1 |
;/************************************************************************** |
;/**************************************************************************** |
2 |
; * |
; * |
3 |
; * XVID MPEG-4 VIDEO CODEC |
; * XVID MPEG-4 VIDEO CODEC |
4 |
; * 3dnow (*but without xmm*) sum of absolute difference |
; * - 3DNow sad operators w/o XMM instructions - |
5 |
; * |
; * |
6 |
; * This program is free software; you can redistribute it and/or modify |
; * Copyright(C) 2002 Peter ross <pross@xvid.org> |
7 |
; * it under the terms of the GNU General Public License as published by |
; * |
8 |
|
; * This program is free software; you can redistribute it and/or modify it |
9 |
|
; * under the terms of the GNU General Public License as published by |
10 |
; * the Free Software Foundation; either version 2 of the License, or |
; * the Free Software Foundation; either version 2 of the License, or |
11 |
; * (at your option) any later version. |
; * (at your option) any later version. |
12 |
; * |
; * |
17 |
; * |
; * |
18 |
; * You should have received a copy of the GNU General Public License |
; * You should have received a copy of the GNU General Public License |
19 |
; * along with this program; if not, write to the Free Software |
; * along with this program; if not, write to the Free Software |
20 |
; * Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA. |
; * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA |
|
; * |
|
|
; *************************************************************************/ |
|
|
|
|
|
;/************************************************************************** |
|
|
; * |
|
|
; * History: |
|
21 |
; * |
; * |
22 |
; * 23.07.2002 sad[16,8]bi_3dn; <pross@xvid.org> |
; * $Id$ |
23 |
; * |
; * |
24 |
; *************************************************************************/ |
; ***************************************************************************/ |
25 |
|
|
26 |
bits 32 |
BITS 32 |
27 |
|
|
28 |
%macro cglobal 1 |
%macro cglobal 1 |
29 |
%ifdef PREFIX |
%ifdef PREFIX |
30 |
|
%ifdef MARK_FUNCS |
31 |
|
global _%1:function |
32 |
|
%define %1 _%1:function |
33 |
|
%else |
34 |
global _%1 |
global _%1 |
35 |
%define %1 _%1 |
%define %1 _%1 |
36 |
|
%endif |
37 |
|
%else |
38 |
|
%ifdef MARK_FUNCS |
39 |
|
global %1:function |
40 |
%else |
%else |
41 |
global %1 |
global %1 |
42 |
%endif |
%endif |
43 |
|
%endif |
44 |
%endmacro |
%endmacro |
45 |
|
|
46 |
section .data |
;============================================================================= |
47 |
|
; Read only data |
48 |
align 16 |
;============================================================================= |
|
mmx_one times 4 dw 1 |
|
49 |
|
|
50 |
section .text |
%ifdef FORMAT_COFF |
51 |
|
SECTION .rodata |
52 |
cglobal sad16bi_3dn |
%else |
53 |
cglobal sad8bi_3dn |
SECTION .rodata align=16 |
54 |
|
%endif |
|
;=========================================================================== |
|
|
; |
|
|
; uint32_t sad16bi_3dn(const uint8_t * const cur, |
|
|
; const uint8_t * const ref1, |
|
|
; const uint8_t * const ref2, |
|
|
; const uint32_t stride); |
|
|
; |
|
|
;=========================================================================== |
|
55 |
|
|
56 |
|
ALIGN 16 |
57 |
|
mmx_one: |
58 |
|
times 4 dw 1 |
59 |
|
|
60 |
|
;============================================================================= |
61 |
|
; Helper macros |
62 |
|
;============================================================================= |
63 |
%macro SADBI_16x16_3DN 0 |
%macro SADBI_16x16_3DN 0 |
