H2(v4u32, __VA_ARGS__)
    959 
    960 #define ILVRL_W2(RTYPE, in0, in1, out0, out1) do {     \
    961  out0 = (RTYPE)__msa_ilvr_w((v4i32)in0, (v4i32)in1);  \
    962  out1 = (RTYPE)__msa_ilvl_w((v4i32)in0, (v4i32)in1);  \
    963 } while (0)
    964 #define ILVRL_W2_UB(...) ILVRL_W2(v16u8, __VA_ARGS__)
    965 #define ILVRL_W2_SH(...) ILVRL_W2(v8i16, __VA_ARGS__)
    966 #define ILVRL_W2_SW(...) ILVRL_W2(v4i32, __VA_ARGS__)
    967 #define ILVRL_W2_UW(...) ILVRL_W2(v4u32, __VA_ARGS__)
    968 
    969 /* Description : Pack even byte elements of vector pairs
    970 *  Arguments   : Inputs  - in0, in1, in2, in3
    971 *                Outputs - out0, out1
    972 *                Return Type - as per RTYPE
    973 *  Details     : Even byte elements of 'in0' are copied to the left half of
    974 *                'out0' & even byte elements of 'in1' are copied to the right
    975 *                half of 'out0'.
    976 */
    977 #define PCKEV_B2(RTYPE, in0, in1, in2, in3, out0, out1) do {  \
    978  out0 = (RTYPE)__msa_pckev_b((v16i8)in0, (v16i8)in1);        \
    979  out1 = (RTYPE)__msa_pckev_b((v16i8)in2, (v16i8)in3);        \
    980 } while (0)
    981 #define PCKEV_B2_SB(...) PCKEV_B2(v16i8, __VA_ARGS__)
    982 #define PCKEV_B2_UB(...) PCKEV_B2(v16u8, __VA_ARGS__)
    983 #define PCKEV_B2_SH(...) PCKEV_B2(v8i16, __VA_ARGS__)
    984 #define PCKEV_B2_SW(...) PCKEV_B2(v4i32, __VA_ARGS__)
    985 
    986 #define PCKEV_B4(RTYPE, in0, in1, in2, in3, in4, in5, in6, in7,  \
    987                 out0, out1, out2, out3) do {                    \
    988  PCKEV_B2(RTYPE, in0, in1, in2, in3, out0, out1);               \
    989  PCKEV_B2(RTYPE, in4, in5, in6, in7, out2, out3);               \
    990 } while (0)
    991 #define PCKEV_B4_SB(...) PCKEV_B4(v16i8, __VA_ARGS__)
    992 #define PCKEV_B4_UB(...) PCKEV_B4(v16u8, __VA_ARGS__)
    993 #define PCKEV_B4_SH(...) PCKEV_B4(v8i16, __VA_ARGS__)
    994 #define PCKEV_B4_SW(...) PCKEV_B4(v4i32, __VA_ARGS__)
    995 
    996 /* Description : Pack even halfword elements of vector pairs
    997 * Arguments   : Inputs  - in0, in1, in2, in3
    998 *               Outputs - out0, out1
    999 *               Return Type - as per RTYPE
   1000 * Details     : Even halfword elements of 'in0' are copied to the left half of
   1001 *               'out0' & even halfword elements of 'in1' are copied to the
   1002 *               right half of 'out0'.
   1003 */
   1004 #define PCKEV_H2(RTYPE, in0, in1, in2, in3, out0, out1) do {  \
   1005  out0 = (RTYPE)__msa_pckev_h((v8i16)in0, (v8i16)in1);        \
   1006  out1 = (RTYPE)__msa_pckev_h((v8i16)in2, (v8i16)in3);        \
   1007 } while (0)
   1008 #define PCKEV_H2_UH(...) PCKEV_H2(v8u16, __VA_ARGS__)
   1009 #define PCKEV_H2_SH(...) PCKEV_H2(v8i16, __VA_ARGS__)
   1010 #define PCKEV_H2_SW(...) PCKEV_H2(v4i32, __VA_ARGS__)
   1011 #define PCKEV_H2_UW(...) PCKEV_H2(v4u32, __VA_ARGS__)
   1012 
   1013 /* Description : Pack even word elements of vector pairs
   1014 * Arguments   : Inputs  - in0, in1, in2, in3
   1015 *               Outputs - out0, out1
   1016 *               Return Type - as per RTYPE
   1017 * Details     : Even word elements of 'in0' are copied to the left half of
   1018 *               'out0' & even word elements of 'in1' are copied to the
   1019 *               right half of 'out0'.
