Add support for lowering fneg when AVX is enabled.

rdar://10566486

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@146625 91177308-0d34-0410-b5e6-96231b3b80d8

diff --git a/lib/Target/X86/X86ISelLowering.cpp b/lib/Target/X86/X86ISelLowering.cpp
index e38de9c0d226200b1b6a72a84a41c1e00429c169..95421fb15ee23ff70849b4a34fc53062362666e9 100644 (file)
--- a/lib/Target/X86/X86ISelLowering.cpp
+++ b/lib/Target/X86/X86ISelLowering.cpp
@@ -7910,19 +7910,18 @@ SDValue X86TargetLowering::LowerFNEG(SDValue Op, SelectionDAG &DAG) const {
   DebugLoc dl = Op.getDebugLoc();
   EVT VT = Op.getValueType();
   EVT EltVT = VT;
-  if (VT.isVector())
+  unsigned NumElts = VT == MVT::f64 ? 2 : 4;
+  if (VT.isVector()) {
     EltVT = VT.getVectorElementType();
-  std::vector<Constant*> CV;
+    NumElts = VT.getVectorNumElements();
+  }
+  SmallVector<Constant*,8> CV;
   if (EltVT == MVT::f64) {
     Constant *C = ConstantFP::get(*Context, APFloat(APInt(64, 1ULL << 63)));
-    CV.push_back(C);
-    CV.push_back(C);
+    CV.assign(NumElts, C);
   } else {
     Constant *C = ConstantFP::get(*Context, APFloat(APInt(32, 1U << 31)));
-    CV.push_back(C);
-    CV.push_back(C);
-    CV.push_back(C);
-    CV.push_back(C);
+    CV.assign(NumElts, C);
   }
   Constant *C = ConstantVector::get(CV);
   SDValue CPIdx = DAG.getConstantPool(C, getPointerTy(), 16);
@@ -7930,11 +7929,12 @@ SDValue X86TargetLowering::LowerFNEG(SDValue Op, SelectionDAG &DAG) const {
                              MachinePointerInfo::getConstantPool(),
                              false, false, false, 16);
   if (VT.isVector()) {
+    MVT XORVT = VT.getSizeInBits() == 128 ? MVT::v2i64 : MVT::v4i64;
     return DAG.getNode(ISD::BITCAST, dl, VT,
-                       DAG.getNode(ISD::XOR, dl, MVT::v2i64,
-                    DAG.getNode(ISD::BITCAST, dl, MVT::v2i64,
+                       DAG.getNode(ISD::XOR, dl, XORVT,
+                    DAG.getNode(ISD::BITCAST, dl, XORVT,
                                 Op.getOperand(0)),
-                    DAG.getNode(ISD::BITCAST, dl, MVT::v2i64, Mask)));
+                    DAG.getNode(ISD::BITCAST, dl, XORVT, Mask)));
   } else {
     return DAG.getNode(X86ISD::FXOR, dl, VT, Op.getOperand(0), Mask);
   }

diff --git a/test/CodeGen/X86/avx-basic.ll b/test/CodeGen/X86/avx-basic.ll
index edbdc06a1c94be3d929c2d09fac2211fe1f19505..78026069df6e26aaf5f9e5d6239b89e01c26b325 100644 (file)
--- a/test/CodeGen/X86/avx-basic.ll
+++ b/test/CodeGen/X86/avx-basic.ll
@@ -105,3 +105,11 @@ allocas:
   ret <8 x i32> %updatedret.i30.i
 }
 
+;;;; Don't crash on fneg
+; rdar://10566486
+; CHECK: fneg
+; CHECK: vxorps
+define <16 x float> @fneg(<16 x float> addrspace(1)* nocapture %out) nounwind {
+  %1 = fsub <16 x float> <float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00>, <float 1.000000e+00, float 1.000000e+00, float 1.000000e+00, float 1.000000e+00, float 1.000000e+00, float 1.000000e+00, float 1.000000e+00, float 1.000000e+00, float 1.000000e+00, float 1.000000e+00, float 1.000000e+00, float 1.000000e+00, float 1.000000e+00, float 1.000000e+00, float 1.000000e+00, float 1.000000e+00>
+  ret <16 x float> %1
+}