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.
52638 Commits
32 Branches
552MB
Tree: dbcf7e9ef7
Branches Tags
${ item.name }
Create branch ${ searchTerm }
from 'dbcf7e9ef7'
${ noResults }
FFmpeg/libavcodec/x86/hpeldsp_init.c

321 lines
12KB
Raw Blame History 

  1. /*

  2.  * MMX optimized DSP utils

  3.  * Copyright (c) 2000, 2001 Fabrice Bellard

  4.  * Copyright (c) 2002-2004 Michael Niedermayer <michaelni@gmx.at>

  5.  *

  6.  * This file is part of FFmpeg.

  7.  *

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

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

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

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

  12.  *

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

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

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

  16.  * Lesser General Public License for more details.

  17.  *

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

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

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

  21.  *

  22.  * MMX optimization by Nick Kurshev <nickols_k@mail.ru>

  23.  */

  24. 

  25. #include "libavutil/cpu.h"

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

  27. #include "libavcodec/hpeldsp.h"

  28. #include "dsputil_mmx.h"

  29. 

  30. void ff_put_pixels8_x2_mmxext(uint8_t *block, const uint8_t *pixels,

  31.                               ptrdiff_t line_size, int h);

  32. void ff_put_pixels8_x2_3dnow(uint8_t *block, const uint8_t *pixels,

  33.                              ptrdiff_t line_size, int h);

  34. void ff_put_pixels16_x2_mmxext(uint8_t *block, const uint8_t *pixels,

  35.                                ptrdiff_t line_size, int h);

  36. void ff_put_pixels16_x2_3dnow(uint8_t *block, const uint8_t *pixels,

  37.                               ptrdiff_t line_size, int h);

  38. void ff_put_no_rnd_pixels8_x2_mmxext(uint8_t *block, const uint8_t *pixels,

  39.                                      ptrdiff_t line_size, int h);

  40. void ff_put_no_rnd_pixels8_x2_3dnow(uint8_t *block, const uint8_t *pixels,

  41.                                     ptrdiff_t line_size, int h);

  42. void ff_put_no_rnd_pixels8_x2_exact_mmxext(uint8_t *block,

  43.                                            const uint8_t *pixels,

  44.                                            ptrdiff_t line_size, int h);

  45. void ff_put_no_rnd_pixels8_x2_exact_3dnow(uint8_t *block,

  46.                                           const uint8_t *pixels,

  47.                                           ptrdiff_t line_size, int h);

  48. void ff_put_pixels8_y2_mmxext(uint8_t *block, const uint8_t *pixels,

  49.                               ptrdiff_t line_size, int h);

  50. void ff_put_pixels8_y2_3dnow(uint8_t *block, const uint8_t *pixels,

  51.                              ptrdiff_t line_size, int h);

  52. void ff_put_no_rnd_pixels8_y2_mmxext(uint8_t *block, const uint8_t *pixels,

  53.                                      ptrdiff_t line_size, int h);

  54. void ff_put_no_rnd_pixels8_y2_3dnow(uint8_t *block, const uint8_t *pixels,

  55.                                     ptrdiff_t line_size, int h);

  56. void ff_put_no_rnd_pixels8_y2_exact_mmxext(uint8_t *block,

  57.                                            const uint8_t *pixels,

  58.                                            ptrdiff_t line_size, int h);

  59. void ff_put_no_rnd_pixels8_y2_exact_3dnow(uint8_t *block,

  60.                                           const uint8_t *pixels,

  61.                                           ptrdiff_t line_size, int h);

  62. void ff_avg_pixels8_3dnow(uint8_t *block, const uint8_t *pixels,

  63.                           ptrdiff_t line_size, int h);

  64. void ff_avg_pixels8_x2_mmxext(uint8_t *block, const uint8_t *pixels,

  65.                               ptrdiff_t line_size, int h);

  66. void ff_avg_pixels8_x2_3dnow(uint8_t *block, const uint8_t *pixels,

  67.                              ptrdiff_t line_size, int h);

  68. void ff_avg_pixels8_y2_mmxext(uint8_t *block, const uint8_t *pixels,

  69.                               ptrdiff_t line_size, int h);

  70. void ff_avg_pixels8_y2_3dnow(uint8_t *block, const uint8_t *pixels,

  71.                              ptrdiff_t line_size, int h);

  72. void ff_avg_pixels8_xy2_mmxext(uint8_t *block, const uint8_t *pixels,

  73.                                ptrdiff_t line_size, int h);

  74. void ff_avg_pixels8_xy2_3dnow(uint8_t *block, const uint8_t *pixels,

  75.                               ptrdiff_t line_size, int h);

  76. 

