1 |
;/***************************************************************************** |
;/***************************************************************************** |
2 |
; * |
; * |
3 |
; * XVID MPEG-4 VIDEO CODEC |
; * XVID MPEG-4 VIDEO CODEC |
4 |
; * mmx 8x8 block-based halfpel interpolation |
; * - mmx 8x8 block-based halfpel interpolation - |
5 |
; * |
; * |
6 |
; * Copyright(C) 2002 Peter Ross <pross@xvid.org> |
; * Copyright(C) 2001 Peter Ross <pross@xvid.org> |
7 |
|
; * 2002 Michael Militzer <isibaar@xvid.org> |
8 |
; * |
; * |
9 |
; * This file is part of XviD, a free MPEG-4 video encoder/decoder |
; * This program is free software ; you can redistribute it and/or modify |
10 |
; * |
; * it under the terms of the GNU General Public License as published by |
|
; * XviD is free software; you can redistribute it and/or modify it |
|
|
; * under the terms of the GNU General Public License as published by |
|
11 |
; * the Free Software Foundation; either version 2 of the License, or |
; * the Free Software Foundation; either version 2 of the License, or |
12 |
; * (at your option) any later version. |
; * (at your option) any later version. |
13 |
; * |
; * |
20 |
; * along with this program; if not, write to the Free Software |
; * along with this program; if not, write to the Free Software |
21 |
; * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA |
; * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA |
22 |
; * |
; * |
|
; * Under section 8 of the GNU General Public License, the copyright |
|
|
; * holders of XVID explicitly forbid distribution in the following |
|
|
; * countries: |
|
|
; * |
|
|
; * - Japan |
|
|
; * - United States of America |
|
|
; * |
|
|
; * Linking XviD statically or dynamically with other modules is making a |
|
|
; * combined work based on XviD. Thus, the terms and conditions of the |
|
|
; * GNU General Public License cover the whole combination. |
|
|
; * |
|
|
; * As a special exception, the copyright holders of XviD give you |
|
|
; * permission to link XviD with independent modules that communicate with |
|
|
; * XviD solely through the VFW1.1 and DShow interfaces, regardless of the |
|
|
; * license terms of these independent modules, and to copy and distribute |
|
|
; * the resulting combined work under terms of your choice, provided that |
|
|
; * every copy of the combined work is accompanied by a complete copy of |
|
|
; * the source code of XviD (the version of XviD used to produce the |
|
|
; * combined work), being distributed under the terms of the GNU General |
|
|
; * Public License plus this exception. An independent module is a module |
|
|
; * which is not derived from or based on XviD. |
|
|
; * |
|
|
; * Note that people who make modified versions of XviD are not obligated |
|
|
; * to grant this special exception for their modified versions; it is |
|
|
; * their choice whether to do so. The GNU General Public License gives |
|
|
; * permission to release a modified version without this exception; this |
|
|
; * exception also makes it possible to release a modified version which |
|
|
; * carries forward this exception. |
|
|
; * |
|
|
; * $Id: interpolate8x8_mmx.asm,v 1.11 2002-11-17 00:20:30 edgomez Exp $ |
|
|
; * |
|
23 |
; ****************************************************************************/ |
; ****************************************************************************/ |
24 |
|
|
25 |
bits 32 |
BITS 32 |
26 |
|
|
27 |
%macro cglobal 1 |
%macro cglobal 1 |
28 |
%ifdef PREFIX |
%ifdef PREFIX |
33 |
%endif |
%endif |
34 |
%endmacro |
%endmacro |
35 |
|
|
36 |
section .data |
;============================================================================= |
37 |
|
; Read only data |
38 |
|
;============================================================================= |
39 |
|
|
40 |
|
%ifdef FORMAT_COFF |
41 |
|
SECTION .rodata |
42 |
|
%else |
43 |
|
SECTION .rodata align=16 |
44 |
|
%endif |
45 |
|
|
46 |
align 16 |
;----------------------------------------------------------------------------- |
47 |
|
; (16 - r) rounding table |
48 |
|
;----------------------------------------------------------------------------- |
49 |
|
|
50 |
|
ALIGN 16 |
51 |
|
rounding_lowpass_mmx: |
52 |
|
times 4 dw 16 |
53 |
|
times 4 dw 15 |
54 |
|
|
55 |
;=========================================================================== |
;----------------------------------------------------------------------------- |
