src/video/SDL_yuv_mmx.c
author Ryan C. Gordon <icculus@icculus.org>
Fri, 06 Jan 2006 13:20:10 +0000
changeset 1234 73676c1f56ee
parent 1148 63fb2da89a4b
child 1312 c9b51268668f
permissions -rw-r--r--
For sanity's sake, removed the '&' when passing copy_row array to asm.
     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 #define ASM_ARRAY(x) x[] __asm__("_" #x) __attribute__((used))
    34  
    35 static unsigned int  ASM_ARRAY(MMX_0080w)    = {0x00800080, 0x00800080};
    36 static unsigned int  ASM_ARRAY(MMX_00FFw)    = {0x00ff00ff, 0x00ff00ff};
    37 static unsigned int  ASM_ARRAY(MMX_FF00w)    = {0xff00ff00, 0xff00ff00};
    38 
    39 static unsigned short ASM_ARRAY(MMX_Ycoeff)  = {0x004a, 0x004a, 0x004a, 0x004a};
    40 
    41 static unsigned short ASM_ARRAY(MMX_UbluRGB) = {0x0072, 0x0072, 0x0072, 0x0072};
    42 static unsigned short ASM_ARRAY(MMX_VredRGB) = {0x0059, 0x0059, 0x0059, 0x0059};
    43 static unsigned short ASM_ARRAY(MMX_UgrnRGB) = {0xffea, 0xffea, 0xffea, 0xffea};
    44 static unsigned short ASM_ARRAY(MMX_VgrnRGB) = {0xffd2, 0xffd2, 0xffd2, 0xffd2};
    45 
    46 static unsigned short ASM_ARRAY(MMX_Ublu5x5) = {0x0081, 0x0081, 0x0081, 0x0081};
    47 static unsigned short ASM_ARRAY(MMX_Vred5x5) = {0x0066, 0x0066, 0x0066, 0x0066};
    48 static unsigned short ASM_ARRAY(MMX_Ugrn555) = {0xffe7, 0xffe7, 0xffe7, 0xffe7};
    49 static unsigned short ASM_ARRAY(MMX_Vgrn555) = {0xffcc, 0xffcc, 0xffcc, 0xffcc};
    50 static unsigned short ASM_ARRAY(MMX_Ugrn565) = {0xffe8, 0xffe8, 0xffe8, 0xffe8};
    51 static unsigned short ASM_ARRAY(MMX_Vgrn565) = {0xffcd, 0xffcd, 0xffcd, 0xffcd};
    52 
    53 static unsigned short ASM_ARRAY(MMX_red555)  = {0x7c00, 0x7c00, 0x7c00, 0x7c00};
    54 static unsigned short ASM_ARRAY(MMX_red565)  = {0xf800, 0xf800, 0xf800, 0xf800};
    55 static unsigned short ASM_ARRAY(MMX_grn555)  = {0x03e0, 0x03e0, 0x03e0, 0x03e0};
    56 static unsigned short ASM_ARRAY(MMX_grn565)  = {0x07e0, 0x07e0, 0x07e0, 0x07e0};
    57 static unsigned short ASM_ARRAY(MMX_blu5x5)  = {0x001f, 0x001f, 0x001f, 0x001f};
    58 
    59 /**
    60    This MMX assembler is my first assembler/MMX program ever.
    61    Thus it maybe buggy.
    62    Send patches to:
    63    mvogt@rhrk.uni-kl.de
    64 
    65    After it worked fine I have "obfuscated" the code a bit to have
    66    more parallism in the MMX units. This means I moved
    67    initilisation around and delayed other instruction.
    68    Performance measurement did not show that this brought any advantage
    69    but in theory it _should_ be faster this way.
    70 
    71    The overall performanve gain to the C based dither was 30%-40%.
    72    The MMX routine calculates 256bit=8RGB values in each cycle
    73    (4 for row1 & 4 for row2)
    74 
    75    The red/green/blue.. coefficents are taken from the mpeg_play 
    76    player. They look nice, but I dont know if you can have
    77    better values, to avoid integer rounding errors.
