AMDGPU/SI: Add VI patterns to select FLAT instructions for global memory ops

[• tstellarAMD]

Summary

The MUBUF addr64 bit has been removed on VI, so we must use FLAT
instructions when the pointer is stored in VGPRs.

[arsenm]

LGTM

lib/Target/AMDGPU/SIISelLowering.cpp

271–283

This should go under all of the cases, since the code assumes MUBUF for all of these address spaces.

271–283

I'm not sure about using this only for global, but it sort of makes sense

test/CodeGen/AMDGPU/cgp-addressing-modes.ll

7–11

A test that shows an offset that is sunk on CI and not sunk on VI would be useful. I'm surprised that one of these existing test cases didn't break.

lib/Target/AMDGPU/AMDGPUISelDAGToDAG.cpp

 }
 
 bool AMDGPUDAGToDAGISel::SelectMUBUFAddr64(SDValue Addr, SDValue &SRsrc,
                                            SDValue &VAddr, SDValue &SOffset,
                                            SDValue &Offset, SDValue &GLC,
                                            SDValue &SLC, SDValue &TFE) const {
   SDValue Ptr, Offen, Idxen, Addr64;
 
+  // addr64 bit was removed for volcanic islands.
+  if (Subtarget->getGeneration() >= AMDGPUSubtarget::VOLCANIC_ISLANDS)
+    return false;
+
   SelectMUBUF(Addr, Ptr, VAddr, SOffset, Offset, Offen, Idxen, Addr64,
               GLC, SLC, TFE);
 
   ConstantSDNode *C = cast<ConstantSDNode>(Addr64);
   if (C->getSExtValue()) {
     SDLoc DL(Addr);
 
     const SITargetLowering& Lowering =

lib/Target/AMDGPU/SIISelLowering.h

SDValue performSHLPtrCombine(SDNode *N,
                                DAGCombinerInfo &DCI) const;
   SDValue performAndCombine(SDNode *N, DAGCombinerInfo &DCI) const;
   SDValue performOrCombine(SDNode *N, DAGCombinerInfo &DCI) const;
   SDValue performClassCombine(SDNode *N, DAGCombinerInfo &DCI) const;
 
   SDValue performMin3Max3Combine(SDNode *N, DAGCombinerInfo &DCI) const;
   SDValue performSetCCCombine(SDNode *N, DAGCombinerInfo &DCI) const;
 
+  bool isLegalFlatAddressingMode(const AddrMode &AM) const;
 public:
   SITargetLowering(TargetMachine &tm, const AMDGPUSubtarget &STI);
 
   bool isShuffleMaskLegal(const SmallVectorImpl<int> &/*Mask*/,
                           EVT /*VT*/) const override;
 
   bool isLegalAddressingMode(const AddrMode &AM,
                              Type *Ty, unsigned AS) const override;

lib/Target/AMDGPU/SIISelLowering.cpp

 
 bool SITargetLowering::isShuffleMaskLegal(const SmallVectorImpl<int> &,
                                           EVT) const {
   // SI has some legal vector types, but no legal vector operations. Say no
   // shuffles are legal in order to prefer scalarizing some vector operations.
   return false;
 }
 
+bool SITargetLowering::isLegalFlatAddressingMode(const AddrMode &AM) const {
+  // Flat instructions do not have offsets, and only have the register
+  // address.
+  return AM.BaseOffs == 0 && (AM.Scale == 0 || AM.Scale == 1);
+}
+
 bool SITargetLowering::isLegalAddressingMode(const AddrMode &AM,
                                              Type *Ty, unsigned AS) const {
   // No global is ever allowed as a base.
   if (AM.BaseGV)
     return false;
 
   switch (AS) {
   case AMDGPUAS::GLOBAL_ADDRESS:
-  case AMDGPUAS::CONSTANT_ADDRESS: // XXX - Should we assume SMRD instructions?
+    if (Subtarget->getGeneration() >= AMDGPUSubtarget::VOLCANIC_ISLANDS) {
+      // Assume the we will use FLAT for all global memory accesses
+      // on VI.
+      // FIXME: This assumption is currently wrong.  On VI we still use
+      // MUBUF instructions for the r + i addressing mode.  As currently
+      // implemented, the MUBUF instructions only work on buffer < 4GB.
+      // It may be possible to support > 4GB buffers with MUBUF instructions,
+      // by setting the stride value in the resource descriptor which would
+      // increase the size limit to (stride * 4GB).  However, this is risky,
+      // because it has never been validated.
+      return isLegalFlatAddressingMode(AM);
+    }
+    // fall-through
   case AMDGPUAS::PRIVATE_ADDRESS:
+  case AMDGPUAS::CONSTANT_ADDRESS: // XXX - Should we assume SMRD instructions?
   case AMDGPUAS::UNKNOWN_ADDRESS_SPACE: {
     // MUBUF / MTBUF instructions have a 12-bit unsigned byte offset, and
     // additionally can do r + r + i with addr64. 32-bit has more addressing
     // mode options. Depending on the resource constant, it can also do
     // (i64 r0) + (i32 r1) * (i14 i).
     //
     // SMRD instructions have an 8-bit, dword offset.
     //
case AMDGPUAS::REGION_ADDRESS: {
     if (AM.Scale == 0) // r + i or just i, depending on HasBaseReg.
       return true;
 
     if (AM.Scale == 1 && AM.HasBaseReg)
       return true;
 
     return false;
   }
-  case AMDGPUAS::FLAT_ADDRESS: {
-    // Flat instructions do not have offsets, and only have the register
-    // address.
-    return AM.BaseOffs == 0 && (AM.Scale == 0 || AM.Scale == 1);
-  }
+  case AMDGPUAS::FLAT_ADDRESS:
+    return isLegalFlatAddressingMode(AM);
+
   default:
     llvm_unreachable("unhandled address space");
   }
 }
 
 bool SITargetLowering::allowsMisalignedMemoryAccesses(EVT VT,
                                                       unsigned AddrSpace,
                                                       unsigned Align,

lib/Target/AMDGPU/VIInstructions.td

 let Predicates = [isVI] in {
 
 // 1. Offset as 20bit DWORD immediate
 def : Pat <
   (SIload_constant v4i32:$sbase, IMM20bit:$offset),
   (S_BUFFER_LOAD_DWORD_IMM $sbase, (as_i32imm $offset))
 >;
 
+// Patterns for global loads with no offset
+class FlatLoadPat <FLAT inst, SDPatternOperator node, ValueType vt> : Pat <
+  (vt (node i64:$addr)),
+  (inst $addr, 0, 0, 0)
+>;
+
+def : FlatLoadPat <FLAT_LOAD_UBYTE, az_extloadi8_global, i32>;
+def : FlatLoadPat <FLAT_LOAD_SBYTE, sextloadi8_global, i32>;
+def : FlatLoadPat <FLAT_LOAD_USHORT, az_extloadi16_global, i32>;
+def : FlatLoadPat <FLAT_LOAD_SSHORT, sextloadi16_global, i32>;
+def : FlatLoadPat <FLAT_LOAD_DWORD, global_load, i32>;
+def : FlatLoadPat <FLAT_LOAD_DWORDX2, global_load, v2i32>;
+def : FlatLoadPat <FLAT_LOAD_DWORDX4, global_load, v4i32>;
+
+class FlatStorePat <FLAT inst, SDPatternOperator node, ValueType vt> : Pat <
+  (node vt:$data, i64:$addr),
+  (inst $data, $addr, 0, 0, 0)
+>;
+
+def : FlatStorePat <FLAT_STORE_BYTE, truncstorei8_global, i32>;
+def : FlatStorePat <FLAT_STORE_SHORT, truncstorei16_global, i32>;
+def : FlatStorePat <FLAT_STORE_DWORD, global_store, i32>;
+def : FlatStorePat <FLAT_STORE_DWORDX2, global_store, v2i32>;
+def : FlatStorePat <FLAT_STORE_DWORDX4, global_store, v4i32>;
+
+class FlatAtomicPat <FLAT inst, SDPatternOperator node, ValueType vt> : Pat <
+  (vt (node i64:$addr, vt:$data)),
+  (inst $addr, $data, 0, 0)
+>;
+
+def : FlatAtomicPat <FLAT_ATOMIC_ADD_RTN, atomic_add_global, i32>;
+def : FlatAtomicPat <FLAT_ATOMIC_AND_RTN, atomic_and_global, i32>;
+def : FlatAtomicPat <FLAT_ATOMIC_SUB_RTN, atomic_sub_global, i32>;
+def : FlatAtomicPat <FLAT_ATOMIC_SMAX_RTN, atomic_max_global, i32>;
+def : FlatAtomicPat <FLAT_ATOMIC_UMAX_RTN, atomic_umax_global, i32>;
+def : FlatAtomicPat <FLAT_ATOMIC_SMIN_RTN, atomic_min_global, i32>;
+def : FlatAtomicPat <FLAT_ATOMIC_UMIN_RTN, atomic_umin_global, i32>;
+def : FlatAtomicPat <FLAT_ATOMIC_OR_RTN, atomic_or_global, i32>;
+def : FlatAtomicPat <FLAT_ATOMIC_SWAP_RTN, atomic_swap_global, i32>;
+def : FlatAtomicPat <FLAT_ATOMIC_XOR_RTN, atomic_xor_global, i32>;
+
+
 } // End Predicates = [isVI]

test/CodeGen/AMDGPU/cgp-addressing-modes.ll

 ; RUN: opt -S -codegenprepare -mtriple=amdgcn-unknown-unknown < %s | FileCheck -check-prefix=OPT %s
-; RUN: llc -march=amdgcn -mcpu=bonaire -mattr=-promote-alloca < %s | FileCheck -check-prefix=GCN %s
+; RUN: llc -march=amdgcn -mcpu=bonaire -mattr=-promote-alloca < %s | FileCheck -check-prefix=GCN -check-prefix=CI %s
+; RUN: llc -march=amdgcn -mcpu=tonga -mattr=-promote-alloca < %s | FileCheck -check-prefix=GCN -check-prefix=VI %s
 
 declare i32 @llvm.r600.read.tidig.x() #0
 
 ; OPT-LABEL: @test_sink_global_small_offset_i32(
 ; OPT-NOT: getelementptr i32, i32 addrspace(1)* %in
 ; OPT: br i1
 ; OPT: ptrtoint
 
