27 |
; enable dequant saturate [-2048,2047], test purposes only. |
; enable dequant saturate [-2048,2047], test purposes only. |
28 |
%define SATURATE |
%define SATURATE |
29 |
|
|
30 |
; data/text alignment |
BITS 32 |
|
%define ALIGN 8 |
|
|
|
|
|
bits 32 |
|
31 |
|
|
32 |
%macro cglobal 1 |
%macro cglobal 1 |
33 |
%ifdef PREFIX |
%ifdef PREFIX |
38 |
%endif |
%endif |
39 |
%endmacro |
%endmacro |
40 |
|
|
41 |
;*************************************************************************** |
;============================================================================= |
42 |
; Local data |
; Read only Local data |
43 |
;*************************************************************************** |
;============================================================================= |
44 |
|
|
45 |
%ifdef FORMAT_COFF |
%ifdef FORMAT_COFF |
46 |
section .data data |
SECTION .rodata data |
47 |
%else |
%else |
48 |
section .data data align=16 |
SECTION .rodata data align=16 |
49 |
%endif |
%endif |
50 |
|
|
51 |
align 16 |
ALIGN 16 |
52 |
plus_one: |
plus_one: |
53 |
times 8 dw 1 |
times 8 dw 1 |
54 |
|
|
55 |
;=========================================================================== |
;----------------------------------------------------------------------------- |
56 |
; |
; |
57 |
; subtract by Q/2 table |
; subtract by Q/2 table |
58 |
; |
; |
59 |
;=========================================================================== |
;----------------------------------------------------------------------------- |
60 |
|
|
61 |
align 16 |
ALIGN 16 |
62 |
mmx_sub: |
mmx_sub: |
63 |
%assign quant 1 |
%assign quant 1 |
64 |
%rep 31 |
%rep 31 |
66 |
%assign quant quant+1 |
%assign quant quant+1 |
67 |
%endrep |
%endrep |
68 |
|
|
69 |
;=========================================================================== |
;----------------------------------------------------------------------------- |
70 |
; |
; |
71 |
; divide by 2Q table |
; divide by 2Q table |
72 |
; |
; |
74 |
; for q=1, _pmulhw_ will overflow so it is treated seperately |
; for q=1, _pmulhw_ will overflow so it is treated seperately |
75 |
; (3dnow2 provides _pmulhuw_ which wont cause overflow) |
; (3dnow2 provides _pmulhuw_ which wont cause overflow) |
76 |
; |
; |
77 |
;=========================================================================== |
;----------------------------------------------------------------------------- |
78 |
|
|
79 |
align 16 |
ALIGN 16 |
80 |
mmx_div: |
mmx_div: |
81 |
%assign quant 1 |
%assign quant 1 |
82 |
%rep 31 |
%rep 31 |
84 |
%assign quant quant+1 |
%assign quant quant+1 |
85 |
%endrep |
%endrep |
86 |
|
|
87 |
;=========================================================================== |
;----------------------------------------------------------------------------- |
88 |
; |
; |
89 |
; add by (odd(Q) ? Q : Q - 1) table |
; add by (odd(Q) ? Q : Q - 1) table |
90 |
; |
; |
91 |
;=========================================================================== |
;----------------------------------------------------------------------------- |
92 |
|
|
93 |
align 16 |
ALIGN 16 |
94 |
mmx_add: |
mmx_add: |
95 |
%assign quant 1 |
%assign quant 1 |
96 |
%rep 31 |
%rep 31 |
102 |
%assign quant quant+1 |
%assign quant quant+1 |
103 |
%endrep |
%endrep |
104 |
|
|
105 |
;=========================================================================== |
