AMDGPU: Set flat_scratch from flat_scratch_init reg

[arsenm]

Summary

This was hardcoded to the static private size, but this
would be missing the offset and additional size for someday
when we have dynamic sizing.

Also stops always initializing flat_scratch even when unused.

In the future we should stop emitting this unless flat instructions
are used to access private memory. For example this will initialize
it almost always on VI because flat is used for global access.

[• tstellarAMD]

LGTM.

lib/Target/AMDGPU/SIFrameLowering.cpp

78

I'm not sure if this comment was intentional, but I think this should be dropped.

[arsenm]

r260658

lib/Target/AMDGPU/SIFrameLowering.cpp

78

The condition should be there, but right now it doesn't do anything because flat instructions are detected in SILowerControlFlow

lib/Target/AMDGPU/SIFrameLowering.cpp

 #include "llvm/CodeGen/MachineInstrBuilder.h"
 #include "llvm/CodeGen/RegisterScavenging.h"
 
 using namespace llvm;
 
 
 static bool hasOnlySGPRSpills(const SIMachineFunctionInfo *FuncInfo,
                               const MachineFrameInfo *FrameInfo) {
-  if (!FuncInfo->hasSpilledSGPRs())
-    return false;
-
-  if (FuncInfo->hasSpilledVGPRs())
-    return false;
-
-  for (int I = FrameInfo->getObjectIndexBegin(),
-         E = FrameInfo->getObjectIndexEnd(); I != E; ++I) {
-    if (!FrameInfo->isSpillSlotObjectIndex(I))
-      return false;
-  }
-
-  return true;
+  return FuncInfo->hasSpilledSGPRs() &&
+    (!FuncInfo->hasSpilledVGPRs() && !FuncInfo->hasNonSpillStackObjects());
 }
 
 static ArrayRef<MCPhysReg> getAllSGPR128() {
   return makeArrayRef(AMDGPU::SReg_128RegClass.begin(),
                       AMDGPU::SReg_128RegClass.getNumRegs());
 }
 
 static ArrayRef<MCPhysReg> getAllSGPRs() {
void SIFrameLowering::emitPrologue(MachineFunction &MF,
   // somewhere.
   if (hasOnlySGPRSpills(MFI, MF.getFrameInfo()))
     return;
 
   const SIInstrInfo *TII =
       static_cast<const SIInstrInfo *>(MF.getSubtarget().getInstrInfo());
   const SIRegisterInfo *TRI = &TII->getRegisterInfo();
   const AMDGPUSubtarget &ST = MF.getSubtarget<AMDGPUSubtarget>();
+  MachineRegisterInfo &MRI = MF.getRegInfo();
+  MachineBasicBlock::iterator I = MBB.begin();
 
   // We need to insert initialization of the scratch resource descriptor.
   unsigned ScratchRsrcReg = MFI->getScratchRSrcReg();
   assert(ScratchRsrcReg != AMDGPU::NoRegister);
 
   unsigned ScratchWaveOffsetReg = MFI->getScratchWaveOffsetReg();
   assert(ScratchWaveOffsetReg != AMDGPU::NoRegister);
 
   unsigned PreloadedScratchWaveOffsetReg = TRI->getPreloadedValue(
     MF, SIRegisterInfo::PRIVATE_SEGMENT_WAVE_BYTE_OFFSET);
 
   unsigned PreloadedPrivateBufferReg = AMDGPU::NoRegister;
   if (ST.isAmdHsaOS()) {
     PreloadedPrivateBufferReg = TRI->getPreloadedValue(
       MF, SIRegisterInfo::PRIVATE_SEGMENT_BUFFER);
   }
 
+  if (MFI->hasFlatScratchInit() /*&& MFI->hasFlatInstructions()*/) {
+    // We don't need this if we only have spills since there is no user facing
+    // scratch.
+
+    // TODO: If we know we don't have flat instructions earlier, we can omit
+    // this from the input registers.
+    //
+    // TODO: We only need to know if we access scratch space through a flat
+    // pointer. Because we only detect if flat instructions are used at all,
+    // this will be used more often than necessary on VI.
+
+    DebugLoc DL;
+
+    unsigned FlatScratchInitReg
+      = TRI->getPreloadedValue(MF, SIRegisterInfo::FLAT_SCRATCH_INIT);
+
+    MRI.addLiveIn(FlatScratchInitReg);
+    MBB.addLiveIn(FlatScratchInitReg);
+
+    // Copy the size in bytes.
+    unsigned FlatScrInitHi = TRI->getSubReg(FlatScratchInitReg, AMDGPU::sub1);
+    BuildMI(MBB, I, DL, TII->get(AMDGPU::S_MOV_B32), AMDGPU::FLAT_SCR_LO)
+      .addReg(FlatScrInitHi, RegState::Kill);
+
+    unsigned FlatScrInitLo = TRI->getSubReg(FlatScratchInitReg, AMDGPU::sub0);
+
+    // Add wave offset in bytes to private base offset.
+    // See comment in AMDKernelCodeT.h for enable_sgpr_flat_scratch_init.
+    BuildMI(MBB, I, DL, TII->get(AMDGPU::S_ADD_U32), FlatScrInitLo)
+      .addReg(FlatScrInitLo)
+      .addReg(ScratchWaveOffsetReg);
+
+    // Convert offset to 256-byte units.
+    BuildMI(MBB, I, DL, TII->get(AMDGPU::S_LSHR_B32), AMDGPU::FLAT_SCR_HI)
+      .addReg(FlatScrInitLo, RegState::Kill)
+      .addImm(8);
+  }
+
   // If we reserved the original input registers, we don't need to copy to the
   // reserved registers.
   if (ScratchRsrcReg == PreloadedPrivateBufferReg) {
     // We should always reserve these 5 registers at the same time.
     assert(ScratchWaveOffsetReg == PreloadedScratchWaveOffsetReg &&
            "scratch wave offset and private segment buffer inconsistent");
     return;
   }
 
