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 #undef PAVGB
  15
  16 //#undef HAVE_MMX2
  17 //#undef HAVE_MMX
  18 //#undef ARCH_X86
  19 #define DITHER1XBPP
  20 int fullUVIpol=0;
  21 //disables the unscaled height version
  22 int allwaysIpol=0;
  23
  24 #define RET 0xC3 //near return opcode
  25 /*
  26 NOTES
  27
  28 known BUGS with known cause (no bugreports please!, but patches are welcome :) )
  29 horizontal MMX2 scaler reads 1-7 samples too much (might cause a sig11)
  30
  31 Supported output formats BGR15 BGR16 BGR24 BGR32
  32 BGR15 & BGR16 MMX verions support dithering
  33 Special versions: fast Y 1:1 scaling (no interpolation in y direction)
  34
  35 TODO
  36 more intelligent missalignment avoidance for the horizontal scaler
  37 bicubic scaler
  38 dither in C
  39 change the distance of the u & v buffer
  40 */
  41
  42 #define ABS(a) ((a) > 0 ? (a) : (-(a)))
  43 #define MIN(a,b) ((a) > (b) ? (b) : (a))
  44 #define MAX(a,b) ((a) < (b) ? (b) : (a))
  45
  46 #ifdef HAVE_MMX2
  47 #define PAVGB(a,b) "pavgb " #a ", " #b " \n\t"
  48 #elif defined (HAVE_3DNOW)
  49 #define PAVGB(a,b) "pavgusb " #a ", " #b " \n\t"
  50 #endif
  51
  52 #ifdef HAVE_MMX2
  53 #define MOVNTQ(a,b) "movntq " #a ", " #b " \n\t"
  54 #else
  55 #define MOVNTQ(a,b) "movq " #a ", " #b " \n\t"
  56 #endif
  57
  58
  59 #ifdef HAVE_MMX
  60 static uint64_t __attribute__((aligned(8))) yCoeff=    0x2568256825682568LL;
  61 static uint64_t __attribute__((aligned(8))) vrCoeff=   0x3343334333433343LL;
  62 static uint64_t __attribute__((aligned(8))) ubCoeff=   0x40cf40cf40cf40cfLL;
  63 static uint64_t __attribute__((aligned(8))) vgCoeff=   0xE5E2E5E2E5E2E5E2LL;
  64 static uint64_t __attribute__((aligned(8))) ugCoeff=   0xF36EF36EF36EF36ELL;
  65 static uint64_t __attribute__((aligned(8))) bF8=       0xF8F8F8F8F8F8F8F8LL;
  66 static uint64_t __attribute__((aligned(8))) bFC=       0xFCFCFCFCFCFCFCFCLL;
  67 static uint64_t __attribute__((aligned(8))) w400=      0x0400040004000400LL;
  68 static uint64_t __attribute__((aligned(8))) w80=       0x0080008000800080LL;
  69 static uint64_t __attribute__((aligned(8))) w10=       0x0010001000100010LL;
  70 static uint64_t __attribute__((aligned(8))) bm00001111=0x00000000FFFFFFFFLL;
  71 static uint64_t __attribute__((aligned(8))) bm00000111=0x0000000000FFFFFFLL;
  72 static uint64_t __attribute__((aligned(8))) bm11111000=0xFFFFFFFFFF000000LL;
  73
  74 static uint64_t __attribute__((aligned(8))) b16Dither= 0x0004000400040004LL;
  75 static uint64_t __attribute__((aligned(8))) b16Dither1=0x0004000400040004LL;
  76 static uint64_t __attribute__((aligned(8))) b16Dither2=0x0602060206020602LL;
  77 static uint64_t __attribute__((aligned(8))) g16Dither= 0x0002000200020002LL;
  78 static uint64_t __attribute__((aligned(8))) g16Dither1=0x0002000200020002LL;
  79 static uint64_t __attribute__((aligned(8))) g16Dither2=0x0301030103010301LL;
  80
  81 static uint64_t __attribute__((aligned(8))) b16Mask=   0x001F001F001F001FLL;
  82 static uint64_t __attribute__((aligned(8))) g16Mask=   0x07E007E007E007E0LL;
  83 static uint64_t __attribute__((aligned(8))) r16Mask=   0xF800F800F800F800LL;
  84 static uint64_t __attribute__((aligned(8))) b15Mask=   0x001F001F001F001FLL;
  85 static uint64_t __attribute__((aligned(8))) g15Mask=   0x03E003E003E003E0LL;
  86 static uint64_t __attribute__((aligned(8))) r15Mask=   0x7C007C007C007C00LL;
  87
  88 static uint64_t __attribute__((aligned(8))) temp0;
  89 static uint64_t __attribute__((aligned(8))) asm_yalpha1;
  90 static uint64_t __attribute__((aligned(8))) asm_uvalpha1;
  91 #endif
  92
  93 // temporary storage for 4 yuv lines:
  94 // 16bit for now (mmx likes it more compact)
  95 #ifdef HAVE_MMX
  96 static uint16_t __attribute__((aligned(8))) pix_buf_y[4][2048];
  97 static uint16_t __attribute__((aligned(8))) pix_buf_uv[2][2048*2];
  98 #else
  99 static uint16_t pix_buf_y[4][2048];
 100 static uint16_t pix_buf_uv[2][2048*2];
 101 #endif
 102
 103 // clipping helper table for C implementations:
 104 static unsigned char clip_table[768];
 105
 106 static unsigned short clip_table16b[768];
 107 static unsigned short clip_table16g[768];
 108 static unsigned short clip_table16r[768];
 109 static unsigned short clip_table15b[768];
 110 static unsigned short clip_table15g[768];
 111 static unsigned short clip_table15r[768];
 112
 113 // yuv->rgb conversion tables:
 114 static    int yuvtab_2568[256];
 115 static    int yuvtab_3343[256];
 116 static    int yuvtab_0c92[256];
 117 static    int yuvtab_1a1e[256];
 118 static    int yuvtab_40cf[256];
 119
 120 #ifdef HAVE_MMX2
 121 static uint8_t funnyYCode[10000];
 122 static uint8_t funnyUVCode[10000];
 123 #endif
 124
 125 static int canMMX2BeUsed=0;
 126
 127 #define FULL_YSCALEYUV2RGB \
 128                 "pxor %%mm7, %%mm7              \n\t"\
 129                 "movd %6, %%mm6                 \n\t" /*yalpha1*/\
 130                 "punpcklwd %%mm6, %%mm6         \n\t"\
 131                 "punpcklwd %%mm6, %%mm6         \n\t"\
 132                 "movd %7, %%mm5                 \n\t" /*uvalpha1*/\
 133                 "punpcklwd %%mm5, %%mm5         \n\t"\
 134                 "punpcklwd %%mm5, %%mm5         \n\t"\
 135                 "xorl %%eax, %%eax              \n\t"\
 136                 "1:                             \n\t"\
 137                 "movq (%0, %%eax, 2), %%mm0     \n\t" /*buf0[eax]*/\
 138                 "movq (%1, %%eax, 2), %%mm1     \n\t" /*buf1[eax]*/\
 139                 "movq (%2, %%eax,2), %%mm2      \n\t" /* uvbuf0[eax]*/\
 140                 "movq (%3, %%eax,2), %%mm3      \n\t" /* uvbuf1[eax]*/\
 141                 "psubw %%mm1, %%mm0             \n\t" /* buf0[eax] - buf1[eax]*/\
 142                 "psubw %%mm3, %%mm2             \n\t" /* uvbuf0[eax] - uvbuf1[eax]*/\
 143                 "pmulhw %%mm6, %%mm0            \n\t" /* (buf0[eax] - buf1[eax])yalpha1>>16*/\
 144                 "pmulhw %%mm5, %%mm2            \n\t" /* (uvbuf0[eax] - uvbuf1[eax])uvalpha1>>16*/\
 145                 "psraw $4, %%mm1                \n\t" /* buf0[eax] - buf1[eax] >>4*/\
 146                 "movq 4096(%2, %%eax,2), %%mm4  \n\t" /* uvbuf0[eax+2048]*/\
 147                 "psraw $4, %%mm3                \n\t" /* uvbuf0[eax] - uvbuf1[eax] >>4*/\
 148                 "paddw %%mm0, %%mm1             \n\t" /* buf0[eax]yalpha1 + buf1[eax](1-yalpha1) >>16*/\
 149                 "movq 4096(%3, %%eax,2), %%mm0  \n\t" /* uvbuf1[eax+2048]*/\
 150                 "paddw %%mm2, %%mm3             \n\t" /* uvbuf0[eax]uvalpha1 - uvbuf1[eax](1-uvalpha1)*/\
 151                 "psubw %%mm0, %%mm4             \n\t" /* uvbuf0[eax+2048] - uvbuf1[eax+2048]*/\
 152                 "psubw w80, %%mm1               \n\t" /* 8(Y-16)*/\
 153                 "psubw w400, %%mm3              \n\t" /* 8(U-128)*/\
 154                 "pmulhw yCoeff, %%mm1           \n\t"\
 155 \
 156 \
 157                 "pmulhw %%mm5, %%mm4            \n\t" /* (uvbuf0[eax+2048] - uvbuf1[eax+2048])uvalpha1>>16*/\
 158                 "movq %%mm3, %%mm2              \n\t" /* (U-128)8*/\
 159                 "pmulhw ubCoeff, %%mm3          \n\t"\
 160                 "psraw $4, %%mm0                \n\t" /* uvbuf0[eax+2048] - uvbuf1[eax+2048] >>4*/\
 161                 "pmulhw ugCoeff, %%mm2          \n\t"\
 162                 "paddw %%mm4, %%mm0             \n\t" /* uvbuf0[eax+2048]uvalpha1 - uvbuf1[eax+2048](1-uvalpha1)*/\
 163                 "psubw w400, %%mm0              \n\t" /* (V-128)8*/\
 164 \
 165 \
 166                 "movq %%mm0, %%mm4              \n\t" /* (V-128)8*/\
 167                 "pmulhw vrCoeff, %%mm0          \n\t"\
 168                 "pmulhw vgCoeff, %%mm4          \n\t"\
 169                 "paddw %%mm1, %%mm3             \n\t" /* B*/\
 170                 "paddw %%mm1, %%mm0             \n\t" /* R*/\
 171                 "packuswb %%mm3, %%mm3          \n\t"\
 172 \
 173                 "packuswb %%mm0, %%mm0          \n\t"\
 174                 "paddw %%mm4, %%mm2             \n\t"\
 175                 "paddw %%mm2, %%mm1             \n\t" /* G*/\
 176 \
 177                 "packuswb %%mm1, %%mm1          \n\t"
 178
 179 #define YSCALEYUV2RGB \
 180                 "movd %6, %%mm6                 \n\t" /*yalpha1*/\
 181                 "punpcklwd %%mm6, %%mm6         \n\t"\
 182                 "punpcklwd %%mm6, %%mm6         \n\t"\
 183                 "movq %%mm6, asm_yalpha1        \n\t"\
 184                 "movd %7, %%mm5                 \n\t" /*uvalpha1*/\
 185                 "punpcklwd %%mm5, %%mm5         \n\t"\
 186                 "punpcklwd %%mm5, %%mm5         \n\t"\
 187                 "movq %%mm5, asm_uvalpha1       \n\t"\
 188                 "xorl %%eax, %%eax              \n\t"\
 189                 "1:                             \n\t"\
 190                 "movq (%2, %%eax), %%mm2        \n\t" /* uvbuf0[eax]*/\
 191                 "movq (%3, %%eax), %%mm3        \n\t" /* uvbuf1[eax]*/\
 192                 "movq 4096(%2, %%eax), %%mm5    \n\t" /* uvbuf0[eax+2048]*/\
 193                 "movq 4096(%3, %%eax), %%mm4    \n\t" /* uvbuf1[eax+2048]*/\
 194                 "psubw %%mm3, %%mm2             \n\t" /* uvbuf0[eax] - uvbuf1[eax]*/\
 195                 "psubw %%mm4, %%mm5             \n\t" /* uvbuf0[eax+2048] - uvbuf1[eax+2048]*/\
 196                 "movq asm_uvalpha1, %%mm0       \n\t"\
 197                 "pmulhw %%mm0, %%mm2            \n\t" /* (uvbuf0[eax] - uvbuf1[eax])uvalpha1>>16*/\
 198                 "pmulhw %%mm0, %%mm5            \n\t" /* (uvbuf0[eax+2048] - uvbuf1[eax+2048])uvalpha1>>16*/\
 199                 "psraw $4, %%mm3                \n\t" /* uvbuf0[eax] - uvbuf1[eax] >>4*/\
 200                 "psraw $4, %%mm4                \n\t" /* uvbuf0[eax+2048] - uvbuf1[eax+2048] >>4*/\
 201                 "paddw %%mm2, %%mm3             \n\t" /* uvbuf0[eax]uvalpha1 - uvbuf1[eax](1-uvalpha1)*/\
 202                 "paddw %%mm5, %%mm4             \n\t" /* uvbuf0[eax+2048]uvalpha1 - uvbuf1[eax+2048](1-uvalpha1)*/\
 203                 "psubw w400, %%mm3              \n\t" /* (U-128)8*/\
 204                 "psubw w400, %%mm4              \n\t" /* (V-128)8*/\
 205                 "movq %%mm3, %%mm2              \n\t" /* (U-128)8*/\
