21 |
; * along with this program ; if not, write to the Free Software |
; * along with this program ; if not, write to the Free Software |
22 |
; * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA |
; * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA |
23 |
; * |
; * |
24 |
; * $Id: quantize_mpeg_mmx.asm,v 1.1.2.4 2003-11-03 15:51:50 edgomez Exp $ |
; * $Id: quantize_mpeg_mmx.asm,v 1.1.2.5 2003-11-30 16:13:16 edgomez Exp $ |
25 |
; * |
; * |
26 |
; *************************************************************************/ |
; *************************************************************************/ |
27 |
|
|
76 |
%assign quant quant+1 |
%assign quant quant+1 |
77 |
%endrep |
%endrep |
78 |
|
|
|
;----------------------------------------------------------------------------- |
|
|
; intra matrix |
|
|
;----------------------------------------------------------------------------- |
|
|
|
|
|
cextern intra_matrix |
|
|
cextern intra_matrix_fix |
|
|
|
|
|
;----------------------------------------------------------------------------- |
|
|
; inter matrix |
|
|
;----------------------------------------------------------------------------- |
|
|
|
|
|
cextern inter_matrix |
|
|
cextern inter_matrix_fix |
|
|
|
|
|
|
|
79 |
%define VM18P 3 |
%define VM18P 3 |
80 |
%define VM18Q 4 |
%define VM18Q 4 |
81 |
|
|
135 |
; uint32_t quant_mpeg_intra_mmx(int16_t * coeff, |
; uint32_t quant_mpeg_intra_mmx(int16_t * coeff, |
136 |
; const int16_t const * data, |
; const int16_t const * data, |
137 |
; const uint32_t quant, |
; const uint32_t quant, |
138 |
; const uint32_t dcscalar); |
; const uint32_t dcscalar, |
139 |
|
; const uint16_t *mpeg_matrices); |
140 |
; |
; |
141 |
;----------------------------------------------------------------------------- |
;----------------------------------------------------------------------------- |
142 |
|
|
146 |
push ecx |
push ecx |
147 |
push esi |
push esi |
148 |
push edi |
push edi |
149 |
|
push ebx |
150 |
|
|
151 |
mov edi, [esp + 12 + 4] ; coeff |
mov edi, [esp + 16 + 4] ; coeff |
152 |
mov esi, [esp + 12 + 8] ; data |
mov esi, [esp + 16 + 8] ; data |
153 |
mov eax, [esp + 12 + 12] ; quant |
mov eax, [esp + 16 + 12] ; quant |
154 |
|
mov ebx, [esp + 16 + 20] ; mpeg_quant_matrices |
155 |
|
|
156 |
movq mm5, [quantd + eax * 8 - 8] ; quantd -> mm5 |
movq mm5, [quantd + eax * 8 - 8] ; quantd -> mm5 |
157 |
|
|
178 |
psubw mm3, mm4 ; |
psubw mm3, mm4 ; |
179 |
psllw mm0, 4 ; level << 4 |
psllw mm0, 4 ; level << 4 |
180 |
psllw mm3, 4 |
psllw mm3, 4 |
181 |
movq mm2, [intra_matrix + 8*ecx] |
movq mm2, [ebx + 8*ecx] |
182 |
psrlw mm2, 1 ; intra_matrix[i]>>1 |
psrlw mm2, 1 ; intra_matrix[i]>>1 |
183 |
paddw mm0, mm2 |
paddw mm0, mm2 |
184 |
movq mm2, [intra_matrix_fix + ecx*8] |
movq mm2, [ebx + 256 + ecx*8] |
185 |
pmulhw mm0, mm2 ; (level<<4 + intra_matrix[i]>>1) / intra_matrix[i] |
