diff options
Diffstat (limited to 'llvm')
| -rw-r--r-- | llvm/lib/Transforms/Vectorize/LoopVectorize.cpp | 3 | ||||
| -rw-r--r-- | llvm/test/Transforms/LoopVectorize/float-reduction.ll | 22 |
2 files changed, 24 insertions, 1 deletions
diff --git a/llvm/lib/Transforms/Vectorize/LoopVectorize.cpp b/llvm/lib/Transforms/Vectorize/LoopVectorize.cpp index 9f6cfeca2c0..1ab439f724a 100644 --- a/llvm/lib/Transforms/Vectorize/LoopVectorize.cpp +++ b/llvm/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/llvm/test/Transforms/LoopVectorize/float-reduction.ll b/llvm/test/Transforms/LoopVectorize/float-reduction.ll index 0dfbab07279..36eb992f0be 100644 --- a/llvm/test/Transforms/LoopVectorize/float-reduction.ll +++ b/llvm/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 +} |
