AMDGPU: Fix flat atomics

The flat atomics could already be selected, but only
when using flat instructions for global memory. Add
patterns for flat addresses.

llvm-svn: 272345

8 files changed, 2301 insertions, 41 deletions

diff --git a/llvm/lib/Target/AMDGPU/AMDGPUISelDAGToDAG.cpp b/llvm/lib/Target/AMDGPU/AMDGPUISelDAGToDAG.cpp
index 6143631856b..4c8aa4b0c2a 100644
--- a/llvm/lib/Target/AMDGPU/AMDGPUISelDAGToDAG.cpp
+++ b/llvm/lib/Target/AMDGPU/AMDGPUISelDAGToDAG.cpp
@@ -138,6 +138,10 @@ private:
                                   SDValue &ImmOffset) const;
   bool SelectMUBUFIntrinsicVOffset(SDValue Offset, SDValue &SOffset,
                                    SDValue &ImmOffset, SDValue &VOffset) const;
+
+  bool SelectFlat(SDValue Addr, SDValue &VAddr,
+                  SDValue &SLC, SDValue &TFE) const;
+
   bool SelectSMRDOffset(SDValue ByteOffsetNode, SDValue &Offset,
                         bool &Imm) const;
   bool SelectSMRD(SDValue Addr, SDValue &SBase, SDValue &Offset,
@@ -1236,6 +1240,15 @@ bool AMDGPUDAGToDAGISel::SelectMUBUFIntrinsicVOffset(SDValue Offset,
   return true;
 }
 
+bool AMDGPUDAGToDAGISel::SelectFlat(SDValue Addr,
+                                    SDValue &VAddr,
+                                    SDValue &SLC,
+                                    SDValue &TFE) const {
+  VAddr = Addr;
+  TFE = SLC = CurDAG->getTargetConstant(0, SDLoc(), MVT::i1);
+  return true;
+}
+
 ///
 /// \param EncodedOffset This is the immediate value that will be encoded
 ///        directly into the instruction.  On SI/CI the \p EncodedOffset
@@ -1500,6 +1513,10 @@ void AMDGPUDAGToDAGISel::SelectBRCOND(SDNode *N) {
 void AMDGPUDAGToDAGISel::SelectATOMIC_CMP_SWAP(SDNode *N) {
   MemSDNode *Mem = cast<MemSDNode>(N);
   unsigned AS = Mem->getAddressSpace();
+  if (AS == AMDGPUAS::FLAT_ADDRESS) {
+    SelectCode(N);
+    return;
+  }
 
   MVT VT = N->getSimpleValueType(0);
   bool Is32 = (VT == MVT::i32);
diff --git a/llvm/lib/Target/AMDGPU/AMDGPUInstructions.td b/llvm/lib/Target/AMDGPU/AMDGPUInstructions.td
index 2874a55bef3..4b6fc47cdb7 100644
--- a/llvm/lib/Target/AMDGPU/AMDGPUInstructions.td
+++ b/llvm/lib/Target/AMDGPU/AMDGPUInstructions.td
@@ -394,6 +394,12 @@ class global_binary_atomic_op<SDNode atomic_op> : PatFrag<
   [{return cast<MemSDNode>(N)->getAddressSpace() == AMDGPUAS::GLOBAL_ADDRESS;}]
 >;
 
+class flat_binary_atomic_op<SDNode atomic_op> : PatFrag<
+  (ops node:$ptr, node:$value),
+  (atomic_op node:$ptr, node:$value),
+  [{return cast<MemSDNode>(N)->getAddressSpace() == AMDGPUAS::FLAT_ADDRESS;}]
+>;
+
 def atomic_swap_global : global_binary_atomic_op<atomic_swap>;
 def atomic_add_global : global_binary_atomic_op<atomic_load_add>;
 def atomic_and_global : global_binary_atomic_op<atomic_load_and>;
@@ -412,6 +418,19 @@ def atomic_cmp_swap_global_nortn : PatFrag<
   [{ return SDValue(N, 0).use_empty(); }]
 >;
 
+def atomic_swap_flat : flat_binary_atomic_op<atomic_swap>;
+def atomic_add_flat : flat_binary_atomic_op<atomic_load_add>;
+def atomic_and_flat : flat_binary_atomic_op<atomic_load_and>;
+def atomic_max_flat : flat_binary_atomic_op<atomic_load_max>;
+def atomic_min_flat : flat_binary_atomic_op<atomic_load_min>;
+def atomic_or_flat : flat_binary_atomic_op<atomic_load_or>;
+def atomic_sub_flat : flat_binary_atomic_op<atomic_load_sub>;
+def atomic_umax_flat : flat_binary_atomic_op<atomic_load_umax>;
+def atomic_umin_flat : flat_binary_atomic_op<atomic_load_umin>;
+def atomic_xor_flat : flat_binary_atomic_op<atomic_load_xor>;
+
+def atomic_cmp_swap_flat : flat_binary_atomic_op<AMDGPUatomic_cmp_swap>;
+
 //===----------------------------------------------------------------------===//
 // Misc Pattern Fragments
 //===----------------------------------------------------------------------===//
diff --git a/llvm/lib/Target/AMDGPU/CIInstructions.td b/llvm/lib/Target/AMDGPU/CIInstructions.td
index 9fb6e6039e5..f9a9f79126b 100644
--- a/llvm/lib/Target/AMDGPU/CIInstructions.td
+++ b/llvm/lib/Target/AMDGPU/CIInstructions.td
@@ -153,82 +153,84 @@ defm FLAT_STORE_DWORDX3 : FLAT_Store_Helper <
   flat<0x1f, 0x1e>, "flat_store_dwordx3", VReg_96
 >;
 defm FLAT_ATOMIC_SWAP : FLAT_ATOMIC <
-  flat<0x30, 0x40>, "flat_atomic_swap", VGPR_32
+  flat<0x30, 0x40>, "flat_atomic_swap", VGPR_32, i32, atomic_swap_flat
 >;
 defm FLAT_ATOMIC_CMPSWAP : FLAT_ATOMIC <
-  flat<0x31, 0x41>, "flat_atomic_cmpswap", VGPR_32, VReg_64
+  flat<0x31, 0x41>, "flat_atomic_cmpswap", VGPR_32, i32,
+    atomic_cmp_swap_flat, v2i32, VReg_64
 >;
 defm FLAT_ATOMIC_ADD : FLAT_ATOMIC <
-  flat<0x32, 0x42>, "flat_atomic_add", VGPR_32
+  flat<0x32, 0x42>, "flat_atomic_add", VGPR_32, i32, atomic_add_flat
 >;
 defm FLAT_ATOMIC_SUB : FLAT_ATOMIC <
-  flat<0x33, 0x43>, "flat_atomic_sub", VGPR_32
+  flat<0x33, 0x43>, "flat_atomic_sub", VGPR_32, i32, atomic_sub_flat
 >;
 defm FLAT_ATOMIC_SMIN : FLAT_ATOMIC <
-  flat<0x35, 0x44>, "flat_atomic_smin", VGPR_32
+  flat<0x35, 0x44>, "flat_atomic_smin", VGPR_32, i32, atomic_min_flat
 >;
 defm FLAT_ATOMIC_UMIN : FLAT_ATOMIC <
-  flat<0x36, 0x45>, "flat_atomic_umin", VGPR_32
+  flat<0x36, 0x45>, "flat_atomic_umin", VGPR_32, i32, atomic_umin_flat
 >;
 defm FLAT_ATOMIC_SMAX : FLAT_ATOMIC <
-  flat<0x37, 0x46>, "flat_atomic_smax", VGPR_32
+  flat<0x37, 0x46>, "flat_atomic_smax", VGPR_32, i32, atomic_max_flat
 >;
 defm FLAT_ATOMIC_UMAX : FLAT_ATOMIC <
-  flat<0x38, 0x47>, "flat_atomic_umax", VGPR_32
+  flat<0x38, 0x47>, "flat_atomic_umax", VGPR_32, i32, atomic_umax_flat
 >;
 defm FLAT_ATOMIC_AND : FLAT_ATOMIC <
-  flat<0x39, 0x48>, "flat_atomic_and", VGPR_32
+  flat<0x39, 0x48>, "flat_atomic_and", VGPR_32, i32, atomic_and_flat
 >;
 defm FLAT_ATOMIC_OR : FLAT_ATOMIC <
-  flat<0x3a, 0x49>, "flat_atomic_or", VGPR_32
+  flat<0x3a, 0x49>, "flat_atomic_or", VGPR_32, i32, atomic_or_flat
 >;
 defm FLAT_ATOMIC_XOR : FLAT_ATOMIC <
-  flat<0x3b, 0x4a>, "flat_atomic_xor", VGPR_32
+  flat<0x3b, 0x4a>, "flat_atomic_xor", VGPR_32, i32, atomic_xor_flat
 >;
 defm FLAT_ATOMIC_INC : FLAT_ATOMIC <
-  flat<0x3c, 0x4b>, "flat_atomic_inc", VGPR_32
+  flat<0x3c, 0x4b>, "flat_atomic_inc", VGPR_32, i32, atomic_inc_flat
 >;
 defm FLAT_ATOMIC_DEC : FLAT_ATOMIC <
-  flat<0x3d, 0x4c>, "flat_atomic_dec", VGPR_32
+  flat<0x3d, 0x4c>, "flat_atomic_dec", VGPR_32, i32, atomic_dec_flat
 >;
 defm FLAT_ATOMIC_SWAP_X2 : FLAT_ATOMIC <
-  flat<0x50, 0x60>, "flat_atomic_swap_x2", VReg_64
+  flat<0x50, 0x60>, "flat_atomic_swap_x2", VReg_64, i64, atomic_swap_flat
 >;
 defm FLAT_ATOMIC_CMPSWAP_X2 : FLAT_ATOMIC <
-  flat<0x51, 0x61>, "flat_atomic_cmpswap_x2", VReg_64, VReg_128
+  flat<0x51, 0x61>, "flat_atomic_cmpswap_x2", VReg_64, i64,
+    atomic_cmp_swap_flat, v2i64, VReg_128
 >;
 defm FLAT_ATOMIC_ADD_X2 : FLAT_ATOMIC <
-  flat<0x52, 0x62>, "flat_atomic_add_x2", VReg_64
+  flat<0x52, 0x62>, "flat_atomic_add_x2", VReg_64, i64, atomic_add_flat
 >;
 defm FLAT_ATOMIC_SUB_X2 : FLAT_ATOMIC <
-  flat<0x53, 0x63>, "flat_atomic_sub_x2", VReg_64
+  flat<0x53, 0x63>, "flat_atomic_sub_x2", VReg_64, i64, atomic_sub_flat
 >;
 defm FLAT_ATOMIC_SMIN_X2 : FLAT_ATOMIC <
-  flat<0x55, 0x64>, "flat_atomic_smin_x2", VReg_64
+  flat<0x55, 0x64>, "flat_atomic_smin_x2", VReg_64, i64, atomic_min_flat
 >;
 defm FLAT_ATOMIC_UMIN_X2 : FLAT_ATOMIC <
-  flat<0x56, 0x65>, "flat_atomic_umin_x2", VReg_64
+  flat<0x56, 0x65>, "flat_atomic_umin_x2", VReg_64, i64, atomic_umin_flat
 >;
 defm FLAT_ATOMIC_SMAX_X2 : FLAT_ATOMIC <
-  flat<0x57, 0x66>, "flat_atomic_smax_x2", VReg_64
+  flat<0x57, 0x66>, "flat_atomic_smax_x2", VReg_64, i64, atomic_max_flat
 >;
 defm FLAT_ATOMIC_UMAX_X2 : FLAT_ATOMIC <
-  flat<0x58, 0x67>, "flat_atomic_umax_x2", VReg_64
+  flat<0x58, 0x67>, "flat_atomic_umax_x2", VReg_64, i64, atomic_umax_flat
 >;
 defm FLAT_ATOMIC_AND_X2 : FLAT_ATOMIC <
-  flat<0x59, 0x68>, "flat_atomic_and_x2", VReg_64
+  flat<0x59, 0x68>, "flat_atomic_and_x2", VReg_64, i64, atomic_and_flat
 >;
 defm FLAT_ATOMIC_OR_X2 : FLAT_ATOMIC <
-  flat<0x5a, 0x69>, "flat_atomic_or_x2", VReg_64
+  flat<0x5a, 0x69>, "flat_atomic_or_x2", VReg_64, i64, atomic_or_flat
 >;
 defm FLAT_ATOMIC_XOR_X2 : FLAT_ATOMIC <
-  flat<0x5b, 0x6a>, "flat_atomic_xor_x2", VReg_64
+  flat<0x5b, 0x6a>, "flat_atomic_xor_x2", VReg_64, i64, atomic_xor_flat
 >;
 defm FLAT_ATOMIC_INC_X2 : FLAT_ATOMIC <
-  flat<0x5c, 0x6b>, "flat_atomic_inc_x2", VReg_64
+  flat<0x5c, 0x6b>, "flat_atomic_inc_x2", VReg_64, i64, atomic_inc_flat
 >;
 defm FLAT_ATOMIC_DEC_X2 : FLAT_ATOMIC <
-  flat<0x5d, 0x6c>, "flat_atomic_dec_x2", VReg_64
+  flat<0x5d, 0x6c>, "flat_atomic_dec_x2", VReg_64, i64, atomic_dec_flat
 >;
 
 } // End SubtargetPredicate = isCIVI
@@ -238,22 +240,24 @@ defm FLAT_ATOMIC_DEC_X2 : FLAT_ATOMIC <
 let SubtargetPredicate = isCI, VIAssemblerPredicate = DisableInst, DisableVIDecoder = 1 in {
 
 defm FLAT_ATOMIC_FCMPSWAP : FLAT_ATOMIC <
-  flat<0x3e>, "flat_atomic_fcmpswap", VGPR_32, VReg_64
+  flat<0x3e>, "flat_atomic_fcmpswap", VGPR_32, f32,
+    null_frag, v2f32, VReg_64
 >;
 defm FLAT_ATOMIC_FMIN : FLAT_ATOMIC <
-  flat<0x3f>, "flat_atomic_fmin", VGPR_32
+  flat<0x3f>, "flat_atomic_fmin", VGPR_32, f32
 >;
 defm FLAT_ATOMIC_FMAX : FLAT_ATOMIC <
-  flat<0x40>, "flat_atomic_fmax", VGPR_32
+  flat<0x40>, "flat_atomic_fmax", VGPR_32, f32
 >;
 defm FLAT_ATOMIC_FCMPSWAP_X2 : FLAT_ATOMIC <
-  flat<0x5e>, "flat_atomic_fcmpswap_x2", VReg_64, VReg_128
+  flat<0x5e>, "flat_atomic_fcmpswap_x2", VReg_64, f64,
+  null_frag, v2f64, VReg_128
 >;
 defm FLAT_ATOMIC_FMIN_X2 : FLAT_ATOMIC <
-  flat<0x5f>, "flat_atomic_fmin_x2", VReg_64
+  flat<0x5f>, "flat_atomic_fmin_x2", VReg_64, f64
 >;
 defm FLAT_ATOMIC_FMAX_X2 : FLAT_ATOMIC <
-  flat<0x60>, "flat_atomic_fmax_x2", VReg_64
+  flat<0x60>, "flat_atomic_fmax_x2", VReg_64, f64
 >;
 
 } // End SubtargetPredicate = isCI, VIAssemblerPredicate = DisableInst, DisableVIDecoder = 1
@@ -293,7 +297,8 @@ class FlatStorePat <FLAT inst, SDPatternOperator node, ValueType vt> : Pat <
 >;
 
 class FlatStoreAtomicPat <FLAT inst, SDPatternOperator node, ValueType vt> : Pat <
-  // atomic store follows aotmic binop convenction so the address comes first
+  // atomic store follows atomic binop convention so the address comes
+  // first.
   (node i64:$addr, vt:$data),
   (inst $addr, $data, 1, 0, 0)
 >;
diff --git a/llvm/lib/Target/AMDGPU/SIInstrInfo.td b/llvm/lib/Target/AMDGPU/SIInstrInfo.td
index d7805fd8e37..94932a89e13 100644
--- a/llvm/lib/Target/AMDGPU/SIInstrInfo.td
+++ b/llvm/lib/Target/AMDGPU/SIInstrInfo.td
@@ -190,6 +190,9 @@ def smrd_load : PatFrag <(ops node:$ptr), (load node:$ptr), [{
 def atomic_inc_global : global_binary_atomic_op<SIatomic_inc>;
 def atomic_dec_global : global_binary_atomic_op<SIatomic_dec>;
 
+def atomic_inc_flat : flat_binary_atomic_op<SIatomic_inc>;
+def atomic_dec_flat : flat_binary_atomic_op<SIatomic_dec>;
+
 //===----------------------------------------------------------------------===//
 // SDNodes and PatFrag for local loads and stores to enable s_mov_b32 m0, -1
 // to be glued to the memory instructions.
@@ -567,6 +570,7 @@ def DS64Bit4ByteAligned : ComplexPattern<i32, 3, "SelectDS64Bit4ByteAligned">;
 def MUBUFAddr32 : ComplexPattern<i64, 9, "SelectMUBUFAddr32">;
 def MUBUFAddr64 : ComplexPattern<i64, 7, "SelectMUBUFAddr64">;
 def MUBUFAddr64Atomic : ComplexPattern<i64, 5, "SelectMUBUFAddr64">;
+def FLATAtomic : ComplexPattern<i64, 3, "SelectFlat">;
 def MUBUFScratch : ComplexPattern<i64, 4, "SelectMUBUFScratch">;
 def MUBUFOffset : ComplexPattern<i64, 6, "SelectMUBUFOffset">;
 def MUBUFOffsetNoGLC : ComplexPattern<i64, 3, "SelectMUBUFOffset">;
@@ -3210,32 +3214,36 @@ multiclass FLAT_Store_Helper <flat op, string asm_name,
 }
 
 multiclass FLAT_ATOMIC <flat op, string asm_name, RegisterClass vdst_rc,
+    ValueType vt, SDPatternOperator atomic = null_frag,
+    ValueType data_vt = vt,
     RegisterClass data_rc = vdst_rc,
-    dag outs_noret = (outs),
     string asm_noret = asm_name#" $addr, $data"#"$slc"#"$tfe"> {
 
   let mayLoad = 1, mayStore = 1, glc = 0, vdst = 0 in {
-    def "" : FLAT_Pseudo <NAME, outs_noret,
+    def "" : FLAT_Pseudo <NAME, (outs),
                           (ins VReg_64:$addr, data_rc:$data,
                                slc:$slc, tfe:$tfe), []>,
              AtomicNoRet <NAME, 0>;
 
-    def _ci : FLAT_Real_ci <op.CI, NAME, outs_noret,
+    def _ci : FLAT_Real_ci <op.CI, NAME, (outs),
                             (ins VReg_64:$addr, data_rc:$data,
                                  slc:$slc, tfe:$tfe),
                             asm_noret>;
 
-    def _vi : FLAT_Real_vi <op.VI, NAME, outs_noret,
+    def _vi : FLAT_Real_vi <op.VI, NAME, (outs),
                             (ins VReg_64:$addr, data_rc:$data,
                                  slc:$slc, tfe:$tfe),
                             asm_noret>;
   }
 
   let glc = 1, hasPostISelHook = 1 in {
-    defm _RTN : FLAT_AtomicRet_m <op, (outs vdst_rc:$vdst),
-                        (ins VReg_64:$addr, data_rc:$data, slc:$slc,
-                             tfe:$tfe),
-                        asm_name#" $vdst, $addr, $data glc$slc$tfe", []>;
+    defm _RTN : FLAT_AtomicRet_m <
+      op, (outs vdst_rc:$vdst),
+      (ins VReg_64:$addr, data_rc:$data, slc:$slc, tfe:$tfe),
+      asm_name#" $vdst, $addr, $data glc$slc$tfe",
+      [(set vt:$vdst,
+         (atomic (FLATAtomic i64:$addr, i1:$slc, i1:$tfe), data_vt:$data))]
+    >;
   }
 }
 
