git.sesse.net Git - ffmpeg/blob - libavcodec/mips/h264chroma_mmi.c

   1 /*
   2  * Loongson SIMD optimized h264chroma
   3  *
   4  * Copyright (c) 2015 Loongson Technology Corporation Limited
   5  * Copyright (c) 2015 Zhou Xiaoyong <zhouxiaoyong@loongson.cn>
   6  *                    Zhang Shuangshuang <zhangshuangshuang@ict.ac.cn>
   7  *
   8  * This file is part of FFmpeg.
   9  *
  10  * FFmpeg is free software; you can redistribute it and/or
  11  * modify it under the terms of the GNU Lesser General Public
  12  * License as published by the Free Software Foundation; either
  13  * version 2.1 of the License, or (at your option) any later version.
  14  *
  15  * FFmpeg is distributed in the hope that it will be useful,
  16  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  17  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  18  * Lesser General Public License for more details.
  19  *
  20  * You should have received a copy of the GNU Lesser General Public
  21  * License along with FFmpeg; if not, write to the Free Software
  22  * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
  23  */
  24
  25 #include "h264chroma_mips.h"
  26
  27 void ff_put_h264_chroma_mc8_mmi(uint8_t *dst, uint8_t *src, int stride,
  28         int h, int x, int y)
  29 {
  30     const int A = (8 - x) * (8 - y);
  31     const int B = x * (8 - y);
  32     const int C = (8 - x) * y;
  33     const int D = x * y;
  34     const int E = B + C;
  35     int i;
  36
  37     av_assert2(x<8 && y<8 && x>=0 && y>=0);
  38
  39     if (D) {
  40         for (i=0; i<h; i++) {
  41             __asm__ volatile (
  42                 "ldl $2, %2                 \r\n"
  43                 "ldr $2, %1                 \r\n"
  44                 "ldl $3, %4                 \r\n"
  45                 "ldr $3, %3                 \r\n"
  46                 "ldl $4, %6                 \r\n"
  47                 "ldr $4, %5                 \r\n"
  48                 "ldl $5, %8                 \r\n"
  49                 "ldr $5, %7                 \r\n"
  50                 "daddiu $6, $0, 32          \r\n"
  51                 "mtc1 %9, $f6               \r\n"
  52                 "mtc1 %10, $f8              \r\n"
  53                 "mtc1 %11, $f10             \r\n"
  54                 "mtc1 %12, $f12             \r\n"
  55                 "mtc1 $0, $f20              \r\n"
  56                 "mtc1 $2, $f2               \r\n"
  57                 "mtc1 $3, $f4               \r\n"
  58                 "mtc1 $4, $f16              \r\n"
  59                 "mtc1 $5, $f18              \r\n"
  60                 "mtc1 $6, $f14              \r\n"
  61                 "punpcklbh $f2, $f2, $f20   \r\n"
  62                 "punpcklbh $f4, $f4, $f20   \r\n"
  63                 "pshufh $f6, $f6, $f20      \r\n"
  64                 "pshufh $f8, $f8, $f20      \r\n"
  65                 "pshufh $f10, $f10, $f20    \r\n"
  66                 "pshufh $f12, $f12, $f20    \r\n"
  67                 "pshufh $f14, $f14, $f20    \r\n"
  68                 "punpcklbh $f16, $f16, $f20 \r\n"
  69                 "punpcklbh $f18, $f18, $f20 \r\n"
  70                 "daddiu $6, $0, 6           \r\n"
  71                 "mtc1 $6, $f22              \r\n"
  72                 "dsrl32 $2, $2, 0           \r\n"
  73                 "dsrl32 $3, $3, 0           \r\n"
  74                 "dsrl32 $4, $4, 0           \r\n"
  75                 "dsrl32 $5, $5, 0           \r\n"
  76                 "pmullh $f2, $f2, $f6       \r\n"
  77                 "pmullh $f4, $f4, $f8       \r\n"
  78                 "pmullh $f16, $f10, $f16    \r\n"
  79                 "pmullh $f18, $f12, $f18    \r\n"
  80                 "paddh $f2, $f2, $f14       \r\n"
  81                 "paddh $f4, $f4, $f16       \r\n"
  82                 "paddh $f2, $f2, $f18       \r\n"