 206                 "movq %%mm4, %%mm5              \n\t" /* (V-128)8*/\
 207                 "pmulhw ugCoeff, %%mm3          \n\t"\
 208                 "pmulhw vgCoeff, %%mm4          \n\t"\
 209         /* mm2=(U-128)8, mm3=ug, mm4=vg mm5=(V-128)8 */\
 210                 "movq (%0, %%eax, 2), %%mm0     \n\t" /*buf0[eax]*/\
 211                 "movq (%1, %%eax, 2), %%mm1     \n\t" /*buf1[eax]*/\
 212                 "movq 8(%0, %%eax, 2), %%mm6    \n\t" /*buf0[eax]*/\
 213                 "movq 8(%1, %%eax, 2), %%mm7    \n\t" /*buf1[eax]*/\
 214                 "psubw %%mm1, %%mm0             \n\t" /* buf0[eax] - buf1[eax]*/\
 215                 "psubw %%mm7, %%mm6             \n\t" /* buf0[eax] - buf1[eax]*/\
 216                 "pmulhw asm_yalpha1, %%mm0      \n\t" /* (buf0[eax] - buf1[eax])yalpha1>>16*/\
 217                 "pmulhw asm_yalpha1, %%mm6      \n\t" /* (buf0[eax] - buf1[eax])yalpha1>>16*/\
 218                 "psraw $4, %%mm1                \n\t" /* buf0[eax] - buf1[eax] >>4*/\
 219                 "psraw $4, %%mm7                \n\t" /* buf0[eax] - buf1[eax] >>4*/\
 220                 "paddw %%mm0, %%mm1             \n\t" /* buf0[eax]yalpha1 + buf1[eax](1-yalpha1) >>16*/\
 221                 "paddw %%mm6, %%mm7             \n\t" /* buf0[eax]yalpha1 + buf1[eax](1-yalpha1) >>16*/\
 222                 "pmulhw ubCoeff, %%mm2          \n\t"\
 223                 "pmulhw vrCoeff, %%mm5          \n\t"\
 224                 "psubw w80, %%mm1               \n\t" /* 8(Y-16)*/\
 225                 "psubw w80, %%mm7               \n\t" /* 8(Y-16)*/\
 226                 "pmulhw yCoeff, %%mm1           \n\t"\
 227                 "pmulhw yCoeff, %%mm7           \n\t"\
 228         /* mm1= Y1, mm2=ub, mm3=ug, mm4=vg mm5=vr, mm7=Y2 */\
 229                 "paddw %%mm3, %%mm4             \n\t"\
 230                 "movq %%mm2, %%mm0              \n\t"\
 231                 "movq %%mm5, %%mm6              \n\t"\
 232                 "movq %%mm4, %%mm3              \n\t"\
 233                 "punpcklwd %%mm2, %%mm2         \n\t"\
 234                 "punpcklwd %%mm5, %%mm5         \n\t"\
 235                 "punpcklwd %%mm4, %%mm4         \n\t"\
 236                 "paddw %%mm1, %%mm2             \n\t"\
 237                 "paddw %%mm1, %%mm5             \n\t"\
 238                 "paddw %%mm1, %%mm4             \n\t"\
 239                 "punpckhwd %%mm0, %%mm0         \n\t"\
 240                 "punpckhwd %%mm6, %%mm6         \n\t"\
 241                 "punpckhwd %%mm3, %%mm3         \n\t"\
 242                 "paddw %%mm7, %%mm0             \n\t"\
 243                 "paddw %%mm7, %%mm6             \n\t"\
 244                 "paddw %%mm7, %%mm3             \n\t"\
 245                 /* mm0=B1, mm2=B2, mm3=G2, mm4=G1, mm5=R1, mm6=R2 */\
 246                 "packuswb %%mm0, %%mm2          \n\t"\
 247                 "packuswb %%mm6, %%mm5          \n\t"\
 248                 "packuswb %%mm3, %%mm4          \n\t"\
 249                 "pxor %%mm7, %%mm7              \n\t"
 250
 251 #define YSCALEYUV2RGB1 \
 252                 "xorl %%eax, %%eax              \n\t"\
 253                 "1:                             \n\t"\
 254                 "movq (%2, %%eax), %%mm3        \n\t" /* uvbuf0[eax]*/\
 255                 "movq 4096(%2, %%eax), %%mm4    \n\t" /* uvbuf0[eax+2048]*/\
 256                 "psraw $4, %%mm3                \n\t" /* uvbuf0[eax] - uvbuf1[eax] >>4*/\
 257                 "psraw $4, %%mm4                \n\t" /* uvbuf0[eax+2048] - uvbuf1[eax+2048] >>4*/\
 258                 "psubw w400, %%mm3              \n\t" /* (U-128)8*/\
 259                 "psubw w400, %%mm4              \n\t" /* (V-128)8*/\
 260                 "movq %%mm3, %%mm2              \n\t" /* (U-128)8*/\
 261                 "movq %%mm4, %%mm5              \n\t" /* (V-128)8*/\
 262                 "pmulhw ugCoeff, %%mm3          \n\t"\
 263                 "pmulhw vgCoeff, %%mm4          \n\t"\
 264         /* mm2=(U-128)8, mm3=ug, mm4=vg mm5=(V-128)8 */\
 265                 "movq (%0, %%eax, 2), %%mm1     \n\t" /*buf0[eax]*/\
 266                 "movq 8(%0, %%eax, 2), %%mm7    \n\t" /*buf0[eax]*/\
 267                 "psraw $4, %%mm1                \n\t" /* buf0[eax] - buf1[eax] >>4*/\
 268                 "psraw $4, %%mm7                \n\t" /* buf0[eax] - buf1[eax] >>4*/\
 269                 "pmulhw ubCoeff, %%mm2          \n\t"\
 270                 "pmulhw vrCoeff, %%mm5          \n\t"\
 271                 "psubw w80, %%mm1               \n\t" /* 8(Y-16)*/\
 272                 "psubw w80, %%mm7               \n\t" /* 8(Y-16)*/\
 273                 "pmulhw yCoeff, %%mm1           \n\t"\
 274                 "pmulhw yCoeff, %%mm7           \n\t"\
 275         /* mm1= Y1, mm2=ub, mm3=ug, mm4=vg mm5=vr, mm7=Y2 */\
 276                 "paddw %%mm3, %%mm4             \n\t"\
 277                 "movq %%mm2, %%mm0              \n\t"\
 278                 "movq %%mm5, %%mm6              \n\t"\
 279                 "movq %%mm4, %%mm3              \n\t"\
 280                 "punpcklwd %%mm2, %%mm2         \n\t"\
 281                 "punpcklwd %%mm5, %%mm5         \n\t"\
 282                 "punpcklwd %%mm4, %%mm4         \n\t"\
 283                 "paddw %%mm1, %%mm2             \n\t"\
 284                 "paddw %%mm1, %%mm5             \n\t"\
 285                 "paddw %%mm1, %%mm4             \n\t"\
 286                 "punpckhwd %%mm0, %%mm0         \n\t"\
 287                 "punpckhwd %%mm6, %%mm6         \n\t"\
 288                 "punpckhwd %%mm3, %%mm3         \n\t"\
 289                 "paddw %%mm7, %%mm0             \n\t"\
 290                 "paddw %%mm7, %%mm6             \n\t"\
 291                 "paddw %%mm7, %%mm3             \n\t"\
 292                 /* mm0=B1, mm2=B2, mm3=G2, mm4=G1, mm5=R1, mm6=R2 */\
 293                 "packuswb %%mm0, %%mm2          \n\t"\
 294                 "packuswb %%mm6, %%mm5          \n\t"\
 295                 "packuswb %%mm3, %%mm4          \n\t"\
 296                 "pxor %%mm7, %%mm7              \n\t"
 297
 298 // do vertical chrominance interpolation
 299 #define YSCALEYUV2RGB1b \
 300                 "xorl %%eax, %%eax              \n\t"\
 301                 "1:                             \n\t"\
 302                 "movq (%2, %%eax), %%mm2        \n\t" /* uvbuf0[eax]*/\
 303                 "movq (%3, %%eax), %%mm3        \n\t" /* uvbuf1[eax]*/\
 304                 "movq 4096(%2, %%eax), %%mm5    \n\t" /* uvbuf0[eax+2048]*/\
 305                 "movq 4096(%3, %%eax), %%mm4    \n\t" /* uvbuf1[eax+2048]*/\
 306                 "paddw %%mm2, %%mm3             \n\t" /* uvbuf0[eax] + uvbuf1[eax]*/\
 307                 "paddw %%mm5, %%mm4             \n\t" /* uvbuf0[eax+2048] + uvbuf1[eax+2048]*/\
 308                 "psrlw $5, %%mm3                \n\t"\
 309                 "psrlw $5, %%mm4                \n\t"\
 310                 "psubw w400, %%mm3              \n\t" /* (U-128)8*/\
 311                 "psubw w400, %%mm4              \n\t" /* (V-128)8*/\
 312                 "movq %%mm3, %%mm2              \n\t" /* (U-128)8*/\
 313                 "movq %%mm4, %%mm5              \n\t" /* (V-128)8*/\
 314                 "pmulhw ugCoeff, %%mm3          \n\t"\
 315                 "pmulhw vgCoeff, %%mm4          \n\t"\
 316         /* mm2=(U-128)8, mm3=ug, mm4=vg mm5=(V-128)8 */\
 317                 "movq (%0, %%eax, 2), %%mm1     \n\t" /*buf0[eax]*/\
 318                 "movq 8(%0, %%eax, 2), %%mm7    \n\t" /*buf0[eax]*/\
 319                 "psraw $4, %%mm1                \n\t" /* buf0[eax] - buf1[eax] >>4*/\
 320                 "psraw $4, %%mm7                \n\t" /* buf0[eax] - buf1[eax] >>4*/\
 321                 "pmulhw ubCoeff, %%mm2          \n\t"\
 322                 "pmulhw vrCoeff, %%mm5          \n\t"\
 323                 "psubw w80, %%mm1               \n\t" /* 8(Y-16)*/\
 324                 "psubw w80, %%mm7               \n\t" /* 8(Y-16)*/\
 325                 "pmulhw yCoeff, %%mm1           \n\t"\
 326                 "pmulhw yCoeff, %%mm7           \n\t"\
 327         /* mm1= Y1, mm2=ub, mm3=ug, mm4=vg mm5=vr, mm7=Y2 */\
 328                 "paddw %%mm3, %%mm4             \n\t"\
 329                 "movq %%mm2, %%mm0              \n\t"\
 330                 "movq %%mm5, %%mm6              \n\t"\
 331                 "movq %%mm4, %%mm3              \n\t"\
 332                 "punpcklwd %%mm2, %%mm2         \n\t"\
 333                 "punpcklwd %%mm5, %%mm5         \n\t"\
 334                 "punpcklwd %%mm4, %%mm4         \n\t"\
 335                 "paddw %%mm1, %%mm2             \n\t"\
 336                 "paddw %%mm1, %%mm5             \n\t"\
 337                 "paddw %%mm1, %%mm4             \n\t"\
 338                 "punpckhwd %%mm0, %%mm0         \n\t"\
 339                 "punpckhwd %%mm6, %%mm6         \n\t"\
 340                 "punpckhwd %%mm3, %%mm3         \n\t"\
 341                 "paddw %%mm7, %%mm0             \n\t"\
 342                 "paddw %%mm7, %%mm6             \n\t"\
 343                 "paddw %%mm7, %%mm3             \n\t"\
 344                 /* mm0=B1, mm2=B2, mm3=G2, mm4=G1, mm5=R1, mm6=R2 */\
 345                 "packuswb %%mm0, %%mm2          \n\t"\
 346                 "packuswb %%mm6, %%mm5          \n\t"\
 347                 "packuswb %%mm3, %%mm4          \n\t"\
 348                 "pxor %%mm7, %%mm7              \n\t"
 349
 350 #define WRITEBGR32 \
 351                 /* mm2=B, %%mm4=G, %%mm5=R, %%mm7=0 */\
 352                         "movq %%mm2, %%mm1              \n\t" /* B */\
 353                         "movq %%mm5, %%mm6              \n\t" /* R */\
 354                         "punpcklbw %%mm4, %%mm2         \n\t" /* GBGBGBGB 0 */\
 355                         "punpcklbw %%mm7, %%mm5         \n\t" /* 0R0R0R0R 0 */\
 356                         "punpckhbw %%mm4, %%mm1         \n\t" /* GBGBGBGB 2 */\
 357                         "punpckhbw %%mm7, %%mm6         \n\t" /* 0R0R0R0R 2 */\
 358                         "movq %%mm2, %%mm0              \n\t" /* GBGBGBGB 0 */\
 359                         "movq %%mm1, %%mm3              \n\t" /* GBGBGBGB 2 */\
 360                         "punpcklwd %%mm5, %%mm0         \n\t" /* 0RGB0RGB 0 */\
