1/* 2 * Chinese AVS video (AVS1-P2, JiZhun profile) decoder. 3 * Copyright (c) 2006 Stefan Gehrer <stefan.gehrer@gmx.de> 4 * 5 * MMX-optimized DSP functions, based on H.264 optimizations by 6 * Michael Niedermayer and Loren Merritt 7 * 8 * This file is part of FFmpeg. 9 * 10 * FFmpeg is free software; you can redistribute it and/or 11 * modify it under the terms of the GNU Lesser General Public 12 * License as published by the Free Software Foundation; either 13 * version 2.1 of the License, or (at your option) any later version. 14 * 15 * FFmpeg is distributed in the hope that it will be useful, 16 * but WITHOUT ANY WARRANTY; without even the implied warranty of 17 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU 18 * Lesser General Public License for more details. 19 * 20 * You should have received a copy of the GNU Lesser General Public 21 * License along with FFmpeg; if not, write to the Free Software 22 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA 23 */ 24 25#include "libavutil/attributes.h" 26#include "libavutil/common.h" 27#include "libavutil/cpu.h" 28#include "libavutil/x86/asm.h" 29#include "libavutil/x86/cpu.h" 30#include "libavcodec/cavsdsp.h" 31#include "libavcodec/idctdsp.h" 32#include "constants.h" 33#include "fpel.h" 34#include "idctdsp.h" 35#include "config.h" 36 37#if HAVE_MMX_INLINE 38 39/* in/out: mma=mma+mmb, mmb=mmb-mma */ 40#define SUMSUB_BA( a, b ) \ 41 "paddw "#b", "#a" \n\t"\ 42 "paddw "#b", "#b" \n\t"\ 43 "psubw "#a", "#b" \n\t" 44 45/***************************************************************************** 46 * 47 * inverse transform 48 * 49 ****************************************************************************/ 50 51static inline void cavs_idct8_1d(int16_t *block, uint64_t bias) 52{ 53 __asm__ volatile( 54 "movq 112(%0), %%mm4 \n\t" /* mm4 = src7 */ 55 "movq 16(%0), %%mm5 \n\t" /* mm5 = src1 */ 56 "movq 80(%0), %%mm2 \n\t" /* mm2 = src5 */ 57 "movq 48(%0), %%mm7 \n\t" /* mm7 = src3 */ 58 "movq %%mm4, %%mm0 \n\t" 59 "movq %%mm5, %%mm3 \n\t" 60 "movq %%mm2, %%mm6 \n\t" 61 "movq %%mm7, %%mm1 \n\t" 62 63 "paddw %%mm4, %%mm4 \n\t" /* mm4 = 2*src7 */ 64 "paddw %%mm3, %%mm3 \n\t" /* mm3 = 2*src1 */ 65 "paddw %%mm6, %%mm6 \n\t" /* mm6 = 2*src5 */ 66 "paddw %%mm1, %%mm1 \n\t" /* mm1 = 2*src3 */ 67 "paddw %%mm4, %%mm0 \n\t" /* mm0 = 3*src7 */ 68 "paddw %%mm3, %%mm5 \n\t" /* mm5 = 3*src1 */ 69 "paddw %%mm6, %%mm2 \n\t" /* mm2 = 3*src5 */ 70 "paddw %%mm1, %%mm7 \n\t" /* mm7 = 3*src3 */ 71 "psubw %%mm4, %%mm5 \n\t" /* mm5 = 3*src1 - 2*src7 = a0 */ 72 "paddw %%mm6, %%mm7 \n\t" /* mm7 = 3*src3 + 2*src5 = a1 */ 73 "psubw %%mm2, %%mm1 \n\t" /* mm1 = 2*src3 - 3*src5 = a2 */ 74 "paddw %%mm0, %%mm3 \n\t" /* mm3 = 2*src1 + 3*src7 = a3 */ 75 76 "movq %%mm5, %%mm4 \n\t" 77 "movq %%mm7, %%mm6 \n\t" 78 "movq %%mm3, %%mm0 \n\t" 79 "movq %%mm1, %%mm2 \n\t" 80 SUMSUB_BA( %%mm7, %%mm5 ) /* mm7 = a0 + a1 mm5 = a0 - a1 */ 81 "paddw %%mm3, %%mm7 \n\t" /* mm7 = a0 + a1 + a3 */ 82 "paddw %%mm1, %%mm5 \n\t" /* mm5 = a0 - a1 + a2 */ 83 "paddw %%mm7, %%mm7 \n\t" 84 "paddw %%mm5, %%mm5 \n\t" 85 "paddw %%mm6, %%mm7 \n\t" /* mm7 = b4 */ 86 "paddw %%mm4, %%mm5 \n\t" /* mm5 = b5 */ 87 88 SUMSUB_BA( %%mm1, %%mm3 ) /* mm1 = a3 + a2 mm3 = a3 - a2 */ 89 "psubw %%mm1, %%mm4 \n\t" /* mm4 = a0 - a2 - a3 */ 90 "movq %%mm4, %%mm1 \n\t" /* mm1 = a0 - a2 - a3 */ 91 "psubw %%mm6, %%mm3 \n\t" /* mm3 = a3 - a2 - a1 */ 92 "paddw %%mm1, %%mm1 \n\t" 93 "paddw %%mm3, %%mm3 \n\t" 94 "psubw %%mm2, %%mm1 \n\t" /* mm1 = b7 */ 95 "paddw %%mm0, %%mm3 \n\t" /* mm3 = b6 */ 96 97 "movq 32(%0), %%mm2 \n\t" /* mm2 = src2 */ 98 "movq 96(%0), %%mm6 \n\t" /* mm6 = src6 */ 99 "movq %%mm2, %%mm4 \n\t" 100 "movq %%mm6, %%mm0 \n\t" 101 "psllw $2, %%mm4 \n\t" /* mm4 = 4*src2 */ 102 "psllw $2, %%mm6 \n\t" /* mm6 = 4*src6 */ 103 "paddw %%mm4, %%mm2 \n\t" /* mm2 = 5*src2 */ 104 "paddw %%mm6, %%mm0 \n\t" /* mm0 = 5*src6 */ 105 "paddw %%mm2, %%mm2 \n\t" 106 "paddw %%mm0, %%mm0 \n\t" 107 "psubw %%mm0, %%mm4 \n\t" /* mm4 = 4*src2 - 10*src6 = a7 */ 108 "paddw %%mm2, %%mm6 \n\t" /* mm6 = 4*src6 + 10*src2 = a6 */ 109 110 "movq (%0), %%mm2 \n\t" /* mm2 = src0 */ 111 "movq 64(%0), %%mm0 \n\t" /* mm0 = src4 */ 112 SUMSUB_BA( %%mm0, %%mm2 ) /* mm0 = src0+src4 mm2 = src0-src4 */ 113 "psllw $3, %%mm0 \n\t" 114 "psllw $3, %%mm2 \n\t" 115 "paddw %1, %%mm0 \n\t" /* add rounding bias */ 116 "paddw %1, %%mm2 \n\t" /* add rounding bias */ 117 118 SUMSUB_BA( %%mm6, %%mm0 ) /* mm6 = a4 + a6 mm0 = a4 - a6 */ 119 SUMSUB_BA( %%mm4, %%mm2 ) /* mm4 = a5 + a7 mm2 = a5 - a7 */ 120 SUMSUB_BA( %%mm7, %%mm6 ) /* mm7 = dst0 mm6 = dst7 */ 121 SUMSUB_BA( %%mm5, %%mm4 ) /* mm5 = dst1 mm4 = dst6 */ 122 SUMSUB_BA( %%mm3, %%mm2 ) /* mm3 = dst2 mm2 = dst5 */ 123 SUMSUB_BA( %%mm1, %%mm0 ) /* mm1 = dst3 mm0 = dst4 */ 124 :: "r"(block), "m"(bias) 125 ); 126} 127 128#define SBUTTERFLY(a,b,t,n,m)\ 129 "mov" #m " " #a ", " #t " \n\t" /* abcd */\ 130 "punpckl" #n " " #b ", " #a " \n\t" /* aebf */\ 131 "punpckh" #n " " #b ", " #t " \n\t" /* cgdh */\ 132 133#define TRANSPOSE4(a,b,c,d,t)\ 134 SBUTTERFLY(a,b,t,wd,q) /* a=aebf t=cgdh */\ 135 SBUTTERFLY(c,d,b,wd,q) /* c=imjn b=kolp */\ 136 