56 |
; (1 - r) rounding table |
; (1 - r) rounding table |
57 |
;=========================================================================== |
;----------------------------------------------------------------------------- |
58 |
|
|
59 |
rounding1_mmx |
rounding1_mmx: |
60 |
times 4 dw 1 |
times 4 dw 1 |
61 |
times 4 dw 0 |
times 4 dw 0 |
62 |
|
|
63 |
;=========================================================================== |
;----------------------------------------------------------------------------- |
64 |
; (2 - r) rounding table |
; (2 - r) rounding table |
65 |
;=========================================================================== |
;----------------------------------------------------------------------------- |
66 |
|
|
67 |
rounding2_mmx |
rounding2_mmx: |
68 |
times 4 dw 2 |
times 4 dw 2 |
69 |
times 4 dw 1 |
times 4 dw 1 |
70 |
|
|
71 |
mmx_one |
mmx_one: |
72 |
times 8 db 1 |
times 8 db 1 |
73 |
|
|
74 |
section .text |
mmx_two: |
75 |
|
times 8 db 2 |
76 |
|
|
77 |
|
mmx_three: |
78 |
|
times 8 db 3 |
79 |
|
|
80 |
|
mmx_five: |
81 |
|
times 4 dw 5 |
82 |
|
|
83 |
|
mmx_mask: |
84 |
|
times 8 db 254 |
85 |
|
|
86 |
|
mmx_mask2: |
87 |
|
times 8 db 252 |
88 |
|
|
89 |
|
;============================================================================= |
90 |
|
; Code |
91 |
|
;============================================================================= |
92 |
|
|
93 |
|
SECTION .text |
94 |
|
|
95 |
|
cglobal interpolate8x8_halfpel_h_mmx |
96 |
|
cglobal interpolate8x8_halfpel_v_mmx |
97 |
|
cglobal interpolate8x8_halfpel_hv_mmx |
98 |
|
|
99 |
|
cglobal interpolate8x8_avg4_mmx |
100 |
|
cglobal interpolate8x8_avg2_mmx |
101 |
|
|
102 |
|
cglobal interpolate8x8_6tap_lowpass_h_mmx |
103 |
|
cglobal interpolate8x8_6tap_lowpass_v_mmx |
104 |
|
|
105 |
|
cglobal interpolate8x8_halfpel_add_mmx |
106 |
|
cglobal interpolate8x8_halfpel_h_add_mmx |
107 |
|
cglobal interpolate8x8_halfpel_v_add_mmx |
108 |
|
cglobal interpolate8x8_halfpel_hv_add_mmx |
109 |
|
|
110 |
%macro CALC_AVG 6 |
%macro CALC_AVG 6 |
111 |
punpcklbw %3, %6 |
punpcklbw %3, %6 |
118 |
|
|
119 |
psrlw %1, 1 ; mm01 >>= 1 |
psrlw %1, 1 ; mm01 >>= 1 |
120 |
psrlw %2, 1 |
psrlw %2, 1 |
|
|
|
121 |
%endmacro |
%endmacro |
122 |
|
|
123 |
|
|
124 |
;=========================================================================== |
;----------------------------------------------------------------------------- |
125 |
; |
; |
126 |
; void interpolate8x8_halfpel_h_mmx(uint8_t * const dst, |
; void interpolate8x8_halfpel_h_mmx(uint8_t * const dst, |
127 |
; const uint8_t * const src, |
; const uint8_t * const src, |
128 |
; const uint32_t stride, |
; const uint32_t stride, |
129 |
; const uint32_t rounding); |
; const uint32_t rounding); |
130 |
; |
; |
131 |
;=========================================================================== |
;----------------------------------------------------------------------------- |
132 |
|
|
133 |
%macro COPY_H_MMX 0 |
%macro COPY_H_MMX 0 |
134 |
movq mm0, [esi] |
movq mm0, [esi] |
148 |
add edi, edx ; dst += stride |
add edi, edx ; dst += stride |
149 |
%endmacro |
%endmacro |
150 |
|
|
151 |
align 16 |
ALIGN 16 |
152 |
cglobal interpolate8x8_halfpel_h_mmx |
interpolate8x8_halfpel_h_mmx: |
|
interpolate8x8_halfpel_h_mmx |
|
153 |
|
|
154 |
push esi |
push esi |
155 |
push edi |
push edi |
|
|
|
156 |
mov eax, [esp + 8 + 16] ; rounding |
mov eax, [esp + 8 + 16] ; rounding |
157 |
|
|
|
interpolate8x8_halfpel_h_mmx.start |
|
158 |
movq mm7, [rounding1_mmx + eax * 8] |
movq mm7, [rounding1_mmx + eax * 8] |
159 |
|
|
160 |
mov edi, [esp + 8 + 4] ; dst |
mov edi, [esp + 8 + 4] ; dst |
178 |
ret |
ret |
179 |
|
|
180 |
|
|
181 |
;=========================================================================== |
;----------------------------------------------------------------------------- |
182 |
; |
; |
183 |
; void interpolate8x8_halfpel_v_mmx(uint8_t * const dst, |
; void interpolate8x8_halfpel_v_mmx(uint8_t * const dst, |
184 |
; const uint8_t * const src, |
; const uint8_t * const src, |
185 |
; const uint32_t stride, |
; const uint32_t stride, |
186 |
; const uint32_t rounding); |
; const uint32_t rounding); |
187 |
; |
; |
188 |
;=========================================================================== |
;----------------------------------------------------------------------------- |
189 |
|
|
190 |
%macro COPY_V_MMX 0 |
%macro COPY_V_MMX 0 |
191 |
movq mm0, [esi] |
