39 #define OP_AVG(S,D) "pavgb " #S ", " #D " \n\t"
42 #define NORMALIZE_MMX(SHIFT) \
43 "paddw %%mm7, %%mm3 \n\t" \
44 "paddw %%mm7, %%mm4 \n\t" \
45 "psraw "SHIFT", %%mm3 \n\t" \
46 "psraw "SHIFT", %%mm4 \n\t"
48 #define TRANSFER_DO_PACK(OP) \
49 "packuswb %%mm4, %%mm3 \n\t" \
51 "movq %%mm3, (%2) \n\t"
53 #define TRANSFER_DONT_PACK(OP) \
56 "movq %%mm3, 0(%2) \n\t" \
57 "movq %%mm4, 8(%2) \n\t"
60 #define DO_UNPACK(reg) "punpcklbw %%mm0, " reg "\n\t"
61 #define DONT_UNPACK(reg)
64 #define LOAD_ROUNDER_MMX(ROUND) \
65 "movd "ROUND", %%mm7 \n\t" \
66 "punpcklwd %%mm7, %%mm7 \n\t" \
67 "punpckldq %%mm7, %%mm7 \n\t"
69 #define SHIFT2_LINE(OFF, R0,R1,R2,R3) \
70 "paddw %%mm"#R2", %%mm"#R1" \n\t" \
71 "movd (%0,%3), %%mm"#R0" \n\t" \
72 "pmullw %%mm6, %%mm"#R1" \n\t" \
73 "punpcklbw %%mm0, %%mm"#R0" \n\t" \
74 "movd (%0,%2), %%mm"#R3" \n\t" \
75 "psubw %%mm"#R0", %%mm"#R1" \n\t" \
76 "punpcklbw %%mm0, %%mm"#R3" \n\t" \
77 "paddw %%mm7, %%mm"#R1" \n\t" \
78 "psubw %%mm"#R3", %%mm"#R1" \n\t" \
79 "psraw %4, %%mm"#R1" \n\t" \
80 "movq %%mm"#R1", "#OFF"(%1) \n\t" \
84 static void vc1_put_ver_16b_shift2_mmx(int16_t *
dst,
86 int rnd, int64_t
shift)
89 "mov $3, %%"REG_c
" \n\t"
90 LOAD_ROUNDER_MMX(
"%5")
93 "movd (%0), %%mm2 \n\t"
95 "movd (%0), %%mm3 \n\t"
96 "punpcklbw %%mm0, %%mm2 \n\t"
97 "punpcklbw %%mm0, %%mm3 \n\t"
98 SHIFT2_LINE( 0, 1, 2, 3, 4)
99 SHIFT2_LINE( 24, 2, 3, 4, 1)
100 SHIFT2_LINE( 48, 3, 4, 1, 2)
101 SHIFT2_LINE( 72, 4, 1, 2, 3)
102 SHIFT2_LINE( 96, 1, 2, 3, 4)
103 SHIFT2_LINE(120, 2, 3, 4, 1)
104 SHIFT2_LINE(144, 3, 4, 1, 2)
105 SHIFT2_LINE(168, 4, 1, 2, 3)
110 : "+
r"(src), "+
r"(dst)
111 : "
r"(stride), "
r"(-2*stride),
112 "
m"(shift), "
m"(rnd), "
r"(9*stride-4)
121 #define VC1_HOR_16b_SHIFT2(OP, OPNAME)\
122 static void OPNAME ## vc1_hor_16b_shift2_mmx(uint8_t *dst, x86_reg stride,\
123 const int16_t *src, int rnd)\
128 rnd -= (-1+9+9-1)*1024; \
130 LOAD_ROUNDER_MMX("%4")\
131 "movq "MANGLE(ff_pw_128)", %%mm6\n\t"\
132 "movq "MANGLE(ff_pw_9)", %%mm5 \n\t"\
134 "movq 2*0+0(%1), %%mm1 \n\t"\
135 "movq 2*0+8(%1), %%mm2 \n\t"\
136 "movq 2*1+0(%1), %%mm3 \n\t"\
137 "movq 2*1+8(%1), %%mm4 \n\t"\