 361                         "punpckhwd %%mm5, %%mm2         \n\t" /* 0RGB0RGB 1 */\
 362                         "punpcklwd %%mm6, %%mm1         \n\t" /* 0RGB0RGB 2 */\
 363                         "punpckhwd %%mm6, %%mm3         \n\t" /* 0RGB0RGB 3 */\
 364 \
 365                         MOVNTQ(%%mm0, (%4, %%eax, 4))\
 366                         MOVNTQ(%%mm2, 8(%4, %%eax, 4))\
 367                         MOVNTQ(%%mm1, 16(%4, %%eax, 4))\
 368                         MOVNTQ(%%mm3, 24(%4, %%eax, 4))\
 369 \
 370                         "addl $8, %%eax                 \n\t"\
 371                         "cmpl %5, %%eax                 \n\t"\
 372                         " jb 1b                         \n\t"
 373
 374 #define WRITEBGR16 \
 375                         "pand bF8, %%mm2                \n\t" /* B */\
 376                         "pand bFC, %%mm4                \n\t" /* G */\
 377                         "pand bF8, %%mm5                \n\t" /* R */\
 378                         "psrlq $3, %%mm2                \n\t"\
 379 \
 380                         "movq %%mm2, %%mm1              \n\t"\
 381                         "movq %%mm4, %%mm3              \n\t"\
 382 \
 383                         "punpcklbw %%mm7, %%mm3         \n\t"\
 384                         "punpcklbw %%mm5, %%mm2         \n\t"\
 385                         "punpckhbw %%mm7, %%mm4         \n\t"\
 386                         "punpckhbw %%mm5, %%mm1         \n\t"\
 387 \
 388                         "psllq $3, %%mm3                \n\t"\
 389                         "psllq $3, %%mm4                \n\t"\
 390 \
 391                         "por %%mm3, %%mm2               \n\t"\
 392                         "por %%mm4, %%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                         "pand bF8, %%mm2                \n\t" /* B */\
 403                         "pand bF8, %%mm4                \n\t" /* G */\
 404                         "pand bF8, %%mm5                \n\t" /* R */\
 405                         "psrlq $3, %%mm2                \n\t"\
 406                         "psrlq $1, %%mm5                \n\t"\
 407 \
 408                         "movq %%mm2, %%mm1              \n\t"\
 409                         "movq %%mm4, %%mm3              \n\t"\
 410 \
 411                         "punpcklbw %%mm7, %%mm3         \n\t"\
 412                         "punpcklbw %%mm5, %%mm2         \n\t"\
 413                         "punpckhbw %%mm7, %%mm4         \n\t"\
 414                         "punpckhbw %%mm5, %%mm1         \n\t"\
 415 \
 416                         "psllq $2, %%mm3                \n\t"\
 417                         "psllq $2, %%mm4                \n\t"\
 418 \
 419                         "por %%mm3, %%mm2               \n\t"\
 420                         "por %%mm4, %%mm1               \n\t"\
 421 \
 422                         MOVNTQ(%%mm2, (%4, %%eax, 2))\
 423                         MOVNTQ(%%mm1, 8(%4, %%eax, 2))\
 424 \
 425                         "addl $8, %%eax                 \n\t"\
 426                         "cmpl %5, %%eax                 \n\t"\
 427                         " jb 1b                         \n\t"
 428
 429 // FIXME find a faster way to shuffle it to BGR24
 430 #define WRITEBGR24 \
 431                 /* mm2=B, %%mm4=G, %%mm5=R, %%mm7=0 */\
 432                         "movq %%mm2, %%mm1              \n\t" /* B */\
 433                         "movq %%mm5, %%mm6              \n\t" /* R */\
 434                         "punpcklbw %%mm4, %%mm2         \n\t" /* GBGBGBGB 0 */\
 435                         "punpcklbw %%mm7, %%mm5         \n\t" /* 0R0R0R0R 0 */\
 436                         "punpckhbw %%mm4, %%mm1         \n\t" /* GBGBGBGB 2 */\
 437                         "punpckhbw %%mm7, %%mm6         \n\t" /* 0R0R0R0R 2 */\
 438                         "movq %%mm2, %%mm0              \n\t" /* GBGBGBGB 0 */\
 439                         "movq %%mm1, %%mm3              \n\t" /* GBGBGBGB 2 */\
 440                         "punpcklwd %%mm5, %%mm0         \n\t" /* 0RGB0RGB 0 */\
 441                         "punpckhwd %%mm5, %%mm2         \n\t" /* 0RGB0RGB 1 */\
 442                         "punpcklwd %%mm6, %%mm1         \n\t" /* 0RGB0RGB 2 */\
 443                         "punpckhwd %%mm6, %%mm3         \n\t" /* 0RGB0RGB 3 */\
 444 \
 445                         "movq %%mm0, %%mm4              \n\t" /* 0RGB0RGB 0 */\
 446                         "psrlq $8, %%mm0                \n\t" /* 00RGB0RG 0 */\
 447                         "pand bm00000111, %%mm4         \n\t" /* 00000RGB 0 */\
 448                         "pand bm11111000, %%mm0         \n\t" /* 00RGB000 0.5 */\
 449                         "por %%mm4, %%mm0               \n\t" /* 00RGBRGB 0 */\
 450                         "movq %%mm2, %%mm4              \n\t" /* 0RGB0RGB 1 */\
 451                         "psllq $48, %%mm2               \n\t" /* GB000000 1 */\
 452                         "por %%mm2, %%mm0               \n\t" /* GBRGBRGB 0 */\
 453 \
 454                         "movq %%mm4, %%mm2              \n\t" /* 0RGB0RGB 1 */\
 455                         "psrld $16, %%mm4               \n\t" /* 000R000R 1 */\
 456                         "psrlq $24, %%mm2               \n\t" /* 0000RGB0 1.5 */\
 457                         "por %%mm4, %%mm2               \n\t" /* 000RRGBR 1 */\
 458                         "pand bm00001111, %%mm2         \n\t" /* 0000RGBR 1 */\
 459                         "movq %%mm1, %%mm4              \n\t" /* 0RGB0RGB 2 */\
 460                         "psrlq $8, %%mm1                \n\t" /* 00RGB0RG 2 */\
 461                         "pand bm00000111, %%mm4         \n\t" /* 00000RGB 2 */\
 462                         "pand bm11111000, %%mm1         \n\t" /* 00RGB000 2.5 */\
 463                         "por %%mm4, %%mm1               \n\t" /* 00RGBRGB 2 */\
 464                         "movq %%mm1, %%mm4              \n\t" /* 00RGBRGB 2 */\
 465                         "psllq $32, %%mm1               \n\t" /* BRGB0000 2 */\
 466                         "por %%mm1, %%mm2               \n\t" /* BRGBRGBR 1 */\
 467 \
 468                         "psrlq $32, %%mm4               \n\t" /* 000000RG 2.5 */\
 469                         "movq %%mm3, %%mm5              \n\t" /* 0RGB0RGB 3 */\
 470                         "psrlq $8, %%mm3                \n\t" /* 00RGB0RG 3 */\
 471                         "pand bm00000111, %%mm5         \n\t" /* 00000RGB 3 */\
 472                         "pand bm11111000, %%mm3         \n\t" /* 00RGB000 3.5 */\
 473                         "por %%mm5, %%mm3               \n\t" /* 00RGBRGB 3 */\
 474                         "psllq $16, %%mm3               \n\t" /* RGBRGB00 3 */\
 475                         "por %%mm4, %%mm3               \n\t" /* RGBRGBRG 2.5 */\
 476 \
 477                         "leal (%%eax, %%eax, 2), %%ebx  \n\t"\
 478                         MOVNTQ(%%mm0, (%4, %%ebx))\
 479                         MOVNTQ(%%mm2, 8(%4, %%ebx))\
 480                         MOVNTQ(%%mm3, 16(%4, %%ebx))\
 481 \
 482                         "addl $8, %%eax                 \n\t"\
 483                         "cmpl %5, %%eax                 \n\t"\
 484                         " jb 1b                         \n\t"
 485
 486 #ifdef HAVE_MMX
 487 void in_asm_used_var_warning_killer()
 488 {
 489  int i= yCoeff+vrCoeff+ubCoeff+vgCoeff+ugCoeff+bF8+bFC+w400+w80+w10+
 490  bm00001111+bm00000111+bm11111000+b16Dither+b16Dither1+b16Dither2+g16Dither+g16Dither1+
 491  g16Dither2+b16Mask+g16Mask+r16Mask+b15Mask+g15Mask+r15Mask+temp0+asm_yalpha1+ asm_uvalpha1;
 492  if(i) i=0;
 493 }
 494 #endif
 495
 496 static inline void yuv2yuv(uint16_t *buf0, uint16_t *buf1, uint16_t *uvbuf0, uint16_t *uvbuf1,
 497                            uint8_t *dest, uint8_t *uDest, uint8_t *vDest, int dstw, int yalpha, int uvalpha)
 498 {
 499         int yalpha1=yalpha^4095;
 500         int uvalpha1=uvalpha^4095;
 501         int i;
 502
 503         asm volatile ("\n\t"::: "memory");
 504
 505         for(i=0;i<dstw;i++)
 506         {
 507                 ((uint8_t*)dest)[i] = (buf0[i]*yalpha1+buf1[i]*yalpha)>>19;
 508         }
 509
 510         if(uvalpha != -1)
 511         {
 512                 for(i=0; i<(dstw>>1); i++)
 513                 {
 514                         ((uint8_t*)uDest)[i] = (uvbuf0[i]*uvalpha1+uvbuf1[i]*uvalpha)>>19;
 515                         ((uint8_t*)vDest)[i] = (uvbuf0[i+2048]*uvalpha1+uvbuf1[i+2048]*uvalpha)>>19;
 516                 }
 517         }
 518 }
 519
 520 /**
 521  * vertical scale YV12 to RGB
 522  */
 523 static inline void yuv2rgbX(uint16_t *buf0, uint16_t *buf1, uint16_t *uvbuf0, uint16_t *uvbuf1,
 524                             uint8_t *dest, int dstw, int yalpha, int uvalpha, int dstbpp)
 525 {
 526         int yalpha1=yalpha^4095;
 527         int uvalpha1=uvalpha^4095;
 528
 529         if(fullUVIpol)
 530         {
 531
 532 #ifdef HAVE_MMX
 533                 if(dstbpp == 32)
 534                 {
 535                         asm volatile(
 536
 537
 538 FULL_YSCALEYUV2RGB
 539                         "punpcklbw %%mm1, %%mm3         \n\t" // BGBGBGBG
 540                         "punpcklbw %%mm7, %%mm0         \n\t" // R0R0R0R0
 541
 542                         "movq %%mm3, %%mm1              \n\t"
 543                         "punpcklwd %%mm0, %%mm3         \n\t" // BGR0BGR0
 544                         "punpckhwd %%mm0, %%mm1         \n\t" // BGR0BGR0
 545
 546                         MOVNTQ(%%mm3, (%4, %%eax, 4))
 547                         MOVNTQ(%%mm1, 8(%4, %%eax, 4))
 548
 549                         "addl $4, %%eax                 \n\t"
 550                         "cmpl %5, %%eax                 \n\t"
 551                         " jb 1b                         \n\t"
 552
 553
 554                         :: "r" (buf0), "r" (buf1), "r" (uvbuf0), "r" (uvbuf1), "r" (dest), "m" (dstw),
 555                         "m" (yalpha1), "m" (uvalpha1)
 556                         : "%eax"
 557                         );
 558                 }
 559                 else if(dstbpp==24)
 560                 {
 561                         asm volatile(
 562
 563 FULL_YSCALEYUV2RGB
 564
 565                                                                 // lsb ... msb
