Parent Directory | Revision Log
Revision 1586 - (view) (download)
1 : | edgomez | 1586 | ;/**************************************************************************** |
2 : | ; * | ||
3 : | ; * XVID MPEG-4 VIDEO CODEC | ||
4 : | ; * - Interlacing Field test - | ||
5 : | ; * | ||
6 : | ; * Copyright(C) 2002 Daniel Smith <danielsmith@astroboymail.com> | ||
7 : | ; * 2004 Andre Werthmann <wertmann@aei.mpg.de> | ||
8 : | ; * | ||
9 : | ; * This program is free software ; you can redistribute it and/or modify | ||
10 : | ; * it under the terms of the GNU General Public License as published by | ||
11 : | ; * the Free Software Foundation ; either version 2 of the License, or | ||
12 : | ; * (at your option) any later version. | ||
13 : | ; * | ||
14 : | ; * This program is distributed in the hope that it will be useful, | ||
15 : | ; * but WITHOUT ANY WARRANTY ; without even the implied warranty of | ||
16 : | ; * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the | ||
17 : | ; * GNU General Public License for more details. | ||
18 : | ; * | ||
19 : | ; * You should have received a copy of the GNU General Public License | ||
20 : | ; * along with this program ; if not, write to the Free Software | ||
21 : | ; * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA | ||
22 : | ; * | ||
23 : | ; * $Id: interlacing_mmx.asm,v 1.1 2005-01-05 23:02:15 edgomez Exp $ | ||
24 : | ; * | ||
25 : | ; ***************************************************************************/ | ||
26 : | |||
27 : | BITS 64 | ||
28 : | |||
29 : | %macro cglobal 1 | ||
30 : | %ifdef PREFIX | ||
31 : | %ifdef MARK_FUNCS | ||
32 : | global _%1:function %1.endfunc-%1 | ||
33 : | %define %1 _%1:function %1.endfunc-%1 | ||
34 : | %else | ||
35 : | global _%1 | ||
36 : | %define %1 _%1 | ||
37 : | %endif | ||
38 : | %else | ||
39 : | %ifdef MARK_FUNCS | ||
40 : | global %1:function %1.endfunc-%1 | ||
41 : | %else | ||
42 : | global %1 | ||
43 : | %endif | ||
44 : | %endif | ||
45 : | %endmacro | ||
46 : | |||
47 : | ;============================================================================= | ||
48 : | ; Read only data | ||
49 : | ;============================================================================= | ||
50 : | |||
51 : | %ifdef FORMAT_COFF | ||
52 : | SECTION .rodata | ||
53 : | %else | ||
54 : | SECTION .rodata align=16 | ||
55 : | %endif | ||
56 : | |||
57 : | ; advances to next block on right | ||
58 : | ALIGN 16 | ||
59 : | nexts: | ||
60 : | dd 0, 0, 8, 120, 8 | ||
61 : | |||
62 : | ; multiply word sums into dwords | ||
63 : | ALIGN 16 | ||
64 : | ones: | ||
65 : | times 4 dw 1 | ||
66 : | |||
67 : | ;============================================================================= | ||
68 : | ; Code | ||
69 : | ;============================================================================= | ||
70 : | |||
71 : | SECTION .text align=16 | ||
72 : | |||
73 : | cglobal MBFieldTest_x86_64 | ||
74 : | |||
75 : | ; neater | ||
76 : | %define line0 rsi | ||
77 : | %define line1 rsi+16 | ||
78 : | %define line2 rsi+32 | ||
79 : | %define line3 rsi+48 | ||
80 : | %define line4 rsi+64 | ||
81 : | %define line5 rsi+80 | ||
82 : | %define line6 rsi+96 | ||
83 : | %define line7 rsi+112 | ||
84 : | %define line8 rdi | ||
85 : | %define line9 rdi+16 | ||
86 : | %define line10 rdi+32 | ||
87 : | %define line11 rdi+48 | ||
88 : | %define line12 rdi+64 | ||
89 : | %define line13 rdi+80 | ||
90 : | %define line14 rdi+96 | ||
91 : | %define line15 rdi+112 | ||
92 : | |||
93 : | ; keep from losing track which reg holds which line - these never overlap | ||
94 : | %define m00 mm0 | ||
95 : | %define m01 mm1 | ||
96 : | %define m02 mm2 | ||
97 : | %define m03 mm0 | ||
98 : | %define m04 mm1 | ||
99 : | %define m05 mm2 | ||
100 : | %define m06 mm0 | ||
101 : | %define m07 mm1 | ||
102 : | %define m08 mm2 | ||
103 : | %define m09 mm0 | ||
104 : | %define m10 mm1 | ||
105 : | %define m11 mm2 | ||
106 : | %define m12 mm0 | ||
107 : | %define m13 mm1 | ||
