[PATCH] D42078: AMDGPU: Fold inline offset for loads properly in moveToVALU on GFX9

[Marek Olšák via Phabricator via llvm-commits]

mareko added inline comments.


================
Comment at: lib/Target/AMDGPU/SIInstrInfo.cpp:3769
+      if (Add &&
+          (Add->getOpcode() == AMDGPU::V_ADD_I32_e32 ||
+           Add->getOpcode() == AMDGPU::V_ADD_U32_e64)) {
----------------
arsenm wrote:
> It's a defect that at this point the e32 version is ever encountered. We shouldn't be emitting VCC uses up front. Where is this coming from? I thought I fixed most of these places already
moveToVALU selects it from S_ADD.


Repository:
  rL LLVM

https://reviews.llvm.org/D42078