[LVL][CSA] Legalize CSA vectorization

Author: michaelmaitland

llvm/include/llvm/Analysis/TargetTransformInfo.h

@@ -1800,6 +1800,10 @@ class TargetTransformInfo {
         : EVLParamStrategy(EVLParamStrategy), OpStrategy(OpStrategy) {}
   };
 
+  /// \returns true if the loop vectorizer should vectorize conditional
+  /// scalar assignments for the target.
+  bool enableCSAVectorization() const;
+
   /// \returns How the target needs this vector-predicated operation to be
   /// transformed.
   VPLegalization getVPLegalizationStrategy(const VPIntrinsic &PI) const;
@@ -2225,6 +2229,7 @@ class TargetTransformInfo::Concept {
                              SmallVectorImpl<Use *> &OpsToSink) const = 0;
 
   virtual bool isVectorShiftByScalarCheap(Type *Ty) const = 0;
+  virtual bool enableCSAVectorization() const = 0;
   virtual VPLegalization
   getVPLegalizationStrategy(const VPIntrinsic &PI) const = 0;
   virtual bool hasArmWideBranch(bool Thumb) const = 0;
@@ -3020,6 +3025,10 @@ class TargetTransformInfo::Model final : public TargetTransformInfo::Concept {
     return Impl.isVectorShiftByScalarCheap(Ty);
   }
 
+  bool enableCSAVectorization() const override {
+    return Impl.enableCSAVectorization();
+  }
+
   VPLegalization
   getVPLegalizationStrategy(const VPIntrinsic &PI) const override {
     return Impl.getVPLegalizationStrategy(PI);

llvm/include/llvm/Analysis/TargetTransformInfoImpl.h

@@ -997,6 +997,8 @@ class TargetTransformInfoImplBase {
 
   bool isVectorShiftByScalarCheap(Type *Ty) const { return false; }
 
+  bool enableCSAVectorization() const { return false; }
+
   TargetTransformInfo::VPLegalization
   getVPLegalizationStrategy(const VPIntrinsic &PI) const {
     return TargetTransformInfo::VPLegalization(

llvm/include/llvm/Transforms/Vectorize/LoopVectorizationLegality.h

@@ -27,6 +27,7 @@
 #define LLVM_TRANSFORMS_VECTORIZE_LOOPVECTORIZATIONLEGALITY_H
 
 #include "llvm/ADT/MapVector.h"
+#include "llvm/Analysis/CSADescriptors.h"
 #include "llvm/Analysis/LoopAccessAnalysis.h"
 #include "llvm/Support/TypeSize.h"
 #include "llvm/Transforms/Utils/LoopUtils.h"
@@ -269,6 +270,10 @@ class LoopVectorizationLegality {
   /// induction descriptor.
   using InductionList = MapVector<PHINode *, InductionDescriptor>;
 
+  /// CSAList contains the CSA descriptors for all the CSAs that were found
+  /// in the loop, rooted by their phis.
+  using CSAList = MapVector<PHINode *, CSADescriptor>;
+
   /// RecurrenceSet contains the phi nodes that are recurrences other than
   /// inductions and reductions.
   using RecurrenceSet = SmallPtrSet<const PHINode *, 8>;
@@ -321,6 +326,12 @@ class LoopVectorizationLegality {
   /// Returns True if V is a Phi node of an induction variable in this loop.
   bool isInductionPhi(const Value *V) const;
 
+  /// Returns the CSAs found in the loop.
+  const CSAList &getCSAs() const { return CSAs; }
+
+  /// Returns true if Phi is the root of a CSA in the loop.
+  bool isCSAPhi(PHINode *Phi) const { return CSAs.count(Phi) != 0; }
+
   /// Returns a pointer to the induction descriptor, if \p Phi is an integer or
   /// floating point induction.
   const InductionDescriptor *getIntOrFpInductionDescriptor(PHINode *Phi) const;
@@ -545,6 +556,10 @@ class LoopVectorizationLegality {
   void addInductionPhi(PHINode *Phi, const InductionDescriptor &ID,
                        SmallPtrSetImpl<Value *> &AllowedExit);
 
+  // Updates the vetorization state by adding \p Phi to the CSA list.
+  void addCSAPhi(PHINode *Phi, const CSADescriptor &CSADesc,
+                 SmallPtrSetImpl<Value *> &AllowedExit);
+
   /// The loop that we evaluate.
   Loop *TheLoop;
 
@@ -589,6 +604,9 @@ class LoopVectorizationLegality {
   /// variables can be pointers.
   InductionList Inductions;
 
+  /// Holds the conditional scalar assignments
+  CSAList CSAs;
+
   /// Holds all the casts that participate in the update chain of the induction
   /// variables, and that have been proven to be redundant (possibly under a
   /// runtime guard). These casts can be ignored when creating the vectorized

llvm/lib/Analysis/TargetTransformInfo.cpp

@@ -1329,6 +1329,10 @@ bool TargetTransformInfo::preferEpilogueVectorization() const {
   return TTIImpl->preferEpilogueVectorization();
 }
 
