git.sesse.net Git - ffmpeg/blob - libswscale/x86/swscale_mmx.c

   1 /*
   2  * Copyright (C) 2001-2003 Michael Niedermayer <michaelni@gmx.at>
   3  *
   4  * This file is part of FFmpeg.
   5  *
   6  * FFmpeg is free software; you can redistribute it and/or
   7  * modify it under the terms of the GNU Lesser General Public
   8  * License as published by the Free Software Foundation; either
   9  * version 2.1 of the License, or (at your option) any later version.
  10  *
  11  * FFmpeg is distributed in the hope that it will be useful,
  12  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  13  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  14  * Lesser General Public License for more details.
  15  *
  16  * You should have received a copy of the GNU Lesser General Public
  17  * License along with FFmpeg; if not, write to the Free Software
  18  * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
  19  */
  20
  21 #include <inttypes.h>
  22 #include "config.h"
  23 #include "libswscale/swscale.h"
  24 #include "libswscale/swscale_internal.h"
  25 #include "libavutil/intreadwrite.h"
  26 #include "libavutil/x86_cpu.h"
  27 #include "libavutil/cpu.h"
  28 #include "libavutil/pixdesc.h"
  29
  30 DECLARE_ASM_CONST(8, uint64_t, bF8)=       0xF8F8F8F8F8F8F8F8LL;
  31 DECLARE_ASM_CONST(8, uint64_t, bFC)=       0xFCFCFCFCFCFCFCFCLL;
  32 DECLARE_ASM_CONST(8, uint64_t, w10)=       0x0010001000100010LL;
  33 DECLARE_ASM_CONST(8, uint64_t, w02)=       0x0002000200020002LL;
  34 DECLARE_ASM_CONST(8, uint64_t, bm00001111)=0x00000000FFFFFFFFLL;
  35 DECLARE_ASM_CONST(8, uint64_t, bm00000111)=0x0000000000FFFFFFLL;
  36 DECLARE_ASM_CONST(8, uint64_t, bm11111000)=0xFFFFFFFFFF000000LL;
  37 DECLARE_ASM_CONST(8, uint64_t, bm01010101)=0x00FF00FF00FF00FFLL;
  38
  39 const DECLARE_ALIGNED(8, uint64_t, ff_dither4)[2] = {
  40     0x0103010301030103LL,
  41     0x0200020002000200LL,};
  42
  43 const DECLARE_ALIGNED(8, uint64_t, ff_dither8)[2] = {
  44     0x0602060206020602LL,
  45     0x0004000400040004LL,};
  46
  47 DECLARE_ASM_CONST(8, uint64_t, b16Mask)=   0x001F001F001F001FLL;
  48 DECLARE_ASM_CONST(8, uint64_t, g16Mask)=   0x07E007E007E007E0LL;
  49 DECLARE_ASM_CONST(8, uint64_t, r16Mask)=   0xF800F800F800F800LL;
  50 DECLARE_ASM_CONST(8, uint64_t, b15Mask)=   0x001F001F001F001FLL;
  51 DECLARE_ASM_CONST(8, uint64_t, g15Mask)=   0x03E003E003E003E0LL;
  52 DECLARE_ASM_CONST(8, uint64_t, r15Mask)=   0x7C007C007C007C00LL;
  53
  54 DECLARE_ALIGNED(8, const uint64_t, ff_M24A)         = 0x00FF0000FF0000FFLL;
  55 DECLARE_ALIGNED(8, const uint64_t, ff_M24B)         = 0xFF0000FF0000FF00LL;
  56 DECLARE_ALIGNED(8, const uint64_t, ff_M24C)         = 0x0000FF0000FF0000LL;
  57
  58 #ifdef FAST_BGR2YV12
  59 DECLARE_ALIGNED(8, const uint64_t, ff_bgr2YCoeff)   = 0x000000210041000DULL;
  60 DECLARE_ALIGNED(8, const uint64_t, ff_bgr2UCoeff)   = 0x0000FFEEFFDC0038ULL;
  61 DECLARE_ALIGNED(8, const uint64_t, ff_bgr2VCoeff)   = 0x00000038FFD2FFF8ULL;
  62 #else
  63 DECLARE_ALIGNED(8, const uint64_t, ff_bgr2YCoeff)   = 0x000020E540830C8BULL;
