[PATCH] D137542: [AMDGPU] Use aperture registers instead of S_GETREG

[Pierre van Houtryve via Phabricator via llvm-commits]

Pierre-vh added inline comments.


================
Comment at: llvm/lib/Target/AMDGPU/AMDGPUInstructionSelector.cpp:149-150
+  // register with a reference to its artificial HI 32 bits.
+  if (IsApertureRegister64(SrcReg))
+    return RBI.constrainGenericRegister(DstReg, AMDGPU::SGPR_64RegClass, *MRI);
+
----------------
arsenm wrote:
> You shouldn't need to special case this (also note there's no equivalent for the DAG). Either it needs to belong to a different class from SReg_64, or you need to reserve the high bits 
> or you need to reserve the high bits
Aren't they reserved with `reserveRegisterTuples`?

If I create a different class, do you mean I should exclude those aperture registers from SReg and create a superclass with both SReg and the apertures? I'm not sure I understand how the class will help if I shouldn't special-case the uses of those aperture registers


Repository:
  rG LLVM Github Monorepo

CHANGES SINCE LAST ACTION
  https://reviews.llvm.org/D137542/new/

https://reviews.llvm.org/D137542