2 * yuv2rgb.c, Software YUV to RGB converter
4 * Copyright (C) 1999, Aaron Holtzman <aholtzma@ess.engr.uvic.ca>
6 * Functions broken out from display_x11.c and several new modes
7 * added by HÃ¥kan Hjort <d95hjort@dtek.chalmers.se>
9 * 15 & 16 bpp support by Franck Sicard <Franck.Sicard@solsoft.fr>
11 * MMX/MMX2 template stuff (needed for fast movntq support),
12 * 1,4,8bpp support and context / deglobalize stuff
13 * by Michael Niedermayer (michaelni@gmx.at)
15 * This file is part of mpeg2dec, a free MPEG-2 video decoder
17 * mpeg2dec is free software; you can redistribute it and/or modify
18 * it under the terms of the GNU General Public License as published by
19 * the Free Software Foundation; either version 2, or (at your option)
22 * mpeg2dec is distributed in the hope that it will be useful,
23 * but WITHOUT ANY WARRANTY; without even the implied warranty of
24 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
25 * GNU General Public License for more details.
27 * You should have received a copy of the GNU General Public License
28 * along with mpeg2dec; if not, write to the Free Software
29 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
40 #include "swscale_internal.h"
42 #define DITHER1XBPP // only for MMX
44 const uint8_t __attribute__((aligned(8))) dither_2x2_4[2][8]={
45 { 1, 3, 1, 3, 1, 3, 1, 3, },
46 { 2, 0, 2, 0, 2, 0, 2, 0, },
49 const uint8_t __attribute__((aligned(8))) dither_2x2_8[2][8]={
50 { 6, 2, 6, 2, 6, 2, 6, 2, },
51 { 0, 4, 0, 4, 0, 4, 0, 4, },
54 const uint8_t __attribute__((aligned(8))) dither_8x8_32[8][8]={
55 { 17, 9, 23, 15, 16, 8, 22, 14, },
56 { 5, 29, 3, 27, 4, 28, 2, 26, },
57 { 21, 13, 19, 11, 20, 12, 18, 10, },
58 { 0, 24, 6, 30, 1, 25, 7, 31, },
59 { 16, 8, 22, 14, 17, 9, 23, 15, },
60 { 4, 28, 2, 26, 5, 29, 3, 27, },
61 { 20, 12, 18, 10, 21, 13, 19, 11, },
62 { 1, 25, 7, 31, 0, 24, 6, 30, },
66 const uint8_t __attribute__((aligned(8))) dither_8x8_64[8][8]={
67 { 0, 48, 12, 60, 3, 51, 15, 63, },
68 { 32, 16, 44, 28, 35, 19, 47, 31, },
69 { 8, 56, 4, 52, 11, 59, 7, 55, },
70 { 40, 24, 36, 20, 43, 27, 39, 23, },
71 { 2, 50, 14, 62, 1, 49, 13, 61, },
72 { 34, 18, 46, 30, 33, 17, 45, 29, },
73 { 10, 58, 6, 54, 9, 57, 5, 53, },
74 { 42, 26, 38, 22, 41, 25, 37, 21, },
78 const uint8_t __attribute__((aligned(8))) dither_8x8_73[8][8]={
79 { 0, 55, 14, 68, 3, 58, 17, 72, },
80 { 37, 18, 50, 32, 40, 22, 54, 35, },
81 { 9, 64, 5, 59, 13, 67, 8, 63, },
82 { 46, 27, 41, 23, 49, 31, 44, 26, },
83 { 2, 57, 16, 71, 1, 56, 15, 70, },
84 { 39, 21, 52, 34, 38, 19, 51, 33, },
85 { 11, 66, 7, 62, 10, 65, 6, 60, },
