| Line | Branch | Exec | Source |
|---|---|---|---|
| 1 | /* | ||
| 2 | * Copyright (C) 2007 Marco Gerards <marco@gnu.org> | ||
| 3 | * Copyright (C) 2009 David Conrad | ||
| 4 | * | ||
| 5 | * This file is part of FFmpeg. | ||
| 6 | * | ||
| 7 | * FFmpeg is free software; you can redistribute it and/or | ||
| 8 | * modify it under the terms of the GNU Lesser General Public | ||
| 9 | * License as published by the Free Software Foundation; either | ||
| 10 | * version 2.1 of the License, or (at your option) any later version. | ||
| 11 | * | ||
| 12 | * FFmpeg is distributed in the hope that it will be useful, | ||
| 13 | * but WITHOUT ANY WARRANTY; without even the implied warranty of | ||
| 14 | * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU | ||
| 15 | * Lesser General Public License for more details. | ||
| 16 | * | ||
| 17 | * You should have received a copy of the GNU Lesser General Public | ||
| 18 | * License along with FFmpeg; if not, write to the Free Software | ||
| 19 | * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA | ||
| 20 | */ | ||
| 21 | |||
| 22 | /** | ||
| 23 | * @file | ||
| 24 | * Arithmetic decoder for Dirac | ||
| 25 | * @author Marco Gerards <marco@gnu.org> | ||
| 26 | */ | ||
| 27 | |||
| 28 | #ifndef AVCODEC_DIRAC_ARITH_H | ||
| 29 | #define AVCODEC_DIRAC_ARITH_H | ||
| 30 | |||
| 31 | #include "config.h" | ||
| 32 | |||
| 33 | #if ARCH_X86 | ||
| 34 | #include "libavutil/x86/asm.h" | ||
| 35 | #endif | ||
| 36 | #include "bytestream.h" | ||
| 37 | #include "get_bits.h" | ||
| 38 | |||
| 39 | enum dirac_arith_contexts { | ||
| 40 | CTX_ZPZN_F1, | ||
| 41 | CTX_ZPNN_F1, | ||
| 42 | CTX_NPZN_F1, | ||
| 43 | CTX_NPNN_F1, | ||
| 44 | CTX_ZP_F2, | ||
| 45 | CTX_ZP_F3, | ||
| 46 | CTX_ZP_F4, | ||
| 47 | CTX_ZP_F5, | ||
| 48 | CTX_ZP_F6, | ||
| 49 | CTX_NP_F2, | ||
| 50 | CTX_NP_F3, | ||
| 51 | CTX_NP_F4, | ||
| 52 | CTX_NP_F5, | ||
| 53 | CTX_NP_F6, | ||
| 54 | CTX_COEFF_DATA, | ||
| 55 | CTX_SIGN_NEG, | ||
| 56 | CTX_SIGN_ZERO, | ||
| 57 | CTX_SIGN_POS, | ||
| 58 | CTX_ZERO_BLOCK, | ||
| 59 | CTX_DELTA_Q_F, | ||
| 60 | CTX_DELTA_Q_DATA, | ||
| 61 | CTX_DELTA_Q_SIGN, | ||
| 62 | |||
| 63 | DIRAC_CTX_COUNT | ||
| 64 | }; | ||
| 65 | |||
| 66 | // Dirac resets the arith decoder between decoding various types of data, | ||
| 67 | // so many contexts are never used simultaneously. Thus, we can reduce | ||