    78    
    79 
    80    IMPORTANT:
    81    ==========
    82 
    83    It is a requirement that the cr/cb/lum are 8 byte aligned and
    84    the out are 16byte aligned or you will/may get segfaults
    85 
    86 */
    87 
    88 void ColorRGBDitherYV12MMX1X( int *colortab, Uint32 *rgb_2_pix,
    89                               unsigned char *lum, unsigned char *cr,
    90                               unsigned char *cb, unsigned char *out,
    91                               int rows, int cols, int mod )
    92 {
    93     Uint32 *row1;
    94     Uint32 *row2;
    95 
    96     unsigned char* y = lum +cols*rows;    // Pointer to the end
    97     int x=0;
    98     row1 = (Uint32 *)out;                 // 32 bit target
    99     row2 = (Uint32 *)out+cols+mod;        // start of second row 
   100     mod = (mod+cols+mod)*4;               // increment for row1 in byte
   101 
   102     __asm__ __volatile__ (
   103 /* We don't really care about PIC - the code should be rewritten to use
   104    relative addressing for the static tables, so right now we take the
   105    COW hit on the pages this code resides. Big deal.
   106    This spill is just to reduce register pressure in the PIC case. */
   107 		 "pushl %%ebx\n"
   108 		 "movl %0, %%ebx\n"
   109 
   110 	         ".align 8\n"
   111 		 "1:\n"
   112 		
   113 		 // create Cr (result in mm1)
   114 		 "movd (%%ebx), %%mm1\n"   //         0  0  0  0  v3 v2 v1 v0
   115 		 "pxor %%mm7,%%mm7\n"      //         00 00 00 00 00 00 00 00
   116 		 "movd (%2), %%mm2\n"           //    0  0  0  0 l3 l2 l1 l0
   117 		 "punpcklbw %%mm7,%%mm1\n" //         0  v3 0  v2 00 v1 00 v0
   118 		 "punpckldq %%mm1,%%mm1\n" //         00 v1 00 v0 00 v1 00 v0
   119 		 "psubw _MMX_0080w,%%mm1\n"  // mm1-128:r1 r1 r0 r0 r1 r1 r0 r0 
   120 
   121 		 // create Cr_g (result in mm0)
   122 		 "movq %%mm1,%%mm0\n"           // r1 r1 r0 r0 r1 r1 r0 r0
   123 		 "pmullw _MMX_VgrnRGB,%%mm0\n"// red*-46dec=0.7136*64
   124 		 "pmullw _MMX_VredRGB,%%mm1\n"// red*89dec=1.4013*64
   125 		 "psraw  $6, %%mm0\n"           // red=red/64
   126 		 "psraw  $6, %%mm1\n"           // red=red/64
   127 		 
   128 		 // create L1 L2 (result in mm2,mm4)
   129 		 // L2=lum+cols
   130 		 "movq (%2,%4),%%mm3\n"         //    0  0  0  0 L3 L2 L1 L0
   131 		 "punpckldq %%mm3,%%mm2\n"      //   L3 L2 L1 L0 l3 l2 l1 l0
   132 		 "movq %%mm2,%%mm4\n"           //   L3 L2 L1 L0 l3 l2 l1 l0
   133 		 "pand _MMX_FF00w,%%mm2\n"      //   L3 0  L1  0 l3  0 l1  0
   134 		 "pand _MMX_00FFw,%%mm4\n"      //   0  L2  0 L0  0 l2  0 l0
   135 		 "psrlw $8,%%mm2\n"             //   0  L3  0 L1  0 l3  0 l1
   136 
   137 		 // create R (result in mm6)
   138 		 "movq %%mm2,%%mm5\n"           //   0 L3  0 L1  0 l3  0 l1
   139 		 "movq %%mm4,%%mm6\n"           //   0 L2  0 L0  0 l2  0 l0
   140 		 "paddsw  %%mm1, %%mm5\n"       // lum1+red:x R3 x R1 x r3 x r1
   141 		 "paddsw  %%mm1, %%mm6\n"       // lum1+red:x R2 x R0 x r2 x r0
   142 		 "packuswb %%mm5,%%mm5\n"       //  R3 R1 r3 r1 R3 R1 r3 r1
   143 		 "packuswb %%mm6,%%mm6\n"       //  R2 R0 r2 r0 R2 R0 r2 r0
