Statistics
| Branch: | Revision:

ffmpeg / postproc / yuv2rgb.c @ b5a40dc6

History | View | Annotate | Download (12.1 KB)

1
/*
2
 * yuv2rgb.c, Software YUV to RGB coverter
3
 *
4
 *  Copyright (C) 1999, Aaron Holtzman <aholtzma@ess.engr.uvic.ca>
5
 *  All Rights Reserved.
6
 *
7
 *  Functions broken out from display_x11.c and several new modes
8
 *  added by H?kan Hjort <d95hjort@dtek.chalmers.se>
9
 *
10
 *  15 & 16 bpp support by Franck Sicard <Franck.Sicard@solsoft.fr>
11
 *
12
 *  This file is part of mpeg2dec, a free MPEG-2 video decoder
13
 *
14
 *  mpeg2dec is free software; you can redistribute it and/or modify
15
 *  it under the terms of the GNU General Public License as published by
16
 *  the Free Software Foundation; either version 2, or (at your option)
17
 *  any later version.
18
 *
19
 *  mpeg2dec is distributed in the hope that it will be useful,
20
 *  but WITHOUT ANY WARRANTY; without even the implied warranty of
21
 *  MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
22
 *  GNU General Public License for more details.
23
 *
24
 *  You should have received a copy of the GNU General Public License
25
 *  along with GNU Make; see the file COPYING.  If not, write to
26
 *  the Free Software Foundation, 675 Mass Ave, Cambridge, MA 02139, USA.
27
 *
28
 * MMX/MMX2 Template stuff from Michael Niedermayer (michaelni@gmx.at) (needed for fast movntq support)
29
 */
30

    
31
#include <stdio.h>
32
#include <stdlib.h>
33
#include <inttypes.h>
34

    
35
#include "config.h"
36
//#include "video_out.h"
37
#include "rgb2rgb.h"
38
#include "../cpudetect.h"
39
#include "../mangle.h"
40
#include "../mp_msg.h"
41

    
42
#ifdef HAVE_MLIB
43
#include "yuv2rgb_mlib.c"
44
#endif
45

    
46
#define DITHER1XBPP // only for mmx
47

    
48
#ifdef ARCH_X86
49
#define CAN_COMPILE_X86_ASM
50
#endif
51

    
52
#ifdef CAN_COMPILE_X86_ASM
53

    
54
/* hope these constant values are cache line aligned */
55
uint64_t __attribute__((aligned(8))) mmx_80w = 0x0080008000800080;
56
uint64_t __attribute__((aligned(8))) mmx_10w = 0x1010101010101010;
57
uint64_t __attribute__((aligned(8))) mmx_00ffw = 0x00ff00ff00ff00ff;
58
uint64_t __attribute__((aligned(8))) mmx_Y_coeff = 0x253f253f253f253f;
59

    
60
/* hope these constant values are cache line aligned */
61
uint64_t __attribute__((aligned(8))) mmx_U_green = 0xf37df37df37df37d;
62
uint64_t __attribute__((aligned(8))) mmx_U_blue = 0x4093409340934093;
63
uint64_t __attribute__((aligned(8))) mmx_V_red = 0x3312331233123312;
64
uint64_t __attribute__((aligned(8))) mmx_V_green = 0xe5fce5fce5fce5fc;
65

    
66
/* hope these constant values are cache line aligned */
67
uint64_t __attribute__((aligned(8))) mmx_redmask = 0xf8f8f8f8f8f8f8f8;
68
uint64_t __attribute__((aligned(8))) mmx_grnmask = 0xfcfcfcfcfcfcfcfc;
69

    
70
uint64_t __attribute__((aligned(8))) M24A=   0x00FF0000FF0000FFLL;
71
uint64_t __attribute__((aligned(8))) M24B=   0xFF0000FF0000FF00LL;
72
uint64_t __attribute__((aligned(8))) M24C=   0x0000FF0000FF0000LL;
73

    
74
// the volatile is required because gcc otherwise optimizes some writes away not knowing that these
75
// are read in the asm block
76
volatile uint64_t __attribute__((aligned(8))) b5Dither;
77
volatile uint64_t __attribute__((aligned(8))) g5Dither;
78
volatile uint64_t __attribute__((aligned(8))) g6Dither;
79
volatile uint64_t __attribute__((aligned(8))) r5Dither;
80

