[llvm] [AMDGPU] Omit umin on ctlz/cttz if operand is non-zero. (PR #79127)

[Jay Foad via llvm-commits]

================
@@ -3116,9 +3116,14 @@ SDValue AMDGPUTargetLowering::LowerCTLZ_CTTZ(SDValue Op, SelectionDAG &DAG) cons
     // (cttz_zero_undef src) -> (S_FF1_I32_B64 src)
     SDValue NewOpr = DAG.getNode(NewOpc, SL, MVT::i32, Src);
     if (!ZeroUndef) {
-      const SDValue ConstVal = DAG.getConstant(
-          Op.getValueType().getScalarSizeInBits(), SL, MVT::i32);
-      NewOpr = DAG.getNode(ISD::UMIN, SL, MVT::i32, NewOpr, ConstVal);
+      // umin can be omitted if the operand is known to be non-zero.
+      auto KB = DAG.computeKnownBits(Src);
+      auto const IsNonZero = KB.countMinPopulation() > 0u;
+      if (!IsNonZero) {
----------------
jayfoad wrote:

```suggestion
      if (!KB.isNonZero) {
```

https://github.com/llvm/llvm-project/pull/79127