*/
/*
- * Copyright (C) 2012, 2013 Eric Biggers
+ * Copyright (C) 2012, 2013, 2014 Eric Biggers
*
* This file is part of wimlib, a library for working with WIM files.
*
*/
/*
- * LZX is a LZ77 and Huffman-code based compression format that has many
+ * LZX is an LZ77 and Huffman-code based compression format that has many
* similarities to the DEFLATE format used in zlib. The compression ratio is as
- * good or better than DEFLATE. However, in WIM files only up to 32768 bytes of
- * data can ever compressed be in the same LZX block, so a .tar.gz file could
- * potentially be smaller than a WIM file that uses LZX compression because it
- * can use a larger LZ77 window size.
+ * good or better than DEFLATE.
*
* Some notes on the LZX compression format as used in Windows Imaging (WIM)
* files:
*
* A compressed WIM resource consists of a table of chunk offsets followed by
* the compressed chunks themselves. All compressed chunks except possibly the
- * last decompress to WIM_CHUNK_SIZE (= 32768) bytes. This is quite similar to
- * the cabinet (.cab) file format, but they are not the same. According to the
- * cabinet format documentation, the LZX block size is independent from the
- * CFDATA blocks, and a LZX block may span several CFDATA blocks. However, in
- * WIMs, LZX blocks do not appear to ever span multiple WIM chunks. Note that
- * this means any WIM chunk may be decompressed or compressed independently from
- * any other chunk, which is convenient.
- *
- * A LZX compressed WIM chunk contains one or more LZX blocks of the aligned,
+ * last decompress to a fixed number of bytes, by default 32768. This is quite
+ * similar to the cabinet (.cab) file format, but they are not the same.
+ * According to the cabinet format documentation, the LZX block size is
+ * independent from the CFDATA blocks, and an LZX block may span several CFDATA
+ * blocks. However, in WIMs, LZX blocks do not appear to ever span multiple WIM
+ * chunks. Note that this means any WIM chunk may be decompressed or compressed
+ * independently from any other chunk, which allows random access.
+ *
+ * An LZX compressed WIM chunk contains one or more LZX blocks of the aligned,
* verbatim, or uncompressed block types. For aligned and verbatim blocks, the
* size of the block in uncompressed bytes is specified by a bit following the 3
* bits that specify the block type, possibly followed by an additional 16 bits.
- * '1' means to use the default block size (equal to 32768, the size of a WIM
- * chunk--- and this seems to only be valid for the first LZX block in a WIM
- * chunk), while '0' means that the block size is provided by the next 16 bits.
+ * '1' means to use the default block size (equal to 32768, the default size of
+ * a WIM chunk), while '0' means that the block size is provided by the next 16
+ * bits.
*
* The cabinet format, as documented, allows for the possibility that a
* compressed CFDATA chunk is up to 6144 bytes larger than the data it
* defined in the specification.
*
* The LZX document states that aligned offset blocks have their aligned offset
- * huffman tree AFTER the main and length trees. The implementation suggests
+ * Huffman tree AFTER the main and length trees. The implementation suggests
* that the aligned offset tree is BEFORE the main and length trees.
*
* The LZX document decoding algorithm states that, in an aligned offset block,
* if an extra_bits value is 1, 2 or 3, then that number of bits should be read
* and the result added to the match offset. This is correct for 1 and 2, but
- * not 3, where just a huffman symbol (using the aligned tree) should be read.
+ * not 3, where just a Huffman symbol (using the aligned tree) should be read.
*
* Regarding the E8 preprocessing, the LZX document states 'No translation may
* be performed on the last 6 bytes of the input block'. This is correct.
* would cause the next four bytes to be modified, at least one of which would
* be in the last 6 bytes, which is not allowed according to the spec.
*
- * The specification states that the huffman trees must always contain at least
+ * The specification states that the Huffman trees must always contain at least
* one element. However, many CAB files contain blocks where the length tree is
* completely empty (because there are no matches), and this is expected to
* succeed.