 
   // We added live-ins during argument lowering, but since they were not used
   // they were deleted. We're adding the uses now, so add them back.
-  MachineRegisterInfo &MRI = MF.getRegInfo();
   MRI.addLiveIn(PreloadedScratchWaveOffsetReg);
   MBB.addLiveIn(PreloadedScratchWaveOffsetReg);
 
   if (ST.isAmdHsaOS()) {
     MRI.addLiveIn(PreloadedPrivateBufferReg);
     MBB.addLiveIn(PreloadedPrivateBufferReg);
   }
 
if (ScratchWaveOffsetReg == TRI->reservedPrivateSegmentWaveByteOffsetReg(MF)) {
       }
     }
   }
 
 
   assert(!TRI->isSubRegister(ScratchRsrcReg, ScratchWaveOffsetReg));
 
   const MCInstrDesc &SMovB32 = TII->get(AMDGPU::S_MOV_B32);
-  MachineBasicBlock::iterator I = MBB.begin();
   DebugLoc DL;
 
   if (PreloadedScratchWaveOffsetReg != ScratchWaveOffsetReg) {
     // Make sure we emit the copy for the offset first. We may have chosen to copy
     // the buffer resource into a register that aliases the input offset register.
     BuildMI(MBB, I, DL, SMovB32, ScratchWaveOffsetReg)
       .addReg(PreloadedScratchWaveOffsetReg, RegState::Kill);
   }

lib/Target/AMDGPU/SIISelLowering.cpp

SDValue SITargetLowering::LowerFormalArguments(
   }
 
   if (Info->hasKernargSegmentPtr()) {
     unsigned InputPtrReg = Info->addKernargSegmentPtr(*TRI);
     MF.addLiveIn(InputPtrReg, &AMDGPU::SReg_64RegClass);
     CCInfo.AllocateReg(InputPtrReg);
   }
 
+  if (Info->hasFlatScratchInit()) {
+    unsigned FlatScratchInitReg = Info->addFlatScratchInit(*TRI);
+    MF.addLiveIn(FlatScratchInitReg, &AMDGPU::SReg_64RegClass);
+    CCInfo.AllocateReg(FlatScratchInitReg);
+  }
+
   AnalyzeFormalArguments(CCInfo, Splits);
 
   SmallVector<SDValue, 16> Chains;
 
   for (unsigned i = 0, e = Ins.size(), ArgIdx = 0; i != e; ++i) {
 
     const ISD::InputArg &Arg = Ins[i];
     if (Skipped[i]) {
unsigned PrivateSegmentWaveByteOffsetReg
       = Info->addPrivateSegmentWaveByteOffset();
 
     MF.addLiveIn(PrivateSegmentWaveByteOffsetReg, &AMDGPU::SGPR_32RegClass);
     CCInfo.AllocateReg(PrivateSegmentWaveByteOffsetReg);
   }
 
   // Now that we've figured out where the scratch register inputs are, see if
   // should reserve the arguments and use them directly.
-
   bool HasStackObjects = MF.getFrameInfo()->hasStackObjects();
+  // Record that we know we have non-spill stack objects so we don't need to
+  // check all stack objects later.
+  if (HasStackObjects)
+    Info->setHasNonSpillStackObjects(true);
 
   if (ST.isAmdHsaOS()) {
     // TODO: Assume we will spill without optimizations.
     if (HasStackObjects) {
       // If we have stack objects, we unquestionably need the private buffer
       // resource. For the HSA ABI, this will be the first 4 user SGPR
       // inputs. We can reserve those and use them directly.
 

lib/Target/AMDGPU/SILowerControlFlow.cpp

bool SILowerControlFlowPass::runOnMachineFunction(MachineFunction &MF) {
   }
 
   if (NeedWQM && MFI->getShaderType() == ShaderType::PIXEL) {
     MachineBasicBlock &MBB = MF.front();
     BuildMI(MBB, MBB.getFirstNonPHI(), DebugLoc(), TII->get(AMDGPU::S_WQM_B64),
             AMDGPU::EXEC).addReg(AMDGPU::EXEC);
   }
 
