[SVE][CodeGen] Add simple integer add tests for SVE tuple types

[david-arm]

Summary

I have added tests to:

CodeGen/AArch64/sve-intrinsics-int-arith.ll

for doing simple integer add operations on tuple types. Since these
tests introduced new warnings due to incorrect use of
getVectorNumElements() I have also fixed up these warnings in the
same patch. These fixes are:

1. In narrowExtractedVectorBinOp I have changed the code to bail out

early for scalable vector types, since we've not yet hit a case that
proves the optimisations are profitable for scalable vectors.

2. In DAGTypeLegalizer::WidenVecRes_CONCAT_VECTORS I have replaced

calls to getVectorNumElements with getVectorMinNumElements in cases
that work with scalable vectors. For the other cases I have added
asserts that the vector is not scalable because we should not be
using shuffle vectors and build vectors in such cases.

[efriedma]

llvm/lib/CodeGen/SelectionDAG/DAGCombiner.cpp

19078

I think I committed a similar change today.

[david-arm]

Hi @efriedma I've rebased my patch now to pick up the same fix you did.

[efriedma]

LGTM

llvm/lib/CodeGen/SelectionDAG/DAGCombiner.cpp

SDValue DAGCombiner::visitCONCAT_VECTORS(SDNode *N) {
   // at the exact same location.
   //
   // For scalable vectors, EXTRACT_SUBVECTOR indexes are implicitly scaled.
   SDValue SingleSource = SDValue();
   unsigned PartNumElem =
       N->getOperand(0).getValueType().getVectorMinNumElements();
 
   for (unsigned i = 0, e = N->getNumOperands(); i != e; ++i) {
     SDValue Op = N->getOperand(i);
 
     if (Op.isUndef())
       continue;
 
     // Check if this is the identity extract:
     if (Op.getOpcode() != ISD::EXTRACT_SUBVECTOR)
       return SDValue();
 
static SDValue narrowExtractedVectorBinOp(SDNode *Extract, SelectionDAG &DAG) {
   if (BOpcode == ISD::FSUB) {
     auto *C = isConstOrConstSplatFP(BinOp.getOperand(0), /*AllowUndefs*/ true);
     if (C && C->getValueAPF().isNegZero())
       return SDValue();
   }
 
   // The binop must be a vector type, so we can extract some fraction of it.
   EVT WideBVT = BinOp.getValueType();
-  if (!WideBVT.isVector())
+  // The optimisations below currently assume we are dealing with fixed length
+  // vectors. It is possible to add support for scalable vectors, but at the
+  // moment we've done no analysis to prove whether they are profitable or not.
+  if (!WideBVT.isFixedLengthVector())
     return SDValue();
 
   EVT VT = Extract->getValueType(0);
   unsigned ExtractIndex = ExtractIndexC->getZExtValue();
   assert(ExtractIndex % VT.getVectorNumElements() == 0 &&
          "Extract index is not a multiple of the vector length.");
 
   // Bail out if this is not a proper multiple width extraction.

llvm/lib/CodeGen/SelectionDAG/LegalizeVectorTypes.cpp

SDValue DAGTypeLegalizer::WidenVecRes_BUILD_VECTOR(SDNode *N) {
 
   return DAG.getBuildVector(WidenVT, dl, NewOps);
 }
 
 SDValue DAGTypeLegalizer::WidenVecRes_CONCAT_VECTORS(SDNode *N) {
   EVT InVT = N->getOperand(0).getValueType();
   EVT WidenVT = TLI.getTypeToTransformTo(*DAG.getContext(), N->getValueType(0));
   SDLoc dl(N);
-  unsigned WidenNumElts = WidenVT.getVectorNumElements();
-  unsigned NumInElts = InVT.getVectorNumElements();
   unsigned NumOperands = N->getNumOperands();
 
