macros_msa.h 96.3 KB
Newer Older
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18
/*
 *  Copyright (c) 2015 The WebM project authors. All Rights Reserved.
 *
 *  Use of this source code is governed by a BSD-style license
 *  that can be found in the LICENSE file in the root of the source
 *  tree. An additional intellectual property rights grant can be found
 *  in the file PATENTS.  All contributing project authors may
 *  be found in the AUTHORS file in the root of the source tree.
 */

#ifndef VPX_DSP_MIPS_MACROS_MSA_H_
#define VPX_DSP_MIPS_MACROS_MSA_H_

#include <msa.h>

#include "./vpx_config.h"
#include "vpx/vpx_integer.h"

Kaustubh Raste's avatar
Kaustubh Raste committed
19 20 21 22 23 24 25 26 27 28 29 30
#define LD_V(RTYPE, psrc) *((const RTYPE *)(psrc))
#define LD_UB(...) LD_V(v16u8, __VA_ARGS__)
#define LD_SB(...) LD_V(v16i8, __VA_ARGS__)
#define LD_UH(...) LD_V(v8u16, __VA_ARGS__)
#define LD_SH(...) LD_V(v8i16, __VA_ARGS__)
#define LD_SW(...) LD_V(v4i32, __VA_ARGS__)

#define ST_V(RTYPE, in, pdst) *((RTYPE *)(pdst)) = (in)
#define ST_UB(...) ST_V(v16u8, __VA_ARGS__)
#define ST_SB(...) ST_V(v16i8, __VA_ARGS__)
#define ST_SH(...) ST_V(v8i16, __VA_ARGS__)
#define ST_SW(...) ST_V(v4i32, __VA_ARGS__)
Johann's avatar
Johann committed
31

32
#if (__mips_isa_rev >= 6)
Kaustubh Raste's avatar
Kaustubh Raste committed
33 34 35 36
#define LH(psrc)                                   \
  ({                                               \
    uint16_t val_lh_m = *(const uint16_t *)(psrc); \
    val_lh_m;                                      \
clang-format's avatar
clang-format committed
37 38
  })

Kaustubh Raste's avatar
Kaustubh Raste committed
39 40 41 42
#define LW(psrc)                                   \
  ({                                               \
    uint32_t val_lw_m = *(const uint32_t *)(psrc); \
    val_lw_m;                                      \
clang-format's avatar
clang-format committed
43
  })
44 45

#if (__mips == 64)
Kaustubh Raste's avatar
Kaustubh Raste committed
46 47 48 49
#define LD(psrc)                                   \
  ({                                               \
    uint64_t val_ld_m = *(const uint64_t *)(psrc); \
    val_ld_m;                                      \
clang-format's avatar
clang-format committed
50
  })
51
#else  // !(__mips == 64)
Kaustubh Raste's avatar
Kaustubh Raste committed
52 53 54 55 56 57 58 59 60 61 62 63 64 65
#define LD(psrc)                                                  \
  ({                                                              \
    const uint8_t *psrc_ld_m = (const uint8_t *)(psrc);           \
    uint32_t val0_ld_m, val1_ld_m;                                \
    uint64_t val_ld_m = 0;                                        \
                                                                  \
    val0_ld_m = LW(psrc_ld_m);                                    \
    val1_ld_m = LW(psrc_ld_m + 4);                                \
                                                                  \
    val_ld_m = (uint64_t)(val1_ld_m);                             \
    val_ld_m = (uint64_t)((val_ld_m << 32) & 0xFFFFFFFF00000000); \
    val_ld_m = (uint64_t)(val_ld_m | (uint64_t)val0_ld_m);        \
                                                                  \
    val_ld_m;                                                     \
clang-format's avatar
clang-format committed
66 67 68
  })
#endif  // (__mips == 64)

Kaustubh Raste's avatar
Kaustubh Raste committed
69 70 71
#define SH(val, pdst) *(uint16_t *)(pdst) = (val);
#define SW(val, pdst) *(uint32_t *)(pdst) = (val);
#define SD(val, pdst) *(uint64_t *)(pdst) = (val);
clang-format's avatar
clang-format committed
72
#else  // !(__mips_isa_rev >= 6)
Kaustubh Raste's avatar
Kaustubh Raste committed
73 74 75 76 77 78 79 80 81 82 83
#define LH(psrc)                                                 \
  ({                                                             \
    const uint8_t *psrc_lh_m = (const uint8_t *)(psrc);          \
    uint16_t val_lh_m;                                           \
                                                                 \
    __asm__ __volatile__("ulh  %[val_lh_m],  %[psrc_lh_m]  \n\t" \
                                                                 \
                         : [val_lh_m] "=r"(val_lh_m)             \
                         : [psrc_lh_m] "m"(*psrc_lh_m));         \
                                                                 \
    val_lh_m;                                                    \
clang-format's avatar
clang-format committed
84
  })
85

Kaustubh Raste's avatar
Kaustubh Raste committed
86 87 88 89 90 91 92 93 94 95 96
#define LW(psrc)                                                 \
  ({                                                             \
    const uint8_t *psrc_lw_m = (const uint8_t *)(psrc);          \
    uint32_t val_lw_m;                                           \
                                                                 \
    __asm__ __volatile__("ulw  %[val_lw_m],  %[psrc_lw_m]  \n\t" \
                                                                 \
                         : [val_lw_m] "=r"(val_lw_m)             \
                         : [psrc_lw_m] "m"(*psrc_lw_m));         \
                                                                 \
    val_lw_m;                                                    \
clang-format's avatar
clang-format committed
97
  })
98 99

#if (__mips == 64)
Kaustubh Raste's avatar
Kaustubh Raste committed
100 101 102 103 104 105 106 107 108 109 110
#define LD(psrc)                                                 \
  ({                                                             \
    const uint8_t *psrc_ld_m = (const uint8_t *)(psrc);          \
    uint64_t val_ld_m = 0;                                       \
                                                                 \
    __asm__ __volatile__("uld  %[val_ld_m],  %[psrc_ld_m]  \n\t" \
                                                                 \
                         : [val_ld_m] "=r"(val_ld_m)             \
                         : [psrc_ld_m] "m"(*psrc_ld_m));         \
                                                                 \
    val_ld_m;                                                    \
clang-format's avatar
clang-format committed
111 112
  })
#else  // !(__mips == 64)
Kaustubh Raste's avatar
Kaustubh Raste committed
113 114 115 116 117 118 119 120 121 122 123 124 125 126
#define LD(psrc)                                                  \
  ({                                                              \
    const uint8_t *psrc_ld_m = (const uint8_t *)(psrc);           \
    uint32_t val0_ld_m, val1_ld_m;                                \
    uint64_t val_ld_m = 0;                                        \
                                                                  \
    val0_ld_m = LW(psrc_ld_m);                                    \
    val1_ld_m = LW(psrc_ld_m + 4);                                \
                                                                  \
    val_ld_m = (uint64_t)(val1_ld_m);                             \
    val_ld_m = (uint64_t)((val_ld_m << 32) & 0xFFFFFFFF00000000); \
    val_ld_m = (uint64_t)(val_ld_m | (uint64_t)val0_ld_m);        \
                                                                  \
    val_ld_m;                                                     \
clang-format's avatar
clang-format committed
127 128 129
  })
#endif  // (__mips == 64)

Kaustubh Raste's avatar
Kaustubh Raste committed
130 131 132 133 134 135 136 137 138
#define SH(val, pdst)                                            \
  {                                                              \
    uint8_t *pdst_sh_m = (uint8_t *)(pdst);                      \
    const uint16_t val_sh_m = (val);                             \
                                                                 \
    __asm__ __volatile__("ush  %[val_sh_m],  %[pdst_sh_m]  \n\t" \
                                                                 \
                         : [pdst_sh_m] "=m"(*pdst_sh_m)          \
                         : [val_sh_m] "r"(val_sh_m));            \
clang-format's avatar
clang-format committed
139 140
  }

Kaustubh Raste's avatar
Kaustubh Raste committed
141 142 143 144 145 146 147 148 149
#define SW(val, pdst)                                            \
  {                                                              \
    uint8_t *pdst_sw_m = (uint8_t *)(pdst);                      \
    const uint32_t val_sw_m = (val);                             \
                                                                 \
    __asm__ __volatile__("usw  %[val_sw_m],  %[pdst_sw_m]  \n\t" \
                                                                 \
                         : [pdst_sw_m] "=m"(*pdst_sw_m)          \
                         : [val_sw_m] "r"(val_sw_m));            \
clang-format's avatar
clang-format committed
150 151
  }

Kaustubh Raste's avatar
Kaustubh Raste committed
152 153 154 155 156 157 158 159 160 161
#define SD(val, pdst)                                           \
  {                                                             \
    uint8_t *pdst_sd_m = (uint8_t *)(pdst);                     \
    uint32_t val0_sd_m, val1_sd_m;                              \
                                                                \
    val0_sd_m = (uint32_t)((val)&0x00000000FFFFFFFF);           \
    val1_sd_m = (uint32_t)(((val) >> 32) & 0x00000000FFFFFFFF); \
                                                                \
    SW(val0_sd_m, pdst_sd_m);                                   \
    SW(val1_sd_m, pdst_sd_m + 4);                               \
clang-format's avatar
clang-format committed
162
  }
163 164 165 166 167 168 169 170 171 172
#endif  // (__mips_isa_rev >= 6)

/* Description : Load 4 words with stride
   Arguments   : Inputs  - psrc, stride
                 Outputs - out0, out1, out2, out3
   Details     : Load word in 'out0' from (psrc)
                 Load word in 'out1' from (psrc + stride)
                 Load word in 'out2' from (psrc + 2 * stride)
                 Load word in 'out3' from (psrc + 3 * stride)
*/
clang-format's avatar
clang-format committed
173 174 175 176 177 178 179
#define LW4(psrc, stride, out0, out1, out2, out3) \
  {                                               \
    out0 = LW((psrc));                            \
    out1 = LW((psrc) + stride);                   \
    out2 = LW((psrc) + 2 * stride);               \
    out3 = LW((psrc) + 3 * stride);               \
  }
180

181 182 183 184 185 186
/* Description : Load double words with stride
   Arguments   : Inputs  - psrc, stride
                 Outputs - out0, out1
   Details     : Load double word in 'out0' from (psrc)
                 Load double word in 'out1' from (psrc + stride)
*/
clang-format's avatar
clang-format committed
187 188 189 190 191 192 193 194 195 196
#define LD2(psrc, stride, out0, out1) \
  {                                   \
    out0 = LD((psrc));                \
    out1 = LD((psrc) + stride);       \
  }
#define LD4(psrc, stride, out0, out1, out2, out3) \
  {                                               \
    LD2((psrc), stride, out0, out1);              \
    LD2((psrc) + 2 * stride, stride, out2, out3); \
  }
197