   144 		 "pxor %%mm7,%%mm7\n"      //         00 00 00 00 00 00 00 00
   145 		 "punpcklbw %%mm5,%%mm6\n"      //  R3 R2 R1 R0 r3 r2 r1 r0
   146 
   147 		 // create Cb (result in mm1)
   148 		 "movd (%1), %%mm1\n"      //         0  0  0  0  u3 u2 u1 u0
   149 		 "punpcklbw %%mm7,%%mm1\n" //         0  u3 0  u2 00 u1 00 u0
   150 		 "punpckldq %%mm1,%%mm1\n" //         00 u1 00 u0 00 u1 00 u0
   151 		 "psubw _MMX_0080w,%%mm1\n"  // mm1-128:u1 u1 u0 u0 u1 u1 u0 u0 
   152 		 // create Cb_g (result in mm5)
   153 		 "movq %%mm1,%%mm5\n"            // u1 u1 u0 u0 u1 u1 u0 u0
   154 		 "pmullw _MMX_UgrnRGB,%%mm5\n"    // blue*-109dec=1.7129*64
   155 		 "pmullw _MMX_UbluRGB,%%mm1\n"    // blue*114dec=1.78125*64
   156 		 "psraw  $6, %%mm5\n"            // blue=red/64
   157 		 "psraw  $6, %%mm1\n"            // blue=blue/64
   158 
   159 		 // create G (result in mm7)
   160 		 "movq %%mm2,%%mm3\n"      //   0  L3  0 L1  0 l3  0 l1
   161 		 "movq %%mm4,%%mm7\n"      //   0  L2  0 L0  0 l2  0 l1
   162 		 "paddsw  %%mm5, %%mm3\n"  // lum1+Cb_g:x G3t x G1t x g3t x g1t
   163 		 "paddsw  %%mm5, %%mm7\n"  // lum1+Cb_g:x G2t x G0t x g2t x g0t
   164 		 "paddsw  %%mm0, %%mm3\n"  // lum1+Cr_g:x G3  x G1  x g3  x g1
   165 		 "paddsw  %%mm0, %%mm7\n"  // lum1+blue:x G2  x G0  x g2  x g0
   166 		 "packuswb %%mm3,%%mm3\n"  // G3 G1 g3 g1 G3 G1 g3 g1
   167 		 "packuswb %%mm7,%%mm7\n"  // G2 G0 g2 g0 G2 G0 g2 g0
   168 		 "punpcklbw %%mm3,%%mm7\n" // G3 G2 G1 G0 g3 g2 g1 g0
   169 		 
   170 		 // create B (result in mm5)
   171 		 "movq %%mm2,%%mm3\n"         //   0  L3  0 L1  0 l3  0 l1
   172 		 "movq %%mm4,%%mm5\n"         //   0  L2  0 L0  0 l2  0 l1
   173 		 "paddsw  %%mm1, %%mm3\n"     // lum1+blue:x B3 x B1 x b3 x b1
   174 		 "paddsw  %%mm1, %%mm5\n"     // lum1+blue:x B2 x B0 x b2 x b0
   175 		 "packuswb %%mm3,%%mm3\n"     // B3 B1 b3 b1 B3 B1 b3 b1
   176 		 "packuswb %%mm5,%%mm5\n"     // B2 B0 b2 b0 B2 B0 b2 b0
   177 		 "punpcklbw %%mm3,%%mm5\n"    // B3 B2 B1 B0 b3 b2 b1 b0
   178 
   179 		 // fill destination row1 (needed are mm6=Rr,mm7=Gg,mm5=Bb)
   180 
   181 		 "pxor %%mm2,%%mm2\n"           //  0  0  0  0  0  0  0  0
   182 		 "pxor %%mm4,%%mm4\n"           //  0  0  0  0  0  0  0  0
   183 		 "movq %%mm6,%%mm1\n"           // R3 R2 R1 R0 r3 r2 r1 r0
   184 		 "movq %%mm5,%%mm3\n"           // B3 B2 B1 B0 b3 b2 b1 b0
   185 		 // process lower lum
   186 		 "punpcklbw %%mm4,%%mm1\n"      //  0 r3  0 r2  0 r1  0 r0
   187 		 "punpcklbw %%mm4,%%mm3\n"      //  0 b3  0 b2  0 b1  0 b0
   188 		 "movq %%mm1,%%mm2\n"           //  0 r3  0 r2  0 r1  0 r0
   189 		 "movq %%mm3,%%mm0\n"           //  0 b3  0 b2  0 b1  0 b0
   190 		 "punpcklwd %%mm1,%%mm3\n"      //  0 r1  0 b1  0 r0  0 b0
   191 		 "punpckhwd %%mm2,%%mm0\n"      //  0 r3  0 b3  0 r2  0 b2
   192 
   193 		 "pxor %%mm2,%%mm2\n"           //  0  0  0  0  0  0  0  0
   194 		 "movq %%mm7,%%mm1\n"           // G3 G2 G1 G0 g3 g2 g1 g0
   195 		 "punpcklbw %%mm1,%%mm2\n"      // g3  0 g2  0 g1  0 g0  0
   196 		 "punpcklwd %%mm4,%%mm2\n"      //  0  0 g1  0  0  0 g0  0 
   197 		 "por  %%mm3, %%mm2\n"          //  0 r1 g1 b1  0 r0 g0 b0
   198 		 "movq   %%mm2,(%3)\n"          // wrote out ! row1