138 "paddw 2*3+0(%1), %%mm1 \n\t"\
139 "paddw 2*3+8(%1), %%mm2 \n\t"\
140 "paddw 2*2+0(%1), %%mm3 \n\t"\
141 "paddw 2*2+8(%1), %%mm4 \n\t"\
142 "pmullw %%mm5, %%mm3 \n\t"\
143 "pmullw %%mm5, %%mm4 \n\t"\
144 "psubw %%mm1, %%mm3 \n\t"\
145 "psubw %%mm2, %%mm4 \n\t"\
148 "paddw %%mm6, %%mm3 \n\t"\
149 "paddw %%mm6, %%mm4 \n\t"\
150 TRANSFER_DO_PACK(OP)\
155 : "+r"(h), "+r" (src), "+r" (dst)\
156 : "r"(stride), "m"(rnd)\
161 VC1_HOR_16b_SHIFT2(
OP_PUT, put_)
162 VC1_HOR_16b_SHIFT2(
OP_AVG, avg_)
169 #define VC1_SHIFT2(OP, OPNAME)\
170 static void OPNAME ## vc1_shift2_mmx(uint8_t *dst, const uint8_t *src,\
171 x86_reg stride, int rnd, x86_reg offset)\
175 "mov $8, %%"REG_c" \n\t"\
176 LOAD_ROUNDER_MMX("%5")\
177 "movq "MANGLE(ff_pw_9)", %%mm6\n\t"\
179 "movd 0(%0 ), %%mm3 \n\t"\
180 "movd 4(%0 ), %%mm4 \n\t"\
181 "movd 0(%0,%2), %%mm1 \n\t"\
182 "movd 4(%0,%2), %%mm2 \n\t"\
184 "punpcklbw %%mm0, %%mm3 \n\t"\
185 "punpcklbw %%mm0, %%mm4 \n\t"\
186 "punpcklbw %%mm0, %%mm1 \n\t"\
187 "punpcklbw %%mm0, %%mm2 \n\t"\
188 "paddw %%mm1, %%mm3 \n\t"\
189 "paddw %%mm2, %%mm4 \n\t"\
190 "movd 0(%0,%3), %%mm1 \n\t"\
191 "movd 4(%0,%3), %%mm2 \n\t"\
192 "pmullw %%mm6, %%mm3 \n\t" \
193 "pmullw %%mm6, %%mm4 \n\t" \
194 "punpcklbw %%mm0, %%mm1 \n\t"\
195 "punpcklbw %%mm0, %%mm2 \n\t"\
196 "psubw %%mm1, %%mm3 \n\t" \
197 "psubw %%mm2, %%mm4 \n\t" \
198 "movd 0(%0,%2), %%mm1 \n\t"\
199 "movd 4(%0,%2), %%mm2 \n\t"\
200 "punpcklbw %%mm0, %%mm1 \n\t"\
201 "punpcklbw %%mm0, %%mm2 \n\t"\
202 "psubw %%mm1, %%mm3 \n\t" \
203 "psubw %%mm2, %%mm4 \n\t" \
205 "packuswb %%mm4, %%mm3 \n\t"\
207 "movq %%mm3, (%1) \n\t"\
210 "dec %%"REG_c" \n\t"\
212 : "+r"(src), "+r"(dst)\
213 : "r"(offset), "r"(-2*offset), "g"(stride), "m"(rnd),\
215 : "%"REG_c, "memory"\
220 VC1_SHIFT2(OP_AVG, avg_)
232 #define MSPEL_FILTER13_CORE(UNPACK, MOVQ, A1, A2, A3, A4) \
233 MOVQ "*0+"A1", %%mm1 \n\t" \
234 MOVQ "*4+"A1", %%mm2 \n\t" \
237 "pmullw "MANGLE(ff_pw_3)", %%mm1\n\t" \
238 "pmullw "MANGLE(ff_pw_3)", %%mm2\n\t" \
239 MOVQ "*0+"A2", %%mm3 \n\t" \
240 MOVQ "*4+"A2", %%mm4 \n\t" \
243 "pmullw %%mm6, %%mm3 \n\t" \
244 "pmullw %%mm6, %%mm4 \n\t" \
245 "psubw %%mm1, %%mm3 \n\t" \
