falkTX
/
FFmpeg
mirror of https://github.com/falkTX/FFmpeg.git
1
0
Fork 0
Code Issues 0 Releases 338 Wiki Activity
You can not select more than 25 topics Topics must start with a letter or number, can include dashes ('-') and can be up to 35 characters long.
64109 Commits
32 Branches
290MB
Tree: 0dceefc5fa
Branches Tags
${ item.name }
Create branch ${ searchTerm }
from '0dceefc5fa'
${ noResults }
FFmpeg/libswresample/x86/resample_mmx.h

301 lines
11KB
Raw Blame History 

  1. /*

  2.  * Copyright (c) 2012 Michael Niedermayer <michaelni@gmx.at>

  3.  *

  4.  * This file is part of FFmpeg.

  5.  *

  6.  * FFmpeg is free software; you can redistribute it and/or

  7.  * modify it under the terms of the GNU Lesser General Public

  8.  * License as published by the Free Software Foundation; either

  9.  * version 2.1 of the License, or (at your option) any later version.

  10.  *

  11.  * FFmpeg is distributed in the hope that it will be useful,

  12.  * but WITHOUT ANY WARRANTY; without even the implied warranty of

  13.  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU

  14.  * Lesser General Public License for more details.

  15.  *

  16.  * You should have received a copy of the GNU Lesser General Public

  17.  * License along with FFmpeg; if not, write to the Free Software

  18.  * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA

  19.  */

  20. 

  21. #include "libavutil/x86/asm.h"

  22. #include "libavutil/cpu.h"

  23. #include "libswresample/swresample_internal.h"

  24. 

  25. DECLARE_ALIGNED(16, const uint64_t, ff_resample_int16_rounder)[2]    = { 0x0000000000004000ULL, 0x0000000000000000ULL};

  26. 

  27. #define COMMON_CORE_INT16_MMX2 \

  28.     x86_reg len= -2*c->filter_length;\

  29. __asm__ volatile(\

  30.     "movq "MANGLE(ff_resample_int16_rounder)", %%mm0 \n\t"\

  31.     "1:                         \n\t"\

  32.     "movq    (%1, %0), %%mm1    \n\t"\

  33.     "pmaddwd (%2, %0), %%mm1    \n\t"\

  34.     "paddd  %%mm1, %%mm0        \n\t"\

  35.     "add       $8, %0           \n\t"\

  36.     " js 1b                     \n\t"\

  37.     "pshufw $0x0E, %%mm0, %%mm1 \n\t"\

  38.     "paddd %%mm1, %%mm0         \n\t"\

  39.     "psrad    $15, %%mm0        \n\t"\

  40.     "packssdw %%mm0, %%mm0      \n\t"\

  41.     "movd %%mm0, (%3)           \n\t"\

  42.     : "+r" (len)\

  43.     : "r" (((uint8_t*)(src+sample_index))-len),\

  44.       "r" (((uint8_t*)filter)-len),\

  45.       "r" (dst+dst_index)\

  46.       NAMED_CONSTRAINTS_ARRAY_ADD(ff_resample_int16_rounder)\

  47. );

  48. 

