SDL_yuv_mmx.c
1
2
/*
SDL - Simple DirectMedia Layer
3
Copyright (C) 1997-2004 Sam Lantinga
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
This library is free software; you can redistribute it and/or
modify it under the terms of the GNU Library General Public
License as published by the Free Software Foundation; either
version 2 of the License, or (at your option) any later version.
This library is distributed in the hope that it will be useful,
but WITHOUT ANY WARRANTY; without even the implied warranty of
MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
Library General Public License for more details.
You should have received a copy of the GNU Library General Public
License along with this library; if not, write to the Free
Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA
Sam Lantinga
20
slouken@libsdl.org
21
22
23
24
25
26
27
28
29
30
31
32
*/
#ifdef SAVE_RCSID
static char rcsid =
"@(#) $Id$";
#endif
#if defined(i386) && defined(__GNUC__) && defined(USE_ASMBLIT)
#include "SDL_types.h"
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
#if __GNUC__ > 2
# undef GCC2_HACK
#else
# define GCC2_HACK
#endif
#if defined(GCC2_HACK) && defined (__ELF__)
#define ASM_VAR(X) _##X
#else
#define ASM_VAR(X) X
#endif
static volatile unsigned int ASM_VAR(MMX_0080w)[] = {0x00800080, 0x00800080};
static volatile unsigned int ASM_VAR(MMX_00FFw)[] = {0x00ff00ff, 0x00ff00ff};
static volatile unsigned int ASM_VAR(MMX_FF00w)[] = {0xff00ff00, 0xff00ff00};
static volatile unsigned short ASM_VAR(MMX_Ycoeff)[] = {0x004a, 0x004a, 0x004a, 0x004a};
static volatile unsigned short ASM_VAR(MMX_UbluRGB)[] = {0x0072, 0x0072, 0x0072, 0x0072};
static volatile unsigned short ASM_VAR(MMX_VredRGB)[] = {0x0059, 0x0059, 0x0059, 0x0059};
static volatile unsigned short ASM_VAR(MMX_UgrnRGB)[] = {0xffea, 0xffea, 0xffea, 0xffea};
static volatile unsigned short ASM_VAR(MMX_VgrnRGB)[] = {0xffd2, 0xffd2, 0xffd2, 0xffd2};
static volatile unsigned short ASM_VAR(MMX_Ublu5x5)[] = {0x0081, 0x0081, 0x0081, 0x0081};
static volatile unsigned short ASM_VAR(MMX_Vred5x5)[] = {0x0066, 0x0066, 0x0066, 0x0066};
static volatile unsigned short ASM_VAR(MMX_Ugrn555)[] = {0xffe7, 0xffe7, 0xffe7, 0xffe7};
static volatile unsigned short ASM_VAR(MMX_Vgrn555)[] = {0xffcc, 0xffcc, 0xffcc, 0xffcc};
static volatile unsigned short ASM_VAR(MMX_Ugrn565)[] = {0xffe8, 0xffe8, 0xffe8, 0xffe8};
static volatile unsigned short ASM_VAR(MMX_Vgrn565)[] = {0xffcd, 0xffcd, 0xffcd, 0xffcd};
static volatile unsigned short ASM_VAR(MMX_red555)[] = {0x7c00, 0x7c00, 0x7c00, 0x7c00};
static volatile unsigned short ASM_VAR(MMX_red565)[] = {0xf800, 0xf800, 0xf800, 0xf800};
static volatile unsigned short ASM_VAR(MMX_grn555)[] = {0x03e0, 0x03e0, 0x03e0, 0x03e0};
static volatile unsigned short ASM_VAR(MMX_grn565)[] = {0x07e0, 0x07e0, 0x07e0, 0x07e0};
static volatile unsigned short ASM_VAR(MMX_blu5x5)[] = {0x001f, 0x001f, 0x001f, 0x001f};
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
/**
This MMX assembler is my first assembler/MMX program ever.
Thus it maybe buggy.
Send patches to:
mvogt@rhrk.uni-kl.de
After it worked fine I have "obfuscated" the code a bit to have
more parallism in the MMX units. This means I moved
initilisation around and delayed other instruction.
Performance measurement did not show that this brought any advantage
but in theory it _should_ be faster this way.
The overall performanve gain to the C based dither was 30%-40%.