  83                 "paddh $f2, $f2, $f4        \r\n"
  84                 "psrah $f24, $f2, $f22      \r\n"
  85                 "mtc1 $2, $f2               \r\n"
  86                 "mtc1 $3, $f4               \r\n"
  87                 "mtc1 $4, $f16              \r\n"
  88                 "mtc1 $5, $f18              \r\n"
  89                 "punpcklbh $f2, $f2, $f20   \r\n"
  90                 "punpcklbh $f4, $f4, $f20   \r\n"
  91                 "punpcklbh $f16, $f16, $f20 \r\n"
  92                 "punpcklbh $f18, $f18, $f20 \r\n"
  93                 "pmullh $f2, $f2, $f6       \r\n"
  94                 "pmullh $f4, $f4, $f8       \r\n"
  95                 "pmullh $f16, $f10, $f16    \r\n"
  96                 "pmullh $f18, $f12, $f18    \r\n"
  97                 "paddh $f2, $f2, $f14       \r\n"
  98                 "paddh $f4, $f4, $f16       \r\n"
  99                 "paddh $f2, $f2, $f18       \r\n"
 100                 "paddh $f2, $f2, $f4        \r\n"
 101                 "psrah $f2, $f2, $f22       \r\n"
 102                 "packushb $f2, $f24, $f2    \r\n"
 103                 "sdc1 $f2, %0               \r\n"
 104                 : "=m"(*dst)
 105                 : "m"(*src),"m"(*(src+7)),"m"(*(src+1)),"m"(*(src+8)),
 106                   "m"(*(src+stride)),"m"(*(src+stride+7)),
 107                   "m"(*(src+stride+1)),"m"(*(src+stride+8)),
 108                   "r"(A),"r"(B),"r"(C),"r"(D)
 109                 : "$2","$3","$4","$5","$6"
 110             );
 111
 112             dst += stride;
 113             src += stride;
 114         }
 115     } else if (E) {
 116         const int step = C ? stride : 1;
 117
 118         for (i=0; i<h; i++) {
 119             __asm__ volatile (
 120                 "daddiu $6, $0, 32          \r\n"
 121                 "ldl $2, %2                 \r\n"
 122                 "ldr $2, %1                 \r\n"
 123                 "ldl $3, %4                 \r\n"
 124                 "ldr $3, %3                 \r\n"
 125                 "mtc1 $6, $f14              \r\n"
 126                 "mtc1 %5, $f6               \r\n"
 127                 "mtc1 %6, $f8               \r\n"
 128                 "mtc1 $0, $f20              \r\n"
 129                 "mtc1 $2, $f2               \r\n"
 130                 "mtc1 $3, $f4               \r\n"
 131                 "daddiu $6, $0, 6           \r\n"
 132                 "punpcklbh $f2, $f2, $f20   \r\n"
 133                 "punpcklbh $f4, $f4, $f20   \r\n"
 134                 "pshufh $f6, $f6, $f20      \r\n"
 135                 "pshufh $f8, $f8, $f20      \r\n"
 136                 "pshufh $f14, $f14, $f20    \r\n"
 137                 "mtc1 $6, $f22              \r\n"
 138                 "dsrl32 $2, $2, 0           \r\n"
 139                 "dsrl32 $3, $3, 0           \r\n"
 140                 "pmullh $f2, $f2, $f6       \r\n"
 141                 "pmullh $f4, $f4, $f8       \r\n"
 142                 "paddh $f2, $f2, $f14       \r\n"
 143                 "paddh $f2, $f2, $f4        \r\n"
 144                 "psrah $f24, $f2, $f22      \r\n"
 145                 "mtc1 $2, $f2               \r\n"
 146                 "mtc1 $3, $f4               \r\n"
 147                 "punpcklbh $f2, $f2, $f20   \r\n"
 148                 "punpcklbh $f4, $f4, $f20   \r\n"
 149                 "pmullh $f2, $f2, $f6       \r\n"
 150                 "pmullh $f4, $f4, $f8       \r\n"
 151                 "paddh $f2, $f2, $f14       \r\n"
 152                 "paddh $f2, $f2, $f4        \r\n"
 153                 "psrah $f2, $f2, $f22       \r\n"
 154                 "packushb $f2, $f24, $f2    \r\n"
 155                 "sdc1 $f2, %0               \r\n"
 156                 : "=m"(*dst)
 157                 : "m"(*(src)),"m"(*(src+7)),
 158                   "m"(*(src+step)),"m"(*(src+step+7)),
 159                   "r"(A),"r"(E)
