git.sesse.net Git - ffmpeg/blob - postproc/swscale.c

   1
   2 // Software scaling and colorspace conversion routines for MPlayer
   3
   4 // Orginal C implementation by A'rpi/ESP-team <arpi@thot.banki.hu>
   5 // current version mostly by Michael Niedermayer (michaelni@gmx.at)
   6 // the parts written by michael are under GNU GPL
   7
   8 #include <inttypes.h>
   9 #include <string.h>
  10 #include "../config.h"
  11 #include "swscale.h"
  12 #include "../mmx_defs.h"
  13 #undef MOVNTQ
  14
  15 //#undef HAVE_MMX2
  16 //#undef HAVE_MMX
  17 //#undef ARCH_X86
  18 #define DITHER1XBPP
  19 int fullUVIpol=0;
  20 //disables the unscaled height version
  21 int allwaysIpol=0;
  22
  23 #define RET 0xC3 //near return opcode
  24 /*
  25 NOTES
  26
  27 known BUGS with known cause (no bugreports please!, but patches are welcome :) )
  28 horizontal MMX2 scaler reads 1-7 samples too much (might cause a sig11)
  29
  30 Supported output formats BGR15 BGR16 BGR24 BGR32
  31 BGR15 & BGR16 MMX verions support dithering
  32 Special versions: fast Y 1:1 scaling (no interpolation in y direction)
  33
  34 TODO
  35 more intelligent missalignment avoidance for the horizontal scaler
  36 bicubic scaler
  37 dither in C
  38 change the distance of the u & v buffer
  39 */
  40
  41 #define ABS(a) ((a) > 0 ? (a) : (-(a)))
  42 #define MIN(a,b) ((a) > (b) ? (b) : (a))
  43 #define MAX(a,b) ((a) < (b) ? (b) : (a))
  44
  45 #ifdef HAVE_MMX2
  46 #define PAVGB(a,b) "pavgb " #a ", " #b " \n\t"
  47 #elif defined (HAVE_3DNOW)
  48 #define PAVGB(a,b) "pavgusb " #a ", " #b " \n\t"
  49 #endif
  50
  51 #ifdef HAVE_MMX2
  52 #define MOVNTQ(a,b) "movntq " #a ", " #b " \n\t"
  53 #else
  54 #define MOVNTQ(a,b) "movq " #a ", " #b " \n\t"
  55 #endif
  56
  57
  58 #ifdef HAVE_MMX
  59 static uint64_t __attribute__((aligned(8))) yCoeff=    0x2568256825682568LL;
  60 static uint64_t __attribute__((aligned(8))) vrCoeff=   0x3343334333433343LL;
  61 static uint64_t __attribute__((aligned(8))) ubCoeff=   0x40cf40cf40cf40cfLL;
  62 static uint64_t __attribute__((aligned(8))) vgCoeff=   0xE5E2E5E2E5E2E5E2LL;
  63 static uint64_t __attribute__((aligned(8))) ugCoeff=   0xF36EF36EF36EF36ELL;
  64 static uint64_t __attribute__((aligned(8))) bF8=       0xF8F8F8F8F8F8F8F8LL;
  65 static uint64_t __attribute__((aligned(8))) bFC=       0xFCFCFCFCFCFCFCFCLL;
  66 static uint64_t __attribute__((aligned(8))) w400=      0x0400040004000400LL;
  67 static uint64_t __attribute__((aligned(8))) w80=       0x0080008000800080LL;
  68 static uint64_t __attribute__((aligned(8))) w10=       0x0010001000100010LL;
  69 static uint64_t __attribute__((aligned(8))) bm00001111=0x00000000FFFFFFFFLL;
  70 static uint64_t __attribute__((aligned(8))) bm00000111=0x0000000000FFFFFFLL;
  71 static uint64_t __attribute__((aligned(8))) bm11111000=0xFFFFFFFFFF000000LL;
  72
  73 static uint64_t __attribute__((aligned(8))) b16Dither= 0x0004000400040004LL;
  74 static uint64_t __attribute__((aligned(8))) b16Dither1=0x0004000400040004LL;
  75 static uint64_t __attribute__((aligned(8))) b16Dither2=0x0602060206020602LL;
  76 static uint64_t __attribute__((aligned(8))) g16Dither= 0x0002000200020002LL;
  77 static uint64_t __attribute__((aligned(8))) g16Dither1=0x0002000200020002LL;
  78 static uint64_t __attribute__((aligned(8))) g16Dither2=0x0301030103010301LL;
  79
  80 static uint64_t __attribute__((aligned(8))) b16Mask=   0x001F001F001F001FLL;
  81 static uint64_t __attribute__((aligned(8))) g16Mask=   0x07E007E007E007E0LL;
  82 static uint64_t __attribute__((aligned(8))) r16Mask=   0xF800F800F800F800LL;
  83 static uint64_t __attribute__((aligned(8))) b15Mask=   0x001F001F001F001FLL;
  84 static uint64_t __attribute__((aligned(8))) g15Mask=   0x03E003E003E003E0LL;
  85 static uint64_t __attribute__((aligned(8))) r15Mask=   0x7C007C007C007C00LL;
  86
  87 static uint64_t __attribute__((aligned(8))) temp0;
  88 static uint64_t __attribute__((aligned(8))) asm_yalpha1;
  89 static uint64_t __attribute__((aligned(8))) asm_uvalpha1;
  90 #endif
  91
  92 // temporary storage for 4 yuv lines:
  93 // 16bit for now (mmx likes it more compact)
  94 #ifdef HAVE_MMX
  95 static uint16_t __attribute__((aligned(8))) pix_buf_y[4][2048];
  96 static uint16_t __attribute__((aligned(8))) pix_buf_uv[2][2048*2];
  97 #else
  98 static uint16_t pix_buf_y[4][2048];
  99 static uint16_t pix_buf_uv[2][2048*2];
 100 #endif
 101
 102 // clipping helper table for C implementations:
 103 static unsigned char clip_table[768];
 104
 105 static unsigned short clip_table16b[768];
 106 static unsigned short clip_table16g[768];
 107 static unsigned short clip_table16r[768];
 108 static unsigned short clip_table15b[768];
 109 static unsigned short clip_table15g[768];
 110 static unsigned short clip_table15r[768];
 111
 112 // yuv->rgb conversion tables:
 113 static    int yuvtab_2568[256];
 114 static    int yuvtab_3343[256];
 115 static    int yuvtab_0c92[256];
 116 static    int yuvtab_1a1e[256];
 117 static    int yuvtab_40cf[256];
 118
 119 #ifdef HAVE_MMX2
 120 static uint8_t funnyYCode[10000];
 121 static uint8_t funnyUVCode[10000];
 122 #endif
 123
 124 static int canMMX2BeUsed=0;
 125
 126 #define FULL_YSCALEYUV2RGB \
 127                 "pxor %%mm7, %%mm7              \n\t"\
 128                 "movd %6, %%mm6                 \n\t" /*yalpha1*/\
 129                 "punpcklwd %%mm6, %%mm6         \n\t"\
 130                 "punpcklwd %%mm6, %%mm6         \n\t"\
 131                 "movd %7, %%mm5                 \n\t" /*uvalpha1*/\
 132                 "punpcklwd %%mm5, %%mm5         \n\t"\
 133                 "punpcklwd %%mm5, %%mm5         \n\t"\
 134                 "xorl %%eax, %%eax              \n\t"\
 135                 "1:                             \n\t"\
 136                 "movq (%0, %%eax, 2), %%mm0     \n\t" /*buf0[eax]*/\
 137                 "movq (%1, %%eax, 2), %%mm1     \n\t" /*buf1[eax]*/\
 138                 "movq (%2, %%eax,2), %%mm2      \n\t" /* uvbuf0[eax]*/\
 139                 "movq (%3, %%eax,2), %%mm3      \n\t" /* uvbuf1[eax]*/\
 140                 "psubw %%mm1, %%mm0             \n\t" /* buf0[eax] - buf1[eax]*/\
 141                 "psubw %%mm3, %%mm2             \n\t" /* uvbuf0[eax] - uvbuf1[eax]*/\
 142                 "pmulhw %%mm6, %%mm0            \n\t" /* (buf0[eax] - buf1[eax])yalpha1>>16*/\
 143                 "pmulhw %%mm5, %%mm2            \n\t" /* (uvbuf0[eax] - uvbuf1[eax])uvalpha1>>16*/\
 144                 "psraw $4, %%mm1                \n\t" /* buf0[eax] - buf1[eax] >>4*/\
 145                 "movq 4096(%2, %%eax,2), %%mm4  \n\t" /* uvbuf0[eax+2048]*/\
 146                 "psraw $4, %%mm3                \n\t" /* uvbuf0[eax] - uvbuf1[eax] >>4*/\
 147                 "paddw %%mm0, %%mm1             \n\t" /* buf0[eax]yalpha1 + buf1[eax](1-yalpha1) >>16*/\
 148                 "movq 4096(%3, %%eax,2), %%mm0  \n\t" /* uvbuf1[eax+2048]*/\
 149                 "paddw %%mm2, %%mm3             \n\t" /* uvbuf0[eax]uvalpha1 - uvbuf1[eax](1-uvalpha1)*/\
 150                 "psubw %%mm0, %%mm4             \n\t" /* uvbuf0[eax+2048] - uvbuf1[eax+2048]*/\
 151                 "psubw w80, %%mm1               \n\t" /* 8(Y-16)*/\
 152                 "psubw w400, %%mm3              \n\t" /* 8(U-128)*/\
 153                 "pmulhw yCoeff, %%mm1           \n\t"\
 154 \
 155 \
 156                 "pmulhw %%mm5, %%mm4            \n\t" /* (uvbuf0[eax+2048] - uvbuf1[eax+2048])uvalpha1>>16*/\
 157                 "movq %%mm3, %%mm2              \n\t" /* (U-128)8*/\
 158                 "pmulhw ubCoeff, %%mm3          \n\t"\
 159                 "psraw $4, %%mm0                \n\t" /* uvbuf0[eax+2048] - uvbuf1[eax+2048] >>4*/\
 160                 "pmulhw ugCoeff, %%mm2          \n\t"\
 161                 "paddw %%mm4, %%mm0             \n\t" /* uvbuf0[eax+2048]uvalpha1 - uvbuf1[eax+2048](1-uvalpha1)*/\
 162                 "psubw w400, %%mm0              \n\t" /* (V-128)8*/\
 163 \
 164 \
 165                 "movq %%mm0, %%mm4              \n\t" /* (V-128)8*/\
 166                 "pmulhw vrCoeff, %%mm0          \n\t"\
 167                 "pmulhw vgCoeff, %%mm4          \n\t"\
 168                 "paddw %%mm1, %%mm3             \n\t" /* B*/\
 169                 "paddw %%mm1, %%mm0             \n\t" /* R*/\
 170                 "packuswb %%mm3, %%mm3          \n\t"\
 171 \
 172                 "packuswb %%mm0, %%mm0          \n\t"\
 173                 "paddw %%mm4, %%mm2             \n\t"\
 174                 "paddw %%mm2, %%mm1             \n\t" /* G*/\
 175 \
 176                 "packuswb %%mm1, %%mm1          \n\t"
 177
 178 #define YSCALEYUV2RGB \
 179                 "movd %6, %%mm6                 \n\t" /*yalpha1*/\
 180                 "punpcklwd %%mm6, %%mm6         \n\t"\
 181                 "punpcklwd %%mm6, %%mm6         \n\t"\
 182                 "movq %%mm6, asm_yalpha1        \n\t"\
 183                 "movd %7, %%mm5                 \n\t" /*uvalpha1*/\
 184                 "punpcklwd %%mm5, %%mm5         \n\t"\
 185                 "punpcklwd %%mm5, %%mm5         \n\t"\
 186                 "movq %%mm5, asm_uvalpha1       \n\t"\
 187                 "xorl %%eax, %%eax              \n\t"\
 188                 "1:                             \n\t"\
 189                 "movq (%2, %%eax), %%mm2        \n\t" /* uvbuf0[eax]*/\
 190                 "movq (%3, %%eax), %%mm3        \n\t" /* uvbuf1[eax]*/\
 191                 "movq 4096(%2, %%eax), %%mm5    \n\t" /* uvbuf0[eax+2048]*/\
 192                 "movq 4096(%3, %%eax), %%mm4    \n\t" /* uvbuf1[eax+2048]*/\
 193                 "psubw %%mm3, %%mm2             \n\t" /* uvbuf0[eax] - uvbuf1[eax]*/\
 194                 "psubw %%mm4, %%mm5             \n\t" /* uvbuf0[eax+2048] - uvbuf1[eax+2048]*/\
 195                 "movq asm_uvalpha1, %%mm0       \n\t"\
 196                 "pmulhw %%mm0, %%mm2            \n\t" /* (uvbuf0[eax] - uvbuf1[eax])uvalpha1>>16*/\
 197                 "pmulhw %%mm0, %%mm5            \n\t" /* (uvbuf0[eax+2048] - uvbuf1[eax+2048])uvalpha1>>16*/\
 198                 "psraw $4, %%mm3                \n\t" /* uvbuf0[eax] - uvbuf1[eax] >>4*/\
 199                 "psraw $4, %%mm4                \n\t" /* uvbuf0[eax+2048] - uvbuf1[eax+2048] >>4*/\
 200                 "paddw %%mm2, %%mm3             \n\t" /* uvbuf0[eax]uvalpha1 - uvbuf1[eax](1-uvalpha1)*/\
 201                 "paddw %%mm5, %%mm4             \n\t" /* uvbuf0[eax+2048]uvalpha1 - uvbuf1[eax+2048](1-uvalpha1)*/\
