[llvm] [AMDGPU] Generate waterfall for calls with SGPR(inreg) argument (PR #146997)

[Shilei Tian via llvm-commits]

================
@@ -7312,6 +7311,43 @@ void SIInstrInfo::legalizeOperandsVALUt16(MachineInstr &MI,
     legalizeOperandsVALUt16(MI, OpIdx, MRI);
 }
 
+void SIInstrInfo::createWaterFall(MachineInstr *MI, MachineDominatorTree *MDT,
+                                  ArrayRef<MachineOperand *> ScalarOps,
+                                  ArrayRef<Register> PhySGPRs) const {
+
+  MachineBasicBlock *MBB = MI->getParent();
+  if (!MBB)
+    return;
+  MachineRegisterInfo &MRI = MBB->getParent()->getRegInfo();
+  const TargetRegisterInfo *TRI = MRI.getTargetRegisterInfo();
+
+  if (MI->getOpcode() == AMDGPU::SI_CALL_ISEL) {
+    // Move everything between ADJCALLSTACKUP and ADJCALLSTACKDOWN and
+    // following copies, we also need to move copies from and to physical
+    // registers into the loop block.
+    unsigned FrameSetupOpcode = this->getCallFrameSetupOpcode();
+    unsigned FrameDestroyOpcode = this->getCallFrameDestroyOpcode();
+
+    // Also move the copies to physical registers into the loop block
+    MachineBasicBlock &MBB = *MI->getParent();
+    MachineBasicBlock::iterator Start(MI);
+    while (Start->getOpcode() != FrameSetupOpcode)
+      --Start;
+    MachineBasicBlock::iterator End(MI);
+    while (End->getOpcode() != FrameDestroyOpcode)
----------------
shiltian wrote:

I suppose frame destroy is always after setup?

https://github.com/llvm/llvm-project/pull/146997