[AMDGPU] Do not release VGPRs if there may be pending scratch stores

[gandhi21299]

llvm/lib/Target/AMDGPU/SIInsertWaitcnts.cpp

1420

Shouldn't this return true?

[arsenm]

llvm/lib/Target/AMDGPU/SIInsertWaitcnts.cpp

1420

It should be false but the comment should say never not always

[rampitec]

LGTM (modulo the comment).

llvm/lib/Target/AMDGPU/SIInsertWaitcnts.cpp

 struct RegisterEncoding {
   unsigned VGPR0;
   unsigned VGPRL;
   unsigned SGPR0;
   unsigned SGPRL;
 };
 
 enum WaitEventType {
-  VMEM_ACCESS,       // vector-memory read & write
-  VMEM_READ_ACCESS,  // vector-memory read
-  VMEM_WRITE_ACCESS, // vector-memory write
-  LDS_ACCESS,        // lds read & write
-  GDS_ACCESS,        // gds read & write
-  SQ_MESSAGE,        // send message
-  SMEM_ACCESS,       // scalar-memory read & write
-  EXP_GPR_LOCK,      // export holding on its data src
-  GDS_GPR_LOCK,      // GDS holding on its data and addr src
-  EXP_POS_ACCESS,    // write to export position
-  EXP_PARAM_ACCESS,  // write to export parameter
-  VMW_GPR_LOCK,      // vector-memory write holding on its data src
-  EXP_LDS_ACCESS,    // read by ldsdir counting as export
+  VMEM_ACCESS,          // vector-memory read & write
+  VMEM_READ_ACCESS,     // vector-memory read
+  VMEM_WRITE_ACCESS,    // vector-memory write that is not scratch
+  SCRATCH_WRITE_ACCESS, // vector-memory write that may be scratch
+  LDS_ACCESS,           // lds read & write
+  GDS_ACCESS,           // gds read & write
+  SQ_MESSAGE,           // send message
+  SMEM_ACCESS,          // scalar-memory read & write
+  EXP_GPR_LOCK,         // export holding on its data src
+  GDS_GPR_LOCK,         // GDS holding on its data and addr src
+  EXP_POS_ACCESS,       // write to export position
+  EXP_PARAM_ACCESS,     // write to export parameter
+  VMW_GPR_LOCK,         // vector-memory write holding on its data src
+  EXP_LDS_ACCESS,       // read by ldsdir counting as export
   NUM_WAIT_EVENTS,
 };
 
 static const unsigned WaitEventMaskForInst[NUM_INST_CNTS] = {
     (1 << VMEM_ACCESS) | (1 << VMEM_READ_ACCESS),
     (1 << SMEM_ACCESS) | (1 << LDS_ACCESS) | (1 << GDS_ACCESS) |
         (1 << SQ_MESSAGE),
     (1 << EXP_GPR_LOCK) | (1 << GDS_GPR_LOCK) | (1 << VMW_GPR_LOCK) |
         (1 << EXP_PARAM_ACCESS) | (1 << EXP_POS_ACCESS) | (1 << EXP_LDS_ACCESS),
-    (1 << VMEM_WRITE_ACCESS)};
+    (1 << VMEM_WRITE_ACCESS) | (1 << SCRATCH_WRITE_ACCESS)};
 
 // The mapping is:
 //  0                .. SQ_MAX_PGM_VGPRS-1               real VGPRs
 //  SQ_MAX_PGM_VGPRS .. NUM_ALL_VGPRS-1                  extra VGPR-like slots
 //  NUM_ALL_VGPRS    .. NUM_ALL_VGPRS+SQ_MAX_PGM_SGPRS-1 real SGPRs
 // We reserve a fixed number of VGPR slots in the scoring tables for
 // special tokens like SCMEM_LDS (needed for buffer load to LDS).
 enum RegisterMapping {
#endif // NDEBUG
   }
 
   // Return the appropriate VMEM_*_ACCESS type for Inst, which must be a VMEM or
   // FLAT instruction.
   WaitEventType getVmemWaitEventType(const MachineInstr &Inst) const {
     assert(SIInstrInfo::isVMEM(Inst) || SIInstrInfo::isFLAT(Inst));
     if (!ST->hasVscnt())
       return VMEM_ACCESS;
-    if (Inst.mayStore() && !SIInstrInfo::isAtomicRet(Inst))
+    if (Inst.mayStore() && !SIInstrInfo::isAtomicRet(Inst)) {
+      // FLAT and SCRATCH instructions may access scratch. Other VMEM
+      // instructions do not.
+      if (SIInstrInfo::isFLAT(Inst) && mayAccessScratchThroughFlat(Inst))
+        return SCRATCH_WRITE_ACCESS;
       return VMEM_WRITE_ACCESS;
+    }
     return VMEM_READ_ACCESS;
   }
 
   bool mayAccessVMEMThroughFlat(const MachineInstr &MI) const;
   bool mayAccessLDSThroughFlat(const MachineInstr &MI) const;
+  bool mayAccessScratchThroughFlat(const MachineInstr &MI) const;
   bool generateWaitcntInstBefore(MachineInstr &MI,
                                  WaitcntBrackets &ScoreBrackets,
                                  MachineInstr *OldWaitcntInstr,
                                  bool FlushVmCnt);
   bool generateWaitcntBlockEnd(MachineBasicBlock &Block,
                                WaitcntBrackets &ScoreBrackets,
                                MachineInstr *OldWaitcntInstr);
   bool generateWaitcnt(AMDGPU::Waitcnt Wait,
bool SIInsertWaitcnts::generateWaitcntInstBefore(MachineInstr &MI,
   if (MI.getOpcode() == AMDGPU::SI_RETURN_TO_EPILOG ||
       MI.getOpcode() == AMDGPU::SI_RETURN ||
       MI.getOpcode() == AMDGPU::S_SETPC_B64_return ||
       (MI.isReturn() && MI.isCall() && !callWaitsOnFunctionEntry(MI))) {
     Wait = Wait.combined(allZeroWaitcnt());
   }
   // Identify S_ENDPGM instructions which may have to wait for outstanding VMEM
   // stores. In this case it can be useful to send a message to explicitly
-  // release all VGPRs before the stores have completed.
+  // release all VGPRs before the stores have completed, but it is only safe to
+  // do this if there are no outstanding scratch stores.
   else if (MI.getOpcode() == AMDGPU::S_ENDPGM ||
            MI.getOpcode() == AMDGPU::S_ENDPGM_SAVED) {
     if (ST->getGeneration() >= AMDGPUSubtarget::GFX11 &&
-        ScoreBrackets.getScoreRange(VS_CNT) != 0)
+        ScoreBrackets.getScoreRange(VS_CNT) != 0 &&
+        !ScoreBrackets.hasPendingEvent(SCRATCH_WRITE_ACCESS))
       ReleaseVGPRInsts.insert(&MI);
   }
   // Resolve vm waits before gs-done.
   else if ((MI.getOpcode() == AMDGPU::S_SENDMSG ||
             MI.getOpcode() == AMDGPU::S_SENDMSGHALT) &&
            ST->hasLegacyGeometry() &&
            ((MI.getOperand(0).getImm() & AMDGPU::SendMsg::ID_MASK_PreGFX11_) ==
             AMDGPU::SendMsg::ID_GS_DONE_PreGFX11)) {
for (const MachineMemOperand *Memop : MI.memoperands()) {
     unsigned AS = Memop->getAddrSpace();
     if (AS == AMDGPUAS::LOCAL_ADDRESS || AS == AMDGPUAS::FLAT_ADDRESS)
       return true;
   }
 
   return false;
 }
 
+// This is a flat memory operation. Check to see if it has memory tokens for
+// either scratch or FLAT.
+bool SIInsertWaitcnts::mayAccessScratchThroughFlat(
+    const MachineInstr &MI) const {
+  assert(TII->isFLAT(MI));
+
+  // SCRATCH instructions always access scratch.
+  if (TII->isFLATScratch(MI))
+    return true;
+
+  // GLOBAL instructions never access scratch.
+  if (TII->isFLATGlobal(MI))
+    return false;
+
+  // If there are no memory operands then conservatively assume the flat
+  // operation may access scratch.
+  if (MI.memoperands_empty())
+    return true;
+
+  // See if any memory operand specifies an address space that involves scratch.
+  return any_of(MI.memoperands(), [](const MachineMemOperand *Memop) {
+    unsigned AS = Memop->getAddrSpace();
+    return AS == AMDGPUAS::PRIVATE_ADDRESS || AS == AMDGPUAS::FLAT_ADDRESS;
+  });
+}
+
 void SIInsertWaitcnts::updateEventWaitcntAfter(MachineInstr &Inst,
                                                WaitcntBrackets *ScoreBrackets) {
   // Now look at the instruction opcode. If it is a memory access
   // instruction, update the upper-bound of the appropriate counter's
   // bracket and the destination operand scores.
   // TODO: Use the (TSFlags & SIInstrFlags::LGKM_CNT) property everywhere.
   if (TII->isDS(Inst) && TII->usesLGKM_CNT(Inst)) {
     if (TII->isAlwaysGDS(Inst.getOpcode()) ||

llvm/test/CodeGen/AMDGPU/GlobalISel/dynamic-alloca-uniform.ll

 ; GFX11-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-NEXT:    s_lshl2_add_u32 s0, s0, 15
 ; GFX11-NEXT:    s_delay_alu instid0(SALU_CYCLE_1) | instskip(NEXT) | instid1(SALU_CYCLE_1)
 ; GFX11-NEXT:    s_and_b32 s0, s0, -16
 ; GFX11-NEXT:    s_lshl_b32 s0, s0, 5
 ; GFX11-NEXT:    s_delay_alu instid0(SALU_CYCLE_1)
 ; GFX11-NEXT:    s_add_u32 s0, s32, s0
 ; GFX11-NEXT:    scratch_store_b32 off, v0, s0
-; GFX11-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-NEXT:    s_endpgm
   %alloca = alloca i32, i32 %n, align 4, addrspace(5)
   store i32 0, ptr addrspace(5) %alloca
   ret void
 }
 
 define void @func_dynamic_stackalloc_sgpr_align4() {
 ; GFX9-LABEL: func_dynamic_stackalloc_sgpr_align4:

 ; GFX11-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-NEXT:    s_lshl2_add_u32 s0, s0, 15
 ; GFX11-NEXT:    s_delay_alu instid0(SALU_CYCLE_1) | instskip(NEXT) | instid1(SALU_CYCLE_1)
 ; GFX11-NEXT:    s_and_b32 s0, s0, -16
 ; GFX11-NEXT:    s_lshl_b32 s0, s0, 5
 ; GFX11-NEXT:    s_delay_alu instid0(SALU_CYCLE_1)
 ; GFX11-NEXT:    s_add_u32 s0, s32, s0
 ; GFX11-NEXT:    scratch_store_b32 off, v0, s0
-; GFX11-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-NEXT:    s_endpgm
   %alloca = alloca i32, i32 %n, align 16, addrspace(5)
   store i32 0, ptr addrspace(5) %alloca
   ret void
 }
 
 define void @func_dynamic_stackalloc_sgpr_align16() {
 ; GFX9-LABEL: func_dynamic_stackalloc_sgpr_align16:

 ; GFX11-NEXT:    s_lshl2_add_u32 s0, s0, 15
 ; GFX11-NEXT:    s_delay_alu instid0(SALU_CYCLE_1) | instskip(NEXT) | instid1(SALU_CYCLE_1)
 ; GFX11-NEXT:    s_and_b32 s0, s0, -16
 ; GFX11-NEXT:    s_lshl_b32 s0, s0, 5
 ; GFX11-NEXT:    s_delay_alu instid0(SALU_CYCLE_1) | instskip(NEXT) | instid1(SALU_CYCLE_1)
 ; GFX11-NEXT:    s_add_u32 s0, s32, s0
 ; GFX11-NEXT:    s_and_b32 s0, s0, 0xfffffc00
 ; GFX11-NEXT:    scratch_store_b32 off, v0, s0
-; GFX11-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-NEXT:    s_endpgm
   %alloca = alloca i32, i32 %n, align 32, addrspace(5)
   store i32 0, ptr addrspace(5) %alloca
   ret void
 }
 
 define void @func_dynamic_stackalloc_sgpr_align32(ptr addrspace(1) %out) {
 ; GFX9-LABEL: func_dynamic_stackalloc_sgpr_align32:

llvm/test/CodeGen/AMDGPU/GlobalISel/llvm.amdgcn.atomic.inc.ll

 ; GFX11-NEXT:    s_load_b128 s[0:3], s[0:1], 0x0
 ; GFX11-NEXT:    v_mov_b32_e32 v2, 42
 ; GFX11-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-NEXT:    v_dual_mov_b32 v0, s2 :: v_dual_mov_b32 v1, s3
 ; GFX11-NEXT:    flat_atomic_inc_u32 v2, v[0:1], v2 glc
 ; GFX11-NEXT:    v_dual_mov_b32 v0, s0 :: v_dual_mov_b32 v1, s1
 ; GFX11-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX11-NEXT:    flat_store_b32 v[0:1], v2
-; GFX11-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-NEXT:    s_endpgm
   %result = call i32 @llvm.amdgcn.atomic.inc.i32.p0(ptr %ptr, i32 42, i32 0, i32 0, i1 false)
   store i32 %result, ptr %out
   ret void
 }
 
 define amdgpu_kernel void @flat_atomic_inc_ret_i32_offset(ptr %out, ptr %ptr) #0 {
 ; CI-LABEL: flat_atomic_inc_ret_i32_offset:

 ; GFX11-NEXT:    s_load_b128 s[0:3], s[0:1], 0x0
 ; GFX11-NEXT:    v_mov_b32_e32 v2, 42
 ; GFX11-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-NEXT:    v_dual_mov_b32 v0, s2 :: v_dual_mov_b32 v1, s3
 ; GFX11-NEXT:    flat_atomic_inc_u32 v2, v[0:1], v2 offset:16 glc
 ; GFX11-NEXT:    v_dual_mov_b32 v0, s0 :: v_dual_mov_b32 v1, s1
 ; GFX11-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX11-NEXT:    flat_store_b32 v[0:1], v2
-; GFX11-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-NEXT:    s_endpgm
   %gep = getelementptr i32, ptr %ptr, i32 4
   %result = call i32 @llvm.amdgcn.atomic.inc.i32.p0(ptr %gep, i32 42, i32 0, i32 0, i1 false)
   store i32 %result, ptr %out
   ret void
 }
 
 define amdgpu_kernel void @flat_atomic_inc_noret_i32(ptr %ptr) nounwind {

 ;
 ; GFX11-LABEL: flat_atomic_inc_noret_i32:
 ; GFX11:       ; %bb.0:
 ; GFX11-NEXT:    s_load_b64 s[0:1], s[0:1], 0x0
 ; GFX11-NEXT:    v_mov_b32_e32 v2, 42
 ; GFX11-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-NEXT:    v_dual_mov_b32 v0, s0 :: v_dual_mov_b32 v1, s1
 ; GFX11-NEXT:    flat_atomic_inc_u32 v[0:1], v2
-; GFX11-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-NEXT:    s_endpgm
   %result = call i32 @llvm.amdgcn.atomic.inc.i32.p0(ptr %ptr, i32 42, i32 0, i32 0, i1 false)
   ret void
 }
 
 define amdgpu_kernel void @flat_atomic_inc_noret_i32_offset(ptr %ptr) nounwind {
 ; CI-LABEL: flat_atomic_inc_noret_i32_offset:
 ; CI:       ; %bb.0:

 ;
 ; GFX11-LABEL: flat_atomic_inc_noret_i32_offset:
 ; GFX11:       ; %bb.0:
 ; GFX11-NEXT:    s_load_b64 s[0:1], s[0:1], 0x0
 ; GFX11-NEXT:    v_mov_b32_e32 v2, 42
 ; GFX11-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-NEXT:    v_dual_mov_b32 v0, s0 :: v_dual_mov_b32 v1, s1
 ; GFX11-NEXT:    flat_atomic_inc_u32 v[0:1], v2 offset:16
-; GFX11-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-NEXT:    s_endpgm
   %gep = getelementptr i32, ptr %ptr, i32 4
   %result = call i32 @llvm.amdgcn.atomic.inc.i32.p0(ptr %gep, i32 42, i32 0, i32 0, i1 false)
   ret void
 }
 
 define amdgpu_kernel void @flat_atomic_inc_ret_i32_offset_addr64(ptr %out, ptr %ptr) #0 {
 ; CI-LABEL: flat_atomic_inc_ret_i32_offset_addr64:

 ; GFX11-NEXT:    v_add_co_ci_u32_e32 v1, vcc_lo, 0, v1, vcc_lo
 ; GFX11-NEXT:    flat_atomic_inc_u32 v3, v[0:1], v3 offset:20 glc
 ; GFX11-NEXT:    v_dual_mov_b32 v0, s0 :: v_dual_mov_b32 v1, s1
 ; GFX11-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_2)
 ; GFX11-NEXT:    v_add_co_u32 v0, vcc_lo, v0, v2
 ; GFX11-NEXT:    v_add_co_ci_u32_e32 v1, vcc_lo, 0, v1, vcc_lo
 ; GFX11-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX11-NEXT:    flat_store_b32 v[0:1], v3
-; GFX11-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-NEXT:    s_endpgm
   %id = call i32 @llvm.amdgcn.workitem.id.x()
   %gep.tid = getelementptr i32, ptr %ptr, i32 %id
   %out.gep = getelementptr i32, ptr %out, i32 %id
   %gep = getelementptr i32, ptr %gep.tid, i32 5
   %result = call i32 @llvm.amdgcn.atomic.inc.i32.p0(ptr %gep, i32 42, i32 0, i32 0, i1 false)
   store i32 %result, ptr %out.gep
   ret void

 ; GFX11-NEXT:    v_lshlrev_b32_e32 v2, 2, v0
 ; GFX11-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-NEXT:    v_dual_mov_b32 v0, s0 :: v_dual_mov_b32 v1, s1
 ; GFX11-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_2)
 ; GFX11-NEXT:    v_add_co_u32 v0, vcc_lo, v0, v2
 ; GFX11-NEXT:    v_add_co_ci_u32_e32 v1, vcc_lo, 0, v1, vcc_lo
 ; GFX11-NEXT:    v_mov_b32_e32 v2, 42
 ; GFX11-NEXT:    flat_atomic_inc_u32 v[0:1], v2 offset:20
-; GFX11-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-NEXT:    s_endpgm
   %id = call i32 @llvm.amdgcn.workitem.id.x()
   %gep.tid = getelementptr i32, ptr %ptr, i32 %id
   %gep = getelementptr i32, ptr %gep.tid, i32 5
   %result = call i32 @llvm.amdgcn.atomic.inc.i32.p0(ptr %gep, i32 42, i32 0, i32 0, i1 false)
   ret void
 }
 

 ; GFX11-NEXT:    v_mov_b32_e32 v0, 42
 ; GFX11-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-NEXT:    v_dual_mov_b32 v1, 0 :: v_dual_mov_b32 v2, s2
 ; GFX11-NEXT:    v_mov_b32_e32 v3, s3
 ; GFX11-NEXT:    flat_atomic_inc_u64 v[0:1], v[2:3], v[0:1] glc
 ; GFX11-NEXT:    v_dual_mov_b32 v3, s1 :: v_dual_mov_b32 v2, s0
 ; GFX11-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX11-NEXT:    flat_store_b64 v[2:3], v[0:1]
-; GFX11-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-NEXT:    s_endpgm
   %result = call i64 @llvm.amdgcn.atomic.inc.i64.p0(ptr %ptr, i64 42, i32 0, i32 0, i1 false)
   store i64 %result, ptr %out
   ret void
 }
 
 define amdgpu_kernel void @flat_atomic_inc_ret_i64_offset(ptr %out, ptr %ptr) #0 {
 ; CI-LABEL: flat_atomic_inc_ret_i64_offset:

 ; GFX11-NEXT:    v_mov_b32_e32 v0, 42
 ; GFX11-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-NEXT:    v_dual_mov_b32 v1, 0 :: v_dual_mov_b32 v2, s2
 ; GFX11-NEXT:    v_mov_b32_e32 v3, s3
 ; GFX11-NEXT:    flat_atomic_inc_u64 v[0:1], v[2:3], v[0:1] offset:32 glc
 ; GFX11-NEXT:    v_dual_mov_b32 v3, s1 :: v_dual_mov_b32 v2, s0
 ; GFX11-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX11-NEXT:    flat_store_b64 v[2:3], v[0:1]
-; GFX11-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-NEXT:    s_endpgm
   %gep = getelementptr i64, ptr %ptr, i32 4
   %result = call i64 @llvm.amdgcn.atomic.inc.i64.p0(ptr %gep, i64 42, i32 0, i32 0, i1 false)
   store i64 %result, ptr %out
   ret void
 }
 
 define amdgpu_kernel void @flat_atomic_inc_noret_i64(ptr %ptr) nounwind {

 ; GFX11-LABEL: flat_atomic_inc_noret_i64:
 ; GFX11:       ; %bb.0:
 ; GFX11-NEXT:    s_load_b64 s[0:1], s[0:1], 0x0
 ; GFX11-NEXT:    v_mov_b32_e32 v0, 42
 ; GFX11-NEXT:    v_mov_b32_e32 v1, 0
 ; GFX11-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-NEXT:    v_dual_mov_b32 v3, s1 :: v_dual_mov_b32 v2, s0
 ; GFX11-NEXT:    flat_atomic_inc_u64 v[2:3], v[0:1]
-; GFX11-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-NEXT:    s_endpgm
   %result = call i64 @llvm.amdgcn.atomic.inc.i64.p0(ptr %ptr, i64 42, i32 0, i32 0, i1 false)
   ret void
 }
 
 define amdgpu_kernel void @flat_atomic_inc_noret_i64_offset(ptr %ptr) nounwind {
 ; CI-LABEL: flat_atomic_inc_noret_i64_offset:
 ; CI:       ; %bb.0:

 ; GFX11-LABEL: flat_atomic_inc_noret_i64_offset:
 ; GFX11:       ; %bb.0:
 ; GFX11-NEXT:    s_load_b64 s[0:1], s[0:1], 0x0
 ; GFX11-NEXT:    v_mov_b32_e32 v0, 42
 ; GFX11-NEXT:    v_mov_b32_e32 v1, 0
 ; GFX11-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-NEXT:    v_dual_mov_b32 v3, s1 :: v_dual_mov_b32 v2, s0
 ; GFX11-NEXT:    flat_atomic_inc_u64 v[2:3], v[0:1] offset:32
-; GFX11-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-NEXT:    s_endpgm
   %gep = getelementptr i64, ptr %ptr, i32 4
   %result = call i64 @llvm.amdgcn.atomic.inc.i64.p0(ptr %gep, i64 42, i32 0, i32 0, i1 false)
   ret void
 }
 
 define amdgpu_kernel void @flat_atomic_inc_ret_i64_offset_addr64(ptr %out, ptr %ptr) #0 {
 ; CI-LABEL: flat_atomic_inc_ret_i64_offset_addr64:

 ; GFX11-NEXT:    v_add_co_ci_u32_e32 v1, vcc_lo, 0, v1, vcc_lo
 ; GFX11-NEXT:    flat_atomic_inc_u64 v[0:1], v[0:1], v[2:3] offset:40 glc
 ; GFX11-NEXT:    v_dual_mov_b32 v3, s1 :: v_dual_mov_b32 v2, s0
 ; GFX11-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_2)
 ; GFX11-NEXT:    v_add_co_u32 v2, vcc_lo, v2, v4
 ; GFX11-NEXT:    v_add_co_ci_u32_e32 v3, vcc_lo, 0, v3, vcc_lo
 ; GFX11-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX11-NEXT:    flat_store_b64 v[2:3], v[0:1]
-; GFX11-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-NEXT:    s_endpgm
   %id = call i32 @llvm.amdgcn.workitem.id.x()
   %gep.tid = getelementptr i64, ptr %ptr, i32 %id
   %out.gep = getelementptr i64, ptr %out, i32 %id
   %gep = getelementptr i64, ptr %gep.tid, i32 5
   %result = call i64 @llvm.amdgcn.atomic.inc.i64.p0(ptr %gep, i64 42, i32 0, i32 0, i1 false)
   store i64 %result, ptr %out.gep
   ret void

 ; GFX11-NEXT:    v_mov_b32_e32 v2, 42
 ; GFX11-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-NEXT:    v_dual_mov_b32 v3, 0 :: v_dual_mov_b32 v0, s0
 ; GFX11-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX11-NEXT:    s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_2)
 ; GFX11-NEXT:    v_add_co_u32 v0, vcc_lo, v0, v4
 ; GFX11-NEXT:    v_add_co_ci_u32_e32 v1, vcc_lo, 0, v1, vcc_lo
 ; GFX11-NEXT:    flat_atomic_inc_u64 v[0:1], v[2:3] offset:40
-; GFX11-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-NEXT:    s_endpgm
   %id = call i32 @llvm.amdgcn.workitem.id.x()
   %gep.tid = getelementptr i64, ptr %ptr, i32 %id
   %gep = getelementptr i64, ptr %gep.tid, i32 5
   %result = call i64 @llvm.amdgcn.atomic.inc.i64.p0(ptr %gep, i64 42, i32 0, i32 0, i1 false)
   ret void
 }
 

llvm/test/CodeGen/AMDGPU/GlobalISel/llvm.amdgcn.intersect_ray.ll

 ; GFX11-NEXT:    s_mov_b32 s2, 2.0
 ; GFX11-NEXT:    v_dual_mov_b32 v0, s0 :: v_dual_mov_b32 v3, s8
 ; GFX11-NEXT:    v_dual_mov_b32 v1, s1 :: v_dual_mov_b32 v2, s2
 ; GFX11-NEXT:    v_dual_mov_b32 v5, s10 :: v_dual_mov_b32 v4, s9
 ; GFX11-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX11-NEXT:    image_bvh_intersect_ray v[0:3], [v9, v10, v[0:2], v[3:5], v[6:8]], s[4:7]
 ; GFX11-NEXT:    s_waitcnt vmcnt(0)
 ; GFX11-NEXT:    flat_store_b128 v[0:1], v[0:3]
-; GFX11-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-NEXT:    s_endpgm
   %lid = tail call i32 @llvm.amdgcn.workitem.id.x()
   %gep_node_ptr = getelementptr inbounds i32, ptr %p_node_ptr, i32 %lid
   %node_ptr = load i32, ptr %gep_node_ptr, align 4
   %gep_ray = getelementptr inbounds float, ptr %p_ray, i32 %lid
   %ray_extent = load float, ptr %gep_ray, align 4
   %ray_origin0 = insertelement <3 x float> undef, float 0.0, i32 0
   %ray_origin1 = insertelement <3 x float> %ray_origin0, float 1.0, i32 1

 ; GFX11-NEXT:    s_mov_b32 s2, 2.0
 ; GFX11-NEXT:    v_dual_mov_b32 v0, s0 :: v_dual_mov_b32 v3, s8
 ; GFX11-NEXT:    v_dual_mov_b32 v1, s1 :: v_dual_mov_b32 v2, s2
 ; GFX11-NEXT:    v_dual_mov_b32 v5, s10 :: v_dual_mov_b32 v4, s9
 ; GFX11-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX11-NEXT:    image_bvh_intersect_ray v[0:3], [v6, v7, v[0:2], v[3:5]], s[4:7] a16
 ; GFX11-NEXT:    s_waitcnt vmcnt(0)
 ; GFX11-NEXT:    flat_store_b128 v[0:1], v[0:3]
-; GFX11-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-NEXT:    s_endpgm
   %lid = tail call i32 @llvm.amdgcn.workitem.id.x()
   %gep_node_ptr = getelementptr inbounds i32, ptr %p_node_ptr, i32 %lid
   %node_ptr = load i32, ptr %gep_node_ptr, align 4
   %gep_ray = getelementptr inbounds float, ptr %p_ray, i32 %lid
   %ray_extent = load float, ptr %gep_ray, align 4
   %ray_origin0 = insertelement <3 x float> undef, float 0.0, i32 0
   %ray_origin1 = insertelement <3 x float> %ray_origin0, float 1.0, i32 1

 ; GFX11-NEXT:    v_add_co_ci_u32_e32 v1, vcc_lo, 0, v1, vcc_lo
 ; GFX11-NEXT:    flat_load_b32 v11, v[0:1]
 ; GFX11-NEXT:    v_dual_mov_b32 v0, s4 :: v_dual_mov_b32 v1, s5
 ; GFX11-NEXT:    v_mov_b32_e32 v2, s6
 ; GFX11-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX11-NEXT:    image_bvh64_intersect_ray v[0:3], [v[9:10], v11, v[0:2], v[3:5], v[6:8]], s[0:3]
 ; GFX11-NEXT:    s_waitcnt vmcnt(0)
 ; GFX11-NEXT:    flat_store_b128 v[0:1], v[0:3]
-; GFX11-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-NEXT:    s_endpgm
   %lid = tail call i32 @llvm.amdgcn.workitem.id.x()
   %gep_ray = getelementptr inbounds float, ptr %p_ray, i32 %lid
   %ray_extent = load float, ptr %gep_ray, align 4
   %ray_origin0 = insertelement <3 x float> undef, float 0.0, i32 0
   %ray_origin1 = insertelement <3 x float> %ray_origin0, float 1.0, i32 1
   %ray_origin = insertelement <3 x float> %ray_origin1, float 2.0, i32 2
   %ray_dir0 = insertelement <3 x float> undef, float 3.0, i32 0

 ; GFX11-NEXT:    v_add_co_ci_u32_e32 v1, vcc_lo, 0, v1, vcc_lo
 ; GFX11-NEXT:    flat_load_b32 v8, v[0:1]
 ; GFX11-NEXT:    v_dual_mov_b32 v0, s4 :: v_dual_mov_b32 v1, s5
 ; GFX11-NEXT:    v_mov_b32_e32 v2, s6
 ; GFX11-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX11-NEXT:    image_bvh64_intersect_ray v[0:3], [v[6:7], v8, v[0:2], v[3:5]], s[0:3] a16
 ; GFX11-NEXT:    s_waitcnt vmcnt(0)
 ; GFX11-NEXT:    flat_store_b128 v[0:1], v[0:3]
-; GFX11-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-NEXT:    s_endpgm
   %lid = tail call i32 @llvm.amdgcn.workitem.id.x()
   %gep_ray = getelementptr inbounds float, ptr %p_ray, i32 %lid
   %ray_extent = load float, ptr %gep_ray, align 4
   %ray_origin0 = insertelement <3 x float> undef, float 0.0, i32 0
   %ray_origin1 = insertelement <3 x float> %ray_origin0, float 1.0, i32 1
   %ray_origin = insertelement <3 x float> %ray_origin1, float 2.0, i32 2
   %ray_dir0 = insertelement <3 x half> undef, half 3.0, i32 0

llvm/test/CodeGen/AMDGPU/GlobalISel/llvm.amdgcn.trig.preop.ll

 ; GFX11:       ; %bb.0:
 ; GFX11-NEXT:    s_clause 0x1
 ; GFX11-NEXT:    s_load_b64 s[2:3], s[0:1], 0x0
 ; GFX11-NEXT:    s_load_b32 s0, s[0:1], 0x8
 ; GFX11-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-NEXT:    v_trig_preop_f64 v[0:1], s[2:3], s0
 ; GFX11-NEXT:    flat_store_b64 v[0:1], v[0:1] dlc
 ; GFX11-NEXT:    s_waitcnt_vscnt null, 0x0
-; GFX11-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-NEXT:    s_endpgm
   %result = call double @llvm.amdgcn.trig.preop.f64(double %a, i32 %b)
   store volatile double %result, ptr undef
   ret void
 }
 
 define amdgpu_kernel void @s_trig_preop_f64_imm(double %a, i32 %b) {
 ; GCN-LABEL: s_trig_preop_f64_imm:

 ;
 ; GFX11-LABEL: s_trig_preop_f64_imm:
 ; GFX11:       ; %bb.0:
 ; GFX11-NEXT:    s_load_b64 s[0:1], s[0:1], 0x0
 ; GFX11-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-NEXT:    v_trig_preop_f64 v[0:1], s[0:1], 7
 ; GFX11-NEXT:    flat_store_b64 v[0:1], v[0:1] dlc
 ; GFX11-NEXT:    s_waitcnt_vscnt null, 0x0
-; GFX11-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-NEXT:    s_endpgm
   %result = call double @llvm.amdgcn.trig.preop.f64(double %a, i32 7)
   store volatile double %result, ptr undef
   ret void
 }
 
 declare double @llvm.amdgcn.trig.preop.f64(double, i32) #0
 
 attributes #0 = { nounwind readnone speculatable }

llvm/test/CodeGen/AMDGPU/amd.endpgm.ll

 ; GFX11-NEXT:    s_cmp_lt_i32 s2, 1
 ; GFX11-NEXT:    s_cbranch_scc0 .LBB2_2
 ; GFX11-NEXT:  ; %bb.1: ; %else
 ; GFX11-NEXT:    s_load_b64 s[0:1], s[0:1], 0x24
 ; GFX11-NEXT:    v_mov_b32_e32 v2, s2
 ; GFX11-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-NEXT:    v_dual_mov_b32 v0, s0 :: v_dual_mov_b32 v1, s1
 ; GFX11-NEXT:    flat_store_b32 v[0:1], v2
-; GFX11-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-NEXT:    s_endpgm
 ; GFX11-NEXT:  .LBB2_2: ; %then
 ; GFX11-NEXT:    s_endpgm
   %cond = icmp sgt i32 %x, 0
   br i1 %cond, label %then, label %else
 
 then:
   tail call void @llvm.amdgcn.endpgm()
   unreachable
 
 else:
   store i32 %x, ptr %p
   ret void
 }
 
 declare void @llvm.amdgcn.endpgm()

llvm/test/CodeGen/AMDGPU/cc-update.ll

 ; GFX1010-NEXT:    s_waitcnt_vscnt null, 0x0
 ; GFX1010-NEXT:    s_endpgm
 ;
 ; GFX1100-LABEL: test_kern_stack:
 ; GFX1100:       ; %bb.0: ; %entry
 ; GFX1100-NEXT:    v_mov_b32_e32 v0, 0
 ; GFX1100-NEXT:    scratch_store_b32 off, v0, off offset:4 dlc
 ; GFX1100-NEXT:    s_waitcnt_vscnt null, 0x0
-; GFX1100-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX1100-NEXT:    s_endpgm
 entry:
   %x = alloca i32, align 4, addrspace(5)
   store volatile i32 0, ptr addrspace(5) %x, align 4
   ret void
 }
 
 define amdgpu_kernel void @test_kern_call() local_unnamed_addr #0 {

 ; GFX1010-NEXT:    s_endpgm
 ;
 ; GFX1100-LABEL: test_force_fp_kern_stack:
 ; GFX1100:       ; %bb.0: ; %entry
 ; GFX1100-NEXT:    v_mov_b32_e32 v0, 0
 ; GFX1100-NEXT:    s_mov_b32 s33, 0
 ; GFX1100-NEXT:    scratch_store_b32 off, v0, s33 offset:4 dlc
 ; GFX1100-NEXT:    s_waitcnt_vscnt null, 0x0
-; GFX1100-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX1100-NEXT:    s_endpgm
 entry:
   %x = alloca i32, align 4, addrspace(5)
   store volatile i32 0, ptr addrspace(5) %x, align 4
   ret void
 }
 
 define amdgpu_kernel void @test_force_fp_kern_call() local_unnamed_addr #2 {

 ; GFX1100-NEXT:    s_movk_i32 s0, 0x1000
 ; GFX1100-NEXT:    scratch_store_b32 off, v0, s0 ; 4-byte Folded Spill
 ; GFX1100-NEXT:    ;;#ASMSTART
 ; GFX1100-NEXT:    ;;#ASMEND
 ; GFX1100-NEXT:    scratch_load_b32 v0, off, s0 ; 4-byte Folded Reload
 ; GFX1100-NEXT:    s_waitcnt vmcnt(0)
 ; GFX1100-NEXT:    scratch_store_b32 off, v0, off offset:8 dlc
 ; GFX1100-NEXT:    s_waitcnt_vscnt null, 0x0
-; GFX1100-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX1100-NEXT:    s_endpgm
 entry:
   ; Occupy 4096 bytes of scratch, so the offset of the spill of %a does not
   ; fit in the instruction, and has to live in the SGPR offset.
   %alloca = alloca i8, i32 4092, align 4, addrspace(5)
 
   %aptr = getelementptr i32, ptr addrspace(5) %alloca, i32 1
   ; 0x40000 / 64 = 4096 (for wave64)

llvm/test/CodeGen/AMDGPU/cluster_stores.ll

 ; GFX11-NEXT:    s_waitcnt vmcnt(3) lgkmcnt(3)
 ; GFX11-NEXT:    flat_store_b32 v[0:1], v2
 ; GFX11-NEXT:    s_waitcnt vmcnt(2) lgkmcnt(3)
 ; GFX11-NEXT:    flat_store_b32 v[0:1], v3 offset:8
 ; GFX11-NEXT:    s_waitcnt vmcnt(1) lgkmcnt(3)
 ; GFX11-NEXT:    flat_store_b32 v[0:1], v4 offset:16
 ; GFX11-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(3)
 ; GFX11-NEXT:    flat_store_b32 v[0:1], v5 offset:24
-; GFX11-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-NEXT:    s_endpgm
 bb:
   %ld0 = load i32, ptr %lb
   %la1 = getelementptr inbounds i32, ptr %lb, i32 2
   %ld1 = load i32, ptr %la1
   %la2 = getelementptr inbounds i32, ptr %lb, i32 4
   %ld2 = load i32, ptr %la2
   %la3 = getelementptr inbounds i32, ptr %lb, i32 6

 ; GFX11-NEXT:    s_waitcnt vmcnt(2) lgkmcnt(2)
 ; GFX11-NEXT:    s_clause 0x1
 ; GFX11-NEXT:    flat_store_b32 v[0:1], v3
 ; GFX11-NEXT:    flat_store_b32 v[0:1], v2 offset:8
 ; GFX11-NEXT:    s_waitcnt vmcnt(1) lgkmcnt(3)
 ; GFX11-NEXT:    flat_store_b32 v[0:1], v4 offset:16
 ; GFX11-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(3)
 ; GFX11-NEXT:    flat_store_b32 v[0:1], v5 offset:24
-; GFX11-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-NEXT:    s_endpgm
 bb:
   %ld0 = load i32, ptr %lb
   %la1 = getelementptr inbounds i32, ptr %lb, i32 2
   %ld1 = load i32, ptr %la1
   %la2 = getelementptr inbounds i32, ptr %lb, i32 4
   %ld2 = load i32, ptr %la2
   %la3 = getelementptr inbounds i32, ptr %lb, i32 6

llvm/test/CodeGen/AMDGPU/flat-scratch-i8-i16.ll

 ; GFX10-NEXT:    s_endpgm
 ;
 ; GFX11-LABEL: test_scratch_load_i8_zext_v:
 ; GFX11:       ; %bb.0:
 ; GFX11-NEXT:    v_add_nc_u32_e32 v0, 1, v0
 ; GFX11-NEXT:    scratch_load_u8 v0, v0, off
 ; GFX11-NEXT:    s_waitcnt vmcnt(0)
 ; GFX11-NEXT:    flat_store_b32 v[1:2], v0
-; GFX11-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-NEXT:    s_endpgm
   %gep = getelementptr inbounds i8, ptr addrspace(5) %in, i32 1
   %load = load i8, ptr addrspace(5) %gep, align 4
   %ext = zext i8 %load to i32
   store i32 %ext, ptr %out, align 4
   ret void
 }
 

 ; GFX10-NEXT:    s_endpgm
 ;
 ; GFX11-LABEL: test_scratch_load_i8_sext_v:
 ; GFX11:       ; %bb.0:
 ; GFX11-NEXT:    v_add_nc_u32_e32 v0, 1, v0
 ; GFX11-NEXT:    scratch_load_i8 v0, v0, off
 ; GFX11-NEXT:    s_waitcnt vmcnt(0)
 ; GFX11-NEXT:    flat_store_b32 v[1:2], v0
-; GFX11-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-NEXT:    s_endpgm
   %gep = getelementptr inbounds i8, ptr addrspace(5) %in, i32 1
   %load = load i8, ptr addrspace(5) %gep, align 4
   %ext = sext i8 %load to i32
   store i32 %ext, ptr %out, align 4
   ret void
 }
 

 ; GFX10-NEXT:    s_endpgm
 ;
 ; GFX11-LABEL: test_scratch_load_i16_zext_v:
 ; GFX11:       ; %bb.0:
 ; GFX11-NEXT:    v_add_nc_u32_e32 v0, 2, v0
 ; GFX11-NEXT:    scratch_load_u16 v0, v0, off
 ; GFX11-NEXT:    s_waitcnt vmcnt(0)
 ; GFX11-NEXT:    flat_store_b32 v[1:2], v0
-; GFX11-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-NEXT:    s_endpgm
   %gep = getelementptr inbounds i16, ptr addrspace(5) %in, i32 1
   %load = load i16, ptr addrspace(5) %gep, align 4
   %ext = zext i16 %load to i32
   store i32 %ext, ptr %out, align 4
   ret void
 }
 

 ; GFX10-NEXT:    s_endpgm
 ;
 ; GFX11-LABEL: test_scratch_load_i16_sext_v:
 ; GFX11:       ; %bb.0:
 ; GFX11-NEXT:    v_add_nc_u32_e32 v0, 2, v0
 ; GFX11-NEXT:    scratch_load_i16 v0, v0, off
 ; GFX11-NEXT:    s_waitcnt vmcnt(0)
 ; GFX11-NEXT:    flat_store_b32 v[1:2], v0
-; GFX11-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-NEXT:    s_endpgm
   %gep = getelementptr inbounds i16, ptr addrspace(5) %in, i32 1
   %load = load i16, ptr addrspace(5) %gep, align 4
   %ext = sext i16 %load to i32
   store i32 %ext, ptr %out, align 4
   ret void
 }
 

 ; GFX10-NEXT:    s_endpgm
 ;
 ; GFX11-LABEL: test_scratch_load_i8_zext_to_d16_lo_v:
 ; GFX11:       ; %bb.0: ; %bb
 ; GFX11-NEXT:    v_dual_mov_b32 v3, 0xffff0000 :: v_dual_add_nc_u32 v0, 1, v0
 ; GFX11-NEXT:    scratch_load_d16_u8 v3, v0, off
 ; GFX11-NEXT:    s_waitcnt vmcnt(0)
 ; GFX11-NEXT:    flat_store_b32 v[1:2], v3
-; GFX11-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-NEXT:    s_endpgm
 bb:
   %gep = getelementptr i8, ptr addrspace(5) %in, i64 1
   %load_lo = load i8, ptr addrspace(5) %gep
   %ext = zext i8 %load_lo to i16
   %result = insertelement <2 x i16> <i16 -1, i16 -1>, i16 %ext, i32 0
   store <2 x i16> %result, ptr %out, align 4
   ret void

 ; GFX10-NEXT:    s_endpgm
 ;
 ; GFX11-LABEL: test_scratch_load_i8_sext_to_d16_lo_v:
 ; GFX11:       ; %bb.0: ; %bb
 ; GFX11-NEXT:    v_dual_mov_b32 v3, 0xffff0000 :: v_dual_add_nc_u32 v0, 1, v0
 ; GFX11-NEXT:    scratch_load_d16_i8 v3, v0, off
 ; GFX11-NEXT:    s_waitcnt vmcnt(0)
 ; GFX11-NEXT:    flat_store_b32 v[1:2], v3
-; GFX11-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-NEXT:    s_endpgm
 bb:
   %gep = getelementptr i8, ptr addrspace(5) %in, i64 1
   %load_lo = load i8, ptr addrspace(5) %gep
   %ext = sext i8 %load_lo to i16
   %result = insertelement <2 x i16> <i16 -1, i16 -1>, i16 %ext, i32 0
   store <2 x i16> %result, ptr %out, align 4
   ret void

 ; GFX10-NEXT:    s_endpgm
 ;
 ; GFX11-LABEL: test_scratch_load_i16_to_d16_lo_v:
 ; GFX11:       ; %bb.0: ; %bb
 ; GFX11-NEXT:    v_dual_mov_b32 v3, 0xffff0000 :: v_dual_add_nc_u32 v0, 2, v0
 ; GFX11-NEXT:    scratch_load_d16_b16 v3, v0, off
 ; GFX11-NEXT:    s_waitcnt vmcnt(0)
 ; GFX11-NEXT:    flat_store_b32 v[1:2], v3
-; GFX11-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-NEXT:    s_endpgm
 bb:
   %gep = getelementptr i16, ptr addrspace(5) %in, i64 1
   %load_lo = load i16, ptr addrspace(5) %gep
   %result = insertelement <2 x i16> <i16 -1, i16 -1>, i16 %load_lo, i32 0
   store <2 x i16> %result, ptr %out, align 4
   ret void
 }

 ; GFX10-NEXT:    s_endpgm
 ;
 ; GFX11-LABEL: test_scratch_load_i8_zext_to_d16_hi_v:
 ; GFX11:       ; %bb.0: ; %bb
 ; GFX11-NEXT:    v_dual_mov_b32 v3, -1 :: v_dual_add_nc_u32 v0, 1, v0
 ; GFX11-NEXT:    scratch_load_d16_hi_u8 v3, v0, off
 ; GFX11-NEXT:    s_waitcnt vmcnt(0)
 ; GFX11-NEXT:    flat_store_b32 v[1:2], v3
-; GFX11-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-NEXT:    s_endpgm
 bb:
   %gep = getelementptr i8, ptr addrspace(5) %in, i64 1
   %load_lo = load i8, ptr addrspace(5) %gep
   %ext = zext i8 %load_lo to i16
   %result = insertelement <2 x i16> <i16 -1, i16 -1>, i16 %ext, i32 1
   store <2 x i16> %result, ptr %out, align 4
   ret void

 ; GFX10-NEXT:    s_endpgm
 ;
 ; GFX11-LABEL: test_scratch_load_i8_sext_to_d16_hi_v:
 ; GFX11:       ; %bb.0: ; %bb
 ; GFX11-NEXT:    v_dual_mov_b32 v3, -1 :: v_dual_add_nc_u32 v0, 1, v0
 ; GFX11-NEXT:    scratch_load_d16_hi_i8 v3, v0, off
 ; GFX11-NEXT:    s_waitcnt vmcnt(0)
 ; GFX11-NEXT:    flat_store_b32 v[1:2], v3
-; GFX11-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-NEXT:    s_endpgm
 bb:
   %gep = getelementptr i8, ptr addrspace(5) %in, i64 1
   %load_lo = load i8, ptr addrspace(5) %gep
   %ext = sext i8 %load_lo to i16
   %result = insertelement <2 x i16> <i16 -1, i16 -1>, i16 %ext, i32 1
   store <2 x i16> %result, ptr %out, align 4
   ret void

 ; GFX10-NEXT:    s_endpgm
 ;
 ; GFX11-LABEL: test_scratch_load_i16_to_d16_hi_v:
 ; GFX11:       ; %bb.0: ; %bb
 ; GFX11-NEXT:    v_dual_mov_b32 v3, -1 :: v_dual_add_nc_u32 v0, 2, v0
 ; GFX11-NEXT:    scratch_load_d16_hi_b16 v3, v0, off
 ; GFX11-NEXT:    s_waitcnt vmcnt(0)
 ; GFX11-NEXT:    flat_store_b32 v[1:2], v3
-; GFX11-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-NEXT:    s_endpgm
 bb:
   %gep = getelementptr i16, ptr addrspace(5) %in, i64 1
   %load_lo = load i16, ptr addrspace(5) %gep
   %result = insertelement <2 x i16> <i16 -1, i16 -1>, i16 %load_lo, i32 1
   store <2 x i16> %result, ptr %out, align 4
   ret void
 }

 ; GFX10-NEXT:    s_endpgm
 ;
 ; GFX11-LABEL: test_scratch_store_b8_from_d16_hi_v:
 ; GFX11:       ; %bb.0: ; %bb
 ; GFX11-NEXT:    flat_load_b32 v0, v[0:1]
 ; GFX11-NEXT:    v_add_nc_u32_e32 v1, 4, v2
 ; GFX11-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX11-NEXT:    scratch_store_d16_hi_b8 v1, v0, off
-; GFX11-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-NEXT:    s_endpgm
 bb:
   %load = load <4 x i8>, ptr %in
   %element = extractelement <4 x i8> %load, i32 2
   %gep = getelementptr <4 x i8>, ptr addrspace(5) %out, i64 1
   store i8 %element, ptr addrspace(5) %gep, align 4
   ret void
 }

 ; GFX10-NEXT:    s_endpgm
 ;
 ; GFX11-LABEL: test_scratch_store_b16_from_d16_hi_v:
 ; GFX11:       ; %bb.0: ; %bb
 ; GFX11-NEXT:    flat_load_b32 v0, v[0:1]
 ; GFX11-NEXT:    v_add_nc_u32_e32 v1, 2, v2
 ; GFX11-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX11-NEXT:    scratch_store_d16_hi_b16 v1, v0, off
-; GFX11-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-NEXT:    s_endpgm
 bb:
   %load = load <2 x i16>, ptr %in
   %element = extractelement <2 x i16> %load, i32 1
   %gep = getelementptr <2 x i8>, ptr addrspace(5) %out, i64 1
   store i16 %element, ptr addrspace(5) %gep, align 4
   ret void
 }

 ; GFX10-NEXT:    s_endpgm
 ;
 ; GFX11-LABEL: test_scratch_load_i8_zext_s:
 ; GFX11:       ; %bb.0:
 ; GFX11-NEXT:    s_add_i32 s0, s0, 1
 ; GFX11-NEXT:    scratch_load_u8 v2, off, s0
 ; GFX11-NEXT:    s_waitcnt vmcnt(0)
 ; GFX11-NEXT:    flat_store_b32 v[0:1], v2
-; GFX11-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-NEXT:    s_endpgm
   %gep = getelementptr inbounds i8, ptr addrspace(5) %in, i32 1
   %load = load i8, ptr addrspace(5) %gep, align 4
   %ext = zext i8 %load to i32
   store i32 %ext, ptr %out, align 4
   ret void
 }
 

 ; GFX10-NEXT:    s_endpgm
 ;
 ; GFX11-LABEL: test_scratch_load_i8_sext_s:
 ; GFX11:       ; %bb.0:
 ; GFX11-NEXT:    s_add_i32 s0, s0, 1
 ; GFX11-NEXT:    scratch_load_i8 v2, off, s0
 ; GFX11-NEXT:    s_waitcnt vmcnt(0)
 ; GFX11-NEXT:    flat_store_b32 v[0:1], v2
-; GFX11-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-NEXT:    s_endpgm
   %gep = getelementptr inbounds i8, ptr addrspace(5) %in, i32 1
   %load = load i8, ptr addrspace(5) %gep, align 4
   %ext = sext i8 %load to i32
   store i32 %ext, ptr %out, align 4
   ret void
 }
 

 ; GFX10-NEXT:    s_endpgm
 ;
 ; GFX11-LABEL: test_scratch_load_i16_zext_s:
 ; GFX11:       ; %bb.0:
 ; GFX11-NEXT:    s_add_i32 s0, s0, 2
 ; GFX11-NEXT:    scratch_load_u16 v2, off, s0
 ; GFX11-NEXT:    s_waitcnt vmcnt(0)
 ; GFX11-NEXT:    flat_store_b32 v[0:1], v2
-; GFX11-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-NEXT:    s_endpgm
   %gep = getelementptr inbounds i16, ptr addrspace(5) %in, i32 1
   %load = load i16, ptr addrspace(5) %gep, align 4
   %ext = zext i16 %load to i32
   store i32 %ext, ptr %out, align 4
   ret void
 }
 

 ; GFX10-NEXT:    s_endpgm
 ;
 ; GFX11-LABEL: test_scratch_load_i16_sext_s:
 ; GFX11:       ; %bb.0:
 ; GFX11-NEXT:    s_add_i32 s0, s0, 2
 ; GFX11-NEXT:    scratch_load_i16 v2, off, s0
 ; GFX11-NEXT:    s_waitcnt vmcnt(0)
 ; GFX11-NEXT:    flat_store_b32 v[0:1], v2
-; GFX11-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-NEXT:    s_endpgm
   %gep = getelementptr inbounds i16, ptr addrspace(5) %in, i32 1
   %load = load i16, ptr addrspace(5) %gep, align 4
   %ext = sext i16 %load to i32
   store i32 %ext, ptr %out, align 4
   ret void
 }
 

 ;
 ; GFX11-LABEL: test_scratch_load_i8_zext_to_d16_lo_s:
 ; GFX11:       ; %bb.0: ; %bb
 ; GFX11-NEXT:    v_mov_b32_e32 v2, 0xffff0000
 ; GFX11-NEXT:    s_add_i32 s0, s0, 1
 ; GFX11-NEXT:    scratch_load_d16_u8 v2, off, s0
 ; GFX11-NEXT:    s_waitcnt vmcnt(0)
 ; GFX11-NEXT:    flat_store_b32 v[0:1], v2
-; GFX11-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-NEXT:    s_endpgm
 bb:
   %gep = getelementptr i8, ptr addrspace(5) %in, i64 1
   %load_lo = load i8, ptr addrspace(5) %gep
   %ext = zext i8 %load_lo to i16
   %result = insertelement <2 x i16> <i16 -1, i16 -1>, i16 %ext, i32 0
   store <2 x i16> %result, ptr %out, align 4
   ret void

 ;
 ; GFX11-LABEL: test_scratch_load_i8_sext_to_d16_lo_s:
 ; GFX11:       ; %bb.0: ; %bb
 ; GFX11-NEXT:    v_mov_b32_e32 v2, 0xffff0000
 ; GFX11-NEXT:    s_add_i32 s0, s0, 1
 ; GFX11-NEXT:    scratch_load_d16_i8 v2, off, s0
 ; GFX11-NEXT:    s_waitcnt vmcnt(0)
 ; GFX11-NEXT:    flat_store_b32 v[0:1], v2
-; GFX11-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-NEXT:    s_endpgm
 bb:
   %gep = getelementptr i8, ptr addrspace(5) %in, i64 1
   %load_lo = load i8, ptr addrspace(5) %gep
   %ext = sext i8 %load_lo to i16
   %result = insertelement <2 x i16> <i16 -1, i16 -1>, i16 %ext, i32 0
   store <2 x i16> %result, ptr %out, align 4
   ret void

 ;
 ; GFX11-LABEL: test_scratch_load_i16_to_d16_lo_s:
 ; GFX11:       ; %bb.0: ; %bb
 ; GFX11-NEXT:    v_mov_b32_e32 v2, 0xffff0000
 ; GFX11-NEXT:    s_add_i32 s0, s0, 2
 ; GFX11-NEXT:    scratch_load_d16_b16 v2, off, s0
 ; GFX11-NEXT:    s_waitcnt vmcnt(0)
 ; GFX11-NEXT:    flat_store_b32 v[0:1], v2
-; GFX11-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-NEXT:    s_endpgm
 bb:
   %gep = getelementptr i16, ptr addrspace(5) %in, i64 1
   %load_lo = load i16, ptr addrspace(5) %gep
   %result = insertelement <2 x i16> <i16 -1, i16 -1>, i16 %load_lo, i32 0
   store <2 x i16> %result, ptr %out, align 4
   ret void
 }

 ;
 ; GFX11-LABEL: test_scratch_load_i8_zext_to_d16_hi_s:
 ; GFX11:       ; %bb.0: ; %bb
 ; GFX11-NEXT:    v_mov_b32_e32 v2, -1
 ; GFX11-NEXT:    s_add_i32 s0, s0, 1
 ; GFX11-NEXT:    scratch_load_d16_hi_u8 v2, off, s0
 ; GFX11-NEXT:    s_waitcnt vmcnt(0)
 ; GFX11-NEXT:    flat_store_b32 v[0:1], v2
-; GFX11-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-NEXT:    s_endpgm
 bb:
   %gep = getelementptr i8, ptr addrspace(5) %in, i64 1
   %load_lo = load i8, ptr addrspace(5) %gep
   %ext = zext i8 %load_lo to i16
   %result = insertelement <2 x i16> <i16 -1, i16 -1>, i16 %ext, i32 1
   store <2 x i16> %result, ptr %out, align 4
   ret void

 ;
 ; GFX11-LABEL: test_scratch_load_i8_sext_to_d16_hi_s:
 ; GFX11:       ; %bb.0: ; %bb
 ; GFX11-NEXT:    v_mov_b32_e32 v2, -1
 ; GFX11-NEXT:    s_add_i32 s0, s0, 1
 ; GFX11-NEXT:    scratch_load_d16_hi_i8 v2, off, s0
 ; GFX11-NEXT:    s_waitcnt vmcnt(0)
 ; GFX11-NEXT:    flat_store_b32 v[0:1], v2
-; GFX11-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-NEXT:    s_endpgm
 bb:
   %gep = getelementptr i8, ptr addrspace(5) %in, i64 1
   %load_lo = load i8, ptr addrspace(5) %gep
   %ext = sext i8 %load_lo to i16
   %result = insertelement <2 x i16> <i16 -1, i16 -1>, i16 %ext, i32 1
   store <2 x i16> %result, ptr %out, align 4
   ret void

 ;
 ; GFX11-LABEL: test_scratch_load_i16_to_d16_hi_s:
 ; GFX11:       ; %bb.0: ; %bb
 ; GFX11-NEXT:    v_mov_b32_e32 v2, -1
 ; GFX11-NEXT:    s_add_i32 s0, s0, 2
 ; GFX11-NEXT:    scratch_load_d16_hi_b16 v2, off, s0
 ; GFX11-NEXT:    s_waitcnt vmcnt(0)
 ; GFX11-NEXT:    flat_store_b32 v[0:1], v2
-; GFX11-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-NEXT:    s_endpgm
 bb:
   %gep = getelementptr i16, ptr addrspace(5) %in, i64 1
   %load_lo = load i16, ptr addrspace(5) %gep
   %result = insertelement <2 x i16> <i16 -1, i16 -1>, i16 %load_lo, i32 1
   store <2 x i16> %result, ptr %out, align 4
   ret void
 }

 ; GFX10-NEXT:    s_endpgm
 ;
 ; GFX11-LABEL: test_scratch_store_b8_from_d16_hi_s:
 ; GFX11:       ; %bb.0: ; %bb
 ; GFX11-NEXT:    flat_load_b32 v0, v[0:1]
 ; GFX11-NEXT:    s_add_i32 s0, s0, 4
 ; GFX11-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX11-NEXT:    scratch_store_d16_hi_b8 off, v0, s0
-; GFX11-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-NEXT:    s_endpgm
 bb:
   %load = load <4 x i8>, ptr %in
   %element = extractelement <4 x i8> %load, i32 2
   %gep = getelementptr <4 x i8>, ptr addrspace(5) %out, i64 1
   store i8 %element, ptr addrspace(5) %gep, align 4
   ret void
 }

 ; GFX10-NEXT:    s_endpgm
 ;
 ; GFX11-LABEL: test_scratch_store_b16_from_d16_hi_s:
 ; GFX11:       ; %bb.0: ; %bb
 ; GFX11-NEXT:    flat_load_b32 v0, v[0:1]
 ; GFX11-NEXT:    s_add_i32 s0, s0, 2
 ; GFX11-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX11-NEXT:    scratch_store_d16_hi_b16 off, v0, s0
-; GFX11-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-NEXT:    s_endpgm
 bb:
   %load = load <2 x i16>, ptr %in
   %element = extractelement <2 x i16> %load, i32 1
   %gep = getelementptr <2 x i8>, ptr addrspace(5) %out, i64 1
   store i16 %element, ptr addrspace(5) %gep, align 4
   ret void
 }

 ; GFX11-LABEL: test_scratch_load_i8_zext_svs:
 ; GFX11:       ; %bb.0:
 ; GFX11-NEXT:    v_lshlrev_b32_e32 v0, 2, v0
 ; GFX11-NEXT:    s_delay_alu instid0(VALU_DEP_1)
 ; GFX11-NEXT:    v_add3_u32 v0, s0, v0, 1
 ; GFX11-NEXT:    scratch_load_u8 v0, v0, off
 ; GFX11-NEXT:    s_waitcnt vmcnt(0)
 ; GFX11-NEXT:    flat_store_b32 v[1:2], v0
-; GFX11-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-NEXT:    s_endpgm
   %voffset4 = mul i32 %voffset, 4
   %gep0 = getelementptr inbounds i8, ptr addrspace(5) %in, i32 %voffset4
   %gep = getelementptr inbounds i8, ptr addrspace(5) %gep0, i32 1
   %load = load i8, ptr addrspace(5) %gep, align 4
   %ext = zext i8 %load to i32
   store i32 %ext, ptr %out, align 4
   ret void

 ; GFX11-LABEL: test_scratch_load_i8_sext_svs:
 ; GFX11:       ; %bb.0:
 ; GFX11-NEXT:    v_lshlrev_b32_e32 v0, 2, v0
 ; GFX11-NEXT:    s_delay_alu instid0(VALU_DEP_1)
 ; GFX11-NEXT:    v_add3_u32 v0, s0, v0, 1
 ; GFX11-NEXT:    scratch_load_i8 v0, v0, off
 ; GFX11-NEXT:    s_waitcnt vmcnt(0)
 ; GFX11-NEXT:    flat_store_b32 v[1:2], v0
-; GFX11-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-NEXT:    s_endpgm
   %voffset4 = mul i32 %voffset, 4
   %gep0 = getelementptr inbounds i8, ptr addrspace(5) %in, i32 %voffset4
   %gep = getelementptr inbounds i8, ptr addrspace(5) %gep0, i32 1
   %load = load i8, ptr addrspace(5) %gep, align 4
   %ext = sext i8 %load to i32
   store i32 %ext, ptr %out, align 4
   ret void

 ; GFX11-LABEL: test_scratch_load_i16_zext_svs:
 ; GFX11:       ; %bb.0:
 ; GFX11-NEXT:    v_lshlrev_b32_e32 v0, 2, v0
 ; GFX11-NEXT:    s_delay_alu instid0(VALU_DEP_1)
 ; GFX11-NEXT:    v_add3_u32 v0, s0, v0, 2
 ; GFX11-NEXT:    scratch_load_u16 v0, v0, off
 ; GFX11-NEXT:    s_waitcnt vmcnt(0)
 ; GFX11-NEXT:    flat_store_b32 v[1:2], v0
-; GFX11-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-NEXT:    s_endpgm
   %voffset4 = mul i32 %voffset, 4
   %gep0 = getelementptr inbounds i8, ptr addrspace(5) %in, i32 %voffset4
   %gep = getelementptr inbounds i16, ptr addrspace(5) %gep0, i32 1
   %load = load i16, ptr addrspace(5) %gep, align 4
   %ext = zext i16 %load to i32
   store i32 %ext, ptr %out, align 4
   ret void

 ; GFX11-LABEL: test_scratch_load_i16_sext_svs:
 ; GFX11:       ; %bb.0:
 ; GFX11-NEXT:    v_lshlrev_b32_e32 v0, 2, v0
 ; GFX11-NEXT:    s_delay_alu instid0(VALU_DEP_1)
 ; GFX11-NEXT:    v_add3_u32 v0, s0, v0, 2
 ; GFX11-NEXT:    scratch_load_i16 v0, v0, off
 ; GFX11-NEXT:    s_waitcnt vmcnt(0)
 ; GFX11-NEXT:    flat_store_b32 v[1:2], v0
-; GFX11-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-NEXT:    s_endpgm
   %voffset4 = mul i32 %voffset, 4
   %gep0 = getelementptr inbounds i8, ptr addrspace(5) %in, i32 %voffset4
   %gep = getelementptr inbounds i16, ptr addrspace(5) %gep0, i32 1
   %load = load i16, ptr addrspace(5) %gep, align 4
   %ext = sext i16 %load to i32
   store i32 %ext, ptr %out, align 4
   ret void

 ; GFX11-LABEL: test_scratch_load_i8_zext_to_d16_lo_svs:
 ; GFX11:       ; %bb.0: ; %bb
 ; GFX11-NEXT:    v_dual_mov_b32 v3, 0xffff0000 :: v_dual_lshlrev_b32 v0, 2, v0
 ; GFX11-NEXT:    s_delay_alu instid0(VALU_DEP_1)
 ; GFX11-NEXT:    v_add3_u32 v0, s0, v0, 1
 ; GFX11-NEXT:    scratch_load_d16_u8 v3, v0, off
 ; GFX11-NEXT:    s_waitcnt vmcnt(0)
 ; GFX11-NEXT:    flat_store_b32 v[1:2], v3
-; GFX11-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-NEXT:    s_endpgm
 bb:
   %voffset4 = mul i32 %voffset, 4
   %gep0 = getelementptr inbounds i8, ptr addrspace(5) %in, i32 %voffset4
   %gep = getelementptr i8, ptr addrspace(5) %gep0, i64 1
   %load_lo = load i8, ptr addrspace(5) %gep
   %ext = zext i8 %load_lo to i16
   %result = insertelement <2 x i16> <i16 -1, i16 -1>, i16 %ext, i32 0

 ; GFX11-LABEL: test_scratch_load_i8_sext_to_d16_lo_svs:
 ; GFX11:       ; %bb.0: ; %bb
 ; GFX11-NEXT:    v_dual_mov_b32 v3, 0xffff0000 :: v_dual_lshlrev_b32 v0, 2, v0
 ; GFX11-NEXT:    s_delay_alu instid0(VALU_DEP_1)
 ; GFX11-NEXT:    v_add3_u32 v0, s0, v0, 1
 ; GFX11-NEXT:    scratch_load_d16_i8 v3, v0, off
 ; GFX11-NEXT:    s_waitcnt vmcnt(0)
 ; GFX11-NEXT:    flat_store_b32 v[1:2], v3
-; GFX11-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-NEXT:    s_endpgm
 bb:
   %voffset4 = mul i32 %voffset, 4
   %gep0 = getelementptr inbounds i8, ptr addrspace(5) %in, i32 %voffset4
   %gep = getelementptr i8, ptr addrspace(5) %gep0, i64 1
   %load_lo = load i8, ptr addrspace(5) %gep
   %ext = sext i8 %load_lo to i16
   %result = insertelement <2 x i16> <i16 -1, i16 -1>, i16 %ext, i32 0

 ; GFX11-LABEL: test_scratch_load_i16_to_d16_lo_svs:
 ; GFX11:       ; %bb.0: ; %bb
 ; GFX11-NEXT:    v_dual_mov_b32 v3, 0xffff0000 :: v_dual_lshlrev_b32 v0, 2, v0
 ; GFX11-NEXT:    s_delay_alu instid0(VALU_DEP_1)
 ; GFX11-NEXT:    v_add3_u32 v0, s0, v0, 2
 ; GFX11-NEXT:    scratch_load_d16_b16 v3, v0, off
 ; GFX11-NEXT:    s_waitcnt vmcnt(0)
 ; GFX11-NEXT:    flat_store_b32 v[1:2], v3
-; GFX11-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-NEXT:    s_endpgm
 bb:
   %voffset4 = mul i32 %voffset, 4
   %gep0 = getelementptr inbounds i8, ptr addrspace(5) %in, i32 %voffset4
   %gep = getelementptr i16, ptr addrspace(5) %gep0, i64 1
   %load_lo = load i16, ptr addrspace(5) %gep
   %result = insertelement <2 x i16> <i16 -1, i16 -1>, i16 %load_lo, i32 0
   store <2 x i16> %result, ptr %out, align 4

 ; GFX11-LABEL: test_scratch_load_i8_zext_to_d16_hi_svs:
 ; GFX11:       ; %bb.0: ; %bb
 ; GFX11-NEXT:    v_dual_mov_b32 v3, -1 :: v_dual_lshlrev_b32 v0, 2, v0
 ; GFX11-NEXT:    s_delay_alu instid0(VALU_DEP_1)
 ; GFX11-NEXT:    v_add3_u32 v0, s0, v0, 1
 ; GFX11-NEXT:    scratch_load_d16_hi_u8 v3, v0, off
 ; GFX11-NEXT:    s_waitcnt vmcnt(0)
 ; GFX11-NEXT:    flat_store_b32 v[1:2], v3
-; GFX11-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-NEXT:    s_endpgm
 bb:
   %voffset4 = mul i32 %voffset, 4
   %gep0 = getelementptr inbounds i8, ptr addrspace(5) %in, i32 %voffset4
   %gep = getelementptr i8, ptr addrspace(5) %gep0, i64 1
   %load_lo = load i8, ptr addrspace(5) %gep
   %ext = zext i8 %load_lo to i16
   %result = insertelement <2 x i16> <i16 -1, i16 -1>, i16 %ext, i32 1

 ; GFX11-LABEL: test_scratch_load_i8_sext_to_d16_hi_svs:
 ; GFX11:       ; %bb.0: ; %bb
 ; GFX11-NEXT:    v_dual_mov_b32 v3, -1 :: v_dual_lshlrev_b32 v0, 2, v0
 ; GFX11-NEXT:    s_delay_alu instid0(VALU_DEP_1)
 ; GFX11-NEXT:    v_add3_u32 v0, s0, v0, 1
 ; GFX11-NEXT:    scratch_load_d16_hi_i8 v3, v0, off
 ; GFX11-NEXT:    s_waitcnt vmcnt(0)
 ; GFX11-NEXT:    flat_store_b32 v[1:2], v3
-; GFX11-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-NEXT:    s_endpgm
 bb:
   %voffset4 = mul i32 %voffset, 4
   %gep0 = getelementptr inbounds i8, ptr addrspace(5) %in, i32 %voffset4
   %gep = getelementptr i8, ptr addrspace(5) %gep0, i64 1
   %load_lo = load i8, ptr addrspace(5) %gep
   %ext = sext i8 %load_lo to i16
   %result = insertelement <2 x i16> <i16 -1, i16 -1>, i16 %ext, i32 1

 ; GFX11-LABEL: test_scratch_load_i16_to_d16_hi_svs:
 ; GFX11:       ; %bb.0: ; %bb
 ; GFX11-NEXT:    v_dual_mov_b32 v3, -1 :: v_dual_lshlrev_b32 v0, 2, v0
 ; GFX11-NEXT:    s_delay_alu instid0(VALU_DEP_1)
 ; GFX11-NEXT:    v_add3_u32 v0, s0, v0, 2
 ; GFX11-NEXT:    scratch_load_d16_hi_b16 v3, v0, off
 ; GFX11-NEXT:    s_waitcnt vmcnt(0)
 ; GFX11-NEXT:    flat_store_b32 v[1:2], v3
-; GFX11-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-NEXT:    s_endpgm
 bb:
   %voffset4 = mul i32 %voffset, 4
   %gep0 = getelementptr inbounds i8, ptr addrspace(5) %in, i32 %voffset4
   %gep = getelementptr i16, ptr addrspace(5) %gep0, i64 1
   %load_lo = load i16, ptr addrspace(5) %gep
   %result = insertelement <2 x i16> <i16 -1, i16 -1>, i16 %load_lo, i32 1
   store <2 x i16> %result, ptr %out, align 4

 ; GFX11-LABEL: test_scratch_store_b8_from_d16_hi_svs:
 ; GFX11:       ; %bb.0: ; %bb
 ; GFX11-NEXT:    flat_load_b32 v0, v[0:1]
 ; GFX11-NEXT:    v_lshlrev_b32_e32 v1, 2, v2
 ; GFX11-NEXT:    s_delay_alu instid0(VALU_DEP_1)
 ; GFX11-NEXT:    v_add3_u32 v1, s0, v1, 4
 ; GFX11-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX11-NEXT:    scratch_store_d16_hi_b8 v1, v0, off
-; GFX11-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-NEXT:    s_endpgm
 bb:
   %load = load <4 x i8>, ptr %in
   %element = extractelement <4 x i8> %load, i32 2
   %voffset4 = mul i32 %voffset, 4
   %gep0 = getelementptr inbounds i8, ptr addrspace(5) %out, i32 %voffset4
   %gep = getelementptr <4 x i8>, ptr addrspace(5) %gep0, i64 1
   store i8 %element, ptr addrspace(5) %gep, align 4

 ; GFX11-LABEL: test_scratch_store_b16_from_d16_hi_svs:
 ; GFX11:       ; %bb.0: ; %bb
 ; GFX11-NEXT:    flat_load_b32 v0, v[0:1]
 ; GFX11-NEXT:    v_lshlrev_b32_e32 v1, 2, v2
 ; GFX11-NEXT:    s_delay_alu instid0(VALU_DEP_1)
 ; GFX11-NEXT:    v_add3_u32 v1, s0, v1, 2
 ; GFX11-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX11-NEXT:    scratch_store_d16_hi_b16 v1, v0, off
-; GFX11-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-NEXT:    s_endpgm
 bb:
   %load = load <2 x i16>, ptr %in
   %element = extractelement <2 x i16> %load, i32 1
   %voffset4 = mul i32 %voffset, 4
   %gep0 = getelementptr inbounds i8, ptr addrspace(5) %out, i32 %voffset4
   %gep = getelementptr <2 x i8>, ptr addrspace(5) %gep0, i64 1
   store i16 %element, ptr addrspace(5) %gep, align 4
   ret void
 }

llvm/test/CodeGen/AMDGPU/flat-scratch-svs.ll

 ; GFX11-SDAG-NEXT:    v_add_nc_u32_e32 v5, 2, v0
 ; GFX11-SDAG-NEXT:    v_add_nc_u32_e32 v0, 4, v0
 ; GFX11-SDAG-NEXT:    scratch_store_b8 v4, v1, off dlc
 ; GFX11-SDAG-NEXT:    s_waitcnt_vscnt null, 0x0
 ; GFX11-SDAG-NEXT:    scratch_store_b8 v5, v2, off dlc
 ; GFX11-SDAG-NEXT:    s_waitcnt_vscnt null, 0x0
 ; GFX11-SDAG-NEXT:    scratch_store_b8 v0, v3, off dlc
 ; GFX11-SDAG-NEXT:    s_waitcnt_vscnt null, 0x0
-; GFX11-SDAG-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-SDAG-NEXT:    s_endpgm
 ;
 ; GFX11-GISEL-LABEL: soff1_voff1:
 ; GFX11-GISEL:       ; %bb.0: ; %bb
 ; GFX11-GISEL-NEXT:    s_load_b32 s0, s[0:1], 0x24
 ; GFX11-GISEL-NEXT:    v_dual_mov_b32 v1, 1 :: v_dual_mov_b32 v2, 2
 ; GFX11-GISEL-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-GISEL-NEXT:    v_add3_u32 v0, 4, s0, v0
 ; GFX11-GISEL-NEXT:    s_delay_alu instid0(VALU_DEP_1)
 ; GFX11-GISEL-NEXT:    v_dual_mov_b32 v3, 4 :: v_dual_add_nc_u32 v4, 1, v0
 ; GFX11-GISEL-NEXT:    v_add_nc_u32_e32 v5, 2, v0
 ; GFX11-GISEL-NEXT:    v_add_nc_u32_e32 v0, 4, v0
 ; GFX11-GISEL-NEXT:    scratch_store_b8 v4, v1, off dlc
 ; GFX11-GISEL-NEXT:    s_waitcnt_vscnt null, 0x0
 ; GFX11-GISEL-NEXT:    scratch_store_b8 v5, v2, off dlc
 ; GFX11-GISEL-NEXT:    s_waitcnt_vscnt null, 0x0
 ; GFX11-GISEL-NEXT:    scratch_store_b8 v0, v3, off dlc
 ; GFX11-GISEL-NEXT:    s_waitcnt_vscnt null, 0x0
-; GFX11-GISEL-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-GISEL-NEXT:    s_endpgm
 bb:
   %soff1 = mul i32 %soff, 1
   %a = alloca i8, i32 64, align 4, addrspace(5)
   %as = getelementptr i8, ptr addrspace(5) %a, i32 %soff1
   %voff = call i32 @llvm.amdgcn.workitem.id.x()
   %voff1 = mul i32 %voff, 1
   %asv = getelementptr i8, ptr addrspace(5) %as, i32 %voff1

