[PATCH] D141355: [AMDGPU] Add NewPM support to AMDGPUUnifyDivergentExitNodes pass
Matt Arsenault via Phabricator via llvm-commits
llvm-commits at lists.llvm.org
Tue Jan 10 09:41:29 PST 2023
arsenm added inline comments.
================
Comment at: llvm/lib/Target/AMDGPU/AMDGPUUnifyDivergentExitNodes.cpp:118-119
/// XXX - Is there a more efficient way to find this?
-static bool isUniformlyReached(const LegacyDivergenceAnalysis &DA,
- BasicBlock &BB) {
+static bool isUniformlyReached(BasicBlock &BB,
+ std::function<bool(Value *)> IsUniform) {
SmallVector<BasicBlock *, 8> Stack(predecessors(&BB));
----------------
gandhi21299 wrote:
> arsenm wrote:
> > Not sure why you really need to change this, but the function should probably be a template argument. Is DA just missing a layer to present the PM independent analysis?
> The LegacyPM version depends on `LegacyDivergenceAnalysis` info whereas the NewPM version of this pass depends on `DivergenceInfo`.
There should be a common logic DivergenceInfo that both pass versions export
Repository:
rG LLVM Github Monorepo
CHANGES SINCE LAST ACTION
https://reviews.llvm.org/D141355/new/
https://reviews.llvm.org/D141355
More information about the llvm-commits
mailing list