| 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 | #include "mpegvideoencdsp.h" | ||
| 33 | |||
| 34 | #undef SPREADW | ||
| 35 | #undef PMAXW | ||
| 36 | #undef PMAX | ||
| 37 | #undef SAVE_SIGN | ||
| 38 | #undef RESTORE_SIGN | ||
| 39 | |||
| 40 | #define SPREADW(a) \ | ||
| 41 | "pshuflw $0, "a", "a" \n\t"\ | ||
| 42 | "punpcklwd "a", "a" \n\t" | ||
| 43 | #define PMAX(a,b) \ | ||
| 44 | "movhlps "a", "b" \n\t"\ | ||
| 45 | "pmaxsw "b", "a" \n\t"\ | ||
| 46 | "pshuflw $0x0E, "a", "b" \n\t"\ | ||
| 47 | "pmaxsw "b", "a" \n\t"\ | ||
| 48 | "pshuflw $0x01, "a", "b" \n\t"\ | ||
| 49 | "pmaxsw "b", "a" \n\t" | ||
| 50 | |||
| 51 | #if COMPILE_TEMPLATE_SSSE3 | ||
| 52 | #define SAVE_SIGN(a,b) \ | ||
| 53 | "movdqa "b", "a" \n\t"\ | ||
| 54 | "pabsw "b", "b" \n\t" | ||
| 55 | #define RESTORE_SIGN(a,b) \ | ||
| 56 | "psignw "a", "b" \n\t" | ||
| 57 | #else | ||
| 58 | #define SAVE_SIGN(a,b) \ | ||
| 59 | "pxor "a", "a" \n\t"\ | ||
| 60 | "pcmpgtw "b", "a" \n\t" /* block[i] <= 0 ? 0xFF : 0x00 */\ | ||
| 61 | "pxor "a", "b" \n\t"\ | ||
| 62 | "psubw "a", "b" \n\t" /* ABS(block[i]) */ | ||
| 63 | #define RESTORE_SIGN(a,b) \ | ||
| 64 | "pxor "a", "b" \n\t"\ | ||
| 65 | "psubw "a", "b" \n\t" // out=((ABS(block[i])*qmat[0] - bias[0]*qmat[0])>>16)*sign(block[i]) | ||
| 66 | #endif | ||
| 67 | |||
| 68 | 7290 | static int RENAME(dct_quantize)(MPVEncContext *const s, | |
| 69 | int16_t *block, int n, | ||
| 70 | int qscale, int *overflow) | ||
| 71 | { | ||
| 72 | x86_reg last_non_zero_p1; | ||
| 73 | 7290 | int level=0, q; //=0 is because gcc says uninitialized ... | |
| 74 | const uint16_t *qmat; | ||
| 75 | 7290 | LOCAL_ALIGNED_16(int16_t, temp_block, [64]); | |
| 76 | |||
| 77 | //s->fdct (block); | ||
| 78 | 7290 | ff_fdct_sse2(block); // cannot be anything else ... | |
| 79 | |||
| 80 | 7290 | if (s->dct_error_sum) { | |
| 81 | ✗ | const int intra = s->c.mb_intra; | |
| 82 | ✗ | s->dct_count[intra]++; | |
| 83 | #if HAVE_SSE2_EXTERNAL | ||
| 84 | ✗ | ff_mpv_denoise_dct_sse2(block, s->dct_error_sum[intra], s->dct_offset[intra]); | |
| 85 | #else | ||
| 86 | s->mpvencdsp.denoise_dct(block, s->dct_error_sum[intra], s->dct_offset[intra]); | ||
| 87 | #endif | ||
| 88 | } | ||
| 89 | |||
| 90 | 7290 | if (s->c.mb_intra) { | |
| 91 | int dummy; | ||
| 92 | 7290 | if (n < 4){ | |
| 93 | 4860 | q = s->c.y_dc_scale; | |
| 94 | 4860 | qmat = s->q_intra_matrix16[qscale][0]; | |
| 95 | }else{ | ||
| 96 | 2430 | q = s->c.c_dc_scale; | |
| 97 | 2430 | qmat = s->q_chroma_intra_matrix16[qscale][0]; | |
| 98 | } | ||