| 68 | // the number of contexts needed by reusing them. | ||
| 69 | #define CTX_SB_F1 CTX_ZP_F5 | ||
| 70 | #define CTX_SB_DATA 0 | ||
| 71 | #define CTX_PMODE_REF1 0 | ||
| 72 | #define CTX_PMODE_REF2 1 | ||
| 73 | #define CTX_GLOBAL_BLOCK 2 | ||
| 74 | #define CTX_MV_F1 CTX_ZP_F2 | ||
| 75 | #define CTX_MV_DATA 0 | ||
| 76 | #define CTX_DC_F1 CTX_ZP_F5 | ||
| 77 | #define CTX_DC_DATA 0 | ||
| 78 | |||
| 79 | typedef struct { | ||
| 80 | unsigned low; | ||
| 81 | uint16_t range; | ||
| 82 | int16_t counter; | ||
| 83 | |||
| 84 | const uint8_t *bytestream; | ||
| 85 | const uint8_t *bytestream_end; | ||
| 86 | |||
| 87 | uint16_t contexts[DIRAC_CTX_COUNT]; | ||
| 88 | int error; | ||
| 89 | int overread; | ||
| 90 | } DiracArith; | ||
| 91 | |||
| 92 | extern const uint8_t ff_dirac_next_ctx[DIRAC_CTX_COUNT]; | ||
| 93 | extern int16_t ff_dirac_prob_branchless[256][2]; | ||
| 94 | |||
| 95 | 1024041 | static inline void renorm(DiracArith *c) | |
| 96 | { | ||
| 97 | #if HAVE_FAST_CLZ | ||
| 98 | 1024041 | int shift = 14 - av_log2_16bit(c->range-1) + ((c->range-1)>>15); | |
| 99 | |||
| 100 | 1024041 | c->low <<= shift; | |
| 101 | 1024041 | c->range <<= shift; | |
| 102 | 1024041 | c->counter += shift; | |
| 103 | #else | ||
| 104 | while (c->range <= 0x4000) { | ||
| 105 | c->low <<= 1; | ||
| 106 | c->range <<= 1; | ||
| 107 | c->counter++; | ||
| 108 | } | ||
| 109 | #endif | ||
| 110 | 1024041 | } | |
| 111 | |||
| 112 | 1024041 | static inline void refill(DiracArith *c) | |
| 113 | { | ||
| 114 | 1024041 | int counter = c->counter; | |
| 115 | |||
| 116 |
2/2✓ Branch 0 taken 46510 times.
✓ Branch 1 taken 977531 times.
|
1024041 | if (counter >= 0) { |
| 117 | 46510 | int new = bytestream_get_be16(&c->bytestream); | |
| 118 | |||
| 119 | // the spec defines overread bits to be 1, and streams rely on this | ||
| 120 |
2/2✓ Branch 0 taken 495 times.
✓ Branch 1 taken 46015 times.
|
46510 | if (c->bytestream > c->bytestream_end) { |
| 121 | 495 | new |= 0xff; | |
| 122 |
2/2✓ Branch 0 taken 91 times.
✓ Branch 1 taken 404 times.
|
495 | if (c->bytestream > c->bytestream_end+1) |
| 123 | 91 | new |= 0xff00; | |
| 124 | |||
| 125 | 495 | c->bytestream = c->bytestream_end; | |
| 126 | 495 | c->overread ++; | |
| 127 |
1/2✗ Branch 0 not taken.
✓ Branch 1 taken 495 times.
