git.sesse.net Git - vlc/blob - modules/video_chroma/i420_rgb_mmx.h

   1 /*****************************************************************************
   2  * i420_rgb_mmx.h: MMX YUV transformation assembly
   3  *****************************************************************************
   4  * Copyright (C) 1999-2007 the VideoLAN team
   5  * $Id$
   6  *
   7  * Authors: Olie Lho <ollie@sis.com.tw>
   8  *          Gaël Hendryckx <jimmy@via.ecp.fr>
   9  *          Samuel Hocevar <sam@zoy.org>
  10  *          Damien Fouilleul <damienf@videolan.org>
  11  *
  12  * This program is free software; you can redistribute it and/or modify
  13  * it under the terms of the GNU General Public License as published by
  14  * the Free Software Foundation; either version 2 of the License, or
  15  * (at your option) any later version.
  16  *
  17  * This program is distributed in the hope that it will be useful,
  18  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  19  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
  20  * GNU General Public License for more details.
  21  *
  22  * You should have received a copy of the GNU General Public License
  23  * along with this program; if not, write to the Free Software
  24  * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston MA 02110-1301, USA.
  25  *****************************************************************************/
  26
  27 /* hope these constant values are cache line aligned */
  28 static const uint64_t mmx_80w     = 0x0080008000800080ULL; /* Will be referenced as %4 in inline asm */
  29 static const uint64_t mmx_10w     = 0x1010101010101010ULL; /* -- as %5 */
  30 static const uint64_t mmx_00ffw   = 0x00ff00ff00ff00ffULL; /* -- as %6 */
  31 static const uint64_t mmx_Y_coeff = 0x253f253f253f253fULL; /* -- as %7 */
  32
  33 static const uint64_t mmx_U_green = 0xf37df37df37df37dULL; /* -- as %8 */
  34 static const uint64_t mmx_U_blue  = 0x4093409340934093ULL; /* -- as %9 */
  35 static const uint64_t mmx_V_red   = 0x3312331233123312ULL; /* -- as %10 */
  36 static const uint64_t mmx_V_green = 0xe5fce5fce5fce5fcULL; /* -- as %11 */
  37
  38 static const uint64_t mmx_mask_f8 = 0xf8f8f8f8f8f8f8f8ULL; /* -- as %12 */
  39 static const uint64_t mmx_mask_fc = 0xfcfcfcfcfcfcfcfcULL; /* -- as %13 */
  40
  41 #if defined(CAN_COMPILE_MMX)
  42
  43 /* MMX assembly */
  44
  45 #define MMX_CALL(MMX_INSTRUCTIONS)      \
  46     do {                                \
  47     __asm__ __volatile__(               \
  48         ".p2align 3 \n\t"               \
  49         MMX_INSTRUCTIONS                \
  50         :                               \
  51         : "r" (p_y), "r" (p_u),         \
  52           "r" (p_v), "r" (p_buffer),    \
  53           "m" (mmx_80w), "m" (mmx_10w), \
  54           "m" (mmx_00ffw), "m" (mmx_Y_coeff), \
  55           "m" (mmx_U_green), "m" (mmx_U_blue), \
  56           "m" (mmx_V_red), "m" (mmx_V_green), \
  57           "m" (mmx_mask_f8), "m" (mmx_mask_fc) \
  58         : "mm0", "mm1", "mm2", "mm3", "mm4", "mm5", "mm6", "mm7" );  \
  59     } while(0)
  60
  61 #define MMX_END __asm__ __volatile__ ( "emms" )
  62
  63 #define MMX_INIT_16 "                                                       \n\
  64 movd       (%1), %%mm0      # Load 4 Cb       00 00 00 00 u3 u2 u1 u0       \n\
  65 movd       (%2), %%mm1      # Load 4 Cr       00 00 00 00 v3 v2 v1 v0       \n\
  66 pxor      %%mm4, %%mm4      # zero mm4                                      \n\
  67 movq       (%0), %%mm6      # Load 8 Y        Y7 Y6 Y5 Y4 Y3 Y2 Y1 Y0       \n\
  68 "
  69
  70 #define MMX_INIT_16_GRAY "                                                  \n\
  71 movq      (%0), %%mm6       # Load 8 Y        Y7 Y6 Y5 Y4 Y3 Y2 Y1 Y0       \n\
  72 #movl      $0, (%3)         # cache preload for image                       \n\
  73 "
  74
  75 #define MMX_INIT_32 "                                                       \n\
  76 movd      (%1), %%mm0       # Load 4 Cb       00 00 00 00 u3 u2 u1 u0       \n\
  77 movl        $0, (%3)        # cache preload for image                       \n\
