Mercurial > cgi-bin > hgwebdir.cgi > PR > Applications > VSs > VSs__H264__App
diff libavcodec/x86/dsputil_mmx.c @ 2:897f711a7157
rearrange to work with autoconf
| author | Nina Engelhardt <nengel@mailbox.tu-berlin.de> |
|---|---|
| date | Tue, 25 Sep 2012 15:55:33 +0200 |
| parents | |
| children |
line diff
1.1 --- /dev/null Thu Jan 01 00:00:00 1970 +0000 1.2 +++ b/libavcodec/x86/dsputil_mmx.c Tue Sep 25 15:55:33 2012 +0200 1.3 @@ -0,0 +1,821 @@ 1.4 +/* 1.5 + * MMX optimized DSP utils 1.6 + * Copyright (c) 2000, 2001 Fabrice Bellard 1.7 + * Copyright (c) 2002-2004 Michael Niedermayer <michaelni@gmx.at> 1.8 + * 1.9 + * This file is part of FFmpeg. 1.10 + * 1.11 + * FFmpeg is free software; you can redistribute it and/or 1.12 + * modify it under the terms of the GNU Lesser General Public 1.13 + * License as published by the Free Software Foundation; either 1.14 + * version 2.1 of the License, or (at your option) any later version. 1.15 + * 1.16 + * FFmpeg is distributed in the hope that it will be useful, 1.17 + * but WITHOUT ANY WARRANTY; without even the implied warranty of 1.18 + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU 1.19 + * Lesser General Public License for more details. 1.20 + * 1.21 + * You should have received a copy of the GNU Lesser General Public 1.22 + * License along with FFmpeg; if not, write to the Free Software 1.23 + * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA 1.24 + * 1.25 + * MMX optimization by Nick Kurshev <nickols_k@mail.ru> 1.26 + */ 1.27 + 1.28 +#include "libavutil/x86_cpu.h" 1.29 +#include "libavutil/internal.h" 1.30 +#include "libavcodec/dsputil.h" 1.31 +#include "libavcodec/h264_dsp.h" 1.32 +#include "dsputil_mmx.h" 1.33 + 1.34 + 1.35 +//#undef NDEBUG 1.36 +//#include <assert.h> 1.37 + 1.38 +int mm_flags; /* multimedia extension flags */ 1.39 + 1.40 +/* pixel operations */ 1.41 +DECLARE_ALIGNED(8, const uint64_t, ff_bone) = 0x0101010101010101ULL; 1.42 +DECLARE_ALIGNED(8, const uint64_t, ff_wtwo) = 0x0002000200020002ULL; 1.43 + 1.44 +DECLARE_ALIGNED(16, const uint64_t, ff_pdw_80000000)[2] = 1.45 +{0x8000000080000000ULL, 0x8000000080000000ULL}; 1.46 + 1.47 +DECLARE_ALIGNED(8, const uint64_t, ff_pw_3 ) = 0x0003000300030003ULL; 1.48 +DECLARE_ALIGNED(8, const uint64_t, ff_pw_4 ) = 0x0004000400040004ULL; 1.49 +DECLARE_ALIGNED(16, const xmm_reg, ff_pw_5 ) = {0x0005000500050005ULL, 0x0005000500050005ULL}; 1.50 +DECLARE_ALIGNED(16, const xmm_reg, ff_pw_8 ) = {0x0008000800080008ULL, 0x0008000800080008ULL}; 1.51 +DECLARE_ALIGNED(8, const uint64_t, ff_pw_15 ) = 0x000F000F000F000FULL; 1.52 +DECLARE_ALIGNED(16, const xmm_reg, ff_pw_16 ) = {0x0010001000100010ULL, 0x0010001000100010ULL}; 1.53 +DECLARE_ALIGNED(8, const uint64_t, ff_pw_20 ) = 0x0014001400140014ULL; 1.54 +DECLARE_ALIGNED(16, const xmm_reg, ff_pw_28 ) = {0x001C001C001C001CULL, 0x001C001C001C001CULL}; 1.55 +DECLARE_ALIGNED(16, const xmm_reg, ff_pw_32 ) = {0x0020002000200020ULL, 0x0020002000200020ULL}; 1.56 +DECLARE_ALIGNED(8, const uint64_t, ff_pw_42 ) = 0x002A002A002A002AULL; 1.57 +DECLARE_ALIGNED(16, const xmm_reg, ff_pw_64 ) = {0x0040004000400040ULL, 0x0040004000400040ULL}; 1.58 +DECLARE_ALIGNED(8, const uint64_t, ff_pw_96 ) = 0x0060006000600060ULL; 1.59 +DECLARE_ALIGNED(8, const uint64_t, ff_pw_128) = 0x0080008000800080ULL; 1.60 +DECLARE_ALIGNED(8, const uint64_t, ff_pw_255) = 0x00ff00ff00ff00ffULL; 1.61 + 1.62 +DECLARE_ALIGNED(8, const uint64_t, ff_pb_1 ) = 0x0101010101010101ULL; 1.63 +DECLARE_ALIGNED(8, const uint64_t, ff_pb_3 ) = 0x0303030303030303ULL; 1.64 +DECLARE_ALIGNED(8, const uint64_t, ff_pb_7 ) = 0x0707070707070707ULL; 1.65 +DECLARE_ALIGNED(8, const uint64_t, ff_pb_1F ) = 0x1F1F1F1F1F1F1F1FULL; 1.66 +DECLARE_ALIGNED(8, const uint64_t, ff_pb_3F ) = 0x3F3F3F3F3F3F3F3FULL; 1.67 +DECLARE_ALIGNED(8, const uint64_t, ff_pb_81 ) = 0x8181818181818181ULL; 1.68 +DECLARE_ALIGNED(8, const uint64_t, ff_pb_A1 ) = 0xA1A1A1A1A1A1A1A1ULL; 1.69 +DECLARE_ALIGNED(8, const uint64_t, ff_pb_FC ) = 0xFCFCFCFCFCFCFCFCULL; 1.70 + 1.71 +DECLARE_ALIGNED(16, const double, ff_pd_1)[2] = { 1.0, 1.0 }; 1.72 +DECLARE_ALIGNED(16, const double, ff_pd_2)[2] = { 2.0, 2.0 }; 1.73 + 1.74 +#define ASMALIGN(ZEROBITS) ".align 1 << " #ZEROBITS "\n\t" 1.75 +#define JUMPALIGN() __asm__ volatile (ASMALIGN(3)::) 1.76 +#define MOVQ_ZERO(regd) __asm__ volatile ("pxor %%" #regd ", %%" #regd ::) 1.77 + 1.78 +#define MOVQ_BFE(regd) \ 1.79 + __asm__ volatile ( \ 1.80 + "pcmpeqd %%" #regd ", %%" #regd " \n\t"\ 1.81 + "paddb %%" #regd ", %%" #regd " \n\t" ::) 1.82 + 1.83 +#ifndef PIC 1.84 +#define MOVQ_BONE(regd) __asm__ volatile ("movq %0, %%" #regd " \n\t" ::"m"(ff_bone)) 1.85 +#define MOVQ_WTWO(regd) __asm__ volatile ("movq %0, %%" #regd " \n\t" ::"m"(ff_wtwo)) 1.86 +#else 1.87 +// for shared library it's better to use this way for accessing constants 1.88 +// pcmpeqd -> -1 1.89 +#define MOVQ_BONE(regd) \ 1.90 + __asm__ volatile ( \ 1.91 + "pcmpeqd %%" #regd ", %%" #regd " \n\t" \ 1.92 + "psrlw $15, %%" #regd " \n\t" \ 1.93 + "packuswb %%" #regd ", %%" #regd " \n\t" ::) 1.94 + 1.95 +#define MOVQ_WTWO(regd) \ 1.96 + __asm__ volatile ( \ 1.97 + "pcmpeqd %%" #regd ", %%" #regd " \n\t" \ 1.98 + "psrlw $15, %%" #regd " \n\t" \ 1.99 + "psllw $1, %%" #regd " \n\t"::) 1.100 + 1.101 +#endif 1.102 + 1.103 +// using regr as temporary and for the output result 1.104 +// first argument is unmodifed and second is trashed 1.105 +// regfe is supposed to contain 0xfefefefefefefefe 1.106 +#define PAVGB_MMX_NO_RND(rega, regb, regr, regfe) \ 1.107 + "movq " #rega ", " #regr " \n\t"\ 1.108 + "pand " #regb ", " #regr " \n\t"\ 1.109 + "pxor " #rega ", " #regb " \n\t"\ 1.110 + "pand " #regfe "," #regb " \n\t"\ 1.111 + "psrlq $1, " #regb " \n\t"\ 1.112 + "paddb " #regb ", " #regr " \n\t" 1.113 + 1.114 +#define PAVGB_MMX(rega, regb, regr, regfe) \ 1.115 + "movq " #rega ", " #regr " \n\t"\ 1.116 + "por " #regb ", " #regr " \n\t"\ 1.117 + "pxor " #rega ", " #regb " \n\t"\ 1.118 + "pand " #regfe "," #regb " \n\t"\ 1.119 + "psrlq $1, " #regb " \n\t"\ 1.120 + "psubb " #regb ", " #regr " \n\t" 1.121 + 1.122 +// mm6 is supposed to contain 0xfefefefefefefefe 1.123 +#define PAVGBP_MMX_NO_RND(rega, regb, regr, regc, regd, regp) \ 1.124 + "movq " #rega ", " #regr " \n\t"\ 1.125 + "movq " #regc ", " #regp " \n\t"\ 1.126 + "pand " #regb ", " #regr " \n\t"\ 1.127 + "pand " #regd ", " #regp " \n\t"\ 1.128 + "pxor " #rega ", " #regb " \n\t"\ 1.129 + "pxor " #regc ", " #regd " \n\t"\ 1.130 + "pand %%mm6, " #regb " \n\t"\ 1.131 + "pand %%mm6, " #regd " \n\t"\ 1.132 + "psrlq $1, " #regb " \n\t"\ 1.133 + "psrlq $1, " #regd " \n\t"\ 1.134 + "paddb " #regb ", " #regr " \n\t"\ 1.135 + "paddb " #regd ", " #regp " \n\t" 1.136 + 1.137 +#define PAVGBP_MMX(rega, regb, regr, regc, regd, regp) \ 1.138 + "movq " #rega ", " #regr " \n\t"\ 1.139 + "movq " #regc ", " #regp " \n\t"\ 1.140 + "por " #regb ", " #regr " \n\t"\ 1.141 + "por " #regd ", " #regp " \n\t"\ 1.142 + "pxor " #rega ", " #regb " \n\t"\ 1.143 + "pxor " #regc ", " #regd " \n\t"\ 1.144 + "pand %%mm6, " #regb " \n\t"\ 1.145 + "pand %%mm6, " #regd " \n\t"\ 1.146 + "psrlq $1, " #regd " \n\t"\ 1.147 + "psrlq $1, " #regb " \n\t"\ 1.148 + "psubb " #regb ", " #regr " \n\t"\ 1.149 + "psubb " #regd ", " #regp " \n\t" 1.150 + 1.151 +/***********************************/ 1.152 +/* MMX2 specific */ 1.153 + 1.154 +#define DEF(x) x ## _mmx2 1.155 + 1.156 +/* Introduced only in MMX2 set */ 1.157 +#define PAVGB "pavgb" 1.158 +#define OP_AVG PAVGB 1.159 + 1.160 +#include "dsputil_mmx_avg_template.c" 1.161 + 1.162 +#undef DEF 1.163 +#undef PAVGB 1.164 +#undef OP_AVG 1.165 + 1.166 +#define put_no_rnd_pixels16_mmx put_pixels16_mmx 1.167 +#define put_no_rnd_pixels8_mmx put_pixels8_mmx 1.168 +#define put_pixels16_mmx2 put_pixels16_mmx 1.169 +#define put_pixels8_mmx2 put_pixels8_mmx 1.170 +#define put_pixels4_mmx2 put_pixels4_mmx 1.171 +#define put_no_rnd_pixels16_mmx2 put_no_rnd_pixels16_mmx 1.172 +#define put_no_rnd_pixels8_mmx2 put_no_rnd_pixels8_mmx 1.173 +#define put_pixels16_3dnow put_pixels16_mmx 1.174 +#define put_pixels8_3dnow put_pixels8_mmx 1.175 +#define put_pixels4_3dnow put_pixels4_mmx 1.176 +#define put_no_rnd_pixels16_3dnow put_no_rnd_pixels16_mmx 1.177 +#define put_no_rnd_pixels8_3dnow put_no_rnd_pixels8_mmx 1.178 + 1.179 +/***********************************/ 1.180 +/* standard MMX */ 1.181 + 1.182 +void put_pixels_clamped_mmx(const DCTELEM *block, uint8_t *pixels, int line_size) 1.183 +{ 1.184 + const DCTELEM *p; 1.185 + uint8_t *pix; 1.186 + 1.187 + /* read the pixels */ 1.188 + p = block; 1.189 + pix = pixels; 1.190 + /* unrolled loop */ 1.191 + __asm__ volatile( 1.192 + "movq %3, %%mm0 \n\t" 1.193 + "movq 8%3, %%mm1 \n\t" 1.194 + "movq 16%3, %%mm2 \n\t" 1.195 + "movq 24%3, %%mm3 \n\t" 1.196 + "movq 32%3, %%mm4 \n\t" 1.197 + "movq 40%3, %%mm5 \n\t" 1.198 + "movq 48%3, %%mm6 \n\t" 1.199 + "movq 56%3, %%mm7 \n\t" 1.200 + "packuswb %%mm1, %%mm0 \n\t" 1.201 + "packuswb %%mm3, %%mm2 \n\t" 1.202 + "packuswb %%mm5, %%mm4 \n\t" 1.203 + "packuswb %%mm7, %%mm6 \n\t" 1.204 + "movq %%mm0, (%0) \n\t" 1.205 + "movq %%mm2, (%0, %1) \n\t" 1.206 + "movq %%mm4, (%0, %1, 2) \n\t" 1.207 + "movq %%mm6, (%0, %2) \n\t" 1.208 + ::"r" (pix), "r" ((x86_reg)line_size), "r" ((x86_reg)line_size*3), "m"(*p) 1.209 + :"memory"); 1.210 + pix += line_size*4; 1.211 + p += 32; 1.212 + 1.213 + // if here would be an exact copy of the code above 1.214 + // compiler would generate some very strange code 1.215 + // thus using "r" 1.216 + __asm__ volatile( 1.217 + "movq (%3), %%mm0 \n\t" 1.218 + "movq 8(%3), %%mm1 \n\t" 1.219 + "movq 16(%3), %%mm2 \n\t" 1.220 + "movq 24(%3), %%mm3 \n\t" 1.221 + "movq 32(%3), %%mm4 \n\t" 1.222 + "movq 40(%3), %%mm5 \n\t" 1.223 + "movq 48(%3), %%mm6 \n\t" 1.224 + "movq 56(%3), %%mm7 \n\t" 1.225 + "packuswb %%mm1, %%mm0 \n\t" 1.226 + "packuswb %%mm3, %%mm2 \n\t" 1.227 + "packuswb %%mm5, %%mm4 \n\t" 1.228 + "packuswb %%mm7, %%mm6 \n\t" 1.229 + "movq %%mm0, (%0) \n\t" 1.230 + "movq %%mm2, (%0, %1) \n\t" 1.231 + "movq %%mm4, (%0, %1, 2) \n\t" 1.232 + "movq %%mm6, (%0, %2) \n\t" 1.233 + ::"r" (pix), "r" ((x86_reg)line_size), "r" ((x86_reg)line_size*3), "r"(p) 1.234 + :"memory"); 1.235 +} 1.236 + 1.237 +DECLARE_ASM_CONST(8, uint8_t, ff_vector128)[8] = 1.238 + { 0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80 }; 1.239 + 1.240 +#define put_signed_pixels_clamped_mmx_half(off) \ 1.241 + "movq "#off"(%2), %%mm1 \n\t"\ 1.242 + "movq 16+"#off"(%2), %%mm2 \n\t"\ 1.243 + "movq 32+"#off"(%2), %%mm3 \n\t"\ 1.244 + "movq 48+"#off"(%2), %%mm4 \n\t"\ 1.245 + "packsswb 8+"#off"(%2), %%mm1 \n\t"\ 1.246 + "packsswb 24+"#off"(%2), %%mm2 \n\t"\ 1.247 + "packsswb 40+"#off"(%2), %%mm3 \n\t"\ 1.248 + "packsswb 56+"#off"(%2), %%mm4 \n\t"\ 1.249 + "paddb %%mm0, %%mm1 \n\t"\ 1.250 + "paddb %%mm0, %%mm2 \n\t"\ 1.251 + "paddb %%mm0, %%mm3 \n\t"\ 1.252 + "paddb %%mm0, %%mm4 \n\t"\ 1.253 + "movq %%mm1, (%0) \n\t"\ 1.254 + "movq %%mm2, (%0, %3) \n\t"\ 1.255 + "movq %%mm3, (%0, %3, 2) \n\t"\ 1.256 + "movq %%mm4, (%0, %1) \n\t" 1.257 + 1.258 +void put_signed_pixels_clamped_mmx(const DCTELEM *block, uint8_t *pixels, int line_size) 1.259 +{ 1.260 + x86_reg line_skip = line_size; 1.261 + x86_reg line_skip3; 1.262 + 1.263 + __asm__ volatile ( 1.264 + "movq "MANGLE(ff_vector128)", %%mm0 \n\t" 1.265 + "lea (%3, %3, 2), %1 \n\t" 1.266 + put_signed_pixels_clamped_mmx_half(0) 1.267 + "lea (%0, %3, 4), %0 \n\t" 1.268 + put_signed_pixels_clamped_mmx_half(64) 1.269 + :"+&r" (pixels), "=&r" (line_skip3) 1.270 + :"r" (block), "r"(line_skip) 1.271 + :"memory"); 1.272 +} 1.273 + 1.274 +void add_pixels_clamped_mmx(const DCTELEM *block, uint8_t *pixels, int line_size) 1.275 +{ 1.276 + const DCTELEM *p; 1.277 + uint8_t *pix; 1.278 + int i; 1.279 + 1.280 + /* read the pixels */ 1.281 + p = block; 1.282 + pix = pixels; 1.283 + MOVQ_ZERO(mm7); 1.284 + i = 4; 1.285 + do { 1.286 + __asm__ volatile( 1.287 + "movq (%2), %%mm0 \n\t" 1.288 + "movq 8(%2), %%mm1 \n\t" 1.289 + "movq 16(%2), %%mm2 \n\t" 1.290 + "movq 24(%2), %%mm3 \n\t" 1.291 + "movq %0, %%mm4 \n\t" 1.292 + "movq %1, %%mm6 \n\t" 1.293 + "movq %%mm4, %%mm5 \n\t" 1.294 + "punpcklbw %%mm7, %%mm4 \n\t" 1.295 + "punpckhbw %%mm7, %%mm5 \n\t" 1.296 + "paddsw %%mm4, %%mm0 \n\t" 1.297 + "paddsw %%mm5, %%mm1 \n\t" 1.298 + "movq %%mm6, %%mm5 \n\t" 1.299 + "punpcklbw %%mm7, %%mm6 \n\t" 1.300 + "punpckhbw %%mm7, %%mm5 \n\t" 1.301 + "paddsw %%mm6, %%mm2 \n\t" 1.302 + "paddsw %%mm5, %%mm3 \n\t" 1.303 + "packuswb %%mm1, %%mm0 \n\t" 1.304 + "packuswb %%mm3, %%mm2 \n\t" 1.305 + "movq %%mm0, %0 \n\t" 1.306 + "movq %%mm2, %1 \n\t" 1.307 + :"+m"(*pix), "+m"(*(pix+line_size)) 1.308 + :"r"(p) 1.309 + :"memory"); 1.310 + pix += line_size*2; 1.311 + p += 16; 1.312 + } while (--i); 1.313 +} 1.314 + 1.315 +static void put_pixels8_mmx(uint8_t *block, const uint8_t *pixels, int line_size, int h) 1.316 +{ 1.317 + __asm__ volatile( 1.318 + "lea (%3, %3), %%"REG_a" \n\t" 1.319 + ASMALIGN(3) 1.320 + "1: \n\t" 1.321 + "movq (%1), %%mm0 \n\t" 1.322 + "movq (%1, %3), %%mm1 \n\t" 1.323 + "movq %%mm0, (%2) \n\t" 1.324 + "movq %%mm1, (%2, %3) \n\t" 1.325 + "add %%"REG_a", %1 \n\t" 1.326 + "add %%"REG_a", %2 \n\t" 1.327 + "movq (%1), %%mm0 \n\t" 1.328 + "movq (%1, %3), %%mm1 \n\t" 1.329 + "movq %%mm0, (%2) \n\t" 1.330 + "movq %%mm1, (%2, %3) \n\t" 1.331 + "add %%"REG_a", %1 \n\t" 1.332 + "add %%"REG_a", %2 \n\t" 1.333 + "subl $4, %0 \n\t" 1.334 + "jnz 1b \n\t" 1.335 + : "+g"(h), "+r" (pixels), "+r" (block) 1.336 + : "r"((x86_reg)line_size) 1.337 + : "%"REG_a, "memory" 1.338 + ); 1.339 +} 1.340 + 1.341 +static void put_pixels16_sse2(uint8_t *block, const uint8_t *pixels, int line_size, int h) 1.342 +{ 1.343 + __asm__ volatile( 1.344 + "1: \n\t" 1.345 + "movdqu (%1), %%xmm0 \n\t" 1.346 + "movdqu (%1,%3), %%xmm1 \n\t" 1.347 + "movdqu (%1,%3,2), %%xmm2 \n\t" 1.348 + "movdqu (%1,%4), %%xmm3 \n\t" 1.349 + "movdqa %%xmm0, (%2) \n\t" 1.350 + "movdqa %%xmm1, (%2,%3) \n\t" 1.351 + "movdqa %%xmm2, (%2,%3,2) \n\t" 1.352 + "movdqa %%xmm3, (%2,%4) \n\t" 1.353 + "subl $4, %0 \n\t" 1.354 + "lea (%1,%3,4), %1 \n\t" 1.355 + "lea (%2,%3,4), %2 \n\t" 1.356 + "jnz 1b \n\t" 1.357 + : "+g"(h), "+r" (pixels), "+r" (block) 1.358 + : "r"((x86_reg)line_size), "r"((x86_reg)3L*line_size) 1.359 + : "memory" 1.360 + ); 1.361 +} 1.362 + 1.363 +static void avg_pixels16_sse2(uint8_t *block, const uint8_t *pixels, int line_size, int h) 1.364 +{ 1.365 + __asm__ volatile( 1.366 + "1: \n\t" 1.367 + "movdqu (%1), %%xmm0 \n\t" 1.368 + "movdqu (%1,%3), %%xmm1 \n\t" 1.369 + "movdqu (%1,%3,2), %%xmm2 \n\t" 1.370 + "movdqu (%1,%4), %%xmm3 \n\t" 1.371 + "pavgb (%2), %%xmm0 \n\t" 1.372 + "pavgb (%2,%3), %%xmm1 \n\t" 1.373 + "pavgb (%2,%3,2), %%xmm2 \n\t" 1.374 + "pavgb (%2,%4), %%xmm3 \n\t" 1.375 + "movdqa %%xmm0, (%2) \n\t" 1.376 + "movdqa %%xmm1, (%2,%3) \n\t" 1.377 + "movdqa %%xmm2, (%2,%3,2) \n\t" 1.378 + "movdqa %%xmm3, (%2,%4) \n\t" 1.379 + "subl $4, %0 \n\t" 1.380 + "lea (%1,%3,4), %1 \n\t" 1.381 + "lea (%2,%3,4), %2 \n\t" 1.382 + "jnz 1b \n\t" 1.383 + : "+g"(h), "+r" (pixels), "+r" (block) 1.384 + : "r"((x86_reg)line_size), "r"((x86_reg)3L*line_size) 1.385 + : "memory" 1.386 + ); 1.387 +} 1.388 + 1.389 +static void clear_block_sse(DCTELEM *block) 1.390 +{ 1.391 + __asm__ volatile( 1.392 + "xorps %%xmm0, %%xmm0 \n" 1.393 + "movaps %%xmm0, (%0) \n" 1.394 + "movaps %%xmm0, 16(%0) \n" 1.395 + "movaps %%xmm0, 32(%0) \n" 1.396 + "movaps %%xmm0, 48(%0) \n" 1.397 + "movaps %%xmm0, 64(%0) \n" 1.398 + "movaps %%xmm0, 80(%0) \n" 1.399 + "movaps %%xmm0, 96(%0) \n" 1.400 + "movaps %%xmm0, 112(%0) \n" 1.401 + :: "r"(block) 1.402 + : "memory" 1.403 + ); 1.404 +} 1.405 + 1.406 +static void clear_blocks_sse(DCTELEM *blocks) 1.407 +{\ 1.408 + __asm__ volatile( 1.409 + "xorps %%xmm0, %%xmm0 \n" 1.410 + "mov %1, %%"REG_a" \n" 1.411 + "1: \n" 1.412 + "movaps %%xmm0, (%0, %%"REG_a") \n" 1.413 + "movaps %%xmm0, 16(%0, %%"REG_a") \n" 1.414 + "movaps %%xmm0, 32(%0, %%"REG_a") \n" 1.415 + "movaps %%xmm0, 48(%0, %%"REG_a") \n" 1.416 + "movaps %%xmm0, 64(%0, %%"REG_a") \n" 1.417 + "movaps %%xmm0, 80(%0, %%"REG_a") \n" 1.418 + "movaps %%xmm0, 96(%0, %%"REG_a") \n" 1.419 + "movaps %%xmm0, 112(%0, %%"REG_a") \n" 1.420 + "add $128, %%"REG_a" \n" 1.421 + " js 1b \n" 1.422 + : : "r" (((uint8_t *)blocks)+128*6), 1.423 + "i" (-128*6) 1.424 + : "%"REG_a 1.425 + ); 1.426 +} 1.427 + 1.428 +static inline void transpose4x4(uint8_t *dst, uint8_t *src, int dst_stride, int src_stride){ 1.429 + __asm__ volatile( //FIXME could save 1 instruction if done as 8x4 ... 1.430 + "movd %4, %%mm0 \n\t" 1.431 + "movd %5, %%mm1 \n\t" 1.432 + "movd %6, %%mm2 \n\t" 1.433 + "movd %7, %%mm3 \n\t" 1.434 + "punpcklbw %%mm1, %%mm0 \n\t" 1.435 + "punpcklbw %%mm3, %%mm2 \n\t" 1.436 + "movq %%mm0, %%mm1 \n\t" 1.437 + "punpcklwd %%mm2, %%mm0 \n\t" 1.438 + "punpckhwd %%mm2, %%mm1 \n\t" 1.439 + "movd %%mm0, %0 \n\t" 1.440 + "punpckhdq %%mm0, %%mm0 \n\t" 1.441 + "movd %%mm0, %1 \n\t" 1.442 + "movd %%mm1, %2 \n\t" 1.443 + "punpckhdq %%mm1, %%mm1 \n\t" 1.444 + "movd %%mm1, %3 \n\t" 1.445 + 1.446 + : "=m" (*(uint32_t*)(dst + 0*dst_stride)), 1.447 + "=m" (*(uint32_t*)(dst + 1*dst_stride)), 1.448 + "=m" (*(uint32_t*)(dst + 2*dst_stride)), 1.449 + "=m" (*(uint32_t*)(dst + 3*dst_stride)) 1.450 + : "m" (*(uint32_t*)(src + 0*src_stride)), 1.451 + "m" (*(uint32_t*)(src + 1*src_stride)), 1.452 + "m" (*(uint32_t*)(src + 2*src_stride)), 1.453 + "m" (*(uint32_t*)(src + 3*src_stride)) 1.454 + ); 1.455 +} 1.456 + 1.457 +#define QPEL_OP(OPNAME, ROUNDER, RND, OP, MMX)\ 1.458 +\ 1.459 +static void OPNAME ## qpel8_mc00_ ## MMX (uint8_t *dst, uint8_t *src, int stride){\ 1.460 + OPNAME ## pixels8_ ## MMX(dst, src, stride, 8);\ 1.461 +}\ 1.462 +\ 1.463 +static void OPNAME ## qpel8_mc10_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\ 1.464 + uint64_t temp[8];\ 1.465 + uint8_t * const half= (uint8_t*)temp;\ 1.466 + put ## RND ## mpeg4_qpel8_h_lowpass_ ## MMX(half, src, 8, stride, 8);\ 1.467 + OPNAME ## pixels8_l2_ ## MMX(dst, src, half, stride, stride, 8);\ 1.468 +}\ 1.469 +\ 1.470 +static void OPNAME ## qpel8_mc20_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\ 1.471 + OPNAME ## mpeg4_qpel8_h_lowpass_ ## MMX(dst, src, stride, stride, 8);\ 1.472 +}\ 1.473 +\ 1.474 +static void OPNAME ## qpel8_mc30_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\ 1.475 + uint64_t temp[8];\ 1.476 + uint8_t * const half= (uint8_t*)temp;\ 1.477 + put ## RND ## mpeg4_qpel8_h_lowpass_ ## MMX(half, src, 8, stride, 8);\ 1.478 + OPNAME ## pixels8_l2_ ## MMX(dst, src+1, half, stride, stride, 8);\ 1.479 +}\ 1.480 +\ 1.481 +static void OPNAME ## qpel8_mc01_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\ 1.482 + uint64_t temp[8];\ 1.483 + uint8_t * const half= (uint8_t*)temp;\ 1.484 + put ## RND ## mpeg4_qpel8_v_lowpass_ ## MMX(half, src, 8, stride);\ 1.485 + OPNAME ## pixels8_l2_ ## MMX(dst, src, half, stride, stride, 8);\ 1.486 +}\ 1.487 +\ 1.488 +static void OPNAME ## qpel8_mc02_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\ 1.489 + OPNAME ## mpeg4_qpel8_v_lowpass_ ## MMX(dst, src, stride, stride);\ 1.490 +}\ 1.491 +\ 1.492 +static void OPNAME ## qpel8_mc03_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\ 1.493 + uint64_t temp[8];\ 1.494 + uint8_t * const half= (uint8_t*)temp;\ 1.495 + put ## RND ## mpeg4_qpel8_v_lowpass_ ## MMX(half, src, 8, stride);\ 1.496 + OPNAME ## pixels8_l2_ ## MMX(dst, src+stride, half, stride, stride, 8);\ 1.497 +}\ 1.498 +static void OPNAME ## qpel8_mc11_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\ 1.499 + uint64_t half[8 + 9];\ 1.500 + uint8_t * const halfH= ((uint8_t*)half) + 64;\ 1.501 + uint8_t * const halfHV= ((uint8_t*)half);\ 1.502 + put ## RND ## mpeg4_qpel8_h_lowpass_ ## MMX(halfH, src, 8, stride, 9);\ 1.503 + put ## RND ## pixels8_l2_ ## MMX(halfH, src, halfH, 8, stride, 9);\ 1.504 + put ## RND ## mpeg4_qpel8_v_lowpass_ ## MMX(halfHV, halfH, 8, 8);\ 1.505 + OPNAME ## pixels8_l2_ ## MMX(dst, halfH, halfHV, stride, 8, 8);\ 1.506 +}\ 1.507 +static void OPNAME ## qpel8_mc31_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\ 1.508 + uint64_t half[8 + 9];\ 1.509 + uint8_t * const halfH= ((uint8_t*)half) + 64;\ 1.510 + uint8_t * const halfHV= ((uint8_t*)half);\ 1.511 + put ## RND ## mpeg4_qpel8_h_lowpass_ ## MMX(halfH, src, 8, stride, 9);\ 1.512 + put ## RND ## pixels8_l2_ ## MMX(halfH, src+1, halfH, 8, stride, 9);\ 1.513 + put ## RND ## mpeg4_qpel8_v_lowpass_ ## MMX(halfHV, halfH, 8, 8);\ 1.514 + OPNAME ## pixels8_l2_ ## MMX(dst, halfH, halfHV, stride, 8, 8);\ 1.515 +}\ 1.516 +static void OPNAME ## qpel8_mc13_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\ 1.517 + uint64_t half[8 + 9];\ 1.518 + uint8_t * const halfH= ((uint8_t*)half) + 64;\ 1.519 + uint8_t * const halfHV= ((uint8_t*)half);\ 1.520 + put ## RND ## mpeg4_qpel8_h_lowpass_ ## MMX(halfH, src, 8, stride, 9);\ 1.521 + put ## RND ## pixels8_l2_ ## MMX(halfH, src, halfH, 8, stride, 9);\ 1.522 + put ## RND ## mpeg4_qpel8_v_lowpass_ ## MMX(halfHV, halfH, 8, 8);\ 1.523 + OPNAME ## pixels8_l2_ ## MMX(dst, halfH+8, halfHV, stride, 8, 8);\ 1.524 +}\ 1.525 +static void OPNAME ## qpel8_mc33_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\ 1.526 + uint64_t half[8 + 9];\ 1.527 + uint8_t * const halfH= ((uint8_t*)half) + 64;\ 1.528 + uint8_t * const halfHV= ((uint8_t*)half);\ 1.529 + put ## RND ## mpeg4_qpel8_h_lowpass_ ## MMX(halfH, src, 8, stride, 9);\ 1.530 + put ## RND ## pixels8_l2_ ## MMX(halfH, src+1, halfH, 8, stride, 9);\ 1.531 + put ## RND ## mpeg4_qpel8_v_lowpass_ ## MMX(halfHV, halfH, 8, 8);\ 1.532 + OPNAME ## pixels8_l2_ ## MMX(dst, halfH+8, halfHV, stride, 8, 8);\ 1.533 +}\ 1.534 +static void OPNAME ## qpel8_mc21_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\ 1.535 + uint64_t half[8 + 9];\ 1.536 + uint8_t * const halfH= ((uint8_t*)half) + 64;\ 1.537 + uint8_t * const halfHV= ((uint8_t*)half);\ 1.538 + put ## RND ## mpeg4_qpel8_h_lowpass_ ## MMX(halfH, src, 8, stride, 9);\ 1.539 + put ## RND ## mpeg4_qpel8_v_lowpass_ ## MMX(halfHV, halfH, 8, 8);\ 1.540 + OPNAME ## pixels8_l2_ ## MMX(dst, halfH, halfHV, stride, 8, 8);\ 1.541 +}\ 1.542 +static void OPNAME ## qpel8_mc23_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\ 1.543 + uint64_t half[8 + 9];\ 1.544 + uint8_t * const halfH= ((uint8_t*)half) + 64;\ 1.545 + uint8_t * const halfHV= ((uint8_t*)half);\ 1.546 + put ## RND ## mpeg4_qpel8_h_lowpass_ ## MMX(halfH, src, 8, stride, 9);\ 1.547 + put ## RND ## mpeg4_qpel8_v_lowpass_ ## MMX(halfHV, halfH, 8, 8);\ 1.548 + OPNAME ## pixels8_l2_ ## MMX(dst, halfH+8, halfHV, stride, 8, 8);\ 