*/
/*
- * Copyright (C) 2012 Eric Biggers
+ * Copyright (C) 2012, 2013 Eric Biggers
*
* This file is part of wimlib, a library for working with WIM files.
*
#include <string.h>
/*
- * Builds a fast huffman decoding table from an array that gives the length of
- * the codeword for each symbol in the alphabet. Originally based on code
- * written by David Tritscher (taken the original LZX decompression code); also
- * heavily modified to add some optimizations used in the zlib code, as well as
- * more comments.
+ * make_huffman_decode_table: - Builds a fast huffman decoding table from an
+ * array that gives the length of the codeword for each symbol in the alphabet.
+ * Originally based on code written by David Tritscher (taken the original LZX
+ * decompression code); also heavily modified to add some optimizations used in
+ * the zlib code, as well as more comments.
*
* @decode_table: The array in which to create the fast huffman decoding
* table. It must have a length of at least
* (2**table_bits) + 2 * num_syms to guarantee
* that there is enough space.
*
- * @num_syms: Total number of symbols in the Huffman tree.
+ * @num_syms: Number of symbols in the alphabet, including symbols
+ * that do not appear in this particular input chunk.
*
* @table_bits: Any symbols with a code length of table_bits or less can
* be decoded in one lookup of the table. 2**table_bits
* any Huffman codes longer than @table_bits.
*
* @lens: An array of length @num_syms, indexable by symbol, that
- * gives the length of the Huffman codeward for that
+ * gives the length of the Huffman codeword for that
* symbol. Because the Huffman tree is in canonical form,
* it can be reconstructed by only knowing the length of
* the codeword for each symbol. It is assumed, but not
break;
unsigned num_entries = 1 << (table_bits - codeword_len);
- if (num_entries >=
- (sizeof(unsigned long) / sizeof(decode_table[0])))
- {
- wimlib_assert2(decode_table_pos % 4 == 0);
+ const unsigned entries_per_long = sizeof(unsigned long) /
+ sizeof(decode_table[0]);
+ if (num_entries >= entries_per_long) {
+ /* Fill in the Huffman decode table entries one unsigned
+ * long at a time. On 32-bit systems this is 2 entries
+ * per store, while on 64-bit systems this is 4 entries
+ * per store. */
+ wimlib_assert2(decode_table_pos % entries_per_long == 0);
BUILD_BUG_ON(sizeof(unsigned long) != 4 &&
sizeof(unsigned long) != 8);
unsigned long *p = (unsigned long *)&decode_table[decode_table_pos];
- unsigned long n = num_entries /
- (sizeof(unsigned long) /
- sizeof(decode_table[0]));
+ unsigned n = num_entries / entries_per_long;
unsigned long v = sym;
if (sizeof(unsigned long) >= 4)
v |= v << 16;
- if (sizeof(unsigned long) >= 8)
+ if (sizeof(unsigned long) >= 8) {
+ /* This may produce a compiler warning if an
+ * unsigned long is 32 bits, but this won't be
+ * executed unless an unsigned long is at least
+ * 64 bits anyway. */
v |= v << 32;
+ }
do {
*p++ = v;
} while (--n);
decode_table_pos += num_entries;
} else {
+ /* Fill in the Huffman decode table entries one 16-bit
+ * integer at a time. */
do {
decode_table[decode_table_pos++] = sym;
} while (--num_entries);
unsigned sym = sorted_syms[i];
unsigned codeword_len = lens[sym];
unsigned extra_bits = codeword_len - table_bits;
- unsigned extra_mask;
cur_codeword <<= (codeword_len - prev_codeword_len);
prev_codeword_len = codeword_len;
return 0;
}
-/* Reads a Huffman-encoded symbol when it is known there are less than
- * MAX_CODE_LEN bits remaining in the bitstream. */
+/* Reads a Huffman-encoded symbol from the bistream when the number of remaining
+ * bits is less than the maximum codeword length. */
int read_huffsym_near_end_of_input(struct input_bitstream *istream,
const u16 decode_table[],
const u8 lens[],