1 |
;/************************************************************************** |
;/************************************************************************** |
2 |
; * |
; * |
3 |
; * XVID MPEG-4 VIDEO CODEC |
; * XVID MPEG-4 VIDEO CODEC |
4 |
; * - mmx quantization/dequantization - |
; * - 3dne Quantization/Dequantization - |
5 |
; * |
; * |
6 |
; * Copyright(C) 2001-2003 XviD Team <xvid-devel@xvid.org> |
; * Copyright(C) 2002-2003 Jaan Kalda |
7 |
; * |
; * |
8 |
; * This program is free software ; you can redistribute it and/or modify |
; * This program is free software ; you can redistribute it and/or modify |
9 |
; * it under the terms of the GNU General Public License as published by |
; * it under the terms of the GNU General Public License as published by |
22 |
; * $Id$ |
; * $Id$ |
23 |
; * |
; * |
24 |
; *************************************************************************/ |
; *************************************************************************/ |
25 |
|
; |
26 |
; these 3dne functions are compatible with iSSE, but are optimized specifically for |
; these 3dne functions are compatible with iSSE, but are optimized specifically for |
27 |
; K7 pipelines |
; K7 pipelines |
|
; |
|
|
;------------------------------------------------------------------------------ |
|
|
; 09.12.2002 Athlon optimizations contributed by Jaan Kalda |
|
|
;------------------------------------------------------------------------------ |
|
28 |
|
|
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 |
|
|
|
|
|
%ifdef FORMAT_COFF |
|
|
section .data data |
|
|
%else |
|
|
section .data data align=16 |
|
|
%endif |
|
33 |
|
|
34 |
%macro cglobal 1 |
%macro cglobal 1 |
35 |
%ifdef PREFIX |
%ifdef PREFIX |
36 |
|
%ifdef MARK_FUNCS |
37 |
|
global _%1:function |
38 |
|
%define %1 _%1:function |
39 |
|
%else |
40 |
global _%1 |
global _%1 |
41 |
%define %1 _%1 |
%define %1 _%1 |
42 |
|
%endif |
43 |
|
%else |
44 |
|
%ifdef MARK_FUNCS |
45 |
|
global %1:function |
46 |
%else |
%else |
47 |
global %1 |
global %1 |
48 |
%endif |
%endif |
49 |
|
%endif |
50 |
%endmacro |
%endmacro |
51 |
|
|
52 |
|
;============================================================================= |
53 |
|
; Local data |
54 |
|
;============================================================================= |
55 |
|
|
56 |
|
%ifdef FORMAT_COFF |
57 |
|
SECTION .rodata |
58 |
|
%else |
59 |
|
SECTION .rodata align=16 |
60 |
|
%endif |
61 |
|
|
62 |
align 4 |
align 4 |
63 |
int_div |
int_div: |
64 |
dd 0 |
dd 0 |
65 |
%assign i 1 |
%assign i 1 |
66 |
%rep 255 |
%rep 255 |
68 |
%assign i i+1 |
%assign i i+1 |
69 |
%endrep |
%endrep |
70 |
|
|
71 |
align 16 |
ALIGN 16 |
72 |
|
plus_one: |
73 |
plus_one times 8 dw 1 |
times 8 dw 1 |
74 |
|
|
75 |
;=========================================================================== |
;----------------------------------------------------------------------------- |
|
; |
|
76 |
; subtract by Q/2 table |
; subtract by Q/2 table |
77 |
; |
;----------------------------------------------------------------------------- |
