40 #define OP_AVG(S,D) "pavgb " #S ", " #D " \n\t"
43 #define NORMALIZE_MMX(SHIFT) \
44 "paddw %%mm7, %%mm3 \n\t" \
45 "paddw %%mm7, %%mm4 \n\t" \
46 "psraw "SHIFT", %%mm3 \n\t" \
47 "psraw "SHIFT", %%mm4 \n\t"
49 #define TRANSFER_DO_PACK(OP) \
50 "packuswb %%mm4, %%mm3 \n\t" \
52 "movq %%mm3, (%2) \n\t"
54 #define TRANSFER_DONT_PACK(OP) \
57 "movq %%mm3, 0(%2) \n\t" \
58 "movq %%mm4, 8(%2) \n\t"
61 #define DO_UNPACK(reg) "punpcklbw %%mm0, " reg "\n\t"
62 #define DONT_UNPACK(reg)
65 #define LOAD_ROUNDER_MMX(ROUND) \
66 "movd "ROUND", %%mm7 \n\t" \
67 "punpcklwd %%mm7, %%mm7 \n\t" \
68 "punpckldq %%mm7, %%mm7 \n\t"
70 #define SHIFT2_LINE(OFF, R0,R1,R2,R3) \
71 "paddw %%mm"#R2", %%mm"#R1" \n\t" \
72 "movd (%0,%3), %%mm"#R0" \n\t" \
73 "pmullw %%mm6, %%mm"#R1" \n\t" \
74 "punpcklbw %%mm0, %%mm"#R0" \n\t" \
75 "movd (%0,%2), %%mm"#R3" \n\t" \
76 "psubw %%mm"#R0", %%mm"#R1" \n\t" \
77 "punpcklbw %%mm0, %%mm"#R3" \n\t" \
78 "paddw %%mm7, %%mm"#R1" \n\t" \
79 "psubw %%mm"#R3", %%mm"#R1" \n\t" \
80 "psraw %4, %%mm"#R1" \n\t" \
81 "movq %%mm"#R1", "#OFF"(%1) \n\t" \
85 static void vc1_put_ver_16b_shift2_mmx(int16_t *dst,
87 int rnd, int64_t shift)
90 "mov $3, %%"REG_c
" \n\t"
91 LOAD_ROUNDER_MMX(
"%5")
94 "movd (%0), %%mm2 \n\t"
96 "movd (%0), %%mm3 \n\t"
97 "punpcklbw %%mm0, %%mm2 \n\t"
98 "punpcklbw %%mm0, %%mm3 \n\t"
99 SHIFT2_LINE( 0, 1, 2, 3, 4)
100 SHIFT2_LINE( 24, 2, 3, 4, 1)
101 SHIFT2_LINE( 48, 3, 4, 1, 2)
102 SHIFT2_LINE( 72, 4, 1, 2, 3)
103 SHIFT2_LINE( 96, 1, 2, 3, 4)
104 SHIFT2_LINE(120, 2, 3, 4, 1)
105 SHIFT2_LINE(144, 3, 4, 1, 2)
106 SHIFT2_LINE(168, 4, 1, 2, 3)
111 : "+
r"(src), "+
r"(dst)
112 : "
r"(stride), "
r"(-2*stride),
113 "m"(shift), "m"(rnd), "
r"(9*stride-4)
122 #define VC1_HOR_16b_SHIFT2(OP, OPNAME)\
123 static void OPNAME ## vc1_hor_16b_shift2_mmx(uint8_t *dst, x86_reg stride,\
124 const int16_t *src, int rnd)\
129 rnd -= (-1+9+9-1)*1024; \
131 LOAD_ROUNDER_MMX("%4")\
132 "movq "MANGLE(ff_pw_128)", %%mm6\n\t"\
133 "movq "MANGLE(ff_pw_9)", %%mm5 \n\t"\
135 "movq 2*0+0(%1), %%mm1 \n\t"\
136 "movq 2*0+8(%1), %%mm2 \n\t"\
137 "movq 2*1+0(%1), %%mm3 \n\t"\
138 "movq 2*1+8(%1), %%mm4 \n\t"\