   bool InputWidened = false; // Indicates we need to widen the input.
   if (getTypeAction(InVT) != TargetLowering::TypeWidenVector) {
-    if (WidenVT.getVectorNumElements() % InVT.getVectorNumElements() == 0) {
+    unsigned WidenNumElts = WidenVT.getVectorMinNumElements();
+    unsigned NumInElts = InVT.getVectorMinNumElements();
+    if (WidenNumElts % NumInElts == 0) {
       // Add undef vectors to widen to correct length.
-      unsigned NumConcat = WidenVT.getVectorNumElements() /
-                           InVT.getVectorNumElements();
+      unsigned NumConcat = WidenNumElts / NumInElts;
       SDValue UndefVal = DAG.getUNDEF(InVT);
       SmallVector<SDValue, 16> Ops(NumConcat);
       for (unsigned i=0; i < NumOperands; ++i)
         Ops[i] = N->getOperand(i);
       for (unsigned i = NumOperands; i != NumConcat; ++i)
         Ops[i] = UndefVal;
       return DAG.getNode(ISD::CONCAT_VECTORS, dl, WidenVT, Ops);
     }
   } else {
     InputWidened = true;
     if (WidenVT == TLI.getTypeToTransformTo(*DAG.getContext(), InVT)) {
       // The inputs and the result are widen to the same value.
       unsigned i;
       for (i=1; i < NumOperands; ++i)
         if (!N->getOperand(i).isUndef())
           break;
 
       if (i == NumOperands)
         // Everything but the first operand is an UNDEF so just return the
         // widened first operand.
         return GetWidenedVector(N->getOperand(0));
 
       if (NumOperands == 2) {
+        assert(!WidenVT.isScalableVector() &&
+               "Cannot use vector shuffles to widen CONCAT_VECTOR result");
+        unsigned WidenNumElts = WidenVT.getVectorNumElements();
+        unsigned NumInElts = InVT.getVectorNumElements();
+
         // Replace concat of two operands with a shuffle.
         SmallVector<int, 16> MaskOps(WidenNumElts, -1);
         for (unsigned i = 0; i < NumInElts; ++i) {
           MaskOps[i] = i;
           MaskOps[i + NumInElts] = i + WidenNumElts;
         }
         return DAG.getVectorShuffle(WidenVT, dl,
                                     GetWidenedVector(N->getOperand(0)),
                                     GetWidenedVector(N->getOperand(1)),
                                     MaskOps);
       }
     }
   }
 
+  assert(!WidenVT.isScalableVector() &&
+         "Cannot use build vectors to widen CONCAT_VECTOR result");
+  unsigned WidenNumElts = WidenVT.getVectorNumElements();
+  unsigned NumInElts = InVT.getVectorNumElements();
+
   // Fall back to use extracts and build vector.
   EVT EltVT = WidenVT.getVectorElementType();
   SmallVector<SDValue, 16> Ops(WidenNumElts);
   unsigned Idx = 0;
   for (unsigned i=0; i < NumOperands; ++i) {
     SDValue InOp = N->getOperand(i);
     if (InputWidened)
       InOp = GetWidenedVector(InOp);

llvm/test/CodeGen/AArch64/sve-intrinsics-int-arith.ll

 ; CHECK-LABEL: uqsub_i64:
 ; CHECK: uqsub z0.d, z0.d, z1.d
 ; CHECK-NEXT: ret
   %out = call <vscale x 2 x i64> @llvm.aarch64.sve.uqsub.x.nxv2i64(<vscale x 2 x i64> %a,
                                                                    <vscale x 2 x i64> %b)
   ret <vscale x 2 x i64> %out
 }
 