    
81
uint64_t __attribute__((aligned(8))) dither4[2]={
82
        0x0103010301030103LL,
83
        0x0200020002000200LL,};
84

    
85
uint64_t __attribute__((aligned(8))) dither8[2]={
86
        0x0602060206020602LL,
87
        0x0004000400040004LL,};
88

    
89
#undef HAVE_MMX
90
#undef ARCH_X86
91

    
92
//MMX versions
93
#undef RENAME
94
#define HAVE_MMX
95
#undef HAVE_MMX2
96
#undef HAVE_3DNOW
97
#define ARCH_X86
98
#define RENAME(a) a ## _MMX
99
#include "yuv2rgb_template.c"
100

    
101
//MMX2 versions
102
#undef RENAME
103
#define HAVE_MMX
104
#define HAVE_MMX2
105
#undef HAVE_3DNOW
106
#define ARCH_X86
107
#define RENAME(a) a ## _MMX2
108
#include "yuv2rgb_template.c"
109

    
110
#endif // CAN_COMPILE_X86_ASM
111

    
112

    
113
uint32_t matrix_coefficients = 6;
114

    
115
const int32_t Inverse_Table_6_9[8][4] = {
116
    {117504, 138453, 13954, 34903}, /* no sequence_display_extension */
117
    {117504, 138453, 13954, 34903}, /* ITU-R Rec. 709 (1990) */
118
    {104597, 132201, 25675, 53279}, /* unspecified */
119
    {104597, 132201, 25675, 53279}, /* reserved */
120
    {104448, 132798, 24759, 53109}, /* FCC */
121
    {104597, 132201, 25675, 53279}, /* ITU-R Rec. 624-4 System B, G */
122
    {104597, 132201, 25675, 53279}, /* SMPTE 170M */
123
    {117579, 136230, 16907, 35559}  /* SMPTE 240M (1987) */
124
};
125

    
126
static void yuv2rgb_c_init (int bpp, int mode);
127

    
128
yuv2rgb_fun yuv2rgb;
129

    
130
static void (* yuv2rgb_c_internal) (uint8_t *, uint8_t *,
131
                                    uint8_t *, uint8_t *,
132
                                    void *, void *, int);
133

    
134
static void yuv2rgb_c (void * dst, uint8_t * py,
135
                       uint8_t * pu, uint8_t * pv,
136
                       int h_size, int v_size,
137
                       int rgb_stride, int y_stride, int uv_stride)
138
{
139
    v_size >>= 1;
140

    
141
    while (v_size--) {
142
        yuv2rgb_c_internal (py, py + y_stride, pu, pv, dst, dst + rgb_stride,
143
                            h_size);
144

    
145
        py += 2 * y_stride;
146
        pu += uv_stride;
147
        pv += uv_stride;
148
        dst += 2 * rgb_stride;
149
    }
150
}
151

    
152
void yuv2rgb_init (int bpp, int mode)
153
{
154
    yuv2rgb = NULL;
155
#ifdef CAN_COMPILE_X86_ASM
156
    if(gCpuCaps.hasMMX2)
157
    {
158
        if (yuv2rgb == NULL /*&& (config.flags & VO_MMX_ENABLE)*/) {
159
                yuv2rgb = yuv2rgb_init_MMX2 (bpp, mode);
160
                if (yuv2rgb != NULL)
161
                        mp_msg(MSGT_SWS,MSGL_INFO,"Using MMX2 for colorspace transform\n");
162
                else
163
                        mp_msg(MSGT_SWS,MSGL_WARN,"Cannot init MMX2 colorspace transform\n");
164
        }
165
    }
166
    else if(gCpuCaps.hasMMX)
167
    {
168
        if (yuv2rgb == NULL /*&& (config.flags & VO_MMX_ENABLE)*/) {
169
                yuv2rgb = yuv2rgb_init_MMX (bpp, mode);
170
                if (yuv2rgb != NULL)
171
                        mp_msg(MSGT_SWS,MSGL_INFO,"Using MMX for colorspace transform\n");
172
                else
173
                        mp_msg(MSGT_SWS,MSGL_WARN,"Cannot init MMX colorspace transform\n");
174
        }
175
    }
176
#endif
177
#ifdef HAVE_MLIB
178
    if (yuv2rgb == NULL /*&& (config.flags & VO_MLIB_ENABLE)*/) {
179
        yuv2rgb = yuv2rgb_init_mlib (bpp, mode);
180
        if (yuv2rgb != NULL)
181
            mp_msg(MSGT_SWS,MSGL_INFO,"Using mlib for colorspace transform\n");
182
    }
183
#endif
184
    if (yuv2rgb == NULL) {
185
        mp_msg(MSGT_SWS,MSGL_INFO,"No accelerated colorspace conversion found\n");
186
        yuv2rgb_c_init (bpp, mode);
187
        yuv2rgb = (yuv2rgb_fun)yuv2rgb_c;
188
    }
189
}
190