  78 movd      (%2), %%mm1       # Load 4 Cr       00 00 00 00 v3 v2 v1 v0       \n\
  79 pxor     %%mm4, %%mm4       # zero mm4                                      \n\
  80 movq      (%0), %%mm6       # Load 8 Y        Y7 Y6 Y5 Y4 Y3 Y2 Y1 Y0       \n\
  81 "
  82
  83 /*
  84  * Do the multiply part of the conversion for even and odd pixels,
  85  * register usage:
  86  * mm0 -> Cblue, mm1 -> Cred, mm2 -> Cgreen even pixels,
  87  * mm3 -> Cblue, mm4 -> Cred, mm5 -> Cgreen odd  pixels,
  88  * mm6 -> Y even, mm7 -> Y odd
  89  */
  90
  91 #define MMX_YUV_MUL "                                                       \n\
  92 # convert the chroma part                                                   \n\
  93 punpcklbw %%mm4, %%mm0          # scatter 4 Cb    00 u3 00 u2 00 u1 00 u0   \n\
  94 punpcklbw %%mm4, %%mm1          # scatter 4 Cr    00 v3 00 v2 00 v1 00 v0   \n\
  95 psubsw    %4, %%mm0     # Cb -= 128                                 \n\
  96 psubsw    %4, %%mm1     # Cr -= 128                                 \n\
  97 psllw     $3, %%mm0             # Promote precision                         \n\
  98 psllw     $3, %%mm1             # Promote precision                         \n\
  99 movq      %%mm0, %%mm2          # Copy 4 Cb       00 u3 00 u2 00 u1 00 u0   \n\
 100 movq      %%mm1, %%mm3          # Copy 4 Cr       00 v3 00 v2 00 v1 00 v0   \n\
 101 pmulhw    %8, %%mm2 # Mul Cb with green coeff -> Cb green       \n\
 102 pmulhw    %11, %%mm3 # Mul Cr with green coeff -> Cr green       \n\
 103 pmulhw    %9, %%mm0  # Mul Cb -> Cblue 00 b3 00 b2 00 b1 00 b0   \n\
 104 pmulhw    %10, %%mm1   # Mul Cr -> Cred  00 r3 00 r2 00 r1 00 r0   \n\
 105 paddsw    %%mm3, %%mm2          # Cb green + Cr green -> Cgreen             \n\
 106                                                                             \n\
 107 # convert the luma part                                                     \n\
 108 psubusb   %5, %%mm6     # Y -= 16                                   \n\
 109 movq      %%mm6, %%mm7          # Copy 8 Y        Y7 Y6 Y5 Y4 Y3 Y2 Y1 Y0   \n\
 110 pand      %6, %%mm6   # get Y even      00 Y6 00 Y4 00 Y2 00 Y0   \n\
 111 psrlw     $8, %%mm7             # get Y odd       00 Y7 00 Y5 00 Y3 00 Y1   \n\
 112 psllw     $3, %%mm6             # Promote precision                         \n\
 113 psllw     $3, %%mm7             # Promote precision                         \n\
 114 pmulhw    %7, %%mm6 # Mul 4 Y even    00 y6 00 y4 00 y2 00 y0   \n\
 115 pmulhw    %7, %%mm7 # Mul 4 Y odd     00 y7 00 y5 00 y3 00 y1   \n\
 116 "
 117
 118 /*
 119  * Do the addition part of the conversion for even and odd pixels,
 120  * register usage:
 121  * mm0 -> Cblue, mm1 -> Cred, mm2 -> Cgreen even pixels,
 122  * mm3 -> Cblue, mm4 -> Cred, mm5 -> Cgreen odd  pixels,
 123  * mm6 -> Y even, mm7 -> Y odd
 124  */
 125
 126 #define MMX_YUV_ADD "                                                       \n\
 127 # Do horizontal and vertical scaling                                        \n\
 128 movq      %%mm0, %%mm3          # Copy Cblue                                \n\
 129 movq      %%mm1, %%mm4          # Copy Cred                                 \n\
 130 movq      %%mm2, %%mm5          # Copy Cgreen                               \n\
 131 paddsw    %%mm6, %%mm0          # Y even + Cblue  00 B6 00 B4 00 B2 00 B0   \n\
 132 paddsw    %%mm7, %%mm3          # Y odd  + Cblue  00 B7 00 B5 00 B3 00 B1   \n\
 133 paddsw    %%mm6, %%mm1          # Y even + Cred   00 R6 00 R4 00 R2 00 R0   \n\
 134 paddsw    %%mm7, %%mm4          # Y odd  + Cred   00 R7 00 R5 00 R3 00 R1   \n\
 135 paddsw    %%mm6, %%mm2          # Y even + Cgreen 00 G6 00 G4 00 G2 00 G0   \n\
 136 paddsw    %%mm7, %%mm5          # Y odd  + Cgreen 00 G7 00 G5 00 G3 00 G1   \n\
 137                                                                             \n\
 138 # Limit RGB even to 0..255                                                  \n\
 139 packuswb  %%mm0, %%mm0          # B6 B4 B2 B0 / B6 B4 B2 B0                 \n\
