better dithering
[ffmpeg.git] / postproc / swscale_template.c
1
2 // Software scaling and colorspace conversion routines for MPlayer
3
4 // Orginal C implementation by A'rpi/ESP-team <arpi@thot.banki.hu>
5 // current version mostly by Michael Niedermayer (michaelni@gmx.at)
6 // the parts written by michael are under GNU GPL
7
8 #include <inttypes.h>
9 #include <string.h>
10 #include "../config.h"
11 #include "swscale.h"
12 #include "../mmx_defs.h"
13 #undef MOVNTQ
14 #undef PAVGB
15
16 //#undef HAVE_MMX2
17 //#undef HAVE_MMX
18 //#undef ARCH_X86
19 #define DITHER1XBPP
20 int fullUVIpol=0;
21 //disables the unscaled height version
22 int allwaysIpol=0;
23
24 #define RET 0xC3 //near return opcode
25 /*
26 NOTES
27
28 known BUGS with known cause (no bugreports please!, but patches are welcome :) )
29 horizontal MMX2 scaler reads 1-7 samples too much (might cause a sig11)
30
31 Supported output formats BGR15 BGR16 BGR24 BGR32
32 BGR15 & BGR16 MMX verions support dithering
33 Special versions: fast Y 1:1 scaling (no interpolation in y direction)
34
35 TODO
36 more intelligent missalignment avoidance for the horizontal scaler
37 bicubic scaler
38 dither in C
39 change the distance of the u & v buffer
40 */
41
42 #define ABS(a) ((a) > 0 ? (a) : (-(a)))
43 #define MIN(a,b) ((a) > (b) ? (b) : (a))
44 #define MAX(a,b) ((a) < (b) ? (b) : (a))
45
46 #ifdef HAVE_MMX2
47 #define PAVGB(a,b) "pavgb " #a ", " #b " \n\t"
48 #elif defined (HAVE_3DNOW)
49 #define PAVGB(a,b) "pavgusb " #a ", " #b " \n\t"
50 #endif
51
52 #ifdef HAVE_MMX2
53 #define MOVNTQ(a,b) "movntq " #a ", " #b " \n\t"
54 #else
55 #define MOVNTQ(a,b) "movq " #a ", " #b " \n\t"
56 #endif
57
58
59 #ifdef HAVE_MMX
60 static uint64_t __attribute__((aligned(8))) yCoeff=    0x2568256825682568LL;
61 static uint64_t __attribute__((aligned(8))) vrCoeff=   0x3343334333433343LL;
62 static uint64_t __attribute__((aligned(8))) ubCoeff=   0x40cf40cf40cf40cfLL;
63 static uint64_t __attribute__((aligned(8))) vgCoeff=   0xE5E2E5E2E5E2E5E2LL;
64 static uint64_t __attribute__((aligned(8))) ugCoeff=   0xF36EF36EF36EF36ELL;
65 static uint64_t __attribute__((aligned(8))) bF8=       0xF8F8F8F8F8F8F8F8LL;
66 static uint64_t __attribute__((aligned(8))) bFC=       0xFCFCFCFCFCFCFCFCLL;
67 static uint64_t __attribute__((aligned(8))) w400=      0x0400040004000400LL;
68 static uint64_t __attribute__((aligned(8))) w80=       0x0080008000800080LL;
69 static uint64_t __attribute__((aligned(8))) w10=       0x0010001000100010LL;
70 static uint64_t __attribute__((aligned(8))) bm00001111=0x00000000FFFFFFFFLL;
71 static uint64_t __attribute__((aligned(8))) bm00000111=0x0000000000FFFFFFLL;
72 static uint64_t __attribute__((aligned(8))) bm11111000=0xFFFFFFFFFF000000LL;
73
74 static uint64_t __attribute__((aligned(8))) b5Dither;
75 static uint64_t __attribute__((aligned(8))) g5Dither;
76 static uint64_t __attribute__((aligned(8))) g6Dither;
77 static uint64_t __attribute__((aligned(8))) r5Dither;
78
79 static uint64_t __attribute__((aligned(8))) dither4[2]={
80         0x0103010301030103LL,
81         0x0200020002000200LL,};
82
83 static uint64_t __attribute__((aligned(8))) dither8[2]={
84         0x0602060206020602LL,
85         0x0004000400040004LL,};
86
87 static uint64_t __attribute__((aligned(8))) b16Mask=   0x001F001F001F001FLL;
88 static uint64_t __attribute__((aligned(8))) g16Mask=   0x07E007E007E007E0LL;
89 static uint64_t __attribute__((aligned(8))) r16Mask=   0xF800F800F800F800LL;
90 static uint64_t __attribute__((aligned(8))) b15Mask=   0x001F001F001F001FLL;
91 static uint64_t __attribute__((aligned(8))) g15Mask=   0x03E003E003E003E0LL;
92 static uint64_t __attribute__((aligned(8))) r15Mask=   0x7C007C007C007C00LL;
93
94 static uint64_t __attribute__((aligned(8))) M24A=   0x00FF0000FF0000FFLL;
95 static uint64_t __attribute__((aligned(8))) M24B=   0xFF0000FF0000FF00LL;
96 static uint64_t __attribute__((aligned(8))) M24C=   0x0000FF0000FF0000LL;
97
98 static uint64_t __attribute__((aligned(8))) temp0;
99 static uint64_t __attribute__((aligned(8))) asm_yalpha1;
100 static uint64_t __attribute__((aligned(8))) asm_uvalpha1;
101 #endif
102
103 // temporary storage for 4 yuv lines:
104 // 16bit for now (mmx likes it more compact)
105 #ifdef HAVE_MMX
106 static uint16_t __attribute__((aligned(8))) pix_buf_y[4][2048];
107 static uint16_t __attribute__((aligned(8))) pix_buf_uv[2][2048*2];
108 #else
109 static uint16_t pix_buf_y[4][2048];
110 static uint16_t pix_buf_uv[2][2048*2];
111 #endif
112
113 // clipping helper table for C implementations:
114 static unsigned char clip_table[768];
115
116 static unsigned short clip_table16b[768];
117 static unsigned short clip_table16g[768];
118 static unsigned short clip_table16r[768];
119 static unsigned short clip_table15b[768];
120 static unsigned short clip_table15g[768];
121 static unsigned short clip_table15r[768];
122
123 // yuv->rgb conversion tables:
124 static    int yuvtab_2568[256];
125 static    int yuvtab_3343[256];
126 static    int yuvtab_0c92[256];
127 static    int yuvtab_1a1e[256];
128 static    int yuvtab_40cf[256];
129
130 #ifdef HAVE_MMX2
131 static uint8_t funnyYCode[10000];
132 static uint8_t funnyUVCode[10000];
133 #endif
134
135 static int canMMX2BeUsed=0;
136
137 #define FULL_YSCALEYUV2RGB \
138                 "pxor %%mm7, %%mm7              \n\t"\
139                 "movd %6, %%mm6                 \n\t" /*yalpha1*/\
140                 "punpcklwd %%mm6, %%mm6         \n\t"\
141                 "punpcklwd %%mm6, %%mm6         \n\t"\
142                 "movd %7, %%mm5                 \n\t" /*uvalpha1*/\
143                 "punpcklwd %%mm5, %%mm5         \n\t"\
144                 "punpcklwd %%mm5, %%mm5         \n\t"\
145                 "xorl %%eax, %%eax              \n\t"\
146                 "1:                             \n\t"\
147                 "movq (%0, %%eax, 2), %%mm0     \n\t" /*buf0[eax]*/\
148                 "movq (%1, %%eax, 2), %%mm1     \n\t" /*buf1[eax]*/\
149                 "movq (%2, %%eax,2), %%mm2      \n\t" /* uvbuf0[eax]*/\
150                 "movq (%3, %%eax,2), %%mm3      \n\t" /* uvbuf1[eax]*/\
151                 "psubw %%mm1, %%mm0             \n\t" /* buf0[eax] - buf1[eax]*/\
152                 "psubw %%mm3, %%mm2             \n\t" /* uvbuf0[eax] - uvbuf1[eax]*/\
153                 "pmulhw %%mm6, %%mm0            \n\t" /* (buf0[eax] - buf1[eax])yalpha1>>16*/\
154                 "pmulhw %%mm5, %%mm2            \n\t" /* (uvbuf0[eax] - uvbuf1[eax])uvalpha1>>16*/\
155                 "psraw $4, %%mm1                \n\t" /* buf0[eax] - buf1[eax] >>4*/\
156                 "movq 4096(%2, %%eax,2), %%mm4  \n\t" /* uvbuf0[eax+2048]*/\
157                 "psraw $4, %%mm3                \n\t" /* uvbuf0[eax] - uvbuf1[eax] >>4*/\
158                 "paddw %%mm0, %%mm1             \n\t" /* buf0[eax]yalpha1 + buf1[eax](1-yalpha1) >>16*/\
159                 "movq 4096(%3, %%eax,2), %%mm0  \n\t" /* uvbuf1[eax+2048]*/\
160                 "paddw %%mm2, %%mm3             \n\t" /* uvbuf0[eax]uvalpha1 - uvbuf1[eax](1-uvalpha1)*/\
161                 "psubw %%mm0, %%mm4             \n\t" /* uvbuf0[eax+2048] - uvbuf1[eax+2048]*/\
162                 "psubw w80, %%mm1               \n\t" /* 8(Y-16)*/\
163                 "psubw w400, %%mm3              \n\t" /* 8(U-128)*/\
164                 "pmulhw yCoeff, %%mm1           \n\t"\
165 \
166 \
167                 "pmulhw %%mm5, %%mm4            \n\t" /* (uvbuf0[eax+2048] - uvbuf1[eax+2048])uvalpha1>>16*/\
168                 "movq %%mm3, %%mm2              \n\t" /* (U-128)8*/\
169                 "pmulhw ubCoeff, %%mm3          \n\t"\
170                 "psraw $4, %%mm0                \n\t" /* uvbuf0[eax+2048] - uvbuf1[eax+2048] >>4*/\
171                 "pmulhw ugCoeff, %%mm2          \n\t"\
172                 "paddw %%mm4, %%mm0             \n\t" /* uvbuf0[eax+2048]uvalpha1 - uvbuf1[eax+2048](1-uvalpha1)*/\
173                 "psubw w400, %%mm0              \n\t" /* (V-128)8*/\
174 \
175 \
176                 "movq %%mm0, %%mm4              \n\t" /* (V-128)8*/\
177                 "pmulhw vrCoeff, %%mm0          \n\t"\
178                 "pmulhw vgCoeff, %%mm4          \n\t"\
179                 "paddw %%mm1, %%mm3             \n\t" /* B*/\
180                 "paddw %%mm1, %%mm0             \n\t" /* R*/\
181                 "packuswb %%mm3, %%mm3          \n\t"\
182 \
183                 "packuswb %%mm0, %%mm0          \n\t"\
184                 "paddw %%mm4, %%mm2             \n\t"\
185                 "paddw %%mm2, %%mm1             \n\t" /* G*/\
186 \
187                 "packuswb %%mm1, %%mm1          \n\t"
188
189 #define YSCALEYUV2RGB \
190                 "movd %6, %%mm6                 \n\t" /*yalpha1*/\
191                 "punpcklwd %%mm6, %%mm6         \n\t"\
192                 "punpcklwd %%mm6, %%mm6         \n\t"\
193                 "movq %%mm6, asm_yalpha1        \n\t"\
194                 "movd %7, %%mm5                 \n\t" /*uvalpha1*/\
195                 "punpcklwd %%mm5, %%mm5         \n\t"\
196                 "punpcklwd %%mm5, %%mm5         \n\t"\
197                 "movq %%mm5, asm_uvalpha1       \n\t"\
198                 "xorl %%eax, %%eax              \n\t"\
199                 "1:                             \n\t"\
200                 "movq (%2, %%eax), %%mm2        \n\t" /* uvbuf0[eax]*/\
201                 "movq (%3, %%eax), %%mm3        \n\t" /* uvbuf1[eax]*/\
202                 "movq 4096(%2, %%eax), %%mm5    \n\t" /* uvbuf0[eax+2048]*/\
203                 "movq 4096(%3, %%eax), %%mm4    \n\t" /* uvbuf1[eax+2048]*/\
204                 "psubw %%mm3, %%mm2             \n\t" /* uvbuf0[eax] - uvbuf1[eax]*/\
205                 "psubw %%mm4, %%mm5             \n\t" /* uvbuf0[eax+2048] - uvbuf1[eax+2048]*/\
206                 "movq asm_uvalpha1, %%mm0       \n\t"\
207                 "pmulhw %%mm0, %%mm2            \n\t" /* (uvbuf0[eax] - uvbuf1[eax])uvalpha1>>16*/\
208                 "pmulhw %%mm0, %%mm5            \n\t" /* (uvbuf0[eax+2048] - uvbuf1[eax+2048])uvalpha1>>16*/\
209                 "psraw $4, %%mm3                \n\t" /* uvbuf0[eax] - uvbuf1[eax] >>4*/\
210                 "psraw $4, %%mm4                \n\t" /* uvbuf0[eax+2048] - uvbuf1[eax+2048] >>4*/\
211                 "paddw %%mm2, %%mm3             \n\t" /* uvbuf0[eax]uvalpha1 - uvbuf1[eax](1-uvalpha1)*/\
212                 "paddw %%mm5, %%mm4             \n\t" /* uvbuf0[eax+2048]uvalpha1 - uvbuf1[eax+2048](1-uvalpha1)*/\
213                 "psubw w400, %%mm3              \n\t" /* (U-128)8*/\
214                 "psubw w400, %%mm4              \n\t" /* (V-128)8*/\
215                 "movq %%mm3, %%mm2              \n\t" /* (U-128)8*/\
216                 "movq %%mm4, %%mm5              \n\t" /* (V-128)8*/\
217                 "pmulhw ugCoeff, %%mm3          \n\t"\
218                 "pmulhw vgCoeff, %%mm4          \n\t"\
219         /* mm2=(U-128)8, mm3=ug, mm4=vg mm5=(V-128)8 */\
220                 "movq (%0, %%eax, 2), %%mm0     \n\t" /*buf0[eax]*/\
221                 "movq (%1, %%eax, 2), %%mm1     \n\t" /*buf1[eax]*/\
222                 "movq 8(%0, %%eax, 2), %%mm6    \n\t" /*buf0[eax]*/\
223                 "movq 8(%1, %%eax, 2), %%mm7    \n\t" /*buf1[eax]*/\
224                 "psubw %%mm1, %%mm0             \n\t" /* buf0[eax] - buf1[eax]*/\
225                 "psubw %%mm7, %%mm6             \n\t" /* buf0[eax] - buf1[eax]*/\
226                 "pmulhw asm_yalpha1, %%mm0      \n\t" /* (buf0[eax] - buf1[eax])yalpha1>>16*/\
