[SLP] Preserve IR flags for extra args.

Summary:
We should preserve IR flags for extra args. These IR flags should be
taken from original scalar operations, not from the reduction
operations.

Reviewers: mkuper, mzolotukhin, hfinkel

Subscribers: llvm-commits

Differential Revision: https://reviews.llvm.org/D30447

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@296613 91177308-0d34-0410-b5e6-96231b3b80d8

diff --git a/lib/Transforms/Vectorize/SLPVectorizer.cpp b/lib/Transforms/Vectorize/SLPVectorizer.cpp
index c6d26abc76abb356d552bc0aad3a6b7e3ba28aeb..de18a6540465a432917932f819e6c7b7de4f5ece 100644 (file)
--- a/lib/Transforms/Vectorize/SLPVectorizer.cpp
+++ b/lib/Transforms/Vectorize/SLPVectorizer.cpp
@@ -212,14 +212,14 @@ static unsigned getSameOpcode(ArrayRef<Value *> VL) {
 /// Flag set: NSW, NUW, exact, and all of fast-math.
 static void propagateIRFlags(Value *I, ArrayRef<Value *> VL) {
   if (auto *VecOp = dyn_cast<Instruction>(I)) {
-    if (auto *Intersection = dyn_cast<Instruction>(VL[0])) {
-      // Intersection is initialized to the 0th scalar,
-      // so start counting from index '1'.
+    if (auto *I0 = dyn_cast<Instruction>(VL[0])) {
+      // VecOVp is initialized to the 0th scalar, so start counting from index
+      // '1'.
+      VecOp->copyIRFlags(I0);
       for (int i = 1, e = VL.size(); i < e; ++i) {
         if (auto *Scalar = dyn_cast<Instruction>(VL[i]))
-          Intersection->andIRFlags(Scalar);
+          VecOp->andIRFlags(Scalar);
       }
-      VecOp->copyIRFlags(Intersection);
     }
   }
 }
@@ -304,7 +304,8 @@ public:
   typedef SmallVector<Instruction *, 16> InstrList;
   typedef SmallPtrSet<Value *, 16> ValueSet;
   typedef SmallVector<StoreInst *, 8> StoreList;
-  typedef MapVector<Value *, SmallVector<DebugLoc, 2>> ExtraValueToDebugLocsMap;
+  typedef MapVector<Value *, SmallVector<Instruction *, 2>>
+      ExtraValueToDebugLocsMap;
 
   BoUpSLP(Function *Func, ScalarEvolution *Se, TargetTransformInfo *Tti,
           TargetLibraryInfo *TLi, AliasAnalysis *Aa, LoopInfo *Li,
@@ -4430,7 +4431,7 @@ public:
     // The same extra argument may be used several time, so log each attempt
     // to use it.
     for (auto &Pair : ExtraArgs)
-      ExternallyUsedValues[Pair.second].push_back(Pair.first->getDebugLoc());
+      ExternallyUsedValues[Pair.second].push_back(Pair.first);
     while (i < NumReducedVals - ReduxWidth + 1 && ReduxWidth > 2) {
       auto VL = makeArrayRef(&ReducedVals[i], ReduxWidth);
       V.buildTree(VL, ExternallyUsedValues, ReductionOps);
@@ -4481,10 +4482,11 @@ public:
         assert(!Pair.second.empty() &&
                "At least one DebugLoc must be inserted");
         // Add each externally used value to the final reduction.
-        for (auto &DL : Pair.second) {
-          Builder.SetCurrentDebugLocation(DL);
+        for (auto *I : Pair.second) {
+          Builder.SetCurrentDebugLocation(I->getDebugLoc());
           VectorizedTree = Builder.CreateBinOp(ReductionOpcode, VectorizedTree,
                                                Pair.first, "bin.extra");
+          propagateIRFlags(VectorizedTree, I);
         }
       }
       // Update users.

