src/third_party/libyuv/source/rotate_mips.cc

   1 /*
   2  *  Copyright 2011 The LibYuv Project Authors. All rights reserved.
   3  *
   4  *  Use of this source code is governed by a BSD-style license
   5  *  that can be found in the LICENSE file in the root of the source
   6  *  tree. An additional intellectual property rights grant can be found
   7  *  in the file PATENTS. All contributing project authors may
   8  *  be found in the AUTHORS file in the root of the source tree.
   9  */
  10
  11 #include "libyuv/row.h"
  12
  13 #include "libyuv/basic_types.h"
  14
  15 #ifdef __cplusplus
  16 namespace libyuv {
  17 extern "C" {
  18 #endif
  19
  20 #if !defined(LIBYUV_DISABLE_MIPS) && \
  21     defined(__mips_dsp) && (__mips_dsp_rev >= 2)
  22
  23 void TransposeWx8_MIPS_DSPR2(const uint8* src, int src_stride,
  24                              uint8* dst, int dst_stride,
  25                              int width) {
  26    __asm__ __volatile__ (
  27       ".set push                                         \n"
  28       ".set noreorder                                    \n"
  29       "sll              $t2, %[src_stride], 0x1          \n" // src_stride x 2
  30       "sll              $t4, %[src_stride], 0x2          \n" // src_stride x 4
  31       "sll              $t9, %[src_stride], 0x3          \n" // src_stride x 8
  32       "addu             $t3, $t2, %[src_stride]          \n"
  33       "addu             $t5, $t4, %[src_stride]          \n"
  34       "addu             $t6, $t2, $t4                    \n"
  35       "andi             $t0, %[dst], 0x3                 \n"
  36       "andi             $t1, %[dst_stride], 0x3          \n"
  37       "or               $t0, $t0, $t1                    \n"
  38       "bnez             $t0, 11f                         \n"
  39       " subu            $t7, $t9, %[src_stride]          \n"
  40 //dst + dst_stride word aligned
  41     "1:                                                  \n"
  42       "lbu              $t0, 0(%[src])                   \n"
  43       "lbux             $t1, %[src_stride](%[src])       \n"
  44       "lbux             $t8, $t2(%[src])                 \n"
  45       "lbux             $t9, $t3(%[src])                 \n"
  46       "sll              $t1, $t1, 16                     \n"
  47       "sll              $t9, $t9, 16                     \n"
  48       "or               $t0, $t0, $t1                    \n"
  49       "or               $t8, $t8, $t9                    \n"
  50       "precr.qb.ph      $s0, $t8, $t0                    \n"
  51       "lbux             $t0, $t4(%[src])                 \n"
  52       "lbux             $t1, $t5(%[src])                 \n"
  53       "lbux             $t8, $t6(%[src])                 \n"
  54       "lbux             $t9, $t7(%[src])                 \n"
  55       "sll              $t1, $t1, 16                     \n"
  56       "sll              $t9, $t9, 16                     \n"
  57       "or               $t0, $t0, $t1                    \n"
  58       "or               $t8, $t8, $t9                    \n"
  59       "precr.qb.ph      $s1, $t8, $t0                    \n"
  60       "sw               $s0, 0(%[dst])                   \n"
  61       "addiu            %[width], -1                     \n"
  62       "addiu            %[src], 1                        \n"
  63       "sw               $s1, 4(%[dst])                   \n"
  64       "bnez             %[width], 1b                     \n"
  65       " addu            %[dst], %[dst], %[dst_stride]    \n"
  66       "b                2f                               \n"
  67 //dst + dst_stride unaligned
