Remainder operation (%) performance in Linux vs Windows and macOS
#90667
Replies: 3 comments 3 replies
-
Beta Was this translation helpful? Give feedback.
0 replies
-
|
Can you show disasm for both? via |
Beta Was this translation helpful? Give feedback.
3 replies
-
|
Unfortunately GH Actions Linux one failed due to: and GH Actions macOS faced something similar to dotnet/BenchmarkDotNet#1499: Windows one can be found in https://github.com/lynx-chess/Lynx/actions/runs/5882072009 Locally in Ubuntu inside of WSL it's also mysteriously failing after adding the attribute $ sudo dotnet run -c Release
Available Benchmarks:
#0 ArrayCopy
#1 BitBoard_Struct_ReadonlyStruct_Class_Record
#2 Branching
#3 CountBits
#4 DivideByHalf
#5 EncodeMove
#6 EnumCasting
#7 FENGeneration
#8 GetLS1BIndex
#9 GoCommandParsing
#10 GoCommandParsing_GeneratedAttribute
#11 InitializeBishopAndRookAttacks
#12 IsSquareAttacked_local_vs_external
#13 IsSquareAttacked_parallel
#14 LocalVariableIn_vs_NoIn
#15 MakeUnmakeMove_implementation
#16 MakeUnmakeMove_integration
#17 MoveGeneratorArrayVsDictionary
#18 MoveGeneratorParallel
#19 MutableReadonlyBitBoard
#20 OppositeSide
#21 PawnAttacks
#22 PieceOffset_Boolean
#23 PieceOffset_Side
#24 PositionClone
#25 PositionIdGeneration
#26 PriorityQueue_Dequeue
#27 PriorityQueue_EnqueueRange
#28 PVTable_SumVsArrayAccess
#29 ResetLS1B
#30 ResetLS1BvsWithoutLS1B
#31 ToggleBits
#32 TranspositionTableIndex
You should select the target benchmark(s). Please, print a number of a benchmark (e.g. `0`) or a contained benchmark caption (e.g. `ArrayCopy`).
If you want to select few, please separate them with space ` ` (e.g. `1 2 3`).
You can also provide the class name in console arguments by using --filter. (e.g. `--filter '*ArrayCopy*'`).
Enter the asterisk `*` to select all.
32
// Validating benchmarks:
// ***** BenchmarkRunner: Start *****
// ***** Found 2 benchmark(s) in total *****
// ***** Building 1 exe(s) in Parallel: Start *****
// start dotnet restore /p:UseSharedCompilation=false /p:BuildInParallel=false /m:1 /p:Deterministic=true /p:Optimize=true in /home/edu/dev/tmp/Lynx/src/Lynx.Benchmark/bin/Release/net7.0/8562b748-4afb-4e78-af89-3ae3c1ce4253
// command took 2.23 sec and exited with 0
// start dotnet build -c Release --no-restore /p:UseSharedCompilation=false /p:BuildInParallel=false /m:1 /p:Deterministic=true /p:Optimize=true in /home/edu/dev/tmp/Lynx/src/Lynx.Benchmark/bin/Release/net7.0/8562b748-4afb-4e78-af89-3ae3c1ce4253
// command took 5.36 sec and exited with 0
// ***** Done, took 00:00:07 (7.72 sec) *****
// Found 2 benchmarks:
// TranspositionTableIndex.Modulo: DefaultJob [size=2097152]
// TranspositionTableIndex.Modulo: DefaultJob [size=2097153]
// **************************
// Benchmark: TranspositionTableIndex.Modulo: DefaultJob [size=2097152]
// *** Execute ***
// Launch: 1 / 1
// Execute: dotnet 8562b748-4afb-4e78-af89-3ae3c1ce4253.dll --anonymousPipes 122 123 --benchmarkName "Lynx.Benchmark.TranspositionTableIndex.Modulo(size: 2097152)" --job Default --benchmarkId 0 in /home/edu/dev/tmp/Lynx/src/Lynx.Benchmark/bin/Release/net7.0/8562b748-4afb-4e78-af89-3ae3c1ce4253/bin/Release/net7.0
// BeforeAnythingElse
// Benchmark Process Environment Information:
// BenchmarkDotNet v0.13.7
// Runtime=.NET 7.0.9 (7.0.923.32018), X64 RyuJIT AVX2
// GC=Concurrent Workstation