139 "paddw 2*3+0(%1), %%mm1 \n\t"\
140 "paddw 2*3+8(%1), %%mm2 \n\t"\
141 "paddw 2*2+0(%1), %%mm3 \n\t"\
142 "paddw 2*2+8(%1), %%mm4 \n\t"\
143 "pmullw %%mm5, %%mm3 \n\t"\
144 "pmullw %%mm5, %%mm4 \n\t"\
145 "psubw %%mm1, %%mm3 \n\t"\
146 "psubw %%mm2, %%mm4 \n\t"\
149 "paddw %%mm6, %%mm3 \n\t"\
150 "paddw %%mm6, %%mm4 \n\t"\
151 TRANSFER_DO_PACK(OP)\
156 : "+r"(h), "+r" (src), "+r" (dst)\
157 : "r"(stride), "m"(rnd)\
162 VC1_HOR_16b_SHIFT2(
OP_PUT, put_)
163 VC1_HOR_16b_SHIFT2(OP_AVG, avg_)
170 #define VC1_SHIFT2(OP, OPNAME)\
171 static void OPNAME ## vc1_shift2_mmx(uint8_t *dst, const uint8_t *src,\
172 x86_reg stride, int rnd, x86_reg offset)\
176 "mov $8, %%"REG_c" \n\t"\
177 LOAD_ROUNDER_MMX("%5")\
178 "movq "MANGLE(ff_pw_9)", %%mm6\n\t"\
180 "movd 0(%0 ), %%mm3 \n\t"\
181 "movd 4(%0 ), %%mm4 \n\t"\
182 "movd 0(%0,%2), %%mm1 \n\t"\
183 "movd 4(%0,%2), %%mm2 \n\t"\
185 "punpcklbw %%mm0, %%mm3 \n\t"\
186 "punpcklbw %%mm0, %%mm4 \n\t"\
187 "punpcklbw %%mm0, %%mm1 \n\t"\
188 "punpcklbw %%mm0, %%mm2 \n\t"\
189 "paddw %%mm1, %%mm3 \n\t"\
190 "paddw %%mm2, %%mm4 \n\t"\
191 "movd 0(%0,%3), %%mm1 \n\t"\
192 "movd 4(%0,%3), %%mm2 \n\t"\
193 "pmullw %%mm6, %%mm3 \n\t" \
194 "pmullw %%mm6, %%mm4 \n\t" \
195 "punpcklbw %%mm0, %%mm1 \n\t"\
196 "punpcklbw %%mm0, %%mm2 \n\t"\
197 "psubw %%mm1, %%mm3 \n\t" \
198 "psubw %%mm2, %%mm4 \n\t" \
199 "movd 0(%0,%2), %%mm1 \n\t"\
200 "movd 4(%0,%2), %%mm2 \n\t"\
201 "punpcklbw %%mm0, %%mm1 \n\t"\
202 "punpcklbw %%mm0, %%mm2 \n\t"\
203 "psubw %%mm1, %%mm3 \n\t" \
204 "psubw %%mm2, %%mm4 \n\t" \
206 "packuswb %%mm4, %%mm3 \n\t"\
208 "movq %%mm3, (%1) \n\t"\
211 "dec %%"REG_c" \n\t"\
213 : "+r"(src), "+r"(dst)\
214 : "r"(offset), "r"(-2*offset), "g"(stride), "m"(rnd),\
216 : "%"REG_c, "memory"\
221 VC1_SHIFT2(OP_AVG, avg_)
233 #define MSPEL_FILTER13_CORE(UNPACK, MOVQ, A1, A2, A3, A4) \
234 MOVQ "*0+"A1", %%mm1 \n\t" \
235 MOVQ "*4+"A1", %%mm2 \n\t" \
238 "pmullw "MANGLE(ff_pw_3)", %%mm1\n\t" \
239 "pmullw "MANGLE(ff_pw_3)", %%mm2\n\t" \
240 MOVQ "*0+"A2", %%mm3 \n\t" \
241 MOVQ "*4+"A2", %%mm4 \n\t" \
244 "pmullw %%mm6, %%mm3 \n\t" \
245 "pmullw %%mm6, %%mm4 \n\t" \
246 "psubw %%mm1, %%mm3 \n\t" \