pmulhw mm0, mm2 ; (level<<4 + intra_matrix[i]>>1) / intra_matrix[i] |
186 |
movq mm2, [intra_matrix + 8*ecx + 8] |
movq mm2, [ebx + 8*ecx + 8] |
187 |
psrlw mm2, 1 |
psrlw mm2, 1 |
188 |
paddw mm3, mm2 |
paddw mm3, mm2 |
189 |
movq mm2, [intra_matrix_fix + ecx*8 + 8] |
movq mm2, [ebx + 256 + ecx*8 + 8] |
190 |
pmulhw mm3, mm2 |
pmulhw mm3, mm2 |
191 |
paddw mm0, mm5 ; + quantd |
paddw mm0, mm5 ; + quantd |
192 |
paddw mm3, mm5 |
paddw mm3, mm5 |
208 |
|
|
209 |
.done |
.done |
210 |
; caclulate data[0] // (int32_t)dcscalar) |
; caclulate data[0] // (int32_t)dcscalar) |
211 |
mov ecx, [esp + 12 + 16] ; dcscalar |
mov ecx, [esp + 16 + 16] ; dcscalar |
212 |
mov edx, ecx |
mov edx, ecx |
213 |
movsx eax, word [esi] ; data[0] |
movsx eax, word [esi] ; data[0] |
214 |
shr edx, 1 ; edx = dcscalar /2 |
shr edx, 1 ; edx = dcscalar /2 |
225 |
|
|
226 |
mov [edi], ax ; coeff[0] = ax |
mov [edi], ax ; coeff[0] = ax |
227 |
|
|
228 |
|
pop ebx |
229 |
pop edi |
pop edi |
230 |
pop esi |
pop esi |
231 |
pop ecx |
pop ecx |
247 |
psubw mm3, mm4 ; |
psubw mm3, mm4 ; |
248 |
psllw mm0, 4 |
psllw mm0, 4 |
249 |
psllw mm3, 4 |
psllw mm3, 4 |
250 |
movq mm2, [intra_matrix + 8*ecx] |
movq mm2, [ebx + 8*ecx] |
251 |
psrlw mm2, 1 |
psrlw mm2, 1 |
252 |
paddw mm0, mm2 |
paddw mm0, mm2 |
253 |
movq mm2, [intra_matrix_fix + ecx*8] |
movq mm2, [ebx + 256 + ecx*8] |
254 |
pmulhw mm0, mm2 ; (level<<4 + intra_matrix[i]>>1) / intra_matrix[i] |
pmulhw mm0, mm2 ; (level<<4 + intra_matrix[i]>>1) / intra_matrix[i] |
255 |
movq mm2, [intra_matrix + 8*ecx + 8] |
movq mm2, [ebx + 8*ecx + 8] |
256 |
psrlw mm2, 1 |
psrlw mm2, 1 |
257 |
paddw mm3, mm2 |
paddw mm3, mm2 |
258 |
movq mm2, [intra_matrix_fix + ecx*8 + 8] |
movq mm2, [ebx + 256 + ecx*8 + 8] |
259 |
pmulhw mm3, mm2 |
pmulhw mm3, mm2 |
260 |
paddw mm0, mm5 |
paddw mm0, mm5 |
261 |
paddw mm3, mm5 |
paddw mm3, mm5 |
288 |
psubw mm3, mm4 ; |
psubw mm3, mm4 ; |
289 |
psllw mm0, 4 |
psllw mm0, 4 |
290 |
psllw mm3, 4 |
psllw mm3, 4 |
291 |
movq mm2, [intra_matrix + 8*ecx] |
movq mm2, [ebx + 8*ecx] |
292 |
psrlw mm2, 1 |
psrlw mm2, 1 |
293 |
paddw mm0, mm2 |
paddw mm0, mm2 |
294 |
movq mm2, [intra_matrix_fix + ecx*8] |
movq mm2, [ebx + 256 + ecx*8] |
295 |
pmulhw mm0, mm2 ; (level<<4 + intra_matrix[i]>>1) / intra_matrix[i] |
pmulhw mm0, mm2 ; (level<<4 + intra_matrix[i]>>1) / intra_matrix[i] |
296 |
movq mm2, [intra_matrix + 8*ecx + 8] |
movq mm2, [ebx + 8*ecx + 8] |
297 |
psrlw mm2, 1 |
psrlw mm2, 1 |
298 |
paddw mm3, mm2 |
paddw mm3, mm2 |
299 |
movq mm2, [intra_matrix_fix + ecx*8 + 8] |
movq mm2, [ebx + 256 + ecx*8 + 8] |
300 |
pmulhw mm3, mm2 |
pmulhw mm3, mm2 |
301 |
paddw mm0, mm5 |
paddw mm0, mm5 |
302 |
paddw mm3, mm5 |
paddw mm3, mm5 |
319 |
; |
; |
320 |
; uint32_t quant_mpeg_inter_mmx(int16_t * coeff, |
; uint32_t quant_mpeg_inter_mmx(int16_t * coeff, |
321 |
; const int16_t const * data, |
; const int16_t const * data, |
322 |
; const uint32_t quant); |