SBUTTERFLY(a,c,d,dq,q) /* a=aeim d=bfjn */\ 137 SBUTTERFLY(t,b,c,dq,q) /* t=cgko c=dhlp */ 138 139static void cavs_idct8_add_mmx(uint8_t *dst, int16_t *block, int stride) 140{ 141 int i; 142 DECLARE_ALIGNED(8, int16_t, b2)[64]; 143 144 for(i=0; i<2; i++){ 145 DECLARE_ALIGNED(8, uint64_t, tmp); 146 147 cavs_idct8_1d(block+4*i, ff_pw_4.a); 148 149 __asm__ volatile( 150 "psraw $3, %%mm7 \n\t" 151 "psraw $3, %%mm6 \n\t" 152 "psraw $3, %%mm5 \n\t" 153 "psraw $3, %%mm4 \n\t" 154 "psraw $3, %%mm3 \n\t" 155 "psraw $3, %%mm2 \n\t" 156 "psraw $3, %%mm1 \n\t" 157 "psraw $3, %%mm0 \n\t" 158 "movq %%mm7, %0 \n\t" 159 TRANSPOSE4( %%mm0, %%mm2, %%mm4, %%mm6, %%mm7 ) 160 "movq %%mm0, 8(%1) \n\t" 161 "movq %%mm6, 24(%1) \n\t" 162 "movq %%mm7, 40(%1) \n\t" 163 "movq %%mm4, 56(%1) \n\t" 164 "movq %0, %%mm7 \n\t" 165 TRANSPOSE4( %%mm7, %%mm5, %%mm3, %%mm1, %%mm0 ) 166 "movq %%mm7, (%1) \n\t" 167 "movq %%mm1, 16(%1) \n\t" 168 "movq %%mm0, 32(%1) \n\t" 169 "movq %%mm3, 48(%1) \n\t" 170 : "=m"(tmp) 171 : "r"(b2+32*i) 172 : "memory" 173 ); 174 } 175 176 for(i=0; i<2; i++){ 177 cavs_idct8_1d(b2+4*i, ff_pw_64.a); 178 179 __asm__ volatile( 180 "psraw $7, %%mm7 \n\t" 181 "psraw $7, %%mm6 \n\t" 182 "psraw $7, %%mm5 \n\t" 183 "psraw $7, %%mm4 \n\t" 184 "psraw $7, %%mm3 \n\t" 185 "psraw $7, %%mm2 \n\t" 186 "psraw $7, %%mm1 \n\t" 187 "psraw $7, %%mm0 \n\t" 188 "movq %%mm7, (%0) \n\t" 189 "movq %%mm5, 16(%0) \n\t" 190 "movq %%mm3, 32(%0) \n\t" 191 "movq %%mm1, 48(%0) \n\t" 192 "movq %%mm0, 64(%0) \n\t" 193 "movq %%mm2, 80(%0) \n\t" 194 "movq %%mm4, 96(%0) \n\t" 195 "movq %%mm6, 112(%0) \n\t" 196 :: "r"(b2+4*i) 197 : "memory" 198 ); 199 } 200 201 ff_add_pixels_clamped_mmx(b2, dst, stride); 202} 203 204#endif /* HAVE_MMX_INLINE */ 205 206#if (HAVE_MMXEXT_INLINE || HAVE_AMD3DNOW_INLINE) 207 208/***************************************************************************** 209 * 210 * motion compensation 211 * 212 ****************************************************************************/ 213 214/* vertical filter [-1 -2 96 42 -7 0] */ 215#define QPEL_CAVSV1(A,B,C,D,E,F,OP,ADD, MUL1, MUL2) \ 216 "movd (%0), "#F" \n\t"\ 217 "movq "#C", %%mm6 \n\t"\ 218 "pmullw "MANGLE(MUL1)", %%mm6\n\t"\ 219 "movq "#D", %%mm7 \n\t"\ 220 "pmullw "MANGLE(MUL2)", %%mm7\n\t"\ 221 "psllw $3, "#E" \n\t"\ 222 "psubw "#E", %%mm6 \n\t"\ 223 "psraw $3, "#E" \n\t"\ 224 "paddw %%mm7, %%mm6 \n\t"\ 225 "paddw "#E", %%mm6 \n\t"\ 226 "paddw "#B", "#B" \n\t"\ 227 "pxor %%mm7, %%mm7 \n\t"\ 228 "add %2, %0 \n\t"\ 229 "punpcklbw %%mm7, "#F" \n\t"\ 230 "psubw "#B", %%mm6 \n\t"\ 231 "psraw $1, "#B" \n\t"\ 