X-Git-Url: https://wimlib.net/git/?p=wimlib;a=blobdiff_plain;f=src%2Flzx-compress.c;h=5246e1a1ed3b7add7f7c543c0a240177c3c1218e;hp=056dc83bb52944c63ad9b98cdad8f966b2340e76;hb=6d0470be4d1855a0ea254e788e9ced23fc36dfb7;hpb=dbfee435692344cccd48bb4c7deb3af23ac80176 diff --git a/src/lzx-compress.c b/src/lzx-compress.c index 056dc83b..5246e1a1 100644 --- a/src/lzx-compress.c +++ b/src/lzx-compress.c @@ -112,7 +112,7 @@ * Huffman codes that were computed for the block. * * Note: the algorithm does not yet attempt to split the input into multiple LZX - * blocks, instead using a series of blocks of LZX_DIV_BLOCK_SIZE bytes. + * blocks; it instead uses a series of blocks of LZX_DIV_BLOCK_SIZE bytes. * * Fast algorithm * -------------- @@ -131,7 +131,7 @@ * it possible to implement this code: * * - divsufsort (author: Yuta Mori), for the suffix array construction code, - * located in a separate directory (divsufsort/). + * located in a separate file (divsufsort.c). * * - "Linear-Time Longest-Common-Prefix Computation in Suffix Arrays and Its * Applications" (Kasai et al. 2001), for the LCP array computation. @@ -162,8 +162,6 @@ #include "wimlib/lz_sarray.h" #include "wimlib/lzx.h" #include "wimlib/util.h" -#include -#include #include #ifdef ENABLE_LZX_DEBUG @@ -171,7 +169,7 @@ #endif typedef u32 block_cost_t; -#define INFINITE_BLOCK_COST ((block_cost_t)~0U) +#define INFINITE_BLOCK_COST (~(block_cost_t)0) #define LZX_OPTIM_ARRAY_SIZE 4096 @@ -264,50 +262,11 @@ struct lzx_block_spec { struct lzx_codes codes; }; -/* - * An array of these structures is used during the match-choosing algorithm. - * They correspond to consecutive positions in the window and are used to keep - * track of the cost to reach each position, and the match/literal choices that - * need to be chosen to reach that position. - */ -struct lzx_optimal { - /* The approximate minimum cost, in bits, to reach this position in the - * window which has been found so far. */ - block_cost_t cost; - - /* The union here is just for clarity, since the fields are used in two - * slightly different ways. Initially, the @prev structure is filled in - * first, and links go from later in the window to earlier in the - * window. Later, @next structure is filled in and links go from - * earlier in the window to later in the window. */ - union { - struct { - /* Position of the start of the match or literal that - * was taken to get to this position in the approximate - * minimum-cost parse. */ - input_idx_t link; - - /* Offset (as in an LZ (length, offset) pair) of the - * match or literal that was taken to get to this - * position in the approximate minimum-cost parse. */ - input_idx_t match_offset; - } prev; - struct { - /* Position at which the match or literal starting at - * this position ends in the minimum-cost parse. */ - input_idx_t link; - - /* Offset (as in an LZ (length, offset) pair) of the - * match or literal starting at this position in the - * approximate minimum-cost parse. */ - input_idx_t match_offset; - } next; - }; - - /* The match offset LRU queue that will exist when the approximate - * minimum-cost path to reach this position is taken. */ - struct lzx_lru_queue queue; -}; +/* Include template for the match-choosing algorithm. */ +#define LZ_COMPRESSOR struct lzx_compressor +#define LZ_ADAPTIVE_STATE struct lzx_lru_queue +struct lzx_compressor; +#include "wimlib/lz_optimal.h" /* State of the LZX compressor. */ struct lzx_compressor { @@ -326,8 +285,8 @@ struct lzx_compressor { * chunks. * * We reserve a few extra bytes to potentially allow reading off the end - * of the array in the match-finding code for optimization purposes. - */ + * of the array in the match-finding code for optimization purposes + * (currently only needed for the hash chain match-finder). */ u8 *window; /* Number of bytes of data to be compressed, which is the number of @@ -387,23 +346,8 @@ struct lzx_compressor { unsigned cached_matches_pos; bool matches_cached; - /* Slow algorithm only: Temporary space used for match-choosing - * algorithm. - * - * The size of this array must be at least LZX_MAX_MATCH_LEN but - * otherwise is arbitrary. More space simply allows the match-choosing - * algorithm to potentially find better matches (depending on the input, - * as always). */ - struct lzx_optimal *optimum; - - /* Slow algorithm only: Variables used by the match-choosing algorithm. - * - * 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; + /* Match chooser. */ + struct lz_match_chooser mc; }; /* Returns the LZX position slot that corresponds to a given match offset, @@ -471,13 +415,18 @@ lzx_make_huffman_codes(const struct lzx_freqs *freqs, } /* - * Output an LZX match. + * Output a precomputed 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. + * @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, @@ -507,9 +456,6 @@ lzx_write_match(struct output_bitstream *out, int block_type, * MIN_MATCH_LEN. */ if (match_len_minus_2 < LZX_NUM_PRIMARY_LENS) { len_header = match_len_minus_2; - /* No length footer-- mark it with a special - * value. */ - len_footer = (unsigned)(-1); } else { len_header = LZX_NUM_PRIMARY_LENS; len_footer = match_len_minus_2 - LZX_NUM_PRIMARY_LENS; @@ -529,10 +475,9 @@ lzx_write_match(struct output_bitstream *out, int block_type, /* If there is a length footer, output it using the * length Huffman code. */ - if (len_footer != (unsigned)(-1)) { + if (len_header == LZX_NUM_PRIMARY_LENS) bitstream_put_bits(out, codes->codewords.len[len_footer], codes->lens.len[len_footer]); - } num_extra_bits = lzx_get_num_extra_bits(position_slot); @@ -557,6 +502,16 @@ 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], @@ -696,23 +651,33 @@ lzx_build_precode(const u8 lens[restrict], } /* - * Writes a compressed Huffman code to the output, preceded by the precode for - * it. + * 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. * - * The Huffman code is represented in the output as a series of path lengths - * from which the canonical Huffman code can be reconstructed. The path lengths - * themselves are compressed using a separate Huffman code, the precode, which - * consists of LZX_PRECODE_NUM_SYMBOLS (= 20) symbols that cover all possible - * code lengths, plus extra codes for repeated lengths. The path lengths of the - * precode precede the path lengths of the larger code and are uncompressed, - * consisting of 20 entries of 4 bits each. + * 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 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. + * @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, @@ -771,20 +736,22 @@ lzx_write_compressed_code(struct output_bitstream *out, } /* - * Writes all compressed matches and literal bytes in an LZX block to the the - * output bitstream. + * Write all matches and literal bytes (which were precomputed) in an LZX + * compressed block to the output bitstream in the final compressed + * representation. * * @ostream * The output bitstream. * @block_type - * The type of the block (LZX_BLOCKTYPE_ALIGNED or LZX_BLOCKTYPE_VERBATIM). + * The chosen type of the LZX compressed block (LZX_BLOCKTYPE_ALIGNED or + * LZX_BLOCKTYPE_VERBATIM). * @match_tab - * The array of matches/literals that will be output (length @match_count). + * The array of matches/literals to output. * @match_count - * Number of matches/literals to be output. + * Number of matches/literals to output (length of @match_tab). * @codes - * Pointer to a structure that contains the codewords for the main, length, - * and aligned offset Huffman codes. + * The main, length, and aligned offset Huffman codes for the current + * LZX compressed block. */ static void lzx_write_matches_and_literals(struct output_bitstream *ostream, @@ -796,18 +763,13 @@ lzx_write_matches_and_literals(struct output_bitstream *ostream, 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]); - } + /* 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); } } @@ -996,8 +958,8 @@ lzx_tally_match(unsigned match_len, unsigned match_offset, /* 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); + position_footer = (match_offset + LZX_OFFSET_OFFSET) - + lzx_position_base[position_slot]; /* The match length shall be encoded as a length header (itself encoded * as part of the main symbol) and an optional length footer. */ @@ -1172,7 +1134,7 @@ lzx_set_costs(struct lzx_compressor * ctx, const struct lzx_lens * lens) /* Tell the match-finder to skip the specified number of bytes (@n) in the * input. */ static void -lzx_lz_skip_bytes(struct lzx_compressor *ctx, unsigned n) +lzx_lz_skip_bytes(struct lzx_compressor *ctx, input_idx_t n) { LZX_ASSERT(n <= ctx->match_window_end - ctx->match_window_pos); if (ctx->matches_cached) { @@ -1193,14 +1155,14 @@ lzx_lz_skip_bytes(struct lzx_compressor *ctx, unsigned n) /* Retrieve a list of matches available at the next position in the input. * - * The matches are written to ctx->matches in decreasing order of length, and - * the return value is the number of matches found. */ -static unsigned + * A pointer to the matches array is written into @matches_ret, and the return + * value is the number of matches found. */ +static u32 lzx_lz_get_matches_caching(struct lzx_compressor *ctx, const struct lzx_lru_queue *queue, struct raw_match **matches_ret) { - unsigned num_matches; + u32 num_matches; struct raw_match *matches; LZX_ASSERT(ctx->match_window_pos <= ctx->match_window_end); @@ -1224,7 +1186,7 @@ lzx_lz_get_matches_caching(struct lzx_compressor *ctx, * if it is not the whole window. */ if (ctx->match_window_end < ctx->window_size) { unsigned maxlen = ctx->match_window_end - ctx->match_window_pos; - for (unsigned i = 0; i < num_matches; i++) + for (u32 i = 0; i < num_matches; i++) if (matches[i].len > maxlen) matches[i].len = maxlen; } @@ -1236,7 +1198,7 @@ lzx_lz_get_matches_caching(struct lzx_compressor *ctx, #endif #ifdef ENABLE_LZX_DEBUG - for (unsigned i = 0; i < num_matches; i++) { + for (u32 i = 0; i < num_matches; i++) { 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); @@ -1252,275 +1214,61 @@ lzx_lz_get_matches_caching(struct lzx_compressor *ctx, return num_matches; } -/* - * 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 u32 +lzx_get_prev_literal_cost(struct lzx_compressor *ctx, + struct lzx_lru_queue *queue) { - unsigned prev_link, saved_prev_link; - unsigned prev_match_offset, saved_prev_match_offset; - - ctx->optimum_end_idx = cur_pos; - - saved_prev_link = ctx->optimum[cur_pos].prev.link; - saved_prev_match_offset = ctx->optimum[cur_pos].prev.match_offset; - - do { - prev_link = saved_prev_link; - prev_match_offset = saved_prev_match_offset; - - saved_prev_link = ctx->optimum[prev_link].prev.link; - saved_prev_match_offset = ctx->optimum[prev_link].prev.match_offset; - - ctx->optimum[prev_link].next.link = cur_pos; - ctx->optimum[prev_link].next.match_offset = prev_match_offset; - - cur_pos = prev_link; - } while (cur_pos != 0); - - ctx->optimum_cur_idx = ctx->optimum[0].next.link; + return lzx_literal_cost(ctx->window[ctx->match_window_pos - 1], + &ctx->costs); +} - return (struct raw_match) - { .len = ctx->optimum_cur_idx, - .offset = ctx->optimum[0].next.match_offset, - }; +static u32 +lzx_get_match_cost(struct lzx_compressor *ctx, + struct lzx_lru_queue *queue, + input_idx_t length, input_idx_t offset) +{ + return lzx_match_cost(length, offset, &ctx->costs, queue); } -/* - * lzx_lz_get_near_optimal_match() - - * - * Choose the optimal match or literal to use at the next position in the input. - * - * Unlike a greedy parser that always takes the longest match, or even a - * parser with one match/literal look-ahead like zlib, the algorithm used here - * may look ahead many matches/literals to determine the optimal match/literal to - * output next. The motivation is that the compression ratio is improved if the - * compressor can do things like use a shorter-than-possible match in order to - * allow a longer match later, and also take into account the Huffman code cost - * model rather than simply assuming that longer is better. - * - * Still, this is not truly an optimal parser