git.sesse.net Git - vlc/blob - plugins/yuv/transforms_yuvmmx.h

   1 /*****************************************************************************
   2  * transforms_yuvmmx.h: MMX YUV transformation assembly
   3  *****************************************************************************
   4  * Copyright (C) 1999, 2000 VideoLAN
   5  *
   6  * Authors:
   7  * Olie Lho <ollie@sis.com.tw>
   8  *
   9  * Adapted to VideoLAN by:
  10  * Gaël Hendryckx <jimmy@via.ecp.fr>
  11  * Samuel Hocevar <sam@via.ecp.fr>
  12  *
  13  * This program is free software; you can redistribute it and/or modify
  14  * it under the terms of the GNU General Public License as published by
  15  * the Free Software Foundation; either version 2 of the License, or
  16  * (at your option) any later version.
  17  *
  18  * This program is distributed in the hope that it will be useful,
  19  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  20  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  21  * General Public License for more details.
  22  *
  23  * You should have received a copy of the GNU General Public
  24  * License along with this program; if not, write to the
  25  * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
  26  * Boston, MA 02111-1307, USA.
  27  *****************************************************************************/
  28
  29 /* hope these constant values are cache line aligned */
  30 #define UNUSED_LONGLONG(foo) \
  31     static unsigned long long foo __attribute__((unused))
  32 UNUSED_LONGLONG(mmx_80w)     = 0x0080008000800080;
  33 UNUSED_LONGLONG(mmx_10w)     = 0x1010101010101010;
  34 UNUSED_LONGLONG(mmx_00ffw)   = 0x00ff00ff00ff00ff;
  35 UNUSED_LONGLONG(mmx_Y_coeff) = 0x253f253f253f253f;
  36
  37 UNUSED_LONGLONG(mmx_U_green) = 0xf37df37df37df37d;
  38 UNUSED_LONGLONG(mmx_U_blue)  = 0x4093409340934093;
  39 UNUSED_LONGLONG(mmx_V_red)   = 0x3312331233123312;
  40 UNUSED_LONGLONG(mmx_V_green) = 0xe5fce5fce5fce5fc;
  41
  42 UNUSED_LONGLONG(mmx_redmask) = 0xf8f8f8f8f8f8f8f8;
  43 UNUSED_LONGLONG(mmx_grnmask) = 0xfcfcfcfcfcfcfcfc;
  44 UNUSED_LONGLONG(mmx_grnshift)   = 0x03;
  45 UNUSED_LONGLONG(mmx_blueshift)  = 0x03;
  46 #undef UNUSED_LONGLONG
  47
  48 #define MMX_INIT_16 "                                                       \n\
  49                                                                             \n\
  50 movd      (%1), %%mm0       # Load 4 Cb       00 00 00 00 u3 u2 u1 u0       \n\
  51 movd      (%2), %%mm1       # Load 4 Cr       00 00 00 00 v3 v2 v1 v0       \n\
  52 pxor      %%mm4, %%mm4      # zero mm4                                      \n\
  53 movq      (%0), %%mm6       # Load 8 Y        Y7 Y6 Y5 Y4 Y3 Y2 Y1 Y0       \n\
  54 #movl      $0, (%3)         # cache preload for image                       \n\
  55 "
  56
  57 #define MMX_INIT_16_GRAY "                                                  \n\
  58                                                                             \n\
  59 movq      (%0), %%mm6       # Load 8 Y        Y7 Y6 Y5 Y4 Y3 Y2 Y1 Y0       \n\
  60 #movl      $0, (%3)         # cache preload for image                       \n\
  61 "
  62
  63 #define MMX_INIT_32 "                                                       \n\
  64                                                                             \n\
  65 movd      (%1), %%mm0       # Load 4 Cb       00 00 00 00 u3 u2 u1 u0       \n\
