X-Git-Url: https://wimlib.net/git/?p=wimlib;a=blobdiff_plain;f=src%2Flzx-compress.c;h=00d448603c07f9f9aece71f15fd3165655252323;hp=6b520ae5f05634eb35d3ab3e36a9c97e5edd20fc;hb=af2ea1ab40784caabb36f7aadd334e7dd34d09b9;hpb=9d9c2fe60b5842825002131f4ca926024e6f0038 diff --git a/src/lzx-compress.c b/src/lzx-compress.c index 6b520ae5..00d44860 100644 --- a/src/lzx-compress.c +++ b/src/lzx-compress.c @@ -158,20 +158,18 @@ #include "wimlib/compress_common.h" #include "wimlib/endianness.h" #include "wimlib/error.h" +#include "wimlib/lz_hash.h" +#include "wimlib/lz_sarray.h" #include "wimlib/lzx.h" #include "wimlib/util.h" -#include -#include #include #ifdef ENABLE_LZX_DEBUG # include "wimlib/decompress_common.h" #endif -#include "divsufsort/divsufsort.h" - 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 @@ -239,18 +237,6 @@ struct lzx_match { u32 data; }; -/* Raw LZ match/literal format: just a length and offset. - * - * The length is the number of bytes of the match, and the offset is the number - * of bytes back in the input the match is from the current position. - * - * If @len < LZX_MIN_MATCH_LEN, then it's really just a literal byte and @offset is - * meaningless. */ -struct raw_match { - u16 len; - input_idx_t offset; -}; - /* Specification for an LZX block. */ struct lzx_block_spec { @@ -276,77 +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; -}; - -/* Suffix array link */ -struct salink { - /* Rank of highest ranked suffix that has rank lower than the suffix - * corresponding to this structure and either has a lower position - * (initially) or has a position lower than the highest position at - * which matches have been searched for so far, or -1 if there is no - * such suffix. */ - input_idx_t prev; - - /* Rank of lowest ranked suffix that has rank greater than the suffix - * corresponding to this structure and either has a lower position - * (intially) or has a position lower than the highest position at which - * matches have been searched for so far, or -1 if there is no such - * suffix. */ - input_idx_t next; - - /* Length of longest common prefix between the suffix corresponding to - * this structure and the suffix with rank @prev, or 0 if @prev is -1. - */ - input_idx_t lcpprev; - - /* Length of longest common prefix between the suffix corresponding to - * this structure and the suffix with rank @next, or 0 if @next is -1. - */ - input_idx_t lcpnext; -}; +/* 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 { @@ -407,29 +327,8 @@ struct lzx_compressor { /* Fast algorithm only: Array of hash table links. */ input_idx_t *prev_tab; - /* Suffix array for window. - * This is a mapping from suffix rank to suffix position. */ - input_idx_t *SA; - - /* Inverse suffix array for window. - * This is a mapping from suffix position to suffix rank. - * If 0 <= r < window_size, then ISA[SA[r]] == r. */ - input_idx_t *ISA; - - /* Longest common prefix array corresponding to the suffix array SA. - * LCP[i] is the length of the longest common prefix between the - * suffixes with positions SA[i - 1] and SA[i]. LCP[0] is undefined. - */ - input_idx_t *LCP; - - /* Suffix array links. - * - * During a linear scan of the input string to find matches, this array - * used to keep track of which rank suffixes in the suffix array appear - * before the current position. Instead of searching in the original - * suffix array, scans for matches at a given position traverse a linked - * list containing only suffixes that appear before that position. */ - struct salink *salink; + /* Slow algorithm only: Suffix array match-finder. */ + struct lz_sarray lz_sarray; /* Position in window of next match to return. */ input_idx_t match_window_pos; @@ -447,23 +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, @@ -1178,16 +1062,17 @@ lzx_match_cost(unsigned