Index: include/llvm/IR/IntrinsicsAMDGPU.td
===================================================================
--- include/llvm/IR/IntrinsicsAMDGPU.td
+++ include/llvm/IR/IntrinsicsAMDGPU.td
@@ -65,6 +65,9 @@
 
 let TargetPrefix = "amdgcn" in {
 
+def int_amdgcn_s_barrier : GCCBuiltin<"__builtin_amdgcn_s_barrier">,
+  Intrinsic<[], [], [IntrConvergent]>;
+
 def int_amdgcn_div_scale : GCCBuiltin<"__builtin_amdgcn_div_scale">,
   // 1st parameter: Numerator
   // 2nd parameter: Denominator
Index: lib/Target/AMDGPU/SIInstructions.td
===================================================================
--- lib/Target/AMDGPU/SIInstructions.td
+++ lib/Target/AMDGPU/SIInstructions.td
@@ -481,7 +481,7 @@
 
 let hasSideEffects = 1 in {
 def S_BARRIER : SOPP <0x0000000a, (ins), "s_barrier",
-  [(int_AMDGPU_barrier_local)]
+  [(int_amdgcn_s_barrier)]
 > {
   let SchedRW = [WriteBarrier];
   let simm16 = 0;
@@ -2167,11 +2167,17 @@
 // SOPP Patterns
 //===----------------------------------------------------------------------===//
 
+// FIXME: These should be removed eventually
 def : Pat <
   (int_AMDGPU_barrier_global),
   (S_BARRIER)
 >;
 
+def : Pat <
+  (int_AMDGPU_barrier_local),
+  (S_BARRIER)
+>;
+
 //===----------------------------------------------------------------------===//
 // VOP1 Patterns
 //===----------------------------------------------------------------------===//
Index: test/CodeGen/AMDGPU/llvm.amdgcn.s.barrier.ll
===================================================================
--- /dev/null
+++ test/CodeGen/AMDGPU/llvm.amdgcn.s.barrier.ll
@@ -0,0 +1,28 @@
+; RUN: llc -march=amdgcn -verify-machineinstrs < %s | FileCheck -check-prefix=GCN %s
+
+; GCN-LABEL: {{^}}test_barrier:
+; GCN: buffer_store_dword
+; GCN: s_waitcnt
+; GCN: s_barrier
+define void @test_barrier(i32 addrspace(1)* %out) #0 {
+entry:
+  %tmp = call i32 @llvm.r600.read.tidig.x()
+  %tmp1 = getelementptr i32, i32 addrspace(1)* %out, i32 %tmp
+  store i32 %tmp, i32 addrspace(1)* %tmp1
+  call void @llvm.amdgcn.s.barrier()
+  %tmp2 = call i32 @llvm.r600.read.local.size.x()
+  %tmp3 = sub i32 %tmp2, 1
+  %tmp4 = sub i32 %tmp3, %tmp
+  %tmp5 = getelementptr i32, i32 addrspace(1)* %out, i32 %tmp4
+  %tmp6 = load i32, i32 addrspace(1)* %tmp5
+  store i32 %tmp6, i32 addrspace(1)* %tmp1
+  ret void
+}
+
+declare void @llvm.amdgcn.s.barrier() #1
+declare i32 @llvm.r600.read.tidig.x() #2
+declare i32 @llvm.r600.read.local.size.x() #2
+
+attributes #0 = { nounwind }
+attributes #1 = { convergent nounwind }
+attributes #2 = { nounwind readnone }