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 |
; * Copyright(C) 2002 Peter Ross <pross@xvid.org> |
; * Copyright(C) 2002 Peter ross <pross@xvid.org> |
7 |
; * |
; * |
8 |
; * This program is an implementation of a part of one or more MPEG-4 |
; * This program is free software; you can redistribute it and/or modify it |
9 |
; * Video tools as specified in ISO/IEC 14496-2 standard. Those intending |
; * under the terms of the GNU General Public License as published by |
|
; * to use this software module in hardware or software products are |
|
|
; * advised that its use may infringe existing patents or copyrights, and |
|
|
; * any such use would be at such party's own risk. The original |
|
|
; * developer of this software module and his/her company, and subsequent |
|
|
; * editors and their companies, will have no liability for use of this |
|
|
; * software or modifications or derivatives thereof. |
|
|
; * |
|
|
; * This program is free software; you can redistribute it and/or modify |
|
|
; * it 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 |
; * |
; * |
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., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA |
; * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA |
21 |
; * |
; * |
22 |
; ****************************************************************************/ |
; * $Id: sad_3dn.asm,v 1.11 2008-11-11 20:46:24 Isibaar Exp $ |
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 %1.endfunc-%1 |
32 |
|
%define %1 _%1:function %1.endfunc-%1 |
33 |
|
%define ENDFUNC .endfunc |
34 |
|
%else |
35 |
global _%1 |
global _%1 |
36 |
%define %1 _%1 |
%define %1 _%1 |
37 |
|
%define ENDFUNC |
38 |
|
%endif |
39 |
|
%else |
40 |
|
%ifdef MARK_FUNCS |
41 |
|
global %1:function %1.endfunc-%1 |
42 |
|
%define ENDFUNC .endfunc |
43 |
%else |
%else |
44 |
global %1 |
global %1 |
45 |
|
%define ENDFUNC |
46 |
|
%endif |
47 |
%endif |
%endif |
48 |
%endmacro |
%endmacro |
49 |
|
|
50 |
section .data |
;============================================================================= |
51 |
|
; Read only data |
52 |
align 16 |
;============================================================================= |
|
mmx_one times 4 dw 1 |
|
53 |
|
|
54 |
section .text |
%ifdef FORMAT_COFF |
55 |
|
SECTION .rodata |
56 |
cglobal sad16bi_3dn |
%else |
57 |
cglobal sad8bi_3dn |
SECTION .rodata align=16 |
58 |
|
%endif |
|
;=========================================================================== |
|
|
; |
|
|
; uint32_t sad16bi_3dn(const uint8_t * const cur, |
|
|
; const uint8_t * const ref1, |
|
|
; const uint8_t * const ref2, |
|
|
; const uint32_t stride); |
|
|
; |
|
|
;=========================================================================== |
|
59 |
|
|
60 |
|
ALIGN 16 |
61 |
|
mmx_one: |
62 |
|
times 4 dw 1 |
63 |
|
|
64 |
|
;============================================================================= |
65 |
|
; Helper macros |
66 |
|
;============================================================================= |
67 |
%macro SADBI_16x16_3DN 0 |
%macro SADBI_16x16_3DN 0 |
68 |
movq mm0, [eax] ; src |
movq mm0, [eax] ; src |
69 |
movq mm2, [eax+8] |
movq mm2, [eax+8] |
100 |
paddusw mm6,mm2 |
paddusw mm6,mm2 |
101 |
%endmacro |
%endmacro |
102 |
|
|
103 |
align 16 |
%macro SADBI_8x8_3DN 0 |
104 |
|
movq mm0, [eax] ; src |
105 |
|
movq mm2, [eax+ecx] |
106 |
|
|
107 |
|
