31 |
|
|
32 |
%macro cglobal 1 |
%macro cglobal 1 |
33 |
%ifdef PREFIX |
%ifdef PREFIX |
34 |
|
%ifdef MARK_FUNCS |
35 |
|
global _%1:function %1.endfunc-%1 |
36 |
|
%define %1 _%1:function %1.endfunc-%1 |
37 |
|
%else |
38 |
global _%1 |
global _%1 |
39 |
%define %1 _%1 |
%define %1 _%1 |
40 |
|
%endif |
41 |
|
%else |
42 |
|
%ifdef MARK_FUNCS |
43 |
|
global %1:function %1.endfunc-%1 |
44 |
%else |
%else |
45 |
global %1 |
global %1 |
46 |
%endif |
%endif |
47 |
|
%endif |
48 |
%endmacro |
%endmacro |
49 |
|
|
50 |
%macro cextern 1 |
%macro cextern 1 |
60 |
; Local data (Read Only) |
; Local data (Read Only) |
61 |
;============================================================================= |
;============================================================================= |
62 |
|
|
63 |
|
%ifdef FORMAT_COFF |
64 |
SECTION .rodata |
SECTION .rodata |
65 |
|
%else |
66 |
|
SECTION .rodata align=16 |
67 |
|
%endif |
68 |
|
|
69 |
mmx_one: |
mmx_one: |
70 |
times 4 dw 1 |
times 4 dw 1 |
85 |
%assign quant quant+1 |
%assign quant quant+1 |
86 |
%endrep |
%endrep |
87 |
|
|
|
;----------------------------------------------------------------------------- |
|
|
; intra matrix |
|
|
;----------------------------------------------------------------------------- |
|
|
|
|
|
cextern intra_matrix |
|
|
cextern intra_matrix_fix |
|
|
|
|
|
;----------------------------------------------------------------------------- |
|
|
; inter matrix |
|
|
;----------------------------------------------------------------------------- |
|
|
|
|
|
cextern inter_matrix |
|
|
cextern inter_matrix_fix |
|
|
|
|
|
|
|
88 |
%define VM18P 3 |
%define VM18P 3 |
89 |
%define VM18Q 4 |
%define VM18Q 4 |
90 |
|
|
144 |
; uint32_t quant_mpeg_intra_mmx(int16_t * coeff, |
; uint32_t quant_mpeg_intra_mmx(int16_t * coeff, |
145 |
; const int16_t const * data, |
; const int16_t const * data, |
146 |
; const uint32_t quant, |
; const uint32_t quant, |
147 |
; const uint32_t dcscalar); |
; const uint32_t dcscalar, |
148 |
|
; const uint16_t *mpeg_matrices); |
149 |
; |
; |
150 |
;----------------------------------------------------------------------------- |
;----------------------------------------------------------------------------- |
151 |
|
|
155 |
push ecx |
push ecx |
156 |
push esi |
push esi |
157 |
push edi |
push edi |
158 |
|
push ebx |
159 |
|
|
160 |
mov edi, [esp + 12 + 4] ; coeff |
mov edi, [esp + 16 + 4] ; coeff |
161 |
mov esi, [esp + 12 + 8] ; data |
mov esi, [esp + 16 + 8] ; data |
162 |
mov eax, [esp + 12 + 12] ; quant |
mov eax, [esp + 16 + 12] ; quant |
163 |
|
mov ebx, [esp + 16 + 20] ; mpeg_quant_matrices |
164 |
|
|