227                 "pmulhw asm_yalpha1, %%mm6      \n\t" /* (buf0[eax] - buf1[eax])yalpha1>>16*/\
228                 "psraw $4, %%mm1                \n\t" /* buf0[eax] - buf1[eax] >>4*/\
229                 "psraw $4, %%mm7                \n\t" /* buf0[eax] - buf1[eax] >>4*/\
230                 "paddw %%mm0, %%mm1             \n\t" /* buf0[eax]yalpha1 + buf1[eax](1-yalpha1) >>16*/\
231                 "paddw %%mm6, %%mm7             \n\t" /* buf0[eax]yalpha1 + buf1[eax](1-yalpha1) >>16*/\
232                 "pmulhw ubCoeff, %%mm2          \n\t"\
233                 "pmulhw vrCoeff, %%mm5          \n\t"\
234                 "psubw w80, %%mm1               \n\t" /* 8(Y-16)*/\
235                 "psubw w80, %%mm7               \n\t" /* 8(Y-16)*/\
236                 "pmulhw yCoeff, %%mm1           \n\t"\
237                 "pmulhw yCoeff, %%mm7           \n\t"\
238         /* mm1= Y1, mm2=ub, mm3=ug, mm4=vg mm5=vr, mm7=Y2 */\
239                 "paddw %%mm3, %%mm4             \n\t"\
240                 "movq %%mm2, %%mm0              \n\t"\
241                 "movq %%mm5, %%mm6              \n\t"\
242                 "movq %%mm4, %%mm3              \n\t"\
243                 "punpcklwd %%mm2, %%mm2         \n\t"\
244                 "punpcklwd %%mm5, %%mm5         \n\t"\
245                 "punpcklwd %%mm4, %%mm4         \n\t"\
246                 "paddw %%mm1, %%mm2             \n\t"\
247                 "paddw %%mm1, %%mm5             \n\t"\
248                 "paddw %%mm1, %%mm4             \n\t"\
249                 "punpckhwd %%mm0, %%mm0         \n\t"\
250                 "punpckhwd %%mm6, %%mm6         \n\t"\
251                 "punpckhwd %%mm3, %%mm3         \n\t"\
252                 "paddw %%mm7, %%mm0             \n\t"\
253                 "paddw %%mm7, %%mm6             \n\t"\
254                 "paddw %%mm7, %%mm3             \n\t"\
255                 /* mm0=B1, mm2=B2, mm3=G2, mm4=G1, mm5=R1, mm6=R2 */\
256                 "packuswb %%mm0, %%mm2          \n\t"\
257                 "packuswb %%mm6, %%mm5          \n\t"\
258                 "packuswb %%mm3, %%mm4          \n\t"\
259                 "pxor %%mm7, %%mm7              \n\t"
260
261 #define YSCALEYUV2RGB1 \
262                 "xorl %%eax, %%eax              \n\t"\
263                 "1:                             \n\t"\
264                 "movq (%2, %%eax), %%mm3        \n\t" /* uvbuf0[eax]*/\
265                 "movq 4096(%2, %%eax), %%mm4    \n\t" /* uvbuf0[eax+2048]*/\
266                 "psraw $4, %%mm3                \n\t" /* uvbuf0[eax] - uvbuf1[eax] >>4*/\
267                 "psraw $4, %%mm4                \n\t" /* uvbuf0[eax+2048] - uvbuf1[eax+2048] >>4*/\
268                 "psubw w400, %%mm3              \n\t" /* (U-128)8*/\
269                 "psubw w400, %%mm4              \n\t" /* (V-128)8*/\
270                 "movq %%mm3, %%mm2              \n\t" /* (U-128)8*/\
271                 "movq %%mm4, %%mm5              \n\t" /* (V-128)8*/\
272                 "pmulhw ugCoeff, %%mm3          \n\t"\
273                 "pmulhw vgCoeff, %%mm4          \n\t"\
274         /* mm2=(U-128)8, mm3=ug, mm4=vg mm5=(V-128)8 */\
275                 "movq (%0, %%eax, 2), %%mm1     \n\t" /*buf0[eax]*/\
276                 "movq 8(%0, %%eax, 2), %%mm7    \n\t" /*buf0[eax]*/\
277                 "psraw $4, %%mm1                \n\t" /* buf0[eax] - buf1[eax] >>4*/\
278                 "psraw $4, %%mm7                \n\t" /* buf0[eax] - buf1[eax] >>4*/\
279                 "pmulhw ubCoeff, %%mm2          \n\t"\
280                 "pmulhw vrCoeff, %%mm5          \n\t"\
281                 "psubw w80, %%mm1               \n\t" /* 8(Y-16)*/\
282                 "psubw w80, %%mm7               \n\t" /* 8(Y-16)*/\
283                 "pmulhw yCoeff, %%mm1           \n\t"\
284                 "pmulhw yCoeff, %%mm7           \n\t"\
285         /* mm1= Y1, mm2=ub, mm3=ug, mm4=vg mm5=vr, mm7=Y2 */\
286                 "paddw %%mm3, %%mm4             \n\t"\
287                 "movq %%mm2, %%mm0              \n\t"\
288                 "movq %%mm5, %%mm6              \n\t"\
289                 "movq %%mm4, %%mm3              \n\t"\
290                 "punpcklwd %%mm2, %%mm2         \n\t"\
291                 "punpcklwd %%mm5, %%mm5         \n\t"\
292                 "punpcklwd %%mm4, %%mm4         \n\t"\
293                 "paddw %%mm1, %%mm2             \n\t"\
294                 "paddw %%mm1, %%mm5             \n\t"\
295                 "paddw %%mm1, %%mm4             \n\t"\
296                 "punpckhwd %%mm0, %%mm0         \n\t"\
297                 "punpckhwd %%mm6, %%mm6         \n\t"\
298                 "punpckhwd %%mm3, %%mm3         \n\t"\
299                 "paddw %%mm7, %%mm0             \n\t"\
300                 "paddw %%mm7, %%mm6             \n\t"\
301                 "paddw %%mm7, %%mm3             \n\t"\
302                 /* mm0=B1, mm2=B2, mm3=G2, mm4=G1, mm5=R1, mm6=R2 */\
303                 "packuswb %%mm0, %%mm2          \n\t"\
304                 "packuswb %%mm6, %%mm5          \n\t"\
305                 "packuswb %%mm3, %%mm4          \n\t"\
306                 "pxor %%mm7, %%mm7              \n\t"
307
308 // do vertical chrominance interpolation
309 #define YSCALEYUV2RGB1b \
310                 "xorl %%eax, %%eax              \n\t"\
311                 "1:                             \n\t"\
312                 "movq (%2, %%eax), %%mm2        \n\t" /* uvbuf0[eax]*/\
313                 "movq (%3, %%eax), %%mm3        \n\t" /* uvbuf1[eax]*/\
314                 "movq 4096(%2, %%eax), %%mm5    \n\t" /* uvbuf0[eax+2048]*/\
315                 "movq 4096(%3, %%eax), %%mm4    \n\t" /* uvbuf1[eax+2048]*/\
316                 "paddw %%mm2, %%mm3             \n\t" /* uvbuf0[eax] + uvbuf1[eax]*/\
317                 "paddw %%mm5, %%mm4             \n\t" /* uvbuf0[eax+2048] + uvbuf1[eax+2048]*/\
318                 "psrlw $5, %%mm3                \n\t"\
319                 "psrlw $5, %%mm4                \n\t"\
320                 "psubw w400, %%mm3              \n\t" /* (U-128)8*/\
321                 "psubw w400, %%mm4              \n\t" /* (V-128)8*/\
322                 "movq %%mm3, %%mm2              \n\t" /* (U-128)8*/\
323                 "movq %%mm4, %%mm5              \n\t" /* (V-128)8*/\
324                 "pmulhw ugCoeff, %%mm3          \n\t"\
325                 "pmulhw vgCoeff, %%mm4          \n\t"\
326         /* mm2=(U-128)8, mm3=ug, mm4=vg mm5=(V-128)8 */\
327                 "movq (%0, %%eax, 2), %%mm1     \n\t" /*buf0[eax]*/\
328                 "movq 8(%0, %%eax, 2), %%mm7    \n\t" /*buf0[eax]*/\
329                 "psraw $4, %%mm1                \n\t" /* buf0[eax] - buf1[eax] >>4*/\
330                 "psraw $4, %%mm7                \n\t" /* buf0[eax] - buf1[eax] >>4*/\
331                 "pmulhw ubCoeff, %%mm2          \n\t"\
332                 "pmulhw vrCoeff, %%mm5          \n\t"\
333                 "psubw w80, %%mm1               \n\t" /* 8(Y-16)*/\
334                 "psubw w80, %%mm7               \n\t" /* 8(Y-16)*/\
335                 "pmulhw yCoeff, %%mm1           \n\t"\
336                 "pmulhw yCoeff, %%mm7           \n\t"\
337         /* mm1= Y1, mm2=ub, mm3=ug, mm4=vg mm5=vr, mm7=Y2 */\
338                 "paddw %%mm3, %%mm4             \n\t"\
339                 "movq %%mm2, %%mm0              \n\t"\
340                 "movq %%mm5, %%mm6              \n\t"\
341                 "movq %%mm4, %%mm3              \n\t"\
342                 "punpcklwd %%mm2, %%mm2         \n\t"\
343                 "punpcklwd %%mm5, %%mm5         \n\t"\
344                 "punpcklwd %%mm4, %%mm4         \n\t"\
345                 "paddw %%mm1, %%mm2             \n\t"\
346                 "paddw %%mm1, %%mm5             \n\t"\
347                 "paddw %%mm1, %%mm4             \n\t"\
348                 "punpckhwd %%mm0, %%mm0         \n\t"\
349                 "punpckhwd %%mm6, %%mm6         \n\t"\
350                 "punpckhwd %%mm3, %%mm3         \n\t"\
351                 "paddw %%mm7, %%mm0             \n\t"\
352                 "paddw %%mm7, %%mm6             \n\t"\
353                 "paddw %%mm7, %%mm3             \n\t"\
354                 /* mm0=B1, mm2=B2, mm3=G2, mm4=G1, mm5=R1, mm6=R2 */\
355                 "packuswb %%mm0, %%mm2          \n\t"\
356                 "packuswb %%mm6, %%mm5          \n\t"\
357                 "packuswb %%mm3, %%mm4          \n\t"\
358                 "pxor %%mm7, %%mm7              \n\t"
359
360 #define WRITEBGR32 \
361                 /* mm2=B, %%mm4=G, %%mm5=R, %%mm7=0 */\
362                         "movq %%mm2, %%mm1              \n\t" /* B */\
363                         "movq %%mm5, %%mm6              \n\t" /* R */\
364                         "punpcklbw %%mm4, %%mm2         \n\t" /* GBGBGBGB 0 */\
365                         "punpcklbw %%mm7, %%mm5         \n\t" /* 0R0R0R0R 0 */\
366                         "punpckhbw %%mm4, %%mm1         \n\t" /* GBGBGBGB 2 */\
367                         "punpckhbw %%mm7, %%mm6         \n\t" /* 0R0R0R0R 2 */\
368                         "movq %%mm2, %%mm0              \n\t" /* GBGBGBGB 0 */\
369                         "movq %%mm1, %%mm3              \n\t" /* GBGBGBGB 2 */\
370                         "punpcklwd %%mm5, %%mm0         \n\t" /* 0RGB0RGB 0 */\
371                         "punpckhwd %%mm5, %%mm2         \n\t" /* 0RGB0RGB 1 */\
372                         "punpcklwd %%mm6, %%mm1         \n\t" /* 0RGB0RGB 2 */\
373                         "punpckhwd %%mm6, %%mm3         \n\t" /* 0RGB0RGB 3 */\
374 \
375                         MOVNTQ(%%mm0, (%4, %%eax, 4))\
376                         MOVNTQ(%%mm2, 8(%4, %%eax, 4))\
377                         MOVNTQ(%%mm1, 16(%4, %%eax, 4))\
378                         MOVNTQ(%%mm3, 24(%4, %%eax, 4))\
379 \
380                         "addl $8, %%eax                 \n\t"\
381                         "cmpl %5, %%eax                 \n\t"\
382                         " jb 1b                         \n\t"
383
384 #define WRITEBGR16 \
385                         "pand bF8, %%mm2                \n\t" /* B */\
386                         "pand bFC, %%mm4                \n\t" /* G */\
387                         "pand bF8, %%mm5                \n\t" /* R */\
388                         "psrlq $3, %%mm2                \n\t"\
389 \
390                         "movq %%mm2, %%mm1              \n\t"\
391                         "movq %%mm4, %%mm3              \n\t"\
392 \
393                         "punpcklbw %%mm7, %%mm3         \n\t"\
394                         "punpcklbw %%mm5, %%mm2         \n\t"\
395                         "punpckhbw %%mm7, %%mm4         \n\t"\
396                         "punpckhbw %%mm5, %%mm1         \n\t"\
397 \
398                         "psllq $3, %%mm3                \n\t"\
399                         "psllq $3, %%mm4                \n\t"\
400 \
401                         "por %%mm3, %%mm2               \n\t"\
402                         "por %%mm4, %%mm1               \n\t"\
403 \
404                         MOVNTQ(%%mm2, (%4, %%eax, 2))\
405                         MOVNTQ(%%mm1, 8(%4, %%eax, 2))\
406 \
407                         "addl $8, %%eax                 \n\t"\
408                         "cmpl %5, %%eax                 \n\t"\
409                         " jb 1b                         \n\t"
410
411 #define WRITEBGR15 \
