X-Git-Url: https://wimlib.net/git/?p=wimlib;a=blobdiff_plain;f=src%2Flzx-compress.c;h=cdbb63699004edbed7066f9344f6d1c6f30207d2;hp=ea55fca42f9ab24ae3c87373328ac2b750344af6;hb=10be6c712f4f5f066458dbab11e81f8e64cd2324;hpb=6f1261c57e213d6f12cb7aa8f858f2971bee687e diff --git a/src/lzx-compress.c b/src/lzx-compress.c index ea55fca4..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. * @@ -25,130 +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 - * ====== + * Format Overview * - * First, the primary reference for the LZX compression format is the - * specification released by Microsoft. + * 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. * - * Second, the comments in lzx-decompress.c provide some more information about - * the LZX compression format, including errors in the Microsoft specification. + * 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: * - * 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 preprocesses the data to attempt to make x86 machine code slightly more + * compressible before attempting to compress it further. * - * - LZX preprocesses the data before attempting to compress it. * - 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 footer" (giving, roughly speaking, the order of + * 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 has a minimum match length of 2 rather than 3. - * - In LZX, match offsets 0 through 2 actually represent entries in a LRU queue - * of match offsets. - * - * 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. Determine the best known sequence of LZ77 matches ((offset, length) pairs) - * and literal bytes to divide the input into. Raw match-finding is done - * using a very clever binary tree search based on the "Bt3" algorithm from - * 7-Zip. Parsing, or match-choosing, is solved essentially as a - * minimum-cost path problem, but using a heuristic forward search based on - * the Deflate encoder from 7-Zip rather than a more intuitive backward - * search, the latter of which would naively require that all matches be - * found. This heuristic search, as well as other heuristics such as limits - * on the matches considered, considerably speed up this part of the - * algorithm, which is the main bottleneck. Finally, after matches and - * literals are chosen, the needed Huffman codes needed to output them are - * built. - * - * 3. Up to a certain number of iterations, use the resulting Huffman codes to - * refine a cost model and go back to Step #2 to determine an improved - * sequence of matches and literals. - * - * 4. Up to a certain depth, try splitting the current block to see if the - * compression ratio can be improved. This may be the case if parts of the - * input differ greatly from each other and could benefit from different - * Huffman codes. - * - * 5. Output the resulting block(s) using the match/literal sequences and the - * Huffman codes that were computed for each block. - * - * 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, match choosing, and block splitting. 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() + * - 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"). * - * 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() - * - * Both wimlib_lzx_compress() and wimlib_lzx_compress2() are designed to - * compress an in-memory buffer of up to 32768 bytes. There is no sliding - * window. This is suitable for the WIM format, which uses fixed-size chunks - * that are seemingly always 32768 bytes. If needed, the compressor potentially - * could be extended to support a larger and/or sliding window. - * - * 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 API are exported from the library, although 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 other open-source projects that made it possible - * to implement this code: - * - * - 7-Zip (author: Igor Pavlov), for the binary tree match-finding - * algorithm, the heuristic near-optimal forward match-choosing - * algorithm, and the block splitting algorithm. - * - * - zlib (author: Jean-loup Gailly and Mark Adler), for the hash table - * match-finding algorithm. + * - LZX has a minimum match length of 2 rather than 3. * - * - lzx-compress (author: Matthew T. Russotto), on which some parts of this - * code were originally based. + * - 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. + * + * ---------------------------------------------------------------------------- + * + * 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 @@ -156,48 +232,58 @@ #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 #ifdef ENABLE_LZX_DEBUG -# include +# include "wimlib/decompress_common.h" #endif -#include - -/* Experimental parameters not exposed through the API */ -#define LZX_PARAM_OPTIM_ARRAY_SIZE 1024 -#define LZX_PARAM_ACCOUNT_FOR_LRU 1 -#define LZX_PARAM_DONT_SKIP_MATCHES 0 -#define LZX_PARAM_USE_EMPIRICAL_DEFAULT_COSTS 1 +#define LZX_OPTIM_ARRAY_SIZE 4096 -/* Currently, this constant can't simply be changed because the code currently - * uses a static number of position slots (and may make other assumptions as - * well). */ -#define LZX_MAX_WINDOW_SIZE 32768 +#define LZX_DIV_BLOCK_SIZE 32768 -/* This may be WIM-specific */ -#define LZX_DEFAULT_BLOCK_SIZE 32768 +#define LZX_CACHE_PER_POS 8 -#define LZX_LZ_HASH_BITS 15 -#define LZX_LZ_HASH_SIZE (1 << LZX_LZ_HASH_BITS) -#define LZX_LZ_HASH_MASK (LZX_LZ_HASH_SIZE - 1) -#define LZX_LZ_HASH_SHIFT 5 +#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_MAINTREE_NUM_SYMBOLS]; - u16 len[LZX_LENTREE_NUM_SYMBOLS]; - u16 aligned[LZX_ALIGNEDTREE_NUM_SYMBOLS]; + u32 main[LZX_MAINCODE_MAX_NUM_SYMBOLS]; + u32 len[LZX_LENCODE_NUM_SYMBOLS]; + u32 aligned[LZX_ALIGNEDCODE_NUM_SYMBOLS]; }; -/* Lengths for the LZX main, length, and aligned offset Huffman codes */ +/* Codeword lengths (in bits) for the LZX main, length, and aligned offset + * Huffman codes. + * + * A 0 length means the codeword has zero frequency. + */ struct lzx_lens { - u8 main[LZX_MAINTREE_NUM_SYMBOLS]; - u8 len[LZX_LENTREE_NUM_SYMBOLS]; - u8 aligned[LZX_ALIGNEDTREE_NUM_SYMBOLS]; + u8 main[LZX_MAINCODE_MAX_NUM_SYMBOLS]; + u8 len[LZX_LENCODE_NUM_SYMBOLS]; + u8 aligned[LZX_ALIGNEDCODE_NUM_SYMBOLS]; +}; + +/* Costs for the LZX main, length, and aligned offset Huffman symbols. + * + * 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 many times. */ +struct lzx_costs { + u8 main[LZX_MAINCODE_MAX_NUM_SYMBOLS]; + u8 len[LZX_LENCODE_NUM_SYMBOLS]; + u8 aligned[LZX_ALIGNEDCODE_NUM_SYMBOLS]; }; /* The LZX main, length, and aligned offset Huffman codes */ @@ -208,13 +294,13 @@ struct lzx_codes { /* Tables for tallying symbol frequencies in the three LZX alphabets */ struct lzx_freqs { - freq_t main[LZX_MAINTREE_NUM_SYMBOLS]; - freq_t len[LZX_LENTREE_NUM_SYMBOLS]; - freq_t aligned[LZX_ALIGNEDTREE_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. @@ -224,298 +310,254 @@ struct lzx_match { * * 8-24 position footer. This is the offset of the real formatted * offset from the position base. This can be at most 17 bits - * (since lzx_extra_bits[LZX_NUM_POSITION_SLOTS - 1] is 17). + * (since lzx_extra_bits[LZX_MAX_POSITION_SLOTS - 1] is 17). * * 0-7 length of match, minus 2. This can be at most - * (LZX_MAX_MATCH - 2) == 255, so it will fit in 8 bits. */ + * (LZX_MAX_MATCH_LEN - 2) == 255, so it will fit in 8 bits. */ 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 matched text. - * - * If @len < LZX_MIN_MATCH, then it's really just a literal byte and @offset is - * meaningless. */ -struct raw_match { - u16 len; - u16 offset; -}; - -/* Specification for a LZX block */ +/* Specification for an LZX block. */ struct lzx_block_spec { - /* Set to 1 if this block has been split (in two --- we only considser - * binary splits). In such cases the rest of the fields are - * unimportant, since the relevant information is rather in the - * structures for the sub-blocks. */ - u8 is_split : 1; - /* One of the LZX_BLOCKTYPE_* constants indicating which type of this * block. */ - u8 block_type : 2; + int block_type; /* 0-based position in the window at which this block starts. */ - u16 window_pos; + u32 window_pos; /* The number of bytes of uncompressed data this block represents. */ - u16 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. */ - unsigned 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. */ - u16 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. */ - u32 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. */ - u16 link; - - /* Offset (as in a LZ (length, offset) pair) of the - * match or literal that was taken to get to this - * position in the approximate minimum-cost parse. */ - u16 match_offset; - } prev; - struct { - /* Position at which the match or literal starting at - * this position ends in the minimum-cost parse. */ - u16 link; - - /* Offset (as in a LZ (length, offset) pair) of the - * match or literal starting at this position in the - * approximate minimum-cost parse. */ - u16 match_offset; - } next; - }; -#if LZX_PARAM_ACCOUNT_FOR_LRU - struct lzx_lru_queue queue; -#endif -}; - -/* State of the LZX compressor */ +/* 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. * * 0xe8 byte preprocessing is done directly on the data here before * further compression. * - * Note that this compressor does *not* use a sliding window!!!! + * Note that this compressor does *not* use a real sliding window!!!! * It's not needed in the WIM format, since every chunk is compressed * independently. This is by design, to allow random access to the * 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. - */ - u8 window[LZX_MAX_WINDOW_SIZE + 12]; + * 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. */ - unsigned window_size; + u32 window_size; + + /* Allocated size of the @window. */ + u32 max_window_size; + + /* Number of symbols in the main alphabet (depends on the + * @max_window_size since it determines the maximum allowed offset). */ + unsigned num_main_syms; /* The current match offset LRU queue. */ struct lzx_lru_queue queue; - /* Space for sequence of matches/literals that were chosen. - * - * Each LZX_MAX_WINDOW_SIZE-sized portion of this array is used for a - * different block splitting level. */ - struct lzx_match *chosen_matches; + /* Space for the sequences of matches/literals that were chosen for each + * block. */ + struct lzx_item *chosen_items; - /* Structures used during block splitting. - * - * This can be thought of as a binary tree. block_specs[(1) - 1] - * represents to the top-level block (root node), and block_specs[(i*2) - * - 1] and block_specs[(i*2+1) - 1] represent the sub-blocks (child - * nodes) resulting from a binary split of the block represented by - * block_spec[(i) - 1]. - */ + /* Information about the LZX blocks the preprocessed input was divided + * into. */ struct lzx_block_spec *block_specs; + /* Number of LZX blocks the input was divided into; a.k.a. the number of + * elements of @block_specs that are valid. */ + unsigned num_blocks; + /* This is simply filled in with zeroes and used to avoid special-casing * the output of the first compressed Huffman code, which conceptually * has a delta taken from a code with all symbols having zero-length * codewords. */ struct lzx_codes zero_codes; - /* Slow algorithm only: The current cost model. */ - struct lzx_lens costs; - - /* Slow algorithm only: Table that maps the hash codes for 3 character - * sequences to the most recent position that sequence (or a sequence - * sharing the same hash code) appeared in the window. */ - u16 *hash_tab; - - /* Slow algorithm only: Table that maps 2-character