src/video/SDL_yuv_mmx.c
author Ryan C. Gordon <icculus@icculus.org>
Sat, 14 Jul 2007 07:26:34 +0000
branchSDL-1.2
changeset 4064 940fddb81bea
parent 4062 ee5a38285651
child 4159 a1b03ba2fcd0
permissions -rw-r--r--
Mac OS X/x86 won't build the MMX/YUV inline assembly without optimizations
enabled (not enough registers), so for now, we only build it if we see
the __OPTIMIZE__ #define, which GCC provides when you build at -O1 or higher.
slouken@0
     1
/*
slouken@0
     2
    SDL - Simple DirectMedia Layer
slouken@1312
     3
    Copyright (C) 1997-2006 Sam Lantinga
slouken@0
     4
slouken@0
     5
    This library is free software; you can redistribute it and/or
slouken@1312
     6
    modify it under the terms of the GNU Lesser General Public
slouken@0
     7
    License as published by the Free Software Foundation; either
slouken@1312
     8
    version 2.1 of the License, or (at your option) any later version.
slouken@0
     9
slouken@0
    10
    This library is distributed in the hope that it will be useful,
slouken@0
    11
    but WITHOUT ANY WARRANTY; without even the implied warranty of
slouken@0
    12
    MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
slouken@1312
    13
    Lesser General Public License for more details.
slouken@0
    14
slouken@1312
    15
    You should have received a copy of the GNU Lesser General Public
slouken@1312
    16
    License along with this library; if not, write to the Free Software
slouken@1312
    17
    Foundation, Inc., 51 Franklin St, Fifth Floor, Boston, MA  02110-1301  USA
slouken@0
    18
slouken@0
    19
    Sam Lantinga
slouken@252
    20
    slouken@libsdl.org
slouken@0
    21
*/
slouken@1402
    22
#include "SDL_config.h"
slouken@0
    23
icculus@4064
    24
#if (__GNUC__ > 2) && defined(__i386__) && __OPTIMIZE__ && SDL_ASSEMBLY_ROUTINES
slouken@0
    25
slouken@1407
    26
#include "SDL_stdinc.h"
slouken@1407
    27
icculus@4045
    28
#include "mmx.h"
slouken@0
    29
icculus@4047
    30
/* *INDENT-OFF* */
icculus@4047
    31
icculus@4045
    32
static mmx_t MMX_0080w    = { .ud = {0x00800080, 0x00800080} };
icculus@4045
    33
static mmx_t MMX_00FFw    = { .ud = {0x00ff00ff, 0x00ff00ff} };
icculus@4045
    34
static mmx_t MMX_FF00w    = { .ud = {0xff00ff00, 0xff00ff00} };
slouken@0
    35
icculus@4045
    36
static mmx_t MMX_Ycoeff   = { .uw = {0x004a, 0x004a, 0x004a, 0x004a} };
slouken@0
    37
icculus@4045
    38
static mmx_t MMX_UbluRGB  = { .uw = {0x0072, 0x0072, 0x0072, 0x0072} };
icculus@4045
    39
static mmx_t MMX_VredRGB  = { .uw = {0x0059, 0x0059, 0x0059, 0x0059} };
icculus@4045
    40
static mmx_t MMX_UgrnRGB  = { .uw = {0xffea, 0xffea, 0xffea, 0xffea} };
icculus@4045
    41
static mmx_t MMX_VgrnRGB  = { .uw = {0xffd2, 0xffd2, 0xffd2, 0xffd2} };
slouken@0
    42
icculus@4045
    43
static mmx_t MMX_Ublu5x5  = { .uw = {0x0081, 0x0081, 0x0081, 0x0081} };
icculus@4045
    44
static mmx_t MMX_Vred5x5  = { .uw = {0x0066, 0x0066, 0x0066, 0x0066} };
icculus@4045
    45
static mmx_t MMX_Ugrn565  = { .uw = {0xffe8, 0xffe8, 0xffe8, 0xffe8} };
icculus@4045
    46
static mmx_t MMX_Vgrn565  = { .uw = {0xffcd, 0xffcd, 0xffcd, 0xffcd} };
icculus@4045
    47
icculus@4045
    48
static mmx_t MMX_red565   = { .uw = {0xf800, 0xf800, 0xf800, 0xf800} };
icculus@4045
    49
static mmx_t MMX_grn565   = { .uw = {0x07e0, 0x07e0, 0x07e0, 0x07e0} };
slouken@0
    50
slouken@0
    51
/**
slouken@0
    52
   This MMX assembler is my first assembler/MMX program ever.
slouken@0
    53
   Thus it maybe buggy.
