libswscale/yuv2rgb_template.c

   1 /*
   2  * yuv2rgb_mmx.c, software YUV to RGB converter with Intel MMX "technology"
   3  *
   4  * Copyright (C) 2000, Silicon Integrated System Corp
   5  *
   6  * Author: Olie Lho <ollie@sis.com.tw>
   7  *
   8  * 15,24 bpp and dithering from Michael Niedermayer (michaelni@gmx.at)
   9  * MMX/MMX2 Template stuff from Michael Niedermayer (needed for fast movntq support)
  10  * context / deglobalize stuff by Michael Niedermayer
  11  *
  12  * This file is part of mpeg2dec, a free MPEG-2 video decoder
  13  *
  14  * mpeg2dec is free software; you can redistribute it and/or modify
  15  * it under the terms of the GNU General Public License as published by
  16  * the Free Software Foundation; either version 2, or (at your option)
  17  * any later version.
  18  *
  19  * mpeg2dec is distributed in the hope that it will be useful,
  20  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  21  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
  22  * GNU General Public License for more details.
  23  *
  24  * You should have received a copy of the GNU General Public License
  25  * along with mpeg2dec; if not, write to the Free Software
  26  * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
  27  */
  28
  29 #undef MOVNTQ
  30 #undef EMMS
  31 #undef SFENCE
  32
  33 #if HAVE_AMD3DNOW
  34 /* On K6 femms is faster than emms. On K7 femms is directly mapped to emms. */
  35 #define EMMS     "femms"
  36 #else
  37 #define EMMS     "emms"
  38 #endif
  39
  40 #if HAVE_MMX2
  41 #define MOVNTQ "movntq"
  42 #define SFENCE "sfence"
  43 #else
  44 #define MOVNTQ "movq"
  45 #define SFENCE "/nop"
  46 #endif
  47
  48 #define YUV2RGB \
  49     /* Do the multiply part of the conversion for even and odd pixels,
  50        register usage:
  51        mm0 -> Cblue, mm1 -> Cred, mm2 -> Cgreen even pixels,
  52        mm3 -> Cblue, mm4 -> Cred, mm5 -> Cgreen odd pixels,
  53        mm6 -> Y even, mm7 -> Y odd */\
  54     /* convert the chroma part */\
  55     "punpcklbw %%mm4, %%mm0;" /* scatter 4 Cb 00 u3 00 u2 00 u1 00 u0 */ \
  56     "punpcklbw %%mm4, %%mm1;" /* scatter 4 Cr 00 v3 00 v2 00 v1 00 v0 */ \
  57 \
  58     "psllw $3, %%mm0;" /* Promote precision */ \
  59     "psllw $3, %%mm1;" /* Promote precision */ \
  60 \
  61     "psubsw "U_OFFSET"(%4), %%mm0;" /* Cb -= 128 */ \
  62     "psubsw "V_OFFSET"(%4), %%mm1;" /* Cr -= 128 */ \
  63 \
  64     "movq %%mm0, %%mm2;" /* Copy 4 Cb 00 u3 00 u2 00 u1 00 u0 */ \
  65     "movq %%mm1, %%mm3;" /* Copy 4 Cr 00 v3 00 v2 00 v1 00 v0 */ \
  66 \
  67     "pmulhw "UG_COEFF"(%4), %%mm2;" /* Mul Cb with green coeff -> Cb green */ \
  68     "pmulhw "VG_COEFF"(%4), %%mm3;" /* Mul Cr with green coeff -> Cr green */ \
  69 \
  70     "pmulhw "UB_COEFF"(%4), %%mm0;" /* Mul Cb -> Cblue 00 b3 00 b2 00 b1 00 b0 */\
  71     "pmulhw "VR_COEFF"(%4), %%mm1;" /* Mul Cr -> Cred 00 r3 00 r2 00 r1 00 r0 */\
  72 \
  73     "paddsw %%mm3, %%mm2;" /* Cb green + Cr green -> Cgreen */\
  74 \
  75     /* convert the luma part */\
  76     "movq %%mm6, %%mm7;" /* Copy 8 Y Y7 Y6 Y5 Y4 Y3 Y2 Y1 Y0 */\
  77     "pand "MANGLE(mmx_00ffw)", %%mm6;" /* get Y even 00 Y6 00 Y4 00 Y2 00 Y0 */\
  78 \
  79     "psrlw $8, %%mm7;" /* get Y odd 00 Y7 00 Y5 00 Y3 00 Y1 */\
  80 \
  81     "psllw $3, %%mm6;" /* Promote precision */\
  82     "psllw $3, %%mm7;" /* Promote precision */\
  83 \
  84     "psubw "Y_OFFSET"(%4), %%mm6;" /* Y -= 16 */\
  85     "psubw "Y_OFFSET"(%4), %%mm7;" /* Y -= 16 */\
  86 \
  87     "pmulhw "Y_COEFF"(%4), %%mm6;" /* Mul 4 Y even 00 y6 00 y4 00 y2 00 y0 */\