sequences to the - * most recent position that sequence appeared in the window. */ - u16 *digram_tab; + /* The current cost model. */ + struct lzx_costs costs; - /* Slow algorithm only: Table that contains the logical child pointers - * in the binary trees in the match-finding code. - * - * child_tab[i*2] and child_tab[i*2+1] are the left and right pointers, - * respectively, from the binary tree root corresponding to window - * position i. */ - u16 *child_tab; - - /* Slow algorithm only: Matches that were already found and are saved in - * memory for subsequent queries (e.g. when block splitting). */ - struct raw_match *cached_matches; + /* Fast algorithm only: Array of hash table links. */ + u32 *prev_tab; - /* Slow algorithm only: Next position in 'cached_matches' to either - * return or fill in. */ - unsigned cached_matches_pos; + /* Slow algorithm only: Binary tree match-finder. */ + struct lz_bt mf; - /* Slow algorithm only: %true if reading from 'cached_matches'; %false - * if writing to 'cached_matches'. */ - bool matches_already_found; + /* Position in window of next match to return. */ + u32 match_window_pos; - /* Slow algorithm only: Position in window of next match to return. */ - unsigned match_window_pos; - - /* Slow algorithm only: No matches returned shall reach past this + /* The end-of-block position. We can't allow any matches to span this * position. */ - unsigned match_window_end; - - /* Slow algorithm only: Temporary space used for match-choosing - * algorithm. - * - * The size of this array must be at least LZX_MAX_MATCH but otherwise - * is arbitrary. More space simply allows the match-choosing algorithm - * to find better matches (depending on the input, as always). */ - struct lzx_optimal *optimum; - - /* Slow algorithm only: Variables used by the match-choosing algorithm. - * + 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 lz_match *cached_matches; + struct lz_match *cache_ptr; + bool matches_cached; + struct lz_match *cache_limit; + + /* 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; }; -/* Returns the LZX position slot that corresponds to a given formatted offset. - * - * Logically, this returns the smallest i such that - * formatted_offset >= lzx_position_base[i]. +/* + * Match chooser position data: * - * The actual implementation below takes advantage of the regularity of the - * numbers in the lzx_position_base array to calculate the slot directly from - * the formatted offset without actually looking at the array. + * 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 formatted_offset) +lzx_get_position_slot(u32 offset, struct lzx_lru_queue *queue) { -#if 0 - /* - * Slots 36-49 (formatted_offset >= 262144) can be found by - * (formatted_offset/131072) + 34 == (formatted_offset >> 17) + 34; - * however, this check for formatted_offset >= 262144 is commented out - * because WIM chunks cannot be that large. - */ - if (formatted_offset >= 262144) { - return (formatted_offset >> 17) + 34; - } else -#endif - { - /* Note: this part here only works if: - * - * 2 <= formatted_offset < 655360 - * - * It is < 655360 because the frequency of the position bases - * increases starting at the 655360 entry, and it is >= 2 - * because the below calculation fails if the most significant - * bit is lower than the 2's place. */ - LZX_ASSERT(2 <= formatted_offset && formatted_offset < 655360); - unsigned mssb_idx = bsr32(formatted_offset); - return (mssb_idx << 1) | - ((formatted_offset >> (mssb_idx - 1)) & 1); + unsigned position_slot; + + /* See if the offset was recently used. */ + for (int i = 0; i < LZX_NUM_RECENT_OFFSETS; i++) { + if (offset == queue->R[i]) { + /* Found it. */ + + /* Bring the repeat offset to the front of the + * queue. Note: this is, in fact, not a real + * LRU queue because repeat matches are simply + * swapped to the front. */ + swap(queue->R[0], queue->R[i]); + + /* The resulting position slot is simply the first index + * at which the offset was found in the queue. */ + return i; + } } -} -/* Compute the hash code for the next 3-character sequence in the window. */ -static unsigned -lzx_lz_compute_hash(const u8 *window) -{ - unsigned hash; - - hash = window[0]; - hash <<= LZX_LZ_HASH_SHIFT; - hash ^= window[1]; - hash <<= LZX_LZ_HASH_SHIFT; - hash ^= window[2]; - return hash & LZX_LZ_HASH_MASK; + /* The offset was not recently used; look up its real position slot. */ + position_slot = lzx_get_position_slot_raw(offset + LZX_OFFSET_OFFSET); + + /* Bring the new offset to the front of the queue. */ + for (int i = LZX_NUM_RECENT_OFFSETS - 1; i > 0; i--) + queue->R[i] = queue->R[i - 1]; + queue->R[0] = offset; + + return position_slot; } /* Build the main, length, and aligned offset Huffman codes used in LZX. * * This takes as input the frequency tables for each code and produces as output - * a set of tables that map symbols to codewords and lengths. */ + * a set of tables that map symbols to codewords and codeword lengths. */ static void lzx_make_huffman_codes(const struct lzx_freqs *freqs, - struct lzx_codes *codes) + struct lzx_codes *codes, + unsigned num_main_syms) { - make_canonical_huffman_code(LZX_MAINTREE_NUM_SYMBOLS, - LZX_MAX_CODEWORD_LEN, + make_canonical_huffman_code(num_main_syms, + LZX_MAX_MAIN_CODEWORD_LEN, freqs->main, codes->lens.main, codes->codewords.main); - make_canonical_huffman_code(LZX_LENTREE_NUM_SYMBOLS, - LZX_MAX_CODEWORD_LEN, + make_canonical_huffman_code(LZX_LENCODE_NUM_SYMBOLS, + LZX_MAX_LEN_CODEWORD_LEN, freqs->len, codes->lens.len, codes->codewords.len); - make_canonical_huffman_code(LZX_ALIGNEDTREE_NUM_SYMBOLS, 8, + make_canonical_huffman_code(LZX_ALIGNEDCODE_NUM_SYMBOLS, + LZX_MAX_ALIGNED_CODEWORD_LEN, freqs->aligned, codes->lens.aligned, codes->codewords.aligned); } /* - * Output a 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; @@ -525,40 +567,34 @@ lzx_write_match(struct output_bitstream *out, int block_type, unsigned position_slot = (match.data >> 25) & 0x3f; /* 6 bits */ unsigned len_header; unsigned len_footer; - unsigned len_pos_header; unsigned main_symbol; unsigned num_extra_bits; unsigned verbatim_bits; unsigned aligned_bits; - /* If the match length is less than MIN_MATCH (= 2) + + /* If the match length is less than MIN_MATCH_LEN (= 2) + * NUM_PRIMARY_LENS (= 7), the length header contains - * the match length minus MIN_MATCH, and there is no + * the match length minus MIN_MATCH_LEN, and there is no * length footer. * * Otherwise, the length header contains * NUM_PRIMARY_LENS, and the length footer contains * the match length minus NUM_PRIMARY_LENS minus - * MIN_MATCH. */ + * 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; } - /* Combine the position slot with the length header into - * a single symbol that will be encoded with the main - * tree. */ - len_pos_header = (position_slot << 3) | len_header; - - /* The actual main symbol is offset by LZX_NUM_CHARS because - * values under LZX_NUM_CHARS are used to indicate a literal - * byte rather than a match. */ - main_symbol = len_pos_header + LZX_NUM_CHARS; + /* Combine the position slot with the length header into a single symbol + * that will be encoded with the main code. + * + * The actual main symbol is offset by LZX_NUM_CHARS because values + * under LZX_NUM_CHARS are used to indicate a literal byte rather than a + * match. */ + main_symbol = ((position_slot << 3) | len_header) + LZX_NUM_CHARS; /* Output main symbol. */ bitstream_put_bits(out, codes->codewords.main[main_symbol], @@ -566,16 +602,15 @@ 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); /* For aligned offset blocks with at least 3 extra bits, output the * verbatim bits literally, then the aligned bits encoded using the - * aligned offset tree. Otherwise, only the verbatim bits need to be + * aligned offset code. Otherwise, only the verbatim bits need to be * output. */ if ((block_type == LZX_BLOCKTYPE_ALIGNED) && (num_extra_bits >= 3)) { @@ -594,26 +629,28 @@ 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], - unsigned num_syms, - freq_t precode_freqs[restrict LZX_PRETREE_NUM_SYMBOLS], + const unsigned num_syms, + u32 precode_freqs[restrict LZX_PRECODE_NUM_SYMBOLS], u8 output_syms[restrict num_syms], - u8 precode_lens[restrict LZX_PRETREE_NUM_SYMBOLS], - u16 precode_codewords[restrict LZX_PRETREE_NUM_SYMBOLS], - unsigned * num_additional_bits_ret) + u8 precode_lens[restrict LZX_PRECODE_NUM_SYMBOLS], + u32 precode_codewords[restrict LZX_PRECODE_NUM_SYMBOLS], + unsigned *num_additional_bits_ret) { - unsigned output_syms_idx; - unsigned cur_run_len; - unsigned i; - unsigned len_in_run; - unsigned additional_bits; - signed char delta; - unsigned num_additional_bits = 0; - memset(precode_freqs, 0, - LZX_PRETREE_NUM_SYMBOLS * sizeof(precode_freqs[0])); + LZX_PRECODE_NUM_SYMBOLS * sizeof(precode_freqs[0])); /* Since the code word lengths use a form of RLE encoding, the goal here * is to find each run of identical lengths when going through them in @@ -622,14 +659,14 @@ lzx_build_precode(const u8 lens[restrict], * literally. * * output_syms[] will be filled in with the length symbols that will be - * output, including RLE codes, not yet encoded using the pre-tree. + * output, including RLE codes, not yet encoded using the precode. * * cur_run_len keeps track of how many code word lengths are in the - * current run of identical lengths. - */ - output_syms_idx = 0; - cur_run_len = 1; - for (i = 1; i <= num_syms; i++) { + * current run of identical lengths. */ + unsigned output_syms_idx = 0; + unsigned cur_run_len = 1; + unsigned num_additional_bits = 0; + for (unsigned i = 1; i <= num_syms; i++) { if (i != num_syms && lens[i] == lens[i - 1]) { /* Still in a run--- keep going. */ @@ -642,7 +679,7 @@ lzx_build_precode(const u8 lens[restrict], /* The symbol that was repeated in the run--- not to be confused * with the length *of* the run (cur_run_len) */ - len_in_run = lens[i - 1]; + unsigned len_in_run = lens[i - 1]; if (len_in_run == 0) { /* A run of 0's. Encode it in as few length @@ -652,6 +689,7 @@ lzx_build_precode(const u8 lens[restrict], * where n is an uncompressed literal 5-bit integer that * follows the magic length. */ while (cur_run_len >= 20) { + unsigned additional_bits; additional_bits = min(cur_run_len - 20, 0x1f); num_additional_bits += 5; @@ -665,6 +703,8 @@ lzx_build_precode(const u8 lens[restrict], * where n is an uncompressed literal 4-bit integer that * follows the magic length. */ while (cur_run_len >= 4) { + unsigned additional_bits; + additional_bits = min(cur_run_len - 4, 0xf); num_additional_bits += 4; precode_freqs[17]++; @@ -685,9 +725,12 @@ lzx_build_precode(const u8 lens[restrict], * * The extra length symbol is encoded as a difference * from the length of the codeword for the first symbol - * in the run in the previous tree. + * in the run in the previous code. * */ while (cur_run_len >= 4) { + unsigned additional_bits; + signed char delta; + additional_bits = (cur_run_len > 4); num_additional_bits += 1; delta = (signed char)prev_lens[i - cur_run_len] - @@ -705,8 +748,10 @@ lzx_build_precode(const u8 lens[restrict], /* Any remaining lengths in the run are outputted without RLE, * as a difference from the length of that codeword in the - * previous tree. */ + * previous code. */ while (cur_run_len > 0) { + signed char delta; + delta = (signed char)prev_lens[i - cur_run_len] - (signed char)len_in_run; if (delta < 0) @@ -722,35 +767,44 @@ lzx_build_precode(const u8 lens[restrict], /* Build the precode from the frequencies of the length symbols. */ - make_canonical_huffman_code(LZX_PRETREE_NUM_SYMBOLS, - LZX_MAX_CODEWORD_LEN, + make_canonical_huffman_code(LZX_PRECODE_NUM_SYMBOLS, + LZX_MAX_PRE_CODEWORD_LEN, precode_freqs, precode_lens, precode_codewords); - if (num_additional_bits_ret) - *num_additional_bits_ret = num_additional_bits; + *num_additional_bits_ret = num_additional_bits; return output_syms_idx; } /* - * 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_PRETREE_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, @@ -758,13 +812,14 @@ lzx_write_compressed_code(struct output_bitstream *out, const u8 prev_lens[restrict], unsigned num_syms) { - freq_t precode_freqs[LZX_PRETREE_NUM_SYMBOLS]; + u32 precode_freqs[LZX_PRECODE_NUM_SYMBOLS]; u8 output_syms[num_syms]; - u8 precode_lens[LZX_PRETREE_NUM_SYMBOLS]; - u16 precode_codewords[LZX_PRETREE_NUM_SYMBOLS]; + u8 precode_lens[LZX_PRECODE_NUM_SYMBOLS]; + u32 precode_codewords[LZX_PRECODE_NUM_SYMBOLS]; unsigned i; unsigned num_output_syms; u8 precode_sym; + unsigned dummy; num_output_syms = lzx_build_precode(lens, prev_lens, @@ -773,12 +828,12 @@ lzx_write_compressed_code(struct output_bitstream *out, output_syms, precode_lens, precode_codewords, - NULL); + &dummy); /* Write the lengths of the precode codes to the output. */ - for (i = 0; i < LZX_PRETREE_NUM_SYMBOLS; i++) + for (i = 0; i < LZX_PRECODE_NUM_SYMBOLS; i++) bitstream_put_bits(out, precode_lens[i], - LZX_PRETREE_ELEMENT_SIZE); + LZX_PRECODE_ELEMENT_SIZE); /* Write the length symbols, encoded with the precode, to the output. */ @@ -808,90 +863,88 @@ lzx_write_compressed_code(struct output_bitstream *out, } /* - * Writes all compressed matches and literal bytes in a 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); } } - static void -lzx_assert_codes_valid(const struct lzx_codes * codes) +lzx_assert_codes_valid(const struct lzx_codes * codes, unsigned num_main_syms) { #ifdef ENABLE_LZX_DEBUG unsigned i; - for (i = 0; i < LZX_MAINTREE_NUM_SYMBOLS; i++) - LZX_ASSERT(codes->lens.main[i] <= LZX_MAX_CODEWORD_LEN); + for (i = 0; i < num_main_syms; i++) + LZX_ASSERT(codes->lens.main[i] <= LZX_MAX_MAIN_CODEWORD_LEN); - for (i = 0; i < LZX_LENTREE_NUM_SYMBOLS; i++) - LZX_ASSERT(codes->lens.len[i] <= LZX_MAX_CODEWORD_LEN); + for (i = 0; i < LZX_LENCODE_NUM_SYMBOLS; i++) + LZX_ASSERT(codes->lens.len[i] <= LZX_MAX_LEN_CODEWORD_LEN); - for (i = 0; i < LZX_ALIGNEDTREE_NUM_SYMBOLS; i++) - LZX_ASSERT(codes->lens.aligned[i] <= 8); + for (i = 0; i < LZX_ALIGNEDCODE_NUM_SYMBOLS; i++) + LZX_ASSERT(codes->lens.aligned[i] <= LZX_MAX_ALIGNED_CODEWORD_LEN); const unsigned tablebits = 10; u16 decode_table[(1 << tablebits) + - (2 * max(LZX_MAINTREE_NUM_SYMBOLS, LZX_LENTREE_NUM_SYMBOLS))] + (2 * max(num_main_syms, LZX_LENCODE_NUM_SYMBOLS))] _aligned_attribute(DECODE_TABLE_ALIGNMENT); LZX_ASSERT(0 == make_huffman_decode_table(decode_table, - LZX_MAINTREE_NUM_SYMBOLS, - tablebits, + num_main_syms, + min(tablebits, LZX_MAINCODE_TABLEBITS), codes->lens.main, - LZX_MAX_CODEWORD_LEN)); + LZX_MAX_MAIN_CODEWORD_LEN)); LZX_ASSERT(0 == make_huffman_decode_table(decode_table, - LZX_LENTREE_NUM_SYMBOLS, - tablebits, + LZX_LENCODE_NUM_SYMBOLS, + min(tablebits, LZX_LENCODE_TABLEBITS), codes->lens.len, - LZX_MAX_CODEWORD_LEN)); + LZX_MAX_LEN_CODEWORD_LEN)); LZX_ASSERT(0 == make_huffman_decode_table(decode_table, - LZX_ALIGNEDTREE_NUM_SYMBOLS, - min(tablebits, 6), + LZX_ALIGNEDCODE_NUM_SYMBOLS, + min(tablebits, LZX_ALIGNEDCODE_TABLEBITS), codes->lens.aligned, - 8)); + LZX_MAX_ALIGNED_CODEWORD_LEN)); #endif /* ENABLE_LZX_DEBUG */ } -/* Write a LZX aligned offset or verbatim block to the output. */ +/* Write an LZX aligned offset or verbatim block to the output. */ static void lzx_write_compressed_block(int block_type, unsigned block_size, - struct lzx_match * chosen_matches, - unsigned num_chosen_matches, + unsigned max_window_size, + unsigned num_main_syms, + struct lzx_item * chosen_items, + unsigned num_chosen_items, const struct lzx_codes * codes, const struct lzx_codes * prev_codes, struct output_bitstream * ostream) @@ -900,252 +953,215 @@ lzx_write_compressed_block(int block_type, LZX_ASSERT(block_type == LZX_BLOCKTYPE_ALIGNED || block_type == LZX_BLOCKTYPE_VERBATIM); - LZX_ASSERT(block_size <= LZX_MAX_WINDOW_SIZE); - LZX_ASSERT(num_chosen_matches <= LZX_MAX_WINDOW_SIZE); - lzx_assert_codes_valid(codes); + lzx_assert_codes_valid(codes, num_main_syms); /* The first three bits indicate the type of block and are one of the * LZX_BLOCKTYPE_* constants. */ - bitstream_put_bits(ostream, block_type, LZX_BLOCKTYPE_NBITS); + bitstream_put_bits(ostream, block_type, 3); - /* The next bit indicates whether the block size is the default (32768), - * indicated by a 1 bit, or whether the block size is given by the next - * 16 bits, indicated by a 0 bit. */ + /* Output the block size. + * + * The original LZX format seemed to always encode the block size in 3 + * bytes. However, the implementation in WIMGAPI, as used in WIM files, + * uses the first bit to indicate whether the block is the default size + * (32768) or a different size given explicitly by the next 16 bits. + * + * By default, this compressor uses a window size of 32768 and therefore + * follows the WIMGAPI behavior. However, this compressor also supports + * window sizes greater than 32768 bytes, which do not appear to be + * supported by WIMGAPI. In such cases, we retain the default size bit + * to mean a size of 32768 bytes but output non-default block size in 24 + * bits rather than 16. The compatibility of this behavior is unknown + * because WIMs created with chunk size greater than 32768 can seemingly + * only be opened by wimlib anyway. */ if (block_size == LZX_DEFAULT_BLOCK_SIZE) { bitstream_put_bits(ostream, 1, 1); } else { bitstream_put_bits(ostream, 0, 1); - bitstream_put_bits(ostream, block_size, LZX_BLOCKSIZE_NBITS); + + if (max_window_size >= 65536) + bitstream_put_bits(ostream, block_size >> 16, 8); + + bitstream_put_bits(ostream, block_size, 16); } /* Write out lengths of the main code. Note that the LZX specification * incorrectly states that the aligned offset code comes after the - * length code, but in fact it is the very first tree to be written + * length code, but in fact it is the very first code to be written * (before the main code). */ if (block_type == LZX_BLOCKTYPE_ALIGNED) - for (i = 0; i < LZX_ALIGNEDTREE_NUM_SYMBOLS; i++) + for (i = 0; i < LZX_ALIGNEDCODE_NUM_SYMBOLS; i++) bitstream_put_bits(ostream, codes->lens.aligned[i], - LZX_ALIGNEDTREE_ELEMENT_SIZE); + LZX_ALIGNEDCODE_ELEMENT_SIZE); LZX_DEBUG("Writing main code..."); - /* Write the pre-tree and lengths for the first LZX_NUM_CHARS symbols in + /* Write the precode and lengths for the first LZX_NUM_CHARS symbols in * the main code, which are the codewords for literal bytes. */ lzx_write_compressed_code(ostream, codes->lens.main, prev_codes->lens.main, LZX_NUM_CHARS); - /* Write the pre-tree and lengths for the rest of the main code, which + /* Write the precode and lengths for the rest of the main code, which * are the codewords for match headers. */ lzx_write_compressed_code(ostream, codes->lens.main + LZX_NUM_CHARS, prev_codes->lens.main + LZX_NUM_CHARS, - LZX_MAINTREE_NUM_SYMBOLS - LZX_NUM_CHARS); + num_main_syms - LZX_NUM_CHARS); LZX_DEBUG("Writing length code..."); - /* Write the pre-tree and lengths for the length code. */ + /* Write the precode and lengths for the length code. */ lzx_write_compressed_code(ostream, codes->lens.len, prev_codes->lens.len, - LZX_LENTREE_NUM_SYMBOLS); + LZX_LENCODE_NUM_SYMBOLS); LZX_DEBUG("Writing matches and literals..."); /* 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."); } -/* Write the LZX block of index @block_number, or write its children recursively - * if it is a split block. - * - * @prev_codes is a pointer to the Huffman codes for the most recent block - * written, or all zeroes if this is the first block. - * - * Return a pointer to the Huffman codes for the last block written. */ -static struct lzx_codes * -lzx_write_block_recursive(struct lzx_compressor *ctx, - unsigned block_number, - struct lzx_codes * prev_codes, - struct output_bitstream *ostream) +/* Write out the LZX blocks that were computed. */ +static void +lzx_write_all_blocks(struct lzx_compressor *ctx, struct output_bitstream *ostream) { - struct lzx_block_spec *spec = &ctx->block_specs[block_number - 1]; - if (spec->is_split) { - prev_codes = lzx_write_block_recursive(ctx, block_number * 2 + 0, - prev_codes, ostream); - prev_codes = lzx_write_block_recursive(ctx, block_number * 2 + 1, - prev_codes, ostream); - } else { - LZX_DEBUG("Writing block #%u (type=%d, size=%u, num_chosen_matches=%u)...", - block_number, spec->block_type, spec->block_size, - spec->num_chosen_matches); + const struct lzx_codes *prev_codes = &ctx->zero_codes; + 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_items=%u)...", + i + 1, ctx->num_blocks, + spec->block_type, spec->block_size, + spec->num_chosen_items); + lzx_write_compressed_block(spec->block_type, spec->block_size, - &ctx->chosen_matches[spec->chosen_matches_start_pos], - spec->num_chosen_matches, + ctx->max_window_size, + ctx->num_main_syms, + spec->chosen_items, + spec->num_chosen_items, &spec->codes, prev_codes, ostream); + prev_codes = &spec->codes; } - return prev_codes; -} - -/* Write out the LZX blocks that were computed. */ -static void -lzx_write_all_blocks(struct lzx_compressor *ctx, struct output_bitstream *ostream) -{ - lzx_write_block_recursive(ctx, 1, &ctx->zero_codes, ostream); } -static u32 -lzx_record_literal(u8 literal, void *_freqs) +/* Constructs an LZX match from a literal byte and updates the main code symbol + * frequencies. */ +static inline u32 +lzx_tally_literal(u8 lit, struct lzx_freqs *freqs) { - struct lzx_freqs *freqs = _freqs; - - freqs->main[literal]++; - - return (u32)literal; + freqs->main[lit]++; + return (u32)lit; } -/* Constructs a match from an offset and a length, and updates the LRU 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 +/* Constructs an LZX match from an offset and a length, and updates the LRU + * 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_record_match(unsigned match_offset, unsigned match_len, - void *_freqs, void *_queue) +static inline u32 +lzx_tally_match(unsigned match_len, u32 match_offset, + struct lzx_freqs *freqs, struct lzx_lru_queue *queue) { - struct lzx_freqs *freqs = _freqs; - struct lzx_lru_queue *queue = _queue; unsigned position_slot; - unsigned position_footer = 0; + unsigned position_footer; u32 len_header; - u32 len_pos_header; + unsigned main_symbol; unsigned len_footer; unsigned adjusted_match_len; - LZX_ASSERT(match_len >= LZX_MIN_MATCH && match_len <= LZX_MAX_MATCH); - - /* If possible, encode this offset as a repeated offset. */ - if (match_offset == queue->R0) { - position_slot = 0; - } else if (match_offset == queue->R1) { - swap(queue->R0, queue->R1); - position_slot = 1; - } else if (match_offset == queue->R2) { - swap(queue->R0, queue->R2); - position_slot = 2; - } else { - /* Not a repeated offset. */ - - /* offsets of 0, 1, and 2 are reserved for the repeated offset - * codes, so non-repeated offsets must be encoded as 3+. The - * minimum offset is 1, so encode the offsets offset by 2. */ - unsigned formatted_offset = match_offset + 2; - - queue->R2 = queue->R1; - queue->R1 = queue->R0; - queue->R0 = match_offset; - - /* The (now-formatted) offset will actually be encoded as a - * small position slot number that maps to a certain hard-coded - * offset (position base), followed by a number of extra bits--- - * the position footer--- that are added to the position base to - * get the original formatted offset. */ - - position_slot = lzx_get_position_slot(formatted_offset); - position_footer = formatted_offset & - ((1 << lzx_get_num_extra_bits(position_slot)) - 1); - } - - adjusted_match_len = match_len - LZX_MIN_MATCH; + LZX_ASSERT(match_len >= LZX_MIN_MATCH_LEN && match_len <= LZX_MAX_MATCH_LEN); + /* The match offset shall be encoded as a position slot (itself encoded + * as part of the main symbol) and a position footer. */ + position_slot = lzx_get_position_slot(match_offset, queue); + position_footer = (match_offset + LZX_OFFSET_OFFSET) & + ((1U << lzx_get_num_extra_bits(position_slot)) - 1); - /* The match length must be at least 2, so let the adjusted match length - * be the match length minus 2. - * - * If it is less than 7, the adjusted match length is encoded as a 3-bit - * number offset by 2. Otherwise, the 3-bit length header is all 1's - * and the actual adjusted length is given as a symbol encoded with the - * length tree, offset by 7. - */ + /* The match length shall be encoded as a length header (itself encoded + * as part of the main symbol) and an optional length footer. */ + adjusted_match_len = match_len - LZX_MIN_MATCH_LEN; if (adjusted_match_len < LZX_NUM_PRIMARY_LENS) { + /* No length footer needed. */ len_header = adjusted_match_len; } else { + /* Length footer needed. It will be encoded using the length + * code. */ len_header = LZX_NUM_PRIMARY_LENS; len_footer = adjusted_match_len - LZX_NUM_PRIMARY_LENS; freqs->len[len_footer]++; } - len_pos_header = (position_slot << 3) | len_header; - freqs->main[len_pos_header + LZX_NUM_CHARS]++; + /* Account for the main symbol. */ + main_symbol = ((position_slot << 3) | len_header) + LZX_NUM_CHARS; + + freqs->main[main_symbol]++; - /* Equivalent to: - * if (lzx_extra_bits[position_slot] >= 3) */ + /* In an aligned offset block, 3 bits of the position footer are output + * as an aligned offset symbol. Account for this, although we may + * ultimately decide to output the block as verbatim. */ + + /* The following check is equivalent to: + * + * if (lzx_extra_bits[position_slot] >= 3) + * + * Note that this correctly excludes position slots that correspond to + * recent offsets. */ if (position_slot >= 8) freqs->aligned[position_footer & 7]++; /* Pack the position slot, position footer, and match length into an - * intermediate representation. - * - * bits description - * ---- ----------------------------------------------------------- - * - * 31 1 if a match, 0 if a literal. - * - * 30-25 position slot. This can be at most 50, so it will fit in 6 - * bits. - * - * 8-24 position footer. This is the offset of the real formatted - * offset from the position base. This can be at most 17 bits - * (since lzx_extra_bits[LZX_NUM_POSITION_SLOTS - 1] is 17). - * - * 0-7 length of match, offset by 2. This can be at most - * (LZX_MAX_MATCH - 2) == 255, so it will fit in 8 bits. */ + * 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); + LZX_ASSERT(LZX_MAX_MATCH_LEN - LZX_MIN_MATCH_LEN + 1 <= 256); + + LZX_ASSERT(position_slot <= (1U << (31 - 25)) - 1); + LZX_ASSERT(position_footer <= (1U << (25 - 8)) - 1); + LZX_ASSERT(adjusted_match_len <= (1U << (8 - 0)) - 1); return 0x80000000 | (position_slot << 25) | (position_footer << 8) | (adjusted_match_len); } -/* Set the cost model @ctx->costs from the Huffman codeword lengths specified in - * @lens. - * - * These are basically the same thing, except that the Huffman codewords with - * length 0 correspond to symbols with zero frequency that still need to be - * assigned actual costs. The specific values assigned are arbitrary, but they - * should be fairly high (near the maximum codeword length) to take into account - * the fact that uses of these symbols are expected to be rare. - */ +struct lzx_record_ctx { + struct lzx_freqs freqs; + struct lzx_lru_queue queue; + struct lzx_item *matches; +}; + static void -lzx_set_costs(struct lzx_compressor * ctx, const struct lzx_lens * lens) +lzx_record_match(unsigned len, unsigned offset, void *_ctx) { - unsigned i; - - memcpy(&ctx->costs, lens, sizeof(struct lzx_lens)); + struct lzx_record_ctx *ctx = _ctx; - for (i = 0; i < LZX_MAINTREE_NUM_SYMBOLS; i++) - if (ctx->costs.main[i] == 0) - ctx->costs.main[i] = ctx->params.alg_params.slow.main_nostat_cost; + (ctx->matches++)->data = lzx_tally_match(len, offset, &ctx->freqs, &ctx->queue); +} - for (i = 0; i < LZX_LENTREE_NUM_SYMBOLS; i++) - if (ctx->costs.len[i] == 0) - ctx->costs.len[i] = ctx->params.alg_params.slow.len_nostat_cost; +static void +lzx_record_literal(u8 lit, void *_ctx) +{ + struct lzx_record_ctx *ctx = _ctx; - for (i = 0; i < LZX_ALIGNEDTREE_NUM_SYMBOLS; i++) - if (ctx->costs.aligned[i] == 0) - ctx->costs.aligned[i] = ctx->params.alg_params.slow.aligned_nostat_cost; + (ctx->matches++)->data = lzx_tally_literal(lit, &ctx->freqs); } +/* Returns the cost, in bits, to output a literal byte using the specified cost + * model. */ static u32 -lzx_literal_cost(u8 c, const struct lzx_lens * costs) +lzx_literal_cost(u8 c, const struct lzx_costs * costs) { return costs->main[c]; } @@ -1153,444 +1169,188 @@ lzx_literal_cost(u8 c, const struct lzx_lens * costs) /* Given a (length, offset) pair that could be turned into a valid LZX match as * 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. */ -static unsigned -lzx_match_cost(unsigned length, unsigned offset, const struct lzx_lens *costs - -#if LZX_PARAM_ACCOUNT_FOR_LRU - , struct lzx_lru_queue *queue -#endif - ) + * match in the compressed output. Take into account the match offset LRU + * 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, len_header, main_symbol; - unsigned cost = 0; - - /* Calculate position slot and length header, then combine them into the - * main symbol. */ - -#if LZX_PARAM_ACCOUNT_FOR_LRU - if (offset == queue->R0) { - position_slot = 0; - } else if (offset == queue->R1) { - swap(queue->R0, queue->R1); - position_slot = 1; - } else if (offset == queue->R2) { - swap(queue->R0, queue->R2); - position_slot = 2; - } else -#endif - position_slot = lzx_get_position_slot(offset + 2); + unsigned position_slot; + unsigned len_header, main_symbol; + unsigned num_extra_bits; + u32 cost = 0; - len_header = min(length - LZX_MIN_MATCH, LZX_NUM_PRIMARY_LENS); + position_slot = lzx_get_position_slot(offset, queue); + + len_header = min(length - LZX_MIN_MATCH_LEN, LZX_NUM_PRIMARY_LENS); main_symbol = ((position_slot << 3) | len_header) + LZX_NUM_CHARS; /* Account for main symbol. */ 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_MIN_MATCH) & 7]; + cost += costs->aligned[(offset + LZX_OFFSET_OFFSET) & 7]; } else { cost += num_extra_bits; } /* Account for extra length information. */ - if (length - LZX_MIN_MATCH >= LZX_NUM_PRIMARY_LENS) - cost += costs->len[length - LZX_MIN_MATCH - LZX_NUM_PRIMARY_LENS]; + if (len_header == LZX_NUM_PRIMARY_LENS) + cost += costs->len[length - LZX_MIN_MATCH_LEN - LZX_NUM_PRIMARY_LENS]; return cost; + } -/* This procedure effectively creates a new binary tree corresponding to the - * current string at the same time that it searches the existing tree nodes for - * matches. This is the same algorithm as that used in GetMatchesSpec1() in - * 7-Zip, but it is hopefully explained a little more clearly below. */ -static unsigned -lzx_lz_get_matches(const u8 window[restrict], - const unsigned bytes_remaining, - const unsigned strstart, - const unsigned max_length, - u16 child_tab[restrict], - unsigned cur_match, - const unsigned prev_len, - struct raw_match * const matches) +/* Set the cost model @ctx->costs from the Huffman codeword lengths specified in + * @lens. + * + * The cost model and codeword lengths are almost the same thing, but the + * Huffman codewords with length 0 correspond to symbols with zero frequency + * that still need to be assigned actual costs. The specific values assigned + * are arbitrary, but they should be fairly high (near the maximum codeword + * length) to take into account the fact that uses of these symbols are expected + * to be rare. */ +static void +lzx_set_costs(struct lzx_compressor * ctx, const struct lzx_lens * lens) { - u16 *new_tree_lt_ptr = &child_tab[strstart * 2]; - u16 *new_tree_gt_ptr = &child_tab[strstart * 2 + 1]; + unsigned i; + unsigned num_main_syms = ctx->num_main_syms; - u16 longest_lt_match_len = 0; - u16 longest_gt_match_len = 0; + /* Main code */ + for (i = 0; i < num_main_syms; i++) { + ctx->costs.main[i] = lens->main[i]; + if (ctx->costs.main[i] == 0) + ctx->costs.main[i] = ctx->params.alg_params.slow.main_nostat_cost; + } - /* Maximum number of nodes to walk down before stopping */ - unsigned depth = max_length; + /* Length code */ + for (i = 0; i < LZX_LENCODE_NUM_SYMBOLS; i++) { + ctx->costs.len[i] = lens->len[i]; + if (ctx->costs.len[i] == 0) + ctx->costs.len[i] = ctx->params.alg_params.slow.len_nostat_cost; + } - /* Length of longest match found so far */ - unsigned longest_match_len = prev_len; - - /* Maximum length of match to return */ - unsigned len_limit = min(bytes_remaining, max_length); - - /* Number of matches found so far */ - unsigned num_matches = 0; - - for (;;) { - - /* Stop if too many nodes were traversed or if there is no next - * node */ - if (depth-- == 0 || cur_match == 0) { - *new_tree_gt_ptr = 0; - *new_tree_lt_ptr = 0; - return num_matches; - } - - /* Load the pointers to the children of the binary tree node - * corresponding to the current match */ - u16 * const cur_match_ptrs = &child_tab[cur_match * 2]; - - /* Set up pointers to the current match and to the current - * string */ - const u8 * const matchptr = &window[cur_match]; - const u8 * const strptr = &window[strstart]; - - /* Determine position at which to start comparing */ - u16 len = min(longest_lt_match_len, - longest_gt_match_len); - - if (matchptr[len] == strptr[len]) { - - /* Extend the match as far as possible. */ - while (++len != len_limit) - if (matchptr[len] != strptr[len]) - break; - - /* Record this match if it is the longest found so far. - */ - if (len > longest_match_len) { - longest_match_len = len; - matches[num_matches].len = len; - matches[num_matches].offset = strstart - cur_match; - num_matches++; - - if (len == len_limit) { - /* Length limit was reached. Link left pointer - * in the new tree with left subtree of current - * match tree, and link the right pointer in the - * new tree with the right subtree of the - * current match tree. This in effect deletes - * the node for the currrent match, which is - * desirable because the current match is the - * same as the current string up until the - * length limit, so in subsequent queries it - * will never be preferable to the current - * position. */ - *new_tree_lt_ptr = cur_match_ptrs[0]; - *new_tree_gt_ptr = cur_match_ptrs[1]; - return num_matches; - } - } - } - - if (matchptr[len] < strptr[len]) { - /* Case 1: The current match is lexicographically less - * than the current string. - * - * Since we are searching the binary tree structures, we - * need to walk down to the *right* subtree of the - * current match's node to get to a match that is - * lexicographically *greater* than the current match - * but still lexicographically *lesser* than the current - * string. - * - * At the same time, we link the entire binary tree - * corresponding to the current match into the - * appropriate place in the new binary tree being built - * for the current string. */ - *new_tree_lt_ptr = cur_match; - new_tree_lt_ptr = &cur_match_ptrs[1]; - cur_match = *new_tree_lt_ptr; - longest_lt_match_len = len; - } else { - /* Case 2: The current match