X-Git-Url: https://wimlib.net/git/?p=wimlib;a=blobdiff_plain;f=src%2Flzx-compress.c;h=cdbb63699004edbed7066f9344f6d1c6f30207d2;hp=04e5fbfb6fe2f1d3e51ffac9cb8ad4d03b4ad638;hb=10be6c712f4f5f066458dbab11e81f8e64cd2324;hpb=4ecf344e77e1f5891055881950a6e89e32b16008 diff --git a/src/lzx-compress.c b/src/lzx-compress.c index 04e5fbfb..cdbb6369 100644 --- a/src/lzx-compress.c +++ b/src/lzx-compress.c @@ -1,11 +1,9 @@ /* * lzx-compress.c - * - * LZX compression routines */ /* - * 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. * @@ -25,163 +23,208 @@ /* - * This file contains a compressor for the LZX compression format, as used in - * the WIM file format. + * This file contains a compressor for the LZX ("Lempel-Ziv eXtended"?) + * compression format, as used in the WIM (Windows IMaging) file format. This + * code may need some slight modifications to be used outside of the WIM format. + * In particular, in other situations the LZX block header might be slightly + * different, and a sliding window rather than a fixed-size window might be + * required. * - * Format - * ====== + * ---------------------------------------------------------------------------- * - * First, the primary reference for the LZX compression format is the - * specification released by Microsoft. + * Format Overview * - * Second, the comments in lzx-decompress.c provide some more information about - * the LZX compression format, including errors in the Microsoft specification. + * The primary reference for LZX is the specification released by Microsoft. + * However, the comments in lzx-decompress.c provide more information about LZX + * and note some errors in the Microsoft specification. * - * Do note that LZX shares many similarities with DEFLATE, the algorithm used by - * zlib and gzip. Both LZX and DEFLATE use LZ77 matching and Huffman coding, - * and certain other details are quite similar, such as the method for storing - * Huffman codes. However, some of the main differences are: + * LZX shares many similarities with DEFLATE, the format used by zlib and gzip. + * Both LZX and DEFLATE use LZ77 matching and Huffman coding. Certain details + * are quite similar, such as the method for storing Huffman codes. However, + * the main differences are: * * - LZX preprocesses the data to attempt to make x86 machine code slightly more * compressible before attempting to compress it further. + * * - LZX uses a "main" alphabet which combines literals and matches, with the * match symbols containing a "length header" (giving all or part of the match * length) and a "position slot" (giving, roughly speaking, the order of * magnitude of the match offset). - * - LZX does not have static Huffman blocks; however it does have two types of - * dynamic Huffman blocks ("aligned offset" and "verbatim"). + * + * - LZX does not have static Huffman blocks (that is, the kind with preset + * Huffman codes); however it does have two types of dynamic Huffman blocks + * ("verbatim" and "aligned"). + * * - LZX has a minimum match length of 2 rather than 3. + * * - In LZX, match offsets 0 through 2 actually represent entries in an LRU * queue of match offsets. This is very useful for certain types of files, * such as binary files that have repeating records. * - * Algorithms - * ========== - * - * There are actually two distinct overall algorithms implemented here. We - * shall refer to them as the "slow" algorithm and the "fast" algorithm. The - * "slow" algorithm spends more time compressing to achieve a higher compression - * ratio compared to the "fast" algorithm. More details are presented below. - * - * Slow algorithm - * -------------- - * - * The "slow" algorithm to generate LZX-compressed data is roughly as follows: - * - * 1. Preprocess the input data to translate the targets of x86 call - * instructions to absolute offsets. - * - * 2. Build the suffix array and inverse suffix array for the input data. The - * suffix array contains the indices of all suffixes of the input data, - * sorted lexcographically by the corresponding suffixes. The "position" of - * a suffix is the index of that suffix in the original string, whereas the - * "rank" of a suffix is the index at which that suffix's position is found - * in the suffix array. - * - * 3. Build the longest common prefix array corresponding to the suffix array. - * - * 4. For each suffix, find the highest lower ranked suffix that has a lower - * position, the lowest higher ranked suffix that has a lower position, and - * the length of the common prefix shared between each. This information is - * later used to link suffix ranks into a doubly-linked list for searching - * the suffix array. - * - * 5. Set a default cost model for matches/literals. - * - * 6. Determine the lowest cost sequence of LZ77 matches ((offset, length) - * pairs) and literal bytes to divide the input into. Raw match-finding is - * done by searching the suffix array using a linked list to avoid - * considering any suffixes that start after the current position. Each run - * of the match-finder returns the approximate lowest-cost longest match as - * well as any shorter matches that have even lower approximate costs. Each - * such run also adds the suffix rank of the current position into the linked - * list being used to search the suffix array. Parsing, or match-choosing, - * is solved as a minimum-cost path problem using a forward "optimal parsing" - * algorithm based on the Deflate encoder from 7-Zip. This algorithm moves - * forward calculating the minimum cost to reach each byte until either a - * very long match is found or until a position is found at which no matches - * start or overlap. - * - * 7. Build the Huffman codes needed to output the matches/literals. - * - * 8. Up to a certain number of iterations, use the resulting Huffman codes to - * refine a cost model and go back to Step #6 to determine an improved - * sequence of matches and literals. - * - * 9. Output the resulting block using the match/literal sequences and the - * Huffman codes that were computed for the block. - * - * Note: the algorithm does not yet attempt to split the input into multiple LZX - * blocks, instead using a series of blocks of LZX_DIV_BLOCK_SIZE bytes. - * - * Fast algorithm - * -------------- - * - * The fast algorithm (and the only one available in wimlib v1.5.1 and earlier) - * spends much less time on the main bottlenecks of the compression process --- - * that is, the match finding and match choosing. Matches are found and chosen - * with hash chains using a greedy parse with one position of look-ahead. No - * block splitting is done; only compressing the full input into an aligned - * offset block is considered. - * - * API - * === - * - * The old API (retained for backward compatibility) consists of just one - * function: - * - * wimlib_lzx_compress() - * - * The new compressor has more potential parameters and needs more memory, so - * the new API ties up memory allocations and compression parameters into a - * context: - * - * wimlib_lzx_alloc_context() - * wimlib_lzx_compress2() - * wimlib_lzx_free_context() - * wimlib_lzx_set_default_params() - * - * Both wimlib_lzx_compress() and wimlib_lzx_compress2() are designed to - * compress an in-memory buffer of up to the window size, which can be any power - * of two between 2^15 and 2^21 inclusively. However, by default, the WIM - * format uses 2^15, and this is seemingly the only value that is compatible - * with WIMGAPI. In any case, the window is not a true "sliding window" since - * no data is ever "slid out" of the window. This is needed for the WIM format, - * which is designed such that chunks may be randomly accessed. - * - * Both wimlib_lzx_compress() and wimlib_lzx_compress2() return 0 if the data - * could not be compressed to less than the size of the uncompressed data. - * Again, this is suitable for the WIM format, which stores such data chunks - * uncompressed. - * - * The functions in this LZX compression API are exported from the library, - * although with the possible exception of wimlib_lzx_set_default_params(), this - * is only in case other programs happen to have uses for it other than WIM - * reading/writing as already handled through the rest of the library. - * - * Acknowledgments - * =============== - * - * Acknowledgments to several open-source projects and research papers that made - * it possible to implement this code: - * - * - divsufsort (author: Yuta Mori), for the suffix array construction code, - * located in a separate directory (divsufsort/). - * - * - "Linear-Time Longest-Common-Prefix Computation in Suffix Arrays and Its - * Applications" (Kasai et al. 2001), for the LCP array computation. - * - * - "LPF computation revisited" (Crochemore et al. 2009) for the prev and next - * array computations. - * - * - 7-Zip (author: Igor Pavlov) for the algorithm for forward optimal parsing - * (match-choosing). - * - * - zlib (author: Jean-loup Gailly and Mark Adler), for the hash table - * match-finding algorithm (used in lz77.c). - * - * - lzx-compress (author: Matthew T. Russotto), on which some parts of this - * code were originally based. + * ---------------------------------------------------------------------------- + * + * Algorithmic Overview + * + * At a high level, any implementation of LZX compression must operate as + * follows: + * + * 1. Preprocess the input data to translate the targets of 32-bit x86 call + * instructions to absolute offsets. (Actually, this is required for WIM, + * but might not be in other places LZX is used.) + * + * 2. Find a sequence of LZ77-style matches and literal bytes that expands to + * the preprocessed data. + * + * 3. Divide the match/literal sequence into one or more LZX blocks, each of + * which may be "uncompressed", "verbatim", or "aligned". + * + * 4. Output each LZX block. + * + * Step (1) is fairly straightforward. It requires looking for 0xe8 bytes in + * the input data and performing a translation on the 4 bytes following each + * one. + * + * Step (4) is complicated, but it is mostly determined by the LZX format. The + * only real choice we have is what algorithm to use to build the length-limited + * canonical Huffman codes. See lzx_write_all_blocks() for details. + * + * That leaves steps (2) and (3) as where all the hard stuff happens. Focusing + * on step (2), we need to do LZ77-style parsing on the input data, or "window", + * to divide it into a sequence of matches and literals. Each position in the + * window might have multiple matches associated with it, and we need to choose + * which one, if any, to actually use. Therefore, the problem can really be + * divided into two areas of concern: (a) finding matches at a given position, + * which we shall call "match-finding", and (b) choosing whether to use a + * match or a literal at a given position, and if using a match, which one (if + * there is more than one available). We shall call this "match-choosing". We + * first consider match-finding, then match-choosing. + * + * ---------------------------------------------------------------------------- + * + * Match-finding + * + * Given a position in the window, we want to find LZ77-style "matches" with + * that position at previous positions in the window. With LZX, the minimum + * match length is 2 and the maximum match length is 257. The only restriction + * on offsets is that LZX does not allow the last 2 bytes of the window to match + * the the beginning of the window. + * + * Depending on how good a compression ratio we want (see the "Match-choosing" + * section), we may want to find: (a) all matches, or (b) just the longest + * match, or (c) just some "promising" matches that we are able to find quickly, + * or (d) just the longest match that we're able to find quickly. Below we + * introduce the match-finding methods that the code currently uses or has + * previously used: + * + * - Hash chains. Maintain a table that maps hash codes, computed from + * fixed-length byte sequences, to linked lists containing previous window + * positions. To search for matches, compute the hash for the current + * position in the window and search the appropriate hash chain. When + * advancing to the next position, prepend the current position to the + * appropriate hash list. This is a good approach for producing matches with + * stategy (d) and is useful for fast compression. Therefore, we provide an + * option to use this method for LZX compression. See lz_hash.c for the + * implementation. + * + * - Binary trees. Similar to hash chains, but each hash bucket contains a + * binary tree of previous window positions rather than a linked list. This + * is a good approach for producing matches with stategy (c) and is useful for + * achieving a good compression ratio. Therefore, we provide an option to use + * this method; see lz_bt.c for the implementation. + * + * - Suffix arrays. This code previously used this method to produce matches + * with stategy (c), but I've dropped it because it was slower than the binary + * trees approach, used more memory, and did not improve the compression ratio + * enough to compensate. Download wimlib v1.6.2 if you want the code. + * However, the suffix array method was basically as follows. Build the + * suffix array for the entire window. The suffix array contains each + * possible window position, sorted by the lexicographic order of the strings + * that begin at those positions. Find the matches at a given position by + * searching the suffix array outwards, in both directions, from the suffix + * array slot for that position. This produces the longest matches first, but + * "matches" that actually occur at later positions in the window must be + * skipped. To do this skipping, use an auxiliary array with dynamically + * constructed linked lists. Also, use the inverse suffix array to quickly + * find the suffix array slot for a given position without doing a binary + * search. + * + * ---------------------------------------------------------------------------- + * + * Match-choosing + * + * Usually, choosing the longest match is best because it encodes the most data + * in that one item. However, sometimes the longest match is not optimal + * because (a) choosing a long match now might prevent using an even longer + * match later, or (b) more generally, what we actually care about is the number + * of bits it will ultimately take to output each match or literal, which is + * actually dependent on the entropy encoding using by the underlying + * compression format. Consequently, a longer match usually, but not always, + * takes fewer bits to encode than multiple shorter matches or literals that + * cover the same data. + * + * This problem of choosing the truly best match/literal sequence is probably + * impossible to solve efficiently when combined with entropy encoding. If we + * knew how many bits it takes to output each match/literal, then we could + * choose the optimal sequence using shortest-path search a la Dijkstra's + * algorithm. However, with entropy encoding, the chosen match/literal sequence + * affects its own encoding. Therefore, we can't know how many bits it will + * take to actually output any one match or literal until we have actually + * chosen the full sequence of matches and literals. + * + * Notwithstanding the entropy encoding problem, we also aren't guaranteed to + * choose the optimal match/literal sequence unless the match-finder (see + * section "Match-finder") provides the match-chooser with all possible matches + * at each position. However, this is not computationally efficient. For + * example, there might be many matches of the same length, and usually (but not + * always) the best choice is the one with the smallest offset. So in practice, + * it's fine to only consider the smallest offset for a given match length at a + * given position. (Actually, for LZX, it's also worth considering repeat + * offsets.) + * + * In addition, as mentioned earlier, in LZX we have the choice of using + * multiple blocks, each of which resets the Huffman codes. This expands the + * search space even further. Therefore, to simplify the problem, we currently + * we don't attempt to actually choose the LZX blocks based on the data. + * Instead, we just divide the data into fixed-size blocks of LZX_DIV_BLOCK_SIZE + * bytes each, and always use verbatim or aligned blocks (never uncompressed). + * A previous version of this code recursively split the input data into + * equal-sized blocks, up to a maximum depth, and chose the lowest-cost block + * divisions. However, this made compression much slower and did not actually + * help very much. It remains an open question whether a sufficiently fast and + * useful block-splitting algorithm is possible for LZX. Essentially the same + * problem also applies to DEFLATE. The Microsoft LZX compressor seemingly does + * do block splitting, although I don't know how fast or useful it is, + * specifically. + * + * Now, back to the entropy encoding problem. The "solution" is to use an + * iterative approach to compute a good, but not necessarily optimal, + * match/literal sequence. Start with a fixed assignment of symbol costs and + * choose an "optimal" match/literal sequence based on those costs, using + * shortest-path seach a la Dijkstra's algorithm. Then, for each iteration of + * the optimization, update the costs based on the entropy encoding of the + * current match/literal sequence, then choose a new match/literal sequence + * based on the updated costs. Usually, the actual cost to output the current + * match/literal sequence will decrease in each iteration until it converges on + * a fixed point. This result may not be the truly optimal match/literal + * sequence, but it usually is much better than one chosen by doing a "greedy" + * parse where we always chooe the longest match. + * + * An alternative to both greedy parsing and iterative, near-optimal parsing is + * "lazy" parsing. Briefly, "lazy" parsing considers just the longest match at + * each position, but it waits to choose that match until it has also examined + * the next position. This is actually a useful approach; it's used by zlib, + * for example. Therefore, for fast compression we combine lazy parsing with + * the hash chain max-finder. For normal/high compression we combine + * near-optimal parsing with the binary tree match-finder. + * + * Anyway, if you've read through this comment, you hopefully should have a + * better idea of why things are done in a certain way in this LZX compressor, + * as well as in other compressors for LZ77-based formats (including third-party + * ones). In my opinion, the phrase "compression algorithm" is often mis-used + * in place of "compression format", since there can be many different + * algorithms that all generate compressed data in the same format. The + * challenge is to design an algorithm that is efficient but still gives a good + * compression ratio. */ #ifdef HAVE_CONFIG_H @@ -189,34 +232,36 @@ #endif #include "wimlib.h" -#include "wimlib/compress.h" +#include "wimlib/compressor_ops.h" +#include "wimlib/compress_common.h" +#include "wimlib/endianness.h" #include "wimlib/error.h" +#include "wimlib/lz.h" +#include "wimlib/lz_hash.h" +#include "wimlib/lz_bt.h" #include "wimlib/lzx.h" #include "wimlib/util.h" -#include -#include #include #ifdef ENABLE_LZX_DEBUG -# include "wimlib/decompress.h" +# include "wimlib/decompress_common.h" #endif -#include "divsufsort/divsufsort.h" - -typedef u32 block_cost_t; -#define INFINITE_BLOCK_COST ((block_cost_t)~0U) - #define LZX_OPTIM_ARRAY_SIZE 4096 #define LZX_DIV_BLOCK_SIZE 32768 -#define LZX_MAX_CACHE_PER_POS 10 +#define LZX_CACHE_PER_POS 8 + +#define LZX_CACHE_LEN (LZX_DIV_BLOCK_SIZE * (LZX_CACHE_PER_POS + 1)) +#define LZX_CACHE_SIZE (LZX_CACHE_LEN * sizeof(struct lz_match)) +#define LZX_MAX_MATCHES_PER_POS (LZX_MAX_MATCH_LEN - LZX_MIN_MATCH_LEN + 1) /* Codewords for the LZX main, length, and aligned offset Huffman codes */ struct lzx_codewords { - u16 main[LZX_MAINCODE_MAX_NUM_SYMBOLS]; - u16 len[LZX_LENCODE_NUM_SYMBOLS]; - u16 aligned[LZX_ALIGNEDCODE_NUM_SYMBOLS]; + u32 main[LZX_MAINCODE_MAX_NUM_SYMBOLS]; + u32 len[LZX_LENCODE_NUM_SYMBOLS]; + u32 aligned[LZX_ALIGNEDCODE_NUM_SYMBOLS]; }; /* Codeword lengths (in bits) for the LZX main, length, and aligned offset @@ -234,7 +279,7 @@ struct lzx_lens { * * If a codeword has zero frequency, it must still be assigned some nonzero cost * --- generally a high cost, since even if it gets used in the next iteration, - * it probably will not be used very times. */ + * it probably will not be used very many times. */ struct lzx_costs { u8 main[LZX_MAINCODE_MAX_NUM_SYMBOLS]; u8 len[LZX_LENCODE_NUM_SYMBOLS]; @@ -249,13 +294,13 @@ struct lzx_codes { /* Tables for tallying symbol frequencies in the three LZX alphabets */ struct lzx_freqs { - freq_t main[LZX_MAINCODE_MAX_NUM_SYMBOLS]; - freq_t len[LZX_LENCODE_NUM_SYMBOLS]; - freq_t aligned[LZX_ALIGNEDCODE_NUM_SYMBOLS]; + u32 main[LZX_MAINCODE_MAX_NUM_SYMBOLS]; + u32 len[LZX_LENCODE_NUM_SYMBOLS]; + u32 aligned[LZX_ALIGNEDCODE_NUM_SYMBOLS]; }; /* LZX intermediate match/literal format */ -struct lzx_match { +struct lzx_item { /* Bit Description * * 31 1 if a match, 0 if a literal. @@ -272,18 +317,6 @@ struct lzx_match { u32 data; }; -/* Raw LZ match/literal format: just a length and offset. - * - * The length is the number of bytes of the match, and the offset is the number - * of bytes back in the input the match is from the current position. - * - * If @len < LZX_MIN_MATCH_LEN, then it's really just a literal byte and @offset is - * meaningless. */ -struct raw_match { - u16 len; - input_idx_t offset; -}; - /* Specification for an LZX block. */ struct lzx_block_spec { @@ -292,100 +325,26 @@ struct lzx_block_spec { int block_type; /* 0-based position in the window at which this block starts. */ - input_idx_t window_pos; + u32 window_pos; /* The number of bytes of uncompressed data this block represents. */ - input_idx_t block_size; + u32 block_size; - /* The position in the 'chosen_matches' array in the `struct - * lzx_compressor' at which the match/literal specifications for - * this block begin. */ - input_idx_t chosen_matches_start_pos; + /* The match/literal sequence for this block. */ + struct lzx_item *chosen_items; - /* The number of match/literal specifications for this block. */ - input_idx_t num_chosen_matches; + /* The length of the @chosen_items sequence. */ + u32 num_chosen_items; /* Huffman codes for this block. */ struct lzx_codes codes; }; -/* - * An array of these structures is used during the match-choosing algorithm. - * They correspond to consecutive positions in the window and are used to keep - * track of the cost to reach each position, and the match/literal choices that - * need to be chosen to reach that position. - */ -struct lzx_optimal { - /* The approximate minimum cost, in bits, to reach this position in the - * window which has been found so far. */ - block_cost_t cost; - - /* The union here is just for clarity, since the fields are used in two - * slightly different ways. Initially, the @prev structure is filled in - * first, and links go from later in the window to earlier in the - * window. Later, @next structure is filled in and links go from - * earlier in the window to later in the window. */ - union { - struct { - /* Position of the start of the match or literal that - * was taken to get to this position in the approximate - * minimum-cost parse. */ - input_idx_t link; - - /* Offset (as in an LZ (length, offset) pair) of the - * match or literal that was taken to get to this - * position in the approximate minimum-cost parse. */ - input_idx_t match_offset; - } prev; - struct { - /* Position at which the match or literal starting at - * this position ends in the minimum-cost parse. */ - input_idx_t link; - - /* Offset (as in an LZ (length, offset) pair) of the - * match or literal starting at this position in the - * approximate minimum-cost parse. */ - input_idx_t match_offset; - } next; - }; - - /* The match offset LRU queue that will exist when the approximate - * minimum-cost path to reach this position is taken. */ - struct lzx_lru_queue queue; -}; - -/* Suffix array link */ -struct salink { - /* Rank of highest ranked suffix that has rank lower than the suffix - * corresponding to this structure and either has a lower position - * (initially) or has a position lower than the highest position at - * which matches have been searched for so far, or -1 if there is no - * such suffix. */ - input_idx_t prev; - - /* Rank of lowest ranked suffix that has rank greater than the suffix - * corresponding to this structure and either has a lower position - * (intially) or has a position lower than the highest position at which - * matches have been searched for so far, or -1 if there is no such - * suffix. */ - input_idx_t next; - - /* Length of longest common prefix between the suffix corresponding to - * this structure and the suffix with rank @prev, or 0 if @prev is -1. - */ - input_idx_t lcpprev; - - /* Length of longest common prefix between the suffix corresponding to - * this structure and the suffix with rank @next, or 0 if @next is -1. - */ - input_idx_t lcpnext; -}; - /* State of the LZX compressor. */ struct lzx_compressor { /* The parameters that were used to create the compressor. */ - struct wimlib_lzx_params params; + struct wimlib_lzx_compressor_params params; /* The buffer of data to be compressed. * @@ -398,16 +357,16 @@ struct lzx_compressor { * chunks. * * We reserve a few extra bytes to potentially allow reading off the end - * of the array in the match-finding code for optimization purposes. - */ + * of the array in the match-finding code for optimization purposes + * (currently only needed for the hash chain match-finder). */ u8 *window; /* Number of bytes of data to be compressed, which is the number of * bytes of data in @window that are actually valid. */ - input_idx_t window_size; + u32 window_size; /* Allocated size of the @window. */ - input_idx_t max_window_size; + u32 max_window_size; /* Number of symbols in the main alphabet (depends on the * @max_window_size since it determines the maximum allowed offset). */ @@ -418,7 +377,7 @@ struct lzx_compressor { /* Space for the sequences of matches/literals that were chosen for each * block. */ - struct lzx_match *chosen_matches; + struct lzx_item *chosen_items; /* Information about the LZX blocks the preprocessed input was divided * into. */ @@ -438,77 +397,96 @@ struct lzx_compressor { struct lzx_costs costs; /* Fast algorithm only: Array of hash table links. */ - input_idx_t *prev_tab; + u32 *prev_tab; - /* Suffix array for window. - * This is a mapping from suffix rank to suffix position. */ - input_idx_t *SA; - - /* Inverse