Mercurial > libavcodec.hg
annotate i386/mpegvideo_mmx_template.c @ 594:7e26eb37bbbc libavcodec
supporting rare overflow mess even with error_resilience>=0
author | michaelni |
---|---|
date | Sun, 04 Aug 2002 16:46:18 +0000 |
parents | 718a22dc121f |
children | b1a191202f96 |
rev | line source |
---|---|
220 | 1 /* |
429 | 2 * MPEG video MMX templates |
3 * | |
4 * Copyright (c) 2002 Michael Niedermayer <michaelni@gmx.at> | |
5 * | |
6 * This library is free software; you can redistribute it and/or | |
7 * modify it under the terms of the GNU Lesser General Public | |
8 * License as published by the Free Software Foundation; either | |
9 * version 2 of the License, or (at your option) any later version. | |
10 * | |
11 * This library is distributed in the hope that it will be useful, | |
12 * but WITHOUT ANY WARRANTY; without even the implied warranty of | |
13 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU | |
14 * Lesser General Public License for more details. | |
15 * | |
16 * You should have received a copy of the GNU Lesser General Public | |
17 * License along with this library; if not, write to the Free Software | |
18 * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA | |
19 */ | |
220 | 20 #undef SPREADW |
21 #undef PMAXW | |
22 #ifdef HAVE_MMX2 | |
23 #define SPREADW(a) "pshufw $0, " #a ", " #a " \n\t" | |
24 #define PMAXW(a,b) "pmaxsw " #a ", " #b " \n\t" | |
25 | |
26 #else | |
27 #define SPREADW(a) \ | |
28 "punpcklwd " #a ", " #a " \n\t"\ | |
29 "punpcklwd " #a ", " #a " \n\t" | |
30 #define PMAXW(a,b) \ | |
31 "psubusw " #a ", " #b " \n\t"\ | |
32 "paddw " #a ", " #b " \n\t" | |
33 #endif | |
34 | |
35 static int RENAME(dct_quantize)(MpegEncContext *s, | |
36 DCTELEM *block, int n, | |
344 | 37 int qscale, int *overflow) |
220 | 38 { |
344 | 39 int level=0, last_non_zero_p1, q; //=0 is cuz gcc says uninitalized ... |
40 const UINT16 *qmat, *bias; | |
220 | 41 static __align8 INT16 temp_block[64]; |
42 | |
43 av_fdct (block); | |
344 | 44 |
220 | 45 if (s->mb_intra) { |
46 int dummy; | |
47 if (n < 4) | |
48 q = s->y_dc_scale; | |
49 else | |
50 q = s->c_dc_scale; | |
51 /* note: block[0] is assumed to be positive */ | |
350
6ebbecc10063
- Advanced Intra Coding (AIC) support for H.263+ encoder, just DC by now.
pulento
parents:
344
diff
changeset
|
52 if (!s->h263_aic) { |
220 | 53 #if 1 |
344 | 54 asm volatile ( |
55 "xorl %%edx, %%edx \n\t" | |
56 "mul %%ecx \n\t" | |
57 : "=d" (level), "=a"(dummy) | |
58 : "a" (block[0] + (q >> 1)), "c" (inverse[q]) | |
59 ); | |
220 | 60 #else |
344 | 61 asm volatile ( |
62 "xorl %%edx, %%edx \n\t" | |
63 "divw %%cx \n\t" | |
64 "movzwl %%ax, %%eax \n\t" | |
65 : "=a" (level) | |
66 : "a" (block[0] + (q >> 1)), "c" (q) | |
67 : "%edx" | |
68 ); | |
220 | 69 #endif |
350
6ebbecc10063
- Advanced Intra Coding (AIC) support for H.263+ encoder, just DC by now.
pulento
parents:
344
diff
changeset
|
70 } else |
6ebbecc10063
- Advanced Intra Coding (AIC) support for H.263+ encoder, just DC by now.
pulento
parents:
344
diff
changeset
|
71 /* For AIC we skip quant/dequant of INTRADC */ |
6ebbecc10063
- Advanced Intra Coding (AIC) support for H.263+ encoder, just DC by now.
pulento
parents:
344
diff
changeset
|
72 level = block[0]; |
6ebbecc10063
- Advanced Intra Coding (AIC) support for H.263+ encoder, just DC by now.