 140 packuswb  %%mm1, %%mm1          # R6 R4 R2 R0 / R6 R4 R2 R0                 \n\
 141 packuswb  %%mm2, %%mm2          # G6 G4 G2 G0 / G6 G4 G2 G0                 \n\
 142                                                                             \n\
 143 # Limit RGB odd to 0..255                                                   \n\
 144 packuswb  %%mm3, %%mm3          # B7 B5 B3 B1 / B7 B5 B3 B1                 \n\
 145 packuswb  %%mm4, %%mm4          # R7 R5 R3 R1 / R7 R5 R3 R1                 \n\
 146 packuswb  %%mm5, %%mm5          # G7 G5 G3 G1 / G7 G5 G3 G1                 \n\
 147                                                                             \n\
 148 # Interleave RGB even and odd                                               \n\
 149 punpcklbw %%mm3, %%mm0          #                 B7 B6 B5 B4 B3 B2 B1 B0   \n\
 150 punpcklbw %%mm4, %%mm1          #                 R7 R6 R5 R4 R3 R2 R1 R0   \n\
 151 punpcklbw %%mm5, %%mm2          #                 G7 G6 G5 G4 G3 G2 G1 G0   \n\
 152 "
 153
 154 /*
 155  * Grayscale case, only use Y
 156  */
 157
 158 #define MMX_YUV_GRAY "                                                      \n\
 159 # convert the luma part                                                     \n\
 160 psubusb   %5, %%mm6                                                 \n\
 161 movq      %%mm6, %%mm7                                                      \n\
 162 pand      %6, %%mm6                                               \n\
 163 psrlw     $8, %%mm7                                                         \n\
 164 psllw     $3, %%mm6                                                         \n\
 165 psllw     $3, %%mm7                                                         \n\
 166 pmulhw    %7, %%mm6                                             \n\
 167 pmulhw    %7, %%mm7                                             \n\
 168 packuswb  %%mm6, %%mm6                                                      \n\
 169 packuswb  %%mm7, %%mm7                                                      \n\
 170 punpcklbw %%mm7, %%mm6                                                      \n\
 171 "
 172
 173 #define MMX_UNPACK_16_GRAY "                                                \n\
 174 movq      %%mm6, %%mm5                                                      \n\
 175 pand      %12, %%mm6                                             \n\
 176 pand      %13, %%mm5                                             \n\
 177 movq      %%mm6, %%mm7                                                      \n\
 178 psrlw     $3, %%mm7                                                         \n\
 179 pxor      %%mm3, %%mm3                                                      \n\
 180 movq      %%mm7, %%mm2                                                      \n\
 181 movq      %%mm5, %%mm0                                                      \n\
 182 punpcklbw %%mm3, %%mm5                                                      \n\
 183 punpcklbw %%mm6, %%mm7                                                      \n\
 184 psllw     $3, %%mm5                                                         \n\
 185 por       %%mm5, %%mm7                                                      \n\
 186 movq      %%mm7, (%3)                                                       \n\
 187 punpckhbw %%mm3, %%mm0                                                      \n\
 188 punpckhbw %%mm6, %%mm2                                                      \n\
 189 psllw     $3, %%mm0                                                         \n\
 190 movq      8(%0), %%mm6                                                      \n\
 191 por       %%mm0, %%mm2                                                      \n\
 192 movq      %%mm2, 8(%3)                                                      \n\
 193 "
 194
 195
 196 /*
 197  * convert RGB plane to RGB 15 bits,
 198  * mm0 -> B, mm1 -> R, mm2 -> G,
 199  * mm4 -> GB, mm5 -> AR pixel 4-7,
 200  * mm6 -> GB, mm7 -> AR pixel 0-3
 201  */
 202
 203 #define MMX_UNPACK_15 "                                                     \n\
