diff options
Diffstat (limited to 'llvm/lib')
| -rw-r--r-- | llvm/lib/CodeGen/SelectionDAG/DAGCombiner.cpp | 2 | ||||
| -rw-r--r-- | llvm/lib/Target/X86/X86ISelLowering.cpp | 14 | ||||
| -rw-r--r-- | llvm/lib/Target/X86/X86ISelLowering.h | 3 |
3 files changed, 15 insertions, 4 deletions
diff --git a/llvm/lib/CodeGen/SelectionDAG/DAGCombiner.cpp b/llvm/lib/CodeGen/SelectionDAG/DAGCombiner.cpp index 185a9a72fa3..9d8850b59f5 100644 --- a/llvm/lib/CodeGen/SelectionDAG/DAGCombiner.cpp +++ b/llvm/lib/CodeGen/SelectionDAG/DAGCombiner.cpp @@ -3556,7 +3556,7 @@ SDValue DAGCombiner::visitMUL(SDNode *N) { // x * 15 --> (x << 4) - x // x * -33 --> -((x << 5) + x) // x * -15 --> -((x << 4) - x) ; this reduces --> x - (x << 4) - if (N1IsConst && TLI.decomposeMulByConstant(VT, N1)) { + if (N1IsConst && TLI.decomposeMulByConstant(*DAG.getContext(), VT, N1)) { // TODO: We could handle more general decomposition of any constant by // having the target set a limit on number of ops and making a // callback to determine that sequence (similar to sqrt expansion). diff --git a/llvm/lib/Target/X86/X86ISelLowering.cpp b/llvm/lib/Target/X86/X86ISelLowering.cpp index 4b8987b0a02..46b31894df7 100644 --- a/llvm/lib/Target/X86/X86ISelLowering.cpp +++ b/llvm/lib/Target/X86/X86ISelLowering.cpp @@ -4869,15 +4869,25 @@ bool X86TargetLowering::convertSelectOfConstantsToMath(EVT VT) const { return true; } -bool X86TargetLowering::decomposeMulByConstant(EVT VT, SDValue C) const { +bool X86TargetLowering::decomposeMulByConstant(LLVMContext &Context, EVT VT, + SDValue C) const { // TODO: We handle scalars using custom code, but generic combining could make // that unnecessary. APInt MulC; if (!ISD::isConstantSplatVector(C.getNode(), MulC)) return false; + // Find the type this will be legalized too. Otherwise we might prematurely + // convert this to shl+add/sub and then still have to type legalize those ops. + // Another choice would be to defer the decision for illegal types until + // after type legalization. But constant splat vectors of i64 can't make it + // through type legalization on 32-bit targets so we would need to special + // case vXi64. + while (getTypeAction(Context, VT) != TypeLegal) + VT = getTypeToTransformTo(Context, VT); + // If vector multiply is legal, assume that's faster than shl + add/sub. - // TODO: Multiply is a complex op with higher latency and lower througput in + // TODO: Multiply is a complex op with higher latency and lower throughput in // most implementations, so this check could be loosened based on type // and/or a CPU attribute. if (isOperationLegal(ISD::MUL, VT)) diff --git a/llvm/lib/Target/X86/X86ISelLowering.h b/llvm/lib/Target/X86/X86ISelLowering.h index 625b42d3515..8dc58a188dd 100644 --- a/llvm/lib/Target/X86/X86ISelLowering.h +++ b/llvm/lib/Target/X86/X86ISelLowering.h @@ -1105,7 +1105,8 @@ namespace llvm { bool convertSelectOfConstantsToMath(EVT VT) const override; - bool decomposeMulByConstant(EVT VT, SDValue C) const override; + bool decomposeMulByConstant(LLVMContext &Context, EVT VT, + SDValue C) const override; bool shouldUseStrictFP_TO_INT(EVT FpVT, EVT IntVT, bool IsSigned) const override; |