412                         "pand bF8, %%mm2                \n\t" /* B */\
413                         "pand bF8, %%mm4                \n\t" /* G */\
414                         "pand bF8, %%mm5                \n\t" /* R */\
415                         "psrlq $3, %%mm2                \n\t"\
416                         "psrlq $1, %%mm5                \n\t"\
417 \
418                         "movq %%mm2, %%mm1              \n\t"\
419                         "movq %%mm4, %%mm3              \n\t"\
420 \
421                         "punpcklbw %%mm7, %%mm3         \n\t"\
422                         "punpcklbw %%mm5, %%mm2         \n\t"\
423                         "punpckhbw %%mm7, %%mm4         \n\t"\
424                         "punpckhbw %%mm5, %%mm1         \n\t"\
425 \
426                         "psllq $2, %%mm3                \n\t"\
427                         "psllq $2, %%mm4                \n\t"\
428 \
429                         "por %%mm3, %%mm2               \n\t"\
430                         "por %%mm4, %%mm1               \n\t"\
431 \
432                         MOVNTQ(%%mm2, (%4, %%eax, 2))\
433                         MOVNTQ(%%mm1, 8(%4, %%eax, 2))\
434 \
435                         "addl $8, %%eax                 \n\t"\
436                         "cmpl %5, %%eax                 \n\t"\
437                         " jb 1b                         \n\t"
438
439 #define WRITEBGR24OLD \
440                 /* mm2=B, %%mm4=G, %%mm5=R, %%mm7=0 */\
441                         "movq %%mm2, %%mm1              \n\t" /* B */\
442                         "movq %%mm5, %%mm6              \n\t" /* R */\
443                         "punpcklbw %%mm4, %%mm2         \n\t" /* GBGBGBGB 0 */\
444                         "punpcklbw %%mm7, %%mm5         \n\t" /* 0R0R0R0R 0 */\
445                         "punpckhbw %%mm4, %%mm1         \n\t" /* GBGBGBGB 2 */\
446                         "punpckhbw %%mm7, %%mm6         \n\t" /* 0R0R0R0R 2 */\
447                         "movq %%mm2, %%mm0              \n\t" /* GBGBGBGB 0 */\
448                         "movq %%mm1, %%mm3              \n\t" /* GBGBGBGB 2 */\
449                         "punpcklwd %%mm5, %%mm0         \n\t" /* 0RGB0RGB 0 */\
450                         "punpckhwd %%mm5, %%mm2         \n\t" /* 0RGB0RGB 1 */\
451                         "punpcklwd %%mm6, %%mm1         \n\t" /* 0RGB0RGB 2 */\
452                         "punpckhwd %%mm6, %%mm3         \n\t" /* 0RGB0RGB 3 */\
453 \
454                         "movq %%mm0, %%mm4              \n\t" /* 0RGB0RGB 0 */\
455                         "psrlq $8, %%mm0                \n\t" /* 00RGB0RG 0 */\
456                         "pand bm00000111, %%mm4         \n\t" /* 00000RGB 0 */\
457                         "pand bm11111000, %%mm0         \n\t" /* 00RGB000 0.5 */\
458                         "por %%mm4, %%mm0               \n\t" /* 00RGBRGB 0 */\
459                         "movq %%mm2, %%mm4              \n\t" /* 0RGB0RGB 1 */\
460                         "psllq $48, %%mm2               \n\t" /* GB000000 1 */\
461                         "por %%mm2, %%mm0               \n\t" /* GBRGBRGB 0 */\
462 \
463                         "movq %%mm4, %%mm2              \n\t" /* 0RGB0RGB 1 */\
464                         "psrld $16, %%mm4               \n\t" /* 000R000R 1 */\
465                         "psrlq $24, %%mm2               \n\t" /* 0000RGB0 1.5 */\
466                         "por %%mm4, %%mm2               \n\t" /* 000RRGBR 1 */\
467                         "pand bm00001111, %%mm2         \n\t" /* 0000RGBR 1 */\
468                         "movq %%mm1, %%mm4              \n\t" /* 0RGB0RGB 2 */\
469                         "psrlq $8, %%mm1                \n\t" /* 00RGB0RG 2 */\
470                         "pand bm00000111, %%mm4         \n\t" /* 00000RGB 2 */\
471                         "pand bm11111000, %%mm1         \n\t" /* 00RGB000 2.5 */\
472                         "por %%mm4, %%mm1               \n\t" /* 00RGBRGB 2 */\
473                         "movq %%mm1, %%mm4              \n\t" /* 00RGBRGB 2 */\
474                         "psllq $32, %%mm1               \n\t" /* BRGB0000 2 */\
475                         "por %%mm1, %%mm2               \n\t" /* BRGBRGBR 1 */\
476 \
477                         "psrlq $32, %%mm4               \n\t" /* 000000RG 2.5 */\
478                         "movq %%mm3, %%mm5              \n\t" /* 0RGB0RGB 3 */\
479                         "psrlq $8, %%mm3                \n\t" /* 00RGB0RG 3 */\
480                         "pand bm00000111, %%mm5         \n\t" /* 00000RGB 3 */\
481                         "pand bm11111000, %%mm3         \n\t" /* 00RGB000 3.5 */\
482                         "por %%mm5, %%mm3               \n\t" /* 00RGBRGB 3 */\
483                         "psllq $16, %%mm3               \n\t" /* RGBRGB00 3 */\
484                         "por %%mm4, %%mm3               \n\t" /* RGBRGBRG 2.5 */\
485 \
486                         MOVNTQ(%%mm0, (%%ebx))\
487                         MOVNTQ(%%mm2, 8(%%ebx))\
488                         MOVNTQ(%%mm3, 16(%%ebx))\
489                         "addl $24, %%ebx                \n\t"\
490 \
491                         "addl $8, %%eax                 \n\t"\
492                         "cmpl %5, %%eax                 \n\t"\
493                         " jb 1b                         \n\t"
494
495 #define WRITEBGR24MMX \
496                 /* mm2=B, %%mm4=G, %%mm5=R, %%mm7=0 */\
497                         "movq %%mm2, %%mm1              \n\t" /* B */\
498                         "movq %%mm5, %%mm6              \n\t" /* R */\
499                         "punpcklbw %%mm4, %%mm2         \n\t" /* GBGBGBGB 0 */\
500                         "punpcklbw %%mm7, %%mm5         \n\t" /* 0R0R0R0R 0 */\
501                         "punpckhbw %%mm4, %%mm1         \n\t" /* GBGBGBGB 2 */\
502                         "punpckhbw %%mm7, %%mm6         \n\t" /* 0R0R0R0R 2 */\
503                         "movq %%mm2, %%mm0              \n\t" /* GBGBGBGB 0 */\
504                         "movq %%mm1, %%mm3              \n\t" /* GBGBGBGB 2 */\
505                         "punpcklwd %%mm5, %%mm0         \n\t" /* 0RGB0RGB 0 */\
506                         "punpckhwd %%mm5, %%mm2         \n\t" /* 0RGB0RGB 1 */\
507                         "punpcklwd %%mm6, %%mm1         \n\t" /* 0RGB0RGB 2 */\
508                         "punpckhwd %%mm6, %%mm3         \n\t" /* 0RGB0RGB 3 */\
509 \
510                         "movq %%mm0, %%mm4              \n\t" /* 0RGB0RGB 0 */\
511                         "movq %%mm2, %%mm6              \n\t" /* 0RGB0RGB 1 */\
512                         "movq %%mm1, %%mm5              \n\t" /* 0RGB0RGB 2 */\
513                         "movq %%mm3, %%mm7              \n\t" /* 0RGB0RGB 3 */\
514 \
515                         "psllq $40, %%mm0               \n\t" /* RGB00000 0 */\
516                         "psllq $40, %%mm2               \n\t" /* RGB00000 1 */\
517                         "psllq $40, %%mm1               \n\t" /* RGB00000 2 */\
518                         "psllq $40, %%mm3               \n\t" /* RGB00000 3 */\
519 \
520                         "punpckhdq %%mm4, %%mm0         \n\t" /* 0RGBRGB0 0 */\
521                         "punpckhdq %%mm6, %%mm2         \n\t" /* 0RGBRGB0 1 */\
522                         "punpckhdq %%mm5, %%mm1         \n\t" /* 0RGBRGB0 2 */\
523                         "punpckhdq %%mm7, %%mm3         \n\t" /* 0RGBRGB0 3 */\
524 \
525                         "psrlq $8, %%mm0                \n\t" /* 00RGBRGB 0 */\
526                         "movq %%mm2, %%mm6              \n\t" /* 0RGBRGB0 1 */\
527                         "psllq $40, %%mm2               \n\t" /* GB000000 1 */\
528                         "por %%mm2, %%mm0               \n\t" /* GBRGBRGB 0 */\
529                         MOVNTQ(%%mm0, (%%ebx))\
530 \
531                         "psrlq $24, %%mm6               \n\t" /* 0000RGBR 1 */\
532                         "movq %%mm1, %%mm5              \n\t" /* 0RGBRGB0 2 */\
533                         "psllq $24, %%mm1               \n\t" /* BRGB0000 2 */\
534                         "por %%mm1, %%mm6               \n\t" /* BRGBRGBR 1 */\
535                         MOVNTQ(%%mm6, 8(%%ebx))\
536 \
537                         "psrlq $40, %%mm5               \n\t" /* 000000RG 2 */\
538                         "psllq $8, %%mm3                \n\t" /* RGBRGB00 3 */\
539                         "por %%mm3, %%mm5               \n\t" /* RGBRGBRG 2 */\
540                         MOVNTQ(%%mm5, 16(%%ebx))\
541 \
542                         "addl $24, %%ebx                \n\t"\
543 \
544                         "addl $8, %%eax                 \n\t"\
545                         "cmpl %5, %%eax                 \n\t"\
546                         " jb 1b                         \n\t"
547
548 #define WRITEBGR24MMX2 \
549                 /* mm2=B, %%mm4=G, %%mm5=R, %%mm7=0 */\
550                         "movq M24A, %%mm0               \n\t"\
551                         "movq M24C, %%mm7               \n\t"\
552                         "pshufw $0x50, %%mm2, %%mm1     \n\t" /* B3 B2 B3 B2  B1 B0 B1 B0 */\
553                         "pshufw $0x50, %%mm4, %%mm3     \n\t" /* G3 G2 G3 G2  G1 G0 G1 G0 */\
554                         "pshufw $0x00, %%mm5, %%mm6     \n\t" /* R1 R0 R1 R0  R1 R0 R1 R0 */\
555 \
556                         "pand %%mm0, %%mm1              \n\t" /*    B2        B1       B0 */\
557                         "pand %%mm0, %%mm3              \n\t" /*    G2        G1       G0 */\
558                         "pand %%mm7, %%mm6              \n\t" /*       R1        R0       */\
559 \
560                         "psllq $8, %%mm3                \n\t" /* G2        G1       G0    */\
561                         "por %%mm1, %%mm6               \n\t"\
562                         "por %%mm3, %%mm6               \n\t"\
563                         MOVNTQ(%%mm6, (%%ebx))\
564 \
565                         "psrlq $8, %%mm4                \n\t" /* 00 G7 G6 G5  G4 G3 G2 G1 */\
566                         "pshufw $0xA5, %%mm2, %%mm1     \n\t" /* B5 B4 B5 B4  B3 B2 B3 B2 */\
567                         "pshufw $0x55, %%mm4, %%mm3     \n\t" /* G4 G3 G4 G3  G4 G3 G4 G3 */\
568                         "pshufw $0xA5, %%mm5, %%mm6     \n\t" /* R5 R4 R5 R4  R3 R2 R3 R2 */\
569 \
570                         "pand M24B, %%mm1               \n\t" /* B5       B4        B3    */\
571                         "pand %%mm7, %%mm3              \n\t" /*       G4        G3       */\
572                         "pand %%mm0, %%mm6              \n\t" /*    R4        R3       R2 */\
573 \
574                         "por %%mm1, %%mm3               \n\t" /* B5    G4 B4     G3 B3    */\
575                         "por %%mm3, %%mm6               \n\t"\
576                         MOVNTQ(%%mm6, 8(%%ebx))\
577 \
578                         "pshufw $0xFF, %%mm2, %%mm1     \n\t" /* B7 B6 B7 B6  B7 B6 B6 B7 */\
579                         "pshufw $0xFA, %%mm4, %%mm3     \n\t" /* 00 G7 00 G7  G6 G5 G6 G5 */\
580                         "pshufw $0xFA, %%mm5, %%mm6     \n\t" /* R7 R6 R7 R6  R5 R4 R5 R4 */\
581 \
582                         "pand %%mm7, %%mm1              \n\t" /*       B7        B6       */\
583                         "pand %%mm0, %%mm3              \n\t" /*    G7        G6       G5 */\
584                         "pand M24B, %%mm6               \n\t" /* R7       R6        R5    */\
585 \
