Skip to content
/ AdventOfPerformance Public

C# performance optimization techniques via Advent of Code puzzles. Includes benchmarks across multiple .NET versions and Native AOT.

License

MIT license
2 stars 0 forks Branches Tags Activity
Star
Notifications You must be signed in to change notification settings

eNeRGy164/AdventOfPerformance

BranchesTags

Repository files navigation

A Season for Speed: Turning Puzzles into C# Performance Win

Welcome to the repository accompanying my conference talk "A Season for Speed: Turning Puzzles into C# Performance Win". This project showcases my journey of solving Advent of Code puzzles while learning and applying C# performance optimization techniques.

Presentation

Get the slides I used at previous presentations.

Getting started

Prerequisites

Ensure you have the following installed:

For most benchmarks:

  • .NET 8.0 SDK

For the cross-framework comparison benchmark:

  • .NET 8.0 SDK
  • .NET 7.0 SDK
  • .NET 6.0 SDK
  • .NET 5.0 SDK
  • .NET Framework 4.8.1 SDK

Cloning the repository

git clone https://github.com/eNeRGy164/AdventOfPerformance
cd AdventOfPerformance

Running the benchmarks

Native AOT

For running with the Native AOT, as I did for my talk, use:

dotnet run -c Release -f net8.0 --framework nativeaot8.0 --filter *

This takes about 16 minutes on an AMD Ryzen 5 7600.

Important

On processors with the AVX-512F capability, the current version of BenchmarkDotNet (0.12.3) may generate invalid code and not work correctly. This issue will be resolved in the next version of BenchmarkDotNet.

Without specifying a framework

To run all the benchmarks, use the following command:

dotnet run -c Release -f net8.0 --filter *

This takes about 16 minutes on an AMD Ryzen 5 7600.

Only the puzzle benchmarks

To run all the benchmarks, use the following command:

dotnet run -c Release -f net8.0 --framework nativeaot8.0 --filter AdventOfPerformance.Puzzle*

This takes about 16 minutes on an AMD Ryzen 5 7600.

Only the framework benchmarks

dotnet run -c Release -f net8.0 --filter AdventOfPerformance.Framework*

This takes about 1.5 minutes on an AMD Ryzen 5 7600.

Benchmark results

Puzzle part 1

PuzzlePart1Benchmarks.cs

BenchmarkDotNet v0.13.13-nightly.20240601.156, Windows 11 (10.0.22631.3737/23H2/2023Update/SunValley3)
AMD Ryzen 5 7600, 1 CPU, 12 logical and 6 physical cores
.NET SDK 8.0.302
  [Host]     : .NET 8.0.6 (8.0.624.26715), X64 RyuJIT AVX-512F+CD+BW+DQ+VL+VBMI
  Job-MUAWXP : .NET 8.0.6, X64 NativeAOT AVX-512F+CD+BW+DQ+VL+VBMI

Runtime=NativeAOT 8.0  Toolchain=Latest ILCompiler
Method Median Ratio Gen0 Gen1 Allocated Alloc Ratio
LinqQuery 528,551.5 ns 1.000 62.5000 - 1048328 B 1.000
LinqQueryIndex 420,219.8 ns 0.792 49.3164 - 830576 B 0.792
Foreach 185,389.9 ns 0.349 0.4883 - 10472 B 0.010
ForeachInnerRange 13,049.5 ns 0.025 5.1575 0.0153 86472 B 0.082
For 10,302.7 ns 0.019 0.0458 - 872 B 0.001
Pointers 6,196.6 ns 0.012 - - - 0.000
IntermediateVariables 6,055.8 ns 0.011 0.0458 - 872 B 0.001
ForGoto 5,920.1 ns 0.011 0.0458 - 872 B 0.001
Refs 5,567.1 ns 0.011 0.0458 - 824 B 0.001
InitializeArray 4,394.5 ns 0.008 0.0458 - 824 B 0.001
CustomIntParser 3,519.3 ns 0.007 0.0458 - 824 B 0.001
LowerMemory 3,494.3 ns 0.007 0.0229 - 424 B 0.000
TwoPointer 3,010.2 ns 0.006 0.0496 - 872 B 0.001
Hashing 1,923.7 ns 0.004 0.2327 - 3944 B 0.004
Bits 245.4 ns 0.000 0.0186 - 312 B 0.000

