Parent Directory | Revision Log
Revision 1.7 - (view) (download)
1 : | edgomez | 1.6 | ;/**************************************************************************** |
2 : | suxen_drol | 1.1 | ; * |
3 : | edgomez | 1.6 | ; * XVID MPEG-4 VIDEO CODEC |
4 : | ; * - 3DNow sad operators w/o XMM instructions - | ||
5 : | suxen_drol | 1.1 | ; * |
6 : | edgomez | 1.6 | ; * Copyright(C) 2002 Peter ross <pross@xvid.org> |
7 : | suxen_drol | 1.1 | ; * |
8 : | edgomez | 1.6 | ; * 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 : | chl | 1.2 | ; * |
13 : | edgomez | 1.6 | ; * 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 : | edgomez | 1.4 | ; * |
18 : | edgomez | 1.6 | ; * 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 : | edgomez | 1.5 | ; * |
22 : | edgomez | 1.6 | ; * $Id$ |
23 : | edgomez | 1.5 | ; * |
24 : | edgomez | 1.6 | ; ***************************************************************************/ |
25 : | suxen_drol | 1.1 | |
26 : | edgomez | 1.6 | BITS 32 |
27 : | suxen_drol | 1.1 | |
28 : | edgomez | 1.6 | %macro cglobal 1 |
29 : | suxen_drol | 1.1 | %ifdef PREFIX |
30 : | edgomez | 1.6 | global _%1 |
31 : | suxen_drol | 1.1 | %define %1 _%1 |
32 : | %else | ||
33 : | global %1 | ||
34 : | %endif | ||
35 : | %endmacro | ||
36 : | |||
37 : | edgomez | 1.6 | ;============================================================================= |
38 : | ; Read only data | ||
39 : | ;============================================================================= | ||
40 : | |||
41 : | %ifdef FORMAT_COFF | ||
42 : | edgomez | 1.7 | SECTION .rodata |
43 : | edgomez | 1.6 | %else |
44 : | edgomez | 1.7 | SECTION .rodata align=16 |
45 : | edgomez | 1.6 | %endif |
46 : | |||
47 : | ALIGN 16 | ||
48 : | mmx_one: | ||
49 : | times 4 dw 1 | ||
50 : | |||
51 : | ;============================================================================= | ||
52 : | ; Helper macros | ||
53 : | ;============================================================================= | ||
54 : | %macro SADBI_16x16_3DN 0 | ||
55 : | movq mm0, [eax] ; src | ||
56 : | movq mm2, [eax+8] | ||
57 : | |||
58 : | movq mm1, [edx] ; ref1 | ||
59 : | movq mm3, [edx+8] | ||
60 : | pavgusb mm1, [ebx] ; ref2 | ||
61 : | lea edx, [edx+ecx] | ||
62 : | pavgusb mm3, [ebx+8] | ||
63 : | lea ebx, [ebx+ecx] | ||
64 : | |||
65 : | movq mm4, mm0 | ||
66 : | lea eax, [eax+ecx] | ||
67 : | psubusb mm0, mm1 | ||
68 : | movq mm5, mm2 | ||
69 : | psubusb mm2, mm3 | ||
70 : | |||
71 : | psubusb mm1, mm4 | ||
72 : | por mm0, mm1 | ||
73 : | psubusb mm3, mm5 | ||
74 : | por mm2, mm3 | ||
75 : | |||
76 : | movq mm1, mm0 | ||
77 : | movq mm3, mm2 | ||
78 : | |||
79 : | punpcklbw mm0,mm7 | ||
80 : | punpckhbw mm1,mm7 | ||
81 : | punpcklbw mm2,mm7 | ||
82 : | punpckhbw mm3,mm7 | ||
83 : | |||
84 : | paddusw mm0,mm1 | ||
85 : | paddusw mm2,mm3 | ||
86 : | paddusw mm6,mm0 | ||
87 : | paddusw mm6,mm2 | ||
88 : | %endmacro | ||
89 : | |||
90 : | %macro SADBI_8x8_3DN 0 | ||
91 : | movq mm0, [eax] ; src | ||
92 : | movq mm2, [eax+ecx] | ||
93 : | |||
94 : | movq mm1, [edx] ; ref1 | ||
95 : | movq mm3, [edx+ecx] | ||
96 : | pavgusb mm1, [ebx] ; ref2 | ||
97 : | lea edx, [edx+2*ecx] | ||
98 : | pavgusb mm3, [ebx+ecx] | ||
99 : | lea ebx, [ebx+2*ecx] | ||
100 : | |||
101 : | movq mm4, mm0 | ||
102 : | lea eax, [eax+2*ecx] | ||
103 : | psubusb mm0, mm1 | ||
104 : | movq mm5, mm2 | ||
105 : | psubusb mm2, mm3 | ||
106 : | |||