*/
-#include "util.h"
-#include "lzx.h"
-#include "decompress.h"
+#ifdef HAVE_CONFIG_H
+# include "config.h"
+#endif
+
+#include "wimlib.h"
+#include "wimlib/decompressor_ops.h"
+#include "wimlib/decompress_common.h"
+#include "wimlib/lzx.h"
+#include "wimlib/util.h"
+
#include <string.h>
+#ifdef __SSE2__
+# include <emmintrin.h>
+#endif
+
/* Huffman decoding tables and maps from symbols to code lengths. */
struct lzx_tables {
- u16 maintree_decode_table[(1 << LZX_MAINTREE_TABLEBITS) +
- (LZX_MAINTREE_NUM_SYMBOLS * 2)];
- u8 maintree_lens[LZX_MAINTREE_NUM_SYMBOLS];
+ u16 maintree_decode_table[(1 << LZX_MAINCODE_TABLEBITS) +
+ (LZX_MAINCODE_MAX_NUM_SYMBOLS * 2)]
+ _aligned_attribute(DECODE_TABLE_ALIGNMENT);
+ u8 maintree_lens[LZX_MAINCODE_MAX_NUM_SYMBOLS];
- u16 lentree_decode_table[(1 << LZX_LENTREE_TABLEBITS) +
- (LZX_LENTREE_NUM_SYMBOLS * 2)];
- u8 lentree_lens[LZX_LENTREE_NUM_SYMBOLS];
+ u16 lentree_decode_table[(1 << LZX_LENCODE_TABLEBITS) +
+ (LZX_LENCODE_NUM_SYMBOLS * 2)]
+ _aligned_attribute(DECODE_TABLE_ALIGNMENT);
+ u8 lentree_lens[LZX_LENCODE_NUM_SYMBOLS];
- u16 alignedtree_decode_table[(1 << LZX_ALIGNEDTREE_TABLEBITS) +
- (LZX_ALIGNEDTREE_NUM_SYMBOLS * 2)];
- u8 alignedtree_lens[LZX_ALIGNEDTREE_NUM_SYMBOLS];
-};
+ u16 alignedtree_decode_table[(1 << LZX_ALIGNEDCODE_TABLEBITS) +
+ (LZX_ALIGNEDCODE_NUM_SYMBOLS * 2)]
+ _aligned_attribute(DECODE_TABLE_ALIGNMENT);
+ u8 alignedtree_lens[LZX_ALIGNEDCODE_NUM_SYMBOLS];
+} _aligned_attribute(DECODE_TABLE_ALIGNMENT);
+struct lzx_decompressor {
+ u32 max_window_size;
+ unsigned num_main_syms;
+ struct lzx_tables tables;
+};
/*
* Reads a Huffman-encoded symbol using the pre-tree.
const u8 pretree_lens[], unsigned *n)
{
return read_huffsym(istream, pretree_decode_table, pretree_lens,
- LZX_PRETREE_NUM_SYMBOLS, LZX_PRETREE_TABLEBITS, n,
- LZX_MAX_CODEWORD_LEN);
+ LZX_PRECODE_NUM_SYMBOLS, LZX_PRECODE_TABLEBITS, n,
+ LZX_MAX_PRE_CODEWORD_LEN);
}
/* Reads a Huffman-encoded symbol using the main tree. */
static inline int
read_huffsym_using_maintree(struct input_bitstream *istream,
const struct lzx_tables *tables,
- unsigned *n)
+ unsigned *n,
+ unsigned num_main_syms)
{
return read_huffsym(istream, tables->maintree_decode_table,
- tables->maintree_lens, LZX_MAINTREE_NUM_SYMBOLS,
- LZX_MAINTREE_TABLEBITS, n, LZX_MAX_CODEWORD_LEN);
+ tables->maintree_lens, num_main_syms,
+ LZX_MAINCODE_TABLEBITS, n, LZX_MAX_MAIN_CODEWORD_LEN);
}
/* Reads a Huffman-encoded symbol using the length tree. */
unsigned *n)
{
return read_huffsym(istream, tables->lentree_decode_table,
- tables->lentree_lens, LZX_LENTREE_NUM_SYMBOLS,
- LZX_LENTREE_TABLEBITS, n, LZX_MAX_CODEWORD_LEN);
+ tables->lentree_lens, LZX_LENCODE_NUM_SYMBOLS,
+ LZX_LENCODE_TABLEBITS, n, LZX_MAX_LEN_CODEWORD_LEN);
}
/* Reads a Huffman-encoded symbol using the aligned offset tree. */
{
return read_huffsym(istream, tables->alignedtree_decode_table,
tables->alignedtree_lens,
- LZX_ALIGNEDTREE_NUM_SYMBOLS,
- LZX_ALIGNEDTREE_TABLEBITS, n, 8);
+ LZX_ALIGNEDCODE_NUM_SYMBOLS,
+ LZX_ALIGNEDCODE_TABLEBITS, n,
+ LZX_MAX_ALIGNED_CODEWORD_LEN);
}
/*
* code length values from the input.
*
* @istream: The bit stream for the input. It is positioned on the beginning
- * of the pretree for the code length values.
+ * of the pretree for the code length values.
* @lens: An array that contains the length values from the previous time
- * the code lengths for this Huffman tree were read, or all
- * 0's if this is the first time.
+ * the code lengths for this Huffman tree were read, or all
+ * 0's if this is the first time.
* @num_lens: Number of length values to decode and return.