  88     "pmulhw "Y_COEFF"(%4), %%mm7;" /* Mul 4 Y odd 00 y7 00 y5 00 y3 00 y1 */\
  89 \
  90     /* Do the addition part of the conversion for even and odd pixels,
  91        register usage:
  92        mm0 -> Cblue, mm1 -> Cred, mm2 -> Cgreen even pixels,
  93        mm3 -> Cblue, mm4 -> Cred, mm5 -> Cgreen odd pixels,
  94        mm6 -> Y even, mm7 -> Y odd */\
  95     "movq %%mm0, %%mm3;" /* Copy Cblue */\
  96     "movq %%mm1, %%mm4;" /* Copy Cred */\
  97     "movq %%mm2, %%mm5;" /* Copy Cgreen */\
  98 \
  99     "paddsw %%mm6, %%mm0;" /* Y even + Cblue 00 B6 00 B4 00 B2 00 B0 */\
 100     "paddsw %%mm7, %%mm3;" /* Y odd + Cblue 00 B7 00 B5 00 B3 00 B1 */\
 101 \
 102     "paddsw %%mm6, %%mm1;" /* Y even + Cred 00 R6 00 R4 00 R2 00 R0 */\
 103     "paddsw %%mm7, %%mm4;" /* Y odd + Cred 00 R7 00 R5 00 R3 00 R1 */\
 104 \
 105     "paddsw %%mm6, %%mm2;" /* Y even + Cgreen 00 G6 00 G4 00 G2 00 G0 */\
 106     "paddsw %%mm7, %%mm5;" /* Y odd + Cgreen 00 G7 00 G5 00 G3 00 G1 */\
 107 \
 108     /* Limit RGB even to 0..255 */\
 109     "packuswb %%mm0, %%mm0;" /* B6 B4 B2 B0  B6 B4 B2 B0 */\
 110     "packuswb %%mm1, %%mm1;" /* R6 R4 R2 R0  R6 R4 R2 R0 */\
 111     "packuswb %%mm2, %%mm2;" /* G6 G4 G2 G0  G6 G4 G2 G0 */\
 112 \
 113     /* Limit RGB odd to 0..255 */\
 114     "packuswb %%mm3, %%mm3;" /* B7 B5 B3 B1  B7 B5 B3 B1 */\
 115     "packuswb %%mm4, %%mm4;" /* R7 R5 R3 R1  R7 R5 R3 R1 */\
 116     "packuswb %%mm5, %%mm5;" /* G7 G5 G3 G1  G7 G5 G3 G1 */\
 117 \
 118     /* Interleave RGB even and odd */\
 119     "punpcklbw %%mm3, %%mm0;" /* B7 B6 B5 B4 B3 B2 B1 B0 */\
 120     "punpcklbw %%mm4, %%mm1;" /* R7 R6 R5 R4 R3 R2 R1 R0 */\
 121     "punpcklbw %%mm5, %%mm2;" /* G7 G6 G5 G4 G3 G2 G1 G0 */\
 122
 123
 124 static inline int RENAME(yuv420_rgb16)(SwsContext *c, uint8_t* src[], int srcStride[], int srcSliceY,
 125                                        int srcSliceH, uint8_t* dst[], int dstStride[]){
 126     int y, h_size;
 127
 128     if(c->srcFormat == PIX_FMT_YUV422P){
 129         srcStride[1] *= 2;
 130         srcStride[2] *= 2;
 131     }
 132
 133     h_size= (c->dstW+7)&~7;
 134     if(h_size*2 > FFABS(dstStride[0])) h_size-=8;
 135
 136     __asm__ volatile ("pxor %mm4, %mm4;" /* zero mm4 */ );
 137     //printf("%X %X %X %X %X %X %X %X %X %X\n", (int)&c->redDither, (int)&c->blueDither, (int)src[0], (int)src[1], (int)src[2], (int)dst[0],
 138     //srcStride[0],srcStride[1],srcStride[2],dstStride[0]);
 139     for (y= 0; y<srcSliceH; y++ ) {
 140         uint8_t *image = dst[0] + (y+srcSliceY)*dstStride[0];
 141         uint8_t *py = src[0] + y*srcStride[0];
 142         uint8_t *pu = src[1] + (y>>1)*srcStride[1];
 143         uint8_t *pv = src[2] + (y>>1)*srcStride[2];
 144         long index= -h_size/2;
 145
 146         c->blueDither= ff_dither8[y&1];
 147         c->greenDither= ff_dither4[y&1];
 148         c->redDither= ff_dither8[(y+1)&1];
 149         /* This MMX assembly code deals with a SINGLE scan line at a time,
 150          * it converts 8 pixels in each iteration. */
 151         __asm__ volatile (
 152         /* load data for start of next scan line */
 153         "movd    (%2, %0), %%mm0;" /* Load 4 Cb 00 00 00 00 u3 u2 u1 u0 */
 154         "movd    (%3, %0), %%mm1;" /* Load 4 Cr 00 00 00 00 v3 v2 v1 v0 */
 155         "movq (%5, %0, 2), %%mm6;" /* Load 8  Y Y7 Y6 Y5 Y4 Y3 Y2 Y1 Y0 */
 156         //".balign 16     \n\t"
 157         "1:             \n\t"
 158         /* No speed difference on my p3@500 with prefetch,
 159          * if it is faster for anyone with -benchmark then tell me.