165 |
movq mm5, [quantd + eax * 8 - 8] ; quantd -> mm5 |
movq mm5, [quantd + eax * 8 - 8] ; quantd -> mm5 |
166 |
|
|
187 |
psubw mm3, mm4 ; |
psubw mm3, mm4 ; |
188 |
psllw mm0, 4 ; level << 4 |
psllw mm0, 4 ; level << 4 |
189 |
psllw mm3, 4 |
psllw mm3, 4 |
190 |
movq mm2, [intra_matrix + 8*ecx] |
movq mm2, [ebx + 8*ecx] |
191 |
psrlw mm2, 1 ; intra_matrix[i]>>1 |
psrlw mm2, 1 ; intra_matrix[i]>>1 |
192 |
paddw mm0, mm2 |
paddw mm0, mm2 |
193 |
movq mm2, [intra_matrix_fix + ecx*8] |
movq mm2, [ebx + 256 + ecx*8] |
194 |
pmulhw mm0, mm2 ; (level<<4 + intra_matrix[i]>>1) / intra_matrix[i] |
pmulhw mm0, mm2 ; (level<<4 + intra_matrix[i]>>1) / intra_matrix[i] |
195 |
movq mm2, [intra_matrix + 8*ecx + 8] |
movq mm2, [ebx + 8*ecx + 8] |
196 |
psrlw mm2, 1 |
psrlw mm2, 1 |
197 |
paddw mm3, mm2 |
paddw mm3, mm2 |
198 |
movq mm2, [intra_matrix_fix + ecx*8 + 8] |
movq mm2, [ebx + 256 + ecx*8 + 8] |
199 |
pmulhw mm3, mm2 |
pmulhw mm3, mm2 |
200 |
paddw mm0, mm5 ; + quantd |
paddw mm0, mm5 ; + quantd |
201 |
paddw mm3, mm5 |
paddw mm3, mm5 |
217 |
|
|
218 |
.done |
.done |
219 |
; caclulate data[0] // (int32_t)dcscalar) |
; caclulate data[0] // (int32_t)dcscalar) |
220 |
mov ecx, [esp + 12 + 16] ; dcscalar |
mov ecx, [esp + 16 + 16] ; dcscalar |
221 |
mov edx, ecx |
mov edx, ecx |
222 |
movsx eax, word [esi] ; data[0] |
movsx eax, word [esi] ; data[0] |
223 |
shr edx, 1 ; edx = dcscalar /2 |
shr edx, 1 ; edx = dcscalar /2 |
234 |
|
|
235 |
mov [edi], ax ; coeff[0] = ax |
mov [edi], ax ; coeff[0] = ax |
236 |
|
|
237 |
|
pop ebx |
238 |
pop edi |
pop edi |
239 |
pop esi |
pop esi |
240 |
pop ecx |
pop ecx |
256 |
psubw mm3, mm4 ; |
psubw mm3, mm4 ; |
257 |
psllw mm0, 4 |
psllw mm0, 4 |
258 |
psllw mm3, 4 |
psllw mm3, 4 |
259 |
movq mm2, [intra_matrix + 8*ecx] |
movq mm2, [ebx + 8*ecx] |
260 |
psrlw mm2, 1 |
psrlw mm2, 1 |
261 |
paddw mm0, mm2 |
paddw mm0, mm2 |
262 |
movq mm2, [intra_matrix_fix + ecx*8] |
movq mm2, [ebx + 256 + ecx*8] |
263 |
pmulhw mm0, mm2 ; (level<<4 + intra_matrix[i]>>1) / intra_matrix[i] |
pmulhw mm0, mm2 ; (level<<4 + intra_matrix[i]>>1) / intra_matrix[i] |
264 |
movq mm2, [intra_matrix + 8*ecx + 8] |
movq mm2, [ebx + 8*ecx + 8] |
265 |
psrlw mm2, 1 |
psrlw mm2, 1 |
266 |
paddw mm3, mm2 |
paddw mm3, mm2 |
267 |
movq mm2, [intra_matrix_fix + ecx*8 + 8] |
movq mm2, [ebx + 256 + ecx*8 + 8] |
268 |
pmulhw mm3, mm2 |
pmulhw mm3, mm2 |
269 |
paddw mm0, mm5 |
paddw mm0, mm5 |
270 |
paddw mm3, mm5 |
paddw mm3, mm5 |
297 |
psubw mm3, mm4 ; |
psubw mm3, mm4 ; |
298 |
psllw mm0, 4 |
psllw mm0, 4 |
299 |
psllw mm3, 4 |
psllw mm3, 4 |
300 |
movq mm2, [intra_matrix + 8*ecx] |
movq mm2, [ebx + 8*ecx] |
301 |