diff --git a/llvm/test/CodeGen/AMDGPU/flat_atomics.ll b/llvm/test/CodeGen/AMDGPU/flat_atomics.ll
new file mode 100644
index 00000000000..7400dbcf890
--- /dev/null
+++ b/llvm/test/CodeGen/AMDGPU/flat_atomics.ll
@@ -0,0 +1,968 @@
+; RUN: llc -march=amdgcn -mcpu=bonaire -verify-machineinstrs < %s | FileCheck -check-prefix=GCN %s
+; RUN: llc -march=amdgcn -mcpu=tonga -verify-machineinstrs < %s | FileCheck -check-prefix=GCN %s
+
+; GCN-LABEL: {{^}}atomic_add_i32_offset:
+; GCN: flat_atomic_add v[{{[0-9]+}}:{{[0-9]+}}], v{{[0-9]+}}{{$}}
+define void @atomic_add_i32_offset(i32 addrspace(4)* %out, i32 %in) {
+entry:
+  %gep = getelementptr i32, i32 addrspace(4)* %out, i32 4
+  %val = atomicrmw volatile add i32 addrspace(4)* %gep, i32 %in seq_cst
+  ret void
+}
+
+; GCN-LABEL: {{^}}atomic_add_i32_ret_offset:
+; GCN: flat_atomic_add [[RET:v[0-9]+]], v[{{[0-9]+}}:{{[0-9]+}}], v{{[0-9]+}} glc{{$}}
+; GCN: flat_store_dword v{{\[[0-9]+:[0-9]+\]}}, [[RET]]
+define void @atomic_add_i32_ret_offset(i32 addrspace(4)* %out, i32 addrspace(4)* %out2, i32 %in) {
+entry:
+  %gep = getelementptr i32, i32 addrspace(4)* %out, i32 4
+  %val = atomicrmw volatile add i32 addrspace(4)* %gep, i32 %in seq_cst
+  store i32 %val, i32 addrspace(4)* %out2
+  ret void
+}
+
+; GCN-LABEL: {{^}}atomic_add_i32_addr64_offset:
+; GCN: flat_atomic_add v[{{[0-9]+:[0-9]+}}], v{{[0-9]+$}}
+define void @atomic_add_i32_addr64_offset(i32 addrspace(4)* %out, i32 %in, i64 %index) {
+entry:
+  %ptr = getelementptr i32, i32 addrspace(4)* %out, i64 %index
+  %gep = getelementptr i32, i32 addrspace(4)* %ptr, i32 4
+  %val = atomicrmw volatile add i32 addrspace(4)* %gep, i32 %in seq_cst
+  ret void
+}
+
+; GCN-LABEL: {{^}}atomic_add_i32_ret_addr64_offset:
+; GCN: flat_atomic_add [[RET:v[0-9]+]], v[{{[0-9]+:[0-9]+}}], v{{[0-9]+}} glc{{$}}
+; GCN: flat_store_dword v{{\[[0-9]+:[0-9]+\]}}, [[RET]]
+define void @atomic_add_i32_ret_addr64_offset(i32 addrspace(4)* %out, i32 addrspace(4)* %out2, i32 %in, i64 %index) {
+entry:
+  %ptr = getelementptr i32, i32 addrspace(4)* %out, i64 %index
+  %gep = getelementptr i32, i32 addrspace(4)* %ptr, i32 4
+  %val = atomicrmw volatile add i32 addrspace(4)* %gep, i32 %in seq_cst
+  store i32 %val, i32 addrspace(4)* %out2
+  ret void
+}
+
+; GCN-LABEL: {{^}}atomic_add_i32:
+; GCN: flat_atomic_add v[{{[0-9]+:[0-9]+}}], v{{[0-9]+$}}
+define void @atomic_add_i32(i32 addrspace(4)* %out, i32 %in) {
+entry:
+  %val = atomicrmw volatile add i32 addrspace(4)* %out, i32 %in seq_cst
+  ret void
+}
+
+; GCN-LABEL: {{^}}atomic_add_i32_ret:
+; GCN: flat_atomic_add [[RET:v[0-9]+]], v[{{[0-9]+:[0-9]+}}], v{{[0-9]+}} glc{{$}}
+; GCN: flat_store_dword v{{\[[0-9]+:[0-9]+\]}}, [[RET]]
+define void @atomic_add_i32_ret(i32 addrspace(4)* %out, i32 addrspace(4)* %out2, i32 %in) {
+entry:
+  %val = atomicrmw volatile add i32 addrspace(4)* %out, i32 %in seq_cst
+  store i32 %val, i32 addrspace(4)* %out2
+  ret void
+}
+
+; GCN-LABEL: {{^}}atomic_add_i32_addr64:
+; GCN: flat_atomic_add v[{{[0-9]+:[0-9]+}}], v{{[0-9]+$}}
+define void @atomic_add_i32_addr64(i32 addrspace(4)* %out, i32 %in, i64 %index) {
+entry:
+  %ptr = getelementptr i32, i32 addrspace(4)* %out, i64 %index
+  %val = atomicrmw volatile add i32 addrspace(4)* %ptr, i32 %in seq_cst
+  ret void
+}
+
+; GCN-LABEL: {{^}}atomic_add_i32_ret_addr64:
+; GCN: flat_atomic_add [[RET:v[0-9]+]], v[{{[0-9]+:[0-9]+}}], v{{[0-9]+}} glc{{$}}
+; GCN: flat_store_dword v{{\[[0-9]+:[0-9]+\]}}, [[RET]]
+define void @atomic_add_i32_ret_addr64(i32 addrspace(4)* %out, i32 addrspace(4)* %out2, i32 %in, i64 %index) {
+entry:
+  %ptr = getelementptr i32, i32 addrspace(4)* %out, i64 %index
+  %val = atomicrmw volatile add i32 addrspace(4)* %ptr, i32 %in seq_cst
+  store i32 %val, i32 addrspace(4)* %out2
+  ret void
+}
+
+; GCN-LABEL: {{^}}atomic_and_i32_offset:
+; GCN: flat_atomic_and v[{{[0-9]+:[0-9]+}}], v{{[0-9]+$}}
+define void @atomic_and_i32_offset(i32 addrspace(4)* %out, i32 %in) {
+entry:
+  %gep = getelementptr i32, i32 addrspace(4)* %out, i32 4
+  %val = atomicrmw volatile and i32 addrspace(4)* %gep, i32 %in seq_cst
+  ret void
+}
+
+; GCN-LABEL: {{^}}atomic_and_i32_ret_offset:
+; GCN: flat_atomic_and [[RET:v[0-9]]], v[{{[0-9]+:[0-9]+}}], v{{[0-9]+}} glc{{$}}
+; GCN: flat_store_dword v{{\[[0-9]+:[0-9]+\]}}, [[RET]]
+define void @atomic_and_i32_ret_offset(i32 addrspace(4)* %out, i32 addrspace(4)* %out2, i32 %in) {
+entry:
+  %gep = getelementptr i32, i32 addrspace(4)* %out, i32 4
+  %val = atomicrmw volatile and i32 addrspace(4)* %gep, i32 %in seq_cst
+  store i32 %val, i32 addrspace(4)* %out2
+  ret void
+}
+
+; GCN-LABEL: {{^}}atomic_and_i32_addr64_offset:
+; GCN: flat_atomic_and v[{{[0-9]+:[0-9]+}}], v{{[0-9]+$}}
+define void @atomic_and_i32_addr64_offset(i32 addrspace(4)* %out, i32 %in, i64 %index) {
+entry:
+  %ptr = getelementptr i32, i32 addrspace(4)* %out, i64 %index
+  %gep = getelementptr i32, i32 addrspace(4)* %ptr, i32 4
+  %val = atomicrmw volatile and i32 addrspace(4)* %gep, i32 %in seq_cst
+  ret void
+}
+
+; GCN-LABEL: {{^}}atomic_and_i32_ret_addr64_offset:
+; GCN: flat_atomic_and [[RET:v[0-9]]], v[{{[0-9]+:[0-9]+}}], v{{[0-9]+}} glc{{$}}
+; GCN: flat_store_dword v{{\[[0-9]+:[0-9]+\]}}, [[RET]]
+define void @atomic_and_i32_ret_addr64_offset(i32 addrspace(4)* %out, i32 addrspace(4)* %out2, i32 %in, i64 %index) {
+entry:
+  %ptr = getelementptr i32, i32 addrspace(4)* %out, i64 %index
+  %gep = getelementptr i32, i32 addrspace(4)* %ptr, i32 4
+  %val = atomicrmw volatile and i32 addrspace(4)* %gep, i32 %in seq_cst
+  store i32 %val, i32 addrspace(4)* %out2
+  ret void
+}
+
+; GCN-LABEL: {{^}}atomic_and_i32:
+; GCN: flat_atomic_and v[{{[0-9]+:[0-9]+}}], v{{[0-9]+$}}
+define void @atomic_and_i32(i32 addrspace(4)* %out, i32 %in) {
+entry:
+  %val = atomicrmw volatile and i32 addrspace(4)* %out, i32 %in seq_cst
+  ret void
+}
+
+; GCN-LABEL: {{^}}atomic_and_i32_ret:
+; GCN: flat_atomic_and [[RET:v[0-9]+]], v[{{[0-9]+:[0-9]+}}], v{{[0-9]+}} glc{{$}}
+; GCN: flat_store_dword v{{\[[0-9]+:[0-9]+\]}}, [[RET]]
+define void @atomic_and_i32_ret(i32 addrspace(4)* %out, i32 addrspace(4)* %out2, i32 %in) {
+entry:
+  %val = atomicrmw volatile and i32 addrspace(4)* %out, i32 %in seq_cst
+  store i32 %val, i32 addrspace(4)* %out2
+  ret void
+}
+
+; GCN-LABEL: {{^}}atomic_and_i32_addr64:
+; GCN: flat_atomic_and v[{{[0-9]+:[0-9]+}}], v{{[0-9]+$}}
+define void @atomic_and_i32_addr64(i32 addrspace(4)* %out, i32 %in, i64 %index) {
+entry:
+  %ptr = getelementptr i32, i32 addrspace(4)* %out, i64 %index
+  %val = atomicrmw volatile and i32 addrspace(4)* %ptr, i32 %in seq_cst
+  ret void
+}
+
+; GCN-LABEL: {{^}}atomic_and_i32_ret_addr64:
+; GCN: flat_atomic_and [[RET:v[0-9]+]], v[{{[0-9]+:[0-9]+}}], v{{[0-9]+}} glc{{$}}
+; GCN: flat_store_dword v{{\[[0-9]+:[0-9]+\]}}, [[RET]]
+define void @atomic_and_i32_ret_addr64(i32 addrspace(4)* %out, i32 addrspace(4)* %out2, i32 %in, i64 %index) {
+entry:
+  %ptr = getelementptr i32, i32 addrspace(4)* %out, i64 %index
+  %val = atomicrmw volatile and i32 addrspace(4)* %ptr, i32 %in seq_cst
+  store i32 %val, i32 addrspace(4)* %out2
+  ret void
+}
+
+; GCN-LABEL: {{^}}atomic_sub_i32_offset:
+; GCN: flat_atomic_sub v[{{[0-9]+:[0-9]+}}], v{{[0-9]+$}}
+define void @atomic_sub_i32_offset(i32 addrspace(4)* %out, i32 %in) {
+entry:
+  %gep = getelementptr i32, i32 addrspace(4)* %out, i32 4
+  %val = atomicrmw volatile sub i32 addrspace(4)* %gep, i32 %in seq_cst
+  ret void
+}
+
+; GCN-LABEL: {{^}}atomic_sub_i32_ret_offset:
+; GCN: flat_atomic_sub [[RET:v[0-9]+]], v[{{[0-9]+:[0-9]+}}], v{{[0-9]+}} glc{{$}}
+; GCN: flat_store_dword v{{\[[0-9]+:[0-9]+\]}}, [[RET]]
+define void @atomic_sub_i32_ret_offset(i32 addrspace(4)* %out, i32 addrspace(4)* %out2, i32 %in) {
+entry:
+  %gep = getelementptr i32, i32 addrspace(4)* %out, i32 4
+  %val = atomicrmw volatile sub i32 addrspace(4)* %gep, i32 %in seq_cst
+  store i32 %val, i32 addrspace(4)* %out2
+  ret void
+}
+
+; GCN-LABEL: {{^}}atomic_sub_i32_addr64_offset:
+; GCN: flat_atomic_sub v[{{[0-9]+:[0-9]+}}], v{{[0-9]+$}}
+define void @atomic_sub_i32_addr64_offset(i32 addrspace(4)* %out, i32 %in, i64 %index) {
+entry:
+  %ptr = getelementptr i32, i32 addrspace(4)* %out, i64 %index
+  %gep = getelementptr i32, i32 addrspace(4)* %ptr, i32 4
+  %val = atomicrmw volatile sub i32 addrspace(4)* %gep, i32 %in seq_cst
+  ret void
+}
+
+; GCN-LABEL: {{^}}atomic_sub_i32_ret_addr64_offset:
+; GCN: flat_atomic_sub [[RET:v[0-9]+]], v[{{[0-9]+:[0-9]+}}], v{{[0-9]+}} glc{{$}}
+; GCN: flat_store_dword v{{\[[0-9]+:[0-9]+\]}}, [[RET]]
+define void @atomic_sub_i32_ret_addr64_offset(i32 addrspace(4)* %out, i32 addrspace(4)* %out2, i32 %in, i64 %index) {
+entry:
+  %ptr = getelementptr i32, i32 addrspace(4)* %out, i64 %index
+  %gep = getelementptr i32, i32 addrspace(4)* %ptr, i32 4
+  %val = atomicrmw volatile sub i32 addrspace(4)* %gep, i32 %in seq_cst
+  store i32 %val, i32 addrspace(4)* %out2
+  ret void
+}
+
+; GCN-LABEL: {{^}}atomic_sub_i32:
+; GCN: flat_atomic_sub v[{{[0-9]+:[0-9]+}}], v{{[0-9]+$}}
+define void @atomic_sub_i32(i32 addrspace(4)* %out, i32 %in) {
+entry:
+  %val = atomicrmw volatile sub i32 addrspace(4)* %out, i32 %in seq_cst
+  ret void
+}
+
+; GCN-LABEL: {{^}}atomic_sub_i32_ret:
+; GCN: flat_atomic_sub [[RET:v[0-9]+]], v[{{[0-9]+:[0-9]+}}], v{{[0-9]+}} glc{{$}}
+; GCN: flat_store_dword v{{\[[0-9]+:[0-9]+\]}}, [[RET]]
+define void @atomic_sub_i32_ret(i32 addrspace(4)* %out, i32 addrspace(4)* %out2, i32 %in) {
+entry:
+  %val = atomicrmw volatile sub i32 addrspace(4)* %out, i32 %in seq_cst
+  store i32 %val, i32 addrspace(4)* %out2
+  ret void
+}
+
+; GCN-LABEL: {{^}}atomic_sub_i32_addr64:
+; GCN: flat_atomic_sub v[{{[0-9]+:[0-9]+}}], v{{[0-9]+$}}
+define void @atomic_sub_i32_addr64(i32 addrspace(4)* %out, i32 %in, i64 %index) {
+entry:
+  %ptr = getelementptr i32, i32 addrspace(4)* %out, i64 %index
+  %val = atomicrmw volatile sub i32 addrspace(4)* %ptr, i32 %in seq_cst
+  ret void
+}
+
+; GCN-LABEL: {{^}}atomic_sub_i32_ret_addr64:
+; GCN: flat_atomic_sub [[RET:v[0-9]+]], v[{{[0-9]+:[0-9]+}}], v{{[0-9]+}} glc{{$}}
+; GCN: flat_store_dword v{{\[[0-9]+:[0-9]+\]}}, [[RET]]
+define void @atomic_sub_i32_ret_addr64(i32 addrspace(4)* %out, i32 addrspace(4)* %out2, i32 %in, i64 %index) {
+entry:
+  %ptr = getelementptr i32, i32 addrspace(4)* %out, i64 %index
+  %val = atomicrmw volatile sub i32 addrspace(4)* %ptr, i32 %in seq_cst
+  store i32 %val, i32 addrspace(4)* %out2
+  ret void
+}
+
+; GCN-LABEL: {{^}}atomic_max_i32_offset:
+; GCN: flat_atomic_smax v[{{[0-9]+:[0-9]+}}], v{{[0-9]+$}}
+define void @atomic_max_i32_offset(i32 addrspace(4)* %out, i32 %in) {
+entry:
+  %gep = getelementptr i32, i32 addrspace(4)* %out, i32 4
+  %val = atomicrmw volatile max i32 addrspace(4)* %gep, i32 %in seq_cst
+  ret void
+}
+
+; GCN-LABEL: {{^}}atomic_max_i32_ret_offset:
+; GCN: flat_atomic_smax [[RET:v[0-9]+]], v[{{[0-9]+:[0-9]+}}], v{{[0-9]+}} glc{{$}}
+; GCN: flat_store_dword v{{\[[0-9]+:[0-9]+\]}}, [[RET]]
+define void @atomic_max_i32_ret_offset(i32 addrspace(4)* %out, i32 addrspace(4)* %out2, i32 %in) {
+entry:
+  %gep = getelementptr i32, i32 addrspace(4)* %out, i32 4
+  %val = atomicrmw volatile max i32 addrspace(4)* %gep, i32 %in seq_cst
+  store i32 %val, i32 addrspace(4)* %out2
+  ret void
+}
+
+; GCN-LABEL: {{^}}atomic_max_i32_addr64_offset:
+; GCN: flat_atomic_smax v[{{[0-9]+:[0-9]+}}], v{{[0-9]+$}}
+define void @atomic_max_i32_addr64_offset(i32 addrspace(4)* %out, i32 %in, i64 %index) {
+entry:
+  %ptr = getelementptr i32, i32 addrspace(4)* %out, i64 %index
+  %gep = getelementptr i32, i32 addrspace(4)* %ptr, i32 4
+  %val = atomicrmw volatile max i32 addrspace(4)* %gep, i32 %in seq_cst
+  ret void
+}
+
+; GCN-LABEL: {{^}}atomic_max_i32_ret_addr64_offset:
+; GCN: flat_atomic_smax [[RET:v[0-9]+]], v[{{[0-9]+:[0-9]+}}], v{{[0-9]+}} glc{{$}}
+; GCN: flat_store_dword v{{\[[0-9]+:[0-9]+\]}}, [[RET]]
+define void @atomic_max_i32_ret_addr64_offset(i32 addrspace(4)* %out, i32 addrspace(4)* %out2, i32 %in, i64 %index) {
+entry:
+  %ptr = getelementptr i32, i32 addrspace(4)* %out, i64 %index
+  %gep = getelementptr i32, i32 addrspace(4)* %ptr, i32 4
+  %val = atomicrmw volatile max i32 addrspace(4)* %gep, i32 %in seq_cst
+  store i32 %val, i32 addrspace(4)* %out2
+  ret void
+}
+
+; GCN-LABEL: {{^}}atomic_max_i32:
+; GCN: flat_atomic_smax v[{{[0-9]+:[0-9]+}}], v{{[0-9]+$}}
+define void @atomic_max_i32(i32 addrspace(4)* %out, i32 %in) {
+entry:
+  %val = atomicrmw volatile max i32 addrspace(4)* %out, i32 %in seq_cst
+  ret void
+}
+
+; GCN-LABEL: {{^}}atomic_max_i32_ret:
+; GCN: flat_atomic_smax [[RET:v[0-9]+]], v[{{[0-9]+:[0-9]+}}], v{{[0-9]+}} glc{{$}}
+; GCN: flat_store_dword v{{\[[0-9]+:[0-9]+\]}}, [[RET]]
+define void @atomic_max_i32_ret(i32 addrspace(4)* %out, i32 addrspace(4)* %out2, i32 %in) {
+entry:
+  %val = atomicrmw volatile max i32 addrspace(4)* %out, i32 %in seq_cst
+  store i32 %val, i32 addrspace(4)* %out2
+  ret void
+}
+
+; GCN-LABEL: {{^}}atomic_max_i32_addr64:
+; GCN: flat_atomic_smax v[{{[0-9]+:[0-9]+}}], v{{[0-9]+$}}
+define void @atomic_max_i32_addr64(i32 addrspace(4)* %out, i32 %in, i64 %index) {
+entry:
+  %ptr = getelementptr i32, i32 addrspace(4)* %out, i64 %index
+  %val = atomicrmw volatile max i32 addrspace(4)* %ptr, i32 %in seq_cst
+  ret void
+}
+
+; GCN-LABEL: {{^}}atomic_max_i32_ret_addr64:
+; GCN: flat_atomic_smax [[RET:v[0-9]+]], v[{{[0-9]+:[0-9]+}}], v{{[0-9]+}} glc{{$}}
+; GCN: flat_store_dword v{{\[[0-9]+:[0-9]+\]}}, [[RET]]
+define void @atomic_max_i32_ret_addr64(i32 addrspace(4)* %out, i32 addrspace(4)* %out2, i32 %in, i64 %index) {
+entry:
+  %ptr = getelementptr i32, i32 addrspace(4)* %out, i64 %index
+  %val = atomicrmw volatile max i32 addrspace(4)* %ptr, i32 %in seq_cst
+  store i32 %val, i32 addrspace(4)* %out2
+  ret void
+}
+
+; GCN-LABEL: {{^}}atomic_umax_i32_offset:
+; GCN: flat_atomic_umax v[{{[0-9]+:[0-9]+}}], v{{[0-9]+$}}
+define void @atomic_umax_i32_offset(i32 addrspace(4)* %out, i32 %in) {
+entry:
+  %gep = getelementptr i32, i32 addrspace(4)* %out, i32 4
+  %val = atomicrmw volatile umax i32 addrspace(4)* %gep, i32 %in seq_cst
+  ret void
+}
+