 204 # mask unneeded bits off                                                    \n\
 205 pand      %12, %%mm0 # b7b6b5b4 b3______ b7b6b5b4 b3______       \n\
 206 psrlw     $3,%%mm0              # ______b7 b6b5b4b3 ______b7 b6b5b4b3       \n\
 207 pand      %12, %%mm2 # g7g6g5g4 g3______ g7g6g5g4 g3______       \n\
 208 pand      %12, %%mm1 # r7r6r5r4 r3______ r7r6r5r4 r3______       \n\
 209 psrlw     $1,%%mm1              # __r7r6r5 r4r3____ __r7r6r5 r4r3____       \n\
 210 pxor      %%mm4, %%mm4          # zero mm4                                  \n\
 211 movq      %%mm0, %%mm5          # Copy B7-B0                                \n\
 212 movq      %%mm2, %%mm7          # Copy G7-G0                                \n\
 213                                                                             \n\
 214 # convert rgb24 plane to rgb15 pack for pixel 0-3                           \n\
 215 punpcklbw %%mm4, %%mm2          # ________ ________ g7g6g5g4 g3______       \n\
 216 punpcklbw %%mm1, %%mm0          # r7r6r5r4 r3______ ______b7 b6b5b4b3       \n\
 217 psllw     $2,%%mm2              # ________ ____g7g6 g5g4g3__ ________       \n\
 218 por       %%mm2, %%mm0          # r7r6r5r4 r3__g7g6 g5g4g3b7 b6b5b4b3       \n\
 219 movq      8(%0), %%mm6          # Load 8 Y        Y7 Y6 Y5 Y4 Y3 Y2 Y1 Y0   \n\
 220 movq      %%mm0, (%3)           # store pixel 0-3                           \n\
 221                                                                             \n\
 222 # convert rgb24 plane to rgb16 pack for pixel 0-3                           \n\
 223 punpckhbw %%mm4, %%mm7          # ________ ________ g7g6g5g4 g3______       \n\
 224 punpckhbw %%mm1, %%mm5          # r7r6r5r4 r3______ ______b7 b6b5b4b3       \n\
 225 psllw     $2,%%mm7              # ________ ____g7g6 g5g4g3__ ________       \n\
 226 movd      4(%1), %%mm0          # Load 4 Cb       __ __ __ __ u3 u2 u1 u0   \n\
 227 por       %%mm7, %%mm5          # r7r6r5r4 r3__g7g6 g5g4g3b7 b6b5b4b3       \n\
 228 movd      4(%2), %%mm1          # Load 4 Cr       __ __ __ __ v3 v2 v1 v0   \n\
 229 movq      %%mm5, 8(%3)          # store pixel 4-7                           \n\
 230 "
 231
 232 /*
 233  * convert RGB plane to RGB 16 bits,
 234  * mm0 -> B, mm1 -> R, mm2 -> G,
 235  * mm4 -> GB, mm5 -> AR pixel 4-7,
 236  * mm6 -> GB, mm7 -> AR pixel 0-3
 237  */
 238
 239 #define MMX_UNPACK_16 "                                                     \n\
 240 # mask unneeded bits off                                                    \n\
 241 pand      %12, %%mm0 # b7b6b5b4 b3______ b7b6b5b4 b3______       \n\
 242 pand      %13, %%mm2 # g7g6g5g4 g3g2____ g7g6g5g4 g3g2____       \n\
 243 pand      %12, %%mm1 # r7r6r5r4 r3______ r7r6r5r4 r3______       \n\
 244 psrlw     $3,%%mm0              # ______b7 b6b5b4b3 ______b7 b6b5b4b3       \n\
 245 pxor      %%mm4, %%mm4          # zero mm4                                  \n\
 246 movq      %%mm0, %%mm5          # Copy B7-B0                                \n\
 247 movq      %%mm2, %%mm7          # Copy G7-G0                                \n\
 248                                                                             \n\
 249 # convert rgb24 plane to rgb16 pack for pixel 0-3                           \n\
 250 punpcklbw %%mm4, %%mm2          # ________ ________ g7g6g5g4 g3g2____       \n\
 251 punpcklbw %%mm1, %%mm0          # r7r6r5r4 r3______ ______b7 b6b5b4b3       \n\
 252 psllw     $3,%%mm2              # ________ __g7g6g5 g4g3g2__ ________       \n\
 253 por       %%mm2, %%mm0          # r7r6r5r4 r3g7g6g5 g4g3g2b7 b6b5b4b3       \n\
 254 movq      8(%0), %%mm6          # Load 8 Y        Y7 Y6 Y5 Y4 Y3 Y2 Y1 Y0   \n\
 255 movq      %%mm0, (%3)           # store pixel 0-3                           \n\
 256                                                                             \n\
 257 # convert rgb24 plane to rgb16 pack for pixel 0-3                           \n\
 258 punpckhbw %%mm4, %%mm7          # ________ ________ g7g6g5g4 g3g2____       \n\
 259 punpckhbw %%mm1, %%mm5          # r7r6r5r4 r3______ ______b7 b6b5b4b3       \n\
