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/mem_internal.h" 29#include "libavutil/x86/asm.h" 30#include "libavutil/x86/cpu.h" 31#include "libavcodec/cavsdsp.h" 32#include "libavcodec/idctdsp.h" 33#include "constants.h" 34#include "fpel.h" 35#include "idctdsp.h" 36#include "config.h" 37 38 39#if HAVE_SSE2_EXTERNAL 40 41void ff_cavs_idct8_sse2(int16_t *out, const int16_t *in); 42 43static void cavs_idct8_add_sse2(uint8_t *dst, int16_t *block, ptrdiff_t stride) 44{ 45 LOCAL_ALIGNED(16, int16_t, b2, [64]); 46 ff_cavs_idct8_sse2(b2, block); 47 ff_add_pixels_clamped_sse2(b2, dst, stride); 48} 49 50#endif /* HAVE_SSE2_EXTERNAL */ 51 52#if HAVE_MMXEXT_INLINE 53 54/***************************************************************************** 55 * 56 * motion compensation 57 * 58 ****************************************************************************/ 59 60/* vertical filter [-1 -2 96 42 -7 0] */ 61#define QPEL_CAVSV1(A,B,C,D,E,F,OP,ADD, MUL1, MUL2) \ 62 "movd (%0), "#F" \n\t"\ 63 "movq "#C", %%mm6 \n\t"\ 64 "pmullw "MANGLE(MUL1)", %%mm6\n\t"\ 65 "movq "#D", %%mm7 \n\t"\ 66 "pmullw "MANGLE(MUL2)", %%mm7\n\t"\ 67 "psllw $3, "#E" \n\t"\ 68 "psubw "#E", %%mm6 \n\t"\ 69 "psraw $3, "#E" \n\t"\ 70 "paddw %%mm7, %%mm6 \n\t"\ 71 "paddw "#E", %%mm6 \n\t"\ 72 "paddw "#B", "#B" \n\t"\ 73 "pxor %%mm7, %%mm7 \n\t"\ 74 "add %2, %0 \n\t"\ 75 "punpcklbw %%mm7, "#F" \n\t"\ 76 "psubw "#B", %%mm6 \n\t"\ 77 "psraw $1, "#B" \n\t"\ 78 "psubw "#A", %%mm6 \n\t"\ 79 "paddw "MANGLE(ADD)", %%mm6 \n\t"\ 80 "psraw $7, %%mm6 \n\t"\ 81 "packuswb %%mm6, %%mm6 \n\t"\ 82 OP(%%mm6, (%1), A, d) \ 83 "add %3, %1 \n\t" 84 85/* vertical filter [ 0 -1 5 5 -1 0] */ 86#define QPEL_CAVSV2(A,B,C,D,E,F,OP,ADD, MUL1, MUL2) \ 87 "movd (%0), "#F" \n\t"\ 88 "movq "#C", %%mm6 \n\t"\ 89 "paddw "#D", %%mm6 \n\t"\ 90 "pmullw "MANGLE(MUL1)", %%mm6\n\t"\ 91 "add %2, %0 \n\t"\ 92 "punpcklbw %%mm7, "#F" \n\t"\ 93 "psubw "#B", %%mm6 \n\t"\ 94 "psubw "#E", %%mm6 \n\t"\ 95 "paddw "MANGLE(ADD)", %%mm6 \n\t"\ 96 "psraw $3, %%mm6 \n\t"\ 97 "packuswb %%mm6, %%mm6 \n\t"\ 98 OP(%%mm6, (%1), A, d) \ 99 "add %3, %1 \n\t" 100 101/* vertical filter [ 0 -7 42 96 -2 -1] */ 102#define