 160                 : "$2","$3","$4","$5","$6"
 161             );
 162
 163             dst += stride;
 164             src += stride;
 165         }
 166     } else {
 167         for (i = 0; i < h; i++) {
 168             __asm__ volatile (
 169                 "daddiu $6, $0, 32          \r\n"
 170                 "ldl $2, %2                 \r\n"
 171                 "ldr $2, %1                 \r\n"
 172                 "mtc1 $6, $f14              \r\n"
 173                 "mtc1 %3, $f6               \r\n"
 174                 "mtc1 $0, $f20              \r\n"
 175                 "mtc1 $2, $f2               \r\n"
 176                 "daddiu $6, $0, 6           \r\n"
 177                 "punpcklbh $f2, $f2, $f20   \r\n"
 178                 "pshufh $f6, $f6, $f20      \r\n"
 179                 "pshufh $f14, $f14, $f20    \r\n"
 180                 "mtc1 $6, $f22              \r\n"
 181                 "dsrl32 $2, $2, 0           \r\n"
 182                 "pmullh $f2, $f2, $f6       \r\n"
 183                 "paddh $f2, $f2, $f14       \r\n"
 184                 "psrah $f24, $f2, $f22      \r\n"
 185                 "mtc1 $2, $f2               \r\n"
 186                 "punpcklbh $f2, $f2, $f20   \r\n"
 187                 "pmullh $f2, $f2, $f6       \r\n"
 188                 "paddh $f2, $f2, $f14       \r\n"
 189                 "psrah $f2, $f2, $f22       \r\n"
 190                 "packushb $f2, $f24, $f2    \r\n"
 191                 "sdc1 $f2, %0               \r\n"
 192                 :"=m"(*dst)
 193                 :"m"(*src),"m"(*(src+7)),"r"(A)
 194                 :"$2"
 195             );
 196
 197             dst += stride;
 198             src += stride;
 199         }
 200     }
 201 }
 202
 203 void ff_avg_h264_chroma_mc8_mmi(uint8_t *dst, uint8_t *src, int stride,
 204         int h, int x, int y)
 205 {
 206     const int A = (8 - x) * (8 - y);
 207     const int B = x * (8 - y);
 208     const int C = (8 - x) * y;
 209     const int D = x * y;
 210     const int E = B + C;
 211     int i;
 212
 213     av_assert2(x<8 && y<8 && x>=0 && y>=0);
 214
 215     if (D) {
 216         for (i=0; i<h; i++) {
 217             __asm__ volatile (
 218                 "ldl $2, %2                 \r\n"
 219                 "ldr $2, %1                 \r\n"
 220                 "ldl $3, %4                 \r\n"
 221                 "ldr $3, %3                 \r\n"
 222                 "ldl $4, %6                 \r\n"
 223                 "ldr $4, %5                 \r\n"
 224                 "ldl $5, %8                 \r\n"
 225                 "ldr $5, %7                 \r\n"
 226                 "daddiu $6, $0, 32          \r\n"
 227                 "mtc1 %9, $f6               \r\n"
 228                 "mtc1 %10, $f8              \r\n"
 229                 "mtc1 %11, $f10             \r\n"
 230                 "mtc1 %12, $f12             \r\n"
 231                 "mtc1 $0, $f20              \r\n"
 232                 "mtc1 $2, $f2               \r\n"
 233                 "mtc1 $3, $f4               \r\n"
 234                 "mtc1 $4, $f16              \r\n"
 235                 "mtc1 $5, $f18              \r\n"
 236                 "mtc1 $6, $f14              \r\n"
 237                 "punpcklbh $f2, $f2, $f20   \r\n"
 238                 "punpcklbh $f4, $f4, $f20   \r\n"
 239                 "pshufh $f6, $f6, $f20      \r\n"
 240                 "pshufh $f8, $f8, $f20      \r\n"
 241                 "pshufh $f10, $f10, $f20    \r\n"
 242                 "pshufh $f12, $f12, $f20    \r\n"
 243                 "pshufh $f14, $f14, $f20    \r\n"
 244                 "punpcklbh $f16, $f16, $f20 \r\n"
 245                 "punpcklbh $f18, $f18, $f20 \r\n"
 246                 "daddiu $6, $0, 6           \r\n"
 247                 "mtc1 $6, $f22              \r\n"