  77. 

  78. #if HAVE_INLINE_ASM

  79. 

  80. /***********************************/

  81. /* MMX no rounding */

  82. #define NO_RND 1

  83. #define DEF(x, y) x ## _no_rnd_ ## y ## _mmx

  84. #define SET_RND  MOVQ_WONE

  85. #define PAVGBP(a, b, c, d, e, f)        PAVGBP_MMX_NO_RND(a, b, c, d, e, f)

  86. #define PAVGB(a, b, c, e)               PAVGB_MMX_NO_RND(a, b, c, e)

  87. #define OP_AVG(a, b, c, e)              PAVGB_MMX(a, b, c, e)

  88. 

  89. #include "hpeldsp_rnd_template.c"

  90. 

  91. #undef DEF

  92. #undef SET_RND

  93. #undef PAVGBP

  94. #undef PAVGB

  95. #undef NO_RND

  96. /***********************************/

  97. /* MMX rounding */

  98. 

  99. #define DEF(x, y) x ## _ ## y ## _mmx

  100. #define SET_RND  MOVQ_WTWO

  101. #define PAVGBP(a, b, c, d, e, f)        PAVGBP_MMX(a, b, c, d, e, f)

  102. #define PAVGB(a, b, c, e)               PAVGB_MMX(a, b, c, e)

  103. 

  104. #include "hpeldsp_rnd_template.c"

  105. 

  106. #undef DEF

  107. #undef SET_RND

  108. #undef PAVGBP

  109. #undef PAVGB

  110. #undef OP_AVG

  111. 

  112. #endif /* HAVE_INLINE_ASM */

  113. 

  114. 

  115. #if HAVE_YASM

  116. /***********************************/

  117. /* 3Dnow specific */

  118. 

  119. #define DEF(x) x ## _3dnow

  120. 

  121. #include "hpeldsp_avg_template.c"

  122. 

  123. #undef DEF

  124. 

  125. /***********************************/

  126. /* MMXEXT specific */

  127. 

  128. #define DEF(x) x ## _mmxext

  129. 

  130. #include "hpeldsp_avg_template.c"

  131. 

  132. #undef DEF

  133. 

  134. #endif /* HAVE_YASM */

  135. 

  136. 

  137. #if HAVE_INLINE_ASM

  138. #define put_no_rnd_pixels16_mmx put_pixels16_mmx

  139. #define put_no_rnd_pixels8_mmx put_pixels8_mmx

  140. 

  141. static void put_pixels8_mmx(uint8_t *block, const uint8_t *pixels,

  142.                             ptrdiff_t line_size, int h)

  143. {

  144.     __asm__ volatile (

  145.         "lea   (%3, %3), %%"REG_a"      \n\t"

  146.         ".p2align     3                 \n\t"

  147.         "1:                             \n\t"

  148.         "movq  (%1    ), %%mm0          \n\t"

  149.         "movq  (%1, %3), %%mm1          \n\t"

  150.         "movq     %%mm0, (%2)           \n\t"

  151.         "movq     %%mm1, (%2, %3)       \n\t"

  152.         "add  %%"REG_a", %1             \n\t"

  153.         "add  %%"REG_a", %2             \n\t"

  154.         "movq  (%1    ), %%mm0          \n\t"

  155.         "movq  (%1, %3), %%mm1          \n\t"

  156.         "movq     %%mm0, (%2)           \n\t"

  157.         "movq     %%mm1, (%2, %3)       \n\t"

  158.         "add  %%"REG_a", %1             \n\t"

  159.         "add  %%"REG_a", %2             \n\t"

  160.         "subl        $4, %0             \n\t"

  161.         "jnz         1b                 \n\t"

  162.         : "+g"(h), "+r"(pixels),  "+r"(block)

  163.         : "r"((x86_reg)line_size)

  164.         : "%"REG_a, "memory"

  165.         );

  166. }

  167. 

  168. static void put_pixels16_mmx(uint8_t *block, const uint8_t *pixels,

  169.                              ptrdiff_t line_size, int h)

  170. {

  171.     __asm__ volatile (

  172.         "lea   (%3, %3), %%"REG_a"      \n\t"

  173.         ".p2align     3                 \n\t"

  174.         "1:                             \n\t"

  175.         "movq  (%1    ), %%mm0          \n\t"

  176.         "movq 8(%1    ), %%mm4          \n\t"

  177.         "movq  (%1, %3), %%mm1          \n\t"

  178.         "movq 8(%1, %3), %%mm5          \n\t"

  179.         "movq     %%mm0,  (%2)          \n\t"

  180.         "movq     %%mm4, 8(%2)          \n\t"