Puzzle part 2

PuzzlePart2Benchmarks.cs

BenchmarkDotNet v0.13.13-nightly.20240601.156, Windows 11 (10.0.22631.3737/23H2/2023Update/SunValley3)
AMD Ryzen 5 7600, 1 CPU, 12 logical and 6 physical cores
.NET SDK 8.0.302
  [Host]     : .NET 8.0.6 (8.0.624.26715), X64 RyuJIT AVX-512F+CD+BW+DQ+VL+VBMI
  Job-MUAWXP : .NET 8.0.6, X64 NativeAOT AVX-512F+CD+BW+DQ+VL+VBMI

Runtime=NativeAOT 8.0  Toolchain=Latest ILCompiler
Method Median Ratio Gen0 Allocated Alloc Ratio
LinqQuery 18,929.987 μs 1.000 2156.2500 36410868 B 1.000
CustomIntParser 63.120 μs 0.003 - 824 B 0.000
TwoPointer 3.020 μs 0.000 0.0496 872 B 0.000

Compare frameworks

FrameworksTestBenchmark.cs

BenchmarkDotNet v0.13.13-nightly.20240601.156, Windows 11 (10.0.22631.3737/23H2/2023Update/SunValley3)
AMD Ryzen 5 7600, 1 CPU, 12 logical and 6 physical cores
.NET SDK 8.0.302
  [Host]               : .NET 8.0.6 (8.0.624.26715), X64 RyuJIT AVX-512F+CD+BW+DQ+VL+VBMI
  .NET Framework 4.8.1 : .NET Framework 4.8.1 (4.8.9241.0), X64 RyuJIT VectorSize=256
  .NET 5.0             : .NET 5.0.17 (5.0.1722.21314), X64 RyuJIT AVX2
  .NET 7.0             : .NET 7.0.20 (7.0.2024.26716), X64 RyuJIT AVX2
  .NET 6.0             : .NET 6.0.31 (6.0.3124.26714), X64 RyuJIT AVX2
  .NET 8.0             : .NET 8.0.6 (8.0.624.26715), X64 RyuJIT AVX-512F+CD+BW+DQ+VL+VBMI
Method Job Runtime Median Gen0 Allocated
LinqQuery .NET Framework 4.8.1 .NET Framework 4.8.1 994.5 μs 166.0156 1026.81 KB
LinqQuery .NET 5.0 .NET 5.0 491.0 μs 62.5000 1023.76 KB
LinqQuery .NET 7.0 .NET 7.0 440.3 μs 62.5000 1023.76 KB
LinqQuery .NET 6.0 .NET 6.0 437.1 μs 62.5000 1023.76 KB
LinqQuery .NET 8.0 .NET 8.0 294.1 μs 62.5000 1023.76 KB

Walking a 2D Array

WalkThroughMultiDimArraysBenchmarks.cs

BenchmarkDotNet v0.13.13-nightly.20240601.156, Windows 11 (10.0.22631.3737/23H2/2023Update/SunValley3)
AMD Ryzen 5 7600, 1 CPU, 12 logical and 6 physical cores
.NET SDK 8.0.302
  [Host]     : .NET 8.0.6 (8.0.624.26715), X64 RyuJIT AVX-512F+CD+BW+DQ+VL+VBMI
  Job-MUAWXP : .NET 8.0.6, X64 NativeAOT AVX-512F+CD+BW+DQ+VL+VBMI

Runtime=NativeAOT 8.0  Toolchain=Latest ILCompiler
Method Median Allocated
'Walk through 2D Array, column by column' 233.8 μs -
'Walk through 2D Array, row by row' 138.7 μs -
'Use enumerator of the 2D Array' 122.9 μs -

Writing a 2D Array

SerializeMultiDimArraysBenchmarks.cs

BenchmarkDotNet v0.13.13-nightly.20240601.156, Windows 11 (10.0.22631.3737/23H2/2023Update/SunValley3)
AMD Ryzen 5 7600, 1 CPU, 12 logical and 6 physical cores
.NET SDK 8.0.302
  [Host]     : .NET 8.0.6 (8.0.624.26715), X64 RyuJIT AVX-512F+CD+BW+DQ+VL+VBMI
  Job-MUAWXP : .NET 8.0.6, X64 NativeAOT AVX-512F+CD+BW+DQ+VL+VBMI

Runtime=NativeAOT 8.0  Toolchain=Latest ILCompiler
Method Median Gen0 Gen1 Gen2 Allocated
WriteCharYX 636.8 μs 166.0156 166.0156 166.0156 1031.03 KB
WriteCharForeach 584.4 μs 166.0156 166.0156 166.0156 1031.03 KB
WriteCharStringCreate 403.1 μs 166.5039 166.5039 166.5039 512.08 KB
WriteCharSpan 195.2 μs 166.5039 166.5039 166.5039 512.08 KB