 248                 "dsrl32 $2, $2, 0           \r\n"
 249                 "dsrl32 $3, $3, 0           \r\n"
 250                 "dsrl32 $4, $4, 0           \r\n"
 251                 "dsrl32 $5, $5, 0           \r\n"
 252                 "pmullh $f2, $f2, $f6       \r\n"
 253                 "pmullh $f4, $f4, $f8       \r\n"
 254                 "pmullh $f16, $f10, $f16    \r\n"
 255                 "pmullh $f18, $f12, $f18    \r\n"
 256                 "paddh $f2, $f2, $f14       \r\n"
 257                 "paddh $f4, $f4, $f16       \r\n"
 258                 "paddh $f2, $f2, $f18       \r\n"
 259                 "paddh $f2, $f2, $f4        \r\n"
 260                 "psrah $f24, $f2, $f22      \r\n"
 261                 "mtc1 $2, $f2               \r\n"
 262                 "mtc1 $3, $f4               \r\n"
 263                 "mtc1 $4, $f16              \r\n"
 264                 "mtc1 $5, $f18              \r\n"
 265                 "punpcklbh $f2, $f2, $f20   \r\n"
 266                 "punpcklbh $f4, $f4, $f20   \r\n"
 267                 "punpcklbh $f16, $f16, $f20 \r\n"
 268                 "punpcklbh $f18, $f18, $f20 \r\n"
 269                 "pmullh $f2, $f2, $f6       \r\n"
 270                 "pmullh $f4, $f4, $f8       \r\n"
 271                 "pmullh $f16, $f10, $f16    \r\n"
 272                 "pmullh $f18, $f12, $f18    \r\n"
 273                 "paddh $f2, $f2, $f14       \r\n"
 274                 "paddh $f4, $f4, $f16       \r\n"
 275                 "paddh $f2, $f2, $f18       \r\n"
 276                 "paddh $f2, $f2, $f4        \r\n"
 277                 "psrah $f2, $f2, $f22       \r\n"
 278                 "packushb $f2, $f24, $f2    \r\n"
 279                 "ldc1 $f4, %0               \r\n"
 280                 "pavgb $f2, $f2, $f4        \r\n"
 281                 "sdc1 $f2, %0               \r\n"
 282                 : "=m"(*dst)
 283                 : "m"(*(src)),"m"(*(src+7)),"m"(*(src+1)),"m"(*(src+8)),
 284                   "m"(*(src+stride)),"m"(*(src+stride+7)),
 285                   "m"(*(src+stride+1)),"m"(*(src+stride+8)),
 286                   "r"(A),"r"(B),"r"(C),"r"(D)
 287                 : "$2","$3","$4","$5","$6"
 288             );
 289
 290             dst += stride;
 291             src += stride;
 292         }
 293     } else {
 294         const int step = C ? stride : 1;
 295
 296         for (i=0; i<h; i++) {
 297             __asm__ volatile (
 298                 "daddiu $6, $0, 32          \r\n"
 299                 "ldl $2, %2                 \r\n"
 300                 "ldr $2, %1                 \r\n"
 301                 "ldl $3, %4                 \r\n"
 302                 "ldr $3, %3                 \r\n"
 303                 "mtc1 $6, $f14              \r\n"
 304                 "mtc1 %5, $f6               \r\n"
 305                 "mtc1 %6, $f8               \r\n"
 306                 "mtc1 $0, $f20              \r\n"
 307                 "mtc1 $2, $f2               \r\n"
 308                 "mtc1 $3, $f4               \r\n"
 309                 "daddiu $6, $0, 6           \r\n"
 310                 "punpcklbh $f2, $f2, $f20   \r\n"
 311                 "punpcklbh $f4, $f4, $f20   \r\n"
 312                 "pshufh $f6, $f6, $f20      \r\n"
 313                 "pshufh $f8, $f8, $f20      \r\n"
 314                 "pshufh $f14, $f14, $f20    \r\n"
 315                 "mtc1 $6, $f22              \r\n"
 316                 "dsrl32 $2, $2, 0           \r\n"
 317                 "dsrl32 $3, $3, 0           \r\n"
 318                 "pmullh $f2, $f2, $f6       \r\n"
 319                 "pmullh $f4, $f4, $f8       \r\n"
 320                 "paddh $f2, $f2, $f14       \r\n"
 321                 "paddh $f2, $f2, $f4        \r\n"
 322                 "psrah $f24, $f2, $f22      \r\n"
 323                 "mtc1 $2, $f2               \r\n"
