1 |
;/************************************************************************** |
;/***************************************************************************** |
2 |
; * |
; * |
3 |
; * XVID MPEG-4 VIDEO CODEC |
; * XVID MPEG-4 VIDEO CODEC |
4 |
; * xmm 8x8 block-based halfpel interpolation |
; * - mmx 8x8 block-based halfpel interpolation - |
5 |
|
; * |
6 |
|
; * Copyright(C) 2002 Michael Militzer <isibaar@xvid.org> |
7 |
|
; * 2002 Pascal Massimino <skal@planet-d.net> |
8 |
; * |
; * |
9 |
; * This program is free software; you can redistribute it and/or modify |
; * 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 |
; * it under the terms of the GNU General Public License as published by |
18 |
; * |
; * |
19 |
; * You should have received a copy of the GNU General Public License |
; * You should have received a copy of the GNU General Public License |
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., 675 Mass Ave, Cambridge, MA 02139, USA. |
; * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA |
|
; * |
|
|
; *************************************************************************/ |
|
|
|
|
|
;/************************************************************************** |
|
|
; * |
|
|
; * History: |
|
22 |
; * |
; * |
23 |
; * 04.06.2002 rewrote some funcs, mostly XMM. -Skal- |
; ****************************************************************************/ |
|
; * Heavily tuned for overlap and AGI-stalls avoidance |
|
|
; * 04.02.2002 initial version (Isibaar) |
|
|
; * |
|
|
; *************************************************************************/ |
|
|
|
|
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 |
|
SECTION .rodata |
41 |
|
|
42 |
align 16 |
ALIGN 16 |
43 |
|
mmx_one: |
|
mmx_one |
|
44 |
times 8 db 1 |
times 8 db 1 |
45 |
|
|
46 |
section .text |
SECTION .text |
47 |
|
|
48 |
cglobal interpolate8x8_halfpel_h_xmm |
cglobal interpolate8x8_halfpel_h_xmm |
49 |
cglobal interpolate8x8_halfpel_v_xmm |
cglobal interpolate8x8_halfpel_v_xmm |
88 |
movq [ecx+edx], mm1 |
movq [ecx+edx], mm1 |
89 |
%endmacro |
%endmacro |
90 |
|
|
91 |
align 16 |
ALIGN 16 |
92 |
interpolate8x8_halfpel_h_xmm: |
interpolate8x8_halfpel_h_xmm: |
93 |
|
|
94 |
mov eax, [esp+16]; rounding |
mov eax, [esp+16]; rounding |
158 |
movq [ecx+edx], mm1 |
movq [ecx+edx], mm1 |
159 |
%endmacro |
%endmacro |
160 |
|
|
161 |
align 16 |
ALIGN 16 |
162 |
interpolate8x8_halfpel_v_xmm: |
interpolate8x8_halfpel_v_xmm: |
163 |
|
|
164 |
mov eax, [esp+16]; rounding |
mov eax, [esp+16]; rounding |
168 |
mov edx, [esp+12] ; stride |
mov edx, [esp+12] ; stride |
169 |
|
|
170 |
; we process 2 line at a time |
; we process 2 line at a time |
|
|
|
171 |
jnz near .rounding1 |
jnz near .rounding1 |
172 |
|
|
173 |
COPY_V_SSE_RND0 |
COPY_V_SSE_RND0 |
294 |
movq [ecx], mm0 |
movq [ecx], mm0 |
295 |
%endmacro |
%endmacro |
296 |
|
|
297 |
align 16 |
ALIGN 16 |
298 |
interpolate8x8_halfpel_hv_xmm: |
interpolate8x8_halfpel_hv_xmm: |
299 |
mov eax, [esp+16] ; rounding |
mov eax, [esp+16] ; rounding |
300 |
mov ecx, [esp+ 4] ; Dst |
mov ecx, [esp+ 4] ; Dst |