+bool TargetTransformInfo::enableCSAVectorization() const {
+  return TTIImpl->enableCSAVectorization();
+}
+
 TargetTransformInfo::VPLegalization
 TargetTransformInfo::getVPLegalizationStrategy(const VPIntrinsic &VPI) const {
   return TTIImpl->getVPLegalizationStrategy(VPI);

llvm/lib/Target/RISCV/RISCVTargetTransformInfo.cpp

@@ -2315,6 +2315,11 @@ bool RISCVTTIImpl::isLegalMaskedExpandLoad(Type *DataTy, Align Alignment) {
   return true;
 }
 
+bool RISCVTTIImpl::enableCSAVectorization() const {
+  return ST->hasVInstructions() &&
+         ST->getProcFamily() == RISCVSubtarget::SiFive7;
+}
+
 bool RISCVTTIImpl::isLegalMaskedCompressStore(Type *DataTy, Align Alignment) {
   auto *VTy = dyn_cast<VectorType>(DataTy);
   if (!VTy || VTy->isScalableTy())

llvm/lib/Target/RISCV/RISCVTargetTransformInfo.h

@@ -308,6 +308,10 @@ class RISCVTTIImpl : public BasicTTIImplBase<RISCVTTIImpl> {
     return TLI->isVScaleKnownToBeAPowerOfTwo();
   }
 
+  /// \returns true if the loop vectorizer should vectorize conditional
+  /// scalar assignments for the target.
+  bool enableCSAVectorization() const;
+
   /// \returns How the target needs this vector-predicated operation to be
   /// transformed.
   TargetTransformInfo::VPLegalization

llvm/lib/Transforms/Vectorize/LoopVectorizationLegality.cpp

@@ -83,6 +83,10 @@ static cl::opt<bool> EnableHistogramVectorization(
     "enable-histogram-loop-vectorization", cl::init(false), cl::Hidden,
     cl::desc("Enables autovectorization of some loops containing histograms"));
 
+static cl::opt<bool>
+    EnableCSA("enable-csa-vectorization", cl::init(false), cl::Hidden,
+              cl::desc("Control whether CSA loop vectorization is enabled"));
+
 /// Maximum vectorization interleave count.
 static const unsigned MaxInterleaveFactor = 16;
 
@@ -750,6 +754,15 @@ bool LoopVectorizationLegality::setupOuterLoopInductions() {
   return llvm::all_of(Header->phis(), IsSupportedPhi);
 }
 
+void LoopVectorizationLegality::addCSAPhi(
+    PHINode *Phi, const CSADescriptor &CSADesc,
+    SmallPtrSetImpl<Value *> &AllowedExit) {
+  assert(CSADesc.isValid() && "Expected Valid CSADescriptor");
+  LLVM_DEBUG(dbgs() << "LV: found legal CSA opportunity" << *Phi << "\n");
+  AllowedExit.insert(Phi);
+  CSAs.insert({Phi, CSADesc});
+}
+
 /// Checks if a function is scalarizable according to the TLI, in
 /// the sense that it should be vectorized and then expanded in
 /// multiple scalar calls. This is represented in the
@@ -867,14 +880,23 @@ bool LoopVectorizationLegality::canVectorizeInstrs() {
           continue;
         }
 
-        // As a last resort, coerce the PHI to a AddRec expression
-        // and re-try classifying it a an induction PHI.
+        // Try to coerce the PHI to a AddRec expression and re-try classifying
+        // it a an induction PHI.
         if (InductionDescriptor::isInductionPHI(Phi, TheLoop, PSE, ID, true) &&
             !IsDisallowedStridedPointerInduction(ID)) {
           addInductionPhi(Phi, ID, AllowedExit);
           continue;
         }
 
+        // Check if the PHI can be classified as a CSA PHI.
+        if (EnableCSA || (TTI->enableCSAVectorization() &&
+                          EnableCSA.getNumOccurrences() == 0)) {
+          if (auto CSADesc = CSADescriptor::isCSAPhi(Phi, TheLoop)) {
+            addCSAPhi(Phi, CSADesc, AllowedExit);
+            continue;
+          }
+        }
+
         reportVectorizationFailure("Found an unidentified PHI",
             "value that could not be identified as "
             "reduction is used outside the loop",
@@ -1846,11 +1868,15 @@ bool LoopVectorizationLegality::canFoldTailByMasking() const {
   for (const auto &Reduction : getReductionVars())
     ReductionLiveOuts.insert(Reduction.second.getLoopExitInstr());
 
+  SmallPtrSet<const Value *, 8> CSALiveOuts;
+  for (const auto &CSA : getCSAs())
+    CSALiveOuts.insert(CSA.second.getAssignment());
+
   // TODO: handle non-reduction outside users when tail is folded by masking.
   for (auto *AE : AllowedExit) {
     // Check that all users of allowed exit values are inside the loop or
-    // are the live-out of a reduction.
-    if (ReductionLiveOuts.count(AE))
+    // are the live-out of a reduction or a CSA
+    if (ReductionLiveOuts.count(AE) || CSALiveOuts.count(AE))
       continue;
     for (User *U : AE->users()) {
       Instruction *UI = cast<Instruction>(U);