Line |
Branch |
Exec |
Source |
1 |
|
|
/* |
2 |
|
|
* MPEG video MMX templates |
3 |
|
|
* |
4 |
|
|
* Copyright (c) 2002 Michael Niedermayer <michaelni@gmx.at> |
5 |
|
|
* |
6 |
|
|
* This file is part of FFmpeg. |
7 |
|
|
* |
8 |
|
|
* FFmpeg is free software; you can redistribute it and/or |
9 |
|
|
* modify it under the terms of the GNU Lesser General Public |
10 |
|
|
* License as published by the Free Software Foundation; either |
11 |
|
|
* version 2.1 of the License, or (at your option) any later version. |
12 |
|
|
* |
13 |
|
|
* FFmpeg 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 GNU |
16 |
|
|
* Lesser General Public License for more details. |
17 |
|
|
* |
18 |
|
|
* You should have received a copy of the GNU Lesser General Public |
19 |
|
|
* License along with FFmpeg; if not, write to the Free Software |
20 |
|
|
* Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA |
21 |
|
|
*/ |
22 |
|
|
|
23 |
|
|
#include <stdint.h> |
24 |
|
|
|
25 |
|
|
#include "libavutil/internal.h" |
26 |
|
|
#include "libavutil/mem_internal.h" |
27 |
|
|
#include "libavutil/x86/asm.h" |
28 |
|
|
#include "libavcodec/mathops.h" |
29 |
|
|
#include "libavcodec/mpegutils.h" |
30 |
|
|
#include "libavcodec/mpegvideoenc.h" |
31 |
|
|
#include "fdct.h" |
32 |
|
|
|
33 |
|
|
#undef SPREADW |
34 |
|
|
#undef PMAXW |
35 |
|
|
#undef PMAX |
36 |
|
|
#undef SAVE_SIGN |
37 |
|
|
#undef RESTORE_SIGN |
38 |
|
|
|
39 |
|
|
#define SPREADW(a) \ |
40 |
|
|
"pshuflw $0, "a", "a" \n\t"\ |
41 |
|
|
"punpcklwd "a", "a" \n\t" |
42 |
|
|
#define PMAX(a,b) \ |
43 |
|
|
"movhlps "a", "b" \n\t"\ |
44 |
|
|
"pmaxsw "b", "a" \n\t"\ |
45 |
|
|
"pshuflw $0x0E, "a", "b" \n\t"\ |
46 |
|
|
"pmaxsw "b", "a" \n\t"\ |
47 |
|
|
"pshuflw $0x01, "a", "b" \n\t"\ |
48 |
|
|
"pmaxsw "b", "a" \n\t" |
49 |
|
|
|
50 |
|
|
#if COMPILE_TEMPLATE_SSSE3 |
51 |
|
|
#define SAVE_SIGN(a,b) \ |
52 |
|
|
"movdqa "b", "a" \n\t"\ |
53 |
|
|
"pabsw "b", "b" \n\t" |
54 |
|
|
#define RESTORE_SIGN(a,b) \ |
55 |
|
|
"psignw "a", "b" \n\t" |
56 |
|
|
#else |
57 |
|
|
#define SAVE_SIGN(a,b) \ |
58 |
|
|
"pxor "a", "a" \n\t"\ |
59 |
|
|
"pcmpgtw "b", "a" \n\t" /* block[i] <= 0 ? 0xFF : 0x00 */\ |
60 |
|
|
"pxor "a", "b" \n\t"\ |
61 |
|
|
"psubw "a", "b" \n\t" /* ABS(block[i]) */ |
62 |
|
|
#define RESTORE_SIGN(a,b) \ |
63 |
|
|
"pxor "a", "b" \n\t"\ |
64 |
|
|
"psubw "a", "b" \n\t" // out=((ABS(block[i])*qmat[0] - bias[0]*qmat[0])>>16)*sign(block[i]) |
65 |
|
|
#endif |
66 |
|
|
|
67 |
|
✗ |
static int RENAME(dct_quantize)(MPVEncContext *const s, |
68 |
|
|
int16_t *block, int n, |
69 |
|
|
int qscale, int *overflow) |
70 |