; const uint32_t quant, |
323 |
|
; const uint16_t *mpeg_matrices); |
324 |
; |
; |
325 |
;----------------------------------------------------------------------------- |
;----------------------------------------------------------------------------- |
326 |
|
|
330 |
push ecx |
push ecx |
331 |
push esi |
push esi |
332 |
push edi |
push edi |
333 |
|
push ebx |
334 |
|
|
335 |
mov edi, [esp + 12 + 4] ; coeff |
mov edi, [esp + 16 + 4] ; coeff |
336 |
mov esi, [esp + 12 + 8] ; data |
mov esi, [esp + 16 + 8] ; data |
337 |
mov eax, [esp + 12 + 12] ; quant |
mov eax, [esp + 16 + 12] ; quant |
338 |
|
mov ebx, [esp + 16 + 16] ; mpeg_quant_matrices |
339 |
|
|
340 |
xor ecx, ecx |
xor ecx, ecx |
341 |
|
|
363 |
psubw mm3, mm4 ; |
psubw mm3, mm4 ; |
364 |
psllw mm0, 4 |
psllw mm0, 4 |
365 |
psllw mm3, 4 |
psllw mm3, 4 |
366 |
movq mm2, [inter_matrix + 8*ecx] |
movq mm2, [ebx + 512 + 8*ecx] |
367 |
psrlw mm2, 1 |
psrlw mm2, 1 |
368 |
paddw mm0, mm2 |
paddw mm0, mm2 |
369 |
movq mm2, [inter_matrix_fix + ecx*8] |
movq mm2, [ebx + 768 + ecx*8] |
370 |
pmulhw mm0, mm2 ; (level<<4 + inter_matrix[i]>>1) / inter_matrix[i] |
pmulhw mm0, mm2 ; (level<<4 + inter_matrix[i]>>1) / inter_matrix[i] |
371 |
movq mm2, [inter_matrix + 8*ecx + 8] |
movq mm2, [ebx + 512 + 8*ecx + 8] |
372 |
psrlw mm2, 1 |
psrlw mm2, 1 |
373 |
paddw mm3, mm2 |
paddw mm3, mm2 |
374 |
movq mm2, [inter_matrix_fix + ecx*8 + 8] |
movq mm2, [ebx + 768 + ecx*8 + 8] |
375 |
pmulhw mm3, mm2 |
pmulhw mm3, mm2 |
376 |
pmulhw mm0, mm7 ; mm0 = (mm0 / 2Q) >> 16 |
pmulhw mm0, mm7 ; mm0 = (mm0 / 2Q) >> 16 |
377 |
pmulhw mm3, mm7 ; |
pmulhw mm3, mm7 ; |
397 |
paddd mm0, mm5 |
paddd mm0, mm5 |
398 |
movd eax, mm0 ; return sum |
movd eax, mm0 ; return sum |
399 |
|
|
400 |
|
pop ebx |
401 |
pop edi |
pop edi |
402 |
pop esi |
pop esi |
403 |
pop ecx |
pop ecx |
418 |
psubw mm3, mm4 ; |
psubw mm3, mm4 ; |
419 |
psllw mm0, 4 |
psllw mm0, 4 |
420 |
psllw mm3, 4 |
psllw mm3, 4 |
421 |
movq mm2, [inter_matrix + 8*ecx] |
movq mm2, [ebx + 512 + 8*ecx] |
422 |
psrlw mm2, 1 |
psrlw mm2, 1 |
423 |
paddw mm0, mm2 |
paddw mm0, mm2 |
424 |
movq mm2, [inter_matrix_fix + ecx*8] |
movq mm2, [ebx + 768 + ecx*8] |
425 |
pmulhw mm0, mm2 ; (level<<4 + inter_matrix[i]>>1) / inter_matrix[i] |
pmulhw mm0, mm2 ; (level<<4 + inter_matrix[i]>>1) / inter_matrix[i] |
426 |
movq mm2, [inter_matrix + 8*ecx + 8] |
movq mm2, [ebx + 512 + 8*ecx + 8] |
427 |
psrlw mm2, 1 |
psrlw mm2, 1 |
428 |
paddw mm3, mm2 |
paddw mm3, mm2 |
429 |
movq mm2, [inter_matrix_fix + ecx*8 + 8] |
movq mm2, [ebx + 768 + ecx*8 + 8] |
430 |
pmulhw mm3, mm2 |
pmulhw mm3, mm2 |
431 |
psrlw mm0, 1 ; mm0 >>= 1 (/2) |
psrlw mm0, 1 ; mm0 >>= 1 (/2) |
432 |
psrlw mm3, 1 ; |
psrlw mm3, 1 ; |
460 |
psubw mm3, mm4 ; |
psubw mm3, mm4 ; |
461 |
psllw mm0, 4 |
psllw mm0, 4 |
462 |
psllw mm3, 4 |
psllw mm3, 4 |
463 |
movq mm2, [inter_matrix + 8*ecx] |
movq mm2, [ebx + 512 + 8*ecx] |
