[llvm] [LV] Add support for cmp reductions with decreasing IVs. (PR #140451)

[Ramkumar Ramachandra via llvm-commits]

================
@@ -710,21 +713,39 @@ RecurrenceDescriptor::isFindLastIVPattern(Loop *TheLoop, PHINode *OrigPhi,
       return false;
 
     const SCEV *Step = AR->getStepRecurrence(SE);
-    if (!SE.isKnownPositive(Step))
+    if (Kind == RecurKind::FindFirstIVUMin ||
+        Kind == RecurKind::FindFirstIVSMin) {
+      if (!SE.isKnownNegative(Step))
+        return false;
+    } else if (!SE.isKnownPositive(Step))
       return false;
 
     const ConstantRange IVRange = SE.getSignedRange(AR);
     unsigned NumBits = Ty->getIntegerBitWidth();
-    // Keep the minimum value of the recurrence type as the sentinel value.
-    // The maximum acceptable range for the increasing induction variable,
-    // called the valid range, will be defined as
+    // Keep the minimum (FindLast) or maximum (FindFirst) value of the
+    // recurrence type as the sentinel value. The maximum acceptable range for
+    // the induction variable, called the valid range, will be defined as
     //   [<sentinel value> + 1, <sentinel value>)
-    // where <sentinel value> is SignedMin(<recurrence type>)
+    // where <sentinel value> is SignedMin(<recurrence type>) for FindLast or
+    // UnsignedMax(<recurrence type>) for FindFirst.
     // TODO: This range restriction can be lifted by adding an additional
     // virtual OR reduction.
-    const APInt Sentinel = APInt::getSignedMinValue(NumBits);
-    const ConstantRange ValidRange =
-        ConstantRange::getNonEmpty(Sentinel + 1, Sentinel);
+    const APInt Sentinel = Kind == RecurKind::FindFirstIVUMin
+                               ? APInt::getMaxValue(NumBits)
+                               : (Kind == RecurKind::FindFirstIVSMin
+                                      ? APInt::getSignedMaxValue(NumBits)
+                                      : APInt::getSignedMinValue(NumBits));
----------------
artagnon wrote:

This ternary expression also reads badly?

https://github.com/llvm/llvm-project/pull/140451