  68    "11:                                                  \n"
  69       "lbu              $t0, 0(%[src])                   \n"
  70       "lbux             $t1, %[src_stride](%[src])       \n"
  71       "lbux             $t8, $t2(%[src])                 \n"
  72       "lbux             $t9, $t3(%[src])                 \n"
  73       "sll              $t1, $t1, 16                     \n"
  74       "sll              $t9, $t9, 16                     \n"
  75       "or               $t0, $t0, $t1                    \n"
  76       "or               $t8, $t8, $t9                    \n"
  77       "precr.qb.ph      $s0, $t8, $t0                    \n"
  78       "lbux             $t0, $t4(%[src])                 \n"
  79       "lbux             $t1, $t5(%[src])                 \n"
  80       "lbux             $t8, $t6(%[src])                 \n"
  81       "lbux             $t9, $t7(%[src])                 \n"
  82       "sll              $t1, $t1, 16                     \n"
  83       "sll              $t9, $t9, 16                     \n"
  84       "or               $t0, $t0, $t1                    \n"
  85       "or               $t8, $t8, $t9                    \n"
  86       "precr.qb.ph      $s1, $t8, $t0                    \n"
  87       "swr              $s0, 0(%[dst])                   \n"
  88       "swl              $s0, 3(%[dst])                   \n"
  89       "addiu            %[width], -1                     \n"
  90       "addiu            %[src], 1                        \n"
  91       "swr              $s1, 4(%[dst])                   \n"
  92       "swl              $s1, 7(%[dst])                   \n"
  93       "bnez             %[width], 11b                    \n"
  94        "addu             %[dst], %[dst], %[dst_stride]   \n"
  95     "2:                                                  \n"
  96       ".set pop                                          \n"
  97       :[src] "+r" (src),
  98        [dst] "+r" (dst),
  99        [width] "+r" (width)
 100       :[src_stride] "r" (src_stride),
 101        [dst_stride] "r" (dst_stride)
 102       : "t0", "t1",  "t2", "t3", "t4", "t5",
 103         "t6", "t7", "t8", "t9",
 104         "s0", "s1"
 105   );
 106 }
 107
 108 void TransposeWx8_FAST_MIPS_DSPR2(const uint8* src, int src_stride,
 109                                   uint8* dst, int dst_stride,
 110                                   int width) {
 111   __asm__ __volatile__ (
 112       ".set noat                                         \n"
 113       ".set push                                         \n"
 114       ".set noreorder                                    \n"
 115       "beqz             %[width], 2f                     \n"
 116       " sll             $t2, %[src_stride], 0x1          \n"  // src_stride x 2
 117       "sll              $t4, %[src_stride], 0x2          \n"  // src_stride x 4
 118       "sll              $t9, %[src_stride], 0x3          \n"  // src_stride x 8
 119       "addu             $t3, $t2, %[src_stride]          \n"
 120       "addu             $t5, $t4, %[src_stride]          \n"
 121       "addu             $t6, $t2, $t4                    \n"
 122
 123       "srl              $AT, %[width], 0x2               \n"
 124       "andi             $t0, %[dst], 0x3                 \n"
 125       "andi             $t1, %[dst_stride], 0x3          \n"
 126       "or               $t0, $t0, $t1                    \n"
 127       "bnez             $t0, 11f                         \n"
 128       " subu            $t7, $t9, %[src_stride]          \n"
 129 //dst + dst_stride word aligned
 130       "1:                                                \n"
 131       "lw               $t0, 0(%[src])                   \n"
