diff --git a/llvm/lib/Target/AMDGPU/AMDGPUInstructionSelector.h b/llvm/lib/Target/AMDGPU/AMDGPUInstructionSelector.h
--- a/llvm/lib/Target/AMDGPU/AMDGPUInstructionSelector.h
+++ b/llvm/lib/Target/AMDGPU/AMDGPUInstructionSelector.h
@@ -144,7 +144,8 @@
   bool selectG_INSERT_VECTOR_ELT(MachineInstr &I) const;
   bool selectG_SHUFFLE_VECTOR(MachineInstr &I) const;
   bool selectAMDGPU_BUFFER_ATOMIC_FADD(MachineInstr &I) const;
-  bool selectGlobalAtomicFaddIntrinsic(MachineInstr &I) const;
+  bool selectGlobalAtomicFadd(MachineInstr &I, MachineOperand &AddrOp,
+                              MachineOperand &DataOp) const;
   bool selectBVHIntrinsic(MachineInstr &I) const;
 
   std::pair<Register, unsigned> selectVOP3ModsImpl(MachineOperand &Root,
diff --git a/llvm/lib/Target/AMDGPU/AMDGPUInstructionSelector.cpp b/llvm/lib/Target/AMDGPU/AMDGPUInstructionSelector.cpp
--- a/llvm/lib/Target/AMDGPU/AMDGPUInstructionSelector.cpp
+++ b/llvm/lib/Target/AMDGPU/AMDGPUInstructionSelector.cpp
@@ -1749,7 +1749,7 @@
   case Intrinsic::amdgcn_s_barrier:
     return selectSBarrier(I);
   case Intrinsic::amdgcn_global_atomic_fadd:
-    return selectGlobalAtomicFaddIntrinsic(I);
+    return selectGlobalAtomicFadd(I, I.getOperand(2), I.getOperand(3));
   default: {
     return selectImpl(I, *CoverageInfo);
   }
@@ -2352,6 +2352,13 @@
 
 bool AMDGPUInstructionSelector::selectG_LOAD_STORE_ATOMICRMW(
   MachineInstr &I) const {
+  if (I.getOpcode() == TargetOpcode::G_ATOMICRMW_FADD) {
+    const LLT PtrTy = MRI->getType(I.getOperand(1).getReg());
+    unsigned AS = PtrTy.getAddressSpace();
+    if (AS == AMDGPUAS::GLOBAL_ADDRESS)
+      return selectGlobalAtomicFadd(I, I.getOperand(1), I.getOperand(2));
+  }
+
   initM0(I);
   return selectImpl(I, *CoverageInfo);
 }
@@ -2994,11 +3001,14 @@
   return true;
 }
 
-bool AMDGPUInstructionSelector::selectGlobalAtomicFaddIntrinsic(
-  MachineInstr &MI) const{
+bool AMDGPUInstructionSelector::selectGlobalAtomicFadd(
+  MachineInstr &MI, MachineOperand &AddrOp, MachineOperand &DataOp) const {
 
-  if (STI.hasGFX90AInsts())
+  if (STI.hasGFX90AInsts()) {
+    // gfx90a adds return versions of the global atomic fadd instructions so no
+    // special handling is required.
     return selectImpl(MI, *CoverageInfo);
+  }
 
   MachineBasicBlock *MBB = MI.getParent();
   const DebugLoc &DL = MI.getDebugLoc();
@@ -3015,9 +3025,9 @@
   // FIXME: This is only needed because tablegen requires number of dst operands
   // in match and replace pattern to be the same. Otherwise patterns can be
   // exported from SDag path.
-  auto Addr = selectFlatOffsetImpl<true>(MI.getOperand(2));
+  auto Addr = selectFlatOffsetImpl<true>(AddrOp);
 
-  Register Data = MI.getOperand(3).getReg();
+  Register Data = DataOp.getReg();
   const unsigned Opc = MRI->getType(Data).isVector() ?
     AMDGPU::GLOBAL_ATOMIC_PK_ADD_F16 : AMDGPU::GLOBAL_ATOMIC_ADD_F32;
   auto MIB = BuildMI(*MBB, &MI, DL, TII.get(Opc))
diff --git a/llvm/lib/Target/AMDGPU/AMDGPULegalizerInfo.cpp b/llvm/lib/Target/AMDGPU/AMDGPULegalizerInfo.cpp
--- a/llvm/lib/Target/AMDGPU/AMDGPULegalizerInfo.cpp
+++ b/llvm/lib/Target/AMDGPU/AMDGPULegalizerInfo.cpp
@@ -1289,12 +1289,14 @@
     Atomics.legalFor({{S32, FlatPtr}, {S64, FlatPtr}});
   }
 
+  auto &Atomic = getActionDefinitionsBuilder(G_ATOMICRMW_FADD);
   if (ST.hasLDSFPAtomics()) {
-    auto &Atomic = getActionDefinitionsBuilder(G_ATOMICRMW_FADD)
-      .legalFor({{S32, LocalPtr}, {S32, RegionPtr}});
+    Atomic.legalFor({{S32, LocalPtr}, {S32, RegionPtr}});
     if (ST.hasGFX90AInsts())
       Atomic.legalFor({{S64, LocalPtr}});
   }
+  if (ST.hasAtomicFaddInsts())
+    Atomic.legalFor({{S32, GlobalPtr}});
 
   // BUFFER/FLAT_ATOMIC_CMP_SWAP on GCN GPUs needs input marshalling, and output
   // demarshalling
diff --git a/llvm/test/CodeGen/AMDGPU/GlobalISel/legalize-atomicrmw-fadd-global.mir b/llvm/test/CodeGen/AMDGPU/GlobalISel/legalize-atomicrmw-fadd-global.mir
new file mode 100644
--- /dev/null
+++ b/llvm/test/CodeGen/AMDGPU/GlobalISel/legalize-atomicrmw-fadd-global.mir
@@ -0,0 +1,22 @@
+# NOTE: Assertions have been autogenerated by utils/update_mir_test_checks.py
+# RUN: llc -mtriple=amdgcn-mesa-mesa3d -mcpu=gfx908 -O0 -run-pass=legalizer %s -o - | FileCheck %s
+# RUN: llc -mtriple=amdgcn-mesa-mesa3d -mcpu=gfx90a -O0 -run-pass=legalizer %s -o - | FileCheck %s
+
+# RUN: llc -mtriple=amdgcn-mesa-mesa3d -mcpu=fiji -O0 -run-pass=legalizer -global-isel-abort=2 -pass-remarks-missed='gisel.*' -o /dev/null %s 2>&1 | FileCheck -check-prefix=ERR %s
+
+# ERR: remark: <unknown>:0:0: unable to legalize instruction: %2:_(s32) = G_ATOMICRMW_FADD %0:_(p1), %1:_ :: (load store seq_cst 4, addrspace 1) (in function: atomicrmw_fadd_global_i32)
+
+---
+name: atomicrmw_fadd_global_i32
+
+body: |
+  bb.0:
+    liveins: $sgpr0_sgpr1, $sgpr2
+    ; CHECK-LABEL: name: atomicrmw_fadd_global_i32
+    ; CHECK: [[COPY:%[0-9]+]]:_(p1) = COPY $sgpr0_sgpr1
+    ; CHECK: [[COPY1:%[0-9]+]]:_(s32) = COPY $sgpr2
+    ; CHECK: [[ATOMICRMW_FADD:%[0-9]+]]:_(s32) = G_ATOMICRMW_FADD [[COPY]](p1), [[COPY1]] :: (load store seq_cst 4, addrspace 1)
+    %0:_(p1) = COPY $sgpr0_sgpr1
+    %1:_(s32) = COPY $sgpr2
+    %2:_(s32) = G_ATOMICRMW_FADD %0, %1 :: (load store seq_cst 4, addrspace 1)
+...