232 "psubw "#A", %%mm6 \n\t"\ 233 "paddw "MANGLE(ADD)", %%mm6 \n\t"\ 234 "psraw $7, %%mm6 \n\t"\ 235 "packuswb %%mm6, %%mm6 \n\t"\ 236 OP(%%mm6, (%1), A, d) \ 237 "add %3, %1 \n\t" 238 239/* vertical filter [ 0 -1 5 5 -1 0] */ 240#define QPEL_CAVSV2(A,B,C,D,E,F,OP,ADD, MUL1, MUL2) \ 241 "movd (%0), "#F" \n\t"\ 242 "movq "#C", %%mm6 \n\t"\ 243 "paddw "#D", %%mm6 \n\t"\ 244 "pmullw "MANGLE(MUL1)", %%mm6\n\t"\ 245 "add %2, %0 \n\t"\ 246 "punpcklbw %%mm7, "#F" \n\t"\ 247 "psubw "#B", %%mm6 \n\t"\ 248 "psubw "#E", %%mm6 \n\t"\ 249 "paddw "MANGLE(ADD)", %%mm6 \n\t"\ 250 "psraw $3, %%mm6 \n\t"\ 251 "packuswb %%mm6, %%mm6 \n\t"\ 252 OP(%%mm6, (%1), A, d) \ 253 "add %3, %1 \n\t" 254 255/* vertical filter [ 0 -7 42 96 -2 -1] */ 256#define QPEL_CAVSV3(A,B,C,D,E,F,OP,ADD, MUL1, MUL2) \ 257 "movd (%0), "#F" \n\t"\ 258 "movq "#C", %%mm6 \n\t"\ 259 "pmullw "MANGLE(MUL2)", %%mm6\n\t"\ 260 "movq "#D", %%mm7 \n\t"\ 261 "pmullw "MANGLE(MUL1)", %%mm7\n\t"\ 262 "psllw $3, "#B" \n\t"\ 263 "psubw "#B", %%mm6 \n\t"\ 264 "psraw $3, "#B" \n\t"\ 265 "paddw %%mm7, %%mm6 \n\t"\ 266 "paddw "#B", %%mm6 \n\t"\ 267 "paddw "#E", "#E" \n\t"\ 268 "pxor %%mm7, %%mm7 \n\t"\ 269 "add %2, %0 \n\t"\ 270 "punpcklbw %%mm7, "#F" \n\t"\ 271 "psubw "#E", %%mm6 \n\t"\ 272 "psraw $1, "#E" \n\t"\ 273 "psubw "#F", %%mm6 \n\t"\ 274 "paddw "MANGLE(ADD)", %%mm6 \n\t"\ 275 "psraw $7, %%mm6 \n\t"\ 276 "packuswb %%mm6, %%mm6 \n\t"\ 277 OP(%%mm6, (%1), A, d) \ 278 "add %3, %1 \n\t" 279 280 281#define QPEL_CAVSVNUM(VOP,OP,ADD,MUL1,MUL2)\ 282 int w= 2;\ 283 src -= 2*srcStride;\ 284 \ 285 while(w--){\ 286 __asm__ volatile(\ 287 "pxor %%mm7, %%mm7 \n\t"\ 288 "movd (%0), %%mm0 \n\t"\ 289 "add %2, %0 \n\t"\ 290 "movd (%0), %%mm1 \n\t"\ 291 "add %2, %0 \n\t"\ 292 "movd (%0), %%mm2 \n\t"\ 293 "add %2, %0 \n\t"\ 294 "movd (%0), %%mm3 \n\t"\ 295 "add %2, %0 \n\t"\ 296 "movd (%0), %%mm4 \n\t"\ 297 "add %2, %0 \n\t"\ 298 "punpcklbw %%mm7, %%mm0 \n\t"\ 299 "punpcklbw %%mm7, %%mm1 \n\t"\ 300 "punpcklbw %%mm7, %%mm2 \n\t"\ 301 "punpcklbw %%mm7, %%mm3 \n\t"\ 302 "punpcklbw %%mm7, %%mm4 \n\t"\ 303 VOP(%%mm0, %%mm1, %%mm2, %%mm3, %%mm4, %%mm5, OP, ADD, MUL1, MUL2)\ 304 VOP(%%mm1, %%mm2, %%mm3, %%mm4, %%mm5, %%mm0, OP, ADD, MUL1, MUL2)\ 305 VOP(%%mm2, %%mm3, %%mm4, %%mm5, %%mm0, %%mm1, OP, ADD, MUL1, MUL2)\ 306 VOP(%%mm3, %%mm4, %%mm5, %%mm0, %%mm1, %%mm2, OP, ADD, MUL1, MUL2)\ 307 VOP(%%mm4, %%mm5, %%mm0, %%mm1, %%mm2, %%mm3, OP, ADD, MUL1, MUL2)\ 308 VOP(%%mm5, %%mm0, %%mm1, %%mm2, %%mm3, %%mm4, OP, ADD, MUL1, MUL2)\ 309 VOP(%%mm0, %%mm1, %%mm2, %%mm3, %%mm4, %%mm5, OP, ADD, MUL1, MUL2)\ 