 ; GCN-LABEL: {{^}}test_sink_global_small_offset_i32:
 ; GCN: {{^}}BB0_2:
 define void @test_sink_global_small_offset_i32(i32 addrspace(1)* %out, i32 addrspace(1)* %in, i32 %cond) {
 entry:
   %out.gep = getelementptr i32, i32 addrspace(1)* %out, i64 999999
   %in.gep = getelementptr i32, i32 addrspace(1)* %in, i64 7
   %tmp0 = icmp eq i32 %cond, 0
   br i1 %tmp0, label %endif, label %if
endif:
   store i32 %load, i32 addrspace(1)* %out.gep.1
   br label %done
 
 done:
   ret void
 }
 
 ; GCN-LABEL: {{^}}test_sink_global_vreg_sreg_i32:
+; VI-DAG: s_movk_i32 flat_scratch_lo, 0x0
+; VI-DAG: s_movk_i32 flat_scratch_hi, 0x0
 ; GCN: s_and_saveexec_b64
-; GCN: buffer_load_dword {{v[0-9]+}}, {{v\[[0-9]+:[0-9]+\]}}, {{s\[[0-9]+:[0-9]+\]}}, 0 addr64{{$}}
+; CI: buffer_load_dword {{v[0-9]+}}, {{v\[[0-9]+:[0-9]+\]}}, {{s\[[0-9]+:[0-9]+\]}}, 0 addr64{{$}}
+; VI: flat_load_dword v{{[0-9]+}}, v[{{[0-9]+:[0-9]+}}]
 ; GCN: {{^}}BB7_2:
 define void @test_sink_global_vreg_sreg_i32(i32 addrspace(1)* %out, i32 addrspace(1)* %in, i32 %offset, i32 %cond) {
 entry:
   %offset.ext = zext i32 %offset to i64
   %out.gep = getelementptr i32, i32 addrspace(1)* %out, i64 999999
   %in.gep = getelementptr i32, i32 addrspace(1)* %in, i64 %offset.ext
   %tmp0 = icmp eq i32 %cond, 0
   br i1 %tmp0, label %endif, label %if

test/CodeGen/AMDGPU/global_atomics.ll

-; RUN: llc < %s -march=amdgcn -mcpu=SI -verify-machineinstrs | FileCheck --check-prefix=SI --check-prefix=FUNC %s
+; RUN: llc < %s -march=amdgcn -mcpu=SI -verify-machineinstrs | FileCheck --check-prefix=GCN --check-prefix=SI --check-prefix=FUNC %s
+; RUN: llc < %s -march=amdgcn -mcpu=tonga -verify-machineinstrs | FileCheck --check-prefix=GCN --check-prefix=VI --check-prefix=FUNC %s
+
 
 ; FUNC-LABEL: {{^}}atomic_add_i32_offset:
-; SI: buffer_atomic_add v{{[0-9]+}}, s[{{[0-9]+}}:{{[0-9]+}}], 0 offset:16{{$}}
+; GCN: buffer_atomic_add v{{[0-9]+}}, s[{{[0-9]+}}:{{[0-9]+}}], 0 offset:16{{$}}
 define void @atomic_add_i32_offset(i32 addrspace(1)* %out, i32 %in) {
 entry:
   %gep = getelementptr i32, i32 addrspace(1)* %out, i32 4
   %0  = atomicrmw volatile add i32 addrspace(1)* %gep, i32 %in seq_cst
   ret void
 }
 
 ; FUNC-LABEL: {{^}}atomic_add_i32_ret_offset:
-; SI: buffer_atomic_add [[RET:v[0-9]+]], s[{{[0-9]+}}:{{[0-9]+}}], 0 offset:16 glc {{$}}
-; SI: buffer_store_dword [[RET]]
+; GCN: buffer_atomic_add [[RET:v[0-9]+]], s[{{[0-9]+}}:{{[0-9]+}}], 0 offset:16 glc {{$}}
+; GCN: buffer_store_dword [[RET]]
 define void @atomic_add_i32_ret_offset(i32 addrspace(1)* %out, i32 addrspace(1)* %out2, i32 %in) {
 entry:
   %gep = getelementptr i32, i32 addrspace(1)* %out, i32 4
   %0  = atomicrmw volatile add i32 addrspace(1)* %gep, i32 %in seq_cst
   store i32 %0, i32 addrspace(1)* %out2
   ret void
 }
 
 ; FUNC-LABEL: {{^}}atomic_add_i32_addr64_offset:
 ; SI: buffer_atomic_add v{{[0-9]+}}, v[{{[0-9]+}}:{{[0-9]+}}], s[{{[0-9]+}}:{{[0-9]+}}], 0 addr64 offset:16{{$}}
+; VI: s_movk_i32 flat_scratch_lo, 0x0
+; VI: s_movk_i32 flat_scratch_hi, 0x0
+; VI: flat_atomic_add v[{{[0-9]+:[0-9]+}}], v{{[0-9]+$}}
+
 define void @atomic_add_i32_addr64_offset(i32 addrspace(1)* %out, i32 %in, i64 %index) {
 entry:
   %ptr = getelementptr i32, i32 addrspace(1)* %out, i64 %index
   %gep = getelementptr i32, i32 addrspace(1)* %ptr, i32 4
   %0  = atomicrmw volatile add i32 addrspace(1)* %gep, i32 %in seq_cst
   ret void
 }
 
 ; FUNC-LABEL: {{^}}atomic_add_i32_ret_addr64_offset:
 ; SI: buffer_atomic_add [[RET:v[0-9]+]], v[{{[0-9]+}}:{{[0-9]+}}], s[{{[0-9]+}}:{{[0-9]+}}], 0 addr64 offset:16 glc{{$}}
-; SI: buffer_store_dword [[RET]]
+; VI: s_movk_i32 flat_scratch_lo, 0x0
+; VI: s_movk_i32 flat_scratch_hi, 0x0
+; VI: flat_atomic_add [[RET:v[0-9]+]], v[{{[0-9]+:[0-9]+}}], v{{[0-9]+}} glc{{$}}
+; GCN: buffer_store_dword [[RET]]
 define void @atomic_add_i32_ret_addr64_offset(i32 addrspace(1)* %out, i32 addrspace(1)* %out2, i32 %in, i64 %index) {
 entry:
   %ptr = getelementptr i32, i32 addrspace(1)* %out, i64 %index
   %gep = getelementptr i32, i32 addrspace(1)* %ptr, i32 4
   %0  = atomicrmw volatile add i32 addrspace(1)* %gep, i32 %in seq_cst
   store i32 %0, i32 addrspace(1)* %out2
   ret void
 }
 
 ; FUNC-LABEL: {{^}}atomic_add_i32:
-; SI: buffer_atomic_add v{{[0-9]+}}, s[{{[0-9]+}}:{{[0-9]+}}], 0{{$}}
+; GCN: buffer_atomic_add v{{[0-9]+}}, s[{{[0-9]+}}:{{[0-9]+}}], 0{{$}}
 define void @atomic_add_i32(i32 addrspace(1)* %out, i32 %in) {
 entry:
   %0  = atomicrmw volatile add i32 addrspace(1)* %out, i32 %in seq_cst
   ret void
 }
 
 ; FUNC-LABEL: {{^}}atomic_add_i32_ret:
-; SI: buffer_atomic_add [[RET:v[0-9]+]], s[{{[0-9]+}}:{{[0-9]+}}], 0 glc
-; SI: buffer_store_dword [[RET]]
+; GCN: buffer_atomic_add [[RET:v[0-9]+]], s[{{[0-9]+}}:{{[0-9]+}}], 0 glc
+; GCN: buffer_store_dword [[RET]]
 define void @atomic_add_i32_ret(i32 addrspace(1)* %out, i32 addrspace(1)* %out2, i32 %in) {
 entry:
   %0  = atomicrmw volatile add i32 addrspace(1)* %out, i32 %in seq_cst
   store i32 %0, i32 addrspace(1)* %out2
   ret void
 }
 
 ; FUNC-LABEL: {{^}}atomic_add_i32_addr64:
 ; SI: buffer_atomic_add v{{[0-9]+}}, v[{{[0-9]+}}:{{[0-9]+}}], s[{{[0-9]+}}:{{[0-9]+}}], 0 addr64{{$}}
+; VI: s_movk_i32 flat_scratch_lo, 0x0
+; VI: s_movk_i32 flat_scratch_hi, 0x0
+; VI: flat_atomic_add v[{{[0-9]+:[0-9]+}}], v{{[0-9]+$}}
 define void @atomic_add_i32_addr64(i32 addrspace(1)* %out, i32 %in, i64 %index) {
 entry:
   %ptr = getelementptr i32, i32 addrspace(1)* %out, i64 %index
   %0  = atomicrmw volatile add i32 addrspace(1)* %ptr, i32 %in seq_cst
   ret void
 }
 
 ; FUNC-LABEL: {{^}}atomic_add_i32_ret_addr64:
 ; SI: buffer_atomic_add [[RET:v[0-9]+]], v[{{[0-9]+}}:{{[0-9]+}}], s[{{[0-9]+}}:{{[0-9]+}}], 0 addr64 glc{{$}}
-; SI: buffer_store_dword [[RET]]
+; VI: s_movk_i32 flat_scratch_lo, 0x0
+; VI: s_movk_i32 flat_scratch_hi, 0x0
+; VI: flat_atomic_add [[RET:v[0-9]+]], v[{{[0-9]+:[0-9]+}}], v{{[0-9]+}} glc{{$}}
+; GCN: buffer_store_dword [[RET]]
 define void @atomic_add_i32_ret_addr64(i32 addrspace(1)* %out, i32 addrspace(1)* %out2, i32 %in, i64 %index) {
 entry:
   %ptr = getelementptr i32, i32 addrspace(1)* %out, i64 %index
   %0  = atomicrmw volatile add i32 addrspace(1)* %ptr, i32 %in seq_cst
   store i32 %0, i32 addrspace(1)* %out2
   ret void
 }
 
 ; FUNC-LABEL: {{^}}atomic_and_i32_offset:
-; SI: buffer_atomic_and v{{[0-9]+}}, s[{{[0-9]+}}:{{[0-9]+}}], 0 offset:16{{$}}
+; GCN: buffer_atomic_and v{{[0-9]+}}, s[{{[0-9]+}}:{{[0-9]+}}], 0 offset:16{{$}}
 define void @atomic_and_i32_offset(i32 addrspace(1)* %out, i32 %in) {
 entry:
   %gep = getelementptr i32, i32 addrspace(1)* %out, i32 4
   %0  = atomicrmw volatile and i32 addrspace(1)* %gep, i32 %in seq_cst
   ret void
 }
 
 ; FUNC-LABEL: {{^}}atomic_and_i32_ret_offset:
-; SI: buffer_atomic_and [[RET:v[0-9]+]], s[{{[0-9]+}}:{{[0-9]+}}], 0 offset:16 glc {{$}}
-; SI: buffer_store_dword [[RET]]
+; GCN: buffer_atomic_and [[RET:v[0-9]+]], s[{{[0-9]+}}:{{[0-9]+}}], 0 offset:16 glc {{$}}
+; GCN: buffer_store_dword [[RET]]
 define void @atomic_and_i32_ret_offset(i32 addrspace(1)* %out, i32 addrspace(1)* %out2, i32 %in) {
 entry:
   %gep = getelementptr i32, i32 addrspace(1)* %out, i32 4
   %0  = atomicrmw volatile and i32 addrspace(1)* %gep, i32 %in seq_cst
   store i32 %0, i32 addrspace(1)* %out2
   ret void
 }
 