464 |
psrlw mm2, 1 |
psrlw mm2, 1 |
465 |
paddw mm0, mm2 |
paddw mm0, mm2 |
466 |
movq mm2, [inter_matrix_fix + ecx*8] |
movq mm2, [ebx + 768 + ecx*8] |
467 |
pmulhw mm0, mm2 ; (level<<4 + inter_matrix[i]>>1) / inter_matrix[i] |
pmulhw mm0, mm2 ; (level<<4 + inter_matrix[i]>>1) / inter_matrix[i] |
468 |
movq mm2, [inter_matrix + 8*ecx + 8] |
movq mm2, [ebx + 512 + 8*ecx + 8] |
469 |
psrlw mm2, 1 |
psrlw mm2, 1 |
470 |
paddw mm3, mm2 |
paddw mm3, mm2 |
471 |
movq mm2, [inter_matrix_fix + ecx*8 + 8] |
movq mm2, [ebx + 768 + ecx*8 + 8] |
472 |
pmulhw mm3, mm2 |
pmulhw mm3, mm2 |
473 |
psrlw mm0, 2 ; mm0 >>= 1 (/2) |
psrlw mm0, 2 ; mm0 >>= 1 (/2) |
474 |
psrlw mm3, 2 ; |
psrlw mm3, 2 ; |
493 |
; uint32_t dequant_mpeg_intra_mmx(int16_t *data, |
; uint32_t dequant_mpeg_intra_mmx(int16_t *data, |
494 |
; const int16_t const *coeff, |
; const int16_t const *coeff, |
495 |
; const uint32_t quant, |
; const uint32_t quant, |
496 |
; const uint32_t dcscalar); |
; const uint32_t dcscalar, |
497 |
|
; const uint16_t *mpeg_matrices); |
498 |
; |
; |
499 |
;----------------------------------------------------------------------------- |
;----------------------------------------------------------------------------- |
500 |
|
|
517 |
psubw mm0, mm1 ; -> mm0 = abs(coeff[i]), mm1 = sign of coeff[i] |
psubw mm0, mm1 ; -> mm0 = abs(coeff[i]), mm1 = sign of coeff[i] |
518 |
|
|
519 |
movq mm2, mm7 ; mm2 = quant |
movq mm2, mm7 ; mm2 = quant |
520 |
pmullw mm2, [intra_matrix + 8*eax + 8*16 ] ; matrix[i]*quant. |
pmullw mm2, [ebx + 8*eax + 8*16 ] ; matrix[i]*quant. |
521 |
|
|
522 |
movq mm6, mm2 |
movq mm6, mm2 |
523 |
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) |
540 |
ALIGN 16 |
ALIGN 16 |
541 |
dequant_mpeg_intra_mmx: |
dequant_mpeg_intra_mmx: |
542 |
|
|
543 |
mov edx, [esp+4] ; data |
push ebx |
544 |
mov ecx, [esp+8] ; coeff |
|
545 |
mov eax, [esp+12] ; quant |
mov edx, [esp + 4 + 4] ; data |
546 |
|
mov ecx, [esp + 4 + 8] ; coeff |
547 |
|
mov eax, [esp + 4 + 12] ; quant |
548 |
|
mov ebx, [esp + 4 + 20] ; mpeg_quant_matrices |
549 |
|
|
550 |
movq mm7, [mmx_mul_quant + eax*8 - 8] |
movq mm7, [mmx_mul_quant + eax*8 - 8] |
551 |
mov eax, -16 ; to keep ALIGNed, we regularly process coeff[0] |
mov eax, -16 ; to keep ALIGNed, we regularly process coeff[0] |
562 |
movq mm2, mm7 ; mm2 = quant |
movq mm2, mm7 ; mm2 = quant |
563 |
|
|
564 |
pcmpgtw mm4, mm3 ; mm4 = sgn(c') |
pcmpgtw mm4, mm3 ; mm4 = sgn(c') |
565 |
pmullw mm2, [intra_matrix + 8*eax + 8*16 ] ; matrix[i]*quant |
pmullw mm2, [ebx + 8*eax + 8*16 ] ; matrix[i]*quant |
566 |
|
|
567 |
pxor mm0, mm1 ; negate if negative |
pxor mm0, mm1 ; negate if negative |
568 |
pxor mm3, mm4 ; negate if negative |
pxor mm3, mm4 ; negate if negative |
578 |
pmulhw mm0, mm5 ; high of coeff*(matrix*quant) |
pmulhw mm0, mm5 ; high of coeff*(matrix*quant) |
579 |
movq mm5, mm7 ; mm2 = quant |
movq mm5, mm7 ; mm2 = quant |