psrlw mm2, 1 |
psrlw mm2, 1 |
302 |
paddw mm0, mm2 |
paddw mm0, mm2 |
303 |
movq mm2, [intra_matrix_fix + ecx*8] |
movq mm2, [ebx + 256 + ecx*8] |
304 |
pmulhw mm0, mm2 ; (level<<4 + intra_matrix[i]>>1) / intra_matrix[i] |
pmulhw mm0, mm2 ; (level<<4 + intra_matrix[i]>>1) / intra_matrix[i] |
305 |
movq mm2, [intra_matrix + 8*ecx + 8] |
movq mm2, [ebx + 8*ecx + 8] |
306 |
psrlw mm2, 1 |
psrlw mm2, 1 |
307 |
paddw mm3, mm2 |
paddw mm3, mm2 |
308 |
movq mm2, [intra_matrix_fix + ecx*8 + 8] |
movq mm2, [ebx + 256 + ecx*8 + 8] |
309 |
pmulhw mm3, mm2 |
pmulhw mm3, mm2 |
310 |
paddw mm0, mm5 |
paddw mm0, mm5 |
311 |
paddw mm3, mm5 |
paddw mm3, mm5 |
322 |
cmp ecx,16 |
cmp ecx,16 |
323 |
jnz near .q2loop |
jnz near .q2loop |
324 |
jmp near .done |
jmp near .done |
325 |
|
.endfunc |
326 |
|
|
327 |
;----------------------------------------------------------------------------- |
;----------------------------------------------------------------------------- |
328 |
; |
; |
329 |
; uint32_t quant_mpeg_inter_mmx(int16_t * coeff, |
; uint32_t quant_mpeg_inter_mmx(int16_t * coeff, |
330 |
; const int16_t const * data, |
; const int16_t const * data, |
331 |
; const uint32_t quant); |
; const uint32_t quant, |
332 |
|
; const uint16_t *mpeg_matrices); |
333 |
; |
; |
334 |
;----------------------------------------------------------------------------- |
;----------------------------------------------------------------------------- |
335 |
|
|
339 |
push ecx |
push ecx |
340 |
push esi |
push esi |
341 |
push edi |
push edi |
342 |
|
push ebx |
343 |
|
|
344 |
mov edi, [esp + 12 + 4] ; coeff |
mov edi, [esp + 16 + 4] ; coeff |
345 |
mov esi, [esp + 12 + 8] ; data |
mov esi, [esp + 16 + 8] ; data |
346 |
mov eax, [esp + 12 + 12] ; quant |
mov eax, [esp + 16 + 12] ; quant |
347 |
|
mov ebx, [esp + 16 + 16] ; mpeg_quant_matrices |
348 |
|
|
349 |
xor ecx, ecx |
xor ecx, ecx |
350 |
|
|
372 |
psubw mm3, mm4 ; |
psubw mm3, mm4 ; |
373 |
psllw mm0, 4 |
psllw mm0, 4 |
374 |
psllw mm3, 4 |
psllw mm3, 4 |
375 |
movq mm2, [inter_matrix + 8*ecx] |
movq mm2, [ebx + 512 + 8*ecx] |
376 |
psrlw mm2, 1 |
psrlw mm2, 1 |
377 |
paddw mm0, mm2 |
paddw mm0, mm2 |
378 |
movq mm2, [inter_matrix_fix + ecx*8] |
movq mm2, [ebx + 768 + ecx*8] |
379 |
pmulhw mm0, mm2 ; (level<<4 + inter_matrix[i]>>1) / inter_matrix[i] |
pmulhw mm0, mm2 ; (level<<4 + inter_matrix[i]>>1) / inter_matrix[i] |
380 |
movq mm2, [inter_matrix + 8*ecx + 8] |
movq mm2, [ebx + 512 + 8*ecx + 8] |
381 |
psrlw mm2, 1 |
psrlw mm2, 1 |
382 |
paddw mm3, mm2 |
paddw mm3, mm2 |
383 |
movq mm2, [inter_matrix_fix + ecx*8 + 8] |
movq mm2, [ebx + 768 + ecx*8 + 8] |
384 |
pmulhw mm3, mm2 |
