[SLP] Preserve IR flags when vectorizing horizontal reductions.

Summary:
The SLP vectorizer should propagate IR-level optimization hints/flags
(nsw, nuw, exact, fast-math) when converting scalar horizontal
reductions instructions into vectors, just like for other vectorized
instructions.
It doe not include IR propagation for extra arguments, we need to handle
original scalar operations for extra args to propagate correct flags.

Reviewers: mkuper, mzolotukhin, hfinkel

Subscribers: llvm-commits

Differential Revision: https://reviews.llvm.org/D30418

llvm-svn: 296614

3 files changed, 21 insertions, 18 deletions

diff --git a/llvm/lib/Transforms/Vectorize/SLPVectorizer.cpp b/llvm/lib/Transforms/Vectorize/SLPVectorizer.cpp
index de18a654046..b0c70285dd0 100644
--- a/llvm/lib/Transforms/Vectorize/SLPVectorizer.cpp
+++ b/llvm/lib/Transforms/Vectorize/SLPVectorizer.cpp
@@ -4459,11 +4459,12 @@ public:
 
       // Emit a reduction.
       Value *ReducedSubTree =
-          emitReduction(VectorizedRoot, Builder, ReduxWidth);
+          emitReduction(VectorizedRoot, Builder, ReduxWidth, ReductionOps);
       if (VectorizedTree) {
         Builder.SetCurrentDebugLocation(Loc);
         VectorizedTree = Builder.CreateBinOp(ReductionOpcode, VectorizedTree,
                                              ReducedSubTree, "bin.rdx");
+        propagateIRFlags(VectorizedTree, ReductionOps);
       } else
         VectorizedTree = ReducedSubTree;
       i += ReduxWidth;
@@ -4477,6 +4478,7 @@ public:
         Builder.SetCurrentDebugLocation(I->getDebugLoc());
         VectorizedTree =
             Builder.CreateBinOp(ReductionOpcode, VectorizedTree, I);
+        propagateIRFlags(VectorizedTree, ReductionOps);
       }
       for (auto &Pair : ExternallyUsedValues) {
         assert(!Pair.second.empty() &&
@@ -4527,7 +4529,7 @@ private:
 
   /// \brief Emit a horizontal reduction of the vectorized value.
   Value *emitReduction(Value *VectorizedValue, IRBuilder<> &Builder,
-                       unsigned ReduxWidth) {
+                       unsigned ReduxWidth, ArrayRef<Value *> RedOps) {
     assert(VectorizedValue && "Need to have a vectorized tree node");
     assert(isPowerOf2_32(ReduxWidth) &&
            "We only handle power-of-two reductions for now");
@@ -4554,6 +4556,7 @@ private:
           TmpVec, UndefValue::get(TmpVec->getType()), UpperHalf, "rdx.shuf");
         TmpVec = Builder.CreateBinOp(ReductionOpcode, TmpVec, Shuf, "bin.rdx");
       }
+      propagateIRFlags(TmpVec, RedOps);
     }
 