   199 
   200 		 "pxor %%mm2,%%mm2\n"           //  0  0  0  0  0  0  0  0
   201 		 "punpcklbw %%mm1,%%mm4\n"      // g3  0 g2  0 g1  0 g0  0
   202 		 "punpckhwd %%mm2,%%mm4\n"      //  0  0 g3  0  0  0 g2  0 
   203 		 "por  %%mm0, %%mm4\n"          //  0 r3 g3 b3  0 r2 g2 b2
   204 		 "movq   %%mm4,8(%3)\n"         // wrote out ! row1
   205 		 
   206 		 // fill destination row2 (needed are mm6=Rr,mm7=Gg,mm5=Bb)
   207 		 // this can be done "destructive"
   208 		 "pxor %%mm2,%%mm2\n"           //  0  0  0  0  0  0  0  0
   209 		 "punpckhbw %%mm2,%%mm6\n"      //  0 R3  0 R2  0 R1  0 R0
   210 		 "punpckhbw %%mm1,%%mm5\n"      // G3 B3 G2 B2 G1 B1 G0 B0
   211 		 "movq %%mm5,%%mm1\n"           // G3 B3 G2 B2 G1 B1 G0 B0
   212 		 "punpcklwd %%mm6,%%mm1\n"      //  0 R1 G1 B1  0 R0 G0 B0
   213 		 "movq   %%mm1,(%5)\n"          // wrote out ! row2
   214 		 "punpckhwd %%mm6,%%mm5\n"      //  0 R3 G3 B3  0 R2 G2 B2
   215 		 "movq   %%mm5,8(%5)\n"         // wrote out ! row2
   216 		 
   217 		 "addl  $4,%2\n"            // lum+4
   218 		 "leal  16(%3),%3\n"        // row1+16
   219 		 "leal  16(%5),%5\n"        // row2+16
   220 		 "addl  $2, %%ebx\n"        // cr+2
   221 		 "addl  $2, %1\n"           // cb+2
   222 
   223 		 "addl  $4,%6\n"            // x+4
   224 		 "cmpl  %4,%6\n"
   225 
   226 		 "jl    1b\n"
   227 		 "addl           %4,     %2\n" // lum += cols 
   228 		 "addl           %8,     %3\n" // row1+= mod
   229 		 "addl           %8,     %5\n" // row2+= mod
   230 		 "movl           $0,     %6\n" // x=0
   231 		 "cmpl           %7,     %2\n"
   232 		 "jl             1b\n"
   233 		 "emms\n"
   234 		 "popl %%ebx\n"
   235 		 :
   236 		 : "m" (cr), "r"(cb),"r"(lum),
   237 		 "r"(row1),"r"(cols),"r"(row2),"m"(x),"m"(y),"m"(mod));
   238 }
   239 
   240 void Color565DitherYV12MMX1X( int *colortab, Uint32 *rgb_2_pix,
   241                              unsigned char *lum, unsigned char *cr,
   242                              unsigned char *cb, unsigned char *out,
   243                              int rows, int cols, int mod )
   244 {
   245     Uint16 *row1;
   246     Uint16 *row2;
   247 
   248     unsigned char* y = lum +cols*rows;    /* Pointer to the end */
   249     int x=0;
   250     row1 = (Uint16 *)out;                 /* 16 bit target */
   251     row2 = (Uint16 *)out+cols+mod;        /* start of second row  */
   252     mod = (mod+cols+mod)*2;               /* increment for row1 in byte */
   253 
   254 
   255       __asm__ __volatile__(
   256          "pushl %%ebx\n"
   257 	 "movl %0, %%ebx\n"
   258 
   259          ".align 8\n"
   260          "1:\n"
   261          "movd           (%1),                   %%mm0\n" // 4 Cb         0  0  0  0 u3 u2 u1 u0
   262          "pxor           %%mm7,                  %%mm7\n"
   263          "movd           (%%ebx),                %%mm1\n" // 4 Cr                0  0  0  0 v3 v2 v1 v0
   264          "punpcklbw      %%mm7,                  %%mm0\n" // 4 W cb   0 u3  0 u2  0 u1  0 u0
   265          "punpcklbw      %%mm7,                  %%mm1\n" // 4 W cr   0 v3  0 v2  0 v1  0 v0