 202                 "psubw w400, %%mm3              \n\t" /* (U-128)8*/\
 203                 "psubw w400, %%mm4              \n\t" /* (V-128)8*/\
 204                 "movq %%mm3, %%mm2              \n\t" /* (U-128)8*/\
 205                 "movq %%mm4, %%mm5              \n\t" /* (V-128)8*/\
 206                 "pmulhw ugCoeff, %%mm3          \n\t"\
 207                 "pmulhw vgCoeff, %%mm4          \n\t"\
 208         /* mm2=(U-128)8, mm3=ug, mm4=vg mm5=(V-128)8 */\
 209                 "movq (%0, %%eax, 2), %%mm0     \n\t" /*buf0[eax]*/\
 210                 "movq (%1, %%eax, 2), %%mm1     \n\t" /*buf1[eax]*/\
 211                 "movq 8(%0, %%eax, 2), %%mm6    \n\t" /*buf0[eax]*/\
 212                 "movq 8(%1, %%eax, 2), %%mm7    \n\t" /*buf1[eax]*/\
 213                 "psubw %%mm1, %%mm0             \n\t" /* buf0[eax] - buf1[eax]*/\
 214                 "psubw %%mm7, %%mm6             \n\t" /* buf0[eax] - buf1[eax]*/\
 215                 "pmulhw asm_yalpha1, %%mm0      \n\t" /* (buf0[eax] - buf1[eax])yalpha1>>16*/\
 216                 "pmulhw asm_yalpha1, %%mm6      \n\t" /* (buf0[eax] - buf1[eax])yalpha1>>16*/\
 217                 "psraw $4, %%mm1                \n\t" /* buf0[eax] - buf1[eax] >>4*/\
 218                 "psraw $4, %%mm7                \n\t" /* buf0[eax] - buf1[eax] >>4*/\
 219                 "paddw %%mm0, %%mm1             \n\t" /* buf0[eax]yalpha1 + buf1[eax](1-yalpha1) >>16*/\
 220                 "paddw %%mm6, %%mm7             \n\t" /* buf0[eax]yalpha1 + buf1[eax](1-yalpha1) >>16*/\
 221                 "pmulhw ubCoeff, %%mm2          \n\t"\
 222                 "pmulhw vrCoeff, %%mm5          \n\t"\
 223                 "psubw w80, %%mm1               \n\t" /* 8(Y-16)*/\
 224                 "psubw w80, %%mm7               \n\t" /* 8(Y-16)*/\
 225                 "pmulhw yCoeff, %%mm1           \n\t"\
 226                 "pmulhw yCoeff, %%mm7           \n\t"\
 227         /* mm1= Y1, mm2=ub, mm3=ug, mm4=vg mm5=vr, mm7=Y2 */\
 228                 "paddw %%mm3, %%mm4             \n\t"\
 229                 "movq %%mm2, %%mm0              \n\t"\
 230                 "movq %%mm5, %%mm6              \n\t"\
 231                 "movq %%mm4, %%mm3              \n\t"\
 232                 "punpcklwd %%mm2, %%mm2         \n\t"\
 233                 "punpcklwd %%mm5, %%mm5         \n\t"\
 234                 "punpcklwd %%mm4, %%mm4         \n\t"\
 235                 "paddw %%mm1, %%mm2             \n\t"\
 236                 "paddw %%mm1, %%mm5             \n\t"\
 237                 "paddw %%mm1, %%mm4             \n\t"\
 238                 "punpckhwd %%mm0, %%mm0         \n\t"\
 239                 "punpckhwd %%mm6, %%mm6         \n\t"\
 240                 "punpckhwd %%mm3, %%mm3         \n\t"\
 241                 "paddw %%mm7, %%mm0             \n\t"\
 242                 "paddw %%mm7, %%mm6             \n\t"\
 243                 "paddw %%mm7, %%mm3             \n\t"\
 244                 /* mm0=B1, mm2=B2, mm3=G2, mm4=G1, mm5=R1, mm6=R2 */\
 245                 "packuswb %%mm0, %%mm2          \n\t"\
 246                 "packuswb %%mm6, %%mm5          \n\t"\
 247                 "packuswb %%mm3, %%mm4          \n\t"\
 248                 "pxor %%mm7, %%mm7              \n\t"
 249
 250 #define YSCALEYUV2RGB1 \
 251                 "xorl %%eax, %%eax              \n\t"\
 252                 "1:                             \n\t"\
 253                 "movq (%2, %%eax), %%mm3        \n\t" /* uvbuf0[eax]*/\
 254                 "movq 4096(%2, %%eax), %%mm4    \n\t" /* uvbuf0[eax+2048]*/\
 255                 "psraw $4, %%mm3                \n\t" /* uvbuf0[eax] - uvbuf1[eax] >>4*/\
 256                 "psraw $4, %%mm4                \n\t" /* uvbuf0[eax+2048] - uvbuf1[eax+2048] >>4*/\
 257                 "psubw w400, %%mm3              \n\t" /* (U-128)8*/\
 258                 "psubw w400, %%mm4              \n\t" /* (V-128)8*/\
 259                 "movq %%mm3, %%mm2              \n\t" /* (U-128)8*/\
 260                 "movq %%mm4, %%mm5              \n\t" /* (V-128)8*/\
 261                 "pmulhw ugCoeff, %%mm3          \n\t"\
 262                 "pmulhw vgCoeff, %%mm4          \n\t"\
 263         /* mm2=(U-128)8, mm3=ug, mm4=vg mm5=(V-128)8 */\
 264                 "movq (%0, %%eax, 2), %%mm1     \n\t" /*buf0[eax]*/\
 265                 "movq 8(%0, %%eax, 2), %%mm7    \n\t" /*buf0[eax]*/\
 266                 "psraw $4, %%mm1                \n\t" /* buf0[eax] - buf1[eax] >>4*/\
 267                 "psraw $4, %%mm7                \n\t" /* buf0[eax] - buf1[eax] >>4*/\
 268                 "pmulhw ubCoeff, %%mm2          \n\t"\
 269                 "pmulhw vrCoeff, %%mm5          \n\t"\
 270                 "psubw w80, %%mm1               \n\t" /* 8(Y-16)*/\
 271                 "psubw w80, %%mm7               \n\t" /* 8(Y-16)*/\
 272                 "pmulhw yCoeff, %%mm1           \n\t"\
 273                 "pmulhw yCoeff, %%mm7           \n\t"\
 274         /* mm1= Y1, mm2=ub, mm3=ug, mm4=vg mm5=vr, mm7=Y2 */\
 275                 "paddw %%mm3, %%mm4             \n\t"\
 276                 "movq %%mm2, %%mm0              \n\t"\
 277                 "movq %%mm5, %%mm6              \n\t"\
 278                 "movq %%mm4, %%mm3              \n\t"\
 279                 "punpcklwd %%mm2, %%mm2         \n\t"\
 280                 "punpcklwd %%mm5, %%mm5         \n\t"\
 281                 "punpcklwd %%mm4, %%mm4         \n\t"\
 282                 "paddw %%mm1, %%mm2             \n\t"\
 283                 "paddw %%mm1, %%mm5             \n\t"\
 284                 "paddw %%mm1, %%mm4             \n\t"\
 285                 "punpckhwd %%mm0, %%mm0         \n\t"\
 286                 "punpckhwd %%mm6, %%mm6         \n\t"\
 287                 "punpckhwd %%mm3, %%mm3         \n\t"\
 288                 "paddw %%mm7, %%mm0             \n\t"\
 289                 "paddw %%mm7, %%mm6             \n\t"\
 290                 "paddw %%mm7, %%mm3             \n\t"\
 291                 /* mm0=B1, mm2=B2, mm3=G2, mm4=G1, mm5=R1, mm6=R2 */\
 292                 "packuswb %%mm0, %%mm2          \n\t"\
 293                 "packuswb %%mm6, %%mm5          \n\t"\
 294                 "packuswb %%mm3, %%mm4          \n\t"\
 295                 "pxor %%mm7, %%mm7              \n\t"
 296
 297 // do vertical chrominance interpolation
 298 #define YSCALEYUV2RGB1b \
 299                 "xorl %%eax, %%eax              \n\t"\
 300                 "1:                             \n\t"\
 301                 "movq (%2, %%eax), %%mm2        \n\t" /* uvbuf0[eax]*/\
 302                 "movq (%3, %%eax), %%mm3        \n\t" /* uvbuf1[eax]*/\
 303                 "movq 4096(%2, %%eax), %%mm5    \n\t" /* uvbuf0[eax+2048]*/\
 304                 "movq 4096(%3, %%eax), %%mm4    \n\t" /* uvbuf1[eax+2048]*/\
 305                 "paddw %%mm2, %%mm3             \n\t" /* uvbuf0[eax] + uvbuf1[eax]*/\
 306                 "paddw %%mm5, %%mm4             \n\t" /* uvbuf0[eax+2048] + uvbuf1[eax+2048]*/\
 307                 "psrlw $5, %%mm3                \n\t"\
 308                 "psrlw $5, %%mm4                \n\t"\
 309                 "psubw w400, %%mm3              \n\t" /* (U-128)8*/\
 310                 "psubw w400, %%mm4              \n\t" /* (V-128)8*/\
 311                 "movq %%mm3, %%mm2              \n\t" /* (U-128)8*/\
 312                 "movq %%mm4, %%mm5              \n\t" /* (V-128)8*/\
 313                 "pmulhw ugCoeff, %%mm3          \n\t"\
 314                 "pmulhw vgCoeff, %%mm4          \n\t"\
 315         /* mm2=(U-128)8, mm3=ug, mm4=vg mm5=(V-128)8 */\
 316                 "movq (%0, %%eax, 2), %%mm1     \n\t" /*buf0[eax]*/\
 317                 "movq 8(%0, %%eax, 2), %%mm7    \n\t" /*buf0[eax]*/\
 318                 "psraw $4, %%mm1                \n\t" /* buf0[eax] - buf1[eax] >>4*/\
 319                 "psraw $4, %%mm7                \n\t" /* buf0[eax] - buf1[eax] >>4*/\
 320                 "pmulhw ubCoeff, %%mm2          \n\t"\
 321                 "pmulhw vrCoeff, %%mm5          \n\t"\
 322                 "psubw w80, %%mm1               \n\t" /* 8(Y-16)*/\
 323                 "psubw w80, %%mm7               \n\t" /* 8(Y-16)*/\
 324                 "pmulhw yCoeff, %%mm1           \n\t"\
 325                 "pmulhw yCoeff, %%mm7           \n\t"\
 326         /* mm1= Y1, mm2=ub, mm3=ug, mm4=vg mm5=vr, mm7=Y2 */\
 327                 "paddw %%mm3, %%mm4             \n\t"\
 328                 "movq %%mm2, %%mm0              \n\t"\
 329                 "movq %%mm5, %%mm6              \n\t"\
 330                 "movq %%mm4, %%mm3              \n\t"\
 331                 "punpcklwd %%mm2, %%mm2         \n\t"\
 332                 "punpcklwd %%mm5, %%mm5         \n\t"\
 333                 "punpcklwd %%mm4, %%mm4         \n\t"\
 334                 "paddw %%mm1, %%mm2             \n\t"\
 335                 "paddw %%mm1, %%mm5             \n\t"\
 336                 "paddw %%mm1, %%mm4             \n\t"\
 337                 "punpckhwd %%mm0, %%mm0         \n\t"\
 338                 "punpckhwd %%mm6, %%mm6         \n\t"\
 339                 "punpckhwd %%mm3, %%mm3         \n\t"\
 340                 "paddw %%mm7, %%mm0             \n\t"\
 341                 "paddw %%mm7, %%mm6             \n\t"\
 342                 "paddw %%mm7, %%mm3             \n\t"\
 343                 /* mm0=B1, mm2=B2, mm3=G2, mm4=G1, mm5=R1, mm6=R2 */\
 344                 "packuswb %%mm0, %%mm2          \n\t"\
 345                 "packuswb %%mm6, %%mm5          \n\t"\
 346                 "packuswb %%mm3, %%mm4          \n\t"\
 347                 "pxor %%mm7, %%mm7              \n\t"
 348
 349 #define WRITEBGR32 \
 350                 /* mm2=B, %%mm4=G, %%mm5=R, %%mm7=0 */\
 351                         "movq %%mm2, %%mm1              \n\t" /* B */\
 352                         "movq %%mm5, %%mm6              \n\t" /* R */\
 353                         "punpcklbw %%mm4, %%mm2         \n\t" /* GBGBGBGB 0 */\
 354                         "punpcklbw %%mm7, %%mm5         \n\t" /* 0R0R0R0R 0 */\
 355                         "punpckhbw %%mm4, %%mm1         \n\t" /* GBGBGBGB 2 */\
 356                         "punpckhbw %%mm7, %%mm6         \n\t" /* 0R0R0R0R 2 */\
 357                         "movq %%mm2, %%mm0              \n\t" /* GBGBGBGB 0 */\
 358                         "movq %%mm1, %%mm3              \n\t" /* GBGBGBGB 2 */\