Johann's avatar
Johann committed
198 199 200 201 202 203 204
/* Description : Store 4 words with stride
   Arguments   : Inputs - in0, in1, in2, in3, pdst, stride
   Details     : Store word from 'in0' to (pdst)
                 Store word from 'in1' to (pdst + stride)
                 Store word from 'in2' to (pdst + 2 * stride)
                 Store word from 'in3' to (pdst + 3 * stride)
*/
clang-format's avatar
clang-format committed
205 206 207 208 209 210 211
#define SW4(in0, in1, in2, in3, pdst, stride) \
  {                                           \
    SW(in0, (pdst))                           \
    SW(in1, (pdst) + stride);                 \
    SW(in2, (pdst) + 2 * stride);             \
    SW(in3, (pdst) + 3 * stride);             \
  }
Johann's avatar
Johann committed
212 213 214 215 216 217 218 219

/* Description : Store 4 double words with stride
   Arguments   : Inputs - in0, in1, in2, in3, pdst, stride
   Details     : Store double word from 'in0' to (pdst)
                 Store double word from 'in1' to (pdst + stride)
                 Store double word from 'in2' to (pdst + 2 * stride)
                 Store double word from 'in3' to (pdst + 3 * stride)
*/
clang-format's avatar
clang-format committed
220 221 222 223 224 225 226
#define SD4(in0, in1, in2, in3, pdst, stride) \
  {                                           \
    SD(in0, (pdst))                           \
    SD(in1, (pdst) + stride);                 \
    SD(in2, (pdst) + 2 * stride);             \
    SD(in3, (pdst) + 3 * stride);             \
  }
Johann's avatar
Johann committed
227

Kaustubh Raste's avatar
Kaustubh Raste committed
228
/* Description : Load vector elements with stride
229 230 231 232 233 234
   Arguments   : Inputs  - psrc, stride
                 Outputs - out0, out1
                 Return Type - as per RTYPE
   Details     : Load 16 byte elements in 'out0' from (psrc)
                 Load 16 byte elements in 'out1' from (psrc + stride)
*/
Kaustubh Raste's avatar
Kaustubh Raste committed
235
#define LD_V2(RTYPE, psrc, stride, out0, out1) \
clang-format's avatar
clang-format committed
236
  {                                            \
Kaustubh Raste's avatar
Kaustubh Raste committed
237 238
    out0 = LD_V(RTYPE, (psrc));                \
    out1 = LD_V(RTYPE, (psrc) + stride);       \
clang-format's avatar
clang-format committed
239
  }
Kaustubh Raste's avatar
Kaustubh Raste committed
240 241 242 243
#define LD_UB2(...) LD_V2(v16u8, __VA_ARGS__)
#define LD_SB2(...) LD_V2(v16i8, __VA_ARGS__)
#define LD_SH2(...) LD_V2(v8i16, __VA_ARGS__)
#define LD_SW2(...) LD_V2(v4i32, __VA_ARGS__)
244

Kaustubh Raste's avatar
Kaustubh Raste committed
245
#define LD_V3(RTYPE, psrc, stride, out0, out1, out2) \
clang-format's avatar
clang-format committed
246
  {                                                  \
Kaustubh Raste's avatar
Kaustubh Raste committed
247 248
    LD_V2(RTYPE, (psrc), stride, out0, out1);        \
    out2 = LD_V(RTYPE, (psrc) + 2 * stride);         \
clang-format's avatar
clang-format committed
249
  }
Kaustubh Raste's avatar
Kaustubh Raste committed
250
#define LD_UB3(...) LD_V3(v16u8, __VA_ARGS__)
251

Kaustubh Raste's avatar
Kaustubh Raste committed
252
#define LD_V4(RTYPE, psrc, stride, out0, out1, out2, out3) \
clang-format's avatar
clang-format committed
253
  {                                                        \
Kaustubh Raste's avatar
Kaustubh Raste committed
254 255
    LD_V2(RTYPE, (psrc), stride, out0, out1);              \
    LD_V2(RTYPE, (psrc) + 2 * stride, stride, out2, out3); \
clang-format's avatar
clang-format committed
256
  }
Kaustubh Raste's avatar
Kaustubh Raste committed
257 258 259
#define LD_UB4(...) LD_V4(v16u8, __VA_ARGS__)
#define LD_SB4(...) LD_V4(v16i8, __VA_ARGS__)
#define LD_SH4(...) LD_V4(v8i16, __VA_ARGS__)
260

Kaustubh Raste's avatar
Kaustubh Raste committed
261
#define LD_V5(RTYPE, psrc, stride, out0, out1, out2, out3, out4) \
clang-format's avatar
clang-format committed
262
  {                                                              \
Kaustubh Raste's avatar
Kaustubh Raste committed
263 264
    LD_V4(RTYPE, (psrc), stride, out0, out1, out2, out3);        \
    out4 = LD_V(RTYPE, (psrc) + 4 * stride);                     \
clang-format's avatar
clang-format committed
265
  }
Kaustubh Raste's avatar
Kaustubh Raste committed
266 267
#define LD_UB5(...) LD_V5(v16u8, __VA_ARGS__)
#define LD_SB5(...) LD_V5(v16i8, __VA_ARGS__)
268

Kaustubh Raste's avatar
Kaustubh Raste committed
269
#define LD_V7(RTYPE, psrc, stride, out0, out1, out2, out3, out4, out5, out6) \
clang-format's avatar
clang-format committed
270
  {                                                                          \
Kaustubh Raste's avatar
Kaustubh Raste committed
271 272
    LD_V5(RTYPE, (psrc), stride, out0, out1, out2, out3, out4);              \
    LD_V2(RTYPE, (psrc) + 5 * stride, stride, out5, out6);                   \
clang-format's avatar
clang-format committed
273
  }
Kaustubh Raste's avatar
Kaustubh Raste committed
274
#define LD_SB7(...) LD_V7(v16i8, __VA_ARGS__)
275

Kaustubh Raste's avatar
Kaustubh Raste committed
276
#define LD_V8(RTYPE, psrc, stride, out0, out1, out2, out3, out4, out5, out6, \
clang-format's avatar
clang-format committed
277 278
              out7)                                                          \
  {                                                                          \
Kaustubh Raste's avatar
Kaustubh Raste committed
279 280
    LD_V4(RTYPE, (psrc), stride, out0, out1, out2, out3);                    \
    LD_V4(RTYPE, (psrc) + 4 * stride, stride, out4, out5, out6, out7);       \
clang-format's avatar
clang-format committed
281
  }
Kaustubh Raste's avatar
Kaustubh Raste committed
282 283 284
#define LD_UB8(...) LD_V8(v16u8, __VA_ARGS__)
#define LD_SB8(...) LD_V8(v16i8, __VA_ARGS__)
#define LD_SH8(...) LD_V8(v8i16, __VA_ARGS__)
Johann's avatar
Johann committed
285

Kaustubh Raste's avatar
Kaustubh Raste committed
286
#define LD_V16(RTYPE, psrc, stride, out0, out1, out2, out3, out4, out5, out6,  \
clang-format's avatar
clang-format committed
287 288
               out7, out8, out9, out10, out11, out12, out13, out14, out15)     \
  {                                                                            \
Kaustubh Raste's avatar
Kaustubh Raste committed
289
    LD_V8(RTYPE, (psrc), stride, out0, out1, out2, out3, out4, out5, out6,     \
clang-format's avatar
clang-format committed
290
          out7);                                                               \
Kaustubh Raste's avatar
Kaustubh Raste committed
291
    LD_V8(RTYPE, (psrc) + 8 * stride, stride, out8, out9, out10, out11, out12, \
clang-format's avatar
clang-format committed
292 293
          out13, out14, out15);                                                \
  }
Kaustubh Raste's avatar
Kaustubh Raste committed
294
#define LD_SH16(...) LD_V16(v8i16, __VA_ARGS__)
Johann's avatar
Johann committed
295 296 297 298 299 300

/* Description : Load 4x4 block of signed halfword elements from 1D source
                 data into 4 vectors (Each vector with 4 signed halfwords)
   Arguments   : Input   - psrc
                 Outputs - out0, out1, out2, out3
*/
clang-format's avatar
clang-format committed
301 302 303 304 305 306 307
#define LD4x4_SH(psrc, out0, out1, out2, out3)            \
  {                                                       \
    out0 = LD_SH(psrc);                                   \
    out2 = LD_SH(psrc + 8);                               \
    out1 = (v8i16)__msa_ilvl_d((v2i64)out0, (v2i64)out0); \
    out3 = (v8i16)__msa_ilvl_d((v2i64)out2, (v2i64)out2); \
  }
Johann's avatar
Johann committed
308

Kaustubh Raste's avatar
Kaustubh Raste committed
309
/* Description : Store vectors with stride
Johann's avatar
Johann committed
310 311 312 313
   Arguments   : Inputs - in0, in1, pdst, stride
   Details     : Store 16 byte elements from 'in0' to (pdst)
                 Store 16 byte elements from 'in1' to (pdst + stride)
*/
Kaustubh Raste's avatar
Kaustubh Raste committed
314
#define ST_V2(RTYPE, in0, in1, pdst, stride) \
clang-format's avatar
clang-format committed
315
  {                                          \
Kaustubh Raste's avatar
Kaustubh Raste committed
316 317
    ST_V(RTYPE, in0, (pdst));                \
    ST_V(RTYPE, in1, (pdst) + stride);       \
clang-format's avatar
clang-format committed
318
  }
Kaustubh Raste's avatar
Kaustubh Raste committed
319 320 321
#define ST_UB2(...) ST_V2(v16u8, __VA_ARGS__)
#define ST_SH2(...) ST_V2(v8i16, __VA_ARGS__)
#define ST_SW2(...) ST_V2(v4i32, __VA_ARGS__)
Johann's avatar
Johann committed
322

Kaustubh Raste's avatar
Kaustubh Raste committed
323
#define ST_V4(RTYPE, in0, in1, in2, in3, pdst, stride)   \
clang-format's avatar
clang-format committed
324
  {                                                      \
Kaustubh Raste's avatar
Kaustubh Raste committed
325 326
    ST_V2(RTYPE, in0, in1, (pdst), stride);              \
    ST_V2(RTYPE, in2, in3, (pdst) + 2 * stride, stride); \
clang-format's avatar
clang-format committed
327
  }
Kaustubh Raste's avatar
Kaustubh Raste committed
328 329
#define ST_UB4(...) ST_V4(v16u8, __VA_ARGS__)
#define ST_SH4(...) ST_V4(v8i16, __VA_ARGS__)
Johann's avatar
Johann committed
330