|
;=========================================================================== |
|
|
|
|
|
%macro MMX_SUB 1 |
|
|
times 4 dw %1 / 2 |
|
|
%endmacro |
|
|
|
|
78 |
|
|
79 |
align 16 |
ALIGN 16 |
80 |
mmx_sub |
mmx_sub: |
81 |
%assign i 1 |
%assign i 1 |
82 |
%rep 31 |
%rep 31 |
83 |
times 4 dw i / 2 |
times 4 dw i / 2 |
85 |
%endrep |
%endrep |
86 |
|
|
87 |
|
|
88 |
;=========================================================================== |
;----------------------------------------------------------------------------- |
89 |
; |
; |
90 |
; divide by 2Q table |
; divide by 2Q table |
91 |
; |
; |
93 |
; for q=1, _pmulhw_ will overflow so it is treated seperately |
; for q=1, _pmulhw_ will overflow so it is treated seperately |
94 |
; (3dnow2 provides _pmulhuw_ which wont cause overflow) |
; (3dnow2 provides _pmulhuw_ which wont cause overflow) |
95 |
; |
; |
96 |
;=========================================================================== |
;----------------------------------------------------------------------------- |
|
|
|
|
align 16 |
|
|
mmx_div |
|
97 |
|
|
98 |
|
ALIGN 16 |
99 |
|
mmx_div: |
100 |
%assign i 1 |
%assign i 1 |
101 |
%rep 31 |
%rep 31 |
102 |
times 4 dw (1 << 16) / (i * 2) + 1 |
times 4 dw (1 << 16) / (i * 2) + 1 |
103 |
%assign i i+1 |
%assign i i+1 |
104 |
%endrep |
%endrep |
105 |
|
|
106 |
;=========================================================================== |
;----------------------------------------------------------------------------- |
|
; |
|
107 |
; add by (odd(Q) ? Q : Q - 1) table |
; add by (odd(Q) ? Q : Q - 1) table |
108 |
; |
;----------------------------------------------------------------------------- |
|
;=========================================================================== |
|
|
|
|
|
%macro MMX_ADD 1 |
|
|
%if %1 % 2 != 0 |
|
|
times 4 dw %1 |
|
|
%else |
|
|
times 4 dw %1 - 1 |
|
|
%endif |
|
|
%endmacro |
|
|
|
|
|
align 16 |
|
|
mmx_add |
|
109 |
|
|
110 |
|
ALIGN 16 |
111 |
|
mmx_add: |
112 |
%assign i 1 |
%assign i 1 |
113 |
%rep 31 |
%rep 31 |
114 |
MMX_ADD i |
%if i % 2 != 0 |
115 |
|
times 4 dw i |
116 |
|
%else |
117 |
|
times 4 dw i - 1 |
118 |
|
%endif |
119 |
%assign i i+1 |
%assign i i+1 |
120 |
%endrep |
%endrep |
121 |
|
|
122 |
;=========================================================================== |
;----------------------------------------------------------------------------- |
|
; |
|
123 |
; multiple by 2Q table |
; multiple by 2Q table |
124 |
; |
;----------------------------------------------------------------------------- |
|
;=========================================================================== |
|
|
|
|
|
%macro MMX_MUL 1 |
|
|
times 4 dw %1 * 2 |
|
|
%endmacro |
|
|
|
|
|
align 16 |
|
|
mmx_mul |
|
125 |
|
|
126 |
|
ALIGN 16 |
127 |
|
mmx_mul: |
128 |
%assign i 1 |
%assign i 1 |
129 |
%rep 31 |
%rep 31 |
130 |
times 4 dw i * 2 |
times 4 dw i * 2 |
131 |
%assign i i+1 |
%assign i i+1 |
132 |
%endrep |
%endrep |
133 |
|
|
134 |
;=========================================================================== |
