[llvm] [RISC-V][GISEL] Select G_BITCAST for scalable vectors (PR #101486)

[Michael Maitland via llvm-commits]

================
@@ -757,13 +758,21 @@ bool RISCVInstructionSelector::replacePtrWithInt(MachineOperand &Op,
                                                  MachineIRBuilder &MIB,
                                                  MachineRegisterInfo &MRI) {
   Register PtrReg = Op.getReg();
-  assert(MRI.getType(PtrReg).isPointer() && "Operand is not a pointer!");
-
+  const LLT PtrTy = MRI.getType(PtrReg);
   const LLT sXLen = LLT::scalar(STI.getXLen());
-  auto PtrToInt = MIB.buildPtrToInt(sXLen, PtrReg);
-  MRI.setRegBank(PtrToInt.getReg(0), RBI.getRegBank(RISCV::GPRBRegBankID));
-  Op.setReg(PtrToInt.getReg(0));
-  return select(*PtrToInt);
+  if (PtrTy.isPointer()) {
+    auto PtrToInt = MIB.buildPtrToInt(sXLen, PtrReg);
+    MRI.setRegBank(PtrToInt.getReg(0), RBI.getRegBank(RISCV::GPRBRegBankID));
+    Op.setReg(PtrToInt.getReg(0));
+    return select(*PtrToInt);
+  }
+  assert(PtrTy.isPointerVector() &&
+         "Operand must be a pointer of a vector of pointers");
+  assert(PtrTy.isScalableVector() &&
+         "Currently only working for scalable vector of pointers now");
+  MRI.setType(PtrReg, LLT::scalable_vector(
----------------
michaelmaitland wrote:

It [looks like AArch64](https://github.com/llvm/llvm-project/blob/3fbf6f8bb183ad8b9157e50c442479f4ca7a9b8d/llvm/lib/Target/AArch64/GISel/AArch64InstructionSelector.cpp#L2194) changes the type without building a PtrToInt. They say it works because all users have been selected already (bottom up selection) so the type does not matter for them. I think we can do the same thing, which is what you have done here.

I wonder if we really needed to build the PtrToInt above for the same reason. I can take a look at that in a follow up patch.

https://github.com/llvm/llvm-project/pull/101486