 ; FUNC-LABEL: {{^}}atomic_and_i32_addr64_offset:
 ; SI: buffer_atomic_and v{{[0-9]+}}, v[{{[0-9]+}}:{{[0-9]+}}], s[{{[0-9]+}}:{{[0-9]+}}], 0 addr64 offset:16{{$}}
+; VI: s_movk_i32 flat_scratch_lo, 0x0
+; VI: s_movk_i32 flat_scratch_hi, 0x0
+; VI: flat_atomic_and v[{{[0-9]+:[0-9]+}}], v{{[0-9]+$}}
 define void @atomic_and_i32_addr64_offset(i32 addrspace(1)* %out, i32 %in, i64 %index) {
 entry:
   %ptr = getelementptr i32, i32 addrspace(1)* %out, i64 %index
   %gep = getelementptr i32, i32 addrspace(1)* %ptr, i32 4
   %0  = atomicrmw volatile and i32 addrspace(1)* %gep, i32 %in seq_cst
   ret void
 }
 
 ; FUNC-LABEL: {{^}}atomic_and_i32_ret_addr64_offset:
 ; SI: buffer_atomic_and [[RET:v[0-9]+]], v[{{[0-9]+}}:{{[0-9]+}}], s[{{[0-9]+}}:{{[0-9]+}}], 0 addr64 offset:16 glc{{$}}
-; SI: buffer_store_dword [[RET]]
+; VI: s_movk_i32 flat_scratch_lo, 0x0
+; VI: s_movk_i32 flat_scratch_hi, 0x0
+; VI: flat_atomic_and [[RET:v[0-9]]], v[{{[0-9]+:[0-9]+}}], v{{[0-9]+}} glc{{$}}
+; GCN: buffer_store_dword [[RET]]
 define void @atomic_and_i32_ret_addr64_offset(i32 addrspace(1)* %out, i32 addrspace(1)* %out2, i32 %in, i64 %index) {
 entry:
   %ptr = getelementptr i32, i32 addrspace(1)* %out, i64 %index
   %gep = getelementptr i32, i32 addrspace(1)* %ptr, i32 4
   %0  = atomicrmw volatile and i32 addrspace(1)* %gep, i32 %in seq_cst
   store i32 %0, i32 addrspace(1)* %out2
   ret void
 }
 
 ; FUNC-LABEL: {{^}}atomic_and_i32:
-; SI: buffer_atomic_and v{{[0-9]+}}, s[{{[0-9]+}}:{{[0-9]+}}], 0{{$}}
+; GCN: buffer_atomic_and v{{[0-9]+}}, s[{{[0-9]+}}:{{[0-9]+}}], 0{{$}}
 define void @atomic_and_i32(i32 addrspace(1)* %out, i32 %in) {
 entry:
   %0  = atomicrmw volatile and i32 addrspace(1)* %out, i32 %in seq_cst
   ret void
 }
 
 ; FUNC-LABEL: {{^}}atomic_and_i32_ret:
-; SI: buffer_atomic_and [[RET:v[0-9]+]], s[{{[0-9]+}}:{{[0-9]+}}], 0 glc
-; SI: buffer_store_dword [[RET]]
+; GCN: buffer_atomic_and [[RET:v[0-9]+]], s[{{[0-9]+}}:{{[0-9]+}}], 0 glc
+; GCN: buffer_store_dword [[RET]]
 define void @atomic_and_i32_ret(i32 addrspace(1)* %out, i32 addrspace(1)* %out2, i32 %in) {
 entry:
   %0  = atomicrmw volatile and i32 addrspace(1)* %out, i32 %in seq_cst
   store i32 %0, i32 addrspace(1)* %out2
   ret void
 }
 
 ; FUNC-LABEL: {{^}}atomic_and_i32_addr64:
 ; SI: buffer_atomic_and v{{[0-9]+}}, v[{{[0-9]+}}:{{[0-9]+}}], s[{{[0-9]+}}:{{[0-9]+}}], 0 addr64{{$}}
+; VI: s_movk_i32 flat_scratch_lo, 0x0
+; VI: s_movk_i32 flat_scratch_hi, 0x0
+; VI: flat_atomic_and v[{{[0-9]+:[0-9]+}}], v{{[0-9]+$}}
 define void @atomic_and_i32_addr64(i32 addrspace(1)* %out, i32 %in, i64 %index) {
 entry:
   %ptr = getelementptr i32, i32 addrspace(1)* %out, i64 %index
   %0  = atomicrmw volatile and i32 addrspace(1)* %ptr, i32 %in seq_cst
   ret void
 }
 
 ; FUNC-LABEL: {{^}}atomic_and_i32_ret_addr64:
 ; SI: buffer_atomic_and [[RET:v[0-9]+]], v[{{[0-9]+}}:{{[0-9]+}}], s[{{[0-9]+}}:{{[0-9]+}}], 0 addr64 glc{{$}}
-; SI: buffer_store_dword [[RET]]
+; VI: s_movk_i32 flat_scratch_lo, 0x0
+; VI: s_movk_i32 flat_scratch_hi, 0x0
+; VI: flat_atomic_and [[RET:v[0-9]+]], v[{{[0-9]+:[0-9]+}}], v{{[0-9]+}} glc{{$}}
+; GCN: buffer_store_dword [[RET]]
 define void @atomic_and_i32_ret_addr64(i32 addrspace(1)* %out, i32 addrspace(1)* %out2, i32 %in, i64 %index) {
 entry:
   %ptr = getelementptr i32, i32 addrspace(1)* %out, i64 %index
   %0  = atomicrmw volatile and i32 addrspace(1)* %ptr, i32 %in seq_cst
   store i32 %0, i32 addrspace(1)* %out2
   ret void
 }
 
 ; FUNC-LABEL: {{^}}atomic_sub_i32_offset:
-; SI: buffer_atomic_sub v{{[0-9]+}}, s[{{[0-9]+}}:{{[0-9]+}}], 0 offset:16{{$}}
+; GCN: buffer_atomic_sub v{{[0-9]+}}, s[{{[0-9]+}}:{{[0-9]+}}], 0 offset:16{{$}}
 define void @atomic_sub_i32_offset(i32 addrspace(1)* %out, i32 %in) {
 entry:
   %gep = getelementptr i32, i32 addrspace(1)* %out, i32 4
   %0  = atomicrmw volatile sub i32 addrspace(1)* %gep, i32 %in seq_cst
   ret void
 }
 
 ; FUNC-LABEL: {{^}}atomic_sub_i32_ret_offset:
-; SI: buffer_atomic_sub [[RET:v[0-9]+]], s[{{[0-9]+}}:{{[0-9]+}}], 0 offset:16 glc {{$}}
-; SI: buffer_store_dword [[RET]]
+; GCN: buffer_atomic_sub [[RET:v[0-9]+]], s[{{[0-9]+}}:{{[0-9]+}}], 0 offset:16 glc {{$}}
+; GCN: buffer_store_dword [[RET]]
 define void @atomic_sub_i32_ret_offset(i32 addrspace(1)* %out, i32 addrspace(1)* %out2, i32 %in) {
 entry:
   %gep = getelementptr i32, i32 addrspace(1)* %out, i32 4
   %0  = atomicrmw volatile sub i32 addrspace(1)* %gep, i32 %in seq_cst
   store i32 %0, i32 addrspace(1)* %out2
   ret void
 }
 
 ; FUNC-LABEL: {{^}}atomic_sub_i32_addr64_offset:
 ; SI: buffer_atomic_sub v{{[0-9]+}}, v[{{[0-9]+}}:{{[0-9]+}}], s[{{[0-9]+}}:{{[0-9]+}}], 0 addr64 offset:16{{$}}
+; VI: s_movk_i32 flat_scratch_lo, 0x0
+; VI: s_movk_i32 flat_scratch_hi, 0x0
+; VI: flat_atomic_sub v[{{[0-9]+:[0-9]+}}], v{{[0-9]+$}}
 define void @atomic_sub_i32_addr64_offset(i32 addrspace(1)* %out, i32 %in, i64 %index) {
 entry:
   %ptr = getelementptr i32, i32 addrspace(1)* %out, i64 %index
   %gep = getelementptr i32, i32 addrspace(1)* %ptr, i32 4
   %0  = atomicrmw volatile sub i32 addrspace(1)* %gep, i32 %in seq_cst
   ret void
 }
 
 ; FUNC-LABEL: {{^}}atomic_sub_i32_ret_addr64_offset:
 ; SI: buffer_atomic_sub [[RET:v[0-9]+]], v[{{[0-9]+}}:{{[0-9]+}}], s[{{[0-9]+}}:{{[0-9]+}}], 0 addr64 offset:16 glc{{$}}
-; SI: buffer_store_dword [[RET]]
+; VI: s_movk_i32 flat_scratch_lo, 0x0
+; VI: s_movk_i32 flat_scratch_hi, 0x0
+; VI: flat_atomic_sub [[RET:v[0-9]+]], v[{{[0-9]+:[0-9]+}}], v{{[0-9]+}} glc{{$}}
+; GCN: buffer_store_dword [[RET]]
 define void @atomic_sub_i32_ret_addr64_offset(i32 addrspace(1)* %out, i32 addrspace(1)* %out2, i32 %in, i64 %index) {
 entry:
   %ptr = getelementptr i32, i32 addrspace(1)* %out, i64 %index
   %gep = getelementptr i32, i32 addrspace(1)* %ptr, i32 4
   %0  = atomicrmw volatile sub i32 addrspace(1)* %gep, i32 %in seq_cst
   store i32 %0, i32 addrspace(1)* %out2
   ret void
 }
 
 ; FUNC-LABEL: {{^}}atomic_sub_i32:
-; SI: buffer_atomic_sub v{{[0-9]+}}, s[{{[0-9]+}}:{{[0-9]+}}], 0{{$}}
+; GCN: buffer_atomic_sub v{{[0-9]+}}, s[{{[0-9]+}}:{{[0-9]+}}], 0{{$}}
 define void @atomic_sub_i32(i32 addrspace(1)* %out, i32 %in) {
 entry:
   %0  = atomicrmw volatile sub i32 addrspace(1)* %out, i32 %in seq_cst
   ret void
 }
 
