[llvm] c008647 - [VPlan] Introduce isHeaderMask helper (NFCI).

[Florian Hahn via llvm-commits]

Author: Florian Hahn
Date: 2024-06-19T20:17:01+01:00
New Revision: c008647b3a60efdbb5499c3df2b3e403728c29da

URL: https://github.com/llvm/llvm-project/commit/c008647b3a60efdbb5499c3df2b3e403728c29da
DIFF: https://github.com/llvm/llvm-project/commit/c008647b3a60efdbb5499c3df2b3e403728c29da.diff

LOG: [VPlan] Introduce isHeaderMask helper (NFCI).

Split off from https://github.com/llvm/llvm-project/pull/92555
and slightly generalized to more precisely check for a header mask.

Use it to replace manual checks in collectHeaderMasks.

Added: 
    

Modified: 
    llvm/lib/Transforms/Vectorize/VPlan.cpp
    llvm/lib/Transforms/Vectorize/VPlan.h
    llvm/lib/Transforms/Vectorize/VPlanPatternMatch.h
    llvm/lib/Transforms/Vectorize/VPlanTransforms.cpp

Removed: 
    


################################################################################
diff  --git a/llvm/lib/Transforms/Vectorize/VPlan.cpp b/llvm/lib/Transforms/Vectorize/VPlan.cpp
index 348a2be5072b4..0016fecb78540 100644
--- a/llvm/lib/Transforms/Vectorize/VPlan.cpp
+++ b/llvm/lib/Transforms/Vectorize/VPlan.cpp
@@ -1459,3 +1459,23 @@ VPValue *vputils::getOrCreateVPValueForSCEVExpr(VPlan &Plan, const SCEV *Expr,
   Plan.addSCEVExpansion(Expr, Expanded);
   return Expanded;
 }
+
+bool vputils::isHeaderMask(VPValue *V, VPlan &Plan) {
+  if (isa<VPActiveLaneMaskPHIRecipe>(V))
+    return true;
+
+  auto IsWideCanonicalIV = [](VPValue *A) {
+    return isa<VPWidenCanonicalIVRecipe>(A) ||
+           (isa<VPWidenIntOrFpInductionRecipe>(A) &&
+            cast<VPWidenIntOrFpInductionRecipe>(A)->isCanonical());
+  };
+
+  VPValue *A, *B;
+  if (match(V, m_ActiveLaneMask(m_VPValue(A), m_VPValue(B))))
+    return B == Plan.getTripCount() &&
+           (match(A, m_ScalarIVSteps(m_CanonicalIV(), m_SpecificInt(1))) ||
+            IsWideCanonicalIV(A));
+
+  return match(V, m_Binary<Instruction::ICmp>(m_VPValue(A), m_VPValue(B))) &&
+         IsWideCanonicalIV(A) && B == Plan.getOrCreateBackedgeTakenCount();
+}

diff  --git a/llvm/lib/Transforms/Vectorize/VPlan.h b/llvm/lib/Transforms/Vectorize/VPlan.h
index 5bb88e4a57dc3..fc25ed907b153 100644
--- a/llvm/lib/Transforms/Vectorize/VPlan.h
+++ b/llvm/lib/Transforms/Vectorize/VPlan.h
@@ -3665,6 +3665,10 @@ inline bool isUniformAfterVectorization(VPValue *VPV) {
     return VPI->isVectorToScalar();
   return false;
 }
+
+/// Return true if \p V is a header mask in \p Plan.
+bool isHeaderMask(VPValue *V, VPlan &Plan);
+
 } // end namespace vputils
 
 } // end namespace llvm

diff  --git a/llvm/lib/Transforms/Vectorize/VPlanPatternMatch.h b/llvm/lib/Transforms/Vectorize/VPlanPatternMatch.h
index d6b4acb2bdba2..9cd7712624bac 100644
--- a/llvm/lib/Transforms/Vectorize/VPlanPatternMatch.h
+++ b/llvm/lib/Transforms/Vectorize/VPlanPatternMatch.h
@@ -303,6 +303,46 @@ inline BinaryVPInstruction_match<Op0_t, Op1_t, VPInstruction::LogicalAnd>
 m_LogicalAnd(const Op0_t &Op0, const Op1_t &Op1) {
   return m_VPInstruction<VPInstruction::LogicalAnd, Op0_t, Op1_t>(Op0, Op1);
 }
+
+struct VPCanonicalIVPHI_match {
+  bool match(const VPValue *V) {
+    auto *DefR = V->getDefiningRecipe();
+    return DefR && match(DefR);
+  }
+
+  bool match(const VPRecipeBase *R) { return isa<VPCanonicalIVPHIRecipe>(R); }
+};
+
+inline VPCanonicalIVPHI_match m_CanonicalIV() {
+  return VPCanonicalIVPHI_match();
+}
+
+template <typename Op0_t, typename Op1_t> struct VPScalarIVSteps_match {
+  Op0_t Op0;
+  Op1_t Op1;
+
+  VPScalarIVSteps_match(Op0_t Op0, Op1_t Op1) : Op0(Op0), Op1(Op1) {}
+
+  bool match(const VPValue *V) {
+    auto *DefR = V->getDefiningRecipe();
+    return DefR && match(DefR);
+  }
+
+  bool match(const VPRecipeBase *R) {
+    if (!isa<VPScalarIVStepsRecipe>(R))
+      return false;
+    assert(R->getNumOperands() == 2 &&
+           "VPScalarIVSteps must have exactly 2 operands");
+    return Op0.match(R->getOperand(0)) && Op1.match(R->getOperand(1));
+  }
+};
+
+template <typename Op0_t, typename Op1_t>
+inline VPScalarIVSteps_match<Op0_t, Op1_t> m_ScalarIVSteps(const Op0_t &Op0,
+                                                           const Op1_t &Op1) {
+  return VPScalarIVSteps_match<Op0_t, Op1_t>(Op0, Op1);
+}
+
 } // namespace VPlanPatternMatch
 } // namespace llvm
 

diff  --git a/llvm/lib/Transforms/Vectorize/VPlanTransforms.cpp b/llvm/lib/Transforms/Vectorize/VPlanTransforms.cpp
index 8ec67eb2f54bd..e2b7b0c7a219d 100644
--- a/llvm/lib/Transforms/Vectorize/VPlanTransforms.cpp
+++ b/llvm/lib/Transforms/Vectorize/VPlanTransforms.cpp
@@ -1334,13 +1334,10 @@ static SmallVector<VPValue *> collectAllHeaderMasks(VPlan &Plan) {
   // Walk users of wide canonical IVs and collect to all compares of the form
   // (ICMP_ULE, WideCanonicalIV, backedge-taken-count).
   SmallVector<VPValue *> HeaderMasks;
-  VPValue *BTC = Plan.getOrCreateBackedgeTakenCount();
   for (auto *Wide : WideCanonicalIVs) {
     for (VPUser *U : SmallVector<VPUser *>(Wide->users())) {
       auto *HeaderMask = dyn_cast<VPInstruction>(U);
-      if (!HeaderMask || HeaderMask->getOpcode() != Instruction::ICmp ||
-          HeaderMask->getPredicate() != CmpInst::ICMP_ULE ||
-          HeaderMask->getOperand(1) != BTC)
+      if (!HeaderMask || !vputils::isHeaderMask(HeaderMask, Plan))
         continue;
 
       assert(HeaderMask->getOperand(0) == Wide &&