is lexicographically - * greater than the current string. - * - * This is analogous to Case 1 above, but everything - * happens in the other direction. - */ - *new_tree_gt_ptr = cur_match; - new_tree_gt_ptr = &cur_match_ptrs[0]; - cur_match = *new_tree_gt_ptr; - longest_gt_match_len = len; - } - } -} - -/* Equivalent to lzx_lz_get_matches(), but only updates the tree and doesn't - * return matches. See that function for details (including comments). */ -static void -lzx_lz_skip_matches(const u8 window[restrict], - const unsigned bytes_remaining, - const unsigned strstart, - const unsigned max_length, - u16 child_tab[restrict], - unsigned cur_match, - const unsigned prev_len) -{ - u16 *new_tree_lt_ptr = &child_tab[strstart * 2]; - u16 *new_tree_gt_ptr = &child_tab[strstart * 2 + 1]; - - u16 longest_lt_match_len = 0; - u16 longest_gt_match_len = 0; - - unsigned depth = max_length; - - unsigned longest_match_len = prev_len; - - unsigned len_limit = min(bytes_remaining, max_length); - - for (;;) { - if (depth-- == 0 || cur_match == 0) { - *new_tree_gt_ptr = 0; - *new_tree_lt_ptr = 0; - return; - } - - u16 * const cur_match_ptrs = &child_tab[cur_match * 2]; - - const u8 * const matchptr = &window[cur_match]; - const u8 * const strptr = &window[strstart]; - - u16 len = min(longest_lt_match_len, - longest_gt_match_len); - - if (matchptr[len] == strptr[len]) { - while (++len != len_limit) - if (matchptr[len] != strptr[len]) - break; - - if (len > longest_match_len) { - longest_match_len = len; - - if (len == len_limit) { - *new_tree_lt_ptr = cur_match_ptrs[0]; - *new_tree_gt_ptr = cur_match_ptrs[1]; - return; - } - } - } - - if (matchptr[len] < strptr[len]) { - *new_tree_lt_ptr = cur_match; - new_tree_lt_ptr = &cur_match_ptrs[1]; - cur_match = *new_tree_lt_ptr; - longest_lt_match_len = len; - } else { - *new_tree_gt_ptr = cur_match; - new_tree_gt_ptr = &cur_match_ptrs[0]; - cur_match = *new_tree_gt_ptr; - longest_gt_match_len = len; - } - } -} - -static unsigned -lzx_lz_get_matches_caching(struct lzx_compressor *ctx, - struct raw_match **matches_ret); - -/* 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) -{ - -#if LZX_PARAM_DONT_SKIP_MATCHES - /* Option 1: Still cache the matches from the positions skipped. They - * will then be available in later passes. */ - struct raw_match *matches; - while (n--) - lzx_lz_get_matches_caching(ctx, &matches); -#else - /* Option 2: Mark the positions skipped as having no matches available, - * but we still need to update the binary tree in case subsequent - * positions have matches at the current position. */ - LZX_ASSERT(n <= ctx->match_window_end - ctx->match_window_pos); - if (ctx->matches_already_found) { - while (n--) { - LZX_ASSERT(ctx->cached_matches[ctx->cached_matches_pos].offset == - ctx->match_window_pos); - ctx->cached_matches_pos += ctx->cached_matches[ctx->cached_matches_pos].len + 1; - ctx->match_window_pos++; - } - } else { - while (n--) { - if (ctx->params.alg_params.slow.use_len2_matches && - ctx->match_window_end - ctx->match_window_pos >= 2) { - unsigned c1 = ctx->window[ctx->match_window_pos]; - unsigned c2 = ctx->window[ctx->match_window_pos + 1]; - unsigned digram = c1 | (c2 << 8); - ctx->digram_tab[digram] = ctx->match_window_pos; - } - if (ctx->match_window_end - ctx->match_window_pos >= 3) { - unsigned hash; - unsigned cur_match; - - hash = lzx_lz_compute_hash(&ctx->window[ctx->match_window_pos]); - - cur_match = ctx->hash_tab[hash]; - ctx->hash_tab[hash] = ctx->match_window_pos; - - lzx_lz_skip_matches(ctx->window, - ctx->match_window_end - ctx->match_window_pos, - ctx->match_window_pos, - ctx->params.alg_params.slow.num_fast_bytes, - ctx->child_tab, - cur_match, 1); - } - ctx->cached_matches[ctx->cached_matches_pos].len = 0; - ctx->cached_matches[ctx->cached_matches_pos].offset = ctx->match_window_pos; - ctx->cached_matches_pos++; - ctx->match_window_pos++; - } + /* Aligned offset code */ + for (i = 0; i < LZX_ALIGNEDCODE_NUM_SYMBOLS; i++) { + ctx->costs.aligned[i] = lens->aligned[i]; + if (ctx->costs.aligned[i] == 0) + ctx->costs.aligned[i] = ctx->params.alg_params.slow.aligned_nostat_cost; } -#endif /* !LZX_PARAM_DONT_SKIP_MATCHES */ } /* Retrieve a list of matches available at the next position in the input. * - * The return value is the number of matches found, and a pointer to them is - * written to @matches_ret. The matches will be sorted in order by length. - * - * This is essentially a wrapper around lzx_lz_get_matches() that caches its - * output the first time and also performs the needed hashing. - */ + * 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_caching(struct lzx_compressor *ctx, - struct raw_match **matches_ret) +lzx_get_matches(struct lzx_compressor *ctx, + const struct lz_match **matches_ret) { + struct lz_match *cache_ptr; + struct lz_match *matches; unsigned num_matches; - struct raw_match *matches; - - LZX_ASSERT(ctx->match_window_end >= ctx->match_window_pos); - matches = &ctx->cached_matches[ctx->cached_matches_pos + 1]; + LZX_ASSERT(ctx->match_window_pos < ctx->match_window_end); - if (ctx->matches_already_found) { - num_matches = ctx->cached_matches[ctx->cached_matches_pos].len; - LZX_ASSERT(ctx->cached_matches[ctx->cached_matches_pos].offset == ctx->match_window_pos); - - for (int i = (int)num_matches - 1; i >= 0; i--) { - if (ctx->match_window_pos + matches[i].len > ctx->match_window_end) - matches[i].len = ctx->match_window_end - ctx->match_window_pos; - else - break; + 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 { - unsigned prev_len = 1; - struct raw_match * matches_ret = &ctx->cached_matches[ctx->cached_matches_pos + 1]; num_matches = 0; + } - if (ctx->params.alg_params.slow.use_len2_matches && - ctx->match_window_end - ctx->match_window_pos >= 3) { - unsigned c1 = ctx->window[ctx->match_window_pos]; - unsigned c2 = ctx->window[ctx->match_window_pos + 1]; - unsigned digram = c1 | (c2 << 8); - unsigned cur_match; - - cur_match = ctx->digram_tab[digram]; - ctx->digram_tab[digram] = ctx->match_window_pos; - if (cur_match != 0 && - ctx->window[cur_match + 2] != ctx->window[ctx->match_window_pos + 2]) - { - matches_ret->len = 2; - matches_ret->offset = ctx->match_window_pos - cur_match; - matches_ret++; - num_matches++; - prev_len = 2; - } - } - if (ctx->match_window_end - ctx->match_window_pos >= 3) { - unsigned hash; - unsigned cur_match; - - hash = lzx_lz_compute_hash(&ctx->window[ctx->match_window_pos]); - - cur_match = ctx->hash_tab[hash]; - ctx->hash_tab[hash] = ctx->match_window_pos; - num_matches += lzx_lz_get_matches(ctx->window, - ctx->match_window_end - ctx->match_window_pos, - ctx->match_window_pos, - ctx->params.alg_params.slow.num_fast_bytes, - ctx->child_tab, - cur_match, - prev_len, - matches_ret); - } + /* 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; - ctx->cached_matches[ctx->cached_matches_pos].len = num_matches; - ctx->cached_matches[ctx->cached_matches_pos].offset = ctx->match_window_pos; + if (limit >= LZX_MIN_MATCH_LEN) { - if (num_matches) { - struct raw_match *longest_match_ptr = - &ctx->cached_matches[ctx->cached_matches_pos + 1 + - num_matches - 1]; - u16 len = longest_match_ptr->len; - - /* If the longest match returned by the match-finder - * reached the number of fast bytes, extend it as much - * as possible. */ - if (len == ctx->params.alg_params.slow.num_fast_bytes) { - const unsigned maxlen = - min(ctx->match_window_end - ctx->match_window_pos, - LZX_MAX_MATCH); - - const u8 * const matchptr = - &ctx->window[ctx->match_window_pos - longest_match_ptr->offset]; - - const u8 * const strptr = - &ctx->window[ctx->match_window_pos]; - - while (len < maxlen && matchptr[len] == strptr[len]) - len++; - } - longest_match_ptr->len = len; + unsigned i = num_matches - 1; + do { + if (matches[i].len >= limit) { + matches[i].len = limit; + + /* Truncation might produce multiple + * matches with length 'limit'. Keep at + * most 1. */ + num_matches = i + 1; + } + } while (i--); + } else { + num_matches = 0; } + cache_ptr->len = num_matches; } - ctx->cached_matches_pos += num_matches + 1; - *matches_ret = matches; #if 0 - printf("\n"); + fprintf(stderr, "Pos %u/%u: %u matches\n", + ctx->match_window_pos, ctx->window_size, num_matches); for (unsigned i = 0; i < num_matches; i++) - { - printf("Len %u Offset %u\n", matches[i].len, matches[i].offset); - } + fprintf(stderr, "\tLen %u Offset %u\n", matches[i].len, matches[i].offset); #endif +#ifdef ENABLE_LZX_DEBUG for (unsigned i = 0; i < num_matches; i++) { - LZX_ASSERT(matches[i].len <= LZX_MAX_MATCH); - if (matches[i].len >= LZX_MIN_MATCH) { - LZX_ASSERT(matches[i].offset <= ctx->match_window_pos); - LZX_ASSERT(matches[i].len <= ctx->match_window_end - ctx->match_window_pos); - LZX_ASSERT(!memcmp(&ctx->window[ctx->match_window_pos], - &ctx->window[ctx->match_window_pos - matches[i].offset], - matches[i].len)); + LZX_ASSERT(matches[i].len >= LZX_MIN_MATCH_LEN); + LZX_ASSERT(matches[i].len <= LZX_MAX_MATCH_LEN); + LZX_ASSERT(matches[i].len <= ctx->match_window_end - ctx->match_window_pos); + LZX_ASSERT(matches[i].offset > 0); + LZX_ASSERT(matches[i].offset <= ctx->match_window_pos); + 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; @@ -1615,88 +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 "best" match or literal to use at the next position in the input. + * Choose an approximately optimal match or literal to use at the next position + * in the string, or "window", being LZ-encoded. * - * Unlike a "greedy" parser that always takes the longest match, or even a + * This is based on algorithms used in 7-Zip, including the DEFLATE encoder + * and the LZMA encoder, written by Igor Pavlov. + * + * 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 best 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. It is not a true - * "optimal" parser, however, since some shortcuts can be taken; for example, if - * a match is very long, the parser just chooses it immediately before too much - * time is wasting considering many different alternatives that are unlikely to - * be better. - * - * This algorithm is based on that used in 7-Zip's DEFLATE encoder. + * 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) - * ctx->hash_tab (must set to 0 before first call) - * ctx->cached_matches (internal state; leave uninitialized) - * ctx->cached_matches_pos (initialize to 0 before first call; save and - * restore value if restarting parse from a - * certain position) - * ctx->match_window_pos (must initialize to position of next match to - * return; subsequent calls return subsequent - * matches) - * ctx->match_window_end (must initialize to limit of match-finding region; - * subsequent calls use the same limit) + * function. * * The return value is a (length, offset) pair specifying the match or literal - * chosen. For literals, length is either 0 or 1 and 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) { -#if 0 - /* Testing: literals only */ - ctx->match_window_pos++; - return (struct raw_match) { .len = 0 }; -#elif 0 - /* Testing: greedy parsing */ - struct raw_match *matches; unsigned num_matches; - struct raw_match match = {.len = 0}; - - num_matches = lzx_lz_get_matches_caching(ctx, &matches); - if (num_matches) { - match = matches[num_matches - 1]; - lzx_lz_skip_bytes(ctx, match.len - 1); - } - return match; -#else - unsigned num_possible_matches; - struct raw_match *possible_matches; - struct raw_match match; - unsigned longest_match_len; - unsigned len, match_idx; + 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. */ @@ -1713,693 +1470,502 @@ 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, &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 by length. Get the length of - * the longest one. */ - longest_match_len = possible_matches[num_possible_matches - 1].