 359                         "punpcklwd %%mm5, %%mm0         \n\t" /* 0RGB0RGB 0 */\
 360                         "punpckhwd %%mm5, %%mm2         \n\t" /* 0RGB0RGB 1 */\
 361                         "punpcklwd %%mm6, %%mm1         \n\t" /* 0RGB0RGB 2 */\
 362                         "punpckhwd %%mm6, %%mm3         \n\t" /* 0RGB0RGB 3 */\
 363 \
 364                         MOVNTQ(%%mm0, (%4, %%eax, 4))\
 365                         MOVNTQ(%%mm2, 8(%4, %%eax, 4))\
 366                         MOVNTQ(%%mm1, 16(%4, %%eax, 4))\
 367                         MOVNTQ(%%mm3, 24(%4, %%eax, 4))\
 368 \
 369                         "addl $8, %%eax                 \n\t"\
 370                         "cmpl %5, %%eax                 \n\t"\
 371                         " jb 1b                         \n\t"
 372
 373 #define WRITEBGR16 \
 374                         "pand bF8, %%mm2                \n\t" /* B */\
 375                         "pand bFC, %%mm4                \n\t" /* G */\
 376                         "pand bF8, %%mm5                \n\t" /* R */\
 377                         "psrlq $3, %%mm2                \n\t"\
 378 \
 379                         "movq %%mm2, %%mm1              \n\t"\
 380                         "movq %%mm4, %%mm3              \n\t"\
 381 \
 382                         "punpcklbw %%mm7, %%mm3         \n\t"\
 383                         "punpcklbw %%mm5, %%mm2         \n\t"\
 384                         "punpckhbw %%mm7, %%mm4         \n\t"\
 385                         "punpckhbw %%mm5, %%mm1         \n\t"\
 386 \
 387                         "psllq $3, %%mm3                \n\t"\
 388                         "psllq $3, %%mm4                \n\t"\
 389 \
 390                         "por %%mm3, %%mm2               \n\t"\
 391                         "por %%mm4, %%mm1               \n\t"\
 392 \
 393                         MOVNTQ(%%mm2, (%4, %%eax, 2))\
 394                         MOVNTQ(%%mm1, 8(%4, %%eax, 2))\
 395 \
 396                         "addl $8, %%eax                 \n\t"\
 397                         "cmpl %5, %%eax                 \n\t"\
 398                         " jb 1b                         \n\t"
 399
 400 #define WRITEBGR15 \
 401                         "pand bF8, %%mm2                \n\t" /* B */\
 402                         "pand bF8, %%mm4                \n\t" /* G */\
 403                         "pand bF8, %%mm5                \n\t" /* R */\
 404                         "psrlq $3, %%mm2                \n\t"\
 405                         "psrlq $1, %%mm5                \n\t"\
 406 \
 407                         "movq %%mm2, %%mm1              \n\t"\
 408                         "movq %%mm4, %%mm3              \n\t"\
 409 \
 410                         "punpcklbw %%mm7, %%mm3         \n\t"\
 411                         "punpcklbw %%mm5, %%mm2         \n\t"\
 412                         "punpckhbw %%mm7, %%mm4         \n\t"\
 413                         "punpckhbw %%mm5, %%mm1         \n\t"\
 414 \
 415                         "psllq $2, %%mm3                \n\t"\
 416                         "psllq $2, %%mm4                \n\t"\
 417 \
 418                         "por %%mm3, %%mm2               \n\t"\
 419                         "por %%mm4, %%mm1               \n\t"\
 420 \
 421                         MOVNTQ(%%mm2, (%4, %%eax, 2))\
 422                         MOVNTQ(%%mm1, 8(%4, %%eax, 2))\
 423 \
 424                         "addl $8, %%eax                 \n\t"\
 425                         "cmpl %5, %%eax                 \n\t"\
 426                         " jb 1b                         \n\t"
 427
 428 // FIXME find a faster way to shuffle it to BGR24
 429 #define WRITEBGR24 \
 430                 /* mm2=B, %%mm4=G, %%mm5=R, %%mm7=0 */\
 431                         "movq %%mm2, %%mm1              \n\t" /* B */\
 432                         "movq %%mm5, %%mm6              \n\t" /* R */\
 433                         "punpcklbw %%mm4, %%mm2         \n\t" /* GBGBGBGB 0 */\
 434                         "punpcklbw %%mm7, %%mm5         \n\t" /* 0R0R0R0R 0 */\
 435                         "punpckhbw %%mm4, %%mm1         \n\t" /* GBGBGBGB 2 */\
 436                         "punpckhbw %%mm7, %%mm6         \n\t" /* 0R0R0R0R 2 */\
 437                         "movq %%mm2, %%mm0              \n\t" /* GBGBGBGB 0 */\
 438                         "movq %%mm1, %%mm3              \n\t" /* GBGBGBGB 2 */\
 439                         "punpcklwd %%mm5, %%mm0         \n\t" /* 0RGB0RGB 0 */\
 440                         "punpckhwd %%mm5, %%mm2         \n\t" /* 0RGB0RGB 1 */\
 441                         "punpcklwd %%mm6, %%mm1         \n\t" /* 0RGB0RGB 2 */\
 442                         "punpckhwd %%mm6, %%mm3         \n\t" /* 0RGB0RGB 3 */\
 443 \
 444                         "movq %%mm0, %%mm4              \n\t" /* 0RGB0RGB 0 */\
 445                         "psrlq $8, %%mm0                \n\t" /* 00RGB0RG 0 */\
 446                         "pand bm00000111, %%mm4         \n\t" /* 00000RGB 0 */\
 447                         "pand bm11111000, %%mm0         \n\t" /* 00RGB000 0.5 */\
 448                         "por %%mm4, %%mm0               \n\t" /* 00RGBRGB 0 */\
 449                         "movq %%mm2, %%mm4              \n\t" /* 0RGB0RGB 1 */\
 450                         "psllq $48, %%mm2               \n\t" /* GB000000 1 */\
 451                         "por %%mm2, %%mm0               \n\t" /* GBRGBRGB 0 */\
 452 \
 453                         "movq %%mm4, %%mm2              \n\t" /* 0RGB0RGB 1 */\
 454                         "psrld $16, %%mm4               \n\t" /* 000R000R 1 */\
 455                         "psrlq $24, %%mm2               \n\t" /* 0000RGB0 1.5 */\
 456                         "por %%mm4, %%mm2               \n\t" /* 000RRGBR 1 */\
 457                         "pand bm00001111, %%mm2         \n\t" /* 0000RGBR 1 */\
 458                         "movq %%mm1, %%mm4              \n\t" /* 0RGB0RGB 2 */\
 459                         "psrlq $8, %%mm1                \n\t" /* 00RGB0RG 2 */\
 460                         "pand bm00000111, %%mm4         \n\t" /* 00000RGB 2 */\
 461                         "pand bm11111000, %%mm1         \n\t" /* 00RGB000 2.5 */\
 462                         "por %%mm4, %%mm1               \n\t" /* 00RGBRGB 2 */\
 463                         "movq %%mm1, %%mm4              \n\t" /* 00RGBRGB 2 */\
 464                         "psllq $32, %%mm1               \n\t" /* BRGB0000 2 */\
 465                         "por %%mm1, %%mm2               \n\t" /* BRGBRGBR 1 */\
 466 \
 467                         "psrlq $32, %%mm4               \n\t" /* 000000RG 2.5 */\
 468                         "movq %%mm3, %%mm5              \n\t" /* 0RGB0RGB 3 */\
 469                         "psrlq $8, %%mm3                \n\t" /* 00RGB0RG 3 */\
 470                         "pand bm00000111, %%mm5         \n\t" /* 00000RGB 3 */\
 471                         "pand bm11111000, %%mm3         \n\t" /* 00RGB000 3.5 */\
 472                         "por %%mm5, %%mm3               \n\t" /* 00RGBRGB 3 */\
 473                         "psllq $16, %%mm3               \n\t" /* RGBRGB00 3 */\
 474                         "por %%mm4, %%mm3               \n\t" /* RGBRGBRG 2.5 */\
 475 \
 476                         "leal (%%eax, %%eax, 2), %%ebx  \n\t"\
 477                         MOVNTQ(%%mm0, (%4, %%ebx))\
 478                         MOVNTQ(%%mm2, 8(%4, %%ebx))\
 479                         MOVNTQ(%%mm3, 16(%4, %%ebx))\
 480 \
 481                         "addl $8, %%eax                 \n\t"\
 482                         "cmpl %5, %%eax                 \n\t"\
 483                         " jb 1b                         \n\t"
 484
 485 #ifdef HAVE_MMX
 486 void in_asm_used_var_warning_killer()
 487 {
 488  int i= yCoeff+vrCoeff+ubCoeff+vgCoeff+ugCoeff+bF8+bFC+w400+w80+w10+
 489  bm00001111+bm00000111+bm11111000+b16Dither+b16Dither1+b16Dither2+g16Dither+g16Dither1+
 490  g16Dither2+b16Mask+g16Mask+r16Mask+b15Mask+g15Mask+r15Mask+temp0+asm_yalpha1+ asm_uvalpha1;
 491  if(i) i=0;
 492 }
 493 #endif
 494
 495 static inline void yuv2yuv(uint16_t *buf0, uint16_t *buf1, uint16_t *uvbuf0, uint16_t *uvbuf1,
 496                            uint8_t *dest, uint8_t *uDest, uint8_t *vDest, int dstw, int yalpha, int uvalpha)
 497 {
 498         int yalpha1=yalpha^4095;
 499         int uvalpha1=uvalpha^4095;
 500         int i;
 501
 502         asm volatile ("\n\t"::: "memory");
 503
 504         for(i=0;i<dstw;i++)
 505         {
 506                 ((uint8_t*)dest)[i] = (buf0[i]*yalpha1+buf1[i]*yalpha)>>19;
 507         }
 508
 509         if(uvalpha != -1)
 510         {
 511                 for(i=0; i<(dstw>>1); i++)
 512                 {
 513                         ((uint8_t*)uDest)[i] = (uvbuf0[i]*uvalpha1+uvbuf1[i]*uvalpha)>>19;
 514                         ((uint8_t*)vDest)[i] = (uvbuf0[i+2048]*uvalpha1+uvbuf1[i+2048]*uvalpha)>>19;
 515                 }
 516         }
 517 }
 518
 519 /**
 520  * vertical scale YV12 to RGB
 521  */
 522 static inline void yuv2rgbX(uint16_t *buf0, uint16_t *buf1, uint16_t *uvbuf0, uint16_t *uvbuf1,
 523                             uint8_t *dest, int dstw, int yalpha, int uvalpha, int dstbpp)
 524 {
 525         int yalpha1=yalpha^4095;
 526         int uvalpha1=uvalpha^4095;
 527
 528         if(fullUVIpol)
 529         {
 530
 531 #ifdef HAVE_MMX
 532                 if(dstbpp == 32)
 533                 {
 534                         asm volatile(
 535
 536
 537 FULL_YSCALEYUV2RGB
 538                         "punpcklbw %%mm1, %%mm3         \n\t" // BGBGBGBG
 539                         "punpcklbw %%mm7, %%mm0         \n\t" // R0R0R0R0
 540
 541                         "movq %%mm3, %%mm1              \n\t"
 542                         "punpcklwd %%mm0, %%mm3         \n\t" // BGR0BGR0
 543                         "punpckhwd %%mm0, %%mm1         \n\t" // BGR0BGR0
 544
 545                         MOVNTQ(%%mm3, (%4, %%eax, 4))
 546                         MOVNTQ(%%mm1, 8(%4, %%eax, 4))
 547
 548                         "addl $4, %%eax                 \n\t"
 549                         "cmpl %5, %%eax                 \n\t"
 550                         " jb 1b                         \n\t"
 551
 552
 553                         :: "r" (buf0), "r" (buf1), "r" (uvbuf0), "r" (uvbuf1), "r" (dest), "m" (dstw),
 554                         "m" (yalpha1), "m" (uvalpha1)
