wimlib.net Git - wimlib/blob - src/xpress-compress.c

   1 /*
   2  * xpress-compress.c
   3  *
   4  * XPRESS compression routines.
   5  *
   6  * See the comments in xpress-decompress.c about the XPRESS format.
   7  */
   8
   9 /*
  10  * Copyright (C) 2012, 2013 Eric Biggers
  11  *
  12  * This file is part of wimlib, a library for working with WIM files.
  13  *
  14  * wimlib is free software; you can redistribute it and/or modify it under the
  15  * terms of the GNU General Public License as published by the Free
  16  * Software Foundation; either version 3 of the License, or (at your option)
  17  * any later version.
  18  *
  19  * wimlib is distributed in the hope that it will be useful, but WITHOUT ANY
  20  * WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS FOR
  21  * A PARTICULAR PURPOSE. See the GNU General Public License for more
  22  * details.
  23  *
  24  * You should have received a copy of the GNU General Public License
  25  * along with wimlib; if not, see http://www.gnu.org/licenses/.
  26  */
  27
  28 #ifdef HAVE_CONFIG_H
  29 #  include "config.h"
  30 #endif
  31
  32 #include "wimlib.h"
  33 #include "wimlib/assert.h"
  34 #include "wimlib/compress.h"
  35 #include "wimlib/error.h"
  36 #include "wimlib/util.h"
  37 #include "wimlib/xpress.h"
  38
  39 #include <stdlib.h>
  40 #include <string.h>
  41
  42 /*
  43  * Writes @match, which is a match given in the intermediate representation for
  44  * XPRESS matches, to the output stream @ostream.
  45  *
  46  * @codewords and @lens provide the Huffman code that is being used.
  47  */
  48 static int
  49 xpress_write_match(struct output_bitstream *restrict ostream,
  50                    u32 match,
  51                    const u16 codewords[restrict],
  52                    const u8 lens[restrict])
  53 {
  54         u32 adjusted_match_len = match & 0xffff;
  55         u32 match_offset = match >> 16;
  56         u32 len_hdr = min(adjusted_match_len, 0xf);
  57         u32 offset_bsr = bsr32(match_offset);
  58         u32 sym = len_hdr | (offset_bsr << 4) | XPRESS_NUM_CHARS;
  59         int ret;
  60
  61         ret = bitstream_put_bits(ostream, codewords[sym], lens[sym]);
  62         if (ret)
  63                 return ret;
  64
  65         if (adjusted_match_len >= 0xf) {
  66                 u8 byte1 = min(adjusted_match_len - 0xf, 0xff);
  67                 ret = bitstream_put_byte(ostream, byte1);
  68                 if (ret)
  69                         return ret;
  70                 if (byte1 == 0xff) {
  71                         ret = bitstream_put_two_bytes(ostream, adjusted_match_len);
  72                         if (ret)
  73                                 return ret;
  74                 }
  75         }
  76         return bitstream_put_bits(ostream,
  77                                   match_offset ^ (1 << offset_bsr), offset_bsr);
  78 }
  79
  80 static int
  81 xpress_write_compressed_literals(struct output_bitstream *ostream,
  82                                  const u32 match_tab[restrict],
  83                                  unsigned num_matches,
  84                                  const u16 codewords[restrict],
  85                                  const u8 lens[restrict])
  86 {
  87         for (unsigned i = 0; i < num_matches; i++) {
  88                 int ret;
  89                 u32 match = match_tab[i];
  90
  91                 if (match >= XPRESS_NUM_CHARS) /* match */
  92                         ret = xpress_write_match(ostream, match, codewords,
  93                                                  lens);
  94                 else /* literal byte */
