[llvm] [GISel][RISCV]Implement indirect parameter passing for large scalars (PR #95429)

[Michael Maitland via llvm-commits]

================
@@ -764,19 +765,28 @@ bool CallLowering::handleAssignments(ValueHandler &Handler,
       // If we can't directly assign the register, we need one or more
       // intermediate values.
       Args[i].Regs.resize(NumParts);
-
-      // For each split register, create and assign a vreg that will store
-      // the incoming component of the larger value. These will later be
-      // merged to form the final vreg.
-      for (unsigned Part = 0; Part < NumParts; ++Part)
-        Args[i].Regs[Part] = MRI.createGenericVirtualRegister(NewLLT);
+    
+      // When we have indirect parameter passing we are receiving a pointer,
+      // that points to the actual value. In that case we need a pointer.
+      if (VA.getLocInfo() == CCValAssign::Indirect &&
+          Args[i].Flags[0].isSplit()) {
+        if (Handler.isIncomingArgumentHandler())
+          Args[i].Regs[0] = MRI.createGenericVirtualRegister(PointerTy);
+      } else {
+        // For each split register, create and assign a vreg that will store
+        // the incoming component of the larger value. These will later be
+        // merged to form the final vreg.
+        for (unsigned Part = 0; Part < NumParts; ++Part)
+          Args[i].Regs[Part] = MRI.createGenericVirtualRegister(NewLLT);
+      }
     }
 
     assert((j + (NumParts - 1)) < ArgLocs.size() &&
            "Too many regs for number of args");
 
     // Coerce into outgoing value types before register assignment.
-    if (!Handler.isIncomingArgumentHandler() && OrigTy != ValTy) {
+    if (!Handler.isIncomingArgumentHandler() && OrigTy != ValTy &&
----------------
michaelmaitland wrote:

Can you explain why this change is needed? Above we only handle the case when `Handler.isIncomingArgumentHandler()`. Why do we need to avoid `buildCopyToRegs` for outgoing indirect assignment?

https://github.com/llvm/llvm-project/pull/95429