faster bgr24 output
[ffmpeg.git] / postproc / swscale.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))) b16Dither= 0x0004000400040004LL;
75 static uint64_t __attribute__((aligned(8))) b16Dither1=0x0004000400040004LL;
76 static uint64_t __attribute__((aligned(8))) b16Dither2=0x0602060206020602LL;
77 static uint64_t __attribute__((aligned(8))) g16Dither= 0x0002000200020002LL;
78 static uint64_t __attribute__((aligned(8))) g16Dither1=0x0002000200020002LL;
79 static uint64_t __attribute__((aligned(8))) g16Dither2=0x0301030103010301LL;
80
81 static uint64_t __attribute__((aligned(8))) b16Mask=   0x001F001F001F001FLL;
82 static uint64_t __attribute__((aligned(8))) g16Mask=   0x07E007E007E007E0LL;
83 static uint64_t __attribute__((aligned(8))) r16Mask=   0xF800F800F800F800LL;
84 static uint64_t __attribute__((aligned(8))) b15Mask=   0x001F001F001F001FLL;
85 static uint64_t __attribute__((aligned(8))) g15Mask=   0x03E003E003E003E0LL;
86 static uint64_t __attribute__((aligned(8))) r15Mask=   0x7C007C007C007C00LL;
87
88 static uint64_t __attribute__((aligned(8))) M24A=   0x00FF0000FF0000FFLL;
89 static uint64_t __attribute__((aligned(8))) M24B=   0xFF0000FF0000FF00LL;
90 static uint64_t __attribute__((aligned(8))) M24C=   0x0000FF0000FF0000LL;
91
92 static uint64_t __attribute__((aligned(8))) temp0;
93 static uint64_t __attribute__((aligned(8))) asm_yalpha1;
94 static uint64_t __attribute__((aligned(8))) asm_uvalpha1;
95 #endif
96
97 // temporary storage for 4 yuv lines:
98 // 16bit for now (mmx likes it more compact)
99 #ifdef HAVE_MMX
100 static uint16_t __attribute__((aligned(8))) pix_buf_y[4][2048];
101 static uint16_t __attribute__((aligned(8))) pix_buf_uv[2][2048*2];
102 #else
103 static uint16_t pix_buf_y[4][2048];
104 static uint16_t pix_buf_uv[2][2048*2];
105 #endif
106
107 // clipping helper table for C implementations:
108 static unsigned char clip_table[768];
109
110 static unsigned short clip_table16b[768];
111 static unsigned short clip_table16g[768];
112 static unsigned short clip_table16r[768];
113 static unsigned short clip_table15b[768];
114 static unsigned short clip_table15g[768];
115 static unsigned short clip_table15r[768];
116
117 // yuv->rgb conversion tables:
118 static    int yuvtab_2568[256];
119 static    int yuvtab_3343[256];
120 static    int yuvtab_0c92[256];
121 static    int yuvtab_1a1e[256];
122 static    int yuvtab_40cf[256];
123
124 #ifdef HAVE_MMX2
125 static uint8_t funnyYCode[10000];
126 static uint8_t funnyUVCode[10000];
127 #endif
128
129 static int canMMX2BeUsed=0;
130
131 #define FULL_YSCALEYUV2RGB \
132                 "pxor %%mm7, %%mm7              \n\t"\
133                 "movd %6, %%mm6                 \n\t" /*yalpha1*/\
134                 "punpcklwd %%mm6, %%mm6         \n\t"\
135                 "punpcklwd %%mm6, %%mm6         \n\t"\
136                 "movd %7, %%mm5                 \n\t" /*uvalpha1*/\
137                 "punpcklwd %%mm5, %%mm5         \n\t"\
138                 "punpcklwd %%mm5, %%mm5         \n\t"\
139                 "xorl %%eax, %%eax              \n\t"\
140                 "1:                             \n\t"\
141                 "movq (%0, %%eax, 2), %%mm0     \n\t" /*buf0[eax]*/\
142                 "movq (%1, %%eax, 2), %%mm1     \n\t" /*buf1[eax]*/\
143                 "movq (%2, %%eax,2), %%mm2      \n\t" /* uvbuf0[eax]*/\
144                 "movq (%3, %%eax,2), %%mm3      \n\t" /* uvbuf1[eax]*/\
145                 "psubw %%mm1, %%mm0             \n\t" /* buf0[eax] - buf1[eax]*/\
146                 "psubw %%mm3, %%mm2             \n\t" /* uvbuf0[eax] - uvbuf1[eax]*/\
147                 "pmulhw %%mm6, %%mm0            \n\t" /* (buf0[eax] - buf1[eax])yalpha1>>16*/\
148                 "pmulhw %%mm5, %%mm2            \n\t" /* (uvbuf0[eax] - uvbuf1[eax])uvalpha1>>16*/\
149                 "psraw $4, %%mm1                \n\t" /* buf0[eax] - buf1[eax] >>4*/\
150                 "movq 4096(%2, %%eax,2), %%mm4  \n\t" /* uvbuf0[eax+2048]*/\
151                 "psraw $4, %%mm3                \n\t" /* uvbuf0[eax] - uvbuf1[eax] >>4*/\
152                 "paddw %%mm0, %%mm1             \n\t" /* buf0[eax]yalpha1 + buf1[eax](1-yalpha1) >>16*/\
153                 "movq 4096(%3, %%eax,2), %%mm0  \n\t" /* uvbuf1[eax+2048]*/\
154                 "paddw %%mm2, %%mm3             \n\t" /* uvbuf0[eax]uvalpha1 - uvbuf1[eax](1-uvalpha1)*/\
155                 "psubw %%mm0, %%mm4             \n\t" /* uvbuf0[eax+2048] - uvbuf1[eax+2048]*/\
156                 "psubw w80, %%mm1               \n\t" /* 8(Y-16)*/\
157                 "psubw w400, %%mm3              \n\t" /* 8(U-128)*/\
158                 "pmulhw yCoeff, %%mm1           \n\t"\
159 \
160 \
161                 "pmulhw %%mm5, %%mm4            \n\t" /* (uvbuf0[eax+2048] - uvbuf1[eax+2048])uvalpha1>>16*/\
162                 "movq %%mm3, %%mm2              \n\t" /* (U-128)8*/\
163                 "pmulhw ubCoeff, %%mm3          \n\t"\
164                 "psraw $4, %%mm0                \n\t" /* uvbuf0[eax+2048] - uvbuf1[eax+2048] >>4*/\
165                 "pmulhw ugCoeff, %%mm2          \n\t"\
166                 "paddw %%mm4, %%mm0             \n\t" /* uvbuf0[eax+2048]uvalpha1 - uvbuf1[eax+2048](1-uvalpha1)*/\
167                 "psubw w400, %%mm0              \n\t" /* (V-128)8*/\
168 \
169 \
170                 "movq %%mm0, %%mm4              \n\t" /* (V-128)8*/\
171                 "pmulhw vrCoeff, %%mm0          \n\t"\
172                 "pmulhw vgCoeff, %%mm4          \n\t"\
173                 "paddw %%mm1, %%mm3             \n\t" /* B*/\
174                 "paddw %%mm1, %%mm0             \n\t" /* R*/\
175                 "packuswb %%mm3, %%mm3          \n\t"\
176 \
177                 "packuswb %%mm0, %%mm0          \n\t"\
178                 "paddw %%mm4, %%mm2             \n\t"\
179                 "paddw %%mm2, %%mm1             \n\t" /* G*/\
180 \
181                 "packuswb %%mm1, %%mm1          \n\t"
182
183 #define YSCALEYUV2RGB \
184                 "movd %6, %%mm6                 \n\t" /*yalpha1*/\
185                 "punpcklwd %%mm6, %%mm6         \n\t"\
186                 "punpcklwd %%mm6, %%mm6         \n\t"\
187                 "movq %%mm6, asm_yalpha1        \n\t"\
188                 "movd %7, %%mm5                 \n\t" /*uvalpha1*/\
189                 "punpcklwd %%mm5, %%mm5         \n\t"\
190                 "punpcklwd %%mm5, %%mm5         \n\t"\
191                 "movq %%mm5, asm_uvalpha1       \n\t"\
192                 "xorl %%eax, %%eax              \n\t"\
193                 "1:                             \n\t"\
194                 "movq (%2, %%eax), %%mm2        \n\t" /* uvbuf0[eax]*/\
195                 "movq (%3, %%eax), %%mm3        \n\t" /* uvbuf1[eax]*/\
196                 "movq 4096(%2, %%eax), %%mm5    \n\t" /* uvbuf0[eax+2048]*/\
197                 "movq 4096(%3, %%eax), %%mm4    \n\t" /* uvbuf1[eax+2048]*/\
198                 "psubw %%mm3, %%mm2             \n\t" /* uvbuf0[eax] - uvbuf1[eax]*/\
199                 "psubw %%mm4, %%mm5             \n\t" /* uvbuf0[eax+2048] - uvbuf1[eax+2048]*/\
200                 "movq asm_uvalpha1, %%mm0       \n\t"\
201                 "pmulhw %%mm0, %%mm2            \n\t" /* (uvbuf0[eax] - uvbuf1[eax])uvalpha1>>16*/\
202                 "pmulhw %%mm0, %%mm5            \n\t" /* (uvbuf0[eax+2048] - uvbuf1[eax+2048])uvalpha1>>16*/\
203                 "psraw $4, %%mm3                \n\t" /* uvbuf0[eax] - uvbuf1[eax] >>4*/\
204                 "psraw $4, %%mm4                \n\t" /* uvbuf0[eax+2048] - uvbuf1[eax+2048] >>4*/\
205                 "paddw %%mm2, %%mm3             \n\t" /* uvbuf0[eax]uvalpha1 - uvbuf1[eax](1-uvalpha1)*/\
206                 "paddw %%mm5, %%mm4             \n\t" /* uvbuf0[eax+2048]uvalpha1 - uvbuf1[eax+2048](1-uvalpha1)*/\
207                 "psubw w400, %%mm3              \n\t" /* (U-128)8*/\
208                 "psubw w400, %%mm4              \n\t" /* (V-128)8*/\
209                 "movq %%mm3, %%mm2              \n\t" /* (U-128)8*/\
210                 "movq %%mm4, %%mm5              \n\t" /* (V-128)8*/\
211                 "pmulhw ugCoeff, %%mm3          \n\t"\
212                 "pmulhw vgCoeff, %%mm4          \n\t"\
213         /* mm2=(U-128)8, mm3=ug, mm4=vg mm5=(V-128)8 */\
214                 "movq (%0, %%eax, 2), %%mm0     \n\t" /*buf0[eax]*/\
215                 "movq (%1, %%eax, 2), %%mm1     \n\t" /*buf1[eax]*/\
216                 "movq 8(%0, %%eax, 2), %%mm6    \n\t" /*buf0[eax]*/\
217                 "movq 8(%1, %%eax, 2), %%mm7    \n\t" /*buf1[eax]*/\
218                 "psubw %%mm1, %%mm0             \n\t" /* buf0[eax] - buf1[eax]*/\
219                 "psubw %%mm7, %%mm6             \n\t" /* buf0[eax] - buf1[eax]*/\
220                 "pmulhw asm_yalpha1, %%mm0      \n\t" /* (buf0[eax] - buf1[eax])yalpha1>>16*/\
221                 "pmulhw asm_yalpha1, %%mm6      \n\t" /* (buf0[eax] - buf1[eax])yalpha1>>16*/\
222                 "psraw $4, %%mm1                \n\t" /* buf0[eax] - buf1[eax] >>4*/\
223                 "psraw $4, %%mm7                \n\t" /* buf0[eax] - buf1[eax] >>4*/\
224                 "paddw %%mm0, %%mm1             \n\t" /* buf0[eax]yalpha1 + buf1[eax](1-yalpha1) >>16*/\
225                 "paddw %%mm6, %%mm7             \n\t" /* buf0[eax]yalpha1 + buf1[eax](1-yalpha1) >>16*/\
226                 "pmulhw ubCoeff, %%mm2          \n\t"\
227                 "pmulhw vrCoeff, %%mm5          \n\t"\
228                 "psubw w80, %%mm1               \n\t" /* 8(Y-16)*/\
229                 "psubw w80, %%mm7               \n\t" /* 8(Y-16)*/\
230                 "pmulhw yCoeff, %%mm1           \n\t"\
231                 "pmulhw yCoeff, %%mm7           \n\t"\
232         /* mm1= Y1, mm2=ub, mm3=ug, mm4=vg mm5=vr, mm7=Y2 */\
233                 "paddw %%mm3, %%mm4             \n\t"\
234                 "movq %%mm2, %%mm0              \n\t"\
235                 "movq %%mm5, %%mm6              \n\t"\
236                 "movq %%mm4, %%mm3              \n\t"\
237                 "punpcklwd %%mm2, %%mm2         \n\t"\
238                 "punpcklwd %%mm5, %%mm5         \n\t"\
239                 "punpcklwd %%mm4, %%mm4         \n\t"\
240                 "paddw %%mm1, %%mm2             \n\t"\
241                 "paddw %%mm1, %%mm5             \n\t"\
242                 "paddw %%mm1, %%mm4             \n\t"\
243                 "punpckhwd %%mm0, %%mm0         \n\t"\
244                 "punpckhwd %%mm6, %%mm6         \n\t"\
245                 "punpckhwd %%mm3, %%mm3         \n\t"\
246                 "paddw %%mm7, %%mm0             \n\t"\
247                 "paddw %%mm7, %%mm6             \n\t"\
248                 "paddw %%mm7, %%mm3             \n\t"\
249                 /* mm0=B1, mm2=B2, mm3=G2, mm4=G1, mm5=R1, mm6=R2 */\
250                 "packuswb %%mm0, %%mm2          \n\t"\
251                 "packuswb %%mm6, %%mm5          \n\t"\
252                 "packuswb %%mm3, %%mm4          \n\t"\
253                 "pxor %%mm7, %%mm7              \n\t"
254
