src/video/SDL_yuv_mmx.c
author Sam Lantinga <slouken@libsdl.org>
Sat, 21 Aug 2004 14:54:49 +0000
changeset 946 8520712f8ef0
parent 887 b4b64bb88f2f
child 949 e0d96eb0af19
permissions -rw-r--r--
*** empty log message ***
     1 /*
     2     SDL - Simple DirectMedia Layer
     3     Copyright (C) 1997-2004 Sam Lantinga
     4 
     5     This library is free software; you can redistribute it and/or
     6     modify it under the terms of the GNU Library General Public
     7     License as published by the Free Software Foundation; either
     8     version 2 of the License, or (at your option) any later version.
     9 
    10     This library is distributed in the hope that it will be useful,
    11     but WITHOUT ANY WARRANTY; without even the implied warranty of
    12     MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
    13     Library General Public License for more details.
    14 
    15     You should have received a copy of the GNU Library General Public
    16     License along with this library; if not, write to the Free
    17     Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA  02111-1307  USA
    18 
    19     Sam Lantinga
    20     slouken@libsdl.org
    21 */
    22 
    23 #ifdef SAVE_RCSID
    24 static char rcsid =
    25  "@(#) $Id$";
    26 #endif
    27 
    28 
    29 #if defined(i386) && defined(__GNUC__) && defined(USE_ASMBLIT)
    30 
    31 #include "SDL_types.h"
    32 
    33 static unsigned int  MMX_0080w[]    = {0x00800080, 0x00800080};
    34 static unsigned int  MMX_00FFw[]    = {0x00ff00ff, 0x00ff00ff}; 
    35 static unsigned int  MMX_FF00w[]    = {0xff00ff00, 0xff00ff00}; 
    36 
    37 static unsigned short MMX_Ycoeff[]  = {0x004a, 0x004a, 0x004a, 0x004a}; 
    38 
    39 static unsigned short MMX_UbluRGB[] = {0x0072, 0x0072, 0x0072, 0x0072};    
    40 static unsigned short MMX_VredRGB[] = {0x0059, 0x0059, 0x0059, 0x0059};  
    41 static unsigned short MMX_UgrnRGB[] = {0xffea, 0xffea, 0xffea, 0xffea}; 
    42 static unsigned short MMX_VgrnRGB[] = {0xffd2, 0xffd2, 0xffd2, 0xffd2};  
    43 
    44 static unsigned short MMX_Ublu5x5[] = {0x0081, 0x0081, 0x0081, 0x0081};
    45 static unsigned short MMX_Vred5x5[] = {0x0066, 0x0066, 0x0066, 0x0066};
    46 static unsigned short MMX_Ugrn555[] = {0xffe7, 0xffe7, 0xffe7, 0xffe7};
    47 static unsigned short MMX_Vgrn555[] = {0xffcc, 0xffcc, 0xffcc, 0xffcc};
    48 static unsigned short MMX_Ugrn565[] = {0xffe8, 0xffe8, 0xffe8, 0xffe8};
    49 static unsigned short MMX_Vgrn565[] = {0xffcd, 0xffcd, 0xffcd, 0xffcd};
    50 
    51 static unsigned short MMX_red555[]  = {0x7c00, 0x7c00, 0x7c00, 0x7c00};
    52 static unsigned short MMX_red565[]  = {0xf800, 0xf800, 0xf800, 0xf800};
    53 static unsigned short MMX_grn555[]  = {0x03e0, 0x03e0, 0x03e0, 0x03e0};
    54 static unsigned short MMX_grn565[]  = {0x07e0, 0x07e0, 0x07e0, 0x07e0};
    55 static unsigned short MMX_blu5x5[]  = {0x001f, 0x001f, 0x001f, 0x001f};
    56 
    57 /**
    58    This MMX assembler is my first assembler/MMX program ever.
    59    Thus it maybe buggy.
    60    Send patches to:
    61    mvogt@rhrk.uni-kl.de
    62 
    63    After it worked fine I have "obfuscated" the code a bit to have
    64    more parallism in the MMX units. This means I moved
    65    initilisation around and delayed other instruction.
    66    Performance measurement did not show that this brought any advantage
    67    but in theory it _should_ be faster this way.