slouken@0
    54
   Send patches to:
slouken@0
    55
   mvogt@rhrk.uni-kl.de
slouken@0
    56
slouken@0
    57
   After it worked fine I have "obfuscated" the code a bit to have
slouken@0
    58
   more parallism in the MMX units. This means I moved
slouken@0
    59
   initilisation around and delayed other instruction.
slouken@0
    60
   Performance measurement did not show that this brought any advantage
slouken@0
    61
   but in theory it _should_ be faster this way.
slouken@0
    62
slouken@0
    63
   The overall performanve gain to the C based dither was 30%-40%.
slouken@0
    64
   The MMX routine calculates 256bit=8RGB values in each cycle
slouken@0
    65
   (4 for row1 & 4 for row2)
slouken@0
    66
slouken@0
    67
   The red/green/blue.. coefficents are taken from the mpeg_play 
slouken@0
    68
   player. They look nice, but I dont know if you can have
slouken@0
    69
   better values, to avoid integer rounding errors.
slouken@0
    70
   
slouken@0
    71
slouken@0
    72
   IMPORTANT:
slouken@0
    73
   ==========
slouken@0
    74
slouken@0
    75
   It is a requirement that the cr/cb/lum are 8 byte aligned and
slouken@0
    76
   the out are 16byte aligned or you will/may get segfaults
slouken@0
    77
slouken@0
    78
*/
slouken@0
    79
slouken@0
    80
void ColorRGBDitherYV12MMX1X( int *colortab, Uint32 *rgb_2_pix,
slouken@0
    81
                              unsigned char *lum, unsigned char *cr,
slouken@0
    82
                              unsigned char *cb, unsigned char *out,
slouken@0
    83
                              int rows, int cols, int mod )
slouken@0
    84
{
icculus@4045
    85
	Uint32 *row1;
icculus@4045
    86
	Uint32 *row2;
slouken@0
    87
icculus@4045
    88
	unsigned char* y = lum +cols*rows;    // Pointer to the end
icculus@4045
    89
	int x = 0;
icculus@4045
    90
	row1 = (Uint32 *)out;                 // 32 bit target
icculus@4045
    91
	row2 = (Uint32 *)out+cols+mod;        // start of second row
icculus@4045
    92
	mod = (mod+cols+mod)*4;               // increment for row1 in byte
slouken@0
    93
icculus@4045
    94
	__asm__ __volatile__ (
icculus@4046
    95
		// tap dance to workaround the inability to use %%ebx at will...
icculus@4046
    96
		//  move one thing to the stack...
icculus@4046
    97
		"pushl $0\n"  // save a slot on the stack.
icculus@4046
    98
		"pushl %%ebx\n"  // save %%ebx.
icculus@4046
    99
		"movl %0, %%ebx\n"  // put the thing in ebx.
icculus@4046
   100
		"movl %%ebx,4(%%esp)\n"  // put the thing in the stack slot.
icculus@4046
   101
		"popl %%ebx\n"  // get back %%ebx (the PIC register).
slouken@0
   102
icculus@4046
   103
		".align 8\n"
icculus@4046
   104
		"1:\n"
icculus@4046
   105
icculus@4046
   106
		// create Cr (result in mm1)
icculus@4046
   107
		"pushl %%ebx\n"
icculus@4046
   108
		"movl 4(%%esp),%%ebx\n"
icculus@4046
   109
		"movd (%%ebx),%%mm1\n"   //         0  0  0  0  v3 v2 v1 v0
icculus@4046
   110
		"popl %%ebx\n"
icculus@4046
   111
		"pxor %%mm7,%%mm7\n"      //         00 00 00 00 00 00 00 00
icculus@4046
   112
		"movd (%2), %%mm2\n"           //    0  0  0  0 l3 l2 l1 l0
icculus@4046
   113
