Parent Directory
|
Revision Log
Revision 1.10.2.1 - (view) (download)
1 : | edgomez | 1.3 | ;/**************************************************************************** |
2 : | edgomez | 1.2 | ; * |
3 : | edgomez | 1.3 | ; * XVID MPEG-4 VIDEO CODEC |
4 : | ; * - Interlacing Field test - | ||
5 : | edgomez | 1.2 | ; * |
6 : | edgomez | 1.3 | ; * Copyright(C) 2002 Daniel Smith <danielsmith@astroboymail.com> |
7 : | edgomez | 1.2 | ; * |
8 : | Isibaar | 1.9 | ; * This program is free software ; you can r_EDIstribute it and/or modify |
9 : | edgomez | 1.3 | ; * 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 | ||
11 : | ; * (at your option) any later version. | ||
12 : | edgomez | 1.2 | ; * |
13 : | edgomez | 1.3 | ; * 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 | 1.2 | ; * |
18 : | edgomez | 1.3 | ; * 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 | 1.2 | ; * |
22 : | Isibaar | 1.10.2.1 | ; * $Id: interlacing_mmx.asm,v 1.10 2008/11/27 18:35:36 Isibaar Exp $ |
23 : | edgomez | 1.2 | ; * |
24 : | edgomez | 1.3 | ; ***************************************************************************/ |
25 : | edgomez | 1.2 | |
26 : | Isibaar | 1.9 | %include "nasm.inc" |
27 : | edgomez | 1.2 | |
28 : | edgomez | 1.3 | ;============================================================================= |
29 : | ; Read only data | ||
30 : | ;============================================================================= | ||
31 : | |||
32 : | Isibaar | 1.9 | DATA |
33 : | edgomez | 1.3 | |
34 : | ; advances to next block on right | ||
35 : | Isibaar | 1.9 | ALIGN SECTION_ALIGN |
36 : | edgomez | 1.3 | nexts: |
37 : | dd 0, 0, 8, 120, 8 | ||
38 : | |||
39 : | ; multiply word sums into dwords | ||
40 : | Isibaar | 1.9 | ALIGN SECTION_ALIGN |
41 : | edgomez | 1.3 | ones: |
42 : | times 4 dw 1 | ||
43 : | |||
44 : | ;============================================================================= | ||
45 : | ; Code | ||
46 : | ;============================================================================= | ||
47 : | edgomez | 1.2 | |
48 : | Isibaar | 1.10.2.1 | TEXT |
49 : | edgomez | 1.2 | |
50 : | cglobal MBFieldTest_mmx | ||
51 : | |||
52 : | ; neater | ||
53 : | Isibaar | 1.9 | %define line0 _ESI |
54 : | %define line1 _ESI+16 | ||
55 : | %define line2 _ESI+32 | ||
56 : | %define line3 _ESI+48 | ||
57 : | %define line4 _ESI+64 | ||
58 : | %define line5 _ESI+80 | ||
59 : | %define line6 _ESI+96 | ||
60 : | %define line7 _ESI+112 | ||
61 : | %define line8 _EDI | ||
62 : | %define line9 _EDI+16 | ||
63 : | %define line10 _EDI+32 | ||
64 : | %define line11 _EDI+48 | ||
65 : | %define line12 _EDI+64 | ||
66 : | %define line13 _EDI+80 | ||
67 : | %define line14 _EDI+96 | ||
68 : | %define line15 _EDI+112 | ||
69 : | edgomez | 1.2 | |
70 : | ; keep from losing track which reg holds which line - these never overlap | ||
71 : | %define m00 mm0 | ||
72 : | %define m01 mm1 | ||
73 : | %define m02 mm2 | ||
74 : | %define m03 mm0 | ||
75 : | %define m04 mm1 | ||
76 : | %define m05 mm2 | ||
77 : | %define m06 mm0 | ||
78 : | %define m07 mm1 | ||
79 : | %define m08 mm2 | ||
80 : | %define m09 mm0 | ||
81 : | %define m10 mm1 | ||
82 : | %define m11 mm2 | ||
83 : | %define m12 mm0 | ||
84 : | %define m13 mm1 | ||
85 : | %define m14 mm2 | ||
86 : | %define m15 mm0 | ||
87 : | |||
88 : | ; gets diff between three lines low(%2),mid(%3),hi(%4): frame = mid-low, field = hi-low | ||
89 : | %macro ABS8 4 | ||
90 : | edgomez | 1.3 | movq %4, [%1] ; m02 = hi |
91 : | movq mm3, %2 ; mm3 = low copy | ||
92 : | edgomez | 1.2 | |
93 : | edgomez | 1.3 | pxor mm4, mm4 ; mm4 = 0 |
94 : | pxor mm5, mm5 ; mm5 = 0 | ||
95 : | edgomez | 1.2 | |
96 : | edgomez | 1.3 | psubw %2, %3 ; diff(med,low) for frame |
97 : | psubw mm3, %4 ; diff(hi,low) for field | ||
98 : | edgomez | 1.2 | |
99 : | edgomez | 1.3 | pcmpgtw mm4, %2 ; if (diff<0), mm4 will be all 1's, else all 0's |
100 : | pcmpgtw mm5, mm3 | ||
101 : | pxor %2, mm4 ; this will get abs(), but off by 1 if (diff<0) | ||
102 : | pxor mm3, mm5 | ||
