29 |
; enable dequant saturate [-2048,2047], test purposes only. |
; enable dequant saturate [-2048,2047], test purposes only. |
30 |
%define SATURATE |
%define SATURATE |
31 |
|
|
32 |
; data/text alignment |
BITS 32 |
|
%define ALIGN 16 |
|
|
|
|
|
bits 32 |
|
33 |
|
|
34 |
%macro cglobal 1 |
%macro cglobal 1 |
35 |
%ifdef PREFIX |
%ifdef PREFIX |
40 |
%endif |
%endif |
41 |
%endmacro |
%endmacro |
42 |
|
|
43 |
;*************************************************************************** |
;============================================================================= |
44 |
; Local data |
; Local data |
45 |
;*************************************************************************** |
;============================================================================= |
46 |
|
|
47 |
%ifdef FORMAT_COFF |
%ifdef FORMAT_COFF |
48 |
section .data data |
SECTION .rodata |
49 |
%else |
%else |
50 |
section .data data align=16 |
SECTION .rodata align=16 |
51 |
%endif |
%endif |
52 |
|
|
53 |
align 4 |
align 4 |
59 |
%assign i i+1 |
%assign i i+1 |
60 |
%endrep |
%endrep |
61 |
|
|
62 |
align 16 |
ALIGN 16 |
63 |
plus_one: |
plus_one: |
64 |
times 8 dw 1 |
times 8 dw 1 |
65 |
|
|
66 |
;=========================================================================== |
;----------------------------------------------------------------------------- |
|
; |
|
67 |
; subtract by Q/2 table |
; subtract by Q/2 table |
68 |
; |
;----------------------------------------------------------------------------- |
|
;=========================================================================== |
|
69 |
|
|
70 |
align 16 |
ALIGN 16 |
71 |
mmx_sub: |
mmx_sub: |
72 |
%assign i 1 |
%assign i 1 |
73 |
%rep 31 |
%rep 31 |
76 |
%endrep |
%endrep |
77 |
|
|
78 |
|
|
79 |
;=========================================================================== |
;----------------------------------------------------------------------------- |
80 |
; |
; |
81 |
; divide by 2Q table |
; divide by 2Q table |
82 |
; |
; |
84 |
; for q=1, _pmulhw_ will overflow so it is treated seperately |
; for q=1, _pmulhw_ will overflow so it is treated seperately |
85 |
; (3dnow2 provides _pmulhuw_ which wont cause overflow) |
; (3dnow2 provides _pmulhuw_ which wont cause overflow) |
86 |
; |
; |
87 |
;=========================================================================== |
;----------------------------------------------------------------------------- |
88 |
|
|
89 |
align 16 |
ALIGN 16 |
90 |
mmx_div: |
mmx_div: |
91 |
%assign i 1 |
%assign i 1 |
92 |
%rep 31 |
%rep 31 |
94 |
%assign i i+1 |
%assign i i+1 |
95 |
%endrep |
%endrep |
96 |
|
|
97 |
;=========================================================================== |
;----------------------------------------------------------------------------- |
|
; |
|
98 |
; add by (odd(Q) ? Q : Q - 1) table |
; add by (odd(Q) ? Q : Q - 1) table |
99 |
; |
;----------------------------------------------------------------------------- |
|
;=========================================================================== |
|
100 |
|
|
101 |
align 16 |
ALIGN 16 |
102 |
mmx_add: |
mmx_add: |
103 |
%assign i 1 |
%assign i 1 |
104 |
%rep 31 |
%rep 31 |
110 |
%assign i i+1 |
%assign i i+1 |
