yuv2rgb_template.c
Go to the documentation of this file.
1 /*
2  * software YUV to RGB converter
3  *
4  * Copyright (C) 2001-2007 Michael Niedermayer
5  * (c) 2010 Konstantin Shishkov
6  *
7  * This file is part of Libav.
8  *
9  * Libav is free software; you can redistribute it and/or
10  * modify it under the terms of the GNU Lesser General Public
11  * License as published by the Free Software Foundation; either
12  * version 2.1 of the License, or (at your option) any later version.
13  *
14  * Libav is distributed in the hope that it will be useful,
15  * but WITHOUT ANY WARRANTY; without even the implied warranty of
16  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
17  * Lesser General Public License for more details.
18  *
19  * You should have received a copy of the GNU Lesser General Public
20  * License along with Libav; if not, write to the Free Software
21  * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
22  */
23 
24 #undef MOVNTQ
25 #undef EMMS
26 #undef SFENCE
27 
28 #if COMPILE_TEMPLATE_MMX2
29 #define MOVNTQ "movntq"
30 #define SFENCE "sfence"
31 #else
32 #define MOVNTQ "movq"
33 #define SFENCE " # nop"
34 #endif
35 
36 #define REG_BLUE "0"
37 #define REG_RED "1"
38 #define REG_GREEN "2"
39 #define REG_ALPHA "3"
40 
41 #define YUV2RGB_LOOP(depth) \
42  h_size = (c->dstW + 7) & ~7; \
43  if (h_size * depth > FFABS(dstStride[0])) \
44  h_size -= 8; \
45  \
46  if (c->srcFormat == PIX_FMT_YUV422P) { \
47  srcStride[1] *= 2; \
48  srcStride[2] *= 2; \
49  } \
50  \
51  __asm__ volatile ("pxor %mm4, %mm4\n\t"); \
52  for (y = 0; y < srcSliceH; y++) { \
53  uint8_t *image = dst[0] + (y + srcSliceY) * dstStride[0]; \
54  const uint8_t *py = src[0] + y * srcStride[0]; \
55  const uint8_t *pu = src[1] + (y >> 1) * srcStride[1]; \
56  const uint8_t *pv = src[2] + (y >> 1) * srcStride[2]; \
57  x86_reg index = -h_size / 2; \
58 
59 #define YUV2RGB_INITIAL_LOAD \
60  __asm__ volatile ( \
61  "movq (%5, %0, 2), %%mm6\n\t" \
62  "movd (%2, %0), %%mm0\n\t" \
63  "movd (%3, %0), %%mm1\n\t" \
64  "1: \n\t" \
65 
66 /* YUV2RGB core
67  * Conversion is performed in usual way:
68  * R = Y' * Ycoef + Vred * V'
69  * G = Y' * Ycoef + Vgreen * V' + Ugreen * U'
70  * B = Y' * Ycoef + Ublue * U'
71  *
72  * where X' = X * 8 - Xoffset (multiplication is performed to increase
73  * precision a bit).
74  * Since it operates in YUV420 colorspace, Y component is additionally
75  * split into Y1 and Y2 for even and odd pixels.
