[PATCH] D147408: [AMDGPU] Enable AMDGPU Atomic Optimizer Pass by default.

[Ruiling, Song via Phabricator via llvm-commits]

ruiling added inline comments.


================
Comment at: llvm/lib/Target/AMDGPU/AMDGPUAtomicOptimizer.cpp:444
+// readlane and writelane intrinsics.
+std::pair<Value *, Value *> AMDGPUAtomicOptimizer::buildScanIteratively(
+    IRBuilder<> &B, AtomicRMWInst::BinOp Op, Value *const Identity, Value *V,
----------------
pravinjagtap wrote:
> Hello @ruiling.
> 
> Your suggestions i.e., loop based iterative approach have been implemented to perform scan operation. Now, we iterate over only active lanes using @llvm.cttz and clear the associated bit when processed so that for the next iteration we will be branching out to next active lane. 
This part LGTM.


Repository:
  rG LLVM Github Monorepo

CHANGES SINCE LAST ACTION
  https://reviews.llvm.org/D147408/new/

https://reviews.llvm.org/D147408