yading@10: /*
yading@10:  * Copyright (C) 2007 Marco Gerards <marco@gnu.org>
yading@10:  * Copyright (C) 2009 David Conrad
yading@10:  *
yading@10:  * This file is part of FFmpeg.
yading@10:  *
yading@10:  * FFmpeg is free software; you can redistribute it and/or
yading@10:  * modify it under the terms of the GNU Lesser General Public
yading@10:  * License as published by the Free Software Foundation; either
yading@10:  * version 2.1 of the License, or (at your option) any later version.
yading@10:  *
yading@10:  * FFmpeg is distributed in the hope that it will be useful,
yading@10:  * but WITHOUT ANY WARRANTY; without even the implied warranty of
yading@10:  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
yading@10:  * Lesser General Public License for more details.
yading@10:  *
yading@10:  * You should have received a copy of the GNU Lesser General Public
yading@10:  * License along with FFmpeg; if not, write to the Free Software
yading@10:  * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
yading@10:  */
yading@10: 
yading@10: /**
yading@10:  * @file
yading@10:  * Arithmetic decoder for Dirac
yading@10:  * @author Marco Gerards <marco@gnu.org>
yading@10:  */
yading@10: 
yading@10: #ifndef AVCODEC_DIRAC_ARITH_H
yading@10: #define AVCODEC_DIRAC_ARITH_H
yading@10: 
yading@10: #include "bytestream.h"
yading@10: #include "get_bits.h"
yading@10: 
yading@10: enum dirac_arith_contexts {
yading@10:     CTX_ZPZN_F1,
yading@10:     CTX_ZPNN_F1,
yading@10:     CTX_NPZN_F1,
yading@10:     CTX_NPNN_F1,
yading@10:     CTX_ZP_F2,
yading@10:     CTX_ZP_F3,
yading@10:     CTX_ZP_F4,
yading@10:     CTX_ZP_F5,
yading@10:     CTX_ZP_F6,
yading@10:     CTX_NP_F2,
yading@10:     CTX_NP_F3,
yading@10:     CTX_NP_F4,
yading@10:     CTX_NP_F5,
yading@10:     CTX_NP_F6,
yading@10:     CTX_COEFF_DATA,
yading@10:     CTX_SIGN_NEG,
yading@10:     CTX_SIGN_ZERO,
yading@10:     CTX_SIGN_POS,
yading@10:     CTX_ZERO_BLOCK,
yading@10:     CTX_DELTA_Q_F,
yading@10:     CTX_DELTA_Q_DATA,
yading@10:     CTX_DELTA_Q_SIGN,
yading@10: 
yading@10:     DIRAC_CTX_COUNT
yading@10: };
yading@10: 
yading@10: // Dirac resets the arith decoder between decoding various types of data,
yading@10: // so many contexts are never used simultaneously. Thus, we can reduce
yading@10: // the number of contexts needed by reusing them.
yading@10: #define CTX_SB_F1        CTX_ZP_F5
yading@10: #define CTX_SB_DATA      0
yading@10: #define CTX_PMODE_REF1   0
yading@10: #define CTX_PMODE_REF2   1
yading@10: #define CTX_GLOBAL_BLOCK 2
yading@10: #define CTX_MV_F1        CTX_ZP_F2
yading@10: #define CTX_MV_DATA      0
yading@10: #define CTX_DC_F1        CTX_ZP_F5
yading@10: #define CTX_DC_DATA      0
yading@10: 
yading@10: typedef struct {
yading@10:     unsigned low;
yading@10:     uint16_t range;
yading@10:     int16_t  counter;
yading@10: 
yading@10:     const uint8_t *bytestream;
yading@10:     const uint8_t *bytestream_end;
yading@10: 
yading@10:     uint16_t contexts[DIRAC_CTX_COUNT];
yading@10: } DiracArith;
yading@10: 
yading@10: extern const uint8_t ff_dirac_next_ctx[DIRAC_CTX_COUNT];
yading@10: extern const uint16_t ff_dirac_prob[256];
yading@10: extern int16_t ff_dirac_prob_branchless[256][2];
yading@10: 
yading@10: static inline void renorm(DiracArith *c)
yading@10: {
yading@10: #if HAVE_FAST_CLZ
yading@10:     int shift = 14 - av_log2_16bit(c->range-1) + ((c->range-1)>>15);
yading@10: 
