[PATCH] D81211: [AMDGPU] Enable structurizer workarounds by default

[Sameer Sahasrabuddhe via Phabricator via llvm-commits]

sameerds marked an inline comment as done.
sameerds added inline comments.


================
Comment at: llvm/lib/Target/AMDGPU/AMDGPUTargetMachine.cpp:197
 static cl::opt<bool> EnableStructurizerWorkarounds(
     "amdgpu-enable-structurizer-workarounds",
+    cl::desc("Enable workarounds for the StructurizeCFG pass"), cl::init(true),
----------------
cfang wrote:
> Can we come out with a name that is closer to what was implemented, like amdgpu-unify-loop-exits, or even amdgpu-normalize-loops(cfg)? "workarounds" is kind of misleading.
This one option controls more than one pass. The intention of naming this option is to indicate to the user that additional things are being done to work around known structurizer limitations. Each of these pass is used to avoid a specific situation where the structurizer is know to fail. We could have one option per pass like the names you suggest, but that sounds like overkill. We might even add more passes to the set of workarounds in the future. 


Repository:
  rG LLVM Github Monorepo

CHANGES SINCE LAST ACTION
  https://reviews.llvm.org/D81211/new/

https://reviews.llvm.org/D81211