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.
53036 Commits
32 Branches
283MB
Tree: eda9d97b7a
Branches Tags
${ item.name }
Create branch ${ searchTerm }
from 'eda9d97b7a'
${ noResults }
FFmpeg/libavcodec/x86/hpeldsp_init.c

269 lines
11KB
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. #define avg_pixels8_mmx         ff_avg_pixels8_mmx

  78. #define avg_pixels8_x2_mmx      ff_avg_pixels8_x2_mmx

  79. #define avg_pixels16_mmx        ff_avg_pixels16_mmx

  80. #define avg_pixels8_xy2_mmx     ff_avg_pixels8_xy2_mmx

  81. #define avg_pixels16_xy2_mmx    ff_avg_pixels16_xy2_mmx

  82. #define put_pixels8_mmx         ff_put_pixels8_mmx

  83. #define put_pixels16_mmx        ff_put_pixels16_mmx

  84. #define put_pixels8_xy2_mmx     ff_put_pixels8_xy2_mmx

  85. #define put_pixels16_xy2_mmx    ff_put_pixels16_xy2_mmx

  86. #define avg_no_rnd_pixels16_mmx ff_avg_pixels16_mmx

  87. #define put_no_rnd_pixels8_mmx  ff_put_pixels8_mmx

  88. #define put_no_rnd_pixels16_mmx ff_put_pixels16_mmx

  89. 

  90. #if HAVE_INLINE_ASM

  91. 

  92. /***********************************/

  93. /* MMX no rounding */

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

  95. #define SET_RND  MOVQ_WONE

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

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

  98. #define STATIC static

  99. 

  100. #include "rnd_template.c"

  101. #include "hpeldsp_rnd_template.c"

  102. 

  103. #undef DEF

  104. #undef SET_RND

  105. #undef PAVGBP

  106. #undef PAVGB

  107. #undef STATIC

  108. 

  109. PIXELS16(static, avg_no_rnd, , _y2, _mmx)

  110. PIXELS16(static, put_no_rnd, , _y2, _mmx)

  111. 

  112. PIXELS16(static, avg_no_rnd, , _xy2, _mmx)

  113. PIXELS16(static, put_no_rnd, , _xy2, _mmx)

  114. 

  115. /***********************************/

  116. /* MMX rounding */

  117. 

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

  119. #define SET_RND  MOVQ_WTWO

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

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

  122. 

  123. #include "hpeldsp_rnd_template.c"

  124. 

  125. #undef DEF

  126. #undef SET_RND

  127. #undef PAVGBP

  128. #undef PAVGB

  129. 

  130. PIXELS16(static, avg, , _y2, _mmx)

  131. PIXELS16(static, put, , _y2, _mmx)

  132. 

  133. #endif /* HAVE_INLINE_ASM */

  134. 

  135. 

  136. #if HAVE_YASM

  137. 

  138. #define HPELDSP_AVG_PIXELS16(CPUEXT)                \

  139.     PIXELS16(static, put_no_rnd, ff_,  _x2, CPUEXT) \

  140.     PIXELS16(static, put,        ff_,  _y2, CPUEXT) \

  141.     PIXELS16(static, put_no_rnd, ff_,  _y2, CPUEXT) \

  142.     PIXELS16(static, avg,        ff_,     , CPUEXT) \

  143.     PIXELS16(static, avg,        ff_,  _x2, CPUEXT) \

  144.     PIXELS16(static, avg,        ff_,  _y2, CPUEXT) \

  145.     PIXELS16(static, avg,        ff_, _xy2, CPUEXT)

  146. 

  147. HPELDSP_AVG_PIXELS16(_3dnow)

  148. HPELDSP_AVG_PIXELS16(_mmxext)

  149. 

  150. #endif /* HAVE_YASM */

  151. 

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

  153.     do {                                                                        \

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

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

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

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

  158.     } while (0)

  159. 

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

  161. {

  162. #if HAVE_MMX_INLINE

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

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

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

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

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

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

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

  170. #endif /* HAVE_MMX_INLINE */

  171. }

  172. 

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

  174. {

  175. #if HAVE_MMXEXT_EXTERNAL

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

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

  178. 

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

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

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

  182. 

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

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

  185. 

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

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

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

  189. 

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

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

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

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

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

  195. 

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

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

  198.     }

  199. 

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

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

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

  203.     }

  204. #endif /* HAVE_MMXEXT_EXTERNAL */

  205. }

  206. 

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

  208. {

  209. #if HAVE_AMD3DNOW_EXTERNAL

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

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

  212. 

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

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

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

  216. 

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

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

  219. 

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

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

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

  223. 

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

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

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

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

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

  229. 

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

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

  232.     }

  233. 

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

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

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

  237.     }

  238. #endif /* HAVE_AMD3DNOW_EXTERNAL */

  239. }

  240. 

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

  242. {

  243. #if HAVE_SSE2_EXTERNAL

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

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

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

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

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

  249.     }

  250. #endif /* HAVE_SSE2_EXTERNAL */

  251. }

  252. 

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

  254. {

  255.     int mm_flags = av_get_cpu_flags();

  256. 

  257.     if (HAVE_MMX && mm_flags & AV_CPU_FLAG_MMX)

  258.         hpeldsp_init_mmx(c, flags, mm_flags);

  259. 

  260.     if (mm_flags & AV_CPU_FLAG_MMXEXT)

  261.         hpeldsp_init_mmxext(c, flags, mm_flags);

  262. 

  263.     if (mm_flags & AV_CPU_FLAG_3DNOW)

  264.         hpeldsp_init_3dnow(c, flags, mm_flags);

  265. 

  266.     if (mm_flags & AV_CPU_FLAG_SSE2)

  267.         hpeldsp_init_sse2(c, flags, mm_flags);

  268. }