1 |
;/***************************************************************************** |
;/************************************************************************** |
2 |
; * |
; * |
3 |
; * XVID MPEG-4 VIDEO CODEC |
; * XVID MPEG-4 VIDEO CODEC |
4 |
; * xmm (extended mmx) sum of absolute difference |
; * xmm sum of absolute difference |
|
; * |
|
|
; * Copyright(C) 2002 Peter Ross <pross@xvid.org> |
|
|
; * Copyright(C) 2002 Michael Militzer <michael@xvid.org> |
|
|
; * Copyright(C) 2002 Pascal Massimino <skal@planet-d.net> |
|
5 |
; * |
; * |
6 |
; * This program is an implementation of a part of one or more MPEG-4 |
; * This program is an implementation of a part of one or more MPEG-4 |
7 |
; * Video tools as specified in ISO/IEC 14496-2 standard. Those intending |
; * Video tools as specified in ISO/IEC 14496-2 standard. Those intending |
24 |
; * |
; * |
25 |
; * You should have received a copy of the GNU General Public License |
; * You should have received a copy of the GNU General Public License |
26 |
; * along with this program; if not, write to the Free Software |
; * along with this program; if not, write to the Free Software |
27 |
; * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA |
; * Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA. |
28 |
|
; * |
29 |
|
; *************************************************************************/ |
30 |
|
|
31 |
|
;/************************************************************************** |
32 |
|
; * |
33 |
|
; * History: |
34 |
|
; * |
35 |
|
; * 23.07.2002 sad8bi_xmm; <pross@xvid.org> |
36 |
|
; * 04.06.2002 rewrote some funcs (XMM mainly) -Skal- |
37 |
|
; * 17.11.2001 bugfix and small improvement for dev16_xmm, |
38 |
|
; * removed terminate early in sad16_xmm (Isibaar) |
39 |
|
; * 12.11.2001 inital version; (c)2001 peter ross <pross@cs.rmit.edu.au> |
40 |
; * |
; * |
41 |
; *************************************************************************/ |
; *************************************************************************/ |
42 |
|
|
357 |
|
|
358 |
mov eax, [esp+ 4] ; Src |
mov eax, [esp+ 4] ; Src |
359 |
|
|
360 |
|
|
361 |
pxor mm5, mm5 ; sums |
pxor mm5, mm5 ; sums |
362 |
pxor mm6, mm6 |
pxor mm6, mm6 |
363 |
|
|
386 |
|
|
387 |
movd eax, mm6 |
movd eax, mm6 |
388 |
ret |
ret |
389 |
|
|
390 |
|
cglobal sad16v_xmm |
391 |
|
|
392 |
|
;=========================================================================== |
393 |
|
;int sad16v_xmm(const uint8_t * const cur, |
394 |
|
; const uint8_t * const ref, |
395 |
|
; const uint32_t stride, |
396 |
|
; int* sad8); |
397 |
|
;=========================================================================== |
398 |
|
align 16 |
399 |
|
sad16v_xmm: |
400 |
|
push ebx |
401 |
|
mov eax, [esp+4+ 4] ; Src1 |
402 |
|
mov edx, [esp+4+ 8] ; Src2 |
403 |
|
mov ecx, [esp+4+12] ; Stride |
404 |
|
mov ebx, [esp+4+16] ; sad ptr |
405 |
|
|
406 |
|
pxor mm5, mm5 ; accum1 |
407 |
|
pxor mm6, mm6 ; accum2 |
408 |
|
pxor mm7, mm7 ; total |
409 |
|
SAD_16x16_SSE |
410 |
|
SAD_16x16_SSE |
411 |
|
SAD_16x16_SSE |
412 |
|
SAD_16x16_SSE |
413 |
|
SAD_16x16_SSE |
414 |
|
SAD_16x16_SSE |
415 |
|
SAD_16x16_SSE |
416 |
|
SAD_16x16_SSE |
417 |
|
paddusw mm7, mm5 |
418 |
|
paddusw mm7, mm6 |
419 |
|
movd [ebx], mm5 |
420 |
|
movd [ebx+4], mm6 |
421 |
|
|
422 |
|
pxor mm5, mm5 ; accum1 |
423 |
|
pxor mm6, mm6 ; accum2 |
424 |
|
SAD_16x16_SSE |
425 |
|
SAD_16x16_SSE |
426 |
|
SAD_16x16_SSE |
427 |
|
SAD_16x16_SSE |
428 |
|
SAD_16x16_SSE |
429 |
|
SAD_16x16_SSE |
430 |
|
SAD_16x16_SSE |
431 |
|
SAD_16x16_SSE |
432 |
|
paddusw mm7, mm5 |
433 |
|
paddusw mm7, mm6 |
434 |
|
movd [ebx+8], mm5 |
435 |
|
movd [ebx+12], mm6 |
436 |
|
|
437 |
|
movd eax, mm7 |
438 |
|
pop ebx |
439 |
|
ret |
440 |
|
;-------- |
441 |
|
|
442 |
|
|