 555                         : "%eax"
 556                         );
 557                 }
 558                 else if(dstbpp==24)
 559                 {
 560                         asm volatile(
 561
 562 FULL_YSCALEYUV2RGB
 563
 564                                                                 // lsb ... msb
 565                         "punpcklbw %%mm1, %%mm3         \n\t" // BGBGBGBG
 566                         "punpcklbw %%mm7, %%mm0         \n\t" // R0R0R0R0
 567
 568                         "movq %%mm3, %%mm1              \n\t"
 569                         "punpcklwd %%mm0, %%mm3         \n\t" // BGR0BGR0
 570                         "punpckhwd %%mm0, %%mm1         \n\t" // BGR0BGR0
 571
 572                         "movq %%mm3, %%mm2              \n\t" // BGR0BGR0
 573                         "psrlq $8, %%mm3                \n\t" // GR0BGR00
 574                         "pand bm00000111, %%mm2         \n\t" // BGR00000
 575                         "pand bm11111000, %%mm3         \n\t" // 000BGR00
 576                         "por %%mm2, %%mm3               \n\t" // BGRBGR00
 577                         "movq %%mm1, %%mm2              \n\t"
 578                         "psllq $48, %%mm1               \n\t" // 000000BG
 579                         "por %%mm1, %%mm3               \n\t" // BGRBGRBG
 580
 581                         "movq %%mm2, %%mm1              \n\t" // BGR0BGR0
 582                         "psrld $16, %%mm2               \n\t" // R000R000
 583                         "psrlq $24, %%mm1               \n\t" // 0BGR0000
 584                         "por %%mm2, %%mm1               \n\t" // RBGRR000
 585
 586                         "movl %4, %%ebx                 \n\t"
 587                         "addl %%eax, %%ebx              \n\t"
 588
 589 #ifdef HAVE_MMX2
 590                         //FIXME Alignment
 591                         "movntq %%mm3, (%%ebx, %%eax, 2)\n\t"
 592                         "movntq %%mm1, 8(%%ebx, %%eax, 2)\n\t"
 593 #else
 594                         "movd %%mm3, (%%ebx, %%eax, 2)  \n\t"
 595                         "psrlq $32, %%mm3               \n\t"
 596                         "movd %%mm3, 4(%%ebx, %%eax, 2) \n\t"
 597                         "movd %%mm1, 8(%%ebx, %%eax, 2) \n\t"
 598 #endif
 599                         "addl $4, %%eax                 \n\t"
 600                         "cmpl %5, %%eax                 \n\t"
 601                         " jb 1b                         \n\t"
 602
 603                         :: "r" (buf0), "r" (buf1), "r" (uvbuf0), "r" (uvbuf1), "m" (dest), "m" (dstw),
 604                         "m" (yalpha1), "m" (uvalpha1)
 605                         : "%eax", "%ebx"
 606                         );
 607                 }
 608                 else if(dstbpp==15)
 609                 {
 610                         asm volatile(
 611
 612 FULL_YSCALEYUV2RGB
 613 #ifdef DITHER1XBPP
 614                         "paddusb b16Dither, %%mm1       \n\t"
 615                         "paddusb b16Dither, %%mm0       \n\t"
 616                         "paddusb b16Dither, %%mm3       \n\t"
 617 #endif
 618                         "punpcklbw %%mm7, %%mm1         \n\t" // 0G0G0G0G
 619                         "punpcklbw %%mm7, %%mm3         \n\t" // 0B0B0B0B
 620                         "punpcklbw %%mm7, %%mm0         \n\t" // 0R0R0R0R
 621
 622                         "psrlw $3, %%mm3                \n\t"
 623                         "psllw $2, %%mm1                \n\t"
 624                         "psllw $7, %%mm0                \n\t"
 625                         "pand g15Mask, %%mm1            \n\t"
 626                         "pand r15Mask, %%mm0            \n\t"
 627
 628                         "por %%mm3, %%mm1               \n\t"
 629                         "por %%mm1, %%mm0               \n\t"
 630
 631                         MOVNTQ(%%mm0, (%4, %%eax, 2))
 632
 633                         "addl $4, %%eax                 \n\t"
 634                         "cmpl %5, %%eax                 \n\t"
 635                         " jb 1b                         \n\t"
 636
 637                         :: "r" (buf0), "r" (buf1), "r" (uvbuf0), "r" (uvbuf1), "r" (dest), "m" (dstw),
 638                         "m" (yalpha1), "m" (uvalpha1)
 639                         : "%eax"
 640                         );
 641                 }
 642                 else if(dstbpp==16)
 643                 {
 644                         asm volatile(
 645
 646 FULL_YSCALEYUV2RGB
 647 #ifdef DITHER1XBPP
 648                         "paddusb g16Dither, %%mm1       \n\t"
 649                         "paddusb b16Dither, %%mm0       \n\t"
 650                         "paddusb b16Dither, %%mm3       \n\t"
 651 #endif
 652                         "punpcklbw %%mm7, %%mm1         \n\t" // 0G0G0G0G
 653                         "punpcklbw %%mm7, %%mm3         \n\t" // 0B0B0B0B
 654                         "punpcklbw %%mm7, %%mm0         \n\t" // 0R0R0R0R
 655
 656                         "psrlw $3, %%mm3                \n\t"
 657                         "psllw $3, %%mm1                \n\t"
 658                         "psllw $8, %%mm0                \n\t"
 659                         "pand g16Mask, %%mm1            \n\t"
 660                         "pand r16Mask, %%mm0            \n\t"
 661
 662                         "por %%mm3, %%mm1               \n\t"
 663                         "por %%mm1, %%mm0               \n\t"
 664
 665                         MOVNTQ(%%mm0, (%4, %%eax, 2))
 666
 667                         "addl $4, %%eax                 \n\t"
 668                         "cmpl %5, %%eax                 \n\t"
 669                         " jb 1b                         \n\t"
 670
 671                         :: "r" (buf0), "r" (buf1), "r" (uvbuf0), "r" (uvbuf1), "r" (dest), "m" (dstw),
 672                         "m" (yalpha1), "m" (uvalpha1)
 673                         : "%eax"
 674                         );
 675                 }
 676 #else
 677                 asm volatile ("\n\t"::: "memory");
 678
 679                 if(dstbpp==32 || dstbpp==24)
 680                 {
 681                         int i;
 682                         for(i=0;i<dstw;i++){
 683                                 // vertical linear interpolation && yuv2rgb in a single step:
 684                                 int Y=yuvtab_2568[((buf0[i]*yalpha1+buf1[i]*yalpha)>>19)];
 685                                 int U=((uvbuf0[i]*uvalpha1+uvbuf1[i]*uvalpha)>>19);
 686                                 int V=((uvbuf0[i+2048]*uvalpha1+uvbuf1[i+2048]*uvalpha)>>19);
 687                                 dest[0]=clip_table[((Y + yuvtab_40cf[U]) >>13)];
 688                                 dest[1]=clip_table[((Y + yuvtab_1a1e[V] + yuvtab_0c92[U]) >>13)];
 689                                 dest[2]=clip_table[((Y + yuvtab_3343[V]) >>13)];
 690                                 dest+=dstbpp>>3;
 691                         }
 692                 }
 693                 else if(dstbpp==16)
 694                 {
 695                         int i;
 696                         for(i=0;i<dstw;i++){
 697                                 // vertical linear interpolation && yuv2rgb in a single step:
 698                                 int Y=yuvtab_2568[((buf0[i]*yalpha1+buf1[i]*yalpha)>>19)];
 699                                 int U=((uvbuf0[i]*uvalpha1+uvbuf1[i]*uvalpha)>>19);
 700                                 int V=((uvbuf0[i+2048]*uvalpha1+uvbuf1[i+2048]*uvalpha)>>19);
 701
 702                                 ((uint16_t*)dest)[i] =
 703                                         clip_table16b[(Y + yuvtab_40cf[U]) >>13] |
 704                                         clip_table16g[(Y + yuvtab_1a1e[V] + yuvtab_0c92[U]) >>13] |
 705                                         clip_table16r[(Y + yuvtab_3343[V]) >>13];
 706                         }
 707                 }
 708                 else if(dstbpp==15)
 709                 {
 710                         int i;
 711                         for(i=0;i<dstw;i++){
 712                                 // vertical linear interpolation && yuv2rgb in a single step:
 713                                 int Y=yuvtab_2568[((buf0[i]*yalpha1+buf1[i]*yalpha)>>19)];
 714                                 int U=((uvbuf0[i]*uvalpha1+uvbuf1[i]*uvalpha)>>19);
 715                                 int V=((uvbuf0[i+2048]*uvalpha1+uvbuf1[i+2048]*uvalpha)>>19);
 716
 717                                 ((uint16_t*)dest)[i] =
 718                                         clip_table15b[(Y + yuvtab_40cf[U]) >>13] |
 719                                         clip_table15g[(Y + yuvtab_1a1e[V] + yuvtab_0c92[U]) >>13] |
 720                                         clip_table15r[(Y + yuvtab_3343[V]) >>13];
 721                         }
 722                 }
 723 #endif
 724         }//FULL_UV_IPOL
 725         else
 726         {
 727 #ifdef HAVE_MMX
 728                 if(dstbpp == 32)
 729                 {
 730                         asm volatile(
 731                                 YSCALEYUV2RGB
 732                                 WRITEBGR32
 733
 734                         :: "r" (buf0), "r" (buf1), "r" (uvbuf0), "r" (uvbuf1), "r" (dest), "m" (dstw),
 735                         "m" (yalpha1), "m" (uvalpha1)
 736                         : "%eax"
 737                         );
 738                 }
 739                 else if(dstbpp==24)
 740                 {
 741                         asm volatile(
 742                                 YSCALEYUV2RGB
 743                                 WRITEBGR24
 744
 745                         :: "r" (buf0), "r" (buf1), "r" (uvbuf0), "r" (uvbuf1), "r" (dest), "m" (dstw),
 746                         "m" (yalpha1), "m" (uvalpha1)
 747                         : "%eax", "%ebx"
 748                         );
 749                 }
 750                 else if(dstbpp==15)
 751                 {
 752                         asm volatile(
 753                                 YSCALEYUV2RGB
 754                 /* mm2=B, %%mm4=G, %%mm5=R, %%mm7=0 */
 755 #ifdef DITHER1XBPP
 756                                 "paddusb b16Dither, %%mm2       \n\t"
 757                                 "paddusb b16Dither, %%mm4       \n\t"
 758                                 "paddusb b16Dither, %%mm5       \n\t"
 759 #endif
 760
 761                                 WRITEBGR15
 762