movq mm0, [esi] |
205 |
add edi, edx ; dst += stride |
add edi, edx ; dst += stride |
206 |
%endmacro |
%endmacro |
207 |
|
|
208 |
align 16 |
ALIGN 16 |
209 |
cglobal interpolate8x8_halfpel_v_mmx |
interpolate8x8_halfpel_v_mmx: |
|
interpolate8x8_halfpel_v_mmx |
|
210 |
|
|
211 |
push esi |
push esi |
212 |
push edi |
push edi |
213 |
|
|
214 |
mov eax, [esp + 8 + 16] ; rounding |
mov eax, [esp + 8 + 16] ; rounding |
215 |
|
|
|
interpolate8x8_halfpel_v_mmx.start |
|
216 |
movq mm7, [rounding1_mmx + eax * 8] |
movq mm7, [rounding1_mmx + eax * 8] |
217 |
|
|
218 |
mov edi, [esp + 8 + 4] ; dst |
mov edi, [esp + 8 + 4] ; dst |
237 |
ret |
ret |
238 |
|
|
239 |
|
|
240 |
;=========================================================================== |
;----------------------------------------------------------------------------- |
241 |
; |
; |
242 |
; void interpolate8x8_halfpel_hv_mmx(uint8_t * const dst, |
; void interpolate8x8_halfpel_hv_mmx(uint8_t * const dst, |
243 |
; const uint8_t * const src, |
; const uint8_t * const src, |
245 |
; const uint32_t rounding); |
; const uint32_t rounding); |
246 |
; |
; |
247 |
; |
; |
248 |
;=========================================================================== |
;----------------------------------------------------------------------------- |
249 |
|
|
250 |
%macro COPY_HV_MMX 0 |
%macro COPY_HV_MMX 0 |
251 |
; current row |
; current row |
|
|
|
252 |
movq mm0, [esi] |
movq mm0, [esi] |
253 |
movq mm2, [esi + 1] |
movq mm2, [esi + 1] |
254 |
|
|
264 |
paddusw mm1, mm3 |
paddusw mm1, mm3 |
265 |
|
|
266 |
; next row |
; next row |
|
|
|
267 |
movq mm4, [esi + edx] |
movq mm4, [esi + edx] |
268 |
movq mm2, [esi + edx + 1] |
movq mm2, [esi + edx + 1] |
269 |
|
|
279 |
paddusw mm5, mm3 |
paddusw mm5, mm3 |
280 |
|
|
281 |
; add current + next row |
; add current + next row |
|
|
|
282 |
paddusw mm0, mm4 ; mm01 += mm45 |
paddusw mm0, mm4 ; mm01 += mm45 |
283 |
paddusw mm1, mm5 |
paddusw mm1, mm5 |
284 |
paddusw mm0, mm7 ; mm01 += rounding2 |
paddusw mm0, mm7 ; mm01 += rounding2 |
294 |
add edi, edx ; dst += stride |
add edi, edx ; dst += stride |
295 |
%endmacro |
%endmacro |
296 |
|
|
297 |
align 16 |
ALIGN 16 |
298 |
cglobal interpolate8x8_halfpel_hv_mmx |
interpolate8x8_halfpel_hv_mmx: |
|
interpolate8x8_halfpel_hv_mmx |
|
299 |
|
|
300 |
push esi |
push esi |
301 |
push edi |
push edi |
302 |
|
|
303 |
mov eax, [esp + 8 + 16] ; rounding |
mov eax, [esp + 8 + 16] ; rounding |
|
interpolate8x8_halfpel_hv_mmx.start |
|
304 |
|
|
305 |
movq mm7, [rounding2_mmx + eax * 8] |
movq mm7, [rounding2_mmx + eax * 8] |
306 |
|
|
326 |
pop esi |
pop esi |
327 |
|
|
328 |
ret |
ret |
329 |
|
|
330 |
|
;----------------------------------------------------------------------------- |
331 |
|
; |
332 |
|
; void interpolate8x8_avg2_mmx(uint8_t const *dst, |
333 |
|
; const uint8_t * const src1, |
334 |
|
; const uint8_t * const src2, |
335 |
|
; const uint32_t stride, |
336 |
|
; const uint32_t rounding, |
337 |
|
; const uint32_t height); |
338 |
|
; |
339 |
|
;----------------------------------------------------------------------------- |
340 |
|
|
341 |
|
%macro AVG2_MMX_RND0 0 |
342 |
|
movq mm0, [eax] ; src1 -> mm0 |
343 |
|
movq mm1, [ebx] ; src2 -> mm1 |
344 |
|
|
345 |
|
movq mm4, [eax+edx] |
346 |
|
movq mm5, [ebx+edx] |
347 |
|
|
348 |
|
movq mm2, mm0 ; src1 -> mm2 |
349 |
|
movq mm3, mm1 ; src2 -> mm3 |
350 |
|
|
351 |
|
pand mm2, mm7 ; isolate the lsb |
352 |
|
pand mm3, mm7 ; isolate the lsb |
353 |
|
|
354 |
|
por mm2, mm3 ; ODD(src1) OR ODD(src2) -> mm2 |
355 |
|
|
356 |
|
movq mm3, mm4 |
357 |
|
movq mm6, mm5 |
358 |
|
|
359 |
|
pand mm3, mm7 |
360 |
|
pand mm6, mm7 |
361 |
|
|
362 |
|
por mm3, mm6 |
363 |
|
|
364 |
|
pand mm0, [mmx_mask] |
365 |
|
pand mm1, [mmx_mask] |
366 |
|
pand mm4, [mmx_mask] |
367 |
|
pand mm5, [mmx_mask] |
368 |
|
|
369 |
|
psrlq mm0, 1 ; src1 / 2 |
370 |
|
psrlq mm1, 1 ; src2 / 2 |
371 |
|
|
372 |
|
psrlq mm4, 1 |
373 |
|
psrlq mm5, 1 |
374 |
|
|
375 |
|
paddb mm0, mm1 ; src1/2 + src2/2 -> mm0 |
376 |
|
paddb mm0, mm2 ; correct rounding error |
377 |
|
|
378 |
|
paddb mm4, mm5 |
379 |
|
paddb mm4, mm3 |
380 |
|
|
381 |
|
lea eax, [eax+2*edx] |
382 |
|
lea ebx, [ebx+2*edx] |
383 |
|
|
384 |
|
movq [ecx], mm0 ; (src1 + src2 + 1) / 2 -> dst |
385 |
|
movq [ecx+edx], mm4 |
386 |
|
%endmacro |
387 |
|
|
388 |
|
%macro AVG2_MMX_RND1 0 |
389 |
|
movq mm0, [eax] ; src1 -> mm0 |
390 |
|
movq mm1, [ebx] ; src2 -> mm1 |