The MMX routine calculates 256bit=8RGB values in each cycle
(4 for row1 & 4 for row2)
The red/green/blue.. coefficents are taken from the mpeg_play
player. They look nice, but I dont know if you can have
better values, to avoid integer rounding errors.
IMPORTANT:
==========
It is a requirement that the cr/cb/lum are 8 byte aligned and
the out are 16byte aligned or you will/may get segfaults
*/
void ColorRGBDitherYV12MMX1X( int *colortab, Uint32 *rgb_2_pix,
unsigned char *lum, unsigned char *cr,
unsigned char *cb, unsigned char *out,
int rows, int cols, int mod )
{
Uint32 *row1;
Uint32 *row2;
unsigned char* y = lum +cols*rows; // Pointer to the end
int x=0;
row1 = (Uint32 *)out; // 32 bit target
row2 = (Uint32 *)out+cols+mod; // start of second row
mod = (mod+cols+mod)*4; // increment for row1 in byte
__asm__ __volatile__ (
/* We don't really care about PIC - the code should be rewritten to use
relative addressing for the static tables, so right now we take the
COW hit on the pages this code resides. Big deal.
This spill is just to reduce register pressure in the PIC case. */
"pushl %%ebx\n"
"movl %0, %%ebx\n"
".align 8\n"
"1:\n"
// create Cr (result in mm1)
"movd (%%ebx), %%mm1\n" // 0 0 0 0 v3 v2 v1 v0
"pxor %%mm7,%%mm7\n" // 00 00 00 00 00 00 00 00
"movd (%2), %%mm2\n" // 0 0 0 0 l3 l2 l1 l0
"punpcklbw %%mm7,%%mm1\n" // 0 v3 0 v2 00 v1 00 v0
"punpckldq %%mm1,%%mm1\n" // 00 v1 00 v0 00 v1 00 v0
130
131
132
#ifdef GCC2_HACK
"psubw _MMX_0080w,%%mm1\n" // mm1-128:r1 r1 r0 r0 r1 r1 r0 r0
#else
133
"psubw %[_MMX_0080w],%%mm1\n" // mm1-128:r1 r1 r0 r0 r1 r1 r0 r0
135
136
137
// create Cr_g (result in mm0)
"movq %%mm1,%%mm0\n" // r1 r1 r0 r0 r1 r1 r0 r0
138
139
140
141
#ifdef GCC2_HACK
"pmullw _MMX_VgrnRGB,%%mm0\n"// red*-46dec=0.7136*64
"pmullw _MMX_VredRGB,%%mm1\n"// red*89dec=1.4013*64
#else
142
143
"pmullw %[_MMX_VgrnRGB],%%mm0\n"// red*-46dec=0.7136*64
"pmullw %[_MMX_VredRGB],%%mm1\n"// red*89dec=1.4013*64
145
146
147
148
149
150
151
152
"psraw $6, %%mm0\n" // red=red/64
"psraw $6, %%mm1\n" // red=red/64
// create L1 L2 (result in mm2,mm4)
// L2=lum+cols
"movq (%2,%4),%%mm3\n" // 0 0 0 0 L3 L2 L1 L0
"punpckldq %%mm3,%%mm2\n" // L3 L2 L1 L0 l3 l2 l1 l0
"movq %%mm2,%%mm4\n" // L3 L2 L1 L0 l3 l2 l1 l0
153
154
155
156
#ifdef GCC2_HACK
"pand _MMX_FF00w,%%mm2\n" // L3 0 L1 0 l3 0 l1 0
"pand _MMX_00FFw,%%mm4\n" // 0 L2 0 L0 0 l2 0 l0
#else
157
158
