FFmpeg coverage


Directory: ../../../ffmpeg/
File: src/libavcodec/x86/mpegvideoenc_template.c
Date: 2025-05-08 14:53:23
Exec Total Coverage
Lines: 0 170 0.0%
Functions: 0 0 -%
Branches: 0 0 -%

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