 566                         "punpcklbw %%mm1, %%mm3         \n\t" // BGBGBGBG
 567                         "punpcklbw %%mm7, %%mm0         \n\t" // R0R0R0R0
 568
 569                         "movq %%mm3, %%mm1              \n\t"
 570                         "punpcklwd %%mm0, %%mm3         \n\t" // BGR0BGR0
 571                         "punpckhwd %%mm0, %%mm1         \n\t" // BGR0BGR0
 572
 573                         "movq %%mm3, %%mm2              \n\t" // BGR0BGR0
 574                         "psrlq $8, %%mm3                \n\t" // GR0BGR00
 575                         "pand bm00000111, %%mm2         \n\t" // BGR00000
 576                         "pand bm11111000, %%mm3         \n\t" // 000BGR00
 577                         "por %%mm2, %%mm3               \n\t" // BGRBGR00
 578                         "movq %%mm1, %%mm2              \n\t"
 579                         "psllq $48, %%mm1               \n\t" // 000000BG
 580                         "por %%mm1, %%mm3               \n\t" // BGRBGRBG
 581
 582                         "movq %%mm2, %%mm1              \n\t" // BGR0BGR0
 583                         "psrld $16, %%mm2               \n\t" // R000R000
 584                         "psrlq $24, %%mm1               \n\t" // 0BGR0000
 585                         "por %%mm2, %%mm1               \n\t" // RBGRR000
 586
 587                         "movl %4, %%ebx                 \n\t"
 588                         "addl %%eax, %%ebx              \n\t"
 589
 590 #ifdef HAVE_MMX2
 591                         //FIXME Alignment
 592                         "movntq %%mm3, (%%ebx, %%eax, 2)\n\t"
 593                         "movntq %%mm1, 8(%%ebx, %%eax, 2)\n\t"
 594 #else
 595                         "movd %%mm3, (%%ebx, %%eax, 2)  \n\t"
 596                         "psrlq $32, %%mm3               \n\t"
 597                         "movd %%mm3, 4(%%ebx, %%eax, 2) \n\t"
 598                         "movd %%mm1, 8(%%ebx, %%eax, 2) \n\t"
 599 #endif
 600                         "addl $4, %%eax                 \n\t"
 601                         "cmpl %5, %%eax                 \n\t"
 602                         " jb 1b                         \n\t"
 603
 604                         :: "r" (buf0), "r" (buf1), "r" (uvbuf0), "r" (uvbuf1), "m" (dest), "m" (dstw),
 605                         "m" (yalpha1), "m" (uvalpha1)
 606                         : "%eax", "%ebx"
 607                         );
 608                 }
 609                 else if(dstbpp==15)
 610                 {
 611                         asm volatile(
 612
 613 FULL_YSCALEYUV2RGB
 614 #ifdef DITHER1XBPP
 615                         "paddusb b16Dither, %%mm1       \n\t"
 616                         "paddusb b16Dither, %%mm0       \n\t"
 617                         "paddusb b16Dither, %%mm3       \n\t"
 618 #endif
 619                         "punpcklbw %%mm7, %%mm1         \n\t" // 0G0G0G0G
 620                         "punpcklbw %%mm7, %%mm3         \n\t" // 0B0B0B0B
 621                         "punpcklbw %%mm7, %%mm0         \n\t" // 0R0R0R0R
 622
 623                         "psrlw $3, %%mm3                \n\t"
 624                         "psllw $2, %%mm1                \n\t"
 625                         "psllw $7, %%mm0                \n\t"
 626                         "pand g15Mask, %%mm1            \n\t"
 627                         "pand r15Mask, %%mm0            \n\t"
 628
 629                         "por %%mm3, %%mm1               \n\t"
 630                         "por %%mm1, %%mm0               \n\t"
 631
 632                         MOVNTQ(%%mm0, (%4, %%eax, 2))
 633
 634                         "addl $4, %%eax                 \n\t"
 635                         "cmpl %5, %%eax                 \n\t"
 636                         " jb 1b                         \n\t"
 637
 638                         :: "r" (buf0), "r" (buf1), "r" (uvbuf0), "r" (uvbuf1), "r" (dest), "m" (dstw),
 639                         "m" (yalpha1), "m" (uvalpha1)
 640                         : "%eax"
 641                         );
 642                 }
 643                 else if(dstbpp==16)
 644                 {
 645                         asm volatile(
 646
 647 FULL_YSCALEYUV2RGB
 648 #ifdef DITHER1XBPP
 649                         "paddusb g16Dither, %%mm1       \n\t"
 650                         "paddusb b16Dither, %%mm0       \n\t"
 651                         "paddusb b16Dither, %%mm3       \n\t"
 652 #endif
 653                         "punpcklbw %%mm7, %%mm1         \n\t" // 0G0G0G0G
 654                         "punpcklbw %%mm7, %%mm3         \n\t" // 0B0B0B0B
 655                         "punpcklbw %%mm7, %%mm0         \n\t" // 0R0R0R0R
 656
 657                         "psrlw $3, %%mm3                \n\t"
 658                         "psllw $3, %%mm1                \n\t"
 659                         "psllw $8, %%mm0                \n\t"
 660                         "pand g16Mask, %%mm1            \n\t"
 661                         "pand r16Mask, %%mm0            \n\t"
 662
 663                         "por %%mm3, %%mm1               \n\t"
 664                         "por %%mm1, %%mm0               \n\t"
 665
 666                         MOVNTQ(%%mm0, (%4, %%eax, 2))
 667
 668                         "addl $4, %%eax                 \n\t"
 669                         "cmpl %5, %%eax                 \n\t"
 670                         " jb 1b                         \n\t"
 671
 672                         :: "r" (buf0), "r" (buf1), "r" (uvbuf0), "r" (uvbuf1), "r" (dest), "m" (dstw),
 673                         "m" (yalpha1), "m" (uvalpha1)
 674                         : "%eax"
 675                         );
 676                 }
 677 #else
 678                 asm volatile ("\n\t"::: "memory");
 679
 680                 if(dstbpp==32 || dstbpp==24)
 681                 {
 682                         int i;
 683                         for(i=0;i<dstw;i++){
 684                                 // vertical linear interpolation && yuv2rgb in a single step:
 685                                 int Y=yuvtab_2568[((buf0[i]*yalpha1+buf1[i]*yalpha)>>19)];
 686                                 int U=((uvbuf0[i]*uvalpha1+uvbuf1[i]*uvalpha)>>19);
 687                                 int V=((uvbuf0[i+2048]*uvalpha1+uvbuf1[i+2048]*uvalpha)>>19);
 688                                 dest[0]=clip_table[((Y + yuvtab_40cf[U]) >>13)];
 689                                 dest[1]=clip_table[((Y + yuvtab_1a1e[V] + yuvtab_0c92[U]) >>13)];
 690                                 dest[2]=clip_table[((Y + yuvtab_3343[V]) >>13)];
 691                                 dest+=dstbpp>>3;
 692                         }
 693                 }
 694                 else if(dstbpp==16)
 695                 {
 696                         int i;
 697                         for(i=0;i<dstw;i++){
 698                                 // vertical linear interpolation && yuv2rgb in a single step:
 699                                 int Y=yuvtab_2568[((buf0[i]*yalpha1+buf1[i]*yalpha)>>19)];
 700                                 int U=((uvbuf0[i]*uvalpha1+uvbuf1[i]*uvalpha)>>19);
 701                                 int V=((uvbuf0[i+2048]*uvalpha1+uvbuf1[i+2048]*uvalpha)>>19);
 702
 703                                 ((uint16_t*)dest)[i] =
 704                                         clip_table16b[(Y + yuvtab_40cf[U]) >>13] |
 705                                         clip_table16g[(Y + yuvtab_1a1e[V] + yuvtab_0c92[U]) >>13] |
 706                                         clip_table16r[(Y + yuvtab_3343[V]) >>13];
 707                         }
 708                 }
 709                 else if(dstbpp==15)
 710                 {
 711                         int i;
 712                         for(i=0;i<dstw;i++){
 713                                 // vertical linear interpolation && yuv2rgb in a single step:
 714                                 int Y=yuvtab_2568[((buf0[i]*yalpha1+buf1[i]*yalpha)>>19)];
 715                                 int U=((uvbuf0[i]*uvalpha1+uvbuf1[i]*uvalpha)>>19);
 716                                 int V=((uvbuf0[i+2048]*uvalpha1+uvbuf1[i+2048]*uvalpha)>>19);
 717
 718                                 ((uint16_t*)dest)[i] =
 719                                         clip_table15b[(Y + yuvtab_40cf[U]) >>13] |
 720                                         clip_table15g[(Y + yuvtab_1a1e[V] + yuvtab_0c92[U]) >>13] |
 721                                         clip_table15r[(Y + yuvtab_3343[V]) >>13];
 722                         }
 723                 }
 724 #endif
 725         }//FULL_UV_IPOL
 726         else
 727         {
 728 #ifdef HAVE_MMX
 729                 if(dstbpp == 32)
 730                 {
 731                         asm volatile(
 732                                 YSCALEYUV2RGB
 733                                 WRITEBGR32
 734
 735                         :: "r" (buf0), "r" (buf1), "r" (uvbuf0), "r" (uvbuf1), "r" (dest), "m" (dstw),
 736                         "m" (yalpha1), "m" (uvalpha1)
 737                         : "%eax"
 738                         );
 739                 }
 740                 else if(dstbpp==24)
 741                 {
 742                         asm volatile(
 743                                 YSCALEYUV2RGB
 744                                 WRITEBGR24
 745
 746                         :: "r" (buf0), "r" (buf1), "r" (uvbuf0), "r" (uvbuf1), "r" (dest), "m" (dstw),
 747                         "m" (yalpha1), "m" (uvalpha1)
 748                         : "%eax", "%ebx"
 749                         );
 750                 }
 751                 else if(dstbpp==15)
 752                 {
 753                         asm volatile(
 754                                 YSCALEYUV2RGB
 755                 /* mm2=B, %%mm4=G, %%mm5=R, %%mm7=0 */
 756 #ifdef DITHER1XBPP
 757                                 "paddusb b16Dither, %%mm2       \n\t"
 758                                 "paddusb b16Dither, %%mm4       \n\t"
 759                                 "paddusb b16Dither, %%mm5       \n\t"
 760 #endif
 761
 762                                 WRITEBGR15
 763
 764                         :: "r" (buf0), "r" (buf1), "r" (uvbuf0), "r" (uvbuf1), "r" (dest), "m" (dstw),
 765                         "m" (yalpha1), "m" (uvalpha1)
 766                         : "%eax"