247 "psubw %%mm2, %%mm4 \n\t" \
248 MOVQ "*0+"A4", %%mm1 \n\t" \
249 MOVQ "*4+"A4", %%mm2 \n\t" \
252 "psllw $2, %%mm1 \n\t" \
253 "psllw $2, %%mm2 \n\t" \
254 "psubw %%mm1, %%mm3 \n\t" \
255 "psubw %%mm2, %%mm4 \n\t" \
256 MOVQ "*0+"A3", %%mm1 \n\t" \
257 MOVQ "*4+"A3", %%mm2 \n\t" \
260 "pmullw %%mm5, %%mm1 \n\t" \
261 "pmullw %%mm5, %%mm2 \n\t" \
262 "paddw %%mm1, %%mm3 \n\t" \
263 "paddw %%mm2, %%mm4 \n\t"
273 #define MSPEL_FILTER13_VER_16B(NAME, A1, A2, A3, A4) \
275 vc1_put_ver_16b_ ## NAME ## _mmx(int16_t *dst, const uint8_t *src, \
276 x86_reg src_stride, \
277 int rnd, int64_t shift) \
282 LOAD_ROUNDER_MMX("%5") \
283 "movq "MANGLE(ff_pw_53)", %%mm5\n\t" \
284 "movq "MANGLE(ff_pw_18)", %%mm6\n\t" \
287 MSPEL_FILTER13_CORE(DO_UNPACK, "movd 1", A1, A2, A3, A4) \
288 NORMALIZE_MMX("%6") \
289 TRANSFER_DONT_PACK(OP_PUT) \
291 "movd 8+"A1", %%mm1 \n\t" \
293 "movq %%mm1, %%mm3 \n\t" \
294 "paddw %%mm1, %%mm1 \n\t" \
295 "paddw %%mm3, %%mm1 \n\t" \
296 "movd 8+"A2", %%mm3 \n\t" \
298 "pmullw %%mm6, %%mm3 \n\t" \
299 "psubw %%mm1, %%mm3 \n\t" \
300 "movd 8+"A3", %%mm1 \n\t" \
302 "pmullw %%mm5, %%mm1 \n\t" \
303 "paddw %%mm1, %%mm3 \n\t" \
304 "movd 8+"A4", %%mm1 \n\t" \
306 "psllw $2, %%mm1 \n\t" \
307 "psubw %%mm1, %%mm3 \n\t" \
308 "paddw %%mm7, %%mm3 \n\t" \
309 "psraw %6, %%mm3 \n\t" \
310 "movq %%mm3, 16(%2) \n\t" \
315 : "+r"(h), "+r" (src), "+r" (dst) \
316 : "r"(src_stride), "r"(3*src_stride), \
317 "m"(rnd), "m"(shift) \
329 #define MSPEL_FILTER13_HOR_16B(NAME, A1, A2, A3, A4, OP, OPNAME) \
331 OPNAME ## vc1_hor_16b_ ## NAME ## _mmx(uint8_t *dst, x86_reg stride, \
332 const int16_t *src, int rnd) \
336 rnd -= (-4+58+13-3)*256; \
338 LOAD_ROUNDER_MMX("%4") \
339 "movq "MANGLE(ff_pw_18)", %%mm6 \n\t" \
340 "movq "MANGLE(ff_pw_53)", %%mm5 \n\t" \
343 MSPEL_FILTER13_CORE(DONT_UNPACK, "movq 2", A1, A2, A3, A4) \
344 NORMALIZE_MMX("$7") \
346 "paddw "MANGLE(ff_pw_128)", %%mm3 \n\t" \
347 "paddw "MANGLE(ff_pw_128)", %%mm4 \n\t" \
348 TRANSFER_DO_PACK(OP) \
353 : "+r"(h), "+r" (src), "+r" (dst) \
354 : "r"(stride), "m"(rnd) \
367 #define MSPEL_FILTER13_8B(NAME, A1, A2, A3, A4, OP, OPNAME) \