     // The result is in the first element of the vector.
diff --git a/llvm/test/Transforms/SLPVectorizer/X86/horizontal.ll b/llvm/test/Transforms/SLPVectorizer/X86/horizontal.ll
index 715da3e3abf..080f850f91c 100644
--- a/llvm/test/Transforms/SLPVectorizer/X86/horizontal.ll
+++ b/llvm/test/Transforms/SLPVectorizer/X86/horizontal.ll
@@ -624,9 +624,9 @@ define void @i32_red_example4(i32* %res) {
 ; STORE-LABEL: @i32_red_example4(
 ; STORE:         [[TMP0:%.*]] = load <4 x i32>, <4 x i32>* bitcast ([32 x i32]* @arr_i32 to <4 x i32>*), align 16
 ; STORE:         [[RDX_SHUF:%.*]] = shufflevector <4 x i32> [[TMP0]], <4 x i32> undef, <4 x i32> <i32 2, i32 3, i32 undef, i32 undef>
-; STORE-NEXT:    [[BIN_RDX:%.*]] = add <4 x i32> [[TMP0]], [[RDX_SHUF]]
+; STORE-NEXT:    [[BIN_RDX:%.*]] = add nsw <4 x i32> [[TMP0]], [[RDX_SHUF]]
 ; STORE-NEXT:    [[RDX_SHUF1:%.*]] = shufflevector <4 x i32> [[BIN_RDX]], <4 x i32> undef, <4 x i32> <i32 1, i32 undef, i32 undef, i32 undef>
-; STORE-NEXT:    [[BIN_RDX2:%.*]] = add <4 x i32> [[BIN_RDX]], [[RDX_SHUF1]]
+; STORE-NEXT:    [[BIN_RDX2:%.*]] = add nsw <4 x i32> [[BIN_RDX]], [[RDX_SHUF1]]
 ; STORE-NEXT:    [[TMP1:%.*]] = extractelement <4 x i32> [[BIN_RDX2]], i32 0
 ; STORE:         store i32 [[TMP1]], i32* %res, align 16
 ; STORE-NEXT:    ret void
@@ -647,11 +647,11 @@ define void @i32_red_example8(i32* %res) {
 ; STORE-LABEL: @i32_red_example8(
 ; STORE:         [[TMP0:%.*]] = load <8 x i32>, <8 x i32>* bitcast ([32 x i32]* @arr_i32 to <8 x i32>*), align 16
 ; STORE:         [[RDX_SHUF:%.*]] = shufflevector <8 x i32> [[TMP0]], <8 x i32> undef, <8 x i32> <i32 4, i32 5, i32 6, i32 7, i32 undef, i32 undef, i32 undef, i32 undef>
-; STORE-NEXT:    [[BIN_RDX:%.*]] = add <8 x i32> [[TMP0]], [[RDX_SHUF]]
+; STORE-NEXT:    [[BIN_RDX:%.*]] = add nsw <8 x i32> [[TMP0]], [[RDX_SHUF]]
 ; STORE-NEXT:    [[RDX_SHUF1:%.*]] = shufflevector <8 x i32> [[BIN_RDX]], <8 x i32> undef, <8 x i32> <i32 2, i32 3, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
-; STORE-NEXT:    [[BIN_RDX2:%.*]] = add <8 x i32> [[BIN_RDX]], [[RDX_SHUF1]]
+; STORE-NEXT:    [[BIN_RDX2:%.*]] = add nsw <8 x i32> [[BIN_RDX]], [[RDX_SHUF1]]
 ; STORE-NEXT:    [[RDX_SHUF3:%.*]] = shufflevector <8 x i32> [[BIN_RDX2]], <8 x i32> undef, <8 x i32> <i32 1, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
-; STORE-NEXT:    [[BIN_RDX4:%.*]] = add <8 x i32> [[BIN_RDX2]], [[RDX_SHUF3]]
+; STORE-NEXT:    [[BIN_RDX4:%.*]] = add nsw <8 x i32> [[BIN_RDX2]], [[RDX_SHUF3]]
 ; STORE-NEXT:    [[TMP1:%.*]] = extractelement <8 x i32> [[BIN_RDX4]], i32 0
 ; STORE:         store i32 [[TMP1]], i32* %res, align 16
 ; STORE-NEXT:    ret void
@@ -680,13 +680,13 @@ define void @i32_red_example16(i32* %res) {
 ; STORE-LABEL: @i32_red_example16(
 ; STORE:         [[TMP0:%.*]] = load <16 x i32>, <16 x i32>* bitcast ([32 x i32]* @arr_i32 to <16 x i32>*), align 16
 ; STORE:         [[RDX_SHUF:%.*]] = shufflevector <16 x i32> [[TMP0]], <16 x i32> undef, <16 x i32> <i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
-; STORE-NEXT:    [[BIN_RDX:%.*]] = add <16 x i32> [[TMP0]], [[RDX_SHUF]]
+; STORE-NEXT:    [[BIN_RDX:%.*]] = add nsw <16 x i32> [[TMP0]], [[RDX_SHUF]]
 ; STORE-NEXT:    [[RDX_SHUF1:%.*]] = shufflevector <16 x i32> [[BIN_RDX]], <16 x i32> undef, <16 x i32> <i32 4, i32 5, i32 6, i32 7, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
-; STORE-NEXT:    [[BIN_RDX2:%.*]] = add <16 x i32> [[BIN_RDX]], [[RDX_SHUF1]]
+; STORE-NEXT:    [[BIN_RDX2:%.*]] = add nsw <16 x i32> [[BIN_RDX]], [[RDX_SHUF1]]
 ; STORE-NEXT:    [[RDX_SHUF3:%.*]] = shufflevector <16 x i32> [[BIN_RDX2]], <16 x i32> undef, <16 x i32> <i32 2, i32 3, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
-; STORE-NEXT:    [[BIN_RDX4:%.*]] = add <16 x i32> [[BIN_RDX2]], [[RDX_SHUF3]]