108 : | %define m14 mm2 | ||
109 : | %define m15 mm0 | ||
110 : | |||
111 : | ; gets diff between three lines low(%2),mid(%3),hi(%4): frame = mid-low, field = hi-low | ||
112 : | %macro ABS8 4 | ||
113 : | movq %4, [%1] ; m02 = hi | ||
114 : | movq mm3, %2 ; mm3 = low copy | ||
115 : | |||
116 : | pxor mm4, mm4 ; mm4 = 0 | ||
117 : | pxor mm5, mm5 ; mm5 = 0 | ||
118 : | |||
119 : | psubw %2, %3 ; diff(med,low) for frame | ||
120 : | psubw mm3, %4 ; diff(hi,low) for field | ||
121 : | |||
122 : | pcmpgtw mm4, %2 ; if (diff<0), mm4 will be all 1's, else all 0's | ||
123 : | pcmpgtw mm5, mm3 | ||
124 : | pxor %2, mm4 ; this will get abs(), but off by 1 if (diff<0) | ||
125 : | pxor mm3, mm5 | ||
126 : | psubw %2, mm4 ; correct abs being off by 1 when (diff<0) | ||
127 : | psubw mm3, mm5 | ||
128 : | |||
129 : | paddw mm6, %2 ; add to totals | ||
130 : | paddw mm7, mm3 | ||
131 : | %endmacro | ||
132 : | |||
133 : | ;----------------------------------------------------------------------------- | ||
134 : | ; | ||
135 : | ; uint32_t MBFieldTest_mmx(int16_t * const data); | ||
136 : | ; | ||
137 : | ;----------------------------------------------------------------------------- | ||
138 : | |||
139 : | ALIGN 16 | ||
140 : | MBFieldTest_x86_64: | ||
141 : | mov rsi, rdi ; rsi = top left block | ||
142 : | ; rdi is already data[] | ||
143 : | add rdi, 256 ; rdi = bottom left block | ||
144 : | |||
145 : | pxor mm6, mm6 ; frame total | ||
146 : | pxor mm7, mm7 ; field total | ||
147 : | |||
148 : | mov rax, 4 ; we do left 8 bytes of data[0*64], then right 8 bytes | ||
149 : | ; then left 8 bytes of data[1*64], then last 8 bytes | ||
150 : | .loop: | ||
151 : | movq m00, [line0] ; line0 | ||
152 : | movq m01, [line1] ; line1 | ||
153 : | |||
154 : | ABS8 line2, m00, m01, m02 ; frame += (line2-line1), field += (line2-line0) | ||
155 : | ABS8 line3, m01, m02, m03 | ||
156 : | ABS8 line4, m02, m03, m04 | ||
157 : | ABS8 line5, m03, m04, m05 | ||
158 : | ABS8 line6, m04, m05, m06 | ||
159 : | ABS8 line7, m05, m06, m07 | ||
160 : | ABS8 line8, m06, m07, m08 | ||
161 : | |||
162 : | movq m09, [line9] ; line9-line7, no frame comp for line9-line8! | ||
163 : | pxor mm4, mm4 | ||
164 : | psubw m07, m09 | ||
165 : | pcmpgtw mm4, mm1 | ||
166 : | pxor m07, mm4 | ||
167 : | psubw m07, mm4 | ||
168 : | paddw mm7, m07 ; add to field total | ||
169 : | |||
170 : | ABS8 line10, m08, m09, m10 ; frame += (line10-line9), field += (line10-line8) | ||
171 : | ABS8 line11, m09, m10, m11 | ||
172 : | ABS8 line12, m10, m11, m12 | ||
173 : | ABS8 line13, m11, m12, m13 | ||
174 : | ABS8 line14, m12, m13, m14 | ||
175 : | ABS8 line15, m13, m14, m15 | ||
176 : | |||
177 : | pxor mm4, mm4 ; line15-line14, we're done with field comps! | ||
178 : | psubw m14, m15 | ||
179 : | pcmpgtw mm4, m14 | ||
180 : | pxor m14, mm4 | ||
181 : | psubw m14, mm4 | ||
182 : | paddw mm6, m14 ; add to frame total | ||
183 : | |||
184 : | lea r9, [nexts wrt rip] ; pic code | ||
185 : | mov rcx, [r9+rax*4] ; move rsi/rdi 8 pixels to the right | ||
186 : | add rsi, rcx | ||
187 : | add rdi, rcx | ||
188 : | |||
189 : | dec rax | ||
190 : | jnz near .loop | ||
191 : | |||
192 : | .decide: | ||
193 : | movq mm0, [ones wrt rip] ; add packed words into single dwords | ||
194 : | pmaddwd mm6, mm0 | ||
195 : | pmaddwd mm7, mm0 | ||
196 : | |||
197 : | movq mm0, mm6 ; ecx will be frame total, edx field | ||
198 : | movq mm1, mm7 | ||
199 : | psrlq mm0, 32 | ||
200 : | psrlq mm1, 32 | ||
201 : | paddd mm0, mm6 | ||
202 : | paddd mm1, mm7 | ||
203 : | movd rcx, mm0 | ||
204 : | movd rdx, mm1 | ||
205 : | |||
206 : | add rdx, 350 ; add bias against field decision | ||
207 : | cmp rcx, rdx | ||
208 : | jb .end ; if frame<field, don't use field dct | ||
209 : | inc rax ; if frame>=field, use field dct (return 1) | ||
210 : | |||
211 : | .end: | ||
212 : | ret | ||
213 : | .endfunc | ||
214 : |
No admin address has been configured | ViewVC Help |
Powered by ViewVC 1.0.4 |