  95                         ret = bitstream_put_bits(ostream, codewords[match],
  96                                                  lens[match]);
  97                 if (ret)
  98                         return ret;
  99         }
 100         return bitstream_put_bits(ostream, codewords[XPRESS_END_OF_DATA],
 101                                   lens[XPRESS_END_OF_DATA]);
 102 }
 103
 104 static u32
 105 xpress_record_literal(u8 literal, void *_freq_tab)
 106 {
 107         freq_t *freq_tab = _freq_tab;
 108         freq_tab[literal]++;
 109         return literal;
 110 }
 111
 112 static u32
 113 xpress_record_match(unsigned match_offset, unsigned match_len,
 114                     void *freq_tab, void *_ignore)
 115 {
 116         wimlib_assert(match_len >= XPRESS_MIN_MATCH &&
 117                       match_len <= XPRESS_MAX_MATCH);
 118         wimlib_assert(match_offset >= XPRESS_MIN_OFFSET &&
 119                       match_offset <= XPRESS_MAX_OFFSET);
 120
 121         /*
 122          * The intermediate representation of XPRESS matches is as follows:
 123          *
 124          * bits    description
 125          * ----    -----------------------------------------------------------
 126          *
 127          * 16-31   match offset (XPRESS_MIN_OFFSET < x < XPRESS_MAX_OFFSET)
 128          *
 129          * 0-15    adjusted match length (0 <= x <= XPRESS_MAX_MATCH - XPRESS_MIN_MATCH)
 130          *
 131          * Literals are simply represented as themselves and can be
 132          * distinguished from matches by the fact that only literals will have
 133          * the upper three bytes completely clear. */
 134
 135         u32 adjusted_match_len = match_len - XPRESS_MIN_MATCH;
 136         u32 len_hdr = min(adjusted_match_len, 0xf);
 137         u32 offset_bsr = bsr32(match_offset);
 138         u32 sym = len_hdr | (offset_bsr << 4) | XPRESS_NUM_CHARS;
 139         ((freq_t*)freq_tab)[sym]++;
 140         return adjusted_match_len | (match_offset << 16);
 141 }
 142
 143 static const struct lz_params xpress_lz_params = {
 144         .min_match      = XPRESS_MIN_MATCH,
 145         .max_match      = XPRESS_MAX_MATCH,
 146         .good_match     = 16,
 147         .nice_match     = 32,
 148         .max_chain_len  = 16,
 149         .max_lazy_match = 16,
 150         .too_far        = 4096,
 151 };
 152
 153 /* Documented in wimlib.h */
 154 WIMLIBAPI unsigned
 155 wimlib_xpress_compress(const void * restrict _uncompressed_data,
 156                        unsigned uncompressed_len,
 157                        void * restrict _compressed_data)
 158 {
 159         u8 *compressed_data = _compressed_data;
 160         struct output_bitstream ostream;
 161         u32 match_tab[uncompressed_len];
 162         freq_t freq_tab[XPRESS_NUM_SYMBOLS];
 163         u16 codewords[XPRESS_NUM_SYMBOLS];
 164         u8 lens[XPRESS_NUM_SYMBOLS];
 165         unsigned num_matches;
 166         unsigned compressed_len;
 167         unsigned i;
 168         int ret;
 169         u8 uncompressed_data[uncompressed_len + 8];
 170
 171         memcpy(uncompressed_data, _uncompressed_data, uncompressed_len);
 172         memset(uncompressed_data + uncompressed_len, 0, 8);
 173
 174         wimlib_assert(uncompressed_len <= 32768);
 175
 176         /* XPRESS requires 256 bytes of overhead for the Huffman tables, so it's
 177          * impossible cannot compress 256 bytes or less of data to less than the
 178          * input size.
 179          *
 180          * +1 to take into account that the buffer for compressed data is 1 byte
 181          * smaller than the buffer for uncompressed data.
