Add a convenience method to copy wrapping, exact, and fast-math flags (NFC).

The loop vectorizer preserves wrapping, exact, and fast-math properties of scalar instructions.
This patch adds a convenience method to make that operation easier because we need to do this
in the loop vectorizer, SLP vectorizer, and possibly other places.

Although this is a 'no functional change' patch, I've added a testcase to verify that the exact
flag is preserved by the loop vectorizer. The wrapping and fast-math flags are already checked
in existing testcases.

Differential Revision: http://reviews.llvm.org/D5138

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@216886 91177308-0d34-0410-b5e6-96231b3b80d8

diff --git a/include/llvm/IR/InstrTypes.h b/include/llvm/IR/InstrTypes.h
index 981aad852b292747e90e89dccda23490cc7450c2..799f684349f125f261047253b3fafde89fb793e5 100644 (file)
--- a/include/llvm/IR/InstrTypes.h
+++ b/include/llvm/IR/InstrTypes.h
@@ -358,6 +358,10 @@ public:
   /// isExact - Determine whether the exact flag is set.
   bool isExact() const;
 
+  /// Convenience method to copy wrapping, exact, and fast-math flag values
+  /// from V to this instruction.
+  void copyFlags(const Value *V);
+
   // Methods for support type inquiry through isa, cast, and dyn_cast:
   static inline bool classof(const Instruction *I) {
     return I->isBinaryOp();

diff --git a/lib/IR/Instructions.cpp b/lib/IR/Instructions.cpp
index 9553252f4e96343f45175a9d277d1e106dab97a3..16993f1bba545d074bce5bec4b9c24e32e975c9d 100644 (file)
--- a/lib/IR/Instructions.cpp
+++ b/lib/IR/Instructions.cpp
@@ -2030,6 +2030,22 @@ bool BinaryOperator::isExact() const {
   return cast<PossiblyExactOperator>(this)->isExact();
 }
 
+void BinaryOperator::copyFlags(const Value *V) {
+  // Copy the wrapping flags.
+  if (auto *OB = dyn_cast<OverflowingBinaryOperator>(V)) {
+    setHasNoSignedWrap(OB->hasNoSignedWrap());
+    setHasNoUnsignedWrap(OB->hasNoUnsignedWrap());
+  }
+
+  // Copy the exact flag.
+  if (auto *PE = dyn_cast<PossiblyExactOperator>(V))
+    setIsExact(PE->isExact());
+  
+  // Copy the fast-math flags.
+  if (auto *FP = dyn_cast<FPMathOperator>(V))
+    setFastMathFlags(FP->getFastMathFlags());
+}
+
 //===----------------------------------------------------------------------===//
 //                             FPMathOperator Class
 //===----------------------------------------------------------------------===//

diff --git a/lib/Transforms/Vectorize/LoopVectorize.cpp b/lib/Transforms/Vectorize/LoopVectorize.cpp
index 5ca9106e65275880bb0e8a85bed0bf8b012baff7..ba3301287882c5468cb5169395d96e79c93acffe 100644 (file)
--- a/lib/Transforms/Vectorize/LoopVectorize.cpp
+++ b/lib/Transforms/Vectorize/LoopVectorize.cpp
@@ -3248,19 +3248,9 @@ void InnerLoopVectorizer::vectorizeBlockInLoop(BasicBlock *BB, PhiVector *PV) {
       for (unsigned Part = 0; Part < UF; ++Part) {
         Value *V = Builder.CreateBinOp(BinOp->getOpcode(), A[Part], B[Part]);
 
-        // Update the NSW, NUW and Exact flags. Notice: V can be an Undef.
-        BinaryOperator *VecOp = dyn_cast<BinaryOperator>(V);
-        if (VecOp && isa<OverflowingBinaryOperator>(BinOp)) {
-          VecOp->setHasNoSignedWrap(BinOp->hasNoSignedWrap());
-          VecOp->setHasNoUnsignedWrap(BinOp->hasNoUnsignedWrap());
-        }
-        if (VecOp && isa<PossiblyExactOperator>(VecOp))
-          VecOp->setIsExact(BinOp->isExact());
-
-        // Copy the fast-math flags.
-        if (VecOp && isa<FPMathOperator>(V))
-          VecOp->setFastMathFlags(it->getFastMathFlags());
-
+        if (BinaryOperator *VecOp = dyn_cast<BinaryOperator>(V))
+          VecOp->copyFlags(BinOp);
+        
         Entry[Part] = V;
       }
 

diff --git a/test/Transforms/LoopVectorize/exact.ll b/test/Transforms/LoopVectorize/exact.ll
new file mode 100644 (file)
index 0000000..0a8fbf3
--- /dev/null
+++ b/test/Transforms/LoopVectorize/exact.ll
@@ -0,0 +1,24 @@
+; RUN: opt < %s  -loop-vectorize -force-vector-width=4 -S | FileCheck %s
+
+target datalayout = "e-m:o-i64:64-f80:128-n8:16:32:64-S128"
+target triple = "x86_64-apple-macosx10.9.0"
+
+; CHECK-LABEL: @lshr_exact(
+; CHECK: lshr exact <4 x i32>
+define void @lshr_exact(i32* %x) {
+entry:
+  br label %for.body
+
+for.body:
+  %iv = phi i64 [ 0, %entry ], [ %iv.next, %for.body ]
+  %arrayidx = getelementptr inbounds i32* %x, i64 %iv
+  %0 = load i32* %arrayidx, align 4
+  %conv1 = lshr exact i32 %0, 1
+  store i32 %conv1, i32* %arrayidx, align 4
+  %iv.next = add nuw nsw i64 %iv, 1
+  %exitcond = icmp eq i64 %iv.next, 256
+  br i1 %exitcond, label %for.end, label %for.body
+
+for.end:
+  ret void
+}