Tanner Gooding

> I'm not sure what if anything to do about it. I think this is an acceptable tradeoff. The measured difference is just over 1ns which would be about 2-4...

Also CC. @stephentoub This is something that both x86/x64 (`cmpxchg`) and Arm64 (`cas`, `casb`, and `cash`) support so I don't see anything that would block this. - I'm still of...

This is probably related to or the same as https://github.com/dotnet/runtime/issues/64375 which was fixed by @SingleAccretion in https://github.com/dotnet/runtime/pull/69992.

Is this one that should be backported to .NET 6? CC. @BruceForstall

The only changes here that I'm aware of are CSE related. @adamsitnik do you have the disassembly for 5.0 vs 6.0 on hand?

Right. There is a general consideration that most "native" compilers (and most shipping apps) target the "lowest common machine" for portability. For x86/x64 this is `x86-x64-v1` which is a machine...

> you might get wildly different perf characteristics based on the machine the code is deployed to. Yes, although in practice this isn't the case. All major cloud providers are...

> BenchmarkDotNet is tuned to measuring the best possible case - that's why it does all the warmup with JIT by default. With tiered PGO you'll get the best possible...

The purpose of benchmarks isn't to make ourselves look good. It is a tool meant to help users best productize their own apps and to track potential regressions Other languages...

Then the industry standard is wrong. It will not help the community achieve success for their actual deployments. Instead, it will directly lead them into actual pit of failures by...