};
bool VectorLegalizer::Run() {
+ // Before we start legalizing vector nodes, check if there are any vectors.
+ bool HasVectors = false;
+ for (SelectionDAG::allnodes_iterator I = DAG.allnodes_begin(),
+ E = prior(DAG.allnodes_end()); I != llvm::next(E); ++I) {
+ // Check if the values of the nodes contain vectors. We don't need to check
+ // the operands because we are going to check their values at some point.
+ for (SDNode::value_iterator J = I->value_begin(), E = I->value_end();
+ J != E; ++J)
+ HasVectors |= J->isVector();
+
+ // If we found a vector node we can start the legalization.
+ if (HasVectors)
+ break;
+ }
+
+ // If this basic block has no vectors then no need to legalize vectors.
+ if (!HasVectors)
+ return false;
+
// The legalize process is inherently a bottom-up recursive process (users
// legalize their uses before themselves). Given infinite stack space, we
// could just start legalizing on the root and traverse the whole graph. In
SDValue VectorLegalizer::ExpandVSELECT(SDValue Op) {
// Implement VSELECT in terms of XOR, AND, OR
// on platforms which do not support blend natively.
- EVT VT = Op.getOperand(0).getValueType();
DebugLoc DL = Op.getDebugLoc();
SDValue Mask = Op.getOperand(0);
SDValue Op1 = Op.getOperand(1);
SDValue Op2 = Op.getOperand(2);
+ EVT VT = Mask.getValueType();
+
// If we can't even use the basic vector operations of
// AND,OR,XOR, we will have to scalarize the op.
// Notice that the operation may be 'promoted' which means that it is
TargetLowering::ZeroOrNegativeOneBooleanContent)
return DAG.UnrollVectorOp(Op.getNode());
- assert(VT.getSizeInBits() == Op1.getValueType().getSizeInBits()
- && "Invalid mask size");
+ // If the mask and the type are different sizes, unroll the vector op. This
+ // can occur when getSetCCResultType returns something that is different in
+ // size from the operand types. For example, v4i8 = select v4i32, v4i8, v4i8.
+ if (VT.getSizeInBits() != Op1.getValueType().getSizeInBits())
+ return DAG.UnrollVectorOp(Op.getNode());
+
// Bitcast the operands to be the same type as the mask.
// This is needed when we select between FP types because
// the mask is a vector of integers.