 767                         );
 768                 }
 769                 else if(dstbpp==16)
 770                 {
 771                         asm volatile(
 772                                 YSCALEYUV2RGB
 773                 /* mm2=B, %%mm4=G, %%mm5=R, %%mm7=0 */
 774 #ifdef DITHER1XBPP
 775                                 "paddusb g16Dither, %%mm2       \n\t"
 776                                 "paddusb b16Dither, %%mm4       \n\t"
 777                                 "paddusb b16Dither, %%mm5       \n\t"
 778 #endif
 779
 780                                 WRITEBGR16
 781
 782                         :: "r" (buf0), "r" (buf1), "r" (uvbuf0), "r" (uvbuf1), "r" (dest), "m" (dstw),
 783                         "m" (yalpha1), "m" (uvalpha1)
 784                         : "%eax"
 785                         );
 786                 }
 787 #else
 788                 asm volatile ("\n\t"::: "memory");
 789
 790                 if(dstbpp==32)
 791                 {
 792                         int i;
 793                         for(i=0; i<dstw-1; i+=2){
 794                                 // vertical linear interpolation && yuv2rgb in a single step:
 795                                 int Y1=yuvtab_2568[((buf0[i]*yalpha1+buf1[i]*yalpha)>>19)];
 796                                 int Y2=yuvtab_2568[((buf0[i+1]*yalpha1+buf1[i+1]*yalpha)>>19)];
 797                                 int U=((uvbuf0[i>>1]*uvalpha1+uvbuf1[i>>1]*uvalpha)>>19);
 798                                 int V=((uvbuf0[(i>>1)+2048]*uvalpha1+uvbuf1[(i>>1)+2048]*uvalpha)>>19);
 799
 800                                 int Cb= yuvtab_40cf[U];
 801                                 int Cg= yuvtab_1a1e[V] + yuvtab_0c92[U];
 802                                 int Cr= yuvtab_3343[V];
 803
 804                                 dest[4*i+0]=clip_table[((Y1 + Cb) >>13)];
 805                                 dest[4*i+1]=clip_table[((Y1 + Cg) >>13)];
 806                                 dest[4*i+2]=clip_table[((Y1 + Cr) >>13)];
 807
 808                                 dest[4*i+4]=clip_table[((Y2 + Cb) >>13)];
 809                                 dest[4*i+5]=clip_table[((Y2 + Cg) >>13)];
 810                                 dest[4*i+6]=clip_table[((Y2 + Cr) >>13)];
 811                         }
 812                 }
 813                 if(dstbpp==24)
 814                 {
 815                         int i;
 816                         for(i=0; i<dstw-1; i+=2){
 817                                 // vertical linear interpolation && yuv2rgb in a single step:
 818                                 int Y1=yuvtab_2568[((buf0[i]*yalpha1+buf1[i]*yalpha)>>19)];
 819                                 int Y2=yuvtab_2568[((buf0[i+1]*yalpha1+buf1[i+1]*yalpha)>>19)];
 820                                 int U=((uvbuf0[i>>1]*uvalpha1+uvbuf1[i>>1]*uvalpha)>>19);
 821                                 int V=((uvbuf0[(i>>1)+2048]*uvalpha1+uvbuf1[(i>>1)+2048]*uvalpha)>>19);
 822
 823                                 int Cb= yuvtab_40cf[U];
 824                                 int Cg= yuvtab_1a1e[V] + yuvtab_0c92[U];
 825                                 int Cr= yuvtab_3343[V];
 826
 827                                 dest[0]=clip_table[((Y1 + Cb) >>13)];
 828                                 dest[1]=clip_table[((Y1 + Cg) >>13)];
 829                                 dest[2]=clip_table[((Y1 + Cr) >>13)];
 830
 831                                 dest[3]=clip_table[((Y2 + Cb) >>13)];
 832                                 dest[4]=clip_table[((Y2 + Cg) >>13)];
 833                                 dest[5]=clip_table[((Y2 + Cr) >>13)];
 834                                 dest+=6;
 835                         }
 836                 }
 837                 else if(dstbpp==16)
 838                 {
 839                         int i;
 840                         for(i=0; i<dstw-1; i+=2){
 841                                 // vertical linear interpolation && yuv2rgb in a single step:
 842                                 int Y1=yuvtab_2568[((buf0[i]*yalpha1+buf1[i]*yalpha)>>19)];
 843                                 int Y2=yuvtab_2568[((buf0[i+1]*yalpha1+buf1[i+1]*yalpha)>>19)];
 844                                 int U=((uvbuf0[i>>1]*uvalpha1+uvbuf1[i>>1]*uvalpha)>>19);
 845                                 int V=((uvbuf0[(i>>1)+2048]*uvalpha1+uvbuf1[(i>>1)+2048]*uvalpha)>>19);
 846
 847                                 int Cb= yuvtab_40cf[U];
 848                                 int Cg= yuvtab_1a1e[V] + yuvtab_0c92[U];
 849                                 int Cr= yuvtab_3343[V];
 850
 851                                 ((uint16_t*)dest)[i] =
 852                                         clip_table16b[(Y1 + Cb) >>13] |
 853                                         clip_table16g[(Y1 + Cg) >>13] |
 854                                         clip_table16r[(Y1 + Cr) >>13];
 855
 856                                 ((uint16_t*)dest)[i+1] =
 857                                         clip_table16b[(Y2 + Cb) >>13] |
 858                                         clip_table16g[(Y2 + Cg) >>13] |
 859                                         clip_table16r[(Y2 + Cr) >>13];
 860                         }
 861                 }
 862                 else if(dstbpp==15)
 863                 {
 864                         int i;
 865                         for(i=0; i<dstw-1; i+=2){
 866                                 // vertical linear interpolation && yuv2rgb in a single step:
 867                                 int Y1=yuvtab_2568[((buf0[i]*yalpha1+buf1[i]*yalpha)>>19)];
 868                                 int Y2=yuvtab_2568[((buf0[i+1]*yalpha1+buf1[i+1]*yalpha)>>19)];
 869                                 int U=((uvbuf0[i>>1]*uvalpha1+uvbuf1[i>>1]*uvalpha)>>19);
 870                                 int V=((uvbuf0[(i>>1)+2048]*uvalpha1+uvbuf1[(i>>1)+2048]*uvalpha)>>19);
 871
 872                                 int Cb= yuvtab_40cf[U];
 873                                 int Cg= yuvtab_1a1e[V] + yuvtab_0c92[U];
 874                                 int Cr= yuvtab_3343[V];
 875
 876                                 ((uint16_t*)dest)[i] =
 877                                         clip_table15b[(Y1 + Cb) >>13] |
 878                                         clip_table15g[(Y1 + Cg) >>13] |
 879                                         clip_table15r[(Y1 + Cr) >>13];
 880
 881                                 ((uint16_t*)dest)[i+1] =
 882                                         clip_table15b[(Y2 + Cb) >>13] |
 883                                         clip_table15g[(Y2 + Cg) >>13] |
 884                                         clip_table15r[(Y2 + Cr) >>13];
 885                         }
 886                 }
 887 #endif
 888         } //!FULL_UV_IPOL
 889 }
 890
 891 /**
 892  * YV12 to RGB without scaling or interpolating
 893  */
 894 static inline void yuv2rgb1(uint16_t *buf0, uint16_t *buf1, uint16_t *uvbuf0, uint16_t *uvbuf1,
 895                             uint8_t *dest, int dstw, int yalpha, int uvalpha, int dstbpp)
 896 {
 897         int uvalpha1=uvalpha^4095;
 898 #ifdef HAVE_MMX
 899         int yalpha1=yalpha^4095;
 900 #endif
 901
 902         if(fullUVIpol || allwaysIpol)
 903         {
 904                 yuv2rgbX(buf0, buf1, uvbuf0, uvbuf1, dest, dstw, yalpha, uvalpha, dstbpp);
 905                 return;
 906         }
 907         if( yalpha > 2048 ) buf0 = buf1;
 908
 909 #ifdef HAVE_MMX
 910         if( uvalpha < 2048 ) // note this is not correct (shifts chrominance by 0.5 pixels) but its a bit faster
 911         {
 912                 if(dstbpp == 32)
 913                 {
 914                         asm volatile(
 915                                 YSCALEYUV2RGB1
 916                                 WRITEBGR32
 917                         :: "r" (buf0), "r" (buf1), "r" (uvbuf0), "r" (uvbuf1), "r" (dest), "m" (dstw),
 918                         "m" (yalpha1), "m" (uvalpha1)
 919                         : "%eax"
 920                         );
 921                 }
 922                 else if(dstbpp==24)
 923                 {
 924                         asm volatile(
 925                                 YSCALEYUV2RGB1
 926                                 WRITEBGR24
 927                         :: "r" (buf0), "r" (buf1), "r" (uvbuf0), "r" (uvbuf1), "r" (dest), "m" (dstw),
 928                         "m" (yalpha1), "m" (uvalpha1)
 929                         : "%eax", "%ebx"
 930                         );
 931                 }
 932                 else if(dstbpp==15)
 933                 {
 934                         asm volatile(
 935                                 YSCALEYUV2RGB1
 936                 /* mm2=B, %%mm4=G, %%mm5=R, %%mm7=0 */
 937 #ifdef DITHER1XBPP
 938                                 "paddusb b16Dither, %%mm2       \n\t"
 939                                 "paddusb b16Dither, %%mm4       \n\t"
 940                                 "paddusb b16Dither, %%mm5       \n\t"
 941 #endif
 942                                 WRITEBGR15
 943                         :: "r" (buf0), "r" (buf1), "r" (uvbuf0), "r" (uvbuf1), "r" (dest), "m" (dstw),
 944                         "m" (yalpha1), "m" (uvalpha1)
 945                         : "%eax"
 946                         );
 947                 }
 948                 else if(dstbpp==16)
 949                 {
 950                         asm volatile(
 951                                 YSCALEYUV2RGB1
 952                 /* mm2=B, %%mm4=G, %%mm5=R, %%mm7=0 */
 953 #ifdef DITHER1XBPP
 954                                 "paddusb g16Dither, %%mm2       \n\t"
 955                                 "paddusb b16Dither, %%mm4       \n\t"
 956                                 "paddusb b16Dither, %%mm5       \n\t"
 957 #endif
 958
 959                                 WRITEBGR16
 960                         :: "r" (buf0), "r" (buf1), "r" (uvbuf0), "r" (uvbuf1), "r" (dest), "m" (dstw),
 961                         "m" (yalpha1), "m" (uvalpha1)
 962                         : "%eax"
 963                         );
 964                 }
 965         }
 966         else
 967         {
 968                 if(dstbpp == 32)
 969                 {
 970                         asm volatile(
 971                                 YSCALEYUV2RGB1b