yading@10:     c->low    <<= shift;
yading@10:     c->range  <<= shift;
yading@10:     c->counter += shift;
yading@10: #else
yading@10:     while (c->range <= 0x4000) {
yading@10:         c->low   <<= 1;
yading@10:         c->range <<= 1;
yading@10:         c->counter++;
yading@10:     }
yading@10: #endif
yading@10: }
yading@10: 
yading@10: static inline void refill(DiracArith *c)
yading@10: {
yading@10:     int counter = c->counter;
yading@10: 
yading@10:     if (counter >= 0) {
yading@10:         int new = bytestream_get_be16(&c->bytestream);
yading@10: 
yading@10:         // the spec defines overread bits to be 1, and streams rely on this
yading@10:         if (c->bytestream > c->bytestream_end) {
yading@10:             new |= 0xff;
yading@10:             if (c->bytestream > c->bytestream_end+1)
yading@10:                 new |= 0xff00;
yading@10: 
yading@10:             c->bytestream = c->bytestream_end;
yading@10:         }
yading@10: 
yading@10:         c->low += new << counter;
yading@10:         counter -= 16;
yading@10:     }
yading@10:     c->counter = counter;
yading@10: }
yading@10: 
yading@10: static inline int dirac_get_arith_bit(DiracArith *c, int ctx)
yading@10: {
yading@10:     int prob_zero = c->contexts[ctx];
yading@10:     int range_times_prob, bit;
yading@10:     unsigned low = c->low;
yading@10:     int    range = c->range;
yading@10: 
yading@10:     range_times_prob = (c->range * prob_zero) >> 16;
yading@10: 
yading@10: #if HAVE_FAST_CMOV && HAVE_INLINE_ASM
yading@10:     low   -= range_times_prob << 16;
yading@10:     range -= range_times_prob;
yading@10:     bit = 0;
yading@10:     __asm__(
yading@10:         "cmpl   %5, %4 \n\t"
yading@10:         "setae  %b0    \n\t"
yading@10:         "cmovb  %3, %2 \n\t"
yading@10:         "cmovb  %5, %1 \n\t"
yading@10:         : "+q"(bit), "+r"(range), "+r"(low)
yading@10:         : "r"(c->low), "r"(c->low>>16),
yading@10:           "r"(range_times_prob)
yading@10:     );
yading@10: #else
yading@10:     bit = (low >> 16) >= range_times_prob;
yading@10:     if (bit) {
yading@10:         low   -= range_times_prob << 16;
yading@10:         range -= range_times_prob;
yading@10:     } else {
yading@10:         range  = range_times_prob;
yading@10:     }
yading@10: #endif
yading@10: 
yading@10:     c->contexts[ctx] += ff_dirac_prob_branchless[prob_zero>>8][bit];
yading@10:     c->low   = low;
yading@10:     c->range = range;
yading@10: 
yading@10:     renorm(c);
yading@10:     refill(c);
yading@10:     return bit;
yading@10: }
yading@10: 
yading@10: static inline int dirac_get_arith_uint(DiracArith *c, int follow_ctx, int data_ctx)
yading@10: {
yading@10:     int ret = 1;
yading@10:     while (!dirac_get_arith_bit(c, follow_ctx)) {
yading@10:         ret <<= 1;
yading@10:         ret += dirac_get_arith_bit(c, data_ctx);
yading@10:         follow_ctx = ff_dirac_next_ctx[follow_ctx];
yading@10:     }
yading@10:     return ret-1;
yading@10: }
yading@10: 
yading@10: static inline int dirac_get_arith_int(DiracArith *c, int follow_ctx, int data_ctx)
yading@10: {
yading@10:     int ret = dirac_get_arith_uint(c, follow_ctx, data_ctx);
yading@10:     if (ret && dirac_get_arith_bit(c, data_ctx+1))
yading@10:         ret = -ret;
yading@10:     return ret;
yading@10: }
yading@10: 
yading@10: void ff_dirac_init_arith_decoder(DiracArith *c, GetBitContext *gb, int length);
yading@10: 
yading@10: #endif /* AVCODEC_DIRAC_ARITH_H */