diff options
Diffstat (limited to 'llvm/lib/Target/AMDGPU/AMDGPUTargetTransformInfo.cpp')
| -rw-r--r-- | llvm/lib/Target/AMDGPU/AMDGPUTargetTransformInfo.cpp | 12 |
1 files changed, 11 insertions, 1 deletions
diff --git a/llvm/lib/Target/AMDGPU/AMDGPUTargetTransformInfo.cpp b/llvm/lib/Target/AMDGPU/AMDGPUTargetTransformInfo.cpp index 81d083c1c88a..cb877a4695f1 100644 --- a/llvm/lib/Target/AMDGPU/AMDGPUTargetTransformInfo.cpp +++ b/llvm/lib/Target/AMDGPU/AMDGPUTargetTransformInfo.cpp @@ -650,6 +650,15 @@ InstructionCost GCNTTIImpl::getArithmeticInstrCost( return LT.first * Cost * NElts; } + if (SLT == MVT::f32 && ((CxtI && CxtI->hasApproxFunc()) || + TLI->getTargetMachine().Options.UnsafeFPMath)) { + // Fast unsafe fdiv lowering: + // f32 rcp + // f32 fmul + int Cost = getQuarterRateInstrCost(CostKind) + getFullRateInstrCost(); + return LT.first * Cost * NElts; + } + if (SLT == MVT::f32 || SLT == MVT::f16) { // 4 more v_cvt_* insts without f16 insts support int Cost = (SLT == MVT::f16 ? 14 : 10) * getFullRateInstrCost() + @@ -1114,7 +1123,8 @@ InstructionCost GCNTTIImpl::getShuffleCost(TTI::ShuffleKind Kind, TTI::TargetCostKind CostKind, int Index, VectorType *SubTp, ArrayRef<const Value *> Args) { - Kind = improveShuffleKindFromMask(Kind, Mask); + Kind = improveShuffleKindFromMask(Kind, Mask, VT, Index, SubTp); + if (ST->hasVOP3PInsts()) { if (cast<FixedVectorType>(VT)->getNumElements() == 2 && DL.getTypeSizeInBits(VT->getElementType()) == 16) { |