310 VOP(%%mm1, %%mm2, %%mm3, %%mm4, %%mm5, %%mm0, OP, ADD, MUL1, MUL2)\ 311 \ 312 : "+a"(src), "+c"(dst)\ 313 : "S"((x86_reg)srcStride), "r"((x86_reg)dstStride)\ 314 NAMED_CONSTRAINTS_ADD(ADD,MUL1,MUL2)\ 315 : "memory"\ 316 );\ 317 if(h==16){\ 318 __asm__ volatile(\ 319 VOP(%%mm2, %%mm3, %%mm4, %%mm5, %%mm0, %%mm1, OP, ADD, MUL1, MUL2)\ 320 VOP(%%mm3, %%mm4, %%mm5, %%mm0, %%mm1, %%mm2, OP, ADD, MUL1, MUL2)\ 321 VOP(%%mm4, %%mm5, %%mm0, %%mm1, %%mm2, %%mm3, OP, ADD, MUL1, MUL2)\ 322 VOP(%%mm5, %%mm0, %%mm1, %%mm2, %%mm3, %%mm4, OP, ADD, MUL1, MUL2)\ 323 VOP(%%mm0, %%mm1, %%mm2, %%mm3, %%mm4, %%mm5, OP, ADD, MUL1, MUL2)\ 324 VOP(%%mm1, %%mm2, %%mm3, %%mm4, %%mm5, %%mm0, OP, ADD, MUL1, MUL2)\ 325 VOP(%%mm2, %%mm3, %%mm4, %%mm5, %%mm0, %%mm1, OP, ADD, MUL1, MUL2)\ 326 VOP(%%mm3, %%mm4, %%mm5, %%mm0, %%mm1, %%mm2, OP, ADD, MUL1, MUL2)\ 327 \ 328 : "+a"(src), "+c"(dst)\ 329 : "S"((x86_reg)srcStride), "r"((x86_reg)dstStride)\ 330 NAMED_CONSTRAINTS_ADD(ADD,MUL1,MUL2)\ 331 : "memory"\ 332 );\ 333 }\ 334 src += 4-(h+5)*srcStride;\ 335 dst += 4-h*dstStride;\ 336 } 337 338#define QPEL_CAVS(OPNAME, OP, MMX)\ 339static void OPNAME ## cavs_qpel8_h_ ## MMX(uint8_t *dst, uint8_t *src, int dstStride, int srcStride){\ 340 int h=8;\ 341 __asm__ volatile(\ 342 "pxor %%mm7, %%mm7 \n\t"\ 343 "movq "MANGLE(ff_pw_5)", %%mm6\n\t"\ 344 "1: \n\t"\ 345 "movq (%0), %%mm0 \n\t"\ 346 "movq 1(%0), %%mm2 \n\t"\ 347 "movq %%mm0, %%mm1 \n\t"\ 348 "movq %%mm2, %%mm3 \n\t"\ 349 "punpcklbw %%mm7, %%mm0 \n\t"\ 350 "punpckhbw %%mm7, %%mm1 \n\t"\ 351 "punpcklbw %%mm7, %%mm2 \n\t"\ 352 "punpckhbw %%mm7, %%mm3 \n\t"\ 353 "paddw %%mm2, %%mm0 \n\t"\ 354 "paddw %%mm3, %%mm1 \n\t"\ 355 "pmullw %%mm6, %%mm0 \n\t"\ 356 "pmullw %%mm6, %%mm1 \n\t"\ 357 "movq -1(%0), %%mm2 \n\t"\ 358 "movq 2(%0), %%mm4 \n\t"\ 359 "movq %%mm2, %%mm3 \n\t"\ 360 "movq %%mm4, %%mm5 \n\t"\ 361 "punpcklbw %%mm7, %%mm2 \n\t"\ 362 "punpckhbw %%mm7, %%mm3 \n\t"\ 363 "punpcklbw %%mm7, %%mm4 \n\t"\ 364 "punpckhbw %%mm7, %%mm5 \n\t"\ 365 "paddw %%mm4, %%mm2 \n\t"\ 366 "paddw %%mm3, %%mm5 \n\t"\ 367 "psubw %%mm2, %%mm0 \n\t"\ 368 "psubw %%mm5, %%mm1 \n\t"\ 369 "movq "MANGLE(ff_pw_4)", %%mm5\n\t"\ 370 "paddw %%mm5, %%mm0 \n\t"\ 371 "paddw %%mm5, %%mm1 \n\t"\ 372 "psraw $3, %%mm0 \n\t"\ 373 "psraw $3, %%mm1 \n\t"\ 374 "packuswb %%mm1, %%mm0 \n\t"\ 375 OP(%%mm0, (%1),%%mm5, q) \ 376 "add %3, %0 \n\t"\ 377 "add %4, %1 \n\t"\ 378 "decl %2 \n\t"\ 379 " jnz 1b \n\t"\ 380 : "+a"(src), "+c"(dst), "+m"(h)\ 381 : "d"((x86_reg)srcStride), "S"((x86_reg)dstStride)\ 382 NAMED_CONSTRAINTS_ADD(ff_pw_4,ff_pw_5)\ 