length, unsigned offset, const struct lzx_costs *costs, /* Fast heuristic cost evaluation to use in the inner loop of the match-finder. * Unlike lzx_match_cost() which does a true cost evaluation, this simply * prioritize matches based on their offset. */ -static block_cost_t -lzx_match_cost_fast(unsigned offset, const struct lzx_lru_queue *queue) +static input_idx_t +lzx_match_cost_fast(input_idx_t length, input_idx_t offset, const void *_queue) { + const struct lzx_lru_queue *queue = _queue; + /* It seems well worth it to take the time to give priority to recently * used offsets. */ - for (unsigned i = 0; i < LZX_NUM_RECENT_OFFSETS; i++) + for (input_idx_t i = 0; i < LZX_NUM_RECENT_OFFSETS; i++) if (offset == queue->R[i]) return i; - BUILD_BUG_ON(LZX_MAX_WINDOW_SIZE >= (block_cost_t)~0U); return offset; } @@ -1228,223 +1113,10 @@ lzx_set_costs(struct lzx_compressor * ctx, const struct lzx_lens * lens) } } -/* Advance the suffix array match-finder to the next position. */ -static void -lzx_lz_update_salink(input_idx_t i, - const input_idx_t SA[restrict], - const input_idx_t ISA[restrict], - struct salink link[restrict]) -{ - /* r = Rank of the suffix at the current position. */ - const input_idx_t r = ISA[i]; - - /* next = rank of LOWEST ranked suffix that is ranked HIGHER than the - * current suffix AND has a LOWER position, or -1 if none exists. */ - const input_idx_t next = link[r].next; - - /* prev = rank of HIGHEST ranked suffix that is ranked LOWER than the - * current suffix AND has a LOWER position, or -1 if none exists. */ - const input_idx_t prev = link[r].prev; - - /* Link the suffix at the current position into the linked list that - * contains all suffixes in the suffix array that are appear at or - * before the current position, sorted by rank. - * - * Save the values of all fields we overwrite so that rollback is - * possible. */ - if (next != (input_idx_t)~0U) { - - link[next].prev = r; - link[next].lcpprev = link[r].lcpnext; - } - - if (prev != (input_idx_t)~0U) { - - link[prev].next = r; - link[prev].lcpnext = link[r].lcpprev; - } -} - -/* - * Use the suffix array match-finder to retrieve a list of LZ matches at the - * current position. - * - * [in] @i Current position in the window. - * [in] @SA Suffix array for the window. - * [in] @ISA Inverse suffix array for the window. - * [inout] @link Suffix array links used internally by the match-finder. - * [out] @matches The (length, offset) pairs of the resulting matches will - * be written here, sorted in decreasing order by - * length. All returned lengths will be unique. - * [in] @queue Recently used match offsets, used when evaluating the - * cost of matches. - * [in] @min_match_len Minimum match length to return. - * [in] @max_matches_to_consider Maximum number of matches to consider at - * the position. - * [in] @max_matches_to_return Maximum number of matches to return. - * - * The return value is the number of matches found and written to @matches. - */ -static unsigned -lzx_lz_get_matches(const input_idx_t i, - const input_idx_t SA[const restrict], - const input_idx_t ISA[const restrict], - struct salink link[const restrict], - struct raw_match matches[const restrict], - const struct lzx_lru_queue * const restrict queue, - const unsigned min_match_len, - const u32 max_matches_to_consider, - const u32 max_matches_to_return) -{ - /* r = Rank of the suffix at the current position. */ - const input_idx_t r = ISA[i]; - - /* Prepare for searching the current position. */ - lzx_lz_update_salink(i, SA, ISA, link); - - /* L = rank of next suffix to the left; - * R = rank of next suffix to the right; - * lenL = length of match between current position and the suffix with rank L; - * lenR = length of match between current position and the suffix with rank R. - * - * This is left and right