255 #define YSCALEYUV2RGB1 \
256                 "xorl %%eax, %%eax              \n\t"\
257                 "1:                             \n\t"\
258                 "movq (%2, %%eax), %%mm3        \n\t" /* uvbuf0[eax]*/\
259                 "movq 4096(%2, %%eax), %%mm4    \n\t" /* uvbuf0[eax+2048]*/\
260                 "psraw $4, %%mm3                \n\t" /* uvbuf0[eax] - uvbuf1[eax] >>4*/\
261                 "psraw $4, %%mm4                \n\t" /* uvbuf0[eax+2048] - uvbuf1[eax+2048] >>4*/\
262                 "psubw w400, %%mm3              \n\t" /* (U-128)8*/\
263                 "psubw w400, %%mm4              \n\t" /* (V-128)8*/\
264                 "movq %%mm3, %%mm2              \n\t" /* (U-128)8*/\
265                 "movq %%mm4, %%mm5              \n\t" /* (V-128)8*/\
266                 "pmulhw ugCoeff, %%mm3          \n\t"\
267                 "pmulhw vgCoeff, %%mm4          \n\t"\
268         /* mm2=(U-128)8, mm3=ug, mm4=vg mm5=(V-128)8 */\
269                 "movq (%0, %%eax, 2), %%mm1     \n\t" /*buf0[eax]*/\
270                 "movq 8(%0, %%eax, 2), %%mm7    \n\t" /*buf0[eax]*/\
271                 "psraw $4, %%mm1                \n\t" /* buf0[eax] - buf1[eax] >>4*/\
272                 "psraw $4, %%mm7                \n\t" /* buf0[eax] - buf1[eax] >>4*/\
273                 "pmulhw ubCoeff, %%mm2          \n\t"\
274                 "pmulhw vrCoeff, %%mm5          \n\t"\
275                 "psubw w80, %%mm1               \n\t" /* 8(Y-16)*/\
276                 "psubw w80, %%mm7               \n\t" /* 8(Y-16)*/\
277                 "pmulhw yCoeff, %%mm1           \n\t"\
278                 "pmulhw yCoeff, %%mm7           \n\t"\
279         /* mm1= Y1, mm2=ub, mm3=ug, mm4=vg mm5=vr, mm7=Y2 */\
280                 "paddw %%mm3, %%mm4             \n\t"\
281                 "movq %%mm2, %%mm0              \n\t"\
282                 "movq %%mm5, %%mm6              \n\t"\
283                 "movq %%mm4, %%mm3              \n\t"\
284                 "punpcklwd %%mm2, %%mm2         \n\t"\
285                 "punpcklwd %%mm5, %%mm5         \n\t"\
286                 "punpcklwd %%mm4, %%mm4         \n\t"\
287                 "paddw %%mm1, %%mm2             \n\t"\
288                 "paddw %%mm1, %%mm5             \n\t"\
289                 "paddw %%mm1, %%mm4             \n\t"\
290                 "punpckhwd %%mm0, %%mm0         \n\t"\
291                 "punpckhwd %%mm6, %%mm6         \n\t"\
292                 "punpckhwd %%mm3, %%mm3         \n\t"\
293                 "paddw %%mm7, %%mm0             \n\t"\
294                 "paddw %%mm7, %%mm6             \n\t"\
295                 "paddw %%mm7, %%mm3             \n\t"\
296                 /* mm0=B1, mm2=B2, mm3=G2, mm4=G1, mm5=R1, mm6=R2 */\
297                 "packuswb %%mm0, %%mm2          \n\t"\
298                 "packuswb %%mm6, %%mm5          \n\t"\
299                 "packuswb %%mm3, %%mm4          \n\t"\
300                 "pxor %%mm7, %%mm7              \n\t"
301
302 // do vertical chrominance interpolation
303 #define YSCALEYUV2RGB1b \
304                 "xorl %%eax, %%eax              \n\t"\
305                 "1:                             \n\t"\
306                 "movq (%2, %%eax), %%mm2        \n\t" /* uvbuf0[eax]*/\
307                 "movq (%3, %%eax), %%mm3        \n\t" /* uvbuf1[eax]*/\
308                 "movq 4096(%2, %%eax), %%mm5    \n\t" /* uvbuf0[eax+2048]*/\
309                 "movq 4096(%3, %%eax), %%mm4    \n\t" /* uvbuf1[eax+2048]*/\
310                 "paddw %%mm2, %%mm3             \n\t" /* uvbuf0[eax] + uvbuf1[eax]*/\
311                 "paddw %%mm5, %%mm4             \n\t" /* uvbuf0[eax+2048] + uvbuf1[eax+2048]*/\
312                 "psrlw $5, %%mm3                \n\t"\
313                 "psrlw $5, %%mm4                \n\t"\
314                 "psubw w400, %%mm3              \n\t" /* (U-128)8*/\
315                 "psubw w400, %%mm4              \n\t" /* (V-128)8*/\
316                 "movq %%mm3, %%mm2              \n\t" /* (U-128)8*/\
317                 "movq %%mm4, %%mm5              \n\t" /* (V-128)8*/\
318                 "pmulhw ugCoeff, %%mm3          \n\t"\
319                 "pmulhw vgCoeff, %%mm4          \n\t"\
320         /* mm2=(U-128)8, mm3=ug, mm4=vg mm5=(V-128)8 */\
321                 "movq (%0, %%eax, 2), %%mm1     \n\t" /*buf0[eax]*/\
322                 "movq 8(%0, %%eax, 2), %%mm7    \n\t" /*buf0[eax]*/\
323                 "psraw $4, %%mm1                \n\t" /* buf0[eax] - buf1[eax] >>4*/\
324                 "psraw $4, %%mm7                \n\t" /* buf0[eax] - buf1[eax] >>4*/\
325                 "pmulhw ubCoeff, %%mm2          \n\t"\
326                 "pmulhw vrCoeff, %%mm5          \n\t"\
327                 "psubw w80, %%mm1               \n\t" /* 8(Y-16)*/\
328                 "psubw w80, %%mm7               \n\t" /* 8(Y-16)*/\
329                 "pmulhw yCoeff, %%mm1           \n\t"\
330                 "pmulhw yCoeff, %%mm7           \n\t"\
331         /* mm1= Y1, mm2=ub, mm3=ug, mm4=vg mm5=vr, mm7=Y2 */\
332                 "paddw %%mm3, %%mm4             \n\t"\
333                 "movq %%mm2, %%mm0              \n\t"\
334                 "movq %%mm5, %%mm6              \n\t"\
335                 "movq %%mm4, %%mm3              \n\t"\
336                 "punpcklwd %%mm2, %%mm2         \n\t"\
337                 "punpcklwd %%mm5, %%mm5         \n\t"\
338                 "punpcklwd %%mm4, %%mm4         \n\t"\
339                 "paddw %%mm1, %%mm2             \n\t"\
340                 "paddw %%mm1, %%mm5             \n\t"\
341                 "paddw %%mm1, %%mm4             \n\t"\
342                 "punpckhwd %%mm0, %%mm0         \n\t"\
343                 "punpckhwd %%mm6, %%mm6         \n\t"\
344                 "punpckhwd %%mm3, %%mm3         \n\t"\
345                 "paddw %%mm7, %%mm0             \n\t"\
346                 "paddw %%mm7, %%mm6             \n\t"\
347                 "paddw %%mm7, %%mm3             \n\t"\
348                 /* mm0=B1, mm2=B2, mm3=G2, mm4=G1, mm5=R1, mm6=R2 */\
349                 "packuswb %%mm0, %%mm2          \n\t"\
350                 "packuswb %%mm6, %%mm5          \n\t"\
351                 "packuswb %%mm3, %%mm4          \n\t"\
352                 "pxor %%mm7, %%mm7              \n\t"
353
354 #define WRITEBGR32 \
355                 /* mm2=B, %%mm4=G, %%mm5=R, %%mm7=0 */\
356                         "movq %%mm2, %%mm1              \n\t" /* B */\
357                         "movq %%mm5, %%mm6              \n\t" /* R */\
358                         "punpcklbw %%mm4, %%mm2         \n\t" /* GBGBGBGB 0 */\
359                         "punpcklbw %%mm7, %%mm5         \n\t" /* 0R0R0R0R 0 */\
360                         "punpckhbw %%mm4, %%mm1         \n\t" /* GBGBGBGB 2 */\
361                         "punpckhbw %%mm7, %%mm6         \n\t" /* 0R0R0R0R 2 */\
362                         "movq %%mm2, %%mm0              \n\t" /* GBGBGBGB 0 */\
363                         "movq %%mm1, %%mm3              \n\t" /* GBGBGBGB 2 */\
364                         "punpcklwd %%mm5, %%mm0         \n\t" /* 0RGB0RGB 0 */\
365                         "punpckhwd %%mm5, %%mm2         \n\t" /* 0RGB0RGB 1 */\
366                         "punpcklwd %%mm6, %%mm1         \n\t" /* 0RGB0RGB 2 */\
367                         "punpckhwd %%mm6, %%mm3         \n\t" /* 0RGB0RGB 3 */\
368 \
369                         MOVNTQ(%%mm0, (%4, %%eax, 4))\
370                         MOVNTQ(%%mm2, 8(%4, %%eax, 4))\
371                         MOVNTQ(%%mm1, 16(%4, %%eax, 4))\
372                         MOVNTQ(%%mm3, 24(%4, %%eax, 4))\
373 \
374                         "addl $8, %%eax                 \n\t"\
375                         "cmpl %5, %%eax                 \n\t"\
376                         " jb 1b                         \n\t"
377
378 #define WRITEBGR16 \
379                         "pand bF8, %%mm2                \n\t" /* B */\
380                         "pand bFC, %%mm4                \n\t" /* G */\
381                         "pand bF8, %%mm5                \n\t" /* R */\
382                         "psrlq $3, %%mm2                \n\t"\
383 \
384                         "movq %%mm2, %%mm1              \n\t"\
385                         "movq %%mm4, %%mm3              \n\t"\
386 \
387                         "punpcklbw %%mm7, %%mm3         \n\t"\
388                         "punpcklbw %%mm5, %%mm2         \n\t"\
389                         "punpckhbw %%mm7, %%mm4         \n\t"\
390                         "punpckhbw %%mm5, %%mm1         \n\t"\
391 \
392                         "psllq $3, %%mm3                \n\t"\
393                         "psllq $3, %%mm4                \n\t"\
394 \
395                         "por %%mm3, %%mm2               \n\t"\
396                         "por %%mm4, %%mm1               \n\t"\
397 \
398                         MOVNTQ(%%mm2, (%4, %%eax, 2))\
399                         MOVNTQ(%%mm1, 8(%4, %%eax, 2))\
400 \
401                         "addl $8, %%eax                 \n\t"\
402                         "cmpl %5, %%eax                 \n\t"\
403                         " jb 1b                         \n\t"
404
405 #define WRITEBGR15 \
406                         "pand bF8, %%mm2                \n\t" /* B */\
407                         "pand bF8, %%mm4                \n\t" /* G */\
408                         "pand bF8, %%mm5                \n\t" /* R */\
409                         "psrlq $3, %%mm2                \n\t"\
410                         "psrlq $1, %%mm5                \n\t"\
411 \
412                         "movq %%mm2, %%mm1              \n\t"\
413                         "movq %%mm4, %%mm3              \n\t"\
414 \
415                         "punpcklbw %%mm7, %%mm3         \n\t"\
416                         "punpcklbw %%mm5, %%mm2         \n\t"\
417                         "punpckhbw %%mm7, %%mm4         \n\t"\
418                         "punpckhbw %%mm5, %%mm1         \n\t"\
419 \
420                         "psllq $2, %%mm3                \n\t"\
421                         "psllq $2, %%mm4                \n\t"\
422 \
423                         "por %%mm3, %%mm2               \n\t"\
424                         "por %%mm4, %%mm1               \n\t"\
425 \
426                         MOVNTQ(%%mm2, (%4, %%eax, 2))\
427                         MOVNTQ(%%mm1, 8(%4, %%eax, 2))\
428 \
429                         "addl $8, %%eax                 \n\t"\
430                         "cmpl %5, %%eax                 \n\t"\
431                         " jb 1b                         \n\t"
432
433 #define WRITEBGR24OLD \
434                 /* mm2=B, %%mm4=G, %%mm5=R, %%mm7=0 */\
435                         "movq %%mm2, %%mm1              \n\t" /* B */\
436                         "movq %%mm5, %%mm6              \n\t" /* R */\
437                         "punpcklbw %%mm4, %%mm2         \n\t" /* GBGBGBGB 0 */\
438                         "punpcklbw %%mm7, %%mm5         \n\t" /* 0R0R0R0R 0 */\
439                         "punpckhbw %%mm4, %%mm1         \n\t" /* GBGBGBGB 2 */\