1.549 +}\ 1.550 +static void OPNAME ## qpel8_mc12_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\ 1.551 + uint64_t half[8 + 9];\ 1.552 + uint8_t * const halfH= ((uint8_t*)half);\ 1.553 + put ## RND ## mpeg4_qpel8_h_lowpass_ ## MMX(halfH, src, 8, stride, 9);\ 1.554 + put ## RND ## pixels8_l2_ ## MMX(halfH, src, halfH, 8, stride, 9);\ 1.555 + OPNAME ## mpeg4_qpel8_v_lowpass_ ## MMX(dst, halfH, stride, 8);\ 1.556 +}\ 1.557 +static void OPNAME ## qpel8_mc32_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\ 1.558 + uint64_t half[8 + 9];\ 1.559 + uint8_t * const halfH= ((uint8_t*)half);\ 1.560 + put ## RND ## mpeg4_qpel8_h_lowpass_ ## MMX(halfH, src, 8, stride, 9);\ 1.561 + put ## RND ## pixels8_l2_ ## MMX(halfH, src+1, halfH, 8, stride, 9);\ 1.562 + OPNAME ## mpeg4_qpel8_v_lowpass_ ## MMX(dst, halfH, stride, 8);\ 1.563 +}\ 1.564 +static void OPNAME ## qpel8_mc22_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\ 1.565 + uint64_t half[9];\ 1.566 + uint8_t * const halfH= ((uint8_t*)half);\ 1.567 + put ## RND ## mpeg4_qpel8_h_lowpass_ ## MMX(halfH, src, 8, stride, 9);\ 1.568 + OPNAME ## mpeg4_qpel8_v_lowpass_ ## MMX(dst, halfH, stride, 8);\ 1.569 +}\ 1.570 +static void OPNAME ## qpel16_mc00_ ## MMX (uint8_t *dst, uint8_t *src, int stride){\ 1.571 + OPNAME ## pixels16_ ## MMX(dst, src, stride, 16);\ 1.572 +}\ 1.573 +\ 1.574 +static void OPNAME ## qpel16_mc10_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\ 1.575 + uint64_t temp[32];\ 1.576 + uint8_t * const half= (uint8_t*)temp;\ 1.577 + put ## RND ## mpeg4_qpel16_h_lowpass_ ## MMX(half, src, 16, stride, 16);\ 1.578 + OPNAME ## pixels16_l2_ ## MMX(dst, src, half, stride, stride, 16);\ 1.579 +}\ 1.580 +\ 1.581 +static void OPNAME ## qpel16_mc20_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\ 1.582 + OPNAME ## mpeg4_qpel16_h_lowpass_ ## MMX(dst, src, stride, stride, 16);\ 1.583 +}\ 1.584 +\ 1.585 +static void OPNAME ## qpel16_mc30_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\ 1.586 + uint64_t temp[32];\ 1.587 + uint8_t * const half= (uint8_t*)temp;\ 1.588 + put ## RND ## mpeg4_qpel16_h_lowpass_ ## MMX(half, src, 16, stride, 16);\ 1.589 + OPNAME ## pixels16_l2_ ## MMX(dst, src+1, half, stride, stride, 16);\ 1.590 +}\ 1.591 +\ 1.592 +static void OPNAME ## qpel16_mc01_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\ 1.593 + uint64_t temp[32];\ 1.594 + uint8_t * const half= (uint8_t*)temp;\ 1.595 + put ## RND ## mpeg4_qpel16_v_lowpass_ ## MMX(half, src, 16, stride);\ 1.596 + OPNAME ## pixels16_l2_ ## MMX(dst, src, half, stride, stride, 16);\ 1.597 +}\ 1.598 +\ 1.599 +static void OPNAME ## qpel16_mc02_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\ 1.600 + OPNAME ## mpeg4_qpel16_v_lowpass_ ## MMX(dst, src, stride, stride);\ 1.601 +}\ 1.602 +\ 1.603 +static void OPNAME ## qpel16_mc03_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\ 1.604 + uint64_t temp[32];\ 1.605 + uint8_t * const half= (uint8_t*)temp;\ 1.606 + put ## RND ## mpeg4_qpel16_v_lowpass_ ## MMX(half, src, 16, stride);\ 1.607 + OPNAME ## pixels16_l2_ ## MMX(dst, src+stride, half, stride, stride, 16);\ 1.608 +}\ 1.609 +static void OPNAME ## qpel16_mc11_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\ 1.610 + uint64_t half[16*2 + 17*2];\ 1.611 + uint8_t * const halfH= ((uint8_t*)half) + 256;\ 1.612 + uint8_t * const halfHV= ((uint8_t*)half);\ 1.613 + put ## RND ## mpeg4_qpel16_h_lowpass_ ## MMX(halfH, src, 16, stride, 17);\ 1.614 + put ## RND ## pixels16_l2_ ## MMX(halfH, src, halfH, 16, stride, 17);\ 1.615 + put ## RND ## mpeg4_qpel16_v_lowpass_ ## MMX(halfHV, halfH, 16, 16);\ 1.616 + OPNAME ## pixels16_l2_ ## MMX(dst, halfH, halfHV, stride, 16, 16);\ 1.617 +}\ 1.618 +static void OPNAME ## qpel16_mc31_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\ 1.619 + uint64_t half[16*2 + 17*2];\ 1.620 + uint8_t * const halfH= ((uint8_t*)half) + 256;\ 1.621 + uint8_t * const halfHV= ((uint8_t*)half);\ 1.622 + put ## RND ## mpeg4_qpel16_h_lowpass_ ## MMX(halfH, src, 16, stride, 17);\ 1.623 + put ## RND ## pixels16_l2_ ## MMX(halfH, src+1, halfH, 16, stride, 17);\ 1.624 + put ## RND ## mpeg4_qpel16_v_lowpass_ ## MMX(halfHV, halfH, 16, 16);\ 1.625 + OPNAME ## pixels16_l2_ ## MMX(dst, halfH, halfHV, stride, 16, 16);\ 1.626 +}\ 1.627 +static void OPNAME ## qpel16_mc13_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\ 1.628 + uint64_t half[16*2 + 17*2];\ 1.629 + uint8_t * const halfH= ((uint8_t*)half) + 256;\ 1.630 + uint8_t * const halfHV= ((uint8_t*)half);\ 