|
|
{ |
71 |
|
|
x86_reg last_non_zero_p1; |
72 |
|
✗ |
int level=0, q; //=0 is because gcc says uninitialized ... |
73 |
|
|
const uint16_t *qmat, *bias; |
74 |
|
✗ |
LOCAL_ALIGNED_16(int16_t, temp_block, [64]); |
75 |
|
|
|
76 |
|
|
//s->fdct (block); |
77 |
|
✗ |
ff_fdct_sse2(block); // cannot be anything else ... |
78 |
|
|
|
79 |
|
✗ |
if(s->dct_error_sum) |
80 |
|
✗ |
s->denoise_dct(s, block); |
81 |
|
|
|
82 |
|
✗ |
if (s->c.mb_intra) { |
83 |
|
|
int dummy; |
84 |
|
✗ |
if (n < 4){ |
85 |
|
✗ |
q = s->c.y_dc_scale; |
86 |
|
✗ |
bias = s->q_intra_matrix16[qscale][1]; |
87 |
|
✗ |
qmat = s->q_intra_matrix16[qscale][0]; |
88 |
|
|
}else{ |
89 |
|
✗ |
q = s->c.c_dc_scale; |
90 |
|
✗ |
bias = s->q_chroma_intra_matrix16[qscale][1]; |
91 |
|
✗ |
qmat = s->q_chroma_intra_matrix16[qscale][0]; |
92 |
|
|
} |
93 |
|
|
/* note: block[0] is assumed to be positive */ |
94 |
|
✗ |
if (!s->c.h263_aic) { |
95 |
|
✗ |
__asm__ volatile ( |
96 |
|
|
"mul %%ecx \n\t" |
97 |
|
|
: "=d" (level), "=a"(dummy) |
98 |
|
✗ |
: "a" ((block[0]>>2) + q), "c" (ff_inverse[q<<1]) |
99 |
|
|
); |
100 |
|
|
} else |
101 |
|
|
/* For AIC we skip quant/dequant of INTRADC */ |
102 |
|
✗ |
level = (block[0] + 4)>>3; |
103 |
|
|
|
104 |
|
✗ |
block[0]=0; //avoid fake overflow |
105 |
|
|
// temp_block[0] = (block[0] + (q >> 1)) / q; |
106 |
|
✗ |
last_non_zero_p1 = 1; |
107 |
|
|
} else { |
108 |
|
✗ |
last_non_zero_p1 = 0; |
109 |
|
✗ |
bias = s->q_inter_matrix16[qscale][1]; |
110 |
|
✗ |
qmat = s->q_inter_matrix16[qscale][0]; |
111 |
|
|
} |
112 |
|
|
|
113 |
|
✗ |
if ((s->c.out_format == FMT_H263 || s->c.out_format == FMT_H261) && !s->mpeg_quant) { |
114 |
|
✗ |
__asm__ volatile( |
115 |
|
|
"movd %%"FF_REG_a", %%xmm3 \n\t" // last_non_zero_p1 |
116 |
|
|
SPREADW("%%xmm3") |
117 |
|
|
"pxor %%xmm7, %%xmm7 \n\t" // 0 |
118 |
|
|
"pxor %%xmm4, %%xmm4 \n\t" // 0 |
119 |
|
|
"movdqa (%2), %%xmm5 \n\t" // qmat[0] |
120 |
|
|
"pxor %%xmm6, %%xmm6 \n\t" |
121 |
|
|
"psubw (%3), %%xmm6 \n\t" // -bias[0] |
122 |
|
|
"mov $-128, %%"FF_REG_a" \n\t" |
123 |
|
|
".p2align 4 \n\t" |
124 |
|
|
"1: \n\t" |
125 |
|
|
"movdqa (%1, %%"FF_REG_a"), %%xmm0 \n\t" // block[i] |
126 |
|
|
SAVE_SIGN("%%xmm1", "%%xmm0") // ABS(block[i]) |
127 |
|
|
"psubusw %%xmm6, %%xmm0 \n\t" // ABS(block[i]) + bias[0] |
128 |
|
|
"pmulhw %%xmm5, %%xmm0 \n\t" // (ABS(block[i])*qmat[0] - bias[0]*qmat[0])>>16 |
129 |
|
|
"por %%xmm0, %%xmm4 \n\t" |
130 |
|
|
RESTORE_SIGN("%%xmm1", "%%xmm0") // out=((ABS(block[i])*qmat[0] - bias[0]*qmat[0])>>16)*sign(block[i]) |
131 |
|
|
"movdqa %%xmm0, (%5, %%"FF_REG_a") \n\t" |
132 |
|
|