 324                 "mtc1 $3, $f4               \r\n"
 325                 "punpcklbh $f2, $f2, $f20   \r\n"
 326                 "punpcklbh $f4, $f4, $f20   \r\n"
 327                 "pmullh $f2, $f2, $f6       \r\n"
 328                 "pmullh $f4, $f4, $f8       \r\n"
 329                 "paddh $f2, $f2, $f14       \r\n"
 330                 "paddh $f2, $f2, $f4        \r\n"
 331                 "psrah $f2, $f2, $f22       \r\n"
 332                 "packushb $f2, $f24, $f2    \r\n"
 333                 "ldc1 $f4, %0               \r\n"
 334                 "pavgb $f2, $f2, $f4        \r\n"
 335                 "sdc1 $f2, %0               \r\n"
 336                 : "=m"(*dst)
 337                 : "m"(*(src)),"m"(*(src+7)),
 338                   "m"(*(src+step)),"m"(*(src+step+7)),"r"(A),"r"(E)
 339                 : "$2","$3","$4","$5","$6"
 340             );
 341
 342             dst += stride;
 343             src += stride;
 344         }
 345     }
 346 }
 347
 348 void ff_put_h264_chroma_mc4_mmi(uint8_t *dst, uint8_t *src, int stride,
 349         int h, int x, int y)
 350 {
 351     const int A = (8 - x) * (8 - y);
 352     const int B = x * (8 - y);
 353     const int C = (8 - x) *  y;
 354     const int D = x *  y;
 355     const int E = B + C;
 356     int i;
 357
 358     av_assert2(x<8 && y<8 && x>=0 && y>=0);
 359
 360     if (D) {
 361         for (i=0; i<h; i++) {
 362             __asm__ volatile (
 363                 "ldl $2, %2                 \r\n"
 364                 "ldr $2, %1                 \r\n"
 365                 "ldl $3, %4                 \r\n"
 366                 "ldr $3, %3                 \r\n"
 367                 "ldl $4, %6                 \r\n"
 368                 "ldr $4, %5                 \r\n"
 369                 "ldl $5, %8                 \r\n"
 370                 "ldr $5, %7                 \r\n"
 371                 "daddiu $6, $0, 32          \r\n"
 372                 "mtc1 %9, $f6               \r\n"
 373                 "mtc1 %10, $f8              \r\n"
 374                 "mtc1 %11, $f10             \r\n"
 375                 "mtc1 %12, $f12             \r\n"
 376                 "mtc1 $0, $f20              \r\n"
 377                 "mtc1 $2, $f2               \r\n"
 378                 "mtc1 $3, $f4               \r\n"
 379                 "mtc1 $4, $f16              \r\n"
 380                 "mtc1 $5, $f18              \r\n"
 381                 "mtc1 $6, $f14              \r\n"
 382                 "punpcklbh $f2, $f2, $f20   \r\n"
 383                 "punpcklbh $f4, $f4, $f20   \r\n"
 384                 "pshufh $f6, $f6, $f20      \r\n"
 385                 "pshufh $f8, $f8, $f20      \r\n"
 386                 "pshufh $f10, $f10, $f20    \r\n"
 387                 "pshufh $f12, $f12, $f20    \r\n"
 388                 "pshufh $f14, $f14, $f20    \r\n"
 389                 "punpcklbh $f16, $f16, $f20 \r\n"
 390                 "punpcklbh $f18, $f18, $f20 \r\n"
 391                 "daddiu $6, $0, 6           \r\n"
 392                 "mtc1 $6, $f22              \r\n"
 393                 "pmullh $f2, $f2, $f6       \r\n"
 394                 "pmullh $f4, $f4, $f8       \r\n"
 395                 "pmullh $f16, $f10, $f16    \r\n"
 396                 "pmullh $f18, $f12, $f18    \r\n"
 397                 "paddh $f2, $f2, $f14       \r\n"
 398                 "paddh $f4, $f4, $f16       \r\n"
 399                 "paddh $f2, $f2, $f18       \r\n"
 400                 "paddh $f2, $f2, $f4        \r\n"
 401                 "psrah $f2, $f2, $f22       \r\n"
 402                 "packushb $f2, $f2, $f2     \r\n"
 403                 "swc1 $f2, %0               \r\n"
 404                 : "=m"(*dst)
 405                 : "m"(*(src)),"m"(*(src+7)),"m"(*(src+1)),"m"(*(src+8)),
 406                   "m"(*(src+stride)),"m"(*(src+stride+7)),
