ICU-23061 Transliterator improve scalability by avoiding monitor contention

stevenschlansker · echeran · commit fd27c483fc55 · 2025-06-26T14:28:16.000-07:00
Current implementation of lower, upper, title, etc synchronizes the whole
operation, so it is always single-threaded

Removing the shared state lets us utilize multiple CPU cores
diff --git a/icu4j/main/translit/src/main/java/com/ibm/icu/text/BreakTransliterator.java b/icu4j/main/translit/src/main/java/com/ibm/icu/text/BreakTransliterator.java
@@ -22,8 +22,6 @@
 final class BreakTransliterator extends Transliterator {
     private BreakIterator bi;
     private String insertion;
-    private int[] boundaries = new int[50];
-    private int boundaryCount = 0;
 
     public BreakTransliterator(String ID, UnicodeFilter filter, BreakIterator bi, String insertion) {
         super(ID, filter);
@@ -52,8 +50,9 @@ public void setInsertion(String insertion) {
     public BreakIterator getBreakIterator() {
         // Defer initialization of BreakIterator because it is slow,
         // typically over 2000 ms.
-        if (bi == null) bi = BreakIterator.getWordInstance(new ULocale("th_TH"));
-        return bi;
+        // Using a holder class for safe init without a volatile-read.
+        if (bi == null) bi = WordBreakIteratorHolder.BI;
+        return (BreakIterator) bi.clone();
     }
 
     ///CLOVER:OFF
@@ -74,10 +73,11 @@ public void setBreakIterator(BreakIterator bi) {
         | (1<<Character.ENCLOSING_MARK)
         ;
     @Override
-    protected synchronized void handleTransliterate(Replaceable text, Position pos, boolean incremental) {
-        boundaryCount = 0;
+    protected void handleTransliterate(Replaceable text, Position pos, boolean incremental) {
+        int[] boundaries = new int[50];
+        int boundaryCount = 0;
         int boundary = 0;
-        getBreakIterator(); // Lazy-create it if necessary
+        BreakIterator bi = getBreakIterator(); // Lazy-create it if necessary
         bi.setText(new ReplaceableCharacterIterator(text, pos.start, pos.limit, pos.start));
         // TODO: fix clumsy workaround used below.
         /*
@@ -416,4 +416,7 @@ public void addSourceTargetSet(UnicodeSet inputFilter, UnicodeSet sourceSet, Uni
         }
     }
 
+    private static class WordBreakIteratorHolder {
+        static final BreakIterator BI = BreakIterator.getWordInstance(new ULocale("th_TH"));
+    }
 }
diff --git a/icu4j/main/translit/src/main/java/com/ibm/icu/text/CaseFoldTransliterator.java b/icu4j/main/translit/src/main/java/com/ibm/icu/text/CaseFoldTransliterator.java
@@ -39,8 +39,6 @@ public Transliterator getInstance(String ID) {
     }
 
     private final UCaseProps csp;
-    private ReplaceableContextIterator iter;
-    private StringBuilder result;
 
     /**
      * Constructs a transliterator.
@@ -49,15 +47,13 @@ public Transliterator getInstance(String ID) {
     public CaseFoldTransliterator() {
         super(_ID, null);
         csp=UCaseProps.INSTANCE;
-        iter=new ReplaceableContextIterator();
-        result = new StringBuilder();
     }
 
     /**
      * Implements {@link Transliterator#handleTransliterate}.
      */
     @Override
-    protected synchronized void handleTransliterate(Replaceable text,
+    protected void handleTransliterate(Replaceable text,
                                        Position offsets, boolean isIncremental) {
         if(csp==null) {
             return;
@@ -67,8 +63,10 @@ protected synchronized void handleTransliterate(Replaceable text,
             return;
         }
 
+        ReplaceableContextIterator iter = new ReplaceableContextIterator();
+        StringBuilder result = new StringBuilder();
+
         iter.setText(text);
-        result.setLength(0);
         int c, delta;
 
         // Walk through original string
diff --git a/icu4j/main/translit/src/main/java/com/ibm/icu/text/LowercaseTransliterator.java b/icu4j/main/translit/src/main/java/com/ibm/icu/text/LowercaseTransliterator.java
@@ -42,8 +42,6 @@ public Transliterator getInstance(String ID) {
     private final ULocale locale;
 
     private final UCaseProps csp;
-    private ReplaceableContextIterator iter;
-    private StringBuilder result;
     private int caseLocale;
 