;----------------------------------------------------------------------------- |
|
; |
|
135 |
; saturation limits |
; saturation limits |
136 |
; |
;----------------------------------------------------------------------------- |
137 |
;=========================================================================== |
|
138 |
|
ALIGN 8 |
139 |
|
mmx_32768_minus_2048: |
140 |
|
times 4 dw (32768-2048) |
141 |
|
mmx_32767_minus_2047: |
142 |
|
times 4 dw (32767-2047) |
143 |
|
|
144 |
align 8 |
ALIGN 16 |
145 |
mmx_32768_minus_2048 times 4 dw (32768-2048) |
mmx_2047: |
146 |
mmx_32767_minus_2047 times 4 dw (32767-2047) |
times 4 dw 2047 |
147 |
|
|
148 |
align 16 |
ALIGN 8 |
149 |
mmx_2047 times 4 dw 2047 |
mmzero: |
150 |
|
dd 0, 0 |
151 |
|
int2047: |
152 |
|
dd 2047 |
153 |
|
int_2048: |
154 |
|
dd -2048 |
155 |
|
|
156 |
align 8 |
;============================================================================= |
157 |
mmzero dd 0, 0 |
; Code |
158 |
int2047 dd 2047 |
;============================================================================= |
|
int_2048 dd -2048 |
|
159 |
|
|
160 |
section .text |
SECTION .text |
161 |
|
|
162 |
|
|
163 |
;=========================================================================== |
;----------------------------------------------------------------------------- |
164 |
; |
; |
165 |
; void quant_intra_3dne(int16_t * coeff, |
; uint32_t quant_h263_intra_3dne(int16_t * coeff, |
166 |
; const int16_t const * data, |
; const int16_t const * data, |
167 |
; const uint32_t quant, |
; const uint32_t quant, |
168 |
; const uint32_t dcscalar); |
; const uint32_t dcscalar, |
169 |
|
; const uint16_t *mpeg_matrices); |
170 |
; |
; |
171 |
;=========================================================================== |
;----------------------------------------------------------------------------- |
172 |
;This is Athlon-optimized code (ca 70 clk per call) |
;This is Athlon-optimized code (ca 70 clk per call) |
|
;Optimized by Jaan, 30 Nov 2002 |
|
173 |
|
|
174 |
%macro quant_intra1 1 |
%macro quant_intra1 1 |
|
|
|
175 |
psubw mm1, mm0 ;A3 |
psubw mm1, mm0 ;A3 |
176 |
psubw mm3, mm2 ;B3 |
psubw mm3, mm2 ;B3 |
177 |
%if (%1) |
%if (%1) |
179 |
psubw mm7, mm6 ;D8 |
psubw mm7, mm6 ;D8 |
180 |
%endif |
%endif |
181 |
|
|
182 |
align 8 |
ALIGN 8 |
183 |
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 |
184 |
pmaxsw mm1, mm0 ;A4 |
pmaxsw mm1, mm0 ;A4 |
185 |
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 |
253 |
psubw mm7, mm6 ;D8 |
psubw mm7, mm6 ;D8 |
254 |
%endif |
%endif |
255 |
|
|
256 |
align 8 |
ALIGN 8 |
257 |
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 |
258 |
pmaxsw mm1, mm0 ;A4 |
pmaxsw mm1, mm0 ;A4 |
259 |
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 |
320 |
%endmacro |
%endmacro |
321 |
|
|
322 |
|
|
323 |
align ALIGN |
ALIGN 16 |
324 |
cglobal quant_h263_intra_3dne |
cglobal quant_h263_intra_3dne |
325 |
quant_h263_intra_3dne: |
quant_h263_intra_3dne: |
326 |
|
|
372 |
movq [edx + 3 * 32 + 16], mm5 ;C9 |