369 OPNAME ## vc1_## NAME ## _mmx(uint8_t *dst, const uint8_t *src, \
370 x86_reg stride, int rnd, x86_reg offset) \
376 LOAD_ROUNDER_MMX("%6") \
377 "movq "MANGLE(ff_pw_53)", %%mm5 \n\t" \
378 "movq "MANGLE(ff_pw_18)", %%mm6 \n\t" \
381 MSPEL_FILTER13_CORE(DO_UNPACK, "movd 1", A1, A2, A3, A4) \
382 NORMALIZE_MMX("$6") \
383 TRANSFER_DO_PACK(OP) \
388 : "+r"(h), "+r" (src), "+r" (dst) \
389 : "r"(offset), "r"(3*offset), "g"(stride), "m"(rnd) \
395 MSPEL_FILTER13_8B (
shift1,
"0(%1,%4 )",
"0(%1,%3,2)",
"0(%1,%3 )",
"0(%1 )",
OP_PUT, put_)
396 MSPEL_FILTER13_8B (
shift1, "0(%1,%4 )", "0(%1,%3,2)", "0(%1,%3 )", "0(%1 )", OP_AVG, avg_)
397 MSPEL_FILTER13_VER_16B(shift1, "0(%1,%4 )", "0(%1,%3,2)", "0(%1,%3 )", "0(%1 )")
398 MSPEL_FILTER13_HOR_16B(shift1, "2*3(%1)", "2*2(%1)", "2*1(%1)", "2*0(%1)",
OP_PUT, put_)
399 MSPEL_FILTER13_HOR_16B(shift1, "2*3(%1)", "2*2(%1)", "2*1(%1)", "2*0(%1)", OP_AVG, avg_)
402 MSPEL_FILTER13_8B (shift3, "0(%1 )", "0(%1,%3 )", "0(%1,%3,2)", "0(%1,%4 )", OP_PUT, put_)
403 MSPEL_FILTER13_8B (shift3, "0(%1 )", "0(%1,%3 )", "0(%1,%3,2)", "0(%1,%4 )", OP_AVG, avg_)
404 MSPEL_FILTER13_VER_16B(shift3, "0(%1 )", "0(%1,%3 )", "0(%1,%3,2)", "0(%1,%4 )")
405 MSPEL_FILTER13_HOR_16B(shift3, "2*0(%1)", "2*1(%1)", "2*2(%1)", "2*3(%1)", OP_PUT, put_)
406 MSPEL_FILTER13_HOR_16B(shift3, "2*0(%1)", "2*1(%1)", "2*2(%1)", "2*3(%1)", OP_AVG, avg_)
408 typedef
void (*vc1_mspel_mc_filter_ver_16bits)(int16_t *dst, const
uint8_t *src,
x86_reg src_stride,
int rnd, int64_t shift);
409 typedef
void (*vc1_mspel_mc_filter_hor_16bits)(
uint8_t *dst,
x86_reg dst_stride, const int16_t *src,
int rnd);
423 #define VC1_MSPEL_MC(OP)\
424 static void OP ## vc1_mspel_mc(uint8_t *dst, const uint8_t *src, int stride,\
425 int hmode, int vmode, int rnd)\
427 static const vc1_mspel_mc_filter_ver_16bits vc1_put_shift_ver_16bits[] =\
428 { NULL, vc1_put_ver_16b_shift1_mmx, vc1_put_ver_16b_shift2_mmx, vc1_put_ver_16b_shift3_mmx };\
429 static const vc1_mspel_mc_filter_hor_16bits vc1_put_shift_hor_16bits[] =\
430 { NULL, OP ## vc1_hor_16b_shift1_mmx, OP ## vc1_hor_16b_shift2_mmx, OP ## vc1_hor_16b_shift3_mmx };\