 260 psllw     $3,%%mm7              # ________ __g7g6g5 g4g3g2__ ________       \n\
 261 movd      4(%1), %%mm0          # Load 4 Cb       __ __ __ __ u3 u2 u1 u0   \n\
 262 por       %%mm7, %%mm5          # r7r6r5r4 r3g7g6g5 g4g3g2b7 b6b5b4b3       \n\
 263 movd      4(%2), %%mm1          # Load 4 Cr       __ __ __ __ v3 v2 v1 v0   \n\
 264 movq      %%mm5, 8(%3)          # store pixel 4-7                           \n\
 265 "
 266
 267 /*
 268  * convert RGB plane to RGB packed format,
 269  * mm0 -> B, mm1 -> R, mm2 -> G
 270  */
 271
 272 #define MMX_UNPACK_32_ARGB "                                                \n\
 273 pxor      %%mm3, %%mm3  # zero mm3                                          \n\
 274 movq      %%mm0, %%mm4  #                 B7 B6 B5 B4 B3 B2 B1 B0           \n\
 275 punpcklbw %%mm2, %%mm4  #                 G3 B3 G2 B2 G1 B1 G0 B0           \n\
 276 movq      %%mm1, %%mm5  #                 R7 R6 R5 R4 R3 R2 R1 R0           \n\
 277 punpcklbw %%mm3, %%mm5  #                 00 R3 00 R2 00 R1 00 R0           \n\
 278 movq      %%mm4, %%mm6  #                 G3 B3 G2 B2 G1 B1 G0 B0           \n\
 279 punpcklwd %%mm5, %%mm4  #                 00 R1 B1 G1 00 R0 B0 G0           \n\
 280 movq      %%mm4, (%3)   # Store ARGB1 ARGB0                                 \n\
 281 punpckhwd %%mm5, %%mm6  #                 00 R3 B3 G3 00 R2 B2 G2           \n\
 282 movq      %%mm6, 8(%3)  # Store ARGB3 ARGB2                                 \n\
 283 punpckhbw %%mm2, %%mm0  #                 G7 B7 G6 B6 G5 B5 G4 B4           \n\
 284 punpckhbw %%mm3, %%mm1  #                 00 R7 00 R6 00 R5 00 R4           \n\
 285 movq      %%mm0, %%mm5  #                 G7 B7 G6 B6 G5 B5 G4 B4           \n\
 286 punpcklwd %%mm1, %%mm5  #                 00 R5 B5 G5 00 R4 B4 G4           \n\
 287 movq      %%mm5, 16(%3) # Store ARGB5 ARGB4                                 \n\
 288 punpckhwd %%mm1, %%mm0  #                 00 R7 B7 G7 00 R6 B6 G6           \n\
 289 movq      %%mm0, 24(%3) # Store ARGB7 ARGB6                                 \n\
 290 "
 291
 292 #define MMX_UNPACK_32_RGBA "                                                \n\
 293 pxor      %%mm3, %%mm3  # zero mm3                                          \n\
 294 movq      %%mm2, %%mm4  #                 G7 G6 G5 G4 G3 G2 G1 G0           \n\
 295 punpcklbw %%mm1, %%mm4  #                 R3 G3 R2 G2 R1 G1 R0 G0           \n\
 296 punpcklbw %%mm0, %%mm3  #                 B3 00 B2 00 B1 00 B0 00           \n\
 297 movq      %%mm3, %%mm5  #                 R3 00 R2 00 R1 00 R0 00           \n\
 298 punpcklwd %%mm4, %%mm3  #                 R1 G1 B1 00 R0 G0 B0 00           \n\
 299 movq      %%mm3, (%3)   # Store RGBA1 RGBA0                                 \n\
 300 punpckhwd %%mm4, %%mm5  #                 R3 G3 B3 00 R2 G2 B2 00           \n\
 301 movq      %%mm5, 8(%3)  # Store RGBA3 RGBA2                                 \n\
 302 pxor      %%mm6, %%mm6  # zero mm6                                          \n\
 303 punpckhbw %%mm1, %%mm2  #                 R7 G7 R6 G6 R5 G5 R4 G4           \n\
 304 punpckhbw %%mm0, %%mm6  #                 B7 00 B6 00 B5 00 B4 00           \n\
 305 movq      %%mm6, %%mm0  #                 B7 00 B6 00 B5 00 B4 00           \n\
 306 punpcklwd %%mm2, %%mm6  #                 R5 G5 B5 00 R4 G4 B4 00           \n\
 307 movq      %%mm6, 16(%3) # Store RGBA5 RGBA4                                 \n\
 308 punpckhwd %%mm2, %%mm0  #                 R7 G7 B7 00 R6 G6 B6 00           \n\
 309 movq      %%mm0, 24(%3) # Store RGBA7 RGBA6                                 \n\
 310 "
 311
 312 #define MMX_UNPACK_32_BGRA "                                                \n\
 313 pxor      %%mm3, %%mm3  # zero mm3                                          \n\
 314 movq      %%mm2, %%mm4  #                 G7 G6 G5 G4 G3 G2 G1 G0           \n\
 315 punpcklbw %%mm0, %%mm4  #                 B3 G3 B2 G2 B1 G1 B0 G0           \n\
 316 punpcklbw %%mm1, %%mm3  #                 R3 00 R2 00 R1 00 R0 00           \n\
 317 movq      %%mm3, %%mm5  #                 R3 00 R2 00 R1 00 R0 00           \n\