;----------------------------------------------------------------------------- |
106 |
; |
; |
107 |
; multiple by 2Q table |
; multiple by 2Q table |
108 |
; |
; |
109 |
;=========================================================================== |
;----------------------------------------------------------------------------- |
110 |
|
|
111 |
align 16 |
ALIGN 16 |
112 |
mmx_mul: |
mmx_mul: |
113 |
%assign quant 1 |
%assign quant 1 |
114 |
%rep 31 |
%rep 31 |
116 |
%assign quant quant+1 |
%assign quant quant+1 |
117 |
%endrep |
%endrep |
118 |
|
|
119 |
;=========================================================================== |
;----------------------------------------------------------------------------- |
120 |
; |
; |
121 |
; saturation limits |
; saturation limits |
122 |
; |
; |
123 |
;=========================================================================== |
;----------------------------------------------------------------------------- |
124 |
|
|
125 |
align 16 |
ALIGN 16 |
126 |
sse2_2047: |
sse2_2047: |
127 |
times 8 dw 2047 |
times 8 dw 2047 |
128 |
|
|
129 |
align 16 |
ALIGN 16 |
130 |
mmx_2047: |
mmx_2047: |
131 |
times 4 dw 2047 |
times 4 dw 2047 |
132 |
|
|
133 |
align 8 |
ALIGN 8 |
134 |
mmx_32768_minus_2048: |
mmx_32768_minus_2048: |
135 |
times 4 dw (32768-2048) |
times 4 dw (32768-2048) |
136 |
|
|
138 |
times 4 dw (32767-2047) |
times 4 dw (32767-2047) |
139 |
|
|
140 |
|
|
141 |
;*************************************************************************** |
;============================================================================= |
142 |
; Code |
; Code |
143 |
;*************************************************************************** |
;============================================================================= |
144 |
|
|
145 |
section .text |
SECTION .text |
146 |
|
|
147 |
|
cglobal quant_h263_intra_mmx |
148 |
|
cglobal quant_h263_intra_sse2 |
149 |
|
cglobal quant_h263_inter_mmx |
150 |
|
cglobal quant_h263_inter_sse2 |
151 |
|
cglobal dequant_h263_intra_mmx |
152 |
|
cglobal dequant_h263_intra_xmm |
153 |
|
cglobal dequant_h263_intra_sse2 |
154 |
|
cglobal dequant_h263_inter_mmx |
155 |
|
cglobal dequant_h263_inter_xmm |
156 |
|
cglobal dequant_h263_inter_sse2 |
157 |
|
|
158 |
;=========================================================================== |
;----------------------------------------------------------------------------- |
159 |
; |
; |
160 |
; uint32_t quant_h263_intra_mmx(int16_t * coeff, |
; uint32_t quant_h263_intra_mmx(int16_t * coeff, |
161 |
; const int16_t const * data, |
; const int16_t const * data, |
162 |
; const uint32_t quant, |
; const uint32_t quant, |
163 |
; const uint32_t dcscalar); |
; const uint32_t dcscalar, |
164 |
|
; const uint16_t *mpeg_matrices); |
165 |
; |
; |
166 |
;=========================================================================== |
;----------------------------------------------------------------------------- |
167 |
|
|
168 |
align ALIGN |
ALIGN 16 |
|
cglobal quant_h263_intra_mmx |
|
169 |
quant_h263_intra_mmx: |
quant_h263_intra_mmx: |
170 |
|
|
171 |
push ecx |
push ecx |
182 |
|
|
183 |
movq mm7, [mmx_div + eax * 8 - 8] |
movq mm7, [mmx_div + eax * 8 - 8] |
184 |
|
|
185 |
align ALIGN |
ALIGN 16 |
186 |
.loop |
.loop |
187 |
movq mm0, [esi + 8*ecx] ; mm0 = [1st] |
movq mm0, [esi + 8*ecx] ; mm0 = [1st] |
188 |
movq mm3, [esi + 8*ecx + 8] |