Kaustubh Raste's avatar
Kaustubh Raste committed
331
#define ST_V8(RTYPE, in0, in1, in2, in3, in4, in5, in6, in7, pdst, stride) \
clang-format's avatar
clang-format committed
332
  {                                                                        \
Kaustubh Raste's avatar
Kaustubh Raste committed
333 334
    ST_V4(RTYPE, in0, in1, in2, in3, pdst, stride);                        \
    ST_V4(RTYPE, in4, in5, in6, in7, (pdst) + 4 * stride, stride);         \
clang-format's avatar
clang-format committed
335
  }
Kaustubh Raste's avatar
Kaustubh Raste committed
336 337
#define ST_UB8(...) ST_V8(v16u8, __VA_ARGS__)
#define ST_SH8(...) ST_V8(v8i16, __VA_ARGS__)
Johann's avatar
Johann committed
338 339 340 341 342 343 344 345 346 347 348 349

/* Description : Store 2x4 byte block to destination memory from input vector
   Arguments   : Inputs - in, stidx, pdst, stride
   Details     : Index 'stidx' halfword element from 'in' vector is copied to
                 the GP register and stored to (pdst)
                 Index 'stidx+1' halfword element from 'in' vector is copied to
                 the GP register and stored to (pdst + stride)
                 Index 'stidx+2' halfword element from 'in' vector is copied to
                 the GP register and stored to (pdst + 2 * stride)
                 Index 'stidx+3' halfword element from 'in' vector is copied to
                 the GP register and stored to (pdst + 3 * stride)
*/
clang-format's avatar
clang-format committed
350 351 352 353 354 355 356 357 358 359 360 361 362 363 364
#define ST2x4_UB(in, stidx, pdst, stride)            \
  {                                                  \
    uint16_t out0_m, out1_m, out2_m, out3_m;         \
    uint8_t *pblk_2x4_m = (uint8_t *)(pdst);         \
                                                     \
    out0_m = __msa_copy_u_h((v8i16)in, (stidx));     \
    out1_m = __msa_copy_u_h((v8i16)in, (stidx + 1)); \
    out2_m = __msa_copy_u_h((v8i16)in, (stidx + 2)); \
    out3_m = __msa_copy_u_h((v8i16)in, (stidx + 3)); \
                                                     \
    SH(out0_m, pblk_2x4_m);                          \
    SH(out1_m, pblk_2x4_m + stride);                 \
    SH(out2_m, pblk_2x4_m + 2 * stride);             \
    SH(out3_m, pblk_2x4_m + 3 * stride);             \
  }
Johann's avatar
Johann committed
365 366 367 368 369 370 371 372

/* Description : Store 4x2 byte block to destination memory from input vector
   Arguments   : Inputs - in, pdst, stride
   Details     : Index 0 word element from 'in' vector is copied to the GP
                 register and stored to (pdst)
                 Index 1 word element from 'in' vector is copied to the GP
                 register and stored to (pdst + stride)
*/
clang-format's avatar
clang-format committed
373 374 375 376 377 378 379 380 381 382 383
#define ST4x2_UB(in, pdst, stride)           \
  {                                          \
    uint32_t out0_m, out1_m;                 \
    uint8_t *pblk_4x2_m = (uint8_t *)(pdst); \
                                             \
    out0_m = __msa_copy_u_w((v4i32)in, 0);   \
    out1_m = __msa_copy_u_w((v4i32)in, 1);   \
                                             \
    SW(out0_m, pblk_4x2_m);                  \
    SW(out1_m, pblk_4x2_m + stride);         \
  }
Johann's avatar
Johann committed
384 385 386 387 388 389 390 391 392 393 394 395

/* Description : Store 4x4 byte block to destination memory from input vector
   Arguments   : Inputs - in0, in1, pdst, stride
   Details     : 'Idx0' word element from input vector 'in0' is copied to the
                 GP register and stored to (pdst)
                 'Idx1' word element from input vector 'in0' is copied to the
                 GP register and stored to (pdst + stride)
                 'Idx2' word element from input vector 'in0' is copied to the
                 GP register and stored to (pdst + 2 * stride)
                 'Idx3' word element from input vector 'in0' is copied to the
                 GP register and stored to (pdst + 3 * stride)
*/
clang-format's avatar
clang-format committed
396 397 398 399 400 401 402 403 404 405 406 407 408 409 410 411 412 413 414
#define ST4x4_UB(in0, in1, idx0, idx1, idx2, idx3, pdst, stride) \
  {                                                              \
    uint32_t out0_m, out1_m, out2_m, out3_m;                     \
    uint8_t *pblk_4x4_m = (uint8_t *)(pdst);                     \
                                                                 \
    out0_m = __msa_copy_u_w((v4i32)in0, idx0);                   \
    out1_m = __msa_copy_u_w((v4i32)in0, idx1);                   \
    out2_m = __msa_copy_u_w((v4i32)in1, idx2);                   \
    out3_m = __msa_copy_u_w((v4i32)in1, idx3);                   \
                                                                 \
    SW4(out0_m, out1_m, out2_m, out3_m, pblk_4x4_m, stride);     \
  }
#define ST4x8_UB(in0, in1, pdst, stride)                           \
  {                                                                \
    uint8_t *pblk_4x8 = (uint8_t *)(pdst);                         \
                                                                   \
    ST4x4_UB(in0, in0, 0, 1, 2, 3, pblk_4x8, stride);              \
    ST4x4_UB(in1, in1, 0, 1, 2, 3, pblk_4x8 + 4 * stride, stride); \
  }
Johann's avatar
Johann committed
415 416 417 418 419 420

/* Description : Store 8x1 byte block to destination memory from input vector
   Arguments   : Inputs - in, pdst
   Details     : Index 0 double word element from 'in' vector is copied to the
                 GP register and stored to (pdst)
*/
clang-format's avatar
clang-format committed
421 422 423 424 425 426 427
#define ST8x1_UB(in, pdst)                 \
  {                                        \
    uint64_t out0_m;                       \
                                           \
    out0_m = __msa_copy_u_d((v2i64)in, 0); \
    SD(out0_m, pdst);                      \
  }
Johann's avatar
Johann committed
428 429 430 431 432 433 434 435

/* Description : Store 8x2 byte block to destination memory from input vector
   Arguments   : Inputs - in, pdst, stride
   Details     : Index 0 double word element from 'in' vector is copied to the
                 GP register and stored to (pdst)
                 Index 1 double word element from 'in' vector is copied to the
                 GP register and stored to (pdst + stride)
*/
clang-format's avatar
clang-format committed
436 437 438 439 440 441 442 443 444 445 446
#define ST8x2_UB(in, pdst, stride)           \
  {                                          \
    uint64_t out0_m, out1_m;                 \
    uint8_t *pblk_8x2_m = (uint8_t *)(pdst); \
                                             \
    out0_m = __msa_copy_u_d((v2i64)in, 0);   \
    out1_m = __msa_copy_u_d((v2i64)in, 1);   \
                                             \
    SD(out0_m, pblk_8x2_m);                  \
    SD(out1_m, pblk_8x2_m + stride);         \
  }
Johann's avatar
Johann committed
447 448 449 450 451 452 453 454 455 456 457 458 459

/* Description : Store 8x4 byte block to destination memory from input
                 vectors
   Arguments   : Inputs - in0, in1, pdst, stride
   Details     : Index 0 double word element from 'in0' vector is copied to the
                 GP register and stored to (pdst)
                 Index 1 double word element from 'in0' vector is copied to the
                 GP register and stored to (pdst + stride)
                 Index 0 double word element from 'in1' vector is copied to the
                 GP register and stored to (pdst + 2 * stride)
                 Index 1 double word element from 'in1' vector is copied to the
                 GP register and stored to (pdst + 3 * stride)
*/
clang-format's avatar
clang-format committed
460 461 462 463 464 465 466 467 468 469 470 471
#define ST8x4_UB(in0, in1, pdst, stride)                     \
  {                                                          \
    uint64_t out0_m, out1_m, out2_m, out3_m;                 \
    uint8_t *pblk_8x4_m = (uint8_t *)(pdst);                 \
                                                             \
    out0_m = __msa_copy_u_d((v2i64)in0, 0);                  \
    out1_m = __msa_copy_u_d((v2i64)in0, 1);                  \
    out2_m = __msa_copy_u_d((v2i64)in1, 0);                  \
    out3_m = __msa_copy_u_d((v2i64)in1, 1);                  \
                                                             \
    SD4(out0_m, out1_m, out2_m, out3_m, pblk_8x4_m, stride); \
  }
Johann's avatar
Johann committed
472

473 474 475 476 477 478 479 480
/* Description : average with rounding (in0 + in1 + 1) / 2.
   Arguments   : Inputs  - in0, in1, in2, in3,
                 Outputs - out0, out1
                 Return Type - as per RTYPE
   Details     : Each unsigned byte element from 'in0' vector is added with
                 each unsigned byte element from 'in1' vector. Then the average
                 with rounding is calculated and written to 'out0'
*/
clang-format's avatar
clang-format committed
481 482 483 484 485
#define AVER_UB2(RTYPE, in0, in1, in2, in3, out0, out1)   \
  {                                                       \
    out0 = (RTYPE)__msa_aver_u_b((v16u8)in0, (v16u8)in1); \
    out1 = (RTYPE)__msa_aver_u_b((v16u8)in2, (v16u8)in3); \
  }
486 487
#define AVER_UB2_UB(...) AVER_UB2(v16u8, __VA_ARGS__)

clang-format's avatar
clang-format committed
488 489 490 491 492 493
#define AVER_UB4(RTYPE, in0, in1, in2, in3, in4, in5, in6, in7, out0, out1, \
                 out2, out3)                                                \
  {                                                                         \
    AVER_UB2(RTYPE, in0, in1, in2, in3, out0, out1)                         \
    AVER_UB2(RTYPE, in4, in5, in6, in7, out2, out3)                         \
  }
494 495
#define AVER_UB4_UB(...) AVER_UB4(v16u8, __VA_ARGS__)