 ; GFX11-SDAG-NEXT:    v_add_nc_u32_e32 v5, 2, v0
 ; GFX11-SDAG-NEXT:    v_add_nc_u32_e32 v0, 4, v0
 ; GFX11-SDAG-NEXT:    scratch_store_b8 v4, v1, off dlc
 ; GFX11-SDAG-NEXT:    s_waitcnt_vscnt null, 0x0
 ; GFX11-SDAG-NEXT:    scratch_store_b8 v5, v2, off dlc
 ; GFX11-SDAG-NEXT:    s_waitcnt_vscnt null, 0x0
 ; GFX11-SDAG-NEXT:    scratch_store_b8 v0, v3, off dlc
 ; GFX11-SDAG-NEXT:    s_waitcnt_vscnt null, 0x0
-; GFX11-SDAG-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-SDAG-NEXT:    s_endpgm
 ;
 ; GFX11-GISEL-LABEL: soff1_voff2:
 ; GFX11-GISEL:       ; %bb.0: ; %bb
 ; GFX11-GISEL-NEXT:    s_load_b32 s0, s[0:1], 0x24
 ; GFX11-GISEL-NEXT:    v_dual_mov_b32 v1, 1 :: v_dual_lshlrev_b32 v0, 1, v0
 ; GFX11-GISEL-NEXT:    v_dual_mov_b32 v2, 2 :: v_dual_mov_b32 v3, 4
 ; GFX11-GISEL-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-GISEL-NEXT:    s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_1)
 ; GFX11-GISEL-NEXT:    v_add3_u32 v0, 4, s0, v0
 ; GFX11-GISEL-NEXT:    v_add_nc_u32_e32 v4, 1, v0
 ; GFX11-GISEL-NEXT:    v_add_nc_u32_e32 v5, 2, v0
 ; GFX11-GISEL-NEXT:    v_add_nc_u32_e32 v0, 4, v0
 ; GFX11-GISEL-NEXT:    scratch_store_b8 v4, v1, off dlc
 ; GFX11-GISEL-NEXT:    s_waitcnt_vscnt null, 0x0
 ; GFX11-GISEL-NEXT:    scratch_store_b8 v5, v2, off dlc
 ; GFX11-GISEL-NEXT:    s_waitcnt_vscnt null, 0x0
 ; GFX11-GISEL-NEXT:    scratch_store_b8 v0, v3, off dlc
 ; GFX11-GISEL-NEXT:    s_waitcnt_vscnt null, 0x0
-; GFX11-GISEL-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-GISEL-NEXT:    s_endpgm
 bb:
   %soff1 = mul i32 %soff, 1
   %a = alloca i8, i32 64, align 4, addrspace(5)
   %as = getelementptr i8, ptr addrspace(5) %a, i32 %soff1
   %voff = call i32 @llvm.amdgcn.workitem.id.x()
   %voff2 = mul i32 %voff, 2
   %asv = getelementptr i8, ptr addrspace(5) %as, i32 %voff2

 ; GFX11-SDAG-NEXT:    v_add_nc_u32_e32 v5, 2, v0
 ; GFX11-SDAG-NEXT:    v_add_nc_u32_e32 v0, 4, v0
 ; GFX11-SDAG-NEXT:    scratch_store_b8 v4, v1, off dlc
 ; GFX11-SDAG-NEXT:    s_waitcnt_vscnt null, 0x0
 ; GFX11-SDAG-NEXT:    scratch_store_b8 v5, v2, off dlc
 ; GFX11-SDAG-NEXT:    s_waitcnt_vscnt null, 0x0
 ; GFX11-SDAG-NEXT:    scratch_store_b8 v0, v3, off dlc
 ; GFX11-SDAG-NEXT:    s_waitcnt_vscnt null, 0x0
-; GFX11-SDAG-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-SDAG-NEXT:    s_endpgm
 ;
 ; GFX11-GISEL-LABEL: soff1_voff4:
 ; GFX11-GISEL:       ; %bb.0: ; %bb
 ; GFX11-GISEL-NEXT:    s_load_b32 s0, s[0:1], 0x24
 ; GFX11-GISEL-NEXT:    v_dual_mov_b32 v1, 1 :: v_dual_lshlrev_b32 v0, 2, v0
 ; GFX11-GISEL-NEXT:    v_dual_mov_b32 v2, 2 :: v_dual_mov_b32 v3, 4
 ; GFX11-GISEL-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-GISEL-NEXT:    s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_1)
 ; GFX11-GISEL-NEXT:    v_add3_u32 v0, 4, s0, v0
 ; GFX11-GISEL-NEXT:    v_add_nc_u32_e32 v4, 1, v0
 ; GFX11-GISEL-NEXT:    v_add_nc_u32_e32 v5, 2, v0
 ; GFX11-GISEL-NEXT:    v_add_nc_u32_e32 v0, 4, v0
 ; GFX11-GISEL-NEXT:    scratch_store_b8 v4, v1, off dlc
 ; GFX11-GISEL-NEXT:    s_waitcnt_vscnt null, 0x0
 ; GFX11-GISEL-NEXT:    scratch_store_b8 v5, v2, off dlc
 ; GFX11-GISEL-NEXT:    s_waitcnt_vscnt null, 0x0
 ; GFX11-GISEL-NEXT:    scratch_store_b8 v0, v3, off dlc
 ; GFX11-GISEL-NEXT:    s_waitcnt_vscnt null, 0x0
-; GFX11-GISEL-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-GISEL-NEXT:    s_endpgm
 bb:
   %soff1 = mul i32 %soff, 1
   %a = alloca i8, i32 64, align 4, addrspace(5)
   %as = getelementptr i8, ptr addrspace(5) %a, i32 %soff1
   %voff = call i32 @llvm.amdgcn.workitem.id.x()
   %voff4 = mul i32 %voff, 4
   %asv = getelementptr i8, ptr addrspace(5) %as, i32 %voff4

 ; GFX11-SDAG-NEXT:    v_add_nc_u32_e32 v5, 2, v0
 ; GFX11-SDAG-NEXT:    v_add_nc_u32_e32 v0, 4, v0
 ; GFX11-SDAG-NEXT:    scratch_store_b8 v4, v1, off dlc
 ; GFX11-SDAG-NEXT:    s_waitcnt_vscnt null, 0x0
 ; GFX11-SDAG-NEXT:    scratch_store_b8 v5, v2, off dlc
 ; GFX11-SDAG-NEXT:    s_waitcnt_vscnt null, 0x0
 ; GFX11-SDAG-NEXT:    scratch_store_b8 v0, v3, off dlc
 ; GFX11-SDAG-NEXT:    s_waitcnt_vscnt null, 0x0
-; GFX11-SDAG-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-SDAG-NEXT:    s_endpgm
 ;
 ; GFX11-GISEL-LABEL: soff2_voff1:
 ; GFX11-GISEL:       ; %bb.0: ; %bb
 ; GFX11-GISEL-NEXT:    s_load_b32 s0, s[0:1], 0x24
 ; GFX11-GISEL-NEXT:    v_dual_mov_b32 v1, 1 :: v_dual_mov_b32 v2, 2
 ; GFX11-GISEL-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-GISEL-NEXT:    s_lshl_b32 s0, s0, 1
 ; GFX11-GISEL-NEXT:    s_delay_alu instid0(SALU_CYCLE_1) | instskip(NEXT) | instid1(VALU_DEP_1)
 ; GFX11-GISEL-NEXT:    v_add3_u32 v0, 4, s0, v0
 ; GFX11-GISEL-NEXT:    v_dual_mov_b32 v3, 4 :: v_dual_add_nc_u32 v4, 1, v0
 ; GFX11-GISEL-NEXT:    v_add_nc_u32_e32 v5, 2, v0
 ; GFX11-GISEL-NEXT:    v_add_nc_u32_e32 v0, 4, v0
 ; GFX11-GISEL-NEXT:    scratch_store_b8 v4, v1, off dlc
 ; GFX11-GISEL-NEXT:    s_waitcnt_vscnt null, 0x0
 ; GFX11-GISEL-NEXT:    scratch_store_b8 v5, v2, off dlc
 ; GFX11-GISEL-NEXT:    s_waitcnt_vscnt null, 0x0
 ; GFX11-GISEL-NEXT:    scratch_store_b8 v0, v3, off dlc
 ; GFX11-GISEL-NEXT:    s_waitcnt_vscnt null, 0x0
-; GFX11-GISEL-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-GISEL-NEXT:    s_endpgm
 bb:
   %soff2 = mul i32 %soff, 2
   %a = alloca i8, i32 64, align 4, addrspace(5)
   %as = getelementptr i8, ptr addrspace(5) %a, i32 %soff2
   %voff = call i32 @llvm.amdgcn.workitem.id.x()
   %voff1 = mul i32 %voff, 1
   %asv = getelementptr i8, ptr addrspace(5) %as, i32 %voff1

 ; GFX11-SDAG-NEXT:    v_add_nc_u32_e32 v5, 2, v0
 ; GFX11-SDAG-NEXT:    v_add_nc_u32_e32 v0, 4, v0
 ; GFX11-SDAG-NEXT:    scratch_store_b8 v4, v1, off dlc
 ; GFX11-SDAG-NEXT:    s_waitcnt_vscnt null, 0x0
 ; GFX11-SDAG-NEXT:    scratch_store_b8 v5, v2, off dlc
 ; GFX11-SDAG-NEXT:    s_waitcnt_vscnt null, 0x0
 ; GFX11-SDAG-NEXT:    scratch_store_b8 v0, v3, off dlc
 ; GFX11-SDAG-NEXT:    s_waitcnt_vscnt null, 0x0
-; GFX11-SDAG-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-SDAG-NEXT:    s_endpgm
 ;
 ; GFX11-GISEL-LABEL: soff2_voff2:
 ; GFX11-GISEL:       ; %bb.0: ; %bb
 ; GFX11-GISEL-NEXT:    s_load_b32 s0, s[0:1], 0x24
 ; GFX11-GISEL-NEXT:    v_dual_mov_b32 v1, 1 :: v_dual_lshlrev_b32 v0, 1, v0
 ; GFX11-GISEL-NEXT:    v_dual_mov_b32 v2, 2 :: v_dual_mov_b32 v3, 4
 ; GFX11-GISEL-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-GISEL-NEXT:    s_lshl_b32 s0, s0, 1
 ; GFX11-GISEL-NEXT:    s_delay_alu instid0(VALU_DEP_2) | instid1(SALU_CYCLE_1)
 ; GFX11-GISEL-NEXT:    v_add3_u32 v0, 4, s0, v0
 ; GFX11-GISEL-NEXT:    s_delay_alu instid0(VALU_DEP_1)
 ; GFX11-GISEL-NEXT:    v_add_nc_u32_e32 v4, 1, v0
 ; GFX11-GISEL-NEXT:    v_add_nc_u32_e32 v5, 2, v0
 ; GFX11-GISEL-NEXT:    v_add_nc_u32_e32 v0, 4, v0
 ; GFX11-GISEL-NEXT:    scratch_store_b8 v4, v1, off dlc
 ; GFX11-GISEL-NEXT:    s_waitcnt_vscnt null, 0x0
 ; GFX11-GISEL-NEXT:    scratch_store_b8 v5, v2, off dlc
 ; GFX11-GISEL-NEXT:    s_waitcnt_vscnt null, 0x0
 ; GFX11-GISEL-NEXT:    scratch_store_b8 v0, v3, off dlc
 ; GFX11-GISEL-NEXT:    s_waitcnt_vscnt null, 0x0
-; GFX11-GISEL-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-GISEL-NEXT:    s_endpgm
 bb:
   %soff2 = mul i32 %soff, 2
   %a = alloca i8, i32 64, align 4, addrspace(5)
   %as = getelementptr i8, ptr addrspace(5) %a, i32 %soff2
   %voff = call i32 @llvm.amdgcn.workitem.id.x()
   %voff2 = mul i32 %voff, 2
   %asv = getelementptr i8, ptr addrspace(5) %as, i32 %voff2

 ; GFX11-SDAG-NEXT:    v_add_nc_u32_e32 v5, 2, v0
 ; GFX11-SDAG-NEXT:    v_add_nc_u32_e32 v0, 4, v0
 ; GFX11-SDAG-NEXT:    scratch_store_b8 v4, v1, off dlc
 ; GFX11-SDAG-NEXT:    s_waitcnt_vscnt null, 0x0
 ; GFX11-SDAG-NEXT:    scratch_store_b8 v5, v2, off dlc
 ; GFX11-SDAG-NEXT:    s_waitcnt_vscnt null, 0x0
 ; GFX11-SDAG-NEXT:    scratch_store_b8 v0, v3, off dlc
 ; GFX11-SDAG-NEXT:    s_waitcnt_vscnt null, 0x0
-; GFX11-SDAG-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-SDAG-NEXT:    s_endpgm
 ;
 ; GFX11-GISEL-LABEL: soff2_voff4:
 ; GFX11-GISEL:       ; %bb.0: ; %bb
 ; GFX11-GISEL-NEXT:    s_load_b32 s0, s[0:1], 0x24
 ; GFX11-GISEL-NEXT:    v_dual_mov_b32 v1, 1 :: v_dual_lshlrev_b32 v0, 2, v0
 ; GFX11-GISEL-NEXT:    v_dual_mov_b32 v2, 2 :: v_dual_mov_b32 v3, 4
 ; GFX11-GISEL-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-GISEL-NEXT:    s_lshl_b32 s0, s0, 1
 ; GFX11-GISEL-NEXT:    s_delay_alu instid0(VALU_DEP_2) | instid1(SALU_CYCLE_1)
 ; GFX11-GISEL-NEXT:    v_add3_u32 v0, 4, s0, v0
 ; GFX11-GISEL-NEXT:    s_delay_alu instid0(VALU_DEP_1)
 ; GFX11-GISEL-NEXT:    v_add_nc_u32_e32 v4, 1, v0
 ; GFX11-GISEL-NEXT:    v_add_nc_u32_e32 v5, 2, v0
 ; GFX11-GISEL-NEXT:    v_add_nc_u32_e32 v0, 4, v0
 ; GFX11-GISEL-NEXT:    scratch_store_b8 v4, v1, off dlc
 ; GFX11-GISEL-NEXT:    s_waitcnt_vscnt null, 0x0
 ; GFX11-GISEL-NEXT:    scratch_store_b8 v5, v2, off dlc
 ; GFX11-GISEL-NEXT:    s_waitcnt_vscnt null, 0x0
 ; GFX11-GISEL-NEXT:    scratch_store_b8 v0, v3, off dlc
 ; GFX11-GISEL-NEXT:    s_waitcnt_vscnt null, 0x0
-; GFX11-GISEL-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-GISEL-NEXT:    s_endpgm
 bb:
   %soff2 = mul i32 %soff, 2
   %a = alloca i8, i32 64, align 4, addrspace(5)
   %as = getelementptr i8, ptr addrspace(5) %a, i32 %soff2
   %voff = call i32 @llvm.amdgcn.workitem.id.x()
   %voff4 = mul i32 %voff, 4
   %asv = getelementptr i8, ptr addrspace(5) %as, i32 %voff4

 ; GFX11-SDAG-NEXT:    v_add_nc_u32_e32 v5, 2, v0
 ; GFX11-SDAG-NEXT:    v_add_nc_u32_e32 v0, 4, v0
 ; GFX11-SDAG-NEXT:    scratch_store_b8 v4, v1, off dlc
 ; GFX11-SDAG-NEXT:    s_waitcnt_vscnt null, 0x0
 ; GFX11-SDAG-NEXT:    scratch_store_b8 v5, v2, off dlc
 ; GFX11-SDAG-NEXT:    s_waitcnt_vscnt null, 0x0
 ; GFX11-SDAG-NEXT:    scratch_store_b8 v0, v3, off dlc
 ; GFX11-SDAG-NEXT:    s_waitcnt_vscnt null, 0x0
-; GFX11-SDAG-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-SDAG-NEXT:    s_endpgm
 ;
 ; GFX11-GISEL-LABEL: soff4_voff1:
 ; GFX11-GISEL:       ; %bb.0: ; %bb
 ; GFX11-GISEL-NEXT:    s_load_b32 s0, s[0:1], 0x24
 ; GFX11-GISEL-NEXT:    v_dual_mov_b32 v1, 1 :: v_dual_mov_b32 v2, 2
 ; GFX11-GISEL-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-GISEL-NEXT:    s_lshl_b32 s0, s0, 2
 ; GFX11-GISEL-NEXT:    s_delay_alu instid0(SALU_CYCLE_1) | instskip(NEXT) | instid1(VALU_DEP_1)
 ; GFX11-GISEL-NEXT:    v_add3_u32 v0, 4, s0, v0
 ; GFX11-GISEL-NEXT:    v_dual_mov_b32 v3, 4 :: v_dual_add_nc_u32 v4, 1, v0
 ; GFX11-GISEL-NEXT:    v_add_nc_u32_e32 v5, 2, v0
 ; GFX11-GISEL-NEXT:    v_add_nc_u32_e32 v0, 4, v0
 ; GFX11-GISEL-NEXT:    scratch_store_b8 v4, v1, off dlc
 ; GFX11-GISEL-NEXT:    s_waitcnt_vscnt null, 0x0
 ; GFX11-GISEL-NEXT:    scratch_store_b8 v5, v2, off dlc
 ; GFX11-GISEL-NEXT:    s_waitcnt_vscnt null, 0x0
 ; GFX11-GISEL-NEXT:    scratch_store_b8 v0, v3, off dlc
 ; GFX11-GISEL-NEXT:    s_waitcnt_vscnt null, 0x0
-; GFX11-GISEL-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-GISEL-NEXT:    s_endpgm
 bb:
   %soff4 = mul i32 %soff, 4
   %a = alloca i8, i32 64, align 4, addrspace(5)
   %as = getelementptr i8, ptr addrspace(5) %a, i32 %soff4
   %voff = call i32 @llvm.amdgcn.workitem.id.x()
   %voff1 = mul i32 %voff, 1
   %asv = getelementptr i8, ptr addrspace(5) %as, i32 %voff1

 ; GFX11-SDAG-NEXT:    v_add_nc_u32_e32 v5, 2, v0
 ; GFX11-SDAG-NEXT:    v_add_nc_u32_e32 v0, 4, v0
 ; GFX11-SDAG-NEXT:    scratch_store_b8 v4, v1, off dlc
 ; GFX11-SDAG-NEXT:    s_waitcnt_vscnt null, 0x0
 ; GFX11-SDAG-NEXT:    scratch_store_b8 v5, v2, off dlc
 ; GFX11-SDAG-NEXT:    s_waitcnt_vscnt null, 0x0
 ; GFX11-SDAG-NEXT:    scratch_store_b8 v0, v3, off dlc
 ; GFX11-SDAG-NEXT:    s_waitcnt_vscnt null, 0x0
-; GFX11-SDAG-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-SDAG-NEXT:    s_endpgm
 ;
 ; GFX11-GISEL-LABEL: soff4_voff2:
 ; GFX11-GISEL:       ; %bb.0: ; %bb
 ; GFX11-GISEL-NEXT:    s_load_b32 s0, s[0:1], 0x24
 ; GFX11-GISEL-NEXT:    v_dual_mov_b32 v1, 1 :: v_dual_lshlrev_b32 v0, 1, v0
 ; GFX11-GISEL-NEXT:    v_dual_mov_b32 v2, 2 :: v_dual_mov_b32 v3, 4
 ; GFX11-GISEL-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-GISEL-NEXT:    s_lshl_b32 s0, s0, 2
 ; GFX11-GISEL-NEXT:    s_delay_alu instid0(VALU_DEP_2) | instid1(SALU_CYCLE_1)
 ; GFX11-GISEL-NEXT:    v_add3_u32 v0, 4, s0, v0
 ; GFX11-GISEL-NEXT:    s_delay_alu instid0(VALU_DEP_1)
 ; GFX11-GISEL-NEXT:    v_add_nc_u32_e32 v4, 1, v0
 ; GFX11-GISEL-NEXT:    v_add_nc_u32_e32 v5, 2, v0
 ; GFX11-GISEL-NEXT:    v_add_nc_u32_e32 v0, 4, v0
 ; GFX11-GISEL-NEXT:    scratch_store_b8 v4, v1, off dlc
 ; GFX11-GISEL-NEXT:    s_waitcnt_vscnt null, 0x0
 ; GFX11-GISEL-NEXT:    scratch_store_b8 v5, v2, off dlc
 ; GFX11-GISEL-NEXT:    s_waitcnt_vscnt null, 0x0
 ; GFX11-GISEL-NEXT:    scratch_store_b8 v0, v3, off dlc
 ; GFX11-GISEL-NEXT:    s_waitcnt_vscnt null, 0x0
-; GFX11-GISEL-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-GISEL-NEXT:    s_endpgm
 bb:
   %soff4 = mul i32 %soff, 4
   %a = alloca i8, i32 64, align 4, addrspace(5)
   %as = getelementptr i8, ptr addrspace(5) %a, i32 %soff4
   %voff = call i32 @llvm.amdgcn.workitem.id.x()
   %voff2 = mul i32 %voff, 2
   %asv = getelementptr i8, ptr addrspace(5) %as, i32 %voff2

 ; GFX11-SDAG-NEXT:    v_or_b32_e32 v5, 2, v0
 ; GFX11-SDAG-NEXT:    v_add_nc_u32_e32 v0, 4, v0
 ; GFX11-SDAG-NEXT:    scratch_store_b8 v4, v1, off dlc
 ; GFX11-SDAG-NEXT:    s_waitcnt_vscnt null, 0x0
 ; GFX11-SDAG-NEXT:    scratch_store_b8 v5, v2, off dlc
 ; GFX11-SDAG-NEXT:    s_waitcnt_vscnt null, 0x0
 ; GFX11-SDAG-NEXT:    scratch_store_b8 v0, v3, off dlc
 ; GFX11-SDAG-NEXT:    s_waitcnt_vscnt null, 0x0
-; GFX11-SDAG-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-SDAG-NEXT:    s_endpgm
 ;
 ; GFX11-GISEL-LABEL: soff4_voff4:
 ; GFX11-GISEL:       ; %bb.0: ; %bb
 ; GFX11-GISEL-NEXT:    s_load_b32 s0, s[0:1], 0x24
 ; GFX11-GISEL-NEXT:    v_dual_mov_b32 v1, 1 :: v_dual_lshlrev_b32 v0, 2, v0
 ; GFX11-GISEL-NEXT:    v_dual_mov_b32 v2, 2 :: v_dual_mov_b32 v3, 4
 ; GFX11-GISEL-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-GISEL-NEXT:    s_lshl_b32 s0, s0, 2
 ; GFX11-GISEL-NEXT:    s_delay_alu instid0(VALU_DEP_2) | instid1(SALU_CYCLE_1)
 ; GFX11-GISEL-NEXT:    v_add3_u32 v0, 4, s0, v0
 ; GFX11-GISEL-NEXT:    s_delay_alu instid0(VALU_DEP_1)
 ; GFX11-GISEL-NEXT:    v_add_nc_u32_e32 v4, 1, v0
 ; GFX11-GISEL-NEXT:    v_add_nc_u32_e32 v5, 2, v0
 ; GFX11-GISEL-NEXT:    v_add_nc_u32_e32 v0, 4, v0
 ; GFX11-GISEL-NEXT:    scratch_store_b8 v4, v1, off dlc
 ; GFX11-GISEL-NEXT:    s_waitcnt_vscnt null, 0x0
 ; GFX11-GISEL-NEXT:    scratch_store_b8 v5, v2, off dlc
 ; GFX11-GISEL-NEXT:    s_waitcnt_vscnt null, 0x0
 ; GFX11-GISEL-NEXT:    scratch_store_b8 v0, v3, off dlc
 ; GFX11-GISEL-NEXT:    s_waitcnt_vscnt null, 0x0
-; GFX11-GISEL-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-GISEL-NEXT:    s_endpgm
 bb:
   %soff4 = mul i32 %soff, 4
   %a = alloca i8, i32 64, align 4, addrspace(5)
   %as = getelementptr i8, ptr addrspace(5) %a, i32 %soff4
   %voff = call i32 @llvm.amdgcn.workitem.id.x()
   %voff4 = mul i32 %voff, 4
   %asv = getelementptr i8, ptr addrspace(5) %as, i32 %voff4
   %p1 = getelementptr i8, ptr addrspace(5) %asv, i32 1
   store volatile i8 1, ptr addrspace(5) %p1
   %p2 = getelementptr i8, ptr addrspace(5) %asv, i32 2
   store volatile i8 2, ptr addrspace(5) %p2
   %p4 = getelementptr i8, ptr addrspace(5) %asv, i32 4
   store volatile i8 4, ptr addrspace(5) %p4
   ret void
 }

llvm/test/CodeGen/AMDGPU/flat-scratch.ll

 ; GFX11-NEXT:    s_mov_b32 s3, s0
 ; GFX11-NEXT:    v_dual_mov_b32 v0, s0 :: v_dual_mov_b32 v1, s1
 ; GFX11-NEXT:    v_dual_mov_b32 v2, s2 :: v_dual_mov_b32 v3, s3
 ; GFX11-NEXT:    s_clause 0x3
 ; GFX11-NEXT:    scratch_store_b128 off, v[0:3], off offset:52
 ; GFX11-NEXT:    scratch_store_b128 off, v[0:3], off offset:36
 ; GFX11-NEXT:    scratch_store_b128 off, v[0:3], off offset:20
 ; GFX11-NEXT:    scratch_store_b128 off, v[0:3], off offset:4
-; GFX11-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-NEXT:    s_endpgm
 ;
 ; GFX9-PAL-LABEL: zero_init_kernel:
 ; GFX9-PAL:       ; %bb.0:
 ; GFX9-PAL-NEXT:    s_getpc_b64 s[2:3]
 ; GFX9-PAL-NEXT:    s_mov_b32 s2, s0
 ; GFX9-PAL-NEXT:    s_load_dwordx2 s[2:3], s[2:3], 0x0
 ; GFX9-PAL-NEXT:    s_mov_b32 s0, 0

 ; GFX11-PAL-NEXT:    s_mov_b32 s3, s0
 ; GFX11-PAL-NEXT:    v_dual_mov_b32 v0, s0 :: v_dual_mov_b32 v1, s1
 ; GFX11-PAL-NEXT:    v_dual_mov_b32 v2, s2 :: v_dual_mov_b32 v3, s3
 ; GFX11-PAL-NEXT:    s_clause 0x3
 ; GFX11-PAL-NEXT:    scratch_store_b128 off, v[0:3], off offset:52
 ; GFX11-PAL-NEXT:    scratch_store_b128 off, v[0:3], off offset:36
 ; GFX11-PAL-NEXT:    scratch_store_b128 off, v[0:3], off offset:20
 ; GFX11-PAL-NEXT:    scratch_store_b128 off, v[0:3], off offset:4
-; GFX11-PAL-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-PAL-NEXT:    s_endpgm
   %alloca = alloca [32 x i16], align 2, addrspace(5)
   call void @llvm.memset.p5.i64(ptr addrspace(5) align 2 dereferenceable(64) %alloca, i8 0, i64 64, i1 false)
   ret void
 }
 
 define void @zero_init_foo() {
 ; GFX9-LABEL: zero_init_foo:

 ; GFX11-NEXT:    s_mov_b32 s3, s0
 ; GFX11-NEXT:    v_dual_mov_b32 v0, s0 :: v_dual_mov_b32 v1, s1
 ; GFX11-NEXT:    v_dual_mov_b32 v2, s2 :: v_dual_mov_b32 v3, s3
 ; GFX11-NEXT:    s_clause 0x3
 ; GFX11-NEXT:    scratch_store_b128 off, v[0:3], off offset:260
 ; GFX11-NEXT:    scratch_store_b128 off, v[0:3], off offset:276
 ; GFX11-NEXT:    scratch_store_b128 off, v[0:3], off offset:292
 ; GFX11-NEXT:    scratch_store_b128 off, v[0:3], off offset:308
-; GFX11-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-NEXT:    s_endpgm
 ;
 ; GFX9-PAL-LABEL: zero_init_small_offset_kernel:
 ; GFX9-PAL:       ; %bb.0:
 ; GFX9-PAL-NEXT:    s_getpc_b64 s[2:3]
 ; GFX9-PAL-NEXT:    s_mov_b32 s2, s0
 ; GFX9-PAL-NEXT:    s_load_dwordx2 s[2:3], s[2:3], 0x0
 ; GFX9-PAL-NEXT:    s_mov_b32 s0, 0

 ; GFX11-PAL-NEXT:    s_mov_b32 s3, s0
 ; GFX11-PAL-NEXT:    v_dual_mov_b32 v0, s0 :: v_dual_mov_b32 v1, s1
 ; GFX11-PAL-NEXT:    v_dual_mov_b32 v2, s2 :: v_dual_mov_b32 v3, s3
 ; GFX11-PAL-NEXT:    s_clause 0x3
 ; GFX11-PAL-NEXT:    scratch_store_b128 off, v[0:3], off offset:260
 ; GFX11-PAL-NEXT:    scratch_store_b128 off, v[0:3], off offset:276
 ; GFX11-PAL-NEXT:    scratch_store_b128 off, v[0:3], off offset:292
 ; GFX11-PAL-NEXT:    scratch_store_b128 off, v[0:3], off offset:308
-; GFX11-PAL-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-PAL-NEXT:    s_endpgm
   %padding = alloca [64 x i32], align 4, addrspace(5)
   %alloca = alloca [32 x i16], align 2, addrspace(5)
   %pad_gep = getelementptr inbounds [64 x i32], ptr addrspace(5) %padding, i32 0, i32 undef
   %pad_load = load volatile i32, ptr addrspace(5) %pad_gep, align 4
   call void @llvm.memset.p5.i64(ptr addrspace(5) align 2 dereferenceable(64) %alloca, i8 0, i64 64, i1 false)
   ret void
 }

 ; GFX11-NEXT:    v_dual_mov_b32 v0, s0 :: v_dual_mov_b32 v1, s1
 ; GFX11-NEXT:    v_dual_mov_b32 v2, s2 :: v_dual_mov_b32 v3, s3
 ; GFX11-NEXT:    s_movk_i32 s0, 0x4004
 ; GFX11-NEXT:    s_clause 0x3
 ; GFX11-NEXT:    scratch_store_b128 off, v[0:3], s0
 ; GFX11-NEXT:    scratch_store_b128 off, v[0:3], s0 offset:16
 ; GFX11-NEXT:    scratch_store_b128 off, v[0:3], s0 offset:32
 ; GFX11-NEXT:    scratch_store_b128 off, v[0:3], s0 offset:48
-; GFX11-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-NEXT:    s_endpgm
 ;
 ; GFX9-PAL-LABEL: zero_init_large_offset_kernel:
 ; GFX9-PAL:       ; %bb.0:
 ; GFX9-PAL-NEXT:    s_getpc_b64 s[2:3]
 ; GFX9-PAL-NEXT:    s_mov_b32 s2, s0
 ; GFX9-PAL-NEXT:    s_load_dwordx2 s[2:3], s[2:3], 0x0
 ; GFX9-PAL-NEXT:    s_mov_b32 s0, 0

 ; GFX11-PAL-NEXT:    v_dual_mov_b32 v0, s0 :: v_dual_mov_b32 v1, s1
 ; GFX11-PAL-NEXT:    v_dual_mov_b32 v2, s2 :: v_dual_mov_b32 v3, s3
 ; GFX11-PAL-NEXT:    s_movk_i32 s0, 0x4004
 ; GFX11-PAL-NEXT:    s_clause 0x3
 ; GFX11-PAL-NEXT:    scratch_store_b128 off, v[0:3], s0
 ; GFX11-PAL-NEXT:    scratch_store_b128 off, v[0:3], s0 offset:16
 ; GFX11-PAL-NEXT:    scratch_store_b128 off, v[0:3], s0 offset:32
 ; GFX11-PAL-NEXT:    scratch_store_b128 off, v[0:3], s0 offset:48
-; GFX11-PAL-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-PAL-NEXT:    s_endpgm
   %padding = alloca [4096 x i32], align 4, addrspace(5)
   %alloca = alloca [32 x i16], align 2, addrspace(5)
   %pad_gep = getelementptr inbounds [4096 x i32], ptr addrspace(5) %padding, i32 0, i32 undef
   %pad_load = load volatile i32, ptr addrspace(5) %pad_gep, align 4
   call void @llvm.memset.p5.i64(ptr addrspace(5) align 2 dereferenceable(64) %alloca, i8 0, i64 64, i1 false)
   ret void
 }

llvm/test/CodeGen/AMDGPU/fp-min-max-flat-atomics.ll

 ; GFX10-LABEL: flat_atomic_fmin_f32_noret:
 ; GFX10:       ; %bb.0:
 ; GFX10-NEXT:    flat_atomic_fmin v[0:1], v2
 ; GFX10-NEXT:    s_endpgm
 ;
 ; GFX11-LABEL: flat_atomic_fmin_f32_noret:
 ; GFX11:       ; %bb.0:
 ; GFX11-NEXT:    flat_atomic_min_f32 v[0:1], v2
-; GFX11-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-NEXT:    s_endpgm
   %ret = call float @llvm.amdgcn.flat.atomic.fmin.f32.p1.f32(ptr %ptr, float %data)
   ret void
 }
 
 define amdgpu_cs void @flat_atomic_fmax_f32_noret(ptr %ptr, float %data) {
 ; GFX10-LABEL: flat_atomic_fmax_f32_noret:
 ; GFX10:       ; %bb.0:
 ; GFX10-NEXT:    flat_atomic_fmax v[0:1], v2
 ; GFX10-NEXT:    s_endpgm
 ;
 ; GFX11-LABEL: flat_atomic_fmax_f32_noret:
 ; GFX11:       ; %bb.0:
 ; GFX11-NEXT:    flat_atomic_max_f32 v[0:1], v2
-; GFX11-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-NEXT:    s_endpgm
   %ret = call float @llvm.amdgcn.flat.atomic.fmax.f32.p1.f32(ptr %ptr, float %data)
   ret void
 }
 
 define amdgpu_cs float @flat_atomic_fmin_f32_rtn(ptr %ptr, float %data, ptr %out) {
 ; GFX10-LABEL: flat_atomic_fmin_f32_rtn:
 ; GFX10:       ; %bb.0:

llvm/test/CodeGen/AMDGPU/lds-atomic-fmin-fmax.ll

 ; GFX11-NEXT:    s_lshl_b32 s2, s2, 4
 ; GFX11-NEXT:    v_dual_mov_b32 v3, s1 :: v_dual_mov_b32 v2, s2
 ; GFX11-NEXT:    ds_min_rtn_f32 v1, v1, v0 offset:32
 ; GFX11-NEXT:    ds_min_f32 v2, v0 offset:64
 ; GFX11-NEXT:    s_waitcnt lgkmcnt(1)
 ; GFX11-NEXT:    ds_min_rtn_f32 v0, v3, v1
 ; GFX11-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-NEXT:    scratch_store_b32 off, v0, s0
-; GFX11-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-NEXT:    s_endpgm
 ;
 ; G_SI-LABEL: lds_ds_fmin:
 ; G_SI:       ; %bb.0:
 ; G_SI-NEXT:    s_mov_b32 s4, SCRATCH_RSRC_DWORD0
 ; G_SI-NEXT:    s_mov_b32 s5, SCRATCH_RSRC_DWORD1
 ; G_SI-NEXT:    s_load_dword s2, s[0:1], 0xb
 ; G_SI-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9

 ; G_GFX11-NEXT:    v_mov_b32_e32 v0, s3
 ; G_GFX11-NEXT:    v_mov_b32_e32 v2, s2
 ; G_GFX11-NEXT:    ds_min_rtn_f32 v0, v0, v1
 ; G_GFX11-NEXT:    ds_min_f32 v2, v1
 ; G_GFX11-NEXT:    s_waitcnt lgkmcnt(1)
 ; G_GFX11-NEXT:    ds_min_rtn_f32 v0, v3, v0
 ; G_GFX11-NEXT:    s_waitcnt lgkmcnt(0)
 ; G_GFX11-NEXT:    scratch_store_b32 off, v0, s0
-; G_GFX11-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; G_GFX11-NEXT:    s_endpgm
   %idx.add = add nuw i32 %idx, 4
   %shl0 = shl i32 %idx.add, 3
   %shl1 = shl i32 %idx.add, 4
   %ptr0 = inttoptr i32 %shl0 to ptr addrspace(3)
   %ptr1 = inttoptr i32 %shl1 to ptr addrspace(3)
   %a1 = call float @llvm.amdgcn.ds.fmin.f32(ptr addrspace(3) %ptr0, float 4.2e+1, i32 0, i32 0, i1 false)
   %a2 = call float @llvm.amdgcn.ds.fmin.f32(ptr addrspace(3) %ptr1, float 4.2e+1, i32 0, i32 0, i1 false)

 ; GFX11-NEXT:    s_lshl_b32 s2, s2, 4
 ; GFX11-NEXT:    v_dual_mov_b32 v3, s1 :: v_dual_mov_b32 v2, s2
 ; GFX11-NEXT:    ds_max_rtn_f32 v1, v1, v0 offset:32
 ; GFX11-NEXT:    ds_max_f32 v2, v0 offset:64
 ; GFX11-NEXT:    s_waitcnt lgkmcnt(1)
 ; GFX11-NEXT:    ds_max_rtn_f32 v0, v3, v1
 ; GFX11-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-NEXT:    scratch_store_b32 off, v0, s0
-; GFX11-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-NEXT:    s_endpgm
 ;
 ; G_SI-LABEL: lds_ds_fmax:
 ; G_SI:       ; %bb.0:
 ; G_SI-NEXT:    s_mov_b32 s4, SCRATCH_RSRC_DWORD0
 ; G_SI-NEXT:    s_mov_b32 s5, SCRATCH_RSRC_DWORD1
 ; G_SI-NEXT:    s_load_dword s2, s[0:1], 0xb
 ; G_SI-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9

 ; G_GFX11-NEXT:    v_mov_b32_e32 v0, s3
 ; G_GFX11-NEXT:    v_mov_b32_e32 v2, s2
 ; G_GFX11-NEXT:    ds_max_rtn_f32 v0, v0, v1
 ; G_GFX11-NEXT:    ds_max_f32 v2, v1
 ; G_GFX11-NEXT:    s_waitcnt lgkmcnt(1)
 ; G_GFX11-NEXT:    ds_max_rtn_f32 v0, v3, v0
 ; G_GFX11-NEXT:    s_waitcnt lgkmcnt(0)
 ; G_GFX11-NEXT:    scratch_store_b32 off, v0, s0
-; G_GFX11-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; G_GFX11-NEXT:    s_endpgm
   %idx.add = add nuw i32 %idx, 4
   %shl0 = shl i32 %idx.add, 3
   %shl1 = shl i32 %idx.add, 4
   %ptr0 = inttoptr i32 %shl0 to ptr addrspace(3)
   %ptr1 = inttoptr i32 %shl1 to ptr addrspace(3)
   %a1 = call float @llvm.amdgcn.ds.fmax.f32(ptr addrspace(3) %ptr0, float 4.2e+1, i32 0, i32 0, i1 false)
   %a2 = call float @llvm.amdgcn.ds.fmax.f32(ptr addrspace(3) %ptr1, float 4.2e+1, i32 0, i32 0, i1 false)

 ; GFX11-NEXT:    s_delay_alu instid0(SALU_CYCLE_1)
 ; GFX11-NEXT:    v_mov_b32_e32 v4, s2
 ; GFX11-NEXT:    ds_min_rtn_f64 v[2:3], v2, v[0:1] offset:32
 ; GFX11-NEXT:    ds_min_f64 v4, v[0:1] offset:64
 ; GFX11-NEXT:    s_waitcnt lgkmcnt(1)
 ; GFX11-NEXT:    ds_min_rtn_f64 v[0:1], v5, v[2:3]
 ; GFX11-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-NEXT:    scratch_store_b64 off, v[0:1], s0
-; GFX11-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-NEXT:    s_endpgm
 ;
 ; G_SI-LABEL: lds_ds_fmin_f64:
 ; G_SI:       ; %bb.0:
 ; G_SI-NEXT:    s_mov_b32 s8, SCRATCH_RSRC_DWORD0
 ; G_SI-NEXT:    s_mov_b32 s9, SCRATCH_RSRC_DWORD1
 ; G_SI-NEXT:    s_load_dword s4, s[0:1], 0xb
 ; G_SI-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9

 ; G_GFX11-NEXT:    s_delay_alu instid0(SALU_CYCLE_1)
 ; G_GFX11-NEXT:    v_mov_b32_e32 v4, s2
 ; G_GFX11-NEXT:    ds_min_rtn_f64 v[2:3], v2, v[0:1]
 ; G_GFX11-NEXT:    ds_min_f64 v4, v[0:1]
 ; G_GFX11-NEXT:    s_waitcnt lgkmcnt(1)
 ; G_GFX11-NEXT:    ds_min_rtn_f64 v[0:1], v5, v[2:3]
 ; G_GFX11-NEXT:    s_waitcnt lgkmcnt(0)
 ; G_GFX11-NEXT:    scratch_store_b64 off, v[0:1], s0
-; G_GFX11-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; G_GFX11-NEXT:    s_endpgm
   %idx.add = add nuw i32 %idx, 4
   %shl0 = shl i32 %idx.add, 3
   %shl1 = shl i32 %idx.add, 4
   %ptr0 = inttoptr i32 %shl0 to ptr addrspace(3)
   %ptr1 = inttoptr i32 %shl1 to ptr addrspace(3)
   %a1 = call double @llvm.amdgcn.ds.fmin.f64(ptr addrspace(3) %ptr0, double 4.2e+1, i32 0, i32 0, i1 false)
   %a2 = call double @llvm.amdgcn.ds.fmin.f64(ptr addrspace(3) %ptr1, double 4.2e+1, i32 0, i32 0, i1 false)

 ; GFX11-NEXT:    s_delay_alu instid0(SALU_CYCLE_1)
 ; GFX11-NEXT:    v_mov_b32_e32 v4, s2
 ; GFX11-NEXT:    ds_max_rtn_f64 v[2:3], v2, v[0:1] offset:32
 ; GFX11-NEXT:    ds_max_f64 v4, v[0:1] offset:64
 ; GFX11-NEXT:    s_waitcnt lgkmcnt(1)
 ; GFX11-NEXT:    ds_max_rtn_f64 v[0:1], v5, v[2:3]
 ; GFX11-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-NEXT:    scratch_store_b64 off, v[0:1], s0
-; GFX11-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-NEXT:    s_endpgm
 ;
 ; G_SI-LABEL: lds_ds_fmax_f64:
 ; G_SI:       ; %bb.0:
 ; G_SI-NEXT:    s_mov_b32 s8, SCRATCH_RSRC_DWORD0
 ; G_SI-NEXT:    s_mov_b32 s9, SCRATCH_RSRC_DWORD1
 ; G_SI-NEXT:    s_load_dword s4, s[0:1], 0xb
 ; G_SI-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9

 ; G_GFX11-NEXT:    s_delay_alu instid0(SALU_CYCLE_1)
 ; G_GFX11-NEXT:    v_mov_b32_e32 v4, s2
 ; G_GFX11-NEXT:    ds_max_rtn_f64 v[2:3], v2, v[0:1]
 ; G_GFX11-NEXT:    ds_max_f64 v4, v[0:1]
 ; G_GFX11-NEXT:    s_waitcnt lgkmcnt(1)
 ; G_GFX11-NEXT:    ds_max_rtn_f64 v[0:1], v5, v[2:3]
 ; G_GFX11-NEXT:    s_waitcnt lgkmcnt(0)
 ; G_GFX11-NEXT:    scratch_store_b64 off, v[0:1], s0
-; G_GFX11-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; G_GFX11-NEXT:    s_endpgm
   %idx.add = add nuw i32 %idx, 4
   %shl0 = shl i32 %idx.add, 3
   %shl1 = shl i32 %idx.add, 4
   %ptr0 = inttoptr i32 %shl0 to ptr addrspace(3)
   %ptr1 = inttoptr i32 %shl1 to ptr addrspace(3)
   %a1 = call double @llvm.amdgcn.ds.fmax.f64(ptr addrspace(3) %ptr0, double 4.2e+1, i32 0, i32 0, i1 false)
   %a2 = call double @llvm.amdgcn.ds.fmax.f64(ptr addrspace(3) %ptr1, double 4.2e+1, i32 0, i32 0, i1 false)
   %a3 = call double @llvm.amdgcn.ds.fmax.f64(ptr addrspace(3) %ptrf, double %a1, i32 0, i32 0, i1 false)
   store double %a3, ptr addrspace(5) %out
   ret void
 }

llvm/test/CodeGen/AMDGPU/llvm.amdgcn.fdot2.bf16.bf16.ll

 ; SDAG-GFX11-NEXT:    s_load_b128 s[0:3], s[0:1], 0x24
 ; SDAG-GFX11-NEXT:    s_waitcnt lgkmcnt(0)
 ; SDAG-GFX11-NEXT:    scratch_load_b32 v0, off, s2
 ; SDAG-GFX11-NEXT:    scratch_load_u16 v1, off, s3
 ; SDAG-GFX11-NEXT:    scratch_load_b32 v2, off, s1
 ; SDAG-GFX11-NEXT:    s_waitcnt vmcnt(0)
 ; SDAG-GFX11-NEXT:    v_dot2_bf16_bf16_e64_dpp v0, v2, v0, v1 quad_perm:[1,0,0,0] row_mask:0xf bank_mask:0xf bound_ctrl:1
 ; SDAG-GFX11-NEXT:    scratch_store_b16 off, v0, s0
-; SDAG-GFX11-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; SDAG-GFX11-NEXT:    s_endpgm
 ;
 ; GISEL-GFX11-LABEL: test_llvm_amdgcn_fdot2_bf16_bf16_dpp:
 ; GISEL-GFX11:       ; %bb.0: ; %entry
 ; GISEL-GFX11-NEXT:    s_load_b128 s[0:3], s[0:1], 0x24
 ; GISEL-GFX11-NEXT:    s_waitcnt lgkmcnt(0)
 ; GISEL-GFX11-NEXT:    scratch_load_b32 v0, off, s1
 ; GISEL-GFX11-NEXT:    scratch_load_b32 v1, off, s2
 ; GISEL-GFX11-NEXT:    scratch_load_u16 v2, off, s3
 ; GISEL-GFX11-NEXT:    s_waitcnt vmcnt(0)
 ; GISEL-GFX11-NEXT:    v_dot2_bf16_bf16_e64_dpp v0, v0, v1, v2 quad_perm:[1,0,0,0] row_mask:0xf bank_mask:0xf bound_ctrl:1
 ; GISEL-GFX11-NEXT:    scratch_store_b16 off, v0, s0
-; GISEL-GFX11-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GISEL-GFX11-NEXT:    s_endpgm
     ptr addrspace(5) %r,
     ptr addrspace(5) %a,
     ptr addrspace(5) %b,
     ptr addrspace(5) %c) {
 entry:
   %a.val = load <2 x i16>, ptr addrspace(5) %a
   %b.val = load <2 x i16>, ptr addrspace(5) %b

llvm/test/CodeGen/AMDGPU/llvm.amdgcn.fdot2.f16.f16.ll

 ; SDAG-GFX11-NEXT:    s_load_b128 s[0:3], s[0:1], 0x24
 ; SDAG-GFX11-NEXT:    s_waitcnt lgkmcnt(0)
 ; SDAG-GFX11-NEXT:    scratch_load_b32 v0, off, s2
 ; SDAG-GFX11-NEXT:    scratch_load_u16 v1, off, s3
 ; SDAG-GFX11-NEXT:    scratch_load_b32 v2, off, s1
 ; SDAG-GFX11-NEXT:    s_waitcnt vmcnt(0)
 ; SDAG-GFX11-NEXT:    v_dot2_f16_f16_e64_dpp v0, v2, v0, v1 quad_perm:[1,0,0,0] row_mask:0xf bank_mask:0xf bound_ctrl:1
 ; SDAG-GFX11-NEXT:    scratch_store_b16 off, v0, s0
-; SDAG-GFX11-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; SDAG-GFX11-NEXT:    s_endpgm
 ;
 ; GISEL-GFX11-LABEL: test_llvm_amdgcn_fdot2_f16_f16_dpp:
 ; GISEL-GFX11:       ; %bb.0: ; %entry
 ; GISEL-GFX11-NEXT:    s_load_b128 s[0:3], s[0:1], 0x24
 ; GISEL-GFX11-NEXT:    s_waitcnt lgkmcnt(0)
 ; GISEL-GFX11-NEXT:    scratch_load_b32 v0, off, s1
 ; GISEL-GFX11-NEXT:    scratch_load_b32 v1, off, s2
 ; GISEL-GFX11-NEXT:    scratch_load_u16 v2, off, s3
 ; GISEL-GFX11-NEXT:    s_waitcnt vmcnt(0)
 ; GISEL-GFX11-NEXT:    v_dot2_f16_f16_e64_dpp v0, v0, v1, v2 quad_perm:[1,0,0,0] row_mask:0xf bank_mask:0xf bound_ctrl:1
 ; GISEL-GFX11-NEXT:    scratch_store_b16 off, v0, s0
-; GISEL-GFX11-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GISEL-GFX11-NEXT:    s_endpgm
     ptr addrspace(5) %r,
     ptr addrspace(5) %a,
     ptr addrspace(5) %b,
     ptr addrspace(5) %c) {
 entry:
   %a.val = load <2 x half>, ptr addrspace(5) %a
   %b.val = load <2 x half>, ptr addrspace(5) %b

llvm/test/CodeGen/AMDGPU/llvm.amdgcn.intersect_ray.ll

 ; GFX11-NEXT:    v_mov_b32_e32 v0, 0x40c00000
 ; GFX11-NEXT:    v_mov_b32_e32 v1, 0x40e00000
 ; GFX11-NEXT:    v_mov_b32_e32 v2, 0x41000000
 ; GFX11-NEXT:    v_mov_b32_e32 v3, 0x40400000
 ; GFX11-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX11-NEXT:    image_bvh_intersect_ray v[0:3], [v9, v10, v[6:8], v[3:5], v[0:2]], s[4:7]
 ; GFX11-NEXT:    s_waitcnt vmcnt(0)
 ; GFX11-NEXT:    flat_store_b128 v[0:1], v[0:3]
-; GFX11-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-NEXT:    s_endpgm
 main_body:
   %lid = tail call i32 @llvm.amdgcn.workitem.id.x()
   %gep_node_ptr = getelementptr inbounds i32, ptr %p_node_ptr, i32 %lid
   %node_ptr = load i32, ptr %gep_node_ptr, align 4
   %gep_ray = getelementptr inbounds float, ptr %p_ray, i32 %lid
   %ray_extent = load float, ptr %gep_ray, align 4
   %ray_origin0 = insertelement <3 x float> undef, float 0.0, i32 0

 ; GFX11-NEXT:    flat_load_b32 v7, v[2:3]
 ; GFX11-NEXT:    v_mov_b32_e32 v1, 0x47004400
 ; GFX11-NEXT:    v_dual_mov_b32 v0, 0x46004200 :: v_dual_mov_b32 v3, 0
 ; GFX11-NEXT:    v_mov_b32_e32 v2, 0x48004500
 ; GFX11-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX11-NEXT:    image_bvh_intersect_ray v[0:3], [v6, v7, v[3:5], v[0:2]], s[4:7] a16
 ; GFX11-NEXT:    s_waitcnt vmcnt(0)
 ; GFX11-NEXT:    flat_store_b128 v[0:1], v[0:3]
-; GFX11-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-NEXT:    s_endpgm
 main_body:
   %lid = tail call i32 @llvm.amdgcn.workitem.id.x()
   %gep_node_ptr = getelementptr inbounds i32, ptr %p_node_ptr, i32 %lid
   %node_ptr = load i32, ptr %gep_node_ptr, align 4
   %gep_ray = getelementptr inbounds float, ptr %p_ray, i32 %lid
   %ray_extent = load float, ptr %gep_ray, align 4
   %ray_origin0 = insertelement <3 x float> undef, float 0.0, i32 0

 ; GFX11-NEXT:    v_add_co_ci_u32_e64 v1, null, s5, 0, s4
 ; GFX11-NEXT:    flat_load_b32 v11, v[0:1]
 ; GFX11-NEXT:    v_mov_b32_e32 v0, 0x40c00000
 ; GFX11-NEXT:    v_mov_b32_e32 v1, 0x40e00000
 ; GFX11-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX11-NEXT:    image_bvh64_intersect_ray v[0:3], [v[9:10], v11, v[6:8], v[3:5], v[0:2]], s[0:3]
 ; GFX11-NEXT:    s_waitcnt vmcnt(0)
 ; GFX11-NEXT:    flat_store_b128 v[0:1], v[0:3]
-; GFX11-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-NEXT:    s_endpgm
 main_body:
   %lid = tail call i32 @llvm.amdgcn.workitem.id.x()
   %gep_ray = getelementptr inbounds float, ptr %p_ray, i32 %lid
   %ray_extent = load float, ptr %gep_ray, align 4
   %ray_origin0 = insertelement <3 x float> undef, float 0.0, i32 0
   %ray_origin1 = insertelement <3 x float> %ray_origin0, float 1.0, i32 1
   %ray_origin = insertelement <3 x float> %ray_origin1, float 2.0, i32 2

 ; GFX11-NEXT:    v_add_co_ci_u32_e64 v1, null, s5, 0, s4
 ; GFX11-NEXT:    flat_load_b32 v8, v[0:1]
 ; GFX11-NEXT:    v_mov_b32_e32 v0, 0x46004200
 ; GFX11-NEXT:    v_mov_b32_e32 v1, 0x47004400
 ; GFX11-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX11-NEXT:    image_bvh64_intersect_ray v[0:3], [v[6:7], v8, v[3:5], v[0:2]], s[0:3] a16
 ; GFX11-NEXT:    s_waitcnt vmcnt(0)
 ; GFX11-NEXT:    flat_store_b128 v[0:1], v[0:3]
-; GFX11-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-NEXT:    s_endpgm
 main_body:
   %lid = tail call i32 @llvm.amdgcn.workitem.id.x()
   %gep_ray = getelementptr inbounds float, ptr %p_ray, i32 %lid
   %ray_extent = load float, ptr %gep_ray, align 4
   %ray_origin0 = insertelement <3 x float> undef, float 0.0, i32 0
   %ray_origin1 = insertelement <3 x float> %ray_origin0, float 1.0, i32 1
   %ray_origin = insertelement <3 x float> %ray_origin1, float 2.0, i32 2

llvm/test/CodeGen/AMDGPU/memory-legalizer-flat-agent.ll

 ; GFX11-WGP:       ; %bb.0: ; %entry
 ; GFX11-WGP-NEXT:    s_load_b128 s[0:3], s[0:1], 0x0
 ; GFX11-WGP-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-WGP-NEXT:    v_dual_mov_b32 v0, s0 :: v_dual_mov_b32 v1, s1
 ; GFX11-WGP-NEXT:    flat_load_b32 v2, v[0:1]
 ; GFX11-WGP-NEXT:    v_dual_mov_b32 v0, s2 :: v_dual_mov_b32 v1, s3
 ; GFX11-WGP-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX11-WGP-NEXT:    flat_store_b32 v[0:1], v2
-; GFX11-WGP-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-WGP-NEXT:    s_endpgm
 ;
 ; GFX11-CU-LABEL: flat_agent_unordered_load:
 ; GFX11-CU:       ; %bb.0: ; %entry
 ; GFX11-CU-NEXT:    s_load_b128 s[0:3], s[0:1], 0x0
 ; GFX11-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v0, s0 :: v_dual_mov_b32 v1, s1
 ; GFX11-CU-NEXT:    flat_load_b32 v2, v[0:1]
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v0, s2 :: v_dual_mov_b32 v1, s3
 ; GFX11-CU-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX11-CU-NEXT:    flat_store_b32 v[0:1], v2
-; GFX11-CU-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-CU-NEXT:    s_endpgm
     ptr %in, ptr %out) {
 entry:
   %val = load atomic i32, ptr %in syncscope("agent") unordered, align 4
   store i32 %val, ptr %out
   ret void
 }
 

 ; GFX11-WGP:       ; %bb.0: ; %entry
 ; GFX11-WGP-NEXT:    s_load_b128 s[0:3], s[0:1], 0x0
 ; GFX11-WGP-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-WGP-NEXT:    v_dual_mov_b32 v0, s0 :: v_dual_mov_b32 v1, s1
 ; GFX11-WGP-NEXT:    flat_load_b32 v2, v[0:1] glc
 ; GFX11-WGP-NEXT:    v_dual_mov_b32 v0, s2 :: v_dual_mov_b32 v1, s3
 ; GFX11-WGP-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX11-WGP-NEXT:    flat_store_b32 v[0:1], v2
-; GFX11-WGP-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-WGP-NEXT:    s_endpgm
 ;
 ; GFX11-CU-LABEL: flat_agent_monotonic_load:
 ; GFX11-CU:       ; %bb.0: ; %entry
 ; GFX11-CU-NEXT:    s_load_b128 s[0:3], s[0:1], 0x0
 ; GFX11-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v0, s0 :: v_dual_mov_b32 v1, s1
 ; GFX11-CU-NEXT:    flat_load_b32 v2, v[0:1] glc
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v0, s2 :: v_dual_mov_b32 v1, s3
 ; GFX11-CU-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX11-CU-NEXT:    flat_store_b32 v[0:1], v2
-; GFX11-CU-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-CU-NEXT:    s_endpgm
     ptr %in, ptr %out) {
 entry:
   %val = load atomic i32, ptr %in syncscope("agent") monotonic, align 4
   store i32 %val, ptr %out
   ret void
 }
 

 ; GFX11-WGP-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-WGP-NEXT:    v_dual_mov_b32 v0, s0 :: v_dual_mov_b32 v1, s1
 ; GFX11-WGP-NEXT:    flat_load_b32 v2, v[0:1] glc
 ; GFX11-WGP-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX11-WGP-NEXT:    buffer_gl0_inv
 ; GFX11-WGP-NEXT:    buffer_gl1_inv
 ; GFX11-WGP-NEXT:    v_dual_mov_b32 v0, s2 :: v_dual_mov_b32 v1, s3
 ; GFX11-WGP-NEXT:    flat_store_b32 v[0:1], v2
-; GFX11-WGP-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-WGP-NEXT:    s_endpgm
 ;
 ; GFX11-CU-LABEL: flat_agent_acquire_load:
 ; GFX11-CU:       ; %bb.0: ; %entry
 ; GFX11-CU-NEXT:    s_load_b128 s[0:3], s[0:1], 0x0
 ; GFX11-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v0, s0 :: v_dual_mov_b32 v1, s1
 ; GFX11-CU-NEXT:    flat_load_b32 v2, v[0:1] glc
 ; GFX11-CU-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX11-CU-NEXT:    buffer_gl0_inv
 ; GFX11-CU-NEXT:    buffer_gl1_inv
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v0, s2 :: v_dual_mov_b32 v1, s3
 ; GFX11-CU-NEXT:    flat_store_b32 v[0:1], v2
-; GFX11-CU-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-CU-NEXT:    s_endpgm
     ptr %in, ptr %out) {
 entry:
   %val = load atomic i32, ptr %in syncscope("agent") acquire, align 4
   store i32 %val, ptr %out
   ret void
 }
 

 ; GFX11-WGP-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX11-WGP-NEXT:    s_waitcnt_vscnt null, 0x0
 ; GFX11-WGP-NEXT:    flat_load_b32 v2, v[0:1] glc
 ; GFX11-WGP-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX11-WGP-NEXT:    buffer_gl0_inv
 ; GFX11-WGP-NEXT:    buffer_gl1_inv
 ; GFX11-WGP-NEXT:    v_dual_mov_b32 v0, s2 :: v_dual_mov_b32 v1, s3
 ; GFX11-WGP-NEXT:    flat_store_b32 v[0:1], v2
-; GFX11-WGP-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-WGP-NEXT:    s_endpgm
 ;
 ; GFX11-CU-LABEL: flat_agent_seq_cst_load:
 ; GFX11-CU:       ; %bb.0: ; %entry
 ; GFX11-CU-NEXT:    s_load_b128 s[0:3], s[0:1], 0x0
 ; GFX11-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v0, s0 :: v_dual_mov_b32 v1, s1
 ; GFX11-CU-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX11-CU-NEXT:    s_waitcnt_vscnt null, 0x0
 ; GFX11-CU-NEXT:    flat_load_b32 v2, v[0:1] glc
 ; GFX11-CU-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX11-CU-NEXT:    buffer_gl0_inv
 ; GFX11-CU-NEXT:    buffer_gl1_inv
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v0, s2 :: v_dual_mov_b32 v1, s3
 ; GFX11-CU-NEXT:    flat_store_b32 v[0:1], v2
-; GFX11-CU-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-CU-NEXT:    s_endpgm
     ptr %in, ptr %out) {
 entry:
   %val = load atomic i32, ptr %in syncscope("agent") seq_cst, align 4
   store i32 %val, ptr %out
   ret void
 }
 

 ; GFX11-WGP:       ; %bb.0: ; %entry
 ; GFX11-WGP-NEXT:    s_clause 0x1
 ; GFX11-WGP-NEXT:    s_load_b64 s[2:3], s[0:1], 0x8
 ; GFX11-WGP-NEXT:    s_load_b32 s0, s[0:1], 0x0
 ; GFX11-WGP-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-WGP-NEXT:    v_dual_mov_b32 v0, s2 :: v_dual_mov_b32 v1, s3
 ; GFX11-WGP-NEXT:    v_mov_b32_e32 v2, s0
 ; GFX11-WGP-NEXT:    flat_store_b32 v[0:1], v2
-; GFX11-WGP-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-WGP-NEXT:    s_endpgm
 ;
 ; GFX11-CU-LABEL: flat_agent_unordered_store:
 ; GFX11-CU:       ; %bb.0: ; %entry
 ; GFX11-CU-NEXT:    s_clause 0x1
 ; GFX11-CU-NEXT:    s_load_b64 s[2:3], s[0:1], 0x8
 ; GFX11-CU-NEXT:    s_load_b32 s0, s[0:1], 0x0
 ; GFX11-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v0, s2 :: v_dual_mov_b32 v1, s3
 ; GFX11-CU-NEXT:    v_mov_b32_e32 v2, s0
 ; GFX11-CU-NEXT:    flat_store_b32 v[0:1], v2
-; GFX11-CU-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-CU-NEXT:    s_endpgm
     i32 %in, ptr %out) {
 entry:
   store atomic i32 %in, ptr %out syncscope("agent") unordered, align 4
   ret void
 }
 
 define amdgpu_kernel void @flat_agent_monotonic_store(

 ; GFX11-WGP:       ; %bb.0: ; %entry
 ; GFX11-WGP-NEXT:    s_clause 0x1
 ; GFX11-WGP-NEXT:    s_load_b64 s[2:3], s[0:1], 0x8
 ; GFX11-WGP-NEXT:    s_load_b32 s0, s[0:1], 0x0
 ; GFX11-WGP-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-WGP-NEXT:    v_dual_mov_b32 v0, s2 :: v_dual_mov_b32 v1, s3
 ; GFX11-WGP-NEXT:    v_mov_b32_e32 v2, s0
 ; GFX11-WGP-NEXT:    flat_store_b32 v[0:1], v2
-; GFX11-WGP-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-WGP-NEXT:    s_endpgm
 ;
 ; GFX11-CU-LABEL: flat_agent_monotonic_store:
 ; GFX11-CU:       ; %bb.0: ; %entry
 ; GFX11-CU-NEXT:    s_clause 0x1
 ; GFX11-CU-NEXT:    s_load_b64 s[2:3], s[0:1], 0x8
 ; GFX11-CU-NEXT:    s_load_b32 s0, s[0:1], 0x0
 ; GFX11-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v0, s2 :: v_dual_mov_b32 v1, s3
 ; GFX11-CU-NEXT:    v_mov_b32_e32 v2, s0
 ; GFX11-CU-NEXT:    flat_store_b32 v[0:1], v2
-; GFX11-CU-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-CU-NEXT:    s_endpgm
     i32 %in, ptr %out) {
 entry:
   store atomic i32 %in, ptr %out syncscope("agent") monotonic, align 4
   ret void
 }
 
 define amdgpu_kernel void @flat_agent_release_store(

 ; GFX11-WGP-NEXT:    s_load_b64 s[2:3], s[0:1], 0x8
 ; GFX11-WGP-NEXT:    s_load_b32 s0, s[0:1], 0x0
 ; GFX11-WGP-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-WGP-NEXT:    v_dual_mov_b32 v0, s2 :: v_dual_mov_b32 v1, s3
 ; GFX11-WGP-NEXT:    v_mov_b32_e32 v2, s0
 ; GFX11-WGP-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX11-WGP-NEXT:    s_waitcnt_vscnt null, 0x0
 ; GFX11-WGP-NEXT:    flat_store_b32 v[0:1], v2
-; GFX11-WGP-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-WGP-NEXT:    s_endpgm
 ;
 ; GFX11-CU-LABEL: flat_agent_release_store:
 ; GFX11-CU:       ; %bb.0: ; %entry
 ; GFX11-CU-NEXT:    s_clause 0x1
 ; GFX11-CU-NEXT:    s_load_b64 s[2:3], s[0:1], 0x8
 ; GFX11-CU-NEXT:    s_load_b32 s0, s[0:1], 0x0
 ; GFX11-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v0, s2 :: v_dual_mov_b32 v1, s3
 ; GFX11-CU-NEXT:    v_mov_b32_e32 v2, s0
 ; GFX11-CU-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX11-CU-NEXT:    s_waitcnt_vscnt null, 0x0
 ; GFX11-CU-NEXT:    flat_store_b32 v[0:1], v2
-; GFX11-CU-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-CU-NEXT:    s_endpgm
     i32 %in, ptr %out) {
 entry:
   store atomic i32 %in, ptr %out syncscope("agent") release, align 4
   ret void
 }
 
 define amdgpu_kernel void @flat_agent_seq_cst_store(

 ; GFX11-WGP-NEXT:    s_load_b64 s[2:3], s[0:1], 0x8
 ; GFX11-WGP-NEXT:    s_load_b32 s0, s[0:1], 0x0
 ; GFX11-WGP-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-WGP-NEXT:    v_dual_mov_b32 v0, s2 :: v_dual_mov_b32 v1, s3
 ; GFX11-WGP-NEXT:    v_mov_b32_e32 v2, s0
 ; GFX11-WGP-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX11-WGP-NEXT:    s_waitcnt_vscnt null, 0x0
 ; GFX11-WGP-NEXT:    flat_store_b32 v[0:1], v2
-; GFX11-WGP-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-WGP-NEXT:    s_endpgm
 ;
 ; GFX11-CU-LABEL: flat_agent_seq_cst_store:
 ; GFX11-CU:       ; %bb.0: ; %entry
 ; GFX11-CU-NEXT:    s_clause 0x1
 ; GFX11-CU-NEXT:    s_load_b64 s[2:3], s[0:1], 0x8
 ; GFX11-CU-NEXT:    s_load_b32 s0, s[0:1], 0x0
 ; GFX11-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v0, s2 :: v_dual_mov_b32 v1, s3
 ; GFX11-CU-NEXT:    v_mov_b32_e32 v2, s0
 ; GFX11-CU-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX11-CU-NEXT:    s_waitcnt_vscnt null, 0x0
 ; GFX11-CU-NEXT:    flat_store_b32 v[0:1], v2
-; GFX11-CU-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-CU-NEXT:    s_endpgm
     i32 %in, ptr %out) {
 entry:
   store atomic i32 %in, ptr %out syncscope("agent") seq_cst, align 4
   ret void
 }
 
 define amdgpu_kernel void @flat_agent_monotonic_atomicrmw(

 ; GFX11-WGP:       ; %bb.0: ; %entry
 ; GFX11-WGP-NEXT:    s_clause 0x1
 ; GFX11-WGP-NEXT:    s_load_b64 s[2:3], s[0:1], 0x0
 ; GFX11-WGP-NEXT:    s_load_b32 s0, s[0:1], 0x8
 ; GFX11-WGP-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-WGP-NEXT:    v_dual_mov_b32 v0, s2 :: v_dual_mov_b32 v1, s3
 ; GFX11-WGP-NEXT:    v_mov_b32_e32 v2, s0
 ; GFX11-WGP-NEXT:    flat_atomic_swap_b32 v[0:1], v2
-; GFX11-WGP-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-WGP-NEXT:    s_endpgm
 ;
 ; GFX11-CU-LABEL: flat_agent_monotonic_atomicrmw:
 ; GFX11-CU:       ; %bb.0: ; %entry
 ; GFX11-CU-NEXT:    s_clause 0x1
 ; GFX11-CU-NEXT:    s_load_b64 s[2:3], s[0:1], 0x0
 ; GFX11-CU-NEXT:    s_load_b32 s0, s[0:1], 0x8
 ; GFX11-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v0, s2 :: v_dual_mov_b32 v1, s3
 ; GFX11-CU-NEXT:    v_mov_b32_e32 v2, s0
 ; GFX11-CU-NEXT:    flat_atomic_swap_b32 v[0:1], v2
-; GFX11-CU-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-CU-NEXT:    s_endpgm
     ptr %out, i32 %in) {
 entry:
   %val = atomicrmw volatile xchg ptr %out, i32 %in syncscope("agent") monotonic
   ret void
 }
 
 define amdgpu_kernel void @flat_agent_acquire_atomicrmw(

 ; GFX11-WGP-NEXT:    s_load_b64 s[2:3], s[0:1], 0x0
 ; GFX11-WGP-NEXT:    s_load_b32 s0, s[0:1], 0x8
 ; GFX11-WGP-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-WGP-NEXT:    v_dual_mov_b32 v0, s2 :: v_dual_mov_b32 v1, s3
 ; GFX11-WGP-NEXT:    v_mov_b32_e32 v2, s0
 ; GFX11-WGP-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX11-WGP-NEXT:    s_waitcnt_vscnt null, 0x0
 ; GFX11-WGP-NEXT:    flat_atomic_swap_b32 v[0:1], v2
-; GFX11-WGP-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-WGP-NEXT:    s_endpgm
 ;
 ; GFX11-CU-LABEL: flat_agent_release_atomicrmw:
 ; GFX11-CU:       ; %bb.0: ; %entry
 ; GFX11-CU-NEXT:    s_clause 0x1
 ; GFX11-CU-NEXT:    s_load_b64 s[2:3], s[0:1], 0x0
 ; GFX11-CU-NEXT:    s_load_b32 s0, s[0:1], 0x8
 ; GFX11-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v0, s2 :: v_dual_mov_b32 v1, s3
 ; GFX11-CU-NEXT:    v_mov_b32_e32 v2, s0
 ; GFX11-CU-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX11-CU-NEXT:    s_waitcnt_vscnt null, 0x0
 ; GFX11-CU-NEXT:    flat_atomic_swap_b32 v[0:1], v2
-; GFX11-CU-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-CU-NEXT:    s_endpgm
     ptr %out, i32 %in) {
 entry:
   %val = atomicrmw volatile xchg ptr %out, i32 %in syncscope("agent") release
   ret void
 }
 
 define amdgpu_kernel void @flat_agent_acq_rel_atomicrmw(

 ; GFX11-WGP-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-WGP-NEXT:    v_dual_mov_b32 v0, s2 :: v_dual_mov_b32 v1, s3
 ; GFX11-WGP-NEXT:    v_mov_b32_e32 v2, s0
 ; GFX11-WGP-NEXT:    flat_atomic_swap_b32 v2, v[0:1], v2 glc
 ; GFX11-WGP-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX11-WGP-NEXT:    buffer_gl0_inv
 ; GFX11-WGP-NEXT:    buffer_gl1_inv
 ; GFX11-WGP-NEXT:    flat_store_b32 v[0:1], v2
-; GFX11-WGP-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-WGP-NEXT:    s_endpgm
 ;
 ; GFX11-CU-LABEL: flat_agent_acquire_ret_atomicrmw:
 ; GFX11-CU:       ; %bb.0: ; %entry
 ; GFX11-CU-NEXT:    s_clause 0x1
 ; GFX11-CU-NEXT:    s_load_b64 s[2:3], s[0:1], 0x0
 ; GFX11-CU-NEXT:    s_load_b32 s0, s[0:1], 0x8
 ; GFX11-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v0, s2 :: v_dual_mov_b32 v1, s3
 ; GFX11-CU-NEXT:    v_mov_b32_e32 v2, s0
 ; GFX11-CU-NEXT:    flat_atomic_swap_b32 v2, v[0:1], v2 glc
 ; GFX11-CU-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX11-CU-NEXT:    buffer_gl0_inv
 ; GFX11-CU-NEXT:    buffer_gl1_inv
 ; GFX11-CU-NEXT:    flat_store_b32 v[0:1], v2
-; GFX11-CU-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-CU-NEXT:    s_endpgm
     ptr %out, i32 %in) {
 entry:
   %val = atomicrmw volatile xchg ptr %out, i32 %in syncscope("agent") acquire
   store i32 %val, ptr %out, align 4
   ret void
 }
 

