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.
43415 Commits
32 Branches
283MB
Tree: 82b7525173
Branches Tags
${ item.name }
Create branch ${ searchTerm }
from '82b7525173'
${ noResults }
FFmpeg/libavcodec/x86/hpeldsp_init.c

282 lines
11KB
Raw Blame History 

  1. /*

  2.  * SIMD-optimized halfpel functions

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

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

  5.  *

  6.  * This file is part of Libav.

  7.  *

  8.  * Libav 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.  * Libav 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 Libav; 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/attributes.h"

  26. #include "libavutil/cpu.h"

  27. #include "libavutil/x86/cpu.h"

  28. #include "libavcodec/avcodec.h"

  29. #include "libavcodec/hpeldsp.h"

  30. #include "libavcodec/pixels.h"

  31. #include "fpel.h"

  32. #include "hpeldsp.h"

  33. 

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

  35.                               ptrdiff_t line_size, int h);

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

  37.                              ptrdiff_t line_size, int h);

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

  39.                                ptrdiff_t line_size, int h);

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

  41.                               ptrdiff_t line_size, int h);

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

  43.                                      ptrdiff_t line_size, int h);

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

  45.                                     ptrdiff_t line_size, int h);

  46. void ff_put_no_rnd_pixels8_x2_exact_mmxext(uint8_t *block,

  47.                                            const uint8_t *pixels,

  48.                                            ptrdiff_t line_size, int h);

  49. void ff_put_no_rnd_pixels8_x2_exact_3dnow(uint8_t *block,

  50.                                           const uint8_t *pixels,

  51.                                           ptrdiff_t line_size, int h);

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

  53.                               ptrdiff_t line_size, int h);

  54. void ff_put_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_mmxext(uint8_t *block, const uint8_t *pixels,

  57.                                      ptrdiff_t line_size, int h);

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

  59.                                     ptrdiff_t line_size, int h);

  60. void ff_put_no_rnd_pixels8_y2_exact_mmxext(uint8_t *block,

  61.                                            const uint8_t *pixels,

  62.                                            ptrdiff_t line_size, int h);

  63. void ff_put_no_rnd_pixels8_y2_exact_3dnow(uint8_t *block,

  64.                                           const uint8_t *pixels,

  65.                                           ptrdiff_t line_size, int h);

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

  67.                           ptrdiff_t line_size, int h);

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

  69.                               ptrdiff_t line_size, int h);

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

  71.                              ptrdiff_t line_size, int h);

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

  73.                               ptrdiff_t line_size, int h);

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

  75.                              ptrdiff_t line_size, int h);

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

  77.                                ptrdiff_t line_size, int h);

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

  79.                               ptrdiff_t line_size, int h);

  80. 

  81. #define avg_pixels8_mmx         ff_avg_pixels8_mmx

  82. #define avg_pixels8_x2_mmx      ff_avg_pixels8_x2_mmx

  83. #define avg_pixels16_mmx        ff_avg_pixels16_mmx

  84. #define avg_pixels8_xy2_mmx     ff_avg_pixels8_xy2_mmx

  85. #define avg_pixels16_xy2_mmx    ff_avg_pixels16_xy2_mmx

  86. #define put_pixels8_mmx         ff_put_pixels8_mmx

  87. #define put_pixels16_mmx        ff_put_pixels16_mmx

  88. #define put_pixels8_xy2_mmx     ff_put_pixels8_xy2_mmx

  89. #define put_pixels16_xy2_mmx    ff_put_pixels16_xy2_mmx

  90. #define avg_no_rnd_pixels16_mmx ff_avg_pixels16_mmx

  91. #define put_no_rnd_pixels8_mmx  ff_put_pixels8_mmx

  92. #define put_no_rnd_pixels16_mmx ff_put_pixels16_mmx

  93. 

  94. #if HAVE_INLINE_ASM

  95. 

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

  97. /* MMX no rounding */

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

  99. #define SET_RND  MOVQ_WONE

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

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

  102. #define STATIC static

  103. 

  104. #include "rnd_template.c"

  105. #include "hpeldsp_rnd_template.c"

  106. 

  107. #undef DEF

  108. #undef SET_RND

  109. #undef PAVGBP

  110. #undef PAVGB

  111. #undef STATIC

  112. 

  113. CALL_2X_PIXELS(avg_no_rnd_pixels16_y2_mmx, avg_no_rnd_pixels8_y2_mmx, 8)

  114. CALL_2X_PIXELS(put_no_rnd_pixels16_y2_mmx, put_no_rnd_pixels8_y2_mmx, 8)

  115. 

  116. CALL_2X_PIXELS(avg_no_rnd_pixels16_xy2_mmx, avg_no_rnd_pixels8_xy2_mmx, 8)

  117. CALL_2X_PIXELS(put_no_rnd_pixels16_xy2_mmx, put_no_rnd_pixels8_xy2_mmx, 8)

  118. 

  119. /***********************************/

  120. /* MMX rounding */

  121. 

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

  123. #define SET_RND  MOVQ_WTWO

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

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

  126. 

  127. #include "hpeldsp_rnd_template.c"

  128. 

  129. #undef DEF

  130. #define DEF(x, y) ff_ ## x ## _ ## y ## _mmx

  131. #define STATIC

  132. 

  133. #include "rnd_template.c"

  134. 