|
495 | if (c->overread > 4) |
| 128 | ✗ | c->error = AVERROR_INVALIDDATA; | |
| 129 | } | ||
| 130 | |||
| 131 | 46510 | c->low += new << counter; | |
| 132 | 46510 | counter -= 16; | |
| 133 | } | ||
| 134 | 1024041 | c->counter = counter; | |
| 135 | 1024041 | } | |
| 136 | |||
| 137 | 1024041 | static inline int dirac_get_arith_bit(DiracArith *c, int ctx) | |
| 138 | { | ||
| 139 | 1024041 | int prob_zero = c->contexts[ctx]; | |
| 140 | int range_times_prob, bit; | ||
| 141 | 1024041 | unsigned low = c->low; | |
| 142 | 1024041 | int range = c->range; | |
| 143 | |||
| 144 | 1024041 | range_times_prob = (c->range * prob_zero) >> 16; | |
| 145 | |||
| 146 | #if ARCH_X86 && HAVE_FAST_CMOV && HAVE_INLINE_ASM && HAVE_6REGS | ||
| 147 | 1024041 | low -= range_times_prob << 16; | |
| 148 | 1024041 | range -= range_times_prob; | |
| 149 | 1024041 | bit = 0; | |
| 150 | 1024041 | __asm__( | |
| 151 | "cmpl %5, %4 \n\t" | ||
| 152 | "setae %b0 \n\t" | ||
| 153 | "cmovb %3, %2 \n\t" | ||
| 154 | "cmovb %5, %1 \n\t" | ||
| 155 | : "+q"(bit), "+r"(range), "+r"(low) | ||
| 156 | 1024041 | : "r"(c->low), "r"(c->low>>16), | |
| 157 | "r"(range_times_prob) | ||
| 158 | ); | ||
| 159 | #else | ||
| 160 | bit = (low >> 16) >= range_times_prob; | ||
| 161 | if (bit) { | ||
| 162 | low -= range_times_prob << 16; | ||
| 163 | range -= range_times_prob; | ||
| 164 | } else { | ||
| 165 | range = range_times_prob; | ||
| 166 | } | ||
| 167 | #endif | ||
| 168 | |||
| 169 | 1024041 | c->contexts[ctx] += ff_dirac_prob_branchless[prob_zero>>8][bit]; | |
| 170 | 1024041 | c->low = low; | |
| 171 | 1024041 | c->range = range; | |
| 172 | |||
| 173 | 1024041 | renorm(c); | |
| 174 | 1024041 | refill(c); | |
| 175 | 1024041 | return bit; | |
| 176 | } | ||
| 177 | |||
| 178 | 353584 | static inline int dirac_get_arith_uint(DiracArith *c, int follow_ctx, int data_ctx) | |
| 179 | { | ||
| 180 | 353584 | int ret = 1; | |
| 181 |
2/2✓ Branch 1 taken 219922 times.
✓ Branch 2 taken 353584 times.
|
573506 | while (!dirac_get_arith_bit(c, follow_ctx)) { |
| 182 |
1/2✗ Branch 0 not taken.
✓ Branch 1 taken 219922 times.
|
219922 | if (ret >= 0x40000000) { |
| 183 | ✗ | av_log(NULL, AV_LOG_ERROR, "dirac_get_arith_uint overflow\n"); | |
| 184 | ✗ | c->error = AVERROR_INVALIDDATA; | |
| 185 | ✗ | return -1; | |
| 186 | } | ||
| 187 | 219922 | ret <<= 1; | |
| 188 | 219922 | ret += dirac_get_arith_bit(c, data_ctx); | |
| 189 | 219922 | follow_ctx = ff_dirac_next_ctx[follow_ctx]; | |
| 190 | } | ||
| 191 | 353584 | return ret-1; | |
| 192 | } | ||
| 193 | |||
| 194 | 14164 | static inline int dirac_get_arith_int(DiracArith *c, int follow_ctx, int data_ctx) | |
| 195 | { | ||
| 196 | 14164 | int ret = dirac_get_arith_uint(c, follow_ctx, data_ctx); | |
| 197 |
4/4✓ Branch 0 taken 2620 times.
✓ Branch 1 taken 11544 times.
✓ Branch 3 taken 1346 times.
✓ Branch 4 taken 1274 times.
|
14164 | if (ret && dirac_get_arith_bit(c, data_ctx+1)) |
| 198 | 1346 | ret = -ret; | |
| 199 | 14164 | return ret; | |
| 200 | } | ||
| 201 | |||
| 202 | void ff_dirac_init_arith_tables(void); | ||
| 203 | void ff_dirac_init_arith_decoder(DiracArith *c, GetBitContext *gb, int length); | ||
| 204 | |||
| 205 | #endif /* AVCODEC_DIRAC_ARITH_H */ | ||
| 206 |