246 "psubw %%mm2, %%mm4 \n\t" \
247 MOVQ "*0+"A4", %%mm1 \n\t" \
248 MOVQ "*4+"A4", %%mm2 \n\t" \
251 "psllw $2, %%mm1 \n\t" \
252 "psllw $2, %%mm2 \n\t" \
253 "psubw %%mm1, %%mm3 \n\t" \
254 "psubw %%mm2, %%mm4 \n\t" \
255 MOVQ "*0+"A3", %%mm1 \n\t" \
256 MOVQ "*4+"A3", %%mm2 \n\t" \
259 "pmullw %%mm5, %%mm1 \n\t" \
260 "pmullw %%mm5, %%mm2 \n\t" \
261 "paddw %%mm1, %%mm3 \n\t" \
262 "paddw %%mm2, %%mm4 \n\t"
272 #define MSPEL_FILTER13_VER_16B(NAME, A1, A2, A3, A4) \
274 vc1_put_ver_16b_ ## NAME ## _mmx(int16_t *dst, const uint8_t *src, \
275 x86_reg src_stride, \
276 int rnd, int64_t shift) \
281 LOAD_ROUNDER_MMX("%5") \
282 "movq "MANGLE(ff_pw_53)", %%mm5\n\t" \
283 "movq "MANGLE(ff_pw_18)", %%mm6\n\t" \
286 MSPEL_FILTER13_CORE(DO_UNPACK, "movd 1", A1, A2, A3, A4) \
287 NORMALIZE_MMX("%6") \
288 TRANSFER_DONT_PACK(OP_PUT) \
290 "movd 8+"A1", %%mm1 \n\t" \
292 "movq %%mm1, %%mm3 \n\t" \
293 "paddw %%mm1, %%mm1 \n\t" \
294 "paddw %%mm3, %%mm1 \n\t" \
295 "movd 8+"A2", %%mm3 \n\t" \
297 "pmullw %%mm6, %%mm3 \n\t" \
298 "psubw %%mm1, %%mm3 \n\t" \
299 "movd 8+"A3", %%mm1 \n\t" \
301 "pmullw %%mm5, %%mm1 \n\t" \
302 "paddw %%mm1, %%mm3 \n\t" \
303 "movd 8+"A4", %%mm1 \n\t" \
305 "psllw $2, %%mm1 \n\t" \
306 "psubw %%mm1, %%mm3 \n\t" \
307 "paddw %%mm7, %%mm3 \n\t" \
308 "psraw %6, %%mm3 \n\t" \
309 "movq %%mm3, 16(%2) \n\t" \
314 : "+r"(h), "+r" (src), "+r" (dst) \
315 : "r"(src_stride), "r"(3*src_stride), \
316 "m"(rnd), "m"(shift) \
328 #define MSPEL_FILTER13_HOR_16B(NAME, A1, A2, A3, A4, OP, OPNAME) \
330 OPNAME ## vc1_hor_16b_ ## NAME ## _mmx(uint8_t *dst, x86_reg stride, \
331 const int16_t *src, int rnd) \
335 rnd -= (-4+58+13-3)*256; \
337 LOAD_ROUNDER_MMX("%4") \
338 "movq "MANGLE(ff_pw_18)", %%mm6 \n\t" \
339 "movq "MANGLE(ff_pw_53)", %%mm5 \n\t" \
342 MSPEL_FILTER13_CORE(DONT_UNPACK, "movq 2", A1, A2, A3, A4) \
343 NORMALIZE_MMX("$7") \
345 "paddw "MANGLE(ff_pw_128)", %%mm3 \n\t" \
346 "paddw "MANGLE(ff_pw_128)", %%mm4 \n\t" \
347 TRANSFER_DO_PACK(OP) \
352 : "+r"(h), "+r" (src), "+r" (dst) \
353 : "r"(stride), "m"(rnd) \
366 #define MSPEL_FILTER13_8B(NAME, A1, A2, A3, A4, OP, OPNAME) \