 972                                 WRITEBGR32
 973                         :: "r" (buf0), "r" (buf1), "r" (uvbuf0), "r" (uvbuf1), "r" (dest), "m" (dstw),
 974                         "m" (yalpha1), "m" (uvalpha1)
 975                         : "%eax"
 976                         );
 977                 }
 978                 else if(dstbpp==24)
 979                 {
 980                         asm volatile(
 981                                 YSCALEYUV2RGB1b
 982                                 WRITEBGR24
 983                         :: "r" (buf0), "r" (buf1), "r" (uvbuf0), "r" (uvbuf1), "r" (dest), "m" (dstw),
 984                         "m" (yalpha1), "m" (uvalpha1)
 985                         : "%eax", "%ebx"
 986                         );
 987                 }
 988                 else if(dstbpp==15)
 989                 {
 990                         asm volatile(
 991                                 YSCALEYUV2RGB1b
 992                 /* mm2=B, %%mm4=G, %%mm5=R, %%mm7=0 */
 993 #ifdef DITHER1XBPP
 994                                 "paddusb b16Dither, %%mm2       \n\t"
 995                                 "paddusb b16Dither, %%mm4       \n\t"
 996                                 "paddusb b16Dither, %%mm5       \n\t"
 997 #endif
 998                                 WRITEBGR15
 999                         :: "r" (buf0), "r" (buf1), "r" (uvbuf0), "r" (uvbuf1), "r" (dest), "m" (dstw),
1000                         "m" (yalpha1), "m" (uvalpha1)
1001                         : "%eax"
1002                         );
1003                 }
1004                 else if(dstbpp==16)
1005                 {
1006                         asm volatile(
1007                                 YSCALEYUV2RGB1b
1008                 /* mm2=B, %%mm4=G, %%mm5=R, %%mm7=0 */
1009 #ifdef DITHER1XBPP
1010                                 "paddusb g16Dither, %%mm2       \n\t"
1011                                 "paddusb b16Dither, %%mm4       \n\t"
1012                                 "paddusb b16Dither, %%mm5       \n\t"
1013 #endif
1014
1015                                 WRITEBGR16
1016                         :: "r" (buf0), "r" (buf1), "r" (uvbuf0), "r" (uvbuf1), "r" (dest), "m" (dstw),
1017                         "m" (yalpha1), "m" (uvalpha1)
1018                         : "%eax"
1019                         );
1020                 }
1021         }
1022 #else
1023 //FIXME write 2 versions (for even & odd lines)
1024         asm volatile ("\n\t"::: "memory");
1025
1026         if(dstbpp==32)
1027         {
1028                 int i;
1029                 for(i=0; i<dstw-1; i+=2){
1030                         // vertical linear interpolation && yuv2rgb in a single step:
1031                         int Y1=yuvtab_2568[buf0[i]>>7];
1032                         int Y2=yuvtab_2568[buf0[i+1]>>7];
1033                         int U=((uvbuf0[i>>1]*uvalpha1+uvbuf1[i>>1]*uvalpha)>>19);
1034                         int V=((uvbuf0[(i>>1)+2048]*uvalpha1+uvbuf1[(i>>1)+2048]*uvalpha)>>19);
1035
1036                         int Cb= yuvtab_40cf[U];
1037                         int Cg= yuvtab_1a1e[V] + yuvtab_0c92[U];
1038                         int Cr= yuvtab_3343[V];
1039
1040                         dest[4*i+0]=clip_table[((Y1 + Cb) >>13)];
1041                         dest[4*i+1]=clip_table[((Y1 + Cg) >>13)];
1042                         dest[4*i+2]=clip_table[((Y1 + Cr) >>13)];
1043
1044                         dest[4*i+4]=clip_table[((Y2 + Cb) >>13)];
1045                         dest[4*i+5]=clip_table[((Y2 + Cg) >>13)];
1046                         dest[4*i+6]=clip_table[((Y2 + Cr) >>13)];
1047                 }
1048         }
1049         if(dstbpp==24)
1050         {
1051                 int i;
1052                 for(i=0; i<dstw-1; i+=2){
1053                         // vertical linear interpolation && yuv2rgb in a single step:
1054                         int Y1=yuvtab_2568[buf0[i]>>7];
1055                         int Y2=yuvtab_2568[buf0[i+1]>>7];
1056                         int U=((uvbuf0[i>>1]*uvalpha1+uvbuf1[i>>1]*uvalpha)>>19);
1057                         int V=((uvbuf0[(i>>1)+2048]*uvalpha1+uvbuf1[(i>>1)+2048]*uvalpha)>>19);
1058
1059                         int Cb= yuvtab_40cf[U];
1060                         int Cg= yuvtab_1a1e[V] + yuvtab_0c92[U];
1061                         int Cr= yuvtab_3343[V];
1062
1063                         dest[0]=clip_table[((Y1 + Cb) >>13)];
1064                         dest[1]=clip_table[((Y1 + Cg) >>13)];
1065                         dest[2]=clip_table[((Y1 + Cr) >>13)];
1066
1067                         dest[3]=clip_table[((Y2 + Cb) >>13)];
1068                         dest[4]=clip_table[((Y2 + Cg) >>13)];
1069                         dest[5]=clip_table[((Y2 + Cr) >>13)];
1070                         dest+=6;
1071                 }
1072         }
1073         else if(dstbpp==16)
1074         {
1075                 int i;
1076                 for(i=0; i<dstw-1; i+=2){
1077                         // vertical linear interpolation && yuv2rgb in a single step:
1078                         int Y1=yuvtab_2568[buf0[i]>>7];
1079                         int Y2=yuvtab_2568[buf0[i+1]>>7];
1080                         int U=((uvbuf0[i>>1]*uvalpha1+uvbuf1[i>>1]*uvalpha)>>19);
1081                         int V=((uvbuf0[(i>>1)+2048]*uvalpha1+uvbuf1[(i>>1)+2048]*uvalpha)>>19);
1082
1083                         int Cb= yuvtab_40cf[U];
1084                         int Cg= yuvtab_1a1e[V] + yuvtab_0c92[U];
1085                         int Cr= yuvtab_3343[V];
1086
1087                         ((uint16_t*)dest)[i] =
1088                                 clip_table16b[(Y1 + Cb) >>13] |
1089                                 clip_table16g[(Y1 + Cg) >>13] |
1090                                 clip_table16r[(Y1 + Cr) >>13];
1091
1092                         ((uint16_t*)dest)[i+1] =
1093                                 clip_table16b[(Y2 + Cb) >>13] |
1094                                 clip_table16g[(Y2 + Cg) >>13] |
1095                                 clip_table16r[(Y2 + Cr) >>13];
1096                 }
1097         }
1098         else if(dstbpp==15)
1099         {
1100                 int i;
1101                 for(i=0; i<dstw-1; i+=2){
1102                         // vertical linear interpolation && yuv2rgb in a single step:
1103                         int Y1=yuvtab_2568[buf0[i]>>7];
1104                         int Y2=yuvtab_2568[buf0[i+1]>>7];
1105                         int U=((uvbuf0[i>>1]*uvalpha1+uvbuf1[i>>1]*uvalpha)>>19);
1106                         int V=((uvbuf0[(i>>1)+2048]*uvalpha1+uvbuf1[(i>>1)+2048]*uvalpha)>>19);
1107
1108                         int Cb= yuvtab_40cf[U];
1109                         int Cg= yuvtab_1a1e[V] + yuvtab_0c92[U];
1110                         int Cr= yuvtab_3343[V];
1111
1112                         ((uint16_t*)dest)[i] =
1113                                 clip_table15b[(Y1 + Cb) >>13] |
1114                                 clip_table15g[(Y1 + Cg) >>13] |
1115                                 clip_table15r[(Y1 + Cr) >>13];
1116
1117                         ((uint16_t*)dest)[i+1] =
1118                                 clip_table15b[(Y2 + Cb) >>13] |
1119                                 clip_table15g[(Y2 + Cg) >>13] |
1120                                 clip_table15r[(Y2 + Cr) >>13];
1121                 }
1122         }
1123 #endif
1124 }
1125
1126
1127 static inline void hyscale(uint16_t *dst, int dstWidth, uint8_t *src, int srcWidth, int xInc)
1128 {
1129       // *** horizontal scale Y line to temp buffer
1130 #ifdef ARCH_X86
1131 #ifdef HAVE_MMX2
1132         int i;
1133         if(canMMX2BeUsed)
1134         {
1135                 asm volatile(
1136                         "pxor %%mm7, %%mm7              \n\t"
1137                         "pxor %%mm2, %%mm2              \n\t" // 2*xalpha
1138                         "movd %5, %%mm6                 \n\t" // xInc&0xFFFF
1139                         "punpcklwd %%mm6, %%mm6         \n\t"
1140                         "punpcklwd %%mm6, %%mm6         \n\t"
1141                         "movq %%mm6, %%mm2              \n\t"
1142                         "psllq $16, %%mm2               \n\t"
1143                         "paddw %%mm6, %%mm2             \n\t"
1144                         "psllq $16, %%mm2               \n\t"
1145                         "paddw %%mm6, %%mm2             \n\t"
1146                         "psllq $16, %%mm2               \n\t" //0,t,2t,3t               t=xInc&0xFF
1147                         "movq %%mm2, temp0              \n\t"
1148                         "movd %4, %%mm6                 \n\t" //(xInc*4)&0xFFFF
1149                         "punpcklwd %%mm6, %%mm6         \n\t"
1150                         "punpcklwd %%mm6, %%mm6         \n\t"
1151                         "xorl %%eax, %%eax              \n\t" // i
1152                         "movl %0, %%esi                 \n\t" // src
1153                         "movl %1, %%edi                 \n\t" // buf1
1154                         "movl %3, %%edx                 \n\t" // (xInc*4)>>16
1155                         "xorl %%ecx, %%ecx              \n\t"
1156                         "xorl %%ebx, %%ebx              \n\t"
1157                         "movw %4, %%bx                  \n\t" // (xInc*4)&0xFFFF
1158
1159 #define FUNNY_Y_CODE \
1160                         PREFETCH" 1024(%%esi)           \n\t"\
1161                         PREFETCH" 1056(%%esi)           \n\t"\
1162                         PREFETCH" 1088(%%esi)           \n\t"\
1163                         "call funnyYCode                \n\t"\
1164                         "movq temp0, %%mm2              \n\t"\
1165                         "xorl %%ecx, %%ecx              \n\t"
1166
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 FUNNY_Y_CODE
1175
1176                         :: "m" (src), "m" (dst), "m" (dstWidth), "m" ((xInc*4)>>16),
1177                         "m" ((xInc*4)&0xFFFF), "m" (xInc&0xFFFF)
1178                         : "%eax", "%ebx", "%ecx", "%edx", "%esi", "%edi"
1179                 );
1180                 for(i=dstWidth-1; (i*xInc)>>16 >=srcWidth-1; i--) dst[i] = src[srcWidth-1]*128;
1181         }
1182         else
1183         {
1184 #endif
1185         //NO MMX just normal asm ...