 132       "lwx              $t1, %[src_stride](%[src])       \n"
 133       "lwx              $t8, $t2(%[src])                 \n"
 134       "lwx              $t9, $t3(%[src])                 \n"
 135
 136 // t0 = | 30 | 20 | 10 | 00 |
 137 // t1 = | 31 | 21 | 11 | 01 |
 138 // t8 = | 32 | 22 | 12 | 02 |
 139 // t9 = | 33 | 23 | 13 | 03 |
 140
 141       "precr.qb.ph     $s0, $t1, $t0                     \n"
 142       "precr.qb.ph     $s1, $t9, $t8                     \n"
 143       "precrq.qb.ph    $s2, $t1, $t0                     \n"
 144       "precrq.qb.ph    $s3, $t9, $t8                     \n"
 145
 146   // s0 = | 21 | 01 | 20 | 00 |
 147   // s1 = | 23 | 03 | 22 | 02 |
 148   // s2 = | 31 | 11 | 30 | 10 |
 149   // s3 = | 33 | 13 | 32 | 12 |
 150
 151       "precr.qb.ph     $s4, $s1, $s0                     \n"
 152       "precrq.qb.ph    $s5, $s1, $s0                     \n"
 153       "precr.qb.ph     $s6, $s3, $s2                     \n"
 154       "precrq.qb.ph    $s7, $s3, $s2                     \n"
 155
 156   // s4 = | 03 | 02 | 01 | 00 |
 157   // s5 = | 23 | 22 | 21 | 20 |
 158   // s6 = | 13 | 12 | 11 | 10 |
 159   // s7 = | 33 | 32 | 31 | 30 |
 160
 161       "lwx              $t0, $t4(%[src])                 \n"
 162       "lwx              $t1, $t5(%[src])                 \n"
 163       "lwx              $t8, $t6(%[src])                 \n"
 164       "lwx              $t9, $t7(%[src])                 \n"
 165
 166 // t0 = | 34 | 24 | 14 | 04 |
 167 // t1 = | 35 | 25 | 15 | 05 |
 168 // t8 = | 36 | 26 | 16 | 06 |
 169 // t9 = | 37 | 27 | 17 | 07 |
 170
 171       "precr.qb.ph     $s0, $t1, $t0                     \n"
 172       "precr.qb.ph     $s1, $t9, $t8                     \n"
 173       "precrq.qb.ph    $s2, $t1, $t0                     \n"
 174       "precrq.qb.ph    $s3, $t9, $t8                     \n"
 175
 176   // s0 = | 25 | 05 | 24 | 04 |
 177   // s1 = | 27 | 07 | 26 | 06 |
 178   // s2 = | 35 | 15 | 34 | 14 |
 179   // s3 = | 37 | 17 | 36 | 16 |
 180
 181       "precr.qb.ph     $t0, $s1, $s0                     \n"
 182       "precrq.qb.ph    $t1, $s1, $s0                     \n"
 183       "precr.qb.ph     $t8, $s3, $s2                     \n"
 184       "precrq.qb.ph    $t9, $s3, $s2                     \n"
 185
 186   // t0 = | 07 | 06 | 05 | 04 |
 187   // t1 = | 27 | 26 | 25 | 24 |
 188   // t8 = | 17 | 16 | 15 | 14 |
 189   // t9 = | 37 | 36 | 35 | 34 |
 190
 191       "addu            $s0, %[dst], %[dst_stride]        \n"
 192       "addu            $s1, $s0, %[dst_stride]           \n"
 193       "addu            $s2, $s1, %[dst_stride]           \n"
 194
 195       "sw              $s4, 0(%[dst])                    \n"
 196       "sw              $t0, 4(%[dst])                    \n"
 197       "sw              $s6, 0($s0)                       \n"
 198       "sw              $t8, 4($s0)                       \n"
 199       "sw              $s5, 0($s1)                       \n"
 200       "sw              $t1, 4($s1)                       \n"
 201       "sw              $s7, 0($s2)                       \n"
 202       "sw              $t9, 4($s2)                       \n"
 203
 204       "addiu            $AT, -1                          \n"
 205       "addiu            %[src], 4                        \n"
 206
 207       "bnez             $AT, 1b                          \n"
 208       " addu            %[dst], $s2, %[dst_stride]       \n"
 209       "b                2f                               \n"
 210 //dst + dst_stride unaligned
 211       "11:                                               \n"
 212       "lw               $t0, 0(%[src])                   \n"