440                         "punpckhbw %%mm7, %%mm6         \n\t" /* 0R0R0R0R 2 */\
441                         "movq %%mm2, %%mm0              \n\t" /* GBGBGBGB 0 */\
442                         "movq %%mm1, %%mm3              \n\t" /* GBGBGBGB 2 */\
443                         "punpcklwd %%mm5, %%mm0         \n\t" /* 0RGB0RGB 0 */\
444                         "punpckhwd %%mm5, %%mm2         \n\t" /* 0RGB0RGB 1 */\
445                         "punpcklwd %%mm6, %%mm1         \n\t" /* 0RGB0RGB 2 */\
446                         "punpckhwd %%mm6, %%mm3         \n\t" /* 0RGB0RGB 3 */\
447 \
448                         "movq %%mm0, %%mm4              \n\t" /* 0RGB0RGB 0 */\
449                         "psrlq $8, %%mm0                \n\t" /* 00RGB0RG 0 */\
450                         "pand bm00000111, %%mm4         \n\t" /* 00000RGB 0 */\
451                         "pand bm11111000, %%mm0         \n\t" /* 00RGB000 0.5 */\
452                         "por %%mm4, %%mm0               \n\t" /* 00RGBRGB 0 */\
453                         "movq %%mm2, %%mm4              \n\t" /* 0RGB0RGB 1 */\
454                         "psllq $48, %%mm2               \n\t" /* GB000000 1 */\
455                         "por %%mm2, %%mm0               \n\t" /* GBRGBRGB 0 */\
456 \
457                         "movq %%mm4, %%mm2              \n\t" /* 0RGB0RGB 1 */\
458                         "psrld $16, %%mm4               \n\t" /* 000R000R 1 */\
459                         "psrlq $24, %%mm2               \n\t" /* 0000RGB0 1.5 */\
460                         "por %%mm4, %%mm2               \n\t" /* 000RRGBR 1 */\
461                         "pand bm00001111, %%mm2         \n\t" /* 0000RGBR 1 */\
462                         "movq %%mm1, %%mm4              \n\t" /* 0RGB0RGB 2 */\
463                         "psrlq $8, %%mm1                \n\t" /* 00RGB0RG 2 */\
464                         "pand bm00000111, %%mm4         \n\t" /* 00000RGB 2 */\
465                         "pand bm11111000, %%mm1         \n\t" /* 00RGB000 2.5 */\
466                         "por %%mm4, %%mm1               \n\t" /* 00RGBRGB 2 */\
467                         "movq %%mm1, %%mm4              \n\t" /* 00RGBRGB 2 */\
468                         "psllq $32, %%mm1               \n\t" /* BRGB0000 2 */\
469                         "por %%mm1, %%mm2               \n\t" /* BRGBRGBR 1 */\
470 \
471                         "psrlq $32, %%mm4               \n\t" /* 000000RG 2.5 */\
472                         "movq %%mm3, %%mm5              \n\t" /* 0RGB0RGB 3 */\
473                         "psrlq $8, %%mm3                \n\t" /* 00RGB0RG 3 */\
474                         "pand bm00000111, %%mm5         \n\t" /* 00000RGB 3 */\
475                         "pand bm11111000, %%mm3         \n\t" /* 00RGB000 3.5 */\
476                         "por %%mm5, %%mm3               \n\t" /* 00RGBRGB 3 */\
477                         "psllq $16, %%mm3               \n\t" /* RGBRGB00 3 */\
478                         "por %%mm4, %%mm3               \n\t" /* RGBRGBRG 2.5 */\
479 \
480                         MOVNTQ(%%mm0, (%%ebx))\
481                         MOVNTQ(%%mm2, 8(%%ebx))\
482                         MOVNTQ(%%mm3, 16(%%ebx))\
483                         "addl $24, %%ebx                \n\t"\
484 \
485                         "addl $8, %%eax                 \n\t"\
486                         "cmpl %5, %%eax                 \n\t"\
487                         " jb 1b                         \n\t"
488
489 #define WRITEBGR24MMX \
490                 /* mm2=B, %%mm4=G, %%mm5=R, %%mm7=0 */\
491                         "movq %%mm2, %%mm1              \n\t" /* B */\
492                         "movq %%mm5, %%mm6              \n\t" /* R */\
493                         "punpcklbw %%mm4, %%mm2         \n\t" /* GBGBGBGB 0 */\
494                         "punpcklbw %%mm7, %%mm5         \n\t" /* 0R0R0R0R 0 */\
495                         "punpckhbw %%mm4, %%mm1         \n\t" /* GBGBGBGB 2 */\
496                         "punpckhbw %%mm7, %%mm6         \n\t" /* 0R0R0R0R 2 */\
497                         "movq %%mm2, %%mm0              \n\t" /* GBGBGBGB 0 */\
498                         "movq %%mm1, %%mm3              \n\t" /* GBGBGBGB 2 */\
499                         "punpcklwd %%mm5, %%mm0         \n\t" /* 0RGB0RGB 0 */\
500                         "punpckhwd %%mm5, %%mm2         \n\t" /* 0RGB0RGB 1 */\
501                         "punpcklwd %%mm6, %%mm1         \n\t" /* 0RGB0RGB 2 */\
502                         "punpckhwd %%mm6, %%mm3         \n\t" /* 0RGB0RGB 3 */\
503 \
504                         "movq %%mm0, %%mm4              \n\t" /* 0RGB0RGB 0 */\
505                         "movq %%mm2, %%mm6              \n\t" /* 0RGB0RGB 1 */\
506                         "movq %%mm1, %%mm5              \n\t" /* 0RGB0RGB 2 */\
507                         "movq %%mm3, %%mm7              \n\t" /* 0RGB0RGB 3 */\
508 \
509                         "psllq $40, %%mm0               \n\t" /* RGB00000 0 */\
510                         "psllq $40, %%mm2               \n\t" /* RGB00000 1 */\
511                         "psllq $40, %%mm1               \n\t" /* RGB00000 2 */\
512                         "psllq $40, %%mm3               \n\t" /* RGB00000 3 */\
513 \
514                         "punpckhdq %%mm4, %%mm0         \n\t" /* 0RGBRGB0 0 */\
515                         "punpckhdq %%mm6, %%mm2         \n\t" /* 0RGBRGB0 1 */\
516                         "punpckhdq %%mm5, %%mm1         \n\t" /* 0RGBRGB0 2 */\
517                         "punpckhdq %%mm7, %%mm3         \n\t" /* 0RGBRGB0 3 */\
518 \
519                         "psrlq $8, %%mm0                \n\t" /* 00RGBRGB 0 */\
520                         "movq %%mm2, %%mm6              \n\t" /* 0RGBRGB0 1 */\
521                         "psllq $40, %%mm2               \n\t" /* GB000000 1 */\
522                         "por %%mm2, %%mm0               \n\t" /* GBRGBRGB 0 */\
523                         MOVNTQ(%%mm0, (%%ebx))\
524 \
525                         "psrlq $24, %%mm6               \n\t" /* 0000RGBR 1 */\
526                         "movq %%mm1, %%mm5              \n\t" /* 0RGBRGB0 2 */\
527                         "psllq $24, %%mm1               \n\t" /* BRGB0000 2 */\
528                         "por %%mm1, %%mm6               \n\t" /* BRGBRGBR 1 */\
529                         MOVNTQ(%%mm6, 8(%%ebx))\
530 \
531                         "psrlq $40, %%mm5               \n\t" /* 000000RG 2 */\
532                         "psllq $8, %%mm3                \n\t" /* RGBRGB00 3 */\
533                         "por %%mm3, %%mm5               \n\t" /* RGBRGBRG 2 */\
534                         MOVNTQ(%%mm5, 16(%%ebx))\
535 \
536                         "addl $24, %%ebx                \n\t"\
537 \
538                         "addl $8, %%eax                 \n\t"\
539                         "cmpl %5, %%eax                 \n\t"\
540                         " jb 1b                         \n\t"
541
542 #define WRITEBGR24MMX2 \
543                 /* mm2=B, %%mm4=G, %%mm5=R, %%mm7=0 */\
544                         "movq M24A, %%mm0               \n\t"\
545                         "movq M24C, %%mm7               \n\t"\
546                         "pshufw $0x50, %%mm2, %%mm1     \n\t" /* B3 B2 B3 B2  B1 B0 B1 B0 */\
547                         "pshufw $0x50, %%mm4, %%mm3     \n\t" /* G3 G2 G3 G2  G1 G0 G1 G0 */\
548                         "pshufw $0x00, %%mm5, %%mm6     \n\t" /* R1 R0 R1 R0  R1 R0 R1 R0 */\
549 \
550                         "pand %%mm0, %%mm1              \n\t" /*    B2        B1       B0 */\
551                         "pand %%mm0, %%mm3              \n\t" /*    G2        G1       G0 */\
552                         "pand %%mm7, %%mm6              \n\t" /*       R1        R0       */\
553 \
554                         "psllq $8, %%mm3                \n\t" /* G2        G1       G0    */\
555                         "por %%mm1, %%mm6               \n\t"\
556                         "por %%mm3, %%mm6               \n\t"\
557                         MOVNTQ(%%mm6, (%%ebx))\
558 \
559                         "psrlq $8, %%mm4                \n\t" /* 00 G7 G6 G5  G4 G3 G2 G1 */\
560                         "pshufw $0xA5, %%mm2, %%mm1     \n\t" /* B5 B4 B5 B4  B3 B2 B3 B2 */\
561                         "pshufw $0x55, %%mm4, %%mm3     \n\t" /* G4 G3 G4 G3  G4 G3 G4 G3 */\
562                         "pshufw $0xA5, %%mm5, %%mm6     \n\t" /* R5 R4 R5 R4  R3 R2 R3 R2 */\
563 \
564                         "pand M24B, %%mm1               \n\t" /* B5       B4        B3    */\
565                         "pand %%mm7, %%mm3              \n\t" /*       G4        G3       */\
566                         "pand %%mm0, %%mm6              \n\t" /*    R4        R3       R2 */\
567 \
568                         "por %%mm1, %%mm3               \n\t" /* B5    G4 B4     G3 B3    */\
569                         "por %%mm3, %%mm6               \n\t"\
570                         MOVNTQ(%%mm6, 8(%%ebx))\
571 \
572                         "pshufw $0xFF, %%mm2, %%mm1     \n\t" /* B7 B6 B7 B6  B7 B6 B6 B7 */\
573                         "pshufw $0xFA, %%mm4, %%mm3     \n\t" /* 00 G7 00 G7  G6 G5 G6 G5 */\
574                         "pshufw $0xFA, %%mm5, %%mm6     \n\t" /* R7 R6 R7 R6  R5 R4 R5 R4 */\
575 \
576                         "pand %%mm7, %%mm1              \n\t" /*       B7        B6       */\
577                         "pand %%mm0, %%mm3              \n\t" /*    G7        G6       G5 */\
578                         "pand M24B, %%mm6               \n\t" /* R7       R6        R5    */\
579 \
580                         "por %%mm1, %%mm3               \n\t"\
581                         "por %%mm3, %%mm6               \n\t"\
582                         MOVNTQ(%%mm6, 16(%%ebx))\
583 \
584                         "addl $24, %%ebx                \n\t"\
585 \
586                         "addl $8, %%eax                 \n\t"\
587                         "cmpl %5, %%eax                 \n\t"\
588                         " jb 1b                         \n\t"
589
590 #ifdef HAVE_MMX2
591 #define WRITEBGR24 WRITEBGR24MMX2
592 #else
593 #define WRITEBGR24 WRITEBGR24MMX
594 #endif
595
596 #ifdef HAVE_MMX
597 void in_asm_used_var_warning_killer()
598 {
599  int i= yCoeff+vrCoeff+ubCoeff+vgCoeff+ugCoeff+bF8+bFC+w400+w80+w10+
600  bm00001111+bm00000111+bm11111000+b16Dither+b16Dither1+b16Dither2+g16Dither+g16Dither1+
601  g16Dither2+b16Mask+g16Mask+r16Mask+b15Mask+g15Mask+r15Mask+temp0+asm_yalpha1+ asm_uvalpha1+
602  M24A+M24B+M24C;
603  if(i) i=0;
604 }
605 #endif
606
607 static inline void yuv2yuv(uint16_t *buf0, uint16_t *buf1, uint16_t *uvbuf0, uint16_t *uvbuf1,
608                            uint8_t *dest, uint8_t *uDest, uint8_t *vDest, int dstw, int yalpha, int uvalpha)
609 {
610         int yalpha1=yalpha^4095;
611         int uvalpha1=uvalpha^4095;
612         int i;
613
614         asm volatile ("\n\t"::: "memory");
615
616         for(i=0;i<dstw;i++)
617         {