 ; FUNC-LABEL: {{^}}atomic_sub_i32_ret:
-; SI: buffer_atomic_sub [[RET:v[0-9]+]], s[{{[0-9]+}}:{{[0-9]+}}], 0 glc
-; SI: buffer_store_dword [[RET]]
+; GCN: buffer_atomic_sub [[RET:v[0-9]+]], s[{{[0-9]+}}:{{[0-9]+}}], 0 glc
+; GCN: buffer_store_dword [[RET]]
 define void @atomic_sub_i32_ret(i32 addrspace(1)* %out, i32 addrspace(1)* %out2, i32 %in) {
 entry:
   %0  = atomicrmw volatile sub i32 addrspace(1)* %out, i32 %in seq_cst
   store i32 %0, i32 addrspace(1)* %out2
   ret void
 }
 
 ; FUNC-LABEL: {{^}}atomic_sub_i32_addr64:
 ; SI: buffer_atomic_sub v{{[0-9]+}}, v[{{[0-9]+}}:{{[0-9]+}}], s[{{[0-9]+}}:{{[0-9]+}}], 0 addr64{{$}}
+; VI: s_movk_i32 flat_scratch_lo, 0x0
+; VI: s_movk_i32 flat_scratch_hi, 0x0
+; VI: flat_atomic_sub v[{{[0-9]+:[0-9]+}}], v{{[0-9]+$}}
 define void @atomic_sub_i32_addr64(i32 addrspace(1)* %out, i32 %in, i64 %index) {
 entry:
   %ptr = getelementptr i32, i32 addrspace(1)* %out, i64 %index
   %0  = atomicrmw volatile sub i32 addrspace(1)* %ptr, i32 %in seq_cst
   ret void
 }
 
 ; FUNC-LABEL: {{^}}atomic_sub_i32_ret_addr64:
 ; SI: buffer_atomic_sub [[RET:v[0-9]+]], v[{{[0-9]+}}:{{[0-9]+}}], s[{{[0-9]+}}:{{[0-9]+}}], 0 addr64 glc{{$}}
-; SI: buffer_store_dword [[RET]]
+; VI: s_movk_i32 flat_scratch_lo, 0x0
+; VI: s_movk_i32 flat_scratch_hi, 0x0
+; VI: flat_atomic_sub [[RET:v[0-9]+]], v[{{[0-9]+:[0-9]+}}], v{{[0-9]+}} glc{{$}}
+; GCN: buffer_store_dword [[RET]]
 define void @atomic_sub_i32_ret_addr64(i32 addrspace(1)* %out, i32 addrspace(1)* %out2, i32 %in, i64 %index) {
 entry:
   %ptr = getelementptr i32, i32 addrspace(1)* %out, i64 %index
   %0  = atomicrmw volatile sub i32 addrspace(1)* %ptr, i32 %in seq_cst
   store i32 %0, i32 addrspace(1)* %out2
   ret void
 }
 
 ; FUNC-LABEL: {{^}}atomic_max_i32_offset:
-; SI: buffer_atomic_smax v{{[0-9]+}}, s[{{[0-9]+}}:{{[0-9]+}}], 0 offset:16{{$}}
+; GCN: buffer_atomic_smax v{{[0-9]+}}, s[{{[0-9]+}}:{{[0-9]+}}], 0 offset:16{{$}}
 define void @atomic_max_i32_offset(i32 addrspace(1)* %out, i32 %in) {
 entry:
   %gep = getelementptr i32, i32 addrspace(1)* %out, i32 4
   %0  = atomicrmw volatile max i32 addrspace(1)* %gep, i32 %in seq_cst
   ret void
 }
 
 ; FUNC-LABEL: {{^}}atomic_max_i32_ret_offset:
-; SI: buffer_atomic_smax [[RET:v[0-9]+]], s[{{[0-9]+}}:{{[0-9]+}}], 0 offset:16 glc {{$}}
-; SI: buffer_store_dword [[RET]]
+; GCN: buffer_atomic_smax [[RET:v[0-9]+]], s[{{[0-9]+}}:{{[0-9]+}}], 0 offset:16 glc {{$}}
+; GCN: buffer_store_dword [[RET]]
 define void @atomic_max_i32_ret_offset(i32 addrspace(1)* %out, i32 addrspace(1)* %out2, i32 %in) {
 entry:
   %gep = getelementptr i32, i32 addrspace(1)* %out, i32 4
   %0  = atomicrmw volatile max i32 addrspace(1)* %gep, i32 %in seq_cst
   store i32 %0, i32 addrspace(1)* %out2
   ret void
 }
 
 ; FUNC-LABEL: {{^}}atomic_max_i32_addr64_offset:
 ; SI: buffer_atomic_smax v{{[0-9]+}}, v[{{[0-9]+}}:{{[0-9]+}}], s[{{[0-9]+}}:{{[0-9]+}}], 0 addr64 offset:16{{$}}
+; VI: s_movk_i32 flat_scratch_lo, 0x0
+; VI: s_movk_i32 flat_scratch_hi, 0x0
+; VI: flat_atomic_smax v[{{[0-9]+:[0-9]+}}], v{{[0-9]+$}}
 define void @atomic_max_i32_addr64_offset(i32 addrspace(1)* %out, i32 %in, i64 %index) {
 entry:
   %ptr = getelementptr i32, i32 addrspace(1)* %out, i64 %index
   %gep = getelementptr i32, i32 addrspace(1)* %ptr, i32 4
   %0  = atomicrmw volatile max i32 addrspace(1)* %gep, i32 %in seq_cst
   ret void
 }
 
 ; FUNC-LABEL: {{^}}atomic_max_i32_ret_addr64_offset:
 ; SI: buffer_atomic_smax [[RET:v[0-9]+]], v[{{[0-9]+}}:{{[0-9]+}}], s[{{[0-9]+}}:{{[0-9]+}}], 0 addr64 offset:16 glc{{$}}
-; SI: buffer_store_dword [[RET]]
+; VI: s_movk_i32 flat_scratch_lo, 0x0
+; VI: s_movk_i32 flat_scratch_hi, 0x0
+; VI: flat_atomic_smax [[RET:v[0-9]+]], v[{{[0-9]+:[0-9]+}}], v{{[0-9]+}} glc{{$}}
+; GCN: buffer_store_dword [[RET]]
 define void @atomic_max_i32_ret_addr64_offset(i32 addrspace(1)* %out, i32 addrspace(1)* %out2, i32 %in, i64 %index) {
 entry:
   %ptr = getelementptr i32, i32 addrspace(1)* %out, i64 %index
   %gep = getelementptr i32, i32 addrspace(1)* %ptr, i32 4
   %0  = atomicrmw volatile max i32 addrspace(1)* %gep, i32 %in seq_cst
   store i32 %0, i32 addrspace(1)* %out2
   ret void
 }
 
 ; FUNC-LABEL: {{^}}atomic_max_i32:
-; SI: buffer_atomic_smax v{{[0-9]+}}, s[{{[0-9]+}}:{{[0-9]+}}], 0{{$}}
+; GCN: buffer_atomic_smax v{{[0-9]+}}, s[{{[0-9]+}}:{{[0-9]+}}], 0{{$}}
 define void @atomic_max_i32(i32 addrspace(1)* %out, i32 %in) {
 entry:
   %0  = atomicrmw volatile max i32 addrspace(1)* %out, i32 %in seq_cst
   ret void
 }
 
 ; FUNC-LABEL: {{^}}atomic_max_i32_ret:
-; SI: buffer_atomic_smax [[RET:v[0-9]+]], s[{{[0-9]+}}:{{[0-9]+}}], 0 glc
-; SI: buffer_store_dword [[RET]]
+; GCN: buffer_atomic_smax [[RET:v[0-9]+]], s[{{[0-9]+}}:{{[0-9]+}}], 0 glc
+; GCN: buffer_store_dword [[RET]]
 define void @atomic_max_i32_ret(i32 addrspace(1)* %out, i32 addrspace(1)* %out2, i32 %in) {
 entry:
   %0  = atomicrmw volatile max i32 addrspace(1)* %out, i32 %in seq_cst
   store i32 %0, i32 addrspace(1)* %out2
   ret void
 }
 
 ; FUNC-LABEL: {{^}}atomic_max_i32_addr64:
 ; SI: buffer_atomic_smax v{{[0-9]+}}, v[{{[0-9]+}}:{{[0-9]+}}], s[{{[0-9]+}}:{{[0-9]+}}], 0 addr64{{$}}
+; VI: s_movk_i32 flat_scratch_lo, 0x0
+; VI: s_movk_i32 flat_scratch_hi, 0x0
+; VI: flat_atomic_smax v[{{[0-9]+:[0-9]+}}], v{{[0-9]+$}}
 define void @atomic_max_i32_addr64(i32 addrspace(1)* %out, i32 %in, i64 %index) {
 entry:
   %ptr = getelementptr i32, i32 addrspace(1)* %out, i64 %index
   %0  = atomicrmw volatile max i32 addrspace(1)* %ptr, i32 %in seq_cst
   ret void
 }
 
 ; FUNC-LABEL: {{^}}atomic_max_i32_ret_addr64:
 ; SI: buffer_atomic_smax [[RET:v[0-9]+]], v[{{[0-9]+}}:{{[0-9]+}}], s[{{[0-9]+}}:{{[0-9]+}}], 0 addr64 glc{{$}}
-; SI: buffer_store_dword [[RET]]
+; VI: s_movk_i32 flat_scratch_lo, 0x0
+; VI: s_movk_i32 flat_scratch_hi, 0x0
+; VI: flat_atomic_smax [[RET:v[0-9]+]], v[{{[0-9]+:[0-9]+}}], v{{[0-9]+}} glc{{$}}
+; GCN: buffer_store_dword [[RET]]
 define void @atomic_max_i32_ret_addr64(i32 addrspace(1)* %out, i32 addrspace(1)* %out2, i32 %in, i64 %index) {
 entry:
   %ptr = getelementptr i32, i32 addrspace(1)* %out, i64 %index
   %0  = atomicrmw volatile max i32 addrspace(1)* %ptr, i32 %in seq_cst
   store i32 %0, i32 addrspace(1)* %out2
   ret void
 }
 
 ; FUNC-LABEL: {{^}}atomic_umax_i32_offset:
-; SI: buffer_atomic_umax v{{[0-9]+}}, s[{{[0-9]+}}:{{[0-9]+}}], 0 offset:16{{$}}
+; GCN: buffer_atomic_umax v{{[0-9]+}}, s[{{[0-9]+}}:{{[0-9]+}}], 0 offset:16{{$}}
 define void @atomic_umax_i32_offset(i32 addrspace(1)* %out, i32 %in) {
 entry:
   %gep = getelementptr i32, i32 addrspace(1)* %out, i32 4
   %0  = atomicrmw volatile umax i32 addrspace(1)* %gep, i32 %in seq_cst
   ret void
 }
 
