jpountz commented on code in PR #13636: URL: https://github.com/apache/lucene/pull/13636#discussion_r1709362066
########## lucene/core/src/java/org/apache/lucene/internal/vectorization/DefaultPostingDecodingUtil.java: ########## @@ -0,0 +1,41 @@ +/* + * 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.internal.vectorization; + +import java.io.IOException; +import org.apache.lucene.store.IndexInput; + +final class DefaultPostingDecodingUtil extends PostingDecodingUtil { + + protected final IndexInput in; + + public DefaultPostingDecodingUtil(IndexInput in) { + this.in = in; + } + + @Override + public void splitLongs(int count, long[] b, int bShift, long bMask, long[] c, long cMask) + throws IOException { + assert count <= 64; + in.readLongs(c, 0, count); + // The below loop is auto-vectorized + for (int i = 0; i < count; ++i) { + b[i] = (c[i] >>> bShift) & bMask; + c[i] &= cMask; Review Comment: I tried a different approach that consists of only specializing 8 and 16. Otherwise, we keep applying the mask all, but then take advantage of the fact that this mask has been applied when decoding tail values using scalar code. This seems to yield a small performance improvement when the vector module is disabled and make performance similar to what it was before. This effectively generalizes the current optimization we have to all numbers of bits per value, so I removed the specialization. ``` Benchmark (bpv) Mode Cnt Score Error Units PostingIndexInputBenchmark.decode 5 thrpt 15 34.857 ± 0.678 ops/us PostingIndexInputBenchmark.decode 6 thrpt 15 33.752 ± 1.060 ops/us PostingIndexInputBenchmark.decode 7 thrpt 15 28.966 ± 4.743 ops/us PostingIndexInputBenchmark.decode 8 thrpt 15 47.009 ± 0.518 ops/us PostingIndexInputBenchmark.decode 9 thrpt 15 28.328 ± 0.978 ops/us PostingIndexInputBenchmark.decode 10 thrpt 15 32.007 ± 1.173 ops/us PostingIndexInputBenchmark.decodeAndPrefixSum 5 thrpt 15 18.161 ± 0.244 ops/us PostingIndexInputBenchmark.decodeAndPrefixSum 6 thrpt 15 17.470 ± 1.149 ops/us PostingIndexInputBenchmark.decodeAndPrefixSum 7 thrpt 15 16.907 ± 1.788 ops/us PostingIndexInputBenchmark.decodeAndPrefixSum 8 thrpt 15 20.259 ± 1.487 ops/us PostingIndexInputBenchmark.decodeAndPrefixSum 9 thrpt 15 14.214 ± 0.521 ops/us PostingIndexInputBenchmark.decodeAndPrefixSum 10 thrpt 15 15.618 ± 1.410 ops/us ``` -- 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