		"punpcklbw %%mm7,%%mm1\n" //         0  v3 0  v2 00 v1 00 v0
icculus@4046
   114
		"punpckldq %%mm1,%%mm1\n" //         00 v1 00 v0 00 v1 00 v0
icculus@4046
   115
		"psubw %9,%%mm1\n"        // mm1-128:r1 r1 r0 r0 r1 r1 r0 r0
slouken@0
   116
slouken@0
   117
		 // create Cr_g (result in mm0)
icculus@4046
   118
		"movq %%mm1,%%mm0\n"           // r1 r1 r0 r0 r1 r1 r0 r0
icculus@4046
   119
		"pmullw %10,%%mm0\n"           // red*-46dec=0.7136*64
icculus@4046
   120
		"pmullw %11,%%mm1\n"           // red*89dec=1.4013*64
icculus@4046
   121
		"psraw  $6, %%mm0\n"           // red=red/64
icculus@4046
   122
		"psraw  $6, %%mm1\n"           // red=red/64
icculus@4046
   123
icculus@4046
   124
		// create L1 L2 (result in mm2,mm4)
icculus@4046
   125
		// L2=lum+cols
icculus@4046
   126
		"movq (%2,%4),%%mm3\n"         //    0  0  0  0 L3 L2 L1 L0
icculus@4046
   127
		"punpckldq %%mm3,%%mm2\n"      //   L3 L2 L1 L0 l3 l2 l1 l0
icculus@4046
   128
		"movq %%mm2,%%mm4\n"           //   L3 L2 L1 L0 l3 l2 l1 l0
icculus@4046
   129
		"pand %12,%%mm2\n"             //   L3 0  L1  0 l3  0 l1  0
icculus@4046
   130
		"pand %13,%%mm4\n"             //   0  L2  0 L0  0 l2  0 l0
icculus@4046
   131
		"psrlw $8,%%mm2\n"             //   0  L3  0 L1  0 l3  0 l1
icculus@4046
   132
icculus@4046
   133
		// create R (result in mm6)
icculus@4046
   134
		"movq %%mm2,%%mm5\n"           //   0 L3  0 L1  0 l3  0 l1
icculus@4046
   135
		"movq %%mm4,%%mm6\n"           //   0 L2  0 L0  0 l2  0 l0
icculus@4046
   136
		"paddsw  %%mm1, %%mm5\n"       // lum1+red:x R3 x R1 x r3 x r1
icculus@4046
   137
		"paddsw  %%mm1, %%mm6\n"       // lum1+red:x R2 x R0 x r2 x r0
icculus@4046
   138
		"packuswb %%mm5,%%mm5\n"       //  R3 R1 r3 r1 R3 R1 r3 r1
icculus@4046
   139
		"packuswb %%mm6,%%mm6\n"       //  R2 R0 r2 r0 R2 R0 r2 r0
icculus@4046
   140
		"pxor %%mm7,%%mm7\n"      //         00 00 00 00 00 00 00 00
icculus@4046
   141
		"punpcklbw %%mm5,%%mm6\n"      //  R3 R2 R1 R0 r3 r2 r1 r0
icculus@4046
   142
icculus@4046
   143
		// create Cb (result in mm1)
icculus@4046
   144
		"movd (%1), %%mm1\n"      //         0  0  0  0  u3 u2 u1 u0
icculus@4046
   145
		"punpcklbw %%mm7,%%mm1\n" //         0  u3 0  u2 00 u1 00 u0
icculus@4046
   146
		"punpckldq %%mm1,%%mm1\n" //         00 u1 00 u0 00 u1 00 u0
icculus@4046
   147
		"psubw %9,%%mm1\n"        // mm1-128:u1 u1 u0 u0 u1 u1 u0 u0
icculus@4046
   148
		// create Cb_g (result in mm5)
icculus@4046
   149
		"movq %%mm1,%%mm5\n"            // u1 u1 u0 u0 u1 u1 u0 u0
icculus@4046
   150
		"pmullw %14,%%mm5\n"            // blue*-109dec=1.7129*64
icculus@4046
   151
		"pmullw %15,%%mm1\n"            // blue*114dec=1.78125*64
icculus@4046
   152
		"psraw  $6, %%mm5\n"            // blue=red/64
icculus@4046
   153
		"psraw  $6, %%mm1\n"            // blue=blue/64
icculus@4046
   154
icculus@4046
   155
		// create G (result in mm7)
icculus@4046
   156
		"movq %%mm2,%%mm3\n"      //   0  L3  0 L1  0 l3  0 l1
icculus@4046
   157
		"movq %%mm4,%%mm7\n"      //   0  L2  0 L0  0 l2  0 l1
icculus@4046
   158
		"paddsw  %%mm5, %%mm3\n"  // lum1+Cb_g:x G3t x G1t x g3t x g1t
icculus@4046
   159