86 { 48, 30, 43, 25, 47, 29, 42, 24, },
90 const uint8_t __attribute__((aligned(8))) dither_8x8_128[8][8]={
91 { 68, 36, 92, 60, 66, 34, 90, 58, },
92 { 20, 116, 12, 108, 18, 114, 10, 106, },
93 { 84, 52, 76, 44, 82, 50, 74, 42, },
94 { 0, 96, 24, 120, 6, 102, 30, 126, },
95 { 64, 32, 88, 56, 70, 38, 94, 62, },
96 { 16, 112, 8, 104, 22, 118, 14, 110, },
97 { 80, 48, 72, 40, 86, 54, 78, 46, },
98 { 4, 100, 28, 124, 2, 98, 26, 122, },
103 const uint8_t __attribute__((aligned(8))) dither_8x8_220[8][8]={
104 {117, 62, 158, 103, 113, 58, 155, 100, },
105 { 34, 199, 21, 186, 31, 196, 17, 182, },
106 {144, 89, 131, 76, 141, 86, 127, 72, },
107 { 0, 165, 41, 206, 10, 175, 52, 217, },
108 {110, 55, 151, 96, 120, 65, 162, 107, },
109 { 28, 193, 14, 179, 38, 203, 24, 189, },
110 {138, 83, 124, 69, 148, 93, 134, 79, },
111 { 7, 172, 48, 213, 3, 168, 45, 210, },
114 // tries to correct a gamma of 1.5
115 const uint8_t __attribute__((aligned(8))) dither_8x8_220[8][8]={
116 { 0, 143, 18, 200, 2, 156, 25, 215, },
117 { 78, 28, 125, 64, 89, 36, 138, 74, },
118 { 10, 180, 3, 161, 16, 195, 8, 175, },
119 {109, 51, 93, 38, 121, 60, 105, 47, },
120 { 1, 152, 23, 210, 0, 147, 20, 205, },
121 { 85, 33, 134, 71, 81, 30, 130, 67, },
122 { 14, 190, 6, 171, 12, 185, 5, 166, },
123 {117, 57, 101, 44, 113, 54, 97, 41, },
126 // tries to correct a gamma of 2.0
127 const uint8_t __attribute__((aligned(8))) dither_8x8_220[8][8]={
128 { 0, 124, 8, 193, 0, 140, 12, 213, },
129 { 55, 14, 104, 42, 66, 19, 119, 52, },
130 { 3, 168, 1, 145, 6, 187, 3, 162, },
131 { 86, 31, 70, 21, 99, 39, 82, 28, },
132 { 0, 134, 11, 206, 0, 129, 9, 200, },
133 { 62, 17, 114, 48, 58, 16, 109, 45, },
134 { 5, 181, 2, 157, 4, 175, 1, 151, },
135 { 95, 36, 78, 26, 90, 34, 74, 24, },
138 // tries to correct a gamma of 2.5
139 const uint8_t __attribute__((aligned(8))) dither_8x8_220[8][8]={
140 { 0, 107, 3, 187, 0, 125, 6, 212, },
141 { 39, 7, 86, 28, 49, 11, 102, 36, },
142 { 1, 158, 0, 131, 3, 180, 1, 151, },
143 { 68, 19, 52, 12, 81, 25, 64, 17, },
144 { 0, 119, 5, 203, 0, 113, 4, 195, },
145 { 45, 9, 96, 33, 42, 8, 91, 30, },
146 { 2, 172, 1, 144, 2, 165, 0, 137, },
147 { 77, 23, 60, 15, 72, 21, 56, 14, },
153 /* hope these constant values are cache line aligned */
154 DECLARE_ASM_CONST(8, uint64_t, mmx_00ffw) = 0x00ff00ff00ff00ffULL;
155 DECLARE_ASM_CONST(8, uint64_t, mmx_redmask) = 0xf8f8f8f8f8f8f8f8ULL;
156 DECLARE_ASM_CONST(8, uint64_t, mmx_grnmask) = 0xfcfcfcfcfcfcfcfcULL;
158 // The volatile is required because gcc otherwise optimizes some writes away
159 // not knowing that these are read in the ASM block.