  64 DECLARE_ALIGNED(8, const uint64_t, ff_bgr2UCoeff)   = 0x0000ED0FDAC23831ULL;
  65 DECLARE_ALIGNED(8, const uint64_t, ff_bgr2VCoeff)   = 0x00003831D0E6F6EAULL;
  66 #endif /* FAST_BGR2YV12 */
  67 DECLARE_ALIGNED(8, const uint64_t, ff_bgr2YOffset)  = 0x1010101010101010ULL;
  68 DECLARE_ALIGNED(8, const uint64_t, ff_bgr2UVOffset) = 0x8080808080808080ULL;
  69 DECLARE_ALIGNED(8, const uint64_t, ff_w1111)        = 0x0001000100010001ULL;
  70
  71 DECLARE_ASM_CONST(8, uint64_t, ff_bgr24toY1Coeff) = 0x0C88000040870C88ULL;
  72 DECLARE_ASM_CONST(8, uint64_t, ff_bgr24toY2Coeff) = 0x20DE4087000020DEULL;
  73 DECLARE_ASM_CONST(8, uint64_t, ff_rgb24toY1Coeff) = 0x20DE0000408720DEULL;
  74 DECLARE_ASM_CONST(8, uint64_t, ff_rgb24toY2Coeff) = 0x0C88408700000C88ULL;
  75 DECLARE_ASM_CONST(8, uint64_t, ff_bgr24toYOffset) = 0x0008010000080100ULL;
  76
  77 DECLARE_ASM_CONST(8, uint64_t, ff_bgr24toUV)[2][4] = {
  78     {0x38380000DAC83838ULL, 0xECFFDAC80000ECFFULL, 0xF6E40000D0E3F6E4ULL, 0x3838D0E300003838ULL},
  79     {0xECFF0000DAC8ECFFULL, 0x3838DAC800003838ULL, 0x38380000D0E33838ULL, 0xF6E4D0E30000F6E4ULL},
  80 };
  81
  82 DECLARE_ASM_CONST(8, uint64_t, ff_bgr24toUVOffset)= 0x0040010000400100ULL;
  83
  84 //MMX versions
  85 #if HAVE_MMX
  86 #undef RENAME
  87 #define COMPILE_TEMPLATE_MMX2 0
  88 #define RENAME(a) a ## _MMX
  89 #include "swscale_template.c"
  90 #endif
  91
  92 //MMX2 versions
  93 #if HAVE_MMX2
  94 #undef RENAME
  95 #undef COMPILE_TEMPLATE_MMX2
  96 #define COMPILE_TEMPLATE_MMX2 1
  97 #define RENAME(a) a ## _MMX2
  98 #include "swscale_template.c"
  99 #endif
 100
 101 void updateMMXDitherTables(SwsContext *c, int dstY, int lumBufIndex, int chrBufIndex,
 102                            int lastInLumBuf, int lastInChrBuf)
 103 {
 104     const int dstH= c->dstH;
 105     const int flags= c->flags;
 106     int16_t **lumPixBuf= c->lumPixBuf;
 107     int16_t **chrUPixBuf= c->chrUPixBuf;
 108     int16_t **alpPixBuf= c->alpPixBuf;
 109     const int vLumBufSize= c->vLumBufSize;
 110     const int vChrBufSize= c->vChrBufSize;
 111     int16_t *vLumFilterPos= c->vLumFilterPos;
 112     int16_t *vChrFilterPos= c->vChrFilterPos;
 113     int16_t *vLumFilter= c->vLumFilter;
 114     int16_t *vChrFilter= c->vChrFilter;
 115     int32_t *lumMmxFilter= c->lumMmxFilter;
 116     int32_t *chrMmxFilter= c->chrMmxFilter;
 117     int32_t av_unused *alpMmxFilter= c->alpMmxFilter;
 118     const int vLumFilterSize= c->vLumFilterSize;
 119     const int vChrFilterSize= c->vChrFilterSize;
 120     const int chrDstY= dstY>>c->chrDstVSubSample;
 121     const int firstLumSrcY= vLumFilterPos[dstY]; //First line needed as input
 122     const int firstChrSrcY= vChrFilterPos[chrDstY]; //First line needed as input
 123
 124     c->blueDither= ff_dither8[dstY&1];
 125     if (c->dstFormat == PIX_FMT_RGB555 || c->dstFormat == PIX_FMT_BGR555)
 126         c->greenDither= ff_dither8[dstY&1];
 127     else
 128         c->greenDither= ff_dither4[dstY&1];
 129     c->redDither= ff_dither8[(dstY+1)&1];
 130     if (dstY < dstH - 2) {
