Github user cloud-fan commented on a diff in the pull request:

    https://github.com/apache/spark/pull/19222#discussion_r170305944
  
    --- Diff: 
common/unsafe/src/main/java/org/apache/spark/unsafe/memory/ByteArrayMemoryBlock.java
 ---
    @@ -0,0 +1,190 @@
    +/*
    + * 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.spark.unsafe.memory;
    +
    +import org.apache.spark.unsafe.Platform;
    +
    +/**
    + * A consecutive block of memory with a byte array on Java heap.
    + */
    +public final class ByteArrayMemoryBlock extends MemoryBlock {
    +
    +  private final byte[] array;
    +
    +  public ByteArrayMemoryBlock(byte[] obj, long offset, long length) {
    +    super(obj, offset, length);
    +    this.array = obj;
    +  }
    +
    +  @Override
    +  public MemoryBlock allocate(long offset, long size) {
    +    return new ByteArrayMemoryBlock(array, offset, size);
    +  }
    +
    +  public byte[] getByteArray() { return array; }
    +
    +  /**
    +   * Creates a memory block pointing to the memory used by the byte array.
    +   */
    +  public static ByteArrayMemoryBlock fromArray(final byte[] array) {
    +    return new ByteArrayMemoryBlock(array, Platform.BYTE_ARRAY_OFFSET, 
array.length);
    +  }
    +
    +
    +  public final int getInt(long offset) {
    +    // UTF8String.getPrefix() assumes data is 4-byte aligned
    +    assert(offset + 4 - Platform.BYTE_ARRAY_OFFSET <= ((array.length + 3) 
/ 4) * 4);
    +    return Platform.getInt(array, offset);
    +  }
    +
    +  public final void putInt(long offset, int value) {
    +    assert(offset + 4 - Platform.BYTE_ARRAY_OFFSET <= array.length);
    +    Platform.putInt(array, offset, value);
    +  }
    +
    +  public final boolean getBoolean(long offset) {
    +    assert(offset + 1 - Platform.BYTE_ARRAY_OFFSET <= array.length);
    +    return Platform.getBoolean(array, offset);
    +  }
    +
    +  public final void putBoolean(long offset, boolean value) {
    +    assert(offset + 1 - Platform.BYTE_ARRAY_OFFSET <= array.length);
    +    Platform.putBoolean(array, offset, value);
    +  }
    +
    +  public final byte getByte(long offset) {
    +    return array[(int)(offset - Platform.BYTE_ARRAY_OFFSET)];
    +  }
    +
    +  public final void putByte(long offset, byte value) {
    +    array[(int)(offset - Platform.BYTE_ARRAY_OFFSET)] = value;
    +  }
    +
    +  public final short getShort(long offset) {
    +    assert(offset + 2 - Platform.BYTE_ARRAY_OFFSET <= array.length);
    +    return Platform.getShort(array, offset);
    +  }
    +
    +  public final void putShort(long offset, short value) {
    +    assert(offset + 2 - Platform.BYTE_ARRAY_OFFSET <= array.length);
    +    Platform.putShort(array, offset, value);
    +  }
    +
    +  public final long getLong(long offset) {
    +    // UTF8String.getPrefix() assumes data is 8-byte aligned
    +    assert(offset + 8 - Platform.BYTE_ARRAY_OFFSET <= ((array.length + 7) 
/ 8) * 8);
    +    return Platform.getLong(array, offset);
    +  }
    +
    +  public final void putLong(long offset, long value) {
    +    assert(offset + 8 - Platform.BYTE_ARRAY_OFFSET <= array.length);
    +    Platform.putLong(array, offset, value);
    +  }
    +
    +  public final float getFloat(long offset) {
    +    assert(offset + 4 - Platform.BYTE_ARRAY_OFFSET <= array.length);
    +    return Platform.getFloat(array, offset);
    +  }
    +
    +  public final void putFloat(long offset, float value) {
    +    assert(offset + 4 - Platform.BYTE_ARRAY_OFFSET <= array.length);
    +    Platform.putFloat(array, offset, value);
    +  }
    +
    +  public final double getDouble(long offset) {
    +    assert(offset + 8 - Platform.BYTE_ARRAY_OFFSET <= array.length);
    +    return Platform.getDouble(array, offset);
    +  }
    +
    +  public final void putDouble(long offset, double value) {
    +    assert(offset + 8 - Platform.BYTE_ARRAY_OFFSET <= array.length);
    +    Platform.putDouble(array, offset, value);
    +  }
    +
    +  public final void copyFrom(byte[] src, long srcOffset, long dstOffset, 
long length) {
    --- End diff --
    
    let me raise my question again. Since `MemoryBlock` has `baseObj` and 
`baseOffset`, We can implement `getInt`, `copyTo`, etc. in `MemoryBlock` as 
final method. However we put them in concrete implementations, for performance 
concerns. This is not ideal as the implementation code are very similar, so we 
should only do it if it really has perf benefit.
    
    I'd like to create a benchmark for `MemoryBlock` and keep the benchmark in 
the codebase, so it's easy to verify the perf gains.
    
    Particularly, I'm a little worried about the `copyFrom`. By the theory, 
`Platform.copyMemory` should run faster if both the `srcObj` and `dstObj` has 
concrete type. To make `srcObj` has concrete type, we create many varieties of 
`copyFrom`, like `copyFom(byte[] ...)`, `copyFrom(long[] ...)`, etc. To make 
`dstObj` has concrete type, we implement various `copyFrom` in all kinds of 
memory blocks. My question is, do both of them have perf benefit?


---

---------------------------------------------------------------------
To unsubscribe, e-mail: reviews-unsubscr...@spark.apache.org
For additional commands, e-mail: reviews-h...@spark.apache.org

Reply via email to