   266          "psubw          _MMX_0080w,             %%mm0\n"
   267          "psubw          _MMX_0080w,             %%mm1\n"
   268          "movq           %%mm0,                  %%mm2\n" // Cb                   0 u3  0 u2  0 u1  0 u0
   269          "movq           %%mm1,                  %%mm3\n" // Cr
   270          "pmullw         _MMX_Ugrn565,           %%mm2\n" // Cb2green 0 R3  0 R2  0 R1  0 R0
   271          "movq           (%2),                   %%mm6\n" // L1      l7 L6 L5 L4 L3 L2 L1 L0
   272          "pmullw         _MMX_Ublu5x5,           %%mm0\n" // Cb2blue
   273          "pand           _MMX_00FFw,             %%mm6\n" // L1      00 L6 00 L4 00 L2 00 L0
   274          "pmullw         _MMX_Vgrn565,           %%mm3\n" // Cr2green
   275          "movq           (%2),                   %%mm7\n" // L2
   276          "pmullw         _MMX_Vred5x5,           %%mm1\n" // Cr2red
   277          "psrlw          $8,                     %%mm7\n"        // L2           00 L7 00 L5 00 L3 00 L1
   278          "pmullw         _MMX_Ycoeff,            %%mm6\n" // lum1
   279          "paddw          %%mm3,                  %%mm2\n" // Cb2green + Cr2green == green
   280          "pmullw         _MMX_Ycoeff,            %%mm7\n" // lum2
   281 
   282          "movq           %%mm6,                  %%mm4\n" // lum1
   283          "paddw          %%mm0,                  %%mm6\n" // lum1 +blue 00 B6 00 B4 00 B2 00 B0
   284          "movq           %%mm4,                  %%mm5\n" // lum1
   285          "paddw          %%mm1,                  %%mm4\n" // lum1 +red  00 R6 00 R4 00 R2 00 R0
   286          "paddw          %%mm2,                  %%mm5\n" // lum1 +green 00 G6 00 G4 00 G2 00 G0
   287          "psraw          $6,                     %%mm4\n" // R1 0 .. 64
   288          "movq           %%mm7,                  %%mm3\n" // lum2                       00 L7 00 L5 00 L3 00 L1
   289          "psraw          $6,                     %%mm5\n" // G1  - .. +
   290          "paddw          %%mm0,                  %%mm7\n" // Lum2 +blue 00 B7 00 B5 00 B3 00 B1
   291          "psraw          $6,                     %%mm6\n" // B1         0 .. 64
   292          "packuswb       %%mm4,                  %%mm4\n" // R1 R1
   293          "packuswb       %%mm5,                  %%mm5\n" // G1 G1
   294          "packuswb       %%mm6,                  %%mm6\n" // B1 B1
   295          "punpcklbw      %%mm4,                  %%mm4\n"
   296          "punpcklbw      %%mm5,                  %%mm5\n"
   297 
   298          "pand           _MMX_red565,            %%mm4\n"
   299          "psllw          $3,                     %%mm5\n" // GREEN       1
   300          "punpcklbw      %%mm6,                  %%mm6\n"
   301          "pand           _MMX_grn565,            %%mm5\n"
   302          "pand           _MMX_red565,            %%mm6\n"
   303          "por            %%mm5,                  %%mm4\n" //
   304          "psrlw          $11,                    %%mm6\n" // BLUE        1
   305          "movq           %%mm3,                  %%mm5\n" // lum2
   306          "paddw          %%mm1,                  %%mm3\n" // lum2 +red      00 R7 00 R5 00 R3 00 R1