 ; FUNC-LABEL: {{^}}atomic_umax_i32_ret_offset:
-; SI: buffer_atomic_umax [[RET:v[0-9]+]], s[{{[0-9]+}}:{{[0-9]+}}], 0 offset:16 glc {{$}}
-; SI: buffer_store_dword [[RET]]
+; GCN: buffer_atomic_umax [[RET:v[0-9]+]], s[{{[0-9]+}}:{{[0-9]+}}], 0 offset:16 glc {{$}}
+; GCN: buffer_store_dword [[RET]]
 define void @atomic_umax_i32_ret_offset(i32 addrspace(1)* %out, i32 addrspace(1)* %out2, i32 %in) {
 entry:
   %gep = getelementptr i32, i32 addrspace(1)* %out, i32 4
   %0  = atomicrmw volatile umax i32 addrspace(1)* %gep, i32 %in seq_cst
   store i32 %0, i32 addrspace(1)* %out2
   ret void
 }
 
 ; FUNC-LABEL: {{^}}atomic_umax_i32_addr64_offset:
 ; SI: buffer_atomic_umax v{{[0-9]+}}, v[{{[0-9]+}}:{{[0-9]+}}], s[{{[0-9]+}}:{{[0-9]+}}], 0 addr64 offset:16{{$}}
+; VI: s_movk_i32 flat_scratch_lo, 0x0
+; VI: s_movk_i32 flat_scratch_hi, 0x0
+; VI: flat_atomic_umax v[{{[0-9]+:[0-9]+}}], v{{[0-9]+$}}
 define void @atomic_umax_i32_addr64_offset(i32 addrspace(1)* %out, i32 %in, i64 %index) {
 entry:
   %ptr = getelementptr i32, i32 addrspace(1)* %out, i64 %index
   %gep = getelementptr i32, i32 addrspace(1)* %ptr, i32 4
   %0  = atomicrmw volatile umax i32 addrspace(1)* %gep, i32 %in seq_cst
   ret void
 }
 
 ; FUNC-LABEL: {{^}}atomic_umax_i32_ret_addr64_offset:
 ; SI: buffer_atomic_umax [[RET:v[0-9]+]], v[{{[0-9]+}}:{{[0-9]+}}], s[{{[0-9]+}}:{{[0-9]+}}], 0 addr64 offset:16 glc{{$}}
-; SI: buffer_store_dword [[RET]]
+; VI: s_movk_i32 flat_scratch_lo, 0x0
+; VI: s_movk_i32 flat_scratch_hi, 0x0
+; VI: flat_atomic_umax [[RET:v[0-9]+]], v[{{[0-9]+:[0-9]+}}], v{{[0-9]+}} glc{{$}}
+; GCN: buffer_store_dword [[RET]]
 define void @atomic_umax_i32_ret_addr64_offset(i32 addrspace(1)* %out, i32 addrspace(1)* %out2, i32 %in, i64 %index) {
 entry:
   %ptr = getelementptr i32, i32 addrspace(1)* %out, i64 %index
   %gep = getelementptr i32, i32 addrspace(1)* %ptr, i32 4
   %0  = atomicrmw volatile umax i32 addrspace(1)* %gep, i32 %in seq_cst
   store i32 %0, i32 addrspace(1)* %out2
   ret void
 }
 
 ; FUNC-LABEL: {{^}}atomic_umax_i32:
-; SI: buffer_atomic_umax v{{[0-9]+}}, s[{{[0-9]+}}:{{[0-9]+}}], 0{{$}}
+; GCN: buffer_atomic_umax v{{[0-9]+}}, s[{{[0-9]+}}:{{[0-9]+}}], 0{{$}}
 define void @atomic_umax_i32(i32 addrspace(1)* %out, i32 %in) {
 entry:
   %0  = atomicrmw volatile umax i32 addrspace(1)* %out, i32 %in seq_cst
   ret void
 }
 
 ; FUNC-LABEL: {{^}}atomic_umax_i32_ret:
-; SI: buffer_atomic_umax [[RET:v[0-9]+]], s[{{[0-9]+}}:{{[0-9]+}}], 0 glc
-; SI: buffer_store_dword [[RET]]
+; GCN: buffer_atomic_umax [[RET:v[0-9]+]], s[{{[0-9]+}}:{{[0-9]+}}], 0 glc
+; GCN: buffer_store_dword [[RET]]
 define void @atomic_umax_i32_ret(i32 addrspace(1)* %out, i32 addrspace(1)* %out2, i32 %in) {
 entry:
   %0  = atomicrmw volatile umax i32 addrspace(1)* %out, i32 %in seq_cst
   store i32 %0, i32 addrspace(1)* %out2
   ret void
 }
 
 ; FUNC-LABEL: {{^}}atomic_umax_i32_addr64:
 ; SI: buffer_atomic_umax v{{[0-9]+}}, v[{{[0-9]+}}:{{[0-9]+}}], s[{{[0-9]+}}:{{[0-9]+}}], 0 addr64{{$}}
+; VI: s_movk_i32 flat_scratch_lo, 0x0
+; VI: s_movk_i32 flat_scratch_hi, 0x0
+; VI: flat_atomic_umax v[{{[0-9]+:[0-9]+}}], v{{[0-9]+$}}
 define void @atomic_umax_i32_addr64(i32 addrspace(1)* %out, i32 %in, i64 %index) {
 entry:
   %ptr = getelementptr i32, i32 addrspace(1)* %out, i64 %index
   %0  = atomicrmw volatile umax i32 addrspace(1)* %ptr, i32 %in seq_cst
   ret void
 }
 
 ; FUNC-LABEL: {{^}}atomic_umax_i32_ret_addr64:
 ; SI: buffer_atomic_umax [[RET:v[0-9]+]], v[{{[0-9]+}}:{{[0-9]+}}], s[{{[0-9]+}}:{{[0-9]+}}], 0 addr64 glc{{$}}
-; SI: buffer_store_dword [[RET]]
+; VI: s_movk_i32 flat_scratch_lo, 0x0
+; VI: s_movk_i32 flat_scratch_hi, 0x0
+; VI: flat_atomic_umax [[RET:v[0-9]+]], v[{{[0-9]+:[0-9]+}}], v{{[0-9]+}} glc{{$}}
+; GCN: buffer_store_dword [[RET]]
 define void @atomic_umax_i32_ret_addr64(i32 addrspace(1)* %out, i32 addrspace(1)* %out2, i32 %in, i64 %index) {
 entry:
   %ptr = getelementptr i32, i32 addrspace(1)* %out, i64 %index
   %0  = atomicrmw volatile umax i32 addrspace(1)* %ptr, i32 %in seq_cst
   store i32 %0, i32 addrspace(1)* %out2
   ret void
 }
 
 ; FUNC-LABEL: {{^}}atomic_min_i32_offset:
-; SI: buffer_atomic_smin v{{[0-9]+}}, s[{{[0-9]+}}:{{[0-9]+}}], 0 offset:16{{$}}
+; GCN: buffer_atomic_smin v{{[0-9]+}}, s[{{[0-9]+}}:{{[0-9]+}}], 0 offset:16{{$}}
 define void @atomic_min_i32_offset(i32 addrspace(1)* %out, i32 %in) {
 entry:
   %gep = getelementptr i32, i32 addrspace(1)* %out, i32 4
   %0  = atomicrmw volatile min i32 addrspace(1)* %gep, i32 %in seq_cst
   ret void
 }
 
 ; FUNC-LABEL: {{^}}atomic_min_i32_ret_offset:
-; SI: buffer_atomic_smin [[RET:v[0-9]+]], s[{{[0-9]+}}:{{[0-9]+}}], 0 offset:16 glc {{$}}
-; SI: buffer_store_dword [[RET]]
+; GCN: buffer_atomic_smin [[RET:v[0-9]+]], s[{{[0-9]+}}:{{[0-9]+}}], 0 offset:16 glc {{$}}
+; GCN: buffer_store_dword [[RET]]
 define void @atomic_min_i32_ret_offset(i32 addrspace(1)* %out, i32 addrspace(1)* %out2, i32 %in) {
 entry:
   %gep = getelementptr i32, i32 addrspace(1)* %out, i32 4
   %0  = atomicrmw volatile min i32 addrspace(1)* %gep, i32 %in seq_cst
   store i32 %0, i32 addrspace(1)* %out2
   ret void
 }
 
 ; FUNC-LABEL: {{^}}atomic_min_i32_addr64_offset:
 ; SI: buffer_atomic_smin v{{[0-9]+}}, v[{{[0-9]+}}:{{[0-9]+}}], s[{{[0-9]+}}:{{[0-9]+}}], 0 addr64 offset:16{{$}}
+; VI: s_movk_i32 flat_scratch_lo, 0x0
+; VI: s_movk_i32 flat_scratch_hi, 0x0
+; VI: flat_atomic_smin v[{{[0-9]+:[0-9]+}}], v{{[0-9]+$}}
 define void @atomic_min_i32_addr64_offset(i32 addrspace(1)* %out, i32 %in, i64 %index) {
 entry:
   %ptr = getelementptr i32, i32 addrspace(1)* %out, i64 %index
   %gep = getelementptr i32, i32 addrspace(1)* %ptr, i32 4
   %0  = atomicrmw volatile min i32 addrspace(1)* %gep, i32 %in seq_cst
   ret void
 }
 
 ; FUNC-LABEL: {{^}}atomic_min_i32_ret_addr64_offset:
 ; SI: buffer_atomic_smin [[RET:v[0-9]+]], v[{{[0-9]+}}:{{[0-9]+}}], s[{{[0-9]+}}:{{[0-9]+}}], 0 addr64 offset:16 glc{{$}}
-; SI: buffer_store_dword [[RET]]
+; VI: s_movk_i32 flat_scratch_lo, 0x0
+; VI: s_movk_i32 flat_scratch_hi, 0x0
+; VI: flat_atomic_smin [[RET:v[0-9]+]], v[{{[0-9]+:[0-9]+}}], v{{[0-9]+}} glc{{$}}
+; GCN: buffer_store_dword [[RET]]
 define void @atomic_min_i32_ret_addr64_offset(i32 addrspace(1)* %out, i32 addrspace(1)* %out2, i32 %in, i64 %index) {
 entry:
   %ptr = getelementptr i32, i32 addrspace(1)* %out, i64 %index
   %gep = getelementptr i32, i32 addrspace(1)* %ptr, i32 4
   %0  = atomicrmw volatile min i32 addrspace(1)* %gep, i32 %in seq_cst
   store i32 %0, i32 addrspace(1)* %out2
   ret void
 }
 