"pand %[_MMX_FF00w],%%mm2\n" // L3 0 L1 0 l3 0 l1 0
"pand %[_MMX_00FFw],%%mm4\n" // 0 L2 0 L0 0 l2 0 l0
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
"psrlw $8,%%mm2\n" // 0 L3 0 L1 0 l3 0 l1
// create R (result in mm6)
"movq %%mm2,%%mm5\n" // 0 L3 0 L1 0 l3 0 l1
"movq %%mm4,%%mm6\n" // 0 L2 0 L0 0 l2 0 l0
"paddsw %%mm1, %%mm5\n" // lum1+red:x R3 x R1 x r3 x r1
"paddsw %%mm1, %%mm6\n" // lum1+red:x R2 x R0 x r2 x r0
"packuswb %%mm5,%%mm5\n" // R3 R1 r3 r1 R3 R1 r3 r1
"packuswb %%mm6,%%mm6\n" // R2 R0 r2 r0 R2 R0 r2 r0
"pxor %%mm7,%%mm7\n" // 00 00 00 00 00 00 00 00
"punpcklbw %%mm5,%%mm6\n" // R3 R2 R1 R0 r3 r2 r1 r0
// create Cb (result in mm1)
"movd (%1), %%mm1\n" // 0 0 0 0 u3 u2 u1 u0
"punpcklbw %%mm7,%%mm1\n" // 0 u3 0 u2 00 u1 00 u0
"punpckldq %%mm1,%%mm1\n" // 00 u1 00 u0 00 u1 00 u0
176
177
178
#ifdef GCC2_HACK
"psubw _MMX_0080w,%%mm1\n" // mm1-128:u1 u1 u0 u0 u1 u1 u0 u0
#else
179
"psubw %[_MMX_0080w],%%mm1\n" // mm1-128:u1 u1 u0 u0 u1 u1 u0 u0
181
182
// create Cb_g (result in mm5)
"movq %%mm1,%%mm5\n" // u1 u1 u0 u0 u1 u1 u0 u0
183
184
185
186
#ifdef GCC2_HACK
"pmullw _MMX_UgrnRGB,%%mm5\n" // blue*-109dec=1.7129*64
"pmullw _MMX_UbluRGB,%%mm1\n" // blue*114dec=1.78125*64
#else
187
188
"pmullw %[_MMX_UgrnRGB],%%mm5\n" // blue*-109dec=1.7129*64
"pmullw %[_MMX_UbluRGB],%%mm1\n" // blue*114dec=1.78125*64
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
"psraw $6, %%mm5\n" // blue=red/64
"psraw $6, %%mm1\n" // blue=blue/64
// create G (result in mm7)
"movq %%mm2,%%mm3\n" // 0 L3 0 L1 0 l3 0 l1
"movq %%mm4,%%mm7\n" // 0 L2 0 L0 0 l2 0 l1
"paddsw %%mm5, %%mm3\n" // lum1+Cb_g:x G3t x G1t x g3t x g1t
"paddsw %%mm5, %%mm7\n" // lum1+Cb_g:x G2t x G0t x g2t x g0t
"paddsw %%mm0, %%mm3\n" // lum1+Cr_g:x G3 x G1 x g3 x g1
"paddsw %%mm0, %%mm7\n" // lum1+blue:x G2 x G0 x g2 x g0
"packuswb %%mm3,%%mm3\n" // G3 G1 g3 g1 G3 G1 g3 g1
"packuswb %%mm7,%%mm7\n" // G2 G0 g2 g0 G2 G0 g2 g0
"punpcklbw %%mm3,%%mm7\n" // G3 G2 G1 G0 g3 g2 g1 g0
// create B (result in mm5)
"movq %%mm2,%%mm3\n" // 0 L3 0 L1 0 l3 0 l1
"movq %%mm4,%%mm5\n" // 0 L2 0 L0 0 l2 0 l1
"paddsw %%mm1, %%mm3\n" // lum1+blue:x B3 x B1 x b3 x b1
"paddsw %%mm1, %%mm5\n" // lum1+blue:x B2 x B0 x b2 x b0
"packuswb %%mm3,%%mm3\n" // B3 B1 b3 b1 B3 B1 b3 b1
"packuswb %%mm5,%%mm5\n" // B2 B0 b2 b0 B2 B0 b2 b0
"punpcklbw %%mm3,%%mm5\n" // B3 B2 B1 B0 b3 b2 b1 b0
// fill destination row1 (needed are mm6=Rr,mm7=Gg,mm5=Bb)
