jpountz commented on code in PR #14935: URL: https://github.com/apache/lucene/pull/14935#discussion_r2200809196
########## lucene/benchmark-jmh/src/java/org/apache/lucene/benchmark/jmh/BitsetToArrayBenchmark.java: ########## @@ -0,0 +1,379 @@ +/* + * Licensed to the Apache Software Foundation (ASF) under one or more + * contributor license agreements. See the NOTICE file distributed with + * this work for additional information regarding copyright ownership. + * The ASF licenses this file to You under the Apache License, Version 2.0 + * (the "License"); you may not use this file except in compliance with + * the License. You may obtain a copy of the License at + * + * http://www.apache.org/licenses/LICENSE-2.0 + * + * Unless required by applicable law or agreed to in writing, software + * distributed under the License is distributed on an "AS IS" BASIS, + * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. + * See the License for the specific language governing permissions and + * limitations under the License. + */ +package org.apache.lucene.benchmark.jmh; + +import java.util.SplittableRandom; +import java.util.concurrent.TimeUnit; +import java.util.stream.IntStream; +import org.openjdk.jmh.annotations.Benchmark; +import org.openjdk.jmh.annotations.BenchmarkMode; +import org.openjdk.jmh.annotations.Fork; +import org.openjdk.jmh.annotations.Level; +import org.openjdk.jmh.annotations.Measurement; +import org.openjdk.jmh.annotations.Mode; +import org.openjdk.jmh.annotations.OutputTimeUnit; +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.Warmup; + +@BenchmarkMode(Mode.Throughput) +@OutputTimeUnit(TimeUnit.MICROSECONDS) +@State(Scope.Benchmark) +@Warmup(iterations = 3, time = 1) +@Measurement(iterations = 5, time = 1) +@Fork( + value = 1, + jvmArgsAppend = {"-Xmx1g", "-Xms1g", "-XX:+AlwaysPreTouch"}) +public class BitsetToArrayBenchmark { + + private final SplittableRandom R = new SplittableRandom(4314123142L); + + @Param({"5", "10", "20", "30", "40", "50", "60"}) + int bitCount; + + private final int[] scratch = new int[64]; + private long word; + private int[] resultArray; + private int base; + private int offset; + + @Setup(Level.Trial) + public void setup() { + base = R.nextInt(1000); + resultArray = new int[bitCount + Long.SIZE]; + } + + @Setup(Level.Invocation) + public void setupInvocation() { + word = 0L; + while (Long.bitCount(word) < bitCount) { + word |= 1L << R.nextInt(64); + } + offset = R.nextInt(64); + } + + @Benchmark + public int whileLoop() { + return _whileLoop(word, resultArray, offset, base); + } + + @Benchmark + public int forLoop() { + return _forLoop(word, resultArray, offset, base); + } + + @Benchmark + public int forLoopManualUnrolling() { + return _forLoopManualUnrolling(word, resultArray, offset, base); + } + + @Benchmark + public int dense() { + return _dense(word, resultArray, offset, base); + } + + @Benchmark + public int denseBranchLess() { + return _denseBranchLess(word, resultArray, offset, base); + } + + @Benchmark + public int denseBranchLessUnrolling() { + return _denseBranchLessUnrolling(word, resultArray, offset, base); + } + + @Benchmark + public int denseBranchLessVectorized() { + return _denseBranchLessVectorized(word, resultArray, offset, base, scratch); + } + + @Benchmark + public int denseInvert() { + return _denseInvert(word, resultArray, offset, base); + } + + @Benchmark + public int hybrid() { + return _hybrid(word, resultArray, offset, base, scratch); + } + + private static int _whileLoop(long word, int[] resultArray, int offset, int base) { + while (word != 0) { + int bit = Long.numberOfTrailingZeros(word); + resultArray[offset++] = base + bit; + word ^= 1L << bit; + } + return offset; + } + + private static int _forLoop(long word, int[] resultArray, int offset, int base) { + int to = offset + Long.bitCount(word); + for (int i = offset; i < to; i++) { + int bit = Long.numberOfTrailingZeros(word); + resultArray[i] = base + bit; + word ^= 1L << bit; + } + return to; + } + + private static int _forLoopManualUnrolling(long word, int[] resultArray, int offset, int