*
*/
unsigned num_lens)
{
/* Declare the decoding table and length table for the pretree. */
- u16 pretree_decode_table[(1 << LZX_PRETREE_TABLEBITS) +
- (LZX_PRETREE_NUM_SYMBOLS * 2)];
- u8 pretree_lens[LZX_PRETREE_NUM_SYMBOLS];
+ u16 pretree_decode_table[(1 << LZX_PRECODE_TABLEBITS) +
+ (LZX_PRECODE_NUM_SYMBOLS * 2)]
+ _aligned_attribute(DECODE_TABLE_ALIGNMENT);
+ u8 pretree_lens[LZX_PRECODE_NUM_SYMBOLS];
unsigned i;
- unsigned len;
+ u32 len;
int ret;
/* Read the code lengths of the pretree codes. There are 20 lengths of
* 4 bits each. */
- for (i = 0; i < LZX_PRETREE_NUM_SYMBOLS; i++) {
- ret = bitstream_read_bits(istream, LZX_PRETREE_ELEMENT_SIZE,
+ for (i = 0; i < LZX_PRECODE_NUM_SYMBOLS; i++) {
+ ret = bitstream_read_bits(istream, LZX_PRECODE_ELEMENT_SIZE,
&len);
- if (ret != 0)
+ if (ret)
return ret;
pretree_lens[i] = len;
}
/* Make the decoding table for the pretree. */
ret = make_huffman_decode_table(pretree_decode_table,
- LZX_PRETREE_NUM_SYMBOLS,
- LZX_PRETREE_TABLEBITS,
+ LZX_PRECODE_NUM_SYMBOLS,
+ LZX_PRECODE_TABLEBITS,
pretree_lens,
- LZX_MAX_CODEWORD_LEN);
- if (ret != 0)
+ LZX_MAX_PRE_CODEWORD_LEN);
+ if (ret)
return ret;
/* Pointer past the last length value that needs to be filled in. */
* the next lengths are all equal to the next symbol in the
* input. */
unsigned tree_code;
- unsigned num_zeroes;
+ u32 num_zeroes;
unsigned code;
- unsigned num_same;
- char value;
+ u32 num_same;
+ signed char value;
ret = read_huffsym_using_pretree(istream, pretree_decode_table,
pretree_lens, &tree_code);
- if (ret != 0)
+ if (ret)
return ret;
switch (tree_code) {
case 17: /* Run of 0's */
ret = bitstream_read_bits(istream, 4, &num_zeroes);
- if (ret != 0)
+ if (ret)
return ret;
num_zeroes += 4;
while (num_zeroes--) {
break;
case 18: /* Longer run of 0's */
ret = bitstream_read_bits(istream, 5, &num_zeroes);
- if (ret != 0)
+ if (ret)
return ret;
num_zeroes += 20;
while (num_zeroes--) {
break;
case 19: /* Run of identical lengths */
ret = bitstream_read_bits(istream, 1, &num_same);
- if (ret != 0)
+ if (ret)
return ret;
num_same += 4;
ret = read_huffsym_using_pretree(istream,
pretree_decode_table,
pretree_lens,
&code);
- if (ret != 0)
+ if (ret)
return ret;
- value = (char)*lens - (char)code;
+ value = (signed char)*lens - (signed char)code;
if (value < 0)
value += 17;
while (num_same--) {
}
break;
default: /* Difference from old length. */
- value = (char)*lens - (char)tree_code;
+ value = (signed char)*lens - (signed char)tree_code;
if (value < 0)
value += 17;
*lens = value;
*
* @istream: The input bitstream.
* @block_size_ret: A pointer to an int into which the size of the block,
- * in bytes, will be returned.
+ * in bytes, will be returned.
* @block_type_ret: A pointer to an int into which the type of the block
- * (LZX_BLOCKTYPE_*) will be returned.
- * @tables: A pointer to a lzx_tables structure in which the
- * main tree, the length tree, and possibly the
- * aligned offset tree will be constructed.
+ * (LZX_BLOCKTYPE_*) will be returned.
+ * @tables: A pointer to an lzx_tables structure in which the
+ * main tree, the length tree, and possibly the
+ * aligned offset tree will be constructed.
* @queue: A pointer to the least-recently-used queue into which
- * R0, R1, and R2 will be written (only for uncompressed
- * blocks, which contain this information in the header)
+ * R0, R1, and R2 will be written (only for uncompressed
+ * blocks, which contain this information in the header)
*/
static int
lzx_read_block_header(struct input_bitstream *istream,
+ unsigned num_main_syms,
+ unsigned max_window_size,
unsigned *block_size_ret,
unsigned *block_type_ret,
struct lzx_tables *tables,
- struct lru_queue *queue)
+ struct lzx_lru_queue *queue)
{
int ret;
unsigned block_type;
unsigned block_size;
- unsigned s;
- unsigned i;
- unsigned len;
ret = bitstream_ensure_bits(istream, 4);
- if (ret != 0) {
- ERROR("LZX input stream overrun");
+ if (ret)
return ret;
- }
/* The first three bits tell us what kind of block it is, and are one
* of the LZX_BLOCKTYPE_* values. */
block_type = bitstream_read_bits_nocheck(istream, 3);
- /* The next bit indicates whether the block size is the default (32768),
- * indicated by a 1 bit, or whether the block size is given by the next
- * 16 bits, indicated by a 0 bit. */
- s = bitstream_read_bits_nocheck(istream, 1);
-
- if (s) {
- block_size = 32768;
+ /* Read the block size. This mirrors the behavior
+ * lzx_write_compressed_block() in lzx-compress.c; see that for more
+ * details. */
+ if (bitstream_read_bits_nocheck(istream, 1)) {
+ block_size = LZX_DEFAULT_BLOCK_SIZE;
} else {
- ret = bitstream_read_bits(istream, 16, &block_size);
- if (ret != 0)
+ u32 tmp;
+ block_size = 0;
+
+ ret = bitstream_read_bits(istream, 8, &tmp);
