1// Copyright 2014 Google Inc. All Rights Reserved.
2//
3// Use of this source code is governed by a BSD-style license
4// that can be found in the COPYING file in the root of the source
5// tree. An additional intellectual property rights grant can be found
6// in the file PATENTS. All contributing project authors may
7// be found in the AUTHORS file in the root of the source tree.
8// -----------------------------------------------------------------------------
9//
10// Specific inlined methods for boolean decoder [VP8GetBit() ...]
11// This file should be included by the .c sources that actually need to call
12// these methods.
13//
14// Author: Skal (pascal.massimino@gmail.com)
15
16#ifndef WEBP_UTILS_BIT_READER_INL_H_
17#define WEBP_UTILS_BIT_READER_INL_H_
18
19#ifdef HAVE_CONFIG_H
20#include "../webp/config.h"
21#endif
22
23#ifdef WEBP_FORCE_ALIGNED
24#include <string.h>  // memcpy
25#endif
26
27#include "../dsp/dsp.h"
28#include "./bit_reader.h"
29#include "./endian_inl.h"
30
31#ifdef __cplusplus
32extern "C" {
33#endif
34
35//------------------------------------------------------------------------------
36// Derived type lbit_t = natural type for memory I/O
37
38#if   (BITS > 32)
39typedef uint64_t lbit_t;
40#elif (BITS > 16)
41typedef uint32_t lbit_t;
42#elif (BITS >  8)
43typedef uint16_t lbit_t;
44#else
45typedef uint8_t lbit_t;
46#endif
47
48extern const uint8_t kVP8Log2Range[128];
49extern const uint8_t kVP8NewRange[128];
50
51// special case for the tail byte-reading
52void VP8LoadFinalBytes(VP8BitReader* const br);
53
54//------------------------------------------------------------------------------
55// Inlined critical functions
56
57// makes sure br->value_ has at least BITS bits worth of data
58static WEBP_INLINE void VP8LoadNewBytes(VP8BitReader* const br) {
59  assert(br != NULL && br->buf_ != NULL);
60  // Read 'BITS' bits at a time if possible.
61  if (br->buf_ < br->buf_max_) {
62    // convert memory type to register type (with some zero'ing!)
63    bit_t bits;
64#if defined(WEBP_FORCE_ALIGNED)
65    lbit_t in_bits;
66    memcpy(&in_bits, br->buf_, sizeof(in_bits));
67#elif defined(WEBP_USE_MIPS32)
68    // This is needed because of un-aligned read.
69    lbit_t in_bits;
70    lbit_t* p_buf_ = (lbit_t*)br->buf_;
71    __asm__ volatile(
72      ".set   push                             \n\t"
73      ".set   at                               \n\t"
74      ".set   macro                            \n\t"
75      "ulw    %[in_bits], 0(%[p_buf_])         \n\t"
76      ".set   pop                              \n\t"
77      : [in_bits]"=r"(in_bits)
78      : [p_buf_]"r"(p_buf_)
79      : "memory", "at"
80    );
81#else
82    const lbit_t in_bits = *(const lbit_t*)br->buf_;
83#endif
84    br->buf_ += BITS >> 3;
85#if !defined(WORDS_BIGENDIAN)
86#if (BITS > 32)
87    bits = BSwap64(in_bits);
88    bits >>= 64 - BITS;
89#elif (BITS >= 24)
90    bits = BSwap32(in_bits);
91    bits >>= (32 - BITS);
92#elif (BITS == 16)
93    bits = BSwap16(in_bits);
94#else   // BITS == 8
95    bits = (bit_t)in_bits;
96#endif  // BITS > 32
97#else    // WORDS_BIGENDIAN
98    bits = (bit_t)in_bits;
99    if (BITS != 8 * sizeof(bit_t)) bits >>= (8 * sizeof(bit_t) - BITS);
100#endif
101    br->value_ = bits | (br->value_ << BITS);
102    br->bits_ += BITS;
103  } else {
104    VP8LoadFinalBytes(br);    // no need to be inlined
105  }
106}
107
108// Read a bit with proba 'prob'. Speed-critical function!
109static WEBP_INLINE int VP8GetBit(VP8BitReader* const br, int prob) {
110  // Don't move this declaration! It makes a big speed difference to store
111  // 'range' *before* calling VP8LoadNewBytes(), even if this function doesn't
112  // alter br->range_ value.
113  range_t range = br->range_;
114  if (br->bits_ < 0) {
115    VP8LoadNewBytes(br);
116  }
117  {
118    const int pos = br->bits_;
119    const range_t split = (range * prob) >> 8;
120    const range_t value = (range_t)(br->value_ >> pos);
121#if defined(__arm__) || defined(_M_ARM)      // ARM-specific
122    const int bit = ((int)(split - value) >> 31) & 1;
123    if (value > split) {
124      range -= split + 1;
125      br->value_ -= (bit_t)(split + 1) << pos;
126    } else {
127      range = split;
128    }
129#else  // faster version on x86
130    int bit;  // Don't use 'const int bit = (value > split);", it's slower.
131    if (value > split) {
132      range -= split + 1;
133      br->value_ -= (bit_t)(split + 1) << pos;
134      bit = 1;
135    } else {
136      range = split;
137      bit = 0;
138    }
139#endif
140    if (range <= (range_t)0x7e) {
141      const int shift = kVP8Log2Range[range];
142      range = kVP8NewRange[range];
143      br->bits_ -= shift;
144    }
145    br->range_ = range;
146    return bit;
147  }
148}
149
150// simplified version of VP8GetBit() for prob=0x80 (note shift is always 1 here)
151static WEBP_INLINE int VP8GetSigned(VP8BitReader* const br, int v) {
152  if (br->bits_ < 0) {
153    VP8LoadNewBytes(br);
154  }
155  {
156    const int pos = br->bits_;
157    const range_t split = br->range_ >> 1;
158    const range_t value = (range_t)(br->value_ >> pos);
159    const int32_t mask = (int32_t)(split - value) >> 31;  // -1 or 0
160    br->bits_ -= 1;
161    br->range_ += mask;
162    br->range_ |= 1;
163    br->value_ -= (bit_t)((split + 1) & mask) << pos;
164    return (v ^ mask) - mask;
165  }
166}
167
168#ifdef __cplusplus
169}    // extern "C"
170#endif
171
172#endif   // WEBP_UTILS_BIT_READER_INL_H_
173