[PATCH] D109870: [AMDGPU] Enable the pass "amdgpu-replace-lds-use-with-pointer"

[Mahesha S via Phabricator via llvm-commits]

hsmhsm added inline comments.


================
Comment at: llvm/lib/Target/AMDGPU/AMDGPUReplaceLDSUseWithPointer.cpp:199
+    auto *EI = &(*(EBB.getFirstInsertionPt()));
+    BasicBlock::reverse_iterator RIT(EBB.getTerminator());
+    while (!isa<AllocaInst>(*RIT) && (&*RIT != EI))
----------------
arsenm wrote:
> Why is this using a reverse iterator? Why not just forward iterate since that's where the allocas are?
(At least in the failed openmp test case), what I noticed is that alloca at the beginning are not contiguous one after other, but there are in between addressspace casts like below (I am not sure, if there is a possibility of other in-between instructions too other than addressspace). So I went with the reverse iteration. But, actually if we are able to successfully forward iterate, that would very neat.

alloca
addresss-space-cast 
alloca
addresss-space-cast 
....


Repository:
  rG LLVM Github Monorepo

CHANGES SINCE LAST ACTION
  https://reviews.llvm.org/D109870/new/

https://reviews.llvm.org/D109870