| 99 | /* note: block[0] is assumed to be positive */ | ||
| 100 | 7290 | if (!s->c.h263_aic) { | |
| 101 | 7290 | __asm__ volatile ( | |
| 102 | "mul %%ecx \n\t" | ||
| 103 | : "=d" (level), "=a"(dummy) | ||
| 104 | 7290 | : "a" ((block[0]>>2) + q), "c" (ff_inverse[q<<1]) | |
| 105 | ); | ||
| 106 | } else | ||
| 107 | /* For AIC we skip quant/dequant of INTRADC */ | ||
| 108 | ✗ | level = (block[0] + 4)>>3; | |
| 109 | |||
| 110 | 7290 | block[0]=0; //avoid fake overflow | |
| 111 | // temp_block[0] = (block[0] + (q >> 1)) / q; | ||
| 112 | 7290 | last_non_zero_p1 = 1; | |
| 113 | } else { | ||
| 114 | ✗ | last_non_zero_p1 = 0; | |
| 115 | ✗ | qmat = s->q_inter_matrix16[qscale][0]; | |
| 116 | } | ||
| 117 | |||
| 118 | 7290 | if ((s->c.out_format == FMT_H263 || s->c.out_format == FMT_H261) && !s->mpeg_quant) { | |
| 119 | 2970 | __asm__ volatile( | |
| 120 | "movd %%"FF_REG_a", %%xmm3 \n\t" // last_non_zero_p1 | ||
| 121 | SPREADW("%%xmm3") | ||
| 122 | "pxor %%xmm2, %%xmm2 \n\t" // 0 | ||
| 123 | "pxor %%xmm4, %%xmm4 \n\t" // 0 | ||
| 124 | "movdqa (%2), %%xmm5 \n\t" // qmat[0] | ||
| 125 | "pxor %%xmm6, %%xmm6 \n\t" | ||
| 126 | "psubw 128(%2), %%xmm6 \n\t" // -bias[0] | ||
| 127 | "mov $-128, %%"FF_REG_a" \n\t" | ||
| 128 | ".p2align 4 \n\t" | ||
| 129 | "1: \n\t" | ||
| 130 | "movdqa (%1, %%"FF_REG_a"), %%xmm0 \n\t" // block[i] | ||
| 131 | SAVE_SIGN("%%xmm1", "%%xmm0") // ABS(block[i]) | ||
| 132 | "psubusw %%xmm6, %%xmm0 \n\t" // ABS(block[i]) + bias[0] | ||
| 133 | "pmulhw %%xmm5, %%xmm0 \n\t" // (ABS(block[i])*qmat[0] - bias[0]*qmat[0])>>16 | ||
| 134 | "por %%xmm0, %%xmm4 \n\t" | ||
| 135 | RESTORE_SIGN("%%xmm1", "%%xmm0") // out=((ABS(block[i])*qmat[0] - bias[0]*qmat[0])>>16)*sign(block[i]) | ||
| 136 | "movdqa %%xmm0, (%4, %0) \n\t" | ||
| 137 | "pcmpeqw %%xmm2, %%xmm0 \n\t" // out==0 ? 0xFF : 0x00 | ||
| 138 | "movdqa (%3, %0), %%xmm1 \n\t" | ||
| 139 | "movdqa %%xmm2, (%1, %%"FF_REG_a") \n\t" // 0 | ||
| 140 | "pandn %%xmm1, %%xmm0 \n\t" | ||
| 141 | "pmaxsw %%xmm0, %%xmm3 \n\t" | ||
| 142 | "add $16, %%"FF_REG_a" \n\t" | ||
| 143 | " js 1b \n\t" | ||
| 144 | PMAX("%%xmm3", "%%xmm0") | ||
| 145 | "movd %%xmm3, %%"FF_REG_a" \n\t" | ||
| 146 | "movzbl %%al, %%eax \n\t" // last_non_zero_p1 | ||
| 147 | : "+a" (last_non_zero_p1) | ||
| 148 | 2970 | : "r" (block+64), "r" (qmat), | |
| 149 | 2970 | "r" (inv_zigzag_direct16 + 64), "r" (temp_block + 64) | |
| 150 | XMM_CLOBBERS_ONLY("%xmm0", "%xmm1", "%xmm2", "%xmm3", | ||
| 151 | "%xmm4", "%xmm5", "%xmm6") | ||
| 152 | ); | ||
| 153 | }else{ // FMT_H263 | ||