QPEL_CAVSV3(A,B,C,D,E,F,OP,ADD, MUL1, MUL2) \ 103 "movd (%0), "#F" \n\t"\ 104 "movq "#C", %%mm6 \n\t"\ 105 "pmullw "MANGLE(MUL2)", %%mm6\n\t"\ 106 "movq "#D", %%mm7 \n\t"\ 107 "pmullw "MANGLE(MUL1)", %%mm7\n\t"\ 108 "psllw $3, "#B" \n\t"\ 109 "psubw "#B", %%mm6 \n\t"\ 110 "psraw $3, "#B" \n\t"\ 111 "paddw %%mm7, %%mm6 \n\t"\ 112 "paddw "#B", %%mm6 \n\t"\ 113 "paddw "#E", "#E" \n\t"\ 114 "pxor %%mm7, %%mm7 \n\t"\ 115 "add %2, %0 \n\t"\ 116 "punpcklbw %%mm7, "#F" \n\t"\ 117 "psubw "#E", %%mm6 \n\t"\ 118 "psraw $1, "#E" \n\t"\ 119 "psubw "#F", %%mm6 \n\t"\ 120 "paddw "MANGLE(ADD)", %%mm6 \n\t"\ 121 "psraw $7, %%mm6 \n\t"\ 122 "packuswb %%mm6, %%mm6 \n\t"\ 123 OP(%%mm6, (%1), A, d) \ 124 "add %3, %1 \n\t" 125 126 127#define QPEL_CAVSVNUM(VOP,OP,ADD,MUL1,MUL2)\ 128 int w= 2;\ 129 src -= 2*srcStride;\ 130 \ 131 while(w--){\ 132 __asm__ volatile(\ 133 "pxor %%mm7, %%mm7 \n\t"\ 134 "movd (%0), %%mm0 \n\t"\ 135 "add %2, %0 \n\t"\ 136 "movd (%0), %%mm1 \n\t"\ 137 "add %2, %0 \n\t"\ 138 "movd (%0), %%mm2 \n\t"\ 139 "add %2, %0 \n\t"\ 140 "movd (%0), %%mm3 \n\t"\ 141 "add %2, %0 \n\t"\ 142 "movd (%0), %%mm4 \n\t"\ 143 "add %2, %0 \n\t"\ 144 "punpcklbw %%mm7, %%mm0 \n\t"\ 145 "punpcklbw %%mm7, %%mm1 \n\t"\ 146 "punpcklbw %%mm7, %%mm2 \n\t"\ 147 "punpcklbw %%mm7, %%mm3 \n\t"\ 148 "punpcklbw %%mm7, %%mm4 \n\t"\ 149 VOP(%%mm0, %%mm1, %%mm2, %%mm3, %%mm4, %%mm5, OP, ADD, MUL1, MUL2)\ 150 VOP(%%mm1, %%mm2, %%mm3, %%mm4, %%mm5, %%mm0, OP, ADD, MUL1, MUL2)\ 151 VOP(%%mm2, %%mm3, %%mm4, %%mm5, %%mm0, %%mm1, OP, ADD, MUL1, MUL2)\ 152 VOP(%%mm3, %%mm4, %%mm5, %%mm0, %%mm1, %%mm2, OP, ADD, MUL1, MUL2)\ 153 VOP(%%mm4, %%mm5, %%mm0, %%mm1, %%mm2, %%mm3, OP, ADD, MUL1, MUL2)\ 154 VOP(%%mm5, %%mm0, %%mm1, %%mm2, %%mm3, %%mm4, OP, ADD, MUL1, MUL2)\ 155 VOP(%%mm0, %%mm1, %%mm2, %%mm3, %%mm4, %%mm5, OP, ADD, MUL1, MUL2)\ 156 VOP(%%mm1, %%mm2, %%mm3, %%mm4, %%mm5, %%mm0, OP, ADD, MUL1, MUL2)\ 157 \ 158 : "+a"(src), "+c"(dst)\ 159 : "S"((x86_reg)srcStride), "r"((x86_reg)dstStride)\ 160 NAMED_CONSTRAINTS_ADD(ADD,MUL1,MUL2)\ 161 : "memory"\ 162 );\ 163 if(h==16){\ 164 __asm__ volatile(\ 165 VOP(%%mm2, %%mm3, %%mm4, %%mm5, %%mm0, %%mm1, OP, ADD, MUL1, MUL2)\ 166 