base) { + int to = offset + Long.bitCount(word); + int i = offset; + + for (; i < to - 3; i += 4) { + int ntz = Long.numberOfTrailingZeros(word); + resultArray[i] = base + ntz; + word ^= 1L << ntz; + ntz = Long.numberOfTrailingZeros(word); + resultArray[i] = base + ntz; + word ^= 1L << ntz; + ntz = Long.numberOfTrailingZeros(word); + resultArray[i] = base + ntz; + word ^= 1L << ntz; + ntz = Long.numberOfTrailingZeros(word); + resultArray[i] = base + ntz; + word ^= 1L << ntz; + } + + for (; i < to; i++) { + int ntz = Long.numberOfTrailingZeros(word); + resultArray[i] = base + ntz; + word ^= 1L << ntz; + } + + return to; + } + + private static int _dense(long word, int[] resultArray, int offset, int base) { + for (int i = 0; i < Long.SIZE; i++) { + if ((word & (1L << i)) != 0) { + resultArray[offset++] = base + i; + } + } + return offset; + } + + private static int _denseBranchLess(long word, int[] resultArray, int offset, int base) { + int lWord = (int) word; + int hWord = (int) (word >>> 32); + for (int i = 0; i < Integer.SIZE; i++) { + resultArray[offset] = base + i; + offset += lWord & 1; + lWord >>>= 1; + } + for (int i = Integer.SIZE; i < Long.SIZE; i++) { + resultArray[offset] = base + i; + offset += hWord & 1; + hWord >>>= 1; + } + return offset; + } + + private static int _denseBranchLessUnrolling(long word, int[] resultArray, int offset, int base) { + int lWord = (int) word; + int hWord = (int) (word >>> 32); + + resultArray[offset] = base + 0; + offset += (lWord >>> 0) & 1; + resultArray[offset] = base + 1; + offset += (lWord >>> 1) & 1; + resultArray[offset] = base + 2; + offset += (lWord >>> 2) & 1; + resultArray[offset] = base + 3; + offset += (lWord >>> 3) & 1; + resultArray[offset] = base + 4; + offset += (lWord >>> 4) & 1; + resultArray[offset] = base + 5; + offset += (lWord >>> 5) & 1; + resultArray[offset] = base + 6; + offset += (lWord >>> 6) & 1; + resultArray[offset] = base + 7; + offset += (lWord >>> 7) & 1; + resultArray[offset] = base + 8; + offset += (lWord >>> 8) & 1; + resultArray[offset] = base + 9; + offset += (lWord >>> 9) & 1; + resultArray[offset] = base + 10; + offset += (lWord >>> 10) & 1; + resultArray[offset] = base + 11; + offset += (lWord >>> 11) & 1; + resultArray[offset] = base + 12; + offset += (lWord >>> 12) & 1; + resultArray[offset] = base + 13; + offset += (lWord >>> 13) & 1; + resultArray[offset] = base + 14; + offset += (lWord >>> 14) & 1; + resultArray[offset] = base + 15; + offset += (lWord >>> 15) & 1; + resultArray[offset] = base + 16; + offset += (lWord >>> 16) & 1; + resultArray[offset] = base + 17; + offset += (lWord >>> 17) & 1; + resultArray[offset] = base + 18; + offset += (lWord >>> 18) & 1; + resultArray[offset] = base + 19; + offset += (lWord >>> 19) & 1; + resultArray[offset] = base + 20; + offset += (lWord >>> 20) & 1; + resultArray[offset] = base + 21; + offset += (lWord >>> 21) & 1; + resultArray[offset] = base + 22; + offset += (lWord >>> 22) & 1; + resultArray[offset] = base + 23; + offset += (lWord >>> 23) & 1; + resultArray[offset] = base + 24; + offset += (lWord >>> 24) & 1; + resultArray[offset] = base + 25; + offset += (lWord >>> 25) & 1; + resultArray[offset] = base + 26; + offset += (lWord >>> 26) & 1; + resultArray[offset] = base + 27; + offset += (lWord >>> 27) & 1; + resultArray[offset] = base + 28; + offset += (lWord >>> 28) & 1; + resultArray[offset] = base + 29; + offset += (lWord >>> 29) & 1; + resultArray[offset] = base + 30; + offset += (lWord >>> 30) & 1; + resultArray[offset] = base + 31; + offset += (lWord >>> 31) & 1; + + resultArray[offset] = base + 32; + offset += (hWord >>> 0) & 1; + resultArray[offset] = base + 33; + offset += (hWord >>> 1) & 1; + resultArray[offset] = base + 34; + offset += (hWord >>> 2) & 1; + resultArray[offset] = base + 35; + offset += (hWord >>> 3) & 1; + resultArray[offset] = base + 36; + offset += (hWord >>> 4) & 1; + resultArray[offset] = base + 37; + offset += (hWord >>> 5) & 1; + resultArray[offset] = base + 38; + offset += (hWord >>> 6) & 1; + resultArray[offset] = base + 39; + offset += (hWord >>> 