| 154 | 4320 | __asm__ volatile( | |
| 155 | "movd %%"FF_REG_a", %%xmm3 \n\t" // last_non_zero_p1 | ||
| 156 | SPREADW("%%xmm3") | ||
| 157 | "pxor %%xmm2, %%xmm2 \n\t" // 0 | ||
| 158 | "pxor %%xmm4, %%xmm4 \n\t" // 0 | ||
| 159 | "mov $-128, %%"FF_REG_a" \n\t" | ||
| 160 | ".p2align 4 \n\t" | ||
| 161 | "1: \n\t" | ||
| 162 | "movdqa (%1, %%"FF_REG_a"), %%xmm0 \n\t" // block[i] | ||
| 163 | SAVE_SIGN("%%xmm1", "%%xmm0") // ABS(block[i]) | ||
| 164 | "movdqa 128(%2, %0), %%xmm6 \n\t" // bias[i] | ||
| 165 | "paddusw %%xmm6, %%xmm0 \n\t" // ABS(block[i]) + bias[0] | ||
| 166 | "movdqa (%2, %%"FF_REG_a"), %%xmm5 \n\t" // qmat[i] | ||
| 167 | "pmulhw %%xmm5, %%xmm0 \n\t" // (ABS(block[i])*qmat[0] + bias[0]*qmat[0])>>16 | ||
| 168 | "por %%xmm0, %%xmm4 \n\t" | ||
| 169 | RESTORE_SIGN("%%xmm1", "%%xmm0") // out=((ABS(block[i])*qmat[0] - bias[0]*qmat[0])>>16)*sign(block[i]) | ||
| 170 | "movdqa %%xmm0, (%4, %0) \n\t" | ||
| 171 | "pcmpeqw %%xmm2, %%xmm0 \n\t" // out==0 ? 0xFF : 0x00 | ||
| 172 | "movdqa (%3, %0), %%xmm1 \n\t" | ||
| 173 | "movdqa %%xmm2, (%1, %%"FF_REG_a") \n\t" // 0 | ||
| 174 | "pandn %%xmm1, %%xmm0 \n\t" | ||
| 175 | "pmaxsw %%xmm0, %%xmm3 \n\t" | ||
| 176 | "add $16, %%"FF_REG_a" \n\t" | ||
| 177 | " js 1b \n\t" | ||
| 178 | PMAX("%%xmm3", "%%xmm0") | ||
| 179 | "movd %%xmm3, %%"FF_REG_a" \n\t" | ||
| 180 | "movzbl %%al, %%eax \n\t" // last_non_zero_p1 | ||
| 181 | : "+a" (last_non_zero_p1) | ||
| 182 | 4320 | : "r" (block+64), "r" (qmat+64), | |
| 183 | 4320 | "r" (inv_zigzag_direct16 + 64), "r" (temp_block + 64) | |
| 184 | XMM_CLOBBERS_ONLY("%xmm0", "%xmm1", "%xmm2", "%xmm3", | ||
| 185 | "%xmm4", "%xmm5", "%xmm6") | ||
| 186 | ); | ||
| 187 | } | ||
| 188 | 7290 | __asm__ volatile( | |
| 189 | "movd %1, %%xmm1 \n\t" // max_qcoeff | ||
| 190 | SPREADW("%%xmm1") | ||
| 191 | "psubusw %%xmm1, %%xmm4 \n\t" | ||
| 192 | "packuswb %%xmm4, %%xmm4 \n\t" | ||
| 193 | "packsswb %%xmm4, %%xmm4 \n\t" | ||
| 194 | "movd %%xmm4, %0 \n\t" // *overflow | ||
| 195 | : "=g" (*overflow) | ||
| 196 | 7290 | : "g" (s->max_qcoeff) | |
| 197 | ); | ||
| 198 | |||
| 199 | 7290 | block[0] = s->c.mb_intra ? level : temp_block[0]; | |
| 200 | |||
| 201 | av_assert2(ARCH_X86_32 || s->c.idsp.perm_type != FF_IDCT_PERM_SIMPLE); | ||
| 202 | if (ARCH_X86_32 && s->c.idsp.perm_type == FF_IDCT_PERM_SIMPLE) { | ||
| 203 | if(last_non_zero_p1 <= 1) goto end; | ||
| 204 | block[0x08] = temp_block[0x01]; block[0x10] = temp_block[0x08]; | ||
| 205 | block[0x20] = temp_block[0x10]; | ||
| 206 | if(last_non_zero_p1 <= 4) goto end; | ||