VOP(%%mm3, %%mm4, %%mm5, %%mm0, %%mm1, %%mm2, OP, ADD, MUL1, MUL2)\ 167 VOP(%%mm4, %%mm5, %%mm0, %%mm1, %%mm2, %%mm3, OP, ADD, MUL1, MUL2)\ 168 VOP(%%mm5, %%mm0, %%mm1, %%mm2, %%mm3, %%mm4, OP, ADD, MUL1, MUL2)\ 169 VOP(%%mm0, %%mm1, %%mm2, %%mm3, %%mm4, %%mm5, OP, ADD, MUL1, MUL2)\ 170 VOP(%%mm1, %%mm2, %%mm3, %%mm4, %%mm5, %%mm0, OP, ADD, MUL1, MUL2)\ 171 VOP(%%mm2, %%mm3, %%mm4, %%mm5, %%mm0, %%mm1, OP, ADD, MUL1, MUL2)\ 172 VOP(%%mm3, %%mm4, %%mm5, %%mm0, %%mm1, %%mm2, OP, ADD, MUL1, MUL2)\ 173 \ 174 : "+a"(src), "+c"(dst)\ 175 : "S"((x86_reg)srcStride), "r"((x86_reg)dstStride)\ 176 NAMED_CONSTRAINTS_ADD(ADD,MUL1,MUL2)\ 177 : "memory"\ 178 );\ 179 }\ 180 src += 4-(h+5)*srcStride;\ 181 dst += 4-h*dstStride;\ 182 } 183 184#define QPEL_CAVS(OPNAME, OP, MMX)\ 185static void OPNAME ## cavs_qpel8_h_ ## MMX(uint8_t *dst, const uint8_t *src, ptrdiff_t dstStride, ptrdiff_t srcStride)\ 186{\ 187 int h=8;\ 188 __asm__ volatile(\ 189 "pxor %%mm7, %%mm7 \n\t"\ 190 "movq "MANGLE(ff_pw_5)", %%mm6\n\t"\ 191 "1: \n\t"\ 192 "movq (%0), %%mm0 \n\t"\ 193 "movq 1(%0), %%mm2 \n\t"\ 194 "movq %%mm0, %%mm1 \n\t"\ 195 "movq %%mm2, %%mm3 \n\t"\ 196 "punpcklbw %%mm7, %%mm0 \n\t"\ 197 "punpckhbw %%mm7, %%mm1 \n\t"\ 198 "punpcklbw %%mm7, %%mm2 \n\t"\ 199 "punpckhbw %%mm7, %%mm3 \n\t"\ 200 "paddw %%mm2, %%mm0 \n\t"\ 201 "paddw %%mm3, %%mm1 \n\t"\ 202 "pmullw %%mm6, %%mm0 \n\t"\ 203 "pmullw %%mm6, %%mm1 \n\t"\ 204 "movq -1(%0), %%mm2 \n\t"\ 205 "movq 2(%0), %%mm4 \n\t"\ 206 "movq %%mm2, %%mm3 \n\t"\ 207 "movq %%mm4, %%mm5 \n\t"\ 208 "punpcklbw %%mm7, %%mm2 \n\t"\ 209 "punpckhbw %%mm7, %%mm3 \n\t"\ 210 "punpcklbw %%mm7, %%mm4 \n\t"\ 211 "punpckhbw %%mm7, %%mm5 \n\t"\ 212 "paddw %%mm4, %%mm2 \n\t"\ 213 "paddw %%mm3, %%mm5 \n\t"\ 214 "psubw %%mm2, %%mm0 \n\t"\ 215 "psubw %%mm5, %%mm1 \n\t"\ 216 "movq "MANGLE(ff_pw_4)", %%mm5\n\t"\ 217 "paddw %%mm5, %%mm0 \n\t"\ 218 "paddw %%mm5, %%mm1 \n\t"\ 219 "psraw $3, %%mm0 \n\t"\ 220 "psraw $3, %%mm1 \n\t"\ 221 "packuswb %%mm1, %%mm0 \n\t"\ 222 OP(%%mm0, (%1),%%mm5, q) \ 223 "add %3, %0 \n\t"\ 224 "add %4, %1 \n\t"\ 225 "decl %2 \n\t"\ 226 " jnz 1b \n\t"\ 227 : "+a"(src), "+c"(dst), "+m"(h)\ 228 : "d"((x86_reg)srcStride), "S"((x86_reg)dstStride)\ 