Parent Directory | Revision Log
Revision 1.14 - (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 : | Isibaar | 1.14 | ; * $Id: sad_3dn.asm,v 1.13 2008/12/04 14:41:50 Isibaar Exp $ |
23 : | edgomez | 1.5 | ; * |
24 : | edgomez | 1.6 | ; ***************************************************************************/ |
25 : | suxen_drol | 1.1 | |
26 : | Isibaar | 1.12 | %include "nasm.inc" |
27 : | suxen_drol | 1.1 | |
28 : | edgomez | 1.6 | ;============================================================================= |
29 : | ; Read only data | ||
30 : | ;============================================================================= | ||
31 : | |||
32 : | Isibaar | 1.12 | DATA |
33 : | edgomez | 1.6 | |
34 : | Isibaar | 1.12 | ALIGN SECTION_ALIGN |
35 : | edgomez | 1.6 | mmx_one: |
36 : | times 4 dw 1 | ||
37 : | |||
38 : | ;============================================================================= | ||
39 : | ; Helper macros | ||
40 : | ;============================================================================= | ||
41 : | %macro SADBI_16x16_3DN 0 | ||
42 : | Isibaar | 1.12 | movq mm0, [_EAX] ; src |
43 : | movq mm2, [_EAX+8] | ||
44 : | edgomez | 1.6 | |
45 : | Isibaar | 1.12 | movq mm1, [TMP1] ; ref1 |
46 : | movq mm3, [TMP1+8] | ||
47 : | pavgusb mm1, [_EBX] ; ref2 | ||
48 : | lea TMP1, [TMP1+TMP0] | ||
49 : | pavgusb mm3, [_EBX+8] | ||
50 : | lea _EBX, [_EBX+TMP0] | ||
51 : | edgomez | 1.6 | |
52 : | movq mm4, mm0 | ||
53 : | Isibaar | 1.12 | lea _EAX, [_EAX+TMP0] |
54 : | edgomez | 1.6 | psubusb mm0, mm1 |
55 : | movq mm5, mm2 | ||
56 : | psubusb mm2, mm3 | ||
57 : | |||
58 : | psubusb mm1, mm4 | ||
59 : | por mm0, mm1 | ||
60 : | psubusb mm3, mm5 | ||
61 : | por mm2, mm3 | ||
62 : | |||
63 : | movq mm1, mm0 | ||
64 : | movq mm3, mm2 | ||
65 : | |||
66 : | punpcklbw mm0,mm7 | ||
67 : | punpckhbw mm1,mm7 | ||
68 : | punpcklbw mm2,mm7 | ||
69 : | punpckhbw mm3,mm7 | ||
70 : | |||
71 : | paddusw mm0,mm1 | ||
72 : | paddusw mm2,mm3 | ||
73 : | paddusw mm6,mm0 | ||
74 : | paddusw mm6,mm2 | ||
75 : | %endmacro | ||
76 : | |||
77 : | %macro SADBI_8x8_3DN 0 | ||
78 : | Isibaar | 1.12 | movq mm0, [_EAX] ; src |
79 : | movq mm2, [_EAX+TMP0] | ||
80 : | edgomez | 1.6 | |
81 : | Isibaar | 1.12 | movq mm1, [TMP1] ; ref1 |
82 : | movq mm3, [TMP1+TMP0] | ||
83 : | pavgusb mm1, [_EBX] ; ref2 | ||
84 : | lea TMP1, [TMP1+2*TMP0] | ||
85 : | pavgusb mm3, [_EBX+TMP0] | ||
86 : | lea _EBX, [_EBX+2*TMP0] | ||
87 : | edgomez | 1.6 | |
88 : | movq mm4, mm0 | ||
89 : | Isibaar | 1.12 | lea _EAX, [_EAX+2*TMP0] |
90 : | edgomez | 1.6 | psubusb mm0, mm1 |
91 : | movq mm5, mm2 | ||
92 : | psubusb mm2, mm3 | ||
93 : | |||
94 : | psubusb mm1, mm4 | ||
95 : | por mm0, mm1 | ||
96 : | psubusb mm3, mm5 | ||
97 : | por mm2, mm3 | ||
98 : | |||
99 : | movq mm1, mm0 | ||
100 : | movq mm3, mm2 | ||
101 : | |||
102 : | punpcklbw mm0,mm7 | ||
103 : | punpckhbw mm1,mm7 | ||
104 : | punpcklbw mm2,mm7 | ||
105 : | punpckhbw mm3,mm7 | ||
106 : | |||
107 : | paddusw mm0,mm1 | ||
108 : | paddusw mm2,mm3 | ||
109 : | paddusw mm6,mm0 | ||
110 : | paddusw mm6,mm2 | ||
111 : | %endmacro | ||
112 : | suxen_drol | 1.1 | |
113 : | edgomez | 1.6 | ;============================================================================= |
114 : | ; Code | ||
115 : | ;============================================================================= | ||