 ; GFX11-WGP-NEXT:    v_mov_b32_e32 v2, s0
 ; GFX11-WGP-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX11-WGP-NEXT:    s_waitcnt_vscnt null, 0x0
 ; GFX11-WGP-NEXT:    flat_atomic_swap_b32 v2, v[0:1], v2 glc
 ; GFX11-WGP-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX11-WGP-NEXT:    buffer_gl0_inv
 ; GFX11-WGP-NEXT:    buffer_gl1_inv
 ; GFX11-WGP-NEXT:    flat_store_b32 v[0:1], v2
-; GFX11-WGP-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-WGP-NEXT:    s_endpgm
 ;
 ; GFX11-CU-LABEL: flat_agent_acq_rel_ret_atomicrmw:
 ; GFX11-CU:       ; %bb.0: ; %entry
 ; GFX11-CU-NEXT:    s_clause 0x1
 ; GFX11-CU-NEXT:    s_load_b64 s[2:3], s[0:1], 0x0
 ; GFX11-CU-NEXT:    s_load_b32 s0, s[0:1], 0x8
 ; GFX11-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v0, s2 :: v_dual_mov_b32 v1, s3
 ; GFX11-CU-NEXT:    v_mov_b32_e32 v2, s0
 ; GFX11-CU-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX11-CU-NEXT:    s_waitcnt_vscnt null, 0x0
 ; GFX11-CU-NEXT:    flat_atomic_swap_b32 v2, v[0:1], v2 glc
 ; GFX11-CU-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX11-CU-NEXT:    buffer_gl0_inv
 ; GFX11-CU-NEXT:    buffer_gl1_inv
 ; GFX11-CU-NEXT:    flat_store_b32 v[0:1], v2
-; GFX11-CU-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-CU-NEXT:    s_endpgm
     ptr %out, i32 %in) {
 entry:
   %val = atomicrmw volatile xchg ptr %out, i32 %in syncscope("agent") acq_rel
   store i32 %val, ptr %out, align 4
   ret void
 }
 

 ; GFX11-WGP-NEXT:    v_mov_b32_e32 v2, s0
 ; GFX11-WGP-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX11-WGP-NEXT:    s_waitcnt_vscnt null, 0x0
 ; GFX11-WGP-NEXT:    flat_atomic_swap_b32 v2, v[0:1], v2 glc
 ; GFX11-WGP-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX11-WGP-NEXT:    buffer_gl0_inv
 ; GFX11-WGP-NEXT:    buffer_gl1_inv
 ; GFX11-WGP-NEXT:    flat_store_b32 v[0:1], v2
-; GFX11-WGP-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-WGP-NEXT:    s_endpgm
 ;
 ; GFX11-CU-LABEL: flat_agent_seq_cst_ret_atomicrmw:
 ; GFX11-CU:       ; %bb.0: ; %entry
 ; GFX11-CU-NEXT:    s_clause 0x1
 ; GFX11-CU-NEXT:    s_load_b64 s[2:3], s[0:1], 0x0
 ; GFX11-CU-NEXT:    s_load_b32 s0, s[0:1], 0x8
 ; GFX11-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v0, s2 :: v_dual_mov_b32 v1, s3
 ; GFX11-CU-NEXT:    v_mov_b32_e32 v2, s0
 ; GFX11-CU-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX11-CU-NEXT:    s_waitcnt_vscnt null, 0x0
 ; GFX11-CU-NEXT:    flat_atomic_swap_b32 v2, v[0:1], v2 glc
 ; GFX11-CU-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX11-CU-NEXT:    buffer_gl0_inv
 ; GFX11-CU-NEXT:    buffer_gl1_inv
 ; GFX11-CU-NEXT:    flat_store_b32 v[0:1], v2
-; GFX11-CU-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-CU-NEXT:    s_endpgm
     ptr %out, i32 %in) {
 entry:
   %val = atomicrmw volatile xchg ptr %out, i32 %in syncscope("agent") seq_cst
   store i32 %val, ptr %out, align 4
   ret void
 }
 

 ;
 ; GFX11-WGP-LABEL: flat_agent_monotonic_monotonic_cmpxchg:
 ; GFX11-WGP:       ; %bb.0: ; %entry
 ; GFX11-WGP-NEXT:    s_load_b128 s[0:3], s[0:1], 0x0
 ; GFX11-WGP-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-WGP-NEXT:    v_dual_mov_b32 v0, s0 :: v_dual_mov_b32 v1, s1
 ; GFX11-WGP-NEXT:    v_dual_mov_b32 v2, s2 :: v_dual_mov_b32 v3, s3
 ; GFX11-WGP-NEXT:    flat_atomic_cmpswap_b32 v[0:1], v[2:3] offset:16
-; GFX11-WGP-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-WGP-NEXT:    s_endpgm
 ;
 ; GFX11-CU-LABEL: flat_agent_monotonic_monotonic_cmpxchg:
 ; GFX11-CU:       ; %bb.0: ; %entry
 ; GFX11-CU-NEXT:    s_load_b128 s[0:3], s[0:1], 0x0
 ; GFX11-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v0, s0 :: v_dual_mov_b32 v1, s1
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v2, s2 :: v_dual_mov_b32 v3, s3
 ; GFX11-CU-NEXT:    flat_atomic_cmpswap_b32 v[0:1], v[2:3] offset:16
-; GFX11-CU-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-CU-NEXT:    s_endpgm
     ptr %out, i32 %in, i32 %old) {
 entry:
   %gep = getelementptr i32, ptr %out, i32 4
   %val = cmpxchg volatile ptr %gep, i32 %old, i32 %in syncscope("agent") monotonic monotonic
   ret void
 }
 

 ; GFX11-WGP:       ; %bb.0: ; %entry
 ; GFX11-WGP-NEXT:    s_load_b128 s[0:3], s[0:1], 0x0
 ; GFX11-WGP-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-WGP-NEXT:    v_dual_mov_b32 v0, s0 :: v_dual_mov_b32 v1, s1
 ; GFX11-WGP-NEXT:    v_dual_mov_b32 v2, s2 :: v_dual_mov_b32 v3, s3
 ; GFX11-WGP-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX11-WGP-NEXT:    s_waitcnt_vscnt null, 0x0
 ; GFX11-WGP-NEXT:    flat_atomic_cmpswap_b32 v[0:1], v[2:3] offset:16
-; GFX11-WGP-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-WGP-NEXT:    s_endpgm
 ;
 ; GFX11-CU-LABEL: flat_agent_release_monotonic_cmpxchg:
 ; GFX11-CU:       ; %bb.0: ; %entry
 ; GFX11-CU-NEXT:    s_load_b128 s[0:3], s[0:1], 0x0
 ; GFX11-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v0, s0 :: v_dual_mov_b32 v1, s1
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v2, s2 :: v_dual_mov_b32 v3, s3
 ; GFX11-CU-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX11-CU-NEXT:    s_waitcnt_vscnt null, 0x0
 ; GFX11-CU-NEXT:    flat_atomic_cmpswap_b32 v[0:1], v[2:3] offset:16
-; GFX11-CU-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-CU-NEXT:    s_endpgm
     ptr %out, i32 %in, i32 %old) {
 entry:
   %gep = getelementptr i32, ptr %out, i32 4
   %val = cmpxchg volatile ptr %gep, i32 %old, i32 %in syncscope("agent") release monotonic
   ret void
 }
 

 ; GFX11-WGP:       ; %bb.0: ; %entry
 ; GFX11-WGP-NEXT:    s_load_b128 s[0:3], s[0:1], 0x0
 ; GFX11-WGP-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-WGP-NEXT:    v_dual_mov_b32 v0, s0 :: v_dual_mov_b32 v1, s1
 ; GFX11-WGP-NEXT:    v_dual_mov_b32 v2, s2 :: v_dual_mov_b32 v3, s3
 ; GFX11-WGP-NEXT:    flat_atomic_cmpswap_b32 v2, v[0:1], v[2:3] offset:16 glc
 ; GFX11-WGP-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX11-WGP-NEXT:    flat_store_b32 v[0:1], v2
-; GFX11-WGP-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-WGP-NEXT:    s_endpgm
 ;
 ; GFX11-CU-LABEL: flat_agent_monotonic_monotonic_ret_cmpxchg:
 ; GFX11-CU:       ; %bb.0: ; %entry
 ; GFX11-CU-NEXT:    s_load_b128 s[0:3], s[0:1], 0x0
 ; GFX11-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v0, s0 :: v_dual_mov_b32 v1, s1
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v2, s2 :: v_dual_mov_b32 v3, s3
 ; GFX11-CU-NEXT:    flat_atomic_cmpswap_b32 v2, v[0:1], v[2:3] offset:16 glc
 ; GFX11-CU-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX11-CU-NEXT:    flat_store_b32 v[0:1], v2
-; GFX11-CU-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-CU-NEXT:    s_endpgm
     ptr %out, i32 %in, i32 %old) {
 entry:
   %gep = getelementptr i32, ptr %out, i32 4
   %val = cmpxchg volatile ptr %gep, i32 %old, i32 %in syncscope("agent") monotonic monotonic
   %val0 = extractvalue { i32, i1 } %val, 0
   store i32 %val0, ptr %out, align 4
   ret void

 ; GFX11-WGP-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-WGP-NEXT:    v_dual_mov_b32 v0, s0 :: v_dual_mov_b32 v1, s1
 ; GFX11-WGP-NEXT:    v_dual_mov_b32 v2, s2 :: v_dual_mov_b32 v3, s3
 ; GFX11-WGP-NEXT:    flat_atomic_cmpswap_b32 v2, v[0:1], v[2:3] offset:16 glc
 ; GFX11-WGP-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX11-WGP-NEXT:    buffer_gl0_inv
 ; GFX11-WGP-NEXT:    buffer_gl1_inv
 ; GFX11-WGP-NEXT:    flat_store_b32 v[0:1], v2
-; GFX11-WGP-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-WGP-NEXT:    s_endpgm
 ;
 ; GFX11-CU-LABEL: flat_agent_acquire_monotonic_ret_cmpxchg:
 ; GFX11-CU:       ; %bb.0: ; %entry
 ; GFX11-CU-NEXT:    s_load_b128 s[0:3], s[0:1], 0x0
 ; GFX11-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v0, s0 :: v_dual_mov_b32 v1, s1
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v2, s2 :: v_dual_mov_b32 v3, s3
 ; GFX11-CU-NEXT:    flat_atomic_cmpswap_b32 v2, v[0:1], v[2:3] offset:16 glc
 ; GFX11-CU-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX11-CU-NEXT:    buffer_gl0_inv
 ; GFX11-CU-NEXT:    buffer_gl1_inv
 ; GFX11-CU-NEXT:    flat_store_b32 v[0:1], v2
-; GFX11-CU-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-CU-NEXT:    s_endpgm
     ptr %out, i32 %in, i32 %old) {
 entry:
   %gep = getelementptr i32, ptr %out, i32 4
   %val = cmpxchg volatile ptr %gep, i32 %old, i32 %in syncscope("agent") acquire monotonic
   %val0 = extractvalue { i32, i1 } %val, 0
   store i32 %val0, ptr %out, align 4
   ret void

 ; GFX11-WGP-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-WGP-NEXT:    v_dual_mov_b32 v0, s0 :: v_dual_mov_b32 v1, s1
 ; GFX11-WGP-NEXT:    v_dual_mov_b32 v2, s2 :: v_dual_mov_b32 v3, s3
 ; GFX11-WGP-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX11-WGP-NEXT:    s_waitcnt_vscnt null, 0x0
 ; GFX11-WGP-NEXT:    flat_atomic_cmpswap_b32 v2, v[0:1], v[2:3] offset:16 glc
 ; GFX11-WGP-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX11-WGP-NEXT:    flat_store_b32 v[0:1], v2
-; GFX11-WGP-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-WGP-NEXT:    s_endpgm
 ;
 ; GFX11-CU-LABEL: flat_agent_release_monotonic_ret_cmpxchg:
 ; GFX11-CU:       ; %bb.0: ; %entry
 ; GFX11-CU-NEXT:    s_load_b128 s[0:3], s[0:1], 0x0
 ; GFX11-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v0, s0 :: v_dual_mov_b32 v1, s1
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v2, s2 :: v_dual_mov_b32 v3, s3
 ; GFX11-CU-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX11-CU-NEXT:    s_waitcnt_vscnt null, 0x0
 ; GFX11-CU-NEXT:    flat_atomic_cmpswap_b32 v2, v[0:1], v[2:3] offset:16 glc
 ; GFX11-CU-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX11-CU-NEXT:    flat_store_b32 v[0:1], v2
-; GFX11-CU-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-CU-NEXT:    s_endpgm
     ptr %out, i32 %in, i32 %old) {
 entry:
   %gep = getelementptr i32, ptr %out, i32 4
   %val = cmpxchg volatile ptr %gep, i32 %old, i32 %in syncscope("agent") release monotonic
   %val0 = extractvalue { i32, i1 } %val, 0
   store i32 %val0, ptr %out, align 4
   ret void

 ; GFX11-WGP-NEXT:    v_dual_mov_b32 v2, s2 :: v_dual_mov_b32 v3, s3
 ; GFX11-WGP-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX11-WGP-NEXT:    s_waitcnt_vscnt null, 0x0
 ; GFX11-WGP-NEXT:    flat_atomic_cmpswap_b32 v2, v[0:1], v[2:3] offset:16 glc
 ; GFX11-WGP-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX11-WGP-NEXT:    buffer_gl0_inv
 ; GFX11-WGP-NEXT:    buffer_gl1_inv
 ; GFX11-WGP-NEXT:    flat_store_b32 v[0:1], v2
-; GFX11-WGP-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-WGP-NEXT:    s_endpgm
 ;
 ; GFX11-CU-LABEL: flat_agent_acq_rel_monotonic_ret_cmpxchg:
 ; GFX11-CU:       ; %bb.0: ; %entry
 ; GFX11-CU-NEXT:    s_load_b128 s[0:3], s[0:1], 0x0
 ; GFX11-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v0, s0 :: v_dual_mov_b32 v1, s1
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v2, s2 :: v_dual_mov_b32 v3, s3
 ; GFX11-CU-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX11-CU-NEXT:    s_waitcnt_vscnt null, 0x0
 ; GFX11-CU-NEXT:    flat_atomic_cmpswap_b32 v2, v[0:1], v[2:3] offset:16 glc
 ; GFX11-CU-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX11-CU-NEXT:    buffer_gl0_inv
 ; GFX11-CU-NEXT:    buffer_gl1_inv
 ; GFX11-CU-NEXT:    flat_store_b32 v[0:1], v2
-; GFX11-CU-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-CU-NEXT:    s_endpgm
     ptr %out, i32 %in, i32 %old) {
 entry:
   %gep = getelementptr i32, ptr %out, i32 4
   %val = cmpxchg volatile ptr %gep, i32 %old, i32 %in syncscope("agent") acq_rel monotonic
   %val0 = extractvalue { i32, i1 } %val, 0
   store i32 %val0, ptr %out, align 4
   ret void

 ; GFX11-WGP-NEXT:    v_dual_mov_b32 v2, s2 :: v_dual_mov_b32 v3, s3
 ; GFX11-WGP-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX11-WGP-NEXT:    s_waitcnt_vscnt null, 0x0
 ; GFX11-WGP-NEXT:    flat_atomic_cmpswap_b32 v2, v[0:1], v[2:3] offset:16 glc
 ; GFX11-WGP-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX11-WGP-NEXT:    buffer_gl0_inv
 ; GFX11-WGP-NEXT:    buffer_gl1_inv
 ; GFX11-WGP-NEXT:    flat_store_b32 v[0:1], v2
-; GFX11-WGP-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-WGP-NEXT:    s_endpgm
 ;
 ; GFX11-CU-LABEL: flat_agent_seq_cst_monotonic_ret_cmpxchg:
 ; GFX11-CU:       ; %bb.0: ; %entry
 ; GFX11-CU-NEXT:    s_load_b128 s[0:3], s[0:1], 0x0
 ; GFX11-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v0, s0 :: v_dual_mov_b32 v1, s1
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v2, s2 :: v_dual_mov_b32 v3, s3
 ; GFX11-CU-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX11-CU-NEXT:    s_waitcnt_vscnt null, 0x0
 ; GFX11-CU-NEXT:    flat_atomic_cmpswap_b32 v2, v[0:1], v[2:3] offset:16 glc
 ; GFX11-CU-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX11-CU-NEXT:    buffer_gl0_inv
 ; GFX11-CU-NEXT:    buffer_gl1_inv
 ; GFX11-CU-NEXT:    flat_store_b32 v[0:1], v2
-; GFX11-CU-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-CU-NEXT:    s_endpgm
     ptr %out, i32 %in, i32 %old) {
 entry:
   %gep = getelementptr i32, ptr %out, i32 4
   %val = cmpxchg volatile ptr %gep, i32 %old, i32 %in syncscope("agent") seq_cst monotonic
   %val0 = extractvalue { i32, i1 } %val, 0
   store i32 %val0, ptr %out, align 4
   ret void

 ; GFX11-WGP-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-WGP-NEXT:    v_dual_mov_b32 v0, s0 :: v_dual_mov_b32 v1, s1
 ; GFX11-WGP-NEXT:    v_dual_mov_b32 v2, s2 :: v_dual_mov_b32 v3, s3
 ; GFX11-WGP-NEXT:    flat_atomic_cmpswap_b32 v2, v[0:1], v[2:3] offset:16 glc
 ; GFX11-WGP-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX11-WGP-NEXT:    buffer_gl0_inv
 ; GFX11-WGP-NEXT:    buffer_gl1_inv
 ; GFX11-WGP-NEXT:    flat_store_b32 v[0:1], v2
-; GFX11-WGP-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-WGP-NEXT:    s_endpgm
 ;
 ; GFX11-CU-LABEL: flat_agent_monotonic_acquire_ret_cmpxchg:
 ; GFX11-CU:       ; %bb.0: ; %entry
 ; GFX11-CU-NEXT:    s_load_b128 s[0:3], s[0:1], 0x0
 ; GFX11-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v0, s0 :: v_dual_mov_b32 v1, s1
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v2, s2 :: v_dual_mov_b32 v3, s3
 ; GFX11-CU-NEXT:    flat_atomic_cmpswap_b32 v2, v[0:1], v[2:3] offset:16 glc
 ; GFX11-CU-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX11-CU-NEXT:    buffer_gl0_inv
 ; GFX11-CU-NEXT:    buffer_gl1_inv
 ; GFX11-CU-NEXT:    flat_store_b32 v[0:1], v2
-; GFX11-CU-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-CU-NEXT:    s_endpgm
     ptr %out, i32 %in, i32 %old) {
 entry:
   %gep = getelementptr i32, ptr %out, i32 4
   %val = cmpxchg volatile ptr %gep, i32 %old, i32 %in syncscope("agent") monotonic acquire
   %val0 = extractvalue { i32, i1 } %val, 0
   store i32 %val0, ptr %out, align 4
   ret void

 ; GFX11-WGP-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-WGP-NEXT:    v_dual_mov_b32 v0, s0 :: v_dual_mov_b32 v1, s1
 ; GFX11-WGP-NEXT:    v_dual_mov_b32 v2, s2 :: v_dual_mov_b32 v3, s3
 ; GFX11-WGP-NEXT:    flat_atomic_cmpswap_b32 v2, v[0:1], v[2:3] offset:16 glc
 ; GFX11-WGP-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX11-WGP-NEXT:    buffer_gl0_inv
 ; GFX11-WGP-NEXT:    buffer_gl1_inv
 ; GFX11-WGP-NEXT:    flat_store_b32 v[0:1], v2
-; GFX11-WGP-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-WGP-NEXT:    s_endpgm
 ;
 ; GFX11-CU-LABEL: flat_agent_acquire_acquire_ret_cmpxchg:
 ; GFX11-CU:       ; %bb.0: ; %entry
 ; GFX11-CU-NEXT:    s_load_b128 s[0:3], s[0:1], 0x0
 ; GFX11-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v0, s0 :: v_dual_mov_b32 v1, s1
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v2, s2 :: v_dual_mov_b32 v3, s3
 ; GFX11-CU-NEXT:    flat_atomic_cmpswap_b32 v2, v[0:1], v[2:3] offset:16 glc
 ; GFX11-CU-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX11-CU-NEXT:    buffer_gl0_inv
 ; GFX11-CU-NEXT:    buffer_gl1_inv
 ; GFX11-CU-NEXT:    flat_store_b32 v[0:1], v2
-; GFX11-CU-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-CU-NEXT:    s_endpgm
     ptr %out, i32 %in, i32 %old) {
 entry:
   %gep = getelementptr i32, ptr %out, i32 4
   %val = cmpxchg volatile ptr %gep, i32 %old, i32 %in syncscope("agent") acquire acquire
   %val0 = extractvalue { i32, i1 } %val, 0
   store i32 %val0, ptr %out, align 4
   ret void

 ; GFX11-WGP-NEXT:    v_dual_mov_b32 v2, s2 :: v_dual_mov_b32 v3, s3
 ; GFX11-WGP-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX11-WGP-NEXT:    s_waitcnt_vscnt null, 0x0
 ; GFX11-WGP-NEXT:    flat_atomic_cmpswap_b32 v2, v[0:1], v[2:3] offset:16 glc
 ; GFX11-WGP-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX11-WGP-NEXT:    buffer_gl0_inv
 ; GFX11-WGP-NEXT:    buffer_gl1_inv
 ; GFX11-WGP-NEXT:    flat_store_b32 v[0:1], v2
-; GFX11-WGP-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-WGP-NEXT:    s_endpgm
 ;
 ; GFX11-CU-LABEL: flat_agent_release_acquire_ret_cmpxchg:
 ; GFX11-CU:       ; %bb.0: ; %entry
 ; GFX11-CU-NEXT:    s_load_b128 s[0:3], s[0:1], 0x0
 ; GFX11-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v0, s0 :: v_dual_mov_b32 v1, s1
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v2, s2 :: v_dual_mov_b32 v3, s3
 ; GFX11-CU-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX11-CU-NEXT:    s_waitcnt_vscnt null, 0x0
 ; GFX11-CU-NEXT:    flat_atomic_cmpswap_b32 v2, v[0:1], v[2:3] offset:16 glc
 ; GFX11-CU-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX11-CU-NEXT:    buffer_gl0_inv
 ; GFX11-CU-NEXT:    buffer_gl1_inv
 ; GFX11-CU-NEXT:    flat_store_b32 v[0:1], v2
-; GFX11-CU-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-CU-NEXT:    s_endpgm
     ptr %out, i32 %in, i32 %old) {
 entry:
   %gep = getelementptr i32, ptr %out, i32 4
   %val = cmpxchg volatile ptr %gep, i32 %old, i32 %in syncscope("agent") release acquire
   %val0 = extractvalue { i32, i1 } %val, 0
   store i32 %val0, ptr %out, align 4
   ret void

 ; GFX11-WGP-NEXT:    v_dual_mov_b32 v2, s2 :: v_dual_mov_b32 v3, s3
 ; GFX11-WGP-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX11-WGP-NEXT:    s_waitcnt_vscnt null, 0x0
 ; GFX11-WGP-NEXT:    flat_atomic_cmpswap_b32 v2, v[0:1], v[2:3] offset:16 glc
 ; GFX11-WGP-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX11-WGP-NEXT:    buffer_gl0_inv
 ; GFX11-WGP-NEXT:    buffer_gl1_inv
 ; GFX11-WGP-NEXT:    flat_store_b32 v[0:1], v2
-; GFX11-WGP-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-WGP-NEXT:    s_endpgm
 ;
 ; GFX11-CU-LABEL: flat_agent_acq_rel_acquire_ret_cmpxchg:
 ; GFX11-CU:       ; %bb.0: ; %entry
 ; GFX11-CU-NEXT:    s_load_b128 s[0:3], s[0:1], 0x0
 ; GFX11-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v0, s0 :: v_dual_mov_b32 v1, s1
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v2, s2 :: v_dual_mov_b32 v3, s3
 ; GFX11-CU-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX11-CU-NEXT:    s_waitcnt_vscnt null, 0x0
 ; GFX11-CU-NEXT:    flat_atomic_cmpswap_b32 v2, v[0:1], v[2:3] offset:16 glc
 ; GFX11-CU-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX11-CU-NEXT:    buffer_gl0_inv
 ; GFX11-CU-NEXT:    buffer_gl1_inv
 ; GFX11-CU-NEXT:    flat_store_b32 v[0:1], v2
-; GFX11-CU-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-CU-NEXT:    s_endpgm
     ptr %out, i32 %in, i32 %old) {
 entry:
   %gep = getelementptr i32, ptr %out, i32 4
   %val = cmpxchg volatile ptr %gep, i32 %old, i32 %in syncscope("agent") acq_rel acquire
   %val0 = extractvalue { i32, i1 } %val, 0
   store i32 %val0, ptr %out, align 4
   ret void

 ; GFX11-WGP-NEXT:    v_dual_mov_b32 v2, s2 :: v_dual_mov_b32 v3, s3
 ; GFX11-WGP-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX11-WGP-NEXT:    s_waitcnt_vscnt null, 0x0
 ; GFX11-WGP-NEXT:    flat_atomic_cmpswap_b32 v2, v[0:1], v[2:3] offset:16 glc
 ; GFX11-WGP-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX11-WGP-NEXT:    buffer_gl0_inv
 ; GFX11-WGP-NEXT:    buffer_gl1_inv
 ; GFX11-WGP-NEXT:    flat_store_b32 v[0:1], v2
-; GFX11-WGP-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-WGP-NEXT:    s_endpgm
 ;
 ; GFX11-CU-LABEL: flat_agent_seq_cst_acquire_ret_cmpxchg:
 ; GFX11-CU:       ; %bb.0: ; %entry
 ; GFX11-CU-NEXT:    s_load_b128 s[0:3], s[0:1], 0x0
 ; GFX11-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v0, s0 :: v_dual_mov_b32 v1, s1
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v2, s2 :: v_dual_mov_b32 v3, s3
 ; GFX11-CU-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX11-CU-NEXT:    s_waitcnt_vscnt null, 0x0
 ; GFX11-CU-NEXT:    flat_atomic_cmpswap_b32 v2, v[0:1], v[2:3] offset:16 glc
 ; GFX11-CU-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX11-CU-NEXT:    buffer_gl0_inv
 ; GFX11-CU-NEXT:    buffer_gl1_inv
 ; GFX11-CU-NEXT:    flat_store_b32 v[0:1], v2
-; GFX11-CU-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-CU-NEXT:    s_endpgm
     ptr %out, i32 %in, i32 %old) {
 entry:
   %gep = getelementptr i32, ptr %out, i32 4
   %val = cmpxchg volatile ptr %gep, i32 %old, i32 %in syncscope("agent") seq_cst acquire
   %val0 = extractvalue { i32, i1 } %val, 0
   store i32 %val0, ptr %out, align 4
   ret void

 ; GFX11-WGP-NEXT:    v_dual_mov_b32 v2, s2 :: v_dual_mov_b32 v3, s3
 ; GFX11-WGP-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX11-WGP-NEXT:    s_waitcnt_vscnt null, 0x0
 ; GFX11-WGP-NEXT:    flat_atomic_cmpswap_b32 v2, v[0:1], v[2:3] offset:16 glc
 ; GFX11-WGP-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX11-WGP-NEXT:    buffer_gl0_inv
 ; GFX11-WGP-NEXT:    buffer_gl1_inv
 ; GFX11-WGP-NEXT:    flat_store_b32 v[0:1], v2
-; GFX11-WGP-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-WGP-NEXT:    s_endpgm
 ;
 ; GFX11-CU-LABEL: flat_agent_monotonic_seq_cst_ret_cmpxchg:
 ; GFX11-CU:       ; %bb.0: ; %entry
 ; GFX11-CU-NEXT:    s_load_b128 s[0:3], s[0:1], 0x0
 ; GFX11-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v0, s0 :: v_dual_mov_b32 v1, s1
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v2, s2 :: v_dual_mov_b32 v3, s3
 ; GFX11-CU-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX11-CU-NEXT:    s_waitcnt_vscnt null, 0x0
 ; GFX11-CU-NEXT:    flat_atomic_cmpswap_b32 v2, v[0:1], v[2:3] offset:16 glc
 ; GFX11-CU-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX11-CU-NEXT:    buffer_gl0_inv
 ; GFX11-CU-NEXT:    buffer_gl1_inv
 ; GFX11-CU-NEXT:    flat_store_b32 v[0:1], v2
-; GFX11-CU-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-CU-NEXT:    s_endpgm
     ptr %out, i32 %in, i32 %old) {
 entry:
   %gep = getelementptr i32, ptr %out, i32 4
   %val = cmpxchg volatile ptr %gep, i32 %old, i32 %in syncscope("agent") monotonic seq_cst
   %val0 = extractvalue { i32, i1 } %val, 0
   store i32 %val0, ptr %out, align 4
   ret void

 ; GFX11-WGP-NEXT:    v_dual_mov_b32 v2, s2 :: v_dual_mov_b32 v3, s3
 ; GFX11-WGP-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX11-WGP-NEXT:    s_waitcnt_vscnt null, 0x0
 ; GFX11-WGP-NEXT:    flat_atomic_cmpswap_b32 v2, v[0:1], v[2:3] offset:16 glc
 ; GFX11-WGP-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX11-WGP-NEXT:    buffer_gl0_inv
 ; GFX11-WGP-NEXT:    buffer_gl1_inv
 ; GFX11-WGP-NEXT:    flat_store_b32 v[0:1], v2
-; GFX11-WGP-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-WGP-NEXT:    s_endpgm
 ;
 ; GFX11-CU-LABEL: flat_agent_acquire_seq_cst_ret_cmpxchg:
 ; GFX11-CU:       ; %bb.0: ; %entry
 ; GFX11-CU-NEXT:    s_load_b128 s[0:3], s[0:1], 0x0
 ; GFX11-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v0, s0 :: v_dual_mov_b32 v1, s1
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v2, s2 :: v_dual_mov_b32 v3, s3
 ; GFX11-CU-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX11-CU-NEXT:    s_waitcnt_vscnt null, 0x0
 ; GFX11-CU-NEXT:    flat_atomic_cmpswap_b32 v2, v[0:1], v[2:3] offset:16 glc
 ; GFX11-CU-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX11-CU-NEXT:    buffer_gl0_inv
 ; GFX11-CU-NEXT:    buffer_gl1_inv
 ; GFX11-CU-NEXT:    flat_store_b32 v[0:1], v2
-; GFX11-CU-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-CU-NEXT:    s_endpgm
     ptr %out, i32 %in, i32 %old) {
 entry:
   %gep = getelementptr i32, ptr %out, i32 4
   %val = cmpxchg volatile ptr %gep, i32 %old, i32 %in syncscope("agent") acquire seq_cst
   %val0 = extractvalue { i32, i1 } %val, 0
   store i32 %val0, ptr %out, align 4
   ret void

 ; GFX11-WGP-NEXT:    v_dual_mov_b32 v2, s2 :: v_dual_mov_b32 v3, s3
 ; GFX11-WGP-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX11-WGP-NEXT:    s_waitcnt_vscnt null, 0x0
 ; GFX11-WGP-NEXT:    flat_atomic_cmpswap_b32 v2, v[0:1], v[2:3] offset:16 glc
 ; GFX11-WGP-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX11-WGP-NEXT:    buffer_gl0_inv
 ; GFX11-WGP-NEXT:    buffer_gl1_inv
 ; GFX11-WGP-NEXT:    flat_store_b32 v[0:1], v2
-; GFX11-WGP-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-WGP-NEXT:    s_endpgm
 ;
 ; GFX11-CU-LABEL: flat_agent_release_seq_cst_ret_cmpxchg:
 ; GFX11-CU:       ; %bb.0: ; %entry
 ; GFX11-CU-NEXT:    s_load_b128 s[0:3], s[0:1], 0x0
 ; GFX11-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v0, s0 :: v_dual_mov_b32 v1, s1
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v2, s2 :: v_dual_mov_b32 v3, s3
 ; GFX11-CU-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX11-CU-NEXT:    s_waitcnt_vscnt null, 0x0
 ; GFX11-CU-NEXT:    flat_atomic_cmpswap_b32 v2, v[0:1], v[2:3] offset:16 glc
 ; GFX11-CU-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX11-CU-NEXT:    buffer_gl0_inv
 ; GFX11-CU-NEXT:    buffer_gl1_inv
 ; GFX11-CU-NEXT:    flat_store_b32 v[0:1], v2
-; GFX11-CU-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-CU-NEXT:    s_endpgm
     ptr %out, i32 %in, i32 %old) {
 entry:
   %gep = getelementptr i32, ptr %out, i32 4
   %val = cmpxchg volatile ptr %gep, i32 %old, i32 %in syncscope("agent") release seq_cst
   %val0 = extractvalue { i32, i1 } %val, 0
   store i32 %val0, ptr %out, align 4
   ret void

 ; GFX11-WGP-NEXT:    v_dual_mov_b32 v2, s2 :: v_dual_mov_b32 v3, s3
 ; GFX11-WGP-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX11-WGP-NEXT:    s_waitcnt_vscnt null, 0x0
 ; GFX11-WGP-NEXT:    flat_atomic_cmpswap_b32 v2, v[0:1], v[2:3] offset:16 glc
 ; GFX11-WGP-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX11-WGP-NEXT:    buffer_gl0_inv
 ; GFX11-WGP-NEXT:    buffer_gl1_inv
 ; GFX11-WGP-NEXT:    flat_store_b32 v[0:1], v2
-; GFX11-WGP-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-WGP-NEXT:    s_endpgm
 ;
 ; GFX11-CU-LABEL: flat_agent_acq_rel_seq_cst_ret_cmpxchg:
 ; GFX11-CU:       ; %bb.0: ; %entry
 ; GFX11-CU-NEXT:    s_load_b128 s[0:3], s[0:1], 0x0
 ; GFX11-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v0, s0 :: v_dual_mov_b32 v1, s1
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v2, s2 :: v_dual_mov_b32 v3, s3
 ; GFX11-CU-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX11-CU-NEXT:    s_waitcnt_vscnt null, 0x0
 ; GFX11-CU-NEXT:    flat_atomic_cmpswap_b32 v2, v[0:1], v[2:3] offset:16 glc
 ; GFX11-CU-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX11-CU-NEXT:    buffer_gl0_inv
 ; GFX11-CU-NEXT:    buffer_gl1_inv
 ; GFX11-CU-NEXT:    flat_store_b32 v[0:1], v2
-; GFX11-CU-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-CU-NEXT:    s_endpgm
     ptr %out, i32 %in, i32 %old) {
 entry:
   %gep = getelementptr i32, ptr %out, i32 4
   %val = cmpxchg volatile ptr %gep, i32 %old, i32 %in syncscope("agent") acq_rel seq_cst
   %val0 = extractvalue { i32, i1 } %val, 0
   store i32 %val0, ptr %out, align 4
   ret void

 ; GFX11-WGP-NEXT:    v_dual_mov_b32 v2, s2 :: v_dual_mov_b32 v3, s3
 ; GFX11-WGP-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX11-WGP-NEXT:    s_waitcnt_vscnt null, 0x0
 ; GFX11-WGP-NEXT:    flat_atomic_cmpswap_b32 v2, v[0:1], v[2:3] offset:16 glc
 ; GFX11-WGP-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX11-WGP-NEXT:    buffer_gl0_inv
 ; GFX11-WGP-NEXT:    buffer_gl1_inv
 ; GFX11-WGP-NEXT:    flat_store_b32 v[0:1], v2
-; GFX11-WGP-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-WGP-NEXT:    s_endpgm
 ;
 ; GFX11-CU-LABEL: flat_agent_seq_cst_seq_cst_ret_cmpxchg:
 ; GFX11-CU:       ; %bb.0: ; %entry
 ; GFX11-CU-NEXT:    s_load_b128 s[0:3], s[0:1], 0x0
 ; GFX11-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v0, s0 :: v_dual_mov_b32 v1, s1
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v2, s2 :: v_dual_mov_b32 v3, s3
 ; GFX11-CU-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX11-CU-NEXT:    s_waitcnt_vscnt null, 0x0
 ; GFX11-CU-NEXT:    flat_atomic_cmpswap_b32 v2, v[0:1], v[2:3] offset:16 glc
 ; GFX11-CU-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX11-CU-NEXT:    buffer_gl0_inv
 ; GFX11-CU-NEXT:    buffer_gl1_inv
 ; GFX11-CU-NEXT:    flat_store_b32 v[0:1], v2
-; GFX11-CU-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-CU-NEXT:    s_endpgm
     ptr %out, i32 %in, i32 %old) {
 entry:
   %gep = getelementptr i32, ptr %out, i32 4
   %val = cmpxchg volatile ptr %gep, i32 %old, i32 %in syncscope("agent") seq_cst seq_cst
   %val0 = extractvalue { i32, i1 } %val, 0
   store i32 %val0, ptr %out, align 4
   ret void

 ; GFX11-WGP:       ; %bb.0: ; %entry
 ; GFX11-WGP-NEXT:    s_load_b128 s[0:3], s[0:1], 0x0
 ; GFX11-WGP-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-WGP-NEXT:    v_dual_mov_b32 v0, s0 :: v_dual_mov_b32 v1, s1
 ; GFX11-WGP-NEXT:    flat_load_b32 v2, v[0:1]
 ; GFX11-WGP-NEXT:    v_dual_mov_b32 v0, s2 :: v_dual_mov_b32 v1, s3
 ; GFX11-WGP-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX11-WGP-NEXT:    flat_store_b32 v[0:1], v2
-; GFX11-WGP-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-WGP-NEXT:    s_endpgm
 ;
 ; GFX11-CU-LABEL: flat_agent_one_as_unordered_load:
 ; GFX11-CU:       ; %bb.0: ; %entry
 ; GFX11-CU-NEXT:    s_load_b128 s[0:3], s[0:1], 0x0
 ; GFX11-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v0, s0 :: v_dual_mov_b32 v1, s1
 ; GFX11-CU-NEXT:    flat_load_b32 v2, v[0:1]
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v0, s2 :: v_dual_mov_b32 v1, s3
 ; GFX11-CU-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX11-CU-NEXT:    flat_store_b32 v[0:1], v2
-; GFX11-CU-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-CU-NEXT:    s_endpgm
     ptr %in, ptr %out) {
 entry:
   %val = load atomic i32, ptr %in syncscope("agent-one-as") unordered, align 4
   store i32 %val, ptr %out
   ret void
 }
 

 ; GFX11-WGP:       ; %bb.0: ; %entry
 ; GFX11-WGP-NEXT:    s_load_b128 s[0:3], s[0:1], 0x0
 ; GFX11-WGP-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-WGP-NEXT:    v_dual_mov_b32 v0, s0 :: v_dual_mov_b32 v1, s1
 ; GFX11-WGP-NEXT:    flat_load_b32 v2, v[0:1] glc
 ; GFX11-WGP-NEXT:    v_dual_mov_b32 v0, s2 :: v_dual_mov_b32 v1, s3
 ; GFX11-WGP-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX11-WGP-NEXT:    flat_store_b32 v[0:1], v2
-; GFX11-WGP-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-WGP-NEXT:    s_endpgm
 ;
 ; GFX11-CU-LABEL: flat_agent_one_as_monotonic_load:
 ; GFX11-CU:       ; %bb.0: ; %entry
 ; GFX11-CU-NEXT:    s_load_b128 s[0:3], s[0:1], 0x0
 ; GFX11-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v0, s0 :: v_dual_mov_b32 v1, s1
 ; GFX11-CU-NEXT:    flat_load_b32 v2, v[0:1] glc
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v0, s2 :: v_dual_mov_b32 v1, s3
 ; GFX11-CU-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX11-CU-NEXT:    flat_store_b32 v[0:1], v2
-; GFX11-CU-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-CU-NEXT:    s_endpgm
     ptr %in, ptr %out) {
 entry:
   %val = load atomic i32, ptr %in syncscope("agent-one-as") monotonic, align 4
   store i32 %val, ptr %out
   ret void
 }
 

 ; GFX11-WGP-NEXT:    v_dual_mov_b32 v0, s0 :: v_dual_mov_b32 v1, s1
 ; GFX11-WGP-NEXT:    flat_load_b32 v2, v[0:1] glc
 ; GFX11-WGP-NEXT:    s_waitcnt vmcnt(0)
 ; GFX11-WGP-NEXT:    buffer_gl0_inv
 ; GFX11-WGP-NEXT:    buffer_gl1_inv
 ; GFX11-WGP-NEXT:    v_dual_mov_b32 v0, s2 :: v_dual_mov_b32 v1, s3
 ; GFX11-WGP-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-WGP-NEXT:    flat_store_b32 v[0:1], v2
-; GFX11-WGP-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-WGP-NEXT:    s_endpgm
 ;
 ; GFX11-CU-LABEL: flat_agent_one_as_acquire_load:
 ; GFX11-CU:       ; %bb.0: ; %entry
 ; GFX11-CU-NEXT:    s_load_b128 s[0:3], s[0:1], 0x0
 ; GFX11-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v0, s0 :: v_dual_mov_b32 v1, s1
 ; GFX11-CU-NEXT:    flat_load_b32 v2, v[0:1] glc
 ; GFX11-CU-NEXT:    s_waitcnt vmcnt(0)
 ; GFX11-CU-NEXT:    buffer_gl0_inv
 ; GFX11-CU-NEXT:    buffer_gl1_inv
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v0, s2 :: v_dual_mov_b32 v1, s3
 ; GFX11-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-CU-NEXT:    flat_store_b32 v[0:1], v2
-; GFX11-CU-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-CU-NEXT:    s_endpgm
     ptr %in, ptr %out) {
 entry:
   %val = load atomic i32, ptr %in syncscope("agent-one-as") acquire, align 4
   store i32 %val, ptr %out
   ret void
 }
 

 ; GFX11-WGP-NEXT:    s_waitcnt_vscnt null, 0x0
 ; GFX11-WGP-NEXT:    flat_load_b32 v2, v[0:1] glc
 ; GFX11-WGP-NEXT:    s_waitcnt vmcnt(0)
 ; GFX11-WGP-NEXT:    buffer_gl0_inv
 ; GFX11-WGP-NEXT:    buffer_gl1_inv
 ; GFX11-WGP-NEXT:    v_dual_mov_b32 v0, s2 :: v_dual_mov_b32 v1, s3
 ; GFX11-WGP-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-WGP-NEXT:    flat_store_b32 v[0:1], v2
-; GFX11-WGP-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-WGP-NEXT:    s_endpgm
 ;
 ; GFX11-CU-LABEL: flat_agent_one_as_seq_cst_load:
 ; GFX11-CU:       ; %bb.0: ; %entry
 ; GFX11-CU-NEXT:    s_load_b128 s[0:3], s[0:1], 0x0
 ; GFX11-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v0, s0 :: v_dual_mov_b32 v1, s1
 ; GFX11-CU-NEXT:    s_waitcnt vmcnt(0)
 ; GFX11-CU-NEXT:    s_waitcnt_vscnt null, 0x0
 ; GFX11-CU-NEXT:    flat_load_b32 v2, v[0:1] glc
 ; GFX11-CU-NEXT:    s_waitcnt vmcnt(0)
 ; GFX11-CU-NEXT:    buffer_gl0_inv
 ; GFX11-CU-NEXT:    buffer_gl1_inv
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v0, s2 :: v_dual_mov_b32 v1, s3
 ; GFX11-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-CU-NEXT:    flat_store_b32 v[0:1], v2
-; GFX11-CU-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-CU-NEXT:    s_endpgm
     ptr %in, ptr %out) {
 entry:
   %val = load atomic i32, ptr %in syncscope("agent-one-as") seq_cst, align 4
   store i32 %val, ptr %out
   ret void
 }
 

 ; GFX11-WGP:       ; %bb.0: ; %entry
 ; GFX11-WGP-NEXT:    s_clause 0x1
 ; GFX11-WGP-NEXT:    s_load_b64 s[2:3], s[0:1], 0x8
 ; GFX11-WGP-NEXT:    s_load_b32 s0, s[0:1], 0x0
 ; GFX11-WGP-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-WGP-NEXT:    v_dual_mov_b32 v0, s2 :: v_dual_mov_b32 v1, s3
 ; GFX11-WGP-NEXT:    v_mov_b32_e32 v2, s0
 ; GFX11-WGP-NEXT:    flat_store_b32 v[0:1], v2
-; GFX11-WGP-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-WGP-NEXT:    s_endpgm
 ;
 ; GFX11-CU-LABEL: flat_agent_one_as_unordered_store:
 ; GFX11-CU:       ; %bb.0: ; %entry
 ; GFX11-CU-NEXT:    s_clause 0x1
 ; GFX11-CU-NEXT:    s_load_b64 s[2:3], s[0:1], 0x8
 ; GFX11-CU-NEXT:    s_load_b32 s0, s[0:1], 0x0
 ; GFX11-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v0, s2 :: v_dual_mov_b32 v1, s3
 ; GFX11-CU-NEXT:    v_mov_b32_e32 v2, s0
 ; GFX11-CU-NEXT:    flat_store_b32 v[0:1], v2
-; GFX11-CU-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-CU-NEXT:    s_endpgm
     i32 %in, ptr %out) {
 entry:
   store atomic i32 %in, ptr %out syncscope("agent-one-as") unordered, align 4
   ret void
 }
 
 define amdgpu_kernel void @flat_agent_one_as_monotonic_store(

 ; GFX11-WGP:       ; %bb.0: ; %entry
 ; GFX11-WGP-NEXT:    s_clause 0x1
 ; GFX11-WGP-NEXT:    s_load_b64 s[2:3], s[0:1], 0x8
 ; GFX11-WGP-NEXT:    s_load_b32 s0, s[0:1], 0x0
 ; GFX11-WGP-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-WGP-NEXT:    v_dual_mov_b32 v0, s2 :: v_dual_mov_b32 v1, s3
 ; GFX11-WGP-NEXT:    v_mov_b32_e32 v2, s0
 ; GFX11-WGP-NEXT:    flat_store_b32 v[0:1], v2
-; GFX11-WGP-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-WGP-NEXT:    s_endpgm
 ;
 ; GFX11-CU-LABEL: flat_agent_one_as_monotonic_store:
 ; GFX11-CU:       ; %bb.0: ; %entry
 ; GFX11-CU-NEXT:    s_clause 0x1
 ; GFX11-CU-NEXT:    s_load_b64 s[2:3], s[0:1], 0x8
 ; GFX11-CU-NEXT:    s_load_b32 s0, s[0:1], 0x0
 ; GFX11-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v0, s2 :: v_dual_mov_b32 v1, s3
 ; GFX11-CU-NEXT:    v_mov_b32_e32 v2, s0
 ; GFX11-CU-NEXT:    flat_store_b32 v[0:1], v2
-; GFX11-CU-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-CU-NEXT:    s_endpgm
     i32 %in, ptr %out) {
 entry:
   store atomic i32 %in, ptr %out syncscope("agent-one-as") monotonic, align 4
   ret void
 }
 
 define amdgpu_kernel void @flat_agent_one_as_release_store(

 ; GFX11-WGP-NEXT:    s_load_b64 s[2:3], s[0:1], 0x8
 ; GFX11-WGP-NEXT:    s_load_b32 s0, s[0:1], 0x0
 ; GFX11-WGP-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-WGP-NEXT:    v_dual_mov_b32 v0, s2 :: v_dual_mov_b32 v1, s3
 ; GFX11-WGP-NEXT:    v_mov_b32_e32 v2, s0
 ; GFX11-WGP-NEXT:    s_waitcnt vmcnt(0)
 ; GFX11-WGP-NEXT:    s_waitcnt_vscnt null, 0x0
 ; GFX11-WGP-NEXT:    flat_store_b32 v[0:1], v2
-; GFX11-WGP-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-WGP-NEXT:    s_endpgm
 ;
 ; GFX11-CU-LABEL: flat_agent_one_as_release_store:
 ; GFX11-CU:       ; %bb.0: ; %entry
 ; GFX11-CU-NEXT:    s_clause 0x1
 ; GFX11-CU-NEXT:    s_load_b64 s[2:3], s[0:1], 0x8
 ; GFX11-CU-NEXT:    s_load_b32 s0, s[0:1], 0x0
 ; GFX11-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v0, s2 :: v_dual_mov_b32 v1, s3
 ; GFX11-CU-NEXT:    v_mov_b32_e32 v2, s0
 ; GFX11-CU-NEXT:    s_waitcnt vmcnt(0)
 ; GFX11-CU-NEXT:    s_waitcnt_vscnt null, 0x0
 ; GFX11-CU-NEXT:    flat_store_b32 v[0:1], v2
-; GFX11-CU-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-CU-NEXT:    s_endpgm
     i32 %in, ptr %out) {
 entry:
   store atomic i32 %in, ptr %out syncscope("agent-one-as") release, align 4
   ret void
 }
 
 define amdgpu_kernel void @flat_agent_one_as_seq_cst_store(

 ; GFX11-WGP-NEXT:    s_load_b64 s[2:3], s[0:1], 0x8
 ; GFX11-WGP-NEXT:    s_load_b32 s0, s[0:1], 0x0
 ; GFX11-WGP-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-WGP-NEXT:    v_dual_mov_b32 v0, s2 :: v_dual_mov_b32 v1, s3
 ; GFX11-WGP-NEXT:    v_mov_b32_e32 v2, s0
 ; GFX11-WGP-NEXT:    s_waitcnt vmcnt(0)
 ; GFX11-WGP-NEXT:    s_waitcnt_vscnt null, 0x0
 ; GFX11-WGP-NEXT:    flat_store_b32 v[0:1], v2
-; GFX11-WGP-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-WGP-NEXT:    s_endpgm
 ;
 ; GFX11-CU-LABEL: flat_agent_one_as_seq_cst_store:
 ; GFX11-CU:       ; %bb.0: ; %entry
 ; GFX11-CU-NEXT:    s_clause 0x1
 ; GFX11-CU-NEXT:    s_load_b64 s[2:3], s[0:1], 0x8
 ; GFX11-CU-NEXT:    s_load_b32 s0, s[0:1], 0x0
 ; GFX11-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v0, s2 :: v_dual_mov_b32 v1, s3
 ; GFX11-CU-NEXT:    v_mov_b32_e32 v2, s0
 ; GFX11-CU-NEXT:    s_waitcnt vmcnt(0)
 ; GFX11-CU-NEXT:    s_waitcnt_vscnt null, 0x0
 ; GFX11-CU-NEXT:    flat_store_b32 v[0:1], v2
-; GFX11-CU-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-CU-NEXT:    s_endpgm
     i32 %in, ptr %out) {
 entry:
   store atomic i32 %in, ptr %out syncscope("agent-one-as") seq_cst, align 4
   ret void
 }
 
 define amdgpu_kernel void @flat_agent_one_as_monotonic_atomicrmw(

 ; GFX11-WGP:       ; %bb.0: ; %entry
 ; GFX11-WGP-NEXT:    s_clause 0x1
 ; GFX11-WGP-NEXT:    s_load_b64 s[2:3], s[0:1], 0x0
 ; GFX11-WGP-NEXT:    s_load_b32 s0, s[0:1], 0x8
 ; GFX11-WGP-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-WGP-NEXT:    v_dual_mov_b32 v0, s2 :: v_dual_mov_b32 v1, s3
 ; GFX11-WGP-NEXT:    v_mov_b32_e32 v2, s0
 ; GFX11-WGP-NEXT:    flat_atomic_swap_b32 v[0:1], v2
-; GFX11-WGP-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-WGP-NEXT:    s_endpgm
 ;
 ; GFX11-CU-LABEL: flat_agent_one_as_monotonic_atomicrmw:
 ; GFX11-CU:       ; %bb.0: ; %entry
 ; GFX11-CU-NEXT:    s_clause 0x1
 ; GFX11-CU-NEXT:    s_load_b64 s[2:3], s[0:1], 0x0
 ; GFX11-CU-NEXT:    s_load_b32 s0, s[0:1], 0x8
 ; GFX11-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v0, s2 :: v_dual_mov_b32 v1, s3
 ; GFX11-CU-NEXT:    v_mov_b32_e32 v2, s0
 ; GFX11-CU-NEXT:    flat_atomic_swap_b32 v[0:1], v2
-; GFX11-CU-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-CU-NEXT:    s_endpgm
     ptr %out, i32 %in) {
 entry:
   %val = atomicrmw volatile xchg ptr %out, i32 %in syncscope("agent-one-as") monotonic
   ret void
 }
 
 define amdgpu_kernel void @flat_agent_one_as_acquire_atomicrmw(

 ; GFX11-WGP-NEXT:    s_load_b64 s[2:3], s[0:1], 0x0
 ; GFX11-WGP-NEXT:    s_load_b32 s0, s[0:1], 0x8
 ; GFX11-WGP-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-WGP-NEXT:    v_dual_mov_b32 v0, s2 :: v_dual_mov_b32 v1, s3
 ; GFX11-WGP-NEXT:    v_mov_b32_e32 v2, s0
 ; GFX11-WGP-NEXT:    s_waitcnt vmcnt(0)
 ; GFX11-WGP-NEXT:    s_waitcnt_vscnt null, 0x0
 ; GFX11-WGP-NEXT:    flat_atomic_swap_b32 v[0:1], v2
-; GFX11-WGP-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-WGP-NEXT:    s_endpgm
 ;
 ; GFX11-CU-LABEL: flat_agent_one_as_release_atomicrmw:
 ; GFX11-CU:       ; %bb.0: ; %entry
 ; GFX11-CU-NEXT:    s_clause 0x1
 ; GFX11-CU-NEXT:    s_load_b64 s[2:3], s[0:1], 0x0
 ; GFX11-CU-NEXT:    s_load_b32 s0, s[0:1], 0x8
 ; GFX11-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v0, s2 :: v_dual_mov_b32 v1, s3
 ; GFX11-CU-NEXT:    v_mov_b32_e32 v2, s0
 ; GFX11-CU-NEXT:    s_waitcnt vmcnt(0)
 ; GFX11-CU-NEXT:    s_waitcnt_vscnt null, 0x0
 ; GFX11-CU-NEXT:    flat_atomic_swap_b32 v[0:1], v2
-; GFX11-CU-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-CU-NEXT:    s_endpgm
     ptr %out, i32 %in) {
 entry:
   %val = atomicrmw volatile xchg ptr %out, i32 %in syncscope("agent-one-as") release
   ret void
 }
 
 define amdgpu_kernel void @flat_agent_one_as_acq_rel_atomicrmw(

 ; GFX11-WGP-NEXT:    v_dual_mov_b32 v0, s2 :: v_dual_mov_b32 v1, s3
 ; GFX11-WGP-NEXT:    v_mov_b32_e32 v2, s0
 ; GFX11-WGP-NEXT:    flat_atomic_swap_b32 v2, v[0:1], v2 glc
 ; GFX11-WGP-NEXT:    s_waitcnt vmcnt(0)
 ; GFX11-WGP-NEXT:    buffer_gl0_inv
 ; GFX11-WGP-NEXT:    buffer_gl1_inv
 ; GFX11-WGP-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-WGP-NEXT:    flat_store_b32 v[0:1], v2
-; GFX11-WGP-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-WGP-NEXT:    s_endpgm
 ;
 ; GFX11-CU-LABEL: flat_agent_one_as_acquire_ret_atomicrmw:
 ; GFX11-CU:       ; %bb.0: ; %entry
 ; GFX11-CU-NEXT:    s_clause 0x1
 ; GFX11-CU-NEXT:    s_load_b64 s[2:3], s[0:1], 0x0
 ; GFX11-CU-NEXT:    s_load_b32 s0, s[0:1], 0x8
 ; GFX11-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v0, s2 :: v_dual_mov_b32 v1, s3
 ; GFX11-CU-NEXT:    v_mov_b32_e32 v2, s0
 ; GFX11-CU-NEXT:    flat_atomic_swap_b32 v2, v[0:1], v2 glc
 ; GFX11-CU-NEXT:    s_waitcnt vmcnt(0)
 ; GFX11-CU-NEXT:    buffer_gl0_inv
 ; GFX11-CU-NEXT:    buffer_gl1_inv
 ; GFX11-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-CU-NEXT:    flat_store_b32 v[0:1], v2
-; GFX11-CU-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-CU-NEXT:    s_endpgm
     ptr %out, i32 %in) {
 entry:
   %val = atomicrmw volatile xchg ptr %out, i32 %in syncscope("agent-one-as") acquire
   store i32 %val, ptr %out, align 4
   ret void
 }
 

 ; GFX11-WGP-NEXT:    s_waitcnt vmcnt(0)
 ; GFX11-WGP-NEXT:    s_waitcnt_vscnt null, 0x0
 ; GFX11-WGP-NEXT:    flat_atomic_swap_b32 v2, v[0:1], v2 glc
 ; GFX11-WGP-NEXT:    s_waitcnt vmcnt(0)
 ; GFX11-WGP-NEXT:    buffer_gl0_inv
 ; GFX11-WGP-NEXT:    buffer_gl1_inv
 ; GFX11-WGP-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-WGP-NEXT:    flat_store_b32 v[0:1], v2
-; GFX11-WGP-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-WGP-NEXT:    s_endpgm
 ;
 ; GFX11-CU-LABEL: flat_agent_one_as_acq_rel_ret_atomicrmw:
 ; GFX11-CU:       ; %bb.0: ; %entry
 ; GFX11-CU-NEXT:    s_clause 0x1
 ; GFX11-CU-NEXT:    s_load_b64 s[2:3], s[0:1], 0x0
 ; GFX11-CU-NEXT:    s_load_b32 s0, s[0:1], 0x8
 ; GFX11-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v0, s2 :: v_dual_mov_b32 v1, s3
 ; GFX11-CU-NEXT:    v_mov_b32_e32 v2, s0
 ; GFX11-CU-NEXT:    s_waitcnt vmcnt(0)
 ; GFX11-CU-NEXT:    s_waitcnt_vscnt null, 0x0
 ; GFX11-CU-NEXT:    flat_atomic_swap_b32 v2, v[0:1], v2 glc
 ; GFX11-CU-NEXT:    s_waitcnt vmcnt(0)
 ; GFX11-CU-NEXT:    buffer_gl0_inv
 ; GFX11-CU-NEXT:    buffer_gl1_inv
 ; GFX11-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-CU-NEXT:    flat_store_b32 v[0:1], v2
-; GFX11-CU-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-CU-NEXT:    s_endpgm
     ptr %out, i32 %in) {
 entry:
   %val = atomicrmw volatile xchg ptr %out, i32 %in syncscope("agent-one-as") acq_rel
   store i32 %val, ptr %out, align 4
   ret void
 }
 

 ; GFX11-WGP-NEXT:    s_waitcnt vmcnt(0)
 ; GFX11-WGP-NEXT:    s_waitcnt_vscnt null, 0x0
 ; GFX11-WGP-NEXT:    flat_atomic_swap_b32 v2, v[0:1], v2 glc
 ; GFX11-WGP-NEXT:    s_waitcnt vmcnt(0)
 ; GFX11-WGP-NEXT:    buffer_gl0_inv
 ; GFX11-WGP-NEXT:    buffer_gl1_inv
 ; GFX11-WGP-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-WGP-NEXT:    flat_store_b32 v[0:1], v2
-; GFX11-WGP-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-WGP-NEXT:    s_endpgm
 ;
 ; GFX11-CU-LABEL: flat_agent_one_as_seq_cst_ret_atomicrmw:
 ; GFX11-CU:       ; %bb.0: ; %entry
 ; GFX11-CU-NEXT:    s_clause 0x1
 ; GFX11-CU-NEXT:    s_load_b64 s[2:3], s[0:1], 0x0
 ; GFX11-CU-NEXT:    s_load_b32 s0, s[0:1], 0x8
 ; GFX11-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v0, s2 :: v_dual_mov_b32 v1, s3
 ; GFX11-CU-NEXT:    v_mov_b32_e32 v2, s0
 ; GFX11-CU-NEXT:    s_waitcnt vmcnt(0)
 ; GFX11-CU-NEXT:    s_waitcnt_vscnt null, 0x0
 ; GFX11-CU-NEXT:    flat_atomic_swap_b32 v2, v[0:1], v2 glc
 ; GFX11-CU-NEXT:    s_waitcnt vmcnt(0)
 ; GFX11-CU-NEXT:    buffer_gl0_inv
 ; GFX11-CU-NEXT:    buffer_gl1_inv
 ; GFX11-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-CU-NEXT:    flat_store_b32 v[0:1], v2
-; GFX11-CU-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-CU-NEXT:    s_endpgm
     ptr %out, i32 %in) {
 entry:
   %val = atomicrmw volatile xchg ptr %out, i32 %in syncscope("agent-one-as") seq_cst
   store i32 %val, ptr %out, align 4
   ret void
 }
 

 ;
 ; GFX11-WGP-LABEL: flat_agent_one_as_monotonic_monotonic_cmpxchg:
 ; GFX11-WGP:       ; %bb.0: ; %entry
 ; GFX11-WGP-NEXT:    s_load_b128 s[0:3], s[0:1], 0x0
 ; GFX11-WGP-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-WGP-NEXT:    v_dual_mov_b32 v0, s0 :: v_dual_mov_b32 v1, s1
 ; GFX11-WGP-NEXT:    v_dual_mov_b32 v2, s2 :: v_dual_mov_b32 v3, s3
 ; GFX11-WGP-NEXT:    flat_atomic_cmpswap_b32 v[0:1], v[2:3] offset:16
-; GFX11-WGP-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-WGP-NEXT:    s_endpgm
 ;
 ; GFX11-CU-LABEL: flat_agent_one_as_monotonic_monotonic_cmpxchg:
 ; GFX11-CU:       ; %bb.0: ; %entry
 ; GFX11-CU-NEXT:    s_load_b128 s[0:3], s[0:1], 0x0
 ; GFX11-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v0, s0 :: v_dual_mov_b32 v1, s1
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v2, s2 :: v_dual_mov_b32 v3, s3
 ; GFX11-CU-NEXT:    flat_atomic_cmpswap_b32 v[0:1], v[2:3] offset:16
-; GFX11-CU-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-CU-NEXT:    s_endpgm
     ptr %out, i32 %in, i32 %old) {
 entry:
   %gep = getelementptr i32, ptr %out, i32 4
   %val = cmpxchg volatile ptr %gep, i32 %old, i32 %in syncscope("agent-one-as") monotonic monotonic
   ret void
 }
 

 ; GFX11-WGP:       ; %bb.0: ; %entry
 ; GFX11-WGP-NEXT:    s_load_b128 s[0:3], s[0:1], 0x0
 ; GFX11-WGP-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-WGP-NEXT:    v_dual_mov_b32 v0, s0 :: v_dual_mov_b32 v1, s1
 ; GFX11-WGP-NEXT:    v_dual_mov_b32 v2, s2 :: v_dual_mov_b32 v3, s3
 ; GFX11-WGP-NEXT:    s_waitcnt vmcnt(0)
 ; GFX11-WGP-NEXT:    s_waitcnt_vscnt null, 0x0
 ; GFX11-WGP-NEXT:    flat_atomic_cmpswap_b32 v[0:1], v[2:3] offset:16
-; GFX11-WGP-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-WGP-NEXT:    s_endpgm
 ;
 ; GFX11-CU-LABEL: flat_agent_one_as_release_monotonic_cmpxchg:
 ; GFX11-CU:       ; %bb.0: ; %entry
 ; GFX11-CU-NEXT:    s_load_b128 s[0:3], s[0:1], 0x0
 ; GFX11-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v0, s0 :: v_dual_mov_b32 v1, s1
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v2, s2 :: v_dual_mov_b32 v3, s3
 ; GFX11-CU-NEXT:    s_waitcnt vmcnt(0)
 ; GFX11-CU-NEXT:    s_waitcnt_vscnt null, 0x0
 ; GFX11-CU-NEXT:    flat_atomic_cmpswap_b32 v[0:1], v[2:3] offset:16
-; GFX11-CU-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-CU-NEXT:    s_endpgm
     ptr %out, i32 %in, i32 %old) {
 entry:
   %gep = getelementptr i32, ptr %out, i32 4
   %val = cmpxchg volatile ptr %gep, i32 %old, i32 %in syncscope("agent-one-as") release monotonic
   ret void
 }
 

 ; GFX11-WGP:       ; %bb.0: ; %entry
 ; GFX11-WGP-NEXT:    s_load_b128 s[0:3], s[0:1], 0x0
 ; GFX11-WGP-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-WGP-NEXT:    v_dual_mov_b32 v0, s0 :: v_dual_mov_b32 v1, s1
 ; GFX11-WGP-NEXT:    v_dual_mov_b32 v2, s2 :: v_dual_mov_b32 v3, s3
 ; GFX11-WGP-NEXT:    flat_atomic_cmpswap_b32 v2, v[0:1], v[2:3] offset:16 glc
 ; GFX11-WGP-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX11-WGP-NEXT:    flat_store_b32 v[0:1], v2
-; GFX11-WGP-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-WGP-NEXT:    s_endpgm
 ;
 ; GFX11-CU-LABEL: flat_agent_one_as_monotonic_monotonic_ret_cmpxchg:
 ; GFX11-CU:       ; %bb.0: ; %entry
 ; GFX11-CU-NEXT:    s_load_b128 s[0:3], s[0:1], 0x0
 ; GFX11-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v0, s0 :: v_dual_mov_b32 v1, s1
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v2, s2 :: v_dual_mov_b32 v3, s3
 ; GFX11-CU-NEXT:    flat_atomic_cmpswap_b32 v2, v[0:1], v[2:3] offset:16 glc
 ; GFX11-CU-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX11-CU-NEXT:    flat_store_b32 v[0:1], v2
-; GFX11-CU-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-CU-NEXT:    s_endpgm
     ptr %out, i32 %in, i32 %old) {
 entry:
   %gep = getelementptr i32, ptr %out, i32 4
   %val = cmpxchg volatile ptr %gep, i32 %old, i32 %in syncscope("agent-one-as") monotonic monotonic
   %val0 = extractvalue { i32, i1 } %val, 0
   store i32 %val0, ptr %out, align 4
   ret void

 ; GFX11-WGP-NEXT:    v_dual_mov_b32 v0, s0 :: v_dual_mov_b32 v1, s1
 ; GFX11-WGP-NEXT:    v_dual_mov_b32 v2, s2 :: v_dual_mov_b32 v3, s3
 ; GFX11-WGP-NEXT:    flat_atomic_cmpswap_b32 v2, v[0:1], v[2:3] offset:16 glc
 ; GFX11-WGP-NEXT:    s_waitcnt vmcnt(0)
 ; GFX11-WGP-NEXT:    buffer_gl0_inv
 ; GFX11-WGP-NEXT:    buffer_gl1_inv
 ; GFX11-WGP-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-WGP-NEXT:    flat_store_b32 v[0:1], v2
-; GFX11-WGP-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-WGP-NEXT:    s_endpgm
 ;
 ; GFX11-CU-LABEL: flat_agent_one_as_acquire_monotonic_ret_cmpxchg:
 ; GFX11-CU:       ; %bb.0: ; %entry
 ; GFX11-CU-NEXT:    s_load_b128 s[0:3], s[0:1], 0x0
 ; GFX11-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v0, s0 :: v_dual_mov_b32 v1, s1
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v2, s2 :: v_dual_mov_b32 v3, s3
 ; GFX11-CU-NEXT:    flat_atomic_cmpswap_b32 v2, v[0:1], v[2:3] offset:16 glc
 ; GFX11-CU-NEXT:    s_waitcnt vmcnt(0)
 ; GFX11-CU-NEXT:    buffer_gl0_inv
 ; GFX11-CU-NEXT:    buffer_gl1_inv
 ; GFX11-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-CU-NEXT:    flat_store_b32 v[0:1], v2
-; GFX11-CU-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-CU-NEXT:    s_endpgm
     ptr %out, i32 %in, i32 %old) {
 entry:
   %gep = getelementptr i32, ptr %out, i32 4
   %val = cmpxchg volatile ptr %gep, i32 %old, i32 %in syncscope("agent-one-as") acquire monotonic
   %val0 = extractvalue { i32, i1 } %val, 0
   store i32 %val0, ptr %out, align 4
   ret void

 ; GFX11-WGP-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-WGP-NEXT:    v_dual_mov_b32 v0, s0 :: v_dual_mov_b32 v1, s1
 ; GFX11-WGP-NEXT:    v_dual_mov_b32 v2, s2 :: v_dual_mov_b32 v3, s3
 ; GFX11-WGP-NEXT:    s_waitcnt vmcnt(0)
 ; GFX11-WGP-NEXT:    s_waitcnt_vscnt null, 0x0
 ; GFX11-WGP-NEXT:    flat_atomic_cmpswap_b32 v2, v[0:1], v[2:3] offset:16 glc
 ; GFX11-WGP-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX11-WGP-NEXT:    flat_store_b32 v[0:1], v2
-; GFX11-WGP-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-WGP-NEXT:    s_endpgm
 ;
 ; GFX11-CU-LABEL: flat_agent_one_as_release_monotonic_ret_cmpxchg:
 ; GFX11-CU:       ; %bb.0: ; %entry
 ; GFX11-CU-NEXT:    s_load_b128 s[0:3], s[0:1], 0x0
 ; GFX11-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v0, s0 :: v_dual_mov_b32 v1, s1
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v2, s2 :: v_dual_mov_b32 v3, s3
 ; GFX11-CU-NEXT:    s_waitcnt vmcnt(0)
 ; GFX11-CU-NEXT:    s_waitcnt_vscnt null, 0x0
 ; GFX11-CU-NEXT:    flat_atomic_cmpswap_b32 v2, v[0:1], v[2:3] offset:16 glc
 ; GFX11-CU-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX11-CU-NEXT:    flat_store_b32 v[0:1], v2
-; GFX11-CU-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-CU-NEXT:    s_endpgm
     ptr %out, i32 %in, i32 %old) {
 entry:
   %gep = getelementptr i32, ptr %out, i32 4
   %val = cmpxchg volatile ptr %gep, i32 %old, i32 %in syncscope("agent-one-as") release monotonic
   %val0 = extractvalue { i32, i1 } %val, 0
   store i32 %val0, ptr %out, align 4
   ret void

 ; GFX11-WGP-NEXT:    s_waitcnt vmcnt(0)
 ; GFX11-WGP-NEXT:    s_waitcnt_vscnt null, 0x0
 ; GFX11-WGP-NEXT:    flat_atomic_cmpswap_b32 v2, v[0:1], v[2:3] offset:16 glc
 ; GFX11-WGP-NEXT:    s_waitcnt vmcnt(0)
 ; GFX11-WGP-NEXT:    buffer_gl0_inv
 ; GFX11-WGP-NEXT:    buffer_gl1_inv
 ; GFX11-WGP-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-WGP-NEXT:    flat_store_b32 v[0:1], v2
-; GFX11-WGP-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-WGP-NEXT:    s_endpgm
 ;
 ; GFX11-CU-LABEL: flat_agent_one_as_acq_rel_monotonic_ret_cmpxchg:
 ; GFX11-CU:       ; %bb.0: ; %entry
 ; GFX11-CU-NEXT:    s_load_b128 s[0:3], s[0:1], 0x0
 ; GFX11-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v0, s0 :: v_dual_mov_b32 v1, s1
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v2, s2 :: v_dual_mov_b32 v3, s3
 ; GFX11-CU-NEXT:    s_waitcnt vmcnt(0)
 ; GFX11-CU-NEXT:    s_waitcnt_vscnt null, 0x0
 ; GFX11-CU-NEXT:    flat_atomic_cmpswap_b32 v2, v[0:1], v[2:3] offset:16 glc
 ; GFX11-CU-NEXT:    s_waitcnt vmcnt(0)
 ; GFX11-CU-NEXT:    buffer_gl0_inv
 ; GFX11-CU-NEXT:    buffer_gl1_inv
 ; GFX11-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-CU-NEXT:    flat_store_b32 v[0:1], v2
-; GFX11-CU-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-CU-NEXT:    s_endpgm
     ptr %out, i32 %in, i32 %old) {
 entry:
   %gep = getelementptr i32, ptr %out, i32 4
   %val = cmpxchg volatile ptr %gep, i32 %old, i32 %in syncscope("agent-one-as") acq_rel monotonic
   %val0 = extractvalue { i32, i1 } %val, 0
   store i32 %val0, ptr %out, align 4
   ret void

 ; GFX11-WGP-NEXT:    s_waitcnt vmcnt(0)
 ; GFX11-WGP-NEXT:    s_waitcnt_vscnt null, 0x0
 ; GFX11-WGP-NEXT:    flat_atomic_cmpswap_b32 v2, v[0:1], v[2:3] offset:16 glc
 ; GFX11-WGP-NEXT:    s_waitcnt vmcnt(0)
 ; GFX11-WGP-NEXT:    buffer_gl0_inv
 ; GFX11-WGP-NEXT:    buffer_gl1_inv
 ; GFX11-WGP-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-WGP-NEXT:    flat_store_b32 v[0:1], v2
-; GFX11-WGP-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-WGP-NEXT:    s_endpgm
 ;
 ; GFX11-CU-LABEL: flat_agent_one_as_seq_cst_monotonic_ret_cmpxchg:
 ; GFX11-CU:       ; %bb.0: ; %entry
 ; GFX11-CU-NEXT:    s_load_b128 s[0:3], s[0:1], 0x0
 ; GFX11-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v0, s0 :: v_dual_mov_b32 v1, s1
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v2, s2 :: v_dual_mov_b32 v3, s3
 ; GFX11-CU-NEXT:    s_waitcnt vmcnt(0)
 ; GFX11-CU-NEXT:    s_waitcnt_vscnt null, 0x0
 ; GFX11-CU-NEXT:    flat_atomic_cmpswap_b32 v2, v[0:1], v[2:3] offset:16 glc
 ; GFX11-CU-NEXT:    s_waitcnt vmcnt(0)
 ; GFX11-CU-NEXT:    buffer_gl0_inv
 ; GFX11-CU-NEXT:    buffer_gl1_inv
 ; GFX11-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-CU-NEXT:    flat_store_b32 v[0:1], v2
-; GFX11-CU-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-CU-NEXT:    s_endpgm
     ptr %out, i32 %in, i32 %old) {
 entry:
   %gep = getelementptr i32, ptr %out, i32 4
   %val = cmpxchg volatile ptr %gep, i32 %old, i32 %in syncscope("agent-one-as") seq_cst monotonic
   %val0 = extractvalue { i32, i1 } %val, 0
   store i32 %val0, ptr %out, align 4
   ret void