 ; FUNC-LABEL: {{^}}atomic_min_i32:
-; SI: buffer_atomic_smin v{{[0-9]+}}, s[{{[0-9]+}}:{{[0-9]+}}], 0{{$}}
+; GCN: buffer_atomic_smin v{{[0-9]+}}, s[{{[0-9]+}}:{{[0-9]+}}], 0{{$}}
 define void @atomic_min_i32(i32 addrspace(1)* %out, i32 %in) {
 entry:
   %0  = atomicrmw volatile min i32 addrspace(1)* %out, i32 %in seq_cst
   ret void
 }
 
 ; FUNC-LABEL: {{^}}atomic_min_i32_ret:
-; SI: buffer_atomic_smin [[RET:v[0-9]+]], s[{{[0-9]+}}:{{[0-9]+}}], 0 glc
-; SI: buffer_store_dword [[RET]]
+; GCN: buffer_atomic_smin [[RET:v[0-9]+]], s[{{[0-9]+}}:{{[0-9]+}}], 0 glc
+; GCN: buffer_store_dword [[RET]]
 define void @atomic_min_i32_ret(i32 addrspace(1)* %out, i32 addrspace(1)* %out2, i32 %in) {
 entry:
   %0  = atomicrmw volatile min i32 addrspace(1)* %out, i32 %in seq_cst
   store i32 %0, i32 addrspace(1)* %out2
   ret void
 }
 
 ; FUNC-LABEL: {{^}}atomic_min_i32_addr64:
 ; SI: buffer_atomic_smin v{{[0-9]+}}, v[{{[0-9]+}}:{{[0-9]+}}], s[{{[0-9]+}}:{{[0-9]+}}], 0 addr64{{$}}
+; VI: s_movk_i32 flat_scratch_lo, 0x0
+; VI: s_movk_i32 flat_scratch_hi, 0x0
+; VI: flat_atomic_smin v[{{[0-9]+:[0-9]+}}], v{{[0-9]+$}}
 define void @atomic_min_i32_addr64(i32 addrspace(1)* %out, i32 %in, i64 %index) {
 entry:
   %ptr = getelementptr i32, i32 addrspace(1)* %out, i64 %index
   %0  = atomicrmw volatile min i32 addrspace(1)* %ptr, i32 %in seq_cst
   ret void
 }
 
 ; FUNC-LABEL: {{^}}atomic_min_i32_ret_addr64:
 ; SI: buffer_atomic_smin [[RET:v[0-9]+]], v[{{[0-9]+}}:{{[0-9]+}}], s[{{[0-9]+}}:{{[0-9]+}}], 0 addr64 glc{{$}}
-; SI: buffer_store_dword [[RET]]
+; VI: s_movk_i32 flat_scratch_lo, 0x0
+; VI: s_movk_i32 flat_scratch_hi, 0x0
+; VI: flat_atomic_smin [[RET:v[0-9]+]], v[{{[0-9]+:[0-9]+}}], v{{[0-9]+}} glc{{$}}
+; GCN: buffer_store_dword [[RET]]
 define void @atomic_min_i32_ret_addr64(i32 addrspace(1)* %out, i32 addrspace(1)* %out2, i32 %in, i64 %index) {
 entry:
   %ptr = getelementptr i32, i32 addrspace(1)* %out, i64 %index
   %0  = atomicrmw volatile min i32 addrspace(1)* %ptr, i32 %in seq_cst
   store i32 %0, i32 addrspace(1)* %out2
   ret void
 }
 
 ; FUNC-LABEL: {{^}}atomic_umin_i32_offset:
-; SI: buffer_atomic_umin v{{[0-9]+}}, s[{{[0-9]+}}:{{[0-9]+}}], 0 offset:16{{$}}
+; GCN: buffer_atomic_umin v{{[0-9]+}}, s[{{[0-9]+}}:{{[0-9]+}}], 0 offset:16{{$}}
 define void @atomic_umin_i32_offset(i32 addrspace(1)* %out, i32 %in) {
 entry:
   %gep = getelementptr i32, i32 addrspace(1)* %out, i32 4
   %0  = atomicrmw volatile umin i32 addrspace(1)* %gep, i32 %in seq_cst
   ret void
 }
 
 ; FUNC-LABEL: {{^}}atomic_umin_i32_ret_offset:
-; SI: buffer_atomic_umin [[RET:v[0-9]+]], s[{{[0-9]+}}:{{[0-9]+}}], 0 offset:16 glc {{$}}
-; SI: buffer_store_dword [[RET]]
+; GCN: buffer_atomic_umin [[RET:v[0-9]+]], s[{{[0-9]+}}:{{[0-9]+}}], 0 offset:16 glc {{$}}
+; GCN: buffer_store_dword [[RET]]
 define void @atomic_umin_i32_ret_offset(i32 addrspace(1)* %out, i32 addrspace(1)* %out2, i32 %in) {
 entry:
   %gep = getelementptr i32, i32 addrspace(1)* %out, i32 4
   %0  = atomicrmw volatile umin i32 addrspace(1)* %gep, i32 %in seq_cst
   store i32 %0, i32 addrspace(1)* %out2
   ret void
 }
 
 ; FUNC-LABEL: {{^}}atomic_umin_i32_addr64_offset:
 ; SI: buffer_atomic_umin v{{[0-9]+}}, v[{{[0-9]+}}:{{[0-9]+}}], s[{{[0-9]+}}:{{[0-9]+}}], 0 addr64 offset:16{{$}}
+; VI: s_movk_i32 flat_scratch_lo, 0x0
+; VI: s_movk_i32 flat_scratch_hi, 0x0
+; VI: flat_atomic_umin v[{{[0-9]+:[0-9]+}}], v{{[0-9]+$}}
 define void @atomic_umin_i32_addr64_offset(i32 addrspace(1)* %out, i32 %in, i64 %index) {
 entry:
   %ptr = getelementptr i32, i32 addrspace(1)* %out, i64 %index
   %gep = getelementptr i32, i32 addrspace(1)* %ptr, i32 4
   %0  = atomicrmw volatile umin i32 addrspace(1)* %gep, i32 %in seq_cst
   ret void
 }
 
 ; FUNC-LABEL: {{^}}atomic_umin_i32_ret_addr64_offset:
 ; SI: buffer_atomic_umin [[RET:v[0-9]+]], v[{{[0-9]+}}:{{[0-9]+}}], s[{{[0-9]+}}:{{[0-9]+}}], 0 addr64 offset:16 glc{{$}}
-; SI: buffer_store_dword [[RET]]
+; VI: s_movk_i32 flat_scratch_lo, 0x0
+; VI: s_movk_i32 flat_scratch_hi, 0x0
+; VI: flat_atomic_umin [[RET:v[0-9]+]], v[{{[0-9]+:[0-9]+}}], v{{[0-9]+}} glc{{$}}
+; GCN: buffer_store_dword [[RET]]
 define void @atomic_umin_i32_ret_addr64_offset(i32 addrspace(1)* %out, i32 addrspace(1)* %out2, i32 %in, i64 %index) {
 entry:
   %ptr = getelementptr i32, i32 addrspace(1)* %out, i64 %index
   %gep = getelementptr i32, i32 addrspace(1)* %ptr, i32 4
   %0  = atomicrmw volatile umin i32 addrspace(1)* %gep, i32 %in seq_cst
   store i32 %0, i32 addrspace(1)* %out2
   ret void
 }
 
 ; FUNC-LABEL: {{^}}atomic_umin_i32:
-; SI: buffer_atomic_umin v{{[0-9]+}}, s[{{[0-9]+}}:{{[0-9]+}}], 0{{$}}
+; GCN: buffer_atomic_umin v{{[0-9]+}}, s[{{[0-9]+}}:{{[0-9]+}}], 0{{$}}
 define void @atomic_umin_i32(i32 addrspace(1)* %out, i32 %in) {
 entry:
   %0  = atomicrmw volatile umin i32 addrspace(1)* %out, i32 %in seq_cst
   ret void
 }
 
 ; FUNC-LABEL: {{^}}atomic_umin_i32_ret:
 ; SI: buffer_atomic_umin [[RET:v[0-9]+]], s[{{[0-9]+}}:{{[0-9]+}}], 0 glc
-; SI: buffer_store_dword [[RET]]
+; GCN: buffer_store_dword [[RET]]
 define void @atomic_umin_i32_ret(i32 addrspace(1)* %out, i32 addrspace(1)* %out2, i32 %in) {
 entry:
   %0  = atomicrmw volatile umin i32 addrspace(1)* %out, i32 %in seq_cst
   store i32 %0, i32 addrspace(1)* %out2
   ret void
 }
 
 ; FUNC-LABEL: {{^}}atomic_umin_i32_addr64:
 ; SI: buffer_atomic_umin v{{[0-9]+}}, v[{{[0-9]+}}:{{[0-9]+}}], s[{{[0-9]+}}:{{[0-9]+}}], 0 addr64{{$}}
+; VI: s_movk_i32 flat_scratch_lo, 0x0
+; VI: s_movk_i32 flat_scratch_hi, 0x0
+; VI: flat_atomic_umin v[{{[0-9]+:[0-9]+}}], v{{[0-9]+$}}
 define void @atomic_umin_i32_addr64(i32 addrspace(1)* %out, i32 %in, i64 %index) {
 entry:
   %ptr = getelementptr i32, i32 addrspace(1)* %out, i64 %index
   %0  = atomicrmw volatile umin i32 addrspace(1)* %ptr, i32 %in seq_cst
   ret void
 }
 
 ; FUNC-LABEL: {{^}}atomic_umin_i32_ret_addr64:
 ; SI: buffer_atomic_umin [[RET:v[0-9]+]], v[{{[0-9]+}}:{{[0-9]+}}], s[{{[0-9]+}}:{{[0-9]+}}], 0 addr64 glc{{$}}
-; SI: buffer_store_dword [[RET]]
+; VI: s_movk_i32 flat_scratch_lo, 0x0
+; VI: s_movk_i32 flat_scratch_hi, 0x0
+; VI: flat_atomic_umin [[RET:v[0-9]+]], v[{{[0-9]+:[0-9]+}}], v{{[0-9]+}} glc{{$}}
+; GCN: buffer_store_dword [[RET]]
 define void @atomic_umin_i32_ret_addr64(i32 addrspace(1)* %out, i32 addrspace(1)* %out2, i32 %in, i64 %index) {
 entry:
   %ptr = getelementptr i32, i32 addrspace(1)* %out, i64 %index
   %0  = atomicrmw volatile umin i32 addrspace(1)* %ptr, i32 %in seq_cst
   store i32 %0, i32 addrspace(1)* %out2
   ret void
 }
 
