src/render/SDL_yuv_mmx.c
author Sam Lantinga <slouken@libsdl.org>
Fri, 11 Feb 2011 22:37:15 -0800
changeset 5262 b530ef003506
parent 5156 307ccc9c135e
child 5535 96594ac5fd1a
permissions -rw-r--r--
Happy 2011! :)
     1 /*
     2     SDL - Simple DirectMedia Layer
     3     Copyright (C) 1997-2011 Sam Lantinga
     4 
     5     This library is free software; you can redistribute it and/or
     6     modify it under the terms of the GNU Lesser General Public
     7     License as published by the Free Software Foundation; either
     8     version 2.1 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     Lesser General Public License for more details.
    14 
    15     You should have received a copy of the GNU Lesser General Public
    16     License along with this library; if not, write to the Free Software
    17     Foundation, Inc., 51 Franklin St, Fifth Floor, Boston, MA  02110-1301  USA
    18 
    19     Sam Lantinga
    20     slouken@libsdl.org
    21 */
    22 #include "SDL_config.h"
    23 
    24 #if (__GNUC__ > 2) && defined(__i386__) && __OPTIMIZE__ && SDL_ASSEMBLY_ROUTINES
    25 
    26 #include "SDL_stdinc.h"
    27 
    28 #include "mmx.h"
    29 
    30 /* *INDENT-OFF* */
    31 
    32 static mmx_t MMX_0080w    = { .ud = {0x00800080, 0x00800080} };
    33 static mmx_t MMX_00FFw    = { .ud = {0x00ff00ff, 0x00ff00ff} };
    34 static mmx_t MMX_FF00w    = { .ud = {0xff00ff00, 0xff00ff00} };
    35 
    36 static mmx_t MMX_Ycoeff   = { .uw = {0x004a, 0x004a, 0x004a, 0x004a} };
    37 
    38 static mmx_t MMX_UbluRGB  = { .uw = {0x0072, 0x0072, 0x0072, 0x0072} };
    39 static mmx_t MMX_VredRGB  = { .uw = {0x0059, 0x0059, 0x0059, 0x0059} };
    40 static mmx_t MMX_UgrnRGB  = { .uw = {0xffea, 0xffea, 0xffea, 0xffea} };
    41 static mmx_t MMX_VgrnRGB  = { .uw = {0xffd2, 0xffd2, 0xffd2, 0xffd2} };
    42 
    43 static mmx_t MMX_Ublu5x5  = { .uw = {0x0081, 0x0081, 0x0081, 0x0081} };
    44 static mmx_t MMX_Vred5x5  = { .uw = {0x0066, 0x0066, 0x0066, 0x0066} };
    45 static mmx_t MMX_Ugrn565  = { .uw = {0xffe8, 0xffe8, 0xffe8, 0xffe8} };
    46 static mmx_t MMX_Vgrn565  = { .uw = {0xffcd, 0xffcd, 0xffcd, 0xffcd} };
    47 
    48 static mmx_t MMX_red565   = { .uw = {0xf800, 0xf800, 0xf800, 0xf800} };
    49 static mmx_t MMX_grn565   = { .uw = {0x07e0, 0x07e0, 0x07e0, 0x07e0} };
    50 
    51 /**
    52    This MMX assembler is my first assembler/MMX program ever.
    53    Thus it maybe buggy.
    54    Send patches to:
    55    mvogt@rhrk.uni-kl.de
    56 
    57    After it worked fine I have "obfuscated" the code a bit to have
    58    more parallism in the MMX units. This means I moved
    59    initilisation around and delayed other instruction.
    60    Performance measurement did not show that this brought any advantage
    61    but in theory it _should_ be faster this way.
    62 
    63    The overall performanve gain to the C based dither was 30%-40%.
    64    The MMX routine calculates 256bit=8RGB values in each cycle
    65    (4 for row1 & 4 for row2)
    66 
    67    The red/green/blue.. coefficents are taken from the mpeg_play 
    68    player. They look nice, but I dont know if you can have
    69    better values, to avoid integer rounding errors.