160 static volatile uint64_t attribute_used __attribute__((aligned(8))) b5Dither;
161 static volatile uint64_t attribute_used __attribute__((aligned(8))) g5Dither;
162 static volatile uint64_t attribute_used __attribute__((aligned(8))) g6Dither;
163 static volatile uint64_t attribute_used __attribute__((aligned(8))) r5Dither;
172 #define RENAME(a) a ## _MMX
173 #include "yuv2rgb_template.c"
180 #define RENAME(a) a ## _MMX2
181 #include "yuv2rgb_template.c"
183 #endif /* HAVE_MMX */
185 const int32_t Inverse_Table_6_9[8][4] = {
186 {117504, 138453, 13954, 34903}, /* no sequence_display_extension */
187 {117504, 138453, 13954, 34903}, /* ITU-R Rec. 709 (1990) */
188 {104597, 132201, 25675, 53279}, /* unspecified */
189 {104597, 132201, 25675, 53279}, /* reserved */
190 {104448, 132798, 24759, 53109}, /* FCC */
191 {104597, 132201, 25675, 53279}, /* ITU-R Rec. 624-4 System B, G */
192 {104597, 132201, 25675, 53279}, /* SMPTE 170M */
193 {117579, 136230, 16907, 35559} /* SMPTE 240M (1987) */
199 r = (void *)c->table_rV[V]; \
200 g = (void *)(c->table_gU[U] + c->table_gV[V]); \
201 b = (void *)c->table_bU[U];
205 dst_1[2*i] = r[Y] + g[Y] + b[Y]; \
207 dst_1[2*i+1] = r[Y] + g[Y] + b[Y];
211 dst_2[2*i] = r[Y] + g[Y] + b[Y]; \
213 dst_2[2*i+1] = r[Y] + g[Y] + b[Y];
217 dst_1[6*i] = r[Y]; dst_1[6*i+1] = g[Y]; dst_1[6*i+2] = b[Y]; \
219 dst_1[6*i+3] = r[Y]; dst_1[6*i+4] = g[Y]; dst_1[6*i+5] = b[Y];
223 dst_2[6*i] = r[Y]; dst_2[6*i+1] = g[Y]; dst_2[6*i+2] = b[Y]; \
225 dst_2[6*i+3] = r[Y]; dst_2[6*i+4] = g[Y]; dst_2[6*i+5] = b[Y];
229 dst_1[6*i] = b[Y]; dst_1[6*i+1] = g[Y]; dst_1[6*i+2] = r[Y]; \
231 dst_1[6*i+3] = b[Y]; dst_1[6*i+4] = g[Y]; dst_1[6*i+5] = r[Y];
235 dst_2[6*i] = b[Y]; dst_2[6*i+1] = g[Y]; dst_2[6*i+2] = r[Y]; \
237 dst_2[6*i+3] = b[Y]; dst_2[6*i+4] = g[Y]; dst_2[6*i+5] = r[Y];
239 #define PROLOG(func_name, dst_type) \
240 static int func_name(SwsContext *c, uint8_t* src[], int srcStride[], int srcSliceY, \
241 int srcSliceH, uint8_t* dst[], int dstStride[]){\
244 if (c->srcFormat == PIX_FMT_YUV422P){\
248 for (y=0; y<srcSliceH; y+=2){\
249 dst_type *dst_1= (dst_type*)(dst[0] + (y+srcSliceY )*dstStride[0]);\
250 dst_type *dst_2= (dst_type*)(dst[0] + (y+srcSliceY+1)*dstStride[0]);\
251 dst_type av_unused *r, *b;\
253 uint8_t *py_1= src[0] + y*srcStride[0];\
254 uint8_t *py_2= py_1 + srcStride[0];\
255 uint8_t *pu= src[1] + (y>>1)*srcStride[1];\
256 uint8_t *pv= src[2] + (y>>1)*srcStride[2];\
257 unsigned int h_size= c->dstW>>3;\
262 #define EPILOG1(dst_delta)\
271 int av_unused Y, U, V;\
279 #define EPILOG(dst_delta)\
283 PROLOG(yuv2rgb_c_32, uint32_t)
309 PROLOG(yuv2rgb_c_24_rgb, uint8_t)
335 // only trivial mods from yuv2rgb_c_24_rgb
336 PROLOG(yuv2rgb_c_24_bgr, uint8_t)
362 // This is exactly the same code as yuv2rgb_c_32 except for the types of