  66 movl      $0, (%3)          # cache preload for image                       \n\
  67 movd      (%2), %%mm1       # Load 4 Cr       00 00 00 00 v3 v2 v1 v0       \n\
  68 pxor      %%mm4, %%mm4      # zero mm4                                      \n\
  69 movq      (%0), %%mm6       # Load 8 Y        Y7 Y6 Y5 Y4 Y3 Y2 Y1 Y0       \n\
  70 "
  71
  72 /*
  73  * Do the multiply part of the conversion for even and odd pixels,
  74  * register usage:
  75  * mm0 -> Cblue, mm1 -> Cred, mm2 -> Cgreen even pixels,
  76  * mm3 -> Cblue, mm4 -> Cred, mm5 -> Cgreen odd  pixels,
  77  * mm6 -> Y even, mm7 -> Y odd
  78  */
  79
  80 #define MMX_YUV_MUL "                                                       \n\
  81                                                                             \n\
  82 # convert the chroma part                                                   \n\
  83 punpcklbw %%mm4, %%mm0          # scatter 4 Cb    00 u3 00 u2 00 u1 00 u0   \n\
  84 punpcklbw %%mm4, %%mm1          # scatter 4 Cr    00 v3 00 v2 00 v1 00 v0   \n\
  85 psubsw    mmx_80w, %%mm0        # Cb -= 128                                 \n\
  86 psubsw    mmx_80w, %%mm1        # Cr -= 128                                 \n\
  87 psllw     $3, %%mm0             # Promote precision                         \n\
  88 psllw     $3, %%mm1             # Promote precision                         \n\
  89 movq      %%mm0, %%mm2          # Copy 4 Cb       00 u3 00 u2 00 u1 00 u0   \n\
  90 movq      %%mm1, %%mm3          # Copy 4 Cr       00 v3 00 v2 00 v1 00 v0   \n\
  91 pmulhw    mmx_U_green, %%mm2    # Mul Cb with green coeff -> Cb green       \n\
  92 pmulhw    mmx_V_green, %%mm3    # Mul Cr with green coeff -> Cr green       \n\
  93 pmulhw    mmx_U_blue, %%mm0     # Mul Cb -> Cblue 00 b3 00 b2 00 b1 00 b0   \n\
  94 pmulhw    mmx_V_red, %%mm1      # Mul Cr -> Cred  00 r3 00 r2 00 r1 00 r0   \n\
  95 paddsw    %%mm3, %%mm2          # Cb green + Cr green -> Cgreen             \n\
  96                                                                             \n\
  97 # convert the luma part                                                     \n\
  98 psubusb   mmx_10w, %%mm6        # Y -= 16                                   \n\
  99 movq      %%mm6, %%mm7          # Copy 8 Y        Y7 Y6 Y5 Y4 Y3 Y2 Y1 Y0   \n\
 100 pand      mmx_00ffw, %%mm6      # get Y even      00 Y6 00 Y4 00 Y2 00 Y0   \n\
 101 psrlw     $8, %%mm7             # get Y odd       00 Y7 00 Y5 00 Y3 00 Y1   \n\
 102 psllw     $3, %%mm6             # Promote precision                         \n\
 103 psllw     $3, %%mm7             # Promote precision                         \n\
 104 pmulhw    mmx_Y_coeff, %%mm6    # Mul 4 Y even    00 y6 00 y4 00 y2 00 y0   \n\
 105 pmulhw    mmx_Y_coeff, %%mm7    # Mul 4 Y odd     00 y7 00 y5 00 y3 00 y1   \n\
 106 "
 107
 108 #define MMX_YUV_YCBR_422 "                                                  \n\
 109                                                                             \n\
 110 punpcklbw %%mm1, %%mm0                                                      \n\
 111 movq %%mm6, %%mm2                                                           \n\
 112 punpckhbw %%mm0, %%mm6                                                      \n\
 113 punpcklbw %%mm0, %%mm2                                                      \n\
 114 movq %%mm2, (%3)                                                            \n\
