swscale_template.c 75.9 KB
Newer Older
1 2 3

// Software scaling and colorspace conversion routines for MPlayer

arpi's avatar
arpi committed
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

michael's avatar
michael committed
8
#undef MOVNTQ
michael's avatar
michael committed
9
#undef PAVGB
michael's avatar
oops  
michael committed
10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37
#undef PREFETCH
#undef PREFETCHW
#undef EMMS
#undef SFENCE

#ifdef HAVE_3DNOW
/* On K6 femms is faster of emms. On K7 femms is directly mapped on emms. */
#define EMMS     "femms"
#else
#define EMMS     "emms"
#endif

#ifdef HAVE_3DNOW
#define PREFETCH  "prefetch"
#define PREFETCHW "prefetchw"
#elif defined ( HAVE_MMX2 )
#define PREFETCH "prefetchnta"
#define PREFETCHW "prefetcht0"
#else
#define PREFETCH "/nop"
#define PREFETCHW "/nop"
#endif

#ifdef HAVE_MMX2
#define SFENCE "sfence"
#else
#define SFENCE "/nop"
#endif
38

39 40 41 42 43
#ifdef HAVE_MMX2
#define PAVGB(a,b) "pavgb " #a ", " #b " \n\t"
#elif defined (HAVE_3DNOW)
#define PAVGB(a,b) "pavgusb " #a ", " #b " \n\t"
#endif
44

45 46 47 48 49 50
#ifdef HAVE_MMX2
#define MOVNTQ(a,b) "movntq " #a ", " #b " \n\t"
#else
#define MOVNTQ(a,b) "movq " #a ", " #b " \n\t"
#endif

michael's avatar
michael committed
51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173

#define YSCALEYUV2YV12X(x) \
			"xorl %%eax, %%eax		\n\t"\
			"pxor %%mm3, %%mm3		\n\t"\
			"pxor %%mm4, %%mm4		\n\t"\
			"movl %0, %%edx			\n\t"\
			".balign 16			\n\t" /* FIXME Unroll? */\
			"1:				\n\t"\
			"movl (%1, %%edx, 4), %%esi	\n\t"\
			"movq (%2, %%edx, 8), %%mm0	\n\t" /* filterCoeff */\
			"movq " #x "(%%esi, %%eax, 2), %%mm2	\n\t" /* srcData */\
			"movq 8+" #x "(%%esi, %%eax, 2), %%mm5	\n\t" /* srcData */\
			"pmulhw %%mm0, %%mm2		\n\t"\
			"pmulhw %%mm0, %%mm5		\n\t"\
			"paddw %%mm2, %%mm3		\n\t"\
			"paddw %%mm5, %%mm4		\n\t"\
			"addl $1, %%edx			\n\t"\
			" jnz 1b			\n\t"\
			"psraw $3, %%mm3		\n\t"\
			"psraw $3, %%mm4		\n\t"\
			"packuswb %%mm4, %%mm3		\n\t"\
			MOVNTQ(%%mm3, (%3, %%eax))\
			"addl $8, %%eax			\n\t"\
			"cmpl %4, %%eax			\n\t"\
			"pxor %%mm3, %%mm3		\n\t"\
			"pxor %%mm4, %%mm4		\n\t"\
			"movl %0, %%edx			\n\t"\
			"jb 1b				\n\t"

#define YSCALEYUV2YV121 \
			"movl %2, %%eax			\n\t"\
			".balign 16			\n\t" /* FIXME Unroll? */\
			"1:				\n\t"\
			"movq (%0, %%eax, 2), %%mm0	\n\t"\
			"movq 8(%0, %%eax, 2), %%mm1	\n\t"\
			"psraw $7, %%mm0		\n\t"\
			"psraw $7, %%mm1		\n\t"\
			"packuswb %%mm1, %%mm0		\n\t"\
			MOVNTQ(%%mm0, (%1, %%eax))\
			"addl $8, %%eax			\n\t"\
			"jnc 1b				\n\t"

/*
			:: "m" (-lumFilterSize), "m" (-chrFilterSize),
			   "m" (lumMmxFilter+lumFilterSize*4), "m" (chrMmxFilter+chrFilterSize*4),
			   "r" (dest), "m" (dstW),
			   "m" (lumSrc+lumFilterSize), "m" (chrSrc+chrFilterSize)
			: "%eax", "%ebx", "%ecx", "%edx", "%esi"
*/
#define YSCALEYUV2RGBX \
		"xorl %%eax, %%eax		\n\t"\
		".balign 16			\n\t"\
		"1:				\n\t"\
		"movl %1, %%edx			\n\t" /* -chrFilterSize */\
		"movl %3, %%ebx			\n\t" /* chrMmxFilter+lumFilterSize */\
		"movl %7, %%ecx			\n\t" /* chrSrc+lumFilterSize */\
		"pxor %%mm3, %%mm3		\n\t"\
		"pxor %%mm4, %%mm4		\n\t"\
		"2:				\n\t"\
		"movl (%%ecx, %%edx, 4), %%esi	\n\t"\
		"movq (%%ebx, %%edx, 8), %%mm0	\n\t" /* filterCoeff */\
		"movq (%%esi, %%eax), %%mm2	\n\t" /* UsrcData */\
		"movq 4096(%%esi, %%eax), %%mm5	\n\t" /* VsrcData */\
		"pmulhw %%mm0, %%mm2		\n\t"\
		"pmulhw %%mm0, %%mm5		\n\t"\
		"paddw %%mm2, %%mm3		\n\t"\
		"paddw %%mm5, %%mm4		\n\t"\
		"addl $1, %%edx			\n\t"\
		" jnz 2b			\n\t"\
\
		"movl %0, %%edx			\n\t" /* -lumFilterSize */\
		"movl %2, %%ebx			\n\t" /* lumMmxFilter+lumFilterSize */\
		"movl %6, %%ecx			\n\t" /* lumSrc+lumFilterSize */\
		"pxor %%mm1, %%mm1		\n\t"\
		"pxor %%mm7, %%mm7		\n\t"\
		"2:				\n\t"\
		"movl (%%ecx, %%edx, 4), %%esi	\n\t"\
		"movq (%%ebx, %%edx, 8), %%mm0	\n\t" /* filterCoeff */\
		"movq (%%esi, %%eax, 2), %%mm2	\n\t" /* Y1srcData */\
		"movq 8(%%esi, %%eax, 2), %%mm5	\n\t" /* Y2srcData */\
		"pmulhw %%mm0, %%mm2		\n\t"\
		"pmulhw %%mm0, %%mm5		\n\t"\
		"paddw %%mm2, %%mm1		\n\t"\
		"paddw %%mm5, %%mm7		\n\t"\
		"addl $1, %%edx			\n\t"\
		" jnz 2b			\n\t"\
\
		"psubw w400, %%mm3		\n\t" /* (U-128)8*/\
		"psubw w400, %%mm4		\n\t" /* (V-128)8*/\
		"movq %%mm3, %%mm2		\n\t" /* (U-128)8*/\
		"movq %%mm4, %%mm5		\n\t" /* (V-128)8*/\
		"pmulhw ugCoeff, %%mm3		\n\t"\
		"pmulhw vgCoeff, %%mm4		\n\t"\
	/* mm2=(U-128)8, mm3=ug, mm4=vg mm5=(V-128)8 */\
		"pmulhw ubCoeff, %%mm2		\n\t"\
		"pmulhw vrCoeff, %%mm5		\n\t"\
		"psubw w80, %%mm1		\n\t" /* 8(Y-16)*/\
		"psubw w80, %%mm7		\n\t" /* 8(Y-16)*/\
		"pmulhw yCoeff, %%mm1		\n\t"\
		"pmulhw yCoeff, %%mm7		\n\t"\
	/* mm1= Y1, mm2=ub, mm3=ug, mm4=vg mm5=vr, mm7=Y2 */\
		"paddw %%mm3, %%mm4		\n\t"\
		"movq %%mm2, %%mm0		\n\t"\
		"movq %%mm5, %%mm6		\n\t"\
		"movq %%mm4, %%mm3		\n\t"\
		"punpcklwd %%mm2, %%mm2		\n\t"\
		"punpcklwd %%mm5, %%mm5		\n\t"\
		"punpcklwd %%mm4, %%mm4		\n\t"\
		"paddw %%mm1, %%mm2		\n\t"\
		"paddw %%mm1, %%mm5		\n\t"\
		"paddw %%mm1, %%mm4		\n\t"\
		"punpckhwd %%mm0, %%mm0		\n\t"\
		"punpckhwd %%mm6, %%mm6		\n\t"\
		"punpckhwd %%mm3, %%mm3		\n\t"\
		"paddw %%mm7, %%mm0		\n\t"\
		"paddw %%mm7, %%mm6		\n\t"\
		"paddw %%mm7, %%mm3		\n\t"\
		/* mm0=B1, mm2=B2, mm3=G2, mm4=G1, mm5=R1, mm6=R2 */\
		"packuswb %%mm0, %%mm2		\n\t"\
		"packuswb %%mm6, %%mm5		\n\t"\
		"packuswb %%mm3, %%mm4		\n\t"\
		"pxor %%mm7, %%mm7		\n\t"

174 175 176 177 178 179 180 181 182
#define FULL_YSCALEYUV2RGB \
		"pxor %%mm7, %%mm7		\n\t"\
		"movd %6, %%mm6			\n\t" /*yalpha1*/\
		"punpcklwd %%mm6, %%mm6		\n\t"\
		"punpcklwd %%mm6, %%mm6		\n\t"\
		"movd %7, %%mm5			\n\t" /*uvalpha1*/\
		"punpcklwd %%mm5, %%mm5		\n\t"\
		"punpcklwd %%mm5, %%mm5		\n\t"\
		"xorl %%eax, %%eax		\n\t"\
183
		".balign 16			\n\t"\
184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236
		"1:				\n\t"\
		"movq (%0, %%eax, 2), %%mm0	\n\t" /*buf0[eax]*/\
		"movq (%1, %%eax, 2), %%mm1	\n\t" /*buf1[eax]*/\
		"movq (%2, %%eax,2), %%mm2	\n\t" /* uvbuf0[eax]*/\
		"movq (%3, %%eax,2), %%mm3	\n\t" /* uvbuf1[eax]*/\
		"psubw %%mm1, %%mm0		\n\t" /* buf0[eax] - buf1[eax]*/\
		"psubw %%mm3, %%mm2		\n\t" /* uvbuf0[eax] - uvbuf1[eax]*/\
		"pmulhw %%mm6, %%mm0		\n\t" /* (buf0[eax] - buf1[eax])yalpha1>>16*/\
		"pmulhw %%mm5, %%mm2		\n\t" /* (uvbuf0[eax] - uvbuf1[eax])uvalpha1>>16*/\
		"psraw $4, %%mm1		\n\t" /* buf0[eax] - buf1[eax] >>4*/\
		"movq 4096(%2, %%eax,2), %%mm4	\n\t" /* uvbuf0[eax+2048]*/\
		"psraw $4, %%mm3		\n\t" /* uvbuf0[eax] - uvbuf1[eax] >>4*/\
		"paddw %%mm0, %%mm1		\n\t" /* buf0[eax]yalpha1 + buf1[eax](1-yalpha1) >>16*/\
		"movq 4096(%3, %%eax,2), %%mm0	\n\t" /* uvbuf1[eax+2048]*/\
		"paddw %%mm2, %%mm3		\n\t" /* uvbuf0[eax]uvalpha1 - uvbuf1[eax](1-uvalpha1)*/\
		"psubw %%mm0, %%mm4		\n\t" /* uvbuf0[eax+2048] - uvbuf1[eax+2048]*/\
		"psubw w80, %%mm1		\n\t" /* 8(Y-16)*/\
		"psubw w400, %%mm3		\n\t" /* 8(U-128)*/\
		"pmulhw yCoeff, %%mm1		\n\t"\
\
\
		"pmulhw %%mm5, %%mm4		\n\t" /* (uvbuf0[eax+2048] - uvbuf1[eax+2048])uvalpha1>>16*/\
		"movq %%mm3, %%mm2		\n\t" /* (U-128)8*/\
		"pmulhw ubCoeff, %%mm3		\n\t"\
		"psraw $4, %%mm0		\n\t" /* uvbuf0[eax+2048] - uvbuf1[eax+2048] >>4*/\
		"pmulhw ugCoeff, %%mm2		\n\t"\
		"paddw %%mm4, %%mm0		\n\t" /* uvbuf0[eax+2048]uvalpha1 - uvbuf1[eax+2048](1-uvalpha1)*/\
		"psubw w400, %%mm0		\n\t" /* (V-128)8*/\
\
\
		"movq %%mm0, %%mm4		\n\t" /* (V-128)8*/\
		"pmulhw vrCoeff, %%mm0		\n\t"\
		"pmulhw vgCoeff, %%mm4		\n\t"\
		"paddw %%mm1, %%mm3		\n\t" /* B*/\
		"paddw %%mm1, %%mm0		\n\t" /* R*/\
		"packuswb %%mm3, %%mm3		\n\t"\
\
		"packuswb %%mm0, %%mm0		\n\t"\
		"paddw %%mm4, %%mm2		\n\t"\
		"paddw %%mm2, %%mm1		\n\t" /* G*/\
\
		"packuswb %%mm1, %%mm1		\n\t"

#define YSCALEYUV2RGB \
		"movd %6, %%mm6			\n\t" /*yalpha1*/\
		"punpcklwd %%mm6, %%mm6		\n\t"\
		"punpcklwd %%mm6, %%mm6		\n\t"\
		"movq %%mm6, asm_yalpha1	\n\t"\
		"movd %7, %%mm5			\n\t" /*uvalpha1*/\
		"punpcklwd %%mm5, %%mm5		\n\t"\
		"punpcklwd %%mm5, %%mm5		\n\t"\
		"movq %%mm5, asm_uvalpha1	\n\t"\
		"xorl %%eax, %%eax		\n\t"\
237
		".balign 16			\n\t"\
238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301
		"1:				\n\t"\
		"movq (%2, %%eax), %%mm2	\n\t" /* uvbuf0[eax]*/\
		"movq (%3, %%eax), %%mm3	\n\t" /* uvbuf1[eax]*/\
		"movq 4096(%2, %%eax), %%mm5	\n\t" /* uvbuf0[eax+2048]*/\
		"movq 4096(%3, %%eax), %%mm4	\n\t" /* uvbuf1[eax+2048]*/\
		"psubw %%mm3, %%mm2		\n\t" /* uvbuf0[eax] - uvbuf1[eax]*/\
		"psubw %%mm4, %%mm5		\n\t" /* uvbuf0[eax+2048] - uvbuf1[eax+2048]*/\
		"movq asm_uvalpha1, %%mm0	\n\t"\
		"pmulhw %%mm0, %%mm2		\n\t" /* (uvbuf0[eax] - uvbuf1[eax])uvalpha1>>16*/\
		"pmulhw %%mm0, %%mm5		\n\t" /* (uvbuf0[eax+2048] - uvbuf1[eax+2048])uvalpha1>>16*/\
		"psraw $4, %%mm3		\n\t" /* uvbuf0[eax] - uvbuf1[eax] >>4*/\
		"psraw $4, %%mm4		\n\t" /* uvbuf0[eax+2048] - uvbuf1[eax+2048] >>4*/\
		"paddw %%mm2, %%mm3		\n\t" /* uvbuf0[eax]uvalpha1 - uvbuf1[eax](1-uvalpha1)*/\
		"paddw %%mm5, %%mm4		\n\t" /* uvbuf0[eax+2048]uvalpha1 - uvbuf1[eax+2048](1-uvalpha1)*/\
		"psubw w400, %%mm3		\n\t" /* (U-128)8*/\
		"psubw w400, %%mm4		\n\t" /* (V-128)8*/\
		"movq %%mm3, %%mm2		\n\t" /* (U-128)8*/\
		"movq %%mm4, %%mm5		\n\t" /* (V-128)8*/\
		"pmulhw ugCoeff, %%mm3		\n\t"\
		"pmulhw vgCoeff, %%mm4		\n\t"\
	/* mm2=(U-128)8, mm3=ug, mm4=vg mm5=(V-128)8 */\
		"movq (%0, %%eax, 2), %%mm0	\n\t" /*buf0[eax]*/\
		"movq (%1, %%eax, 2), %%mm1	\n\t" /*buf1[eax]*/\
		"movq 8(%0, %%eax, 2), %%mm6	\n\t" /*buf0[eax]*/\
		"movq 8(%1, %%eax, 2), %%mm7	\n\t" /*buf1[eax]*/\
		"psubw %%mm1, %%mm0		\n\t" /* buf0[eax] - buf1[eax]*/\
		"psubw %%mm7, %%mm6		\n\t" /* buf0[eax] - buf1[eax]*/\
		"pmulhw asm_yalpha1, %%mm0	\n\t" /* (buf0[eax] - buf1[eax])yalpha1>>16*/\
		"pmulhw asm_yalpha1, %%mm6	\n\t" /* (buf0[eax] - buf1[eax])yalpha1>>16*/\
		"psraw $4, %%mm1		\n\t" /* buf0[eax] - buf1[eax] >>4*/\
		"psraw $4, %%mm7		\n\t" /* buf0[eax] - buf1[eax] >>4*/\
		"paddw %%mm0, %%mm1		\n\t" /* buf0[eax]yalpha1 + buf1[eax](1-yalpha1) >>16*/\
		"paddw %%mm6, %%mm7		\n\t" /* buf0[eax]yalpha1 + buf1[eax](1-yalpha1) >>16*/\
		"pmulhw ubCoeff, %%mm2		\n\t"\
		"pmulhw vrCoeff, %%mm5		\n\t"\
		"psubw w80, %%mm1		\n\t" /* 8(Y-16)*/\
		"psubw w80, %%mm7		\n\t" /* 8(Y-16)*/\
		"pmulhw yCoeff, %%mm1		\n\t"\
		"pmulhw yCoeff, %%mm7		\n\t"\
	/* mm1= Y1, mm2=ub, mm3=ug, mm4=vg mm5=vr, mm7=Y2 */\
		"paddw %%mm3, %%mm4		\n\t"\
		"movq %%mm2, %%mm0		\n\t"\
		"movq %%mm5, %%mm6		\n\t"\
		"movq %%mm4, %%mm3		\n\t"\
		"punpcklwd %%mm2, %%mm2		\n\t"\
		"punpcklwd %%mm5, %%mm5		\n\t"\
		"punpcklwd %%mm4, %%mm4		\n\t"\
		"paddw %%mm1, %%mm2		\n\t"\
		"paddw %%mm1, %%mm5		\n\t"\
		"paddw %%mm1, %%mm4		\n\t"\
		"punpckhwd %%mm0, %%mm0		\n\t"\
		"punpckhwd %%mm6, %%mm6		\n\t"\
		"punpckhwd %%mm3, %%mm3		\n\t"\
		"paddw %%mm7, %%mm0		\n\t"\
		"paddw %%mm7, %%mm6		\n\t"\
		"paddw %%mm7, %%mm3		\n\t"\
		/* mm0=B1, mm2=B2, mm3=G2, mm4=G1, mm5=R1, mm6=R2 */\
		"packuswb %%mm0, %%mm2		\n\t"\
		"packuswb %%mm6, %%mm5		\n\t"\
		"packuswb %%mm3, %%mm4		\n\t"\
		"pxor %%mm7, %%mm7		\n\t"