368 OPNAME ## vc1_## NAME ## _mmx(uint8_t *dst, const uint8_t *src, \
369 x86_reg stride, int rnd, x86_reg offset) \
375 LOAD_ROUNDER_MMX("%6") \
376 "movq "MANGLE(ff_pw_53)", %%mm5 \n\t" \
377 "movq "MANGLE(ff_pw_18)", %%mm6 \n\t" \
380 MSPEL_FILTER13_CORE(DO_UNPACK, "movd 1", A1, A2, A3, A4) \
381 NORMALIZE_MMX("$6") \
382 TRANSFER_DO_PACK(OP) \
387 : "+r"(h), "+r" (src), "+r" (dst) \
388 : "r"(offset), "r"(3*offset), "g"(stride), "m"(rnd) \
394 MSPEL_FILTER13_8B (
shift1,
"0(%1,%4 )",
"0(%1,%3,2)",
"0(%1,%3 )",
"0(%1 )",
OP_PUT, put_)
395 MSPEL_FILTER13_8B (
shift1, "0(%1,%4 )", "0(%1,%3,2)", "0(%1,%3 )", "0(%1 )", OP_AVG, avg_)
396 MSPEL_FILTER13_VER_16B(shift1, "0(%1,%4 )", "0(%1,%3,2)", "0(%1,%3 )", "0(%1 )")
397 MSPEL_FILTER13_HOR_16B(shift1, "2*3(%1)", "2*2(%1)", "2*1(%1)", "2*0(%1)",
OP_PUT, put_)
398 MSPEL_FILTER13_HOR_16B(shift1, "2*3(%1)", "2*2(%1)", "2*1(%1)", "2*0(%1)", OP_AVG, avg_)
401 MSPEL_FILTER13_8B (shift3, "0(%1 )", "0(%1,%3 )", "0(%1,%3,2)", "0(%1,%4 )", OP_PUT, put_)
402 MSPEL_FILTER13_8B (shift3, "0(%1 )", "0(%1,%3 )", "0(%1,%3,2)", "0(%1,%4 )", OP_AVG, avg_)
403 MSPEL_FILTER13_VER_16B(shift3, "0(%1 )", "0(%1,%3 )", "0(%1,%3,2)", "0(%1,%4 )")
404 MSPEL_FILTER13_HOR_16B(shift3, "2*0(%1)", "2*1(%1)", "2*2(%1)", "2*3(%1)", OP_PUT, put_)
405 MSPEL_FILTER13_HOR_16B(shift3, "2*0(%1)", "2*1(%1)", "2*2(%1)", "2*3(%1)", OP_AVG, avg_)
407 typedef
void (*vc1_mspel_mc_filter_ver_16bits)(int16_t *dst, const
uint8_t *src,
x86_reg src_stride,
int rnd, int64_t shift);
408 typedef
void (*vc1_mspel_mc_filter_hor_16bits)(
uint8_t *dst,
x86_reg dst_stride, const int16_t *src,
int rnd);
422 #define VC1_MSPEL_MC(OP)\
423 static void OP ## vc1_mspel_mc(uint8_t *dst, const uint8_t *src, int stride,\
424 int hmode, int vmode, int rnd)\
426 static const vc1_mspel_mc_filter_ver_16bits vc1_put_shift_ver_16bits[] =\
427 { NULL, vc1_put_ver_16b_shift1_mmx, vc1_put_ver_16b_shift2_mmx, vc1_put_ver_16b_shift3_mmx };\
428 static const vc1_mspel_mc_filter_hor_16bits vc1_put_shift_hor_16bits[] =\
429 { NULL, OP ## vc1_hor_16b_shift1_mmx, OP ## vc1_hor_16b_shift2_mmx, OP ## vc1_hor_16b_shift3_mmx };\