Parsing input

InputParseBenchmarks.cs

BenchmarkDotNet v0.13.13-nightly.20240601.156, Windows 11 (10.0.22631.3737/23H2/2023Update/SunValley3)
AMD Ryzen 5 7600, 1 CPU, 12 logical and 6 physical cores
.NET SDK 8.0.302
  [Host]     : .NET 8.0.6 (8.0.624.26715), X64 RyuJIT AVX-512F+CD+BW+DQ+VL+VBMI
  Job-MUAWXP : .NET 8.0.6, X64 NativeAOT AVX-512F+CD+BW+DQ+VL+VBMI

Runtime=NativeAOT 8.0  Toolchain=Latest ILCompiler
Method Median Ratio Gen0 Gen1 Allocated Alloc Ratio
Regex 484.23 μs 1.00 103.0273 12.6953 1683.3 KB 1.00
RegexGenerated 343.51 μs 0.71 103.0273 12.6953 1683.3 KB 1.00
CharSplit 236.39 μs 0.49 51.2695 6.3477 840.38 KB 0.50
CharsSplit 137.35 μs 0.28 34.1797 4.1504 560.55 KB 0.33
Spans 32.04 μs 0.07 2.1362 - 35.11 KB 0.02
SpansAnalyzeInput 10.69 μs 0.02 2.1362 - 35.11 KB 0.02

Creating objects and storing in a HashSet

CreateObjectsInHashSetBenchmark.cs

BenchmarkDotNet v0.13.13-nightly.20240601.156, Windows 11 (10.0.22631.3737/23H2/2023Update/SunValley3)
AMD Ryzen 5 7600, 1 CPU, 12 logical and 6 physical cores
.NET SDK 8.0.302
  [Host]     : .NET 8.0.6 (8.0.624.26715), X64 RyuJIT AVX-512F+CD+BW+DQ+VL+VBMI
  Job-MUAWXP : .NET 8.0.6, X64 NativeAOT AVX-512F+CD+BW+DQ+VL+VBMI

Runtime=NativeAOT 8.0  Toolchain=Latest ILCompiler
Method Median Ratio Gen0 Gen1 Gen2 Allocated Alloc Ratio
'Class w/ properties assigned in ctor' 139.1 μs 1.00 49.8047 49.8047 49.8047 314.64 KB 1.00
'Struct w/ readonly fields assigned in ctor' 136.5 μs 0.98 49.8047 49.8047 49.8047 314.92 KB 1.00
'Readonly Struct w/ properties assigned in ctor' 135.6 μs 0.97 49.8047 49.8047 49.8047 314.92 KB 1.00
'Struct w/ properties assigned in the ctor' 135.1 μs 0.97 49.8047 49.8047 49.8047 314.92 KB 1.00
'Sealed Class w/ properties assigned in ctor' 134.5 μs 0.96 49.8047 49.8047 49.8047 314.64 KB 1.00
'Class w/ fields assigned in the ctor' 133.6 μs 0.96 49.8047 49.8047 49.8047 314.64 KB 1.00
'Readonly Struct w/ '[in]' properties assigned in ctor' 133.7 μs 0.96 49.8047 49.8047 49.8047 314.92 KB 1.00
'Record Class w/ properties assigned via primary ctor' 127.3 μs 0.91 49.8047 49.8047 49.8047 314.64 KB 1.00
'Sealed Record Class w/ properties assigned via the primary ctor' 127.0 μs 0.91 49.8047 49.8047 49.8047 314.64 KB 1.00
'Readonly Record Class w/ '[in]' properties assigned via primary ctor' 104.3 μs 0.75 49.9268 49.9268 49.9268 197.45 KB 0.63
'Readonly Record Struct w/ properties assigned via primary ctor' 103.6 μs 0.74 49.9268 49.9268 49.9268 197.45 KB 0.63
'Record Struct w/ properties assigned via primary ctor' 103.6 μs 0.74 49.9268 49.9268 49.9268 197.45 KB 0.63

Contributing

Feel free to submit issues or pull requests if you have suggestions or improvements.

License

This project is licensed under the MIT License - see the LICENSE file for details.

About

C# performance optimization techniques via Advent of Code puzzles. Includes benchmarks across multiple .NET versions and Native AOT.

Topics

benchmark performance

Resources

Readme

License

MIT license
Activity

Stars

2 stars

Watchers

1 watching

Forks

0 forks
Report repository

Contributors 2

  •  
  •  

Languages