363 // r, g, b, dst_1, dst_2
364 PROLOG(yuv2rgb_c_16, uint16_t)
382 // This is exactly the same code as yuv2rgb_c_32 except for the types of
383 // r, g, b, dst_1, dst_2
384 PROLOG(yuv2rgb_c_8, uint8_t)
402 // r, g, b, dst_1, dst_2
403 PROLOG(yuv2rgb_c_8_ordered_dither, uint8_t)
404 const uint8_t *d32= dither_8x8_32[y&7];
405 const uint8_t *d64= dither_8x8_73[y&7];
406 #define DST1bpp8(i,o) \
408 dst_1[2*i] = r[Y+d32[0+o]] + g[Y+d32[0+o]] + b[Y+d64[0+o]]; \
410 dst_1[2*i+1] = r[Y+d32[1+o]] + g[Y+d32[1+o]] + b[Y+d64[1+o]];
412 #define DST2bpp8(i,o) \
414 dst_2[2*i] = r[Y+d32[8+o]] + g[Y+d32[8+o]] + b[Y+d64[8+o]]; \
416 dst_2[2*i+1] = r[Y+d32[9+o]] + g[Y+d32[9+o]] + b[Y+d64[9+o]];
437 // This is exactly the same code as yuv2rgb_c_32 except for the types of
438 // r, g, b, dst_1, dst_2
439 PROLOG(yuv2rgb_c_4, uint8_t)
443 acc = r[Y] + g[Y] + b[Y]; \
445 acc |= (r[Y] + g[Y] + b[Y])<<4; \
450 acc = r[Y] + g[Y] + b[Y]; \
452 acc |= (r[Y] + g[Y] + b[Y])<<4; \
472 PROLOG(yuv2rgb_c_4_ordered_dither, uint8_t)
473 const uint8_t *d64= dither_8x8_73[y&7];
474 const uint8_t *d128=dither_8x8_220[y&7];
477 #define DST1bpp4(i,o) \
479 acc = r[Y+d128[0+o]] + g[Y+d64[0+o]] + b[Y+d128[0+o]]; \
481 acc |= (r[Y+d128[1+o]] + g[Y+d64[1+o]] + b[Y+d128[1+o]])<<4; \
484 #define DST2bpp4(i,o) \
486 acc = r[Y+d128[8+o]] + g[Y+d64[8+o]] + b[Y+d128[8+o]]; \
488 acc |= (r[Y+d128[9+o]] + g[Y+d64[9+o]] + b[Y+d128[9+o]])<<4; \
509 // This is exactly the same code as yuv2rgb_c_32 except for the types of
510 // r, g, b, dst_1, dst_2
511 PROLOG(yuv2rgb_c_4b, uint8_t)
529 PROLOG(yuv2rgb_c_4b_ordered_dither, uint8_t)
530 const uint8_t *d64= dither_8x8_73[y&7];
531 const uint8_t *d128=dither_8x8_220[y&7];
533 #define DST1bpp4b(i,o) \
535 dst_1[2*i] = r[Y+d128[0+o]] + g[Y+d64[0+o]] + b[Y+d128[0+o]]; \
537 dst_1[2*i+1] = r[Y+d128[1+o]] + g[Y+d64[1+o]] + b[Y+d128[1+o]];
539 #define DST2bpp4b(i,o) \
541 dst_2[2*i] = r[Y+d128[8+o]] + g[Y+d64[8+o]] + b[Y+d128[8+o]]; \
543 dst_2[2*i+1] = r[Y+d128[9+o]] + g[Y+d64[9+o]] + b[Y+d128[9+o]];
563 PROLOG(yuv2rgb_c_1_ordered_dither, uint8_t)
564 const uint8_t *d128=dither_8x8_220[y&7];
565 char out_1=0, out_2=0;
566 g= c->table_gU[128] + c->table_gV[128];
568 #define DST1bpp1(i,o) \
570 out_1+= out_1 + g[Y+d128[0+o]]; \
572 out_1+= out_1 + g[Y+d128[1+o]];
574 #define DST2bpp1(i,o) \
576 out_2+= out_2 + g[Y+d128[8+o]]; \
578 out_2+= out_2 + g[Y+d128[9+o]];
596 SwsFunc yuv2rgb_get_func_ptr (SwsContext *c)
598 #if defined(HAVE_MMX2) || defined(HAVE_MMX)
599 if (c->flags & SWS_CPU_CAPS_MMX2){
600 switch(c->dstFormat){
601 case PIX_FMT_RGB32: return yuv420_rgb32_MMX2;
602 case PIX_FMT_BGR24: return yuv420_rgb24_MMX2;
603 case PIX_FMT_BGR565: return yuv420_rgb16_MMX2;
604 case PIX_FMT_BGR555: return yuv420_rgb15_MMX2;
607 if (c->flags & SWS_CPU_CAPS_MMX){
608 switch(c->dstFormat){