 ; GFX11-WGP-NEXT:    v_dual_mov_b32 v0, s0 :: v_dual_mov_b32 v1, s1
 ; GFX11-WGP-NEXT:    v_dual_mov_b32 v2, s2 :: v_dual_mov_b32 v3, s3
 ; GFX11-WGP-NEXT:    flat_atomic_cmpswap_b32 v2, v[0:1], v[2:3] offset:16 glc
 ; GFX11-WGP-NEXT:    s_waitcnt vmcnt(0)
 ; GFX11-WGP-NEXT:    buffer_gl0_inv
 ; GFX11-WGP-NEXT:    buffer_gl1_inv
 ; GFX11-WGP-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-WGP-NEXT:    flat_store_b32 v[0:1], v2
-; GFX11-WGP-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-WGP-NEXT:    s_endpgm
 ;
 ; GFX11-CU-LABEL: flat_agent_one_as_monotonic_acquire_ret_cmpxchg:
 ; GFX11-CU:       ; %bb.0: ; %entry
 ; GFX11-CU-NEXT:    s_load_b128 s[0:3], s[0:1], 0x0
 ; GFX11-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v0, s0 :: v_dual_mov_b32 v1, s1
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v2, s2 :: v_dual_mov_b32 v3, s3
 ; GFX11-CU-NEXT:    flat_atomic_cmpswap_b32 v2, v[0:1], v[2:3] offset:16 glc
 ; GFX11-CU-NEXT:    s_waitcnt vmcnt(0)
 ; GFX11-CU-NEXT:    buffer_gl0_inv
 ; GFX11-CU-NEXT:    buffer_gl1_inv
 ; GFX11-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-CU-NEXT:    flat_store_b32 v[0:1], v2
-; GFX11-CU-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-CU-NEXT:    s_endpgm
     ptr %out, i32 %in, i32 %old) {
 entry:
   %gep = getelementptr i32, ptr %out, i32 4
   %val = cmpxchg volatile ptr %gep, i32 %old, i32 %in syncscope("agent-one-as") monotonic acquire
   %val0 = extractvalue { i32, i1 } %val, 0
   store i32 %val0, ptr %out, align 4
   ret void

 ; GFX11-WGP-NEXT:    v_dual_mov_b32 v0, s0 :: v_dual_mov_b32 v1, s1
 ; GFX11-WGP-NEXT:    v_dual_mov_b32 v2, s2 :: v_dual_mov_b32 v3, s3
 ; GFX11-WGP-NEXT:    flat_atomic_cmpswap_b32 v2, v[0:1], v[2:3] offset:16 glc
 ; GFX11-WGP-NEXT:    s_waitcnt vmcnt(0)
 ; GFX11-WGP-NEXT:    buffer_gl0_inv
 ; GFX11-WGP-NEXT:    buffer_gl1_inv
 ; GFX11-WGP-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-WGP-NEXT:    flat_store_b32 v[0:1], v2
-; GFX11-WGP-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-WGP-NEXT:    s_endpgm
 ;
 ; GFX11-CU-LABEL: flat_agent_one_as_acquire_acquire_ret_cmpxchg:
 ; GFX11-CU:       ; %bb.0: ; %entry
 ; GFX11-CU-NEXT:    s_load_b128 s[0:3], s[0:1], 0x0
 ; GFX11-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v0, s0 :: v_dual_mov_b32 v1, s1
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v2, s2 :: v_dual_mov_b32 v3, s3
 ; GFX11-CU-NEXT:    flat_atomic_cmpswap_b32 v2, v[0:1], v[2:3] offset:16 glc
 ; GFX11-CU-NEXT:    s_waitcnt vmcnt(0)
 ; GFX11-CU-NEXT:    buffer_gl0_inv
 ; GFX11-CU-NEXT:    buffer_gl1_inv
 ; GFX11-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-CU-NEXT:    flat_store_b32 v[0:1], v2
-; GFX11-CU-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-CU-NEXT:    s_endpgm
     ptr %out, i32 %in, i32 %old) {
 entry:
   %gep = getelementptr i32, ptr %out, i32 4
   %val = cmpxchg volatile ptr %gep, i32 %old, i32 %in syncscope("agent-one-as") acquire acquire
   %val0 = extractvalue { i32, i1 } %val, 0
   store i32 %val0, ptr %out, align 4
   ret void

 ; GFX11-WGP-NEXT:    s_waitcnt vmcnt(0)
 ; GFX11-WGP-NEXT:    s_waitcnt_vscnt null, 0x0
 ; GFX11-WGP-NEXT:    flat_atomic_cmpswap_b32 v2, v[0:1], v[2:3] offset:16 glc
 ; GFX11-WGP-NEXT:    s_waitcnt vmcnt(0)
 ; GFX11-WGP-NEXT:    buffer_gl0_inv
 ; GFX11-WGP-NEXT:    buffer_gl1_inv
 ; GFX11-WGP-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-WGP-NEXT:    flat_store_b32 v[0:1], v2
-; GFX11-WGP-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-WGP-NEXT:    s_endpgm
 ;
 ; GFX11-CU-LABEL: flat_agent_one_as_release_acquire_ret_cmpxchg:
 ; GFX11-CU:       ; %bb.0: ; %entry
 ; GFX11-CU-NEXT:    s_load_b128 s[0:3], s[0:1], 0x0
 ; GFX11-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v0, s0 :: v_dual_mov_b32 v1, s1
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v2, s2 :: v_dual_mov_b32 v3, s3
 ; GFX11-CU-NEXT:    s_waitcnt vmcnt(0)
 ; GFX11-CU-NEXT:    s_waitcnt_vscnt null, 0x0
 ; GFX11-CU-NEXT:    flat_atomic_cmpswap_b32 v2, v[0:1], v[2:3] offset:16 glc
 ; GFX11-CU-NEXT:    s_waitcnt vmcnt(0)
 ; GFX11-CU-NEXT:    buffer_gl0_inv
 ; GFX11-CU-NEXT:    buffer_gl1_inv
 ; GFX11-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-CU-NEXT:    flat_store_b32 v[0:1], v2
-; GFX11-CU-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-CU-NEXT:    s_endpgm
     ptr %out, i32 %in, i32 %old) {
 entry:
   %gep = getelementptr i32, ptr %out, i32 4
   %val = cmpxchg volatile ptr %gep, i32 %old, i32 %in syncscope("agent-one-as") release acquire
   %val0 = extractvalue { i32, i1 } %val, 0
   store i32 %val0, ptr %out, align 4
   ret void

 ; GFX11-WGP-NEXT:    s_waitcnt vmcnt(0)
 ; GFX11-WGP-NEXT:    s_waitcnt_vscnt null, 0x0
 ; GFX11-WGP-NEXT:    flat_atomic_cmpswap_b32 v2, v[0:1], v[2:3] offset:16 glc
 ; GFX11-WGP-NEXT:    s_waitcnt vmcnt(0)
 ; GFX11-WGP-NEXT:    buffer_gl0_inv
 ; GFX11-WGP-NEXT:    buffer_gl1_inv
 ; GFX11-WGP-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-WGP-NEXT:    flat_store_b32 v[0:1], v2
-; GFX11-WGP-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-WGP-NEXT:    s_endpgm
 ;
 ; GFX11-CU-LABEL: flat_agent_one_as_acq_rel_acquire_ret_cmpxchg:
 ; GFX11-CU:       ; %bb.0: ; %entry
 ; GFX11-CU-NEXT:    s_load_b128 s[0:3], s[0:1], 0x0
 ; GFX11-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v0, s0 :: v_dual_mov_b32 v1, s1
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v2, s2 :: v_dual_mov_b32 v3, s3
 ; GFX11-CU-NEXT:    s_waitcnt vmcnt(0)
 ; GFX11-CU-NEXT:    s_waitcnt_vscnt null, 0x0
 ; GFX11-CU-NEXT:    flat_atomic_cmpswap_b32 v2, v[0:1], v[2:3] offset:16 glc
 ; GFX11-CU-NEXT:    s_waitcnt vmcnt(0)
 ; GFX11-CU-NEXT:    buffer_gl0_inv
 ; GFX11-CU-NEXT:    buffer_gl1_inv
 ; GFX11-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-CU-NEXT:    flat_store_b32 v[0:1], v2
-; GFX11-CU-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-CU-NEXT:    s_endpgm
     ptr %out, i32 %in, i32 %old) {
 entry:
   %gep = getelementptr i32, ptr %out, i32 4
   %val = cmpxchg volatile ptr %gep, i32 %old, i32 %in syncscope("agent-one-as") acq_rel acquire
   %val0 = extractvalue { i32, i1 } %val, 0
   store i32 %val0, ptr %out, align 4
   ret void

 ; GFX11-WGP-NEXT:    s_waitcnt vmcnt(0)
 ; GFX11-WGP-NEXT:    s_waitcnt_vscnt null, 0x0
 ; GFX11-WGP-NEXT:    flat_atomic_cmpswap_b32 v2, v[0:1], v[2:3] offset:16 glc
 ; GFX11-WGP-NEXT:    s_waitcnt vmcnt(0)
 ; GFX11-WGP-NEXT:    buffer_gl0_inv
 ; GFX11-WGP-NEXT:    buffer_gl1_inv
 ; GFX11-WGP-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-WGP-NEXT:    flat_store_b32 v[0:1], v2
-; GFX11-WGP-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-WGP-NEXT:    s_endpgm
 ;
 ; GFX11-CU-LABEL: flat_agent_one_as_seq_cst_acquire_ret_cmpxchg:
 ; GFX11-CU:       ; %bb.0: ; %entry
 ; GFX11-CU-NEXT:    s_load_b128 s[0:3], s[0:1], 0x0
 ; GFX11-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v0, s0 :: v_dual_mov_b32 v1, s1
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v2, s2 :: v_dual_mov_b32 v3, s3
 ; GFX11-CU-NEXT:    s_waitcnt vmcnt(0)
 ; GFX11-CU-NEXT:    s_waitcnt_vscnt null, 0x0
 ; GFX11-CU-NEXT:    flat_atomic_cmpswap_b32 v2, v[0:1], v[2:3] offset:16 glc
 ; GFX11-CU-NEXT:    s_waitcnt vmcnt(0)
 ; GFX11-CU-NEXT:    buffer_gl0_inv
 ; GFX11-CU-NEXT:    buffer_gl1_inv
 ; GFX11-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-CU-NEXT:    flat_store_b32 v[0:1], v2
-; GFX11-CU-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-CU-NEXT:    s_endpgm
     ptr %out, i32 %in, i32 %old) {
 entry:
   %gep = getelementptr i32, ptr %out, i32 4
   %val = cmpxchg volatile ptr %gep, i32 %old, i32 %in syncscope("agent-one-as") seq_cst acquire
   %val0 = extractvalue { i32, i1 } %val, 0
   store i32 %val0, ptr %out, align 4
   ret void

 ; GFX11-WGP-NEXT:    s_waitcnt vmcnt(0)
 ; GFX11-WGP-NEXT:    s_waitcnt_vscnt null, 0x0
 ; GFX11-WGP-NEXT:    flat_atomic_cmpswap_b32 v2, v[0:1], v[2:3] offset:16 glc
 ; GFX11-WGP-NEXT:    s_waitcnt vmcnt(0)
 ; GFX11-WGP-NEXT:    buffer_gl0_inv
 ; GFX11-WGP-NEXT:    buffer_gl1_inv
 ; GFX11-WGP-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-WGP-NEXT:    flat_store_b32 v[0:1], v2
-; GFX11-WGP-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-WGP-NEXT:    s_endpgm
 ;
 ; GFX11-CU-LABEL: flat_agent_one_as_monotonic_seq_cst_ret_cmpxchg:
 ; GFX11-CU:       ; %bb.0: ; %entry
 ; GFX11-CU-NEXT:    s_load_b128 s[0:3], s[0:1], 0x0
 ; GFX11-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v0, s0 :: v_dual_mov_b32 v1, s1
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v2, s2 :: v_dual_mov_b32 v3, s3
 ; GFX11-CU-NEXT:    s_waitcnt vmcnt(0)
 ; GFX11-CU-NEXT:    s_waitcnt_vscnt null, 0x0
 ; GFX11-CU-NEXT:    flat_atomic_cmpswap_b32 v2, v[0:1], v[2:3] offset:16 glc
 ; GFX11-CU-NEXT:    s_waitcnt vmcnt(0)
 ; GFX11-CU-NEXT:    buffer_gl0_inv
 ; GFX11-CU-NEXT:    buffer_gl1_inv
 ; GFX11-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-CU-NEXT:    flat_store_b32 v[0:1], v2
-; GFX11-CU-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-CU-NEXT:    s_endpgm
     ptr %out, i32 %in, i32 %old) {
 entry:
   %gep = getelementptr i32, ptr %out, i32 4
   %val = cmpxchg volatile ptr %gep, i32 %old, i32 %in syncscope("agent-one-as") monotonic seq_cst
   %val0 = extractvalue { i32, i1 } %val, 0
   store i32 %val0, ptr %out, align 4
   ret void

 ; GFX11-WGP-NEXT:    s_waitcnt vmcnt(0)
 ; GFX11-WGP-NEXT:    s_waitcnt_vscnt null, 0x0
 ; GFX11-WGP-NEXT:    flat_atomic_cmpswap_b32 v2, v[0:1], v[2:3] offset:16 glc
 ; GFX11-WGP-NEXT:    s_waitcnt vmcnt(0)
 ; GFX11-WGP-NEXT:    buffer_gl0_inv
 ; GFX11-WGP-NEXT:    buffer_gl1_inv
 ; GFX11-WGP-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-WGP-NEXT:    flat_store_b32 v[0:1], v2
-; GFX11-WGP-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-WGP-NEXT:    s_endpgm
 ;
 ; GFX11-CU-LABEL: flat_agent_one_as_acquire_seq_cst_ret_cmpxchg:
 ; GFX11-CU:       ; %bb.0: ; %entry
 ; GFX11-CU-NEXT:    s_load_b128 s[0:3], s[0:1], 0x0
 ; GFX11-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v0, s0 :: v_dual_mov_b32 v1, s1
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v2, s2 :: v_dual_mov_b32 v3, s3
 ; GFX11-CU-NEXT:    s_waitcnt vmcnt(0)
 ; GFX11-CU-NEXT:    s_waitcnt_vscnt null, 0x0
 ; GFX11-CU-NEXT:    flat_atomic_cmpswap_b32 v2, v[0:1], v[2:3] offset:16 glc
 ; GFX11-CU-NEXT:    s_waitcnt vmcnt(0)
 ; GFX11-CU-NEXT:    buffer_gl0_inv
 ; GFX11-CU-NEXT:    buffer_gl1_inv
 ; GFX11-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-CU-NEXT:    flat_store_b32 v[0:1], v2
-; GFX11-CU-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-CU-NEXT:    s_endpgm
     ptr %out, i32 %in, i32 %old) {
 entry:
   %gep = getelementptr i32, ptr %out, i32 4
   %val = cmpxchg volatile ptr %gep, i32 %old, i32 %in syncscope("agent-one-as") acquire seq_cst
   %val0 = extractvalue { i32, i1 } %val, 0
   store i32 %val0, ptr %out, align 4
   ret void

 ; GFX11-WGP-NEXT:    s_waitcnt vmcnt(0)
 ; GFX11-WGP-NEXT:    s_waitcnt_vscnt null, 0x0
 ; GFX11-WGP-NEXT:    flat_atomic_cmpswap_b32 v2, v[0:1], v[2:3] offset:16 glc
 ; GFX11-WGP-NEXT:    s_waitcnt vmcnt(0)
 ; GFX11-WGP-NEXT:    buffer_gl0_inv
 ; GFX11-WGP-NEXT:    buffer_gl1_inv
 ; GFX11-WGP-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-WGP-NEXT:    flat_store_b32 v[0:1], v2
-; GFX11-WGP-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-WGP-NEXT:    s_endpgm
 ;
 ; GFX11-CU-LABEL: flat_agent_one_as_release_seq_cst_ret_cmpxchg:
 ; GFX11-CU:       ; %bb.0: ; %entry
 ; GFX11-CU-NEXT:    s_load_b128 s[0:3], s[0:1], 0x0
 ; GFX11-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v0, s0 :: v_dual_mov_b32 v1, s1
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v2, s2 :: v_dual_mov_b32 v3, s3
 ; GFX11-CU-NEXT:    s_waitcnt vmcnt(0)
 ; GFX11-CU-NEXT:    s_waitcnt_vscnt null, 0x0
 ; GFX11-CU-NEXT:    flat_atomic_cmpswap_b32 v2, v[0:1], v[2:3] offset:16 glc
 ; GFX11-CU-NEXT:    s_waitcnt vmcnt(0)
 ; GFX11-CU-NEXT:    buffer_gl0_inv
 ; GFX11-CU-NEXT:    buffer_gl1_inv
 ; GFX11-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-CU-NEXT:    flat_store_b32 v[0:1], v2
-; GFX11-CU-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-CU-NEXT:    s_endpgm
     ptr %out, i32 %in, i32 %old) {
 entry:
   %gep = getelementptr i32, ptr %out, i32 4
   %val = cmpxchg volatile ptr %gep, i32 %old, i32 %in syncscope("agent-one-as") release seq_cst
   %val0 = extractvalue { i32, i1 } %val, 0
   store i32 %val0, ptr %out, align 4
   ret void

 ; GFX11-WGP-NEXT:    s_waitcnt vmcnt(0)
 ; GFX11-WGP-NEXT:    s_waitcnt_vscnt null, 0x0
 ; GFX11-WGP-NEXT:    flat_atomic_cmpswap_b32 v2, v[0:1], v[2:3] offset:16 glc
 ; GFX11-WGP-NEXT:    s_waitcnt vmcnt(0)
 ; GFX11-WGP-NEXT:    buffer_gl0_inv
 ; GFX11-WGP-NEXT:    buffer_gl1_inv
 ; GFX11-WGP-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-WGP-NEXT:    flat_store_b32 v[0:1], v2
-; GFX11-WGP-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-WGP-NEXT:    s_endpgm
 ;
 ; GFX11-CU-LABEL: flat_agent_one_as_acq_rel_seq_cst_ret_cmpxchg:
 ; GFX11-CU:       ; %bb.0: ; %entry
 ; GFX11-CU-NEXT:    s_load_b128 s[0:3], s[0:1], 0x0
 ; GFX11-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v0, s0 :: v_dual_mov_b32 v1, s1
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v2, s2 :: v_dual_mov_b32 v3, s3
 ; GFX11-CU-NEXT:    s_waitcnt vmcnt(0)
 ; GFX11-CU-NEXT:    s_waitcnt_vscnt null, 0x0
 ; GFX11-CU-NEXT:    flat_atomic_cmpswap_b32 v2, v[0:1], v[2:3] offset:16 glc
 ; GFX11-CU-NEXT:    s_waitcnt vmcnt(0)
 ; GFX11-CU-NEXT:    buffer_gl0_inv
 ; GFX11-CU-NEXT:    buffer_gl1_inv
 ; GFX11-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-CU-NEXT:    flat_store_b32 v[0:1], v2
-; GFX11-CU-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-CU-NEXT:    s_endpgm
     ptr %out, i32 %in, i32 %old) {
 entry:
   %gep = getelementptr i32, ptr %out, i32 4
   %val = cmpxchg volatile ptr %gep, i32 %old, i32 %in syncscope("agent-one-as") acq_rel seq_cst
   %val0 = extractvalue { i32, i1 } %val, 0
   store i32 %val0, ptr %out, align 4
   ret void

 ; GFX11-WGP-NEXT:    s_waitcnt vmcnt(0)
 ; GFX11-WGP-NEXT:    s_waitcnt_vscnt null, 0x0
 ; GFX11-WGP-NEXT:    flat_atomic_cmpswap_b32 v2, v[0:1], v[2:3] offset:16 glc
 ; GFX11-WGP-NEXT:    s_waitcnt vmcnt(0)
 ; GFX11-WGP-NEXT:    buffer_gl0_inv
 ; GFX11-WGP-NEXT:    buffer_gl1_inv
 ; GFX11-WGP-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-WGP-NEXT:    flat_store_b32 v[0:1], v2
-; GFX11-WGP-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-WGP-NEXT:    s_endpgm
 ;
 ; GFX11-CU-LABEL: flat_agent_one_as_seq_cst_seq_cst_ret_cmpxchg:
 ; GFX11-CU:       ; %bb.0: ; %entry
 ; GFX11-CU-NEXT:    s_load_b128 s[0:3], s[0:1], 0x0
 ; GFX11-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v0, s0 :: v_dual_mov_b32 v1, s1
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v2, s2 :: v_dual_mov_b32 v3, s3
 ; GFX11-CU-NEXT:    s_waitcnt vmcnt(0)
 ; GFX11-CU-NEXT:    s_waitcnt_vscnt null, 0x0
 ; GFX11-CU-NEXT:    flat_atomic_cmpswap_b32 v2, v[0:1], v[2:3] offset:16 glc
 ; GFX11-CU-NEXT:    s_waitcnt vmcnt(0)
 ; GFX11-CU-NEXT:    buffer_gl0_inv
 ; GFX11-CU-NEXT:    buffer_gl1_inv
 ; GFX11-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-CU-NEXT:    flat_store_b32 v[0:1], v2
-; GFX11-CU-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-CU-NEXT:    s_endpgm
     ptr %out, i32 %in, i32 %old) {
 entry:
   %gep = getelementptr i32, ptr %out, i32 4
   %val = cmpxchg volatile ptr %gep, i32 %old, i32 %in syncscope("agent-one-as") seq_cst seq_cst
   %val0 = extractvalue { i32, i1 } %val, 0
   store i32 %val0, ptr %out, align 4
   ret void
 }
 

llvm/test/CodeGen/AMDGPU/memory-legalizer-flat-nontemporal.ll

 ; GFX11-WGP:       ; %bb.0: ; %entry
 ; GFX11-WGP-NEXT:    s_load_b128 s[0:3], s[0:1], 0x0
 ; GFX11-WGP-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-WGP-NEXT:    v_dual_mov_b32 v0, s0 :: v_dual_mov_b32 v1, s1
 ; GFX11-WGP-NEXT:    flat_load_b32 v2, v[0:1] slc dlc
 ; GFX11-WGP-NEXT:    v_dual_mov_b32 v0, s2 :: v_dual_mov_b32 v1, s3
 ; GFX11-WGP-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX11-WGP-NEXT:    flat_store_b32 v[0:1], v2
-; GFX11-WGP-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-WGP-NEXT:    s_endpgm
 ;
 ; GFX11-CU-LABEL: flat_nontemporal_load_0:
 ; GFX11-CU:       ; %bb.0: ; %entry
 ; GFX11-CU-NEXT:    s_load_b128 s[0:3], s[0:1], 0x0
 ; GFX11-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v0, s0 :: v_dual_mov_b32 v1, s1
 ; GFX11-CU-NEXT:    flat_load_b32 v2, v[0:1] slc dlc
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v0, s2 :: v_dual_mov_b32 v1, s3
 ; GFX11-CU-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX11-CU-NEXT:    flat_store_b32 v[0:1], v2
-; GFX11-CU-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-CU-NEXT:    s_endpgm
     ptr %in, ptr %out) {
 entry:
   %val = load i32, ptr %in, align 4, !nontemporal !0
   store i32 %val, ptr %out
   ret void
 }
 

 ; GFX11-WGP-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-WGP-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
 ; GFX11-WGP-NEXT:    v_add_co_u32 v0, s0, s0, v0
 ; GFX11-WGP-NEXT:    v_add_co_ci_u32_e64 v1, null, s1, 0, s0
 ; GFX11-WGP-NEXT:    flat_load_b32 v2, v[0:1] slc dlc
 ; GFX11-WGP-NEXT:    v_dual_mov_b32 v1, s3 :: v_dual_mov_b32 v0, s2
 ; GFX11-WGP-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX11-WGP-NEXT:    flat_store_b32 v[0:1], v2
-; GFX11-WGP-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-WGP-NEXT:    s_endpgm
 ;
 ; GFX11-CU-LABEL: flat_nontemporal_load_1:
 ; GFX11-CU:       ; %bb.0: ; %entry
 ; GFX11-CU-NEXT:    s_load_b128 s[0:3], s[0:1], 0x0
 ; GFX11-CU-NEXT:    v_lshlrev_b32_e32 v0, 2, v0
 ; GFX11-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-CU-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
 ; GFX11-CU-NEXT:    v_add_co_u32 v0, s0, s0, v0
 ; GFX11-CU-NEXT:    v_add_co_ci_u32_e64 v1, null, s1, 0, s0
 ; GFX11-CU-NEXT:    flat_load_b32 v2, v[0:1] slc dlc
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v1, s3 :: v_dual_mov_b32 v0, s2
 ; GFX11-CU-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX11-CU-NEXT:    flat_store_b32 v[0:1], v2
-; GFX11-CU-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-CU-NEXT:    s_endpgm
     ptr %in, ptr %out) {
 entry:
   %tid = call i32 @llvm.amdgcn.workitem.id.x()
   %val.gep = getelementptr inbounds i32, ptr %in, i32 %tid
   %val = load i32, ptr %val.gep, align 4, !nontemporal !0
   store i32 %val, ptr %out
   ret void

 ; GFX11-WGP:       ; %bb.0: ; %entry
 ; GFX11-WGP-NEXT:    s_load_b128 s[0:3], s[0:1], 0x0
 ; GFX11-WGP-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-WGP-NEXT:    v_dual_mov_b32 v0, s0 :: v_dual_mov_b32 v1, s1
 ; GFX11-WGP-NEXT:    flat_load_b32 v2, v[0:1]
 ; GFX11-WGP-NEXT:    v_dual_mov_b32 v0, s2 :: v_dual_mov_b32 v1, s3
 ; GFX11-WGP-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX11-WGP-NEXT:    flat_store_b32 v[0:1], v2 glc slc dlc
-; GFX11-WGP-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-WGP-NEXT:    s_endpgm
 ;
 ; GFX11-CU-LABEL: flat_nontemporal_store_0:
 ; GFX11-CU:       ; %bb.0: ; %entry
 ; GFX11-CU-NEXT:    s_load_b128 s[0:3], s[0:1], 0x0
 ; GFX11-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v0, s0 :: v_dual_mov_b32 v1, s1
 ; GFX11-CU-NEXT:    flat_load_b32 v2, v[0:1]
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v0, s2 :: v_dual_mov_b32 v1, s3
 ; GFX11-CU-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX11-CU-NEXT:    flat_store_b32 v[0:1], v2 glc slc dlc
-; GFX11-CU-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-CU-NEXT:    s_endpgm
     ptr %in, ptr %out) {
 entry:
   %val = load i32, ptr %in, align 4
   store i32 %val, ptr %out, !nontemporal !0
   ret void
 }
 

 ; GFX11-WGP-NEXT:    v_dual_mov_b32 v1, s0 :: v_dual_lshlrev_b32 v0, 2, v0
 ; GFX11-WGP-NEXT:    v_mov_b32_e32 v2, s1
 ; GFX11-WGP-NEXT:    s_delay_alu instid0(VALU_DEP_2)
 ; GFX11-WGP-NEXT:    v_add_co_u32 v0, s0, s2, v0
 ; GFX11-WGP-NEXT:    flat_load_b32 v2, v[1:2]
 ; GFX11-WGP-NEXT:    v_add_co_ci_u32_e64 v1, null, s3, 0, s0
 ; GFX11-WGP-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX11-WGP-NEXT:    flat_store_b32 v[0:1], v2 glc slc dlc
-; GFX11-WGP-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-WGP-NEXT:    s_endpgm
 ;
 ; GFX11-CU-LABEL: flat_nontemporal_store_1:
 ; GFX11-CU:       ; %bb.0: ; %entry
 ; GFX11-CU-NEXT:    s_load_b128 s[0:3], s[0:1], 0x0
 ; GFX11-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v1, s0 :: v_dual_lshlrev_b32 v0, 2, v0
 ; GFX11-CU-NEXT:    v_mov_b32_e32 v2, s1
 ; GFX11-CU-NEXT:    s_delay_alu instid0(VALU_DEP_2)
 ; GFX11-CU-NEXT:    v_add_co_u32 v0, s0, s2, v0
 ; GFX11-CU-NEXT:    flat_load_b32 v2, v[1:2]
 ; GFX11-CU-NEXT:    v_add_co_ci_u32_e64 v1, null, s3, 0, s0
 ; GFX11-CU-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX11-CU-NEXT:    flat_store_b32 v[0:1], v2 glc slc dlc
-; GFX11-CU-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-CU-NEXT:    s_endpgm
     ptr %in, ptr %out) {
 entry:
   %tid = call i32 @llvm.amdgcn.workitem.id.x()
   %val = load i32, ptr %in, align 4
   %out.gep = getelementptr inbounds i32, ptr %out, i32 %tid
   store i32 %val, ptr %out.gep, !nontemporal !0
   ret void
 }
 
 !0 = !{i32 1}
 declare i32 @llvm.amdgcn.workitem.id.x()

llvm/test/CodeGen/AMDGPU/memory-legalizer-flat-singlethread.ll

 ; GFX11-WGP:       ; %bb.0: ; %entry
 ; GFX11-WGP-NEXT:    s_load_b128 s[0:3], s[0:1], 0x0
 ; GFX11-WGP-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-WGP-NEXT:    v_dual_mov_b32 v0, s0 :: v_dual_mov_b32 v1, s1
 ; GFX11-WGP-NEXT:    flat_load_b32 v2, v[0:1]
 ; GFX11-WGP-NEXT:    v_dual_mov_b32 v0, s2 :: v_dual_mov_b32 v1, s3
 ; GFX11-WGP-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX11-WGP-NEXT:    flat_store_b32 v[0:1], v2
-; GFX11-WGP-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-WGP-NEXT:    s_endpgm
 ;
 ; GFX11-CU-LABEL: flat_singlethread_unordered_load:
 ; GFX11-CU:       ; %bb.0: ; %entry
 ; GFX11-CU-NEXT:    s_load_b128 s[0:3], s[0:1], 0x0
 ; GFX11-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v0, s0 :: v_dual_mov_b32 v1, s1
 ; GFX11-CU-NEXT:    flat_load_b32 v2, v[0:1]
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v0, s2 :: v_dual_mov_b32 v1, s3
 ; GFX11-CU-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX11-CU-NEXT:    flat_store_b32 v[0:1], v2
-; GFX11-CU-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-CU-NEXT:    s_endpgm
     ptr %in, ptr %out) {
 entry:
   %val = load atomic i32, ptr %in syncscope("singlethread") unordered, align 4
   store i32 %val, ptr %out
   ret void
 }
 

 ; GFX11-WGP:       ; %bb.0: ; %entry
 ; GFX11-WGP-NEXT:    s_load_b128 s[0:3], s[0:1], 0x0
 ; GFX11-WGP-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-WGP-NEXT:    v_dual_mov_b32 v0, s0 :: v_dual_mov_b32 v1, s1
 ; GFX11-WGP-NEXT:    flat_load_b32 v2, v[0:1]
 ; GFX11-WGP-NEXT:    v_dual_mov_b32 v0, s2 :: v_dual_mov_b32 v1, s3
 ; GFX11-WGP-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX11-WGP-NEXT:    flat_store_b32 v[0:1], v2
-; GFX11-WGP-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-WGP-NEXT:    s_endpgm
 ;
 ; GFX11-CU-LABEL: flat_singlethread_monotonic_load:
 ; GFX11-CU:       ; %bb.0: ; %entry
 ; GFX11-CU-NEXT:    s_load_b128 s[0:3], s[0:1], 0x0
 ; GFX11-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v0, s0 :: v_dual_mov_b32 v1, s1
 ; GFX11-CU-NEXT:    flat_load_b32 v2, v[0:1]
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v0, s2 :: v_dual_mov_b32 v1, s3
 ; GFX11-CU-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX11-CU-NEXT:    flat_store_b32 v[0:1], v2
-; GFX11-CU-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-CU-NEXT:    s_endpgm
     ptr %in, ptr %out) {
 entry:
   %val = load atomic i32, ptr %in syncscope("singlethread") monotonic, align 4
   store i32 %val, ptr %out
   ret void
 }
 

 ; GFX11-WGP:       ; %bb.0: ; %entry
 ; GFX11-WGP-NEXT:    s_load_b128 s[0:3], s[0:1], 0x0
 ; GFX11-WGP-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-WGP-NEXT:    v_dual_mov_b32 v0, s0 :: v_dual_mov_b32 v1, s1
 ; GFX11-WGP-NEXT:    flat_load_b32 v2, v[0:1]
 ; GFX11-WGP-NEXT:    v_dual_mov_b32 v0, s2 :: v_dual_mov_b32 v1, s3
 ; GFX11-WGP-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX11-WGP-NEXT:    flat_store_b32 v[0:1], v2
-; GFX11-WGP-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-WGP-NEXT:    s_endpgm
 ;
 ; GFX11-CU-LABEL: flat_singlethread_acquire_load:
 ; GFX11-CU:       ; %bb.0: ; %entry
 ; GFX11-CU-NEXT:    s_load_b128 s[0:3], s[0:1], 0x0
 ; GFX11-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v0, s0 :: v_dual_mov_b32 v1, s1
 ; GFX11-CU-NEXT:    flat_load_b32 v2, v[0:1]
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v0, s2 :: v_dual_mov_b32 v1, s3
 ; GFX11-CU-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX11-CU-NEXT:    flat_store_b32 v[0:1], v2
-; GFX11-CU-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-CU-NEXT:    s_endpgm
     ptr %in, ptr %out) {
 entry:
   %val = load atomic i32, ptr %in syncscope("singlethread") acquire, align 4
   store i32 %val, ptr %out
   ret void
 }
 

 ; GFX11-WGP:       ; %bb.0: ; %entry
 ; GFX11-WGP-NEXT:    s_load_b128 s[0:3], s[0:1], 0x0
 ; GFX11-WGP-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-WGP-NEXT:    v_dual_mov_b32 v0, s0 :: v_dual_mov_b32 v1, s1
 ; GFX11-WGP-NEXT:    flat_load_b32 v2, v[0:1]
 ; GFX11-WGP-NEXT:    v_dual_mov_b32 v0, s2 :: v_dual_mov_b32 v1, s3
 ; GFX11-WGP-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX11-WGP-NEXT:    flat_store_b32 v[0:1], v2
-; GFX11-WGP-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-WGP-NEXT:    s_endpgm
 ;
 ; GFX11-CU-LABEL: flat_singlethread_seq_cst_load:
 ; GFX11-CU:       ; %bb.0: ; %entry
 ; GFX11-CU-NEXT:    s_load_b128 s[0:3], s[0:1], 0x0
 ; GFX11-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v0, s0 :: v_dual_mov_b32 v1, s1
 ; GFX11-CU-NEXT:    flat_load_b32 v2, v[0:1]
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v0, s2 :: v_dual_mov_b32 v1, s3
 ; GFX11-CU-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX11-CU-NEXT:    flat_store_b32 v[0:1], v2
-; GFX11-CU-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-CU-NEXT:    s_endpgm
     ptr %in, ptr %out) {
 entry:
   %val = load atomic i32, ptr %in syncscope("singlethread") seq_cst, align 4
   store i32 %val, ptr %out
   ret void
 }
 

 ; GFX11-WGP:       ; %bb.0: ; %entry
 ; GFX11-WGP-NEXT:    s_clause 0x1
 ; GFX11-WGP-NEXT:    s_load_b64 s[2:3], s[0:1], 0x8
 ; GFX11-WGP-NEXT:    s_load_b32 s0, s[0:1], 0x0
 ; GFX11-WGP-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-WGP-NEXT:    v_dual_mov_b32 v0, s2 :: v_dual_mov_b32 v1, s3
 ; GFX11-WGP-NEXT:    v_mov_b32_e32 v2, s0
 ; GFX11-WGP-NEXT:    flat_store_b32 v[0:1], v2
-; GFX11-WGP-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-WGP-NEXT:    s_endpgm
 ;
 ; GFX11-CU-LABEL: flat_singlethread_unordered_store:
 ; GFX11-CU:       ; %bb.0: ; %entry
 ; GFX11-CU-NEXT:    s_clause 0x1
 ; GFX11-CU-NEXT:    s_load_b64 s[2:3], s[0:1], 0x8
 ; GFX11-CU-NEXT:    s_load_b32 s0, s[0:1], 0x0
 ; GFX11-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v0, s2 :: v_dual_mov_b32 v1, s3
 ; GFX11-CU-NEXT:    v_mov_b32_e32 v2, s0
 ; GFX11-CU-NEXT:    flat_store_b32 v[0:1], v2
-; GFX11-CU-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-CU-NEXT:    s_endpgm
     i32 %in, ptr %out) {
 entry:
   store atomic i32 %in, ptr %out syncscope("singlethread") unordered, align 4
   ret void
 }
 
 define amdgpu_kernel void @flat_singlethread_monotonic_store(

 ; GFX11-WGP:       ; %bb.0: ; %entry
 ; GFX11-WGP-NEXT:    s_clause 0x1
 ; GFX11-WGP-NEXT:    s_load_b64 s[2:3], s[0:1], 0x8
 ; GFX11-WGP-NEXT:    s_load_b32 s0, s[0:1], 0x0
 ; GFX11-WGP-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-WGP-NEXT:    v_dual_mov_b32 v0, s2 :: v_dual_mov_b32 v1, s3
 ; GFX11-WGP-NEXT:    v_mov_b32_e32 v2, s0
 ; GFX11-WGP-NEXT:    flat_store_b32 v[0:1], v2
-; GFX11-WGP-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-WGP-NEXT:    s_endpgm
 ;
 ; GFX11-CU-LABEL: flat_singlethread_monotonic_store:
 ; GFX11-CU:       ; %bb.0: ; %entry
 ; GFX11-CU-NEXT:    s_clause 0x1
 ; GFX11-CU-NEXT:    s_load_b64 s[2:3], s[0:1], 0x8
 ; GFX11-CU-NEXT:    s_load_b32 s0, s[0:1], 0x0
 ; GFX11-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v0, s2 :: v_dual_mov_b32 v1, s3
 ; GFX11-CU-NEXT:    v_mov_b32_e32 v2, s0
 ; GFX11-CU-NEXT:    flat_store_b32 v[0:1], v2
-; GFX11-CU-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-CU-NEXT:    s_endpgm
     i32 %in, ptr %out) {
 entry:
   store atomic i32 %in, ptr %out syncscope("singlethread") monotonic, align 4
   ret void
 }
 
 define amdgpu_kernel void @flat_singlethread_release_store(

 ; GFX11-WGP:       ; %bb.0: ; %entry
 ; GFX11-WGP-NEXT:    s_clause 0x1
 ; GFX11-WGP-NEXT:    s_load_b64 s[2:3], s[0:1], 0x8
 ; GFX11-WGP-NEXT:    s_load_b32 s0, s[0:1], 0x0
 ; GFX11-WGP-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-WGP-NEXT:    v_dual_mov_b32 v0, s2 :: v_dual_mov_b32 v1, s3
 ; GFX11-WGP-NEXT:    v_mov_b32_e32 v2, s0
 ; GFX11-WGP-NEXT:    flat_store_b32 v[0:1], v2
-; GFX11-WGP-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-WGP-NEXT:    s_endpgm
 ;
 ; GFX11-CU-LABEL: flat_singlethread_release_store:
 ; GFX11-CU:       ; %bb.0: ; %entry
 ; GFX11-CU-NEXT:    s_clause 0x1
 ; GFX11-CU-NEXT:    s_load_b64 s[2:3], s[0:1], 0x8
 ; GFX11-CU-NEXT:    s_load_b32 s0, s[0:1], 0x0
 ; GFX11-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v0, s2 :: v_dual_mov_b32 v1, s3
 ; GFX11-CU-NEXT:    v_mov_b32_e32 v2, s0
 ; GFX11-CU-NEXT:    flat_store_b32 v[0:1], v2
-; GFX11-CU-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-CU-NEXT:    s_endpgm
     i32 %in, ptr %out) {
 entry:
   store atomic i32 %in, ptr %out syncscope("singlethread") release, align 4
   ret void
 }
 
 define amdgpu_kernel void @flat_singlethread_seq_cst_store(

 ; GFX11-WGP:       ; %bb.0: ; %entry
 ; GFX11-WGP-NEXT:    s_clause 0x1
 ; GFX11-WGP-NEXT:    s_load_b64 s[2:3], s[0:1], 0x8
 ; GFX11-WGP-NEXT:    s_load_b32 s0, s[0:1], 0x0
 ; GFX11-WGP-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-WGP-NEXT:    v_dual_mov_b32 v0, s2 :: v_dual_mov_b32 v1, s3
 ; GFX11-WGP-NEXT:    v_mov_b32_e32 v2, s0
 ; GFX11-WGP-NEXT:    flat_store_b32 v[0:1], v2
-; GFX11-WGP-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-WGP-NEXT:    s_endpgm
 ;
 ; GFX11-CU-LABEL: flat_singlethread_seq_cst_store:
 ; GFX11-CU:       ; %bb.0: ; %entry
 ; GFX11-CU-NEXT:    s_clause 0x1
 ; GFX11-CU-NEXT:    s_load_b64 s[2:3], s[0:1], 0x8
 ; GFX11-CU-NEXT:    s_load_b32 s0, s[0:1], 0x0
 ; GFX11-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v0, s2 :: v_dual_mov_b32 v1, s3
 ; GFX11-CU-NEXT:    v_mov_b32_e32 v2, s0
 ; GFX11-CU-NEXT:    flat_store_b32 v[0:1], v2
-; GFX11-CU-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-CU-NEXT:    s_endpgm
     i32 %in, ptr %out) {
 entry:
   store atomic i32 %in, ptr %out syncscope("singlethread") seq_cst, align 4
   ret void
 }
 
 define amdgpu_kernel void @flat_singlethread_monotonic_atomicrmw(

 ; GFX11-WGP:       ; %bb.0: ; %entry
 ; GFX11-WGP-NEXT:    s_clause 0x1
 ; GFX11-WGP-NEXT:    s_load_b64 s[2:3], s[0:1], 0x0
 ; GFX11-WGP-NEXT:    s_load_b32 s0, s[0:1], 0x8
 ; GFX11-WGP-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-WGP-NEXT:    v_dual_mov_b32 v0, s2 :: v_dual_mov_b32 v1, s3
 ; GFX11-WGP-NEXT:    v_mov_b32_e32 v2, s0
 ; GFX11-WGP-NEXT:    flat_atomic_swap_b32 v[0:1], v2
-; GFX11-WGP-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-WGP-NEXT:    s_endpgm
 ;
 ; GFX11-CU-LABEL: flat_singlethread_monotonic_atomicrmw:
 ; GFX11-CU:       ; %bb.0: ; %entry
 ; GFX11-CU-NEXT:    s_clause 0x1
 ; GFX11-CU-NEXT:    s_load_b64 s[2:3], s[0:1], 0x0
 ; GFX11-CU-NEXT:    s_load_b32 s0, s[0:1], 0x8
 ; GFX11-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v0, s2 :: v_dual_mov_b32 v1, s3
 ; GFX11-CU-NEXT:    v_mov_b32_e32 v2, s0
 ; GFX11-CU-NEXT:    flat_atomic_swap_b32 v[0:1], v2
-; GFX11-CU-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-CU-NEXT:    s_endpgm
     ptr %out, i32 %in) {
 entry:
   %val = atomicrmw volatile xchg ptr %out, i32 %in syncscope("singlethread") monotonic
   ret void
 }
 
 define amdgpu_kernel void @flat_singlethread_acquire_atomicrmw(

 ; GFX11-WGP:       ; %bb.0: ; %entry
 ; GFX11-WGP-NEXT:    s_clause 0x1
 ; GFX11-WGP-NEXT:    s_load_b64 s[2:3], s[0:1], 0x0
 ; GFX11-WGP-NEXT:    s_load_b32 s0, s[0:1], 0x8
 ; GFX11-WGP-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-WGP-NEXT:    v_dual_mov_b32 v0, s2 :: v_dual_mov_b32 v1, s3
 ; GFX11-WGP-NEXT:    v_mov_b32_e32 v2, s0
 ; GFX11-WGP-NEXT:    flat_atomic_swap_b32 v[0:1], v2
-; GFX11-WGP-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-WGP-NEXT:    s_endpgm
 ;
 ; GFX11-CU-LABEL: flat_singlethread_acquire_atomicrmw:
 ; GFX11-CU:       ; %bb.0: ; %entry
 ; GFX11-CU-NEXT:    s_clause 0x1
 ; GFX11-CU-NEXT:    s_load_b64 s[2:3], s[0:1], 0x0
 ; GFX11-CU-NEXT:    s_load_b32 s0, s[0:1], 0x8
 ; GFX11-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v0, s2 :: v_dual_mov_b32 v1, s3
 ; GFX11-CU-NEXT:    v_mov_b32_e32 v2, s0
 ; GFX11-CU-NEXT:    flat_atomic_swap_b32 v[0:1], v2
-; GFX11-CU-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-CU-NEXT:    s_endpgm
     ptr %out, i32 %in) {
 entry:
   %val = atomicrmw volatile xchg ptr %out, i32 %in syncscope("singlethread") acquire
   ret void
 }
 
 define amdgpu_kernel void @flat_singlethread_release_atomicrmw(

 ; GFX11-WGP:       ; %bb.0: ; %entry
 ; GFX11-WGP-NEXT:    s_clause 0x1
 ; GFX11-WGP-NEXT:    s_load_b64 s[2:3], s[0:1], 0x0
 ; GFX11-WGP-NEXT:    s_load_b32 s0, s[0:1], 0x8
 ; GFX11-WGP-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-WGP-NEXT:    v_dual_mov_b32 v0, s2 :: v_dual_mov_b32 v1, s3
 ; GFX11-WGP-NEXT:    v_mov_b32_e32 v2, s0
 ; GFX11-WGP-NEXT:    flat_atomic_swap_b32 v[0:1], v2
-; GFX11-WGP-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-WGP-NEXT:    s_endpgm
 ;
 ; GFX11-CU-LABEL: flat_singlethread_release_atomicrmw:
 ; GFX11-CU:       ; %bb.0: ; %entry
 ; GFX11-CU-NEXT:    s_clause 0x1
 ; GFX11-CU-NEXT:    s_load_b64 s[2:3], s[0:1], 0x0
 ; GFX11-CU-NEXT:    s_load_b32 s0, s[0:1], 0x8
 ; GFX11-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v0, s2 :: v_dual_mov_b32 v1, s3
 ; GFX11-CU-NEXT:    v_mov_b32_e32 v2, s0
 ; GFX11-CU-NEXT:    flat_atomic_swap_b32 v[0:1], v2
-; GFX11-CU-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-CU-NEXT:    s_endpgm
     ptr %out, i32 %in) {
 entry:
   %val = atomicrmw volatile xchg ptr %out, i32 %in syncscope("singlethread") release
   ret void
 }
 
 define amdgpu_kernel void @flat_singlethread_acq_rel_atomicrmw(

 ; GFX11-WGP:       ; %bb.0: ; %entry
 ; GFX11-WGP-NEXT:    s_clause 0x1
 ; GFX11-WGP-NEXT:    s_load_b64 s[2:3], s[0:1], 0x0
 ; GFX11-WGP-NEXT:    s_load_b32 s0, s[0:1], 0x8
 ; GFX11-WGP-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-WGP-NEXT:    v_dual_mov_b32 v0, s2 :: v_dual_mov_b32 v1, s3
 ; GFX11-WGP-NEXT:    v_mov_b32_e32 v2, s0
 ; GFX11-WGP-NEXT:    flat_atomic_swap_b32 v[0:1], v2
-; GFX11-WGP-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-WGP-NEXT:    s_endpgm
 ;
 ; GFX11-CU-LABEL: flat_singlethread_acq_rel_atomicrmw:
 ; GFX11-CU:       ; %bb.0: ; %entry
 ; GFX11-CU-NEXT:    s_clause 0x1
 ; GFX11-CU-NEXT:    s_load_b64 s[2:3], s[0:1], 0x0
 ; GFX11-CU-NEXT:    s_load_b32 s0, s[0:1], 0x8
 ; GFX11-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v0, s2 :: v_dual_mov_b32 v1, s3
 ; GFX11-CU-NEXT:    v_mov_b32_e32 v2, s0
 ; GFX11-CU-NEXT:    flat_atomic_swap_b32 v[0:1], v2
-; GFX11-CU-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-CU-NEXT:    s_endpgm
     ptr %out, i32 %in) {
 entry:
   %val = atomicrmw volatile xchg ptr %out, i32 %in syncscope("singlethread") acq_rel
   ret void
 }
 
 define amdgpu_kernel void @flat_singlethread_seq_cst_atomicrmw(

 ; GFX11-WGP:       ; %bb.0: ; %entry
 ; GFX11-WGP-NEXT:    s_clause 0x1
 ; GFX11-WGP-NEXT:    s_load_b64 s[2:3], s[0:1], 0x0
 ; GFX11-WGP-NEXT:    s_load_b32 s0, s[0:1], 0x8
 ; GFX11-WGP-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-WGP-NEXT:    v_dual_mov_b32 v0, s2 :: v_dual_mov_b32 v1, s3
 ; GFX11-WGP-NEXT:    v_mov_b32_e32 v2, s0
 ; GFX11-WGP-NEXT:    flat_atomic_swap_b32 v[0:1], v2
-; GFX11-WGP-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-WGP-NEXT:    s_endpgm
 ;
 ; GFX11-CU-LABEL: flat_singlethread_seq_cst_atomicrmw:
 ; GFX11-CU:       ; %bb.0: ; %entry
 ; GFX11-CU-NEXT:    s_clause 0x1
 ; GFX11-CU-NEXT:    s_load_b64 s[2:3], s[0:1], 0x0
 ; GFX11-CU-NEXT:    s_load_b32 s0, s[0:1], 0x8
 ; GFX11-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v0, s2 :: v_dual_mov_b32 v1, s3
 ; GFX11-CU-NEXT:    v_mov_b32_e32 v2, s0
 ; GFX11-CU-NEXT:    flat_atomic_swap_b32 v[0:1], v2
-; GFX11-CU-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-CU-NEXT:    s_endpgm
     ptr %out, i32 %in) {
 entry:
   %val = atomicrmw volatile xchg ptr %out, i32 %in syncscope("singlethread") seq_cst
   ret void
 }
 
 define amdgpu_kernel void @flat_singlethread_acquire_ret_atomicrmw(

 ; GFX11-WGP-NEXT:    s_load_b64 s[2:3], s[0:1], 0x0
 ; GFX11-WGP-NEXT:    s_load_b32 s0, s[0:1], 0x8
 ; GFX11-WGP-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-WGP-NEXT:    v_dual_mov_b32 v0, s2 :: v_dual_mov_b32 v1, s3
 ; GFX11-WGP-NEXT:    v_mov_b32_e32 v2, s0
 ; GFX11-WGP-NEXT:    flat_atomic_swap_b32 v2, v[0:1], v2 glc
 ; GFX11-WGP-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX11-WGP-NEXT:    flat_store_b32 v[0:1], v2
-; GFX11-WGP-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-WGP-NEXT:    s_endpgm
 ;
 ; GFX11-CU-LABEL: flat_singlethread_acquire_ret_atomicrmw:
 ; GFX11-CU:       ; %bb.0: ; %entry
 ; GFX11-CU-NEXT:    s_clause 0x1
 ; GFX11-CU-NEXT:    s_load_b64 s[2:3], s[0:1], 0x0
 ; GFX11-CU-NEXT:    s_load_b32 s0, s[0:1], 0x8
 ; GFX11-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v0, s2 :: v_dual_mov_b32 v1, s3
 ; GFX11-CU-NEXT:    v_mov_b32_e32 v2, s0
 ; GFX11-CU-NEXT:    flat_atomic_swap_b32 v2, v[0:1], v2 glc
 ; GFX11-CU-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX11-CU-NEXT:    flat_store_b32 v[0:1], v2
-; GFX11-CU-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-CU-NEXT:    s_endpgm
     ptr %out, i32 %in) {
 entry:
   %val = atomicrmw volatile xchg ptr %out, i32 %in syncscope("singlethread") acquire
   store i32 %val, ptr %out, align 4
   ret void
 }
 

 ; GFX11-WGP-NEXT:    s_load_b64 s[2:3], s[0:1], 0x0
 ; GFX11-WGP-NEXT:    s_load_b32 s0, s[0:1], 0x8
 ; GFX11-WGP-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-WGP-NEXT:    v_dual_mov_b32 v0, s2 :: v_dual_mov_b32 v1, s3
 ; GFX11-WGP-NEXT:    v_mov_b32_e32 v2, s0
 ; GFX11-WGP-NEXT:    flat_atomic_swap_b32 v2, v[0:1], v2 glc
 ; GFX11-WGP-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX11-WGP-NEXT:    flat_store_b32 v[0:1], v2
-; GFX11-WGP-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-WGP-NEXT:    s_endpgm
 ;
 ; GFX11-CU-LABEL: flat_singlethread_acq_rel_ret_atomicrmw:
 ; GFX11-CU:       ; %bb.0: ; %entry
 ; GFX11-CU-NEXT:    s_clause 0x1
 ; GFX11-CU-NEXT:    s_load_b64 s[2:3], s[0:1], 0x0
 ; GFX11-CU-NEXT:    s_load_b32 s0, s[0:1], 0x8
 ; GFX11-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v0, s2 :: v_dual_mov_b32 v1, s3
 ; GFX11-CU-NEXT:    v_mov_b32_e32 v2, s0
 ; GFX11-CU-NEXT:    flat_atomic_swap_b32 v2, v[0:1], v2 glc
 ; GFX11-CU-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX11-CU-NEXT:    flat_store_b32 v[0:1], v2
-; GFX11-CU-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-CU-NEXT:    s_endpgm
     ptr %out, i32 %in) {
 entry:
   %val = atomicrmw volatile xchg ptr %out, i32 %in syncscope("singlethread") acq_rel
   store i32 %val, ptr %out, align 4
   ret void
 }
 

 ; GFX11-WGP-NEXT:    s_load_b64 s[2:3], s[0:1], 0x0
 ; GFX11-WGP-NEXT:    s_load_b32 s0, s[0:1], 0x8
 ; GFX11-WGP-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-WGP-NEXT:    v_dual_mov_b32 v0, s2 :: v_dual_mov_b32 v1, s3
 ; GFX11-WGP-NEXT:    v_mov_b32_e32 v2, s0
 ; GFX11-WGP-NEXT:    flat_atomic_swap_b32 v2, v[0:1], v2 glc
 ; GFX11-WGP-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX11-WGP-NEXT:    flat_store_b32 v[0:1], v2
-; GFX11-WGP-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-WGP-NEXT:    s_endpgm
 ;
 ; GFX11-CU-LABEL: flat_singlethread_seq_cst_ret_atomicrmw:
 ; GFX11-CU:       ; %bb.0: ; %entry
 ; GFX11-CU-NEXT:    s_clause 0x1
 ; GFX11-CU-NEXT:    s_load_b64 s[2:3], s[0:1], 0x0
 ; GFX11-CU-NEXT:    s_load_b32 s0, s[0:1], 0x8
 ; GFX11-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v0, s2 :: v_dual_mov_b32 v1, s3
 ; GFX11-CU-NEXT:    v_mov_b32_e32 v2, s0
 ; GFX11-CU-NEXT:    flat_atomic_swap_b32 v2, v[0:1], v2 glc
 ; GFX11-CU-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX11-CU-NEXT:    flat_store_b32 v[0:1], v2
-; GFX11-CU-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-CU-NEXT:    s_endpgm
     ptr %out, i32 %in) {
 entry:
   %val = atomicrmw volatile xchg ptr %out, i32 %in syncscope("singlethread") seq_cst
   store i32 %val, ptr %out, align 4
   ret void
 }
 

 ;
 ; GFX11-WGP-LABEL: flat_singlethread_monotonic_monotonic_cmpxchg:
 ; GFX11-WGP:       ; %bb.0: ; %entry
 ; GFX11-WGP-NEXT:    s_load_b128 s[0:3], s[0:1], 0x0
 ; GFX11-WGP-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-WGP-NEXT:    v_dual_mov_b32 v0, s0 :: v_dual_mov_b32 v1, s1
 ; GFX11-WGP-NEXT:    v_dual_mov_b32 v2, s2 :: v_dual_mov_b32 v3, s3
 ; GFX11-WGP-NEXT:    flat_atomic_cmpswap_b32 v[0:1], v[2:3] offset:16
-; GFX11-WGP-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-WGP-NEXT:    s_endpgm
 ;
 ; GFX11-CU-LABEL: flat_singlethread_monotonic_monotonic_cmpxchg:
 ; GFX11-CU:       ; %bb.0: ; %entry
 ; GFX11-CU-NEXT:    s_load_b128 s[0:3], s[0:1], 0x0
 ; GFX11-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v0, s0 :: v_dual_mov_b32 v1, s1
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v2, s2 :: v_dual_mov_b32 v3, s3
 ; GFX11-CU-NEXT:    flat_atomic_cmpswap_b32 v[0:1], v[2:3] offset:16
-; GFX11-CU-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-CU-NEXT:    s_endpgm
     ptr %out, i32 %in, i32 %old) {
 entry:
   %gep = getelementptr i32, ptr %out, i32 4
   %val = cmpxchg volatile ptr %gep, i32 %old, i32 %in syncscope("singlethread") monotonic monotonic
   ret void
 }
 

 ;
 ; GFX11-WGP-LABEL: flat_singlethread_acquire_monotonic_cmpxchg:
 ; GFX11-WGP:       ; %bb.0: ; %entry
 ; GFX11-WGP-NEXT:    s_load_b128 s[0:3], s[0:1], 0x0
 ; GFX11-WGP-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-WGP-NEXT:    v_dual_mov_b32 v0, s0 :: v_dual_mov_b32 v1, s1
 ; GFX11-WGP-NEXT:    v_dual_mov_b32 v2, s2 :: v_dual_mov_b32 v3, s3
 ; GFX11-WGP-NEXT:    flat_atomic_cmpswap_b32 v[0:1], v[2:3] offset:16
-; GFX11-WGP-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-WGP-NEXT:    s_endpgm
 ;
 ; GFX11-CU-LABEL: flat_singlethread_acquire_monotonic_cmpxchg:
 ; GFX11-CU:       ; %bb.0: ; %entry
 ; GFX11-CU-NEXT:    s_load_b128 s[0:3], s[0:1], 0x0
 ; GFX11-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v0, s0 :: v_dual_mov_b32 v1, s1
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v2, s2 :: v_dual_mov_b32 v3, s3
 ; GFX11-CU-NEXT:    flat_atomic_cmpswap_b32 v[0:1], v[2:3] offset:16
-; GFX11-CU-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-CU-NEXT:    s_endpgm
     ptr %out, i32 %in, i32 %old) {
 entry:
   %gep = getelementptr i32, ptr %out, i32 4
   %val = cmpxchg volatile ptr %gep, i32 %old, i32 %in syncscope("singlethread") acquire monotonic
   ret void
 }
 

 ;
 ; GFX11-WGP-LABEL: flat_singlethread_release_monotonic_cmpxchg:
 ; GFX11-WGP:       ; %bb.0: ; %entry
 ; GFX11-WGP-NEXT:    s_load_b128 s[0:3], s[0:1], 0x0
 ; GFX11-WGP-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-WGP-NEXT:    v_dual_mov_b32 v0, s0 :: v_dual_mov_b32 v1, s1
 ; GFX11-WGP-NEXT:    v_dual_mov_b32 v2, s2 :: v_dual_mov_b32 v3, s3
 ; GFX11-WGP-NEXT:    flat_atomic_cmpswap_b32 v[0:1], v[2:3] offset:16
-; GFX11-WGP-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-WGP-NEXT:    s_endpgm
 ;
 ; GFX11-CU-LABEL: flat_singlethread_release_monotonic_cmpxchg:
 ; GFX11-CU:       ; %bb.0: ; %entry
 ; GFX11-CU-NEXT:    s_load_b128 s[0:3], s[0:1], 0x0
 ; GFX11-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v0, s0 :: v_dual_mov_b32 v1, s1
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v2, s2 :: v_dual_mov_b32 v3, s3
 ; GFX11-CU-NEXT:    flat_atomic_cmpswap_b32 v[0:1], v[2:3] offset:16
-; GFX11-CU-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-CU-NEXT:    s_endpgm
     ptr %out, i32 %in, i32 %old) {
 entry:
   %gep = getelementptr i32, ptr %out, i32 4
   %val = cmpxchg volatile ptr %gep, i32 %old, i32 %in syncscope("singlethread") release monotonic
   ret void
 }
 

 ;
 ; GFX11-WGP-LABEL: flat_singlethread_acq_rel_monotonic_cmpxchg:
 ; GFX11-WGP:       ; %bb.0: ; %entry
 ; GFX11-WGP-NEXT:    s_load_b128 s[0:3], s[0:1], 0x0
 ; GFX11-WGP-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-WGP-NEXT:    v_dual_mov_b32 v0, s0 :: v_dual_mov_b32 v1, s1
 ; GFX11-WGP-NEXT:    v_dual_mov_b32 v2, s2 :: v_dual_mov_b32 v3, s3
 ; GFX11-WGP-NEXT:    flat_atomic_cmpswap_b32 v[0:1], v[2:3] offset:16
-; GFX11-WGP-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-WGP-NEXT:    s_endpgm
 ;
 ; GFX11-CU-LABEL: flat_singlethread_acq_rel_monotonic_cmpxchg:
 ; GFX11-CU:       ; %bb.0: ; %entry
 ; GFX11-CU-NEXT:    s_load_b128 s[0:3], s[0:1], 0x0
 ; GFX11-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v0, s0 :: v_dual_mov_b32 v1, s1
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v2, s2 :: v_dual_mov_b32 v3, s3
 ; GFX11-CU-NEXT:    flat_atomic_cmpswap_b32 v[0:1], v[2:3] offset:16
-; GFX11-CU-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-CU-NEXT:    s_endpgm
     ptr %out, i32 %in, i32 %old) {
 entry:
   %gep = getelementptr i32, ptr %out, i32 4
   %val = cmpxchg volatile ptr %gep, i32 %old, i32 %in syncscope("singlethread") acq_rel monotonic
   ret void
 }
 

 ;
 ; GFX11-WGP-LABEL: flat_singlethread_seq_cst_monotonic_cmpxchg:
 ; GFX11-WGP:       ; %bb.0: ; %entry
 ; GFX11-WGP-NEXT:    s_load_b128 s[0:3], s[0:1], 0x0
 ; GFX11-WGP-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-WGP-NEXT:    v_dual_mov_b32 v0, s0 :: v_dual_mov_b32 v1, s1
 ; GFX11-WGP-NEXT:    v_dual_mov_b32 v2, s2 :: v_dual_mov_b32 v3, s3
 ; GFX11-WGP-NEXT:    flat_atomic_cmpswap_b32 v[0:1], v[2:3] offset:16
-; GFX11-WGP-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-WGP-NEXT:    s_endpgm
 ;
 ; GFX11-CU-LABEL: flat_singlethread_seq_cst_monotonic_cmpxchg:
 ; GFX11-CU:       ; %bb.0: ; %entry
 ; GFX11-CU-NEXT:    s_load_b128 s[0:3], s[0:1], 0x0
 ; GFX11-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v0, s0 :: v_dual_mov_b32 v1, s1
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v2, s2 :: v_dual_mov_b32 v3, s3
 ; GFX11-CU-NEXT:    flat_atomic_cmpswap_b32 v[0:1], v[2:3] offset:16
-; GFX11-CU-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-CU-NEXT:    s_endpgm
     ptr %out, i32 %in, i32 %old) {
 entry:
   %gep = getelementptr i32, ptr %out, i32 4
   %val = cmpxchg volatile ptr %gep, i32 %old, i32 %in syncscope("singlethread") seq_cst monotonic
   ret void
 }
 

 ;
 ; GFX11-WGP-LABEL: flat_singlethread_monotonic_acquire_cmpxchg:
 ; GFX11-WGP:       ; %bb.0: ; %entry
 ; GFX11-WGP-NEXT:    s_load_b128 s[0:3], s[0:1], 0x0
 ; GFX11-WGP-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-WGP-NEXT:    v_dual_mov_b32 v0, s0 :: v_dual_mov_b32 v1, s1
 ; GFX11-WGP-NEXT:    v_dual_mov_b32 v2, s2 :: v_dual_mov_b32 v3, s3
 ; GFX11-WGP-NEXT:    flat_atomic_cmpswap_b32 v[0:1], v[2:3] offset:16
-; GFX11-WGP-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-WGP-NEXT:    s_endpgm
 ;
 ; GFX11-CU-LABEL: flat_singlethread_monotonic_acquire_cmpxchg:
 ; GFX11-CU:       ; %bb.0: ; %entry
 ; GFX11-CU-NEXT:    s_load_b128 s[0:3], s[0:1], 0x0
 ; GFX11-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v0, s0 :: v_dual_mov_b32 v1, s1
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v2, s2 :: v_dual_mov_b32 v3, s3
 ; GFX11-CU-NEXT:    flat_atomic_cmpswap_b32 v[0:1], v[2:3] offset:16
-; GFX11-CU-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-CU-NEXT:    s_endpgm
     ptr %out, i32 %in, i32 %old) {
 entry:
   %gep = getelementptr i32, ptr %out, i32 4
   %val = cmpxchg volatile ptr %gep, i32 %old, i32 %in syncscope("singlethread") monotonic acquire
   ret void
 }
 

 ;
 ; GFX11-WGP-LABEL: flat_singlethread_acquire_acquire_cmpxchg:
 ; GFX11-WGP:       ; %bb.0: ; %entry
 ; GFX11-WGP-NEXT:    s_load_b128 s[0:3], s[0:1], 0x0
 ; GFX11-WGP-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-WGP-NEXT:    v_dual_mov_b32 v0, s0 :: v_dual_mov_b32 v1, s1
 ; GFX11-WGP-NEXT:    v_dual_mov_b32 v2, s2 :: v_dual_mov_b32 v3, s3
 ; GFX11-WGP-NEXT:    flat_atomic_cmpswap_b32 v[0:1], v[2:3] offset:16
-; GFX11-WGP-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-WGP-NEXT:    s_endpgm
 ;
 ; GFX11-CU-LABEL: flat_singlethread_acquire_acquire_cmpxchg:
 ; GFX11-CU:       ; %bb.0: ; %entry
 ; GFX11-CU-NEXT:    s_load_b128 s[0:3], s[0:1], 0x0
 ; GFX11-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v0, s0 :: v_dual_mov_b32 v1, s1
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v2, s2 :: v_dual_mov_b32 v3, s3
 ; GFX11-CU-NEXT:    flat_atomic_cmpswap_b32 v[0:1], v[2:3] offset:16
-; GFX11-CU-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-CU-NEXT:    s_endpgm
     ptr %out, i32 %in, i32 %old) {
 entry:
   %gep = getelementptr i32, ptr %out, i32 4
   %val = cmpxchg volatile ptr %gep, i32 %old, i32 %in syncscope("singlethread") acquire acquire
   ret void
 }
 

 ;
 ; GFX11-WGP-LABEL: flat_singlethread_release_acquire_cmpxchg:
 ; GFX11-WGP:       ; %bb.0: ; %entry
 ; GFX11-WGP-NEXT:    s_load_b128 s[0:3], s[0:1], 0x0
 ; GFX11-WGP-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-WGP-NEXT:    v_dual_mov_b32 v0, s0 :: v_dual_mov_b32 v1, s1
 ; GFX11-WGP-NEXT:    v_dual_mov_b32 v2, s2 :: v_dual_mov_b32 v3, s3
 ; GFX11-WGP-NEXT:    flat_atomic_cmpswap_b32 v[0:1], v[2:3] offset:16
-; GFX11-WGP-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-WGP-NEXT:    s_endpgm
 ;
 ; GFX11-CU-LABEL: flat_singlethread_release_acquire_cmpxchg:
 ; GFX11-CU:       ; %bb.0: ; %entry
 ; GFX11-CU-NEXT:    s_load_b128 s[0:3], s[0:1], 0x0
 ; GFX11-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v0, s0 :: v_dual_mov_b32 v1, s1
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v2, s2 :: v_dual_mov_b32 v3, s3
 ; GFX11-CU-NEXT:    flat_atomic_cmpswap_b32 v[0:1], v[2:3] offset:16
-; GFX11-CU-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-CU-NEXT:    s_endpgm
     ptr %out, i32 %in, i32 %old) {
 entry:
   %gep = getelementptr i32, ptr %out, i32 4
   %val = cmpxchg volatile ptr %gep, i32 %old, i32 %in syncscope("singlethread") release acquire
   ret void
 }
 

 ;
 ; GFX11-WGP-LABEL: flat_singlethread_acq_rel_acquire_cmpxchg:
 ; GFX11-WGP:       ; %bb.0: ; %entry
 ; GFX11-WGP-NEXT:    s_load_b128 s[0:3], s[0:1], 0x0
 ; GFX11-WGP-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-WGP-NEXT:    v_dual_mov_b32 v0, s0 :: v_dual_mov_b32 v1, s1
 ; GFX11-WGP-NEXT:    v_dual_mov_b32 v2, s2 :: v_dual_mov_b32 v3, s3
 ; GFX11-WGP-NEXT:    flat_atomic_cmpswap_b32 v[0:1], v[2:3] offset:16
-; GFX11-WGP-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-WGP-NEXT:    s_endpgm
 ;
 ; GFX11-CU-LABEL: flat_singlethread_acq_rel_acquire_cmpxchg:
 ; GFX11-CU:       ; %bb.0: ; %entry
 ; GFX11-CU-NEXT:    s_load_b128 s[0:3], s[0:1], 0x0
 ; GFX11-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v0, s0 :: v_dual_mov_b32 v1, s1
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v2, s2 :: v_dual_mov_b32 v3, s3
 ; GFX11-CU-NEXT:    flat_atomic_cmpswap_b32 v[0:1], v[2:3] offset:16
-; GFX11-CU-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-CU-NEXT:    s_endpgm
     ptr %out, i32 %in, i32 %old) {
 entry:
   %gep = getelementptr i32, ptr %out, i32 4
   %val = cmpxchg volatile ptr %gep, i32 %old, i32 %in syncscope("singlethread") acq_rel acquire
   ret void
 }
 

 ;
 ; GFX11-WGP-LABEL: flat_singlethread_seq_cst_acquire_cmpxchg:
 ; GFX11-WGP:       ; %bb.0: ; %entry
 ; GFX11-WGP-NEXT:    s_load_b128 s[0:3], s[0:1], 0x0
 ; GFX11-WGP-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-WGP-NEXT:    v_dual_mov_b32 v0, s0 :: v_dual_mov_b32 v1, s1
 ; GFX11-WGP-NEXT:    v_dual_mov_b32 v2, s2 :: v_dual_mov_b32 v3, s3
 ; GFX11-WGP-NEXT:    flat_atomic_cmpswap_b32 v[0:1], v[2:3] offset:16
-; GFX11-WGP-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-WGP-NEXT:    s_endpgm
 ;
 ; GFX11-CU-LABEL: flat_singlethread_seq_cst_acquire_cmpxchg:
 ; GFX11-CU:       ; %bb.0: ; %entry
 ; GFX11-CU-NEXT:    s_load_b128 s[0:3], s[0:1], 0x0
 ; GFX11-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v0, s0 :: v_dual_mov_b32 v1, s1
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v2, s2 :: v_dual_mov_b32 v3, s3
 ; GFX11-CU-NEXT:    flat_atomic_cmpswap_b32 v[0:1], v[2:3] offset:16
-; GFX11-CU-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-CU-NEXT:    s_endpgm
     ptr %out, i32 %in, i32 %old) {
 entry:
   %gep = getelementptr i32, ptr %out, i32 4
   %val = cmpxchg volatile ptr %gep, i32 %old, i32 %in syncscope("singlethread") seq_cst acquire
   ret void
 }
 

 ;
 ; GFX11-WGP-LABEL: flat_singlethread_monotonic_seq_cst_cmpxchg:
 ; GFX11-WGP:       ; %bb.0: ; %entry
 ; GFX11-WGP-NEXT:    s_load_b128 s[0:3], s[0:1], 0x0
 ; GFX11-WGP-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-WGP-NEXT:    v_dual_mov_b32 v0, s0 :: v_dual_mov_b32 v1, s1
 ; GFX11-WGP-NEXT:    v_dual_mov_b32 v2, s2 :: v_dual_mov_b32 v3, s3
 ; GFX11-WGP-NEXT:    flat_atomic_cmpswap_b32 v[0:1], v[2:3] offset:16
-; GFX11-WGP-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-WGP-NEXT:    s_endpgm
 ;
 ; GFX11-CU-LABEL: flat_singlethread_monotonic_seq_cst_cmpxchg:
 ; GFX11-CU:       ; %bb.0: ; %entry
 ; GFX11-CU-NEXT:    s_load_b128 s[0:3], s[0:1], 0x0
 ; GFX11-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v0, s0 :: v_dual_mov_b32 v1, s1
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v2, s2 :: v_dual_mov_b32 v3, s3
 ; GFX11-CU-NEXT:    flat_atomic_cmpswap_b32 v[0:1], v[2:3] offset:16
-; GFX11-CU-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-CU-NEXT:    s_endpgm
     ptr %out, i32 %in, i32 %old) {
 entry:
   %gep = getelementptr i32, ptr %out, i32 4
   %val = cmpxchg volatile ptr %gep, i32 %old, i32 %in syncscope("singlethread") monotonic seq_cst
   ret void
 }
 