391 |
|
|
392 |
|
movq mm4, [eax+edx] |
393 |
|
movq mm5, [ebx+edx] |
394 |
|
|
395 |
|
movq mm2, mm0 ; src1 -> mm2 |
396 |
|
movq mm3, mm1 ; src2 -> mm3 |
397 |
|
|
398 |
|
pand mm2, mm7 ; isolate the lsb |
399 |
|
pand mm3, mm7 ; isolate the lsb |
400 |
|
|
401 |
|
pand mm2, mm3 ; ODD(src1) AND ODD(src2) -> mm2 |
402 |
|
|
403 |
|
movq mm3, mm4 |
404 |
|
movq mm6, mm5 |
405 |
|
|
406 |
|
pand mm3, mm7 |
407 |
|
pand mm6, mm7 |
408 |
|
|
409 |
|
pand mm3, mm6 |
410 |
|
|
411 |
|
pand mm0, [mmx_mask] |
412 |
|
pand mm1, [mmx_mask] |
413 |
|
pand mm4, [mmx_mask] |
414 |
|
pand mm5, [mmx_mask] |
415 |
|
|
416 |
|
psrlq mm0, 1 ; src1 / 2 |
417 |
|
psrlq mm1, 1 ; src2 / 2 |
418 |
|
|
419 |
|
psrlq mm4, 1 |
420 |
|
psrlq mm5, 1 |
421 |
|
|
422 |
|
paddb mm0, mm1 ; src1/2 + src2/2 -> mm0 |
423 |
|
paddb mm0, mm2 ; correct rounding error |
424 |
|
|
425 |
|
paddb mm4, mm5 |
426 |
|
paddb mm4, mm3 |
427 |
|
|
428 |
|
lea eax, [eax+2*edx] |
429 |
|
lea ebx, [ebx+2*edx] |
430 |
|
|
431 |
|
movq [ecx], mm0 ; (src1 + src2 + 1) / 2 -> dst |
432 |
|
movq [ecx+edx], mm4 |
433 |
|
%endmacro |
434 |
|
|
435 |
|
ALIGN 16 |
436 |
|
interpolate8x8_avg2_mmx: |
437 |
|
|
438 |
|
push ebx |
439 |
|
|
440 |
|
mov eax, [esp + 4 + 20] ; rounding |
441 |
|
test eax, eax |
442 |
|
|
443 |
|
jnz near .rounding1 |
444 |
|
|
445 |
|
mov eax, [esp + 4 + 24] ; height -> eax |
446 |
|
sub eax, 8 |
447 |
|
test eax, eax |
448 |
|
|
449 |
|
mov ecx, [esp + 4 + 4] ; dst -> edi |
450 |
|
mov eax, [esp + 4 + 8] ; src1 -> esi |
451 |
|
mov ebx, [esp + 4 + 12] ; src2 -> eax |
452 |
|
mov edx, [esp + 4 + 16] ; stride -> edx |
453 |
|
|
454 |
|
movq mm7, [mmx_one] |
455 |
|
|
456 |
|
jz near .start0 |
457 |
|
|
458 |
|
AVG2_MMX_RND0 |
459 |
|
lea ecx, [ecx+2*edx] |
460 |
|
|
461 |
|
.start0 |
462 |
|
|
463 |
|
AVG2_MMX_RND0 |
464 |
|
lea ecx, [ecx+2*edx] |
465 |
|
AVG2_MMX_RND0 |
466 |
|
lea ecx, [ecx+2*edx] |
467 |
|
AVG2_MMX_RND0 |
468 |
|
lea ecx, [ecx+2*edx] |
469 |
|
AVG2_MMX_RND0 |
470 |
|
|
471 |
|
pop ebx |
472 |
|
ret |
473 |
|
|
474 |
|
.rounding1 |
475 |
|
mov eax, [esp + 4 + 24] ; height -> eax |
476 |
|
sub eax, 8 |
477 |
|
test eax, eax |
478 |
|
|
479 |
|
mov ecx, [esp + 4 + 4] ; dst -> edi |
480 |
|
mov eax, [esp + 4 + 8] ; src1 -> esi |
481 |
|
mov ebx, [esp + 4 + 12] ; src2 -> eax |
482 |
|
mov edx, [esp + 4 + 16] ; stride -> edx |
483 |
|
|
484 |
|
movq mm7, [mmx_one] |
485 |
|
|
486 |
|
jz near .start1 |
487 |
|
|
488 |
|
AVG2_MMX_RND1 |
489 |
|
lea ecx, [ecx+2*edx] |
490 |
|
|
491 |
|
.start1 |
492 |
|
|
493 |
|
AVG2_MMX_RND1 |
494 |
|
lea ecx, [ecx+2*edx] |
495 |
|
AVG2_MMX_RND1 |
496 |
|
lea ecx, [ecx+2*edx] |
497 |
|
AVG2_MMX_RND1 |
498 |
|
lea ecx, [ecx+2*edx] |
499 |
|
AVG2_MMX_RND1 |
500 |
|
|
501 |
|
pop ebx |
502 |
|
ret |
503 |
|
|
504 |
|
|
505 |
|
;----------------------------------------------------------------------------- |
506 |
|
; |
507 |
|
; void interpolate8x8_avg4_mmx(uint8_t const *dst, |
508 |
|
; const uint8_t * const src1, |
509 |
|
; const uint8_t * const src2, |
510 |
|
; const uint8_t * const src3, |
511 |
|
; const uint8_t * const src4, |
512 |
|
; const uint32_t stride, |
513 |
|
; const uint32_t rounding); |
514 |
|
; |
515 |
|
;----------------------------------------------------------------------------- |
516 |
|
|
517 |
|
%macro AVG4_MMX_RND0 0 |
518 |
|
movq mm0, [eax] ; src1 -> mm0 |
519 |
|
movq mm1, [ebx] ; src2 -> mm1 |
520 |
|
|
521 |
|
movq mm2, mm0 |
522 |
|
movq mm3, mm1 |
523 |
|
|
524 |
|
pand mm2, [mmx_three] |
525 |
|
pand mm3, [mmx_three] |
526 |
|
|
527 |
|
pand mm0, [mmx_mask2] |
528 |
|
pand mm1, [mmx_mask2] |
529 |
|
|
530 |
|
psrlq mm0, 2 |
531 |
|
psrlq mm1, 2 |
532 |
|
|
533 |
|
lea eax, [eax+edx] |
534 |
|
lea ebx, [ebx+edx] |
535 |
|
|
536 |
|
paddb mm0, mm1 |
537 |
|
paddb mm2, mm3 |
538 |
|
|
539 |
|
movq mm4, [esi] ; src3 -> mm0 |
540 |
|
movq mm5, [edi] ; src4 -> mm1 |
541 |
|
|
542 |
|
movq mm1, mm4 |
543 |
|
movq mm3, mm5 |
544 |
|
|
545 |
|
pand mm1, [mmx_three] |
546 |
|
pand mm3, [mmx_three] |
547 |
|
|
548 |
|
pand mm4, [mmx_mask2] |
549 |
|
pand mm5, [mmx_mask2] |
550 |
|
|
551 |
|
psrlq mm4, 2 |
552 |
|
psrlq mm5, 2 |
553 |
|
|
554 |
|
paddb mm4, mm5 |
555 |
|
paddb mm0, mm4 |
556 |
|
|
557 |
|
paddb mm1, mm3 |
558 |
|
paddb mm2, mm1 |
559 |
|
|
560 |
|
paddb mm2, [mmx_two] |
561 |
|
pand mm2, [mmx_mask2] |
562 |
|
|
563 |
|
psrlq mm2, 2 |
564 |
|
paddb mm0, mm2 |
565 |
|
|
566 |
|
lea esi, [esi+edx] |
567 |
|