76  *
77  * Input:
78  * mm0 - U (4 elems), mm1 - V (4 elems), mm6 - Y (8 elems), mm4 - zero register
79  * Output:
80  * mm1 - R, mm2 - G, mm0 - B
81  */
82 #define YUV2RGB \
83  /* convert Y, U, V into Y1', Y2', U', V' */ \
84  "movq %%mm6, %%mm7\n\t" \
85  "punpcklbw %%mm4, %%mm0\n\t" \
86  "punpcklbw %%mm4, %%mm1\n\t" \
87  "pand "MANGLE(mmx_00ffw)", %%mm6\n\t" \
88  "psrlw $8, %%mm7\n\t" \
89  "psllw $3, %%mm0\n\t" \
90  "psllw $3, %%mm1\n\t" \
91  "psllw $3, %%mm6\n\t" \
92  "psllw $3, %%mm7\n\t" \
93  "psubsw "U_OFFSET"(%4), %%mm0\n\t" \
94  "psubsw "V_OFFSET"(%4), %%mm1\n\t" \
95  "psubw "Y_OFFSET"(%4), %%mm6\n\t" \
96  "psubw "Y_OFFSET"(%4), %%mm7\n\t" \
97 \
98  /* multiply by coefficients */ \
99  "movq %%mm0, %%mm2\n\t" \
100  "movq %%mm1, %%mm3\n\t" \
101  "pmulhw "UG_COEFF"(%4), %%mm2\n\t" \
102  "pmulhw "VG_COEFF"(%4), %%mm3\n\t" \
103  "pmulhw "Y_COEFF" (%4), %%mm6\n\t" \
104  "pmulhw "Y_COEFF" (%4), %%mm7\n\t" \
105  "pmulhw "UB_COEFF"(%4), %%mm0\n\t" \
106  "pmulhw "VR_COEFF"(%4), %%mm1\n\t" \
107  "paddsw %%mm3, %%mm2\n\t" \
108  /* now: mm0 = UB, mm1 = VR, mm2 = CG */ \
109  /* mm6 = Y1, mm7 = Y2 */ \
110 \
111  /* produce RGB */ \
112  "movq %%mm7, %%mm3\n\t" \
113  "movq %%mm7, %%mm5\n\t" \
114  "paddsw %%mm0, %%mm3\n\t" \
115  "paddsw %%mm1, %%mm5\n\t" \
116  "paddsw %%mm2, %%mm7\n\t" \
117  "paddsw %%mm6, %%mm0\n\t" \
118  "paddsw %%mm6, %%mm1\n\t" \
119  "paddsw %%mm6, %%mm2\n\t" \
120 
121 #define RGB_PACK_INTERLEAVE \
122  /* pack and interleave even/odd pixels */ \
123  "packuswb %%mm1, %%mm0\n\t" \
124  "packuswb %%mm5, %%mm3\n\t" \
125  "packuswb %%mm2, %%mm2\n\t" \
126  "movq %%mm0, %%mm1\n\n" \
127  "packuswb %%mm7, %%mm7\n\t" \
128  "punpcklbw %%mm3, %%mm0\n\t" \
129  "punpckhbw %%mm3, %%mm1\n\t" \
130  "punpcklbw %%mm7, %%mm2\n\t" \
131 
132 #define YUV2RGB_ENDLOOP(depth) \
133  "movq 8 (%5, %0, 2), %%mm6\n\t" \
134  "movd 4 (%3, %0), %%mm1\n\t" \
135  "movd 4 (%2, %0), %%mm0\n\t" \
136  "add $"AV_STRINGIFY(depth * 8)", %1\n\t" \
137  "add $4, %0\n\t" \
138  "js 1b\n\t" \
139 
140 #define YUV2RGB_OPERANDS \
141  : "+r" (index), "+r" (image) \
142  : "r" (pu - index), "r" (pv - index), "r"(&c->redDither), \
143  "r" (py - 2*index) \
144  ); \
145  } \
146 
147 #define YUV2RGB_OPERANDS_ALPHA \
148  : "+r" (index), "+r" (image) \
149  : "r" (pu - index), "r" (pv - index), "r"(&c->redDither), \
150  "r" (py - 2*index), "r" (pa - 2*index) \
151  ); \
152  } \
153 
154 #define YUV2RGB_ENDFUNC \
155  __asm__ volatile (SFENCE"\n\t" \
156  "emms \n\t"); \