7) & 1; + resultArray[offset] = base + 40; + offset += (hWord >>> 8) & 1; + resultArray[offset] = base + 41; + offset += (hWord >>> 9) & 1; + resultArray[offset] = base + 42; + offset += (hWord >>> 10) & 1; + resultArray[offset] = base + 43; + offset += (hWord >>> 11) & 1; + resultArray[offset] = base + 44; + offset += (hWord >>> 12) & 1; + resultArray[offset] = base + 45; + offset += (hWord >>> 13) & 1; + resultArray[offset] = base + 46; + offset += (hWord >>> 14) & 1; + resultArray[offset] = base + 47; + offset += (hWord >>> 15) & 1; + resultArray[offset] = base + 48; + offset += (hWord >>> 16) & 1; + resultArray[offset] = base + 49; + offset += (hWord >>> 17) & 1; + resultArray[offset] = base + 50; + offset += (hWord >>> 18) & 1; + resultArray[offset] = base + 51; + offset += (hWord >>> 19) & 1; + resultArray[offset] = base + 52; + offset += (hWord >>> 20) & 1; + resultArray[offset] = base + 53; + offset += (hWord >>> 21) & 1; + resultArray[offset] = base + 54; + offset += (hWord >>> 22) & 1; + resultArray[offset] = base + 55; + offset += (hWord >>> 23) & 1; + resultArray[offset] = base + 56; + offset += (hWord >>> 24) & 1; + resultArray[offset] = base + 57; + offset += (hWord >>> 25) & 1; + resultArray[offset] = base + 58; + offset += (hWord >>> 26) & 1; + resultArray[offset] = base + 59; + offset += (hWord >>> 27) & 1; + resultArray[offset] = base + 60; + offset += (hWord >>> 28) & 1; + resultArray[offset] = base + 61; + offset += (hWord >>> 29) & 1; + resultArray[offset] = base + 62; + offset += (hWord >>> 30) & 1; + resultArray[offset] = base + 63; + offset += (hWord >>> 31) & 1; + + return offset; + } + + static int[] IDENTITY = IntStream.range(0, 32).toArray(); + + private static int _denseBranchLessVectorized( + long word, int[] resultArray, int offset, int base, int[] scratch) { + int lWord = (int) word; + int hWord = (int) (word >>> 32); + + // vectorized loop + for (int i = 0; i < 32; i++) { + scratch[i] = (lWord >>> IDENTITY[i]) & 1; + scratch[i + 32] = (hWord >>> IDENTITY[i]) & 1; Review Comment: Did it perform better with `IDENTITY[i]` than just `i`? ########## lucene/benchmark-jmh/src/java/org/apache/lucene/benchmark/jmh/BitsetToArrayBenchmark.java: ########## @@ -0,0 +1,379 @@ +/* + * Licensed to the Apache Software Foundation (ASF) under one or more + * contributor license agreements. See the NOTICE file distributed with + * this work for additional information regarding copyright ownership. + * The ASF licenses this file to You under the Apache License, Version 2.0 + * (the "License"); you may not use this file except in compliance with + * the License. You may obtain a copy of the License at + * + * http://www.apache.org/licenses/LICENSE-2.0 + * + * Unless required by applicable law or agreed to in writing, software + * distributed under the License is distributed on an "AS IS" BASIS, + * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. + * See the License for the specific language governing permissions and + * limitations under the License. + */ +package org.apache.lucene.benchmark.jmh; + +import java.util.SplittableRandom; +import java.util.concurrent.TimeUnit; +import java.util.stream.IntStream; +import org.openjdk.jmh.annotations.Benchmark; +import org.openjdk.jmh.annotations.BenchmarkMode; +import org.openjdk.jmh.annotations.Fork; +import org.openjdk.jmh.annotations.Level; +import org.openjdk.jmh.annotations.Measurement; +import org.openjdk.jmh.annotations.Mode; +import org.openjdk.jmh.annotations.OutputTimeUnit; +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.Warmup; + +@BenchmarkMode(Mode.Throughput) +@OutputTimeUnit(TimeUnit.MICROSECONDS) +@State(Scope.Benchmark) +@Warmup(iterations = 3, time = 1) +@Measurement(iterations = 5, time = 1) +@Fork( + value = 1, + jvmArgsAppend = {"-Xmx1g", "-Xms1g", "-XX:+AlwaysPreTouch"}) +public class BitsetToArrayBenchmark { + + private final SplittableRandom R = new SplittableRandom(4314123142L); + + @Param({"5", "10", "20", "30", "40", "50", "60"}) + int bitCount; + + private final int[] scratch = new int[64]; + private long word; + private