"pcmpeqw %%xmm7, %%xmm0 \n\t" // out==0 ? 0xFF : 0x00 |
133 |
|
|
"movdqa (%4, %%"FF_REG_a"), %%xmm1 \n\t" |
134 |
|
|
"movdqa %%xmm7, (%1, %%"FF_REG_a") \n\t" // 0 |
135 |
|
|
"pandn %%xmm1, %%xmm0 \n\t" |
136 |
|
|
"pmaxsw %%xmm0, %%xmm3 \n\t" |
137 |
|
|
"add $16, %%"FF_REG_a" \n\t" |
138 |
|
|
" js 1b \n\t" |
139 |
|
|
PMAX("%%xmm3", "%%xmm0") |
140 |
|
|
"movd %%xmm3, %%"FF_REG_a" \n\t" |
141 |
|
|
"movzbl %%al, %%eax \n\t" // last_non_zero_p1 |
142 |
|
|
: "+a" (last_non_zero_p1) |
143 |
|
✗ |
: "r" (block+64), "r" (qmat), "r" (bias), |
144 |
|
✗ |
"r" (inv_zigzag_direct16 + 64), "r" (temp_block + 64) |
145 |
|
|
XMM_CLOBBERS_ONLY("%xmm0", "%xmm1", "%xmm2", "%xmm3", |
146 |
|
|
"%xmm4", "%xmm5", "%xmm6", "%xmm7") |
147 |
|
|
); |
148 |
|
|
}else{ // FMT_H263 |
149 |
|
✗ |
__asm__ volatile( |
150 |
|
|
"movd %%"FF_REG_a", %%xmm3 \n\t" // last_non_zero_p1 |
151 |
|
|
SPREADW("%%xmm3") |
152 |
|
|
"pxor %%xmm7, %%xmm7 \n\t" // 0 |
153 |
|
|
"pxor %%xmm4, %%xmm4 \n\t" // 0 |
154 |
|
|
"mov $-128, %%"FF_REG_a" \n\t" |
155 |
|
|
".p2align 4 \n\t" |
156 |
|
|
"1: \n\t" |
157 |
|
|
"movdqa (%1, %%"FF_REG_a"), %%xmm0 \n\t" // block[i] |
158 |
|
|
SAVE_SIGN("%%xmm1", "%%xmm0") // ABS(block[i]) |
159 |
|
|
"movdqa (%3, %%"FF_REG_a"), %%xmm6 \n\t" // bias[0] |
160 |
|
|
"paddusw %%xmm6, %%xmm0 \n\t" // ABS(block[i]) + bias[0] |
161 |
|
|
"movdqa (%2, %%"FF_REG_a"), %%xmm5 \n\t" // qmat[i] |
162 |
|
|
"pmulhw %%xmm5, %%xmm0 \n\t" // (ABS(block[i])*qmat[0] + bias[0]*qmat[0])>>16 |
163 |
|
|
"por %%xmm0, %%xmm4 \n\t" |
164 |
|
|
RESTORE_SIGN("%%xmm1", "%%xmm0") // out=((ABS(block[i])*qmat[0] - bias[0]*qmat[0])>>16)*sign(block[i]) |
165 |
|
|
"movdqa %%xmm0, (%5, %%"FF_REG_a") \n\t" |
166 |
|
|
"pcmpeqw %%xmm7, %%xmm0 \n\t" // out==0 ? 0xFF : 0x00 |
167 |
|
|
"movdqa (%4, %%"FF_REG_a"), %%xmm1 \n\t" |
168 |
|
|
"movdqa %%xmm7, (%1, %%"FF_REG_a") \n\t" // 0 |
169 |
|
|
"pandn %%xmm1, %%xmm0 \n\t" |
170 |
|
|
"pmaxsw %%xmm0, %%xmm3 \n\t" |
171 |
|
|
"add $16, %%"FF_REG_a" \n\t" |
172 |
|
|
" js 1b \n\t" |
173 |
|
|
PMAX("%%xmm3", "%%xmm0") |
174 |
|
|
"movd %%xmm3, %%"FF_REG_a" \n\t" |
175 |
|
|
"movzbl %%al, %%eax \n\t" // last_non_zero_p1 |
176 |
|
|
: "+a" (last_non_zero_p1) |
177 |
|
✗ |
: "r" (block+64), "r" (qmat+64), "r" (bias+64), |
178 |
|
✗ |
"r" (inv_zigzag_direct16 + 64), "r" (temp_block + 64) |
179 |
|
|
XMM_CLOBBERS_ONLY("%xmm0", "%xmm1", "%xmm2", "%xmm3", |
180 |
|
|
"%xmm4", "%xmm5", "%xmm6", "%xmm7") |
181 |
|
|
); |
182 |
|
|
} |
183 |
|
✗ |
__asm__ volatile( |
184 |
|
|
"movd %1, %%xmm1 \n\t" // max_qcoeff |
185 |
|
|
SPREADW("%%xmm1") |
186 |
|
|