diff --git a/llvm/test/CodeGen/AMDGPU/GlobalISel/legalize-atomicrmw-fadd.mir b/llvm/test/CodeGen/AMDGPU/GlobalISel/legalize-atomicrmw-fadd-local.mir
rename from llvm/test/CodeGen/AMDGPU/GlobalISel/legalize-atomicrmw-fadd.mir
rename to llvm/test/CodeGen/AMDGPU/GlobalISel/legalize-atomicrmw-fadd-local.mir
diff --git a/llvm/test/CodeGen/AMDGPU/global-atomics-fp.ll b/llvm/test/CodeGen/AMDGPU/global-atomics-fp.ll
--- a/llvm/test/CodeGen/AMDGPU/global-atomics-fp.ll
+++ b/llvm/test/CodeGen/AMDGPU/global-atomics-fp.ll
@@ -1,340 +1,656 @@
 ; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
-; RUN: llc -march=amdgcn -mcpu=gfx900 -verify-machineinstrs < %s | FileCheck -enable-var-scope -check-prefixes=GCN,GFX900 %s
-; RUN: llc -march=amdgcn -mcpu=gfx908 -verify-machineinstrs < %s | FileCheck -enable-var-scope -check-prefixes=GCN,GFX908 %s
-; RUN: llc -march=amdgcn -mcpu=gfx90a -verify-machineinstrs < %s | FileCheck -enable-var-scope -check-prefixes=GCN,GFX90A %s
+; RUN: llc -march=amdgcn -mcpu=gfx900 -verify-machineinstrs < %s | FileCheck -enable-var-scope -check-prefixes=SDAG,SDAG-GFX900 %s
+; RUN: llc -march=amdgcn -mcpu=gfx908 -verify-machineinstrs < %s | FileCheck -enable-var-scope -check-prefixes=SDAG,SDAG-GFX908 %s
+; RUN: llc -march=amdgcn -mcpu=gfx90a -verify-machineinstrs < %s | FileCheck -enable-var-scope -check-prefixes=SDAG,SDAG-GFX90A %s
+; RUN: llc -global-isel -march=amdgcn -mcpu=gfx900 -verify-machineinstrs < %s | FileCheck -enable-var-scope -check-prefixes=GISEL,GISEL-GFX900 %s
+; RUN: llc -global-isel -march=amdgcn -mcpu=gfx908 -verify-machineinstrs < %s | FileCheck -enable-var-scope -check-prefixes=GISEL,GISEL-GFX908 %s
+; RUN: llc -global-isel -march=amdgcn -mcpu=gfx90a -verify-machineinstrs < %s | FileCheck -enable-var-scope -check-prefixes=GISEL,GISEL-GFX90A %s
 
 define amdgpu_kernel void @global_atomic_fadd_ret_f32(float addrspace(1)* %ptr) #0 {
-; GFX900-LABEL: global_atomic_fadd_ret_f32:
-; GFX900:       ; %bb.0:
-; GFX900-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x24
-; GFX900-NEXT:    s_mov_b64 s[2:3], 0
-; GFX900-NEXT:    s_waitcnt lgkmcnt(0)
-; GFX900-NEXT:    s_load_dword s4, s[0:1], 0x0
-; GFX900-NEXT:    s_waitcnt lgkmcnt(0)
-; GFX900-NEXT:    v_mov_b32_e32 v0, s4
-; GFX900-NEXT:  BB0_1: ; %atomicrmw.start
-; GFX900-NEXT:    ; =>This Inner Loop Header: Depth=1
-; GFX900-NEXT:    v_mov_b32_e32 v1, v0
-; GFX900-NEXT:    v_mov_b32_e32 v2, 0
-; GFX900-NEXT:    v_add_f32_e32 v0, 4.0, v1
-; GFX900-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
-; GFX900-NEXT:    global_atomic_cmpswap v0, v2, v[0:1], s[0:1] glc
-; GFX900-NEXT:    s_waitcnt vmcnt(0)
-; GFX900-NEXT:    buffer_wbinvl1_vol
-; GFX900-NEXT:    v_cmp_eq_u32_e32 vcc, v0, v1
-; GFX900-NEXT:    s_or_b64 s[2:3], vcc, s[2:3]
-; GFX900-NEXT:    s_andn2_b64 exec, exec, s[2:3]
-; GFX900-NEXT:    s_cbranch_execnz BB0_1
-; GFX900-NEXT:  ; %bb.2: ; %atomicrmw.end
-; GFX900-NEXT:    s_or_b64 exec, exec, s[2:3]
-; GFX900-NEXT:    global_store_dword v[0:1], v0, off
-; GFX900-NEXT:    s_endpgm
+; SDAG-GFX900-LABEL: global_atomic_fadd_ret_f32:
+; SDAG-GFX900:       ; %bb.0:
+; SDAG-GFX900-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x24
+; SDAG-GFX900-NEXT:    s_mov_b64 s[2:3], 0
+; SDAG-GFX900-NEXT:    s_waitcnt lgkmcnt(0)
+; SDAG-GFX900-NEXT:    s_load_dword s4, s[0:1], 0x0
+; SDAG-GFX900-NEXT:    s_waitcnt lgkmcnt(0)
+; SDAG-GFX900-NEXT:    v_mov_b32_e32 v0, s4
+; SDAG-GFX900-NEXT:  BB0_1: ; %atomicrmw.start
+; SDAG-GFX900-NEXT:    ; =>This Inner Loop Header: Depth=1
+; SDAG-GFX900-NEXT:    v_mov_b32_e32 v1, v0
+; SDAG-GFX900-NEXT:    v_mov_b32_e32 v2, 0
+; SDAG-GFX900-NEXT:    v_add_f32_e32 v0, 4.0, v1
+; SDAG-GFX900-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
+; SDAG-GFX900-NEXT:    global_atomic_cmpswap v0, v2, v[0:1], s[0:1] glc
+; SDAG-GFX900-NEXT:    s_waitcnt vmcnt(0)
+; SDAG-GFX900-NEXT:    buffer_wbinvl1_vol
+; SDAG-GFX900-NEXT:    v_cmp_eq_u32_e32 vcc, v0, v1
+; SDAG-GFX900-NEXT:    s_or_b64 s[2:3], vcc, s[2:3]
+; SDAG-GFX900-NEXT:    s_andn2_b64 exec, exec, s[2:3]
+; SDAG-GFX900-NEXT:    s_cbranch_execnz BB0_1
+; SDAG-GFX900-NEXT:  ; %bb.2: ; %atomicrmw.end
+; SDAG-GFX900-NEXT:    s_or_b64 exec, exec, s[2:3]
+; SDAG-GFX900-NEXT:    global_store_dword v[0:1], v0, off
+; SDAG-GFX900-NEXT:    s_endpgm
 ;
-; GFX908-LABEL: global_atomic_fadd_ret_f32:
-; GFX908:       ; %bb.0:
-; GFX908-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x24
-; GFX908-NEXT:    s_mov_b64 s[2:3], 0
-; GFX908-NEXT:    s_waitcnt lgkmcnt(0)
-; GFX908-NEXT:    s_load_dword s4, s[0:1], 0x0
-; GFX908-NEXT:    s_waitcnt lgkmcnt(0)
-; GFX908-NEXT:    v_mov_b32_e32 v0, s4
-; GFX908-NEXT:  BB0_1: ; %atomicrmw.start
-; GFX908-NEXT:    ; =>This Inner Loop Header: Depth=1
-; GFX908-NEXT:    v_mov_b32_e32 v1, v0
-; GFX908-NEXT:    v_mov_b32_e32 v2, 0
-; GFX908-NEXT:    v_add_f32_e32 v0, 4.0, v1
-; GFX908-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
-; GFX908-NEXT:    global_atomic_cmpswap v0, v2, v[0:1], s[0:1] glc
-; GFX908-NEXT:    s_waitcnt vmcnt(0)
-; GFX908-NEXT:    buffer_wbinvl1_vol
-; GFX908-NEXT:    v_cmp_eq_u32_e32 vcc, v0, v1
-; GFX908-NEXT:    s_or_b64 s[2:3], vcc, s[2:3]
-; GFX908-NEXT:    s_andn2_b64 exec, exec, s[2:3]
-; GFX908-NEXT:    s_cbranch_execnz BB0_1
-; GFX908-NEXT:  ; %bb.2: ; %atomicrmw.end
-; GFX908-NEXT:    s_or_b64 exec, exec, s[2:3]
-; GFX908-NEXT:    global_store_dword v[0:1], v0, off
-; GFX908-NEXT:    s_endpgm
+; SDAG-GFX908-LABEL: global_atomic_fadd_ret_f32:
+; SDAG-GFX908:       ; %bb.0:
+; SDAG-GFX908-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x24
+; SDAG-GFX908-NEXT:    s_mov_b64 s[2:3], 0
+; SDAG-GFX908-NEXT:    s_waitcnt lgkmcnt(0)
+; SDAG-GFX908-NEXT:    s_load_dword s4, s[0:1], 0x0
+; SDAG-GFX908-NEXT:    s_waitcnt lgkmcnt(0)
+; SDAG-GFX908-NEXT:    v_mov_b32_e32 v0, s4
+; SDAG-GFX908-NEXT:  BB0_1: ; %atomicrmw.start
+; SDAG-GFX908-NEXT:    ; =>This Inner Loop Header: Depth=1
+; SDAG-GFX908-NEXT:    v_mov_b32_e32 v1, v0
+; SDAG-GFX908-NEXT:    v_mov_b32_e32 v2, 0
+; SDAG-GFX908-NEXT:    v_add_f32_e32 v0, 4.0, v1
+; SDAG-GFX908-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
+; SDAG-GFX908-NEXT:    global_atomic_cmpswap v0, v2, v[0:1], s[0:1] glc
+; SDAG-GFX908-NEXT:    s_waitcnt vmcnt(0)
+; SDAG-GFX908-NEXT:    buffer_wbinvl1_vol
+; SDAG-GFX908-NEXT:    v_cmp_eq_u32_e32 vcc, v0, v1
+; SDAG-GFX908-NEXT:    s_or_b64 s[2:3], vcc, s[2:3]
+; SDAG-GFX908-NEXT:    s_andn2_b64 exec, exec, s[2:3]
+; SDAG-GFX908-NEXT:    s_cbranch_execnz BB0_1
+; SDAG-GFX908-NEXT:  ; %bb.2: ; %atomicrmw.end