    
191
void * table_rV[256];
192
void * table_gU[256];
193
int table_gV[256];
194
void * table_bU[256];
195

    
196
#define RGB(i)                                        \
197
        U = pu[i];                                \
198
        V = pv[i];                                \
199
        r = table_rV[V];                        \
200
        g = table_gU[U] + table_gV[V];                \
201
        b = table_bU[U];
202

    
203
#define DST1(i)                                        \
204
        Y = py_1[2*i];                                \
205
        dst_1[2*i] = r[Y] + g[Y] + b[Y];        \
206
        Y = py_1[2*i+1];                        \
207
        dst_1[2*i+1] = r[Y] + g[Y] + b[Y];
208

    
209
#define DST2(i)                                        \
210
        Y = py_2[2*i];                                \
211
        dst_2[2*i] = r[Y] + g[Y] + b[Y];        \
212
        Y = py_2[2*i+1];                        \
213
        dst_2[2*i+1] = r[Y] + g[Y] + b[Y];
214

    
215
#define DST1RGB(i)                                                        \
216
        Y = py_1[2*i];                                                        \
217
        dst_1[6*i] = r[Y]; dst_1[6*i+1] = g[Y]; dst_1[6*i+2] = b[Y];        \
218
        Y = py_1[2*i+1];                                                \
219
        dst_1[6*i+3] = r[Y]; dst_1[6*i+4] = g[Y]; dst_1[6*i+5] = b[Y];
220

    
221
#define DST2RGB(i)                                                        \
222
        Y = py_2[2*i];                                                        \
223
        dst_2[6*i] = r[Y]; dst_2[6*i+1] = g[Y]; dst_2[6*i+2] = b[Y];        \
224
        Y = py_2[2*i+1];                                                \
225
        dst_2[6*i+3] = r[Y]; dst_2[6*i+4] = g[Y]; dst_2[6*i+5] = b[Y];
226

    
227
#define DST1BGR(i)                                                        \
228
        Y = py_1[2*i];                                                        \
229
        dst_1[6*i] = b[Y]; dst_1[6*i+1] = g[Y]; dst_1[6*i+2] = r[Y];        \
230
        Y = py_1[2*i+1];                                                \
231
        dst_1[6*i+3] = b[Y]; dst_1[6*i+4] = g[Y]; dst_1[6*i+5] = r[Y];
232

    
233
#define DST2BGR(i)                                                        \
234
        Y = py_2[2*i];                                                        \
235
        dst_2[6*i] = b[Y]; dst_2[6*i+1] = g[Y]; dst_2[6*i+2] = r[Y];        \
236
        Y = py_2[2*i+1];                                                \
237
        dst_2[6*i+3] = b[Y]; dst_2[6*i+4] = g[Y]; dst_2[6*i+5] = r[Y];
238

    
239
static void yuv2rgb_c_32 (uint8_t * py_1, uint8_t * py_2,
240
                          uint8_t * pu, uint8_t * pv,
241
                          void * _dst_1, void * _dst_2, int h_size)
242
{
243
    int U, V, Y;
244
    uint32_t * r, * g, * b;
245
    uint32_t * dst_1, * dst_2;
246

    
247
    h_size >>= 3;
248
    dst_1 = _dst_1;
249
    dst_2 = _dst_2;
250

    
251
    while (h_size--) {
252
        RGB(0);
253
        DST1(0);
254
        DST2(0);
255

    
256
        RGB(1);
257
        DST2(1);
258
        DST1(1);
259

    
260
        RGB(2);
261
        DST1(2);
262
        DST2(2);
263

    
264
        RGB(3);
265
        DST2(3);
266
        DST1(3);
267

    
268
        pu += 4;
269
        pv += 4;
270
        py_1 += 8;
271
        py_2 += 8;
272
        dst_1 += 8;
273
        dst_2 += 8;
274
    }
275
}
276