 131         const int16_t **lumSrcPtr= (const int16_t **) lumPixBuf + lumBufIndex + firstLumSrcY - lastInLumBuf + vLumBufSize;
 132         const int16_t **chrUSrcPtr= (const int16_t **) chrUPixBuf + chrBufIndex + firstChrSrcY - lastInChrBuf + vChrBufSize;
 133         const int16_t **alpSrcPtr= (CONFIG_SWSCALE_ALPHA && alpPixBuf) ? (const int16_t **) alpPixBuf + lumBufIndex + firstLumSrcY - lastInLumBuf + vLumBufSize : NULL;
 134         int i;
 135         if (flags & SWS_ACCURATE_RND) {
 136             int s= APCK_SIZE / 8;
 137             for (i=0; i<vLumFilterSize; i+=2) {
 138                 *(const void**)&lumMmxFilter[s*i              ]= lumSrcPtr[i  ];
 139                 *(const void**)&lumMmxFilter[s*i+APCK_PTR2/4  ]= lumSrcPtr[i+(vLumFilterSize>1)];
 140                 lumMmxFilter[s*i+APCK_COEF/4  ]=
 141                 lumMmxFilter[s*i+APCK_COEF/4+1]= vLumFilter[dstY*vLumFilterSize + i    ]
 142                 + (vLumFilterSize>1 ? vLumFilter[dstY*vLumFilterSize + i + 1]<<16 : 0);
 143                 if (CONFIG_SWSCALE_ALPHA && alpPixBuf) {
 144                     *(const void**)&alpMmxFilter[s*i              ]= alpSrcPtr[i  ];
 145                     *(const void**)&alpMmxFilter[s*i+APCK_PTR2/4  ]= alpSrcPtr[i+(vLumFilterSize>1)];
 146                     alpMmxFilter[s*i+APCK_COEF/4  ]=
 147                     alpMmxFilter[s*i+APCK_COEF/4+1]= lumMmxFilter[s*i+APCK_COEF/4  ];
 148                 }
 149             }
 150             for (i=0; i<vChrFilterSize; i+=2) {
 151                 *(const void**)&chrMmxFilter[s*i              ]= chrUSrcPtr[i  ];
 152                 *(const void**)&chrMmxFilter[s*i+APCK_PTR2/4  ]= chrUSrcPtr[i+(vChrFilterSize>1)];
 153                 chrMmxFilter[s*i+APCK_COEF/4  ]=
 154                 chrMmxFilter[s*i+APCK_COEF/4+1]= vChrFilter[chrDstY*vChrFilterSize + i    ]
 155                 + (vChrFilterSize>1 ? vChrFilter[chrDstY*vChrFilterSize + i + 1]<<16 : 0);
 156             }
 157         } else {
 158             for (i=0; i<vLumFilterSize; i++) {
 159                 *(const void**)&lumMmxFilter[4*i+0]= lumSrcPtr[i];
 160                 lumMmxFilter[4*i+2]=
 161                 lumMmxFilter[4*i+3]=
 162                 ((uint16_t)vLumFilter[dstY*vLumFilterSize + i])*0x10001;
 163                 if (CONFIG_SWSCALE_ALPHA && alpPixBuf) {
 164                     *(const void**)&alpMmxFilter[4*i+0]= alpSrcPtr[i];
 165                     alpMmxFilter[4*i+2]=
 166                     alpMmxFilter[4*i+3]= lumMmxFilter[4*i+2];
 167                 }
 168             }
 169             for (i=0; i<vChrFilterSize; i++) {
 170                 *(const void**)&chrMmxFilter[4*i+0]= chrUSrcPtr[i];
 171                 chrMmxFilter[4*i+2]=
 172                 chrMmxFilter[4*i+3]=
 173                 ((uint16_t)vChrFilter[chrDstY*vChrFilterSize + i])*0x10001;
 174             }
 175         }
 176     }
 177 }
 178
 179 #define SCALE_FUNC(filter_n, from_bpc, to_bpc, opt) \
 180 extern void ff_hscale ## from_bpc ## to ## to_bpc ## _ ## filter_n ## _ ## opt( \
 181                                                 SwsContext *c, int16_t *data, \