+; SDAG-GFX908-NEXT:    s_or_b64 exec, exec, s[2:3]
+; SDAG-GFX908-NEXT:    global_store_dword v[0:1], v0, off
+; SDAG-GFX908-NEXT:    s_endpgm
 ;
-; GFX90A-LABEL: global_atomic_fadd_ret_f32:
-; GFX90A:       ; %bb.0:
-; GFX90A-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x24
-; GFX90A-NEXT:    v_mov_b32_e32 v0, 0
-; GFX90A-NEXT:    v_mov_b32_e32 v1, 4.0
-; GFX90A-NEXT:    buffer_wbl2
-; GFX90A-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
-; GFX90A-NEXT:    global_atomic_add_f32 v0, v0, v1, s[0:1] glc scc
-; GFX90A-NEXT:    s_waitcnt vmcnt(0)
-; GFX90A-NEXT:    buffer_invl2
-; GFX90A-NEXT:    s_waitcnt vmcnt(0)
-; GFX90A-NEXT:    buffer_wbinvl1_vol
-; GFX90A-NEXT:    global_store_dword v[0:1], v0, off
-; GFX90A-NEXT:    s_endpgm
+; SDAG-GFX90A-LABEL: global_atomic_fadd_ret_f32:
+; SDAG-GFX90A:       ; %bb.0:
+; SDAG-GFX90A-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x24
+; SDAG-GFX90A-NEXT:    v_mov_b32_e32 v0, 0
+; SDAG-GFX90A-NEXT:    v_mov_b32_e32 v1, 4.0
+; SDAG-GFX90A-NEXT:    buffer_wbl2
+; SDAG-GFX90A-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
+; SDAG-GFX90A-NEXT:    global_atomic_add_f32 v0, v0, v1, s[0:1] glc scc
+; SDAG-GFX90A-NEXT:    s_waitcnt vmcnt(0)
+; SDAG-GFX90A-NEXT:    buffer_invl2
+; SDAG-GFX90A-NEXT:    s_waitcnt vmcnt(0)
+; SDAG-GFX90A-NEXT:    buffer_wbinvl1_vol
+; SDAG-GFX90A-NEXT:    global_store_dword v[0:1], v0, off
+; SDAG-GFX90A-NEXT:    s_endpgm
+;
+; GISEL-GFX900-LABEL: global_atomic_fadd_ret_f32:
+; GISEL-GFX900:       ; %bb.0:
+; GISEL-GFX900-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x24
+; GISEL-GFX900-NEXT:    s_mov_b64 s[2:3], 0
+; GISEL-GFX900-NEXT:    s_waitcnt lgkmcnt(0)
+; GISEL-GFX900-NEXT:    s_load_dword s4, s[0:1], 0x0
+; GISEL-GFX900-NEXT:    s_waitcnt lgkmcnt(0)
+; GISEL-GFX900-NEXT:    v_mov_b32_e32 v0, s4
+; GISEL-GFX900-NEXT:  BB0_1: ; %atomicrmw.start
+; GISEL-GFX900-NEXT:    ; =>This Inner Loop Header: Depth=1
+; GISEL-GFX900-NEXT:    v_mov_b32_e32 v1, v0
+; GISEL-GFX900-NEXT:    v_mov_b32_e32 v2, 0
+; GISEL-GFX900-NEXT:    v_add_f32_e32 v0, 4.0, v1
+; GISEL-GFX900-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
+; GISEL-GFX900-NEXT:    global_atomic_cmpswap v0, v2, v[0:1], s[0:1] glc
+; GISEL-GFX900-NEXT:    s_waitcnt vmcnt(0)
+; GISEL-GFX900-NEXT:    buffer_wbinvl1_vol
+; GISEL-GFX900-NEXT:    v_cmp_eq_u32_e32 vcc, v0, v1
+; GISEL-GFX900-NEXT:    s_or_b64 s[2:3], vcc, s[2:3]
+; GISEL-GFX900-NEXT:    s_andn2_b64 exec, exec, s[2:3]
+; GISEL-GFX900-NEXT:    s_cbranch_execnz BB0_1
+; GISEL-GFX900-NEXT:  ; %bb.2: ; %atomicrmw.end
+; GISEL-GFX900-NEXT:    s_or_b64 exec, exec, s[2:3]
+; GISEL-GFX900-NEXT:    global_store_dword v[0:1], v0, off
+; GISEL-GFX900-NEXT:    s_endpgm
+;
+; GISEL-GFX908-LABEL: global_atomic_fadd_ret_f32:
+; GISEL-GFX908:       ; %bb.0:
+; GISEL-GFX908-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x24
+; GISEL-GFX908-NEXT:    s_mov_b64 s[2:3], 0
+; GISEL-GFX908-NEXT:    s_waitcnt lgkmcnt(0)
+; GISEL-GFX908-NEXT:    s_load_dword s4, s[0:1], 0x0
+; GISEL-GFX908-NEXT:    s_waitcnt lgkmcnt(0)
+; GISEL-GFX908-NEXT:    v_mov_b32_e32 v0, s4
+; GISEL-GFX908-NEXT:  BB0_1: ; %atomicrmw.start
+; GISEL-GFX908-NEXT:    ; =>This Inner Loop Header: Depth=1
+; GISEL-GFX908-NEXT:    v_mov_b32_e32 v1, v0
+; GISEL-GFX908-NEXT:    v_mov_b32_e32 v2, 0
+; GISEL-GFX908-NEXT:    v_add_f32_e32 v0, 4.0, v1
+; GISEL-GFX908-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
+; GISEL-GFX908-NEXT:    global_atomic_cmpswap v0, v2, v[0:1], s[0:1] glc
+; GISEL-GFX908-NEXT:    s_waitcnt vmcnt(0)
+; GISEL-GFX908-NEXT:    buffer_wbinvl1_vol
+; GISEL-GFX908-NEXT:    v_cmp_eq_u32_e32 vcc, v0, v1
+; GISEL-GFX908-NEXT:    s_or_b64 s[2:3], vcc, s[2:3]
+; GISEL-GFX908-NEXT:    s_andn2_b64 exec, exec, s[2:3]
+; GISEL-GFX908-NEXT:    s_cbranch_execnz BB0_1
+; GISEL-GFX908-NEXT:  ; %bb.2: ; %atomicrmw.end
+; GISEL-GFX908-NEXT:    s_or_b64 exec, exec, s[2:3]
+; GISEL-GFX908-NEXT:    global_store_dword v[0:1], v0, off
+; GISEL-GFX908-NEXT:    s_endpgm
+;
+; GISEL-GFX90A-LABEL: global_atomic_fadd_ret_f32:
+; GISEL-GFX90A:       ; %bb.0:
+; GISEL-GFX90A-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x24
+; GISEL-GFX90A-NEXT:    v_mov_b32_e32 v0, 4.0
+; GISEL-GFX90A-NEXT:    v_mov_b32_e32 v1, 0
+; GISEL-GFX90A-NEXT:    buffer_wbl2
+; GISEL-GFX90A-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
+; GISEL-GFX90A-NEXT:    global_atomic_add_f32 v0, v1, v0, s[0:1] glc scc
+; GISEL-GFX90A-NEXT:    s_waitcnt vmcnt(0)
+; GISEL-GFX90A-NEXT:    buffer_invl2
+; GISEL-GFX90A-NEXT:    s_waitcnt vmcnt(0)
+; GISEL-GFX90A-NEXT:    buffer_wbinvl1_vol
+; GISEL-GFX90A-NEXT:    global_store_dword v[0:1], v0, off
+; GISEL-GFX90A-NEXT:    s_endpgm
   %result = atomicrmw fadd float addrspace(1)* %ptr, float 4.0 seq_cst
   store float %result, float addrspace(1)* undef
   ret void
 }
 
 define amdgpu_kernel void @global_atomic_fadd_ret_f32_ieee(float addrspace(1)* %ptr) #2 {
-; GFX900-LABEL: global_atomic_fadd_ret_f32_ieee:
-; GFX900:       ; %bb.0:
-; GFX900-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x24
-; GFX900-NEXT:    s_mov_b64 s[2:3], 0
-; GFX900-NEXT:    s_waitcnt lgkmcnt(0)
-; GFX900-NEXT:    s_load_dword s4, s[0:1], 0x0
-; GFX900-NEXT:    s_waitcnt lgkmcnt(0)
-; GFX900-NEXT:    v_mov_b32_e32 v0, s4
-; GFX900-NEXT:  BB1_1: ; %atomicrmw.start
-; GFX900-NEXT:    ; =>This Inner Loop Header: Depth=1
-; GFX900-NEXT:    v_mov_b32_e32 v1, v0
-; GFX900-NEXT:    v_mov_b32_e32 v2, 0
-; GFX900-NEXT:    v_add_f32_e32 v0, 4.0, v1
-; GFX900-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
-; GFX900-NEXT:    global_atomic_cmpswap v0, v2, v[0:1], s[0:1] glc
-; GFX900-NEXT:    s_waitcnt vmcnt(0)
-; GFX900-NEXT:    buffer_wbinvl1_vol
-; GFX900-NEXT:    v_cmp_eq_u32_e32 vcc, v0, v1
-; GFX900-NEXT:    s_or_b64 s[2:3], vcc, s[2:3]
-; GFX900-NEXT:    s_andn2_b64 exec, exec, s[2:3]
-; GFX900-NEXT:    s_cbranch_execnz BB1_1
-; GFX900-NEXT:  ; %bb.2: ; %atomicrmw.end
-; GFX900-NEXT:    s_or_b64 exec, exec, s[2:3]
-; GFX900-NEXT:    global_store_dword v[0:1], v0, off
-; GFX900-NEXT:    s_endpgm
+; SDAG-GFX900-LABEL: global_atomic_fadd_ret_f32_ieee:
+; SDAG-GFX900:       ; %bb.0:
+; SDAG-GFX900-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x24
+; SDAG-GFX900-NEXT:    s_mov_b64 s[2:3], 0
+; SDAG-GFX900-NEXT:    s_waitcnt lgkmcnt(0)
+; SDAG-GFX900-NEXT:    s_load_dword s4, s[0:1], 0x0
+; SDAG-GFX900-NEXT:    s_waitcnt lgkmcnt(0)
+; SDAG-GFX900-NEXT:    v_mov_b32_e32 v0, s4
+; SDAG-GFX900-NEXT:  BB1_1: ; %atomicrmw.start