   1020 */
   1021 #define PCKEV_W2(RTYPE, in0, in1, in2, in3, out0, out1) do {  \
   1022  out0 = (RTYPE)__msa_pckev_w((v4i32)in0, (v4i32)in1);        \
   1023  out1 = (RTYPE)__msa_pckev_w((v4i32)in2, (v4i32)in3);        \
   1024 } while (0)
   1025 #define PCKEV_W2_UH(...) PCKEV_W2(v8u16, __VA_ARGS__)
   1026 #define PCKEV_W2_SH(...) PCKEV_W2(v8i16, __VA_ARGS__)
   1027 #define PCKEV_W2_SW(...) PCKEV_W2(v4i32, __VA_ARGS__)
   1028 #define PCKEV_W2_UW(...) PCKEV_W2(v4u32, __VA_ARGS__)
   1029 
   1030 /* Description : Pack odd halfword elements of vector pairs
   1031 * Arguments   : Inputs  - in0, in1, in2, in3
   1032 *               Outputs - out0, out1
   1033 *               Return Type - as per RTYPE
   1034 * Details     : Odd halfword elements of 'in0' are copied to the left half of
   1035 *               'out0' & odd halfword elements of 'in1' are copied to the
   1036 *               right half of 'out0'.
   1037 */
   1038 #define PCKOD_H2(RTYPE, in0, in1, in2, in3, out0, out1) do {  \
   1039  out0 = (RTYPE)__msa_pckod_h((v8i16)in0, (v8i16)in1);        \
   1040  out1 = (RTYPE)__msa_pckod_h((v8i16)in2, (v8i16)in3);        \
   1041 } while (0)
   1042 #define PCKOD_H2_UH(...) PCKOD_H2(v8u16, __VA_ARGS__)
   1043 #define PCKOD_H2_SH(...) PCKOD_H2(v8i16, __VA_ARGS__)
   1044 #define PCKOD_H2_SW(...) PCKOD_H2(v4i32, __VA_ARGS__)
   1045 #define PCKOD_H2_UW(...) PCKOD_H2(v4u32, __VA_ARGS__)
   1046 
   1047 /* Description : Arithmetic immediate shift right all elements of word vector
   1048 * Arguments   : Inputs  - in0, in1, shift
   1049 *               Outputs - in place operation
   1050 *               Return Type - as per input vector RTYPE
   1051 * Details     : Each element of vector 'in0' is right shifted by 'shift' and
   1052 *               the result is written in-place. 'shift' is a GP variable.
   1053 */
   1054 #define SRAI_W2(RTYPE, in0, in1, shift_val) do {  \
   1055  in0 = (RTYPE)SRAI_W(in0, shift_val);            \
   1056  in1 = (RTYPE)SRAI_W(in1, shift_val);            \
   1057 } while (0)
   1058 #define SRAI_W2_SW(...) SRAI_W2(v4i32, __VA_ARGS__)
   1059 #define SRAI_W2_UW(...) SRAI_W2(v4u32, __VA_ARGS__)
   1060 
   1061 #define SRAI_W4(RTYPE, in0, in1, in2, in3, shift_val) do {  \
   1062  SRAI_W2(RTYPE, in0, in1, shift_val);                      \
   1063  SRAI_W2(RTYPE, in2, in3, shift_val);                      \
   1064 } while (0)
   1065 #define SRAI_W4_SW(...) SRAI_W4(v4i32, __VA_ARGS__)
   1066 #define SRAI_W4_UW(...) SRAI_W4(v4u32, __VA_ARGS__)
   1067 
   1068 /* Description : Arithmetic shift right all elements of half-word vector
   1069 * Arguments   : Inputs  - in0, in1, shift
   1070 *               Outputs - in place operation
   1071 *               Return Type - as per input vector RTYPE
   1072 * Details     : Each element of vector 'in0' is right shifted by 'shift' and
   1073 *               the result is written in-place. 'shift' is a GP variable.
