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