    68 
    69    The overall performanve gain to the C based dither was 30%-40%.
    70    The MMX routine calculates 256bit=8RGB values in each cycle
    71    (4 for row1 & 4 for row2)
    72 
    73    The red/green/blue.. coefficents are taken from the mpeg_play 
    74    player. They look nice, but I dont know if you can have
    75    better values, to avoid integer rounding errors.
    76    
    77 
    78    IMPORTANT:
    79    ==========
    80 
    81    It is a requirement that the cr/cb/lum are 8 byte aligned and
    82    the out are 16byte aligned or you will/may get segfaults
    83 
    84 */
    85 
    86 void ColorRGBDitherYV12MMX1X( int *colortab, Uint32 *rgb_2_pix,
    87                               unsigned char *lum, unsigned char *cr,
    88                               unsigned char *cb, unsigned char *out,
    89                               int rows, int cols, int mod )
    90 {
    91     Uint32 *row1;
    92     Uint32 *row2;
    93 
    94     unsigned char* y = lum +cols*rows;    // Pointer to the end
    95     int x=0;
    96     row1 = (Uint32 *)out;                 // 32 bit target
    97     row2 = (Uint32 *)out+cols+mod;        // start of second row 
    98     mod = (mod+cols+mod)*4;               // increment for row1 in byte
    99 
   100     __asm__ __volatile__ (
   101 /* We don't really care about PIC - the code should be rewritten to use
   102    relative addressing for the static tables, so right now we take the
   103    COW hit on the pages this code resides. Big deal.
   104    This spill is just to reduce register pressure in the PIC case. */
   105 		 "pushl %%ebx\n"
   106 		 "movl %0, %%ebx\n"
   107 
   108 	         ".align 8\n"
   109 		 "1:\n"
   110 		
   111 		 // create Cr (result in mm1)
   112 		 "movd (%%ebx), %%mm1\n"   //         0  0  0  0  v3 v2 v1 v0
   113 		 "pxor %%mm7,%%mm7\n"      //         00 00 00 00 00 00 00 00
   114 		 "movd (%2), %%mm2\n"           //    0  0  0  0 l3 l2 l1 l0
   115 		 "punpcklbw %%mm7,%%mm1\n" //         0  v3 0  v2 00 v1 00 v0
   116 		 "punpckldq %%mm1,%%mm1\n" //         00 v1 00 v0 00 v1 00 v0
   117 		 "psubw %[_MMX_0080w],%%mm1\n"  // mm1-128:r1 r1 r0 r0 r1 r1 r0 r0 
   118 
   119 		 // create Cr_g (result in mm0)
   120 		 "movq %%mm1,%%mm0\n"           // r1 r1 r0 r0 r1 r1 r0 r0
   121 		 "pmullw %[_MMX_VgrnRGB],%%mm0\n"// red*-46dec=0.7136*64
   122 		 "pmullw %[_MMX_VredRGB],%%mm1\n"// red*89dec=1.4013*64
   123 		 "psraw  $6, %%mm0\n"           // red=red/64
   124 		 "psraw  $6, %%mm1\n"           // red=red/64
   125 		 
   126 		 // create L1 L2 (result in mm2,mm4)
   127 		 // L2=lum+cols
   128 		 "movq (%2,%4),%%mm3\n"         //    0  0  0  0 L3 L2 L1 L0
   129 		 "punpckldq %%mm3,%%mm2\n"      //   L3 L2 L1 L0 l3 l2 l1 l0
   130 		 "movq %%mm2,%%mm4\n"           //   L3 L2 L1 L0 l3 l2 l1 l0
   131 		 "pand %[_MMX_FF00w],%%mm2\n"      //   L3 0  L1  0 l3  0 l1  0
   132 		 "pand %[_MMX_00FFw],%%mm4\n"      //   0  L2  0 L0  0 l2  0 l0
   133 		 "psrlw $8,%%mm2\n"             //   0  L3  0 L1  0 l3  0 l1
   134 
   135 		 // create R (result in mm6)
   136 		 "movq %%mm2,%%mm5\n"           //   0 L3  0 L1  0 l3  0 l1
   137 		 "movq %%mm4,%%mm6\n"           //   0 L2  0 L0  0 l2  0 l0
   138 		 "paddsw  %%mm1, %%mm5\n"       // lum1+red:x R3 x R1 x r3 x r1