     /**
@@ -54,16 +52,14 @@ public LowercaseTransliterator(ULocale loc) {
         super(_ID, null);
         locale = loc;
         csp=UCaseProps.INSTANCE;
-        iter=new ReplaceableContextIterator();
-        result = new StringBuilder();
         caseLocale = UCaseProps.getCaseLocale(locale);
     }
 
     /**
      * Implements {@link Transliterator#handleTransliterate}.
      */
     @Override
-    protected synchronized void handleTransliterate(Replaceable text,
+    protected void handleTransliterate(Replaceable text,
                                        Position offsets, boolean isIncremental) {
         if(csp==null) {
             return;
@@ -73,8 +69,10 @@ protected synchronized void handleTransliterate(Replaceable text,
             return;
         }
 
+        ReplaceableContextIterator iter = new ReplaceableContextIterator();
+        StringBuilder result = new StringBuilder();
+
         iter.setText(text);
-        result.setLength(0);
         int c, delta;
 
         // Walk through original string
diff --git a/icu4j/main/translit/src/main/java/com/ibm/icu/text/TitlecaseTransliterator.java b/icu4j/main/translit/src/main/java/com/ibm/icu/text/TitlecaseTransliterator.java
@@ -40,8 +40,6 @@ public Transliterator getInstance(String ID) {
     private final ULocale locale;
 
     private final UCaseProps csp;
-    private ReplaceableContextIterator iter;
-    private StringBuilder result;
     private int caseLocale;
 
    /**
@@ -53,16 +51,14 @@ public TitlecaseTransliterator(ULocale loc) {
         // Need to look back 2 characters in the case of "can't"
         setMaximumContextLength(2);
         csp=UCaseProps.INSTANCE;
-        iter=new ReplaceableContextIterator();
-        result = new StringBuilder();
         caseLocale = UCaseProps.getCaseLocale(locale);
     }
 
     /**
      * Implements {@link Transliterator#handleTransliterate}.
      */
     @Override
-    protected synchronized void handleTransliterate(Replaceable text,
+    protected void handleTransliterate(Replaceable text,
                                        Position offsets, boolean isIncremental) {
         // TODO reimplement, see ustrcase.c
         // using a real word break iterator
@@ -75,6 +71,9 @@ protected synchronized void handleTransliterate(Replaceable text,
             return;
         }
 
+        ReplaceableContextIterator iter = new ReplaceableContextIterator();
+        StringBuilder result = new StringBuilder();
+
         // case type: >0 cased (UCaseProps.LOWER etc.)  ==0 uncased  <0 case-ignorable
         int type;
 
@@ -108,8 +107,6 @@ protected synchronized void handleTransliterate(Replaceable text,
         iter.setLimit(offsets.limit);
         iter.setContextLimits(offsets.contextStart, offsets.contextLimit);
 
-        result.setLength(0);
-
         // Walk through original string
         // If there is a case change, modify corresponding position in replaceable
         int delta;
diff --git a/icu4j/main/translit/src/main/java/com/ibm/icu/text/UppercaseTransliterator.java b/icu4j/main/translit/src/main/java/com/ibm/icu/text/UppercaseTransliterator.java
@@ -39,8 +39,6 @@ public Transliterator getInstance(String ID) {
     private final ULocale locale;
 
     private final UCaseProps csp;
-    private ReplaceableContextIterator iter;
-    private StringBuilder result;
     private int caseLocale;
 
     /**
@@ -50,16 +48,14 @@ public UppercaseTransliterator(ULocale loc) {
         super(_ID, null);
         locale = loc;
         csp=UCaseProps.INSTANCE;
-        iter=new ReplaceableContextIterator();
-        result = new StringBuilder();
         caseLocale = UCaseProps.getCaseLocale(locale);
     }
 
     /**
      * Implements {@link Transliterator#handleTransliterate}.
      */
     @Override
-    protected synchronized void handleTransliterate(Replaceable text,
+    protected void handleTransliterate(Replaceable text,
                 Position offsets, boolean isIncremental) {
         if(csp==null) {
             return;
@@ -69,8 +65,10 @@ protected synchronized void handleTransliterate(Replaceable text,
             return;
         }
 