    
277
// This is very near from the yuv2rgb_c_32 code
278
static void yuv2rgb_c_24_rgb (uint8_t * py_1, uint8_t * py_2,
279
                              uint8_t * pu, uint8_t * pv,
280
                              void * _dst_1, void * _dst_2, int h_size)
281
{
282
    int U, V, Y;
283
    uint8_t * r, * g, * b;
284
    uint8_t * dst_1, * dst_2;
285

    
286
    h_size >>= 3;
287
    dst_1 = _dst_1;
288
    dst_2 = _dst_2;
289

    
290
    while (h_size--) {
291
        RGB(0);
292
        DST1RGB(0);
293
        DST2RGB(0);
294

    
295
        RGB(1);
296
        DST2RGB(1);
297
        DST1RGB(1);
298

    
299
        RGB(2);
300
        DST1RGB(2);
301
        DST2RGB(2);
302

    
303
        RGB(3);
304
        DST2RGB(3);
305
        DST1RGB(3);
306

    
307
        pu += 4;
308
        pv += 4;
309
        py_1 += 8;
310
        py_2 += 8;
311
        dst_1 += 24;
312
        dst_2 += 24;
313
    }
314
}
315

    
316
// only trivial mods from yuv2rgb_c_24_rgb
317
static void yuv2rgb_c_24_bgr (uint8_t * py_1, uint8_t * py_2,
318
                              uint8_t * pu, uint8_t * pv,
319
                              void * _dst_1, void * _dst_2, int h_size)
320
{
321
    int U, V, Y;
322
    uint8_t * r, * g, * b;
323
    uint8_t * dst_1, * dst_2;
324

    
325
    h_size >>= 3;
326
    dst_1 = _dst_1;
327
    dst_2 = _dst_2;
328

    
329
    while (h_size--) {
330
        RGB(0);
331
        DST1BGR(0);
332
        DST2BGR(0);
333

    
334
        RGB(1);
335
        DST2BGR(1);
336
        DST1BGR(1);
337

    
338
        RGB(2);
339
        DST1BGR(2);
340
        DST2BGR(2);
341

    
342
        RGB(3);
343
        DST2BGR(3);
344
        DST1BGR(3);
345

    
346
        pu += 4;
347
        pv += 4;
348
        py_1 += 8;
349
        py_2 += 8;
350
        dst_1 += 24;
351
        dst_2 += 24;
352
    }
353
}
354

    
355
// This is exactly the same code as yuv2rgb_c_32 except for the types of
356
// r, g, b, dst_1, dst_2
357
static void yuv2rgb_c_16 (uint8_t * py_1, uint8_t * py_2,
358
                          uint8_t * pu, uint8_t * pv,
359
                          void * _dst_1, void * _dst_2, int h_size)
360
{
361
    int U, V, Y;
362
    uint16_t * r, * g, * b;
363
    uint16_t * dst_1, * dst_2;
364

    
365
    h_size >>= 3;
366
    dst_1 = _dst_1;
367
    dst_2 = _dst_2;
368

    
369
    while (h_size--) {
370
        RGB(0);
371
        DST1(0);
372
        DST2(0);
373

    
374
        RGB(1);
375
        DST2(1);
376
        DST1(1);
377

    
378
        RGB(2);
379
        DST1(2);
380
        DST2(2);
381

    
382
        RGB(3);
383
        DST2(3);
384
        DST1(3);
385

    
386
        pu += 4;
387
        pv += 4;
388
        py_1 += 8;
389
        py_2 += 8;
390
        dst_1 += 8;
391
        dst_2 += 8;
392
    }
393
}
394

    
395
static int div_round (int dividend, int divisor)
396
{
397
    if (dividend > 0)
398
        return (dividend + (divisor>>1)) / divisor;
399
    else
400
        return -((-dividend + (divisor>>1)) / divisor);
401
}
402

    
403
static void yuv2rgb_c_init (int bpp, int mode)
404
{  
405
    int i;
406
    uint8_t table_Y[1024];
407
    uint32_t *table_32 = 0;
408
    uint16_t *table_16 = 0;
409
    uint8_t *table_8 = 0;
410
    int entry_size = 0;
411
    void *table_r = 0, *table_g = 0, *table_b = 0;
412