    70    
    71 
    72    IMPORTANT:
    73    ==========
    74 
    75    It is a requirement that the cr/cb/lum are 8 byte aligned and
    76    the out are 16byte aligned or you will/may get segfaults
    77 
    78 */
    79 
    80 void ColorRGBDitherYV12MMX1X( int *colortab, Uint32 *rgb_2_pix,
    81                               unsigned char *lum, unsigned char *cr,
    82                               unsigned char *cb, unsigned char *out,
    83                               int rows, int cols, int mod )
    84 {
    85     Uint32 *row1;
    86     Uint32 *row2;
    87 
    88     unsigned char* y = lum +cols*rows;    // Pointer to the end
    89     int x = 0;
    90     row1 = (Uint32 *)out;                 // 32 bit target
    91     row2 = (Uint32 *)out+cols+mod;        // start of second row
    92     mod = (mod+cols+mod)*4;               // increment for row1 in byte
    93 
    94     __asm__ __volatile__ (
    95         // tap dance to workaround the inability to use %%ebx at will...
    96         //  move one thing to the stack...
    97         "pushl $0\n"  // save a slot on the stack.
    98         "pushl %%ebx\n"  // save %%ebx.
    99         "movl %0, %%ebx\n"  // put the thing in ebx.
   100         "movl %%ebx,4(%%esp)\n"  // put the thing in the stack slot.
   101         "popl %%ebx\n"  // get back %%ebx (the PIC register).
   102 
   103         ".align 8\n"
   104         "1:\n"
   105 
   106         // create Cr (result in mm1)
   107         "pushl %%ebx\n"
   108         "movl 4(%%esp),%%ebx\n"
   109         "movd (%%ebx),%%mm1\n"   //         0  0  0  0  v3 v2 v1 v0
   110         "popl %%ebx\n"
   111         "pxor %%mm7,%%mm7\n"      //         00 00 00 00 00 00 00 00
   112         "movd (%2), %%mm2\n"           //    0  0  0  0 l3 l2 l1 l0
   113         "punpcklbw %%mm7,%%mm1\n" //         0  v3 0  v2 00 v1 00 v0
   114         "punpckldq %%mm1,%%mm1\n" //         00 v1 00 v0 00 v1 00 v0
   115         "psubw %9,%%mm1\n"        // mm1-128:r1 r1 r0 r0 r1 r1 r0 r0
   116 
   117         // create Cr_g (result in mm0)
   118         "movq %%mm1,%%mm0\n"           // r1 r1 r0 r0 r1 r1 r0 r0
   119         "pmullw %10,%%mm0\n"           // red*-46dec=0.7136*64
   120         "pmullw %11,%%mm1\n"           // red*89dec=1.4013*64
   121         "psraw  $6, %%mm0\n"           // red=red/64
   122         "psraw  $6, %%mm1\n"           // red=red/64
   123 
   124         // create L1 L2 (result in mm2,mm4)
   125         // L2=lum+cols
   126         "movq (%2,%4),%%mm3\n"         //    0  0  0  0 L3 L2 L1 L0
   127         "punpckldq %%mm3,%%mm2\n"      //   L3 L2 L1 L0 l3 l2 l1 l0
   128         "movq %%mm2,%%mm4\n"           //   L3 L2 L1 L0 l3 l2 l1 l0
   129         "pand %12,%%mm2\n"             //   L3 0  L1  0 l3  0 l1  0
   130         "pand %13,%%mm4\n"             //   0  L2  0 L0  0 l2  0 l0
   131         "psrlw $8,%%mm2\n"             //   0  L3  0 L1  0 l3  0 l1
   132 
   133         // create R (result in mm6)
   134         "movq %%mm2,%%mm5\n"           //   0 L3  0 L1  0 l3  0 l1
   135         "movq %%mm4,%%mm6\n"           //   0 L2  0 L0  0 l2  0 l0
