performance improvements

the heap refill code was recursively implemented with two methods.
I merged both methods.

replace recursion in heap refill method with iterative approach

use array for list of LongQueue
This way there is no precondition when accessing the elements

build.gradle

@@ -7,7 +7,7 @@ buildscript {
     jcenter()
   }
   dependencies {
-    // run with gradle --no-daemon clean jmh
+    // run with Java 11 and ./gradlew --no-daemon clean jmh
     classpath "me.champeau.gradle:jmh-gradle-plugin:0.5.2"
   }
 }
@@ -16,9 +16,13 @@ plugins {
   // usage: gradle dependencyUpdates -Drevision=release
   id "com.github.ben-manes.versions" version "0.34.0"
 }
+
 apply plugin: 'java'
 apply plugin: 'eclipse'
 
+// java compatibility version
+sourceCompatibility = 11
+
 /*
  * The shared configuration for all sub-projects:
  */
@@ -78,7 +82,7 @@ subprojects {
 	jmh {
 		//jvmArgsAppend = ""
 		resultFormat = "JSON"
-		//include = ['.*Union.*'] // include pattern (regular expression) for benchmarks to be executed
+		include = ['.*BenchmarkMultiwayMerge.*'] // include pattern (regular expression) for benchmarks to be executed
 		//exclude = ['some regular expression'] // exclude pattern (regular expression) for benchmarks to be executed
 	}
 }

primitiveCollections/src/jmh/java/org/lucares/collections/BenchmarkMultiwayMerge.java

@@ -0,0 +1,85 @@
+package org.lucares.collections;
+
+import java.util.ArrayList;
+import java.util.List;
+import java.util.concurrent.TimeUnit;
+import java.util.stream.LongStream;
+
+import org.lucares.collections.MultiwayLongMerger.LongQueue;
+import org.openjdk.jmh.annotations.Benchmark;
+import org.openjdk.jmh.annotations.BenchmarkMode;
+import org.openjdk.jmh.annotations.Fork;
+import org.openjdk.jmh.annotations.Measurement;
+import org.openjdk.jmh.annotations.Mode;
+import org.openjdk.jmh.annotations.Param;
+import org.openjdk.jmh.annotations.Scope;
+import org.openjdk.jmh.annotations.Setup;
+import org.openjdk.jmh.annotations.State;
+import org.openjdk.jmh.annotations.TearDown;
+import org.openjdk.jmh.annotations.Warmup;
+
+@State(Scope.Benchmark)
+@BenchmarkMode(Mode.Throughput)
+@Warmup(iterations = 10, time = 500, timeUnit = TimeUnit.MILLISECONDS)
+@Measurement(iterations = 5, time = 500, timeUnit = TimeUnit.MILLISECONDS)
+@Fork(3)
+public class BenchmarkMultiwayMerge {
+
+	@Param({ "10000"/*, "20000" */})
+	private int values;
+	
+	@Param({/*"3","5",*/"10"})
+	private int numLists;
+
+
+	private List<LongList> longSorted = null;
+
+	@Setup
+	public void setup() throws Exception {
+		longSorted = new ArrayList<>();
+		for(int i = 0; i < numLists; i++) {
+			LongList list = LongList.of();
+			LongStream.range(0, values).forEachOrdered(list::add);
+			list.sort();
+			longSorted.add(list);
+		}
+	}
+
+	@TearDown
+	public void tearDown() {
+		longSorted = null;
+	}
+
+	@Benchmark
+	public void testUnionSortedLists_MultiwayMerge() throws Exception {
+
+		LongList.union(longSorted);
+	}
+
+	//@Benchmark
+	public void testUnionSortedLists_TwowayMergeImplementation() throws Exception {
+
+		twowayMerge(longSorted);
+	}
+	
+	private void twowayMerge(List<LongList> longLists) {
+		LongList result = longLists.get(0);
+		for(int i =1; i < longLists.size(); i++) {
+			result = LongList.union(result, longLists.get(i));
+		}
+	}
+	
+	public static void main(String[] args) throws Exception {
+		System.out.println("\n\n----------------\nstart");
+		// -XX:+PrintCompilation
+		for (int i = 0; i < 200; i++) {
+			BenchmarkMultiwayMerge benchmark = new BenchmarkMultiwayMerge();
+			benchmark.setup();
+			//System.out.println("\n\n----------------\n"+i);
+			for (int j = 0; j < 100000; j++) {
+			benchmark.testUnionSortedLists_MultiwayMerge();
+			}
+		}
+		System.out.println("done");
+	}
+}

primitiveCollections/src/main/java/org/lucares/collections/MultiwayLongMerger.java

@@ -8,7 +8,7 @@ import java.util.List;
 class MultiwayLongMerger {
 	private static final long UNSET = Long.MAX_VALUE;
 