609 case PIX_FMT_RGB32: return yuv420_rgb32_MMX;
610 case PIX_FMT_BGR24: return yuv420_rgb24_MMX;
611 case PIX_FMT_BGR565: return yuv420_rgb16_MMX;
612 case PIX_FMT_BGR555: return yuv420_rgb15_MMX;
618 SwsFunc t= yuv2rgb_init_vis(c);
624 SwsFunc t= yuv2rgb_init_mlib(c);
629 if (c->flags & SWS_CPU_CAPS_ALTIVEC)
631 SwsFunc t = yuv2rgb_init_altivec(c);
637 if (c->flags & SWS_CPU_CAPS_BFIN)
639 SwsFunc t = ff_bfin_yuv2rgb_get_func_ptr (c);
644 av_log(c, AV_LOG_WARNING, "No accelerated colorspace conversion found.\n");
646 switch(c->dstFormat){
648 case PIX_FMT_RGB32: return yuv2rgb_c_32;
649 case PIX_FMT_RGB24: return yuv2rgb_c_24_rgb;
650 case PIX_FMT_BGR24: return yuv2rgb_c_24_bgr;
654 case PIX_FMT_BGR555: return yuv2rgb_c_16;
656 case PIX_FMT_BGR8: return yuv2rgb_c_8_ordered_dither;
658 case PIX_FMT_BGR4: return yuv2rgb_c_4_ordered_dither;
659 case PIX_FMT_RGB4_BYTE:
660 case PIX_FMT_BGR4_BYTE: return yuv2rgb_c_4b_ordered_dither;
661 case PIX_FMT_MONOBLACK: return yuv2rgb_c_1_ordered_dither;
668 static int div_round (int dividend, int divisor)
671 return (dividend + (divisor>>1)) / divisor;
673 return -((-dividend + (divisor>>1)) / divisor);
676 int yuv2rgb_c_init_tables (SwsContext *c, const int inv_table[4], int fullRange, int brightness, int contrast, int saturation)
678 const int isRgb = isBGR(c->dstFormat);
679 const int bpp = fmt_depth(c->dstFormat);
681 uint8_t table_Y[1024];
682 uint32_t *table_32 = 0;
683 uint16_t *table_16 = 0;
684 uint8_t *table_8 = 0;
685 uint8_t *table_332 = 0;
686 uint8_t *table_121 = 0;
687 uint8_t *table_1 = 0;
689 void *table_r = 0, *table_g = 0, *table_b = 0;
692 int64_t crv = inv_table[0];
693 int64_t cbu = inv_table[1];
694 int64_t cgu = -inv_table[2];
695 int64_t cgv = -inv_table[3];
699 //printf("%lld %lld %lld %lld %lld\n", cy, crv, cbu, cgu, cgv);
704 crv= (crv*224) / 255;
705 cbu= (cbu*224) / 255;
706 cgu= (cgu*224) / 255;
707 cgv= (cgv*224) / 255;
710 cy = (cy *contrast )>>16;
711 crv= (crv*contrast * saturation)>>32;
712 cbu= (cbu*contrast * saturation)>>32;
713 cgu= (cgu*contrast * saturation)>>32;
714 cgv= (cgv*contrast * saturation)>>32;
715 //printf("%lld %lld %lld %lld %lld\n", cy, crv, cbu, cgu, cgv);
716 oy -= 256*brightness;
718 for (i = 0; i < 1024; i++) {
721 j= (cy*(((i - 384)<<16) - oy) + (1<<31))>>32;
722 j = (j < 0) ? 0 : ((j > 255) ? 255 : j);
728 table_start= table_32 = av_malloc ((197 + 2*682 + 256 + 132) * sizeof (uint32_t));
730 entry_size = sizeof (uint32_t);
731 table_r = table_32 + 197;
732 table_b = table_32 + 197 + 685;
733 table_g = table_32 + 197 + 2*682;
735 for (i = -197; i < 256+197; i++)
736 ((uint32_t *)table_r)[i] = table_Y[i+384] << (isRgb ? 16 : 0);
737 for (i = -132; i < 256+132; i++)
738 ((uint32_t *)table_g)[i] = table_Y[i+384] << 8;
739 for (i = -232; i < 256+232; i++)
