X-Git-Url: https://wimlib.net/git/?p=wimlib;a=blobdiff_plain;f=src%2Flzx_compress.c;h=36e27dbed6c2ecdda17b4206e3d8a75c0b299ceb;hp=98142962db6d21c866c01367de675a527e28057d;hb=c9be4d389724d00cec9f5e444efb965a449d2ba8;hpb=e9a04c1cb384cf3cf23d70107e85f79c4ac0a555

diff --git a/src/lzx_compress.c b/src/lzx_compress.c
index 98142962..36e27dbe 100644
--- a/src/lzx_compress.c
+++ b/src/lzx_compress.c
@@ -114,10 +114,9 @@
 #define LZX_BIT_COST		16
 
 /*
- * Consideration of aligned offset costs is disabled for now, due to
- * insufficient benefit gained from the time spent.
+ * Should the compressor take into account the costs of aligned offset symbols?
  */
-#define LZX_CONSIDER_ALIGNED_COSTS	0
+#define LZX_CONSIDER_ALIGNED_COSTS	1
 
 /*
  * LZX_MAX_FAST_LEVEL is the maximum compression level at which we use the
@@ -1557,16 +1556,18 @@ lzx_find_min_cost_path(struct lzx_compressor * const restrict c,
 				u32 offset_data = offset + LZX_OFFSET_ADJUSTMENT;
 				unsigned offset_slot = lzx_comp_get_offset_slot(c, offset_data,
 										is_16_bit);
+				u32 base_cost = cur_node->cost;
+
+			#if LZX_CONSIDER_ALIGNED_COSTS
+				if (offset_data >= 16)
+					base_cost += c->costs.aligned[offset_data &
+								      LZX_ALIGNED_OFFSET_BITMASK];
+			#endif
+
 				do {
-					u32 cost = cur_node->cost +
+					u32 cost = base_cost +
 						   c->costs.match_cost[offset_slot][
 								next_len - LZX_MIN_MATCH_LEN];
-				#if LZX_CONSIDER_ALIGNED_COSTS
-					if (lzx_extra_offset_bits[offset_slot] >=
-					    LZX_NUM_ALIGNED_OFFSET_BITS)
-						cost += c->costs.aligned[offset_data &
-									 LZX_ALIGNED_OFFSET_BITMASK];
-				#endif
 					if (cost < (cur_node + next_len)->cost) {
 						(cur_node + next_len)->cost = cost;
 						(cur_node + next_len)->item =
@@ -1635,7 +1636,7 @@ lzx_compute_match_costs(struct lzx_compressor *c)
 		unsigned i;
 
 	#if LZX_CONSIDER_ALIGNED_COSTS
-		if (lzx_extra_offset_bits[offset_slot] >= LZX_NUM_ALIGNED_OFFSET_BITS)
+		if (offset_slot >= 8)
 			extra_cost -= LZX_NUM_ALIGNED_OFFSET_BITS * LZX_BIT_COST;
 	#endif
 
@@ -1778,9 +1779,9 @@ lzx_compress_near_optimal(struct lzx_compressor *c,
 	const u8 * const in_begin = c->in_buffer;
 	const u8 *	 in_next = in_begin;
 	const u8 * const in_end  = in_begin + c->in_nbytes;
-	unsigned max_len = LZX_MAX_MATCH_LEN;
-	unsigned nice_len = min(c->nice_match_length, max_len);
-	u32 next_hash = 0;
+	u32 max_len = LZX_MAX_MATCH_LEN;
+	u32 nice_len = min(c->nice_match_length, max_len);
+	u32 next_hashes[2] = {};
 	struct lzx_lru_queue queue;
 
 	CALL_BT_MF(is_16_bit, c, bt_matchfinder_init);
@@ -1796,20 +1797,14 @@ lzx_compress_near_optimal(struct lzx_compressor *c,
 		struct lz_match *cache_ptr = c->match_cache;
 		do {
 			struct lz_match *lz_matchptr;
-			unsigned best_len;
+			u32 best_len;
 
 			/* If approaching the end of the input buffer, adjust
 			 * 'max_len' and 'nice_len' accordingly.  */
 			if (unlikely(max_len > in_end - in_next)) {
 				max_len = in_end - in_next;
 				nice_len = min(max_len, nice_len);
-
-				/* This extra check is needed to ensure that we
-				 * never output a length 2 match of the very
-				 * last two bytes with the very first two bytes,
-				 * since such a match has an offset too large to
-				 * be represented.  */
-				if (unlikely(max_len < 3)) {
+				if (unlikely(max_len < 5)) {
 					in_next++;
 					cache_ptr->length = 0;
 					cache_ptr++;
@@ -1824,7 +1819,7 @@ lzx_compress_near_optimal(struct lzx_compressor *c,
 						 max_len,
 						 nice_len,
 						 c->max_search_depth,
-						 &next_hash,
+						 next_hashes,
 						 &best_len,
 						 cache_ptr + 1);
 			in_next++;
@@ -1849,7 +1844,7 @@ lzx_compress_near_optimal(struct lzx_compressor *c,
 					if (unlikely(max_len > in_end - in_next)) {
 						max_len = in_end - in_next;
 						nice_len = min(max_len, nice_len);
-						if (unlikely(max_len < 3)) {
+						if (unlikely(max_len < 5)) {
 							in_next++;
 							cache_ptr->length = 0;
 							cache_ptr++;
@@ -1862,7 +1857,7 @@ lzx_compress_near_optimal(struct lzx_compressor *c,
 						   max_len,
 						   nice_len,
 						   c->max_search_depth,
-						   &next_hash);
+						   next_hashes);
 					in_next++;
 					cache_ptr->length = 0;
 					cache_ptr++;