4 * A very fast decompressor for XPRESS (Huffman version).
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
72 #include "wimlib/decompressor_ops.h"
73 #include "wimlib/decompress_common.h"
74 #include "wimlib/error.h"
75 #include "wimlib/xpress.h"
77 /* This value is chosen for fast decompression. */
78 #define XPRESS_TABLEBITS 12
80 /* Decode the matches and literal bytes in a region of XPRESS-encoded data. */
82 xpress_decode_window(struct input_bitstream *istream, const u16 *decode_table,
83 u8 *window, unsigned window_size)
85 u8 *window_ptr = window;
86 u8 *window_end = &window[window_size];
90 unsigned match_offset;
92 while (window_ptr != window_end) {
94 sym = read_huffsym(istream, decode_table,
95 XPRESS_TABLEBITS, XPRESS_MAX_CODEWORD_LEN);
96 if (sym < XPRESS_NUM_CHARS) {
103 match_len = sym & 0xf;
104 offset_bsr = (sym >> 4) & 0xf;
106 bitstream_ensure_bits(istream, 16);
108 match_offset = (1 << offset_bsr) |
109 bitstream_pop_bits(istream, offset_bsr);
111 if (match_len == 0xf) {
112 match_len += bitstream_read_byte(istream);
113 if (match_len == 0xf + 0xff)
114 match_len = bitstream_read_u16(istream);
116 match_len += XPRESS_MIN_MATCH_LEN;
118 if (unlikely(match_offset > window_ptr - window))
121 if (unlikely(match_len > window_end - window_ptr))
124 lz_copy(window_ptr, match_len, match_offset, window_end);
126 window_ptr += match_len;
132 xpress_decompress(const void *compressed_data, size_t compressed_size,
133 void *uncompressed_data, size_t uncompressed_size, void *_ctx)
135 const u8 *cdata = compressed_data;
136 u8 lens[XPRESS_NUM_SYMBOLS];
138 u16 decode_table[(1 << XPRESS_TABLEBITS) + 2 * XPRESS_NUM_SYMBOLS]
139 _aligned_attribute(DECODE_TABLE_ALIGNMENT);
140 struct input_bitstream istream;
142 /* XPRESS uses only one Huffman code. It contains 512 symbols, and the
143 * code lengths of these symbols are given literally as 4-bit integers
144 * in the first 256 bytes of the compressed data. */
145 if (compressed_size < XPRESS_NUM_SYMBOLS / 2)
149 for (unsigned i = 0; i < XPRESS_NUM_SYMBOLS / 2; i++) {
150 *lens_p++ = cdata[i] & 0xf;
151 *lens_p++ = cdata[i] >> 4;
154 if (make_huffman_decode_table(decode_table, XPRESS_NUM_SYMBOLS,
155 XPRESS_TABLEBITS, lens,
156 XPRESS_MAX_CODEWORD_LEN))
159 init_input_bitstream(&istream, cdata + XPRESS_NUM_SYMBOLS / 2,
160 compressed_size - XPRESS_NUM_SYMBOLS / 2);
162 return xpress_decode_window(&istream, decode_table,
163 uncompressed_data, uncompressed_size);
167 xpress_create_decompressor(size_t max_block_size, void **dec_ret)
169 if (max_block_size > XPRESS_MAX_OFFSET + 1)
170 return WIMLIB_ERR_INVALID_PARAM;
175 const struct decompressor_ops xpress_decompressor_ops = {
176 .create_decompressor = xpress_create_decompressor,
177 .decompress = xpress_decompress,