movq mm3, [esi + 8*ecx + 8] |
234 |
|
|
235 |
ret |
ret |
236 |
|
|
237 |
align ALIGN |
ALIGN 16 |
238 |
.q1loop |
.q1loop |
239 |
movq mm0, [esi + 8*ecx] ; mm0 = [1st] |
movq mm0, [esi + 8*ecx] ; mm0 = [1st] |
240 |
movq mm3, [esi + 8*ecx + 8] |
movq mm3, [esi + 8*ecx + 8] |
262 |
|
|
263 |
|
|
264 |
|
|
265 |
;=========================================================================== |
;----------------------------------------------------------------------------- |
266 |
; |
; |
267 |
; uint32_t quant_h263_intra_sse2(int16_t * coeff, |
; uint32_t quant_h263_intra_sse2(int16_t * coeff, |
268 |
; const int16_t const * data, |
; const int16_t const * data, |
269 |
; const uint32_t quant, |
; const uint32_t quant, |
270 |
; const uint32_t dcscalar); |
; const uint32_t dcscalar, |
271 |
|
; const uint16_t *mpeg_matrices); |
272 |
; |
; |
273 |
;=========================================================================== |
;----------------------------------------------------------------------------- |
274 |
|
|
275 |
align ALIGN |
ALIGN 16 |
|
cglobal quant_h263_intra_sse2 |
|
276 |
quant_h263_intra_sse2: |
quant_h263_intra_sse2: |
277 |
|
|
278 |
push esi |
push esi |
291 |
movq2dq xmm7, mm7 |
movq2dq xmm7, mm7 |
292 |
movlhps xmm7, xmm7 |
movlhps xmm7, xmm7 |
293 |
|
|
294 |
align 16 |
ALIGN 16 |
295 |
.qas2_loop |
.qas2_loop |
296 |
movdqa xmm0, [esi + ecx*8] ; xmm0 = [1st] |
movdqa xmm0, [esi + ecx*8] ; xmm0 = [1st] |
297 |
movdqa xmm3, [esi + ecx*8 + 16] ; xmm3 = [2nd] |
movdqa xmm3, [esi + ecx*8 + 16] ; xmm3 = [2nd] |
342 |
|
|
343 |
ret |
ret |
344 |
|
|
345 |
align 16 |
ALIGN 16 |
346 |
.qas2_q1loop |
.qas2_q1loop |
347 |
movdqa xmm0, [esi + ecx*8] ; xmm0 = [1st] |
movdqa xmm0, [esi + ecx*8] ; xmm0 = [1st] |
348 |
movdqa xmm3, [esi + ecx*8 + 16] ; xmm3 = [2nd] |
movdqa xmm3, [esi + ecx*8 + 16] ; xmm3 = [2nd] |
370 |
|
|
371 |
|
|
372 |
|
|
373 |
;=========================================================================== |
;----------------------------------------------------------------------------- |
374 |
; |
; |
375 |
; uint32_t quant_h263_inter_mmx(int16_t * coeff, |
; uint32_t quant_h263_inter_mmx(int16_t * coeff, |
376 |
; const int16_t const * data, |
; const int16_t const * data, |
377 |
; const uint32_t quant); |
; const uint32_t quant, |
378 |
|
; const uint16_t *mpeg_matrices); |
379 |
; |
; |
380 |
;=========================================================================== |
;----------------------------------------------------------------------------- |
381 |
|
|
382 |
align ALIGN |
ALIGN 16 |
|
cglobal quant_h263_inter_mmx |
|
383 |
quant_h263_inter_mmx: |
quant_h263_inter_mmx: |
384 |
|
|
385 |
push ecx |
push ecx |
400 |
|
|
401 |
movq mm7, [mmx_div + eax * 8 - 8] ; divider |
movq mm7, [mmx_div + eax * 8 - 8] ; divider |
402 |
|
|
403 |
align ALIGN |
ALIGN 8 |
404 |
.loop |
.loop |
405 |
movq mm0, [esi + 8*ecx] ; mm0 = [1st] |
movq mm0, [esi + 8*ecx] ; mm0 = [1st] |
406 |
movq mm3, [esi + 8*ecx + 8] |
movq mm3, [esi + 8*ecx + 8] |
442 |
|
|
443 |
ret |
ret |
444 |
|
|
445 |
align ALIGN |
ALIGN 8 |
446 |
.q1loop |
.q1loop |
447 |
movq mm0, [esi + 8*ecx] ; mm0 = [1st] |
movq mm0, [esi + 8*ecx] ; mm0 = [1st] |
448 |
movq mm3, [esi + 8*ecx+ 8] ; |
movq mm3, [esi + 8*ecx+ 8] ; |
475 |
|
|
476 |
|
|
477 |
|
|
478 |
;=========================================================================== |