740 ((uint32_t *)table_b)[i] = table_Y[i+384] << (isRgb ? 0 : 16);
744 table_start= table_8 = av_malloc ((256 + 2*232) * sizeof (uint8_t));
746 entry_size = sizeof (uint8_t);
747 table_r = table_g = table_b = table_8 + 232;
749 for (i = -232; i < 256+232; i++)
750 ((uint8_t * )table_b)[i] = table_Y[i+384];
755 table_start= table_16 = av_malloc ((197 + 2*682 + 256 + 132) * sizeof (uint16_t));
757 entry_size = sizeof (uint16_t);
758 table_r = table_16 + 197;
759 table_b = table_16 + 197 + 685;
760 table_g = table_16 + 197 + 2*682;
762 for (i = -197; i < 256+197; i++) {
763 int j = table_Y[i+384] >> 3;
766 j <<= ((bpp==16) ? 11 : 10);
768 ((uint16_t *)table_r)[i] = j;
770 for (i = -132; i < 256+132; i++) {
771 int j = table_Y[i+384] >> ((bpp==16) ? 2 : 3);
773 ((uint16_t *)table_g)[i] = j << 5;
775 for (i = -232; i < 256+232; i++) {
776 int j = table_Y[i+384] >> 3;
779 j <<= ((bpp==16) ? 11 : 10);
781 ((uint16_t *)table_b)[i] = j;
786 table_start= table_332 = av_malloc ((197 + 2*682 + 256 + 132) * sizeof (uint8_t));
788 entry_size = sizeof (uint8_t);
789 table_r = table_332 + 197;
790 table_b = table_332 + 197 + 685;
791 table_g = table_332 + 197 + 2*682;
793 for (i = -197; i < 256+197; i++) {
794 int j = (table_Y[i+384 - 16] + 18)/36;
799 ((uint8_t *)table_r)[i] = j;
801 for (i = -132; i < 256+132; i++) {
802 int j = (table_Y[i+384 - 16] + 18)/36;
807 ((uint8_t *)table_g)[i] = j << 2;
809 for (i = -232; i < 256+232; i++) {
810 int j = (table_Y[i+384 - 37] + 43)/85;
815 ((uint8_t *)table_b)[i] = j;
820 table_start= table_121 = av_malloc ((197 + 2*682 + 256 + 132) * sizeof (uint8_t));
822 entry_size = sizeof (uint8_t);
823 table_r = table_121 + 197;
824 table_b = table_121 + 197 + 685;
825 table_g = table_121 + 197 + 2*682;
827 for (i = -197; i < 256+197; i++) {
828 int j = table_Y[i+384 - 110] >> 7;
833 ((uint8_t *)table_r)[i] = j;
835 for (i = -132; i < 256+132; i++) {
836 int j = (table_Y[i+384 - 37]+ 43)/85;
838 ((uint8_t *)table_g)[i] = j << 1;
840 for (i = -232; i < 256+232; i++) {
841 int j =table_Y[i+384 - 110] >> 7;
846 ((uint8_t *)table_b)[i] = j;
851 table_start= table_1 = av_malloc (256*2 * sizeof (uint8_t));
853 entry_size = sizeof (uint8_t);
855 table_r = table_b = NULL;
857 for (i = 0; i < 256+256; i++) {
858 int j = table_Y[i + 384 - 110]>>7;
860 ((uint8_t *)table_g)[i] = j;
866 av_log(c, AV_LOG_ERROR, "%ibpp not supported by yuv2rgb\n", bpp);
871 for (i = 0; i < 256; i++) {
872 c->table_rV[i] = (uint8_t *)table_r + entry_size * div_round (crv * (i-128), 76309);
873 c->table_gU[i] = (uint8_t *)table_g + entry_size * div_round (cgu * (i-128), 76309);
874 c->table_gV[i] = entry_size * div_round (cgv * (i-128), 76309);
875 c->table_bU[i] = (uint8_t *)table_b + entry_size * div_round (cbu * (i-128), 76309);
878 av_free(c->yuvTable);
879 c->yuvTable= table_start;