+ if (ret)
return ret;
- block_size = le16_to_cpu(block_size);
+ block_size |= tmp;
+
+ ret = bitstream_read_bits(istream, 8, &tmp);
+ if (ret)
+ return ret;
+ block_size <<= 8;
+ block_size |= tmp;
+
+ if (max_window_size >= 65536) {
+ ret = bitstream_read_bits(istream, 8, &tmp);
+ if (ret)
+ return ret;
+ block_size <<= 8;
+ block_size |= tmp;
+ }
}
switch (block_type) {
/* Read the path lengths for the elements of the aligned tree,
* then build it. */
- for (i = 0; i < LZX_ALIGNEDTREE_NUM_SYMBOLS; i++) {
+ for (unsigned i = 0; i < LZX_ALIGNEDCODE_NUM_SYMBOLS; i++) {
+ u32 len;
+
ret = bitstream_read_bits(istream,
- LZX_ALIGNEDTREE_ELEMENT_SIZE,
+ LZX_ALIGNEDCODE_ELEMENT_SIZE,
&len);
- if (ret != 0)
+ if (ret)
return ret;
tables->alignedtree_lens[i] = len;
}
LZX_DEBUG("Building the aligned tree.");
ret = make_huffman_decode_table(tables->alignedtree_decode_table,
- LZX_ALIGNEDTREE_NUM_SYMBOLS,
- LZX_ALIGNEDTREE_TABLEBITS,
+ LZX_ALIGNEDCODE_NUM_SYMBOLS,
+ LZX_ALIGNEDCODE_TABLEBITS,
tables->alignedtree_lens,
- 8);
- if (ret != 0) {
- ERROR("lzx_decompress(): Failed to make the decode "
- "table for the aligned offset tree");
+ LZX_MAX_ALIGNED_CODEWORD_LEN);
+ if (ret) {
+ LZX_DEBUG("Failed to make the decode table for the "
+ "aligned offset tree");
return ret;
}
* tree. */
ret = lzx_read_code_lens(istream, tables->maintree_lens,
LZX_NUM_CHARS);
- if (ret != 0) {
- ERROR("lzx_decompress(): Failed to read the code "
- "lengths for the first 256 elements of the "
- "main tree");
+ if (ret) {
+ LZX_DEBUG("Failed to read the code lengths for the "
+ "first 256 elements of the main tree");
return ret;
}
* tree. */
LZX_DEBUG("Reading path lengths for remaining elements of "
"main tree (%d elements).",
- LZX_MAINTREE_NUM_SYMBOLS - LZX_NUM_CHARS);
+ num_main_syms - LZX_NUM_CHARS);
ret = lzx_read_code_lens(istream,
tables->maintree_lens + LZX_NUM_CHARS,
- LZX_MAINTREE_NUM_SYMBOLS - LZX_NUM_CHARS);
- if (ret != 0) {
- ERROR("lzx_decompress(): Failed to read the path "
- "lengths for the remaining elements of the main "
- "tree");
+ num_main_syms - LZX_NUM_CHARS);
+ if (ret) {
+ LZX_DEBUG("Failed to read the path lengths for the "
+ "remaining elements of the main tree");
return ret;
}
"table for the main tree.");
ret = make_huffman_decode_table(tables->maintree_decode_table,
- LZX_MAINTREE_NUM_SYMBOLS,
- LZX_MAINTREE_TABLEBITS,
+ num_main_syms,
+ LZX_MAINCODE_TABLEBITS,
tables->maintree_lens,
- LZX_MAX_CODEWORD_LEN);
- if (ret != 0) {
- ERROR("lzx_decompress(): Failed to make the decode "
- "table for the main tree");
+ LZX_MAX_MAIN_CODEWORD_LEN);
+ if (ret) {
+ LZX_DEBUG("Failed to make the decode "
+ "table for the main tree");
return ret;
}
LZX_DEBUG("Reading path lengths for the length tree.");
ret = lzx_read_code_lens(istream, tables->lentree_lens,
- LZX_LENTREE_NUM_SYMBOLS);
- if (ret != 0) {
- ERROR("lzx_decompress(): Failed to read the path "
- "lengths for the length tree");
+ LZX_LENCODE_NUM_SYMBOLS);
+ if (ret) {
+ LZX_DEBUG("Failed to read the path "
+ "lengths for the length tree");
return ret;
}
LZX_DEBUG("Building the length tree.");
ret = make_huffman_decode_table(tables->lentree_decode_table,
- LZX_LENTREE_NUM_SYMBOLS,
- LZX_LENTREE_TABLEBITS,
+ LZX_LENCODE_NUM_SYMBOLS,
+ LZX_LENCODE_TABLEBITS,
tables->lentree_lens,
- LZX_MAX_CODEWORD_LEN);
- if (ret != 0) {
- ERROR("lzx_decompress(): Failed to build the length "
- "Huffman tree");
+ LZX_MAX_LEN_CODEWORD_LEN);
+ if (ret) {
+ LZX_DEBUG("Failed to build the length Huffman tree");
return ret;
}
/* The bitstream of compressed literals and matches for this
* *already* aligned, the correct thing to do is to throw away
* the next 16 bits. */
if (istream->bitsleft == 0) {
- if (istream->data_bytes_left < 14)
+ if (istream->data_bytes_left < 14) {
+ LZX_DEBUG("Insufficient length in "
+ "uncompressed block");
return -1;
+ }
istream->data += 2;
istream->data_bytes_left -= 2;
} else {
- if (istream->data_bytes_left < 12)
+ if (istream->data_bytes_left < 12) {
+ LZX_DEBUG("Insufficient length in "
+ "uncompressed block");
return -1;
+ }
istream->bitsleft = 0;
istream->bitbuf = 0;
}
- queue->R0 = le32_to_cpu(*(u32*)(istream->data + 0));
- queue->R1 = le32_to_cpu(*(u32*)(istream->data + 4));
- queue->R2 = le32_to_cpu(*(u32*)(istream->data + 8));
+ queue->R[0] = le32_to_cpu(*(le32*)(istream->data + 0));
+ queue->R[1] = le32_to_cpu(*(le32*)(istream->data + 4));
+ queue->R[2] = le32_to_cpu(*(le32*)(istream->data + 8));
istream->data += 12;
istream->data_bytes_left -= 12;
/* The uncompressed data of this block directly follows and will
* be read in lzx_decompress(). */
break;
default:
- ERROR("lzx_decompress(): Found invalid block");
+ LZX_DEBUG("Found invalid block");
return -1;
}
*block_type_ret = block_type;
* tree.