 407                   "m"(*(src+stride+1)),"m"(*(src+stride+8)),
 408                   "r"(A),"r"(B),"r"(C),"r"(D)
 409                 : "$2","$3","$4","$5","$6"
 410             );
 411
 412             dst += stride;
 413             src += stride;
 414         }
 415     } else if (E) {
 416         const int step = C ? stride : 1;
 417
 418         for (i=0; i<h; i++) {
 419             __asm__ volatile (
 420                 "ldl $2, %2                 \r\n"
 421                 "ldr $2, %1                 \r\n"
 422                 "ldl $3, %4                 \r\n"
 423                 "ldr $3, %3                 \r\n"
 424                 "daddiu $4, $0, 32          \r\n"
 425                 "mtc1 %5, $f6               \r\n"
 426                 "mtc1 %6, $f8               \r\n"
 427                 "mtc1 $0, $f20              \r\n"
 428                 "mtc1 $2, $f2               \r\n"
 429                 "mtc1 $3, $f4               \r\n"
 430                 "mtc1 $4, $f10              \r\n"
 431                 "punpcklbh $f2, $f2, $f20   \r\n"
 432                 "punpcklbh $f4, $f4, $f20   \r\n"
 433                 "pshufh $f6, $f6, $f20      \r\n"
 434                 "pshufh $f8, $f8, $f20      \r\n"
 435                 "pshufh $f10, $f10, $f20    \r\n"
 436                 "daddiu $4, $0, 6           \r\n"
 437                 "mtc1 $4, $f22              \r\n"
 438                 "pmullh $f2, $f2, $f6       \r\n"
 439                 "pmullh $f4, $f4, $f8       \r\n"
 440                 "paddh $f2, $f2, $f10       \r\n"
 441                 "paddh $f2, $f2, $f4        \r\n"
 442                 "psrah $f2, $f2, $f22       \r\n"
 443                 "packushb $f2, $f2, $f20    \r\n"
 444                 "swc1 $f2, %0               \r\n"
 445                 : "=m"(*dst)
 446                 : "m"(*(src)),"m"(*(src+7)),"m"(*(src+step)),
 447                   "m"(*(src+step+7)),"r"(A),"r"(E)
 448                 : "$2","$3","$4","$5","$6"
 449             );
 450
 451             dst += stride;
 452             src += stride;
 453         }
 454     } else {
 455         for (i=0; i<h; i++) {
 456             __asm__ volatile (
 457                 "lwl $2, %2                 \r\n"
 458                 "lwr $2, %1                 \r\n"
 459                 "sw $2, %0                  \r\n"
 460                 : "=m"(*dst)
 461                 : "m"(*src),"m"(*(src+3))
 462                 : "$2"
 463             );
 464
 465             dst += stride;
 466             src += stride;
 467         }
 468     }
 469 }
 470
 471 void ff_avg_h264_chroma_mc4_mmi(uint8_t *dst, uint8_t *src, int stride,
 472         int h, int x, int y)
 473 {
 474     const int A = (8 - x) *(8 - y);
 475     const int B = x * (8 - y);
 476     const int C = (8 - x) * y;
 477     const int D = x * y;
 478     int i;
 479
 480     av_assert2(x<8 && y<8 && x>=0 && y>=0);
 481
 482     if (D) {
 483         for (i=0; i<h; i++) {
 484             __asm__ volatile (
 485                 "ldl $2, %2                 \r\n"
 486                 "ldr $2, %1                 \r\n"
 487                 "ldl $3, %4                 \r\n"
 488                 "ldr $3, %3                 \r\n"
 489                 "ldl $4, %6                 \r\n"
 490                 "ldr $4, %5                 \r\n"
 491                 "ldl $5, %8                 \r\n"
 492                 "ldr $5, %7                 \r\n"
 493                 "daddiu $6, $0, 32          \r\n"
 494                 "mtc1 %9, $f6               \r\n"
 495                 "mtc1 %10, $f8              \r\n"
 496                 "mtc1 %11, $f10             \r\n"
 497                 "mtc1 %12, $f12             \r\n"
 498                 "mtc1 $0, $f20              \r\n"
 499                 "mtc1 $2, $f2               \r\n"
 500                 "mtc1 $3, $f4               \r\n"
