X-Git-Url: https://wimlib.net/git/?a=blobdiff_plain;f=src%2Fxpress_compress.c;h=a04bc52d207adf35db24bcb3725cefdcef8d897a;hb=e2f7bf9e5406b8784a2dd1e4947398c2ba9ddbe5;hp=bce0901f523665b8b95aa1ae73864d1e8ccfc600;hpb=e8d16043cd1185a0631dc04b75b59b27f4510929;p=wimlib diff --git a/src/xpress_compress.c b/src/xpress_compress.c index bce0901f..a04bc52d 100644 --- a/src/xpress_compress.c +++ b/src/xpress_compress.c @@ -906,27 +906,23 @@ xpress_find_matches(struct xpress_compressor * restrict c, { const u8 * const in_begin = in; const u8 *in_next = in_begin; - const u8 * const in_end = in_begin + in_nbytes; struct lz_match *cache_ptr = c->match_cache; - u32 next_hash = 0; + u32 next_hashes[2] = {}; + u32 max_len = in_nbytes; + u32 nice_len = min(max_len, c->nice_match_length); bt_matchfinder_init(&c->bt_mf); - do { + for (;;) { struct lz_match *matches; - unsigned best_len; + u32 best_len; /* If we've found so many matches that the cache might overflow * if we keep finding more, then stop finding matches. This * case is very unlikely. */ - if (unlikely(cache_ptr >= c->cache_overflow_mark)) { - do { - cache_ptr->length = 0; - cache_ptr->offset = *in_next++; - cache_ptr++; - } while (in_next != in_end); - return cache_ptr; - } + if (unlikely(cache_ptr >= c->cache_overflow_mark || + max_len < BT_MATCHFINDER_REQUIRED_NBYTES)) + break; matches = cache_ptr; @@ -937,16 +933,17 @@ xpress_find_matches(struct xpress_compressor * restrict c, bt_matchfinder_get_matches(&c->bt_mf, in_begin, in_next - in_begin, - in_end - in_next, - min(in_end - in_next, c->nice_match_length), + max_len, + nice_len, c->max_search_depth, - &next_hash, + next_hashes, &best_len, cache_ptr); cache_ptr->length = cache_ptr - matches; - cache_ptr->offset = *in_next; - in_next++; + cache_ptr->offset = *in_next++; cache_ptr++; + max_len--; + nice_len = min(nice_len, max_len); /* * If there was a very long match found, then don't cache any @@ -958,24 +955,33 @@ xpress_find_matches(struct xpress_compressor * restrict c, * very much. If there's a long match, then the data must be * highly compressible, so it doesn't matter as much what we do. */ - if (best_len >= c->nice_match_length) { + if (best_len >= nice_len) { + if (unlikely(best_len + + BT_MATCHFINDER_REQUIRED_NBYTES >= max_len)) + break; --best_len; do { bt_matchfinder_skip_position(&c->bt_mf, in_begin, in_next - in_begin, - in_end - in_next, - min(in_end - in_next, - c->nice_match_length), + max_len, + nice_len, c->max_search_depth, - &next_hash); - + next_hashes); cache_ptr->length = 0; cache_ptr->offset = *in_next++; cache_ptr++; + max_len--; + nice_len = min(nice_len, max_len); } while (--best_len); } - } while (in_next != in_end); + } + + while (max_len--) { + cache_ptr->length = 0; + cache_ptr->offset = *in_next++; + cache_ptr++; + } return cache_ptr; } @@ -1086,12 +1092,12 @@ xpress_create_compressor(size_t max_bufsize, unsigned compression_level, if (compression_level < 30) { c->impl = xpress_compress_greedy; - c->max_search_depth = (compression_level * 24) / 16; - c->nice_match_length = (compression_level * 48) / 16; + c->max_search_depth = (compression_level * 30) / 16; + c->nice_match_length = (compression_level * 60) / 16; } else { c->impl = xpress_compress_lazy; - c->max_search_depth = (compression_level * 24) / 32; - c->nice_match_length = (compression_level * 48) / 32; + c->max_search_depth = (compression_level * 30) / 32; + c->nice_match_length = (compression_level * 60) / 32; /* xpress_compress_lazy() needs max_search_depth >= 2 * because it halves the max_search_depth when @@ -1118,8 +1124,8 @@ xpress_create_compressor(size_t max_bufsize, unsigned compression_level, &c->match_cache[max_bufsize * CACHE_RESERVE_PER_POS]; c->impl = xpress_compress_near_optimal; - c->max_search_depth = (compression_level * 32) / 100; - c->nice_match_length = (compression_level * 50) / 100; + c->max_search_depth = (compression_level * 28) / 100; + c->nice_match_length = (compression_level * 56) / 100; c->num_optim_passes = compression_level / 40; } #endif /* SUPPORT_NEAR_OPTIMAL_PARSING */