157  return srcSliceH; \
158 
159 #define IF0(x)
160 #define IF1(x) x
161 
162 #define RGB_PACK16(gmask, is15) \
163  "pand "MANGLE(mmx_redmask)", %%mm0\n\t" \
164  "pand "MANGLE(mmx_redmask)", %%mm1\n\t" \
165  "movq %%mm2, %%mm3\n\t" \
166  "psllw $"AV_STRINGIFY(3-is15)", %%mm2\n\t" \
167  "psrlw $"AV_STRINGIFY(5+is15)", %%mm3\n\t" \
168  "psrlw $3, %%mm0\n\t" \
169  IF##is15("psrlw $1, %%mm1\n\t") \
170  "pand "MANGLE(pb_e0)", %%mm2\n\t" \
171  "pand "MANGLE(gmask)", %%mm3\n\t" \
172  "por %%mm2, %%mm0\n\t" \
173  "por %%mm3, %%mm1\n\t" \
174  "movq %%mm0, %%mm2\n\t" \
175  "punpcklbw %%mm1, %%mm0\n\t" \
176  "punpckhbw %%mm1, %%mm2\n\t" \
177  MOVNTQ " %%mm0, (%1)\n\t" \
178  MOVNTQ " %%mm2, 8(%1)\n\t" \
179 
180 #define DITHER_RGB \
181  "paddusb "BLUE_DITHER"(%4), %%mm0\n\t" \
182  "paddusb "GREEN_DITHER"(%4), %%mm2\n\t" \
183  "paddusb "RED_DITHER"(%4), %%mm1\n\t" \
184 
185 #if !COMPILE_TEMPLATE_MMX2
186 static inline int RENAME(yuv420_rgb15)(SwsContext *c, const uint8_t *src[],
187  int srcStride[],
188  int srcSliceY, int srcSliceH,
189  uint8_t *dst[], int dstStride[])
190 {
191  int y, h_size;
192 
193  YUV2RGB_LOOP(2)
194 
195 #ifdef DITHER1XBPP
196  c->blueDither = ff_dither8[y & 1];
197  c->greenDither = ff_dither8[y & 1];
198  c->redDither = ff_dither8[(y + 1) & 1];
199 #endif
200 
202  YUV2RGB
204 #ifdef DITHER1XBPP
205  DITHER_RGB
206 #endif
207  RGB_PACK16(pb_03, 1)
208 
209  YUV2RGB_ENDLOOP(2)
212 }
213 
214 static inline int RENAME(yuv420_rgb16)(SwsContext *c, const uint8_t *src[],
215  int srcStride[],
216  int srcSliceY, int srcSliceH,
217  uint8_t *dst[], int dstStride[])
218 {
219  int y, h_size;
220 
221  YUV2RGB_LOOP(2)
222 
223 #ifdef DITHER1XBPP
224  c->blueDither = ff_dither8[y & 1];
225  c->greenDither = ff_dither4[y & 1];
226  c->redDither = ff_dither8[(y + 1) & 1];
227 #endif
228 
230  YUV2RGB
232 #ifdef DITHER1XBPP
233  DITHER_RGB
234 #endif
235  RGB_PACK16(pb_07, 0)
236 
237  YUV2RGB_ENDLOOP(2)
240 }
241 #endif /* !COMPILE_TEMPLATE_MMX2 */
242 
243 #define RGB_PACK24(blue, red)\
244  "packuswb %%mm3, %%mm0 \n" /* R0 R2 R4 R6 R1 R3 R5 R7 */\
245  "packuswb %%mm5, %%mm1 \n" /* B0 B2 B4 B6 B1 B3 B5 B7 */\
246  "packuswb %%mm7, %%mm2 \n" /* G0 G2 G4 G6 G1 G3 G5 G7 */\
247  "movq %%mm"red", %%mm3 \n"\
248  "movq %%mm"blue", %%mm6 \n"\
249  "psrlq $32, %%mm"red" \n" /* R1 R3 R5 R7 */\
250  "punpcklbw %%mm2, %%mm3 \n" /* R0 G0 R2 G2 R4 G4 R6 G6 */\
251  "punpcklbw %%mm"red", %%mm6 \n" /* B0 R1 B2 R3 B4 R5 B6 R7 */\
252  "movq %%mm3, %%mm5 \n"\