   1074 */
   1075 #define SRAI_H2(RTYPE, in0, in1, shift_val) do {  \
   1076  in0 = (RTYPE)SRAI_H(in0, shift_val);            \
   1077  in1 = (RTYPE)SRAI_H(in1, shift_val);            \
   1078 } while (0)
   1079 #define SRAI_H2_SH(...) SRAI_H2(v8i16, __VA_ARGS__)
   1080 #define SRAI_H2_UH(...) SRAI_H2(v8u16, __VA_ARGS__)
   1081 
   1082 /* Description : Arithmetic rounded shift right all elements of word vector
   1083 * Arguments   : Inputs  - in0, in1, shift
   1084 *               Outputs - in place operation
   1085 *               Return Type - as per input vector RTYPE
   1086 * Details     : Each element of vector 'in0' is right shifted by 'shift' and
   1087 *               the result is written in-place. 'shift' is a GP variable.
   1088 */
   1089 #define SRARI_W2(RTYPE, in0, in1, shift) do {     \
   1090  in0 = (RTYPE)__msa_srari_w((v4i32)in0, shift);  \
   1091  in1 = (RTYPE)__msa_srari_w((v4i32)in1, shift);  \
   1092 } while (0)
   1093 #define SRARI_W2_SW(...) SRARI_W2(v4i32, __VA_ARGS__)
   1094 
   1095 #define SRARI_W4(RTYPE, in0, in1, in2, in3, shift) do {  \
   1096  SRARI_W2(RTYPE, in0, in1, shift);                      \
   1097  SRARI_W2(RTYPE, in2, in3, shift);                      \
   1098 } while (0)
   1099 #define SRARI_W4_SH(...) SRARI_W4(v8i16, __VA_ARGS__)
   1100 #define SRARI_W4_UW(...) SRARI_W4(v4u32, __VA_ARGS__)
   1101 #define SRARI_W4_SW(...) SRARI_W4(v4i32, __VA_ARGS__)
   1102 
   1103 /* Description : Shift right arithmetic rounded double words
   1104 * Arguments   : Inputs  - in0, in1, shift
   1105 *               Outputs - in place operation
   1106 *               Return Type - as per RTYPE
   1107 * Details     : Each element of vector 'in0' is shifted right arithmetically by
   1108 *               the number of bits in the corresponding element in the vector
   1109 *               'shift'. The last discarded bit is added to shifted value for
   1110 *               rounding and the result is written in-place.
   1111 *               'shift' is a vector.
   1112 */
   1113 #define SRAR_D2(RTYPE, in0, in1, shift) do {            \
   1114  in0 = (RTYPE)__msa_srar_d((v2i64)in0, (v2i64)shift);  \
   1115  in1 = (RTYPE)__msa_srar_d((v2i64)in1, (v2i64)shift);  \
   1116 } while (0)
   1117 #define SRAR_D2_SW(...) SRAR_D2(v4i32, __VA_ARGS__)
   1118 #define SRAR_D2_SD(...) SRAR_D2(v2i64, __VA_ARGS__)
   1119 #define SRAR_D2_UD(...) SRAR_D2(v2u64, __VA_ARGS__)
   1120 
   1121 #define SRAR_D4(RTYPE, in0, in1, in2, in3, shift) do {  \
   1122  SRAR_D2(RTYPE, in0, in1, shift);                      \
   1123  SRAR_D2(RTYPE, in2, in3, shift);                      \
   1124 } while (0)
   1125 #define SRAR_D4_SD(...) SRAR_D4(v2i64, __VA_ARGS__)
   1126 #define SRAR_D4_UD(...) SRAR_D4(v2u64, __VA_ARGS__)
   1127 
   1128 /* Description : Addition of 2 pairs of half-word vectors
   1129 * Arguments   : Inputs  - in0, in1, in2, in3
   1130 *               Outputs - out0, out1
   1131 * Details     : Each element in 'in0' is added to 'in1' and result is written
   1132 *               to 'out0'.
   1133 */
   1134 #define ADDVI_H2(RTYPE, in0, in1, in2, in3, out0, out1) do {  \
   1135  out0 = (RTYPE)ADDVI_H(in0, in1);                            \
   1136  out1 = (RTYPE)ADDVI_H(in2, in3);                            \
   1137 } while (0)
   1138 #define ADDVI_H2_SH(...) ADDVI_H2(v8i16, __VA_ARGS__)
   1139 #define ADDVI_H2_UH(...) ADDVI_H2(v8u16, __VA_ARGS__)
   1140 
   1141 /* Description : Addition of 2 pairs of word vectors
   1142 * Arguments   : Inputs  - in0, in1, in2, in3
   1143 *               Outputs - out0, out1
   1144 * Details     : Each element in 'in0' is added to 'in1' and result is written
   1145 *               to 'out0'.