;----------------------------------------------------------------------------- |
479 |
; |
; |
480 |
; uint32_t quant_h263_inter_sse2(int16_t * coeff, |
; uint32_t quant_h263_inter_sse2(int16_t * coeff, |
481 |
; const int16_t const * data, |
; const int16_t const * data, |
482 |
; const uint32_t quant); |
; const uint32_t quant, |
483 |
|
; const uint16_t *mpeg_matrices); |
484 |
; |
; |
485 |
;=========================================================================== |
;----------------------------------------------------------------------------- |
486 |
|
|
487 |
align 16 |
ALIGN 16 |
|
cglobal quant_h263_inter_sse2 |
|
488 |
quant_h263_inter_sse2: |
quant_h263_inter_sse2: |
489 |
|
|
490 |
push esi |
push esi |
510 |
movq2dq xmm7, mm0 |
movq2dq xmm7, mm0 |
511 |
movlhps xmm7, xmm7 |
movlhps xmm7, xmm7 |
512 |
|
|
513 |
align 16 |
ALIGN 16 |
514 |
.qes2_loop |
.qes2_loop |
515 |
movdqa xmm0, [esi + ecx*8] ; xmm0 = [1st] |
movdqa xmm0, [esi + ecx*8] ; xmm0 = [1st] |
516 |
movdqa xmm3, [esi + ecx*8 + 16] ; xmm3 = [2nd] |
movdqa xmm3, [esi + ecx*8 + 16] ; xmm3 = [2nd] |
557 |
|
|
558 |
ret |
ret |
559 |
|
|
560 |
align 16 |
ALIGN 16 |
561 |
.qes2_q1loop |
.qes2_q1loop |
562 |
movdqa xmm0, [esi + ecx*8] ; xmm0 = [1st] |
movdqa xmm0, [esi + ecx*8] ; xmm0 = [1st] |
563 |
movdqa xmm3, [esi + ecx*8 + 16] ; xmm3 = [2nd] |
movdqa xmm3, [esi + ecx*8 + 16] ; xmm3 = [2nd] |
588 |
jmp .qes2_done |
jmp .qes2_done |
589 |
|
|
590 |
|
|
591 |
;=========================================================================== |
;----------------------------------------------------------------------------- |
592 |
; |
; |
593 |
; uint32_t dequant_h263_intra_mmx(int16_t *data, |
; uint32_t dequant_h263_intra_mmx(int16_t *data, |
594 |
; const int16_t const *coeff, |
; const int16_t const *coeff, |
595 |
; const uint32_t quant, |
; const uint32_t quant, |
596 |
; const uint32_t dcscalar); |
; const uint32_t dcscalar, |
597 |
|
; const uint16_t *mpeg_matrices); |
598 |
; |
; |
599 |
;=========================================================================== |
;----------------------------------------------------------------------------- |
600 |
|
|
601 |
; note: we only saturate to +2047 *before* restoring the sign. |
; note: we only saturate to +2047 *before* restoring the sign. |
602 |
; Hence, final clamp really is [-2048,2047] |
; Hence, final clamp really is [-2048,2047] |
603 |
|
|
604 |
align ALIGN |
ALIGN 16 |
|
cglobal dequant_h263_intra_mmx |
|
605 |
dequant_h263_intra_mmx: |
dequant_h263_intra_mmx: |
606 |
|
|
607 |
mov edx, [esp+ 4] ; data |
mov edx, [esp+ 4] ; data |
611 |
movq mm7, [mmx_mul + eax*8 - 8] ; 2*quant |
movq mm7, [mmx_mul + eax*8 - 8] ; 2*quant |
612 |
mov eax, -16 |
mov eax, -16 |
613 |
|
|
614 |
align ALIGN |
ALIGN 16 |
615 |
.loop |
.loop |
616 |
movq mm0, [ecx+8*eax+8*16] ; c = coeff[i] |
movq mm0, [ecx+8*eax+8*16] ; c = coeff[i] |
617 |
movq mm3, [ecx+8*eax+8*16 + 8] ; c' = coeff[i+1] |
movq mm3, [ecx+8*eax+8*16 + 8] ; c' = coeff[i+1] |
665 |
xor eax, eax ; return(0); |
xor eax, eax ; return(0); |
666 |
ret |
ret |
667 |
|
|
668 |
;=========================================================================== |
;----------------------------------------------------------------------------- |
669 |
; |
; |
670 |
; uint32_t dequant_h263_intra_xmm(int16_t *data, |
; uint32_t dequant_h263_intra_xmm(int16_t *data, |
671 |
; const int16_t const *coeff, |