suffix array for window. - * This is a mapping from suffix position to suffix rank. - * If 0 <= r < window_size, then ISA[SA[r]] == r. */ - input_idx_t *ISA; - - /* Longest common prefix array corresponding to the suffix array SA. - * LCP[i] is the length of the longest common prefix between the - * suffixes with positions SA[i - 1] and SA[i]. LCP[0] is undefined. - */ - input_idx_t *LCP; - - /* Suffix array links. - * - * During a linear scan of the input string to find matches, this array - * used to keep track of which rank suffixes in the suffix array appear - * before the current position. Instead of searching in the original - * suffix array, scans for matches at a given position traverse a linked - * list containing only suffixes that appear before that position. */ - struct salink *salink; + /* Slow algorithm only: Binary tree match-finder. */ + struct lz_bt mf; /* Position in window of next match to return. */ - input_idx_t match_window_pos; + u32 match_window_pos; - /* The match-finder shall ensure the length of matches does not exceed - * this position in the input. */ - input_idx_t match_window_end; + /* The end-of-block position. We can't allow any matches to span this + * position. */ + u32 match_window_end; /* Matches found by the match-finder are cached in the following array * to achieve a slight speedup when the same matches are needed on * subsequent passes. This is suboptimal because different matches may * be preferred with different cost models, but seems to be a worthwhile * speedup. */ - struct raw_match *cached_matches; - unsigned cached_matches_pos; + struct lz_match *cached_matches; + struct lz_match *cache_ptr; bool matches_cached; + struct lz_match *cache_limit; - /* Slow algorithm only: Temporary space used for match-choosing - * algorithm. - * - * The size of this array must be at least LZX_MAX_MATCH_LEN but - * otherwise is arbitrary. More space simply allows the match-choosing - * algorithm to potentially find better matches (depending on the input, - * as always). */ - struct lzx_optimal *optimum; - - /* Slow algorithm only: Variables used by the match-choosing algorithm. - * + /* Match-chooser state. * When matches have been chosen, optimum_cur_idx is set to the position * in the window of the next match/literal to return and optimum_end_idx * is set to the position in the window at the end of the last * match/literal to return. */ - u32 optimum_cur_idx; - u32 optimum_end_idx; + struct lzx_mc_pos_data *optimum; + unsigned optimum_cur_idx; + unsigned optimum_end_idx; +}; + +/* + * Match chooser position data: + * + * An array of these structures is used during the match-choosing algorithm. + * They correspond to consecutive positions in the window and are used to keep + * track of the cost to reach each position, and the match/literal choices that + * need to be chosen to reach that position. + */ +struct lzx_mc_pos_data { + /* The approximate minimum cost, in bits, to reach this position in the + * window which has been found so far. */ + u32 cost; +#define MC_INFINITE_COST ((u32)~0UL) + + /* The union here is just for clarity, since the fields are used in two + * slightly different ways. Initially, the @prev structure is filled in + * first, and links go from later in the window to earlier in the + * window. Later, @next structure is filled in and links go from + * earlier in the window to later in the window. */ + union { + struct { + /* Position of the start of the match or literal that + * was taken to get to this position in the approximate + * minimum-cost parse. */ + u32 link; + + /* Offset (as in an LZ (length, offset) pair) of the + * match or literal that was taken to get to this + * position in the approximate minimum-cost parse. */ + u32 match_offset; + } prev; + struct { + /* Position at which the match or literal starting at + * this position ends in the minimum-cost parse. */ + u32 link; + + /* Offset (as in an LZ (length, offset) pair) of the + * match or literal starting at this position in the + * approximate minimum-cost parse. */ + u32 match_offset; + } next; + }; + + /* Adaptive state that exists after an approximate minimum-cost path to + * reach this position is taken. */ + struct lzx_lru_queue queue; }; /* Returns the LZX position slot that corresponds to a given match offset, * taking into account the recent offset queue and updating it if the offset is * found in it. */ static unsigned -lzx_get_position_slot(unsigned offset, struct lzx_lru_queue *queue) +lzx_get_position_slot(u32 offset, struct lzx_lru_queue *queue) { unsigned position_slot; /* See if the offset was recently used. */ - for (unsigned i = 0; i < LZX_NUM_RECENT_OFFSETS; i++) { + for (int i = 0; i < LZX_NUM_RECENT_OFFSETS; i++) { if (offset == queue->R[i]) { /* Found it. */ @@ -528,7 +506,7 @@ lzx_get_position_slot(unsigned offset, struct lzx_lru_queue *queue) position_slot = lzx_get_position_slot_raw(offset + LZX_OFFSET_OFFSET); /* Bring the new offset to the front of the queue. */ - for (unsigned i = LZX_NUM_RECENT_OFFSETS - 1; i > 0; i--) + for (int i = LZX_NUM_RECENT_OFFSETS - 1; i > 0; i--) queue->R[i] = queue->R[i - 1]; queue->R[0] = offset; @@ -564,17 +542,22 @@ lzx_make_huffman_codes(const struct lzx_freqs *freqs, } /* - * Output an LZX match. - * - * @out: The bitstream to write the match to. - * @block_type: The type of the LZX block (LZX_BLOCKTYPE_ALIGNED or LZX_BLOCKTYPE_VERBATIM) - * @match: The match. - * @codes: Pointer to a structure that contains the codewords for the - * main, length, and aligned offset Huffman codes. + * Output a precomputed LZX match. + * + * @out: + * The bitstream to which to write the match. + * @block_type: + * The type of the LZX block (LZX_BLOCKTYPE_ALIGNED or + * LZX_BLOCKTYPE_VERBATIM) + * @match: + * The match, as a (length, offset) pair. + * @codes: + * Pointer to a structure that contains the codewords for the main, length, + * and aligned offset Huffman codes for the current LZX compressed block. */ static void lzx_write_match(struct output_bitstream *out, int block_type, - struct lzx_match match, const struct lzx_codes *codes) + struct lzx_item match, const struct lzx_codes *codes) { /* low 8 bits are the match length minus 2 */ unsigned match_len_minus_2 = match.data & 0xff; @@ -600,9 +583,6 @@ lzx_write_match(struct output_bitstream *out, int block_type, * MIN_MATCH_LEN. */ if (match_len_minus_2 < LZX_NUM_PRIMARY_LENS) { len_header = match_len_minus_2; - /* No length footer-- mark it with a special - * value. */ - len_footer = (unsigned)(-1); } else { len_header = LZX_NUM_PRIMARY_LENS; len_footer = match_len_minus_2 - LZX_NUM_PRIMARY_LENS; @@ -622,10 +602,9 @@ lzx_write_match(struct output_bitstream *out, int block_type, /* If there is a length footer, output it using the * length Huffman code. */ - if (len_footer != (unsigned)(-1)) { + if (len_header == LZX_NUM_PRIMARY_LENS) bitstream_put_bits(out, codes->codewords.len[len_footer], codes->lens.len[len_footer]); - } num_extra_bits = lzx_get_num_extra_bits(position_slot); @@ -650,14 +629,24 @@ lzx_write_match(struct output_bitstream *out, int block_type, } } +/* Output an LZX literal (encoded with the main Huffman code). */ +static void +lzx_write_literal(struct output_bitstream *out, u8 literal, + const struct lzx_codes *codes) +{ + bitstream_put_bits(out, + codes->codewords.main[literal], + codes->lens.main[literal]); +} + static unsigned lzx_build_precode(const u8 lens[restrict], const u8 prev_lens[restrict], const unsigned num_syms, - freq_t precode_freqs[restrict LZX_PRECODE_NUM_SYMBOLS], + u32 precode_freqs[restrict LZX_PRECODE_NUM_SYMBOLS], u8 output_syms[restrict num_syms], u8 precode_lens[restrict LZX_PRECODE_NUM_SYMBOLS], - u16 precode_codewords[restrict LZX_PRECODE_NUM_SYMBOLS], + u32 precode_codewords[restrict LZX_PRECODE_NUM_SYMBOLS], unsigned *num_additional_bits_ret) { memset(precode_freqs, 0, @@ -789,23 +778,33 @@ lzx_build_precode(const u8 lens[restrict], } /* - * Writes a compressed Huffman code to the output, preceded by the precode for - * it. - * - * The Huffman code is represented in the output as a series of path lengths - * from which the canonical Huffman code can be reconstructed. The path lengths - * themselves are compressed using a separate Huffman code, the precode, which - * consists of LZX_PRECODE_NUM_SYMBOLS (= 20) symbols that cover all possible - * code lengths, plus extra codes for repeated lengths. The path lengths of the - * precode precede the path lengths of the larger code and are uncompressed, - * consisting of 20 entries of 4 bits each. - * - * @out: Bitstream to write the code to. - * @lens: The code lengths for the Huffman code, indexed by symbol. - * @prev_lens: Code lengths for this Huffman code, indexed by symbol, - * in the *previous block*, or all zeroes if this is the - * first block. - * @num_syms: The number of symbols in the code. + * Output a Huffman code in the compressed form used in LZX. + * + * The Huffman code is represented in the output as a logical series of codeword + * lengths from which the Huffman code, which must be in canonical form, can be + * reconstructed. + * + * The codeword lengths are themselves compressed using a separate Huffman code, + * the "precode", which contains a symbol for each possible codeword length in + * the larger code as well as several special symbols to represent repeated + * codeword lengths (a form of run-length encoding). The precode is itself + * constructed in canonical form, and its codeword lengths are represented + * literally in 20 4-bit fields that immediately precede the compressed codeword + * lengths of the larger code. + * + * Furthermore, the codeword lengths of the larger code are actually represented + * as deltas from the codeword lengths of the corresponding code in the previous + * block. + * + * @out: + * Bitstream to which to write the compressed Huffman code. + * @lens: + * The codeword lengths, indexed by symbol, in the Huffman code. + * @prev_lens: + * The codeword lengths, indexed by symbol, in the corresponding Huffman + * code in the previous block, or all zeroes if this is the first block. + * @num_syms: + * The number of symbols in the Huffman code. */ static void lzx_write_compressed_code(struct output_bitstream *out, @@ -813,10 +812,10 @@ lzx_write_compressed_code(struct output_bitstream *out, const u8 prev_lens[restrict], unsigned num_syms) { - freq_t precode_freqs[LZX_PRECODE_NUM_SYMBOLS]; + u32 precode_freqs[LZX_PRECODE_NUM_SYMBOLS]; u8 output_syms[num_syms]; u8 precode_lens[LZX_PRECODE_NUM_SYMBOLS]; - u16 precode_codewords[LZX_PRECODE_NUM_SYMBOLS]; + u32 precode_codewords[LZX_PRECODE_NUM_SYMBOLS]; unsigned i; unsigned num_output_syms; u8 precode_sym; @@ -864,43 +863,40 @@ lzx_write_compressed_code(struct output_bitstream *out, } /* - * Writes all compressed matches and literal bytes in an LZX block to the the - * output bitstream. + * Write all matches and literal bytes (which were precomputed) in an LZX + * compressed block to the output bitstream in the final compressed + * representation. * * @ostream * The output bitstream. * @block_type - * The type of the block (LZX_BLOCKTYPE_ALIGNED or LZX_BLOCKTYPE_VERBATIM). + * The chosen type of the LZX compressed block (LZX_BLOCKTYPE_ALIGNED or + * LZX_BLOCKTYPE_VERBATIM). * @match_tab - * The array of matches/literals that will be output (length @match_count). + * The array of matches/literals to output. * @match_count - * Number of matches/literals to be output. + * Number of matches/literals to output (length of @match_tab). * @codes - * Pointer to a structure that contains the codewords for the main, length, - * and aligned offset Huffman codes. + * The main, length, and aligned offset Huffman codes for the current + * LZX compressed block. */ static void lzx_write_matches_and_literals(struct output_bitstream *ostream, int block_type, - const struct lzx_match match_tab[], + const struct lzx_item match_tab[], unsigned match_count, const struct lzx_codes *codes) { for (unsigned i = 0; i < match_count; i++) { - struct lzx_match match = match_tab[i]; - - /* High bit of the match indicates whether the match is an - * actual match (1) or a literal uncompressed byte (0) */ - if (match.data & 0x80000000) { - /* match */ - lzx_write_match(ostream, block_type, - match, codes); - } else { - /* literal byte */ - bitstream_put_bits(ostream, - codes->codewords.main[match.data], - codes->lens.main[match.data]); - } + struct lzx_item match = match_tab[i]; + + /* The high bit of the 32-bit intermediate representation + * indicates whether the item is an actual LZ-style match (1) or + * a literal byte (0). */ + if (match.data & 0x80000000) + lzx_write_match(ostream, block_type, match, codes); + else + lzx_write_literal(ostream, match.data, codes); } } @@ -947,8 +943,8 @@ lzx_write_compressed_block(int block_type, unsigned block_size, unsigned max_window_size, unsigned num_main_syms, - struct lzx_match * chosen_matches, - unsigned num_chosen_matches, + struct lzx_item * chosen_items, + unsigned num_chosen_items, const struct lzx_codes * codes, const struct lzx_codes * prev_codes, struct output_bitstream * ostream) @@ -1026,7 +1022,7 @@ lzx_write_compressed_block(int block_type, /* Write the actual matches and literals. */ lzx_write_matches_and_literals(ostream, block_type, - chosen_matches, num_chosen_matches, + chosen_items, num_chosen_items, codes); LZX_DEBUG("Done writing block."); @@ -1041,17 +1037,17 @@ lzx_write_all_blocks(struct lzx_compressor *ctx, struct output_bitstream *ostrea for (unsigned i = 0; i < ctx->num_blocks; i++) { const struct lzx_block_spec *spec = &ctx->block_specs[i]; - LZX_DEBUG("Writing block %u/%u (type=%d, size=%u, num_chosen_matches=%u)...", + LZX_DEBUG("Writing block %u/%u (type=%d, size=%u, num_chosen_items=%u)...", i + 1, ctx->num_blocks, spec->block_type, spec->block_size, - spec->num_chosen_matches); + spec->num_chosen_items); lzx_write_compressed_block(spec->block_type, spec->block_size, ctx->max_window_size, ctx->num_main_syms, - &ctx->chosen_matches[spec->chosen_matches_start_pos], - spec->num_chosen_matches, + spec->chosen_items, + spec->num_chosen_items, &spec->codes, prev_codes, ostream); @@ -1062,7 +1058,7 @@ lzx_write_all_blocks(struct lzx_compressor *ctx, struct output_bitstream *ostrea /* Constructs an LZX match from a literal byte and updates the main code symbol * frequencies. */ -static u32 +static inline u32 lzx_tally_literal(u8 lit, struct lzx_freqs *freqs) { freqs->main[lit]++; @@ -1073,8 +1069,8 @@ lzx_tally_literal(u8 lit, struct lzx_freqs *freqs) * queue and the frequency of symbols in the main, length, and aligned offset * alphabets. The return value is a 32-bit number that provides the match in an * intermediate representation documented below. */ -static u32 -lzx_tally_match(unsigned match_len, unsigned match_offset, +static inline u32 +lzx_tally_match(unsigned match_len, u32 match_offset, struct lzx_freqs *freqs, struct lzx_lru_queue *queue) { unsigned position_slot; @@ -1125,7 +1121,7 @@ lzx_tally_match(unsigned match_len, unsigned match_offset, freqs->aligned[position_footer & 7]++; /* Pack the position slot, position footer, and match length into an - * intermediate representation. See `struct lzx_match' for details. + * intermediate representation. See `struct lzx_item' for details. */ LZX_ASSERT(LZX_MAX_POSITION_SLOTS <= 64); LZX_ASSERT(lzx_get_num_extra_bits(LZX_MAX_POSITION_SLOTS - 1) <= 17); @@ -1143,7 +1139,7 @@ lzx_tally_match(unsigned match_len, unsigned match_offset, struct lzx_record_ctx { struct lzx_freqs freqs; struct lzx_lru_queue queue; - struct lzx_match *matches; + struct lzx_item *matches; }; static void @@ -1164,7 +1160,7 @@ lzx_record_literal(u8 lit, void *_ctx) /* Returns the cost, in bits, to output a literal byte using the specified cost * model. */ -static unsigned +static u32 lzx_literal_cost(u8 c, const struct lzx_costs * costs) { return costs->main[c]; @@ -1174,14 +1170,15 @@ lzx_literal_cost(u8 c, const struct lzx_costs * costs) * well as costs for the codewords in the main, length, and aligned Huffman * codes, return the approximate number of bits it will take to represent this * match in the compressed output. Take into account the match offset LRU - * queue and optionally update it. */ -static unsigned -lzx_match_cost(unsigned length, unsigned offset, const struct lzx_costs *costs, + * queue and also updates it. */ +static u32 +lzx_match_cost(unsigned length, u32 offset, const struct lzx_costs *costs, struct lzx_lru_queue *queue) { unsigned position_slot; unsigned len_header, main_symbol; - unsigned cost = 0; + unsigned num_extra_bits; + u32 cost = 0; position_slot = lzx_get_position_slot(offset, queue); @@ -1192,7 +1189,7 @@ lzx_match_cost(unsigned length, unsigned offset, const struct lzx_costs *costs, cost += costs->main[main_symbol]; /* Account for extra position information. */ - unsigned num_extra_bits = lzx_get_num_extra_bits(position_slot); + num_extra_bits = lzx_get_num_extra_bits(position_slot); if (num_extra_bits >= 3) { cost += num_extra_bits - 3; cost += costs->aligned[(offset + LZX_OFFSET_OFFSET) & 7]; @@ -1208,22 +1205,6 @@ lzx_match_cost(unsigned length, unsigned offset, const struct lzx_costs *costs, } -/* Fast heuristic cost evaluation to use in the inner loop of the match-finder. - * Unlike lzx_match_cost() which does a true cost evaluation, this simply - * prioritize matches based on their offset. */ -static block_cost_t -lzx_match_cost_fast(unsigned offset, const struct lzx_lru_queue *queue) -{ - /* It seems well worth it to take the time to give priority to recently - * used offsets. */ - for (unsigned i = 0; i < LZX_NUM_RECENT_OFFSETS; i++) - if (offset == queue->R[i]) - return i; - - BUILD_BUG_ON(LZX_MAX_WINDOW_SIZE >= (block_cost_t)~0U); - return offset; -} - /* Set the cost model @ctx->costs from the Huffman codeword lengths specified in * @lens. * @@ -1261,293 +1242,59 @@ lzx_set_costs(struct lzx_compressor * ctx, const struct lzx_lens * lens) } } -/* Advance the suffix array match-finder to the next position. */ -static void -lzx_lz_update_salink(input_idx_t i, - const input_idx_t SA[restrict], - const input_idx_t ISA[restrict], - struct salink link[restrict]) -{ - /* r = Rank of the suffix at the current position. */ - const input_idx_t r = ISA[i]; - - /* next = rank of LOWEST ranked suffix that is ranked HIGHER than the - * current suffix AND has a LOWER position, or -1 if none exists. */ - const input_idx_t next = link[r].next; - - /* prev = rank of HIGHEST ranked suffix that is ranked LOWER than the - * current suffix AND has a LOWER position, or -1 if none exists. */ - const input_idx_t prev = link[r].prev; - - /* Link the suffix at the current position into the linked list that - * contains all suffixes in the suffix array that are appear at or - * before the current position, sorted by rank. - * - * Save the values of all fields we overwrite so that rollback is - * possible. */ - if (next != (input_idx_t)~0U) { - - link[next].prev = r; - link[next].lcpprev = link[r].lcpnext; - } - - if (prev != (input_idx_t)~0U) { - - link[prev].next = r; - link[prev].lcpnext = link[r].lcpprev; - } -} - -/* - * Use the suffix array match-finder to retrieve a list of LZ matches at the - * current position. - * - * [in] @i Current position in the window. - * [in] @SA Suffix array for the window. - * [in] @ISA Inverse suffix array for the window. - * [inout] @link Suffix array links used internally by the match-finder. - * [out] @matches The (length, offset) pairs of the resulting matches will - * be written here, sorted in decreasing order by - * length. All returned lengths will be unique. - * [in] @queue Recently used match offsets, used when evaluating the - * cost of matches. - * [in] @min_match_len Minimum match length to return. - * [in] @max_matches_to_consider Maximum number of matches to consider at - * the position. - * [in] @max_matches_to_return Maximum number of matches to return. - * - * The return value is the number of matches found and written to @matches. - */ +/* Retrieve a list of matches available at the next position in the input. + * + * A pointer to the matches array is written into @matches_ret, and the return + * value is the number of matches found. */ static unsigned -lzx_lz_get_matches(const input_idx_t i, - const input_idx_t SA[const restrict], - const input_idx_t ISA[const restrict], - struct salink link[const restrict], - struct raw_match matches[const restrict], - const struct lzx_lru_queue * const restrict queue, - const unsigned min_match_len, - const u32 max_matches_to_consider, - const u32 max_matches_to_return) +lzx_get_matches(struct lzx_compressor *ctx, + const struct lz_match **matches_ret) { - /* r = Rank of the suffix at the current position. */ - const input_idx_t r = ISA[i]; - - /* Prepare for searching the current position. */ - lzx_lz_update_salink(i, SA, ISA, link); - - /* L = rank of next suffix to the left; - * R = rank of next suffix to the right; - * lenL = length of match between current position and the suffix with rank L; - * lenR = length of match between current position and the suffix with rank R. - * - * This is left and right relative to the rank of the current suffix. - * Since the suffixes in the suffix array are sorted, the longest - * matches are immediately to the left and right (using the linked list - * to ignore all suffixes that occur later in the window). The match - * length decreases the farther left and right we go. We shall keep the - * length on both sides in sync in order to choose the lowest-cost match - * of each length. - */ - input_idx_t L = link[r].prev; - input_idx_t R = link[r].next; - input_idx_t lenL = link[r].lcpprev; - input_idx_t lenR = link[r].lcpnext; - - /* nmatches = number of matches found so far. */ - unsigned nmatches = 0; - - /* best_cost = cost of lowest-cost match found so far. - * - * We keep track of this so that we can ignore shorter matches that do - * not have lower costs than a longer matches already found. - */ - block_cost_t best_cost = INFINITE_BLOCK_COST; - - /* count_remaining = maximum number of possible matches remaining to be - * considered. */ - u32 count_remaining = max_matches_to_consider; - - /* pending = match currently being considered for a specific length. */ - struct raw_match pending; - block_cost_t pending_cost; - - while (lenL >= min_match_len || lenR >= min_match_len) - { - pending.len = lenL; - pending_cost = INFINITE_BLOCK_COST; - block_cost_t cost; - - /* Extend left. */ - if (lenL >= min_match_len && lenL >= lenR) { - for (;;) { - - if (--count_remaining == 0) - goto out_save_pending; - - input_idx_t offset = i - SA[L]; - - /* Save match if it has smaller cost. */ - cost = lzx_match_cost_fast(offset, queue); - if (cost < pending_cost) { - pending.offset = offset; - pending_cost = cost; - } - - if (link[L].lcpprev < lenL) { - /* Match length decreased. */ - - lenL = link[L].lcpprev; - - /* Save the pending match unless the - * right side still may have matches of - * this length to be scanned, or if a - * previous (longer) match had lower - * cost. */ - if (pending.len > lenR) { - if (pending_cost < best_cost) { - best_cost = pending_cost; - matches[nmatches++] = pending; - if (nmatches == max_matches_to_return) - return nmatches; - } - pending.len = lenL; - pending_cost = INFINITE_BLOCK_COST; - } - if (lenL < min_match_len || lenL < lenR) - break; - } - L = link[L].prev; - } - } - - pending.len = lenR; - - /* Extend right. */ - if (lenR >= min_match_len && lenR > lenL) { - for (;;) { - - if (--count_remaining == 0) - goto out_save_pending; - - input_idx_t offset = i - SA[R]; - - /* Save match if it has smaller cost. */ - cost = lzx_match_cost_fast(offset, queue); - if (cost < pending_cost) { - pending.offset = offset; - pending_cost = cost; - } - - if (link[R].lcpnext < lenR) { - /* Match length decreased. */ - - lenR = link[R].lcpnext; - - /* Save the pending match unless a - * previous (longer) match had lower - * cost. */ - if (pending_cost < best_cost) { - matches[nmatches++] = pending; - best_cost = pending_cost; - if (nmatches == max_matches_to_return) - return nmatches; - } - - if (lenR < min_match_len || lenR <= lenL) - break; - - pending.len = lenR; - pending_cost = INFINITE_BLOCK_COST; - } - R = link[R].next; - } - } - } - goto out; - -out_save_pending: - if (pending_cost != INFINITE_BLOCK_COST) - matches[nmatches++] = pending; - -out: - return nmatches; -} + struct lz_match *cache_ptr; + struct lz_match *matches; + unsigned num_matches; + LZX_ASSERT(ctx->match_window_pos < ctx->match_window_end); -/* Tell the match-finder to skip the specified number of bytes (@n) in the - * input. */ -static void -lzx_lz_skip_bytes(struct lzx_compressor *ctx, unsigned n) -{ - LZX_ASSERT(n <= ctx->match_window_end - ctx->match_window_pos); - if (ctx->matches_cached) { - ctx->match_window_pos += n; - while (n--) { - ctx->cached_matches_pos += - ctx->cached_matches[ctx->cached_matches_pos].len + 1; + cache_ptr = ctx->cache_ptr; + matches = cache_ptr + 1; + if (likely(cache_ptr <= ctx->cache_limit)) { + if (ctx->matches_cached) { + num_matches = cache_ptr->len; + } else { + num_matches = lz_bt_get_matches(&ctx->mf, matches); + cache_ptr->len = num_matches; } } else { - while (n--) { - ctx->cached_matches[ctx->cached_matches_pos++].len = 0; - lzx_lz_update_salink(ctx->match_window_pos++, ctx->SA, - ctx->ISA, ctx->salink); - } + num_matches = 0; } -} -/* Retrieve a list of matches available at the next position in the input. - * - * The matches are written to ctx->matches in decreasing order of length, and - * the return value is the number of matches found. */ -static unsigned -lzx_lz_get_matches_caching(struct lzx_compressor *ctx, - const struct lzx_lru_queue *queue, - struct raw_match **matches_ret) -{ - unsigned num_matches; - struct raw_match *matches; - - LZX_ASSERT(ctx->match_window_pos <= ctx->match_window_end); + /* Don't allow matches to span the end of an LZX block. */ + if (ctx->match_window_end < ctx->window_size && num_matches != 0) { + unsigned limit = ctx->match_window_end - ctx->match_window_pos; - matches = &ctx->cached_matches[ctx->cached_matches_pos + 1]; + if (limit >= LZX_MIN_MATCH_LEN) { - if (ctx->matches_cached) { - num_matches = matches[-1].len; - } else { - unsigned min_match_len = LZX_MIN_MATCH_LEN; - if (!ctx->params.alg_params.slow.use_len2_matches) - min_match_len = max(min_match_len, 3); - const u32 max_search_depth = ctx->params.alg_params.slow.max_search_depth; - const u32 max_matches_per_pos = ctx->params.alg_params.slow.max_matches_per_pos; + unsigned i = num_matches - 1; + do { + if (matches[i].len >= limit) { + matches[i].len = limit; - if (unlikely(max_search_depth == 0 || max_matches_per_pos == 0)) + /* Truncation might produce multiple + * matches with length 'limit'. Keep at + * most 1. */ + num_matches = i + 1; + } + } while (i--); + } else { num_matches = 0; - else - num_matches = lzx_lz_get_matches(ctx->match_window_pos, - ctx->SA, - ctx->ISA, - ctx->salink, - matches, - queue, - min_match_len, - max_search_depth, - max_matches_per_pos); - matches[-1].len = num_matches; + } + cache_ptr->len = num_matches; } - ctx->cached_matches_pos += num_matches + 1; - *matches_ret = matches; - /* Cap the length of returned matches to the number of bytes remaining, - * if it is not the whole window. */ - if (ctx->match_window_end < ctx->window_size) { - unsigned maxlen = ctx->match_window_end - ctx->match_window_pos; - for (unsigned i = 0; i < num_matches; i++) - if (matches[i].len > maxlen) - matches[i].len = maxlen; - } #if 0 fprintf(stderr, "Pos %u/%u: %u matches\n", - ctx->match_window_pos, ctx->match_window_end, num_matches); + ctx->match_window_pos, ctx->window_size, num_matches); for (unsigned i = 0; i < num_matches; i++) fprintf(stderr, "\tLen %u Offset %u\n", matches[i].len, matches[i].offset); #endif @@ -1562,22 +1309,48 @@ lzx_lz_get_matches_caching(struct lzx_compressor *ctx, LZX_ASSERT(!memcmp(&ctx->window[ctx->match_window_pos], &ctx->window[ctx->match_window_pos - matches[i].offset], matches[i].len)); + if (i) { + LZX_ASSERT(matches[i].len > matches[i - 1].len); + LZX_ASSERT(matches[i].offset > matches[i - 1].offset); + } } #endif - ctx->match_window_pos++; + ctx->cache_ptr = matches + num_matches; + *matches_ret = matches; return num_matches; } +static void +lzx_skip_bytes(struct lzx_compressor *ctx, unsigned n) +{ + struct lz_match *cache_ptr; + + LZX_ASSERT(n <= ctx->match_window_end - ctx->match_window_pos); + + cache_ptr = ctx->cache_ptr; + ctx->match_window_pos += n; + if (ctx->matches_cached) { + while (n-- && cache_ptr <= ctx->cache_limit) + cache_ptr += 1 + cache_ptr->len; + } else { + lz_bt_skip_positions(&ctx->mf, n); + while (n-- && cache_ptr <= ctx->cache_limit) { + cache_ptr->len = 0; + cache_ptr += 1; + } + } + ctx->cache_ptr = cache_ptr; +} + /* * Reverse the linked list of near-optimal matches so that they can be returned * in forwards order. * * Returns the first match in the list. */ -static struct raw_match -lzx_lz_reverse_near_optimal_match_list(struct lzx_compressor *ctx, - unsigned cur_pos) +static struct lz_match +lzx_match_chooser_reverse_list(struct lzx_compressor *ctx, unsigned cur_pos) { unsigned prev_link, saved_prev_link; unsigned prev_match_offset, saved_prev_match_offset; @@ -1602,64 +1375,85 @@ lzx_lz_reverse_near_optimal_match_list(struct lzx_compressor *ctx, ctx->optimum_cur_idx = ctx->optimum[0].next.link; - return (struct raw_match) + return (struct lz_match) { .len = ctx->optimum_cur_idx, .offset = ctx->optimum[0].next.match_offset, }; } /* - * lzx_lz_get_near_optimal_match() - + * lzx_get_near_optimal_match() - * - * Choose the optimal match or literal to use at the next position in the input. - * - * Unlike a greedy parser that always takes the longest match, or even a - * parser with one match/literal look-ahead like zlib, the algorithm used here - * may look ahead many matches/literals to determine the optimal match/literal to - * output next. The motivation is that the compression ratio is improved if the - * compressor can do things like use a shorter-than-possible match in order to - * allow a longer match later, and also take into account the Huffman code cost - * model rather than simply assuming that longer is better. + * Choose an approximately optimal match or literal to use at the next position + * in the string, or "window", being LZ-encoded. * - * Still, this is not truly an optimal parser because very long matches are - * taken immediately, and the raw match-finder takes some shortcuts. This is - * done to avoid considering many different alternatives that are unlikely to - * be significantly better. + * This is based on algorithms used in 7-Zip, including the DEFLATE encoder + * and the LZMA encoder, written by Igor Pavlov. * - * This algorithm is based on that used in 7-Zip's DEFLATE encoder. + * Unlike a greedy parser that always takes the longest match, or even a "lazy" + * parser with one match/literal look-ahead like zlib, the algorithm used here + * may look ahead many matches/literals to determine the approximately optimal + * match/literal to code next. The motivation is that the compression ratio is + * improved if the compressor can do things like use a shorter-than-possible + * match in order to allow a longer match later, and also take into account the + * estimated real cost of coding each match/literal based on the underlying + * entropy encoding. + * + * Still, this is not a true optimal parser for several reasons: + * + * - Real compression formats use entropy encoding of the literal/match + * sequence, so the real cost of coding each match or literal is unknown until + * the parse is fully determined. It can be approximated based on iterative + * parses, but the end result is not guaranteed to be globally optimal. + * + * - Very long matches are chosen immediately. This is because locations with + * long matches are likely to have many possible alternatives that would cause + * slow optimal parsing, but also such locations are already highly + * compressible so it is not too harmful to just grab the longest match. + * + * - Not all possible matches at each location are considered because the + * underlying match-finder limits the number and type of matches produced at + * each position. For example, for a given match length it's usually not + * worth it to only consider matches other than the lowest-offset match, + * except in the case of a repeat offset. + * + * - Although we take into account the adaptive state (in LZX, the recent offset + * queue), coding decisions made with respect to the adaptive state will be + * locally optimal but will not necessarily be globally optimal. This is + * because the algorithm only keeps the least-costly path to get to a given + * location and does not take into account that a slightly more costly path + * could result in a different adaptive state that ultimately results in a + * lower global cost. + * + * - The array space used by this function is bounded, so in degenerate cases it + * is forced to start returning matches/literals before the algorithm has + * really finished. * * Each call to this function does one of two things: * - * 1. Build a near-optimal sequence of matches/literals, up to some point, that + * 1. Build a sequence of near-optimal matches/literals, up to some point, that * will be returned by subsequent calls to this function, then return the * first one. * * OR * * 2. Return the next match/literal previously computed by a call to this - * function; - * - * This function relies on the following state in the compressor context: - * - * ctx->window (read-only: preprocessed data being compressed) - * ctx->cost (read-only: cost model to use) - * ctx->optimum (internal state; leave uninitialized) - * ctx->optimum_cur_idx (must set to 0 before first call) - * ctx->optimum_end_idx (must set to 0 before first call) - * - * Plus any state used by the raw match-finder. + * function. * * The return value is a (length, offset) pair specifying the match or literal - * chosen. For literals, the length is less than LZX_MIN_MATCH_LEN and the - * offset is meaningless. + * chosen. For literals, the length is 0 or 1 and the offset is meaningless. */ -static struct raw_match -lzx_lz_get_near_optimal_match(struct lzx_compressor * ctx) +static struct lz_match +lzx_get_near_optimal_match(struct lzx_compressor *ctx) { - unsigned num_possible_matches; - struct raw_match *possible_matches; - struct raw_match match; - unsigned longest_match_len; + unsigned num_matches; + const struct lz_match *matches; + struct lz_match match; + unsigned longest_len; + unsigned longest_rep_len; + u32 longest_rep_offset; + unsigned cur_pos; + unsigned end_pos; if (ctx->optimum_cur_idx != ctx->optimum_end_idx) { /* Case 2: Return the next match/literal already found. */ @@ -1676,168 +1470,361 @@ lzx_lz_get_near_optimal_match(struct lzx_compressor * ctx) ctx->optimum_cur_idx = 0; ctx->optimum_end_idx = 0; - /* Get matches at this position. */ - num_possible_matches = lzx_lz_get_matches_caching(ctx, &ctx->queue, &possible_matches); + /* Search for matches at recent offsets. Only keep the one with the + * longest match length. */ + longest_rep_len = LZX_MIN_MATCH_LEN - 1; + if (ctx->match_window_pos >= 1) { + unsigned limit = min(LZX_MAX_MATCH_LEN, + ctx->match_window_end - ctx->match_window_pos); + for (int i = 0; i < LZX_NUM_RECENT_OFFSETS; i++) { + u32 offset = ctx->queue.R[i]; + const u8 *strptr = &ctx->window[ctx->match_window_pos]; + const u8 *matchptr = strptr - offset; + unsigned len = 0; + while (len < limit && strptr[len] == matchptr[len]) + len++; + if (len > longest_rep_len) { + longest_rep_len = len; + longest_rep_offset = offset; + } + } + } - /* If no matches found, return literal. */ - if (num_possible_matches == 0) - return (struct raw_match){ .len = 0 }; + /* If there's a long match with a recent offset, take it. */ + if (longest_rep_len >= ctx->params.alg_params.slow.nice_match_length) { + lzx_skip_bytes(ctx, longest_rep_len); + return (struct lz_match) { + .len = longest_rep_len, + .offset = longest_rep_offset, + }; + } - /* The matches that were found are sorted in decreasing order by length. - * Get the length of the longest one. */ - longest_match_len = possible_matches[0].len; + /* Search other matches. */ + num_matches = lzx_get_matches(ctx, &matches); - /* Greedy heuristic: if the longest match that was found is greater - * than the number of fast bytes, return it immediately; don't both - * doing more work. */ - if (longest_match_len > ctx->params.alg_params.slow.num_fast_bytes) { - lzx_lz_skip_bytes(ctx, longest_match_len - 1); - return possible_matches[0]; + /* If there's a long match, take it. */ + if (num_matches) { + longest_len = matches[num_matches - 1].len; + if (longest_len >= ctx->params.alg_params.slow.nice_match_length) { + lzx_skip_bytes(ctx, longest_len - 1); + return matches[num_matches - 1]; + } + } else { + longest_len = 1; } - /* Calculate the cost to reach the next position by outputting a - * literal. */ - ctx->optimum[0].queue = ctx->queue; - ctx->optimum[1].queue = ctx->optimum[0].queue; - ctx->optimum[1].cost = lzx_literal_cost(ctx->window[ctx->match_window_pos], + /* Calculate the cost to reach the next position by coding a literal. + */ + ctx->optimum[1].queue = ctx->queue; + ctx->optimum[1].cost = lzx_literal_cost(ctx->window[ctx->match_window_pos - 1], &ctx->costs); ctx->optimum[1].prev.link = 0; /* Calculate the cost to reach any position up to and including that - * reached by the longest match, using the shortest (i.e. closest) match - * that reaches each position. */ - BUILD_BUG_ON(LZX_MIN_MATCH_LEN != 2); - for (unsigned len = LZX_MIN_MATCH_LEN, match_idx = num_possible_matches - 1; - len <= longest_match_len; len++) { - - LZX_ASSERT(match_idx < num_possible_matches); - - ctx->optimum[len].queue = ctx->optimum[0].queue; - ctx->optimum[len].prev.link = 0; - ctx->optimum[len].prev.match_offset = possible_matches[match_idx].offset; - ctx->optimum[len].cost = lzx_match_cost(len, - possible_matches[match_idx].offset, - &ctx->costs, - &ctx->optimum[len].queue); - if (len == possible_matches[match_idx].len) - match_idx--; - } + * reached by the longest match. + * + * Note: We consider only the lowest-offset match that reaches each + * position. + * + * Note: Some of the cost calculation stays the same for each offset, + * regardless of how many lengths it gets used for. Therefore, to + * improve performance, we hand-code the cost calculation instead of + * calling lzx_match_cost() to do a from-scratch cost evaluation at each + * length. */ + for (unsigned i = 0, len = 2; i < num_matches; i++) { + u32 offset; + struct lzx_lru_queue queue; + u32 position_cost; + unsigned position_slot; + unsigned num_extra_bits; + + offset = matches[i].offset; + queue = ctx->queue; + position_cost = 0; + + position_slot = lzx_get_position_slot(offset, &queue); + num_extra_bits = lzx_get_num_extra_bits(position_slot); + if (num_extra_bits >= 3) { + position_cost += num_extra_bits - 3; + position_cost += ctx->costs.aligned[(offset + LZX_OFFSET_OFFSET) & 7]; + } else { + position_cost += num_extra_bits; + } + + do { + unsigned len_header; + unsigned main_symbol; + u32 cost; - unsigned cur_pos = 0; + cost = position_cost; - /* len_end: greatest index forward at which costs have been calculated - * so far */ - unsigned len_end = longest_match_len; + len_header = min(len - LZX_MIN_MATCH_LEN, LZX_NUM_PRIMARY_LENS); + main_symbol = ((position_slot << 3) | len_header) + LZX_NUM_CHARS; + cost += ctx->costs.main[main_symbol]; + if (len_header == LZX_NUM_PRIMARY_LENS) + cost += ctx->costs.len[len - LZX_MIN_MATCH_LEN - LZX_NUM_PRIMARY_LENS]; + ctx->optimum[len].queue = queue; + ctx->optimum[len].prev.link = 0; + ctx->optimum[len].prev.match_offset = offset; + ctx->optimum[len].cost = cost; + } while (++len <= matches[i].len); + } + end_pos = longest_len; + + if (longest_rep_len >= LZX_MIN_MATCH_LEN) { + struct lzx_lru_queue queue; + u32 cost; + + while (end_pos < longest_rep_len) + ctx->optimum[++end_pos].cost = MC_INFINITE_COST; + + queue = ctx->queue; + cost = lzx_match_cost(longest_rep_len, longest_rep_offset, + &ctx->costs, &queue); + if (cost <= ctx->optimum[longest_rep_len].cost) { + ctx->optimum[longest_rep_len].queue = queue; + ctx->optimum[longest_rep_len].prev.link = 0; + ctx->optimum[longest_rep_len].prev.match_offset = longest_rep_offset; + ctx->optimum[longest_rep_len].cost = cost; + } + } + + /* Step forward, calculating the estimated minimum cost to reach each + * position. The algorithm may find multiple paths to reach each + * position; only the lowest-cost path is saved. + * + * The progress of the parse is tracked in the @ctx->optimum array, which + * for each position contains the minimum cost to reach that position, + * the index of the start of the match/literal taken to reach that + * position through the minimum-cost path, the offset of the match taken + * (not relevant for literals), and the adaptive state that will exist + * at that position after the minimum-cost path is taken. The @cur_pos + * variable stores the position at which the algorithm is currently + * considering coding choices, and the @end_pos variable stores the + * greatest position at which the costs of coding choices have been + * saved. (Actually, the algorithm guarantees that all positions up to + * and including @end_pos are reachable by at least one path.) + * + * The loop terminates when any one of the following conditions occurs: + * + * 1. A match with length greater than or equal to @nice_match_length is + * found. When this occurs, the algorithm chooses this match + * unconditionally, and consequently the near-optimal match/literal + * sequence up to and including that match is fully determined and it + * can begin returning the match/literal list. + * + * 2. @cur_pos reaches a position not overlapped by a preceding match. + * In such cases, the near-optimal match/literal sequence up to + * @cur_pos is fully determined and it can begin returning the + * match/literal list. + * + * 3. Failing either of the above in a degenerate case, the loop + * terminates when space in the @ctx->optimum array is exhausted. + * This terminates the algorithm and forces it to start returning + * matches/literals even though they may not be globally optimal. + * + * Upon loop termination, a nonempty list of matches/literals will have + * been produced and stored in the @optimum array. These + * matches/literals are linked in reverse order, so the last thing this + * function does is reverse this list and return the first + * match/literal, leaving the rest to be returned immediately by + * subsequent calls to this function. + */ + cur_pos = 0; for (;;) { + u32 cost; + /* Advance to next position. */ cur_pos++; - if (cur_pos == len_end || cur_pos == LZX_OPTIM_ARRAY_SIZE) - return lzx_lz_reverse_near_optimal_match_list(ctx, cur_pos); + /* Check termination conditions (2) and (3) noted above. */ + if (cur_pos == end_pos || cur_pos == LZX_OPTIM_ARRAY_SIZE) + return lzx_match_chooser_reverse_list(ctx, cur_pos); + + /* Search for matches at recent offsets. */ + longest_rep_len = LZX_MIN_MATCH_LEN - 1; + unsigned limit = min(LZX_MAX_MATCH_LEN, + ctx->match_window_end - ctx->match_window_pos); + for (int i = 0; i < LZX_NUM_RECENT_OFFSETS; i++) { + u32 offset = ctx->optimum[cur_pos].queue.R[i]; + const u8 *strptr = &ctx->window[ctx->match_window_pos]; + const u8 *matchptr = strptr - offset; + unsigned len = 0; + while (len < limit && strptr[len] == matchptr[len]) + len++; + if (len > longest_rep_len) { + longest_rep_len = len; + longest_rep_offset = offset; + } + } - /* retrieve the number of matches available at this position */ - num_possible_matches = lzx_lz_get_matches_caching(ctx, &ctx->optimum[cur_pos].queue, - &possible_matches); + /* If we found a long match at a recent offset, choose it + * immediately. */ + if (longest_rep_len >= ctx->params.alg_params.slow.nice_match_length) { + /* Build the list of matches to return and get + * the first one. */ + match = lzx_match_chooser_reverse_list(ctx, cur_pos); - unsigned new_len = 0; + /* Append the long match to the end of the list. */ + ctx->optimum[cur_pos].next.match_offset = longest_rep_offset; + ctx->optimum[cur_pos].next.link = cur_pos + longest_rep_len; + ctx->optimum_end_idx = cur_pos + longest_rep_len; - if (num_possible_matches != 0) { - new_len = possible_matches[0].len; + /* Skip over the remaining bytes of the long match. */ + lzx_skip_bytes(ctx, longest_rep_len); + + /* Return first match in the list. */ + return match; + } - /* Greedy heuristic: if we found a match greater than - * the number of fast bytes, stop immediately. */ - if (new_len > ctx->params.alg_params.slow.num_fast_bytes) { + /* Search other matches. */ + num_matches = lzx_get_matches(ctx, &matches); + /* If there's a long match, take it. */ + if (num_matches) { + longest_len = matches[num_matches - 1].len; + if (longest_len >= ctx->params.alg_params.slow.nice_match_length) { /* Build the list of matches to return and get * the first one. */ - match = lzx_lz_reverse_near_optimal_match_list(ctx, cur_pos); + match = lzx_match_chooser_reverse_list(ctx, cur_pos); /* Append the long match to the end of the list. */ ctx->optimum[cur_pos].next.match_offset = - possible_matches[0].offset; - ctx->optimum[cur_pos].next.link = cur_pos + new_len; - ctx->optimum_end_idx = cur_pos + new_len; + matches[num_matches - 1].offset; + ctx->optimum[cur_pos].next.link = cur_pos + longest_len; + ctx->optimum_end_idx = cur_pos + longest_len; /* Skip over the remaining bytes of the long match. */ - lzx_lz_skip_bytes(ctx, new_len - 1); + lzx_skip_bytes(ctx, longest_len - 1); - /* Return first match in the list */ + /* Return first match in the list. */ return match; } + } else { + longest_len = 1; } - /* Consider proceeding with a literal byte. */ - block_cost_t cur_cost = ctx->optimum[cur_pos].cost; - block_cost_t cur_plus_literal_cost = cur_cost + + while (end_pos < cur_pos + longest_len) + ctx->optimum[++end_pos].cost = MC_INFINITE_COST; + + /* Consider coding a literal. */ + cost = ctx->optimum[cur_pos].cost + lzx_literal_cost(ctx->window[ctx->match_window_pos - 1], &ctx->costs); - if (cur_plus_literal_cost < ctx->optimum[cur_pos + 1].cost) { - ctx->optimum[cur_pos + 1].cost = cur_plus_literal_cost; - ctx->optimum[cur_pos + 1].prev.link = cur_pos; + if (cost < ctx->optimum[cur_pos + 1].cost) { ctx->optimum[cur_pos + 1].queue = ctx->optimum[cur_pos].queue; + ctx->optimum[cur_pos + 1].cost = cost; + ctx->optimum[cur_pos + 1].prev.link = cur_pos; } - if (num_possible_matches == 0) - continue; - - /* Consider proceeding with a match. */ - - while (len_end < cur_pos + new_len) - ctx->optimum[++len_end].cost = INFINITE_BLOCK_COST; - - for (unsigned len = LZX_MIN_MATCH_LEN, match_idx = num_possible_matches - 1; - len <= new_len; len++) { - LZX_ASSERT(match_idx < num_possible_matches); - struct lzx_lru_queue q = ctx->optimum[cur_pos].queue; - block_cost_t cost = cur_cost + lzx_match_cost(len, - possible_matches[match_idx].offset, - &ctx->costs, - &q); - - if (cost < ctx->optimum[cur_pos + len].cost) { - ctx->optimum[cur_pos + len].cost = cost; - ctx->optimum[cur_pos + len].prev.link = cur_pos; - ctx->optimum[cur_pos + len].prev.match_offset = - possible_matches[match_idx].offset; - ctx->optimum[cur_pos + len].queue = q; + /* Consider coding a match. + * + * The hard-coded cost calculation is done for the same reason + * stated in the comment for the similar loop earlier. + * Actually, it is *this* one that has the biggest effect on + * performance; overall LZX compression is > 10% faster with + * this code compared to calling lzx_match_cost() with each + * length. */ + for (unsigned i = 0, len = 2; i < num_matches; i++) { + u32 offset; + struct lzx_lru_queue queue; + u32 position_cost; + unsigned position_slot; + unsigned num_extra_bits; + + offset = matches[i].offset; + queue = ctx->optimum[cur_pos].queue; + position_cost = ctx->optimum[cur_pos].cost; + + position_slot = lzx_get_position_slot(offset, &queue); + num_extra_bits = lzx_get_num_extra_bits(position_slot); + if (num_extra_bits >= 3) { + position_cost += num_extra_bits - 3; + position_cost += ctx->costs.aligned[ + (offset + LZX_OFFSET_OFFSET) & 7]; + } else { + position_cost += num_extra_bits; } - if (len == possible_matches[match_idx].len) - match_idx--; + do { + unsigned len_header; + unsigned main_symbol; + u32 cost; + + cost = position_cost; + + len_header = min(len - LZX_MIN_MATCH_LEN, + LZX_NUM_PRIMARY_LENS); + main_symbol = ((position_slot << 3) | len_header) + + LZX_NUM_CHARS; + cost += ctx->costs.main[main_symbol]; + if (len_header == LZX_NUM_PRIMARY_LENS) { + cost += ctx->costs.len[len - + LZX_MIN_MATCH_LEN - + LZX_NUM_PRIMARY_LENS]; + } + if (cost < ctx->optimum[cur_pos + len].cost) { + ctx->optimum[cur_pos + len].queue = queue; + ctx->optimum[cur_pos + len].prev.link = cur_pos; + ctx->optimum[cur_pos + len].prev.match_offset = offset; + ctx->optimum[cur_pos + len].cost = cost; + } + } while (++len <= matches[i].len); + } + + if (longest_rep_len >= LZX_MIN_MATCH_LEN) { + struct lzx_lru_queue queue; + + while (end_pos < cur_pos + longest_rep_len) + ctx->optimum[++end_pos].cost = MC_INFINITE_COST; + + queue = ctx->optimum[cur_pos].queue; + + cost = ctx->optimum[cur_pos].cost + + lzx_match_cost(longest_rep_len, longest_rep_offset, + &ctx->costs, &queue); + if (cost <= ctx->optimum[cur_pos + longest_rep_len].cost) { + ctx->optimum[cur_pos + longest_rep_len].queue = + queue; + ctx->optimum[cur_pos + longest_rep_len].prev.link = + cur_pos; + ctx->optimum[cur_pos + longest_rep_len].prev.match_offset = + longest_rep_offset; + ctx->optimum[cur_pos + longest_rep_len].cost = + cost; + } } } } -/* - * Set default symbol costs. - */ +/* Set default symbol costs for the LZX Huffman codes. */ static void lzx_set_default_costs(struct lzx_costs * costs, unsigned num_main_syms) { unsigned i; - /* Literal symbols */ + /* Main code (part 1): Literal symbols */ for (i = 0; i < LZX_NUM_CHARS; i++) costs->main[i] = 8; - /* Match header symbols */ + /* Main code (part 2): Match header symbols */ for (; i < num_main_syms; i++) costs->main[i] = 10; - /* Length symbols */ + /* Length code */ for (i = 0; i < LZX_LENCODE_NUM_SYMBOLS; i++) costs->len[i] = 8; - /* Aligned offset symbols */ + /* Aligned offset code */ for (i = 0; i < LZX_ALIGNEDCODE_NUM_SYMBOLS; i++) costs->aligned[i] = 3; } -/* Given the frequencies of symbols in a compressed block and the corresponding - * Huffman codes, return LZX_BLOCKTYPE_ALIGNED or LZX_BLOCKTYPE_VERBATIM if an - * aligned offset or verbatim block, respectively, will take fewer bits to - * output. */ +/* Given the frequencies of symbols in an LZX-compressed block and the + * corresponding Huffman codes, return LZX_BLOCKTYPE_ALIGNED or + * LZX_BLOCKTYPE_VERBATIM if an aligned offset or verbatim block, respectively, + * will take fewer bits to output. */ static int lzx_choose_verbatim_or_aligned(const struct lzx_freqs * freqs, const struct lzx_codes * codes) @@ -1847,8 +1834,8 @@ lzx_choose_verbatim_or_aligned(const struct lzx_freqs * freqs, /* Verbatim blocks have a constant 3 bits per position footer. Aligned * offset blocks have an aligned offset symbol per position footer, plus - * an extra 24 bits to output the lengths necessary to reconstruct the - * aligned offset code itself. */ + * an extra 24 bits per block to output the lengths necessary to + * reconstruct the aligned offset code itself. */ for (unsigned i = 0; i < LZX_ALIGNEDCODE_NUM_SYMBOLS; i++) { verbatim_cost += 3 * freqs->aligned[i]; aligned_cost += codes->lens.aligned[i] * freqs->aligned[i]; @@ -1861,255 +1848,124 @@ lzx_choose_verbatim_or_aligned(const struct lzx_freqs * freqs, } /* Find a near-optimal sequence of matches/literals with which to output the - * specified LZX block, then set its type to that which has the minimum cost to - * output. */ + * specified LZX block, then set the block's type to that which has the minimum + * cost to output (either verbatim or aligned). */ static void lzx_optimize_block(struct lzx_compressor *ctx, struct lzx_block_spec *spec, unsigned num_passes) { const struct lzx_lru_queue orig_queue = ctx->queue; + unsigned num_passes_remaining = num_passes; struct lzx_freqs freqs; + const u8 *window_ptr; + const u8 *window_end; + struct lzx_item *next_chosen_match; + struct lz_match lz_match; + struct lzx_item lzx_item; - unsigned orig_window_pos = spec->window_pos; - unsigned orig_cached_pos = ctx->cached_matches_pos; - - LZX_ASSERT(ctx->match_window_pos == spec->window_pos); + LZX_ASSERT(num_passes >= 1); + LZX_ASSERT(lz_bt_get_position(&ctx->mf) == spec->window_pos); ctx->match_window_end = spec->window_pos + spec->block_size; - spec->chosen_matches_start_pos = spec->window_pos; - - LZX_ASSERT(num_passes >= 1); + ctx->matches_cached = false; /* The first optimal parsing pass is done using the cost model already * set in ctx->costs. Each later pass is done using a cost model - * computed from the previous pass. */ - for (unsigned pass = 0; pass < num_passes; pass++) { + * computed from the previous pass. + * + * To improve performance we only generate the array containing the + * matches and literals in intermediate form on the final pass. */ - ctx->match_window_pos = orig_window_pos; - ctx->cached_matches_pos = orig_cached_pos; - ctx->queue = orig_queue; - spec->num_chosen_matches = 0; + while (--num_passes_remaining) { + ctx->match_window_pos = spec->window_pos; + ctx->cache_ptr = ctx->cached_matches; memset(&freqs, 0, sizeof(freqs)); + window_ptr = &ctx->window[spec->window_pos]; + window_end = window_ptr + spec->block_size; + + while (window_ptr != window_end) { - for (unsigned i = spec->window_pos; i < spec->window_pos + spec->block_size; ) { - struct raw_match raw_match; - struct lzx_match lzx_match; + lz_match = lzx_get_near_optimal_match(ctx); - raw_match = lzx_lz_get_near_optimal_match(ctx); - if (raw_match.len >= LZX_MIN_MATCH_LEN) { - lzx_match.data = lzx_tally_match(raw_match.len, raw_match.offset, - &freqs, &ctx->queue); - i += raw_match.len; + LZX_ASSERT(!