*
* @block_type: The type of the block (LZX_BLOCKTYPE_ALIGNED or
- * LZX_BLOCKTYPE_VERBATIM)
+ * LZX_BLOCKTYPE_VERBATIM)
*
* @bytes_remaining: The amount of uncompressed data remaining to be
- * uncompressed in this block. It is an error if the match
- * is longer than this number.
+ * uncompressed in this block. It is an error if the match
+ * is longer than this number.
*
* @window: A pointer to the window into which the uncompressed
- * data is being written.
+ * data is being written.
*
* @window_pos: The current byte offset in the window.
*
*
* Returns the length of the match, or a negative number on error. The possible
* error cases are:
- * - Match would exceed the amount of data remaining to be uncompressed.
- * - Match refers to data before the window.
- * - The input bitstream ended unexpectedly.
+ * - Match would exceed the amount of data remaining to be uncompressed.
+ * - Match refers to data before the window.
+ * - The input bitstream ended unexpectedly.
*/
static int
lzx_decode_match(unsigned main_element, int block_type,
unsigned bytes_remaining, u8 *window,
unsigned window_pos,
const struct lzx_tables *tables,
- struct lru_queue *queue,
+ struct lzx_lru_queue *queue,
struct input_bitstream *istream)
{
unsigned length_header;
unsigned match_offset;
unsigned additional_len;
unsigned num_extra_bits;
- unsigned verbatim_bits;
- unsigned aligned_bits;
+ u32 verbatim_bits;
+ u32 aligned_bits;
unsigned i;
int ret;
u8 *match_dest;
position_slot = main_element >> 3;
/* If the length_header is less than LZX_NUM_PRIMARY_LENS (= 7), it
- * gives the match length as the offset from LZX_MIN_MATCH. Otherwise,
- * the length is given by an additional symbol encoded using the length
- * tree, offset by 9 (LZX_MIN_MATCH + LZX_NUM_PRIMARY_LENS) */
- match_len = LZX_MIN_MATCH + length_header;
+ * gives the match length as the offset from LZX_MIN_MATCH_LEN.
+ * Otherwise, the length is given by an additional symbol encoded using
+ * the length tree, offset by 9 (LZX_MIN_MATCH_LEN +
+ * LZX_NUM_PRIMARY_LENS) */
+ match_len = LZX_MIN_MATCH_LEN + length_header;
if (length_header == LZX_NUM_PRIMARY_LENS) {
ret = read_huffsym_using_lentree(istream, tables,
&additional_len);
- if (ret != 0)
+ if (ret)
return ret;
match_len += additional_len;
}
* queue. */
switch (position_slot) {
case 0:
- match_offset = queue->R0;
+ match_offset = queue->R[0];
break;
case 1:
- match_offset = queue->R1;
- swap(queue->R0, queue->R1);
+ match_offset = queue->R[1];
+ swap(queue->R[0], queue->R[1]);
break;
case 2:
/* The queue doesn't work quite the same as a real LRU queue,
* since using the R2 offset doesn't bump the R1 offset down to
* R2. */
- match_offset = queue->R2;
- swap(queue->R0, queue->R2);
+ match_offset = queue->R[2];
+ swap(queue->R[0], queue->R[2]);
break;
default:
/* Otherwise, the offset was not encoded as one the offsets in
* will just set it to 0. ) */
ret = bitstream_read_bits(istream, num_extra_bits - 3,
&verbatim_bits);
- if (ret != 0)
+ if (ret)
return ret;
verbatim_bits <<= 3;
ret = read_huffsym_using_alignedtree(istream, tables,
&aligned_bits);
- if (ret != 0)
+ if (ret)
return ret;
} else {
/* For non-aligned blocks, or for aligned blocks with
* the alignment is taken to be 0. */
ret = bitstream_read_bits(istream, num_extra_bits,
&verbatim_bits);
- if (ret != 0)
+ if (ret)
return ret;
aligned_bits = 0;
/* Calculate the match offset. */
match_offset = lzx_position_base[position_slot] +
- verbatim_bits + aligned_bits - 2;
+ verbatim_bits + aligned_bits - LZX_OFFSET_OFFSET;
/* Update the LRU queue. */
- queue->R2 = queue->R1;
- queue->R1 = queue->R0;
- queue->R0 = match_offset;
+ queue->R[2] = queue->R[1];
+ queue->R[1] = queue->R[0];
+ queue->R[0] = match_offset;
