Add DOTNET_JitDisasmOnlyOptimized knob

[EgorBo]

We already have DOTNET_JitDumpTier0 to hide unoptimized dumps, this PR add the same for JitDisasm. Users often don't care about unoptimized asm.

[ghost]

Tagging subscribers to this area: @JulieLeeMSFT, @jakobbotsch
See info in area-owners.md if you want to be subscribed.

Issue Details

---

We already have DOTNET_JitDumpTier0 to hide unoptimized dumps, this PR add the same for JitDisasm. Users often don't care about unoptimized asm.

Author:	EgorBo
Assignees:	EgorBo
Labels:	area-CodeGen-coreclr
Milestone:	-

[jakobbotsch]

The name is DisasmTier0 but the behavior is DisasmOnlyTier1. I'd suggest renaming it to JitDisasmOnlyOptimized and switch the check to OptimizationEnabled() instead.

[EgorBo]

The name is DisasmTier0 but the behavior is DisasmOnlyTier1. I'd suggest renaming it to DisasmOnlyOptimized and switch the check to OptimizationEnabled() instead.

OptimizationEnabled() triggered an assert because it's not yet initialized so I consumed the flag directly to avoid changing too much code unnecessarily. Tier1 flag means "optimizations are allowed", I'm fine renaming the knob though

[jakobbotsch]

I don't think the check is correct as it stands. It does not cover scenarios where the JIT switches from unoptimized code to optimized code or vice versa. Also, do crossgen2/ILC even set the Tier1 flag?

I think you should just delay checking for these conditions until we've made all the decisions around optimization level, including switching between optimized/deoptimizing, and switch to OptimizationEnabled(). We can do that for JitDisasm, but we cannot really do it for JitDump.

[EgorBo]

Failure is #96904

[BruceForstall]

Don't forget to cc jit-contrib on PRs.

[EgorBo]

Don't forget to cc jit-contrib on PRs.

Yep, I forgot because Jakob approved it before I requested for review to anyone 🙂