430 static const vc1_mspel_mc_filter_8bits vc1_put_shift_8bits[] =\
431 { NULL, OP ## vc1_shift1_mmx, OP ## vc1_shift2_mmx, OP ## vc1_shift3_mmx };\
434 "pxor %%mm0, %%mm0 \n\t"\
440 static const int shift_value[] = { 0, 5, 1, 5 };\
441 int shift = (shift_value[hmode]+shift_value[vmode])>>1;\
443 DECLARE_ALIGNED(16, int16_t, tmp)[12*8];\
445 r = (1<<(shift-1)) + rnd-1;\
446 vc1_put_shift_ver_16bits[vmode](tmp, src-1, stride, r, shift);\
448 vc1_put_shift_hor_16bits[hmode](dst, stride, tmp+1, 64-rnd);\
452 vc1_put_shift_8bits[vmode](dst, src, stride, 1-rnd, stride);\
458 vc1_put_shift_8bits[hmode](dst, src, stride, rnd, 1);\
465 #define DECLARE_FUNCTION(a, b) \
466 static void put_vc1_mspel_mc ## a ## b ## _mmx(uint8_t *dst, const uint8_t *src, int stride, int rnd) { \
467 put_vc1_mspel_mc(dst, src, stride, a, b, rnd); \
469 static void avg_vc1_mspel_mc ## a ## b ## _mmxext(uint8_t *dst, \
470 const uint8_t *src, \
471 int stride, int rnd) \
473 avg_vc1_mspel_mc(dst, src, stride, a, b, rnd); \
476 DECLARE_FUNCTION(0, 1)
477 DECLARE_FUNCTION(0, 2)
478 DECLARE_FUNCTION(0, 3)
480 DECLARE_FUNCTION(1, 0)
481 DECLARE_FUNCTION(1, 1)
482 DECLARE_FUNCTION(1, 2)
483 DECLARE_FUNCTION(1, 3)
485 DECLARE_FUNCTION(2, 0)
486 DECLARE_FUNCTION(2, 1)
487 DECLARE_FUNCTION(2, 2)
488 DECLARE_FUNCTION(2, 3)
490 DECLARE_FUNCTION(3, 0)
491 DECLARE_FUNCTION(3, 1)
492 DECLARE_FUNCTION(3, 2)
493 DECLARE_FUNCTION(3, 3)
495 static
void vc1_inv_trans_4x4_dc_mmxext(
uint8_t *dest,
int linesize,
499 dc = (17 * dc + 4) >> 3;
500 dc = (17 * dc + 64) >> 7;
502 "movd %0, %%mm0 \n\t"
503 "pshufw $0, %%mm0, %%mm0 \n\t"
504 "pxor %%mm1, %%mm1 \n\t"
505 "psubw %%mm0, %%mm1 \n\t"
506 "packuswb %%mm0, %%mm0 \n\t"
507 "packuswb %%mm1, %%mm1 \n\t"
511 "movd %0, %%mm2 \n\t"
512 "movd %1, %%mm3 \n\t"
513 "movd %2, %%mm4 \n\t"
514 "movd %3, %%mm5 \n\t"
515 "paddusb %%mm0, %%mm2 \n\t"
516 "paddusb %%mm0, %%mm3 \n\t"
517 "paddusb %%mm0, %%mm4 \n\t"
518 "paddusb %%mm0, %%mm5 \n\t"
519 "psubusb %%mm1, %%mm2 \n\t"
520 "psubusb %%mm1, %%mm3 \n\t"
521 "psubusb %%mm1, %%mm4 \n\t"
522 "psubusb %%mm1, %%mm5 \n\t"
523 "movd %%mm2, %0 \n\t"
524 "movd %%mm3, %1 \n\t"
525 "movd %%mm4, %2 \n\t"
