X-Git-Url: https://pd.if.org/git/?a=blobdiff_plain;f=lzma%2Flzma%2Flzma_encoder_optimum_fast.c;fp=lzma%2Flzma%2Flzma_encoder_optimum_fast.c;h=9b30347cbc4279d4e66828c5bff605483de12fc7;hb=32b8a6b26ed8843828e03e505d2256960bda0980;hp=0000000000000000000000000000000000000000;hpb=d48fc23a4bcf8ca3c406d6e8c8a6f8c6b0fa2f1e;p=zpackage

diff --git a/lzma/lzma/lzma_encoder_optimum_fast.c b/lzma/lzma/lzma_encoder_optimum_fast.c
new file mode 100644
index 0000000..9b30347
--- /dev/null
+++ b/lzma/lzma/lzma_encoder_optimum_fast.c
@@ -0,0 +1,169 @@
+///////////////////////////////////////////////////////////////////////////////
+//
+/// \file       lzma_encoder_optimum_fast.c
+//
+//  Author:     Igor Pavlov
+//
+//  This file has been put into the public domain.
+//  You can do whatever you want with this file.
+//
+///////////////////////////////////////////////////////////////////////////////
+
+#include "lzma_encoder_private.h"
+#include "memcmplen.h"
+
+
+#define change_pair(small_dist, big_dist) \
+	(((big_dist) >> 7) > (small_dist))
+
+
+extern void
+lzma_lzma_optimum_fast(lzma_coder *restrict coder, lzma_mf *restrict mf,
+		uint32_t *restrict back_res, uint32_t *restrict len_res)
+{
+	const uint32_t nice_len = mf->nice_len;
+
+	uint32_t len_main;
+	uint32_t matches_count;
+	if (mf->read_ahead == 0) {
+		len_main = mf_find(mf, &matches_count, coder->matches);
+	} else {
+		assert(mf->read_ahead == 1);
+		len_main = coder->longest_match_length;
+		matches_count = coder->matches_count;
+	}
+
+	const uint8_t *buf = mf_ptr(mf) - 1;
+	const uint32_t buf_avail = my_min(mf_avail(mf) + 1, MATCH_LEN_MAX);
+
+	if (buf_avail < 2) {
+		// There's not enough input left to encode a match.
+		*back_res = UINT32_MAX;
+		*len_res = 1;
+		return;
+	}
+
+	// Look for repeated matches; scan the previous four match distances
+	uint32_t rep_len = 0;
+	uint32_t rep_index = 0;
+
+	for (uint32_t i = 0; i < REPS; ++i) {
+		// Pointer to the beginning of the match candidate
+		const uint8_t *const buf_back = buf - coder->reps[i] - 1;
+
+		// If the first two bytes (2 == MATCH_LEN_MIN) do not match,
+		// this rep is not useful.
+		if (not_equal_16(buf, buf_back))
+			continue;
+
+		// The first two bytes matched.
+		// Calculate the length of the match.
+		const uint32_t len = lzma_memcmplen(
+				buf, buf_back, 2, buf_avail);
+
+		// If we have found a repeated match that is at least
+		// nice_len long, return it immediately.
+		if (len >= nice_len) {
+			*back_res = i;
+			*len_res = len;
+			mf_skip(mf, len - 1);
+			return;
+		}
+
+		if (len > rep_len) {
+			rep_index = i;
+			rep_len = len;
+		}
+	}
+
+	// We didn't find a long enough repeated match. Encode it as a normal
+	// match if the match length is at least nice_len.
+	if (len_main >= nice_len) {
+		*back_res = coder->matches[matches_count - 1].dist + REPS;
+		*len_res = len_main;
+		mf_skip(mf, len_main - 1);
+		return;
+	}
+
+	uint32_t back_main = 0;
+	if (len_main >= 2) {
+		back_main = coder->matches[matches_count - 1].dist;
+
+		while (matches_count > 1 && len_main ==
+				coder->matches[matches_count - 2].len + 1) {
+			if (!change_pair(coder->matches[
+						matches_count - 2].dist,
+					back_main))
+				break;
+
+			--matches_count;
+			len_main = coder->matches[matches_count - 1].len;
+			back_main = coder->matches[matches_count - 1].dist;
+		}
+
+		if (len_main == 2 && back_main >= 0x80)
+			len_main = 1;
+	}
+
+	if (rep_len >= 2) {
+		if (rep_len + 1 >= len_main
+				|| (rep_len + 2 >= len_main
+					&& back_main > (UINT32_C(1) << 9))
+				|| (rep_len + 3 >= len_main
+					&& back_main > (UINT32_C(1) << 15))) {
+			*back_res = rep_index;
+			*len_res = rep_len;
+			mf_skip(mf, rep_len - 1);
+			return;
+		}
+	}
+
+	if (len_main < 2 || buf_avail <= 2) {
+		*back_res = UINT32_MAX;
+		*len_res = 1;
+		return;
+	}
+
+	// Get the matches for the next byte. If we find a better match,
+	// the current byte is encoded as a literal.
+	coder->longest_match_length = mf_find(mf,
+			&coder->matches_count, coder->matches);
+
+	if (coder->longest_match_length >= 2) {
+		const uint32_t new_dist = coder->matches[
+				coder->matches_count - 1].dist;
+
+		if ((coder->longest_match_length >= len_main
+					&& new_dist < back_main)
+				|| (coder->longest_match_length == len_main + 1
+					&& !change_pair(back_main, new_dist))
+				|| (coder->longest_match_length > len_main + 1)
+				|| (coder->longest_match_length + 1 >= len_main
+					&& len_main >= 3
+					&& change_pair(new_dist, back_main))) {
+			*back_res = UINT32_MAX;
+			*len_res = 1;
+			return;
+		}
+	}
+
+	// In contrast to LZMA SDK, dictionary could not have been moved
+	// between mf_find() calls, thus it is safe to just increment
+	// the old buf pointer instead of recalculating it with mf_ptr().
+	++buf;
+
+	const uint32_t limit = my_max(2, len_main - 1);
+
+	for (uint32_t i = 0; i < REPS; ++i) {
+		if (memcmp(buf, buf - coder->reps[i] - 1, limit) == 0) {
+			*back_res = UINT32_MAX;
+			*len_res = 1;
+			return;
+		}
+	}
+
+	*back_res = back_main + REPS;
+	*len_res = len_main;
+	mf_skip(mf, len_main - 2);
+	return;
+}