+; SDAG-GFX900-NEXT:    ; =>This Inner Loop Header: Depth=1
+; SDAG-GFX900-NEXT:    v_mov_b32_e32 v1, v0
+; SDAG-GFX900-NEXT:    v_mov_b32_e32 v2, 0
+; SDAG-GFX900-NEXT:    v_add_f32_e32 v0, 4.0, v1
+; SDAG-GFX900-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
+; SDAG-GFX900-NEXT:    global_atomic_cmpswap v0, v2, v[0:1], s[0:1] glc
+; SDAG-GFX900-NEXT:    s_waitcnt vmcnt(0)
+; SDAG-GFX900-NEXT:    buffer_wbinvl1_vol
+; SDAG-GFX900-NEXT:    v_cmp_eq_u32_e32 vcc, v0, v1
+; SDAG-GFX900-NEXT:    s_or_b64 s[2:3], vcc, s[2:3]
+; SDAG-GFX900-NEXT:    s_andn2_b64 exec, exec, s[2:3]
+; SDAG-GFX900-NEXT:    s_cbranch_execnz BB1_1
+; SDAG-GFX900-NEXT:  ; %bb.2: ; %atomicrmw.end
+; SDAG-GFX900-NEXT:    s_or_b64 exec, exec, s[2:3]
+; SDAG-GFX900-NEXT:    global_store_dword v[0:1], v0, off
+; SDAG-GFX900-NEXT:    s_endpgm
+;
+; SDAG-GFX908-LABEL: global_atomic_fadd_ret_f32_ieee:
+; SDAG-GFX908:       ; %bb.0:
+; SDAG-GFX908-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x24
+; SDAG-GFX908-NEXT:    s_mov_b64 s[2:3], 0
+; SDAG-GFX908-NEXT:    s_waitcnt lgkmcnt(0)
+; SDAG-GFX908-NEXT:    s_load_dword s4, s[0:1], 0x0
+; SDAG-GFX908-NEXT:    s_waitcnt lgkmcnt(0)
+; SDAG-GFX908-NEXT:    v_mov_b32_e32 v0, s4
+; SDAG-GFX908-NEXT:  BB1_1: ; %atomicrmw.start
+; SDAG-GFX908-NEXT:    ; =>This Inner Loop Header: Depth=1
+; SDAG-GFX908-NEXT:    v_mov_b32_e32 v1, v0
+; SDAG-GFX908-NEXT:    v_mov_b32_e32 v2, 0
+; SDAG-GFX908-NEXT:    v_add_f32_e32 v0, 4.0, v1
+; SDAG-GFX908-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
+; SDAG-GFX908-NEXT:    global_atomic_cmpswap v0, v2, v[0:1], s[0:1] glc
+; SDAG-GFX908-NEXT:    s_waitcnt vmcnt(0)
+; SDAG-GFX908-NEXT:    buffer_wbinvl1_vol
+; SDAG-GFX908-NEXT:    v_cmp_eq_u32_e32 vcc, v0, v1
+; SDAG-GFX908-NEXT:    s_or_b64 s[2:3], vcc, s[2:3]
+; SDAG-GFX908-NEXT:    s_andn2_b64 exec, exec, s[2:3]
+; SDAG-GFX908-NEXT:    s_cbranch_execnz BB1_1
+; SDAG-GFX908-NEXT:  ; %bb.2: ; %atomicrmw.end
+; SDAG-GFX908-NEXT:    s_or_b64 exec, exec, s[2:3]
+; SDAG-GFX908-NEXT:    global_store_dword v[0:1], v0, off
+; SDAG-GFX908-NEXT:    s_endpgm
 ;
-; GFX908-LABEL: global_atomic_fadd_ret_f32_ieee:
-; GFX908:       ; %bb.0:
-; GFX908-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x24
-; GFX908-NEXT:    s_mov_b64 s[2:3], 0
-; GFX908-NEXT:    s_waitcnt lgkmcnt(0)
-; GFX908-NEXT:    s_load_dword s4, s[0:1], 0x0
-; GFX908-NEXT:    s_waitcnt lgkmcnt(0)
-; GFX908-NEXT:    v_mov_b32_e32 v0, s4
-; GFX908-NEXT:  BB1_1: ; %atomicrmw.start
-; GFX908-NEXT:    ; =>This Inner Loop Header: Depth=1
-; GFX908-NEXT:    v_mov_b32_e32 v1, v0
-; GFX908-NEXT:    v_mov_b32_e32 v2, 0
-; GFX908-NEXT:    v_add_f32_e32 v0, 4.0, v1
-; GFX908-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
-; GFX908-NEXT:    global_atomic_cmpswap v0, v2, v[0:1], s[0:1] glc
-; GFX908-NEXT:    s_waitcnt vmcnt(0)
-; GFX908-NEXT:    buffer_wbinvl1_vol
-; GFX908-NEXT:    v_cmp_eq_u32_e32 vcc, v0, v1
-; GFX908-NEXT:    s_or_b64 s[2:3], vcc, s[2:3]
-; GFX908-NEXT:    s_andn2_b64 exec, exec, s[2:3]
-; GFX908-NEXT:    s_cbranch_execnz BB1_1
-; GFX908-NEXT:  ; %bb.2: ; %atomicrmw.end
-; GFX908-NEXT:    s_or_b64 exec, exec, s[2:3]
-; GFX908-NEXT:    global_store_dword v[0:1], v0, off
-; GFX908-NEXT:    s_endpgm
+; SDAG-GFX90A-LABEL: global_atomic_fadd_ret_f32_ieee:
+; SDAG-GFX90A:       ; %bb.0:
+; SDAG-GFX90A-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x24
+; SDAG-GFX90A-NEXT:    s_mov_b64 s[2:3], 0
+; SDAG-GFX90A-NEXT:    s_waitcnt lgkmcnt(0)
+; SDAG-GFX90A-NEXT:    s_load_dword s4, s[0:1], 0x0
+; SDAG-GFX90A-NEXT:    s_waitcnt lgkmcnt(0)
+; SDAG-GFX90A-NEXT:    v_mov_b32_e32 v0, s4
+; SDAG-GFX90A-NEXT:  BB1_1: ; %atomicrmw.start
+; SDAG-GFX90A-NEXT:    ; =>This Inner Loop Header: Depth=1
+; SDAG-GFX90A-NEXT:    v_mov_b32_e32 v1, v0
+; SDAG-GFX90A-NEXT:    v_mov_b32_e32 v2, 0
+; SDAG-GFX90A-NEXT:    v_add_f32_e32 v0, 4.0, v1
+; SDAG-GFX90A-NEXT:    buffer_wbl2
+; SDAG-GFX90A-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
+; SDAG-GFX90A-NEXT:    global_atomic_cmpswap v0, v2, v[0:1], s[0:1] glc scc
+; SDAG-GFX90A-NEXT:    s_waitcnt vmcnt(0)
+; SDAG-GFX90A-NEXT:    buffer_invl2
+; SDAG-GFX90A-NEXT:    s_waitcnt vmcnt(0)
+; SDAG-GFX90A-NEXT:    buffer_wbinvl1_vol
+; SDAG-GFX90A-NEXT:    v_cmp_eq_u32_e32 vcc, v0, v1
+; SDAG-GFX90A-NEXT:    s_or_b64 s[2:3], vcc, s[2:3]
+; SDAG-GFX90A-NEXT:    s_andn2_b64 exec, exec, s[2:3]
+; SDAG-GFX90A-NEXT:    s_cbranch_execnz BB1_1
+; SDAG-GFX90A-NEXT:  ; %bb.2: ; %atomicrmw.end
+; SDAG-GFX90A-NEXT:    s_or_b64 exec, exec, s[2:3]
+; SDAG-GFX90A-NEXT:    global_store_dword v[0:1], v0, off
+; SDAG-GFX90A-NEXT:    s_endpgm
 ;
-; GFX90A-LABEL: global_atomic_fadd_ret_f32_ieee:
-; GFX90A:       ; %bb.0:
-; GFX90A-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x24
-; GFX90A-NEXT:    s_mov_b64 s[2:3], 0
-; GFX90A-NEXT:    s_waitcnt lgkmcnt(0)
-; GFX90A-NEXT:    s_load_dword s4, s[0:1], 0x0
-; GFX90A-NEXT:    s_waitcnt lgkmcnt(0)
-; GFX90A-NEXT:    v_mov_b32_e32 v0, s4
-; GFX90A-NEXT:  BB1_1: ; %atomicrmw.start
-; GFX90A-NEXT:    ; =>This Inner Loop Header: Depth=1
-; GFX90A-NEXT:    v_mov_b32_e32 v1, v0
-; GFX90A-NEXT:    v_mov_b32_e32 v2, 0
-; GFX90A-NEXT:    v_add_f32_e32 v0, 4.0, v1
-; GFX90A-NEXT:    buffer_wbl2
-; GFX90A-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
-; GFX90A-NEXT:    global_atomic_cmpswap v0, v2, v[0:1], s[0:1] glc scc
-; GFX90A-NEXT:    s_waitcnt vmcnt(0)
-; GFX90A-NEXT:    buffer_invl2
-; GFX90A-NEXT:    s_waitcnt vmcnt(0)
-; GFX90A-NEXT:    buffer_wbinvl1_vol
-; GFX90A-NEXT:    v_cmp_eq_u32_e32 vcc, v0, v1
-; GFX90A-NEXT:    s_or_b64 s[2:3], vcc, s[2:3]
-; GFX90A-NEXT:    s_andn2_b64 exec, exec, s[2:3]
-; GFX90A-NEXT:    s_cbranch_execnz BB1_1
-; GFX90A-NEXT:  ; %bb.2: ; %atomicrmw.end
-; GFX90A-NEXT:    s_or_b64 exec, exec, s[2:3]
-; GFX90A-NEXT:    global_store_dword v[0:1], v0, off
-; GFX90A-NEXT:    s_endpgm
+; GISEL-GFX900-LABEL: global_atomic_fadd_ret_f32_ieee:
+; GISEL-GFX900:       ; %bb.0:
+; GISEL-GFX900-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x24
+; GISEL-GFX900-NEXT:    s_mov_b64 s[2:3], 0
+; GISEL-GFX900-NEXT:    s_waitcnt lgkmcnt(0)
+; GISEL-GFX900-NEXT:    s_load_dword s4, s[0:1], 0x0
+; GISEL-GFX900-NEXT:    s_waitcnt lgkmcnt(0)
+; GISEL-GFX900-NEXT:    v_mov_b32_e32 v0, s4
+; GISEL-GFX900-NEXT:  BB1_1: ; %atomicrmw.start
+; GISEL-GFX900-NEXT:    ; =>This Inner Loop Header: Depth=1
+; GISEL-GFX900-NEXT:    v_mov_b32_e32 v1, v0
+; GISEL-GFX900-NEXT:    v_mov_b32_e32 v2, 0
+; GISEL-GFX900-NEXT:    v_add_f32_e32 v0, 4.0, v1