(lz_match.len == LZX_MIN_MATCH_LEN && + lz_match.offset == ctx->max_window_size - + LZX_MIN_MATCH_LEN)); + if (lz_match.len >= LZX_MIN_MATCH_LEN) { + lzx_tally_match(lz_match.len, lz_match.offset, + &freqs, &ctx->queue); + window_ptr += lz_match.len; } else { - lzx_match.data = lzx_tally_literal(ctx->window[i], &freqs); - i += 1; + lzx_tally_literal(*window_ptr, &freqs); + window_ptr += 1; } - ctx->chosen_matches[spec->chosen_matches_start_pos + - spec->num_chosen_matches++] = lzx_match; } - - lzx_make_huffman_codes(&freqs, &spec->codes, - ctx->num_main_syms); - if (pass < num_passes - 1) - lzx_set_costs(ctx, &spec->codes.lens); + lzx_make_huffman_codes(&freqs, &spec->codes, ctx->num_main_syms); + lzx_set_costs(ctx, &spec->codes.lens); + ctx->queue = orig_queue; ctx->matches_cached = true; } - spec->block_type = lzx_choose_verbatim_or_aligned(&freqs, &spec->codes); - ctx->matches_cached = false; -} - -static void -lzx_optimize_blocks(struct lzx_compressor *ctx) -{ - lzx_lru_queue_init(&ctx->queue); - ctx->optimum_cur_idx = 0; - ctx->optimum_end_idx = 0; - - const unsigned num_passes = ctx->params.alg_params.slow.num_optim_passes; - - for (unsigned i = 0; i < ctx->num_blocks; i++) - lzx_optimize_block(ctx, &ctx->block_specs[i], num_passes); -} - -/* Initialize the suffix array match-finder for the specified input. */ -static void -lzx_lz_init_matchfinder(const u8 T[const restrict], - const input_idx_t n, - input_idx_t SA[const restrict], - input_idx_t ISA[const restrict], - input_idx_t LCP[const restrict], - struct salink link[const restrict], - const unsigned max_match_len) -{ - /* Compute SA (Suffix Array). */ - - { - /* ISA and link are used as temporary space. */ - BUILD_BUG_ON(LZX_MIN_WINDOW_SIZE * sizeof(ISA[0]) < 256 * sizeof(saidx_t)); - BUILD_BUG_ON(LZX_MIN_WINDOW_SIZE * 2 * sizeof(link[0]) < 256 * 256 * sizeof(saidx_t)); - if (sizeof(input_idx_t) == sizeof(saidx_t)) { - divsufsort(T, SA, n, (saidx_t*)ISA, (saidx_t*)link); + ctx->match_window_pos = spec->window_pos; + ctx->cache_ptr = ctx->cached_matches; + memset(&freqs, 0, sizeof(freqs)); + window_ptr = &ctx->window[spec->window_pos]; + window_end = window_ptr + spec->block_size; + + spec->chosen_items = &ctx->chosen_items[spec->window_pos]; + next_chosen_match = spec->chosen_items; + + while (window_ptr != window_end) { + lz_match = lzx_get_near_optimal_match(ctx); + + LZX_ASSERT(!(lz_match.len == LZX_MIN_MATCH_LEN && + lz_match.offset == ctx->max_window_size - + LZX_MIN_MATCH_LEN)); + if (lz_match.len >= LZX_MIN_MATCH_LEN) { + lzx_item.data = lzx_tally_match(lz_match.len, + lz_match.offset, + &freqs, &ctx->queue); + window_ptr += lz_match.len; } else { - saidx_t sa[n]; - divsufsort(T, sa, n, (saidx_t*)ISA, (saidx_t*)link); - for (input_idx_t i = 0; i < n; i++) - SA[i] = sa[i]; + lzx_item.data = lzx_tally_literal(*window_ptr, &freqs); + window_ptr += 1; } + *next_chosen_match++ = lzx_item; } - -#ifdef ENABLE_LZX_DEBUG - - LZX_ASSERT(n > 0); - - /* Verify suffix array. */ - { - bool found[n]; - ZERO_ARRAY(found); - for (input_idx_t r = 0; r < n; r++) { - input_idx_t i = SA[r]; - LZX_ASSERT(i < n); - LZX_ASSERT(!found[i]); - found[i] = true; - } - } - - for (input_idx_t r = 0; r < n - 1; r++) { - - input_idx_t i1 = SA[r]; - input_idx_t i2 = SA[r + 1]; - - input_idx_t n1 = n - i1; - input_idx_t n2 = n - i2; - - LZX_ASSERT(memcmp(&T[i1], &T[i2], min(n1, n2)) <= 0); - } - LZX_DEBUG("Verified SA (len %u)", n); -#endif /* ENABLE_LZX_DEBUG */ - - /* Compute ISA (Inverse Suffix Array) */ - for (input_idx_t r = 0; r < n; r++) - ISA[SA[r]] = r; - - /* Compute LCP (longest common prefix) array. - * - * Algorithm adapted from Kasai et al. 2001: "Linear-Time - * Longest-Common-Prefix Computation in Suffix Arrays and Its - * Applications". */ - { - input_idx_t h = 0; - for (input_idx_t i = 0; i < n; i++) { - input_idx_t r = ISA[i]; - if (r > 0) { - input_idx_t j = SA[r - 1]; - - input_idx_t lim = min(n - i, n - j); - - while (h < lim && T[i + h] == T[j + h]) - h++; - LCP[r] = h; - if (h > 0) - h--; - } - } - } - -#ifdef ENABLE_LZX_DEBUG - /* Verify LCP array. */ - for (input_idx_t r = 0; r < n - 1; r++) { - LZX_ASSERT(ISA[SA[r]] == r); - LZX_ASSERT(ISA[SA[r + 1]] == r + 1); - - input_idx_t i1 = SA[r]; - input_idx_t i2 = SA[r + 1]; - input_idx_t lcp = LCP[r + 1]; - - input_idx_t n1 = n - i1; - input_idx_t n2 = n - i2; - - LZX_ASSERT(lcp <= min(n1, n2)); - - LZX_ASSERT(memcmp(&T[i1], &T[i2], lcp) == 0); - if (lcp < min(n1, n2)) - LZX_ASSERT(T[i1 + lcp] != T[i2 + lcp]); - } -#endif /* ENABLE_LZX_DEBUG */ - - /* Compute salink.next and salink.lcpnext. - * - * Algorithm adapted from Crochemore et al. 2009: - * "LPF computation revisited". - * - * Note: we cap lcpnext to the maximum match length so that the - * match-finder need not worry about it later. */ - link[n - 1].next = (input_idx_t)~0U; - link[n - 1].prev = (input_idx_t)~0U; - link[n - 1].lcpnext = 0; - link[n - 1].lcpprev = 0; - for (input_idx_t r = n - 2; r != (input_idx_t)~0U; r--) { - input_idx_t t = r + 1; - input_idx_t l = LCP[t]; - while (t != (input_idx_t)~0 && SA[t] > SA[r]) { - l = min(l, link[t].lcpnext); - t = link[t].next; - } - link[r].next = t; - link[r].lcpnext = min(l, max_match_len); - LZX_ASSERT(t == (input_idx_t)~0U || l <= n - SA[t]); - LZX_ASSERT(l <= n - SA[r]); - LZX_ASSERT(memcmp(&T[SA[r]], &T[SA[t]], l) == 0); - } - - /* Compute salink.prev and salink.lcpprev. - * - * Algorithm adapted from Crochemore et al. 2009: - * "LPF computation revisited". - * - * Note: we cap lcpprev to the maximum match length so that the - * match-finder need not worry about it later. */ - link[0].prev = (input_idx_t)~0; - link[0].next = (input_idx_t)~0; - link[0].lcpprev = 0; - link[0].lcpnext = 0; - for (input_idx_t r = 1; r < n; r++) { - input_idx_t t = r - 1; - input_idx_t l = LCP[r]; - while (t != (input_idx_t)~0 && SA[t] > SA[r]) { - l = min(l, link[t].lcpprev); - t = link[t].prev; - } - link[r].prev = t; - link[r].lcpprev = min(l, max_match_len); - LZX_ASSERT(t == (input_idx_t)~0 || l <= n - SA[t]); - LZX_ASSERT(l <= n - SA[r]); - LZX_ASSERT(memcmp(&T[SA[r]], &T[SA[t]], l) == 0); - } + spec->num_chosen_items = next_chosen_match - spec->chosen_items; + lzx_make_huffman_codes(&freqs, &spec->codes, ctx->num_main_syms); + spec->block_type = lzx_choose_verbatim_or_aligned(&freqs, &spec->codes); } /* Prepare the input window into one or more LZX blocks ready to be output. */ static void lzx_prepare_blocks(struct lzx_compressor * ctx) { - /* Initialize the match-finder. */ - lzx_lz_init_matchfinder(ctx->window, ctx->window_size, - ctx->SA, ctx->ISA, ctx->LCP, ctx->salink, - LZX_MAX_MATCH_LEN); - ctx->cached_matches_pos = 0; - ctx->matches_cached = false; - ctx->match_window_pos = 0; - /* Set up a default cost model. */ lzx_set_default_costs(&ctx->costs, ctx->num_main_syms); + /* Set up the block specifications. + * TODO: The compression ratio could be slightly improved by performing + * data-dependent block splitting instead of using fixed-size blocks. + * Doing so well is a computationally hard problem, however. */ ctx->num_blocks = DIV_ROUND_UP(ctx->window_size, LZX_DIV_BLOCK_SIZE); for (unsigned i = 0; i < ctx->num_blocks; i++) { unsigned pos = LZX_DIV_BLOCK_SIZE * i; ctx->block_specs[i].window_pos = pos; - ctx->block_specs[i].block_size = min(ctx->window_size - pos, LZX_DIV_BLOCK_SIZE); + ctx->block_specs[i].block_size = min(ctx->window_size - pos, + LZX_DIV_BLOCK_SIZE); } + /* Load the window into the match-finder. */ + lz_bt_load_window(&ctx->mf, ctx->window, ctx->window_size); + /* Determine sequence of matches/literals to output for each block. */ - lzx_optimize_blocks(ctx); + lzx_lru_queue_init(&ctx->queue); + ctx->optimum_cur_idx = 0; + ctx->optimum_end_idx = 0; + for (unsigned i = 0; i < ctx->num_blocks; i++) { + lzx_optimize_block(ctx, &ctx->block_specs[i], + ctx->params.alg_params.slow.num_optim_passes); + } } /* @@ -2127,7 +1983,7 @@ lzx_prepare_blocks(struct lzx_compressor * ctx) * * ctx->block_specs[] * ctx->num_blocks - * ctx->chosen_matches[] + * ctx->chosen_items[] */ static void lzx_prepare_block_fast(struct lzx_compressor * ctx) @@ -2142,7 +1998,7 @@ lzx_prepare_block_fast(struct lzx_compressor * ctx) * aren't worth choosing when using greedy or lazy parsing. */ .min_match = 3, .max_match = LZX_MAX_MATCH_LEN, - .max_offset = 32768, + .max_offset = LZX_MAX_WINDOW_SIZE, .good_match = LZX_MAX_MATCH_LEN, .nice_match = LZX_MAX_MATCH_LEN, .max_chain_len = LZX_MAX_MATCH_LEN, @@ -2153,7 +2009,7 @@ lzx_prepare_block_fast(struct lzx_compressor * ctx) /* Initialize symbol frequencies and match offset LRU queue. */ memset(&record_ctx.freqs, 0, sizeof(struct lzx_freqs)); lzx_lru_queue_init(&record_ctx.queue); - record_ctx.matches = ctx->chosen_matches; + record_ctx.matches = ctx->chosen_items; /* Determine series of matches/literals to output. */ lz_analyze_block(ctx->window, @@ -2169,75 +2025,39 @@ lzx_prepare_block_fast(struct lzx_compressor * ctx) spec->block_type = LZX_BLOCKTYPE_ALIGNED; spec->window_pos = 0; spec->block_size = ctx->window_size; - spec->num_chosen_matches = (record_ctx.matches - ctx->chosen_matches); - spec->chosen_matches_start_pos = 0; + spec->num_chosen_items = (record_ctx.matches - ctx->chosen_items); + spec->chosen_items = ctx->chosen_items; lzx_make_huffman_codes(&record_ctx.freqs, &spec->codes, ctx->num_main_syms); ctx->num_blocks = 1; } -static void -do_call_insn_translation(u32 *call_insn_target, int input_pos, - s32 file_size) -{ - s32 abs_offset; - s32 rel_offset; - - rel_offset = le32_to_cpu(*call_insn_target); - if (rel_offset >= -input_pos && rel_offset < file_size) { - if (rel_offset < file_size - input_pos) { - /* "good translation" */ - abs_offset = rel_offset + input_pos; - } else { - /* "compensating translation" */ - abs_offset = rel_offset - file_size; - } - *call_insn_target = cpu_to_le32(abs_offset); - } -} - -/* This is the reverse of undo_call_insn_preprocessing() in lzx-decompress.c. - * See the comment above that function for more information. */ -static void -do_call_insn_preprocessing(u8 data[], int size) +static size_t +lzx_compress(const void *uncompressed_data, size_t uncompressed_size, + void *compressed_data, size_t compressed_size_avail, void *_ctx) { - for (int i = 0; i < size - 10; i++) { - if (data[i] == 0xe8) { - do_call_insn_translation((u32*)&data[i + 1], i, - LZX_WIM_MAGIC_FILESIZE); - i += 4; - } - } -} - -/* API function documented in wimlib.h */ -WIMLIBAPI unsigned -wimlib_lzx_compress2(const void * const restrict uncompressed_data, - unsigned const uncompressed_len, - void * const restrict compressed_data, - struct wimlib_lzx_context * const restrict lzx_ctx) -{ - struct lzx_compressor *ctx = (struct lzx_compressor*)lzx_ctx; + struct lzx_compressor *ctx = _ctx; struct output_bitstream ostream; - input_idx_t compressed_len; + size_t compressed_size; - if (uncompressed_len < 100) { + if (uncompressed_size < 100) { LZX_DEBUG("Too small to bother compressing."); return 0; } - if (uncompressed_len > ctx->max_window_size) { - LZX_DEBUG("Can't compress %u bytes using window of %u bytes!", - uncompressed_len, ctx->max_window_size); + if (uncompressed_size > ctx->max_window_size) { + LZX_DEBUG("Can't compress %zu bytes using window of %u bytes!", + uncompressed_size, ctx->max_window_size); return 0; } - LZX_DEBUG("Attempting to compress %u bytes...", uncompressed_len); + LZX_DEBUG("Attempting to compress %zu bytes...", + uncompressed_size); /* The input data must be preprocessed. To avoid changing the original * input, copy it to a temporary buffer. */ - memcpy(ctx->window, uncompressed_data, uncompressed_len); - ctx->window_size = uncompressed_len; + memcpy(ctx->window, uncompressed_data, uncompressed_size); + ctx->window_size = uncompressed_size; /* This line is unnecessary; it just avoids inconsequential accesses of * uninitialized memory that would show up in memory-checking tools such @@ -2248,7 +2068,7 @@ wimlib_lzx_compress2(const void * const restrict uncompressed_data, /* Before doing any actual compression, do the call instruction (0xe8 * byte) translation on the uncompressed data. */ - do_call_insn_preprocessing(ctx->window, ctx->window_size); + lzx_do_e8_preprocessing(ctx->window, ctx->window_size); LZX_DEBUG("Preparing blocks..."); @@ -2261,18 +2081,19 @@ wimlib_lzx_compress2(const void * const restrict uncompressed_data, LZX_DEBUG("Writing compressed blocks..."); /* Generate the compressed data. */ - init_output_bitstream(&ostream, compressed_data, ctx->window_size - 1); + init_output_bitstream(&ostream, compressed_data, compressed_size_avail); lzx_write_all_blocks(ctx, &ostream); LZX_DEBUG("Flushing bitstream..."); - compressed_len = flush_output_bitstream(&ostream); - if (compressed_len == ~(input_idx_t)0) { - LZX_DEBUG("Data did not compress to less than original length!"); + compressed_size = flush_output_bitstream(&ostream); + if (compressed_size == (u32)~0UL) { + LZX_DEBUG("Data did not compress to %zu