586                         "por %%mm1, %%mm3               \n\t"\
587                         "por %%mm3, %%mm6               \n\t"\
588                         MOVNTQ(%%mm6, 16(%%ebx))\
589 \
590                         "addl $24, %%ebx                \n\t"\
591 \
592                         "addl $8, %%eax                 \n\t"\
593                         "cmpl %5, %%eax                 \n\t"\
594                         " jb 1b                         \n\t"
595
596 #ifdef HAVE_MMX2
597 #define WRITEBGR24 WRITEBGR24MMX2
598 #else
599 #define WRITEBGR24 WRITEBGR24MMX
600 #endif
601
602 #ifdef HAVE_MMX
603 void in_asm_used_var_warning_killer()
604 {
605  int i= yCoeff+vrCoeff+ubCoeff+vgCoeff+ugCoeff+bF8+bFC+w400+w80+w10+
606  bm00001111+bm00000111+bm11111000+b16Mask+g16Mask+r16Mask+b15Mask+g15Mask+r15Mask+temp0+asm_yalpha1+ asm_uvalpha1+
607  M24A+M24B+M24C;
608  if(i) i=0;
609 }
610 #endif
611
612 static inline void yuv2yuv(uint16_t *buf0, uint16_t *buf1, uint16_t *uvbuf0, uint16_t *uvbuf1,
613                            uint8_t *dest, uint8_t *uDest, uint8_t *vDest, int dstw, int yalpha, int uvalpha)
614 {
615         int yalpha1=yalpha^4095;
616         int uvalpha1=uvalpha^4095;
617         int i;
618
619         asm volatile ("\n\t"::: "memory");
620
621         for(i=0;i<dstw;i++)
622         {
623                 ((uint8_t*)dest)[i] = (buf0[i]*yalpha1+buf1[i]*yalpha)>>19;
624         }
625
626         if(uvalpha != -1)
627         {
628                 for(i=0; i<(dstw>>1); i++)
629                 {
630                         ((uint8_t*)uDest)[i] = (uvbuf0[i]*uvalpha1+uvbuf1[i]*uvalpha)>>19;
631                         ((uint8_t*)vDest)[i] = (uvbuf0[i+2048]*uvalpha1+uvbuf1[i+2048]*uvalpha)>>19;
632                 }
633         }
634 }
635
636 /**
637  * vertical scale YV12 to RGB
638  */
639 static inline void yuv2rgbX(uint16_t *buf0, uint16_t *buf1, uint16_t *uvbuf0, uint16_t *uvbuf1,
640                             uint8_t *dest, int dstw, int yalpha, int uvalpha, int dstbpp)
641 {
642         int yalpha1=yalpha^4095;
643         int uvalpha1=uvalpha^4095;
644
645         if(fullUVIpol)
646         {
647
648 #ifdef HAVE_MMX
649                 if(dstbpp == 32)
650                 {
651                         asm volatile(
652
653
654 FULL_YSCALEYUV2RGB
655                         "punpcklbw %%mm1, %%mm3         \n\t" // BGBGBGBG
656                         "punpcklbw %%mm7, %%mm0         \n\t" // R0R0R0R0
657
658                         "movq %%mm3, %%mm1              \n\t"
659                         "punpcklwd %%mm0, %%mm3         \n\t" // BGR0BGR0
660                         "punpckhwd %%mm0, %%mm1         \n\t" // BGR0BGR0
661
662                         MOVNTQ(%%mm3, (%4, %%eax, 4))
663                         MOVNTQ(%%mm1, 8(%4, %%eax, 4))
664
665                         "addl $4, %%eax                 \n\t"
666                         "cmpl %5, %%eax                 \n\t"
667                         " jb 1b                         \n\t"
668
669
670                         :: "r" (buf0), "r" (buf1), "r" (uvbuf0), "r" (uvbuf1), "r" (dest), "m" (dstw),
671                         "m" (yalpha1), "m" (uvalpha1)
672                         : "%eax"
673                         );
674                 }
675                 else if(dstbpp==24)
676                 {
677                         asm volatile(
678
679 FULL_YSCALEYUV2RGB
680
681                                                                 // lsb ... msb
682                         "punpcklbw %%mm1, %%mm3         \n\t" // BGBGBGBG
683                         "punpcklbw %%mm7, %%mm0         \n\t" // R0R0R0R0
684
685                         "movq %%mm3, %%mm1              \n\t"
686                         "punpcklwd %%mm0, %%mm3         \n\t" // BGR0BGR0
687                         "punpckhwd %%mm0, %%mm1         \n\t" // BGR0BGR0
688
689                         "movq %%mm3, %%mm2              \n\t" // BGR0BGR0
690                         "psrlq $8, %%mm3                \n\t" // GR0BGR00
691                         "pand bm00000111, %%mm2         \n\t" // BGR00000
692                         "pand bm11111000, %%mm3         \n\t" // 000BGR00
693                         "por %%mm2, %%mm3               \n\t" // BGRBGR00
694                         "movq %%mm1, %%mm2              \n\t"
695                         "psllq $48, %%mm1               \n\t" // 000000BG
696                         "por %%mm1, %%mm3               \n\t" // BGRBGRBG
697
698                         "movq %%mm2, %%mm1              \n\t" // BGR0BGR0
699                         "psrld $16, %%mm2               \n\t" // R000R000
700                         "psrlq $24, %%mm1               \n\t" // 0BGR0000
701                         "por %%mm2, %%mm1               \n\t" // RBGRR000
702
703                         "movl %4, %%ebx                 \n\t"
704                         "addl %%eax, %%ebx              \n\t"
705
706 #ifdef HAVE_MMX2
707                         //FIXME Alignment
708                         "movntq %%mm3, (%%ebx, %%eax, 2)\n\t"
709                         "movntq %%mm1, 8(%%ebx, %%eax, 2)\n\t"
710 #else
711                         "movd %%mm3, (%%ebx, %%eax, 2)  \n\t"
712                         "psrlq $32, %%mm3               \n\t"
713                         "movd %%mm3, 4(%%ebx, %%eax, 2) \n\t"
714                         "movd %%mm1, 8(%%ebx, %%eax, 2) \n\t"
715 #endif
716                         "addl $4, %%eax                 \n\t"
717                         "cmpl %5, %%eax                 \n\t"
718                         " jb 1b                         \n\t"
719
720                         :: "r" (buf0), "r" (buf1), "r" (uvbuf0), "r" (uvbuf1), "m" (dest), "m" (dstw),
721                         "m" (yalpha1), "m" (uvalpha1)
722                         : "%eax", "%ebx"
723                         );
724                 }
725                 else if(dstbpp==15)
726                 {
727                         asm volatile(
728
729 FULL_YSCALEYUV2RGB
730 #ifdef DITHER1XBPP
731                         "paddusb g5Dither, %%mm1        \n\t"
732                         "paddusb r5Dither, %%mm0        \n\t"
733                         "paddusb b5Dither, %%mm3        \n\t"
734 #endif
735                         "punpcklbw %%mm7, %%mm1         \n\t" // 0G0G0G0G
736                         "punpcklbw %%mm7, %%mm3         \n\t" // 0B0B0B0B
737                         "punpcklbw %%mm7, %%mm0         \n\t" // 0R0R0R0R
738
739                         "psrlw $3, %%mm3                \n\t"
740                         "psllw $2, %%mm1                \n\t"
741                         "psllw $7, %%mm0                \n\t"
742                         "pand g15Mask, %%mm1            \n\t"
743                         "pand r15Mask, %%mm0            \n\t"
744
745                         "por %%mm3, %%mm1               \n\t"
746                         "por %%mm1, %%mm0               \n\t"
747
748                         MOVNTQ(%%mm0, (%4, %%eax, 2))
749
750                         "addl $4, %%eax                 \n\t"
751                         "cmpl %5, %%eax                 \n\t"
752                         " jb 1b                         \n\t"
753
754                         :: "r" (buf0), "r" (buf1), "r" (uvbuf0), "r" (uvbuf1), "r" (dest), "m" (dstw),
755                         "m" (yalpha1), "m" (uvalpha1)
756                         : "%eax"
757                         );
758                 }
759                 else if(dstbpp==16)
760                 {
761                         asm volatile(
762
763 FULL_YSCALEYUV2RGB
764 #ifdef DITHER1XBPP
765                         "paddusb g6Dither, %%mm1        \n\t"
766                         "paddusb r5Dither, %%mm0        \n\t"
767                         "paddusb b5Dither, %%mm3        \n\t"
768 #endif
769                         "punpcklbw %%mm7, %%mm1         \n\t" // 0G0G0G0G
770                         "punpcklbw %%mm7, %%mm3         \n\t" // 0B0B0B0B
771                         "punpcklbw %%mm7, %%mm0         \n\t" // 0R0R0R0R
772
773                         "psrlw $3, %%mm3                \n\t"
774                         "psllw $3, %%mm1                \n\t"
775                         "psllw $8, %%mm0                \n\t"
776                         "pand g16Mask, %%mm1            \n\t"
777                         "pand r16Mask, %%mm0            \n\t"
778
779                         "por %%mm3, %%mm1               \n\t"
780                         "por %%mm1, %%mm0               \n\t"
781
782                         MOVNTQ(%%mm0, (%4, %%eax, 2))
783
784                         "addl $4, %%eax                 \n\t"
785                         "cmpl %5, %%eax                 \n\t"
786                         " jb 1b                         \n\t"
787
788                         :: "r" (buf0), "r" (buf1), "r" (uvbuf0), "r" (uvbuf1), "r" (dest), "m" (dstw),
789                         "m" (yalpha1), "m" (uvalpha1)
790                         : "%eax"
791                         );
792                 }
793 #else
794                 asm volatile ("\n\t"::: "memory");
795
796                 if(dstbpp==32 || dstbpp==24)
797                 {
798                         int i;
799                         for(i=0;i<dstw;i++){
800                                 // vertical linear interpolation && yuv2rgb in a single step:
801                                 int Y=yuvtab_2568[((buf0[i]*yalpha1+buf1[i]*yalpha)>>19)];
802                                 int U=((uvbuf0[i]*uvalpha1+uvbuf1[i]*uvalpha)>>19);
803                                 int V=((uvbuf0[i+2048]*uvalpha1+uvbuf1[i+2048]*uvalpha)>>19);
804                                 dest[0]=clip_table[((Y + yuvtab_40cf[U]) >>13)];
805                                 dest[1]=clip_table[((Y + yuvtab_1a1e[V] + yuvtab_0c92[U]) >>13)];
806                                 dest[2]=clip_table[((Y + yuvtab_3343[V]) >>13)];
807                                 dest+=dstbpp>>3;
808                         }
809                 }
810                 else if(dstbpp==16)
811                 {
812                         int i;
813                         for(i=0;i<dstw;i++){
814                                 // vertical linear interpolation && yuv2rgb in a single step:
815                                 int Y=yuvtab_2568[((buf0[i]*yalpha1+buf1[i]*yalpha)>>19)];
816                                 int U=((uvbuf0[i]*uvalpha1+uvbuf1[i]*uvalpha)>>19);
817                                 int V=((uvbuf0[i+2048]*uvalpha1+uvbuf1[i+2048]*uvalpha)>>19);
818
819                                 ((uint16_t*)dest)[i] =
820                                         clip_table16b[(Y + yuvtab_40cf[U]) >>13] |
821                                         clip_table16g[(Y + yuvtab_1a1e[V] + yuvtab_0c92[U]) >>13] |
822                                         clip_table16r[(Y + yuvtab_3343[V]) >>13];
823                         }
824                 }
825                 else if(dstbpp==15)
826                 {
827                         int i;
828                         for(i=0;i<dstw;i++){
829                                 // vertical linear interpolation && yuv2rgb in a single step:
830                                 int Y=yuvtab_2568[((buf0[i]*yalpha1+buf1[i]*yalpha)>>19)];
831                                 int U=((uvbuf0[i]*uvalpha1+uvbuf1[i]*uvalpha)>>19);
832                                 int V=((uvbuf0[i+2048]*uvalpha1+uvbuf1[i+2048]*uvalpha)>>19);
833