#define YSCALEYUV2RGB1 \
		"xorl %%eax, %%eax		\n\t"\
302
		".balign 16			\n\t"\
303 304 305 306 307 308 309 310 311 312 313 314
		"1:				\n\t"\
		"movq (%2, %%eax), %%mm3	\n\t" /* uvbuf0[eax]*/\
		"movq 4096(%2, %%eax), %%mm4	\n\t" /* uvbuf0[eax+2048]*/\
		"psraw $4, %%mm3		\n\t" /* uvbuf0[eax] - uvbuf1[eax] >>4*/\
		"psraw $4, %%mm4		\n\t" /* uvbuf0[eax+2048] - uvbuf1[eax+2048] >>4*/\
		"psubw w400, %%mm3		\n\t" /* (U-128)8*/\
		"psubw w400, %%mm4		\n\t" /* (V-128)8*/\
		"movq %%mm3, %%mm2		\n\t" /* (U-128)8*/\
		"movq %%mm4, %%mm5		\n\t" /* (V-128)8*/\
		"pmulhw ugCoeff, %%mm3		\n\t"\
		"pmulhw vgCoeff, %%mm4		\n\t"\
	/* mm2=(U-128)8, mm3=ug, mm4=vg mm5=(V-128)8 */\
315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 335 336 337 338 339 340 341 342 343 344 345 346 347 348 349 350
		"movq (%0, %%eax, 2), %%mm1	\n\t" /*buf0[eax]*/\
		"movq 8(%0, %%eax, 2), %%mm7	\n\t" /*buf0[eax]*/\
		"psraw $4, %%mm1		\n\t" /* buf0[eax] - buf1[eax] >>4*/\
		"psraw $4, %%mm7		\n\t" /* buf0[eax] - buf1[eax] >>4*/\
		"pmulhw ubCoeff, %%mm2		\n\t"\
		"pmulhw vrCoeff, %%mm5		\n\t"\
		"psubw w80, %%mm1		\n\t" /* 8(Y-16)*/\
		"psubw w80, %%mm7		\n\t" /* 8(Y-16)*/\
		"pmulhw yCoeff, %%mm1		\n\t"\
		"pmulhw yCoeff, %%mm7		\n\t"\
	/* mm1= Y1, mm2=ub, mm3=ug, mm4=vg mm5=vr, mm7=Y2 */\
		"paddw %%mm3, %%mm4		\n\t"\
		"movq %%mm2, %%mm0		\n\t"\
		"movq %%mm5, %%mm6		\n\t"\
		"movq %%mm4, %%mm3		\n\t"\
		"punpcklwd %%mm2, %%mm2		\n\t"\
		"punpcklwd %%mm5, %%mm5		\n\t"\
		"punpcklwd %%mm4, %%mm4		\n\t"\
		"paddw %%mm1, %%mm2		\n\t"\
		"paddw %%mm1, %%mm5		\n\t"\
		"paddw %%mm1, %%mm4		\n\t"\
		"punpckhwd %%mm0, %%mm0		\n\t"\
		"punpckhwd %%mm6, %%mm6		\n\t"\
		"punpckhwd %%mm3, %%mm3		\n\t"\
		"paddw %%mm7, %%mm0		\n\t"\
		"paddw %%mm7, %%mm6		\n\t"\
		"paddw %%mm7, %%mm3		\n\t"\
		/* mm0=B1, mm2=B2, mm3=G2, mm4=G1, mm5=R1, mm6=R2 */\
		"packuswb %%mm0, %%mm2		\n\t"\
		"packuswb %%mm6, %%mm5		\n\t"\
		"packuswb %%mm3, %%mm4		\n\t"\
		"pxor %%mm7, %%mm7		\n\t"

// do vertical chrominance interpolation
#define YSCALEYUV2RGB1b \
		"xorl %%eax, %%eax		\n\t"\
351
		".balign 16			\n\t"\
352 353 354 355 356
		"1:				\n\t"\
		"movq (%2, %%eax), %%mm2	\n\t" /* uvbuf0[eax]*/\
		"movq (%3, %%eax), %%mm3	\n\t" /* uvbuf1[eax]*/\
		"movq 4096(%2, %%eax), %%mm5	\n\t" /* uvbuf0[eax+2048]*/\
		"movq 4096(%3, %%eax), %%mm4	\n\t" /* uvbuf1[eax+2048]*/\
michael's avatar
michael committed
357 358
		"paddw %%mm2, %%mm3		\n\t" /* uvbuf0[eax] + uvbuf1[eax]*/\
		"paddw %%mm5, %%mm4		\n\t" /* uvbuf0[eax+2048] + uvbuf1[eax+2048]*/\
michael's avatar
michael committed
359 360
		"psrlw $5, %%mm3		\n\t" /*FIXME might overflow*/\
		"psrlw $5, %%mm4		\n\t" /*FIXME might overflow*/\
361 362 363 364 365 366 367 368 369
		"psubw w400, %%mm3		\n\t" /* (U-128)8*/\
		"psubw w400, %%mm4		\n\t" /* (V-128)8*/\
		"movq %%mm3, %%mm2		\n\t" /* (U-128)8*/\
		"movq %%mm4, %%mm5		\n\t" /* (V-128)8*/\
		"pmulhw ugCoeff, %%mm3		\n\t"\
		"pmulhw vgCoeff, %%mm4		\n\t"\
	/* mm2=(U-128)8, mm3=ug, mm4=vg mm5=(V-128)8 */\
		"movq (%0, %%eax, 2), %%mm1	\n\t" /*buf0[eax]*/\
		"movq 8(%0, %%eax, 2), %%mm7	\n\t" /*buf0[eax]*/\
370 371 372 373 374 375 376 377 378 379 380 381 382 383 384 385 386 387 388 389 390 391 392 393 394 395 396 397 398 399 400 401 402 403 404 405 406 407 408 409 410 411 412 413 414 415 416 417 418 419 420 421 422 423 424 425
		"psraw $4, %%mm1		\n\t" /* buf0[eax] - buf1[eax] >>4*/\
		"psraw $4, %%mm7		\n\t" /* buf0[eax] - buf1[eax] >>4*/\
		"pmulhw ubCoeff, %%mm2		\n\t"\
		"pmulhw vrCoeff, %%mm5		\n\t"\
		"psubw w80, %%mm1		\n\t" /* 8(Y-16)*/\
		"psubw w80, %%mm7		\n\t" /* 8(Y-16)*/\
		"pmulhw yCoeff, %%mm1		\n\t"\
		"pmulhw yCoeff, %%mm7		\n\t"\
	/* mm1= Y1, mm2=ub, mm3=ug, mm4=vg mm5=vr, mm7=Y2 */\
		"paddw %%mm3, %%mm4		\n\t"\
		"movq %%mm2, %%mm0		\n\t"\
		"movq %%mm5, %%mm6		\n\t"\
		"movq %%mm4, %%mm3		\n\t"\
		"punpcklwd %%mm2, %%mm2		\n\t"\
		"punpcklwd %%mm5, %%mm5		\n\t"\
		"punpcklwd %%mm4, %%mm4		\n\t"\
		"paddw %%mm1, %%mm2		\n\t"\
		"paddw %%mm1, %%mm5		\n\t"\
		"paddw %%mm1, %%mm4		\n\t"\
		"punpckhwd %%mm0, %%mm0		\n\t"\
		"punpckhwd %%mm6, %%mm6		\n\t"\
		"punpckhwd %%mm3, %%mm3		\n\t"\
		"paddw %%mm7, %%mm0		\n\t"\
		"paddw %%mm7, %%mm6		\n\t"\
		"paddw %%mm7, %%mm3		\n\t"\
		/* mm0=B1, mm2=B2, mm3=G2, mm4=G1, mm5=R1, mm6=R2 */\
		"packuswb %%mm0, %%mm2		\n\t"\
		"packuswb %%mm6, %%mm5		\n\t"\
		"packuswb %%mm3, %%mm4		\n\t"\
		"pxor %%mm7, %%mm7		\n\t"

#define WRITEBGR32 \
		/* mm2=B, %%mm4=G, %%mm5=R, %%mm7=0 */\
			"movq %%mm2, %%mm1		\n\t" /* B */\
			"movq %%mm5, %%mm6		\n\t" /* R */\
			"punpcklbw %%mm4, %%mm2		\n\t" /* GBGBGBGB 0 */\
			"punpcklbw %%mm7, %%mm5		\n\t" /* 0R0R0R0R 0 */\
			"punpckhbw %%mm4, %%mm1		\n\t" /* GBGBGBGB 2 */\
			"punpckhbw %%mm7, %%mm6		\n\t" /* 0R0R0R0R 2 */\
			"movq %%mm2, %%mm0		\n\t" /* GBGBGBGB 0 */\
			"movq %%mm1, %%mm3		\n\t" /* GBGBGBGB 2 */\
			"punpcklwd %%mm5, %%mm0		\n\t" /* 0RGB0RGB 0 */\
			"punpckhwd %%mm5, %%mm2		\n\t" /* 0RGB0RGB 1 */\
			"punpcklwd %%mm6, %%mm1		\n\t" /* 0RGB0RGB 2 */\
			"punpckhwd %%mm6, %%mm3		\n\t" /* 0RGB0RGB 3 */\
\
			MOVNTQ(%%mm0, (%4, %%eax, 4))\
			MOVNTQ(%%mm2, 8(%4, %%eax, 4))\
			MOVNTQ(%%mm1, 16(%4, %%eax, 4))\
			MOVNTQ(%%mm3, 24(%4, %%eax, 4))\
\
			"addl $8, %%eax			\n\t"\
			"cmpl %5, %%eax			\n\t"\
			" jb 1b				\n\t"

#define WRITEBGR16 \
michael's avatar
michael committed
426 427 428 429
			"pand bF8, %%mm2		\n\t" /* B */\
			"pand bFC, %%mm4		\n\t" /* G */\
			"pand bF8, %%mm5		\n\t" /* R */\
			"psrlq $3, %%mm2		\n\t"\
430
\
michael's avatar
michael committed
431 432
			"movq %%mm2, %%mm1		\n\t"\
			"movq %%mm4, %%mm3		\n\t"\
433
\
michael's avatar
michael committed
434 435 436 437
			"punpcklbw %%mm7, %%mm3		\n\t"\
			"punpcklbw %%mm5, %%mm2		\n\t"\
			"punpckhbw %%mm7, %%mm4		\n\t"\
			"punpckhbw %%mm5, %%mm1		\n\t"\
438
\
michael's avatar
michael committed
439 440
			"psllq $3, %%mm3		\n\t"\
			"psllq $3, %%mm4		\n\t"\
441 442 443 444 445 446 447 448 449 450 451 452
\
			"por %%mm3, %%mm2		\n\t"\
			"por %%mm4, %%mm1		\n\t"\
\
			MOVNTQ(%%mm2, (%4, %%eax, 2))\
			MOVNTQ(%%mm1, 8(%4, %%eax, 2))\
\
			"addl $8, %%eax			\n\t"\
			"cmpl %5, %%eax			\n\t"\
			" jb 1b				\n\t"

#define WRITEBGR15 \
michael's avatar
michael committed
453 454 455 456 457
			"pand bF8, %%mm2		\n\t" /* B */\
			"pand bF8, %%mm4		\n\t" /* G */\
			"pand bF8, %%mm5		\n\t" /* R */\
			"psrlq $3, %%mm2		\n\t"\
			"psrlq $1, %%mm5		\n\t"\
458
\
michael's avatar
michael committed
459 460
			"movq %%mm2, %%mm1		\n\t"\
			"movq %%mm4, %%mm3		\n\t"\
461
\
michael's avatar
michael committed
462 463 464 465
			"punpcklbw %%mm7, %%mm3		\n\t"\
			"punpcklbw %%mm5, %%mm2		\n\t"\
			"punpckhbw %%mm7, %%mm4		\n\t"\
			"punpckhbw %%mm5, %%mm1		\n\t"\
466
\
michael's avatar
michael committed
467 468
			"psllq $2, %%mm3		\n\t"\
			"psllq $2, %%mm4		\n\t"\
469 470 471 472 473 474 475 476 477 478
\
			"por %%mm3, %%mm2		\n\t"\
			"por %%mm4, %%mm1		\n\t"\
\
			MOVNTQ(%%mm2, (%4, %%eax, 2))\
			MOVNTQ(%%mm1, 8(%4, %%eax, 2))\
\
			"addl $8, %%eax			\n\t"\
			"cmpl %5, %%eax			\n\t"\
			" jb 1b				\n\t"
michael's avatar
michael committed
479

michael's avatar
michael committed
480
#define WRITEBGR24OLD \
481 482 483 484 485 486 487 488 489
		/* mm2=B, %%mm4=G, %%mm5=R, %%mm7=0 */\
			"movq %%mm2, %%mm1		\n\t" /* B */\
			"movq %%mm5, %%mm6		\n\t" /* R */\
			"punpcklbw %%mm4, %%mm2		\n\t" /* GBGBGBGB 0 */\
			"punpcklbw %%mm7, %%mm5		\n\t" /* 0R0R0R0R 0 */\
			"punpckhbw %%mm4, %%mm1		\n\t" /* GBGBGBGB 2 */\
			"punpckhbw %%mm7, %%mm6		\n\t" /* 0R0R0R0R 2 */\
			"movq %%mm2, %%mm0		\n\t" /* GBGBGBGB 0 */\
			"movq %%mm1, %%mm3		\n\t" /* GBGBGBGB 2 */\
michael's avatar
michael committed
490 491 492 493
			"punpcklwd %%mm5, %%mm0		\n\t" /* 0RGB0RGB 0 */\
			"punpckhwd %%mm5, %%mm2		\n\t" /* 0RGB0RGB 1 */\
			"punpcklwd %%mm6, %%mm1		\n\t" /* 0RGB0RGB 2 */\
			"punpckhwd %%mm6, %%mm3		\n\t" /* 0RGB0RGB 3 */\
494 495 496 497 498 499 500 501 502 503 504 505 506 507 508 509 510 511 512 513 514 515 516 517 518 519 520 521 522 523 524 525 526
\
			"movq %%mm0, %%mm4		\n\t" /* 0RGB0RGB 0 */\
			"psrlq $8, %%mm0		\n\t" /* 00RGB0RG 0 */\
			"pand bm00000111, %%mm4		\n\t" /* 00000RGB 0 */\
			"pand bm11111000, %%mm0		\n\t" /* 00RGB000 0.5 */\
			"por %%mm4, %%mm0		\n\t" /* 00RGBRGB 0 */\
			"movq %%mm2, %%mm4		\n\t" /* 0RGB0RGB 1 */\
			"psllq $48, %%mm2		\n\t" /* GB000000 1 */\
			"por %%mm2, %%mm0		\n\t" /* GBRGBRGB 0 */\
\
			"movq %%mm4, %%mm2		\n\t" /* 0RGB0RGB 1 */\
			"psrld $16, %%mm4		\n\t" /* 000R000R 1 */\
			"psrlq $24, %%mm2		\n\t" /* 0000RGB0 1.5 */\
			"por %%mm4, %%mm2		\n\t" /* 000RRGBR 1 */\
			"pand bm00001111, %%mm2		\n\t" /* 0000RGBR 1 */\
			"movq %%mm1, %%mm4		\n\t" /* 0RGB0RGB 2 */\
			"psrlq $8, %%mm1		\n\t" /* 00RGB0RG 2 */\
			"pand bm00000111, %%mm4		\n\t" /* 00000RGB 2 */\
			"pand bm11111000, %%mm1		\n\t" /* 00RGB000 2.5 */\
			"por %%mm4, %%mm1		\n\t" /* 00RGBRGB 2 */\
			"movq %%mm1, %%mm4		\n\t" /* 00RGBRGB 2 */\
			"psllq $32, %%mm1		\n\t" /* BRGB0000 2 */\
			"por %%mm1, %%mm2		\n\t" /* BRGBRGBR 1 */\
\
			"psrlq $32, %%mm4		\n\t" /* 000000RG 2.5 */\
			"movq %%mm3, %%mm5		\n\t" /* 0RGB0RGB 3 */\
			"psrlq $8, %%mm3		\n\t" /* 00RGB0RG 3 */\
			"pand bm00000111, %%mm5		\n\t" /* 00000RGB 3 */\
			"pand bm11111000, %%mm3		\n\t" /* 00RGB000 3.5 */\
			"por %%mm5, %%mm3		\n\t" /* 00RGBRGB 3 */\
			"psllq $16, %%mm3		\n\t" /* RGBRGB00 3 */\
			"por %%mm4, %%mm3		\n\t" /* RGBRGBRG 2.5 */\
\
michael's avatar
michael committed
527 528 529 530
			MOVNTQ(%%mm0, (%%ebx))\
			MOVNTQ(%%mm2, 8(%%ebx))\
			MOVNTQ(%%mm3, 16(%%ebx))\
			"addl $24, %%ebx		\n\t"\
531 532 533 534 535
\
			"addl $8, %%eax			\n\t"\
			"cmpl %5, %%eax			\n\t"\
			" jb 1b				\n\t"

michael's avatar
michael committed
536 537 538 539 540 541 542 543 544 545 546 547 548 549 550 551 552 553 554 555 556 557 558 559 560 561 562 563 564 565 566 567 568 569 570 571 572 573 574 575 576 577 578 579 580 581 582 583 584 585 586 587 588 589 590 591 592 593 594 595 596 597 598 599 600 601 602 603 604 605 606 607 608 609 610 611 612 613 614 615 616 617 618 619 620 621 622 623 624 625 626 627 628 629 630 631 632 633 634 635 636 637
#define WRITEBGR24MMX \
		/* mm2=B, %%mm4=G, %%mm5=R, %%mm7=0 */\
			"movq %%mm2, %%mm1		\n\t" /* B */\
			"movq %%mm5, %%mm6		\n\t" /* R */\
			"punpcklbw %%mm4, %%mm2		\n\t" /* GBGBGBGB 0 */\
			"punpcklbw %%mm7, %%mm5		\n\t" /* 0R0R0R0R 0 */\
			"punpckhbw %%mm4, %%mm1		\n\t" /* GBGBGBGB 2 */\
			"punpckhbw %%mm7, %%mm6		\n\t" /* 0R0R0R0R 2 */\
			"movq %%mm2, %%mm0		\n\t" /* GBGBGBGB 0 */\
			"movq %%mm1, %%mm3		\n\t" /* GBGBGBGB 2 */\
			"punpcklwd %%mm5, %%mm0		\n\t" /* 0RGB0RGB 0 */\
			"punpckhwd %%mm5, %%mm2		\n\t" /* 0RGB0RGB 1 */\
			"punpcklwd %%mm6, %%mm1		\n\t" /* 0RGB0RGB 2 */\
			"punpckhwd %%mm6, %%mm3		\n\t" /* 0RGB0RGB 3 */\
\
			"movq %%mm0, %%mm4		\n\t" /* 0RGB0RGB 0 */\
			"movq %%mm2, %%mm6		\n\t" /* 0RGB0RGB 1 */\
			"movq %%mm1, %%mm5		\n\t" /* 0RGB0RGB 2 */\
			"movq %%mm3, %%mm7		\n\t" /* 0RGB0RGB 3 */\
\
			"psllq $40, %%mm0		\n\t" /* RGB00000 0 */\
			"psllq $40, %%mm2		\n\t" /* RGB00000 1 */\
			"psllq $40, %%mm1		\n\t" /* RGB00000 2 */\
			"psllq $40, %%mm3		\n\t" /* RGB00000 3 */\
\
			"punpckhdq %%mm4, %%mm0		\n\t" /* 0RGBRGB0 0 */\
			"punpckhdq %%mm6, %%mm2		\n\t" /* 0RGBRGB0 1 */\
			"punpckhdq %%mm5, %%mm1		\n\t" /* 0RGBRGB0 2 */\
			"punpckhdq %%mm7, %%mm3		\n\t" /* 0RGBRGB0 3 */\
\
			"psrlq $8, %%mm0		\n\t" /* 00RGBRGB 0 */\
			"movq %%mm2, %%mm6		\n\t" /* 0RGBRGB0 1 */\
			"psllq $40, %%mm2		\n\t" /* GB000000 1 */\
			"por %%mm2, %%mm0		\n\t" /* GBRGBRGB 0 */\
			MOVNTQ(%%mm0, (%%ebx))\
\
			"psrlq $24, %%mm6		\n\t" /* 0000RGBR 1 */\
			"movq %%mm1, %%mm5		\n\t" /* 0RGBRGB0 2 */\
			"psllq $24, %%mm1		\n\t" /* BRGB0000 2 */\
			"por %%mm1, %%mm6		\n\t" /* BRGBRGBR 1 */\
			MOVNTQ(%%mm6, 8(%%ebx))\
\
			"psrlq $40, %%mm5		\n\t" /* 000000RG 2 */\
			"psllq $8, %%mm3		\n\t" /* RGBRGB00 3 */\
			"por %%mm3, %%mm5		\n\t" /* RGBRGBRG 2 */\
			MOVNTQ(%%mm5, 16(%%ebx))\
\
			"addl $24, %%ebx		\n\t"\
\
			"addl $8, %%eax			\n\t"\
			"cmpl %5, %%eax			\n\t"\
			" jb 1b				\n\t"