Johann's avatar
Johann committed
496 497 498 499 500 501 502
/* Description : Immediate number of elements to slide with zero
   Arguments   : Inputs  - in0, in1, slide_val
                 Outputs - out0, out1
                 Return Type - as per RTYPE
   Details     : Byte elements from 'zero_m' vector are slid into 'in0' by
                 value specified in the 'slide_val'
*/
clang-format's avatar
clang-format committed
503 504 505 506 507 508
#define SLDI_B2_0(RTYPE, in0, in1, out0, out1, slide_val)             \
  {                                                                   \
    v16i8 zero_m = { 0 };                                             \
    out0 = (RTYPE)__msa_sldi_b((v16i8)zero_m, (v16i8)in0, slide_val); \
    out1 = (RTYPE)__msa_sldi_b((v16i8)zero_m, (v16i8)in1, slide_val); \
  }
Johann's avatar
Johann committed
509 510
#define SLDI_B2_0_SW(...) SLDI_B2_0(v4i32, __VA_ARGS__)

clang-format's avatar
clang-format committed
511 512 513 514 515 516
#define SLDI_B4_0(RTYPE, in0, in1, in2, in3, out0, out1, out2, out3, \
                  slide_val)                                         \
  {                                                                  \
    SLDI_B2_0(RTYPE, in0, in1, out0, out1, slide_val);               \
    SLDI_B2_0(RTYPE, in2, in3, out2, out3, slide_val);               \
  }
Johann's avatar
Johann committed
517 518
#define SLDI_B4_0_UB(...) SLDI_B4_0(v16u8, __VA_ARGS__)

519 520 521 522 523 524 525
/* Description : Immediate number of elements to slide
   Arguments   : Inputs  - in0_0, in0_1, in1_0, in1_1, slide_val
                 Outputs - out0, out1
                 Return Type - as per RTYPE
   Details     : Byte elements from 'in0_0' vector are slid into 'in1_0' by
                 value specified in the 'slide_val'
*/
clang-format's avatar
clang-format committed
526 527 528 529 530
#define SLDI_B2(RTYPE, in0_0, in0_1, in1_0, in1_1, out0, out1, slide_val) \
  {                                                                       \
    out0 = (RTYPE)__msa_sldi_b((v16i8)in0_0, (v16i8)in1_0, slide_val);    \
    out1 = (RTYPE)__msa_sldi_b((v16i8)in0_1, (v16i8)in1_1, slide_val);    \
  }
531
#define SLDI_B2_UB(...) SLDI_B2(v16u8, __VA_ARGS__)
Johann's avatar
Johann committed
532 533
#define SLDI_B2_SH(...) SLDI_B2(v8i16, __VA_ARGS__)

clang-format's avatar
clang-format committed
534 535 536 537 538 539
#define SLDI_B3(RTYPE, in0_0, in0_1, in0_2, in1_0, in1_1, in1_2, out0, out1, \
                out2, slide_val)                                             \
  {                                                                          \
    SLDI_B2(RTYPE, in0_0, in0_1, in1_0, in1_1, out0, out1, slide_val)        \
    out2 = (RTYPE)__msa_sldi_b((v16i8)in0_2, (v16i8)in1_2, slide_val);       \
  }
Johann's avatar
Johann committed
540 541 542 543 544 545 546 547 548 549
#define SLDI_B3_SB(...) SLDI_B3(v16i8, __VA_ARGS__)
#define SLDI_B3_UH(...) SLDI_B3(v8u16, __VA_ARGS__)

/* Description : Shuffle byte vector elements as per mask vector
   Arguments   : Inputs  - in0, in1, in2, in3, mask0, mask1
                 Outputs - out0, out1
                 Return Type - as per RTYPE
   Details     : Byte elements from 'in0' & 'in1' are copied selectively to
                 'out0' as per control vector 'mask0'
*/
clang-format's avatar
clang-format committed
550 551 552 553 554
#define VSHF_B2(RTYPE, in0, in1, in2, in3, mask0, mask1, out0, out1)  \
  {                                                                   \
    out0 = (RTYPE)__msa_vshf_b((v16i8)mask0, (v16i8)in1, (v16i8)in0); \
    out1 = (RTYPE)__msa_vshf_b((v16i8)mask1, (v16i8)in3, (v16i8)in2); \
  }
Johann's avatar
Johann committed
555 556 557
#define VSHF_B2_UB(...) VSHF_B2(v16u8, __VA_ARGS__)
#define VSHF_B2_SB(...) VSHF_B2(v16i8, __VA_ARGS__)
#define VSHF_B2_UH(...) VSHF_B2(v8u16, __VA_ARGS__)
558
#define VSHF_B2_SH(...) VSHF_B2(v8i16, __VA_ARGS__)
Johann's avatar
Johann committed
559

clang-format's avatar
clang-format committed
560 561 562 563 564 565
#define VSHF_B4(RTYPE, in0, in1, mask0, mask1, mask2, mask3, out0, out1, out2, \
                out3)                                                          \
  {                                                                            \
    VSHF_B2(RTYPE, in0, in1, in0, in1, mask0, mask1, out0, out1);              \
    VSHF_B2(RTYPE, in0, in1, in0, in1, mask2, mask3, out2, out3);              \
  }
Johann's avatar
Johann committed
566 567 568 569 570 571 572 573 574 575 576 577 578
#define VSHF_B4_SB(...) VSHF_B4(v16i8, __VA_ARGS__)
#define VSHF_B4_SH(...) VSHF_B4(v8i16, __VA_ARGS__)

/* Description : Dot product of byte vector elements
   Arguments   : Inputs  - mult0, mult1, cnst0, cnst1
                 Outputs - out0, out1
                 Return Type - as per RTYPE
   Details     : Unsigned byte elements from 'mult0' are multiplied with
                 unsigned byte elements from 'cnst0' producing a result
                 twice the size of input i.e. unsigned halfword.
                 The multiplication result of adjacent odd-even elements
                 are added together and written to the 'out0' vector
*/
clang-format's avatar
clang-format committed
579 580 581 582 583
#define DOTP_UB2(RTYPE, mult0, mult1, cnst0, cnst1, out0, out1) \
  {                                                             \
    out0 = (RTYPE)__msa_dotp_u_h((v16u8)mult0, (v16u8)cnst0);   \
    out1 = (RTYPE)__msa_dotp_u_h((v16u8)mult1, (v16u8)cnst1);   \
  }
Johann's avatar
Johann committed
584 585
#define DOTP_UB2_UH(...) DOTP_UB2(v8u16, __VA_ARGS__)

clang-format's avatar
clang-format committed
586 587 588 589 590 591
#define DOTP_UB4(RTYPE, mult0, mult1, mult2, mult3, cnst0, cnst1, cnst2, \
                 cnst3, out0, out1, out2, out3)                          \
  {                                                                      \
    DOTP_UB2(RTYPE, mult0, mult1, cnst0, cnst1, out0, out1);             \
    DOTP_UB2(RTYPE, mult2, mult3, cnst2, cnst3, out2, out3);             \
  }
Johann's avatar
Johann committed
592 593 594 595 596 597 598 599 600 601 602 603
#define DOTP_UB4_UH(...) DOTP_UB4(v8u16, __VA_ARGS__)

/* Description : Dot product of byte vector elements
   Arguments   : Inputs  - mult0, mult1, cnst0, cnst1
                 Outputs - out0, out1
                 Return Type - as per RTYPE
   Details     : Signed byte elements from 'mult0' are multiplied with
                 signed byte elements from 'cnst0' producing a result
                 twice the size of input i.e. signed halfword.
                 The multiplication result of adjacent odd-even elements
                 are added together and written to the 'out0' vector
*/
clang-format's avatar
clang-format committed
604 605 606 607 608
#define DOTP_SB2(RTYPE, mult0, mult1, cnst0, cnst1, out0, out1) \
  {                                                             \
    out0 = (RTYPE)__msa_dotp_s_h((v16i8)mult0, (v16i8)cnst0);   \
    out1 = (RTYPE)__msa_dotp_s_h((v16i8)mult1, (v16i8)cnst1);   \
  }
Johann's avatar
Johann committed
609 610
#define DOTP_SB2_SH(...) DOTP_SB2(v8i16, __VA_ARGS__)

clang-format's avatar
clang-format committed
611 612 613 614 615 616
#define DOTP_SB4(RTYPE, mult0, mult1, mult2, mult3, cnst0, cnst1, cnst2, \
                 cnst3, out0, out1, out2, out3)                          \
  {                                                                      \
    DOTP_SB2(RTYPE, mult0, mult1, cnst0, cnst1, out0, out1);             \
    DOTP_SB2(RTYPE, mult2, mult3, cnst2, cnst3, out2, out3);             \
  }
Johann's avatar
Johann committed
617 618 619 620 621 622 623 624 625 626 627 628
#define DOTP_SB4_SH(...) DOTP_SB4(v8i16, __VA_ARGS__)

/* Description : Dot product of halfword vector elements
   Arguments   : Inputs  - mult0, mult1, cnst0, cnst1
                 Outputs - out0, out1
                 Return Type - as per RTYPE
   Details     : Signed halfword elements from 'mult0' are multiplied with
                 signed halfword elements from 'cnst0' producing a result
                 twice the size of input i.e. signed word.
                 The multiplication result of adjacent odd-even elements
                 are added together and written to the 'out0' vector
*/
clang-format's avatar
clang-format committed
629 630 631 632 633
#define DOTP_SH2(RTYPE, mult0, mult1, cnst0, cnst1, out0, out1) \
  {                                                             \
    out0 = (RTYPE)__msa_dotp_s_w((v8i16)mult0, (v8i16)cnst0);   \
    out1 = (RTYPE)__msa_dotp_s_w((v8i16)mult1, (v8i16)cnst1);   \
  }
Johann's avatar
Johann committed
634 635
#define DOTP_SH2_SW(...) DOTP_SH2(v4i32, __VA_ARGS__)

clang-format's avatar
clang-format committed
636 637 638 639 640 641
#define DOTP_SH4(RTYPE, mult0, mult1, mult2, mult3, cnst0, cnst1, cnst2, \
                 cnst3, out0, out1, out2, out3)                          \
  {                                                                      \
    DOTP_SH2(RTYPE, mult0, mult1, cnst0, cnst1, out0, out1);             \
    DOTP_SH2(RTYPE, mult2, mult3, cnst2, cnst3, out2, out3);             \
  }
Johann's avatar
Johann committed
642 643 644 645 646 647 648 649 650 651 652 653
#define DOTP_SH4_SW(...) DOTP_SH4(v4i32, __VA_ARGS__)

/* Description : Dot product of word vector elements
   Arguments   : Inputs  - mult0, mult1, cnst0, cnst1
                 Outputs - out0, out1
                 Return Type - as per RTYPE
   Details     : Signed word elements from 'mult0' are multiplied with
                 signed word elements from 'cnst0' producing a result
                 twice the size of input i.e. signed double word.
                 The multiplication result of adjacent odd-even elements
                 are added together and written to the 'out0' vector
*/
clang-format's avatar
clang-format committed
654 655 656 657 658
#define DOTP_SW2(RTYPE, mult0, mult1, cnst0, cnst1, out0, out1) \
  {                                                             \
    out0 = (RTYPE)__msa_dotp_s_d((v4i32)mult0, (v4i32)cnst0);   \
    out1 = (RTYPE)__msa_dotp_s_d((v4i32)mult1, (v4i32)cnst1);   \
  }
Johann's avatar
Johann committed
659 660 661 662 663 664 665 666 667 668 669 670
#define DOTP_SW2_SD(...) DOTP_SW2(v2i64, __VA_ARGS__)