 318 punpcklwd %%mm4, %%mm3  #                 B1 G1 R1 00 B0 G0 R0 00           \n\
 319 movq      %%mm3, (%3)   # Store BGRA1 BGRA0                                 \n\
 320 punpckhwd %%mm4, %%mm5  #                 B3 G3 R3 00 B2 G2 R2 00           \n\
 321 movq      %%mm5, 8(%3)  # Store BGRA3 BGRA2                                 \n\
 322 pxor      %%mm6, %%mm6  # zero mm6                                          \n\
 323 punpckhbw %%mm0, %%mm2  #                 B7 G7 B6 G6 B5 G5 B4 G4           \n\
 324 punpckhbw %%mm1, %%mm6  #                 R7 00 R6 00 R5 00 R4 00           \n\
 325 movq      %%mm6, %%mm0  #                 R7 00 R6 00 R5 00 R4 00           \n\
 326 punpcklwd %%mm2, %%mm6  #                 B5 G5 R5 00 B4 G4 R4 00           \n\
 327 movq      %%mm6, 16(%3) # Store BGRA5 BGRA4                                 \n\
 328 punpckhwd %%mm2, %%mm0  #                 B7 G7 R7 00 B6 G6 R6 00           \n\
 329 movq      %%mm0, 24(%3) # Store BGRA7 BGRA6                                 \n\
 330 "
 331
 332 #define MMX_UNPACK_32_ABGR "                                                \n\
 333 pxor      %%mm3, %%mm3  # zero mm3                                          \n\
 334 movq      %%mm1, %%mm4  #                 R7 R6 R5 R4 R3 R2 R1 R0           \n\
 335 punpcklbw %%mm2, %%mm4  #                 G3 R3 G2 R2 G1 R1 G0 R0           \n\
 336 movq      %%mm0, %%mm5  #                 B7 B6 B5 B4 B3 B2 B1 B0           \n\
 337 punpcklbw %%mm3, %%mm5  #                 00 B3 00 B2 00 B1 00 B0           \n\
 338 movq      %%mm4, %%mm6  #                 G3 R3 G2 R2 G1 R1 G0 R0           \n\
 339 punpcklwd %%mm5, %%mm4  #                 00 B1 G1 R1 00 B0 G0 R0           \n\
 340 movq      %%mm4, (%3)   # Store ABGR1 ABGR0                                 \n\
 341 punpckhwd %%mm5, %%mm6  #                 00 B3 G3 R3 00 B2 G2 R2           \n\
 342 movq      %%mm6, 8(%3)  # Store ABGR3 ABGR2                                 \n\
 343 punpckhbw %%mm2, %%mm1  #                 G7 R7 G6 R6 G5 R5 G4 R4           \n\
 344 punpckhbw %%mm3, %%mm0  #                 00 B7 00 B6 00 B5 00 B4           \n\
 345 movq      %%mm1, %%mm2  #                 G7 R7 G6 R6 G5 R5 G4 R4           \n\
 346 punpcklwd %%mm0, %%mm1  #                 00 B5 G5 R5 00 B4 G4 R4           \n\
 347 movq      %%mm1, 16(%3) # Store ABGR5 ABGR4                                 \n\
 348 punpckhwd %%mm0, %%mm2  #                 B7 G7 R7 00 B6 G6 R6 00           \n\
 349 movq      %%mm2, 24(%3) # Store ABGR7 ABGR6                                 \n\
 350 "
 351
 352 #elif defined(HAVE_MMX_INTRINSICS)
 353
 354 /* MMX intrinsics */
 355
 356 #include <mmintrin.h>
 357
 358 #define MMX_CALL(MMX_INSTRUCTIONS)  \
 359     do {                            \
 360         __m64 mm0, mm1, mm2, mm3,   \
 361               mm4, mm5, mm6, mm7;   \
 362         MMX_INSTRUCTIONS            \
 363     } while(0)
 364
 365 #define MMX_END _mm_empty()
 366
 367 #define MMX_INIT_16                     \
 368     mm0 = _mm_cvtsi32_si64(*(int*)p_u); \
 369     mm1 = _mm_cvtsi32_si64(*(int*)p_v); \
 370     mm4 = _mm_setzero_si64();           \
 371     mm6 = (__m64)*(uint64_t *)p_y;
 372
 373 #define MMX_INIT_32                     \
 374     mm0 = _mm_cvtsi32_si64(*(int*)p_u); \
 375     *(uint16_t *)p_buffer = 0;          \
 376     mm1 = _mm_cvtsi32_si64(*(int*)p_v); \
 377     mm4 = _mm_setzero_si64();           \
 378     mm6 = (__m64)*(uint64_t *)p_y;
 379
 380 #define MMX_YUV_MUL                                 \
 381     mm0 = _mm_unpacklo_pi8(mm0, mm4);               \
 382     mm1 = _mm_unpacklo_pi8(mm1, mm4);               \
 383     mm0 = _mm_subs_pi16(mm0, (__m64)mmx_80w);       \
 384     mm1 = _mm_subs_pi16(mm1, (__m64)mmx_80w);       \
 385     mm0 = _mm_slli_pi16(mm0, 3);                    \
 386     mm1 = _mm_slli_pi16(mm1, 3);                    \
 387     mm2 = mm0;                                      \
 388     mm3 = mm1;                                      \
 389     mm2 = _mm_mulhi_pi16(mm2, (__m64)mmx_U_green);  \