   139 		 "paddsw  %%mm1, %%mm6\n"       // lum1+red:x R2 x R0 x r2 x r0
   140 		 "packuswb %%mm5,%%mm5\n"       //  R3 R1 r3 r1 R3 R1 r3 r1
   141 		 "packuswb %%mm6,%%mm6\n"       //  R2 R0 r2 r0 R2 R0 r2 r0
   142 		 "pxor %%mm7,%%mm7\n"      //         00 00 00 00 00 00 00 00
   143 		 "punpcklbw %%mm5,%%mm6\n"      //  R3 R2 R1 R0 r3 r2 r1 r0
   144 
   145 		 // create Cb (result in mm1)
   146 		 "movd (%1), %%mm1\n"      //         0  0  0  0  u3 u2 u1 u0
   147 		 "punpcklbw %%mm7,%%mm1\n" //         0  u3 0  u2 00 u1 00 u0
   148 		 "punpckldq %%mm1,%%mm1\n" //         00 u1 00 u0 00 u1 00 u0
   149 		 "psubw %[_MMX_0080w],%%mm1\n"  // mm1-128:u1 u1 u0 u0 u1 u1 u0 u0 
   150 		 // create Cb_g (result in mm5)
   151 		 "movq %%mm1,%%mm5\n"            // u1 u1 u0 u0 u1 u1 u0 u0
   152 		 "pmullw %[_MMX_UgrnRGB],%%mm5\n"    // blue*-109dec=1.7129*64
   153 		 "pmullw %[_MMX_UbluRGB],%%mm1\n"    // blue*114dec=1.78125*64
   154 		 "psraw  $6, %%mm5\n"            // blue=red/64
   155 		 "psraw  $6, %%mm1\n"            // blue=blue/64
   156 
   157 		 // create G (result in mm7)
   158 		 "movq %%mm2,%%mm3\n"      //   0  L3  0 L1  0 l3  0 l1
   159 		 "movq %%mm4,%%mm7\n"      //   0  L2  0 L0  0 l2  0 l1
   160 		 "paddsw  %%mm5, %%mm3\n"  // lum1+Cb_g:x G3t x G1t x g3t x g1t
   161 		 "paddsw  %%mm5, %%mm7\n"  // lum1+Cb_g:x G2t x G0t x g2t x g0t
   162 		 "paddsw  %%mm0, %%mm3\n"  // lum1+Cr_g:x G3  x G1  x g3  x g1
   163 		 "paddsw  %%mm0, %%mm7\n"  // lum1+blue:x G2  x G0  x g2  x g0
   164 		 "packuswb %%mm3,%%mm3\n"  // G3 G1 g3 g1 G3 G1 g3 g1
   165 		 "packuswb %%mm7,%%mm7\n"  // G2 G0 g2 g0 G2 G0 g2 g0
   166 		 "punpcklbw %%mm3,%%mm7\n" // G3 G2 G1 G0 g3 g2 g1 g0
   167 		 
   168 		 // create B (result in mm5)
   169 		 "movq %%mm2,%%mm3\n"         //   0  L3  0 L1  0 l3  0 l1
   170 		 "movq %%mm4,%%mm5\n"         //   0  L2  0 L0  0 l2  0 l1
   171 		 "paddsw  %%mm1, %%mm3\n"     // lum1+blue:x B3 x B1 x b3 x b1
   172 		 "paddsw  %%mm1, %%mm5\n"     // lum1+blue:x B2 x B0 x b2 x b0
   173 		 "packuswb %%mm3,%%mm3\n"     // B3 B1 b3 b1 B3 B1 b3 b1
   174 		 "packuswb %%mm5,%%mm5\n"     // B2 B0 b2 b0 B2 B0 b2 b0
   175 		 "punpcklbw %%mm3,%%mm5\n"    // B3 B2 B1 B0 b3 b2 b1 b0
   176 
   177 		 // fill destination row1 (needed are mm6=Rr,mm7=Gg,mm5=Bb)
   178 
   179 		 "pxor %%mm2,%%mm2\n"           //  0  0  0  0  0  0  0  0
   180 		 "pxor %%mm4,%%mm4\n"           //  0  0  0  0  0  0  0  0
   181 		 "movq %%mm6,%%mm1\n"           // R3 R2 R1 R0 r3 r2 r1 r0
   182 		 "movq %%mm5,%%mm3\n"           // B3 B2 B1 B0 b3 b2 b1 b0
   183 		 // process lower lum
   184 		 "punpcklbw %%mm4,%%mm1\n"      //  0 r3  0 r2  0 r1  0 r0
   185 		 "punpcklbw %%mm4,%%mm3\n"      //  0 b3  0 b2  0 b1  0 b0
   186 		 "movq %%mm1,%%mm2\n"           //  0 r3  0 r2  0 r1  0 r0
   187 		 "movq %%mm3,%%mm0\n"           //  0 b3  0 b2  0 b1  0 b0
   188 		 "punpcklwd %%mm1,%%mm3\n"      //  0 r1  0 b1  0 r0  0 b0
   189 		 "punpckhwd %%mm2,%%mm0\n"      //  0 r3  0 b3  0 r2  0 b2
   190 
   191 		 "pxor %%mm2,%%mm2\n"           //  0  0  0  0  0  0  0  0
   192 		 "movq %%mm7,%%mm1\n"           // G3 G2 G1 G0 g3 g2 g1 g0
   193 		 "punpcklbw %%mm1,%%mm2\n"      // g3  0 g2  0 g1  0 g0  0
   194 		 "punpcklwd %%mm4,%%mm2\n"      //  0  0 g1  0  0  0 g0  0 
   195 		 "por  %%mm3, %%mm2\n"          //  0 r1 g1 b1  0 r0 g0 b0
