[SLP] Preserve IR flags when vectorizing horizontal reductions.

[ABataev]

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.

[mkuper]

LGTM

lib/Transforms/Vectorize/SLPVectorizer.cpp

while (i < NumReducedVals - ReduxWidth + 1 && ReduxWidth > 2) {
                    << ". (HorRdx)\n");
 
       // Vectorize a tree.
       DebugLoc Loc = cast<Instruction>(ReducedVals[i])->getDebugLoc();
       Value *VectorizedRoot = V.vectorizeTree(ExternallyUsedValues);
 
       // 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;
       ReduxWidth = PowerOf2Floor(NumReducedVals - i);
     }
 
     if (VectorizedTree) {
       // Finish the reduction.
       for (; i < NumReducedVals; ++i) {
         auto *I = cast<Instruction>(ReducedVals[i]);
         Builder.SetCurrentDebugLocation(I->getDebugLoc());
         VectorizedTree =
             Builder.CreateBinOp(ReductionOpcode, VectorizedTree, I);
+        propagateIRFlags(VectorizedTree, ReductionOps);
       }
       for (auto &Pair : ExternallyUsedValues) {
         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);
           VectorizedTree = Builder.CreateBinOp(ReductionOpcode, VectorizedTree,
DEBUG(dbgs() << "SLP: Adding cost " << VecReduxCost - ScalarReduxCost
                  << (IsPairwiseReduction ? "pairwise" : "splitting")
                  << " reduction)\n");
 
     return VecReduxCost - ScalarReduxCost;
   }
 
   /// \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");
 
     Value *TmpVec = VectorizedValue;
     for (unsigned i = ReduxWidth / 2; i != 0; i >>= 1) {
       if (IsPairwiseReduction) {
         Value *LeftMask =
for (unsigned i = ReduxWidth / 2; i != 0; i >>= 1) {
                                      "bin.rdx");
       } else {
         Value *UpperHalf =
           createRdxShuffleMask(ReduxWidth, i, false, false, Builder);
         Value *Shuf = Builder.CreateShuffleVector(
           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.
     return Builder.CreateExtractElement(TmpVec, Builder.getInt32(0));
   }
 };
 } // end anonymous namespace
 

test/Transforms/SLPVectorizer/X86/horizontal.ll

entry:
   store float %add.14, float* %res, align 16
   ret void
 }
 
 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
 ;
 entry:
   %0 = load i32, i32* getelementptr inbounds ([32 x i32], [32 x i32]* @arr_i32, i64 0, i64 0), align 16
   %1 = load i32, i32* getelementptr inbounds ([32 x i32], [32 x i32]* @arr_i32, i64 0, i64 1), align 4
   %add = add nsw i32 %1, %0
   %2 = load i32, i32* getelementptr inbounds ([32 x i32], [32 x i32]* @arr_i32, i64 0, i64 2), align 8
   %add.1 = add nsw i32 %2, %add
   %3 = load i32, i32* getelementptr inbounds ([32 x i32], [32 x i32]* @arr_i32, i64 0, i64 3), align 4
   %add.2 = add nsw i32 %3, %add.1
   store i32 %add.2, i32* %res, align 16
   ret void
 }
 
 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
 ;
 entry:
   %0 = load i32, i32* getelementptr inbounds ([32 x i32], [32 x i32]* @arr_i32, i64 0, i64 0), align 16
   %1 = load i32, i32* getelementptr inbounds ([32 x i32], [32 x i32]* @arr_i32, i64 0, i64 1), align 4
   %add = add nsw i32 %1, %0
entry:
   store i32 %add.6, i32* %res, align 16
   ret void
 }
 
 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
 ;
 entry:
   %0 = load i32, i32* getelementptr inbounds ([32 x i32], [32 x i32]* @arr_i32, i64 0, i64 0), align 16
   %1 = load i32, i32* getelementptr inbounds ([32 x i32], [32 x i32]* @arr_i32, i64 0, i64 1), align 4
   %add = add nsw i32 %1, %0
entry:
   store i32 %add.14, i32* %res, align 16
   ret void
 }
 
 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
 ;
 entry:
   %0 = load i32, i32* getelementptr inbounds ([32 x i32], [32 x i32]* @arr_i32, i64 0, i64 0), align 16
   %1 = load i32, i32* getelementptr inbounds ([32 x i32], [32 x i32]* @arr_i32, i64 0, i64 1), align 4
   %add = add nsw i32 %1, %0

test/Transforms/SLPVectorizer/X86/scheduling.ll

 ; NOTE: Assertions have been autogenerated by utils/update_test_checks.py
 ; RUN: opt < %s -basicaa -slp-vectorizer -S -mtriple=i386-apple-macosx10.8.0 -mcpu=corei7-avx | FileCheck %s
 
 define i32 @foo(i32* nocapture readonly %diff) #0 {
 ; CHECK-LABEL: @foo(
 ; CHECK: load <4 x i32>
 ; CHECK: load <4 x i32>
 ; 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-NEXT:    [[ADD52:%.*]] = add nsw i32 [[TMP15]],
 ; CHECK:          ret i32 [[ADD52]]
 ;
 entry:
   %m2 = alloca [8 x [8 x i32]], align 16
   %0 = bitcast [8 x [8 x i32]]* %m2 to i8*
   br label %for.body