  49. #define LINEAR_CORE_INT16_MMX2 \

  50.     x86_reg len= -2*c->filter_length;\

  51. __asm__ volatile(\

  52.     "pxor          %%mm0, %%mm0 \n\t"\

  53.     "pxor          %%mm2, %%mm2 \n\t"\

  54.     "1:                         \n\t"\

  55.     "movq       (%3, %0), %%mm1 \n\t"\

  56.     "movq          %%mm1, %%mm3 \n\t"\

  57.     "pmaddwd    (%4, %0), %%mm1 \n\t"\

  58.     "pmaddwd    (%5, %0), %%mm3 \n\t"\

  59.     "paddd         %%mm1, %%mm0 \n\t"\

  60.     "paddd         %%mm3, %%mm2 \n\t"\

  61.     "add              $8, %0    \n\t"\

  62.     " js 1b                     \n\t"\

  63.     "pshufw $0x0E, %%mm0, %%mm1 \n\t"\

  64.     "pshufw $0x0E, %%mm2, %%mm3 \n\t"\

  65.     "paddd         %%mm1, %%mm0 \n\t"\

  66.     "paddd         %%mm3, %%mm2 \n\t"\

  67.     "movd          %%mm0, %1    \n\t"\

  68.     "movd          %%mm2, %2    \n\t"\

  69.     : "+r" (len),\

  70.       "=r" (val),\

  71.       "=r" (v2)\

  72.     : "r" (((uint8_t*)(src+sample_index))-len),\

  73.       "r" (((uint8_t*)filter)-len),\

  74.       "r" (((uint8_t*)(filter+c->filter_alloc))-len)\

  75. );

  76. 

  77. #define COMMON_CORE_INT16_SSE2 \

  78.     x86_reg len= -2*c->filter_length;\

  79. __asm__ volatile(\

  80.     "movdqa "MANGLE(ff_resample_int16_rounder)", %%xmm0 \n\t"\

  81.     "1:                           \n\t"\

  82.     "movdqu  (%1, %0), %%xmm1     \n\t"\

  83.     "pmaddwd (%2, %0), %%xmm1     \n\t"\

  84.     "paddd  %%xmm1, %%xmm0        \n\t"\

  85.     "add       $16, %0            \n\t"\

  86.     " js 1b                       \n\t"\

  87.     "pshufd $0x0E, %%xmm0, %%xmm1 \n\t"\

  88.     "paddd %%xmm1, %%xmm0         \n\t"\

  89.     "pshufd $0x01, %%xmm0, %%xmm1 \n\t"\

  90.     "paddd %%xmm1, %%xmm0         \n\t"\

  91.     "psrad    $15, %%xmm0         \n\t"\

  92.     "packssdw %%xmm0, %%xmm0      \n\t"\

  93.     "movd %%xmm0, (%3)            \n\t"\

  94.     : "+r" (len)\

  95.     : "r" (((uint8_t*)(src+sample_index))-len),\

  96.       "r" (((uint8_t*)filter)-len),\

  97.       "r" (dst+dst_index)\

  98.       NAMED_CONSTRAINTS_ARRAY_ADD(ff_resample_int16_rounder)\

  99.       XMM_CLOBBERS_ONLY("%xmm0", "%xmm1")\

  100. );

  101. 