because very long matches are - * taken immediately, and the raw match-finder takes some shortcuts. This is - * done to avoid considering many different alternatives that are unlikely to - * be significantly better. - * - * This algorithm is based on that used in 7-Zip's DEFLATE encoder. - * - * Each call to this function does one of two things: - * - * 1. Build a near-optimal sequence of matches/literals, up to some point, that - * will be returned by subsequent calls to this function, then return the - * first one. - * - * OR - * - * 2. Return the next match/literal previously computed by a call to this - * function; - * - * This function relies on the following state in the compressor context: - * - * ctx->window (read-only: preprocessed data being compressed) - * ctx->cost (read-only: cost model to use) - * ctx->optimum (internal state; leave uninitialized) - * ctx->optimum_cur_idx (must set to 0 before first call) - * ctx->optimum_end_idx (must set to 0 before first call) - * - * Plus any state used by the raw match-finder. - * - * The return value is a (length, offset) pair specifying the match or literal - * chosen. For literals, the length is less than LZX_MIN_MATCH_LEN and the - * offset is meaningless. - */ static struct raw_match -lzx_lz_get_near_optimal_match(struct lzx_compressor * ctx) +lzx_lz_get_near_optimal_match(struct lzx_compressor *ctx) { - unsigned num_possible_matches; - struct raw_match *possible_matches; - struct raw_match match; - unsigned longest_match_len; - - if (ctx->optimum_cur_idx != ctx->optimum_end_idx) { - /* Case 2: Return the next match/literal already found. */ - match.len = ctx->optimum[ctx->optimum_cur_idx].next.link - - ctx->optimum_cur_idx; - match.offset = ctx->optimum[ctx->optimum_cur_idx].next.match_offset; - - ctx->optimum_cur_idx = ctx->optimum[ctx->optimum_cur_idx].next.link; - return match; - } - - /* Case 1: Compute a new list of matches/literals to return. */ - - ctx->optimum_cur_idx = 0; - ctx->optimum_end_idx = 0; - - /* Get matches at this position. */ - num_possible_matches = lzx_lz_get_matches_caching(ctx, &ctx->queue, &possible_matches); - - /* If no matches found, return literal. */ - if (num_possible_matches == 0) - return (struct raw_match){ .len = 0 }; - - /* The matches that were found are sorted in decreasing order by length. - * Get the length of the longest one. */ - longest_match_len = possible_matches[0].len; - - /* Greedy heuristic: if the longest match that was found is greater - * than the number of fast bytes, return it immediately; don't both - * doing more work. */ - if (longest_match_len > ctx->params.alg_params.slow.num_fast_bytes) { - lzx_lz_skip_bytes(ctx, longest_match_len - 1); - return possible_matches[0]; - } - - /* Calculate the cost to reach the next position by outputting a - * literal. */ - ctx->optimum[0].queue = ctx->queue; - ctx->optimum[1].queue = ctx->optimum[0].queue; - ctx->optimum[1].cost = lzx_literal_cost(ctx->window[ctx->match_window_pos], - &ctx->costs); - ctx->optimum[1].prev.link = 0; - - /* Calculate the cost to reach any position up to and including that - * reached by the longest match, using the shortest (i.e. closest) match - * that reaches each position. */ - BUILD_BUG_ON(LZX_MIN_MATCH_LEN != 2); - for (unsigned len = LZX_MIN_MATCH_LEN, match_idx = num_possible_matches - 1; - len <= longest_match_len; len++) { - - LZX_ASSERT(match_idx < num_possible_matches); - - ctx->optimum[len].queue = ctx->optimum[0].queue; - ctx->optimum[len].prev.link = 0; - ctx->optimum[len].prev.match_offset = possible_matches[match_idx].offset; - ctx->optimum[len].cost = lzx_match_cost(len, - possible_matches[match_idx].offset, - &ctx->costs, - &ctx->optimum[len].queue); - if (len == possible_matches[match_idx].len) - match_idx--; - } - - unsigned cur_pos = 0; - - /* len_end: greatest index forward at which costs have been calculated - * so far */ - unsigned len_end = longest_match_len; - - for (;;) { - /* Advance to next position. */ - cur_pos++; - - if (cur_pos == len_end || cur_pos == LZX_OPTIM_ARRAY_SIZE) - return lzx_lz_reverse_near_optimal_match_list(ctx, cur_pos); - - /* retrieve the number of matches available at this position */ - num_possible_matches = lzx_lz_get_matches_caching(ctx, &ctx->optimum[cur_pos].queue, - &possible_matches); - - unsigned new_len = 0; - - if (num_possible_matches != 0) { - new_len = possible_matches[0].