This patch takes some of the code from D49837 to allow us to enable ISD::ABS support for all SSE vector types.
|46 ↗||(On Diff #181048)|
This sequence is legal, but it isn't actually a good idea; the new sequence is one instruction longer, and has much higher latency.
AArch64TargetLowering::LowerINTRINSIC_WO_CHAIN should probably generate a v1i64 ISD::ABS, not an i64 ISD::ABS.
|758 ↗||(On Diff #181156)|
Need to check ISD::ABS in the switch in VectorLegalizer::LegalizeOp that controls the call to TLI.getOperationAction. Otherwise its not considered a vector op and gets delayed to LegalizeDAG.