/* Description : Dot product & addition of byte vector elements
   Arguments   : Inputs  - mult0, mult1, cnst0, cnst1
                 Outputs - out0, out1
                 Return Type - as per RTYPE
   Details     : Signed byte elements from 'mult0' are multiplied with
                 signed byte elements from 'cnst0' producing a result
                 twice the size of input i.e. signed halfword.
                 The multiplication result of adjacent odd-even elements
                 are added to the 'out0' vector
*/
clang-format's avatar
clang-format committed
671 672 673 674 675
#define DPADD_SB2(RTYPE, mult0, mult1, cnst0, cnst1, out0, out1)            \
  {                                                                         \
    out0 = (RTYPE)__msa_dpadd_s_h((v8i16)out0, (v16i8)mult0, (v16i8)cnst0); \
    out1 = (RTYPE)__msa_dpadd_s_h((v8i16)out1, (v16i8)mult1, (v16i8)cnst1); \
  }
Johann's avatar
Johann committed
676 677
#define DPADD_SB2_SH(...) DPADD_SB2(v8i16, __VA_ARGS__)

clang-format's avatar
clang-format committed
678 679 680 681 682 683
#define DPADD_SB4(RTYPE, mult0, mult1, mult2, mult3, cnst0, cnst1, cnst2, \
                  cnst3, out0, out1, out2, out3)                          \
  {                                                                       \
    DPADD_SB2(RTYPE, mult0, mult1, cnst0, cnst1, out0, out1);             \
    DPADD_SB2(RTYPE, mult2, mult3, cnst2, cnst3, out2, out3);             \
  }
Johann's avatar
Johann committed
684
#define DPADD_SB4_SH(...) DPADD_SB4(v8i16, __VA_ARGS__)
685

686 687 688 689 690 691 692 693 694 695
/* Description : Dot product & addition of halfword vector elements
   Arguments   : Inputs  - mult0, mult1, cnst0, cnst1
                 Outputs - out0, out1
                 Return Type - as per RTYPE
   Details     : Signed halfword elements from 'mult0' are multiplied with
                 signed halfword elements from 'cnst0' producing a result
                 twice the size of input i.e. signed word.
                 The multiplication result of adjacent odd-even elements
                 are added to the 'out0' vector
*/
clang-format's avatar
clang-format committed
696 697 698 699 700
#define DPADD_SH2(RTYPE, mult0, mult1, cnst0, cnst1, out0, out1)            \
  {                                                                         \
    out0 = (RTYPE)__msa_dpadd_s_w((v4i32)out0, (v8i16)mult0, (v8i16)cnst0); \
    out1 = (RTYPE)__msa_dpadd_s_w((v4i32)out1, (v8i16)mult1, (v8i16)cnst1); \
  }
701 702 703 704 705 706 707
#define DPADD_SH2_SW(...) DPADD_SH2(v4i32, __VA_ARGS__)

/* Description : Dot product & addition of double word vector elements
   Arguments   : Inputs  - mult0, mult1
                 Outputs - out0, out1
                 Return Type - as per RTYPE
   Details     : Each signed word element from 'mult0' is multiplied with itself
Johann's avatar
Johann committed
708
                 producing an intermediate result twice the size of input
709 710 711 712
                 i.e. signed double word
                 The multiplication result of adjacent odd-even elements
                 are added to the 'out0' vector
*/
clang-format's avatar
clang-format committed
713 714 715 716 717
#define DPADD_SD2(RTYPE, mult0, mult1, out0, out1)                          \
  {                                                                         \
    out0 = (RTYPE)__msa_dpadd_s_d((v2i64)out0, (v4i32)mult0, (v4i32)mult0); \
    out1 = (RTYPE)__msa_dpadd_s_d((v2i64)out1, (v4i32)mult1, (v4i32)mult1); \
  }
718 719
#define DPADD_SD2_SD(...) DPADD_SD2(v2i64, __VA_ARGS__)

Johann's avatar
Johann committed
720 721 722 723 724 725 726 727
/* Description : Minimum values between unsigned elements of
                 either vector are copied to the output vector
   Arguments   : Inputs  - in0, in1, min_vec
                 Outputs - in place operation
                 Return Type - as per RTYPE
   Details     : Minimum of unsigned halfword element values from 'in0' and
                 'min_vec' are written to output vector 'in0'
*/
clang-format's avatar
clang-format committed
728 729 730 731 732
#define MIN_UH2(RTYPE, in0, in1, min_vec)            \
  {                                                  \
    in0 = (RTYPE)__msa_min_u_h((v8u16)in0, min_vec); \
    in1 = (RTYPE)__msa_min_u_h((v8u16)in1, min_vec); \
  }
Johann's avatar
Johann committed
733 734
#define MIN_UH2_UH(...) MIN_UH2(v8u16, __VA_ARGS__)

clang-format's avatar
clang-format committed
735 736 737 738 739
#define MIN_UH4(RTYPE, in0, in1, in2, in3, min_vec) \
  {                                                 \
    MIN_UH2(RTYPE, in0, in1, min_vec);              \
    MIN_UH2(RTYPE, in2, in3, min_vec);              \
  }
Johann's avatar
Johann committed
740 741 742 743 744 745 746 747
#define MIN_UH4_UH(...) MIN_UH4(v8u16, __VA_ARGS__)

/* Description : Clips all signed halfword elements of input vector
                 between 0 & 255
   Arguments   : Input  - in
                 Output - out_m
                 Return Type - signed halfword
*/
clang-format's avatar
clang-format committed
748 749 750 751 752 753 754 755 756 757 758 759 760 761 762 763 764 765 766
#define CLIP_SH_0_255(in)                              \
  ({                                                   \
    v8i16 max_m = __msa_ldi_h(255);                    \
    v8i16 out_m;                                       \
                                                       \
    out_m = __msa_maxi_s_h((v8i16)in, 0);              \
    out_m = __msa_min_s_h((v8i16)max_m, (v8i16)out_m); \
    out_m;                                             \
  })
#define CLIP_SH2_0_255(in0, in1) \
  {                              \
    in0 = CLIP_SH_0_255(in0);    \
    in1 = CLIP_SH_0_255(in1);    \
  }
#define CLIP_SH4_0_255(in0, in1, in2, in3) \
  {                                        \
    CLIP_SH2_0_255(in0, in1);              \
    CLIP_SH2_0_255(in2, in3);              \
  }
Johann's avatar
Johann committed
767

768 769 770 771 772 773 774
/* Description : Horizontal addition of 4 signed word elements of input vector
   Arguments   : Input  - in       (signed word vector)
                 Output - sum_m    (i32 sum)
                 Return Type - signed word (GP)
   Details     : 4 signed word elements of 'in' vector are added together and
                 the resulting integer sum is returned
*/
clang-format's avatar
clang-format committed
775 776 777 778 779 780 781 782 783 784 785
#define HADD_SW_S32(in)                            \
  ({                                               \
    v2i64 res0_m, res1_m;                          \
    int32_t sum_m;                                 \
                                                   \
    res0_m = __msa_hadd_s_d((v4i32)in, (v4i32)in); \
    res1_m = __msa_splati_d(res0_m, 1);            \
    res0_m = res0_m + res1_m;                      \
    sum_m = __msa_copy_s_w((v4i32)res0_m, 0);      \
    sum_m;                                         \
  })
786

787 788 789 790 791 792
/* Description : Horizontal addition of 4 unsigned word elements
   Arguments   : Input  - in       (unsigned word vector)
                 Output - sum_m    (u32 sum)
                 Return Type - unsigned word (GP)
   Details     : 4 unsigned word elements of 'in' vector are added together and
                 the resulting integer sum is returned
793
*/
794
#define HADD_UW_U32(in)                               \
clang-format's avatar
clang-format committed
795 796 797 798
  ({                                                  \
    v2u64 res0_m, res1_m;                             \
    uint32_t sum_m;                                   \
                                                      \
799
    res0_m = __msa_hadd_u_d((v4u32)in, (v4u32)in);    \
clang-format's avatar
clang-format committed
800
    res1_m = (v2u64)__msa_splati_d((v2i64)res0_m, 1); \
801
    res0_m += res1_m;                                 \
clang-format's avatar
clang-format committed
802 803 804
    sum_m = __msa_copy_u_w((v4i32)res0_m, 0);         \
    sum_m;                                            \
  })
805

806 807 808 809 810 811 812 813 814 815 816 817 818 819 820 821 822
/* Description : Horizontal addition of 8 unsigned halfword elements
   Arguments   : Input  - in       (unsigned halfword vector)
                 Output - sum_m    (u32 sum)
                 Return Type - unsigned word
   Details     : 8 unsigned halfword elements of 'in' vector are added
                 together and the resulting integer sum is returned
*/
#define HADD_UH_U32(in)                           \
  ({                                              \
    v4u32 res_m;                                  \
    uint32_t sum_m;                               \
                                                  \
    res_m = __msa_hadd_u_w((v8u16)in, (v8u16)in); \
    sum_m = HADD_UW_U32(res_m);                   \
    sum_m;                                        \
  })

Johann's avatar
Johann committed
823
/* Description : Horizontal addition of unsigned byte vector elements
824 825 826
   Arguments   : Inputs  - in0, in1
                 Outputs - out0, out1
                 Return Type - as per RTYPE
Johann's avatar
Johann committed
827 828 829 830
   Details     : Each unsigned odd byte element from 'in0' is added to
                 even unsigned byte element from 'in0' (pairwise) and the
                 halfword result is written to 'out0'
*/
clang-format's avatar
clang-format committed
831 832 833 834 835
#define HADD_UB2(RTYPE, in0, in1, out0, out1)             \
  {                                                       \
    out0 = (RTYPE)__msa_hadd_u_h((v16u8)in0, (v16u8)in0); \
    out1 = (RTYPE)__msa_hadd_u_h((v16u8)in1, (v16u8)in1); \
  }
Johann's avatar
Johann committed
836 837
#define HADD_UB2_UH(...) HADD_UB2(v8u16, __VA_ARGS__)

clang-format's avatar
clang-format committed
838 839 840 841 842
#define HADD_UB4(RTYPE, in0, in1, in2, in3, out0, out1, out2, out3) \
  {                                                                 \
    HADD_UB2(RTYPE, in0, in1, out0, out1);                          \
    HADD_UB2(RTYPE, in2, in3, out2, out3);                          \
  }
Johann's avatar
Johann committed
843 844 845 846 847 848 849
#define HADD_UB4_UH(...) HADD_UB4(v8u16, __VA_ARGS__)