 115 movq %%mm6, 8(%3)                                                           \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                                                                             \n\
 128 # Do horizontal and vertical scaling                                        \n\
 129 movq      %%mm0, %%mm3          # Copy Cblue                                \n\
 130 movq      %%mm1, %%mm4          # Copy Cred                                 \n\
 131 movq      %%mm2, %%mm5          # Copy Cgreen                               \n\
 132 paddsw    %%mm6, %%mm0          # Y even + Cblue  00 B6 00 B4 00 B2 00 B0   \n\
 133 paddsw    %%mm7, %%mm3          # Y odd  + Cblue  00 B7 00 B5 00 B3 00 B1   \n\
 134 paddsw    %%mm6, %%mm1          # Y even + Cred   00 R6 00 R4 00 R2 00 R0   \n\
 135 paddsw    %%mm7, %%mm4          # Y odd  + Cred   00 R7 00 R5 00 R3 00 R1   \n\
 136 paddsw    %%mm6, %%mm2          # Y even + Cgreen 00 G6 00 G4 00 G2 00 G0   \n\
 137 paddsw    %%mm7, %%mm5          # Y odd  + Cgreen 00 G7 00 G5 00 G3 00 G1   \n\
 138                                                                             \n\
 139 # Limit RGB even to 0..255                                                  \n\
 140 packuswb  %%mm0, %%mm0          # B6 B4 B2 B0 | B6 B4 B2 B0                 \n\
 141 packuswb  %%mm1, %%mm1          # R6 R4 R2 R0 | R6 R4 R2 R0                 \n\
 142 packuswb  %%mm2, %%mm2          # G6 G4 G2 G0 | G6 G4 G2 G0                 \n\
 143                                                                             \n\
 144 # Limit RGB odd to 0..255                                                   \n\
 145 packuswb  %%mm3, %%mm3          # B7 B5 B3 B1 | B7 B5 B3 B1                 \n\
 146 packuswb  %%mm4, %%mm4          # R7 R5 R3 R1 | R7 R5 R3 R1                 \n\
 147 packuswb  %%mm5, %%mm5          # G7 G5 G3 G1 | G7 G5 G3 G1                 \n\
 148                                                                             \n\
 149 # Interleave RGB even and odd                                               \n\
 150 punpcklbw %%mm3, %%mm0          #                 B7 B6 B5 B4 B3 B2 B1 B0   \n\
 151 punpcklbw %%mm4, %%mm1          #                 R7 R6 R5 R4 R3 R2 R1 R0   \n\
 152 punpcklbw %%mm5, %%mm2          #                 G7 G6 G5 G4 G3 G2 G1 G0   \n\
 153 "
 154
 155 /*
 156  * Grayscale case, only use Y
 157  */
 158
 159 #define MMX_YUV_GRAY "                                                      \n\
 160                                                                             \n\
 161 # convert the luma part                                                     \n\
 162 psubusb   mmx_10w, %%mm6                                                    \n\
 163 movq      %%mm6, %%mm7                                                      \n\
 164 pand      mmx_00ffw, %%mm6                                                  \n\
 165 psrlw     $8, %%mm7                                                         \n\
 166 psllw     $3, %%mm6                                                         \n\
 167 psllw     $3, %%mm7                                                         \n\
 168 pmulhw    mmx_Y_coeff, %%mm6                                                \n\
 169 pmulhw    mmx_Y_coeff, %%mm7                                                \n\
 170 packuswb  %%mm6, %%mm6                                                      \n\
 171 packuswb  %%mm7, %%mm7                                                      \n\