 213       "lwx              $t1, %[src_stride](%[src])       \n"
 214       "lwx              $t8, $t2(%[src])                 \n"
 215       "lwx              $t9, $t3(%[src])                 \n"
 216
 217 // t0 = | 30 | 20 | 10 | 00 |
 218 // t1 = | 31 | 21 | 11 | 01 |
 219 // t8 = | 32 | 22 | 12 | 02 |
 220 // t9 = | 33 | 23 | 13 | 03 |
 221
 222       "precr.qb.ph     $s0, $t1, $t0                     \n"
 223       "precr.qb.ph     $s1, $t9, $t8                     \n"
 224       "precrq.qb.ph    $s2, $t1, $t0                     \n"
 225       "precrq.qb.ph    $s3, $t9, $t8                     \n"
 226
 227   // s0 = | 21 | 01 | 20 | 00 |
 228   // s1 = | 23 | 03 | 22 | 02 |
 229   // s2 = | 31 | 11 | 30 | 10 |
 230   // s3 = | 33 | 13 | 32 | 12 |
 231
 232       "precr.qb.ph     $s4, $s1, $s0                     \n"
 233       "precrq.qb.ph    $s5, $s1, $s0                     \n"
 234       "precr.qb.ph     $s6, $s3, $s2                     \n"
 235       "precrq.qb.ph    $s7, $s3, $s2                     \n"
 236
 237   // s4 = | 03 | 02 | 01 | 00 |
 238   // s5 = | 23 | 22 | 21 | 20 |
 239   // s6 = | 13 | 12 | 11 | 10 |
 240   // s7 = | 33 | 32 | 31 | 30 |
 241
 242       "lwx              $t0, $t4(%[src])                 \n"
 243       "lwx              $t1, $t5(%[src])                 \n"
 244       "lwx              $t8, $t6(%[src])                 \n"
 245       "lwx              $t9, $t7(%[src])                 \n"
 246
 247 // t0 = | 34 | 24 | 14 | 04 |
 248 // t1 = | 35 | 25 | 15 | 05 |
 249 // t8 = | 36 | 26 | 16 | 06 |
 250 // t9 = | 37 | 27 | 17 | 07 |
 251
 252       "precr.qb.ph     $s0, $t1, $t0                     \n"
 253       "precr.qb.ph     $s1, $t9, $t8                     \n"
 254       "precrq.qb.ph    $s2, $t1, $t0                     \n"
 255       "precrq.qb.ph    $s3, $t9, $t8                     \n"
 256
 257   // s0 = | 25 | 05 | 24 | 04 |
 258   // s1 = | 27 | 07 | 26 | 06 |
 259   // s2 = | 35 | 15 | 34 | 14 |
 260   // s3 = | 37 | 17 | 36 | 16 |
 261
 262       "precr.qb.ph     $t0, $s1, $s0                     \n"
 263       "precrq.qb.ph    $t1, $s1, $s0                     \n"
 264       "precr.qb.ph     $t8, $s3, $s2                     \n"
 265       "precrq.qb.ph    $t9, $s3, $s2                     \n"
 266
 267   // t0 = | 07 | 06 | 05 | 04 |
 268   // t1 = | 27 | 26 | 25 | 24 |
 269   // t8 = | 17 | 16 | 15 | 14 |
 270   // t9 = | 37 | 36 | 35 | 34 |
 271
 272       "addu            $s0, %[dst], %[dst_stride]        \n"
 273       "addu            $s1, $s0, %[dst_stride]           \n"
 274       "addu            $s2, $s1, %[dst_stride]           \n"
 275
 276       "swr              $s4, 0(%[dst])                   \n"
 277       "swl              $s4, 3(%[dst])                   \n"
 278       "swr              $t0, 4(%[dst])                   \n"
 279       "swl              $t0, 7(%[dst])                   \n"
 280       "swr              $s6, 0($s0)                      \n"
 281       "swl              $s6, 3($s0)                      \n"
 282       "swr              $t8, 4($s0)                      \n"
 283       "swl              $t8, 7($s0)                      \n"
 284       "swr              $s5, 0($s1)                      \n"
 285       "swl              $s5, 3($s1)                      \n"
 286       "swr              $t1, 4($s1)                      \n"
 287       "swl              $t1, 7($s1)                      \n"
 288       "swr              $s7, 0($s2)                      \n"
 289       "swl              $s7, 3($s2)                      \n"
 290       "swr              $t9, 4($s2)                      \n"
 291       "swl              $t9, 7($s2)                      \n"