580 |
|
|
581 |
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 |
582 |
|
|
583 |
movq mm6, mm5 |
movq mm6, mm5 |
584 |
add eax,2 ; z-flag will be tested later |
add eax,2 ; z-flag will be tested later |
610 |
|
|
611 |
; deal with DC |
; deal with DC |
612 |
movd mm0, [ecx] |
movd mm0, [ecx] |
613 |
pmullw mm0, [esp+16] ; dcscalar |
pmullw mm0, [esp + 4 + 16] ; dcscalar |
614 |
movq mm2, [mmx_32767_minus_2047] |
movq mm2, [mmx_32767_minus_2047] |
615 |
paddsw mm0, mm2 |
paddsw mm0, mm2 |
616 |
psubsw mm0, mm2 |
psubsw mm0, mm2 |
621 |
mov [edx], ax |
mov [edx], ax |
622 |
|
|
623 |
xor eax, eax |
xor eax, eax |
624 |
|
|
625 |
|
pop ebx |
626 |
|
|
627 |
ret |
ret |
628 |
|
|
629 |
;----------------------------------------------------------------------------- |
;----------------------------------------------------------------------------- |
630 |
; |
; |
631 |
; uint32_t dequant_mpeg_inter_mmx(int16_t * data, |
; uint32_t dequant_mpeg_inter_mmx(int16_t * data, |
632 |
; const int16_t * const coeff, |
; const int16_t * const coeff, |
633 |
; const uint32_t quant); |
; const uint32_t quant, |
634 |
|
; const uint16_t *mpeg_matrices); |
635 |
; |
; |
636 |
;----------------------------------------------------------------------------- |
;----------------------------------------------------------------------------- |
637 |
|
|
643 |
ALIGN 16 |
ALIGN 16 |
644 |
dequant_mpeg_inter_mmx: |
dequant_mpeg_inter_mmx: |
645 |
|
|
646 |
mov edx, [esp+ 4] ; data |
push ebx |
647 |
mov ecx, [esp+ 8] ; coeff |
|
648 |
mov eax, [esp+12] ; quant |
mov edx, [esp + 4 + 4] ; data |
649 |
|
mov ecx, [esp + 4 + 8] ; coeff |
650 |
|
mov eax, [esp + 4 + 12] ; quant |
651 |
|
mov ebx, [esp + 4 + 16] ; mpeg_quant_matrices |
652 |
|
|
653 |
movq mm7, [mmx_mul_quant + eax*8 - 8] |
movq mm7, [mmx_mul_quant + eax*8 - 8] |
654 |
mov eax, -16 |
mov eax, -16 |
655 |
paddw mm7, mm7 ; << 1 |
paddw mm7, mm7 ; << 1 |
684 |
; we're short on register, here. Poor pairing... |
; we're short on register, here. Poor pairing... |
685 |
|
|
686 |
movq mm4, mm7 ; (matrix*quant) |
movq mm4, mm7 ; (matrix*quant) |
687 |
pmullw mm4, [inter_matrix + 8*eax + 8*16 -2*8] |
pmullw mm4, [ebx + 512 + 8*eax + 8*16 -2*8] |
688 |
movq mm5, mm4 |
movq mm5, mm4 |
689 |
pmulhw mm5, mm0 ; high of c*(matrix*quant) |
pmulhw mm5, mm0 ; high of c*(matrix*quant) |
690 |
pmullw mm0, mm4 ; low of c*(matrix*quant) |
pmullw mm0, mm4 ; low of c*(matrix*quant) |
691 |
|
|
692 |
movq mm4, mm7 ; (matrix*quant) |
movq mm4, mm7 ; (matrix*quant) |
693 |
pmullw mm4, [inter_matrix + 8*eax + 8*16 -2*8 + 8] |
pmullw mm4, [ebx + 512 + 8*eax + 8*16 -2*8 + 8] |
694 |
|
|
695 |
pcmpgtw mm5, [zero] |
pcmpgtw mm5, [zero] |
696 |
paddusw mm0, mm5 |
paddusw mm0, mm5 |
734 |
xor word [edx + 2*63], ax |
xor word [edx + 2*63], ax |
735 |
|
|
736 |
xor eax, eax |
xor eax, eax |
737 |
|
|
738 |
|
pop ebx |
739 |
|
|
740 |
ret |
ret |