103 : | psubw %2, mm4 ; correct abs being off by 1 when (diff<0) | ||
104 : | psubw mm3, mm5 | ||
105 : | edgomez | 1.2 | |
106 : | edgomez | 1.3 | paddw mm6, %2 ; add to totals |
107 : | paddw mm7, mm3 | ||
108 : | edgomez | 1.2 | %endmacro |
109 : | |||
110 : | edgomez | 1.3 | ;----------------------------------------------------------------------------- |
111 : | edgomez | 1.2 | ; |
112 : | ; uint32_t MBFieldTest_mmx(int16_t * const data); | ||
113 : | ; | ||
114 : | edgomez | 1.3 | ;----------------------------------------------------------------------------- |
115 : | edgomez | 1.2 | |
116 : | Isibaar | 1.9 | ALIGN SECTION_ALIGN |
117 : | edgomez | 1.2 | MBFieldTest_mmx: |
118 : | |||
119 : | Isibaar | 1.9 | mov _EAX, prm1 |
120 : | |||
121 : | push _ESI | ||
122 : | push _EDI | ||
123 : | edgomez | 1.2 | |
124 : | Isibaar | 1.9 | mov _ESI, _EAX ; _ESI = top left block |
125 : | mov _EDI, _ESI | ||
126 : | add _EDI, 256 ; _EDI = bottom left block | ||
127 : | edgomez | 1.3 | |
128 : | pxor mm6, mm6 ; frame total | ||
129 : | pxor mm7, mm7 ; field total | ||
130 : | |||
131 : | Isibaar | 1.9 | mov _EAX, 4 ; we do left 8 bytes of data[0*64], then right 8 bytes |
132 : | edgomez | 1.3 | ; then left 8 bytes of data[1*64], then last 8 bytes |
133 : | .loop: | ||
134 : | movq m00, [line0] ; line0 | ||
135 : | movq m01, [line1] ; line1 | ||
136 : | |||
137 : | ABS8 line2, m00, m01, m02 ; frame += (line2-line1), field += (line2-line0) | ||
138 : | ABS8 line3, m01, m02, m03 | ||
139 : | ABS8 line4, m02, m03, m04 | ||
140 : | ABS8 line5, m03, m04, m05 | ||
141 : | ABS8 line6, m04, m05, m06 | ||
142 : | ABS8 line7, m05, m06, m07 | ||
143 : | ABS8 line8, m06, m07, m08 | ||
144 : | |||
145 : | movq m09, [line9] ; line9-line7, no frame comp for line9-line8! | ||
146 : | pxor mm4, mm4 | ||
147 : | psubw m07, m09 | ||
148 : | pcmpgtw mm4, mm1 | ||
149 : | pxor m07, mm4 | ||
150 : | psubw m07, mm4 | ||
151 : | paddw mm7, m07 ; add to field total | ||
152 : | |||
153 : | ABS8 line10, m08, m09, m10 ; frame += (line10-line9), field += (line10-line8) | ||
154 : | ABS8 line11, m09, m10, m11 | ||
155 : | ABS8 line12, m10, m11, m12 | ||
156 : | ABS8 line13, m11, m12, m13 | ||
157 : | ABS8 line14, m12, m13, m14 | ||
158 : | ABS8 line15, m13, m14, m15 | ||
159 : | |||
160 : | pxor mm4, mm4 ; line15-line14, we're done with field comps! | ||
161 : | psubw m14, m15 | ||
162 : | pcmpgtw mm4, m14 | ||
163 : | pxor m14, mm4 | ||
164 : | psubw m14, mm4 | ||
165 : | paddw mm6, m14 ; add to frame total | ||
166 : | |||
167 : | Isibaar | 1.9 | lea TMP0, [nexts] |
168 : | Isibaar | 1.10 | mov TMP0d, dword [TMP0+_EAX*4] ; move _ESI/_EDI 8 pixels to the right |
169 : | Isibaar | 1.9 | add _ESI, TMP0 |
170 : | add _EDI, TMP0 | ||
171 : | edgomez | 1.3 | |
172 : | Isibaar | 1.9 | dec _EAX |
173 : | edgomez | 1.3 | jnz near .loop |
174 : | |||
175 : | .decide: | ||
176 : | movq mm0, [ones] ; add packed words into single dwords | ||
177 : | pmaddwd mm6, mm0 | ||
178 : | pmaddwd mm7, mm0 | ||
179 : | |||
180 : | Isibaar | 1.9 | movq mm0, mm6 ; TMP0 will be frame total, TMP1 field |
181 : | edgomez | 1.3 | movq mm1, mm7 |
182 : | psrlq mm0, 32 | ||
183 : | psrlq mm1, 32 | ||
184 : | paddd mm0, mm6 | ||
185 : | paddd mm1, mm7 | ||
186 : | Isibaar | 1.9 | movd TMP0d, mm0 |
187 : | movd TMP1d, mm1 | ||
188 : | edgomez | 1.3 | |
189 : | Isibaar | 1.9 | add TMP1, 350 ; add bias against field decision |
190 : | cmp TMP0, TMP1 | ||
191 : | edgomez | 1.3 | jb .end ; if frame<field, don't use field dct |
192 : | Isibaar | 1.9 | inc _EAX ; if frame>=field, use field dct (return 1) |
193 : | edgomez | 1.3 | |
194 : | .end: | ||
195 : | Isibaar | 1.9 | pop _EDI |
196 : | pop _ESI | ||
197 : | edgomez | 1.2 | |
198 : | edgomez | 1.3 | ret |
199 : | Isibaar | 1.8 | ENDFUNC |
200 : | edgomez | 1.6 | |
201 : | Isibaar | 1.7 | |
202 : | %ifidn __OUTPUT_FORMAT__,elf | ||
203 : | section ".note.GNU-stack" noalloc noexec nowrite progbits | ||
204 : | %endif | ||
205 : |
No admin address has been configured | ViewVC Help |
Powered by ViewVC 1.0.4 |