| 207 | block[0x18] = temp_block[0x09]; block[0x04] = temp_block[0x02]; | ||
| 208 | block[0x09] = temp_block[0x03]; | ||
| 209 | if(last_non_zero_p1 <= 7) goto end; | ||
| 210 | block[0x14] = temp_block[0x0A]; block[0x28] = temp_block[0x11]; | ||
| 211 | block[0x12] = temp_block[0x18]; block[0x02] = temp_block[0x20]; | ||
| 212 | if(last_non_zero_p1 <= 11) goto end; | ||
| 213 | block[0x1A] = temp_block[0x19]; block[0x24] = temp_block[0x12]; | ||
| 214 | block[0x19] = temp_block[0x0B]; block[0x01] = temp_block[0x04]; | ||
| 215 | block[0x0C] = temp_block[0x05]; | ||
| 216 | if(last_non_zero_p1 <= 16) goto end; | ||
| 217 | block[0x11] = temp_block[0x0C]; block[0x29] = temp_block[0x13]; | ||
| 218 | block[0x16] = temp_block[0x1A]; block[0x0A] = temp_block[0x21]; | ||
| 219 | block[0x30] = temp_block[0x28]; block[0x22] = temp_block[0x30]; | ||
| 220 | block[0x38] = temp_block[0x29]; block[0x06] = temp_block[0x22]; | ||
| 221 | if(last_non_zero_p1 <= 24) goto end; | ||
| 222 | block[0x1B] = temp_block[0x1B]; block[0x21] = temp_block[0x14]; | ||
| 223 | block[0x1C] = temp_block[0x0D]; block[0x05] = temp_block[0x06]; | ||
| 224 | block[0x0D] = temp_block[0x07]; block[0x15] = temp_block[0x0E]; | ||
| 225 | block[0x2C] = temp_block[0x15]; block[0x13] = temp_block[0x1C]; | ||
| 226 | if(last_non_zero_p1 <= 32) goto end; | ||
| 227 | block[0x0B] = temp_block[0x23]; block[0x34] = temp_block[0x2A]; | ||
| 228 | block[0x2A] = temp_block[0x31]; block[0x32] = temp_block[0x38]; | ||
| 229 | block[0x3A] = temp_block[0x39]; block[0x26] = temp_block[0x32]; | ||
| 230 | block[0x39] = temp_block[0x2B]; block[0x03] = temp_block[0x24]; | ||
| 231 | if(last_non_zero_p1 <= 40) goto end; | ||
| 232 | block[0x1E] = temp_block[0x1D]; block[0x25] = temp_block[0x16]; | ||
| 233 | block[0x1D] = temp_block[0x0F]; block[0x2D] = temp_block[0x17]; | ||
| 234 | block[0x17] = temp_block[0x1E]; block[0x0E] = temp_block[0x25]; | ||
| 235 | block[0x31] = temp_block[0x2C]; block[0x2B] = temp_block[0x33]; | ||
| 236 | if(last_non_zero_p1 <= 48) goto end; | ||
| 237 | block[0x36] = temp_block[0x3A]; block[0x3B] = temp_block[0x3B]; | ||
| 238 | block[0x23] = temp_block[0x34]; block[0x3C] = temp_block[0x2D]; | ||
| 239 | block[0x07] = temp_block[0x26]; block[0x1F] = temp_block[0x1F]; | ||
| 240 | block[0x0F] = temp_block[0x27]; block[0x35] = temp_block[0x2E]; | ||
| 241 | if(last_non_zero_p1 <= 56) goto end; | ||
| 242 | block[0x2E] = temp_block[0x35]; block[0x33] = temp_block[0x3C]; | ||
| 243 | block[0x3E] = temp_block[0x3D]; block[0x27] = temp_block[0x36]; | ||