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[num_possible_matches - 1]; + /* 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. */ -#if LZX_PARAM_ACCOUNT_FOR_LRU - ctx->optimum[0].queue = ctx->queue; - ctx->optimum[1].queue = ctx->optimum[0].queue; -#endif - 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. */ - match_idx = 0; - BUILD_BUG_ON(LZX_MIN_MATCH != 2); - for (len = LZX_MIN_MATCH; len <= longest_match_len; len++) { - - LZX_ASSERT(match_idx < num_possible_matches); + * 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; + } - #if LZX_PARAM_ACCOUNT_FOR_LRU - ctx->optimum[len].queue = ctx->optimum[0].queue; - #endif - 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 - #if LZX_PARAM_ACCOUNT_FOR_LRU - , &ctx->optimum[len].queue - #endif - ); - if (len == possible_matches[match_idx].len) - match_idx++; - } + 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_PARAM_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, - &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[num_possible_matches - 1].len; + /* Skip over the remaining bytes of the long match. */ + lzx_skip_bytes(ctx, longest_rep_len); - /* 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) { + /* Return first match in the list. */ + return match; + } + + /* 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[num_possible_matches - 1].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. */ - u32 cur_cost = ctx->optimum[cur_pos].cost; - u32 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 LZX_PARAM_ACCOUNT_FOR_LRU + if (cost < ctx->optimum[cur_pos + 1].cost) { ctx->optimum[cur_pos + 1].queue = ctx->optimum[cur_pos].queue; - #endif + 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 = ~(u32)0; - - match_idx = 0; - for (len = LZX_MIN_MATCH; len <= new_len; len++) { - LZX_ASSERT(match_idx < num_possible_matches); - #if LZX_PARAM_ACCOUNT_FOR_LRU - struct lzx_lru_queue q = ctx->optimum[cur_pos].queue; - #endif - u32 cost = cur_cost + lzx_match_cost(len, - possible_matches[match_idx].offset, - &ctx->costs - #if LZX_PARAM_ACCOUNT_FOR_LRU - , &q - #endif - ); - - 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; - #if LZX_PARAM_ACCOUNT_FOR_LRU - ctx->optimum[cur_pos + len].queue = q; - #endif + /* 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); } - } -#endif -} - -static unsigned -lzx_huffman_code_output_cost(const u8 lens[restrict], - const freq_t freqs[restrict], - unsigned num_syms) -{ - unsigned cost = 0; - - for (unsigned i = 0; i < num_syms; i++) - cost += (unsigned)lens[i] * (unsigned)freqs[i]; - - return cost; -} -/* Return the number of bits required to output the lengths for the specified - * Huffman code in compressed format (encoded with a precode). */ -static unsigned -lzx_code_cost(const u8 lens[], const u8 prev_lens[], unsigned num_syms) -{ - u8 output_syms[num_syms]; - freq_t precode_freqs[LZX_PRETREE_NUM_SYMBOLS]; - u8 precode_lens[LZX_PRETREE_NUM_SYMBOLS]; - u16 precode_codewords[LZX_PRETREE_NUM_SYMBOLS]; - unsigned cost = 0; - unsigned num_additional_bits; - - /* Acount for the lengths of the precode itself. */ - cost += LZX_PRETREE_NUM_SYMBOLS * LZX_PRETREE_ELEMENT_SIZE; - - lzx_build_precode(lens, prev_lens, num_syms, - precode_freqs, output_syms, - precode_lens, precode_codewords, - &num_additional_bits); - - /* Account for all precode symbols output. */ - cost += lzx_huffman_code_output_cost(precode_lens, precode_freqs, - LZX_PRETREE_NUM_SYMBOLS); - - /* Account for additional bits. */ - cost += num_additional_bits; - - return cost; + 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; + } + } + } } -/* Account for extra bits in the main symbols. */ +/* Set default symbol costs for the LZX Huffman codes. */ static void -lzx_update_mainsym_match_costs(int block_type, - u8 main_lens[LZX_MAINTREE_NUM_SYMBOLS]) +lzx_set_default_costs(struct lzx_costs * costs, unsigned num_main_syms) { unsigned i; - LZX_ASSERT(block_type == LZX_BLOCKTYPE_ALIGNED || - block_type == LZX_BLOCKTYPE_VERBATIM); + /* Main code (part 1): Literal symbols */ + for (i = 0; i < LZX_NUM_CHARS; i++) + costs->main[i] = 8; - for (i = LZX_NUM_CHARS; i < LZX_MAINTREE_NUM_SYMBOLS; i++) { - unsigned position_slot = (i >> 3) & 0x1f; + /* Main code (part 2): Match header symbols */ + for (; i < num_main_syms; i++) + costs->main[i] = 10; - /* If it's a verbatim block, add the number of extra bits - * corresponding to the position slot. - * - * If it's an aligned block and there would normally be at least - * 3 extra bits, count 3 less because they will be output as an - * aligned offset symbol instead. */ - unsigned num_extra_bits = lzx_get_num_extra_bits(position_slot); - - if (block_type == LZX_BLOCKTYPE_ALIGNED && num_extra_bits >= 3) - num_extra_bits -= 3; - main_lens[i] += num_extra_bits; - } + /* Length code */ + for (i = 0; i < LZX_LENCODE_NUM_SYMBOLS; i++) + costs->len[i] = 8; + + /* Aligned offset code */ + for (i = 0; i < LZX_ALIGNEDCODE_NUM_SYMBOLS; i++) + costs->aligned[i] = 3; } -/* - * Compute the costs, in bits, to output a compressed block as aligned offset - * and verbatim. - * - * @block_size - * Number of bytes of uncompressed data the block represents. - * @codes - * Huffman codes that will be used when outputting the block. - * @prev_codes - * Huffman codes for the previous block, or all zeroes if this is the first - * block. - * @freqs - * Frequencies of Huffman symbols that will be output in the block. - * @aligned_cost_ret - * Cost of aligned block will be returned here. - * @verbatim_cost_ret - * Cost of verbatim block will be returned here. - */ -static void -lzx_compute_compressed_block_costs(unsigned block_size, - const struct lzx_codes *codes, - const struct lzx_codes *prev_codes, - const struct lzx_freqs *freqs, - unsigned * aligned_cost_ret, - unsigned * verbatim_cost_ret) +/* 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) { - unsigned common_cost = 0; unsigned aligned_cost = 0; unsigned verbatim_cost = 0; - u8 updated_main_lens[LZX_MAINTREE_NUM_SYMBOLS]; - - /* Account for cost of block header. */ - common_cost += LZX_BLOCKTYPE_NBITS; - if (block_size == LZX_DEFAULT_BLOCK_SIZE) - common_cost += 1; + /* 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 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]; + } + aligned_cost += LZX_ALIGNEDCODE_ELEMENT_SIZE * LZX_ALIGNEDCODE_NUM_SYMBOLS; + if (aligned_cost < verbatim_cost) + return LZX_BLOCKTYPE_ALIGNED; else - common_cost += LZX_BLOCKSIZE_NBITS; - - /* Account for cost of outputting aligned offset code. */ - aligned_cost += LZX_ALIGNEDTREE_NUM_SYMBOLS * LZX_ALIGNEDTREE_ELEMENT_SIZE; - - /* Account for cost of outputting main and length codes. */ - common_cost += lzx_code_cost(codes->lens.main, - prev_codes->lens.main, - LZX_NUM_CHARS); - common_cost += lzx_code_cost(codes->lens.main + LZX_NUM_CHARS, - prev_codes->lens.main + LZX_NUM_CHARS, - LZX_MAINTREE_NUM_SYMBOLS - LZX_NUM_CHARS); - common_cost += lzx_code_cost(codes->lens.len, - prev_codes->lens.len, - LZX_LENTREE_NUM_SYMBOLS); - - /* Account for cost to output main, length, and aligned symbols, taking - * into account extra position bits. */ - - memcpy(updated_main_lens, codes->lens.main, LZX_MAINTREE_NUM_SYMBOLS); - lzx_update_mainsym_match_costs(LZX_BLOCKTYPE_VERBATIM, updated_main_lens); - verbatim_cost += lzx_huffman_code_output_cost(updated_main_lens, - freqs->main, - LZX_MAINTREE_NUM_SYMBOLS); - memcpy(updated_main_lens, codes->lens.main, LZX_MAINTREE_NUM_SYMBOLS); - lzx_update_mainsym_match_costs(LZX_BLOCKTYPE_ALIGNED, updated_main_lens); - aligned_cost += lzx_huffman_code_output_cost(updated_main_lens, - freqs->main, - LZX_MAINTREE_NUM_SYMBOLS); - - common_cost += lzx_huffman_code_output_cost(codes->lens.len, - freqs->len, - LZX_LENTREE_NUM_SYMBOLS); - - aligned_cost += lzx_huffman_code_output_cost(codes->lens.aligned, - freqs->aligned, - LZX_ALIGNEDTREE_NUM_SYMBOLS); - - *aligned_cost_ret = aligned_cost + common_cost; - *verbatim_cost_ret = verbatim_cost + common_cost; + return LZX_BLOCKTYPE_VERBATIM; } -/* Prepare a (nonsplit) compressed block. */ -static unsigned -lzx_prepare_compressed_block(struct lzx_compressor *ctx, unsigned block_number, - struct lzx_codes *prev_codes) +/* Find a near-optimal sequence of matches/literals with which to output the + * 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) { - struct lzx_block_spec *spec = &ctx->block_specs[block_number - 1]; - unsigned orig_cached_matches_pos = ctx->cached_matches_pos; - struct lzx_lru_queue orig_queue = ctx->queue; + const struct lzx_lru_queue orig_queue = ctx->queue; + unsigned num_passes_remaining = num_passes; struct lzx_freqs freqs; - unsigned cost; - - /* Here's where the real work happens. The following loop runs one or - * more times, each time using a cost model based on the Huffman codes - * computed from the previous iteration (the first iteration uses a - * default model). Each iteration of the loop uses a heuristic - * algorithm to divide the block into near-optimal matches/literals from - * beginning to end. */ - LZX_ASSERT(ctx->params.alg_params.slow.num_optim_passes >= 1); - spec->num_chosen_matches = 0; - for (unsigned pass = 0; pass < ctx->params.alg_params.slow.num_optim_passes; pass++) - { - LZX_DEBUG("Block %u: Match-choosing pass %u of %u", - block_number, pass + 1, - ctx->params.alg_params.slow.num_optim_passes); - - /* Reset frequency tables. */ - memset(&freqs, 0, sizeof(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; - /* Reset match offset LRU queue. */ - ctx->queue = orig_queue; + LZX_ASSERT(num_passes >= 1); + LZX_ASSERT(lz_bt_get_position(&ctx->mf) == spec->window_pos); - /* Reset match-finding position. */ - ctx->cached_matches_pos = orig_cached_matches_pos; - ctx->match_window_pos = spec->window_pos; - ctx->match_window_end = spec->window_pos + spec->block_size; - - /* Set cost model. */ - lzx_set_costs(ctx, &spec->codes.lens); + ctx->match_window_end = spec->window_pos + spec->block_size; + ctx->matches_cached = false; - unsigned window_pos = spec->window_pos; - unsigned end = window_pos + spec->block_size; - - while (window_pos < end) { - struct raw_match match; - struct lzx_match lzx_match; - - match = lzx_lz_get_near_optimal_match(ctx); - - if (match.len >= LZX_MIN_MATCH) { + /* 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. + * + * To improve performance we only generate the array containing the + * matches and literals in intermediate form on the final pass. */ - /* Best to output a match here. */ + 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; - LZX_ASSERT(match.len <= LZX_MAX_MATCH); - LZX_ASSERT(!memcmp(&ctx->window[window_pos], - &ctx->window[window_pos - match.offset], - match.len)); + while (window_ptr != window_end) { - /* Tally symbol frequencies. */ - lzx_match.data = lzx_record_match(match.offset, - match.len, - &freqs, - &ctx->queue); + lz_match = lzx_get_near_optimal_match(ctx); - window_pos += 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 { - /* Best to output a literal here. */ - - /* Tally symbol frequencies. */ - lzx_match.data = lzx_record_literal(ctx->window[window_pos], - &freqs); - - window_pos += 1; - } - - /* If it's the last pass, save the match/literal in - * intermediate form. */ - if (pass == ctx->params.alg_params.slow.num_optim_passes - 1) { - ctx->chosen_matches[spec->chosen_matches_start_pos + - spec->num_chosen_matches] = lzx_match; - - spec->num_chosen_matches++; + lzx_tally_literal(*window_ptr, &freqs); + window_ptr += 1; } } - LZX_ASSERT(window_pos == end); - - /* Build Huffman codes using the new frequencies. */ - lzx_make_huffman_codes(&freqs, &spec->codes); - - /* The first time we get here is when the full input has been - * processed, so the match-finding is done. */ - ctx->matches_already_found = true; - } - - LZX_DEBUG("Block %u: saved %u matches/literals @ %u", - block_number, spec->num_chosen_matches, - spec->chosen_matches_start_pos); - - unsigned aligned_cost; - unsigned verbatim_cost; - - lzx_compute_compressed_block_costs(spec->block_size, - &spec->codes, - prev_codes, - &freqs, - &aligned_cost, - &verbatim_cost); - - /* Choose whether to make the block aligned offset or verbatim. */ - if (aligned_cost < verbatim_cost) { - spec->block_type = LZX_BLOCKTYPE_ALIGNED; - cost = aligned_cost; - LZX_DEBUG("Using aligned block (cost %u vs %u for verbatim)", - aligned_cost, verbatim_cost); - } else { - spec->block_type = LZX_BLOCKTYPE_VERBATIM; - cost = verbatim_cost; - LZX_DEBUG("Using verbatim block (cost %u vs %u for aligned)", - verbatim_cost, aligned_cost); - } - - LZX_DEBUG("Block %u is %u => %u bytes unsplit.", - block_number, spec->block_size, cost / 8); - - return cost; -} - -/* - * lzx_prepare_block_recursive() - - * - * Given a (possibly nonproper) sub-sequence of the preprocessed input, compute - * the LZX block(s) that it should be output as. - * - * This function initially considers the case where the given sub-sequence of - * the preprocessed input be output as a single block. This block is calculated - * and its cost (number of bits required to output it) is computed. - * - * Then, if @max_split_level is greater than zero, a split into two evenly sized - * subblocks is considered. The block is recursively split in this way, - * potentially up to the depth specified by @max_split_level. The cost of the - * split block is compared to the cost of the single block, and the lower cost - * solution is used. - * - * For each compressed output block computed, the sequence of matches/literals - * and the corresponding Huffman codes for the block are produced and saved. - * - * The return value is the approximate number of bits the block (or all - * subblocks, in the case that the split block had lower cost), will take up - * when written to the compressed output. - */ -static unsigned -lzx_prepare_block_recursive(struct lzx_compressor * ctx, - unsigned block_number, - unsigned max_split_level, - struct lzx_codes **prev_codes_p) -{ - struct lzx_block_spec *spec = &ctx->block_specs[block_number - 1]; - unsigned cost; - unsigned orig_cached_matches_pos; - struct lzx_lru_queue orig_queue, nonsplit_queue; - struct lzx_codes *prev_codes = *prev_codes_p; - - LZX_DEBUG("Preparing block %u...", block_number); - - /* Save positions of chosen and cached matches, and the match offset LRU - * queue, so that they can be restored if splitting is attempted. */ - orig_cached_matches_pos = ctx->cached_matches_pos; - orig_queue = ctx->queue; - - /* Consider outputting the input subsequence as a single block. */ - spec->is_split = 0; - cost = lzx_prepare_compressed_block(ctx, block_number, prev_codes); - nonsplit_queue = ctx->queue; - - *prev_codes_p = &spec->codes; - - /* If the maximum split level is at least one, consider splitting the - * block in two. */ - if (max_split_level--) { - - LZX_DEBUG("Calculating split of block %u...", block_number); - - struct lzx_block_spec *spec1, *spec2; - unsigned split_cost; - - ctx->cached_matches_pos = orig_cached_matches_pos; + 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; + } - /* Prepare and get the cost of the first sub-block. */ - spec1 = &ctx->block_specs[block_number * 2 - 1]; - spec1->codes.lens = spec->codes.lens; - spec1->window_pos = spec->window_pos; - spec1->block_size = spec->block_size / 2; - spec1->chosen_matches_start_pos = spec->chosen_matches_start_pos + - LZX_MAX_WINDOW_SIZE; - split_cost = lzx_prepare_block_recursive(ctx, - block_number * 2, - max_split_level, - &prev_codes); - - /* Prepare and get the cost of the second sub-block. */ - spec2 = spec1 + 1; - spec2->codes.lens = spec->codes.lens; - spec2->window_pos = spec->window_pos + spec1->block_size; - spec2->block_size = spec->block_size - spec1->block_size; - spec2->chosen_matches_start_pos = spec1->chosen_matches_start_pos + - spec1->block_size; - split_cost += lzx_prepare_block_recursive(ctx, - block_number * 2 + 1, - max_split_level, - &prev_codes); - - /* Compare the cost of the whole block with that of the split - * block. Choose the lower cost solution. */ - if (split_cost < cost) { - LZX_DEBUG("Splitting block %u is worth it " - "(%u => %u bytes).", - block_number, cost / 8, split_cost / 8); - spec->is_split = 1; - cost = split_cost; - *prev_codes_p = prev_codes; + 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 { - LZX_DEBUG("Splitting block %u is NOT worth it " - "(%u => %u bytes).", - block_number, cost / 8, split_cost / 8); - ctx->queue = nonsplit_queue; + lzx_item.data = lzx_tally_literal(*window_ptr, &freqs); + window_ptr += 1; } + *next_chosen_match++ = lzx_item; } - - return cost; + 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); } -/* Empirical averages */ -static const u8 lzx_default_mainsym_costs[LZX_MAINTREE_NUM_SYMBOLS] = { - 7, 9, 9, 10, 9, 10, 10, 10, 9, 10, 9, 10, 10, 9, 10, 10, 9, 10, 10, 11, - 10, 10, 10, 11, 10, 11, 11, 11, 10, 11, 11, 11, 8, 11, 9, 10, 9, 10, 11, - 11, 9, 9, 11, 10, 10, 9, 9, 9, 8, 8, 8, 8, 8, 9, 9, 9, 8, 8, 9, 9, 9, 9, - 10, 10, 10, 8, 9, 8, 8, 8, 8, 9, 9, 9, 10, 10, 8, 8, 9, 9, 8, 10, 9, 8, - 8, 9, 8, 9, 9, 10, 10, 10, 9, 10, 11, 9, 10, 8, 9, 8, 8, 8, 8, 9, 8, 8, - 9, 9, 8, 8, 8, 8, 8, 10, 8, 8, 7, 8, 9, 9, 9, 9, 10, 11, 10, 10, 11, 11, - 10, 11, 11, 10, 10, 11, 11, 11, 10, 10, 11, 10, 11, 10, 11, 11, 10, 11, - 11, 12, 11, 11, 11, 12, 11, 11, 11, 11, 11, 11, 11, 12, 10, 11, 11, 11, - 11, 11, 11, 12, 11, 11, 11, 11, 11, 12, 11, 11, 10, 11, 11, 11, 11, 11, - 11, 11, 10, 11, 11, 11, 11, 11, 11, 11, 10, 11, 11, 11, 11, 11, 11, 11, - 10, 11, 11, 11, 11, 11, 11, 11, 10, 11, 11, 11, 11, 12, 11, 11, 10, 11, - 11, 11, 11, 12, 11, 11, 10, 11, 11, 11, 10, 12, 11, 11, 10, 10, 11, 10, - 10, 11, 11, 11, 10, 11, 11, 11, 10, 11, 11, 11, 10, 11, 11, 11, 10, 11, - 10, 9, 8, 7, 10, 10, 11, 10, 11, 7, 9, 9, 11, 11, 11, 12, 11, 9, 10, 10, - 12, 12, 13, 13, 12, 11, 10, 12, 12, 14, 14, 14, 13, 12, 9, 12, 13, 14, - 14, 14, 14, 14, 9, 10, 13, 14, 14, 14, 14, 14, 9, 9, 11, 11, 13, 13, 13, - 14, 9, 9, 11, 12, 12, 13, 13, 13, 8, 8, 11, 11, 12, 12, 12, 11, 9, 9, - 10, 11, 12, 12, 12, 11, 8, 9, 10, 10, 11, 12, 11, 10, 9, 9, 10, 11, 11, - 12, 11, 10, 8, 9, 10, 10, 11, 11, 11, 9, 9, 9, 10, 11, 11, 11, 11, 9, 8, - 8, 10, 10, 11, 11, 11, 9, 9, 9, 10, 10, 11, 11, 11, 9, 9, 8, 9, 10, 11, - 11, 11, 9, 10, 9, 10, 11, 11, 11, 11, 9, 14, 9, 9, 10, 10, 11, 10, 9, - 14, 9, 10, 11, 11, 11, 11, 9, 14, 9, 10, 10, 11, 11, 11, 9, 14, 10, 10, - 11, 11, 12, 11, 10, 14, 10, 10, 10, 11, 11, 11, 10, 14, 11, 11, 11, 11, - 12, 12, 10, 14, 10, 11, 11, 11, 12, 11, 10, 14, 11, 11, 11, 12, 12, 12, - 11, 15, 11, 11, 11, 12, 12, 12, 11, 14, 12, 12, 12, 12, 13, 12, 11, 15, - 12, 12, 12, 13, 13, 13, 12, 15, 14, 13, 14, 14, 14, 14, 13, -}; - -/* Empirical averages */ -static const u8 lzx_default_lensym_costs[LZX_LENTREE_NUM_SYMBOLS] = { - 5, 5, 5, 5, 5, 6, 5, 5, 6, 7, 7, 7, 8, 8, 7, 8, 9, 9, 9, 9, 10, 9, 9, - 10, 9, 10, 10, 10, 10, 11, 11, 11, 11, 11, 11, 12, 12, 12, 11, 12, 12, - 12, 12, 12, 12, 13, 12, 12, 12, 13, 12, 13, 13, 12, 12, 13, 12, 13, 13, - 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 14, 13, 14, 13, 14, 13, - 14, 13, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, - 14, 13, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, - 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, - 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, - 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, - 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, - 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, - 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, - 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, - 14, 14, 14, 14, 14, 14, 14, 14, 14, 10, -}; - -/* - * Set default symbol costs. - */ +/* Prepare the input window into one or more LZX blocks ready to be output. */ static void -lzx_set_default_costs(struct lzx_lens * lens) -{ - unsigned i; - -#if LZX_PARAM_USE_EMPIRICAL_DEFAULT_COSTS - memcpy(&lens->main, lzx_default_mainsym_costs, LZX_MAINTREE_NUM_SYMBOLS); - memcpy(&lens->len, lzx_default_lensym_costs, LZX_LENTREE_NUM_SYMBOLS); - -#else - /* Literal symbols */ - for (i = 0; i < LZX_NUM_CHARS; i++) - lens->main[i] = 8; - - /* Match header symbols */ - for (; i < LZX_MAINTREE_NUM_SYMBOLS; i++) - lens->main[i] = 10; - - /* Length symbols */ - for (i = 0; i < LZX_LENTREE_NUM_SYMBOLS; i++) - lens->len[i] = 8; -#endif - - /* Aligned offset symbols */ - for (i = 0; i < LZX_ALIGNEDTREE_NUM_SYMBOLS; i++) - lens->aligned[i] = 3; -} - -/* - * lzx_prepare_blocks() - - * - * Calculate the blocks to split the preprocessed data into. - * - * Input --- the preprocessed data: - * - * ctx->window[] - * ctx->window_size - * - * Working space: - * Match finding: - * ctx->hash_tab - * ctx->child_tab - * ctx->cached_matches - * ctx->cached_matches_pos - * ctx->matches_already_found - * - * Block cost modeling: - * ctx->costs - * ctx->block_specs (also an output) - * - * Match choosing: - * ctx->optimum - * ctx->optimum_cur_idx - * ctx->optimum_end_idx - * ctx->chosen_matches (also an output) - * - * Output --- the block specifications and the corresponding match/literal data: - * - * ctx->block_specs[] - * ctx->chosen_matches[] - * - * The return value is the approximate number of bits the compressed data will - * take up. - */ -static unsigned lzx_prepare_blocks(struct lzx_compressor * ctx) { - /* This function merely does some initializations, then passes control - * to lzx_prepare_block_recursive(). */ - - /* 1. Initialize match-finding variables. */ - - /* Zero all entries in the hash table, indicating that no length-3 - * character sequences have been discovered in the input yet. */ - memset(ctx->hash_tab, 0, LZX_LZ_HASH_SIZE * 2 * sizeof(ctx->hash_tab[0])); - if (ctx->params.alg_params.slow.use_len2_matches) - memset(ctx->digram_tab, 0, 256 * 256 * sizeof(ctx->digram_tab[0])); - /* Note: ctx->child_tab need not be initialized. */ + /* 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); + } - /* No matches have been found and cached yet. */ - ctx->cached_matches_pos = 0; - ctx->matches_already_found = false; + /* Load the window into the match-finder. */ + lz_bt_load_window(&ctx->mf, ctx->window, ctx->window_size); - /* 2. Initialize match-choosing variables. */ + /* Determine sequence of matches/literals to output for each block. */ + lzx_lru_queue_init(&ctx->queue); ctx->optimum_cur_idx = 0; ctx->optimum_end_idx = 0; - /* Note: ctx->optimum need not be initialized. */ - ctx->block_specs[0].chosen_matches_start_pos = 0; - - /* 3. Set block 1 (index 0) to represent the entire input data. */ - ctx->block_specs[0].block_size = ctx->window_size; - ctx->block_specs[0].window_pos = 0; - - /* 4. Set up a default Huffman symbol cost model for block 1 (index 0). - * The model will be refined later. */ - lzx_set_default_costs(&ctx->block_specs[0].codes.lens); - - /* 5. Initialize the match offset LRU queue. */ - ctx->queue = (struct lzx_lru_queue){1, 1, 1}; - - /* 6. Pass control to recursive procedure. */ - struct lzx_codes * prev_codes = &ctx->zero_codes; - return lzx_prepare_block_recursive(ctx, 1, - ctx->params.alg_params.slow.num_split_passes, - &prev_codes); + 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); + } } /* @@ -2413,125 +1979,85 @@ lzx_prepare_blocks(struct lzx_compressor * ctx) * ctx->window[] * ctx->window_size * - * Working space: - * ctx->queue - * - * Output --- the block specifications and the corresponding match/literal data: + * Output --- the block specification and the corresponding match/literal data: * * ctx->block_specs[] - * ctx->chosen_matches[] + * ctx->num_blocks + * ctx->chosen_items[] */ static void lzx_prepare_block_fast(struct lzx_compressor * ctx) { - unsigned num_matches; - struct lzx_freqs freqs; + struct lzx_record_ctx record_ctx; struct lzx_block_spec *spec; - /* Parameters to hash chain LZ match finder */ + /* Parameters to hash chain LZ match finder + * (lazy with 1 match lookahead) */ static const struct lz_params lzx_lz_params = { - /* LZX_MIN_MATCH == 2, but 2-character matches are rarely - * useful; the minimum match for compression is set to 3 - * instead. */ + /* Although LZX_MIN_MATCH_LEN == 2, length 2 matches typically + * aren't worth choosing when using greedy or lazy parsing. */ .min_match = 3, - .max_match = LZX_MAX_MATCH, - .good_match = LZX_MAX_MATCH, - .nice_match = LZX_MAX_MATCH, - .max_chain_len = LZX_MAX_MATCH, - .max_lazy_match = LZX_MAX_MATCH, + .max_match = LZX_MAX_MATCH_LEN, + .