   136         "paddsw  %%mm1, %%mm5\n"       // lum1+red:x R3 x R1 x r3 x r1
   137         "paddsw  %%mm1, %%mm6\n"       // lum1+red:x R2 x R0 x r2 x r0
   138         "packuswb %%mm5,%%mm5\n"       //  R3 R1 r3 r1 R3 R1 r3 r1
   139         "packuswb %%mm6,%%mm6\n"       //  R2 R0 r2 r0 R2 R0 r2 r0
   140         "pxor %%mm7,%%mm7\n"      //         00 00 00 00 00 00 00 00
   141         "punpcklbw %%mm5,%%mm6\n"      //  R3 R2 R1 R0 r3 r2 r1 r0
   142 
   143         // create Cb (result in mm1)
   144         "movd (%1), %%mm1\n"      //         0  0  0  0  u3 u2 u1 u0
   145         "punpcklbw %%mm7,%%mm1\n" //         0  u3 0  u2 00 u1 00 u0
   146         "punpckldq %%mm1,%%mm1\n" //         00 u1 00 u0 00 u1 00 u0
   147         "psubw %9,%%mm1\n"        // mm1-128:u1 u1 u0 u0 u1 u1 u0 u0
   148 
   149         // create Cb_g (result in mm5)
   150         "movq %%mm1,%%mm5\n"            // u1 u1 u0 u0 u1 u1 u0 u0
   151         "pmullw %14,%%mm5\n"            // blue*-109dec=1.7129*64
   152         "pmullw %15,%%mm1\n"            // blue*114dec=1.78125*64
   153         "psraw  $6, %%mm5\n"            // blue=red/64
   154         "psraw  $6, %%mm1\n"            // blue=blue/64
   155 
   156         // create G (result in mm7)
   157         "movq %%mm2,%%mm3\n"      //   0  L3  0 L1  0 l3  0 l1
   158         "movq %%mm4,%%mm7\n"      //   0  L2  0 L0  0 l2  0 l1
   159         "paddsw  %%mm5, %%mm3\n"  // lum1+Cb_g:x G3t x G1t x g3t x g1t
   160         "paddsw  %%mm5, %%mm7\n"  // lum1+Cb_g:x G2t x G0t x g2t x g0t
   161         "paddsw  %%mm0, %%mm3\n"  // lum1+Cr_g:x G3  x G1  x g3  x g1
   162         "paddsw  %%mm0, %%mm7\n"  // lum1+blue:x G2  x G0  x g2  x g0
   163         "packuswb %%mm3,%%mm3\n"  // G3 G1 g3 g1 G3 G1 g3 g1
   164         "packuswb %%mm7,%%mm7\n"  // G2 G0 g2 g0 G2 G0 g2 g0
   165         "punpcklbw %%mm3,%%mm7\n" // G3 G2 G1 G0 g3 g2 g1 g0
   166 
   167         // create B (result in mm5)
   168         "movq %%mm2,%%mm3\n"         //   0  L3  0 L1  0 l3  0 l1
   169         "movq %%mm4,%%mm5\n"         //   0  L2  0 L0  0 l2  0 l1
   170         "paddsw  %%mm1, %%mm3\n"     // lum1+blue:x B3 x B1 x b3 x b1
   171         "paddsw  %%mm1, %%mm5\n"     // lum1+blue:x B2 x B0 x b2 x b0
   172         "packuswb %%mm3,%%mm3\n"     // B3 B1 b3 b1 B3 B1 b3 b1
   173         "packuswb %%mm5,%%mm5\n"     // B2 B0 b2 b0 B2 B0 b2 b0
   174         "punpcklbw %%mm3,%%mm5\n"    // B3 B2 B1 B0 b3 b2 b1 b0
   175 
   176         // fill destination row1 (needed are mm6=Rr,mm7=Gg,mm5=Bb)
   177 
   178         "pxor %%mm2,%%mm2\n"           //  0  0  0  0  0  0  0  0
   179         "pxor %%mm4,%%mm4\n"           //  0  0  0  0  0  0  0  0
   180         "movq %%mm6,%%mm1\n"           // R3 R2 R1 R0 r3 r2 r1 r0
   181         "movq %%mm5,%%mm3\n"           // B3 B2 B1 B0 b3 b2 b1 b0
   182 
   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,(%%esp)\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 
   232         "addl $4,%%esp\n"  // get rid of the stack slot we reserved.
   233         "emms\n"  // reset MMX registers.