834                                 ((uint16_t*)dest)[i] =
835                                         clip_table15b[(Y + yuvtab_40cf[U]) >>13] |
836                                         clip_table15g[(Y + yuvtab_1a1e[V] + yuvtab_0c92[U]) >>13] |
837                                         clip_table15r[(Y + yuvtab_3343[V]) >>13];
838                         }
839                 }
840 #endif
841         }//FULL_UV_IPOL
842         else
843         {
844 #ifdef HAVE_MMX
845                 if(dstbpp == 32)
846                 {
847                         asm volatile(
848                                 YSCALEYUV2RGB
849                                 WRITEBGR32
850
851                         :: "r" (buf0), "r" (buf1), "r" (uvbuf0), "r" (uvbuf1), "r" (dest), "m" (dstw),
852                         "m" (yalpha1), "m" (uvalpha1)
853                         : "%eax"
854                         );
855                 }
856                 else if(dstbpp==24)
857                 {
858                         asm volatile(
859                                 "movl %4, %%ebx                 \n\t"
860                                 YSCALEYUV2RGB
861                                 WRITEBGR24
862
863                         :: "r" (buf0), "r" (buf1), "r" (uvbuf0), "r" (uvbuf1), "m" (dest), "m" (dstw),
864                         "m" (yalpha1), "m" (uvalpha1)
865                         : "%eax", "%ebx"
866                         );
867                 }
868                 else if(dstbpp==15)
869                 {
870                         asm volatile(
871                                 YSCALEYUV2RGB
872                 /* mm2=B, %%mm4=G, %%mm5=R, %%mm7=0 */
873 #ifdef DITHER1XBPP
874                                 "paddusb b5Dither, %%mm2        \n\t"
875                                 "paddusb g5Dither, %%mm4        \n\t"
876                                 "paddusb r5Dither, %%mm5        \n\t"
877 #endif
878
879                                 WRITEBGR15
880
881                         :: "r" (buf0), "r" (buf1), "r" (uvbuf0), "r" (uvbuf1), "r" (dest), "m" (dstw),
882                         "m" (yalpha1), "m" (uvalpha1)
883                         : "%eax"
884                         );
885                 }
886                 else if(dstbpp==16)
887                 {
888                         asm volatile(
889                                 YSCALEYUV2RGB
890                 /* mm2=B, %%mm4=G, %%mm5=R, %%mm7=0 */
891 #ifdef DITHER1XBPP
892                                 "paddusb b5Dither, %%mm2        \n\t"
893                                 "paddusb g6Dither, %%mm4        \n\t"
894                                 "paddusb r5Dither, %%mm5        \n\t"
895 #endif
896
897                                 WRITEBGR16
898
899                         :: "r" (buf0), "r" (buf1), "r" (uvbuf0), "r" (uvbuf1), "r" (dest), "m" (dstw),
900                         "m" (yalpha1), "m" (uvalpha1)
901                         : "%eax"
902                         );
903                 }
904 #else
905                 asm volatile ("\n\t"::: "memory");
906
907                 if(dstbpp==32)
908                 {
909                         int i;
910                         for(i=0; i<dstw-1; i+=2){
911                                 // vertical linear interpolation && yuv2rgb in a single step:
912                                 int Y1=yuvtab_2568[((buf0[i]*yalpha1+buf1[i]*yalpha)>>19)];
913                                 int Y2=yuvtab_2568[((buf0[i+1]*yalpha1+buf1[i+1]*yalpha)>>19)];
914                                 int U=((uvbuf0[i>>1]*uvalpha1+uvbuf1[i>>1]*uvalpha)>>19);
915                                 int V=((uvbuf0[(i>>1)+2048]*uvalpha1+uvbuf1[(i>>1)+2048]*uvalpha)>>19);
916
917                                 int Cb= yuvtab_40cf[U];
918                                 int Cg= yuvtab_1a1e[V] + yuvtab_0c92[U];
919                                 int Cr= yuvtab_3343[V];
920
921                                 dest[4*i+0]=clip_table[((Y1 + Cb) >>13)];
922                                 dest[4*i+1]=clip_table[((Y1 + Cg) >>13)];
923                                 dest[4*i+2]=clip_table[((Y1 + Cr) >>13)];
924
925                                 dest[4*i+4]=clip_table[((Y2 + Cb) >>13)];
926                                 dest[4*i+5]=clip_table[((Y2 + Cg) >>13)];
927                                 dest[4*i+6]=clip_table[((Y2 + Cr) >>13)];
928                         }
929                 }
930                 if(dstbpp==24)
931                 {
932                         int i;
933                         for(i=0; i<dstw-1; i+=2){
934                                 // vertical linear interpolation && yuv2rgb in a single step:
935                                 int Y1=yuvtab_2568[((buf0[i]*yalpha1+buf1[i]*yalpha)>>19)];
936                                 int Y2=yuvtab_2568[((buf0[i+1]*yalpha1+buf1[i+1]*yalpha)>>19)];
937                                 int U=((uvbuf0[i>>1]*uvalpha1+uvbuf1[i>>1]*uvalpha)>>19);
938                                 int V=((uvbuf0[(i>>1)+2048]*uvalpha1+uvbuf1[(i>>1)+2048]*uvalpha)>>19);
939
940                                 int Cb= yuvtab_40cf[U];
941                                 int Cg= yuvtab_1a1e[V] + yuvtab_0c92[U];
942                                 int Cr= yuvtab_3343[V];
943
944                                 dest[0]=clip_table[((Y1 + Cb) >>13)];
945                                 dest[1]=clip_table[((Y1 + Cg) >>13)];
946                                 dest[2]=clip_table[((Y1 + Cr) >>13)];
947
948                                 dest[3]=clip_table[((Y2 + Cb) >>13)];
949                                 dest[4]=clip_table[((Y2 + Cg) >>13)];
950                                 dest[5]=clip_table[((Y2 + Cr) >>13)];
951                                 dest+=6;
952                         }
953                 }
954                 else if(dstbpp==16)
955                 {
956                         int i;
957                         for(i=0; i<dstw-1; i+=2){
958                                 // vertical linear interpolation && yuv2rgb in a single step:
959                                 int Y1=yuvtab_2568[((buf0[i]*yalpha1+buf1[i]*yalpha)>>19)];
960                                 int Y2=yuvtab_2568[((buf0[i+1]*yalpha1+buf1[i+1]*yalpha)>>19)];
961                                 int U=((uvbuf0[i>>1]*uvalpha1+uvbuf1[i>>1]*uvalpha)>>19);
962                                 int V=((uvbuf0[(i>>1)+2048]*uvalpha1+uvbuf1[(i>>1)+2048]*uvalpha)>>19);
963
964                                 int Cb= yuvtab_40cf[U];
965                                 int Cg= yuvtab_1a1e[V] + yuvtab_0c92[U];
966                                 int Cr= yuvtab_3343[V];
967
968                                 ((uint16_t*)dest)[i] =
969                                         clip_table16b[(Y1 + Cb) >>13] |
970                                         clip_table16g[(Y1 + Cg) >>13] |
971                                         clip_table16r[(Y1 + Cr) >>13];
972
973                                 ((uint16_t*)dest)[i+1] =
974                                         clip_table16b[(Y2 + Cb) >>13] |
975                                         clip_table16g[(Y2 + Cg) >>13] |
976                                         clip_table16r[(Y2 + Cr) >>13];
977                         }
978                 }
979                 else if(dstbpp==15)
980                 {
981                         int i;
982                         for(i=0; i<dstw-1; i+=2){
983                                 // vertical linear interpolation && yuv2rgb in a single step:
984                                 int Y1=yuvtab_2568[((buf0[i]*yalpha1+buf1[i]*yalpha)>>19)];
985                                 int Y2=yuvtab_2568[((buf0[i+1]*yalpha1+buf1[i+1]*yalpha)>>19)];
986                                 int U=((uvbuf0[i>>1]*uvalpha1+uvbuf1[i>>1]*uvalpha)>>19);
987                                 int V=((uvbuf0[(i>>1)+2048]*uvalpha1+uvbuf1[(i>>1)+2048]*uvalpha)>>19);
988
989                                 int Cb= yuvtab_40cf[U];
990                                 int Cg= yuvtab_1a1e[V] + yuvtab_0c92[U];
991                                 int Cr= yuvtab_3343[V];
992
993                                 ((uint16_t*)dest)[i] =
994                                         clip_table15b[(Y1 + Cb) >>13] |
995                                         clip_table15g[(Y1 + Cg) >>13] |
996                                         clip_table15r[(Y1 + Cr) >>13];
997
998                                 ((uint16_t*)dest)[i+1] =
999                                         clip_table15b[(Y2 + Cb) >>13] |
1000                                         clip_table15g[(Y2 + Cg) >>13] |
1001                                         clip_table15r[(Y2 + Cr) >>13];
1002                         }
1003                 }
1004 #endif
1005         } //!FULL_UV_IPOL
1006 }
1007
1008 /**
1009  * YV12 to RGB without scaling or interpolating
1010  */
1011 static inline void yuv2rgb1(uint16_t *buf0, uint16_t *buf1, uint16_t *uvbuf0, uint16_t *uvbuf1,
1012                             uint8_t *dest, int dstw, int yalpha, int uvalpha, int dstbpp)
1013 {
1014         int uvalpha1=uvalpha^4095;
1015 #ifdef HAVE_MMX
1016         int yalpha1=yalpha^4095;
1017 #endif
1018
1019         if(fullUVIpol || allwaysIpol)
1020         {
1021                 yuv2rgbX(buf0, buf1, uvbuf0, uvbuf1, dest, dstw, yalpha, uvalpha, dstbpp);
1022                 return;
1023         }
1024         if( yalpha > 2048 ) buf0 = buf1;
1025
1026 #ifdef HAVE_MMX
1027         if( uvalpha < 2048 ) // note this is not correct (shifts chrominance by 0.5 pixels) but its a bit faster
1028         {
1029                 if(dstbpp == 32)
1030                 {
1031                         asm volatile(
1032                                 YSCALEYUV2RGB1
1033                                 WRITEBGR32
1034                         :: "r" (buf0), "r" (buf1), "r" (uvbuf0), "r" (uvbuf1), "r" (dest), "m" (dstw),
1035                         "m" (yalpha1), "m" (uvalpha1)
1036                         : "%eax"
1037                         );
1038                 }
1039                 else if(dstbpp==24)
1040                 {
1041                         asm volatile(
1042                                 "movl %4, %%ebx                 \n\t"
1043                                 YSCALEYUV2RGB1
1044                                 WRITEBGR24
1045                         :: "r" (buf0), "r" (buf1), "r" (uvbuf0), "r" (uvbuf1), "m" (dest), "m" (dstw),
1046                         "m" (yalpha1), "m" (uvalpha1)
1047                         : "%eax", "%ebx"
1048                         );
1049                 }
1050                 else if(dstbpp==15)
1051                 {
1052                         asm volatile(
1053                                 YSCALEYUV2RGB1
1054                 /* mm2=B, %%mm4=G, %%mm5=R, %%mm7=0 */
1055 #ifdef DITHER1XBPP
1056                                 "paddusb b5Dither, %%mm2        \n\t"
1057                                 "paddusb g5Dither, %%mm4        \n\t"
1058                                 "paddusb r5Dither, %%mm5        \n\t"
1059 #endif
1060                                 WRITEBGR15
1061                         :: "r" (buf0), "r" (buf1), "r" (uvbuf0), "r" (uvbuf1), "r" (dest), "m" (dstw),
1062                         "m" (yalpha1), "m" (uvalpha1)
1063                         : "%eax"
1064                         );
1065                 }