#define WRITEBGR24MMX2 \
		/* mm2=B, %%mm4=G, %%mm5=R, %%mm7=0 */\
			"movq M24A, %%mm0		\n\t"\
			"movq M24C, %%mm7		\n\t"\
			"pshufw $0x50, %%mm2, %%mm1	\n\t" /* B3 B2 B3 B2  B1 B0 B1 B0 */\
			"pshufw $0x50, %%mm4, %%mm3	\n\t" /* G3 G2 G3 G2  G1 G0 G1 G0 */\
			"pshufw $0x00, %%mm5, %%mm6	\n\t" /* R1 R0 R1 R0  R1 R0 R1 R0 */\
\
			"pand %%mm0, %%mm1		\n\t" /*    B2        B1       B0 */\
			"pand %%mm0, %%mm3		\n\t" /*    G2        G1       G0 */\
			"pand %%mm7, %%mm6		\n\t" /*       R1        R0       */\
\
			"psllq $8, %%mm3		\n\t" /* G2        G1       G0    */\
			"por %%mm1, %%mm6		\n\t"\
			"por %%mm3, %%mm6		\n\t"\
			MOVNTQ(%%mm6, (%%ebx))\
\
			"psrlq $8, %%mm4		\n\t" /* 00 G7 G6 G5  G4 G3 G2 G1 */\
			"pshufw $0xA5, %%mm2, %%mm1	\n\t" /* B5 B4 B5 B4  B3 B2 B3 B2 */\
			"pshufw $0x55, %%mm4, %%mm3	\n\t" /* G4 G3 G4 G3  G4 G3 G4 G3 */\
			"pshufw $0xA5, %%mm5, %%mm6	\n\t" /* R5 R4 R5 R4  R3 R2 R3 R2 */\
\
			"pand M24B, %%mm1		\n\t" /* B5       B4        B3    */\
			"pand %%mm7, %%mm3		\n\t" /*       G4        G3       */\
			"pand %%mm0, %%mm6		\n\t" /*    R4        R3       R2 */\
\
			"por %%mm1, %%mm3		\n\t" /* B5    G4 B4     G3 B3    */\
			"por %%mm3, %%mm6		\n\t"\
			MOVNTQ(%%mm6, 8(%%ebx))\
\
			"pshufw $0xFF, %%mm2, %%mm1	\n\t" /* B7 B6 B7 B6  B7 B6 B6 B7 */\
			"pshufw $0xFA, %%mm4, %%mm3	\n\t" /* 00 G7 00 G7  G6 G5 G6 G5 */\
			"pshufw $0xFA, %%mm5, %%mm6	\n\t" /* R7 R6 R7 R6  R5 R4 R5 R4 */\
\
			"pand %%mm7, %%mm1		\n\t" /*       B7        B6       */\
			"pand %%mm0, %%mm3		\n\t" /*    G7        G6       G5 */\
			"pand M24B, %%mm6		\n\t" /* R7       R6        R5    */\
\
			"por %%mm1, %%mm3		\n\t"\
			"por %%mm3, %%mm6		\n\t"\
			MOVNTQ(%%mm6, 16(%%ebx))\
\
			"addl $24, %%ebx		\n\t"\
\
			"addl $8, %%eax			\n\t"\
			"cmpl %5, %%eax			\n\t"\
			" jb 1b				\n\t"

#ifdef HAVE_MMX2
michael's avatar
michael committed
638
#undef WRITEBGR24
michael's avatar
michael committed
639 640
#define WRITEBGR24 WRITEBGR24MMX2
#else
michael's avatar
michael committed
641
#undef WRITEBGR24
michael's avatar
michael committed
642 643 644
#define WRITEBGR24 WRITEBGR24MMX
#endif

michael's avatar
michael committed
645 646 647 648
static inline void RENAME(yuv2yuvX)(int16_t *lumFilter, int16_t **lumSrc, int lumFilterSize,
				    int16_t *chrFilter, int16_t **chrSrc, int chrFilterSize,
				    uint8_t *dest, uint8_t *uDest, uint8_t *vDest, int dstW,
				    int16_t * lumMmxFilter, int16_t * chrMmxFilter)
michael's avatar
michael committed
649
{
michael's avatar
michael committed
650 651 652 653 654 655 656 657 658 659 660 661 662 663 664 665 666 667 668 669 670 671 672 673 674
#ifdef HAVE_MMX
	if(uDest != NULL)
	{
		asm volatile(
				YSCALEYUV2YV12X(0)
				:: "m" (-chrFilterSize), "r" (chrSrc+chrFilterSize),
				"r" (chrMmxFilter+chrFilterSize*4), "r" (uDest), "m" (dstW>>1)
				: "%eax", "%edx", "%esi"
			);

		asm volatile(
				YSCALEYUV2YV12X(4096)
				:: "m" (-chrFilterSize), "r" (chrSrc+chrFilterSize),
				"r" (chrMmxFilter+chrFilterSize*4), "r" (vDest), "m" (dstW>>1)
				: "%eax", "%edx", "%esi"
			);
	}

	asm volatile(
			YSCALEYUV2YV12X(0)
			:: "m" (-lumFilterSize), "r" (lumSrc+lumFilterSize),
			   "r" (lumMmxFilter+lumFilterSize*4), "r" (dest), "m" (dstW)
			: "%eax", "%edx", "%esi"
		);
#else
michael's avatar
michael committed
675 676 677
yuv2yuvXinC(lumFilter, lumSrc, lumFilterSize,
	    chrFilter, chrSrc, chrFilterSize,
	    dest, uDest, vDest, dstW);
michael's avatar
michael committed
678
#endif
michael's avatar
michael committed
679
}
680

michael's avatar
michael committed
681 682 683 684 685
static inline void RENAME(yuv2yuv1)(int16_t *lumSrc, int16_t *chrSrc,
				    uint8_t *dest, uint8_t *uDest, uint8_t *vDest, int dstW)
{
#ifdef HAVE_MMX
	if(uDest != NULL)
michael's avatar
michael committed
686
	{
michael's avatar
michael committed
687 688 689 690 691 692 693 694 695 696 697 698 699
		asm volatile(
				YSCALEYUV2YV121
				:: "r" (chrSrc + (dstW>>1)), "r" (uDest + (dstW>>1)),
				"g" (-(dstW>>1))
				: "%eax"
			);

		asm volatile(
				YSCALEYUV2YV121
				:: "r" (chrSrc + 2048 + (dstW>>1)), "r" (vDest + (dstW>>1)),
				"g" (-(dstW>>1))
				: "%eax"
			);
michael's avatar
michael committed
700 701
	}

michael's avatar
michael committed
702 703 704 705 706 707 708 709 710 711 712
	asm volatile(
		YSCALEYUV2YV121
		:: "r" (lumSrc + dstW), "r" (dest + dstW),
		"g" (-dstW)
		: "%eax"
	);
#else
	//FIXME Optimize (just quickly writen not opti..)
	//FIXME replace MINMAX with LUTs
	int i;
	for(i=0; i<dstW; i++)
michael's avatar
michael committed
713
	{
michael's avatar
michael committed
714 715 716 717 718 719
		int val= lumSrc[i]>>7;

		dest[i]= MIN(MAX(val>>19, 0), 255);
	}

	if(uDest != NULL)
michael's avatar
michael committed
720
		for(i=0; i<(dstW>>1); i++)
michael's avatar
michael committed
721
		{
michael's avatar
michael committed
722 723 724 725 726
			int u=chrSrc[i]>>7;
			int v=chrSrc[i + 2048]>>7;

			uDest[i]= MIN(MAX(u>>19, 0), 255);
			vDest[i]= MIN(MAX(v>>19, 0), 255);
michael's avatar
michael committed
727
		}
michael's avatar
michael committed
728
#endif
michael's avatar
michael committed
729 730
}

michael's avatar
michael committed
731

732 733 734
/**
 * vertical scale YV12 to RGB
 */
michael's avatar
michael committed
735 736 737 738 739 740 741 742 743 744 745 746 747 748 749 750 751 752 753 754 755 756 757 758 759 760 761 762 763 764 765 766 767 768 769 770 771 772 773 774 775 776 777 778 779 780 781 782 783 784 785 786 787 788 789 790 791 792 793 794 795 796 797 798 799 800 801 802 803 804 805 806 807 808 809 810 811 812 813 814
static inline void RENAME(yuv2rgbX)(int16_t *lumFilter, int16_t **lumSrc, int lumFilterSize,
				    int16_t *chrFilter, int16_t **chrSrc, int chrFilterSize,
			    uint8_t *dest, int dstW, int dstbpp, int16_t * lumMmxFilter, int16_t * chrMmxFilter)
{
	if(fullUVIpol)
	{
//FIXME
	}//FULL_UV_IPOL
	else
	{
#ifdef HAVE_MMX
		if(dstbpp == 32) //FIXME untested
		{
			asm volatile(
				YSCALEYUV2RGBX
				WRITEBGR32

			:: "m" (-lumFilterSize), "m" (-chrFilterSize),
			   "m" (lumMmxFilter+lumFilterSize*4), "m" (chrMmxFilter+chrFilterSize*4),
			   "r" (dest), "m" (dstW),
			   "m" (lumSrc+lumFilterSize), "m" (chrSrc+chrFilterSize)
			: "%eax", "%ebx", "%ecx", "%edx", "%esi"
			);
		}
		else if(dstbpp==24) //FIXME untested
		{
			asm volatile(
				YSCALEYUV2RGBX
				"leal (%%eax, %%eax, 2), %%ebx	\n\t" //FIXME optimize
				"addl %4, %%ebx			\n\t"
				WRITEBGR24

			:: "m" (-lumFilterSize), "m" (-chrFilterSize),
			   "m" (lumMmxFilter+lumFilterSize*4), "m" (chrMmxFilter+chrFilterSize*4),
			   "r" (dest), "m" (dstW),
			   "m" (lumSrc+lumFilterSize), "m" (chrSrc+chrFilterSize)
			: "%eax", "%ebx", "%ecx", "%edx", "%esi"
			);
		}
		else if(dstbpp==15)
		{
			asm volatile(
				YSCALEYUV2RGBX
		/* mm2=B, %%mm4=G, %%mm5=R, %%mm7=0 */
#ifdef DITHER1XBPP
				"paddusb b5Dither, %%mm2	\n\t"
				"paddusb g5Dither, %%mm4	\n\t"
				"paddusb r5Dither, %%mm5	\n\t"
#endif

				WRITEBGR15

			:: "m" (-lumFilterSize), "m" (-chrFilterSize),
			   "m" (lumMmxFilter+lumFilterSize*4), "m" (chrMmxFilter+chrFilterSize*4),
			   "r" (dest), "m" (dstW),
			   "m" (lumSrc+lumFilterSize), "m" (chrSrc+chrFilterSize)
			: "%eax", "%ebx", "%ecx", "%edx", "%esi"
			);
		}
		else if(dstbpp==16)
		{
			asm volatile(
				YSCALEYUV2RGBX
		/* mm2=B, %%mm4=G, %%mm5=R, %%mm7=0 */
#ifdef DITHER1XBPP
				"paddusb b5Dither, %%mm2	\n\t"
				"paddusb g6Dither, %%mm4	\n\t"
				"paddusb r5Dither, %%mm5	\n\t"
#endif

				WRITEBGR16

			:: "m" (-lumFilterSize), "m" (-chrFilterSize),
			   "m" (lumMmxFilter+lumFilterSize*4), "m" (chrMmxFilter+chrFilterSize*4),
			   "r" (dest), "m" (dstW),
			   "m" (lumSrc+lumFilterSize), "m" (chrSrc+chrFilterSize)
			: "%eax", "%ebx", "%ecx", "%edx", "%esi"
			);
		}
#else
michael's avatar
michael committed
815 816 817
yuv2rgbXinC(lumFilter, lumSrc, lumFilterSize,
	    chrFilter, chrSrc, chrFilterSize,
	    dest, dstW, dstbpp);
michael's avatar
michael committed
818 819 820 821 822 823 824 825 826 827

#endif
	} //!FULL_UV_IPOL
}


/**
 * vertical bilinear scale YV12 to RGB
 */
static inline void RENAME(yuv2rgb2)(uint16_t *buf0, uint16_t *buf1, uint16_t *uvbuf0, uint16_t *uvbuf1,
michael's avatar
michael committed
828
			    uint8_t *dest, int dstW, int yalpha, int uvalpha, int dstbpp)
829 830 831 832 833 834 835 836 837 838 839 840 841 842 843 844 845 846 847 848 849 850 851 852 853 854 855 856 857
{
	int yalpha1=yalpha^4095;
	int uvalpha1=uvalpha^4095;

	if(fullUVIpol)
	{

#ifdef HAVE_MMX
		if(dstbpp == 32)
		{
			asm volatile(


FULL_YSCALEYUV2RGB
			"punpcklbw %%mm1, %%mm3		\n\t" // BGBGBGBG
			"punpcklbw %%mm7, %%mm0		\n\t" // R0R0R0R0

			"movq %%mm3, %%mm1		\n\t"
			"punpcklwd %%mm0, %%mm3		\n\t" // BGR0BGR0
			"punpckhwd %%mm0, %%mm1		\n\t" // BGR0BGR0

			MOVNTQ(%%mm3, (%4, %%eax, 4))
			MOVNTQ(%%mm1, 8(%4, %%eax, 4))

			"addl $4, %%eax			\n\t"
			"cmpl %5, %%eax			\n\t"
			" jb 1b				\n\t"


michael's avatar
michael committed
858
			:: "r" (buf0), "r" (buf1), "r" (uvbuf0), "r" (uvbuf1), "r" (dest), "m" (dstW),
859 860 861 862 863 864 865 866 867 868 869 870 871 872 873 874 875 876 877 878 879 880 881 882 883 884 885 886 887 888 889 890 891 892 893 894 895 896 897 898 899 900 901 902 903 904 905 906 907
			"m" (yalpha1), "m" (uvalpha1)
			: "%eax"
			);
		}
		else if(dstbpp==24)
		{
			asm volatile(

FULL_YSCALEYUV2RGB

								// lsb ... msb
			"punpcklbw %%mm1, %%mm3		\n\t" // BGBGBGBG
			"punpcklbw %%mm7, %%mm0		\n\t" // R0R0R0R0

			"movq %%mm3, %%mm1		\n\t"
			"punpcklwd %%mm0, %%mm3		\n\t" // BGR0BGR0
			"punpckhwd %%mm0, %%mm1		\n\t" // BGR0BGR0

			"movq %%mm3, %%mm2		\n\t" // BGR0BGR0
			"psrlq $8, %%mm3		\n\t" // GR0BGR00
			"pand bm00000111, %%mm2		\n\t" // BGR00000
			"pand bm11111000, %%mm3		\n\t" // 000BGR00
			"por %%mm2, %%mm3		\n\t" // BGRBGR00
			"movq %%mm1, %%mm2		\n\t"
			"psllq $48, %%mm1		\n\t" // 000000BG
			"por %%mm1, %%mm3		\n\t" // BGRBGRBG

			"movq %%mm2, %%mm1		\n\t" // BGR0BGR0
			"psrld $16, %%mm2		\n\t" // R000R000
			"psrlq $24, %%mm1		\n\t" // 0BGR0000
			"por %%mm2, %%mm1		\n\t" // RBGRR000

			"movl %4, %%ebx			\n\t"
			"addl %%eax, %%ebx		\n\t"

#ifdef HAVE_MMX2
			//FIXME Alignment
			"movntq %%mm3, (%%ebx, %%eax, 2)\n\t"
			"movntq %%mm1, 8(%%ebx, %%eax, 2)\n\t"
#else
			"movd %%mm3, (%%ebx, %%eax, 2)	\n\t"
			"psrlq $32, %%mm3		\n\t"
			"movd %%mm3, 4(%%ebx, %%eax, 2)	\n\t"
			"movd %%mm1, 8(%%ebx, %%eax, 2)	\n\t"
#endif
			"addl $4, %%eax			\n\t"
			"cmpl %5, %%eax			\n\t"
			" jb 1b				\n\t"

michael's avatar
michael committed
908
			:: "r" (buf0), "r" (buf1), "r" (uvbuf0), "r" (uvbuf1), "m" (dest), "m" (dstW),
909 910 911 912 913 914 915 916 917 918
			"m" (yalpha1), "m" (uvalpha1)
			: "%eax", "%ebx"
			);
		}
		else if(dstbpp==15)
		{
			asm volatile(