lea edi, [edi+edx] |
568 |
|
|
569 |
|
movq [ecx], mm0 ; (src1 + src2 + src3 + src4 + 2) / 4 -> dst |
570 |
|
%endmacro |
571 |
|
|
572 |
|
%macro AVG4_MMX_RND1 0 |
573 |
|
movq mm0, [eax] ; src1 -> mm0 |
574 |
|
movq mm1, [ebx] ; src2 -> mm1 |
575 |
|
|
576 |
|
movq mm2, mm0 |
577 |
|
movq mm3, mm1 |
578 |
|
|
579 |
|
pand mm2, [mmx_three] |
580 |
|
pand mm3, [mmx_three] |
581 |
|
|
582 |
|
pand mm0, [mmx_mask2] |
583 |
|
pand mm1, [mmx_mask2] |
584 |
|
|
585 |
|
psrlq mm0, 2 |
586 |
|
psrlq mm1, 2 |
587 |
|
|
588 |
|
lea eax,[eax+edx] |
589 |
|
lea ebx,[ebx+edx] |
590 |
|
|
591 |
|
paddb mm0, mm1 |
592 |
|
paddb mm2, mm3 |
593 |
|
|
594 |
|
movq mm4, [esi] ; src3 -> mm0 |
595 |
|
movq mm5, [edi] ; src4 -> mm1 |
596 |
|
|
597 |
|
movq mm1, mm4 |
598 |
|
movq mm3, mm5 |
599 |
|
|
600 |
|
pand mm1, [mmx_three] |
601 |
|
pand mm3, [mmx_three] |
602 |
|
|
603 |
|
pand mm4, [mmx_mask2] |
604 |
|
pand mm5, [mmx_mask2] |
605 |
|
|
606 |
|
psrlq mm4, 2 |
607 |
|
psrlq mm5, 2 |
608 |
|
|
609 |
|
paddb mm4, mm5 |
610 |
|
paddb mm0, mm4 |
611 |
|
|
612 |
|
paddb mm1, mm3 |
613 |
|
paddb mm2, mm1 |
614 |
|
|
615 |
|
paddb mm2, [mmx_one] |
616 |
|
pand mm2, [mmx_mask2] |
617 |
|
|
618 |
|
psrlq mm2, 2 |
619 |
|
paddb mm0, mm2 |
620 |
|
|
621 |
|
lea esi,[esi+edx] |
622 |
|
lea edi,[edi+edx] |
623 |
|
|
624 |
|
movq [ecx], mm0 ; (src1 + src2 + src3 + src4 + 2) / 4 -> dst |
625 |
|
%endmacro |
626 |
|
|
627 |
|
ALIGN 16 |
628 |
|
interpolate8x8_avg4_mmx: |
629 |
|
|
630 |
|
push ebx |
631 |
|
push edi |
632 |
|
push esi |
633 |
|
|
634 |
|
mov eax, [esp + 12 + 28] ; rounding |
635 |
|
|
636 |
|
test eax, eax |
637 |
|
|
638 |
|
mov ecx, [esp + 12 + 4] ; dst -> edi |
639 |
|
mov eax, [esp + 12 + 8] ; src1 -> esi |
640 |
|
mov ebx, [esp + 12 + 12] ; src2 -> eax |
641 |
|
mov esi, [esp + 12 + 16] ; src3 -> esi |
642 |
|
mov edi, [esp + 12 + 20] ; src4 -> edi |
643 |
|
mov edx, [esp + 12 + 24] ; stride -> edx |
644 |
|
|
645 |
|
movq mm7, [mmx_one] |
646 |
|
|
647 |
|
jnz near .rounding1 |
648 |
|
|
649 |
|
AVG4_MMX_RND0 |
650 |
|
lea ecx, [ecx+edx] |
651 |
|
AVG4_MMX_RND0 |
652 |
|
lea ecx, [ecx+edx] |
653 |
|
AVG4_MMX_RND0 |
654 |
|
lea ecx, [ecx+edx] |
655 |
|
AVG4_MMX_RND0 |
656 |
|
lea ecx, [ecx+edx] |
657 |
|
AVG4_MMX_RND0 |
658 |
|
lea ecx, [ecx+edx] |
659 |
|
AVG4_MMX_RND0 |
660 |
|
lea ecx, [ecx+edx] |
661 |
|
AVG4_MMX_RND0 |
662 |
|
lea ecx, [ecx+edx] |
663 |
|
AVG4_MMX_RND0 |
664 |
|
|
665 |
|
pop esi |
666 |
|
pop edi |
667 |
|
pop ebx |
668 |
|
ret |
669 |
|
|
670 |
|
.rounding1 |
671 |
|
AVG4_MMX_RND1 |
672 |
|
lea ecx, [ecx+edx] |
673 |
|
AVG4_MMX_RND1 |
674 |
|
lea ecx, [ecx+edx] |
675 |
|
AVG4_MMX_RND1 |
676 |
|
lea ecx, [ecx+edx] |
677 |
|
AVG4_MMX_RND1 |
678 |
|
lea ecx, [ecx+edx] |
679 |
|
AVG4_MMX_RND1 |
680 |
|
lea ecx, [ecx+edx] |
681 |
|
AVG4_MMX_RND1 |
682 |
|
lea ecx, [ecx+edx] |
683 |
|
AVG4_MMX_RND1 |
684 |
|
lea ecx, [ecx+edx] |
685 |
|
AVG4_MMX_RND1 |
686 |
|
|
687 |
|
pop esi |
688 |
|
pop edi |
689 |
|
pop ebx |
690 |
|
ret |
691 |
|
|
692 |
|
|
693 |
|
;----------------------------------------------------------------------------- |
694 |
|
; |
695 |
|
; void interpolate8x8_6tap_lowpass_h_mmx(uint8_t const *dst, |
696 |
|
; const uint8_t * const src, |
697 |
|
; const uint32_t stride, |
698 |
|
; const uint32_t rounding); |
699 |
|
; |
700 |
|
;----------------------------------------------------------------------------- |
701 |
|
|
702 |
|
%macro LOWPASS_6TAP_H_MMX 0 |
703 |
|
movq mm0, [eax] |
704 |
|
movq mm2, [eax+1] |
705 |
|
|
706 |
|
movq mm1, mm0 |
707 |
|
movq mm3, mm2 |
708 |
|
|
709 |
|
punpcklbw mm0, mm7 |
710 |
|
punpcklbw mm2, mm7 |
711 |
|
|
712 |
|
punpckhbw mm1, mm7 |
713 |
|
punpckhbw mm3, mm7 |
714 |
|
|
715 |
|
paddw mm0, mm2 |
716 |
|
paddw mm1, mm3 |
717 |
|
|
718 |
|
psllw mm0, 2 |
719 |
|
psllw mm1, 2 |
720 |
|
|
721 |
|
movq mm2, [eax-1] |
722 |
|
movq mm4, [eax+2] |
723 |
|
|
724 |
|
movq mm3, mm2 |
725 |
|
movq mm5, mm4 |
726 |
|
|
727 |
|
punpcklbw mm2, mm7 |
728 |
|
punpcklbw mm4, mm7 |
729 |
|
|
730 |
|
punpckhbw mm3, mm7 |
731 |
|
punpckhbw mm5, mm7 |
732 |
|
|
733 |
|
paddw mm2, mm4 |
734 |
|
paddw mm3, mm5 |
735 |
|
|
736 |
|
psubsw mm0, mm2 |
737 |
|
psubsw mm1, mm3 |
738 |
|
|
739 |
|
pmullw mm0, [mmx_five] |
740 |
|
pmullw mm1, [mmx_five] |
741 |
|
|
742 |
|
movq mm2, [eax-2] |
743 |
|
movq mm4, [eax+3] |
744 |
|
|
745 |
|
movq mm3, mm2 |
746 |
|
movq mm5, mm4 |
747 |
|
|
748 |
|
punpcklbw mm2, mm7 |
749 |
|
punpcklbw mm4, mm7 |