1066                 else if(dstbpp==16)
1067                 {
1068                         asm volatile(
1069                                 YSCALEYUV2RGB1
1070                 /* mm2=B, %%mm4=G, %%mm5=R, %%mm7=0 */
1071 #ifdef DITHER1XBPP
1072                                 "paddusb b5Dither, %%mm2        \n\t"
1073                                 "paddusb g6Dither, %%mm4        \n\t"
1074                                 "paddusb r5Dither, %%mm5        \n\t"
1075 #endif
1076
1077                                 WRITEBGR16
1078                         :: "r" (buf0), "r" (buf1), "r" (uvbuf0), "r" (uvbuf1), "r" (dest), "m" (dstw),
1079                         "m" (yalpha1), "m" (uvalpha1)
1080                         : "%eax"
1081                         );
1082                 }
1083         }
1084         else
1085         {
1086                 if(dstbpp == 32)
1087                 {
1088                         asm volatile(
1089                                 YSCALEYUV2RGB1b
1090                                 WRITEBGR32
1091                         :: "r" (buf0), "r" (buf1), "r" (uvbuf0), "r" (uvbuf1), "r" (dest), "m" (dstw),
1092                         "m" (yalpha1), "m" (uvalpha1)
1093                         : "%eax"
1094                         );
1095                 }
1096                 else if(dstbpp==24)
1097                 {
1098                         asm volatile(
1099                                 "movl %4, %%ebx                 \n\t"
1100                                 YSCALEYUV2RGB1b
1101                                 WRITEBGR24
1102                         :: "r" (buf0), "r" (buf1), "r" (uvbuf0), "r" (uvbuf1), "m" (dest), "m" (dstw),
1103                         "m" (yalpha1), "m" (uvalpha1)
1104                         : "%eax", "%ebx"
1105                         );
1106                 }
1107                 else if(dstbpp==15)
1108                 {
1109                         asm volatile(
1110                                 YSCALEYUV2RGB1b
1111                 /* mm2=B, %%mm4=G, %%mm5=R, %%mm7=0 */
1112 #ifdef DITHER1XBPP
1113                                 "paddusb b5Dither, %%mm2        \n\t"
1114                                 "paddusb g5Dither, %%mm4        \n\t"
1115                                 "paddusb r5Dither, %%mm5        \n\t"
1116 #endif
1117                                 WRITEBGR15
1118                         :: "r" (buf0), "r" (buf1), "r" (uvbuf0), "r" (uvbuf1), "r" (dest), "m" (dstw),
1119                         "m" (yalpha1), "m" (uvalpha1)
1120                         : "%eax"
1121                         );
1122                 }
1123                 else if(dstbpp==16)
1124                 {
1125                         asm volatile(
1126                                 YSCALEYUV2RGB1b
1127                 /* mm2=B, %%mm4=G, %%mm5=R, %%mm7=0 */
1128 #ifdef DITHER1XBPP
1129                                 "paddusb b5Dither, %%mm2        \n\t"
1130                                 "paddusb g6Dither, %%mm4        \n\t"
1131                                 "paddusb r5Dither, %%mm5        \n\t"
1132 #endif
1133
1134                                 WRITEBGR16
1135                         :: "r" (buf0), "r" (buf1), "r" (uvbuf0), "r" (uvbuf1), "r" (dest), "m" (dstw),
1136                         "m" (yalpha1), "m" (uvalpha1)
1137                         : "%eax"
1138                         );
1139                 }
1140         }
1141 #else
1142 //FIXME write 2 versions (for even & odd lines)
1143         asm volatile ("\n\t"::: "memory");
1144
1145         if(dstbpp==32)
1146         {
1147                 int i;
1148                 for(i=0; i<dstw-1; i+=2){
1149                         // vertical linear interpolation && yuv2rgb in a single step:
1150                         int Y1=yuvtab_2568[buf0[i]>>7];
1151                         int Y2=yuvtab_2568[buf0[i+1]>>7];
1152                         int U=((uvbuf0[i>>1]*uvalpha1+uvbuf1[i>>1]*uvalpha)>>19);
1153                         int V=((uvbuf0[(i>>1)+2048]*uvalpha1+uvbuf1[(i>>1)+2048]*uvalpha)>>19);
1154
1155                         int Cb= yuvtab_40cf[U];
1156                         int Cg= yuvtab_1a1e[V] + yuvtab_0c92[U];
1157                         int Cr= yuvtab_3343[V];
1158
1159                         dest[4*i+0]=clip_table[((Y1 + Cb) >>13)];
1160                         dest[4*i+1]=clip_table[((Y1 + Cg) >>13)];
1161                         dest[4*i+2]=clip_table[((Y1 + Cr) >>13)];
1162
1163                         dest[4*i+4]=clip_table[((Y2 + Cb) >>13)];
1164                         dest[4*i+5]=clip_table[((Y2 + Cg) >>13)];
1165                         dest[4*i+6]=clip_table[((Y2 + Cr) >>13)];
1166                 }
1167         }
1168         if(dstbpp==24)
1169         {
1170                 int i;
1171                 for(i=0; i<dstw-1; i+=2){
1172                         // vertical linear interpolation && yuv2rgb in a single step:
1173                         int Y1=yuvtab_2568[buf0[i]>>7];
1174                         int Y2=yuvtab_2568[buf0[i+1]>>7];
1175                         int U=((uvbuf0[i>>1]*uvalpha1+uvbuf1[i>>1]*uvalpha)>>19);
1176                         int V=((uvbuf0[(i>>1)+2048]*uvalpha1+uvbuf1[(i>>1)+2048]*uvalpha)>>19);
1177
1178                         int Cb= yuvtab_40cf[U];
1179                         int Cg= yuvtab_1a1e[V] + yuvtab_0c92[U];
1180                         int Cr= yuvtab_3343[V];
1181
1182                         dest[0]=clip_table[((Y1 + Cb) >>13)];
1183                         dest[1]=clip_table[((Y1 + Cg) >>13)];
1184                         dest[2]=clip_table[((Y1 + Cr) >>13)];
1185
1186                         dest[3]=clip_table[((Y2 + Cb) >>13)];
1187                         dest[4]=clip_table[((Y2 + Cg) >>13)];
1188                         dest[5]=clip_table[((Y2 + Cr) >>13)];
1189                         dest+=6;
1190                 }
1191         }
1192         else if(dstbpp==16)
1193         {
1194                 int i;
1195                 for(i=0; i<dstw-1; i+=2){
1196                         // vertical linear interpolation && yuv2rgb in a single step:
1197                         int Y1=yuvtab_2568[buf0[i]>>7];
1198                         int Y2=yuvtab_2568[buf0[i+1]>>7];
1199                         int U=((uvbuf0[i>>1]*uvalpha1+uvbuf1[i>>1]*uvalpha)>>19);
1200                         int V=((uvbuf0[(i>>1)+2048]*uvalpha1+uvbuf1[(i>>1)+2048]*uvalpha)>>19);
1201
1202                         int Cb= yuvtab_40cf[U];
1203                         int Cg= yuvtab_1a1e[V] + yuvtab_0c92[U];
1204                         int Cr= yuvtab_3343[V];
1205
1206                         ((uint16_t*)dest)[i] =
1207                                 clip_table16b[(Y1 + Cb) >>13] |
1208                                 clip_table16g[(Y1 + Cg) >>13] |
1209                                 clip_table16r[(Y1 + Cr) >>13];
1210
1211                         ((uint16_t*)dest)[i+1] =
1212                                 clip_table16b[(Y2 + Cb) >>13] |
1213                                 clip_table16g[(Y2 + Cg) >>13] |
1214                                 clip_table16r[(Y2 + Cr) >>13];
1215                 }
1216         }
1217         else if(dstbpp==15)
1218         {
1219                 int i;
1220                 for(i=0; i<dstw-1; i+=2){
1221                         // vertical linear interpolation && yuv2rgb in a single step:
1222                         int Y1=yuvtab_2568[buf0[i]>>7];
1223                         int Y2=yuvtab_2568[buf0[i+1]>>7];
1224                         int U=((uvbuf0[i>>1]*uvalpha1+uvbuf1[i>>1]*uvalpha)>>19);
1225                         int V=((uvbuf0[(i>>1)+2048]*uvalpha1+uvbuf1[(i>>1)+2048]*uvalpha)>>19);
1226
1227                         int Cb= yuvtab_40cf[U];
1228                         int Cg= yuvtab_1a1e[V] + yuvtab_0c92[U];
1229                         int Cr= yuvtab_3343[V];
1230
1231                         ((uint16_t*)dest)[i] =
1232                                 clip_table15b[(Y1 + Cb) >>13] |
1233                                 clip_table15g[(Y1 + Cg) >>13] |
1234                                 clip_table15r[(Y1 + Cr) >>13];
1235
1236                         ((uint16_t*)dest)[i+1] =
1237                                 clip_table15b[(Y2 + Cb) >>13] |
1238                                 clip_table15g[(Y2 + Cg) >>13] |
1239                                 clip_table15r[(Y2 + Cr) >>13];
1240                 }
1241         }
1242 #endif
1243 }
1244
1245
1246 static inline void hyscale(uint16_t *dst, int dstWidth, uint8_t *src, int srcWidth, int xInc)
1247 {
1248       // *** horizontal scale Y line to temp buffer
1249 #ifdef ARCH_X86
1250 #ifdef HAVE_MMX2
1251         int i;
1252         if(canMMX2BeUsed)
1253         {
1254                 asm volatile(
1255                         "pxor %%mm7, %%mm7              \n\t"
1256                         "pxor %%mm2, %%mm2              \n\t" // 2*xalpha
1257                         "movd %5, %%mm6                 \n\t" // xInc&0xFFFF
1258                         "punpcklwd %%mm6, %%mm6         \n\t"
1259                         "punpcklwd %%mm6, %%mm6         \n\t"
1260                         "movq %%mm6, %%mm2              \n\t"
1261                         "psllq $16, %%mm2               \n\t"
1262                         "paddw %%mm6, %%mm2             \n\t"
1263                         "psllq $16, %%mm2               \n\t"
1264                         "paddw %%mm6, %%mm2             \n\t"
1265                         "psllq $16, %%mm2               \n\t" //0,t,2t,3t               t=xInc&0xFF
1266                         "movq %%mm2, temp0              \n\t"
1267                         "movd %4, %%mm6                 \n\t" //(xInc*4)&0xFFFF
1268                         "punpcklwd %%mm6, %%mm6         \n\t"
1269                         "punpcklwd %%mm6, %%mm6         \n\t"
1270                         "xorl %%eax, %%eax              \n\t" // i
1271                         "movl %0, %%esi                 \n\t" // src
1272                         "movl %1, %%edi                 \n\t" // buf1
1273                         "movl %3, %%edx                 \n\t" // (xInc*4)>>16
1274                         "xorl %%ecx, %%ecx              \n\t"
1275                         "xorl %%ebx, %%ebx              \n\t"
1276                         "movw %4, %%bx                  \n\t" // (xInc*4)&0xFFFF
1277
1278 #define FUNNY_Y_CODE \
1279                         PREFETCH" 1024(%%esi)           \n\t"\
1280                         PREFETCH" 1056(%%esi)           \n\t"\
1281                         PREFETCH" 1088(%%esi)           \n\t"\
1282                         "call funnyYCode                \n\t"\
1283                         "movq temp0, %%mm2              \n\t"\
1284                         "xorl %%ecx, %%ecx              \n\t"
1285
1286 FUNNY_Y_CODE
1287 FUNNY_Y_CODE
1288 FUNNY_Y_CODE
1289 FUNNY_Y_CODE
1290 FUNNY_Y_CODE
1291 FUNNY_Y_CODE
1292 FUNNY_Y_CODE
1293 FUNNY_Y_CODE
1294
1295                         :: "m" (src), "m" (dst), "m" (dstWidth), "m" ((xInc*4)>>16),
1296                         "m" ((xInc*4)&0xFFFF), "m" (xInc&0xFFFF)
1297                         : "%eax", "%ebx", "%ecx", "%edx", "%esi", "%edi"
1298                 );
1299                 for(i=dstWidth-1; (i*xInc)>>16 >=srcWidth-1; i--) dst[i] = src[srcWidth-1]*128;
1300         }
1301         else
1302         {
1303 #endif
1304         //NO MMX just normal asm ...