 ;
 ; GFX11-WGP-LABEL: flat_singlethread_acquire_seq_cst_cmpxchg:
 ; GFX11-WGP:       ; %bb.0: ; %entry
 ; GFX11-WGP-NEXT:    s_load_b128 s[0:3], s[0:1], 0x0
 ; GFX11-WGP-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-WGP-NEXT:    v_dual_mov_b32 v0, s0 :: v_dual_mov_b32 v1, s1
 ; GFX11-WGP-NEXT:    v_dual_mov_b32 v2, s2 :: v_dual_mov_b32 v3, s3
 ; GFX11-WGP-NEXT:    flat_atomic_cmpswap_b32 v[0:1], v[2:3] offset:16
-; GFX11-WGP-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-WGP-NEXT:    s_endpgm
 ;
 ; GFX11-CU-LABEL: flat_singlethread_acquire_seq_cst_cmpxchg:
 ; GFX11-CU:       ; %bb.0: ; %entry
 ; GFX11-CU-NEXT:    s_load_b128 s[0:3], s[0:1], 0x0
 ; GFX11-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v0, s0 :: v_dual_mov_b32 v1, s1
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v2, s2 :: v_dual_mov_b32 v3, s3
 ; GFX11-CU-NEXT:    flat_atomic_cmpswap_b32 v[0:1], v[2:3] offset:16
-; GFX11-CU-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-CU-NEXT:    s_endpgm
     ptr %out, i32 %in, i32 %old) {
 entry:
   %gep = getelementptr i32, ptr %out, i32 4
   %val = cmpxchg volatile ptr %gep, i32 %old, i32 %in syncscope("singlethread") acquire seq_cst
   ret void
 }
 

 ;
 ; GFX11-WGP-LABEL: flat_singlethread_release_seq_cst_cmpxchg:
 ; GFX11-WGP:       ; %bb.0: ; %entry
 ; GFX11-WGP-NEXT:    s_load_b128 s[0:3], s[0:1], 0x0
 ; GFX11-WGP-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-WGP-NEXT:    v_dual_mov_b32 v0, s0 :: v_dual_mov_b32 v1, s1
 ; GFX11-WGP-NEXT:    v_dual_mov_b32 v2, s2 :: v_dual_mov_b32 v3, s3
 ; GFX11-WGP-NEXT:    flat_atomic_cmpswap_b32 v[0:1], v[2:3] offset:16
-; GFX11-WGP-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-WGP-NEXT:    s_endpgm
 ;
 ; GFX11-CU-LABEL: flat_singlethread_release_seq_cst_cmpxchg:
 ; GFX11-CU:       ; %bb.0: ; %entry
 ; GFX11-CU-NEXT:    s_load_b128 s[0:3], s[0:1], 0x0
 ; GFX11-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v0, s0 :: v_dual_mov_b32 v1, s1
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v2, s2 :: v_dual_mov_b32 v3, s3
 ; GFX11-CU-NEXT:    flat_atomic_cmpswap_b32 v[0:1], v[2:3] offset:16
-; GFX11-CU-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-CU-NEXT:    s_endpgm
     ptr %out, i32 %in, i32 %old) {
 entry:
   %gep = getelementptr i32, ptr %out, i32 4
   %val = cmpxchg volatile ptr %gep, i32 %old, i32 %in syncscope("singlethread") release seq_cst
   ret void
 }
 

 ;
 ; GFX11-WGP-LABEL: flat_singlethread_acq_rel_seq_cst_cmpxchg:
 ; GFX11-WGP:       ; %bb.0: ; %entry
 ; GFX11-WGP-NEXT:    s_load_b128 s[0:3], s[0:1], 0x0
 ; GFX11-WGP-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-WGP-NEXT:    v_dual_mov_b32 v0, s0 :: v_dual_mov_b32 v1, s1
 ; GFX11-WGP-NEXT:    v_dual_mov_b32 v2, s2 :: v_dual_mov_b32 v3, s3
 ; GFX11-WGP-NEXT:    flat_atomic_cmpswap_b32 v[0:1], v[2:3] offset:16
-; GFX11-WGP-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-WGP-NEXT:    s_endpgm
 ;
 ; GFX11-CU-LABEL: flat_singlethread_acq_rel_seq_cst_cmpxchg:
 ; GFX11-CU:       ; %bb.0: ; %entry
 ; GFX11-CU-NEXT:    s_load_b128 s[0:3], s[0:1], 0x0
 ; GFX11-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v0, s0 :: v_dual_mov_b32 v1, s1
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v2, s2 :: v_dual_mov_b32 v3, s3
 ; GFX11-CU-NEXT:    flat_atomic_cmpswap_b32 v[0:1], v[2:3] offset:16
-; GFX11-CU-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-CU-NEXT:    s_endpgm
     ptr %out, i32 %in, i32 %old) {
 entry:
   %gep = getelementptr i32, ptr %out, i32 4
   %val = cmpxchg volatile ptr %gep, i32 %old, i32 %in syncscope("singlethread") acq_rel seq_cst
   ret void
 }
 

 ;
 ; GFX11-WGP-LABEL: flat_singlethread_seq_cst_seq_cst_cmpxchg:
 ; GFX11-WGP:       ; %bb.0: ; %entry
 ; GFX11-WGP-NEXT:    s_load_b128 s[0:3], s[0:1], 0x0
 ; GFX11-WGP-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-WGP-NEXT:    v_dual_mov_b32 v0, s0 :: v_dual_mov_b32 v1, s1
 ; GFX11-WGP-NEXT:    v_dual_mov_b32 v2, s2 :: v_dual_mov_b32 v3, s3
 ; GFX11-WGP-NEXT:    flat_atomic_cmpswap_b32 v[0:1], v[2:3] offset:16
-; GFX11-WGP-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-WGP-NEXT:    s_endpgm
 ;
 ; GFX11-CU-LABEL: flat_singlethread_seq_cst_seq_cst_cmpxchg:
 ; GFX11-CU:       ; %bb.0: ; %entry
 ; GFX11-CU-NEXT:    s_load_b128 s[0:3], s[0:1], 0x0
 ; GFX11-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v0, s0 :: v_dual_mov_b32 v1, s1
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v2, s2 :: v_dual_mov_b32 v3, s3
 ; GFX11-CU-NEXT:    flat_atomic_cmpswap_b32 v[0:1], v[2:3] offset:16
-; GFX11-CU-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-CU-NEXT:    s_endpgm
     ptr %out, i32 %in, i32 %old) {
 entry:
   %gep = getelementptr i32, ptr %out, i32 4
   %val = cmpxchg volatile ptr %gep, i32 %old, i32 %in syncscope("singlethread") seq_cst seq_cst
   ret void
 }
 

 ; GFX11-WGP:       ; %bb.0: ; %entry
 ; GFX11-WGP-NEXT:    s_load_b128 s[0:3], s[0:1], 0x0
 ; GFX11-WGP-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-WGP-NEXT:    v_dual_mov_b32 v0, s0 :: v_dual_mov_b32 v1, s1
 ; GFX11-WGP-NEXT:    v_dual_mov_b32 v2, s2 :: v_dual_mov_b32 v3, s3
 ; GFX11-WGP-NEXT:    flat_atomic_cmpswap_b32 v2, v[0:1], v[2:3] offset:16 glc
 ; GFX11-WGP-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX11-WGP-NEXT:    flat_store_b32 v[0:1], v2
-; GFX11-WGP-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-WGP-NEXT:    s_endpgm
 ;
 ; GFX11-CU-LABEL: flat_singlethread_monotonic_monotonic_ret_cmpxchg:
 ; GFX11-CU:       ; %bb.0: ; %entry
 ; GFX11-CU-NEXT:    s_load_b128 s[0:3], s[0:1], 0x0
 ; GFX11-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v0, s0 :: v_dual_mov_b32 v1, s1
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v2, s2 :: v_dual_mov_b32 v3, s3
 ; GFX11-CU-NEXT:    flat_atomic_cmpswap_b32 v2, v[0:1], v[2:3] offset:16 glc
 ; GFX11-CU-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX11-CU-NEXT:    flat_store_b32 v[0:1], v2
-; GFX11-CU-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-CU-NEXT:    s_endpgm
     ptr %out, i32 %in, i32 %old) {
 entry:
   %gep = getelementptr i32, ptr %out, i32 4
   %val = cmpxchg volatile ptr %gep, i32 %old, i32 %in syncscope("singlethread") monotonic monotonic
   %val0 = extractvalue { i32, i1 } %val, 0
   store i32 %val0, ptr %out, align 4
   ret void

 ; GFX11-WGP:       ; %bb.0: ; %entry
 ; GFX11-WGP-NEXT:    s_load_b128 s[0:3], s[0:1], 0x0
 ; GFX11-WGP-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-WGP-NEXT:    v_dual_mov_b32 v0, s0 :: v_dual_mov_b32 v1, s1
 ; GFX11-WGP-NEXT:    v_dual_mov_b32 v2, s2 :: v_dual_mov_b32 v3, s3
 ; GFX11-WGP-NEXT:    flat_atomic_cmpswap_b32 v2, v[0:1], v[2:3] offset:16 glc
 ; GFX11-WGP-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX11-WGP-NEXT:    flat_store_b32 v[0:1], v2
-; GFX11-WGP-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-WGP-NEXT:    s_endpgm
 ;
 ; GFX11-CU-LABEL: flat_singlethread_acquire_monotonic_ret_cmpxchg:
 ; GFX11-CU:       ; %bb.0: ; %entry
 ; GFX11-CU-NEXT:    s_load_b128 s[0:3], s[0:1], 0x0
 ; GFX11-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v0, s0 :: v_dual_mov_b32 v1, s1
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v2, s2 :: v_dual_mov_b32 v3, s3
 ; GFX11-CU-NEXT:    flat_atomic_cmpswap_b32 v2, v[0:1], v[2:3] offset:16 glc
 ; GFX11-CU-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX11-CU-NEXT:    flat_store_b32 v[0:1], v2
-; GFX11-CU-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-CU-NEXT:    s_endpgm
     ptr %out, i32 %in, i32 %old) {
 entry:
   %gep = getelementptr i32, ptr %out, i32 4
   %val = cmpxchg volatile ptr %gep, i32 %old, i32 %in syncscope("singlethread") acquire monotonic
   %val0 = extractvalue { i32, i1 } %val, 0
   store i32 %val0, ptr %out, align 4
   ret void

 ; GFX11-WGP:       ; %bb.0: ; %entry
 ; GFX11-WGP-NEXT:    s_load_b128 s[0:3], s[0:1], 0x0
 ; GFX11-WGP-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-WGP-NEXT:    v_dual_mov_b32 v0, s0 :: v_dual_mov_b32 v1, s1
 ; GFX11-WGP-NEXT:    v_dual_mov_b32 v2, s2 :: v_dual_mov_b32 v3, s3
 ; GFX11-WGP-NEXT:    flat_atomic_cmpswap_b32 v2, v[0:1], v[2:3] offset:16 glc
 ; GFX11-WGP-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX11-WGP-NEXT:    flat_store_b32 v[0:1], v2
-; GFX11-WGP-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-WGP-NEXT:    s_endpgm
 ;
 ; GFX11-CU-LABEL: flat_singlethread_release_monotonic_ret_cmpxchg:
 ; GFX11-CU:       ; %bb.0: ; %entry
 ; GFX11-CU-NEXT:    s_load_b128 s[0:3], s[0:1], 0x0
 ; GFX11-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v0, s0 :: v_dual_mov_b32 v1, s1
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v2, s2 :: v_dual_mov_b32 v3, s3
 ; GFX11-CU-NEXT:    flat_atomic_cmpswap_b32 v2, v[0:1], v[2:3] offset:16 glc
 ; GFX11-CU-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX11-CU-NEXT:    flat_store_b32 v[0:1], v2
-; GFX11-CU-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-CU-NEXT:    s_endpgm
     ptr %out, i32 %in, i32 %old) {
 entry:
   %gep = getelementptr i32, ptr %out, i32 4
   %val = cmpxchg volatile ptr %gep, i32 %old, i32 %in syncscope("singlethread") release monotonic
   %val0 = extractvalue { i32, i1 } %val, 0
   store i32 %val0, ptr %out, align 4
   ret void

 ; GFX11-WGP:       ; %bb.0: ; %entry
 ; GFX11-WGP-NEXT:    s_load_b128 s[0:3], s[0:1], 0x0
 ; GFX11-WGP-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-WGP-NEXT:    v_dual_mov_b32 v0, s0 :: v_dual_mov_b32 v1, s1
 ; GFX11-WGP-NEXT:    v_dual_mov_b32 v2, s2 :: v_dual_mov_b32 v3, s3
 ; GFX11-WGP-NEXT:    flat_atomic_cmpswap_b32 v2, v[0:1], v[2:3] offset:16 glc
 ; GFX11-WGP-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX11-WGP-NEXT:    flat_store_b32 v[0:1], v2
-; GFX11-WGP-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-WGP-NEXT:    s_endpgm
 ;
 ; GFX11-CU-LABEL: flat_singlethread_acq_rel_monotonic_ret_cmpxchg:
 ; GFX11-CU:       ; %bb.0: ; %entry
 ; GFX11-CU-NEXT:    s_load_b128 s[0:3], s[0:1], 0x0
 ; GFX11-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v0, s0 :: v_dual_mov_b32 v1, s1
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v2, s2 :: v_dual_mov_b32 v3, s3
 ; GFX11-CU-NEXT:    flat_atomic_cmpswap_b32 v2, v[0:1], v[2:3] offset:16 glc
 ; GFX11-CU-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX11-CU-NEXT:    flat_store_b32 v[0:1], v2
-; GFX11-CU-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-CU-NEXT:    s_endpgm
     ptr %out, i32 %in, i32 %old) {
 entry:
   %gep = getelementptr i32, ptr %out, i32 4
   %val = cmpxchg volatile ptr %gep, i32 %old, i32 %in syncscope("singlethread") acq_rel monotonic
   %val0 = extractvalue { i32, i1 } %val, 0
   store i32 %val0, ptr %out, align 4
   ret void

 ; GFX11-WGP:       ; %bb.0: ; %entry
 ; GFX11-WGP-NEXT:    s_load_b128 s[0:3], s[0:1], 0x0
 ; GFX11-WGP-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-WGP-NEXT:    v_dual_mov_b32 v0, s0 :: v_dual_mov_b32 v1, s1
 ; GFX11-WGP-NEXT:    v_dual_mov_b32 v2, s2 :: v_dual_mov_b32 v3, s3
 ; GFX11-WGP-NEXT:    flat_atomic_cmpswap_b32 v2, v[0:1], v[2:3] offset:16 glc
 ; GFX11-WGP-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX11-WGP-NEXT:    flat_store_b32 v[0:1], v2
-; GFX11-WGP-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-WGP-NEXT:    s_endpgm
 ;
 ; GFX11-CU-LABEL: flat_singlethread_seq_cst_monotonic_ret_cmpxchg:
 ; GFX11-CU:       ; %bb.0: ; %entry
 ; GFX11-CU-NEXT:    s_load_b128 s[0:3], s[0:1], 0x0
 ; GFX11-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v0, s0 :: v_dual_mov_b32 v1, s1
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v2, s2 :: v_dual_mov_b32 v3, s3
 ; GFX11-CU-NEXT:    flat_atomic_cmpswap_b32 v2, v[0:1], v[2:3] offset:16 glc
 ; GFX11-CU-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX11-CU-NEXT:    flat_store_b32 v[0:1], v2
-; GFX11-CU-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-CU-NEXT:    s_endpgm
     ptr %out, i32 %in, i32 %old) {
 entry:
   %gep = getelementptr i32, ptr %out, i32 4
   %val = cmpxchg volatile ptr %gep, i32 %old, i32 %in syncscope("singlethread") seq_cst monotonic
   %val0 = extractvalue { i32, i1 } %val, 0
   store i32 %val0, ptr %out, align 4
   ret void

 ; GFX11-WGP:       ; %bb.0: ; %entry
 ; GFX11-WGP-NEXT:    s_load_b128 s[0:3], s[0:1], 0x0
 ; GFX11-WGP-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-WGP-NEXT:    v_dual_mov_b32 v0, s0 :: v_dual_mov_b32 v1, s1
 ; GFX11-WGP-NEXT:    v_dual_mov_b32 v2, s2 :: v_dual_mov_b32 v3, s3
 ; GFX11-WGP-NEXT:    flat_atomic_cmpswap_b32 v2, v[0:1], v[2:3] offset:16 glc
 ; GFX11-WGP-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX11-WGP-NEXT:    flat_store_b32 v[0:1], v2
-; GFX11-WGP-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-WGP-NEXT:    s_endpgm
 ;
 ; GFX11-CU-LABEL: flat_singlethread_monotonic_acquire_ret_cmpxchg:
 ; GFX11-CU:       ; %bb.0: ; %entry
 ; GFX11-CU-NEXT:    s_load_b128 s[0:3], s[0:1], 0x0
 ; GFX11-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v0, s0 :: v_dual_mov_b32 v1, s1
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v2, s2 :: v_dual_mov_b32 v3, s3
 ; GFX11-CU-NEXT:    flat_atomic_cmpswap_b32 v2, v[0:1], v[2:3] offset:16 glc
 ; GFX11-CU-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX11-CU-NEXT:    flat_store_b32 v[0:1], v2
-; GFX11-CU-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-CU-NEXT:    s_endpgm
     ptr %out, i32 %in, i32 %old) {
 entry:
   %gep = getelementptr i32, ptr %out, i32 4
   %val = cmpxchg volatile ptr %gep, i32 %old, i32 %in syncscope("singlethread") monotonic acquire
   %val0 = extractvalue { i32, i1 } %val, 0
   store i32 %val0, ptr %out, align 4
   ret void

 ; GFX11-WGP:       ; %bb.0: ; %entry
 ; GFX11-WGP-NEXT:    s_load_b128 s[0:3], s[0:1], 0x0
 ; GFX11-WGP-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-WGP-NEXT:    v_dual_mov_b32 v0, s0 :: v_dual_mov_b32 v1, s1
 ; GFX11-WGP-NEXT:    v_dual_mov_b32 v2, s2 :: v_dual_mov_b32 v3, s3
 ; GFX11-WGP-NEXT:    flat_atomic_cmpswap_b32 v2, v[0:1], v[2:3] offset:16 glc
 ; GFX11-WGP-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX11-WGP-NEXT:    flat_store_b32 v[0:1], v2
-; GFX11-WGP-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-WGP-NEXT:    s_endpgm
 ;
 ; GFX11-CU-LABEL: flat_singlethread_acquire_acquire_ret_cmpxchg:
 ; GFX11-CU:       ; %bb.0: ; %entry
 ; GFX11-CU-NEXT:    s_load_b128 s[0:3], s[0:1], 0x0
 ; GFX11-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v0, s0 :: v_dual_mov_b32 v1, s1
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v2, s2 :: v_dual_mov_b32 v3, s3
 ; GFX11-CU-NEXT:    flat_atomic_cmpswap_b32 v2, v[0:1], v[2:3] offset:16 glc
 ; GFX11-CU-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX11-CU-NEXT:    flat_store_b32 v[0:1], v2
-; GFX11-CU-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-CU-NEXT:    s_endpgm
     ptr %out, i32 %in, i32 %old) {
 entry:
   %gep = getelementptr i32, ptr %out, i32 4
   %val = cmpxchg volatile ptr %gep, i32 %old, i32 %in syncscope("singlethread") acquire acquire
   %val0 = extractvalue { i32, i1 } %val, 0
   store i32 %val0, ptr %out, align 4
   ret void

 ; GFX11-WGP:       ; %bb.0: ; %entry
 ; GFX11-WGP-NEXT:    s_load_b128 s[0:3], s[0:1], 0x0
 ; GFX11-WGP-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-WGP-NEXT:    v_dual_mov_b32 v0, s0 :: v_dual_mov_b32 v1, s1
 ; GFX11-WGP-NEXT:    v_dual_mov_b32 v2, s2 :: v_dual_mov_b32 v3, s3
 ; GFX11-WGP-NEXT:    flat_atomic_cmpswap_b32 v2, v[0:1], v[2:3] offset:16 glc
 ; GFX11-WGP-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX11-WGP-NEXT:    flat_store_b32 v[0:1], v2
-; GFX11-WGP-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-WGP-NEXT:    s_endpgm
 ;
 ; GFX11-CU-LABEL: flat_singlethread_release_acquire_ret_cmpxchg:
 ; GFX11-CU:       ; %bb.0: ; %entry
 ; GFX11-CU-NEXT:    s_load_b128 s[0:3], s[0:1], 0x0
 ; GFX11-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v0, s0 :: v_dual_mov_b32 v1, s1
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v2, s2 :: v_dual_mov_b32 v3, s3
 ; GFX11-CU-NEXT:    flat_atomic_cmpswap_b32 v2, v[0:1], v[2:3] offset:16 glc
 ; GFX11-CU-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX11-CU-NEXT:    flat_store_b32 v[0:1], v2
-; GFX11-CU-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-CU-NEXT:    s_endpgm
     ptr %out, i32 %in, i32 %old) {
 entry:
   %gep = getelementptr i32, ptr %out, i32 4
   %val = cmpxchg volatile ptr %gep, i32 %old, i32 %in syncscope("singlethread") release acquire
   %val0 = extractvalue { i32, i1 } %val, 0
   store i32 %val0, ptr %out, align 4
   ret void

 ; GFX11-WGP:       ; %bb.0: ; %entry
 ; GFX11-WGP-NEXT:    s_load_b128 s[0:3], s[0:1], 0x0
 ; GFX11-WGP-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-WGP-NEXT:    v_dual_mov_b32 v0, s0 :: v_dual_mov_b32 v1, s1
 ; GFX11-WGP-NEXT:    v_dual_mov_b32 v2, s2 :: v_dual_mov_b32 v3, s3
 ; GFX11-WGP-NEXT:    flat_atomic_cmpswap_b32 v2, v[0:1], v[2:3] offset:16 glc
 ; GFX11-WGP-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX11-WGP-NEXT:    flat_store_b32 v[0:1], v2
-; GFX11-WGP-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-WGP-NEXT:    s_endpgm
 ;
 ; GFX11-CU-LABEL: flat_singlethread_acq_rel_acquire_ret_cmpxchg:
 ; GFX11-CU:       ; %bb.0: ; %entry
 ; GFX11-CU-NEXT:    s_load_b128 s[0:3], s[0:1], 0x0
 ; GFX11-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v0, s0 :: v_dual_mov_b32 v1, s1
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v2, s2 :: v_dual_mov_b32 v3, s3
 ; GFX11-CU-NEXT:    flat_atomic_cmpswap_b32 v2, v[0:1], v[2:3] offset:16 glc
 ; GFX11-CU-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX11-CU-NEXT:    flat_store_b32 v[0:1], v2
-; GFX11-CU-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-CU-NEXT:    s_endpgm
     ptr %out, i32 %in, i32 %old) {
 entry:
   %gep = getelementptr i32, ptr %out, i32 4
   %val = cmpxchg volatile ptr %gep, i32 %old, i32 %in syncscope("singlethread") acq_rel acquire
   %val0 = extractvalue { i32, i1 } %val, 0
   store i32 %val0, ptr %out, align 4
   ret void

 ; GFX11-WGP:       ; %bb.0: ; %entry
 ; GFX11-WGP-NEXT:    s_load_b128 s[0:3], s[0:1], 0x0
 ; GFX11-WGP-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-WGP-NEXT:    v_dual_mov_b32 v0, s0 :: v_dual_mov_b32 v1, s1
 ; GFX11-WGP-NEXT:    v_dual_mov_b32 v2, s2 :: v_dual_mov_b32 v3, s3
 ; GFX11-WGP-NEXT:    flat_atomic_cmpswap_b32 v2, v[0:1], v[2:3] offset:16 glc
 ; GFX11-WGP-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX11-WGP-NEXT:    flat_store_b32 v[0:1], v2
-; GFX11-WGP-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-WGP-NEXT:    s_endpgm
 ;
 ; GFX11-CU-LABEL: flat_singlethread_seq_cst_acquire_ret_cmpxchg:
 ; GFX11-CU:       ; %bb.0: ; %entry
 ; GFX11-CU-NEXT:    s_load_b128 s[0:3], s[0:1], 0x0
 ; GFX11-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v0, s0 :: v_dual_mov_b32 v1, s1
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v2, s2 :: v_dual_mov_b32 v3, s3
 ; GFX11-CU-NEXT:    flat_atomic_cmpswap_b32 v2, v[0:1], v[2:3] offset:16 glc
 ; GFX11-CU-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX11-CU-NEXT:    flat_store_b32 v[0:1], v2
-; GFX11-CU-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-CU-NEXT:    s_endpgm
     ptr %out, i32 %in, i32 %old) {
 entry:
   %gep = getelementptr i32, ptr %out, i32 4
   %val = cmpxchg volatile ptr %gep, i32 %old, i32 %in syncscope("singlethread") seq_cst acquire
   %val0 = extractvalue { i32, i1 } %val, 0
   store i32 %val0, ptr %out, align 4
   ret void

 ; GFX11-WGP:       ; %bb.0: ; %entry
 ; GFX11-WGP-NEXT:    s_load_b128 s[0:3], s[0:1], 0x0
 ; GFX11-WGP-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-WGP-NEXT:    v_dual_mov_b32 v0, s0 :: v_dual_mov_b32 v1, s1
 ; GFX11-WGP-NEXT:    v_dual_mov_b32 v2, s2 :: v_dual_mov_b32 v3, s3
 ; GFX11-WGP-NEXT:    flat_atomic_cmpswap_b32 v2, v[0:1], v[2:3] offset:16 glc
 ; GFX11-WGP-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX11-WGP-NEXT:    flat_store_b32 v[0:1], v2
-; GFX11-WGP-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-WGP-NEXT:    s_endpgm
 ;
 ; GFX11-CU-LABEL: flat_singlethread_monotonic_seq_cst_ret_cmpxchg:
 ; GFX11-CU:       ; %bb.0: ; %entry
 ; GFX11-CU-NEXT:    s_load_b128 s[0:3], s[0:1], 0x0
 ; GFX11-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v0, s0 :: v_dual_mov_b32 v1, s1
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v2, s2 :: v_dual_mov_b32 v3, s3
 ; GFX11-CU-NEXT:    flat_atomic_cmpswap_b32 v2, v[0:1], v[2:3] offset:16 glc
 ; GFX11-CU-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX11-CU-NEXT:    flat_store_b32 v[0:1], v2
-; GFX11-CU-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-CU-NEXT:    s_endpgm
     ptr %out, i32 %in, i32 %old) {
 entry:
   %gep = getelementptr i32, ptr %out, i32 4
   %val = cmpxchg volatile ptr %gep, i32 %old, i32 %in syncscope("singlethread") monotonic seq_cst
   %val0 = extractvalue { i32, i1 } %val, 0
   store i32 %val0, ptr %out, align 4
   ret void

 ; GFX11-WGP:       ; %bb.0: ; %entry
 ; GFX11-WGP-NEXT:    s_load_b128 s[0:3], s[0:1], 0x0
 ; GFX11-WGP-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-WGP-NEXT:    v_dual_mov_b32 v0, s0 :: v_dual_mov_b32 v1, s1
 ; GFX11-WGP-NEXT:    v_dual_mov_b32 v2, s2 :: v_dual_mov_b32 v3, s3
 ; GFX11-WGP-NEXT:    flat_atomic_cmpswap_b32 v2, v[0:1], v[2:3] offset:16 glc
 ; GFX11-WGP-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX11-WGP-NEXT:    flat_store_b32 v[0:1], v2
-; GFX11-WGP-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-WGP-NEXT:    s_endpgm
 ;
 ; GFX11-CU-LABEL: flat_singlethread_acquire_seq_cst_ret_cmpxchg:
 ; GFX11-CU:       ; %bb.0: ; %entry
 ; GFX11-CU-NEXT:    s_load_b128 s[0:3], s[0:1], 0x0
 ; GFX11-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v0, s0 :: v_dual_mov_b32 v1, s1
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v2, s2 :: v_dual_mov_b32 v3, s3
 ; GFX11-CU-NEXT:    flat_atomic_cmpswap_b32 v2, v[0:1], v[2:3] offset:16 glc
 ; GFX11-CU-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX11-CU-NEXT:    flat_store_b32 v[0:1], v2
-; GFX11-CU-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-CU-NEXT:    s_endpgm
     ptr %out, i32 %in, i32 %old) {
 entry:
   %gep = getelementptr i32, ptr %out, i32 4
   %val = cmpxchg volatile ptr %gep, i32 %old, i32 %in syncscope("singlethread") acquire seq_cst
   %val0 = extractvalue { i32, i1 } %val, 0
   store i32 %val0, ptr %out, align 4
   ret void

 ; GFX11-WGP:       ; %bb.0: ; %entry
 ; GFX11-WGP-NEXT:    s_load_b128 s[0:3], s[0:1], 0x0
 ; GFX11-WGP-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-WGP-NEXT:    v_dual_mov_b32 v0, s0 :: v_dual_mov_b32 v1, s1
 ; GFX11-WGP-NEXT:    v_dual_mov_b32 v2, s2 :: v_dual_mov_b32 v3, s3
 ; GFX11-WGP-NEXT:    flat_atomic_cmpswap_b32 v2, v[0:1], v[2:3] offset:16 glc
 ; GFX11-WGP-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX11-WGP-NEXT:    flat_store_b32 v[0:1], v2
-; GFX11-WGP-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-WGP-NEXT:    s_endpgm
 ;
 ; GFX11-CU-LABEL: flat_singlethread_release_seq_cst_ret_cmpxchg:
 ; GFX11-CU:       ; %bb.0: ; %entry
 ; GFX11-CU-NEXT:    s_load_b128 s[0:3], s[0:1], 0x0
 ; GFX11-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v0, s0 :: v_dual_mov_b32 v1, s1
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v2, s2 :: v_dual_mov_b32 v3, s3
 ; GFX11-CU-NEXT:    flat_atomic_cmpswap_b32 v2, v[0:1], v[2:3] offset:16 glc
 ; GFX11-CU-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX11-CU-NEXT:    flat_store_b32 v[0:1], v2
-; GFX11-CU-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-CU-NEXT:    s_endpgm
     ptr %out, i32 %in, i32 %old) {
 entry:
   %gep = getelementptr i32, ptr %out, i32 4
   %val = cmpxchg volatile ptr %gep, i32 %old, i32 %in syncscope("singlethread") release seq_cst
   %val0 = extractvalue { i32, i1 } %val, 0
   store i32 %val0, ptr %out, align 4
   ret void

 ; GFX11-WGP:       ; %bb.0: ; %entry
 ; GFX11-WGP-NEXT:    s_load_b128 s[0:3], s[0:1], 0x0
 ; GFX11-WGP-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-WGP-NEXT:    v_dual_mov_b32 v0, s0 :: v_dual_mov_b32 v1, s1
 ; GFX11-WGP-NEXT:    v_dual_mov_b32 v2, s2 :: v_dual_mov_b32 v3, s3
 ; GFX11-WGP-NEXT:    flat_atomic_cmpswap_b32 v2, v[0:1], v[2:3] offset:16 glc
 ; GFX11-WGP-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX11-WGP-NEXT:    flat_store_b32 v[0:1], v2
-; GFX11-WGP-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-WGP-NEXT:    s_endpgm
 ;
 ; GFX11-CU-LABEL: flat_singlethread_acq_rel_seq_cst_ret_cmpxchg:
 ; GFX11-CU:       ; %bb.0: ; %entry
 ; GFX11-CU-NEXT:    s_load_b128 s[0:3], s[0:1], 0x0
 ; GFX11-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v0, s0 :: v_dual_mov_b32 v1, s1
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v2, s2 :: v_dual_mov_b32 v3, s3
 ; GFX11-CU-NEXT:    flat_atomic_cmpswap_b32 v2, v[0:1], v[2:3] offset:16 glc
 ; GFX11-CU-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX11-CU-NEXT:    flat_store_b32 v[0:1], v2
-; GFX11-CU-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-CU-NEXT:    s_endpgm
     ptr %out, i32 %in, i32 %old) {
 entry:
   %gep = getelementptr i32, ptr %out, i32 4
   %val = cmpxchg volatile ptr %gep, i32 %old, i32 %in syncscope("singlethread") acq_rel seq_cst
   %val0 = extractvalue { i32, i1 } %val, 0
   store i32 %val0, ptr %out, align 4
   ret void

 ; GFX11-WGP:       ; %bb.0: ; %entry
 ; GFX11-WGP-NEXT:    s_load_b128 s[0:3], s[0:1], 0x0
 ; GFX11-WGP-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-WGP-NEXT:    v_dual_mov_b32 v0, s0 :: v_dual_mov_b32 v1, s1
 ; GFX11-WGP-NEXT:    v_dual_mov_b32 v2, s2 :: v_dual_mov_b32 v3, s3
 ; GFX11-WGP-NEXT:    flat_atomic_cmpswap_b32 v2, v[0:1], v[2:3] offset:16 glc
 ; GFX11-WGP-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX11-WGP-NEXT:    flat_store_b32 v[0:1], v2
-; GFX11-WGP-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-WGP-NEXT:    s_endpgm
 ;
 ; GFX11-CU-LABEL: flat_singlethread_seq_cst_seq_cst_ret_cmpxchg:
 ; GFX11-CU:       ; %bb.0: ; %entry
 ; GFX11-CU-NEXT:    s_load_b128 s[0:3], s[0:1], 0x0
 ; GFX11-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v0, s0 :: v_dual_mov_b32 v1, s1
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v2, s2 :: v_dual_mov_b32 v3, s3
 ; GFX11-CU-NEXT:    flat_atomic_cmpswap_b32 v2, v[0:1], v[2:3] offset:16 glc
 ; GFX11-CU-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX11-CU-NEXT:    flat_store_b32 v[0:1], v2
-; GFX11-CU-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-CU-NEXT:    s_endpgm
     ptr %out, i32 %in, i32 %old) {
 entry:
   %gep = getelementptr i32, ptr %out, i32 4
   %val = cmpxchg volatile ptr %gep, i32 %old, i32 %in syncscope("singlethread") seq_cst seq_cst
   %val0 = extractvalue { i32, i1 } %val, 0
   store i32 %val0, ptr %out, align 4
   ret void

 ; GFX11-WGP:       ; %bb.0: ; %entry
 ; GFX11-WGP-NEXT:    s_load_b128 s[0:3], s[0:1], 0x0
 ; GFX11-WGP-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-WGP-NEXT:    v_dual_mov_b32 v0, s0 :: v_dual_mov_b32 v1, s1
 ; GFX11-WGP-NEXT:    flat_load_b32 v2, v[0:1]
 ; GFX11-WGP-NEXT:    v_dual_mov_b32 v0, s2 :: v_dual_mov_b32 v1, s3
 ; GFX11-WGP-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX11-WGP-NEXT:    flat_store_b32 v[0:1], v2
-; GFX11-WGP-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-WGP-NEXT:    s_endpgm
 ;
 ; GFX11-CU-LABEL: flat_singlethread_one_as_unordered_load:
 ; GFX11-CU:       ; %bb.0: ; %entry
 ; GFX11-CU-NEXT:    s_load_b128 s[0:3], s[0:1], 0x0
 ; GFX11-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v0, s0 :: v_dual_mov_b32 v1, s1
 ; GFX11-CU-NEXT:    flat_load_b32 v2, v[0:1]
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v0, s2 :: v_dual_mov_b32 v1, s3
 ; GFX11-CU-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX11-CU-NEXT:    flat_store_b32 v[0:1], v2
-; GFX11-CU-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-CU-NEXT:    s_endpgm
     ptr %in, ptr %out) {
 entry:
   %val = load atomic i32, ptr %in syncscope("singlethread-one-as") unordered, align 4
   store i32 %val, ptr %out
   ret void
 }
 

 ; GFX11-WGP:       ; %bb.0: ; %entry
 ; GFX11-WGP-NEXT:    s_load_b128 s[0:3], s[0:1], 0x0
 ; GFX11-WGP-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-WGP-NEXT:    v_dual_mov_b32 v0, s0 :: v_dual_mov_b32 v1, s1
 ; GFX11-WGP-NEXT:    flat_load_b32 v2, v[0:1]
 ; GFX11-WGP-NEXT:    v_dual_mov_b32 v0, s2 :: v_dual_mov_b32 v1, s3
 ; GFX11-WGP-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX11-WGP-NEXT:    flat_store_b32 v[0:1], v2
-; GFX11-WGP-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-WGP-NEXT:    s_endpgm
 ;
 ; GFX11-CU-LABEL: flat_singlethread_one_as_monotonic_load:
 ; GFX11-CU:       ; %bb.0: ; %entry
 ; GFX11-CU-NEXT:    s_load_b128 s[0:3], s[0:1], 0x0
 ; GFX11-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v0, s0 :: v_dual_mov_b32 v1, s1
 ; GFX11-CU-NEXT:    flat_load_b32 v2, v[0:1]
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v0, s2 :: v_dual_mov_b32 v1, s3
 ; GFX11-CU-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX11-CU-NEXT:    flat_store_b32 v[0:1], v2
-; GFX11-CU-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-CU-NEXT:    s_endpgm
     ptr %in, ptr %out) {
 entry:
   %val = load atomic i32, ptr %in syncscope("singlethread-one-as") monotonic, align 4
   store i32 %val, ptr %out
   ret void
 }
 

 ; GFX11-WGP:       ; %bb.0: ; %entry
 ; GFX11-WGP-NEXT:    s_load_b128 s[0:3], s[0:1], 0x0
 ; GFX11-WGP-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-WGP-NEXT:    v_dual_mov_b32 v0, s0 :: v_dual_mov_b32 v1, s1
 ; GFX11-WGP-NEXT:    flat_load_b32 v2, v[0:1]
 ; GFX11-WGP-NEXT:    v_dual_mov_b32 v0, s2 :: v_dual_mov_b32 v1, s3
 ; GFX11-WGP-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX11-WGP-NEXT:    flat_store_b32 v[0:1], v2
-; GFX11-WGP-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-WGP-NEXT:    s_endpgm
 ;
 ; GFX11-CU-LABEL: flat_singlethread_one_as_acquire_load:
 ; GFX11-CU:       ; %bb.0: ; %entry
 ; GFX11-CU-NEXT:    s_load_b128 s[0:3], s[0:1], 0x0
 ; GFX11-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v0, s0 :: v_dual_mov_b32 v1, s1
 ; GFX11-CU-NEXT:    flat_load_b32 v2, v[0:1]
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v0, s2 :: v_dual_mov_b32 v1, s3
 ; GFX11-CU-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX11-CU-NEXT:    flat_store_b32 v[0:1], v2
-; GFX11-CU-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-CU-NEXT:    s_endpgm
     ptr %in, ptr %out) {
 entry:
   %val = load atomic i32, ptr %in syncscope("singlethread-one-as") acquire, align 4
   store i32 %val, ptr %out
   ret void
 }
 

 ; GFX11-WGP:       ; %bb.0: ; %entry
 ; GFX11-WGP-NEXT:    s_load_b128 s[0:3], s[0:1], 0x0
 ; GFX11-WGP-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-WGP-NEXT:    v_dual_mov_b32 v0, s0 :: v_dual_mov_b32 v1, s1
 ; GFX11-WGP-NEXT:    flat_load_b32 v2, v[0:1]
 ; GFX11-WGP-NEXT:    v_dual_mov_b32 v0, s2 :: v_dual_mov_b32 v1, s3
 ; GFX11-WGP-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX11-WGP-NEXT:    flat_store_b32 v[0:1], v2
-; GFX11-WGP-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-WGP-NEXT:    s_endpgm
 ;
 ; GFX11-CU-LABEL: flat_singlethread_one_as_seq_cst_load:
 ; GFX11-CU:       ; %bb.0: ; %entry
 ; GFX11-CU-NEXT:    s_load_b128 s[0:3], s[0:1], 0x0
 ; GFX11-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v0, s0 :: v_dual_mov_b32 v1, s1
 ; GFX11-CU-NEXT:    flat_load_b32 v2, v[0:1]
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v0, s2 :: v_dual_mov_b32 v1, s3
 ; GFX11-CU-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX11-CU-NEXT:    flat_store_b32 v[0:1], v2
-; GFX11-CU-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-CU-NEXT:    s_endpgm
     ptr %in, ptr %out) {
 entry:
   %val = load atomic i32, ptr %in syncscope("singlethread-one-as") seq_cst, align 4
   store i32 %val, ptr %out
   ret void
 }
 

 ; GFX11-WGP:       ; %bb.0: ; %entry
 ; GFX11-WGP-NEXT:    s_clause 0x1
 ; GFX11-WGP-NEXT:    s_load_b64 s[2:3], s[0:1], 0x8
 ; GFX11-WGP-NEXT:    s_load_b32 s0, s[0:1], 0x0
 ; GFX11-WGP-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-WGP-NEXT:    v_dual_mov_b32 v0, s2 :: v_dual_mov_b32 v1, s3
 ; GFX11-WGP-NEXT:    v_mov_b32_e32 v2, s0
 ; GFX11-WGP-NEXT:    flat_store_b32 v[0:1], v2
-; GFX11-WGP-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-WGP-NEXT:    s_endpgm
 ;
 ; GFX11-CU-LABEL: flat_singlethread_one_as_unordered_store:
 ; GFX11-CU:       ; %bb.0: ; %entry
 ; GFX11-CU-NEXT:    s_clause 0x1
 ; GFX11-CU-NEXT:    s_load_b64 s[2:3], s[0:1], 0x8
 ; GFX11-CU-NEXT:    s_load_b32 s0, s[0:1], 0x0
 ; GFX11-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v0, s2 :: v_dual_mov_b32 v1, s3
 ; GFX11-CU-NEXT:    v_mov_b32_e32 v2, s0
 ; GFX11-CU-NEXT:    flat_store_b32 v[0:1], v2
-; GFX11-CU-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-CU-NEXT:    s_endpgm
     i32 %in, ptr %out) {
 entry:
   store atomic i32 %in, ptr %out syncscope("singlethread-one-as") unordered, align 4
   ret void
 }
 
 define amdgpu_kernel void @flat_singlethread_one_as_monotonic_store(

 ; GFX11-WGP:       ; %bb.0: ; %entry
 ; GFX11-WGP-NEXT:    s_clause 0x1
 ; GFX11-WGP-NEXT:    s_load_b64 s[2:3], s[0:1], 0x8
 ; GFX11-WGP-NEXT:    s_load_b32 s0, s[0:1], 0x0
 ; GFX11-WGP-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-WGP-NEXT:    v_dual_mov_b32 v0, s2 :: v_dual_mov_b32 v1, s3
 ; GFX11-WGP-NEXT:    v_mov_b32_e32 v2, s0
 ; GFX11-WGP-NEXT:    flat_store_b32 v[0:1], v2
-; GFX11-WGP-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-WGP-NEXT:    s_endpgm
 ;
 ; GFX11-CU-LABEL: flat_singlethread_one_as_monotonic_store:
 ; GFX11-CU:       ; %bb.0: ; %entry
 ; GFX11-CU-NEXT:    s_clause 0x1
 ; GFX11-CU-NEXT:    s_load_b64 s[2:3], s[0:1], 0x8
 ; GFX11-CU-NEXT:    s_load_b32 s0, s[0:1], 0x0
 ; GFX11-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v0, s2 :: v_dual_mov_b32 v1, s3
 ; GFX11-CU-NEXT:    v_mov_b32_e32 v2, s0
 ; GFX11-CU-NEXT:    flat_store_b32 v[0:1], v2
-; GFX11-CU-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-CU-NEXT:    s_endpgm
     i32 %in, ptr %out) {
 entry:
   store atomic i32 %in, ptr %out syncscope("singlethread-one-as") monotonic, align 4
   ret void
 }
 
 define amdgpu_kernel void @flat_singlethread_one_as_release_store(

 ; GFX11-WGP:       ; %bb.0: ; %entry
 ; GFX11-WGP-NEXT:    s_clause 0x1
 ; GFX11-WGP-NEXT:    s_load_b64 s[2:3], s[0:1], 0x8
 ; GFX11-WGP-NEXT:    s_load_b32 s0, s[0:1], 0x0
 ; GFX11-WGP-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-WGP-NEXT:    v_dual_mov_b32 v0, s2 :: v_dual_mov_b32 v1, s3
 ; GFX11-WGP-NEXT:    v_mov_b32_e32 v2, s0
 ; GFX11-WGP-NEXT:    flat_store_b32 v[0:1], v2
-; GFX11-WGP-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-WGP-NEXT:    s_endpgm
 ;
 ; GFX11-CU-LABEL: flat_singlethread_one_as_release_store:
 ; GFX11-CU:       ; %bb.0: ; %entry
 ; GFX11-CU-NEXT:    s_clause 0x1
 ; GFX11-CU-NEXT:    s_load_b64 s[2:3], s[0:1], 0x8
 ; GFX11-CU-NEXT:    s_load_b32 s0, s[0:1], 0x0
 ; GFX11-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v0, s2 :: v_dual_mov_b32 v1, s3
 ; GFX11-CU-NEXT:    v_mov_b32_e32 v2, s0
 ; GFX11-CU-NEXT:    flat_store_b32 v[0:1], v2
-; GFX11-CU-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-CU-NEXT:    s_endpgm
     i32 %in, ptr %out) {
 entry:
   store atomic i32 %in, ptr %out syncscope("singlethread-one-as") release, align 4
   ret void
 }
 
 define amdgpu_kernel void @flat_singlethread_one_as_seq_cst_store(

 ; GFX11-WGP:       ; %bb.0: ; %entry
 ; GFX11-WGP-NEXT:    s_clause 0x1
 ; GFX11-WGP-NEXT:    s_load_b64 s[2:3], s[0:1], 0x8
 ; GFX11-WGP-NEXT:    s_load_b32 s0, s[0:1], 0x0
 ; GFX11-WGP-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-WGP-NEXT:    v_dual_mov_b32 v0, s2 :: v_dual_mov_b32 v1, s3
 ; GFX11-WGP-NEXT:    v_mov_b32_e32 v2, s0
 ; GFX11-WGP-NEXT:    flat_store_b32 v[0:1], v2
-; GFX11-WGP-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-WGP-NEXT:    s_endpgm
 ;
 ; GFX11-CU-LABEL: flat_singlethread_one_as_seq_cst_store:
 ; GFX11-CU:       ; %bb.0: ; %entry
 ; GFX11-CU-NEXT:    s_clause 0x1
 ; GFX11-CU-NEXT:    s_load_b64 s[2:3], s[0:1], 0x8
 ; GFX11-CU-NEXT:    s_load_b32 s0, s[0:1], 0x0
 ; GFX11-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v0, s2 :: v_dual_mov_b32 v1, s3
 ; GFX11-CU-NEXT:    v_mov_b32_e32 v2, s0
 ; GFX11-CU-NEXT:    flat_store_b32 v[0:1], v2
-; GFX11-CU-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-CU-NEXT:    s_endpgm
     i32 %in, ptr %out) {
 entry:
   store atomic i32 %in, ptr %out syncscope("singlethread-one-as") seq_cst, align 4
   ret void
 }
 
 define amdgpu_kernel void @flat_singlethread_one_as_monotonic_atomicrmw(

 ; GFX11-WGP:       ; %bb.0: ; %entry
 ; GFX11-WGP-NEXT:    s_clause 0x1
 ; GFX11-WGP-NEXT:    s_load_b64 s[2:3], s[0:1], 0x0
 ; GFX11-WGP-NEXT:    s_load_b32 s0, s[0:1], 0x8
 ; GFX11-WGP-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-WGP-NEXT:    v_dual_mov_b32 v0, s2 :: v_dual_mov_b32 v1, s3
 ; GFX11-WGP-NEXT:    v_mov_b32_e32 v2, s0
 ; GFX11-WGP-NEXT:    flat_atomic_swap_b32 v[0:1], v2
-; GFX11-WGP-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-WGP-NEXT:    s_endpgm
 ;
 ; GFX11-CU-LABEL: flat_singlethread_one_as_monotonic_atomicrmw:
 ; GFX11-CU:       ; %bb.0: ; %entry
 ; GFX11-CU-NEXT:    s_clause 0x1
 ; GFX11-CU-NEXT:    s_load_b64 s[2:3], s[0:1], 0x0
 ; GFX11-CU-NEXT:    s_load_b32 s0, s[0:1], 0x8
 ; GFX11-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v0, s2 :: v_dual_mov_b32 v1, s3
 ; GFX11-CU-NEXT:    v_mov_b32_e32 v2, s0
 ; GFX11-CU-NEXT:    flat_atomic_swap_b32 v[0:1], v2
-; GFX11-CU-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-CU-NEXT:    s_endpgm
     ptr %out, i32 %in) {
 entry:
   %val = atomicrmw volatile xchg ptr %out, i32 %in syncscope("singlethread-one-as") monotonic
   ret void
 }
 
 define amdgpu_kernel void @flat_singlethread_one_as_acquire_atomicrmw(

 ; GFX11-WGP:       ; %bb.0: ; %entry
 ; GFX11-WGP-NEXT:    s_clause 0x1
 ; GFX11-WGP-NEXT:    s_load_b64 s[2:3], s[0:1], 0x0
 ; GFX11-WGP-NEXT:    s_load_b32 s0, s[0:1], 0x8
 ; GFX11-WGP-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-WGP-NEXT:    v_dual_mov_b32 v0, s2 :: v_dual_mov_b32 v1, s3
 ; GFX11-WGP-NEXT:    v_mov_b32_e32 v2, s0
 ; GFX11-WGP-NEXT:    flat_atomic_swap_b32 v[0:1], v2
-; GFX11-WGP-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-WGP-NEXT:    s_endpgm
 ;
 ; GFX11-CU-LABEL: flat_singlethread_one_as_acquire_atomicrmw:
 ; GFX11-CU:       ; %bb.0: ; %entry
 ; GFX11-CU-NEXT:    s_clause 0x1
 ; GFX11-CU-NEXT:    s_load_b64 s[2:3], s[0:1], 0x0
 ; GFX11-CU-NEXT:    s_load_b32 s0, s[0:1], 0x8
 ; GFX11-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v0, s2 :: v_dual_mov_b32 v1, s3
 ; GFX11-CU-NEXT:    v_mov_b32_e32 v2, s0
 ; GFX11-CU-NEXT:    flat_atomic_swap_b32 v[0:1], v2
-; GFX11-CU-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-CU-NEXT:    s_endpgm
     ptr %out, i32 %in) {
 entry:
   %val = atomicrmw volatile xchg ptr %out, i32 %in syncscope("singlethread-one-as") acquire
   ret void
 }
 
 define amdgpu_kernel void @flat_singlethread_one_as_release_atomicrmw(

 ; GFX11-WGP:       ; %bb.0: ; %entry
 ; GFX11-WGP-NEXT:    s_clause 0x1
 ; GFX11-WGP-NEXT:    s_load_b64 s[2:3], s[0:1], 0x0
 ; GFX11-WGP-NEXT:    s_load_b32 s0, s[0:1], 0x8
 ; GFX11-WGP-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-WGP-NEXT:    v_dual_mov_b32 v0, s2 :: v_dual_mov_b32 v1, s3
 ; GFX11-WGP-NEXT:    v_mov_b32_e32 v2, s0
 ; GFX11-WGP-NEXT:    flat_atomic_swap_b32 v[0:1], v2
-; GFX11-WGP-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-WGP-NEXT:    s_endpgm
 ;
 ; GFX11-CU-LABEL: flat_singlethread_one_as_release_atomicrmw:
 ; GFX11-CU:       ; %bb.0: ; %entry
 ; GFX11-CU-NEXT:    s_clause 0x1
 ; GFX11-CU-NEXT:    s_load_b64 s[2:3], s[0:1], 0x0
 ; GFX11-CU-NEXT:    s_load_b32 s0, s[0:1], 0x8
 ; GFX11-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v0, s2 :: v_dual_mov_b32 v1, s3
 ; GFX11-CU-NEXT:    v_mov_b32_e32 v2, s0
 ; GFX11-CU-NEXT:    flat_atomic_swap_b32 v[0:1], v2
-; GFX11-CU-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-CU-NEXT:    s_endpgm
     ptr %out, i32 %in) {
 entry:
   %val = atomicrmw volatile xchg ptr %out, i32 %in syncscope("singlethread-one-as") release
   ret void
 }
 
 define amdgpu_kernel void @flat_singlethread_one_as_acq_rel_atomicrmw(

 ; GFX11-WGP:       ; %bb.0: ; %entry
 ; GFX11-WGP-NEXT:    s_clause 0x1
 ; GFX11-WGP-NEXT:    s_load_b64 s[2:3], s[0:1], 0x0
 ; GFX11-WGP-NEXT:    s_load_b32 s0, s[0:1], 0x8
 ; GFX11-WGP-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-WGP-NEXT:    v_dual_mov_b32 v0, s2 :: v_dual_mov_b32 v1, s3
 ; GFX11-WGP-NEXT:    v_mov_b32_e32 v2, s0
 ; GFX11-WGP-NEXT:    flat_atomic_swap_b32 v[0:1], v2
-; GFX11-WGP-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-WGP-NEXT:    s_endpgm
 ;
 ; GFX11-CU-LABEL: flat_singlethread_one_as_acq_rel_atomicrmw:
 ; GFX11-CU:       ; %bb.0: ; %entry
 ; GFX11-CU-NEXT:    s_clause 0x1
 ; GFX11-CU-NEXT:    s_load_b64 s[2:3], s[0:1], 0x0
 ; GFX11-CU-NEXT:    s_load_b32 s0, s[0:1], 0x8
 ; GFX11-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v0, s2 :: v_dual_mov_b32 v1, s3
 ; GFX11-CU-NEXT:    v_mov_b32_e32 v2, s0
 ; GFX11-CU-NEXT:    flat_atomic_swap_b32 v[0:1], v2
-; GFX11-CU-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-CU-NEXT:    s_endpgm
     ptr %out, i32 %in) {
 entry:
   %val = atomicrmw volatile xchg ptr %out, i32 %in syncscope("singlethread-one-as") acq_rel
   ret void
 }
 
 define amdgpu_kernel void @flat_singlethread_one_as_seq_cst_atomicrmw(

 ; GFX11-WGP:       ; %bb.0: ; %entry
 ; GFX11-WGP-NEXT:    s_clause 0x1
 ; GFX11-WGP-NEXT:    s_load_b64 s[2:3], s[0:1], 0x0
 ; GFX11-WGP-NEXT:    s_load_b32 s0, s[0:1], 0x8
 ; GFX11-WGP-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-WGP-NEXT:    v_dual_mov_b32 v0, s2 :: v_dual_mov_b32 v1, s3
 ; GFX11-WGP-NEXT:    v_mov_b32_e32 v2, s0
 ; GFX11-WGP-NEXT:    flat_atomic_swap_b32 v[0:1], v2
-; GFX11-WGP-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-WGP-NEXT:    s_endpgm
 ;
 ; GFX11-CU-LABEL: flat_singlethread_one_as_seq_cst_atomicrmw:
 ; GFX11-CU:       ; %bb.0: ; %entry
 ; GFX11-CU-NEXT:    s_clause 0x1
 ; GFX11-CU-NEXT:    s_load_b64 s[2:3], s[0:1], 0x0
 ; GFX11-CU-NEXT:    s_load_b32 s0, s[0:1], 0x8
 ; GFX11-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v0, s2 :: v_dual_mov_b32 v1, s3
 ; GFX11-CU-NEXT:    v_mov_b32_e32 v2, s0
 ; GFX11-CU-NEXT:    flat_atomic_swap_b32 v[0:1], v2
-; GFX11-CU-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-CU-NEXT:    s_endpgm
     ptr %out, i32 %in) {
 entry:
   %val = atomicrmw volatile xchg ptr %out, i32 %in syncscope("singlethread-one-as") seq_cst
   ret void
 }
 
 define amdgpu_kernel void @flat_singlethread_one_as_acquire_ret_atomicrmw(

 ; GFX11-WGP-NEXT:    s_load_b64 s[2:3], s[0:1], 0x0
 ; GFX11-WGP-NEXT:    s_load_b32 s0, s[0:1], 0x8
 ; GFX11-WGP-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-WGP-NEXT:    v_dual_mov_b32 v0, s2 :: v_dual_mov_b32 v1, s3
 ; GFX11-WGP-NEXT:    v_mov_b32_e32 v2, s0
 ; GFX11-WGP-NEXT:    flat_atomic_swap_b32 v2, v[0:1], v2 glc
 ; GFX11-WGP-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX11-WGP-NEXT:    flat_store_b32 v[0:1], v2
-; GFX11-WGP-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-WGP-NEXT:    s_endpgm
 ;
 ; GFX11-CU-LABEL: flat_singlethread_one_as_acquire_ret_atomicrmw:
 ; GFX11-CU:       ; %bb.0: ; %entry
 ; GFX11-CU-NEXT:    s_clause 0x1
 ; GFX11-CU-NEXT:    s_load_b64 s[2:3], s[0:1], 0x0
 ; GFX11-CU-NEXT:    s_load_b32 s0, s[0:1], 0x8
 ; GFX11-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v0, s2 :: v_dual_mov_b32 v1, s3
 ; GFX11-CU-NEXT:    v_mov_b32_e32 v2, s0
 ; GFX11-CU-NEXT:    flat_atomic_swap_b32 v2, v[0:1], v2 glc
 ; GFX11-CU-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX11-CU-NEXT:    flat_store_b32 v[0:1], v2
-; GFX11-CU-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-CU-NEXT:    s_endpgm
     ptr %out, i32 %in) {
 entry:
   %val = atomicrmw volatile xchg ptr %out, i32 %in syncscope("singlethread-one-as") acquire
   store i32 %val, ptr %out, align 4
   ret void
 }
 

 ; GFX11-WGP-NEXT:    s_load_b64 s[2:3], s[0:1], 0x0
 ; GFX11-WGP-NEXT:    s_load_b32 s0, s[0:1], 0x8
 ; GFX11-WGP-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-WGP-NEXT:    v_dual_mov_b32 v0, s2 :: v_dual_mov_b32 v1, s3
 ; GFX11-WGP-NEXT:    v_mov_b32_e32 v2, s0
 ; GFX11-WGP-NEXT:    flat_atomic_swap_b32 v2, v[0:1], v2 glc
 ; GFX11-WGP-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX11-WGP-NEXT:    flat_store_b32 v[0:1], v2
-; GFX11-WGP-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-WGP-NEXT:    s_endpgm
 ;
 ; GFX11-CU-LABEL: flat_singlethread_one_as_acq_rel_ret_atomicrmw:
 ; GFX11-CU:       ; %bb.0: ; %entry
 ; GFX11-CU-NEXT:    s_clause 0x1
 ; GFX11-CU-NEXT:    s_load_b64 s[2:3], s[0:1], 0x0
 ; GFX11-CU-NEXT:    s_load_b32 s0, s[0:1], 0x8
 ; GFX11-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v0, s2 :: v_dual_mov_b32 v1, s3
 ; GFX11-CU-NEXT:    v_mov_b32_e32 v2, s0
 ; GFX11-CU-NEXT:    flat_atomic_swap_b32 v2, v[0:1], v2 glc
 ; GFX11-CU-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX11-CU-NEXT:    flat_store_b32 v[0:1], v2
-; GFX11-CU-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-CU-NEXT:    s_endpgm
     ptr %out, i32 %in) {
 entry:
   %val = atomicrmw volatile xchg ptr %out, i32 %in syncscope("singlethread-one-as") acq_rel
   store i32 %val, ptr %out, align 4
   ret void
 }
 

 ; GFX11-WGP-NEXT:    s_load_b64 s[2:3], s[0:1], 0x0
 ; GFX11-WGP-NEXT:    s_load_b32 s0, s[0:1], 0x8
 ; GFX11-WGP-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-WGP-NEXT:    v_dual_mov_b32 v0, s2 :: v_dual_mov_b32 v1, s3
 ; GFX11-WGP-NEXT:    v_mov_b32_e32 v2, s0
 ; GFX11-WGP-NEXT:    flat_atomic_swap_b32 v2, v[0:1], v2 glc
 ; GFX11-WGP-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX11-WGP-NEXT:    flat_store_b32 v[0:1], v2
-; GFX11-WGP-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-WGP-NEXT:    s_endpgm
 ;
 ; GFX11-CU-LABEL: flat_singlethread_one_as_seq_cst_ret_atomicrmw:
 ; GFX11-CU:       ; %bb.0: ; %entry
 ; GFX11-CU-NEXT:    s_clause 0x1
 ; GFX11-CU-NEXT:    s_load_b64 s[2:3], s[0:1], 0x0
 ; GFX11-CU-NEXT:    s_load_b32 s0, s[0:1], 0x8
 ; GFX11-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v0, s2 :: v_dual_mov_b32 v1, s3
 ; GFX11-CU-NEXT:    v_mov_b32_e32 v2, s0
 ; GFX11-CU-NEXT:    flat_atomic_swap_b32 v2, v[0:1], v2 glc
 ; GFX11-CU-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX11-CU-NEXT:    flat_store_b32 v[0:1], v2
-; GFX11-CU-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-CU-NEXT:    s_endpgm
     ptr %out, i32 %in) {
 entry:
   %val = atomicrmw volatile xchg ptr %out, i32 %in syncscope("singlethread-one-as") seq_cst
   store i32 %val, ptr %out, align 4
   ret void
 }
 

 ;
 ; GFX11-WGP-LABEL: flat_singlethread_one_as_monotonic_monotonic_cmpxchg:
 ; GFX11-WGP:       ; %bb.0: ; %entry
 ; GFX11-WGP-NEXT:    s_load_b128 s[0:3], s[0:1], 0x0
 ; GFX11-WGP-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-WGP-NEXT:    v_dual_mov_b32 v0, s0 :: v_dual_mov_b32 v1, s1
 ; GFX11-WGP-NEXT:    v_dual_mov_b32 v2, s2 :: v_dual_mov_b32 v3, s3
 ; GFX11-WGP-NEXT:    flat_atomic_cmpswap_b32 v[0:1], v[2:3] offset:16
-; GFX11-WGP-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-WGP-NEXT:    s_endpgm
 ;
 ; GFX11-CU-LABEL: flat_singlethread_one_as_monotonic_monotonic_cmpxchg:
 ; GFX11-CU:       ; %bb.0: ; %entry
 ; GFX11-CU-NEXT:    s_load_b128 s[0:3], s[0:1], 0x0
 ; GFX11-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v0, s0 :: v_dual_mov_b32 v1, s1
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v2, s2 :: v_dual_mov_b32 v3, s3
 ; GFX11-CU-NEXT:    flat_atomic_cmpswap_b32 v[0:1], v[2:3] offset:16
-; GFX11-CU-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-CU-NEXT:    s_endpgm
     ptr %out, i32 %in, i32 %old) {
 entry:
   %gep = getelementptr i32, ptr %out, i32 4
   %val = cmpxchg volatile ptr %gep, i32 %old, i32 %in syncscope("singlethread-one-as") monotonic monotonic
   ret void
 }
 

 ;
 ; GFX11-WGP-LABEL: flat_singlethread_one_as_acquire_monotonic_cmpxchg:
 ; GFX11-WGP:       ; %bb.0: ; %entry
 ; GFX11-WGP-NEXT:    s_load_b128 s[0:3], s[0:1], 0x0
 ; GFX11-WGP-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-WGP-NEXT:    v_dual_mov_b32 v0, s0 :: v_dual_mov_b32 v1, s1
 ; GFX11-WGP-NEXT:    v_dual_mov_b32 v2, s2 :: v_dual_mov_b32 v3, s3
 ; GFX11-WGP-NEXT:    flat_atomic_cmpswap_b32 v[0:1], v[2:3] offset:16
-; GFX11-WGP-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-WGP-NEXT:    s_endpgm
 ;
 ; GFX11-CU-LABEL: flat_singlethread_one_as_acquire_monotonic_cmpxchg:
 ; GFX11-CU:       ; %bb.0: ; %entry
 ; GFX11-CU-NEXT:    s_load_b128 s[0:3], s[0:1], 0x0
 ; GFX11-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v0, s0 :: v_dual_mov_b32 v1, s1
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v2, s2 :: v_dual_mov_b32 v3, s3
 ; GFX11-CU-NEXT:    flat_atomic_cmpswap_b32 v[0:1], v[2:3] offset:16
-; GFX11-CU-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-CU-NEXT:    s_endpgm
     ptr %out, i32 %in, i32 %old) {
 entry:
   %gep = getelementptr i32, ptr %out, i32 4
   %val = cmpxchg volatile ptr %gep, i32 %old, i32 %in syncscope("singlethread-one-as") acquire monotonic
   ret void
 }
 

 ;
 ; GFX11-WGP-LABEL: flat_singlethread_one_as_release_monotonic_cmpxchg:
 ; GFX11-WGP:       ; %bb.0: ; %entry
 ; GFX11-WGP-NEXT:    s_load_b128 s[0:3], s[0:1], 0x0
 ; GFX11-WGP-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-WGP-NEXT:    v_dual_mov_b32 v0, s0 :: v_dual_mov_b32 v1, s1
 ; GFX11-WGP-NEXT:    v_dual_mov_b32 v2, s2 :: v_dual_mov_b32 v3, s3
 ; GFX11-WGP-NEXT:    flat_atomic_cmpswap_b32 v[0:1], v[2:3] offset:16
-; GFX11-WGP-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-WGP-NEXT:    s_endpgm
 ;
 ; GFX11-CU-LABEL: flat_singlethread_one_as_release_monotonic_cmpxchg:
 ; GFX11-CU:       ; %bb.0: ; %entry
 ; GFX11-CU-NEXT:    s_load_b128 s[0:3], s[0:1], 0x0
 ; GFX11-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v0, s0 :: v_dual_mov_b32 v1, s1
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v2, s2 :: v_dual_mov_b32 v3, s3
 ; GFX11-CU-NEXT:    flat_atomic_cmpswap_b32 v[0:1], v[2:3] offset:16
-; GFX11-CU-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-CU-NEXT:    s_endpgm
     ptr %out, i32 %in, i32 %old) {
 entry:
   %gep = getelementptr i32, ptr %out, i32 4
   %val = cmpxchg volatile ptr %gep, i32 %old, i32 %in syncscope("singlethread-one-as") release monotonic
   ret void
 }
 

 ;
 ; GFX11-WGP-LABEL: flat_singlethread_one_as_acq_rel_monotonic_cmpxchg:
 ; GFX11-WGP:       ; %bb.0: ; %entry
 ; GFX11-WGP-NEXT:    s_load_b128 s[0:3], s[0:1], 0x0
 ; GFX11-WGP-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-WGP-NEXT:    v_dual_mov_b32 v0, s0 :: v_dual_mov_b32 v1, s1
 ; GFX11-WGP-NEXT:    v_dual_mov_b32 v2, s2 :: v_dual_mov_b32 v3, s3
 ; GFX11-WGP-NEXT:    flat_atomic_cmpswap_b32 v[0:1], v[2:3] offset:16
-; GFX11-WGP-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-WGP-NEXT:    s_endpgm
 ;
 ; GFX11-CU-LABEL: flat_singlethread_one_as_acq_rel_monotonic_cmpxchg:
 ; GFX11-CU:       ; %bb.0: ; %entry
 ; GFX11-CU-NEXT:    s_load_b128 s[0:3], s[0:1], 0x0
 ; GFX11-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v0, s0 :: v_dual_mov_b32 v1, s1
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v2, s2 :: v_dual_mov_b32 v3, s3
 ; GFX11-CU-NEXT:    flat_atomic_cmpswap_b32 v[0:1], v[2:3] offset:16
-; GFX11-CU-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-CU-NEXT:    s_endpgm
     ptr %out, i32 %in, i32 %old) {
 entry:
   %gep = getelementptr i32, ptr %out, i32 4
   %val = cmpxchg volatile ptr %gep, i32 %old, i32 %in syncscope("singlethread-one-as") acq_rel monotonic
   ret void
 }
 

 ;
 ; GFX11-WGP-LABEL: flat_singlethread_one_as_seq_cst_monotonic_cmpxchg:
 ; GFX11-WGP:       ; %bb.0: ; %entry
 ; GFX11-WGP-NEXT:    s_load_b128 s[0:3], s[0:1], 0x0
 ; GFX11-WGP-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-WGP-NEXT:    v_dual_mov_b32 v0, s0 :: v_dual_mov_b32 v1, s1
 ; GFX11-WGP-NEXT:    v_dual_mov_b32 v2, s2 :: v_dual_mov_b32 v3, s3
 ; GFX11-WGP-NEXT:    flat_atomic_cmpswap_b32 v[0:1], v[2:3] offset:16
-; GFX11-WGP-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-WGP-NEXT:    s_endpgm
 ;
 ; GFX11-CU-LABEL: flat_singlethread_one_as_seq_cst_monotonic_cmpxchg:
 ; GFX11-CU:       ; %bb.0: ; %entry
 ; GFX11-CU-NEXT:    s_load_b128 s[0:3], s[0:1], 0x0
 ; GFX11-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v0, s0 :: v_dual_mov_b32 v1, s1
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v2, s2 :: v_dual_mov_b32 v3, s3
 ; GFX11-CU-NEXT:    flat_atomic_cmpswap_b32 v[0:1], v[2:3] offset:16
-; GFX11-CU-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-CU-NEXT:    s_endpgm
     ptr %out, i32 %in, i32 %old) {
 entry:
   %gep = getelementptr i32, ptr %out, i32 4
   %val = cmpxchg volatile ptr %gep, i32 %old, i32 %in syncscope("singlethread-one-as") seq_cst monotonic
   ret void
 }
 

 ;
 ; GFX11-WGP-LABEL: flat_singlethread_one_as_monotonic_acquire_cmpxchg:
 ; GFX11-WGP:       ; %bb.0: ; %entry
 ; GFX11-WGP-NEXT:    s_load_b128 s[0:3], s[0:1], 0x0
 ; GFX11-WGP-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-WGP-NEXT:    v_dual_mov_b32 v0, s0 :: v_dual_mov_b32 v1, s1
 ; GFX11-WGP-NEXT:    v_dual_mov_b32 v2, s2 :: v_dual_mov_b32 v3, s3
 ; GFX11-WGP-NEXT:    flat_atomic_cmpswap_b32 v[0:1], v[2:3] offset:16
-; GFX11-WGP-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-WGP-NEXT:    s_endpgm
 ;
 ; GFX11-CU-LABEL: flat_singlethread_one_as_monotonic_acquire_cmpxchg:
 ; GFX11-CU:       ; %bb.0: ; %entry
 ; GFX11-CU-NEXT:    s_load_b128 s[0:3], s[0:1], 0x0
 ; GFX11-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v0, s0 :: v_dual_mov_b32 v1, s1
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v2, s2 :: v_dual_mov_b32 v3, s3
 ; GFX11-CU-NEXT:    flat_atomic_cmpswap_b32 v[0:1], v[2:3] offset:16
-; GFX11-CU-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-CU-NEXT:    s_endpgm
     ptr %out, i32 %in, i32 %old) {
 entry:
   %gep = getelementptr i32, ptr %out, i32 4
   %val = cmpxchg volatile ptr %gep, i32 %old, i32 %in syncscope("singlethread-one-as") monotonic acquire
   ret void
 }
 

 ;
 ; GFX11-WGP-LABEL: flat_singlethread_one_as_acquire_acquire_cmpxchg:
 ; GFX11-WGP:       ; %bb.0: ; %entry
 ; GFX11-WGP-NEXT:    s_load_b128 s[0:3], s[0:1], 0x0
 ; GFX11-WGP-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-WGP-NEXT:    v_dual_mov_b32 v0, s0 :: v_dual_mov_b32 v1, s1
 ; GFX11-WGP-NEXT:    v_dual_mov_b32 v2, s2 :: v_dual_mov_b32 v3, s3
 ; GFX11-WGP-NEXT:    flat_atomic_cmpswap_b32 v[0:1], v[2:3] offset:16
-; GFX11-WGP-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-WGP-NEXT:    s_endpgm
 ;
 ; GFX11-CU-LABEL: flat_singlethread_one_as_acquire_acquire_cmpxchg:
 ; GFX11-CU:       ; %bb.0: ; %entry
 ; GFX11-CU-NEXT:    s_load_b128 s[0:3], s[0:1], 0x0
 ; GFX11-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v0, s0 :: v_dual_mov_b32 v1, s1
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v2, s2 :: v_dual_mov_b32 v3, s3
 ; GFX11-CU-NEXT:    flat_atomic_cmpswap_b32 v[0:1], v[2:3] offset:16
-; GFX11-CU-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-CU-NEXT:    s_endpgm
     ptr %out, i32 %in, i32 %old) {
 entry:
   %gep = getelementptr i32, ptr %out, i32 4
   %val = cmpxchg volatile ptr %gep, i32 %old, i32 %in syncscope("singlethread-one-as") acquire acquire
   ret void
 }
 

 ;
 ; GFX11-WGP-LABEL: flat_singlethread_one_as_release_acquire_cmpxchg:
 ; GFX11-WGP:       ; %bb.0: ; %entry
 ; GFX11-WGP-NEXT:    s_load_b128 s[0:3], s[0:1], 0x0
 ; GFX11-WGP-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-WGP-NEXT:    v_dual_mov_b32 v0, s0 :: v_dual_mov_b32 v1, s1
 ; GFX11-WGP-NEXT:    v_dual_mov_b32 v2, s2 :: v_dual_mov_b32 v3, s3
 ; GFX11-WGP-NEXT:    flat_atomic_cmpswap_b32 v[0:1], v[2:3] offset:16
-; GFX11-WGP-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-WGP-NEXT:    s_endpgm
 ;
 ; GFX11-CU-LABEL: flat_singlethread_one_as_release_acquire_cmpxchg:
 ; GFX11-CU:       ; %bb.0: ; %entry
 ; GFX11-CU-NEXT:    s_load_b128 s[0:3], s[0:1], 0x0
 ; GFX11-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v0, s0 :: v_dual_mov_b32 v1, s1
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v2, s2 :: v_dual_mov_b32 v3, s3
 ; GFX11-CU-NEXT:    flat_atomic_cmpswap_b32 v[0:1], v[2:3] offset:16
-; GFX11-CU-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-CU-NEXT:    s_endpgm
     ptr %out, i32 %in, i32 %old) {
 entry:
   %gep = getelementptr i32, ptr %out, i32 4
   %val = cmpxchg volatile ptr %gep, i32 %old, i32 %in syncscope("singlethread-one-as") release acquire
   ret void
 }
 

