[AVX] Add INSERT_SUBVECTOR and support it on x86. This provides a
default implementation for x86, going through the stack in a similr
fashion to how the codegen implements BUILD_VECTOR. Eventually this
will get matched to VINSERTF128 if AVX is available.
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@124307 91177308-0d34-0410-b5e6-96231b3b80d8
diff --git a/lib/CodeGen/SelectionDAG/LegalizeDAG.cpp b/lib/CodeGen/SelectionDAG/LegalizeDAG.cpp
index d06a8b4..6b7f83f 100644
--- a/lib/CodeGen/SelectionDAG/LegalizeDAG.cpp
+++ b/lib/CodeGen/SelectionDAG/LegalizeDAG.cpp
@@ -171,6 +171,7 @@
SDValue ExpandBitCount(unsigned Opc, SDValue Op, DebugLoc dl);
SDValue ExpandExtractFromVectorThroughStack(SDValue Op);
+ SDValue ExpandInsertToVectorThroughStack(SDValue Op);
SDValue ExpandVectorBuildThroughStack(SDNode* Node);
std::pair<SDValue, SDValue> ExpandAtomic(SDNode *Node);
@@ -1588,6 +1589,50 @@
false, false, 0);
}
+SDValue SelectionDAGLegalize::ExpandInsertToVectorThroughStack(SDValue Op) {
+ assert(Op.getValueType().isVector() && "Non-vector insert subvector!");
+
+ SDValue Vec = Op.getOperand(0);
+ SDValue Part = Op.getOperand(1);
+ SDValue Idx = Op.getOperand(2);
+ DebugLoc dl = Op.getDebugLoc();
+
+ // Store the value to a temporary stack slot, then LOAD the returned part.
+
+ SDValue StackPtr = DAG.CreateStackTemporary(Vec.getValueType());
+ int FI = cast<FrameIndexSDNode>(StackPtr.getNode())->getIndex();
+ MachinePointerInfo PtrInfo = MachinePointerInfo::getFixedStack(FI);
+
+ // First store the whole vector.
+ SDValue Ch = DAG.getStore(DAG.getEntryNode(), dl, Vec, StackPtr, PtrInfo,
+ false, false, 0);
+
+ // Then store the inserted part.
+
+ // Add the offset to the index.
+ unsigned EltSize =
+ Vec.getValueType().getVectorElementType().getSizeInBits()/8;
+
+ Idx = DAG.getNode(ISD::MUL, dl, Idx.getValueType(), Idx,
+ DAG.getConstant(EltSize, Idx.getValueType()));
+
+ if (Idx.getValueType().bitsGT(TLI.getPointerTy()))
+ Idx = DAG.getNode(ISD::TRUNCATE, dl, TLI.getPointerTy(), Idx);
+ else
+ Idx = DAG.getNode(ISD::ZERO_EXTEND, dl, TLI.getPointerTy(), Idx);
+
+ SDValue SubStackPtr = DAG.getNode(ISD::ADD, dl, Idx.getValueType(), Idx,
+ StackPtr);
+
+ // Store the subvector.
+ Ch = DAG.getStore(DAG.getEntryNode(), dl, Part, SubStackPtr,
+ MachinePointerInfo(), false, false, 0);
+
+ // Finally, load the updated vector.
+ return DAG.getLoad(Op.getValueType(), dl, Ch, StackPtr, PtrInfo,
+ false, false, 0);
+}
+
SDValue SelectionDAGLegalize::ExpandVectorBuildThroughStack(SDNode* Node) {
// We can't handle this case efficiently. Allocate a sufficiently
// aligned object on the stack, store each element into it, then load
@@ -2806,6 +2851,9 @@
case ISD::EXTRACT_SUBVECTOR:
Results.push_back(ExpandExtractFromVectorThroughStack(SDValue(Node, 0)));
break;
+ case ISD::INSERT_SUBVECTOR:
+ Results.push_back(ExpandInsertToVectorThroughStack(SDValue(Node, 0)));
+ break;
case ISD::CONCAT_VECTORS: {
Results.push_back(ExpandVectorBuildThroughStack(Node));
break;
diff --git a/lib/CodeGen/SelectionDAG/SelectionDAG.cpp b/lib/CodeGen/SelectionDAG/SelectionDAG.cpp
index 68d7621..661f810 100644
--- a/lib/CodeGen/SelectionDAG/SelectionDAG.cpp
+++ b/lib/CodeGen/SelectionDAG/SelectionDAG.cpp
@@ -3082,6 +3082,30 @@
case ISD::VECTOR_SHUFFLE:
llvm_unreachable("should use getVectorShuffle constructor!");
break;
+ case ISD::INSERT_SUBVECTOR: {
+ SDValue Index = N3;
+ if (VT.isSimple() && N1.getValueType().isSimple()
+ && N2.getValueType().isSimple()) {
+ assert(VT.isVector() && N1.getValueType().isVector() &&
+ N2.getValueType().isVector() &&
+ "Insert subvector VTs must be a vectors");
+ assert(VT == N1.getValueType() &&
+ "Dest and insert subvector source types must match!");
+ assert(N2.getValueType().getSimpleVT() <= N1.getValueType().getSimpleVT() &&
+ "Insert subvector must be from smaller vector to larger vector!");
+ if (ConstantSDNode *CSD = dyn_cast<ConstantSDNode>(Index.getNode())) {
+ uint64_t Idx = CSD->getZExtValue();
+ assert((N2.getValueType().getVectorNumElements() + Idx
+ <= VT.getVectorNumElements())
+ && "Insert subvector overflow!");
+ }
+
+ // Trivial insertion.
+ if (VT.getSimpleVT() == N2.getValueType().getSimpleVT())
+ return N2;
+ }
+ break;
+ }
case ISD::BITCAST:
// Fold bit_convert nodes from a type to themselves.
if (N1.getValueType() == VT)
@@ -5811,6 +5835,7 @@
case ISD::INSERT_VECTOR_ELT: return "insert_vector_elt";
case ISD::EXTRACT_VECTOR_ELT: return "extract_vector_elt";
case ISD::CONCAT_VECTORS: return "concat_vectors";
+ case ISD::INSERT_SUBVECTOR: return "insert_subvector";
case ISD::EXTRACT_SUBVECTOR: return "extract_subvector";
case ISD::SCALAR_TO_VECTOR: return "scalar_to_vector";
case ISD::VECTOR_SHUFFLE: return "vector_shuffle";