[llvm] [RISCV] Use vsetivli instead of `x0,x0` form to retain SEW/LMUL when AVL is imm (PR #169307)

[Luke Lau via llvm-commits]

================
@@ -1725,6 +1737,14 @@ void RISCVInsertVSETVLI::doPRE(MachineBasicBlock &MBB) {
                 AvailableInfo, OldExit);
 }
 
+bool RISCVInsertVSETVLI::isVLPreserving(const MachineInstr &PrevMI,
+                                        const MachineInstr &MI) const {
+  return MI.getOpcode() == RISCV::PseudoVSETIVLI &&
+         PrevMI.getOpcode() == RISCV::PseudoVSETIVLI &&
+         getInfoForVSETVLI(PrevMI).getAVLImm() ==
+             getInfoForVSETVLI(MI).getAVLImm();
----------------
lukel97 wrote:

Nit/premature optimization :), instead of constructing the VSETVLIInfo I think you can just check PrevMI.getOperand(1).getImm() == MI.getOperand(1).getImm(). And then I think you can make the function static

https://github.com/llvm/llvm-project/pull/169307