+        ReplaceableContextIterator iter = new ReplaceableContextIterator();
+        StringBuilder result = new StringBuilder();
+
         iter.setText(text);
-        result.setLength(0);
         int c, delta;
 
         // Walk through original string
diff --git a/icu4j/perf-tests/README.txt b/icu4j/perf-tests/README.txt
@@ -43,6 +43,9 @@ COLLATION TESTS
     The collation tests run only on the command line with tabular output:
     perl collationperf.pl |& tee collation_output.txt
 
+JMH
+Some performance tests run using OpenJDK JMH. Example invocation:
+    mvn clean package exec:java -pl perf-tests -Pjmh_benchmark
 
 OTHER COMMAND LINE TESTS
 Additional tests are run from the command line, each producing an HTML
diff --git a/icu4j/perf-tests/pom.xml b/icu4j/perf-tests/pom.xml
@@ -43,6 +43,73 @@
       <artifactId>commons-cli</artifactId>
       <version>${commons-cli.version}</version>
     </dependency>
+    <dependency>
+      <groupId>org.openjdk.jmh</groupId>
+      <artifactId>jmh-core</artifactId>
+      <version>${jmh.version}</version>
+    </dependency>
+    <dependency>
+      <groupId>org.openjdk.jmh</groupId>
+      <artifactId>jmh-generator-annprocess</artifactId>
+      <version>${jmh.version}</version>
+      <scope>provided</scope>
+    </dependency>
   </dependencies>
 