526 "movd %%mm5, %3 \n\t"
527 :
"+m"(*(uint32_t*)(dest+0*linesize)),
528 "+m"(*(uint32_t*)(dest+1*linesize)),
529 "+m"(*(uint32_t*)(dest+2*linesize)),
530 "+m"(*(uint32_t*)(dest+3*linesize))
534 static void vc1_inv_trans_4x8_dc_mmxext(
uint8_t *dest,
int linesize,
538 dc = (17 * dc + 4) >> 3;
539 dc = (12 * dc + 64) >> 7;
541 "movd %0, %%mm0 \n\t"
542 "pshufw $0, %%mm0, %%mm0 \n\t"
543 "pxor %%mm1, %%mm1 \n\t"
544 "psubw %%mm0, %%mm1 \n\t"
545 "packuswb %%mm0, %%mm0 \n\t"
546 "packuswb %%mm1, %%mm1 \n\t"
550 "movd %0, %%mm2 \n\t"
551 "movd %1, %%mm3 \n\t"
552 "movd %2, %%mm4 \n\t"
553 "movd %3, %%mm5 \n\t"
554 "paddusb %%mm0, %%mm2 \n\t"
555 "paddusb %%mm0, %%mm3 \n\t"
556 "paddusb %%mm0, %%mm4 \n\t"
557 "paddusb %%mm0, %%mm5 \n\t"
558 "psubusb %%mm1, %%mm2 \n\t"
559 "psubusb %%mm1, %%mm3 \n\t"
560 "psubusb %%mm1, %%mm4 \n\t"
561 "psubusb %%mm1, %%mm5 \n\t"
562 "movd %%mm2, %0 \n\t"
563 "movd %%mm3, %1 \n\t"
564 "movd %%mm4, %2 \n\t"
565 "movd %%mm5, %3 \n\t"
566 :
"+m"(*(uint32_t*)(dest+0*linesize)),
567 "+m"(*(uint32_t*)(dest+1*linesize)),
568 "+m"(*(uint32_t*)(dest+2*linesize)),
569 "+m"(*(uint32_t*)(dest+3*linesize))
573 "movd %0, %%mm2 \n\t"
574 "movd %1, %%mm3 \n\t"
575 "movd %2, %%mm4 \n\t"
576 "movd %3, %%mm5 \n\t"
577 "paddusb %%mm0, %%mm2 \n\t"
578 "paddusb %%mm0, %%mm3 \n\t"
579 "paddusb %%mm0, %%mm4 \n\t"
580 "paddusb %%mm0, %%mm5 \n\t"
581 "psubusb %%mm1, %%mm2 \n\t"
582 "psubusb %%mm1, %%mm3 \n\t"
583 "psubusb %%mm1, %%mm4 \n\t"
584 "psubusb %%mm1, %%mm5 \n\t"
585 "movd %%mm2, %0 \n\t"
586 "movd %%mm3, %1 \n\t"
587 "movd %%mm4, %2 \n\t"
588 "movd %%mm5, %3 \n\t"
589 :
"+m"(*(uint32_t*)(dest+0*linesize)),
590 "+m"(*(uint32_t*)(dest+1*linesize)),
591 "+m"(*(uint32_t*)(dest+2*linesize)),
592 "+m"(*(uint32_t*)(dest+3*linesize))
596 static void vc1_inv_trans_8x4_dc_mmxext(
uint8_t *dest,
int linesize,
600 dc = ( 3 * dc + 1) >> 1;
601 dc = (17 * dc + 64) >> 7;
603 "movd %0, %%mm0 \n\t"
604 "pshufw $0, %%mm0, %%mm0 \n\t"
605 "pxor %%mm1, %%mm1 \n\t"
606 "psubw %%mm0, %%mm1 \n\t"
607 "packuswb %%mm0, %%mm0 \n\t"
608 "packuswb %%mm1, %%mm1 \n\t"
612 "movq %0, %%mm2 \n\t"
613 "movq %1, %%mm3 \n\t"