+; GCN-LABEL: {{^}}atomic_umax_i32_ret_offset:
+; GCN: flat_atomic_umax [[RET:v[0-9]+]], v[{{[0-9]+:[0-9]+}}], v{{[0-9]+}} glc{{$}}
+; GCN: flat_store_dword v{{\[[0-9]+:[0-9]+\]}}, [[RET]]
+define void @atomic_umax_i32_ret_offset(i32 addrspace(4)* %out, i32 addrspace(4)* %out2, i32 %in) {
+entry:
+  %gep = getelementptr i32, i32 addrspace(4)* %out, i32 4
+  %val = atomicrmw volatile umax i32 addrspace(4)* %gep, i32 %in seq_cst
+  store i32 %val, i32 addrspace(4)* %out2
+  ret void
+}
+
+; GCN-LABEL: {{^}}atomic_umax_i32_addr64_offset:
+; GCN: flat_atomic_umax v[{{[0-9]+:[0-9]+}}], v{{[0-9]+$}}
+define void @atomic_umax_i32_addr64_offset(i32 addrspace(4)* %out, i32 %in, i64 %index) {
+entry:
+  %ptr = getelementptr i32, i32 addrspace(4)* %out, i64 %index
+  %gep = getelementptr i32, i32 addrspace(4)* %ptr, i32 4
+  %val = atomicrmw volatile umax i32 addrspace(4)* %gep, i32 %in seq_cst
+  ret void
+}
+
+; GCN-LABEL: {{^}}atomic_umax_i32_ret_addr64_offset:
+; GCN: flat_atomic_umax [[RET:v[0-9]+]], v[{{[0-9]+:[0-9]+}}], v{{[0-9]+}} glc{{$}}
+; GCN: flat_store_dword v{{\[[0-9]+:[0-9]+\]}}, [[RET]]
+define void @atomic_umax_i32_ret_addr64_offset(i32 addrspace(4)* %out, i32 addrspace(4)* %out2, i32 %in, i64 %index) {
+entry:
+  %ptr = getelementptr i32, i32 addrspace(4)* %out, i64 %index
+  %gep = getelementptr i32, i32 addrspace(4)* %ptr, i32 4
+  %val = atomicrmw volatile umax i32 addrspace(4)* %gep, i32 %in seq_cst
+  store i32 %val, i32 addrspace(4)* %out2
+  ret void
+}
+
+; GCN-LABEL: {{^}}atomic_umax_i32:
+; GCN: flat_atomic_umax v[{{[0-9]+:[0-9]+}}], v{{[0-9]+$}}
+define void @atomic_umax_i32(i32 addrspace(4)* %out, i32 %in) {
+entry:
+  %val = atomicrmw volatile umax i32 addrspace(4)* %out, i32 %in seq_cst
+  ret void
+}
+
+; GCN-LABEL: {{^}}atomic_umax_i32_ret:
+; GCN: flat_atomic_umax [[RET:v[0-9]+]], v[{{[0-9]+:[0-9]+}}], v{{[0-9]+}} glc{{$}}
+; GCN: flat_store_dword v{{\[[0-9]+:[0-9]+\]}}, [[RET]]
+define void @atomic_umax_i32_ret(i32 addrspace(4)* %out, i32 addrspace(4)* %out2, i32 %in) {
+entry:
+  %val = atomicrmw volatile umax i32 addrspace(4)* %out, i32 %in seq_cst
+  store i32 %val, i32 addrspace(4)* %out2
+  ret void
+}
+
+; GCN-LABEL: {{^}}atomic_umax_i32_addr64:
+; GCN: flat_atomic_umax v[{{[0-9]+:[0-9]+}}], v{{[0-9]+$}}
+define void @atomic_umax_i32_addr64(i32 addrspace(4)* %out, i32 %in, i64 %index) {
+entry:
+  %ptr = getelementptr i32, i32 addrspace(4)* %out, i64 %index
+  %val = atomicrmw volatile umax i32 addrspace(4)* %ptr, i32 %in seq_cst
+  ret void
+}
+
+; GCN-LABEL: {{^}}atomic_umax_i32_ret_addr64:
+; GCN: flat_atomic_umax [[RET:v[0-9]+]], v[{{[0-9]+:[0-9]+}}], v{{[0-9]+}} glc{{$}}
+; GCN: flat_store_dword v{{\[[0-9]+:[0-9]+\]}}, [[RET]]
+define void @atomic_umax_i32_ret_addr64(i32 addrspace(4)* %out, i32 addrspace(4)* %out2, i32 %in, i64 %index) {
+entry:
+  %ptr = getelementptr i32, i32 addrspace(4)* %out, i64 %index
+  %val = atomicrmw volatile umax i32 addrspace(4)* %ptr, i32 %in seq_cst
+  store i32 %val, i32 addrspace(4)* %out2
+  ret void
+}
+
+; GCN-LABEL: {{^}}atomic_min_i32_offset:
+; GCN: flat_atomic_smin v[{{[0-9]+:[0-9]+}}], v{{[0-9]+$}}
+define void @atomic_min_i32_offset(i32 addrspace(4)* %out, i32 %in) {
+entry:
+  %gep = getelementptr i32, i32 addrspace(4)* %out, i32 4
+  %val = atomicrmw volatile min i32 addrspace(4)* %gep, i32 %in seq_cst
+  ret void
+}
+
+; GCN-LABEL: {{^}}atomic_min_i32_ret_offset:
+; GCN: flat_atomic_smin [[RET:v[0-9]+]], v[{{[0-9]+:[0-9]+}}], v{{[0-9]+}} glc{{$}}
+; GCN: flat_store_dword v{{\[[0-9]+:[0-9]+\]}}, [[RET]]
+define void @atomic_min_i32_ret_offset(i32 addrspace(4)* %out, i32 addrspace(4)* %out2, i32 %in) {
+entry:
+  %gep = getelementptr i32, i32 addrspace(4)* %out, i32 4
+  %val = atomicrmw volatile min i32 addrspace(4)* %gep, i32 %in seq_cst
+  store i32 %val, i32 addrspace(4)* %out2
+  ret void
+}
+
+; GCN-LABEL: {{^}}atomic_min_i32_addr64_offset:
+; GCN: flat_atomic_smin v[{{[0-9]+:[0-9]+}}], v{{[0-9]+$}}
+define void @atomic_min_i32_addr64_offset(i32 addrspace(4)* %out, i32 %in, i64 %index) {
+entry:
+  %ptr = getelementptr i32, i32 addrspace(4)* %out, i64 %index
+  %gep = getelementptr i32, i32 addrspace(4)* %ptr, i32 4
+  %val = atomicrmw volatile min i32 addrspace(4)* %gep, i32 %in seq_cst
+  ret void
+}
+
+; GCN-LABEL: {{^}}atomic_min_i32_ret_addr64_offset:
+; GCN: flat_atomic_smin [[RET:v[0-9]+]], v[{{[0-9]+:[0-9]+}}], v{{[0-9]+}} glc{{$}}
+; GCN: flat_store_dword v{{\[[0-9]+:[0-9]+\]}}, [[RET]]
+define void @atomic_min_i32_ret_addr64_offset(i32 addrspace(4)* %out, i32 addrspace(4)* %out2, i32 %in, i64 %index) {
+entry:
+  %ptr = getelementptr i32, i32 addrspace(4)* %out, i64 %index
+  %gep = getelementptr i32, i32 addrspace(4)* %ptr, i32 4
+  %val = atomicrmw volatile min i32 addrspace(4)* %gep, i32 %in seq_cst
+  store i32 %val, i32 addrspace(4)* %out2
+  ret void
+}
+
+; GCN-LABEL: {{^}}atomic_min_i32:
+; GCN: flat_atomic_smin v[{{[0-9]+:[0-9]+}}], v{{[0-9]+$}}
+define void @atomic_min_i32(i32 addrspace(4)* %out, i32 %in) {
+entry:
+  %val = atomicrmw volatile min i32 addrspace(4)* %out, i32 %in seq_cst
+  ret void
+}
+
+; GCN-LABEL: {{^}}atomic_min_i32_ret:
+; GCN: flat_atomic_smin [[RET:v[0-9]+]], v[{{[0-9]+:[0-9]+}}], v{{[0-9]+}} glc{{$}}
+; GCN: flat_store_dword v{{\[[0-9]+:[0-9]+\]}}, [[RET]]
+define void @atomic_min_i32_ret(i32 addrspace(4)* %out, i32 addrspace(4)* %out2, i32 %in) {
+entry:
+  %val = atomicrmw volatile min i32 addrspace(4)* %out, i32 %in seq_cst
+  store i32 %val, i32 addrspace(4)* %out2
+  ret void
+}
+
+; GCN-LABEL: {{^}}atomic_min_i32_addr64:
+; GCN: flat_atomic_smin v[{{[0-9]+:[0-9]+}}], v{{[0-9]+$}}
+define void @atomic_min_i32_addr64(i32 addrspace(4)* %out, i32 %in, i64 %index) {
+entry:
+  %ptr = getelementptr i32, i32 addrspace(4)* %out, i64 %index
+  %val = atomicrmw volatile min i32 addrspace(4)* %ptr, i32 %in seq_cst
+  ret void
+}
+
+; GCN-LABEL: {{^}}atomic_min_i32_ret_addr64:
+; GCN: flat_atomic_smin [[RET:v[0-9]+]], v[{{[0-9]+:[0-9]+}}], v{{[0-9]+}} glc{{$}}
+; GCN: flat_store_dword v{{\[[0-9]+:[0-9]+\]}}, [[RET]]
+define void @atomic_min_i32_ret_addr64(i32 addrspace(4)* %out, i32 addrspace(4)* %out2, i32 %in, i64 %index) {
+entry:
+  %ptr = getelementptr i32, i32 addrspace(4)* %out, i64 %index
+  %val = atomicrmw volatile min i32 addrspace(4)* %ptr, i32 %in seq_cst
+  store i32 %val, i32 addrspace(4)* %out2
+  ret void
+}
+
+; GCN-LABEL: {{^}}atomic_umin_i32_offset:
+; GCN: flat_atomic_umin v[{{[0-9]+:[0-9]+}}], v{{[0-9]+$}}
+define void @atomic_umin_i32_offset(i32 addrspace(4)* %out, i32 %in) {
+entry:
+  %gep = getelementptr i32, i32 addrspace(4)* %out, i32 4
+  %val = atomicrmw volatile umin i32 addrspace(4)* %gep, i32 %in seq_cst
+  ret void
+}
+
+; GCN-LABEL: {{^}}atomic_umin_i32_ret_offset:
+; GCN: flat_atomic_umin [[RET:v[0-9]+]], v[{{[0-9]+:[0-9]+}}], v{{[0-9]+}} glc{{$}}
+; GCN: flat_store_dword v{{\[[0-9]+:[0-9]+\]}}, [[RET]]
+define void @atomic_umin_i32_ret_offset(i32 addrspace(4)* %out, i32 addrspace(4)* %out2, i32 %in) {
+entry:
+  %gep = getelementptr i32, i32 addrspace(4)* %out, i32 4
+  %val = atomicrmw volatile umin i32 addrspace(4)* %gep, i32 %in seq_cst
+  store i32 %val, i32 addrspace(4)* %out2
+  ret void
+}
+
+; GCN-LABEL: {{^}}atomic_umin_i32_addr64_offset:
+; GCN: flat_atomic_umin v[{{[0-9]+:[0-9]+}}], v{{[0-9]+$}}
+define void @atomic_umin_i32_addr64_offset(i32 addrspace(4)* %out, i32 %in, i64 %index) {
+entry:
+  %ptr = getelementptr i32, i32 addrspace(4)* %out, i64 %index
+  %gep = getelementptr i32, i32 addrspace(4)* %ptr, i32 4
+  %val = atomicrmw volatile umin i32 addrspace(4)* %gep, i32 %in seq_cst
+  ret void
+}
+
+; GCN-LABEL: {{^}}atomic_umin_i32_ret_addr64_offset:
+; GCN: flat_atomic_umin [[RET:v[0-9]+]], v[{{[0-9]+:[0-9]+}}], v{{[0-9]+}} glc{{$}}
+; GCN: flat_store_dword v{{\[[0-9]+:[0-9]+\]}}, [[RET]]
+define void @atomic_umin_i32_ret_addr64_offset(i32 addrspace(4)* %out, i32 addrspace(4)* %out2, i32 %in, i64 %index) {
+entry:
+  %ptr = getelementptr i32, i32 addrspace(4)* %out, i64 %index
+  %gep = getelementptr i32, i32 addrspace(4)* %ptr, i32 4
+  %val = atomicrmw volatile umin i32 addrspace(4)* %gep, i32 %in seq_cst
+  store i32 %val, i32 addrspace(4)* %out2
+  ret void
+}
+
+; GCN-LABEL: {{^}}atomic_umin_i32:
+; GCN: flat_atomic_umin v[{{[0-9]+:[0-9]+}}], v{{[0-9]+$}}
+define void @atomic_umin_i32(i32 addrspace(4)* %out, i32 %in) {
+entry:
+  %val = atomicrmw volatile umin i32 addrspace(4)* %out, i32 %in seq_cst
+  ret void
+}
+
+; GCN-LABEL: {{^}}atomic_umin_i32_ret:
+; GCN: flat_atomic_umin v{{[0-9]+}}, v[{{[0-9]+:[0-9]+}}], v{{[0-9]+}} glc{{$}}
+; GCN: flat_store_dword v{{\[[0-9]+:[0-9]+\]}}, [[RET]]
+define void @atomic_umin_i32_ret(i32 addrspace(4)* %out, i32 addrspace(4)* %out2, i32 %in) {
+entry:
+  %val = atomicrmw volatile umin i32 addrspace(4)* %out, i32 %in seq_cst
+  store i32 %val, i32 addrspace(4)* %out2
+  ret void
+}
+
+; GCN-LABEL: {{^}}atomic_umin_i32_addr64:
+; GCN: flat_atomic_umin v[{{[0-9]+:[0-9]+}}], v{{[0-9]+$}}
+define void @atomic_umin_i32_addr64(i32 addrspace(4)* %out, i32 %in, i64 %index) {
+entry:
+  %ptr = getelementptr i32, i32 addrspace(4)* %out, i64 %index
+  %val = atomicrmw volatile umin i32 addrspace(4)* %ptr, i32 %in seq_cst
+  ret void
+}
+
+; GCN-LABEL: {{^}}atomic_umin_i32_ret_addr64:
+; GCN: flat_atomic_umin [[RET:v[0-9]+]], v[{{[0-9]+:[0-9]+}}], v{{[0-9]+}} glc{{$}}
+; GCN: flat_store_dword v{{\[[0-9]+:[0-9]+\]}}, [[RET]]{{$}}
+  define void @atomic_umin_i32_ret_addr64(i32 addrspace(4)* %out, i32 addrspace(4)* %out2, i32 %in, i64 %index) {
+entry:
+  %ptr = getelementptr i32, i32 addrspace(4)* %out, i64 %index
+  %val = atomicrmw volatile umin i32 addrspace(4)* %ptr, i32 %in seq_cst
+  store i32 %val, i32 addrspace(4)* %out2
+  ret void
+}
+
+; GCN-LABEL: {{^}}atomic_or_i32_offset:
+; GCN: flat_atomic_or v[{{[0-9]+:[0-9]+}}], v{{[0-9]+}}{{$}}
+define void @atomic_or_i32_offset(i32 addrspace(4)* %out, i32 %in) {
+entry:
+  %gep = getelementptr i32, i32 addrspace(4)* %out, i32 4
+  %val = atomicrmw volatile or i32 addrspace(4)* %gep, i32 %in seq_cst
+  ret void
+}
+
+; GCN-LABEL: {{^}}atomic_or_i32_ret_offset:
+; GCN: flat_atomic_or [[RET:v[0-9]+]], v[{{[0-9]+:[0-9]+}}], v{{[0-9]+}} glc{{$}}
+; GCN: flat_store_dword v{{\[[0-9]+:[0-9]+\]}}, [[RET]]
+define void @atomic_or_i32_ret_offset(i32 addrspace(4)* %out, i32 addrspace(4)* %out2, i32 %in) {
+entry:
+  %gep = getelementptr i32, i32 addrspace(4)* %out, i32 4
+  %val = atomicrmw volatile or i32 addrspace(4)* %gep, i32 %in seq_cst
+  store i32 %val, i32 addrspace(4)* %out2
+  ret void
+}
+
+; GCN-LABEL: {{^}}atomic_or_i32_addr64_offset:
+; GCN: flat_atomic_or v[{{[0-9]+:[0-9]+}}], v{{[0-9]+}}{{$}}
+define void @atomic_or_i32_addr64_offset(i32 addrspace(4)* %out, i32 %in, i64 %index) {
+entry:
+  %ptr = getelementptr i32, i32 addrspace(4)* %out, i64 %index
+  %gep = getelementptr i32, i32 addrspace(4)* %ptr, i32 4
+  %val = atomicrmw volatile or i32 addrspace(4)* %gep, i32 %in seq_cst
+  ret void
+}
+
+; GCN-LABEL: {{^}}atomic_or_i32_ret_addr64_offset:
+; GCN: flat_atomic_or [[RET:v[0-9]+]], v[{{[0-9]+:[0-9]+}}], v{{[0-9]+}} glc{{$}}
+; GCN: flat_store_dword v{{\[[0-9]+:[0-9]+\]}}, [[RET]]
+define void @atomic_or_i32_ret_addr64_offset(i32 addrspace(4)* %out, i32 addrspace(4)* %out2, i32 %in, i64 %index) {
+entry:
+  %ptr = getelementptr i32, i32 addrspace(4)* %out, i64 %index
+  %gep = getelementptr i32, i32 addrspace(4)* %ptr, i32 4
+  %val = atomicrmw volatile or i32 addrspace(4)* %gep, i32 %in seq_cst
+  store i32 %val, i32 addrspace(4)* %out2
+  ret void
+}
+
+; GCN-LABEL: {{^}}atomic_or_i32:
+; GCN: flat_atomic_or v[{{[0-9]+:[0-9]+}}], v{{[0-9]+$}}
+define void @atomic_or_i32(i32 addrspace(4)* %out, i32 %in) {
+entry:
+  %val = atomicrmw volatile or i32 addrspace(4)* %out, i32 %in seq_cst
+  ret void
+}
+
+; GCN-LABEL: {{^}}atomic_or_i32_ret:
+; GCN: flat_atomic_or [[RET:v[0-9]+]], v[{{[0-9]+:[0-9]+}}], v{{[0-9]+}} glc{{$}}
+; GCN: flat_store_dword v{{\[[0-9]+:[0-9]+\]}}, [[RET]]
+define void @atomic_or_i32_ret(i32 addrspace(4)* %out, i32 addrspace(4)* %out2, i32 %in) {
+entry:
+  %val = atomicrmw volatile or i32 addrspace(4)* %out, i32 %in seq_cst
+  store i32 %val, i32 addrspace(4)* %out2
+  ret void
+}
+
+; GCN-LABEL: {{^}}atomic_or_i32_addr64:
+; GCN: flat_atomic_or v[{{[0-9]+:[0-9]+}}], v{{[0-9]+$}}
+define void @atomic_or_i32_addr64(i32 addrspace(4)* %out, i32 %in, i64 %index) {
+entry:
+  %ptr = getelementptr i32, i32 addrspace(4)* %out, i64 %index
+  %val = atomicrmw volatile or i32 addrspace(4)* %ptr, i32 %in seq_cst
+  ret void
+}
+
+; GCN-LABEL: {{^}}atomic_or_i32_ret_addr64:
+; GCN: flat_atomic_or [[RET:v[0-9]+]], v[{{[0-9]+:[0-9]+}}], v{{[0-9]+}} glc{{$}}
+; GCN: flat_store_dword v{{\[[0-9]+:[0-9]+\]}}, [[RET]]
+define void @atomic_or_i32_ret_addr64(i32 addrspace(4)* %out, i32 addrspace(4)* %out2, i32 %in, i64 %index) {
+entry:
+  %ptr = getelementptr i32, i32 addrspace(4)* %out, i64 %index
+  %val = atomicrmw volatile or i32 addrspace(4)* %ptr, i32 %in seq_cst
+  store i32 %val, i32 addrspace(4)* %out2
+  ret void
+}
+
+; GCN-LABEL: {{^}}atomic_xchg_i32_offset:
+; GCN: flat_atomic_swap v[{{[0-9]+:[0-9]+}}], v{{[0-9]+}}{{$}}
+define void @atomic_xchg_i32_offset(i32 addrspace(4)* %out, i32 %in) {
+entry:
+  %gep = getelementptr i32, i32 addrspace(4)* %out, i32 4
+  %val = atomicrmw volatile xchg i32 addrspace(4)* %gep, i32 %in seq_cst
+  ret void
+}
+
+; GCN-LABEL: {{^}}atomic_xchg_i32_ret_offset:
+; GCN: flat_atomic_swap [[RET:v[0-9]+]], v[{{[0-9]+:[0-9]+}}], v{{[0-9]+}} glc{{$}}
+; GCN: flat_store_dword v{{\[[0-9]+:[0-9]+\]}}, [[RET]]
+define void @atomic_xchg_i32_ret_offset(i32 addrspace(4)* %out, i32 addrspace(4)* %out2, i32 %in) {
+entry:
+  %gep = getelementptr i32, i32 addrspace(4)* %out, i32 4
+  %val = atomicrmw volatile xchg i32 addrspace(4)* %gep, i32 %in seq_cst
+  store i32 %val, i32 addrspace(4)* %out2
+  ret void
+}
+
+; GCN-LABEL: {{^}}atomic_xchg_i32_addr64_offset:
+; GCN: flat_atomic_swap v[{{[0-9]+:[0-9]+}}], v{{[0-9]+}}{{$}}
+define void @atomic_xchg_i32_addr64_offset(i32 addrspace(4)* %out, i32 %in, i64 %index) {
+entry:
+  %ptr = getelementptr i32, i32 addrspace(4)* %out, i64 %index