+; GISEL-GFX900-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
+; GISEL-GFX900-NEXT:    global_atomic_cmpswap v0, v2, v[0:1], s[0:1] glc
+; GISEL-GFX900-NEXT:    s_waitcnt vmcnt(0)
+; GISEL-GFX900-NEXT:    buffer_wbinvl1_vol
+; GISEL-GFX900-NEXT:    v_cmp_eq_u32_e32 vcc, v0, v1
+; GISEL-GFX900-NEXT:    s_or_b64 s[2:3], vcc, s[2:3]
+; GISEL-GFX900-NEXT:    s_andn2_b64 exec, exec, s[2:3]
+; GISEL-GFX900-NEXT:    s_cbranch_execnz BB1_1
+; GISEL-GFX900-NEXT:  ; %bb.2: ; %atomicrmw.end
+; GISEL-GFX900-NEXT:    s_or_b64 exec, exec, s[2:3]
+; GISEL-GFX900-NEXT:    global_store_dword v[0:1], v0, off
+; GISEL-GFX900-NEXT:    s_endpgm
+;
+; GISEL-GFX908-LABEL: global_atomic_fadd_ret_f32_ieee:
+; GISEL-GFX908:       ; %bb.0:
+; GISEL-GFX908-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x24
+; GISEL-GFX908-NEXT:    s_mov_b64 s[2:3], 0
+; GISEL-GFX908-NEXT:    s_waitcnt lgkmcnt(0)
+; GISEL-GFX908-NEXT:    s_load_dword s4, s[0:1], 0x0
+; GISEL-GFX908-NEXT:    s_waitcnt lgkmcnt(0)
+; GISEL-GFX908-NEXT:    v_mov_b32_e32 v0, s4
+; GISEL-GFX908-NEXT:  BB1_1: ; %atomicrmw.start
+; GISEL-GFX908-NEXT:    ; =>This Inner Loop Header: Depth=1
+; GISEL-GFX908-NEXT:    v_mov_b32_e32 v1, v0
+; GISEL-GFX908-NEXT:    v_mov_b32_e32 v2, 0
+; GISEL-GFX908-NEXT:    v_add_f32_e32 v0, 4.0, v1
+; GISEL-GFX908-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
+; GISEL-GFX908-NEXT:    global_atomic_cmpswap v0, v2, v[0:1], s[0:1] glc
+; GISEL-GFX908-NEXT:    s_waitcnt vmcnt(0)
+; GISEL-GFX908-NEXT:    buffer_wbinvl1_vol
+; GISEL-GFX908-NEXT:    v_cmp_eq_u32_e32 vcc, v0, v1
+; GISEL-GFX908-NEXT:    s_or_b64 s[2:3], vcc, s[2:3]
+; GISEL-GFX908-NEXT:    s_andn2_b64 exec, exec, s[2:3]
+; GISEL-GFX908-NEXT:    s_cbranch_execnz BB1_1
+; GISEL-GFX908-NEXT:  ; %bb.2: ; %atomicrmw.end
+; GISEL-GFX908-NEXT:    s_or_b64 exec, exec, s[2:3]
+; GISEL-GFX908-NEXT:    global_store_dword v[0:1], v0, off
+; GISEL-GFX908-NEXT:    s_endpgm
+;
+; GISEL-GFX90A-LABEL: global_atomic_fadd_ret_f32_ieee:
+; GISEL-GFX90A:       ; %bb.0:
+; GISEL-GFX90A-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x24
+; GISEL-GFX90A-NEXT:    s_mov_b64 s[2:3], 0
+; GISEL-GFX90A-NEXT:    s_waitcnt lgkmcnt(0)
+; GISEL-GFX90A-NEXT:    s_load_dword s4, s[0:1], 0x0
+; GISEL-GFX90A-NEXT:    s_waitcnt lgkmcnt(0)
+; GISEL-GFX90A-NEXT:    v_mov_b32_e32 v0, s4
+; GISEL-GFX90A-NEXT:  BB1_1: ; %atomicrmw.start
+; GISEL-GFX90A-NEXT:    ; =>This Inner Loop Header: Depth=1
+; GISEL-GFX90A-NEXT:    v_mov_b32_e32 v1, v0
+; GISEL-GFX90A-NEXT:    v_mov_b32_e32 v2, 0
+; GISEL-GFX90A-NEXT:    v_add_f32_e32 v0, 4.0, v1
+; GISEL-GFX90A-NEXT:    buffer_wbl2
+; GISEL-GFX90A-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
+; GISEL-GFX90A-NEXT:    global_atomic_cmpswap v0, v2, v[0:1], s[0:1] glc scc
+; GISEL-GFX90A-NEXT:    s_waitcnt vmcnt(0)
+; GISEL-GFX90A-NEXT:    buffer_invl2
+; GISEL-GFX90A-NEXT:    s_waitcnt vmcnt(0)
+; GISEL-GFX90A-NEXT:    buffer_wbinvl1_vol
+; GISEL-GFX90A-NEXT:    v_cmp_eq_u32_e32 vcc, v0, v1
+; GISEL-GFX90A-NEXT:    s_or_b64 s[2:3], vcc, s[2:3]
+; GISEL-GFX90A-NEXT:    s_andn2_b64 exec, exec, s[2:3]
+; GISEL-GFX90A-NEXT:    s_cbranch_execnz BB1_1
+; GISEL-GFX90A-NEXT:  ; %bb.2: ; %atomicrmw.end
+; GISEL-GFX90A-NEXT:    s_or_b64 exec, exec, s[2:3]
+; GISEL-GFX90A-NEXT:    global_store_dword v[0:1], v0, off
+; GISEL-GFX90A-NEXT:    s_endpgm
   %result = atomicrmw fadd float addrspace(1)* %ptr, float 4.0 seq_cst
   store float %result, float addrspace(1)* undef
   ret void
 }
 
 define amdgpu_kernel void @global_atomic_fadd_noret_f32(float addrspace(1)* %ptr) #0 {
-; GFX900-LABEL: global_atomic_fadd_noret_f32:
-; GFX900:       ; %bb.0:
-; GFX900-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x24
-; GFX900-NEXT:    s_mov_b64 s[2:3], 0
-; GFX900-NEXT:    s_waitcnt lgkmcnt(0)
-; GFX900-NEXT:    s_load_dword s4, s[0:1], 0x0
-; GFX900-NEXT:    s_waitcnt lgkmcnt(0)
-; GFX900-NEXT:    v_mov_b32_e32 v1, s4
-; GFX900-NEXT:  BB2_1: ; %atomicrmw.start
-; GFX900-NEXT:    ; =>This Inner Loop Header: Depth=1
-; GFX900-NEXT:    v_mov_b32_e32 v2, 0
-; GFX900-NEXT:    v_add_f32_e32 v0, 4.0, v1
-; GFX900-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
-; GFX900-NEXT:    global_atomic_cmpswap v0, v2, v[0:1], s[0:1] glc
-; GFX900-NEXT:    s_waitcnt vmcnt(0)
-; GFX900-NEXT:    buffer_wbinvl1_vol
-; GFX900-NEXT:    v_cmp_eq_u32_e32 vcc, v0, v1
-; GFX900-NEXT:    s_or_b64 s[2:3], vcc, s[2:3]
-; GFX900-NEXT:    v_mov_b32_e32 v1, v0
-; GFX900-NEXT:    s_andn2_b64 exec, exec, s[2:3]
-; GFX900-NEXT:    s_cbranch_execnz BB2_1
-; GFX900-NEXT:  ; %bb.2: ; %atomicrmw.end
-; GFX900-NEXT:    s_endpgm
+; SDAG-GFX900-LABEL: global_atomic_fadd_noret_f32:
+; SDAG-GFX900:       ; %bb.0:
+; SDAG-GFX900-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x24
+; SDAG-GFX900-NEXT:    s_mov_b64 s[2:3], 0
+; SDAG-GFX900-NEXT:    s_waitcnt lgkmcnt(0)
+; SDAG-GFX900-NEXT:    s_load_dword s4, s[0:1], 0x0
+; SDAG-GFX900-NEXT:    s_waitcnt lgkmcnt(0)
+; SDAG-GFX900-NEXT:    v_mov_b32_e32 v1, s4
+; SDAG-GFX900-NEXT:  BB2_1: ; %atomicrmw.start
+; SDAG-GFX900-NEXT:    ; =>This Inner Loop Header: Depth=1
+; SDAG-GFX900-NEXT:    v_mov_b32_e32 v2, 0
+; SDAG-GFX900-NEXT:    v_add_f32_e32 v0, 4.0, v1
+; SDAG-GFX900-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
+; SDAG-GFX900-NEXT:    global_atomic_cmpswap v0, v2, v[0:1], s[0:1] glc
+; SDAG-GFX900-NEXT:    s_waitcnt vmcnt(0)
+; SDAG-GFX900-NEXT:    buffer_wbinvl1_vol
+; SDAG-GFX900-NEXT:    v_cmp_eq_u32_e32 vcc, v0, v1
+; SDAG-GFX900-NEXT:    s_or_b64 s[2:3], vcc, s[2:3]
+; SDAG-GFX900-NEXT:    v_mov_b32_e32 v1, v0
+; SDAG-GFX900-NEXT:    s_andn2_b64 exec, exec, s[2:3]
+; SDAG-GFX900-NEXT:    s_cbranch_execnz BB2_1
+; SDAG-GFX900-NEXT:  ; %bb.2: ; %atomicrmw.end
+; SDAG-GFX900-NEXT:    s_endpgm
+;
+; SDAG-GFX908-LABEL: global_atomic_fadd_noret_f32:
+; SDAG-GFX908:       ; %bb.0:
+; SDAG-GFX908-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x24
+; SDAG-GFX908-NEXT:    v_mov_b32_e32 v0, 0
+; SDAG-GFX908-NEXT:    v_mov_b32_e32 v1, 4.0
+; SDAG-GFX908-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
+; SDAG-GFX908-NEXT:    global_atomic_add_f32 v0, v1, s[0:1]
+; SDAG-GFX908-NEXT:    s_waitcnt vmcnt(0)
+; SDAG-GFX908-NEXT:    buffer_wbinvl1_vol
+; SDAG-GFX908-NEXT:    s_endpgm
 ;
-; GFX908-LABEL: global_atomic_fadd_noret_f32:
-; GFX908:       ; %bb.0:
-; GFX908-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x24
-; GFX908-NEXT:    v_mov_b32_e32 v0, 0
-; GFX908-NEXT:    v_mov_b32_e32 v1, 4.0
-; GFX908-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