		"paddsw  %%mm5, %%mm7\n"  // lum1+Cb_g:x G2t x G0t x g2t x g0t
icculus@4046
   160
		"paddsw  %%mm0, %%mm3\n"  // lum1+Cr_g:x G3  x G1  x g3  x g1
icculus@4046
   161
		"paddsw  %%mm0, %%mm7\n"  // lum1+blue:x G2  x G0  x g2  x g0
icculus@4046
   162
		"packuswb %%mm3,%%mm3\n"  // G3 G1 g3 g1 G3 G1 g3 g1
icculus@4046
   163
		"packuswb %%mm7,%%mm7\n"  // G2 G0 g2 g0 G2 G0 g2 g0
icculus@4046
   164
		"punpcklbw %%mm3,%%mm7\n" // G3 G2 G1 G0 g3 g2 g1 g0
icculus@4046
   165
icculus@4046
   166
		// create B (result in mm5)
icculus@4046
   167
		"movq %%mm2,%%mm3\n"         //   0  L3  0 L1  0 l3  0 l1
icculus@4046
   168
		"movq %%mm4,%%mm5\n"         //   0  L2  0 L0  0 l2  0 l1
icculus@4046
   169
		"paddsw  %%mm1, %%mm3\n"     // lum1+blue:x B3 x B1 x b3 x b1
icculus@4046
   170
		"paddsw  %%mm1, %%mm5\n"     // lum1+blue:x B2 x B0 x b2 x b0
icculus@4046
   171
		"packuswb %%mm3,%%mm3\n"     // B3 B1 b3 b1 B3 B1 b3 b1
icculus@4046
   172
		"packuswb %%mm5,%%mm5\n"     // B2 B0 b2 b0 B2 B0 b2 b0
icculus@4046
   173
		"punpcklbw %%mm3,%%mm5\n"    // B3 B2 B1 B0 b3 b2 b1 b0
icculus@4046
   174
icculus@4046
   175
		// fill destination row1 (needed are mm6=Rr,mm7=Gg,mm5=Bb)
icculus@4046
   176
icculus@4046
   177
		"pxor %%mm2,%%mm2\n"           //  0  0  0  0  0  0  0  0
icculus@4046
   178
		"pxor %%mm4,%%mm4\n"           //  0  0  0  0  0  0  0  0
icculus@4046
   179
		"movq %%mm6,%%mm1\n"           // R3 R2 R1 R0 r3 r2 r1 r0
icculus@4046
   180
		"movq %%mm5,%%mm3\n"           // B3 B2 B1 B0 b3 b2 b1 b0
icculus@4046
   181
		// process lower lum
icculus@4046
   182
		"punpcklbw %%mm4,%%mm1\n"      //  0 r3  0 r2  0 r1  0 r0
icculus@4046
   183
		"punpcklbw %%mm4,%%mm3\n"      //  0 b3  0 b2  0 b1  0 b0
icculus@4046
   184
		"movq %%mm1,%%mm2\n"           //  0 r3  0 r2  0 r1  0 r0
icculus@4046
   185
		"movq %%mm3,%%mm0\n"           //  0 b3  0 b2  0 b1  0 b0
icculus@4046
   186
		"punpcklwd %%mm1,%%mm3\n"      //  0 r1  0 b1  0 r0  0 b0
icculus@4046
   187
		"punpckhwd %%mm2,%%mm0\n"      //  0 r3  0 b3  0 r2  0 b2
icculus@4046
   188
icculus@4046
   189
		"pxor %%mm2,%%mm2\n"           //  0  0  0  0  0  0  0  0
icculus@4046
   190
		"movq %%mm7,%%mm1\n"           // G3 G2 G1 G0 g3 g2 g1 g0
icculus@4046
   191
		"punpcklbw %%mm1,%%mm2\n"      // g3  0 g2  0 g1  0 g0  0
icculus@4046
   192
		"punpcklwd %%mm4,%%mm2\n"      //  0  0 g1  0  0  0 g0  0
icculus@4046
   193
		"por %%mm3, %%mm2\n"          //  0 r1 g1 b1  0 r0 g0 b0
icculus@4046
   194
		"movq %%mm2,(%3)\n"          // wrote out ! row1
icculus@4046
   195
icculus@4046
   196
		"pxor %%mm2,%%mm2\n"           //  0  0  0  0  0  0  0  0
icculus@4046
   197
		"punpcklbw %%mm1,%%mm4\n"      // g3  0 g2  0 g1  0 g0  0
icculus@4046
   198
		"punpckhwd %%mm2,%%mm4\n"      //  0  0 g3  0  0  0 g2  0
icculus@4046
   199
		"por %%mm0, %%mm4\n"          //  0 r3 g3 b3  0 r2 g2 b2
icculus@4046
   200
		"movq %%mm4,8(%3)\n"         // wrote out ! row1
slouken@0
   201
		 
icculus@4046
   202
		// fill destination row2 (needed are mm6=Rr,mm7=Gg,mm5=Bb)
icculus@4046
   203
		// this can be done "destructive"
icculus@4046
   204
		"pxor %%mm2,%%mm2\n"           //  0  0  0  0  0  0  0  0
icculus@4046
