[llvm] [AMDGPU] Fold copies of constant physical registers into their uses (PR #154410)

[Stanislav Mekhanoshin via llvm-commits]

================
@@ -727,8 +728,12 @@ bool SIFoldOperandsImpl::updateOperand(FoldCandidate &Fold) const {
   // 16-bit SGPRs instead of 32-bit ones.
   if (Old.getSubReg() == AMDGPU::lo16 && TRI->isSGPRReg(*MRI, New->getReg()))
     Old.setSubReg(AMDGPU::NoSubRegister);
-  Old.substVirtReg(New->getReg(), New->getSubReg(), *TRI);
-  Old.setIsUndef(New->isUndef());
+  if (New->getReg().isPhysical()) {
+    Old.substPhysReg(New->getReg(), *TRI);
+  } else {
+    Old.substVirtReg(New->getReg(), New->getSubReg(), *TRI);
+    Old.setIsUndef(New->isUndef());
----------------
rampitec wrote:

Well, it does not really change anything. But if you think how it looks in assembly... There is already branching. Yes, it is not necessary. It just hides code which will be dead into a branch which is already there.

https://github.com/llvm/llvm-project/pull/154410