383 : "memory"\ 384 );\ 385}\ 386\ 387static inline void OPNAME ## cavs_qpel8or16_v1_ ## MMX(uint8_t *dst, uint8_t *src, int dstStride, int srcStride, int h){\ 388 QPEL_CAVSVNUM(QPEL_CAVSV1,OP,ff_pw_64,ff_pw_96,ff_pw_42) \ 389}\ 390\ 391static inline void OPNAME ## cavs_qpel8or16_v2_ ## MMX(uint8_t *dst, uint8_t *src, int dstStride, int srcStride, int h){\ 392 QPEL_CAVSVNUM(QPEL_CAVSV2,OP,ff_pw_4,ff_pw_5,ff_pw_42) \ 393}\ 394\ 395static inline void OPNAME ## cavs_qpel8or16_v3_ ## MMX(uint8_t *dst, uint8_t *src, int dstStride, int srcStride, int h){\ 396 QPEL_CAVSVNUM(QPEL_CAVSV3,OP,ff_pw_64,ff_pw_96,ff_pw_42) \ 397}\ 398\ 399static void OPNAME ## cavs_qpel8_v1_ ## MMX(uint8_t *dst, uint8_t *src, int dstStride, int srcStride){\ 400 OPNAME ## cavs_qpel8or16_v1_ ## MMX(dst , src , dstStride, srcStride, 8);\ 401}\ 402static void OPNAME ## cavs_qpel16_v1_ ## MMX(uint8_t *dst, uint8_t *src, int dstStride, int srcStride){\ 403 OPNAME ## cavs_qpel8or16_v1_ ## MMX(dst , src , dstStride, srcStride, 16);\ 404 OPNAME ## cavs_qpel8or16_v1_ ## MMX(dst+8, src+8, dstStride, srcStride, 16);\ 405}\ 406\ 407static void OPNAME ## cavs_qpel8_v2_ ## MMX(uint8_t *dst, uint8_t *src, int dstStride, int srcStride){\ 408 OPNAME ## cavs_qpel8or16_v2_ ## MMX(dst , src , dstStride, srcStride, 8);\ 409}\ 410static void OPNAME ## cavs_qpel16_v2_ ## MMX(uint8_t *dst, uint8_t *src, int dstStride, int srcStride){\ 411 OPNAME ## cavs_qpel8or16_v2_ ## MMX(dst , src , dstStride, srcStride, 16);\ 412 OPNAME ## cavs_qpel8or16_v2_ ## MMX(dst+8, src+8, dstStride, srcStride, 16);\ 413}\ 414\ 415static void OPNAME ## cavs_qpel8_v3_ ## MMX(uint8_t *dst, uint8_t *src, int dstStride, int srcStride){\ 416 OPNAME ## cavs_qpel8or16_v3_ ## MMX(dst , src , dstStride, srcStride, 8);\ 417}\ 418static void OPNAME ## cavs_qpel16_v3_ ## MMX(uint8_t *dst, uint8_t *src, int dstStride, int srcStride){\ 419 OPNAME ## cavs_qpel8or16_v3_ ## MMX(dst , src , dstStride, srcStride, 16);\ 420 OPNAME ## cavs_qpel8or16_v3_ ## MMX(dst+8, src+8, dstStride, srcStride, 16);\ 421}\ 422\ 423static void OPNAME ## cavs_qpel16_h_ ## MMX(uint8_t *dst, uint8_t *src, int dstStride, int srcStride){\ 424 OPNAME ## cavs_qpel8_h_ ## MMX(dst , src , dstStride, srcStride);\ 425 OPNAME ## cavs_qpel8_h_ ## MMX(dst+8, src+8, dstStride, srcStride);\ 426 src += 8*srcStride;\ 427 dst += 8*dstStride;\ 428 OPNAME ## cavs_qpel8_h_ ## MMX(dst , src , dstStride, srcStride);\ 429 OPNAME ## cavs_qpel8_h_ ## MMX(dst+8, src+8, dstStride, srcStride);\ 430}\ 431 432#define CAVS_MC(OPNAME, SIZE, MMX) \ 433static void