| 244 | block[0x3D] = temp_block[0x2F]; block[0x2F] = temp_block[0x37]; | ||
| 245 | block[0x37] = temp_block[0x3E]; block[0x3F] = temp_block[0x3F]; | ||
| 246 | 7290 | } else if (s->c.idsp.perm_type == FF_IDCT_PERM_LIBMPEG2) { | |
| 247 | ✗ | if(last_non_zero_p1 <= 1) goto end; | |
| 248 | ✗ | block[0x04] = temp_block[0x01]; | |
| 249 | ✗ | block[0x08] = temp_block[0x08]; block[0x10] = temp_block[0x10]; | |
| 250 | ✗ | if(last_non_zero_p1 <= 4) goto end; | |
| 251 | ✗ | block[0x0C] = temp_block[0x09]; block[0x01] = temp_block[0x02]; | |
| 252 | ✗ | block[0x05] = temp_block[0x03]; | |
| 253 | ✗ | if(last_non_zero_p1 <= 7) goto end; | |
| 254 | ✗ | block[0x09] = temp_block[0x0A]; block[0x14] = temp_block[0x11]; | |
| 255 | ✗ | block[0x18] = temp_block[0x18]; block[0x20] = temp_block[0x20]; | |
| 256 | ✗ | if(last_non_zero_p1 <= 11) goto end; | |
| 257 | ✗ | block[0x1C] = temp_block[0x19]; | |
| 258 | ✗ | block[0x11] = temp_block[0x12]; block[0x0D] = temp_block[0x0B]; | |
| 259 | ✗ | block[0x02] = temp_block[0x04]; block[0x06] = temp_block[0x05]; | |
| 260 | ✗ | if(last_non_zero_p1 <= 16) goto end; | |
| 261 | ✗ | block[0x0A] = temp_block[0x0C]; block[0x15] = temp_block[0x13]; | |
| 262 | ✗ | block[0x19] = temp_block[0x1A]; block[0x24] = temp_block[0x21]; | |
| 263 | ✗ | block[0x28] = temp_block[0x28]; block[0x30] = temp_block[0x30]; | |
| 264 | ✗ | block[0x2C] = temp_block[0x29]; block[0x21] = temp_block[0x22]; | |
| 265 | ✗ | if(last_non_zero_p1 <= 24) goto end; | |
| 266 | ✗ | block[0x1D] = temp_block[0x1B]; block[0x12] = temp_block[0x14]; | |
| 267 | ✗ | block[0x0E] = temp_block[0x0D]; block[0x03] = temp_block[0x06]; | |
| 268 | ✗ | block[0x07] = temp_block[0x07]; block[0x0B] = temp_block[0x0E]; | |
| 269 | ✗ | block[0x16] = temp_block[0x15]; block[0x1A] = temp_block[0x1C]; | |
| 270 | ✗ | if(last_non_zero_p1 <= 32) goto end; | |
| 271 | ✗ | block[0x25] = temp_block[0x23]; block[0x29] = temp_block[0x2A]; | |
| 272 | ✗ | block[0x34] = temp_block[0x31]; block[0x38] = temp_block[0x38]; | |
| 273 | ✗ | block[0x3C] = temp_block[0x39]; block[0x31] = temp_block[0x32]; | |
| 274 | ✗ | block[0x2D] = temp_block[0x2B]; block[0x22] = temp_block[0x24]; | |
| 275 | ✗ | if(last_non_zero_p1 <= 40) goto end; | |
| 276 | ✗ | block[0x1E] = temp_block[0x1D]; block[0x13] = temp_block[0x16]; | |
| 277 | ✗ | block[0x0F] = temp_block[0x0F]; block[0x17] = temp_block[0x17]; | |
| 278 | ✗ | block[0x1B] = temp_block[0x1E]; block[0x26] = temp_block[0x25]; | |
| 279 | ✗ | block[0x2A] = temp_block[0x2C]; block[0x35] = temp_block[0x33]; | |