431 static const vc1_mspel_mc_filter_8bits vc1_put_shift_8bits[] =\
432 { NULL, OP ## vc1_shift1_mmx, OP ## vc1_shift2_mmx, OP ## vc1_shift3_mmx };\
435 "pxor %%mm0, %%mm0 \n\t"\
441 static const int shift_value[] = { 0, 5, 1, 5 };\
442 int shift = (shift_value[hmode]+shift_value[vmode])>>1;\
444 DECLARE_ALIGNED(16, int16_t, tmp)[12*8];\
446 r = (1<<(shift-1)) + rnd-1;\
447 vc1_put_shift_ver_16bits[vmode](tmp, src-1, stride, r, shift);\
449 vc1_put_shift_hor_16bits[hmode](dst, stride, tmp+1, 64-rnd);\
453 vc1_put_shift_8bits[vmode](dst, src, stride, 1-rnd, stride);\
459 vc1_put_shift_8bits[hmode](dst, src, stride, rnd, 1);\
466 #define DECLARE_FUNCTION(a, b) \
467 static void put_vc1_mspel_mc ## a ## b ## _mmx(uint8_t *dst, const uint8_t *src, int stride, int rnd) { \
468 put_vc1_mspel_mc(dst, src, stride, a, b, rnd); \
470 static void avg_vc1_mspel_mc ## a ## b ## _mmxext(uint8_t *dst, \
471 const uint8_t *src, \
472 int stride, int rnd) \
474 avg_vc1_mspel_mc(dst, src, stride, a, b, rnd); \
477 DECLARE_FUNCTION(0, 1)
478 DECLARE_FUNCTION(0, 2)
479 DECLARE_FUNCTION(0, 3)
481 DECLARE_FUNCTION(1, 0)
482 DECLARE_FUNCTION(1, 1)
483 DECLARE_FUNCTION(1, 2)
484 DECLARE_FUNCTION(1, 3)
486 DECLARE_FUNCTION(2, 0)
487 DECLARE_FUNCTION(2, 1)
488 DECLARE_FUNCTION(2, 2)
489 DECLARE_FUNCTION(2, 3)
491 DECLARE_FUNCTION(3, 0)
492 DECLARE_FUNCTION(3, 1)
493 DECLARE_FUNCTION(3, 2)
494 DECLARE_FUNCTION(3, 3)
496 static
void vc1_inv_trans_4x4_dc_mmxext(
uint8_t *dest,
int linesize,
500 dc = (17 * dc + 4) >> 3;
501 dc = (17 * dc + 64) >> 7;
503 "movd %0, %%mm0 \n\t"
504 "pshufw $0, %%mm0, %%mm0 \n\t"
505 "pxor %%mm1, %%mm1 \n\t"
506 "psubw %%mm0, %%mm1 \n\t"
507 "packuswb %%mm0, %%mm0 \n\t"
508 "packuswb %%mm1, %%mm1 \n\t"
512 "movd %0, %%mm2 \n\t"
513 "movd %1, %%mm3 \n\t"
514 "movd %2, %%mm4 \n\t"
515 "movd %3, %%mm5 \n\t"
516 "paddusb %%mm0, %%mm2 \n\t"
517 "paddusb %%mm0, %%mm3 \n\t"
518 "paddusb %%mm0, %%mm4 \n\t"
519 "paddusb %%mm0, %%mm5 \n\t"
520 "psubusb %%mm1, %%mm2 \n\t"
521 "psubusb %%mm1, %%mm3 \n\t"
522 "psubusb %%mm1, %%mm4 \n\t"
523 "psubusb %%mm1, %%mm5 \n\t"
524 "movd %%mm2, %0 \n\t"
525 "movd %%mm3, %1 \n\t"
526 "movd %%mm4, %2 \n\t"