 172 punpcklbw %%mm7, %%mm6                                                      \n\
 173 "
 174
 175 #define MMX_UNPACK_16_GRAY "                                                \n\
 176 movq      %%mm6, %%mm5                                                      \n\
 177 pand      mmx_redmask, %%mm6                                                \n\
 178 pand      mmx_grnmask, %%mm5                                                \n\
 179 movq      %%mm6, %%mm7                                                      \n\
 180 psrlw     mmx_blueshift, %%mm7                                              \n\
 181 pxor      %%mm3, %%mm3                                                      \n\
 182 movq      %%mm7, %%mm2                                                      \n\
 183 movq      %%mm5, %%mm0                                                      \n\
 184 punpcklbw %%mm3, %%mm5                                                      \n\
 185 punpcklbw %%mm6, %%mm7                                                      \n\
 186 psllw     mmx_blueshift, %%mm5                                              \n\
 187 por       %%mm5, %%mm7                                                      \n\
 188 movq      %%mm7, (%3)                                                       \n\
 189 punpckhbw %%mm3, %%mm0                                                      \n\
 190 punpckhbw %%mm6, %%mm2                                                      \n\
 191 psllw     mmx_blueshift, %%mm0                                              \n\
 192 movq      8(%0), %%mm6                                                      \n\
 193 por       %%mm0, %%mm2                                                      \n\
 194 movq      %%mm2, 8(%3)                                                      \n\
 195 "
 196
 197
 198 /*
 199  * convert RGB plane to RGB 16 bits,
 200  * mm0 -> B, mm1 -> R, mm2 -> G,
 201  * mm4 -> GB, mm5 -> AR pixel 4-7,
 202  * mm6 -> GB, mm7 -> AR pixel 0-3
 203  */
 204
 205 #define MMX_UNPACK_16 "                                                     \n\
 206                                                                             \n\
 207 # mask unneeded bits off                                                    \n\
 208 pand      mmx_redmask, %%mm0    # b7b6b5b4 b3______ b7b6b5b4 b3______       \n\
 209 pand      mmx_grnmask, %%mm2    # g7g6g5g4 g3g2____ g7g6g5g4 g3g2____       \n\
 210 pand      mmx_redmask, %%mm1    # r7r6r5r4 r3______ r7r6r5r4 r3______       \n\
 211 psrlw     mmx_blueshift,%%mm0   # ______b7 b6b5b4b3 ______b7 b6b5b4b3       \n\
 212 pxor      %%mm4, %%mm4          # zero mm4                                  \n\
 213 movq      %%mm0, %%mm5          # Copy B7-B0                                \n\
 214 movq      %%mm2, %%mm7          # Copy G7-G0                                \n\
 215                                                                             \n\
 216 # convert rgb24 plane to rgb16 pack for pixel 0-3                           \n\
 217 punpcklbw %%mm4, %%mm2          # ________ ________ g7g6g5g4 g3g2____       \n\
 218 punpcklbw %%mm1, %%mm0          # r7r6r5r4 r3______ ______b7 b6b5b4b3       \n\
 219 psllw     mmx_blueshift,%%mm2   # ________ __g7g6g5 g4g3g2__ ________       \n\
 220 por       %%mm2, %%mm0          # r7r6r5r4 r3g7g6g5 g4g3g2b7 b6b5b4b3       \n\
 221 movq      8(%0), %%mm6          # Load 8 Y        Y7 Y6 Y5 Y4 Y3 Y2 Y1 Y0   \n\