 763                         :: "r" (buf0), "r" (buf1), "r" (uvbuf0), "r" (uvbuf1), "r" (dest), "m" (dstw),
 764                         "m" (yalpha1), "m" (uvalpha1)
 765                         : "%eax"
 766                         );
 767                 }
 768                 else if(dstbpp==16)
 769                 {
 770                         asm volatile(
 771                                 YSCALEYUV2RGB
 772                 /* mm2=B, %%mm4=G, %%mm5=R, %%mm7=0 */
 773 #ifdef DITHER1XBPP
 774                                 "paddusb g16Dither, %%mm2       \n\t"
 775                                 "paddusb b16Dither, %%mm4       \n\t"
 776                                 "paddusb b16Dither, %%mm5       \n\t"
 777 #endif
 778
 779                                 WRITEBGR16
 780
 781                         :: "r" (buf0), "r" (buf1), "r" (uvbuf0), "r" (uvbuf1), "r" (dest), "m" (dstw),
 782                         "m" (yalpha1), "m" (uvalpha1)
 783                         : "%eax"
 784                         );
 785                 }
 786 #else
 787                 asm volatile ("\n\t"::: "memory");
 788
 789                 if(dstbpp==32)
 790                 {
 791                         int i;
 792                         for(i=0; i<dstw-1; i+=2){
 793                                 // vertical linear interpolation && yuv2rgb in a single step:
 794                                 int Y1=yuvtab_2568[((buf0[i]*yalpha1+buf1[i]*yalpha)>>19)];
 795                                 int Y2=yuvtab_2568[((buf0[i+1]*yalpha1+buf1[i+1]*yalpha)>>19)];
 796                                 int U=((uvbuf0[i>>1]*uvalpha1+uvbuf1[i>>1]*uvalpha)>>19);
 797                                 int V=((uvbuf0[(i>>1)+2048]*uvalpha1+uvbuf1[(i>>1)+2048]*uvalpha)>>19);
 798
 799                                 int Cb= yuvtab_40cf[U];
 800                                 int Cg= yuvtab_1a1e[V] + yuvtab_0c92[U];
 801                                 int Cr= yuvtab_3343[V];
 802
 803                                 dest[4*i+0]=clip_table[((Y1 + Cb) >>13)];
 804                                 dest[4*i+1]=clip_table[((Y1 + Cg) >>13)];
 805                                 dest[4*i+2]=clip_table[((Y1 + Cr) >>13)];
 806
 807                                 dest[4*i+4]=clip_table[((Y2 + Cb) >>13)];
 808                                 dest[4*i+5]=clip_table[((Y2 + Cg) >>13)];
 809                                 dest[4*i+6]=clip_table[((Y2 + Cr) >>13)];
 810                         }
 811                 }
 812                 if(dstbpp==24)
 813                 {
 814                         int i;
 815                         for(i=0; i<dstw-1; i+=2){
 816                                 // vertical linear interpolation && yuv2rgb in a single step:
 817                                 int Y1=yuvtab_2568[((buf0[i]*yalpha1+buf1[i]*yalpha)>>19)];
 818                                 int Y2=yuvtab_2568[((buf0[i+1]*yalpha1+buf1[i+1]*yalpha)>>19)];
 819                                 int U=((uvbuf0[i>>1]*uvalpha1+uvbuf1[i>>1]*uvalpha)>>19);
 820                                 int V=((uvbuf0[(i>>1)+2048]*uvalpha1+uvbuf1[(i>>1)+2048]*uvalpha)>>19);
 821
 822                                 int Cb= yuvtab_40cf[U];
 823                                 int Cg= yuvtab_1a1e[V] + yuvtab_0c92[U];
 824                                 int Cr= yuvtab_3343[V];
 825
 826                                 dest[0]=clip_table[((Y1 + Cb) >>13)];
 827                                 dest[1]=clip_table[((Y1 + Cg) >>13)];
 828                                 dest[2]=clip_table[((Y1 + Cr) >>13)];
 829
 830                                 dest[3]=clip_table[((Y2 + Cb) >>13)];
 831                                 dest[4]=clip_table[((Y2 + Cg) >>13)];
 832                                 dest[5]=clip_table[((Y2 + Cr) >>13)];
 833                                 dest+=6;
 834                         }
 835                 }
 836                 else if(dstbpp==16)
 837                 {
 838                         int i;
 839                         for(i=0; i<dstw-1; i+=2){
 840                                 // vertical linear interpolation && yuv2rgb in a single step:
 841                                 int Y1=yuvtab_2568[((buf0[i]*yalpha1+buf1[i]*yalpha)>>19)];
 842                                 int Y2=yuvtab_2568[((buf0[i+1]*yalpha1+buf1[i+1]*yalpha)>>19)];
 843                                 int U=((uvbuf0[i>>1]*uvalpha1+uvbuf1[i>>1]*uvalpha)>>19);
 844                                 int V=((uvbuf0[(i>>1)+2048]*uvalpha1+uvbuf1[(i>>1)+2048]*uvalpha)>>19);
 845
 846                                 int Cb= yuvtab_40cf[U];
 847                                 int Cg= yuvtab_1a1e[V] + yuvtab_0c92[U];
 848                                 int Cr= yuvtab_3343[V];
 849
 850                                 ((uint16_t*)dest)[i] =
 851                                         clip_table16b[(Y1 + Cb) >>13] |
 852                                         clip_table16g[(Y1 + Cg) >>13] |
 853                                         clip_table16r[(Y1 + Cr) >>13];
 854
 855                                 ((uint16_t*)dest)[i+1] =
 856                                         clip_table16b[(Y2 + Cb) >>13] |
 857                                         clip_table16g[(Y2 + Cg) >>13] |
 858                                         clip_table16r[(Y2 + Cr) >>13];
 859                         }
 860                 }
 861                 else if(dstbpp==15)
 862                 {
 863                         int i;
 864                         for(i=0; i<dstw-1; i+=2){
 865                                 // vertical linear interpolation && yuv2rgb in a single step:
 866                                 int Y1=yuvtab_2568[((buf0[i]*yalpha1+buf1[i]*yalpha)>>19)];
 867                                 int Y2=yuvtab_2568[((buf0[i+1]*yalpha1+buf1[i+1]*yalpha)>>19)];
 868                                 int U=((uvbuf0[i>>1]*uvalpha1+uvbuf1[i>>1]*uvalpha)>>19);
 869                                 int V=((uvbuf0[(i>>1)+2048]*uvalpha1+uvbuf1[(i>>1)+2048]*uvalpha)>>19);
 870
 871                                 int Cb= yuvtab_40cf[U];
 872                                 int Cg= yuvtab_1a1e[V] + yuvtab_0c92[U];
 873                                 int Cr= yuvtab_3343[V];
 874
 875                                 ((uint16_t*)dest)[i] =
 876                                         clip_table15b[(Y1 + Cb) >>13] |
 877                                         clip_table15g[(Y1 + Cg) >>13] |
 878                                         clip_table15r[(Y1 + Cr) >>13];
 879
 880                                 ((uint16_t*)dest)[i+1] =
 881                                         clip_table15b[(Y2 + Cb) >>13] |
 882                                         clip_table15g[(Y2 + Cg) >>13] |
 883                                         clip_table15r[(Y2 + Cr) >>13];
 884                         }
 885                 }
 886 #endif
 887         } //!FULL_UV_IPOL
 888 }
 889
 890 /**
 891  * YV12 to RGB without scaling or interpolating
 892  */
 893 static inline void yuv2rgb1(uint16_t *buf0, uint16_t *buf1, uint16_t *uvbuf0, uint16_t *uvbuf1,
 894                             uint8_t *dest, int dstw, int yalpha, int uvalpha, int dstbpp)
 895 {
 896         int uvalpha1=uvalpha^4095;
 897 #ifdef HAVE_MMX
 898         int yalpha1=yalpha^4095;
 899 #endif
 900
 901         if(fullUVIpol || allwaysIpol)
 902         {
 903                 yuv2rgbX(buf0, buf1, uvbuf0, uvbuf1, dest, dstw, yalpha, uvalpha, dstbpp);
 904                 return;
 905         }
 906         if( yalpha > 2048 ) buf0 = buf1;
 907
 908 #ifdef HAVE_MMX
 909         if( uvalpha < 2048 ) // note this is not correct (shifts chrominance by 0.5 pixels) but its a bit faster
 910         {
 911                 if(dstbpp == 32)
 912                 {
 913                         asm volatile(
 914                                 YSCALEYUV2RGB1
 915                                 WRITEBGR32
 916                         :: "r" (buf0), "r" (buf1), "r" (uvbuf0), "r" (uvbuf1), "r" (dest), "m" (dstw),
 917                         "m" (yalpha1), "m" (uvalpha1)
 918                         : "%eax"
 919                         );
 920                 }
 921                 else if(dstbpp==24)
 922                 {
 923                         asm volatile(
 924                                 YSCALEYUV2RGB1
 925                                 WRITEBGR24
 926                         :: "r" (buf0), "r" (buf1), "r" (uvbuf0), "r" (uvbuf1), "r" (dest), "m" (dstw),
 927                         "m" (yalpha1), "m" (uvalpha1)
 928                         : "%eax", "%ebx"
 929                         );
 930                 }
 931                 else if(dstbpp==15)
 932                 {
 933                         asm volatile(
 934                                 YSCALEYUV2RGB1
 935                 /* mm2=B, %%mm4=G, %%mm5=R, %%mm7=0 */
 936 #ifdef DITHER1XBPP
 937                                 "paddusb b16Dither, %%mm2       \n\t"
 938                                 "paddusb b16Dither, %%mm4       \n\t"
 939                                 "paddusb b16Dither, %%mm5       \n\t"
 940 #endif
 941                                 WRITEBGR15
 942                         :: "r" (buf0), "r" (buf1), "r" (uvbuf0), "r" (uvbuf1), "r" (dest), "m" (dstw),
 943                         "m" (yalpha1), "m" (uvalpha1)
 944                         : "%eax"
 945                         );
 946                 }
 947                 else if(dstbpp==16)
 948                 {
 949                         asm volatile(
 950                                 YSCALEYUV2RGB1
 951                 /* mm2=B, %%mm4=G, %%mm5=R, %%mm7=0 */
 952 #ifdef DITHER1XBPP
 953                                 "paddusb g16Dither, %%mm2       \n\t"
 954                                 "paddusb b16Dither, %%mm4       \n\t"
 955                                 "paddusb b16Dither, %%mm5       \n\t"
 956 #endif
 957
 958                                 WRITEBGR16
 959                         :: "r" (buf0), "r" (buf1), "r" (uvbuf0), "r" (uvbuf1), "r" (dest), "m" (dstw),
 960                         "m" (yalpha1), "m" (uvalpha1)
 961                         : "%eax"
 962                         );
 963                 }
 964         }
 965         else
 966         {
 967                 if(dstbpp == 32)
 968                 {
 969                         asm volatile(
