Parent Directory | Revision Log
Revision 1790 - (view) (download)
1 : | edgomez | 1382 | ;/**************************************************************************** |
2 : | suxen_drol | 329 | ; * |
3 : | edgomez | 1382 | ; * XVID MPEG-4 VIDEO CODEC |
4 : | ; * - 3DNow sad operators w/o XMM instructions - | ||
5 : | suxen_drol | 329 | ; * |
6 : | edgomez | 1382 | ; * Copyright(C) 2002 Peter ross <pross@xvid.org> |
7 : | suxen_drol | 329 | ; * |
8 : | edgomez | 1382 | ; * 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 | ||
11 : | ; * (at your option) any later version. | ||
12 : | suxen_drol | 329 | ; * |
13 : | edgomez | 1382 | ; * This program is distributed in the hope that it will be useful, |
14 : | ; * but WITHOUT ANY WARRANTY; without even the implied warranty of | ||
15 : | ; * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the | ||
16 : | ; * GNU General Public License for more details. | ||
17 : | suxen_drol | 329 | ; * |
18 : | edgomez | 1382 | ; * You should have received a copy of the GNU General Public License |
19 : | ; * along with this program; if not, write to the Free Software | ||
20 : | ; * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA | ||
21 : | suxen_drol | 329 | ; * |
22 : | Isibaar | 1790 | ; * $Id: sad_3dn.asm,v 1.10 2008-08-19 09:06:48 Isibaar Exp $ |
23 : | suxen_drol | 329 | ; * |
24 : | edgomez | 1382 | ; ***************************************************************************/ |
25 : | suxen_drol | 329 | |
26 : | edgomez | 1382 | BITS 32 |
27 : | suxen_drol | 329 | |
28 : | edgomez | 1382 | %macro cglobal 1 |
29 : | suxen_drol | 329 | %ifdef PREFIX |
30 : | edgomez | 1535 | %ifdef MARK_FUNCS |
31 : | edgomez | 1540 | global _%1:function %1.endfunc-%1 |
32 : | %define %1 _%1:function %1.endfunc-%1 | ||
33 : | edgomez | 1535 | %else |
34 : | global _%1 | ||
35 : | %define %1 _%1 | ||
36 : | %endif | ||
37 : | suxen_drol | 329 | %else |
38 : | edgomez | 1535 | %ifdef MARK_FUNCS |
39 : | edgomez | 1540 | global %1:function %1.endfunc-%1 |
40 : | edgomez | 1535 | %else |
41 : | global %1 | ||
42 : | %endif | ||
43 : | suxen_drol | 329 | %endif |
44 : | %endmacro | ||
45 : | |||
46 : | edgomez | 1382 | ;============================================================================= |
47 : | ; Read only data | ||
48 : | ;============================================================================= | ||
49 : | suxen_drol | 329 | |
50 : | edgomez | 1382 | %ifdef FORMAT_COFF |
51 : | edgomez | 1519 | SECTION .rodata |
52 : | edgomez | 1382 | %else |
53 : | edgomez | 1519 | SECTION .rodata align=16 |
54 : | edgomez | 1382 | %endif |
55 : | suxen_drol | 329 | |
56 : | edgomez | 1382 | ALIGN 16 |
57 : | mmx_one: | ||
58 : | times 4 dw 1 | ||
59 : | suxen_drol | 329 | |
60 : | edgomez | 1382 | ;============================================================================= |
61 : | ; Helper macros | ||