+  %gep = getelementptr i32, i32 addrspace(4)* %ptr, i32 4
+  %val = atomicrmw volatile xchg i32 addrspace(4)* %gep, i32 %in seq_cst
+  ret void
+}
+
+; GCN-LABEL: {{^}}atomic_xchg_i32_ret_addr64_offset:
+; GCN: flat_atomic_swap [[RET:v[0-9]+]], v[{{[0-9]+:[0-9]+}}], v{{[0-9]+}} glc{{$}}
+; GCN: flat_store_dword v{{\[[0-9]+:[0-9]+\]}}, [[RET]]
+define void @atomic_xchg_i32_ret_addr64_offset(i32 addrspace(4)* %out, i32 addrspace(4)* %out2, i32 %in, i64 %index) {
+entry:
+  %ptr = getelementptr i32, i32 addrspace(4)* %out, i64 %index
+  %gep = getelementptr i32, i32 addrspace(4)* %ptr, i32 4
+  %val = atomicrmw volatile xchg i32 addrspace(4)* %gep, i32 %in seq_cst
+  store i32 %val, i32 addrspace(4)* %out2
+  ret void
+}
+
+; GCN-LABEL: {{^}}atomic_xchg_i32:
+; GCN: flat_atomic_swap v[{{[0-9]+}}:{{[0-9]+}}], v{{[0-9]+}}{{$}}
+define void @atomic_xchg_i32(i32 addrspace(4)* %out, i32 %in) {
+entry:
+  %val = atomicrmw volatile xchg i32 addrspace(4)* %out, i32 %in seq_cst
+  ret void
+}
+
+; GCN-LABEL: {{^}}atomic_xchg_i32_ret:
+; GCN: flat_atomic_swap [[RET:v[0-9]+]], v[{{[0-9]+}}:{{[0-9]+}}], v{{[0-9]+}} glc{{$}}
+; GCN: flat_store_dword v{{\[[0-9]+:[0-9]+\]}}, [[RET]]
+define void @atomic_xchg_i32_ret(i32 addrspace(4)* %out, i32 addrspace(4)* %out2, i32 %in) {
+entry:
+  %val = atomicrmw volatile xchg i32 addrspace(4)* %out, i32 %in seq_cst
+  store i32 %val, i32 addrspace(4)* %out2
+  ret void
+}
+
+; GCN-LABEL: {{^}}atomic_xchg_i32_addr64:
+; GCN: flat_atomic_swap v[{{[0-9]+:[0-9]+}}], v{{[0-9]+$}}
+define void @atomic_xchg_i32_addr64(i32 addrspace(4)* %out, i32 %in, i64 %index) {
+entry:
+  %ptr = getelementptr i32, i32 addrspace(4)* %out, i64 %index
+  %val = atomicrmw volatile xchg i32 addrspace(4)* %ptr, i32 %in seq_cst
+  ret void
+}
+
+; GCN-LABEL: {{^}}atomic_xchg_i32_ret_addr64:
+; GCN: flat_atomic_swap [[RET:v[0-9]+]],  v[{{[0-9]+:[0-9]+}}], v{{[0-9]+}} glc{{$}}
+; GCN: flat_store_dword v{{\[[0-9]+:[0-9]+\]}}, [[RET]]
+define void @atomic_xchg_i32_ret_addr64(i32 addrspace(4)* %out, i32 addrspace(4)* %out2, i32 %in, i64 %index) {
+entry:
+  %ptr = getelementptr i32, i32 addrspace(4)* %out, i64 %index
+  %val = atomicrmw volatile xchg i32 addrspace(4)* %ptr, i32 %in seq_cst
+  store i32 %val, i32 addrspace(4)* %out2
+  ret void
+}
+
+; CMP_SWAP
+
+; GCN-LABEL: {{^}}atomic_cmpxchg_i32_offset:
+; GCN: flat_atomic_cmpswap v[{{[0-9]+\:[0-9]+}}], v[{{[0-9]+}}:{{[0-9]+}}]{{$}}
+define void @atomic_cmpxchg_i32_offset(i32 addrspace(4)* %out, i32 %in, i32 %old) {
+entry:
+  %gep = getelementptr i32, i32 addrspace(4)* %out, i32 4
+  %val = cmpxchg volatile i32 addrspace(4)* %gep, i32 %old, i32 %in seq_cst seq_cst
+  ret void
+}
+
+; GCN-LABEL: {{^}}atomic_cmpxchg_i32_ret_offset:
+; GCN: flat_atomic_cmpswap v[[RET:[0-9]+]], v[{{[0-9]+}}:{{[0-9]+}}], v[{{[0-9]+}}:{{[0-9]+}}] glc{{$}}
+; GCN: flat_store_dword v{{\[[0-9]+:[0-9]+\]}}, v[[RET]]
+define void @atomic_cmpxchg_i32_ret_offset(i32 addrspace(4)* %out, i32 addrspace(4)* %out2, i32 %in, i32 %old) {
+entry:
+  %gep = getelementptr i32, i32 addrspace(4)* %out, i32 4
+  %val = cmpxchg volatile i32 addrspace(4)* %gep, i32 %old, i32 %in seq_cst seq_cst
+  %flag = extractvalue { i32, i1 } %val, 0
+  store i32 %flag, i32 addrspace(4)* %out2
+  ret void
+}
+
+; GCN-LABEL: {{^}}atomic_cmpxchg_i32_addr64_offset:
+; GCN: flat_atomic_cmpswap v[{{[0-9]+\:[0-9]+}}], v[{{[0-9]+}}:{{[0-9]+}}]{{$}}
+define void @atomic_cmpxchg_i32_addr64_offset(i32 addrspace(4)* %out, i32 %in, i64 %index, i32 %old) {
+entry:
+  %ptr = getelementptr i32, i32 addrspace(4)* %out, i64 %index
+  %gep = getelementptr i32, i32 addrspace(4)* %ptr, i32 4
+  %val  = cmpxchg volatile i32 addrspace(4)* %gep, i32 %old, i32 %in seq_cst seq_cst
+  ret void
+}
+
+; GCN-LABEL: {{^}}atomic_cmpxchg_i32_ret_addr64_offset:
+; GCN: flat_atomic_cmpswap v[[RET:[0-9]+]], v[{{[0-9]+:[0-9]+}}], v[{{[0-9]+:[0-9]+}}] glc{{$}}
+; GCN: flat_store_dword v{{\[[0-9]+:[0-9]+\]}}, v[[RET]]
+define void @atomic_cmpxchg_i32_ret_addr64_offset(i32 addrspace(4)* %out, i32 addrspace(4)* %out2, i32 %in, i64 %index, i32 %old) {
+entry:
+  %ptr = getelementptr i32, i32 addrspace(4)* %out, i64 %index
+  %gep = getelementptr i32, i32 addrspace(4)* %ptr, i32 4
+  %val  = cmpxchg volatile i32 addrspace(4)* %gep, i32 %old, i32 %in seq_cst seq_cst
+  %flag = extractvalue { i32, i1 } %val, 0
+  store i32 %flag, i32 addrspace(4)* %out2
+  ret void
+}
+
+; GCN-LABEL: {{^}}atomic_cmpxchg_i32:
+; GCN: flat_atomic_cmpswap v[{{[0-9]+}}:{{[0-9]+}}], v[{{[0-9]+}}:{{[0-9]+}}]{{$}}
+define void @atomic_cmpxchg_i32(i32 addrspace(4)* %out, i32 %in, i32 %old) {
+entry:
+  %val = cmpxchg volatile i32 addrspace(4)* %out, i32 %old, i32 %in seq_cst seq_cst
+  ret void
+}
+
+; GCN-LABEL: {{^}}atomic_cmpxchg_i32_ret:
+; GCN: flat_atomic_cmpswap v[[RET:[0-9]+]], v[{{[0-9]+}}:{{[0-9]+}}], v[{{[0-9]+}}:{{[0-9]+}}] glc
+; GCN: flat_store_dword v{{\[[0-9]+:[0-9]+\]}}, v[[RET]]
+define void @atomic_cmpxchg_i32_ret(i32 addrspace(4)* %out, i32 addrspace(4)* %out2, i32 %in, i32 %old) {
+entry:
+  %val = cmpxchg volatile i32 addrspace(4)* %out, i32 %old, i32 %in seq_cst seq_cst
+  %flag = extractvalue { i32, i1 } %val, 0
+  store i32 %flag, i32 addrspace(4)* %out2
+  ret void
+}
+
+; GCN-LABEL: {{^}}atomic_cmpxchg_i32_addr64:
+; GCN: flat_atomic_cmpswap v[{{[0-9]+:[0-9]+}}], v[{{[0-9]+:[0-9]+}}]{{$}}
+define void @atomic_cmpxchg_i32_addr64(i32 addrspace(4)* %out, i32 %in, i64 %index, i32 %old) {
+entry:
+  %ptr = getelementptr i32, i32 addrspace(4)* %out, i64 %index
+  %val = cmpxchg volatile i32 addrspace(4)* %ptr, i32 %old, i32 %in seq_cst seq_cst
+  ret void
+}
+
+; GCN-LABEL: {{^}}atomic_cmpxchg_i32_ret_addr64:
+; GCN: flat_atomic_cmpswap v[[RET:[0-9]+]], v[{{[0-9]+:[0-9]+}}], v[{{[0-9]+:[0-9]+}}] glc{{$}}
+; GCN: flat_store_dword v{{\[[0-9]+:[0-9]+\]}}, v[[RET]]
+define void @atomic_cmpxchg_i32_ret_addr64(i32 addrspace(4)* %out, i32 addrspace(4)* %out2, i32 %in, i64 %index, i32 %old) {
+entry:
+  %ptr = getelementptr i32, i32 addrspace(4)* %out, i64 %index
+  %val = cmpxchg volatile i32 addrspace(4)* %ptr, i32 %old, i32 %in seq_cst seq_cst
+  %flag = extractvalue { i32, i1 } %val, 0
+  store i32 %flag, i32 addrspace(4)* %out2
+  ret void
+}
+
+; GCN-LABEL: {{^}}atomic_xor_i32_offset:
+; GCN: flat_atomic_xor v[{{[0-9]+}}:{{[0-9]+}}], v{{[0-9]+}}{{$}}
+define void @atomic_xor_i32_offset(i32 addrspace(4)* %out, i32 %in) {
+entry:
+  %gep = getelementptr i32, i32 addrspace(4)* %out, i32 4
+  %val = atomicrmw volatile xor i32 addrspace(4)* %gep, i32 %in seq_cst
+  ret void
+}
+
+; GCN-LABEL: {{^}}atomic_xor_i32_ret_offset:
+; GCN: flat_atomic_xor [[RET:v[0-9]+]], v[{{[0-9]+}}:{{[0-9]+}}], v{{[0-9]+}} glc{{$}}
+; GCN: flat_store_dword v{{\[[0-9]+:[0-9]+\]}}, [[RET]]
+define void @atomic_xor_i32_ret_offset(i32 addrspace(4)* %out, i32 addrspace(4)* %out2, i32 %in) {
+entry:
+  %gep = getelementptr i32, i32 addrspace(4)* %out, i32 4
+  %val = atomicrmw volatile xor i32 addrspace(4)* %gep, i32 %in seq_cst
+  store i32 %val, i32 addrspace(4)* %out2
+  ret void
+}
+
+; GCN-LABEL: {{^}}atomic_xor_i32_addr64_offset:
+; GCN: flat_atomic_xor v[{{[0-9]+:[0-9]+}}], v{{[0-9]+$}}
+define void @atomic_xor_i32_addr64_offset(i32 addrspace(4)* %out, i32 %in, i64 %index) {
+entry:
+  %ptr = getelementptr i32, i32 addrspace(4)* %out, i64 %index
+  %gep = getelementptr i32, i32 addrspace(4)* %ptr, i32 4
+  %val = atomicrmw volatile xor i32 addrspace(4)* %gep, i32 %in seq_cst
+  ret void
+}
+
+; GCN-LABEL: {{^}}atomic_xor_i32_ret_addr64_offset:
+; GCN: flat_atomic_xor [[RET:v[0-9]+]], v[{{[0-9]+:[0-9]+}}], v{{[0-9]+}} glc{{$}}
+; GCN: flat_store_dword v{{\[[0-9]+:[0-9]+\]}}, [[RET]]
+define void @atomic_xor_i32_ret_addr64_offset(i32 addrspace(4)* %out, i32 addrspace(4)* %out2, i32 %in, i64 %index) {
+entry:
+  %ptr = getelementptr i32, i32 addrspace(4)* %out, i64 %index
+  %gep = getelementptr i32, i32 addrspace(4)* %ptr, i32 4
+  %val = atomicrmw volatile xor i32 addrspace(4)* %gep, i32 %in seq_cst
+  store i32 %val, i32 addrspace(4)* %out2
+  ret void
+}
+
+; GCN-LABEL: {{^}}atomic_xor_i32:
+; GCN: flat_atomic_xor v[{{[0-9]+}}:{{[0-9]+}}], v{{[0-9]+}}{{$}}
+define void @atomic_xor_i32(i32 addrspace(4)* %out, i32 %in) {
+entry:
+  %val = atomicrmw volatile xor i32 addrspace(4)* %out, i32 %in seq_cst
+  ret void
+}
+
+; GCN-LABEL: {{^}}atomic_xor_i32_ret:
+; GCN: flat_atomic_xor [[RET:v[0-9]+]], v[{{[0-9]+}}:{{[0-9]+}}], v{{[0-9]+}} glc{{$}}
+; GCN: flat_store_dword v{{\[[0-9]+:[0-9]+\]}}, [[RET]]
+define void @atomic_xor_i32_ret(i32 addrspace(4)* %out, i32 addrspace(4)* %out2, i32 %in) {
+entry:
+  %val = atomicrmw volatile xor i32 addrspace(4)* %out, i32 %in seq_cst
+  store i32 %val, i32 addrspace(4)* %out2
+  ret void
+}
+
+; GCN-LABEL: {{^}}atomic_xor_i32_addr64:
+; GCN: flat_atomic_xor v[{{[0-9]+:[0-9]+}}], v{{[0-9]+$}}
+define void @atomic_xor_i32_addr64(i32 addrspace(4)* %out, i32 %in, i64 %index) {
+entry:
+  %ptr = getelementptr i32, i32 addrspace(4)* %out, i64 %index
+  %val = atomicrmw volatile xor i32 addrspace(4)* %ptr, i32 %in seq_cst
+  ret void
+}
+
+; GCN-LABEL: {{^}}atomic_xor_i32_ret_addr64:
+; GCN: flat_atomic_xor [[RET:v[0-9]+]], v[{{[0-9]+:[0-9]+}}], v{{[0-9]+}} glc{{$}}
+; GCN: flat_store_dword v{{\[[0-9]+:[0-9]+\]}}, [[RET]]
+define void @atomic_xor_i32_ret_addr64(i32 addrspace(4)* %out, i32 addrspace(4)* %out2, i32 %in, i64 %index) {
+entry:
+  %ptr = getelementptr i32, i32 addrspace(4)* %out, i64 %index
+  %val = atomicrmw volatile xor i32 addrspace(4)* %ptr, i32 %in seq_cst
+  store i32 %val, i32 addrspace(4)* %out2
+  ret void
+}
+
+; GCN-LABEL: {{^}}atomic_load_i32_offset:
+; GCN: flat_load_dword [[RET:v[0-9]+]], v[{{[0-9]+}}:{{[0-9]+}}] glc{{$}}
+; GCN: flat_store_dword v{{\[[0-9]+:[0-9]+\]}}, [[RET]]
+define void @atomic_load_i32_offset(i32 addrspace(4)* %in, i32 addrspace(4)* %out) {
+entry:
+  %gep = getelementptr i32, i32 addrspace(4)* %in, i32 4
+  %val = load atomic i32, i32 addrspace(4)* %gep  seq_cst, align 4
+  store i32 %val, i32 addrspace(4)* %out
+  ret void
+}
+
+; GCN-LABEL: {{^}}atomic_load_i32:
+; GCN: flat_load_dword [[RET:v[0-9]+]], v[{{[0-9]+}}:{{[0-9]+}}] glc
+; GCN: flat_store_dword v{{\[[0-9]+:[0-9]+\]}}, [[RET]]
+define void @atomic_load_i32(i32 addrspace(4)* %in, i32 addrspace(4)* %out) {
+entry:
+  %val = load atomic i32, i32 addrspace(4)* %in seq_cst, align 4
+  store i32 %val, i32 addrspace(4)* %out
+  ret void
+}
+
+; GCN-LABEL: {{^}}atomic_load_i32_addr64_offset:
+; GCN: flat_load_dword [[RET:v[0-9]+]], v[{{[0-9]+:[0-9]+}}] glc{{$}}
+; GCN: flat_store_dword v{{\[[0-9]+:[0-9]+\]}}, [[RET]]
+define void @atomic_load_i32_addr64_offset(i32 addrspace(4)* %in, i32 addrspace(4)* %out, i64 %index) {
+entry:
+  %ptr = getelementptr i32, i32 addrspace(4)* %in, i64 %index
+  %gep = getelementptr i32, i32 addrspace(4)* %ptr, i32 4
+  %val = load atomic i32, i32 addrspace(4)* %gep seq_cst, align 4
+  store i32 %val, i32 addrspace(4)* %out
+  ret void
+}
+
+; GCN-LABEL: {{^}}atomic_load_i32_addr64:
+; GCN: flat_load_dword [[RET:v[0-9]+]], v[{{[0-9]+:[0-9]+}}] glc{{$}}
+; GCN: flat_store_dword v{{\[[0-9]+:[0-9]+\]}}, [[RET]]
+define void @atomic_load_i32_addr64(i32 addrspace(4)* %in, i32 addrspace(4)* %out, i64 %index) {
+entry:
+  %ptr = getelementptr i32, i32 addrspace(4)* %in, i64 %index
+  %val = load atomic i32, i32 addrspace(4)* %ptr seq_cst, align 4
+  store i32 %val, i32 addrspace(4)* %out
+  ret void
+}
+
+; GCN-LABEL: {{^}}atomic_store_i32_offset:
+; GCN: flat_store_dword v[{{[0-9]+}}:{{[0-9]+}}], {{v[0-9]+}} glc{{$}}
+define void @atomic_store_i32_offset(i32 %in, i32 addrspace(4)* %out) {
+entry:
+  %gep = getelementptr i32, i32 addrspace(4)* %out, i32 4
+  store atomic i32 %in, i32 addrspace(4)* %gep  seq_cst, align 4
+  ret void
+}
+
+; GCN-LABEL: {{^}}atomic_store_i32:
+; GCN: flat_store_dword v[{{[0-9]+}}:{{[0-9]+}}], {{v[0-9]+}} glc{{$}}
+define void @atomic_store_i32(i32 %in, i32 addrspace(4)* %out) {
+entry:
+  store atomic i32 %in, i32 addrspace(4)* %out seq_cst, align 4
+  ret void
+}
+
+; GCN-LABEL: {{^}}atomic_store_i32_addr64_offset:
+; GCN: flat_store_dword v[{{[0-9]+}}:{{[0-9]+}}], {{v[0-9]+}} glc{{$}}
+define void @atomic_store_i32_addr64_offset(i32 %in, i32 addrspace(4)* %out, i64 %index) {
+entry:
+  %ptr = getelementptr i32, i32 addrspace(4)* %out, i64 %index
+  %gep = getelementptr i32, i32 addrspace(4)* %ptr, i32 4
+  store atomic i32 %in, i32 addrspace(4)* %gep seq_cst, align 4
+  ret void
+}
+
+; GCN-LABEL: {{^}}atomic_store_i32_addr64:
+; GCN: flat_store_dword v[{{[0-9]+}}:{{[0-9]+}}], {{v[0-9]+}} glc{{$}}
+define void @atomic_store_i32_addr64(i32 %in, i32 addrspace(4)* %out, i64 %index) {
+entry:
+  %ptr = getelementptr i32, i32 addrspace(4)* %out, i64 %index
+  store atomic i32 %in, i32 addrspace(4)* %ptr seq_cst, align 4
+  ret void
+}
diff --git a/llvm/test/CodeGen/AMDGPU/flat_atomics_i64.ll b/llvm/test/CodeGen/AMDGPU/flat_atomics_i64.ll
new file mode 100644
index 00000000000..0bd6c2dd5b8
--- /dev/null
+++ b/llvm/test/CodeGen/AMDGPU/flat_atomics_i64.ll
@@ -0,0 +1,975 @@
+; RUN: llc -march=amdgcn -mcpu=bonaire -verify-machineinstrs < %s | FileCheck -check-prefix=GCN %s
+; RUN: llc -march=amdgcn -mcpu=tonga -verify-machineinstrs < %s | FileCheck -check-prefix=GCN %s
+
+; GCN-LABEL: {{^}}atomic_add_i64_offset:
+; GCN: flat_atomic_add_x2 v[{{[0-9]+:[0-9]+}}], v{{\[[0-9]+:[0-9]+\]}}{{$}}
+define void @atomic_add_i64_offset(i64 addrspace(4)* %out, i64 %in) {
+entry:
+  %gep = getelementptr i64, i64 addrspace(4)* %out, i64 4
+  %tmp0 = atomicrmw volatile add i64 addrspace(4)* %gep, i64 %in seq_cst
+  ret void
+}
+
+; GCN-LABEL: {{^}}atomic_add_i64_ret_offset:
+; GCN: flat_atomic_add_x2 [[RET:v\[[0-9]+:[0-9]+\]]], v[{{[0-9]+:[0-9]+}}], v{{\[[0-9]+:[0-9]+\]}} glc{{$}}
+; GCN: flat_store_dwordx2 v{{\[[0-9]+:[0-9]+\]}}, [[RET]]
+define void @atomic_add_i64_ret_offset(i64 addrspace(4)* %out, i64 addrspace(4)* %out2, i64 %in) {
+entry:
+  %gep = getelementptr i64, i64 addrspace(4)* %out, i64 4
+  %tmp0 = atomicrmw volatile add i64 addrspace(4)* %gep, i64 %in seq_cst
+  store i64 %tmp0, i64 addrspace(4)* %out2
+  ret void
+}
+