"psubusw %%xmm1, %%xmm4 \n\t" |
187 |
|
|
"packuswb %%xmm4, %%xmm4 \n\t" |
188 |
|
|
"packsswb %%xmm4, %%xmm4 \n\t" |
189 |
|
|
"movd %%xmm4, %0 \n\t" // *overflow |
190 |
|
|
: "=g" (*overflow) |
191 |
|
✗ |
: "g" (s->max_qcoeff) |
192 |
|
|
); |
193 |
|
|
|
194 |
|
✗ |
block[0] = s->c.mb_intra ? level : temp_block[0]; |
195 |
|
|
|
196 |
|
|
av_assert2(ARCH_X86_32 || s->c.idsp.perm_type != FF_IDCT_PERM_SIMPLE); |
197 |
|
|
if (ARCH_X86_32 && s->c.idsp.perm_type == FF_IDCT_PERM_SIMPLE) { |
198 |
|
|
if(last_non_zero_p1 <= 1) goto end; |
199 |
|
|
block[0x08] = temp_block[0x01]; block[0x10] = temp_block[0x08]; |
200 |
|
|
block[0x20] = temp_block[0x10]; |
201 |
|
|
if(last_non_zero_p1 <= 4) goto end; |
202 |
|
|
block[0x18] = temp_block[0x09]; block[0x04] = temp_block[0x02]; |
203 |
|
|
block[0x09] = temp_block[0x03]; |
204 |
|
|
if(last_non_zero_p1 <= 7) goto end; |
205 |
|
|
block[0x14] = temp_block[0x0A]; block[0x28] = temp_block[0x11]; |
206 |
|
|
block[0x12] = temp_block[0x18]; block[0x02] = temp_block[0x20]; |
207 |
|
|
if(last_non_zero_p1 <= 11) goto end; |
208 |
|
|
block[0x1A] = temp_block[0x19]; block[0x24] = temp_block[0x12]; |
209 |
|
|
block[0x19] = temp_block[0x0B]; block[0x01] = temp_block[0x04]; |
210 |
|
|
block[0x0C] = temp_block[0x05]; |
211 |
|
|
if(last_non_zero_p1 <= 16) goto end; |
212 |
|
|
block[0x11] = temp_block[0x0C]; block[0x29] = temp_block[0x13]; |
213 |
|
|
block[0x16] = temp_block[0x1A]; block[0x0A] = temp_block[0x21]; |
214 |
|
|
block[0x30] = temp_block[0x28]; block[0x22] = temp_block[0x30]; |
215 |
|
|
block[0x38] = temp_block[0x29]; block[0x06] = temp_block[0x22]; |
216 |
|
|
if(last_non_zero_p1 <= 24) goto end; |
217 |
|
|
block[0x1B] = temp_block[0x1B]; block[0x21] = temp_block[0x14]; |
218 |
|
|
block[0x1C] = temp_block[0x0D]; block[0x05] = temp_block[0x06]; |
219 |
|
|
block[0x0D] = temp_block[0x07]; block[0x15] = temp_block[0x0E]; |
220 |
|
|
block[0x2C] = temp_block[0x15]; block[0x13] = temp_block[0x1C]; |
221 |
|
|
if(last_non_zero_p1 <= 32) goto end; |
222 |
|
|
block[0x0B] = temp_block[0x23]; block[0x34] = temp_block[0x2A]; |
223 |
|
|
block[0x2A] = temp_block[0x31]; block[0x32] = temp_block[0x38]; |
224 |
|
|
block[0x3A] = temp_block[0x39]; block[0x26] = temp_block[0x32]; |
225 |
|
|
block[0x39] = temp_block[0x2B]; block[0x03] = temp_block[0x24]; |
226 |
|
|
if(last_non_zero_p1 <= 40) goto end; |
227 |
|
|
block[0x1E] = temp_block[0x1D]; block[0x25] = temp_block[0x16]; |
228 |
|
|
block[0x1D] = temp_block[0x0F]; block[0x2D] = temp_block[0x17]; |
229 |
|
|
block[0x17] = temp_block[0x1E]; block[0x0E] = temp_block[0x25]; |
230 |
|
|