  102. #define LINEAR_CORE_INT16_SSE2 \

  103.     x86_reg len= -2*c->filter_length;\

  104. __asm__ volatile(\

  105.     "pxor          %%xmm0, %%xmm0 \n\t"\

  106.     "pxor          %%xmm2, %%xmm2 \n\t"\

  107.     "1:                           \n\t"\

  108.     "movdqu      (%3, %0), %%xmm1 \n\t"\

  109.     "movdqa        %%xmm1, %%xmm3 \n\t"\

  110.     "pmaddwd     (%4, %0), %%xmm1 \n\t"\

  111.     "pmaddwd     (%5, %0), %%xmm3 \n\t"\

  112.     "paddd         %%xmm1, %%xmm0 \n\t"\

  113.     "paddd         %%xmm3, %%xmm2 \n\t"\

  114.     "add              $16, %0     \n\t"\

  115.     " js 1b                       \n\t"\

  116.     "pshufd $0x0E, %%xmm0, %%xmm1 \n\t"\

  117.     "pshufd $0x0E, %%xmm2, %%xmm3 \n\t"\

  118.     "paddd         %%xmm1, %%xmm0 \n\t"\

  119.     "paddd         %%xmm3, %%xmm2 \n\t"\

  120.     "pshufd $0x01, %%xmm0, %%xmm1 \n\t"\

  121.     "pshufd $0x01, %%xmm2, %%xmm3 \n\t"\

  122.     "paddd         %%xmm1, %%xmm0 \n\t"\

  123.     "paddd         %%xmm3, %%xmm2 \n\t"\

  124.     "movd          %%xmm0, %1     \n\t"\

  125.     "movd          %%xmm2, %2     \n\t"\

  126.     : "+r" (len),\

  127.       "=r" (val),\

  128.       "=r" (v2)\

  129.     : "r" (((uint8_t*)(src+sample_index))-len),\

  130.       "r" (((uint8_t*)filter)-len),\

  131.       "r" (((uint8_t*)(filter+c->filter_alloc))-len)\

  132.     XMM_CLOBBERS_ONLY("%xmm0", "%xmm1", "%xmm2", "%xmm3")\

  133. );

  134. 

  135. #define COMMON_CORE_FLT_SSE \

  136.     x86_reg len= -4*c->filter_length;\

  137. __asm__ volatile(\

  138.     "xorps     %%xmm0, %%xmm0     \n\t"\

  139.     "1:                           \n\t"\

  140.     "movups  (%1, %0), %%xmm1     \n\t"\

  141.     "mulps   (%2, %0), %%xmm1     \n\t"\

  142.     "addps     %%xmm1, %%xmm0     \n\t"\

  143.     "add       $16, %0            \n\t"\

  144.     " js 1b                       \n\t"\

  145.     "movhlps   %%xmm0, %%xmm1     \n\t"\

  146.     "addps     %%xmm1, %%xmm0     \n\t"\

  147.     "movss     %%xmm0, %%xmm1     \n\t"\

  148.     "shufps $1, %%xmm0, %%xmm0    \n\t"\

  149.     "addps     %%xmm1, %%xmm0     \n\t"\

  150.     "movss     %%xmm0, (%3)       \n\t"\

  151.     : "+r" (len)\

  152.     : "r" (((uint8_t*)(src+sample_index))-len),\

  153.       "r" (((uint8_t*)filter)-len),\

  154.       "r" (dst+dst_index)\

  155.     XMM_CLOBBERS_ONLY("%xmm0", "%xmm1")\

  156. );

  157. 

  158. #define LINEAR_CORE_FLT_SSE \

  159.     x86_reg len= -4*c->filter_length;\

  160. __asm__ volatile(\

  161.     "xorps      %%xmm0, %%xmm0    \n\t"\

  162.     "xorps      %%xmm2, %%xmm2    \n\t"\

  163.     "1:                           \n\t"\

  164.     "movups   (%3, %0), %%xmm1    \n\t"\

  165.     "movaps     %%xmm1, %%xmm3    \n\t"\

  166.     "mulps    (%4, %0), %%xmm1    \n\t"\

  167.     "mulps    (%5, %0), %%xmm3    \n\t"\

  168.     "addps      %%xmm1, %%xmm0    \n\t"\

  169.     "addps      %%xmm3, %%xmm2    \n\t"\

  170.     "add           $16, %0        \n\t"\

  171.     " js 1b                       \n\t"\

  172.     "movhlps    %%xmm0, %%xmm1    \n\t"\

  173.     "movhlps    %%xmm2, %%xmm3    \n\t"\

  174.     "addps      %%xmm1, %%xmm0    \n\t"\

  175.     "addps      %%xmm3, %%xmm2    \n\t"\

  176.     "movss      %%xmm0, %%xmm1    \n\t"\

  177.     "movss      %%xmm2, %%xmm3    \n\t"\

  178.     "shufps $1, %%xmm0, %%xmm0    \n\t"\

  179.     "shufps $1, %%xmm2, %%xmm2    \n\t"\

  180.     "addps      %%xmm1, %%xmm0    \n\t"\

  181.     "addps      %%xmm3, %%xmm2    \n\t"\

  182.     "movss      %%xmm0, %1        \n\t"\

  183.     "movss      %%xmm2, %2        \n\t"\

  184.     : "+r" (len),\

  185.       "=m" (val),\

  186.       "=m" (v2)\

  187.     : "r" (((uint8_t*)(src+sample_index))-len),\

  188.       "r" (((uint8_t*)filter)-len),\

  189.       "r" (((uint8_t*)(filter+c->filter_alloc))-len)\

  190.     XMM_CLOBBERS_ONLY("%xmm0", "%xmm1", "%xmm2", "%xmm3")\

  191. );

  192. 