pmulhw mm3, mm2 |
385 |
pmulhw mm0, mm7 ; mm0 = (mm0 / 2Q) >> 16 |
pmulhw mm0, mm7 ; mm0 = (mm0 / 2Q) >> 16 |
386 |
pmulhw mm3, mm7 ; |
pmulhw mm3, mm7 ; |
406 |
paddd mm0, mm5 |
paddd mm0, mm5 |
407 |
movd eax, mm0 ; return sum |
movd eax, mm0 ; return sum |
408 |
|
|
409 |
|
pop ebx |
410 |
pop edi |
pop edi |
411 |
pop esi |
pop esi |
412 |
pop ecx |
pop ecx |
427 |
psubw mm3, mm4 ; |
psubw mm3, mm4 ; |
428 |
psllw mm0, 4 |
psllw mm0, 4 |
429 |
psllw mm3, 4 |
psllw mm3, 4 |
430 |
movq mm2, [inter_matrix + 8*ecx] |
movq mm2, [ebx + 512 + 8*ecx] |
431 |
psrlw mm2, 1 |
psrlw mm2, 1 |
432 |
paddw mm0, mm2 |
paddw mm0, mm2 |
433 |
movq mm2, [inter_matrix_fix + ecx*8] |
movq mm2, [ebx + 768 + ecx*8] |
434 |
pmulhw mm0, mm2 ; (level<<4 + inter_matrix[i]>>1) / inter_matrix[i] |
pmulhw mm0, mm2 ; (level<<4 + inter_matrix[i]>>1) / inter_matrix[i] |
435 |
movq mm2, [inter_matrix + 8*ecx + 8] |
movq mm2, [ebx + 512 + 8*ecx + 8] |
436 |
psrlw mm2, 1 |
psrlw mm2, 1 |
437 |
paddw mm3, mm2 |
paddw mm3, mm2 |
438 |
movq mm2, [inter_matrix_fix + ecx*8 + 8] |
movq mm2, [ebx + 768 + ecx*8 + 8] |
439 |
pmulhw mm3, mm2 |
pmulhw mm3, mm2 |
440 |
psrlw mm0, 1 ; mm0 >>= 1 (/2) |
psrlw mm0, 1 ; mm0 >>= 1 (/2) |
441 |
psrlw mm3, 1 ; |
psrlw mm3, 1 ; |
454 |
|
|
455 |
jmp .done |
jmp .done |
456 |
|
|
|
|
|
457 |
ALIGN 16 |
ALIGN 16 |
458 |
.q2loop |
.q2loop |
459 |
movq mm0, [esi + 8*ecx] ; mm0 = [1st] |
movq mm0, [esi + 8*ecx] ; mm0 = [1st] |
468 |
psubw mm3, mm4 ; |
psubw mm3, mm4 ; |
469 |
psllw mm0, 4 |
psllw mm0, 4 |
470 |
psllw mm3, 4 |
psllw mm3, 4 |
471 |
movq mm2, [inter_matrix + 8*ecx] |
movq mm2, [ebx + 512 + 8*ecx] |
472 |
psrlw mm2, 1 |
psrlw mm2, 1 |
473 |
paddw mm0, mm2 |
paddw mm0, mm2 |
474 |
movq mm2, [inter_matrix_fix + ecx*8] |
movq mm2, [ebx + 768 + ecx*8] |
475 |
pmulhw mm0, mm2 ; (level<<4 + inter_matrix[i]>>1) / inter_matrix[i] |
pmulhw mm0, mm2 ; (level<<4 + inter_matrix[i]>>1) / inter_matrix[i] |
476 |
movq mm2, [inter_matrix + 8*ecx + 8] |
movq mm2, [ebx + 512 + 8*ecx + 8] |
477 |
psrlw mm2, 1 |
psrlw mm2, 1 |
478 |
paddw mm3, mm2 |
paddw mm3, mm2 |
479 |
movq mm2, [inter_matrix_fix + ecx*8 + 8] |
movq mm2, [ebx + 768 + ecx*8 + 8] |
480 |
pmulhw mm3, mm2 |
pmulhw mm3, mm2 |
481 |
psrlw mm0, 2 ; mm0 >>= 1 (/2) |
psrlw mm0, 2 ; mm0 >>= 1 (/2) |
482 |
psrlw mm3, 2 ; |
psrlw mm3, 2 ; |
494 |
jnz near .q2loop |
jnz near .q2loop |
495 |
|
|
496 |
jmp .done |
jmp .done |
497 |
|
.endfunc |
498 |
|
|
499 |
|
|
500 |
;----------------------------------------------------------------------------- |
;----------------------------------------------------------------------------- |
502 |