   205
		"punpckhbw %%mm2,%%mm6\n"      //  0 R3  0 R2  0 R1  0 R0
icculus@4046
   206
		"punpckhbw %%mm1,%%mm5\n"      // G3 B3 G2 B2 G1 B1 G0 B0
icculus@4046
   207
		"movq %%mm5,%%mm1\n"           // G3 B3 G2 B2 G1 B1 G0 B0
icculus@4046
   208
		"punpcklwd %%mm6,%%mm1\n"      //  0 R1 G1 B1  0 R0 G0 B0
icculus@4046
   209
		"movq %%mm1,(%5)\n"          // wrote out ! row2
icculus@4046
   210
		"punpckhwd %%mm6,%%mm5\n"      //  0 R3 G3 B3  0 R2 G2 B2
icculus@4046
   211
		"movq %%mm5,8(%5)\n"         // wrote out ! row2
icculus@4046
   212
		 
icculus@4046
   213
		"addl $4,%2\n"            // lum+4
icculus@4046
   214
		"leal 16(%3),%3\n"        // row1+16
icculus@4046
   215
		"leal 16(%5),%5\n"        // row2+16
icculus@4046
   216
		"addl $2,(%%esp)\n"        // cr+2
icculus@4046
   217
		"addl $2,%1\n"           // cb+2
slouken@0
   218
icculus@4046
   219
		"addl $4,%6\n"            // x+4
icculus@4046
   220
		"cmpl %4,%6\n"
slouken@0
   221
icculus@4046
   222
		"jl 1b\n"
icculus@4046
   223
		"addl %4,%2\n" // lum += cols
icculus@4046
   224
		"addl %8,%3\n" // row1+= mod
icculus@4046
   225
		"addl %8,%5\n" // row2+= mod
icculus@4046
   226
		"movl $0,%6\n" // x=0
icculus@4046
   227
		"cmpl %7,%2\n"
icculus@4046
   228
		"jl 1b\n"
slouken@0
   229
icculus@4046
   230
		"addl $4,%%esp\n"  // get rid of the stack slot we reserved.
icculus@4046
   231
		"emms\n"  // reset MMX registers.
icculus@4046
   232
		:
icculus@4046
   233
		: "m" (cr), "r"(cb),"r"(lum),
icculus@4046
   234
		  "r"(row1),"r"(cols),"r"(row2),"m"(x),"m"(y),"m"(mod),
icculus@4046
   235
		  "m"(MMX_0080w),"m"(MMX_VgrnRGB),"m"(MMX_VredRGB),
icculus@4046
   236
		  "m"(MMX_FF00w),"m"(MMX_00FFw),"m"(MMX_UgrnRGB),
icculus@4046
   237
		  "m"(MMX_UbluRGB)
icculus@4045
   238
	);
slouken@0
   239
}
slouken@0
   240
slouken@0
   241
void Color565DitherYV12MMX1X( int *colortab, Uint32 *rgb_2_pix,
slouken@0
   242
                             unsigned char *lum, unsigned char *cr,
slouken@0
   243
                             unsigned char *cb, unsigned char *out,
slouken@0
   244
                             int rows, int cols, int mod )
slouken@0
   245
{
icculus@4046
   246
	Uint16 *row1;
icculus@4046
   247
	Uint16 *row2;
slouken@0
   248
icculus@4046
   249
	unsigned char* y = lum +cols*rows;    /* Pointer to the end */
icculus@4046
   250
	int x = 0;
icculus@4046
   251
	row1 = (Uint16 *)out;                 /* 16 bit target */
icculus@4046
   252
	row2 = (Uint16 *)out+cols+mod;        /* start of second row  */
icculus@4046
   253
	mod = (mod+cols+mod)*2;               /* increment for row1 in byte */
slouken@0
   254
icculus@4046
   255
	__asm__ __volatile__(
icculus@4046
   256
		// tap dance to workaround the inability to use %%ebx at will...
icculus@4046
   257
		//  move one thing to the stack...
icculus@4046
   258
		"pushl $0\n"  // save a slot on the stack.
icculus@4046
   259
		"pushl %%ebx\n"  // save %%ebx.
icculus@4046
   260
		"movl %0, %%ebx\n"  // put the thing in ebx.
icculus@4046
   261
		"movl %%ebx, 4(%%esp)\n"  // put the thing in the stack slot.
icculus@4046
   262
		"popl %%ebx\n"  // get back %%ebx (the PIC register).