-  // FIXME: This seems inappropriate to do here.
   if (NeedFlat && MFI->IsKernel) {
-    // Insert the prologue initializing the SGPRs pointing to the scratch space
-    // for flat accesses.
-    const MachineFrameInfo *FrameInfo = MF.getFrameInfo();
-
     // TODO: What to use with function calls?
-
-    // FIXME: This is reporting stack size that is used in a scratch buffer
-    // rather than registers as well.
-    uint64_t StackSizeBytes = FrameInfo->getStackSize();
-
-    int IndirectBegin
-      = static_cast<const AMDGPUInstrInfo*>(TII)->getIndirectIndexBegin(MF);
-    // Convert register index to 256-byte unit.
-    uint64_t StackOffset = IndirectBegin < 0 ? 0 : (4 * IndirectBegin / 256);
-
-    assert((StackSizeBytes < 0xffff) && StackOffset < 0xffff &&
-           "Stack limits should be smaller than 16-bits");
-
-    // Initialize the flat scratch register pair.
-    // TODO: Can we use one s_mov_b64 here?
-
-    // Offset is in units of 256-bytes.
-    MachineBasicBlock &MBB = MF.front();
-    DebugLoc NoDL;
-    MachineBasicBlock::iterator Start = MBB.getFirstNonPHI();
-    const MCInstrDesc &SMovK = TII->get(AMDGPU::S_MOVK_I32);
-
-    assert(isInt<16>(StackOffset) && isInt<16>(StackSizeBytes));
-
-    BuildMI(MBB, Start, NoDL, SMovK, AMDGPU::FLAT_SCR_LO)
-      .addImm(StackOffset);
-
-    // Documentation says size is "per-thread scratch size in bytes"
-    BuildMI(MBB, Start, NoDL, SMovK, AMDGPU::FLAT_SCR_HI)
-      .addImm(StackSizeBytes);
+    // We will need to Initialize the flat scratch register pair.
+    if (NeedFlat)
+      MFI->setHasFlatInstructions(true);
   }
 
   return true;
 }

lib/Target/AMDGPU/SIMachineFunctionInfo.h

public:
   std::map<unsigned, unsigned> LaneVGPRs;
   unsigned ScratchOffsetReg;
   unsigned NumUserSGPRs;
   unsigned NumSystemSGPRs;
 
 private:
   bool HasSpilledSGPRs;
   bool HasSpilledVGPRs;
+  bool HasNonSpillStackObjects;
+  bool HasFlatInstructions;
 
   // Feature bits required for inputs passed in user SGPRs.
   bool PrivateSegmentBuffer : 1;
   bool DispatchPtr : 1;
   bool QueuePtr : 1;
   bool DispatchID : 1;
   bool KernargSegmentPtr : 1;
   bool FlatScratchInit : 1;
public:
   unsigned getTIDReg() const { return TIDReg; };
   void setTIDReg(unsigned Reg) { TIDReg = Reg; }
 
   // Add user SGPRs.
   unsigned addPrivateSegmentBuffer(const SIRegisterInfo &TRI);
   unsigned addDispatchPtr(const SIRegisterInfo &TRI);
   unsigned addQueuePtr(const SIRegisterInfo &TRI);
   unsigned addKernargSegmentPtr(const SIRegisterInfo &TRI);
+  unsigned addFlatScratchInit(const SIRegisterInfo &TRI);
 
   // Add system SGPRs.
   unsigned addWorkGroupIDX() {
     WorkGroupIDXSystemSGPR = getNextSystemSGPR();
     NumSystemSGPRs += 1;
     return WorkGroupIDXSystemSGPR;
   }
 
public:
   bool hasSpilledVGPRs() const {
     return HasSpilledVGPRs;
   }
 
   void setHasSpilledVGPRs(bool Spill = true) {
     HasSpilledVGPRs = Spill;
   }
 
+  bool hasNonSpillStackObjects() const {
+    return HasNonSpillStackObjects;
+  }
+
+  void setHasNonSpillStackObjects(bool StackObject = true) {
+    HasNonSpillStackObjects = StackObject;
+  }
+
+  bool hasFlatInstructions() const {
+    return HasFlatInstructions;
+  }
+
+  void setHasFlatInstructions(bool UseFlat = true) {
+    HasFlatInstructions = UseFlat;
+  }
+
   unsigned getPSInputAddr() const {
     return PSInputAddr;
   }
 
   bool isPSInputAllocated(unsigned Index) const {
     return PSInputAddr & (1 << Index);
   }
 

lib/Target/AMDGPU/SIMachineFunctionInfo.cpp

: AMDGPUMachineFunction(MF),
     PSInputAddr(0),
     ReturnsVoid(true),
     LDSWaveSpillSize(0),
     PSInputEna(0),
     NumUserSGPRs(0),
     NumSystemSGPRs(0),
     HasSpilledSGPRs(false),
     HasSpilledVGPRs(false),
+    HasNonSpillStackObjects(false),
+    HasFlatInstructions(false),
     PrivateSegmentBuffer(false),
     DispatchPtr(false),
     QueuePtr(false),
     DispatchID(false),
     KernargSegmentPtr(false),
     FlatScratchInit(false),
     GridWorkgroupCountX(false),
     GridWorkgroupCountY(false),
if (F->hasFnAttribute("amdgpu-work-group-id-z"))
     WorkGroupIDZ = true;
 
   if (F->hasFnAttribute("amdgpu-work-item-id-y"))
     WorkItemIDY = true;
 
   if (F->hasFnAttribute("amdgpu-work-item-id-z"))
     WorkItemIDZ = true;
 