1186         asm volatile(
1187                 "xorl %%eax, %%eax              \n\t" // i
1188                 "xorl %%ebx, %%ebx              \n\t" // xx
1189                 "xorl %%ecx, %%ecx              \n\t" // 2*xalpha
1190                 "1:                             \n\t"
1191                 "movzbl  (%0, %%ebx), %%edi     \n\t" //src[xx]
1192                 "movzbl 1(%0, %%ebx), %%esi     \n\t" //src[xx+1]
1193                 "subl %%edi, %%esi              \n\t" //src[xx+1] - src[xx]
1194                 "imull %%ecx, %%esi             \n\t" //(src[xx+1] - src[xx])*2*xalpha
1195                 "shll $16, %%edi                \n\t"
1196                 "addl %%edi, %%esi              \n\t" //src[xx+1]*2*xalpha + src[xx]*(1-2*xalpha)
1197                 "movl %1, %%edi                 \n\t"
1198                 "shrl $9, %%esi                 \n\t"
1199                 "movw %%si, (%%edi, %%eax, 2)   \n\t"
1200                 "addw %4, %%cx                  \n\t" //2*xalpha += xInc&0xFF
1201                 "adcl %3, %%ebx                 \n\t" //xx+= xInc>>8 + carry
1202
1203                 "movzbl (%0, %%ebx), %%edi      \n\t" //src[xx]
1204                 "movzbl 1(%0, %%ebx), %%esi     \n\t" //src[xx+1]
1205                 "subl %%edi, %%esi              \n\t" //src[xx+1] - src[xx]
1206                 "imull %%ecx, %%esi             \n\t" //(src[xx+1] - src[xx])*2*xalpha
1207                 "shll $16, %%edi                \n\t"
1208                 "addl %%edi, %%esi              \n\t" //src[xx+1]*2*xalpha + src[xx]*(1-2*xalpha)
1209                 "movl %1, %%edi                 \n\t"
1210                 "shrl $9, %%esi                 \n\t"
1211                 "movw %%si, 2(%%edi, %%eax, 2)  \n\t"
1212                 "addw %4, %%cx                  \n\t" //2*xalpha += xInc&0xFF
1213                 "adcl %3, %%ebx                 \n\t" //xx+= xInc>>8 + carry
1214
1215
1216                 "addl $2, %%eax                 \n\t"
1217                 "cmpl %2, %%eax                 \n\t"
1218                 " jb 1b                         \n\t"
1219
1220
1221                 :: "r" (src), "m" (dst), "m" (dstWidth), "m" (xInc>>16), "m" (xInc&0xFFFF)
1222                 : "%eax", "%ebx", "%ecx", "%edi", "%esi"
1223                 );
1224 #ifdef HAVE_MMX2
1225         } //if MMX2 cant be used
1226 #endif
1227 #else
1228         int i;
1229         unsigned int xpos=0;
1230         for(i=0;i<dstWidth;i++)
1231         {
1232                 register unsigned int xx=xpos>>16;
1233                 register unsigned int xalpha=(xpos&0xFFFF)>>9;
1234                 dst[i]= (src[xx]<<7) + (src[xx+1] - src[xx])*xalpha;
1235                 xpos+=xInc;
1236         }
1237 #endif
1238 }
1239
1240 inline static void hcscale(uint16_t *dst, int dstWidth,
1241                                 uint8_t *src1, uint8_t *src2, int srcWidth, int xInc)
1242 {
1243 #ifdef ARCH_X86
1244 #ifdef HAVE_MMX2
1245         int i;
1246         if(canMMX2BeUsed)
1247         {
1248                 asm volatile(
1249                 "pxor %%mm7, %%mm7              \n\t"
1250                 "pxor %%mm2, %%mm2              \n\t" // 2*xalpha
1251                 "movd %5, %%mm6                 \n\t" // xInc&0xFFFF
1252                 "punpcklwd %%mm6, %%mm6         \n\t"
1253                 "punpcklwd %%mm6, %%mm6         \n\t"
1254                 "movq %%mm6, %%mm2              \n\t"
1255                 "psllq $16, %%mm2               \n\t"
1256                 "paddw %%mm6, %%mm2             \n\t"
1257                 "psllq $16, %%mm2               \n\t"
1258                 "paddw %%mm6, %%mm2             \n\t"
1259                 "psllq $16, %%mm2               \n\t" //0,t,2t,3t               t=xInc&0xFFFF
1260                 "movq %%mm2, temp0              \n\t"
1261                 "movd %4, %%mm6                 \n\t" //(xInc*4)&0xFFFF
1262                 "punpcklwd %%mm6, %%mm6         \n\t"
1263                 "punpcklwd %%mm6, %%mm6         \n\t"
1264                 "xorl %%eax, %%eax              \n\t" // i
1265                 "movl %0, %%esi                 \n\t" // src
1266                 "movl %1, %%edi                 \n\t" // buf1
1267                 "movl %3, %%edx                 \n\t" // (xInc*4)>>16
1268                 "xorl %%ecx, %%ecx              \n\t"
1269                 "xorl %%ebx, %%ebx              \n\t"
1270                 "movw %4, %%bx                  \n\t" // (xInc*4)&0xFFFF
1271
1272 #define FUNNYUVCODE \
1273                         PREFETCH" 1024(%%esi)           \n\t"\
1274                         PREFETCH" 1056(%%esi)           \n\t"\
1275                         PREFETCH" 1088(%%esi)           \n\t"\
1276                         "call funnyUVCode               \n\t"\
1277                         "movq temp0, %%mm2              \n\t"\
1278                         "xorl %%ecx, %%ecx              \n\t"
1279
1280 FUNNYUVCODE
1281 FUNNYUVCODE
1282 FUNNYUVCODE
1283 FUNNYUVCODE
1284
1285 FUNNYUVCODE
1286 FUNNYUVCODE
1287 FUNNYUVCODE
1288 FUNNYUVCODE
1289                 "xorl %%eax, %%eax              \n\t" // i
1290                 "movl %6, %%esi                 \n\t" // src
1291                 "movl %1, %%edi                 \n\t" // buf1
1292                 "addl $4096, %%edi              \n\t"
1293
1294 FUNNYUVCODE
1295 FUNNYUVCODE
1296 FUNNYUVCODE
1297 FUNNYUVCODE
1298
1299 FUNNYUVCODE
1300 FUNNYUVCODE
1301 FUNNYUVCODE
1302 FUNNYUVCODE
1303
1304                 :: "m" (src1), "m" (dst), "m" (dstWidth), "m" ((xInc*4)>>16),
1305                   "m" ((xInc*4)&0xFFFF), "m" (xInc&0xFFFF), "m" (src2)
1306                 : "%eax", "%ebx", "%ecx", "%edx", "%esi", "%edi"
1307         );
1308                 for(i=dstWidth-1; (i*xInc)>>16 >=srcWidth/2-1; i--)
1309                 {
1310                         dst[i] = src1[srcWidth/2-1]*128;
1311                         dst[i+2048] = src2[srcWidth/2-1]*128;
1312                 }
1313         }
1314         else
1315         {
1316 #endif
1317         asm volatile(
1318                 "xorl %%eax, %%eax              \n\t" // i
1319                 "xorl %%ebx, %%ebx              \n\t" // xx
1320                 "xorl %%ecx, %%ecx              \n\t" // 2*xalpha
1321                 "1:                             \n\t"
1322                 "movl %0, %%esi                 \n\t"
1323                 "movzbl  (%%esi, %%ebx), %%edi  \n\t" //src[xx]
1324                 "movzbl 1(%%esi, %%ebx), %%esi  \n\t" //src[xx+1]
1325                 "subl %%edi, %%esi              \n\t" //src[xx+1] - src[xx]
1326                 "imull %%ecx, %%esi             \n\t" //(src[xx+1] - src[xx])*2*xalpha
1327                 "shll $16, %%edi                \n\t"
1328                 "addl %%edi, %%esi              \n\t" //src[xx+1]*2*xalpha + src[xx]*(1-2*xalpha)
1329                 "movl %1, %%edi                 \n\t"
1330                 "shrl $9, %%esi                 \n\t"
1331                 "movw %%si, (%%edi, %%eax, 2)   \n\t"
1332
1333                 "movzbl  (%5, %%ebx), %%edi     \n\t" //src[xx]
1334                 "movzbl 1(%5, %%ebx), %%esi     \n\t" //src[xx+1]
1335                 "subl %%edi, %%esi              \n\t" //src[xx+1] - src[xx]
1336                 "imull %%ecx, %%esi             \n\t" //(src[xx+1] - src[xx])*2*xalpha
1337                 "shll $16, %%edi                \n\t"
1338                 "addl %%edi, %%esi              \n\t" //src[xx+1]*2*xalpha + src[xx]*(1-2*xalpha)
1339                 "movl %1, %%edi                 \n\t"
1340                 "shrl $9, %%esi                 \n\t"
1341                 "movw %%si, 4096(%%edi, %%eax, 2)\n\t"
1342
1343                 "addw %4, %%cx                  \n\t" //2*xalpha += xInc&0xFF
1344                 "adcl %3, %%ebx                 \n\t" //xx+= xInc>>8 + carry
1345                 "addl $1, %%eax                 \n\t"
1346                 "cmpl %2, %%eax                 \n\t"
1347                 " jb 1b                         \n\t"
1348
1349                 :: "m" (src1), "m" (dst), "m" (dstWidth), "m" (xInc>>16), "m" (xInc&0xFFFF),
1350                 "r" (src2)
1351                 : "%eax", "%ebx", "%ecx", "%edi", "%esi"
1352                 );
1353 #ifdef HAVE_MMX2
1354         } //if MMX2 cant be used
1355 #endif
1356 #else
1357         int i;
1358         unsigned int xpos=0;
1359         for(i=0;i<dstWidth;i++)
1360         {
1361                 register unsigned int xx=xpos>>16;
1362                 register unsigned int xalpha=(xpos&0xFFFF)>>9;
1363                 dst[i]=(src1[xx]*(xalpha^127)+src1[xx+1]*xalpha);
1364                 dst[i+2048]=(src2[xx]*(xalpha^127)+src2[xx+1]*xalpha);
1365 /* slower
1366           dst[i]= (src1[xx]<<7) + (src1[xx+1] - src1[xx])*xalpha;
1367           dst[i+2048]=(src2[xx]<<7) + (src2[xx+1] - src2[xx])*xalpha;
1368 */
1369                 xpos+=xInc;
1370         }
1371 #endif
1372 }
1373
1374
1375 // *** bilinear scaling and yuv->rgb or yuv->yuv conversion of yv12 slices:
1376 // *** Note: it's called multiple times while decoding a frame, first time y==0
1377 // *** Designed to upscale, but may work for downscale too.
1378 // s_xinc = (src_width << 16) / dst_width
1379 // s_yinc = (src_height << 16) / dst_height
1380 void SwScale_YV12slice(unsigned char* srcptr[],int stride[], int y, int h,
1381                              uint8_t* dstptr[], int dststride, int dstw, int dstbpp,
1382                              unsigned int s_xinc,unsigned int s_yinc){
1383
1384 // scaling factors:
1385 //static int s_yinc=(vo_dga_src_height<<16)/vo_dga_vp_height;
1386 //static int s_xinc=(vo_dga_src_width<<8)/vo_dga_vp_width;
1387
1388 unsigned int s_xinc2;
1389
1390 static int s_srcypos; // points to the dst Pixels center in the source (0 is the center of pixel 0,0 in src)
1391 static int s_ypos;
1392
1393 // last horzontally interpolated lines, used to avoid unnecessary calculations
1394 static int s_last_ypos;
1395 static int s_last_y1pos;
1396
1397 #ifdef HAVE_MMX2
1398 // used to detect a horizontal size change
1399 static int old_dstw= -1;
1400 static int old_s_xinc= -1;
1401 #endif
1402
1403 int srcWidth;
1404 int dstUVw;
1405 int i;
1406
1407 if(((dstw + 7)&(~7)) >= dststride) dstw&= ~7;
1408
1409 srcWidth= (dstw*s_xinc + 0x8000)>>16;
1410 dstUVw= fullUVIpol ? dstw : dstw/2;
1411
1412 #ifdef HAVE_MMX2
1413 canMMX2BeUsed= (s_xinc <= 0x10000 && (dstw&31)==0 && (srcWidth&15)==0) ? 1 : 0;
1414 #endif
1415
1416 // match pixel 0 of the src to pixel 0 of dst and match pixel n-2 of src to pixel n-2 of dst
1417 // n-2 is the last chrominance sample available
1418 // FIXME this is not perfect, but noone shuld notice the difference, the more correct variant
1419 // would be like the vertical one, but that would require some special code for the
1420 // first and last pixel
1421 if(canMMX2BeUsed)       s_xinc+= 20;
1422 else                    s_xinc = ((srcWidth-2)<<16)/(dstw-2) - 20;
1423
1424 if(fullUVIpol && !(dstbpp==12))         s_xinc2= s_xinc>>1;
1425 else                                    s_xinc2= s_xinc;
1426   // force calculation of the horizontal interpolation of the first line
1427
1428   if(y==0){
1429 //      printf("dstw %d, srcw %d, mmx2 %d\n", dstw, srcWidth, canMMX2BeUsed);
1430         s_last_ypos=-99;
1431         s_last_y1pos=-99;
1432         s_srcypos= s_yinc/2 - 0x8000;
1433         s_ypos=0;
1434
1435         // clean the buffers so that no green stuff is drawen if the width is not sane (%8=0)
1436         for(i=dstw-2; i<dstw+20; i++)
1437         {
1438                 pix_buf_uv[0][i] = pix_buf_uv[1][i]
1439                 = pix_buf_uv[0][2048+i] = pix_buf_uv[1][2048+i] = 128*128;
1440                 pix_buf_uv[0][i/2] = pix_buf_uv[1][i/2]
1441                 = pix_buf_uv[0][2048+i/2] = pix_buf_uv[1][2048+i/2] = 128*128;
1442                 pix_buf_y[0][i]= pix_buf_y[1][i]= 0;
1443         }
1444
1445 #ifdef HAVE_MMX2
1446 // cant downscale !!!