750 |
|
|
751 |
|
punpckhbw mm3, mm7 |
752 |
|
punpckhbw mm5, mm7 |
753 |
|
|
754 |
|
paddw mm2, mm4 |
755 |
|
paddw mm3, mm5 |
756 |
|
|
757 |
|
paddsw mm0, mm2 |
758 |
|
paddsw mm1, mm3 |
759 |
|
|
760 |
|
paddsw mm0, mm6 |
761 |
|
paddsw mm1, mm6 |
762 |
|
|
763 |
|
psraw mm0, 5 |
764 |
|
psraw mm1, 5 |
765 |
|
|
766 |
|
lea eax, [eax+edx] |
767 |
|
packuswb mm0, mm1 |
768 |
|
movq [ecx], mm0 |
769 |
|
%endmacro |
770 |
|
|
771 |
|
ALIGN 16 |
772 |
|
interpolate8x8_6tap_lowpass_h_mmx: |
773 |
|
|
774 |
|
mov eax, [esp + 16] ; rounding |
775 |
|
|
776 |
|
movq mm6, [rounding_lowpass_mmx + eax * 8] |
777 |
|
|
778 |
|
mov ecx, [esp + 4] ; dst -> edi |
779 |
|
mov eax, [esp + 8] ; src -> esi |
780 |
|
mov edx, [esp + 12] ; stride -> edx |
781 |
|
|
782 |
|
pxor mm7, mm7 |
783 |
|
|
784 |
|
LOWPASS_6TAP_H_MMX |
785 |
|
lea ecx, [ecx+edx] |
786 |
|
LOWPASS_6TAP_H_MMX |
787 |
|
lea ecx, [ecx+edx] |
788 |
|
LOWPASS_6TAP_H_MMX |
789 |
|
lea ecx, [ecx+edx] |
790 |
|
LOWPASS_6TAP_H_MMX |
791 |
|
lea ecx, [ecx+edx] |
792 |
|
LOWPASS_6TAP_H_MMX |
793 |
|
lea ecx, [ecx+edx] |
794 |
|
LOWPASS_6TAP_H_MMX |
795 |
|
lea ecx, [ecx+edx] |
796 |
|
LOWPASS_6TAP_H_MMX |
797 |
|
lea ecx, [ecx+edx] |
798 |
|
LOWPASS_6TAP_H_MMX |
799 |
|
|
800 |
|
ret |
801 |
|
|
802 |
|
;----------------------------------------------------------------------------- |
803 |
|
; |
804 |
|
; void interpolate8x8_6tap_lowpass_v_mmx(uint8_t const *dst, |
805 |
|
; const uint8_t * const src, |
806 |
|
; const uint32_t stride, |
807 |
|
; const uint32_t rounding); |
808 |
|
; |
809 |
|
;----------------------------------------------------------------------------- |
810 |
|
|
811 |
|
%macro LOWPASS_6TAP_V_MMX 0 |
812 |
|
movq mm0, [eax] |
813 |
|
movq mm2, [eax+edx] |
814 |
|
|
815 |
|
movq mm1, mm0 |
816 |
|
movq mm3, mm2 |
817 |
|
|
818 |
|
punpcklbw mm0, mm7 |
819 |
|
punpcklbw mm2, mm7 |
820 |
|
|
821 |
|
punpckhbw mm1, mm7 |
822 |
|
punpckhbw mm3, mm7 |
823 |
|
|
824 |
|
paddw mm0, mm2 |
825 |
|
paddw mm1, mm3 |
826 |
|
|
827 |
|
psllw mm0, 2 |
828 |
|
psllw mm1, 2 |
829 |
|
|
830 |
|
movq mm4, [eax+2*edx] |
831 |
|
sub eax, ebx |
832 |
|
movq mm2, [eax+2*edx] |
833 |
|
|
834 |
|
movq mm3, mm2 |
835 |
|
movq mm5, mm4 |
836 |
|
|
837 |
|
punpcklbw mm2, mm7 |
838 |
|
punpcklbw mm4, mm7 |
839 |
|
|
840 |
|
punpckhbw mm3, mm7 |
841 |
|
punpckhbw mm5, mm7 |
842 |
|
|
843 |
|
paddw mm2, mm4 |
844 |
|
paddw mm3, mm5 |
845 |
|
|
846 |
|
psubsw mm0, mm2 |
847 |
|
psubsw mm1, mm3 |
848 |
|
|
849 |
|
pmullw mm0, [mmx_five] |
850 |
|
pmullw mm1, [mmx_five] |
851 |
|
|
852 |
|
movq mm2, [eax+edx] |
853 |
|
movq mm4, [eax+2*ebx] |
854 |
|
|
855 |
|
movq mm3, mm2 |
856 |
|
movq mm5, mm4 |
857 |
|
|
858 |
|
punpcklbw mm2, mm7 |
859 |
|
punpcklbw mm4, mm7 |
860 |
|
|
861 |
|
punpckhbw mm3, mm7 |
862 |
|
punpckhbw mm5, mm7 |
863 |
|
|
864 |
|
paddw mm2, mm4 |
865 |
|
paddw mm3, mm5 |
866 |
|
|
867 |
|
paddsw mm0, mm2 |
868 |
|
paddsw mm1, mm3 |
869 |
|
|
870 |
|
paddsw mm0, mm6 |
871 |
|
paddsw mm1, mm6 |
872 |
|
|
873 |
|
psraw mm0, 5 |
874 |
|
psraw mm1, 5 |
875 |
|
|
876 |
|
lea eax, [eax+4*edx] |
877 |
|
packuswb mm0, mm1 |
878 |
|
movq [ecx], mm0 |
879 |
|
%endmacro |
880 |
|
|
881 |
|
ALIGN 16 |
882 |
|
interpolate8x8_6tap_lowpass_v_mmx: |
883 |
|
|
884 |
|
push ebx |
885 |
|
|
886 |
|
mov eax, [esp + 4 + 16] ; rounding |
887 |
|
|
888 |
|
movq mm6, [rounding_lowpass_mmx + eax * 8] |
889 |
|
|
890 |
|
mov ecx, [esp + 4 + 4] ; dst -> edi |
891 |
|
mov eax, [esp + 4 + 8] ; src -> esi |
892 |
|
mov edx, [esp + 4 + 12] ; stride -> edx |
893 |
|
|
894 |
|
mov ebx, edx |
895 |
|
shl ebx, 1 |
896 |
|
add ebx, edx |
897 |
|
|
898 |
|
pxor mm7, mm7 |
899 |
|
|
900 |
|
LOWPASS_6TAP_V_MMX |
901 |
|
lea ecx, [ecx+edx] |
902 |
|
LOWPASS_6TAP_V_MMX |
903 |
|
lea ecx, [ecx+edx] |
904 |
|
LOWPASS_6TAP_V_MMX |
905 |
|
lea ecx, [ecx+edx] |
906 |
|
LOWPASS_6TAP_V_MMX |
907 |
|
lea ecx, [ecx+edx] |
908 |
|
LOWPASS_6TAP_V_MMX |
909 |
|
lea ecx, [ecx+edx] |
910 |
|
LOWPASS_6TAP_V_MMX |
911 |
|
lea ecx, [ecx+edx] |
912 |
|
LOWPASS_6TAP_V_MMX |
913 |
|
lea ecx, [ecx+edx] |
914 |
|
LOWPASS_6TAP_V_MMX |
915 |
|
|
916 |
|
pop ebx |
917 |
|
ret |
918 |
|
|
919 |
|
;=========================================================================== |
920 |
|
; |
921 |
|
; The next functions combine both source halfpel interpolation step and the |
922 |
|
; averaging (with rouding) step to avoid wasting memory bandwidth computing |
923 |
|