break;
}
/* Verify that the match is in the bounds of the part of the window
* currently in use, then copy the source of the match to the current
* position. */
- match_dest = window + window_pos;
- match_src = match_dest - match_offset;
if (match_len > bytes_remaining) {
- ERROR("lzx_decode_match(): Match of length %u bytes overflows "
- "uncompressed block size", match_len);
+ LZX_DEBUG("Match of length %u bytes overflows "
+ "uncompressed block size", match_len);
return -1;
}
- if (match_src < window) {
- ERROR("lzx_decode_match(): Match of length %u bytes references "
- "data before window (match_offset = %u, window_pos = %u)",
- match_len, match_offset, window_pos);
+ if (match_offset > window_pos) {
+ LZX_DEBUG("Match of length %u bytes references "
+ "data before window (match_offset = %u, "
+ "window_pos = %u)",
+ match_len, match_offset, window_pos);
return -1;
}
+ match_dest = window + window_pos;
+ match_src = match_dest - match_offset;
+
#if 0
printf("Match: src %u, dst %u, len %u\n", match_src - window,
match_dest - window,
}
static void
-undo_call_insn_translation(u32 *call_insn_target, int input_pos,
- int32_t file_size)
+undo_call_insn_translation(u32 *call_insn_target, s32 input_pos)
{
- int32_t abs_offset;
- int32_t rel_offset;
+ s32 abs_offset;
+ s32 rel_offset;
abs_offset = le32_to_cpu(*call_insn_target);
- if (abs_offset >= -input_pos && abs_offset < file_size) {
+ if (abs_offset >= -input_pos && abs_offset < LZX_WIM_MAGIC_FILESIZE) {
if (abs_offset >= 0) {
/* "good translation" */
rel_offset = abs_offset - input_pos;
} else {
/* "compensating translation" */
- rel_offset = abs_offset + file_size;
+ rel_offset = abs_offset + LZX_WIM_MAGIC_FILESIZE;
}
*call_insn_target = cpu_to_le32(rel_offset);
}
* as it is used in calculating the translated jump targets. But in WIM files,
* this file size is always the same (LZX_WIM_MAGIC_FILESIZE == 12000000).*/
static void
-undo_call_insn_preprocessing(u8 uncompressed_data[], int uncompressed_data_len)
+undo_call_insn_preprocessing(u8 *uncompressed_data, size_t uncompressed_size)
{
- for (int i = 0; i < uncompressed_data_len - 10; i++) {
- if (uncompressed_data[i] == 0xe8) {
- undo_call_insn_translation((u32*)&uncompressed_data[i + 1],
- i,
- LZX_WIM_MAGIC_FILESIZE);
- i += 4;
- }
+#ifdef __SSE2__
+
+ /* SSE2 vectorized implementation for x86_64. This speeds up LZX
+ * decompression by about 5-8% overall. (Usually --- the performance
+ * actually regresses slightly in the degenerate case that the data
+ * consists entirely of 0xe8 bytes.) */
+ __m128i *p128 = (__m128i *)uncompressed_data;
+ u32 valid_mask = 0xFFFFFFFF;
+
+ if (uncompressed_size >= 32 &&
+ ((uintptr_t)uncompressed_data % 16 == 0))
+ {
+ __m128i * const end128 = p128 + uncompressed_size / 16 - 1;
+
+ /* Create a vector of all 0xe8 bytes */
+ const __m128i e8_bytes = _mm_set1_epi8(0xe8);
+
+ /* Iterate through the 16-byte vectors in the input. */
+ do {
+ /* Compare the current 16-byte vector with the vector of
+ * all 0xe8 bytes. This produces 0xff where the byte is
+ * 0xe8 and 0x00 where it is not. */
+ __m128i cmpresult = _mm_cmpeq_epi8(*p128, e8_bytes);
+
+ /* Map the comparison results into a single 16-bit
+ * number. It will contain a 1 bit when the
+ * corresponding byte in the current 16-byte vector is
+ * an e8 byte. Note: the low-order bit corresponds to
+ * the first (lowest address) byte. */
+ u32 e8_mask = _mm_movemask_epi8(cmpresult);
+
+ if (!e8_mask) {
+ /* If e8_mask is 0, then none of these 16 bytes
+ * have value 0xe8. No e8 translation is
+ * needed, and there is no restriction that
+ * carries over to the next 16 bytes. */
+ valid_mask = 0xFFFFFFFF;
+ } else {
+ /* At least one byte has value 0xe8.