1447         if((old_s_xinc != s_xinc || old_dstw!=dstw) && canMMX2BeUsed)
1448         {
1449                 uint8_t *fragment;
1450                 int imm8OfPShufW1;
1451                 int imm8OfPShufW2;
1452                 int fragmentLength;
1453
1454                 int xpos, i;
1455
1456                 old_s_xinc= s_xinc;
1457                 old_dstw= dstw;
1458
1459                 // create an optimized horizontal scaling routine
1460
1461                 //code fragment
1462
1463                 asm volatile(
1464                         "jmp 9f                         \n\t"
1465                 // Begin
1466                         "0:                             \n\t"
1467                         "movq (%%esi), %%mm0            \n\t" //FIXME Alignment
1468                         "movq %%mm0, %%mm1              \n\t"
1469                         "psrlq $8, %%mm0                \n\t"
1470                         "punpcklbw %%mm7, %%mm1 \n\t"
1471                         "movq %%mm2, %%mm3              \n\t"
1472                         "punpcklbw %%mm7, %%mm0 \n\t"
1473                         "addw %%bx, %%cx                \n\t" //2*xalpha += (4*s_xinc)&0xFFFF
1474                         "pshufw $0xFF, %%mm1, %%mm1     \n\t"
1475                         "1:                             \n\t"
1476                         "adcl %%edx, %%esi              \n\t" //xx+= (4*s_xinc)>>16 + carry
1477                         "pshufw $0xFF, %%mm0, %%mm0     \n\t"
1478                         "2:                             \n\t"
1479                         "psrlw $9, %%mm3                \n\t"
1480                         "psubw %%mm1, %%mm0             \n\t"
1481                         "pmullw %%mm3, %%mm0            \n\t"
1482                         "paddw %%mm6, %%mm2             \n\t" // 2*alpha += xpos&0xFFFF
1483                         "psllw $7, %%mm1                \n\t"
1484                         "paddw %%mm1, %%mm0             \n\t"
1485
1486                         "movq %%mm0, (%%edi, %%eax)     \n\t"
1487
1488                         "addl $8, %%eax                 \n\t"
1489                 // End
1490                         "9:                             \n\t"
1491 //              "int $3\n\t"
1492                         "leal 0b, %0                    \n\t"
1493                         "leal 1b, %1                    \n\t"
1494                         "leal 2b, %2                    \n\t"
1495                         "decl %1                        \n\t"
1496                         "decl %2                        \n\t"
1497                         "subl %0, %1                    \n\t"
1498                         "subl %0, %2                    \n\t"
1499                         "leal 9b, %3                    \n\t"
1500                         "subl %0, %3                    \n\t"
1501                         :"=r" (fragment), "=r" (imm8OfPShufW1), "=r" (imm8OfPShufW2),
1502                          "=r" (fragmentLength)
1503                 );
1504
1505                 xpos= 0; //s_xinc/2 - 0x8000; // difference between pixel centers
1506
1507                 /* choose xinc so that all 8 parts fit exactly
1508                    Note: we cannot use just 1 part because it would not fit in the code cache */
1509 //              s_xinc2_diff= -((((s_xinc2*(dstw/8))&0xFFFF))/(dstw/8))-10;
1510 //              s_xinc_diff= -((((s_xinc*(dstw/8))&0xFFFF))/(dstw/8));
1511 #ifdef ALT_ERROR
1512 //              s_xinc2_diff+= ((0x10000/(dstw/8)));
1513 #endif
1514 //              s_xinc_diff= s_xinc2_diff*2;
1515
1516 //              s_xinc2+= s_xinc2_diff;
1517 //              s_xinc+= s_xinc_diff;
1518
1519 //              old_s_xinc= s_xinc;
1520
1521                 for(i=0; i<dstw/8; i++)
1522                 {
1523                         int xx=xpos>>16;
1524
1525                         if((i&3) == 0)
1526                         {
1527                                 int a=0;
1528                                 int b=((xpos+s_xinc)>>16) - xx;
1529                                 int c=((xpos+s_xinc*2)>>16) - xx;
1530                                 int d=((xpos+s_xinc*3)>>16) - xx;
1531
1532                                 memcpy(funnyYCode + fragmentLength*i/4, fragment, fragmentLength);
1533
1534                                 funnyYCode[fragmentLength*i/4 + imm8OfPShufW1]=
1535                                 funnyYCode[fragmentLength*i/4 + imm8OfPShufW2]=
1536                                         a | (b<<2) | (c<<4) | (d<<6);
1537
1538                                 // if we dont need to read 8 bytes than dont :), reduces the chance of
1539                                 // crossing a cache line
1540                                 if(d<3) funnyYCode[fragmentLength*i/4 + 1]= 0x6E;
1541
1542                                 funnyYCode[fragmentLength*(i+4)/4]= RET;
1543                         }
1544                         xpos+=s_xinc;
1545                 }
1546
1547                 xpos= 0; //s_xinc2/2 - 0x10000; // difference between centers of chrom samples
1548                 for(i=0; i<dstUVw/8; i++)
1549                 {
1550                         int xx=xpos>>16;
1551
1552                         if((i&3) == 0)
1553                         {
1554                                 int a=0;
1555                                 int b=((xpos+s_xinc2)>>16) - xx;
1556                                 int c=((xpos+s_xinc2*2)>>16) - xx;
1557                                 int d=((xpos+s_xinc2*3)>>16) - xx;
1558
1559                                 memcpy(funnyUVCode + fragmentLength*i/4, fragment, fragmentLength);
1560
1561                                 funnyUVCode[fragmentLength*i/4 + imm8OfPShufW1]=
1562                                 funnyUVCode[fragmentLength*i/4 + imm8OfPShufW2]=
1563                                         a | (b<<2) | (c<<4) | (d<<6);
1564
1565                                 // if we dont need to read 8 bytes than dont :), reduces the chance of
1566                                 // crossing a cache line
1567                                 if(d<3) funnyUVCode[fragmentLength*i/4 + 1]= 0x6E;
1568
1569                                 funnyUVCode[fragmentLength*(i+4)/4]= RET;
1570                         }
1571                         xpos+=s_xinc2;
1572                 }
1573 //              funnyCode[0]= RET;
1574         }
1575
1576 #endif // HAVE_MMX2
1577   } // reset counters
1578
1579   while(1){
1580     unsigned char *dest =dstptr[0]+dststride*s_ypos;
1581     unsigned char *uDest=dstptr[1]+(dststride>>1)*(s_ypos>>1);
1582     unsigned char *vDest=dstptr[2]+(dststride>>1)*(s_ypos>>1);
1583
1584     int y0=(s_srcypos + 0xFFFF)>>16;  // first luminance source line number below the dst line
1585         // points to the dst Pixels center in the source (0 is the center of pixel 0,0 in src)
1586     int srcuvpos= dstbpp==12 ?  s_srcypos + s_yinc/2 - 0x8000 :
1587                                 s_srcypos - 0x8000;
1588     int y1=(srcuvpos + 0x1FFFF)>>17; // first chrominance source line number below the dst line
1589     int yalpha=((s_srcypos-1)&0xFFFF)>>4;
1590     int uvalpha=((srcuvpos-1)&0x1FFFF)>>5;
1591     uint16_t *buf0=pix_buf_y[y0&1];             // top line of the interpolated slice
1592     uint16_t *buf1=pix_buf_y[((y0+1)&1)];       // bottom line of the interpolated slice
1593     uint16_t *uvbuf0=pix_buf_uv[y1&1];          // top line of the interpolated slice
1594     uint16_t *uvbuf1=pix_buf_uv[(y1+1)&1];      // bottom line of the interpolated slice
1595
1596     if(y0>=y+h) break; // FIXME wrong, skips last lines, but they are dupliactes anyway
1597
1598     if((y0&1) && dstbpp==12) uvalpha=-1; // there is no alpha if there is no line
1599
1600     s_ypos++; s_srcypos+=s_yinc;
1601
1602     //only interpolate the src line horizontally if we didnt do it allready
1603         if(s_last_ypos!=y0)
1604         {
1605                 unsigned char *src;
1606                 // skip if first line has been horiz scaled alleady
1607                 if(s_last_ypos != y0-1)
1608                 {
1609                         // check if first line is before any available src lines
1610                         if(y0-1 < y)    src=srcptr[0]+(0     )*stride[0];
1611                         else            src=srcptr[0]+(y0-y-1)*stride[0];
1612
1613                         hyscale(buf0, dstw, src, srcWidth, s_xinc);
1614                 }
1615                 // check if second line is after any available src lines
1616                 if(y0-y >= h)   src=srcptr[0]+(h-1)*stride[0];
1617                 else            src=srcptr[0]+(y0-y)*stride[0];
1618
1619                 // the min() is required to avoid reuseing lines which where not available
1620                 s_last_ypos= MIN(y0, y+h-1);
1621                 hyscale(buf1, dstw, src, srcWidth, s_xinc);
1622         }
1623 //      printf("%d %d %d %d\n", y, y1, s_last_y1pos, h);
1624       // *** horizontal scale U and V lines to temp buffer
1625         if(s_last_y1pos!=y1)
1626         {
1627                 uint8_t *src1, *src2;
1628                 // skip if first line has been horiz scaled alleady
1629                 if(s_last_y1pos != y1-1)
1630                 {
1631                         // check if first line is before any available src lines
1632                         if(y1-y/2-1 < 0)
1633                         {
1634                                 src1= srcptr[1]+(0)*stride[1];
1635                                 src2= srcptr[2]+(0)*stride[2];
1636                         }else{
1637                                 src1= srcptr[1]+(y1-y/2-1)*stride[1];
1638                                 src2= srcptr[2]+(y1-y/2-1)*stride[2];
1639                         }
1640                         hcscale(uvbuf0, dstUVw, src1, src2, srcWidth, s_xinc2);
1641                 }
1642
1643                 // check if second line is after any available src lines
1644                 if(y1 - y/2 >= h/2)
1645                 {
1646                         src1= srcptr[1]+(h/2-1)*stride[1];
1647                         src2= srcptr[2]+(h/2-1)*stride[2];
1648                 }else{
1649                         src1= srcptr[1]+(y1-y/2)*stride[1];
1650                         src2= srcptr[2]+(y1-y/2)*stride[2];
1651                 }
1652                 hcscale(uvbuf1, dstUVw, src1, src2, srcWidth, s_xinc2);
1653
1654                 // the min() is required to avoid reuseing lines which where not available
1655                 s_last_y1pos= MIN(y1, y/2+h/2-1);
1656         }
1657
1658         if(dstbpp==12) //YV12
1659                 yuv2yuv(buf0, buf1, uvbuf0, uvbuf1, dest, uDest, vDest, dstw, yalpha, uvalpha);
1660         else if(ABS(s_yinc - 0x10000) < 10)
1661                 yuv2rgb1(buf0, buf1, uvbuf0, uvbuf1, dest, dstw, yalpha, uvalpha, dstbpp);
1662         else
1663                 yuv2rgbX(buf0, buf1, uvbuf0, uvbuf1, dest, dstw, yalpha, uvalpha, dstbpp);
1664
1665 #ifdef HAVE_MMX
1666         b16Dither= b16Dither1;
1667         b16Dither1= b16Dither2;
1668         b16Dither2= b16Dither;
1669
1670         g16Dither= g16Dither1;
1671         g16Dither1= g16Dither2;
1672         g16Dither2= g16Dither;
1673 #endif
1674   }
1675
1676 #ifdef HAVE_MMX
1677         __asm __volatile(SFENCE:::"memory");
1678         __asm __volatile(EMMS:::"memory");
1679 #endif
1680 }
1681
1682
1683 void SwScale_Init(){
1684     // generating tables:
1685     int i;
1686     for(i=0;i<256;i++){
1687         clip_table[i]=0;
1688         clip_table[i+256]=i;
1689         clip_table[i+512]=255;
1690         yuvtab_2568[i]=(0x2568*(i-16))+(256<<13);
1691         yuvtab_3343[i]=0x3343*(i-128);
1692         yuvtab_0c92[i]=-0x0c92*(i-128);
1693         yuvtab_1a1e[i]=-0x1a1e*(i-128);
1694         yuvtab_40cf[i]=0x40cf*(i-128);
1695     }
1696
1697     for(i=0; i<768; i++)
1698     {
1699         int v= clip_table[i];
1700         clip_table16b[i]= v>>3;
1701         clip_table16g[i]= (v<<3)&0x07E0;
1702         clip_table16r[i]= (v<<8)&0xF800;
1703         clip_table15b[i]= v>>3;
1704         clip_table15g[i]= (v<<2)&0x03E0;
1705         clip_table15r[i]= (v<<7)&0x7C00;
1706     }
1707
1708 }