527 "movd %%mm5, %3 \n\t"
528 :
"+m"(*(uint32_t*)(dest+0*linesize)),
529 "+m"(*(uint32_t*)(dest+1*linesize)),
530 "+m"(*(uint32_t*)(dest+2*linesize)),
531 "+m"(*(uint32_t*)(dest+3*linesize))
535 static void vc1_inv_trans_4x8_dc_mmxext(
uint8_t *dest,
int linesize,
539 dc = (17 * dc + 4) >> 3;
540 dc = (12 * dc + 64) >> 7;
542 "movd %0, %%mm0 \n\t"
543 "pshufw $0, %%mm0, %%mm0 \n\t"
544 "pxor %%mm1, %%mm1 \n\t"
545 "psubw %%mm0, %%mm1 \n\t"
546 "packuswb %%mm0, %%mm0 \n\t"
547 "packuswb %%mm1, %%mm1 \n\t"
551 "movd %0, %%mm2 \n\t"
552 "movd %1, %%mm3 \n\t"
553 "movd %2, %%mm4 \n\t"
554 "movd %3, %%mm5 \n\t"
555 "paddusb %%mm0, %%mm2 \n\t"
556 "paddusb %%mm0, %%mm3 \n\t"
557 "paddusb %%mm0, %%mm4 \n\t"
558 "paddusb %%mm0, %%mm5 \n\t"
559 "psubusb %%mm1, %%mm2 \n\t"
560 "psubusb %%mm1, %%mm3 \n\t"
561 "psubusb %%mm1, %%mm4 \n\t"
562 "psubusb %%mm1, %%mm5 \n\t"
563 "movd %%mm2, %0 \n\t"
564 "movd %%mm3, %1 \n\t"
565 "movd %%mm4, %2 \n\t"
566 "movd %%mm5, %3 \n\t"
567 :
"+m"(*(uint32_t*)(dest+0*linesize)),
568 "+m"(*(uint32_t*)(dest+1*linesize)),
569 "+m"(*(uint32_t*)(dest+2*linesize)),
570 "+m"(*(uint32_t*)(dest+3*linesize))
574 "movd %0, %%mm2 \n\t"
575 "movd %1, %%mm3 \n\t"
576 "movd %2, %%mm4 \n\t"
577 "movd %3, %%mm5 \n\t"
578 "paddusb %%mm0, %%mm2 \n\t"
579 "paddusb %%mm0, %%mm3 \n\t"
580 "paddusb %%mm0, %%mm4 \n\t"
581 "paddusb %%mm0, %%mm5 \n\t"
582 "psubusb %%mm1, %%mm2 \n\t"
583 "psubusb %%mm1, %%mm3 \n\t"
584 "psubusb %%mm1, %%mm4 \n\t"
585 "psubusb %%mm1, %%mm5 \n\t"
586 "movd %%mm2, %0 \n\t"
587 "movd %%mm3, %1 \n\t"
588 "movd %%mm4, %2 \n\t"
589 "movd %%mm5, %3 \n\t"
590 :
"+m"(*(uint32_t*)(dest+0*linesize)),
591 "+m"(*(uint32_t*)(dest+1*linesize)),
592 "+m"(*(uint32_t*)(dest+2*linesize)),
593 "+m"(*(uint32_t*)(dest+3*linesize))
597 static void vc1_inv_trans_8x4_dc_mmxext(
uint8_t *dest,
int linesize,
601 dc = ( 3 * dc + 1) >> 1;
602 dc = (17 * dc + 64) >> 7;
604 "movd %0, %%mm0 \n\t"
605 "pshufw $0, %%mm0, %%mm0 \n\t"
606 "pxor %%mm1, %%mm1 \n\t"
607 "psubw %%mm0, %%mm1 \n\t"
608 "packuswb %%mm0, %%mm0 \n\t"
609 "packuswb %%mm1, %%mm1 \n\t"
613 "movq %0, %%mm2 \n\t"
614 "movq %1, %%mm3 \n\t"