 970                                 YSCALEYUV2RGB1b
 971                                 WRITEBGR32
 972                         :: "r" (buf0), "r" (buf1), "r" (uvbuf0), "r" (uvbuf1), "r" (dest), "m" (dstw),
 973                         "m" (yalpha1), "m" (uvalpha1)
 974                         : "%eax"
 975                         );
 976                 }
 977                 else if(dstbpp==24)
 978                 {
 979                         asm volatile(
 980                                 YSCALEYUV2RGB1b
 981                                 WRITEBGR24
 982                         :: "r" (buf0), "r" (buf1), "r" (uvbuf0), "r" (uvbuf1), "r" (dest), "m" (dstw),
 983                         "m" (yalpha1), "m" (uvalpha1)
 984                         : "%eax", "%ebx"
 985                         );
 986                 }
 987                 else if(dstbpp==15)
 988                 {
 989                         asm volatile(
 990                                 YSCALEYUV2RGB1b
 991                 /* mm2=B, %%mm4=G, %%mm5=R, %%mm7=0 */
 992 #ifdef DITHER1XBPP
 993                                 "paddusb b16Dither, %%mm2       \n\t"
 994                                 "paddusb b16Dither, %%mm4       \n\t"
 995                                 "paddusb b16Dither, %%mm5       \n\t"
 996 #endif
 997                                 WRITEBGR15
 998                         :: "r" (buf0), "r" (buf1), "r" (uvbuf0), "r" (uvbuf1), "r" (dest), "m" (dstw),
 999                         "m" (yalpha1), "m" (uvalpha1)
1000                         : "%eax"
1001                         );
1002                 }
1003                 else if(dstbpp==16)
1004                 {
1005                         asm volatile(
1006                                 YSCALEYUV2RGB1b
1007                 /* mm2=B, %%mm4=G, %%mm5=R, %%mm7=0 */
1008 #ifdef DITHER1XBPP
1009                                 "paddusb g16Dither, %%mm2       \n\t"
1010                                 "paddusb b16Dither, %%mm4       \n\t"
1011                                 "paddusb b16Dither, %%mm5       \n\t"
1012 #endif
1013
1014                                 WRITEBGR16
1015                         :: "r" (buf0), "r" (buf1), "r" (uvbuf0), "r" (uvbuf1), "r" (dest), "m" (dstw),
1016                         "m" (yalpha1), "m" (uvalpha1)
1017                         : "%eax"
1018                         );
1019                 }
1020         }
1021 #else
1022 //FIXME write 2 versions (for even & odd lines)
1023         asm volatile ("\n\t"::: "memory");
1024
1025         if(dstbpp==32)
1026         {
1027                 int i;
1028                 for(i=0; i<dstw-1; i+=2){
1029                         // vertical linear interpolation && yuv2rgb in a single step:
1030                         int Y1=yuvtab_2568[buf0[i]>>7];
1031                         int Y2=yuvtab_2568[buf0[i+1]>>7];
1032                         int U=((uvbuf0[i>>1]*uvalpha1+uvbuf1[i>>1]*uvalpha)>>19);
1033                         int V=((uvbuf0[(i>>1)+2048]*uvalpha1+uvbuf1[(i>>1)+2048]*uvalpha)>>19);
1034
1035                         int Cb= yuvtab_40cf[U];
1036                         int Cg= yuvtab_1a1e[V] + yuvtab_0c92[U];
1037                         int Cr= yuvtab_3343[V];
1038
1039                         dest[4*i+0]=clip_table[((Y1 + Cb) >>13)];
1040                         dest[4*i+1]=clip_table[((Y1 + Cg) >>13)];
1041                         dest[4*i+2]=clip_table[((Y1 + Cr) >>13)];
1042
1043                         dest[4*i+4]=clip_table[((Y2 + Cb) >>13)];
1044                         dest[4*i+5]=clip_table[((Y2 + Cg) >>13)];
1045                         dest[4*i+6]=clip_table[((Y2 + Cr) >>13)];
1046                 }
1047         }
1048         if(dstbpp==24)
1049         {
1050                 int i;
1051                 for(i=0; i<dstw-1; i+=2){
1052                         // vertical linear interpolation && yuv2rgb in a single step:
1053                         int Y1=yuvtab_2568[buf0[i]>>7];
1054                         int Y2=yuvtab_2568[buf0[i+1]>>7];
1055                         int U=((uvbuf0[i>>1]*uvalpha1+uvbuf1[i>>1]*uvalpha)>>19);
1056                         int V=((uvbuf0[(i>>1)+2048]*uvalpha1+uvbuf1[(i>>1)+2048]*uvalpha)>>19);
1057
1058                         int Cb= yuvtab_40cf[U];
1059                         int Cg= yuvtab_1a1e[V] + yuvtab_0c92[U];
1060                         int Cr= yuvtab_3343[V];
1061
1062                         dest[0]=clip_table[((Y1 + Cb) >>13)];
1063                         dest[1]=clip_table[((Y1 + Cg) >>13)];
1064                         dest[2]=clip_table[((Y1 + Cr) >>13)];
1065
1066                         dest[3]=clip_table[((Y2 + Cb) >>13)];
1067                         dest[4]=clip_table[((Y2 + Cg) >>13)];
1068                         dest[5]=clip_table[((Y2 + Cr) >>13)];
1069                         dest+=6;
1070                 }
1071         }
1072         else if(dstbpp==16)
1073         {
1074                 int i;
1075                 for(i=0; i<dstw-1; i+=2){
1076                         // vertical linear interpolation && yuv2rgb in a single step:
1077                         int Y1=yuvtab_2568[buf0[i]>>7];
1078                         int Y2=yuvtab_2568[buf0[i+1]>>7];
1079                         int U=((uvbuf0[i>>1]*uvalpha1+uvbuf1[i>>1]*uvalpha)>>19);
1080                         int V=((uvbuf0[(i>>1)+2048]*uvalpha1+uvbuf1[(i>>1)+2048]*uvalpha)>>19);
1081
1082                         int Cb= yuvtab_40cf[U];
1083                         int Cg= yuvtab_1a1e[V] + yuvtab_0c92[U];
1084                         int Cr= yuvtab_3343[V];
1085
1086                         ((uint16_t*)dest)[i] =
1087                                 clip_table16b[(Y1 + Cb) >>13] |
1088                                 clip_table16g[(Y1 + Cg) >>13] |
1089                                 clip_table16r[(Y1 + Cr) >>13];
1090
1091                         ((uint16_t*)dest)[i+1] =
1092                                 clip_table16b[(Y2 + Cb) >>13] |
1093                                 clip_table16g[(Y2 + Cg) >>13] |
1094                                 clip_table16r[(Y2 + Cr) >>13];
1095                 }
1096         }
1097         else if(dstbpp==15)
1098         {
1099                 int i;
1100                 for(i=0; i<dstw-1; i+=2){
1101                         // vertical linear interpolation && yuv2rgb in a single step:
1102                         int Y1=yuvtab_2568[buf0[i]>>7];
1103                         int Y2=yuvtab_2568[buf0[i+1]>>7];
1104                         int U=((uvbuf0[i>>1]*uvalpha1+uvbuf1[i>>1]*uvalpha)>>19);
1105                         int V=((uvbuf0[(i>>1)+2048]*uvalpha1+uvbuf1[(i>>1)+2048]*uvalpha)>>19);
1106
1107                         int Cb= yuvtab_40cf[U];
1108                         int Cg= yuvtab_1a1e[V] + yuvtab_0c92[U];
1109                         int Cr= yuvtab_3343[V];
1110
1111                         ((uint16_t*)dest)[i] =
1112                                 clip_table15b[(Y1 + Cb) >>13] |
1113                                 clip_table15g[(Y1 + Cg) >>13] |
1114                                 clip_table15r[(Y1 + Cr) >>13];
1115
1116                         ((uint16_t*)dest)[i+1] =
1117                                 clip_table15b[(Y2 + Cb) >>13] |
1118                                 clip_table15g[(Y2 + Cg) >>13] |
1119                                 clip_table15r[(Y2 + Cr) >>13];
1120                 }
1121         }
1122 #endif
1123 }
1124
1125
1126 static inline void hyscale(uint16_t *dst, int dstWidth, uint8_t *src, int srcWidth, int xInc)
1127 {
1128       // *** horizontal scale Y line to temp buffer
1129 #ifdef ARCH_X86
1130 #ifdef HAVE_MMX2
1131         int i;
1132         if(canMMX2BeUsed)
1133         {
1134                 asm volatile(
1135                         "pxor %%mm7, %%mm7              \n\t"
1136                         "pxor %%mm2, %%mm2              \n\t" // 2*xalpha
1137                         "movd %5, %%mm6                 \n\t" // xInc&0xFFFF
1138                         "punpcklwd %%mm6, %%mm6         \n\t"
1139                         "punpcklwd %%mm6, %%mm6         \n\t"
1140                         "movq %%mm6, %%mm2              \n\t"
1141                         "psllq $16, %%mm2               \n\t"
1142                         "paddw %%mm6, %%mm2             \n\t"
1143                         "psllq $16, %%mm2               \n\t"
1144                         "paddw %%mm6, %%mm2             \n\t"
1145                         "psllq $16, %%mm2               \n\t" //0,t,2t,3t               t=xInc&0xFF
1146                         "movq %%mm2, temp0              \n\t"
1147                         "movd %4, %%mm6                 \n\t" //(xInc*4)&0xFFFF
1148                         "punpcklwd %%mm6, %%mm6         \n\t"
1149                         "punpcklwd %%mm6, %%mm6         \n\t"
1150                         "xorl %%eax, %%eax              \n\t" // i
1151                         "movl %0, %%esi                 \n\t" // src
1152                         "movl %1, %%edi                 \n\t" // buf1
1153                         "movl %3, %%edx                 \n\t" // (xInc*4)>>16
1154                         "xorl %%ecx, %%ecx              \n\t"
1155                         "xorl %%ebx, %%ebx              \n\t"
1156                         "movw %4, %%bx                  \n\t" // (xInc*4)&0xFFFF
1157
1158 #define FUNNY_Y_CODE \
1159                         PREFETCH" 1024(%%esi)           \n\t"\
1160                         PREFETCH" 1056(%%esi)           \n\t"\
1161                         PREFETCH" 1088(%%esi)           \n\t"\
1162                         "call funnyYCode                \n\t"\
1163                         "movq temp0, %%mm2              \n\t"\
1164                         "xorl %%ecx, %%ecx              \n\t"
1165
1166 FUNNY_Y_CODE
1167 FUNNY_Y_CODE
1168 FUNNY_Y_CODE
1169 FUNNY_Y_CODE
1170 FUNNY_Y_CODE
1171 FUNNY_Y_CODE
1172 FUNNY_Y_CODE
1173 FUNNY_Y_CODE
1174
1175                         :: "m" (src), "m" (dst), "m" (dstWidth), "m" ((xInc*4)>>16),
1176                         "m" ((xInc*4)&0xFFFF), "m" (xInc&0xFFFF)
1177                         : "%eax", "%ebx", "%ecx", "%edx", "%esi", "%edi"
1178                 );
1179                 for(i=dstWidth-1; (i*xInc)>>16 >=srcWidth-1; i--) dst[i] = src[srcWidth-1]*128;
1180         }
1181         else
1182         {
1183 #endif
1184         //NO MMX just normal asm ...