block[0x31] = temp_block[0x2C]; block[0x2B] = temp_block[0x33]; |
231 |
|
|
if(last_non_zero_p1 <= 48) goto end; |
232 |
|
|
block[0x36] = temp_block[0x3A]; block[0x3B] = temp_block[0x3B]; |
233 |
|
|
block[0x23] = temp_block[0x34]; block[0x3C] = temp_block[0x2D]; |
234 |
|
|
block[0x07] = temp_block[0x26]; block[0x1F] = temp_block[0x1F]; |
235 |
|
|
block[0x0F] = temp_block[0x27]; block[0x35] = temp_block[0x2E]; |
236 |
|
|
if(last_non_zero_p1 <= 56) goto end; |
237 |
|
|
block[0x2E] = temp_block[0x35]; block[0x33] = temp_block[0x3C]; |
238 |
|
|
block[0x3E] = temp_block[0x3D]; block[0x27] = temp_block[0x36]; |
239 |
|
|
block[0x3D] = temp_block[0x2F]; block[0x2F] = temp_block[0x37]; |
240 |
|
|
block[0x37] = temp_block[0x3E]; block[0x3F] = temp_block[0x3F]; |
241 |
|
✗ |
} else if (s->c.idsp.perm_type == FF_IDCT_PERM_LIBMPEG2) { |
242 |
|
✗ |
if(last_non_zero_p1 <= 1) goto end; |
243 |
|
✗ |
block[0x04] = temp_block[0x01]; |
244 |
|
✗ |
block[0x08] = temp_block[0x08]; block[0x10] = temp_block[0x10]; |
245 |
|
✗ |
if(last_non_zero_p1 <= 4) goto end; |
246 |
|
✗ |
block[0x0C] = temp_block[0x09]; block[0x01] = temp_block[0x02]; |
247 |
|
✗ |
block[0x05] = temp_block[0x03]; |
248 |
|
✗ |
if(last_non_zero_p1 <= 7) goto end; |
249 |
|
✗ |
block[0x09] = temp_block[0x0A]; block[0x14] = temp_block[0x11]; |
250 |
|
✗ |
block[0x18] = temp_block[0x18]; block[0x20] = temp_block[0x20]; |
251 |
|
✗ |
if(last_non_zero_p1 <= 11) goto end; |
252 |
|
✗ |
block[0x1C] = temp_block[0x19]; |
253 |
|
✗ |
block[0x11] = temp_block[0x12]; block[0x0D] = temp_block[0x0B]; |
254 |
|
✗ |
block[0x02] = temp_block[0x04]; block[0x06] = temp_block[0x05]; |
255 |
|
✗ |
if(last_non_zero_p1 <= 16) goto end; |
256 |
|
✗ |
block[0x0A] = temp_block[0x0C]; block[0x15] = temp_block[0x13]; |
257 |
|
✗ |
block[0x19] = temp_block[0x1A]; block[0x24] = temp_block[0x21]; |
258 |
|
✗ |
block[0x28] = temp_block[0x28]; block[0x30] = temp_block[0x30]; |
259 |
|
✗ |
block[0x2C] = temp_block[0x29]; block[0x21] = temp_block[0x22]; |
260 |
|
✗ |
if(last_non_zero_p1 <= 24) goto end; |
261 |
|
✗ |
block[0x1D] = temp_block[0x1B]; block[0x12] = temp_block[0x14]; |
262 |
|
✗ |
block[0x0E] = temp_block[0x0D]; block[0x03] = temp_block[0x06]; |
263 |
|
✗ |
block[0x07] = temp_block[0x07]; block[0x0B] = temp_block[0x0E]; |
264 |
|
✗ |
block[0x16] = temp_block[0x15]; block[0x1A] = temp_block[0x1C]; |
265 |
|
✗ |
if(last_non_zero_p1 <= 32) goto end; |
266 |
|
✗ |
block[0x25] = temp_block[0x23]; block[0x29] = temp_block[0x2A]; |
267 |
|
✗ |
block[0x34] = temp_block[0x31]; block[0x38] = temp_block[0x38]; |
268 |
|
✗ |