   307          "paddw          %%mm2,                  %%mm5\n" // lum2 +green 00 G7 00 G5 00 G3 00 G1
   308          "psraw          $6,                     %%mm3\n" // R2
   309          "por            %%mm6,                  %%mm4\n" // MM4
   310          "psraw          $6,                     %%mm5\n" // G2
   311          "movq           (%2, %4),               %%mm6\n" // L3 load lum2
   312          "psraw          $6,                     %%mm7\n"
   313          "packuswb       %%mm3,                  %%mm3\n"
   314          "packuswb       %%mm5,                  %%mm5\n"
   315          "packuswb       %%mm7,                  %%mm7\n"
   316          "pand           _MMX_00FFw,             %%mm6\n" // L3
   317          "punpcklbw      %%mm3,                  %%mm3\n"
   318          "punpcklbw      %%mm5,                  %%mm5\n"
   319          "pmullw         _MMX_Ycoeff,            %%mm6\n" // lum3
   320          "punpcklbw      %%mm7,                  %%mm7\n"
   321          "psllw          $3,                     %%mm5\n" // GREEN 2
   322          "pand           _MMX_red565,            %%mm7\n"
   323          "pand           _MMX_red565,            %%mm3\n"
   324          "psrlw          $11,                    %%mm7\n" // BLUE  2
   325          "pand           _MMX_grn565,            %%mm5\n"
   326          "por            %%mm7,                  %%mm3\n"
   327          "movq           (%2,%4),                %%mm7\n" // L4 load lum2
   328          "por            %%mm5,                  %%mm3\n" //
   329          "psrlw          $8,                     %%mm7\n" // L4
   330          "movq           %%mm4,                  %%mm5\n"
   331          "punpcklwd      %%mm3,                  %%mm4\n"
   332          "pmullw         _MMX_Ycoeff,            %%mm7\n" // lum4
   333          "punpckhwd      %%mm3,                  %%mm5\n"
   334 
   335          "movq           %%mm4,                  (%3)\n"  // write row1
   336          "movq           %%mm5,                  8(%3)\n" // write row1
   337 
   338          "movq           %%mm6,                  %%mm4\n" // Lum3
   339          "paddw          %%mm0,                  %%mm6\n" // Lum3 +blue
   340 
   341          "movq           %%mm4,                  %%mm5\n" // Lum3
   342          "paddw          %%mm1,                  %%mm4\n" // Lum3 +red
   343          "paddw          %%mm2,                  %%mm5\n" // Lum3 +green
   344          "psraw          $6,                     %%mm4\n"
   345          "movq           %%mm7,                  %%mm3\n" // Lum4
   346          "psraw          $6,                     %%mm5\n"
   347          "paddw          %%mm0,                  %%mm7\n" // Lum4 +blue
   348          "psraw          $6,                     %%mm6\n" // Lum3 +blue
   349          "movq           %%mm3,                  %%mm0\n" // Lum4
   350          "packuswb       %%mm4,                  %%mm4\n"
   351          "paddw          %%mm1,                  %%mm3\n" // Lum4 +red
   352          "packuswb       %%mm5,                  %%mm5\n"
   353          "paddw          %%mm2,                  %%mm0\n" // Lum4 +green
