00001 /* 00002 * MMX optimized DSP utils 00003 * Copyright (c) 2000, 2001 Fabrice Bellard 00004 * Copyright (c) 2002-2004 Michael Niedermayer <michaelni@gmx.at> 00005 * 00006 * This file is part of FFmpeg. 00007 * 00008 * FFmpeg is free software; you can redistribute it and/or 00009 * modify it under the terms of the GNU Lesser General Public 00010 * License as published by the Free Software Foundation; either 00011 * version 2.1 of the License, or (at your option) any later version. 00012 * 00013 * FFmpeg is distributed in the hope that it will be useful, 00014 * but WITHOUT ANY WARRANTY; without even the implied warranty of 00015 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU 00016 * Lesser General Public License for more details. 00017 * 00018 * You should have received a copy of the GNU Lesser General Public 00019 * License along with FFmpeg; if not, write to the Free Software 00020 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA 00021 * 00022 * MMX optimization by Nick Kurshev <nickols_k@mail.ru> 00023 */ 00024 00025 #include "libavutil/x86_cpu.h" 00026 #include "libavcodec/dsputil.h" 00027 #include "libavcodec/h263.h" 00028 #include "libavcodec/mpegvideo.h" 00029 #include "libavcodec/simple_idct.h" 00030 #include "dsputil_mmx.h" 00031 #include "mmx.h" 00032 #include "vp3dsp_mmx.h" 00033 #include "vp3dsp_sse2.h" 00034 #include "vp6dsp_mmx.h" 00035 #include "vp6dsp_sse2.h" 00036 #include "idct_xvid.h" 00037 00038 //#undef NDEBUG 00039 //#include <assert.h> 00040 00041 int mm_flags; /* multimedia extension flags */ 00042 00043 /* pixel operations */ 00044 DECLARE_ALIGNED_8 (const uint64_t, ff_bone) = 0x0101010101010101ULL; 00045 DECLARE_ALIGNED_8 (const uint64_t, ff_wtwo) = 0x0002000200020002ULL; 00046 00047 DECLARE_ALIGNED_16(const uint64_t, ff_pdw_80000000[2]) = 00048 {0x8000000080000000ULL, 0x8000000080000000ULL}; 00049 00050 DECLARE_ALIGNED_8 (const uint64_t, ff_pw_3 ) = 0x0003000300030003ULL; 00051 DECLARE_ALIGNED_8 (const uint64_t, ff_pw_4 ) = 0x0004000400040004ULL; 00052 DECLARE_ALIGNED_16(const xmm_reg, ff_pw_5 ) = {0x0005000500050005ULL, 0x0005000500050005ULL}; 00053 DECLARE_ALIGNED_16(const xmm_reg, ff_pw_8 ) = {0x0008000800080008ULL, 0x0008000800080008ULL}; 00054 DECLARE_ALIGNED_8 (const uint64_t, ff_pw_15 ) = 0x000F000F000F000FULL; 00055 DECLARE_ALIGNED_16(const xmm_reg, ff_pw_16 ) = {0x0010001000100010ULL, 0x0010001000100010ULL}; 00056 DECLARE_ALIGNED_8 (const uint64_t, ff_pw_20 ) = 0x0014001400140014ULL; 00057 DECLARE_ALIGNED_16(const xmm_reg, ff_pw_28 ) = {0x001C001C001C001CULL, 0x001C001C001C001CULL}; 00058 DECLARE_ALIGNED_16(const xmm_reg, ff_pw_32 ) = {0x0020002000200020ULL, 0x0020002000200020ULL}; 00059 DECLARE_ALIGNED_8 (const uint64_t, ff_pw_42 ) = 0x002A002A002A002AULL; 00060 DECLARE_ALIGNED_16(const xmm_reg, ff_pw_64 ) = {0x0040004000400040ULL, 0x0040004000400040ULL}; 00061 DECLARE_ALIGNED_8 (const uint64_t, ff_pw_96 ) = 0x0060006000600060ULL; 00062 DECLARE_ALIGNED_8 (const uint64_t, ff_pw_128) = 0x0080008000800080ULL; 00063 DECLARE_ALIGNED_8 (const uint64_t, ff_pw_255) = 0x00ff00ff00ff00ffULL; 00064 00065 DECLARE_ALIGNED_8 (const uint64_t, ff_pb_1 ) = 0x0101010101010101ULL; 00066 DECLARE_ALIGNED_8 (const uint64_t, ff_pb_3 ) = 0x0303030303030303ULL; 00067 DECLARE_ALIGNED_8 (const uint64_t, ff_pb_7 ) = 0x0707070707070707ULL; 00068 DECLARE_ALIGNED_8 (const uint64_t, ff_pb_1F ) = 0x1F1F1F1F1F1F1F1FULL; 00069 DECLARE_ALIGNED_8 (const uint64_t, ff_pb_3F ) = 0x3F3F3F3F3F3F3F3FULL; 00070 DECLARE_ALIGNED_8 (const uint64_t, ff_pb_81 ) = 0x8181818181818181ULL; 00071 DECLARE_ALIGNED_8 (const uint64_t, ff_pb_A1 ) = 0xA1A1A1A1A1A1A1A1ULL; 00072 DECLARE_ALIGNED_8 (const uint64_t, ff_pb_FC ) = 0xFCFCFCFCFCFCFCFCULL; 00073 00074 DECLARE_ALIGNED_16(const double, ff_pd_1[2]) = { 1.0, 1.0 }; 00075 DECLARE_ALIGNED_16(const double, ff_pd_2[2]) = { 2.0, 2.0 }; 00076 00077 #define JUMPALIGN() __asm__ volatile (ASMALIGN(3)::) 00078 #define MOVQ_ZERO(regd) __asm__ volatile ("pxor %%" #regd ", %%" #regd ::) 00079 00080 #define MOVQ_BFE(regd) \ 00081 __asm__ volatile ( \ 00082 "pcmpeqd %%" #regd ", %%" #regd " \n\t"\ 00083 "paddb %%" #regd ", %%" #regd " \n\t" ::) 00084 00085 #ifndef PIC 00086 #define MOVQ_BONE(regd) __asm__ volatile ("movq %0, %%" #regd " \n\t" ::"m"(ff_bone)) 00087 #define MOVQ_WTWO(regd) __asm__ volatile ("movq %0, %%" #regd " \n\t" ::"m"(ff_wtwo)) 00088 #else 00089 // for shared library it's better to use this way for accessing constants 00090 // pcmpeqd -> -1 00091 #define MOVQ_BONE(regd) \ 00092 __asm__ volatile ( \ 00093 "pcmpeqd %%" #regd ", %%" #regd " \n\t" \ 00094 "psrlw 15,ドル %%" #regd " \n\t" \ 00095 "packuswb %%" #regd ", %%" #regd " \n\t" ::) 00096 00097 #define MOVQ_WTWO(regd) \ 00098 __asm__ volatile ( \ 00099 "pcmpeqd %%" #regd ", %%" #regd " \n\t" \ 00100 "psrlw 15,ドル %%" #regd " \n\t" \ 00101 "psllw 1,ドル %%" #regd " \n\t"::) 00102 00103 #endif 00104 00105 // using regr as temporary and for the output result 00106 // first argument is unmodifed and second is trashed 00107 // regfe is supposed to contain 0xfefefefefefefefe 00108 #define PAVGB_MMX_NO_RND(rega, regb, regr, regfe) \ 00109 "movq " #rega ", " #regr " \n\t"\ 00110 "pand " #regb ", " #regr " \n\t"\ 00111 "pxor " #rega ", " #regb " \n\t"\ 00112 "pand " #regfe "," #regb " \n\t"\ 00113 "psrlq 1,ドル " #regb " \n\t"\ 00114 "paddb " #regb ", " #regr " \n\t" 00115 00116 #define PAVGB_MMX(rega, regb, regr, regfe) \ 00117 "movq " #rega ", " #regr " \n\t"\ 00118 "por " #regb ", " #regr " \n\t"\ 00119 "pxor " #rega ", " #regb " \n\t"\ 00120 "pand " #regfe "," #regb " \n\t"\ 00121 "psrlq 1,ドル " #regb " \n\t"\ 00122 "psubb " #regb ", " #regr " \n\t" 00123 00124 // mm6 is supposed to contain 0xfefefefefefefefe 00125 #define PAVGBP_MMX_NO_RND(rega, regb, regr, regc, regd, regp) \ 00126 "movq " #rega ", " #regr " \n\t"\ 00127 "movq " #regc ", " #regp " \n\t"\ 00128 "pand " #regb ", " #regr " \n\t"\ 00129 "pand " #regd ", " #regp " \n\t"\ 00130 "pxor " #rega ", " #regb " \n\t"\ 00131 "pxor " #regc ", " #regd " \n\t"\ 00132 "pand %%mm6, " #regb " \n\t"\ 00133 "pand %%mm6, " #regd " \n\t"\ 00134 "psrlq 1,ドル " #regb " \n\t"\ 00135 "psrlq 1,ドル " #regd " \n\t"\ 00136 "paddb " #regb ", " #regr " \n\t"\ 00137 "paddb " #regd ", " #regp " \n\t" 00138 00139 #define PAVGBP_MMX(rega, regb, regr, regc, regd, regp) \ 00140 "movq " #rega ", " #regr " \n\t"\ 00141 "movq " #regc ", " #regp " \n\t"\ 00142 "por " #regb ", " #regr " \n\t"\ 00143 "por " #regd ", " #regp " \n\t"\ 00144 "pxor " #rega ", " #regb " \n\t"\ 00145 "pxor " #regc ", " #regd " \n\t"\ 00146 "pand %%mm6, " #regb " \n\t"\ 00147 "pand %%mm6, " #regd " \n\t"\ 00148 "psrlq 1,ドル " #regd " \n\t"\ 00149 "psrlq 1,ドル " #regb " \n\t"\ 00150 "psubb " #regb ", " #regr " \n\t"\ 00151 "psubb " #regd ", " #regp " \n\t" 00152 00153 /***********************************/ 00154 /* MMX no rounding */ 00155 #define DEF(x, y) x ## _no_rnd_ ## y ##_mmx 00156 #define SET_RND MOVQ_WONE 00157 #define PAVGBP(a, b, c, d, e, f) PAVGBP_MMX_NO_RND(a, b, c, d, e, f) 00158 #define PAVGB(a, b, c, e) PAVGB_MMX_NO_RND(a, b, c, e) 00159 00160 #include "dsputil_mmx_rnd_template.c" 00161 00162 #undef DEF 00163 #undef SET_RND 00164 #undef PAVGBP 00165 #undef PAVGB 00166 /***********************************/ 00167 /* MMX rounding */ 00168 00169 #define DEF(x, y) x ## _ ## y ##_mmx 00170 #define SET_RND MOVQ_WTWO 00171 #define PAVGBP(a, b, c, d, e, f) PAVGBP_MMX(a, b, c, d, e, f) 00172 #define PAVGB(a, b, c, e) PAVGB_MMX(a, b, c, e) 00173 00174 #include "dsputil_mmx_rnd_template.c" 00175 00176 #undef DEF 00177 #undef SET_RND 00178 #undef PAVGBP 00179 #undef PAVGB 00180 00181 /***********************************/ 00182 /* 3Dnow specific */ 00183 00184 #define DEF(x) x ## _3dnow 00185 #define PAVGB "pavgusb" 00186 00187 #include "dsputil_mmx_avg_template.c" 00188 00189 #undef DEF 00190 #undef PAVGB 00191 00192 /***********************************/ 00193 /* MMX2 specific */ 00194 00195 #define DEF(x) x ## _mmx2 00196 00197 /* Introduced only in MMX2 set */ 00198 #define PAVGB "pavgb" 00199 00200 #include "dsputil_mmx_avg_template.c" 00201 00202 #undef DEF 00203 #undef PAVGB 00204 00205 #define put_no_rnd_pixels16_mmx put_pixels16_mmx 00206 #define put_no_rnd_pixels8_mmx put_pixels8_mmx 00207 #define put_pixels16_mmx2 put_pixels16_mmx 00208 #define put_pixels8_mmx2 put_pixels8_mmx 00209 #define put_pixels4_mmx2 put_pixels4_mmx 00210 #define put_no_rnd_pixels16_mmx2 put_no_rnd_pixels16_mmx 00211 #define put_no_rnd_pixels8_mmx2 put_no_rnd_pixels8_mmx 00212 #define put_pixels16_3dnow put_pixels16_mmx 00213 #define put_pixels8_3dnow put_pixels8_mmx 00214 #define put_pixels4_3dnow put_pixels4_mmx 00215 #define put_no_rnd_pixels16_3dnow put_no_rnd_pixels16_mmx 00216 #define put_no_rnd_pixels8_3dnow put_no_rnd_pixels8_mmx 00217 00218 /***********************************/ 00219 /* standard MMX */ 00220 00221 void put_pixels_clamped_mmx(const DCTELEM *block, uint8_t *pixels, int line_size) 00222 { 00223 const DCTELEM *p; 00224 uint8_t *pix; 00225 00226 /* read the pixels */ 00227 p = block; 00228 pix = pixels; 00229 /* unrolled loop */ 00230 __asm__ volatile( 00231 "movq %3, %%mm0 \n\t" 00232 "movq 8%3, %%mm1 \n\t" 00233 "movq 16%3, %%mm2 \n\t" 00234 "movq 24%3, %%mm3 \n\t" 00235 "movq 32%3, %%mm4 \n\t" 00236 "movq 40%3, %%mm5 \n\t" 00237 "movq 48%3, %%mm6 \n\t" 00238 "movq 56%3, %%mm7 \n\t" 00239 "packuswb %%mm1, %%mm0 \n\t" 00240 "packuswb %%mm3, %%mm2 \n\t" 00241 "packuswb %%mm5, %%mm4 \n\t" 00242 "packuswb %%mm7, %%mm6 \n\t" 00243 "movq %%mm0, (%0) \n\t" 00244 "movq %%mm2, (%0, %1) \n\t" 00245 "movq %%mm4, (%0, %1, 2) \n\t" 00246 "movq %%mm6, (%0, %2) \n\t" 00247 ::"r" (pix), "r" ((x86_reg)line_size), "r" ((x86_reg)line_size*3), "m"(*p) 00248 :"memory"); 00249 pix += line_size*4; 00250 p += 32; 00251 00252 // if here would be an exact copy of the code above 00253 // compiler would generate some very strange code 00254 // thus using "r" 00255 __asm__ volatile( 00256 "movq (%3), %%mm0 \n\t" 00257 "movq 8(%3), %%mm1 \n\t" 00258 "movq 16(%3), %%mm2 \n\t" 00259 "movq 24(%3), %%mm3 \n\t" 00260 "movq 32(%3), %%mm4 \n\t" 00261 "movq 40(%3), %%mm5 \n\t" 00262 "movq 48(%3), %%mm6 \n\t" 00263 "movq 56(%3), %%mm7 \n\t" 00264 "packuswb %%mm1, %%mm0 \n\t" 00265 "packuswb %%mm3, %%mm2 \n\t" 00266 "packuswb %%mm5, %%mm4 \n\t" 00267 "packuswb %%mm7, %%mm6 \n\t" 00268 "movq %%mm0, (%0) \n\t" 00269 "movq %%mm2, (%0, %1) \n\t" 00270 "movq %%mm4, (%0, %1, 2) \n\t" 00271 "movq %%mm6, (%0, %2) \n\t" 00272 ::"r" (pix), "r" ((x86_reg)line_size), "r" ((x86_reg)line_size*3), "r"(p) 00273 :"memory"); 00274 } 00275 00276 static DECLARE_ALIGNED_8(const unsigned char, vector128[8]) = 00277 { 0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80 }; 00278 00279 void put_signed_pixels_clamped_mmx(const DCTELEM *block, uint8_t *pixels, int line_size) 00280 { 00281 int i; 00282 00283 movq_m2r(*vector128, mm1); 00284 for (i = 0; i < 8; i++) { 00285 movq_m2r(*(block), mm0); 00286 packsswb_m2r(*(block + 4), mm0); 00287 block += 8; 00288 paddb_r2r(mm1, mm0); 00289 movq_r2m(mm0, *pixels); 00290 pixels += line_size; 00291 } 00292 } 00293 00294 void add_pixels_clamped_mmx(const DCTELEM *block, uint8_t *pixels, int line_size) 00295 { 00296 const DCTELEM *p; 00297 uint8_t *pix; 00298 int i; 00299 00300 /* read the pixels */ 00301 p = block; 00302 pix = pixels; 00303 MOVQ_ZERO(mm7); 00304 i = 4; 00305 do { 00306 __asm__ volatile( 00307 "movq (%2), %%mm0 \n\t" 00308 "movq 8(%2), %%mm1 \n\t" 00309 "movq 16(%2), %%mm2 \n\t" 00310 "movq 24(%2), %%mm3 \n\t" 00311 "movq %0, %%mm4 \n\t" 00312 "movq %1, %%mm6 \n\t" 00313 "movq %%mm4, %%mm5 \n\t" 00314 "punpcklbw %%mm7, %%mm4 \n\t" 00315 "punpckhbw %%mm7, %%mm5 \n\t" 00316 "paddsw %%mm4, %%mm0 \n\t" 00317 "paddsw %%mm5, %%mm1 \n\t" 00318 "movq %%mm6, %%mm5 \n\t" 00319 "punpcklbw %%mm7, %%mm6 \n\t" 00320 "punpckhbw %%mm7, %%mm5 \n\t" 00321 "paddsw %%mm6, %%mm2 \n\t" 00322 "paddsw %%mm5, %%mm3 \n\t" 00323 "packuswb %%mm1, %%mm0 \n\t" 00324 "packuswb %%mm3, %%mm2 \n\t" 00325 "movq %%mm0, %0 \n\t" 00326 "movq %%mm2, %1 \n\t" 00327 :"+m"(*pix), "+m"(*(pix+line_size)) 00328 :"r"(p) 00329 :"memory"); 00330 pix += line_size*2; 00331 p += 16; 00332 } while (--i); 00333 } 00334 00335 static void put_pixels4_mmx(uint8_t *block, const uint8_t *pixels, int line_size, int h) 00336 { 00337 __asm__ volatile( 00338 "lea (%3, %3), %%"REG_a" \n\t" 00339 ASMALIGN(3) 00340 "1: \n\t" 00341 "movd (%1), %%mm0 \n\t" 00342 "movd (%1, %3), %%mm1 \n\t" 00343 "movd %%mm0, (%2) \n\t" 00344 "movd %%mm1, (%2, %3) \n\t" 00345 "add %%"REG_a", %1 \n\t" 00346 "add %%"REG_a", %2 \n\t" 00347 "movd (%1), %%mm0 \n\t" 00348 "movd (%1, %3), %%mm1 \n\t" 00349 "movd %%mm0, (%2) \n\t" 00350 "movd %%mm1, (%2, %3) \n\t" 00351 "add %%"REG_a", %1 \n\t" 00352 "add %%"REG_a", %2 \n\t" 00353 "subl 4,ドル %0 \n\t" 00354 "jnz 1b \n\t" 00355 : "+g"(h), "+r" (pixels), "+r" (block) 00356 : "r"((x86_reg)line_size) 00357 : "%"REG_a, "memory" 00358 ); 00359 } 00360 00361 static void put_pixels8_mmx(uint8_t *block, const uint8_t *pixels, int line_size, int h) 00362 { 00363 __asm__ volatile( 00364 "lea (%3, %3), %%"REG_a" \n\t" 00365 ASMALIGN(3) 00366 "1: \n\t" 00367 "movq (%1), %%mm0 \n\t" 00368 "movq (%1, %3), %%mm1 \n\t" 00369 "movq %%mm0, (%2) \n\t" 00370 "movq %%mm1, (%2, %3) \n\t" 00371 "add %%"REG_a", %1 \n\t" 00372 "add %%"REG_a", %2 \n\t" 00373 "movq (%1), %%mm0 \n\t" 00374 "movq (%1, %3), %%mm1 \n\t" 00375 "movq %%mm0, (%2) \n\t" 00376 "movq %%mm1, (%2, %3) \n\t" 00377 "add %%"REG_a", %1 \n\t" 00378 "add %%"REG_a", %2 \n\t" 00379 "subl 4,ドル %0 \n\t" 00380 "jnz 1b \n\t" 00381 : "+g"(h), "+r" (pixels), "+r" (block) 00382 : "r"((x86_reg)line_size) 00383 : "%"REG_a, "memory" 00384 ); 00385 } 00386 00387 static void put_pixels16_mmx(uint8_t *block, const uint8_t *pixels, int line_size, int h) 00388 { 00389 __asm__ volatile( 00390 "lea (%3, %3), %%"REG_a" \n\t" 00391 ASMALIGN(3) 00392 "1: \n\t" 00393 "movq (%1), %%mm0 \n\t" 00394 "movq 8(%1), %%mm4 \n\t" 00395 "movq (%1, %3), %%mm1 \n\t" 00396 "movq 8(%1, %3), %%mm5 \n\t" 00397 "movq %%mm0, (%2) \n\t" 00398 "movq %%mm4, 8(%2) \n\t" 00399 "movq %%mm1, (%2, %3) \n\t" 00400 "movq %%mm5, 8(%2, %3) \n\t" 00401 "add %%"REG_a", %1 \n\t" 00402 "add %%"REG_a", %2 \n\t" 00403 "movq (%1), %%mm0 \n\t" 00404 "movq 8(%1), %%mm4 \n\t" 00405 "movq (%1, %3), %%mm1 \n\t" 00406 "movq 8(%1, %3), %%mm5 \n\t" 00407 "movq %%mm0, (%2) \n\t" 00408 "movq %%mm4, 8(%2) \n\t" 00409 "movq %%mm1, (%2, %3) \n\t" 00410 "movq %%mm5, 8(%2, %3) \n\t" 00411 "add %%"REG_a", %1 \n\t" 00412 "add %%"REG_a", %2 \n\t" 00413 "subl 4,ドル %0 \n\t" 00414 "jnz 1b \n\t" 00415 : "+g"(h), "+r" (pixels), "+r" (block) 00416 : "r"((x86_reg)line_size) 00417 : "%"REG_a, "memory" 00418 ); 00419 } 00420 00421 static void put_pixels16_sse2(uint8_t *block, const uint8_t *pixels, int line_size, int h) 00422 { 00423 __asm__ volatile( 00424 "1: \n\t" 00425 "movdqu (%1), %%xmm0 \n\t" 00426 "movdqu (%1,%3), %%xmm1 \n\t" 00427 "movdqu (%1,%3,2), %%xmm2 \n\t" 00428 "movdqu (%1,%4), %%xmm3 \n\t" 00429 "movdqa %%xmm0, (%2) \n\t" 00430 "movdqa %%xmm1, (%2,%3) \n\t" 00431 "movdqa %%xmm2, (%2,%3,2) \n\t" 00432 "movdqa %%xmm3, (%2,%4) \n\t" 00433 "subl 4,ドル %0 \n\t" 00434 "lea (%1,%3,4), %1 \n\t" 00435 "lea (%2,%3,4), %2 \n\t" 00436 "jnz 1b \n\t" 00437 : "+g"(h), "+r" (pixels), "+r" (block) 00438 : "r"((x86_reg)line_size), "r"((x86_reg)3L*line_size) 00439 : "memory" 00440 ); 00441 } 00442 00443 static void avg_pixels16_sse2(uint8_t *block, const uint8_t *pixels, int line_size, int h) 00444 { 00445 __asm__ volatile( 00446 "1: \n\t" 00447 "movdqu (%1), %%xmm0 \n\t" 00448 "movdqu (%1,%3), %%xmm1 \n\t" 00449 "movdqu (%1,%3,2), %%xmm2 \n\t" 00450 "movdqu (%1,%4), %%xmm3 \n\t" 00451 "pavgb (%2), %%xmm0 \n\t" 00452 "pavgb (%2,%3), %%xmm1 \n\t" 00453 "pavgb (%2,%3,2), %%xmm2 \n\t" 00454 "pavgb (%2,%4), %%xmm3 \n\t" 00455 "movdqa %%xmm0, (%2) \n\t" 00456 "movdqa %%xmm1, (%2,%3) \n\t" 00457 "movdqa %%xmm2, (%2,%3,2) \n\t" 00458 "movdqa %%xmm3, (%2,%4) \n\t" 00459 "subl 4,ドル %0 \n\t" 00460 "lea (%1,%3,4), %1 \n\t" 00461 "lea (%2,%3,4), %2 \n\t" 00462 "jnz 1b \n\t" 00463 : "+g"(h), "+r" (pixels), "+r" (block) 00464 : "r"((x86_reg)line_size), "r"((x86_reg)3L*line_size) 00465 : "memory" 00466 ); 00467 } 00468 00469 #define CLEAR_BLOCKS(name,n) \ 00470 static void name(DCTELEM *blocks)\ 00471 {\ 00472 __asm__ volatile(\ 00473 "pxor %%mm7, %%mm7 \n\t"\ 00474 "mov %1, %%"REG_a" \n\t"\ 00475 "1: \n\t"\ 00476 "movq %%mm7, (%0, %%"REG_a") \n\t"\ 00477 "movq %%mm7, 8(%0, %%"REG_a") \n\t"\ 00478 "movq %%mm7, 16(%0, %%"REG_a") \n\t"\ 00479 "movq %%mm7, 24(%0, %%"REG_a") \n\t"\ 00480 "add 32,ドル %%"REG_a" \n\t"\ 00481 " js 1b \n\t"\ 00482 : : "r" (((uint8_t *)blocks)+128*n),\ 00483 "i" (-128*n)\ 00484 : "%"REG_a\ 00485 );\ 00486 } 00487 CLEAR_BLOCKS(clear_blocks_mmx, 6) 00488 CLEAR_BLOCKS(clear_block_mmx, 1) 00489 00490 static void clear_block_sse(DCTELEM *block) 00491 { 00492 __asm__ volatile( 00493 "xorps %%xmm0, %%xmm0 \n" 00494 "movaps %%xmm0, (%0) \n" 00495 "movaps %%xmm0, 16(%0) \n" 00496 "movaps %%xmm0, 32(%0) \n" 00497 "movaps %%xmm0, 48(%0) \n" 00498 "movaps %%xmm0, 64(%0) \n" 00499 "movaps %%xmm0, 80(%0) \n" 00500 "movaps %%xmm0, 96(%0) \n" 00501 "movaps %%xmm0, 112(%0) \n" 00502 :: "r"(block) 00503 : "memory" 00504 ); 00505 } 00506 00507 static void add_bytes_mmx(uint8_t *dst, uint8_t *src, int w){ 00508 x86_reg i=0; 00509 __asm__ volatile( 00510 "jmp 2f \n\t" 00511 "1: \n\t" 00512 "movq (%1, %0), %%mm0 \n\t" 00513 "movq (%2, %0), %%mm1 \n\t" 00514 "paddb %%mm0, %%mm1 \n\t" 00515 "movq %%mm1, (%2, %0) \n\t" 00516 "movq 8(%1, %0), %%mm0 \n\t" 00517 "movq 8(%2, %0), %%mm1 \n\t" 00518 "paddb %%mm0, %%mm1 \n\t" 00519 "movq %%mm1, 8(%2, %0) \n\t" 00520 "add 16,ドル %0 \n\t" 00521 "2: \n\t" 00522 "cmp %3, %0 \n\t" 00523 " js 1b \n\t" 00524 : "+r" (i) 00525 : "r"(src), "r"(dst), "r"((x86_reg)w-15) 00526 ); 00527 for(; i<w; i++) 00528 dst[i+0] += src[i+0]; 00529 } 00530 00531 static void add_bytes_l2_mmx(uint8_t *dst, uint8_t *src1, uint8_t *src2, int w){ 00532 x86_reg i=0; 00533 __asm__ volatile( 00534 "jmp 2f \n\t" 00535 "1: \n\t" 00536 "movq (%2, %0), %%mm0 \n\t" 00537 "movq 8(%2, %0), %%mm1 \n\t" 00538 "paddb (%3, %0), %%mm0 \n\t" 00539 "paddb 8(%3, %0), %%mm1 \n\t" 00540 "movq %%mm0, (%1, %0) \n\t" 00541 "movq %%mm1, 8(%1, %0) \n\t" 00542 "add 16,ドル %0 \n\t" 00543 "2: \n\t" 00544 "cmp %4, %0 \n\t" 00545 " js 1b \n\t" 00546 : "+r" (i) 00547 : "r"(dst), "r"(src1), "r"(src2), "r"((x86_reg)w-15) 00548 ); 00549 for(; i<w; i++) 00550 dst[i] = src1[i] + src2[i]; 00551 } 00552 00553 #if HAVE_7REGS && HAVE_TEN_OPERANDS 00554 static void add_hfyu_median_prediction_cmov(uint8_t *dst, uint8_t *top, uint8_t *diff, int w, int *left, int *left_top) { 00555 x86_reg w2 = -w; 00556 x86_reg x; 00557 int l = *left & 0xff; 00558 int tl = *left_top & 0xff; 00559 int t; 00560 __asm__ volatile( 00561 "mov %7, %3 \n" 00562 "1: \n" 00563 "movzx (%3,%4), %2 \n" 00564 "mov %2, %k3 \n" 00565 "sub %b1, %b3 \n" 00566 "add %b0, %b3 \n" 00567 "mov %2, %1 \n" 00568 "cmp %0, %2 \n" 00569 "cmovg %0, %2 \n" 00570 "cmovg %1, %0 \n" 00571 "cmp %k3, %0 \n" 00572 "cmovg %k3, %0 \n" 00573 "mov %7, %3 \n" 00574 "cmp %2, %0 \n" 00575 "cmovl %2, %0 \n" 00576 "add (%6,%4), %b0 \n" 00577 "mov %b0, (%5,%4) \n" 00578 "inc %4 \n" 00579 "jl 1b \n" 00580 :"+&q"(l), "+&q"(tl), "=&r"(t), "=&q"(x), "+&r"(w2) 00581 :"r"(dst+w), "r"(diff+w), "rm"(top+w) 00582 ); 00583 *left = l; 00584 *left_top = tl; 00585 } 00586 #endif 00587 00588 #define H263_LOOP_FILTER \ 00589 "pxor %%mm7, %%mm7 \n\t"\ 00590 "movq %0, %%mm0 \n\t"\ 00591 "movq %0, %%mm1 \n\t"\ 00592 "movq %3, %%mm2 \n\t"\ 00593 "movq %3, %%mm3 \n\t"\ 00594 "punpcklbw %%mm7, %%mm0 \n\t"\ 00595 "punpckhbw %%mm7, %%mm1 \n\t"\ 00596 "punpcklbw %%mm7, %%mm2 \n\t"\ 00597 "punpckhbw %%mm7, %%mm3 \n\t"\ 00598 "psubw %%mm2, %%mm0 \n\t"\ 00599 "psubw %%mm3, %%mm1 \n\t"\ 00600 "movq %1, %%mm2 \n\t"\ 00601 "movq %1, %%mm3 \n\t"\ 00602 "movq %2, %%mm4 \n\t"\ 00603 "movq %2, %%mm5 \n\t"\ 00604 "punpcklbw %%mm7, %%mm2 \n\t"\ 00605 "punpckhbw %%mm7, %%mm3 \n\t"\ 00606 "punpcklbw %%mm7, %%mm4 \n\t"\ 00607 "punpckhbw %%mm7, %%mm5 \n\t"\ 00608 "psubw %%mm2, %%mm4 \n\t"\ 00609 "psubw %%mm3, %%mm5 \n\t"\ 00610 "psllw 2,ドル %%mm4 \n\t"\ 00611 "psllw 2,ドル %%mm5 \n\t"\ 00612 "paddw %%mm0, %%mm4 \n\t"\ 00613 "paddw %%mm1, %%mm5 \n\t"\ 00614 "pxor %%mm6, %%mm6 \n\t"\ 00615 "pcmpgtw %%mm4, %%mm6 \n\t"\ 00616 "pcmpgtw %%mm5, %%mm7 \n\t"\ 00617 "pxor %%mm6, %%mm4 \n\t"\ 00618 "pxor %%mm7, %%mm5 \n\t"\ 00619 "psubw %%mm6, %%mm4 \n\t"\ 00620 "psubw %%mm7, %%mm5 \n\t"\ 00621 "psrlw 3,ドル %%mm4 \n\t"\ 00622 "psrlw 3,ドル %%mm5 \n\t"\ 00623 "packuswb %%mm5, %%mm4 \n\t"\ 00624 "packsswb %%mm7, %%mm6 \n\t"\ 00625 "pxor %%mm7, %%mm7 \n\t"\ 00626 "movd %4, %%mm2 \n\t"\ 00627 "punpcklbw %%mm2, %%mm2 \n\t"\ 00628 "punpcklbw %%mm2, %%mm2 \n\t"\ 00629 "punpcklbw %%mm2, %%mm2 \n\t"\ 00630 "psubusb %%mm4, %%mm2 \n\t"\ 00631 "movq %%mm2, %%mm3 \n\t"\ 00632 "psubusb %%mm4, %%mm3 \n\t"\ 00633 "psubb %%mm3, %%mm2 \n\t"\ 00634 "movq %1, %%mm3 \n\t"\ 00635 "movq %2, %%mm4 \n\t"\ 00636 "pxor %%mm6, %%mm3 \n\t"\ 00637 "pxor %%mm6, %%mm4 \n\t"\ 00638 "paddusb %%mm2, %%mm3 \n\t"\ 00639 "psubusb %%mm2, %%mm4 \n\t"\ 00640 "pxor %%mm6, %%mm3 \n\t"\ 00641 "pxor %%mm6, %%mm4 \n\t"\ 00642 "paddusb %%mm2, %%mm2 \n\t"\ 00643 "packsswb %%mm1, %%mm0 \n\t"\ 00644 "pcmpgtb %%mm0, %%mm7 \n\t"\ 00645 "pxor %%mm7, %%mm0 \n\t"\ 00646 "psubb %%mm7, %%mm0 \n\t"\ 00647 "movq %%mm0, %%mm1 \n\t"\ 00648 "psubusb %%mm2, %%mm0 \n\t"\ 00649 "psubb %%mm0, %%mm1 \n\t"\ 00650 "pand %5, %%mm1 \n\t"\ 00651 "psrlw 2,ドル %%mm1 \n\t"\ 00652 "pxor %%mm7, %%mm1 \n\t"\ 00653 "psubb %%mm7, %%mm1 \n\t"\ 00654 "movq %0, %%mm5 \n\t"\ 00655 "movq %3, %%mm6 \n\t"\ 00656 "psubb %%mm1, %%mm5 \n\t"\ 00657 "paddb %%mm1, %%mm6 \n\t" 00658 00659 static void h263_v_loop_filter_mmx(uint8_t *src, int stride, int qscale){ 00660 if(CONFIG_ANY_H263) { 00661 const int strength= ff_h263_loop_filter_strength[qscale]; 00662 00663 __asm__ volatile( 00664 00665 H263_LOOP_FILTER 00666 00667 "movq %%mm3, %1 \n\t" 00668 "movq %%mm4, %2 \n\t" 00669 "movq %%mm5, %0 \n\t" 00670 "movq %%mm6, %3 \n\t" 00671 : "+m" (*(uint64_t*)(src - 2*stride)), 00672 "+m" (*(uint64_t*)(src - 1*stride)), 00673 "+m" (*(uint64_t*)(src + 0*stride)), 00674 "+m" (*(uint64_t*)(src + 1*stride)) 00675 : "g" (2*strength), "m"(ff_pb_FC) 00676 ); 00677 } 00678 } 00679 00680 static inline void transpose4x4(uint8_t *dst, uint8_t *src, int dst_stride, int src_stride){ 00681 __asm__ volatile( //FIXME could save 1 instruction if done as 8x4 ... 00682 "movd %4, %%mm0 \n\t" 00683 "movd %5, %%mm1 \n\t" 00684 "movd %6, %%mm2 \n\t" 00685 "movd %7, %%mm3 \n\t" 00686 "punpcklbw %%mm1, %%mm0 \n\t" 00687 "punpcklbw %%mm3, %%mm2 \n\t" 00688 "movq %%mm0, %%mm1 \n\t" 00689 "punpcklwd %%mm2, %%mm0 \n\t" 00690 "punpckhwd %%mm2, %%mm1 \n\t" 00691 "movd %%mm0, %0 \n\t" 00692 "punpckhdq %%mm0, %%mm0 \n\t" 00693 "movd %%mm0, %1 \n\t" 00694 "movd %%mm1, %2 \n\t" 00695 "punpckhdq %%mm1, %%mm1 \n\t" 00696 "movd %%mm1, %3 \n\t" 00697 00698 : "=m" (*(uint32_t*)(dst + 0*dst_stride)), 00699 "=m" (*(uint32_t*)(dst + 1*dst_stride)), 00700 "=m" (*(uint32_t*)(dst + 2*dst_stride)), 00701 "=m" (*(uint32_t*)(dst + 3*dst_stride)) 00702 : "m" (*(uint32_t*)(src + 0*src_stride)), 00703 "m" (*(uint32_t*)(src + 1*src_stride)), 00704 "m" (*(uint32_t*)(src + 2*src_stride)), 00705 "m" (*(uint32_t*)(src + 3*src_stride)) 00706 ); 00707 } 00708 00709 static void h263_h_loop_filter_mmx(uint8_t *src, int stride, int qscale){ 00710 if(CONFIG_ANY_H263) { 00711 const int strength= ff_h263_loop_filter_strength[qscale]; 00712 DECLARE_ALIGNED(8, uint64_t, temp[4]); 00713 uint8_t *btemp= (uint8_t*)temp; 00714 00715 src -= 2; 00716 00717 transpose4x4(btemp , src , 8, stride); 00718 transpose4x4(btemp+4, src + 4*stride, 8, stride); 00719 __asm__ volatile( 00720 H263_LOOP_FILTER // 5 3 4 6 00721 00722 : "+m" (temp[0]), 00723 "+m" (temp[1]), 00724 "+m" (temp[2]), 00725 "+m" (temp[3]) 00726 : "g" (2*strength), "m"(ff_pb_FC) 00727 ); 00728 00729 __asm__ volatile( 00730 "movq %%mm5, %%mm1 \n\t" 00731 "movq %%mm4, %%mm0 \n\t" 00732 "punpcklbw %%mm3, %%mm5 \n\t" 00733 "punpcklbw %%mm6, %%mm4 \n\t" 00734 "punpckhbw %%mm3, %%mm1 \n\t" 00735 "punpckhbw %%mm6, %%mm0 \n\t" 00736 "movq %%mm5, %%mm3 \n\t" 00737 "movq %%mm1, %%mm6 \n\t" 00738 "punpcklwd %%mm4, %%mm5 \n\t" 00739 "punpcklwd %%mm0, %%mm1 \n\t" 00740 "punpckhwd %%mm4, %%mm3 \n\t" 00741 "punpckhwd %%mm0, %%mm6 \n\t" 00742 "movd %%mm5, (%0) \n\t" 00743 "punpckhdq %%mm5, %%mm5 \n\t" 00744 "movd %%mm5, (%0,%2) \n\t" 00745 "movd %%mm3, (%0,%2,2) \n\t" 00746 "punpckhdq %%mm3, %%mm3 \n\t" 00747 "movd %%mm3, (%0,%3) \n\t" 00748 "movd %%mm1, (%1) \n\t" 00749 "punpckhdq %%mm1, %%mm1 \n\t" 00750 "movd %%mm1, (%1,%2) \n\t" 00751 "movd %%mm6, (%1,%2,2) \n\t" 00752 "punpckhdq %%mm6, %%mm6 \n\t" 00753 "movd %%mm6, (%1,%3) \n\t" 00754 :: "r" (src), 00755 "r" (src + 4*stride), 00756 "r" ((x86_reg) stride ), 00757 "r" ((x86_reg)(3*stride)) 00758 ); 00759 } 00760 } 00761 00762 /* draw the edges of width 'w' of an image of size width, height 00763 this mmx version can only handle w==8 || w==16 */ 00764 static void draw_edges_mmx(uint8_t *buf, int wrap, int width, int height, int w) 00765 { 00766 uint8_t *ptr, *last_line; 00767 int i; 00768 00769 last_line = buf + (height - 1) * wrap; 00770 /* left and right */ 00771 ptr = buf; 00772 if(w==8) 00773 { 00774 __asm__ volatile( 00775 "1: \n\t" 00776 "movd (%0), %%mm0 \n\t" 00777 "punpcklbw %%mm0, %%mm0 \n\t" 00778 "punpcklwd %%mm0, %%mm0 \n\t" 00779 "punpckldq %%mm0, %%mm0 \n\t" 00780 "movq %%mm0, -8(%0) \n\t" 00781 "movq -8(%0, %2), %%mm1 \n\t" 00782 "punpckhbw %%mm1, %%mm1 \n\t" 00783 "punpckhwd %%mm1, %%mm1 \n\t" 00784 "punpckhdq %%mm1, %%mm1 \n\t" 00785 "movq %%mm1, (%0, %2) \n\t" 00786 "add %1, %0 \n\t" 00787 "cmp %3, %0 \n\t" 00788 " jb 1b \n\t" 00789 : "+r" (ptr) 00790 : "r" ((x86_reg)wrap), "r" ((x86_reg)width), "r" (ptr + wrap*height) 00791 ); 00792 } 00793 else 00794 { 00795 __asm__ volatile( 00796 "1: \n\t" 00797 "movd (%0), %%mm0 \n\t" 00798 "punpcklbw %%mm0, %%mm0 \n\t" 00799 "punpcklwd %%mm0, %%mm0 \n\t" 00800 "punpckldq %%mm0, %%mm0 \n\t" 00801 "movq %%mm0, -8(%0) \n\t" 00802 "movq %%mm0, -16(%0) \n\t" 00803 "movq -8(%0, %2), %%mm1 \n\t" 00804 "punpckhbw %%mm1, %%mm1 \n\t" 00805 "punpckhwd %%mm1, %%mm1 \n\t" 00806 "punpckhdq %%mm1, %%mm1 \n\t" 00807 "movq %%mm1, (%0, %2) \n\t" 00808 "movq %%mm1, 8(%0, %2) \n\t" 00809 "add %1, %0 \n\t" 00810 "cmp %3, %0 \n\t" 00811 " jb 1b \n\t" 00812 : "+r" (ptr) 00813 : "r" ((x86_reg)wrap), "r" ((x86_reg)width), "r" (ptr + wrap*height) 00814 ); 00815 } 00816 00817 for(i=0;i<w;i+=4) { 00818 /* top and bottom (and hopefully also the corners) */ 00819 ptr= buf - (i + 1) * wrap - w; 00820 __asm__ volatile( 00821 "1: \n\t" 00822 "movq (%1, %0), %%mm0 \n\t" 00823 "movq %%mm0, (%0) \n\t" 00824 "movq %%mm0, (%0, %2) \n\t" 00825 "movq %%mm0, (%0, %2, 2) \n\t" 00826 "movq %%mm0, (%0, %3) \n\t" 00827 "add 8,ドル %0 \n\t" 00828 "cmp %4, %0 \n\t" 00829 " jb 1b \n\t" 00830 : "+r" (ptr) 00831 : "r" ((x86_reg)buf - (x86_reg)ptr - w), "r" ((x86_reg)-wrap), "r" ((x86_reg)-wrap*3), "r" (ptr+width+2*w) 00832 ); 00833 ptr= last_line + (i + 1) * wrap - w; 00834 __asm__ volatile( 00835 "1: \n\t" 00836 "movq (%1, %0), %%mm0 \n\t" 00837 "movq %%mm0, (%0) \n\t" 00838 "movq %%mm0, (%0, %2) \n\t" 00839 "movq %%mm0, (%0, %2, 2) \n\t" 00840 "movq %%mm0, (%0, %3) \n\t" 00841 "add 8,ドル %0 \n\t" 00842 "cmp %4, %0 \n\t" 00843 " jb 1b \n\t" 00844 : "+r" (ptr) 00845 : "r" ((x86_reg)last_line - (x86_reg)ptr - w), "r" ((x86_reg)wrap), "r" ((x86_reg)wrap*3), "r" (ptr+width+2*w) 00846 ); 00847 } 00848 } 00849 00850 #define PAETH(cpu, abs3)\ 00851 static void add_png_paeth_prediction_##cpu(uint8_t *dst, uint8_t *src, uint8_t *top, int w, int bpp)\ 00852 {\ 00853 x86_reg i = -bpp;\ 00854 x86_reg end = w-3;\ 00855 __asm__ volatile(\ 00856 "pxor %%mm7, %%mm7 \n"\ 00857 "movd (%1,%0), %%mm0 \n"\ 00858 "movd (%2,%0), %%mm1 \n"\ 00859 "punpcklbw %%mm7, %%mm0 \n"\ 00860 "punpcklbw %%mm7, %%mm1 \n"\ 00861 "add %4, %0 \n"\ 00862 "1: \n"\ 00863 "movq %%mm1, %%mm2 \n"\ 00864 "movd (%2,%0), %%mm1 \n"\ 00865 "movq %%mm2, %%mm3 \n"\ 00866 "punpcklbw %%mm7, %%mm1 \n"\ 00867 "movq %%mm2, %%mm4 \n"\ 00868 "psubw %%mm1, %%mm3 \n"\ 00869 "psubw %%mm0, %%mm4 \n"\ 00870 "movq %%mm3, %%mm5 \n"\ 00871 "paddw %%mm4, %%mm5 \n"\ 00872 abs3\ 00873 "movq %%mm4, %%mm6 \n"\ 00874 "pminsw %%mm5, %%mm6 \n"\ 00875 "pcmpgtw %%mm6, %%mm3 \n"\ 00876 "pcmpgtw %%mm5, %%mm4 \n"\ 00877 "movq %%mm4, %%mm6 \n"\ 00878 "pand %%mm3, %%mm4 \n"\ 00879 "pandn %%mm3, %%mm6 \n"\ 00880 "pandn %%mm0, %%mm3 \n"\ 00881 "movd (%3,%0), %%mm0 \n"\ 00882 "pand %%mm1, %%mm6 \n"\ 00883 "pand %%mm4, %%mm2 \n"\ 00884 "punpcklbw %%mm7, %%mm0 \n"\ 00885 "movq %6, %%mm5 \n"\ 00886 "paddw %%mm6, %%mm0 \n"\ 00887 "paddw %%mm2, %%mm3 \n"\ 00888 "paddw %%mm3, %%mm0 \n"\ 00889 "pand %%mm5, %%mm0 \n"\ 00890 "movq %%mm0, %%mm3 \n"\ 00891 "packuswb %%mm3, %%mm3 \n"\ 00892 "movd %%mm3, (%1,%0) \n"\ 00893 "add %4, %0 \n"\ 00894 "cmp %5, %0 \n"\ 00895 "jle 1b \n"\ 00896 :"+r"(i)\ 00897 :"r"(dst), "r"(top), "r"(src), "r"((x86_reg)bpp), "g"(end),\ 00898 "m"(ff_pw_255)\ 00899 :"memory"\ 00900 );\ 00901 } 00902 00903 #define ABS3_MMX2\ 00904 "psubw %%mm5, %%mm7 \n"\ 00905 "pmaxsw %%mm7, %%mm5 \n"\ 00906 "pxor %%mm6, %%mm6 \n"\ 00907 "pxor %%mm7, %%mm7 \n"\ 00908 "psubw %%mm3, %%mm6 \n"\ 00909 "psubw %%mm4, %%mm7 \n"\ 00910 "pmaxsw %%mm6, %%mm3 \n"\ 00911 "pmaxsw %%mm7, %%mm4 \n"\ 00912 "pxor %%mm7, %%mm7 \n" 00913 00914 #define ABS3_SSSE3\ 00915 "pabsw %%mm3, %%mm3 \n"\ 00916 "pabsw %%mm4, %%mm4 \n"\ 00917 "pabsw %%mm5, %%mm5 \n" 00918 00919 PAETH(mmx2, ABS3_MMX2) 00920 #if HAVE_SSSE3 00921 PAETH(ssse3, ABS3_SSSE3) 00922 #endif 00923 00924 #define QPEL_V_LOW(m3,m4,m5,m6, pw_20, pw_3, rnd, in0, in1, in2, in7, out, OP)\ 00925 "paddw " #m4 ", " #m3 " \n\t" /* x1 */\ 00926 "movq "MANGLE(ff_pw_20)", %%mm4 \n\t" /* 20 */\ 00927 "pmullw " #m3 ", %%mm4 \n\t" /* 20x1 */\ 00928 "movq "#in7", " #m3 " \n\t" /* d */\ 00929 "movq "#in0", %%mm5 \n\t" /* D */\ 00930 "paddw " #m3 ", %%mm5 \n\t" /* x4 */\ 00931 "psubw %%mm5, %%mm4 \n\t" /* 20x1 - x4 */\ 00932 "movq "#in1", %%mm5 \n\t" /* C */\ 00933 "movq "#in2", %%mm6 \n\t" /* B */\ 00934 "paddw " #m6 ", %%mm5 \n\t" /* x3 */\ 00935 "paddw " #m5 ", %%mm6 \n\t" /* x2 */\ 00936 "paddw %%mm6, %%mm6 \n\t" /* 2x2 */\ 00937 "psubw %%mm6, %%mm5 \n\t" /* -2x2 + x3 */\ 00938 "pmullw "MANGLE(ff_pw_3)", %%mm5 \n\t" /* -6x2 + 3x3 */\ 00939 "paddw " #rnd ", %%mm4 \n\t" /* x2 */\ 00940 "paddw %%mm4, %%mm5 \n\t" /* 20x1 - 6x2 + 3x3 - x4 */\ 00941 "psraw 5,ドル %%mm5 \n\t"\ 00942 "packuswb %%mm5, %%mm5 \n\t"\ 00943 OP(%%mm5, out, %%mm7, d) 00944 00945 #define QPEL_BASE(OPNAME, ROUNDER, RND, OP_MMX2, OP_3DNOW)\ 00946 static void OPNAME ## mpeg4_qpel16_h_lowpass_mmx2(uint8_t *dst, uint8_t *src, int dstStride, int srcStride, int h){\ 00947 uint64_t temp;\ 00948 \ 00949 __asm__ volatile(\ 00950 "pxor %%mm7, %%mm7 \n\t"\ 00951 "1: \n\t"\ 00952 "movq (%0), %%mm0 \n\t" /* ABCDEFGH */\ 00953 "movq %%mm0, %%mm1 \n\t" /* ABCDEFGH */\ 00954 "movq %%mm0, %%mm2 \n\t" /* ABCDEFGH */\ 00955 "punpcklbw %%mm7, %%mm0 \n\t" /* 0A0B0C0D */\ 00956 "punpckhbw %%mm7, %%mm1 \n\t" /* 0E0F0G0H */\ 00957 "pshufw 0ドルx90, %%mm0, %%mm5 \n\t" /* 0A0A0B0C */\ 00958 "pshufw 0ドルx41, %%mm0, %%mm6 \n\t" /* 0B0A0A0B */\ 00959 "movq %%mm2, %%mm3 \n\t" /* ABCDEFGH */\ 00960 "movq %%mm2, %%mm4 \n\t" /* ABCDEFGH */\ 00961 "psllq 8,ドル %%mm2 \n\t" /* 0ABCDEFG */\ 00962 "psllq 16,ドル %%mm3 \n\t" /* 00ABCDEF */\ 00963 "psllq 24,ドル %%mm4 \n\t" /* 000ABCDE */\ 00964 "punpckhbw %%mm7, %%mm2 \n\t" /* 0D0E0F0G */\ 00965 "punpckhbw %%mm7, %%mm3 \n\t" /* 0C0D0E0F */\ 00966 "punpckhbw %%mm7, %%mm4 \n\t" /* 0B0C0D0E */\ 00967 "paddw %%mm3, %%mm5 \n\t" /* b */\ 00968 "paddw %%mm2, %%mm6 \n\t" /* c */\ 00969 "paddw %%mm5, %%mm5 \n\t" /* 2b */\ 00970 "psubw %%mm5, %%mm6 \n\t" /* c - 2b */\ 00971 "pshufw 0ドルx06, %%mm0, %%mm5 \n\t" /* 0C0B0A0A */\ 00972 "pmullw "MANGLE(ff_pw_3)", %%mm6 \n\t" /* 3c - 6b */\ 00973 "paddw %%mm4, %%mm0 \n\t" /* a */\ 00974 "paddw %%mm1, %%mm5 \n\t" /* d */\ 00975 "pmullw "MANGLE(ff_pw_20)", %%mm0 \n\t" /* 20a */\ 00976 "psubw %%mm5, %%mm0 \n\t" /* 20a - d */\ 00977 "paddw %6, %%mm6 \n\t"\ 00978 "paddw %%mm6, %%mm0 \n\t" /* 20a - 6b + 3c - d */\ 00979 "psraw 5,ドル %%mm0 \n\t"\ 00980 "movq %%mm0, %5 \n\t"\ 00981 /* mm1=EFGH, mm2=DEFG, mm3=CDEF, mm4=BCDE, mm7=0 */\ 00982 \ 00983 "movq 5(%0), %%mm0 \n\t" /* FGHIJKLM */\ 00984 "movq %%mm0, %%mm5 \n\t" /* FGHIJKLM */\ 00985 "movq %%mm0, %%mm6 \n\t" /* FGHIJKLM */\ 00986 "psrlq 8,ドル %%mm0 \n\t" /* GHIJKLM0 */\ 00987 "psrlq 16,ドル %%mm5 \n\t" /* HIJKLM00 */\ 00988 "punpcklbw %%mm7, %%mm0 \n\t" /* 0G0H0I0J */\ 00989 "punpcklbw %%mm7, %%mm5 \n\t" /* 0H0I0J0K */\ 00990 "paddw %%mm0, %%mm2 \n\t" /* b */\ 00991 "paddw %%mm5, %%mm3 \n\t" /* c */\ 00992 "paddw %%mm2, %%mm2 \n\t" /* 2b */\ 00993 "psubw %%mm2, %%mm3 \n\t" /* c - 2b */\ 00994 "movq %%mm6, %%mm2 \n\t" /* FGHIJKLM */\ 00995 "psrlq 24,ドル %%mm6 \n\t" /* IJKLM000 */\ 00996 "punpcklbw %%mm7, %%mm2 \n\t" /* 0F0G0H0I */\ 00997 "punpcklbw %%mm7, %%mm6 \n\t" /* 0I0J0K0L */\ 00998 "pmullw "MANGLE(ff_pw_3)", %%mm3 \n\t" /* 3c - 6b */\ 00999 "paddw %%mm2, %%mm1 \n\t" /* a */\ 01000 "paddw %%mm6, %%mm4 \n\t" /* d */\ 01001 "pmullw "MANGLE(ff_pw_20)", %%mm1 \n\t" /* 20a */\ 01002 "psubw %%mm4, %%mm3 \n\t" /* - 6b +3c - d */\ 01003 "paddw %6, %%mm1 \n\t"\ 01004 "paddw %%mm1, %%mm3 \n\t" /* 20a - 6b +3c - d */\ 01005 "psraw 5,ドル %%mm3 \n\t"\ 01006 "movq %5, %%mm1 \n\t"\ 01007 "packuswb %%mm3, %%mm1 \n\t"\ 01008 OP_MMX2(%%mm1, (%1),%%mm4, q)\ 01009 /* mm0= GHIJ, mm2=FGHI, mm5=HIJK, mm6=IJKL, mm7=0 */\ 01010 \ 01011 "movq 9(%0), %%mm1 \n\t" /* JKLMNOPQ */\ 01012 "movq %%mm1, %%mm4 \n\t" /* JKLMNOPQ */\ 01013 "movq %%mm1, %%mm3 \n\t" /* JKLMNOPQ */\ 01014 "psrlq 8,ドル %%mm1 \n\t" /* KLMNOPQ0 */\ 01015 "psrlq 16,ドル %%mm4 \n\t" /* LMNOPQ00 */\ 01016 "punpcklbw %%mm7, %%mm1 \n\t" /* 0K0L0M0N */\ 01017 "punpcklbw %%mm7, %%mm4 \n\t" /* 0L0M0N0O */\ 01018 "paddw %%mm1, %%mm5 \n\t" /* b */\ 01019 "paddw %%mm4, %%mm0 \n\t" /* c */\ 01020 "paddw %%mm5, %%mm5 \n\t" /* 2b */\ 01021 "psubw %%mm5, %%mm0 \n\t" /* c - 2b */\ 01022 "movq %%mm3, %%mm5 \n\t" /* JKLMNOPQ */\ 01023 "psrlq 24,ドル %%mm3 \n\t" /* MNOPQ000 */\ 01024 "pmullw "MANGLE(ff_pw_3)", %%mm0 \n\t" /* 3c - 6b */\ 01025 "punpcklbw %%mm7, %%mm3 \n\t" /* 0M0N0O0P */\ 01026 "paddw %%mm3, %%mm2 \n\t" /* d */\ 01027 "psubw %%mm2, %%mm0 \n\t" /* -6b + 3c - d */\ 01028 "movq %%mm5, %%mm2 \n\t" /* JKLMNOPQ */\ 01029 "punpcklbw %%mm7, %%mm2 \n\t" /* 0J0K0L0M */\ 01030 "punpckhbw %%mm7, %%mm5 \n\t" /* 0N0O0P0Q */\ 01031 "paddw %%mm2, %%mm6 \n\t" /* a */\ 01032 "pmullw "MANGLE(ff_pw_20)", %%mm6 \n\t" /* 20a */\ 01033 "paddw %6, %%mm0 \n\t"\ 01034 "paddw %%mm6, %%mm0 \n\t" /* 20a - 6b + 3c - d */\ 01035 "psraw 5,ドル %%mm0 \n\t"\ 01036 /* mm1=KLMN, mm2=JKLM, mm3=MNOP, mm4=LMNO, mm5=NOPQ mm7=0 */\ 01037 \ 01038 "paddw %%mm5, %%mm3 \n\t" /* a */\ 01039 "pshufw 0ドルxF9, %%mm5, %%mm6 \n\t" /* 0O0P0Q0Q */\ 01040 "paddw %%mm4, %%mm6 \n\t" /* b */\ 01041 "pshufw 0ドルxBE, %%mm5, %%mm4 \n\t" /* 0P0Q0Q0P */\ 01042 "pshufw 0ドルx6F, %%mm5, %%mm5 \n\t" /* 0Q0Q0P0O */\ 01043 "paddw %%mm1, %%mm4 \n\t" /* c */\ 01044 "paddw %%mm2, %%mm5 \n\t" /* d */\ 01045 "paddw %%mm6, %%mm6 \n\t" /* 2b */\ 01046 "psubw %%mm6, %%mm4 \n\t" /* c - 2b */\ 01047 "pmullw "MANGLE(ff_pw_20)", %%mm3 \n\t" /* 20a */\ 01048 "pmullw "MANGLE(ff_pw_3)", %%mm4 \n\t" /* 3c - 6b */\ 01049 "psubw %%mm5, %%mm3 \n\t" /* -6b + 3c - d */\ 01050 "paddw %6, %%mm4 \n\t"\ 01051 "paddw %%mm3, %%mm4 \n\t" /* 20a - 6b + 3c - d */\ 01052 "psraw 5,ドル %%mm4 \n\t"\ 01053 "packuswb %%mm4, %%mm0 \n\t"\ 01054 OP_MMX2(%%mm0, 8(%1), %%mm4, q)\ 01055 \ 01056 "add %3, %0 \n\t"\ 01057 "add %4, %1 \n\t"\ 01058 "decl %2 \n\t"\ 01059 " jnz 1b \n\t"\ 01060 : "+a"(src), "+c"(dst), "+D"(h)\ 01061 : "d"((x86_reg)srcStride), "S"((x86_reg)dstStride), /*"m"(ff_pw_20), "m"(ff_pw_3),*/ "m"(temp), "m"(ROUNDER)\ 01062 : "memory"\ 01063 );\ 01064 }\ 01065 \ 01066 static void OPNAME ## mpeg4_qpel16_h_lowpass_3dnow(uint8_t *dst, uint8_t *src, int dstStride, int srcStride, int h){\ 01067 int i;\ 01068 int16_t temp[16];\ 01069 /* quick HACK, XXX FIXME MUST be optimized */\ 01070 for(i=0; i<h; i++)\ 01071 {\ 01072 temp[ 0]= (src[ 0]+src[ 1])*20 - (src[ 0]+src[ 2])*6 + (src[ 1]+src[ 3])*3 - (src[ 2]+src[ 4]);\ 01073 temp[ 1]= (src[ 1]+src[ 2])*20 - (src[ 0]+src[ 3])*6 + (src[ 0]+src[ 4])*3 - (src[ 1]+src[ 5]);\ 01074 temp[ 2]= (src[ 2]+src[ 3])*20 - (src[ 1]+src[ 4])*6 + (src[ 0]+src[ 5])*3 - (src[ 0]+src[ 6]);\ 01075 temp[ 3]= (src[ 3]+src[ 4])*20 - (src[ 2]+src[ 5])*6 + (src[ 1]+src[ 6])*3 - (src[ 0]+src[ 7]);\ 01076 temp[ 4]= (src[ 4]+src[ 5])*20 - (src[ 3]+src[ 6])*6 + (src[ 2]+src[ 7])*3 - (src[ 1]+src[ 8]);\ 01077 temp[ 5]= (src[ 5]+src[ 6])*20 - (src[ 4]+src[ 7])*6 + (src[ 3]+src[ 8])*3 - (src[ 2]+src[ 9]);\ 01078 temp[ 6]= (src[ 6]+src[ 7])*20 - (src[ 5]+src[ 8])*6 + (src[ 4]+src[ 9])*3 - (src[ 3]+src[10]);\ 01079 temp[ 7]= (src[ 7]+src[ 8])*20 - (src[ 6]+src[ 9])*6 + (src[ 5]+src[10])*3 - (src[ 4]+src[11]);\ 01080 temp[ 8]= (src[ 8]+src[ 9])*20 - (src[ 7]+src[10])*6 + (src[ 6]+src[11])*3 - (src[ 5]+src[12]);\ 01081 temp[ 9]= (src[ 9]+src[10])*20 - (src[ 8]+src[11])*6 + (src[ 7]+src[12])*3 - (src[ 6]+src[13]);\ 01082 temp[10]= (src[10]+src[11])*20 - (src[ 9]+src[12])*6 + (src[ 8]+src[13])*3 - (src[ 7]+src[14]);\ 01083 temp[11]= (src[11]+src[12])*20 - (src[10]+src[13])*6 + (src[ 9]+src[14])*3 - (src[ 8]+src[15]);\ 01084 temp[12]= (src[12]+src[13])*20 - (src[11]+src[14])*6 + (src[10]+src[15])*3 - (src[ 9]+src[16]);\ 01085 temp[13]= (src[13]+src[14])*20 - (src[12]+src[15])*6 + (src[11]+src[16])*3 - (src[10]+src[16]);\ 01086 temp[14]= (src[14]+src[15])*20 - (src[13]+src[16])*6 + (src[12]+src[16])*3 - (src[11]+src[15]);\ 01087 temp[15]= (src[15]+src[16])*20 - (src[14]+src[16])*6 + (src[13]+src[15])*3 - (src[12]+src[14]);\ 01088 __asm__ volatile(\ 01089 "movq (%0), %%mm0 \n\t"\ 01090 "movq 8(%0), %%mm1 \n\t"\ 01091 "paddw %2, %%mm0 \n\t"\ 01092 "paddw %2, %%mm1 \n\t"\ 01093 "psraw 5,ドル %%mm0 \n\t"\ 01094 "psraw 5,ドル %%mm1 \n\t"\ 01095 "packuswb %%mm1, %%mm0 \n\t"\ 01096 OP_3DNOW(%%mm0, (%1), %%mm1, q)\ 01097 "movq 16(%0), %%mm0 \n\t"\ 01098 "movq 24(%0), %%mm1 \n\t"\ 01099 "paddw %2, %%mm0 \n\t"\ 01100 "paddw %2, %%mm1 \n\t"\ 01101 "psraw 5,ドル %%mm0 \n\t"\ 01102 "psraw 5,ドル %%mm1 \n\t"\ 01103 "packuswb %%mm1, %%mm0 \n\t"\ 01104 OP_3DNOW(%%mm0, 8(%1), %%mm1, q)\ 01105 :: "r"(temp), "r"(dst), "m"(ROUNDER)\ 01106 : "memory"\ 01107 );\ 01108 dst+=dstStride;\ 01109 src+=srcStride;\ 01110 }\ 01111 }\ 01112 \ 01113 static void OPNAME ## mpeg4_qpel8_h_lowpass_mmx2(uint8_t *dst, uint8_t *src, int dstStride, int srcStride, int h){\ 01114 __asm__ volatile(\ 01115 "pxor %%mm7, %%mm7 \n\t"\ 01116 "1: \n\t"\ 01117 "movq (%0), %%mm0 \n\t" /* ABCDEFGH */\ 01118 "movq %%mm0, %%mm1 \n\t" /* ABCDEFGH */\ 01119 "movq %%mm0, %%mm2 \n\t" /* ABCDEFGH */\ 01120 "punpcklbw %%mm7, %%mm0 \n\t" /* 0A0B0C0D */\ 01121 "punpckhbw %%mm7, %%mm1 \n\t" /* 0E0F0G0H */\ 01122 "pshufw 0ドルx90, %%mm0, %%mm5 \n\t" /* 0A0A0B0C */\ 01123 "pshufw 0ドルx41, %%mm0, %%mm6 \n\t" /* 0B0A0A0B */\ 01124 "movq %%mm2, %%mm3 \n\t" /* ABCDEFGH */\ 01125 "movq %%mm2, %%mm4 \n\t" /* ABCDEFGH */\ 01126 "psllq 8,ドル %%mm2 \n\t" /* 0ABCDEFG */\ 01127 "psllq 16,ドル %%mm3 \n\t" /* 00ABCDEF */\ 01128 "psllq 24,ドル %%mm4 \n\t" /* 000ABCDE */\ 01129 "punpckhbw %%mm7, %%mm2 \n\t" /* 0D0E0F0G */\ 01130 "punpckhbw %%mm7, %%mm3 \n\t" /* 0C0D0E0F */\ 01131 "punpckhbw %%mm7, %%mm4 \n\t" /* 0B0C0D0E */\ 01132 "paddw %%mm3, %%mm5 \n\t" /* b */\ 01133 "paddw %%mm2, %%mm6 \n\t" /* c */\ 01134 "paddw %%mm5, %%mm5 \n\t" /* 2b */\ 01135 "psubw %%mm5, %%mm6 \n\t" /* c - 2b */\ 01136 "pshufw 0ドルx06, %%mm0, %%mm5 \n\t" /* 0C0B0A0A */\ 01137 "pmullw "MANGLE(ff_pw_3)", %%mm6 \n\t" /* 3c - 6b */\ 01138 "paddw %%mm4, %%mm0 \n\t" /* a */\ 01139 "paddw %%mm1, %%mm5 \n\t" /* d */\ 01140 "pmullw "MANGLE(ff_pw_20)", %%mm0 \n\t" /* 20a */\ 01141 "psubw %%mm5, %%mm0 \n\t" /* 20a - d */\ 01142 "paddw %5, %%mm6 \n\t"\ 01143 "paddw %%mm6, %%mm0 \n\t" /* 20a - 6b + 3c - d */\ 01144 "psraw 5,ドル %%mm0 \n\t"\ 01145 /* mm1=EFGH, mm2=DEFG, mm3=CDEF, mm4=BCDE, mm7=0 */\ 01146 \ 01147 "movd 5(%0), %%mm5 \n\t" /* FGHI */\ 01148 "punpcklbw %%mm7, %%mm5 \n\t" /* 0F0G0H0I */\ 01149 "pshufw 0ドルxF9, %%mm5, %%mm6 \n\t" /* 0G0H0I0I */\ 01150 "paddw %%mm5, %%mm1 \n\t" /* a */\ 01151 "paddw %%mm6, %%mm2 \n\t" /* b */\ 01152 "pshufw 0ドルxBE, %%mm5, %%mm6 \n\t" /* 0H0I0I0H */\ 01153 "pshufw 0ドルx6F, %%mm5, %%mm5 \n\t" /* 0I0I0H0G */\ 01154 "paddw %%mm6, %%mm3 \n\t" /* c */\ 01155 "paddw %%mm5, %%mm4 \n\t" /* d */\ 01156 "paddw %%mm2, %%mm2 \n\t" /* 2b */\ 01157 "psubw %%mm2, %%mm3 \n\t" /* c - 2b */\ 01158 "pmullw "MANGLE(ff_pw_20)", %%mm1 \n\t" /* 20a */\ 01159 "pmullw "MANGLE(ff_pw_3)", %%mm3 \n\t" /* 3c - 6b */\ 01160 "psubw %%mm4, %%mm3 \n\t" /* -6b + 3c - d */\ 01161 "paddw %5, %%mm1 \n\t"\ 01162 "paddw %%mm1, %%mm3 \n\t" /* 20a - 6b + 3c - d */\ 01163 "psraw 5,ドル %%mm3 \n\t"\ 01164 "packuswb %%mm3, %%mm0 \n\t"\ 01165 OP_MMX2(%%mm0, (%1), %%mm4, q)\ 01166 \ 01167 "add %3, %0 \n\t"\ 01168 "add %4, %1 \n\t"\ 01169 "decl %2 \n\t"\ 01170 " jnz 1b \n\t"\ 01171 : "+a"(src), "+c"(dst), "+d"(h)\ 01172 : "S"((x86_reg)srcStride), "D"((x86_reg)dstStride), /*"m"(ff_pw_20), "m"(ff_pw_3),*/ "m"(ROUNDER)\ 01173 : "memory"\ 01174 );\ 01175 }\ 01176 \ 01177 static void OPNAME ## mpeg4_qpel8_h_lowpass_3dnow(uint8_t *dst, uint8_t *src, int dstStride, int srcStride, int h){\ 01178 int i;\ 01179 int16_t temp[8];\ 01180 /* quick HACK, XXX FIXME MUST be optimized */\ 01181 for(i=0; i<h; i++)\ 01182 {\ 01183 temp[ 0]= (src[ 0]+src[ 1])*20 - (src[ 0]+src[ 2])*6 + (src[ 1]+src[ 3])*3 - (src[ 2]+src[ 4]);\ 01184 temp[ 1]= (src[ 1]+src[ 2])*20 - (src[ 0]+src[ 3])*6 + (src[ 0]+src[ 4])*3 - (src[ 1]+src[ 5]);\ 01185 temp[ 2]= (src[ 2]+src[ 3])*20 - (src[ 1]+src[ 4])*6 + (src[ 0]+src[ 5])*3 - (src[ 0]+src[ 6]);\ 01186 temp[ 3]= (src[ 3]+src[ 4])*20 - (src[ 2]+src[ 5])*6 + (src[ 1]+src[ 6])*3 - (src[ 0]+src[ 7]);\ 01187 temp[ 4]= (src[ 4]+src[ 5])*20 - (src[ 3]+src[ 6])*6 + (src[ 2]+src[ 7])*3 - (src[ 1]+src[ 8]);\ 01188 temp[ 5]= (src[ 5]+src[ 6])*20 - (src[ 4]+src[ 7])*6 + (src[ 3]+src[ 8])*3 - (src[ 2]+src[ 8]);\ 01189 temp[ 6]= (src[ 6]+src[ 7])*20 - (src[ 5]+src[ 8])*6 + (src[ 4]+src[ 8])*3 - (src[ 3]+src[ 7]);\ 01190 temp[ 7]= (src[ 7]+src[ 8])*20 - (src[ 6]+src[ 8])*6 + (src[ 5]+src[ 7])*3 - (src[ 4]+src[ 6]);\ 01191 __asm__ volatile(\ 01192 "movq (%0), %%mm0 \n\t"\ 01193 "movq 8(%0), %%mm1 \n\t"\ 01194 "paddw %2, %%mm0 \n\t"\ 01195 "paddw %2, %%mm1 \n\t"\ 01196 "psraw 5,ドル %%mm0 \n\t"\ 01197 "psraw 5,ドル %%mm1 \n\t"\ 01198 "packuswb %%mm1, %%mm0 \n\t"\ 01199 OP_3DNOW(%%mm0, (%1), %%mm1, q)\ 01200 :: "r"(temp), "r"(dst), "m"(ROUNDER)\ 01201 :"memory"\ 01202 );\ 01203 dst+=dstStride;\ 01204 src+=srcStride;\ 01205 }\ 01206 } 01207 01208 #define