// HardwareIntrinsics=AVX2,AES,BMI1,BMI2,FMA,LZCNT,PCLMUL,POPCNT VectorSize=256
// Job: DefaultJob
OverheadJitting 1: 1 op, 588800.00 ns, 588.8000 us/op
WorkloadJitting 1: 1 op, 721400.00 ns, 721.4000 us/op
OverheadJitting 2: 16 op, 1038300.00 ns, 64.8937 us/op
WorkloadJitting 2: 16 op, 2404300.00 ns, 150.2688 us/op
WorkloadPilot 1: 16 op, 1554500.00 ns, 97.1562 us/op
WorkloadPilot 2: 32 op, 3867500.00 ns, 120.8594 us/op
WorkloadPilot 3: 64 op, 6136800.00 ns, 95.8875 us/op
WorkloadPilot 4: 128 op, 18774600.00 ns, 146.6766 us/op
WorkloadPilot 5: 256 op, 27941800.00 ns, 109.1477 us/op
WorkloadPilot 6: 512 op, 53809700.00 ns, 105.0971 us/op
WorkloadPilot 7: 1024 op, 103322100.00 ns, 100.9005 us/op
WorkloadPilot 8: 2048 op, 226087300.00 ns, 110.3942 us/op
WorkloadPilot 9: 4096 op, 403528500.00 ns, 98.5177 us/op
WorkloadPilot 10: 8192 op, 786790600.00 ns, 96.0438 us/op
OverheadWarmup 1: 8192 op, 34700.00 ns, 4.2358 ns/op
OverheadWarmup 2: 8192 op, 45700.00 ns, 5.5786 ns/op
OverheadWarmup 3: 8192 op, 46600.00 ns, 5.6885 ns/op
OverheadWarmup 4: 8192 op, 44900.00 ns, 5.4810 ns/op
OverheadWarmup 5: 8192 op, 41600.00 ns, 5.0781 ns/op
OverheadWarmup 6: 8192 op, 40400.00 ns, 4.9316 ns/op
OverheadWarmup 7: 8192 op, 40600.00 ns, 4.9561 ns/op
OverheadWarmup 8: 8192 op, 45100.00 ns, 5.5054 ns/op
OverheadWarmup 9: 8192 op, 77100.00 ns, 9.4116 ns/op
OverheadWarmup 10: 8192 op, 46500.00 ns, 5.6763 ns/op
OverheadActual 1: 8192 op, 46200.00 ns, 5.6396 ns/op
OverheadActual 2: 8192 op, 40400.00 ns, 4.9316 ns/op
OverheadActual 3: 8192 op, 32200.00 ns, 3.9307 ns/op
OverheadActual 4: 8192 op, 41000.00 ns, 5.0049 ns/op
OverheadActual 5: 8192 op, 40700.00 ns, 4.9683 ns/op
OverheadActual 6: 8192 op, 45200.00 ns, 5.5176 ns/op
OverheadActual 7: 8192 op, 46400.00 ns, 5.6641 ns/op
OverheadActual 8: 8192 op, 32200.00 ns, 3.9307 ns/op
OverheadActual 9: 8192 op, 31900.00 ns, 3.8940 ns/op
OverheadActual 10: 8192 op, 31800.00 ns, 3.8818 ns/op
OverheadActual 11: 8192 op, 31900.00 ns, 3.8940 ns/op
OverheadActual 12: 8192 op, 32000.00 ns, 3.9062 ns/op
OverheadActual 13: 8192 op, 44100.00 ns, 5.3833 ns/op
OverheadActual 14: 8192 op, 31900.00 ns, 3.8940 ns/op
OverheadActual 15: 8192 op, 34100.00 ns, 4.1626 ns/op
OverheadActual 16: 8192 op, 32100.00 ns, 3.9185 ns/op
OverheadActual 17: 8192 op, 33600.00 ns, 4.1016 ns/op
OverheadActual 18: 8192 op, 41300.00 ns, 5.0415 ns/op
OverheadActual 19: 8192 op, 40300.00 ns, 4.9194 ns/op
OverheadActual 20: 8192 op, 32200.00 ns, 3.9307 ns/op
WorkloadWarmup 1: 8192 op, 789246700.00 ns, 96.3436 us/op
WorkloadWarmup 2: 8192 op, 798522000.00 ns, 97.4758 us/op
WorkloadWarmup 3: 8192 op, 792974900.00 ns, 96.7987 us/op
WorkloadWarmup 4: 8192 op, 787944800.00 ns, 96.1847 us/op
WorkloadWarmup 5: 8192 op, 858571800.00 ns, 104.8061 us/op
WorkloadWarmup 6: 8192 op, 792109600.00 ns, 96.6931 us/op
// BeforeActualRun
WorkloadActual 1: 8192 op, 776998500.00 ns, 94.8484 us/op
WorkloadActual 2: 8192 op, 790785400.00 ns, 96.5314 us/op
WorkloadActual 3: 8192 op, 794547900.00 ns, 96.9907 us/op
WorkloadActual 4: 8192 op, 772093200.00 ns, 94.2497 us/op
WorkloadActual 5: 8192 op, 796677800.00 ns, 97.2507 us/op
WorkloadActual 6: 8192 op, 794390100.00 ns, 96.9714 us/op
WorkloadActual 7: 8192 op, 776298400.00 ns, 94.7630 us/op
WorkloadActual 8: 8192 op, 799839000.00 ns, 97.6366 us/op
WorkloadActual 9: 8192 op, 821026900.00 ns, 100.2230 us/op