1.631 + put ## RND ## mpeg4_qpel16_h_lowpass_ ## MMX(halfH, src, 16, stride, 17);\ 1.632 + put ## RND ## pixels16_l2_ ## MMX(halfH, src, halfH, 16, stride, 17);\ 1.633 + put ## RND ## mpeg4_qpel16_v_lowpass_ ## MMX(halfHV, halfH, 16, 16);\ 1.634 + OPNAME ## pixels16_l2_ ## MMX(dst, halfH+16, halfHV, stride, 16, 16);\ 1.635 +}\ 1.636 +static void OPNAME ## qpel16_mc33_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\ 1.637 + uint64_t half[16*2 + 17*2];\ 1.638 + uint8_t * const halfH= ((uint8_t*)half) + 256;\ 1.639 + uint8_t * const halfHV= ((uint8_t*)half);\ 1.640 + put ## RND ## mpeg4_qpel16_h_lowpass_ ## MMX(halfH, src, 16, stride, 17);\ 1.641 + put ## RND ## pixels16_l2_ ## MMX(halfH, src+1, halfH, 16, stride, 17);\ 1.642 + put ## RND ## mpeg4_qpel16_v_lowpass_ ## MMX(halfHV, halfH, 16, 16);\ 1.643 + OPNAME ## pixels16_l2_ ## MMX(dst, halfH+16, halfHV, stride, 16, 16);\ 1.644 +}\ 1.645 +static void OPNAME ## qpel16_mc21_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\ 1.646 + uint64_t half[16*2 + 17*2];\ 1.647 + uint8_t * const halfH= ((uint8_t*)half) + 256;\ 1.648 + uint8_t * const halfHV= ((uint8_t*)half);\ 1.649 + put ## RND ## mpeg4_qpel16_h_lowpass_ ## MMX(halfH, src, 16, stride, 17);\ 1.650 + put ## RND ## mpeg4_qpel16_v_lowpass_ ## MMX(halfHV, halfH, 16, 16);\ 1.651 + OPNAME ## pixels16_l2_ ## MMX(dst, halfH, halfHV, stride, 16, 16);\ 1.652 +}\ 1.653 +static void OPNAME ## qpel16_mc23_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\ 1.654 + uint64_t half[16*2 + 17*2];\ 1.655 + uint8_t * const halfH= ((uint8_t*)half) + 256;\ 1.656 + uint8_t * const halfHV= ((uint8_t*)half);\ 1.657 + put ## RND ## mpeg4_qpel16_h_lowpass_ ## MMX(halfH, src, 16, stride, 17);\ 1.658 + put ## RND ## mpeg4_qpel16_v_lowpass_ ## MMX(halfHV, halfH, 16, 16);\ 1.659 + OPNAME ## pixels16_l2_ ## MMX(dst, halfH+16, halfHV, stride, 16, 16);\ 1.660 +}\ 1.661 +static void OPNAME ## qpel16_mc12_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\ 1.662 + uint64_t half[17*2];\ 1.663 + uint8_t * const halfH= ((uint8_t*)half);\ 1.664 + put ## RND ## mpeg4_qpel16_h_lowpass_ ## MMX(halfH, src, 16, stride, 17);\ 1.665 + put ## RND ## pixels16_l2_ ## MMX(halfH, src, halfH, 16, stride, 17);\ 1.666 + OPNAME ## mpeg4_qpel16_v_lowpass_ ## MMX(dst, halfH, stride, 16);\ 1.667 +}\ 1.668 +static void OPNAME ## qpel16_mc32_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\ 1.669 + uint64_t half[17*2];\ 1.670 + uint8_t * const halfH= ((uint8_t*)half);\ 1.671 + put ## RND ## mpeg4_qpel16_h_lowpass_ ## MMX(halfH, src, 16, stride, 17);\ 1.672 + put ## RND ## pixels16_l2_ ## MMX(halfH, src+1, halfH, 16, stride, 17);\ 1.673 + OPNAME ## mpeg4_qpel16_v_lowpass_ ## MMX(dst, halfH, stride, 16);\ 1.674 +}\ 1.675 +static void OPNAME ## qpel16_mc22_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\ 1.676 + uint64_t half[17*2];\ 1.677 + uint8_t * const halfH= ((uint8_t*)half);\ 1.678 + put ## RND ## mpeg4_qpel16_h_lowpass_ ## MMX(halfH, src, 16, stride, 17);\ 1.679 + OPNAME ## mpeg4_qpel16_v_lowpass_ ## MMX(dst, halfH, stride, 16);\ 1.680 +} 1.681 + 1.682 +#define PUT_OP(a,b,temp, size) "mov" #size " " #a ", " #b " \n\t" 1.683 +#define AVG_3DNOW_OP(a,b,temp, size) \ 1.684 +"mov" #size " " #b ", " #temp " \n\t"\ 1.685 +"pavgusb " #temp ", " #a " \n\t"\ 1.686 +"mov" #size " " #a ", " #b " \n\t" 1.687 +#define AVG_MMX2_OP(a,b,temp, size) \ 1.688 +"mov" #size " " #b ", " #temp " \n\t"\ 1.689 +"pavgb " #temp ", " #a " \n\t"\ 1.690 +"mov" #size " " #a ", " #b " \n\t" 1.691 + 1.692 +#define PREFETCH(name, op) \ 1.693 +static void name(void *mem, int stride, int h){\ 1.694 + const uint8_t *p= mem;\ 1.695 + do{\ 1.696 + __asm__ volatile(#op" %0" :: "m"(*p));\ 1.697 + p+= stride;\ 1.698 + }while(--h);\ 1.699 +} 1.700 +PREFETCH(prefetch_mmx2, prefetcht0) 1.701 +#undef PREFETCH 1.702 + 1.703 +#include "h264dsp_mmx.c" 1.704 + 1.705 +void ff_x264_deblock_v_luma_sse2(uint8_t *pix, int stride, int alpha, int beta, int8_t *tc0); 1.706 +void ff_x264_deblock_h_luma_sse2(uint8_t *pix, int stride, int alpha, int beta, int8_t *tc0); 1.707 +void ff_x264_deblock_h_luma_intra_mmxext(uint8_t *pix, int stride, int alpha, int beta); 1.708 +void ff_x264_deblock_v_luma_intra_sse2(uint8_t *pix, int stride, int alpha, int beta); 1.709 +void ff_x264_deblock_h_luma_intra_sse2(uint8_t *pix, int stride, int alpha, int beta); 1.710 + 1.711 +void dsputil_init_mmx(DSPContext* c) 1.712 +{ 1.713 + mm_flags = mm_support(); 1.714 + 1.715 + if (mm_flags & FF_MM_MMX) { 1.716 + c->clear_block = clear_block_sse; 1.717 + c->clear_blocks = clear_blocks_sse; 1.718 + c->prefetch = prefetch_mmx2; 1.719 + 1.720 + 1.721 +#define H264_QPEL_FUNCS(x, y, CPU)\ 1.722 + c->put_h264_qpel_pixels_tab[0][x+y*4] = put_h264_qpel16_mc##x##y##_##CPU;\ 1.723 + c->put_h264_qpel_pixels_tab[1][x+y*4] = put_h264_qpel8_mc##x##y##_##CPU;\ 1.724 + c->avg_h264_qpel_pixels_tab[0][x+y*4] = avg_h264_qpel16_mc##x##y##_##CPU;\ 1.725 + c->avg_h264_qpel_pixels_tab[1][x+y*4] = avg_h264_qpel8_mc##x##y##_##CPU; 1.726 + 1.727 + if((mm_flags & FF_MM_SSE2)){ 1.728 + c->put_pixels_tab[0][0] = put_pixels16_sse2; 1.729 + c->avg_pixels_tab[0][0] = avg_pixels16_sse2; 1.730 + 1.731 + } 1.732 + if(mm_flags & FF_MM_SSE2){ 1.733 + H264_QPEL_FUNCS(0, 1, sse2); 1.734 + H264_QPEL_FUNCS(0, 2, sse2); 1.735 + H264_QPEL_FUNCS(0, 3, sse2); 1.736 + H264_QPEL_FUNCS(1, 1, sse2); 1.737 + H264_QPEL_FUNCS(1, 2, sse2); 1.738 + H264_QPEL_FUNCS(1, 3, sse2); 1.739 + H264_QPEL_FUNCS(2, 1, sse2); 1.740 + H264_QPEL_FUNCS(2, 2, sse2); 1.741 + H264_QPEL_FUNCS(2, 3, sse2); 1.742 + H264_QPEL_FUNCS(3, 1, sse2); 1.743 + H264_QPEL_FUNCS(3, 2, sse2); 1.744 + H264_QPEL_FUNCS(3, 3, sse2); 1.745 + } 1.746 +#if HAVE_SSSE3 1.747 + if(mm_flags & FF_MM_SSSE3){ 1.748 + H264_QPEL_FUNCS(1, 0, ssse3); 1.749 + H264_QPEL_FUNCS(1, 1, ssse3); 1.750 + H264_QPEL_FUNCS(1, 2, ssse3); 1.751 + H264_QPEL_FUNCS(1, 3, ssse3); 1.752 + H264_QPEL_FUNCS(2, 0, ssse3); 1.753 + H264_QPEL_FUNCS(2, 1, ssse3); 1.754 + H264_QPEL_FUNCS(2, 2, ssse3); 1.755 + H264_QPEL_FUNCS(2, 3, ssse3); 1.756 + H264_QPEL_FUNCS(3, 0, ssse3); 1.757 + H264_QPEL_FUNCS(3, 1, ssse3); 1.758 + H264_QPEL_FUNCS(3, 2, ssse3); 1.759 + H264_QPEL_FUNCS(3, 3, ssse3); 1.760 + 1.761 + c->put_h264_chroma_pixels_tab[0]= put_h264_chroma_mc8_ssse3_rnd; 1.762 + c->avg_h264_chroma_pixels_tab[0]= avg_h264_chroma_mc8_ssse3_rnd; 1.763 + c->put_h264_chroma_pixels_tab[1]= put_h264_chroma_mc4_ssse3; 1.764 + c->avg_h264_chroma_pixels_tab[1]= avg_h264_chroma_mc4_ssse3; 1.765 + } 1.766 +#endif 1.767 + 1.768 + 1.769 + } 1.770 +} 1.771 + 1.772 +void ff_h264dsp_init_x86(H264DSPContext *c) 1.773 +{ 1.774 + mm_flags = mm_support(); 1.775 + 1.776 + if (mm_flags & FF_MM_MMX) { 1.777 + c->h264_idct_dc_add= 1.778 + c->h264_idct_add= ff_h264_idct_add_mmx; 1.779 + c->h264_idct8_dc_add= 1.780 + c->h264_idct8_add= ff_h264_idct8_add_mmx; 1.781 + 1.782 + if (mm_flags & FF_MM_MMX2) { 1.783 + c->h264_idct_dc_add= ff_h264_idct_dc_add_mmx2; 1.784 + c->h264_idct_add8 = ff_h264_idct_add8_mmx2; 1.785 + c->h264_idct_add16 = ff_h264_idct_add16_mmx2; 1.786 + c->h264_idct_add16intra= ff_h264_idct_add16intra_mmx2; 1.787 + 1.788 + c->h264_idct8_dc_add= ff_h264_idct8_dc_add_mmx2; 1.789 + c->h264_idct8_add4 = ff_h264_idct8_add4_mmx2; 1.790 + 1.791 + c->h264_v_loop_filter_luma= h264_v_loop_filter_luma_mmx2; 1.792 + c->h264_h_loop_filter_luma= h264_h_loop_filter_luma_mmx2; 1.793 + c->h264_v_loop_filter_chroma= h264_v_loop_filter_chroma_mmx2; 1.794 + c->h264_h_loop_filter_chroma= h264_h_loop_filter_chroma_mmx2; 1.795 + c->h264_v_loop_filter_chroma_intra= h264_v_loop_filter_chroma_intra_mmx2; 1.796 + c->h264_h_loop_filter_chroma_intra= h264_h_loop_filter_chroma_intra_mmx2; 1.797 + c->h264_loop_filter_strength= h264_loop_filter_strength_mmx2; 1.798 + 1.799 + c->weight_h264_pixels_tab[0]= ff_h264_weight_16x16_mmx2; 1.800 + c->weight_h264_pixels_tab[1]= ff_h264_weight_16x8_mmx2; 1.801 + c->weight_h264_pixels_tab[2]= ff_h264_weight_8x16_mmx2; 1.802 + c->weight_h264_pixels_tab[3]= ff_h264_weight_8x8_mmx2; 1.803 + c->weight_h264_pixels_tab[4]= ff_h264_weight_8x4_mmx2; 1.804 + c->weight_h264_pixels_tab[5]= ff_h264_weight_4x8_mmx2; 1.805 + c->weight_h264_pixels_tab[6]= ff_h264_weight_4x4_mmx2; 1.806 + c->weight_h264_pixels_tab[7]= ff_h264_weight_4x2_mmx2; 1.807 + 1.808 + c->biweight_h264_pixels_tab[0]= ff_h264_biweight_16x16_mmx2; 1.809 + c->biweight_h264_pixels_tab[1]= ff_h264_biweight_16x8_mmx2; 1.810 + c->biweight_h264_pixels_tab[2]= ff_h264_biweight_8x16_mmx2; 1.811 + c->biweight_h264_pixels_tab[3]= ff_h264_biweight_8x8_mmx2; 1.812 + c->biweight_h264_pixels_tab[4]= ff_h264_biweight_8x4_mmx2; 1.813 + c->biweight_h264_pixels_tab[5]= ff_h264_biweight_4x8_mmx2; 1.814 + c->biweight_h264_pixels_tab[6]= ff_h264_biweight_4x4_mmx2; 1.815 + c->biweight_h264_pixels_tab[7]= ff_h264_biweight_4x2_mmx2; 1.816 + } 1.817 + if(mm_flags & FF_MM_SSE2){ 1.818 + c->h264_idct8_add = ff_h264_idct8_add_sse2; 1.819 + c->h264_idct8_add4= ff_h264_idct8_add4_sse2; 1.820 + } 1.821 + 1.822 + } 1.823 +} 1.824 +