1 files changed, 21 insertions, 12 deletions

diff --git a/lib/Target/R600/AMDGPUISelDAGToDAG.cpp b/lib/Target/R600/AMDGPUISelDAGToDAG.cpp
index ce498e79e2..85d0f9dd69 100644
--- a/lib/Target/R600/AMDGPUISelDAGToDAG.cpp
+++ b/lib/Target/R600/AMDGPUISelDAGToDAG.cpp
@@ -85,7 +85,7 @@ private:
   bool SelectADDRVTX_READ(SDValue Addr, SDValue &Base, SDValue &Offset);
   bool SelectADDRIndirect(SDValue Addr, SDValue &Base, SDValue &Offset);
 
-  SDNode *SelectADD_I64(SDNode *N);
+  SDNode *SelectADD_SUB_I64(SDNode *N);
 
   // Include the pieces autogenerated from the target description.
 #include "AMDGPUGenDAGISel.inc"
@@ -214,12 +214,13 @@ SDNode *AMDGPUDAGToDAGISel::Select(SDNode *N) {
   // We are selecting i64 ADD here instead of custom lower it during
   // DAG legalization, so we can fold some i64 ADDs used for address
   // calculation into the LOAD and STORE instructions.
-  case ISD::ADD: {
+  case ISD::ADD:
+  case ISD::SUB: {
     if (N->getValueType(0) != MVT::i64 ||
         ST.getGeneration() < AMDGPUSubtarget::SOUTHERN_ISLANDS)
       break;
 
-    return SelectADD_I64(N);
+    return SelectADD_SUB_I64(N);
   }
   case ISD::SCALAR_TO_VECTOR:
   case AMDGPUISD::BUILD_VERTICAL_VECTOR:
@@ -646,11 +647,13 @@ bool AMDGPUDAGToDAGISel::SelectADDRIndirect(SDValue Addr, SDValue &Base,
   return true;
 }
 
-SDNode *AMDGPUDAGToDAGISel::SelectADD_I64(SDNode *N) {
+SDNode *AMDGPUDAGToDAGISel::SelectADD_SUB_I64(SDNode *N) {
   SDLoc DL(N);
   SDValue LHS = N->getOperand(0);
   SDValue RHS = N->getOperand(1);
 
+  bool IsAdd = (N->getOpcode() == ISD::ADD);
+
   SDValue Sub0 = CurDAG->getTargetConstant(AMDGPU::sub0, MVT::i32);
   SDValue Sub1 = CurDAG->getTargetConstant(AMDGPU::sub1, MVT::i32);
 
@@ -665,16 +668,22 @@ SDNode *AMDGPUDAGToDAGISel::SelectADD_I64(SDNode *N) {
                                        DL, MVT::i32, RHS, Sub1);
 
   SDVTList VTList = CurDAG->getVTList(MVT::i32, MVT::Glue);
-
   SDValue AddLoArgs[] = { SDValue(Lo0, 0), SDValue(Lo1, 0) };
 
-  SDNode *AddLo = CurDAG->getMachineNode(
-    isCFDepth0() ? AMDGPU::S_ADD_I32 : AMDGPU::V_ADD_I32_e32,
-    DL, VTList, AddLoArgs);
-  SDValue Carry = SDValue(AddLo, 1);
-  SDNode *AddHi = CurDAG->getMachineNode(
-    isCFDepth0() ? AMDGPU::S_ADDC_U32 : AMDGPU::V_ADDC_U32_e32,
-    DL, MVT::i32, SDValue(Hi0, 0), SDValue(Hi1, 0), Carry);
+
+  unsigned Opc = IsAdd ? AMDGPU::S_ADD_I32 : AMDGPU::S_SUB_I32;
+  unsigned CarryOpc = IsAdd ? AMDGPU::S_ADDC_U32 : AMDGPU::S_SUBB_U32;
+
+  if (!isCFDepth0()) {
+    Opc = IsAdd ? AMDGPU::V_ADD_I32_e32 : AMDGPU::V_SUB_I32_e32;
+    CarryOpc = IsAdd ? AMDGPU::V_ADDC_U32_e32 : AMDGPU::V_SUBB_U32_e32;
+  }
+
+  SDNode *AddLo = CurDAG->getMachineNode( Opc, DL, VTList, AddLoArgs);
+  SDValue Carry(AddLo, 1);
+  SDNode *AddHi
+    = CurDAG->getMachineNode(CarryOpc, DL, MVT::i32,
+                             SDValue(Hi0, 0), SDValue(Hi1, 0), Carry);
 
   SDValue Args[5] = {
     CurDAG->getTargetConstant(AMDGPU::SReg_64RegClassID, MVT::i32),