]> wimlib.net Git - wimlib/commitdiff
lzx_compress.c: optimize aligned cost consideration and enable by default
authorEric Biggers <ebiggers3@gmail.com>
Sat, 19 Sep 2015 18:56:13 +0000 (13:56 -0500)
committerEric Biggers <ebiggers3@gmail.com>
Sun, 27 Sep 2015 14:41:28 +0000 (09:41 -0500)
src/lzx_compress.c

index 98142962db6d21c866c01367de675a527e28057d..3e19bd191018ef11f28c7be61debfaf4a80181e2 100644 (file)
 #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