+; GCN-LABEL: {{^}}atomic_add_i64_addr64_offset:
+; GCN: flat_atomic_add_x2 v[{{[0-9]+:[0-9]+}}], v{{\[[0-9]+:[0-9]+\]}}{{$}}
+define void @atomic_add_i64_addr64_offset(i64 addrspace(4)* %out, i64 %in, i64 %index) {
+entry:
+  %ptr = getelementptr i64, i64 addrspace(4)* %out, i64 %index
+  %gep = getelementptr i64, i64 addrspace(4)* %ptr, i64 4
+  %tmp0 = atomicrmw volatile add i64 addrspace(4)* %gep, i64 %in seq_cst
+  ret void
+}
+
+; GCN-LABEL: {{^}}atomic_add_i64_ret_addr64_offset:
+; GCN: flat_atomic_add_x2 [[RET:v\[[0-9]+:[0-9]+\]]], v[{{[0-9]+:[0-9]+}}], v{{\[[0-9]+:[0-9]+\]}} glc{{$}}
+; GCN: flat_store_dwordx2 v{{\[[0-9]+:[0-9]+\]}}, [[RET]]
+define void @atomic_add_i64_ret_addr64_offset(i64 addrspace(4)* %out, i64 addrspace(4)* %out2, i64 %in, i64 %index) {
+entry:
+  %ptr = getelementptr i64, i64 addrspace(4)* %out, i64 %index
+  %gep = getelementptr i64, i64 addrspace(4)* %ptr, i64 4
+  %tmp0 = atomicrmw volatile add i64 addrspace(4)* %gep, i64 %in seq_cst
+  store i64 %tmp0, i64 addrspace(4)* %out2
+  ret void
+}
+
+; GCN-LABEL: {{^}}atomic_add_i64:
+; GCN: flat_atomic_add_x2 v[{{[0-9]+:[0-9]+}}], v{{\[[0-9]+:[0-9]+\]$}}
+define void @atomic_add_i64(i64 addrspace(4)* %out, i64 %in) {
+entry:
+  %tmp0 = atomicrmw volatile add i64 addrspace(4)* %out, i64 %in seq_cst
+  ret void
+}
+
+; GCN-LABEL: {{^}}atomic_add_i64_ret:
+; GCN: flat_atomic_add_x2 [[RET:v\[[0-9]+:[0-9]+\]]], v[{{[0-9]+:[0-9]+}}], v{{\[[0-9]+:[0-9]+\]}} glc{{$}}
+; GCN: flat_store_dwordx2 v{{\[[0-9]+:[0-9]+\]}}, [[RET]]
+define void @atomic_add_i64_ret(i64 addrspace(4)* %out, i64 addrspace(4)* %out2, i64 %in) {
+entry:
+  %tmp0 = atomicrmw volatile add i64 addrspace(4)* %out, i64 %in seq_cst
+  store i64 %tmp0, i64 addrspace(4)* %out2
+  ret void
+}
+
+; GCN-LABEL: {{^}}atomic_add_i64_addr64:
+; GCN: flat_atomic_add_x2 v[{{[0-9]+:[0-9]+}}], v{{\[[0-9]+:[0-9]+\]$}}
+define void @atomic_add_i64_addr64(i64 addrspace(4)* %out, i64 %in, i64 %index) {
+entry:
+  %ptr = getelementptr i64, i64 addrspace(4)* %out, i64 %index
+  %tmp0 = atomicrmw volatile add i64 addrspace(4)* %ptr, i64 %in seq_cst
+  ret void
+}
+
+; GCN-LABEL: {{^}}atomic_add_i64_ret_addr64:
+; GCN: flat_atomic_add_x2 [[RET:v\[[0-9]+:[0-9]+\]]], v[{{[0-9]+:[0-9]+}}], v{{\[[0-9]+:[0-9]+\]}} glc{{$}}
+; GCN: flat_store_dwordx2 v{{\[[0-9]+:[0-9]+\]}}, [[RET]]
+define void @atomic_add_i64_ret_addr64(i64 addrspace(4)* %out, i64 addrspace(4)* %out2, i64 %in, i64 %index) {
+entry:
+  %ptr = getelementptr i64, i64 addrspace(4)* %out, i64 %index
+  %tmp0 = atomicrmw volatile add i64 addrspace(4)* %ptr, i64 %in seq_cst
+  store i64 %tmp0, i64 addrspace(4)* %out2
+  ret void
+}
+
+; GCN-LABEL: {{^}}atomic_and_i64_offset:
+; GCN: flat_atomic_and_x2 v[{{[0-9]+:[0-9]+}}], v{{\[[0-9]+:[0-9]+\]$}}
+define void @atomic_and_i64_offset(i64 addrspace(4)* %out, i64 %in) {
+entry:
+  %gep = getelementptr i64, i64 addrspace(4)* %out, i64 4
+  %tmp0 = atomicrmw volatile and i64 addrspace(4)* %gep, i64 %in seq_cst
+  ret void
+}
+
+; GCN-LABEL: {{^}}atomic_and_i64_ret_offset:
+; GCN: flat_atomic_and_x2 [[RET:v\[[0-9]+:[0-9]+\]]], v[{{[0-9]+:[0-9]+}}], v{{\[[0-9]+:[0-9]+\]}} glc{{$}}
+; GCN: flat_store_dwordx2 v{{\[[0-9]+:[0-9]+\]}}, [[RET]]
+define void @atomic_and_i64_ret_offset(i64 addrspace(4)* %out, i64 addrspace(4)* %out2, i64 %in) {
+entry:
+  %gep = getelementptr i64, i64 addrspace(4)* %out, i64 4
+  %tmp0 = atomicrmw volatile and i64 addrspace(4)* %gep, i64 %in seq_cst
+  store i64 %tmp0, i64 addrspace(4)* %out2
+  ret void
+}
+
+; GCN-LABEL: {{^}}atomic_and_i64_addr64_offset:
+; GCN: flat_atomic_and_x2 v[{{[0-9]+:[0-9]+}}], v{{\[[0-9]+:[0-9]+\]$}}
+define void @atomic_and_i64_addr64_offset(i64 addrspace(4)* %out, i64 %in, i64 %index) {
+entry:
+  %ptr = getelementptr i64, i64 addrspace(4)* %out, i64 %index
+  %gep = getelementptr i64, i64 addrspace(4)* %ptr, i64 4
+  %tmp0 = atomicrmw volatile and i64 addrspace(4)* %gep, i64 %in seq_cst
+  ret void
+}
+
+; GCN-LABEL: {{^}}atomic_and_i64_ret_addr64_offset:
+; GCN: flat_atomic_and_x2 [[RET:v\[[0-9]+:[0-9]+\]]], v[{{[0-9]+:[0-9]+}}], v{{\[[0-9]+:[0-9]+\]}} glc{{$}}
+; GCN: flat_store_dwordx2 v{{\[[0-9]+:[0-9]+\]}}, [[RET]]
+define void @atomic_and_i64_ret_addr64_offset(i64 addrspace(4)* %out, i64 addrspace(4)* %out2, i64 %in, i64 %index) {
+entry:
+  %ptr = getelementptr i64, i64 addrspace(4)* %out, i64 %index
+  %gep = getelementptr i64, i64 addrspace(4)* %ptr, i64 4
+  %tmp0 = atomicrmw volatile and i64 addrspace(4)* %gep, i64 %in seq_cst
+  store i64 %tmp0, i64 addrspace(4)* %out2
+  ret void
+}
+
+; GCN-LABEL: {{^}}atomic_and_i64:
+; GCN: flat_atomic_and_x2 v[{{[0-9]+:[0-9]+}}], v{{\[[0-9]+:[0-9]+\]$}}
+define void @atomic_and_i64(i64 addrspace(4)* %out, i64 %in) {
+entry:
+  %tmp0 = atomicrmw volatile and i64 addrspace(4)* %out, i64 %in seq_cst
+  ret void
+}
+
+; GCN-LABEL: {{^}}atomic_and_i64_ret:
+; GCN: flat_atomic_and_x2 [[RET:v\[[0-9]+:[0-9]+\]]], v[{{[0-9]+:[0-9]+}}], v{{\[[0-9]+:[0-9]+\]}} glc{{$}}
+; GCN: flat_store_dwordx2 v{{\[[0-9]+:[0-9]+\]}}, [[RET]]
+define void @atomic_and_i64_ret(i64 addrspace(4)* %out, i64 addrspace(4)* %out2, i64 %in) {
+entry:
+  %tmp0 = atomicrmw volatile and i64 addrspace(4)* %out, i64 %in seq_cst
+  store i64 %tmp0, i64 addrspace(4)* %out2
+  ret void
+}
+
+; GCN-LABEL: {{^}}atomic_and_i64_addr64:
+; GCN: flat_atomic_and_x2 v[{{[0-9]+:[0-9]+}}], v{{\[[0-9]+:[0-9]+\]$}}
+define void @atomic_and_i64_addr64(i64 addrspace(4)* %out, i64 %in, i64 %index) {
+entry:
+  %ptr = getelementptr i64, i64 addrspace(4)* %out, i64 %index
+  %tmp0 = atomicrmw volatile and i64 addrspace(4)* %ptr, i64 %in seq_cst
+  ret void
+}
+
+; GCN-LABEL: {{^}}atomic_and_i64_ret_addr64:
+; GCN: flat_atomic_and_x2 [[RET:v\[[0-9]+:[0-9]+\]]], v[{{[0-9]+:[0-9]+}}], v{{\[[0-9]+:[0-9]+\]}} glc{{$}}
+; GCN: flat_store_dwordx2 v{{\[[0-9]+:[0-9]+\]}}, [[RET]]
+define void @atomic_and_i64_ret_addr64(i64 addrspace(4)* %out, i64 addrspace(4)* %out2, i64 %in, i64 %index) {
+entry:
+  %ptr = getelementptr i64, i64 addrspace(4)* %out, i64 %index
+  %tmp0 = atomicrmw volatile and i64 addrspace(4)* %ptr, i64 %in seq_cst
+  store i64 %tmp0, i64 addrspace(4)* %out2
+  ret void
+}
+
+; GCN-LABEL: {{^}}atomic_sub_i64_offset:
+; GCN: flat_atomic_sub_x2 v[{{[0-9]+:[0-9]+}}], v{{\[[0-9]+:[0-9]+\]$}}
+define void @atomic_sub_i64_offset(i64 addrspace(4)* %out, i64 %in) {
+entry:
+  %gep = getelementptr i64, i64 addrspace(4)* %out, i64 4
+  %tmp0 = atomicrmw volatile sub i64 addrspace(4)* %gep, i64 %in seq_cst
+  ret void
+}
+
+; GCN-LABEL: {{^}}atomic_sub_i64_ret_offset:
+; GCN: flat_atomic_sub_x2 [[RET:v\[[0-9]+:[0-9]+\]]], v[{{[0-9]+:[0-9]+}}], v{{\[[0-9]+:[0-9]+\]}} glc{{$}}
+; GCN: flat_store_dwordx2 v{{\[[0-9]+:[0-9]+\]}}, [[RET]]
+define void @atomic_sub_i64_ret_offset(i64 addrspace(4)* %out, i64 addrspace(4)* %out2, i64 %in) {
+entry:
+  %gep = getelementptr i64, i64 addrspace(4)* %out, i64 4
+  %tmp0 = atomicrmw volatile sub i64 addrspace(4)* %gep, i64 %in seq_cst
+  store i64 %tmp0, i64 addrspace(4)* %out2
+  ret void
+}
+
+; GCN-LABEL: {{^}}atomic_sub_i64_addr64_offset:
+; GCN: flat_atomic_sub_x2 v[{{[0-9]+:[0-9]+}}], v{{\[[0-9]+:[0-9]+\]$}}
+define void @atomic_sub_i64_addr64_offset(i64 addrspace(4)* %out, i64 %in, i64 %index) {
+entry:
+  %ptr = getelementptr i64, i64 addrspace(4)* %out, i64 %index
+  %gep = getelementptr i64, i64 addrspace(4)* %ptr, i64 4
+  %tmp0 = atomicrmw volatile sub i64 addrspace(4)* %gep, i64 %in seq_cst
+  ret void
+}
+
+; GCN-LABEL: {{^}}atomic_sub_i64_ret_addr64_offset:
+; GCN: flat_atomic_sub_x2 [[RET:v\[[0-9]+:[0-9]+\]]], v[{{[0-9]+:[0-9]+}}], v{{\[[0-9]+:[0-9]+\]}} glc{{$}}
+; GCN: flat_store_dwordx2 v{{\[[0-9]+:[0-9]+\]}}, [[RET]]
+define void @atomic_sub_i64_ret_addr64_offset(i64 addrspace(4)* %out, i64 addrspace(4)* %out2, i64 %in, i64 %index) {
+entry:
+  %ptr = getelementptr i64, i64 addrspace(4)* %out, i64 %index
+  %gep = getelementptr i64, i64 addrspace(4)* %ptr, i64 4
+  %tmp0 = atomicrmw volatile sub i64 addrspace(4)* %gep, i64 %in seq_cst
+  store i64 %tmp0, i64 addrspace(4)* %out2
+  ret void
+}
+
+; GCN-LABEL: {{^}}atomic_sub_i64:
+; GCN: flat_atomic_sub_x2 v[{{[0-9]+:[0-9]+}}], v{{\[[0-9]+:[0-9]+\]$}}
+define void @atomic_sub_i64(i64 addrspace(4)* %out, i64 %in) {
+entry:
+  %tmp0 = atomicrmw volatile sub i64 addrspace(4)* %out, i64 %in seq_cst
+  ret void
+}
+
+; GCN-LABEL: {{^}}atomic_sub_i64_ret:
+; GCN: flat_atomic_sub_x2 [[RET:v\[[0-9]+:[0-9]+\]]], v[{{[0-9]+:[0-9]+}}], v{{\[[0-9]+:[0-9]+\]}} glc{{$}}
+; GCN: flat_store_dwordx2 v{{\[[0-9]+:[0-9]+\]}}, [[RET]]
+define void @atomic_sub_i64_ret(i64 addrspace(4)* %out, i64 addrspace(4)* %out2, i64 %in) {
+entry:
+  %tmp0 = atomicrmw volatile sub i64 addrspace(4)* %out, i64 %in seq_cst
+  store i64 %tmp0, i64 addrspace(4)* %out2
+  ret void
+}
+
+; GCN-LABEL: {{^}}atomic_sub_i64_addr64:
+; GCN: flat_atomic_sub_x2 v[{{[0-9]+:[0-9]+}}], v{{\[[0-9]+:[0-9]+\]$}}
+define void @atomic_sub_i64_addr64(i64 addrspace(4)* %out, i64 %in, i64 %index) {
+entry:
+  %ptr = getelementptr i64, i64 addrspace(4)* %out, i64 %index
+  %tmp0 = atomicrmw volatile sub i64 addrspace(4)* %ptr, i64 %in seq_cst
+  ret void
+}
+
+; GCN-LABEL: {{^}}atomic_sub_i64_ret_addr64:
+; GCN: flat_atomic_sub_x2 [[RET:v\[[0-9]+:[0-9]+\]]], v[{{[0-9]+:[0-9]+}}], v{{\[[0-9]+:[0-9]+\]}} glc{{$}}
+; GCN: flat_store_dwordx2 v{{\[[0-9]+:[0-9]+\]}}, [[RET]]
+define void @atomic_sub_i64_ret_addr64(i64 addrspace(4)* %out, i64 addrspace(4)* %out2, i64 %in, i64 %index) {
+entry:
+  %ptr = getelementptr i64, i64 addrspace(4)* %out, i64 %index
+  %tmp0 = atomicrmw volatile sub i64 addrspace(4)* %ptr, i64 %in seq_cst
+  store i64 %tmp0, i64 addrspace(4)* %out2
+  ret void
+}
+
+; GCN-LABEL: {{^}}atomic_max_i64_offset:
+; GCN: flat_atomic_smax_x2 v[{{[0-9]+:[0-9]+}}], v{{\[[0-9]+:[0-9]+\]$}}
+define void @atomic_max_i64_offset(i64 addrspace(4)* %out, i64 %in) {
+entry:
+  %gep = getelementptr i64, i64 addrspace(4)* %out, i64 4
+  %tmp0 = atomicrmw volatile max i64 addrspace(4)* %gep, i64 %in seq_cst
+  ret void
+}
+
+; GCN-LABEL: {{^}}atomic_max_i64_ret_offset:
+; GCN: flat_atomic_smax_x2 [[RET:v\[[0-9]+:[0-9]+\]]], v[{{[0-9]+:[0-9]+}}], v{{\[[0-9]+:[0-9]+\]}} glc{{$}}
+; GCN: flat_store_dwordx2 v{{\[[0-9]+:[0-9]+\]}}, [[RET]]
+define void @atomic_max_i64_ret_offset(i64 addrspace(4)* %out, i64 addrspace(4)* %out2, i64 %in) {
+entry:
+  %gep = getelementptr i64, i64 addrspace(4)* %out, i64 4
+  %tmp0 = atomicrmw volatile max i64 addrspace(4)* %gep, i64 %in seq_cst
+  store i64 %tmp0, i64 addrspace(4)* %out2
+  ret void
+}
+
+; GCN-LABEL: {{^}}atomic_max_i64_addr64_offset:
+; GCN: flat_atomic_smax_x2 v[{{[0-9]+:[0-9]+}}], v{{\[[0-9]+:[0-9]+\]$}}
+define void @atomic_max_i64_addr64_offset(i64 addrspace(4)* %out, i64 %in, i64 %index) {
+entry:
+  %ptr = getelementptr i64, i64 addrspace(4)* %out, i64 %index
+  %gep = getelementptr i64, i64 addrspace(4)* %ptr, i64 4
+  %tmp0 = atomicrmw volatile max i64 addrspace(4)* %gep, i64 %in seq_cst
+  ret void
+}
+
+; GCN-LABEL: {{^}}atomic_max_i64_ret_addr64_offset:
+; GCN: flat_atomic_smax_x2 [[RET:v\[[0-9]+:[0-9]+\]]], v[{{[0-9]+:[0-9]+}}], v{{\[[0-9]+:[0-9]+\]}} glc{{$}}
+; GCN: flat_store_dwordx2 v{{\[[0-9]+:[0-9]+\]}}, [[RET]]
+define void @atomic_max_i64_ret_addr64_offset(i64 addrspace(4)* %out, i64 addrspace(4)* %out2, i64 %in, i64 %index) {
+entry:
+  %ptr = getelementptr i64, i64 addrspace(4)* %out, i64 %index
+  %gep = getelementptr i64, i64 addrspace(4)* %ptr, i64 4
+  %tmp0 = atomicrmw volatile max i64 addrspace(4)* %gep, i64 %in seq_cst
+  store i64 %tmp0, i64 addrspace(4)* %out2
+  ret void
+}
+
+; GCN-LABEL: {{^}}atomic_max_i64:
+; GCN: flat_atomic_smax_x2 v[{{[0-9]+:[0-9]+}}], v{{\[[0-9]+:[0-9]+\]$}}
+define void @atomic_max_i64(i64 addrspace(4)* %out, i64 %in) {
+entry:
+  %tmp0 = atomicrmw volatile max i64 addrspace(4)* %out, i64 %in seq_cst
+  ret void
+}
+
+; GCN-LABEL: {{^}}atomic_max_i64_ret:
+; GCN: flat_atomic_smax_x2 [[RET:v\[[0-9]+:[0-9]+\]]], v[{{[0-9]+:[0-9]+}}], v{{\[[0-9]+:[0-9]+\]}} glc{{$}}
+; GCN: flat_store_dwordx2 v{{\[[0-9]+:[0-9]+\]}}, [[RET]]
+define void @atomic_max_i64_ret(i64 addrspace(4)* %out, i64 addrspace(4)* %out2, i64 %in) {
+entry:
+  %tmp0 = atomicrmw volatile max i64 addrspace(4)* %out, i64 %in seq_cst
+  store i64 %tmp0, i64 addrspace(4)* %out2
+  ret void
+}
+
+; GCN-LABEL: {{^}}atomic_max_i64_addr64:
+; GCN: flat_atomic_smax_x2 v[{{[0-9]+:[0-9]+}}], v{{\[[0-9]+:[0-9]+\]$}}
+define void @atomic_max_i64_addr64(i64 addrspace(4)* %out, i64 %in, i64 %index) {
+entry:
+  %ptr = getelementptr i64, i64 addrspace(4)* %out, i64 %index
+  %tmp0 = atomicrmw volatile max i64 addrspace(4)* %ptr, i64 %in seq_cst
+  ret void
+}
+
+; GCN-LABEL: {{^}}atomic_max_i64_ret_addr64:
+; GCN: flat_atomic_smax_x2 [[RET:v\[[0-9]+:[0-9]+\]]], v[{{[0-9]+:[0-9]+}}], v{{\[[0-9]+:[0-9]+\]}} glc{{$}}
+; GCN: flat_store_dwordx2 v{{\[[0-9]+:[0-9]+\]}}, [[RET]]
+define void @atomic_max_i64_ret_addr64(i64 addrspace(4)* %out, i64 addrspace(4)* %out2, i64 %in, i64 %index) {
+entry:
+  %ptr = getelementptr i64, i64 addrspace(4)* %out, i64 %index
+  %tmp0 = atomicrmw volatile max i64 addrspace(4)* %ptr, i64 %in seq_cst
+  store i64 %tmp0, i64 addrspace(4)* %out2
+  ret void
+}
+
+; GCN-LABEL: {{^}}atomic_umax_i64_offset:
+; GCN: flat_atomic_umax_x2 v[{{[0-9]+:[0-9]+}}], v{{\[[0-9]+:[0-9]+\]$}}
+define void @atomic_umax_i64_offset(i64 addrspace(4)* %out, i64 %in) {
+entry:
+  %gep = getelementptr i64, i64 addrspace(4)* %out, i64 4
+  %tmp0 = atomicrmw volatile umax i64 addrspace(4)* %gep, i64 %in seq_cst
+  ret void
+}
+
+; GCN-LABEL: {{^}}atomic_umax_i64_ret_offset:
+; GCN: flat_atomic_umax_x2 [[RET:v\[[0-9]+:[0-9]+\]]], v[{{[0-9]+:[0-9]+}}], v{{\[[0-9]+:[0-9]+\]}} glc{{$}}
+; GCN: flat_store_dwordx2 v{{\[[0-9]+:[0-9]+\]}}, [[RET]]
+define void @atomic_umax_i64_ret_offset(i64 addrspace(4)* %out, i64 addrspace(4)* %out2, i64 %in) {
+entry:
+  %gep = getelementptr i64, i64 addrspace(4)* %out, i64 4
+  %tmp0 = atomicrmw volatile umax i64 addrspace(4)* %gep, i64 %in seq_cst
+  store i64 %tmp0, i64 addrspace(4)* %out2
+  ret void
+}
+
+; GCN-LABEL: {{^}}atomic_umax_i64_addr64_offset:
+; GCN: flat_atomic_umax_x2 v[{{[0-9]+:[0-9]+}}], v{{\[[0-9]+:[0-9]+\]$}}
+define void @atomic_umax_i64_addr64_offset(i64 addrspace(4)* %out, i64 %in, i64 %index) {
+entry:
+  %ptr = getelementptr i64, i64 addrspace(4)* %out, i64 %index
+  %gep = getelementptr i64, i64 addrspace(4)* %ptr, i64 4
+  %tmp0 = atomicrmw volatile umax i64 addrspace(4)* %gep, i64 %in seq_cst
+  ret void
+}
+
+; GCN-LABEL: {{^}}atomic_umax_i64_ret_addr64_offset:
+; GCN: flat_atomic_umax_x2 [[RET:v\[[0-9]+:[0-9]+\]]], v[{{[0-9]+:[0-9]+}}], v{{\[[0-9]+:[0-9]+\]}} glc{{$}}
+; GCN: flat_store_dwordx2 v{{\[[0-9]+:[0-9]+\]}}, [[RET]]
+define void @atomic_umax_i64_ret_addr64_offset(i64 addrspace(4)* %out, i64 addrspace(4)* %out2, i64 %in, i64 %index) {
+entry:
+  %ptr = getelementptr i64, i64 addrspace(4)* %out, i64 %index
+  %gep = getelementptr i64, i64 addrspace(4)* %ptr, i64 4
+  %tmp0 = atomicrmw volatile umax i64 addrspace(4)* %gep, i64 %in seq_cst
+  store i64 %tmp0, i64 addrspace(4)* %out2
+  ret void
+}
+
+; GCN-LABEL: {{^}}atomic_umax_i64:
+; GCN: flat_atomic_umax_x2 v[{{[0-9]+:[0-9]+}}], v{{\[[0-9]+:[0-9]+\]$}}
+define void @atomic_umax_i64(i64 addrspace(4)* %out, i64 %in) {
+entry:
+  %tmp0 = atomicrmw volatile umax i64 addrspace(4)* %out, i64 %in seq_cst
+  ret void
+}
+
+; GCN-LABEL: {{^}}atomic_umax_i64_ret:
+; GCN: flat_atomic_umax_x2 [[RET:v\[[0-9]+:[0-9]+\]]], v[{{[0-9]+:[0-9]+}}], v{{\[[0-9]+:[0-9]+\]}} glc{{$}}
+; GCN: flat_store_dwordx2 v{{\[[0-9]+:[0-9]+\]}}, [[RET]]
+define void @atomic_umax_i64_ret(i64 addrspace(4)* %out, i64 addrspace(4)* %out2, i64 %in) {
+entry:
+  %tmp0 = atomicrmw volatile umax i64 addrspace(4)* %out, i64 %in seq_cst
+  store i64 %tmp0, i64 addrspace(4)* %out2
+  ret void
+}
+
+; GCN-LABEL: {{^}}atomic_umax_i64_addr64:
+; GCN: flat_atomic_umax_x2 v[{{[0-9]+:[0-9]+}}], v{{\[[0-9]+:[0-9]+\]$}}
+define void @atomic_umax_i64_addr64(i64 addrspace(4)* %out, i64 %in, i64 %index) {
+entry:
+  %ptr = getelementptr i64, i64 addrspace(4)* %out, i64 %index
+  %tmp0 = atomicrmw volatile umax i64 addrspace(4)* %ptr, i64 %in seq_cst
+  ret void
+}
+
+; GCN-LABEL: {{^}}atomic_umax_i64_ret_addr64:
+; GCN: flat_atomic_umax_x2 [[RET:v\[[0-9]+:[0-9]+\]]], v[{{[0-9]+:[0-9]+}}], v{{\[[0-9]+:[0-9]+\]}} glc{{$}}
+; GCN: flat_store_dwordx2 v{{\[[0-9]+:[0-9]+\]}}, [[RET]]
+define void @atomic_umax_i64_ret_addr64(i64 addrspace(4)* %out, i64 addrspace(4)* %out2, i64 %in, i64 %index) {
+entry:
+  %ptr = getelementptr i64, i64 addrspace(4)* %out, i64 %index
+  %tmp0 = atomicrmw volatile umax i64 addrspace(4)* %ptr, i64 %in seq_cst
+  store i64 %tmp0, i64 addrspace(4)* %out2
+  ret void
+}
+
+; GCN-LABEL: {{^}}atomic_min_i64_offset:
+; GCN: flat_atomic_smin_x2 v[{{[0-9]+:[0-9]+}}], v{{\[[0-9]+:[0-9]+\]$}}
+define void @atomic_min_i64_offset(i64 addrspace(4)* %out, i64 %in) {
+entry:
+  %gep = getelementptr i64, i64 addrspace(4)* %out, i64 4
+  %tmp0 = atomicrmw volatile min i64 addrspace(4)* %gep, i64 %in seq_cst
+  ret void
+}
+
+; GCN-LABEL: {{^}}atomic_min_i64_ret_offset:
+; GCN: flat_atomic_smin_x2 [[RET:v\[[0-9]+:[0-9]+\]]], v[{{[0-9]+:[0-9]+}}], v{{\[[0-9]+:[0-9]+\]}} glc{{$}}
+; GCN: flat_store_dwordx2 v{{\[[0-9]+:[0-9]+\]}}, [[RET]]
+define void @atomic_min_i64_ret_offset(i64 addrspace(4)* %out, i64 addrspace(4)* %out2, i64 %in) {
+entry:
+  %gep = getelementptr i64, i64 addrspace(4)* %out, i64 4
+  %tmp0 = atomicrmw volatile min i64 addrspace(4)* %gep, i64 %in seq_cst
+  store i64 %tmp0, i64 addrspace(4)* %out2
+  ret void
+}
+
+; GCN-LABEL: {{^}}atomic_min_i64_addr64_offset:
+; GCN: flat_atomic_smin_x2 v[{{[0-9]+:[0-9]+}}], v{{\[[0-9]+:[0-9]+\]$}}
+define void @atomic_min_i64_addr64_offset(i64 addrspace(4)* %out, i64 %in, i64 %index) {
+entry:
+  %ptr = getelementptr i64, i64 addrspace(4)* %out, i64 %index
+  %gep = getelementptr i64, i64 addrspace(4)* %ptr, i64 4
+  %tmp0 = atomicrmw volatile min i64 addrspace(4)* %gep, i64 %in seq_cst
+  ret void
+}
+
+; GCN-LABEL: {{^}}atomic_min_i64_ret_addr64_offset:
+; GCN: flat_atomic_smin_x2 [[RET:v\[[0-9]+:[0-9]+\]]], v[{{[0-9]+:[0-9]+}}], v{{\[[0-9]+:[0-9]+\]}} glc{{$}}
+; GCN: flat_store_dwordx2 v{{\[[0-9]+:[0-9]+\]}}, [[RET]]
+define void @atomic_min_i64_ret_addr64_offset(i64 addrspace(4)* %out, i64 addrspace(4)* %out2, i64 %in, i64 %index) {
+entry:
+  %ptr = getelementptr i64, i64 addrspace(4)* %out, i64 %index
+  %gep = getelementptr i64, i64 addrspace(4)* %ptr, i64 4
+  %tmp0 = atomicrmw volatile min i64 addrspace(4)* %gep, i64 %in seq_cst
+  store i64 %tmp0, i64 addrspace(4)* %out2
+  ret void
+}
+
+; GCN-LABEL: {{^}}atomic_min_i64:
+; GCN: flat_atomic_smin_x2 v[{{[0-9]+:[0-9]+}}], v{{\[[0-9]+:[0-9]+\]$}}
+define void @atomic_min_i64(i64 addrspace(4)* %out, i64 %in) {
+entry:
+  %tmp0 = atomicrmw volatile min i64 addrspace(4)* %out, i64 %in seq_cst
+  ret void
+}
+
+; GCN-LABEL: {{^}}atomic_min_i64_ret:
+; GCN: flat_atomic_smin_x2 [[RET:v\[[0-9]+:[0-9]+\]]], v[{{[0-9]+:[0-9]+}}], v{{\[[0-9]+:[0-9]+\]}} glc{{$}}
+; GCN: flat_store_dwordx2 v{{\[[0-9]+:[0-9]+\]}}, [[RET]]
+define void @atomic_min_i64_ret(i64 addrspace(4)* %out, i64 addrspace(4)* %out2, i64 %in) {
+entry:
+  %tmp0 = atomicrmw volatile min i64 addrspace(4)* %out, i64 %in seq_cst
+  store i64 %tmp0, i64 addrspace(4)* %out2
+  ret void
+}
+
+; GCN-LABEL: {{^}}atomic_min_i64_addr64:
+; GCN: flat_atomic_smin_x2 v[{{[0-9]+:[0-9]+}}], v{{\[[0-9]+:[0-9]+\]$}}
+define void @atomic_min_i64_addr64(i64 addrspace(4)* %out, i64 %in, i64 %index) {
+entry:
+  %ptr = getelementptr i64, i64 addrspace(4)* %out, i64 %index
+  %tmp0 = atomicrmw volatile min i64 addrspace(4)* %ptr, i64 %in seq_cst
+  ret void
+}
+
+; GCN-LABEL: {{^}}atomic_min_i64_ret_addr64:
+; GCN: flat_atomic_smin_x2 [[RET:v\[[0-9]+:[0-9]+\]]], v[{{[0-9]+:[0-9]+}}], v{{\[[0-9]+:[0-9]+\]}} glc{{$}}
+; GCN: flat_store_dwordx2 v{{\[[0-9]+:[0-9]+\]}}, [[RET]]
+define void @atomic_min_i64_ret_addr64(i64 addrspace(4)* %out, i64 addrspace(4)* %out2, i64 %in, i64 %index) {
+entry:
+  %ptr = getelementptr i64, i64 addrspace(4)* %out, i64 %index
+  %tmp0 = atomicrmw volatile min i64 addrspace(4)* %ptr, i64 %in seq_cst
+  store i64 %tmp0, i64 addrspace(4)* %out2
+  ret void
+}
+
+; GCN-LABEL: {{^}}atomic_umin_i64_offset:
+; GCN: flat_atomic_umin_x2 v[{{[0-9]+:[0-9]+}}], v{{\[[0-9]+:[0-9]+\]$}}
+define void @atomic_umin_i64_offset(i64 addrspace(4)* %out, i64 %in) {
+entry:
+  %gep = getelementptr i64, i64 addrspace(4)* %out, i64 4
+  %tmp0 = atomicrmw volatile umin i64 addrspace(4)* %gep, i64 %in seq_cst
+  ret void
+}
+
+; GCN-LABEL: {{^}}atomic_umin_i64_ret_offset:
+; GCN: flat_atomic_umin_x2 [[RET:v\[[0-9]+:[0-9]+\]]], v[{{[0-9]+:[0-9]+}}], v{{\[[0-9]+:[0-9]+\]}} glc{{$}}
+; GCN: flat_store_dwordx2 v{{\[[0-9]+:[0-9]+\]}}, [[RET]]
+define void @atomic_umin_i64_ret_offset(i64 addrspace(4)* %out, i64 addrspace(4)* %out2, i64 %in) {
+entry:
+  %gep = getelementptr i64, i64 addrspace(4)* %out, i64 4
+  %tmp0 = atomicrmw volatile umin i64 addrspace(4)* %gep, i64 %in seq_cst
+  store i64 %tmp0, i64 addrspace(4)* %out2
+  ret void
+}
+
+; GCN-LABEL: {{^}}atomic_umin_i64_addr64_offset:
+; GCN: flat_atomic_umin_x2 v[{{[0-9]+:[0-9]+}}], v{{\[[0-9]+:[0-9]+\]$}}
+define void @atomic_umin_i64_addr64_offset(i64 addrspace(4)* %out, i64 %in, i64 %index) {
+entry:
+  %ptr = getelementptr i64, i64 addrspace(4)* %out, i64 %index
+  %gep = getelementptr i64, i64 addrspace(4)* %ptr, i64 4
+  %tmp0 = atomicrmw volatile umin i64 addrspace(4)* %gep, i64 %in seq_cst
+  ret void
+}
+
+; GCN-LABEL: {{^}}atomic_umin_i64_ret_addr64_offset:
+; GCN: flat_atomic_umin_x2 [[RET:v\[[0-9]+:[0-9]+\]]], v[{{[0-9]+:[0-9]+}}], v{{\[[0-9]+:[0-9]+\]}} glc{{$}}
+; GCN: flat_store_dwordx2 v{{\[[0-9]+:[0-9]+\]}}, [[RET]]
+define void @atomic_umin_i64_ret_addr64_offset(i64 addrspace(4)* %out, i64 addrspace(4)* %out2, i64 %in, i64 %index) {
+entry:
+  %ptr = getelementptr i64, i64 addrspace(4)* %out, i64 %index
+  %gep = getelementptr i64, i64 addrspace(4)* %ptr, i64 4
+  %tmp0 = atomicrmw volatile umin i64 addrspace(4)* %gep, i64 %in seq_cst
+  store i64 %tmp0, i64 addrspace(4)* %out2
+  ret void
+}
+
+; GCN-LABEL: {{^}}atomic_umin_i64:
+; GCN: flat_atomic_umin_x2 v[{{[0-9]+:[0-9]+}}], v{{\[[0-9]+:[0-9]+\]$}}
+define void @atomic_umin_i64(i64 addrspace(4)* %out, i64 %in) {
+entry:
+  %tmp0 = atomicrmw volatile umin i64 addrspace(4)* %out, i64 %in seq_cst
+  ret void
+}
+
+; GCN-LABEL: {{^}}atomic_umin_i64_ret:
+; GCN: flat_atomic_umin_x2 [[RET:v\[[0-9]+:[0-9]+\]]], v[{{[0-9]+:[0-9]+}}], v{{\[[0-9]+:[0-9]+\]}} glc{{$}}
+; GCN: flat_store_dwordx2 v{{\[[0-9]+:[0-9]+\]}}, [[RET]]
+define void @atomic_umin_i64_ret(i64 addrspace(4)* %out, i64 addrspace(4)* %out2, i64 %in) {
+entry:
+  %tmp0 = atomicrmw volatile umin i64 addrspace(4)* %out, i64 %in seq_cst
+  store i64 %tmp0, i64 addrspace(4)* %out2
+  ret void
+}
+
+; GCN-LABEL: {{^}}atomic_umin_i64_addr64:
+; GCN: flat_atomic_umin_x2 v[{{[0-9]+:[0-9]+}}], v{{\[[0-9]+:[0-9]+\]$}}
+define void @atomic_umin_i64_addr64(i64 addrspace(4)* %out, i64 %in, i64 %index) {
+entry:
+  %ptr = getelementptr i64, i64 addrspace(4)* %out, i64 %index
+  %tmp0 = atomicrmw volatile umin i64 addrspace(4)* %ptr, i64 %in seq_cst
+  ret void
+}
+
+; GCN-LABEL: {{^}}atomic_umin_i64_ret_addr64:
+; GCN: flat_atomic_umin_x2 [[RET:v\[[0-9]+:[0-9]+\]]], v[{{[0-9]+:[0-9]+}}], v{{\[[0-9]+:[0-9]+\]}} glc{{$}}
+; GCN: flat_store_dwordx2 v{{\[[0-9]+:[0-9]+\]}}, [[RET]]
+define void @atomic_umin_i64_ret_addr64(i64 addrspace(4)* %out, i64 addrspace(4)* %out2, i64 %in, i64 %index) {
+entry:
+  %ptr = getelementptr i64, i64 addrspace(4)* %out, i64 %index
+  %tmp0 = atomicrmw volatile umin i64 addrspace(4)* %ptr, i64 %in seq_cst
+  store i64 %tmp0, i64 addrspace(4)* %out2
+  ret void
+}
+
+; GCN-LABEL: {{^}}atomic_or_i64_offset:
+; GCN: flat_atomic_or_x2 v[{{[0-9]+:[0-9]+}}], v{{\[[0-9]+:[0-9]+\]$}}
+define void @atomic_or_i64_offset(i64 addrspace(4)* %out, i64 %in) {
+entry:
+  %gep = getelementptr i64, i64 addrspace(4)* %out, i64 4
+  %tmp0 = atomicrmw volatile or i64 addrspace(4)* %gep, i64 %in seq_cst
+  ret void
+}
+
+; GCN-LABEL: {{^}}atomic_or_i64_ret_offset:
+; GCN: flat_atomic_or_x2 [[RET:v\[[0-9]+:[0-9]+\]]], v[{{[0-9]+:[0-9]+}}], v{{\[[0-9]+:[0-9]+\]}} glc{{$}}
+; GCN: flat_store_dwordx2 v{{\[[0-9]+:[0-9]+\]}}, [[RET]]
+define void @atomic_or_i64_ret_offset(i64 addrspace(4)* %out, i64 addrspace(4)* %out2, i64 %in) {
+entry:
+  %gep = getelementptr i64, i64 addrspace(4)* %out, i64 4
+  %tmp0 = atomicrmw volatile or i64 addrspace(4)* %gep, i64 %in seq_cst
+  store i64 %tmp0, i64 addrspace(4)* %out2
+  ret void
+}
+
+; GCN-LABEL: {{^}}atomic_or_i64_addr64_offset:
+; GCN: flat_atomic_or_x2 v[{{[0-9]+:[0-9]+}}], v{{\[[0-9]+:[0-9]+\]$}}
+define void @atomic_or_i64_addr64_offset(i64 addrspace(4)* %out, i64 %in, i64 %index) {
+entry:
+  %ptr = getelementptr i64, i64 addrspace(4)* %out, i64 %index
+  %gep = getelementptr i64, i64 addrspace(4)* %ptr, i64 4
+  %tmp0 = atomicrmw volatile or i64 addrspace(4)* %gep, i64 %in seq_cst
+  ret void
+}
+
+; GCN-LABEL: {{^}}atomic_or_i64_ret_addr64_offset:
+; GCN: flat_atomic_or_x2 [[RET:v\[[0-9]+:[0-9]+\]]], v[{{[0-9]+:[0-9]+}}], v{{\[[0-9]+:[0-9]+\]}} glc{{$}}
+; GCN: flat_store_dwordx2 v{{\[[0-9]+:[0-9]+\]}}, [[RET]]
+define void @atomic_or_i64_ret_addr64_offset(i64 addrspace(4)* %out, i64 addrspace(4)* %out2, i64 %in, i64 %index) {
+entry:
+  %ptr = getelementptr i64, i64 addrspace(4)* %out, i64 %index
+  %gep = getelementptr i64, i64 addrspace(4)* %ptr, i64 4
+  %tmp0 = atomicrmw volatile or i64 addrspace(4)* %gep, i64 %in seq_cst
+  store i64 %tmp0, i64 addrspace(4)* %out2
+  ret void
+}
+
+; GCN-LABEL: {{^}}atomic_or_i64:
+; GCN: flat_atomic_or_x2 v[{{[0-9]+:[0-9]+}}], v{{\[[0-9]+:[0-9]+\]$}}
+define void @atomic_or_i64(i64 addrspace(4)* %out, i64 %in) {
+entry:
+  %tmp0 = atomicrmw volatile or i64 addrspace(4)* %out, i64 %in seq_cst
+  ret void
+}
+
+; GCN-LABEL: {{^}}atomic_or_i64_ret:
+; GCN: flat_atomic_or_x2 [[RET:v\[[0-9]+:[0-9]+\]]], v[{{[0-9]+:[0-9]+}}], v{{\[[0-9]+:[0-9]+\]}} glc{{$}}
+; GCN: flat_store_dwordx2 v{{\[[0-9]+:[0-9]+\]}}, [[RET]]
+define void @atomic_or_i64_ret(i64 addrspace(4)* %out, i64 addrspace(4)* %out2, i64 %in) {
+entry:
+  %tmp0 = atomicrmw volatile or i64 addrspace(4)* %out, i64 %in seq_cst
+  store i64 %tmp0, i64 addrspace(4)* %out2
+  ret void
+}
+
+; GCN-LABEL: {{^}}atomic_or_i64_addr64:
+; GCN: flat_atomic_or_x2 v[{{[0-9]+:[0-9]+}}], v{{\[[0-9]+:[0-9]+\]$}}
+define void @atomic_or_i64_addr64(i64 addrspace(4)* %out, i64 %in, i64 %index) {
+entry:
+  %ptr = getelementptr i64, i64 addrspace(4)* %out, i64 %index
+  %tmp0 = atomicrmw volatile or i64 addrspace(4)* %ptr, i64 %in seq_cst
+  ret void
+}
+
+; GCN-LABEL: {{^}}atomic_or_i64_ret_addr64:
+; GCN: flat_atomic_or_x2 [[RET:v\[[0-9]+:[0-9]+\]]], v[{{[0-9]+:[0-9]+}}], v{{\[[0-9]+:[0-9]+\]}} glc{{$}}
+; GCN: flat_store_dwordx2 v{{\[[0-9]+:[0-9]+\]}}, [[RET]]
+define void @atomic_or_i64_ret_addr64(i64 addrspace(4)* %out, i64 addrspace(4)* %out2, i64 %in, i64 %index) {
+entry:
+  %ptr = getelementptr i64, i64 addrspace(4)* %out, i64 %index
+  %tmp0 = atomicrmw volatile or i64 addrspace(4)* %ptr, i64 %in seq_cst
+  store i64 %tmp0, i64 addrspace(4)* %out2
+  ret void
+}
+
+; GCN-LABEL: {{^}}atomic_xchg_i64_offset:
+; GCN: flat_atomic_swap_x2 v[{{[0-9]+:[0-9]+}}], v{{\[[0-9]+:[0-9]+\]$}}
+define void @atomic_xchg_i64_offset(i64 addrspace(4)* %out, i64 %in) {
+entry:
+  %gep = getelementptr i64, i64 addrspace(4)* %out, i64 4
+  %tmp0 = atomicrmw volatile xchg i64 addrspace(4)* %gep, i64 %in seq_cst
+  ret void
+}
+
+; GCN-LABEL: {{^}}atomic_xchg_i64_ret_offset:
+; GCN: flat_atomic_swap_x2 [[RET:v\[[0-9]+:[0-9]+\]]], v[{{[0-9]+:[0-9]+}}], v{{\[[0-9]+:[0-9]+\]}} glc{{$}}
+; GCN: flat_store_dwordx2 v{{\[[0-9]+:[0-9]+\]}}, [[RET]]
+define void @atomic_xchg_i64_ret_offset(i64 addrspace(4)* %out, i64 addrspace(4)* %out2, i64 %in) {
+entry:
+  %gep = getelementptr i64, i64 addrspace(4)* %out, i64 4
+  %tmp0 = atomicrmw volatile xchg i64 addrspace(4)* %gep, i64 %in seq_cst
+  store i64 %tmp0, i64 addrspace(4)* %out2
+  ret void
+}
+
+; GCN-LABEL: {{^}}atomic_xchg_i64_addr64_offset:
+; GCN: flat_atomic_swap_x2 v[{{[0-9]+:[0-9]+}}], v{{\[[0-9]+:[0-9]+\]$}}