-; GFX908-NEXT:    global_atomic_add_f32 v0, v1, s[0:1]
-; GFX908-NEXT:    s_waitcnt vmcnt(0)
-; GFX908-NEXT:    buffer_wbinvl1_vol
-; GFX908-NEXT:    s_endpgm
+; SDAG-GFX90A-LABEL: global_atomic_fadd_noret_f32:
+; SDAG-GFX90A:       ; %bb.0:
+; SDAG-GFX90A-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x24
+; SDAG-GFX90A-NEXT:    v_mov_b32_e32 v0, 0
+; SDAG-GFX90A-NEXT:    v_mov_b32_e32 v1, 4.0
+; SDAG-GFX90A-NEXT:    buffer_wbl2
+; SDAG-GFX90A-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
+; SDAG-GFX90A-NEXT:    global_atomic_add_f32 v0, v1, s[0:1] scc
+; SDAG-GFX90A-NEXT:    s_waitcnt vmcnt(0)
+; SDAG-GFX90A-NEXT:    buffer_invl2
+; SDAG-GFX90A-NEXT:    s_waitcnt vmcnt(0)
+; SDAG-GFX90A-NEXT:    buffer_wbinvl1_vol
+; SDAG-GFX90A-NEXT:    s_endpgm
 ;
-; GFX90A-LABEL: global_atomic_fadd_noret_f32:
-; GFX90A:       ; %bb.0:
-; GFX90A-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x24
-; GFX90A-NEXT:    v_mov_b32_e32 v0, 0
-; GFX90A-NEXT:    v_mov_b32_e32 v1, 4.0
-; GFX90A-NEXT:    buffer_wbl2
-; GFX90A-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
-; GFX90A-NEXT:    global_atomic_add_f32 v0, v1, s[0:1] scc
-; GFX90A-NEXT:    s_waitcnt vmcnt(0)
-; GFX90A-NEXT:    buffer_invl2
-; GFX90A-NEXT:    s_waitcnt vmcnt(0)
-; GFX90A-NEXT:    buffer_wbinvl1_vol
-; GFX90A-NEXT:    s_endpgm
+; GISEL-GFX900-LABEL: global_atomic_fadd_noret_f32:
+; GISEL-GFX900:       ; %bb.0:
+; GISEL-GFX900-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x24
+; GISEL-GFX900-NEXT:    s_mov_b64 s[2:3], 0
+; GISEL-GFX900-NEXT:    s_waitcnt lgkmcnt(0)
+; GISEL-GFX900-NEXT:    s_load_dword s4, s[0:1], 0x0
+; GISEL-GFX900-NEXT:    s_waitcnt lgkmcnt(0)
+; GISEL-GFX900-NEXT:    v_mov_b32_e32 v1, s4
+; GISEL-GFX900-NEXT:  BB2_1: ; %atomicrmw.start
+; GISEL-GFX900-NEXT:    ; =>This Inner Loop Header: Depth=1
+; GISEL-GFX900-NEXT:    v_add_f32_e32 v0, 4.0, v1
+; GISEL-GFX900-NEXT:    v_mov_b32_e32 v2, 0
+; GISEL-GFX900-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
+; GISEL-GFX900-NEXT:    global_atomic_cmpswap v0, v2, v[0:1], s[0:1] glc
+; GISEL-GFX900-NEXT:    s_waitcnt vmcnt(0)
+; GISEL-GFX900-NEXT:    buffer_wbinvl1_vol
+; GISEL-GFX900-NEXT:    v_cmp_eq_u32_e32 vcc, v0, v1
+; GISEL-GFX900-NEXT:    s_or_b64 s[2:3], vcc, s[2:3]
+; GISEL-GFX900-NEXT:    v_mov_b32_e32 v1, v0
+; GISEL-GFX900-NEXT:    s_andn2_b64 exec, exec, s[2:3]
+; GISEL-GFX900-NEXT:    s_cbranch_execnz BB2_1
+; GISEL-GFX900-NEXT:  ; %bb.2: ; %atomicrmw.end
+; GISEL-GFX900-NEXT:    s_endpgm
+;
+; GISEL-GFX908-LABEL: global_atomic_fadd_noret_f32:
+; GISEL-GFX908:       ; %bb.0:
+; GISEL-GFX908-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x24
+; GISEL-GFX908-NEXT:    v_mov_b32_e32 v2, 4.0
+; GISEL-GFX908-NEXT:    s_waitcnt lgkmcnt(0)
+; GISEL-GFX908-NEXT:    v_mov_b32_e32 v0, s0
+; GISEL-GFX908-NEXT:    v_mov_b32_e32 v1, s1
+; GISEL-GFX908-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
+; GISEL-GFX908-NEXT:    global_atomic_add_f32 v[0:1], v2, off
+; GISEL-GFX908-NEXT:    s_waitcnt vmcnt(0)
+; GISEL-GFX908-NEXT:    buffer_wbinvl1_vol
+; GISEL-GFX908-NEXT:    s_endpgm
+;
+; GISEL-GFX90A-LABEL: global_atomic_fadd_noret_f32:
+; GISEL-GFX90A:       ; %bb.0:
+; GISEL-GFX90A-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x24
+; GISEL-GFX90A-NEXT:    v_mov_b32_e32 v0, 4.0
+; GISEL-GFX90A-NEXT:    v_mov_b32_e32 v1, 0
+; GISEL-GFX90A-NEXT:    buffer_wbl2
+; GISEL-GFX90A-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
+; GISEL-GFX90A-NEXT:    global_atomic_add_f32 v0, v1, v0, s[0:1] glc scc
+; GISEL-GFX90A-NEXT:    s_waitcnt vmcnt(0)
+; GISEL-GFX90A-NEXT:    buffer_invl2
+; GISEL-GFX90A-NEXT:    s_waitcnt vmcnt(0)
+; GISEL-GFX90A-NEXT:    buffer_wbinvl1_vol
+; GISEL-GFX90A-NEXT:    s_endpgm
   %result = atomicrmw fadd float addrspace(1)* %ptr, float 4.0 seq_cst
   ret void
 }
 
 define amdgpu_kernel void @global_atomic_fadd_noret_f32_ieee(float addrspace(1)* %ptr) #2 {
-; GFX900-LABEL: global_atomic_fadd_noret_f32_ieee:
-; GFX900:       ; %bb.0:
-; GFX900-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x24
-; GFX900-NEXT:    s_mov_b64 s[2:3], 0
-; GFX900-NEXT:    s_waitcnt lgkmcnt(0)
-; GFX900-NEXT:    s_load_dword s4, s[0:1], 0x0
-; GFX900-NEXT:    s_waitcnt lgkmcnt(0)
-; GFX900-NEXT:    v_mov_b32_e32 v1, s4
-; GFX900-NEXT:  BB3_1: ; %atomicrmw.start
-; GFX900-NEXT:    ; =>This Inner Loop Header: Depth=1
-; GFX900-NEXT:    v_mov_b32_e32 v2, 0
-; GFX900-NEXT:    v_add_f32_e32 v0, 4.0, v1
-; GFX900-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
-; GFX900-NEXT:    global_atomic_cmpswap v0, v2, v[0:1], s[0:1] glc
-; GFX900-NEXT:    s_waitcnt vmcnt(0)
-; GFX900-NEXT:    buffer_wbinvl1_vol
-; GFX900-NEXT:    v_cmp_eq_u32_e32 vcc, v0, v1
-; GFX900-NEXT:    s_or_b64 s[2:3], vcc, s[2:3]
-; GFX900-NEXT:    v_mov_b32_e32 v1, v0
-; GFX900-NEXT:    s_andn2_b64 exec, exec, s[2:3]
-; GFX900-NEXT:    s_cbranch_execnz BB3_1
-; GFX900-NEXT:  ; %bb.2: ; %atomicrmw.end
-; GFX900-NEXT:    s_endpgm
+; SDAG-GFX900-LABEL: global_atomic_fadd_noret_f32_ieee:
+; SDAG-GFX900:       ; %bb.0:
+; SDAG-GFX900-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x24
+; SDAG-GFX900-NEXT:    s_mov_b64 s[2:3], 0
+; SDAG-GFX900-NEXT:    s_waitcnt lgkmcnt(0)
+; SDAG-GFX900-NEXT:    s_load_dword s4, s[0:1], 0x0
+; SDAG-GFX900-NEXT:    s_waitcnt lgkmcnt(0)
+; SDAG-GFX900-NEXT:    v_mov_b32_e32 v1, s4
+; SDAG-GFX900-NEXT:  BB3_1: ; %atomicrmw.start
+; SDAG-GFX900-NEXT:    ; =>This Inner Loop Header: Depth=1
+; SDAG-GFX900-NEXT:    v_mov_b32_e32 v2, 0
+; SDAG-GFX900-NEXT:    v_add_f32_e32 v0, 4.0, v1
+; SDAG-GFX900-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
+; SDAG-GFX900-NEXT:    global_atomic_cmpswap v0, v2, v[0:1], s[0:1] glc
+; SDAG-GFX900-NEXT:    s_waitcnt vmcnt(0)
+; SDAG-GFX900-NEXT:    buffer_wbinvl1_vol
+; SDAG-GFX900-NEXT:    v_cmp_eq_u32_e32 vcc, v0, v1
+; SDAG-GFX900-NEXT:    s_or_b64 s[2:3], vcc, s[2:3]
+; SDAG-GFX900-NEXT:    v_mov_b32_e32 v1, v0
+; SDAG-GFX900-NEXT:    s_andn2_b64 exec, exec, s[2:3]
+; SDAG-GFX900-NEXT:    s_cbranch_execnz BB3_1
+; SDAG-GFX900-NEXT:  ; %bb.2: ; %atomicrmw.end
+; SDAG-GFX900-NEXT:    s_endpgm
+;
+; SDAG-GFX908-LABEL: global_atomic_fadd_noret_f32_ieee:
+; SDAG-GFX908:       ; %bb.0:
+; SDAG-GFX908-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x24
+; SDAG-GFX908-NEXT:    s_mov_b64 s[2:3], 0
+; SDAG-GFX908-NEXT:    s_waitcnt lgkmcnt(0)
+; SDAG-GFX908-NEXT:    s_load_dword s4, s[0:1], 0x0
+; SDAG-GFX908-NEXT:    s_waitcnt lgkmcnt(0)
+; SDAG-GFX908-NEXT:    v_mov_b32_e32 v1, s4
+; SDAG-GFX908-NEXT:  BB3_1: ; %atomicrmw.start