relative to the rank of the current suffix. - * Since the suffixes in the suffix array are sorted, the longest - * matches are immediately to the left and right (using the linked list - * to ignore all suffixes that occur later in the window). The match - * length decreases the farther left and right we go. We shall keep the - * length on both sides in sync in order to choose the lowest-cost match - * of each length. - */ - input_idx_t L = link[r].prev; - input_idx_t R = link[r].next; - input_idx_t lenL = link[r].lcpprev; - input_idx_t lenR = link[r].lcpnext; - - /* nmatches = number of matches found so far. */ - unsigned nmatches = 0; - - /* best_cost = cost of lowest-cost match found so far. - * - * We keep track of this so that we can ignore shorter matches that do - * not have lower costs than a longer matches already found. - */ - block_cost_t best_cost = INFINITE_BLOCK_COST; - - /* count_remaining = maximum number of possible matches remaining to be - * considered. */ - u32 count_remaining = max_matches_to_consider; - - /* pending = match currently being considered for a specific length. */ - struct raw_match pending; - block_cost_t pending_cost; - - while (lenL >= min_match_len || lenR >= min_match_len) - { - pending.len = lenL; - pending_cost = INFINITE_BLOCK_COST; - block_cost_t cost; - - /* Extend left. */ - if (lenL >= min_match_len && lenL >= lenR) { - for (;;) { - - if (--count_remaining == 0) - goto out_save_pending; - - input_idx_t offset = i - SA[L]; - - /* Save match if it has smaller cost. */ - cost = lzx_match_cost_fast(offset, queue); - if (cost < pending_cost) { - pending.offset = offset; - pending_cost = cost; - } - - if (link[L].lcpprev < lenL) { - /* Match length decreased. */ - - lenL = link[L].lcpprev; - - /* Save the pending match unless the - * right side still may have matches of - * this length to be scanned, or if a - * previous (longer) match had lower - * cost. */ - if (pending.len > lenR) { - if (pending_cost < best_cost) { - best_cost = pending_cost; - matches[nmatches++] = pending; - if (nmatches == max_matches_to_return) - return nmatches; - } - pending.len = lenL; - pending_cost = INFINITE_BLOCK_COST; - } - if (lenL < min_match_len || lenL < lenR) - break; - } - L = link[L].prev; - } - } - - pending.len = lenR; - - /* Extend right. */ - if (lenR >= min_match_len && lenR > lenL) { - for (;;) { - - if (--count_remaining == 0) - goto out_save_pending; - - input_idx_t offset = i - SA[R]; - - /* Save match if it has smaller cost. */ - cost = lzx_match_cost_fast(offset, queue); - if (cost < pending_cost) { - pending.offset = offset; - pending_cost = cost; - } - - if (link[R].lcpnext < lenR) { - /* Match length decreased. */ - - lenR = link[R].lcpnext; - - /* Save the pending match unless a - * previous (longer) match had lower - * cost. */ - if (pending_cost < best_cost) { - matches[nmatches++] = pending; - best_cost = pending_cost; - if (nmatches == max_matches_to_return) - return nmatches; - } - - if (lenR < min_match_len || lenR <= lenL) - break; - - pending.len = lenR; - pending_cost = INFINITE_BLOCK_COST; - } - R = link[R].next; - } - } - } - goto out; - -out_save_pending: - if (pending_cost != INFINITE_BLOCK_COST) - matches[nmatches++] = pending; - -out: - return nmatches; -} - - /* 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) { @@ -1456,9 +1128,10 @@ lzx_lz_skip_bytes(struct lzx_compressor *ctx, unsigned n) } else { while (n--) { ctx->cached_matches[ctx->cached_matches_pos++].len = 0; - lzx_lz_update_salink(ctx->match_window_pos++, ctx->SA, - ctx->ISA, ctx->salink); + lz_sarray_skip_position(&ctx->lz_sarray); + ctx->match_window_pos++; } + LZX_ASSERT(lz_sarray_get_pos(&ctx->lz_sarray) == ctx->match_window_pos); } } @@ -1466,12 +1139,12 @@ lzx_lz_skip_bytes(struct lzx_compressor *ctx, unsigned