+  // X, XY, and XYZ are the only supported combinations, so make sure Y is
+  // enabled if Z is.
+  if (WorkItemIDZ)
+    WorkItemIDY = true;
+
   bool MaySpill = ST.isVGPRSpillingEnabled(this);
   bool HasStackObjects = FrameInfo->hasStackObjects();
 
   if (HasStackObjects || MaySpill)
     PrivateSegmentWaveByteOffset = true;
 
   if (ST.isAmdHsaOS()) {
     if (HasStackObjects || MaySpill)
       PrivateSegmentBuffer = true;
 
     if (F->hasFnAttribute("amdgpu-dispatch-ptr"))
       DispatchPtr = true;
   }
 
-  // X, XY, and XYZ are the only supported combinations, so make sure Y is
-  // enabled if Z is.
-  if (WorkItemIDZ)
-    WorkItemIDY = true;
+  // We don't need to worry about accessing spills with flat instructions.
+  // TODO: On VI where we must use flat for global, we should be able to omit
+  // this if it is never used for generic access.
+  if (HasStackObjects && ST.getGeneration() >= AMDGPUSubtarget::SEA_ISLANDS &&
+      ST.isAmdHsaOS())
+    FlatScratchInit = true;
 }
 
 unsigned SIMachineFunctionInfo::addPrivateSegmentBuffer(
   const SIRegisterInfo &TRI) {
   PrivateSegmentBufferUserSGPR = TRI.getMatchingSuperReg(
     getNextUserSGPR(), AMDGPU::sub0, &AMDGPU::SReg_128RegClass);
   NumUserSGPRs += 4;
   return PrivateSegmentBufferUserSGPR;

 
 unsigned SIMachineFunctionInfo::addKernargSegmentPtr(const SIRegisterInfo &TRI) {
   KernargSegmentPtrUserSGPR = TRI.getMatchingSuperReg(
     getNextUserSGPR(), AMDGPU::sub0, &AMDGPU::SReg_64RegClass);
   NumUserSGPRs += 2;
   return KernargSegmentPtrUserSGPR;
 }
 
+unsigned SIMachineFunctionInfo::addFlatScratchInit(const SIRegisterInfo &TRI) {
+  FlatScratchInitUserSGPR = TRI.getMatchingSuperReg(
+    getNextUserSGPR(), AMDGPU::sub0, &AMDGPU::SReg_64RegClass);
+  NumUserSGPRs += 2;
+  return FlatScratchInitUserSGPR;
+}
+
 SIMachineFunctionInfo::SpilledReg SIMachineFunctionInfo::getSpilledReg(
                                                        MachineFunction *MF,
                                                        unsigned FrameIndex,
                                                        unsigned SubIdx) {
   const MachineFrameInfo *FrameInfo = MF->getFrameInfo();
   const SIRegisterInfo *TRI = static_cast<const SIRegisterInfo *>(
       MF->getSubtarget<AMDGPUSubtarget>().getRegisterInfo());
   MachineRegisterInfo &MRI = MF->getRegInfo();

lib/Target/AMDGPU/SIRegisterInfo.h

public:
 
   /// \returns True if operands defined with this operand type can accept
   /// an inline constant. i.e. An integer value in the range (-16, 64) or
   /// -4.0f, -2.0f, -1.0f, -0.5f, 0.0f, 0.5f, 1.0f, 2.0f, 4.0f.
   bool opCanUseInlineConstant(unsigned OpType) const;
 
   enum PreloadedValue {
     // SGPRS:
-    PRIVATE_SEGMENT_BUFFER =  0,
+    PRIVATE_SEGMENT_BUFFER = 0,
     DISPATCH_PTR        =  1,
     QUEUE_PTR           =  2,
     KERNARG_SEGMENT_PTR =  3,
+    DISPATCH_ID         =  4,
+    FLAT_SCRATCH_INIT   =  5,
     WORKGROUP_ID_X      = 10,
     WORKGROUP_ID_Y      = 11,
     WORKGROUP_ID_Z      = 12,
     PRIVATE_SEGMENT_WAVE_BYTE_OFFSET = 14,
 
     // VGPRS:
     FIRST_VGPR_VALUE    = 15,
     WORKITEM_ID_X       = FIRST_VGPR_VALUE,

lib/Target/AMDGPU/SIRegisterInfo.cpp

case SIRegisterInfo::PRIVATE_SEGMENT_WAVE_BYTE_OFFSET:
     return MFI->PrivateSegmentWaveByteOffsetSystemSGPR;
   case SIRegisterInfo::PRIVATE_SEGMENT_BUFFER:
     assert(ST.isAmdHsaOS() && "Non-HSA ABI currently uses relocations");
     assert(MFI->hasPrivateSegmentBuffer());
     return MFI->PrivateSegmentBufferUserSGPR;
   case SIRegisterInfo::KERNARG_SEGMENT_PTR:
     assert(MFI->hasKernargSegmentPtr());
     return MFI->KernargSegmentPtrUserSGPR;
+  case SIRegisterInfo::DISPATCH_ID:
+    llvm_unreachable("unimplemented");
+  case SIRegisterInfo::FLAT_SCRATCH_INIT:
+    assert(MFI->hasFlatScratchInit());
+    return MFI->FlatScratchInitUserSGPR;
   case SIRegisterInfo::DISPATCH_PTR:
     assert(MFI->hasDispatchPtr());
     return MFI->DispatchPtrUserSGPR;
   case SIRegisterInfo::QUEUE_PTR:
     llvm_unreachable("not implemented");
   case SIRegisterInfo::WORKITEM_ID_X:
     assert(MFI->hasWorkItemIDX());
     return AMDGPU::VGPR0;