FULL_YSCALEYUV2RGB
#ifdef DITHER1XBPP
michael's avatar
michael committed
919 920 921
			"paddusb g5Dither, %%mm1	\n\t"
			"paddusb r5Dither, %%mm0	\n\t"
			"paddusb b5Dither, %%mm3	\n\t"
922 923 924 925 926 927 928 929 930 931 932 933 934 935 936 937 938 939 940 941
#endif
			"punpcklbw %%mm7, %%mm1		\n\t" // 0G0G0G0G
			"punpcklbw %%mm7, %%mm3		\n\t" // 0B0B0B0B
			"punpcklbw %%mm7, %%mm0		\n\t" // 0R0R0R0R

			"psrlw $3, %%mm3		\n\t"
			"psllw $2, %%mm1		\n\t"
			"psllw $7, %%mm0		\n\t"
			"pand g15Mask, %%mm1		\n\t"
			"pand r15Mask, %%mm0		\n\t"

			"por %%mm3, %%mm1		\n\t"
			"por %%mm1, %%mm0		\n\t"

			MOVNTQ(%%mm0, (%4, %%eax, 2))

			"addl $4, %%eax			\n\t"
			"cmpl %5, %%eax			\n\t"
			" jb 1b				\n\t"

michael's avatar
michael committed
942
			:: "r" (buf0), "r" (buf1), "r" (uvbuf0), "r" (uvbuf1), "r" (dest), "m" (dstW),
943 944 945 946 947 948 949 950 951 952
			"m" (yalpha1), "m" (uvalpha1)
			: "%eax"
			);
		}
		else if(dstbpp==16)
		{
			asm volatile(

FULL_YSCALEYUV2RGB
#ifdef DITHER1XBPP
michael's avatar
michael committed
953 954 955
			"paddusb g6Dither, %%mm1	\n\t"
			"paddusb r5Dither, %%mm0	\n\t"
			"paddusb b5Dither, %%mm3	\n\t"
956 957 958 959 960 961 962 963 964 965 966 967 968 969 970 971 972 973 974 975
#endif
			"punpcklbw %%mm7, %%mm1		\n\t" // 0G0G0G0G
			"punpcklbw %%mm7, %%mm3		\n\t" // 0B0B0B0B
			"punpcklbw %%mm7, %%mm0		\n\t" // 0R0R0R0R

			"psrlw $3, %%mm3		\n\t"
			"psllw $3, %%mm1		\n\t"
			"psllw $8, %%mm0		\n\t"
			"pand g16Mask, %%mm1		\n\t"
			"pand r16Mask, %%mm0		\n\t"

			"por %%mm3, %%mm1		\n\t"
			"por %%mm1, %%mm0		\n\t"

			MOVNTQ(%%mm0, (%4, %%eax, 2))

			"addl $4, %%eax			\n\t"
			"cmpl %5, %%eax			\n\t"
			" jb 1b				\n\t"

michael's avatar
michael committed
976
			:: "r" (buf0), "r" (buf1), "r" (uvbuf0), "r" (uvbuf1), "r" (dest), "m" (dstW),
977 978 979 980 981 982 983
			"m" (yalpha1), "m" (uvalpha1)
			: "%eax"
			);
		}
#else
		if(dstbpp==32 || dstbpp==24)
		{
michael's avatar
michael committed
984
			int i;
michael's avatar
michael committed
985
			for(i=0;i<dstW;i++){
986 987 988 989
				// vertical linear interpolation && yuv2rgb in a single step:
				int Y=yuvtab_2568[((buf0[i]*yalpha1+buf1[i]*yalpha)>>19)];
				int U=((uvbuf0[i]*uvalpha1+uvbuf1[i]*uvalpha)>>19);
				int V=((uvbuf0[i+2048]*uvalpha1+uvbuf1[i+2048]*uvalpha)>>19);
michael's avatar
michael committed
990 991 992
				dest[0]=clip_table[((Y + yuvtab_40cf[U]) >>13)];
				dest[1]=clip_table[((Y + yuvtab_1a1e[V] + yuvtab_0c92[U]) >>13)];
				dest[2]=clip_table[((Y + yuvtab_3343[V]) >>13)];
993 994 995 996 997
				dest+=dstbpp>>3;
			}
		}
		else if(dstbpp==16)
		{
michael's avatar
michael committed
998
			int i;
michael's avatar
michael committed
999
			for(i=0;i<dstW;i++){
1000 1001 1002 1003 1004
				// vertical linear interpolation && yuv2rgb in a single step:
				int Y=yuvtab_2568[((buf0[i]*yalpha1+buf1[i]*yalpha)>>19)];
				int U=((uvbuf0[i]*uvalpha1+uvbuf1[i]*uvalpha)>>19);
				int V=((uvbuf0[i+2048]*uvalpha1+uvbuf1[i+2048]*uvalpha)>>19);

1005
				((uint16_t*)dest)[i] =
michael's avatar
michael committed
1006 1007 1008
					clip_table16b[(Y + yuvtab_40cf[U]) >>13] |
					clip_table16g[(Y + yuvtab_1a1e[V] + yuvtab_0c92[U]) >>13] |
					clip_table16r[(Y + yuvtab_3343[V]) >>13];
1009 1010 1011 1012
			}
		}
		else if(dstbpp==15)
		{
michael's avatar
michael committed
1013
			int i;
michael's avatar
michael committed
1014
			for(i=0;i<dstW;i++){
1015 1016 1017 1018 1019
				// vertical linear interpolation && yuv2rgb in a single step:
				int Y=yuvtab_2568[((buf0[i]*yalpha1+buf1[i]*yalpha)>>19)];
				int U=((uvbuf0[i]*uvalpha1+uvbuf1[i]*uvalpha)>>19);
				int V=((uvbuf0[i+2048]*uvalpha1+uvbuf1[i+2048]*uvalpha)>>19);

1020
				((uint16_t*)dest)[i] =
michael's avatar
michael committed
1021 1022 1023
					clip_table15b[(Y + yuvtab_40cf[U]) >>13] |
					clip_table15g[(Y + yuvtab_1a1e[V] + yuvtab_0c92[U]) >>13] |
					clip_table15r[(Y + yuvtab_3343[V]) >>13];
1024 1025 1026 1027 1028 1029 1030 1031 1032 1033 1034 1035 1036
			}
		}
#endif
	}//FULL_UV_IPOL
	else
	{
#ifdef HAVE_MMX
		if(dstbpp == 32)
		{
			asm volatile(
				YSCALEYUV2RGB
				WRITEBGR32

michael's avatar
michael committed
1037
			:: "r" (buf0), "r" (buf1), "r" (uvbuf0), "r" (uvbuf1), "r" (dest), "m" (dstW),
1038 1039 1040 1041 1042 1043 1044
			"m" (yalpha1), "m" (uvalpha1)
			: "%eax"
			);
		}
		else if(dstbpp==24)
		{
			asm volatile(
michael's avatar
michael committed
1045
				"movl %4, %%ebx			\n\t"
1046 1047 1048
				YSCALEYUV2RGB
				WRITEBGR24

michael's avatar
michael committed
1049
			:: "r" (buf0), "r" (buf1), "r" (uvbuf0), "r" (uvbuf1), "m" (dest), "m" (dstW),
1050 1051 1052 1053 1054 1055 1056 1057 1058 1059
			"m" (yalpha1), "m" (uvalpha1)
			: "%eax", "%ebx"
			);
		}
		else if(dstbpp==15)
		{
			asm volatile(
				YSCALEYUV2RGB
		/* mm2=B, %%mm4=G, %%mm5=R, %%mm7=0 */
#ifdef DITHER1XBPP
michael's avatar
michael committed
1060 1061 1062
				"paddusb b5Dither, %%mm2	\n\t"
				"paddusb g5Dither, %%mm4	\n\t"
				"paddusb r5Dither, %%mm5	\n\t"
1063 1064 1065 1066
#endif

				WRITEBGR15

michael's avatar
michael committed
1067
			:: "r" (buf0), "r" (buf1), "r" (uvbuf0), "r" (uvbuf1), "r" (dest), "m" (dstW),
1068 1069 1070 1071 1072 1073 1074 1075 1076 1077
			"m" (yalpha1), "m" (uvalpha1)
			: "%eax"
			);
		}
		else if(dstbpp==16)
		{
			asm volatile(
				YSCALEYUV2RGB
		/* mm2=B, %%mm4=G, %%mm5=R, %%mm7=0 */
#ifdef DITHER1XBPP
michael's avatar
michael committed
1078 1079 1080
				"paddusb b5Dither, %%mm2	\n\t"
				"paddusb g6Dither, %%mm4	\n\t"
				"paddusb r5Dither, %%mm5	\n\t"
1081 1082 1083 1084
#endif

				WRITEBGR16

michael's avatar
michael committed
1085
			:: "r" (buf0), "r" (buf1), "r" (uvbuf0), "r" (uvbuf1), "r" (dest), "m" (dstW),
1086 1087 1088 1089 1090
			"m" (yalpha1), "m" (uvalpha1)
			: "%eax"
			);
		}
#else
michael's avatar
michael committed
1091
		if(dstbpp==32)
1092
		{
michael's avatar
michael committed
1093
			int i;
michael's avatar
michael committed
1094
			for(i=0; i<dstW-1; i+=2){
1095
				// vertical linear interpolation && yuv2rgb in a single step:
michael's avatar
michael committed
1096 1097
				int Y1=yuvtab_2568[((buf0[i]*yalpha1+buf1[i]*yalpha)>>19)];
				int Y2=yuvtab_2568[((buf0[i+1]*yalpha1+buf1[i+1]*yalpha)>>19)];
michael's avatar
michael committed
1098 1099
				int U=((uvbuf0[i>>1]*uvalpha1+uvbuf1[i>>1]*uvalpha)>>19);
				int V=((uvbuf0[(i>>1)+2048]*uvalpha1+uvbuf1[(i>>1)+2048]*uvalpha)>>19);
michael's avatar
michael committed
1100 1101 1102 1103 1104 1105 1106 1107 1108 1109 1110 1111 1112 1113

				int Cb= yuvtab_40cf[U];
				int Cg= yuvtab_1a1e[V] + yuvtab_0c92[U];
				int Cr= yuvtab_3343[V];

				dest[4*i+0]=clip_table[((Y1 + Cb) >>13)];
				dest[4*i+1]=clip_table[((Y1 + Cg) >>13)];
				dest[4*i+2]=clip_table[((Y1 + Cr) >>13)];

				dest[4*i+4]=clip_table[((Y2 + Cb) >>13)];
				dest[4*i+5]=clip_table[((Y2 + Cg) >>13)];
				dest[4*i+6]=clip_table[((Y2 + Cr) >>13)];
			}
		}
michael's avatar
michael committed
1114
		else if(dstbpp==24)
michael's avatar
michael committed
1115
		{
michael's avatar
michael committed
1116
			int i;
michael's avatar
michael committed
1117
			for(i=0; i<dstW-1; i+=2){
michael's avatar
michael committed
1118 1119 1120
				// vertical linear interpolation && yuv2rgb in a single step:
				int Y1=yuvtab_2568[((buf0[i]*yalpha1+buf1[i]*yalpha)>>19)];
				int Y2=yuvtab_2568[((buf0[i+1]*yalpha1+buf1[i+1]*yalpha)>>19)];
michael's avatar
michael committed
1121 1122
				int U=((uvbuf0[i>>1]*uvalpha1+uvbuf1[i>>1]*uvalpha)>>19);
				int V=((uvbuf0[(i>>1)+2048]*uvalpha1+uvbuf1[(i>>1)+2048]*uvalpha)>>19);
michael's avatar
michael committed
1123 1124 1125 1126 1127 1128 1129 1130 1131 1132 1133 1134 1135

				int Cb= yuvtab_40cf[U];
				int Cg= yuvtab_1a1e[V] + yuvtab_0c92[U];
				int Cr= yuvtab_3343[V];

				dest[0]=clip_table[((Y1 + Cb) >>13)];
				dest[1]=clip_table[((Y1 + Cg) >>13)];
				dest[2]=clip_table[((Y1 + Cr) >>13)];

				dest[3]=clip_table[((Y2 + Cb) >>13)];
				dest[4]=clip_table[((Y2 + Cg) >>13)];
				dest[5]=clip_table[((Y2 + Cr) >>13)];
				dest+=6;
1136 1137 1138 1139
			}
		}
		else if(dstbpp==16)
		{
michael's avatar
michael committed
1140
			int i;
michael's avatar
michael committed
1141
			for(i=0; i<dstW-1; i+=2){
1142
				// vertical linear interpolation && yuv2rgb in a single step:
michael's avatar
michael committed
1143 1144
				int Y1=yuvtab_2568[((buf0[i]*yalpha1+buf1[i]*yalpha)>>19)];
				int Y2=yuvtab_2568[((buf0[i+1]*yalpha1+buf1[i+1]*yalpha)>>19)];
michael's avatar
michael committed
1145 1146
				int U=((uvbuf0[i>>1]*uvalpha1+uvbuf1[i>>1]*uvalpha)>>19);
				int V=((uvbuf0[(i>>1)+2048]*uvalpha1+uvbuf1[(i>>1)+2048]*uvalpha)>>19);
1147

michael's avatar
michael committed
1148 1149 1150 1151
				int Cb= yuvtab_40cf[U];
				int Cg= yuvtab_1a1e[V] + yuvtab_0c92[U];
				int Cr= yuvtab_3343[V];

1152
				((uint16_t*)dest)[i] =
michael's avatar
michael committed
1153 1154 1155
					clip_table16b[(Y1 + Cb) >>13] |
					clip_table16g[(Y1 + Cg) >>13] |
					clip_table16r[(Y1 + Cr) >>13];
michael's avatar
michael committed
1156 1157

				((uint16_t*)dest)[i+1] =
michael's avatar
michael committed
1158 1159 1160
					clip_table16b[(Y2 + Cb) >>13] |
					clip_table16g[(Y2 + Cg) >>13] |
					clip_table16r[(Y2 + Cr) >>13];
1161 1162 1163 1164
			}
		}
		else if(dstbpp==15)
		{
michael's avatar
michael committed
1165
			int i;
michael's avatar
michael committed
1166
			for(i=0; i<dstW-1; i+=2){
1167
				// vertical linear interpolation && yuv2rgb in a single step:
michael's avatar
michael committed
1168 1169
				int Y1=yuvtab_2568[((buf0[i]*yalpha1+buf1[i]*yalpha)>>19)];
				int Y2=yuvtab_2568[((buf0[i+1]*yalpha1+buf1[i+1]*yalpha)>>19)];
michael's avatar
michael committed
1170 1171
				int U=((uvbuf0[i>>1]*uvalpha1+uvbuf1[i>>1]*uvalpha)>>19);
				int V=((uvbuf0[(i>>1)+2048]*uvalpha1+uvbuf1[(i>>1)+2048]*uvalpha)>>19);
1172

michael's avatar
michael committed
1173 1174 1175 1176
				int Cb= yuvtab_40cf[U];
				int Cg= yuvtab_1a1e[V] + yuvtab_0c92[U];
				int Cr= yuvtab_3343[V];

1177
				((uint16_t*)dest)[i] =
michael's avatar
michael committed
1178 1179 1180 1181
					clip_table15b[(Y1 + Cb) >>13] |
					clip_table15g[(Y1 + Cg) >>13] |
					clip_table15r[(Y1 + Cr) >>13];

michael's avatar
michael committed
1182
				((uint16_t*)dest)[i+1] =
michael's avatar
michael committed
1183 1184 1185
					clip_table15b[(Y2 + Cb) >>13] |
					clip_table15g[(Y2 + Cg) >>13] |
					clip_table15r[(Y2 + Cr) >>13];
1186 1187 1188 1189 1190 1191 1192 1193 1194
			}
		}
#endif
	} //!FULL_UV_IPOL
}

/**
 * YV12 to RGB without scaling or interpolating
 */
michael's avatar
michael committed
1195 1196
static inline void RENAME(yuv2rgb1)(uint16_t *buf0, uint16_t *uvbuf0, uint16_t *uvbuf1,
			    uint8_t *dest, int dstW, int uvalpha, int dstbpp)
1197 1198
{
	int uvalpha1=uvalpha^4095;
michael's avatar
michael committed
1199
	const int yalpha1=0;
michael's avatar
michael committed
1200

1201 1202
	if(fullUVIpol || allwaysIpol)
	{
michael's avatar
michael committed
1203
		RENAME(yuv2rgb2)(buf0, buf0, uvbuf0, uvbuf1, dest, dstW, 0, uvalpha, dstbpp);
1204 1205
		return;
	}
michael's avatar
michael committed
1206 1207

#ifdef HAVE_MMX
1208 1209
	if( uvalpha < 2048 ) // note this is not correct (shifts chrominance by 0.5 pixels) but its a bit faster
	{
1210 1211 1212 1213 1214
		if(dstbpp == 32)
		{
			asm volatile(
				YSCALEYUV2RGB1
				WRITEBGR32
michael's avatar
michael committed
1215
			:: "r" (buf0), "r" (buf0), "r" (uvbuf0), "r" (uvbuf1), "r" (dest), "m" (dstW),
1216 1217 1218 1219 1220 1221 1222
			"m" (yalpha1), "m" (uvalpha1)
			: "%eax"
			);
		}
		else if(dstbpp==24)
		{
			asm volatile(
michael's avatar
michael committed
1223
				"movl %4, %%ebx			\n\t"
1224 1225
				YSCALEYUV2RGB1
				WRITEBGR24
michael's avatar
michael committed
1226
			:: "r" (buf0), "r" (buf0), "r" (uvbuf0), "r" (uvbuf1), "m" (dest), "m" (dstW),
1227 1228 1229 1230 1231 1232 1233 1234 1235 1236
			"m" (yalpha1), "m" (uvalpha1)
			: "%eax", "%ebx"
			);
		}
		else if(dstbpp==15)
		{
			asm volatile(
				YSCALEYUV2RGB1
		/* mm2=B, %%mm4=G, %%mm5=R, %%mm7=0 */
#ifdef DITHER1XBPP
michael's avatar
michael committed
1237 1238 1239
				"paddusb b5Dither, %%mm2	\n\t"
				"paddusb g5Dither, %%mm4	\n\t"
				"paddusb r5Dither, %%mm5	\n\t"
1240 1241
#endif
				WRITEBGR15
michael's avatar
michael committed
1242
			:: "r" (buf0), "r" (buf0), "r" (uvbuf0), "r" (uvbuf1), "r" (dest), "m" (dstW),
1243 1244 1245 1246 1247 1248 1249 1250 1251 1252
			"m" (yalpha1), "m" (uvalpha1)
			: "%eax"
			);
		}
		else if(dstbpp==16)
		{
			asm volatile(
				YSCALEYUV2RGB1
		/* mm2=B, %%mm4=G, %%mm5=R, %%mm7=0 */
#ifdef DITHER1XBPP
michael's avatar
michael committed
1253 1254 1255
				"paddusb b5Dither, %%mm2	\n\t"
				"paddusb g6Dither, %%mm4	\n\t"
				"paddusb r5Dither, %%mm5	\n\t"
1256 1257 1258
#endif

