[llvm] [VPlan] Expand SCEV min/max via scalar intrinsic VPInstructions. (PR #207842)

[Ramkumar Ramachandra via llvm-commits]

================
@@ -842,6 +842,44 @@ VPValue *VPSCEVExpander::tryToExpand(const SCEV *S) {
     }
     return Builder.createScalarCast(Opcode, Op, S->getType(), DL);
   }
+  case scUMaxExpr:
+  case scSMaxExpr:
+  case scUMinExpr:
+  case scSMinExpr: {
+    auto *MinMax = cast<SCEVMinMaxExpr>(S);
+    Intrinsic::ID IntrinsicID;
+    switch (S->getSCEVType()) {
+    case scUMaxExpr:
+      IntrinsicID = Intrinsic::umax;
+      break;
+    case scSMaxExpr:
+      IntrinsicID = Intrinsic::smax;
+      break;
+    case scUMinExpr:
+      IntrinsicID = Intrinsic::umin;
+      break;
+    case scSMinExpr:
+      IntrinsicID = Intrinsic::smin;
+      break;
+    default:
+      llvm_unreachable("not a min/max SCEV");
+    }
+    // Chain operands in reverse order matching SCEVExpander's expansion of
+    // min/max expressions.
+    VPValue *Result =
+        tryToExpand(MinMax->getOperand(MinMax->getNumOperands() - 1));
+    if (!Result)
+      return nullptr;
+    Type *Ty = MinMax->getType();
+    for (const SCEVUse &Op : drop_begin(reverse(MinMax->operands()))) {
+      VPValue *RHS = tryToExpand(Op);
+      if (!RHS)
+        return nullptr;
+      Result =
+          Builder.createScalarIntrinsic(IntrinsicID, {Result, RHS}, Ty, DL);
+    }
----------------
artagnon wrote:

Hm, seems wrong, considering that we could bail in between? Shouldn't we collect all the VPValues from tryToExpand (with the bailout) first, before creating scalar intrinsics?

https://github.com/llvm/llvm-project/pull/207842