int[] resultArray; + private int base; + private int offset; + + @Setup(Level.Trial) + public void setup() { + base = R.nextInt(1000); + resultArray = new int[bitCount + Long.SIZE]; + } + + @Setup(Level.Invocation) + public void setupInvocation() { + word = 0L; + while (Long.bitCount(word) < bitCount) { + word |= 1L << R.nextInt(64); + } + offset = R.nextInt(64); + } + + @Benchmark + public int whileLoop() { + return _whileLoop(word, resultArray, offset, base); + } + + @Benchmark + public int forLoop() { + return _forLoop(word, resultArray, offset, base); + } + + @Benchmark + public int forLoopManualUnrolling() { + return _forLoopManualUnrolling(word, resultArray, offset, base); + } + + @Benchmark + public int dense() { + return _dense(word, resultArray, offset, base); + } + + @Benchmark + public int denseBranchLess() { + return _denseBranchLess(word, resultArray, offset, base); + } + + @Benchmark + public int denseBranchLessUnrolling() { + return _denseBranchLessUnrolling(word, resultArray, offset, base); + } + + @Benchmark + public int denseBranchLessVectorized() { + return _denseBranchLessVectorized(word, resultArray, offset, base, scratch); + } + + @Benchmark + public int denseInvert() { + return _denseInvert(word, resultArray, offset, base); + } + + @Benchmark + public int hybrid() { + return _hybrid(word, resultArray, offset, base, scratch); + } + + private static int _whileLoop(long word, int[] resultArray, int offset, int base) { + while (word != 0) { + int bit = Long.numberOfTrailingZeros(word); + resultArray[offset++] = base + bit; + word ^= 1L << bit; + } + return offset; + } + + private static int _forLoop(long word, int[] resultArray, int offset, int base) { + int to = offset + Long.bitCount(word); + for (int i = offset; i < to; i++) { + int bit = Long.numberOfTrailingZeros(word); + resultArray[i] = base + bit; + word ^= 1L << bit; + } + return to; + } + + private static int _forLoopManualUnrolling(long word, int[] resultArray, int offset, int base) { + int to = offset + Long.bitCount(word); + int i = offset; + + for (; i < to - 3; i += 4) { + int ntz = Long.numberOfTrailingZeros(word); + resultArray[i] = base + ntz; + word ^= 1L << ntz; + ntz = Long.numberOfTrailingZeros(word); + resultArray[i] = base + ntz; + word ^= 1L << ntz; + ntz = Long.numberOfTrailingZeros(word); + resultArray[i] = base + ntz; + word ^= 1L << ntz; + ntz = Long.numberOfTrailingZeros(word); + resultArray[i] = base + ntz; + word ^= 1L << ntz; + } + + for (; i < to; i++) { + int ntz = Long.numberOfTrailingZeros(word); + resultArray[i] = base + ntz; + word ^= 1L << ntz; + } + + return to; + } + + private static int _dense(long word, int[] resultArray, int offset, int base) { + for (int i = 0; i < Long.SIZE; i++) { + if ((word & (1L << i)) != 0) { + resultArray[offset++] = base + i; + } + } + return offset; + } + + private static int _denseBranchLess(long word, int[] resultArray, int offset, int base) { + int lWord = (int) word; + int hWord = (int) (word >>> 32); + for (int i = 0; i < Integer.SIZE; i++) { + resultArray[offset] = base + i; + offset += lWord & 1; + lWord >>>= 1; + } + for (int i = Integer.SIZE; i < Long.SIZE; i++) { + resultArray[offset] = base + i; + offset += hWord & 1; + hWord >>>= 1; + } + return offset; Review Comment: Another way to make it branchless may be to try to get the compiler to use a cmov, e.g. I'm curious how something like this would perform: ```java for (int j = 0; j < Long.SIZE; ++j) { resultArray[offset] = base + j; if ((word & (1L << j)) != 0L) { // hopefully compiles as cmov? offset++; } } ``` -- This is an automated message from the Apache Git Service. To respond to the message, please log on to GitHub and use the URL above to go to the specific comment. To unsubscribe, e-mail: issues-unsubscr...@lucene.apache.org For queries about this service, please contact Infrastructure at: us...@infra.apache.org --------------------------------------------------------------------- To unsubscribe, e-mail: issues-unsubscr...@lucene.apache.org For additional commands, e-mail: issues-h...@lucene.apache.org