+; ADD (tuples)
+
+define <vscale x 4 x i64> @add_i64_tuple2(<vscale x 4 x i64>* %out, <vscale x 2 x i64> %in1, <vscale x 2 x i64> %in2) {
+; CHECK-LABEL: add_i64_tuple2
+; CHECK: add z0.d, z0.d, z0.d
+; CHECK: add z1.d, z1.d, z1.d
+  %tuple = tail call <vscale x 4 x i64> @llvm.aarch64.sve.tuple.create2.nxv4i64.nxv2i64(<vscale x 2 x i64> %in1, <vscale x 2 x i64> %in2)
+  %res = add <vscale x 4 x i64> %tuple, %tuple
+  ret <vscale x 4 x i64> %res
+}
+
+define <vscale x 6 x i64> @add_i64_tuple3(<vscale x 6 x i64>* %out, <vscale x 2 x i64> %in1, <vscale x 2 x i64> %in2, <vscale x 2 x i64> %in3) {
+; CHECK-LABEL: add_i64_tuple3
+; CHECK: add z0.d, z0.d, z0.d
+; CHECK: add z1.d, z1.d, z1.d
+; CHECK: add z2.d, z2.d, z2.d
+  %tuple = tail call <vscale x 6 x i64> @llvm.aarch64.sve.tuple.create3.nxv6i64.nxv2i64(<vscale x 2 x i64> %in1, <vscale x 2 x i64> %in2, <vscale x 2 x i64> %in3)
+  %res = add <vscale x 6 x i64> %tuple, %tuple
+  ret <vscale x 6 x i64> %res
+}
+
+define <vscale x 8 x i64> @add_i64_tuple4(<vscale x 8 x i64>* %out, <vscale x 2 x i64> %in1, <vscale x 2 x i64> %in2, <vscale x 2 x i64> %in3, <vscale x 2 x i64> %in4) {
+; CHECK-LABEL: add_i64_tuple4
+; CHECK: add z0.d, z0.d, z0.d
+; CHECK: add z1.d, z1.d, z1.d
+; CHECK: add z2.d, z2.d, z2.d
+; CHECK: add z3.d, z3.d, z3.d
+  %tuple = tail call <vscale x 8 x i64> @llvm.aarch64.sve.tuple.create4.nxv8i64.nxv2i64(<vscale x 2 x i64> %in1, <vscale x 2 x i64> %in2, <vscale x 2 x i64> %in3, <vscale x 2 x i64> %in4)
+  %res = add <vscale x 8 x i64> %tuple, %tuple
+  ret <vscale x 8 x i64> %res
+}
+
+
 declare <vscale x 16 x i8> @llvm.aarch64.sve.abs.nxv16i8(<vscale x 16 x i8>, <vscale x 16 x i1>, <vscale x 16 x i8>)
 declare <vscale x 8 x i16> @llvm.aarch64.sve.abs.nxv8i16(<vscale x 8 x i16>, <vscale x 8 x i1>, <vscale x 8 x i16>)
 declare <vscale x 4 x i32> @llvm.aarch64.sve.abs.nxv4i32(<vscale x 4 x i32>, <vscale x 4 x i1>, <vscale x 4 x i32>)
 declare <vscale x 2 x i64> @llvm.aarch64.sve.abs.nxv2i64(<vscale x 2 x i64>, <vscale x 2 x i1>, <vscale x 2 x i64>)
 
 declare <vscale x 16 x i8> @llvm.aarch64.sve.neg.nxv16i8(<vscale x 16 x i8>, <vscale x 16 x i1>, <vscale x 16 x i8>)
 declare <vscale x 8 x i16> @llvm.aarch64.sve.neg.nxv8i16(<vscale x 8 x i16>, <vscale x 8 x i1>, <vscale x 8 x i16>)
 declare <vscale x 4 x i32> @llvm.aarch64.sve.neg.nxv4i32(<vscale x 4 x i32>, <vscale x 4 x i1>, <vscale x 4 x i32>)

 declare <vscale x 8 x i16> @llvm.aarch64.sve.uqadd.x.nxv8i16(<vscale x 8 x i16>, <vscale x 8 x i16>)
 declare <vscale x 4 x i32> @llvm.aarch64.sve.uqadd.x.nxv4i32(<vscale x 4 x i32>, <vscale x 4 x i32>)
 declare <vscale x 2 x i64> @llvm.aarch64.sve.uqadd.x.nxv2i64(<vscale x 2 x i64>, <vscale x 2 x i64>)
 
 declare <vscale x 16 x i8> @llvm.aarch64.sve.uqsub.x.nxv16i8(<vscale x 16 x i8>, <vscale x 16 x i8>)
 declare <vscale x 8 x i16> @llvm.aarch64.sve.uqsub.x.nxv8i16(<vscale x 8 x i16>, <vscale x 8 x i16>)
 declare <vscale x 4 x i32> @llvm.aarch64.sve.uqsub.x.nxv4i32(<vscale x 4 x i32>, <vscale x 4 x i32>)
 declare <vscale x 2 x i64> @llvm.aarch64.sve.uqsub.x.nxv2i64(<vscale x 2 x i64>, <vscale x 2 x i64>)
+
+declare <vscale x 4 x i64> @llvm.aarch64.sve.tuple.create2.nxv4i64.nxv2i64(<vscale x 2 x i64>, <vscale x 2 x i64>)
+declare <vscale x 6 x i64> @llvm.aarch64.sve.tuple.create3.nxv6i64.nxv2i64(<vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>)
+declare <vscale x 8 x i64> @llvm.aarch64.sve.tuple.create4.nxv8i64.nxv2i64(<vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>)