OPNAME ## cavs_qpel ## SIZE ## _mc20_ ## MMX(uint8_t *dst, uint8_t *src, ptrdiff_t stride)\ 434{\ 435 OPNAME ## cavs_qpel ## SIZE ## _h_ ## MMX(dst, src, stride, stride);\ 436}\ 437\ 438static void OPNAME ## cavs_qpel ## SIZE ## _mc01_ ## MMX(uint8_t *dst, uint8_t *src, ptrdiff_t stride)\ 439{\ 440 OPNAME ## cavs_qpel ## SIZE ## _v1_ ## MMX(dst, src, stride, stride);\ 441}\ 442\ 443static void OPNAME ## cavs_qpel ## SIZE ## _mc02_ ## MMX(uint8_t *dst, uint8_t *src, ptrdiff_t stride)\ 444{\ 445 OPNAME ## cavs_qpel ## SIZE ## _v2_ ## MMX(dst, src, stride, stride);\ 446}\ 447\ 448static void OPNAME ## cavs_qpel ## SIZE ## _mc03_ ## MMX(uint8_t *dst, uint8_t *src, ptrdiff_t stride)\ 449{\ 450 OPNAME ## cavs_qpel ## SIZE ## _v3_ ## MMX(dst, src, stride, stride);\ 451}\ 452 453#define PUT_OP(a,b,temp, size) "mov" #size " " #a ", " #b " \n\t" 454#define AVG_3DNOW_OP(a,b,temp, size) \ 455"mov" #size " " #b ", " #temp " \n\t"\ 456"pavgusb " #temp ", " #a " \n\t"\ 457"mov" #size " " #a ", " #b " \n\t" 458#define AVG_MMXEXT_OP(a, b, temp, size) \ 459"mov" #size " " #b ", " #temp " \n\t"\ 460"pavgb " #temp ", " #a " \n\t"\ 461"mov" #size " " #a ", " #b " \n\t" 462 463#endif /* (HAVE_MMXEXT_INLINE || HAVE_AMD3DNOW_INLINE) */ 464 465#if HAVE_MMX_EXTERNAL 466static void put_cavs_qpel8_mc00_mmx(uint8_t *dst, uint8_t *src, 467 ptrdiff_t stride) 468{ 469 ff_put_pixels8_mmx(dst, src, stride, 8); 470} 471 472static void avg_cavs_qpel8_mc00_mmx(uint8_t *dst, uint8_t *src, 473 ptrdiff_t stride) 474{ 475 ff_avg_pixels8_mmx(dst, src, stride, 8); 476} 477 478static void avg_cavs_qpel8_mc00_mmxext(uint8_t *dst, uint8_t *src, 479 ptrdiff_t stride) 480{ 481 ff_avg_pixels8_mmxext(dst, src, stride, 8); 482} 483 484static void put_cavs_qpel16_mc00_mmx(uint8_t *dst, uint8_t *src, 485 ptrdiff_t stride) 486{ 487 ff_put_pixels16_mmx(dst, src, stride, 16); 488} 489 490static void avg_cavs_qpel16_mc00_mmx(uint8_t *dst, uint8_t *src, 491 ptrdiff_t stride) 492{ 493 ff_avg_pixels16_mmx(dst, src, stride, 16); 494} 495 496static void avg_cavs_qpel16_mc00_mmxext(uint8_t *dst, uint8_t *src, 497 ptrdiff_t stride) 498{ 499 ff_avg_pixels16_mmxext(dst, src, stride, 16); 500} 501 502static void put_cavs_qpel16_mc00_sse2(uint8_t *dst, uint8_t *src, 503 ptrdiff_t stride) 504{ 505 ff_put_pixels16_sse2(dst, src, stride, 16); 506} 507 508static void avg_cavs_qpel16_mc00_sse2(uint8_t *dst, uint8_t *src, 509 ptrdiff_t stride) 510{ 511 ff_avg_pixels16_sse2(dst, src, stride, 16); 512} 513#endif 514 515static av_cold void cavsdsp_init_mmx(CAVSDSPContext *c, 516 AVCodecContext *avctx) 517{ 518#if HAVE_MMX_EXTERNAL 519 c->put_cavs_qpel_pixels_tab[0][0] = put_cavs_qpel16_mc00_mmx; 520 c->put_cavs_qpel_pixels_tab[1][0] = put_cavs_qpel8_mc00_mmx; 521 c->avg_cavs_qpel_pixels_tab[0][0] = avg_cavs_qpel16_mc00_mmx; 522 c->avg_cavs_qpel_pixels_tab[1][0] = avg_cavs_qpel8_mc00_mmx; 523#endif 524 525#if HAVE_MMX_INLINE 526 c->cavs_idct8_add = cavs_idct8_add_mmx; 527 c->idct_perm = FF_TRANSPOSE_IDCT_PERM; 528#endif /* HAVE_MMX_INLINE */ 529} 530 531#define DSPFUNC(PFX, IDX, NUM, EXT) \ 532 c->PFX ## _cavs_qpel_pixels_tab[IDX][ 2] = PFX ## _cavs_qpel ## NUM ## _mc20_ ## EXT; \ 533 c->PFX ## _cavs_qpel_pixels_tab[IDX][ 4] = PFX ## _cavs_qpel ## NUM ## _mc01_ ## EXT; \ 534 c->PFX ## _cavs_qpel_pixels_tab[IDX][ 8] = PFX ## _cavs_qpel ## NUM ## _mc02_ ## EXT; \ 535 c->PFX ## _cavs_qpel_pixels_tab[IDX][12] = PFX ## _cavs_qpel ## NUM ## _mc03_ ## EXT; \ 536 537#if HAVE_MMXEXT_INLINE 538QPEL_CAVS(put_, PUT_OP, mmxext) 539QPEL_CAVS(avg_, AVG_MMXEXT_OP, mmxext) 540 541CAVS_MC(put_, 8, mmxext) 542CAVS_MC(put_, 16, mmxext) 543CAVS_MC(avg_, 8, mmxext) 544CAVS_MC(avg_, 16, mmxext) 545#endif /* HAVE_MMXEXT_INLINE */ 546 547#if HAVE_AMD3DNOW_INLINE 548QPEL_CAVS(put_, PUT_OP, 3dnow) 549QPEL_CAVS(avg_, AVG_3DNOW_OP, 3dnow) 550 551CAVS_MC(put_, 8, 3dnow) 552CAVS_MC(put_, 16,3dnow) 553CAVS_MC(avg_, 8, 3dnow) 554CAVS_MC(avg_, 16,3dnow) 555 556static av_cold void cavsdsp_init_3dnow(CAVSDSPContext *c, 557 AVCodecContext *avctx) 558{ 559 DSPFUNC(put, 0, 16, 3dnow); 560 DSPFUNC(put, 1, 8, 3dnow); 561 DSPFUNC(avg, 0, 16, 3dnow); 562 DSPFUNC(avg, 1, 8, 3dnow); 563} 564#endif /* HAVE_AMD3DNOW_INLINE */ 565 566av_cold void ff_cavsdsp_init_x86(CAVSDSPContext *c, AVCodecContext *avctx) 567{ 568 int cpu_flags = av_get_cpu_flags(); 569 570 cavsdsp_init_mmx(c, avctx); 571#if HAVE_AMD3DNOW_INLINE 572 if (INLINE_AMD3DNOW(cpu_flags)) 573 cavsdsp_init_3dnow(c, avctx); 574#endif /* HAVE_AMD3DNOW_INLINE */ 575#if HAVE_MMXEXT_INLINE 576 if (INLINE_MMXEXT(cpu_flags)) { 577 DSPFUNC(put, 0, 16, mmxext); 578 DSPFUNC(put, 1, 8, mmxext); 579 DSPFUNC(avg, 0, 16, mmxext); 580 DSPFUNC(avg, 1, 8, mmxext); 581 } 582#endif 583#if HAVE_MMX_EXTERNAL 584 if (EXTERNAL_MMXEXT(cpu_flags)) { 585 c->avg_cavs_qpel_pixels_tab[0][0] = avg_cavs_qpel16_mc00_mmxext; 586 c->avg_cavs_qpel_pixels_tab[1][0] = avg_cavs_qpel8_mc00_mmxext; 587 } 588#endif 589#if HAVE_SSE2_EXTERNAL 590 if (EXTERNAL_SSE2(cpu_flags)) { 591 c->put_cavs_qpel_pixels_tab[0][0] = put_cavs_qpel16_mc00_sse2; 592 c->avg_cavs_qpel_pixels_tab[0][0] = avg_cavs_qpel16_mc00_sse2; 593 } 594#endif 595} 596