 ;
 ; GFX11-WGP-LABEL: flat_singlethread_one_as_acq_rel_acquire_cmpxchg:
 ; GFX11-WGP:       ; %bb.0: ; %entry
 ; GFX11-WGP-NEXT:    s_load_b128 s[0:3], s[0:1], 0x0
 ; GFX11-WGP-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-WGP-NEXT:    v_dual_mov_b32 v0, s0 :: v_dual_mov_b32 v1, s1
 ; GFX11-WGP-NEXT:    v_dual_mov_b32 v2, s2 :: v_dual_mov_b32 v3, s3
 ; GFX11-WGP-NEXT:    flat_atomic_cmpswap_b32 v[0:1], v[2:3] offset:16
-; GFX11-WGP-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-WGP-NEXT:    s_endpgm
 ;
 ; GFX11-CU-LABEL: flat_singlethread_one_as_acq_rel_acquire_cmpxchg:
 ; GFX11-CU:       ; %bb.0: ; %entry
 ; GFX11-CU-NEXT:    s_load_b128 s[0:3], s[0:1], 0x0
 ; GFX11-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v0, s0 :: v_dual_mov_b32 v1, s1
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v2, s2 :: v_dual_mov_b32 v3, s3
 ; GFX11-CU-NEXT:    flat_atomic_cmpswap_b32 v[0:1], v[2:3] offset:16
-; GFX11-CU-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-CU-NEXT:    s_endpgm
     ptr %out, i32 %in, i32 %old) {
 entry:
   %gep = getelementptr i32, ptr %out, i32 4
   %val = cmpxchg volatile ptr %gep, i32 %old, i32 %in syncscope("singlethread-one-as") acq_rel acquire
   ret void
 }
 

 ;
 ; GFX11-WGP-LABEL: flat_singlethread_one_as_seq_cst_acquire_cmpxchg:
 ; GFX11-WGP:       ; %bb.0: ; %entry
 ; GFX11-WGP-NEXT:    s_load_b128 s[0:3], s[0:1], 0x0
 ; GFX11-WGP-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-WGP-NEXT:    v_dual_mov_b32 v0, s0 :: v_dual_mov_b32 v1, s1
 ; GFX11-WGP-NEXT:    v_dual_mov_b32 v2, s2 :: v_dual_mov_b32 v3, s3
 ; GFX11-WGP-NEXT:    flat_atomic_cmpswap_b32 v[0:1], v[2:3] offset:16
-; GFX11-WGP-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-WGP-NEXT:    s_endpgm
 ;
 ; GFX11-CU-LABEL: flat_singlethread_one_as_seq_cst_acquire_cmpxchg:
 ; GFX11-CU:       ; %bb.0: ; %entry
 ; GFX11-CU-NEXT:    s_load_b128 s[0:3], s[0:1], 0x0
 ; GFX11-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v0, s0 :: v_dual_mov_b32 v1, s1
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v2, s2 :: v_dual_mov_b32 v3, s3
 ; GFX11-CU-NEXT:    flat_atomic_cmpswap_b32 v[0:1], v[2:3] offset:16
-; GFX11-CU-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-CU-NEXT:    s_endpgm
     ptr %out, i32 %in, i32 %old) {
 entry:
   %gep = getelementptr i32, ptr %out, i32 4
   %val = cmpxchg volatile ptr %gep, i32 %old, i32 %in syncscope("singlethread-one-as") seq_cst acquire
   ret void
 }
 

 ;
 ; GFX11-WGP-LABEL: flat_singlethread_one_as_monotonic_seq_cst_cmpxchg:
 ; GFX11-WGP:       ; %bb.0: ; %entry
 ; GFX11-WGP-NEXT:    s_load_b128 s[0:3], s[0:1], 0x0
 ; GFX11-WGP-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-WGP-NEXT:    v_dual_mov_b32 v0, s0 :: v_dual_mov_b32 v1, s1
 ; GFX11-WGP-NEXT:    v_dual_mov_b32 v2, s2 :: v_dual_mov_b32 v3, s3
 ; GFX11-WGP-NEXT:    flat_atomic_cmpswap_b32 v[0:1], v[2:3] offset:16
-; GFX11-WGP-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-WGP-NEXT:    s_endpgm
 ;
 ; GFX11-CU-LABEL: flat_singlethread_one_as_monotonic_seq_cst_cmpxchg:
 ; GFX11-CU:       ; %bb.0: ; %entry
 ; GFX11-CU-NEXT:    s_load_b128 s[0:3], s[0:1], 0x0
 ; GFX11-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v0, s0 :: v_dual_mov_b32 v1, s1
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v2, s2 :: v_dual_mov_b32 v3, s3
 ; GFX11-CU-NEXT:    flat_atomic_cmpswap_b32 v[0:1], v[2:3] offset:16
-; GFX11-CU-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-CU-NEXT:    s_endpgm
     ptr %out, i32 %in, i32 %old) {
 entry:
   %gep = getelementptr i32, ptr %out, i32 4
   %val = cmpxchg volatile ptr %gep, i32 %old, i32 %in syncscope("singlethread-one-as") monotonic seq_cst
   ret void
 }
 

 ;
 ; GFX11-WGP-LABEL: flat_singlethread_one_as_acquire_seq_cst_cmpxchg:
 ; GFX11-WGP:       ; %bb.0: ; %entry
 ; GFX11-WGP-NEXT:    s_load_b128 s[0:3], s[0:1], 0x0
 ; GFX11-WGP-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-WGP-NEXT:    v_dual_mov_b32 v0, s0 :: v_dual_mov_b32 v1, s1
 ; GFX11-WGP-NEXT:    v_dual_mov_b32 v2, s2 :: v_dual_mov_b32 v3, s3
 ; GFX11-WGP-NEXT:    flat_atomic_cmpswap_b32 v[0:1], v[2:3] offset:16
-; GFX11-WGP-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-WGP-NEXT:    s_endpgm
 ;
 ; GFX11-CU-LABEL: flat_singlethread_one_as_acquire_seq_cst_cmpxchg:
 ; GFX11-CU:       ; %bb.0: ; %entry
 ; GFX11-CU-NEXT:    s_load_b128 s[0:3], s[0:1], 0x0
 ; GFX11-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v0, s0 :: v_dual_mov_b32 v1, s1
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v2, s2 :: v_dual_mov_b32 v3, s3
 ; GFX11-CU-NEXT:    flat_atomic_cmpswap_b32 v[0:1], v[2:3] offset:16
-; GFX11-CU-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-CU-NEXT:    s_endpgm
     ptr %out, i32 %in, i32 %old) {
 entry:
   %gep = getelementptr i32, ptr %out, i32 4
   %val = cmpxchg volatile ptr %gep, i32 %old, i32 %in syncscope("singlethread-one-as") acquire seq_cst
   ret void
 }
 

 ;
 ; GFX11-WGP-LABEL: flat_singlethread_one_as_release_seq_cst_cmpxchg:
 ; GFX11-WGP:       ; %bb.0: ; %entry
 ; GFX11-WGP-NEXT:    s_load_b128 s[0:3], s[0:1], 0x0
 ; GFX11-WGP-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-WGP-NEXT:    v_dual_mov_b32 v0, s0 :: v_dual_mov_b32 v1, s1
 ; GFX11-WGP-NEXT:    v_dual_mov_b32 v2, s2 :: v_dual_mov_b32 v3, s3
 ; GFX11-WGP-NEXT:    flat_atomic_cmpswap_b32 v[0:1], v[2:3] offset:16
-; GFX11-WGP-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-WGP-NEXT:    s_endpgm
 ;
 ; GFX11-CU-LABEL: flat_singlethread_one_as_release_seq_cst_cmpxchg:
 ; GFX11-CU:       ; %bb.0: ; %entry
 ; GFX11-CU-NEXT:    s_load_b128 s[0:3], s[0:1], 0x0
 ; GFX11-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v0, s0 :: v_dual_mov_b32 v1, s1
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v2, s2 :: v_dual_mov_b32 v3, s3
 ; GFX11-CU-NEXT:    flat_atomic_cmpswap_b32 v[0:1], v[2:3] offset:16
-; GFX11-CU-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-CU-NEXT:    s_endpgm
     ptr %out, i32 %in, i32 %old) {
 entry:
   %gep = getelementptr i32, ptr %out, i32 4
   %val = cmpxchg volatile ptr %gep, i32 %old, i32 %in syncscope("singlethread-one-as") release seq_cst
   ret void
 }
 

 ;
 ; GFX11-WGP-LABEL: flat_singlethread_one_as_acq_rel_seq_cst_cmpxchg:
 ; GFX11-WGP:       ; %bb.0: ; %entry
 ; GFX11-WGP-NEXT:    s_load_b128 s[0:3], s[0:1], 0x0
 ; GFX11-WGP-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-WGP-NEXT:    v_dual_mov_b32 v0, s0 :: v_dual_mov_b32 v1, s1
 ; GFX11-WGP-NEXT:    v_dual_mov_b32 v2, s2 :: v_dual_mov_b32 v3, s3
 ; GFX11-WGP-NEXT:    flat_atomic_cmpswap_b32 v[0:1], v[2:3] offset:16
-; GFX11-WGP-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-WGP-NEXT:    s_endpgm
 ;
 ; GFX11-CU-LABEL: flat_singlethread_one_as_acq_rel_seq_cst_cmpxchg:
 ; GFX11-CU:       ; %bb.0: ; %entry
 ; GFX11-CU-NEXT:    s_load_b128 s[0:3], s[0:1], 0x0
 ; GFX11-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v0, s0 :: v_dual_mov_b32 v1, s1
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v2, s2 :: v_dual_mov_b32 v3, s3
 ; GFX11-CU-NEXT:    flat_atomic_cmpswap_b32 v[0:1], v[2:3] offset:16
-; GFX11-CU-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-CU-NEXT:    s_endpgm
     ptr %out, i32 %in, i32 %old) {
 entry:
   %gep = getelementptr i32, ptr %out, i32 4
   %val = cmpxchg volatile ptr %gep, i32 %old, i32 %in syncscope("singlethread-one-as") acq_rel seq_cst
   ret void
 }
 

 ;
 ; GFX11-WGP-LABEL: flat_singlethread_one_as_seq_cst_seq_cst_cmpxchg:
 ; GFX11-WGP:       ; %bb.0: ; %entry
 ; GFX11-WGP-NEXT:    s_load_b128 s[0:3], s[0:1], 0x0
 ; GFX11-WGP-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-WGP-NEXT:    v_dual_mov_b32 v0, s0 :: v_dual_mov_b32 v1, s1
 ; GFX11-WGP-NEXT:    v_dual_mov_b32 v2, s2 :: v_dual_mov_b32 v3, s3
 ; GFX11-WGP-NEXT:    flat_atomic_cmpswap_b32 v[0:1], v[2:3] offset:16
-; GFX11-WGP-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-WGP-NEXT:    s_endpgm
 ;
 ; GFX11-CU-LABEL: flat_singlethread_one_as_seq_cst_seq_cst_cmpxchg:
 ; GFX11-CU:       ; %bb.0: ; %entry
 ; GFX11-CU-NEXT:    s_load_b128 s[0:3], s[0:1], 0x0
 ; GFX11-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v0, s0 :: v_dual_mov_b32 v1, s1
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v2, s2 :: v_dual_mov_b32 v3, s3
 ; GFX11-CU-NEXT:    flat_atomic_cmpswap_b32 v[0:1], v[2:3] offset:16
-; GFX11-CU-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-CU-NEXT:    s_endpgm
     ptr %out, i32 %in, i32 %old) {
 entry:
   %gep = getelementptr i32, ptr %out, i32 4
   %val = cmpxchg volatile ptr %gep, i32 %old, i32 %in syncscope("singlethread-one-as") seq_cst seq_cst
   ret void
 }
 

 ; GFX11-WGP:       ; %bb.0: ; %entry
 ; GFX11-WGP-NEXT:    s_load_b128 s[0:3], s[0:1], 0x0
 ; GFX11-WGP-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-WGP-NEXT:    v_dual_mov_b32 v0, s0 :: v_dual_mov_b32 v1, s1
 ; GFX11-WGP-NEXT:    v_dual_mov_b32 v2, s2 :: v_dual_mov_b32 v3, s3
 ; GFX11-WGP-NEXT:    flat_atomic_cmpswap_b32 v2, v[0:1], v[2:3] offset:16 glc
 ; GFX11-WGP-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX11-WGP-NEXT:    flat_store_b32 v[0:1], v2
-; GFX11-WGP-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-WGP-NEXT:    s_endpgm
 ;
 ; GFX11-CU-LABEL: flat_singlethread_one_as_monotonic_monotonic_ret_cmpxchg:
 ; GFX11-CU:       ; %bb.0: ; %entry
 ; GFX11-CU-NEXT:    s_load_b128 s[0:3], s[0:1], 0x0
 ; GFX11-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v0, s0 :: v_dual_mov_b32 v1, s1
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v2, s2 :: v_dual_mov_b32 v3, s3
 ; GFX11-CU-NEXT:    flat_atomic_cmpswap_b32 v2, v[0:1], v[2:3] offset:16 glc
 ; GFX11-CU-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX11-CU-NEXT:    flat_store_b32 v[0:1], v2
-; GFX11-CU-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-CU-NEXT:    s_endpgm
     ptr %out, i32 %in, i32 %old) {
 entry:
   %gep = getelementptr i32, ptr %out, i32 4
   %val = cmpxchg volatile ptr %gep, i32 %old, i32 %in syncscope("singlethread-one-as") monotonic monotonic
   %val0 = extractvalue { i32, i1 } %val, 0
   store i32 %val0, ptr %out, align 4
   ret void

 ; GFX11-WGP:       ; %bb.0: ; %entry
 ; GFX11-WGP-NEXT:    s_load_b128 s[0:3], s[0:1], 0x0
 ; GFX11-WGP-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-WGP-NEXT:    v_dual_mov_b32 v0, s0 :: v_dual_mov_b32 v1, s1
 ; GFX11-WGP-NEXT:    v_dual_mov_b32 v2, s2 :: v_dual_mov_b32 v3, s3
 ; GFX11-WGP-NEXT:    flat_atomic_cmpswap_b32 v2, v[0:1], v[2:3] offset:16 glc
 ; GFX11-WGP-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX11-WGP-NEXT:    flat_store_b32 v[0:1], v2
-; GFX11-WGP-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-WGP-NEXT:    s_endpgm
 ;
 ; GFX11-CU-LABEL: flat_singlethread_one_as_acquire_monotonic_ret_cmpxchg:
 ; GFX11-CU:       ; %bb.0: ; %entry
 ; GFX11-CU-NEXT:    s_load_b128 s[0:3], s[0:1], 0x0
 ; GFX11-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v0, s0 :: v_dual_mov_b32 v1, s1
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v2, s2 :: v_dual_mov_b32 v3, s3
 ; GFX11-CU-NEXT:    flat_atomic_cmpswap_b32 v2, v[0:1], v[2:3] offset:16 glc
 ; GFX11-CU-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX11-CU-NEXT:    flat_store_b32 v[0:1], v2
-; GFX11-CU-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-CU-NEXT:    s_endpgm
     ptr %out, i32 %in, i32 %old) {
 entry:
   %gep = getelementptr i32, ptr %out, i32 4
   %val = cmpxchg volatile ptr %gep, i32 %old, i32 %in syncscope("singlethread-one-as") acquire monotonic
   %val0 = extractvalue { i32, i1 } %val, 0
   store i32 %val0, ptr %out, align 4
   ret void

 ; GFX11-WGP:       ; %bb.0: ; %entry
 ; GFX11-WGP-NEXT:    s_load_b128 s[0:3], s[0:1], 0x0
 ; GFX11-WGP-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-WGP-NEXT:    v_dual_mov_b32 v0, s0 :: v_dual_mov_b32 v1, s1
 ; GFX11-WGP-NEXT:    v_dual_mov_b32 v2, s2 :: v_dual_mov_b32 v3, s3
 ; GFX11-WGP-NEXT:    flat_atomic_cmpswap_b32 v2, v[0:1], v[2:3] offset:16 glc
 ; GFX11-WGP-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX11-WGP-NEXT:    flat_store_b32 v[0:1], v2
-; GFX11-WGP-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-WGP-NEXT:    s_endpgm
 ;
 ; GFX11-CU-LABEL: flat_singlethread_one_as_release_monotonic_ret_cmpxchg:
 ; GFX11-CU:       ; %bb.0: ; %entry
 ; GFX11-CU-NEXT:    s_load_b128 s[0:3], s[0:1], 0x0
 ; GFX11-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v0, s0 :: v_dual_mov_b32 v1, s1
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v2, s2 :: v_dual_mov_b32 v3, s3
 ; GFX11-CU-NEXT:    flat_atomic_cmpswap_b32 v2, v[0:1], v[2:3] offset:16 glc
 ; GFX11-CU-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX11-CU-NEXT:    flat_store_b32 v[0:1], v2
-; GFX11-CU-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-CU-NEXT:    s_endpgm
     ptr %out, i32 %in, i32 %old) {
 entry:
   %gep = getelementptr i32, ptr %out, i32 4
   %val = cmpxchg volatile ptr %gep, i32 %old, i32 %in syncscope("singlethread-one-as") release monotonic
   %val0 = extractvalue { i32, i1 } %val, 0
   store i32 %val0, ptr %out, align 4
   ret void

 ; GFX11-WGP:       ; %bb.0: ; %entry
 ; GFX11-WGP-NEXT:    s_load_b128 s[0:3], s[0:1], 0x0
 ; GFX11-WGP-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-WGP-NEXT:    v_dual_mov_b32 v0, s0 :: v_dual_mov_b32 v1, s1
 ; GFX11-WGP-NEXT:    v_dual_mov_b32 v2, s2 :: v_dual_mov_b32 v3, s3
 ; GFX11-WGP-NEXT:    flat_atomic_cmpswap_b32 v2, v[0:1], v[2:3] offset:16 glc
 ; GFX11-WGP-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX11-WGP-NEXT:    flat_store_b32 v[0:1], v2
-; GFX11-WGP-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-WGP-NEXT:    s_endpgm
 ;
 ; GFX11-CU-LABEL: flat_singlethread_one_as_acq_rel_monotonic_ret_cmpxchg:
 ; GFX11-CU:       ; %bb.0: ; %entry
 ; GFX11-CU-NEXT:    s_load_b128 s[0:3], s[0:1], 0x0
 ; GFX11-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v0, s0 :: v_dual_mov_b32 v1, s1
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v2, s2 :: v_dual_mov_b32 v3, s3
 ; GFX11-CU-NEXT:    flat_atomic_cmpswap_b32 v2, v[0:1], v[2:3] offset:16 glc
 ; GFX11-CU-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX11-CU-NEXT:    flat_store_b32 v[0:1], v2
-; GFX11-CU-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-CU-NEXT:    s_endpgm
     ptr %out, i32 %in, i32 %old) {
 entry:
   %gep = getelementptr i32, ptr %out, i32 4
   %val = cmpxchg volatile ptr %gep, i32 %old, i32 %in syncscope("singlethread-one-as") acq_rel monotonic
   %val0 = extractvalue { i32, i1 } %val, 0
   store i32 %val0, ptr %out, align 4
   ret void

 ; GFX11-WGP:       ; %bb.0: ; %entry
 ; GFX11-WGP-NEXT:    s_load_b128 s[0:3], s[0:1], 0x0
 ; GFX11-WGP-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-WGP-NEXT:    v_dual_mov_b32 v0, s0 :: v_dual_mov_b32 v1, s1
 ; GFX11-WGP-NEXT:    v_dual_mov_b32 v2, s2 :: v_dual_mov_b32 v3, s3
 ; GFX11-WGP-NEXT:    flat_atomic_cmpswap_b32 v2, v[0:1], v[2:3] offset:16 glc
 ; GFX11-WGP-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX11-WGP-NEXT:    flat_store_b32 v[0:1], v2
-; GFX11-WGP-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-WGP-NEXT:    s_endpgm
 ;
 ; GFX11-CU-LABEL: flat_singlethread_one_as_seq_cst_monotonic_ret_cmpxchg:
 ; GFX11-CU:       ; %bb.0: ; %entry
 ; GFX11-CU-NEXT:    s_load_b128 s[0:3], s[0:1], 0x0
 ; GFX11-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v0, s0 :: v_dual_mov_b32 v1, s1
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v2, s2 :: v_dual_mov_b32 v3, s3
 ; GFX11-CU-NEXT:    flat_atomic_cmpswap_b32 v2, v[0:1], v[2:3] offset:16 glc
 ; GFX11-CU-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX11-CU-NEXT:    flat_store_b32 v[0:1], v2
-; GFX11-CU-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-CU-NEXT:    s_endpgm
     ptr %out, i32 %in, i32 %old) {
 entry:
   %gep = getelementptr i32, ptr %out, i32 4
   %val = cmpxchg volatile ptr %gep, i32 %old, i32 %in syncscope("singlethread-one-as") seq_cst monotonic
   %val0 = extractvalue { i32, i1 } %val, 0
   store i32 %val0, ptr %out, align 4
   ret void

 ; GFX11-WGP:       ; %bb.0: ; %entry
 ; GFX11-WGP-NEXT:    s_load_b128 s[0:3], s[0:1], 0x0
 ; GFX11-WGP-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-WGP-NEXT:    v_dual_mov_b32 v0, s0 :: v_dual_mov_b32 v1, s1
 ; GFX11-WGP-NEXT:    v_dual_mov_b32 v2, s2 :: v_dual_mov_b32 v3, s3
 ; GFX11-WGP-NEXT:    flat_atomic_cmpswap_b32 v2, v[0:1], v[2:3] offset:16 glc
 ; GFX11-WGP-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX11-WGP-NEXT:    flat_store_b32 v[0:1], v2
-; GFX11-WGP-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-WGP-NEXT:    s_endpgm
 ;
 ; GFX11-CU-LABEL: flat_singlethread_one_as_monotonic_acquire_ret_cmpxchg:
 ; GFX11-CU:       ; %bb.0: ; %entry
 ; GFX11-CU-NEXT:    s_load_b128 s[0:3], s[0:1], 0x0
 ; GFX11-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v0, s0 :: v_dual_mov_b32 v1, s1
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v2, s2 :: v_dual_mov_b32 v3, s3
 ; GFX11-CU-NEXT:    flat_atomic_cmpswap_b32 v2, v[0:1], v[2:3] offset:16 glc
 ; GFX11-CU-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX11-CU-NEXT:    flat_store_b32 v[0:1], v2
-; GFX11-CU-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-CU-NEXT:    s_endpgm
     ptr %out, i32 %in, i32 %old) {
 entry:
   %gep = getelementptr i32, ptr %out, i32 4
   %val = cmpxchg volatile ptr %gep, i32 %old, i32 %in syncscope("singlethread-one-as") monotonic acquire
   %val0 = extractvalue { i32, i1 } %val, 0
   store i32 %val0, ptr %out, align 4
   ret void

 ; GFX11-WGP:       ; %bb.0: ; %entry
 ; GFX11-WGP-NEXT:    s_load_b128 s[0:3], s[0:1], 0x0
 ; GFX11-WGP-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-WGP-NEXT:    v_dual_mov_b32 v0, s0 :: v_dual_mov_b32 v1, s1
 ; GFX11-WGP-NEXT:    v_dual_mov_b32 v2, s2 :: v_dual_mov_b32 v3, s3
 ; GFX11-WGP-NEXT:    flat_atomic_cmpswap_b32 v2, v[0:1], v[2:3] offset:16 glc
 ; GFX11-WGP-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX11-WGP-NEXT:    flat_store_b32 v[0:1], v2
-; GFX11-WGP-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-WGP-NEXT:    s_endpgm
 ;
 ; GFX11-CU-LABEL: flat_singlethread_one_as_acquire_acquire_ret_cmpxchg:
 ; GFX11-CU:       ; %bb.0: ; %entry
 ; GFX11-CU-NEXT:    s_load_b128 s[0:3], s[0:1], 0x0
 ; GFX11-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v0, s0 :: v_dual_mov_b32 v1, s1
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v2, s2 :: v_dual_mov_b32 v3, s3
 ; GFX11-CU-NEXT:    flat_atomic_cmpswap_b32 v2, v[0:1], v[2:3] offset:16 glc
 ; GFX11-CU-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX11-CU-NEXT:    flat_store_b32 v[0:1], v2
-; GFX11-CU-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-CU-NEXT:    s_endpgm
     ptr %out, i32 %in, i32 %old) {
 entry:
   %gep = getelementptr i32, ptr %out, i32 4
   %val = cmpxchg volatile ptr %gep, i32 %old, i32 %in syncscope("singlethread-one-as") acquire acquire
   %val0 = extractvalue { i32, i1 } %val, 0
   store i32 %val0, ptr %out, align 4
   ret void

 ; GFX11-WGP:       ; %bb.0: ; %entry
 ; GFX11-WGP-NEXT:    s_load_b128 s[0:3], s[0:1], 0x0
 ; GFX11-WGP-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-WGP-NEXT:    v_dual_mov_b32 v0, s0 :: v_dual_mov_b32 v1, s1
 ; GFX11-WGP-NEXT:    v_dual_mov_b32 v2, s2 :: v_dual_mov_b32 v3, s3
 ; GFX11-WGP-NEXT:    flat_atomic_cmpswap_b32 v2, v[0:1], v[2:3] offset:16 glc
 ; GFX11-WGP-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX11-WGP-NEXT:    flat_store_b32 v[0:1], v2
-; GFX11-WGP-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-WGP-NEXT:    s_endpgm
 ;
 ; GFX11-CU-LABEL: flat_singlethread_one_as_release_acquire_ret_cmpxchg:
 ; GFX11-CU:       ; %bb.0: ; %entry
 ; GFX11-CU-NEXT:    s_load_b128 s[0:3], s[0:1], 0x0
 ; GFX11-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v0, s0 :: v_dual_mov_b32 v1, s1
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v2, s2 :: v_dual_mov_b32 v3, s3
 ; GFX11-CU-NEXT:    flat_atomic_cmpswap_b32 v2, v[0:1], v[2:3] offset:16 glc
 ; GFX11-CU-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX11-CU-NEXT:    flat_store_b32 v[0:1], v2
-; GFX11-CU-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-CU-NEXT:    s_endpgm
     ptr %out, i32 %in, i32 %old) {
 entry:
   %gep = getelementptr i32, ptr %out, i32 4
   %val = cmpxchg volatile ptr %gep, i32 %old, i32 %in syncscope("singlethread-one-as") release acquire
   %val0 = extractvalue { i32, i1 } %val, 0
   store i32 %val0, ptr %out, align 4
   ret void

 ; GFX11-WGP:       ; %bb.0: ; %entry
 ; GFX11-WGP-NEXT:    s_load_b128 s[0:3], s[0:1], 0x0
 ; GFX11-WGP-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-WGP-NEXT:    v_dual_mov_b32 v0, s0 :: v_dual_mov_b32 v1, s1
 ; GFX11-WGP-NEXT:    v_dual_mov_b32 v2, s2 :: v_dual_mov_b32 v3, s3
 ; GFX11-WGP-NEXT:    flat_atomic_cmpswap_b32 v2, v[0:1], v[2:3] offset:16 glc
 ; GFX11-WGP-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX11-WGP-NEXT:    flat_store_b32 v[0:1], v2
-; GFX11-WGP-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-WGP-NEXT:    s_endpgm
 ;
 ; GFX11-CU-LABEL: flat_singlethread_one_as_acq_rel_acquire_ret_cmpxchg:
 ; GFX11-CU:       ; %bb.0: ; %entry
 ; GFX11-CU-NEXT:    s_load_b128 s[0:3], s[0:1], 0x0
 ; GFX11-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v0, s0 :: v_dual_mov_b32 v1, s1
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v2, s2 :: v_dual_mov_b32 v3, s3
 ; GFX11-CU-NEXT:    flat_atomic_cmpswap_b32 v2, v[0:1], v[2:3] offset:16 glc
 ; GFX11-CU-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX11-CU-NEXT:    flat_store_b32 v[0:1], v2
-; GFX11-CU-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-CU-NEXT:    s_endpgm
     ptr %out, i32 %in, i32 %old) {
 entry:
   %gep = getelementptr i32, ptr %out, i32 4
   %val = cmpxchg volatile ptr %gep, i32 %old, i32 %in syncscope("singlethread-one-as") acq_rel acquire
   %val0 = extractvalue { i32, i1 } %val, 0
   store i32 %val0, ptr %out, align 4
   ret void

 ; GFX11-WGP:       ; %bb.0: ; %entry
 ; GFX11-WGP-NEXT:    s_load_b128 s[0:3], s[0:1], 0x0
 ; GFX11-WGP-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-WGP-NEXT:    v_dual_mov_b32 v0, s0 :: v_dual_mov_b32 v1, s1
 ; GFX11-WGP-NEXT:    v_dual_mov_b32 v2, s2 :: v_dual_mov_b32 v3, s3
 ; GFX11-WGP-NEXT:    flat_atomic_cmpswap_b32 v2, v[0:1], v[2:3] offset:16 glc
 ; GFX11-WGP-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX11-WGP-NEXT:    flat_store_b32 v[0:1], v2
-; GFX11-WGP-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-WGP-NEXT:    s_endpgm
 ;
 ; GFX11-CU-LABEL: flat_singlethread_one_as_seq_cst_acquire_ret_cmpxchg:
 ; GFX11-CU:       ; %bb.0: ; %entry
 ; GFX11-CU-NEXT:    s_load_b128 s[0:3], s[0:1], 0x0
 ; GFX11-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v0, s0 :: v_dual_mov_b32 v1, s1
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v2, s2 :: v_dual_mov_b32 v3, s3
 ; GFX11-CU-NEXT:    flat_atomic_cmpswap_b32 v2, v[0:1], v[2:3] offset:16 glc
 ; GFX11-CU-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX11-CU-NEXT:    flat_store_b32 v[0:1], v2
-; GFX11-CU-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-CU-NEXT:    s_endpgm
     ptr %out, i32 %in, i32 %old) {
 entry:
   %gep = getelementptr i32, ptr %out, i32 4
   %val = cmpxchg volatile ptr %gep, i32 %old, i32 %in syncscope("singlethread-one-as") seq_cst acquire
   %val0 = extractvalue { i32, i1 } %val, 0
   store i32 %val0, ptr %out, align 4
   ret void

 ; GFX11-WGP:       ; %bb.0: ; %entry
 ; GFX11-WGP-NEXT:    s_load_b128 s[0:3], s[0:1], 0x0
 ; GFX11-WGP-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-WGP-NEXT:    v_dual_mov_b32 v0, s0 :: v_dual_mov_b32 v1, s1
 ; GFX11-WGP-NEXT:    v_dual_mov_b32 v2, s2 :: v_dual_mov_b32 v3, s3
 ; GFX11-WGP-NEXT:    flat_atomic_cmpswap_b32 v2, v[0:1], v[2:3] offset:16 glc
 ; GFX11-WGP-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX11-WGP-NEXT:    flat_store_b32 v[0:1], v2
-; GFX11-WGP-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-WGP-NEXT:    s_endpgm
 ;
 ; GFX11-CU-LABEL: flat_singlethread_one_as_monotonic_seq_cst_ret_cmpxchg:
 ; GFX11-CU:       ; %bb.0: ; %entry
 ; GFX11-CU-NEXT:    s_load_b128 s[0:3], s[0:1], 0x0
 ; GFX11-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v0, s0 :: v_dual_mov_b32 v1, s1
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v2, s2 :: v_dual_mov_b32 v3, s3
 ; GFX11-CU-NEXT:    flat_atomic_cmpswap_b32 v2, v[0:1], v[2:3] offset:16 glc
 ; GFX11-CU-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX11-CU-NEXT:    flat_store_b32 v[0:1], v2
-; GFX11-CU-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-CU-NEXT:    s_endpgm
     ptr %out, i32 %in, i32 %old) {
 entry:
   %gep = getelementptr i32, ptr %out, i32 4
   %val = cmpxchg volatile ptr %gep, i32 %old, i32 %in syncscope("singlethread-one-as") monotonic seq_cst
   %val0 = extractvalue { i32, i1 } %val, 0
   store i32 %val0, ptr %out, align 4
   ret void

 ; GFX11-WGP:       ; %bb.0: ; %entry
 ; GFX11-WGP-NEXT:    s_load_b128 s[0:3], s[0:1], 0x0
 ; GFX11-WGP-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-WGP-NEXT:    v_dual_mov_b32 v0, s0 :: v_dual_mov_b32 v1, s1
 ; GFX11-WGP-NEXT:    v_dual_mov_b32 v2, s2 :: v_dual_mov_b32 v3, s3
 ; GFX11-WGP-NEXT:    flat_atomic_cmpswap_b32 v2, v[0:1], v[2:3] offset:16 glc
 ; GFX11-WGP-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX11-WGP-NEXT:    flat_store_b32 v[0:1], v2
-; GFX11-WGP-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-WGP-NEXT:    s_endpgm
 ;
 ; GFX11-CU-LABEL: flat_singlethread_one_as_acquire_seq_cst_ret_cmpxchg:
 ; GFX11-CU:       ; %bb.0: ; %entry
 ; GFX11-CU-NEXT:    s_load_b128 s[0:3], s[0:1], 0x0
 ; GFX11-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v0, s0 :: v_dual_mov_b32 v1, s1
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v2, s2 :: v_dual_mov_b32 v3, s3
 ; GFX11-CU-NEXT:    flat_atomic_cmpswap_b32 v2, v[0:1], v[2:3] offset:16 glc
 ; GFX11-CU-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX11-CU-NEXT:    flat_store_b32 v[0:1], v2
-; GFX11-CU-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-CU-NEXT:    s_endpgm
     ptr %out, i32 %in, i32 %old) {
 entry:
   %gep = getelementptr i32, ptr %out, i32 4
   %val = cmpxchg volatile ptr %gep, i32 %old, i32 %in syncscope("singlethread-one-as") acquire seq_cst
   %val0 = extractvalue { i32, i1 } %val, 0
   store i32 %val0, ptr %out, align 4
   ret void

 ; GFX11-WGP:       ; %bb.0: ; %entry
 ; GFX11-WGP-NEXT:    s_load_b128 s[0:3], s[0:1], 0x0
 ; GFX11-WGP-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-WGP-NEXT:    v_dual_mov_b32 v0, s0 :: v_dual_mov_b32 v1, s1
 ; GFX11-WGP-NEXT:    v_dual_mov_b32 v2, s2 :: v_dual_mov_b32 v3, s3
 ; GFX11-WGP-NEXT:    flat_atomic_cmpswap_b32 v2, v[0:1], v[2:3] offset:16 glc
 ; GFX11-WGP-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX11-WGP-NEXT:    flat_store_b32 v[0:1], v2
-; GFX11-WGP-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-WGP-NEXT:    s_endpgm
 ;
 ; GFX11-CU-LABEL: flat_singlethread_one_as_release_seq_cst_ret_cmpxchg:
 ; GFX11-CU:       ; %bb.0: ; %entry
 ; GFX11-CU-NEXT:    s_load_b128 s[0:3], s[0:1], 0x0
 ; GFX11-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v0, s0 :: v_dual_mov_b32 v1, s1
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v2, s2 :: v_dual_mov_b32 v3, s3
 ; GFX11-CU-NEXT:    flat_atomic_cmpswap_b32 v2, v[0:1], v[2:3] offset:16 glc
 ; GFX11-CU-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX11-CU-NEXT:    flat_store_b32 v[0:1], v2
-; GFX11-CU-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-CU-NEXT:    s_endpgm
     ptr %out, i32 %in, i32 %old) {
 entry:
   %gep = getelementptr i32, ptr %out, i32 4
   %val = cmpxchg volatile ptr %gep, i32 %old, i32 %in syncscope("singlethread-one-as") release seq_cst
   %val0 = extractvalue { i32, i1 } %val, 0
   store i32 %val0, ptr %out, align 4
   ret void

 ; GFX11-WGP:       ; %bb.0: ; %entry
 ; GFX11-WGP-NEXT:    s_load_b128 s[0:3], s[0:1], 0x0
 ; GFX11-WGP-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-WGP-NEXT:    v_dual_mov_b32 v0, s0 :: v_dual_mov_b32 v1, s1
 ; GFX11-WGP-NEXT:    v_dual_mov_b32 v2, s2 :: v_dual_mov_b32 v3, s3
 ; GFX11-WGP-NEXT:    flat_atomic_cmpswap_b32 v2, v[0:1], v[2:3] offset:16 glc
 ; GFX11-WGP-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX11-WGP-NEXT:    flat_store_b32 v[0:1], v2
-; GFX11-WGP-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-WGP-NEXT:    s_endpgm
 ;
 ; GFX11-CU-LABEL: flat_singlethread_one_as_acq_rel_seq_cst_ret_cmpxchg:
 ; GFX11-CU:       ; %bb.0: ; %entry
 ; GFX11-CU-NEXT:    s_load_b128 s[0:3], s[0:1], 0x0
 ; GFX11-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v0, s0 :: v_dual_mov_b32 v1, s1
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v2, s2 :: v_dual_mov_b32 v3, s3
 ; GFX11-CU-NEXT:    flat_atomic_cmpswap_b32 v2, v[0:1], v[2:3] offset:16 glc
 ; GFX11-CU-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX11-CU-NEXT:    flat_store_b32 v[0:1], v2
-; GFX11-CU-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-CU-NEXT:    s_endpgm
     ptr %out, i32 %in, i32 %old) {
 entry:
   %gep = getelementptr i32, ptr %out, i32 4
   %val = cmpxchg volatile ptr %gep, i32 %old, i32 %in syncscope("singlethread-one-as") acq_rel seq_cst
   %val0 = extractvalue { i32, i1 } %val, 0
   store i32 %val0, ptr %out, align 4
   ret void

 ; GFX11-WGP:       ; %bb.0: ; %entry
 ; GFX11-WGP-NEXT:    s_load_b128 s[0:3], s[0:1], 0x0
 ; GFX11-WGP-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-WGP-NEXT:    v_dual_mov_b32 v0, s0 :: v_dual_mov_b32 v1, s1
 ; GFX11-WGP-NEXT:    v_dual_mov_b32 v2, s2 :: v_dual_mov_b32 v3, s3
 ; GFX11-WGP-NEXT:    flat_atomic_cmpswap_b32 v2, v[0:1], v[2:3] offset:16 glc
 ; GFX11-WGP-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX11-WGP-NEXT:    flat_store_b32 v[0:1], v2
-; GFX11-WGP-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-WGP-NEXT:    s_endpgm
 ;
 ; GFX11-CU-LABEL: flat_singlethread_one_as_seq_cst_seq_cst_ret_cmpxchg:
 ; GFX11-CU:       ; %bb.0: ; %entry
 ; GFX11-CU-NEXT:    s_load_b128 s[0:3], s[0:1], 0x0
 ; GFX11-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v0, s0 :: v_dual_mov_b32 v1, s1
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v2, s2 :: v_dual_mov_b32 v3, s3
 ; GFX11-CU-NEXT:    flat_atomic_cmpswap_b32 v2, v[0:1], v[2:3] offset:16 glc
 ; GFX11-CU-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX11-CU-NEXT:    flat_store_b32 v[0:1], v2
-; GFX11-CU-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-CU-NEXT:    s_endpgm
     ptr %out, i32 %in, i32 %old) {
 entry:
   %gep = getelementptr i32, ptr %out, i32 4
   %val = cmpxchg volatile ptr %gep, i32 %old, i32 %in syncscope("singlethread-one-as") seq_cst seq_cst
   %val0 = extractvalue { i32, i1 } %val, 0
   store i32 %val0, ptr %out, align 4
   ret void
 }
 

llvm/test/CodeGen/AMDGPU/memory-legalizer-flat-system.ll

 ; GFX11-WGP:       ; %bb.0: ; %entry
 ; GFX11-WGP-NEXT:    s_load_b128 s[0:3], s[0:1], 0x0
 ; GFX11-WGP-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-WGP-NEXT:    v_dual_mov_b32 v0, s0 :: v_dual_mov_b32 v1, s1
 ; GFX11-WGP-NEXT:    flat_load_b32 v2, v[0:1]
 ; GFX11-WGP-NEXT:    v_dual_mov_b32 v0, s2 :: v_dual_mov_b32 v1, s3
 ; GFX11-WGP-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX11-WGP-NEXT:    flat_store_b32 v[0:1], v2
-; GFX11-WGP-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-WGP-NEXT:    s_endpgm
 ;
 ; GFX11-CU-LABEL: flat_system_unordered_load:
 ; GFX11-CU:       ; %bb.0: ; %entry
 ; GFX11-CU-NEXT:    s_load_b128 s[0:3], s[0:1], 0x0
 ; GFX11-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v0, s0 :: v_dual_mov_b32 v1, s1
 ; GFX11-CU-NEXT:    flat_load_b32 v2, v[0:1]
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v0, s2 :: v_dual_mov_b32 v1, s3
 ; GFX11-CU-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX11-CU-NEXT:    flat_store_b32 v[0:1], v2
-; GFX11-CU-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-CU-NEXT:    s_endpgm
     ptr %in, ptr %out) {
 entry:
   %val = load atomic i32, ptr %in unordered, align 4
   store i32 %val, ptr %out
   ret void
 }
 

 ; GFX11-WGP:       ; %bb.0: ; %entry
 ; GFX11-WGP-NEXT:    s_load_b128 s[0:3], s[0:1], 0x0
 ; GFX11-WGP-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-WGP-NEXT:    v_dual_mov_b32 v0, s0 :: v_dual_mov_b32 v1, s1
 ; GFX11-WGP-NEXT:    flat_load_b32 v2, v[0:1] glc
 ; GFX11-WGP-NEXT:    v_dual_mov_b32 v0, s2 :: v_dual_mov_b32 v1, s3
 ; GFX11-WGP-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX11-WGP-NEXT:    flat_store_b32 v[0:1], v2
-; GFX11-WGP-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-WGP-NEXT:    s_endpgm
 ;
 ; GFX11-CU-LABEL: flat_system_monotonic_load:
 ; GFX11-CU:       ; %bb.0: ; %entry
 ; GFX11-CU-NEXT:    s_load_b128 s[0:3], s[0:1], 0x0
 ; GFX11-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v0, s0 :: v_dual_mov_b32 v1, s1
 ; GFX11-CU-NEXT:    flat_load_b32 v2, v[0:1] glc
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v0, s2 :: v_dual_mov_b32 v1, s3
 ; GFX11-CU-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX11-CU-NEXT:    flat_store_b32 v[0:1], v2
-; GFX11-CU-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-CU-NEXT:    s_endpgm
     ptr %in, ptr %out) {
 entry:
   %val = load atomic i32, ptr %in monotonic, align 4
   store i32 %val, ptr %out
   ret void
 }
 

 ; GFX11-WGP-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-WGP-NEXT:    v_dual_mov_b32 v0, s0 :: v_dual_mov_b32 v1, s1
 ; GFX11-WGP-NEXT:    flat_load_b32 v2, v[0:1] glc
 ; GFX11-WGP-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX11-WGP-NEXT:    buffer_gl0_inv
 ; GFX11-WGP-NEXT:    buffer_gl1_inv
 ; GFX11-WGP-NEXT:    v_dual_mov_b32 v0, s2 :: v_dual_mov_b32 v1, s3
 ; GFX11-WGP-NEXT:    flat_store_b32 v[0:1], v2
-; GFX11-WGP-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-WGP-NEXT:    s_endpgm
 ;
 ; GFX11-CU-LABEL: flat_system_acquire_load:
 ; GFX11-CU:       ; %bb.0: ; %entry
 ; GFX11-CU-NEXT:    s_load_b128 s[0:3], s[0:1], 0x0
 ; GFX11-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v0, s0 :: v_dual_mov_b32 v1, s1
 ; GFX11-CU-NEXT:    flat_load_b32 v2, v[0:1] glc
 ; GFX11-CU-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX11-CU-NEXT:    buffer_gl0_inv
 ; GFX11-CU-NEXT:    buffer_gl1_inv
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v0, s2 :: v_dual_mov_b32 v1, s3
 ; GFX11-CU-NEXT:    flat_store_b32 v[0:1], v2
-; GFX11-CU-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-CU-NEXT:    s_endpgm
     ptr %in, ptr %out) {
 entry:
   %val = load atomic i32, ptr %in acquire, align 4
   store i32 %val, ptr %out
   ret void
 }
 

 ; GFX11-WGP-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX11-WGP-NEXT:    s_waitcnt_vscnt null, 0x0
 ; GFX11-WGP-NEXT:    flat_load_b32 v2, v[0:1] glc
 ; GFX11-WGP-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX11-WGP-NEXT:    buffer_gl0_inv
 ; GFX11-WGP-NEXT:    buffer_gl1_inv
 ; GFX11-WGP-NEXT:    v_dual_mov_b32 v0, s2 :: v_dual_mov_b32 v1, s3
 ; GFX11-WGP-NEXT:    flat_store_b32 v[0:1], v2
-; GFX11-WGP-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-WGP-NEXT:    s_endpgm
 ;
 ; GFX11-CU-LABEL: flat_system_seq_cst_load:
 ; GFX11-CU:       ; %bb.0: ; %entry
 ; GFX11-CU-NEXT:    s_load_b128 s[0:3], s[0:1], 0x0
 ; GFX11-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v0, s0 :: v_dual_mov_b32 v1, s1
 ; GFX11-CU-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX11-CU-NEXT:    s_waitcnt_vscnt null, 0x0
 ; GFX11-CU-NEXT:    flat_load_b32 v2, v[0:1] glc
 ; GFX11-CU-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX11-CU-NEXT:    buffer_gl0_inv
 ; GFX11-CU-NEXT:    buffer_gl1_inv
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v0, s2 :: v_dual_mov_b32 v1, s3
 ; GFX11-CU-NEXT:    flat_store_b32 v[0:1], v2
-; GFX11-CU-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-CU-NEXT:    s_endpgm
     ptr %in, ptr %out) {
 entry:
   %val = load atomic i32, ptr %in seq_cst, align 4
   store i32 %val, ptr %out
   ret void
 }
 

 ; GFX11-WGP:       ; %bb.0: ; %entry
 ; GFX11-WGP-NEXT:    s_clause 0x1
 ; GFX11-WGP-NEXT:    s_load_b64 s[2:3], s[0:1], 0x8
 ; GFX11-WGP-NEXT:    s_load_b32 s0, s[0:1], 0x0
 ; GFX11-WGP-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-WGP-NEXT:    v_dual_mov_b32 v0, s2 :: v_dual_mov_b32 v1, s3
 ; GFX11-WGP-NEXT:    v_mov_b32_e32 v2, s0
 ; GFX11-WGP-NEXT:    flat_store_b32 v[0:1], v2
-; GFX11-WGP-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-WGP-NEXT:    s_endpgm
 ;
 ; GFX11-CU-LABEL: flat_system_unordered_store:
 ; GFX11-CU:       ; %bb.0: ; %entry
 ; GFX11-CU-NEXT:    s_clause 0x1
 ; GFX11-CU-NEXT:    s_load_b64 s[2:3], s[0:1], 0x8
 ; GFX11-CU-NEXT:    s_load_b32 s0, s[0:1], 0x0
 ; GFX11-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v0, s2 :: v_dual_mov_b32 v1, s3
 ; GFX11-CU-NEXT:    v_mov_b32_e32 v2, s0
 ; GFX11-CU-NEXT:    flat_store_b32 v[0:1], v2
-; GFX11-CU-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-CU-NEXT:    s_endpgm
     i32 %in, ptr %out) {
 entry:
   store atomic i32 %in, ptr %out unordered, align 4
   ret void
 }
 
 define amdgpu_kernel void @flat_system_monotonic_store(

 ; GFX11-WGP:       ; %bb.0: ; %entry
 ; GFX11-WGP-NEXT:    s_clause 0x1
 ; GFX11-WGP-NEXT:    s_load_b64 s[2:3], s[0:1], 0x8
 ; GFX11-WGP-NEXT:    s_load_b32 s0, s[0:1], 0x0
 ; GFX11-WGP-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-WGP-NEXT:    v_dual_mov_b32 v0, s2 :: v_dual_mov_b32 v1, s3
 ; GFX11-WGP-NEXT:    v_mov_b32_e32 v2, s0
 ; GFX11-WGP-NEXT:    flat_store_b32 v[0:1], v2
-; GFX11-WGP-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-WGP-NEXT:    s_endpgm
 ;
 ; GFX11-CU-LABEL: flat_system_monotonic_store:
 ; GFX11-CU:       ; %bb.0: ; %entry
 ; GFX11-CU-NEXT:    s_clause 0x1
 ; GFX11-CU-NEXT:    s_load_b64 s[2:3], s[0:1], 0x8
 ; GFX11-CU-NEXT:    s_load_b32 s0, s[0:1], 0x0
 ; GFX11-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v0, s2 :: v_dual_mov_b32 v1, s3
 ; GFX11-CU-NEXT:    v_mov_b32_e32 v2, s0
 ; GFX11-CU-NEXT:    flat_store_b32 v[0:1], v2
-; GFX11-CU-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-CU-NEXT:    s_endpgm
     i32 %in, ptr %out) {
 entry:
   store atomic i32 %in, ptr %out monotonic, align 4
   ret void
 }
 
 define amdgpu_kernel void @flat_system_release_store(

 ; GFX11-WGP-NEXT:    s_load_b64 s[2:3], s[0:1], 0x8
 ; GFX11-WGP-NEXT:    s_load_b32 s0, s[0:1], 0x0
 ; GFX11-WGP-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-WGP-NEXT:    v_dual_mov_b32 v0, s2 :: v_dual_mov_b32 v1, s3
 ; GFX11-WGP-NEXT:    v_mov_b32_e32 v2, s0
 ; GFX11-WGP-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX11-WGP-NEXT:    s_waitcnt_vscnt null, 0x0
 ; GFX11-WGP-NEXT:    flat_store_b32 v[0:1], v2
-; GFX11-WGP-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-WGP-NEXT:    s_endpgm
 ;
 ; GFX11-CU-LABEL: flat_system_release_store:
 ; GFX11-CU:       ; %bb.0: ; %entry
 ; GFX11-CU-NEXT:    s_clause 0x1
 ; GFX11-CU-NEXT:    s_load_b64 s[2:3], s[0:1], 0x8
 ; GFX11-CU-NEXT:    s_load_b32 s0, s[0:1], 0x0
 ; GFX11-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v0, s2 :: v_dual_mov_b32 v1, s3
 ; GFX11-CU-NEXT:    v_mov_b32_e32 v2, s0
 ; GFX11-CU-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX11-CU-NEXT:    s_waitcnt_vscnt null, 0x0
 ; GFX11-CU-NEXT:    flat_store_b32 v[0:1], v2
-; GFX11-CU-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-CU-NEXT:    s_endpgm
     i32 %in, ptr %out) {
 entry:
   store atomic i32 %in, ptr %out release, align 4
   ret void
 }
 
 define amdgpu_kernel void @flat_system_seq_cst_store(

 ; GFX11-WGP-NEXT:    s_load_b64 s[2:3], s[0:1], 0x8
 ; GFX11-WGP-NEXT:    s_load_b32 s0, s[0:1], 0x0
 ; GFX11-WGP-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-WGP-NEXT:    v_dual_mov_b32 v0, s2 :: v_dual_mov_b32 v1, s3
 ; GFX11-WGP-NEXT:    v_mov_b32_e32 v2, s0
 ; GFX11-WGP-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX11-WGP-NEXT:    s_waitcnt_vscnt null, 0x0
 ; GFX11-WGP-NEXT:    flat_store_b32 v[0:1], v2
-; GFX11-WGP-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-WGP-NEXT:    s_endpgm
 ;
 ; GFX11-CU-LABEL: flat_system_seq_cst_store:
 ; GFX11-CU:       ; %bb.0: ; %entry
 ; GFX11-CU-NEXT:    s_clause 0x1
 ; GFX11-CU-NEXT:    s_load_b64 s[2:3], s[0:1], 0x8
 ; GFX11-CU-NEXT:    s_load_b32 s0, s[0:1], 0x0
 ; GFX11-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v0, s2 :: v_dual_mov_b32 v1, s3
 ; GFX11-CU-NEXT:    v_mov_b32_e32 v2, s0
 ; GFX11-CU-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX11-CU-NEXT:    s_waitcnt_vscnt null, 0x0
 ; GFX11-CU-NEXT:    flat_store_b32 v[0:1], v2
-; GFX11-CU-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-CU-NEXT:    s_endpgm
     i32 %in, ptr %out) {
 entry:
   store atomic i32 %in, ptr %out seq_cst, align 4
   ret void
 }
 
 define amdgpu_kernel void @flat_system_monotonic_atomicrmw(

 ; GFX11-WGP:       ; %bb.0: ; %entry
 ; GFX11-WGP-NEXT:    s_clause 0x1
 ; GFX11-WGP-NEXT:    s_load_b64 s[2:3], s[0:1], 0x0
 ; GFX11-WGP-NEXT:    s_load_b32 s0, s[0:1], 0x8
 ; GFX11-WGP-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-WGP-NEXT:    v_dual_mov_b32 v0, s2 :: v_dual_mov_b32 v1, s3
 ; GFX11-WGP-NEXT:    v_mov_b32_e32 v2, s0
 ; GFX11-WGP-NEXT:    flat_atomic_swap_b32 v[0:1], v2
-; GFX11-WGP-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-WGP-NEXT:    s_endpgm
 ;
 ; GFX11-CU-LABEL: flat_system_monotonic_atomicrmw:
 ; GFX11-CU:       ; %bb.0: ; %entry
 ; GFX11-CU-NEXT:    s_clause 0x1
 ; GFX11-CU-NEXT:    s_load_b64 s[2:3], s[0:1], 0x0
 ; GFX11-CU-NEXT:    s_load_b32 s0, s[0:1], 0x8
 ; GFX11-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v0, s2 :: v_dual_mov_b32 v1, s3
 ; GFX11-CU-NEXT:    v_mov_b32_e32 v2, s0
 ; GFX11-CU-NEXT:    flat_atomic_swap_b32 v[0:1], v2
-; GFX11-CU-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-CU-NEXT:    s_endpgm
     ptr %out, i32 %in) {
 entry:
   %val = atomicrmw volatile xchg ptr %out, i32 %in monotonic
   ret void
 }
 
 define amdgpu_kernel void @flat_system_acquire_atomicrmw(

 ; GFX11-WGP-NEXT:    s_load_b64 s[2:3], s[0:1], 0x0
 ; GFX11-WGP-NEXT:    s_load_b32 s0, s[0:1], 0x8
 ; GFX11-WGP-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-WGP-NEXT:    v_dual_mov_b32 v0, s2 :: v_dual_mov_b32 v1, s3
 ; GFX11-WGP-NEXT:    v_mov_b32_e32 v2, s0
 ; GFX11-WGP-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX11-WGP-NEXT:    s_waitcnt_vscnt null, 0x0
 ; GFX11-WGP-NEXT:    flat_atomic_swap_b32 v[0:1], v2
-; GFX11-WGP-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-WGP-NEXT:    s_endpgm
 ;
 ; GFX11-CU-LABEL: flat_system_release_atomicrmw:
 ; GFX11-CU:       ; %bb.0: ; %entry
 ; GFX11-CU-NEXT:    s_clause 0x1
 ; GFX11-CU-NEXT:    s_load_b64 s[2:3], s[0:1], 0x0
 ; GFX11-CU-NEXT:    s_load_b32 s0, s[0:1], 0x8
 ; GFX11-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v0, s2 :: v_dual_mov_b32 v1, s3
 ; GFX11-CU-NEXT:    v_mov_b32_e32 v2, s0
 ; GFX11-CU-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX11-CU-NEXT:    s_waitcnt_vscnt null, 0x0
 ; GFX11-CU-NEXT:    flat_atomic_swap_b32 v[0:1], v2
-; GFX11-CU-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-CU-NEXT:    s_endpgm
     ptr %out, i32 %in) {
 entry:
   %val = atomicrmw volatile xchg ptr %out, i32 %in release
   ret void
 }
 
 define amdgpu_kernel void @flat_system_acq_rel_atomicrmw(

 ; GFX11-WGP-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-WGP-NEXT:    v_dual_mov_b32 v0, s2 :: v_dual_mov_b32 v1, s3
 ; GFX11-WGP-NEXT:    v_mov_b32_e32 v2, s0
 ; GFX11-WGP-NEXT:    flat_atomic_swap_b32 v2, v[0:1], v2 glc
 ; GFX11-WGP-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX11-WGP-NEXT:    buffer_gl0_inv
 ; GFX11-WGP-NEXT:    buffer_gl1_inv
 ; GFX11-WGP-NEXT:    flat_store_b32 v[0:1], v2
-; GFX11-WGP-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-WGP-NEXT:    s_endpgm
 ;
 ; GFX11-CU-LABEL: flat_system_acquire_ret_atomicrmw:
 ; GFX11-CU:       ; %bb.0: ; %entry
 ; GFX11-CU-NEXT:    s_clause 0x1
 ; GFX11-CU-NEXT:    s_load_b64 s[2:3], s[0:1], 0x0
 ; GFX11-CU-NEXT:    s_load_b32 s0, s[0:1], 0x8
 ; GFX11-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v0, s2 :: v_dual_mov_b32 v1, s3
 ; GFX11-CU-NEXT:    v_mov_b32_e32 v2, s0
 ; GFX11-CU-NEXT:    flat_atomic_swap_b32 v2, v[0:1], v2 glc
 ; GFX11-CU-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX11-CU-NEXT:    buffer_gl0_inv
 ; GFX11-CU-NEXT:    buffer_gl1_inv
 ; GFX11-CU-NEXT:    flat_store_b32 v[0:1], v2
-; GFX11-CU-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-CU-NEXT:    s_endpgm
     ptr %out, i32 %in) {
 entry:
   %val = atomicrmw volatile xchg ptr %out, i32 %in acquire
   store i32 %val, ptr %out, align 4
   ret void
 }
 

 ; GFX11-WGP-NEXT:    v_mov_b32_e32 v2, s0
 ; GFX11-WGP-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX11-WGP-NEXT:    s_waitcnt_vscnt null, 0x0
 ; GFX11-WGP-NEXT:    flat_atomic_swap_b32 v2, v[0:1], v2 glc
 ; GFX11-WGP-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX11-WGP-NEXT:    buffer_gl0_inv
 ; GFX11-WGP-NEXT:    buffer_gl1_inv
 ; GFX11-WGP-NEXT:    flat_store_b32 v[0:1], v2
-; GFX11-WGP-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-WGP-NEXT:    s_endpgm
 ;
 ; GFX11-CU-LABEL: flat_system_acq_rel_ret_atomicrmw:
 ; GFX11-CU:       ; %bb.0: ; %entry
 ; GFX11-CU-NEXT:    s_clause 0x1
 ; GFX11-CU-NEXT:    s_load_b64 s[2:3], s[0:1], 0x0
 ; GFX11-CU-NEXT:    s_load_b32 s0, s[0:1], 0x8
 ; GFX11-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v0, s2 :: v_dual_mov_b32 v1, s3
 ; GFX11-CU-NEXT:    v_mov_b32_e32 v2, s0
 ; GFX11-CU-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX11-CU-NEXT:    s_waitcnt_vscnt null, 0x0
 ; GFX11-CU-NEXT:    flat_atomic_swap_b32 v2, v[0:1], v2 glc
 ; GFX11-CU-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX11-CU-NEXT:    buffer_gl0_inv
 ; GFX11-CU-NEXT:    buffer_gl1_inv
 ; GFX11-CU-NEXT:    flat_store_b32 v[0:1], v2
-; GFX11-CU-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-CU-NEXT:    s_endpgm
     ptr %out, i32 %in) {
 entry:
   %val = atomicrmw volatile xchg ptr %out, i32 %in acq_rel
   store i32 %val, ptr %out, align 4
   ret void
 }
 

 ; GFX11-WGP-NEXT:    v_mov_b32_e32 v2, s0
 ; GFX11-WGP-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX11-WGP-NEXT:    s_waitcnt_vscnt null, 0x0
 ; GFX11-WGP-NEXT:    flat_atomic_swap_b32 v2, v[0:1], v2 glc
 ; GFX11-WGP-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX11-WGP-NEXT:    buffer_gl0_inv
 ; GFX11-WGP-NEXT:    buffer_gl1_inv
 ; GFX11-WGP-NEXT:    flat_store_b32 v[0:1], v2
-; GFX11-WGP-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-WGP-NEXT:    s_endpgm
 ;
 ; GFX11-CU-LABEL: flat_system_seq_cst_ret_atomicrmw:
 ; GFX11-CU:       ; %bb.0: ; %entry
 ; GFX11-CU-NEXT:    s_clause 0x1
 ; GFX11-CU-NEXT:    s_load_b64 s[2:3], s[0:1], 0x0
 ; GFX11-CU-NEXT:    s_load_b32 s0, s[0:1], 0x8
 ; GFX11-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v0, s2 :: v_dual_mov_b32 v1, s3
 ; GFX11-CU-NEXT:    v_mov_b32_e32 v2, s0
 ; GFX11-CU-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX11-CU-NEXT:    s_waitcnt_vscnt null, 0x0
 ; GFX11-CU-NEXT:    flat_atomic_swap_b32 v2, v[0:1], v2 glc
 ; GFX11-CU-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX11-CU-NEXT:    buffer_gl0_inv
 ; GFX11-CU-NEXT:    buffer_gl1_inv
 ; GFX11-CU-NEXT:    flat_store_b32 v[0:1], v2
-; GFX11-CU-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-CU-NEXT:    s_endpgm
     ptr %out, i32 %in) {
 entry:
   %val = atomicrmw volatile xchg ptr %out, i32 %in seq_cst
   store i32 %val, ptr %out, align 4
   ret void
 }
 

 ;
 ; GFX11-WGP-LABEL: flat_system_monotonic_monotonic_cmpxchg:
 ; GFX11-WGP:       ; %bb.0: ; %entry
 ; GFX11-WGP-NEXT:    s_load_b128 s[0:3], s[0:1], 0x0
 ; GFX11-WGP-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-WGP-NEXT:    v_dual_mov_b32 v0, s0 :: v_dual_mov_b32 v1, s1
 ; GFX11-WGP-NEXT:    v_dual_mov_b32 v2, s2 :: v_dual_mov_b32 v3, s3
 ; GFX11-WGP-NEXT:    flat_atomic_cmpswap_b32 v[0:1], v[2:3] offset:16
-; GFX11-WGP-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-WGP-NEXT:    s_endpgm
 ;
 ; GFX11-CU-LABEL: flat_system_monotonic_monotonic_cmpxchg:
 ; GFX11-CU:       ; %bb.0: ; %entry
 ; GFX11-CU-NEXT:    s_load_b128 s[0:3], s[0:1], 0x0
 ; GFX11-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v0, s0 :: v_dual_mov_b32 v1, s1
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v2, s2 :: v_dual_mov_b32 v3, s3
 ; GFX11-CU-NEXT:    flat_atomic_cmpswap_b32 v[0:1], v[2:3] offset:16
-; GFX11-CU-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-CU-NEXT:    s_endpgm
     ptr %out, i32 %in, i32 %old) {
 entry:
   %gep = getelementptr i32, ptr %out, i32 4
   %val = cmpxchg volatile ptr %gep, i32 %old, i32 %in monotonic monotonic
   ret void
 }
 

 ; GFX11-WGP:       ; %bb.0: ; %entry
 ; GFX11-WGP-NEXT:    s_load_b128 s[0:3], s[0:1], 0x0
 ; GFX11-WGP-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-WGP-NEXT:    v_dual_mov_b32 v0, s0 :: v_dual_mov_b32 v1, s1
 ; GFX11-WGP-NEXT:    v_dual_mov_b32 v2, s2 :: v_dual_mov_b32 v3, s3
 ; GFX11-WGP-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX11-WGP-NEXT:    s_waitcnt_vscnt null, 0x0
 ; GFX11-WGP-NEXT:    flat_atomic_cmpswap_b32 v[0:1], v[2:3] offset:16
-; GFX11-WGP-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-WGP-NEXT:    s_endpgm
 ;
 ; GFX11-CU-LABEL: flat_system_release_monotonic_cmpxchg:
 ; GFX11-CU:       ; %bb.0: ; %entry
 ; GFX11-CU-NEXT:    s_load_b128 s[0:3], s[0:1], 0x0
 ; GFX11-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v0, s0 :: v_dual_mov_b32 v1, s1
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v2, s2 :: v_dual_mov_b32 v3, s3
 ; GFX11-CU-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX11-CU-NEXT:    s_waitcnt_vscnt null, 0x0
 ; GFX11-CU-NEXT:    flat_atomic_cmpswap_b32 v[0:1], v[2:3] offset:16
-; GFX11-CU-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-CU-NEXT:    s_endpgm
     ptr %out, i32 %in, i32 %old) {
 entry:
   %gep = getelementptr i32, ptr %out, i32 4
   %val = cmpxchg volatile ptr %gep, i32 %old, i32 %in release monotonic
   ret void
 }
 

 ; GFX11-WGP:       ; %bb.0: ; %entry
 ; GFX11-WGP-NEXT:    s_load_b128 s[0:3], s[0:1], 0x0
 ; GFX11-WGP-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-WGP-NEXT:    v_dual_mov_b32 v0, s0 :: v_dual_mov_b32 v1, s1
 ; GFX11-WGP-NEXT:    v_dual_mov_b32 v2, s2 :: v_dual_mov_b32 v3, s3
 ; GFX11-WGP-NEXT:    flat_atomic_cmpswap_b32 v2, v[0:1], v[2:3] offset:16 glc
 ; GFX11-WGP-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX11-WGP-NEXT:    flat_store_b32 v[0:1], v2
-; GFX11-WGP-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-WGP-NEXT:    s_endpgm
 ;
 ; GFX11-CU-LABEL: flat_system_monotonic_monotonic_ret_cmpxchg:
 ; GFX11-CU:       ; %bb.0: ; %entry
 ; GFX11-CU-NEXT:    s_load_b128 s[0:3], s[0:1], 0x0
 ; GFX11-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v0, s0 :: v_dual_mov_b32 v1, s1
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v2, s2 :: v_dual_mov_b32 v3, s3
 ; GFX11-CU-NEXT:    flat_atomic_cmpswap_b32 v2, v[0:1], v[2:3] offset:16 glc
 ; GFX11-CU-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX11-CU-NEXT:    flat_store_b32 v[0:1], v2
-; GFX11-CU-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-CU-NEXT:    s_endpgm
     ptr %out, i32 %in, i32 %old) {
 entry:
   %gep = getelementptr i32, ptr %out, i32 4
   %val = cmpxchg volatile ptr %gep, i32 %old, i32 %in monotonic monotonic
   %val0 = extractvalue { i32, i1 } %val, 0
   store i32 %val0, ptr %out, align 4
   ret void

 ; GFX11-WGP-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-WGP-NEXT:    v_dual_mov_b32 v0, s0 :: v_dual_mov_b32 v1, s1
 ; GFX11-WGP-NEXT:    v_dual_mov_b32 v2, s2 :: v_dual_mov_b32 v3, s3
 ; GFX11-WGP-NEXT:    flat_atomic_cmpswap_b32 v2, v[0:1], v[2:3] offset:16 glc
 ; GFX11-WGP-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX11-WGP-NEXT:    buffer_gl0_inv
 ; GFX11-WGP-NEXT:    buffer_gl1_inv
 ; GFX11-WGP-NEXT:    flat_store_b32 v[0:1], v2
-; GFX11-WGP-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-WGP-NEXT:    s_endpgm
 ;
 ; GFX11-CU-LABEL: flat_system_acquire_monotonic_ret_cmpxchg:
 ; GFX11-CU:       ; %bb.0: ; %entry
 ; GFX11-CU-NEXT:    s_load_b128 s[0:3], s[0:1], 0x0
 ; GFX11-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v0, s0 :: v_dual_mov_b32 v1, s1
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v2, s2 :: v_dual_mov_b32 v3, s3
 ; GFX11-CU-NEXT:    flat_atomic_cmpswap_b32 v2, v[0:1], v[2:3] offset:16 glc
 ; GFX11-CU-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX11-CU-NEXT:    buffer_gl0_inv
 ; GFX11-CU-NEXT:    buffer_gl1_inv
 ; GFX11-CU-NEXT:    flat_store_b32 v[0:1], v2
-; GFX11-CU-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-CU-NEXT:    s_endpgm
     ptr %out, i32 %in, i32 %old) {
 entry:
   %gep = getelementptr i32, ptr %out, i32 4
   %val = cmpxchg volatile ptr %gep, i32 %old, i32 %in acquire monotonic
   %val0 = extractvalue { i32, i1 } %val, 0
   store i32 %val0, ptr %out, align 4
   ret void

 ; GFX11-WGP-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-WGP-NEXT:    v_dual_mov_b32 v0, s0 :: v_dual_mov_b32 v1, s1
 ; GFX11-WGP-NEXT:    v_dual_mov_b32 v2, s2 :: v_dual_mov_b32 v3, s3
 ; GFX11-WGP-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX11-WGP-NEXT:    s_waitcnt_vscnt null, 0x0
 ; GFX11-WGP-NEXT:    flat_atomic_cmpswap_b32 v2, v[0:1], v[2:3] offset:16 glc
 ; GFX11-WGP-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX11-WGP-NEXT:    flat_store_b32 v[0:1], v2
-; GFX11-WGP-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-WGP-NEXT:    s_endpgm
 ;
 ; GFX11-CU-LABEL: flat_system_release_monotonic_ret_cmpxchg:
 ; GFX11-CU:       ; %bb.0: ; %entry
 ; GFX11-CU-NEXT:    s_load_b128 s[0:3], s[0:1], 0x0
 ; GFX11-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v0, s0 :: v_dual_mov_b32 v1, s1
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v2, s2 :: v_dual_mov_b32 v3, s3
 ; GFX11-CU-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX11-CU-NEXT:    s_waitcnt_vscnt null, 0x0
 ; GFX11-CU-NEXT:    flat_atomic_cmpswap_b32 v2, v[0:1], v[2:3] offset:16 glc
 ; GFX11-CU-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX11-CU-NEXT:    flat_store_b32 v[0:1], v2
-; GFX11-CU-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-CU-NEXT:    s_endpgm
     ptr %out, i32 %in, i32 %old) {
 entry:
   %gep = getelementptr i32, ptr %out, i32 4
   %val = cmpxchg volatile ptr %gep, i32 %old, i32 %in release monotonic
   %val0 = extractvalue { i32, i1 } %val, 0
   store i32 %val0, ptr %out, align 4
   ret void

 ; GFX11-WGP-NEXT:    v_dual_mov_b32 v2, s2 :: v_dual_mov_b32 v3, s3
 ; GFX11-WGP-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX11-WGP-NEXT:    s_waitcnt_vscnt null, 0x0
 ; GFX11-WGP-NEXT:    flat_atomic_cmpswap_b32 v2, v[0:1], v[2:3] offset:16 glc
 ; GFX11-WGP-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX11-WGP-NEXT:    buffer_gl0_inv
 ; GFX11-WGP-NEXT:    buffer_gl1_inv
 ; GFX11-WGP-NEXT:    flat_store_b32 v[0:1], v2
-; GFX11-WGP-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-WGP-NEXT:    s_endpgm
 ;
 ; GFX11-CU-LABEL: flat_system_acq_rel_monotonic_ret_cmpxchg:
 ; GFX11-CU:       ; %bb.0: ; %entry
 ; GFX11-CU-NEXT:    s_load_b128 s[0:3], s[0:1], 0x0
 ; GFX11-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v0, s0 :: v_dual_mov_b32 v1, s1
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v2, s2 :: v_dual_mov_b32 v3, s3
 ; GFX11-CU-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX11-CU-NEXT:    s_waitcnt_vscnt null, 0x0
 ; GFX11-CU-NEXT:    flat_atomic_cmpswap_b32 v2, v[0:1], v[2:3] offset:16 glc
 ; GFX11-CU-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX11-CU-NEXT:    buffer_gl0_inv
 ; GFX11-CU-NEXT:    buffer_gl1_inv
 ; GFX11-CU-NEXT:    flat_store_b32 v[0:1], v2
-; GFX11-CU-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-CU-NEXT:    s_endpgm
     ptr %out, i32 %in, i32 %old) {
 entry:
   %gep = getelementptr i32, ptr %out, i32 4
   %val = cmpxchg volatile ptr %gep, i32 %old, i32 %in acq_rel monotonic
   %val0 = extractvalue { i32, i1 } %val, 0
   store i32 %val0, ptr %out, align 4
   ret void

 ; GFX11-WGP-NEXT:    v_dual_mov_b32 v2, s2 :: v_dual_mov_b32 v3, s3
 ; GFX11-WGP-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX11-WGP-NEXT:    s_waitcnt_vscnt null, 0x0
 ; GFX11-WGP-NEXT:    flat_atomic_cmpswap_b32 v2, v[0:1], v[2:3] offset:16 glc
 ; GFX11-WGP-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX11-WGP-NEXT:    buffer_gl0_inv
 ; GFX11-WGP-NEXT:    buffer_gl1_inv
 ; GFX11-WGP-NEXT:    flat_store_b32 v[0:1], v2
-; GFX11-WGP-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-WGP-NEXT:    s_endpgm
 ;
 ; GFX11-CU-LABEL: flat_system_seq_cst_monotonic_ret_cmpxchg:
 ; GFX11-CU:       ; %bb.0: ; %entry
 ; GFX11-CU-NEXT:    s_load_b128 s[0:3], s[0:1], 0x0
 ; GFX11-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v0, s0 :: v_dual_mov_b32 v1, s1
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v2, s2 :: v_dual_mov_b32 v3, s3
 ; GFX11-CU-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX11-CU-NEXT:    s_waitcnt_vscnt null, 0x0
 ; GFX11-CU-NEXT:    flat_atomic_cmpswap_b32 v2, v[0:1], v[2:3] offset:16 glc
 ; GFX11-CU-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX11-CU-NEXT:    buffer_gl0_inv
 ; GFX11-CU-NEXT:    buffer_gl1_inv
 ; GFX11-CU-NEXT:    flat_store_b32 v[0:1], v2
-; GFX11-CU-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-CU-NEXT:    s_endpgm
     ptr %out, i32 %in, i32 %old) {
 entry:
   %gep = getelementptr i32, ptr %out, i32 4
   %val = cmpxchg volatile ptr %gep, i32 %old, i32 %in seq_cst monotonic
   %val0 = extractvalue { i32, i1 } %val, 0
   store i32 %val0, ptr %out, align 4
   ret void