-	private static class LongQueue {
+	static class LongQueue {
 		private static final LongQueue EMPTY = new LongQueue(LongList.of());
 
 		final LongList wrapped;
@@ -109,7 +109,9 @@ class MultiwayLongMerger {
 
 	private static class MinValuePriorityQueue {
 
-		private List<LongQueue> longQueues;
+		private static final LongQueue[] EMPTY_QUEUE = new LongQueue[0];
+		
+		private LongQueue[] longQueues;
 
 		/*
 		 * a classic heap where the nodes are layed out in breath first order. First the
@@ -122,7 +124,7 @@ class MultiwayLongMerger {
 		private final int firstLeafIndex;
 
 		public MinValuePriorityQueue(final Collection<LongQueue> longQueues) {
-			this.longQueues = new ArrayList<>(longQueues);
+			final List<LongQueue> tmpQueues = new ArrayList<>(longQueues);
 			size = longQueues.size();
 			heap = new long[2 * nextPowOfTwo(size) - 1];
 
@@ -133,9 +135,10 @@ class MultiwayLongMerger {
 			// fill the longQueues list with empty queues, so that we
 			// have a queue for every leaf in the heap. This makes fillWithMinOfChildren()
 			// easier, because it removes special cases.
-			for (int i = size ; i < nextPowOfTwo(size); i++) {
+			for (int i = size; i < nextPowOfTwo(size); i++) {
-				this.longQueues.add(LongQueue.EMPTY);
+				tmpQueues.add(LongQueue.EMPTY);
 			}
+			this.longQueues = tmpQueues.toArray(EMPTY_QUEUE);
 
 			init();
 		}
@@ -180,7 +183,7 @@ class MultiwayLongMerger {
 			// fill leaf nodes
 			int offset = firstLeafIndex;
 			for (int j = 0; j < size; j++) {
-				final LongQueue q = longQueues.get(j);
+				final LongQueue q = longQueues[j];
 				heap[offset + j] = q.isEmpty() ? UNSET : q.pop();
 			}
 
@@ -211,37 +214,40 @@ class MultiwayLongMerger {
 		}
 
 		private void fillWithMinOfChildren(int index) {
-			final int leftChildIndex = index * 2 + 1; // leftChildIndex(index);
-			final int rightChildIndex = leftChildIndex + 1;// rightChildIndex(index);
 
-			final long valueOfLeftChild = heap[leftChildIndex];
+			int firstLeafOffset = firstLeafIndex;
-			final long valueOfRightChild = heap[rightChildIndex];
 			
-			final int chosenValue;
+			int currentIndex = index;
+			while (true) {
+				final int leftChildIndex = currentIndex * 2 + 1; // leftChildIndex(index);
+				final int rightChildIndex = leftChildIndex + 1;// rightChildIndex(index);
 
+				final long valueOfLeftChild = heap[leftChildIndex];
+				final long valueOfRightChild = heap[rightChildIndex];
 
-			if (valueOfLeftChild < valueOfRightChild) {
+//				final boolean chooseLeft = valueOfLeftChild < valueOfRightChild;
-				// left < right
+//				heap[currentIndex] = chooseLeft ? valueOfLeftChild : valueOfRightChild;
-				heap[index] = valueOfLeftChild;
+//				currentIndex = chooseLeft ? leftChildIndex : rightChildIndex;
-				chosenValue = leftChildIndex;
+				
-			} else {
+				if (valueOfLeftChild < valueOfRightChild) {
-				// left >= right
+					// left < right
-				heap[index] = valueOfRightChild;
+					heap[currentIndex] = valueOfLeftChild;
-				chosenValue = rightChildIndex;
+					currentIndex = leftChildIndex;
+				} else {
+					// left >= right
+					heap[currentIndex] = valueOfRightChild;
+					currentIndex = rightChildIndex;
+				}
+
+				if (currentIndex >= firstLeafOffset) {
+					final int listIndex = currentIndex - firstLeafIndex; // leafIndexToListIndex(index);
+					final LongQueue queue = longQueues[listIndex];
+					heap[currentIndex] = queue.isEmpty() ? UNSET : queue.pop();
+					return;
+				}
 			}
-			
-			refillValue(chosenValue);
 		}
 
-		private void refillValue(int index) {
-			if (isLeaf(index)) {
-				final int listIndex = index - firstLeafIndex; // leafIndexToListIndex(index);
-				final LongQueue queue = longQueues.get(listIndex);
-				heap[index] = queue.isEmpty() ? UNSET : queue.pop();
-				return;
-			}
-			fillWithMinOfChildren(index);
-		}
 	}
 
 	public static void main(String[] args) {