618                 ((uint8_t*)dest)[i] = (buf0[i]*yalpha1+buf1[i]*yalpha)>>19;
619         }
620
621         if(uvalpha != -1)
622         {
623                 for(i=0; i<(dstw>>1); i++)
624                 {
625                         ((uint8_t*)uDest)[i] = (uvbuf0[i]*uvalpha1+uvbuf1[i]*uvalpha)>>19;
626                         ((uint8_t*)vDest)[i] = (uvbuf0[i+2048]*uvalpha1+uvbuf1[i+2048]*uvalpha)>>19;
627                 }
628         }
629 }
630
631 /**
632  * vertical scale YV12 to RGB
633  */
634 static inline void yuv2rgbX(uint16_t *buf0, uint16_t *buf1, uint16_t *uvbuf0, uint16_t *uvbuf1,
635                             uint8_t *dest, int dstw, int yalpha, int uvalpha, int dstbpp)
636 {
637         int yalpha1=yalpha^4095;
638         int uvalpha1=uvalpha^4095;
639
640         if(fullUVIpol)
641         {
642
643 #ifdef HAVE_MMX
644                 if(dstbpp == 32)
645                 {
646                         asm volatile(
647
648
649 FULL_YSCALEYUV2RGB
650                         "punpcklbw %%mm1, %%mm3         \n\t" // BGBGBGBG
651                         "punpcklbw %%mm7, %%mm0         \n\t" // R0R0R0R0
652
653                         "movq %%mm3, %%mm1              \n\t"
654                         "punpcklwd %%mm0, %%mm3         \n\t" // BGR0BGR0
655                         "punpckhwd %%mm0, %%mm1         \n\t" // BGR0BGR0
656
657                         MOVNTQ(%%mm3, (%4, %%eax, 4))
658                         MOVNTQ(%%mm1, 8(%4, %%eax, 4))
659
660                         "addl $4, %%eax                 \n\t"
661                         "cmpl %5, %%eax                 \n\t"
662                         " jb 1b                         \n\t"
663
664
665                         :: "r" (buf0), "r" (buf1), "r" (uvbuf0), "r" (uvbuf1), "r" (dest), "m" (dstw),
666                         "m" (yalpha1), "m" (uvalpha1)
667                         : "%eax"
668                         );
669                 }
670                 else if(dstbpp==24)
671                 {
672                         asm volatile(
673
674 FULL_YSCALEYUV2RGB
675
676                                                                 // lsb ... msb
677                         "punpcklbw %%mm1, %%mm3         \n\t" // BGBGBGBG
678                         "punpcklbw %%mm7, %%mm0         \n\t" // R0R0R0R0
679
680                         "movq %%mm3, %%mm1              \n\t"
681                         "punpcklwd %%mm0, %%mm3         \n\t" // BGR0BGR0
682                         "punpckhwd %%mm0, %%mm1         \n\t" // BGR0BGR0
683
684                         "movq %%mm3, %%mm2              \n\t" // BGR0BGR0
685                         "psrlq $8, %%mm3                \n\t" // GR0BGR00
686                         "pand bm00000111, %%mm2         \n\t" // BGR00000
687                         "pand bm11111000, %%mm3         \n\t" // 000BGR00
688                         "por %%mm2, %%mm3               \n\t" // BGRBGR00
689                         "movq %%mm1, %%mm2              \n\t"
690                         "psllq $48, %%mm1               \n\t" // 000000BG
691                         "por %%mm1, %%mm3               \n\t" // BGRBGRBG
692
693                         "movq %%mm2, %%mm1              \n\t" // BGR0BGR0
694                         "psrld $16, %%mm2               \n\t" // R000R000
695                         "psrlq $24, %%mm1               \n\t" // 0BGR0000
696                         "por %%mm2, %%mm1               \n\t" // RBGRR000
697
698                         "movl %4, %%ebx                 \n\t"
699                         "addl %%eax, %%ebx              \n\t"
700
701 #ifdef HAVE_MMX2
702                         //FIXME Alignment
703                         "movntq %%mm3, (%%ebx, %%eax, 2)\n\t"
704                         "movntq %%mm1, 8(%%ebx, %%eax, 2)\n\t"
705 #else
706                         "movd %%mm3, (%%ebx, %%eax, 2)  \n\t"
707                         "psrlq $32, %%mm3               \n\t"
708                         "movd %%mm3, 4(%%ebx, %%eax, 2) \n\t"
709                         "movd %%mm1, 8(%%ebx, %%eax, 2) \n\t"
710 #endif
711                         "addl $4, %%eax                 \n\t"
712                         "cmpl %5, %%eax                 \n\t"
713                         " jb 1b                         \n\t"
714
715                         :: "r" (buf0), "r" (buf1), "r" (uvbuf0), "r" (uvbuf1), "m" (dest), "m" (dstw),
716                         "m" (yalpha1), "m" (uvalpha1)
717                         : "%eax", "%ebx"
718                         );
719                 }
720                 else if(dstbpp==15)
721                 {
722                         asm volatile(
723
724 FULL_YSCALEYUV2RGB
725 #ifdef DITHER1XBPP
726                         "paddusb b16Dither, %%mm1       \n\t"
727                         "paddusb b16Dither, %%mm0       \n\t"
728                         "paddusb b16Dither, %%mm3       \n\t"
729 #endif
730                         "punpcklbw %%mm7, %%mm1         \n\t" // 0G0G0G0G
731                         "punpcklbw %%mm7, %%mm3         \n\t" // 0B0B0B0B
732                         "punpcklbw %%mm7, %%mm0         \n\t" // 0R0R0R0R
733
734                         "psrlw $3, %%mm3                \n\t"
735                         "psllw $2, %%mm1                \n\t"
736                         "psllw $7, %%mm0                \n\t"
737                         "pand g15Mask, %%mm1            \n\t"
738                         "pand r15Mask, %%mm0            \n\t"
739
740                         "por %%mm3, %%mm1               \n\t"
741                         "por %%mm1, %%mm0               \n\t"
742
743                         MOVNTQ(%%mm0, (%4, %%eax, 2))
744
745                         "addl $4, %%eax                 \n\t"
746                         "cmpl %5, %%eax                 \n\t"
747                         " jb 1b                         \n\t"
748
749                         :: "r" (buf0), "r" (buf1), "r" (uvbuf0), "r" (uvbuf1), "r" (dest), "m" (dstw),
750                         "m" (yalpha1), "m" (uvalpha1)
751                         : "%eax"
752                         );
753                 }
754                 else if(dstbpp==16)
755                 {
756                         asm volatile(
757
758 FULL_YSCALEYUV2RGB
759 #ifdef DITHER1XBPP
760                         "paddusb g16Dither, %%mm1       \n\t"
761                         "paddusb b16Dither, %%mm0       \n\t"
762                         "paddusb b16Dither, %%mm3       \n\t"
763 #endif
764                         "punpcklbw %%mm7, %%mm1         \n\t" // 0G0G0G0G
765                         "punpcklbw %%mm7, %%mm3         \n\t" // 0B0B0B0B
766                         "punpcklbw %%mm7, %%mm0         \n\t" // 0R0R0R0R
767
768                         "psrlw $3, %%mm3                \n\t"
769                         "psllw $3, %%mm1                \n\t"
770                         "psllw $8, %%mm0                \n\t"
771                         "pand g16Mask, %%mm1            \n\t"
772                         "pand r16Mask, %%mm0            \n\t"
773
774                         "por %%mm3, %%mm1               \n\t"
775                         "por %%mm1, %%mm0               \n\t"
776
777                         MOVNTQ(%%mm0, (%4, %%eax, 2))
778
779                         "addl $4, %%eax                 \n\t"
780                         "cmpl %5, %%eax                 \n\t"
781                         " jb 1b                         \n\t"
782
783                         :: "r" (buf0), "r" (buf1), "r" (uvbuf0), "r" (uvbuf1), "r" (dest), "m" (dstw),
784                         "m" (yalpha1), "m" (uvalpha1)
785                         : "%eax"
786                         );
787                 }
788 #else
789                 asm volatile ("\n\t"::: "memory");
790
791                 if(dstbpp==32 || dstbpp==24)
792                 {
793                         int i;
794                         for(i=0;i<dstw;i++){
795                                 // vertical linear interpolation && yuv2rgb in a single step:
796                                 int Y=yuvtab_2568[((buf0[i]*yalpha1+buf1[i]*yalpha)>>19)];
797                                 int U=((uvbuf0[i]*uvalpha1+uvbuf1[i]*uvalpha)>>19);
798                                 int V=((uvbuf0[i+2048]*uvalpha1+uvbuf1[i+2048]*uvalpha)>>19);
799                                 dest[0]=clip_table[((Y + yuvtab_40cf[U]) >>13)];
800                                 dest[1]=clip_table[((Y + yuvtab_1a1e[V] + yuvtab_0c92[U]) >>13)];
801                                 dest[2]=clip_table[((Y + yuvtab_3343[V]) >>13)];
802                                 dest+=dstbpp>>3;
803                         }
804                 }
805                 else if(dstbpp==16)
806                 {
807                         int i;
808                         for(i=0;i<dstw;i++){
809                                 // vertical linear interpolation && yuv2rgb in a single step:
810                                 int Y=yuvtab_2568[((buf0[i]*yalpha1+buf1[i]*yalpha)>>19)];
811                                 int U=((uvbuf0[i]*uvalpha1+uvbuf1[i]*uvalpha)>>19);
812                                 int V=((uvbuf0[i+2048]*uvalpha1+uvbuf1[i+2048]*uvalpha)>>19);
813
814                                 ((uint16_t*)dest)[i] =
815                                         clip_table16b[(Y + yuvtab_40cf[U]) >>13] |
816                                         clip_table16g[(Y + yuvtab_1a1e[V] + yuvtab_0c92[U]) >>13] |
817                                         clip_table16r[(Y + yuvtab_3343[V]) >>13];
818                         }
819                 }
820                 else if(dstbpp==15)
821                 {
822                         int i;
823                         for(i=0;i<dstw;i++){
824                                 // vertical linear interpolation && yuv2rgb in a single step:
825                                 int Y=yuvtab_2568[((buf0[i]*yalpha1+buf1[i]*yalpha)>>19)];
826                                 int U=((uvbuf0[i]*uvalpha1+uvbuf1[i]*uvalpha)>>19);
827                                 int V=((uvbuf0[i+2048]*uvalpha1+uvbuf1[i+2048]*uvalpha)>>19);
828
829                                 ((uint16_t*)dest)[i] =
830                                         clip_table15b[(Y + yuvtab_40cf[U]) >>13] |
831                                         clip_table15g[(Y + yuvtab_1a1e[V] + yuvtab_0c92[U]) >>13] |
832                                         clip_table15r[(Y + yuvtab_3343[V]) >>13];
833                         }
834                 }
835 #endif
836         }//FULL_UV_IPOL
837         else
838         {
839 #ifdef HAVE_MMX
840                 if(dstbpp == 32)
841                 {
842                         asm volatile(
843                                 YSCALEYUV2RGB
844                                 WRITEBGR32
845
846                         :: "r" (buf0), "r" (buf1), "r" (uvbuf0), "r" (uvbuf1), "r" (dest), "m" (dstw),
847                         "m" (yalpha1), "m" (uvalpha1)
848                         : "%eax"