 ; FUNC-LABEL: {{^}}atomic_or_i32_offset:
-; SI: buffer_atomic_or v{{[0-9]+}}, s[{{[0-9]+}}:{{[0-9]+}}], 0 offset:16{{$}}
+; GCN: buffer_atomic_or v{{[0-9]+}}, s[{{[0-9]+}}:{{[0-9]+}}], 0 offset:16{{$}}
 define void @atomic_or_i32_offset(i32 addrspace(1)* %out, i32 %in) {
 entry:
   %gep = getelementptr i32, i32 addrspace(1)* %out, i32 4
   %0  = atomicrmw volatile or i32 addrspace(1)* %gep, i32 %in seq_cst
   ret void
 }
 
 ; FUNC-LABEL: {{^}}atomic_or_i32_ret_offset:
-; SI: buffer_atomic_or [[RET:v[0-9]+]], s[{{[0-9]+}}:{{[0-9]+}}], 0 offset:16 glc {{$}}
-; SI: buffer_store_dword [[RET]]
+; GCN: buffer_atomic_or [[RET:v[0-9]+]], s[{{[0-9]+}}:{{[0-9]+}}], 0 offset:16 glc {{$}}
+; GCN: buffer_store_dword [[RET]]
 define void @atomic_or_i32_ret_offset(i32 addrspace(1)* %out, i32 addrspace(1)* %out2, i32 %in) {
 entry:
   %gep = getelementptr i32, i32 addrspace(1)* %out, i32 4
   %0  = atomicrmw volatile or i32 addrspace(1)* %gep, i32 %in seq_cst
   store i32 %0, i32 addrspace(1)* %out2
   ret void
 }
 
 ; FUNC-LABEL: {{^}}atomic_or_i32_addr64_offset:
 ; SI: buffer_atomic_or v{{[0-9]+}}, v[{{[0-9]+}}:{{[0-9]+}}], s[{{[0-9]+}}:{{[0-9]+}}], 0 addr64 offset:16{{$}}
+; VI: s_movk_i32 flat_scratch_lo, 0x0
+; VI: s_movk_i32 flat_scratch_hi, 0x0
+; VI: flat_atomic_or v[{{[0-9]+:[0-9]+}}], v{{[0-9]+$}}
 define void @atomic_or_i32_addr64_offset(i32 addrspace(1)* %out, i32 %in, i64 %index) {
 entry:
   %ptr = getelementptr i32, i32 addrspace(1)* %out, i64 %index
   %gep = getelementptr i32, i32 addrspace(1)* %ptr, i32 4
   %0  = atomicrmw volatile or i32 addrspace(1)* %gep, i32 %in seq_cst
   ret void
 }
 
 ; FUNC-LABEL: {{^}}atomic_or_i32_ret_addr64_offset:
 ; SI: buffer_atomic_or [[RET:v[0-9]+]], v[{{[0-9]+}}:{{[0-9]+}}], s[{{[0-9]+}}:{{[0-9]+}}], 0 addr64 offset:16 glc{{$}}
-; SI: buffer_store_dword [[RET]]
+; VI: s_movk_i32 flat_scratch_lo, 0x0
+; VI: s_movk_i32 flat_scratch_hi, 0x0
+; VI: flat_atomic_or [[RET:v[0-9]+]], v[{{[0-9]+:[0-9]+}}], v{{[0-9]+}} glc{{$}}
+; GCN: buffer_store_dword [[RET]]
 define void @atomic_or_i32_ret_addr64_offset(i32 addrspace(1)* %out, i32 addrspace(1)* %out2, i32 %in, i64 %index) {
 entry:
   %ptr = getelementptr i32, i32 addrspace(1)* %out, i64 %index
   %gep = getelementptr i32, i32 addrspace(1)* %ptr, i32 4
   %0  = atomicrmw volatile or i32 addrspace(1)* %gep, i32 %in seq_cst
   store i32 %0, i32 addrspace(1)* %out2
   ret void
 }
 
 ; FUNC-LABEL: {{^}}atomic_or_i32:
-; SI: buffer_atomic_or v{{[0-9]+}}, s[{{[0-9]+}}:{{[0-9]+}}], 0{{$}}
+; GCN: buffer_atomic_or v{{[0-9]+}}, s[{{[0-9]+}}:{{[0-9]+}}], 0{{$}}
 define void @atomic_or_i32(i32 addrspace(1)* %out, i32 %in) {
 entry:
   %0  = atomicrmw volatile or i32 addrspace(1)* %out, i32 %in seq_cst
   ret void
 }
 
 ; FUNC-LABEL: {{^}}atomic_or_i32_ret:
-; SI: buffer_atomic_or [[RET:v[0-9]+]], s[{{[0-9]+}}:{{[0-9]+}}], 0 glc
-; SI: buffer_store_dword [[RET]]
+; GCN: buffer_atomic_or [[RET:v[0-9]+]], s[{{[0-9]+}}:{{[0-9]+}}], 0 glc
+; GCN: buffer_store_dword [[RET]]
 define void @atomic_or_i32_ret(i32 addrspace(1)* %out, i32 addrspace(1)* %out2, i32 %in) {
 entry:
   %0  = atomicrmw volatile or i32 addrspace(1)* %out, i32 %in seq_cst
   store i32 %0, i32 addrspace(1)* %out2
   ret void
 }
 
 ; FUNC-LABEL: {{^}}atomic_or_i32_addr64:
 ; SI: buffer_atomic_or v{{[0-9]+}}, v[{{[0-9]+}}:{{[0-9]+}}], s[{{[0-9]+}}:{{[0-9]+}}], 0 addr64{{$}}
+; VI: s_movk_i32 flat_scratch_lo, 0x0
+; VI: s_movk_i32 flat_scratch_hi, 0x0
+; VI: flat_atomic_or v[{{[0-9]+:[0-9]+}}], v{{[0-9]+$}}
 define void @atomic_or_i32_addr64(i32 addrspace(1)* %out, i32 %in, i64 %index) {
 entry:
   %ptr = getelementptr i32, i32 addrspace(1)* %out, i64 %index
   %0  = atomicrmw volatile or i32 addrspace(1)* %ptr, i32 %in seq_cst
   ret void
 }
 
 ; FUNC-LABEL: {{^}}atomic_or_i32_ret_addr64:
 ; SI: buffer_atomic_or [[RET:v[0-9]+]], v[{{[0-9]+}}:{{[0-9]+}}], s[{{[0-9]+}}:{{[0-9]+}}], 0 addr64 glc{{$}}
-; SI: buffer_store_dword [[RET]]
+; VI: s_movk_i32 flat_scratch_lo, 0x0
+; VI: s_movk_i32 flat_scratch_hi, 0x0
+; VI: flat_atomic_or [[RET:v[0-9]+]], v[{{[0-9]+:[0-9]+}}], v{{[0-9]+}} glc{{$}}
+; GCN: buffer_store_dword [[RET]]
 define void @atomic_or_i32_ret_addr64(i32 addrspace(1)* %out, i32 addrspace(1)* %out2, i32 %in, i64 %index) {
 entry:
   %ptr = getelementptr i32, i32 addrspace(1)* %out, i64 %index
   %0  = atomicrmw volatile or i32 addrspace(1)* %ptr, i32 %in seq_cst
   store i32 %0, i32 addrspace(1)* %out2
   ret void
 }
 
 ; FUNC-LABEL: {{^}}atomic_xchg_i32_offset:
-; SI: buffer_atomic_swap v{{[0-9]+}}, s[{{[0-9]+}}:{{[0-9]+}}], 0 offset:16{{$}}
+; GCN: buffer_atomic_swap v{{[0-9]+}}, s[{{[0-9]+}}:{{[0-9]+}}], 0 offset:16{{$}}
 define void @atomic_xchg_i32_offset(i32 addrspace(1)* %out, i32 %in) {
 entry:
   %gep = getelementptr i32, i32 addrspace(1)* %out, i32 4
   %0  = atomicrmw volatile xchg i32 addrspace(1)* %gep, i32 %in seq_cst
   ret void
 }
 
 ; FUNC-LABEL: {{^}}atomic_xchg_i32_ret_offset:
-; SI: buffer_atomic_swap [[RET:v[0-9]+]], s[{{[0-9]+}}:{{[0-9]+}}], 0 offset:16 glc {{$}}
-; SI: buffer_store_dword [[RET]]
+; GCN: buffer_atomic_swap [[RET:v[0-9]+]], s[{{[0-9]+}}:{{[0-9]+}}], 0 offset:16 glc {{$}}
+; GCN: buffer_store_dword [[RET]]
 define void @atomic_xchg_i32_ret_offset(i32 addrspace(1)* %out, i32 addrspace(1)* %out2, i32 %in) {
 entry:
   %gep = getelementptr i32, i32 addrspace(1)* %out, i32 4
   %0  = atomicrmw volatile xchg i32 addrspace(1)* %gep, i32 %in seq_cst
   store i32 %0, i32 addrspace(1)* %out2
   ret void
 }
 
 ; FUNC-LABEL: {{^}}atomic_xchg_i32_addr64_offset:
 ; SI: buffer_atomic_swap v{{[0-9]+}}, v[{{[0-9]+}}:{{[0-9]+}}], s[{{[0-9]+}}:{{[0-9]+}}], 0 addr64 offset:16{{$}}
 define void @atomic_xchg_i32_addr64_offset(i32 addrspace(1)* %out, i32 %in, i64 %index) {
 entry:
   %ptr = getelementptr i32, i32 addrspace(1)* %out, i64 %index
   %gep = getelementptr i32, i32 addrspace(1)* %ptr, i32 4
   %0  = atomicrmw volatile xchg i32 addrspace(1)* %gep, i32 %in seq_cst
   ret void
 }
 
 ; FUNC-LABEL: {{^}}atomic_xchg_i32_ret_addr64_offset:
 ; SI: buffer_atomic_swap [[RET:v[0-9]+]], v[{{[0-9]+}}:{{[0-9]+}}], s[{{[0-9]+}}:{{[0-9]+}}], 0 addr64 offset:16 glc{{$}}
-; SI: buffer_store_dword [[RET]]
+; VI: s_movk_i32 flat_scratch_lo, 0x0
+; VI: s_movk_i32 flat_scratch_hi, 0x0
+; VI: flat_atomic_swap [[RET:v[0-9]+]], v[{{[0-9]+:[0-9]+}}], v{{[0-9]+}} glc{{$}}
+; GCN: buffer_store_dword [[RET]]
 define void @atomic_xchg_i32_ret_addr64_offset(i32 addrspace(1)* %out, i32 addrspace(1)* %out2, i32 %in, i64 %index) {
 entry:
   %ptr = getelementptr i32, i32 addrspace(1)* %out, i64 %index
   %gep = getelementptr i32, i32 addrspace(1)* %ptr, i32 4
   %0  = atomicrmw volatile xchg i32 addrspace(1)* %gep, i32 %in seq_cst
   store i32 %0, i32 addrspace(1)* %out2
   ret void
 }
 