62 : | ;============================================================================= | ||
63 : | %macro SADBI_16x16_3DN 0 | ||
64 : | movq mm0, [eax] ; src | ||
65 : | movq mm2, [eax+8] | ||
66 : | suxen_drol | 329 | |
67 : | edgomez | 1382 | movq mm1, [edx] ; ref1 |
68 : | movq mm3, [edx+8] | ||
69 : | pavgusb mm1, [ebx] ; ref2 | ||
70 : | lea edx, [edx+ecx] | ||
71 : | pavgusb mm3, [ebx+8] | ||
72 : | lea ebx, [ebx+ecx] | ||
73 : | suxen_drol | 329 | |
74 : | edgomez | 1382 | movq mm4, mm0 |
75 : | lea eax, [eax+ecx] | ||
76 : | psubusb mm0, mm1 | ||
77 : | movq mm5, mm2 | ||
78 : | psubusb mm2, mm3 | ||
79 : | suxen_drol | 329 | |
80 : | edgomez | 1382 | psubusb mm1, mm4 |
81 : | por mm0, mm1 | ||
82 : | psubusb mm3, mm5 | ||
83 : | por mm2, mm3 | ||
84 : | suxen_drol | 329 | |
85 : | edgomez | 1382 | movq mm1, mm0 |
86 : | movq mm3, mm2 | ||
87 : | suxen_drol | 329 | |
88 : | edgomez | 1382 | punpcklbw mm0,mm7 |
89 : | punpckhbw mm1,mm7 | ||
90 : | punpcklbw mm2,mm7 | ||
91 : | punpckhbw mm3,mm7 | ||
92 : | suxen_drol | 329 | |
93 : | edgomez | 1382 | paddusw mm0,mm1 |
94 : | paddusw mm2,mm3 | ||
95 : | paddusw mm6,mm0 | ||
96 : | paddusw mm6,mm2 | ||
97 : | %endmacro | ||
98 : | suxen_drol | 329 | |
99 : | edgomez | 1382 | %macro SADBI_8x8_3DN 0 |
100 : | movq mm0, [eax] ; src | ||
101 : | movq mm2, [eax+ecx] | ||
102 : | suxen_drol | 329 | |
103 : | edgomez | 1382 | 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 : | suxen_drol | 329 | |
110 : | edgomez | 1382 | movq mm4, mm0 |
111 : | lea eax, [eax+2*ecx] | ||
112 : | psubusb mm0, mm1 | ||
113 : | movq mm5, mm2 | ||
114 : | psubusb mm2, mm3 | ||
115 : | suxen_drol | 329 | |
116 : | edgomez | 1382 | psubusb mm1, mm4 |
117 : | por mm0, mm1 | ||
118 : | psubusb mm3, mm5 | ||
119 : | por mm2, mm3 | ||
120 : | suxen_drol | 329 | |
121 : | edgomez | 1382 | movq mm1, mm0 |
122 : | movq mm3, mm2 | ||
123 : | suxen_drol | 329 | |
124 : | edgomez | 1382 | punpcklbw mm0,mm7 |
125 : | punpckhbw mm1,mm7 | ||
126 : | punpcklbw mm2,mm7 | ||
127 : | punpckhbw mm3,mm7 | ||
128 : | suxen_drol | 329 | |
129 : | edgomez | 1382 | paddusw mm0,mm1 |
130 : | paddusw mm2,mm3 | ||
131 : | paddusw mm6,mm0 | ||
132 : | paddusw mm6,mm2 | ||
133 : | %endmacro | ||
134 : | suxen_drol | 329 | |
135 : | edgomez | 1382 | ;============================================================================= |
136 : | ; Code | ||
137 : | ;============================================================================= | ||
138 : | suxen_drol | 329 | |
139 : | edgomez | 1382 | SECTION .text |
140 : | suxen_drol | 329 | |
141 : | edgomez | 1382 | cglobal sad16bi_3dn |
142 : | cglobal sad8bi_3dn | ||
143 : | suxen_drol | 329 | |
144 : | edgomez | 1382 | ;----------------------------------------------------------------------------- |
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 : | suxen_drol | 329 | |
