release v1.1.1

CHANGELOG.md

@@ -1,5 +1,9 @@
 ## Changelog
 
+### [1.1.1] - 2022-07-29
+#### Performance
+- improve performance for banded Levenshtein implementation
+
 ### [1.1.0] - 2022-07-29
 #### Fixed
 - fix banded Levenshtein implementation

CMakeLists.txt

@@ -20,7 +20,7 @@ if (CMAKE_BINARY_DIR STREQUAL CMAKE_CURRENT_SOURCE_DIR)
     message(FATAL_ERROR "Building in-source is not supported! Create a build dir and remove ${CMAKE_SOURCE_DIR}/CMakeCache.txt")
 endif()
 
-project(rapidfuzz LANGUAGES CXX VERSION 1.1.0)
+project(rapidfuzz LANGUAGES CXX VERSION 1.1.1)
 
 list(APPEND CMAKE_MODULE_PATH "${CMAKE_CURRENT_LIST_DIR}/cmake")
 include(GNUInstallDirs)

rapidfuzz/distance/Levenshtein.impl

@@ -228,7 +228,8 @@ int64_t levenshtein_hyrroe2003_small_band(const BlockPatternMatchVector& PM, Ran
     ptrdiff_t start_pos = max + 1 - 64;
 
     /* Searching */
-    for (ptrdiff_t i = 0; i < s2.size(); ++i,++start_pos) {
+    ptrdiff_t i = 0;
+    for (; i < s1.size() - max; ++i,++start_pos) {
         /* Step 1: Computing D0 */
         uint64_t PM_j = 0;
         if (start_pos < 0)
@@ -253,16 +254,41 @@ int64_t levenshtein_hyrroe2003_small_band(const BlockPatternMatchVector& PM, Ran
         uint64_t HN = D0 & VP;
 
         /* Step 3: Computing the value D[m,j] */
-        if (i < s1.size() - max)
+        currDist += !bool(D0 & diagonal_mask);
+
+        /* Step 4: Computing Vp and VN */
+        VP = HN | ~((D0 >> 1) | HP);
+        VN = (D0 >> 1) & HP;
+    }
+
+    for (; i < s2.size(); ++i,++start_pos) {
+        /* Step 1: Computing D0 */
+        uint64_t PM_j = 0;
+        if (start_pos < 0)
         {
-            currDist += !bool(D0 & diagonal_mask);
+            PM_j = PM.get(0, s2[i]) << (-start_pos);
         }
         else
         {
-            currDist += bool(HP & horizontal_mask);
-            currDist -= bool(HN & horizontal_mask);
-            horizontal_mask >>= 1;
+            size_t word = static_cast<size_t>(start_pos) / 64;
+            size_t word_pos = static_cast<size_t>(start_pos) % 64;
+
+            PM_j = PM.get(word, s2[i]) >> word_pos;
+
+            if (word + 1 < words && word_pos != 0)
+                PM_j |= PM.get(word + 1, s2[i]) << (64 - word_pos);
         }
+        uint64_t X = PM_j;
+        uint64_t D0 = (((X & VP) + VP) ^ VP) | X | VN;
+
+        /* Step 2: Computing HP and HN */
+        uint64_t HP = VN | ~(D0 | VP);
+        uint64_t HN = D0 & VP;
+
+        /* Step 3: Computing the value D[m,j] */
+        currDist += bool(HP & horizontal_mask);
+        currDist -= bool(HN & horizontal_mask);
+        horizontal_mask >>= 1;
 
         /* Step 4: Computing Vp and VN */
         VP = HN | ~((D0 >> 1) | HP);