 222 movq      %%mm0, (%3)           # store pixel 0-3                           \n\
 223                                                                             \n\
 224 # convert rgb24 plane to rgb16 pack for pixel 0-3                           \n\
 225 punpckhbw %%mm4, %%mm7          # ________ ________ g7g6g5g4 g3g2____       \n\
 226 punpckhbw %%mm1, %%mm5          # r7r6r5r4 r3______ ______b7 b6b5b4b3       \n\
 227 psllw     mmx_blueshift,%%mm7   # ________ __g7g6g5 g4g3g2__ ________       \n\
 228 movd      4(%1), %%mm0          # Load 4 Cb       __ __ __ __ u3 u2 u1 u0   \n\
 229 por       %%mm7, %%mm5          # r7r6r5r4 r3g7g6g5 g4g3g2b7 b6b5b4b3       \n\
 230 movd      4(%2), %%mm1          # Load 4 Cr       __ __ __ __ v3 v2 v1 v0   \n\
 231 movq      %%mm5, 8(%3)          # store pixel 4-7                           \n\
 232 "
 233
 234 /*
 235  * convert RGB plane to RGB packed format,
 236  * mm0 -> B, mm1 -> R, mm2 -> G, mm3 -> 0,
 237  * mm4 -> GB, mm5 -> AR pixel 4-7,
 238  * mm6 -> GB, mm7 -> AR pixel 0-3
 239  */
 240
 241 #define MMX_UNPACK_32 "                                                     \n\
 242                                                                             \n\
 243 pxor      %%mm3, %%mm3  # zero mm3                                          \n\
 244 movq      %%mm0, %%mm6  #                 B7 B6 B5 B4 B3 B2 B1 B0           \n\
 245 movq      %%mm1, %%mm7  #                 R7 R6 R5 R4 R3 R2 R1 R0           \n\
 246 movq      %%mm0, %%mm4  #                 B7 B6 B5 B4 B3 B2 B1 B0           \n\
 247 movq      %%mm1, %%mm5  #                 R7 R6 R5 R4 R3 R2 R1 R0           \n\
 248 punpcklbw %%mm2, %%mm6  #                 G3 B3 G2 B2 G1 B1 G0 B0           \n\
 249 punpcklbw %%mm3, %%mm7  #                 00 R3 00 R2 00 R1 00 R0           \n\
 250 punpcklwd %%mm7, %%mm6  #                 00 R1 B1 G1 00 R0 B0 G0           \n\
 251 movq      %%mm6, (%3)   # Store ARGB1 ARGB0                                 \n\
 252 movq      %%mm0, %%mm6  #                 B7 B6 B5 B4 B3 B2 B1 B0           \n\
 253 punpcklbw %%mm2, %%mm6  #                 G3 B3 G2 B2 G1 B1 G0 B0           \n\
 254 punpckhwd %%mm7, %%mm6  #                 00 R3 G3 B3 00 R2 B3 G2           \n\
 255 movq      %%mm6, 8(%3)  # Store ARGB3 ARGB2                                 \n\
 256 punpckhbw %%mm2, %%mm4  #                 G7 B7 G6 B6 G5 B5 G4 B4           \n\
 257 punpckhbw %%mm3, %%mm5  #                 00 R7 00 R6 00 R5 00 R4           \n\
 258 punpcklwd %%mm5, %%mm4  #                 00 R5 B5 G5 00 R4 B4 G4           \n\
 259 movq      %%mm4, 16(%3) # Store ARGB5 ARGB4                                 \n\
 260 movq      %%mm0, %%mm4  #                 B7 B6 B5 B4 B3 B2 B1 B0           \n\
 261 punpckhbw %%mm2, %%mm4  #                 G7 B7 G6 B6 G5 B5 G4 B4           \n\
 262 punpckhwd %%mm5, %%mm4  #                 00 R7 G7 B7 00 R6 B6 G6           \n\
 263 movq      %%mm4, 24(%3) # Store ARGB7 ARGB6                                 \n\
 264                                                                             \n\
 265 #movd      4(%1), %%mm0  # Load 4 Cb       00 00 00 00 u3 u2 u1 u0           \n\
 266 #movd      4(%2), %%mm1  # Load 4 Cr       00 00 00 00 v3 v2 v1 v0           \n\
 267 #pxor      %%mm4, %%mm4  # zero mm4                                          \n\
 268 #movq      8(%0), %%mm6  # Load 8 Y        Y7 Y6 Y5 Y4 Y3 Y2 Y1 Y0           \n\
 269 "