| 280 | ✗ | if(last_non_zero_p1 <= 48) goto end; | |
| 281 | ✗ | block[0x39] = temp_block[0x3A]; block[0x3D] = temp_block[0x3B]; | |
| 282 | ✗ | block[0x32] = temp_block[0x34]; block[0x2E] = temp_block[0x2D]; | |
| 283 | ✗ | block[0x23] = temp_block[0x26]; block[0x1F] = temp_block[0x1F]; | |
| 284 | ✗ | block[0x27] = temp_block[0x27]; block[0x2B] = temp_block[0x2E]; | |
| 285 | ✗ | if(last_non_zero_p1 <= 56) goto end; | |
| 286 | ✗ | block[0x36] = temp_block[0x35]; block[0x3A] = temp_block[0x3C]; | |
| 287 | ✗ | block[0x3E] = temp_block[0x3D]; block[0x33] = temp_block[0x36]; | |
| 288 | ✗ | block[0x2F] = temp_block[0x2F]; block[0x37] = temp_block[0x37]; | |
| 289 | ✗ | block[0x3B] = temp_block[0x3E]; block[0x3F] = temp_block[0x3F]; | |
| 290 | 7290 | } else if (s->c.idsp.perm_type == FF_IDCT_PERM_NONE) { | |
| 291 | ✗ | if(last_non_zero_p1 <= 1) goto end; | |
| 292 | ✗ | block[0x01] = temp_block[0x01]; | |
| 293 | ✗ | block[0x08] = temp_block[0x08]; block[0x10] = temp_block[0x10]; | |
| 294 | ✗ | if(last_non_zero_p1 <= 4) goto end; | |
| 295 | ✗ | block[0x09] = temp_block[0x09]; block[0x02] = temp_block[0x02]; | |
| 296 | ✗ | block[0x03] = temp_block[0x03]; | |
| 297 | ✗ | if(last_non_zero_p1 <= 7) goto end; | |
| 298 | ✗ | block[0x0A] = temp_block[0x0A]; block[0x11] = temp_block[0x11]; | |
| 299 | ✗ | block[0x18] = temp_block[0x18]; block[0x20] = temp_block[0x20]; | |
| 300 | ✗ | if(last_non_zero_p1 <= 11) goto end; | |
| 301 | ✗ | block[0x19] = temp_block[0x19]; | |
| 302 | ✗ | block[0x12] = temp_block[0x12]; block[0x0B] = temp_block[0x0B]; | |
| 303 | ✗ | block[0x04] = temp_block[0x04]; block[0x05] = temp_block[0x05]; | |
| 304 | ✗ | if(last_non_zero_p1 <= 16) goto end; | |
| 305 | ✗ | block[0x0C] = temp_block[0x0C]; block[0x13] = temp_block[0x13]; | |
| 306 | ✗ | block[0x1A] = temp_block[0x1A]; block[0x21] = temp_block[0x21]; | |
| 307 | ✗ | block[0x28] = temp_block[0x28]; block[0x30] = temp_block[0x30]; | |
| 308 | ✗ | block[0x29] = temp_block[0x29]; block[0x22] = temp_block[0x22]; | |
| 309 | ✗ | if(last_non_zero_p1 <= 24) goto end; | |
| 310 | ✗ | block[0x1B] = temp_block[0x1B]; block[0x14] = temp_block[0x14]; | |
| 311 | ✗ | block[0x0D] = temp_block[0x0D]; block[0x06] = temp_block[0x06]; | |
| 312 | ✗ | block[0x07] = temp_block[0x07]; block[0x0E] = temp_block[0x0E]; | |
| 313 | ✗ | block[0x15] = temp_block[0x15]; block[0x1C] = temp_block[0x1C]; | |
| 314 | ✗ | if(last_non_zero_p1 <= 32) goto end; | |
| 315 | ✗ | block[0x23] = temp_block[0x23]; block[0x2A] = temp_block[0x2A]; | |