slouken@0
   263
icculus@4046
   264
		".align 8\n"
icculus@4046
   265
		"1:\n"
icculus@4046
   266
		"movd           (%1),                   %%mm0\n" // 4 Cb         0  0  0  0 u3 u2 u1 u0
icculus@4046
   267
		"pxor           %%mm7,                  %%mm7\n"
icculus@4046
   268
		"pushl %%ebx\n"
icculus@4046
   269
		"movl 4(%%esp), %%ebx\n"
icculus@4046
   270
		"movd (%%ebx), %%mm1\n"   // 4 Cr                0  0  0  0 v3 v2 v1 v0
icculus@4046
   271
		"popl %%ebx\n"
slouken@0
   272
icculus@4046
   273
		"punpcklbw      %%mm7,                  %%mm0\n" // 4 W cb   0 u3  0 u2  0 u1  0 u0
icculus@4046
   274
		"punpcklbw      %%mm7,                  %%mm1\n" // 4 W cr   0 v3  0 v2  0 v1  0 v0
icculus@4046
   275
		"psubw          %9,                     %%mm0\n"
icculus@4046
   276
		"psubw          %9,                     %%mm1\n"
icculus@4046
   277
		"movq           %%mm0,                  %%mm2\n" // Cb                   0 u3  0 u2  0 u1  0 u0
icculus@4046
   278
		"movq           %%mm1,                  %%mm3\n" // Cr
icculus@4046
   279
		"pmullw         %10,                    %%mm2\n" // Cb2green 0 R3  0 R2  0 R1  0 R0
icculus@4046
   280
		"movq           (%2),                   %%mm6\n" // L1      l7 L6 L5 L4 L3 L2 L1 L0
icculus@4046
   281
		"pmullw         %11,                    %%mm0\n" // Cb2blue
icculus@4046
   282
		"pand           %12,                    %%mm6\n" // L1      00 L6 00 L4 00 L2 00 L0
icculus@4046
   283
		"pmullw         %13,                    %%mm3\n" // Cr2green
icculus@4046
   284
		"movq           (%2),                   %%mm7\n" // L2
icculus@4046
   285
		"pmullw         %14,                    %%mm1\n" // Cr2red
icculus@4046
   286
		"psrlw          $8,                     %%mm7\n"        // L2           00 L7 00 L5 00 L3 00 L1
icculus@4046
   287
		"pmullw         %15,                    %%mm6\n" // lum1
icculus@4046
   288
		"paddw          %%mm3,                  %%mm2\n" // Cb2green + Cr2green == green
icculus@4046
   289
		"pmullw         %15,                    %%mm7\n" // lum2
icculus@4045
   290
icculus@4046
   291
		"movq           %%mm6,                  %%mm4\n" // lum1
icculus@4046
   292
		"paddw          %%mm0,                  %%mm6\n" // lum1 +blue 00 B6 00 B4 00 B2 00 B0
icculus@4046
   293
		"movq           %%mm4,                  %%mm5\n" // lum1
icculus@4046
   294
		"paddw          %%mm1,                  %%mm4\n" // lum1 +red  00 R6 00 R4 00 R2 00 R0
icculus@4046
   295
		"paddw          %%mm2,                  %%mm5\n" // lum1 +green 00 G6 00 G4 00 G2 00 G0
icculus@4046
   296
		"psraw          $6,                     %%mm4\n" // R1 0 .. 64
icculus@4046
   297
		"movq           %%mm7,                  %%mm3\n" // lum2                       00 L7 00 L5 00 L3 00 L1
icculus@4046
   298
		"psraw          $6,                     %%mm5\n" // G1  - .. +
icculus@4046
   299
		"paddw          %%mm0,                  %%mm7\n" // Lum2 +blue 00 B7 00 B5 00 B3 00 B1
icculus@4046
   300
		"psraw          $6,                     %%mm6\n" // B1         0 .. 64
icculus@4046
   301
		"packuswb       %%mm4,                  %%mm4\n" // R1 R1
icculus@4046
   302
		"packuswb       %%mm5,                  %%mm5\n" // G1 G1
icculus@4046
   303
		"packuswb       %%mm6,                  %%mm6\n" // B1 B1
icculus@4046
   304
		"punpcklbw      %%mm4,                  %%mm4\n"
icculus@4046
   305
		"punpcklbw      %%mm5,                  %%mm5\n"
slouken@0
   306
icculus@4046
   307
		"pand           %16,                    %%mm4\n"
icculus@4046
   308
		"psllw          $3,                     %%mm5\n" // GREEN       1
icculus@4046
   309
		"punpcklbw      %%mm6,                  %%mm6\n"
icculus@4046
   310
		"pand           %17,                    %%mm5\n"
icculus@4046
   311
		"pand           %16,                    %%mm6\n"
icculus@4046
   312