116 : | suxen_drol | 1.1 | |
117 : | Isibaar | 1.13 | TEXT |
118 : | suxen_drol | 1.1 | |
119 : | cglobal sad16bi_3dn | ||
120 : | cglobal sad8bi_3dn | ||
121 : | |||
122 : | edgomez | 1.6 | ;----------------------------------------------------------------------------- |
123 : | ; | ||
124 : | ; uint32_t sad16bi_3dn(const uint8_t * const cur, | ||
125 : | ; const uint8_t * const ref1, | ||
126 : | ; const uint8_t * const ref2, | ||
127 : | ; const uint32_t stride); | ||
128 : | ; | ||
129 : | ;----------------------------------------------------------------------------- | ||
130 : | |||
131 : | Isibaar | 1.12 | ALIGN SECTION_ALIGN |
132 : | edgomez | 1.6 | sad16bi_3dn: |
133 : | Isibaar | 1.12 | mov _EAX, prm1 ; Src |
134 : | mov TMP1, prm2 ; Ref1 | ||
135 : | mov TMP0, prm4 ; Stride | ||
136 : | |||
137 : | push _EBX | ||
138 : | %ifdef ARCH_IS_X86_64 | ||
139 : | mov _EBX, prm3 | ||
140 : | %else | ||
141 : | mov _EBX, [_ESP+4+12] ; Ref2 | ||
142 : | %endif | ||
143 : | edgomez | 1.6 | |
144 : | pxor mm6, mm6 ; accum2 | ||
145 : | pxor mm7, mm7 | ||
146 : | Isibaar | 1.11 | .Loop: |
147 : | edgomez | 1.6 | SADBI_16x16_3DN |
148 : | SADBI_16x16_3DN | ||
149 : | SADBI_16x16_3DN | ||
150 : | SADBI_16x16_3DN | ||
151 : | SADBI_16x16_3DN | ||
152 : | SADBI_16x16_3DN | ||
153 : | SADBI_16x16_3DN | ||
154 : | SADBI_16x16_3DN | ||
155 : | |||
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 : | SADBI_16x16_3DN | ||
164 : | |||
165 : | pmaddwd mm6, [mmx_one] ; collapse | ||
166 : | movq mm7, mm6 | ||
167 : | psrlq mm7, 32 | ||
168 : | paddd mm6, mm7 | ||
169 : | |||
170 : | movd eax, mm6 | ||
171 : | |||
172 : | Isibaar | 1.12 | pop _EBX |
173 : | edgomez | 1.6 | |
174 : | ret | ||
175 : | Isibaar | 1.11 | ENDFUNC |
176 : | edgomez | 1.6 | |
177 : | ;----------------------------------------------------------------------------- | ||
178 : | ; | ||
179 : | ; uint32_t sad8bi_3dn(const uint8_t * const cur, | ||
180 : | ; const uint8_t * const ref1, | ||
181 : | ; const uint8_t * const ref2, | ||
182 : | ; const uint32_t stride); | ||
183 : | ; | ||
184 : | ;----------------------------------------------------------------------------- | ||
185 : | |||
186 : | Isibaar | 1.12 | ALIGN SECTION_ALIGN |
187 : | edgomez | 1.6 | sad8bi_3dn: |
188 : | Isibaar | 1.12 | mov _EAX, prm1 ; Src |
189 : | mov TMP1, prm2 ; Ref1 | ||
190 : | mov TMP0, prm4 ; Stride | ||
191 : | |||
192 : | push _EBX | ||
193 : | %ifdef ARCH_IS_X86_64 | ||
194 : | mov _EBX, prm3 | ||
195 : | %else | ||
196 : | mov _EBX, [_ESP+4+12] ; Ref2 | ||
197 : | %endif | ||
198 : | edgomez | 1.6 | |
199 : | pxor mm6, mm6 ; accum2 | ||
200 : | pxor mm7, mm7 | ||
201 : | Isibaar | 1.11 | .Loop: |
202 : | edgomez | 1.6 | SADBI_8x8_3DN |
203 : | SADBI_8x8_3DN | ||
204 : | SADBI_8x8_3DN | ||
205 : | SADBI_8x8_3DN | ||
206 : | |||
207 : | pmaddwd mm6, [mmx_one] ; collapse | ||
208 : | movq mm7, mm6 | ||
209 : | psrlq mm7, 32 | ||
210 : | paddd mm6, mm7 | ||
211 : | suxen_drol | 1.1 | |
212 : | edgomez | 1.6 | movd eax, mm6 |
213 : | suxen_drol | 1.1 | |
214 : | Isibaar | 1.12 | pop _EBX |
215 : | suxen_drol | 1.1 | |
216 : | edgomez | 1.6 | ret |
217 : | Isibaar | 1.11 | ENDFUNC |
218 : | edgomez | 1.9 | |
219 : | Isibaar | 1.14 | NON_EXEC_STACK |
No admin address has been configured | ViewVC Help |
Powered by ViewVC 1.0.4 |