1305         asm volatile(
1306                 "xorl %%eax, %%eax              \n\t" // i
1307                 "xorl %%ebx, %%ebx              \n\t" // xx
1308                 "xorl %%ecx, %%ecx              \n\t" // 2*xalpha
1309                 "1:                             \n\t"
1310                 "movzbl  (%0, %%ebx), %%edi     \n\t" //src[xx]
1311                 "movzbl 1(%0, %%ebx), %%esi     \n\t" //src[xx+1]
1312                 "subl %%edi, %%esi              \n\t" //src[xx+1] - src[xx]
1313                 "imull %%ecx, %%esi             \n\t" //(src[xx+1] - src[xx])*2*xalpha
1314                 "shll $16, %%edi                \n\t"
1315                 "addl %%edi, %%esi              \n\t" //src[xx+1]*2*xalpha + src[xx]*(1-2*xalpha)
1316                 "movl %1, %%edi                 \n\t"
1317                 "shrl $9, %%esi                 \n\t"
1318                 "movw %%si, (%%edi, %%eax, 2)   \n\t"
1319                 "addw %4, %%cx                  \n\t" //2*xalpha += xInc&0xFF
1320                 "adcl %3, %%ebx                 \n\t" //xx+= xInc>>8 + carry
1321
1322                 "movzbl (%0, %%ebx), %%edi      \n\t" //src[xx]
1323                 "movzbl 1(%0, %%ebx), %%esi     \n\t" //src[xx+1]
1324                 "subl %%edi, %%esi              \n\t" //src[xx+1] - src[xx]
1325                 "imull %%ecx, %%esi             \n\t" //(src[xx+1] - src[xx])*2*xalpha
1326                 "shll $16, %%edi                \n\t"
1327                 "addl %%edi, %%esi              \n\t" //src[xx+1]*2*xalpha + src[xx]*(1-2*xalpha)
1328                 "movl %1, %%edi                 \n\t"
1329                 "shrl $9, %%esi                 \n\t"
1330                 "movw %%si, 2(%%edi, %%eax, 2)  \n\t"
1331                 "addw %4, %%cx                  \n\t" //2*xalpha += xInc&0xFF
1332                 "adcl %3, %%ebx                 \n\t" //xx+= xInc>>8 + carry
1333
1334
1335                 "addl $2, %%eax                 \n\t"
1336                 "cmpl %2, %%eax                 \n\t"
1337                 " jb 1b                         \n\t"
1338
1339
1340                 :: "r" (src), "m" (dst), "m" (dstWidth), "m" (xInc>>16), "m" (xInc&0xFFFF)
1341                 : "%eax", "%ebx", "%ecx", "%edi", "%esi"
1342                 );
1343 #ifdef HAVE_MMX2
1344         } //if MMX2 cant be used
1345 #endif
1346 #else
1347         int i;
1348         unsigned int xpos=0;
1349         for(i=0;i<dstWidth;i++)
1350         {
1351                 register unsigned int xx=xpos>>16;
1352                 register unsigned int xalpha=(xpos&0xFFFF)>>9;
1353                 dst[i]= (src[xx]<<7) + (src[xx+1] - src[xx])*xalpha;
1354                 xpos+=xInc;
1355         }
1356 #endif
1357 }
1358
1359 inline static void hcscale(uint16_t *dst, int dstWidth,
1360                                 uint8_t *src1, uint8_t *src2, int srcWidth, int xInc)
1361 {
1362 #ifdef ARCH_X86
1363 #ifdef HAVE_MMX2
1364         int i;
1365         if(canMMX2BeUsed)
1366         {
1367                 asm volatile(
1368                 "pxor %%mm7, %%mm7              \n\t"
1369                 "pxor %%mm2, %%mm2              \n\t" // 2*xalpha
1370                 "movd %5, %%mm6                 \n\t" // xInc&0xFFFF
1371                 "punpcklwd %%mm6, %%mm6         \n\t"
1372                 "punpcklwd %%mm6, %%mm6         \n\t"
1373                 "movq %%mm6, %%mm2              \n\t"
1374                 "psllq $16, %%mm2               \n\t"
1375                 "paddw %%mm6, %%mm2             \n\t"
1376                 "psllq $16, %%mm2               \n\t"
1377                 "paddw %%mm6, %%mm2             \n\t"
1378                 "psllq $16, %%mm2               \n\t" //0,t,2t,3t               t=xInc&0xFFFF
1379                 "movq %%mm2, temp0              \n\t"
1380                 "movd %4, %%mm6                 \n\t" //(xInc*4)&0xFFFF
1381                 "punpcklwd %%mm6, %%mm6         \n\t"
1382                 "punpcklwd %%mm6, %%mm6         \n\t"
1383                 "xorl %%eax, %%eax              \n\t" // i
1384                 "movl %0, %%esi                 \n\t" // src
1385                 "movl %1, %%edi                 \n\t" // buf1
1386                 "movl %3, %%edx                 \n\t" // (xInc*4)>>16
1387                 "xorl %%ecx, %%ecx              \n\t"
1388                 "xorl %%ebx, %%ebx              \n\t"
1389                 "movw %4, %%bx                  \n\t" // (xInc*4)&0xFFFF
1390
1391 #define FUNNYUVCODE \
1392                         PREFETCH" 1024(%%esi)           \n\t"\
1393                         PREFETCH" 1056(%%esi)           \n\t"\
1394                         PREFETCH" 1088(%%esi)           \n\t"\
1395                         "call funnyUVCode               \n\t"\
1396                         "movq temp0, %%mm2              \n\t"\
1397                         "xorl %%ecx, %%ecx              \n\t"
1398
1399 FUNNYUVCODE
1400 FUNNYUVCODE
1401 FUNNYUVCODE
1402 FUNNYUVCODE
1403
1404 FUNNYUVCODE
1405 FUNNYUVCODE
1406 FUNNYUVCODE
1407 FUNNYUVCODE
1408                 "xorl %%eax, %%eax              \n\t" // i
1409                 "movl %6, %%esi                 \n\t" // src
1410                 "movl %1, %%edi                 \n\t" // buf1
1411                 "addl $4096, %%edi              \n\t"
1412
1413 FUNNYUVCODE
1414 FUNNYUVCODE
1415 FUNNYUVCODE
1416 FUNNYUVCODE
1417
1418 FUNNYUVCODE
1419 FUNNYUVCODE
1420 FUNNYUVCODE
1421 FUNNYUVCODE
1422
1423                 :: "m" (src1), "m" (dst), "m" (dstWidth), "m" ((xInc*4)>>16),
1424                   "m" ((xInc*4)&0xFFFF), "m" (xInc&0xFFFF), "m" (src2)
1425                 : "%eax", "%ebx", "%ecx", "%edx", "%esi", "%edi"
1426         );
1427                 for(i=dstWidth-1; (i*xInc)>>16 >=srcWidth/2-1; i--)
1428                 {
1429                         dst[i] = src1[srcWidth/2-1]*128;
1430                         dst[i+2048] = src2[srcWidth/2-1]*128;
1431                 }
1432         }
1433         else
1434         {
1435 #endif
1436         asm volatile(
1437                 "xorl %%eax, %%eax              \n\t" // i
1438                 "xorl %%ebx, %%ebx              \n\t" // xx
1439                 "xorl %%ecx, %%ecx              \n\t" // 2*xalpha
1440                 "1:                             \n\t"
1441                 "movl %0, %%esi                 \n\t"
1442                 "movzbl  (%%esi, %%ebx), %%edi  \n\t" //src[xx]
1443                 "movzbl 1(%%esi, %%ebx), %%esi  \n\t" //src[xx+1]
1444                 "subl %%edi, %%esi              \n\t" //src[xx+1] - src[xx]
1445                 "imull %%ecx, %%esi             \n\t" //(src[xx+1] - src[xx])*2*xalpha
1446                 "shll $16, %%edi                \n\t"
1447                 "addl %%edi, %%esi              \n\t" //src[xx+1]*2*xalpha + src[xx]*(1-2*xalpha)
1448                 "movl %1, %%edi                 \n\t"
1449                 "shrl $9, %%esi                 \n\t"
1450                 "movw %%si, (%%edi, %%eax, 2)   \n\t"
1451
1452                 "movzbl  (%5, %%ebx), %%edi     \n\t" //src[xx]
1453                 "movzbl 1(%5, %%ebx), %%esi     \n\t" //src[xx+1]
1454                 "subl %%edi, %%esi              \n\t" //src[xx+1] - src[xx]
1455                 "imull %%ecx, %%esi             \n\t" //(src[xx+1] - src[xx])*2*xalpha
1456                 "shll $16, %%edi                \n\t"
1457                 "addl %%edi, %%esi              \n\t" //src[xx+1]*2*xalpha + src[xx]*(1-2*xalpha)
1458                 "movl %1, %%edi                 \n\t"
1459                 "shrl $9, %%esi                 \n\t"
1460                 "movw %%si, 4096(%%edi, %%eax, 2)\n\t"
1461
1462                 "addw %4, %%cx                  \n\t" //2*xalpha += xInc&0xFF
1463                 "adcl %3, %%ebx                 \n\t" //xx+= xInc>>8 + carry
1464                 "addl $1, %%eax                 \n\t"
1465                 "cmpl %2, %%eax                 \n\t"
1466                 " jb 1b                         \n\t"
1467
1468                 :: "m" (src1), "m" (dst), "m" (dstWidth), "m" (xInc>>16), "m" (xInc&0xFFFF),
1469                 "r" (src2)
1470                 : "%eax", "%ebx", "%ecx", "%edi", "%esi"
1471                 );
1472 #ifdef HAVE_MMX2
1473         } //if MMX2 cant be used
1474 #endif
1475 #else
1476         int i;
1477         unsigned int xpos=0;
1478         for(i=0;i<dstWidth;i++)
1479         {
1480                 register unsigned int xx=xpos>>16;
1481                 register unsigned int xalpha=(xpos&0xFFFF)>>9;
1482                 dst[i]=(src1[xx]*(xalpha^127)+src1[xx+1]*xalpha);
1483                 dst[i+2048]=(src2[xx]*(xalpha^127)+src2[xx+1]*xalpha);
1484 /* slower
1485           dst[i]= (src1[xx]<<7) + (src1[xx+1] - src1[xx])*xalpha;
1486           dst[i+2048]=(src2[xx]<<7) + (src2[xx+1] - src2[xx])*xalpha;
1487 */
1488                 xpos+=xInc;
1489         }
1490 #endif
1491 }
1492
1493
1494 // *** bilinear scaling and yuv->rgb or yuv->yuv conversion of yv12 slices:
1495 // *** Note: it's called multiple times while decoding a frame, first time y==0
1496 // *** Designed to upscale, but may work for downscale too.
1497 // s_xinc = (src_width << 16) / dst_width
1498 // s_yinc = (src_height << 16) / dst_height
1499 void SwScale_YV12slice(unsigned char* srcptr[],int stride[], int y, int h,
1500                              uint8_t* dstptr[], int dststride, int dstw, int dstbpp,
1501                              unsigned int s_xinc,unsigned int s_yinc){
1502
1503 // scaling factors:
1504 //static int s_yinc=(vo_dga_src_height<<16)/vo_dga_vp_height;
1505 //static int s_xinc=(vo_dga_src_width<<8)/vo_dga_vp_width;
1506
1507 unsigned int s_xinc2;
1508
1509 static int s_srcypos; // points to the dst Pixels center in the source (0 is the center of pixel 0,0 in src)
1510 static int s_ypos;
1511
1512 // last horzontally interpolated lines, used to avoid unnecessary calculations
1513 static int s_last_ypos;
1514 static int s_last_y1pos;
1515
1516 #ifdef HAVE_MMX2
1517 // used to detect a horizontal size change
1518 static int old_dstw= -1;
1519 static int old_s_xinc= -1;
1520 #endif
1521
1522 int srcWidth;
1523 int dstUVw;
1524 int i;
1525
1526 if(((dstw + 7)&(~7)) >= dststride) dstw&= ~7;
1527
1528 srcWidth= (dstw*s_xinc + 0x8000)>>16;
1529 dstUVw= fullUVIpol ? dstw : dstw/2;
1530
1531 #ifdef HAVE_MMX2
1532 canMMX2BeUsed= (s_xinc <= 0x10000 && (dstw&31)==0 && (srcWidth&15)==0) ? 1 : 0;
1533 #endif
1534
1535 // match pixel 0 of the src to pixel 0 of dst and match pixel n-2 of src to pixel n-2 of dst
1536 // n-2 is the last chrominance sample available
1537 // FIXME this is not perfect, but noone shuld notice the difference, the more correct variant
1538 // would be like the vertical one, but that would require some special code for the
1539 // first and last pixel
1540 if(canMMX2BeUsed)       s_xinc+= 20;
1541 else                    s_xinc = ((srcWidth-2)<<16)/(dstw-2) - 20;
1542
1543 if(fullUVIpol && !(dstbpp==12))         s_xinc2= s_xinc>>1;
1544 else                                    s_xinc2= s_xinc;
1545   // force calculation of the horizontal interpolation of the first line
1546
1547   if(y==0){
1548 //      printf("dstw %d, srcw %d, mmx2 %d\n", dstw, srcWidth, canMMX2BeUsed);
1549         s_last_ypos=-99;
1550         s_last_y1pos=-99;
1551         s_srcypos= s_yinc/2 - 0x8000;
1552         s_ypos=0;
1553
1554         // clean the buffers so that no green stuff is drawen if the width is not sane (%8=0)
1555         for(i=dstw-2; i<dstw+20; i++)
1556         {
1557                 pix_buf_uv[0][i] = pix_buf_uv[1][i]
1558                 = pix_buf_uv[0][2048+i] = pix_buf_uv[1][2048+i] = 128*128;
1559                 pix_buf_uv[0][i/2] = pix_buf_uv[1][i/2]
1560                 = pix_buf_uv[0][2048+i/2] = pix_buf_uv[1][2048+i/2] = 128*128;
1561                 pix_buf_y[0][i]= pix_buf_y[1][i]= 0;
1562         }
1563
1564 #ifdef HAVE_MMX2
1565 // cant downscale !!!