107 : | psubusb mm1, mm4 | ||
108 : | por mm0, mm1 | ||
109 : | psubusb mm3, mm5 | ||
110 : | por mm2, mm3 | ||
111 : | |||
112 : | movq mm1, mm0 | ||
113 : | movq mm3, mm2 | ||
114 : | |||
115 : | punpcklbw mm0,mm7 | ||
116 : | punpckhbw mm1,mm7 | ||
117 : | punpcklbw mm2,mm7 | ||
118 : | punpckhbw mm3,mm7 | ||
119 : | |||
120 : | paddusw mm0,mm1 | ||
121 : | paddusw mm2,mm3 | ||
122 : | paddusw mm6,mm0 | ||
123 : | paddusw mm6,mm2 | ||
124 : | %endmacro | ||
125 : | suxen_drol | 1.1 | |
126 : | edgomez | 1.6 | ;============================================================================= |
127 : | ; Code | ||
128 : | ;============================================================================= | ||
129 : | suxen_drol | 1.1 | |
130 : | edgomez | 1.6 | SECTION .text |
131 : | suxen_drol | 1.1 | |
132 : | cglobal sad16bi_3dn | ||
133 : | cglobal sad8bi_3dn | ||
134 : | |||
135 : | edgomez | 1.6 | ;----------------------------------------------------------------------------- |
136 : | ; | ||
137 : | ; uint32_t sad16bi_3dn(const uint8_t * const cur, | ||
138 : | ; const uint8_t * const ref1, | ||
139 : | ; const uint8_t * const ref2, | ||
140 : | ; const uint32_t stride); | ||
141 : | ; | ||
142 : | ;----------------------------------------------------------------------------- | ||
143 : | |||
144 : | ALIGN 16 | ||
145 : | sad16bi_3dn: | ||
146 : | push ebx | ||
147 : | mov eax, [esp+4+ 4] ; Src | ||
148 : | mov edx, [esp+4+ 8] ; Ref1 | ||
149 : | mov ebx, [esp+4+12] ; Ref2 | ||
150 : | mov ecx, [esp+4+16] ; Stride | ||
151 : | |||
152 : | pxor mm6, mm6 ; accum2 | ||
153 : | pxor mm7, mm7 | ||
154 : | .Loop | ||
155 : | SADBI_16x16_3DN | ||
156 : | SADBI_16x16_3DN | ||
157 : | SADBI_16x16_3DN | ||
158 : | SADBI_16x16_3DN | ||
159 : | SADBI_16x16_3DN | ||
160 : | SADBI_16x16_3DN | ||
161 : | SADBI_16x16_3DN | ||
162 : | SADBI_16x16_3DN | ||
163 : | |||
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 : | |||
173 : | pmaddwd mm6, [mmx_one] ; collapse | ||
174 : | movq mm7, mm6 | ||
175 : | psrlq mm7, 32 | ||
176 : | paddd mm6, mm7 | ||
177 : | |||
178 : | movd eax, mm6 | ||
179 : | |||
180 : | pop ebx | ||
181 : | |||
182 : | ret | ||
183 : | |||
184 : | ;----------------------------------------------------------------------------- | ||
185 : | ; | ||
186 : | ; uint32_t sad8bi_3dn(const uint8_t * const cur, | ||
187 : | ; const uint8_t * const ref1, | ||
188 : | ; const uint8_t * const ref2, | ||
189 : | ; const uint32_t stride); | ||
190 : | ; | ||
191 : | ;----------------------------------------------------------------------------- | ||
192 : | |||
193 : | ALIGN 16 | ||
194 : | sad8bi_3dn: | ||
195 : | push ebx | ||
196 : | mov eax, [esp+4+ 4] ; Src | ||
197 : | mov edx, [esp+4+ 8] ; Ref1 | ||
198 : | mov ebx, [esp+4+12] ; Ref2 | ||
199 : | mov ecx, [esp+4+16] ; Stride | ||
200 : | |||
201 : | pxor mm6, mm6 ; accum2 | ||
202 : | pxor mm7, mm7 | ||
203 : | .Loop | ||
204 : | SADBI_8x8_3DN | ||
205 : | SADBI_8x8_3DN | ||
206 : | SADBI_8x8_3DN | ||
207 : | SADBI_8x8_3DN | ||
208 : | |||
209 : | pmaddwd mm6, [mmx_one] ; collapse | ||
210 : | movq mm7, mm6 | ||
211 : | psrlq mm7, 32 | ||
212 : | paddd mm6, mm7 | ||
213 : | suxen_drol | 1.1 | |
214 : | edgomez | 1.6 | movd eax, mm6 |
215 : | suxen_drol | 1.1 | |
216 : | edgomez | 1.6 | pop ebx |
217 : | suxen_drol | 1.1 | |
218 : | edgomez | 1.6 | ret |
No admin address has been configured | ViewVC Help |
Powered by ViewVC 1.0.4 |