111 |
%endrep |
%endrep |
112 |
|
|
113 |
;=========================================================================== |
;----------------------------------------------------------------------------- |
|
; |
|
114 |
; multiple by 2Q table |
; multiple by 2Q table |
115 |
; |
;----------------------------------------------------------------------------- |
|
;=========================================================================== |
|
116 |
|
|
117 |
align 16 |
ALIGN 16 |
118 |
mmx_mul: |
mmx_mul: |
119 |
%assign i 1 |
%assign i 1 |
120 |
%rep 31 |
%rep 31 |
122 |
%assign i i+1 |
%assign i i+1 |
123 |
%endrep |
%endrep |
124 |
|
|
125 |
;=========================================================================== |
;----------------------------------------------------------------------------- |
|
; |
|
126 |
; saturation limits |
; saturation limits |
127 |
; |
;----------------------------------------------------------------------------- |
|
;=========================================================================== |
|
128 |
|
|
129 |
align 8 |
ALIGN 8 |
130 |
mmx_32768_minus_2048: |
mmx_32768_minus_2048: |
131 |
times 4 dw (32768-2048) |
times 4 dw (32768-2048) |
132 |
mmx_32767_minus_2047: |
mmx_32767_minus_2047: |
133 |
times 4 dw (32767-2047) |
times 4 dw (32767-2047) |
134 |
|
|
135 |
align 16 |
ALIGN 16 |
136 |
mmx_2047: |
mmx_2047: |
137 |
times 4 dw 2047 |
times 4 dw 2047 |
138 |
|
|
139 |
align 8 |
ALIGN 8 |
140 |
mmzero: |
mmzero: |
141 |
dd 0, 0 |
dd 0, 0 |
142 |
int2047: |
int2047: |
144 |
int_2048: |
int_2048: |
145 |
dd -2048 |
dd -2048 |
146 |
|
|
147 |
;*************************************************************************** |
;============================================================================= |
148 |
; Code |
; Code |
149 |
;*************************************************************************** |
;============================================================================= |
150 |
|
|
151 |
section .text |
SECTION .text |
152 |
|
|
153 |
|
|
154 |
;=========================================================================== |
;----------------------------------------------------------------------------- |
155 |
; |
; |
156 |
; uint32_t quant_h263_intra_3dne(int16_t * coeff, |
; uint32_t quant_h263_intra_3dne(int16_t * coeff, |
157 |
; const int16_t const * data, |
; const int16_t const * data, |
158 |
; const uint32_t quant, |
; const uint32_t quant, |
159 |
; const uint32_t dcscalar); |
; const uint32_t dcscalar, |
160 |
|
; const uint16_t *mpeg_matrices); |
161 |
; |
; |
162 |
;=========================================================================== |
;----------------------------------------------------------------------------- |
163 |
;This is Athlon-optimized code (ca 70 clk per call) |
;This is Athlon-optimized code (ca 70 clk per call) |
164 |
|
|
165 |
%macro quant_intra1 1 |
%macro quant_intra1 1 |
|
|
|
166 |
psubw mm1, mm0 ;A3 |
psubw mm1, mm0 ;A3 |
167 |
psubw mm3, mm2 ;B3 |
psubw mm3, mm2 ;B3 |
168 |
%if (%1) |
%if (%1) |
170 |
psubw mm7, mm6 ;D8 |
psubw mm7, mm6 ;D8 |
171 |
%endif |
%endif |
172 |
|
|
173 |
align 8 |
ALIGN 8 |
174 |
db 0Fh, 6Fh, 64h, 21h, (%1 * 32 +16) ;movq mm4, [ecx + %1 * 32 +16+32] ;C1 |