 160         PREFETCH" 64(%0) \n\t"
 161         PREFETCH" 64(%1) \n\t"
 162         PREFETCH" 64(%2) \n\t"
 163         */
 164 YUV2RGB
 165
 166 #ifdef DITHER1XBPP
 167         "paddusb "BLUE_DITHER"(%4), %%mm0;"
 168         "paddusb "GREEN_DITHER"(%4), %%mm2;"
 169         "paddusb "RED_DITHER"(%4), %%mm1;"
 170 #endif
 171         /* mask unneeded bits off */
 172         "pand "MANGLE(mmx_redmask)", %%mm0;" /* b7b6b5b4 b3_0_0_0 b7b6b5b4 b3_0_0_0 */
 173         "pand "MANGLE(mmx_grnmask)", %%mm2;" /* g7g6g5g4 g3g2_0_0 g7g6g5g4 g3g2_0_0 */
 174         "pand "MANGLE(mmx_redmask)", %%mm1;" /* r7r6r5r4 r3_0_0_0 r7r6r5r4 r3_0_0_0 */
 175
 176         "psrlw   $3, %%mm0;" /* 0_0_0_b7 b6b5b4b3 0_0_0_b7 b6b5b4b3 */
 177         "pxor %%mm4, %%mm4;" /* zero mm4 */
 178
 179         "movq %%mm0, %%mm5;" /* Copy B7-B0 */
 180         "movq %%mm2, %%mm7;" /* Copy G7-G0 */
 181
 182         /* convert RGB24 plane to RGB16 pack for pixel 0-3 */
 183         "punpcklbw %%mm4, %%mm2;" /* 0_0_0_0 0_0_0_0 g7g6g5g4 g3g2_0_0 */
 184         "punpcklbw %%mm1, %%mm0;" /* r7r6r5r4 r3_0_0_0 0_0_0_b7 b6b5b4b3 */
 185
 186         "psllw  $3, %%mm2;" /* 0_0_0_0 0_g7g6g5 g4g3g2_0 0_0_0_0 */
 187         "por %%mm2, %%mm0;" /* r7r6r5r4 r3g7g6g5 g4g3g2b7 b6b5b4b3 */
 188
 189         "movq 8 (%5, %0, 2), %%mm6;" /* Load 8 Y Y7 Y6 Y5 Y4 Y3 Y2 Y1 Y0 */
 190         MOVNTQ "      %%mm0, (%1);" /* store pixel 0-3 */
 191
 192         /* convert RGB24 plane to RGB16 pack for pixel 0-3 */
 193         "punpckhbw %%mm4, %%mm7;" /* 0_0_0_0 0_0_0_0 g7g6g5g4 g3g2_0_0 */
 194         "punpckhbw %%mm1, %%mm5;" /* r7r6r5r4 r3_0_0_0 0_0_0_b7 b6b5b4b3 */
 195
 196         "psllw        $3, %%mm7;" /* 0_0_0_0 0_g7g6g5 g4g3g2_0 0_0_0_0 */
 197         "movd 4 (%2, %0), %%mm0;" /* Load 4 Cb 00 00 00 00 u3 u2 u1 u0 */
 198
 199         "por       %%mm7, %%mm5;" /* r7r6r5r4 r3g7g6g5 g4g3g2b7 b6b5b4b3 */
 200         "movd 4 (%3, %0), %%mm1;" /* Load 4 Cr 00 00 00 00 v3 v2 v1 v0 */
 201
 202         MOVNTQ "   %%mm5, 8 (%1);" /* store pixel 4-7 */
 203
 204         "add $16, %1    \n\t"
 205         "add  $4, %0    \n\t"
 206         " js  1b        \n\t"
 207
 208         : "+r" (index), "+r" (image)
 209         : "r" (pu - index), "r" (pv - index), "r"(&c->redDither), "r" (py - 2*index)
 210         );
 211     }
 212
 213     __asm__ volatile (EMMS);
 214
 215     return srcSliceH;
 216 }
 217
 218 static inline int RENAME(yuv420_rgb15)(SwsContext *c, uint8_t* src[], int srcStride[], int srcSliceY,
 219                                        int srcSliceH, uint8_t* dst[], int dstStride[]){
 220     int y, h_size;
 221
 222     if(c->srcFormat == PIX_FMT_YUV422P){
 223         srcStride[1] *= 2;
 224         srcStride[2] *= 2;
 225     }
 226
 227     h_size= (c->dstW+7)&~7;
 228     if(h_size*2 > FFABS(dstStride[0])) h_size-=8;
 229
 230     __asm__ volatile ("pxor %mm4, %mm4;" /* zero mm4 */ );