 292
 293       "addiu            $AT, -1                          \n"
 294       "addiu            %[src], 4                        \n"
 295
 296       "bnez             $AT, 11b                         \n"
 297       " addu            %[dst], $s2, %[dst_stride]       \n"
 298       "2:                                                \n"
 299       ".set pop                                          \n"
 300       ".set at                                           \n"
 301       :[src] "+r" (src),
 302        [dst] "+r" (dst),
 303        [width] "+r" (width)
 304       :[src_stride] "r" (src_stride),
 305        [dst_stride] "r" (dst_stride)
 306       : "t0", "t1",  "t2", "t3",  "t4", "t5",
 307         "t6", "t7", "t8", "t9",
 308         "s0", "s1", "s2", "s3", "s4",
 309         "s5", "s6", "s7"
 310   );
 311 }
 312
 313 void TransposeUVWx8_MIPS_DSPR2(const uint8* src, int src_stride,
 314                                uint8* dst_a, int dst_stride_a,
 315                                uint8* dst_b, int dst_stride_b,
 316                                int width) {
 317   __asm__ __volatile__ (
 318       ".set push                                         \n"
 319       ".set noreorder                                    \n"
 320       "beqz            %[width], 2f                      \n"
 321       " sll            $t2, %[src_stride], 0x1           \n" // src_stride x 2
 322       "sll             $t4, %[src_stride], 0x2           \n" // src_stride x 4
 323       "sll             $t9, %[src_stride], 0x3           \n" // src_stride x 8
 324       "addu            $t3, $t2, %[src_stride]           \n"
 325       "addu            $t5, $t4, %[src_stride]           \n"
 326       "addu            $t6, $t2, $t4                     \n"
 327       "subu            $t7, $t9, %[src_stride]           \n"
 328       "srl             $t1, %[width], 1                  \n"
 329
 330 // check word aligment for dst_a, dst_b, dst_stride_a and dst_stride_b
 331       "andi            $t0, %[dst_a], 0x3                \n"
 332       "andi            $t8, %[dst_b], 0x3                \n"
 333       "or              $t0, $t0, $t8                     \n"
 334       "andi            $t8, %[dst_stride_a], 0x3         \n"
 335       "andi            $s5, %[dst_stride_b], 0x3         \n"
 336       "or              $t8, $t8, $s5                     \n"
 337       "or              $t0, $t0, $t8                     \n"
 338       "bnez            $t0, 11f                          \n"
 339       " nop                                              \n"
 340 // dst + dst_stride word aligned (both, a & b dst addresses)
 341     "1:                                                  \n"
 342       "lw              $t0, 0(%[src])                    \n" // |B0|A0|b0|a0|
 343       "lwx             $t8, %[src_stride](%[src])        \n" // |B1|A1|b1|a1|
 344       "addu            $s5, %[dst_a], %[dst_stride_a]    \n"
 345       "lwx             $t9, $t2(%[src])                  \n" // |B2|A2|b2|a2|
 346       "lwx             $s0, $t3(%[src])                  \n" // |B3|A3|b3|a3|
 347       "addu            $s6, %[dst_b], %[dst_stride_b]    \n"
 348
 349       "precrq.ph.w     $s1, $t8, $t0                     \n" // |B1|A1|B0|A0|
 350       "precrq.ph.w     $s2, $s0, $t9                     \n" // |B3|A3|B2|A2|
 351       "precr.qb.ph     $s3, $s2, $s1                     \n" // |A3|A2|A1|A0|
 352       "precrq.qb.ph    $s4, $s2, $s1                     \n" // |B3|B2|B1|B0|
 353
 354       "sll             $t0, $t0, 16                      \n"
 355       "packrl.ph       $s1, $t8, $t0                     \n" // |b1|a1|b0|a0|
 356       "sll             $t9, $t9, 16                      \n"