1185         asm volatile(
1186                 "xorl %%eax, %%eax              \n\t" // i
1187                 "xorl %%ebx, %%ebx              \n\t" // xx
1188                 "xorl %%ecx, %%ecx              \n\t" // 2*xalpha
1189                 "1:                             \n\t"
1190                 "movzbl  (%0, %%ebx), %%edi     \n\t" //src[xx]
1191                 "movzbl 1(%0, %%ebx), %%esi     \n\t" //src[xx+1]
1192                 "subl %%edi, %%esi              \n\t" //src[xx+1] - src[xx]
1193                 "imull %%ecx, %%esi             \n\t" //(src[xx+1] - src[xx])*2*xalpha
1194                 "shll $16, %%edi                \n\t"
1195                 "addl %%edi, %%esi              \n\t" //src[xx+1]*2*xalpha + src[xx]*(1-2*xalpha)
1196                 "movl %1, %%edi                 \n\t"
1197                 "shrl $9, %%esi                 \n\t"
1198                 "movw %%si, (%%edi, %%eax, 2)   \n\t"
1199                 "addw %4, %%cx                  \n\t" //2*xalpha += xInc&0xFF
1200                 "adcl %3, %%ebx                 \n\t" //xx+= xInc>>8 + carry
1201
1202                 "movzbl (%0, %%ebx), %%edi      \n\t" //src[xx]
1203                 "movzbl 1(%0, %%ebx), %%esi     \n\t" //src[xx+1]
1204                 "subl %%edi, %%esi              \n\t" //src[xx+1] - src[xx]
1205                 "imull %%ecx, %%esi             \n\t" //(src[xx+1] - src[xx])*2*xalpha
1206                 "shll $16, %%edi                \n\t"
1207                 "addl %%edi, %%esi              \n\t" //src[xx+1]*2*xalpha + src[xx]*(1-2*xalpha)
1208                 "movl %1, %%edi                 \n\t"
1209                 "shrl $9, %%esi                 \n\t"
1210                 "movw %%si, 2(%%edi, %%eax, 2)  \n\t"
1211                 "addw %4, %%cx                  \n\t" //2*xalpha += xInc&0xFF
1212                 "adcl %3, %%ebx                 \n\t" //xx+= xInc>>8 + carry
1213
1214
1215                 "addl $2, %%eax                 \n\t"
1216                 "cmpl %2, %%eax                 \n\t"
1217                 " jb 1b                         \n\t"
1218
1219
1220                 :: "r" (src), "m" (dst), "m" (dstWidth), "m" (xInc>>16), "m" (xInc&0xFFFF)
1221                 : "%eax", "%ebx", "%ecx", "%edi", "%esi"
1222                 );
1223 #ifdef HAVE_MMX2
1224         } //if MMX2 cant be used
1225 #endif
1226 #else
1227         int i;
1228         unsigned int xpos=0;
1229         for(i=0;i<dstWidth;i++)
1230         {
1231                 register unsigned int xx=xpos>>16;
1232                 register unsigned int xalpha=(xpos&0xFFFF)>>9;
1233                 dst[i]= (src[xx]<<7) + (src[xx+1] - src[xx])*xalpha;
1234                 xpos+=xInc;
1235         }
1236 #endif
1237 }
1238
1239 inline static void hcscale(uint16_t *dst, int dstWidth,
1240                                 uint8_t *src1, uint8_t *src2, int srcWidth, int xInc)
1241 {
1242 #ifdef ARCH_X86
1243 #ifdef HAVE_MMX2
1244         int i;
1245         if(canMMX2BeUsed)
1246         {
1247                 asm volatile(
1248                 "pxor %%mm7, %%mm7              \n\t"
1249                 "pxor %%mm2, %%mm2              \n\t" // 2*xalpha
1250                 "movd %5, %%mm6                 \n\t" // xInc&0xFFFF
1251                 "punpcklwd %%mm6, %%mm6         \n\t"
1252                 "punpcklwd %%mm6, %%mm6         \n\t"
1253                 "movq %%mm6, %%mm2              \n\t"
1254                 "psllq $16, %%mm2               \n\t"
1255                 "paddw %%mm6, %%mm2             \n\t"
1256                 "psllq $16, %%mm2               \n\t"
1257                 "paddw %%mm6, %%mm2             \n\t"
1258                 "psllq $16, %%mm2               \n\t" //0,t,2t,3t               t=xInc&0xFFFF
1259                 "movq %%mm2, temp0              \n\t"
1260                 "movd %4, %%mm6                 \n\t" //(xInc*4)&0xFFFF
1261                 "punpcklwd %%mm6, %%mm6         \n\t"
1262                 "punpcklwd %%mm6, %%mm6         \n\t"
1263                 "xorl %%eax, %%eax              \n\t" // i
1264                 "movl %0, %%esi                 \n\t" // src
1265                 "movl %1, %%edi                 \n\t" // buf1
1266                 "movl %3, %%edx                 \n\t" // (xInc*4)>>16
1267                 "xorl %%ecx, %%ecx              \n\t"
1268                 "xorl %%ebx, %%ebx              \n\t"
1269                 "movw %4, %%bx                  \n\t" // (xInc*4)&0xFFFF
1270
1271 #define FUNNYUVCODE \
1272                         PREFETCH" 1024(%%esi)           \n\t"\
1273                         PREFETCH" 1056(%%esi)           \n\t"\
1274                         PREFETCH" 1088(%%esi)           \n\t"\
1275                         "call funnyUVCode               \n\t"\
1276                         "movq temp0, %%mm2              \n\t"\
1277                         "xorl %%ecx, %%ecx              \n\t"
1278
1279 FUNNYUVCODE
1280 FUNNYUVCODE
1281 FUNNYUVCODE
1282 FUNNYUVCODE
1283
1284 FUNNYUVCODE
1285 FUNNYUVCODE
1286 FUNNYUVCODE
1287 FUNNYUVCODE
1288                 "xorl %%eax, %%eax              \n\t" // i
1289                 "movl %6, %%esi                 \n\t" // src
1290                 "movl %1, %%edi                 \n\t" // buf1
1291                 "addl $4096, %%edi              \n\t"
1292
1293 FUNNYUVCODE
1294 FUNNYUVCODE
1295 FUNNYUVCODE
1296 FUNNYUVCODE
1297
1298 FUNNYUVCODE
1299 FUNNYUVCODE
1300 FUNNYUVCODE
1301 FUNNYUVCODE
1302
1303                 :: "m" (src1), "m" (dst), "m" (dstWidth), "m" ((xInc*4)>>16),
1304                   "m" ((xInc*4)&0xFFFF), "m" (xInc&0xFFFF), "m" (src2)
1305                 : "%eax", "%ebx", "%ecx", "%edx", "%esi", "%edi"
1306         );
1307                 for(i=dstWidth-1; (i*xInc)>>16 >=srcWidth/2-1; i--)
1308                 {
1309                         dst[i] = src1[srcWidth/2-1]*128;
1310                         dst[i+2048] = src2[srcWidth/2-1]*128;
1311                 }
1312         }
1313         else
1314         {
1315 #endif
1316         asm volatile(
1317                 "xorl %%eax, %%eax              \n\t" // i
1318                 "xorl %%ebx, %%ebx              \n\t" // xx
1319                 "xorl %%ecx, %%ecx              \n\t" // 2*xalpha
1320                 "1:                             \n\t"
1321                 "movl %0, %%esi                 \n\t"
1322                 "movzbl  (%%esi, %%ebx), %%edi  \n\t" //src[xx]
1323                 "movzbl 1(%%esi, %%ebx), %%esi  \n\t" //src[xx+1]
1324                 "subl %%edi, %%esi              \n\t" //src[xx+1] - src[xx]
1325                 "imull %%ecx, %%esi             \n\t" //(src[xx+1] - src[xx])*2*xalpha
1326                 "shll $16, %%edi                \n\t"
1327                 "addl %%edi, %%esi              \n\t" //src[xx+1]*2*xalpha + src[xx]*(1-2*xalpha)
1328                 "movl %1, %%edi                 \n\t"
1329                 "shrl $9, %%esi                 \n\t"
1330                 "movw %%si, (%%edi, %%eax, 2)   \n\t"
1331
1332                 "movzbl  (%5, %%ebx), %%edi     \n\t" //src[xx]
1333                 "movzbl 1(%5, %%ebx), %%esi     \n\t" //src[xx+1]
1334                 "subl %%edi, %%esi              \n\t" //src[xx+1] - src[xx]
1335                 "imull %%ecx, %%esi             \n\t" //(src[xx+1] - src[xx])*2*xalpha
1336                 "shll $16, %%edi                \n\t"
1337                 "addl %%edi, %%esi              \n\t" //src[xx+1]*2*xalpha + src[xx]*(1-2*xalpha)
1338                 "movl %1, %%edi                 \n\t"
1339                 "shrl $9, %%esi                 \n\t"
1340                 "movw %%si, 4096(%%edi, %%eax, 2)\n\t"
1341
1342                 "addw %4, %%cx                  \n\t" //2*xalpha += xInc&0xFF
1343                 "adcl %3, %%ebx                 \n\t" //xx+= xInc>>8 + carry
1344                 "addl $1, %%eax                 \n\t"
1345                 "cmpl %2, %%eax                 \n\t"
1346                 " jb 1b                         \n\t"
1347
1348                 :: "m" (src1), "m" (dst), "m" (dstWidth), "m" (xInc>>16), "m" (xInc&0xFFFF),
1349                 "r" (src2)
1350                 : "%eax", "%ebx", "%ecx", "%edi", "%esi"
1351                 );
1352 #ifdef HAVE_MMX2
1353         } //if MMX2 cant be used
1354 #endif
1355 #else
1356         int i;
1357         unsigned int xpos=0;
1358         for(i=0;i<dstWidth;i++)
1359         {
1360                 register unsigned int xx=xpos>>16;
1361                 register unsigned int xalpha=(xpos&0xFFFF)>>9;
1362                 dst[i]=(src1[xx]*(xalpha^127)+src1[xx+1]*xalpha);
1363                 dst[i+2048]=(src2[xx]*(xalpha^127)+src2[xx+1]*xalpha);
1364 /* slower
1365           dst[i]= (src1[xx]<<7) + (src1[xx+1] - src1[xx])*xalpha;
1366           dst[i+2048]=(src2[xx]<<7) + (src2[xx+1] - src2[xx])*xalpha;
1367 */
1368                 xpos+=xInc;
1369         }
1370 #endif
1371 }
1372
1373
1374 // *** bilinear scaling and yuv->rgb or yuv->yuv conversion of yv12 slices:
1375 // *** Note: it's called multiple times while decoding a frame, first time y==0
1376 // *** Designed to upscale, but may work for downscale too.
1377 // s_xinc = (src_width << 16) / dst_width
1378 // s_yinc = (src_height << 16) / dst_height
1379 void SwScale_YV12slice(unsigned char* srcptr[],int stride[], int y, int h,
1380                              uint8_t* dstptr[], int dststride, int dstw, int dstbpp,
1381                              unsigned int s_xinc,unsigned int s_yinc){
1382
1383 // scaling factors:
1384 //static int s_yinc=(vo_dga_src_height<<16)/vo_dga_vp_height;
1385 //static int s_xinc=(vo_dga_src_width<<8)/vo_dga_vp_width;
1386
1387 unsigned int s_xinc2;
1388
1389 static int s_srcypos; // points to the dst Pixels center in the source (0 is the center of pixel 0,0 in src)
1390 static int s_ypos;
1391
1392 // last horzontally interpolated lines, used to avoid unnecessary calculations
1393 static int s_last_ypos;
1394 static int s_last_y1pos;
1395
1396 #ifdef HAVE_MMX2
1397 // used to detect a horizontal size change
1398 static int old_dstw= -1;
1399 static int old_s_xinc= -1;
1400 #endif
1401
1402 int srcWidth= (dstw*s_xinc + 0x8000)>>16;
1403 int dstUVw= fullUVIpol ? dstw : dstw/2;
1404 int i;
1405
1406 #ifdef HAVE_MMX2
1407 canMMX2BeUsed= (s_xinc <= 0x10000 && (dstw&31)==0 && (srcWidth&15)==0) ? 1 : 0;
1408 #endif
1409
1410 // match pixel 0 of the src to pixel 0 of dst and match pixel n-2 of src to pixel n-2 of dst
1411 // n-2 is the last chrominance sample available
1412 // FIXME this is not perfect, but noone shuld notice the difference, the more correct variant
1413 // would be like the vertical one, but that would require some special code for the
1414 // first and last pixel
1415 if(canMMX2BeUsed)       s_xinc+= 20;
1416 else                    s_xinc = ((srcWidth-2)<<16)/(dstw-2) - 20;
1417
1418 if(fullUVIpol && !(dstbpp==12))         s_xinc2= s_xinc>>1;
1419 else                                    s_xinc2= s_xinc;
1420   // force calculation of the horizontal interpolation of the first line
1421
1422   if(y==0){
1423         s_last_ypos=-99;
1424         s_last_y1pos=-99;
1425         s_srcypos= s_yinc/2 - 0x8000;
1426         s_ypos=0;
1427
1428         // clean the buffers so that no green stuff is drawen if the width is not sane (%8=0)
1429         for(i=dstw-2; i<dstw+20; i++)
1430         {
1431                 pix_buf_uv[0][i] = pix_buf_uv[1][i]
1432                 = pix_buf_uv[0][2048+i] = pix_buf_uv[1][2048+i] = 128;
1433                 pix_buf_uv[0][i/2] = pix_buf_uv[1][i/2]
1434                 = pix_buf_uv[0][2048+i/2] = pix_buf_uv[1][2048+i/2] = 128;
1435                 pix_buf_y[0][i]= pix_buf_y[1][i]= 0;
1436         }
1437
1438 #ifdef HAVE_MMX2
1439 // cant downscale !!!