+ *
+ * The AND with valid_mask accounts for the fact
+ * that we can't start an e8 translation that
+ * overlaps the previous one. */
+ while ((e8_mask &= valid_mask)) {
+
+ /* Count the number of trailing zeroes
+ * in e8_mask. This will produce the
+ * index of the byte, within the 16, at
+ * which the next e8 translation should
+ * be done. */
+ u32 bit = __builtin_ctz(e8_mask);
+
+ /* Do the e8 translation. */
+ u8 *p8 = (u8 *)p128 + bit;
+ undo_call_insn_translation((s32 *)(p8 + 1),
+ p8 - uncompressed_data);
+
+ /* Don't start an e8 translation in the
+ * next 4 bytes. */
+ valid_mask &= ~((u32)0x1F << bit);
+ }
+ /* Moving on to the next vector. Shift and set
+ * valid_mask accordingly. */
+ valid_mask >>= 16;
+ valid_mask |= 0xFFFF0000;
+ }
+ } while (++p128 < end128);
+ }
+
+ u8 *p8 = (u8 *)p128;
+ while (!(valid_mask & 1)) {
+ p8++;
+ valid_mask >>= 1;
+ }
+#else /* __SSE2__ */
+ u8 *p8 = uncompressed_data;
+#endif /* !__SSE2__ */
+
+ if (uncompressed_size > 10) {
+ /* Finish any bytes that weren't processed by the vectorized
+ * implementation. */
+ u8 *p8_end = uncompressed_data + uncompressed_size - 10;
+ do {
+ if (*p8 == 0xe8) {
+ undo_call_insn_translation((s32 *)(p8 + 1),
+ p8 - uncompressed_data);
+ p8 += 5;
+ } else {
+ p8++;
+ }
+ } while (p8 < p8_end);
}
}
/*
- * Decompresses a LZX-compressed block of data from which the header has already
+ * Decompresses an LZX-compressed block of data from which the header has already
* been read.
*
* @block_type: The type of the block (LZX_BLOCKTYPE_VERBATIM or
- * LZX_BLOCKTYPE_ALIGNED)
+ * LZX_BLOCKTYPE_ALIGNED)
* @block_size: The size of the block, in bytes.
+ * @num_main_syms: Number of symbols in the main alphabet.
* @window: Pointer to the decompression window.
* @window_pos: The current position in the window. Will be 0 for the first
- * block.
+ * block.
* @tables: The Huffman decoding tables for the block (main, length, and
- * aligned offset, the latter only for LZX_BLOCKTYPE_ALIGNED)
+ * aligned offset, the latter only for LZX_BLOCKTYPE_ALIGNED)
* @queue: The least-recently-used queue for match offsets.
* @istream: The input bitstream for the compressed literals.
*/
static int
lzx_decompress_block(int block_type, unsigned block_size,
+ unsigned num_main_syms,
u8 *window,
unsigned window_pos,
const struct lzx_tables *tables,
- struct lru_queue *queue,
+ struct lzx_lru_queue *queue,
struct input_bitstream *istream)
{
unsigned main_element;
end = window_pos + block_size;
while (window_pos < end) {
ret = read_huffsym_using_maintree(istream, tables,
- &main_element);
- if (ret != 0)
+ &main_element,
+ num_main_syms);
+ if (ret)
return ret;
if (main_element < LZX_NUM_CHARS) {
/* literal: 0 to LZX_NUM_CHARS - 1 */
window[window_pos++] = main_element;
} else {
- /* match: LZX_NUM_CHARS to LZX_MAINTREE_NUM_SYMBOLS - 1 */
+ /* match: LZX_NUM_CHARS to num_main_syms - 1 */
match_len = lzx_decode_match(main_element,
block_type,
end - window_pos,
return 0;
}
-/*
- * Decompresses a block of LZX-compressed data as used in the WIM file format.
- *
- * Note that this will NOT work unmodified for LZX as used in the cabinet
- * format, which is not the same as in the WIM format!
- *
- * @compressed_data: A pointer to the compressed data.
- *
- * @compressed_len: The length of the compressed data, in bytes.
- *
- * @uncompressed_data: A pointer to the buffer into which to write the
- * uncompressed data.
- *
- * @uncompressed_len: The length of the uncompressed data. It must be
- * 32768 bytes or less.
- *
- * Return 0 on success; non-zero on failure.
