[PATCH] D121371: [AMDGPUPromoteAlloca] Make compatible with opaque pointers

[Matt Arsenault via Phabricator via llvm-commits]

arsenm accepted this revision.
arsenm added inline comments.
This revision is now accepted and ready to land.


================
Comment at: llvm/lib/Target/AMDGPU/AMDGPUPromoteAlloca.cpp:433
+      // This is a store of the pointer, not to the pointer.
+      if (U->get() != Ptr)
         return false;
----------------
nikic wrote:
> arsenm wrote:
> > Doesn't this need to check the exact operand number in case a pointer is being stored to itself?
> Yeah, at least far as intent goes. I don't think it's really relevant for correctness, but possibly for profitability.
A captured pointer shouldn't be handled for correctness so I guess it doesn't matter


================
Comment at: llvm/test/CodeGen/AMDGPU/promote-alloca-array-aggregate.ll:40
 ;
   %i = alloca i32
   %f1 = alloca [1 x float]
----------------
nikic wrote:
> arsenm wrote:
> > Weird this test isn't using the right address space
> Yeah, though I don't think it really matters for the transform. Should I adjust the test to use addrspace(5) for the allocas?
It shouldn't matter. I can fix this test later


CHANGES SINCE LAST ACTION
  https://reviews.llvm.org/D121371/new/

https://reviews.llvm.org/D121371