 182                                                 int dstW, const uint8_t *src, \
 183                                                 const int16_t *filter, \
 184                                                 const int16_t *filterPos, int filterSize);
 185
 186 #define SCALE_FUNCS(filter_n, opt) \
 187     SCALE_FUNC(filter_n,  8, 15, opt); \
 188     SCALE_FUNC(filter_n,  9, 15, opt); \
 189     SCALE_FUNC(filter_n, 10, 15, opt); \
 190     SCALE_FUNC(filter_n, 14, 15, opt); \
 191     SCALE_FUNC(filter_n, 16, 15, opt); \
 192     SCALE_FUNC(filter_n,  8, 19, opt); \
 193     SCALE_FUNC(filter_n,  9, 19, opt); \
 194     SCALE_FUNC(filter_n, 10, 19, opt); \
 195     SCALE_FUNC(filter_n, 14, 19, opt); \
 196     SCALE_FUNC(filter_n, 16, 19, opt)
 197
 198 #define SCALE_FUNCS_MMX(opt) \
 199     SCALE_FUNCS(4, opt); \
 200     SCALE_FUNCS(8, opt); \
 201     SCALE_FUNCS(X, opt)
 202
 203 #define SCALE_FUNCS_SSE(opt) \
 204     SCALE_FUNCS(4, opt); \
 205     SCALE_FUNCS(8, opt); \
 206     SCALE_FUNCS(X4, opt); \
 207     SCALE_FUNCS(X8, opt)
 208
 209 #if ARCH_X86_32
 210 SCALE_FUNCS_MMX(mmx);
 211 #endif
 212 SCALE_FUNCS_SSE(sse2);
 213 SCALE_FUNCS_SSE(ssse3);
 214 SCALE_FUNCS_SSE(sse4);
 215
 216 #define VSCALEX_FUNC(size, opt) \
 217 extern void ff_yuv2planeX_ ## size ## _ ## opt(const int16_t *filter, int filterSize, \
 218                                                const int16_t **src, uint8_t *dest, int dstW, \
 219                                                const uint8_t *dither, int offset)
 220 #define VSCALEX_FUNCS(opt1, opt2) \
 221     VSCALEX_FUNC(8,  opt1); \
 222     VSCALEX_FUNC(9,  opt2); \
 223     VSCALEX_FUNC(10, opt2)
 224
 225 #if ARCH_X86_32
 226 VSCALEX_FUNCS(mmx,  mmx2);
 227 #endif
 228 VSCALEX_FUNCS(sse2, sse2);
 229 VSCALEX_FUNCS(sse4, sse4);
 230 VSCALEX_FUNC(16, sse4);
 231 VSCALEX_FUNCS(avx,  avx);
 232
 233 void ff_sws_init_swScale_mmx(SwsContext *c)
 234 {
 235     int cpu_flags = av_get_cpu_flags();
 236
 237     if (cpu_flags & AV_CPU_FLAG_MMX)
 238         sws_init_swScale_MMX(c);
 239 #if HAVE_MMX2
 240     if (cpu_flags & AV_CPU_FLAG_MMX2)
 241         sws_init_swScale_MMX2(c);
 242 #endif
 243
 244 #if HAVE_YASM
 245 #define ASSIGN_SCALE_FUNC2(hscalefn, filtersize, opt1, opt2) do { \
 246     if (c->srcBpc == 8) { \
 247         hscalefn = c->dstBpc <= 10 ? ff_hscale8to15_ ## filtersize ## _ ## opt2 : \
 248                                      ff_hscale8to19_ ## filtersize ## _ ## opt1; \
 249     } else if (c->srcBpc == 9) { \
 250         hscalefn = c->dstBpc <= 10 ? ff_hscale9to15_ ## filtersize ## _ ## opt2 : \
 251                                      ff_hscale9to19_ ## filtersize ## _ ## opt1; \
 252     } else if (c->srcBpc == 10) { \
 253         hscalefn = c->dstBpc <= 10 ? ff_hscale10to15_ ## filtersize ## _ ## opt2 : \
 254                                      ff_hscale10to19_ ## filtersize ## _ ## opt1; \
 255     } else if (c->srcBpc == 14 || ((c->srcFormat==PIX_FMT_PAL8||isAnyRGB(c->srcFormat)) && av_pix_fmt_descriptors[c->srcFormat].comp[0].depth_minus1<15)) { \
 256         hscalefn = c->dstBpc <= 10 ? ff_hscale14to15_ ## filtersize ## _ ## opt2 : \