614 "movq %2, %%mm4 \n\t"
615 "movq %3, %%mm5 \n\t"
616 "paddusb %%mm0, %%mm2 \n\t"
617 "paddusb %%mm0, %%mm3 \n\t"
618 "paddusb %%mm0, %%mm4 \n\t"
619 "paddusb %%mm0, %%mm5 \n\t"
620 "psubusb %%mm1, %%mm2 \n\t"
621 "psubusb %%mm1, %%mm3 \n\t"
622 "psubusb %%mm1, %%mm4 \n\t"
623 "psubusb %%mm1, %%mm5 \n\t"
624 "movq %%mm2, %0 \n\t"
625 "movq %%mm3, %1 \n\t"
626 "movq %%mm4, %2 \n\t"
627 "movq %%mm5, %3 \n\t"
628 :
"+m"(*(uint32_t*)(dest+0*linesize)),
629 "+m"(*(uint32_t*)(dest+1*linesize)),
630 "+m"(*(uint32_t*)(dest+2*linesize)),
631 "+m"(*(uint32_t*)(dest+3*linesize))
635 static void vc1_inv_trans_8x8_dc_mmxext(
uint8_t *dest,
int linesize,
639 dc = (3 * dc + 1) >> 1;
640 dc = (3 * dc + 16) >> 5;
642 "movd %0, %%mm0 \n\t"
643 "pshufw $0, %%mm0, %%mm0 \n\t"
644 "pxor %%mm1, %%mm1 \n\t"
645 "psubw %%mm0, %%mm1 \n\t"
646 "packuswb %%mm0, %%mm0 \n\t"
647 "packuswb %%mm1, %%mm1 \n\t"
651 "movq %0, %%mm2 \n\t"
652 "movq %1, %%mm3 \n\t"
653 "movq %2, %%mm4 \n\t"
654 "movq %3, %%mm5 \n\t"
655 "paddusb %%mm0, %%mm2 \n\t"
656 "paddusb %%mm0, %%mm3 \n\t"
657 "paddusb %%mm0, %%mm4 \n\t"
658 "paddusb %%mm0, %%mm5 \n\t"
659 "psubusb %%mm1, %%mm2 \n\t"
660 "psubusb %%mm1, %%mm3 \n\t"
661 "psubusb %%mm1, %%mm4 \n\t"
662 "psubusb %%mm1, %%mm5 \n\t"
663 "movq %%mm2, %0 \n\t"
664 "movq %%mm3, %1 \n\t"
665 "movq %%mm4, %2 \n\t"
666 "movq %%mm5, %3 \n\t"
667 :
"+m"(*(uint32_t*)(dest+0*linesize)),
668 "+m"(*(uint32_t*)(dest+1*linesize)),
669 "+m"(*(uint32_t*)(dest+2*linesize)),
670 "+m"(*(uint32_t*)(dest+3*linesize))
674 "movq %0, %%mm2 \n\t"
675 "movq %1, %%mm3 \n\t"
676 "movq %2, %%mm4 \n\t"
677 "movq %3, %%mm5 \n\t"
678 "paddusb %%mm0, %%mm2 \n\t"
679 "paddusb %%mm0, %%mm3 \n\t"
680 "paddusb %%mm0, %%mm4 \n\t"
681 "paddusb %%mm0, %%mm5 \n\t"
682 "psubusb %%mm1, %%mm2 \n\t"
683 "psubusb %%mm1, %%mm3 \n\t"
684 "psubusb %%mm1, %%mm4 \n\t"
685 "psubusb %%mm1, %%mm5 \n\t"
686 "movq %%mm2, %0 \n\t"
687 "movq %%mm3, %1 \n\t"
688 "movq %%mm4, %2 \n\t"
689 "movq %%mm5, %3 \n\t"
690 :
"+m"(*(uint32_t*)(dest+0*linesize)),
691 "+m"(*(uint32_t*)(dest+1*linesize)),
692 "+m"(*(uint32_t*)(dest+2*linesize)),
693 "+m"(*(uint32_t*)(dest+3*linesize))