+define void @atomic_xchg_i64_addr64_offset(i64 addrspace(4)* %out, i64 %in, i64 %index) {
+entry:
+  %ptr = getelementptr i64, i64 addrspace(4)* %out, i64 %index
+  %gep = getelementptr i64, i64 addrspace(4)* %ptr, i64 4
+  %tmp0 = atomicrmw volatile xchg i64 addrspace(4)* %gep, i64 %in seq_cst
+  ret void
+}
+
+; GCN-LABEL: {{^}}atomic_xchg_i64_ret_addr64_offset:
+; GCN: flat_atomic_swap_x2 [[RET:v\[[0-9]+:[0-9]+\]]], v[{{[0-9]+:[0-9]+}}], v{{\[[0-9]+:[0-9]+\]}} glc{{$}}
+; GCN: flat_store_dwordx2 v{{\[[0-9]+:[0-9]+\]}}, [[RET]]
+define void @atomic_xchg_i64_ret_addr64_offset(i64 addrspace(4)* %out, i64 addrspace(4)* %out2, i64 %in, i64 %index) {
+entry:
+  %ptr = getelementptr i64, i64 addrspace(4)* %out, i64 %index
+  %gep = getelementptr i64, i64 addrspace(4)* %ptr, i64 4
+  %tmp0 = atomicrmw volatile xchg i64 addrspace(4)* %gep, i64 %in seq_cst
+  store i64 %tmp0, i64 addrspace(4)* %out2
+  ret void
+}
+
+; GCN-LABEL: {{^}}atomic_xchg_i64:
+; GCN: flat_atomic_swap_x2 v[{{[0-9]+:[0-9]+}}], v{{\[[0-9]+:[0-9]+\]$}}
+define void @atomic_xchg_i64(i64 addrspace(4)* %out, i64 %in) {
+entry:
+  %tmp0 = atomicrmw volatile xchg i64 addrspace(4)* %out, i64 %in seq_cst
+  ret void
+}
+
+; GCN-LABEL: {{^}}atomic_xchg_i64_ret:
+; GCN: flat_atomic_swap_x2 [[RET:v\[[0-9]+:[0-9]+\]]], v[{{[0-9]+:[0-9]+}}], v{{\[[0-9]+:[0-9]+\]}} glc{{$}}
+; GCN: flat_store_dwordx2 v{{\[[0-9]+:[0-9]+\]}}, [[RET]]
+define void @atomic_xchg_i64_ret(i64 addrspace(4)* %out, i64 addrspace(4)* %out2, i64 %in) {
+entry:
+  %tmp0 = atomicrmw volatile xchg i64 addrspace(4)* %out, i64 %in seq_cst
+  store i64 %tmp0, i64 addrspace(4)* %out2
+  ret void
+}
+
+; GCN-LABEL: {{^}}atomic_xchg_i64_addr64:
+; GCN: flat_atomic_swap_x2 v[{{[0-9]+:[0-9]+}}], v{{\[[0-9]+:[0-9]+\]$}}
+define void @atomic_xchg_i64_addr64(i64 addrspace(4)* %out, i64 %in, i64 %index) {
+entry:
+  %ptr = getelementptr i64, i64 addrspace(4)* %out, i64 %index
+  %tmp0 = atomicrmw volatile xchg i64 addrspace(4)* %ptr, i64 %in seq_cst
+  ret void
+}
+
+; GCN-LABEL: {{^}}atomic_xchg_i64_ret_addr64:
+; GCN: flat_atomic_swap_x2 [[RET:v\[[0-9]+:[0-9]+\]]],  v[{{[0-9]+:[0-9]+}}], v{{\[[0-9]+:[0-9]+\]}} glc{{$}}
+; GCN: flat_store_dwordx2 v{{\[[0-9]+:[0-9]+\]}}, [[RET]]
+define void @atomic_xchg_i64_ret_addr64(i64 addrspace(4)* %out, i64 addrspace(4)* %out2, i64 %in, i64 %index) {
+entry:
+  %ptr = getelementptr i64, i64 addrspace(4)* %out, i64 %index
+  %tmp0 = atomicrmw volatile xchg i64 addrspace(4)* %ptr, i64 %in seq_cst
+  store i64 %tmp0, i64 addrspace(4)* %out2
+  ret void
+}
+
+; GCN-LABEL: {{^}}atomic_xor_i64_offset:
+; GCN: flat_atomic_xor_x2 v[{{[0-9]+:[0-9]+}}], v{{\[[0-9]+:[0-9]+\]$}}
+define void @atomic_xor_i64_offset(i64 addrspace(4)* %out, i64 %in) {
+entry:
+  %gep = getelementptr i64, i64 addrspace(4)* %out, i64 4
+  %tmp0 = atomicrmw volatile xor i64 addrspace(4)* %gep, i64 %in seq_cst
+  ret void
+}
+
+; GCN-LABEL: {{^}}atomic_xor_i64_ret_offset:
+; GCN: flat_atomic_xor_x2 [[RET:v\[[0-9]+:[0-9]+\]]], v[{{[0-9]+:[0-9]+}}], v{{\[[0-9]+:[0-9]+\]}} glc{{$}}
+; GCN: flat_store_dwordx2 v{{\[[0-9]+:[0-9]+\]}}, [[RET]]
+define void @atomic_xor_i64_ret_offset(i64 addrspace(4)* %out, i64 addrspace(4)* %out2, i64 %in) {
+entry:
+  %gep = getelementptr i64, i64 addrspace(4)* %out, i64 4
+  %tmp0 = atomicrmw volatile xor i64 addrspace(4)* %gep, i64 %in seq_cst
+  store i64 %tmp0, i64 addrspace(4)* %out2
+  ret void
+}
+
+; GCN-LABEL: {{^}}atomic_xor_i64_addr64_offset:
+; GCN: flat_atomic_xor_x2 v[{{[0-9]+:[0-9]+}}], v{{\[[0-9]+:[0-9]+\]$}}
+define void @atomic_xor_i64_addr64_offset(i64 addrspace(4)* %out, i64 %in, i64 %index) {
+entry:
+  %ptr = getelementptr i64, i64 addrspace(4)* %out, i64 %index
+  %gep = getelementptr i64, i64 addrspace(4)* %ptr, i64 4
+  %tmp0 = atomicrmw volatile xor i64 addrspace(4)* %gep, i64 %in seq_cst
+  ret void
+}
+
+; GCN-LABEL: {{^}}atomic_xor_i64_ret_addr64_offset:
+; GCN: flat_atomic_xor_x2 [[RET:v\[[0-9]+:[0-9]+\]]], v[{{[0-9]+:[0-9]+}}], v{{\[[0-9]+:[0-9]+\]}} glc{{$}}
+; GCN: flat_store_dwordx2 v{{\[[0-9]+:[0-9]+\]}}, [[RET]]
+define void @atomic_xor_i64_ret_addr64_offset(i64 addrspace(4)* %out, i64 addrspace(4)* %out2, i64 %in, i64 %index) {
+entry:
+  %ptr = getelementptr i64, i64 addrspace(4)* %out, i64 %index
+  %gep = getelementptr i64, i64 addrspace(4)* %ptr, i64 4
+  %tmp0 = atomicrmw volatile xor i64 addrspace(4)* %gep, i64 %in seq_cst
+  store i64 %tmp0, i64 addrspace(4)* %out2
+  ret void
+}
+
+; GCN-LABEL: {{^}}atomic_xor_i64:
+; GCN: flat_atomic_xor_x2 v[{{[0-9]+:[0-9]+}}], v{{\[[0-9]+:[0-9]+\]$}}
+define void @atomic_xor_i64(i64 addrspace(4)* %out, i64 %in) {
+entry:
+  %tmp0 = atomicrmw volatile xor i64 addrspace(4)* %out, i64 %in seq_cst
+  ret void
+}
+
+; GCN-LABEL: {{^}}atomic_xor_i64_ret:
+; GCN: flat_atomic_xor_x2 [[RET:v\[[0-9]+:[0-9]+\]]], v[{{[0-9]+:[0-9]+}}], v{{\[[0-9]+:[0-9]+\]}} glc{{$}}
+; GCN: flat_store_dwordx2 v{{\[[0-9]+:[0-9]+\]}}, [[RET]]
+define void @atomic_xor_i64_ret(i64 addrspace(4)* %out, i64 addrspace(4)* %out2, i64 %in) {
+entry:
+  %tmp0 = atomicrmw volatile xor i64 addrspace(4)* %out, i64 %in seq_cst
+  store i64 %tmp0, i64 addrspace(4)* %out2
+  ret void
+}
+
+; GCN-LABEL: {{^}}atomic_xor_i64_addr64:
+; GCN: flat_atomic_xor_x2 v[{{[0-9]+:[0-9]+}}], v{{\[[0-9]+:[0-9]+\]$}}
+define void @atomic_xor_i64_addr64(i64 addrspace(4)* %out, i64 %in, i64 %index) {
+entry:
+  %ptr = getelementptr i64, i64 addrspace(4)* %out, i64 %index
+  %tmp0 = atomicrmw volatile xor i64 addrspace(4)* %ptr, i64 %in seq_cst
+  ret void
+}
+
+; GCN-LABEL: {{^}}atomic_xor_i64_ret_addr64:
+; GCN: flat_atomic_xor_x2 [[RET:v\[[0-9]+:[0-9]+\]]], v[{{[0-9]+:[0-9]+}}], v{{\[[0-9]+:[0-9]+\]}} glc{{$}}
+; GCN: flat_store_dwordx2 v{{\[[0-9]+:[0-9]+\]}}, [[RET]]
+define void @atomic_xor_i64_ret_addr64(i64 addrspace(4)* %out, i64 addrspace(4)* %out2, i64 %in, i64 %index) {
+entry:
+  %ptr = getelementptr i64, i64 addrspace(4)* %out, i64 %index
+  %tmp0 = atomicrmw volatile xor i64 addrspace(4)* %ptr, i64 %in seq_cst
+  store i64 %tmp0, i64 addrspace(4)* %out2
+  ret void
+}
+
+; GCN-LABEL: {{^}}atomic_load_i64_offset:
+; GCN: flat_load_dwordx2 [[RET:v\[[0-9]+:[0-9]\]]], v[{{[0-9]+}}:{{[0-9]+}}] glc{{$}}
+; GCN: flat_store_dwordx2 v{{\[[0-9]+:[0-9]+\]}}, [[RET]]
+define void @atomic_load_i64_offset(i64 addrspace(4)* %in, i64 addrspace(4)* %out) {
+entry:
+  %gep = getelementptr i64, i64 addrspace(4)* %in, i64 4
+  %val = load atomic i64, i64 addrspace(4)* %gep  seq_cst, align 8
+  store i64 %val, i64 addrspace(4)* %out
+  ret void
+}
+
+; GCN-LABEL: {{^}}atomic_load_i64:
+; GCN: flat_load_dwordx2 [[RET:v\[[0-9]+:[0-9]\]]], v[{{[0-9]+}}:{{[0-9]+}}] glc
+; GCN: flat_store_dwordx2 v{{\[[0-9]+:[0-9]+\]}}, [[RET]]
+define void @atomic_load_i64(i64 addrspace(4)* %in, i64 addrspace(4)* %out) {
+entry:
+  %val = load atomic i64, i64 addrspace(4)* %in seq_cst, align 8
+  store i64 %val, i64 addrspace(4)* %out
+  ret void
+}
+
+; GCN-LABEL: {{^}}atomic_load_i64_addr64_offset:
+; GCN: flat_load_dwordx2 [[RET:v\[[0-9]+:[0-9]+\]]], v[{{[0-9]+:[0-9]+}}] glc{{$}}
+; GCN: flat_store_dwordx2 v{{\[[0-9]+:[0-9]+\]}}, [[RET]]
+define void @atomic_load_i64_addr64_offset(i64 addrspace(4)* %in, i64 addrspace(4)* %out, i64 %index) {
+entry:
+  %ptr = getelementptr i64, i64 addrspace(4)* %in, i64 %index
+  %gep = getelementptr i64, i64 addrspace(4)* %ptr, i64 4
+  %val = load atomic i64, i64 addrspace(4)* %gep seq_cst, align 8
+  store i64 %val, i64 addrspace(4)* %out
+  ret void
+}
+
+; GCN-LABEL: {{^}}atomic_load_i64_addr64:
+; GCN: flat_load_dwordx2 [[RET:v\[[0-9]+:[0-9]+\]]], v[{{[0-9]+:[0-9]+}}] glc{{$}}
+; GCN: flat_store_dwordx2 v{{\[[0-9]+:[0-9]+\]}}, [[RET]]
+define void @atomic_load_i64_addr64(i64 addrspace(4)* %in, i64 addrspace(4)* %out, i64 %index) {
+entry:
+  %ptr = getelementptr i64, i64 addrspace(4)* %in, i64 %index
+  %val = load atomic i64, i64 addrspace(4)* %ptr seq_cst, align 8
+  store i64 %val, i64 addrspace(4)* %out
+  ret void
+}
+
+; GCN-LABEL: {{^}}atomic_store_i64_offset:
+; GCN: flat_store_dwordx2 [[RET:v\[[0-9]+:[0-9]\]]], v[{{[0-9]+}}:{{[0-9]+}}] glc{{$}}
+define void @atomic_store_i64_offset(i64 %in, i64 addrspace(4)* %out) {
+entry:
+  %gep = getelementptr i64, i64 addrspace(4)* %out, i64 4
+  store atomic i64 %in, i64 addrspace(4)* %gep  seq_cst, align 8
+  ret void
+}
+
+; GCN-LABEL: {{^}}atomic_store_i64:
+; GCN: flat_store_dwordx2 {{v\[[0-9]+:[0-9]\]}}, v[{{[0-9]+}}:{{[0-9]+}}] glc
+define void @atomic_store_i64(i64 %in, i64 addrspace(4)* %out) {
+entry:
+  store atomic i64 %in, i64 addrspace(4)* %out seq_cst, align 8
+  ret void
+}
+
+; GCN-LABEL: {{^}}atomic_store_i64_addr64_offset:
+; GCN: flat_store_dwordx2 {{v\[[0-9]+:[0-9]+\]}}, v[{{[0-9]+:[0-9]+}}] glc{{$}}
+define void @atomic_store_i64_addr64_offset(i64 %in, i64 addrspace(4)* %out, i64 %index) {
+entry:
+  %ptr = getelementptr i64, i64 addrspace(4)* %out, i64 %index
+  %gep = getelementptr i64, i64 addrspace(4)* %ptr, i64 4
+  store atomic i64 %in, i64 addrspace(4)* %gep seq_cst, align 8
+  ret void
+}
+
+; GCN-LABEL: {{^}}atomic_store_i64_addr64:
+; GCN: flat_store_dwordx2 {{v\[[0-9]+:[0-9]+\]}}, v[{{[0-9]+:[0-9]+}}] glc{{$}}
+define void @atomic_store_i64_addr64(i64 %in, i64 addrspace(4)* %out, i64 %index) {
+entry:
+  %ptr = getelementptr i64, i64 addrspace(4)* %out, i64 %index
+  store atomic i64 %in, i64 addrspace(4)* %ptr seq_cst, align 8
+  ret void
+}
+
+; GCN-LABEL: {{^}}atomic_cmpxchg_i64_offset:
+; GCN: flat_atomic_cmpswap_x2 v[{{[0-9]+\:[0-9]+}}], v[{{[0-9]+}}:{{[0-9]+}}]{{$}}
+define void @atomic_cmpxchg_i64_offset(i64 addrspace(4)* %out, i64 %in, i64 %old) {
+entry:
+  %gep = getelementptr i64, i64 addrspace(4)* %out, i64 4
+  %val = cmpxchg volatile i64 addrspace(4)* %gep, i64 %old, i64 %in seq_cst seq_cst
+  ret void
+}
+
+; GCN-LABEL: {{^}}atomic_cmpxchg_i64_soffset:
+; GCN: flat_atomic_cmpswap_x2 v[{{[0-9]+}}:{{[0-9]+}}], v[{{[0-9]+}}:{{[0-9]+}}]{{$}}
+define void @atomic_cmpxchg_i64_soffset(i64 addrspace(4)* %out, i64 %in, i64 %old) {
+entry:
+  %gep = getelementptr i64, i64 addrspace(4)* %out, i64 9000
+  %val = cmpxchg volatile i64 addrspace(4)* %gep, i64 %old, i64 %in seq_cst seq_cst
+  ret void
+}
+
+; GCN-LABEL: {{^}}atomic_cmpxchg_i64_ret_offset:
+; GCN: flat_atomic_cmpswap_x2 v{{\[}}[[RET:[0-9]+]]{{:[0-9]+}}], v[{{[0-9]+:[0-9]+}}], v[{{[0-9]+:[0-9]+}}] glc{{$}}
+; GCN: flat_store_dwordx2 v{{\[[0-9]+:[0-9]+\]}}, v{{\[}}[[RET]]:
+define void @atomic_cmpxchg_i64_ret_offset(i64 addrspace(4)* %out, i64 addrspace(4)* %out2, i64 %in, i64 %old) {
+entry:
+  %gep = getelementptr i64, i64 addrspace(4)* %out, i64 4
+  %val = cmpxchg volatile i64 addrspace(4)* %gep, i64 %old, i64 %in seq_cst seq_cst
+  %extract0 = extractvalue { i64, i1 } %val, 0
+  store i64 %extract0, i64 addrspace(4)* %out2
+  ret void
+}
+
+; GCN-LABEL: {{^}}atomic_cmpxchg_i64_addr64_offset:
+; GCN: flat_atomic_cmpswap_x2 v[{{[0-9]+\:[0-9]+}}], v[{{[0-9]+}}:{{[0-9]+}}]{{$}}
+define void @atomic_cmpxchg_i64_addr64_offset(i64 addrspace(4)* %out, i64 %in, i64 %index, i64 %old) {
+entry:
+  %ptr = getelementptr i64, i64 addrspace(4)* %out, i64 %index
+  %gep = getelementptr i64, i64 addrspace(4)* %ptr, i64 4
+  %val = cmpxchg volatile i64 addrspace(4)* %gep, i64 %old, i64 %in seq_cst seq_cst
+  ret void
+}
+
+; GCN-LABEL: {{^}}atomic_cmpxchg_i64_ret_addr64_offset:
+; GCN: flat_atomic_cmpswap_x2 v{{\[}}[[RET:[0-9]+]]:{{[0-9]+\]}}, v[{{[0-9]+:[0-9]+}}], v[{{[0-9]+:[0-9]+}}] glc{{$}}
+; GCN: flat_store_dwordx2 v{{\[[0-9]+:[0-9]+\]}}, v{{\[}}[[RET]]:
+define void @atomic_cmpxchg_i64_ret_addr64_offset(i64 addrspace(4)* %out, i64 addrspace(4)* %out2, i64 %in, i64 %index, i64 %old) {
+entry:
+  %ptr = getelementptr i64, i64 addrspace(4)* %out, i64 %index
+  %gep = getelementptr i64, i64 addrspace(4)* %ptr, i64 4
+  %val = cmpxchg volatile i64 addrspace(4)* %gep, i64 %old, i64 %in seq_cst seq_cst
+  %extract0 = extractvalue { i64, i1 } %val, 0
+  store i64 %extract0, i64 addrspace(4)* %out2
+  ret void
+}
+
+; GCN-LABEL: {{^}}atomic_cmpxchg_i64:
+; GCN: flat_atomic_cmpswap_x2 v[{{[0-9]+:[0-9]+}}], v[{{[0-9]+:[0-9]+}}]{{$}}
+define void @atomic_cmpxchg_i64(i64 addrspace(4)* %out, i64 %in, i64 %old) {
+entry:
+  %val = cmpxchg volatile i64 addrspace(4)* %out, i64 %old, i64 %in seq_cst seq_cst
+  ret void
+}
+
+; GCN-LABEL: {{^}}atomic_cmpxchg_i64_ret:
+; GCN: flat_atomic_cmpswap_x2 v{{\[}}[[RET:[0-9]+]]:{{[0-9]+\]}}, v[{{[0-9]+:[0-9]+}}], v[{{[0-9]+:[0-9]+}}] glc{{$}}
+; GCN: flat_store_dwordx2 {{v\[[0-9]+:[0-9]+\]}}, v{{\[}}[[RET]]:
+define void @atomic_cmpxchg_i64_ret(i64 addrspace(4)* %out, i64 addrspace(4)* %out2, i64 %in, i64 %old) {
+entry:
+  %val = cmpxchg volatile i64 addrspace(4)* %out, i64 %old, i64 %in seq_cst seq_cst
+  %extract0 = extractvalue { i64, i1 } %val, 0
+  store i64 %extract0, i64 addrspace(4)* %out2
+  ret void
+}
+
+; GCN-LABEL: {{^}}atomic_cmpxchg_i64_addr64:
+; GCN: flat_atomic_cmpswap_x2 v[{{[0-9]+:[0-9]+}}], v[{{[0-9]+:[0-9]+}}]{{$}}
+define void @atomic_cmpxchg_i64_addr64(i64 addrspace(4)* %out, i64 %in, i64 %index, i64 %old) {
+entry:
+  %ptr = getelementptr i64, i64 addrspace(4)* %out, i64 %index
+  %val = cmpxchg volatile i64 addrspace(4)* %ptr, i64 %old, i64 %in seq_cst seq_cst
+  ret void
+}
+
+; GCN-LABEL: {{^}}atomic_cmpxchg_i64_ret_addr64:
+; GCN: flat_atomic_cmpswap_x2 v{{\[}}[[RET:[0-9]+]]:{{[0-9]+\]}}, v[{{[0-9]+:[0-9]+}}], v[{{[0-9]+:[0-9]+}}] glc{{$}}
+; GCN: flat_store_dwordx2 {{v\[[0-9]+:[0-9]+\]}}, v{{\[}}[[RET]]:
+define void @atomic_cmpxchg_i64_ret_addr64(i64 addrspace(4)* %out, i64 addrspace(4)* %out2, i64 %in, i64 %index, i64 %old) {
+entry:
+  %ptr = getelementptr i64, i64 addrspace(4)* %out, i64 %index
+  %val = cmpxchg volatile i64 addrspace(4)* %ptr, i64 %old, i64 %in seq_cst seq_cst
+  %extract0 = extractvalue { i64, i1 } %val, 0
+  store i64 %extract0, i64 addrspace(4)* %out2
+  ret void
+}
diff --git a/llvm/test/CodeGen/AMDGPU/llvm.amdgcn.atomic.dec.ll b/llvm/test/CodeGen/AMDGPU/llvm.amdgcn.atomic.dec.ll
index d17179e96bc..93911d4a91f 100644
--- a/llvm/test/CodeGen/AMDGPU/llvm.amdgcn.atomic.dec.ll
+++ b/llvm/test/CodeGen/AMDGPU/llvm.amdgcn.atomic.dec.ll
@@ -3,9 +3,11 @@
 