229 NAMED_CONSTRAINTS_ADD(ff_pw_4,ff_pw_5)\ 230 : "memory"\ 231 );\ 232}\ 233\ 234static inline void OPNAME ## cavs_qpel8or16_v1_ ## MMX(uint8_t *dst, const uint8_t *src, ptrdiff_t dstStride, ptrdiff_t srcStride, int h)\ 235{ \ 236 QPEL_CAVSVNUM(QPEL_CAVSV1,OP,ff_pw_64,ff_pw_96,ff_pw_42) \ 237}\ 238\ 239static inline void OPNAME ## cavs_qpel8or16_v2_ ## MMX(uint8_t *dst, const uint8_t *src, ptrdiff_t dstStride, ptrdiff_t srcStride, int h)\ 240{ \ 241 QPEL_CAVSVNUM(QPEL_CAVSV2,OP,ff_pw_4,ff_pw_5,ff_pw_42) \ 242}\ 243\ 244static inline void OPNAME ## cavs_qpel8or16_v3_ ## MMX(uint8_t *dst, const uint8_t *src, ptrdiff_t dstStride, ptrdiff_t srcStride, int h)\ 245{ \ 246 QPEL_CAVSVNUM(QPEL_CAVSV3,OP,ff_pw_64,ff_pw_96,ff_pw_42) \ 247}\ 248\ 249static void OPNAME ## cavs_qpel8_v1_ ## MMX(uint8_t *dst, const uint8_t *src, ptrdiff_t dstStride, ptrdiff_t srcStride)\ 250{ \ 251 OPNAME ## cavs_qpel8or16_v1_ ## MMX(dst , src , dstStride, srcStride, 8);\ 252}\ 253static void OPNAME ## cavs_qpel16_v1_ ## MMX(uint8_t *dst, const uint8_t *src, ptrdiff_t dstStride, ptrdiff_t srcStride)\ 254{ \ 255 OPNAME ## cavs_qpel8or16_v1_ ## MMX(dst , src , dstStride, srcStride, 16);\ 256 OPNAME ## cavs_qpel8or16_v1_ ## MMX(dst+8, src+8, dstStride, srcStride, 16);\ 257}\ 258\ 259static void OPNAME ## cavs_qpel8_v2_ ## MMX(uint8_t *dst, const uint8_t *src, ptrdiff_t dstStride, ptrdiff_t srcStride)\ 260{ \ 261 OPNAME ## cavs_qpel8or16_v2_ ## MMX(dst , src , dstStride, srcStride, 8);\ 262}\ 263static void OPNAME ## cavs_qpel16_v2_ ## MMX(uint8_t *dst, const uint8_t *src, ptrdiff_t dstStride, ptrdiff_t srcStride)\ 264{ \ 265 OPNAME ## cavs_qpel8or16_v2_ ## MMX(dst , src , dstStride, srcStride, 16);\ 266 OPNAME ## cavs_qpel8or16_v2_ ## MMX(dst+8, src+8, dstStride, srcStride, 16);\ 267}\ 268\ 269static void OPNAME ## cavs_qpel8_v3_ ## MMX(uint8_t *dst, const uint8_t *src, ptrdiff_t dstStride, ptrdiff_t srcStride)\ 270{ \ 271 OPNAME ## cavs_qpel8or16_v3_ ## MMX(dst , src , dstStride, srcStride, 8);\ 272}\ 273static void OPNAME ## cavs_qpel16_v3_ ## MMX(uint8_t *dst, const uint8_t *src, ptrdiff_t dstStride, ptrdiff_t srcStride)\ 274{ \ 275 OPNAME ## cavs_qpel8or16_v3_ ## MMX(dst , src , dstStride, srcStride, 16);\ 276 OPNAME ## cavs_qpel8or16_v3_ ## MMX(dst+8, src+8, dstStride, srcStride, 16);\ 277}\ 278\ 279static void OPNAME ## cavs_qpel16_h_ ## MMX(uint8_t *dst, const uint8_t *src, ptrdiff_t dstStride, ptrdiff_t srcStride)\ 280{ \ 281 OPNAME ## cavs_qpel8_h_ ## MMX(dst , src , dstStride, srcStride);\ 282 OPNAME ## cavs_qpel8_h_ ## MMX(dst+8, src+8, dstStride, srcStride);\ 283 src += 8*srcStride;\ 284 dst += 8*dstStride;\ 285 OPNAME ## cavs_qpel8_h_ ## MMX(dst , src , dstStride, srcStride);\ 286 OPNAME ## cavs_qpel8_h_ ## MMX(dst+8, src+8, dstStride, srcStride);\ 287}\ 288 289#define CAVS_MC(OPNAME, SIZE, MMX) \ 290static void OPNAME ## cavs_qpel ## SIZE ## _mc20_ ## MMX(uint8_t *dst, const uint8_t *src, ptrdiff_t stride)\ 291{\ 292 OPNAME ## cavs_qpel ## SIZE ## _h_ ## MMX(dst, src, stride, stride);\ 293}\ 294\ 295static void OPNAME ## cavs_qpel ## SIZE ## _mc01_ ## MMX(uint8_t *dst, const uint8_t *src, ptrdiff_t stride)\ 296{\ 297 OPNAME ## cavs_qpel ## SIZE ## _v1_ ## MMX(dst, src, stride, stride);\ 298}\ 299\ 300static void OPNAME ## cavs_qpel ## SIZE ## _mc02_ ## MMX(uint8_t *dst, const uint8_t *src, ptrdiff_t stride)\ 301{\ 302 OPNAME ## cavs_qpel ## SIZE ## _v2_ ## MMX(dst, src, stride, stride);\ 303}\ 304\ 305static void OPNAME ## cavs_qpel ## SIZE ## _mc03_ ## MMX(uint8_t *dst, const uint8_t *src, ptrdiff_t stride)\ 306{\ 307 OPNAME ## cavs_qpel ## SIZE ## _v3_ ## MMX(dst, src, stride, stride);\ 308}\ 309 310#define PUT_OP(a,b,temp, size) "mov" #size " " #a ", " #b " \n\t" 311#define AVG_3DNOW_OP(a,b,temp, size) \ 312"mov" #size " " #b ", " #temp " \n\t"\ 313"pavgusb " #temp ", " #a " \n\t"\ 314"mov" #size " " #a ", " #b " \n\t" 315#define AVG_MMXEXT_OP(a, b, temp, size) \ 316"mov" #size " " #b ", " #temp " \n\t"\ 317"pavgb " #temp ", " #a " \n\t"\ 318"mov" #size " " #a ", " #b " \n\t" 319 320#endif /* HAVE_MMXEXT_INLINE */ 321 322#if HAVE_MMX_EXTERNAL 323static void put_cavs_qpel8_mc00_mmx(uint8_t *dst, const uint8_t *src, 324 ptrdiff_t stride) 325{ 326 ff_put_pixels8_mmx(dst, src, stride, 8); 327} 328 329static void avg_cavs_qpel8_mc00_mmxext(uint8_t *dst, const uint8_t *src, 330 