"pxor %%mm2,%%mm2\n" // 0 0 0 0 0 0 0 0
"pxor %%mm4,%%mm4\n" // 0 0 0 0 0 0 0 0
"movq %%mm6,%%mm1\n" // R3 R2 R1 R0 r3 r2 r1 r0
"movq %%mm5,%%mm3\n" // B3 B2 B1 B0 b3 b2 b1 b0
// process lower lum
"punpcklbw %%mm4,%%mm1\n" // 0 r3 0 r2 0 r1 0 r0
"punpcklbw %%mm4,%%mm3\n" // 0 b3 0 b2 0 b1 0 b0
"movq %%mm1,%%mm2\n" // 0 r3 0 r2 0 r1 0 r0
"movq %%mm3,%%mm0\n" // 0 b3 0 b2 0 b1 0 b0
"punpcklwd %%mm1,%%mm3\n" // 0 r1 0 b1 0 r0 0 b0
"punpckhwd %%mm2,%%mm0\n" // 0 r3 0 b3 0 r2 0 b2
"pxor %%mm2,%%mm2\n" // 0 0 0 0 0 0 0 0
"movq %%mm7,%%mm1\n" // G3 G2 G1 G0 g3 g2 g1 g0
"punpcklbw %%mm1,%%mm2\n" // g3 0 g2 0 g1 0 g0 0
"punpcklwd %%mm4,%%mm2\n" // 0 0 g1 0 0 0 g0 0
"por %%mm3, %%mm2\n" // 0 r1 g1 b1 0 r0 g0 b0
"movq %%mm2,(%3)\n" // wrote out ! row1
"pxor %%mm2,%%mm2\n" // 0 0 0 0 0 0 0 0
"punpcklbw %%mm1,%%mm4\n" // g3 0 g2 0 g1 0 g0 0
"punpckhwd %%mm2,%%mm4\n" // 0 0 g3 0 0 0 g2 0
"por %%mm0, %%mm4\n" // 0 r3 g3 b3 0 r2 g2 b2
"movq %%mm4,8(%3)\n" // wrote out ! row1
// fill destination row2 (needed are mm6=Rr,mm7=Gg,mm5=Bb)
// this can be done "destructive"
"pxor %%mm2,%%mm2\n" // 0 0 0 0 0 0 0 0
"punpckhbw %%mm2,%%mm6\n" // 0 R3 0 R2 0 R1 0 R0
"punpckhbw %%mm1,%%mm5\n" // G3 B3 G2 B2 G1 B1 G0 B0
"movq %%mm5,%%mm1\n" // G3 B3 G2 B2 G1 B1 G0 B0
"punpcklwd %%mm6,%%mm1\n" // 0 R1 G1 B1 0 R0 G0 B0
"movq %%mm1,(%5)\n" // wrote out ! row2
"punpckhwd %%mm6,%%mm5\n" // 0 R3 G3 B3 0 R2 G2 B2
"movq %%mm5,8(%5)\n" // wrote out ! row2
"addl $4,%2\n" // lum+4
"leal 16(%3),%3\n" // row1+16
"leal 16(%5),%5\n" // row2+16
"addl $2, %%ebx\n" // cr+2
"addl $2, %1\n" // cb+2
"addl $4,%6\n" // x+4
"cmpl %4,%6\n"
"jl 1b\n"
"addl %4, %2\n" // lum += cols
"addl %8, %3\n" // row1+= mod
"addl %8, %5\n" // row2+= mod
"movl $0, %6\n" // x=0
"cmpl %7, %2\n"
"jl 1b\n"
"emms\n"
"popl %%ebx\n"
:
: "m" (cr), "r"(cb),"r"(lum),
271
272
273
274
275
276
277
278
279
280
281
"r"(row1),"r"(cols),"r"(row2),"m"(x),"m"(y),"m"(mod)
#ifndef GCC2_HACK
,[_MMX_0080w] "m" (*MMX_0080w),
[_MMX_00FFw] "m" (*MMX_00FFw),
[_MMX_FF00w] "m" (*MMX_FF00w),
[_MMX_VgrnRGB] "m" (*MMX_VgrnRGB),
[_MMX_VredRGB] "m" (*MMX_VredRGB),
[_MMX_UgrnRGB] "m" (*MMX_UgrnRGB),
[_MMX_UbluRGB] "m" (*MMX_UbluRGB)
#endif
);
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
}
void Color565DitherYV12MMX1X( int *colortab, Uint32 *rgb_2_pix,
unsigned char *lum, unsigned char *cr,
unsigned char *cb, unsigned char *out,
int rows, int cols, int mod )
{
Uint16 *row1;
