[PATCH] D77344: [AMDGPU] Propagate AGPR RC from PHI to its PHI operands

[Stanislav Mekhanoshin via Phabricator via llvm-commits]

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


================
Comment at: llvm/test/CodeGen/AMDGPU/mfma-loop.ll:534
+}
+
 declare <32 x float> @llvm.amdgcn.mfma.f32.32x32x1f32(float, float, <32 x float>, i32, i32, i32)
----------------
rampitec wrote:
> arsenm wrote:
> > Should also have a test where a phi directly refers to itself
> Is it a real case?! This is SSA, value should be defined earlier.
I cannot create a relevant test. If PHI refers itself it does not refer an MFMA inside the loop which is needed to start propagation. I.e. such PHI never triggers this code.


CHANGES SINCE LAST ACTION
  https://reviews.llvm.org/D77344/new/

https://reviews.llvm.org/D77344