				WRITEBGR16
michael's avatar
michael committed
1259
			:: "r" (buf0), "r" (buf0), "r" (uvbuf0), "r" (uvbuf1), "r" (dest), "m" (dstW),
1260 1261 1262 1263
			"m" (yalpha1), "m" (uvalpha1)
			: "%eax"
			);
		}
1264 1265 1266 1267
	}
	else
	{
		if(dstbpp == 32)
1268
		{
1269 1270 1271
			asm volatile(
				YSCALEYUV2RGB1b
				WRITEBGR32
michael's avatar
michael committed
1272
			:: "r" (buf0), "r" (buf0), "r" (uvbuf0), "r" (uvbuf1), "r" (dest), "m" (dstW),
1273 1274 1275
			"m" (yalpha1), "m" (uvalpha1)
			: "%eax"
			);
1276
		}
1277
		else if(dstbpp==24)
1278
		{
1279
			asm volatile(
michael's avatar
michael committed
1280
				"movl %4, %%ebx			\n\t"
1281 1282
				YSCALEYUV2RGB1b
				WRITEBGR24
michael's avatar
michael committed
1283
			:: "r" (buf0), "r" (buf0), "r" (uvbuf0), "r" (uvbuf1), "m" (dest), "m" (dstW),
1284 1285 1286
			"m" (yalpha1), "m" (uvalpha1)
			: "%eax", "%ebx"
			);
1287 1288 1289
		}
		else if(dstbpp==15)
		{
1290 1291 1292 1293
			asm volatile(
				YSCALEYUV2RGB1b
		/* mm2=B, %%mm4=G, %%mm5=R, %%mm7=0 */
#ifdef DITHER1XBPP
michael's avatar
michael committed
1294 1295 1296
				"paddusb b5Dither, %%mm2	\n\t"
				"paddusb g5Dither, %%mm4	\n\t"
				"paddusb r5Dither, %%mm5	\n\t"
1297 1298
#endif
				WRITEBGR15
michael's avatar
michael committed
1299
			:: "r" (buf0), "r" (buf0), "r" (uvbuf0), "r" (uvbuf1), "r" (dest), "m" (dstW),
1300 1301 1302 1303 1304 1305 1306 1307 1308 1309
			"m" (yalpha1), "m" (uvalpha1)
			: "%eax"
			);
		}
		else if(dstbpp==16)
		{
			asm volatile(
				YSCALEYUV2RGB1b
		/* mm2=B, %%mm4=G, %%mm5=R, %%mm7=0 */
#ifdef DITHER1XBPP
michael's avatar
michael committed
1310 1311 1312
				"paddusb b5Dither, %%mm2	\n\t"
				"paddusb g6Dither, %%mm4	\n\t"
				"paddusb r5Dither, %%mm5	\n\t"
1313
#endif
1314

1315
				WRITEBGR16
michael's avatar
michael committed
1316
			:: "r" (buf0), "r" (buf0), "r" (uvbuf0), "r" (uvbuf1), "r" (dest), "m" (dstW),
1317 1318 1319
			"m" (yalpha1), "m" (uvalpha1)
			: "%eax"
			);
1320
		}
1321 1322
	}
#else
michael's avatar
michael committed
1323
//FIXME write 2 versions (for even & odd lines)
1324

michael's avatar
michael committed
1325
	if(dstbpp==32)
1326
	{
michael's avatar
michael committed
1327
		int i;
michael's avatar
michael committed
1328
		for(i=0; i<dstW-1; i+=2){
1329
			// vertical linear interpolation && yuv2rgb in a single step:
michael's avatar
michael committed
1330 1331
			int Y1=yuvtab_2568[buf0[i]>>7];
			int Y2=yuvtab_2568[buf0[i+1]>>7];
michael's avatar
michael committed
1332 1333
			int U=((uvbuf0[i>>1]*uvalpha1+uvbuf1[i>>1]*uvalpha)>>19);
			int V=((uvbuf0[(i>>1)+2048]*uvalpha1+uvbuf1[(i>>1)+2048]*uvalpha)>>19);
michael's avatar
michael committed
1334 1335 1336 1337 1338 1339 1340 1341 1342 1343 1344 1345 1346 1347

			int Cb= yuvtab_40cf[U];
			int Cg= yuvtab_1a1e[V] + yuvtab_0c92[U];
			int Cr= yuvtab_3343[V];

			dest[4*i+0]=clip_table[((Y1 + Cb) >>13)];
			dest[4*i+1]=clip_table[((Y1 + Cg) >>13)];
			dest[4*i+2]=clip_table[((Y1 + Cr) >>13)];

			dest[4*i+4]=clip_table[((Y2 + Cb) >>13)];
			dest[4*i+5]=clip_table[((Y2 + Cg) >>13)];
			dest[4*i+6]=clip_table[((Y2 + Cr) >>13)];
		}
	}
michael's avatar
michael committed
1348
	else if(dstbpp==24)
michael's avatar
michael committed
1349
	{
michael's avatar
michael committed
1350
		int i;
michael's avatar
michael committed
1351
		for(i=0; i<dstW-1; i+=2){
michael's avatar
michael committed
1352 1353 1354
			// vertical linear interpolation && yuv2rgb in a single step:
			int Y1=yuvtab_2568[buf0[i]>>7];
			int Y2=yuvtab_2568[buf0[i+1]>>7];
michael's avatar
michael committed
1355 1356
			int U=((uvbuf0[i>>1]*uvalpha1+uvbuf1[i>>1]*uvalpha)>>19);
			int V=((uvbuf0[(i>>1)+2048]*uvalpha1+uvbuf1[(i>>1)+2048]*uvalpha)>>19);
michael's avatar
michael committed
1357 1358 1359 1360 1361 1362 1363 1364 1365 1366 1367 1368 1369

			int Cb= yuvtab_40cf[U];
			int Cg= yuvtab_1a1e[V] + yuvtab_0c92[U];
			int Cr= yuvtab_3343[V];

			dest[0]=clip_table[((Y1 + Cb) >>13)];
			dest[1]=clip_table[((Y1 + Cg) >>13)];
			dest[2]=clip_table[((Y1 + Cr) >>13)];

			dest[3]=clip_table[((Y2 + Cb) >>13)];
			dest[4]=clip_table[((Y2 + Cg) >>13)];
			dest[5]=clip_table[((Y2 + Cr) >>13)];
			dest+=6;
1370 1371 1372 1373
		}
	}
	else if(dstbpp==16)
	{
michael's avatar
michael committed
1374
		int i;
michael's avatar
michael committed
1375
		for(i=0; i<dstW-1; i+=2){
1376
			// vertical linear interpolation && yuv2rgb in a single step:
michael's avatar
michael committed
1377 1378
			int Y1=yuvtab_2568[buf0[i]>>7];
			int Y2=yuvtab_2568[buf0[i+1]>>7];
michael's avatar
michael committed
1379 1380
			int U=((uvbuf0[i>>1]*uvalpha1+uvbuf1[i>>1]*uvalpha)>>19);
			int V=((uvbuf0[(i>>1)+2048]*uvalpha1+uvbuf1[(i>>1)+2048]*uvalpha)>>19);
1381

michael's avatar
michael committed
1382 1383 1384 1385
			int Cb= yuvtab_40cf[U];
			int Cg= yuvtab_1a1e[V] + yuvtab_0c92[U];
			int Cr= yuvtab_3343[V];

1386
			((uint16_t*)dest)[i] =
michael's avatar
michael committed
1387 1388 1389
				clip_table16b[(Y1 + Cb) >>13] |
				clip_table16g[(Y1 + Cg) >>13] |
				clip_table16r[(Y1 + Cr) >>13];
michael's avatar
michael committed
1390 1391

			((uint16_t*)dest)[i+1] =
michael's avatar
michael committed
1392 1393 1394
				clip_table16b[(Y2 + Cb) >>13] |
				clip_table16g[(Y2 + Cg) >>13] |
				clip_table16r[(Y2 + Cr) >>13];
1395 1396 1397 1398
		}
	}
	else if(dstbpp==15)
	{
michael's avatar
michael committed
1399
		int i;
michael's avatar
michael committed
1400
		for(i=0; i<dstW-1; i+=2){
1401
			// vertical linear interpolation && yuv2rgb in a single step:
michael's avatar
michael committed
1402 1403
			int Y1=yuvtab_2568[buf0[i]>>7];
			int Y2=yuvtab_2568[buf0[i+1]>>7];
michael's avatar
michael committed
1404 1405
			int U=((uvbuf0[i>>1]*uvalpha1+uvbuf1[i>>1]*uvalpha)>>19);
			int V=((uvbuf0[(i>>1)+2048]*uvalpha1+uvbuf1[(i>>1)+2048]*uvalpha)>>19);
1406

michael's avatar
michael committed
1407 1408 1409 1410
			int Cb= yuvtab_40cf[U];
			int Cg= yuvtab_1a1e[V] + yuvtab_0c92[U];
			int Cr= yuvtab_3343[V];

1411
			((uint16_t*)dest)[i] =
michael's avatar
michael committed
1412 1413 1414 1415
				clip_table15b[(Y1 + Cb) >>13] |
				clip_table15g[(Y1 + Cg) >>13] |
				clip_table15r[(Y1 + Cr) >>13];

michael's avatar
michael committed
1416
			((uint16_t*)dest)[i+1] =
michael's avatar
michael committed
1417 1418 1419
				clip_table15b[(Y2 + Cb) >>13] |
				clip_table15g[(Y2 + Cg) >>13] |
				clip_table15r[(Y2 + Cr) >>13];
1420 1421
		}
	}
1422 1423 1424
#endif
}

1425 1426 1427
// Bilinear / Bicubic scaling
static inline void RENAME(hScale)(int16_t *dst, int dstW, uint8_t *src, int srcW, int xInc,
				  int16_t *filter, int16_t *filterPos, int filterSize)
michael's avatar
michael committed
1428
{
1429 1430 1431 1432 1433 1434 1435 1436 1437 1438 1439 1440 1441 1442 1443 1444 1445 1446 1447 1448 1449 1450 1451 1452 1453 1454 1455 1456 1457 1458 1459 1460
#ifdef HAVE_MMX
	if(filterSize==4) // allways true for upscaling, sometimes for down too
	{
		int counter= -2*dstW;
		filter-= counter*2;
		filterPos-= counter/2;
		dst-= counter/2;
		asm volatile(
			"pxor %%mm7, %%mm7		\n\t"
			"movq w02, %%mm6		\n\t"
			"pushl %%ebp			\n\t" // we use 7 regs here ...
			"movl %%eax, %%ebp		\n\t"
			".balign 16			\n\t"
			"1:				\n\t"
			"movzwl (%2, %%ebp), %%eax	\n\t"
			"movzwl 2(%2, %%ebp), %%ebx	\n\t"
			"movq (%1, %%ebp, 4), %%mm1	\n\t"
			"movq 8(%1, %%ebp, 4), %%mm3	\n\t"
			"movd (%3, %%eax), %%mm0	\n\t"
			"movd (%3, %%ebx), %%mm2	\n\t"
			"punpcklbw %%mm7, %%mm0		\n\t"
			"punpcklbw %%mm7, %%mm2		\n\t"
			"pmaddwd %%mm1, %%mm0		\n\t"
			"pmaddwd %%mm2, %%mm3		\n\t"
			"psrad $8, %%mm0		\n\t"
			"psrad $8, %%mm3		\n\t"
			"packssdw %%mm3, %%mm0		\n\t"
			"pmaddwd %%mm6, %%mm0		\n\t"
			"packssdw %%mm0, %%mm0		\n\t"
			"movd %%mm0, (%4, %%ebp)	\n\t"
			"addl $4, %%ebp			\n\t"
			" jnc 1b			\n\t"
michael's avatar
michael committed
1461

1462 1463 1464 1465 1466 1467 1468 1469 1470 1471 1472 1473 1474 1475 1476 1477 1478 1479 1480 1481 1482 1483 1484 1485 1486 1487 1488 1489 1490 1491 1492 1493 1494 1495 1496 1497 1498 1499 1500 1501 1502 1503 1504 1505 1506 1507 1508 1509 1510
			"popl %%ebp			\n\t"
			: "+a" (counter)
			: "c" (filter), "d" (filterPos), "S" (src), "D" (dst)
			: "%ebx"
		);
	}
	else if(filterSize==8)
	{
		int counter= -2*dstW;
		filter-= counter*4;
		filterPos-= counter/2;
		dst-= counter/2;
		asm volatile(
			"pxor %%mm7, %%mm7		\n\t"
			"movq w02, %%mm6		\n\t"
			"pushl %%ebp			\n\t" // we use 7 regs here ...
			"movl %%eax, %%ebp		\n\t"
			".balign 16			\n\t"
			"1:				\n\t"
			"movzwl (%2, %%ebp), %%eax	\n\t"
			"movzwl 2(%2, %%ebp), %%ebx	\n\t"
			"movq (%1, %%ebp, 8), %%mm1	\n\t"
			"movq 16(%1, %%ebp, 8), %%mm3	\n\t"
			"movd (%3, %%eax), %%mm0	\n\t"
			"movd (%3, %%ebx), %%mm2	\n\t"
			"punpcklbw %%mm7, %%mm0		\n\t"
			"punpcklbw %%mm7, %%mm2		\n\t"
			"pmaddwd %%mm1, %%mm0		\n\t"
			"pmaddwd %%mm2, %%mm3		\n\t"

			"movq 8(%1, %%ebp, 8), %%mm1	\n\t"
			"movq 24(%1, %%ebp, 8), %%mm5	\n\t"
			"movd 4(%3, %%eax), %%mm4	\n\t"
			"movd 4(%3, %%ebx), %%mm2	\n\t"
			"punpcklbw %%mm7, %%mm4		\n\t"
			"punpcklbw %%mm7, %%mm2		\n\t"
			"pmaddwd %%mm1, %%mm4		\n\t"
			"pmaddwd %%mm2, %%mm5		\n\t"
			"paddd %%mm4, %%mm0		\n\t"
			"paddd %%mm5, %%mm3		\n\t"
						
			"psrad $8, %%mm0		\n\t"
			"psrad $8, %%mm3		\n\t"
			"packssdw %%mm3, %%mm0		\n\t"
			"pmaddwd %%mm6, %%mm0		\n\t"
			"packssdw %%mm0, %%mm0		\n\t"
			"movd %%mm0, (%4, %%ebp)	\n\t"
			"addl $4, %%ebp			\n\t"
			" jnc 1b			\n\t"
michael's avatar
michael committed
1511

1512 1513 1514 1515 1516 1517 1518 1519 1520 1521 1522 1523 1524 1525 1526 1527 1528 1529 1530 1531 1532 1533 1534 1535 1536 1537 1538 1539 1540 1541 1542 1543 1544 1545 1546 1547 1548 1549 1550 1551 1552 1553 1554 1555 1556 1557 1558 1559
			"popl %%ebp			\n\t"
			: "+a" (counter)
			: "c" (filter), "d" (filterPos), "S" (src), "D" (dst)
			: "%ebx"
		);
	}
	else
	{
		int counter= -2*dstW;
//		filter-= counter*filterSize/2;
		filterPos-= counter/2;
		dst-= counter/2;
		asm volatile(
			"pxor %%mm7, %%mm7		\n\t"
			"movq w02, %%mm6		\n\t"
			".balign 16			\n\t"
			"1:				\n\t"
			"movl %2, %%ecx			\n\t"
			"movzwl (%%ecx, %0), %%eax	\n\t"
			"movzwl 2(%%ecx, %0), %%ebx	\n\t"
			"movl %5, %%ecx			\n\t"
			"pxor %%mm4, %%mm4		\n\t"
			"pxor %%mm5, %%mm5		\n\t"
			"2:				\n\t"
			"movq (%1), %%mm1		\n\t"
			"movq (%1, %6), %%mm3		\n\t"
			"movd (%%ecx, %%eax), %%mm0	\n\t"
			"movd (%%ecx, %%ebx), %%mm2	\n\t"
			"punpcklbw %%mm7, %%mm0		\n\t"
			"punpcklbw %%mm7, %%mm2		\n\t"
			"pmaddwd %%mm1, %%mm0		\n\t"
			"pmaddwd %%mm2, %%mm3		\n\t"
			"paddd %%mm3, %%mm5		\n\t"
			"paddd %%mm0, %%mm4		\n\t"
			"addl $8, %1			\n\t"
			"addl $4, %%ecx			\n\t"
			"cmpl %4, %%ecx			\n\t"
			" jb 2b				\n\t"
			"addl %6, %1			\n\t"
			"psrad $8, %%mm4		\n\t"
			"psrad $8, %%mm5		\n\t"
			"packssdw %%mm5, %%mm4		\n\t"
			"pmaddwd %%mm6, %%mm4		\n\t"
			"packssdw %%mm4, %%mm4		\n\t"
			"movl %3, %%eax			\n\t"
			"movd %%mm4, (%%eax, %0)	\n\t"
			"addl $4, %0			\n\t"
			" jnc 1b			\n\t"
michael's avatar
michael committed
1560