153 : | edgomez | 1382 | ALIGN 16 |
154 : | sad16bi_3dn: | ||
155 : | push ebx | ||
156 : | mov eax, [esp+4+ 4] ; Src | ||
157 : | mov edx, [esp+4+ 8] ; Ref1 | ||
158 : | mov ebx, [esp+4+12] ; Ref2 | ||
159 : | mov ecx, [esp+4+16] ; Stride | ||
160 : | suxen_drol | 329 | |
161 : | edgomez | 1382 | pxor mm6, mm6 ; accum2 |
162 : | pxor mm7, mm7 | ||
163 : | .Loop | ||
164 : | SADBI_16x16_3DN | ||
165 : | SADBI_16x16_3DN | ||
166 : | SADBI_16x16_3DN | ||
167 : | SADBI_16x16_3DN | ||
168 : | SADBI_16x16_3DN | ||
169 : | SADBI_16x16_3DN | ||
170 : | SADBI_16x16_3DN | ||
171 : | SADBI_16x16_3DN | ||
172 : | suxen_drol | 329 | |
173 : | edgomez | 1382 | SADBI_16x16_3DN |
174 : | SADBI_16x16_3DN | ||
175 : | SADBI_16x16_3DN | ||
176 : | SADBI_16x16_3DN | ||
177 : | SADBI_16x16_3DN | ||
178 : | SADBI_16x16_3DN | ||
179 : | SADBI_16x16_3DN | ||
180 : | SADBI_16x16_3DN | ||
181 : | suxen_drol | 329 | |
182 : | edgomez | 1382 | pmaddwd mm6, [mmx_one] ; collapse |
183 : | movq mm7, mm6 | ||
184 : | psrlq mm7, 32 | ||
185 : | paddd mm6, mm7 | ||
186 : | suxen_drol | 329 | |
187 : | edgomez | 1382 | movd eax, mm6 |
188 : | suxen_drol | 329 | |
189 : | edgomez | 1382 | pop ebx |
190 : | suxen_drol | 329 | |
191 : | edgomez | 1382 | ret |
192 : | edgomez | 1540 | .endfunc |
193 : | suxen_drol | 329 | |
194 : | edgomez | 1382 | ;----------------------------------------------------------------------------- |
195 : | ; | ||
196 : | ; uint32_t sad8bi_3dn(const uint8_t * const cur, | ||
197 : | ; const uint8_t * const ref1, | ||
198 : | ; const uint8_t * const ref2, | ||
199 : | ; const uint32_t stride); | ||
200 : | ; | ||
201 : | ;----------------------------------------------------------------------------- | ||
202 : | suxen_drol | 329 | |
203 : | edgomez | 1382 | ALIGN 16 |
204 : | sad8bi_3dn: | ||
205 : | push ebx | ||
206 : | mov eax, [esp+4+ 4] ; Src | ||
207 : | mov edx, [esp+4+ 8] ; Ref1 | ||
208 : | mov ebx, [esp+4+12] ; Ref2 | ||
209 : | mov ecx, [esp+4+16] ; Stride | ||
210 : | suxen_drol | 329 | |
211 : | edgomez | 1382 | pxor mm6, mm6 ; accum2 |
212 : | pxor mm7, mm7 | ||
213 : | .Loop | ||
214 : | SADBI_8x8_3DN | ||
215 : | SADBI_8x8_3DN | ||
216 : | SADBI_8x8_3DN | ||
217 : | SADBI_8x8_3DN | ||
218 : | suxen_drol | 329 | |
219 : | edgomez | 1382 | pmaddwd mm6, [mmx_one] ; collapse |
220 : | movq mm7, mm6 | ||
221 : | psrlq mm7, 32 | ||
222 : | paddd mm6, mm7 | ||
223 : | suxen_drol | 329 | |
224 : | edgomez | 1382 | movd eax, mm6 |
225 : | suxen_drol | 329 | |
226 : | edgomez | 1382 | pop ebx |
227 : | suxen_drol | 329 | |
228 : | edgomez | 1382 | ret |
229 : | edgomez | 1540 | .endfunc |
230 : | |||
231 : | Isibaar | 1790 | |
232 : | %ifidn __OUTPUT_FORMAT__,elf | ||
233 : | section ".note.GNU-stack" noalloc noexec nowrite progbits | ||
234 : | %endif | ||
235 : |
No admin address has been configured | ViewVC Help |
Powered by ViewVC 1.0.4 |