 231     //printf("%X %X %X %X %X %X %X %X %X %X\n", (int)&c->redDither, (int)&c->blueDither, (int)src[0], (int)src[1], (int)src[2], (int)dst[0],
 232     //srcStride[0],srcStride[1],srcStride[2],dstStride[0]);
 233     for (y= 0; y<srcSliceH; y++ ) {
 234         uint8_t *image = dst[0] + (y+srcSliceY)*dstStride[0];
 235         uint8_t *py = src[0] + y*srcStride[0];
 236         uint8_t *pu = src[1] + (y>>1)*srcStride[1];
 237         uint8_t *pv = src[2] + (y>>1)*srcStride[2];
 238         long index= -h_size/2;
 239
 240         c->blueDither= ff_dither8[y&1];
 241         c->greenDither= ff_dither8[y&1];
 242         c->redDither= ff_dither8[(y+1)&1];
 243         /* This MMX assembly code deals with a SINGLE scan line at a time,
 244          * it converts 8 pixels in each iteration. */
 245         __asm__ volatile (
 246         /* load data for start of next scan line */
 247         "movd    (%2, %0), %%mm0;" /* Load 4 Cb 00 00 00 00 u3 u2 u1 u0 */
 248         "movd    (%3, %0), %%mm1;" /* Load 4 Cr 00 00 00 00 v3 v2 v1 v0 */
 249         "movq (%5, %0, 2), %%mm6;" /* Load 8  Y Y7 Y6 Y5 Y4 Y3 Y2 Y1 Y0 */
 250         //".balign 16     \n\t"
 251         "1:             \n\t"
 252 YUV2RGB
 253
 254 #ifdef DITHER1XBPP
 255         "paddusb "BLUE_DITHER"(%4), %%mm0  \n\t"
 256         "paddusb "GREEN_DITHER"(%4), %%mm2  \n\t"
 257         "paddusb "RED_DITHER"(%4), %%mm1  \n\t"
 258 #endif
 259
 260         /* mask unneeded bits off */
 261         "pand "MANGLE(mmx_redmask)", %%mm0;" /* b7b6b5b4 b3_0_0_0 b7b6b5b4 b3_0_0_0 */
 262         "pand "MANGLE(mmx_redmask)", %%mm2;" /* g7g6g5g4 g3_0_0_0 g7g6g5g4 g3_0_0_0 */
 263         "pand "MANGLE(mmx_redmask)", %%mm1;" /* r7r6r5r4 r3_0_0_0 r7r6r5r4 r3_0_0_0 */
 264
 265         "psrlw   $3, %%mm0;" /* 0_0_0_b7 b6b5b4b3 0_0_0_b7 b6b5b4b3 */
 266         "psrlw   $1, %%mm1;" /* 0_r7r6r5  r4r3_0_0 0_r7r6r5 r4r3_0_0 */
 267         "pxor %%mm4, %%mm4;" /* zero mm4 */
 268
 269         "movq %%mm0, %%mm5;" /* Copy B7-B0 */
 270         "movq %%mm2, %%mm7;" /* Copy G7-G0 */
 271
 272         /* convert RGB24 plane to RGB16 pack for pixel 0-3 */
 273         "punpcklbw %%mm4, %%mm2;" /* 0_0_0_0 0_0_0_0 g7g6g5g4 g3_0_0_0 */
 274         "punpcklbw %%mm1, %%mm0;" /* r7r6r5r4 r3_0_0_0 0_0_0_b7 b6b5b4b3 */
 275
 276         "psllw  $2, %%mm2;" /* 0_0_0_0 0_0_g7g6 g5g4g3_0 0_0_0_0 */
 277         "por %%mm2, %%mm0;" /* 0_r7r6r5 r4r3g7g6 g5g4g3b7 b6b5b4b3 */
 278
 279         "movq 8 (%5, %0, 2), %%mm6;" /* Load 8 Y Y7 Y6 Y5 Y4 Y3 Y2 Y1 Y0 */
 280         MOVNTQ "      %%mm0, (%1);"  /* store pixel 0-3 */
 281
 282         /* convert RGB24 plane to RGB16 pack for pixel 0-3 */
 283         "punpckhbw %%mm4, %%mm7;" /* 0_0_0_0 0_0_0_0 0_g7g6g5 g4g3_0_0 */
 284         "punpckhbw %%mm1, %%mm5;" /* r7r6r5r4 r3_0_0_0 0_0_0_b7 b6b5b4b3 */
 285
 286         "psllw        $2, %%mm7;" /* 0_0_0_0 0_0_g7g6 g5g4g3_0 0_0_0_0 */
