+ // 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)));
+ }
+ }
+