QPEL_OP(OPNAME, ROUNDER, RND, OP, MMX)\ 01209 \ 01210 static void OPNAME ## mpeg4_qpel16_v_lowpass_ ## MMX(uint8_t *dst, uint8_t *src, int dstStride, int srcStride){\ 01211 uint64_t temp[17*4];\ 01212 uint64_t *temp_ptr= temp;\ 01213 int count= 17;\ 01214 \ 01215 /*FIXME unroll */\ 01216 __asm__ volatile(\ 01217 "pxor %%mm7, %%mm7 \n\t"\ 01218 "1: \n\t"\ 01219 "movq (%0), %%mm0 \n\t"\ 01220 "movq (%0), %%mm1 \n\t"\ 01221 "movq 8(%0), %%mm2 \n\t"\ 01222 "movq 8(%0), %%mm3 \n\t"\ 01223 "punpcklbw %%mm7, %%mm0 \n\t"\ 01224 "punpckhbw %%mm7, %%mm1 \n\t"\ 01225 "punpcklbw %%mm7, %%mm2 \n\t"\ 01226 "punpckhbw %%mm7, %%mm3 \n\t"\ 01227 "movq %%mm0, (%1) \n\t"\ 01228 "movq %%mm1, 17*8(%1) \n\t"\ 01229 "movq %%mm2, 2*17*8(%1) \n\t"\ 01230 "movq %%mm3, 3*17*8(%1) \n\t"\ 01231 "add 8,ドル %1 \n\t"\ 01232 "add %3, %0 \n\t"\ 01233 "decl %2 \n\t"\ 01234 " jnz 1b \n\t"\ 01235 : "+r" (src), "+r" (temp_ptr), "+r"(count)\ 01236 : "r" ((x86_reg)srcStride)\ 01237 : "memory"\ 01238 );\ 01239 \ 01240 temp_ptr= temp;\ 01241 count=4;\ 01242 \ 01243 /*FIXME reorder for speed */\ 01244 __asm__ volatile(\ 01245 /*"pxor %%mm7, %%mm7 \n\t"*/\ 01246 "1: \n\t"\ 01247 "movq (%0), %%mm0 \n\t"\ 01248 "movq 8(%0), %%mm1 \n\t"\ 01249 "movq 16(%0), %%mm2 \n\t"\ 01250 "movq 24(%0), %%mm3 \n\t"\ 01251 QPEL_V_LOW(%%mm0, %%mm1, %%mm2, %%mm3, %5, %6, %5, 16(%0), 8(%0), (%0), 32(%0), (%1), OP)\ 01252 QPEL_V_LOW(%%mm1, %%mm2, %%mm3, %%mm0, %5, %6, %5, 8(%0), (%0), (%0), 40(%0), (%1, %3), OP)\ 01253 "add %4, %1 \n\t"\ 01254 QPEL_V_LOW(%%mm2, %%mm3, %%mm0, %%mm1, %5, %6, %5, (%0), (%0), 8(%0), 48(%0), (%1), OP)\ 01255 \ 01256 QPEL_V_LOW(%%mm3, %%mm0, %%mm1, %%mm2, %5, %6, %5, (%0), 8(%0), 16(%0), 56(%0), (%1, %3), OP)\ 01257 "add %4, %1 \n\t"\ 01258 QPEL_V_LOW(%%mm0, %%mm1, %%mm2, %%mm3, %5, %6, %5, 8(%0), 16(%0), 24(%0), 64(%0), (%1), OP)\ 01259 QPEL_V_LOW(%%mm1, %%mm2, %%mm3, %%mm0, %5, %6, %5, 16(%0), 24(%0), 32(%0), 72(%0), (%1, %3), OP)\ 01260 "add %4, %1 \n\t"\ 01261 QPEL_V_LOW(%%mm2, %%mm3, %%mm0, %%mm1, %5, %6, %5, 24(%0), 32(%0), 40(%0), 80(%0), (%1), OP)\ 01262 QPEL_V_LOW(%%mm3, %%mm0, %%mm1, %%mm2, %5, %6, %5, 32(%0), 40(%0), 48(%0), 88(%0), (%1, %3), OP)\ 01263 "add %4, %1 \n\t"\ 01264 QPEL_V_LOW(%%mm0, %%mm1, %%mm2, %%mm3, %5, %6, %5, 40(%0), 48(%0), 56(%0), 96(%0), (%1), OP)\ 01265 QPEL_V_LOW(%%mm1, %%mm2, %%mm3, %%mm0, %5, %6, %5, 48(%0), 56(%0), 64(%0),104(%0), (%1, %3), OP)\ 01266 "add %4, %1 \n\t"\ 01267 QPEL_V_LOW(%%mm2, %%mm3, %%mm0, %%mm1, %5, %6, %5, 56(%0), 64(%0), 72(%0),112(%0), (%1), OP)\ 01268 QPEL_V_LOW(%%mm3, %%mm0, %%mm1, %%mm2, %5, %6, %5, 64(%0), 72(%0), 80(%0),120(%0), (%1, %3), OP)\ 01269 "add %4, %1 \n\t"\ 01270 QPEL_V_LOW(%%mm0, %%mm1, %%mm2, %%mm3, %5, %6, %5, 72(%0), 80(%0), 88(%0),128(%0), (%1), OP)\ 01271 \ 01272 QPEL_V_LOW(%%mm1, %%mm2, %%mm3, %%mm0, %5, %6, %5, 80(%0), 88(%0), 96(%0),128(%0), (%1, %3), OP)\ 01273 "add %4, %1 \n\t" \ 01274 QPEL_V_LOW(%%mm2, %%mm3, %%mm0, %%mm1, %5, %6, %5, 88(%0), 96(%0),104(%0),120(%0), (%1), OP)\ 01275 QPEL_V_LOW(%%mm3, %%mm0, %%mm1, %%mm2, %5, %6, %5, 96(%0),104(%0),112(%0),112(%0), (%1, %3), OP)\ 01276 \ 01277 "add 136,ドル %0 \n\t"\ 01278 "add %6, %1 \n\t"\ 01279 "decl %2 \n\t"\ 01280 " jnz 1b \n\t"\ 01281 \ 01282 : "+r"(temp_ptr), "+r"(dst), "+g"(count)\ 01283 : "r"((x86_reg)dstStride), "r"(2*(x86_reg)dstStride), /*"m"(ff_pw_20), "m"(ff_pw_3),*/ "m"(ROUNDER), "g"(4-14*(x86_reg)dstStride)\ 01284 :"memory"\ 01285 );\ 01286 }\ 01287 \ 01288 static void OPNAME ## mpeg4_qpel8_v_lowpass_ ## MMX(uint8_t *dst, uint8_t *src, int dstStride, int srcStride){\ 01289 uint64_t temp[9*2];\ 01290 uint64_t *temp_ptr= temp;\ 01291 int count= 9;\ 01292 \ 01293 /*FIXME unroll */\ 01294 __asm__ volatile(\ 01295 "pxor %%mm7, %%mm7 \n\t"\ 01296 "1: \n\t"\ 01297 "movq (%0), %%mm0 \n\t"\ 01298 "movq (%0), %%mm1 \n\t"\ 01299 "punpcklbw %%mm7, %%mm0 \n\t"\ 01300 "punpckhbw %%mm7, %%mm1 \n\t"\ 01301 "movq %%mm0, (%1) \n\t"\ 01302 "movq %%mm1, 9*8(%1) \n\t"\ 01303 "add 8,ドル %1 \n\t"\ 01304 "add %3, %0 \n\t"\ 01305 "decl %2 \n\t"\ 01306 " jnz 1b \n\t"\ 01307 : "+r" (src), "+r" (temp_ptr), "+r"(count)\ 01308 : "r" ((x86_reg)srcStride)\ 01309 : "memory"\ 01310 );\ 01311 \ 01312 temp_ptr= temp;\ 01313 count=2;\ 01314 \ 01315 /*FIXME reorder for speed */\ 01316 __asm__ volatile(\ 01317 /*"pxor %%mm7, %%mm7 \n\t"*/\ 01318 "1: \n\t"\ 01319 "movq (%0), %%mm0 \n\t"\ 01320 "movq 8(%0), %%mm1 \n\t"\ 01321 "movq 16(%0), %%mm2 \n\t"\ 01322 "movq 24(%0), %%mm3 \n\t"\ 01323 QPEL_V_LOW(%%mm0, %%mm1, %%mm2, %%mm3, %5, %6, %5, 16(%0), 8(%0), (%0), 32(%0), (%1), OP)\ 01324 QPEL_V_LOW(%%mm1, %%mm2, %%mm3, %%mm0, %5, %6, %5, 8(%0), (%0), (%0), 40(%0), (%1, %3), OP)\ 01325 "add %4, %1 \n\t"\ 01326 QPEL_V_LOW(%%mm2, %%mm3, %%mm0, %%mm1, %5, %6, %5, (%0), (%0), 8(%0), 48(%0), (%1), OP)\ 01327 \ 01328 QPEL_V_LOW(%%mm3, %%mm0, %%mm1, %%mm2, %5, %6, %5, (%0), 8(%0), 16(%0), 56(%0), (%1, %3), OP)\ 01329 "add %4, %1 \n\t"\ 01330 QPEL_V_LOW(%%mm0, %%mm1, %%mm2, %%mm3, %5, %6, %5, 8(%0), 16(%0), 24(%0), 64(%0), (%1), OP)\ 01331 \ 01332 QPEL_V_LOW(%%mm1, %%mm2, %%mm3, %%mm0, %5, %6, %5, 16(%0), 24(%0), 32(%0), 64(%0), (%1, %3), OP)\ 01333 "add %4, %1 \n\t"\ 01334 QPEL_V_LOW(%%mm2, %%mm3, %%mm0, %%mm1, %5, %6, %5, 24(%0), 32(%0), 40(%0), 56(%0), (%1), OP)\ 01335 QPEL_V_LOW(%%mm3, %%mm0, %%mm1, %%mm2, %5, %6, %5, 32(%0), 40(%0), 48(%0), 48(%0), (%1, %3), OP)\ 01336 \ 01337 "add 72,ドル %0 \n\t"\ 01338 "add %6, %1 \n\t"\ 01339 "decl %2 \n\t"\ 01340 " jnz 1b \n\t"\ 01341 \ 01342 : "+r"(temp_ptr), "+r"(dst), "+g"(count)\ 01343 : "r"((x86_reg)dstStride), "r"(2*(x86_reg)dstStride), /*"m"(ff_pw_20), "m"(ff_pw_3),*/ "m"(ROUNDER), "g"(4-6*(x86_reg)dstStride)\ 01344 : "memory"\ 01345 );\ 01346 }\ 01347 \ 01348 static void OPNAME ## qpel8_mc00_ ## MMX (uint8_t *dst, uint8_t *src, int stride){\ 01349 OPNAME ## pixels8_ ## MMX(dst, src, stride, 8);\ 01350 }\ 01351 \ 01352 static void OPNAME ## qpel8_mc10_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\ 01353 uint64_t temp[8];\ 01354 uint8_t * const half= (uint8_t*)temp;\ 01355 put ## RND ## mpeg4_qpel8_h_lowpass_ ## MMX(half, src, 8, stride, 8);\ 01356 OPNAME ## pixels8_l2_ ## MMX(dst, src, half, stride, stride, 8);\ 01357 }\ 01358 \ 01359 static void OPNAME ## qpel8_mc20_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\ 01360 OPNAME ## mpeg4_qpel8_h_lowpass_ ## MMX(dst, src, stride, stride, 8);\ 01361 }\ 01362 \ 01363 static void OPNAME ## qpel8_mc30_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\ 01364 uint64_t temp[8];\ 01365 uint8_t * const half= (uint8_t*)temp;\ 01366 put ## RND ## mpeg4_qpel8_h_lowpass_ ## MMX(half, src, 8, stride, 8);\ 01367 OPNAME ## pixels8_l2_ ## MMX(dst, src+1, half, stride, stride, 8);\ 01368 }\ 01369 \ 01370 static void OPNAME ## qpel8_mc01_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\ 01371 uint64_t temp[8];\ 01372 uint8_t * const half= (uint8_t*)temp;\ 01373 put ## RND ## mpeg4_qpel8_v_lowpass_ ## MMX(half, src, 8, stride);\ 01374 OPNAME ## pixels8_l2_ ## MMX(dst, src, half, stride, stride, 8);\ 01375 }\ 01376 \ 01377 static void OPNAME ## qpel8_mc02_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\ 01378 OPNAME ## mpeg4_qpel8_v_lowpass_ ## MMX(dst, src, stride, stride);\ 01379 }\ 01380 \ 01381 static void OPNAME ## qpel8_mc03_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\ 01382 uint64_t temp[8];\ 01383 uint8_t * const half= (uint8_t*)temp;\ 01384 put ## RND ## mpeg4_qpel8_v_lowpass_ ## MMX(half, src, 8, stride);\ 01385 OPNAME ## pixels8_l2_ ## MMX(dst, src+stride, half, stride, stride, 8);\ 01386 }\ 01387 static void OPNAME ## qpel8_mc11_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\ 01388 uint64_t half[8 + 9];\ 01389 uint8_t * const halfH= ((uint8_t*)half) + 64;\ 01390 uint8_t * const halfHV= ((uint8_t*)half);\ 01391 put ## RND ## mpeg4_qpel8_h_lowpass_ ## MMX(halfH, src, 8, stride, 9);\ 01392 put ## RND ## pixels8_l2_ ## MMX(halfH, src, halfH, 8, stride, 9);\ 01393 put ## RND ## mpeg4_qpel8_v_lowpass_ ## MMX(halfHV, halfH, 8, 8);\ 01394 OPNAME ## pixels8_l2_ ## MMX(dst, halfH, halfHV, stride, 8, 8);\ 01395 }\ 01396 static void OPNAME ## qpel8_mc31_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\ 01397 uint64_t half[8 + 9];\ 01398 uint8_t * const halfH= ((uint8_t*)half) + 64;\ 01399 uint8_t * const halfHV= ((uint8_t*)half);\ 01400 put ## RND ## mpeg4_qpel8_h_lowpass_ ## MMX(halfH, src, 8, stride, 9);\ 01401 put ## RND ## pixels8_l2_ ## MMX(halfH, src+1, halfH, 8, stride, 9);\ 01402 put ## RND ## mpeg4_qpel8_v_lowpass_ ## MMX(halfHV, halfH, 8, 8);\ 01403 OPNAME ## pixels8_l2_ ## MMX(dst, halfH, halfHV, stride, 8, 8);\ 01404 }\ 01405 static void OPNAME ## qpel8_mc13_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\ 01406 uint64_t half[8 + 9];\ 01407 uint8_t * const halfH= ((uint8_t*)half) + 64;\ 01408 uint8_t * const halfHV= ((uint8_t*)half);\ 01409 put ## RND ## mpeg4_qpel8_h_lowpass_ ## MMX(halfH, src, 8, stride, 9);\ 01410 put ## RND ## pixels8_l2_ ## MMX(halfH, src, halfH, 8, stride, 9);\ 01411 put ## RND ## mpeg4_qpel8_v_lowpass_ ## MMX(halfHV, halfH, 8, 8);\ 01412 OPNAME ## pixels8_l2_ ## MMX(dst, halfH+8, halfHV, stride, 8, 8);\ 01413 }\ 01414 static void OPNAME ## qpel8_mc33_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\ 01415 uint64_t half[8 + 9];\ 01416 uint8_t * const halfH= ((uint8_t*)half) + 64;\ 01417 uint8_t * const halfHV= ((uint8_t*)half);\ 01418 put ## RND ## mpeg4_qpel8_h_lowpass_ ## MMX(halfH, src, 8, stride, 9);\ 01419 put ## RND ## pixels8_l2_ ## MMX(halfH, src+1, halfH, 8, stride, 9);\ 01420 put ## RND ## mpeg4_qpel8_v_lowpass_ ## MMX(halfHV, halfH, 8, 8);\ 01421 OPNAME ## pixels8_l2_ ## MMX(dst, halfH+8, halfHV, stride, 8, 8);\ 01422 }\ 01423 static void OPNAME ## qpel8_mc21_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\ 01424 uint64_t half[8 + 9];\ 01425 uint8_t * const halfH= ((uint8_t*)half) + 64;\ 01426 uint8_t * const halfHV= ((uint8_t*)half);\ 01427 put ## RND ## mpeg4_qpel8_h_lowpass_ ## MMX(halfH, src, 8, stride, 9);\ 01428 put ## RND ## mpeg4_qpel8_v_lowpass_ ## MMX(halfHV, halfH, 8, 8);\ 01429 OPNAME ## pixels8_l2_ ## MMX(dst, halfH, halfHV, stride, 8, 8);\ 01430 }\ 01431 static void OPNAME ## qpel8_mc23_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\ 01432 uint64_t half[8 + 9];\ 01433 uint8_t * const halfH= ((uint8_t*)half) + 64;\ 01434 uint8_t * const halfHV= ((uint8_t*)half);\ 01435 put ## RND ## mpeg4_qpel8_h_lowpass_ ## MMX(halfH, src, 8, stride, 9);\ 01436 put ## RND ## mpeg4_qpel8_v_lowpass_ ## MMX(halfHV, halfH, 8, 8);\ 01437 OPNAME ## pixels8_l2_ ## MMX(dst, halfH+8, halfHV, stride, 8, 8);\ 01438 }\ 01439 static void OPNAME ## qpel8_mc12_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\ 01440 uint64_t half[8 + 9];\ 01441 uint8_t * const halfH= ((uint8_t*)half);\ 01442 put ## RND ## mpeg4_qpel8_h_lowpass_ ## MMX(halfH, src, 8, stride, 9);\ 01443 put ## RND ## pixels8_l2_ ## MMX(halfH, src, halfH, 8, stride, 9);\ 01444 OPNAME ## mpeg4_qpel8_v_lowpass_ ## MMX(dst, halfH, stride, 8);\ 01445 }\ 01446 static void OPNAME ## qpel8_mc32_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\ 01447 uint64_t half[8 + 9];\ 01448 uint8_t * const halfH= ((uint8_t*)half);\ 01449 put ## RND ## mpeg4_qpel8_h_lowpass_ ## MMX(halfH, src, 8, stride, 9);\ 01450 put ## RND ## pixels8_l2_ ## MMX(halfH, src+1, halfH, 8, stride, 9);\ 01451 OPNAME ## mpeg4_qpel8_v_lowpass_ ## MMX(dst, halfH, stride, 8);\ 01452 }\ 01453 static void OPNAME ## qpel8_mc22_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\ 01454 uint64_t half[9];\ 01455 uint8_t * const halfH= ((uint8_t*)half);\ 01456 put ## RND ## mpeg4_qpel8_h_lowpass_ ## MMX(halfH, src, 8, stride, 9);\ 01457 OPNAME ## mpeg4_qpel8_v_lowpass_ ## MMX(dst, halfH, stride, 8);\ 01458 }\ 01459 static void OPNAME ## qpel16_mc00_ ## MMX (uint8_t *dst, uint8_t *src, int stride){\ 01460 OPNAME ## pixels16_ ## MMX(dst, src, stride, 16);\ 01461 }\ 01462 \ 01463 static void OPNAME ## qpel16_mc10_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\ 01464 uint64_t temp[32];\ 01465 uint8_t * const half= (uint8_t*)temp;\ 01466 put ## RND ## mpeg4_qpel16_h_lowpass_ ## MMX(half, src, 16, stride, 16);\ 01467 OPNAME ## pixels16_l2_ ## MMX(dst, src, half, stride, stride, 16);\ 01468 }\ 01469 \ 01470 static void OPNAME ## qpel16_mc20_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\ 01471 OPNAME ## mpeg4_qpel16_h_lowpass_ ## MMX(dst, src, stride, stride, 16);\ 01472 }\ 01473 \ 01474 static void OPNAME ## qpel16_mc30_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\ 01475 uint64_t temp[32];\ 01476 uint8_t * const half= (uint8_t*)temp;\ 01477 put ## RND ## mpeg4_qpel16_h_lowpass_ ## MMX(half, src, 16, stride, 16);\ 01478 OPNAME ## pixels16_l2_ ## MMX(dst, src+1, half, stride, stride, 16);\ 01479 }\ 01480 \ 01481 static void OPNAME ## qpel16_mc01_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\ 01482 uint64_t temp[32];\ 01483 uint8_t * const half= (uint8_t*)temp;\ 01484 put ## RND ## mpeg4_qpel16_v_lowpass_ ## MMX(half, src, 16, stride);\ 01485 OPNAME ## pixels16_l2_ ## MMX(dst, src, half, stride, stride, 16);\ 01486 }\ 01487 \ 01488 static void OPNAME ## qpel16_mc02_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\ 01489 OPNAME ## mpeg4_qpel16_v_lowpass_ ## MMX(dst, src, stride, stride);\ 01490 }\ 01491 \ 01492 static void OPNAME ## qpel16_mc03_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\ 01493 uint64_t temp[32];\ 01494 uint8_t * const half= (uint8_t*)temp;\ 01495 put ## RND ## mpeg4_qpel16_v_lowpass_ ## MMX(half, src, 16, stride);\ 01496 OPNAME ## pixels16_l2_ ## MMX(dst, src+stride, half, stride, stride, 16);\ 01497 }\ 01498 static void OPNAME ## qpel16_mc11_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\ 01499 uint64_t half[16*2 + 17*2];\ 01500 uint8_t * const halfH= ((uint8_t*)half) + 256;\ 01501 uint8_t * const halfHV= ((uint8_t*)half);\ 01502 put ## RND ## mpeg4_qpel16_h_lowpass_ ## MMX(halfH, src, 16, stride, 17);\ 01503 put ## RND ## pixels16_l2_ ## MMX(halfH, src, halfH, 16, stride, 17);\ 01504 put ## RND ## mpeg4_qpel16_v_lowpass_ ## MMX(halfHV, halfH, 16, 16);\ 01505 OPNAME ## pixels16_l2_ ## MMX(dst, halfH, halfHV, stride, 16, 16);\ 01506 }\ 01507 static void OPNAME ## qpel16_mc31_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\ 01508 uint64_t half[16*2 + 17*2];\ 01509 uint8_t * const halfH= ((uint8_t*)half) + 256;\ 01510 uint8_t * const halfHV= ((uint8_t*)half);\ 01511 put ## RND ## mpeg4_qpel16_h_lowpass_ ## MMX(halfH, src, 16, stride, 17);\ 01512 put ## RND ## pixels16_l2_ ## MMX(halfH, src+1, halfH, 16, stride, 17);\ 01513 put ## RND ## mpeg4_qpel16_v_lowpass_ ## MMX(halfHV, halfH, 16, 16);\ 01514 OPNAME ## pixels16_l2_ ## MMX(dst, halfH, halfHV, stride, 16, 16);\ 01515 }\ 01516 static void OPNAME ## qpel16_mc13_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\ 01517 uint64_t half[16*2 + 17*2];\ 01518 uint8_t * const halfH= ((uint8_t*)half) + 256;\ 01519 uint8_t * const halfHV= ((uint8_t*)half);\ 01520 put ## RND ## mpeg4_qpel16_h_lowpass_ ## MMX(halfH, src, 16, stride, 17);\ 01521 put ## RND ## pixels16_l2_ ## MMX(halfH, src, halfH, 16, stride, 17);\ 01522 put ## RND ## mpeg4_qpel16_v_lowpass_ ## MMX(halfHV, halfH, 16, 16);\ 01523 OPNAME ## pixels16_l2_ ## MMX(dst, halfH+16, halfHV, stride, 16, 16);\ 01524 }\ 01525 static void OPNAME ## qpel16_mc33_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\ 01526 uint64_t half[16*2 + 17*2];\ 01527 uint8_t * const halfH= ((uint8_t*)half) + 256;\ 01528 uint8_t * const halfHV= ((uint8_t*)half);\ 01529 put ## RND ## mpeg4_qpel16_h_lowpass_ ## MMX(halfH, src, 16, stride, 17);\ 01530 put ## RND ## pixels16_l2_ ## MMX(halfH, src+1, halfH, 16, stride, 17);\ 01531 put ## RND ## mpeg4_qpel16_v_lowpass_ ## MMX(halfHV, halfH, 16, 16);\ 01532 OPNAME ## pixels16_l2_ ## MMX(dst, halfH+16, halfHV, stride, 16, 16);\ 01533 }\ 01534 static void OPNAME ## qpel16_mc21_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\ 01535 uint64_t half[16*2 + 17*2];\ 01536 uint8_t * const halfH= ((uint8_t*)half) + 256;\ 01537 uint8_t * const halfHV= ((uint8_t*)half);\ 01538 put ## RND ## mpeg4_qpel16_h_lowpass_ ## MMX(halfH, src, 16, stride, 17);\ 01539 put ## RND ## mpeg4_qpel16_v_lowpass_ ## MMX(halfHV, halfH, 16, 16);\ 01540 OPNAME ## pixels16_l2_ ## MMX(dst, halfH, halfHV, stride, 16, 16);\ 01541 }\ 01542 static void OPNAME ## qpel16_mc23_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\ 01543 uint64_t half[16*2 + 17*2];\ 01544 uint8_t * const halfH= ((uint8_t*)half) + 256;\ 01545 uint8_t * const halfHV= ((uint8_t*)half);\ 01546 put ## RND ## mpeg4_qpel16_h_lowpass_ ## MMX(halfH, src, 16, stride, 17);\ 01547 put ## RND ## mpeg4_qpel16_v_lowpass_ ## MMX(halfHV, halfH, 16, 16);\ 01548 OPNAME ## pixels16_l2_ ## MMX(dst, halfH+16, halfHV, stride, 16, 16);\ 01549 }\ 01550 static void OPNAME ## qpel16_mc12_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\ 01551 uint64_t half[17*2];\ 01552 uint8_t * const halfH= ((uint8_t*)half);\ 01553 put ## RND ## mpeg4_qpel16_h_lowpass_ ## MMX(halfH, src, 16, stride, 17);\ 01554 put ## RND ## pixels16_l2_ ## MMX(halfH, src, halfH, 16, stride, 17);\ 01555 OPNAME ## mpeg4_qpel16_v_lowpass_ ## MMX(dst, halfH, stride, 16);\ 01556 }\ 01557 static void OPNAME ## qpel16_mc32_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\ 01558 uint64_t half[17*2];\ 01559 uint8_t * const halfH= ((uint8_t*)half);\ 01560 put ## RND ## mpeg4_qpel16_h_lowpass_ ## MMX(halfH, src, 16, stride, 17);\ 01561 put ## RND ## pixels16_l2_ ## MMX(halfH, src+1, halfH, 16, stride, 17);\ 01562 OPNAME ## mpeg4_qpel16_v_lowpass_ ## MMX(dst, halfH, stride, 16);\ 01563 }\ 01564 static void OPNAME ## qpel16_mc22_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\ 01565 uint64_t half[17*2];\ 01566 uint8_t * const halfH= ((uint8_t*)half);\ 01567 put ## RND ## mpeg4_qpel16_h_lowpass_ ## MMX(halfH, src, 16, stride, 17);\ 01568 OPNAME ## mpeg4_qpel16_v_lowpass_ ## MMX(dst, halfH, stride, 16);\ 01569 } 01570 01571 #define PUT_OP(a,b,temp, size) "mov" #size " " #a ", " #b " \n\t" 01572 #define AVG_3DNOW_OP(a,b,temp, size) \ 01573 "mov" #size " " #b ", " #temp " \n\t"\ 01574 "pavgusb " #temp ", " #a " \n\t"\ 01575 "mov" #size " " #a ", " #b " \n\t" 01576 #define AVG_MMX2_OP(a,b,temp, size) \ 01577 "mov" #size " " #b ", " #temp " \n\t"\ 01578 "pavgb " #temp ", " #a " \n\t"\ 01579 "mov" #size " " #a ", " #b " \n\t" 01580 01581 QPEL_BASE(put_ , ff_pw_16, _ , PUT_OP, PUT_OP) 01582 QPEL_BASE(avg_ , ff_pw_16, _ , AVG_MMX2_OP, AVG_3DNOW_OP) 01583 QPEL_BASE(put_no_rnd_, ff_pw_15, _no_rnd_, PUT_OP, PUT_OP) 01584 QPEL_OP(put_ , ff_pw_16, _ , PUT_OP, 3dnow) 01585 QPEL_OP(avg_ , ff_pw_16, _ , AVG_3DNOW_OP, 3dnow) 01586 QPEL_OP(put_no_rnd_, ff_pw_15, _no_rnd_, PUT_OP, 3dnow) 01587 QPEL_OP(put_ , ff_pw_16, _ , PUT_OP, mmx2) 01588 QPEL_OP(avg_ , ff_pw_16, _ , AVG_MMX2_OP, mmx2) 01589 QPEL_OP(put_no_rnd_, ff_pw_15, _no_rnd_, PUT_OP, mmx2) 01590 01591 /***********************************/ 01592 /* bilinear qpel: not compliant to any spec, only for -lavdopts fast */ 01593 01594 #define QPEL_2TAP_XY(OPNAME, SIZE, MMX, XY, HPEL)\ 01595 static void OPNAME ## 2tap_qpel ## SIZE ## _mc ## XY ## _ ## MMX(uint8_t *dst, uint8_t *src, int stride){\ 01596 OPNAME ## pixels ## SIZE ## HPEL(dst, src, stride, SIZE);\ 01597 } 01598 #define QPEL_2TAP_L3(OPNAME, SIZE, MMX, XY, S0, S1, S2)\ 01599 static void OPNAME ## 2tap_qpel ## SIZE ## _mc ## XY ## _ ## MMX(uint8_t *dst, uint8_t *src, int stride){\ 01600 OPNAME ## 2tap_qpel ## SIZE ## _l3_ ## MMX(dst, src+S0, stride, SIZE, S1, S2);\ 01601 } 01602 01603 #define QPEL_2TAP(OPNAME, SIZE, MMX)\ 01604 QPEL_2TAP_XY(OPNAME, SIZE, MMX, 20, _x2_ ## MMX)\ 01605 QPEL_2TAP_XY(OPNAME, SIZE, MMX, 02, _y2_ ## MMX)\ 01606 QPEL_2TAP_XY(OPNAME, SIZE, MMX, 22, _xy2_mmx)\ 01607 static const qpel_mc_func OPNAME ## 2tap_qpel ## SIZE ## _mc00_ ## MMX =\ 01608 OPNAME ## qpel ## SIZE ## _mc00_ ## MMX;\ 01609 static const qpel_mc_func OPNAME ## 2tap_qpel ## SIZE ## _mc21_ ## MMX =\ 01610 OPNAME ## 2tap_qpel ## SIZE ## _mc20_ ## MMX;\ 01611 static const qpel_mc_func OPNAME ## 2tap_qpel ## SIZE ## _mc12_ ## MMX =\ 01612 OPNAME ## 2tap_qpel ## SIZE ## _mc02_ ## MMX;\ 01613 static void OPNAME ## 2tap_qpel ## SIZE ## _mc32_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\ 01614 OPNAME ## pixels ## SIZE ## _y2_ ## MMX(dst, src+1, stride, SIZE);\ 01615 }\ 01616 static void OPNAME ## 2tap_qpel ## SIZE ## _mc23_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\ 01617 OPNAME ## pixels ## SIZE ## _x2_ ## MMX(dst, src+stride, stride, SIZE);\ 01618 }\ 01619 QPEL_2TAP_L3(OPNAME, SIZE, MMX, 10, 0, 1, 0)\ 01620 QPEL_2TAP_L3(OPNAME, SIZE, MMX, 30, 1, -1, 0)\ 01621 QPEL_2TAP_L3(OPNAME, SIZE, MMX, 01, 0, stride, 0)\ 01622 QPEL_2TAP_L3(OPNAME, SIZE, MMX, 03, stride, -stride, 0)\ 01623 QPEL_2TAP_L3(OPNAME, SIZE, MMX, 11, 0, stride, 1)\ 01624 QPEL_2TAP_L3(OPNAME, SIZE, MMX, 31, 1, stride, -1)\ 01625 QPEL_2TAP_L3(OPNAME, SIZE, MMX, 13, stride, -stride, 1)\ 01626 QPEL_2TAP_L3(OPNAME, SIZE, MMX, 33, stride+1, -stride, -1)\ 01627 01628 QPEL_2TAP(put_, 16, mmx2) 01629 QPEL_2TAP(avg_, 16, mmx2) 01630 QPEL_2TAP(put_, 8, mmx2) 01631 QPEL_2TAP(avg_, 8, mmx2) 01632 QPEL_2TAP(put_, 16, 3dnow) 01633 QPEL_2TAP(avg_, 16, 3dnow) 01634 QPEL_2TAP(put_, 8, 3dnow) 01635 QPEL_2TAP(avg_, 8, 3dnow) 01636 01637 01638 #if 0 01639 static void just_return(void) { return; } 01640 #endif 01641 01642 static void gmc_mmx(uint8_t *dst, uint8_t *src, int stride, int h, int ox, int oy, 01643 int dxx, int dxy, int dyx, int dyy, int shift, int r, int width, int height){ 01644 const int w = 8; 01645 const int ix = ox>>(16+shift); 01646 const int iy = oy>>(16+shift); 01647 const int oxs = ox>>4; 01648 const int oys = oy>>4; 01649 const int dxxs = dxx>>4; 01650 const int dxys = dxy>>4; 01651 const int dyxs = dyx>>4; 01652 const int dyys = dyy>>4; 01653 const uint16_t r4[4] = {r,r,r,r}; 01654 const uint16_t dxy4[4] = {dxys,dxys,dxys,dxys}; 01655 const uint16_t dyy4[4] = {dyys,dyys,dyys,dyys}; 01656 const uint64_t shift2 = 2*shift; 01657 uint8_t edge_buf[(h+1)*stride]; 01658 int x, y; 01659 01660 const int dxw = (dxx-(1<<(16+shift)))*(w-1); 01661 const int dyh = (dyy-(1<<(16+shift)))*(h-1); 01662 const int dxh = dxy*(h-1); 01663 const int dyw = dyx*(w-1); 01664 if( // non-constant fullpel offset (3% of blocks) 01665 ((ox^(ox+dxw)) | (ox^(ox+dxh)) | (ox^(ox+dxw+dxh)) | 01666 (oy^(oy+dyw)) | (oy^(oy+dyh)) | (oy^(oy+dyw+dyh))) >> (16+shift) 01667 // uses more than 16 bits of subpel mv (only at huge resolution) 01668 || (dxx|dxy|dyx|dyy)&15 ) 01669 { 01670 //FIXME could still use mmx for some of the rows 01671 ff_gmc_c(dst, src, stride, h, ox, oy, dxx, dxy, dyx, dyy, shift, r, width, height); 01672 return; 01673 } 01674 01675 src += ix + iy*stride; 01676 if( (unsigned)ix >= width-w || 01677 (unsigned)iy >= height-h ) 01678 { 01679 ff_emulated_edge_mc(edge_buf, src, stride, w+1, h+1, ix, iy, width, height); 01680 src = edge_buf; 01681 } 01682 01683 __asm__ volatile( 01684 "movd %0, %%mm6 \n\t" 01685 "pxor %%mm7, %%mm7 \n\t" 01686 "punpcklwd %%mm6, %%mm6 \n\t" 01687 "punpcklwd %%mm6, %%mm6 \n\t" 01688 :: "r"(1<<shift) 01689 ); 01690 01691 for(x=0; x<w; x+=4){ 01692 uint16_t dx4[4] = { oxs - dxys + dxxs*(x+0), 01693 oxs - dxys + dxxs*(x+1), 01694 oxs - dxys + dxxs*(x+2), 01695 oxs - dxys + dxxs*(x+3) }; 01696 uint16_t dy4[4] = { oys - dyys + dyxs*(x+0), 01697 oys - dyys + dyxs*(x+1), 01698 oys - dyys + dyxs*(x+2), 01699 oys - dyys + dyxs*(x+3) }; 01700 01701 for(y=0; y<h; y++){ 01702 __asm__ volatile( 01703 "movq %0, %%mm4 \n\t" 01704 "movq %1, %%mm5 \n\t" 01705 "paddw %2, %%mm4 \n\t" 01706 "paddw %3, %%mm5 \n\t" 01707 "movq %%mm4, %0 \n\t" 01708 "movq %%mm5, %1 \n\t" 01709 "psrlw 12,ドル %%mm4 \n\t" 01710 "psrlw 12,ドル %%mm5 \n\t" 01711 : "+m"(*dx4), "+m"(*dy4) 01712 : "m"(*dxy4), "m"(*dyy4) 01713 ); 01714 01715 __asm__ volatile( 01716 "movq %%mm6, %%mm2 \n\t" 01717 "movq %%mm6, %%mm1 \n\t" 01718 "psubw %%mm4, %%mm2 \n\t" 01719 "psubw %%mm5, %%mm1 \n\t" 01720 "movq %%mm2, %%mm0 \n\t" 01721 "movq %%mm4, %%mm3 \n\t" 01722 "pmullw %%mm1, %%mm0 \n\t" // (s-dx)*(s-dy) 01723 "pmullw %%mm5, %%mm3 \n\t" // dx*dy 01724 "pmullw %%mm5, %%mm2 \n\t" // (s-dx)*dy 01725 "pmullw %%mm4, %%mm1 \n\t" // dx*(s-dy) 01726 01727 "movd %4, %%mm5 \n\t" 01728 "movd %3, %%mm4 \n\t" 01729 "punpcklbw %%mm7, %%mm5 \n\t" 01730 "punpcklbw %%mm7, %%mm4 \n\t" 01731 "pmullw %%mm5, %%mm3 \n\t" // src[1,1] * dx*dy 01732 "pmullw %%mm4, %%mm2 \n\t" // src[0,1] * (s-dx)*dy 01733 01734 "movd %2, %%mm5 \n\t" 01735 "movd %1, %%mm4 \n\t" 01736 "punpcklbw %%mm7, %%mm5 \n\t" 01737 "punpcklbw %%mm7, %%mm4 \n\t" 01738 "pmullw %%mm5, %%mm1 \n\t" // src[1,0] * dx*(s-dy) 01739 "pmullw %%mm4, %%mm0 \n\t" // src[0,0] * (s-dx)*(s-dy) 01740 "paddw %5, %%mm1 \n\t" 01741 "paddw %%mm3, %%mm2 \n\t" 01742 "paddw %%mm1, %%mm0 \n\t" 01743 "paddw %%mm2, %%mm0 \n\t" 01744 01745 "psrlw %6, %%mm0 \n\t" 01746 "packuswb %%mm0, %%mm0 \n\t" 01747 "movd %%mm0, %0 \n\t" 01748 01749 : "=m"(dst[x+y*stride]) 01750 : "m"(src[0]), "m"(src[1]), 01751 "m"(src[stride]), "m"(src[stride+1]), 01752 "m"(*r4), "m"(shift2) 01753 ); 01754 src += stride; 01755 } 01756 src += 4-h*stride; 01757 } 01758 } 01759 01760 #define PREFETCH(name, op) \ 01761 static void name(void *mem, int stride, int h){\ 01762 const uint8_t *p= mem;\ 01763 do{\ 01764 __asm__ volatile(#op" %0" :: "m"(*p));\ 01765 p+= stride;\ 01766 }while(--h);\ 01767 } 01768 PREFETCH(prefetch_mmx2, prefetcht0) 01769 PREFETCH(prefetch_3dnow, prefetch) 01770 #undef PREFETCH 01771 01772 #include "h264dsp_mmx.c" 01773 #include "rv40dsp_mmx.c" 01774 01775 /* CAVS specific */ 01776 void ff_cavsdsp_init_mmx2(DSPContext* c, AVCodecContext *avctx); 01777 void ff_cavsdsp_init_3dnow(DSPContext* c, AVCodecContext *avctx); 01778 01779 void ff_put_cavs_qpel8_mc00_mmx2(uint8_t *dst, uint8_t *src, int stride) { 01780 put_pixels8_mmx(dst, src, stride, 8); 01781 } 01782 void ff_avg_cavs_qpel8_mc00_mmx2(uint8_t *dst, uint8_t *src, int stride) { 01783 avg_pixels8_mmx(dst, src, stride, 8); 01784 } 01785 void ff_put_cavs_qpel16_mc00_mmx2(uint8_t *dst, uint8_t *src, int stride) { 01786 put_pixels16_mmx(dst, src, stride, 16); 01787 } 01788 void ff_avg_cavs_qpel16_mc00_mmx2(uint8_t *dst, uint8_t *src, int stride) { 01789 avg_pixels16_mmx(dst, src, stride, 16); 01790 } 01791 01792 /* VC1 specific */ 01793 void ff_vc1dsp_init_mmx(DSPContext* dsp, AVCodecContext *avctx); 01794 01795 void ff_put_vc1_mspel_mc00_mmx(uint8_t *dst, const uint8_t *src, int stride, int rnd) { 01796 put_pixels8_mmx(dst, src, stride, 8); 01797 } 01798 01799 /* external functions, from idct_mmx.c */ 01800 void ff_mmx_idct(DCTELEM *block); 01801 void ff_mmxext_idct(DCTELEM *block); 01802 01803 /* XXX: those functions should be suppressed ASAP when all IDCTs are 01804 converted */ 01805 #if CONFIG_GPL 01806 static void ff_libmpeg2mmx_idct_put(uint8_t *dest, int line_size, DCTELEM *block) 01807 { 01808 ff_mmx_idct (block); 01809 put_pixels_clamped_mmx(block, dest, line_size); 01810 } 01811 static void ff_libmpeg2mmx_idct_add(uint8_t *dest, int line_size, DCTELEM *block) 01812 { 01813 ff_mmx_idct (block); 01814 add_pixels_clamped_mmx(block, dest, line_size); 01815 } 01816 static void ff_libmpeg2mmx2_idct_put(uint8_t *dest, int line_size, DCTELEM *block) 01817 { 01818 ff_mmxext_idct (block); 01819 put_pixels_clamped_mmx(block, dest, line_size); 01820 } 01821 static void ff_libmpeg2mmx2_idct_add(uint8_t *dest, int line_size, DCTELEM *block) 01822 { 01823 ff_mmxext_idct (block); 01824 add_pixels_clamped_mmx(block, dest, line_size); 01825 } 01826 #endif 01827 static void ff_idct_xvid_mmx_put(uint8_t *dest, int line_size, DCTELEM *block) 01828 { 01829 ff_idct_xvid_mmx (block); 01830 put_pixels_clamped_mmx(block, dest, line_size); 01831 } 01832 static void ff_idct_xvid_mmx_add(uint8_t *dest, int line_size, DCTELEM *block) 01833 { 01834 ff_idct_xvid_mmx (block); 01835 add_pixels_clamped_mmx(block, dest, line_size); 01836 } 01837 static void ff_idct_xvid_mmx2_put(uint8_t *dest, int line_size, DCTELEM *block) 01838 { 01839 ff_idct_xvid_mmx2 (block); 01840 put_pixels_clamped_mmx(block, dest, line_size); 01841 } 01842 static void ff_idct_xvid_mmx2_add(uint8_t *dest, int line_size, DCTELEM *block) 01843 { 01844 ff_idct_xvid_mmx2 (block); 01845 add_pixels_clamped_mmx(block, dest, line_size); 01846 } 01847 01848 static void vorbis_inverse_coupling_3dnow(float *mag, float *ang, int blocksize) 01849 { 01850 int i; 01851 __asm__ volatile("pxor %%mm7, %%mm7":); 01852 for(i=0; i<blocksize; i+=2) { 01853 __asm__ volatile( 01854 "movq %0, %%mm0 \n\t" 01855 "movq %1, %%mm1 \n\t" 01856 "movq %%mm0, %%mm2 \n\t" 01857 "movq %%mm1, %%mm3 \n\t" 01858 "pfcmpge %%mm7, %%mm2 \n\t" // m <= 0.0 01859 "pfcmpge %%mm7, %%mm3 \n\t" // a <= 0.0 01860 "pslld 31,ドル %%mm2 \n\t" // keep only the sign bit 01861 "pxor %%mm2, %%mm1 \n\t" 01862 "movq %%mm3, %%mm4 \n\t" 01863 "pand %%mm1, %%mm3 \n\t" 01864 "pandn %%mm1, %%mm4 \n\t" 01865 "pfadd %%mm0, %%mm3 \n\t" // a = m + ((a<0) & (a ^ sign(m))) 01866 "pfsub %%mm4, %%mm0 \n\t" // m = m + ((a>0) & (a ^ sign(m))) 01867 "movq %%mm3, %1 \n\t" 01868 "movq %%mm0, %0 \n\t" 01869 :"+m"(mag[i]), "+m"(ang[i]) 01870 ::"memory" 01871 ); 01872 } 01873 __asm__ volatile("femms"); 01874 } 01875 static void vorbis_inverse_coupling_sse(float *mag, float *ang, int blocksize) 01876 { 01877 int i; 01878 01879 __asm__ volatile( 01880 "movaps %0, %%xmm5 \n\t" 01881 ::"m"(ff_pdw_80000000[0]) 01882 ); 01883 for(i=0; i<blocksize; i+=4) { 01884 __asm__ volatile( 01885 "movaps %0, %%xmm0 \n\t" 01886 "movaps %1, %%xmm1 \n\t" 01887 "xorps %%xmm2, %%xmm2 \n\t" 01888 "xorps %%xmm3, %%xmm3 \n\t" 01889 "cmpleps %%xmm0, %%xmm2 \n\t" // m <= 0.0 01890 "cmpleps %%xmm1, %%xmm3 \n\t" // a <= 0.0 01891 "andps %%xmm5, %%xmm2 \n\t" // keep only the sign bit 01892 "xorps %%xmm2, %%xmm1 \n\t" 01893 "movaps %%xmm3, %%xmm4 \n\t" 01894 "andps %%xmm1, %%xmm3 \n\t" 01895 "andnps %%xmm1, %%xmm4 \n\t" 01896 "addps %%xmm0, %%xmm3 \n\t" // a = m + ((a<0) & (a ^ sign(m))) 01897 "subps %%xmm4, %%xmm0 \n\t" // m = m + ((a>0) & (a ^ sign(m))) 01898 "movaps %%xmm3, %1 \n\t" 01899 "movaps %%xmm0, %0 \n\t" 01900 :"+m"(mag[i]), "+m"(ang[i]) 01901 ::"memory" 01902 ); 01903 } 01904 } 01905 01906 #define IF1(x) x 01907 #define IF0(x) 01908 01909 #define MIX5(mono,stereo)\ 01910 __asm__ volatile(\ 01911 "movss 0(%2), %%xmm5 \n"\ 01912 "movss 8(%2), %%xmm6 \n"\ 01913 "movss 24(%2), %%xmm7 \n"\ 01914 "shufps 0,ドル %%xmm5, %%xmm5 \n"\ 01915 "shufps 0,ドル %%xmm6, %%xmm6 \n"\ 01916 "shufps 0,ドル %%xmm7, %%xmm7 \n"\ 01917 "1: \n"\ 01918 "movaps (%0,%1), %%xmm0 \n"\ 01919 "movaps 0x400(%0,%1), %%xmm1 \n"\ 01920 "movaps 0x800(%0,%1), %%xmm2 \n"\ 01921 "movaps 0xc00(%0,%1), %%xmm3 \n"\ 01922 "movaps 0x1000(%0,%1), %%xmm4 \n"\ 01923 "mulps %%xmm5, %%xmm0 \n"\ 01924 "mulps %%xmm6, %%xmm1 \n"\ 01925 "mulps %%xmm5, %%xmm2 \n"\ 01926 "mulps %%xmm7, %%xmm3 \n"\ 01927 "mulps %%xmm7, %%xmm4 \n"\ 01928 stereo("addps %%xmm1, %%xmm0 \n")\ 01929 "addps %%xmm1, %%xmm2 \n"\ 01930 "addps %%xmm3, %%xmm0 \n"\ 01931 "addps %%xmm4, %%xmm2 \n"\ 01932 mono("addps %%xmm2, %%xmm0 \n")\ 01933 "movaps %%xmm0, (%0,%1) \n"\ 01934 stereo("movaps %%xmm2, 0x400(%0,%1) \n")\ 01935 "add 16,ドル %0 \n"\ 01936 "jl 1b \n"\ 01937 :"+&r"(i)\ 01938 :"r"(samples[0]+len), "r"(matrix)\ 01939 :"memory"\ 01940 ); 01941 01942 #define MIX_MISC(stereo)\ 01943 __asm__ volatile(\ 01944 "1: \n"\ 01945 "movaps (%3,%0), %%xmm0 \n"\ 01946 stereo("movaps %%xmm0, %%xmm1 \n")\ 01947 "mulps %%xmm6, %%xmm0 \n"\ 01948 stereo("mulps %%xmm7, %%xmm1 \n")\ 01949 "lea 1024(%3,%0), %1 \n"\ 01950 "mov %5, %2 \n"\ 01951 "2: \n"\ 01952 "movaps (%1), %%xmm2 \n"\ 01953 stereo("movaps %%xmm2, %%xmm3 \n")\ 01954 "mulps (%4,%2), %%xmm2 \n"\ 01955 stereo("mulps 16(%4,%2), %%xmm3 \n")\ 01956 "addps %%xmm2, %%xmm0 \n"\ 01957 stereo("addps %%xmm3, %%xmm1 \n")\ 01958 "add 1024,ドル %1 \n"\ 01959 "add 32,ドル %2 \n"\ 01960 "jl 2b \n"\ 01961 "movaps %%xmm0, (%3,%0) \n"\ 01962 stereo("movaps %%xmm1, 1024(%3,%0) \n")\ 01963 "add 16,ドル %0 \n"\ 01964 "jl 1b \n"\ 01965 :"+&r"(i), "=&r"(j), "=&r"(k)\ 01966 :"r"(samples[0]+len), "r"(matrix_simd+in_ch), "g"((intptr_t)-32*(in_ch-1))\ 01967 :"memory"\ 01968 ); 01969 01970 static void ac3_downmix_sse(float (*samples)[256], float (*matrix)[2], int out_ch, int in_ch, int len) 01971 { 01972 int (*matrix_cmp)[2] = (int(*)[2])matrix; 01973 intptr_t i,j,k; 01974 01975 i = -len*sizeof(float); 01976 if(in_ch == 5 && out_ch == 2 && !(matrix_cmp[0][1]|matrix_cmp[2][0]|matrix_cmp[3][1]|matrix_cmp[4][0]|(matrix_cmp[1][0]^matrix_cmp[1][1])|(matrix_cmp[0][0]^matrix_cmp[2][1]))) { 01977 MIX5(IF0,IF1); 01978 } else if(in_ch == 5 && out_ch == 1 && matrix_cmp[0][0]==matrix_cmp[2][0] && matrix_cmp[3][0]==matrix_cmp[4][0]) { 01979 MIX5(IF1,IF0); 01980 } else { 01981 DECLARE_ALIGNED_16(float, matrix_simd[in_ch][2][4]); 01982 j = 2*in_ch*sizeof(float); 01983 __asm__ volatile( 01984 "1: \n" 01985 "sub 8,ドル %0 \n" 01986 "movss (%2,%0), %%xmm6 \n" 01987 "movss 4(%2,%0), %%xmm7 \n" 01988 "shufps 0,ドル %%xmm6, %%xmm6 \n" 01989 "shufps 0,ドル %%xmm7, %%xmm7 \n" 01990 "movaps %%xmm6, (%1,%0,4) \n" 01991 "movaps %%xmm7, 16(%1,%0,4) \n" 01992 "jg 1b \n" 01993 :"+&r"(j) 01994 :"r"(matrix_simd), "r"(matrix) 01995 :"memory" 01996 ); 01997 if(out_ch == 2) { 01998 MIX_MISC(IF1); 01999 } else { 02000 MIX_MISC(IF0); 02001 } 02002 } 02003 } 02004 02005 static void vector_fmul_3dnow(float *dst, const float *src, int len){ 02006 x86_reg i = (len-4)*4; 02007 __asm__ volatile( 02008 "1: \n\t" 02009 "movq (%1,%0), %%mm0 \n\t" 02010 "movq 8(%1,%0), %%mm1 \n\t" 02011 "pfmul (%2,%0), %%mm0 \n\t" 02012 "pfmul 8(%2,%0), %%mm1 \n\t" 02013 "movq %%mm0, (%1,%0) \n\t" 02014 "movq %%mm1, 8(%1,%0) \n\t" 02015 "sub 16,ドル %0 \n\t" 02016 "jge 1b \n\t" 02017 "femms \n\t" 02018 :"+r"(i) 02019 :"r"(dst), "r"(src) 02020 :"memory" 02021 ); 02022 } 02023 static void vector_fmul_sse(float *dst, const float *src, int len){ 02024 x86_reg i = (len-8)*4; 02025 __asm__ volatile( 02026 "1: \n\t" 02027 "movaps (%1,%0), %%xmm0 \n\t" 02028 "movaps 16(%1,%0), %%xmm1 \n\t" 02029 "mulps (%2,%0), %%xmm0 \n\t" 02030 "mulps 16(%2,%0), %%xmm1 \n\t" 02031 "movaps %%xmm0, (%1,%0) \n\t" 02032 "movaps %%xmm1, 16(%1,%0) \n\t" 02033 "sub 32,ドル %0 \n\t" 02034 "jge 1b \n\t" 02035 :"+r"(i) 02036 :"r"(dst), "r"(src) 02037 :"memory" 02038 ); 02039 } 02040 02041 static void vector_fmul_reverse_3dnow2(float *dst, const float *src0, const float *src1, int len){ 02042 x86_reg i = len*4-16; 02043 __asm__ volatile( 02044 "1: \n\t" 02045 "pswapd 8(%1), %%mm0 \n\t" 02046 "pswapd (%1), %%mm1 \n\t" 02047 "pfmul (%3,%0), %%mm0 \n\t" 02048 "pfmul 8(%3,%0), %%mm1 \n\t" 02049 "movq %%mm0, (%2,%0) \n\t" 02050 "movq %%mm1, 8(%2,%0) \n\t" 02051 "add 16,ドル %1 \n\t" 02052 "sub 16,ドル %0 \n\t" 02053 "jge 1b \n\t" 02054 :"+r"(i), "+r"(src1) 02055 :"r"(dst), "r"(src0) 02056 ); 02057 __asm__ volatile("femms"); 02058 } 02059 static void vector_fmul_reverse_sse(float *dst, const float *src0, const float *src1, int len){ 02060 x86_reg i = len*4-32; 02061 __asm__ volatile( 02062 "1: \n\t" 02063 "movaps 16(%1), %%xmm0 \n\t" 02064 "movaps (%1), %%xmm1 \n\t" 02065 "shufps 0ドルx1b, %%xmm0, %%xmm0 \n\t" 02066 "shufps 0ドルx1b, %%xmm1, %%xmm1 \n\t" 02067 "mulps (%3,%0), %%xmm0 \n\t" 02068 "mulps 16(%3,%0), %%xmm1 \n\t" 02069 "movaps %%xmm0, (%2,%0) \n\t" 02070 "movaps %%xmm1, 16(%2,%0) \n\t" 02071 "add 32,ドル %1 \n\t" 02072 "sub 32,ドル %0 \n\t" 02073 "jge 1b \n\t" 02074 :"+r"(i), "+r"(src1) 02075 :"r"(dst), "r"(src0) 02076 ); 02077 } 02078 02079 static void vector_fmul_add_add_3dnow(float *dst, const float *src0, const float *src1, 02080 const float *src2, int src3, int len, int step){ 02081 x86_reg i = (len-4)*4; 02082 if(step == 2 && src3 == 0){ 02083 dst += (len-4)*2; 02084 __asm__ volatile( 02085 "1: \n\t" 02086 "movq (%2,%0), %%mm0 \n\t" 02087 "movq 8(%2,%0), %%mm1 \n\t" 02088 "pfmul (%3,%0), %%mm0 \n\t" 02089 "pfmul 8(%3,%0), %%mm1 \n\t" 02090 "pfadd (%4,%0), %%mm0 \n\t" 02091 "pfadd 8(%4,%0), %%mm1 \n\t" 02092 "movd %%mm0, (%1) \n\t" 02093 "movd %%mm1, 16(%1) \n\t" 02094 "psrlq 32,ドル %%mm0 \n\t" 02095 "psrlq 32,ドル %%mm1 \n\t" 02096 "movd %%mm0, 8(%1) \n\t" 02097 "movd %%mm1, 24(%1) \n\t" 02098 "sub 32,ドル %1 \n\t" 02099 "sub 16,ドル %0 \n\t" 02100 "jge 1b \n\t" 02101 :"+r"(i), "+r"(dst) 02102 :"r"(src0), "r"(src1), "r"(src2) 02103 :"memory" 02104 ); 02105 } 02106 else if(step == 1 && src3 == 0){ 02107 __asm__ volatile( 02108 "1: \n\t" 02109 "movq (%2,%0), %%mm0 \n\t" 02110 "movq 8(%2,%0), %%mm1 \n\t" 02111 "pfmul (%3,%0), %%mm0 \n\t" 02112 "pfmul 8(%3,%0), %%mm1 \n\t" 02113 "pfadd (%4,%0), %%mm0 \n\t" 02114 "pfadd 8(%4,%0), %%mm1 \n\t" 02115 "movq %%mm0, (%1,%0) \n\t" 02116 "movq %%mm1, 8(%1,%0) \n\t" 02117 "sub 16,ドル %0 \n\t" 02118 "jge 1b \n\t" 02119 :"+r"(i) 02120 :"r"(dst), "r"(src0), "r"(src1), "r"(src2) 02121 :"memory" 02122 ); 02123 } 02124 else 02125 ff_vector_fmul_add_add_c(dst, src0, src1, src2, src3, len, step); 02126 __asm__ volatile("femms"); 02127 } 02128 static void vector_fmul_add_add_sse(float *dst, const float *src0, const float *src1, 02129 const float *src2, int src3, int len, int step){ 02130 x86_reg i = (len-8)*4; 02131 if(step == 2 && src3 == 0){ 02132 dst += (len-8)*2; 02133 __asm__ volatile( 02134 "1: \n\t" 02135 "movaps (%2,%0), %%xmm0 \n\t" 02136 "movaps 16(%2,%0), %%xmm1 \n\t" 02137 "mulps (%3,%0), %%xmm0 \n\t" 02138 "mulps 16(%3,%0), %%xmm1 \n\t" 02139 "addps (%4,%0), %%xmm0 \n\t" 02140 "addps 16(%4,%0), %%xmm1 \n\t" 02141 "movss %%xmm0, (%1) \n\t" 02142 "movss %%xmm1, 32(%1) \n\t" 02143 "movhlps %%xmm0, %%xmm2 \n\t" 02144 "movhlps %%xmm1, %%xmm3 \n\t" 02145 "movss %%xmm2, 16(%1) \n\t" 02146 "movss %%xmm3, 48(%1) \n\t" 02147 "shufps 0ドルxb1, %%xmm0, %%xmm0 \n\t" 02148 "shufps 0ドルxb1, %%xmm1, %%xmm1 \n\t" 02149 "movss %%xmm0, 8(%1) \n\t" 02150 "movss %%xmm1, 40(%1) \n\t" 02151 "movhlps %%xmm0, %%xmm2 \n\t" 02152 "movhlps %%xmm1, %%xmm3 \n\t" 02153 "movss %%xmm2, 24(%1) \n\t" 02154 "movss %%xmm3, 56(%1) \n\t" 02155 "sub 64,ドル %1 \n\t" 02156 "sub 32,ドル %0 \n\t" 02157 "jge 1b \n\t" 02158 :"+r"(i), "+r"(dst) 02159 :"r"(src0), "r"(src1), "r"(src2) 02160 :"memory" 02161 ); 02162 } 02163 else if(step == 1 && src3 == 0){ 02164 __asm__ volatile( 02165 "1: \n\t" 02166 "movaps (%2,%0), %%xmm0 \n\t" 02167 "movaps 16(%2,%0), %%xmm1 \n\t" 02168 "mulps (%3,%0), %%xmm0 \n\t" 02169 "mulps 16(%3,%0), %%xmm1 \n\t" 02170 "addps (%4,%0), %%xmm0 \n\t" 02171 "addps 16(%4,%0), %%xmm1 \n\t" 02172 "movaps %%xmm0, (%1,%0) \n\t" 02173 "movaps %%xmm1, 16(%1,%0) \n\t" 02174 "sub 32,ドル %0 \n\t" 02175 "jge 1b \n\t" 02176 :"+r"(i) 02177 :"r"(dst), "r"(src0), "r"(src1), "r"(src2) 02178 :"memory" 02179 ); 02180 } 02181 else 02182 ff_vector_fmul_add_add_c(dst, src0, src1, src2, src3, len, step); 02183 } 02184 02185 static void vector_fmul_window_3dnow2(float *dst, const float *src0, const float *src1, 02186 const float *win, float add_bias, int len){ 02187 #if HAVE_6REGS 02188 if(add_bias == 0){ 02189 x86_reg i = -len*4; 02190 x86_reg j = len*4-8; 02191 __asm__ volatile( 02192 "1: \n" 02193 "pswapd (%5,%1), %%mm1 \n" 02194 "movq (%5,%0), %%mm0 \n" 02195 "pswapd (%4,%1), %%mm5 \n" 02196 "movq (%3,%0), %%mm4 \n" 02197 "movq %%mm0, %%mm2 \n" 02198 "movq %%mm1, %%mm3 \n" 02199 "pfmul %%mm4, %%mm2 \n" // src0[len+i]*win[len+i] 02200 "pfmul %%mm5, %%mm3 \n" // src1[ j]*win[len+j] 02201 "pfmul %%mm4, %%mm1 \n" // src0[len+i]*win[len+j] 02202 "pfmul %%mm5, %%mm0 \n" // src1[ j]*win[len+i] 02203 "pfadd %%mm3, %%mm2 \n" 02204 "pfsub %%mm0, %%mm1 \n" 02205 "pswapd %%mm2, %%mm2 \n" 02206 "movq %%mm1, (%2,%0) \n" 02207 "movq %%mm2, (%2,%1) \n" 02208 "sub 8,ドル %1 \n" 02209 "add 8,ドル %0 \n" 02210 "jl 1b \n" 02211 "femms \n" 02212 :"+r"(i), "+r"(j) 02213 :"r"(dst+len), "r"(src0+len), "r"(src1), "r"(win+len) 02214 ); 02215 }else 02216 #endif 02217 ff_vector_fmul_window_c(dst, src0, src1, win, add_bias, len); 02218 } 02219 02220 static void vector_fmul_window_sse(float *dst, const float *src0, const float *src1, 02221 const float *win, float add_bias, int len){ 02222 #if HAVE_6REGS 02223 if(add_bias == 0){ 02224 x86_reg i = -len*4; 02225 x86_reg j = len*4-16; 02226 __asm__ volatile( 02227 "1: \n" 02228 "movaps (%5,%1), %%xmm1 \n" 02229 "movaps (%5,%0), %%xmm0 \n" 02230 "movaps (%4,%1), %%xmm5 \n" 02231 "movaps (%3,%0), %%xmm4 \n" 02232 "shufps 0ドルx1b, %%xmm1, %%xmm1 \n" 02233 "shufps 0ドルx1b, %%xmm5, %%xmm5 \n" 02234 "movaps %%xmm0, %%xmm2 \n" 02235 "movaps %%xmm1, %%xmm3 \n" 02236 "mulps %%xmm4, %%xmm2 \n" // src0[len+i]*win[len+i] 02237 "mulps %%xmm5, %%xmm3 \n" // src1[ j]*win[len+j] 02238 "mulps %%xmm4, %%xmm1 \n" // src0[len+i]*win[len+j] 02239 "mulps %%xmm5, %%xmm0 \n" // src1[ j]*win[len+i] 02240 "addps %%xmm3, %%xmm2 \n" 02241 "subps %%xmm0, %%xmm1 \n" 02242 "shufps 0ドルx1b, %%xmm2, %%xmm2 \n" 02243 "movaps %%xmm1, (%2,%0) \n" 02244 "movaps %%xmm2, (%2,%1) \n" 02245 "sub 16,ドル %1 \n" 02246 "add 16,ドル %0 \n" 02247 "jl 1b \n" 02248 :"+r"(i), "+r"(j) 02249 :"r"(dst+len), "r"(src0+len), "r"(src1), "r"(win+len) 02250 ); 02251 }else 02252 #endif 02253 ff_vector_fmul_window_c(dst, src0, src1, win, add_bias, len); 02254 } 02255 02256 static void int32_to_float_fmul_scalar_sse(float *dst, const int *src, float mul, int len) 02257 { 02258 x86_reg i = -4*len; 02259 __asm__ volatile( 02260 "movss %3, %%xmm4 \n" 02261 "shufps 0,ドル %%xmm4, %%xmm4 \n" 02262 "1: \n" 02263 "cvtpi2ps (%2,%0), %%xmm0 \n" 02264 "cvtpi2ps 8(%2,%0), %%xmm1 \n" 02265 "cvtpi2ps 16(%2,%0), %%xmm2 \n" 02266 "cvtpi2ps 24(%2,%0), %%xmm3 \n" 02267 "movlhps %%xmm1, %%xmm0 \n" 02268 "movlhps %%xmm3, %%xmm2 \n" 02269 "mulps %%xmm4, %%xmm0 \n" 02270 "mulps %%xmm4, %%xmm2 \n" 02271 "movaps %%xmm0, (%1,%0) \n" 02272 "movaps %%xmm2, 16(%1,%0) \n" 02273 "add 32,ドル %0 \n" 02274 "jl 1b \n" 02275 :"+r"(i) 02276 :"r"(dst+len), "r"(src+len), "m"(mul) 02277 ); 02278 } 02279 02280 static void int32_to_float_fmul_scalar_sse2(float *dst, const int *src, float mul, int len) 02281 { 02282 x86_reg i = -4*len; 02283 __asm__ volatile( 02284 "movss %3, %%xmm4 \n" 02285 "shufps 0,ドル %%xmm4, %%xmm4 \n" 02286 "1: \n" 02287 "cvtdq2ps (%2,%0), %%xmm0 \n" 02288 "cvtdq2ps 16(%2,%0), %%xmm1 \n" 02289 "mulps %%xmm4, %%xmm0 \n" 02290 "mulps %%xmm4, %%xmm1 \n" 02291 "movaps %%xmm0, (%1,%0) \n" 02292 "movaps %%xmm1, 16(%1,%0) \n" 02293 "add 32,ドル %0 \n" 02294 "jl 1b \n" 02295 :"+r"(i) 02296 :"r"(dst+len), "r"(src+len), "m"(mul) 02297 ); 02298 } 02299 02300 static void float_to_int16_3dnow(int16_t *dst, const float *src, long len){ 02301 x86_reg reglen = len; 02302 // not bit-exact: pf2id uses different rounding than C and SSE 02303 __asm__ volatile( 02304 "add %0 , %0 \n\t" 02305 "lea (%2,%0,2) , %2 \n\t" 02306 "add %0 , %1 \n\t" 02307 "neg %0 \n\t" 02308 "1: \n\t" 02309 "pf2id (%2,%0,2) , %%mm0 \n\t" 02310 "pf2id 8(%2,%0,2) , %%mm1 \n\t" 02311 "pf2id 16(%2,%0,2) , %%mm2 \n\t" 02312 "pf2id 24(%2,%0,2) , %%mm3 \n\t" 02313 "packssdw %%mm1 , %%mm0 \n\t" 02314 "packssdw %%mm3 , %%mm2 \n\t" 02315 "movq %%mm0 , (%1,%0) \n\t" 02316 "movq %%mm2 , 