849                         );
850                 }
851                 else if(dstbpp==24)
852                 {
853                         asm volatile(
854                                 "movl %4, %%ebx                 \n\t"
855                                 YSCALEYUV2RGB
856                                 WRITEBGR24
857
858                         :: "r" (buf0), "r" (buf1), "r" (uvbuf0), "r" (uvbuf1), "m" (dest), "m" (dstw),
859                         "m" (yalpha1), "m" (uvalpha1)
860                         : "%eax", "%ebx"
861                         );
862                 }
863                 else if(dstbpp==15)
864                 {
865                         asm volatile(
866                                 YSCALEYUV2RGB
867                 /* mm2=B, %%mm4=G, %%mm5=R, %%mm7=0 */
868 #ifdef DITHER1XBPP
869                                 "paddusb b16Dither, %%mm2       \n\t"
870                                 "paddusb b16Dither, %%mm4       \n\t"
871                                 "paddusb b16Dither, %%mm5       \n\t"
872 #endif
873
874                                 WRITEBGR15
875
876                         :: "r" (buf0), "r" (buf1), "r" (uvbuf0), "r" (uvbuf1), "r" (dest), "m" (dstw),
877                         "m" (yalpha1), "m" (uvalpha1)
878                         : "%eax"
879                         );
880                 }
881                 else if(dstbpp==16)
882                 {
883                         asm volatile(
884                                 YSCALEYUV2RGB
885                 /* mm2=B, %%mm4=G, %%mm5=R, %%mm7=0 */
886 #ifdef DITHER1XBPP
887                                 "paddusb g16Dither, %%mm2       \n\t"
888                                 "paddusb b16Dither, %%mm4       \n\t"
889                                 "paddusb b16Dither, %%mm5       \n\t"
890 #endif
891
892                                 WRITEBGR16
893
894                         :: "r" (buf0), "r" (buf1), "r" (uvbuf0), "r" (uvbuf1), "r" (dest), "m" (dstw),
895                         "m" (yalpha1), "m" (uvalpha1)
896                         : "%eax"
897                         );
898                 }
899 #else
900                 asm volatile ("\n\t"::: "memory");
901
902                 if(dstbpp==32)
903                 {
904                         int i;
905                         for(i=0; i<dstw-1; i+=2){
906                                 // vertical linear interpolation && yuv2rgb in a single step:
907                                 int Y1=yuvtab_2568[((buf0[i]*yalpha1+buf1[i]*yalpha)>>19)];
908                                 int Y2=yuvtab_2568[((buf0[i+1]*yalpha1+buf1[i+1]*yalpha)>>19)];
909                                 int U=((uvbuf0[i>>1]*uvalpha1+uvbuf1[i>>1]*uvalpha)>>19);
910                                 int V=((uvbuf0[(i>>1)+2048]*uvalpha1+uvbuf1[(i>>1)+2048]*uvalpha)>>19);
911
912                                 int Cb= yuvtab_40cf[U];
913                                 int Cg= yuvtab_1a1e[V] + yuvtab_0c92[U];
914                                 int Cr= yuvtab_3343[V];
915
916                                 dest[4*i+0]=clip_table[((Y1 + Cb) >>13)];
917                                 dest[4*i+1]=clip_table[((Y1 + Cg) >>13)];
918                                 dest[4*i+2]=clip_table[((Y1 + Cr) >>13)];
919
920                                 dest[4*i+4]=clip_table[((Y2 + Cb) >>13)];
921                                 dest[4*i+5]=clip_table[((Y2 + Cg) >>13)];
922                                 dest[4*i+6]=clip_table[((Y2 + Cr) >>13)];
923                         }
924                 }
925                 if(dstbpp==24)
926                 {
927                         int i;
928                         for(i=0; i<dstw-1; i+=2){
929                                 // vertical linear interpolation && yuv2rgb in a single step:
930                                 int Y1=yuvtab_2568[((buf0[i]*yalpha1+buf1[i]*yalpha)>>19)];
931                                 int Y2=yuvtab_2568[((buf0[i+1]*yalpha1+buf1[i+1]*yalpha)>>19)];
932                                 int U=((uvbuf0[i>>1]*uvalpha1+uvbuf1[i>>1]*uvalpha)>>19);
933                                 int V=((uvbuf0[(i>>1)+2048]*uvalpha1+uvbuf1[(i>>1)+2048]*uvalpha)>>19);
934
935                                 int Cb= yuvtab_40cf[U];
936                                 int Cg= yuvtab_1a1e[V] + yuvtab_0c92[U];
937                                 int Cr= yuvtab_3343[V];
938
939                                 dest[0]=clip_table[((Y1 + Cb) >>13)];
940                                 dest[1]=clip_table[((Y1 + Cg) >>13)];
941                                 dest[2]=clip_table[((Y1 + Cr) >>13)];
942
943                                 dest[3]=clip_table[((Y2 + Cb) >>13)];
944                                 dest[4]=clip_table[((Y2 + Cg) >>13)];
945                                 dest[5]=clip_table[((Y2 + Cr) >>13)];
946                                 dest+=6;
947                         }
948                 }
949                 else if(dstbpp==16)
950                 {
951                         int i;
952                         for(i=0; i<dstw-1; i+=2){
953                                 // vertical linear interpolation && yuv2rgb in a single step:
954                                 int Y1=yuvtab_2568[((buf0[i]*yalpha1+buf1[i]*yalpha)>>19)];
955                                 int Y2=yuvtab_2568[((buf0[i+1]*yalpha1+buf1[i+1]*yalpha)>>19)];
956                                 int U=((uvbuf0[i>>1]*uvalpha1+uvbuf1[i>>1]*uvalpha)>>19);
957                                 int V=((uvbuf0[(i>>1)+2048]*uvalpha1+uvbuf1[(i>>1)+2048]*uvalpha)>>19);
958
959                                 int Cb= yuvtab_40cf[U];
960                                 int Cg= yuvtab_1a1e[V] + yuvtab_0c92[U];
961                                 int Cr= yuvtab_3343[V];
962
963                                 ((uint16_t*)dest)[i] =
964                                         clip_table16b[(Y1 + Cb) >>13] |
965                                         clip_table16g[(Y1 + Cg) >>13] |
966                                         clip_table16r[(Y1 + Cr) >>13];
967
968                                 ((uint16_t*)dest)[i+1] =
969                                         clip_table16b[(Y2 + Cb) >>13] |
970                                         clip_table16g[(Y2 + Cg) >>13] |
971                                         clip_table16r[(Y2 + Cr) >>13];
972                         }
973                 }
974                 else if(dstbpp==15)
975                 {
976                         int i;
977                         for(i=0; i<dstw-1; i+=2){
978                                 // vertical linear interpolation && yuv2rgb in a single step:
979                                 int Y1=yuvtab_2568[((buf0[i]*yalpha1+buf1[i]*yalpha)>>19)];
980                                 int Y2=yuvtab_2568[((buf0[i+1]*yalpha1+buf1[i+1]*yalpha)>>19)];
981                                 int U=((uvbuf0[i>>1]*uvalpha1+uvbuf1[i>>1]*uvalpha)>>19);
982                                 int V=((uvbuf0[(i>>1)+2048]*uvalpha1+uvbuf1[(i>>1)+2048]*uvalpha)>>19);
983
984                                 int Cb= yuvtab_40cf[U];
985                                 int Cg= yuvtab_1a1e[V] + yuvtab_0c92[U];
986                                 int Cr= yuvtab_3343[V];
987
988                                 ((uint16_t*)dest)[i] =
989                                         clip_table15b[(Y1 + Cb) >>13] |
990                                         clip_table15g[(Y1 + Cg) >>13] |
991                                         clip_table15r[(Y1 + Cr) >>13];
992
993                                 ((uint16_t*)dest)[i+1] =
994                                         clip_table15b[(Y2 + Cb) >>13] |
995                                         clip_table15g[(Y2 + Cg) >>13] |
996                                         clip_table15r[(Y2 + Cr) >>13];
997                         }
998                 }
999 #endif
1000         } //!FULL_UV_IPOL
1001 }
1002
1003 /**
1004  * YV12 to RGB without scaling or interpolating
1005  */
1006 static inline void yuv2rgb1(uint16_t *buf0, uint16_t *buf1, uint16_t *uvbuf0, uint16_t *uvbuf1,
1007                             uint8_t *dest, int dstw, int yalpha, int uvalpha, int dstbpp)
1008 {
1009         int uvalpha1=uvalpha^4095;
1010 #ifdef HAVE_MMX
1011         int yalpha1=yalpha^4095;
1012 #endif
1013
1014         if(fullUVIpol || allwaysIpol)
1015         {
1016                 yuv2rgbX(buf0, buf1, uvbuf0, uvbuf1, dest, dstw, yalpha, uvalpha, dstbpp);
1017                 return;
1018         }
1019         if( yalpha > 2048 ) buf0 = buf1;
1020
1021 #ifdef HAVE_MMX
1022         if( uvalpha < 2048 ) // note this is not correct (shifts chrominance by 0.5 pixels) but its a bit faster
1023         {
1024                 if(dstbpp == 32)
1025                 {
1026                         asm volatile(
1027                                 YSCALEYUV2RGB1
1028                                 WRITEBGR32
1029                         :: "r" (buf0), "r" (buf1), "r" (uvbuf0), "r" (uvbuf1), "r" (dest), "m" (dstw),
1030                         "m" (yalpha1), "m" (uvalpha1)
1031                         : "%eax"
1032                         );
1033                 }
1034                 else if(dstbpp==24)
1035                 {
1036                         asm volatile(
1037                                 "movl %4, %%ebx                 \n\t"
1038                                 YSCALEYUV2RGB1
1039                                 WRITEBGR24
1040                         :: "r" (buf0), "r" (buf1), "r" (uvbuf0), "r" (uvbuf1), "m" (dest), "m" (dstw),
1041                         "m" (yalpha1), "m" (uvalpha1)
1042                         : "%eax", "%ebx"
1043                         );
1044                 }
1045                 else if(dstbpp==15)
1046                 {
1047                         asm volatile(
1048                                 YSCALEYUV2RGB1
1049                 /* mm2=B, %%mm4=G, %%mm5=R, %%mm7=0 */
1050 #ifdef DITHER1XBPP
1051                                 "paddusb b16Dither, %%mm2       \n\t"
1052                                 "paddusb b16Dither, %%mm4       \n\t"
1053                                 "paddusb b16Dither, %%mm5       \n\t"
1054 #endif
1055                                 WRITEBGR15
1056                         :: "r" (buf0), "r" (buf1), "r" (uvbuf0), "r" (uvbuf1), "r" (dest), "m" (dstw),
1057                         "m" (yalpha1), "m" (uvalpha1)
1058                         : "%eax"
1059                         );
1060                 }
1061                 else if(dstbpp==16)
1062                 {
1063                         asm volatile(
1064                                 YSCALEYUV2RGB1
1065                 /* mm2=B, %%mm4=G, %%mm5=R, %%mm7=0 */
1066 #ifdef DITHER1XBPP
1067                                 "paddusb g16Dither, %%mm2       \n\t"
1068                                 "paddusb b16Dither, %%mm4       \n\t"
1069                                 "paddusb b16Dither, %%mm5       \n\t"