615 "movq %2, %%mm4 \n\t"
616 "movq %3, %%mm5 \n\t"
617 "paddusb %%mm0, %%mm2 \n\t"
618 "paddusb %%mm0, %%mm3 \n\t"
619 "paddusb %%mm0, %%mm4 \n\t"
620 "paddusb %%mm0, %%mm5 \n\t"
621 "psubusb %%mm1, %%mm2 \n\t"
622 "psubusb %%mm1, %%mm3 \n\t"
623 "psubusb %%mm1, %%mm4 \n\t"
624 "psubusb %%mm1, %%mm5 \n\t"
625 "movq %%mm2, %0 \n\t"
626 "movq %%mm3, %1 \n\t"
627 "movq %%mm4, %2 \n\t"
628 "movq %%mm5, %3 \n\t"
629 :
"+m"(*(uint32_t*)(dest+0*linesize)),
630 "+m"(*(uint32_t*)(dest+1*linesize)),
631 "+m"(*(uint32_t*)(dest+2*linesize)),
632 "+m"(*(uint32_t*)(dest+3*linesize))
636 static void vc1_inv_trans_8x8_dc_mmxext(
uint8_t *dest,
int linesize,
640 dc = (3 * dc + 1) >> 1;
641 dc = (3 * dc + 16) >> 5;
643 "movd %0, %%mm0 \n\t"
644 "pshufw $0, %%mm0, %%mm0 \n\t"
645 "pxor %%mm1, %%mm1 \n\t"
646 "psubw %%mm0, %%mm1 \n\t"
647 "packuswb %%mm0, %%mm0 \n\t"
648 "packuswb %%mm1, %%mm1 \n\t"
652 "movq %0, %%mm2 \n\t"
653 "movq %1, %%mm3 \n\t"
654 "movq %2, %%mm4 \n\t"
655 "movq %3, %%mm5 \n\t"
656 "paddusb %%mm0, %%mm2 \n\t"
657 "paddusb %%mm0, %%mm3 \n\t"
658 "paddusb %%mm0, %%mm4 \n\t"
659 "paddusb %%mm0, %%mm5 \n\t"
660 "psubusb %%mm1, %%mm2 \n\t"
661 "psubusb %%mm1, %%mm3 \n\t"
662 "psubusb %%mm1, %%mm4 \n\t"
663 "psubusb %%mm1, %%mm5 \n\t"
664 "movq %%mm2, %0 \n\t"
665 "movq %%mm3, %1 \n\t"
666 "movq %%mm4, %2 \n\t"
667 "movq %%mm5, %3 \n\t"
668 :
"+m"(*(uint32_t*)(dest+0*linesize)),
669 "+m"(*(uint32_t*)(dest+1*linesize)),
670 "+m"(*(uint32_t*)(dest+2*linesize)),
671 "+m"(*(uint32_t*)(dest+3*linesize))
675 "movq %0, %%mm2 \n\t"
676 "movq %1, %%mm3 \n\t"
677 "movq %2, %%mm4 \n\t"
678 "movq %3, %%mm5 \n\t"
679 "paddusb %%mm0, %%mm2 \n\t"
680 "paddusb %%mm0, %%mm3 \n\t"
681 "paddusb %%mm0, %%mm4 \n\t"
682 "paddusb %%mm0, %%mm5 \n\t"
683 "psubusb %%mm1, %%mm2 \n\t"
684 "psubusb %%mm1, %%mm3 \n\t"
685 "psubusb %%mm1, %%mm4 \n\t"
686 "psubusb %%mm1, %%mm5 \n\t"
687 "movq %%mm2, %0 \n\t"
688 "movq %%mm3, %1 \n\t"
689 "movq %%mm4, %2 \n\t"
690 "movq %%mm5, %3 \n\t"
691 :
"+m"(*(uint32_t*)(dest+0*linesize)),
692 "+m"(*(uint32_t*)(dest+1*linesize)),
693 "+m"(*(uint32_t*)(dest+2*linesize)),
694 "+m"(*(uint32_t*)(dest+3*linesize))