diff --git a/lib/Target/R600/SIISelLowering.cpp b/lib/Target/R600/SIISelLowering.cpp
index 10e6c17f6cc..f042aaa70c0 100644
--- a/lib/Target/R600/SIISelLowering.cpp
+++ b/lib/Target/R600/SIISelLowering.cpp
@@ -1484,6 +1484,28 @@ SDValue SITargetLowering::PerformDAGCombine(SDNode *N,
 
         return DAG.getNode(AMDGPUISD::MAD, DL, VT, A, B, C);
       }
+
+      if (LHS.getOpcode() == ISD::FADD) {
+        // (fsub (fadd a, a), c) -> mad 2.0, a, (fneg c)
+
+        SDValue A = LHS.getOperand(0);
+        if (A == LHS.getOperand(1)) {
+          const SDValue Two = DAG.getTargetConstantFP(2.0, MVT::f32);
+          SDValue NegRHS = DAG.getNode(ISD::FNEG, DL, VT, RHS);
+
+          return DAG.getNode(AMDGPUISD::MAD, DL, VT, Two, A, NegRHS);
+        }
+      }
+
+      if (RHS.getOpcode() == ISD::FADD) {
+        // (fsub c, (fadd a, a)) -> mad -2.0, a, c
+
+        SDValue A = RHS.getOperand(0);
+        if (A == RHS.getOperand(1)) {
+          const SDValue NegTwo = DAG.getTargetConstantFP(-2.0, MVT::f32);
+          return DAG.getNode(AMDGPUISD::MAD, DL, VT, NegTwo, A, LHS);
+        }
+      }
     }
 
     break;
diff --git a/test/CodeGen/R600/fmuladd.ll b/test/CodeGen/R600/fmuladd.ll
index 6f581f21d9b..ea559c3ddc6 100644
--- a/test/CodeGen/R600/fmuladd.ll
+++ b/test/CodeGen/R600/fmuladd.ll
@@ -175,3 +175,25 @@ define void @fmuladd_2.0_neg_a_b_f32(float addrspace(1)* %out, float addrspace(1
   store float %r3, float addrspace(1)* %gep.out
   ret void
 }
+
+
+; CHECK-LABEL: @fmuladd_2.0_a_neg_b_f32
+; CHECK-DAG: BUFFER_LOAD_DWORD [[R1:v[0-9]+]], {{v\[[0-9]+:[0-9]+\]}}, {{s\[[0-9]+:[0-9]+\]}}, 0 addr64{{$}}
+; CHECK-DAG: BUFFER_LOAD_DWORD [[R2:v[0-9]+]], {{v\[[0-9]+:[0-9]+\]}}, {{s\[[0-9]+:[0-9]+\]}}, 0 addr64 offset:0x4
+; CHECK: V_MAD_F32 [[RESULT:v[0-9]+]], 2.0, [[R1]], -[[R2]]
+; CHECK: BUFFER_STORE_DWORD [[RESULT]]
+define void @fmuladd_2.0_a_neg_b_f32(float addrspace(1)* %out, float addrspace(1)* %in) {
+  %tid = call i32 @llvm.r600.read.tidig.x() nounwind readnone
+  %gep.0 = getelementptr float addrspace(1)* %out, i32 %tid
+  %gep.1 = getelementptr float addrspace(1)* %gep.0, i32 1
+  %gep.out = getelementptr float addrspace(1)* %out, i32 %tid
+
+  %r1 = load float addrspace(1)* %gep.0
+  %r2 = load float addrspace(1)* %gep.1
+
+  %r2.fneg = fsub float -0.000000e+00, %r2
+
+  %r3 = tail call float @llvm.fmuladd.f32(float 2.0, float %r1, float %r2.fneg)
+  store float %r3, float addrspace(1)* %gep.out
+  ret void
+}
diff --git a/test/CodeGen/R600/mad-sub.ll b/test/CodeGen/R600/mad-sub.ll
index 08069e2f4f1..444bb02b66c 100644
--- a/test/CodeGen/R600/mad-sub.ll
+++ b/test/CodeGen/R600/mad-sub.ll
@@ -169,5 +169,47 @@ define void @mad_fabs_sub_f32(float addrspace(1)* noalias nocapture %out, float
   ret void
 }
 
+; FUNC-LABEL: @fsub_c_fadd_a_a
+; SI-DAG: BUFFER_LOAD_DWORD [[R1:v[0-9]+]], {{v\[[0-9]+:[0-9]+\]}}, {{s\[[0-9]+:[0-9]+\]}}, 0 addr64{{$}}
+; SI-DAG: BUFFER_LOAD_DWORD [[R2:v[0-9]+]], {{v\[[0-9]+:[0-9]+\]}}, {{s\[[0-9]+:[0-9]+\]}}, 0 addr64 offset:0x4
+; SI: V_MAD_F32 [[RESULT:v[0-9]+]], -2.0, [[R1]], [[R2]]
+; SI: BUFFER_STORE_DWORD [[RESULT]]
+define void @fsub_c_fadd_a_a(float addrspace(1)* %out, float addrspace(1)* %in) {
+  %tid = call i32 @llvm.r600.read.tidig.x() nounwind readnone
+  %gep.0 = getelementptr float addrspace(1)* %out, i32 %tid
+  %gep.1 = getelementptr float addrspace(1)* %gep.0, i32 1
+  %gep.out = getelementptr float addrspace(1)* %out, i32 %tid
+
+  %r1 = load float addrspace(1)* %gep.0
+  %r2 = load float addrspace(1)* %gep.1
+
+  %add = fadd float %r1, %r1
+  %r3 = fsub float %r2, %add
+
+  store float %r3, float addrspace(1)* %gep.out
+  ret void
+}
+
+; FUNC-LABEL: @fsub_fadd_a_a_c
+; SI-DAG: BUFFER_LOAD_DWORD [[R1:v[0-9]+]], {{v\[[0-9]+:[0-9]+\]}}, {{s\[[0-9]+:[0-9]+\]}}, 0 addr64{{$}}
+; SI-DAG: BUFFER_LOAD_DWORD [[R2:v[0-9]+]], {{v\[[0-9]+:[0-9]+\]}}, {{s\[[0-9]+:[0-9]+\]}}, 0 addr64 offset:0x4
+; SI: V_MAD_F32 [[RESULT:v[0-9]+]], 2.0, [[R1]], -[[R2]]
+; SI: BUFFER_STORE_DWORD [[RESULT]]
+define void @fsub_fadd_a_a_c(float addrspace(1)* %out, float addrspace(1)* %in) {
+  %tid = call i32 @llvm.r600.read.tidig.x() nounwind readnone
+  %gep.0 = getelementptr float addrspace(1)* %out, i32 %tid
+  %gep.1 = getelementptr float addrspace(1)* %gep.0, i32 1
+  %gep.out = getelementptr float addrspace(1)* %out, i32 %tid
+
+  %r1 = load float addrspace(1)* %gep.0
+  %r2 = load float addrspace(1)* %gep.1
+
+  %add = fadd float %r1, %r1
+  %r3 = fsub float %add, %r2
+
+  store float %r3, float addrspace(1)* %gep.out
+  ret void
+}
+
 attributes #0 = { nounwind readnone }
 attributes #1 = { nounwind }