; uint32_t dequant_mpeg_intra_mmx(int16_t *data, |
; uint32_t dequant_mpeg_intra_mmx(int16_t *data, |
503 |
; const int16_t const *coeff, |
; const int16_t const *coeff, |
504 |
; const uint32_t quant, |
; const uint32_t quant, |
505 |
; const uint32_t dcscalar); |
; const uint32_t dcscalar, |
506 |
|
; const uint16_t *mpeg_matrices); |
507 |
; |
; |
508 |
;----------------------------------------------------------------------------- |
;----------------------------------------------------------------------------- |
509 |
|
|
526 |
psubw mm0, mm1 ; -> mm0 = abs(coeff[i]), mm1 = sign of coeff[i] |
psubw mm0, mm1 ; -> mm0 = abs(coeff[i]), mm1 = sign of coeff[i] |
527 |
|
|
528 |
movq mm2, mm7 ; mm2 = quant |
movq mm2, mm7 ; mm2 = quant |
529 |
pmullw mm2, [intra_matrix + 8*eax + 8*16 ] ; matrix[i]*quant. |
pmullw mm2, [ebx + 8*eax + 8*16 ] ; matrix[i]*quant. |
530 |
|
|
531 |
movq mm6, mm2 |
movq mm6, mm2 |
532 |
pmulhw mm2, mm0 ; high of coeff*(matrix*quant) (should be 0 if no overflow) |
pmulhw mm2, mm0 ; high of coeff*(matrix*quant) (should be 0 if no overflow) |
549 |
ALIGN 16 |
ALIGN 16 |
550 |
dequant_mpeg_intra_mmx: |
dequant_mpeg_intra_mmx: |
551 |
|
|
552 |
mov edx, [esp+4] ; data |
push ebx |
553 |
mov ecx, [esp+8] ; coeff |
|
554 |
mov eax, [esp+12] ; quant |
mov edx, [esp + 4 + 4] ; data |
555 |
|
mov ecx, [esp + 4 + 8] ; coeff |
556 |
|
mov eax, [esp + 4 + 12] ; quant |
557 |
|
mov ebx, [esp + 4 + 20] ; mpeg_quant_matrices |
558 |
|
|
559 |
movq mm7, [mmx_mul_quant + eax*8 - 8] |
movq mm7, [mmx_mul_quant + eax*8 - 8] |
560 |
mov eax, -16 ; to keep ALIGNed, we regularly process coeff[0] |
mov eax, -16 ; to keep ALIGNed, we regularly process coeff[0] |
571 |
movq mm2, mm7 ; mm2 = quant |
movq mm2, mm7 ; mm2 = quant |
572 |
|
|
573 |
pcmpgtw mm4, mm3 ; mm4 = sgn(c') |
pcmpgtw mm4, mm3 ; mm4 = sgn(c') |
574 |
pmullw mm2, [intra_matrix + 8*eax + 8*16 ] ; matrix[i]*quant |
pmullw mm2, [ebx + 8*eax + 8*16 ] ; matrix[i]*quant |
575 |
|
|
576 |
pxor mm0, mm1 ; negate if negative |
pxor mm0, mm1 ; negate if negative |
577 |
pxor mm3, mm4 ; negate if negative |
pxor mm3, mm4 ; negate if negative |
587 |
pmulhw mm0, mm5 ; high of coeff*(matrix*quant) |
pmulhw mm0, mm5 ; high of coeff*(matrix*quant) |
588 |
movq mm5, mm7 ; mm2 = quant |
movq mm5, mm7 ; mm2 = quant |
589 |
|
|
590 |
pmullw mm5, [intra_matrix + 8*eax + 8*16 +8] ; matrix[i+1]*quant |
pmullw mm5, [ebx + 8*eax + 8*16 +8] ; matrix[i+1]*quant |
591 |
|
|
592 |
movq mm6, mm5 |
movq mm6, mm5 |
593 |
add eax,2 ; z-flag will be tested later |
add eax,2 ; z-flag will be tested later |
619 |
|
|
620 |
; deal with DC |
; deal with DC |
621 |
movd mm0, [ecx] |
movd mm0, [ecx] |