Uint16 *row2;
unsigned char* y = lum +cols*rows; /* Pointer to the end */
int x=0;
row1 = (Uint16 *)out; /* 16 bit target */
row2 = (Uint16 *)out+cols+mod; /* start of second row */
mod = (mod+cols+mod)*2; /* increment for row1 in byte */
__asm__ __volatile__(
"pushl %%ebx\n"
"movl %0, %%ebx\n"
".align 8\n"
"1:\n"
"movd (%1), %%mm0\n" // 4 Cb 0 0 0 0 u3 u2 u1 u0
"pxor %%mm7, %%mm7\n"
"movd (%%ebx), %%mm1\n" // 4 Cr 0 0 0 0 v3 v2 v1 v0
"punpcklbw %%mm7, %%mm0\n" // 4 W cb 0 u3 0 u2 0 u1 0 u0
"punpcklbw %%mm7, %%mm1\n" // 4 W cr 0 v3 0 v2 0 v1 0 v0
310
311
312
313
#ifdef GCC2_HACK
"psubw _MMX_0080w, %%mm0\n"
"psubw _MMX_0080w, %%mm1\n"
#else
314
315
"psubw %[_MMX_0080w], %%mm0\n"
"psubw %[_MMX_0080w], %%mm1\n"
317
318
"movq %%mm0, %%mm2\n" // Cb 0 u3 0 u2 0 u1 0 u0
"movq %%mm1, %%mm3\n" // Cr
319
320
321
#ifdef GCC2_HACK
"pmullw _MMX_Ugrn565, %%mm2\n" // Cb2green 0 R3 0 R2 0 R1 0 R0
#else
322
"pmullw %[_MMX_Ugrn565], %%mm2\n" // Cb2green 0 R3 0 R2 0 R1 0 R0
324
"movq (%2), %%mm6\n" // L1 l7 L6 L5 L4 L3 L2 L1 L0
325
326
327
328
329
#ifdef GCC2_HACK
"pmullw _MMX_Ublu5x5, %%mm0\n" // Cb2blue
"pand _MMX_00FFw, %%mm6\n" // L1 00 L6 00 L4 00 L2 00 L0
"pmullw _MMX_Vgrn565, %%mm3\n" // Cr2green
#else
330
331
332
"pmullw %[_MMX_Ublu5x5], %%mm0\n" // Cb2blue
"pand %[_MMX_00FFw], %%mm6\n" // L1 00 L6 00 L4 00 L2 00 L0
"pmullw %[_MMX_Vgrn565], %%mm3\n" // Cr2green
334
"movq (%2), %%mm7\n" // L2
335
336
337
#ifdef GCC2_HACK
"pmullw _MMX_Vred5x5, %%mm1\n" // Cr2red
#else
338
"pmullw %[_MMX_Vred5x5], %%mm1\n" // Cr2red
340
"psrlw $8, %%mm7\n" // L2 00 L7 00 L5 00 L3 00 L1
341
342
343
#ifdef GCC2_HACK
"pmullw _MMX_Ycoeff, %%mm6\n" // lum1
#else
344
"pmullw %[_MMX_Ycoeff], %%mm6\n" // lum1
346
"paddw %%mm3, %%mm2\n" // Cb2green + Cr2green == green
347
348
349
#ifdef GCC2_HACK
"pmullw _MMX_Ycoeff, %%mm7\n" // lum2
#else
350
"pmullw %[_MMX_Ycoeff], %%mm7\n" // lum2
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
"movq %%mm6, %%mm4\n" // lum1
"paddw %%mm0, %%mm6\n" // lum1 +blue 00 B6 00 B4 00 B2 00 B0
"movq %%mm4, %%mm5\n" // lum1
"paddw %%mm1, %%mm4\n" // lum1 +red 00 R6 00 R4 00 R2 00 R0
"paddw %%mm2, %%mm5\n" // lum1 +green 00 G6 00 G4 00 G2 00 G0
"psraw $6, %%mm4\n" // R1 0 .. 64
"movq %%mm7, %%mm3\n" // lum2 00 L7 00 L5 00 L3 00 L1
"psraw $6, %%mm5\n" // G1 - .. +
"paddw %%mm0, %%mm7\n" // Lum2 +blue 00 B7 00 B5 00 B3 00 B1
"psraw $6, %%mm6\n" // B1 0 .. 64