+; SDAG-GFX908-NEXT:    ; =>This Inner Loop Header: Depth=1
+; SDAG-GFX908-NEXT:    v_mov_b32_e32 v2, 0
+; SDAG-GFX908-NEXT:    v_add_f32_e32 v0, 4.0, v1
+; SDAG-GFX908-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
+; SDAG-GFX908-NEXT:    global_atomic_cmpswap v0, v2, v[0:1], s[0:1] glc
+; SDAG-GFX908-NEXT:    s_waitcnt vmcnt(0)
+; SDAG-GFX908-NEXT:    buffer_wbinvl1_vol
+; SDAG-GFX908-NEXT:    v_cmp_eq_u32_e32 vcc, v0, v1
+; SDAG-GFX908-NEXT:    s_or_b64 s[2:3], vcc, s[2:3]
+; SDAG-GFX908-NEXT:    v_mov_b32_e32 v1, v0
+; SDAG-GFX908-NEXT:    s_andn2_b64 exec, exec, s[2:3]
+; SDAG-GFX908-NEXT:    s_cbranch_execnz BB3_1
+; SDAG-GFX908-NEXT:  ; %bb.2: ; %atomicrmw.end
+; SDAG-GFX908-NEXT:    s_endpgm
+;
+; SDAG-GFX90A-LABEL: global_atomic_fadd_noret_f32_ieee:
+; SDAG-GFX90A:       ; %bb.0:
+; SDAG-GFX90A-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x24
+; SDAG-GFX90A-NEXT:    s_mov_b64 s[2:3], 0
+; SDAG-GFX90A-NEXT:    s_waitcnt lgkmcnt(0)
+; SDAG-GFX90A-NEXT:    s_load_dword s4, s[0:1], 0x0
+; SDAG-GFX90A-NEXT:    s_waitcnt lgkmcnt(0)
+; SDAG-GFX90A-NEXT:    v_mov_b32_e32 v1, s4
+; SDAG-GFX90A-NEXT:  BB3_1: ; %atomicrmw.start
+; SDAG-GFX90A-NEXT:    ; =>This Inner Loop Header: Depth=1
+; SDAG-GFX90A-NEXT:    v_mov_b32_e32 v2, 0
+; SDAG-GFX90A-NEXT:    v_add_f32_e32 v0, 4.0, v1
+; SDAG-GFX90A-NEXT:    buffer_wbl2
+; SDAG-GFX90A-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
+; SDAG-GFX90A-NEXT:    global_atomic_cmpswap v0, v2, v[0:1], s[0:1] glc scc
+; SDAG-GFX90A-NEXT:    s_waitcnt vmcnt(0)
+; SDAG-GFX90A-NEXT:    buffer_invl2
+; SDAG-GFX90A-NEXT:    s_waitcnt vmcnt(0)
+; SDAG-GFX90A-NEXT:    buffer_wbinvl1_vol
+; SDAG-GFX90A-NEXT:    v_cmp_eq_u32_e32 vcc, v0, v1
+; SDAG-GFX90A-NEXT:    s_or_b64 s[2:3], vcc, s[2:3]
+; SDAG-GFX90A-NEXT:    v_mov_b32_e32 v1, v0
+; SDAG-GFX90A-NEXT:    s_andn2_b64 exec, exec, s[2:3]
+; SDAG-GFX90A-NEXT:    s_cbranch_execnz BB3_1
+; SDAG-GFX90A-NEXT:  ; %bb.2: ; %atomicrmw.end
+; SDAG-GFX90A-NEXT:    s_endpgm
 ;
-; GFX908-LABEL: global_atomic_fadd_noret_f32_ieee:
-; GFX908:       ; %bb.0:
-; GFX908-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x24
-; GFX908-NEXT:    s_mov_b64 s[2:3], 0
-; GFX908-NEXT:    s_waitcnt lgkmcnt(0)
-; GFX908-NEXT:    s_load_dword s4, s[0:1], 0x0
-; GFX908-NEXT:    s_waitcnt lgkmcnt(0)
-; GFX908-NEXT:    v_mov_b32_e32 v1, s4
-; GFX908-NEXT:  BB3_1: ; %atomicrmw.start
-; GFX908-NEXT:    ; =>This Inner Loop Header: Depth=1
-; GFX908-NEXT:    v_mov_b32_e32 v2, 0
-; GFX908-NEXT:    v_add_f32_e32 v0, 4.0, v1
-; GFX908-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
-; GFX908-NEXT:    global_atomic_cmpswap v0, v2, v[0:1], s[0:1] glc
-; GFX908-NEXT:    s_waitcnt vmcnt(0)
-; GFX908-NEXT:    buffer_wbinvl1_vol
-; GFX908-NEXT:    v_cmp_eq_u32_e32 vcc, v0, v1
-; GFX908-NEXT:    s_or_b64 s[2:3], vcc, s[2:3]
-; GFX908-NEXT:    v_mov_b32_e32 v1, v0
-; GFX908-NEXT:    s_andn2_b64 exec, exec, s[2:3]
-; GFX908-NEXT:    s_cbranch_execnz BB3_1
-; GFX908-NEXT:  ; %bb.2: ; %atomicrmw.end
-; GFX908-NEXT:    s_endpgm
+; GISEL-GFX900-LABEL: global_atomic_fadd_noret_f32_ieee:
+; GISEL-GFX900:       ; %bb.0:
+; GISEL-GFX900-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x24
+; GISEL-GFX900-NEXT:    s_mov_b64 s[2:3], 0
+; GISEL-GFX900-NEXT:    s_waitcnt lgkmcnt(0)
+; GISEL-GFX900-NEXT:    s_load_dword s4, s[0:1], 0x0
+; GISEL-GFX900-NEXT:    s_waitcnt lgkmcnt(0)
+; GISEL-GFX900-NEXT:    v_mov_b32_e32 v1, s4
+; GISEL-GFX900-NEXT:  BB3_1: ; %atomicrmw.start
+; GISEL-GFX900-NEXT:    ; =>This Inner Loop Header: Depth=1
+; GISEL-GFX900-NEXT:    v_add_f32_e32 v0, 4.0, v1
+; GISEL-GFX900-NEXT:    v_mov_b32_e32 v2, 0
+; GISEL-GFX900-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
+; GISEL-GFX900-NEXT:    global_atomic_cmpswap v0, v2, v[0:1], s[0:1] glc
+; GISEL-GFX900-NEXT:    s_waitcnt vmcnt(0)
+; GISEL-GFX900-NEXT:    buffer_wbinvl1_vol
+; GISEL-GFX900-NEXT:    v_cmp_eq_u32_e32 vcc, v0, v1
+; GISEL-GFX900-NEXT:    s_or_b64 s[2:3], vcc, s[2:3]
+; GISEL-GFX900-NEXT:    v_mov_b32_e32 v1, v0
+; GISEL-GFX900-NEXT:    s_andn2_b64 exec, exec, s[2:3]
+; GISEL-GFX900-NEXT:    s_cbranch_execnz BB3_1
+; GISEL-GFX900-NEXT:  ; %bb.2: ; %atomicrmw.end
+; GISEL-GFX900-NEXT:    s_endpgm
 ;
-; GFX90A-LABEL: global_atomic_fadd_noret_f32_ieee:
-; GFX90A:       ; %bb.0:
-; GFX90A-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x24
-; GFX90A-NEXT:    s_mov_b64 s[2:3], 0
-; GFX90A-NEXT:    s_waitcnt lgkmcnt(0)
-; GFX90A-NEXT:    s_load_dword s4, s[0:1], 0x0
-; GFX90A-NEXT:    s_waitcnt lgkmcnt(0)
-; GFX90A-NEXT:    v_mov_b32_e32 v1, s4
-; GFX90A-NEXT:  BB3_1: ; %atomicrmw.start
-; GFX90A-NEXT:    ; =>This Inner Loop Header: Depth=1
-; GFX90A-NEXT:    v_mov_b32_e32 v2, 0
-; GFX90A-NEXT:    v_add_f32_e32 v0, 4.0, v1
-; GFX90A-NEXT:    buffer_wbl2
-; GFX90A-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
-; GFX90A-NEXT:    global_atomic_cmpswap v0, v2, v[0:1], s[0:1] glc scc
-; GFX90A-NEXT:    s_waitcnt vmcnt(0)
-; GFX90A-NEXT:    buffer_invl2
-; GFX90A-NEXT:    s_waitcnt vmcnt(0)
-; GFX90A-NEXT:    buffer_wbinvl1_vol
-; GFX90A-NEXT:    v_cmp_eq_u32_e32 vcc, v0, v1
-; GFX90A-NEXT:    s_or_b64 s[2:3], vcc, s[2:3]
-; GFX90A-NEXT:    v_mov_b32_e32 v1, v0
-; GFX90A-NEXT:    s_andn2_b64 exec, exec, s[2:3]
-; GFX90A-NEXT:    s_cbranch_execnz BB3_1
-; GFX90A-NEXT:  ; %bb.2: ; %atomicrmw.end
-; GFX90A-NEXT:    s_endpgm
+; GISEL-GFX908-LABEL: global_atomic_fadd_noret_f32_ieee:
+; GISEL-GFX908:       ; %bb.0:
+; GISEL-GFX908-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x24
+; GISEL-GFX908-NEXT:    s_mov_b64 s[2:3], 0
+; GISEL-GFX908-NEXT:    s_waitcnt lgkmcnt(0)
+; GISEL-GFX908-NEXT:    s_load_dword s4, s[0:1], 0x0
+; GISEL-GFX908-NEXT:    s_waitcnt lgkmcnt(0)
+; GISEL-GFX908-NEXT:    v_mov_b32_e32 v1, s4
+; GISEL-GFX908-NEXT:  BB3_1: ; %atomicrmw.start
+; GISEL-GFX908-NEXT:    ; =>This Inner Loop Header: Depth=1
+; GISEL-GFX908-NEXT:    v_add_f32_e32 v0, 4.0, v1
+; GISEL-GFX908-NEXT:    v_mov_b32_e32 v2, 0
+; GISEL-GFX908-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
+; GISEL-GFX908-NEXT:    global_atomic_cmpswap v0, v2, v[0:1], s[0:1] glc
+; GISEL-GFX908-NEXT:    s_waitcnt vmcnt(0)
+; GISEL-GFX908-NEXT:    buffer_wbinvl1_vol
+; GISEL-GFX908-NEXT:    v_cmp_eq_u32_e32 vcc, v0, v1
+; GISEL-GFX908-NEXT:    s_or_b64 s[2:3], vcc, s[2:3]
+; GISEL-GFX908-NEXT:    v_mov_b32_e32 v1, v0
+; GISEL-GFX908-NEXT:    s_andn2_b64 exec, exec, s[2:3]
+; GISEL-GFX908-NEXT:    s_cbranch_execnz BB3_1
+; GISEL-GFX908-NEXT:  ; %bb.2: ; %atomicrmw.end
+; GISEL-GFX908-NEXT:    s_endpgm