   354          "packuswb       %%mm6,                  %%mm6\n"
   355          "punpcklbw      %%mm4,                  %%mm4\n"
   356          "punpcklbw      %%mm5,                  %%mm5\n"
   357          "punpcklbw      %%mm6,                  %%mm6\n"
   358          "psllw          $3,                     %%mm5\n" // GREEN 3
   359          "pand           _MMX_red565,            %%mm4\n"
   360          "psraw          $6,                     %%mm3\n" // psr 6
   361          "psraw          $6,                     %%mm0\n"
   362          "pand           _MMX_red565,            %%mm6\n" // BLUE
   363          "pand           _MMX_grn565,            %%mm5\n"
   364          "psrlw          $11,                    %%mm6\n" // BLUE  3
   365          "por            %%mm5,                  %%mm4\n"
   366          "psraw          $6,                     %%mm7\n"
   367          "por            %%mm6,                  %%mm4\n"
   368          "packuswb       %%mm3,                  %%mm3\n"
   369          "packuswb       %%mm0,                  %%mm0\n"
   370          "packuswb       %%mm7,                  %%mm7\n"
   371          "punpcklbw      %%mm3,                  %%mm3\n"
   372          "punpcklbw      %%mm0,                  %%mm0\n"
   373          "punpcklbw      %%mm7,                  %%mm7\n"
   374          "pand           _MMX_red565,            %%mm3\n"
   375          "pand           _MMX_red565,            %%mm7\n" // BLUE
   376          "psllw          $3,                     %%mm0\n" // GREEN 4
   377          "psrlw          $11,                    %%mm7\n"
   378          "pand           _MMX_grn565,            %%mm0\n"
   379          "por            %%mm7,                  %%mm3\n"
   380          "por            %%mm0,                  %%mm3\n"
   381 
   382          "movq           %%mm4,                  %%mm5\n"
   383 
   384          "punpcklwd      %%mm3,                  %%mm4\n"
   385          "punpckhwd      %%mm3,                  %%mm5\n"
   386 
   387          "movq           %%mm4,                  (%5)\n"
   388 	 "movq           %%mm5,                  8(%5)\n"
   389 
   390          "addl           $8,                     %6\n"
   391          "addl           $8,                     %2\n"
   392          "addl           $4,                     %%ebx\n"
   393          "addl           $4,                     %1\n"
   394          "cmpl           %4,                     %6\n"
   395          "leal           16(%3),                 %3\n"
   396 	 "leal           16(%5),%5\n" // row2+16
   397 
   398 
   399          "jl             1b\n"
   400 	 "addl           %4,     %2\n" // lum += cols 
   401 	 "addl           %8,     %3\n" // row1+= mod
   402 	 "addl           %8,     %5\n" // row2+= mod
   403 	 "movl           $0,     %6\n" // x=0
   404 	 "cmpl           %7,     %2\n"
   405 	 "jl             1b\n"
   406          "emms\n"
   407 	 "popl %%ebx\n"
   408          :
   409          :"m" (cr), "r"(cb),"r"(lum),
   410 	 "r"(row1),"r"(cols),"r"(row2),"m"(x),"m"(y),"m"(mod));
   411 }
   412 
   413 #endif /* GCC i386 inline assembly */