[LoopVectorizer] Enable support for floating-point subtraction reductions

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@215200 91177308-0d34-0410-b5e6-96231b3b80d8

diff --git a/lib/Transforms/Vectorize/LoopVectorize.cpp b/lib/Transforms/Vectorize/LoopVectorize.cpp
index 9f6cfeca2c02b6d7d3609f5fb9cccfad4f671a12..1ab439f724a43427de8c6676cbff6a4e4783cbe8 100644 (file)
--- a/lib/Transforms/Vectorize/LoopVectorize.cpp
+++ b/lib/Transforms/Vectorize/LoopVectorize.cpp
@@ -5136,7 +5136,7 @@ LoopVectorizationLegality::isReductionInstr(Instruction *I,
                                             ReductionKind Kind,
                                             ReductionInstDesc &Prev) {
   bool FP = I->getType()->isFloatingPointTy();
-  bool FastMath = (FP && I->isCommutative() && I->isAssociative());
+  bool FastMath = FP && I->hasUnsafeAlgebra();
   switch (I->getOpcode()) {
   default:
     return ReductionInstDesc(false, I);
@@ -5158,6 +5158,7 @@ LoopVectorizationLegality::isReductionInstr(Instruction *I,
     return ReductionInstDesc(Kind == RK_IntegerXor, I);
   case Instruction::FMul:
     return ReductionInstDesc(Kind == RK_FloatMult && FastMath, I);
+  case Instruction::FSub:
   case Instruction::FAdd:
     return ReductionInstDesc(Kind == RK_FloatAdd && FastMath, I);
   case Instruction::FCmp:

diff --git a/test/Transforms/LoopVectorize/float-reduction.ll b/test/Transforms/LoopVectorize/float-reduction.ll
index 0dfbab07279ac0bb1277e8f0a78a60aef4e9ffaa..36eb992f0be151b309fbe5da69989b7af6f43bd3 100644 (file)
--- a/test/Transforms/LoopVectorize/float-reduction.ll
+++ b/test/Transforms/LoopVectorize/float-reduction.ll
@@ -23,3 +23,25 @@ for.body:                                         ; preds = %for.body, %entry
 for.end:                                          ; preds = %for.body
   ret float %add
 }
+
+;CHECK-LABEL: @foosub(
+;CHECK: fsub fast <4 x float>
+;CHECK: ret
+define float @foosub(float* nocapture %A, i32* nocapture %n) nounwind uwtable readonly ssp {
+entry:
+  br label %for.body
+
+for.body:                                         ; preds = %for.body, %entry
+  %indvars.iv = phi i64 [ 0, %entry ], [ %indvars.iv.next, %for.body ]
+  %sum.04 = phi float [ 0.000000e+00, %entry ], [ %sub, %for.body ]
+  %arrayidx = getelementptr inbounds float* %A, i64 %indvars.iv
+  %0 = load float* %arrayidx, align 4
+  %sub = fsub fast float %sum.04, %0
+  %indvars.iv.next = add i64 %indvars.iv, 1
+  %lftr.wideiv = trunc i64 %indvars.iv.next to i32
+  %exitcond = icmp eq i32 %lftr.wideiv, 200
+  br i1 %exitcond, label %for.end, label %for.body
+
+for.end:                                          ; preds = %for.body
+  ret float %sub
+}