 287         "movd 4 (%2, %0), %%mm0;" /* Load 4 Cb 00 00 00 00 u3 u2 u1 u0 */
 288
 289         "por       %%mm7, %%mm5;" /* 0_r7r6r5 r4r3g7g6 g5g4g3b7 b6b5b4b3 */
 290         "movd 4 (%3, %0), %%mm1;" /* Load 4 Cr 00 00 00 00 v3 v2 v1 v0 */
 291
 292         MOVNTQ " %%mm5, 8 (%1);" /* store pixel 4-7 */
 293
 294         "add $16, %1            \n\t"
 295         "add $4, %0             \n\t"
 296         " js 1b                 \n\t"
 297         : "+r" (index), "+r" (image)
 298         : "r" (pu - index), "r" (pv - index), "r"(&c->redDither), "r" (py - 2*index)
 299         );
 300     }
 301
 302     __asm__ volatile (EMMS);
 303     return srcSliceH;
 304 }
 305
 306 static inline int RENAME(yuv420_rgb24)(SwsContext *c, uint8_t* src[], int srcStride[], int srcSliceY,
 307                                        int srcSliceH, uint8_t* dst[], int dstStride[]){
 308     int y, h_size;
 309
 310     if(c->srcFormat == PIX_FMT_YUV422P){
 311         srcStride[1] *= 2;
 312         srcStride[2] *= 2;
 313     }
 314
 315     h_size= (c->dstW+7)&~7;
 316     if(h_size*3 > FFABS(dstStride[0])) h_size-=8;
 317
 318     __asm__ volatile ("pxor %mm4, %mm4;" /* zero mm4 */ );
 319
 320     for (y= 0; y<srcSliceH; y++ ) {
 321         uint8_t *image = dst[0] + (y+srcSliceY)*dstStride[0];
 322         uint8_t *py = src[0] + y*srcStride[0];
 323         uint8_t *pu = src[1] + (y>>1)*srcStride[1];
 324         uint8_t *pv = src[2] + (y>>1)*srcStride[2];
 325         long index= -h_size/2;
 326
 327         /* This MMX assembly code deals with a SINGLE scan line at a time,
 328          * it converts 8 pixels in each iteration. */
 329         __asm__ volatile (
 330         /* load data for start of next scan line */
 331         "movd    (%2, %0), %%mm0;" /* Load 4 Cb 00 00 00 00 u3 u2 u1 u0 */
 332         "movd    (%3, %0), %%mm1;" /* Load 4 Cr 00 00 00 00 v3 v2 v1 v0 */
 333         "movq (%5, %0, 2), %%mm6;" /* Load 8  Y Y7 Y6 Y5 Y4 Y3 Y2 Y1 Y0 */
 334         //".balign 16     \n\t"
 335         "1:             \n\t"
 336 YUV2RGB
 337         /* mm0=B, %%mm2=G, %%mm1=R */
 338 #if HAVE_MMX2
 339         "movq "MANGLE(ff_M24A)", %%mm4     \n\t"
 340         "movq "MANGLE(ff_M24C)", %%mm7     \n\t"
 341         "pshufw $0x50, %%mm0, %%mm5     \n\t" /* B3 B2 B3 B2  B1 B0 B1 B0 */
 342         "pshufw $0x50, %%mm2, %%mm3     \n\t" /* G3 G2 G3 G2  G1 G0 G1 G0 */
 343         "pshufw $0x00, %%mm1, %%mm6     \n\t" /* R1 R0 R1 R0  R1 R0 R1 R0 */
 344
 345         "pand   %%mm4, %%mm5            \n\t" /*    B2        B1       B0 */
 346         "pand   %%mm4, %%mm3            \n\t" /*    G2        G1       G0 */
 347         "pand   %%mm7, %%mm6            \n\t" /*       R1        R0       */
 348
 349         "psllq     $8, %%mm3            \n\t" /* G2        G1       G0    */
 350         "por    %%mm5, %%mm6            \n\t"