WorkloadActual 10: 8192 op, 797089600.00 ns, 97.3010 us/op
WorkloadActual 11: 8192 op, 777251900.00 ns, 94.8794 us/op
WorkloadActual 12: 8192 op, 797092800.00 ns, 97.3014 us/op
WorkloadActual 13: 8192 op, 796822300.00 ns, 97.2683 us/op
WorkloadActual 14: 8192 op, 794410600.00 ns, 96.9740 us/op
WorkloadActual 15: 8192 op, 822079900.00 ns, 100.3516 us/op
// AfterActualRun
WorkloadResult 1: 8192 op, 776964650.00 ns, 94.8443 us/op
WorkloadResult 2: 8192 op, 790751550.00 ns, 96.5273 us/op
WorkloadResult 3: 8192 op, 794514050.00 ns, 96.9866 us/op
WorkloadResult 4: 8192 op, 772059350.00 ns, 94.2455 us/op
WorkloadResult 5: 8192 op, 796643950.00 ns, 97.2466 us/op
WorkloadResult 6: 8192 op, 794356250.00 ns, 96.9673 us/op
WorkloadResult 7: 8192 op, 776264550.00 ns, 94.7589 us/op
WorkloadResult 8: 8192 op, 799805150.00 ns, 97.6325 us/op
WorkloadResult 9: 8192 op, 797055750.00 ns, 97.2968 us/op
WorkloadResult 10: 8192 op, 777218050.00 ns, 94.8753 us/op
WorkloadResult 11: 8192 op, 797058950.00 ns, 97.2972 us/op
WorkloadResult 12: 8192 op, 796788450.00 ns, 97.2642 us/op
WorkloadResult 13: 8192 op, 794376750.00 ns, 96.9698 us/op
// GC: 0 0 0 936 8192
// Threading: 0 0 8192
// AfterAll
If case anyone gets time/interest, it's branch dotnet/90667 |
Beta Was this translation helpful? Give feedback.
-
You need to run the benchmark with |
Beta Was this translation helpful? Give feedback.
-
Good point, I assumed it wasn't possible in GH action runners/containers, but it is. |
Beta Was this translation helpful? Give feedback.
-
|
I was hinted this could 'just' be because of the different processors used on GH actions runners: And I just verified that's the case: Ubuntu run in Intel Xeon CPU E5-2673 v4 2.30GHz, with similar results than same benchmarks in Windows/Ubuntu and equivalent processorBenchmarkDotNet v0.13.7, Ubuntu 22.04.3 LTS (Jammy Jellyfish) Intel Xeon CPU E5-2673 v4 2.30GHz, 1 CPU, 2 logical and 2 physical cores .NET SDK 8.0.100-preview.7.23376.3 [Host] : .NET 8.0.0 (8.0.23.37506), X64 RyuJIT AVX2 DefaultJob : .NET 8.0.0 (8.0.23.37506), X64 RyuJIT AVX2
|
Beta Was this translation helpful? Give feedback.
0 replies
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Uh oh!
There was an error while loading. Please reload this page.
Uh oh!
There was an error while loading. Please reload this page.
-
While running some unrelated benchmarks (to compare
x % yandx & (y-1)when y is a power of two), I found some surprising results in the%part:Remainder operations (
%) in Ubuntu seems to be twice faster than the same operations in Windows or macOS, at least for the test cases I tried.Am I approaching this wrong? Is there a documented reason for this to happen?
And more importantly, is there room for improvement?
Edit: I've been hinted this could be because of the different processors used on GH actions runners:
Intel Xeon CPU E5-2673 v4 2.30GHzandIntel Xeon CPU E5-1650 v2 3.50GHzin Windows and mac vsIntel Xeon Platinum 8370C CPU 2.80GHzin Linux.Benchmark code
Benchmark results (using GH Actions)
Results for (2,4,8,16,32,64,128,256,512,1024) * 1024 * 1024 (power of two)
https://github.com/lynx-chess/Lynx/actions/runs/5879633929
Results for (3,5,9,17,33,65,129,257,513,1025) * 1024 * 1024
https://github.com/lynx-chess/Lynx/actions/runs/5879866516
Beta Was this translation helpful? Give feedback.
All reactions