		"por            %%mm5,                  %%mm4\n" //
icculus@4046
   313
		"psrlw          $11,                    %%mm6\n" // BLUE        1
icculus@4046
   314
		"movq           %%mm3,                  %%mm5\n" // lum2
icculus@4046
   315
		"paddw          %%mm1,                  %%mm3\n" // lum2 +red      00 R7 00 R5 00 R3 00 R1
icculus@4046
   316
		"paddw          %%mm2,                  %%mm5\n" // lum2 +green 00 G7 00 G5 00 G3 00 G1
icculus@4046
   317
		"psraw          $6,                     %%mm3\n" // R2
icculus@4046
   318
		"por            %%mm6,                  %%mm4\n" // MM4
icculus@4046
   319
		"psraw          $6,                     %%mm5\n" // G2
icculus@4046
   320
		"movq           (%2, %4),               %%mm6\n" // L3 load lum2
icculus@4046
   321
		"psraw          $6,                     %%mm7\n"
icculus@4046
   322
		"packuswb       %%mm3,                  %%mm3\n"
icculus@4046
   323
		"packuswb       %%mm5,                  %%mm5\n"
icculus@4046
   324
		"packuswb       %%mm7,                  %%mm7\n"
icculus@4046
   325
		"pand           %12,                    %%mm6\n" // L3
icculus@4046
   326
		"punpcklbw      %%mm3,                  %%mm3\n"
icculus@4046
   327
		"punpcklbw      %%mm5,                  %%mm5\n"
icculus@4046
   328
		"pmullw         %15,                    %%mm6\n" // lum3
icculus@4046
   329
		"punpcklbw      %%mm7,                  %%mm7\n"
icculus@4046
   330
		"psllw          $3,                     %%mm5\n" // GREEN 2
icculus@4046
   331
		"pand           %16,                    %%mm7\n"
icculus@4046
   332
		"pand           %16,                    %%mm3\n"
icculus@4046
   333
		"psrlw          $11,                    %%mm7\n" // BLUE  2
icculus@4046
   334
		"pand           %17,                    %%mm5\n"
icculus@4046
   335
		"por            %%mm7,                  %%mm3\n"
icculus@4046
   336
		"movq           (%2,%4),                %%mm7\n" // L4 load lum2
icculus@4046
   337
		"por            %%mm5,                  %%mm3\n" //
icculus@4046
   338
		"psrlw          $8,                     %%mm7\n" // L4
icculus@4046
   339
		"movq           %%mm4,                  %%mm5\n"
icculus@4046
   340
		"punpcklwd      %%mm3,                  %%mm4\n"
icculus@4046
   341
		"pmullw         %15,                    %%mm7\n" // lum4
icculus@4046
   342
		"punpckhwd      %%mm3,                  %%mm5\n"
slouken@0
   343
icculus@4046
   344
		"movq           %%mm4,                  (%3)\n"  // write row1
icculus@4046
   345
		"movq           %%mm5,                  8(%3)\n" // write row1
slouken@0
   346
icculus@4046
   347
		"movq           %%mm6,                  %%mm4\n" // Lum3
icculus@4046
   348
		"paddw          %%mm0,                  %%mm6\n" // Lum3 +blue
slouken@0
   349
icculus@4046
   350
		"movq           %%mm4,                  %%mm5\n" // Lum3
icculus@4046
   351
		"paddw          %%mm1,                  %%mm4\n" // Lum3 +red
icculus@4046
   352
		"paddw          %%mm2,                  %%mm5\n" // Lum3 +green
icculus@4046
   353
		"psraw          $6,                     %%mm4\n"
icculus@4046
   354
		"movq           %%mm7,                  %%mm3\n" // Lum4
icculus@4046
   355
		"psraw          $6,                     %%mm5\n"
icculus@4046
   356
		"paddw          %%mm0,                  %%mm7\n" // Lum4 +blue
icculus@4046
   357
		"psraw          $6,                     %%mm6\n" // Lum3 +blue
icculus@4046
   358
		"movq           %%mm3,                  %%mm0\n" // Lum4
icculus@4046
   359
		"packuswb       %%mm4,                  %%mm4\n"
icculus@4046
   360
		"paddw          %%mm1,                  %%mm3\n" // Lum4 +red
icculus@4046
   361
		"packuswb       %%mm5,                  %%mm5\n"
icculus@4046
   362
		"paddw          %%mm2,                  %%mm0\n" // Lum4 +green
icculus@4046
   363
		"packuswb       %%mm6,                  %%mm6\n"
icculus@4046
   364
		"punpcklbw      %%mm4,                  %%mm4\n"
icculus@4046
   365
		"punpcklbw      %%mm5,                  %%mm5\n"
icculus@4046
   366
		"punpcklbw      %%mm6,                  %%mm6\n"
icculus@4046
   367
		"psllw          $3,                     %%mm5\n" // GREEN 3
icculus@4046
   368
		"pand           %16,                    %%mm4\n"
icculus@4046
   369
		"psraw          $6,                     %%mm3\n" // psr 6
icculus@4046
   370
		"psraw          $6,                     %%mm0\n"
icculus@4046
   371
		"pand           %16,                    %%mm6\n" // BLUE
icculus@4046
   372
		"pand           %17,                    %%mm5\n"
icculus@4046
   373
		"psrlw          $11,                    %%mm6\n" // BLUE  3
icculus@4046
   374
		"por            %%mm5,                  %%mm4\n"
icculus@4046
   375
		"psraw          $6,                     %%mm7\n"
icculus@4046
   376
		"por            %%mm6,                  %%mm4\n"
icculus@4046
   377
		"packuswb       %%mm3,                  %%mm3\n"
icculus@4046
   378
		"packuswb       %%mm0,                  %%mm0\n"
icculus@4046
   379
		"packuswb       %%mm7,                  %%mm7\n"
icculus@4046
   380
		"punpcklbw      %%mm3,                  %%mm3\n"
icculus@4046
   381
		"punpcklbw      %%mm0,                  %%mm0\n"
icculus@4046
   382
		"punpcklbw      %%mm7,                  %%mm7\n"
icculus@4046
   383
		"pand           %16,                    %%mm3\n"
icculus@4046
   384
		"pand           %16,                    %%mm7\n" // BLUE
icculus@4046
   385
		"psllw          $3,                     %%mm0\n" // GREEN 4
icculus@4046
   386
		"psrlw          $11,                    %%mm7\n"
icculus@4046
   387
		"pand           %17,                    %%mm0\n"
icculus@4046
   388
		"por            %%mm7,                  %%mm3\n"
icculus@4046
   389
		"por            %%mm0,                  %%mm3\n"
slouken@0
   390
icculus@4046
   391
		"movq           %%mm4,                  %%mm5\n"
slouken@0
   392
icculus@4046
   393
		"punpcklwd      %%mm3,                  %%mm4\n"
icculus@4046
   394
		"punpckhwd      %%mm3,                  %%mm5\n"
slouken@0
   395
icculus@4046
   396
		"movq           %%mm4,                  (%5)\n"
icculus@4046
   397
		"movq           %%mm5,                  8(%5)\n"
slouken@0
   398
icculus@4046
   399
		"addl           $8,                     %6\n"
icculus@4046
   400
		"addl           $8,                     %2\n"
icculus@4046
   401
		"addl           $4,                     (%%esp)\n"
icculus@4046
   402
		"addl           $4,                     %1\n"
icculus@4046
   403
		"cmpl           %4,                     %6\n"
icculus@4046
   404
		"leal           16(%3),                 %3\n"
icculus@4046
   405
		"leal           16(%5),%5\n" // row2+16
slouken@0
   406
icculus@4046
   407
		"jl             1b\n"
icculus@4046
   408
		"addl           %4,     %2\n" // lum += cols
icculus@4046
   409
		"addl           %8,     %3\n" // row1+= mod
icculus@4046
   410
		"addl           %8,     %5\n" // row2+= mod
icculus@4046
   411
		"movl           $0,     %6\n" // x=0
icculus@4046
   412
		"cmpl           %7,     %2\n"
icculus@4046
   413
		"jl             1b\n"
icculus@4046
   414
		"addl $4, %%esp\n"  // get rid of the stack slot we reserved.
icculus@4046
   415
		"emms\n"
icculus@4046
   416
		:
icculus@4046
   417
		: "m" (cr), "r"(cb),"r"(lum),
icculus@4046
   418
		  "r"(row1),"r"(cols),"r"(row2),"m"(x),"m"(y),"m"(mod),
icculus@4046
   419
		  "m"(MMX_0080w),"m"(MMX_Ugrn565),"m"(MMX_Ublu5x5),
icculus@4046
   420
		  "m"(MMX_00FFw),"m"(MMX_Vgrn565),"m"(MMX_Vred5x5),
icculus@4046
   421
		  "m"(MMX_Ycoeff),"m"(MMX_red565),"m"(MMX_grn565)
icculus@4046
   422
	);
slouken@0
   423
}
slouken@0
   424
icculus@4047
   425
/* *INDENT-ON* */
icculus@4047
   426
icculus@4045
   427
#endif /* GCC3 i386 inline assembly */
icculus@4045
   428