diff --git a/test/Transforms/SLPVectorizer/X86/horizontal-list.ll b/test/Transforms/SLPVectorizer/X86/horizontal-list.ll
index 3553bfbd829ee7ecc64d09fe8a818544694f92d5..73844037f12eaa8364ec898e058a7b5c339c3b06 100644 (file)
--- a/test/Transforms/SLPVectorizer/X86/horizontal-list.ll
+++ b/test/Transforms/SLPVectorizer/X86/horizontal-list.ll
@@ -1679,8 +1679,8 @@ define i32 @wobble(i32 %arg, i32 %bar) {
 ; CHECK-NEXT:    [[RDX_SHUF1:%.*]] = shufflevector <4 x i32> [[BIN_RDX]], <4 x i32> undef, <4 x i32> <i32 1, i32 undef, i32 undef, i32 undef>
 ; CHECK-NEXT:    [[BIN_RDX2:%.*]] = add <4 x i32> [[BIN_RDX]], [[RDX_SHUF1]]
 ; CHECK-NEXT:    [[TMP12:%.*]] = extractelement <4 x i32> [[BIN_RDX2]], i32 0
-; CHECK-NEXT:    [[BIN_EXTRA:%.*]] = add i32 [[TMP12]], [[ARG]]
-; CHECK-NEXT:    [[BIN_EXTRA3:%.*]] = add i32 [[BIN_EXTRA]], [[TMP9]]
+; CHECK-NEXT:    [[BIN_EXTRA:%.*]] = add nuw i32 [[TMP12]], [[ARG]]
+; CHECK-NEXT:    [[BIN_EXTRA3:%.*]] = add nsw i32 [[BIN_EXTRA]], [[TMP9]]
 ; CHECK-NEXT:    [[R5:%.*]] = add nsw i32 [[R4]], undef
 ; CHECK-NEXT:    ret i32 [[BIN_EXTRA3]]
 ;
@@ -1707,8 +1707,8 @@ define i32 @wobble(i32 %arg, i32 %bar) {
 ; THRESHOLD-NEXT:    [[RDX_SHUF1:%.*]] = shufflevector <4 x i32> [[BIN_RDX]], <4 x i32> undef, <4 x i32> <i32 1, i32 undef, i32 undef, i32 undef>
 ; THRESHOLD-NEXT:    [[BIN_RDX2:%.*]] = add <4 x i32> [[BIN_RDX]], [[RDX_SHUF1]]
 ; THRESHOLD-NEXT:    [[TMP12:%.*]] = extractelement <4 x i32> [[BIN_RDX2]], i32 0
-; THRESHOLD-NEXT:    [[BIN_EXTRA:%.*]] = add i32 [[TMP12]], [[ARG]]
-; THRESHOLD-NEXT:    [[BIN_EXTRA3:%.*]] = add i32 [[BIN_EXTRA]], [[TMP9]]
+; THRESHOLD-NEXT:    [[BIN_EXTRA:%.*]] = add nuw i32 [[TMP12]], [[ARG]]
+; THRESHOLD-NEXT:    [[BIN_EXTRA3:%.*]] = add nsw i32 [[BIN_EXTRA]], [[TMP9]]
 ; THRESHOLD-NEXT:    [[R5:%.*]] = add nsw i32 [[R4]], undef
 ; THRESHOLD-NEXT:    ret i32 [[BIN_EXTRA3]]
 ;

diff --git a/test/Transforms/SLPVectorizer/X86/scheduling.ll b/test/Transforms/SLPVectorizer/X86/scheduling.ll
index a01646b9dbefad2e0fde5eb06671ee41495c86eb..8395401c5df0b44a5c33b705e04d1c7ece9669d9 100644 (file)
--- a/test/Transforms/SLPVectorizer/X86/scheduling.ll
+++ b/test/Transforms/SLPVectorizer/X86/scheduling.ll
@@ -12,7 +12,7 @@ define i32 @foo(i32* nocapture readonly %diff) #0 {
 ; CHECK-NEXT:    [[RDX_SHUF1:%.*]] = shufflevector <4 x i32> [[BIN_RDX]], <4 x i32> undef, <4 x i32> <i32 1, i32 undef, i32 undef, i32 undef>
 ; CHECK-NEXT:    [[BIN_RDX2:%.*]] = add <4 x i32> [[BIN_RDX]], [[RDX_SHUF1]]
 ; CHECK-NEXT:    [[TMP15:%.*]] = extractelement <4 x i32> [[BIN_RDX2]], i32 0
-; CHECK:         [[ADD52:%.*]] = add i32 [[TMP15]],
+; CHECK:         [[ADD52:%.*]] = add nsw i32 [[TMP15]],
 ; CHECK:          ret i32 [[ADD52]]
 ;
 entry: