diff options
| author | Ahmed Bougacha <ahmed.bougacha@gmail.com> | 2015-03-17 03:23:09 +0000 |
|---|---|---|
| committer | Ahmed Bougacha <ahmed.bougacha@gmail.com> | 2015-03-17 03:23:09 +0000 |
| commit | e0afb1fe6c2facf927321f8f8e13851293df9796 (patch) | |
| tree | 93406fa6fcc767b9f2b1df045a26471a50bd7f4f /llvm/lib/Target/AArch64/AArch64ISelLowering.cpp | |
| parent | e33e6c979c0974368d69e9edad35843d4a39fcef (diff) | |
| download | bcm5719-llvm-e0afb1fe6c2facf927321f8f8e13851293df9796.tar.gz bcm5719-llvm-e0afb1fe6c2facf927321f8f8e13851293df9796.zip | |
[AArch64] Use intermediate step for concat_vectors of illegal truncs.
Optimize concat_vectors of truncated vectors, where the intermediate
type is illegal, to avoid said illegality, e.g.,
(v4i16 (concat_vectors (v2i16 (truncate (v2i64))),
(v2i16 (truncate (v2i64)))))
->
(v4i16 (truncate (v4i32 (concat_vectors (v2i32 (truncate (v2i64))),
(v2i32 (truncate (v2i64)))))))
This isn't really target-specific, and, as such, would best go in the
DAGCombiner. However, ISD::TRUNCATE legality isn't keyed on both input
and result type, so we might generate worse code when we don't know
better. On AArch64 we know it's fine for v2i64->v4i16 and v4i32->v8i8.
rdar://20022387
llvm-svn: 232459
Diffstat (limited to 'llvm/lib/Target/AArch64/AArch64ISelLowering.cpp')
| -rw-r--r-- | llvm/lib/Target/AArch64/AArch64ISelLowering.cpp | 31 |
1 files changed, 31 insertions, 0 deletions
diff --git a/llvm/lib/Target/AArch64/AArch64ISelLowering.cpp b/llvm/lib/Target/AArch64/AArch64ISelLowering.cpp index 43f443fc400..f1073beb32f 100644 --- a/llvm/lib/Target/AArch64/AArch64ISelLowering.cpp +++ b/llvm/lib/Target/AArch64/AArch64ISelLowering.cpp @@ -7180,6 +7180,37 @@ static SDValue performConcatVectorsCombine(SDNode *N, EVT VT = N->getValueType(0); SDValue N0 = N->getOperand(0), N1 = N->getOperand(1); + // Optimize concat_vectors of truncated vectors, where the intermediate + // type is illegal, to avoid said illegality, e.g., + // (v4i16 (concat_vectors (v2i16 (truncate (v2i64))), + // (v2i16 (truncate (v2i64))))) + // -> + // (v4i16 (truncate (v4i32 (concat_vectors (v2i32 (truncate (v2i64))), + // (v2i32 (truncate (v2i64))))))) + // This isn't really target-specific, but ISD::TRUNCATE legality isn't keyed + // on both input and result type, so we might generate worse code. + // On AArch64 we know it's fine for v2i64->v4i16 and v4i32->v8i8. + if (N->getNumOperands() == 2 && + N0->getOpcode() == ISD::TRUNCATE && + N1->getOpcode() == ISD::TRUNCATE) { + SDValue N00 = N0->getOperand(0); + SDValue N10 = N1->getOperand(0); + EVT N00VT = N00.getValueType(); + + if (N00VT == N10.getValueType() && + (N00VT == MVT::v2i64 || N00VT == MVT::v4i32) && + N00VT.getScalarSizeInBits() == 4 * VT.getScalarSizeInBits()) { + MVT MidVT = (N00VT == MVT::v2i64 ? MVT::v2i32 : MVT::v4i16); + MVT ConcatMidVT = MVT::getVectorVT(MidVT.getVectorElementType(), + MidVT.getVectorNumElements() * 2); + return DAG.getNode( + ISD::TRUNCATE, dl, VT, + DAG.getNode(ISD::CONCAT_VECTORS, dl, ConcatMidVT, + DAG.getNode(ISD::TRUNCATE, dl, MidVT, N00), + DAG.getNode(ISD::TRUNCATE, dl, MidVT, N10))); + } + } + // Wait 'til after everything is legalized to try this. That way we have // legal vector types and such. if (DCI.isBeforeLegalizeOps()) |