+;
+; GISEL-GFX90A-LABEL: global_atomic_fadd_noret_f32_ieee:
+; GISEL-GFX90A:       ; %bb.0:
+; GISEL-GFX90A-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x24
+; GISEL-GFX90A-NEXT:    s_mov_b64 s[2:3], 0
+; GISEL-GFX90A-NEXT:    s_waitcnt lgkmcnt(0)
+; GISEL-GFX90A-NEXT:    s_load_dword s4, s[0:1], 0x0
+; GISEL-GFX90A-NEXT:    s_waitcnt lgkmcnt(0)
+; GISEL-GFX90A-NEXT:    v_mov_b32_e32 v1, s4
+; GISEL-GFX90A-NEXT:  BB3_1: ; %atomicrmw.start
+; GISEL-GFX90A-NEXT:    ; =>This Inner Loop Header: Depth=1
+; GISEL-GFX90A-NEXT:    v_add_f32_e32 v0, 4.0, v1
+; GISEL-GFX90A-NEXT:    v_mov_b32_e32 v2, 0
+; GISEL-GFX90A-NEXT:    buffer_wbl2
+; GISEL-GFX90A-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
+; GISEL-GFX90A-NEXT:    global_atomic_cmpswap v0, v2, v[0:1], s[0:1] glc scc
+; GISEL-GFX90A-NEXT:    s_waitcnt vmcnt(0)
+; GISEL-GFX90A-NEXT:    buffer_invl2
+; GISEL-GFX90A-NEXT:    s_waitcnt vmcnt(0)
+; GISEL-GFX90A-NEXT:    buffer_wbinvl1_vol
+; GISEL-GFX90A-NEXT:    v_cmp_eq_u32_e32 vcc, v0, v1
+; GISEL-GFX90A-NEXT:    s_or_b64 s[2:3], vcc, s[2:3]
+; GISEL-GFX90A-NEXT:    v_mov_b32_e32 v1, v0
+; GISEL-GFX90A-NEXT:    s_andn2_b64 exec, exec, s[2:3]
+; GISEL-GFX90A-NEXT:    s_cbranch_execnz BB3_1
+; GISEL-GFX90A-NEXT:  ; %bb.2: ; %atomicrmw.end
+; GISEL-GFX90A-NEXT:    s_endpgm
   %result = atomicrmw fadd float addrspace(1)* %ptr, float 4.0 seq_cst
   ret void
 }
 
 define amdgpu_kernel void @global_atomic_fadd_ret_f32_wrong_subtarget(float addrspace(1)* %ptr) #1 {
-; GCN-LABEL: global_atomic_fadd_ret_f32_wrong_subtarget:
-; GCN:       ; %bb.0:
-; GCN-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x24
-; GCN-NEXT:    s_mov_b64 s[2:3], 0
-; GCN-NEXT:    s_waitcnt lgkmcnt(0)
-; GCN-NEXT:    s_load_dword s4, s[0:1], 0x0
-; GCN-NEXT:    s_waitcnt lgkmcnt(0)
-; GCN-NEXT:    v_mov_b32_e32 v0, s4
-; GCN-NEXT:  BB4_1: ; %atomicrmw.start
-; GCN-NEXT:    ; =>This Inner Loop Header: Depth=1
-; GCN-NEXT:    v_mov_b32_e32 v1, v0
-; GCN-NEXT:    v_mov_b32_e32 v2, 0
-; GCN-NEXT:    v_add_f32_e32 v0, 4.0, v1
-; GCN-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
-; GCN-NEXT:    global_atomic_cmpswap v0, v2, v[0:1], s[0:1] glc
-; GCN-NEXT:    s_waitcnt vmcnt(0)
-; GCN-NEXT:    buffer_wbinvl1_vol
-; GCN-NEXT:    v_cmp_eq_u32_e32 vcc, v0, v1
-; GCN-NEXT:    s_or_b64 s[2:3], vcc, s[2:3]
-; GCN-NEXT:    s_andn2_b64 exec, exec, s[2:3]
-; GCN-NEXT:    s_cbranch_execnz BB4_1
-; GCN-NEXT:  ; %bb.2: ; %atomicrmw.end
-; GCN-NEXT:    s_or_b64 exec, exec, s[2:3]
-; GCN-NEXT:    global_store_dword v[0:1], v0, off
-; GCN-NEXT:    s_endpgm
+; SDAG-LABEL: global_atomic_fadd_ret_f32_wrong_subtarget:
+; SDAG:       ; %bb.0:
+; SDAG-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x24
+; SDAG-NEXT:    s_mov_b64 s[2:3], 0
+; SDAG-NEXT:    s_waitcnt lgkmcnt(0)
+; SDAG-NEXT:    s_load_dword s4, s[0:1], 0x0
+; SDAG-NEXT:    s_waitcnt lgkmcnt(0)
+; SDAG-NEXT:    v_mov_b32_e32 v0, s4
+; SDAG-NEXT:  BB4_1: ; %atomicrmw.start
+; SDAG-NEXT:    ; =>This Inner Loop Header: Depth=1
+; SDAG-NEXT:    v_mov_b32_e32 v1, v0
+; SDAG-NEXT:    v_mov_b32_e32 v2, 0
+; SDAG-NEXT:    v_add_f32_e32 v0, 4.0, v1
+; SDAG-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
+; SDAG-NEXT:    global_atomic_cmpswap v0, v2, v[0:1], s[0:1] glc
+; SDAG-NEXT:    s_waitcnt vmcnt(0)
+; SDAG-NEXT:    buffer_wbinvl1_vol
+; SDAG-NEXT:    v_cmp_eq_u32_e32 vcc, v0, v1
+; SDAG-NEXT:    s_or_b64 s[2:3], vcc, s[2:3]
+; SDAG-NEXT:    s_andn2_b64 exec, exec, s[2:3]
+; SDAG-NEXT:    s_cbranch_execnz BB4_1
+; SDAG-NEXT:  ; %bb.2: ; %atomicrmw.end
+; SDAG-NEXT:    s_or_b64 exec, exec, s[2:3]
+; SDAG-NEXT:    global_store_dword v[0:1], v0, off
+; SDAG-NEXT:    s_endpgm
+;
+; GISEL-LABEL: global_atomic_fadd_ret_f32_wrong_subtarget:
+; GISEL:       ; %bb.0:
+; GISEL-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x24
+; GISEL-NEXT:    s_mov_b64 s[2:3], 0
+; GISEL-NEXT:    s_waitcnt lgkmcnt(0)
+; GISEL-NEXT:    s_load_dword s4, s[0:1], 0x0
+; GISEL-NEXT:    s_waitcnt lgkmcnt(0)
+; GISEL-NEXT:    v_mov_b32_e32 v0, s4
+; GISEL-NEXT:  BB4_1: ; %atomicrmw.start
+; GISEL-NEXT:    ; =>This Inner Loop Header: Depth=1
+; GISEL-NEXT:    v_mov_b32_e32 v1, v0
+; GISEL-NEXT:    v_mov_b32_e32 v2, 0
+; GISEL-NEXT:    v_add_f32_e32 v0, 4.0, v1
+; GISEL-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
+; GISEL-NEXT:    global_atomic_cmpswap v0, v2, v[0:1], s[0:1] glc
+; GISEL-NEXT:    s_waitcnt vmcnt(0)
+; GISEL-NEXT:    buffer_wbinvl1_vol
+; GISEL-NEXT:    v_cmp_eq_u32_e32 vcc, v0, v1
+; GISEL-NEXT:    s_or_b64 s[2:3], vcc, s[2:3]
+; GISEL-NEXT:    s_andn2_b64 exec, exec, s[2:3]
+; GISEL-NEXT:    s_cbranch_execnz BB4_1
+; GISEL-NEXT:  ; %bb.2: ; %atomicrmw.end
+; GISEL-NEXT:    s_or_b64 exec, exec, s[2:3]
+; GISEL-NEXT:    global_store_dword v[0:1], v0, off
+; GISEL-NEXT:    s_endpgm
   %result = atomicrmw fadd float addrspace(1)* %ptr, float 4.0 seq_cst
   store float %result, float addrspace(1)* undef
   ret void
 }
 
 define amdgpu_kernel void @global_atomic_fadd_noret_f32_wrong_subtarget(float addrspace(1)* %ptr) #1 {
-; GCN-LABEL: global_atomic_fadd_noret_f32_wrong_subtarget:
-; GCN:       ; %bb.0:
-; GCN-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x24
-; GCN-NEXT:    v_mov_b32_e32 v0, 0
-; GCN-NEXT:    v_mov_b32_e32 v1, 4.0
-; GCN-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
-; GCN-NEXT:    global_atomic_add_f32 v0, v1, s[0:1]
-; GCN-NEXT:    s_waitcnt vmcnt(0)
-; GCN-NEXT:    buffer_wbinvl1_vol
-; GCN-NEXT:    s_endpgm
+; SDAG-LABEL: global_atomic_fadd_noret_f32_wrong_subtarget:
+; SDAG:       ; %bb.0:
+; SDAG-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x24
+; SDAG-NEXT:    v_mov_b32_e32 v0, 0
+; SDAG-NEXT:    v_mov_b32_e32 v1, 4.0
+; SDAG-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
+; SDAG-NEXT:    global_atomic_add_f32 v0, v1, s[0:1]
+; SDAG-NEXT:    s_waitcnt vmcnt(0)
+; SDAG-NEXT:    buffer_wbinvl1_vol
+; SDAG-NEXT:    s_endpgm
+;
+; GISEL-LABEL: global_atomic_fadd_noret_f32_wrong_subtarget:
+; GISEL:       ; %bb.0:
+; GISEL-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x24
+; GISEL-NEXT:    v_mov_b32_e32 v2, 4.0
+; GISEL-NEXT:    s_waitcnt lgkmcnt(0)
+; GISEL-NEXT:    v_mov_b32_e32 v0, s0
+; GISEL-NEXT:    v_mov_b32_e32 v1, s1
+; GISEL-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
+; GISEL-NEXT:    global_atomic_add_f32 v[0:1], v2, off
+; GISEL-NEXT:    s_waitcnt vmcnt(0)
+; GISEL-NEXT:    buffer_wbinvl1_vol
+; GISEL-NEXT:    s_endpgm
   %result = atomicrmw fadd float addrspace(1)* %ptr, float 4.0 seq_cst
   ret void
 }