  181.         "movq     %%mm1,  (%2, %3)      \n\t"

  182.         "movq     %%mm5, 8(%2, %3)      \n\t"

  183.         "add  %%"REG_a", %1             \n\t"

  184.         "add  %%"REG_a", %2             \n\t"

  185.         "movq  (%1    ), %%mm0          \n\t"

  186.         "movq 8(%1    ), %%mm4          \n\t"

  187.         "movq  (%1, %3), %%mm1          \n\t"

  188.         "movq 8(%1, %3), %%mm5          \n\t"

  189.         "movq     %%mm0,  (%2)          \n\t"

  190.         "movq     %%mm4, 8(%2)          \n\t"

  191.         "movq     %%mm1,  (%2, %3)      \n\t"

  192.         "movq     %%mm5, 8(%2, %3)      \n\t"

  193.         "add  %%"REG_a", %1             \n\t"

  194.         "add  %%"REG_a", %2             \n\t"

  195.         "subl        $4, %0             \n\t"

  196.         "jnz         1b                 \n\t"

  197.         : "+g"(h), "+r"(pixels),  "+r"(block)

  198.         : "r"((x86_reg)line_size)

  199.         : "%"REG_a, "memory"

  200.         );

  201. }

  202. #endif /* HAVE_INLINE_ASM */

  203. 

  204. #define SET_HPEL_FUNCS(PFX, IDX, SIZE, CPU)                                     \

  205.     do {                                                                        \

  206.         c->PFX ## _pixels_tab IDX [0] = PFX ## _pixels ## SIZE ## _     ## CPU; \

  207.         c->PFX ## _pixels_tab IDX [1] = PFX ## _pixels ## SIZE ## _x2_  ## CPU; \

  208.         c->PFX ## _pixels_tab IDX [2] = PFX ## _pixels ## SIZE ## _y2_  ## CPU; \

  209.         c->PFX ## _pixels_tab IDX [3] = PFX ## _pixels ## SIZE ## _xy2_ ## CPU; \

  210.     } while (0)

  211. 

  212. static void hpeldsp_init_mmx(HpelDSPContext *c, int flags, int mm_flags)

  213. {

  214. #if HAVE_INLINE_ASM

  215.     SET_HPEL_FUNCS(put,        [0], 16, mmx);

  216.     SET_HPEL_FUNCS(put_no_rnd, [0], 16, mmx);

  217.     SET_HPEL_FUNCS(avg,        [0], 16, mmx);

  218.     SET_HPEL_FUNCS(avg_no_rnd,    , 16, mmx);

  219.     SET_HPEL_FUNCS(put,        [1],  8, mmx);

  220.     SET_HPEL_FUNCS(put_no_rnd, [1],  8, mmx);

  221.     SET_HPEL_FUNCS(avg,        [1],  8, mmx);

  222. #endif /* HAVE_INLINE_ASM */

  223. }

  224. 

  225. static void hpeldsp_init_mmxext(HpelDSPContext *c, int flags, int mm_flags)

  226. {

  227. #if HAVE_YASM

  228.     c->put_pixels_tab[0][1] = ff_put_pixels16_x2_mmxext;

  229.     c->put_pixels_tab[0][2] = put_pixels16_y2_mmxext;

  230. 

  231.     c->avg_pixels_tab[0][0] = avg_pixels16_mmxext;

  232.     c->avg_pixels_tab[0][1] = avg_pixels16_x2_mmxext;

  233.     c->avg_pixels_tab[0][2] = avg_pixels16_y2_mmxext;

  234. 

  235.     c->put_pixels_tab[1][1] = ff_put_pixels8_x2_mmxext;

  236.     c->put_pixels_tab[1][2] = ff_put_pixels8_y2_mmxext;

  237. 

  238.     c->avg_pixels_tab[1][0] = ff_avg_pixels8_mmxext;

  239.     c->avg_pixels_tab[1][1] = ff_avg_pixels8_x2_mmxext;

  240.     c->avg_pixels_tab[1][2] = ff_avg_pixels8_y2_mmxext;

  241. 

  242.     if (!(flags & CODEC_FLAG_BITEXACT)) {

  243.         c->put_no_rnd_pixels_tab[0][1] = put_no_rnd_pixels16_x2_mmxext;

  244.         c->put_no_rnd_pixels_tab[0][2] = put_no_rnd_pixels16_y2_mmxext;

  245.         c->put_no_rnd_pixels_tab[1][1] = ff_put_no_rnd_pixels8_x2_mmxext;

  246.         c->put_no_rnd_pixels_tab[1][2] = ff_put_no_rnd_pixels8_y2_mmxext;

  247. 