+  <profiles>
+    <profile>
+      <id>jmh_benchmark</id>
+      <build>
+        <plugins>
+          <plugin>
+            <groupId>org.apache.maven.plugins</groupId>
+            <artifactId>maven-compiler-plugin</artifactId>
+            <configuration>
+              <annotationProcessors>
+                <annotationProcessor>org.openjdk.jmh.generators.BenchmarkProcessor</annotationProcessor>
+              </annotationProcessors>
+            </configuration>
+          </plugin>
+          <plugin>
+            <groupId>org.apache.maven.plugins</groupId>
+            <artifactId>maven-dependency-plugin</artifactId>
+            <executions>
+              <execution>
+                <id>build-jmh-classpath</id>
+                <goals>
+                  <goal>build-classpath</goal>
+                </goals>
+                <configuration>
+                  <includeScope>runtime</includeScope>
+                  <outputProperty>jmhClasspath</outputProperty>
+                </configuration>
+              </execution>
+            </executions>
+          </plugin>
+          <plugin>
+            <groupId>org.codehaus.mojo</groupId>
+            <artifactId>exec-maven-plugin</artifactId>
+            <configuration>
+              <mainClass>org.openjdk.jmh.Main</mainClass>
+              <arguments>
+                <argument>-f</argument>
+                <argument>1</argument>
+                <argument>-wi</argument>
+                <argument>5</argument>
+                <argument>-i</argument>
+                <argument>10</argument>
+              </arguments>
+              <systemProperties>
+                <property>
+                  <key>java.class.path</key>
+                  <value>${project.build.outputDirectory}${path.separator}${jmhClasspath}</value>
+                </property>
+              </systemProperties>
+            </configuration>
+          </plugin>
+        </plugins>
+      </build>
+    </profile>
+  </profiles>
+
 </project>
diff --git a/icu4j/perf-tests/src/main/java/com/ibm/icu/dev/test/perf/BreakTransliteratorPerfTest.java b/icu4j/perf-tests/src/main/java/com/ibm/icu/dev/test/perf/BreakTransliteratorPerfTest.java
@@ -0,0 +1,30 @@
+// © 2025 and later: Unicode, Inc. and others.
+// License & terms of use: http://www.unicode.org/copyright.html
+package com.ibm.icu.dev.test.perf;
+
+import java.util.concurrent.TimeUnit;
+
+import com.ibm.icu.text.BreakTransliteratorAccess;
+import com.ibm.icu.text.Transliterator;
+import org.openjdk.jmh.annotations.Benchmark;
+import org.openjdk.jmh.annotations.BenchmarkMode;
+import org.openjdk.jmh.annotations.Mode;
+import org.openjdk.jmh.annotations.OutputTimeUnit;
+
+@BenchmarkMode(Mode.Throughput)
+@OutputTimeUnit(TimeUnit.MILLISECONDS)
+public class BreakTransliteratorPerfTest {
+
+    static final Transliterator TITLE = BreakTransliteratorAccess.newInstance();
+
+    @Benchmark
+    public String testShort() {
+        return TITLE.transliterate("Cat");
+    }
+
+    @Benchmark
+    public String testSentence() {
+        return TITLE.transliterate("The Quick Brown Fox jumped over the Lazy Dog");
+    }
+
+}
diff --git a/icu4j/perf-tests/src/main/java/com/ibm/icu/dev/test/perf/CaseFoldTransliteratorPerfTest.java b/icu4j/perf-tests/src/main/java/com/ibm/icu/dev/test/perf/CaseFoldTransliteratorPerfTest.java
@@ -0,0 +1,29 @@
+// © 2025 and later: Unicode, Inc. and others.
+// License & terms of use: http://www.unicode.org/copyright.html
+package com.ibm.icu.dev.test.perf;
+
+import java.util.concurrent.TimeUnit;
+
+import com.ibm.icu.text.Transliterator;
+import org.openjdk.jmh.annotations.Benchmark;
+import org.openjdk.jmh.annotations.BenchmarkMode;
+import org.openjdk.jmh.annotations.Mode;
+import org.openjdk.jmh.annotations.OutputTimeUnit;
+
+@BenchmarkMode(Mode.Throughput)
+@OutputTimeUnit(TimeUnit.MILLISECONDS)
+public class CaseFoldTransliteratorPerfTest {
+
+    static final Transliterator CASE = Transliterator.getInstance("CaseFold");
+
+    @Benchmark
+    public String testShort() {
+        return CASE.transliterate("Cat");
+    }
+
+    @Benchmark
+    public String testSentence() {
+        return CASE.transliterate("The Quick Brown Fox Jumped Over The Lazy Dog");
+    }
+
+}
diff --git a/icu4j/perf-tests/src/main/java/com/ibm/icu/dev/test/perf/LowercaseTransliteratorPerfTest.java b/icu4j/perf-tests/src/main/java/com/ibm/icu/dev/test/perf/LowercaseTransliteratorPerfTest.java
@@ -0,0 +1,29 @@
+// © 2025 and later: Unicode, Inc. and others.
+// License & terms of use: http://www.unicode.org/copyright.html
+package com.ibm.icu.dev.test.perf;
+
+import java.util.concurrent.TimeUnit;
+
+import com.ibm.icu.text.Transliterator;
+import org.openjdk.jmh.annotations.Benchmark;
+import org.openjdk.jmh.annotations.BenchmarkMode;
+import org.openjdk.jmh.annotations.Mode;
+import org.openjdk.jmh.annotations.OutputTimeUnit;
+
+@BenchmarkMode(Mode.Throughput)
+@OutputTimeUnit(TimeUnit.MILLISECONDS)
+public class LowercaseTransliteratorPerfTest {
+
+    static final Transliterator LOWER = Transliterator.getInstance("Lower");
+
+    @Benchmark
+    public String testShort() {
+        return LOWER.transliterate("Cat");
+    }
+
+    @Benchmark
+    public String testSentence() {
+        return LOWER.transliterate("The Quick Brown Fox Jumped Over The Lazy Dog");
+    }
+
+}
diff --git a/icu4j/perf-tests/src/main/java/com/ibm/icu/dev/test/perf/TitlecaseTransliteratorPerfTest.java b/icu4j/perf-tests/src/main/java/com/ibm/icu/dev/test/perf/TitlecaseTransliteratorPerfTest.java
diff --git a/icu4j/perf-tests/src/main/java/com/ibm/icu/dev/test/perf/UppercaseTransliteratorPerfTest.java b/icu4j/perf-tests/src/main/java/com/ibm/icu/dev/test/perf/UppercaseTransliteratorPerfTest.java
diff --git a/icu4j/perf-tests/src/main/java/com/ibm/icu/text/BreakTransliteratorAccess.java b/icu4j/perf-tests/src/main/java/com/ibm/icu/text/BreakTransliteratorAccess.java
diff --git a/icu4j/pom.xml b/icu4j/pom.xml