1440         if((old_s_xinc != s_xinc || old_dstw!=dstw) && canMMX2BeUsed)
1441         {
1442                 uint8_t *fragment;
1443                 int imm8OfPShufW1;
1444                 int imm8OfPShufW2;
1445                 int fragmentLength;
1446
1447                 int xpos, i;
1448
1449                 old_s_xinc= s_xinc;
1450                 old_dstw= dstw;
1451
1452                 // create an optimized horizontal scaling routine
1453
1454                 //code fragment
1455
1456                 asm volatile(
1457                         "jmp 9f                         \n\t"
1458                 // Begin
1459                         "0:                             \n\t"
1460                         "movq (%%esi), %%mm0            \n\t" //FIXME Alignment
1461                         "movq %%mm0, %%mm1              \n\t"
1462                         "psrlq $8, %%mm0                \n\t"
1463                         "punpcklbw %%mm7, %%mm1 \n\t"
1464                         "movq %%mm2, %%mm3              \n\t"
1465                         "punpcklbw %%mm7, %%mm0 \n\t"
1466                         "addw %%bx, %%cx                \n\t" //2*xalpha += (4*s_xinc)&0xFFFF
1467                         "pshufw $0xFF, %%mm1, %%mm1     \n\t"
1468                         "1:                             \n\t"
1469                         "adcl %%edx, %%esi              \n\t" //xx+= (4*s_xinc)>>16 + carry
1470                         "pshufw $0xFF, %%mm0, %%mm0     \n\t"
1471                         "2:                             \n\t"
1472                         "psrlw $9, %%mm3                \n\t"
1473                         "psubw %%mm1, %%mm0             \n\t"
1474                         "pmullw %%mm3, %%mm0            \n\t"
1475                         "paddw %%mm6, %%mm2             \n\t" // 2*alpha += xpos&0xFFFF
1476                         "psllw $7, %%mm1                \n\t"
1477                         "paddw %%mm1, %%mm0             \n\t"
1478
1479                         "movq %%mm0, (%%edi, %%eax)     \n\t"
1480
1481                         "addl $8, %%eax                 \n\t"
1482                 // End
1483                         "9:                             \n\t"
1484 //              "int $3\n\t"
1485                         "leal 0b, %0                    \n\t"
1486                         "leal 1b, %1                    \n\t"
1487                         "leal 2b, %2                    \n\t"
1488                         "decl %1                        \n\t"
1489                         "decl %2                        \n\t"
1490                         "subl %0, %1                    \n\t"
1491                         "subl %0, %2                    \n\t"
1492                         "leal 9b, %3                    \n\t"
1493                         "subl %0, %3                    \n\t"
1494                         :"=r" (fragment), "=r" (imm8OfPShufW1), "=r" (imm8OfPShufW2),
1495                          "=r" (fragmentLength)
1496                 );
1497
1498                 xpos= 0; //s_xinc/2 - 0x8000; // difference between pixel centers
1499
1500                 /* choose xinc so that all 8 parts fit exactly
1501                    Note: we cannot use just 1 part because it would not fit in the code cache */
1502 //              s_xinc2_diff= -((((s_xinc2*(dstw/8))&0xFFFF))/(dstw/8))-10;
1503 //              s_xinc_diff= -((((s_xinc*(dstw/8))&0xFFFF))/(dstw/8));
1504 #ifdef ALT_ERROR
1505 //              s_xinc2_diff+= ((0x10000/(dstw/8)));
1506 #endif
1507 //              s_xinc_diff= s_xinc2_diff*2;
1508
1509 //              s_xinc2+= s_xinc2_diff;
1510 //              s_xinc+= s_xinc_diff;
1511
1512 //              old_s_xinc= s_xinc;
1513
1514                 for(i=0; i<dstw/8; i++)
1515                 {
1516                         int xx=xpos>>16;
1517
1518                         if((i&3) == 0)
1519                         {
1520                                 int a=0;
1521                                 int b=((xpos+s_xinc)>>16) - xx;
1522                                 int c=((xpos+s_xinc*2)>>16) - xx;
1523                                 int d=((xpos+s_xinc*3)>>16) - xx;
1524
1525                                 memcpy(funnyYCode + fragmentLength*i/4, fragment, fragmentLength);
1526
1527                                 funnyYCode[fragmentLength*i/4 + imm8OfPShufW1]=
1528                                 funnyYCode[fragmentLength*i/4 + imm8OfPShufW2]=
1529                                         a | (b<<2) | (c<<4) | (d<<6);
1530
1531                                 // if we dont need to read 8 bytes than dont :), reduces the chance of
1532                                 // crossing a cache line
1533                                 if(d<3) funnyYCode[fragmentLength*i/4 + 1]= 0x6E;
1534
1535                                 funnyYCode[fragmentLength*(i+4)/4]= RET;
1536                         }
1537                         xpos+=s_xinc;
1538                 }
1539
1540                 xpos= 0; //s_xinc2/2 - 0x10000; // difference between centers of chrom samples
1541                 for(i=0; i<dstUVw/8; i++)
1542                 {
1543                         int xx=xpos>>16;
1544
1545                         if((i&3) == 0)
1546                         {
1547                                 int a=0;
1548                                 int b=((xpos+s_xinc2)>>16) - xx;
1549                                 int c=((xpos+s_xinc2*2)>>16) - xx;
1550                                 int d=((xpos+s_xinc2*3)>>16) - xx;
1551
1552                                 memcpy(funnyUVCode + fragmentLength*i/4, fragment, fragmentLength);
1553
1554                                 funnyUVCode[fragmentLength*i/4 + imm8OfPShufW1]=
1555                                 funnyUVCode[fragmentLength*i/4 + imm8OfPShufW2]=
1556                                         a | (b<<2) | (c<<4) | (d<<6);
1557
1558                                 // if we dont need to read 8 bytes than dont :), reduces the chance of
1559                                 // crossing a cache line
1560                                 if(d<3) funnyUVCode[fragmentLength*i/4 + 1]= 0x6E;
1561
1562                                 funnyUVCode[fragmentLength*(i+4)/4]= RET;
1563                         }
1564                         xpos+=s_xinc2;
1565                 }
1566 //              funnyCode[0]= RET;
1567         }
1568
1569 #endif // HAVE_MMX2
1570   } // reset counters
1571
1572   while(1){
1573     unsigned char *dest =dstptr[0]+dststride*s_ypos;
1574     unsigned char *uDest=dstptr[1]+(dststride>>1)*(s_ypos>>1);
1575     unsigned char *vDest=dstptr[2]+(dststride>>1)*(s_ypos>>1);
1576
1577     int y0=(s_srcypos + 0xFFFF)>>16;  // first luminance source line number below the dst line
1578         // points to the dst Pixels center in the source (0 is the center of pixel 0,0 in src)
1579     int srcuvpos= dstbpp==12 ?  s_srcypos + s_yinc/2 - 0x8000 :
1580                                 s_srcypos - 0x8000;
1581     int y1=(srcuvpos + 0x1FFFF)>>17; // first chrominance source line number below the dst line
1582     int yalpha=((s_srcypos-1)&0xFFFF)>>4;
1583     int uvalpha=((srcuvpos-1)&0x1FFFF)>>5;
1584     uint16_t *buf0=pix_buf_y[y0&1];             // top line of the interpolated slice
1585     uint16_t *buf1=pix_buf_y[((y0+1)&1)];       // bottom line of the interpolated slice
1586     uint16_t *uvbuf0=pix_buf_uv[y1&1];          // top line of the interpolated slice
1587     uint16_t *uvbuf1=pix_buf_uv[(y1+1)&1];      // bottom line of the interpolated slice
1588
1589     if(y0>=y+h) break; // FIXME wrong, skips last lines, but they are dupliactes anyway
1590
1591     if((y0&1) && dstbpp==12) uvalpha=-1; // there is no alpha if there is no line
1592
1593     s_ypos++; s_srcypos+=s_yinc;
1594
1595     //only interpolate the src line horizontally if we didnt do it allready
1596         if(s_last_ypos!=y0)
1597         {
1598                 unsigned char *src;
1599                 // skip if first line has been horiz scaled alleady
1600                 if(s_last_ypos != y0-1)
1601                 {
1602                         // check if first line is before any available src lines
1603                         if(y0-1 < y)    src=srcptr[0]+(0     )*stride[0];
1604                         else            src=srcptr[0]+(y0-y-1)*stride[0];
1605
1606                         hyscale(buf0, dstw, src, srcWidth, s_xinc);
1607                 }
1608                 // check if second line is after any available src lines
1609                 if(y0-y >= h)   src=srcptr[0]+(h-1)*stride[0];
1610                 else            src=srcptr[0]+(y0-y)*stride[0];
1611
1612                 // the min() is required to avoid reuseing lines which where not available
1613                 s_last_ypos= MIN(y0, y+h-1);
1614                 hyscale(buf1, dstw, src, srcWidth, s_xinc);
1615         }
1616 //      printf("%d %d %d %d\n", y, y1, s_last_y1pos, h);
1617       // *** horizontal scale U and V lines to temp buffer
1618         if(s_last_y1pos!=y1)
1619         {
1620                 uint8_t *src1, *src2;
1621                 // skip if first line has been horiz scaled alleady
1622                 if(s_last_y1pos != y1-1)
1623                 {
1624                         // check if first line is before any available src lines
1625                         if(y1-y/2-1 < 0)
1626                         {
1627                                 src1= srcptr[1]+(0)*stride[1];
1628                                 src2= srcptr[2]+(0)*stride[2];
1629                         }else{
1630                                 src1= srcptr[1]+(y1-y/2-1)*stride[1];
1631                                 src2= srcptr[2]+(y1-y/2-1)*stride[2];
1632                         }
1633                         hcscale(uvbuf0, dstUVw, src1, src2, srcWidth, s_xinc2);
1634                 }
1635
1636                 // check if second line is after any available src lines
1637                 if(y1 - y/2 >= h/2)
1638                 {
1639                         src1= srcptr[1]+(h/2-1)*stride[1];
1640                         src2= srcptr[2]+(h/2-1)*stride[2];
1641                 }else{
1642                         src1= srcptr[1]+(y1-y/2)*stride[1];
1643                         src2= srcptr[2]+(y1-y/2)*stride[2];
1644                 }
1645                 hcscale(uvbuf1, dstUVw, src1, src2, srcWidth, s_xinc2);
1646
1647                 // the min() is required to avoid reuseing lines which where not available
1648                 s_last_y1pos= MIN(y1, y/2+h/2-1);
1649         }
1650
1651         if(dstbpp==12) //YV12
1652                 yuv2yuv(buf0, buf1, uvbuf0, uvbuf1, dest, uDest, vDest, dstw, yalpha, uvalpha);
1653         else if(ABS(s_yinc - 0x10000) < 10)
1654                 yuv2rgb1(buf0, buf1, uvbuf0, uvbuf1, dest, dstw, yalpha, uvalpha, dstbpp);
1655         else
1656                 yuv2rgbX(buf0, buf1, uvbuf0, uvbuf1, dest, dstw, yalpha, uvalpha, dstbpp);
1657
1658 #ifdef HAVE_MMX
1659         b16Dither= b16Dither1;
1660         b16Dither1= b16Dither2;
1661         b16Dither2= b16Dither;
1662
1663         g16Dither= g16Dither1;
1664         g16Dither1= g16Dither2;
1665         g16Dither2= g16Dither;
1666 #endif
1667   }
1668
1669 #ifdef HAVE_MMX
1670         __asm __volatile(SFENCE:::"memory");
1671         __asm __volatile(EMMS:::"memory");
1672 #endif
1673 }
1674
1675
1676 void SwScale_Init(){
1677     // generating tables:
1678     int i;
1679     for(i=0;i<256;i++){
1680         clip_table[i]=0;
1681         clip_table[i+256]=i;
1682         clip_table[i+512]=255;
1683         yuvtab_2568[i]=(0x2568*(i-16))+(256<<13);
1684         yuvtab_3343[i]=0x3343*(i-128);
1685         yuvtab_0c92[i]=-0x0c92*(i-128);
1686         yuvtab_1a1e[i]=-0x1a1e*(i-128);
1687         yuvtab_40cf[i]=0x40cf*(i-128);
1688     }
1689
1690     for(i=0; i<768; i++)
1691     {
1692         int v= clip_table[i];
1693         clip_table16b[i]= v>>3;
1694         clip_table16g[i]= (v<<3)&0x07E0;
1695         clip_table16r[i]= (v<<8)&0xF800;
1696         clip_table15b[i]= v>>3;
1697         clip_table15g[i]= (v<<2)&0x03E0;
1698         clip_table15r[i]= (v<<7)&0x7C00;
1699     }
1700
1701 }