 501                 "mtc1 $4, $f16              \r\n"
 502                 "mtc1 $5, $f18              \r\n"
 503                 "mtc1 $6, $f14              \r\n"
 504                 "punpcklbh $f2, $f2, $f20   \r\n"
 505                 "punpcklbh $f4, $f4, $f20   \r\n"
 506                 "pshufh $f6, $f6, $f20      \r\n"
 507                 "pshufh $f8, $f8, $f20      \r\n"
 508                 "pshufh $f10, $f10, $f20    \r\n"
 509                 "pshufh $f12, $f12, $f20    \r\n"
 510                 "pshufh $f14, $f14, $f20    \r\n"
 511                 "punpcklbh $f16, $f16, $f20 \r\n"
 512                 "punpcklbh $f18, $f18, $f20 \r\n"
 513                 "daddiu $6, $0, 6           \r\n"
 514                 "mtc1 $6, $f22              \r\n"
 515                 "pmullh $f2, $f2, $f6       \r\n"
 516                 "pmullh $f4, $f4, $f8       \r\n"
 517                 "pmullh $f16, $f10, $f16    \r\n"
 518                 "pmullh $f18, $f12, $f18    \r\n"
 519                 "paddh $f2, $f2, $f14       \r\n"
 520                 "paddh $f4, $f4, $f16       \r\n"
 521                 "paddh $f2, $f2, $f18       \r\n"
 522                 "paddh $f2, $f2, $f4        \r\n"
 523                 "psrah $f2, $f2, $f22       \r\n"
 524                 "packushb $f2, $f2, $f2     \r\n"
 525                 "lwc1 $f4, %0               \r\n"
 526                 "pavgb $f2, $f2, $f4        \r\n"
 527                 "swc1 $f2, %0               \r\n"
 528                 : "=m"(*dst)
 529                 : "m"(*(src)),"m"(*(src+7)),"m"(*(src+1)),"m"(*(src+8)),
 530                   "m"(*(src+stride)),"m"(*(src+stride+7)),
 531                   "m"(*(src+stride+1)),"m"(*(src+stride+8)),
 532                   "r"(A),"r"(B),"r"(C),"r"(D)
 533                 : "$2","$3","$4","$5","$6"
 534             );
 535
 536             dst += stride;
 537             src += stride;
 538         }
 539     } else {
 540         const int E = B + C;
 541         const int step = C ? stride : 1;
 542
 543         for (i=0; i<h; i++) {
 544             __asm__ volatile (
 545                 "ldl $2, %2                 \r\n"
 546                 "ldr $2, %1                 \r\n"
 547                 "ldl $3, %4                 \r\n"
 548                 "ldr $3, %3                 \r\n"
 549                 "daddiu $4, $0, 32          \r\n"
 550                 "mtc1 %5, $f6               \r\n"
 551                 "mtc1 %6, $f8               \r\n"
 552                 "mtc1 $0, $f20              \r\n"
 553                 "mtc1 $2, $f2               \r\n"
 554                 "mtc1 $3, $f4               \r\n"
 555                 "mtc1 $4, $f10              \r\n"
 556                 "punpcklbh $f2, $f2, $f20   \r\n"
 557                 "punpcklbh $f4, $f4, $f20   \r\n"
 558                 "pshufh $f6, $f6, $f20      \r\n"
 559                 "pshufh $f8, $f8, $f20      \r\n"
 560                 "pshufh $f10, $f10, $f20    \r\n"
 561                 "daddiu $4, $0, 6           \r\n"
 562                 "mtc1 $4, $f22              \r\n"
 563                 "pmullh $f2, $f2, $f6       \r\n"
 564                 "pmullh $f4, $f4, $f8       \r\n"
 565                 "paddh $f2, $f2, $f10       \r\n"
 566                 "paddh $f2, $f2, $f4        \r\n"
 567                 "psrah $f2, $f2, $f22       \r\n"
 568                 "packushb $f2, $f2, $f20    \r\n"
 569                 "lwc1 $f4, %0               \r\n"
 570                 "pavgb $f2, $f2, $f4        \r\n"
 571                 "swc1 $f2, %0               \r\n"
 572                 : "=m"(*dst)
 573                 : "m"(*(src)),"m"(*(src+7)),"m"(*(src+step)),
 574                   "m"(*(src+step+7)),"r"(A),"r"(E)
 575                 : "$2","$3","$4","$5","$6"
 576             );
 577
 578             dst += stride;
 579             src += stride;
 580         }
 581     }
 582 }