  193. #define COMMON_CORE_FLT_AVX \

  194.     x86_reg len= -4*c->filter_length;\

  195. __asm__ volatile(\

  196.     "vxorps     %%ymm0, %%ymm0, %%ymm0    \n\t"\

  197.     "1:                                   \n\t"\

  198.     "vmovups  (%1, %0), %%ymm1            \n\t"\

  199.     "vmulps   (%2, %0), %%ymm1, %%ymm1    \n\t"\

  200.     "vaddps     %%ymm1, %%ymm0, %%ymm0    \n\t"\

  201.     "add           $32, %0                \n\t"\

  202.     " js 1b                               \n\t"\

  203.     "vextractf128   $1, %%ymm0, %%xmm1    \n\t"\

  204.     "vaddps     %%xmm1, %%xmm0, %%xmm0    \n\t"\

  205.     "vmovhlps   %%xmm0, %%xmm1, %%xmm1    \n\t"\

  206.     "vaddps     %%xmm1, %%xmm0, %%xmm0    \n\t"\

  207.     "vshufps $1, %%xmm0, %%xmm0, %%xmm1   \n\t"\

  208.     "vaddss     %%xmm1, %%xmm0, %%xmm0    \n\t"\

  209.     "vmovss     %%xmm0, (%3)              \n\t"\

  210.     : "+r" (len)\

  211.     : "r" (((uint8_t*)(src+sample_index))-len),\

  212.       "r" (((uint8_t*)filter)-len),\

  213.       "r" (dst+dst_index)\

  214.     XMM_CLOBBERS_ONLY("%xmm0", "%xmm1")\

  215. );

  216. 