1561 1562 1563
			: "+r" (counter)
			: "r" (filter), "m" (filterPos), "m" (dst), "m"(src+filterSize),
			  "m" (src), "r" (filterSize*2)
michael's avatar
typo  
michael committed
1564
			: "%ebx", "%eax", "%ecx"
1565 1566 1567 1568 1569 1570 1571 1572 1573
		);
	}
#else
	int i;
	for(i=0; i<dstW; i++)
	{
		int j;
		int srcPos= filterPos[i];
		int val=0;
michael's avatar
michael committed
1574
//		printf("filterPos: %d\n", filterPos[i]);
1575 1576 1577 1578 1579 1580 1581 1582 1583 1584 1585
		for(j=0; j<filterSize; j++)
		{
//			printf("filter: %d, src: %d\n", filter[i], src[srcPos + j]);
			val += ((int)src[srcPos + j])*filter[filterSize*i + j];
		}
//		filter += hFilterSize;
		dst[i] = MIN(MAX(0, val>>7), (1<<15)-1); // the cubic equation does overflow ...
//		dst[i] = val>>7;
	}
#endif
}
michael's avatar
michael committed
1586
      // *** horizontal scale Y line to temp buffer
1587 1588
static inline void RENAME(hyscale)(uint16_t *dst, int dstWidth, uint8_t *src, int srcW, int xInc)
{
michael's avatar
michael committed
1589 1590 1591 1592
#ifdef HAVE_MMX
	// use the new MMX scaler if th mmx2 cant be used (its faster than the x86asm one)
    if(sws_flags != SWS_FAST_BILINEAR || (!canMMX2BeUsed))
#else
1593
    if(sws_flags != SWS_FAST_BILINEAR)
michael's avatar
michael committed
1594
#endif
1595 1596 1597 1598 1599
    {
    	RENAME(hScale)(dst, dstWidth, src, srcW, xInc, hLumFilter, hLumFilterPos, hLumFilterSize);
    }
    else // Fast Bilinear upscale / crap downscale
    {
michael's avatar
michael committed
1600 1601
#ifdef ARCH_X86
#ifdef HAVE_MMX2
michael's avatar
michael committed
1602
	int i;
michael's avatar
michael committed
1603 1604 1605 1606 1607 1608 1609 1610 1611 1612 1613 1614 1615 1616 1617 1618 1619 1620 1621 1622 1623 1624 1625 1626 1627
	if(canMMX2BeUsed)
	{
		asm volatile(
			"pxor %%mm7, %%mm7		\n\t"
			"pxor %%mm2, %%mm2		\n\t" // 2*xalpha
			"movd %5, %%mm6			\n\t" // xInc&0xFFFF
			"punpcklwd %%mm6, %%mm6		\n\t"
			"punpcklwd %%mm6, %%mm6		\n\t"
			"movq %%mm6, %%mm2		\n\t"
			"psllq $16, %%mm2		\n\t"
			"paddw %%mm6, %%mm2		\n\t"
			"psllq $16, %%mm2		\n\t"
			"paddw %%mm6, %%mm2		\n\t"
			"psllq $16, %%mm2		\n\t" //0,t,2t,3t		t=xInc&0xFF
			"movq %%mm2, temp0		\n\t"
			"movd %4, %%mm6			\n\t" //(xInc*4)&0xFFFF
			"punpcklwd %%mm6, %%mm6		\n\t"
			"punpcklwd %%mm6, %%mm6		\n\t"
			"xorl %%eax, %%eax		\n\t" // i
			"movl %0, %%esi			\n\t" // src
			"movl %1, %%edi			\n\t" // buf1
			"movl %3, %%edx			\n\t" // (xInc*4)>>16
			"xorl %%ecx, %%ecx		\n\t"
			"xorl %%ebx, %%ebx		\n\t"
			"movw %4, %%bx			\n\t" // (xInc*4)&0xFFFF
michael's avatar
michael committed
1628

michael's avatar
michael committed
1629
#define FUNNY_Y_CODE \
michael's avatar
michael committed
1630 1631 1632
			PREFETCH" 1024(%%esi)		\n\t"\
			PREFETCH" 1056(%%esi)		\n\t"\
			PREFETCH" 1088(%%esi)		\n\t"\
michael's avatar
michael committed
1633 1634 1635
			"call funnyYCode		\n\t"\
			"movq temp0, %%mm2		\n\t"\
			"xorl %%ecx, %%ecx		\n\t"
michael's avatar
michael committed
1636

michael's avatar
michael committed
1637 1638 1639 1640 1641 1642 1643 1644 1645 1646 1647 1648 1649
FUNNY_Y_CODE
FUNNY_Y_CODE
FUNNY_Y_CODE
FUNNY_Y_CODE
FUNNY_Y_CODE
FUNNY_Y_CODE
FUNNY_Y_CODE
FUNNY_Y_CODE

			:: "m" (src), "m" (dst), "m" (dstWidth), "m" ((xInc*4)>>16),
			"m" ((xInc*4)&0xFFFF), "m" (xInc&0xFFFF)
			: "%eax", "%ebx", "%ecx", "%edx", "%esi", "%edi"
		);
michael's avatar
michael committed
1650
		for(i=dstWidth-1; (i*xInc)>>16 >=srcW-1; i--) dst[i] = src[srcW-1]*128;
michael's avatar
michael committed
1651 1652 1653 1654 1655 1656 1657 1658 1659
	}
	else
	{
#endif
	//NO MMX just normal asm ...
	asm volatile(
		"xorl %%eax, %%eax		\n\t" // i
		"xorl %%ebx, %%ebx		\n\t" // xx
		"xorl %%ecx, %%ecx		\n\t" // 2*xalpha
1660
		".balign 16			\n\t"
michael's avatar
michael committed
1661 1662 1663 1664 1665 1666 1667 1668 1669 1670 1671 1672 1673 1674 1675 1676 1677 1678 1679 1680 1681 1682 1683 1684 1685 1686 1687 1688 1689 1690 1691 1692 1693 1694 1695 1696 1697 1698
		"1:				\n\t"
		"movzbl  (%0, %%ebx), %%edi	\n\t" //src[xx]
		"movzbl 1(%0, %%ebx), %%esi	\n\t" //src[xx+1]
		"subl %%edi, %%esi		\n\t" //src[xx+1] - src[xx]
		"imull %%ecx, %%esi		\n\t" //(src[xx+1] - src[xx])*2*xalpha
		"shll $16, %%edi		\n\t"
		"addl %%edi, %%esi		\n\t" //src[xx+1]*2*xalpha + src[xx]*(1-2*xalpha)
		"movl %1, %%edi			\n\t"
		"shrl $9, %%esi			\n\t"
		"movw %%si, (%%edi, %%eax, 2)	\n\t"
		"addw %4, %%cx			\n\t" //2*xalpha += xInc&0xFF
		"adcl %3, %%ebx			\n\t" //xx+= xInc>>8 + carry

		"movzbl (%0, %%ebx), %%edi	\n\t" //src[xx]
		"movzbl 1(%0, %%ebx), %%esi	\n\t" //src[xx+1]
		"subl %%edi, %%esi		\n\t" //src[xx+1] - src[xx]
		"imull %%ecx, %%esi		\n\t" //(src[xx+1] - src[xx])*2*xalpha
		"shll $16, %%edi		\n\t"
		"addl %%edi, %%esi		\n\t" //src[xx+1]*2*xalpha + src[xx]*(1-2*xalpha)
		"movl %1, %%edi			\n\t"
		"shrl $9, %%esi			\n\t"
		"movw %%si, 2(%%edi, %%eax, 2)	\n\t"
		"addw %4, %%cx			\n\t" //2*xalpha += xInc&0xFF
		"adcl %3, %%ebx			\n\t" //xx+= xInc>>8 + carry


		"addl $2, %%eax			\n\t"
		"cmpl %2, %%eax			\n\t"
		" jb 1b				\n\t"


		:: "r" (src), "m" (dst), "m" (dstWidth), "m" (xInc>>16), "m" (xInc&0xFFFF)
		: "%eax", "%ebx", "%ecx", "%edi", "%esi"
		);
#ifdef HAVE_MMX2
	} //if MMX2 cant be used
#endif
#else
michael's avatar
michael committed
1699 1700 1701 1702 1703 1704 1705 1706 1707
	int i;
	unsigned int xpos=0;
	for(i=0;i<dstWidth;i++)
	{
		register unsigned int xx=xpos>>16;
		register unsigned int xalpha=(xpos&0xFFFF)>>9;
		dst[i]= (src[xx]<<7) + (src[xx+1] - src[xx])*xalpha;
		xpos+=xInc;
	}
michael's avatar
michael committed
1708
#endif
1709
    }
michael's avatar
michael committed
1710 1711
}

michael's avatar
michael committed
1712
inline static void RENAME(hcscale)(uint16_t *dst, int dstWidth,
michael's avatar
michael committed
1713
				uint8_t *src1, uint8_t *src2, int srcW, int xInc)
michael's avatar
michael committed
1714
{
michael's avatar
michael committed
1715 1716 1717 1718
#ifdef HAVE_MMX
	// use the new MMX scaler if th mmx2 cant be used (its faster than the x86asm one)
    if(sws_flags != SWS_FAST_BILINEAR || (!canMMX2BeUsed))
#else
1719
    if(sws_flags != SWS_FAST_BILINEAR)
michael's avatar
michael committed
1720
#endif
1721 1722 1723 1724 1725 1726
    {
    	RENAME(hScale)(dst     , dstWidth, src1, srcW, xInc, hChrFilter, hChrFilterPos, hChrFilterSize);
    	RENAME(hScale)(dst+2048, dstWidth, src2, srcW, xInc, hChrFilter, hChrFilterPos, hChrFilterSize);
    }
    else // Fast Bilinear upscale / crap downscale
    {
michael's avatar
michael committed
1727 1728
#ifdef ARCH_X86
#ifdef HAVE_MMX2
michael's avatar
michael committed
1729
	int i;
michael's avatar
michael committed
1730 1731 1732 1733 1734 1735 1736 1737 1738 1739 1740 1741 1742 1743 1744 1745 1746 1747 1748 1749 1750 1751 1752 1753 1754 1755 1756
	if(canMMX2BeUsed)
	{
		asm volatile(
		"pxor %%mm7, %%mm7		\n\t"
		"pxor %%mm2, %%mm2		\n\t" // 2*xalpha
		"movd %5, %%mm6			\n\t" // xInc&0xFFFF
		"punpcklwd %%mm6, %%mm6		\n\t"
		"punpcklwd %%mm6, %%mm6		\n\t"
		"movq %%mm6, %%mm2		\n\t"
		"psllq $16, %%mm2		\n\t"
		"paddw %%mm6, %%mm2		\n\t"
		"psllq $16, %%mm2		\n\t"
		"paddw %%mm6, %%mm2		\n\t"
		"psllq $16, %%mm2		\n\t" //0,t,2t,3t		t=xInc&0xFFFF
		"movq %%mm2, temp0		\n\t"
		"movd %4, %%mm6			\n\t" //(xInc*4)&0xFFFF
		"punpcklwd %%mm6, %%mm6		\n\t"
		"punpcklwd %%mm6, %%mm6		\n\t"
		"xorl %%eax, %%eax		\n\t" // i
		"movl %0, %%esi			\n\t" // src
		"movl %1, %%edi			\n\t" // buf1
		"movl %3, %%edx			\n\t" // (xInc*4)>>16
		"xorl %%ecx, %%ecx		\n\t"
		"xorl %%ebx, %%ebx		\n\t"
		"movw %4, %%bx			\n\t" // (xInc*4)&0xFFFF

#define FUNNYUVCODE \
michael's avatar
michael committed
1757 1758 1759
			PREFETCH" 1024(%%esi)		\n\t"\
			PREFETCH" 1056(%%esi)		\n\t"\
			PREFETCH" 1088(%%esi)		\n\t"\
michael's avatar
michael committed
1760 1761 1762 1763 1764 1765 1766 1767 1768 1769 1770 1771 1772 1773 1774 1775 1776 1777 1778 1779 1780 1781 1782 1783 1784 1785 1786 1787 1788 1789 1790 1791
			"call funnyUVCode		\n\t"\
			"movq temp0, %%mm2		\n\t"\
			"xorl %%ecx, %%ecx		\n\t"

FUNNYUVCODE
FUNNYUVCODE
FUNNYUVCODE
FUNNYUVCODE

FUNNYUVCODE
FUNNYUVCODE
FUNNYUVCODE
FUNNYUVCODE
		"xorl %%eax, %%eax		\n\t" // i
		"movl %6, %%esi			\n\t" // src
		"movl %1, %%edi			\n\t" // buf1
		"addl $4096, %%edi		\n\t"

FUNNYUVCODE
FUNNYUVCODE
FUNNYUVCODE
FUNNYUVCODE

FUNNYUVCODE
FUNNYUVCODE
FUNNYUVCODE
FUNNYUVCODE

		:: "m" (src1), "m" (dst), "m" (dstWidth), "m" ((xInc*4)>>16),
		  "m" ((xInc*4)&0xFFFF), "m" (xInc&0xFFFF), "m" (src2)
		: "%eax", "%ebx", "%ecx", "%edx", "%esi", "%edi"
	);
michael's avatar
michael committed
1792
		for(i=dstWidth-1; (i*xInc)>>16 >=srcW-1; i--)
michael's avatar
michael committed
1793
		{
michael's avatar
michael committed
1794 1795 1796
//			printf("%d %d %d\n", dstWidth, i, srcW);
			dst[i] = src1[srcW-1]*128;
			dst[i+2048] = src2[srcW-1]*128;
michael's avatar
michael committed
1797 1798 1799 1800 1801 1802 1803 1804 1805
		}
	}
	else
	{
#endif
	asm volatile(
		"xorl %%eax, %%eax		\n\t" // i
		"xorl %%ebx, %%ebx		\n\t" // xx
		"xorl %%ecx, %%ecx		\n\t" // 2*xalpha
1806
		".balign 16			\n\t"
michael's avatar
michael committed
1807 1808 1809 1810 1811 1812 1813 1814 1815 1816 1817 1818 1819 1820 1821 1822 1823 1824 1825 1826 1827 1828 1829 1830 1831 1832 1833 1834 1835 1836 1837 1838 1839 1840 1841 1842
		"1:				\n\t"
		"movl %0, %%esi			\n\t"
		"movzbl  (%%esi, %%ebx), %%edi	\n\t" //src[xx]
		"movzbl 1(%%esi, %%ebx), %%esi	\n\t" //src[xx+1]
		"subl %%edi, %%esi		\n\t" //src[xx+1] - src[xx]
		"imull %%ecx, %%esi		\n\t" //(src[xx+1] - src[xx])*2*xalpha
		"shll $16, %%edi		\n\t"
		"addl %%edi, %%esi		\n\t" //src[xx+1]*2*xalpha + src[xx]*(1-2*xalpha)
		"movl %1, %%edi			\n\t"
		"shrl $9, %%esi			\n\t"
		"movw %%si, (%%edi, %%eax, 2)	\n\t"

		"movzbl  (%5, %%ebx), %%edi	\n\t" //src[xx]
		"movzbl 1(%5, %%ebx), %%esi	\n\t" //src[xx+1]
		"subl %%edi, %%esi		\n\t" //src[xx+1] - src[xx]
		"imull %%ecx, %%esi		\n\t" //(src[xx+1] - src[xx])*2*xalpha
		"shll $16, %%edi		\n\t"
		"addl %%edi, %%esi		\n\t" //src[xx+1]*2*xalpha + src[xx]*(1-2*xalpha)
		"movl %1, %%edi			\n\t"
		"shrl $9, %%esi			\n\t"
		"movw %%si, 4096(%%edi, %%eax, 2)\n\t"

		"addw %4, %%cx			\n\t" //2*xalpha += xInc&0xFF
		"adcl %3, %%ebx			\n\t" //xx+= xInc>>8 + carry
		"addl $1, %%eax			\n\t"
		"cmpl %2, %%eax			\n\t"
		" jb 1b				\n\t"

		:: "m" (src1), "m" (dst), "m" (dstWidth), "m" (xInc>>16), "m" (xInc&0xFFFF),
		"r" (src2)
		: "%eax", "%ebx", "%ecx", "%edi", "%esi"
		);
#ifdef HAVE_MMX2
	} //if MMX2 cant be used
#endif
#else
michael's avatar
michael committed
1843 1844 1845 1846 1847 1848 1849 1850
	int i;
	unsigned int xpos=0;
	for(i=0;i<dstWidth;i++)
	{
		register unsigned int xx=xpos>>16;
		register unsigned int xalpha=(xpos&0xFFFF)>>9;
		dst[i]=(src1[xx]*(xalpha^127)+src1[xx+1]*xalpha);
		dst[i+2048]=(src2[xx]*(xalpha^127)+src2[xx+1]*xalpha);
michael's avatar
michael committed
1851 1852 1853 1854
/* slower
	  dst[i]= (src1[xx]<<7) + (src1[xx+1] - src1[xx])*xalpha;
	  dst[i+2048]=(src2[xx]<<7) + (src2[xx+1] - src2[xx])*xalpha;
*/
michael's avatar
michael committed
1855 1856
		xpos+=xInc;
	}
michael's avatar
michael committed
1857
#endif
1858 1859 1860
   }
}

michael's avatar
michael committed
1861
static inline void RENAME(initFilter)(int16_t *dstFilter, int16_t *filterPos, int *filterSize, int xInc,
michael's avatar
michael committed
1862
				      int srcW, int dstW, int filterAlign, int one)
1863 1864
{
	int i;
michael's avatar
michael committed
1865
	double filter[8000];
1866
#ifdef HAVE_MMX
michael's avatar
michael committed
1867
	asm volatile("emms\n\t"::: "memory"); //FIXME this shouldnt be required but it IS (even for non mmx versions)
1868
#endif
michael's avatar
michael committed
1869 1870 1871 1872 1873 1874 1875 1876 1877 1878 1879 1880 1881 1882 1883

	if(ABS(xInc - 0x10000) <10) // unscaled
	{
		int i;
		*filterSize= (1 +(filterAlign-1)) & (~(filterAlign-1)); // 1 or 4 normaly
		for(i=0; i<dstW*(*filterSize); i++) filter[i]=0;

		for(i=0; i<dstW; i++)
		{
			filter[i*(*filterSize)]=1;
			filterPos[i]=i;
		}

	}
	else if(xInc <= (1<<16) || sws_flags==SWS_FAST_BILINEAR) // upscale
1884 1885 1886 1887 1888 1889
	{
		int i;
		int xDstInSrc;
		if(sws_flags==SWS_BICUBIC) *filterSize= 4;
		else			   *filterSize= 2;
//		printf("%d %d %d\n", filterSize, srcW, dstW);
michael's avatar
michael committed
1890 1891
		*filterSize= (*filterSize +(filterAlign-1)) & (~(filterAlign-1));

1892 1893 1894 1895 1896
		xDstInSrc= xInc - 0x8000;
		for(i=0; i<dstW; i++)
		{
			int xx= (xDstInSrc>>16) - (*filterSize>>1) + 1;
			int j;
michael's avatar
michael committed
1897

1898 1899
			filterPos[i]= xx;
			if(sws_flags == SWS_BICUBIC)
michael's avatar
michael committed
1900
			{
1901
				double d= ABS(((xx+1)<<16) - xDstInSrc)/(double)(1<<16);
michael's avatar
michael committed
1902
				double y1,y2,y3,y4;
1903 1904
				double A= -0.75;
					// Equation is from VirtualDub
michael's avatar
michael committed
1905 1906 1907 1908
				y1 = (        +     A*d -       2.0*A*d*d +       A*d*d*d);
				y2 = (+ 1.0             -     (A+3.0)*d*d + (A+2.0)*d*d*d);
				y3 = (        -     A*d + (2.0*A+3.0)*d*d - (A+2.0)*d*d*d);
				y4 = (                  +           A*d*d -       A*d*d*d);
michael's avatar
michael committed
1909

1910 1911 1912 1913 1914 1915 1916
//				printf("%d %d %d \n", coeff, (int)d, xDstInSrc);
				filter[i*(*filterSize) + 0]= y1;
				filter[i*(*filterSize) + 1]= y2;
				filter[i*(*filterSize) + 2]= y3;
				filter[i*(*filterSize) + 3]= y4;
//				printf("%1.3f %d, %d, %d, %d\n",d , y1, y2, y3, y4);
			}
michael's avatar
michael committed
1917
			else
1918 1919 1920 1921
			{
				for(j=0; j<*filterSize; j++)
				{
					double d= ABS((xx<<16) - xDstInSrc)/(double)(1<<16);
michael's avatar
michael committed
1922
					double coeff= 1.0 - d;
michael's avatar
michael committed
1923
					if(coeff<0) coeff=0;
1924 1925 1926 1927
	//				printf("%d %d %d \n", coeff, (int)d, xDstInSrc);
					filter[i*(*filterSize) + j]= coeff;
					xx++;
				}
michael's avatar
michael committed
1928 1929
			}
			xDstInSrc+= xInc;
1930 1931 1932 1933 1934 1935 1936 1937
		}
	}
	else // downscale
	{
		int xDstInSrc;
		if(sws_flags==SWS_BICUBIC) *filterSize= (int)ceil(1 + 4.0*srcW / (double)dstW);
		else			   *filterSize= (int)ceil(1 + 2.0*srcW / (double)dstW);
//		printf("%d %d %d\n", *filterSize, srcW, dstW);
michael's avatar
michael committed
1938 1939
		*filterSize= (*filterSize +(filterAlign-1)) & (~(filterAlign-1));

