[PATCH] D124330: Revert "HHVM calling conventions."

[Denis Antrushin via Phabricator via llvm-commits]

dantrushin added inline comments.


================
Comment at: llvm/test/CodeGen/X86/statepoint-fastregalloc.mir:24
 # Same as above but with multiple RegMask operands per instruction.
 # These regmasks have no real meaning and chosen to allow only single register to be assignable ($r12)
 ---
----------------
Amir wrote:
> MatzeB wrote:
> > Amir wrote:
> > > MatzeB wrote:
> > > > I guess this comment needs to be adapted or removed?
> > > If I only replace the register name, would it still make sense? I'm removing csr_64_hhvm and apparently it has an effect on statepoint operands.
> > This feels like they just choose the HHVM convention to show the concept. This feels like they just choose it because HHVM convention only had a single callee-saved register to choose from to make the test more stable. But at a first glance feels fine to switch to any other callee saved reg of the default calling convention instead...
> > You should subscribe the original author(s) of the test to this diff in case they have a comment.
> Thanks!
> 
> @dantrushin – does this change look good to you?
Could you please replace `csr_hhvm`  with `csr_64_cxx_tls_darwin_pe` instead? And register will change to $rbp.


Repository:
  rG LLVM Github Monorepo

CHANGES SINCE LAST ACTION
  https://reviews.llvm.org/D124330/new/

https://reviews.llvm.org/D124330