 351         "por    %%mm3, %%mm6            \n\t"
 352         MOVNTQ" %%mm6, (%1)             \n\t"
 353
 354         "psrlq     $8, %%mm2            \n\t" /* 00 G7 G6 G5  G4 G3 G2 G1 */
 355         "pshufw $0xA5, %%mm0, %%mm5     \n\t" /* B5 B4 B5 B4  B3 B2 B3 B2 */
 356         "pshufw $0x55, %%mm2, %%mm3     \n\t" /* G4 G3 G4 G3  G4 G3 G4 G3 */
 357         "pshufw $0xA5, %%mm1, %%mm6     \n\t" /* R5 R4 R5 R4  R3 R2 R3 R2 */
 358
 359         "pand "MANGLE(ff_M24B)", %%mm5     \n\t" /* B5       B4        B3    */
 360         "pand          %%mm7, %%mm3     \n\t" /*       G4        G3       */
 361         "pand          %%mm4, %%mm6     \n\t" /*    R4        R3       R2 */
 362
 363         "por    %%mm5, %%mm3            \n\t" /* B5    G4 B4     G3 B3    */
 364         "por    %%mm3, %%mm6            \n\t"
 365         MOVNTQ" %%mm6, 8(%1)            \n\t"
 366
 367         "pshufw $0xFF, %%mm0, %%mm5     \n\t" /* B7 B6 B7 B6  B7 B6 B6 B7 */
 368         "pshufw $0xFA, %%mm2, %%mm3     \n\t" /* 00 G7 00 G7  G6 G5 G6 G5 */
 369         "pshufw $0xFA, %%mm1, %%mm6     \n\t" /* R7 R6 R7 R6  R5 R4 R5 R4 */
 370         "movd 4 (%2, %0), %%mm0;" /* Load 4 Cb 00 00 00 00 u3 u2 u1 u0 */
 371
 372         "pand          %%mm7, %%mm5     \n\t" /*       B7        B6       */
 373         "pand          %%mm4, %%mm3     \n\t" /*    G7        G6       G5 */
 374         "pand "MANGLE(ff_M24B)", %%mm6     \n\t" /* R7       R6        R5    */
 375         "movd 4 (%3, %0), %%mm1;" /* Load 4 Cr 00 00 00 00 v3 v2 v1 v0 */
 376 \
 377         "por          %%mm5, %%mm3      \n\t"
 378         "por          %%mm3, %%mm6      \n\t"
 379         MOVNTQ"       %%mm6, 16(%1)     \n\t"
 380         "movq 8 (%5, %0, 2), %%mm6;" /* Load 8 Y Y7 Y6 Y5 Y4 Y3 Y2 Y1 Y0 */
 381         "pxor         %%mm4, %%mm4      \n\t"
 382
 383 #else
 384
 385         "pxor      %%mm4, %%mm4     \n\t"
 386         "movq      %%mm0, %%mm5     \n\t" /* B */
 387         "movq      %%mm1, %%mm6     \n\t" /* R */
 388         "punpcklbw %%mm2, %%mm0     \n\t" /* GBGBGBGB 0 */
 389         "punpcklbw %%mm4, %%mm1     \n\t" /* 0R0R0R0R 0 */
 390         "punpckhbw %%mm2, %%mm5     \n\t" /* GBGBGBGB 2 */
 391         "punpckhbw %%mm4, %%mm6     \n\t" /* 0R0R0R0R 2 */
 392         "movq      %%mm0, %%mm7     \n\t" /* GBGBGBGB 0 */
 393         "movq      %%mm5, %%mm3     \n\t" /* GBGBGBGB 2 */
 394         "punpcklwd %%mm1, %%mm7     \n\t" /* 0RGB0RGB 0 */
 395         "punpckhwd %%mm1, %%mm0     \n\t" /* 0RGB0RGB 1 */
 396         "punpcklwd %%mm6, %%mm5     \n\t" /* 0RGB0RGB 2 */
 397         "punpckhwd %%mm6, %%mm3     \n\t" /* 0RGB0RGB 3 */
 398
 399         "movq      %%mm7, %%mm2     \n\t" /* 0RGB0RGB 0 */
 400         "movq      %%mm0, %%mm6     \n\t" /* 0RGB0RGB 1 */
 401         "movq      %%mm5, %%mm1     \n\t" /* 0RGB0RGB 2 */
 402         "movq      %%mm3, %%mm4     \n\t" /* 0RGB0RGB 3 */