 ; FUNC-LABEL: {{^}}atomic_xchg_i32:
-; SI: buffer_atomic_swap v{{[0-9]+}}, s[{{[0-9]+}}:{{[0-9]+}}], 0{{$}}
+; GCN: buffer_atomic_swap v{{[0-9]+}}, s[{{[0-9]+}}:{{[0-9]+}}], 0{{$}}
 define void @atomic_xchg_i32(i32 addrspace(1)* %out, i32 %in) {
 entry:
   %0  = atomicrmw volatile xchg i32 addrspace(1)* %out, i32 %in seq_cst
   ret void
 }
 
 ; FUNC-LABEL: {{^}}atomic_xchg_i32_ret:
-; SI: buffer_atomic_swap [[RET:v[0-9]+]], s[{{[0-9]+}}:{{[0-9]+}}], 0 glc
-; SI: buffer_store_dword [[RET]]
+; GCN: buffer_atomic_swap [[RET:v[0-9]+]], s[{{[0-9]+}}:{{[0-9]+}}], 0 glc
+; GCN: buffer_store_dword [[RET]]
 define void @atomic_xchg_i32_ret(i32 addrspace(1)* %out, i32 addrspace(1)* %out2, i32 %in) {
 entry:
   %0  = atomicrmw volatile xchg i32 addrspace(1)* %out, i32 %in seq_cst
   store i32 %0, i32 addrspace(1)* %out2
   ret void
 }
 
 ; FUNC-LABEL: {{^}}atomic_xchg_i32_addr64:
 ; SI: buffer_atomic_swap v{{[0-9]+}}, v[{{[0-9]+}}:{{[0-9]+}}], s[{{[0-9]+}}:{{[0-9]+}}], 0 addr64{{$}}
+; VI: s_movk_i32 flat_scratch_lo, 0x0
+; VI: s_movk_i32 flat_scratch_hi, 0x0
+; VI: flat_atomic_swap v[{{[0-9]+:[0-9]+}}], v{{[0-9]+$}}
 define void @atomic_xchg_i32_addr64(i32 addrspace(1)* %out, i32 %in, i64 %index) {
 entry:
   %ptr = getelementptr i32, i32 addrspace(1)* %out, i64 %index
   %0  = atomicrmw volatile xchg i32 addrspace(1)* %ptr, i32 %in seq_cst
   ret void
 }
 
 ; FUNC-LABEL: {{^}}atomic_xchg_i32_ret_addr64:
 ; SI: buffer_atomic_swap [[RET:v[0-9]+]], v[{{[0-9]+}}:{{[0-9]+}}], s[{{[0-9]+}}:{{[0-9]+}}], 0 addr64 glc{{$}}
-; SI: buffer_store_dword [[RET]]
+; VI: s_movk_i32 flat_scratch_lo, 0x0
+; VI: s_movk_i32 flat_scratch_hi, 0x0
+; VI: flat_atomic_swap [[RET:v[0-9]+]],  v[{{[0-9]+:[0-9]+}}], v{{[0-9]+}} glc{{$}}
+; GCN: buffer_store_dword [[RET]]
 define void @atomic_xchg_i32_ret_addr64(i32 addrspace(1)* %out, i32 addrspace(1)* %out2, i32 %in, i64 %index) {
 entry:
   %ptr = getelementptr i32, i32 addrspace(1)* %out, i64 %index
   %0  = atomicrmw volatile xchg i32 addrspace(1)* %ptr, i32 %in seq_cst
   store i32 %0, i32 addrspace(1)* %out2
   ret void
 }
 
 ; FUNC-LABEL: {{^}}atomic_xor_i32_offset:
-; SI: buffer_atomic_xor v{{[0-9]+}}, s[{{[0-9]+}}:{{[0-9]+}}], 0 offset:16{{$}}
+; GCN: buffer_atomic_xor v{{[0-9]+}}, s[{{[0-9]+}}:{{[0-9]+}}], 0 offset:16{{$}}
 define void @atomic_xor_i32_offset(i32 addrspace(1)* %out, i32 %in) {
 entry:
   %gep = getelementptr i32, i32 addrspace(1)* %out, i32 4
   %0  = atomicrmw volatile xor i32 addrspace(1)* %gep, i32 %in seq_cst
   ret void
 }
 
 ; FUNC-LABEL: {{^}}atomic_xor_i32_ret_offset:
-; SI: buffer_atomic_xor [[RET:v[0-9]+]], s[{{[0-9]+}}:{{[0-9]+}}], 0 offset:16 glc {{$}}
-; SI: buffer_store_dword [[RET]]
+; GCN: buffer_atomic_xor [[RET:v[0-9]+]], s[{{[0-9]+}}:{{[0-9]+}}], 0 offset:16 glc {{$}}
+; GCN: buffer_store_dword [[RET]]
 define void @atomic_xor_i32_ret_offset(i32 addrspace(1)* %out, i32 addrspace(1)* %out2, i32 %in) {
 entry:
   %gep = getelementptr i32, i32 addrspace(1)* %out, i32 4
   %0  = atomicrmw volatile xor i32 addrspace(1)* %gep, i32 %in seq_cst
   store i32 %0, i32 addrspace(1)* %out2
   ret void
 }
 
 ; FUNC-LABEL: {{^}}atomic_xor_i32_addr64_offset:
 ; SI: buffer_atomic_xor v{{[0-9]+}}, v[{{[0-9]+}}:{{[0-9]+}}], s[{{[0-9]+}}:{{[0-9]+}}], 0 addr64 offset:16{{$}}
+; VI: s_movk_i32 flat_scratch_lo, 0x0
+; VI: s_movk_i32 flat_scratch_hi, 0x0
+; VI: flat_atomic_xor v[{{[0-9]+:[0-9]+}}], v{{[0-9]+$}}
 define void @atomic_xor_i32_addr64_offset(i32 addrspace(1)* %out, i32 %in, i64 %index) {
 entry:
   %ptr = getelementptr i32, i32 addrspace(1)* %out, i64 %index
   %gep = getelementptr i32, i32 addrspace(1)* %ptr, i32 4
   %0  = atomicrmw volatile xor i32 addrspace(1)* %gep, i32 %in seq_cst
   ret void
 }
 
 ; FUNC-LABEL: {{^}}atomic_xor_i32_ret_addr64_offset:
 ; SI: buffer_atomic_xor [[RET:v[0-9]+]], v[{{[0-9]+}}:{{[0-9]+}}], s[{{[0-9]+}}:{{[0-9]+}}], 0 addr64 offset:16 glc{{$}}
-; SI: buffer_store_dword [[RET]]
+; VI: s_movk_i32 flat_scratch_lo, 0x0
+; VI: s_movk_i32 flat_scratch_hi, 0x0
+; VI: flat_atomic_xor [[RET:v[0-9]+]], v[{{[0-9]+:[0-9]+}}], v{{[0-9]+}} glc{{$}}
+; GCN: buffer_store_dword [[RET]]
 define void @atomic_xor_i32_ret_addr64_offset(i32 addrspace(1)* %out, i32 addrspace(1)* %out2, i32 %in, i64 %index) {
 entry:
   %ptr = getelementptr i32, i32 addrspace(1)* %out, i64 %index
   %gep = getelementptr i32, i32 addrspace(1)* %ptr, i32 4
   %0  = atomicrmw volatile xor i32 addrspace(1)* %gep, i32 %in seq_cst
   store i32 %0, i32 addrspace(1)* %out2
   ret void
 }
 
 ; FUNC-LABEL: {{^}}atomic_xor_i32:
-; SI: buffer_atomic_xor v{{[0-9]+}}, s[{{[0-9]+}}:{{[0-9]+}}], 0{{$}}
+; GCN: buffer_atomic_xor v{{[0-9]+}}, s[{{[0-9]+}}:{{[0-9]+}}], 0{{$}}
 define void @atomic_xor_i32(i32 addrspace(1)* %out, i32 %in) {
 entry:
   %0  = atomicrmw volatile xor i32 addrspace(1)* %out, i32 %in seq_cst
   ret void
 }
 
 ; FUNC-LABEL: {{^}}atomic_xor_i32_ret:
-; SI: buffer_atomic_xor [[RET:v[0-9]+]], s[{{[0-9]+}}:{{[0-9]+}}], 0 glc
-; SI: buffer_store_dword [[RET]]
+; GCN: buffer_atomic_xor [[RET:v[0-9]+]], s[{{[0-9]+}}:{{[0-9]+}}], 0 glc
+; GCN: buffer_store_dword [[RET]]
 define void @atomic_xor_i32_ret(i32 addrspace(1)* %out, i32 addrspace(1)* %out2, i32 %in) {
 entry:
   %0  = atomicrmw volatile xor i32 addrspace(1)* %out, i32 %in seq_cst
   store i32 %0, i32 addrspace(1)* %out2
   ret void
 }
 
 ; FUNC-LABEL: {{^}}atomic_xor_i32_addr64:
 ; SI: buffer_atomic_xor v{{[0-9]+}}, v[{{[0-9]+}}:{{[0-9]+}}], s[{{[0-9]+}}:{{[0-9]+}}], 0 addr64{{$}}
+; VI: s_movk_i32 flat_scratch_lo, 0x0
+; VI: s_movk_i32 flat_scratch_hi, 0x0
+; VI: flat_atomic_xor v[{{[0-9]+:[0-9]+}}], v{{[0-9]+$}}
 define void @atomic_xor_i32_addr64(i32 addrspace(1)* %out, i32 %in, i64 %index) {
 entry:
   %ptr = getelementptr i32, i32 addrspace(1)* %out, i64 %index
   %0  = atomicrmw volatile xor i32 addrspace(1)* %ptr, i32 %in seq_cst
   ret void
 }
 
 ; FUNC-LABEL: {{^}}atomic_xor_i32_ret_addr64:
 ; SI: buffer_atomic_xor [[RET:v[0-9]+]], v[{{[0-9]+}}:{{[0-9]+}}], s[{{[0-9]+}}:{{[0-9]+}}], 0 addr64 glc{{$}}
-; SI: buffer_store_dword [[RET]]
+; VI: s_movk_i32 flat_scratch_lo, 0x0
+; VI: s_movk_i32 flat_scratch_hi, 0x0
+; VI: flat_atomic_xor [[RET:v[0-9]+]], v[{{[0-9]+:[0-9]+}}], v{{[0-9]+}} glc{{$}}
+; GCN: buffer_store_dword [[RET]]
 define void @atomic_xor_i32_ret_addr64(i32 addrspace(1)* %out, i32 addrspace(1)* %out2, i32 %in, i64 %index) {
 entry:
   %ptr = getelementptr i32, i32 addrspace(1)* %out, i64 %index
   %0  = atomicrmw volatile xor i32 addrspace(1)* %ptr, i32 %in seq_cst
   store i32 %0, i32 addrspace(1)* %out2
   ret void
 }