[PATCH] D152399: [CodeGen] Fine tune MachineFunctionSplitPass (MFS) for FSAFDO.

Thu Jul 13 14:29:52 PDT 2023

shenhan added a comment.

In D152399#4464394 <https://reviews.llvm.org/D152399#4464394>, @snehasish wrote:

> lgtm, thanks for your patience with all the suggestions!

In D152399#4498827 <https://reviews.llvm.org/D152399#4498827>, @xur wrote:

> It seems that we now need to specify "-mllvm -mfs-psi-cutoff=0" to enable the MFS for FSAFDO, if we want to get the same optimization behavior.  This is not good for usage adoption. 
> I think if the users want to enable MFS in FSAFDO, they should only need to add -fsplit-machine-functions. They should not need to use ANY extra internal options.
>
> I know the newer change was to unify the heuristic code in isColdBlock. But I don't like the change here.

Thanks. Yes, currently to use FSAFDO MFS, we need these flags "-fsplit-machine-functions -mllvm -mfs-psi-cutoff=0", which is very unfriendly to normal users and that is not the best practice.

> I would propose to replace  HasAccurateProfile parameter in isColdBlock with an enum of profileKind (InstruProfile, Sample Profile, or Sample Profile with FSAFDO etc). And we can have different heuristic for different profile kind.

I'll prepare another CL to do the suggested change.

Repository:
  rG LLVM Github Monorepo

CHANGES SINCE LAST ACTION
  https://reviews.llvm.org/D152399/new/

https://reviews.llvm.org/D152399