diff --git a/lib/Target/R600/AMDGPUInstructions.td b/lib/Target/R600/AMDGPUInstructions.td
index a6086270968..ef656b93ce0 100644
--- a/lib/Target/R600/AMDGPUInstructions.td
+++ b/lib/Target/R600/AMDGPUInstructions.td
@@ -387,6 +387,7 @@ class global_binary_atomic_op<SDNode atomic_op> : PatFrag<
 >;
 
 def atomic_add_global : global_binary_atomic_op<atomic_load_add>;
+def atomic_sub_global : global_binary_atomic_op<atomic_load_sub>;
 
 //===----------------------------------------------------------------------===//
 // Misc Pattern Fragments
diff --git a/lib/Target/R600/SIInstructions.td b/lib/Target/R600/SIInstructions.td
index 2efc7ac5f05..5ef57b8ee61 100644
--- a/lib/Target/R600/SIInstructions.td
+++ b/lib/Target/R600/SIInstructions.td
@@ -900,7 +900,9 @@ defm BUFFER_STORE_DWORDX4 : MUBUF_Store_Helper <
 defm BUFFER_ATOMIC_ADD : MUBUF_Atomic <
   0x00000032, "BUFFER_ATOMIC_ADD", VReg_32, i32, atomic_add_global
 >;
-//def BUFFER_ATOMIC_SUB : MUBUF_ <0x00000033, "BUFFER_ATOMIC_SUB", []>;
+defm BUFFER_ATOMIC_SUB : MUBUF_Atomic <
+  0x00000033, "BUFFER_ATOMIC_SUB", VReg_32, i32, atomic_sub_global
+>;
 //def BUFFER_ATOMIC_RSUB : MUBUF_ <0x00000034, "BUFFER_ATOMIC_RSUB", []>;
 //def BUFFER_ATOMIC_SMIN : MUBUF_ <0x00000035, "BUFFER_ATOMIC_SMIN", []>;
 //def BUFFER_ATOMIC_UMIN : MUBUF_ <0x00000036, "BUFFER_ATOMIC_UMIN", []>;
diff --git a/test/CodeGen/R600/global_atomics.ll b/test/CodeGen/R600/global_atomics.ll
index 698fffb638b..9eb06b9200d 100644
--- a/test/CodeGen/R600/global_atomics.ll
+++ b/test/CodeGen/R600/global_atomics.ll
@@ -79,3 +79,83 @@ entry:
   store i32 %0, i32 addrspace(1)* %out2
   ret void
 }
+
+; FUNC-LABEL: {{^}}atomic_sub_i32_offset:
+; SI: BUFFER_ATOMIC_SUB v{{[0-9]+}}, s[{{[0-9]+}}:{{[0-9]+}}], 0 offset:0x10{{$}}
+define void @atomic_sub_i32_offset(i32 addrspace(1)* %out, i32 %in) {
+entry:
+  %gep = getelementptr i32 addrspace(1)* %out, i32 4
+  %0  = atomicrmw volatile sub i32 addrspace(1)* %gep, i32 %in seq_cst
+  ret void
+}
+
+; FUNC-LABEL: {{^}}atomic_sub_i32_ret_offset:
+; SI: BUFFER_ATOMIC_SUB [[RET:v[0-9]+]], s[{{[0-9]+}}:{{[0-9]+}}], 0 offset:0x10 glc {{$}}
+; SI: BUFFER_STORE_DWORD [[RET]]
+define void @atomic_sub_i32_ret_offset(i32 addrspace(1)* %out, i32 addrspace(1)* %out2, i32 %in) {
+entry:
+  %gep = getelementptr i32 addrspace(1)* %out, i32 4
+  %0  = atomicrmw volatile sub i32 addrspace(1)* %gep, i32 %in seq_cst
+  store i32 %0, i32 addrspace(1)* %out2
+  ret void
+}
+
+; FUNC-LABEL: {{^}}atomic_sub_i32_addr64_offset:
