[llvm] [AMDGPU][GlobalISel] Enable kernel argument preloading (PR #134655)

[Matt Arsenault via llvm-commits]

================
@@ -497,6 +499,66 @@ static void allocateHSAUserSGPRs(CCState &CCInfo,
   // these from the dispatch pointer.
 }
 
+void AMDGPUCallLowering::lowerPreloadedParameter(
+    MachineIRBuilder &B, ArrayRef<Register> VRegs, Type *ArgTy,
+    uint64_t ArgOffset, Align Alignment,
+    ArrayRef<MCRegister> PreloadRegs) const {
+  MachineFunction &MF = B.getMF();
+  const GCNSubtarget *Subtarget = &MF.getSubtarget<GCNSubtarget>();
+  MachineRegisterInfo &MRI = MF.getRegInfo();
+  const SIRegisterInfo *TRI = Subtarget->getRegisterInfo();
+  const DataLayout &DL = B.getDataLayout();
+
+  LLT ResTy = getLLTForType(*ArgTy, DL);
+  LLT ScalarTy = LLT::scalar(DL.getTypeSizeInBits(ArgTy));
----------------
arsenm wrote:

There's a bunch of code to go from IR type to register type, which most calling convention code should be operating in terms of, not the original IR type (i.e. what's in CCValAssign and ISD::InputArg)

https://github.com/llvm/llvm-project/pull/134655