 390     mm3 = _mm_mulhi_pi16(mm3, (__m64)mmx_V_green);  \
 391     mm0 = _mm_mulhi_pi16(mm0, (__m64)mmx_U_blue);   \
 392     mm1 = _mm_mulhi_pi16(mm1, (__m64)mmx_V_red);    \
 393     mm2 = _mm_adds_pi16(mm2, mm3);                  \
 394     \
 395     mm6 = _mm_subs_pu8(mm6, (__m64)mmx_10w);        \
 396     mm7 = mm6;                                      \
 397     mm6 = _mm_and_si64(mm6, (__m64)mmx_00ffw);      \
 398     mm7 = _mm_srli_pi16(mm7, 8);                    \
 399     mm6 = _mm_slli_pi16(mm6, 3);                    \
 400     mm7 = _mm_slli_pi16(mm7, 3);                    \
 401     mm6 = _mm_mulhi_pi16(mm6, (__m64)mmx_Y_coeff);  \
 402     mm7 = _mm_mulhi_pi16(mm7, (__m64)mmx_Y_coeff);
 403
 404 #define MMX_YUV_ADD                     \
 405     mm3 = mm0;                          \
 406     mm4 = mm1;                          \
 407     mm5 = mm2;                          \
 408     mm0 = _mm_adds_pi16(mm0, mm6);      \
 409     mm3 = _mm_adds_pi16(mm3, mm7);      \
 410     mm1 = _mm_adds_pi16(mm1, mm6);      \
 411     mm4 = _mm_adds_pi16(mm4, mm7);      \
 412     mm2 = _mm_adds_pi16(mm2, mm6);      \
 413     mm5 = _mm_adds_pi16(mm5, mm7);      \
 414     \
 415     mm0 = _mm_packs_pu16(mm0, mm0);     \
 416     mm1 = _mm_packs_pu16(mm1, mm1);     \
 417     mm2 = _mm_packs_pu16(mm2, mm2);     \
 418     \
 419     mm3 = _mm_packs_pu16(mm3, mm3);     \
 420     mm4 = _mm_packs_pu16(mm4, mm4);     \
 421     mm5 = _mm_packs_pu16(mm5, mm5);     \
 422     \
 423     mm0 = _mm_unpacklo_pi8(mm0, mm3);   \
 424     mm1 = _mm_unpacklo_pi8(mm1, mm4);   \
 425     mm2 = _mm_unpacklo_pi8(mm2, mm5);
 426
 427 #define MMX_UNPACK_15                               \
 428     mm0 = _mm_and_si64(mm0, (__m64)mmx_mask_f8);    \
 429     mm0 = _mm_srli_pi16(mm0, 3);                    \
 430     mm2 = _mm_and_si64(mm2, (__m64)mmx_mask_f8);    \
 431     mm1 = _mm_and_si64(mm1, (__m64)mmx_mask_f8);    \
 432     mm1 = _mm_srli_pi16(mm1, 1);                    \
 433     mm4 = _mm_setzero_si64();                       \
 434     mm5 = mm0;                                      \
 435     mm7 = mm2;                                      \
 436     \
 437     mm2 = _mm_unpacklo_pi8(mm2, mm4);               \
 438     mm0 = _mm_unpacklo_pi8(mm0, mm1);               \
 439     mm2 = _mm_slli_pi16(mm2, 2);                    \
 440     mm0 = _mm_or_si64(mm0, mm2);                    \
 441     mm6 = (__m64)*(uint64_t *)(p_y + 8);            \
 442     *(uint64_t *)p_buffer = (uint64_t)mm0;          \
 443     \
 444     mm7 = _mm_unpackhi_pi8(mm7, mm4);               \
 445     mm5 = _mm_unpackhi_pi8(mm5, mm1);               \
 446     mm7 = _mm_slli_pi16(mm7, 2);                    \
 447     mm0 = _mm_cvtsi32_si64((int)*(uint32_t *)(p_u + 4)); \
 448     mm5 = _mm_or_si64(mm5, mm7);                    \
 449     mm1 = _mm_cvtsi32_si64((int)*(uint32_t *)(p_v + 4)); \
 450     *(uint64_t *)(p_buffer + 4) = (uint64_t)mm5;
 451
 452 #define MMX_UNPACK_16                               \
 453     mm0 = _mm_and_si64(mm0, (__m64)mmx_mask_f8);    \
 454     mm2 = _mm_and_si64(mm2, (__m64)mmx_mask_fc);    \
 455     mm1 = _mm_and_si64(mm1, (__m64)mmx_mask_f8);    \
 456     mm0 = _mm_srli_pi16(mm0, 3);                    \
 457     mm4 = _mm_setzero_si64();                       \
 458     mm5 = mm0;                                      \
 459     mm7 = mm2;                                      \
 460     \
 461     mm2 = _mm_unpacklo_pi8(mm2, mm4);               \
 462     mm0 = _mm_unpacklo_pi8(mm0, mm1);               \
 463     mm2 = _mm_slli_pi16(mm2, 3);                    \
 464     mm0 = _mm_or_si64(mm0, mm2);                    \
 465     mm6 = (__m64)*(uint64_t *)(p_y + 8);            \
 466     *(uint64_t *)p_buffer = (uint64_t)mm0;          \
 467     \
 468     mm7 = _mm_unpackhi_pi8(mm7, mm4);               \
 469     mm5 = _mm_unpackhi_pi8(mm5, mm1);               \
 470     mm7 = _mm_slli_pi16(mm7, 3);                    \
 471     mm0 = _mm_cvtsi32_si64((int)*(uint32_t *)(p_u + 4)); \
 472     mm5 = _mm_or_si64(mm5, mm7);                    \
