[llvm] [AMDGPU] Use LSH for lowering ctlz_zero_undef.i8/i16 (PR #88512)

[Matt Arsenault via llvm-commits]

================
@@ -4145,6 +4156,24 @@ bool AMDGPULegalizerInfo::legalizeCTLZ_CTTZ(MachineInstr &MI,
   return true;
 }
 
+bool AMDGPULegalizerInfo::legalizeCTLZ_ZERO_UNDEF(MachineInstr &MI,
+                                                  MachineRegisterInfo &MRI,
+                                                  MachineIRBuilder &B) const {
+  Register Dst = MI.getOperand(0).getReg();
+  Register Src = MI.getOperand(1).getReg();
+  LLT SrcTy = MRI.getType(Src);
+  TypeSize NumBits = SrcTy.getSizeInBits();
+
+  assert(NumBits < 32u);
+
+  auto ShiftAmt = B.buildConstant(S32, 32u - NumBits);
+  auto Extend = B.buildAnyExt(S32, {Src}).getReg(0u);
+  auto Shift = B.buildLShr(S32, {Extend}, ShiftAmt).getReg(0u);
----------------
arsenm wrote:

```suggestion
  auto Shift = B.buildLShr(S32, {Extend}, ShiftAmt);;
```

https://github.com/llvm/llvm-project/pull/88512