pulento
parents:
344
diff
changeset
|
73 |
344 | 74 block[0]=0; //avoid fake overflow |
220 | 75 // temp_block[0] = (block[0] + (q >> 1)) / q; |
76 last_non_zero_p1 = 1; | |
344 | 77 bias = s->q_intra_matrix16_bias[qscale]; |
78 qmat = s->q_intra_matrix16[qscale]; | |
220 | 79 } else { |
80 last_non_zero_p1 = 0; | |
344 | 81 bias = s->q_inter_matrix16_bias[qscale]; |
82 qmat = s->q_inter_matrix16[qscale]; | |
220 | 83 } |
84 | |
344 | 85 if(s->out_format == FMT_H263){ |
86 | |
87 asm volatile( | |
88 "movd %%eax, %%mm3 \n\t" // last_non_zero_p1 | |
89 SPREADW(%%mm3) | |
90 "pxor %%mm7, %%mm7 \n\t" // 0 | |
91 "pxor %%mm4, %%mm4 \n\t" // 0 | |
92 "movq (%2), %%mm5 \n\t" // qmat[0] | |
93 "pxor %%mm6, %%mm6 \n\t" | |
94 "psubw (%3), %%mm6 \n\t" // -bias[0] | |
95 "movl $-128, %%eax \n\t" | |
96 ".balign 16 \n\t" | |
97 "1: \n\t" | |
98 "pxor %%mm1, %%mm1 \n\t" // 0 | |
99 "movq (%1, %%eax), %%mm0 \n\t" // block[i] | |
100 "pcmpgtw %%mm0, %%mm1 \n\t" // block[i] <= 0 ? 0xFF : 0x00 | |
101 "pxor %%mm1, %%mm0 \n\t" | |
102 "psubw %%mm1, %%mm0 \n\t" // ABS(block[i]) | |
103 "psubusw %%mm6, %%mm0 \n\t" // ABS(block[i]) + bias[0] | |
104 "pmulhw %%mm5, %%mm0 \n\t" // (ABS(block[i])*qmat[0] - bias[0]*qmat[0])>>16 | |
105 "por %%mm0, %%mm4 \n\t" | |
106 "pxor %%mm1, %%mm0 \n\t" | |
107 "psubw %%mm1, %%mm0 \n\t" // out=((ABS(block[i])*qmat[0] - bias[0]*qmat[0])>>16)*sign(block[i]) | |
108 "movq %%mm0, (%5, %%eax) \n\t" | |
109 "pcmpeqw %%mm7, %%mm0 \n\t" // out==0 ? 0xFF : 0x00 | |
110 "movq (%4, %%eax), %%mm1 \n\t" | |
111 "movq %%mm7, (%1, %%eax) \n\t" // 0 | |
112 "pandn %%mm1, %%mm0 \n\t" | |
113 PMAXW(%%mm0, %%mm3) | |
114 "addl $8, %%eax \n\t" | |
115 " js 1b \n\t" | |
116 "movq %%mm3, %%mm0 \n\t" | |
117 "psrlq $32, %%mm3 \n\t" | |
118 PMAXW(%%mm0, %%mm3) | |
119 "movq %%mm3, %%mm0 \n\t" | |
120 "psrlq $16, %%mm3 \n\t" | |
121 PMAXW(%%mm0, %%mm3) | |
122 "movd %%mm3, %%eax \n\t" | |
123 "movzbl %%al, %%eax \n\t" // last_non_zero_p1 | |
124 : "+a" (last_non_zero_p1) | |
125 : "r" (block+64), "r" (qmat), "r" (bias), | |
126 "r" (inv_zigzag_direct16+64), "r" (temp_block+64) | |
127 ); | |
128 // note the asm is split cuz gcc doesnt like that many operands ... | |
129 asm volatile( | |
130 "movd %1, %%mm1 \n\t" // max_qcoeff | |
131 SPREADW(%%mm1) | |
132 "psubusw %%mm1, %%mm4 \n\t" | |
133 "packuswb %%mm4, %%mm4 \n\t" | |
134 "movd %%mm4, %0 \n\t" // *overflow | |
135 : "=g" (*overflow) | |
136 : "g" (s->max_qcoeff) | |
137 ); | |
138 }else{ // FMT_H263 | |
139 asm volatile( | |
140 "movd %%eax, %%mm3 \n\t" // last_non_zero_p1 | |
141 SPREADW(%%mm3) | |
142 "pxor %%mm7, %%mm7 \n\t" // 0 | |
143 "pxor %%mm4, %%mm4 \n\t" // 0 | |
144 "movl $-128, %%eax \n\t" | |
145 ".balign 16 \n\t" | |
146 "1: \n\t" | |
147 "pxor %%mm1, %%mm1 \n\t" // 0 | |
148 "movq (%1, %%eax), %%mm0 \n\t" // block[i] | |