1940 1941 1942 1943 1944
		xDstInSrc= xInc - 0x8000;
		for(i=0; i<dstW; i++)
		{
			int xx= (int)((double)xDstInSrc/(double)(1<<16) - *filterSize*0.5 + 0.5);
			int j;
michael's avatar
michael committed
1945

1946 1947 1948 1949
			filterPos[i]= xx;
			for(j=0; j<*filterSize; j++)
			{
				double d= ABS((xx<<16) - xDstInSrc)/(double)xInc;
michael's avatar
michael committed
1950
				double coeff;
1951
				if(sws_flags == SWS_BICUBIC)
michael's avatar
michael committed
1952
				{
1953 1954 1955 1956
					double A= -0.75;
//					d*=2;
					// Equation is from VirtualDub
					if(d<1.0)
michael's avatar
michael committed
1957
						coeff = (1.0 - (A+3.0)*d*d + (A+2.0)*d*d*d);
1958
					else if(d<2.0)
michael's avatar
michael committed
1959
						coeff = (-4.0*A + 8.0*A*d - 5.0*A*d*d + A*d*d*d);
1960
					else
michael's avatar
michael committed
1961
						coeff=0.0;
1962
				}
michael's avatar
michael committed
1963
				else
1964
				{
michael's avatar
michael committed
1965
					coeff= 1.0 - d;
michael's avatar
michael committed
1966
					if(coeff<0) coeff=0;
1967
				}
michael's avatar
michael committed
1968
//				if(filterAlign==1) printf("%d %d %d \n", coeff, (int)d, xDstInSrc);
1969 1970 1971 1972 1973 1974
				filter[i*(*filterSize) + j]= coeff;
				xx++;
			}
			xDstInSrc+= xInc;
		}
	}
michael's avatar
michael committed
1975

1976 1977 1978 1979 1980 1981 1982 1983 1984 1985 1986 1987 1988 1989 1990 1991
	//fix borders
	for(i=0; i<dstW; i++)
	{
		int j;
		if(filterPos[i] < 0)
		{
			// Move filter coeffs left to compensate for filterPos
			for(j=1; j<*filterSize; j++)
			{
				int left= MAX(j + filterPos[i], 0);
				filter[i*(*filterSize) + left] += filter[i*(*filterSize) + j];
				filter[i*(*filterSize) + j]=0;
			}
			filterPos[i]= 0;
		}

michael's avatar
michael committed
1992
		if(filterPos[i] + (*filterSize) > srcW)
1993
		{
michael's avatar
michael committed
1994
			int shift= filterPos[i] + (*filterSize) - srcW;
1995
			// Move filter coeffs right to compensate for filterPos
michael's avatar
michael committed
1996
			for(j=(*filterSize)-2; j>=0; j--)
1997
			{
michael's avatar
michael committed
1998
				int right= MIN(j + shift, (*filterSize)-1);
1999 2000 2001
				filter[i*(*filterSize) +right] += filter[i*(*filterSize) +j];
				filter[i*(*filterSize) +j]=0;
			}
michael's avatar
michael committed
2002
			filterPos[i]= srcW - (*filterSize);
2003 2004
		}
	}
michael's avatar
michael committed
2005 2006 2007 2008

	//FIXME try to align filterpos if possible / try to shift filterpos to put zeros at the end
	// and skip these than later

2009 2010 2011 2012 2013
	//Normalize
	for(i=0; i<dstW; i++)
	{
		int j;
		double sum=0;
michael's avatar
michael committed
2014
		double scale= one;
2015 2016 2017 2018 2019 2020 2021
		for(j=0; j<*filterSize; j++)
		{
			sum+= filter[i*(*filterSize) + j];
		}
		scale/= sum;
		for(j=0; j<*filterSize; j++)
		{
michael's avatar
michael committed
2022
			dstFilter[i*(*filterSize) + j]= (int)(filter[i*(*filterSize) + j]*scale);
2023 2024
		}
	}
michael's avatar
michael committed
2025
}
2026

michael's avatar
michael committed
2027 2028 2029 2030 2031 2032 2033 2034 2035 2036 2037 2038 2039 2040 2041 2042 2043 2044 2045 2046 2047 2048 2049 2050 2051 2052 2053 2054 2055 2056 2057 2058 2059 2060 2061 2062 2063 2064 2065 2066 2067 2068 2069 2070 2071 2072 2073 2074 2075 2076 2077 2078 2079 2080 2081 2082 2083 2084 2085 2086 2087 2088 2089 2090 2091 2092 2093 2094 2095 2096 2097 2098 2099 2100 2101 2102 2103 2104 2105 2106 2107 2108 2109 2110 2111 2112 2113 2114 2115 2116 2117 2118 2119 2120 2121 2122 2123 2124 2125 2126 2127 2128 2129 2130 2131 2132 2133 2134 2135 2136 2137 2138 2139 2140 2141
#ifdef HAVE_MMX2
static void initMMX2HScaler(int dstW, int xInc, uint8_t *funnyCode)
{
	uint8_t *fragment;
	int imm8OfPShufW1;
	int imm8OfPShufW2;
	int fragmentLength;

	int xpos, i;

	// create an optimized horizontal scaling routine

	//code fragment

	asm volatile(
		"jmp 9f				\n\t"
	// Begin
		"0:				\n\t"
		"movq (%%esi), %%mm0		\n\t" //FIXME Alignment
		"movq %%mm0, %%mm1		\n\t"
		"psrlq $8, %%mm0		\n\t"
		"punpcklbw %%mm7, %%mm1	\n\t"
		"movq %%mm2, %%mm3		\n\t"
		"punpcklbw %%mm7, %%mm0	\n\t"
		"addw %%bx, %%cx		\n\t" //2*xalpha += (4*lumXInc)&0xFFFF
		"pshufw $0xFF, %%mm1, %%mm1	\n\t"
		"1:				\n\t"
		"adcl %%edx, %%esi		\n\t" //xx+= (4*lumXInc)>>16 + carry
		"pshufw $0xFF, %%mm0, %%mm0	\n\t"
		"2:				\n\t"
		"psrlw $9, %%mm3		\n\t"
		"psubw %%mm1, %%mm0		\n\t"
		"pmullw %%mm3, %%mm0		\n\t"
		"paddw %%mm6, %%mm2		\n\t" // 2*alpha += xpos&0xFFFF
		"psllw $7, %%mm1		\n\t"
		"paddw %%mm1, %%mm0		\n\t"

		"movq %%mm0, (%%edi, %%eax)	\n\t"

		"addl $8, %%eax			\n\t"
	// End
		"9:				\n\t"
//		"int $3\n\t"
		"leal 0b, %0			\n\t"
		"leal 1b, %1			\n\t"
		"leal 2b, %2			\n\t"
		"decl %1			\n\t"
		"decl %2			\n\t"
		"subl %0, %1			\n\t"
		"subl %0, %2			\n\t"
		"leal 9b, %3			\n\t"
		"subl %0, %3			\n\t"
		:"=r" (fragment), "=r" (imm8OfPShufW1), "=r" (imm8OfPShufW2),
		"=r" (fragmentLength)
	);

	xpos= 0; //lumXInc/2 - 0x8000; // difference between pixel centers

	for(i=0; i<dstW/8; i++)
	{
		int xx=xpos>>16;

		if((i&3) == 0)
		{
			int a=0;
			int b=((xpos+xInc)>>16) - xx;
			int c=((xpos+xInc*2)>>16) - xx;
			int d=((xpos+xInc*3)>>16) - xx;

			memcpy(funnyCode + fragmentLength*i/4, fragment, fragmentLength);

			funnyCode[fragmentLength*i/4 + imm8OfPShufW1]=
			funnyCode[fragmentLength*i/4 + imm8OfPShufW2]=
				a | (b<<2) | (c<<4) | (d<<6);

			// if we dont need to read 8 bytes than dont :), reduces the chance of
			// crossing a cache line
			if(d<3) funnyCode[fragmentLength*i/4 + 1]= 0x6E;

			funnyCode[fragmentLength*(i+4)/4]= RET;
		}
		xpos+=xInc;
	}
/*
	xpos= 0; //chrXInc/2 - 0x10000; // difference between centers of chrom samples
	for(i=0; i<dstUVw/8; i++)
	{
		int xx=xpos>>16;

		if((i&3) == 0)
		{
			int a=0;
			int b=((xpos+chrXInc)>>16) - xx;
			int c=((xpos+chrXInc*2)>>16) - xx;
			int d=((xpos+chrXInc*3)>>16) - xx;

			memcpy(funnyUVCode + fragmentLength*i/4, fragment, fragmentLength);

			funnyUVCode[fragmentLength*i/4 + imm8OfPShufW1]=
			funnyUVCode[fragmentLength*i/4 + imm8OfPShufW2]=
				a | (b<<2) | (c<<4) | (d<<6);

			// if we dont need to read 8 bytes than dont :), reduces the chance of
			// crossing a cache line
			if(d<3) funnyUVCode[fragmentLength*i/4 + 1]= 0x6E;

			funnyUVCode[fragmentLength*(i+4)/4]= RET;
		}
		xpos+=chrXInc;
	}
*/
//	funnyCode[0]= RET;
}
#endif // HAVE_MMX2

michael's avatar
michael committed
2142 2143 2144
static void RENAME(SwScale_YV12slice)(unsigned char* srcptr[],int stride[], int srcSliceY ,
			     int srcSliceH, uint8_t* dstptr[], int dststride, int dstbpp,
			     int srcW, int srcH, int dstW, int dstH){
2145 2146


michael's avatar
michael committed
2147 2148 2149 2150
unsigned int lumXInc= (srcW << 16) / dstW;
unsigned int lumYInc= (srcH << 16) / dstH;
unsigned int chrXInc;
unsigned int chrYInc;
2151

michael's avatar
michael committed
2152
static int dstY;
2153

michael's avatar
michael committed
2154
// used to detect a size change
2155 2156
static int oldDstW= -1;
static int oldSrcW= -1;
michael's avatar
michael committed
2157 2158 2159
static int oldDstH= -1;
static int oldSrcH= -1;
static int oldFlags=-1;
2160

michael's avatar
michael committed
2161 2162
static int lastInLumBuf;
static int lastInChrBuf;
2163

michael's avatar
michael committed
2164
int chrDstW, chrDstH;
michael's avatar
michael committed
2165

michael's avatar
michael committed
2166 2167 2168 2169 2170
static int lumBufIndex=0;
static int chrBufIndex=0;

static int firstTime=1;

michael's avatar
michael committed
2171 2172 2173 2174 2175
const int widthAlign= dstbpp==12 ? 16 : 8;
const int bytespp= (dstbpp+1)/8; //(12->1, 15&16->2, 24->3, 32->4)
const int over= dstbpp==12 ? 	  (((dstW+15)&(~15))) - dststride
				: (((dstW+7)&(~7)))*bytespp - dststride;
if(dststride%widthAlign !=0 )
michael's avatar
michael committed
2176 2177 2178
{
	if(firstTime)
		fprintf(stderr, "SwScaler: Warning: dstStride is not a multiple of %d!\n"
michael's avatar
michael committed
2179 2180
				"SwScaler:          ->cannot do aligned memory acesses anymore\n",
				widthAlign);
michael's avatar
michael committed
2181
}
michael's avatar
michael committed
2182

michael's avatar
michael committed
2183 2184 2185 2186 2187 2188 2189 2190 2191 2192 2193
if(over>0)
{
	if(firstTime)
		fprintf(stderr, "SwScaler: Warning: output width is not a multiple of 8 (16 for YV12)\n"
				"SwScaler:          and dststride is not large enough to handle %d extra bytes\n"
				"SwScaler:          ->using unoptimized C version for last line(s)\n",
				over);
}



michael's avatar
michael committed
2194
//printf("%d %d %d %d\n", srcW, srcH, dstW, dstH);
michael's avatar
michael committed
2195
//printf("%d %d %d %d\n", lumXInc, lumYInc, srcSliceY, srcSliceH);
michael's avatar
michael committed
2196

michael's avatar
michael committed
2197
#ifdef HAVE_MMX2
michael's avatar
michael committed
2198 2199 2200
canMMX2BeUsed= (lumXInc <= 0x10000 && (dstW&31)==0 && (srcW&15)==0) ? 1 : 0;
if(!canMMX2BeUsed && lumXInc <= 0x10000 && (srcW&15)==0 && sws_flags==SWS_FAST_BILINEAR)
{
michael's avatar
michael committed
2201
	if(firstTime)
michael's avatar
michael committed
2202 2203
		fprintf(stderr, "SwScaler: output Width is not a multiple of 32 -> no MMX2 scaler\n");
}
michael's avatar
michael committed
2204 2205
#else
canMMX2BeUsed=0; // should be 0 anyway but ...
michael's avatar
michael committed
2206 2207 2208 2209
#endif

if(firstTime)
{
michael's avatar
michael committed
2210 2211
#if defined (DITHER1XBPP) && defined (HAVE_MMX)
	char *dither= " dithered";
michael's avatar
michael committed
2212 2213
#else
	char *dither= "";
michael's avatar
michael committed
2214
#endif
michael's avatar
michael committed
2215 2216 2217 2218 2219 2220 2221 2222 2223 2224
	if(sws_flags==SWS_FAST_BILINEAR)
		fprintf(stderr, "SwScaler: FAST_BILINEAR scaler ");
	else if(sws_flags==SWS_BILINEAR)
		fprintf(stderr, "SwScaler: BILINEAR scaler ");
	else if(sws_flags==SWS_BICUBIC)
		fprintf(stderr, "SwScaler: BICUBIC scaler ");
	else
		fprintf(stderr, "SwScaler: ehh flags invalid?! ");

	if(dstbpp==15)
michael's avatar
michael committed
2225
		fprintf(stderr, "with%s BGR15 output ", dither);
michael's avatar
michael committed
2226
	else if(dstbpp==16)
michael's avatar
michael committed
2227
		fprintf(stderr, "with%s BGR16 output ", dither);
michael's avatar
michael committed
2228 2229 2230 2231 2232 2233 2234 2235 2236 2237 2238 2239 2240 2241 2242 2243
	else if(dstbpp==24)
		fprintf(stderr, "with BGR24 output ");
	else if(dstbpp==32)
		fprintf(stderr, "with BGR32 output ");
	else if(dstbpp==12)
		fprintf(stderr, "with YV12 output ");
	else
		fprintf(stderr, "without output ");

#ifdef HAVE_MMX2
		fprintf(stderr, "using MMX2\n");
#elif defined (HAVE_3DNOW)
		fprintf(stderr, "using 3DNOW\n");
#elif defined (HAVE_MMX)
		fprintf(stderr, "using MMX\n");
#elif defined (ARCH_X86)
michael's avatar
michael committed
2244
		fprintf(stderr, "using X86 ASM\n");
michael's avatar
michael committed
2245 2246 2247 2248 2249
#else
		fprintf(stderr, "using C\n");
#endif
}

michael's avatar
michael committed
2250

2251 2252
// match pixel 0 of the src to pixel 0 of dst and match pixel n-2 of src to pixel n-2 of dst
// n-2 is the last chrominance sample available
michael's avatar
michael committed
2253
// this is not perfect, but noone shuld notice the difference, the more correct variant
2254 2255
// would be like the vertical one, but that would require some special code for the
// first and last pixel
michael's avatar
michael committed
2256 2257 2258
if(sws_flags==SWS_FAST_BILINEAR)
{
	if(canMMX2BeUsed) 	lumXInc+= 20;
michael's avatar
michael committed
2259
#ifndef HAVE_MMX //we dont use the x86asm scaler if mmx is available
michael's avatar
michael committed
2260
	else			lumXInc = ((srcW-2)<<16)/(dstW-2) - 20;
michael's avatar
michael committed
2261
#endif
michael's avatar
michael committed
2262 2263 2264
}

if(fullUVIpol && !(dstbpp==12)) 	chrXInc= lumXInc>>1, chrDstW= dstW;
michael's avatar
michael committed
2265
else					chrXInc= lumXInc,    chrDstW= (dstW+1)>>1;
michael's avatar
michael committed
2266

michael's avatar
michael committed
2267
if(dstbpp==12)	chrYInc= lumYInc,    chrDstH= (dstH+1)>>1;
michael's avatar
michael committed
2268
else		chrYInc= lumYInc>>1, chrDstH= dstH;
2269

michael's avatar
michael committed
2270 2271
  // force calculation of the horizontal interpolation of the first line

michael's avatar
michael committed
2272
  if(srcSliceY ==0){
michael's avatar
michael committed
2273
//	printf("dstW %d, srcw %d, mmx2 %d\n", dstW, srcW, canMMX2BeUsed);
michael's avatar
michael committed
2274 2275
	lumBufIndex=0;
	chrBufIndex=0;
michael's avatar
michael committed
2276
	dstY=0;
2277

2278
	//precalculate horizontal scaler filter coefficients
michael's avatar
michael committed
2279 2280 2281 2282 2283 2284 2285 2286 2287
	if(oldDstW!=dstW || oldSrcW!=srcW || oldFlags!=sws_flags)
	{
#ifdef HAVE_MMX
		const int filterAlign=4;
#else
		const int filterAlign=1;
#endif
		oldDstW= dstW; oldSrcW= srcW; oldFlags= sws_flags;

michael's avatar
michael committed
2288 2289 2290 2291
		RENAME(initFilter)(hLumFilter, hLumFilterPos, &hLumFilterSize, lumXInc,
				srcW   , dstW   , filterAlign, 1<<14);
		RENAME(initFilter)(hChrFilter, hChrFilterPos, &hChrFilterSize, chrXInc,
				(srcW+1)>>1, chrDstW, filterAlign, 1<<14);
2292

