Add support for lowering FABS of vector types.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@163461 91177308-0d34-0410-b5e6-96231b3b80d8

2 files changed, 63 insertions, 12 deletions

diff --git a/lib/Target/X86/X86ISelLowering.cpp b/lib/Target/X86/X86ISelLowering.cpp
index 761fe62..8dccce0 100644
--- a/lib/Target/X86/X86ISelLowering.cpp
+++ b/lib/Target/X86/X86ISelLowering.cpp
@@ -829,6 +829,7 @@ X86TargetLowering::X86TargetLowering(X86TargetMachine &TM)
     setOperationAction(ISD::FDIV,               MVT::v4f32, Legal);
     setOperationAction(ISD::FSQRT,              MVT::v4f32, Legal);
     setOperationAction(ISD::FNEG,               MVT::v4f32, Custom);
+    setOperationAction(ISD::FABS,               MVT::v4f32, Custom);
     setOperationAction(ISD::LOAD,               MVT::v4f32, Legal);
     setOperationAction(ISD::BUILD_VECTOR,       MVT::v4f32, Custom);
     setOperationAction(ISD::VECTOR_SHUFFLE,     MVT::v4f32, Custom);
@@ -862,6 +863,7 @@ X86TargetLowering::X86TargetLowering(X86TargetMachine &TM)
     setOperationAction(ISD::FDIV,               MVT::v2f64, Legal);
     setOperationAction(ISD::FSQRT,              MVT::v2f64, Legal);
     setOperationAction(ISD::FNEG,               MVT::v2f64, Custom);
+    setOperationAction(ISD::FABS,               MVT::v2f64, Custom);
 
     setOperationAction(ISD::SETCC,              MVT::v2i64, Custom);
     setOperationAction(ISD::SETCC,              MVT::v16i8, Custom);
@@ -1022,6 +1024,7 @@ X86TargetLowering::X86TargetLowering(X86TargetMachine &TM)
     setOperationAction(ISD::FDIV,               MVT::v8f32, Legal);
     setOperationAction(ISD::FSQRT,              MVT::v8f32, Legal);
     setOperationAction(ISD::FNEG,               MVT::v8f32, Custom);
+    setOperationAction(ISD::FABS,               MVT::v8f32, Custom);
 
     setOperationAction(ISD::FADD,               MVT::v4f64, Legal);
     setOperationAction(ISD::FSUB,               MVT::v4f64, Legal);
@@ -1029,6 +1032,7 @@ X86TargetLowering::X86TargetLowering(X86TargetMachine &TM)
     setOperationAction(ISD::FDIV,               MVT::v4f64, Legal);
     setOperationAction(ISD::FSQRT,              MVT::v4f64, Legal);
     setOperationAction(ISD::FNEG,               MVT::v4f64, Custom);
+    setOperationAction(ISD::FABS,               MVT::v4f64, Custom);
 
     setOperationAction(ISD::FP_TO_SINT,         MVT::v8i32, Legal);
     setOperationAction(ISD::SINT_TO_FP,         MVT::v8i32, Legal);
@@ -8175,26 +8179,35 @@ SDValue X86TargetLowering::LowerFP_TO_UINT(SDValue Op,
   return FIST;
 }
 
-SDValue X86TargetLowering::LowerFABS(SDValue Op,
-                                     SelectionDAG &DAG) const {
+SDValue X86TargetLowering::LowerFABS(SDValue Op, SelectionDAG &DAG) const {
   LLVMContext *Context = DAG.getContext();
   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();
-  Constant *C;
-  if (EltVT == MVT::f64) {
-    C = ConstantVector::getSplat(2,
-                ConstantFP::get(*Context, APFloat(APInt(64, ~(1ULL << 63)))));
-  } else {
-    C = ConstantVector::getSplat(4,
-               ConstantFP::get(*Context, APFloat(APInt(32, ~(1U << 31)))));
+    NumElts = VT.getVectorNumElements();
   }
-  SDValue CPIdx = DAG.getConstantPool(C, getPointerTy(), 16);
+  Constant *C;
+  if (EltVT == MVT::f64)
+    C = ConstantFP::get(*Context, APFloat(APInt(64, ~(1ULL << 63))));
+  else
+    C = ConstantFP::get(*Context, APFloat(APInt(32, ~(1U << 31))));
+  C = ConstantVector::getSplat(NumElts, C);
+  SDValue CPIdx = DAG.getConstantPool(C, getPointerTy());
+  unsigned Alignment = cast<ConstantPoolSDNode>(CPIdx)->getAlignment();
   SDValue Mask = DAG.getLoad(VT, dl, DAG.getEntryNode(), CPIdx,
                              MachinePointerInfo::getConstantPool(),
-                             false, false, false, 16);
+                             false, false, false, Alignment);
+  if (VT.isVector()) {
+    MVT ANDVT = VT.is128BitVector() ? MVT::v2i64 : MVT::v4i64;
+    return DAG.getNode(ISD::BITCAST, dl, VT,
+                       DAG.getNode(ISD::AND, dl, ANDVT,
+                                   DAG.getNode(ISD::BITCAST, dl, ANDVT,
+                                               Op.getOperand(0)),
+                                   DAG.getNode(ISD::BITCAST, dl, ANDVT, Mask)));
+  }
   return DAG.getNode(X86ISD::FAND, dl, VT, Op.getOperand(0), Mask);
 }
 
diff --git a/test/CodeGen/X86/vec_fabs.ll b/test/CodeGen/X86/vec_fabs.ll
new file mode 100644
index 0000000..82517cb
--- /dev/null
+++ b/test/CodeGen/X86/vec_fabs.ll
@@ -0,0 +1,38 @@
+; RUN: llc < %s -mtriple=x86_64-apple-darwin -march=x86 -mcpu=corei7-avx | FileCheck %s
+
+
+define <2 x double> @fabs_v2f64(<2 x double> %p)
+{
+  ; CHECK: fabs_v2f64
+  ; CHECK: vandps
+  %t = call <2 x double> @llvm.fabs.v2f64(<2 x double> %p)
+  ret <2 x double> %t
+}
+declare <2 x double> @llvm.fabs.v2f64(<2 x double> %p)
+
+define <4 x float> @fabs_v4f32(<4 x float> %p)
+{
+  ; CHECK: fabs_v4f32
+  ; CHECK: vandps
+  %t = call <4 x float> @llvm.fabs.v4f32(<4 x float> %p)
+  ret <4 x float> %t
+}
+declare <4 x float> @llvm.fabs.v4f32(<4 x float> %p)
+
+define <4 x double> @fabs_v4f64(<4 x double> %p)
+{
+  ; CHECK: fabs_v4f64
+  ; CHECK: vandps
+  %t = call <4 x double> @llvm.fabs.v4f64(<4 x double> %p)
+  ret <4 x double> %t
+}
+declare <4 x double> @llvm.fabs.v4f64(<4 x double> %p)
+
+define <8 x float> @fabs_v8f32(<8 x float> %p)
+{
+  ; CHECK: fabs_v8f32
+  ; CHECK: vandps
+  %t = call <8 x float> @llvm.fabs.v8f32(<8 x float> %p)
+  ret <8 x float> %t
+}
+declare <8 x float> @llvm.fabs.v8f32(<8 x float> %p)