/* Description : Horizontal subtraction of unsigned byte vector elements
   Arguments   : Inputs  - in0, in1
                 Outputs - out0, out1
                 Return Type - as per RTYPE
   Details     : Each unsigned odd byte element from 'in0' is subtracted from
850 851 852
                 even unsigned byte element from 'in0' (pairwise) and the
                 halfword result is written to 'out0'
*/
clang-format's avatar
clang-format committed
853 854 855 856 857
#define HSUB_UB2(RTYPE, in0, in1, out0, out1)             \
  {                                                       \
    out0 = (RTYPE)__msa_hsub_u_h((v16u8)in0, (v16u8)in0); \
    out1 = (RTYPE)__msa_hsub_u_h((v16u8)in1, (v16u8)in1); \
  }
858 859
#define HSUB_UB2_SH(...) HSUB_UB2(v8i16, __VA_ARGS__)

860 861 862 863 864 865 866 867
/* Description : SAD (Sum of Absolute Difference)
   Arguments   : Inputs  - in0, in1, ref0, ref1
                 Outputs - sad_m                 (halfword vector)
                 Return Type - unsigned halfword
   Details     : Absolute difference of all the byte elements from 'in0' with
                 'ref0' is calculated and preserved in 'diff0'. Then even-odd
                 pairs are added together to generate 8 halfword results.
*/
clang-format's avatar
clang-format committed
868 869 870 871 872 873 874 875 876 877 878 879 880
#define SAD_UB2_UH(in0, in1, ref0, ref1)                     \
  ({                                                         \
    v16u8 diff0_m, diff1_m;                                  \
    v8u16 sad_m = { 0 };                                     \
                                                             \
    diff0_m = __msa_asub_u_b((v16u8)in0, (v16u8)ref0);       \
    diff1_m = __msa_asub_u_b((v16u8)in1, (v16u8)ref1);       \
                                                             \
    sad_m += __msa_hadd_u_h((v16u8)diff0_m, (v16u8)diff0_m); \
    sad_m += __msa_hadd_u_h((v16u8)diff1_m, (v16u8)diff1_m); \
                                                             \
    sad_m;                                                   \
  })
881

Johann's avatar
Johann committed
882 883 884 885 886 887 888 889
/* Description : Horizontal subtraction of signed halfword vector elements
   Arguments   : Inputs  - in0, in1
                 Outputs - out0, out1
                 Return Type - as per RTYPE
   Details     : Each signed odd halfword element from 'in0' is subtracted from
                 even signed halfword element from 'in0' (pairwise) and the
                 word result is written to 'out0'
*/
clang-format's avatar
clang-format committed
890 891 892 893 894
#define HSUB_UH2(RTYPE, in0, in1, out0, out1)             \
  {                                                       \
    out0 = (RTYPE)__msa_hsub_s_w((v8i16)in0, (v8i16)in0); \
    out1 = (RTYPE)__msa_hsub_s_w((v8i16)in1, (v8i16)in1); \
  }
Johann's avatar
Johann committed
895 896
#define HSUB_UH2_SW(...) HSUB_UH2(v4i32, __VA_ARGS__)

897 898 899 900 901 902
/* Description : Set element n input vector to GPR value
   Arguments   : Inputs - in0, in1, in2, in3
                 Output - out
                 Return Type - as per RTYPE
   Details     : Set element 0 in vector 'out' to value specified in 'in0'
*/
clang-format's avatar
clang-format committed
903 904 905 906 907
#define INSERT_W2(RTYPE, in0, in1, out)              \
  {                                                  \
    out = (RTYPE)__msa_insert_w((v4i32)out, 0, in0); \
    out = (RTYPE)__msa_insert_w((v4i32)out, 1, in1); \
  }
Johann's avatar
Johann committed
908 909
#define INSERT_W2_SB(...) INSERT_W2(v16i8, __VA_ARGS__)

clang-format's avatar
clang-format committed
910 911 912 913 914 915 916
#define INSERT_W4(RTYPE, in0, in1, in2, in3, out)    \
  {                                                  \
    out = (RTYPE)__msa_insert_w((v4i32)out, 0, in0); \
    out = (RTYPE)__msa_insert_w((v4i32)out, 1, in1); \
    out = (RTYPE)__msa_insert_w((v4i32)out, 2, in2); \
    out = (RTYPE)__msa_insert_w((v4i32)out, 3, in3); \
  }
917 918 919
#define INSERT_W4_UB(...) INSERT_W4(v16u8, __VA_ARGS__)
#define INSERT_W4_SB(...) INSERT_W4(v16i8, __VA_ARGS__)

clang-format's avatar
clang-format committed
920 921 922 923 924
#define INSERT_D2(RTYPE, in0, in1, out)              \
  {                                                  \
    out = (RTYPE)__msa_insert_d((v2i64)out, 0, in0); \
    out = (RTYPE)__msa_insert_d((v2i64)out, 1, in1); \
  }
925 926
#define INSERT_D2_UB(...) INSERT_D2(v16u8, __VA_ARGS__)
#define INSERT_D2_SB(...) INSERT_D2(v16i8, __VA_ARGS__)
927
#define INSERT_D2_SH(...) INSERT_D2(v8i16, __VA_ARGS__)
928

Johann's avatar
Johann committed
929 930 931 932 933 934 935
/* Description : Interleave even byte elements from vectors
   Arguments   : Inputs  - in0, in1, in2, in3
                 Outputs - out0, out1
                 Return Type - as per RTYPE
   Details     : Even byte elements of 'in0' and 'in1' are interleaved
                 and written to 'out0'
*/
clang-format's avatar
clang-format committed
936 937 938 939 940
#define ILVEV_B2(RTYPE, in0, in1, in2, in3, out0, out1)  \
  {                                                      \
    out0 = (RTYPE)__msa_ilvev_b((v16i8)in1, (v16i8)in0); \
    out1 = (RTYPE)__msa_ilvev_b((v16i8)in3, (v16i8)in2); \
  }
Johann's avatar
Johann committed
941 942 943 944 945 946 947 948 949 950
#define ILVEV_B2_UB(...) ILVEV_B2(v16u8, __VA_ARGS__)
#define ILVEV_B2_SH(...) ILVEV_B2(v8i16, __VA_ARGS__)

/* Description : Interleave even halfword elements from vectors
   Arguments   : Inputs  - in0, in1, in2, in3
                 Outputs - out0, out1
                 Return Type - as per RTYPE
   Details     : Even halfword elements of 'in0' and 'in1' are interleaved
                 and written to 'out0'
*/
clang-format's avatar
clang-format committed
951 952 953 954 955
#define ILVEV_H2(RTYPE, in0, in1, in2, in3, out0, out1)  \
  {                                                      \
    out0 = (RTYPE)__msa_ilvev_h((v8i16)in1, (v8i16)in0); \
    out1 = (RTYPE)__msa_ilvev_h((v8i16)in3, (v8i16)in2); \
  }
Johann's avatar
Johann committed
956 957 958 959 960 961 962 963 964 965 966
#define ILVEV_H2_UB(...) ILVEV_H2(v16u8, __VA_ARGS__)
#define ILVEV_H2_SH(...) ILVEV_H2(v8i16, __VA_ARGS__)
#define ILVEV_H2_SW(...) ILVEV_H2(v4i32, __VA_ARGS__)

/* Description : Interleave even word elements from vectors
   Arguments   : Inputs  - in0, in1, in2, in3
                 Outputs - out0, out1
                 Return Type - as per RTYPE
   Details     : Even word elements of 'in0' and 'in1' are interleaved
                 and written to 'out0'
*/
clang-format's avatar
clang-format committed
967 968 969 970 971
#define ILVEV_W2(RTYPE, in0, in1, in2, in3, out0, out1)  \
  {                                                      \
    out0 = (RTYPE)__msa_ilvev_w((v4i32)in1, (v4i32)in0); \
    out1 = (RTYPE)__msa_ilvev_w((v4i32)in3, (v4i32)in2); \
  }
Johann's avatar
Johann committed
972 973 974 975 976 977 978 979 980
#define ILVEV_W2_SB(...) ILVEV_W2(v16i8, __VA_ARGS__)

/* Description : Interleave even double word elements from vectors
   Arguments   : Inputs  - in0, in1, in2, in3
                 Outputs - out0, out1
                 Return Type - as per RTYPE
   Details     : Even double word elements of 'in0' and 'in1' are interleaved
                 and written to 'out0'
*/
clang-format's avatar
clang-format committed
981 982 983 984 985
#define ILVEV_D2(RTYPE, in0, in1, in2, in3, out0, out1)  \
  {                                                      \
    out0 = (RTYPE)__msa_ilvev_d((v2i64)in1, (v2i64)in0); \
    out1 = (RTYPE)__msa_ilvev_d((v2i64)in3, (v2i64)in2); \
  }
Johann's avatar
Johann committed
986 987 988 989 990 991 992 993 994
#define ILVEV_D2_UB(...) ILVEV_D2(v16u8, __VA_ARGS__)

/* Description : Interleave left half of byte elements from vectors
   Arguments   : Inputs  - in0, in1, in2, in3
                 Outputs - out0, out1
                 Return Type - as per RTYPE
   Details     : Left half of byte elements of 'in0' and 'in1' are interleaved
                 and written to 'out0'.
*/
clang-format's avatar
clang-format committed
995 996 997 998 999
#define ILVL_B2(RTYPE, in0, in1, in2, in3, out0, out1)  \
  {                                                     \
    out0 = (RTYPE)__msa_ilvl_b((v16i8)in0, (v16i8)in1); \
    out1 = (RTYPE)__msa_ilvl_b((v16i8)in2, (v16i8)in3); \
  }
Johann's avatar
Johann committed
1000 1001 1002 1003 1004
#define ILVL_B2_UB(...) ILVL_B2(v16u8, __VA_ARGS__)
#define ILVL_B2_SB(...) ILVL_B2(v16i8, __VA_ARGS__)
#define ILVL_B2_UH(...) ILVL_B2(v8u16, __VA_ARGS__)
#define ILVL_B2_SH(...) ILVL_B2(v8i16, __VA_ARGS__)

clang-format's avatar
clang-format committed
1005 1006 1007 1008 1009 1010
#define ILVL_B4(RTYPE, in0, in1, in2, in3, in4, in5, in6, in7, out0, out1, \
                out2, out3)                                                \
  {                                                                        \
    ILVL_B2(RTYPE, in0, in1, in2, in3, out0, out1);                        \
    ILVL_B2(RTYPE, in4, in5, in6, in7, out2, out3);                        \
  }
Johann's avatar
Johann committed
1011
#define ILVL_B4_SB(...) ILVL_B4(v16i8, __VA_ARGS__)
1012
#define ILVL_B4_SH(...) ILVL_B4(v8i16, __VA_ARGS__)
Johann's avatar
Johann committed
1013 1014 1015 1016 1017 1018 1019 1020 1021
#define ILVL_B4_UH(...) ILVL_B4(v8u16, __VA_ARGS__)

