jpountz commented on code in PR #13636: URL: https://github.com/apache/lucene/pull/13636#discussion_r1707332904
########## lucene/core/src/java21/org/apache/lucene/internal/vectorization/MemorySegmentPostingDecodingUtil.java: ########## @@ -0,0 +1,71 @@ +/* + * 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 java.lang.foreign.MemorySegment; +import java.nio.ByteOrder; +import java.util.Optional; +import jdk.incubator.vector.LongVector; +import jdk.incubator.vector.VectorOperators; +import jdk.incubator.vector.VectorSpecies; +import org.apache.lucene.store.IndexInput; + +final class MemorySegmentPostingDecodingUtil extends PostingDecodingUtil { + + static Optional<PostingDecodingUtil> wrap(IndexInput in, MemorySegment memorySegment) { + if (64 % LONG_SPECIES.length() != 0) { + // Required to meet PostingDecodingUtil's contract that we do not write entries past index 64 + // for any `count` in 0..64. + return Optional.empty(); + } + return Optional.of(new MemorySegmentPostingDecodingUtil(in, memorySegment)); + } + + private static final VectorSpecies<Long> LONG_SPECIES = VectorSpecies.ofPreferred(long.class); + + private final IndexInput in; + private final MemorySegment memorySegment; + + private MemorySegmentPostingDecodingUtil(IndexInput in, MemorySegment memorySegment) { + this.in = in; + this.memorySegment = memorySegment; + } + + @Override + public void splitLongs(int count, long[] b, int bShift, long bMask, long[] c, long cMask) + throws IOException { + long offset = in.getFilePointer(); + long endOffset = offset + count * Long.BYTES; + int i; + // Note: this loop may apply to more than `count` entries due to the width of the preferred + // species. But doing this is faster than handling the remainder with scalar code. + for (i = 0; + i < count; + i += LONG_SPECIES.length(), offset += LONG_SPECIES.length() * Long.BYTES) { + LongVector vector = + LongVector.fromMemorySegment( + LONG_SPECIES, memorySegment, offset, ByteOrder.LITTLE_ENDIAN); + vector + .lanewise(VectorOperators.LSHR, bShift) + .lanewise(VectorOperators.AND, bMask) + .intoArray(b, i); + vector.lanewise(VectorOperators.AND, cMask).intoArray(c, i); Review Comment: I also tried to run the benchmark with AVX disabled. Decoding runs at ~25 ops/us. My understanding is that it's a a bit slower than the auto-vectorized impl because this auto-vectorized impl also takes advantage of AVX instructions. But it's not a 30x slowdown. ########## lucene/core/src/java21/org/apache/lucene/internal/vectorization/MemorySegmentPostingDecodingUtil.java: ########## @@ -0,0 +1,71 @@ +/* + * 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 java.lang.foreign.MemorySegment; +import java.nio.ByteOrder; +import java.util.Optional; +import jdk.incubator.vector.LongVector; +import jdk.incubator.vector.VectorOperators; +import jdk.incubator.vector.VectorSpecies; +import org.apache.lucene.store.IndexInput; + +final class MemorySegmentPostingDecodingUtil extends PostingDecodingUtil { + + static Optional<PostingDecodingUtil> wrap(IndexInput in, MemorySegment memorySegment) { + if (64 % LONG_SPECIES.length() != 0) { + // Required to meet PostingDecodingUtil's contract that we do not write entries past index 64 + // for any `count` in 0..64. + return Optional.empty(); + } + return Optional.of(new MemorySegmentPostingDecodingUtil(in, memorySegment)); + } + + private static final VectorSpecies<Long> LONG_SPECIES = VectorSpecies.ofPreferred(long.class); + + private final IndexInput in; + private final MemorySegment memorySegment; + + private MemorySegmentPostingDecodingUtil(IndexInput in, MemorySegment memorySegment) { + this.in = in; + this.memorySegment = memorySegment; + } + + @Override + public void splitLongs(int count, long[] b, int bShift, long bMask, long[] c, long cMask) + throws IOException { + long offset = in.getFilePointer(); + long endOffset = offset + count * Long.BYTES; + int i; + // Note: this loop may apply to more than `count` entries due to the width of the preferred + // species. But doing this is faster than handling the remainder with scalar code. + for (i = 0; + i < count; + i += LONG_SPECIES.length(), offset += LONG_SPECIES.length() * Long.BYTES) { + LongVector vector = + LongVector.fromMemorySegment( + LONG_SPECIES, memorySegment, offset, ByteOrder.LITTLE_ENDIAN); + vector + .lanewise(VectorOperators.LSHR, bShift) + .lanewise(VectorOperators.AND, bMask) + .intoArray(b, i); + vector.lanewise(VectorOperators.AND, cMask).intoArray(c, i); Review Comment: Thanks for the pointer, I moved some of the existing logic to PamamaVectorizationProvider to be able to reuse it and I disable the explicitly vectorized impl for decoding postings when `HAS_FAST_INTEGER_VECTORS` is `false`. -- 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