64 |
movq mm0, [eax] ; src |
movq mm0, [eax] ; src |
65 |
movq mm2, [eax+8] |
movq mm2, [eax+8] |
96 |
paddusw mm6,mm2 |
paddusw mm6,mm2 |
97 |
%endmacro |
%endmacro |
98 |
|
|
99 |
align 16 |
%macro SADBI_8x8_3DN 0 |
100 |
|
movq mm0, [eax] ; src |
101 |
|
movq mm2, [eax+ecx] |
102 |
|
|
103 |
|
movq mm1, [edx] ; ref1 |
104 |
|
movq mm3, [edx+ecx] |
105 |
|
pavgusb mm1, [ebx] ; ref2 |
106 |
|
lea edx, [edx+2*ecx] |
107 |
|
pavgusb mm3, [ebx+ecx] |
108 |
|
lea ebx, [ebx+2*ecx] |
109 |
|
|
110 |
|
movq mm4, mm0 |
111 |
|
lea eax, [eax+2*ecx] |
112 |
|
psubusb mm0, mm1 |
113 |
|
movq mm5, mm2 |
114 |
|
psubusb mm2, mm3 |
115 |
|
|
116 |
|
psubusb mm1, mm4 |
117 |
|
por mm0, mm1 |
118 |
|
psubusb mm3, mm5 |
119 |
|
por mm2, mm3 |
120 |
|
|
121 |
|
movq mm1, mm0 |
122 |
|
movq mm3, mm2 |
123 |
|
|
124 |
|
punpcklbw mm0,mm7 |
125 |
|
punpckhbw mm1,mm7 |
126 |
|
punpcklbw mm2,mm7 |
127 |
|
punpckhbw mm3,mm7 |
128 |
|
|
129 |
|
paddusw mm0,mm1 |
130 |
|
paddusw mm2,mm3 |
131 |
|
paddusw mm6,mm0 |
132 |
|
paddusw mm6,mm2 |
133 |
|
%endmacro |
134 |
|
|
135 |
|
;============================================================================= |
136 |
|
; Code |
137 |
|
;============================================================================= |
138 |
|
|
139 |
|
SECTION .text |
140 |
|
|
141 |
|
cglobal sad16bi_3dn |
142 |
|
cglobal sad8bi_3dn |
143 |
|
|
144 |
|
;----------------------------------------------------------------------------- |
145 |
|
; |
146 |
|
; uint32_t sad16bi_3dn(const uint8_t * const cur, |
147 |
|
; const uint8_t * const ref1, |
148 |
|
; const uint8_t * const ref2, |
149 |
|
; const uint32_t stride); |
150 |
|
; |
151 |
|
;----------------------------------------------------------------------------- |
152 |
|
|
153 |
|
ALIGN 16 |
154 |
sad16bi_3dn: |
sad16bi_3dn: |
155 |
push ebx |
push ebx |
156 |
mov eax, [esp+4+ 4] ; Src |
mov eax, [esp+4+ 4] ; Src |
190 |
|
|
191 |
ret |
ret |
192 |
|
|
193 |
|
;----------------------------------------------------------------------------- |
|
|
|
|
;=========================================================================== |
|
194 |
; |
; |
195 |
; uint32_t sad8bi_3dn(const uint8_t * const cur, |
; uint32_t sad8bi_3dn(const uint8_t * const cur, |
196 |
; const uint8_t * const ref1, |
; const uint8_t * const ref1, |
197 |
; const uint8_t * const ref2, |
; const uint8_t * const ref2, |
198 |
; const uint32_t stride); |
; const uint32_t stride); |
199 |
; |
; |
200 |
;=========================================================================== |
;----------------------------------------------------------------------------- |
|
|
|
|
%macro SADBI_8x8_3DN 0 |
|
|
movq mm0, [eax] ; src |
|
|
movq mm2, [eax+ecx] |
|
|
|
|
|
movq mm1, [edx] ; ref1 |
|
|
movq mm3, [edx+ecx] |
|
|
pavgusb mm1, [ebx] ; ref2 |
|
|
lea edx,[edx+2*ecx] |
|
|
pavgusb mm3, [ebx+ecx] |
|
|
lea ebx,[ebx+2*ecx] |
|
|
|
|
|
movq mm4, mm0 |
|
|
lea eax,[eax+2*ecx] |
|
|
psubusb mm0, mm1 |
|
|
movq mm5, mm2 |
|
|
psubusb mm2, mm3 |
|
|
|
|
|
psubusb mm1, mm4 |
|
|
por mm0, mm1 |
|
|
psubusb mm3, mm5 |
|
|
por mm2, mm3 |
|
|
|
|
|
movq mm1,mm0 |
|
|
movq mm3,mm2 |
|
|
|
|
|
punpcklbw mm0,mm7 |
|
|
punpckhbw mm1,mm7 |
|
|
punpcklbw mm2,mm7 |
|
|
punpckhbw mm3,mm7 |
|
|
|
|
|
paddusw mm0,mm1 |
|
|
paddusw mm2,mm3 |
|
|
paddusw mm6,mm0 |
|
|
paddusw mm6,mm2 |
|
|
%endmacro |
|
201 |
|
|
202 |
align 16 |
ALIGN 16 |
203 |
sad8bi_3dn: |
sad8bi_3dn: |
204 |
push ebx |
push ebx |
205 |
mov eax, [esp+4+ 4] ; Src |
mov eax, [esp+4+ 4] ; Src |