  217. #define LINEAR_CORE_FLT_AVX \

  218.     x86_reg len= -4*c->filter_length;\

  219. __asm__ volatile(\

  220.     "vxorps      %%ymm0, %%ymm0, %%ymm0   \n\t"\

  221.     "vxorps      %%ymm2, %%ymm2, %%ymm2   \n\t"\

  222.     "1:                                   \n\t"\

  223.     "vmovups   (%3, %0), %%ymm1           \n\t"\

  224.     "vmulps    (%5, %0), %%ymm1, %%ymm3   \n\t"\

  225.     "vmulps    (%4, %0), %%ymm1, %%ymm1   \n\t"\

  226.     "vaddps      %%ymm1, %%ymm0, %%ymm0   \n\t"\

  227.     "vaddps      %%ymm3, %%ymm2, %%ymm2   \n\t"\

  228.     "add            $32, %0               \n\t"\

  229.     " js 1b                               \n\t"\

  230.     "vextractf128    $1, %%ymm0, %%xmm1   \n\t"\

  231.     "vextractf128    $1, %%ymm2, %%xmm3   \n\t"\

  232.     "vaddps      %%xmm1, %%xmm0, %%xmm0   \n\t"\

  233.     "vaddps      %%xmm3, %%xmm2, %%xmm2   \n\t"\

  234.     "vmovhlps    %%xmm0, %%xmm1, %%xmm1   \n\t"\

  235.     "vmovhlps    %%xmm2, %%xmm3, %%xmm3   \n\t"\

  236.     "vaddps      %%xmm1, %%xmm0, %%xmm0   \n\t"\

  237.     "vaddps      %%xmm3, %%xmm2, %%xmm2   \n\t"\

  238.     "vshufps $1, %%xmm0, %%xmm0, %%xmm1   \n\t"\

  239.     "vshufps $1, %%xmm2, %%xmm2, %%xmm3   \n\t"\

  240.     "vaddss      %%xmm1, %%xmm0, %%xmm0   \n\t"\

  241.     "vaddss      %%xmm3, %%xmm2, %%xmm2   \n\t"\

  242.     "vmovss      %%xmm0, %1               \n\t"\

  243.     "vmovss      %%xmm2, %2               \n\t"\

  244.     : "+r" (len),\

  245.       "=m" (val),\

  246.       "=m" (v2)\

  247.     : "r" (((uint8_t*)(src+sample_index))-len),\

  248.       "r" (((uint8_t*)filter)-len),\

  249.       "r" (((uint8_t*)(filter+c->filter_alloc))-len)\

  250.     XMM_CLOBBERS_ONLY("%xmm0", "%xmm1", "%xmm2", "%xmm3")\

  251. );

  252. 

  253. #define COMMON_CORE_DBL_SSE2 \

  254.     x86_reg len= -8*c->filter_length;\

  255. __asm__ volatile(\

  256.     "xorpd     %%xmm0, %%xmm0     \n\t"\

  257.     "1:                           \n\t"\

  258.     "movupd  (%1, %0), %%xmm1     \n\t"\

  259.     "mulpd   (%2, %0), %%xmm1     \n\t"\

  260.     "addpd     %%xmm1, %%xmm0     \n\t"\

  261.     "add       $16, %0            \n\t"\

  262.     " js 1b                       \n\t"\

  263.     "movhlps   %%xmm0, %%xmm1     \n\t"\

  264.     "addpd     %%xmm1, %%xmm0     \n\t"\

  265.     "movsd     %%xmm0, (%3)       \n\t"\

  266.     : "+r" (len)\

  267.     : "r" (((uint8_t*)(src+sample_index))-len),\

  268.       "r" (((uint8_t*)filter)-len),\

  269.       "r" (dst+dst_index)\

  270.     XMM_CLOBBERS_ONLY("%xmm0", "%xmm1")\

  271. );

  272. 

  273. #define LINEAR_CORE_DBL_SSE2 \

  274.     x86_reg len= -8*c->filter_length;\

  275. __asm__ volatile(\

  276.     "xorpd      %%xmm0, %%xmm0    \n\t"\

  277.     "xorpd      %%xmm2, %%xmm2    \n\t"\

  278.     "1:                           \n\t"\

  279.     "movupd   (%3, %0), %%xmm1    \n\t"\

  280.     "movapd     %%xmm1, %%xmm3    \n\t"\

  281.     "mulpd    (%4, %0), %%xmm1    \n\t"\

  282.     "mulpd    (%5, %0), %%xmm3    \n\t"\

  283.     "addpd      %%xmm1, %%xmm0    \n\t"\

  284.     "addpd      %%xmm3, %%xmm2    \n\t"\

  285.     "add           $16, %0        \n\t"\

  286.     " js 1b                       \n\t"\

  287.     "movhlps    %%xmm0, %%xmm1    \n\t"\

  288.     "movhlps    %%xmm2, %%xmm3    \n\t"\

  289.     "addpd      %%xmm1, %%xmm0    \n\t"\

  290.     "addpd      %%xmm3, %%xmm2    \n\t"\

  291.     "movsd      %%xmm0, %1        \n\t"\

  292.     "movsd      %%xmm2, %2        \n\t"\

  293.     : "+r" (len),\

  294.       "=m" (val),\

  295.       "=m" (v2)\

  296.     : "r" (((uint8_t*)(src+sample_index))-len),\

  297.       "r" (((uint8_t*)filter)-len),\

  298.       "r" (((uint8_t*)(filter+c->filter_alloc))-len)\

  299.     XMM_CLOBBERS_ONLY("%xmm0", "%xmm1", "%xmm2", "%xmm3")\

  300. );