   196 		 "movq   %%mm2,(%3)\n"          // wrote out ! row1
   197 
   198 		 "pxor %%mm2,%%mm2\n"           //  0  0  0  0  0  0  0  0
   199 		 "punpcklbw %%mm1,%%mm4\n"      // g3  0 g2  0 g1  0 g0  0
   200 		 "punpckhwd %%mm2,%%mm4\n"      //  0  0 g3  0  0  0 g2  0 
   201 		 "por  %%mm0, %%mm4\n"          //  0 r3 g3 b3  0 r2 g2 b2
   202 		 "movq   %%mm4,8(%3)\n"         // wrote out ! row1
   203 		 
   204 		 // fill destination row2 (needed are mm6=Rr,mm7=Gg,mm5=Bb)
   205 		 // this can be done "destructive"
   206 		 "pxor %%mm2,%%mm2\n"           //  0  0  0  0  0  0  0  0
   207 		 "punpckhbw %%mm2,%%mm6\n"      //  0 R3  0 R2  0 R1  0 R0
   208 		 "punpckhbw %%mm1,%%mm5\n"      // G3 B3 G2 B2 G1 B1 G0 B0
   209 		 "movq %%mm5,%%mm1\n"           // G3 B3 G2 B2 G1 B1 G0 B0
   210 		 "punpcklwd %%mm6,%%mm1\n"      //  0 R1 G1 B1  0 R0 G0 B0
   211 		 "movq   %%mm1,(%5)\n"          // wrote out ! row2
   212 		 "punpckhwd %%mm6,%%mm5\n"      //  0 R3 G3 B3  0 R2 G2 B2
   213 		 "movq   %%mm5,8(%5)\n"         // wrote out ! row2
   214 		 
   215 		 "addl  $4,%2\n"            // lum+4
   216 		 "leal  16(%3),%3\n"        // row1+16
   217 		 "leal  16(%5),%5\n"        // row2+16
   218 		 "addl  $2, %%ebx\n"        // cr+2
   219 		 "addl  $2, %1\n"           // cb+2
   220 
   221 		 "addl  $4,%6\n"            // x+4
   222 		 "cmpl  %4,%6\n"
   223 
   224 		 "jl    1b\n"
   225 		 "addl           %4,     %2\n" // lum += cols 
   226 		 "addl           %8,     %3\n" // row1+= mod
   227 		 "addl           %8,     %5\n" // row2+= mod
   228 		 "movl           $0,     %6\n" // x=0
   229 		 "cmpl           %7,     %2\n"
   230 		 "jl             1b\n"
   231 		 "emms\n"
   232 		 "popl %%ebx\n"
   233 		 :
   234 		 : "m" (cr), "r"(cb),"r"(lum),
   235 		 "r"(row1),"r"(cols),"r"(row2),"m"(x),"m"(y),"m"(mod),
   236          [_MMX_0080w] "m" (*MMX_0080w),
   237          [_MMX_00FFw] "m" (*MMX_00FFw),
   238          [_MMX_FF00w] "m" (*MMX_FF00w),
   239          [_MMX_VgrnRGB] "m" (*MMX_VgrnRGB),
   240          [_MMX_VredRGB] "m" (*MMX_VredRGB),
   241          [_MMX_UgrnRGB] "m" (*MMX_UgrnRGB),
   242          [_MMX_UbluRGB] "m" (*MMX_UbluRGB)
   243 		 );
   244 }
   245 
   246 void Color565DitherYV12MMX1X( int *colortab, Uint32 *rgb_2_pix,
   247                              unsigned char *lum, unsigned char *cr,
   248                              unsigned char *cb, unsigned char *out,
   249                              int rows, int cols, int mod )
   250 {
   251     Uint16 *row1;
   252     Uint16 *row2;
   253 
   254     unsigned char* y = lum +cols*rows;    /* Pointer to the end */
   255     int x=0;
   256     row1 = (Uint16 *)out;                 /* 16 bit target */
   257     row2 = (Uint16 *)out+cols+mod;        /* start of second row  */
   258     mod = (mod+cols+mod)*2;               /* increment for row1 in byte */
   259 
   260 
   261       __asm__ __volatile__(
   262          "pushl %%ebx\n"
   263 	 "movl %0, %%ebx\n"
   264 
   265          ".align 8\n"
   266          "1:\n"
   267          "movd           (%1),                   %%mm0\n" // 4 Cb         0  0  0  0 u3 u2 u1 u0
   268          "pxor           %%mm7,                  %%mm7\n"
   269          "movd           (%%ebx),                %%mm1\n" // 4 Cr                0  0  0  0 v3 v2 v1 v0