 403
 404         "psllq       $40, %%mm7     \n\t" /* RGB00000 0 */
 405         "psllq       $40, %%mm0     \n\t" /* RGB00000 1 */
 406         "psllq       $40, %%mm5     \n\t" /* RGB00000 2 */
 407         "psllq       $40, %%mm3     \n\t" /* RGB00000 3 */
 408
 409         "punpckhdq %%mm2, %%mm7     \n\t" /* 0RGBRGB0 0 */
 410         "punpckhdq %%mm6, %%mm0     \n\t" /* 0RGBRGB0 1 */
 411         "punpckhdq %%mm1, %%mm5     \n\t" /* 0RGBRGB0 2 */
 412         "punpckhdq %%mm4, %%mm3     \n\t" /* 0RGBRGB0 3 */
 413
 414         "psrlq        $8, %%mm7     \n\t" /* 00RGBRGB 0 */
 415         "movq      %%mm0, %%mm6     \n\t" /* 0RGBRGB0 1 */
 416         "psllq       $40, %%mm0     \n\t" /* GB000000 1 */
 417         "por       %%mm0, %%mm7     \n\t" /* GBRGBRGB 0 */
 418         MOVNTQ"    %%mm7, (%1)      \n\t"
 419
 420         "movd 4 (%2, %0), %%mm0;" /* Load 4 Cb 00 00 00 00 u3 u2 u1 u0 */
 421
 422         "psrlq       $24, %%mm6     \n\t" /* 0000RGBR 1 */
 423         "movq      %%mm5, %%mm1     \n\t" /* 0RGBRGB0 2 */
 424         "psllq       $24, %%mm5     \n\t" /* BRGB0000 2 */
 425         "por       %%mm5, %%mm6     \n\t" /* BRGBRGBR 1 */
 426         MOVNTQ"    %%mm6, 8(%1)     \n\t"
 427
 428         "movq 8 (%5, %0, 2), %%mm6;" /* Load 8 Y Y7 Y6 Y5 Y4 Y3 Y2 Y1 Y0 */
 429
 430         "psrlq       $40, %%mm1     \n\t" /* 000000RG 2 */
 431         "psllq        $8, %%mm3     \n\t" /* RGBRGB00 3 */
 432         "por       %%mm3, %%mm1     \n\t" /* RGBRGBRG 2 */
 433         MOVNTQ"    %%mm1, 16(%1)    \n\t"
 434
 435         "movd 4 (%3, %0), %%mm1;" /* Load 4 Cr 00 00 00 00 v3 v2 v1 v0 */
 436         "pxor      %%mm4, %%mm4     \n\t"
 437 #endif
 438
 439         "add $24, %1    \n\t"
 440         "add  $4, %0    \n\t"
 441         " js  1b        \n\t"
 442
 443         : "+r" (index), "+r" (image)
 444         : "r" (pu - index), "r" (pv - index), "r"(&c->redDither), "r" (py - 2*index)
 445         );
 446     }
 447
 448     __asm__ volatile (EMMS);
 449     return srcSliceH;
 450 }
 451
 452 static inline int RENAME(yuv420_rgb32)(SwsContext *c, uint8_t* src[], int srcStride[], int srcSliceY,
 453                                        int srcSliceH, uint8_t* dst[], int dstStride[]){
 454     int y, h_size;
 455
 456     if(c->srcFormat == PIX_FMT_YUV422P){
 457         srcStride[1] *= 2;
 458         srcStride[2] *= 2;
 459     }
 460
 461     h_size= (c->dstW+7)&~7;
 462     if(h_size*4 > FFABS(dstStride[0])) h_size-=8;
 463
 464     __asm__ volatile ("pxor %mm4, %mm4;" /* zero mm4 */ );
 465
 466     for (y= 0; y<srcSliceH; y++ ) {
 467         uint8_t *image = dst[0] + (y+srcSliceY)*dstStride[0];
 468         uint8_t *py = src[0] + y*srcStride[0];
 469         uint8_t *pu = src[1] + (y>>1)*srcStride[1];
 470         uint8_t *pv = src[2] + (y>>1)*srcStride[2];
 471         long index= -h_size/2;
 472