block[0x3C] = temp_block[0x39]; block[0x31] = temp_block[0x32]; |
269 |
|
✗ |
block[0x2D] = temp_block[0x2B]; block[0x22] = temp_block[0x24]; |
270 |
|
✗ |
if(last_non_zero_p1 <= 40) goto end; |
271 |
|
✗ |
block[0x1E] = temp_block[0x1D]; block[0x13] = temp_block[0x16]; |
272 |
|
✗ |
block[0x0F] = temp_block[0x0F]; block[0x17] = temp_block[0x17]; |
273 |
|
✗ |
block[0x1B] = temp_block[0x1E]; block[0x26] = temp_block[0x25]; |
274 |
|
✗ |
block[0x2A] = temp_block[0x2C]; block[0x35] = temp_block[0x33]; |
275 |
|
✗ |
if(last_non_zero_p1 <= 48) goto end; |
276 |
|
✗ |
block[0x39] = temp_block[0x3A]; block[0x3D] = temp_block[0x3B]; |
277 |
|
✗ |
block[0x32] = temp_block[0x34]; block[0x2E] = temp_block[0x2D]; |
278 |
|
✗ |
block[0x23] = temp_block[0x26]; block[0x1F] = temp_block[0x1F]; |
279 |
|
✗ |
block[0x27] = temp_block[0x27]; block[0x2B] = temp_block[0x2E]; |
280 |
|
✗ |
if(last_non_zero_p1 <= 56) goto end; |
281 |
|
✗ |
block[0x36] = temp_block[0x35]; block[0x3A] = temp_block[0x3C]; |
282 |
|
✗ |
block[0x3E] = temp_block[0x3D]; block[0x33] = temp_block[0x36]; |
283 |
|
✗ |
block[0x2F] = temp_block[0x2F]; block[0x37] = temp_block[0x37]; |
284 |
|
✗ |
block[0x3B] = temp_block[0x3E]; block[0x3F] = temp_block[0x3F]; |
285 |
|
✗ |
} else if (s->c.idsp.perm_type == FF_IDCT_PERM_NONE) { |
286 |
|
✗ |
if(last_non_zero_p1 <= 1) goto end; |
287 |
|
✗ |
block[0x01] = temp_block[0x01]; |
288 |
|
✗ |
block[0x08] = temp_block[0x08]; block[0x10] = temp_block[0x10]; |
289 |
|
✗ |
if(last_non_zero_p1 <= 4) goto end; |
290 |
|
✗ |
block[0x09] = temp_block[0x09]; block[0x02] = temp_block[0x02]; |
291 |
|
✗ |
block[0x03] = temp_block[0x03]; |
292 |
|
✗ |
if(last_non_zero_p1 <= 7) goto end; |
293 |
|
✗ |
block[0x0A] = temp_block[0x0A]; block[0x11] = temp_block[0x11]; |
294 |
|
✗ |
block[0x18] = temp_block[0x18]; block[0x20] = temp_block[0x20]; |
295 |
|
✗ |
if(last_non_zero_p1 <= 11) goto end; |
296 |
|
✗ |
block[0x19] = temp_block[0x19]; |
297 |
|
✗ |
block[0x12] = temp_block[0x12]; block[0x0B] = temp_block[0x0B]; |
298 |
|
✗ |
block[0x04] = temp_block[0x04]; block[0x05] = temp_block[0x05]; |
299 |
|
✗ |
if(last_non_zero_p1 <= 16) goto end; |
300 |
|
✗ |
block[0x0C] = temp_block[0x0C]; block[0x13] = temp_block[0x13]; |
301 |
|
✗ |
block[0x1A] = temp_block[0x1A]; block[0x21] = temp_block[0x21]; |
302 |
|
✗ |
block[0x28] = temp_block[0x28]; block[0x30] = temp_block[0x30]; |
303 |
|
✗ |
block[0x29] = temp_block[0x29]; block[0x22] = temp_block[0x22]; |
304 |
|
✗ |
if(last_non_zero_p1 <= 24) goto end; |
305 |
|
✗ |
block[0x1B] = temp_block[0x1B]; block[0x14] = temp_block[0x14]; |
306 |
|
✗ |
