[ARM] Add instruction selection patterns for vmin/vmax

Summary:
The mid-end was generating vector smin/smax/umin/umax nodes, but
we were using vbsl to generatate the code. This adds the vmin/vmax
patterns and a test to check that we are now generating vmin/vmax
instructions.

Reviewers: rengolin, jmolloy

Subscribers: aemerson, rengolin, llvm-commits

Differential Revision: http://reviews.llvm.org/D12105

llvm-svn: 245439

2 files changed, 24 insertions, 6 deletions

diff --git a/llvm/lib/Target/ARM/ARMISelLowering.cpp b/llvm/lib/Target/ARM/ARMISelLowering.cpp
index fe8da7ffc80..ebc63e0db24 100644
--- a/llvm/lib/Target/ARM/ARMISelLowering.cpp
+++ b/llvm/lib/Target/ARM/ARMISelLowering.cpp
@@ -147,6 +147,11 @@ void ARMTargetLowering::addTypeForNEON(MVT VT, MVT PromotedLdStVT,
     setOperationAction(ISD::SABSDIFF, VT, Legal);
     setOperationAction(ISD::UABSDIFF, VT, Legal);
   }
+  if (!VT.isFloatingPoint() &&
+      VT != MVT::v2i64 && VT != MVT::v1i64)
+    for (unsigned Opcode : {ISD::SMIN, ISD::SMAX, ISD::UMIN, ISD::UMAX})
+      setOperationAction(Opcode, VT, Legal);
+
 }
 
 void ARMTargetLowering::addDRTypeForNEON(MVT VT) {
@@ -2821,11 +2826,24 @@ ARMTargetLowering::LowerINTRINSIC_WO_CHAIN(SDValue Op, SelectionDAG &DAG,
     return DAG.getNode(NewOpc, SDLoc(Op), Op.getValueType(),
                        Op.getOperand(1), Op.getOperand(2));
   }
+  case Intrinsic::arm_neon_vminu:
+  case Intrinsic::arm_neon_vmaxu: {
+    if (Op.getValueType().isFloatingPoint())
+      return SDValue();
+    unsigned NewOpc = (IntNo == Intrinsic::arm_neon_vminu)
+      ? ISD::UMIN : ISD::UMAX;
+    return DAG.getNode(NewOpc, SDLoc(Op), Op.getValueType(),
+                         Op.getOperand(1), Op.getOperand(2));
+  }
   case Intrinsic::arm_neon_vmins:
   case Intrinsic::arm_neon_vmaxs: {
     // v{min,max}s is overloaded between signed integers and floats.
-    if (!Op.getValueType().isFloatingPoint())
-      return SDValue();
+    if (!Op.getValueType().isFloatingPoint()) {
+      unsigned NewOpc = (IntNo == Intrinsic::arm_neon_vmins)
+        ? ISD::SMIN : ISD::SMAX;
+      return DAG.getNode(NewOpc, SDLoc(Op), Op.getValueType(),
+                         Op.getOperand(1), Op.getOperand(2));
+    }
     unsigned NewOpc = (IntNo == Intrinsic::arm_neon_vmins)
       ? ISD::FMINNAN : ISD::FMAXNAN;
     return DAG.getNode(NewOpc, SDLoc(Op), Op.getValueType(),
diff --git a/llvm/lib/Target/ARM/ARMInstrNEON.td b/llvm/lib/Target/ARM/ARMInstrNEON.td
index 889b8a97d25..5e9e3876fe3 100644
--- a/llvm/lib/Target/ARM/ARMInstrNEON.td
+++ b/llvm/lib/Target/ARM/ARMInstrNEON.td
@@ -5026,10 +5026,10 @@ defm VABALu   : N3VLIntExtOp_QHS<1,1,0b0101,0, IIC_VABAD,
 //   VMAX     : Vector Maximum
 defm VMAXs    : N3VInt_QHS<0, 0, 0b0110, 0, N3RegFrm,
                            IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q, IIC_VSUBi4Q,
-                           "vmax", "s", int_arm_neon_vmaxs, 1>;
+                           "vmax", "s", smax, 1>;
 defm VMAXu    : N3VInt_QHS<1, 0, 0b0110, 0, N3RegFrm,
                            IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q, IIC_VSUBi4Q,
-                           "vmax", "u", int_arm_neon_vmaxu, 1>;
+                           "vmax", "u", umax, 1>;
 def  VMAXfd   : N3VDInt<0, 0, 0b00, 0b1111, 0, N3RegFrm, IIC_VBIND,
                         "vmax", "f32",
                         v2f32, v2f32, fmaxnan, 1>;
@@ -5052,10 +5052,10 @@ let PostEncoderMethod = "NEONThumb2V8PostEncoder", DecoderNamespace = "v8NEON" i
 //   VMIN     : Vector Minimum
 defm VMINs    : N3VInt_QHS<0, 0, 0b0110, 1, N3RegFrm,
                            IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q, IIC_VSUBi4Q,
-                           "vmin", "s", int_arm_neon_vmins, 1>;
+                           "vmin", "s", smin, 1>;
 defm VMINu    : N3VInt_QHS<1, 0, 0b0110, 1, N3RegFrm,
                            IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q, IIC_VSUBi4Q,
-                           "vmin", "u", int_arm_neon_vminu, 1>;
+                           "vmin", "u", umin, 1>;
 def  VMINfd   : N3VDInt<0, 0, 0b10, 0b1111, 0, N3RegFrm, IIC_VBIND,
                         "vmin", "f32",
                         v2f32, v2f32, fminnan, 1>;