4 * XPRESS decompression routines.
9 * Copyright (C) 2012, 2013 Eric Biggers
11 * This file is part of wimlib, a library for working with WIM files.
13 * wimlib is free software; you can redistribute it and/or modify it under the
14 * terms of the GNU General Public License as published by the Free
15 * Software Foundation; either version 3 of the License, or (at your option)
18 * wimlib is distributed in the hope that it will be useful, but WITHOUT ANY
19 * WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS FOR
20 * A PARTICULAR PURPOSE. See the GNU General Public License for more
23 * You should have received a copy of the GNU General Public License
24 * along with wimlib; if not, see http://www.gnu.org/licenses/.
29 * The XPRESS compression format is an LZ77 and Huffman-code based algorithm.
30 * That means it is fairly similar to LZX compression, but XPRESS is simpler, so
31 * it is a little faster to compress and decompress.
33 * The XPRESS compression format is mostly documented in a file called "[MS-XCA]
34 * Xpress Compression Algorithm". In the MSDN library, it can currently be
35 * found under Open Specifications => Protocols => Windows Protocols => Windows
36 * Server Protocols => [MS-XCA] Xpress Compression Algorithm". The format in
37 * WIMs is specifically the algorithm labeled as the "LZ77+Huffman Algorithm"
38 * (there apparently are some other versions of XPRESS as well).
40 * If you are already familiar with the LZ77 algorithm and Huffman coding, the
41 * XPRESS format is fairly simple. The compressed data begins with 256 bytes
42 * that contain 512 4-bit integers that are the lengths of the symbols in the
43 * Huffman code used for match/literal headers. In contrast with more
44 * complicated formats such as DEFLATE and LZX, this is the only Huffman code
45 * that is used for the entirety of the XPRESS compressed data, and the codeword
46 * lengths are not encoded with a pretree.
48 * The rest of the compressed data is Huffman-encoded symbols. Values 0 through
49 * 255 represent the corresponding literal bytes. Values 256 through 511
50 * represent matches and may require extra bits or bytes to be read to get the
51 * match offset and match length.
53 * The trickiest part is probably the way in which literal bytes for match
54 * lengths are interleaved in the bitstream.
56 * Also, a caveat--- according to Microsoft's documentation for XPRESS,
58 * "Some implementation of the decompression algorithm expect an extra
59 * symbol to mark the end of the data. Specifically, some implementations
60 * fail during decompression if the Huffman symbol 256 is not found after
63 * This is the case for the implementation in WIMGAPI. However, wimlib's
64 * decompressor in this file currently does not care if this extra symbol is
73 #include "wimlib/decompress.h"
74 #include "wimlib/xpress.h"
77 * Decodes a symbol @sym that begins an XPRESS match.
79 * The low 8 bits of the symbol are divided into:
81 * bits 0-3: length header
82 * bits 4-7: index of high-order bit of match offset
84 * Returns the match length, or -1 if the data is invalid.
87 xpress_decode_match(unsigned sym, input_idx_t window_pos,
88 input_idx_t window_len, u8 window[restrict],
89 struct input_bitstream * restrict istream)
99 unsigned match_offset;
101 sym -= XPRESS_NUM_CHARS;
103 offset_bsr = sym >> 4;
105 if (bitstream_ensure_bits(istream, 16))
108 match_offset = (1U << offset_bsr) | bitstream_pop_bits(istream, offset_bsr);
110 if (len_hdr == 0xf) {
111 ret = bitstream_read_byte(istream);
115 if (unlikely(match_len == 0xff)) {
116 ret = bitstream_read_byte(istream);
121 ret = bitstream_read_byte(istream);
125 match_len |= (ret << 8);
132 match_len += XPRESS_MIN_MATCH_LEN;
135 /* Verify the match is in bounds, then copy its data to the the current
138 if (window_pos + match_len > window_len)
141 if (match_offset > window_pos)
144 match_dest = window + window_pos;
145 match_src = match_dest - match_offset;
147 for (i = 0; i < match_len; i++)
148 match_dest[i] = match_src[i];
153 /* Decodes the Huffman-encoded matches and literal bytes in a region of
154 * XPRESS-encoded data. */
156 xpress_lz_decode(struct input_bitstream * restrict istream,
157 u8 uncompressed_data[restrict],
158 unsigned uncompressed_len,
159 const u8 lens[restrict],
160 const u16 decode_table[restrict])
165 for (curpos = 0; curpos < uncompressed_len; curpos += match_len) {
169 if (unlikely(bitstream_ensure_bits(istream, 16)))
172 if (unlikely(read_huffsym(istream, decode_table, lens,
173 XPRESS_NUM_SYMBOLS, XPRESS_TABLEBITS,
174 &sym, XPRESS_MAX_CODEWORD_LEN)))
177 if (sym < XPRESS_NUM_CHARS) {
179 uncompressed_data[curpos] = sym;
183 ret = xpress_decode_match(sym,
188 if (unlikely(ret < 0))
197 /* API function documented in wimlib.h */
199 wimlib_xpress_decompress(const void * const restrict _compressed_data,
200 const unsigned compressed_len,
201 void * const restrict uncompressed_data,
202 const unsigned uncompressed_len)
204 const u8 *compressed_data = _compressed_data;
205 u8 lens[XPRESS_NUM_SYMBOLS];
207 u16 decode_table[(1 << XPRESS_TABLEBITS) + 2 * XPRESS_NUM_SYMBOLS]
208 _aligned_attribute(DECODE_TABLE_ALIGNMENT);
209 struct input_bitstream istream;
211 /* XPRESS uses only one Huffman code. It contains 512 symbols, and the
212 * code lengths of these symbols are given literally as 4-bit integers
213 * in the first 256 bytes of the compressed data. */
214 if (compressed_len < XPRESS_NUM_SYMBOLS / 2)
218 for (unsigned i = 0; i < XPRESS_NUM_SYMBOLS / 2; i++) {
219 *lens_p++ = compressed_data[i] & 0xf;
220 *lens_p++ = compressed_data[i] >> 4;
223 if (make_huffman_decode_table(decode_table, XPRESS_NUM_SYMBOLS,
224 XPRESS_TABLEBITS, lens,
225 XPRESS_MAX_CODEWORD_LEN))
228 init_input_bitstream(&istream, compressed_data + XPRESS_NUM_SYMBOLS / 2,
229 compressed_len - XPRESS_NUM_SYMBOLS / 2);
231 return xpress_lz_decode(&istream, uncompressed_data,
232 uncompressed_len, lens, decode_table);