block[0x0D] = temp_block[0x0D]; block[0x06] = temp_block[0x06]; |
307 |
|
✗ |
block[0x07] = temp_block[0x07]; block[0x0E] = temp_block[0x0E]; |
308 |
|
✗ |
block[0x15] = temp_block[0x15]; block[0x1C] = temp_block[0x1C]; |
309 |
|
✗ |
if(last_non_zero_p1 <= 32) goto end; |
310 |
|
✗ |
block[0x23] = temp_block[0x23]; block[0x2A] = temp_block[0x2A]; |
311 |
|
✗ |
block[0x31] = temp_block[0x31]; block[0x38] = temp_block[0x38]; |
312 |
|
✗ |
block[0x39] = temp_block[0x39]; block[0x32] = temp_block[0x32]; |
313 |
|
✗ |
block[0x2B] = temp_block[0x2B]; block[0x24] = temp_block[0x24]; |
314 |
|
✗ |
if(last_non_zero_p1 <= 40) goto end; |
315 |
|
✗ |
block[0x1D] = temp_block[0x1D]; block[0x16] = temp_block[0x16]; |
316 |
|
✗ |
block[0x0F] = temp_block[0x0F]; block[0x17] = temp_block[0x17]; |
317 |
|
✗ |
block[0x1E] = temp_block[0x1E]; block[0x25] = temp_block[0x25]; |
318 |
|
✗ |
block[0x2C] = temp_block[0x2C]; block[0x33] = temp_block[0x33]; |
319 |
|
✗ |
if(last_non_zero_p1 <= 48) goto end; |
320 |
|
✗ |
block[0x3A] = temp_block[0x3A]; block[0x3B] = temp_block[0x3B]; |
321 |
|
✗ |
block[0x34] = temp_block[0x34]; block[0x2D] = temp_block[0x2D]; |
322 |
|
✗ |
block[0x26] = temp_block[0x26]; block[0x1F] = temp_block[0x1F]; |
323 |
|
✗ |
block[0x27] = temp_block[0x27]; block[0x2E] = temp_block[0x2E]; |
324 |
|
✗ |
if(last_non_zero_p1 <= 56) goto end; |
325 |
|
✗ |
block[0x35] = temp_block[0x35]; block[0x3C] = temp_block[0x3C]; |
326 |
|
✗ |
block[0x3D] = temp_block[0x3D]; block[0x36] = temp_block[0x36]; |
327 |
|
✗ |
block[0x2F] = temp_block[0x2F]; block[0x37] = temp_block[0x37]; |
328 |
|
✗ |
block[0x3E] = temp_block[0x3E]; block[0x3F] = temp_block[0x3F]; |
329 |
|
✗ |
} else if (s->c.idsp.perm_type == FF_IDCT_PERM_TRANSPOSE) { |
330 |
|
✗ |
if(last_non_zero_p1 <= 1) goto end; |
331 |
|
✗ |
block[0x08] = temp_block[0x01]; |
332 |
|
✗ |
block[0x01] = temp_block[0x08]; block[0x02] = temp_block[0x10]; |
333 |
|
✗ |
if(last_non_zero_p1 <= 4) goto end; |
334 |
|
✗ |
block[0x09] = temp_block[0x09]; block[0x10] = temp_block[0x02]; |
335 |
|
✗ |
block[0x18] = temp_block[0x03]; |
336 |
|
✗ |
if(last_non_zero_p1 <= 7) goto end; |
337 |
|
✗ |
block[0x11] = temp_block[0x0A]; block[0x0A] = temp_block[0x11]; |
338 |
|
✗ |
block[0x03] = temp_block[0x18]; block[0x04] = temp_block[0x20]; |
339 |
|
✗ |
if(last_non_zero_p1 <= 11) goto end; |
340 |
|
✗ |
block[0x0B] = temp_block[0x19]; |
341 |
|
✗ |
block[0x12] = temp_block[0x12]; block[0x19] = temp_block[0x0B]; |
342 |
|
✗ |
block[0x20] = temp_block[0x04]; block[0x28] = temp_block[0x05]; |
343 |
|
✗ |
if(last_non_zero_p1 <= 16) goto end; |
344 |