| 316 | ✗ | block[0x31] = temp_block[0x31]; block[0x38] = temp_block[0x38]; | |
| 317 | ✗ | block[0x39] = temp_block[0x39]; block[0x32] = temp_block[0x32]; | |
| 318 | ✗ | block[0x2B] = temp_block[0x2B]; block[0x24] = temp_block[0x24]; | |
| 319 | ✗ | if(last_non_zero_p1 <= 40) goto end; | |
| 320 | ✗ | block[0x1D] = temp_block[0x1D]; block[0x16] = temp_block[0x16]; | |
| 321 | ✗ | block[0x0F] = temp_block[0x0F]; block[0x17] = temp_block[0x17]; | |
| 322 | ✗ | block[0x1E] = temp_block[0x1E]; block[0x25] = temp_block[0x25]; | |
| 323 | ✗ | block[0x2C] = temp_block[0x2C]; block[0x33] = temp_block[0x33]; | |
| 324 | ✗ | if(last_non_zero_p1 <= 48) goto end; | |
| 325 | ✗ | block[0x3A] = temp_block[0x3A]; block[0x3B] = temp_block[0x3B]; | |
| 326 | ✗ | block[0x34] = temp_block[0x34]; block[0x2D] = temp_block[0x2D]; | |
| 327 | ✗ | block[0x26] = temp_block[0x26]; block[0x1F] = temp_block[0x1F]; | |
| 328 | ✗ | block[0x27] = temp_block[0x27]; block[0x2E] = temp_block[0x2E]; | |
| 329 | ✗ | if(last_non_zero_p1 <= 56) goto end; | |
| 330 | ✗ | block[0x35] = temp_block[0x35]; block[0x3C] = temp_block[0x3C]; | |
| 331 | ✗ | block[0x3D] = temp_block[0x3D]; block[0x36] = temp_block[0x36]; | |
| 332 | ✗ | block[0x2F] = temp_block[0x2F]; block[0x37] = temp_block[0x37]; | |
| 333 | ✗ | block[0x3E] = temp_block[0x3E]; block[0x3F] = temp_block[0x3F]; | |
| 334 | 7290 | } else if (s->c.idsp.perm_type == FF_IDCT_PERM_TRANSPOSE) { | |
| 335 | 7290 | if(last_non_zero_p1 <= 1) goto end; | |
| 336 | 3108 | block[0x08] = temp_block[0x01]; | |
| 337 | 3108 | block[0x01] = temp_block[0x08]; block[0x02] = temp_block[0x10]; | |
| 338 | 3108 | if(last_non_zero_p1 <= 4) goto end; | |
| 339 | 3018 | block[0x09] = temp_block[0x09]; block[0x10] = temp_block[0x02]; | |
| 340 | 3018 | block[0x18] = temp_block[0x03]; | |
| 341 | 3018 | if(last_non_zero_p1 <= 7) goto end; | |
| 342 | 3012 | block[0x11] = temp_block[0x0A]; block[0x0A] = temp_block[0x11]; | |
| 343 | 3012 | block[0x03] = temp_block[0x18]; block[0x04] = temp_block[0x20]; | |
| 344 | 3012 | if(last_non_zero_p1 <= 11) goto end; | |
| 345 | 2994 | block[0x0B] = temp_block[0x19]; | |
| 346 | 2994 | block[0x12] = temp_block[0x12]; block[0x19] = temp_block[0x0B]; | |
| 347 | 2994 | block[0x20] = temp_block[0x04]; block[0x28] = temp_block[0x05]; | |
| 348 | 2994 | if(last_non_zero_p1 <= 16) goto end; | |
| 349 | 2976 | block[0x21] = temp_block[0x0C]; block[0x1A] = temp_block[0x13]; | |
| 350 | 2976 | block[0x13] = temp_block[0x1A]; block[0x0C] = temp_block[0x21]; | |
| 351 | 2976 | block[0x05] = temp_block[0x28]; block[0x06] = temp_block[0x30]; | |