253  "punpckhbw %%mm"blue", %%mm2 \n" /* G1 B1 G3 B3 G5 B5 G7 B7 */\
254  "punpcklwd %%mm6, %%mm3 \n" /* R0 G0 B0 R1 R2 G2 B2 R3 */\
255  "punpckhwd %%mm6, %%mm5 \n" /* R4 G4 B4 R5 R6 G6 B6 R7 */\
256  RGB_PACK24_B
257 
258 #if COMPILE_TEMPLATE_MMX2
259 DECLARE_ASM_CONST(8, int16_t, mask1101[4]) = {-1,-1, 0,-1};
260 DECLARE_ASM_CONST(8, int16_t, mask0010[4]) = { 0, 0,-1, 0};
261 DECLARE_ASM_CONST(8, int16_t, mask0110[4]) = { 0,-1,-1, 0};
262 DECLARE_ASM_CONST(8, int16_t, mask1001[4]) = {-1, 0, 0,-1};
263 DECLARE_ASM_CONST(8, int16_t, mask0100[4]) = { 0,-1, 0, 0};
264 #undef RGB_PACK24_B
265 #define RGB_PACK24_B\
266  "pshufw $0xc6, %%mm2, %%mm1 \n"\
267  "pshufw $0x84, %%mm3, %%mm6 \n"\
268  "pshufw $0x38, %%mm5, %%mm7 \n"\
269  "pand "MANGLE(mask1101)", %%mm6 \n" /* R0 G0 B0 R1 -- -- R2 G2 */\
270  "movq %%mm1, %%mm0 \n"\
271  "pand "MANGLE(mask0110)", %%mm7 \n" /* -- -- R6 G6 B6 R7 -- -- */\
272  "movq %%mm1, %%mm2 \n"\
273  "pand "MANGLE(mask0100)", %%mm1 \n" /* -- -- G3 B3 -- -- -- -- */\
274  "psrlq $48, %%mm3 \n" /* B2 R3 -- -- -- -- -- -- */\
275  "pand "MANGLE(mask0010)", %%mm0 \n" /* -- -- -- -- G1 B1 -- -- */\
276  "psllq $32, %%mm5 \n" /* -- -- -- -- R4 G4 B4 R5 */\
277  "pand "MANGLE(mask1001)", %%mm2 \n" /* G5 B5 -- -- -- -- G7 B7 */\
278  "por %%mm3, %%mm1 \n"\
279  "por %%mm6, %%mm0 \n"\
280  "por %%mm5, %%mm1 \n"\
281  "por %%mm7, %%mm2 \n"\
282  MOVNTQ" %%mm0, (%1) \n"\
283  MOVNTQ" %%mm1, 8(%1) \n"\
284  MOVNTQ" %%mm2, 16(%1) \n"\
285 
286 #else
287 #undef RGB_PACK24_B
288 #define RGB_PACK24_B\
289  "movd %%mm3, (%1) \n" /* R0 G0 B0 R1 */\
290  "movd %%mm2, 4(%1) \n" /* G1 B1 */\
291  "psrlq $32, %%mm3 \n"\
292  "psrlq $16, %%mm2 \n"\
293  "movd %%mm3, 6(%1) \n" /* R2 G2 B2 R3 */\
294  "movd %%mm2, 10(%1) \n" /* G3 B3 */\
295  "psrlq $16, %%mm2 \n"\
296  "movd %%mm5, 12(%1) \n" /* R4 G4 B4 R5 */\
297  "movd %%mm2, 16(%1) \n" /* G5 B5 */\
298  "psrlq $32, %%mm5 \n"\
299  "movd %%mm2, 20(%1) \n" /* -- -- G7 B7 */\
300  "movd %%mm5, 18(%1) \n" /* R6 G6 B6 R7 */\
301 
302 #endif
303 
304 static inline int RENAME(yuv420_rgb24)(SwsContext *c, const uint8_t *src[],
305  int srcStride[],
306  int srcSliceY, int srcSliceH,
307  uint8_t *dst[], int dstStride[])
308 {
309  int y, h_size;
310 
311  YUV2RGB_LOOP(3)
312 
314  YUV2RGB
316 
317  YUV2RGB_ENDLOOP(3)
320 }
321 
322 static inline int RENAME(yuv420_bgr24)(SwsContext *c, const uint8_t *src[],
323  int srcStride[],
324  int srcSliceY, int srcSliceH,
325  uint8_t *dst[], int dstStride[])
326 {
327  int y, h_size;
328 
329  YUV2RGB_LOOP(3)
