range_decoder.h revision 292588 
1///////////////////////////////////////////////////////////////////////////////
2//
3/// \file       range_decoder.h
4/// \brief      Range Decoder
5///
6//  Authors:    Igor Pavlov
7//              Lasse Collin
8//
9//  This file has been put into the public domain.
10//  You can do whatever you want with this file.
11//
12///////////////////////////////////////////////////////////////////////////////
13
14#ifndef LZMA_RANGE_DECODER_H
15#define LZMA_RANGE_DECODER_H
16
17#include "range_common.h"
18
19
20typedef struct {
21	uint32_t range;
22	uint32_t code;
23	uint32_t init_bytes_left;
24} lzma_range_decoder;
25
26
27/// Reads the first five bytes to initialize the range decoder.
28static inline lzma_ret
29rc_read_init(lzma_range_decoder *rc, const uint8_t *restrict in,
30		size_t *restrict in_pos, size_t in_size)
31{
32	while (rc->init_bytes_left > 0) {
33		if (*in_pos == in_size)
34			return LZMA_OK;
35
36		// The first byte is always 0x00. It could have been omitted
37		// in LZMA2 but it wasn't, so one byte is wasted in every
38		// LZMA2 chunk.
39		if (rc->init_bytes_left == 5 && in[*in_pos] != 0x00)
40			return LZMA_DATA_ERROR;
41
42		rc->code = (rc->code << 8) | in[*in_pos];
43		++*in_pos;
44		--rc->init_bytes_left;
45	}
46
47	return LZMA_STREAM_END;
48}
49
50
51/// Makes local copies of range decoder and *in_pos variables. Doing this
52/// improves speed significantly. The range decoder macros expect also
53/// variables `in' and `in_size' to be defined.
54#define rc_to_local(range_decoder, in_pos) \
55	lzma_range_decoder rc = range_decoder; \
56	size_t rc_in_pos = (in_pos); \
57	uint32_t rc_bound
58
59
60/// Stores the local copes back to the range decoder structure.
61#define rc_from_local(range_decoder, in_pos) \
62do { \
63	range_decoder = rc; \
64	in_pos = rc_in_pos; \
65} while (0)
66
67
68/// Resets the range decoder structure.
69#define rc_reset(range_decoder) \
70do { \
71	(range_decoder).range = UINT32_MAX; \
72	(range_decoder).code = 0; \
73	(range_decoder).init_bytes_left = 5; \
74} while (0)
75
76
77/// When decoding has been properly finished, rc.code is always zero unless
78/// the input stream is corrupt. So checking this can catch some corrupt
79/// files especially if they don't have any other integrity check.
80#define rc_is_finished(range_decoder) \
81	((range_decoder).code == 0)
82
83
84/// Read the next input byte if needed. If more input is needed but there is
85/// no more input available, "goto out" is used to jump out of the main
86/// decoder loop.
87#define rc_normalize(seq) \
88do { \
89	if (rc.range < RC_TOP_VALUE) { \
90		if (unlikely(rc_in_pos == in_size)) { \
91			coder->sequence = seq; \
92			goto out; \
93		} \
94		rc.range <<= RC_SHIFT_BITS; \
95		rc.code = (rc.code << RC_SHIFT_BITS) | in[rc_in_pos++]; \
96	} \
97} while (0)
98
99
100/// Start decoding a bit. This must be used together with rc_update_0()
101/// and rc_update_1():
102///
103///     rc_if_0(prob, seq) {
104///         rc_update_0(prob);
105///         // Do something
106///     } else {
107///         rc_update_1(prob);
108///         // Do something else
109///     }
110///
111#define rc_if_0(prob, seq) \
112	rc_normalize(seq); \
113	rc_bound = (rc.range >> RC_BIT_MODEL_TOTAL_BITS) * (prob); \
114	if (rc.code < rc_bound)
115
116
117/// Update the range decoder state and the used probability variable to
118/// match a decoded bit of 0.
119#define rc_update_0(prob) \
120do { \
121	rc.range = rc_bound; \
122	prob += (RC_BIT_MODEL_TOTAL - (prob)) >> RC_MOVE_BITS; \
123} while (0)
124
125
126/// Update the range decoder state and the used probability variable to
127/// match a decoded bit of 1.
128#define rc_update_1(prob) \
129do { \
130	rc.range -= rc_bound; \
131	rc.code -= rc_bound; \
132	prob -= (prob) >> RC_MOVE_BITS; \
133} while (0)
134
135
136/// Decodes one bit and runs action0 or action1 depending on the decoded bit.
137/// This macro is used as the last step in bittree reverse decoders since
138/// those don't use "symbol" for anything else than indexing the probability
139/// arrays.
140#define rc_bit_last(prob, action0, action1, seq) \
141do { \
142	rc_if_0(prob, seq) { \
143		rc_update_0(prob); \
144		action0; \
145	} else { \
146		rc_update_1(prob); \
147		action1; \
148	} \
149} while (0)
150
151
152/// Decodes one bit, updates "symbol", and runs action0 or action1 depending
153/// on the decoded bit.
154#define rc_bit(prob, action0, action1, seq) \
155	rc_bit_last(prob, \
156		symbol <<= 1; action0, \
157		symbol = (symbol << 1) + 1; action1, \
158		seq);
159
160
161/// Like rc_bit() but add "case seq:" as a prefix. This makes the unrolled
162/// loops more readable because the code isn't littered with "case"
163/// statements. On the other hand this also makes it less readable, since
164/// spotting the places where the decoder loop may be restarted is less
165/// obvious.
166#define rc_bit_case(prob, action0, action1, seq) \
167	case seq: rc_bit(prob, action0, action1, seq)
168
169
170/// Decode a bit without using a probability.
171#define rc_direct(dest, seq) \
172do { \
173	rc_normalize(seq); \
174	rc.range >>= 1; \
175	rc.code -= rc.range; \
176	rc_bound = UINT32_C(0) - (rc.code >> 31); \
177	rc.code += rc.range & rc_bound; \
178	dest = (dest << 1) + (rc_bound + 1); \
179} while (0)
180
181
182// NOTE: No macros are provided for bittree decoding. It seems to be simpler
183// to just write them open in the code.
184
185#endif
186