AMDGPU: Use removeAllRegUnitsForPhysReg()

[ruiling]

Summary

I met the issue here when working on something else. I don't look
at it further why I met issue here, but looks like
removeAllRegUnitsForPhysReg() is the right way.

[foad]

llvm/lib/Target/AMDGPU/SIWholeQuadMode.cpp

1591

Can you use it here too?

[ruiling]

also for SCC.

[critson]

LGTM

[arsenm]

llvm/lib/Target/AMDGPU/SIWholeQuadMode.cpp

1591

There's only one regunit so there's no difference

1595

EXEC is reserved, so does the liveness actually exist?

[ruiling]

llvm/lib/Target/AMDGPU/SIWholeQuadMode.cpp

1595

yes, it is a little bit strange, the EXEC_LO_LO16 live range appears after register coalescer, I have not figured out what happened.

[ruiling]

I agree we need to figure out why register coalescer would introduce the liveness for sub-registers of reserved physical register, but I currently don't have enough time to work on that. I think that can be done separately without blocking this change. I would like to apply this change to fix the issue here if no objection. I noticed we already has a bug-report against this issue https://github.com/llvm/llvm-project/issues/54202.

[critson]

No objections from me to you submitting this.
I assume the posted bug report would be sufficient as a starting point for someone (perhaps myself) to try and root cause the why liveness is being introduced for phys subregs?

llvm/lib/Target/AMDGPU/SIWholeQuadMode.cpp

bool SIWholeQuadMode::runOnMachineFunction(MachineFunction &MF) {
 
   // Compute live range for live mask
   if (LiveMaskReg != Exec)
     LIS->createAndComputeVirtRegInterval(LiveMaskReg);
 
   // Physical registers like SCC aren't tracked by default anyway, so just
   // removing the ranges we computed is the simplest option for maintaining
   // the analysis results.
-  LIS->removeRegUnit(*MCRegUnitIterator(MCRegister::from(AMDGPU::SCC), TRI));
+  LIS->removeAllRegUnitsForPhysReg(AMDGPU::SCC);
 
   // If we performed any kills then recompute EXEC
   if (!KillInstrs.empty())
-    LIS->removeRegUnit(*MCRegUnitIterator(AMDGPU::EXEC, TRI));
+    LIS->removeAllRegUnitsForPhysReg(AMDGPU::EXEC);
 
   return true;
 }