// things. Simplifying them may result in a loss of legality.
if (LegalOperations) return SDValue();
- EVT VT = N->getValueType(0);
- assert(VT.isVector() && "SimplifyVBinOp only works on vectors!");
+ assert(N->getValueType(0).isVector() &&
+ "SimplifyVBinOp only works on vectors!");
- EVT EltType = VT.getVectorElementType();
SDValue LHS = N->getOperand(0);
SDValue RHS = N->getOperand(1);
SDValue Shuffle = XformToShuffleWithZero(N);
break;
}
- // If the vector element type is not legal, the BUILD_VECTOR operands
- // are promoted and implicitly truncated. Make that explicit here.
- if (LHSOp.getValueType() != EltType)
- LHSOp = DAG.getNode(ISD::TRUNCATE, LHS.getDebugLoc(), EltType, LHSOp);
- if (RHSOp.getValueType() != EltType)
- RHSOp = DAG.getNode(ISD::TRUNCATE, RHS.getDebugLoc(), EltType, RHSOp);
-
- SDValue FoldOp = DAG.getNode(N->getOpcode(), LHS.getDebugLoc(), EltType,
+ EVT VT = LHSOp.getValueType();
+ assert(RHSOp.getValueType() == VT &&
+ "SimplifyVBinOp with different BUILD_VECTOR element types");
+ SDValue FoldOp = DAG.getNode(N->getOpcode(), LHS.getDebugLoc(), VT,
LHSOp, RHSOp);
if (FoldOp.getOpcode() != ISD::UNDEF &&
FoldOp.getOpcode() != ISD::Constant &&
AddToWorkList(FoldOp.getNode());
}
- if (Ops.size() == LHS.getNumOperands()) {
- EVT VT = LHS.getValueType();
- return DAG.getNode(ISD::BUILD_VECTOR, N->getDebugLoc(), VT,
- &Ops[0], Ops.size());
- }
+ if (Ops.size() == LHS.getNumOperands())
+ return DAG.getNode(ISD::BUILD_VECTOR, N->getDebugLoc(),
+ LHS.getValueType(), &Ops[0], Ops.size());
}
return SDValue();
ret void
}
+; Test folding a binary vector operation with constant BUILD_VECTOR
+; operands with i16 elements.
+define void @test_i16_constant_fold() nounwind optsize {
+entry:
+ %0 = sext <4 x i1> zeroinitializer to <4 x i16>
+ %1 = add <4 x i16> %0, zeroinitializer
+ %2 = shufflevector <4 x i16> %1, <4 x i16> undef, <8 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7>
+ %3 = add <8 x i16> %2, <i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1>
+ %4 = trunc <8 x i16> %3 to <8 x i8>
+ tail call void @llvm.arm.neon.vst1.v8i8(i8* undef, <8 x i8> %4, i32 1)
+ unreachable
+}
+
+declare void @llvm.arm.neon.vst1.v8i8(i8*, <8 x i8>, i32) nounwind