1566         if((old_s_xinc != s_xinc || old_dstw!=dstw) && canMMX2BeUsed)
1567         {
1568                 uint8_t *fragment;
1569                 int imm8OfPShufW1;
1570                 int imm8OfPShufW2;
1571                 int fragmentLength;
1572
1573                 int xpos, i;
1574
1575                 old_s_xinc= s_xinc;
1576                 old_dstw= dstw;
1577
1578                 // create an optimized horizontal scaling routine
1579
1580                 //code fragment
1581
1582                 asm volatile(
1583                         "jmp 9f                         \n\t"
1584                 // Begin
1585                         "0:                             \n\t"
1586                         "movq (%%esi), %%mm0            \n\t" //FIXME Alignment
1587                         "movq %%mm0, %%mm1              \n\t"
1588                         "psrlq $8, %%mm0                \n\t"
1589                         "punpcklbw %%mm7, %%mm1 \n\t"
1590                         "movq %%mm2, %%mm3              \n\t"
1591                         "punpcklbw %%mm7, %%mm0 \n\t"
1592                         "addw %%bx, %%cx                \n\t" //2*xalpha += (4*s_xinc)&0xFFFF
1593                         "pshufw $0xFF, %%mm1, %%mm1     \n\t"
1594                         "1:                             \n\t"
1595                         "adcl %%edx, %%esi              \n\t" //xx+= (4*s_xinc)>>16 + carry
1596                         "pshufw $0xFF, %%mm0, %%mm0     \n\t"
1597                         "2:                             \n\t"
1598                         "psrlw $9, %%mm3                \n\t"
1599                         "psubw %%mm1, %%mm0             \n\t"
1600                         "pmullw %%mm3, %%mm0            \n\t"
1601                         "paddw %%mm6, %%mm2             \n\t" // 2*alpha += xpos&0xFFFF
1602                         "psllw $7, %%mm1                \n\t"
1603                         "paddw %%mm1, %%mm0             \n\t"
1604
1605                         "movq %%mm0, (%%edi, %%eax)     \n\t"
1606
1607                         "addl $8, %%eax                 \n\t"
1608                 // End
1609                         "9:                             \n\t"
1610 //              "int $3\n\t"
1611                         "leal 0b, %0                    \n\t"
1612                         "leal 1b, %1                    \n\t"
1613                         "leal 2b, %2                    \n\t"
1614                         "decl %1                        \n\t"
1615                         "decl %2                        \n\t"
1616                         "subl %0, %1                    \n\t"
1617                         "subl %0, %2                    \n\t"
1618                         "leal 9b, %3                    \n\t"
1619                         "subl %0, %3                    \n\t"
1620                         :"=r" (fragment), "=r" (imm8OfPShufW1), "=r" (imm8OfPShufW2),
1621                          "=r" (fragmentLength)
1622                 );
1623
1624                 xpos= 0; //s_xinc/2 - 0x8000; // difference between pixel centers
1625
1626                 /* choose xinc so that all 8 parts fit exactly
1627                    Note: we cannot use just 1 part because it would not fit in the code cache */
1628 //              s_xinc2_diff= -((((s_xinc2*(dstw/8))&0xFFFF))/(dstw/8))-10;
1629 //              s_xinc_diff= -((((s_xinc*(dstw/8))&0xFFFF))/(dstw/8));
1630 #ifdef ALT_ERROR
1631 //              s_xinc2_diff+= ((0x10000/(dstw/8)));
1632 #endif
1633 //              s_xinc_diff= s_xinc2_diff*2;
1634
1635 //              s_xinc2+= s_xinc2_diff;
1636 //              s_xinc+= s_xinc_diff;
1637
1638 //              old_s_xinc= s_xinc;
1639
1640                 for(i=0; i<dstw/8; i++)
1641                 {
1642                         int xx=xpos>>16;
1643
1644                         if((i&3) == 0)
1645                         {
1646                                 int a=0;
1647                                 int b=((xpos+s_xinc)>>16) - xx;
1648                                 int c=((xpos+s_xinc*2)>>16) - xx;
1649                                 int d=((xpos+s_xinc*3)>>16) - xx;
1650
1651                                 memcpy(funnyYCode + fragmentLength*i/4, fragment, fragmentLength);
1652
1653                                 funnyYCode[fragmentLength*i/4 + imm8OfPShufW1]=
1654                                 funnyYCode[fragmentLength*i/4 + imm8OfPShufW2]=
1655                                         a | (b<<2) | (c<<4) | (d<<6);
1656
1657                                 // if we dont need to read 8 bytes than dont :), reduces the chance of
1658                                 // crossing a cache line
1659                                 if(d<3) funnyYCode[fragmentLength*i/4 + 1]= 0x6E;
1660
1661                                 funnyYCode[fragmentLength*(i+4)/4]= RET;
1662                         }
1663                         xpos+=s_xinc;
1664                 }
1665
1666                 xpos= 0; //s_xinc2/2 - 0x10000; // difference between centers of chrom samples
1667                 for(i=0; i<dstUVw/8; i++)
1668                 {
1669                         int xx=xpos>>16;
1670
1671                         if((i&3) == 0)
1672                         {
1673                                 int a=0;
1674                                 int b=((xpos+s_xinc2)>>16) - xx;
1675                                 int c=((xpos+s_xinc2*2)>>16) - xx;
1676                                 int d=((xpos+s_xinc2*3)>>16) - xx;
1677
1678                                 memcpy(funnyUVCode + fragmentLength*i/4, fragment, fragmentLength);
1679
1680                                 funnyUVCode[fragmentLength*i/4 + imm8OfPShufW1]=
1681                                 funnyUVCode[fragmentLength*i/4 + imm8OfPShufW2]=
1682                                         a | (b<<2) | (c<<4) | (d<<6);
1683
1684                                 // if we dont need to read 8 bytes than dont :), reduces the chance of
1685                                 // crossing a cache line
1686                                 if(d<3) funnyUVCode[fragmentLength*i/4 + 1]= 0x6E;
1687
1688                                 funnyUVCode[fragmentLength*(i+4)/4]= RET;
1689                         }
1690                         xpos+=s_xinc2;
1691                 }
1692 //              funnyCode[0]= RET;
1693         }
1694
1695 #endif // HAVE_MMX2
1696   } // reset counters
1697
1698   while(1){
1699     unsigned char *dest =dstptr[0]+dststride*s_ypos;
1700     unsigned char *uDest=dstptr[1]+(dststride>>1)*(s_ypos>>1);
1701     unsigned char *vDest=dstptr[2]+(dststride>>1)*(s_ypos>>1);
1702
1703     int y0=(s_srcypos + 0xFFFF)>>16;  // first luminance source line number below the dst line
1704         // points to the dst Pixels center in the source (0 is the center of pixel 0,0 in src)
1705     int srcuvpos= dstbpp==12 ?  s_srcypos + s_yinc/2 - 0x8000 :
1706                                 s_srcypos - 0x8000;
1707     int y1=(srcuvpos + 0x1FFFF)>>17; // first chrominance source line number below the dst line
1708     int yalpha=((s_srcypos-1)&0xFFFF)>>4;
1709     int uvalpha=((srcuvpos-1)&0x1FFFF)>>5;
1710     uint16_t *buf0=pix_buf_y[y0&1];             // top line of the interpolated slice
1711     uint16_t *buf1=pix_buf_y[((y0+1)&1)];       // bottom line of the interpolated slice
1712     uint16_t *uvbuf0=pix_buf_uv[y1&1];          // top line of the interpolated slice
1713     uint16_t *uvbuf1=pix_buf_uv[(y1+1)&1];      // bottom line of the interpolated slice
1714
1715     if(y0>=y+h) break; // FIXME wrong, skips last lines, but they are dupliactes anyway
1716
1717     if((y0&1) && dstbpp==12) uvalpha=-1; // there is no alpha if there is no line
1718
1719     s_ypos++; s_srcypos+=s_yinc;
1720
1721     //only interpolate the src line horizontally if we didnt do it allready
1722         if(s_last_ypos!=y0)
1723         {
1724                 unsigned char *src;
1725                 // skip if first line has been horiz scaled alleady
1726                 if(s_last_ypos != y0-1)
1727                 {
1728                         // check if first line is before any available src lines
1729                         if(y0-1 < y)    src=srcptr[0]+(0     )*stride[0];
1730                         else            src=srcptr[0]+(y0-y-1)*stride[0];
1731
1732                         hyscale(buf0, dstw, src, srcWidth, s_xinc);
1733                 }
1734                 // check if second line is after any available src lines
1735                 if(y0-y >= h)   src=srcptr[0]+(h-1)*stride[0];
1736                 else            src=srcptr[0]+(y0-y)*stride[0];
1737
1738                 // the min() is required to avoid reuseing lines which where not available
1739                 s_last_ypos= MIN(y0, y+h-1);
1740                 hyscale(buf1, dstw, src, srcWidth, s_xinc);
1741         }
1742 //      printf("%d %d %d %d\n", y, y1, s_last_y1pos, h);
1743       // *** horizontal scale U and V lines to temp buffer
1744         if(s_last_y1pos!=y1)
1745         {
1746                 uint8_t *src1, *src2;
1747                 // skip if first line has been horiz scaled alleady
1748                 if(s_last_y1pos != y1-1)
1749                 {
1750                         // check if first line is before any available src lines
1751                         if(y1-y/2-1 < 0)
1752                         {
1753                                 src1= srcptr[1]+(0)*stride[1];
1754                                 src2= srcptr[2]+(0)*stride[2];
1755                         }else{
1756                                 src1= srcptr[1]+(y1-y/2-1)*stride[1];
1757                                 src2= srcptr[2]+(y1-y/2-1)*stride[2];
1758                         }
1759                         hcscale(uvbuf0, dstUVw, src1, src2, srcWidth, s_xinc2);
1760                 }
1761
1762                 // check if second line is after any available src lines
1763                 if(y1 - y/2 >= h/2)
1764                 {
1765                         src1= srcptr[1]+(h/2-1)*stride[1];
1766                         src2= srcptr[2]+(h/2-1)*stride[2];
1767                 }else{
1768                         src1= srcptr[1]+(y1-y/2)*stride[1];
1769                         src2= srcptr[2]+(y1-y/2)*stride[2];
1770                 }
1771                 hcscale(uvbuf1, dstUVw, src1, src2, srcWidth, s_xinc2);
1772
1773                 // the min() is required to avoid reuseing lines which where not available
1774                 s_last_y1pos= MIN(y1, y/2+h/2-1);
1775         }
1776 #ifdef HAVE_MMX
1777         b5Dither= dither8[s_ypos&1];
1778         g6Dither= dither4[s_ypos&1];
1779         g5Dither= dither8[s_ypos&1];
1780         r5Dither= dither8[(s_ypos+1)&1];
1781 #endif
1782
1783         if(dstbpp==12) //YV12
1784                 yuv2yuv(buf0, buf1, uvbuf0, uvbuf1, dest, uDest, vDest, dstw, yalpha, uvalpha);
1785         else if(ABS(s_yinc - 0x10000) < 10)
1786                 yuv2rgb1(buf0, buf1, uvbuf0, uvbuf1, dest, dstw, yalpha, uvalpha, dstbpp);
1787         else
1788                 yuv2rgbX(buf0, buf1, uvbuf0, uvbuf1, dest, dstw, yalpha, uvalpha, dstbpp);
1789   }
1790
1791 #ifdef HAVE_MMX
1792         __asm __volatile(SFENCE:::"memory");
1793         __asm __volatile(EMMS:::"memory");
1794 #endif
1795 }
1796
1797
1798 void SwScale_Init(){
1799     // generating tables:
1800     int i;
1801     for(i=0;i<256;i++){
1802         clip_table[i]=0;
1803         clip_table[i+256]=i;
1804         clip_table[i+512]=255;
1805         yuvtab_2568[i]=(0x2568*(i-16))+(256<<13);
1806         yuvtab_3343[i]=0x3343*(i-128);
1807         yuvtab_0c92[i]=-0x0c92*(i-128);
1808         yuvtab_1a1e[i]=-0x1a1e*(i-128);
1809         yuvtab_40cf[i]=0x40cf*(i-128);
1810     }
1811
1812     for(i=0; i<768; i++)
1813     {
1814         int v= clip_table[i];
1815         clip_table16b[i]= v>>3;
1816         clip_table16g[i]= (v<<3)&0x07E0;
1817         clip_table16r[i]= (v<<8)&0xF800;
1818         clip_table15b[i]= v>>3;
1819         clip_table15g[i]= (v<<2)&0x03E0;
1820         clip_table15r[i]= (v<<7)&0x7C00;
1821     }
1822
1823 }