 257                                      ff_hscale14to19_ ## filtersize ## _ ## opt1; \
 258     } else { /* c->srcBpc == 16 */ \
 259         hscalefn = c->dstBpc <= 10 ? ff_hscale16to15_ ## filtersize ## _ ## opt2 : \
 260                                      ff_hscale16to19_ ## filtersize ## _ ## opt1; \
 261     } \
 262 } while (0)
 263 #define ASSIGN_MMX_SCALE_FUNC(hscalefn, filtersize, opt1, opt2) \
 264     switch (filtersize) { \
 265     case 4:  ASSIGN_SCALE_FUNC2(hscalefn, 4, opt1, opt2); break; \
 266     case 8:  ASSIGN_SCALE_FUNC2(hscalefn, 8, opt1, opt2); break; \
 267     default: ASSIGN_SCALE_FUNC2(hscalefn, X, opt1, opt2); break; \
 268     }
 269 #define ASSIGN_VSCALEX_FUNC(vscalefn, opt1, opt2, opt2chk, do_16_case) \
 270 switch(c->dstBpc){ \
 271     case 16:                                     /*do_16_case;*/                           break; \
 272     case 10: if (!isBE(c->dstFormat) && opt2chk) /*vscalefn = ff_yuv2planeX_10_ ## opt2;*/ break; \
 273     case 9:  if (!isBE(c->dstFormat) && opt2chk) /*vscalefn = ff_yuv2planeX_9_  ## opt2;*/ break; \
 274     default:                                     /*vscalefn = ff_yuv2planeX_8_  ## opt1;*/ break; \
 275     }
 276 #if ARCH_X86_32
 277     if (cpu_flags & AV_CPU_FLAG_MMX) {
 278         ASSIGN_MMX_SCALE_FUNC(c->hyScale, c->hLumFilterSize, mmx, mmx);
 279         ASSIGN_MMX_SCALE_FUNC(c->hcScale, c->hChrFilterSize, mmx, mmx);
 280         ASSIGN_VSCALEX_FUNC(c->yuv2planeX, mmx, mmx2, cpu_flags & AV_CPU_FLAG_MMX2,);
 281     }
 282 #endif
 283 #define ASSIGN_SSE_SCALE_FUNC(hscalefn, filtersize, opt1, opt2) \
 284     switch (filtersize) { \
 285     case 4:  ASSIGN_SCALE_FUNC2(hscalefn, 4, opt1, opt2); break; \
 286     case 8:  ASSIGN_SCALE_FUNC2(hscalefn, 8, opt1, opt2); break; \
 287     default: if (filtersize & 4) ASSIGN_SCALE_FUNC2(hscalefn, X4, opt1, opt2); \
 288              else                ASSIGN_SCALE_FUNC2(hscalefn, X8, opt1, opt2); \
 289              break; \
 290     }
 291     if (cpu_flags & AV_CPU_FLAG_SSE2) {
 292         ASSIGN_SSE_SCALE_FUNC(c->hyScale, c->hLumFilterSize, sse2, sse2);
 293         ASSIGN_SSE_SCALE_FUNC(c->hcScale, c->hChrFilterSize, sse2, sse2);
 294         ASSIGN_VSCALEX_FUNC(c->yuv2planeX, sse2, sse2, 1,);
 295     }
 296     if (cpu_flags & AV_CPU_FLAG_SSSE3) {
 297         ASSIGN_SSE_SCALE_FUNC(c->hyScale, c->hLumFilterSize, ssse3, ssse3);
 298         ASSIGN_SSE_SCALE_FUNC(c->hcScale, c->hChrFilterSize, ssse3, ssse3);
 299     }
 300     if (cpu_flags & AV_CPU_FLAG_SSE4) {
 301         /* Xto15 don't need special sse4 functions */
 302         ASSIGN_SSE_SCALE_FUNC(c->hyScale, c->hLumFilterSize, sse4, ssse3);
 303         ASSIGN_SSE_SCALE_FUNC(c->hcScale, c->hChrFilterSize, sse4, ssse3);
 304         ASSIGN_VSCALEX_FUNC(c->yuv2planeX, sse4, sse4, 1,
 305                             if (!isBE(c->dstFormat)) c->yuv2planeX = ff_yuv2planeX_16_sse4);
 306     }
 307
 308     if (cpu_flags & AV_CPU_FLAG_AVX) {
 309         ASSIGN_VSCALEX_FUNC(c->yuv2planeX, avx, avx, 1,);
 310     }
 311 #endif
 312 }