   1146 */
   1147 #define ADDVI_W2(RTYPE, in0, in1, in2, in3, out0, out1) do {  \
   1148  out0 = (RTYPE)ADDVI_W(in0, in1);                            \
   1149  out1 = (RTYPE)ADDVI_W(in2, in3);                            \
   1150 } while (0)
   1151 #define ADDVI_W2_SW(...) ADDVI_W2(v4i32, __VA_ARGS__)
   1152 
   1153 /* Description : Fill 2 pairs of word vectors with GP registers
   1154 * Arguments   : Inputs  - in0, in1
   1155 *               Outputs - out0, out1
   1156 * Details     : GP register in0 is replicated in each word element of out0
   1157 *               GP register in1 is replicated in each word element of out1
   1158 */
   1159 #define FILL_W2(RTYPE, in0, in1, out0, out1) do {  \
   1160  out0 = (RTYPE)__msa_fill_w(in0);                 \
   1161  out1 = (RTYPE)__msa_fill_w(in1);                 \
   1162 } while (0)
   1163 #define FILL_W2_SW(...) FILL_W2(v4i32, __VA_ARGS__)
   1164 
   1165 /* Description : Addition of 2 pairs of vectors
   1166 * Arguments   : Inputs  - in0, in1, in2, in3
   1167 *               Outputs - out0, out1
   1168 * Details     : Each element in 'in0' is added to 'in1' and result is written
   1169 *               to 'out0'.
   1170 */
   1171 #define ADD2(in0, in1, in2, in3, out0, out1) do {  \
   1172  out0 = in0 + in1;                                \
   1173  out1 = in2 + in3;                                \
   1174 } while (0)
   1175 
   1176 #define ADD4(in0, in1, in2, in3, in4, in5, in6, in7,  \
   1177             out0, out1, out2, out3) do {             \
   1178  ADD2(in0, in1, in2, in3, out0, out1);               \
   1179  ADD2(in4, in5, in6, in7, out2, out3);               \
   1180 } while (0)
   1181 
   1182 /* Description : Subtraction of 2 pairs of vectors
   1183 * Arguments   : Inputs  - in0, in1, in2, in3
   1184 *               Outputs - out0, out1
   1185 * Details     : Each element in 'in1' is subtracted from 'in0' and result is
   1186 *               written to 'out0'.
   1187 */
   1188 #define SUB2(in0, in1, in2, in3, out0, out1) do {  \
   1189  out0 = in0 - in1;                                \
   1190  out1 = in2 - in3;                                \
   1191 } while (0)
   1192 
   1193 #define SUB3(in0, in1, in2, in3, in4, in5, out0, out1, out2) do {  \
   1194  out0 = in0 - in1;                                                \
   1195  out1 = in2 - in3;                                                \
   1196  out2 = in4 - in5;                                                \
   1197 } while (0)
   1198 
   1199 #define SUB4(in0, in1, in2, in3, in4, in5, in6, in7,  \
   1200             out0, out1, out2, out3) do {             \
   1201  out0 = in0 - in1;                                   \
   1202  out1 = in2 - in3;                                   \
   1203  out2 = in4 - in5;                                   \
   1204  out3 = in6 - in7;                                   \
   1205 } while (0)
   1206 
   1207 /* Description : Addition - Subtraction of input vectors
   1208 * Arguments   : Inputs  - in0, in1
   1209 *               Outputs - out0, out1
   1210 * Details     : Each element in 'in1' is added to 'in0' and result is
   1211 *               written to 'out0'.
   1212 *               Each element in 'in1' is subtracted from 'in0' and result is
   1213 *               written to 'out1'.