 473         /* This MMX assembly code deals with a SINGLE scan line at a time,
 474          * it converts 8 pixels in each iteration. */
 475         __asm__ volatile (
 476         /* load data for start of next scan line */
 477         "movd    (%2, %0), %%mm0;" /* Load 4 Cb 00 00 00 00 u3 u2 u1 u0 */
 478         "movd    (%3, %0), %%mm1;" /* Load 4 Cr 00 00 00 00 v3 v2 v1 v0 */
 479         "movq (%5, %0, 2), %%mm6;" /* Load 8  Y Y7 Y6 Y5 Y4 Y3 Y2 Y1 Y0 */
 480         //".balign 16     \n\t"
 481         "1:             \n\t"
 482 YUV2RGB
 483         /* convert RGB plane to RGB packed format,
 484            mm0 ->  B, mm1 -> R, mm2 -> G, mm3 -> 0,
 485            mm4 -> GB, mm5 -> AR pixel 4-7,
 486            mm6 -> GB, mm7 -> AR pixel 0-3 */
 487         "pxor      %%mm3, %%mm3;"   /* zero mm3 */
 488
 489         "movq      %%mm0, %%mm6;"   /* B7 B6 B5 B4 B3 B2 B1 B0 */
 490         "movq      %%mm1, %%mm7;"   /* R7 R6 R5 R4 R3 R2 R1 R0 */
 491
 492         "movq      %%mm0, %%mm4;"   /* B7 B6 B5 B4 B3 B2 B1 B0 */
 493         "movq      %%mm1, %%mm5;"   /* R7 R6 R5 R4 R3 R2 R1 R0 */
 494
 495         "punpcklbw %%mm2, %%mm6;"   /* G3 B3 G2 B2 G1 B1 G0 B0 */
 496         "punpcklbw %%mm3, %%mm7;"   /* 00 R3 00 R2 00 R1 00 R0 */
 497
 498         "punpcklwd %%mm7, %%mm6;"   /* 00 R1 B1 G1 00 R0 B0 G0 */
 499         MOVNTQ "   %%mm6, (%1);"    /* Store ARGB1 ARGB0 */
 500
 501         "movq      %%mm0, %%mm6;"   /* B7 B6 B5 B4 B3 B2 B1 B0 */
 502         "punpcklbw %%mm2, %%mm6;"   /* G3 B3 G2 B2 G1 B1 G0 B0 */
 503
 504         "punpckhwd %%mm7, %%mm6;"   /* 00 R3 G3 B3 00 R2 B3 G2 */
 505         MOVNTQ "   %%mm6, 8 (%1);"  /* Store ARGB3 ARGB2 */
 506
 507         "punpckhbw %%mm2, %%mm4;"   /* G7 B7 G6 B6 G5 B5 G4 B4 */
 508         "punpckhbw %%mm3, %%mm5;"   /* 00 R7 00 R6 00 R5 00 R4 */
 509
 510         "punpcklwd %%mm5, %%mm4;"   /* 00 R5 B5 G5 00 R4 B4 G4 */
 511         MOVNTQ "   %%mm4, 16 (%1);" /* Store ARGB5 ARGB4 */
 512
 513         "movq      %%mm0, %%mm4;"   /* B7 B6 B5 B4 B3 B2 B1 B0 */
 514         "punpckhbw %%mm2, %%mm4;"   /* G7 B7 G6 B6 G5 B5 G4 B4 */
 515
 516         "punpckhwd %%mm5, %%mm4;"   /* 00 R7 G7 B7 00 R6 B6 G6 */
 517         MOVNTQ "   %%mm4, 24 (%1);" /* Store ARGB7 ARGB6 */
 518
 519         "movd 4 (%2, %0), %%mm0;"   /* Load 4 Cb 00 00 00 00 u3 u2 u1 u0 */
 520         "movd 4 (%3, %0), %%mm1;"   /* Load 4 Cr 00 00 00 00 v3 v2 v1 v0 */
 521
 522         "pxor         %%mm4, %%mm4;" /* zero mm4 */
 523         "movq 8 (%5, %0, 2), %%mm6;" /* Load 8 Y Y7 Y6 Y5 Y4 Y3 Y2 Y1 Y0 */
 524
 525         "add $32, %1    \n\t"
 526         "add  $4, %0    \n\t"
 527         " js  1b        \n\t"
 528
 529         : "+r" (index), "+r" (image)
 530         : "r" (pu - index), "r" (pv - index), "r"(&c->redDither), "r" (py - 2*index)
 531         );
 532     }
 533
 534     __asm__ volatile (EMMS);
 535     return srcSliceH;
 536 }