   234         :
   235         : "m" (cr), "r"(cb),"r"(lum),
   236           "r"(row1),"r"(cols),"r"(row2),"m"(x),"m"(y),"m"(mod),
   237           "m"(MMX_0080w),"m"(MMX_VgrnRGB),"m"(MMX_VredRGB),
   238           "m"(MMX_FF00w),"m"(MMX_00FFw),"m"(MMX_UgrnRGB),
   239           "m"(MMX_UbluRGB)
   240     );
   241 }
   242 
   243 void Color565DitherYV12MMX1X( int *colortab, Uint32 *rgb_2_pix,
   244                              unsigned char *lum, unsigned char *cr,
   245                              unsigned char *cb, unsigned char *out,
   246                              int rows, int cols, int mod )
   247 {
   248     Uint16 *row1;
   249     Uint16 *row2;
   250 
   251     unsigned char* y = lum +cols*rows;    /* Pointer to the end */
   252     int x = 0;
   253     row1 = (Uint16 *)out;                 /* 16 bit target */
   254     row2 = (Uint16 *)out+cols+mod;        /* start of second row  */
   255     mod = (mod+cols+mod)*2;               /* increment for row1 in byte */
   256 
   257     __asm__ __volatile__(
   258         // tap dance to workaround the inability to use %%ebx at will...
   259         //  move one thing to the stack...
   260         "pushl $0\n"  // save a slot on the stack.
   261         "pushl %%ebx\n"  // save %%ebx.
   262         "movl %0, %%ebx\n"  // put the thing in ebx.
   263         "movl %%ebx, 4(%%esp)\n"  // put the thing in the stack slot.
   264         "popl %%ebx\n"  // get back %%ebx (the PIC register).
   265 
   266         ".align 8\n"
   267         "1:\n"
   268 
   269         "movd           (%1),                   %%mm0\n" // 4 Cb         0  0  0  0 u3 u2 u1 u0
   270         "pxor           %%mm7,                  %%mm7\n"
   271         "pushl %%ebx\n"
   272         "movl 4(%%esp), %%ebx\n"
   273         "movd (%%ebx), %%mm1\n"   // 4 Cr                0  0  0  0 v3 v2 v1 v0
   274         "popl %%ebx\n"
   275 
   276         "punpcklbw      %%mm7,                  %%mm0\n" // 4 W cb   0 u3  0 u2  0 u1  0 u0
   277         "punpcklbw      %%mm7,                  %%mm1\n" // 4 W cr   0 v3  0 v2  0 v1  0 v0
   278         "psubw          %9,                     %%mm0\n"
   279         "psubw          %9,                     %%mm1\n"
   280         "movq           %%mm0,                  %%mm2\n" // Cb                   0 u3  0 u2  0 u1  0 u0
   281         "movq           %%mm1,                  %%mm3\n" // Cr
   282         "pmullw         %10,                    %%mm2\n" // Cb2green 0 R3  0 R2  0 R1  0 R0
   283         "movq           (%2),                   %%mm6\n" // L1      l7 L6 L5 L4 L3 L2 L1 L0
   284         "pmullw         %11,                    %%mm0\n" // Cb2blue
   285         "pand           %12,                    %%mm6\n" // L1      00 L6 00 L4 00 L2 00 L0
   286         "pmullw         %13,                    %%mm3\n" // Cr2green
   287         "movq           (%2),                   %%mm7\n" // L2
   288         "pmullw         %14,                    %%mm1\n" // Cr2red
   289         "psrlw          $8,                     %%mm7\n"        // L2           00 L7 00 L5 00 L3 00 L1
   290         "pmullw         %15,                    %%mm6\n" // lum1
   291         "paddw          %%mm3,                  %%mm2\n" // Cb2green + Cr2green == green
   292         "pmullw         %15,                    %%mm7\n" // lum2
   293 
   294         "movq           %%mm6,                  %%mm4\n" // lum1
   295         "paddw          %%mm0,                  %%mm6\n" // lum1 +blue 00 B6 00 B4 00 B2 00 B0
   296         "movq           %%mm4,                  %%mm5\n" // lum1
   297         "paddw          %%mm1,                  %%mm4\n" // lum1 +red  00 R6 00 R4 00 R2 00 R0
   298         "paddw          %%mm2,                  %%mm5\n" // lum1 +green 00 G6 00 G4 00 G2 00 G0
   299         "psraw          $6,                     %%mm4\n" // R1 0 .. 64
   300         "movq           %%mm7,                  %%mm3\n" // lum2                       00 L7 00 L5 00 L3 00 L1