1070 #endif
1071
1072                                 WRITEBGR16
1073                         :: "r" (buf0), "r" (buf1), "r" (uvbuf0), "r" (uvbuf1), "r" (dest), "m" (dstw),
1074                         "m" (yalpha1), "m" (uvalpha1)
1075                         : "%eax"
1076                         );
1077                 }
1078         }
1079         else
1080         {
1081                 if(dstbpp == 32)
1082                 {
1083                         asm volatile(
1084                                 YSCALEYUV2RGB1b
1085                                 WRITEBGR32
1086                         :: "r" (buf0), "r" (buf1), "r" (uvbuf0), "r" (uvbuf1), "r" (dest), "m" (dstw),
1087                         "m" (yalpha1), "m" (uvalpha1)
1088                         : "%eax"
1089                         );
1090                 }
1091                 else if(dstbpp==24)
1092                 {
1093                         asm volatile(
1094                                 "movl %4, %%ebx                 \n\t"
1095                                 YSCALEYUV2RGB1b
1096                                 WRITEBGR24
1097                         :: "r" (buf0), "r" (buf1), "r" (uvbuf0), "r" (uvbuf1), "m" (dest), "m" (dstw),
1098                         "m" (yalpha1), "m" (uvalpha1)
1099                         : "%eax", "%ebx"
1100                         );
1101                 }
1102                 else if(dstbpp==15)
1103                 {
1104                         asm volatile(
1105                                 YSCALEYUV2RGB1b
1106                 /* mm2=B, %%mm4=G, %%mm5=R, %%mm7=0 */
1107 #ifdef DITHER1XBPP
1108                                 "paddusb b16Dither, %%mm2       \n\t"
1109                                 "paddusb b16Dither, %%mm4       \n\t"
1110                                 "paddusb b16Dither, %%mm5       \n\t"
1111 #endif
1112                                 WRITEBGR15
1113                         :: "r" (buf0), "r" (buf1), "r" (uvbuf0), "r" (uvbuf1), "r" (dest), "m" (dstw),
1114                         "m" (yalpha1), "m" (uvalpha1)
1115                         : "%eax"
1116                         );
1117                 }
1118                 else if(dstbpp==16)
1119                 {
1120                         asm volatile(
1121                                 YSCALEYUV2RGB1b
1122                 /* mm2=B, %%mm4=G, %%mm5=R, %%mm7=0 */
1123 #ifdef DITHER1XBPP
1124                                 "paddusb g16Dither, %%mm2       \n\t"
1125                                 "paddusb b16Dither, %%mm4       \n\t"
1126                                 "paddusb b16Dither, %%mm5       \n\t"
1127 #endif
1128
1129                                 WRITEBGR16
1130                         :: "r" (buf0), "r" (buf1), "r" (uvbuf0), "r" (uvbuf1), "r" (dest), "m" (dstw),
1131                         "m" (yalpha1), "m" (uvalpha1)
1132                         : "%eax"
1133                         );
1134                 }
1135         }
1136 #else
1137 //FIXME write 2 versions (for even & odd lines)
1138         asm volatile ("\n\t"::: "memory");
1139
1140         if(dstbpp==32)
1141         {
1142                 int i;
1143                 for(i=0; i<dstw-1; i+=2){
1144                         // vertical linear interpolation && yuv2rgb in a single step:
1145                         int Y1=yuvtab_2568[buf0[i]>>7];
1146                         int Y2=yuvtab_2568[buf0[i+1]>>7];
1147                         int U=((uvbuf0[i>>1]*uvalpha1+uvbuf1[i>>1]*uvalpha)>>19);
1148                         int V=((uvbuf0[(i>>1)+2048]*uvalpha1+uvbuf1[(i>>1)+2048]*uvalpha)>>19);
1149
1150                         int Cb= yuvtab_40cf[U];
1151                         int Cg= yuvtab_1a1e[V] + yuvtab_0c92[U];
1152                         int Cr= yuvtab_3343[V];
1153
1154                         dest[4*i+0]=clip_table[((Y1 + Cb) >>13)];
1155                         dest[4*i+1]=clip_table[((Y1 + Cg) >>13)];
1156                         dest[4*i+2]=clip_table[((Y1 + Cr) >>13)];
1157
1158                         dest[4*i+4]=clip_table[((Y2 + Cb) >>13)];
1159                         dest[4*i+5]=clip_table[((Y2 + Cg) >>13)];
1160                         dest[4*i+6]=clip_table[((Y2 + Cr) >>13)];
1161                 }
1162         }
1163         if(dstbpp==24)
1164         {
1165                 int i;
1166                 for(i=0; i<dstw-1; i+=2){
1167                         // vertical linear interpolation && yuv2rgb in a single step:
1168                         int Y1=yuvtab_2568[buf0[i]>>7];
1169                         int Y2=yuvtab_2568[buf0[i+1]>>7];
1170                         int U=((uvbuf0[i>>1]*uvalpha1+uvbuf1[i>>1]*uvalpha)>>19);
1171                         int V=((uvbuf0[(i>>1)+2048]*uvalpha1+uvbuf1[(i>>1)+2048]*uvalpha)>>19);
1172
1173                         int Cb= yuvtab_40cf[U];
1174                         int Cg= yuvtab_1a1e[V] + yuvtab_0c92[U];
1175                         int Cr= yuvtab_3343[V];
1176
1177                         dest[0]=clip_table[((Y1 + Cb) >>13)];
1178                         dest[1]=clip_table[((Y1 + Cg) >>13)];
1179                         dest[2]=clip_table[((Y1 + Cr) >>13)];
1180
1181                         dest[3]=clip_table[((Y2 + Cb) >>13)];
1182                         dest[4]=clip_table[((Y2 + Cg) >>13)];
1183                         dest[5]=clip_table[((Y2 + Cr) >>13)];
1184                         dest+=6;
1185                 }
1186         }
1187         else if(dstbpp==16)
1188         {
1189                 int i;
1190                 for(i=0; i<dstw-1; i+=2){
1191                         // vertical linear interpolation && yuv2rgb in a single step:
1192                         int Y1=yuvtab_2568[buf0[i]>>7];
1193                         int Y2=yuvtab_2568[buf0[i+1]>>7];
1194                         int U=((uvbuf0[i>>1]*uvalpha1+uvbuf1[i>>1]*uvalpha)>>19);
1195                         int V=((uvbuf0[(i>>1)+2048]*uvalpha1+uvbuf1[(i>>1)+2048]*uvalpha)>>19);
1196
1197                         int Cb= yuvtab_40cf[U];
1198                         int Cg= yuvtab_1a1e[V] + yuvtab_0c92[U];
1199                         int Cr= yuvtab_3343[V];
1200
1201                         ((uint16_t*)dest)[i] =
1202                                 clip_table16b[(Y1 + Cb) >>13] |
1203                                 clip_table16g[(Y1 + Cg) >>13] |
1204                                 clip_table16r[(Y1 + Cr) >>13];
1205
1206                         ((uint16_t*)dest)[i+1] =
1207                                 clip_table16b[(Y2 + Cb) >>13] |
1208                                 clip_table16g[(Y2 + Cg) >>13] |
1209                                 clip_table16r[(Y2 + Cr) >>13];
1210                 }
1211         }
1212         else if(dstbpp==15)
1213         {
1214                 int i;
1215                 for(i=0; i<dstw-1; i+=2){
1216                         // vertical linear interpolation && yuv2rgb in a single step:
1217                         int Y1=yuvtab_2568[buf0[i]>>7];
1218                         int Y2=yuvtab_2568[buf0[i+1]>>7];
1219                         int U=((uvbuf0[i>>1]*uvalpha1+uvbuf1[i>>1]*uvalpha)>>19);
1220                         int V=((uvbuf0[(i>>1)+2048]*uvalpha1+uvbuf1[(i>>1)+2048]*uvalpha)>>19);
1221
1222                         int Cb= yuvtab_40cf[U];
1223                         int Cg= yuvtab_1a1e[V] + yuvtab_0c92[U];
1224                         int Cr= yuvtab_3343[V];
1225
1226                         ((uint16_t*)dest)[i] =
1227                                 clip_table15b[(Y1 + Cb) >>13] |
1228                                 clip_table15g[(Y1 + Cg) >>13] |
1229                                 clip_table15r[(Y1 + Cr) >>13];
1230
1231                         ((uint16_t*)dest)[i+1] =
1232                                 clip_table15b[(Y2 + Cb) >>13] |
1233                                 clip_table15g[(Y2 + Cg) >>13] |
1234                                 clip_table15r[(Y2 + Cr) >>13];
1235                 }
1236         }
1237 #endif
1238 }
1239
1240
1241 static inline void hyscale(uint16_t *dst, int dstWidth, uint8_t *src, int srcWidth, int xInc)
1242 {
1243       // *** horizontal scale Y line to temp buffer
1244 #ifdef ARCH_X86
1245 #ifdef HAVE_MMX2
1246         int i;
1247         if(canMMX2BeUsed)
1248         {
1249                 asm volatile(
1250                         "pxor %%mm7, %%mm7              \n\t"
1251                         "pxor %%mm2, %%mm2              \n\t" // 2*xalpha
1252                         "movd %5, %%mm6                 \n\t" // xInc&0xFFFF
1253                         "punpcklwd %%mm6, %%mm6         \n\t"
1254                         "punpcklwd %%mm6, %%mm6         \n\t"
1255                         "movq %%mm6, %%mm2              \n\t"
1256                         "psllq $16, %%mm2               \n\t"
1257                         "paddw %%mm6, %%mm2             \n\t"
1258                         "psllq $16, %%mm2               \n\t"
1259                         "paddw %%mm6, %%mm2             \n\t"
1260                         "psllq $16, %%mm2               \n\t" //0,t,2t,3t               t=xInc&0xFF
1261                         "movq %%mm2, temp0              \n\t"
1262                         "movd %4, %%mm6                 \n\t" //(xInc*4)&0xFFFF
1263                         "punpcklwd %%mm6, %%mm6         \n\t"
1264                         "punpcklwd %%mm6, %%mm6         \n\t"
1265                         "xorl %%eax, %%eax              \n\t" // i
1266                         "movl %0, %%esi                 \n\t" // src
1267                         "movl %1, %%edi                 \n\t" // buf1
1268                         "movl %3, %%edx                 \n\t" // (xInc*4)>>16
1269                         "xorl %%ecx, %%ecx              \n\t"
1270                         "xorl %%ebx, %%ebx              \n\t"
1271                         "movw %4, %%bx                  \n\t" // (xInc*4)&0xFFFF
1272
1273 #define FUNNY_Y_CODE \
1274                         PREFETCH" 1024(%%esi)           \n\t"\
1275                         PREFETCH" 1056(%%esi)           \n\t"\
1276                         PREFETCH" 1088(%%esi)           \n\t"\
1277                         "call funnyYCode                \n\t"\
1278                         "movq temp0, %%mm2              \n\t"\
1279                         "xorl %%ecx, %%ecx              \n\t"
1280
1281 FUNNY_Y_CODE
1282 FUNNY_Y_CODE
1283 FUNNY_Y_CODE
1284 FUNNY_Y_CODE
1285 FUNNY_Y_CODE
1286 FUNNY_Y_CODE
1287 FUNNY_Y_CODE
1288 FUNNY_Y_CODE
1289
1290                         :: "m" (src), "m" (dst), "m" (dstWidth), "m" ((xInc*4)>>16),
1291                         "m" ((xInc*4)&0xFFFF), "m" (xInc&0xFFFF)
1292                         : "%eax", "%ebx", "%ecx", "%edx", "%esi", "%edi"
1293                 );
1294                 for(i=dstWidth-1; (i*xInc)>>16 >=srcWidth-1; i--) dst[i] = src[srcWidth-1]*128;
1295         }
1296         else
1297         {
1298 #endif
1299         //NO MMX just normal asm ...