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

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
 ; 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: {{^}}BB6_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

test/CodeGen/AMDGPU/ci-use-flat-for-global.ll

-; RUN: llc < %s -mtriple=amdgcn--amdhsa -mcpu=kaveri | FileCheck -check-prefix=HSA-DEFAULT %s
-; RUN: llc < %s -mtriple=amdgcn--amdhsa -mcpu=kaveri -mattr=-flat-for-global | FileCheck -check-prefix=HSA-NODEFAULT %s
-; RUN: llc < %s -mtriple=amdgcn -mcpu=kaveri | FileCheck -check-prefix=NOHSA-DEFAULT %s
-; RUN: llc < %s -mtriple=amdgcn -mcpu=kaveri -mattr=+flat-for-global | FileCheck -check-prefix=NOHSA-NODEFAULT %s
+; RUN: llc -mtriple=amdgcn--amdhsa -mcpu=kaveri -mattr=+flat-for-global < %s | FileCheck -check-prefix=HSA -check-prefix=HSA-DEFAULT -check-prefix=ALL %s
+; RUN: llc -mtriple=amdgcn--amdhsa -mcpu=kaveri -mattr=-flat-for-global < %s | FileCheck -check-prefix=HSA -check-prefix=HSA-NODEFAULT -check-prefix=ALL %s
+; RUN: llc -mtriple=amdgcn-- -mcpu=kaveri -mattr=-flat-for-global < %s | FileCheck -check-prefix=NOHSA-DEFAULT -check-prefix=ALL %s
+; RUN: llc -mtriple=amdgcn-- -mcpu=kaveri -mattr=+flat-for-global < %s | FileCheck -check-prefix=NOHSA-NODEFAULT -check-prefix=ALL %s
 
 
+; There are no stack objects even though flat is used by default, so
+; flat_scratch_init should be disabled.
+
+; ALL-LABEL: {{^}}test:
+; HSA: .amd_kernel_code_t
+; HSA: enable_sgpr_flat_scratch_init = 0
+; HSA: .end_amd_kernel_code_t
+
+; ALL-NOT: flat_scr
+
 ; HSA-DEFAULT: flat_store_dword
 ; HSA-NODEFAULT: buffer_store_dword
+
 ; NOHSA-DEFAULT: buffer_store_dword
 ; NOHSA-NODEFAULT: flat_store_dword
 define void @test(i32 addrspace(1)* %out) {
 entry:
   store i32 0, i32 addrspace(1)* %out
   ret void
 }

test/CodeGen/AMDGPU/global_atomics.ll

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{{$}}
-; VI: s_movk_i32 flat_scratch_lo, 0x0
-; VI: s_movk_i32 flat_scratch_hi, 0x0
 ; VI: flat_atomic_add [[RET:v[0-9]+]], v[{{[0-9]+:[0-9]+}}], v{{[0-9]+}} glc{{$}}
 ; 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

 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{{$}}
-; VI: s_movk_i32 flat_scratch_lo, 0x0
-; VI: s_movk_i32 flat_scratch_hi, 0x0
 ; VI: flat_atomic_add [[RET:v[0-9]+]], v[{{[0-9]+:[0-9]+}}], v{{[0-9]+}} glc{{$}}
 ; 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
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{{$}}
-; VI: s_movk_i32 flat_scratch_lo, 0x0
-; VI: s_movk_i32 flat_scratch_hi, 0x0
 ; VI: flat_atomic_and [[RET:v[0-9]]], v[{{[0-9]+:[0-9]+}}], v{{[0-9]+}} glc{{$}}
 ; 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

 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{{$}}
-; VI: s_movk_i32 flat_scratch_lo, 0x0
-; VI: s_movk_i32 flat_scratch_hi, 0x0
 ; VI: flat_atomic_and [[RET:v[0-9]+]], v[{{[0-9]+:[0-9]+}}], v{{[0-9]+}} glc{{$}}
 ; 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
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{{$}}
-; VI: s_movk_i32 flat_scratch_lo, 0x0
-; VI: s_movk_i32 flat_scratch_hi, 0x0
 ; VI: flat_atomic_sub [[RET:v[0-9]+]], v[{{[0-9]+:[0-9]+}}], v{{[0-9]+}} glc{{$}}
 ; 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