2293 2294
#ifdef HAVE_MMX2
// cant downscale !!!
michael's avatar
michael committed
2295 2296 2297 2298 2299 2300 2301
		if(canMMX2BeUsed && sws_flags == SWS_FAST_BILINEAR)
		{
			initMMX2HScaler(dstW   , lumXInc, funnyYCode);
			initMMX2HScaler(chrDstW, chrXInc, funnyUVCode);
		}
#endif
	} // Init Horizontal stuff
2302

michael's avatar
michael committed
2303 2304 2305 2306
	if(oldDstH!=dstH || oldSrcH!=srcH || oldFlags!=sws_flags)
	{
		int i;
		oldDstH= dstH; oldSrcH= srcH; oldFlags= sws_flags; //FIXME swsflags conflict with x check
2307

michael's avatar
michael committed
2308 2309 2310
		// deallocate pixbufs
		for(i=0; i<vLumBufSize; i++) free(lumPixBuf[i]);
		for(i=0; i<vChrBufSize; i++) free(chrPixBuf[i]);
2311

michael's avatar
michael committed
2312 2313 2314
		RENAME(initFilter)(vLumFilter, vLumFilterPos, &vLumFilterSize, lumYInc,
				srcH   , dstH,    1, (1<<12)-4);
		RENAME(initFilter)(vChrFilter, vChrFilterPos, &vChrFilterSize, chrYInc,
michael's avatar
michael committed
2315
				(srcH+1)>>1, chrDstH, 1, (1<<12)-4);
2316

michael's avatar
michael committed
2317 2318 2319 2320 2321 2322 2323 2324 2325 2326 2327 2328 2329
		// Calculate Buffer Sizes so that they wont run out while handling these damn slices
		vLumBufSize= vLumFilterSize; vChrBufSize= vChrFilterSize;
		for(i=0; i<dstH; i++)
		{
			int chrI= i*chrDstH / dstH;
			int nextSlice= MAX(vLumFilterPos[i   ] + vLumFilterSize - 1,
					 ((vChrFilterPos[chrI] + vChrFilterSize - 1)<<1));
			nextSlice&= ~1; // Slices start at even boundaries
			if(vLumFilterPos[i   ] + vLumBufSize < nextSlice)
				vLumBufSize= nextSlice - vLumFilterPos[i   ];
			if(vChrFilterPos[chrI] + vChrBufSize < (nextSlice>>1))
				vChrBufSize= (nextSlice>>1) - vChrFilterPos[chrI];
		}
2330

michael's avatar
michael committed
2331 2332 2333 2334 2335 2336
		// allocate pixbufs (we use dynamic allocation because otherwise we would need to
		// allocate several megabytes to handle all possible cases)
		for(i=0; i<vLumBufSize; i++)
			lumPixBuf[i]= lumPixBuf[i+vLumBufSize]= (uint16_t*)memalign(8, 4000);
		for(i=0; i<vChrBufSize; i++)
			chrPixBuf[i]= chrPixBuf[i+vChrBufSize]= (uint16_t*)memalign(8, 8000);
2337

michael's avatar
michael committed
2338 2339 2340
		//try to avoid drawing green stuff between the right end and the stride end
		for(i=0; i<vLumBufSize; i++) memset(lumPixBuf[i], 0, 4000);
		for(i=0; i<vChrBufSize; i++) memset(chrPixBuf[i], 64, 8000);
2341

michael's avatar
michael committed
2342 2343 2344 2345 2346 2347 2348 2349 2350 2351 2352
#ifdef HAVE_MMX
		// pack filter data for mmx code
		for(i=0; i<vLumFilterSize*dstH; i++)
			lumMmxFilter[4*i]=lumMmxFilter[4*i+1]=lumMmxFilter[4*i+2]=lumMmxFilter[4*i+3]=
				vLumFilter[i];

		for(i=0; i<vChrFilterSize*chrDstH; i++)
			chrMmxFilter[4*i]=chrMmxFilter[4*i+1]=chrMmxFilter[4*i+2]=chrMmxFilter[4*i+3]=
				vChrFilter[i];
#endif
	}
2353

michael's avatar
michael committed
2354 2355 2356 2357 2358 2359 2360 2361 2362 2363 2364 2365 2366 2367 2368 2369 2370 2371 2372 2373 2374 2375 2376 2377 2378 2379 2380 2381 2382 2383 2384 2385 2386 2387 2388 2389 2390 2391 2392 2393 2394 2395 2396 2397 2398 2399 2400 2401 2402 2403 2404 2405 2406 2407 2408 2409 2410 2411 2412 2413 2414 2415 2416 2417 2418 2419 2420 2421
	if(firstTime && verbose)
	{
#ifdef HAVE_MMX2
		int mmx2=1;
#else
		int mmx2=0;
#endif
#ifdef HAVE_MMX
		int mmx=1;
#else
		int mmx=0;
#endif

#ifdef HAVE_MMX
		if(canMMX2BeUsed && sws_flags==SWS_FAST_BILINEAR)
			printf("SwScaler: using FAST_BILINEAR MMX2 scaler for horizontal scaling\n");
		else
		{
			if(hLumFilterSize==4)
				printf("SwScaler: using 4-tap MMX scaler for horizontal luminance scaling\n");
			else if(hLumFilterSize==8)
				printf("SwScaler: using 8-tap MMX scaler for horizontal luminance scaling\n");
			else
				printf("SwScaler: using n-tap MMX scaler for horizontal luminance scaling\n");

			if(hChrFilterSize==4)
				printf("SwScaler: using 4-tap MMX scaler for horizontal chrominance scaling\n");
			else if(hChrFilterSize==8)
				printf("SwScaler: using 8-tap MMX scaler for horizontal chrominance scaling\n");
			else
				printf("SwScaler: using n-tap MMX scaler for horizontal chrominance scaling\n");
		}
#elif defined (ARCH_X86)
		printf("SwScaler: using X86-Asm scaler for horizontal scaling\n");
#else
		if(sws_flags==SWS_FAST_BILINEAR)
			printf("SwScaler: using FAST_BILINEAR C scaler for horizontal scaling\n");
		else
			printf("SwScaler: using C scaler for horizontal scaling\n");
#endif

		if(dstbpp==12)
		{
			if(vLumFilterSize==1)
				printf("SwScaler: using 1-tap %s \"scaler\" for vertical scaling (YV12)\n", mmx ? "MMX" : "C");
			else
				printf("SwScaler: using n-tap %s scaler for vertical scaling (YV12)\n", mmx ? "MMX" : "C");
		}
		else
		{
			if(vLumFilterSize==1 && vChrFilterSize==2)
				printf("SwScaler: using 1-tap %s \"scaler\" for vertical luminance scaling (BGR)\n"
				       "SwScaler:       2-tap scaler for vertical chrominance scaling (BGR)\n", mmx ? "MMX" : "C");
			else if(vLumFilterSize==2 && vChrFilterSize==2)
				printf("SwScaler: using 2-tap linear %s scaler for vertical scaling (BGR)\n", mmx ? "MMX" : "C");
			else
				printf("SwScaler: using n-tap %s scaler for vertical scaling (BGR)\n", mmx ? "MMX" : "C");
		}

		if(dstbpp==24)
			printf("SwScaler: using %s YV12->BGR24 Converter\n",
				mmx2 ? "MMX2" : (mmx ? "MMX" : "C"));
		else
			printf("SwScaler: using %s YV12->BGR%d Converter\n", mmx ? "MMX" : "C", dstbpp);

		printf("SwScaler: %dx%d -> %dx%d\n", srcW, srcH, dstW, dstH);
	}

michael's avatar
michael committed
2422 2423 2424
	lastInLumBuf= -1;
	lastInChrBuf= -1;
  } // if(firstLine)
2425

michael's avatar
michael committed
2426 2427 2428 2429 2430
	for(;dstY < dstH; dstY++){
		unsigned char *dest =dstptr[0]+dststride*dstY;
		unsigned char *uDest=dstptr[1]+(dststride>>1)*(dstY>>1);
		unsigned char *vDest=dstptr[2]+(dststride>>1)*(dstY>>1);
		const int chrDstY= dstbpp==12 ? (dstY>>1) : dstY;
2431

michael's avatar
michael committed
2432 2433 2434 2435
		const int firstLumSrcY= vLumFilterPos[dstY]; //First line needed as input
		const int firstChrSrcY= vChrFilterPos[chrDstY]; //First line needed as input
		const int lastLumSrcY= firstLumSrcY + vLumFilterSize -1; // Last line needed as input
		const int lastChrSrcY= firstChrSrcY + vChrFilterSize -1; // Last line needed as input
2436

michael's avatar
michael committed
2437 2438 2439 2440 2441
		if(sws_flags == SWS_FAST_BILINEAR)
		{
			//handle holes
			if(firstLumSrcY > lastInLumBuf) lastInLumBuf= firstLumSrcY-1;
			if(firstChrSrcY > lastInChrBuf) lastInChrBuf= firstChrSrcY-1;
2442 2443
		}

michael's avatar
michael committed
2444 2445
		ASSERT(firstLumSrcY >= lastInLumBuf - vLumBufSize + 1)
		ASSERT(firstChrSrcY >= lastInChrBuf - vChrBufSize + 1)
2446

michael's avatar
michael committed
2447 2448 2449 2450 2451
		// Do we have enough lines in this slice to output the dstY line
		if(lastLumSrcY < srcSliceY + srcSliceH && lastChrSrcY < ((srcSliceY + srcSliceH)>>1))
		{
			//Do horizontal scaling
			while(lastInLumBuf < lastLumSrcY)
2452
			{
michael's avatar
michael committed
2453 2454 2455 2456 2457 2458 2459 2460 2461 2462 2463 2464 2465 2466 2467 2468 2469
				uint8_t *src= srcptr[0]+(lastInLumBuf + 1 - srcSliceY)*stride[0];
				lumBufIndex++;
				ASSERT(lumBufIndex < 2*vLumBufSize)
				ASSERT(lastInLumBuf + 1 - srcSliceY < srcSliceH)
				ASSERT(lastInLumBuf + 1 - srcSliceY >= 0)
//				printf("%d %d\n", lumBufIndex, vLumBufSize);
				RENAME(hyscale)(lumPixBuf[ lumBufIndex ], dstW, src, srcW, lumXInc);
				lastInLumBuf++;
			}
			while(lastInChrBuf < lastChrSrcY)
			{
				uint8_t *src1= srcptr[1]+(lastInChrBuf + 1 - (srcSliceY>>1))*stride[1];
				uint8_t *src2= srcptr[2]+(lastInChrBuf + 1 - (srcSliceY>>1))*stride[2];
				chrBufIndex++;
				ASSERT(chrBufIndex < 2*vChrBufSize)
				ASSERT(lastInChrBuf + 1 - (srcSliceY>>1) < (srcSliceH>>1))
				ASSERT(lastInChrBuf + 1 - (srcSliceY>>1) >= 0)
michael's avatar
michael committed
2470
				RENAME(hcscale)(chrPixBuf[ chrBufIndex ], chrDstW, src1, src2, (srcW+1)>>1, chrXInc);
michael's avatar
michael committed
2471
				lastInChrBuf++;
2472
			}
michael's avatar
michael committed
2473 2474 2475
			//wrap buf index around to stay inside the ring buffer
			if(lumBufIndex >= vLumBufSize ) lumBufIndex-= vLumBufSize;
			if(chrBufIndex >= vChrBufSize ) chrBufIndex-= vChrBufSize;
2476
		}
michael's avatar
michael committed
2477
		else // not enough lines left in this slice -> load the rest in the buffer
michael's avatar
michael committed
2478
		{
michael's avatar
michael committed
2479 2480 2481 2482 2483 2484 2485 2486 2487 2488 2489 2490 2491 2492 2493 2494 2495 2496 2497 2498 2499 2500 2501 2502
/*		printf("%d %d Last:%d %d LastInBuf:%d %d Index:%d %d Y:%d FSize: %d %d BSize: %d %d\n",
			firstChrSrcY,firstLumSrcY,lastChrSrcY,lastLumSrcY,
			lastInChrBuf,lastInLumBuf,chrBufIndex,lumBufIndex,dstY,vChrFilterSize,vLumFilterSize,
			vChrBufSize, vLumBufSize);
*/
			//Do horizontal scaling
			while(lastInLumBuf+1 < srcSliceY + srcSliceH)
			{
				uint8_t *src= srcptr[0]+(lastInLumBuf + 1 - srcSliceY)*stride[0];
				lumBufIndex++;
				ASSERT(lumBufIndex < 2*vLumBufSize)
				ASSERT(lastInLumBuf + 1 - srcSliceY < srcSliceH)
				ASSERT(lastInLumBuf + 1 - srcSliceY >= 0)
				RENAME(hyscale)(lumPixBuf[ lumBufIndex ], dstW, src, srcW, lumXInc);
				lastInLumBuf++;
			}
			while(lastInChrBuf+1 < ((srcSliceY + srcSliceH)>>1))
			{
				uint8_t *src1= srcptr[1]+(lastInChrBuf + 1 - (srcSliceY>>1))*stride[1];
				uint8_t *src2= srcptr[2]+(lastInChrBuf + 1 - (srcSliceY>>1))*stride[2];
				chrBufIndex++;
				ASSERT(chrBufIndex < 2*vChrBufSize)
				ASSERT(lastInChrBuf + 1 - (srcSliceY>>1) < (srcSliceH>>1))
				ASSERT(lastInChrBuf + 1 - (srcSliceY>>1) >= 0)
michael's avatar
michael committed
2503
				RENAME(hcscale)(chrPixBuf[ chrBufIndex ], chrDstW, src1, src2, (srcW+1)>>1, chrXInc);
michael's avatar
michael committed
2504 2505 2506 2507 2508 2509
				lastInChrBuf++;
			}
			//wrap buf index around to stay inside the ring buffer
			if(lumBufIndex >= vLumBufSize ) lumBufIndex-= vLumBufSize;
			if(chrBufIndex >= vChrBufSize ) chrBufIndex-= vChrBufSize;
			break; //we cant output a dstY line so lets try with the next slice
michael's avatar
michael committed
2510
		}
2511

michael's avatar
michael committed
2512 2513 2514 2515 2516 2517
#ifdef HAVE_MMX
		b5Dither= dither8[dstY&1];
		g6Dither= dither4[dstY&1];
		g5Dither= dither8[dstY&1];
		r5Dither= dither8[(dstY+1)&1];
#endif
michael's avatar
michael committed
2518 2519
	    if(dstY < dstH-2 || over<=0)
	    {
michael's avatar
michael committed
2520
		if(dstbpp==12) //YV12
2521
		{
michael's avatar
michael committed
2522 2523
			if(dstY&1) uDest=vDest= NULL; //FIXME split functions in lumi / chromi
			if(vLumFilterSize == 1 && vChrFilterSize == 1) // Unscaled YV12
michael's avatar
michael committed
2524
			{
michael's avatar
michael committed
2525 2526 2527 2528 2529 2530 2531 2532 2533 2534 2535 2536 2537
				int16_t *lumBuf = lumPixBuf[0];
				int16_t *chrBuf= chrPixBuf[0];
				RENAME(yuv2yuv1)(lumBuf, chrBuf, dest, uDest, vDest, dstW);
			}
			else //General YV12
			{
				int16_t **lumSrcPtr= lumPixBuf + lumBufIndex + firstLumSrcY - lastInLumBuf + vLumBufSize;
				int16_t **chrSrcPtr= chrPixBuf + chrBufIndex + firstChrSrcY - lastInChrBuf + vChrBufSize;
				RENAME(yuv2yuvX)(
					vLumFilter+dstY*vLumFilterSize     , lumSrcPtr, vLumFilterSize,
					vChrFilter+(dstY>>1)*vChrFilterSize, chrSrcPtr, vChrFilterSize,
					dest, uDest, vDest, dstW,
					lumMmxFilter+dstY*vLumFilterSize*4, chrMmxFilter+(dstY>>1)*vChrFilterSize*4);
michael's avatar
michael committed
2538
			}
2539
		}
michael's avatar
michael committed
2540
		else
michael's avatar
michael committed
2541
		{
michael's avatar
michael committed
2542 2543
			int16_t **lumSrcPtr= lumPixBuf + lumBufIndex + firstLumSrcY - lastInLumBuf + vLumBufSize;
			int16_t **chrSrcPtr= chrPixBuf + chrBufIndex + firstChrSrcY - lastInChrBuf + vChrBufSize;
2544

michael's avatar
michael committed
2545 2546 2547 2548 2549
			ASSERT(lumSrcPtr + vLumFilterSize - 1 < lumPixBuf + vLumBufSize*2);
			ASSERT(chrSrcPtr + vChrFilterSize - 1 < chrPixBuf + vChrBufSize*2);
			if(vLumFilterSize == 1 && vChrFilterSize == 2) //Unscaled RGB
			{
				int chrAlpha= vChrFilter[2*dstY+1];
michael's avatar
michael committed
2550

michael's avatar
michael committed
2551 2552 2553 2554 2555 2556 2557 2558 2559 2560 2561 2562 2563 2564 2565 2566 2567 2568 2569 2570
				RENAME(yuv2rgb1)(*lumSrcPtr, *chrSrcPtr, *(chrSrcPtr+1),
						 dest, dstW, chrAlpha, dstbpp);
			}
			else if(vLumFilterSize == 2 && vChrFilterSize == 2) //BiLinear Upscale RGB
			{
				int lumAlpha= vLumFilter[2*dstY+1];
				int chrAlpha= vChrFilter[2*dstY+1];

				RENAME(yuv2rgb2)(*lumSrcPtr, *(lumSrcPtr+1), *chrSrcPtr, *(chrSrcPtr+1),
						 dest, dstW, lumAlpha, chrAlpha, dstbpp);
			}
			else //General RGB
			{
				RENAME(yuv2rgbX)(
					vLumFilter+dstY*vLumFilterSize, lumSrcPtr, vLumFilterSize,
					vChrFilter+dstY*vChrFilterSize, chrSrcPtr, vChrFilterSize,
					dest, dstW, dstbpp,
					lumMmxFilter+dstY*vLumFilterSize*4, chrMmxFilter+dstY*vChrFilterSize*4);
			}
		}
michael's avatar
michael committed
2571 2572 2573 2574 2575 2576 2577 2578 2579 2580 2581 2582 2583 2584 2585 2586 2587 2588 2589 2590 2591 2592 2593
            }
	    else // hmm looks like we cant use MMX here without overwriting this arrays tail
	    {
		int16_t **lumSrcPtr= lumPixBuf + lumBufIndex + firstLumSrcY - lastInLumBuf + vLumBufSize;
		int16_t **chrSrcPtr= chrPixBuf + chrBufIndex + firstChrSrcY - lastInChrBuf + vChrBufSize;
		if(dstbpp==12) //YV12
		{
			if(dstY&1) uDest=vDest= NULL; //FIXME split functions in lumi / chromi
			yuv2yuvXinC(
				vLumFilter+dstY*vLumFilterSize     , lumSrcPtr, vLumFilterSize,
				vChrFilter+(dstY>>1)*vChrFilterSize, chrSrcPtr, vChrFilterSize,
				dest, uDest, vDest, dstW);
		}
		else
		{
			ASSERT(lumSrcPtr + vLumFilterSize - 1 < lumPixBuf + vLumBufSize*2);
			ASSERT(chrSrcPtr + vChrFilterSize - 1 < chrPixBuf + vChrBufSize*2);
			yuv2rgbXinC(
				vLumFilter+dstY*vLumFilterSize, lumSrcPtr, vLumFilterSize,
				vChrFilter+dstY*vChrFilterSize, chrSrcPtr, vChrFilterSize,
				dest, dstW, dstbpp);
		}
	    }
michael's avatar
michael committed
2594
	}
2595 2596 2597

#ifdef HAVE_MMX
	__asm __volatile(SFENCE:::"memory");
michael's avatar
michael committed
2598
	__asm __volatile(EMMS:::"memory");
2599
#endif
michael's avatar
michael committed
2600 2601
	firstTime=0;
}