"packuswb %%mm4, %%mm4\n" // R1 R1
"packuswb %%mm5, %%mm5\n" // G1 G1
"packuswb %%mm6, %%mm6\n" // B1 B1
"punpcklbw %%mm4, %%mm4\n"
"punpcklbw %%mm5, %%mm5\n"
369
370
371
#ifdef GCC2_HACK
"pand _MMX_red565, %%mm4\n"
#else
372
"pand %[_MMX_red565], %%mm4\n"
374
375
"psllw $3, %%mm5\n" // GREEN 1
"punpcklbw %%mm6, %%mm6\n"
376
377
378
379
#ifdef GCC2_HACK
"pand _MMX_grn565, %%mm5\n"
"pand _MMX_red565, %%mm6\n"
#else
380
381
"pand %[_MMX_grn565], %%mm5\n"
"pand %[_MMX_red565], %%mm6\n"
383
384
385
386
387
388
389
390
391
392
393
394
395
"por %%mm5, %%mm4\n" //
"psrlw $11, %%mm6\n" // BLUE 1
"movq %%mm3, %%mm5\n" // lum2
"paddw %%mm1, %%mm3\n" // lum2 +red 00 R7 00 R5 00 R3 00 R1
"paddw %%mm2, %%mm5\n" // lum2 +green 00 G7 00 G5 00 G3 00 G1
"psraw $6, %%mm3\n" // R2
"por %%mm6, %%mm4\n" // MM4
"psraw $6, %%mm5\n" // G2
"movq (%2, %4), %%mm6\n" // L3 load lum2
"psraw $6, %%mm7\n"
"packuswb %%mm3, %%mm3\n"
"packuswb %%mm5, %%mm5\n"
"packuswb %%mm7, %%mm7\n"
396
397
398
#ifdef GCC2_HACK
"pand _MMX_00FFw, %%mm6\n" // L3
#else
399
"pand %[_MMX_00FFw], %%mm6\n" // L3
401
402
"punpcklbw %%mm3, %%mm3\n"
"punpcklbw %%mm5, %%mm5\n"
403
404
405
#ifdef GCC2_HACK
"pmullw _MMX_Ycoeff, %%mm6\n" // lum3
#else
406
"pmullw %[_MMX_Ycoeff], %%mm6\n" // lum3
408
409
"punpcklbw %%mm7, %%mm7\n"
"psllw $3, %%mm5\n" // GREEN 2
410
411
412
413
#ifdef GCC2_HACK
"pand _MMX_red565, %%mm7\n"
"pand _MMX_red565, %%mm3\n"
#else
414
415
"pand %[_MMX_red565], %%mm7\n"
"pand %[_MMX_red565], %%mm3\n"
417
"psrlw $11, %%mm7\n" // BLUE 2
418
419
420
#ifdef GCC2_HACK
"pand _MMX_grn565, %%mm5\n"
#else
421
"pand %[_MMX_grn565], %%mm5\n"
423
424
425
426
427
428
"por %%mm7, %%mm3\n"
"movq (%2,%4), %%mm7\n" // L4 load lum2
"por %%mm5, %%mm3\n" //
"psrlw $8, %%mm7\n" // L4
"movq %%mm4, %%mm5\n"
"punpcklwd %%mm3, %%mm4\n"
429
430
431
#ifdef GCC2_HACK
"pmullw _MMX_Ycoeff, %%mm7\n" // lum4
#else
432
"pmullw %[_MMX_Ycoeff], %%mm7\n" // lum4
434
435
436
437
438
439
440
441
442
443
444
445
446
447
448
449
450
451
452
453
454
455
456
457
458
459
"punpckhwd %%mm3, %%mm5\n"
"movq %%mm4, (%3)\n" // write row1
"movq %%mm5, 8(%3)\n" // write row1
"movq %%mm6, %%mm4\n" // Lum3
"paddw %%mm0, %%mm6\n" // Lum3 +blue
"movq %%mm4, %%mm5\n" // Lum3
"paddw %%mm1, %%mm4\n" // Lum3 +red
"paddw %%mm2, %%mm5\n" // Lum3 +green
"psraw $6, %%mm4\n"
"movq %%mm7, %%mm3\n" // Lum4
"psraw $6, %%mm5\n"
"paddw %%mm0, %%mm7\n" // Lum4 +blue
"psraw $6, %%mm6\n" // Lum3 +blue