330 
332  YUV2RGB
334 
335  YUV2RGB_ENDLOOP(3)
338 }
339 
340 
341 #define SET_EMPTY_ALPHA \
342  "pcmpeqd %%mm"REG_ALPHA", %%mm"REG_ALPHA"\n\t" /* set alpha to 0xFF */ \
343 
344 #define LOAD_ALPHA \
345  "movq (%6, %0, 2), %%mm"REG_ALPHA"\n\t" \
346 
347 #define RGB_PACK32(red, green, blue, alpha) \
348  "movq %%mm"blue", %%mm5\n\t" \
349  "movq %%mm"red", %%mm6\n\t" \
350  "punpckhbw %%mm"green", %%mm5\n\t" \
351  "punpcklbw %%mm"green", %%mm"blue"\n\t" \
352  "punpckhbw %%mm"alpha", %%mm6\n\t" \
353  "punpcklbw %%mm"alpha", %%mm"red"\n\t" \
354  "movq %%mm"blue", %%mm"green"\n\t" \
355  "movq %%mm5, %%mm"alpha"\n\t" \
356  "punpcklwd %%mm"red", %%mm"blue"\n\t" \
357  "punpckhwd %%mm"red", %%mm"green"\n\t" \
358  "punpcklwd %%mm6, %%mm5\n\t" \
359  "punpckhwd %%mm6, %%mm"alpha"\n\t" \
360  MOVNTQ " %%mm"blue", 0(%1)\n\t" \
361  MOVNTQ " %%mm"green", 8(%1)\n\t" \
362  MOVNTQ " %%mm5, 16(%1)\n\t" \
363  MOVNTQ " %%mm"alpha", 24(%1)\n\t" \
364 
365 #if !COMPILE_TEMPLATE_MMX2
366 static inline int RENAME(yuv420_rgb32)(SwsContext *c, const uint8_t *src[],
367  int srcStride[],
368  int srcSliceY, int srcSliceH,
369  uint8_t *dst[], int dstStride[])
370 {
371  int y, h_size;
372 
373  YUV2RGB_LOOP(4)
374 
376  YUV2RGB
380 
381  YUV2RGB_ENDLOOP(4)
384 }
385 
386 #if HAVE_7REGS && CONFIG_SWSCALE_ALPHA
387 static inline int RENAME(yuva420_rgb32)(SwsContext *c, const uint8_t *src[],
388  int srcStride[],
389  int srcSliceY, int srcSliceH,
390  uint8_t *dst[], int dstStride[])
391 {
392  int y, h_size;
393 
394  YUV2RGB_LOOP(4)
395 
396  const uint8_t *pa = src[3] + y * srcStride[3];
398  YUV2RGB
400  LOAD_ALPHA
402 
403  YUV2RGB_ENDLOOP(4)
406 }
407 #endif
408 
409 static inline int RENAME(yuv420_bgr32)(SwsContext *c, const uint8_t *src[],
410  int srcStride[],
411  int srcSliceY, int srcSliceH,
412  uint8_t *dst[], int dstStride[])
413 {
414  int y, h_size;
415 
416  YUV2RGB_LOOP(4)
417 
419  YUV2RGB
422  RGB_PACK32(REG_BLUE, REG_GREEN, REG_RED, REG_ALPHA)
423 
424  YUV2RGB_ENDLOOP(4)
427 }
428 
429 #if HAVE_7REGS && CONFIG_SWSCALE_ALPHA
430 static inline int RENAME(yuva420_bgr32)(SwsContext *c, const uint8_t *src[],
431  int srcStride[],
432  int srcSliceY, int srcSliceH,
433  uint8_t *dst[], int dstStride[])
434 {
435  int y, h_size;
436 
437  YUV2RGB_LOOP(4)
438 
439  const uint8_t *pa = src[3] + y * srcStride[3];
441  YUV2RGB
443  LOAD_ALPHA
444  RGB_PACK32(REG_BLUE, REG_GREEN, REG_RED, REG_ALPHA)
445 
446  YUV2RGB_ENDLOOP(4)
449 }
450 #endif
451 
452 #endif /* !COMPILE_TEMPLATE_MMX2 */