 357       "packrl.ph       $s2, $s0, $t9                     \n" // |b3|a3|b2|a2|
 358
 359       "sw              $s3, 0($s5)                       \n"
 360       "sw              $s4, 0($s6)                       \n"
 361
 362       "precr.qb.ph     $s3, $s2, $s1                     \n" // |a3|a2|a1|a0|
 363       "precrq.qb.ph    $s4, $s2, $s1                     \n" // |b3|b2|b1|b0|
 364
 365       "lwx             $t0, $t4(%[src])                  \n" // |B4|A4|b4|a4|
 366       "lwx             $t8, $t5(%[src])                  \n" // |B5|A5|b5|a5|
 367       "lwx             $t9, $t6(%[src])                  \n" // |B6|A6|b6|a6|
 368       "lwx             $s0, $t7(%[src])                  \n" // |B7|A7|b7|a7|
 369       "sw              $s3, 0(%[dst_a])                  \n"
 370       "sw              $s4, 0(%[dst_b])                  \n"
 371
 372       "precrq.ph.w     $s1, $t8, $t0                     \n" // |B5|A5|B4|A4|
 373       "precrq.ph.w     $s2, $s0, $t9                     \n" // |B6|A6|B7|A7|
 374       "precr.qb.ph     $s3, $s2, $s1                     \n" // |A7|A6|A5|A4|
 375       "precrq.qb.ph    $s4, $s2, $s1                     \n" // |B7|B6|B5|B4|
 376
 377       "sll             $t0, $t0, 16                      \n"
 378       "packrl.ph       $s1, $t8, $t0                     \n" // |b5|a5|b4|a4|
 379       "sll             $t9, $t9, 16                      \n"
 380       "packrl.ph       $s2, $s0, $t9                     \n" // |b7|a7|b6|a6|
 381       "sw              $s3, 4($s5)                       \n"
 382       "sw              $s4, 4($s6)                       \n"
 383
 384       "precr.qb.ph     $s3, $s2, $s1                     \n" // |a7|a6|a5|a4|
 385       "precrq.qb.ph    $s4, $s2, $s1                     \n" // |b7|b6|b5|b4|
 386
 387       "addiu           %[src], 4                         \n"
 388       "addiu           $t1, -1                           \n"
 389       "sll             $t0, %[dst_stride_a], 1           \n"
 390       "sll             $t8, %[dst_stride_b], 1           \n"
 391       "sw              $s3, 4(%[dst_a])                  \n"
 392       "sw              $s4, 4(%[dst_b])                  \n"
 393       "addu            %[dst_a], %[dst_a], $t0           \n"
 394       "bnez            $t1, 1b                           \n"
 395       " addu           %[dst_b], %[dst_b], $t8           \n"
 396       "b               2f                                \n"
 397       " nop                                              \n"
 398
 399 // dst_a or dst_b or dst_stride_a or dst_stride_b not word aligned
 400    "11:                                                  \n"
 401       "lw              $t0, 0(%[src])                    \n" // |B0|A0|b0|a0|
 402       "lwx             $t8, %[src_stride](%[src])        \n" // |B1|A1|b1|a1|
 403       "addu            $s5, %[dst_a], %[dst_stride_a]    \n"
 404       "lwx             $t9, $t2(%[src])                  \n" // |B2|A2|b2|a2|
 405       "lwx             $s0, $t3(%[src])                  \n" // |B3|A3|b3|a3|
 406       "addu            $s6, %[dst_b], %[dst_stride_b]    \n"
 407
 408       "precrq.ph.w     $s1, $t8, $t0                     \n" // |B1|A1|B0|A0|
 409       "precrq.ph.w     $s2, $s0, $t9                     \n" // |B3|A3|B2|A2|
 410       "precr.qb.ph     $s3, $s2, $s1                     \n" // |A3|A2|A1|A0|
 411       "precrq.qb.ph    $s4, $s2, $s1                     \n" // |B3|B2|B1|B0|
 412
 413       "sll             $t0, $t0, 16                      \n"
 414       "packrl.ph       $s1, $t8, $t0                     \n" // |b1|a1|b0|a0|
 415       "sll             $t9, $t9, 16                      \n"
