X-Git-Url: https://wimlib.net/git/?p=wimlib;a=blobdiff_plain;f=include%2Fwimlib%2Flzx.h;h=5bd6b1406a3b6e9cdf3e7e54232adc8e829cdf75;hp=b6d1ec237ec78d1ec9a4b3383d3707125afeccb5;hb=2f79af1ffd1f92db74c69117f0c2401826126b92;hpb=9bec19341aac9190cd0bd80a6937b02d923dd566 diff --git a/include/wimlib/lzx.h b/include/wimlib/lzx.h index b6d1ec23..5bd6b140 100644 --- a/include/wimlib/lzx.h +++ b/include/wimlib/lzx.h @@ -1,120 +1,63 @@ +/* + * lzx.h + * + * Declarations shared between LZX compression and decompression. + */ + #ifndef _WIMLIB_LZX_H #define _WIMLIB_LZX_H #include "wimlib/assert.h" +#include "wimlib/bitops.h" +#include "wimlib/compiler.h" +#include "wimlib/lzx_constants.h" +#include "wimlib/util.h" #include "wimlib/types.h" //#define ENABLE_LZX_DEBUG #ifdef ENABLE_LZX_DEBUG -# define LZX_DEBUG DEBUG # define LZX_ASSERT wimlib_assert #else -# define LZX_DEBUG(format, ...) # define LZX_ASSERT(...) #endif -/* Constants, most of which are defined by the LZX specification: */ - -/* The smallest and largest allowed match lengths. */ -#define LZX_MIN_MATCH_LEN 2 -#define LZX_MAX_MATCH_LEN 257 - -/* Number of values an uncompressed literal byte can represent. */ -#define LZX_NUM_CHARS 256 - -/* Each LZX block begins with 3 bits that determines the block type. Below are - * the valid block types. Values 0, and 4 through 7, are invalid. */ -#define LZX_BLOCKTYPE_VERBATIM 1 -#define LZX_BLOCKTYPE_ALIGNED 2 -#define LZX_BLOCKTYPE_UNCOMPRESSED 3 - -#define LZX_NUM_PRIMARY_LENS 7 /* this one missing from spec! */ +extern const u32 lzx_offset_slot_base[LZX_MAX_OFFSET_SLOTS]; -/* NOTE: There are really 51 position slots in the LZX format as a whole, but - * only 30 are needed to allow for the window to be up to 32768 bytes long, - * which is the maximum in the WIM format. */ -#define LZX_NUM_POSITION_SLOTS 30 +extern const u8 lzx_extra_offset_bits[LZX_MAX_OFFSET_SLOTS]; -/* Read the LZX specification for information about the Huffman trees used in - * the LZX compression format. Basically there are 4 of them: The main tree, - * the length tree, the pre tree, and the aligned tree. The main tree and - * length tree are given at the beginning of VERBATIM and ALIGNED blocks as a - * list of *_NUM_SYMBOLS code length values. They are read using the - * read_code_lens() function and built using the make_decode_table() function. - * The decode table is not a real tree but rather a table that we can index by - * some number of bits (*_TABLEBITS) of the input to quickly look up the symbol - * corresponding to a Huffman code. +/* Returns the LZX offset slot that corresponds to a given adjusted offset. * - * The ALIGNED tree is only present on ALIGNED blocks. + * Logically, this returns the smallest i such that + * adjusted_offset >= lzx_offset_slot_base[i]. * - * A PRECODE is used to encode the code lengths for the main tree and the length - * tree. There is a separate pretree for each half of the main tree. */ - -#define LZX_MAINCODE_NUM_SYMBOLS (LZX_NUM_CHARS + \ - (LZX_NUM_POSITION_SLOTS << 3)) -#define LZX_MAINCODE_TABLEBITS 11 - -#define LZX_LENCODE_NUM_SYMBOLS 249 -#define LZX_LENCODE_TABLEBITS 10 - -#define LZX_PRECODE_NUM_SYMBOLS 20 -#define LZX_PRECODE_TABLEBITS 6 -#define LZX_PRECODE_ELEMENT_SIZE 4 - -#define LZX_ALIGNEDCODE_NUM_SYMBOLS 8 -#define LZX_ALIGNEDCODE_TABLEBITS 7 -#define LZX_ALIGNEDCODE_ELEMENT_SIZE 3 - -/* Maximum allowed length of Huffman codewords. */ -#define LZX_MAX_MAIN_CODEWORD_LEN 16 -#define LZX_MAX_LEN_CODEWORD_LEN 16 -#define LZX_MAX_PRE_CODEWORD_LEN 16 -#define LZX_MAX_ALIGNED_CODEWORD_LEN 8 - -/* For the LZX-compressed blocks in WIM files, this value is always used as the - * filesize parameter for the call instruction (0xe8 byte) preprocessing, even - * though the blocks themselves are not this size, and the size of the actual - * file resource in the WIM file is very likely to be something entirely - * different as well. */ -#define LZX_WIM_MAGIC_FILESIZE 12000000 - -#define LZX_BLOCKTYPE_NBITS 3 -#define LZX_BLOCKSIZE_NBITS 16 - -#define USE_LZX_EXTRA_BITS_ARRAY - -#ifdef USE_LZX_EXTRA_BITS_ARRAY -extern const u8 lzx_extra_bits[]; -#endif - -/* Given the number of a LZX position slot, return the number of extra bits that - * are needed to encode the match offset. */ + * The actual implementation below takes advantage of the regularity of the + * numbers in the lzx_offset_slot_base array to calculate the slot directly from + * the adjusted offset without actually looking at the array. + */ static inline unsigned -lzx_get_num_extra_bits(unsigned position_slot) +lzx_get_offset_slot_raw(u32 adjusted_offset) { -#ifdef USE_LZX_EXTRA_BITS_ARRAY - /* Use a table */ - return lzx_extra_bits[position_slot]; -#else - /* Calculate directly using a shift and subtraction. */ - LZX_ASSERT(position_slot >= 2 && position_slot <= 37); - return (position_slot >> 1) - 1; -#endif + if (adjusted_offset >= 196608) { + return (adjusted_offset >> 17) + 34; + } else { + LZX_ASSERT(2 <= adjusted_offset && adjusted_offset < 655360); + unsigned mssb_idx = fls32(adjusted_offset); + return (mssb_idx << 1) | + ((adjusted_offset >> (mssb_idx - 1)) & 1); + } } -extern const u32 lzx_position_base[]; +extern unsigned lzx_get_window_order(size_t max_block_size); -#define LZX_NUM_RECENT_OFFSETS 3 +extern unsigned lzx_get_num_main_syms(unsigned window_order); /* Least-recently used queue for match offsets. */ struct lzx_lru_queue { u32 R[LZX_NUM_RECENT_OFFSETS]; -}; - -/* In the LZX format, an offset of n bytes is actually encoded - * as (n + LZX_OFFSET_OFFSET). */ -#define LZX_OFFSET_OFFSET (LZX_NUM_RECENT_OFFSETS - 1) +} _aligned_attribute(sizeof(unsigned long)); +/* Initialize the LZX least-recently-used match offset queue at the beginning of + * a new window for either decompression or compression. */ static inline void lzx_lru_queue_init(struct lzx_lru_queue *queue) { @@ -122,4 +65,10 @@ lzx_lru_queue_init(struct lzx_lru_queue *queue) queue->R[i] = 1; } +extern void +lzx_do_e8_preprocessing(u8 *data, u32 size); + +extern void +lzx_undo_e8_preprocessing(u8 *data, u32 size); + #endif /* _WIMLIB_LZX_H */