622 |
pmullw mm0, [esp+16] ; dcscalar |
pmullw mm0, [esp + 4 + 16] ; dcscalar |
623 |
movq mm2, [mmx_32767_minus_2047] |
movq mm2, [mmx_32767_minus_2047] |
624 |
paddsw mm0, mm2 |
paddsw mm0, mm2 |
625 |
psubsw mm0, mm2 |
psubsw mm0, mm2 |
630 |
mov [edx], ax |
mov [edx], ax |
631 |
|
|
632 |
xor eax, eax |
xor eax, eax |
633 |
|
|
634 |
|
pop ebx |
635 |
|
|
636 |
ret |
ret |
637 |
|
.endfunc |
638 |
|
|
639 |
;----------------------------------------------------------------------------- |
;----------------------------------------------------------------------------- |
640 |
; |
; |
641 |
; uint32_t dequant_mpeg_inter_mmx(int16_t * data, |
; uint32_t dequant_mpeg_inter_mmx(int16_t * data, |
642 |
; const int16_t * const coeff, |
; const int16_t * const coeff, |
643 |
; const uint32_t quant); |
; const uint32_t quant, |
644 |
|
; const uint16_t *mpeg_matrices); |
645 |
; |
; |
646 |
;----------------------------------------------------------------------------- |
;----------------------------------------------------------------------------- |
647 |
|
|
653 |
ALIGN 16 |
ALIGN 16 |
654 |
dequant_mpeg_inter_mmx: |
dequant_mpeg_inter_mmx: |
655 |
|
|
656 |
mov edx, [esp+ 4] ; data |
push ebx |
657 |
mov ecx, [esp+ 8] ; coeff |
|
658 |
mov eax, [esp+12] ; quant |
mov edx, [esp + 4 + 4] ; data |
659 |
|
mov ecx, [esp + 4 + 8] ; coeff |
660 |
|
mov eax, [esp + 4 + 12] ; quant |
661 |
|
mov ebx, [esp + 4 + 16] ; mpeg_quant_matrices |
662 |
|
|
663 |
movq mm7, [mmx_mul_quant + eax*8 - 8] |
movq mm7, [mmx_mul_quant + eax*8 - 8] |
664 |
mov eax, -16 |
mov eax, -16 |
665 |
paddw mm7, mm7 ; << 1 |
paddw mm7, mm7 ; << 1 |
694 |
; we're short on register, here. Poor pairing... |
; we're short on register, here. Poor pairing... |
695 |
|
|
696 |
movq mm4, mm7 ; (matrix*quant) |
movq mm4, mm7 ; (matrix*quant) |
697 |
pmullw mm4, [inter_matrix + 8*eax + 8*16 -2*8] |
pmullw mm4, [ebx + 512 + 8*eax + 8*16 -2*8] |
698 |
movq mm5, mm4 |
movq mm5, mm4 |
699 |
pmulhw mm5, mm0 ; high of c*(matrix*quant) |
pmulhw mm5, mm0 ; high of c*(matrix*quant) |
700 |
pmullw mm0, mm4 ; low of c*(matrix*quant) |
pmullw mm0, mm4 ; low of c*(matrix*quant) |
701 |
|
|
702 |
movq mm4, mm7 ; (matrix*quant) |
movq mm4, mm7 ; (matrix*quant) |
703 |
pmullw mm4, [inter_matrix + 8*eax + 8*16 -2*8 + 8] |
pmullw mm4, [ebx + 512 + 8*eax + 8*16 -2*8 + 8] |
704 |
|
|
705 |
pcmpgtw mm5, [zero] |
pcmpgtw mm5, [zero] |
706 |
paddusw mm0, mm5 |
paddusw mm0, mm5 |
744 |
xor word [edx + 2*63], ax |
xor word [edx + 2*63], ax |
745 |
|
|
746 |
xor eax, eax |
xor eax, eax |
747 |
|
|
748 |
|
pop ebx |
749 |
|
|
750 |
ret |
ret |
751 |
|
.endfunc |
752 |
|
|