 ; GFX11-WGP-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-WGP-NEXT:    v_dual_mov_b32 v0, s0 :: v_dual_mov_b32 v1, s1
 ; GFX11-WGP-NEXT:    v_dual_mov_b32 v2, s2 :: v_dual_mov_b32 v3, s3
 ; GFX11-WGP-NEXT:    flat_atomic_cmpswap_b32 v2, v[0:1], v[2:3] offset:16 glc
 ; GFX11-WGP-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX11-WGP-NEXT:    buffer_gl0_inv
 ; GFX11-WGP-NEXT:    buffer_gl1_inv
 ; GFX11-WGP-NEXT:    flat_store_b32 v[0:1], v2
-; GFX11-WGP-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-WGP-NEXT:    s_endpgm
 ;
 ; GFX11-CU-LABEL: flat_system_monotonic_acquire_ret_cmpxchg:
 ; GFX11-CU:       ; %bb.0: ; %entry
 ; GFX11-CU-NEXT:    s_load_b128 s[0:3], s[0:1], 0x0
 ; GFX11-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v0, s0 :: v_dual_mov_b32 v1, s1
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v2, s2 :: v_dual_mov_b32 v3, s3
 ; GFX11-CU-NEXT:    flat_atomic_cmpswap_b32 v2, v[0:1], v[2:3] offset:16 glc
 ; GFX11-CU-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX11-CU-NEXT:    buffer_gl0_inv
 ; GFX11-CU-NEXT:    buffer_gl1_inv
 ; GFX11-CU-NEXT:    flat_store_b32 v[0:1], v2
-; GFX11-CU-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-CU-NEXT:    s_endpgm
     ptr %out, i32 %in, i32 %old) {
 entry:
   %gep = getelementptr i32, ptr %out, i32 4
   %val = cmpxchg volatile ptr %gep, i32 %old, i32 %in monotonic acquire
   %val0 = extractvalue { i32, i1 } %val, 0
   store i32 %val0, ptr %out, align 4
   ret void

 ; GFX11-WGP-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-WGP-NEXT:    v_dual_mov_b32 v0, s0 :: v_dual_mov_b32 v1, s1
 ; GFX11-WGP-NEXT:    v_dual_mov_b32 v2, s2 :: v_dual_mov_b32 v3, s3
 ; GFX11-WGP-NEXT:    flat_atomic_cmpswap_b32 v2, v[0:1], v[2:3] offset:16 glc
 ; GFX11-WGP-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX11-WGP-NEXT:    buffer_gl0_inv
 ; GFX11-WGP-NEXT:    buffer_gl1_inv
 ; GFX11-WGP-NEXT:    flat_store_b32 v[0:1], v2
-; GFX11-WGP-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-WGP-NEXT:    s_endpgm
 ;
 ; GFX11-CU-LABEL: flat_system_acquire_acquire_ret_cmpxchg:
 ; GFX11-CU:       ; %bb.0: ; %entry
 ; GFX11-CU-NEXT:    s_load_b128 s[0:3], s[0:1], 0x0
 ; GFX11-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v0, s0 :: v_dual_mov_b32 v1, s1
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v2, s2 :: v_dual_mov_b32 v3, s3
 ; GFX11-CU-NEXT:    flat_atomic_cmpswap_b32 v2, v[0:1], v[2:3] offset:16 glc
 ; GFX11-CU-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX11-CU-NEXT:    buffer_gl0_inv
 ; GFX11-CU-NEXT:    buffer_gl1_inv
 ; GFX11-CU-NEXT:    flat_store_b32 v[0:1], v2
-; GFX11-CU-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-CU-NEXT:    s_endpgm
     ptr %out, i32 %in, i32 %old) {
 entry:
   %gep = getelementptr i32, ptr %out, i32 4
   %val = cmpxchg volatile ptr %gep, i32 %old, i32 %in acquire acquire
   %val0 = extractvalue { i32, i1 } %val, 0
   store i32 %val0, ptr %out, align 4
   ret void

 ; GFX11-WGP-NEXT:    v_dual_mov_b32 v2, s2 :: v_dual_mov_b32 v3, s3
 ; GFX11-WGP-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX11-WGP-NEXT:    s_waitcnt_vscnt null, 0x0
 ; GFX11-WGP-NEXT:    flat_atomic_cmpswap_b32 v2, v[0:1], v[2:3] offset:16 glc
 ; GFX11-WGP-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX11-WGP-NEXT:    buffer_gl0_inv
 ; GFX11-WGP-NEXT:    buffer_gl1_inv
 ; GFX11-WGP-NEXT:    flat_store_b32 v[0:1], v2
-; GFX11-WGP-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-WGP-NEXT:    s_endpgm
 ;
 ; GFX11-CU-LABEL: flat_system_release_acquire_ret_cmpxchg:
 ; GFX11-CU:       ; %bb.0: ; %entry
 ; GFX11-CU-NEXT:    s_load_b128 s[0:3], s[0:1], 0x0
 ; GFX11-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v0, s0 :: v_dual_mov_b32 v1, s1
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v2, s2 :: v_dual_mov_b32 v3, s3
 ; GFX11-CU-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX11-CU-NEXT:    s_waitcnt_vscnt null, 0x0
 ; GFX11-CU-NEXT:    flat_atomic_cmpswap_b32 v2, v[0:1], v[2:3] offset:16 glc
 ; GFX11-CU-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX11-CU-NEXT:    buffer_gl0_inv
 ; GFX11-CU-NEXT:    buffer_gl1_inv
 ; GFX11-CU-NEXT:    flat_store_b32 v[0:1], v2
-; GFX11-CU-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-CU-NEXT:    s_endpgm
     ptr %out, i32 %in, i32 %old) {
 entry:
   %gep = getelementptr i32, ptr %out, i32 4
   %val = cmpxchg volatile ptr %gep, i32 %old, i32 %in release acquire
   %val0 = extractvalue { i32, i1 } %val, 0
   store i32 %val0, ptr %out, align 4
   ret void

 ; GFX11-WGP-NEXT:    v_dual_mov_b32 v2, s2 :: v_dual_mov_b32 v3, s3
 ; GFX11-WGP-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX11-WGP-NEXT:    s_waitcnt_vscnt null, 0x0
 ; GFX11-WGP-NEXT:    flat_atomic_cmpswap_b32 v2, v[0:1], v[2:3] offset:16 glc
 ; GFX11-WGP-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX11-WGP-NEXT:    buffer_gl0_inv
 ; GFX11-WGP-NEXT:    buffer_gl1_inv
 ; GFX11-WGP-NEXT:    flat_store_b32 v[0:1], v2
-; GFX11-WGP-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-WGP-NEXT:    s_endpgm
 ;
 ; GFX11-CU-LABEL: flat_system_acq_rel_acquire_ret_cmpxchg:
 ; GFX11-CU:       ; %bb.0: ; %entry
 ; GFX11-CU-NEXT:    s_load_b128 s[0:3], s[0:1], 0x0
 ; GFX11-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v0, s0 :: v_dual_mov_b32 v1, s1
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v2, s2 :: v_dual_mov_b32 v3, s3
 ; GFX11-CU-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX11-CU-NEXT:    s_waitcnt_vscnt null, 0x0
 ; GFX11-CU-NEXT:    flat_atomic_cmpswap_b32 v2, v[0:1], v[2:3] offset:16 glc
 ; GFX11-CU-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX11-CU-NEXT:    buffer_gl0_inv
 ; GFX11-CU-NEXT:    buffer_gl1_inv
 ; GFX11-CU-NEXT:    flat_store_b32 v[0:1], v2
-; GFX11-CU-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-CU-NEXT:    s_endpgm
     ptr %out, i32 %in, i32 %old) {
 entry:
   %gep = getelementptr i32, ptr %out, i32 4
   %val = cmpxchg volatile ptr %gep, i32 %old, i32 %in acq_rel acquire
   %val0 = extractvalue { i32, i1 } %val, 0
   store i32 %val0, ptr %out, align 4
   ret void

 ; GFX11-WGP-NEXT:    v_dual_mov_b32 v2, s2 :: v_dual_mov_b32 v3, s3
 ; GFX11-WGP-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX11-WGP-NEXT:    s_waitcnt_vscnt null, 0x0
 ; GFX11-WGP-NEXT:    flat_atomic_cmpswap_b32 v2, v[0:1], v[2:3] offset:16 glc
 ; GFX11-WGP-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX11-WGP-NEXT:    buffer_gl0_inv
 ; GFX11-WGP-NEXT:    buffer_gl1_inv
 ; GFX11-WGP-NEXT:    flat_store_b32 v[0:1], v2
-; GFX11-WGP-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-WGP-NEXT:    s_endpgm
 ;
 ; GFX11-CU-LABEL: flat_system_seq_cst_acquire_ret_cmpxchg:
 ; GFX11-CU:       ; %bb.0: ; %entry
 ; GFX11-CU-NEXT:    s_load_b128 s[0:3], s[0:1], 0x0
 ; GFX11-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v0, s0 :: v_dual_mov_b32 v1, s1
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v2, s2 :: v_dual_mov_b32 v3, s3
 ; GFX11-CU-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX11-CU-NEXT:    s_waitcnt_vscnt null, 0x0
 ; GFX11-CU-NEXT:    flat_atomic_cmpswap_b32 v2, v[0:1], v[2:3] offset:16 glc
 ; GFX11-CU-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX11-CU-NEXT:    buffer_gl0_inv
 ; GFX11-CU-NEXT:    buffer_gl1_inv
 ; GFX11-CU-NEXT:    flat_store_b32 v[0:1], v2
-; GFX11-CU-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-CU-NEXT:    s_endpgm
     ptr %out, i32 %in, i32 %old) {
 entry:
   %gep = getelementptr i32, ptr %out, i32 4
   %val = cmpxchg volatile ptr %gep, i32 %old, i32 %in seq_cst acquire
   %val0 = extractvalue { i32, i1 } %val, 0
   store i32 %val0, ptr %out, align 4
   ret void

 ; GFX11-WGP-NEXT:    v_dual_mov_b32 v2, s2 :: v_dual_mov_b32 v3, s3
 ; GFX11-WGP-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX11-WGP-NEXT:    s_waitcnt_vscnt null, 0x0
 ; GFX11-WGP-NEXT:    flat_atomic_cmpswap_b32 v2, v[0:1], v[2:3] offset:16 glc
 ; GFX11-WGP-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX11-WGP-NEXT:    buffer_gl0_inv
 ; GFX11-WGP-NEXT:    buffer_gl1_inv
 ; GFX11-WGP-NEXT:    flat_store_b32 v[0:1], v2
-; GFX11-WGP-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-WGP-NEXT:    s_endpgm
 ;
 ; GFX11-CU-LABEL: flat_system_monotonic_seq_cst_ret_cmpxchg:
 ; GFX11-CU:       ; %bb.0: ; %entry
 ; GFX11-CU-NEXT:    s_load_b128 s[0:3], s[0:1], 0x0
 ; GFX11-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v0, s0 :: v_dual_mov_b32 v1, s1
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v2, s2 :: v_dual_mov_b32 v3, s3
 ; GFX11-CU-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX11-CU-NEXT:    s_waitcnt_vscnt null, 0x0
 ; GFX11-CU-NEXT:    flat_atomic_cmpswap_b32 v2, v[0:1], v[2:3] offset:16 glc
 ; GFX11-CU-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX11-CU-NEXT:    buffer_gl0_inv
 ; GFX11-CU-NEXT:    buffer_gl1_inv
 ; GFX11-CU-NEXT:    flat_store_b32 v[0:1], v2
-; GFX11-CU-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-CU-NEXT:    s_endpgm
     ptr %out, i32 %in, i32 %old) {
 entry:
   %gep = getelementptr i32, ptr %out, i32 4
   %val = cmpxchg volatile ptr %gep, i32 %old, i32 %in monotonic seq_cst
   %val0 = extractvalue { i32, i1 } %val, 0
   store i32 %val0, ptr %out, align 4
   ret void

 ; GFX11-WGP-NEXT:    v_dual_mov_b32 v2, s2 :: v_dual_mov_b32 v3, s3
 ; GFX11-WGP-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX11-WGP-NEXT:    s_waitcnt_vscnt null, 0x0
 ; GFX11-WGP-NEXT:    flat_atomic_cmpswap_b32 v2, v[0:1], v[2:3] offset:16 glc
 ; GFX11-WGP-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX11-WGP-NEXT:    buffer_gl0_inv
 ; GFX11-WGP-NEXT:    buffer_gl1_inv
 ; GFX11-WGP-NEXT:    flat_store_b32 v[0:1], v2
-; GFX11-WGP-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-WGP-NEXT:    s_endpgm
 ;
 ; GFX11-CU-LABEL: flat_system_acquire_seq_cst_ret_cmpxchg:
 ; GFX11-CU:       ; %bb.0: ; %entry
 ; GFX11-CU-NEXT:    s_load_b128 s[0:3], s[0:1], 0x0
 ; GFX11-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v0, s0 :: v_dual_mov_b32 v1, s1
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v2, s2 :: v_dual_mov_b32 v3, s3
 ; GFX11-CU-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX11-CU-NEXT:    s_waitcnt_vscnt null, 0x0
 ; GFX11-CU-NEXT:    flat_atomic_cmpswap_b32 v2, v[0:1], v[2:3] offset:16 glc
 ; GFX11-CU-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX11-CU-NEXT:    buffer_gl0_inv
 ; GFX11-CU-NEXT:    buffer_gl1_inv
 ; GFX11-CU-NEXT:    flat_store_b32 v[0:1], v2
-; GFX11-CU-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-CU-NEXT:    s_endpgm
     ptr %out, i32 %in, i32 %old) {
 entry:
   %gep = getelementptr i32, ptr %out, i32 4
   %val = cmpxchg volatile ptr %gep, i32 %old, i32 %in acquire seq_cst
   %val0 = extractvalue { i32, i1 } %val, 0
   store i32 %val0, ptr %out, align 4
   ret void

 ; GFX11-WGP-NEXT:    v_dual_mov_b32 v2, s2 :: v_dual_mov_b32 v3, s3
 ; GFX11-WGP-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX11-WGP-NEXT:    s_waitcnt_vscnt null, 0x0
 ; GFX11-WGP-NEXT:    flat_atomic_cmpswap_b32 v2, v[0:1], v[2:3] offset:16 glc
 ; GFX11-WGP-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX11-WGP-NEXT:    buffer_gl0_inv
 ; GFX11-WGP-NEXT:    buffer_gl1_inv
 ; GFX11-WGP-NEXT:    flat_store_b32 v[0:1], v2
-; GFX11-WGP-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-WGP-NEXT:    s_endpgm
 ;
 ; GFX11-CU-LABEL: flat_system_release_seq_cst_ret_cmpxchg:
 ; GFX11-CU:       ; %bb.0: ; %entry
 ; GFX11-CU-NEXT:    s_load_b128 s[0:3], s[0:1], 0x0
 ; GFX11-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v0, s0 :: v_dual_mov_b32 v1, s1
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v2, s2 :: v_dual_mov_b32 v3, s3
 ; GFX11-CU-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX11-CU-NEXT:    s_waitcnt_vscnt null, 0x0
 ; GFX11-CU-NEXT:    flat_atomic_cmpswap_b32 v2, v[0:1], v[2:3] offset:16 glc
 ; GFX11-CU-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX11-CU-NEXT:    buffer_gl0_inv
 ; GFX11-CU-NEXT:    buffer_gl1_inv
 ; GFX11-CU-NEXT:    flat_store_b32 v[0:1], v2
-; GFX11-CU-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-CU-NEXT:    s_endpgm
     ptr %out, i32 %in, i32 %old) {
 entry:
   %gep = getelementptr i32, ptr %out, i32 4
   %val = cmpxchg volatile ptr %gep, i32 %old, i32 %in release seq_cst
   %val0 = extractvalue { i32, i1 } %val, 0
   store i32 %val0, ptr %out, align 4
   ret void

 ; GFX11-WGP-NEXT:    v_dual_mov_b32 v2, s2 :: v_dual_mov_b32 v3, s3
 ; GFX11-WGP-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX11-WGP-NEXT:    s_waitcnt_vscnt null, 0x0
 ; GFX11-WGP-NEXT:    flat_atomic_cmpswap_b32 v2, v[0:1], v[2:3] offset:16 glc
 ; GFX11-WGP-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX11-WGP-NEXT:    buffer_gl0_inv
 ; GFX11-WGP-NEXT:    buffer_gl1_inv
 ; GFX11-WGP-NEXT:    flat_store_b32 v[0:1], v2
-; GFX11-WGP-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-WGP-NEXT:    s_endpgm
 ;
 ; GFX11-CU-LABEL: flat_system_acq_rel_seq_cst_ret_cmpxchg:
 ; GFX11-CU:       ; %bb.0: ; %entry
 ; GFX11-CU-NEXT:    s_load_b128 s[0:3], s[0:1], 0x0
 ; GFX11-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v0, s0 :: v_dual_mov_b32 v1, s1
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v2, s2 :: v_dual_mov_b32 v3, s3
 ; GFX11-CU-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX11-CU-NEXT:    s_waitcnt_vscnt null, 0x0
 ; GFX11-CU-NEXT:    flat_atomic_cmpswap_b32 v2, v[0:1], v[2:3] offset:16 glc
 ; GFX11-CU-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX11-CU-NEXT:    buffer_gl0_inv
 ; GFX11-CU-NEXT:    buffer_gl1_inv
 ; GFX11-CU-NEXT:    flat_store_b32 v[0:1], v2
-; GFX11-CU-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-CU-NEXT:    s_endpgm
     ptr %out, i32 %in, i32 %old) {
 entry:
   %gep = getelementptr i32, ptr %out, i32 4
   %val = cmpxchg volatile ptr %gep, i32 %old, i32 %in acq_rel seq_cst
   %val0 = extractvalue { i32, i1 } %val, 0
   store i32 %val0, ptr %out, align 4
   ret void

 ; GFX11-WGP-NEXT:    v_dual_mov_b32 v2, s2 :: v_dual_mov_b32 v3, s3
 ; GFX11-WGP-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX11-WGP-NEXT:    s_waitcnt_vscnt null, 0x0
 ; GFX11-WGP-NEXT:    flat_atomic_cmpswap_b32 v2, v[0:1], v[2:3] offset:16 glc
 ; GFX11-WGP-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX11-WGP-NEXT:    buffer_gl0_inv
 ; GFX11-WGP-NEXT:    buffer_gl1_inv
 ; GFX11-WGP-NEXT:    flat_store_b32 v[0:1], v2
-; GFX11-WGP-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-WGP-NEXT:    s_endpgm
 ;
 ; GFX11-CU-LABEL: flat_system_seq_cst_seq_cst_ret_cmpxchg:
 ; GFX11-CU:       ; %bb.0: ; %entry
 ; GFX11-CU-NEXT:    s_load_b128 s[0:3], s[0:1], 0x0
 ; GFX11-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v0, s0 :: v_dual_mov_b32 v1, s1
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v2, s2 :: v_dual_mov_b32 v3, s3
 ; GFX11-CU-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX11-CU-NEXT:    s_waitcnt_vscnt null, 0x0
 ; GFX11-CU-NEXT:    flat_atomic_cmpswap_b32 v2, v[0:1], v[2:3] offset:16 glc
 ; GFX11-CU-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX11-CU-NEXT:    buffer_gl0_inv
 ; GFX11-CU-NEXT:    buffer_gl1_inv
 ; GFX11-CU-NEXT:    flat_store_b32 v[0:1], v2
-; GFX11-CU-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-CU-NEXT:    s_endpgm
     ptr %out, i32 %in, i32 %old) {
 entry:
   %gep = getelementptr i32, ptr %out, i32 4
   %val = cmpxchg volatile ptr %gep, i32 %old, i32 %in seq_cst seq_cst
   %val0 = extractvalue { i32, i1 } %val, 0
   store i32 %val0, ptr %out, align 4
   ret void

 ; GFX11-WGP:       ; %bb.0: ; %entry
 ; GFX11-WGP-NEXT:    s_load_b128 s[0:3], s[0:1], 0x0
 ; GFX11-WGP-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-WGP-NEXT:    v_dual_mov_b32 v0, s0 :: v_dual_mov_b32 v1, s1
 ; GFX11-WGP-NEXT:    flat_load_b32 v2, v[0:1]
 ; GFX11-WGP-NEXT:    v_dual_mov_b32 v0, s2 :: v_dual_mov_b32 v1, s3
 ; GFX11-WGP-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX11-WGP-NEXT:    flat_store_b32 v[0:1], v2
-; GFX11-WGP-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-WGP-NEXT:    s_endpgm
 ;
 ; GFX11-CU-LABEL: flat_system_one_as_unordered_load:
 ; GFX11-CU:       ; %bb.0: ; %entry
 ; GFX11-CU-NEXT:    s_load_b128 s[0:3], s[0:1], 0x0
 ; GFX11-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v0, s0 :: v_dual_mov_b32 v1, s1
 ; GFX11-CU-NEXT:    flat_load_b32 v2, v[0:1]
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v0, s2 :: v_dual_mov_b32 v1, s3
 ; GFX11-CU-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX11-CU-NEXT:    flat_store_b32 v[0:1], v2
-; GFX11-CU-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-CU-NEXT:    s_endpgm
     ptr %in, ptr %out) {
 entry:
   %val = load atomic i32, ptr %in syncscope("one-as") unordered, align 4
   store i32 %val, ptr %out
   ret void
 }
 

 ; GFX11-WGP:       ; %bb.0: ; %entry
 ; GFX11-WGP-NEXT:    s_load_b128 s[0:3], s[0:1], 0x0
 ; GFX11-WGP-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-WGP-NEXT:    v_dual_mov_b32 v0, s0 :: v_dual_mov_b32 v1, s1
 ; GFX11-WGP-NEXT:    flat_load_b32 v2, v[0:1] glc
 ; GFX11-WGP-NEXT:    v_dual_mov_b32 v0, s2 :: v_dual_mov_b32 v1, s3
 ; GFX11-WGP-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX11-WGP-NEXT:    flat_store_b32 v[0:1], v2
-; GFX11-WGP-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-WGP-NEXT:    s_endpgm
 ;
 ; GFX11-CU-LABEL: flat_system_one_as_monotonic_load:
 ; GFX11-CU:       ; %bb.0: ; %entry
 ; GFX11-CU-NEXT:    s_load_b128 s[0:3], s[0:1], 0x0
 ; GFX11-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v0, s0 :: v_dual_mov_b32 v1, s1
 ; GFX11-CU-NEXT:    flat_load_b32 v2, v[0:1] glc
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v0, s2 :: v_dual_mov_b32 v1, s3
 ; GFX11-CU-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX11-CU-NEXT:    flat_store_b32 v[0:1], v2
-; GFX11-CU-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-CU-NEXT:    s_endpgm
     ptr %in, ptr %out) {
 entry:
   %val = load atomic i32, ptr %in syncscope("one-as") monotonic, align 4
   store i32 %val, ptr %out
   ret void
 }
 

 ; GFX11-WGP-NEXT:    v_dual_mov_b32 v0, s0 :: v_dual_mov_b32 v1, s1
 ; GFX11-WGP-NEXT:    flat_load_b32 v2, v[0:1] glc
 ; GFX11-WGP-NEXT:    s_waitcnt vmcnt(0)
 ; GFX11-WGP-NEXT:    buffer_gl0_inv
 ; GFX11-WGP-NEXT:    buffer_gl1_inv
 ; GFX11-WGP-NEXT:    v_dual_mov_b32 v0, s2 :: v_dual_mov_b32 v1, s3
 ; GFX11-WGP-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-WGP-NEXT:    flat_store_b32 v[0:1], v2
-; GFX11-WGP-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-WGP-NEXT:    s_endpgm
 ;
 ; GFX11-CU-LABEL: flat_system_one_as_acquire_load:
 ; GFX11-CU:       ; %bb.0: ; %entry
 ; GFX11-CU-NEXT:    s_load_b128 s[0:3], s[0:1], 0x0
 ; GFX11-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v0, s0 :: v_dual_mov_b32 v1, s1
 ; GFX11-CU-NEXT:    flat_load_b32 v2, v[0:1] glc
 ; GFX11-CU-NEXT:    s_waitcnt vmcnt(0)
 ; GFX11-CU-NEXT:    buffer_gl0_inv
 ; GFX11-CU-NEXT:    buffer_gl1_inv
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v0, s2 :: v_dual_mov_b32 v1, s3
 ; GFX11-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-CU-NEXT:    flat_store_b32 v[0:1], v2
-; GFX11-CU-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-CU-NEXT:    s_endpgm
     ptr %in, ptr %out) {
 entry:
   %val = load atomic i32, ptr %in syncscope("one-as") acquire, align 4
   store i32 %val, ptr %out
   ret void
 }
 

 ; GFX11-WGP-NEXT:    s_waitcnt_vscnt null, 0x0
 ; GFX11-WGP-NEXT:    flat_load_b32 v2, v[0:1] glc
 ; GFX11-WGP-NEXT:    s_waitcnt vmcnt(0)
 ; GFX11-WGP-NEXT:    buffer_gl0_inv
 ; GFX11-WGP-NEXT:    buffer_gl1_inv
 ; GFX11-WGP-NEXT:    v_dual_mov_b32 v0, s2 :: v_dual_mov_b32 v1, s3
 ; GFX11-WGP-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-WGP-NEXT:    flat_store_b32 v[0:1], v2
-; GFX11-WGP-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-WGP-NEXT:    s_endpgm
 ;
 ; GFX11-CU-LABEL: flat_system_one_as_seq_cst_load:
 ; GFX11-CU:       ; %bb.0: ; %entry
 ; GFX11-CU-NEXT:    s_load_b128 s[0:3], s[0:1], 0x0
 ; GFX11-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v0, s0 :: v_dual_mov_b32 v1, s1
 ; GFX11-CU-NEXT:    s_waitcnt vmcnt(0)
 ; GFX11-CU-NEXT:    s_waitcnt_vscnt null, 0x0
 ; GFX11-CU-NEXT:    flat_load_b32 v2, v[0:1] glc
 ; GFX11-CU-NEXT:    s_waitcnt vmcnt(0)
 ; GFX11-CU-NEXT:    buffer_gl0_inv
 ; GFX11-CU-NEXT:    buffer_gl1_inv
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v0, s2 :: v_dual_mov_b32 v1, s3
 ; GFX11-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-CU-NEXT:    flat_store_b32 v[0:1], v2
-; GFX11-CU-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-CU-NEXT:    s_endpgm
     ptr %in, ptr %out) {
 entry:
   %val = load atomic i32, ptr %in syncscope("one-as") seq_cst, align 4
   store i32 %val, ptr %out
   ret void
 }
 

 ; GFX11-WGP:       ; %bb.0: ; %entry
 ; GFX11-WGP-NEXT:    s_clause 0x1
 ; GFX11-WGP-NEXT:    s_load_b64 s[2:3], s[0:1], 0x8
 ; GFX11-WGP-NEXT:    s_load_b32 s0, s[0:1], 0x0
 ; GFX11-WGP-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-WGP-NEXT:    v_dual_mov_b32 v0, s2 :: v_dual_mov_b32 v1, s3
 ; GFX11-WGP-NEXT:    v_mov_b32_e32 v2, s0
 ; GFX11-WGP-NEXT:    flat_store_b32 v[0:1], v2
-; GFX11-WGP-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-WGP-NEXT:    s_endpgm
 ;
 ; GFX11-CU-LABEL: flat_system_one_as_unordered_store:
 ; GFX11-CU:       ; %bb.0: ; %entry
 ; GFX11-CU-NEXT:    s_clause 0x1
 ; GFX11-CU-NEXT:    s_load_b64 s[2:3], s[0:1], 0x8
 ; GFX11-CU-NEXT:    s_load_b32 s0, s[0:1], 0x0
 ; GFX11-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v0, s2 :: v_dual_mov_b32 v1, s3
 ; GFX11-CU-NEXT:    v_mov_b32_e32 v2, s0
 ; GFX11-CU-NEXT:    flat_store_b32 v[0:1], v2
-; GFX11-CU-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-CU-NEXT:    s_endpgm
     i32 %in, ptr %out) {
 entry:
   store atomic i32 %in, ptr %out syncscope("one-as") unordered, align 4
   ret void
 }
 
 define amdgpu_kernel void @flat_system_one_as_monotonic_store(

 ; GFX11-WGP:       ; %bb.0: ; %entry
 ; GFX11-WGP-NEXT:    s_clause 0x1
 ; GFX11-WGP-NEXT:    s_load_b64 s[2:3], s[0:1], 0x8
 ; GFX11-WGP-NEXT:    s_load_b32 s0, s[0:1], 0x0
 ; GFX11-WGP-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-WGP-NEXT:    v_dual_mov_b32 v0, s2 :: v_dual_mov_b32 v1, s3
 ; GFX11-WGP-NEXT:    v_mov_b32_e32 v2, s0
 ; GFX11-WGP-NEXT:    flat_store_b32 v[0:1], v2
-; GFX11-WGP-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-WGP-NEXT:    s_endpgm
 ;
 ; GFX11-CU-LABEL: flat_system_one_as_monotonic_store:
 ; GFX11-CU:       ; %bb.0: ; %entry
 ; GFX11-CU-NEXT:    s_clause 0x1
 ; GFX11-CU-NEXT:    s_load_b64 s[2:3], s[0:1], 0x8
 ; GFX11-CU-NEXT:    s_load_b32 s0, s[0:1], 0x0
 ; GFX11-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v0, s2 :: v_dual_mov_b32 v1, s3
 ; GFX11-CU-NEXT:    v_mov_b32_e32 v2, s0
 ; GFX11-CU-NEXT:    flat_store_b32 v[0:1], v2
-; GFX11-CU-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-CU-NEXT:    s_endpgm
     i32 %in, ptr %out) {
 entry:
   store atomic i32 %in, ptr %out syncscope("one-as") monotonic, align 4
   ret void
 }
 
 define amdgpu_kernel void @flat_system_one_as_release_store(

 ; GFX11-WGP-NEXT:    s_load_b64 s[2:3], s[0:1], 0x8
 ; GFX11-WGP-NEXT:    s_load_b32 s0, s[0:1], 0x0
 ; GFX11-WGP-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-WGP-NEXT:    v_dual_mov_b32 v0, s2 :: v_dual_mov_b32 v1, s3
 ; GFX11-WGP-NEXT:    v_mov_b32_e32 v2, s0
 ; GFX11-WGP-NEXT:    s_waitcnt vmcnt(0)
 ; GFX11-WGP-NEXT:    s_waitcnt_vscnt null, 0x0
 ; GFX11-WGP-NEXT:    flat_store_b32 v[0:1], v2
-; GFX11-WGP-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-WGP-NEXT:    s_endpgm
 ;
 ; GFX11-CU-LABEL: flat_system_one_as_release_store:
 ; GFX11-CU:       ; %bb.0: ; %entry
 ; GFX11-CU-NEXT:    s_clause 0x1
 ; GFX11-CU-NEXT:    s_load_b64 s[2:3], s[0:1], 0x8
 ; GFX11-CU-NEXT:    s_load_b32 s0, s[0:1], 0x0
 ; GFX11-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v0, s2 :: v_dual_mov_b32 v1, s3
 ; GFX11-CU-NEXT:    v_mov_b32_e32 v2, s0
 ; GFX11-CU-NEXT:    s_waitcnt vmcnt(0)
 ; GFX11-CU-NEXT:    s_waitcnt_vscnt null, 0x0
 ; GFX11-CU-NEXT:    flat_store_b32 v[0:1], v2
-; GFX11-CU-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-CU-NEXT:    s_endpgm
     i32 %in, ptr %out) {
 entry:
   store atomic i32 %in, ptr %out syncscope("one-as") release, align 4
   ret void
 }
 
 define amdgpu_kernel void @flat_system_one_as_seq_cst_store(

 ; GFX11-WGP-NEXT:    s_load_b64 s[2:3], s[0:1], 0x8
 ; GFX11-WGP-NEXT:    s_load_b32 s0, s[0:1], 0x0
 ; GFX11-WGP-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-WGP-NEXT:    v_dual_mov_b32 v0, s2 :: v_dual_mov_b32 v1, s3
 ; GFX11-WGP-NEXT:    v_mov_b32_e32 v2, s0
 ; GFX11-WGP-NEXT:    s_waitcnt vmcnt(0)
 ; GFX11-WGP-NEXT:    s_waitcnt_vscnt null, 0x0
 ; GFX11-WGP-NEXT:    flat_store_b32 v[0:1], v2
-; GFX11-WGP-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-WGP-NEXT:    s_endpgm
 ;
 ; GFX11-CU-LABEL: flat_system_one_as_seq_cst_store:
 ; GFX11-CU:       ; %bb.0: ; %entry
 ; GFX11-CU-NEXT:    s_clause 0x1
 ; GFX11-CU-NEXT:    s_load_b64 s[2:3], s[0:1], 0x8
 ; GFX11-CU-NEXT:    s_load_b32 s0, s[0:1], 0x0
 ; GFX11-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v0, s2 :: v_dual_mov_b32 v1, s3
 ; GFX11-CU-NEXT:    v_mov_b32_e32 v2, s0
 ; GFX11-CU-NEXT:    s_waitcnt vmcnt(0)
 ; GFX11-CU-NEXT:    s_waitcnt_vscnt null, 0x0
 ; GFX11-CU-NEXT:    flat_store_b32 v[0:1], v2
-; GFX11-CU-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-CU-NEXT:    s_endpgm
     i32 %in, ptr %out) {
 entry:
   store atomic i32 %in, ptr %out syncscope("one-as") seq_cst, align 4
   ret void
 }
 
 define amdgpu_kernel void @flat_system_one_as_monotonic_atomicrmw(

 ; GFX11-WGP:       ; %bb.0: ; %entry
 ; GFX11-WGP-NEXT:    s_clause 0x1
 ; GFX11-WGP-NEXT:    s_load_b64 s[2:3], s[0:1], 0x0
 ; GFX11-WGP-NEXT:    s_load_b32 s0, s[0:1], 0x8
 ; GFX11-WGP-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-WGP-NEXT:    v_dual_mov_b32 v0, s2 :: v_dual_mov_b32 v1, s3
 ; GFX11-WGP-NEXT:    v_mov_b32_e32 v2, s0
 ; GFX11-WGP-NEXT:    flat_atomic_swap_b32 v[0:1], v2
-; GFX11-WGP-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-WGP-NEXT:    s_endpgm
 ;
 ; GFX11-CU-LABEL: flat_system_one_as_monotonic_atomicrmw:
 ; GFX11-CU:       ; %bb.0: ; %entry
 ; GFX11-CU-NEXT:    s_clause 0x1
 ; GFX11-CU-NEXT:    s_load_b64 s[2:3], s[0:1], 0x0
 ; GFX11-CU-NEXT:    s_load_b32 s0, s[0:1], 0x8
 ; GFX11-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v0, s2 :: v_dual_mov_b32 v1, s3
 ; GFX11-CU-NEXT:    v_mov_b32_e32 v2, s0
 ; GFX11-CU-NEXT:    flat_atomic_swap_b32 v[0:1], v2
-; GFX11-CU-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-CU-NEXT:    s_endpgm
     ptr %out, i32 %in) {
 entry:
   %val = atomicrmw volatile xchg ptr %out, i32 %in syncscope("one-as") monotonic
   ret void
 }
 
 define amdgpu_kernel void @flat_system_one_as_acquire_atomicrmw(

 ; GFX11-WGP-NEXT:    s_load_b64 s[2:3], s[0:1], 0x0
 ; GFX11-WGP-NEXT:    s_load_b32 s0, s[0:1], 0x8
 ; GFX11-WGP-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-WGP-NEXT:    v_dual_mov_b32 v0, s2 :: v_dual_mov_b32 v1, s3
 ; GFX11-WGP-NEXT:    v_mov_b32_e32 v2, s0
 ; GFX11-WGP-NEXT:    s_waitcnt vmcnt(0)
 ; GFX11-WGP-NEXT:    s_waitcnt_vscnt null, 0x0
 ; GFX11-WGP-NEXT:    flat_atomic_swap_b32 v[0:1], v2
-; GFX11-WGP-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-WGP-NEXT:    s_endpgm
 ;
 ; GFX11-CU-LABEL: flat_system_one_as_release_atomicrmw:
 ; GFX11-CU:       ; %bb.0: ; %entry
 ; GFX11-CU-NEXT:    s_clause 0x1
 ; GFX11-CU-NEXT:    s_load_b64 s[2:3], s[0:1], 0x0
 ; GFX11-CU-NEXT:    s_load_b32 s0, s[0:1], 0x8
 ; GFX11-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v0, s2 :: v_dual_mov_b32 v1, s3
 ; GFX11-CU-NEXT:    v_mov_b32_e32 v2, s0
 ; GFX11-CU-NEXT:    s_waitcnt vmcnt(0)
 ; GFX11-CU-NEXT:    s_waitcnt_vscnt null, 0x0
 ; GFX11-CU-NEXT:    flat_atomic_swap_b32 v[0:1], v2
-; GFX11-CU-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-CU-NEXT:    s_endpgm
     ptr %out, i32 %in) {
 entry:
   %val = atomicrmw volatile xchg ptr %out, i32 %in syncscope("one-as") release
   ret void
 }
 
 define amdgpu_kernel void @flat_system_one_as_acq_rel_atomicrmw(

 ; GFX11-WGP-NEXT:    v_dual_mov_b32 v0, s2 :: v_dual_mov_b32 v1, s3
 ; GFX11-WGP-NEXT:    v_mov_b32_e32 v2, s0
 ; GFX11-WGP-NEXT:    flat_atomic_swap_b32 v2, v[0:1], v2 glc
 ; GFX11-WGP-NEXT:    s_waitcnt vmcnt(0)
 ; GFX11-WGP-NEXT:    buffer_gl0_inv
 ; GFX11-WGP-NEXT:    buffer_gl1_inv
 ; GFX11-WGP-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-WGP-NEXT:    flat_store_b32 v[0:1], v2
-; GFX11-WGP-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-WGP-NEXT:    s_endpgm
 ;
 ; GFX11-CU-LABEL: flat_system_one_as_acquire_ret_atomicrmw:
 ; GFX11-CU:       ; %bb.0: ; %entry
 ; GFX11-CU-NEXT:    s_clause 0x1
 ; GFX11-CU-NEXT:    s_load_b64 s[2:3], s[0:1], 0x0
 ; GFX11-CU-NEXT:    s_load_b32 s0, s[0:1], 0x8
 ; GFX11-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v0, s2 :: v_dual_mov_b32 v1, s3
 ; GFX11-CU-NEXT:    v_mov_b32_e32 v2, s0
 ; GFX11-CU-NEXT:    flat_atomic_swap_b32 v2, v[0:1], v2 glc
 ; GFX11-CU-NEXT:    s_waitcnt vmcnt(0)
 ; GFX11-CU-NEXT:    buffer_gl0_inv
 ; GFX11-CU-NEXT:    buffer_gl1_inv
 ; GFX11-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-CU-NEXT:    flat_store_b32 v[0:1], v2
-; GFX11-CU-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-CU-NEXT:    s_endpgm
     ptr %out, i32 %in) {
 entry:
   %val = atomicrmw volatile xchg ptr %out, i32 %in syncscope("one-as") acquire
   store i32 %val, ptr %out, align 4
   ret void
 }
 

 ; GFX11-WGP-NEXT:    s_waitcnt vmcnt(0)
 ; GFX11-WGP-NEXT:    s_waitcnt_vscnt null, 0x0
 ; GFX11-WGP-NEXT:    flat_atomic_swap_b32 v2, v[0:1], v2 glc
 ; GFX11-WGP-NEXT:    s_waitcnt vmcnt(0)
 ; GFX11-WGP-NEXT:    buffer_gl0_inv
 ; GFX11-WGP-NEXT:    buffer_gl1_inv
 ; GFX11-WGP-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-WGP-NEXT:    flat_store_b32 v[0:1], v2
-; GFX11-WGP-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-WGP-NEXT:    s_endpgm
 ;
 ; GFX11-CU-LABEL: flat_system_one_as_acq_rel_ret_atomicrmw:
 ; GFX11-CU:       ; %bb.0: ; %entry
 ; GFX11-CU-NEXT:    s_clause 0x1
 ; GFX11-CU-NEXT:    s_load_b64 s[2:3], s[0:1], 0x0
 ; GFX11-CU-NEXT:    s_load_b32 s0, s[0:1], 0x8
 ; GFX11-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v0, s2 :: v_dual_mov_b32 v1, s3
 ; GFX11-CU-NEXT:    v_mov_b32_e32 v2, s0
 ; GFX11-CU-NEXT:    s_waitcnt vmcnt(0)
 ; GFX11-CU-NEXT:    s_waitcnt_vscnt null, 0x0
 ; GFX11-CU-NEXT:    flat_atomic_swap_b32 v2, v[0:1], v2 glc
 ; GFX11-CU-NEXT:    s_waitcnt vmcnt(0)
 ; GFX11-CU-NEXT:    buffer_gl0_inv
 ; GFX11-CU-NEXT:    buffer_gl1_inv
 ; GFX11-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-CU-NEXT:    flat_store_b32 v[0:1], v2
-; GFX11-CU-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-CU-NEXT:    s_endpgm
     ptr %out, i32 %in) {
 entry:
   %val = atomicrmw volatile xchg ptr %out, i32 %in syncscope("one-as") acq_rel
   store i32 %val, ptr %out, align 4
   ret void
 }
 

 ; GFX11-WGP-NEXT:    s_waitcnt vmcnt(0)
 ; GFX11-WGP-NEXT:    s_waitcnt_vscnt null, 0x0
 ; GFX11-WGP-NEXT:    flat_atomic_swap_b32 v2, v[0:1], v2 glc
 ; GFX11-WGP-NEXT:    s_waitcnt vmcnt(0)
 ; GFX11-WGP-NEXT:    buffer_gl0_inv
 ; GFX11-WGP-NEXT:    buffer_gl1_inv
 ; GFX11-WGP-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-WGP-NEXT:    flat_store_b32 v[0:1], v2
-; GFX11-WGP-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-WGP-NEXT:    s_endpgm
 ;
 ; GFX11-CU-LABEL: flat_system_one_as_seq_cst_ret_atomicrmw:
 ; GFX11-CU:       ; %bb.0: ; %entry
 ; GFX11-CU-NEXT:    s_clause 0x1
 ; GFX11-CU-NEXT:    s_load_b64 s[2:3], s[0:1], 0x0
 ; GFX11-CU-NEXT:    s_load_b32 s0, s[0:1], 0x8
 ; GFX11-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v0, s2 :: v_dual_mov_b32 v1, s3
 ; GFX11-CU-NEXT:    v_mov_b32_e32 v2, s0
 ; GFX11-CU-NEXT:    s_waitcnt vmcnt(0)
 ; GFX11-CU-NEXT:    s_waitcnt_vscnt null, 0x0
 ; GFX11-CU-NEXT:    flat_atomic_swap_b32 v2, v[0:1], v2 glc
 ; GFX11-CU-NEXT:    s_waitcnt vmcnt(0)
 ; GFX11-CU-NEXT:    buffer_gl0_inv
 ; GFX11-CU-NEXT:    buffer_gl1_inv
 ; GFX11-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-CU-NEXT:    flat_store_b32 v[0:1], v2
-; GFX11-CU-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-CU-NEXT:    s_endpgm
     ptr %out, i32 %in) {
 entry:
   %val = atomicrmw volatile xchg ptr %out, i32 %in syncscope("one-as") seq_cst
   store i32 %val, ptr %out, align 4
   ret void
 }
 

 ;
 ; GFX11-WGP-LABEL: flat_system_one_as_monotonic_monotonic_cmpxchg:
 ; GFX11-WGP:       ; %bb.0: ; %entry
 ; GFX11-WGP-NEXT:    s_load_b128 s[0:3], s[0:1], 0x0
 ; GFX11-WGP-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-WGP-NEXT:    v_dual_mov_b32 v0, s0 :: v_dual_mov_b32 v1, s1
 ; GFX11-WGP-NEXT:    v_dual_mov_b32 v2, s2 :: v_dual_mov_b32 v3, s3
 ; GFX11-WGP-NEXT:    flat_atomic_cmpswap_b32 v[0:1], v[2:3] offset:16
-; GFX11-WGP-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-WGP-NEXT:    s_endpgm
 ;
 ; GFX11-CU-LABEL: flat_system_one_as_monotonic_monotonic_cmpxchg:
 ; GFX11-CU:       ; %bb.0: ; %entry
 ; GFX11-CU-NEXT:    s_load_b128 s[0:3], s[0:1], 0x0
 ; GFX11-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v0, s0 :: v_dual_mov_b32 v1, s1
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v2, s2 :: v_dual_mov_b32 v3, s3
 ; GFX11-CU-NEXT:    flat_atomic_cmpswap_b32 v[0:1], v[2:3] offset:16
-; GFX11-CU-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-CU-NEXT:    s_endpgm
     ptr %out, i32 %in, i32 %old) {
 entry:
   %gep = getelementptr i32, ptr %out, i32 4
   %val = cmpxchg volatile ptr %gep, i32 %old, i32 %in syncscope("one-as") monotonic monotonic
   ret void
 }
 

 ; GFX11-WGP:       ; %bb.0: ; %entry
 ; GFX11-WGP-NEXT:    s_load_b128 s[0:3], s[0:1], 0x0
 ; GFX11-WGP-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-WGP-NEXT:    v_dual_mov_b32 v0, s0 :: v_dual_mov_b32 v1, s1
 ; GFX11-WGP-NEXT:    v_dual_mov_b32 v2, s2 :: v_dual_mov_b32 v3, s3
 ; GFX11-WGP-NEXT:    s_waitcnt vmcnt(0)
 ; GFX11-WGP-NEXT:    s_waitcnt_vscnt null, 0x0
 ; GFX11-WGP-NEXT:    flat_atomic_cmpswap_b32 v[0:1], v[2:3] offset:16
-; GFX11-WGP-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-WGP-NEXT:    s_endpgm
 ;
 ; GFX11-CU-LABEL: flat_system_one_as_release_monotonic_cmpxchg:
 ; GFX11-CU:       ; %bb.0: ; %entry
 ; GFX11-CU-NEXT:    s_load_b128 s[0:3], s[0:1], 0x0
 ; GFX11-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v0, s0 :: v_dual_mov_b32 v1, s1
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v2, s2 :: v_dual_mov_b32 v3, s3
 ; GFX11-CU-NEXT:    s_waitcnt vmcnt(0)
 ; GFX11-CU-NEXT:    s_waitcnt_vscnt null, 0x0
 ; GFX11-CU-NEXT:    flat_atomic_cmpswap_b32 v[0:1], v[2:3] offset:16
-; GFX11-CU-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-CU-NEXT:    s_endpgm
     ptr %out, i32 %in, i32 %old) {
 entry:
   %gep = getelementptr i32, ptr %out, i32 4
   %val = cmpxchg volatile ptr %gep, i32 %old, i32 %in syncscope("one-as") release monotonic
   ret void
 }
 

 ; GFX11-WGP:       ; %bb.0: ; %entry
 ; GFX11-WGP-NEXT:    s_load_b128 s[0:3], s[0:1], 0x0
 ; GFX11-WGP-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-WGP-NEXT:    v_dual_mov_b32 v0, s0 :: v_dual_mov_b32 v1, s1
 ; GFX11-WGP-NEXT:    v_dual_mov_b32 v2, s2 :: v_dual_mov_b32 v3, s3
 ; GFX11-WGP-NEXT:    flat_atomic_cmpswap_b32 v2, v[0:1], v[2:3] offset:16 glc
 ; GFX11-WGP-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX11-WGP-NEXT:    flat_store_b32 v[0:1], v2
-; GFX11-WGP-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-WGP-NEXT:    s_endpgm
 ;
 ; GFX11-CU-LABEL: flat_system_one_as_monotonic_monotonic_ret_cmpxchg:
 ; GFX11-CU:       ; %bb.0: ; %entry
 ; GFX11-CU-NEXT:    s_load_b128 s[0:3], s[0:1], 0x0
 ; GFX11-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v0, s0 :: v_dual_mov_b32 v1, s1
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v2, s2 :: v_dual_mov_b32 v3, s3
 ; GFX11-CU-NEXT:    flat_atomic_cmpswap_b32 v2, v[0:1], v[2:3] offset:16 glc
 ; GFX11-CU-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX11-CU-NEXT:    flat_store_b32 v[0:1], v2
-; GFX11-CU-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-CU-NEXT:    s_endpgm
     ptr %out, i32 %in, i32 %old) {
 entry:
   %gep = getelementptr i32, ptr %out, i32 4
   %val = cmpxchg volatile ptr %gep, i32 %old, i32 %in syncscope("one-as") monotonic monotonic
   %val0 = extractvalue { i32, i1 } %val, 0
   store i32 %val0, ptr %out, align 4
   ret void

 ; GFX11-WGP-NEXT:    v_dual_mov_b32 v0, s0 :: v_dual_mov_b32 v1, s1
 ; GFX11-WGP-NEXT:    v_dual_mov_b32 v2, s2 :: v_dual_mov_b32 v3, s3
 ; GFX11-WGP-NEXT:    flat_atomic_cmpswap_b32 v2, v[0:1], v[2:3] offset:16 glc
 ; GFX11-WGP-NEXT:    s_waitcnt vmcnt(0)
 ; GFX11-WGP-NEXT:    buffer_gl0_inv
 ; GFX11-WGP-NEXT:    buffer_gl1_inv
 ; GFX11-WGP-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-WGP-NEXT:    flat_store_b32 v[0:1], v2
-; GFX11-WGP-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-WGP-NEXT:    s_endpgm
 ;
 ; GFX11-CU-LABEL: flat_system_one_as_acquire_monotonic_ret_cmpxchg:
 ; GFX11-CU:       ; %bb.0: ; %entry
 ; GFX11-CU-NEXT:    s_load_b128 s[0:3], s[0:1], 0x0
 ; GFX11-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v0, s0 :: v_dual_mov_b32 v1, s1
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v2, s2 :: v_dual_mov_b32 v3, s3
 ; GFX11-CU-NEXT:    flat_atomic_cmpswap_b32 v2, v[0:1], v[2:3] offset:16 glc
 ; GFX11-CU-NEXT:    s_waitcnt vmcnt(0)
 ; GFX11-CU-NEXT:    buffer_gl0_inv
 ; GFX11-CU-NEXT:    buffer_gl1_inv
 ; GFX11-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-CU-NEXT:    flat_store_b32 v[0:1], v2
-; GFX11-CU-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-CU-NEXT:    s_endpgm
     ptr %out, i32 %in, i32 %old) {
 entry:
   %gep = getelementptr i32, ptr %out, i32 4
   %val = cmpxchg volatile ptr %gep, i32 %old, i32 %in syncscope("one-as") acquire monotonic
   %val0 = extractvalue { i32, i1 } %val, 0
   store i32 %val0, ptr %out, align 4
   ret void

 ; GFX11-WGP-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-WGP-NEXT:    v_dual_mov_b32 v0, s0 :: v_dual_mov_b32 v1, s1
 ; GFX11-WGP-NEXT:    v_dual_mov_b32 v2, s2 :: v_dual_mov_b32 v3, s3
 ; GFX11-WGP-NEXT:    s_waitcnt vmcnt(0)
 ; GFX11-WGP-NEXT:    s_waitcnt_vscnt null, 0x0
 ; GFX11-WGP-NEXT:    flat_atomic_cmpswap_b32 v2, v[0:1], v[2:3] offset:16 glc
 ; GFX11-WGP-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX11-WGP-NEXT:    flat_store_b32 v[0:1], v2
-; GFX11-WGP-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-WGP-NEXT:    s_endpgm
 ;
 ; GFX11-CU-LABEL: flat_system_one_as_release_monotonic_ret_cmpxchg:
 ; GFX11-CU:       ; %bb.0: ; %entry
 ; GFX11-CU-NEXT:    s_load_b128 s[0:3], s[0:1], 0x0
 ; GFX11-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v0, s0 :: v_dual_mov_b32 v1, s1
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v2, s2 :: v_dual_mov_b32 v3, s3
 ; GFX11-CU-NEXT:    s_waitcnt vmcnt(0)
 ; GFX11-CU-NEXT:    s_waitcnt_vscnt null, 0x0
 ; GFX11-CU-NEXT:    flat_atomic_cmpswap_b32 v2, v[0:1], v[2:3] offset:16 glc
 ; GFX11-CU-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX11-CU-NEXT:    flat_store_b32 v[0:1], v2
-; GFX11-CU-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-CU-NEXT:    s_endpgm
     ptr %out, i32 %in, i32 %old) {
 entry:
   %gep = getelementptr i32, ptr %out, i32 4
   %val = cmpxchg volatile ptr %gep, i32 %old, i32 %in syncscope("one-as") release monotonic
   %val0 = extractvalue { i32, i1 } %val, 0
   store i32 %val0, ptr %out, align 4
   ret void

 ; GFX11-WGP-NEXT:    s_waitcnt vmcnt(0)
 ; GFX11-WGP-NEXT:    s_waitcnt_vscnt null, 0x0
 ; GFX11-WGP-NEXT:    flat_atomic_cmpswap_b32 v2, v[0:1], v[2:3] offset:16 glc
 ; GFX11-WGP-NEXT:    s_waitcnt vmcnt(0)
 ; GFX11-WGP-NEXT:    buffer_gl0_inv
 ; GFX11-WGP-NEXT:    buffer_gl1_inv
 ; GFX11-WGP-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-WGP-NEXT:    flat_store_b32 v[0:1], v2
-; GFX11-WGP-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-WGP-NEXT:    s_endpgm
 ;
 ; GFX11-CU-LABEL: flat_system_one_as_acq_rel_monotonic_ret_cmpxchg:
 ; GFX11-CU:       ; %bb.0: ; %entry
 ; GFX11-CU-NEXT:    s_load_b128 s[0:3], s[0:1], 0x0
 ; GFX11-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v0, s0 :: v_dual_mov_b32 v1, s1
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v2, s2 :: v_dual_mov_b32 v3, s3
 ; GFX11-CU-NEXT:    s_waitcnt vmcnt(0)
 ; GFX11-CU-NEXT:    s_waitcnt_vscnt null, 0x0
 ; GFX11-CU-NEXT:    flat_atomic_cmpswap_b32 v2, v[0:1], v[2:3] offset:16 glc
 ; GFX11-CU-NEXT:    s_waitcnt vmcnt(0)
 ; GFX11-CU-NEXT:    buffer_gl0_inv
 ; GFX11-CU-NEXT:    buffer_gl1_inv
 ; GFX11-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-CU-NEXT:    flat_store_b32 v[0:1], v2
-; GFX11-CU-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-CU-NEXT:    s_endpgm
     ptr %out, i32 %in, i32 %old) {
 entry:
   %gep = getelementptr i32, ptr %out, i32 4
   %val = cmpxchg volatile ptr %gep, i32 %old, i32 %in syncscope("one-as") acq_rel monotonic
   %val0 = extractvalue { i32, i1 } %val, 0
   store i32 %val0, ptr %out, align 4
   ret void

 ; GFX11-WGP-NEXT:    s_waitcnt vmcnt(0)
 ; GFX11-WGP-NEXT:    s_waitcnt_vscnt null, 0x0
 ; GFX11-WGP-NEXT:    flat_atomic_cmpswap_b32 v2, v[0:1], v[2:3] offset:16 glc
 ; GFX11-WGP-NEXT:    s_waitcnt vmcnt(0)
 ; GFX11-WGP-NEXT:    buffer_gl0_inv
 ; GFX11-WGP-NEXT:    buffer_gl1_inv
 ; GFX11-WGP-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-WGP-NEXT:    flat_store_b32 v[0:1], v2
-; GFX11-WGP-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-WGP-NEXT:    s_endpgm
 ;
 ; GFX11-CU-LABEL: flat_system_one_as_seq_cst_monotonic_ret_cmpxchg:
 ; GFX11-CU:       ; %bb.0: ; %entry
 ; GFX11-CU-NEXT:    s_load_b128 s[0:3], s[0:1], 0x0
 ; GFX11-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v0, s0 :: v_dual_mov_b32 v1, s1
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v2, s2 :: v_dual_mov_b32 v3, s3
 ; GFX11-CU-NEXT:    s_waitcnt vmcnt(0)
 ; GFX11-CU-NEXT:    s_waitcnt_vscnt null, 0x0
 ; GFX11-CU-NEXT:    flat_atomic_cmpswap_b32 v2, v[0:1], v[2:3] offset:16 glc
 ; GFX11-CU-NEXT:    s_waitcnt vmcnt(0)
 ; GFX11-CU-NEXT:    buffer_gl0_inv
 ; GFX11-CU-NEXT:    buffer_gl1_inv
 ; GFX11-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-CU-NEXT:    flat_store_b32 v[0:1], v2
-; GFX11-CU-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-CU-NEXT:    s_endpgm
     ptr %out, i32 %in, i32 %old) {
 entry:
   %gep = getelementptr i32, ptr %out, i32 4
   %val = cmpxchg volatile ptr %gep, i32 %old, i32 %in syncscope("one-as") seq_cst monotonic
   %val0 = extractvalue { i32, i1 } %val, 0
   store i32 %val0, ptr %out, align 4
   ret void

 ; GFX11-WGP-NEXT:    v_dual_mov_b32 v0, s0 :: v_dual_mov_b32 v1, s1
 ; GFX11-WGP-NEXT:    v_dual_mov_b32 v2, s2 :: v_dual_mov_b32 v3, s3
 ; GFX11-WGP-NEXT:    flat_atomic_cmpswap_b32 v2, v[0:1], v[2:3] offset:16 glc
 ; GFX11-WGP-NEXT:    s_waitcnt vmcnt(0)
 ; GFX11-WGP-NEXT:    buffer_gl0_inv
 ; GFX11-WGP-NEXT:    buffer_gl1_inv
 ; GFX11-WGP-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-WGP-NEXT:    flat_store_b32 v[0:1], v2
-; GFX11-WGP-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-WGP-NEXT:    s_endpgm
 ;
 ; GFX11-CU-LABEL: flat_system_one_as_monotonic_acquire_ret_cmpxchg:
 ; GFX11-CU:       ; %bb.0: ; %entry
 ; GFX11-CU-NEXT:    s_load_b128 s[0:3], s[0:1], 0x0
 ; GFX11-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v0, s0 :: v_dual_mov_b32 v1, s1
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v2, s2 :: v_dual_mov_b32 v3, s3
 ; GFX11-CU-NEXT:    flat_atomic_cmpswap_b32 v2, v[0:1], v[2:3] offset:16 glc
 ; GFX11-CU-NEXT:    s_waitcnt vmcnt(0)
 ; GFX11-CU-NEXT:    buffer_gl0_inv
 ; GFX11-CU-NEXT:    buffer_gl1_inv
 ; GFX11-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-CU-NEXT:    flat_store_b32 v[0:1], v2
-; GFX11-CU-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-CU-NEXT:    s_endpgm
     ptr %out, i32 %in, i32 %old) {
 entry:
   %gep = getelementptr i32, ptr %out, i32 4
   %val = cmpxchg volatile ptr %gep, i32 %old, i32 %in syncscope("one-as") monotonic acquire
   %val0 = extractvalue { i32, i1 } %val, 0
   store i32 %val0, ptr %out, align 4
   ret void

 ; GFX11-WGP-NEXT:    v_dual_mov_b32 v0, s0 :: v_dual_mov_b32 v1, s1
 ; GFX11-WGP-NEXT:    v_dual_mov_b32 v2, s2 :: v_dual_mov_b32 v3, s3
 ; GFX11-WGP-NEXT:    flat_atomic_cmpswap_b32 v2, v[0:1], v[2:3] offset:16 glc
 ; GFX11-WGP-NEXT:    s_waitcnt vmcnt(0)
 ; GFX11-WGP-NEXT:    buffer_gl0_inv
 ; GFX11-WGP-NEXT:    buffer_gl1_inv
 ; GFX11-WGP-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-WGP-NEXT:    flat_store_b32 v[0:1], v2
-; GFX11-WGP-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-WGP-NEXT:    s_endpgm
 ;
 ; GFX11-CU-LABEL: flat_system_one_as_acquire_acquire_ret_cmpxchg:
 ; GFX11-CU:       ; %bb.0: ; %entry
 ; GFX11-CU-NEXT:    s_load_b128 s[0:3], s[0:1], 0x0
 ; GFX11-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v0, s0 :: v_dual_mov_b32 v1, s1
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v2, s2 :: v_dual_mov_b32 v3, s3
 ; GFX11-CU-NEXT:    flat_atomic_cmpswap_b32 v2, v[0:1], v[2:3] offset:16 glc
 ; GFX11-CU-NEXT:    s_waitcnt vmcnt(0)
 ; GFX11-CU-NEXT:    buffer_gl0_inv
 ; GFX11-CU-NEXT:    buffer_gl1_inv
 ; GFX11-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-CU-NEXT:    flat_store_b32 v[0:1], v2
-; GFX11-CU-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-CU-NEXT:    s_endpgm
     ptr %out, i32 %in, i32 %old) {
 entry:
   %gep = getelementptr i32, ptr %out, i32 4
   %val = cmpxchg volatile ptr %gep, i32 %old, i32 %in syncscope("one-as") acquire acquire
   %val0 = extractvalue { i32, i1 } %val, 0
   store i32 %val0, ptr %out, align 4
   ret void

 ; GFX11-WGP-NEXT:    s_waitcnt vmcnt(0)
 ; GFX11-WGP-NEXT:    s_waitcnt_vscnt null, 0x0
 ; GFX11-WGP-NEXT:    flat_atomic_cmpswap_b32 v2, v[0:1], v[2:3] offset:16 glc
 ; GFX11-WGP-NEXT:    s_waitcnt vmcnt(0)
 ; GFX11-WGP-NEXT:    buffer_gl0_inv
 ; GFX11-WGP-NEXT:    buffer_gl1_inv
 ; GFX11-WGP-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-WGP-NEXT:    flat_store_b32 v[0:1], v2
-; GFX11-WGP-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-WGP-NEXT:    s_endpgm
 ;
 ; GFX11-CU-LABEL: flat_system_one_as_release_acquire_ret_cmpxchg:
 ; GFX11-CU:       ; %bb.0: ; %entry
 ; GFX11-CU-NEXT:    s_load_b128 s[0:3], s[0:1], 0x0
 ; GFX11-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v0, s0 :: v_dual_mov_b32 v1, s1
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v2, s2 :: v_dual_mov_b32 v3, s3
 ; GFX11-CU-NEXT:    s_waitcnt vmcnt(0)
 ; GFX11-CU-NEXT:    s_waitcnt_vscnt null, 0x0
 ; GFX11-CU-NEXT:    flat_atomic_cmpswap_b32 v2, v[0:1], v[2:3] offset:16 glc
 ; GFX11-CU-NEXT:    s_waitcnt vmcnt(0)
 ; GFX11-CU-NEXT:    buffer_gl0_inv
 ; GFX11-CU-NEXT:    buffer_gl1_inv
 ; GFX11-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-CU-NEXT:    flat_store_b32 v[0:1], v2
-; GFX11-CU-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-CU-NEXT:    s_endpgm
     ptr %out, i32 %in, i32 %old) {
 entry:
   %gep = getelementptr i32, ptr %out, i32 4
   %val = cmpxchg volatile ptr %gep, i32 %old, i32 %in syncscope("one-as") release acquire
   %val0 = extractvalue { i32, i1 } %val, 0
   store i32 %val0, ptr %out, align 4
   ret void

 ; GFX11-WGP-NEXT:    s_waitcnt vmcnt(0)
 ; GFX11-WGP-NEXT:    s_waitcnt_vscnt null, 0x0
 ; GFX11-WGP-NEXT:    flat_atomic_cmpswap_b32 v2, v[0:1], v[2:3] offset:16 glc
 ; GFX11-WGP-NEXT:    s_waitcnt vmcnt(0)
 ; GFX11-WGP-NEXT:    buffer_gl0_inv
 ; GFX11-WGP-NEXT:    buffer_gl1_inv
 ; GFX11-WGP-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-WGP-NEXT:    flat_store_b32 v[0:1], v2
-; GFX11-WGP-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-WGP-NEXT:    s_endpgm
 ;
 ; GFX11-CU-LABEL: flat_system_one_as_acq_rel_acquire_ret_cmpxchg:
 ; GFX11-CU:       ; %bb.0: ; %entry
 ; GFX11-CU-NEXT:    s_load_b128 s[0:3], s[0:1], 0x0
 ; GFX11-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v0, s0 :: v_dual_mov_b32 v1, s1
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v2, s2 :: v_dual_mov_b32 v3, s3
 ; GFX11-CU-NEXT:    s_waitcnt vmcnt(0)
 ; GFX11-CU-NEXT:    s_waitcnt_vscnt null, 0x0
 ; GFX11-CU-NEXT:    flat_atomic_cmpswap_b32 v2, v[0:1], v[2:3] offset:16 glc
 ; GFX11-CU-NEXT:    s_waitcnt vmcnt(0)
 ; GFX11-CU-NEXT:    buffer_gl0_inv
 ; GFX11-CU-NEXT:    buffer_gl1_inv
 ; GFX11-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-CU-NEXT:    flat_store_b32 v[0:1], v2
-; GFX11-CU-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-CU-NEXT:    s_endpgm
     ptr %out, i32 %in, i32 %old) {
 entry:
   %gep = getelementptr i32, ptr %out, i32 4
   %val = cmpxchg volatile ptr %gep, i32 %old, i32 %in syncscope("one-as") acq_rel acquire
   %val0 = extractvalue { i32, i1 } %val, 0
   store i32 %val0, ptr %out, align 4
   ret void

 ; GFX11-WGP-NEXT:    s_waitcnt vmcnt(0)
 ; GFX11-WGP-NEXT:    s_waitcnt_vscnt null, 0x0
 ; GFX11-WGP-NEXT:    flat_atomic_cmpswap_b32 v2, v[0:1], v[2:3] offset:16 glc
 ; GFX11-WGP-NEXT:    s_waitcnt vmcnt(0)
 ; GFX11-WGP-NEXT:    buffer_gl0_inv
 ; GFX11-WGP-NEXT:    buffer_gl1_inv
 ; GFX11-WGP-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-WGP-NEXT:    flat_store_b32 v[0:1], v2
-; GFX11-WGP-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-WGP-NEXT:    s_endpgm
 ;
 ; GFX11-CU-LABEL: flat_system_one_as_seq_cst_acquire_ret_cmpxchg:
 ; GFX11-CU:       ; %bb.0: ; %entry
 ; GFX11-CU-NEXT:    s_load_b128 s[0:3], s[0:1], 0x0
 ; GFX11-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v0, s0 :: v_dual_mov_b32 v1, s1
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v2, s2 :: v_dual_mov_b32 v3, s3
 ; GFX11-CU-NEXT:    s_waitcnt vmcnt(0)
 ; GFX11-CU-NEXT:    s_waitcnt_vscnt null, 0x0
 ; GFX11-CU-NEXT:    flat_atomic_cmpswap_b32 v2, v[0:1], v[2:3] offset:16 glc
 ; GFX11-CU-NEXT:    s_waitcnt vmcnt(0)
 ; GFX11-CU-NEXT:    buffer_gl0_inv
 ; GFX11-CU-NEXT:    buffer_gl1_inv
 ; GFX11-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-CU-NEXT:    flat_store_b32 v[0:1], v2
-; GFX11-CU-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-CU-NEXT:    s_endpgm
     ptr %out, i32 %in, i32 %old) {
 entry:
   %gep = getelementptr i32, ptr %out, i32 4
   %val = cmpxchg volatile ptr %gep, i32 %old, i32 %in syncscope("one-as") seq_cst acquire
   %val0 = extractvalue { i32, i1 } %val, 0
   store i32 %val0, ptr %out, align 4
   ret void

 ; GFX11-WGP-NEXT:    s_waitcnt vmcnt(0)
 ; GFX11-WGP-NEXT:    s_waitcnt_vscnt null, 0x0
 ; GFX11-WGP-NEXT:    flat_atomic_cmpswap_b32 v2, v[0:1], v[2:3] offset:16 glc
 ; GFX11-WGP-NEXT:    s_waitcnt vmcnt(0)
 ; GFX11-WGP-NEXT:    buffer_gl0_inv
 ; GFX11-WGP-NEXT:    buffer_gl1_inv
 ; GFX11-WGP-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-WGP-NEXT:    flat_store_b32 v[0:1], v2
-; GFX11-WGP-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-WGP-NEXT:    s_endpgm
 ;
 ; GFX11-CU-LABEL: flat_system_one_as_monotonic_seq_cst_ret_cmpxchg:
 ; GFX11-CU:       ; %bb.0: ; %entry
 ; GFX11-CU-NEXT:    s_load_b128 s[0:3], s[0:1], 0x0
 ; GFX11-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v0, s0 :: v_dual_mov_b32 v1, s1
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v2, s2 :: v_dual_mov_b32 v3, s3
 ; GFX11-CU-NEXT:    s_waitcnt vmcnt(0)
 ; GFX11-CU-NEXT:    s_waitcnt_vscnt null, 0x0
 ; GFX11-CU-NEXT:    flat_atomic_cmpswap_b32 v2, v[0:1], v[2:3] offset:16 glc
 ; GFX11-CU-NEXT:    s_waitcnt vmcnt(0)
 ; GFX11-CU-NEXT:    buffer_gl0_inv
 ; GFX11-CU-NEXT:    buffer_gl1_inv
 ; GFX11-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-CU-NEXT:    flat_store_b32 v[0:1], v2
-; GFX11-CU-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-CU-NEXT:    s_endpgm
     ptr %out, i32 %in, i32 %old) {
 entry:
   %gep = getelementptr i32, ptr %out, i32 4
   %val = cmpxchg volatile ptr %gep, i32 %old, i32 %in syncscope("one-as") monotonic seq_cst
   %val0 = extractvalue { i32, i1 } %val, 0
   store i32 %val0, ptr %out, align 4
   ret void

 ; GFX11-WGP-NEXT:    s_waitcnt vmcnt(0)
 ; GFX11-WGP-NEXT:    s_waitcnt_vscnt null, 0x0
 ; GFX11-WGP-NEXT:    flat_atomic_cmpswap_b32 v2, v[0:1], v[2:3] offset:16 glc
 ; GFX11-WGP-NEXT:    s_waitcnt vmcnt(0)
 ; GFX11-WGP-NEXT:    buffer_gl0_inv
 ; GFX11-WGP-NEXT:    buffer_gl1_inv
 ; GFX11-WGP-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-WGP-NEXT:    flat_store_b32 v[0:1], v2
-; GFX11-WGP-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-WGP-NEXT:    s_endpgm
 ;
 ; GFX11-CU-LABEL: flat_system_one_as_acquire_seq_cst_ret_cmpxchg:
 ; GFX11-CU:       ; %bb.0: ; %entry
 ; GFX11-CU-NEXT:    s_load_b128 s[0:3], s[0:1], 0x0
 ; GFX11-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v0, s0 :: v_dual_mov_b32 v1, s1
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v2, s2 :: v_dual_mov_b32 v3, s3
 ; GFX11-CU-NEXT:    s_waitcnt vmcnt(0)
 ; GFX11-CU-NEXT:    s_waitcnt_vscnt null, 0x0
 ; GFX11-CU-NEXT:    flat_atomic_cmpswap_b32 v2, v[0:1], v[2:3] offset:16 glc
 ; GFX11-CU-NEXT:    s_waitcnt vmcnt(0)
 ; GFX11-CU-NEXT:    buffer_gl0_inv
 ; GFX11-CU-NEXT:    buffer_gl1_inv
 ; GFX11-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-CU-NEXT:    flat_store_b32 v[0:1], v2
-; GFX11-CU-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-CU-NEXT:    s_endpgm
     ptr %out, i32 %in, i32 %old) {
 entry:
   %gep = getelementptr i32, ptr %out, i32 4
   %val = cmpxchg volatile ptr %gep, i32 %old, i32 %in syncscope("one-as") acquire seq_cst
   %val0 = extractvalue { i32, i1 } %val, 0
   store i32 %val0, ptr %out, align 4
   ret void

 ; GFX11-WGP-NEXT:    s_waitcnt vmcnt(0)
 ; GFX11-WGP-NEXT:    s_waitcnt_vscnt null, 0x0
 ; GFX11-WGP-NEXT:    flat_atomic_cmpswap_b32 v2, v[0:1], v[2:3] offset:16 glc
 ; GFX11-WGP-NEXT:    s_waitcnt vmcnt(0)
 ; GFX11-WGP-NEXT:    buffer_gl0_inv
 ; GFX11-WGP-NEXT:    buffer_gl1_inv
 ; GFX11-WGP-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-WGP-NEXT:    flat_store_b32 v[0:1], v2
-; GFX11-WGP-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-WGP-NEXT:    s_endpgm
 ;
 ; GFX11-CU-LABEL: flat_system_one_as_release_seq_cst_ret_cmpxchg:
 ; GFX11-CU:       ; %bb.0: ; %entry
 ; GFX11-CU-NEXT:    s_load_b128 s[0:3], s[0:1], 0x0
 ; GFX11-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v0, s0 :: v_dual_mov_b32 v1, s1
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v2, s2 :: v_dual_mov_b32 v3, s3
 ; GFX11-CU-NEXT:    s_waitcnt vmcnt(0)
 ; GFX11-CU-NEXT:    s_waitcnt_vscnt null, 0x0
 ; GFX11-CU-NEXT:    flat_atomic_cmpswap_b32 v2, v[0:1], v[2:3] offset:16 glc
 ; GFX11-CU-NEXT:    s_waitcnt vmcnt(0)
 ; GFX11-CU-NEXT:    buffer_gl0_inv
 ; GFX11-CU-NEXT:    buffer_gl1_inv
 ; GFX11-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-CU-NEXT:    flat_store_b32 v[0:1], v2
-; GFX11-CU-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-CU-NEXT:    s_endpgm
     ptr %out, i32 %in, i32 %old) {
 entry:
   %gep = getelementptr i32, ptr %out, i32 4
   %val = cmpxchg volatile ptr %gep, i32 %old, i32 %in syncscope("one-as") release seq_cst
   %val0 = extractvalue { i32, i1 } %val, 0
   store i32 %val0, ptr %out, align 4
   ret void