bytes or less!", + compressed_size_avail); return 0; } - LZX_DEBUG("Done: compressed %u => %u bytes.", - uncompressed_len, compressed_len); + LZX_DEBUG("Done: compressed %zu => %zu bytes.", + uncompressed_size, compressed_size); /* Verify that we really get the same thing back when decompressing. * Although this could be disabled by default in all cases, it only @@ -2284,289 +2105,274 @@ wimlib_lzx_compress2(const void * const restrict uncompressed_data, #endif ) { - /* The decompression buffer can be any temporary space that's no - * longer needed. */ - u8 *buf = (u8*)(ctx->SA ? ctx->SA : ctx->prev_tab); + struct wimlib_decompressor *decompressor; - if (wimlib_lzx_decompress2(compressed_data, compressed_len, - buf, uncompressed_len, ctx->max_window_size)) + if (0 == wimlib_create_decompressor(WIMLIB_COMPRESSION_TYPE_LZX, + ctx->max_window_size, + NULL, + &decompressor)) { - ERROR("Failed to decompress data we " - "compressed using LZX algorithm"); - wimlib_assert(0); - return 0; - } - - if (memcmp(uncompressed_data, buf, uncompressed_len)) { - ERROR("Data we compressed using LZX algorithm " - "didn't decompress to original"); - wimlib_assert(0); - return 0; + int ret; + ret = wimlib_decompress(compressed_data, + compressed_size, + ctx->window, + uncompressed_size, + decompressor); + wimlib_free_decompressor(decompressor); + + if (ret) { + ERROR("Failed to decompress data we " + "compressed using LZX algorithm"); + wimlib_assert(0); + return 0; + } + if (memcmp(uncompressed_data, ctx->window, uncompressed_size)) { + ERROR("Data we compressed using LZX algorithm " + "didn't decompress to original"); + wimlib_assert(0); + return 0; + } + } else { + WARNING("Failed to create decompressor for " + "data verification!"); } } - return compressed_len; -} - -static bool -lzx_params_compatible(const struct wimlib_lzx_params *oldparams, - const struct wimlib_lzx_params *newparams) -{ - return 0 == memcmp(oldparams, newparams, sizeof(struct wimlib_lzx_params)); + return compressed_size; } -static struct wimlib_lzx_params lzx_user_default_params; -static struct wimlib_lzx_params *lzx_user_default_params_ptr; - -static bool -lzx_params_valid(const struct wimlib_lzx_params *params) +static void +lzx_free_compressor(void *_ctx) { - /* Validate parameters. */ - if (params->size_of_this != sizeof(struct wimlib_lzx_params)) { - LZX_DEBUG("Invalid parameter structure size!"); - return false; - } - - if (params->algorithm != WIMLIB_LZX_ALGORITHM_SLOW && - params->algorithm != WIMLIB_LZX_ALGORITHM_FAST) - { - LZX_DEBUG("Invalid algorithm."); - return false; - } + struct lzx_compressor *ctx = _ctx; - if (params->algorithm == WIMLIB_LZX_ALGORITHM_SLOW) { - if (params->alg_params.slow.num_optim_passes < 1) - { - LZX_DEBUG("Invalid number of optimization passes!"); - return false; - } - - if (params->alg_params.slow.main_nostat_cost < 1 || - params->alg_params.slow.main_nostat_cost > 16) - { - LZX_DEBUG("Invalid main_nostat_cost!"); - return false; - } - - if (params->alg_params.slow.len_nostat_cost < 1 || - params->alg_params.slow.len_nostat_cost > 16) - { - LZX_DEBUG("Invalid len_nostat_cost!"); - return false; - } - - if (params->alg_params.slow.aligned_nostat_cost < 1 || - params->alg_params.slow.aligned_nostat_cost > 8) - { - LZX_DEBUG("Invalid aligned_nostat_cost!"); - return false; - } + if (ctx) { + FREE(ctx->chosen_items); + FREE(ctx->cached_matches); + FREE(ctx->optimum); + lz_bt_destroy(&ctx->mf); + FREE(ctx->block_specs); + FREE(ctx->prev_tab); + FREE(ctx->window); + FREE(ctx); } - return true; } -/* API function documented in wimlib.h */ -WIMLIBAPI int -wimlib_lzx_set_default_params(const struct wimlib_lzx_params * params) +static const struct wimlib_lzx_compressor_params lzx_fast_default = { + .hdr = { + .size = sizeof(struct wimlib_lzx_compressor_params), + }, + .algorithm = WIMLIB_LZX_ALGORITHM_FAST, + .use_defaults = 0, + .alg_params = { + .fast = { + }, + }, +}; +static const struct wimlib_lzx_compressor_params lzx_slow_default = { + .hdr = { + .size = sizeof(struct wimlib_lzx_compressor_params), + }, + .algorithm = WIMLIB_LZX_ALGORITHM_SLOW, + .use_defaults = 0, + .alg_params = { + .slow = { + .use_len2_matches = 1, + .nice_match_length = 32, + .num_optim_passes = 2, + .max_search_depth = 50, + .main_nostat_cost = 15, + .len_nostat_cost = 15, + .aligned_nostat_cost = 7, + }, + }, +}; + +static const struct wimlib_lzx_compressor_params * +lzx_get_params(const struct wimlib_compressor_params_header *_params) { - if (params) { - if (!lzx_params_valid(params)) - return WIMLIB_ERR_INVALID_PARAM; - lzx_user_default_params = *params; - lzx_user_default_params_ptr = &lzx_user_default_params; + const struct wimlib_lzx_compressor_params *params = + (const struct wimlib_lzx_compressor_params*)_params; + + if (params == NULL) { + LZX_DEBUG("Using default algorithm and parameters."); + params = &lzx_slow_default; } else { - lzx_user_default_params_ptr = NULL; + if (params->use_defaults) { + if (params->algorithm == WIMLIB_LZX_ALGORITHM_SLOW) + params = &lzx_slow_default; + else + params = &lzx_fast_default; + } } - return 0; + return params; } -/* API function documented in wimlib.h */ -WIMLIBAPI int -wimlib_lzx_alloc_context(u32 window_size, - const struct wimlib_lzx_params *params, - struct wimlib_lzx_context **ctx_pp) +static int +lzx_create_compressor(size_t window_size, + const struct wimlib_compressor_params_header *_params, + void **ctx_ret) { + const struct wimlib_lzx_compressor_params *params = lzx_get_params(_params); + struct lzx_compressor *ctx; LZX_DEBUG("Allocating LZX context..."); if (!lzx_window_size_valid(window_size)) return WIMLIB_ERR_INVALID_PARAM; - struct lzx_compressor *ctx; - - static const struct wimlib_lzx_params fast_default = { - .size_of_this = sizeof(struct wimlib_lzx_params), - .algorithm = WIMLIB_LZX_ALGORITHM_FAST, - .use_defaults = 0, - .alg_params = { - .fast = { - }, - }, - }; - static const struct wimlib_lzx_params slow_default = { - .size_of_this = sizeof(struct wimlib_lzx_params), - .algorithm = WIMLIB_LZX_ALGORITHM_SLOW, - .use_defaults = 0, - .alg_params = { - .slow = { - .use_len2_matches = 1, - .num_fast_bytes = 32, - .num_optim_passes = 2, - .max_search_depth = 50, - .max_matches_per_pos = 3, - .main_nostat_cost = 15, - .len_nostat_cost = 15, - .aligned_nostat_cost = 7, - }, - }, - }; - - if (params) { - if (!lzx_params_valid(params)) - return WIMLIB_ERR_INVALID_PARAM; - } else { - LZX_DEBUG("Using default algorithm and parameters."); - if (lzx_user_default_params_ptr) - params = lzx_user_default_params_ptr; - else - params = &slow_default; - } - - if (params->use_defaults) { - if (params->algorithm == WIMLIB_LZX_ALGORITHM_SLOW) - params = &slow_default; - else - params = &fast_default; - } - - if (ctx_pp) { - ctx = *(struct lzx_compressor**)ctx_pp; - - if (ctx && - lzx_params_compatible(&ctx->params, params) && - ctx->max_window_size == window_size) - return 0; - } else { - LZX_DEBUG("Check parameters only."); - return 0; - } - - LZX_DEBUG("Allocating memory."); - ctx = CALLOC(1, sizeof(struct lzx_compressor)); if (ctx == NULL) - goto err; + goto oom; ctx->num_main_syms = lzx_get_num_main_syms(window_size); ctx->max_window_size = window_size; ctx->window = MALLOC(window_size + 12); if (ctx->window == NULL) - goto err; + goto oom; if (params->algorithm == WIMLIB_LZX_ALGORITHM_FAST) { ctx->prev_tab = MALLOC(window_size * sizeof(ctx->prev_tab[0])); if (ctx->prev_tab == NULL) - goto err; + goto oom; } size_t block_specs_length = DIV_ROUND_UP(window_size, LZX_DIV_BLOCK_SIZE); ctx->block_specs = MALLOC(block_specs_length * sizeof(ctx->block_specs[0])); if (ctx->block_specs == NULL) - goto err; + goto oom; if (params->algorithm == WIMLIB_LZX_ALGORITHM_SLOW) { - ctx->SA = MALLOC(3U * window_size * sizeof(ctx->SA[0])); - if (ctx->SA == NULL) - goto err; - ctx->ISA = ctx->SA + window_size; - ctx->LCP = ctx->ISA + window_size; - - ctx->salink = MALLOC(window_size * sizeof(ctx->salink[0])); - if (ctx->salink == NULL) - goto err; + unsigned min_match_len = LZX_MIN_MATCH_LEN; + if (!params->alg_params.slow.use_len2_matches) + min_match_len = max(min_match_len, 3); + + if (!lz_bt_init(&ctx->mf, + window_size, + min_match_len, + LZX_MAX_MATCH_LEN, + params->alg_params.slow.nice_match_length, + params->alg_params.slow.max_search_depth)) + goto oom; } if (params->algorithm == WIMLIB_LZX_ALGORITHM_SLOW) { - ctx->optimum = MALLOC((LZX_OPTIM_ARRAY_SIZE + LZX_MAX_MATCH_LEN) * - sizeof(ctx->optimum[0])); + ctx->optimum = MALLOC((LZX_OPTIM_ARRAY_SIZE + + min(params->alg_params.slow.nice_match_length, + LZX_MAX_MATCH_LEN)) * + sizeof(ctx->optimum[0])); if (ctx->optimum == NULL) - goto err; + goto oom; } if (params->algorithm == WIMLIB_LZX_ALGORITHM_SLOW) { - u32 cache_per_pos; - - cache_per_pos = params->alg_params.slow.max_matches_per_pos; - if (cache_per_pos > LZX_MAX_CACHE_PER_POS) - cache_per_pos = LZX_MAX_CACHE_PER_POS; - - ctx->cached_matches = MALLOC(window_size * (cache_per_pos + 1) * - sizeof(ctx->cached_matches[0])); + ctx->cached_matches = MALLOC(LZX_CACHE_SIZE); if (ctx->cached_matches == NULL) - goto err; + goto oom; + ctx->cache_limit = ctx->cached_matches + + LZX_CACHE_LEN - (LZX_MAX_MATCHES_PER_POS + 1); } - ctx->chosen_matches = MALLOC(window_size * sizeof(ctx->chosen_matches[0])); - if (ctx->chosen_matches == NULL) - goto err; + ctx->chosen_items = MALLOC(window_size * sizeof(ctx->chosen_items[0])); + if (ctx->chosen_items == NULL) + goto oom; - memcpy(&ctx->params, params, sizeof(struct wimlib_lzx_params)); + memcpy(&ctx->params, params, sizeof(struct wimlib_lzx_compressor_params)); memset(&ctx->zero_codes, 0, sizeof(ctx->zero_codes)); LZX_DEBUG("Successfully allocated new LZX context."); - wimlib_lzx_free_context(*ctx_pp); - *ctx_pp = (struct wimlib_lzx_context*)ctx; + *ctx_ret = ctx; return 0; -err: - wimlib_lzx_free_context((struct wimlib_lzx_context*)ctx); - LZX_DEBUG("Ran out of memory."); +oom: + lzx_free_compressor(ctx); return WIMLIB_ERR_NOMEM; } -/* API function documented in wimlib.h */ -WIMLIBAPI void -wimlib_lzx_free_context(struct wimlib_lzx_context *_ctx) +static u64 +lzx_get_needed_memory(size_t max_block_size, + const struct wimlib_compressor_params_header *_params) { - struct lzx_compressor *ctx = (struct lzx_compressor*)_ctx; + const struct wimlib_lzx_compressor_params *params = lzx_get_params(_params); - if (ctx) { - FREE(ctx->chosen_matches); - FREE(ctx->cached_matches); - FREE(ctx->optimum); - FREE(ctx->salink); - FREE(ctx->SA); - FREE(ctx->block_specs); - FREE(ctx->prev_tab); - FREE(ctx->window); - FREE(ctx); + u64 size = 0; + + size += sizeof(struct lzx_compressor); + + size += max_block_size + 12; + + size += DIV_ROUND_UP(max_block_size, LZX_DIV_BLOCK_SIZE) * + sizeof(((struct lzx_compressor*)0)->block_specs[0]); + + if (params->algorithm == WIMLIB_LZX_ALGORITHM_SLOW) { + size += max_block_size * sizeof(((struct lzx_compressor*)0)->chosen_items[0]); + size += lz_bt_get_needed_memory(max_block_size); + size += (LZX_OPTIM_ARRAY_SIZE + + min(params->alg_params.slow.nice_match_length, + LZX_MAX_MATCH_LEN)) * + sizeof(((struct lzx_compressor *)0)->optimum[0]); + size += LZX_CACHE_SIZE; + } else { + size += max_block_size * sizeof(((struct lzx_compressor*)0)->prev_tab[0]); } + return size; } -/* API function documented in wimlib.h */ -WIMLIBAPI unsigned -wimlib_lzx_compress(const void * const restrict uncompressed_data, - unsigned const uncompressed_len, - void * const restrict compressed_data) +static bool +lzx_params_valid(const struct wimlib_compressor_params_header *_params) { - int ret; - struct wimlib_lzx_context *ctx = NULL; - unsigned compressed_len; - - ret = wimlib_lzx_alloc_context(32768, NULL, &ctx); - if (ret) { - wimlib_assert(ret != WIMLIB_ERR_INVALID_PARAM); - WARNING("Couldn't allocate LZX compression context: %"TS"", - wimlib_get_error_string(ret)); - return 0; + const struct wimlib_lzx_compressor_params *params = + (const struct wimlib_lzx_compressor_params*)_params; + + if (params->hdr.size != sizeof(struct wimlib_lzx_compressor_params)) { + LZX_DEBUG("Invalid parameter structure size!"); + return false; } - compressed_len = wimlib_lzx_compress2(uncompressed_data, - uncompressed_len, - compressed_data, - ctx); + if (params->algorithm != WIMLIB_LZX_ALGORITHM_SLOW && + params->algorithm != WIMLIB_LZX_ALGORITHM_FAST) + { + LZX_DEBUG("Invalid algorithm."); + return false; + } + + if (params->algorithm == WIMLIB_LZX_ALGORITHM_SLOW && + !params->use_defaults) + { + if (params->alg_params.slow.num_optim_passes < 1) + { + LZX_DEBUG("Invalid number of optimization passes!"); + return false; + } + + if (params->alg_params.slow.main_nostat_cost < 1 || + params->alg_params.slow.main_nostat_cost > 16) + { + LZX_DEBUG("Invalid main_nostat_cost!"); + return false; + } - wimlib_lzx_free_context(ctx); + if (params->alg_params.slow.len_nostat_cost < 1 || + params->alg_params.slow.len_nostat_cost > 16) + { + LZX_DEBUG("Invalid len_nostat_cost!"); + return false; + } - return compressed_len; + if (params->alg_params.slow.aligned_nostat_cost < 1 || + params->alg_params.slow.aligned_nostat_cost > 8) + { + LZX_DEBUG("Invalid aligned_nostat_cost!"); + return false; + } + } + return true; } + +const struct compressor_ops lzx_compressor_ops = { + .params_valid = lzx_params_valid, + .get_needed_memory = lzx_get_needed_memory, + .create_compressor = lzx_create_compressor, + .compress = lzx_compress, + .free_compressor = lzx_free_compressor, +};