|
✗ |
block[0x21] = temp_block[0x0C]; block[0x1A] = temp_block[0x13]; |
345 |
|
✗ |
block[0x13] = temp_block[0x1A]; block[0x0C] = temp_block[0x21]; |
346 |
|
✗ |
block[0x05] = temp_block[0x28]; block[0x06] = temp_block[0x30]; |
347 |
|
✗ |
block[0x0D] = temp_block[0x29]; block[0x14] = temp_block[0x22]; |
348 |
|
✗ |
if(last_non_zero_p1 <= 24) goto end; |
349 |
|
✗ |
block[0x1B] = temp_block[0x1B]; block[0x22] = temp_block[0x14]; |
350 |
|
✗ |
block[0x29] = temp_block[0x0D]; block[0x30] = temp_block[0x06]; |
351 |
|
✗ |
block[0x38] = temp_block[0x07]; block[0x31] = temp_block[0x0E]; |
352 |
|
✗ |
block[0x2A] = temp_block[0x15]; block[0x23] = temp_block[0x1C]; |
353 |
|
✗ |
if(last_non_zero_p1 <= 32) goto end; |
354 |
|
✗ |
block[0x1C] = temp_block[0x23]; block[0x15] = temp_block[0x2A]; |
355 |
|
✗ |
block[0x0E] = temp_block[0x31]; block[0x07] = temp_block[0x38]; |
356 |
|
✗ |
block[0x0F] = temp_block[0x39]; block[0x16] = temp_block[0x32]; |
357 |
|
✗ |
block[0x1D] = temp_block[0x2B]; block[0x24] = temp_block[0x24]; |
358 |
|
✗ |
if(last_non_zero_p1 <= 40) goto end; |
359 |
|
✗ |
block[0x2B] = temp_block[0x1D]; block[0x32] = temp_block[0x16]; |
360 |
|
✗ |
block[0x39] = temp_block[0x0F]; block[0x3A] = temp_block[0x17]; |
361 |
|
✗ |
block[0x33] = temp_block[0x1E]; block[0x2C] = temp_block[0x25]; |
362 |
|
✗ |
block[0x25] = temp_block[0x2C]; block[0x1E] = temp_block[0x33]; |
363 |
|
✗ |
if(last_non_zero_p1 <= 48) goto end; |
364 |
|
✗ |
block[0x17] = temp_block[0x3A]; block[0x1F] = temp_block[0x3B]; |
365 |
|
✗ |
block[0x26] = temp_block[0x34]; block[0x2D] = temp_block[0x2D]; |
366 |
|
✗ |
block[0x34] = temp_block[0x26]; block[0x3B] = temp_block[0x1F]; |
367 |
|
✗ |
block[0x3C] = temp_block[0x27]; block[0x35] = temp_block[0x2E]; |
368 |
|
✗ |
if(last_non_zero_p1 <= 56) goto end; |
369 |
|
✗ |
block[0x2E] = temp_block[0x35]; block[0x27] = temp_block[0x3C]; |
370 |
|
✗ |
block[0x2F] = temp_block[0x3D]; block[0x36] = temp_block[0x36]; |
371 |
|
✗ |
block[0x3D] = temp_block[0x2F]; block[0x3E] = temp_block[0x37]; |
372 |
|
✗ |
block[0x37] = temp_block[0x3E]; block[0x3F] = temp_block[0x3F]; |
373 |
|
|
} else { |
374 |
|
✗ |
av_log(s->c.avctx, AV_LOG_DEBUG, "s->c.idsp.perm_type: %d\n", |
375 |
|
✗ |
(int)s->c.idsp.perm_type); |
376 |
|
✗ |
av_assert0(s->c.idsp.perm_type == FF_IDCT_PERM_NONE || |
377 |
|
|
s->c.idsp.perm_type == FF_IDCT_PERM_LIBMPEG2 || |
378 |
|
|
s->c.idsp.perm_type == FF_IDCT_PERM_SIMPLE || |
379 |
|
|
s->c.idsp.perm_type == FF_IDCT_PERM_TRANSPOSE); |
380 |
|
|
} |
381 |
|
✗ |
end: |
382 |
|
✗ |
return last_non_zero_p1 - 1; |
383 |
|
|
} |
384 |
|
|
|