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, + .max_lazy_match = LZX_MAX_MATCH_LEN, .too_far = 4096, }; /* Initialize symbol frequencies and match offset LRU queue. */ - memset(&freqs, 0, sizeof(struct lzx_freqs)); - ctx->queue = (struct lzx_lru_queue){ 1, 1, 1 }; + memset(&record_ctx.freqs, 0, sizeof(struct lzx_freqs)); + lzx_lru_queue_init(&record_ctx.queue); + record_ctx.matches = ctx->chosen_items; /* Determine series of matches/literals to output. */ - num_matches = lz_analyze_block(ctx->window, - ctx->window_size, - (u32*)ctx->chosen_matches, - lzx_record_match, - lzx_record_literal, - &freqs, - &ctx->queue, - &freqs, - &lzx_lz_params); - + lz_analyze_block(ctx->window, + ctx->window_size, + lzx_record_match, + lzx_record_literal, + &record_ctx, + &lzx_lz_params, + ctx->prev_tab); /* Set up block specification. */ spec = &ctx->block_specs[0]; - spec->is_split = 0; spec->block_type = LZX_BLOCKTYPE_ALIGNED; spec->window_pos = 0; spec->block_size = ctx->window_size; - spec->num_chosen_matches = num_matches; - spec->chosen_matches_start_pos = 0; - lzx_make_huffman_codes(&freqs, &spec->codes); -} - -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); - } + 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; } -/* 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; - unsigned 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 > 32768) { - LZX_DEBUG("Only up to 32768 bytes of uncompressed data are supported."); + 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; } - wimlib_assert(lzx_ctx != NULL); - - 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 @@ -2542,7 +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..."); @@ -2555,306 +2081,298 @@ 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..."); - if (flush_output_bitstream(&ostream)) { - /* If the bitstream cannot be flushed, then the output space was - * exhausted. */ - 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; } - /* Compute the length of the compressed data. */ - compressed_len = ostream.bit_output - (u8*)compressed_data; + LZX_DEBUG("Done: compressed %zu => %zu bytes.", + uncompressed_size, compressed_size); - LZX_DEBUG("Done: compressed %u => %u bytes.", - uncompressed_len, compressed_len); - -#if defined(ENABLE_LZX_DEBUG) || defined(ENABLE_VERIFY_COMPRESSION) - /* Verify that we really get the same thing back when decompressing. */ + /* Verify that we really get the same thing back when decompressing. + * Although this could be disabled by default in all cases, it only + * takes around 2-3% of the running time of the slow algorithm to do the + * verification. */ + if (ctx->params.algorithm == WIMLIB_LZX_ALGORITHM_SLOW + #if defined(ENABLE_LZX_DEBUG) || defined(ENABLE_VERIFY_COMPRESSION) + || 1 + #endif + ) { - u8 buf[uncompressed_len]; - int ret; - unsigned i; - - ret = wimlib_lzx_decompress(compressed_data, compressed_len, - buf, uncompressed_len); - if (ret) { - ERROR("Failed to decompress data we " - "compressed using LZX algorithm"); - wimlib_assert(0); - return 0; - } + struct wimlib_decompressor *decompressor; - bool bad = false; - const u8 * udata = uncompressed_data; - for (i = 0; i < uncompressed_len; i++) { - if (buf[i] != udata[i]) { - bad = true; + if (0 == wimlib_create_decompressor(WIMLIB_COMPRESSION_TYPE_LZX, + ctx->max_window_size, + NULL, + &decompressor)) + { + 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 " - "(difference at idx %u: c %#02x, u %#02x)", - i, buf[i], udata[i]); + "didn't decompress to original"); + wimlib_assert(0); + return 0; } - } - if (bad) { - wimlib_assert(0); - return 0; + } else { + WARNING("Failed to create decompressor for " + "data verification!"); } } -#endif - 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; } -/* API function documented in wimlib.h */ -WIMLIBAPI int -wimlib_lzx_alloc_context(const struct wimlib_lzx_params *params, - struct wimlib_lzx_context **ctx_pp) +static void +lzx_free_compressor(void *_ctx) { + struct lzx_compressor *ctx = _ctx; - LZX_DEBUG("Allocating LZX context..."); - - struct lzx_compressor *ctx; + 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); + } +} - 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_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_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 = 3, - .num_split_passes = 3, - .main_nostat_cost = 15, - .len_nostat_cost = 15, - .aligned_nostat_cost = 7, - }, + }, +}; +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) +{ + 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 = &slow_default; - } - - if (params->algorithm != WIMLIB_LZX_ALGORITHM_SLOW && - params->algorithm != WIMLIB_LZX_ALGORITHM_FAST) - { - LZX_DEBUG("Invalid algorithm."); - return WIMLIB_ERR_INVALID_PARAM; - } - - if (params->use_defaults) { - if (params->algorithm == WIMLIB_LZX_ALGORITHM_SLOW) - params = &slow_default; - else - params = &fast_default; - } - - if (params->size_of_this != sizeof(struct wimlib_lzx_params)) { - LZX_DEBUG("Invalid parameter structure size!"); - return WIMLIB_ERR_INVALID_PARAM; - } - - if (params->algorithm == WIMLIB_LZX_ALGORITHM_SLOW) { - if (params->alg_params.slow.num_fast_bytes < 3 || - params->alg_params.slow.num_fast_bytes > 257) - { - LZX_DEBUG("Invalid number of fast bytes!"); - return WIMLIB_ERR_INVALID_PARAM; - } - - if (params->alg_params.slow.num_optim_passes < 1) - { - LZX_DEBUG("Invalid number of optimization passes!"); - return WIMLIB_ERR_INVALID_PARAM; - } - - if (params->alg_params.slow.main_nostat_cost < 1 || - params->alg_params.slow.main_nostat_cost > 16) - { - LZX_DEBUG("Invalid main_nostat_cost!"); - return WIMLIB_ERR_INVALID_PARAM; - } - - if (params->alg_params.slow.len_nostat_cost < 1 || - params->alg_params.slow.len_nostat_cost > 16) - { - LZX_DEBUG("Invalid len_nostat_cost!"); - return WIMLIB_ERR_INVALID_PARAM; - } - - if (params->alg_params.slow.aligned_nostat_cost < 1 || - params->alg_params.slow.aligned_nostat_cost > 8) - { - LZX_DEBUG("Invalid aligned_nostat_cost!"); - return WIMLIB_ERR_INVALID_PARAM; + params = &lzx_slow_default; + } else { + if (params->use_defaults) { + if (params->algorithm == WIMLIB_LZX_ALGORITHM_SLOW) + params = &lzx_slow_default; + else + params = &lzx_fast_default; } } + return params; +} - if (ctx_pp == NULL) { - LZX_DEBUG("Check parameters only."); - return 0; - } - - ctx = *(struct lzx_compressor**)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; - if (ctx && lzx_params_compatible(&ctx->params, params)) - return 0; + LZX_DEBUG("Allocating LZX context..."); - LZX_DEBUG("Allocating memory."); + if (!lzx_window_size_valid(window_size)) + return WIMLIB_ERR_INVALID_PARAM; - ctx = MALLOC(sizeof(struct lzx_compressor)); + ctx = CALLOC(1, sizeof(struct lzx_compressor)); if (ctx == NULL) - goto err; - - size_t block_specs_length; + 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 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 oom; + } - if (params->algorithm == WIMLIB_LZX_ALGORITHM_SLOW) - block_specs_length = ((1 << (params->alg_params.slow.num_split_passes + 1)) - 1); - else - block_specs_length = 1; + 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_free_ctx; + goto oom; if (params->algorithm == WIMLIB_LZX_ALGORITHM_SLOW) { - ctx->hash_tab = MALLOC((LZX_LZ_HASH_SIZE + 2 * LZX_MAX_WINDOW_SIZE) * - sizeof(ctx->hash_tab[0])); - if (ctx->hash_tab == NULL) - goto err_free_block_specs; - ctx->child_tab = ctx->hash_tab + LZX_LZ_HASH_SIZE; - } else { - ctx->hash_tab = NULL; - ctx->child_tab = NULL; - } - - if (params->algorithm == WIMLIB_LZX_ALGORITHM_SLOW && - params->alg_params.slow.use_len2_matches) - { - ctx->digram_tab = MALLOC(256 * 256 * sizeof(ctx->digram_tab[0])); - if (ctx->digram_tab == NULL) - goto err_free_hash_tab; - } else { - ctx->digram_tab = NULL; + 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->cached_matches = MALLOC(10 * LZX_MAX_WINDOW_SIZE * - sizeof(ctx->cached_matches[0])); - if (ctx->cached_matches == NULL) - goto err_free_digram_tab; - } else { - ctx->cached_matches = NULL; + 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 oom; } if (params->algorithm == WIMLIB_LZX_ALGORITHM_SLOW) { - ctx->optimum = MALLOC((LZX_PARAM_OPTIM_ARRAY_SIZE + LZX_MAX_MATCH) * - sizeof(ctx->optimum[0])); - if (ctx->optimum == NULL) - goto err_free_cached_matches; - } else { - ctx->optimum = NULL; + ctx->cached_matches = MALLOC(LZX_CACHE_SIZE); + if (ctx->cached_matches == NULL) + goto oom; + ctx->cache_limit = ctx->cached_matches + + LZX_CACHE_LEN - (LZX_MAX_MATCHES_PER_POS + 1); } - size_t chosen_matches_length; - if (params->algorithm == WIMLIB_LZX_ALGORITHM_SLOW) - chosen_matches_length = LZX_MAX_WINDOW_SIZE * - (params->alg_params.slow.num_split_passes + 1); - else - chosen_matches_length = LZX_MAX_WINDOW_SIZE; - - ctx->chosen_matches = MALLOC(chosen_matches_length * - sizeof(ctx->chosen_matches[0])); - if (ctx->chosen_matches == NULL) - goto err_free_optimum; + 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_free_optimum: - FREE(ctx->optimum); -err_free_cached_matches: - FREE(ctx->cached_matches); -err_free_digram_tab: - FREE(ctx->digram_tab); -err_free_hash_tab: - FREE(ctx->hash_tab); -err_free_block_specs: - FREE(ctx->block_specs); -err_free_ctx: - FREE(ctx); -err: - 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->optimum); - FREE(ctx->cached_matches); - FREE(ctx->digram_tab); - FREE(ctx->hash_tab); - FREE(ctx->block_specs); - 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; - unsigned compressed_len; - - ret = wimlib_lzx_alloc_context(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; + } + + if (params->algorithm != WIMLIB_LZX_ALGORITHM_SLOW && + params->algorithm != WIMLIB_LZX_ALGORITHM_FAST) + { + LZX_DEBUG("Invalid algorithm."); + return false; } - compressed_len = wimlib_lzx_compress2(uncompressed_data, - uncompressed_len, - compressed_data, - ctx); + 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; + } - wimlib_lzx_free_context(ctx); + 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; + } - return compressed_len; + 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; + } + } + 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, +};