8(%1,%0) \n\t" 02317 "add 16ドル , %0 \n\t" 02318 " js 1b \n\t" 02319 "femms \n\t" 02320 :"+r"(reglen), "+r"(dst), "+r"(src) 02321 ); 02322 } 02323 static void float_to_int16_sse(int16_t *dst, const float *src, long len){ 02324 x86_reg reglen = len; 02325 __asm__ volatile( 02326 "add %0 , %0 \n\t" 02327 "lea (%2,%0,2) , %2 \n\t" 02328 "add %0 , %1 \n\t" 02329 "neg %0 \n\t" 02330 "1: \n\t" 02331 "cvtps2pi (%2,%0,2) , %%mm0 \n\t" 02332 "cvtps2pi 8(%2,%0,2) , %%mm1 \n\t" 02333 "cvtps2pi 16(%2,%0,2) , %%mm2 \n\t" 02334 "cvtps2pi 24(%2,%0,2) , %%mm3 \n\t" 02335 "packssdw %%mm1 , %%mm0 \n\t" 02336 "packssdw %%mm3 , %%mm2 \n\t" 02337 "movq %%mm0 , (%1,%0) \n\t" 02338 "movq %%mm2 , 8(%1,%0) \n\t" 02339 "add 16ドル , %0 \n\t" 02340 " js 1b \n\t" 02341 "emms \n\t" 02342 :"+r"(reglen), "+r"(dst), "+r"(src) 02343 ); 02344 } 02345 02346 static void float_to_int16_sse2(int16_t *dst, const float *src, long len){ 02347 x86_reg reglen = len; 02348 __asm__ volatile( 02349 "add %0 , %0 \n\t" 02350 "lea (%2,%0,2) , %2 \n\t" 02351 "add %0 , %1 \n\t" 02352 "neg %0 \n\t" 02353 "1: \n\t" 02354 "cvtps2dq (%2,%0,2) , %%xmm0 \n\t" 02355 "cvtps2dq 16(%2,%0,2) , %%xmm1 \n\t" 02356 "packssdw %%xmm1 , %%xmm0 \n\t" 02357 "movdqa %%xmm0 , (%1,%0) \n\t" 02358 "add 16ドル , %0 \n\t" 02359 " js 1b \n\t" 02360 :"+r"(reglen), "+r"(dst), "+r"(src) 02361 ); 02362 } 02363 02364 #if HAVE_YASM 02365 void ff_float_to_int16_interleave6_sse(int16_t *dst, const float **src, int len); 02366 void ff_float_to_int16_interleave6_3dnow(int16_t *dst, const float **src, int len); 02367 void ff_float_to_int16_interleave6_3dn2(int16_t *dst, const float **src, int len); 02368 void ff_add_hfyu_median_prediction_mmx2(uint8_t *dst, uint8_t *top, uint8_t *diff, int w, int *left, int *left_top); 02369 void ff_x264_deblock_v_luma_sse2(uint8_t *pix, int stride, int alpha, int beta, int8_t *tc0); 02370 void ff_x264_deblock_h_luma_sse2(uint8_t *pix, int stride, int alpha, int beta, int8_t *tc0); 02371 void ff_x264_deblock_v8_luma_intra_mmxext(uint8_t *pix, int stride, int alpha, int beta); 02372 void ff_x264_deblock_h_luma_intra_mmxext(uint8_t *pix, int stride, int alpha, int beta); 02373 #if ARCH_X86_32 02374 static void ff_x264_deblock_v_luma_intra_mmxext(uint8_t *pix, int stride, int alpha, int beta) 02375 { 02376 ff_x264_deblock_v8_luma_intra_mmxext(pix+0, stride, alpha, beta); 02377 ff_x264_deblock_v8_luma_intra_mmxext(pix+8, stride, alpha, beta); 02378 } 02379 #endif 02380 void ff_x264_deblock_v_luma_intra_sse2(uint8_t *pix, int stride, int alpha, int beta); 02381 void ff_x264_deblock_h_luma_intra_sse2(uint8_t *pix, int stride, int alpha, int beta); 02382 #else 02383 #define ff_float_to_int16_interleave6_sse(a,b,c) float_to_int16_interleave_misc_sse(a,b,c,6) 02384 #define ff_float_to_int16_interleave6_3dnow(a,b,c) float_to_int16_interleave_misc_3dnow(a,b,c,6) 02385 #define ff_float_to_int16_interleave6_3dn2(a,b,c) float_to_int16_interleave_misc_3dnow(a,b,c,6) 02386 #endif 02387 #define ff_float_to_int16_interleave6_sse2 ff_float_to_int16_interleave6_sse 02388 02389 #define FLOAT_TO_INT16_INTERLEAVE(cpu, body) \ 02390 /* gcc pessimizes register allocation if this is in the same function as float_to_int16_interleave_sse2*/\ 02391 static av_noinline void float_to_int16_interleave_misc_##cpu(int16_t *dst, const float **src, long len, int channels){\ 02392 DECLARE_ALIGNED_16(int16_t, tmp[len]);\ 02393 int i,j,c;\ 02394 for(c=0; c<channels; c++){\ 02395 float_to_int16_##cpu(tmp, src[c], len);\ 02396 for(i=0, j=c; i<len; i++, j+=channels)\ 02397 dst[j] = tmp[i];\ 02398 }\ 02399 }\ 02400 \ 02401 static void float_to_int16_interleave_##cpu(int16_t *dst, const float **src, long len, int channels){\ 02402 if(channels==1)\ 02403 float_to_int16_##cpu(dst, src[0], len);\ 02404 else if(channels==2){\ 02405 x86_reg reglen = len; \ 02406 const float *src0 = src[0];\ 02407 const float *src1 = src[1];\ 02408 __asm__ volatile(\ 02409 "shl 2,ドル %0 \n"\ 02410 "add %0, %1 \n"\ 02411 "add %0, %2 \n"\ 02412 "add %0, %3 \n"\ 02413 "neg %0 \n"\ 02414 body\ 02415 :"+r"(reglen), "+r"(dst), "+r"(src0), "+r"(src1)\ 02416 );\ 02417 }else if(channels==6){\ 02418 ff_float_to_int16_interleave6_##cpu(dst, src, len);\ 02419 }else\ 02420 float_to_int16_interleave_misc_##cpu(dst, src, len, channels);\ 02421 } 02422 02423 FLOAT_TO_INT16_INTERLEAVE(3dnow, 02424 "1: \n" 02425 "pf2id (%2,%0), %%mm0 \n" 02426 "pf2id 8(%2,%0), %%mm1 \n" 02427 "pf2id (%3,%0), %%mm2 \n" 02428 "pf2id 8(%3,%0), %%mm3 \n" 02429 "packssdw %%mm1, %%mm0 \n" 02430 "packssdw %%mm3, %%mm2 \n" 02431 "movq %%mm0, %%mm1 \n" 02432 "punpcklwd %%mm2, %%mm0 \n" 02433 "punpckhwd %%mm2, %%mm1 \n" 02434 "movq %%mm0, (%1,%0)\n" 02435 "movq %%mm1, 8(%1,%0)\n" 02436 "add 16,ドル %0 \n" 02437 "js 1b \n" 02438 "femms \n" 02439 ) 02440 02441 FLOAT_TO_INT16_INTERLEAVE(sse, 02442 "1: \n" 02443 "cvtps2pi (%2,%0), %%mm0 \n" 02444 "cvtps2pi 8(%2,%0), %%mm1 \n" 02445 "cvtps2pi (%3,%0), %%mm2 \n" 02446 "cvtps2pi 8(%3,%0), %%mm3 \n" 02447 "packssdw %%mm1, %%mm0 \n" 02448 "packssdw %%mm3, %%mm2 \n" 02449 "movq %%mm0, %%mm1 \n" 02450 "punpcklwd %%mm2, %%mm0 \n" 02451 "punpckhwd %%mm2, %%mm1 \n" 02452 "movq %%mm0, (%1,%0)\n" 02453 "movq %%mm1, 8(%1,%0)\n" 02454 "add 16,ドル %0 \n" 02455 "js 1b \n" 02456 "emms \n" 02457 ) 02458 02459 FLOAT_TO_INT16_INTERLEAVE(sse2, 02460 "1: \n" 02461 "cvtps2dq (%2,%0), %%xmm0 \n" 02462 "cvtps2dq (%3,%0), %%xmm1 \n" 02463 "packssdw %%xmm1, %%xmm0 \n" 02464 "movhlps %%xmm0, %%xmm1 \n" 02465 "punpcklwd %%xmm1, %%xmm0 \n" 02466 "movdqa %%xmm0, (%1,%0) \n" 02467 "add 16,ドル %0 \n" 02468 "js 1b \n" 02469 ) 02470 02471 static void float_to_int16_interleave_3dn2(int16_t *dst, const float **src, long len, int channels){ 02472 if(channels==6) 02473 ff_float_to_int16_interleave6_3dn2(dst, src, len); 02474 else 02475 float_to_int16_interleave_3dnow(dst, src, len, channels); 02476 } 02477 02478 02479 void ff_snow_horizontal_compose97i_sse2(IDWTELEM *b, int width); 02480 void ff_snow_horizontal_compose97i_mmx(IDWTELEM *b, int width); 02481 void ff_snow_vertical_compose97i_sse2(IDWTELEM *b0, IDWTELEM *b1, IDWTELEM *b2, IDWTELEM *b3, IDWTELEM *b4, IDWTELEM *b5, int width); 02482 void ff_snow_vertical_compose97i_mmx(IDWTELEM *b0, IDWTELEM *b1, IDWTELEM *b2, IDWTELEM *b3, IDWTELEM *b4, IDWTELEM *b5, int width); 02483 void ff_snow_inner_add_yblock_sse2(const uint8_t *obmc, const int obmc_stride, uint8_t * * block, int b_w, int b_h, 02484 int src_x, int src_y, int src_stride, slice_buffer * sb, int add, uint8_t * dst8); 02485 void ff_snow_inner_add_yblock_mmx(const uint8_t *obmc, const int obmc_stride, uint8_t * * block, int b_w, int b_h, 02486 int src_x, int src_y, int src_stride, slice_buffer * sb, int add, uint8_t * dst8); 02487 02488 02489 static void add_int16_sse2(int16_t * v1, int16_t * v2, int order) 02490 { 02491 x86_reg o = -(order << 1); 02492 v1 += order; 02493 v2 += order; 02494 __asm__ volatile( 02495 "1: \n\t" 02496 "movdqu (%1,%2), %%xmm0 \n\t" 02497 "movdqu 16(%1,%2), %%xmm1 \n\t" 02498 "paddw (%0,%2), %%xmm0 \n\t" 02499 "paddw 16(%0,%2), %%xmm1 \n\t" 02500 "movdqa %%xmm0, (%0,%2) \n\t" 02501 "movdqa %%xmm1, 16(%0,%2) \n\t" 02502 "add 32,ドル %2 \n\t" 02503 "js 1b \n\t" 02504 : "+r"(v1), "+r"(v2), "+r"(o) 02505 ); 02506 } 02507 02508 static void sub_int16_sse2(int16_t * v1, int16_t * v2, int order) 02509 { 02510 x86_reg o = -(order << 1); 02511 v1 += order; 02512 v2 += order; 02513 __asm__ volatile( 02514 "1: \n\t" 02515 "movdqa (%0,%2), %%xmm0 \n\t" 02516 "movdqa 16(%0,%2), %%xmm2 \n\t" 02517 "movdqu (%1,%2), %%xmm1 \n\t" 02518 "movdqu 16(%1,%2), %%xmm3 \n\t" 02519 "psubw %%xmm1, %%xmm0 \n\t" 02520 "psubw %%xmm3, %%xmm2 \n\t" 02521 "movdqa %%xmm0, (%0,%2) \n\t" 02522 "movdqa %%xmm2, 16(%0,%2) \n\t" 02523 "add 32,ドル %2 \n\t" 02524 "js 1b \n\t" 02525 : "+r"(v1), "+r"(v2), "+r"(o) 02526 ); 02527 } 02528 02529 static int32_t scalarproduct_int16_sse2(int16_t * v1, int16_t * v2, int order, int shift) 02530 { 02531 int res = 0; 02532 DECLARE_ALIGNED_16(xmm_reg, sh); 02533 x86_reg o = -(order << 1); 02534 02535 v1 += order; 02536 v2 += order; 02537 sh.a = shift; 02538 __asm__ volatile( 02539 "pxor %%xmm7, %%xmm7 \n\t" 02540 "1: \n\t" 02541 "movdqu (%0,%3), %%xmm0 \n\t" 02542 "movdqu 16(%0,%3), %%xmm1 \n\t" 02543 "pmaddwd (%1,%3), %%xmm0 \n\t" 02544 "pmaddwd 16(%1,%3), %%xmm1 \n\t" 02545 "paddd %%xmm0, %%xmm7 \n\t" 02546 "paddd %%xmm1, %%xmm7 \n\t" 02547 "add 32,ドル %3 \n\t" 02548 "js 1b \n\t" 02549 "movhlps %%xmm7, %%xmm2 \n\t" 02550 "paddd %%xmm2, %%xmm7 \n\t" 02551 "psrad %4, %%xmm7 \n\t" 02552 "pshuflw 0ドルx4E, %%xmm7,%%xmm2 \n\t" 02553 "paddd %%xmm2, %%xmm7 \n\t" 02554 "movd %%xmm7, %2 \n\t" 02555 : "+r"(v1), "+r"(v2), "=r"(res), "+r"(o) 02556 : "m"(sh) 02557 ); 02558 return res; 02559 } 02560 02561 void dsputil_init_mmx(DSPContext* c, AVCodecContext *avctx) 02562 { 02563 mm_flags = mm_support(); 02564 02565 if (avctx->dsp_mask) { 02566 if (avctx->dsp_mask & FF_MM_FORCE) 02567 mm_flags |= (avctx->dsp_mask & 0xffff); 02568 else 02569 mm_flags &= ~(avctx->dsp_mask & 0xffff); 02570 } 02571 02572 #if 0 02573 av_log(avctx, AV_LOG_INFO, "libavcodec: CPU flags:"); 02574 if (mm_flags & FF_MM_MMX) 02575 av_log(avctx, AV_LOG_INFO, " mmx"); 02576 if (mm_flags & FF_MM_MMXEXT) 02577 av_log(avctx, AV_LOG_INFO, " mmxext"); 02578 if (mm_flags & FF_MM_3DNOW) 02579 av_log(avctx, AV_LOG_INFO, " 3dnow"); 02580 if (mm_flags & FF_MM_SSE) 02581 av_log(avctx, AV_LOG_INFO, " sse"); 02582 if (mm_flags & FF_MM_SSE2) 02583 av_log(avctx, AV_LOG_INFO, " sse2"); 02584 av_log(avctx, AV_LOG_INFO, "\n"); 02585 #endif 02586 02587 if (mm_flags & FF_MM_MMX) { 02588 const int idct_algo= avctx->idct_algo; 02589 02590 if(avctx->lowres==0){ 02591 if(idct_algo==FF_IDCT_AUTO || idct_algo==FF_IDCT_SIMPLEMMX){ 02592 c->idct_put= ff_simple_idct_put_mmx; 02593 c->idct_add= ff_simple_idct_add_mmx; 02594 c->idct = ff_simple_idct_mmx; 02595 c->idct_permutation_type= FF_SIMPLE_IDCT_PERM; 02596 #if CONFIG_GPL 02597 }else if(idct_algo==FF_IDCT_LIBMPEG2MMX){ 02598 if(mm_flags & FF_MM_MMXEXT){ 02599 c->idct_put= ff_libmpeg2mmx2_idct_put; 02600 c->idct_add= ff_libmpeg2mmx2_idct_add; 02601 c->idct = ff_mmxext_idct; 02602 }else{ 02603 c->idct_put= ff_libmpeg2mmx_idct_put; 02604 c->idct_add= ff_libmpeg2mmx_idct_add; 02605 c->idct = ff_mmx_idct; 02606 } 02607 c->idct_permutation_type= FF_LIBMPEG2_IDCT_PERM; 02608 #endif 02609 }else if((CONFIG_VP3_DECODER || CONFIG_VP5_DECODER || CONFIG_VP6_DECODER || CONFIG_THEORA_DECODER) && 02610 idct_algo==FF_IDCT_VP3){ 02611 if(mm_flags & FF_MM_SSE2){ 02612 c->idct_put= ff_vp3_idct_put_sse2; 02613 c->idct_add= ff_vp3_idct_add_sse2; 02614 c->idct = ff_vp3_idct_sse2; 02615 c->idct_permutation_type= FF_TRANSPOSE_IDCT_PERM; 02616 }else{ 02617 c->idct_put= ff_vp3_idct_put_mmx; 02618 c->idct_add= ff_vp3_idct_add_mmx; 02619 c->idct = ff_vp3_idct_mmx; 02620 c->idct_permutation_type= FF_PARTTRANS_IDCT_PERM; 02621 } 02622 }else if(idct_algo==FF_IDCT_CAVS){ 02623 c->idct_permutation_type= FF_TRANSPOSE_IDCT_PERM; 02624 }else if(idct_algo==FF_IDCT_XVIDMMX){ 02625 if(mm_flags & FF_MM_SSE2){ 02626 c->idct_put= ff_idct_xvid_sse2_put; 02627 c->idct_add= ff_idct_xvid_sse2_add; 02628 c->idct = ff_idct_xvid_sse2; 02629 c->idct_permutation_type= FF_SSE2_IDCT_PERM; 02630 }else if(mm_flags & FF_MM_MMXEXT){ 02631 c->idct_put= ff_idct_xvid_mmx2_put; 02632 c->idct_add= ff_idct_xvid_mmx2_add; 02633 c->idct = ff_idct_xvid_mmx2; 02634 }else{ 02635 c->idct_put= ff_idct_xvid_mmx_put; 02636 c->idct_add= ff_idct_xvid_mmx_add; 02637 c->idct = ff_idct_xvid_mmx; 02638 } 02639 } 02640 } 02641 02642 c->put_pixels_clamped = put_pixels_clamped_mmx; 02643 c->put_signed_pixels_clamped = put_signed_pixels_clamped_mmx; 02644 c->add_pixels_clamped = add_pixels_clamped_mmx; 02645 c->clear_block = clear_block_mmx; 02646 c->clear_blocks = clear_blocks_mmx; 02647 if (mm_flags & FF_MM_SSE) 02648 c->clear_block = clear_block_sse; 02649 02650 #define SET_HPEL_FUNCS(PFX, IDX, SIZE, CPU) \ 02651 c->PFX ## _pixels_tab[IDX][0] = PFX ## _pixels ## SIZE ## _ ## CPU; \ 02652 c->PFX ## _pixels_tab[IDX][1] = PFX ## _pixels ## SIZE ## _x2_ ## CPU; \ 02653 c->PFX ## _pixels_tab[IDX][2] = PFX ## _pixels ## SIZE ## _y2_ ## CPU; \ 02654 c->PFX ## _pixels_tab[IDX][3] = PFX ## _pixels ## SIZE ## _xy2_ ## CPU 02655 02656 SET_HPEL_FUNCS(put, 0, 16, mmx); 02657 SET_HPEL_FUNCS(put_no_rnd, 0, 16, mmx); 02658 SET_HPEL_FUNCS(avg, 0, 16, mmx); 02659 SET_HPEL_FUNCS(avg_no_rnd, 0, 16, mmx); 02660 SET_HPEL_FUNCS(put, 1, 8, mmx); 02661 SET_HPEL_FUNCS(put_no_rnd, 1, 8, mmx); 02662 SET_HPEL_FUNCS(avg, 1, 8, mmx); 02663 SET_HPEL_FUNCS(avg_no_rnd, 1, 8, mmx); 02664 02665 c->gmc= gmc_mmx; 02666 02667 c->add_bytes= add_bytes_mmx; 02668 c->add_bytes_l2= add_bytes_l2_mmx; 02669 02670 c->draw_edges = draw_edges_mmx; 02671 02672 if (CONFIG_ANY_H263) { 02673 c->h263_v_loop_filter= h263_v_loop_filter_mmx; 02674 c->h263_h_loop_filter= h263_h_loop_filter_mmx; 02675 } 02676 c->put_h264_chroma_pixels_tab[0]= put_h264_chroma_mc8_mmx_rnd; 02677 c->put_h264_chroma_pixels_tab[1]= put_h264_chroma_mc4_mmx; 02678 c->put_no_rnd_h264_chroma_pixels_tab[0]= put_h264_chroma_mc8_mmx_nornd; 02679 02680 c->put_rv40_chroma_pixels_tab[0]= put_rv40_chroma_mc8_mmx; 02681 c->put_rv40_chroma_pixels_tab[1]= put_rv40_chroma_mc4_mmx; 02682 02683 c->h264_idct_dc_add= 02684 c->h264_idct_add= ff_h264_idct_add_mmx; 02685 c->h264_idct8_dc_add= 02686 c->h264_idct8_add= ff_h264_idct8_add_mmx; 02687 02688 c->h264_idct_add16 = ff_h264_idct_add16_mmx; 02689 c->h264_idct8_add4 = ff_h264_idct8_add4_mmx; 02690 c->h264_idct_add8 = ff_h264_idct_add8_mmx; 02691 c->h264_idct_add16intra= ff_h264_idct_add16intra_mmx; 02692 02693 if (CONFIG_VP6_DECODER) { 02694 c->vp6_filter_diag4 = ff_vp6_filter_diag4_mmx; 02695 } 02696 02697 if (mm_flags & FF_MM_MMXEXT) { 02698 c->prefetch = prefetch_mmx2; 02699 02700 c->put_pixels_tab[0][1] = put_pixels16_x2_mmx2; 02701 c->put_pixels_tab[0][2] = put_pixels16_y2_mmx2; 02702 02703 c->avg_pixels_tab[0][0] = avg_pixels16_mmx2; 02704 c->avg_pixels_tab[0][1] = avg_pixels16_x2_mmx2; 02705 c->avg_pixels_tab[0][2] = avg_pixels16_y2_mmx2; 02706 02707 c->put_pixels_tab[1][1] = put_pixels8_x2_mmx2; 02708 c->put_pixels_tab[1][2] = put_pixels8_y2_mmx2; 02709 02710 c->avg_pixels_tab[1][0] = avg_pixels8_mmx2; 02711 c->avg_pixels_tab[1][1] = avg_pixels8_x2_mmx2; 02712 c->avg_pixels_tab[1][2] = avg_pixels8_y2_mmx2; 02713 02714 c->h264_idct_dc_add= ff_h264_idct_dc_add_mmx2; 02715 c->h264_idct8_dc_add= ff_h264_idct8_dc_add_mmx2; 02716 c->h264_idct_add16 = ff_h264_idct_add16_mmx2; 02717 c->h264_idct8_add4 = ff_h264_idct8_add4_mmx2; 02718 c->h264_idct_add8 = ff_h264_idct_add8_mmx2; 02719 c->h264_idct_add16intra= ff_h264_idct_add16intra_mmx2; 02720 02721 if(!