db 0Fh, 6Fh, 64h, 21h, (%1 * 32 +16) ;movq mm4, [ecx + %1 * 32 +16+32] ;C1 |
175 |
pmaxsw mm1, mm0 ;A4 |
pmaxsw mm1, mm0 ;A4 |
176 |
db 0Fh, 6Fh, 74h, 21h, (%1 * 32 +24) ;movq mm6, [ecx + %1 * 32 +24+32] ;D1 |
db 0Fh, 6Fh, 74h, 21h, (%1 * 32 +24) ;movq mm6, [ecx + %1 * 32 +24+32] ;D1 |
244 |
psubw mm7, mm6 ;D8 |
psubw mm7, mm6 ;D8 |
245 |
%endif |
%endif |
246 |
|
|
247 |
align 8 |
ALIGN 8 |
248 |
db 0Fh, 6Fh, 64h, 21h, (%1 * 32 +16) ;movq mm4, [ecx + %1 * 32 +16+32] ;C1 |
db 0Fh, 6Fh, 64h, 21h, (%1 * 32 +16) ;movq mm4, [ecx + %1 * 32 +16+32] ;C1 |
249 |
pmaxsw mm1, mm0 ;A4 |
pmaxsw mm1, mm0 ;A4 |
250 |
db 0Fh, 6Fh, 74h, 21h, (%1 * 32 +24) ;movq mm6, [ecx + %1 * 32 +24+32] ;D1 |
db 0Fh, 6Fh, 74h, 21h, (%1 * 32 +24) ;movq mm6, [ecx + %1 * 32 +24+32] ;D1 |
311 |
%endmacro |
%endmacro |
312 |
|
|
313 |
|
|
314 |
align ALIGN |
ALIGN 16 |
315 |
cglobal quant_h263_intra_3dne |
cglobal quant_h263_intra_3dne |
316 |
quant_h263_intra_3dne: |
quant_h263_intra_3dne: |
317 |
|
|
366 |
xor eax, eax |
xor eax, eax |
367 |
ret |
ret |
368 |
|
|
369 |
align 16 |
ALIGN 16 |
370 |
|
|
371 |
.q1loop |
.q1loop |
372 |
quant_intra1 0 |
quant_intra1 0 |
405 |
|
|
406 |
|
|
407 |
|
|
408 |
;=========================================================================== |
;----------------------------------------------------------------------------- |
409 |
; |
; |
410 |
; uint32_t quant_h263_inter_3dne(int16_t * coeff, |
; uint32_t quant_h263_inter_3dne(int16_t * coeff, |
411 |
; const int16_t const * data, |
; const int16_t const * data, |
412 |
; const uint32_t quant); |
; const uint32_t quant, |
413 |
|
; const uint16_t *mpeg_matrices); |
414 |
; |
; |
415 |
;=========================================================================== |
;----------------------------------------------------------------------------- |
416 |
;This is Athlon-optimized code (ca 90 clk per call) |
;This is Athlon-optimized code (ca 90 clk per call) |
417 |
;Optimized by Jaan, 30 Nov 2002 |
;Optimized by Jaan, 30 Nov 2002 |
418 |
|
|
487 |
movq [edx + %1*16+8], mm4 |
movq [edx + %1*16+8], mm4 |
488 |
%endmacro |
%endmacro |
489 |
|
|
490 |
align ALIGN |
ALIGN 16 |
491 |
cglobal quant_h263_inter_3dne |
cglobal quant_h263_inter_3dne |
492 |
quant_h263_inter_3dne: |
quant_h263_inter_3dne: |
493 |
mov edx, [esp + 4] ; coeff |
mov edx, [esp + 4] ; coeff |
504 |
lea eax, [mmzero] |
lea eax, [mmzero] |
505 |
jz near .q1loop |
jz near .q1loop |
506 |
cmp esp, esp |
cmp esp, esp |
507 |
align 8 |
ALIGN 8 |
508 |
movq mm3, [ecx + 120] ;B1 |
movq mm3, [ecx + 120] ;B1 |
509 |
pxor mm4, mm4 ;B2 |
pxor mm4, mm4 ;B2 |
510 |
psubw mm4, mm3 ;B3 |
psubw mm4, mm3 ;B3 |
535 |
|
|
536 |
ret |
ret |
537 |
|
|
538 |
align ALIGN |
ALIGN 16 |
539 |
.q1loop |
.q1loop |
540 |
movq mm6, [byte ebx] |
movq mm6, [byte ebx] |
541 |
|
|
558 |
|
|
559 |
ret |
ret |
560 |
|
|
561 |
;=========================================================================== |
;----------------------------------------------------------------------------- |
562 |
; |
; |
563 |
; uint32_t dequant_h263_intra_3dne(int16_t *data, |
; uint32_t dequant_h263_intra_3dne(int16_t *data, |
564 |
; const int16_t const *coeff, |