   270          "punpcklbw      %%mm7,                  %%mm0\n" // 4 W cb   0 u3  0 u2  0 u1  0 u0
   271          "punpcklbw      %%mm7,                  %%mm1\n" // 4 W cr   0 v3  0 v2  0 v1  0 v0
   272          "psubw          _MMX_0080w,             %%mm0\n"
   273          "psubw          _MMX_0080w,             %%mm1\n"
   274          "movq           %%mm0,                  %%mm2\n" // Cb                   0 u3  0 u2  0 u1  0 u0
   275          "movq           %%mm1,                  %%mm3\n" // Cr
   276          "pmullw         _MMX_Ugrn565,           %%mm2\n" // Cb2green 0 R3  0 R2  0 R1  0 R0
   277          "movq           (%2),                   %%mm6\n" // L1      l7 L6 L5 L4 L3 L2 L1 L0
   278          "pmullw         _MMX_Ublu5x5,           %%mm0\n" // Cb2blue
   279          "pand           _MMX_00FFw,             %%mm6\n" // L1      00 L6 00 L4 00 L2 00 L0
   280          "pmullw         _MMX_Vgrn565,           %%mm3\n" // Cr2green
   281          "movq           (%2),                   %%mm7\n" // L2
   282          "pmullw         _MMX_Vred5x5,           %%mm1\n" // Cr2red
   283          "psrlw          $8,                     %%mm7\n"        // L2           00 L7 00 L5 00 L3 00 L1
   284          "pmullw         _MMX_Ycoeff,            %%mm6\n" // lum1
   285          "paddw          %%mm3,                  %%mm2\n" // Cb2green + Cr2green == green
   286          "pmullw         _MMX_Ycoeff,            %%mm7\n" // lum2
   287 
   288          "movq           %%mm6,                  %%mm4\n" // lum1
   289          "paddw          %%mm0,                  %%mm6\n" // lum1 +blue 00 B6 00 B4 00 B2 00 B0
   290          "movq           %%mm4,                  %%mm5\n" // lum1
   291          "paddw          %%mm1,                  %%mm4\n" // lum1 +red  00 R6 00 R4 00 R2 00 R0
   292          "paddw          %%mm2,                  %%mm5\n" // lum1 +green 00 G6 00 G4 00 G2 00 G0
   293          "psraw          $6,                     %%mm4\n" // R1 0 .. 64
   294          "movq           %%mm7,                  %%mm3\n" // lum2                       00 L7 00 L5 00 L3 00 L1
   295          "psraw          $6,                     %%mm5\n" // G1  - .. +
   296          "paddw          %%mm0,                  %%mm7\n" // Lum2 +blue 00 B7 00 B5 00 B3 00 B1
   297          "psraw          $6,                     %%mm6\n" // B1         0 .. 64
   298          "packuswb       %%mm4,                  %%mm4\n" // R1 R1
   299          "packuswb       %%mm5,                  %%mm5\n" // G1 G1
   300          "packuswb       %%mm6,                  %%mm6\n" // B1 B1
   301          "punpcklbw      %%mm4,                  %%mm4\n"
   302          "punpcklbw      %%mm5,                  %%mm5\n"
   303 
   304          "pand           _MMX_red565,            %%mm4\n"
   305          "psllw          $3,                     %%mm5\n" // GREEN       1
   306          "punpcklbw      %%mm6,                  %%mm6\n"
   307          "pand           _MMX_grn565,            %%mm5\n"
   308          "pand           _MMX_red565,            %%mm6\n"
   309          "por            %%mm5,                  %%mm4\n" //
   310          "psrlw          $11,                    %%mm6\n" // BLUE        1
   311          "movq           %%mm3,                  %%mm5\n" // lum2
   312          "paddw          %%mm1,                  %%mm3\n" // lum2 +red      00 R7 00 R5 00 R3 00 R1