   1214 */
   1215 #define ADDSUB2(in0, in1, out0, out1) do {  \
   1216  out0 = in0 + in1;                         \
   1217  out1 = in0 - in1;                         \
   1218 } while (0)
   1219 
   1220 /* Description : Multiplication of pairs of vectors
   1221 * Arguments   : Inputs  - in0, in1, in2, in3
   1222 *               Outputs - out0, out1
   1223 * Details     : Each element from 'in0' is multiplied with elements from 'in1'
   1224 *               and the result is written to 'out0'
   1225 */
   1226 #define MUL2(in0, in1, in2, in3, out0, out1) do {  \
   1227  out0 = in0 * in1;                                \
   1228  out1 = in2 * in3;                                \
   1229 } while (0)
   1230 
   1231 #define MUL4(in0, in1, in2, in3, in4, in5, in6, in7,  \
   1232             out0, out1, out2, out3) do {             \
   1233  MUL2(in0, in1, in2, in3, out0, out1);               \
   1234  MUL2(in4, in5, in6, in7, out2, out3);               \
   1235 } while (0)
   1236 
   1237 /* Description : Sign extend halfword elements from right half of the vector
   1238 * Arguments   : Input  - in    (halfword vector)
   1239 *               Output - out   (sign extended word vector)
   1240 *               Return Type - signed word
   1241 * Details     : Sign bit of halfword elements from input vector 'in' is
   1242 *               extracted and interleaved with same vector 'in0' to generate
   1243 *               4 word elements keeping sign intact
   1244 */
   1245 #define UNPCK_R_SH_SW(in, out) do {                   \
   1246  const v8i16 sign_m = __msa_clti_s_h((v8i16)in, 0);  \
   1247  out = (v4i32)__msa_ilvr_h(sign_m, (v8i16)in);       \
   1248 } while (0)
   1249 
   1250 /* Description : Sign extend halfword elements from input vector and return
   1251 *               the result in pair of vectors
   1252 * Arguments   : Input   - in            (halfword vector)
   1253 *               Outputs - out0, out1   (sign extended word vectors)
   1254 *               Return Type - signed word
   1255 * Details     : Sign bit of halfword elements from input vector 'in' is
   1256 *               extracted and interleaved right with same vector 'in0' to
   1257 *               generate 4 signed word elements in 'out0'
   1258 *               Then interleaved left with same vector 'in0' to
   1259 *               generate 4 signed word elements in 'out1'
   1260 */
   1261 #define UNPCK_SH_SW(in, out0, out1) do {              \
   1262  const v8i16 tmp_m = __msa_clti_s_h((v8i16)in, 0);   \
   1263  ILVRL_H2_SW(tmp_m, in, out0, out1);                 \
   1264 } while (0)
   1265 
   1266 /* Description : Butterfly of 4 input vectors
   1267 * Arguments   : Inputs  - in0, in1, in2, in3
   1268 *               Outputs - out0, out1, out2, out3
   1269 * Details     : Butterfly operation
   1270 */
   1271 #define BUTTERFLY_4(in0, in1, in2, in3, out0, out1, out2, out3) do {  \
   1272  out0 = in0 + in3;                                                   \
   1273  out1 = in1 + in2;                                                   \
   1274  out2 = in1 - in2;                                                   \
   1275  out3 = in0 - in3;                                                   \
   1276 } while (0)
   1277 
   1278 /* Description : Transpose 16x4 block into 4x16 with byte elements in vectors
   1279 * Arguments   : Inputs  - in0, in1, in2, in3, in4, in5, in6, in7,
   1280 *                         in8, in9, in10, in11, in12, in13, in14, in15
   1281 *               Outputs - out0, out1, out2, out3
   1282 *               Return Type - unsigned byte
   1283 */
   1284 #define TRANSPOSE16x4_UB_UB(in0, in1, in2, in3, in4, in5, in6, in7,        \
   1285                            in8, in9, in10, in11, in12, in13, in14, in15,  \
   1286                            out0, out1, out2, out3) do {                   \
   1287  v2i64 tmp0_m, tmp1_m, tmp2_m, tmp3_m, tmp4_m, tmp5_m;                    \
   1288  ILVEV_W2_SD(in0, in4, in8, in12, tmp2_m, tmp3_m);                        \
   1289  ILVEV_W2_SD(in1, in5, in9, in13, tmp0_m, tmp1_m);                        \
   1290  ILVEV_D2_UB(tmp2_m, tmp3_m, tmp0_m, tmp1_m, out1, out3);                 \