; const int16_t const *coeff, |
672 |
; const uint32_t quant, |
; const uint32_t quant, |
673 |
; const uint32_t dcscalar); |
; const uint32_t dcscalar, |
674 |
|
; const uint16_t *mpeg_matrices); |
675 |
; |
; |
676 |
;=========================================================================== |
;----------------------------------------------------------------------------- |
677 |
|
|
678 |
; this is the same as dequant_inter_mmx, except that we're |
; this is the same as dequant_inter_mmx, except that we're |
679 |
; saturating using 'pminsw' (saves 2 cycles/loop => ~5% faster) |
; saturating using 'pminsw' (saves 2 cycles/loop => ~5% faster) |
680 |
|
|
681 |
align ALIGN |
ALIGN 16 |
|
cglobal dequant_h263_intra_xmm |
|
682 |
dequant_h263_intra_xmm: |
dequant_h263_intra_xmm: |
683 |
|
|
684 |
mov edx, [esp+ 4] ; data |
mov edx, [esp+ 4] ; data |
688 |
movq mm7, [mmx_mul + eax*8 - 8] ; 2*quant |
movq mm7, [mmx_mul + eax*8 - 8] ; 2*quant |
689 |
mov eax, -16 |
mov eax, -16 |
690 |
|
|
691 |
align ALIGN |
ALIGN 16 |
692 |
.loop |
.loop |
693 |
movq mm0, [ecx+8*eax+8*16] ; c = coeff[i] |
movq mm0, [ecx+8*eax+8*16] ; c = coeff[i] |
694 |
movq mm3, [ecx+8*eax+8*16 + 8] ; c' = coeff[i+1] |
movq mm3, [ecx+8*eax+8*16 + 8] ; c' = coeff[i+1] |
741 |
ret |
ret |
742 |
|
|
743 |
|
|
744 |
;=========================================================================== |
;----------------------------------------------------------------------------- |
745 |
; |
; |
746 |
; uint32_t dequant_h263_intra_sse2(int16_t *data, |
; uint32_t dequant_h263_intra_sse2(int16_t *data, |
747 |
; const int16_t const *coeff, |
; const int16_t const *coeff, |
748 |
; const uint32_t quant, |
; const uint32_t quant, |
749 |
; const uint32_t dcscalar); |
; const uint32_t dcscalar, |
750 |
|
; const uint16_t *mpeg_matrices); |
751 |
; |
; |
752 |
;=========================================================================== |
;----------------------------------------------------------------------------- |
753 |
|
|
754 |
align ALIGN |
ALIGN 16 |
|
cglobal dequant_h263_intra_sse2 |
|
755 |
dequant_h263_intra_sse2: |
dequant_h263_intra_sse2: |
756 |
mov edx, [esp+ 4] ; data |
mov edx, [esp+ 4] ; data |
757 |
mov ecx, [esp+ 8] ; coeff |
mov ecx, [esp+ 8] ; coeff |
764 |
movlhps xmm7, xmm7 |
movlhps xmm7, xmm7 |
765 |
mov eax, -16 |
mov eax, -16 |
766 |
|
|
767 |
align ALIGN |
ALIGN 16 |
768 |
.loop |
.loop |
769 |
movdqa xmm0, [ecx + 8*16 + 8*eax] ; c = coeff[i] |
movdqa xmm0, [ecx + 8*16 + 8*eax] ; c = coeff[i] |
770 |
movdqa xmm3, [ecx + 8*16 + 8*eax+ 16] |
movdqa xmm3, [ecx + 8*16 + 8*eax+ 16] |
816 |
xor eax, eax |
xor eax, eax |
817 |
ret |
ret |
818 |
|
|
819 |
;=========================================================================== |
;----------------------------------------------------------------------------- |
820 |
; |
; |
821 |
; uint32t dequant_h263_inter_mmx(int16_t * data, |
; uint32t dequant_h263_inter_mmx(int16_t * data, |
822 |
; const int16_t * const coeff, |
; const int16_t * const coeff, |
823 |
; const uint32_t quant); |
; const uint32_t quant, |
824 |
|
; const uint16_t *mpeg_matrices); |
825 |
; |
; |
826 |
;=========================================================================== |
;----------------------------------------------------------------------------- |
827 |
|
|
828 |
align ALIGN |
ALIGN 16 |
|
cglobal dequant_h263_inter_mmx |
|
829 |