   313          "paddw          %%mm2,                  %%mm5\n" // lum2 +green 00 G7 00 G5 00 G3 00 G1
   314          "psraw          $6,                     %%mm3\n" // R2
   315          "por            %%mm6,                  %%mm4\n" // MM4
   316          "psraw          $6,                     %%mm5\n" // G2
   317          "movq           (%2, %4),               %%mm6\n" // L3 load lum2
   318          "psraw          $6,                     %%mm7\n"
   319          "packuswb       %%mm3,                  %%mm3\n"
   320          "packuswb       %%mm5,                  %%mm5\n"
   321          "packuswb       %%mm7,                  %%mm7\n"
   322          "pand           _MMX_00FFw,             %%mm6\n" // L3
   323          "punpcklbw      %%mm3,                  %%mm3\n"
   324          "punpcklbw      %%mm5,                  %%mm5\n"
   325          "pmullw         _MMX_Ycoeff,            %%mm6\n" // lum3
   326          "punpcklbw      %%mm7,                  %%mm7\n"
   327          "psllw          $3,                     %%mm5\n" // GREEN 2
   328          "pand           _MMX_red565,            %%mm7\n"
   329          "pand           _MMX_red565,            %%mm3\n"
   330          "psrlw          $11,                    %%mm7\n" // BLUE  2
   331          "pand           _MMX_grn565,            %%mm5\n"
   332          "por            %%mm7,                  %%mm3\n"
   333          "movq           (%2,%4),                %%mm7\n" // L4 load lum2
   334          "por            %%mm5,                  %%mm3\n" //
   335          "psrlw          $8,                     %%mm7\n" // L4
   336          "movq           %%mm4,                  %%mm5\n"
   337          "punpcklwd      %%mm3,                  %%mm4\n"
   338          "pmullw         _MMX_Ycoeff,            %%mm7\n" // lum4
   339          "punpckhwd      %%mm3,                  %%mm5\n"
   340 
   341          "movq           %%mm4,                  (%3)\n"  // write row1
   342          "movq           %%mm5,                  8(%3)\n" // write row1
   343 
   344          "movq           %%mm6,                  %%mm4\n" // Lum3
   345          "paddw          %%mm0,                  %%mm6\n" // Lum3 +blue
   346 
   347          "movq           %%mm4,                  %%mm5\n" // Lum3
   348          "paddw          %%mm1,                  %%mm4\n" // Lum3 +red
   349          "paddw          %%mm2,                  %%mm5\n" // Lum3 +green
   350          "psraw          $6,                     %%mm4\n"
   351          "movq           %%mm7,                  %%mm3\n" // Lum4
   352          "psraw          $6,                     %%mm5\n"
   353          "paddw          %%mm0,                  %%mm7\n" // Lum4 +blue
   354          "psraw          $6,                     %%mm6\n" // Lum3 +blue
   355          "movq           %%mm3,                  %%mm0\n" // Lum4
   356          "packuswb       %%mm4,                  %%mm4\n"
   357          "paddw          %%mm1,                  %%mm3\n" // Lum4 +red
   358          "packuswb       %%mm5,                  %%mm5\n"
   359          "paddw          %%mm2,                  %%mm0\n" // Lum4 +green
   360          "packuswb       %%mm6,                  %%mm6\n"
   361          "punpcklbw      %%mm4,                  %%mm4\n"