  135. #undef DEF

  136. #undef SET_RND

  137. #undef PAVGBP

  138. #undef PAVGB

  139. 

  140. CALL_2X_PIXELS(avg_pixels16_y2_mmx, avg_pixels8_y2_mmx, 8)

  141. CALL_2X_PIXELS(put_pixels16_y2_mmx, put_pixels8_y2_mmx, 8)

  142. 

  143. CALL_2X_PIXELS_EXPORT(ff_avg_pixels16_xy2_mmx, ff_avg_pixels8_xy2_mmx, 8)

  144. CALL_2X_PIXELS_EXPORT(ff_put_pixels16_xy2_mmx, ff_put_pixels8_xy2_mmx, 8)

  145. 

  146. #endif /* HAVE_INLINE_ASM */

  147. 

  148. 

  149. #if HAVE_YASM

  150. 

  151. #define HPELDSP_AVG_PIXELS16(CPUEXT)                      \

  152.     CALL_2X_PIXELS(put_no_rnd_pixels16_x2 ## CPUEXT, ff_put_no_rnd_pixels8_x2 ## CPUEXT, 8) \

  153.     CALL_2X_PIXELS(put_pixels16_y2        ## CPUEXT, ff_put_pixels8_y2        ## CPUEXT, 8) \

  154.     CALL_2X_PIXELS(put_no_rnd_pixels16_y2 ## CPUEXT, ff_put_no_rnd_pixels8_y2 ## CPUEXT, 8) \

  155.     CALL_2X_PIXELS(avg_pixels16           ## CPUEXT, ff_avg_pixels8           ## CPUEXT, 8) \

  156.     CALL_2X_PIXELS(avg_pixels16_x2        ## CPUEXT, ff_avg_pixels8_x2        ## CPUEXT, 8) \

  157.     CALL_2X_PIXELS(avg_pixels16_y2        ## CPUEXT, ff_avg_pixels8_y2        ## CPUEXT, 8) \

  158.     CALL_2X_PIXELS(avg_pixels16_xy2       ## CPUEXT, ff_avg_pixels8_xy2       ## CPUEXT, 8)

  159. 

  160. HPELDSP_AVG_PIXELS16(_3dnow)

  161. HPELDSP_AVG_PIXELS16(_mmxext)

  162. 

  163. #endif /* HAVE_YASM */

  164. 

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

  166.     do {                                                                        \

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

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

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

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

  171.     } while (0)

  172. 

  173. static void hpeldsp_init_mmx(HpelDSPContext *c, int flags, int cpu_flags)

  174. {

  175. #if HAVE_MMX_INLINE

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

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

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

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

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

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

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

  183. #endif /* HAVE_MMX_INLINE */

  184. }

  185. 

  186. static void hpeldsp_init_mmxext(HpelDSPContext *c, int flags, int cpu_flags)

  187. {

  188. #if HAVE_MMXEXT_EXTERNAL

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

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

  191. 

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

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

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

  195. 

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

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

  198. 

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

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

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

  202. 

  203.     if (!(flags & AV_CODEC_FLAG_BITEXACT)) {

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

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

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

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

  208. 

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

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

  211.     }

  212. 

  213.     if (flags & AV_CODEC_FLAG_BITEXACT && CONFIG_VP3_DECODER) {

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

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

  216.     }

  217. #endif /* HAVE_MMXEXT_EXTERNAL */

  218. }

  219. 

  220. static void hpeldsp_init_3dnow(HpelDSPContext *c, int flags, int cpu_flags)

  221. {

  222. #if HAVE_AMD3DNOW_EXTERNAL

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

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

  225. 

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

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

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

  229. 

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

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

  232. 

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

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

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

  236. 

  237.     if (!(flags & AV_CODEC_FLAG_BITEXACT)){

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

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

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

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

  242. 

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

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

  245.     }

  246. 

  247.     if (flags & AV_CODEC_FLAG_BITEXACT && CONFIG_VP3_DECODER) {

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

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

  250.     }

  251. #endif /* HAVE_AMD3DNOW_EXTERNAL */

  252. }

  253. 

  254. static void hpeldsp_init_sse2(HpelDSPContext *c, int flags, int cpu_flags)

  255. {

  256. #if HAVE_SSE2_EXTERNAL

  257.     if (!(cpu_flags & AV_CPU_FLAG_SSE2SLOW)) {

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

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

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

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

  262.     }

  263. #endif /* HAVE_SSE2_EXTERNAL */

  264. }

  265. 

  266. av_cold void ff_hpeldsp_init_x86(HpelDSPContext *c, int flags)

  267. {

  268.     int cpu_flags = av_get_cpu_flags();

  269. 

  270.     if (INLINE_MMX(cpu_flags))

  271.         hpeldsp_init_mmx(c, flags, cpu_flags);

  272. 

  273.     if (EXTERNAL_AMD3DNOW(cpu_flags))

  274.         hpeldsp_init_3dnow(c, flags, cpu_flags);

  275. 

  276.     if (EXTERNAL_MMXEXT(cpu_flags))

  277.         hpeldsp_init_mmxext(c, flags, cpu_flags);

  278. 

  279.     if (EXTERNAL_SSE2(cpu_flags))

  280.         hpeldsp_init_sse2(c, flags, cpu_flags);

  281. }