| 352 | 2976 | block[0x0D] = temp_block[0x29]; block[0x14] = temp_block[0x22]; | |
| 353 | 2976 | if(last_non_zero_p1 <= 24) goto end; | |
| 354 | 2958 | block[0x1B] = temp_block[0x1B]; block[0x22] = temp_block[0x14]; | |
| 355 | 2958 | block[0x29] = temp_block[0x0D]; block[0x30] = temp_block[0x06]; | |
| 356 | 2958 | block[0x38] = temp_block[0x07]; block[0x31] = temp_block[0x0E]; | |
| 357 | 2958 | block[0x2A] = temp_block[0x15]; block[0x23] = temp_block[0x1C]; | |
| 358 | 2958 | if(last_non_zero_p1 <= 32) goto end; | |
| 359 | 2232 | block[0x1C] = temp_block[0x23]; block[0x15] = temp_block[0x2A]; | |
| 360 | 2232 | block[0x0E] = temp_block[0x31]; block[0x07] = temp_block[0x38]; | |
| 361 | 2232 | block[0x0F] = temp_block[0x39]; block[0x16] = temp_block[0x32]; | |
| 362 | 2232 | block[0x1D] = temp_block[0x2B]; block[0x24] = temp_block[0x24]; | |
| 363 | 2232 | if(last_non_zero_p1 <= 40) goto end; | |
| 364 | 1848 | block[0x2B] = temp_block[0x1D]; block[0x32] = temp_block[0x16]; | |
| 365 | 1848 | block[0x39] = temp_block[0x0F]; block[0x3A] = temp_block[0x17]; | |
| 366 | 1848 | block[0x33] = temp_block[0x1E]; block[0x2C] = temp_block[0x25]; | |
| 367 | 1848 | block[0x25] = temp_block[0x2C]; block[0x1E] = temp_block[0x33]; | |
| 368 | 1848 | if(last_non_zero_p1 <= 48) goto end; | |
| 369 | 1584 | block[0x17] = temp_block[0x3A]; block[0x1F] = temp_block[0x3B]; | |
| 370 | 1584 | block[0x26] = temp_block[0x34]; block[0x2D] = temp_block[0x2D]; | |
| 371 | 1584 | block[0x34] = temp_block[0x26]; block[0x3B] = temp_block[0x1F]; | |
| 372 | 1584 | block[0x3C] = temp_block[0x27]; block[0x35] = temp_block[0x2E]; | |
| 373 | 1584 | if(last_non_zero_p1 <= 56) goto end; | |
| 374 | 1242 | block[0x2E] = temp_block[0x35]; block[0x27] = temp_block[0x3C]; | |
| 375 | 1242 | block[0x2F] = temp_block[0x3D]; block[0x36] = temp_block[0x36]; | |
| 376 | 1242 | block[0x3D] = temp_block[0x2F]; block[0x3E] = temp_block[0x37]; | |
| 377 | 1242 | block[0x37] = temp_block[0x3E]; block[0x3F] = temp_block[0x3F]; | |
| 378 | } else { | ||
| 379 | ✗ | av_log(s->c.avctx, AV_LOG_DEBUG, "s->c.idsp.perm_type: %d\n", | |
| 380 | ✗ | (int)s->c.idsp.perm_type); | |
| 381 | ✗ | av_assert0(s->c.idsp.perm_type == FF_IDCT_PERM_NONE || | |
| 382 | s->c.idsp.perm_type == FF_IDCT_PERM_LIBMPEG2 || | ||
| 383 | s->c.idsp.perm_type == FF_IDCT_PERM_SIMPLE || | ||
| 384 | s->c.idsp.perm_type == FF_IDCT_PERM_TRANSPOSE); | ||
| 385 | } | ||
| 386 | ✗ | end: | |
| 387 | 7290 | return last_non_zero_p1 - 1; | |
| 388 | } | ||
| 389 |