; const int16_t const *coeff, |
565 |
; const uint32_t quant, |
; const uint32_t quant, |
566 |
; const uint32_t dcscalar); |
; const uint32_t dcscalar, |
567 |
|
; const uint16_t *mpeg_matrices); |
568 |
; |
; |
569 |
;=========================================================================== |
;----------------------------------------------------------------------------- |
570 |
|
|
571 |
; this is the same as dequant_inter_3dne, except that we're |
; this is the same as dequant_inter_3dne, except that we're |
572 |
; saturating using 'pminsw' (saves 2 cycles/loop => ~5% faster) |
; saturating using 'pminsw' (saves 2 cycles/loop => ~5% faster) |
592 |
movq mm4, [esi] ;C1 ;0 |
movq mm4, [esi] ;C1 ;0 |
593 |
mov esp, esp |
mov esp, esp |
594 |
pcmpeqw mm6, [ecx+%1*24] ;A6 (c ==0) ? -1 : 0 (1st) |
pcmpeqw mm6, [ecx+%1*24] ;A6 (c ==0) ? -1 : 0 (1st) |
595 |
align 4 |
ALIGN 4 |
596 |
psraw mm1, 15 ; sign(c) ;A7 (2nd) |
psraw mm1, 15 ; sign(c) ;A7 (2nd) |
597 |
%if (%1) |
%if (%1) |
598 |
movq [edx+%1*24+16-24], mm5 ; C14 (7th) 2later |
movq [edx+%1*24+16-24], mm5 ; C14 (7th) 2later |
638 |
%endmacro |
%endmacro |
639 |
|
|
640 |
|
|
641 |
align ALIGN |
ALIGN 16 |
642 |
cglobal dequant_h263_intra_3dne |
cglobal dequant_h263_intra_3dne |
643 |
dequant_h263_intra_3dne: |
dequant_h263_intra_3dne: |
644 |
mov ecx, [esp+ 8] ; coeff |
mov ecx, [esp+ 8] ; coeff |
665 |
psraw mm3, 15 ; sign(c) ;B7 (2nd) |
psraw mm3, 15 ; sign(c) ;B7 (2nd) |
666 |
mov edx, [esp+ 4+16] ; data |
mov edx, [esp+ 4+16] ; data |
667 |
|
|
668 |
align 8 |
ALIGN 8 |
669 |
dequant 0 |
dequant 0 |
670 |
|
|
671 |
cmp ebp, -2048 |
cmp ebp, -2048 |
709 |
xor eax, eax |
xor eax, eax |
710 |
ret |
ret |
711 |
|
|
712 |
;=========================================================================== |
;----------------------------------------------------------------------------- |
713 |
; |
; |
714 |
; uint32_t dequant_h263_inter_3dne(int16_t * data, |
; uint32_t dequant_h263_inter_3dne(int16_t * data, |
715 |
; const int16_t * const coeff, |
; const int16_t * const coeff, |
716 |
; const uint32_t quant); |
; const uint32_t quant, |
717 |
|
; const uint16_t *mpeg_matrices); |
718 |
; |
; |
719 |
;=========================================================================== |
;----------------------------------------------------------------------------- |
720 |
|
|
721 |
; this is the same as dequant_inter_3dne, |
; this is the same as dequant_inter_3dne, |
722 |
; except that we're saturating using 'pminsw' (saves 2 cycles/loop) |
; except that we're saturating using 'pminsw' (saves 2 cycles/loop) |
723 |
; This is Athlon-optimized code (ca 100 clk per call) |
; This is Athlon-optimized code (ca 100 clk per call) |
724 |
|
|
725 |
align ALIGN |
ALIGN 16 |
726 |
cglobal dequant_h263_inter_3dne |
cglobal dequant_h263_inter_3dne |
727 |
dequant_h263_inter_3dne: |
dequant_h263_inter_3dne: |
728 |
mov ecx, [esp+ 8] ; coeff |
mov ecx, [esp+ 8] ; coeff |
745 |
psraw mm3, 15 ; sign(c) ;B7 (2nd) |
psraw mm3, 15 ; sign(c) ;B7 (2nd) |
746 |
mov edx, [dword esp+ 4+12] ; data |
mov edx, [dword esp+ 4+12] ; data |
747 |
|
|
748 |
align 8 |
ALIGN 8 |
749 |
|
|
750 |
dequant 0 |
dequant 0 |
751 |
dequant 1 |
dequant 1 |