   362          "punpcklbw      %%mm5,                  %%mm5\n"
   363          "punpcklbw      %%mm6,                  %%mm6\n"
   364          "psllw          $3,                     %%mm5\n" // GREEN 3
   365          "pand           _MMX_red565,            %%mm4\n"
   366          "psraw          $6,                     %%mm3\n" // psr 6
   367          "psraw          $6,                     %%mm0\n"
   368          "pand           _MMX_red565,            %%mm6\n" // BLUE
   369          "pand           _MMX_grn565,            %%mm5\n"
   370          "psrlw          $11,                    %%mm6\n" // BLUE  3
   371          "por            %%mm5,                  %%mm4\n"
   372          "psraw          $6,                     %%mm7\n"
   373          "por            %%mm6,                  %%mm4\n"
   374          "packuswb       %%mm3,                  %%mm3\n"
   375          "packuswb       %%mm0,                  %%mm0\n"
   376          "packuswb       %%mm7,                  %%mm7\n"
   377          "punpcklbw      %%mm3,                  %%mm3\n"
   378          "punpcklbw      %%mm0,                  %%mm0\n"
   379          "punpcklbw      %%mm7,                  %%mm7\n"
   380          "pand           _MMX_red565,            %%mm3\n"
   381          "pand           _MMX_red565,            %%mm7\n" // BLUE
   382          "psllw          $3,                     %%mm0\n" // GREEN 4
   383          "psrlw          $11,                    %%mm7\n"
   384          "pand           _MMX_grn565,            %%mm0\n"
   385          "por            %%mm7,                  %%mm3\n"
   386          "por            %%mm0,                  %%mm3\n"
   387 
   388          "movq           %%mm4,                  %%mm5\n"
   389 
   390          "punpcklwd      %%mm3,                  %%mm4\n"
   391          "punpckhwd      %%mm3,                  %%mm5\n"
   392 
   393          "movq           %%mm4,                  (%5)\n"
   394 	 "movq           %%mm5,                  8(%5)\n"
   395 
   396          "addl           $8,                     %6\n"
   397          "addl           $8,                     %2\n"
   398          "addl           $4,                     %%ebx\n"
   399          "addl           $4,                     %1\n"
   400          "cmpl           %4,                     %6\n"
   401          "leal           16(%3),                 %3\n"
   402 	 "leal           16(%5),%5\n" // row2+16
   403 
   404 
   405          "jl             1b\n"
   406 	 "addl           %4,     %2\n" // lum += cols 
   407 	 "addl           %8,     %3\n" // row1+= mod
   408 	 "addl           %8,     %5\n" // row2+= mod
   409 	 "movl           $0,     %6\n" // x=0
   410 	 "cmpl           %7,     %2\n"
   411 	 "jl             1b\n"
   412          "emms\n"
   413 	 "popl %%ebx\n"
   414 	:
   415 	:"m" (cr), "r"(cb),"r"(lum),
   416 	 "r"(row1),"r"(cols),"r"(row2),"m"(x),"m"(y),"m"(mod),
   417 	[_MMX_0080w] "m" (*MMX_0080w),
   418 	[_MMX_Ugrn565] "m" (*MMX_Ugrn565),
   419 	[_MMX_Ublu5x5] "m" (*MMX_Ublu5x5),
   420 	[_MMX_00FFw] "m" (*MMX_00FFw),
   421 	[_MMX_Vgrn565] "m" (*MMX_Vgrn565),
   422 	[_MMX_Vred5x5] "m" (*MMX_Vred5x5),
   423 	[_MMX_Ycoeff] "m" (*MMX_Ycoeff),
   424 	[_MMX_red565] "m" (*MMX_red565),
   425 	[_MMX_grn565] "m" (*MMX_grn565)
   426          );
   427 }
   428 
   429 #endif /* GCC i386 inline assembly */