"movq %%mm3, %%mm0\n" // Lum4
"packuswb %%mm4, %%mm4\n"
"paddw %%mm1, %%mm3\n" // Lum4 +red
"packuswb %%mm5, %%mm5\n"
"paddw %%mm2, %%mm0\n" // Lum4 +green
"packuswb %%mm6, %%mm6\n"
"punpcklbw %%mm4, %%mm4\n"
"punpcklbw %%mm5, %%mm5\n"
"punpcklbw %%mm6, %%mm6\n"
"psllw $3, %%mm5\n" // GREEN 3
460
461
462
#ifdef GCC2_HACK
"pand _MMX_red565, %%mm4\n"
#else
463
"pand %[_MMX_red565], %%mm4\n"
465
466
"psraw $6, %%mm3\n" // psr 6
"psraw $6, %%mm0\n"
467
468
469
470
#ifdef GCC2_HACK
"pand _MMX_red565, %%mm6\n" // BLUE
"pand _MMX_grn565, %%mm5\n"
#else
471
472
"pand %[_MMX_red565], %%mm6\n" // BLUE
"pand %[_MMX_grn565], %%mm5\n"
474
475
476
477
478
479
480
481
482
483
"psrlw $11, %%mm6\n" // BLUE 3
"por %%mm5, %%mm4\n"
"psraw $6, %%mm7\n"
"por %%mm6, %%mm4\n"
"packuswb %%mm3, %%mm3\n"
"packuswb %%mm0, %%mm0\n"
"packuswb %%mm7, %%mm7\n"
"punpcklbw %%mm3, %%mm3\n"
"punpcklbw %%mm0, %%mm0\n"
"punpcklbw %%mm7, %%mm7\n"
484
485
486
487
#ifdef GCC2_HACK
"pand _MMX_red565, %%mm3\n"
"pand _MMX_red565, %%mm7\n" // BLUE
#else
488
489
"pand %[_MMX_red565], %%mm3\n"
"pand %[_MMX_red565], %%mm7\n" // BLUE
491
492
"psllw $3, %%mm0\n" // GREEN 4
"psrlw $11, %%mm7\n"
493
494
495
#ifdef GCC2_HACK
"pand _MMX_grn565, %%mm0\n"
#else
496
"pand %[_MMX_grn565], %%mm0\n"
498
499
500
501
502
503
504
505
506
507
508
509
510
511
512
513
514
515
516
517
518
519
520
521
522
523
524
525
526
"por %%mm7, %%mm3\n"
"por %%mm0, %%mm3\n"
"movq %%mm4, %%mm5\n"
"punpcklwd %%mm3, %%mm4\n"
"punpckhwd %%mm3, %%mm5\n"
"movq %%mm4, (%5)\n"
"movq %%mm5, 8(%5)\n"
"addl $8, %6\n"
"addl $8, %2\n"
"addl $4, %%ebx\n"
"addl $4, %1\n"
"cmpl %4, %6\n"
"leal 16(%3), %3\n"
"leal 16(%5),%5\n" // row2+16
"jl 1b\n"
"addl %4, %2\n" // lum += cols
"addl %8, %3\n" // row1+= mod
"addl %8, %5\n" // row2+= mod
"movl $0, %6\n" // x=0
"cmpl %7, %2\n"
"jl 1b\n"
"emms\n"
"popl %%ebx\n"
527
528
529
530
531
:
:"m" (cr), "r"(cb),"r"(lum),
"r"(row1),"r"(cols),"r"(row2),"m"(x),"m"(y),"m"(mod)
#ifndef GCC2_HACK
,[_MMX_0080w] "m" (*MMX_0080w),
532
533
534
535
536
537
538
539
[_MMX_Ugrn565] "m" (*MMX_Ugrn565),
[_MMX_Ublu5x5] "m" (*MMX_Ublu5x5),
[_MMX_00FFw] "m" (*MMX_00FFw),
[_MMX_Vgrn565] "m" (*MMX_Vgrn565),
[_MMX_Vred5x5] "m" (*MMX_Vred5x5),
[_MMX_Ycoeff] "m" (*MMX_Ycoeff),
[_MMX_red565] "m" (*MMX_red565),
[_MMX_grn565] "m" (*MMX_grn565)
540
541
#endif
);
542
543
}
544
545
#undef GCC2_HACK
546
#endif /* GCC i386 inline assembly */