   1291  ILVEV_W2_SD(in2, in6, in10, in14, tmp4_m, tmp5_m);                       \
   1292  ILVEV_W2_SD(in3, in7, in11, in15, tmp0_m, tmp1_m);                       \
   1293  ILVEV_D2_SD(tmp4_m, tmp5_m, tmp0_m, tmp1_m, tmp2_m, tmp3_m);             \
   1294  ILVEV_B2_SD(out1, out3, tmp2_m, tmp3_m, tmp0_m, tmp1_m);                 \
   1295  ILVEVOD_H2_UB(tmp0_m, tmp1_m, tmp0_m, tmp1_m, out0, out2);               \
   1296  ILVOD_B2_SD(out1, out3, tmp2_m, tmp3_m, tmp0_m, tmp1_m);                 \
   1297  ILVEVOD_H2_UB(tmp0_m, tmp1_m, tmp0_m, tmp1_m, out1, out3);               \
   1298 } while (0)
   1299 
   1300 /* Description : Transpose 16x8 block into 8x16 with byte elements in vectors
   1301 * Arguments   : Inputs  - in0, in1, in2, in3, in4, in5, in6, in7,
   1302 *                         in8, in9, in10, in11, in12, in13, in14, in15
   1303 *               Outputs - out0, out1, out2, out3, out4, out5, out6, out7
   1304 *               Return Type - unsigned byte
   1305 */
   1306 #define TRANSPOSE16x8_UB_UB(in0, in1, in2, in3, in4, in5, in6, in7,        \
   1307                            in8, in9, in10, in11, in12, in13, in14, in15,  \
   1308                            out0, out1, out2, out3, out4, out5,            \
   1309                            out6, out7) do {                               \
   1310  v8i16 tmp0_m, tmp1_m, tmp4_m, tmp5_m, tmp6_m, tmp7_m;                    \
   1311  v4i32 tmp2_m, tmp3_m;                                                    \
   1312  ILVEV_D2_UB(in0, in8, in1, in9, out7, out6);                             \
   1313  ILVEV_D2_UB(in2, in10, in3, in11, out5, out4);                           \
   1314  ILVEV_D2_UB(in4, in12, in5, in13, out3, out2);                           \
   1315  ILVEV_D2_UB(in6, in14, in7, in15, out1, out0);                           \
   1316  ILVEV_B2_SH(out7, out6, out5, out4, tmp0_m, tmp1_m);                     \
   1317  ILVOD_B2_SH(out7, out6, out5, out4, tmp4_m, tmp5_m);                     \
   1318  ILVEV_B2_UB(out3, out2, out1, out0, out5, out7);                         \
   1319  ILVOD_B2_SH(out3, out2, out1, out0, tmp6_m, tmp7_m);                     \
   1320  ILVEV_H2_SW(tmp0_m, tmp1_m, out5, out7, tmp2_m, tmp3_m);                 \
   1321  ILVEVOD_W2_UB(tmp2_m, tmp3_m, tmp2_m, tmp3_m, out0, out4);               \
   1322  ILVOD_H2_SW(tmp0_m, tmp1_m, out5, out7, tmp2_m, tmp3_m);                 \
   1323  ILVEVOD_W2_UB(tmp2_m, tmp3_m, tmp2_m, tmp3_m, out2, out6);               \
   1324  ILVEV_H2_SW(tmp4_m, tmp5_m, tmp6_m, tmp7_m, tmp2_m, tmp3_m);             \
   1325  ILVEVOD_W2_UB(tmp2_m, tmp3_m, tmp2_m, tmp3_m, out1, out5);               \
   1326  ILVOD_H2_SW(tmp4_m, tmp5_m, tmp6_m, tmp7_m, tmp2_m, tmp3_m);             \
   1327  ILVEVOD_W2_UB(tmp2_m, tmp3_m, tmp2_m, tmp3_m, out3, out7);               \
   1328 } while (0)
   1329 
   1330 /* Description : Transpose 4x4 block with word elements in vectors
   1331 * Arguments   : Inputs  - in0, in1, in2, in3
   1332 *                Outputs - out0, out1, out2, out3
   1333 *                Return Type - as per RTYPE
   1334 */
   1335 #define TRANSPOSE4x4_W(RTYPE, in0, in1, in2, in3,                            \
   1336                       out0, out1, out2, out3) do {                          \
   1337  v4i32 s0_m, s1_m, s2_m, s3_m;                                              \
   1338  ILVRL_W2_SW(in1, in0, s0_m, s1_m);                                         \
   1339  ILVRL_W2_SW(in3, in2, s2_m, s3_m);                                         \
   1340  out0 = (RTYPE)__msa_ilvr_d((v2i64)s2_m, (v2i64)s0_m);                      \
   1341  out1 = (RTYPE)__msa_ilvl_d((v2i64)s2_m, (v2i64)s0_m);                      \
   1342  out2 = (RTYPE)__msa_ilvr_d((v2i64)s3_m, (v2i64)s1_m);                      \
   1343  out3 = (RTYPE)__msa_ilvl_d((v2i64)s3_m, (v2i64)s1_m);                      \
   1344 } while (0)
   1345 #define TRANSPOSE4x4_SW_SW(...) TRANSPOSE4x4_W(v4i32, __VA_ARGS__)
   1346 
   1347 /* Description : Add block 4x4
   1348 * Arguments   : Inputs - in0, in1, in2, in3, pdst, stride
   1349 * Details     : Least significant 4 bytes from each input vector are added to
   1350 *               the destination bytes, clipped between 0-255 and stored.
