[TTI] Add a hook for specifying per-target defaults for Interleaved Accesses

Summary:
This adds a hook to TTI which enables us to selectively turn on by default
interleaved access vectorization for targets on which we have have performed
the required benchmarking.

Reviewers: rengolin

Subscribers: rengolin, llvm-commits

Differential Revision: http://reviews.llvm.org/D11901

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@244449 91177308-0d34-0410-b5e6-96231b3b80d8

diff --git a/include/llvm/Analysis/TargetTransformInfo.h b/include/llvm/Analysis/TargetTransformInfo.h
index 31036f71f7e49d6233204ed5f4233e04266b7b26..7bfc3356c268961d433df1231ae6888b1af4655d 100644 (file)
--- a/include/llvm/Analysis/TargetTransformInfo.h
+++ b/include/llvm/Analysis/TargetTransformInfo.h
@@ -357,6 +357,9 @@ public:
   /// \brief Don't restrict interleaved unrolling to small loops.
   bool enableAggressiveInterleaving(bool LoopHasReductions) const;
 
+  /// \brief Enable matching of interleaved access groups.
+  bool enableInterleavedAccessVectorization() const;
+
   /// \brief Return hardware support for population count.
   PopcntSupportKind getPopcntSupport(unsigned IntTyWidthInBit) const;
 
@@ -578,6 +581,7 @@ public:
   virtual unsigned getJumpBufSize() = 0;
   virtual bool shouldBuildLookupTables() = 0;
   virtual bool enableAggressiveInterleaving(bool LoopHasReductions) = 0;
+  virtual bool enableInterleavedAccessVectorization() = 0;
   virtual PopcntSupportKind getPopcntSupport(unsigned IntTyWidthInBit) = 0;
   virtual bool haveFastSqrt(Type *Ty) = 0;
   virtual int getFPOpCost(Type *Ty) = 0;
@@ -719,6 +723,9 @@ public:
   bool enableAggressiveInterleaving(bool LoopHasReductions) override {
     return Impl.enableAggressiveInterleaving(LoopHasReductions);
   }
+  bool enableInterleavedAccessVectorization() override {
+    return Impl.enableInterleavedAccessVectorization();
+  }
   PopcntSupportKind getPopcntSupport(unsigned IntTyWidthInBit) override {
     return Impl.getPopcntSupport(IntTyWidthInBit);
   }

diff --git a/include/llvm/Analysis/TargetTransformInfoImpl.h b/include/llvm/Analysis/TargetTransformInfoImpl.h
index 9805f6bc663a15abbc7f1a25acffb3e15b14da84..e903679259fc289a0f5f477f7afa7abdfda4a3bb 100644 (file)
--- a/include/llvm/Analysis/TargetTransformInfoImpl.h
+++ b/include/llvm/Analysis/TargetTransformInfoImpl.h
@@ -230,6 +230,8 @@ public:
 
   bool enableAggressiveInterleaving(bool LoopHasReductions) { return false; }
 
+  bool enableInterleavedAccessVectorization() { return false; }
+
   TTI::PopcntSupportKind getPopcntSupport(unsigned IntTyWidthInBit) {
     return TTI::PSK_Software;
   }

diff --git a/lib/Analysis/TargetTransformInfo.cpp b/lib/Analysis/TargetTransformInfo.cpp
index 91413687eb36bb9b56c513e54fcff897d545c732..f2984ef1ae261f38b091451dfc30eb5f931326b0 100644 (file)
--- a/lib/Analysis/TargetTransformInfo.cpp
+++ b/lib/Analysis/TargetTransformInfo.cpp
@@ -166,6 +166,10 @@ bool TargetTransformInfo::enableAggressiveInterleaving(bool LoopHasReductions) c
   return TTIImpl->enableAggressiveInterleaving(LoopHasReductions);
 }
 
+bool TargetTransformInfo::enableInterleavedAccessVectorization() const {
+  return TTIImpl->enableInterleavedAccessVectorization();
+}
+
 TargetTransformInfo::PopcntSupportKind
 TargetTransformInfo::getPopcntSupport(unsigned IntTyWidthInBit) const {
   return TTIImpl->getPopcntSupport(IntTyWidthInBit);

diff --git a/lib/Transforms/Vectorize/LoopVectorize.cpp b/lib/Transforms/Vectorize/LoopVectorize.cpp
index 139b7fd96324fac80d5c7d66ea53915c3ba37c7d..53074b0f9d6545f91cee64d53e612df3809eb911 100644 (file)
--- a/lib/Transforms/Vectorize/LoopVectorize.cpp
+++ b/lib/Transforms/Vectorize/LoopVectorize.cpp
@@ -3880,9 +3880,15 @@ bool LoopVectorizationLegality::canVectorize() {
                        : "")
                << "!\n");
 
+  bool UseInterleaved = TTI->enableInterleavedAccessVectorization();
+
+  // If an override option has been passed in for interleaved accesses, use it.
+  if (EnableInterleavedMemAccesses.getNumOccurrences() > 0)
+    UseInterleaved = EnableInterleavedMemAccesses;
+
   // Analyze interleaved memory accesses.
-  if (EnableInterleavedMemAccesses)
-    InterleaveInfo.analyzeInterleaving(Strides);
+  if (UseInterleaved)
+     InterleaveInfo.analyzeInterleaving(Strides);
 
   // Okay! We can vectorize. At this point we don't have any other mem analysis
   // which may limit our maximum vectorization factor, so just return true with