1300         asm volatile(
1301                 "xorl %%eax, %%eax              \n\t" // i
1302                 "xorl %%ebx, %%ebx              \n\t" // xx
1303                 "xorl %%ecx, %%ecx              \n\t" // 2*xalpha
1304                 "1:                             \n\t"
1305                 "movzbl  (%0, %%ebx), %%edi     \n\t" //src[xx]
1306                 "movzbl 1(%0, %%ebx), %%esi     \n\t" //src[xx+1]
1307                 "subl %%edi, %%esi              \n\t" //src[xx+1] - src[xx]
1308                 "imull %%ecx, %%esi             \n\t" //(src[xx+1] - src[xx])*2*xalpha
1309                 "shll $16, %%edi                \n\t"
1310                 "addl %%edi, %%esi              \n\t" //src[xx+1]*2*xalpha + src[xx]*(1-2*xalpha)
1311                 "movl %1, %%edi                 \n\t"
1312                 "shrl $9, %%esi                 \n\t"
1313                 "movw %%si, (%%edi, %%eax, 2)   \n\t"
1314                 "addw %4, %%cx                  \n\t" //2*xalpha += xInc&0xFF
1315                 "adcl %3, %%ebx                 \n\t" //xx+= xInc>>8 + carry
1316
1317                 "movzbl (%0, %%ebx), %%edi      \n\t" //src[xx]
1318                 "movzbl 1(%0, %%ebx), %%esi     \n\t" //src[xx+1]
1319                 "subl %%edi, %%esi              \n\t" //src[xx+1] - src[xx]
1320                 "imull %%ecx, %%esi             \n\t" //(src[xx+1] - src[xx])*2*xalpha
1321                 "shll $16, %%edi                \n\t"
1322                 "addl %%edi, %%esi              \n\t" //src[xx+1]*2*xalpha + src[xx]*(1-2*xalpha)
1323                 "movl %1, %%edi                 \n\t"
1324                 "shrl $9, %%esi                 \n\t"
1325                 "movw %%si, 2(%%edi, %%eax, 2)  \n\t"
1326                 "addw %4, %%cx                  \n\t" //2*xalpha += xInc&0xFF
1327                 "adcl %3, %%ebx                 \n\t" //xx+= xInc>>8 + carry
1328
1329
1330                 "addl $2, %%eax                 \n\t"
1331                 "cmpl %2, %%eax                 \n\t"
1332                 " jb 1b                         \n\t"
1333
1334
1335                 :: "r" (src), "m" (dst), "m" (dstWidth), "m" (xInc>>16), "m" (xInc&0xFFFF)
1336                 : "%eax", "%ebx", "%ecx", "%edi", "%esi"
1337                 );
1338 #ifdef HAVE_MMX2
1339         } //if MMX2 cant be used
1340 #endif
1341 #else
1342         int i;
1343         unsigned int xpos=0;
1344         for(i=0;i<dstWidth;i++)
1345         {
1346                 register unsigned int xx=xpos>>16;
1347                 register unsigned int xalpha=(xpos&0xFFFF)>>9;
1348                 dst[i]= (src[xx]<<7) + (src[xx+1] - src[xx])*xalpha;
1349                 xpos+=xInc;
1350         }
1351 #endif
1352 }
1353
1354 inline static void hcscale(uint16_t *dst, int dstWidth,
1355                                 uint8_t *src1, uint8_t *src2, int srcWidth, int xInc)
1356 {
1357 #ifdef ARCH_X86
1358 #ifdef HAVE_MMX2
1359         int i;
1360         if(canMMX2BeUsed)
1361         {
1362                 asm volatile(
1363                 "pxor %%mm7, %%mm7              \n\t"
1364                 "pxor %%mm2, %%mm2              \n\t" // 2*xalpha
1365                 "movd %5, %%mm6                 \n\t" // xInc&0xFFFF
1366                 "punpcklwd %%mm6, %%mm6         \n\t"
1367                 "punpcklwd %%mm6, %%mm6         \n\t"
1368                 "movq %%mm6, %%mm2              \n\t"
1369                 "psllq $16, %%mm2               \n\t"
1370                 "paddw %%mm6, %%mm2             \n\t"
1371                 "psllq $16, %%mm2               \n\t"
1372                 "paddw %%mm6, %%mm2             \n\t"
1373                 "psllq $16, %%mm2               \n\t" //0,t,2t,3t               t=xInc&0xFFFF
1374                 "movq %%mm2, temp0              \n\t"
1375                 "movd %4, %%mm6                 \n\t" //(xInc*4)&0xFFFF
1376                 "punpcklwd %%mm6, %%mm6         \n\t"
1377                 "punpcklwd %%mm6, %%mm6         \n\t"
1378                 "xorl %%eax, %%eax              \n\t" // i
1379                 "movl %0, %%esi                 \n\t" // src
1380                 "movl %1, %%edi                 \n\t" // buf1
1381                 "movl %3, %%edx                 \n\t" // (xInc*4)>>16
1382                 "xorl %%ecx, %%ecx              \n\t"
1383                 "xorl %%ebx, %%ebx              \n\t"
1384                 "movw %4, %%bx                  \n\t" // (xInc*4)&0xFFFF
1385
1386 #define FUNNYUVCODE \
1387                         PREFETCH" 1024(%%esi)           \n\t"\
1388                         PREFETCH" 1056(%%esi)           \n\t"\
1389                         PREFETCH" 1088(%%esi)           \n\t"\
1390                         "call funnyUVCode               \n\t"\
1391                         "movq temp0, %%mm2              \n\t"\
1392                         "xorl %%ecx, %%ecx              \n\t"
1393
1394 FUNNYUVCODE
1395 FUNNYUVCODE
1396 FUNNYUVCODE
1397 FUNNYUVCODE
1398
1399 FUNNYUVCODE
1400 FUNNYUVCODE
1401 FUNNYUVCODE
1402 FUNNYUVCODE
1403                 "xorl %%eax, %%eax              \n\t" // i
1404                 "movl %6, %%esi                 \n\t" // src
1405                 "movl %1, %%edi                 \n\t" // buf1
1406                 "addl $4096, %%edi              \n\t"
1407
1408 FUNNYUVCODE
1409 FUNNYUVCODE
1410 FUNNYUVCODE
1411 FUNNYUVCODE
1412
1413 FUNNYUVCODE
1414 FUNNYUVCODE
1415 FUNNYUVCODE
1416 FUNNYUVCODE
1417
1418                 :: "m" (src1), "m" (dst), "m" (dstWidth), "m" ((xInc*4)>>16),
1419                   "m" ((xInc*4)&0xFFFF), "m" (xInc&0xFFFF), "m" (src2)
1420                 : "%eax", "%ebx", "%ecx", "%edx", "%esi", "%edi"
1421         );
1422                 for(i=dstWidth-1; (i*xInc)>>16 >=srcWidth/2-1; i--)
1423                 {
1424                         dst[i] = src1[srcWidth/2-1]*128;
1425                         dst[i+2048] = src2[srcWidth/2-1]*128;
1426                 }
1427         }
1428         else
1429         {
1430 #endif
1431         asm volatile(
1432                 "xorl %%eax, %%eax              \n\t" // i
1433                 "xorl %%ebx, %%ebx              \n\t" // xx
1434                 "xorl %%ecx, %%ecx              \n\t" // 2*xalpha
1435                 "1:                             \n\t"
1436                 "movl %0, %%esi                 \n\t"
1437                 "movzbl  (%%esi, %%ebx), %%edi  \n\t" //src[xx]
1438                 "movzbl 1(%%esi, %%ebx), %%esi  \n\t" //src[xx+1]
1439                 "subl %%edi, %%esi              \n\t" //src[xx+1] - src[xx]
1440                 "imull %%ecx, %%esi             \n\t" //(src[xx+1] - src[xx])*2*xalpha
1441                 "shll $16, %%edi                \n\t"
1442                 "addl %%edi, %%esi              \n\t" //src[xx+1]*2*xalpha + src[xx]*(1-2*xalpha)
1443                 "movl %1, %%edi                 \n\t"
1444                 "shrl $9, %%esi                 \n\t"
1445                 "movw %%si, (%%edi, %%eax, 2)   \n\t"
1446
1447                 "movzbl  (%5, %%ebx), %%edi     \n\t" //src[xx]
1448                 "movzbl 1(%5, %%ebx), %%esi     \n\t" //src[xx+1]
1449                 "subl %%edi, %%esi              \n\t" //src[xx+1] - src[xx]
1450                 "imull %%ecx, %%esi             \n\t" //(src[xx+1] - src[xx])*2*xalpha
1451                 "shll $16, %%edi                \n\t"
1452                 "addl %%edi, %%esi              \n\t" //src[xx+1]*2*xalpha + src[xx]*(1-2*xalpha)
1453                 "movl %1, %%edi                 \n\t"
1454                 "shrl $9, %%esi                 \n\t"
1455                 "movw %%si, 4096(%%edi, %%eax, 2)\n\t"
1456
1457                 "addw %4, %%cx                  \n\t" //2*xalpha += xInc&0xFF
1458                 "adcl %3, %%ebx                 \n\t" //xx+= xInc>>8 + carry
1459                 "addl $1, %%eax                 \n\t"
1460                 "cmpl %2, %%eax                 \n\t"
1461                 " jb 1b                         \n\t"
1462
1463                 :: "m" (src1), "m" (dst), "m" (dstWidth), "m" (xInc>>16), "m" (xInc&0xFFFF),
1464                 "r" (src2)
1465                 : "%eax", "%ebx", "%ecx", "%edi", "%esi"
1466                 );
1467 #ifdef HAVE_MMX2
1468         } //if MMX2 cant be used
1469 #endif
1470 #else
1471         int i;
1472         unsigned int xpos=0;
1473         for(i=0;i<dstWidth;i++)
1474         {
1475                 register unsigned int xx=xpos>>16;
1476                 register unsigned int xalpha=(xpos&0xFFFF)>>9;
1477                 dst[i]=(src1[xx]*(xalpha^127)+src1[xx+1]*xalpha);
1478                 dst[i+2048]=(src2[xx]*(xalpha^127)+src2[xx+1]*xalpha);
1479 /* slower
1480           dst[i]= (src1[xx]<<7) + (src1[xx+1] - src1[xx])*xalpha;
1481           dst[i+2048]=(src2[xx]<<7) + (src2[xx+1] - src2[xx])*xalpha;
1482 */
1483                 xpos+=xInc;
1484         }
1485 #endif
1486 }
1487
1488
1489 // *** bilinear scaling and yuv->rgb or yuv->yuv conversion of yv12 slices:
1490 // *** Note: it's called multiple times while decoding a frame, first time y==0
1491 // *** Designed to upscale, but may work for downscale too.
1492 // s_xinc = (src_width << 16) / dst_width
1493 // s_yinc = (src_height << 16) / dst_height
1494 void SwScale_YV12slice(unsigned char* srcptr[],int stride[], int y, int h,
1495                              uint8_t* dstptr[], int dststride, int dstw, int dstbpp,
1496                              unsigned int s_xinc,unsigned int s_yinc){
1497
1498 // scaling factors:
1499 //static int s_yinc=(vo_dga_src_height<<16)/vo_dga_vp_height;
1500 //static int s_xinc=(vo_dga_src_width<<8)/vo_dga_vp_width;
1501
1502 unsigned int s_xinc2;
1503
1504 static int s_srcypos; // points to the dst Pixels center in the source (0 is the center of pixel 0,0 in src)
1505 static int s_ypos;
1506
1507 // last horzontally interpolated lines, used to avoid unnecessary calculations
1508 static int s_last_ypos;
1509 static int s_last_y1pos;
1510
1511 #ifdef HAVE_MMX2
1512 // used to detect a horizontal size change
1513 static int old_dstw= -1;
1514 static int old_s_xinc= -1;
1515 #endif
1516
1517 int srcWidth;
1518 int dstUVw;
1519 int i;
1520
1521 if(((dstw + 7)&(~7)) >= dststride) dstw&= ~7;
1522
1523 srcWidth= (dstw*s_xinc + 0x8000)>>16;
1524 dstUVw= fullUVIpol ? dstw : dstw/2;
1525
1526 #ifdef HAVE_MMX2
1527 canMMX2BeUsed= (s_xinc <= 0x10000 && (dstw&31)==0 && (srcWidth&15)==0) ? 1 : 0;
1528 #endif
1529
1530 // match pixel 0 of the src to pixel 0 of dst and match pixel n-2 of src to pixel n-2 of dst
1531 // n-2 is the last chrominance sample available
1532 // FIXME this is not perfect, but noone shuld notice the difference, the more correct variant
1533 // would be like the vertical one, but that would require some special code for the
1534 // first and last pixel
1535 if(canMMX2BeUsed)       s_xinc+= 20;
1536 else                    s_xinc = ((srcWidth-2)<<16)/(dstw-2) - 20;
1537
1538 if(fullUVIpol && !(dstbpp==12))         s_xinc2= s_xinc>>1;
1539 else                                    s_xinc2= s_xinc;
1540   // force calculation of the horizontal interpolation of the first line
1541
1542   if(y==0){
1543 //      printf("dstw %d, srcw %d, mmx2 %d\n", dstw, srcWidth, canMMX2BeUsed);
1544         s_last_ypos=-99;
1545         s_last_y1pos=-99;
1546         s_srcypos= s_yinc/2 - 0x8000;
1547         s_ypos=0;
1548
1549         // clean the buffers so that no green stuff is drawen if the width is not sane (%8=0)
1550         for(i=dstw-2; i<dstw+20; i++)
1551         {
1552                 pix_buf_uv[0][i] = pix_buf_uv[1][i]
1553                 = pix_buf_uv[0][2048+i] = pix_buf_uv[1][2048+i] = 128*128;
1554                 pix_buf_uv[0][i/2] = pix_buf_uv[1][i/2]
1555                 = pix_buf_uv[0][2048+i/2] = pix_buf_uv[1][2048+i/2] = 128*128;
1556                 pix_buf_y[0][i]= pix_buf_y[1][i]= 0;
1557         }
1558
1559 #ifdef HAVE_MMX2
1560 // cant downscale !!!