ptrdiff_t stride) 331{ 332 ff_avg_pixels8_mmxext(dst, src, stride, 8); 333} 334 335static void put_cavs_qpel16_mc00_sse2(uint8_t *dst, const uint8_t *src, 336 ptrdiff_t stride) 337{ 338 ff_put_pixels16_sse2(dst, src, stride, 16); 339} 340 341static void avg_cavs_qpel16_mc00_sse2(uint8_t *dst, const uint8_t *src, 342 ptrdiff_t stride) 343{ 344 ff_avg_pixels16_sse2(dst, src, stride, 16); 345} 346#endif 347 348static av_cold void cavsdsp_init_mmx(CAVSDSPContext *c, 349 AVCodecContext *avctx) 350{ 351#if HAVE_MMX_EXTERNAL 352 c->put_cavs_qpel_pixels_tab[1][0] = put_cavs_qpel8_mc00_mmx; 353#endif /* HAVE_MMX_EXTERNAL */ 354} 355 356#define DSPFUNC(PFX, IDX, NUM, EXT) \ 357 c->PFX ## _cavs_qpel_pixels_tab[IDX][ 2] = PFX ## _cavs_qpel ## NUM ## _mc20_ ## EXT; \ 358 c->PFX ## _cavs_qpel_pixels_tab[IDX][ 4] = PFX ## _cavs_qpel ## NUM ## _mc01_ ## EXT; \ 359 c->PFX ## _cavs_qpel_pixels_tab[IDX][ 8] = PFX ## _cavs_qpel ## NUM ## _mc02_ ## EXT; \ 360 c->PFX ## _cavs_qpel_pixels_tab[IDX][12] = PFX ## _cavs_qpel ## NUM ## _mc03_ ## EXT; \ 361 362#if HAVE_MMXEXT_INLINE 363QPEL_CAVS(put_, PUT_OP, mmxext) 364QPEL_CAVS(avg_, AVG_MMXEXT_OP, mmxext) 365 366CAVS_MC(put_, 8, mmxext) 367CAVS_MC(put_, 16, mmxext) 368CAVS_MC(avg_, 8, mmxext) 369CAVS_MC(avg_, 16, mmxext) 370#endif /* HAVE_MMXEXT_INLINE */ 371 372av_cold void ff_cavsdsp_init_x86(CAVSDSPContext *c, AVCodecContext *avctx) 373{ 374 av_unused int cpu_flags = av_get_cpu_flags(); 375 376 if (X86_MMX(cpu_flags)) 377 cavsdsp_init_mmx(c, avctx); 378 379#if HAVE_MMXEXT_INLINE 380 if (INLINE_MMXEXT(cpu_flags)) { 381 DSPFUNC(put, 0, 16, mmxext); 382 DSPFUNC(put, 1, 8, mmxext); 383 DSPFUNC(avg, 0, 16, mmxext); 384 DSPFUNC(avg, 1, 8, mmxext); 385 } 386#endif 387#if HAVE_MMX_EXTERNAL 388 if (EXTERNAL_MMXEXT(cpu_flags)) { 389 c->avg_cavs_qpel_pixels_tab[1][0] = avg_cavs_qpel8_mc00_mmxext; 390 } 391#endif 392#if HAVE_SSE2_EXTERNAL 393 if (EXTERNAL_SSE2(cpu_flags)) { 394 c->put_cavs_qpel_pixels_tab[0][0] = put_cavs_qpel16_mc00_sse2; 395 c->avg_cavs_qpel_pixels_tab[0][0] = avg_cavs_qpel16_mc00_sse2; 396 397 c->cavs_idct8_add = cavs_idct8_add_sse2; 398 c->idct_perm = FF_IDCT_PERM_TRANSPOSE; 399 } 400#endif 401} 402