n) * * The matches are written to ctx->matches in decreasing order of length, and * the return value is the number of matches found. */ -static unsigned +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); @@ -1481,24 +1154,11 @@ lzx_lz_get_matches_caching(struct lzx_compressor *ctx, if (ctx->matches_cached) { num_matches = matches[-1].len; } else { - unsigned min_match_len = LZX_MIN_MATCH_LEN; - if (!ctx->params.alg_params.slow.use_len2_matches) - min_match_len = max(min_match_len, 3); - const u32 max_search_depth = ctx->params.alg_params.slow.max_search_depth; - const u32 max_matches_per_pos = ctx->params.alg_params.slow.max_matches_per_pos; - - if (unlikely(max_search_depth == 0 || max_matches_per_pos == 0)) - num_matches = 0; - else - num_matches = lzx_lz_get_matches(ctx->match_window_pos, - ctx->SA, - ctx->ISA, - ctx->salink, - matches, - queue, - min_match_len, - max_search_depth, - max_matches_per_pos); + LZX_ASSERT(lz_sarray_get_pos(&ctx->lz_sarray) == ctx->match_window_pos); + num_matches = lz_sarray_get_matches(&ctx->lz_sarray, + matches, + lzx_match_cost_fast, + queue); matches[-1].len = num_matches; } ctx->cached_matches_pos += num_matches + 1; @@ -1508,7 +1168,7 @@ 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; } @@ -1520,7 +1180,7 @@ 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); @@ -1536,244 +1196,32 @@ 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); } /* @@ -1864,9 +1312,44 @@ 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; @@ -1889,8 +1372,7 @@ 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; @@ -1898,169 +1380,12 @@ lzx_optimize_blocks(struct lzx_compressor *ctx) lzx_optimize_block(ctx, &ctx->block_specs[i], num_passes); } -/* Initialize the suffix array match-finder for the specified input. */ -static void -lzx_lz_init_matchfinder(const u8 T[const restrict], - const input_idx_t n, - input_idx_t SA[const restrict], - input_idx_t ISA[const restrict], - input_idx_t LCP[const restrict], - struct salink link[const restrict], - const unsigned max_match_len) -{ - /* Compute SA (Suffix Array). */ - - { - /* ISA and link are used as temporary space. */ - BUILD_BUG_ON(LZX_MIN_WINDOW_SIZE * sizeof(ISA[0]) < 256 * sizeof(saidx_t)); - BUILD_BUG_ON(LZX_MIN_WINDOW_SIZE * 2 * sizeof(link[0]) < 256 * 256 * sizeof(saidx_t)); - - if (sizeof(input_idx_t) == sizeof(saidx_t)) { - divsufsort(T, SA, n, (saidx_t*)ISA, (saidx_t*)link); - } else { - saidx_t sa[n]; - divsufsort(T, sa, n, (saidx_t*)ISA, (saidx_t*)link); - for (input_idx_t i = 0; i < n; i++) - SA[i] = sa[i]; - } - } - -#ifdef ENABLE_LZX_DEBUG - - LZX_ASSERT(n > 0); - - /* Verify suffix array. */ - { - bool found[n]; - ZERO_ARRAY(found); - for (input_idx_t r = 0; r < n; r++) { - input_idx_t i = SA[r]; - LZX_ASSERT(i < n); - LZX_ASSERT(!found[i]); - found[i] = true; - } - } - - for (input_idx_t r = 0; r < n - 1; r++) { - - input_idx_t i1 = SA[r]; - input_idx_t i2 = SA[r + 1]; - - input_idx_t n1 = n - i1; - input_idx_t n2 = n - i2; - - LZX_ASSERT(memcmp(&T[i1], &T[i2], min(n1, n2)) <= 0); - } - LZX_DEBUG("Verified SA (len %u)", n); -#endif /* ENABLE_LZX_DEBUG */ - - /* Compute ISA (Inverse Suffix Array) */ - for (input_idx_t r = 0; r < n; r++) - ISA[SA[r]] = r; - - /* Compute LCP (longest common prefix) array. - * - * Algorithm adapted from Kasai et al. 2001: "Linear-Time - * Longest-Common-Prefix Computation in Suffix Arrays and Its - * Applications". */ - { - input_idx_t h = 0; - for (input_idx_t i = 0; i < n; i++) { - input_idx_t r = ISA[i]; - if (r > 0) { - input_idx_t j = SA[r - 1]; - - input_idx_t lim = min(n - i, n - j); - - while (h < lim && T[i + h] == T[j + h]) - h++; - LCP[r] = h; - if (h > 0) - h--; - } - } - } - -#ifdef ENABLE_LZX_DEBUG - /* Verify LCP array. */ - for (input_idx_t r = 0; r < n - 1; r++) { - LZX_ASSERT(ISA[SA[r]] == r); - LZX_ASSERT(ISA[SA[r + 1]] == r + 1); - - input_idx_t i1 = SA[r]; - input_idx_t i2 = SA[r + 1]; - input_idx_t lcp = LCP[r + 1]; - - input_idx_t n1 = n - i1; - input_idx_t n2 = n - i2; - - LZX_ASSERT(lcp <= min(n1, n2)); - - LZX_ASSERT(memcmp(&T[i1], &T[i2], lcp) == 0); - if (lcp < min(n1, n2)) - LZX_ASSERT(T[i1 + lcp] != T[i2 + lcp]); - } -#endif /* ENABLE_LZX_DEBUG */ - - /* Compute salink.next and salink.lcpnext. - * - * Algorithm adapted from Crochemore et al. 2009: - * "LPF computation revisited". - * - * Note: we cap lcpnext to the maximum match length so that the - * match-finder need not worry about it later. */ - link[n - 1].next = (input_idx_t)~0U; - link[n - 1].prev = (input_idx_t)~0U; - link[n - 1].lcpnext = 0; - link[n - 1].lcpprev = 0; - for (input_idx_t r = n - 2; r != (input_idx_t)~0U; r--) { - input_idx_t t = r + 1; - input_idx_t l = LCP[t]; - while (t != (input_idx_t)~0 && SA[t] > SA[r]) { - l = min(l, link[t].lcpnext); - t = link[t].next; - } - link[r].next = t; - link[r].lcpnext = min(l, max_match_len); - LZX_ASSERT(t == (input_idx_t)~0U || l <= n - SA[t]); - LZX_ASSERT(l <= n - SA[r]); - LZX_ASSERT(memcmp(&T[SA[r]], &T[SA[t]], l) == 0); - } - - /* Compute salink.prev and salink.lcpprev. - * - * Algorithm adapted from Crochemore et al. 2009: - * "LPF computation revisited". - * - * Note: we cap lcpprev to the maximum match length so that the - * match-finder need not worry about it later. */ - link[0].prev = (input_idx_t)~0; - link[0].next = (input_idx_t)~0; - link[0].lcpprev = 0; - link[0].lcpnext = 0; - for (input_idx_t r = 1; r < n; r++) { - input_idx_t t = r - 1; - input_idx_t l = LCP[r]; - while (t != (input_idx_t)~0 && SA[t] > SA[r]) { - l = min(l, link[t].lcpprev); - t = link[t].prev; - } - link[r].prev = t; - link[r].lcpprev = min(l, max_match_len); - LZX_ASSERT(t == (input_idx_t)~0 || l <= n - SA[t]); - LZX_ASSERT(l <= n - SA[r]); - LZX_ASSERT(memcmp(&T[SA[r]], &T[SA[t]], l) == 0); - } -} - /* Prepare the input window into one or more LZX blocks ready to be output. */ static void lzx_prepare_blocks(struct lzx_compressor * ctx) { /* Initialize the match-finder. */ - lzx_lz_init_matchfinder(ctx->window, ctx->window_size, - ctx->SA, ctx->ISA, ctx->LCP, ctx->salink, - LZX_MAX_MATCH_LEN); + lz_sarray_load_window(&ctx->lz_sarray, ctx->window, ctx->window_size); ctx->cached_matches_pos = 0; ctx->matches_cached = false; ctx->match_window_pos = 0; @@ -2285,53 +1610,6 @@ 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) { @@ -2340,9 +1618,8 @@ lzx_free_compressor(void *_ctx) if (ctx) { FREE(ctx->chosen_matches); FREE(ctx->cached_matches); - FREE(ctx->optimum); - FREE(ctx->salink); - FREE(ctx->SA); + lz_match_chooser_destroy(&ctx->mc); + lz_sarray_destroy(&ctx->lz_sarray); FREE(ctx->block_specs); FREE(ctx->prev_tab); FREE(ctx->window); @@ -2350,13 +1627,63 @@ 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..."); @@ -2364,52 +1691,6 @@ 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)); @@ -2434,21 +1715,24 @@ lzx_create_compressor(size_t window_size, goto oom; if (params->algorithm == WIMLIB_LZX_ALGORITHM_SLOW) { - ctx->SA = MALLOC(3U * window_size * sizeof(ctx->SA[0])); - if (ctx->SA == NULL) - goto oom; - ctx->ISA = ctx->SA + window_size; - ctx->LCP = ctx->ISA + window_size; + unsigned min_match_len = LZX_MIN_MATCH_LEN; + if (!params->alg_params.slow.use_len2_matches) + min_match_len = max(min_match_len, 3); - ctx->salink = MALLOC(window_size * sizeof(ctx->salink[0])); - if (ctx->salink == NULL) + if (!lz_sarray_init(&ctx->lz_sarray, + window_size, + min_match_len, + LZX_MAX_MATCH_LEN, + params->alg_params.slow.max_search_depth, + params->alg_params.slow.max_matches_per_pos)) goto oom; } 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; } @@ -2482,7 +1766,96 @@ 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,