   301         "psraw          $6,                     %%mm5\n" // G1  - .. +
   302         "paddw          %%mm0,                  %%mm7\n" // Lum2 +blue 00 B7 00 B5 00 B3 00 B1
   303         "psraw          $6,                     %%mm6\n" // B1         0 .. 64
   304         "packuswb       %%mm4,                  %%mm4\n" // R1 R1
   305         "packuswb       %%mm5,                  %%mm5\n" // G1 G1
   306         "packuswb       %%mm6,                  %%mm6\n" // B1 B1
   307         "punpcklbw      %%mm4,                  %%mm4\n"
   308         "punpcklbw      %%mm5,                  %%mm5\n"
   309 
   310         "pand           %16,                    %%mm4\n"
   311         "psllw          $3,                     %%mm5\n" // GREEN       1
   312         "punpcklbw      %%mm6,                  %%mm6\n"
   313         "pand           %17,                    %%mm5\n"
   314         "pand           %16,                    %%mm6\n"
   315         "por            %%mm5,                  %%mm4\n" //
   316         "psrlw          $11,                    %%mm6\n" // BLUE        1
   317         "movq           %%mm3,                  %%mm5\n" // lum2
   318         "paddw          %%mm1,                  %%mm3\n" // lum2 +red      00 R7 00 R5 00 R3 00 R1
   319         "paddw          %%mm2,                  %%mm5\n" // lum2 +green 00 G7 00 G5 00 G3 00 G1
   320         "psraw          $6,                     %%mm3\n" // R2
   321         "por            %%mm6,                  %%mm4\n" // MM4
   322         "psraw          $6,                     %%mm5\n" // G2
   323         "movq           (%2, %4),               %%mm6\n" // L3 load lum2
   324         "psraw          $6,                     %%mm7\n"
   325         "packuswb       %%mm3,                  %%mm3\n"
   326         "packuswb       %%mm5,                  %%mm5\n"
   327         "packuswb       %%mm7,                  %%mm7\n"
   328         "pand           %12,                    %%mm6\n" // L3
   329         "punpcklbw      %%mm3,                  %%mm3\n"
   330         "punpcklbw      %%mm5,                  %%mm5\n"
   331         "pmullw         %15,                    %%mm6\n" // lum3
   332         "punpcklbw      %%mm7,                  %%mm7\n"
   333         "psllw          $3,                     %%mm5\n" // GREEN 2
   334         "pand           %16,                    %%mm7\n"
   335         "pand           %16,                    %%mm3\n"
   336         "psrlw          $11,                    %%mm7\n" // BLUE  2
   337         "pand           %17,                    %%mm5\n"
   338         "por            %%mm7,                  %%mm3\n"
   339         "movq           (%2,%4),                %%mm7\n" // L4 load lum2
   340         "por            %%mm5,                  %%mm3\n" //
   341         "psrlw          $8,                     %%mm7\n" // L4
   342         "movq           %%mm4,                  %%mm5\n"
   343         "punpcklwd      %%mm3,                  %%mm4\n"
   344         "pmullw         %15,                    %%mm7\n" // lum4
   345         "punpckhwd      %%mm3,                  %%mm5\n"
   346 
   347         "movq           %%mm4,                  (%3)\n"  // write row1
   348         "movq           %%mm5,                  8(%3)\n" // write row1
   349 
   350         "movq           %%mm6,                  %%mm4\n" // Lum3
   351         "paddw          %%mm0,                  %%mm6\n" // Lum3 +blue
   352 
   353         "movq           %%mm4,                  %%mm5\n" // Lum3
   354         "paddw          %%mm1,                  %%mm4\n" // Lum3 +red
   355         "paddw          %%mm2,                  %%mm5\n" // Lum3 +green
   356         "psraw          $6,                     %%mm4\n"
   357         "movq           %%mm7,                  %%mm3\n" // Lum4
   358         "psraw          $6,                     %%mm5\n"