1561         if((old_s_xinc != s_xinc || old_dstw!=dstw) && canMMX2BeUsed)
1562         {
1563                 uint8_t *fragment;
1564                 int imm8OfPShufW1;
1565                 int imm8OfPShufW2;
1566                 int fragmentLength;
1567
1568                 int xpos, i;
1569
1570                 old_s_xinc= s_xinc;
1571                 old_dstw= dstw;
1572
1573                 // create an optimized horizontal scaling routine
1574
1575                 //code fragment
1576
1577                 asm volatile(
1578                         "jmp 9f                         \n\t"
1579                 // Begin
1580                         "0:                             \n\t"
1581                         "movq (%%esi), %%mm0            \n\t" //FIXME Alignment
1582                         "movq %%mm0, %%mm1              \n\t"
1583                         "psrlq $8, %%mm0                \n\t"
1584                         "punpcklbw %%mm7, %%mm1 \n\t"
1585                         "movq %%mm2, %%mm3              \n\t"
1586                         "punpcklbw %%mm7, %%mm0 \n\t"
1587                         "addw %%bx, %%cx                \n\t" //2*xalpha += (4*s_xinc)&0xFFFF
1588                         "pshufw $0xFF, %%mm1, %%mm1     \n\t"
1589                         "1:                             \n\t"
1590                         "adcl %%edx, %%esi              \n\t" //xx+= (4*s_xinc)>>16 + carry
1591                         "pshufw $0xFF, %%mm0, %%mm0     \n\t"
1592                         "2:                             \n\t"
1593                         "psrlw $9, %%mm3                \n\t"
1594                         "psubw %%mm1, %%mm0             \n\t"
1595                         "pmullw %%mm3, %%mm0            \n\t"
1596                         "paddw %%mm6, %%mm2             \n\t" // 2*alpha += xpos&0xFFFF
1597                         "psllw $7, %%mm1                \n\t"
1598                         "paddw %%mm1, %%mm0             \n\t"
1599
1600                         "movq %%mm0, (%%edi, %%eax)     \n\t"
1601
1602                         "addl $8, %%eax                 \n\t"
1603                 // End
1604                         "9:                             \n\t"
1605 //              "int $3\n\t"
1606                         "leal 0b, %0                    \n\t"
1607                         "leal 1b, %1                    \n\t"
1608                         "leal 2b, %2                    \n\t"
1609                         "decl %1                        \n\t"
1610                         "decl %2                        \n\t"
1611                         "subl %0, %1                    \n\t"
1612                         "subl %0, %2                    \n\t"
1613                         "leal 9b, %3                    \n\t"
1614                         "subl %0, %3                    \n\t"
1615                         :"=r" (fragment), "=r" (imm8OfPShufW1), "=r" (imm8OfPShufW2),
1616                          "=r" (fragmentLength)
1617                 );
1618
1619                 xpos= 0; //s_xinc/2 - 0x8000; // difference between pixel centers
1620
1621                 /* choose xinc so that all 8 parts fit exactly
1622                    Note: we cannot use just 1 part because it would not fit in the code cache */
1623 //              s_xinc2_diff= -((((s_xinc2*(dstw/8))&0xFFFF))/(dstw/8))-10;
1624 //              s_xinc_diff= -((((s_xinc*(dstw/8))&0xFFFF))/(dstw/8));
1625 #ifdef ALT_ERROR
1626 //              s_xinc2_diff+= ((0x10000/(dstw/8)));
1627 #endif
1628 //              s_xinc_diff= s_xinc2_diff*2;
1629
1630 //              s_xinc2+= s_xinc2_diff;
1631 //              s_xinc+= s_xinc_diff;
1632
1633 //              old_s_xinc= s_xinc;
1634
1635                 for(i=0; i<dstw/8; i++)
1636                 {
1637                         int xx=xpos>>16;
1638
1639                         if((i&3) == 0)
1640                         {
1641                                 int a=0;
1642                                 int b=((xpos+s_xinc)>>16) - xx;
1643                                 int c=((xpos+s_xinc*2)>>16) - xx;
1644                                 int d=((xpos+s_xinc*3)>>16) - xx;
1645
1646                                 memcpy(funnyYCode + fragmentLength*i/4, fragment, fragmentLength);
1647
1648                                 funnyYCode[fragmentLength*i/4 + imm8OfPShufW1]=
1649                                 funnyYCode[fragmentLength*i/4 + imm8OfPShufW2]=
1650                                         a | (b<<2) | (c<<4) | (d<<6);
1651
1652                                 // if we dont need to read 8 bytes than dont :), reduces the chance of
1653                                 // crossing a cache line
1654                                 if(d<3) funnyYCode[fragmentLength*i/4 + 1]= 0x6E;
1655
1656                                 funnyYCode[fragmentLength*(i+4)/4]= RET;
1657                         }
1658                         xpos+=s_xinc;
1659                 }
1660
1661                 xpos= 0; //s_xinc2/2 - 0x10000; // difference between centers of chrom samples
1662                 for(i=0; i<dstUVw/8; i++)
1663                 {
1664                         int xx=xpos>>16;
1665
1666                         if((i&3) == 0)
1667                         {
1668                                 int a=0;
1669                                 int b=((xpos+s_xinc2)>>16) - xx;
1670                                 int c=((xpos+s_xinc2*2)>>16) - xx;
1671                                 int d=((xpos+s_xinc2*3)>>16) - xx;
1672
1673                                 memcpy(funnyUVCode + fragmentLength*i/4, fragment, fragmentLength);
1674
1675                                 funnyUVCode[fragmentLength*i/4 + imm8OfPShufW1]=
1676                                 funnyUVCode[fragmentLength*i/4 + imm8OfPShufW2]=
1677                                         a | (b<<2) | (c<<4) | (d<<6);
1678
1679                                 // if we dont need to read 8 bytes than dont :), reduces the chance of
1680                                 // crossing a cache line
1681                                 if(d<3) funnyUVCode[fragmentLength*i/4 + 1]= 0x6E;
1682
1683                                 funnyUVCode[fragmentLength*(i+4)/4]= RET;
1684                         }
1685                         xpos+=s_xinc2;
1686                 }
1687 //              funnyCode[0]= RET;
1688         }
1689
1690 #endif // HAVE_MMX2
1691   } // reset counters
1692
1693   while(1){
1694     unsigned char *dest =dstptr[0]+dststride*s_ypos;
1695     unsigned char *uDest=dstptr[1]+(dststride>>1)*(s_ypos>>1);
1696     unsigned char *vDest=dstptr[2]+(dststride>>1)*(s_ypos>>1);
1697
1698     int y0=(s_srcypos + 0xFFFF)>>16;  // first luminance source line number below the dst line
1699         // points to the dst Pixels center in the source (0 is the center of pixel 0,0 in src)
1700     int srcuvpos= dstbpp==12 ?  s_srcypos + s_yinc/2 - 0x8000 :
1701                                 s_srcypos - 0x8000;
1702     int y1=(srcuvpos + 0x1FFFF)>>17; // first chrominance source line number below the dst line
1703     int yalpha=((s_srcypos-1)&0xFFFF)>>4;
1704     int uvalpha=((srcuvpos-1)&0x1FFFF)>>5;
1705     uint16_t *buf0=pix_buf_y[y0&1];             // top line of the interpolated slice
1706     uint16_t *buf1=pix_buf_y[((y0+1)&1)];       // bottom line of the interpolated slice
1707     uint16_t *uvbuf0=pix_buf_uv[y1&1];          // top line of the interpolated slice
1708     uint16_t *uvbuf1=pix_buf_uv[(y1+1)&1];      // bottom line of the interpolated slice
1709
1710     if(y0>=y+h) break; // FIXME wrong, skips last lines, but they are dupliactes anyway
1711
1712     if((y0&1) && dstbpp==12) uvalpha=-1; // there is no alpha if there is no line
1713
1714     s_ypos++; s_srcypos+=s_yinc;
1715
1716     //only interpolate the src line horizontally if we didnt do it allready
1717         if(s_last_ypos!=y0)
1718         {
1719                 unsigned char *src;
1720                 // skip if first line has been horiz scaled alleady
1721                 if(s_last_ypos != y0-1)
1722                 {
1723                         // check if first line is before any available src lines
1724                         if(y0-1 < y)    src=srcptr[0]+(0     )*stride[0];
1725                         else            src=srcptr[0]+(y0-y-1)*stride[0];
1726
1727                         hyscale(buf0, dstw, src, srcWidth, s_xinc);
1728                 }
1729                 // check if second line is after any available src lines
1730                 if(y0-y >= h)   src=srcptr[0]+(h-1)*stride[0];
1731                 else            src=srcptr[0]+(y0-y)*stride[0];
1732
1733                 // the min() is required to avoid reuseing lines which where not available
1734                 s_last_ypos= MIN(y0, y+h-1);
1735                 hyscale(buf1, dstw, src, srcWidth, s_xinc);
1736         }
1737 //      printf("%d %d %d %d\n", y, y1, s_last_y1pos, h);
1738       // *** horizontal scale U and V lines to temp buffer
1739         if(s_last_y1pos!=y1)
1740         {
1741                 uint8_t *src1, *src2;
1742                 // skip if first line has been horiz scaled alleady
1743                 if(s_last_y1pos != y1-1)
1744                 {
1745                         // check if first line is before any available src lines
1746                         if(y1-y/2-1 < 0)
1747                         {
1748                                 src1= srcptr[1]+(0)*stride[1];
1749                                 src2= srcptr[2]+(0)*stride[2];
1750                         }else{
1751                                 src1= srcptr[1]+(y1-y/2-1)*stride[1];
1752                                 src2= srcptr[2]+(y1-y/2-1)*stride[2];
1753                         }
1754                         hcscale(uvbuf0, dstUVw, src1, src2, srcWidth, s_xinc2);
1755                 }
1756
1757                 // check if second line is after any available src lines
1758                 if(y1 - y/2 >= h/2)
1759                 {
1760                         src1= srcptr[1]+(h/2-1)*stride[1];
1761                         src2= srcptr[2]+(h/2-1)*stride[2];
1762                 }else{
1763                         src1= srcptr[1]+(y1-y/2)*stride[1];
1764                         src2= srcptr[2]+(y1-y/2)*stride[2];
1765                 }
1766                 hcscale(uvbuf1, dstUVw, src1, src2, srcWidth, s_xinc2);
1767
1768                 // the min() is required to avoid reuseing lines which where not available
1769                 s_last_y1pos= MIN(y1, y/2+h/2-1);
1770         }
1771
1772         if(dstbpp==12) //YV12
1773                 yuv2yuv(buf0, buf1, uvbuf0, uvbuf1, dest, uDest, vDest, dstw, yalpha, uvalpha);
1774         else if(ABS(s_yinc - 0x10000) < 10)
1775                 yuv2rgb1(buf0, buf1, uvbuf0, uvbuf1, dest, dstw, yalpha, uvalpha, dstbpp);
1776         else
1777                 yuv2rgbX(buf0, buf1, uvbuf0, uvbuf1, dest, dstw, yalpha, uvalpha, dstbpp);
1778
1779 #ifdef HAVE_MMX
1780         b16Dither= b16Dither1;
1781         b16Dither1= b16Dither2;
1782         b16Dither2= b16Dither;
1783
1784         g16Dither= g16Dither1;
1785         g16Dither1= g16Dither2;
1786         g16Dither2= g16Dither;
1787 #endif
1788   }
1789
1790 #ifdef HAVE_MMX
1791         __asm __volatile(SFENCE:::"memory");
1792         __asm __volatile(EMMS:::"memory");
1793 #endif
1794 }
1795
1796
1797 void SwScale_Init(){
1798     // generating tables:
1799     int i;
1800     for(i=0;i<256;i++){
1801         clip_table[i]=0;
1802         clip_table[i+256]=i;
1803         clip_table[i+512]=255;
1804         yuvtab_2568[i]=(0x2568*(i-16))+(256<<13);
1805         yuvtab_3343[i]=0x3343*(i-128);
1806         yuvtab_0c92[i]=-0x0c92*(i-128);
1807         yuvtab_1a1e[i]=-0x1a1e*(i-128);
1808         yuvtab_40cf[i]=0x40cf*(i-128);
1809     }
1810
1811     for(i=0; i<768; i++)
1812     {
1813         int v= clip_table[i];
1814         clip_table16b[i]= v>>3;
1815         clip_table16g[i]= (v<<3)&0x07E0;
1816         clip_table16r[i]= (v<<8)&0xF800;
1817         clip_table15b[i]= v>>3;
1818         clip_table15g[i]= (v<<2)&0x03E0;
1819         clip_table15r[i]= (v<<7)&0x7C00;
1820     }
1821
1822 }