+; STORE-NEXT:    [[BIN_RDX4:%.*]] = add nsw <16 x i32> [[BIN_RDX2]], [[RDX_SHUF3]]
 ; STORE-NEXT:    [[RDX_SHUF5:%.*]] = shufflevector <16 x i32> [[BIN_RDX4]], <16 x i32> undef, <16 x i32> <i32 1, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
-; STORE-NEXT:    [[BIN_RDX6:%.*]] = add <16 x i32> [[BIN_RDX4]], [[RDX_SHUF5]]
+; STORE-NEXT:    [[BIN_RDX6:%.*]] = add nsw <16 x i32> [[BIN_RDX4]], [[RDX_SHUF5]]
 ; STORE-NEXT:    [[TMP1:%.*]] = extractelement <16 x i32> [[BIN_RDX6]], i32 0
 ; STORE:         store i32 [[TMP1]], i32* %res, align 16
 ; STORE-NEXT:    ret void
@@ -731,15 +731,15 @@ define void @i32_red_example32(i32* %res) {
 ; STORE-LABEL: @i32_red_example32(
 ; STORE:         [[TMP0:%.*]] = load <32 x i32>, <32 x i32>* bitcast ([32 x i32]* @arr_i32 to <32 x i32>*), align 16
 ; STORE:         [[RDX_SHUF:%.*]] = shufflevector <32 x i32> [[TMP0]], <32 x i32> undef, <32 x i32> <i32 16, i32 17, i32 18, i32 19, i32 20, i32 21, i32 22, i32 23, i32 24, i32 25, i32 26, i32 27, i32 28, i32 29, i32 30, i32 31, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
-; STORE-NEXT:    [[BIN_RDX:%.*]] = add <32 x i32> [[TMP0]], [[RDX_SHUF]]
+; STORE-NEXT:    [[BIN_RDX:%.*]] = add nsw <32 x i32> [[TMP0]], [[RDX_SHUF]]
 ; STORE-NEXT:    [[RDX_SHUF1:%.*]] = shufflevector <32 x i32> [[BIN_RDX]], <32 x i32> undef, <32 x i32> <i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
-; STORE-NEXT:    [[BIN_RDX2:%.*]] = add <32 x i32> [[BIN_RDX]], [[RDX_SHUF1]]
+; STORE-NEXT:    [[BIN_RDX2:%.*]] = add nsw <32 x i32> [[BIN_RDX]], [[RDX_SHUF1]]
 ; STORE-NEXT:    [[RDX_SHUF3:%.*]] = shufflevector <32 x i32> [[BIN_RDX2]], <32 x i32> undef, <32 x i32> <i32 4, i32 5, i32 6, i32 7, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
-; STORE-NEXT:    [[BIN_RDX4:%.*]] = add <32 x i32> [[BIN_RDX2]], [[RDX_SHUF3]]
+; STORE-NEXT:    [[BIN_RDX4:%.*]] = add nsw <32 x i32> [[BIN_RDX2]], [[RDX_SHUF3]]
 ; STORE-NEXT:    [[RDX_SHUF5:%.*]] = shufflevector <32 x i32> [[BIN_RDX4]], <32 x i32> undef, <32 x i32> <i32 2, i32 3, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
-; STORE-NEXT:    [[BIN_RDX6:%.*]] = add <32 x i32> [[BIN_RDX4]], [[RDX_SHUF5]]
+; STORE-NEXT:    [[BIN_RDX6:%.*]] = add nsw <32 x i32> [[BIN_RDX4]], [[RDX_SHUF5]]
 ; STORE-NEXT:    [[RDX_SHUF7:%.*]] = shufflevector <32 x i32> [[BIN_RDX6]], <32 x i32> undef, <32 x i32> <i32 1, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
-; STORE-NEXT:    [[BIN_RDX8:%.*]] = add <32 x i32> [[BIN_RDX6]], [[RDX_SHUF7]]
+; STORE-NEXT:    [[BIN_RDX8:%.*]] = add nsw <32 x i32> [[BIN_RDX6]], [[RDX_SHUF7]]
 ; STORE-NEXT:    [[TMP1:%.*]] = extractelement <32 x i32> [[BIN_RDX8]], i32 0
 ; STORE:         store i32 [[TMP1]], i32* %res, align 16
 ; STORE-NEXT:    ret void
diff --git a/llvm/test/Transforms/SLPVectorizer/X86/scheduling.ll b/llvm/test/Transforms/SLPVectorizer/X86/scheduling.ll
index 8395401c5df..c4f521c8963 100644
--- a/llvm/test/Transforms/SLPVectorizer/X86/scheduling.ll
+++ b/llvm/test/Transforms/SLPVectorizer/X86/scheduling.ll
@@ -8,9 +8,9 @@ define i32 @foo(i32* nocapture readonly %diff) #0 {
 ; CHECK: [[S1:%.+]] = add nsw <4 x i32>
 ; CHECK: store <4 x i32> [[S1]],
 ; CHECK:         [[RDX_SHUF:%.*]] = shufflevector <4 x i32> [[S1]], <4 x i32> undef, <4 x i32> <i32 2, i32 3, i32 undef, i32 undef>
-; CHECK-NEXT:    [[BIN_RDX:%.*]] = add <4 x i32> [[S1]], [[RDX_SHUF]]
+; CHECK-NEXT:    [[BIN_RDX:%.*]] = add nsw <4 x i32> [[S1]], [[RDX_SHUF]]
 ; 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:    [[BIN_RDX2:%.*]] = add nsw <4 x i32> [[BIN_RDX]], [[RDX_SHUF1]]
 ; CHECK-NEXT:    [[TMP15:%.*]] = extractelement <4 x i32> [[BIN_RDX2]], i32 0
 ; CHECK:         [[ADD52:%.*]] = add nsw i32 [[TMP15]],
 ; CHECK:          ret i32 [[ADD52]]