[llvm] [AMDGPU] Introduce ordering parameter to atomic intrinsics and introduce new llvm.amdgcn.image.atomic.load intrinsic. (PR #73613)
Matt Arsenault via llvm-commits
llvm-commits at lists.llvm.org
Wed Apr 24 23:36:46 PDT 2024
arsenm wrote:
> Sorry for the late response.
>
> > I'm wondering if we should encode the ordering with metadata or something better than an integer
>
> Is visibility the only reason for this suggestion? @nhaehnle what do you think?
Partially. Previously I was thinking of introducing an IR change to allow arbitrary calls to carry this kind of information. i.e. either a control bit to add the atomicrmw operands, or to encode it as an operand bundle. Really deviating from how atomicrmw represents this is unfortunate
https://github.com/llvm/llvm-project/pull/73613
More information about the llvm-commits
mailing list