   359         "paddw          %%mm0,                  %%mm7\n" // Lum4 +blue
   360         "psraw          $6,                     %%mm6\n" // Lum3 +blue
   361         "movq           %%mm3,                  %%mm0\n" // Lum4
   362         "packuswb       %%mm4,                  %%mm4\n"
   363         "paddw          %%mm1,                  %%mm3\n" // Lum4 +red
   364         "packuswb       %%mm5,                  %%mm5\n"
   365         "paddw          %%mm2,                  %%mm0\n" // Lum4 +green
   366         "packuswb       %%mm6,                  %%mm6\n"
   367         "punpcklbw      %%mm4,                  %%mm4\n"
   368         "punpcklbw      %%mm5,                  %%mm5\n"
   369         "punpcklbw      %%mm6,                  %%mm6\n"
   370         "psllw          $3,                     %%mm5\n" // GREEN 3
   371         "pand           %16,                    %%mm4\n"
   372         "psraw          $6,                     %%mm3\n" // psr 6
   373         "psraw          $6,                     %%mm0\n"
   374         "pand           %16,                    %%mm6\n" // BLUE
   375         "pand           %17,                    %%mm5\n"
   376         "psrlw          $11,                    %%mm6\n" // BLUE  3
   377         "por            %%mm5,                  %%mm4\n"
   378         "psraw          $6,                     %%mm7\n"
   379         "por            %%mm6,                  %%mm4\n"
   380         "packuswb       %%mm3,                  %%mm3\n"
   381         "packuswb       %%mm0,                  %%mm0\n"
   382         "packuswb       %%mm7,                  %%mm7\n"
   383         "punpcklbw      %%mm3,                  %%mm3\n"
   384         "punpcklbw      %%mm0,                  %%mm0\n"
   385         "punpcklbw      %%mm7,                  %%mm7\n"
   386         "pand           %16,                    %%mm3\n"
   387         "pand           %16,                    %%mm7\n" // BLUE
   388         "psllw          $3,                     %%mm0\n" // GREEN 4
   389         "psrlw          $11,                    %%mm7\n"
   390         "pand           %17,                    %%mm0\n"
   391         "por            %%mm7,                  %%mm3\n"
   392         "por            %%mm0,                  %%mm3\n"
   393 
   394         "movq           %%mm4,                  %%mm5\n"
   395 
   396         "punpcklwd      %%mm3,                  %%mm4\n"
   397         "punpckhwd      %%mm3,                  %%mm5\n"
   398 
   399         "movq           %%mm4,                  (%5)\n"
   400         "movq           %%mm5,                  8(%5)\n"
   401 
   402         "addl           $8,                     %6\n"
   403         "addl           $8,                     %2\n"
   404         "addl           $4,                     (%%esp)\n"
   405         "addl           $4,                     %1\n"
   406         "cmpl           %4,                     %6\n"
   407         "leal           16(%3),                 %3\n"
   408         "leal           16(%5),%5\n" // row2+16
   409 
   410         "jl             1b\n"
   411         "addl           %4,     %2\n" // lum += cols
   412         "addl           %8,     %3\n" // row1+= mod
   413         "addl           %8,     %5\n" // row2+= mod
   414         "movl           $0,     %6\n" // x=0
   415         "cmpl           %7,     %2\n"
   416         "jl             1b\n"
   417         "addl $4, %%esp\n"  // get rid of the stack slot we reserved.
   418         "emms\n"
   419         :
   420         : "m" (cr), "r"(cb),"r"(lum),
   421           "r"(row1),"r"(cols),"r"(row2),"m"(x),"m"(y),"m"(mod),
   422           "m"(MMX_0080w),"m"(MMX_Ugrn565),"m"(MMX_Ublu5x5),
   423           "m"(MMX_00FFw),"m"(MMX_Vgrn565),"m"(MMX_Vred5x5),
   424           "m"(MMX_Ycoeff),"m"(MMX_red565),"m"(MMX_grn565)
   425     );
   426 }
   427 
   428 /* *INDENT-ON* */
   429 
   430 #endif /* GCC3 i386 inline assembly */
   431 
   432 /* vi: set ts=4 sw=4 expandtab: */