 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{{$}}
-; VI: s_movk_i32 flat_scratch_lo, 0x0
-; VI: s_movk_i32 flat_scratch_hi, 0x0
 ; VI: flat_atomic_sub [[RET:v[0-9]+]], v[{{[0-9]+:[0-9]+}}], v{{[0-9]+}} glc{{$}}
 ; 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
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{{$}}
-; VI: s_movk_i32 flat_scratch_lo, 0x0
-; VI: s_movk_i32 flat_scratch_hi, 0x0
 ; VI: flat_atomic_smax [[RET:v[0-9]+]], v[{{[0-9]+:[0-9]+}}], v{{[0-9]+}} glc{{$}}
 ; 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

 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{{$}}
-; VI: s_movk_i32 flat_scratch_lo, 0x0
-; VI: s_movk_i32 flat_scratch_hi, 0x0
 ; VI: flat_atomic_smax [[RET:v[0-9]+]], v[{{[0-9]+:[0-9]+}}], v{{[0-9]+}} glc{{$}}
 ; 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
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{{$}}
-; VI: s_movk_i32 flat_scratch_lo, 0x0
-; VI: s_movk_i32 flat_scratch_hi, 0x0
 ; VI: flat_atomic_umax [[RET:v[0-9]+]], v[{{[0-9]+:[0-9]+}}], v{{[0-9]+}} glc{{$}}
 ; 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

 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{{$}}
-; VI: s_movk_i32 flat_scratch_lo, 0x0
-; VI: s_movk_i32 flat_scratch_hi, 0x0
 ; VI: flat_atomic_umax [[RET:v[0-9]+]], v[{{[0-9]+:[0-9]+}}], v{{[0-9]+}} glc{{$}}
 ; 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
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{{$}}
-; VI: s_movk_i32 flat_scratch_lo, 0x0
-; VI: s_movk_i32 flat_scratch_hi, 0x0
 ; VI: flat_atomic_smin [[RET:v[0-9]+]], v[{{[0-9]+:[0-9]+}}], v{{[0-9]+}} glc{{$}}
 ; 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

 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{{$}}
-; VI: s_movk_i32 flat_scratch_lo, 0x0
-; VI: s_movk_i32 flat_scratch_hi, 0x0
 ; VI: flat_atomic_smin [[RET:v[0-9]+]], v[{{[0-9]+:[0-9]+}}], v{{[0-9]+}} glc{{$}}
 ; 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
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{{$}}
-; VI: s_movk_i32 flat_scratch_lo, 0x0
-; VI: s_movk_i32 flat_scratch_hi, 0x0
 ; VI: flat_atomic_umin [[RET:v[0-9]+]], v[{{[0-9]+:[0-9]+}}], v{{[0-9]+}} glc{{$}}
 ; 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

 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{{$}}
-; VI: s_movk_i32 flat_scratch_lo, 0x0
-; VI: s_movk_i32 flat_scratch_hi, 0x0
 ; VI: flat_atomic_umin [[RET:v[0-9]+]], v[{{[0-9]+:[0-9]+}}], v{{[0-9]+}} glc{{$}}
 ; 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
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{{$}}
-; VI: s_movk_i32 flat_scratch_lo, 0x0
-; VI: s_movk_i32 flat_scratch_hi, 0x0
 ; VI: flat_atomic_or [[RET:v[0-9]+]], v[{{[0-9]+:[0-9]+}}], v{{[0-9]+}} glc{{$}}
 ; 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

 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{{$}}
-; VI: s_movk_i32 flat_scratch_lo, 0x0
-; VI: s_movk_i32 flat_scratch_hi, 0x0
 ; VI: flat_atomic_or [[RET:v[0-9]+]], v[{{[0-9]+:[0-9]+}}], v{{[0-9]+}} glc{{$}}
 ; 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
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{{$}}
-; VI: s_movk_i32 flat_scratch_lo, 0x0
-; VI: s_movk_i32 flat_scratch_hi, 0x0
 ; VI: flat_atomic_swap [[RET:v[0-9]+]], v[{{[0-9]+:[0-9]+}}], v{{[0-9]+}} glc{{$}}
 ; 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

 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{{$}}
-; VI: s_movk_i32 flat_scratch_lo, 0x0
-; VI: s_movk_i32 flat_scratch_hi, 0x0
 ; VI: flat_atomic_swap [[RET:v[0-9]+]],  v[{{[0-9]+:[0-9]+}}], v{{[0-9]+}} glc{{$}}
 ; 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
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{{$}}
-; VI: s_movk_i32 flat_scratch_lo, 0x0
-; VI: s_movk_i32 flat_scratch_hi, 0x0
 ; VI: flat_atomic_xor [[RET:v[0-9]+]], v[{{[0-9]+:[0-9]+}}], v{{[0-9]+}} glc{{$}}
 ; 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

 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{{$}}
-; VI: s_movk_i32 flat_scratch_lo, 0x0
-; VI: s_movk_i32 flat_scratch_hi, 0x0
 ; VI: flat_atomic_xor [[RET:v[0-9]+]], v[{{[0-9]+:[0-9]+}}], v{{[0-9]+}} glc{{$}}
 ; 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
 }

test/CodeGen/AMDGPU/hsa.ll

 ; ELF: 0000: 04000000 08000000 01000000 414D4400
 ; ELF: 0010: 01000000 00000000 04000000 1B000000
 ; ELF: 0020: 03000000 414D4400 04000700 07000000
 ; ELF: 0030: 00000000 00000000 414D4400 414D4447
 ; ELF: 0040: 50550000
 