movq mm1, [edx] ; ref1 |
108 |
|
movq mm3, [edx+ecx] |
109 |
|
pavgusb mm1, [ebx] ; ref2 |
110 |
|
lea edx, [edx+2*ecx] |
111 |
|
pavgusb mm3, [ebx+ecx] |
112 |
|
lea ebx, [ebx+2*ecx] |
113 |
|
|
114 |
|
movq mm4, mm0 |
115 |
|
lea eax, [eax+2*ecx] |
116 |
|
psubusb mm0, mm1 |
117 |
|
movq mm5, mm2 |
118 |
|
psubusb mm2, mm3 |
119 |
|
|
120 |
|
psubusb mm1, mm4 |
121 |
|
por mm0, mm1 |
122 |
|
psubusb mm3, mm5 |
123 |
|
por mm2, mm3 |
124 |
|
|
125 |
|
movq mm1, mm0 |
126 |
|
movq mm3, mm2 |
127 |
|
|
128 |
|
punpcklbw mm0,mm7 |
129 |
|
punpckhbw mm1,mm7 |
130 |
|
punpcklbw mm2,mm7 |
131 |
|
punpckhbw mm3,mm7 |
132 |
|
|
133 |
|
paddusw mm0,mm1 |
134 |
|
paddusw mm2,mm3 |
135 |
|
paddusw mm6,mm0 |
136 |
|
paddusw mm6,mm2 |
137 |
|
%endmacro |
138 |
|
|
139 |
|
;============================================================================= |
140 |
|
; Code |
141 |
|
;============================================================================= |
142 |
|
|
143 |
|
SECTION .text |
144 |
|
|
145 |
|
cglobal sad16bi_3dn |
146 |
|
cglobal sad8bi_3dn |
147 |
|
|
148 |
|
;----------------------------------------------------------------------------- |
149 |
|
; |
150 |
|
; uint32_t sad16bi_3dn(const uint8_t * const cur, |
151 |
|
; const uint8_t * const ref1, |
152 |
|
; const uint8_t * const ref2, |
153 |
|
; const uint32_t stride); |
154 |
|
; |
155 |
|
;----------------------------------------------------------------------------- |
156 |
|
|
157 |
|
ALIGN 16 |
158 |
sad16bi_3dn: |
sad16bi_3dn: |
159 |
push ebx |
push ebx |
160 |
mov eax, [esp+4+ 4] ; Src |
mov eax, [esp+4+ 4] ; Src |
164 |
|
|
165 |
pxor mm6, mm6 ; accum2 |
pxor mm6, mm6 ; accum2 |
166 |
pxor mm7, mm7 |
pxor mm7, mm7 |
167 |
.Loop |
.Loop: |
168 |
SADBI_16x16_3DN |
SADBI_16x16_3DN |
169 |
SADBI_16x16_3DN |
SADBI_16x16_3DN |
170 |
SADBI_16x16_3DN |
SADBI_16x16_3DN |
193 |
pop ebx |
pop ebx |
194 |
|
|
195 |
ret |
ret |
196 |
|
ENDFUNC |
197 |
|
|
198 |
|
;----------------------------------------------------------------------------- |
|
|
|
|
;=========================================================================== |
|
199 |
; |
; |
200 |
; uint32_t sad8bi_3dn(const uint8_t * const cur, |
; uint32_t sad8bi_3dn(const uint8_t * const cur, |
201 |
; const uint8_t * const ref1, |
; const uint8_t * const ref1, |
202 |
; const uint8_t * const ref2, |
; const uint8_t * const ref2, |
203 |
; const uint32_t stride); |
; const uint32_t stride); |
204 |
; |
; |
205 |
;=========================================================================== |
;----------------------------------------------------------------------------- |
|
|
|
|
%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 |
|
206 |
|
|
207 |
movq mm1,mm0 |
ALIGN 16 |
|
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 |
|
|
|
|
|
align 16 |
|
208 |
sad8bi_3dn: |
sad8bi_3dn: |
209 |
push ebx |
push ebx |
210 |
mov eax, [esp+4+ 4] ; Src |
mov eax, [esp+4+ 4] ; Src |
214 |
|
|
215 |
pxor mm6, mm6 ; accum2 |
pxor mm6, mm6 ; accum2 |
216 |
pxor mm7, mm7 |
pxor mm7, mm7 |
217 |
.Loop |
.Loop: |
218 |
SADBI_8x8_3DN |
SADBI_8x8_3DN |
219 |
SADBI_8x8_3DN |
SADBI_8x8_3DN |
220 |
SADBI_8x8_3DN |
SADBI_8x8_3DN |
230 |
pop ebx |
pop ebx |
231 |
|
|
232 |
ret |
ret |
233 |
|
ENDFUNC |
234 |
|
|
235 |
|
|
236 |
|
%ifidn __OUTPUT_FORMAT__,elf |
237 |
|
section ".note.GNU-stack" noalloc noexec nowrite progbits |
238 |
|
%endif |
239 |
|
|