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) { - - /* Build the list of matches to return and get - * the first one. */ - match = lzx_lz_reverse_near_optimal_match_list(ctx, cur_pos); - - /* Append the long match to the end of the list. */ - ctx->optimum[cur_pos].next.match_offset = - possible_matches[0].offset; - ctx->optimum[cur_pos].next.link = cur_pos + new_len; - ctx->optimum_end_idx = cur_pos + new_len; - - /* Skip over the remaining bytes of the long match. */ - lzx_lz_skip_bytes(ctx, new_len - 1); - - /* Return first match in the list */ - return match; - } - } - - /* Consider proceeding with a literal byte. */ - block_cost_t cur_cost = ctx->optimum[cur_pos].cost; - block_cost_t cur_plus_literal_cost = cur_cost + - 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; - ctx->optimum[cur_pos + 1].queue = ctx->optimum[cur_pos].queue; - } - - if (num_possible_matches == 0) - continue; - - /* Consider proceeding with a match. */ - - while (len_end < cur_pos + new_len) - ctx->optimum[++len_end].cost = INFINITE_BLOCK_COST; - - for (unsigned len = LZX_MIN_MATCH_LEN, match_idx = num_possible_matches - 1; - len <= new_len; len++) { - LZX_ASSERT(match_idx < num_possible_matches); - struct lzx_lru_queue q = ctx->optimum[cur_pos].queue; - block_cost_t cost = cur_cost + lzx_match_cost(len, - possible_matches[match_idx].offset, - &ctx->costs, - &q); - - if (cost < ctx->optimum[cur_pos + len].cost) { - ctx->optimum[cur_pos + len].cost = cost; - ctx->optimum[cur_pos + len].prev.link = cur_pos; - ctx->optimum[cur_pos + len].prev.match_offset = - possible_matches[match_idx].offset; - ctx->optimum[cur_pos + len].queue = q; - } - - if (len == possible_matches[match_idx].len) - match_idx--; - } - } + return lz_get_near_optimal_match(&ctx->mc, + lzx_lz_get_matches_caching, + lzx_lz_skip_bytes, + lzx_get_prev_literal_cost, + lzx_get_match_cost, + ctx, + &ctx->queue); } -/* - * Set default symbol costs. - */ +/* Set default symbol costs for the LZX Huffman codes. */ static void lzx_set_default_costs(struct lzx_costs * costs, unsigned num_main_syms) { unsigned i; - /* Literal symbols */ + /* Main code (part 1): Literal symbols */ for (i = 0; i < LZX_NUM_CHARS; i++) costs->main[i] = 8; - /* Match header symbols */ + /* Main code (part 2): Match header symbols */ for (; i < num_main_syms; i++) costs->main[i] = 10; - /* Length symbols */ + /* Length code */ for (i = 0; i < LZX_LENCODE_NUM_SYMBOLS; i++) costs->len[i] = 8; - /* Aligned offset symbols */ + /* Aligned offset code */ for (i = 0; i < LZX_ALIGNEDCODE_NUM_SYMBOLS; i++) costs->aligned[i] = 3; } -/* Given the frequencies of symbols in a compressed block and the corresponding - * Huffman codes, return LZX_BLOCKTYPE_ALIGNED or LZX_BLOCKTYPE_VERBATIM if an - * aligned offset or verbatim block, respectively, will take fewer bits to - * output. */ +/* Given the frequencies of symbols in an LZX-compressed block and the + * corresponding Huffman codes, return LZX_BLOCKTYPE_ALIGNED or + * LZX_BLOCKTYPE_VERBATIM if an aligned offset or verbatim block, respectively, + * will take fewer bits to output. */ static int lzx_choose_verbatim_or_aligned(const struct lzx_freqs * freqs, const struct lzx_codes * codes) @@ -1530,8 +1278,8 @@ lzx_choose_verbatim_or_aligned(const struct lzx_freqs * freqs, /* Verbatim blocks have a constant 3 bits per position footer. Aligned * offset blocks have an aligned offset symbol per position footer, plus - * an extra 24 bits to output the lengths necessary to reconstruct the - * aligned offset code itself. */ + * an extra 24 bits per block to output the lengths necessary to + * reconstruct the aligned offset code itself. */ for (unsigned i = 0; i < LZX_ALIGNEDCODE_NUM_SYMBOLS; i++) { verbatim_cost += 3 * freqs->aligned[i]; aligned_cost += codes->lens.aligned[i] * freqs->aligned[i]; @@ -1544,8 +1292,8 @@ lzx_choose_verbatim_or_aligned(const struct lzx_freqs * freqs, } /* Find a near-optimal sequence of matches/literals with which to output the - * specified LZX block, then set its type to that which has the minimum cost to - * output. */ + * specified LZX block, then set the block's type to that which has the minimum + * cost to output (either verbatim or aligned). */ static void lzx_optimize_block(struct lzx_compressor *ctx, struct lzx_block_spec *spec, unsigned num_passes) @@ -1580,9 +1328,44 @@ lzx_optimize_block(struct lzx_compressor *ctx, struct lzx_block_spec *spec, raw_match = lzx_lz_get_near_optimal_match(ctx); if (raw_match.len >= LZX_MIN_MATCH_LEN) { - lzx_match.data = lzx_tally_match(raw_match.len, raw_match.offset, - &freqs, &ctx->queue); - i += raw_match.len; + if (unlikely(raw_match.len == LZX_MIN_MATCH_LEN && + raw_match.offset == ctx->max_window_size - + LZX_MIN_MATCH_LEN)) + { + /* Degenerate case where the parser + * generated the minimum match length + * with the maximum offset. There + * aren't actually enough position slots + * to represent this offset, as noted in + * the comments in + * lzx_get_num_main_syms(), so we cannot + * allow it. Use literals instead. + * + * Note that this case only occurs if + * the match-finder can generate matches + * to the very start of the window. The + * suffix array match-finder can, + * although typical hash chain and + * binary tree match-finders use 0 as a + * null value and therefore cannot + * generate such matches. */ + BUILD_BUG_ON(LZX_MIN_MATCH_LEN != 2); + lzx_match.data = lzx_tally_literal(ctx->window[i], + &freqs); + i += 1; + ctx->chosen_matches[spec->chosen_matches_start_pos + + spec->num_chosen_matches++] + = lzx_match; + lzx_match.data = lzx_tally_literal(ctx->window[i], + &freqs); + i += 1; + } else { + lzx_match.data = lzx_tally_match(raw_match.len, + raw_match.offset, + &freqs, + &ctx->queue); + i += raw_match.len; + } } else { lzx_match.data = lzx_tally_literal(ctx->window[i], &freqs); i += 1; @@ -1605,8 +1388,7 @@ static void lzx_optimize_blocks(struct lzx_compressor *ctx) { lzx_lru_queue_init(&ctx->queue); - ctx->optimum_cur_idx = 0; - ctx->optimum_end_idx = 0; + lz_match_chooser_begin(&ctx->mc); const unsigned num_passes = ctx->params.alg_params.slow.num_optim_passes; @@ -1627,6 +1409,9 @@ lzx_prepare_blocks(struct lzx_compressor * ctx) /* Set up a default cost model. */ lzx_set_default_costs(&ctx->costs, ctx->num_main_syms); + /* 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; @@ -1844,53 +1629,6 @@ lzx_compress(const void *uncompressed_data, size_t uncompressed_size, return compressed_size; } -static bool -lzx_params_valid(const struct wimlib_lzx_compressor_params *params) -{ - /* Validate parameters. */ - 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; - } - - if (params->algorithm == WIMLIB_LZX_ALGORITHM_SLOW) { - if (params->alg_params.slow.num_optim_passes < 1) - { - LZX_DEBUG("Invalid number of optimization passes!"); - return false; - } - - if (params->alg_params.slow.main_nostat_cost < 1 || - params->alg_params.slow.main_nostat_cost > 16) - { - LZX_DEBUG("Invalid main_nostat_cost!"); - return false; - } - - if (params->alg_params.slow.len_nostat_cost < 1 || - params->alg_params.slow.len_nostat_cost > 16) - { - LZX_DEBUG("Invalid len_nostat_cost!"); - return false; - } - - if (params->alg_params.slow.aligned_nostat_cost < 1 || - params->alg_params.slow.aligned_nostat_cost > 8) - { - LZX_DEBUG("Invalid aligned_nostat_cost!"); - return false; - } - } - return true; -} - static void lzx_free_compressor(void *_ctx) { @@ -1899,7 +1637,7 @@ lzx_free_compressor(void *_ctx) if (ctx) { FREE(ctx->chosen_matches); FREE(ctx->cached_matches); - FREE(ctx->optimum); + lz_match_chooser_destroy(&ctx->mc); lz_sarray_destroy(&ctx->lz_sarray); FREE(ctx->block_specs); FREE(ctx->prev_tab); @@ -1908,13 +1646,63 @@ lzx_free_compressor(void *_ctx) } } +static const struct wimlib_lzx_compressor_params lzx_fast_default = { + .hdr = { + .size = sizeof(struct wimlib_lzx_compressor_params), + }, + .algorithm = WIMLIB_LZX_ALGORITHM_FAST, + .use_defaults = 0, + .alg_params = { + .fast = { + }, + }, +}; +static const struct wimlib_lzx_compressor_params lzx_slow_default = { + .hdr = { + .size = sizeof(struct wimlib_lzx_compressor_params), + }, + .algorithm = WIMLIB_LZX_ALGORITHM_SLOW, + .use_defaults = 0, + .alg_params = { + .slow = { + .use_len2_matches = 1, + .nice_match_length = 32, + .num_optim_passes = 2, + .max_search_depth = 50, + .max_matches_per_pos = 3, + .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 = &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; +} + 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 = - (const struct wimlib_lzx_compressor_params*)_params; + const struct wimlib_lzx_compressor_params *params = lzx_get_params(_params); struct lzx_compressor *ctx; LZX_DEBUG("Allocating LZX context..."); @@ -1922,52 +1710,6 @@ lzx_create_compressor(size_t window_size, if (!lzx_window_size_valid(window_size)) return WIMLIB_ERR_INVALID_PARAM; - static const struct wimlib_lzx_compressor_params fast_default = { - .hdr = { - .size = sizeof(struct wimlib_lzx_compressor_params), - }, - .algorithm = WIMLIB_LZX_ALGORITHM_FAST, - .use_defaults = 0, - .alg_params = { - .fast = { - }, - }, - }; - static const struct wimlib_lzx_compressor_params 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, - .num_fast_bytes = 32, - .num_optim_passes = 2, - .max_search_depth = 50, - .max_matches_per_pos = 3, - .main_nostat_cost = 15, - .len_nostat_cost = 15, - .aligned_nostat_cost = 7, - }, - }, - }; - - if (params) { - if (!lzx_params_valid(params)) - return WIMLIB_ERR_INVALID_PARAM; - } else { - LZX_DEBUG("Using default algorithm and parameters."); - params = &slow_default; - } - - if (params->use_defaults) { - if (params->algorithm == WIMLIB_LZX_ALGORITHM_SLOW) - params = &slow_default; - else - params = &fast_default; - } - LZX_DEBUG("Allocating memory."); ctx = CALLOC(1, sizeof(struct lzx_compressor)); @@ -2006,9 +1748,10 @@ lzx_create_compressor(size_t window_size, } if (params->algorithm == WIMLIB_LZX_ALGORITHM_SLOW) { - ctx->optimum = MALLOC((LZX_OPTIM_ARRAY_SIZE + LZX_MAX_MATCH_LEN) * - sizeof(ctx->optimum[0])); - if (ctx->optimum == NULL) + if (!lz_match_chooser_init(&ctx->mc, + LZX_OPTIM_ARRAY_SIZE, + params->alg_params.slow.nice_match_length, + LZX_MAX_MATCH_LEN)) goto oom; } @@ -2042,7 +1785,95 @@ oom: return WIMLIB_ERR_NOMEM; } +static u64 +lzx_get_needed_memory(size_t max_block_size, + const struct wimlib_compressor_params_header *_params) +{ + const struct wimlib_lzx_compressor_params *params = lzx_get_params(_params); + + 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_matches[0]); + size += lz_sarray_get_needed_memory(max_block_size); + size += lz_match_chooser_get_needed_memory(LZX_OPTIM_ARRAY_SIZE, + params->alg_params.slow.nice_match_length, + LZX_MAX_MATCH_LEN); + u32 cache_per_pos; + + cache_per_pos = params->alg_params.slow.max_matches_per_pos; + if (cache_per_pos > LZX_MAX_CACHE_PER_POS) + cache_per_pos = LZX_MAX_CACHE_PER_POS; + + size += max_block_size * (cache_per_pos + 1) * + sizeof(((struct lzx_compressor*)0)->cached_matches[0]); + } else { + size += max_block_size * sizeof(((struct lzx_compressor*)0)->prev_tab[0]); + } + return size; +} + +static bool +lzx_params_valid(const struct wimlib_compressor_params_header *_params) +{ + 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; + } + + if (params->algorithm == WIMLIB_LZX_ALGORITHM_SLOW && + !params->use_defaults) + { + if (params->alg_params.slow.num_optim_passes < 1) + { + LZX_DEBUG("Invalid number of optimization passes!"); + return false; + } + + if (params->alg_params.slow.main_nostat_cost < 1 || + params->alg_params.slow.main_nostat_cost > 16) + { + LZX_DEBUG("Invalid main_nostat_cost!"); + return false; + } + + 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,