 ; ELF: Symbol {
 ; ELF: Name: simple
-; ELF: Size: 296
+; ELF: Size: 288
 ; ELF: Type: AMDGPU_HSA_KERNEL (0xA)
 ; ELF: }
 
 ; HSA: .hsa_code_object_version 1,0
 ; HSA-CI: .hsa_code_object_isa 7,0,0,"AMD","AMDGPU"
 ; HSA-VI: .hsa_code_object_isa 8,0,1,"AMD","AMDGPU"
 
 ; HSA: .hsatext

test/CodeGen/AMDGPU/large-alloca-compute.ll

 ; GCN: s_mov_b32 s10, -1
 ; CI: s_mov_b32 s11, 0x80f000
 ; VI: s_mov_b32 s11, 0x800000
 
 
 ; GCNHSA: .amd_kernel_code_t
 
 ; GCNHSA: compute_pgm_rsrc2_scratch_en = 1
-; GCNHSA: compute_pgm_rsrc2_user_sgpr = 6
+; GCNHSA: compute_pgm_rsrc2_user_sgpr = 8
 ; GCNHSA: compute_pgm_rsrc2_tgid_x_en = 1
 ; GCNHSA: compute_pgm_rsrc2_tgid_y_en = 0
 ; GCNHSA: compute_pgm_rsrc2_tgid_z_en = 0
 ; GCNHSA: compute_pgm_rsrc2_tg_size_en = 0
 ; GCNHSA: compute_pgm_rsrc2_tidig_comp_cnt = 0
 
 ; GCNHSA: enable_sgpr_private_segment_buffer = 1
 ; GCNHSA: enable_sgpr_dispatch_ptr = 0
 ; GCNHSA: enable_sgpr_queue_ptr = 0
 ; GCNHSA: enable_sgpr_kernarg_segment_ptr = 1
 ; GCNHSA: enable_sgpr_dispatch_id = 0
-; GCNHSA: enable_sgpr_flat_scratch_init = 0
+; GCNHSA: enable_sgpr_flat_scratch_init = 1
 ; GCNHSA: enable_sgpr_private_segment_size = 0
 ; GCNHSA: enable_sgpr_grid_workgroup_count_x = 0
 ; GCNHSA: enable_sgpr_grid_workgroup_count_y = 0
 ; GCNHSA: enable_sgpr_grid_workgroup_count_z = 0
 ; GCNHSA: workitem_private_segment_byte_size = 32772
 ; GCNHSA: private_segment_alignment = 4
 ; GCNHSA: .end_amd_kernel_code_t
 
 
-; GCNHSA: buffer_store_dword {{v[0-9]+}}, {{v[0-9]+}}, s[0:3], s7 offen
-; GCNHSA: buffer_load_dword {{v[0-9]+}}, {{v[0-9]+}}, s[0:3], s7 offen
+; GCNHSA: buffer_store_dword {{v[0-9]+}}, {{v[0-9]+}}, s[0:3], s9 offen
+; GCNHSA: buffer_load_dword {{v[0-9]+}}, {{v[0-9]+}}, s[0:3], s9 offen
 
 ; Scratch size = alloca size + emergency stack slot
 ; ALL: ; ScratchSize: 32772
 define void @large_alloca_compute_shader(i32 %x, i32 %y) #0 {
   %large = alloca [8192 x i32], align 4
   %gep = getelementptr [8192 x i32], [8192 x i32]* %large, i32 0, i32 8191
   store volatile i32 %x, i32* %gep
   %gep1 = getelementptr [8192 x i32], [8192 x i32]* %large, i32 0, i32 %y
   %val = load volatile i32, i32* %gep1
   store volatile i32 %val, i32 addrspace(1)* undef
   ret void
 }
 
 attributes #0 = { nounwind  }

test/CodeGen/AMDGPU/private-memory.ll

 ; SI-PROMOTE: ds_read_b32
 
 ; HSA-ALLOCA: .amd_kernel_code_t
 ; FIXME: Creating the emergency stack slots causes us to over-estimate scratch
 ; by 4 bytes.
 ; HSA-ALLOCA: workitem_private_segment_byte_size = 24
 ; HSA-ALLOCA: .end_amd_kernel_code_t
 
+; HSA-ALLOCA: s_mov_b32 flat_scratch_lo, s7
+; HSA-ALLOCA: s_add_u32 s6, s6, s9
+; HSA-ALLOCA: s_lshr_b32 flat_scratch_hi, s6, 8
+
 ; SI-ALLOCA: buffer_store_dword v{{[0-9]+}}, v{{[0-9]+}}, s[{{[0-9]+:[0-9]+}}], s{{[0-9]+}} offen ; encoding: [0x00,0x10,0x70,0xe0
 ; SI-ALLOCA: buffer_store_dword v{{[0-9]+}}, v{{[0-9]+}}, s[{{[0-9]+:[0-9]+}}], s{{[0-9]+}} offen ; encoding: [0x00,0x10,0x70,0xe0
 