 declare i32 @llvm.amdgcn.atomic.dec.i32.p1i32(i32 addrspace(1)* nocapture, i32) #2
 declare i32 @llvm.amdgcn.atomic.dec.i32.p3i32(i32 addrspace(3)* nocapture, i32) #2
+declare i32 @llvm.amdgcn.atomic.dec.i32.p4i32(i32 addrspace(4)* nocapture, i32) #2
 
 declare i64 @llvm.amdgcn.atomic.dec.i64.p1i64(i64 addrspace(1)* nocapture, i64) #2
 declare i64 @llvm.amdgcn.atomic.dec.i64.p3i64(i64 addrspace(3)* nocapture, i64) #2
+declare i64 @llvm.amdgcn.atomic.dec.i64.p4i64(i64 addrspace(4)* nocapture, i64) #2
 
 declare i32 @llvm.amdgcn.workitem.id.x() #1
 
@@ -108,6 +110,131 @@ define void @global_atomic_dec_noret_i32_offset_addr64(i32 addrspace(1)* %ptr) #
   ret void
 }
 
+; GCN-LABEL: {{^}}flat_atomic_dec_ret_i32:
+; GCN: v_mov_b32_e32 [[K:v[0-9]+]], 42
+; GCN: flat_atomic_dec v{{[0-9]+}}, v{{\[[0-9]+:[0-9]+\]}}, [[K]] glc{{$}}
+define void @flat_atomic_dec_ret_i32(i32 addrspace(4)* %out, i32 addrspace(4)* %ptr) #0 {
+  %result = call i32 @llvm.amdgcn.atomic.dec.i32.p4i32(i32 addrspace(4)* %ptr, i32 42)
+  store i32 %result, i32 addrspace(4)* %out
+  ret void
+}
+
+; GCN-LABEL: {{^}}flat_atomic_dec_ret_i32_offset:
+; GCN: v_mov_b32_e32 [[K:v[0-9]+]], 42
+; GCN: flat_atomic_dec v{{[0-9]+}}, v{{\[[0-9]+:[0-9]+\]}}, [[K]] glc{{$}}
+define void @flat_atomic_dec_ret_i32_offset(i32 addrspace(4)* %out, i32 addrspace(4)* %ptr) #0 {
+  %gep = getelementptr i32, i32 addrspace(4)* %ptr, i32 4
+  %result = call i32 @llvm.amdgcn.atomic.dec.i32.p4i32(i32 addrspace(4)* %gep, i32 42)
+  store i32 %result, i32 addrspace(4)* %out
+  ret void
+}
+
+; FUNC-LABEL: {{^}}flat_atomic_dec_noret_i32:
+; GCN: flat_atomic_dec v{{\[[0-9]+:[0-9]+\]}}, [[K]]{{$}}
+define void @flat_atomic_dec_noret_i32(i32 addrspace(4)* %ptr) nounwind {
+  %result = call i32 @llvm.amdgcn.atomic.dec.i32.p4i32(i32 addrspace(4)* %ptr, i32 42)
+  ret void
+}
+
+; FUNC-LABEL: {{^}}flat_atomic_dec_noret_i32_offset:
+; GCN: v_mov_b32_e32 [[K:v[0-9]+]], 42
+; GCN: flat_atomic_dec v{{\[[0-9]+:[0-9]+\]}}, [[K]]{{$}}
+define void @flat_atomic_dec_noret_i32_offset(i32 addrspace(4)* %ptr) nounwind {
+  %gep = getelementptr i32, i32 addrspace(4)* %ptr, i32 4
+  %result = call i32 @llvm.amdgcn.atomic.dec.i32.p4i32(i32 addrspace(4)* %gep, i32 42)
+  ret void
+}
+
+; GCN-LABEL: {{^}}flat_atomic_dec_ret_i32_offset_addr64:
+; GCN: v_mov_b32_e32 [[K:v[0-9]+]], 42
+; GCN: flat_atomic_dec v{{[0-9]+}}, v{{\[[0-9]+:[0-9]+\]}}, [[K]] glc{{$}}
+define void @flat_atomic_dec_ret_i32_offset_addr64(i32 addrspace(4)* %out, i32 addrspace(4)* %ptr) #0 {
+  %id = call i32 @llvm.amdgcn.workitem.id.x()
+  %gep.tid = getelementptr i32, i32 addrspace(4)* %ptr, i32 %id
+  %out.gep = getelementptr i32, i32 addrspace(4)* %out, i32 %id
+  %gep = getelementptr i32, i32 addrspace(4)* %gep.tid, i32 5
+  %result = call i32 @llvm.amdgcn.atomic.dec.i32.p4i32(i32 addrspace(4)* %gep, i32 42)
+  store i32 %result, i32 addrspace(4)* %out.gep
+  ret void
+}
+
+; GCN-LABEL: {{^}}flat_atomic_dec_noret_i32_offset_addr64:
+; GCN: v_mov_b32_e32 [[K:v[0-9]+]], 42
+; GCN: flat_atomic_dec v{{\[[0-9]+:[0-9]+\]}}, [[K]]{{$}}
+define void @flat_atomic_dec_noret_i32_offset_addr64(i32 addrspace(4)* %ptr) #0 {
+  %id = call i32 @llvm.amdgcn.workitem.id.x()
+  %gep.tid = getelementptr i32, i32 addrspace(4)* %ptr, i32 %id
+  %gep = getelementptr i32, i32 addrspace(4)* %gep.tid, i32 5
+  %result = call i32 @llvm.amdgcn.atomic.dec.i32.p4i32(i32 addrspace(4)* %gep, i32 42)
+  ret void
+}
+
+; GCN-LABEL: {{^}}flat_atomic_dec_ret_i64:
+; GCN-DAG: v_mov_b32_e32 v[[KLO:[0-9]+]], 42
+; GCN-DAG: v_mov_b32_e32 v[[KHI:[0-9]+]], 0{{$}}
+; GCN: flat_atomic_dec_x2 v{{\[[0-9]+:[0-9]+\]}}, v{{\[[0-9]+:[0-9]+\]}}, v{{\[}}[[KLO]]:[[KHI]]{{\]}} glc{{$}}
+define void @flat_atomic_dec_ret_i64(i64 addrspace(4)* %out, i64 addrspace(4)* %ptr) #0 {
+  %result = call i64 @llvm.amdgcn.atomic.dec.i64.p4i64(i64 addrspace(4)* %ptr, i64 42)
+  store i64 %result, i64 addrspace(4)* %out
+  ret void
+}
+
+; GCN-LABEL: {{^}}flat_atomic_dec_ret_i64_offset:
+; GCN-DAG: v_mov_b32_e32 v[[KLO:[0-9]+]], 42
+; GCN-DAG: v_mov_b32_e32 v[[KHI:[0-9]+]], 0{{$}}
+; GCN: flat_atomic_dec_x2 v{{\[[0-9]+:[0-9]+\]}}, v{{\[[0-9]+:[0-9]+\]}}, v{{\[}}[[KLO]]:[[KHI]]{{\]}} glc{{$}}
+define void @flat_atomic_dec_ret_i64_offset(i64 addrspace(4)* %out, i64 addrspace(4)* %ptr) #0 {
+  %gep = getelementptr i64, i64 addrspace(4)* %ptr, i32 4
+  %result = call i64 @llvm.amdgcn.atomic.dec.i64.p4i64(i64 addrspace(4)* %gep, i64 42)
+  store i64 %result, i64 addrspace(4)* %out
+  ret void
+}
+
+; FUNC-LABEL: {{^}}flat_atomic_dec_noret_i64:
+; GCN-DAG: v_mov_b32_e32 v[[KLO:[0-9]+]], 42
+; GCN-DAG: v_mov_b32_e32 v[[KHI:[0-9]+]], 0{{$}}
+; GCN: flat_atomic_dec_x2 v{{\[[0-9]+:[0-9]+\]}}, v{{\[}}[[KLO]]:[[KHI]]{{\]$}}
+define void @flat_atomic_dec_noret_i64(i64 addrspace(4)* %ptr) nounwind {
+  %result = call i64 @llvm.amdgcn.atomic.dec.i64.p4i64(i64 addrspace(4)* %ptr, i64 42)
+  ret void
+}
+
+; FUNC-LABEL: {{^}}flat_atomic_dec_noret_i64_offset:
+; GCN-DAG: v_mov_b32_e32 v[[KLO:[0-9]+]], 42
+; GCN-DAG: v_mov_b32_e32 v[[KHI:[0-9]+]], 0{{$}}
+; GCN: flat_atomic_dec_x2 v{{\[[0-9]+:[0-9]+\]}}, v{{\[}}[[KLO]]:[[KHI]]{{\]$}}
+define void @flat_atomic_dec_noret_i64_offset(i64 addrspace(4)* %ptr) nounwind {
+  %gep = getelementptr i64, i64 addrspace(4)* %ptr, i32 4
+  %result = call i64 @llvm.amdgcn.atomic.dec.i64.p4i64(i64 addrspace(4)* %gep, i64 42)
+  ret void
+}
+
+; GCN-LABEL: {{^}}flat_atomic_dec_ret_i64_offset_addr64:
+; GCN-DAG: v_mov_b32_e32 v[[KLO:[0-9]+]], 42
+; GCN-DAG: v_mov_b32_e32 v[[KHI:[0-9]+]], 0{{$}}
+; GCN: flat_atomic_dec_x2 v{{\[[0-9]+:[0-9]+\]}}, v{{\[[0-9]+:[0-9]+\]}}, v{{\[}}[[KLO]]:[[KHI]]{{\]}} glc{{$}}
+define void @flat_atomic_dec_ret_i64_offset_addr64(i64 addrspace(4)* %out, i64 addrspace(4)* %ptr) #0 {
+  %id = call i32 @llvm.amdgcn.workitem.id.x()
+  %gep.tid = getelementptr i64, i64 addrspace(4)* %ptr, i32 %id
+  %out.gep = getelementptr i64, i64 addrspace(4)* %out, i32 %id
+  %gep = getelementptr i64, i64 addrspace(4)* %gep.tid, i32 5
+  %result = call i64 @llvm.amdgcn.atomic.dec.i64.p4i64(i64 addrspace(4)* %gep, i64 42)
+  store i64 %result, i64 addrspace(4)* %out.gep
+  ret void
+}
+
+; GCN-LABEL: {{^}}flat_atomic_dec_noret_i64_offset_addr64:
+; GCN-DAG: v_mov_b32_e32 v[[KLO:[0-9]+]], 42
+; GCN-DAG: v_mov_b32_e32 v[[KHI:[0-9]+]], 0{{$}}
+; GCN: flat_atomic_dec_x2 v{{\[[0-9]+:[0-9]+\]}}, v{{\[}}[[KLO]]:[[KHI]]{{\]$}}
+define void @flat_atomic_dec_noret_i64_offset_addr64(i64 addrspace(4)* %ptr) #0 {
+  %id = call i32 @llvm.amdgcn.workitem.id.x()
+  %gep.tid = getelementptr i64, i64 addrspace(4)* %ptr, i32 %id
+  %gep = getelementptr i64, i64 addrspace(4)* %gep.tid, i32 5
+  %result = call i64 @llvm.amdgcn.atomic.dec.i64.p4i64(i64 addrspace(4)* %gep, i64 42)
+  ret void
+}
+
 @lds0 = addrspace(3) global [512 x i32] undef
 
 ; SI-LABEL: {{^}}atomic_dec_shl_base_lds_0:
@@ -249,3 +376,12 @@ define void @atomic_dec_shl_base_lds_0_i64(i64 addrspace(1)* %out, i32 addrspace
 attributes #0 = { nounwind }
 attributes #1 = { nounwind readnone }
 attributes #2 = { nounwind argmemonly }
+
+
+
+
+
+
+
+
+
diff --git a/llvm/test/CodeGen/AMDGPU/llvm.amdgcn.atomic.inc.ll b/llvm/test/CodeGen/AMDGPU/llvm.amdgcn.atomic.inc.ll
index 93aa1faba2a..181d68c8ea7 100644
--- a/llvm/test/CodeGen/AMDGPU/llvm.amdgcn.atomic.inc.ll
+++ b/llvm/test/CodeGen/AMDGPU/llvm.amdgcn.atomic.inc.ll
@@ -3,9 +3,11 @@
 
 declare i32 @llvm.amdgcn.atomic.inc.i32.p1i32(i32 addrspace(1)* nocapture, i32) #2
 declare i32 @llvm.amdgcn.atomic.inc.i32.p3i32(i32 addrspace(3)* nocapture, i32) #2
+declare i32 @llvm.amdgcn.atomic.inc.i32.p4i32(i32 addrspace(4)* nocapture, i32) #2
 
 declare i64 @llvm.amdgcn.atomic.inc.i64.p1i64(i64 addrspace(1)* nocapture, i64) #2
 declare i64 @llvm.amdgcn.atomic.inc.i64.p3i64(i64 addrspace(3)* nocapture, i64) #2
+declare i64 @llvm.amdgcn.atomic.inc.i64.p4i64(i64 addrspace(4)* nocapture, i64) #2
 
 declare i32 @llvm.amdgcn.workitem.id.x() #1
 
@@ -231,6 +233,65 @@ define void @global_atomic_inc_noret_i64_offset_addr64(i64 addrspace(1)* %ptr) #
   ret void
 }
 
+; GCN-LABEL: {{^}}flat_atomic_inc_ret_i32:
+; GCN: v_mov_b32_e32 [[K:v[0-9]+]], 42
+; GCN: flat_atomic_inc v{{[0-9]+}}, v{{\[[0-9]+:[0-9]+\]}}, [[K]] glc{{$}}
+define void @flat_atomic_inc_ret_i32(i32 addrspace(4)* %out, i32 addrspace(4)* %ptr) #0 {
+  %result = call i32 @llvm.amdgcn.atomic.inc.i32.p4i32(i32 addrspace(4)* %ptr, i32 42)
+  store i32 %result, i32 addrspace(4)* %out
+  ret void
+}
+
+; GCN-LABEL: {{^}}flat_atomic_inc_ret_i32_offset:
+; GCN: v_mov_b32_e32 [[K:v[0-9]+]], 42
+; GCN: flat_atomic_inc v{{[0-9]+}}, v{{\[[0-9]+:[0-9]+\]}}, [[K]] glc{{$}}
+define void @flat_atomic_inc_ret_i32_offset(i32 addrspace(4)* %out, i32 addrspace(4)* %ptr) #0 {
+  %gep = getelementptr i32, i32 addrspace(4)* %ptr, i32 4
+  %result = call i32 @llvm.amdgcn.atomic.inc.i32.p4i32(i32 addrspace(4)* %gep, i32 42)
+  store i32 %result, i32 addrspace(4)* %out
+  ret void
+}
+
+; FUNC-LABEL: {{^}}flat_atomic_inc_noret_i32:
+; GCN: flat_atomic_inc v{{\[[0-9]+:[0-9]+\]}}, [[K]]{{$}}
+define void @flat_atomic_inc_noret_i32(i32 addrspace(4)* %ptr) nounwind {
+  %result = call i32 @llvm.amdgcn.atomic.inc.i32.p4i32(i32 addrspace(4)* %ptr, i32 42)
+  ret void
+}
+
+; FUNC-LABEL: {{^}}flat_atomic_inc_noret_i32_offset:
+; GCN: v_mov_b32_e32 [[K:v[0-9]+]], 42
+; GCN: flat_atomic_inc v{{\[[0-9]+:[0-9]+\]}}, [[K]]{{$}}
+define void @flat_atomic_inc_noret_i32_offset(i32 addrspace(4)* %ptr) nounwind {
+  %gep = getelementptr i32, i32 addrspace(4)* %ptr, i32 4
+  %result = call i32 @llvm.amdgcn.atomic.inc.i32.p4i32(i32 addrspace(4)* %gep, i32 42)
+  ret void
+}
+
+; GCN-LABEL: {{^}}flat_atomic_inc_ret_i32_offset_addr64:
+; GCN: v_mov_b32_e32 [[K:v[0-9]+]], 42
+; GCN: flat_atomic_inc v{{[0-9]+}}, v{{\[[0-9]+:[0-9]+\]}}, [[K]] glc{{$}}
+define void @flat_atomic_inc_ret_i32_offset_addr64(i32 addrspace(4)* %out, i32 addrspace(4)* %ptr) #0 {
+  %id = call i32 @llvm.amdgcn.workitem.id.x()
+  %gep.tid = getelementptr i32, i32 addrspace(4)* %ptr, i32 %id
+  %out.gep = getelementptr i32, i32 addrspace(4)* %out, i32 %id
+  %gep = getelementptr i32, i32 addrspace(4)* %gep.tid, i32 5
+  %result = call i32 @llvm.amdgcn.atomic.inc.i32.p4i32(i32 addrspace(4)* %gep, i32 42)
+  store i32 %result, i32 addrspace(4)* %out.gep
+  ret void
+}
+
+; GCN-LABEL: {{^}}flat_atomic_inc_noret_i32_offset_addr64:
+; GCN: v_mov_b32_e32 [[K:v[0-9]+]], 42
+; GCN: flat_atomic_inc v{{\[[0-9]+:[0-9]+\]}}, [[K]]{{$}}
+define void @flat_atomic_inc_noret_i32_offset_addr64(i32 addrspace(4)* %ptr) #0 {
+  %id = call i32 @llvm.amdgcn.workitem.id.x()
+  %gep.tid = getelementptr i32, i32 addrspace(4)* %ptr, i32 %id
+  %gep = getelementptr i32, i32 addrspace(4)* %gep.tid, i32 5
+  %result = call i32 @llvm.amdgcn.atomic.inc.i32.p4i32(i32 addrspace(4)* %gep, i32 42)
+  ret void
+}
+
 @lds1 = addrspace(3) global [512 x i64] undef, align 8
 
 ; GCN-LABEL: {{^}}atomic_inc_shl_base_lds_0_i64:
@@ -249,3 +310,74 @@ define void @atomic_inc_shl_base_lds_0_i64(i64 addrspace(1)* %out, i32 addrspace
 attributes #0 = { nounwind }
 attributes #1 = { nounwind readnone }
 attributes #2 = { nounwind argmemonly }
+
+
+
+
+
+
+; GCN-LABEL: {{^}}flat_atomic_inc_ret_i64:
+; GCN-DAG: v_mov_b32_e32 v[[KLO:[0-9]+]], 42
+; GCN-DAG: v_mov_b32_e32 v[[KHI:[0-9]+]], 0{{$}}
+; GCN: flat_atomic_inc_x2 v{{\[[0-9]+:[0-9]+\]}}, v{{\[[0-9]+:[0-9]+\]}}, v{{\[}}[[KLO]]:[[KHI]]{{\]}} glc{{$}}
+define void @flat_atomic_inc_ret_i64(i64 addrspace(4)* %out, i64 addrspace(4)* %ptr) #0 {
+  %result = call i64 @llvm.amdgcn.atomic.inc.i64.p4i64(i64 addrspace(4)* %ptr, i64 42)
+  store i64 %result, i64 addrspace(4)* %out
+  ret void
+}
+
+; GCN-LABEL: {{^}}flat_atomic_inc_ret_i64_offset:
+; GCN-DAG: v_mov_b32_e32 v[[KLO:[0-9]+]], 42
+; GCN-DAG: v_mov_b32_e32 v[[KHI:[0-9]+]], 0{{$}}
+; GCN: flat_atomic_inc_x2 v{{\[[0-9]+:[0-9]+\]}}, v{{\[[0-9]+:[0-9]+\]}}, v{{\[}}[[KLO]]:[[KHI]]{{\]}} glc{{$}}
+define void @flat_atomic_inc_ret_i64_offset(i64 addrspace(4)* %out, i64 addrspace(4)* %ptr) #0 {
+  %gep = getelementptr i64, i64 addrspace(4)* %ptr, i32 4
+  %result = call i64 @llvm.amdgcn.atomic.inc.i64.p4i64(i64 addrspace(4)* %gep, i64 42)
+  store i64 %result, i64 addrspace(4)* %out
+  ret void
+}
+
+; FUNC-LABEL: {{^}}flat_atomic_inc_noret_i64:
+; GCN-DAG: v_mov_b32_e32 v[[KLO:[0-9]+]], 42
+; GCN-DAG: v_mov_b32_e32 v[[KHI:[0-9]+]], 0{{$}}
+; GCN: flat_atomic_inc_x2 v{{\[[0-9]+:[0-9]+\]}}, v{{\[}}[[KLO]]:[[KHI]]{{\]$}}
+define void @flat_atomic_inc_noret_i64(i64 addrspace(4)* %ptr) nounwind {
+  %result = call i64 @llvm.amdgcn.atomic.inc.i64.p4i64(i64 addrspace(4)* %ptr, i64 42)
+  ret void
+}
+
+; FUNC-LABEL: {{^}}flat_atomic_inc_noret_i64_offset:
+; GCN-DAG: v_mov_b32_e32 v[[KLO:[0-9]+]], 42
+; GCN-DAG: v_mov_b32_e32 v[[KHI:[0-9]+]], 0{{$}}
+; GCN: flat_atomic_inc_x2 v{{\[[0-9]+:[0-9]+\]}}, v{{\[}}[[KLO]]:[[KHI]]{{\]$}}
+define void @flat_atomic_inc_noret_i64_offset(i64 addrspace(4)* %ptr) nounwind {
+  %gep = getelementptr i64, i64 addrspace(4)* %ptr, i32 4
+  %result = call i64 @llvm.amdgcn.atomic.inc.i64.p4i64(i64 addrspace(4)* %gep, i64 42)
+  ret void
+}
+
+; GCN-LABEL: {{^}}flat_atomic_inc_ret_i64_offset_addr64:
+; GCN-DAG: v_mov_b32_e32 v[[KLO:[0-9]+]], 42
+; GCN-DAG: v_mov_b32_e32 v[[KHI:[0-9]+]], 0{{$}}
+; GCN: flat_atomic_inc_x2 v{{\[[0-9]+:[0-9]+\]}}, v{{\[[0-9]+:[0-9]+\]}}, v{{\[}}[[KLO]]:[[KHI]]{{\]}} glc{{$}}
+define void @flat_atomic_inc_ret_i64_offset_addr64(i64 addrspace(4)* %out, i64 addrspace(4)* %ptr) #0 {
+  %id = call i32 @llvm.amdgcn.workitem.id.x()
+  %gep.tid = getelementptr i64, i64 addrspace(4)* %ptr, i32 %id
+  %out.gep = getelementptr i64, i64 addrspace(4)* %out, i32 %id
+  %gep = getelementptr i64, i64 addrspace(4)* %gep.tid, i32 5
+  %result = call i64 @llvm.amdgcn.atomic.inc.i64.p4i64(i64 addrspace(4)* %gep, i64 42)
+  store i64 %result, i64 addrspace(4)* %out.gep
+  ret void
+}
+
+; GCN-LABEL: {{^}}flat_atomic_inc_noret_i64_offset_addr64:
+; GCN-DAG: v_mov_b32_e32 v[[KLO:[0-9]+]], 42
+; GCN-DAG: v_mov_b32_e32 v[[KHI:[0-9]+]], 0{{$}}
+; GCN: flat_atomic_inc_x2 v{{\[[0-9]+:[0-9]+\]}}, v{{\[}}[[KLO]]:[[KHI]]{{\]$}}
+define void @flat_atomic_inc_noret_i64_offset_addr64(i64 addrspace(4)* %ptr) #0 {
+  %id = call i32 @llvm.amdgcn.workitem.id.x()
+  %gep.tid = getelementptr i64, i64 addrspace(4)* %ptr, i32 %id
+  %gep = getelementptr i64, i64 addrspace(4)* %gep.tid, i32 5
+  %result = call i64 @llvm.amdgcn.atomic.inc.i64.p4i64(i64 addrspace(4)* %gep, i64 42)
+  ret void
+}