; intermediate halfpel images and then averaging them. |
924 |
|
; |
925 |
|
;=========================================================================== |
926 |
|
|
927 |
|
%macro PROLOG0 0 |
928 |
|
mov ecx, [esp+ 4] ; Dst |
929 |
|
mov eax, [esp+ 8] ; Src |
930 |
|
mov edx, [esp+12] ; BpS |
931 |
|
%endmacro |
932 |
|
|
933 |
|
%macro PROLOG 2 ; %1: Rounder, %2 load Dst-Rounder |
934 |
|
pxor mm6, mm6 |
935 |
|
movq mm7, [%1] ; TODO: dangerous! (eax isn't checked) |
936 |
|
%if %2 |
937 |
|
movq mm5, [rounding1_mmx] |
938 |
|
%endif |
939 |
|
|
940 |
|
PROLOG0 |
941 |
|
%endmacro |
942 |
|
|
943 |
|
; performs: mm0 == (mm0+mm2) mm1 == (mm1+mm3) |
944 |
|
%macro MIX 0 |
945 |
|
punpcklbw mm0, mm6 |
946 |
|
punpcklbw mm2, mm6 |
947 |
|
punpckhbw mm1, mm6 |
948 |
|
punpckhbw mm3, mm6 |
949 |
|
paddusw mm0, mm2 |
950 |
|
paddusw mm1, mm3 |
951 |
|
%endmacro |
952 |
|
|
953 |
|
%macro MIX_DST 0 |
954 |
|
movq mm3, mm2 |
955 |
|
paddusw mm0, mm7 ; rounder |
956 |
|
paddusw mm1, mm7 ; rounder |
957 |
|
punpcklbw mm2, mm6 |
958 |
|
punpckhbw mm3, mm6 |
959 |
|
psrlw mm0, 1 |
960 |
|
psrlw mm1, 1 |
961 |
|
|
962 |
|
paddusw mm0, mm2 ; mix Src(mm0/mm1) with Dst(mm2/mm3) |
963 |
|
paddusw mm1, mm3 |
964 |
|
paddusw mm0, mm5 |
965 |
|
paddusw mm1, mm5 |
966 |
|
psrlw mm0, 1 |
967 |
|
psrlw mm1, 1 |
968 |
|
|
969 |
|
packuswb mm0, mm1 |
970 |
|
%endmacro |
971 |
|
|
972 |
|
%macro MIX2 0 |
973 |
|
punpcklbw mm0, mm6 |
974 |
|
punpcklbw mm2, mm6 |
975 |
|
paddusw mm0, mm2 |
976 |
|
paddusw mm0, mm7 |
977 |
|
punpckhbw mm1, mm6 |
978 |
|
punpckhbw mm3, mm6 |
979 |
|
paddusw mm1, mm7 |
980 |
|
paddusw mm1, mm3 |
981 |
|
psrlw mm0, 1 |
982 |
|
psrlw mm1, 1 |
983 |
|
|
984 |
|
packuswb mm0, mm1 |
985 |
|
%endmacro |
986 |
|
|
987 |
|
;=========================================================================== |
988 |
|
; |
989 |
|
; void interpolate8x8_halfpel_add_mmx(uint8_t * const dst, |
990 |
|
; const uint8_t * const src, |
991 |
|
; const uint32_t stride, |
992 |
|
; const uint32_t rounding); |
993 |
|
; |
994 |
|
; |
995 |
|
;=========================================================================== |
996 |
|
|
997 |
|
%macro ADD_FF_MMX 1 |
998 |
|
movq mm0, [eax] |
999 |
|
movq mm2, [ecx] |
1000 |
|
movq mm1, mm0 |
1001 |
|
movq mm3, mm2 |
1002 |
|
%if (%1!=0) |
1003 |
|
lea eax,[eax+%1*edx] |
1004 |
|
%endif |
1005 |
|
MIX |
1006 |
|
paddusw mm0, mm5 ; rounder |
1007 |
|
paddusw mm1, mm5 ; rounder |
1008 |
|
psrlw mm0, 1 |
1009 |
|
psrlw mm1, 1 |
1010 |
|
|
1011 |
|
packuswb mm0, mm1 |
1012 |
|
movq [ecx], mm0 |
1013 |
|
%if (%1!=0) |
1014 |
|
lea ecx,[ecx+%1*edx] |
1015 |
|
%endif |
1016 |
|
%endmacro |
1017 |
|
|
1018 |
|
ALIGN 16 |
1019 |
|
interpolate8x8_halfpel_add_mmx: |
1020 |
|
PROLOG rounding1_mmx, 1 |
1021 |
|
ADD_FF_MMX 1 |
1022 |
|
ADD_FF_MMX 1 |
1023 |
|
ADD_FF_MMX 1 |
1024 |
|
ADD_FF_MMX 1 |
1025 |
|
ADD_FF_MMX 1 |
1026 |
|
ADD_FF_MMX 1 |
1027 |
|
ADD_FF_MMX 1 |
1028 |
|
ADD_FF_MMX 0 |
1029 |
|
ret |
1030 |
|
|
1031 |
|
;=========================================================================== |
1032 |
|
; |
1033 |
|
; void interpolate8x8_halfpel_h_add_mmx(uint8_t * const dst, |
1034 |
|
; const uint8_t * const src, |
1035 |
|
; const uint32_t stride, |
1036 |
|
; const uint32_t rounding); |
1037 |
|
; |
1038 |
|
; |
1039 |
|
;=========================================================================== |
1040 |
|
|
1041 |
|
%macro ADD_FH_MMX 0 |
1042 |
|
movq mm0, [eax] |
1043 |
|
movq mm2, [eax+1] |
1044 |
|
movq mm1, mm0 |
1045 |
|
movq mm3, mm2 |
1046 |
|
|
1047 |
|
lea eax,[eax+edx] |
1048 |
|
|
1049 |
|
MIX |
1050 |
|
movq mm2, [ecx] ; prepare mix with Dst[0] |
1051 |
|
MIX_DST |
1052 |
|
movq [ecx], mm0 |
1053 |
|
%endmacro |
1054 |
|
|
1055 |
|
ALIGN 16 |
1056 |
|
interpolate8x8_halfpel_h_add_mmx: |
1057 |
|
PROLOG rounding1_mmx, 1 |
1058 |
|
|
1059 |
|
ADD_FH_MMX |
1060 |
|
lea ecx,[ecx+edx] |
1061 |
|
ADD_FH_MMX |
1062 |
|
lea ecx,[ecx+edx] |
1063 |
|
ADD_FH_MMX |
1064 |
|
lea ecx,[ecx+edx] |
1065 |
|
ADD_FH_MMX |
1066 |
|
lea ecx,[ecx+edx] |
1067 |
|
ADD_FH_MMX |
1068 |
|
lea ecx,[ecx+edx] |
1069 |
|
ADD_FH_MMX |
1070 |
|
lea ecx,[ecx+edx] |
1071 |
|
ADD_FH_MMX |
1072 |
|
lea ecx,[ecx+edx] |
1073 |
|
ADD_FH_MMX |
1074 |
|
ret |
1075 |
|
|
1076 |
|
;=========================================================================== |
1077 |
|
; |
1078 |
|
; void interpolate8x8_halfpel_v_add_mmx(uint8_t * const dst, |
1079 |
|
; const uint8_t * const src, |
1080 |
|