  248.         c->avg_pixels_tab[0][3] = avg_pixels16_xy2_mmxext;

  249.         c->avg_pixels_tab[1][3] = ff_avg_pixels8_xy2_mmxext;

  250.     }

  251. 

  252.     if (flags & CODEC_FLAG_BITEXACT && CONFIG_VP3_DECODER) {

  253.         c->put_no_rnd_pixels_tab[1][1] = ff_put_no_rnd_pixels8_x2_exact_mmxext;

  254.         c->put_no_rnd_pixels_tab[1][2] = ff_put_no_rnd_pixels8_y2_exact_mmxext;

  255.     }

  256. #endif /* HAVE_YASM */

  257. }

  258. 

  259. static void hpeldsp_init_3dnow(HpelDSPContext *c, int flags, int mm_flags)

  260. {

  261. #if HAVE_YASM

  262.     c->put_pixels_tab[0][1] = ff_put_pixels16_x2_3dnow;

  263.     c->put_pixels_tab[0][2] = put_pixels16_y2_3dnow;

  264. 

  265.     c->avg_pixels_tab[0][0] = avg_pixels16_3dnow;

  266.     c->avg_pixels_tab[0][1] = avg_pixels16_x2_3dnow;

  267.     c->avg_pixels_tab[0][2] = avg_pixels16_y2_3dnow;

  268. 

  269.     c->put_pixels_tab[1][1] = ff_put_pixels8_x2_3dnow;

  270.     c->put_pixels_tab[1][2] = ff_put_pixels8_y2_3dnow;

  271. 

  272.     c->avg_pixels_tab[1][0] = ff_avg_pixels8_3dnow;

  273.     c->avg_pixels_tab[1][1] = ff_avg_pixels8_x2_3dnow;

  274.     c->avg_pixels_tab[1][2] = ff_avg_pixels8_y2_3dnow;

  275. 

  276.     if (!(flags & CODEC_FLAG_BITEXACT)){

  277.         c->put_no_rnd_pixels_tab[0][1] = put_no_rnd_pixels16_x2_3dnow;

  278.         c->put_no_rnd_pixels_tab[0][2] = put_no_rnd_pixels16_y2_3dnow;

  279.         c->put_no_rnd_pixels_tab[1][1] = ff_put_no_rnd_pixels8_x2_3dnow;

  280.         c->put_no_rnd_pixels_tab[1][2] = ff_put_no_rnd_pixels8_y2_3dnow;

  281. 

  282.         c->avg_pixels_tab[0][3] = avg_pixels16_xy2_3dnow;

  283.         c->avg_pixels_tab[1][3] = ff_avg_pixels8_xy2_3dnow;

  284.     }

  285. 

  286.     if (flags & CODEC_FLAG_BITEXACT && CONFIG_VP3_DECODER) {

  287.         c->put_no_rnd_pixels_tab[1][1] = ff_put_no_rnd_pixels8_x2_exact_3dnow;

  288.         c->put_no_rnd_pixels_tab[1][2] = ff_put_no_rnd_pixels8_y2_exact_3dnow;

  289.     }

  290. #endif /* HAVE_YASM */

  291. }

  292. 

  293. static void hpeldsp_init_sse2(HpelDSPContext *c, int flags, int mm_flags)

  294. {

  295. #if HAVE_YASM

  296.     if (!(mm_flags & AV_CPU_FLAG_SSE2SLOW)) {

  297.         // these functions are slower than mmx on AMD, but faster on Intel

  298.         c->put_pixels_tab[0][0]        = ff_put_pixels16_sse2;

  299.         c->put_no_rnd_pixels_tab[0][0] = ff_put_pixels16_sse2;

  300.         c->avg_pixels_tab[0][0]        = ff_avg_pixels16_sse2;

  301.     }

  302. #endif /* HAVE_YASM */

  303. }

  304. 

  305. void ff_hpeldsp_init_x86(HpelDSPContext *c, int flags)

  306. {

  307.     int mm_flags = av_get_cpu_flags();

  308. 

  309.     if (HAVE_MMX && mm_flags & AV_CPU_FLAG_MMX)

  310.         hpeldsp_init_mmx(c, flags, mm_flags);

  311. 

  312.     if (mm_flags & AV_CPU_FLAG_MMXEXT)

  313.         hpeldsp_init_mmxext(c, flags, mm_flags);

  314. 

  315.     if (mm_flags & AV_CPU_FLAG_3DNOW)

  316.         hpeldsp_init_3dnow(c, flags, mm_flags);

  317. 

  318.     if (mm_flags & AV_CPU_FLAG_SSE2)

  319.         hpeldsp_init_sse2(c, flags, mm_flags);

  320. }