[llvm] [LV] Support argmin/argmax with strict predicates. (PR #170223)

[Florian Hahn via llvm-commits]

================
@@ -1390,7 +1392,136 @@ bool VPlanTransforms::handleFindLastReductions(VPlan &Plan) {
   return true;
 }
 
-bool VPlanTransforms::handleMultiUseReductions(VPlan &Plan) {
+/// For argmin/argmax reductions with strict predicates, convert the existing
+/// FindLastIV reduction to a new UMin reduction of a wide canonical IV. If the
+/// original IV was not canonical, a new canonical wide IV is added, and the
+/// final result is scaled back to the original IV.
+static bool handleFirstArgMinOrMax(VPlan &Plan,
+                                   VPReductionPHIRecipe *MinOrMaxPhiR,
+                                   VPReductionPHIRecipe *FindLastIVPhiR,
+                                   VPWidenIntOrFpInductionRecipe *WideIV,
+                                   VPInstruction *MinOrMaxResult) {
+  Type *Ty = Plan.getVectorLoopRegion()->getCanonicalIVType();
+  // TODO: Support non (i.e., narrower than) canonical IV types.
+  if (Ty != VPTypeAnalysis(Plan).inferScalarType(WideIV))
+    return false;
+
+  // If the original wide IV is not canonical, create a new one. The wide IV is
+  // guaranteed to not wrap for all lanes that are active in the vector loop.
+  auto *FindIVSelectR = cast<VPSingleDefRecipe>(
+      FindLastIVPhiR->getBackedgeValue()->getDefiningRecipe());
+  if (!WideIV->isCanonical()) {
+    VPIRValue *Zero = Plan.getConstantInt(Ty, 0);
+    VPIRValue *One = Plan.getConstantInt(Ty, 1);
+    auto *WidenCanIV = new VPWidenIntOrFpInductionRecipe(
+        nullptr, Zero, One, WideIV->getVFValue(),
+        WideIV->getInductionDescriptor(),
+        VPIRFlags::WrapFlagsTy(/*HasNUW=*/true, /*HasNSW=*/false),
+        WideIV->getDebugLoc());
+    WidenCanIV->insertBefore(WideIV);
+
+    // Update the select to use the wide canonical IV.
+    assert(
+        match(FindIVSelectR, m_Select(m_VPValue(), m_VPValue(), m_VPValue())) &&
+        "backedge value must be a select");
+    WideIV->replaceUsesWithIf(WidenCanIV,
+                              [FindIVSelectR](const VPUser &U, unsigned) {
+                                return FindIVSelectR == &U;
+                              });
+  }
+
+  // Create the new UMin reduction recipe to track the minimum index.
+  assert(!FindLastIVPhiR->isInLoop() && !FindLastIVPhiR->isOrdered() &&
+         "inloop and ordered reductions not supported");
+  VPValue *MaxIV =
+      Plan.getConstantInt(APInt::getMaxValue(Ty->getIntegerBitWidth()));
+  assert(FindLastIVPhiR->getVFScaleFactor() == 1 &&
+         "FindIV reduction must not be scaled");
+  ReductionStyle Style = RdxUnordered{1};
+  auto *FirstIdxPhiR = new VPReductionPHIRecipe(
+      dyn_cast_or_null<PHINode>(FindLastIVPhiR->getUnderlyingValue()),
+      RecurKind::UMin, *MaxIV, *FindIVSelectR, Style,
+      FindLastIVPhiR->hasUsesOutsideReductionChain());
----------------
fhahn wrote:

the phis still store the recurrence kind, and leaving it as FindLastXX may be confusing/misleading?

https://github.com/llvm/llvm-project/pull/170223