(avctx->flags & CODEC_FLAG_BITEXACT)){ 02722 c->put_no_rnd_pixels_tab[0][1] = put_no_rnd_pixels16_x2_mmx2; 02723 c->put_no_rnd_pixels_tab[0][2] = put_no_rnd_pixels16_y2_mmx2; 02724 c->put_no_rnd_pixels_tab[1][1] = put_no_rnd_pixels8_x2_mmx2; 02725 c->put_no_rnd_pixels_tab[1][2] = put_no_rnd_pixels8_y2_mmx2; 02726 c->avg_pixels_tab[0][3] = avg_pixels16_xy2_mmx2; 02727 c->avg_pixels_tab[1][3] = avg_pixels8_xy2_mmx2; 02728 02729 if (CONFIG_VP3_DECODER || CONFIG_THEORA_DECODER) { 02730 c->vp3_v_loop_filter= ff_vp3_v_loop_filter_mmx2; 02731 c->vp3_h_loop_filter= ff_vp3_h_loop_filter_mmx2; 02732 } 02733 } 02734 02735 #define SET_QPEL_FUNCS(PFX, IDX, SIZE, CPU) \ 02736 c->PFX ## _pixels_tab[IDX][ 0] = PFX ## SIZE ## _mc00_ ## CPU; \ 02737 c->PFX ## _pixels_tab[IDX][ 1] = PFX ## SIZE ## _mc10_ ## CPU; \ 02738 c->PFX ## _pixels_tab[IDX][ 2] = PFX ## SIZE ## _mc20_ ## CPU; \ 02739 c->PFX ## _pixels_tab[IDX][ 3] = PFX ## SIZE ## _mc30_ ## CPU; \ 02740 c->PFX ## _pixels_tab[IDX][ 4] = PFX ## SIZE ## _mc01_ ## CPU; \ 02741 c->PFX ## _pixels_tab[IDX][ 5] = PFX ## SIZE ## _mc11_ ## CPU; \ 02742 c->PFX ## _pixels_tab[IDX][ 6] = PFX ## SIZE ## _mc21_ ## CPU; \ 02743 c->PFX ## _pixels_tab[IDX][ 7] = PFX ## SIZE ## _mc31_ ## CPU; \ 02744 c->PFX ## _pixels_tab[IDX][ 8] = PFX ## SIZE ## _mc02_ ## CPU; \ 02745 c->PFX ## _pixels_tab[IDX][ 9] = PFX ## SIZE ## _mc12_ ## CPU; \ 02746 c->PFX ## _pixels_tab[IDX][10] = PFX ## SIZE ## _mc22_ ## CPU; \ 02747 c->PFX ## _pixels_tab[IDX][11] = PFX ## SIZE ## _mc32_ ## CPU; \ 02748 c->PFX ## _pixels_tab[IDX][12] = PFX ## SIZE ## _mc03_ ## CPU; \ 02749 c->PFX ## _pixels_tab[IDX][13] = PFX ## SIZE ## _mc13_ ## CPU; \ 02750 c->PFX ## _pixels_tab[IDX][14] = PFX ## SIZE ## _mc23_ ## CPU; \ 02751 c->PFX ## _pixels_tab[IDX][15] = PFX ## SIZE ## _mc33_ ## CPU 02752 02753 SET_QPEL_FUNCS(put_qpel, 0, 16, mmx2); 02754 SET_QPEL_FUNCS(put_qpel, 1, 8, mmx2); 02755 SET_QPEL_FUNCS(put_no_rnd_qpel, 0, 16, mmx2); 02756 SET_QPEL_FUNCS(put_no_rnd_qpel, 1, 8, mmx2); 02757 SET_QPEL_FUNCS(avg_qpel, 0, 16, mmx2); 02758 SET_QPEL_FUNCS(avg_qpel, 1, 8, mmx2); 02759 02760 SET_QPEL_FUNCS(put_h264_qpel, 0, 16, mmx2); 02761 SET_QPEL_FUNCS(put_h264_qpel, 1, 8, mmx2); 02762 SET_QPEL_FUNCS(put_h264_qpel, 2, 4, mmx2); 02763 SET_QPEL_FUNCS(avg_h264_qpel, 0, 16, mmx2); 02764 SET_QPEL_FUNCS(avg_h264_qpel, 1, 8, mmx2); 02765 SET_QPEL_FUNCS(avg_h264_qpel, 2, 4, mmx2); 02766 02767 SET_QPEL_FUNCS(put_2tap_qpel, 0, 16, mmx2); 02768 SET_QPEL_FUNCS(put_2tap_qpel, 1, 8, mmx2); 02769 SET_QPEL_FUNCS(avg_2tap_qpel, 0, 16, mmx2); 02770 SET_QPEL_FUNCS(avg_2tap_qpel, 1, 8, mmx2); 02771 02772 c->avg_rv40_chroma_pixels_tab[0]= avg_rv40_chroma_mc8_mmx2; 02773 c->avg_rv40_chroma_pixels_tab[1]= avg_rv40_chroma_mc4_mmx2; 02774 02775 c->avg_h264_chroma_pixels_tab[0]= avg_h264_chroma_mc8_mmx2_rnd; 02776 c->avg_h264_chroma_pixels_tab[1]= avg_h264_chroma_mc4_mmx2; 02777 c->avg_h264_chroma_pixels_tab[2]= avg_h264_chroma_mc2_mmx2; 02778 c->put_h264_chroma_pixels_tab[2]= put_h264_chroma_mc2_mmx2; 02779 c->h264_v_loop_filter_luma= h264_v_loop_filter_luma_mmx2; 02780 c->h264_h_loop_filter_luma= h264_h_loop_filter_luma_mmx2; 02781 c->h264_v_loop_filter_chroma= h264_v_loop_filter_chroma_mmx2; 02782 c->h264_h_loop_filter_chroma= h264_h_loop_filter_chroma_mmx2; 02783 c->h264_v_loop_filter_chroma_intra= h264_v_loop_filter_chroma_intra_mmx2; 02784 c->h264_h_loop_filter_chroma_intra= h264_h_loop_filter_chroma_intra_mmx2; 02785 c->h264_loop_filter_strength= h264_loop_filter_strength_mmx2; 02786 02787 c->weight_h264_pixels_tab[0]= ff_h264_weight_16x16_mmx2; 02788 c->weight_h264_pixels_tab[1]= ff_h264_weight_16x8_mmx2; 02789 c->weight_h264_pixels_tab[2]= ff_h264_weight_8x16_mmx2; 02790 c->weight_h264_pixels_tab[3]= ff_h264_weight_8x8_mmx2; 02791 c->weight_h264_pixels_tab[4]= ff_h264_weight_8x4_mmx2; 02792 c->weight_h264_pixels_tab[5]= ff_h264_weight_4x8_mmx2; 02793 c->weight_h264_pixels_tab[6]= ff_h264_weight_4x4_mmx2; 02794 c->weight_h264_pixels_tab[7]= ff_h264_weight_4x2_mmx2; 02795 02796 c->biweight_h264_pixels_tab[0]= ff_h264_biweight_16x16_mmx2; 02797 c->biweight_h264_pixels_tab[1]= ff_h264_biweight_16x8_mmx2; 02798 c->biweight_h264_pixels_tab[2]= ff_h264_biweight_8x16_mmx2; 02799 c->biweight_h264_pixels_tab[3]= ff_h264_biweight_8x8_mmx2; 02800 c->biweight_h264_pixels_tab[4]= ff_h264_biweight_8x4_mmx2; 02801 c->biweight_h264_pixels_tab[5]= ff_h264_biweight_4x8_mmx2; 02802 c->biweight_h264_pixels_tab[6]= ff_h264_biweight_4x4_mmx2; 02803 c->biweight_h264_pixels_tab[7]= ff_h264_biweight_4x2_mmx2; 02804 02805 #if HAVE_YASM 02806 c->add_hfyu_median_prediction = ff_add_hfyu_median_prediction_mmx2; 02807 #endif 02808 #if HAVE_7REGS && HAVE_TEN_OPERANDS 02809 if( mm_flags&FF_MM_3DNOW ) 02810 c->add_hfyu_median_prediction = add_hfyu_median_prediction_cmov; 02811 #endif 02812 02813 if (CONFIG_CAVS_DECODER) 02814 ff_cavsdsp_init_mmx2(c, avctx); 02815 02816 if (CONFIG_VC1_DECODER || CONFIG_WMV3_DECODER) 02817 ff_vc1dsp_init_mmx(c, avctx); 02818 02819 c->add_png_paeth_prediction= add_png_paeth_prediction_mmx2; 02820 } else if (mm_flags & FF_MM_3DNOW) { 02821 c->prefetch = prefetch_3dnow; 02822 02823 c->put_pixels_tab[0][1] = put_pixels16_x2_3dnow; 02824 c->put_pixels_tab[0][2] = put_pixels16_y2_3dnow; 02825 02826 c->avg_pixels_tab[0][0] = avg_pixels16_3dnow; 02827 c->avg_pixels_tab[0][1] = avg_pixels16_x2_3dnow; 02828 c->avg_pixels_tab[0][2] = avg_pixels16_y2_3dnow; 02829 02830 c->put_pixels_tab[1][1] = put_pixels8_x2_3dnow; 02831 c->put_pixels_tab[1][2] = put_pixels8_y2_3dnow; 02832 02833 c->avg_pixels_tab[1][0] = avg_pixels8_3dnow; 02834 c->avg_pixels_tab[1][1] = avg_pixels8_x2_3dnow; 02835 c->avg_pixels_tab[1][2] = avg_pixels8_y2_3dnow; 02836 02837 if(!(avctx->flags & CODEC_FLAG_BITEXACT)){ 02838 c->put_no_rnd_pixels_tab[0][1] = put_no_rnd_pixels16_x2_3dnow; 02839 c->put_no_rnd_pixels_tab[0][2] = put_no_rnd_pixels16_y2_3dnow; 02840 c->put_no_rnd_pixels_tab[1][1] = put_no_rnd_pixels8_x2_3dnow; 02841 c->put_no_rnd_pixels_tab[1][2] = put_no_rnd_pixels8_y2_3dnow; 02842 c->avg_pixels_tab[0][3] = avg_pixels16_xy2_3dnow; 02843 c->avg_pixels_tab[1][3] = avg_pixels8_xy2_3dnow; 02844 } 02845 02846 SET_QPEL_FUNCS(put_qpel, 0, 16, 3dnow); 02847 SET_QPEL_FUNCS(put_qpel, 1, 8, 3dnow); 02848 SET_QPEL_FUNCS(put_no_rnd_qpel, 0, 16, 3dnow); 02849 SET_QPEL_FUNCS(put_no_rnd_qpel, 1, 8, 3dnow); 02850 SET_QPEL_FUNCS(avg_qpel, 0, 16, 3dnow); 02851 SET_QPEL_FUNCS(avg_qpel, 1, 8, 3dnow); 02852 02853 SET_QPEL_FUNCS(put_h264_qpel, 0, 16, 3dnow); 02854 SET_QPEL_FUNCS(put_h264_qpel, 1, 8, 3dnow); 02855 SET_QPEL_FUNCS(put_h264_qpel, 2, 4, 3dnow); 02856 SET_QPEL_FUNCS(avg_h264_qpel, 0, 16, 3dnow); 02857 SET_QPEL_FUNCS(avg_h264_qpel, 1, 8, 3dnow); 02858 SET_QPEL_FUNCS(avg_h264_qpel, 2, 4, 3dnow); 02859 02860 SET_QPEL_FUNCS(put_2tap_qpel, 0, 16, 3dnow); 02861 SET_QPEL_FUNCS(put_2tap_qpel, 1, 8, 3dnow); 02862 SET_QPEL_FUNCS(avg_2tap_qpel, 0, 16, 3dnow); 02863 SET_QPEL_FUNCS(avg_2tap_qpel, 1, 8, 3dnow); 02864 02865 c->avg_h264_chroma_pixels_tab[0]= avg_h264_chroma_mc8_3dnow_rnd; 02866 c->avg_h264_chroma_pixels_tab[1]= avg_h264_chroma_mc4_3dnow; 02867 02868 c->avg_rv40_chroma_pixels_tab[0]= avg_rv40_chroma_mc8_3dnow; 02869 c->avg_rv40_chroma_pixels_tab[1]= avg_rv40_chroma_mc4_3dnow; 02870 02871 if (CONFIG_CAVS_DECODER) 02872 ff_cavsdsp_init_3dnow(c, avctx); 02873 } 02874 02875 02876 #define H264_QPEL_FUNCS(x, y, CPU)\ 02877 c->put_h264_qpel_pixels_tab[0][x+y*4] = put_h264_qpel16_mc##x##y##_##CPU;\ 02878 c->put_h264_qpel_pixels_tab[1][x+y*4] = put_h264_qpel8_mc##x##y##_##CPU;\ 02879 c->avg_h264_qpel_pixels_tab[0][x+y*4] = avg_h264_qpel16_mc##x##y##_##CPU;\ 02880 c->avg_h264_qpel_pixels_tab[1][x+y*4] = avg_h264_qpel8_mc##x##y##_##CPU; 02881 if((mm_flags & FF_MM_SSE2) && !(mm_flags & FF_MM_3DNOW)){ 02882 // these functions are slower than mmx on AMD, but faster on Intel 02883 /* FIXME works in most codecs, but crashes svq1 due to unaligned chroma 02884 c->put_pixels_tab[0][0] = put_pixels16_sse2; 02885 c->avg_pixels_tab[0][0] = avg_pixels16_sse2; 02886 */ 02887 H264_QPEL_FUNCS(0, 0, sse2); 02888 } 02889 if(mm_flags & FF_MM_SSE2){ 02890 c->h264_idct8_add = ff_h264_idct8_add_sse2; 02891 c->h264_idct8_add4= ff_h264_idct8_add4_sse2; 02892 02893 H264_QPEL_FUNCS(0, 1, sse2); 02894 H264_QPEL_FUNCS(0, 2, sse2); 02895 H264_QPEL_FUNCS(0, 3, sse2); 02896 H264_QPEL_FUNCS(1, 1, sse2); 02897 H264_QPEL_FUNCS(1, 2, sse2); 02898 H264_QPEL_FUNCS(1, 3, sse2); 02899 H264_QPEL_FUNCS(2, 1, sse2); 02900 H264_QPEL_FUNCS(2, 2, sse2); 02901 H264_QPEL_FUNCS(2, 3, sse2); 02902 H264_QPEL_FUNCS(3, 1, sse2); 02903 H264_QPEL_FUNCS(3, 2, sse2); 02904 H264_QPEL_FUNCS(3, 3, sse2); 02905 02906 if (CONFIG_VP6_DECODER) { 02907 c->vp6_filter_diag4 = ff_vp6_filter_diag4_sse2; 02908 } 02909 } 02910 #if HAVE_SSSE3 02911 if(mm_flags & FF_MM_SSSE3){ 02912 H264_QPEL_FUNCS(1, 0, ssse3); 02913 H264_QPEL_FUNCS(1, 1, ssse3); 02914 H264_QPEL_FUNCS(1, 2, ssse3); 02915 H264_QPEL_FUNCS(1, 3, ssse3); 02916 H264_QPEL_FUNCS(2, 0, ssse3); 02917 H264_QPEL_FUNCS(2, 1, ssse3); 02918 H264_QPEL_FUNCS(2, 2, ssse3); 02919 H264_QPEL_FUNCS(2, 3, ssse3); 02920 H264_QPEL_FUNCS(3, 0, ssse3); 02921 H264_QPEL_FUNCS(3, 1, ssse3); 02922 H264_QPEL_FUNCS(3, 2, ssse3); 02923 H264_QPEL_FUNCS(3, 3, ssse3); 02924 c->put_no_rnd_h264_chroma_pixels_tab[0]= put_h264_chroma_mc8_ssse3_nornd; 02925 c->put_h264_chroma_pixels_tab[0]= put_h264_chroma_mc8_ssse3_rnd; 02926 c->avg_h264_chroma_pixels_tab[0]= avg_h264_chroma_mc8_ssse3_rnd; 02927 c->put_h264_chroma_pixels_tab[1]= put_h264_chroma_mc4_ssse3; 02928 c->avg_h264_chroma_pixels_tab[1]= avg_h264_chroma_mc4_ssse3; 02929 c->add_png_paeth_prediction= add_png_paeth_prediction_ssse3; 02930 } 02931 #endif 02932 02933 #if CONFIG_GPL && HAVE_YASM 02934 if( mm_flags&FF_MM_MMXEXT ){ 02935 #if ARCH_X86_32 02936 c->h264_v_loop_filter_luma_intra = ff_x264_deblock_v_luma_intra_mmxext; 02937 c->h264_h_loop_filter_luma_intra = ff_x264_deblock_h_luma_intra_mmxext; 02938 #endif 02939 if( mm_flags&FF_MM_SSE2 ){ 02940 #if ARCH_X86_64 || !defined(__ICC) || __ICC > 1100 02941 c->h264_v_loop_filter_luma = ff_x264_deblock_v_luma_sse2; 02942 c->h264_h_loop_filter_luma = ff_x264_deblock_h_luma_sse2; 02943 c->h264_v_loop_filter_luma_intra = ff_x264_deblock_v_luma_intra_sse2; 02944 c->h264_h_loop_filter_luma_intra = ff_x264_deblock_h_luma_intra_sse2; 02945 #endif 02946 c->h264_idct_add16 = ff_h264_idct_add16_sse2; 02947 c->h264_idct_add8 = ff_h264_idct_add8_sse2; 02948 c->h264_idct_add16intra = ff_h264_idct_add16intra_sse2; 02949 } 02950 } 02951 #endif 02952 02953 #if CONFIG_SNOW_DECODER 02954 if(mm_flags & FF_MM_SSE2 & 0){ 02955 c->horizontal_compose97i = ff_snow_horizontal_compose97i_sse2; 02956 #if HAVE_7REGS 02957 c->vertical_compose97i = ff_snow_vertical_compose97i_sse2; 02958 #endif 02959 c->inner_add_yblock = ff_snow_inner_add_yblock_sse2; 02960 } 02961 else{ 02962 if(mm_flags & FF_MM_MMXEXT){ 02963 c->horizontal_compose97i = ff_snow_horizontal_compose97i_mmx; 02964 #if HAVE_7REGS 02965 c->vertical_compose97i = ff_snow_vertical_compose97i_mmx; 02966 #endif 02967 } 02968 c->inner_add_yblock = ff_snow_inner_add_yblock_mmx; 02969 } 02970 #endif 02971 02972 if(mm_flags & FF_MM_3DNOW){ 02973 c->vorbis_inverse_coupling = vorbis_inverse_coupling_3dnow; 02974 c->vector_fmul = vector_fmul_3dnow; 02975 if(!(avctx->flags & CODEC_FLAG_BITEXACT)){ 02976 c->float_to_int16 = float_to_int16_3dnow; 02977 c->float_to_int16_interleave = float_to_int16_interleave_3dnow; 02978 } 02979 } 02980 if(mm_flags & FF_MM_3DNOWEXT){ 02981 c->vector_fmul_reverse = vector_fmul_reverse_3dnow2; 02982 c->vector_fmul_window = vector_fmul_window_3dnow2; 02983 if(!(avctx->flags & CODEC_FLAG_BITEXACT)){ 02984 c->float_to_int16_interleave = float_to_int16_interleave_3dn2; 02985 } 02986 } 02987 if(mm_flags & FF_MM_SSE){ 02988 c->vorbis_inverse_coupling = vorbis_inverse_coupling_sse; 02989 c->ac3_downmix = ac3_downmix_sse; 02990 c->vector_fmul = vector_fmul_sse; 02991 c->vector_fmul_reverse = vector_fmul_reverse_sse; 02992 c->vector_fmul_add_add = vector_fmul_add_add_sse; 02993 c->vector_fmul_window = vector_fmul_window_sse; 02994 c->int32_to_float_fmul_scalar = int32_to_float_fmul_scalar_sse; 02995 c->float_to_int16 = float_to_int16_sse; 02996 c->float_to_int16_interleave = float_to_int16_interleave_sse; 02997 } 02998 if(mm_flags & FF_MM_3DNOW) 02999 c->vector_fmul_add_add = vector_fmul_add_add_3dnow; // faster than sse 03000 if(mm_flags & FF_MM_SSE2){ 03001 c->int32_to_float_fmul_scalar = int32_to_float_fmul_scalar_sse2; 03002 c->float_to_int16 = float_to_int16_sse2; 03003 c->float_to_int16_interleave = float_to_int16_interleave_sse2; 03004 c->add_int16 = add_int16_sse2; 03005 c->sub_int16 = sub_int16_sse2; 03006 c->scalarproduct_int16 = scalarproduct_int16_sse2; 03007 } 03008 } 03009 03010 if (CONFIG_ENCODERS) 03011 dsputilenc_init_mmx(c, avctx); 03012 03013 #if 0 03014 // for speed testing 03015 get_pixels = just_return; 03016 put_pixels_clamped = just_return; 03017 add_pixels_clamped = just_return; 03018 03019 pix_abs16x16 = just_return; 03020 pix_abs16x16_x2 = just_return; 03021 pix_abs16x16_y2 = just_return; 03022 pix_abs16x16_xy2 = just_return; 03023 03024 put_pixels_tab[0] = just_return; 03025 put_pixels_tab[1] = just_return; 03026 put_pixels_tab[2] = just_return; 03027 put_pixels_tab[3] = just_return; 03028 03029 put_no_rnd_pixels_tab[0] = just_return; 03030 put_no_rnd_pixels_tab[1] = just_return; 03031 put_no_rnd_pixels_tab[2] = just_return; 03032 put_no_rnd_pixels_tab[3] = just_return; 03033 03034 avg_pixels_tab[0] = just_return; 03035 avg_pixels_tab[1] = just_return; 03036 avg_pixels_tab[2] = just_return; 03037 avg_pixels_tab[3] = just_return; 03038 03039 avg_no_rnd_pixels_tab[0] = just_return; 03040 avg_no_rnd_pixels_tab[1] = just_return; 03041 avg_no_rnd_pixels_tab[2] = just_return; 03042 avg_no_rnd_pixels_tab[3] = just_return; 03043 03044 //av_fdct = just_return; 03045 //ff_idct = just_return; 03046 #endif 03047 }