149 "pcmpgtw %%mm0, %%mm1 \n\t" // block[i] <= 0 ? 0xFF : 0x00 | |
150 "pxor %%mm1, %%mm0 \n\t" | |
151 "psubw %%mm1, %%mm0 \n\t" // ABS(block[i]) | |
152 "movq (%3, %%eax), %%mm6 \n\t" // bias[0] | |
153 "paddusw %%mm6, %%mm0 \n\t" // ABS(block[i]) + bias[0] | |
154 "movq (%2, %%eax), %%mm5 \n\t" // qmat[i] | |
155 "pmulhw %%mm5, %%mm0 \n\t" // (ABS(block[i])*qmat[0] + bias[0]*qmat[0])>>16 | |
156 "por %%mm0, %%mm4 \n\t" | |
157 "pxor %%mm1, %%mm0 \n\t" | |
158 "psubw %%mm1, %%mm0 \n\t" // out=((ABS(block[i])*qmat[0] - bias[0]*qmat[0])>>16)*sign(block[i]) | |
159 "movq %%mm0, (%5, %%eax) \n\t" | |
160 "pcmpeqw %%mm7, %%mm0 \n\t" // out==0 ? 0xFF : 0x00 | |
161 "movq (%4, %%eax), %%mm1 \n\t" | |
162 "movq %%mm7, (%1, %%eax) \n\t" // 0 | |
163 "pandn %%mm1, %%mm0 \n\t" | |
164 PMAXW(%%mm0, %%mm3) | |
165 "addl $8, %%eax \n\t" | |
166 " js 1b \n\t" | |
167 "movq %%mm3, %%mm0 \n\t" | |
168 "psrlq $32, %%mm3 \n\t" | |
169 PMAXW(%%mm0, %%mm3) | |
170 "movq %%mm3, %%mm0 \n\t" | |
171 "psrlq $16, %%mm3 \n\t" | |
172 PMAXW(%%mm0, %%mm3) | |
173 "movd %%mm3, %%eax \n\t" | |
174 "movzbl %%al, %%eax \n\t" // last_non_zero_p1 | |
175 : "+a" (last_non_zero_p1) | |
176 : "r" (block+64), "r" (qmat+64), "r" (bias+64), | |
177 "r" (inv_zigzag_direct16+64), "r" (temp_block+64) | |
178 ); | |
179 // note the asm is split cuz gcc doesnt like that many operands ... | |
180 asm volatile( | |
181 "movd %1, %%mm1 \n\t" // max_qcoeff | |
182 SPREADW(%%mm1) | |
183 "psubusw %%mm1, %%mm4 \n\t" | |
184 "packuswb %%mm4, %%mm4 \n\t" | |
185 "movd %%mm4, %0 \n\t" // *overflow | |
186 : "=g" (*overflow) | |
187 : "g" (s->max_qcoeff) | |
188 ); | |
189 } | |
190 | |
191 if(s->mb_intra) temp_block[0]= level; //FIXME move afer permute | |
220 | 192 // last_non_zero_p1=64; |
193 /* permute for IDCT */ | |
194 asm volatile( | |
344 | 195 "movl %0, %%eax \n\t" |
220 | 196 "pushl %%ebp \n\t" |
197 "movl %%esp, " MANGLE(esp_temp) "\n\t" | |
198 "1: \n\t" | |
199 "movzbl (%1, %%eax), %%ebx \n\t" | |
200 "movzbl 1(%1, %%eax), %%ebp \n\t" | |
201 "movw (%2, %%ebx, 2), %%cx \n\t" | |
202 "movw (%2, %%ebp, 2), %%sp \n\t" | |
203 "movzbl " MANGLE(permutation) "(%%ebx), %%ebx\n\t" | |
204 "movzbl " MANGLE(permutation) "(%%ebp), %%ebp\n\t" | |
205 "movw %%cx, (%3, %%ebx, 2) \n\t" | |
206 "movw %%sp, (%3, %%ebp, 2) \n\t" | |
207 "addl $2, %%eax \n\t" | |
208 " js 1b \n\t" | |
209 "movl " MANGLE(esp_temp) ", %%esp\n\t" | |
210 "popl %%ebp \n\t" | |
211 : | |
212 : "g" (-last_non_zero_p1), "d" (zigzag_direct_noperm+last_non_zero_p1), "S" (temp_block), "D" (block) | |
213 : "%eax", "%ebx", "%ecx" | |
214 ); | |
215 /* | |
216 for(i=0; i<last_non_zero_p1; i++) | |
217 { | |
218 int j= zigzag_direct_noperm[i]; | |
219 block[block_permute_op(j)]= temp_block[j]; | |
220 } | |
221 */ | |
222 //block_permute(block); | |
344 | 223 |
220 | 224 return last_non_zero_p1 - 1; |
225 } |