; const uint32_t stride, |
1081 |
|
; const uint32_t rounding); |
1082 |
|
; |
1083 |
|
; |
1084 |
|
;=========================================================================== |
1085 |
|
|
1086 |
|
%macro ADD_HF_MMX 0 |
1087 |
|
movq mm0, [eax] |
1088 |
|
movq mm2, [eax+edx] |
1089 |
|
movq mm1, mm0 |
1090 |
|
movq mm3, mm2 |
1091 |
|
|
1092 |
|
lea eax,[eax+edx] |
1093 |
|
|
1094 |
|
MIX |
1095 |
|
movq mm2, [ecx] ; prepare mix with Dst[0] |
1096 |
|
MIX_DST |
1097 |
|
movq [ecx], mm0 |
1098 |
|
|
1099 |
|
%endmacro |
1100 |
|
|
1101 |
|
ALIGN 16 |
1102 |
|
interpolate8x8_halfpel_v_add_mmx: |
1103 |
|
PROLOG rounding1_mmx, 1 |
1104 |
|
|
1105 |
|
ADD_HF_MMX |
1106 |
|
lea ecx,[ecx+edx] |
1107 |
|
ADD_HF_MMX |
1108 |
|
lea ecx,[ecx+edx] |
1109 |
|
ADD_HF_MMX |
1110 |
|
lea ecx,[ecx+edx] |
1111 |
|
ADD_HF_MMX |
1112 |
|
lea ecx,[ecx+edx] |
1113 |
|
ADD_HF_MMX |
1114 |
|
lea ecx,[ecx+edx] |
1115 |
|
ADD_HF_MMX |
1116 |
|
lea ecx,[ecx+edx] |
1117 |
|
ADD_HF_MMX |
1118 |
|
lea ecx,[ecx+edx] |
1119 |
|
ADD_HF_MMX |
1120 |
|
ret |
1121 |
|
|
1122 |
|
; The trick is to correct the result of 'pavgb' with some combination of the |
1123 |
|
; lsb's of the 4 input values i,j,k,l, and their intermediate 'pavgb' (s and t). |
1124 |
|
; The boolean relations are: |
1125 |
|
; (i+j+k+l+3)/4 = (s+t+1)/2 - (ij&kl)&st |
1126 |
|
; (i+j+k+l+2)/4 = (s+t+1)/2 - (ij|kl)&st |
1127 |
|
; (i+j+k+l+1)/4 = (s+t+1)/2 - (ij&kl)|st |
1128 |
|
; (i+j+k+l+0)/4 = (s+t+1)/2 - (ij|kl)|st |
1129 |
|
; with s=(i+j+1)/2, t=(k+l+1)/2, ij = i^j, kl = k^l, st = s^t. |
1130 |
|
|
1131 |
|
; Moreover, we process 2 lines at a times, for better overlapping (~15% faster). |
1132 |
|
|
1133 |
|
;=========================================================================== |
1134 |
|
; |
1135 |
|
; void interpolate8x8_halfpel_hv_add_mmx(uint8_t * const dst, |
1136 |
|
; const uint8_t * const src, |
1137 |
|
; const uint32_t stride, |
1138 |
|
; const uint32_t rounding); |
1139 |
|
; |
1140 |
|
; |
1141 |
|
;=========================================================================== |
1142 |
|
|
1143 |
|
%macro ADD_HH_MMX 0 |
1144 |
|
lea eax,[eax+edx] |
1145 |
|
|
1146 |
|
; transfert prev line to mm0/mm1 |
1147 |
|
movq mm0, mm2 |
1148 |
|
movq mm1, mm3 |
1149 |
|
|
1150 |
|
; load new line in mm2/mm3 |
1151 |
|
movq mm2, [eax] |
1152 |
|
movq mm4, [eax+1] |
1153 |
|
movq mm3, mm2 |
1154 |
|
movq mm5, mm4 |
1155 |
|
|
1156 |
|
punpcklbw mm2, mm6 |
1157 |
|
punpcklbw mm4, mm6 |
1158 |
|
paddusw mm2, mm4 |
1159 |
|
punpckhbw mm3, mm6 |
1160 |
|
punpckhbw mm5, mm6 |
1161 |
|
paddusw mm3, mm5 |
1162 |
|
|
1163 |
|
; mix current line (mm2/mm3) with previous (mm0,mm1); |
1164 |
|
; we'll preserve mm2/mm3 for next line... |
1165 |
|
|
1166 |
|
paddusw mm0, mm2 |
1167 |
|
paddusw mm1, mm3 |
1168 |
|
|
1169 |
|
movq mm4, [ecx] ; prepare mix with Dst[0] |
1170 |
|
movq mm5, mm4 |
1171 |
|
|
1172 |
|
paddusw mm0, mm7 ; finish mixing current line |
1173 |
|
paddusw mm1, mm7 |
1174 |
|
|
1175 |
|
punpcklbw mm4, mm6 |
1176 |
|
punpckhbw mm5, mm6 |
1177 |
|
|
1178 |
|
psrlw mm0, 2 |
1179 |
|
psrlw mm1, 2 |
1180 |
|
|
1181 |
|
paddusw mm0, mm4 ; mix Src(mm0/mm1) with Dst(mm2/mm3) |
1182 |
|
paddusw mm1, mm5 |
1183 |
|
|
1184 |
|
paddusw mm0, [rounding1_mmx] |
1185 |
|
paddusw mm1, [rounding1_mmx] |
1186 |
|
|
1187 |
|
psrlw mm0, 1 |
1188 |
|
psrlw mm1, 1 |
1189 |
|
|
1190 |
|
packuswb mm0, mm1 |
1191 |
|
|
1192 |
|
movq [ecx], mm0 |
1193 |
|
%endmacro |
1194 |
|
|
1195 |
|
ALIGN 16 |
1196 |
|
interpolate8x8_halfpel_hv_add_mmx: |
1197 |
|
PROLOG rounding2_mmx, 0 ; mm5 is busy. Don't load dst-rounder |
1198 |
|
|
1199 |
|
; preprocess first line |
1200 |
|
movq mm0, [eax] |
1201 |
|
movq mm2, [eax+1] |
1202 |
|
movq mm1, mm0 |
1203 |
|
movq mm3, mm2 |
1204 |
|
|
1205 |
|
punpcklbw mm0, mm6 |
1206 |
|
punpcklbw mm2, mm6 |
1207 |
|
punpckhbw mm1, mm6 |
1208 |
|
punpckhbw mm3, mm6 |
1209 |
|
paddusw mm2, mm0 |
1210 |
|
paddusw mm3, mm1 |
1211 |
|
|
1212 |
|
; Input: mm2/mm3 contains the value (Src[0]+Src[1]) of previous line |
1213 |
|
|
1214 |
|
ADD_HH_MMX |
1215 |
|
lea ecx,[ecx+edx] |
1216 |
|
ADD_HH_MMX |
1217 |
|
lea ecx,[ecx+edx] |
1218 |
|
ADD_HH_MMX |
1219 |
|
lea ecx,[ecx+edx] |
1220 |
|
ADD_HH_MMX |
1221 |
|
lea ecx,[ecx+edx] |
1222 |
|
ADD_HH_MMX |
1223 |
|
lea ecx,[ecx+edx] |
1224 |
|
ADD_HH_MMX |
1225 |
|
lea ecx,[ecx+edx] |
1226 |
|
ADD_HH_MMX |
1227 |
|
lea ecx,[ecx+edx] |
1228 |
|
ADD_HH_MMX |
1229 |
|
|
1230 |
|
ret |
1231 |
|
|