Improve .NET 8 stream decompression perf

Motivation ---------- .NET 8 has shown a significant regression when decompressing streams, though not blocks. Modifications ------------- Drop some overly aggressive inlining on cold paths, which may provide better enregistration. Also, for .NET 8 use an inline array to avoid allocating the scratch buffer on the heap. Results ------- The regression has been effectively eliminated. BenchmarkDotNet v0.13.10, Windows 11 (10.0.22621.2861/22H2/2022Update/SunValley2) 12th Gen Intel Core i7-1270P, 1 CPU, 16 logical and 12 physical cores .NET SDK 8.0.100 [Host] : .NET 8.0.0 (8.0.23.53103), X64 RyuJIT AVX2 Job-EHUKKX : .NET Framework 4.8.1 (4.8.9181.0), X64 RyuJIT VectorSize=256 Job-SIHOZS : .NET Framework 4.8.1 (4.8.9181.0), X64 RyuJIT VectorSize=256 Job-VAGBBP : .NET 6.0.25 (6.0.2523.51912), X64 RyuJIT AVX2 Job-ZACCID : .NET 6.0.25 (6.0.2523.51912), X64 RyuJIT AVX2 Job-ACXPKE : .NET 8.0.0 (8.0.23.53103), X64 RyuJIT AVX2 Job-JFCQYT : .NET 8.0.0 (8.0.23.53103), X64 RyuJIT AVX2 Job-CWWXAC : .NET 8.0.0 (8.0.23.53103), X64 RyuJIT AVX2 Job-IDGNPX : .NET 8.0.0 (8.0.23.53103), X64 RyuJIT AVX2 | Method | Runtime | BuildConfiguration | PGO | Mean | Error | StdDev | Ratio | Rank | |----------- |------------------- |------------------- |---- |---------:|--------:|--------:|------:|-----:| | Decompress | .NET Framework 4.8 | Previous | N | 342.7 us | 1.73 us | 1.45 us | 1.00 | 2 | | Decompress | .NET Framework 4.8 | Default | N | 334.7 us | 0.85 us | 0.80 us | 0.98 | 1 | | | | | | | | | | | | Decompress | .NET 6.0 | Previous | N | 168.5 us | 1.43 us | 1.34 us | 1.00 | 2 | | Decompress | .NET 6.0 | Default | N | 164.8 us | 0.78 us | 0.73 us | 0.98 | 1 | | | | | | | | | | | | Decompress | .NET 8.0 | Previous | N | 199.9 us | 1.64 us | 1.53 us | 1.00 | 2 | | Decompress | .NET 8.0 | Default | N | 160.7 us | 1.42 us | 1.33 us | 0.80 | 1 | | | | | | | | | | | | Decompress | .NET 8.0 | Previous | Y | 191.3 us | 1.05 us | 0.98 us | 1.00 | 2 | | Decompress | .NET 8.0 | Default | Y | 160.9 us | 1.50 us | 1.40 us | 0.84 | 1 | Note: Benchmarks include other improvements since 1.1.3
brantburnett · Jan 6, 2024 · f9da0f1 · f9da0f1
1 parent 9d07b42
commit f9da0f1
Showing 1 changed file with 25 additions and 10 deletions.
diff --git a/Snappier/Internal/SnappyStreamDecompressor.cs b/Snappier/Internal/SnappyStreamDecompressor.cs
@@ -1,8 +1,6 @@
 using System;
 using System.Buffers.Binary;
 using System.Diagnostics;
-using System.IO;
-using System.Runtime.CompilerServices;
 
 namespace Snappier.Internal
 {
@@ -13,11 +11,30 @@ internal sealed class SnappyStreamDecompressor : IDisposable
     {
         private const int ScratchBufferSize = 4;
 
+#if NET8_0_OR_GREATER
+#pragma warning disable IDE0051
+#pragma warning disable IDE0044
+#pragma warning disable CS0649 // Field is never assigned to, and will always have its default value
+        [System.Runtime.CompilerServices.InlineArray(ScratchBufferSize)]
+        private struct ScratchBuffer
+        {
+            private byte _element0;
+        }
+
+        private ScratchBuffer _scratch;
+#pragma warning restore CS0649 // Field is never assigned to, and will always have its default value
+#pragma warning restore IDE0044
+#pragma warning restore IDE0051
+#else
+        private readonly byte[] _scratch = new byte[ScratchBufferSize];
+#endif
+
+        private Span<byte> Scratch => _scratch;
+
         private SnappyDecompressor? _decompressor = new();
 
         private ReadOnlyMemory<byte> _input;
 
-        private readonly byte[] _scratch = new byte[ScratchBufferSize];
         private int _scratchLength;
         private Constants.ChunkType _chunkType = Constants.ChunkType.Null;
         private int _chunkSize;
@@ -196,14 +213,13 @@ public void SetInput(ReadOnlyMemory<byte> input)
             _input = input;
         }
 
-        [MethodImpl(MethodImplOptions.AggressiveInlining)]
         private uint ReadChunkHeader(ref ReadOnlySpan<byte> buffer)
         {
             if (_scratchLength > 0)
             {
                 int bytesToCopyToScratch = 4 - _scratchLength;
 
-                Span<byte> scratch = _scratch.AsSpan();
+                Span<byte> scratch = Scratch;
                 buffer.Slice(0, bytesToCopyToScratch).CopyTo(scratch.Slice(_scratchLength));
 
                 buffer = buffer.Slice(bytesToCopyToScratch);
@@ -223,7 +239,7 @@ private uint ReadChunkHeader(ref ReadOnlySpan<byte> buffer)
             {
                 // Insufficient data
 
-                buffer.CopyTo(_scratch);
+                buffer.CopyTo(Scratch);
 
                 _scratchLength = buffer.Length;
                 buffer = default;
@@ -240,9 +256,8 @@ private uint ReadChunkHeader(ref ReadOnlySpan<byte> buffer)
 
         /// <summary>
         /// Assuming that we're at the beginning of a chunk, reads the CRC. If partially read, stores the value in
-        /// _scratch for subsequent reads. Should not be called if chunkByteProcessed >= 4.
+        /// Scratch for subsequent reads. Should not be called if chunkByteProcessed >= 4.
         /// </summary>
-        [MethodImpl(MethodImplOptions.AggressiveInlining)]
         private bool ReadChunkCrc(ref ReadOnlySpan<byte> input)
         {
             Debug.Assert(_chunkBytesProcessed < 4);
@@ -259,14 +274,14 @@ private bool ReadChunkCrc(ref ReadOnlySpan<byte> input)
 
             // Copy to scratch
             int crcBytesAvailable = Math.Min(input.Length, 4 - _chunkBytesProcessed);
-            input.Slice(0, crcBytesAvailable).CopyTo(_scratch.AsSpan(_scratchLength));
+            input.Slice(0, crcBytesAvailable).CopyTo(Scratch.Slice(_scratchLength));
             _scratchLength += crcBytesAvailable;
             input = input.Slice(crcBytesAvailable);
             _chunkBytesProcessed += crcBytesAvailable;
 
             if (_scratchLength >= 4)
             {
-                _expectedChunkCrc = BinaryPrimitives.ReadUInt32LittleEndian(_scratch);
+                _expectedChunkCrc = BinaryPrimitives.ReadUInt32LittleEndian(Scratch);
                 _scratchLength = 0;
                 return true;
             }