Benchmarks for dotnet/runtime#127970 (for @EgorBo)

[EgorBot]

Processing benchmark request from PR dotnet/runtime#127970.

Targets: ubuntu24_azure_turin, macos26_helix_arm64
Commits: main;PR_127970

• ubuntu24_azure_turin: live logs
• macos26_helix_arm64: live logs

Results will be posted here once they are ready.

[EgorBot]

Results for macos26_helix_arm64

@EgorBo

BenchmarkDotNet v0.15.8, macOS Sequoia 15.4.1 (24E263) [Darwin 24.4.0]
Apple M2, 1 CPU, 8 logical and 8 physical cores
.NET SDK 11.0.100-preview.5.26257.113
  [Host]     : .NET 10.0.7 (10.0.7, 10.0.726.21808), Arm64 RyuJIT armv8.0-a

Method	Toolchain	Len	Mean	Error	Ratio
Test	PR #127970**	1	1.933 ns	0.0035 ns	1.00
Test	main	1	1.695 ns	0.0320 ns	0.88
Test	PR #127970**	4	1.914 ns	0.0021 ns	1.00
Test	main	4	1.712 ns	0.0128 ns	0.89
Test	PR #127970**	6	1.913 ns	0.0029 ns	1.00
Test	main	6	1.738 ns	0.0128 ns	0.91
Test	PR #127970**	8	1.914 ns	0.0090 ns	1.00
Test	main	8	1.691 ns	0.0128 ns	0.88
Test	PR #127970**	10	1.911 ns	0.0009 ns	1.00
Test	main	10	1.736 ns	0.0233 ns	0.91
Test	PR #127970**	16	1.898 ns	0.0019 ns	1.00
Test	main	16	1.723 ns	0.0340 ns	0.91
Test	PR #127970**	32	3.133 ns	0.0115 ns	1.00
Test	main	32	1.577 ns	0.0019 ns	0.50
Test	PR #127970**	50	1.302 ns	0.0250 ns	1.00
Test	main	50	3.238 ns	0.0154 ns	2.49
Test	PR #127970**	100	4.714 ns	0.0649 ns	1.00
Test	main	100	3.873 ns	0.0064 ns	0.82
Test	PR #127970**	128	4.171 ns	0.1035 ns	1.00
Test	main	128	4.400 ns	0.0812 ns	1.06
Test	PR #127970**	256	6.662 ns	0.1534 ns	1.00
Test	main	256	6.448 ns	0.0251 ns	0.97
Test	PR #127970**	500	7.391 ns	0.0167 ns	1.00
Test	main	500	11.100 ns	0.0376 ns	1.50

Full logs

[EgorBot]

Results for ubuntu24_azure_turin

@EgorBo

BenchmarkDotNet v0.15.8, Linux Ubuntu 24.04.4 LTS (Noble Numbat)
AMD EPYC 9V45 2.60GHz, 1 CPU, 8 logical and 4 physical cores
.NET SDK 11.0.100-preview.5.26257.113
  [Host]     : .NET 10.0.7 (10.0.7, 10.0.726.21808), X64 RyuJIT x86-64-v4

Method	Toolchain	Len	Mean	Error	Ratio
Test	PR #127970**	1	1.345 ns	0.0145 ns	1.00
Test	main	1	1.146 ns	0.0273 ns	0.85
Test	PR #127970**	4	1.337 ns	0.0193 ns	1.00
Test	main	4	1.137 ns	0.0160 ns	0.85
Test	PR #127970**	6	1.351 ns	0.0301 ns	1.00
Test	main	6	1.134 ns	0.0250 ns	0.84
Test	PR #127970**	8	1.361 ns	0.0232 ns	1.00
Test	main	8	1.129 ns	0.0209 ns	0.83
Test	PR #127970**	10	1.362 ns	0.0145 ns	1.00
Test	main	10	1.128 ns	0.0055 ns	0.83
Test	PR #127970**	16	1.317 ns	0.0148 ns	1.00
Test	main	16	1.131 ns	0.0149 ns	0.86
Test	PR #127970**	32	1.528 ns	0.0513 ns	1.00
Test	main	32	1.364 ns	0.0175 ns	0.89
Test	PR #127970**	50	1.369 ns	0.0195 ns	1.00
Test	main	50	1.792 ns	0.0184 ns	1.31
Test	PR #127970**	100	1.348 ns	0.0130 ns	1.00
Test	main	100	2.015 ns	0.0223 ns	1.50
Test	PR #127970**	128	1.372 ns	0.0210 ns	1.00
Test	main	128	2.232 ns	0.0229 ns	1.63
Test	PR #127970**	256	1.427 ns	0.0465 ns	1.00
Test	main	256	3.158 ns	0.0394 ns	2.21
Test	PR #127970**	500	1.360 ns	0.0175 ns	1.00
Test	main	500	6.484 ns	0.0691 ns	4.77

Full logs

[EgorBo]

ugh...
BenchmarkDotNet used the same method for different lengths because of [Params] which is the worst case scenario for PGO...

Although, the impact of misprediction isn't too bad which is a good thing.