/* Description : Interleave left half of halfword elements from vectors
   Arguments   : Inputs  - in0, in1, in2, in3
                 Outputs - out0, out1
                 Return Type - as per RTYPE
   Details     : Left half of halfword elements of 'in0' and 'in1' are
                 interleaved and written to 'out0'.
*/
clang-format's avatar
clang-format committed
1022 1023 1024 1025 1026
#define ILVL_H2(RTYPE, in0, in1, in2, in3, out0, out1)  \
  {                                                     \
    out0 = (RTYPE)__msa_ilvl_h((v8i16)in0, (v8i16)in1); \
    out1 = (RTYPE)__msa_ilvl_h((v8i16)in2, (v8i16)in3); \
  }
Johann's avatar
Johann committed
1027
#define ILVL_H2_SH(...) ILVL_H2(v8i16, __VA_ARGS__)
1028
#define ILVL_H2_SW(...) ILVL_H2(v4i32, __VA_ARGS__)
Johann's avatar
Johann committed
1029 1030 1031 1032 1033 1034 1035 1036

/* Description : Interleave left half of word elements from vectors
   Arguments   : Inputs  - in0, in1, in2, in3
                 Outputs - out0, out1
                 Return Type - as per RTYPE
   Details     : Left half of word elements of 'in0' and 'in1' are interleaved
                 and written to 'out0'.
*/
clang-format's avatar
clang-format committed
1037 1038 1039 1040 1041
#define ILVL_W2(RTYPE, in0, in1, in2, in3, out0, out1)  \
  {                                                     \
    out0 = (RTYPE)__msa_ilvl_w((v4i32)in0, (v4i32)in1); \
    out1 = (RTYPE)__msa_ilvl_w((v4i32)in2, (v4i32)in3); \
  }
Johann's avatar
Johann committed
1042 1043 1044 1045 1046 1047 1048 1049 1050 1051
#define ILVL_W2_UB(...) ILVL_W2(v16u8, __VA_ARGS__)
#define ILVL_W2_SH(...) ILVL_W2(v8i16, __VA_ARGS__)

/* Description : Interleave right half of byte elements from vectors
   Arguments   : Inputs  - in0, in1, in2, in3
                 Outputs - out0, out1
                 Return Type - as per RTYPE
   Details     : Right half of byte elements of 'in0' and 'in1' are interleaved
                 and written to out0.
*/
clang-format's avatar
clang-format committed
1052 1053 1054 1055 1056
#define ILVR_B2(RTYPE, in0, in1, in2, in3, out0, out1)  \
  {                                                     \
    out0 = (RTYPE)__msa_ilvr_b((v16i8)in0, (v16i8)in1); \
    out1 = (RTYPE)__msa_ilvr_b((v16i8)in2, (v16i8)in3); \
  }
Johann's avatar
Johann committed
1057 1058 1059 1060 1061
#define ILVR_B2_UB(...) ILVR_B2(v16u8, __VA_ARGS__)
#define ILVR_B2_SB(...) ILVR_B2(v16i8, __VA_ARGS__)
#define ILVR_B2_UH(...) ILVR_B2(v8u16, __VA_ARGS__)
#define ILVR_B2_SH(...) ILVR_B2(v8i16, __VA_ARGS__)

clang-format's avatar
clang-format committed
1062 1063 1064 1065 1066 1067
#define ILVR_B4(RTYPE, in0, in1, in2, in3, in4, in5, in6, in7, out0, out1, \
                out2, out3)                                                \
  {                                                                        \
    ILVR_B2(RTYPE, in0, in1, in2, in3, out0, out1);                        \
    ILVR_B2(RTYPE, in4, in5, in6, in7, out2, out3);                        \
  }
Johann's avatar
Johann committed
1068 1069 1070 1071 1072
#define ILVR_B4_UB(...) ILVR_B4(v16u8, __VA_ARGS__)
#define ILVR_B4_SB(...) ILVR_B4(v16i8, __VA_ARGS__)
#define ILVR_B4_UH(...) ILVR_B4(v8u16, __VA_ARGS__)
#define ILVR_B4_SH(...) ILVR_B4(v8i16, __VA_ARGS__)

clang-format's avatar
clang-format committed
1073 1074 1075 1076 1077 1078 1079 1080 1081
#define ILVR_B8(RTYPE, in0, in1, in2, in3, in4, in5, in6, in7, in8, in9, in10, \
                in11, in12, in13, in14, in15, out0, out1, out2, out3, out4,    \
                out5, out6, out7)                                              \
  {                                                                            \
    ILVR_B4(RTYPE, in0, in1, in2, in3, in4, in5, in6, in7, out0, out1, out2,   \
            out3);                                                             \
    ILVR_B4(RTYPE, in8, in9, in10, in11, in12, in13, in14, in15, out4, out5,   \
            out6, out7);                                                       \
  }
Johann's avatar
Johann committed
1082 1083 1084 1085 1086 1087 1088 1089 1090
#define ILVR_B8_UH(...) ILVR_B8(v8u16, __VA_ARGS__)

/* Description : Interleave right half of halfword elements from vectors
   Arguments   : Inputs  - in0, in1, in2, in3
                 Outputs - out0, out1
                 Return Type - as per RTYPE
   Details     : Right half of halfword elements of 'in0' and 'in1' are
                 interleaved and written to 'out0'.
*/
clang-format's avatar
clang-format committed
1091 1092 1093 1094 1095
#define ILVR_H2(RTYPE, in0, in1, in2, in3, out0, out1)  \
  {                                                     \
    out0 = (RTYPE)__msa_ilvr_h((v8i16)in0, (v8i16)in1); \
    out1 = (RTYPE)__msa_ilvr_h((v8i16)in2, (v8i16)in3); \
  }
Johann's avatar
Johann committed
1096
#define ILVR_H2_SH(...) ILVR_H2(v8i16, __VA_ARGS__)
1097
#define ILVR_H2_SW(...) ILVR_H2(v4i32, __VA_ARGS__)
Johann's avatar
Johann committed
1098

clang-format's avatar
clang-format committed
1099 1100 1101 1102 1103 1104
#define ILVR_H4(RTYPE, in0, in1, in2, in3, in4, in5, in6, in7, out0, out1, \
                out2, out3)                                                \
  {                                                                        \
    ILVR_H2(RTYPE, in0, in1, in2, in3, out0, out1);                        \
    ILVR_H2(RTYPE, in4, in5, in6, in7, out2, out3);                        \
  }
Johann's avatar
Johann committed
1105 1106
#define ILVR_H4_SH(...) ILVR_H4(v8i16, __VA_ARGS__)

clang-format's avatar
clang-format committed
1107 1108 1109 1110 1111
#define ILVR_W2(RTYPE, in0, in1, in2, in3, out0, out1)  \
  {                                                     \
    out0 = (RTYPE)__msa_ilvr_w((v4i32)in0, (v4i32)in1); \
    out1 = (RTYPE)__msa_ilvr_w((v4i32)in2, (v4i32)in3); \
  }
Johann's avatar
Johann committed
1112 1113 1114
#define ILVR_W2_UB(...) ILVR_W2(v16u8, __VA_ARGS__)
#define ILVR_W2_SH(...) ILVR_W2(v8i16, __VA_ARGS__)

clang-format's avatar
clang-format committed
1115 1116 1117 1118 1119 1120
#define ILVR_W4(RTYPE, in0, in1, in2, in3, in4, in5, in6, in7, out0, out1, \
                out2, out3)                                                \
  {                                                                        \
    ILVR_W2(RTYPE, in0, in1, in2, in3, out0, out1);                        \
    ILVR_W2(RTYPE, in4, in5, in6, in7, out2, out3);                        \
  }
Johann's avatar
Johann committed
1121 1122 1123 1124 1125 1126 1127 1128 1129
#define ILVR_W4_UB(...) ILVR_W4(v16u8, __VA_ARGS__)

/* Description : Interleave right half of double word elements from vectors
   Arguments   : Inputs  - in0, in1, in2, in3
                 Outputs - out0, out1
                 Return Type - as per RTYPE
   Details     : Right half of double word elements of 'in0' and 'in1' are
                 interleaved and written to 'out0'.
*/
clang-format's avatar
clang-format committed
1130 1131 1132 1133 1134
#define ILVR_D2(RTYPE, in0, in1, in2, in3, out0, out1)      \
  {                                                         \
    out0 = (RTYPE)__msa_ilvr_d((v2i64)(in0), (v2i64)(in1)); \
    out1 = (RTYPE)__msa_ilvr_d((v2i64)(in2), (v2i64)(in3)); \
  }
Johann's avatar
Johann committed
1135 1136 1137 1138
#define ILVR_D2_UB(...) ILVR_D2(v16u8, __VA_ARGS__)
#define ILVR_D2_SB(...) ILVR_D2(v16i8, __VA_ARGS__)
#define ILVR_D2_SH(...) ILVR_D2(v8i16, __VA_ARGS__)

clang-format's avatar
clang-format committed
1139 1140 1141 1142 1143
#define ILVR_D3(RTYPE, in0, in1, in2, in3, in4, in5, out0, out1, out2) \
  {                                                                    \
    ILVR_D2(RTYPE, in0, in1, in2, in3, out0, out1);                    \
    out2 = (RTYPE)__msa_ilvr_d((v2i64)(in4), (v2i64)(in5));            \
  }
Johann's avatar
Johann committed
1144 1145
#define ILVR_D3_SB(...) ILVR_D3(v16i8, __VA_ARGS__)

clang-format's avatar
clang-format committed
1146 1147 1148 1149 1150 1151
#define ILVR_D4(RTYPE, in0, in1, in2, in3, in4, in5, in6, in7, out0, out1, \
                out2, out3)                                                \
  {                                                                        \
    ILVR_D2(RTYPE, in0, in1, in2, in3, out0, out1);                        \
    ILVR_D2(RTYPE, in4, in5, in6, in7, out2, out3);                        \
  }
Johann's avatar
Johann committed
1152 1153 1154
#define ILVR_D4_SB(...) ILVR_D4(v16i8, __VA_ARGS__)
#define ILVR_D4_UB(...) ILVR_D4(v16u8, __VA_ARGS__)