 473     mm1 = _mm_cvtsi32_si64((int)*(uint32_t *)(p_v + 4)); \
 474     *(uint64_t *)(p_buffer + 4) = (uint64_t)mm5;
 475
 476 #define MMX_UNPACK_32_ARGB                      \
 477     mm3 = _mm_setzero_si64();                   \
 478     mm4 = mm0;                                  \
 479     mm4 = _mm_unpacklo_pi8(mm4, mm2);           \
 480     mm5 = mm1;                                  \
 481     mm5 = _mm_unpacklo_pi8(mm5, mm3);           \
 482     mm6 = mm4;                                  \
 483     mm4 = _mm_unpacklo_pi16(mm4, mm5);          \
 484     *(uint64_t *)p_buffer = (uint64_t)mm4;      \
 485     mm6 = _mm_unpackhi_pi16(mm6, mm5);          \
 486     *(uint64_t *)(p_buffer + 2) = (uint64_t)mm6;\
 487     mm0 = _mm_unpackhi_pi8(mm0, mm2);           \
 488     mm1 = _mm_unpackhi_pi8(mm1, mm3);           \
 489     mm5 = mm0;                                  \
 490     mm5 = _mm_unpacklo_pi16(mm5, mm1);          \
 491     *(uint64_t *)(p_buffer + 4) = (uint64_t)mm5;\
 492     mm0 = _mm_unpackhi_pi16(mm0, mm1);          \
 493     *(uint64_t *)(p_buffer + 6) = (uint64_t)mm0;
 494
 495 #define MMX_UNPACK_32_RGBA                      \
 496     mm3 = _mm_setzero_si64();                   \
 497     mm4 = mm2;                                  \
 498     mm4 = _mm_unpacklo_pi8(mm4, mm1);           \
 499     mm3 = _mm_unpacklo_pi8(mm3, mm0);           \
 500     mm5 = mm3;                                  \
 501     mm3 = _mm_unpacklo_pi16(mm3, mm4);          \
 502     *(uint64_t *)p_buffer = (uint64_t)mm3;      \
 503     mm5 = _mm_unpackhi_pi16(mm5, mm4);          \
 504     *(uint64_t *)(p_buffer + 2) = (uint64_t)mm5;\
 505     mm6 = _mm_setzero_si64();                   \
 506     mm2 = _mm_unpackhi_pi8(mm2, mm1);           \
 507     mm6 = _mm_unpackhi_pi8(mm6, mm0);           \
 508     mm0 = mm6;                                  \
 509     mm6 = _mm_unpacklo_pi16(mm6, mm2);          \
 510     *(uint64_t *)(p_buffer + 4) = (uint64_t)mm6;\
 511     mm0 = _mm_unpackhi_pi16(mm0, mm2);          \
 512     *(uint64_t *)(p_buffer + 6) = (uint64_t)mm0;
 513
 514 #define MMX_UNPACK_32_BGRA                      \
 515     mm3 = _mm_setzero_si64();                   \
 516     mm4 = mm2;                                  \
 517     mm4 = _mm_unpacklo_pi8(mm4, mm0);           \
 518     mm3 = _mm_unpacklo_pi8(mm3, mm1);           \
 519     mm5 = mm3;                                  \
 520     mm3 = _mm_unpacklo_pi16(mm3, mm4);          \
 521     *(uint64_t *)p_buffer = (uint64_t)mm3;      \
 522     mm5 = _mm_unpackhi_pi16(mm5, mm4);          \
 523     *(uint64_t *)(p_buffer + 2) = (uint64_t)mm5;\
 524     mm6 = _mm_setzero_si64();                   \
 525     mm2 = _mm_unpackhi_pi8(mm2, mm0);           \
 526     mm6 = _mm_unpackhi_pi8(mm6, mm1);           \
 527     mm0 = mm6;                                  \
 528     mm6 = _mm_unpacklo_pi16(mm6, mm2);          \
 529     *(uint64_t *)(p_buffer + 4) = (uint64_t)mm6;\
 530     mm0 = _mm_unpackhi_pi16(mm0, mm2);          \
 531     *(uint64_t *)(p_buffer + 6) = (uint64_t)mm0;
 532
 533 #define MMX_UNPACK_32_ABGR                      \
 534     mm3 = _mm_setzero_si64();                   \
 535     mm4 = mm1;                                  \
 536     mm4 = _mm_unpacklo_pi8(mm4, mm2);           \
 537     mm5 = mm0;                                  \
 538     mm5 = _mm_unpacklo_pi8(mm5, mm3);           \
 539     mm6 = mm4;                                  \
 540     mm4 = _mm_unpacklo_pi16(mm4, mm5);          \
 541     *(uint64_t *)p_buffer = (uint64_t)mm4;      \
 542     mm6 = _mm_unpackhi_pi16(mm6, mm5);          \
 543     *(uint64_t *)(p_buffer + 2) = (uint64_t)mm6;\
 544     mm1 = _mm_unpackhi_pi8(mm1, mm2);           \
 545     mm0 = _mm_unpackhi_pi8(mm0, mm3);           \
 546     mm2 = mm1;                                  \
 547     mm1 = _mm_unpacklo_pi16(mm1, mm0);          \
 548     *(uint64_t *)(p_buffer + 4) = (uint64_t)mm1;\
 549     mm2 = _mm_unpackhi_pi16(mm2, mm0);          \
 550     *(uint64_t *)(p_buffer + 6) = (uint64_t)mm2;
 551
 552 #endif
 553