[llvm] [ValueTracking] improve `isKnownNonZero` precision for `smax` (PR #88170)

[Nikita Popov via llvm-commits]

================
@@ -2828,23 +2828,44 @@ static bool isKnownNonZeroFromOperator(const Operator *I,
       case Intrinsic::uadd_sat:
         return isKnownNonZero(II->getArgOperand(1), DemandedElts, Depth, Q) ||
                isKnownNonZero(II->getArgOperand(0), DemandedElts, Depth, Q);
-      case Intrinsic::smin:
       case Intrinsic::smax: {
-        auto KnownOpImpliesNonZero = [&](const KnownBits &K) {
-          return II->getIntrinsicID() == Intrinsic::smin
-                     ? K.isNegative()
-                     : K.isStrictlyPositive();
+        // If either arg is strictly positive the result is non-zero. Otherwise
+        // the result is non-zero if both ops are non-zero.
+        auto IsNonZero = [&](Value *Op, std::optional<bool> &OpNonZero,
+                             KnownBits OpKnown) {
----------------
nikic wrote:

```suggestion
                             const KnownBits &OpKnown) {
```

https://github.com/llvm/llvm-project/pull/88170