1155 1156 1157 1158 1159 1160 1161
/* Description : Interleave both left and right half of input vectors
   Arguments   : Inputs  - in0, in1
                 Outputs - out0, out1
                 Return Type - as per RTYPE
   Details     : Right half of byte elements from 'in0' and 'in1' are
                 interleaved and written to 'out0'
*/
clang-format's avatar
clang-format committed
1162 1163 1164 1165 1166
#define ILVRL_B2(RTYPE, in0, in1, out0, out1)           \
  {                                                     \
    out0 = (RTYPE)__msa_ilvr_b((v16i8)in0, (v16i8)in1); \
    out1 = (RTYPE)__msa_ilvl_b((v16i8)in0, (v16i8)in1); \
  }
1167
#define ILVRL_B2_UB(...) ILVRL_B2(v16u8, __VA_ARGS__)
Johann's avatar
Johann committed
1168 1169 1170
#define ILVRL_B2_SB(...) ILVRL_B2(v16i8, __VA_ARGS__)
#define ILVRL_B2_UH(...) ILVRL_B2(v8u16, __VA_ARGS__)
#define ILVRL_B2_SH(...) ILVRL_B2(v8i16, __VA_ARGS__)
1171

clang-format's avatar
clang-format committed
1172 1173 1174 1175 1176
#define ILVRL_H2(RTYPE, in0, in1, out0, out1)           \
  {                                                     \
    out0 = (RTYPE)__msa_ilvr_h((v8i16)in0, (v8i16)in1); \
    out1 = (RTYPE)__msa_ilvl_h((v8i16)in0, (v8i16)in1); \
  }
Johann's avatar
Johann committed
1177
#define ILVRL_H2_SH(...) ILVRL_H2(v8i16, __VA_ARGS__)
1178 1179
#define ILVRL_H2_SW(...) ILVRL_H2(v4i32, __VA_ARGS__)

clang-format's avatar
clang-format committed
1180 1181 1182 1183 1184
#define ILVRL_W2(RTYPE, in0, in1, out0, out1)           \
  {                                                     \
    out0 = (RTYPE)__msa_ilvr_w((v4i32)in0, (v4i32)in1); \
    out1 = (RTYPE)__msa_ilvl_w((v4i32)in0, (v4i32)in1); \
  }
1185
#define ILVRL_W2_UB(...) ILVRL_W2(v16u8, __VA_ARGS__)
1186
#define ILVRL_W2_SB(...) ILVRL_W2(v16i8, __VA_ARGS__)
Johann's avatar
Johann committed
1187 1188 1189 1190 1191 1192 1193 1194 1195 1196 1197 1198 1199
#define ILVRL_W2_SH(...) ILVRL_W2(v8i16, __VA_ARGS__)
#define ILVRL_W2_SW(...) ILVRL_W2(v4i32, __VA_ARGS__)

/* Description : Saturate the halfword element values to the max
                 unsigned value of (sat_val + 1) bits
                 The element data width remains unchanged
   Arguments   : Inputs  - in0, in1, sat_val
                 Outputs - in place operation
                 Return Type - as per RTYPE
   Details     : Each unsigned halfword element from 'in0' is saturated to the
                 value generated with (sat_val + 1) bit range.
                 The results are written in place
*/
clang-format's avatar
clang-format committed
1200 1201 1202 1203 1204
#define SAT_UH2(RTYPE, in0, in1, sat_val)            \
  {                                                  \
    in0 = (RTYPE)__msa_sat_u_h((v8u16)in0, sat_val); \
    in1 = (RTYPE)__msa_sat_u_h((v8u16)in1, sat_val); \
  }
Johann's avatar
Johann committed
1205 1206
#define SAT_UH2_UH(...) SAT_UH2(v8u16, __VA_ARGS__)

clang-format's avatar
clang-format committed
1207 1208 1209 1210 1211
#define SAT_UH4(RTYPE, in0, in1, in2, in3, sat_val) \
  {                                                 \
    SAT_UH2(RTYPE, in0, in1, sat_val);              \
    SAT_UH2(RTYPE, in2, in3, sat_val)               \
  }
Johann's avatar
Johann committed
1212 1213 1214 1215 1216 1217 1218 1219 1220 1221 1222 1223
#define SAT_UH4_UH(...) SAT_UH4(v8u16, __VA_ARGS__)

/* Description : Saturate the halfword element values to the max
                 unsigned value of (sat_val + 1) bits
                 The element data width remains unchanged
   Arguments   : Inputs  - in0, in1, sat_val
                 Outputs - in place operation
                 Return Type - as per RTYPE
   Details     : Each unsigned halfword element from 'in0' is saturated to the
                 value generated with (sat_val + 1) bit range
                 The results are written in place
*/
clang-format's avatar
clang-format committed
1224 1225 1226 1227 1228
#define SAT_SH2(RTYPE, in0, in1, sat_val)            \
  {                                                  \
    in0 = (RTYPE)__msa_sat_s_h((v8i16)in0, sat_val); \
    in1 = (RTYPE)__msa_sat_s_h((v8i16)in1, sat_val); \
  }
Johann's avatar
Johann committed
1229 1230
#define SAT_SH2_SH(...) SAT_SH2(v8i16, __VA_ARGS__)

clang-format's avatar
clang-format committed
1231 1232 1233 1234 1235
#define SAT_SH4(RTYPE, in0, in1, in2, in3, sat_val) \
  {                                                 \
    SAT_SH2(RTYPE, in0, in1, sat_val);              \
    SAT_SH2(RTYPE, in2, in3, sat_val);              \
  }
Johann's avatar
Johann committed
1236 1237 1238 1239 1240 1241 1242 1243 1244 1245 1246
#define SAT_SH4_SH(...) SAT_SH4(v8i16, __VA_ARGS__)

/* Description : Indexed halfword element values are replicated to all
                 elements in output vector
   Arguments   : Inputs  - in, idx0, idx1
                 Outputs - out0, out1
                 Return Type - as per RTYPE
   Details     : 'idx0' element value from 'in' vector is replicated to all
                  elements in 'out0' vector
                  Valid index range for halfword operation is 0-7
*/
clang-format's avatar
clang-format committed
1247 1248 1249 1250 1251
#define SPLATI_H2(RTYPE, in, idx0, idx1, out0, out1) \
  {                                                  \
    out0 = (RTYPE)__msa_splati_h((v8i16)in, idx0);   \
    out1 = (RTYPE)__msa_splati_h((v8i16)in, idx1);   \
  }
Johann's avatar
Johann committed
1252 1253
#define SPLATI_H2_SH(...) SPLATI_H2(v8i16, __VA_ARGS__)

clang-format's avatar
clang-format committed
1254 1255 1256 1257 1258
#define SPLATI_H4(RTYPE, in, idx0, idx1, idx2, idx3, out0, out1, out2, out3) \
  {                                                                          \
    SPLATI_H2(RTYPE, in, idx0, idx1, out0, out1);                            \
    SPLATI_H2(RTYPE, in, idx2, idx3, out2, out3);                            \
  }
Johann's avatar
Johann committed
1259 1260 1261 1262 1263 1264 1265 1266 1267 1268 1269
#define SPLATI_H4_SB(...) SPLATI_H4(v16i8, __VA_ARGS__)
#define SPLATI_H4_SH(...) SPLATI_H4(v8i16, __VA_ARGS__)

/* Description : Pack even byte elements of vector pairs
   Arguments   : Inputs  - in0, in1, in2, in3
                 Outputs - out0, out1
                 Return Type - as per RTYPE
   Details     : Even byte elements of 'in0' are copied to the left half of
                 'out0' & even byte elements of 'in1' are copied to the right
                 half of 'out0'.
*/
clang-format's avatar
clang-format committed
1270 1271 1272 1273 1274
#define PCKEV_B2(RTYPE, in0, in1, in2, in3, out0, out1)  \
  {                                                      \
    out0 = (RTYPE)__msa_pckev_b((v16i8)in0, (v16i8)in1); \
    out1 = (RTYPE)__msa_pckev_b((v16i8)in2, (v16i8)in3); \
  }
Johann's avatar
Johann committed
1275 1276 1277 1278
#define PCKEV_B2_SB(...) PCKEV_B2(v16i8, __VA_ARGS__)
#define PCKEV_B2_UB(...) PCKEV_B2(v16u8, __VA_ARGS__)
#define PCKEV_B2_SH(...) PCKEV_B2(v8i16, __VA_ARGS__)

clang-format's avatar
clang-format committed
1279 1280 1281 1282 1283 1284
#define PCKEV_B4(RTYPE, in0, in1, in2, in3, in4, in5, in6, in7, out0, out1, \
                 out2, out3)                                                \
  {                                                                         \
    PCKEV_B2(RTYPE, in0, in1, in2, in3, out0, out1);                        \
    PCKEV_B2(RTYPE, in4, in5, in6, in7, out2, out3);                        \
  }
Johann's avatar
Johann committed
1285 1286 1287 1288 1289 1290 1291 1292 1293 1294 1295 1296
#define PCKEV_B4_SB(...) PCKEV_B4(v16i8, __VA_ARGS__)
#define PCKEV_B4_UB(...) PCKEV_B4(v16u8, __VA_ARGS__)
#define PCKEV_B4_SH(...) PCKEV_B4(v8i16, __VA_ARGS__)

/* Description : Pack even halfword elements of vector pairs
   Arguments   : Inputs  - in0, in1, in2, in3
                 Outputs - out0, out1
                 Return Type - as per RTYPE
   Details     : Even halfword elements of 'in0' are copied to the left half of
                 'out0' & even halfword elements of 'in1' are copied to the
                 right half of 'out0'.
*/
clang-format's avatar
clang-format committed
1297 1298 1299 1300 1301
#define PCKEV_H2(RTYPE, in0, in1, in2, in3, out0, out1)  \
  {                                                      \
    out0 = (RTYPE)__msa_pckev_h((v8i16)in0, (v8i16)in1); \
    out1 = (RTYPE)__msa_pckev_h((v8i16)in2, (v8i16)in3); \
  }
Johann's avatar
Johann committed
1302 1303 1304
#define PCKEV_H2_SH(...) PCKEV_H2(v8i16, __VA_ARGS__)
#define PCKEV_H2_SW(...) PCKEV_H2(v4i32, __VA_ARGS__)

clang-format's avatar
clang-format committed
1305 1306 1307 1308 1309 1310
#define PCKEV_H4(RTYPE, in0, in1, in2, in3, in4, in5, in6, in7, out0, out1, \
                 out2, out3)                                                \
  {                                                                         \
    PCKEV_H2(RTYPE, in0, in1, in2, in3, out0, out1);                        \
    PCKEV_H2(RTYPE, in4, in5, in6, in7, out2, out3);                        \
  }
Johann's avatar
Johann committed
1311 1312
#define PCKEV_H4_SH(...) PCKEV_H4(v8i16, __VA_ARGS__)

1313 1314 1315 1316 1317