   1351 */
   1352 #define ADDBLK_ST4x4_UB(in0, in1, in2, in3, pdst, stride) do {  \
   1353  uint32_t src0_m, src1_m, src2_m, src3_m;                      \
   1354  v8i16 inp0_m, inp1_m, res0_m, res1_m;                         \
   1355  v16i8 dst0_m = { 0 };                                         \
   1356  v16i8 dst1_m = { 0 };                                         \
   1357  const v16i8 zero_m = { 0 };                                   \
   1358  ILVR_D2_SH(in1, in0, in3, in2, inp0_m, inp1_m);               \
   1359  LW4(pdst, stride, src0_m, src1_m, src2_m, src3_m);            \
   1360  INSERT_W2_SB(src0_m, src1_m, dst0_m);                         \
   1361  INSERT_W2_SB(src2_m, src3_m, dst1_m);                         \
   1362  ILVR_B2_SH(zero_m, dst0_m, zero_m, dst1_m, res0_m, res1_m);   \
   1363  ADD2(res0_m, inp0_m, res1_m, inp1_m, res0_m, res1_m);         \
   1364  CLIP_SH2_0_255(res0_m, res1_m);                               \
   1365  PCKEV_B2_SB(res0_m, res0_m, res1_m, res1_m, dst0_m, dst1_m);  \
   1366  ST4x4_UB(dst0_m, dst1_m, 0, 1, 0, 1, pdst, stride);           \
   1367 } while (0)
   1368 
   1369 /* Description : Pack even byte elements, extract 0 & 2 index words from pair
   1370 *               of results and store 4 words in destination memory as per
   1371 *               stride
   1372 * Arguments   : Inputs - in0, in1, in2, in3, pdst, stride
   1373 */
   1374 #define PCKEV_ST4x4_UB(in0, in1, in2, in3, pdst, stride) do {  \
   1375  v16i8 tmp0_m, tmp1_m;                                        \
   1376  PCKEV_B2_SB(in1, in0, in3, in2, tmp0_m, tmp1_m);             \
   1377  ST4x4_UB(tmp0_m, tmp1_m, 0, 2, 0, 2, pdst, stride);          \
   1378 } while (0)
   1379 
   1380 /* Description : average with rounding (in0 + in1 + 1) / 2.
   1381 * Arguments   : Inputs  - in0, in1, in2, in3,
   1382 *               Outputs - out0, out1
   1383 *               Return Type - as per RTYPE
   1384 * Details     : Each unsigned byte element from 'in0' vector is added with
   1385 *               each unsigned byte element from 'in1' vector. Then the average
   1386 *               with rounding is calculated and written to 'out0'
   1387 */
   1388 #define AVER_UB2(RTYPE, in0, in1, in2, in3, out0, out1) do {  \
   1389  out0 = (RTYPE)__msa_aver_u_b((v16u8)in0, (v16u8)in1);       \
   1390  out1 = (RTYPE)__msa_aver_u_b((v16u8)in2, (v16u8)in3);       \
   1391 } while (0)
   1392 #define AVER_UB2_UB(...) AVER_UB2(v16u8, __VA_ARGS__)
   1393 
   1394 #endif  // WEBP_USE_MSA
   1395 #endif  // WEBP_DSP_MSA_MACRO_H_