+; SI: BUFFER_ATOMIC_SUB v{{[0-9]+}}, v[{{[0-9]+}}:{{[0-9]+}}], s[{{[0-9]+}}:{{[0-9]+}}], 0 addr64 offset:0x10{{$}}
+define void @atomic_sub_i32_addr64_offset(i32 addrspace(1)* %out, i32 %in, i64 %index) {
+entry:
+  %ptr = getelementptr i32 addrspace(1)* %out, i64 %index
+  %gep = getelementptr i32 addrspace(1)* %ptr, i32 4
+  %0  = atomicrmw volatile sub i32 addrspace(1)* %gep, i32 %in seq_cst
+  ret void
+}
+
+; FUNC-LABEL: {{^}}atomic_sub_i32_ret_addr64_offset:
+; SI: BUFFER_ATOMIC_SUB [[RET:v[0-9]+]], v[{{[0-9]+}}:{{[0-9]+}}], s[{{[0-9]+}}:{{[0-9]+}}], 0 addr64 offset:0x10 glc{{$}}
+; SI: BUFFER_STORE_DWORD [[RET]]
+define void @atomic_sub_i32_ret_addr64_offset(i32 addrspace(1)* %out, i32 addrspace(1)* %out2, i32 %in, i64 %index) {
+entry:
+  %ptr = getelementptr i32 addrspace(1)* %out, i64 %index
+  %gep = getelementptr i32 addrspace(1)* %ptr, i32 4
+  %0  = atomicrmw volatile sub i32 addrspace(1)* %gep, i32 %in seq_cst
+  store i32 %0, i32 addrspace(1)* %out2
+  ret void
+}
+
+; FUNC-LABEL: {{^}}atomic_sub_i32:
+; SI: BUFFER_ATOMIC_SUB v{{[0-9]+}}, s[{{[0-9]+}}:{{[0-9]+}}], 0{{$}}
+define void @atomic_sub_i32(i32 addrspace(1)* %out, i32 %in) {
+entry:
+  %0  = atomicrmw volatile sub i32 addrspace(1)* %out, i32 %in seq_cst
+  ret void
+}
+
+; FUNC-LABEL: {{^}}atomic_sub_i32_ret:
+; SI: BUFFER_ATOMIC_SUB [[RET:v[0-9]+]], s[{{[0-9]+}}:{{[0-9]+}}], 0 glc
+; SI: BUFFER_STORE_DWORD [[RET]]
+define void @atomic_sub_i32_ret(i32 addrspace(1)* %out, i32 addrspace(1)* %out2, i32 %in) {
+entry:
+  %0  = atomicrmw volatile sub i32 addrspace(1)* %out, i32 %in seq_cst
+  store i32 %0, i32 addrspace(1)* %out2
+  ret void
+}
+
+; FUNC-LABEL: {{^}}atomic_sub_i32_addr64:
+; SI: BUFFER_ATOMIC_SUB v{{[0-9]+}}, v[{{[0-9]+}}:{{[0-9]+}}], s[{{[0-9]+}}:{{[0-9]+}}], 0 addr64{{$}}
+define void @atomic_sub_i32_addr64(i32 addrspace(1)* %out, i32 %in, i64 %index) {
+entry:
+  %ptr = getelementptr i32 addrspace(1)* %out, i64 %index
+  %0  = atomicrmw volatile sub i32 addrspace(1)* %ptr, i32 %in seq_cst
+  ret void
+}
+
+; FUNC-LABEL: {{^}}atomic_sub_i32_ret_addr64:
+; SI: BUFFER_ATOMIC_SUB [[RET:v[0-9]+]], v[{{[0-9]+}}:{{[0-9]+}}], s[{{[0-9]+}}:{{[0-9]+}}], 0 addr64 glc{{$}}
+; SI: BUFFER_STORE_DWORD [[RET]]
+define void @atomic_sub_i32_ret_addr64(i32 addrspace(1)* %out, i32 addrspace(1)* %out2, i32 %in, i64 %index) {
+entry:
+  %ptr = getelementptr i32 addrspace(1)* %out, i64 %index
+  %0  = atomicrmw volatile sub i32 addrspace(1)* %ptr, i32 %in seq_cst
+  store i32 %0, i32 addrspace(1)* %out2
+  ret void
+}