    
413
    int crv = Inverse_Table_6_9[matrix_coefficients][0];
414
    int cbu = Inverse_Table_6_9[matrix_coefficients][1];
415
    int cgu = -Inverse_Table_6_9[matrix_coefficients][2];
416
    int cgv = -Inverse_Table_6_9[matrix_coefficients][3];
417

    
418
    for (i = 0; i < 1024; i++) {
419
        int j;
420

    
421
        j = (76309 * (i - 384 - 16) + 32768) >> 16;
422
        j = (j < 0) ? 0 : ((j > 255) ? 255 : j);
423
        table_Y[i] = j;
424
    }
425

    
426
    switch (bpp) {
427
    case 32:
428
        yuv2rgb_c_internal = yuv2rgb_c_32;
429

    
430
        table_32 = malloc ((197 + 2*682 + 256 + 132) * sizeof (uint32_t));
431

    
432
        entry_size = sizeof (uint32_t);
433
        table_r = table_32 + 197;
434
        table_b = table_32 + 197 + 685;
435
        table_g = table_32 + 197 + 2*682;
436

    
437
        for (i = -197; i < 256+197; i++)
438
            ((uint32_t *)table_r)[i] = table_Y[i+384] << ((mode==MODE_RGB) ? 16 : 0);
439
        for (i = -132; i < 256+132; i++)
440
            ((uint32_t *)table_g)[i] = table_Y[i+384] << 8;
441
        for (i = -232; i < 256+232; i++)
442
            ((uint32_t *)table_b)[i] = table_Y[i+384] << ((mode==MODE_RGB) ? 0 : 16);
443
        break;
444

    
445
    case 24:
446
//        yuv2rgb_c_internal = (mode==MODE_RGB) ? yuv2rgb_c_24_rgb : yuv2rgb_c_24_bgr;
447
        yuv2rgb_c_internal = (mode!=MODE_RGB) ? yuv2rgb_c_24_rgb : yuv2rgb_c_24_bgr;
448

    
449
        table_8 = malloc ((256 + 2*232) * sizeof (uint8_t));
450

    
451
        entry_size = sizeof (uint8_t);
452
        table_r = table_g = table_b = table_8 + 232;
453

    
454
        for (i = -232; i < 256+232; i++)
455
            ((uint8_t * )table_b)[i] = table_Y[i+384];
456
        break;
457

    
458
    case 15:
459
    case 16:
460
        yuv2rgb_c_internal = yuv2rgb_c_16;
461

    
462
        table_16 = malloc ((197 + 2*682 + 256 + 132) * sizeof (uint16_t));
463

    
464
        entry_size = sizeof (uint16_t);
465
        table_r = table_16 + 197;
466
        table_b = table_16 + 197 + 685;
467
        table_g = table_16 + 197 + 2*682;
468

    
469
        for (i = -197; i < 256+197; i++) {
470
            int j = table_Y[i+384] >> 3;
471

    
472
            if (mode == MODE_RGB)
473
                j <<= ((bpp==16) ? 11 : 10);
474

    
475
            ((uint16_t *)table_r)[i] = j;
476
        }
477
        for (i = -132; i < 256+132; i++) {
478
            int j = table_Y[i+384] >> ((bpp==16) ? 2 : 3);
479

    
480
            ((uint16_t *)table_g)[i] = j << 5;
481
        }
482
        for (i = -232; i < 256+232; i++) {
483
            int j = table_Y[i+384] >> 3;
484

    
485
            if (mode == MODE_BGR)
486
                j <<= ((bpp==16) ? 11 : 10);
487

    
488
            ((uint16_t *)table_b)[i] = j;
489
        }
490
        break;
491

    
492
    default:
493
        mp_msg(MSGT_SWS,MSGL_ERR,"%ibpp not supported by yuv2rgb\n", bpp);
494
        //exit (1);
495
    }
496

    
497
    for (i = 0; i < 256; i++) {
498
        table_rV[i] = table_r + entry_size * div_round (crv * (i-128), 76309);
499
        table_gU[i] = table_g + entry_size * div_round (cgu * (i-128), 76309);
500
        table_gV[i] = entry_size * div_round (cgv * (i-128), 76309);
501
        table_bU[i] = table_b + entry_size * div_round (cbu * (i-128), 76309);
502
    }
503
}