movq [edx + 3 * 32 + 16], mm5 ;C9 |
373 |
movq [edx + 3 * 32 + 24], mm7 ;D9 |
movq [edx + 3 * 32 + 24], mm7 ;D9 |
374 |
|
|
375 |
|
xor eax, eax |
376 |
ret |
ret |
377 |
|
|
378 |
align 16 |
ALIGN 16 |
379 |
|
|
380 |
.q1loop |
.q1loop |
381 |
quant_intra1 0 |
quant_intra1 0 |
408 |
movq [edx + 3 * 32 + 16], mm5 ;C9 |
movq [edx + 3 * 32 + 16], mm5 ;C9 |
409 |
movq [edx + 3 * 32 + 24], mm7 ;D9 |
movq [edx + 3 * 32 + 24], mm7 ;D9 |
410 |
|
|
411 |
|
xor eax, eax |
412 |
ret |
ret |
413 |
|
|
414 |
|
|
415 |
|
|
416 |
|
|
417 |
;=========================================================================== |
;----------------------------------------------------------------------------- |
418 |
; |
; |
419 |
; uint32_t quant_inter_3dne(int16_t * coeff, |
; uint32_t quant_h263_inter_3dne(int16_t * coeff, |
420 |
; const int16_t const * data, |
; const int16_t const * data, |
421 |
; const uint32_t quant); |
; const uint32_t quant, |
422 |
|
; const uint16_t *mpeg_matrices); |
423 |
; |
; |
424 |
;=========================================================================== |
;----------------------------------------------------------------------------- |
425 |
;This is Athlon-optimized code (ca 90 clk per call) |
;This is Athlon-optimized code (ca 90 clk per call) |
426 |
;Optimized by Jaan, 30 Nov 2002 |
;Optimized by Jaan, 30 Nov 2002 |
427 |
|
|
496 |
movq [edx + %1*16+8], mm4 |
movq [edx + %1*16+8], mm4 |
497 |
%endmacro |
%endmacro |
498 |
|
|
499 |
align ALIGN |
ALIGN 16 |
500 |
cglobal quant_h263_inter_3dne |
cglobal quant_h263_inter_3dne |
501 |
quant_h263_inter_3dne: |
quant_h263_inter_3dne: |
502 |
mov edx, [esp + 4] ; coeff |
mov edx, [esp + 4] ; coeff |
513 |
lea eax, [mmzero] |
lea eax, [mmzero] |
514 |
jz near .q1loop |
jz near .q1loop |
515 |
cmp esp, esp |
cmp esp, esp |
516 |
align 8 |
ALIGN 8 |
517 |
movq mm3, [ecx + 120] ;B1 |
movq mm3, [ecx + 120] ;B1 |
518 |
pxor mm4, mm4 ;B2 |
pxor mm4, mm4 ;B2 |
519 |
psubw mm4, mm3 ;B3 |
psubw mm4, mm3 ;B3 |
544 |
|
|
545 |
ret |
ret |
546 |
|
|
547 |
align ALIGN |
ALIGN 16 |
548 |
.q1loop |
.q1loop |
549 |
movq mm6, [byte ebx] |
movq mm6, [byte ebx] |
550 |
|
|
567 |
|
|
568 |
ret |
ret |
569 |
|
|
570 |
;=========================================================================== |
;----------------------------------------------------------------------------- |
571 |
; |
; |
572 |
; void dequant_intra_3dne(int16_t *data, |
; uint32_t dequant_h263_intra_3dne(int16_t *data, |
573 |
; const int16_t const *coeff, |
; const int16_t const *coeff, |
574 |
; const uint32_t quant, |
; const uint32_t quant, |
575 |
; const uint32_t dcscalar); |
; const uint32_t dcscalar, |
576 |
|
; const uint16_t *mpeg_matrices); |
577 |
; |
; |
578 |
;=========================================================================== |
;----------------------------------------------------------------------------- |
579 |
|
|
580 |
; this is the same as dequant_inter_3dne, except that we're |
; this is the same as dequant_inter_3dne, except that we're |
581 |