 
 ; HSAOPT: [[DISPATCH_PTR:%[0-9]+]] = call noalias nonnull dereferenceable(64) i8 addrspace(2)* @llvm.amdgcn.dispatch.ptr()
 ; HSAOPT: [[CAST_DISPATCH_PTR:%[0-9]+]] = bitcast i8 addrspace(2)* [[DISPATCH_PTR]] to i32 addrspace(2)*
 ; HSAOPT: [[GEP0:%[0-9]+]] = getelementptr inbounds i32, i32 addrspace(2)* [[CAST_DISPATCH_PTR]], i64 1
 ; HSAOPT: [[LDXY:%[0-9]+]] = load i32, i32 addrspace(2)* [[GEP0]], align 4, !invariant.load !0

test/CodeGen/AMDGPU/vgpr-spill-emergency-stack-slot-compute.ll

-; RUN: llc -march=amdgcn -mcpu=tahiti -mattr=+vgpr-spilling -verify-machineinstrs < %s | FileCheck -check-prefix=GCN -check-prefix=SI %s
-; RUN: llc -march=amdgcn -mcpu=fiji -mattr=+vgpr-spilling -verify-machineinstrs < %s | FileCheck -check-prefix=GCN -check-prefix=VI %s
-; XUN: llc -march=amdgcn -mcpu=hawaii -mtriple=amdgcn-unknown-amdhsa -mattr=+vgpr-spilling -verify-machineinstrs < %s | FileCheck -check-prefix=GCN -check-prefix=CIHSA %s
-; XUN: llc -march=amdgcn -mcpu=fiji -mtriple=amdgcn-unknown-amdhsa -mattr=+vgpr-spilling -verify-machineinstrs < %s | FileCheck -check-prefix=GCN -check-prefix=VIHSA %s
+; RUN: llc -march=amdgcn -mcpu=tahiti -mattr=+vgpr-spilling -verify-machineinstrs < %s | FileCheck -check-prefix=GCN -check-prefix=GCNMESA -check-prefix=SIMESA %s
+; RUN: llc -march=amdgcn -mcpu=fiji -mattr=+vgpr-spilling -verify-machineinstrs < %s | FileCheck -check-prefix=GCN -check-prefix=GCNMESA -check-prefix=VIMESA %s
+; RUN: llc -march=amdgcn -mcpu=hawaii -mtriple=amdgcn-unknown-amdhsa -mattr=+vgpr-spilling -verify-machineinstrs < %s | FileCheck -check-prefix=GCN -check-prefix=CIHSA -check-prefix=HSA %s
+; RUN: llc -march=amdgcn -mcpu=fiji -mtriple=amdgcn-unknown-amdhsa -mattr=+vgpr-spilling -verify-machineinstrs < %s | FileCheck -check-prefix=GCN -check-prefix=VIHSA -check-prefix=HSA %s
 
 ; This ends up using all 256 registers and requires register
 ; scavenging which will fail to find an unsued register.
 
 ; Check the ScratchSize to avoid regressions from spilling
 ; intermediate register class copies.
 
 ; FIXME: The same register is initialized to 0 for every spill.
 
-declare i32 @llvm.r600.read.tgid.x() #1
-declare i32 @llvm.r600.read.tgid.y() #1
-declare i32 @llvm.r600.read.tgid.z() #1
-
 ; GCN-LABEL: {{^}}spill_vgpr_compute:
 
-; GCN: s_mov_b32 s16, s3
-; GCN: s_mov_b32 s12, SCRATCH_RSRC_DWORD0
-; GCN-NEXT: s_mov_b32 s13, SCRATCH_RSRC_DWORD1
-; GCN-NEXT: s_mov_b32 s14, -1
-; SI-NEXT: s_mov_b32 s15, 0x80f000
-; VI-NEXT: s_mov_b32 s15, 0x800000
+; HSA: enable_sgpr_private_segment_buffer = 1
+; HSA: enable_sgpr_flat_scratch_init = 0
+; HSA: workitem_private_segment_byte_size = 1024
+
+; GCN-NOT: flat_scr
+
+; GCNMESA: s_mov_b32 s16, s3
+; GCNMESA: s_mov_b32 s12, SCRATCH_RSRC_DWORD0
+; GCNMESA-NEXT: s_mov_b32 s13, SCRATCH_RSRC_DWORD1
+; GCNMESA-NEXT: s_mov_b32 s14, -1
+; SIMESA-NEXT: s_mov_b32 s15, 0x80f000
+; VIMESA-NEXT: s_mov_b32 s15, 0x800000
 
 
 ; GCN: buffer_store_dword {{v[0-9]+}}, s[12:15], s16 offset:{{[0-9]+}} ; 4-byte Folded Spill
 
 ; GCN: buffer_store_dword {{v[0-9]+}}, {{v[0-9]+}}, s[12:15], s16 offen offset:{{[0-9]+}}
 ; GCN: buffer_load_dwordx4 {{v\[[0-9]+:[0-9]+\]}}, {{v[0-9]+}}, s[12:15], s16 offen offset:{{[0-9]+}}
 
 ; GCN: NumVgprs: 256