 182          *
 183          * +4 to take into account that init_output_bitstream() requires at
 184          * least 4 bytes of data. */
 185         if (uncompressed_len < XPRESS_NUM_SYMBOLS / 2 + 1 + 4)
 186                 return 0;
 187
 188         ZERO_ARRAY(freq_tab);
 189         num_matches = lz_analyze_block(uncompressed_data, uncompressed_len,
 190                                        match_tab, xpress_record_match,
 191                                        xpress_record_literal, freq_tab,
 192                                        NULL, freq_tab,
 193                                        &xpress_lz_params);
 194
 195         freq_tab[XPRESS_END_OF_DATA]++;
 196
 197         make_canonical_huffman_code(XPRESS_NUM_SYMBOLS, XPRESS_MAX_CODEWORD_LEN,
 198                                     freq_tab, lens, codewords);
 199
 200         /* IMPORTANT NOTE:
 201          *
 202          * It's tempting to output the 512 Huffman codeword lengths using the
 203          * bitstream_put_bits() function.  However, this is NOT correct because
 204          * bitstream_put_bits() will output 2 bytes at a time in little-endian
 205          * order, which is the order that is needed for the compressed literals.
 206          * However, the bytes in the lengths table are in order, so they need to
 207          * be written one at a time without using bitstream_put_bits().
 208          *
 209          * Because of this, init_output_bitstream() is not called until after
 210          * the lengths table is output.
 211          */
 212         for (i = 0; i < XPRESS_NUM_SYMBOLS; i += 2)
 213                 *compressed_data++ = (lens[i] & 0xf) | (lens[i + 1] << 4);
 214
 215         init_output_bitstream(&ostream, compressed_data,
 216                               uncompressed_len - XPRESS_NUM_SYMBOLS / 2 - 1);
 217
 218         ret = xpress_write_compressed_literals(&ostream, match_tab,
 219                                                num_matches, codewords, lens);
 220         if (ret)
 221                 return 0;
 222
 223         /* Flush any bits that are buffered. */
 224         ret = flush_output_bitstream(&ostream);
 225         if (ret)
 226                 return 0;
 227
 228         /* Assert that there are no output bytes between the ostream.output
 229          * pointer and the ostream.next_bit_output pointer.  This can only
 230          * happen if bytes had been written at the ostream.output pointer before
 231          * the last bit word was written to the stream.  But, this does not
 232          * occur since xpress_write_match() always finishes by writing some bits
 233          * (a Huffman symbol), and the bitstream was just flushed. */
 234         wimlib_assert(ostream.output - ostream.next_bit_output == 2);
 235
 236         /* The length of the compressed data is supposed to be the value of the
 237          * ostream.output pointer before flushing, which is now the
 238          * output.next_bit_output pointer after flushing.
 239          *
 240          * There will be an extra 2 bytes at the ostream.bit_output pointer,
 241          * which is zeroed out.  (These 2 bytes may be either the last bytes in
 242          * the compressed data, in which case they are actually unnecessary, or
 243          * they may precede a number of bytes embedded into the bitstream.) */
 244         if (ostream.bit_output >
 245             (const u8*)_compressed_data + uncompressed_len - 3)
 246                 return 0;
 247         *(u16*)ostream.bit_output = cpu_to_le16(0);
 248         compressed_len = ostream.next_bit_output - (const u8*)_compressed_data;
 249
 250         wimlib_assert(compressed_len <= uncompressed_len - 1);
 251
 252 #ifdef ENABLE_VERIFY_COMPRESSION
 253         /* Verify that we really get the same thing back when decompressing. */
 254         {
 255                 u8 buf[uncompressed_len];
 256                 ret = wimlib_xpress_decompress(_compressed_data, compressed_len,
 257                                                buf, uncompressed_len);
 258                 if (ret) {
 259                         ERROR("xpress_compress(): Failed to decompress data we "
 260                               "compressed");
 261                         abort();
 262                 }
 263                 for (i = 0; i < uncompressed_len; i++) {
 264                         if (buf[i] != uncompressed_data[i]) {
 265                                 ERROR("xpress_compress(): Data we compressed didn't "
 266                                       "decompress to the original data (difference at "
 267                                       "byte %u of %u)", i + 1, uncompressed_len);
 268                                 abort();
 269                         }
 270                 }
 271         }
 272 #endif
 273         return compressed_len;
 274 }