; saturating using 'pminsw' (saves 2 cycles/loop => ~5% faster) |
; saturating using 'pminsw' (saves 2 cycles/loop => ~5% faster) |
601 |
movq mm4, [esi] ;C1 ;0 |
movq mm4, [esi] ;C1 ;0 |
602 |
mov esp, esp |
mov esp, esp |
603 |
pcmpeqw mm6, [ecx+%1*24] ;A6 (c ==0) ? -1 : 0 (1st) |
pcmpeqw mm6, [ecx+%1*24] ;A6 (c ==0) ? -1 : 0 (1st) |
604 |
align 4 |
ALIGN 4 |
605 |
psraw mm1, 15 ; sign(c) ;A7 (2nd) |
psraw mm1, 15 ; sign(c) ;A7 (2nd) |
606 |
%if (%1) |
%if (%1) |
607 |
movq [edx+%1*24+16-24], mm5 ; C14 (7th) 2later |
movq [edx+%1*24+16-24], mm5 ; C14 (7th) 2later |
647 |
%endmacro |
%endmacro |
648 |
|
|
649 |
|
|
650 |
align ALIGN |
ALIGN 16 |
651 |
cglobal dequant_h263_intra_3dne |
cglobal dequant_h263_intra_3dne |
652 |
dequant_h263_intra_3dne: |
dequant_h263_intra_3dne: |
653 |
mov ecx, [esp+ 8] ; coeff |
mov ecx, [esp+ 8] ; coeff |
674 |
psraw mm3, 15 ; sign(c) ;B7 (2nd) |
psraw mm3, 15 ; sign(c) ;B7 (2nd) |
675 |
mov edx, [esp+ 4+16] ; data |
mov edx, [esp+ 4+16] ; data |
676 |
|
|
677 |
align 8 |
ALIGN 8 |
678 |
dequant 0 |
dequant 0 |
679 |
|
|
680 |
cmp ebp, -2048 |
cmp ebp, -2048 |
714 |
pxor mm3, mm2 ;B13 (6th+) |
pxor mm3, mm2 ;B13 (6th+) |
715 |
movq [edx+4*24+8], mm3 ;B14 (7th) |
movq [edx+4*24+8], mm3 ;B14 (7th) |
716 |
mov [edx], ax |
mov [edx], ax |
717 |
|
|
718 |
|
xor eax, eax |
719 |
ret |
ret |
720 |
|
|
721 |
;=========================================================================== |
;----------------------------------------------------------------------------- |
722 |
; |
; |
723 |
; void dequant_inter_3dne(int16_t * data, |
; uint32_t dequant_h263_inter_3dne(int16_t * data, |
724 |
; const int16_t * const coeff, |
; const int16_t * const coeff, |
725 |
; const uint32_t quant); |
; const uint32_t quant, |
726 |
|
; const uint16_t *mpeg_matrices); |
727 |
; |
; |
728 |
;=========================================================================== |
;----------------------------------------------------------------------------- |
729 |
|
|
730 |
; this is the same as dequant_inter_3dne, |
; this is the same as dequant_inter_3dne, |
731 |
; except that we're saturating using 'pminsw' (saves 2 cycles/loop) |
; except that we're saturating using 'pminsw' (saves 2 cycles/loop) |
732 |
;This is Athlon-optimized code (ca 100 clk per call) |
;This is Athlon-optimized code (ca 100 clk per call) |
|
;Optimized by Jaan, 30 Nov 2002 |
|
733 |
|
|
734 |
align ALIGN |
ALIGN 16 |
735 |
cglobal dequant_h263_inter_3dne |
cglobal dequant_h263_inter_3dne |
736 |
dequant_h263_inter_3dne: |
dequant_h263_inter_3dne: |
737 |
mov ecx, [esp+ 8] ; coeff |
mov ecx, [esp+ 8] ; coeff |
754 |
psraw mm3, 15 ; sign(c) ;B7 (2nd) |
psraw mm3, 15 ; sign(c) ;B7 (2nd) |
755 |
mov edx, [dword esp+ 4+12] ; data |
mov edx, [dword esp+ 4+12] ; data |
756 |
|
|
757 |
align 8 |
ALIGN 8 |
758 |
|
|
759 |
dequant 0 |
dequant 0 |
760 |
dequant 1 |
dequant 1 |
777 |
pxor mm3, mm2 ;B13 (6th+) |
pxor mm3, mm2 ;B13 (6th+) |
778 |
movq [edx+4*24+8], mm3 ;B14 (7th) |
movq [edx+4*24+8], mm3 ;B14 (7th) |
779 |
|
|
780 |
|
xor eax, eax |
781 |
ret |
ret |