 416       "packrl.ph       $s2, $s0, $t9                     \n" // |b3|a3|b2|a2|
 417
 418       "swr             $s3, 0($s5)                       \n"
 419       "swl             $s3, 3($s5)                       \n"
 420       "swr             $s4, 0($s6)                       \n"
 421       "swl             $s4, 3($s6)                       \n"
 422
 423       "precr.qb.ph     $s3, $s2, $s1                     \n" // |a3|a2|a1|a0|
 424       "precrq.qb.ph    $s4, $s2, $s1                     \n" // |b3|b2|b1|b0|
 425
 426       "lwx             $t0, $t4(%[src])                  \n" // |B4|A4|b4|a4|
 427       "lwx             $t8, $t5(%[src])                  \n" // |B5|A5|b5|a5|
 428       "lwx             $t9, $t6(%[src])                  \n" // |B6|A6|b6|a6|
 429       "lwx             $s0, $t7(%[src])                  \n" // |B7|A7|b7|a7|
 430       "swr             $s3, 0(%[dst_a])                  \n"
 431       "swl             $s3, 3(%[dst_a])                  \n"
 432       "swr             $s4, 0(%[dst_b])                  \n"
 433       "swl             $s4, 3(%[dst_b])                  \n"
 434
 435       "precrq.ph.w     $s1, $t8, $t0                     \n" // |B5|A5|B4|A4|
 436       "precrq.ph.w     $s2, $s0, $t9                     \n" // |B6|A6|B7|A7|
 437       "precr.qb.ph     $s3, $s2, $s1                     \n" // |A7|A6|A5|A4|
 438       "precrq.qb.ph    $s4, $s2, $s1                     \n" // |B7|B6|B5|B4|
 439
 440       "sll             $t0, $t0, 16                      \n"
 441       "packrl.ph       $s1, $t8, $t0                     \n" // |b5|a5|b4|a4|
 442       "sll             $t9, $t9, 16                      \n"
 443       "packrl.ph       $s2, $s0, $t9                     \n" // |b7|a7|b6|a6|
 444
 445       "swr             $s3, 4($s5)                       \n"
 446       "swl             $s3, 7($s5)                       \n"
 447       "swr             $s4, 4($s6)                       \n"
 448       "swl             $s4, 7($s6)                       \n"
 449
 450       "precr.qb.ph     $s3, $s2, $s1                     \n" // |a7|a6|a5|a4|
 451       "precrq.qb.ph    $s4, $s2, $s1                     \n" // |b7|b6|b5|b4|
 452
 453       "addiu           %[src], 4                         \n"
 454       "addiu           $t1, -1                           \n"
 455       "sll             $t0, %[dst_stride_a], 1           \n"
 456       "sll             $t8, %[dst_stride_b], 1           \n"
 457       "swr             $s3, 4(%[dst_a])                  \n"
 458       "swl             $s3, 7(%[dst_a])                  \n"
 459       "swr             $s4, 4(%[dst_b])                  \n"
 460       "swl             $s4, 7(%[dst_b])                  \n"
 461       "addu            %[dst_a], %[dst_a], $t0           \n"
 462       "bnez            $t1, 11b                          \n"
 463       " addu           %[dst_b], %[dst_b], $t8           \n"
 464
 465       "2:                                                \n"
 466       ".set pop                                          \n"
 467       : [src] "+r" (src),
 468         [dst_a] "+r" (dst_a),
 469         [dst_b] "+r" (dst_b),
 470         [width] "+r" (width),
 471         [src_stride] "+r" (src_stride)
 472       : [dst_stride_a] "r" (dst_stride_a),
 473         [dst_stride_b] "r" (dst_stride_b)
 474       : "t0", "t1",  "t2", "t3",  "t4", "t5",
 475         "t6", "t7", "t8", "t9",
 476         "s0", "s1", "s2", "s3",
 477         "s4", "s5", "s6"
 478   );
 479 }
 480
 481 #endif  // defined(__mips_dsp) && (__mips_dsp_rev >= 2)
 482
 483 #ifdef __cplusplus
 484 }  // extern "C"
 485 }  // namespace libyuv
 486 #endif