- */
-int
-lzx_decompress(const void *compressed_data, unsigned compressed_len,
- void *uncompressed_data, unsigned uncompressed_len)
+static int
+lzx_decompress(const void *compressed_data, size_t compressed_size,
+ void *uncompressed_data, size_t uncompressed_size,
+ void *_ctx)
{
- struct lzx_tables tables;
+ struct lzx_decompressor *ctx = _ctx;
struct input_bitstream istream;
- struct lru_queue queue;
+ struct lzx_lru_queue queue;
unsigned window_pos;
unsigned block_size;
unsigned block_type;
int ret;
bool e8_preprocessing_done;
- LZX_DEBUG("lzx_decompress (compressed_data = %p, compressed_len = %d, "
- "uncompressed_data = %p, uncompressed_len = %d).",
- compressed_data, compressed_len,
- uncompressed_data, uncompressed_len);
-
- wimlib_assert(uncompressed_len <= 32768);
+ LZX_DEBUG("compressed_data = %p, compressed_size = %zu, "
+ "uncompressed_data = %p, uncompressed_size = %zu, "
+ "max_window_size=%u).",
+ compressed_data, compressed_size,
+ uncompressed_data, uncompressed_size,
+ ctx->max_window_size);
+
+ if (uncompressed_size > ctx->max_window_size) {
+ LZX_DEBUG("Uncompressed size of %zu exceeds "
+ "window size of %u!",
+ uncompressed_size, ctx->max_window_size);
+ return -1;
+ }
- memset(tables.maintree_lens, 0, sizeof(tables.maintree_lens));
- memset(tables.lentree_lens, 0, sizeof(tables.lentree_lens));
- queue.R0 = 1;
- queue.R1 = 1;
- queue.R2 = 1;
- init_input_bitstream(&istream, compressed_data, compressed_len);
+ memset(ctx->tables.maintree_lens, 0, sizeof(ctx->tables.maintree_lens));
+ memset(ctx->tables.lentree_lens, 0, sizeof(ctx->tables.lentree_lens));
+ lzx_lru_queue_init(&queue);
+ init_input_bitstream(&istream, compressed_data, compressed_size);
e8_preprocessing_done = false; /* Set to true if there may be 0xe8 bytes
in the uncompressed data. */
* blocks. */
for (window_pos = 0;
- window_pos < uncompressed_len;
+ window_pos < uncompressed_size;
window_pos += block_size)
{
LZX_DEBUG("Reading block header.");
- ret = lzx_read_block_header(&istream, &block_size,
- &block_type, &tables, &queue);
- if (ret != 0)
+ ret = lzx_read_block_header(&istream, ctx->num_main_syms,
+ ctx->max_window_size, &block_size,
+ &block_type, &ctx->tables, &queue);
+ if (ret)
return ret;
LZX_DEBUG("block_size = %u, window_pos = %u",
block_size, window_pos);
- if (block_size > uncompressed_len - window_pos) {
- ERROR("lzx_decompress(): Expected a block size of at "
- "most %u bytes (found %u bytes)",
- uncompressed_len - window_pos, block_size);
+ if (block_size > uncompressed_size - window_pos) {
+ LZX_DEBUG("Expected a block size of at "
+ "most %zu bytes (found %u bytes)",
+ uncompressed_size - window_pos, block_size);
return -1;
}
LZX_DEBUG("LZX_BLOCKTYPE_ALIGNED");
ret = lzx_decompress_block(block_type,
block_size,
+ ctx->num_main_syms,
uncompressed_data,
window_pos,
- &tables,
+ &ctx->tables,
&queue,
&istream);
- if (ret != 0)
+ if (ret)
return ret;
- if (tables.maintree_lens[0xe8] != 0)
+
+ if (ctx->tables.maintree_lens[0xe8] != 0)
e8_preprocessing_done = true;
break;
case LZX_BLOCKTYPE_UNCOMPRESSED:
LZX_DEBUG("LZX_BLOCKTYPE_UNCOMPRESSED");
if (istream.data_bytes_left < block_size) {
- ERROR("Unexpected end of input when "
- "reading %u bytes from LZX bitstream "
- "(only have %u bytes left)",
- block_size, istream.data_bytes_left);
+ LZX_DEBUG("Unexpected end of input when "
+ "reading %u bytes from LZX bitstream "
+ "(only have %u bytes left)",
+ block_size, istream.data_bytes_left);
return -1;
}
memcpy(&((u8*)uncompressed_data)[window_pos], istream.data,
}
}
if (e8_preprocessing_done)
- undo_call_insn_preprocessing(uncompressed_data, uncompressed_len);
+ undo_call_insn_preprocessing(uncompressed_data, uncompressed_size);
return 0;
}
+
+static void
+lzx_free_decompressor(void *_ctx)
+{
+ struct lzx_decompressor *ctx = _ctx;
+
+ FREE(ctx);
+}
+
+static int
+lzx_create_decompressor(size_t max_window_size,
+ const struct wimlib_decompressor_params_header *params,
+ void **ctx_ret)
+{
+ struct lzx_decompressor *ctx;
+
+ if (!lzx_window_size_valid(max_window_size))
+ return WIMLIB_ERR_INVALID_PARAM;
+
+ ctx = MALLOC(sizeof(struct lzx_decompressor));
+ if (ctx == NULL)
+ return WIMLIB_ERR_NOMEM;
+
+ ctx->max_window_size = max_window_size;
+ ctx->num_main_syms = lzx_get_num_main_syms(max_window_size);
+
+ *ctx_ret = ctx;
+ return 0;
+}
+
+const struct decompressor_ops lzx_decompressor_ops = {
+ .create_decompressor = lzx_create_decompressor,
+ .decompress = lzx_decompress,
+ .free_decompressor = lzx_free_decompressor,
+};
git://wimlib.net / wimlib / blobdiff