dequant_h263_inter_mmx: |
dequant_h263_inter_mmx: |
830 |
|
|
831 |
mov edx, [esp+ 4] ; data |
mov edx, [esp+ 4] ; data |
835 |
movq mm7, [mmx_mul + eax*8 - 8] ; 2*quant |
movq mm7, [mmx_mul + eax*8 - 8] ; 2*quant |
836 |
mov eax, -16 |
mov eax, -16 |
837 |
|
|
838 |
align ALIGN |
ALIGN 16 |
839 |
.loop |
.loop |
840 |
movq mm0, [ecx+8*eax+8*16] ; c = coeff[i] |
movq mm0, [ecx+8*eax+8*16] ; c = coeff[i] |
841 |
movq mm3, [ecx+8*eax+8*16 + 8] ; c' = coeff[i+1] |
movq mm3, [ecx+8*eax+8*16 + 8] ; c' = coeff[i+1] |
876 |
xor eax, eax |
xor eax, eax |
877 |
ret |
ret |
878 |
|
|
879 |
;=========================================================================== |
;----------------------------------------------------------------------------- |
880 |
; |
; |
881 |
; uint32_t dequant_h263_inter_xmm(int16_t * data, |
; uint32_t dequant_h263_inter_xmm(int16_t * data, |
882 |
; const int16_t * const coeff, |
; const int16_t * const coeff, |
883 |
; const uint32_t quant); |
; const uint32_t quant, |
884 |
|
; const uint16_t *mpeg_matrices); |
885 |
; |
; |
886 |
;=========================================================================== |
;----------------------------------------------------------------------------- |
887 |
|
|
888 |
; this is the same as dequant_inter_mmx, |
; this is the same as dequant_inter_mmx, |
889 |
; except that we're saturating using 'pminsw' (saves 2 cycles/loop) |
; except that we're saturating using 'pminsw' (saves 2 cycles/loop) |
890 |
|
|
891 |
align ALIGN |
ALIGN 16 |
|
cglobal dequant_h263_inter_xmm |
|
892 |
dequant_h263_inter_xmm: |
dequant_h263_inter_xmm: |
893 |
|
|
894 |
mov edx, [esp+ 4] ; data |
mov edx, [esp+ 4] ; data |
898 |
movq mm7, [mmx_mul + eax*8 - 8] ; 2*quant |
movq mm7, [mmx_mul + eax*8 - 8] ; 2*quant |
899 |
mov eax, -16 |
mov eax, -16 |
900 |
|
|
901 |
align ALIGN |
ALIGN 16 |
902 |
.loop |
.loop |
903 |
movq mm0, [ecx+8*eax+8*16] ; c = coeff[i] |
movq mm0, [ecx+8*eax+8*16] ; c = coeff[i] |
904 |
movq mm3, [ecx+8*eax+8*16 + 8] ; c' = coeff[i+1] |
movq mm3, [ecx+8*eax+8*16 + 8] ; c' = coeff[i+1] |
937 |
xor eax, eax |
xor eax, eax |
938 |
ret |
ret |
939 |
|
|
940 |
;=========================================================================== |
;----------------------------------------------------------------------------- |
941 |
; |
; |
942 |
; uint32_t dequant_h263_inter_sse2(int16_t * data, |
; uint32_t dequant_h263_inter_sse2(int16_t * data, |
943 |
; const int16_t * const coeff, |
; const int16_t * const coeff, |
944 |
; const uint32_t quant); |
; const uint32_t quant, |
945 |
|
; const uint16_t *mpeg_matrices); |
946 |
; |
; |
947 |
;=========================================================================== |
;----------------------------------------------------------------------------- |
948 |
|
|
949 |
align ALIGN |
ALIGN 16 |
|
cglobal dequant_h263_inter_sse2 |
|
950 |
dequant_h263_inter_sse2: |
dequant_h263_inter_sse2: |
951 |
mov edx, [esp + 4] ; data |
mov edx, [esp + 4] ; data |
952 |
mov ecx, [esp + 8] ; coeff |
mov ecx, [esp + 8] ; coeff |
959 |
movlhps xmm7, xmm7 |
movlhps xmm7, xmm7 |
960 |
mov eax, -16 |
mov eax, -16 |
961 |
|
|
962 |
align ALIGN |
ALIGN 16 |
963 |
.loop |
.loop |
964 |
movdqa xmm0, [ecx + 8*16 + 8*eax] ; c = coeff[i] |
movdqa xmm0, [ecx + 8*16 + 8*eax] ; c = coeff[i] |
965 |
movdqa xmm3, [ecx + 8*16 + 8*eax + 16] |
movdqa xmm3, [ecx + 8*16 + 8*eax + 16] |