 ; GFX11-WGP-NEXT:    s_waitcnt vmcnt(0)
 ; GFX11-WGP-NEXT:    s_waitcnt_vscnt null, 0x0
 ; GFX11-WGP-NEXT:    flat_atomic_cmpswap_b32 v2, v[0:1], v[2:3] offset:16 glc
 ; GFX11-WGP-NEXT:    s_waitcnt vmcnt(0)
 ; GFX11-WGP-NEXT:    buffer_gl0_inv
 ; GFX11-WGP-NEXT:    buffer_gl1_inv
 ; GFX11-WGP-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-WGP-NEXT:    flat_store_b32 v[0:1], v2
-; GFX11-WGP-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-WGP-NEXT:    s_endpgm
 ;
 ; GFX11-CU-LABEL: flat_system_one_as_acq_rel_seq_cst_ret_cmpxchg:
 ; GFX11-CU:       ; %bb.0: ; %entry
 ; GFX11-CU-NEXT:    s_load_b128 s[0:3], s[0:1], 0x0
 ; GFX11-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v0, s0 :: v_dual_mov_b32 v1, s1
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v2, s2 :: v_dual_mov_b32 v3, s3
 ; GFX11-CU-NEXT:    s_waitcnt vmcnt(0)
 ; GFX11-CU-NEXT:    s_waitcnt_vscnt null, 0x0
 ; GFX11-CU-NEXT:    flat_atomic_cmpswap_b32 v2, v[0:1], v[2:3] offset:16 glc
 ; GFX11-CU-NEXT:    s_waitcnt vmcnt(0)
 ; GFX11-CU-NEXT:    buffer_gl0_inv
 ; GFX11-CU-NEXT:    buffer_gl1_inv
 ; GFX11-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-CU-NEXT:    flat_store_b32 v[0:1], v2
-; GFX11-CU-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-CU-NEXT:    s_endpgm
     ptr %out, i32 %in, i32 %old) {
 entry:
   %gep = getelementptr i32, ptr %out, i32 4
   %val = cmpxchg volatile ptr %gep, i32 %old, i32 %in syncscope("one-as") acq_rel seq_cst
   %val0 = extractvalue { i32, i1 } %val, 0
   store i32 %val0, ptr %out, align 4
   ret void

 ; GFX11-WGP-NEXT:    s_waitcnt vmcnt(0)
 ; GFX11-WGP-NEXT:    s_waitcnt_vscnt null, 0x0
 ; GFX11-WGP-NEXT:    flat_atomic_cmpswap_b32 v2, v[0:1], v[2:3] offset:16 glc
 ; GFX11-WGP-NEXT:    s_waitcnt vmcnt(0)
 ; GFX11-WGP-NEXT:    buffer_gl0_inv
 ; GFX11-WGP-NEXT:    buffer_gl1_inv
 ; GFX11-WGP-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-WGP-NEXT:    flat_store_b32 v[0:1], v2
-; GFX11-WGP-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-WGP-NEXT:    s_endpgm
 ;
 ; GFX11-CU-LABEL: flat_system_one_as_seq_cst_seq_cst_ret_cmpxchg:
 ; GFX11-CU:       ; %bb.0: ; %entry
 ; GFX11-CU-NEXT:    s_load_b128 s[0:3], s[0:1], 0x0
 ; GFX11-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v0, s0 :: v_dual_mov_b32 v1, s1
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v2, s2 :: v_dual_mov_b32 v3, s3
 ; GFX11-CU-NEXT:    s_waitcnt vmcnt(0)
 ; GFX11-CU-NEXT:    s_waitcnt_vscnt null, 0x0
 ; GFX11-CU-NEXT:    flat_atomic_cmpswap_b32 v2, v[0:1], v[2:3] offset:16 glc
 ; GFX11-CU-NEXT:    s_waitcnt vmcnt(0)
 ; GFX11-CU-NEXT:    buffer_gl0_inv
 ; GFX11-CU-NEXT:    buffer_gl1_inv
 ; GFX11-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-CU-NEXT:    flat_store_b32 v[0:1], v2
-; GFX11-CU-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-CU-NEXT:    s_endpgm
     ptr %out, i32 %in, i32 %old) {
 entry:
   %gep = getelementptr i32, ptr %out, i32 4
   %val = cmpxchg volatile ptr %gep, i32 %old, i32 %in syncscope("one-as") seq_cst seq_cst
   %val0 = extractvalue { i32, i1 } %val, 0
   store i32 %val0, ptr %out, align 4
   ret void
 }
 

llvm/test/CodeGen/AMDGPU/memory-legalizer-flat-volatile.ll

 ; GFX11-WGP-NEXT:    s_load_b128 s[0:3], s[0:1], 0x0
 ; GFX11-WGP-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-WGP-NEXT:    v_dual_mov_b32 v0, s0 :: v_dual_mov_b32 v1, s1
 ; GFX11-WGP-NEXT:    flat_load_b32 v2, v[0:1] glc dlc
 ; GFX11-WGP-NEXT:    s_waitcnt vmcnt(0)
 ; GFX11-WGP-NEXT:    v_dual_mov_b32 v0, s2 :: v_dual_mov_b32 v1, s3
 ; GFX11-WGP-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-WGP-NEXT:    flat_store_b32 v[0:1], v2
-; GFX11-WGP-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-WGP-NEXT:    s_endpgm
 ;
 ; GFX11-CU-LABEL: flat_nontemporal_load_0:
 ; GFX11-CU:       ; %bb.0: ; %entry
 ; GFX11-CU-NEXT:    s_load_b128 s[0:3], s[0:1], 0x0
 ; GFX11-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v0, s0 :: v_dual_mov_b32 v1, s1
 ; GFX11-CU-NEXT:    flat_load_b32 v2, v[0:1] glc dlc
 ; GFX11-CU-NEXT:    s_waitcnt vmcnt(0)
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v0, s2 :: v_dual_mov_b32 v1, s3
 ; GFX11-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-CU-NEXT:    flat_store_b32 v[0:1], v2
-; GFX11-CU-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-CU-NEXT:    s_endpgm
     ptr %in, ptr %out) {
 entry:
   %val = load volatile i32, ptr %in, align 4
   store i32 %val, ptr %out
   ret void
 }
 

 ; GFX11-WGP-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
 ; GFX11-WGP-NEXT:    v_add_co_u32 v0, s0, s0, v0
 ; GFX11-WGP-NEXT:    v_add_co_ci_u32_e64 v1, null, s1, 0, s0
 ; GFX11-WGP-NEXT:    flat_load_b32 v2, v[0:1] glc dlc
 ; GFX11-WGP-NEXT:    s_waitcnt vmcnt(0)
 ; GFX11-WGP-NEXT:    v_dual_mov_b32 v1, s3 :: v_dual_mov_b32 v0, s2
 ; GFX11-WGP-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-WGP-NEXT:    flat_store_b32 v[0:1], v2
-; GFX11-WGP-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-WGP-NEXT:    s_endpgm
 ;
 ; GFX11-CU-LABEL: flat_nontemporal_load_1:
 ; GFX11-CU:       ; %bb.0: ; %entry
 ; GFX11-CU-NEXT:    s_load_b128 s[0:3], s[0:1], 0x0
 ; GFX11-CU-NEXT:    v_lshlrev_b32_e32 v0, 2, v0
 ; GFX11-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-CU-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
 ; GFX11-CU-NEXT:    v_add_co_u32 v0, s0, s0, v0
 ; GFX11-CU-NEXT:    v_add_co_ci_u32_e64 v1, null, s1, 0, s0
 ; GFX11-CU-NEXT:    flat_load_b32 v2, v[0:1] glc dlc
 ; GFX11-CU-NEXT:    s_waitcnt vmcnt(0)
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v1, s3 :: v_dual_mov_b32 v0, s2
 ; GFX11-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-CU-NEXT:    flat_store_b32 v[0:1], v2
-; GFX11-CU-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-CU-NEXT:    s_endpgm
     ptr %in, ptr %out) {
 entry:
   %tid = call i32 @llvm.amdgcn.workitem.id.x()
   %val.gep = getelementptr inbounds i32, ptr %in, i32 %tid
   %val = load volatile i32, ptr %val.gep, align 4
   store i32 %val, ptr %out
   ret void

 ; GFX11-WGP-NEXT:    s_load_b128 s[0:3], s[0:1], 0x0
 ; GFX11-WGP-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-WGP-NEXT:    v_dual_mov_b32 v0, s0 :: v_dual_mov_b32 v1, s1
 ; GFX11-WGP-NEXT:    flat_load_b32 v2, v[0:1]
 ; GFX11-WGP-NEXT:    v_dual_mov_b32 v0, s2 :: v_dual_mov_b32 v1, s3
 ; GFX11-WGP-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX11-WGP-NEXT:    flat_store_b32 v[0:1], v2 dlc
 ; GFX11-WGP-NEXT:    s_waitcnt_vscnt null, 0x0
-; GFX11-WGP-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-WGP-NEXT:    s_endpgm
 ;
 ; GFX11-CU-LABEL: flat_nontemporal_store_0:
 ; GFX11-CU:       ; %bb.0: ; %entry
 ; GFX11-CU-NEXT:    s_load_b128 s[0:3], s[0:1], 0x0
 ; GFX11-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v0, s0 :: v_dual_mov_b32 v1, s1
 ; GFX11-CU-NEXT:    flat_load_b32 v2, v[0:1]
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v0, s2 :: v_dual_mov_b32 v1, s3
 ; GFX11-CU-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX11-CU-NEXT:    flat_store_b32 v[0:1], v2 dlc
 ; GFX11-CU-NEXT:    s_waitcnt_vscnt null, 0x0
-; GFX11-CU-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-CU-NEXT:    s_endpgm
     ptr %in, ptr %out) {
 entry:
   %val = load i32, ptr %in, align 4
   store volatile i32 %val, ptr %out
   ret void
 }
 

 ; GFX11-WGP-NEXT:    v_mov_b32_e32 v2, s1
 ; GFX11-WGP-NEXT:    s_delay_alu instid0(VALU_DEP_2)
 ; GFX11-WGP-NEXT:    v_add_co_u32 v0, s0, s2, v0
 ; GFX11-WGP-NEXT:    flat_load_b32 v2, v[1:2]
 ; GFX11-WGP-NEXT:    v_add_co_ci_u32_e64 v1, null, s3, 0, s0
 ; GFX11-WGP-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX11-WGP-NEXT:    flat_store_b32 v[0:1], v2 dlc
 ; GFX11-WGP-NEXT:    s_waitcnt_vscnt null, 0x0
-; GFX11-WGP-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-WGP-NEXT:    s_endpgm
 ;
 ; GFX11-CU-LABEL: flat_nontemporal_store_1:
 ; GFX11-CU:       ; %bb.0: ; %entry
 ; GFX11-CU-NEXT:    s_load_b128 s[0:3], s[0:1], 0x0
 ; GFX11-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v1, s0 :: v_dual_lshlrev_b32 v0, 2, v0
 ; GFX11-CU-NEXT:    v_mov_b32_e32 v2, s1
 ; GFX11-CU-NEXT:    s_delay_alu instid0(VALU_DEP_2)
 ; GFX11-CU-NEXT:    v_add_co_u32 v0, s0, s2, v0
 ; GFX11-CU-NEXT:    flat_load_b32 v2, v[1:2]
 ; GFX11-CU-NEXT:    v_add_co_ci_u32_e64 v1, null, s3, 0, s0
 ; GFX11-CU-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX11-CU-NEXT:    flat_store_b32 v[0:1], v2 dlc
 ; GFX11-CU-NEXT:    s_waitcnt_vscnt null, 0x0
-; GFX11-CU-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-CU-NEXT:    s_endpgm
     ptr %in, ptr %out) {
 entry:
   %tid = call i32 @llvm.amdgcn.workitem.id.x()
   %val = load i32, ptr %in, align 4
   %out.gep = getelementptr inbounds i32, ptr %out, i32 %tid
   store volatile i32 %val, ptr %out.gep
   ret void

 ; GFX11-WGP-NEXT:    s_load_b128 s[0:3], s[0:1], 0x0
 ; GFX11-WGP-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-WGP-NEXT:    v_dual_mov_b32 v0, s0 :: v_dual_mov_b32 v1, s1
 ; GFX11-WGP-NEXT:    flat_load_b32 v2, v[0:1] glc
 ; GFX11-WGP-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX11-WGP-NEXT:    buffer_gl0_inv
 ; GFX11-WGP-NEXT:    v_dual_mov_b32 v0, s2 :: v_dual_mov_b32 v1, s3
 ; GFX11-WGP-NEXT:    flat_store_b32 v[0:1], v2
-; GFX11-WGP-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-WGP-NEXT:    s_endpgm
 ;
 ; GFX11-CU-LABEL: flat_volatile_workgroup_acquire_load:
 ; GFX11-CU:       ; %bb.0: ; %entry
 ; GFX11-CU-NEXT:    s_load_b128 s[0:3], s[0:1], 0x0
 ; GFX11-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v0, s0 :: v_dual_mov_b32 v1, s1
 ; GFX11-CU-NEXT:    flat_load_b32 v2, v[0:1]
 ; GFX11-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v0, s2 :: v_dual_mov_b32 v1, s3
 ; GFX11-CU-NEXT:    s_waitcnt vmcnt(0)
 ; GFX11-CU-NEXT:    flat_store_b32 v[0:1], v2
-; GFX11-CU-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-CU-NEXT:    s_endpgm
     ptr %in, ptr %out) {
 entry:
   %val = load atomic volatile i32, ptr %in syncscope("workgroup") acquire, align 4
   store i32 %val, ptr %out
   ret void
 }
 

 ; GFX11-WGP-NEXT:    s_load_b64 s[2:3], s[0:1], 0x8
 ; GFX11-WGP-NEXT:    s_load_b32 s0, s[0:1], 0x0
 ; GFX11-WGP-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-WGP-NEXT:    v_dual_mov_b32 v0, s2 :: v_dual_mov_b32 v1, s3
 ; GFX11-WGP-NEXT:    v_mov_b32_e32 v2, s0
 ; GFX11-WGP-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX11-WGP-NEXT:    s_waitcnt_vscnt null, 0x0
 ; GFX11-WGP-NEXT:    flat_store_b32 v[0:1], v2
-; GFX11-WGP-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-WGP-NEXT:    s_endpgm
 ;
 ; GFX11-CU-LABEL: flat_volatile_workgroup_release_store:
 ; GFX11-CU:       ; %bb.0: ; %entry
 ; GFX11-CU-NEXT:    s_clause 0x1
 ; GFX11-CU-NEXT:    s_load_b64 s[2:3], s[0:1], 0x8
 ; GFX11-CU-NEXT:    s_load_b32 s0, s[0:1], 0x0
 ; GFX11-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v0, s2 :: v_dual_mov_b32 v1, s3
 ; GFX11-CU-NEXT:    v_mov_b32_e32 v2, s0
 ; GFX11-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-CU-NEXT:    flat_store_b32 v[0:1], v2
-; GFX11-CU-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-CU-NEXT:    s_endpgm
    i32 %in, ptr %out) {
 entry:
   store atomic volatile i32 %in, ptr %out syncscope("workgroup") release, align 4
   ret void
 }
 
 declare i32 @llvm.amdgcn.workitem.id.x()

llvm/test/CodeGen/AMDGPU/memory-legalizer-flat-wavefront.ll

 ; GFX11-WGP:       ; %bb.0: ; %entry
 ; GFX11-WGP-NEXT:    s_load_b128 s[0:3], s[0:1], 0x0
 ; GFX11-WGP-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-WGP-NEXT:    v_dual_mov_b32 v0, s0 :: v_dual_mov_b32 v1, s1
 ; GFX11-WGP-NEXT:    flat_load_b32 v2, v[0:1]
 ; GFX11-WGP-NEXT:    v_dual_mov_b32 v0, s2 :: v_dual_mov_b32 v1, s3
 ; GFX11-WGP-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX11-WGP-NEXT:    flat_store_b32 v[0:1], v2
-; GFX11-WGP-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-WGP-NEXT:    s_endpgm
 ;
 ; GFX11-CU-LABEL: flat_wavefront_unordered_load:
 ; GFX11-CU:       ; %bb.0: ; %entry
 ; GFX11-CU-NEXT:    s_load_b128 s[0:3], s[0:1], 0x0
 ; GFX11-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v0, s0 :: v_dual_mov_b32 v1, s1
 ; GFX11-CU-NEXT:    flat_load_b32 v2, v[0:1]
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v0, s2 :: v_dual_mov_b32 v1, s3
 ; GFX11-CU-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX11-CU-NEXT:    flat_store_b32 v[0:1], v2
-; GFX11-CU-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-CU-NEXT:    s_endpgm
     ptr %in, ptr %out) {
 entry:
   %val = load atomic i32, ptr %in syncscope("wavefront") unordered, align 4
   store i32 %val, ptr %out
   ret void
 }
 

 ; GFX11-WGP:       ; %bb.0: ; %entry
 ; GFX11-WGP-NEXT:    s_load_b128 s[0:3], s[0:1], 0x0
 ; GFX11-WGP-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-WGP-NEXT:    v_dual_mov_b32 v0, s0 :: v_dual_mov_b32 v1, s1
 ; GFX11-WGP-NEXT:    flat_load_b32 v2, v[0:1]
 ; GFX11-WGP-NEXT:    v_dual_mov_b32 v0, s2 :: v_dual_mov_b32 v1, s3
 ; GFX11-WGP-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX11-WGP-NEXT:    flat_store_b32 v[0:1], v2
-; GFX11-WGP-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-WGP-NEXT:    s_endpgm
 ;
 ; GFX11-CU-LABEL: flat_wavefront_monotonic_load:
 ; GFX11-CU:       ; %bb.0: ; %entry
 ; GFX11-CU-NEXT:    s_load_b128 s[0:3], s[0:1], 0x0
 ; GFX11-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v0, s0 :: v_dual_mov_b32 v1, s1
 ; GFX11-CU-NEXT:    flat_load_b32 v2, v[0:1]
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v0, s2 :: v_dual_mov_b32 v1, s3
 ; GFX11-CU-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX11-CU-NEXT:    flat_store_b32 v[0:1], v2
-; GFX11-CU-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-CU-NEXT:    s_endpgm
     ptr %in, ptr %out) {
 entry:
   %val = load atomic i32, ptr %in syncscope("wavefront") monotonic, align 4
   store i32 %val, ptr %out
   ret void
 }
 

 ; GFX11-WGP:       ; %bb.0: ; %entry
 ; GFX11-WGP-NEXT:    s_load_b128 s[0:3], s[0:1], 0x0
 ; GFX11-WGP-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-WGP-NEXT:    v_dual_mov_b32 v0, s0 :: v_dual_mov_b32 v1, s1
 ; GFX11-WGP-NEXT:    flat_load_b32 v2, v[0:1]
 ; GFX11-WGP-NEXT:    v_dual_mov_b32 v0, s2 :: v_dual_mov_b32 v1, s3
 ; GFX11-WGP-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX11-WGP-NEXT:    flat_store_b32 v[0:1], v2
-; GFX11-WGP-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-WGP-NEXT:    s_endpgm
 ;
 ; GFX11-CU-LABEL: flat_wavefront_acquire_load:
 ; GFX11-CU:       ; %bb.0: ; %entry
 ; GFX11-CU-NEXT:    s_load_b128 s[0:3], s[0:1], 0x0
 ; GFX11-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v0, s0 :: v_dual_mov_b32 v1, s1
 ; GFX11-CU-NEXT:    flat_load_b32 v2, v[0:1]
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v0, s2 :: v_dual_mov_b32 v1, s3
 ; GFX11-CU-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX11-CU-NEXT:    flat_store_b32 v[0:1], v2
-; GFX11-CU-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-CU-NEXT:    s_endpgm
     ptr %in, ptr %out) {
 entry:
   %val = load atomic i32, ptr %in syncscope("wavefront") acquire, align 4
   store i32 %val, ptr %out
   ret void
 }
 

 ; GFX11-WGP:       ; %bb.0: ; %entry
 ; GFX11-WGP-NEXT:    s_load_b128 s[0:3], s[0:1], 0x0
 ; GFX11-WGP-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-WGP-NEXT:    v_dual_mov_b32 v0, s0 :: v_dual_mov_b32 v1, s1
 ; GFX11-WGP-NEXT:    flat_load_b32 v2, v[0:1]
 ; GFX11-WGP-NEXT:    v_dual_mov_b32 v0, s2 :: v_dual_mov_b32 v1, s3
 ; GFX11-WGP-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX11-WGP-NEXT:    flat_store_b32 v[0:1], v2
-; GFX11-WGP-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-WGP-NEXT:    s_endpgm
 ;
 ; GFX11-CU-LABEL: flat_wavefront_seq_cst_load:
 ; GFX11-CU:       ; %bb.0: ; %entry
 ; GFX11-CU-NEXT:    s_load_b128 s[0:3], s[0:1], 0x0
 ; GFX11-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v0, s0 :: v_dual_mov_b32 v1, s1
 ; GFX11-CU-NEXT:    flat_load_b32 v2, v[0:1]
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v0, s2 :: v_dual_mov_b32 v1, s3
 ; GFX11-CU-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX11-CU-NEXT:    flat_store_b32 v[0:1], v2
-; GFX11-CU-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-CU-NEXT:    s_endpgm
     ptr %in, ptr %out) {
 entry:
   %val = load atomic i32, ptr %in syncscope("wavefront") seq_cst, align 4
   store i32 %val, ptr %out
   ret void
 }
 

 ; GFX11-WGP:       ; %bb.0: ; %entry
 ; GFX11-WGP-NEXT:    s_clause 0x1
 ; GFX11-WGP-NEXT:    s_load_b64 s[2:3], s[0:1], 0x8
 ; GFX11-WGP-NEXT:    s_load_b32 s0, s[0:1], 0x0
 ; GFX11-WGP-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-WGP-NEXT:    v_dual_mov_b32 v0, s2 :: v_dual_mov_b32 v1, s3
 ; GFX11-WGP-NEXT:    v_mov_b32_e32 v2, s0
 ; GFX11-WGP-NEXT:    flat_store_b32 v[0:1], v2
-; GFX11-WGP-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-WGP-NEXT:    s_endpgm
 ;
 ; GFX11-CU-LABEL: flat_wavefront_unordered_store:
 ; GFX11-CU:       ; %bb.0: ; %entry
 ; GFX11-CU-NEXT:    s_clause 0x1
 ; GFX11-CU-NEXT:    s_load_b64 s[2:3], s[0:1], 0x8
 ; GFX11-CU-NEXT:    s_load_b32 s0, s[0:1], 0x0
 ; GFX11-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v0, s2 :: v_dual_mov_b32 v1, s3
 ; GFX11-CU-NEXT:    v_mov_b32_e32 v2, s0
 ; GFX11-CU-NEXT:    flat_store_b32 v[0:1], v2
-; GFX11-CU-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-CU-NEXT:    s_endpgm
     i32 %in, ptr %out) {
 entry:
   store atomic i32 %in, ptr %out syncscope("wavefront") unordered, align 4
   ret void
 }
 
 define amdgpu_kernel void @flat_wavefront_monotonic_store(

 ; GFX11-WGP:       ; %bb.0: ; %entry
 ; GFX11-WGP-NEXT:    s_clause 0x1
 ; GFX11-WGP-NEXT:    s_load_b64 s[2:3], s[0:1], 0x8
 ; GFX11-WGP-NEXT:    s_load_b32 s0, s[0:1], 0x0
 ; GFX11-WGP-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-WGP-NEXT:    v_dual_mov_b32 v0, s2 :: v_dual_mov_b32 v1, s3
 ; GFX11-WGP-NEXT:    v_mov_b32_e32 v2, s0
 ; GFX11-WGP-NEXT:    flat_store_b32 v[0:1], v2
-; GFX11-WGP-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-WGP-NEXT:    s_endpgm
 ;
 ; GFX11-CU-LABEL: flat_wavefront_monotonic_store:
 ; GFX11-CU:       ; %bb.0: ; %entry
 ; GFX11-CU-NEXT:    s_clause 0x1
 ; GFX11-CU-NEXT:    s_load_b64 s[2:3], s[0:1], 0x8
 ; GFX11-CU-NEXT:    s_load_b32 s0, s[0:1], 0x0
 ; GFX11-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v0, s2 :: v_dual_mov_b32 v1, s3
 ; GFX11-CU-NEXT:    v_mov_b32_e32 v2, s0
 ; GFX11-CU-NEXT:    flat_store_b32 v[0:1], v2
-; GFX11-CU-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-CU-NEXT:    s_endpgm
     i32 %in, ptr %out) {
 entry:
   store atomic i32 %in, ptr %out syncscope("wavefront") monotonic, align 4
   ret void
 }
 
 define amdgpu_kernel void @flat_wavefront_release_store(

 ; GFX11-WGP:       ; %bb.0: ; %entry
 ; GFX11-WGP-NEXT:    s_clause 0x1
 ; GFX11-WGP-NEXT:    s_load_b64 s[2:3], s[0:1], 0x8
 ; GFX11-WGP-NEXT:    s_load_b32 s0, s[0:1], 0x0
 ; GFX11-WGP-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-WGP-NEXT:    v_dual_mov_b32 v0, s2 :: v_dual_mov_b32 v1, s3
 ; GFX11-WGP-NEXT:    v_mov_b32_e32 v2, s0
 ; GFX11-WGP-NEXT:    flat_store_b32 v[0:1], v2
-; GFX11-WGP-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-WGP-NEXT:    s_endpgm
 ;
 ; GFX11-CU-LABEL: flat_wavefront_release_store:
 ; GFX11-CU:       ; %bb.0: ; %entry
 ; GFX11-CU-NEXT:    s_clause 0x1
 ; GFX11-CU-NEXT:    s_load_b64 s[2:3], s[0:1], 0x8
 ; GFX11-CU-NEXT:    s_load_b32 s0, s[0:1], 0x0
 ; GFX11-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v0, s2 :: v_dual_mov_b32 v1, s3
 ; GFX11-CU-NEXT:    v_mov_b32_e32 v2, s0
 ; GFX11-CU-NEXT:    flat_store_b32 v[0:1], v2
-; GFX11-CU-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-CU-NEXT:    s_endpgm
     i32 %in, ptr %out) {
 entry:
   store atomic i32 %in, ptr %out syncscope("wavefront") release, align 4
   ret void
 }
 
 define amdgpu_kernel void @flat_wavefront_seq_cst_store(

 ; GFX11-WGP:       ; %bb.0: ; %entry
 ; GFX11-WGP-NEXT:    s_clause 0x1
 ; GFX11-WGP-NEXT:    s_load_b64 s[2:3], s[0:1], 0x8
 ; GFX11-WGP-NEXT:    s_load_b32 s0, s[0:1], 0x0
 ; GFX11-WGP-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-WGP-NEXT:    v_dual_mov_b32 v0, s2 :: v_dual_mov_b32 v1, s3
 ; GFX11-WGP-NEXT:    v_mov_b32_e32 v2, s0
 ; GFX11-WGP-NEXT:    flat_store_b32 v[0:1], v2
-; GFX11-WGP-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-WGP-NEXT:    s_endpgm
 ;
 ; GFX11-CU-LABEL: flat_wavefront_seq_cst_store:
 ; GFX11-CU:       ; %bb.0: ; %entry
 ; GFX11-CU-NEXT:    s_clause 0x1
 ; GFX11-CU-NEXT:    s_load_b64 s[2:3], s[0:1], 0x8
 ; GFX11-CU-NEXT:    s_load_b32 s0, s[0:1], 0x0
 ; GFX11-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v0, s2 :: v_dual_mov_b32 v1, s3
 ; GFX11-CU-NEXT:    v_mov_b32_e32 v2, s0
 ; GFX11-CU-NEXT:    flat_store_b32 v[0:1], v2
-; GFX11-CU-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-CU-NEXT:    s_endpgm
     i32 %in, ptr %out) {
 entry:
   store atomic i32 %in, ptr %out syncscope("wavefront") seq_cst, align 4
   ret void
 }
 
 define amdgpu_kernel void @flat_wavefront_monotonic_atomicrmw(

 ; GFX11-WGP:       ; %bb.0: ; %entry
 ; GFX11-WGP-NEXT:    s_clause 0x1
 ; GFX11-WGP-NEXT:    s_load_b64 s[2:3], s[0:1], 0x0
 ; GFX11-WGP-NEXT:    s_load_b32 s0, s[0:1], 0x8
 ; GFX11-WGP-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-WGP-NEXT:    v_dual_mov_b32 v0, s2 :: v_dual_mov_b32 v1, s3
 ; GFX11-WGP-NEXT:    v_mov_b32_e32 v2, s0
 ; GFX11-WGP-NEXT:    flat_atomic_swap_b32 v[0:1], v2
-; GFX11-WGP-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-WGP-NEXT:    s_endpgm
 ;
 ; GFX11-CU-LABEL: flat_wavefront_monotonic_atomicrmw:
 ; GFX11-CU:       ; %bb.0: ; %entry
 ; GFX11-CU-NEXT:    s_clause 0x1
 ; GFX11-CU-NEXT:    s_load_b64 s[2:3], s[0:1], 0x0
 ; GFX11-CU-NEXT:    s_load_b32 s0, s[0:1], 0x8
 ; GFX11-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v0, s2 :: v_dual_mov_b32 v1, s3
 ; GFX11-CU-NEXT:    v_mov_b32_e32 v2, s0
 ; GFX11-CU-NEXT:    flat_atomic_swap_b32 v[0:1], v2
-; GFX11-CU-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-CU-NEXT:    s_endpgm
     ptr %out, i32 %in) {
 entry:
   %val = atomicrmw volatile xchg ptr %out, i32 %in syncscope("wavefront") monotonic
   ret void
 }
 
 define amdgpu_kernel void @flat_wavefront_acquire_atomicrmw(

 ; GFX11-WGP:       ; %bb.0: ; %entry
 ; GFX11-WGP-NEXT:    s_clause 0x1
 ; GFX11-WGP-NEXT:    s_load_b64 s[2:3], s[0:1], 0x0
 ; GFX11-WGP-NEXT:    s_load_b32 s0, s[0:1], 0x8
 ; GFX11-WGP-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-WGP-NEXT:    v_dual_mov_b32 v0, s2 :: v_dual_mov_b32 v1, s3
 ; GFX11-WGP-NEXT:    v_mov_b32_e32 v2, s0
 ; GFX11-WGP-NEXT:    flat_atomic_swap_b32 v[0:1], v2
-; GFX11-WGP-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-WGP-NEXT:    s_endpgm
 ;
 ; GFX11-CU-LABEL: flat_wavefront_acquire_atomicrmw:
 ; GFX11-CU:       ; %bb.0: ; %entry
 ; GFX11-CU-NEXT:    s_clause 0x1
 ; GFX11-CU-NEXT:    s_load_b64 s[2:3], s[0:1], 0x0
 ; GFX11-CU-NEXT:    s_load_b32 s0, s[0:1], 0x8
 ; GFX11-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v0, s2 :: v_dual_mov_b32 v1, s3
 ; GFX11-CU-NEXT:    v_mov_b32_e32 v2, s0
 ; GFX11-CU-NEXT:    flat_atomic_swap_b32 v[0:1], v2
-; GFX11-CU-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-CU-NEXT:    s_endpgm
     ptr %out, i32 %in) {
 entry:
   %val = atomicrmw volatile xchg ptr %out, i32 %in syncscope("wavefront") acquire
   ret void
 }
 
 define amdgpu_kernel void @flat_wavefront_release_atomicrmw(

 ; GFX11-WGP:       ; %bb.0: ; %entry
 ; GFX11-WGP-NEXT:    s_clause 0x1
 ; GFX11-WGP-NEXT:    s_load_b64 s[2:3], s[0:1], 0x0
 ; GFX11-WGP-NEXT:    s_load_b32 s0, s[0:1], 0x8
 ; GFX11-WGP-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-WGP-NEXT:    v_dual_mov_b32 v0, s2 :: v_dual_mov_b32 v1, s3
 ; GFX11-WGP-NEXT:    v_mov_b32_e32 v2, s0
 ; GFX11-WGP-NEXT:    flat_atomic_swap_b32 v[0:1], v2
-; GFX11-WGP-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-WGP-NEXT:    s_endpgm
 ;
 ; GFX11-CU-LABEL: flat_wavefront_release_atomicrmw:
 ; GFX11-CU:       ; %bb.0: ; %entry
 ; GFX11-CU-NEXT:    s_clause 0x1
 ; GFX11-CU-NEXT:    s_load_b64 s[2:3], s[0:1], 0x0
 ; GFX11-CU-NEXT:    s_load_b32 s0, s[0:1], 0x8
 ; GFX11-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v0, s2 :: v_dual_mov_b32 v1, s3
 ; GFX11-CU-NEXT:    v_mov_b32_e32 v2, s0
 ; GFX11-CU-NEXT:    flat_atomic_swap_b32 v[0:1], v2
-; GFX11-CU-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-CU-NEXT:    s_endpgm
     ptr %out, i32 %in) {
 entry:
   %val = atomicrmw volatile xchg ptr %out, i32 %in syncscope("wavefront") release
   ret void
 }
 
 define amdgpu_kernel void @flat_wavefront_acq_rel_atomicrmw(

 ; GFX11-WGP:       ; %bb.0: ; %entry
 ; GFX11-WGP-NEXT:    s_clause 0x1
 ; GFX11-WGP-NEXT:    s_load_b64 s[2:3], s[0:1], 0x0
 ; GFX11-WGP-NEXT:    s_load_b32 s0, s[0:1], 0x8
 ; GFX11-WGP-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-WGP-NEXT:    v_dual_mov_b32 v0, s2 :: v_dual_mov_b32 v1, s3
 ; GFX11-WGP-NEXT:    v_mov_b32_e32 v2, s0
 ; GFX11-WGP-NEXT:    flat_atomic_swap_b32 v[0:1], v2
-; GFX11-WGP-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-WGP-NEXT:    s_endpgm
 ;
 ; GFX11-CU-LABEL: flat_wavefront_acq_rel_atomicrmw:
 ; GFX11-CU:       ; %bb.0: ; %entry
 ; GFX11-CU-NEXT:    s_clause 0x1
 ; GFX11-CU-NEXT:    s_load_b64 s[2:3], s[0:1], 0x0
 ; GFX11-CU-NEXT:    s_load_b32 s0, s[0:1], 0x8
 ; GFX11-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v0, s2 :: v_dual_mov_b32 v1, s3
 ; GFX11-CU-NEXT:    v_mov_b32_e32 v2, s0
 ; GFX11-CU-NEXT:    flat_atomic_swap_b32 v[0:1], v2
-; GFX11-CU-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-CU-NEXT:    s_endpgm
     ptr %out, i32 %in) {
 entry:
   %val = atomicrmw volatile xchg ptr %out, i32 %in syncscope("wavefront") acq_rel
   ret void
 }
 
 define amdgpu_kernel void @flat_wavefront_seq_cst_atomicrmw(

 ; GFX11-WGP:       ; %bb.0: ; %entry
 ; GFX11-WGP-NEXT:    s_clause 0x1
 ; GFX11-WGP-NEXT:    s_load_b64 s[2:3], s[0:1], 0x0
 ; GFX11-WGP-NEXT:    s_load_b32 s0, s[0:1], 0x8
 ; GFX11-WGP-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-WGP-NEXT:    v_dual_mov_b32 v0, s2 :: v_dual_mov_b32 v1, s3
 ; GFX11-WGP-NEXT:    v_mov_b32_e32 v2, s0
 ; GFX11-WGP-NEXT:    flat_atomic_swap_b32 v[0:1], v2
-; GFX11-WGP-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-WGP-NEXT:    s_endpgm
 ;
 ; GFX11-CU-LABEL: flat_wavefront_seq_cst_atomicrmw:
 ; GFX11-CU:       ; %bb.0: ; %entry
 ; GFX11-CU-NEXT:    s_clause 0x1
 ; GFX11-CU-NEXT:    s_load_b64 s[2:3], s[0:1], 0x0
 ; GFX11-CU-NEXT:    s_load_b32 s0, s[0:1], 0x8
 ; GFX11-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v0, s2 :: v_dual_mov_b32 v1, s3
 ; GFX11-CU-NEXT:    v_mov_b32_e32 v2, s0
 ; GFX11-CU-NEXT:    flat_atomic_swap_b32 v[0:1], v2
-; GFX11-CU-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-CU-NEXT:    s_endpgm
     ptr %out, i32 %in) {
 entry:
   %val = atomicrmw volatile xchg ptr %out, i32 %in syncscope("wavefront") seq_cst
   ret void
 }
 
 define amdgpu_kernel void @flat_wavefront_acquire_ret_atomicrmw(

 ; GFX11-WGP-NEXT:    s_load_b64 s[2:3], s[0:1], 0x0
 ; GFX11-WGP-NEXT:    s_load_b32 s0, s[0:1], 0x8
 ; GFX11-WGP-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-WGP-NEXT:    v_dual_mov_b32 v0, s2 :: v_dual_mov_b32 v1, s3
 ; GFX11-WGP-NEXT:    v_mov_b32_e32 v2, s0
 ; GFX11-WGP-NEXT:    flat_atomic_swap_b32 v2, v[0:1], v2 glc
 ; GFX11-WGP-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX11-WGP-NEXT:    flat_store_b32 v[0:1], v2
-; GFX11-WGP-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-WGP-NEXT:    s_endpgm
 ;
 ; GFX11-CU-LABEL: flat_wavefront_acquire_ret_atomicrmw:
 ; GFX11-CU:       ; %bb.0: ; %entry
 ; GFX11-CU-NEXT:    s_clause 0x1
 ; GFX11-CU-NEXT:    s_load_b64 s[2:3], s[0:1], 0x0
 ; GFX11-CU-NEXT:    s_load_b32 s0, s[0:1], 0x8
 ; GFX11-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v0, s2 :: v_dual_mov_b32 v1, s3
 ; GFX11-CU-NEXT:    v_mov_b32_e32 v2, s0
 ; GFX11-CU-NEXT:    flat_atomic_swap_b32 v2, v[0:1], v2 glc
 ; GFX11-CU-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX11-CU-NEXT:    flat_store_b32 v[0:1], v2
-; GFX11-CU-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-CU-NEXT:    s_endpgm
     ptr %out, i32 %in) {
 entry:
   %val = atomicrmw volatile xchg ptr %out, i32 %in syncscope("wavefront") acquire
   store i32 %val, ptr %out, align 4
   ret void
 }
 

 ; GFX11-WGP-NEXT:    s_load_b64 s[2:3], s[0:1], 0x0
 ; GFX11-WGP-NEXT:    s_load_b32 s0, s[0:1], 0x8
 ; GFX11-WGP-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-WGP-NEXT:    v_dual_mov_b32 v0, s2 :: v_dual_mov_b32 v1, s3
 ; GFX11-WGP-NEXT:    v_mov_b32_e32 v2, s0
 ; GFX11-WGP-NEXT:    flat_atomic_swap_b32 v2, v[0:1], v2 glc
 ; GFX11-WGP-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX11-WGP-NEXT:    flat_store_b32 v[0:1], v2
-; GFX11-WGP-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-WGP-NEXT:    s_endpgm
 ;
 ; GFX11-CU-LABEL: flat_wavefront_acq_rel_ret_atomicrmw:
 ; GFX11-CU:       ; %bb.0: ; %entry
 ; GFX11-CU-NEXT:    s_clause 0x1
 ; GFX11-CU-NEXT:    s_load_b64 s[2:3], s[0:1], 0x0
 ; GFX11-CU-NEXT:    s_load_b32 s0, s[0:1], 0x8
 ; GFX11-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v0, s2 :: v_dual_mov_b32 v1, s3
 ; GFX11-CU-NEXT:    v_mov_b32_e32 v2, s0
 ; GFX11-CU-NEXT:    flat_atomic_swap_b32 v2, v[0:1], v2 glc
 ; GFX11-CU-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX11-CU-NEXT:    flat_store_b32 v[0:1], v2
-; GFX11-CU-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-CU-NEXT:    s_endpgm
     ptr %out, i32 %in) {
 entry:
   %val = atomicrmw volatile xchg ptr %out, i32 %in syncscope("wavefront") acq_rel
   store i32 %val, ptr %out, align 4
   ret void
 }
 

 ; GFX11-WGP-NEXT:    s_load_b64 s[2:3], s[0:1], 0x0
 ; GFX11-WGP-NEXT:    s_load_b32 s0, s[0:1], 0x8
 ; GFX11-WGP-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-WGP-NEXT:    v_dual_mov_b32 v0, s2 :: v_dual_mov_b32 v1, s3
 ; GFX11-WGP-NEXT:    v_mov_b32_e32 v2, s0
 ; GFX11-WGP-NEXT:    flat_atomic_swap_b32 v2, v[0:1], v2 glc
 ; GFX11-WGP-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX11-WGP-NEXT:    flat_store_b32 v[0:1], v2
-; GFX11-WGP-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-WGP-NEXT:    s_endpgm
 ;
 ; GFX11-CU-LABEL: flat_wavefront_seq_cst_ret_atomicrmw:
 ; GFX11-CU:       ; %bb.0: ; %entry
 ; GFX11-CU-NEXT:    s_clause 0x1
 ; GFX11-CU-NEXT:    s_load_b64 s[2:3], s[0:1], 0x0
 ; GFX11-CU-NEXT:    s_load_b32 s0, s[0:1], 0x8
 ; GFX11-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v0, s2 :: v_dual_mov_b32 v1, s3
 ; GFX11-CU-NEXT:    v_mov_b32_e32 v2, s0
 ; GFX11-CU-NEXT:    flat_atomic_swap_b32 v2, v[0:1], v2 glc
 ; GFX11-CU-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX11-CU-NEXT:    flat_store_b32 v[0:1], v2
-; GFX11-CU-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-CU-NEXT:    s_endpgm
     ptr %out, i32 %in) {
 entry:
   %val = atomicrmw volatile xchg ptr %out, i32 %in syncscope("wavefront") seq_cst
   store i32 %val, ptr %out, align 4
   ret void
 }
 

 ;
 ; GFX11-WGP-LABEL: flat_wavefront_monotonic_monotonic_cmpxchg:
 ; GFX11-WGP:       ; %bb.0: ; %entry
 ; GFX11-WGP-NEXT:    s_load_b128 s[0:3], s[0:1], 0x0
 ; GFX11-WGP-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-WGP-NEXT:    v_dual_mov_b32 v0, s0 :: v_dual_mov_b32 v1, s1
 ; GFX11-WGP-NEXT:    v_dual_mov_b32 v2, s2 :: v_dual_mov_b32 v3, s3
 ; GFX11-WGP-NEXT:    flat_atomic_cmpswap_b32 v[0:1], v[2:3] offset:16
-; GFX11-WGP-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-WGP-NEXT:    s_endpgm
 ;
 ; GFX11-CU-LABEL: flat_wavefront_monotonic_monotonic_cmpxchg:
 ; GFX11-CU:       ; %bb.0: ; %entry
 ; GFX11-CU-NEXT:    s_load_b128 s[0:3], s[0:1], 0x0
 ; GFX11-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v0, s0 :: v_dual_mov_b32 v1, s1
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v2, s2 :: v_dual_mov_b32 v3, s3
 ; GFX11-CU-NEXT:    flat_atomic_cmpswap_b32 v[0:1], v[2:3] offset:16
-; GFX11-CU-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-CU-NEXT:    s_endpgm
     ptr %out, i32 %in, i32 %old) {
 entry:
   %gep = getelementptr i32, ptr %out, i32 4
   %val = cmpxchg volatile ptr %gep, i32 %old, i32 %in syncscope("wavefront") monotonic monotonic
   ret void
 }
 

 ;
 ; GFX11-WGP-LABEL: flat_wavefront_acquire_monotonic_cmpxchg:
 ; GFX11-WGP:       ; %bb.0: ; %entry
 ; GFX11-WGP-NEXT:    s_load_b128 s[0:3], s[0:1], 0x0
 ; GFX11-WGP-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-WGP-NEXT:    v_dual_mov_b32 v0, s0 :: v_dual_mov_b32 v1, s1
 ; GFX11-WGP-NEXT:    v_dual_mov_b32 v2, s2 :: v_dual_mov_b32 v3, s3
 ; GFX11-WGP-NEXT:    flat_atomic_cmpswap_b32 v[0:1], v[2:3] offset:16
-; GFX11-WGP-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-WGP-NEXT:    s_endpgm
 ;
 ; GFX11-CU-LABEL: flat_wavefront_acquire_monotonic_cmpxchg:
 ; GFX11-CU:       ; %bb.0: ; %entry
 ; GFX11-CU-NEXT:    s_load_b128 s[0:3], s[0:1], 0x0
 ; GFX11-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v0, s0 :: v_dual_mov_b32 v1, s1
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v2, s2 :: v_dual_mov_b32 v3, s3
 ; GFX11-CU-NEXT:    flat_atomic_cmpswap_b32 v[0:1], v[2:3] offset:16
-; GFX11-CU-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-CU-NEXT:    s_endpgm
     ptr %out, i32 %in, i32 %old) {
 entry:
   %gep = getelementptr i32, ptr %out, i32 4
   %val = cmpxchg volatile ptr %gep, i32 %old, i32 %in syncscope("wavefront") acquire monotonic
   ret void
 }
 

 ;
 ; GFX11-WGP-LABEL: flat_wavefront_release_monotonic_cmpxchg:
 ; GFX11-WGP:       ; %bb.0: ; %entry
 ; GFX11-WGP-NEXT:    s_load_b128 s[0:3], s[0:1], 0x0
 ; GFX11-WGP-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-WGP-NEXT:    v_dual_mov_b32 v0, s0 :: v_dual_mov_b32 v1, s1
 ; GFX11-WGP-NEXT:    v_dual_mov_b32 v2, s2 :: v_dual_mov_b32 v3, s3
 ; GFX11-WGP-NEXT:    flat_atomic_cmpswap_b32 v[0:1], v[2:3] offset:16
-; GFX11-WGP-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-WGP-NEXT:    s_endpgm
 ;
 ; GFX11-CU-LABEL: flat_wavefront_release_monotonic_cmpxchg:
 ; GFX11-CU:       ; %bb.0: ; %entry
 ; GFX11-CU-NEXT:    s_load_b128 s[0:3], s[0:1], 0x0
 ; GFX11-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v0, s0 :: v_dual_mov_b32 v1, s1
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v2, s2 :: v_dual_mov_b32 v3, s3
 ; GFX11-CU-NEXT:    flat_atomic_cmpswap_b32 v[0:1], v[2:3] offset:16
-; GFX11-CU-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-CU-NEXT:    s_endpgm
     ptr %out, i32 %in, i32 %old) {
 entry:
   %gep = getelementptr i32, ptr %out, i32 4
   %val = cmpxchg volatile ptr %gep, i32 %old, i32 %in syncscope("wavefront") release monotonic
   ret void
 }
 

 ;
 ; GFX11-WGP-LABEL: flat_wavefront_acq_rel_monotonic_cmpxchg:
 ; GFX11-WGP:       ; %bb.0: ; %entry
 ; GFX11-WGP-NEXT:    s_load_b128 s[0:3], s[0:1], 0x0
 ; GFX11-WGP-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-WGP-NEXT:    v_dual_mov_b32 v0, s0 :: v_dual_mov_b32 v1, s1
 ; GFX11-WGP-NEXT:    v_dual_mov_b32 v2, s2 :: v_dual_mov_b32 v3, s3
 ; GFX11-WGP-NEXT:    flat_atomic_cmpswap_b32 v[0:1], v[2:3] offset:16
-; GFX11-WGP-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-WGP-NEXT:    s_endpgm
 ;
 ; GFX11-CU-LABEL: flat_wavefront_acq_rel_monotonic_cmpxchg:
 ; GFX11-CU:       ; %bb.0: ; %entry
 ; GFX11-CU-NEXT:    s_load_b128 s[0:3], s[0:1], 0x0
 ; GFX11-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v0, s0 :: v_dual_mov_b32 v1, s1
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v2, s2 :: v_dual_mov_b32 v3, s3
 ; GFX11-CU-NEXT:    flat_atomic_cmpswap_b32 v[0:1], v[2:3] offset:16
-; GFX11-CU-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-CU-NEXT:    s_endpgm
     ptr %out, i32 %in, i32 %old) {
 entry:
   %gep = getelementptr i32, ptr %out, i32 4
   %val = cmpxchg volatile ptr %gep, i32 %old, i32 %in syncscope("wavefront") acq_rel monotonic
   ret void
 }
 

 ;
 ; GFX11-WGP-LABEL: flat_wavefront_seq_cst_monotonic_cmpxchg:
 ; GFX11-WGP:       ; %bb.0: ; %entry
 ; GFX11-WGP-NEXT:    s_load_b128 s[0:3], s[0:1], 0x0
 ; GFX11-WGP-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-WGP-NEXT:    v_dual_mov_b32 v0, s0 :: v_dual_mov_b32 v1, s1
 ; GFX11-WGP-NEXT:    v_dual_mov_b32 v2, s2 :: v_dual_mov_b32 v3, s3
 ; GFX11-WGP-NEXT:    flat_atomic_cmpswap_b32 v[0:1], v[2:3] offset:16
-; GFX11-WGP-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-WGP-NEXT:    s_endpgm
 ;
 ; GFX11-CU-LABEL: flat_wavefront_seq_cst_monotonic_cmpxchg:
 ; GFX11-CU:       ; %bb.0: ; %entry
 ; GFX11-CU-NEXT:    s_load_b128 s[0:3], s[0:1], 0x0
 ; GFX11-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v0, s0 :: v_dual_mov_b32 v1, s1
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v2, s2 :: v_dual_mov_b32 v3, s3
 ; GFX11-CU-NEXT:    flat_atomic_cmpswap_b32 v[0:1], v[2:3] offset:16
-; GFX11-CU-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-CU-NEXT:    s_endpgm
     ptr %out, i32 %in, i32 %old) {
 entry:
   %gep = getelementptr i32, ptr %out, i32 4
   %val = cmpxchg volatile ptr %gep, i32 %old, i32 %in syncscope("wavefront") seq_cst monotonic
   ret void
 }
 

 ;
 ; GFX11-WGP-LABEL: flat_wavefront_monotonic_acquire_cmpxchg:
 ; GFX11-WGP:       ; %bb.0: ; %entry
 ; GFX11-WGP-NEXT:    s_load_b128 s[0:3], s[0:1], 0x0
 ; GFX11-WGP-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-WGP-NEXT:    v_dual_mov_b32 v0, s0 :: v_dual_mov_b32 v1, s1
 ; GFX11-WGP-NEXT:    v_dual_mov_b32 v2, s2 :: v_dual_mov_b32 v3, s3
 ; GFX11-WGP-NEXT:    flat_atomic_cmpswap_b32 v[0:1], v[2:3] offset:16
-; GFX11-WGP-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-WGP-NEXT:    s_endpgm
 ;
 ; GFX11-CU-LABEL: flat_wavefront_monotonic_acquire_cmpxchg:
 ; GFX11-CU:       ; %bb.0: ; %entry
 ; GFX11-CU-NEXT:    s_load_b128 s[0:3], s[0:1], 0x0
 ; GFX11-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v0, s0 :: v_dual_mov_b32 v1, s1
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v2, s2 :: v_dual_mov_b32 v3, s3
 ; GFX11-CU-NEXT:    flat_atomic_cmpswap_b32 v[0:1], v[2:3] offset:16
-; GFX11-CU-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-CU-NEXT:    s_endpgm
     ptr %out, i32 %in, i32 %old) {
 entry:
   %gep = getelementptr i32, ptr %out, i32 4
   %val = cmpxchg volatile ptr %gep, i32 %old, i32 %in syncscope("wavefront") monotonic acquire
   ret void
 }
 

 ;
 ; GFX11-WGP-LABEL: flat_wavefront_acquire_acquire_cmpxchg:
 ; GFX11-WGP:       ; %bb.0: ; %entry
 ; GFX11-WGP-NEXT:    s_load_b128 s[0:3], s[0:1], 0x0
 ; GFX11-WGP-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-WGP-NEXT:    v_dual_mov_b32 v0, s0 :: v_dual_mov_b32 v1, s1
 ; GFX11-WGP-NEXT:    v_dual_mov_b32 v2, s2 :: v_dual_mov_b32 v3, s3
 ; GFX11-WGP-NEXT:    flat_atomic_cmpswap_b32 v[0:1], v[2:3] offset:16
-; GFX11-WGP-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-WGP-NEXT:    s_endpgm
 ;
 ; GFX11-CU-LABEL: flat_wavefront_acquire_acquire_cmpxchg:
 ; GFX11-CU:       ; %bb.0: ; %entry
 ; GFX11-CU-NEXT:    s_load_b128 s[0:3], s[0:1], 0x0
 ; GFX11-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v0, s0 :: v_dual_mov_b32 v1, s1
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v2, s2 :: v_dual_mov_b32 v3, s3
 ; GFX11-CU-NEXT:    flat_atomic_cmpswap_b32 v[0:1], v[2:3] offset:16
-; GFX11-CU-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-CU-NEXT:    s_endpgm
     ptr %out, i32 %in, i32 %old) {
 entry:
   %gep = getelementptr i32, ptr %out, i32 4
   %val = cmpxchg volatile ptr %gep, i32 %old, i32 %in syncscope("wavefront") acquire acquire
   ret void
 }
 

 ;
 ; GFX11-WGP-LABEL: flat_wavefront_release_acquire_cmpxchg:
 ; GFX11-WGP:       ; %bb.0: ; %entry
 ; GFX11-WGP-NEXT:    s_load_b128 s[0:3], s[0:1], 0x0
 ; GFX11-WGP-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-WGP-NEXT:    v_dual_mov_b32 v0, s0 :: v_dual_mov_b32 v1, s1
 ; GFX11-WGP-NEXT:    v_dual_mov_b32 v2, s2 :: v_dual_mov_b32 v3, s3
 ; GFX11-WGP-NEXT:    flat_atomic_cmpswap_b32 v[0:1], v[2:3] offset:16
-; GFX11-WGP-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-WGP-NEXT:    s_endpgm
 ;
 ; GFX11-CU-LABEL: flat_wavefront_release_acquire_cmpxchg:
 ; GFX11-CU:       ; %bb.0: ; %entry
 ; GFX11-CU-NEXT:    s_load_b128 s[0:3], s[0:1], 0x0
 ; GFX11-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v0, s0 :: v_dual_mov_b32 v1, s1
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v2, s2 :: v_dual_mov_b32 v3, s3
 ; GFX11-CU-NEXT:    flat_atomic_cmpswap_b32 v[0:1], v[2:3] offset:16
-; GFX11-CU-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-CU-NEXT:    s_endpgm
     ptr %out, i32 %in, i32 %old) {
 entry:
   %gep = getelementptr i32, ptr %out, i32 4
   %val = cmpxchg volatile ptr %gep, i32 %old, i32 %in syncscope("wavefront") release acquire
   ret void
 }
 

 ;
 ; GFX11-WGP-LABEL: flat_wavefront_acq_rel_acquire_cmpxchg:
 ; GFX11-WGP:       ; %bb.0: ; %entry
 ; GFX11-WGP-NEXT:    s_load_b128 s[0:3], s[0:1], 0x0
 ; GFX11-WGP-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-WGP-NEXT:    v_dual_mov_b32 v0, s0 :: v_dual_mov_b32 v1, s1
 ; GFX11-WGP-NEXT:    v_dual_mov_b32 v2, s2 :: v_dual_mov_b32 v3, s3
 ; GFX11-WGP-NEXT:    flat_atomic_cmpswap_b32 v[0:1], v[2:3] offset:16
-; GFX11-WGP-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-WGP-NEXT:    s_endpgm
 ;
 ; GFX11-CU-LABEL: flat_wavefront_acq_rel_acquire_cmpxchg:
 ; GFX11-CU:       ; %bb.0: ; %entry
 ; GFX11-CU-NEXT:    s_load_b128 s[0:3], s[0:1], 0x0
 ; GFX11-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v0, s0 :: v_dual_mov_b32 v1, s1
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v2, s2 :: v_dual_mov_b32 v3, s3
 ; GFX11-CU-NEXT:    flat_atomic_cmpswap_b32 v[0:1], v[2:3] offset:16
-; GFX11-CU-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-CU-NEXT:    s_endpgm
     ptr %out, i32 %in, i32 %old) {
 entry:
   %gep = getelementptr i32, ptr %out, i32 4
   %val = cmpxchg volatile ptr %gep, i32 %old, i32 %in syncscope("wavefront") acq_rel acquire
   ret void
 }
 

 ;
 ; GFX11-WGP-LABEL: flat_wavefront_seq_cst_acquire_cmpxchg:
 ; GFX11-WGP:       ; %bb.0: ; %entry
 ; GFX11-WGP-NEXT:    s_load_b128 s[0:3], s[0:1], 0x0
 ; GFX11-WGP-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-WGP-NEXT:    v_dual_mov_b32 v0, s0 :: v_dual_mov_b32 v1, s1
 ; GFX11-WGP-NEXT:    v_dual_mov_b32 v2, s2 :: v_dual_mov_b32 v3, s3
 ; GFX11-WGP-NEXT:    flat_atomic_cmpswap_b32 v[0:1], v[2:3] offset:16
-; GFX11-WGP-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-WGP-NEXT:    s_endpgm
 ;
 ; GFX11-CU-LABEL: flat_wavefront_seq_cst_acquire_cmpxchg:
 ; GFX11-CU:       ; %bb.0: ; %entry
 ; GFX11-CU-NEXT:    s_load_b128 s[0:3], s[0:1], 0x0
 ; GFX11-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v0, s0 :: v_dual_mov_b32 v1, s1
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v2, s2 :: v_dual_mov_b32 v3, s3
 ; GFX11-CU-NEXT:    flat_atomic_cmpswap_b32 v[0:1], v[2:3] offset:16
-; GFX11-CU-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-CU-NEXT:    s_endpgm
     ptr %out, i32 %in, i32 %old) {
 entry:
   %gep = getelementptr i32, ptr %out, i32 4
   %val = cmpxchg volatile ptr %gep, i32 %old, i32 %in syncscope("wavefront") seq_cst acquire
   ret void
 }
 

 ;
 ; GFX11-WGP-LABEL: flat_wavefront_monotonic_seq_cst_cmpxchg:
 ; GFX11-WGP:       ; %bb.0: ; %entry
 ; GFX11-WGP-NEXT:    s_load_b128 s[0:3], s[0:1], 0x0
 ; GFX11-WGP-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-WGP-NEXT:    v_dual_mov_b32 v0, s0 :: v_dual_mov_b32 v1, s1
 ; GFX11-WGP-NEXT:    v_dual_mov_b32 v2, s2 :: v_dual_mov_b32 v3, s3
 ; GFX11-WGP-NEXT:    flat_atomic_cmpswap_b32 v[0:1], v[2:3] offset:16
-; GFX11-WGP-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-WGP-NEXT:    s_endpgm
 ;
 ; GFX11-CU-LABEL: flat_wavefront_monotonic_seq_cst_cmpxchg:
 ; GFX11-CU:       ; %bb.0: ; %entry
 ; GFX11-CU-NEXT:    s_load_b128 s[0:3], s[0:1], 0x0
 ; GFX11-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v0, s0 :: v_dual_mov_b32 v1, s1
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v2, s2 :: v_dual_mov_b32 v3, s3
 ; GFX11-CU-NEXT:    flat_atomic_cmpswap_b32 v[0:1], v[2:3] offset:16
-; GFX11-CU-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-CU-NEXT:    s_endpgm
     ptr %out, i32 %in, i32 %old) {
 entry:
   %gep = getelementptr i32, ptr %out, i32 4
   %val = cmpxchg volatile ptr %gep, i32 %old, i32 %in syncscope("wavefront") monotonic seq_cst
   ret void
 }
 

 ;
 ; GFX11-WGP-LABEL: flat_wavefront_acquire_seq_cst_cmpxchg:
 ; GFX11-WGP:       ; %bb.0: ; %entry
 ; GFX11-WGP-NEXT:    s_load_b128 s[0:3], s[0:1], 0x0
 ; GFX11-WGP-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-WGP-NEXT:    v_dual_mov_b32 v0, s0 :: v_dual_mov_b32 v1, s1
 ; GFX11-WGP-NEXT:    v_dual_mov_b32 v2, s2 :: v_dual_mov_b32 v3, s3
 ; GFX11-WGP-NEXT:    flat_atomic_cmpswap_b32 v[0:1], v[2:3] offset:16
-; GFX11-WGP-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-WGP-NEXT:    s_endpgm
 ;
 ; GFX11-CU-LABEL: flat_wavefront_acquire_seq_cst_cmpxchg:
 ; GFX11-CU:       ; %bb.0: ; %entry
 ; GFX11-CU-NEXT:    s_load_b128 s[0:3], s[0:1], 0x0
 ; GFX11-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v0, s0 :: v_dual_mov_b32 v1, s1
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v2, s2 :: v_dual_mov_b32 v3, s3
 ; GFX11-CU-NEXT:    flat_atomic_cmpswap_b32 v[0:1], v[2:3] offset:16
-; GFX11-CU-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-CU-NEXT:    s_endpgm
     ptr %out, i32 %in, i32 %old) {
 entry:
   %gep = getelementptr i32, ptr %out, i32 4
   %val = cmpxchg volatile ptr %gep, i32 %old, i32 %in syncscope("wavefront") acquire seq_cst
   ret void
 }
 

 ;
 ; GFX11-WGP-LABEL: flat_wavefront_release_seq_cst_cmpxchg:
 ; GFX11-WGP:       ; %bb.0: ; %entry
 ; GFX11-WGP-NEXT:    s_load_b128 s[0:3], s[0:1], 0x0
 ; GFX11-WGP-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-WGP-NEXT:    v_dual_mov_b32 v0, s0 :: v_dual_mov_b32 v1, s1
 ; GFX11-WGP-NEXT:    v_dual_mov_b32 v2, s2 :: v_dual_mov_b32 v3, s3
 ; GFX11-WGP-NEXT:    flat_atomic_cmpswap_b32 v[0:1], v[2:3] offset:16
-; GFX11-WGP-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-WGP-NEXT:    s_endpgm
 ;
 ; GFX11-CU-LABEL: flat_wavefront_release_seq_cst_cmpxchg:
 ; GFX11-CU:       ; %bb.0: ; %entry
 ; GFX11-CU-NEXT:    s_load_b128 s[0:3], s[0:1], 0x0
 ; GFX11-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v0, s0 :: v_dual_mov_b32 v1, s1
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v2, s2 :: v_dual_mov_b32 v3, s3
 ; GFX11-CU-NEXT:    flat_atomic_cmpswap_b32 v[0:1], v[2:3] offset:16
-; GFX11-CU-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-CU-NEXT:    s_endpgm
     ptr %out, i32 %in, i32 %old) {
 entry:
   %gep = getelementptr i32, ptr %out, i32 4
   %val = cmpxchg volatile ptr %gep, i32 %old, i32 %in syncscope("wavefront") release seq_cst
   ret void
 }
 

 ;
 ; GFX11-WGP-LABEL: flat_wavefront_acq_rel_seq_cst_cmpxchg:
 ; GFX11-WGP:       ; %bb.0: ; %entry
 ; GFX11-WGP-NEXT:    s_load_b128 s[0:3], s[0:1], 0x0
 ; GFX11-WGP-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-WGP-NEXT:    v_dual_mov_b32 v0, s0 :: v_dual_mov_b32 v1, s1
 ; GFX11-WGP-NEXT:    v_dual_mov_b32 v2, s2 :: v_dual_mov_b32 v3, s3
 ; GFX11-WGP-NEXT:    flat_atomic_cmpswap_b32 v[0:1], v[2:3] offset:16
-; GFX11-WGP-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-WGP-NEXT:    s_endpgm
 ;
 ; GFX11-CU-LABEL: flat_wavefront_acq_rel_seq_cst_cmpxchg:
 ; GFX11-CU:       ; %bb.0: ; %entry
 ; GFX11-CU-NEXT:    s_load_b128 s[0:3], s[0:1], 0x0
 ; GFX11-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v0, s0 :: v_dual_mov_b32 v1, s1
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v2, s2 :: v_dual_mov_b32 v3, s3
 ; GFX11-CU-NEXT:    flat_atomic_cmpswap_b32 v[0:1], v[2:3] offset:16
-; GFX11-CU-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-CU-NEXT:    s_endpgm
     ptr %out, i32 %in, i32 %old) {
 entry:
   %gep = getelementptr i32, ptr %out, i32 4
   %val = cmpxchg volatile ptr %gep, i32 %old, i32 %in syncscope("wavefront") acq_rel seq_cst
   ret void
 }
 

 ;
 ; GFX11-WGP-LABEL: flat_wavefront_seq_cst_seq_cst_cmpxchg:
 ; GFX11-WGP:       ; %bb.0: ; %entry
 ; GFX11-WGP-NEXT:    s_load_b128 s[0:3], s[0:1], 0x0
 ; GFX11-WGP-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-WGP-NEXT:    v_dual_mov_b32 v0, s0 :: v_dual_mov_b32 v1, s1
 ; GFX11-WGP-NEXT:    v_dual_mov_b32 v2, s2 :: v_dual_mov_b32 v3, s3
 ; GFX11-WGP-NEXT:    flat_atomic_cmpswap_b32 v[0:1], v[2:3] offset:16
-; GFX11-WGP-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-WGP-NEXT:    s_endpgm
 ;
 ; GFX11-CU-LABEL: flat_wavefront_seq_cst_seq_cst_cmpxchg:
 ; GFX11-CU:       ; %bb.0: ; %entry
 ; GFX11-CU-NEXT:    s_load_b128 s[0:3], s[0:1], 0x0
 ; GFX11-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v0, s0 :: v_dual_mov_b32 v1, s1
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v2, s2 :: v_dual_mov_b32 v3, s3
 ; GFX11-CU-NEXT:    flat_atomic_cmpswap_b32 v[0:1], v[2:3] offset:16
-; GFX11-CU-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-CU-NEXT:    s_endpgm
     ptr %out, i32 %in, i32 %old) {
 entry:
   %gep = getelementptr i32, ptr %out, i32 4
   %val = cmpxchg volatile ptr %gep, i32 %old, i32 %in syncscope("wavefront") seq_cst seq_cst
   ret void
 }
 

 ; GFX11-WGP:       ; %bb.0: ; %entry
 ; GFX11-WGP-NEXT:    s_load_b128 s[0:3], s[0:1], 0x0
 ; GFX11-WGP-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-WGP-NEXT:    v_dual_mov_b32 v0, s0 :: v_dual_mov_b32 v1, s1
 ; GFX11-WGP-NEXT:    v_dual_mov_b32 v2, s2 :: v_dual_mov_b32 v3, s3
 ; GFX11-WGP-NEXT:    flat_atomic_cmpswap_b32 v2, v[0:1], v[2:3] offset:16 glc
 ; GFX11-WGP-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX11-WGP-NEXT:    flat_store_b32 v[0:1], v2
-; GFX11-WGP-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-WGP-NEXT:    s_endpgm
 ;
 ; GFX11-CU-LABEL: flat_wavefront_monotonic_monotonic_ret_cmpxchg:
 ; GFX11-CU:       ; %bb.0: ; %entry
 ; GFX11-CU-NEXT:    s_load_b128 s[0:3], s[0:1], 0x0
 ; GFX11-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v0, s0 :: v_dual_mov_b32 v1, s1
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v2, s2 :: v_dual_mov_b32 v3, s3
 ; GFX11-CU-NEXT:    flat_atomic_cmpswap_b32 v2, v[0:1], v[2:3] offset:16 glc
 ; GFX11-CU-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX11-CU-NEXT:    flat_store_b32 v[0:1], v2
-; GFX11-CU-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-CU-NEXT:    s_endpgm
     ptr %out, i32 %in, i32 %old) {
 entry:
   %gep = getelementptr i32, ptr %out, i32 4
   %val = cmpxchg volatile ptr %gep, i32 %old, i32 %in syncscope("wavefront") monotonic monotonic
   %val0 = extractvalue { i32, i1 } %val, 0
   store i32 %val0, ptr %out, align 4
   ret void

 ; GFX11-WGP:       ; %bb.0: ; %entry
 ; GFX11-WGP-NEXT:    s_load_b128 s[0:3], s[0:1], 0x0
 ; GFX11-WGP-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-WGP-NEXT:    v_dual_mov_b32 v0, s0 :: v_dual_mov_b32 v1, s1
 ; GFX11-WGP-NEXT:    v_dual_mov_b32 v2, s2 :: v_dual_mov_b32 v3, s3
 ; GFX11-WGP-NEXT:    flat_atomic_cmpswap_b32 v2, v[0:1], v[2:3] offset:16 glc
 ; GFX11-WGP-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX11-WGP-NEXT:    flat_store_b32 v[0:1], v2
-; GFX11-WGP-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-WGP-NEXT:    s_endpgm
 ;
 ; GFX11-CU-LABEL: flat_wavefront_acquire_monotonic_ret_cmpxchg:
 ; GFX11-CU:       ; %bb.0: ; %entry
 ; GFX11-CU-NEXT:    s_load_b128 s[0:3], s[0:1], 0x0
 ; GFX11-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v0, s0 :: v_dual_mov_b32 v1, s1
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v2, s2 :: v_dual_mov_b32 v3, s3
 ; GFX11-CU-NEXT:    flat_atomic_cmpswap_b32 v2, v[0:1], v[2:3] offset:16 glc
 ; GFX11-CU-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX11-CU-NEXT:    flat_store_b32 v[0:1], v2
-; GFX11-CU-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-CU-NEXT:    s_endpgm
     ptr %out, i32 %in, i32 %old) {
 entry:
   %gep = getelementptr i32, ptr %out, i32 4
   %val = cmpxchg volatile ptr %gep, i32 %old, i32 %in syncscope("wavefront") acquire monotonic
   %val0 = extractvalue { i32, i1 } %val, 0
   store i32 %val0, ptr %out, align 4
   ret void

 ; GFX11-WGP:       ; %bb.0: ; %entry
 ; GFX11-WGP-NEXT:    s_load_b128 s[0:3], s[0:1], 0x0
 ; GFX11-WGP-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-WGP-NEXT:    v_dual_mov_b32 v0, s0 :: v_dual_mov_b32 v1, s1
 ; GFX11-WGP-NEXT:    v_dual_mov_b32 v2, s2 :: v_dual_mov_b32 v3, s3
 ; GFX11-WGP-NEXT:    flat_atomic_cmpswap_b32 v2, v[0:1], v[2:3] offset:16 glc
 ; GFX11-WGP-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX11-WGP-NEXT:    flat_store_b32 v[0:1], v2
-; GFX11-WGP-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-WGP-NEXT:    s_endpgm
 ;
 ; GFX11-CU-LABEL: flat_wavefront_release_monotonic_ret_cmpxchg:
 ; GFX11-CU:       ; %bb.0: ; %entry
 ; GFX11-CU-NEXT:    s_load_b128 s[0:3], s[0:1], 0x0
 ; GFX11-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v0, s0 :: v_dual_mov_b32 v1, s1
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v2, s2 :: v_dual_mov_b32 v3, s3
 ; GFX11-CU-NEXT:    flat_atomic_cmpswap_b32 v2, v[0:1], v[2:3] offset:16 glc
 ; GFX11-CU-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX11-CU-NEXT:    flat_store_b32 v[0:1], v2
-; GFX11-CU-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-CU-NEXT:    s_endpgm
     ptr %out, i32 %in, i32 %old) {
 entry:
   %gep = getelementptr i32, ptr %out, i32 4
   %val = cmpxchg volatile ptr %gep, i32 %old, i32 %in syncscope("wavefront") release monotonic
   %val0 = extractvalue { i32, i1 } %val, 0
   store i32 %val0, ptr %out, align 4
   ret void

 ; GFX11-WGP:       ; %bb.0: ; %entry
 ; GFX11-WGP-NEXT:    s_load_b128 s[0:3], s[0:1], 0x0
 ; GFX11-WGP-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-WGP-NEXT:    v_dual_mov_b32 v0, s0 :: v_dual_mov_b32 v1, s1
 ; GFX11-WGP-NEXT:    v_dual_mov_b32 v2, s2 :: v_dual_mov_b32 v3, s3
 ; GFX11-WGP-NEXT:    flat_atomic_cmpswap_b32 v2, v[0:1], v[2:3] offset:16 glc
 ; GFX11-WGP-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX11-WGP-NEXT:    flat_store_b32 v[0:1], v2
-; GFX11-WGP-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-WGP-NEXT:    s_endpgm
 ;
 ; GFX11-CU-LABEL: flat_wavefront_acq_rel_monotonic_ret_cmpxchg:
 ; GFX11-CU:       ; %bb.0: ; %entry
 ; GFX11-CU-NEXT:    s_load_b128 s[0:3], s[0:1], 0x0
 ; GFX11-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v0, s0 :: v_dual_mov_b32 v1, s1
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v2, s2 :: v_dual_mov_b32 v3, s3
 ; GFX11-CU-NEXT:    flat_atomic_cmpswap_b32 v2, v[0:1], v[2:3] offset:16 glc
 ; GFX11-CU-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX11-CU-NEXT:    flat_store_b32 v[0:1], v2
-; GFX11-CU-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-CU-NEXT:    s_endpgm
     ptr %out, i32 %in, i32 %old) {
 entry:
   %gep = getelementptr i32, ptr %out, i32 4
   %val = cmpxchg volatile ptr %gep, i32 %old, i32 %in syncscope("wavefront") acq_rel monotonic
   %val0 = extractvalue { i32, i1 } %val, 0
   store i32 %val0, ptr %out, align 4
   ret void

 ; GFX11-WGP:       ; %bb.0: ; %entry
 ; GFX11-WGP-NEXT:    s_load_b128 s[0:3], s[0:1], 0x0
 ; GFX11-WGP-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-WGP-NEXT:    v_dual_mov_b32 v0, s0 :: v_dual_mov_b32 v1, s1
 ; GFX11-WGP-NEXT:    v_dual_mov_b32 v2, s2 :: v_dual_mov_b32 v3, s3
 ; GFX11-WGP-NEXT:    flat_atomic_cmpswap_b32 v2, v[0:1], v[2:3] offset:16 glc
 ; GFX11-WGP-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX11-WGP-NEXT:    flat_store_b32 v[0:1], v2
-; GFX11-WGP-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-WGP-NEXT:    s_endpgm
 ;
 ; GFX11-CU-LABEL: flat_wavefront_seq_cst_monotonic_ret_cmpxchg:
 ; GFX11-CU:       ; %bb.0: ; %entry
 ; GFX11-CU-NEXT:    s_load_b128 s[0:3], s[0:1], 0x0
 ; GFX11-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v0, s0 :: v_dual_mov_b32 v1, s1
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v2, s2 :: v_dual_mov_b32 v3, s3
 ; GFX11-CU-NEXT:    flat_atomic_cmpswap_b32 v2, v[0:1], v[2:3] offset:16 glc
 ; GFX11-CU-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX11-CU-NEXT:    flat_store_b32 v[0:1], v2
-; GFX11-CU-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-CU-NEXT:    s_endpgm
     ptr %out, i32 %in, i32 %old) {
 entry:
   %gep = getelementptr i32, ptr %out, i32 4
   %val = cmpxchg volatile ptr %gep, i32 %old, i32 %in syncscope("wavefront") seq_cst monotonic
   %val0 = extractvalue { i32, i1 } %val, 0
   store i32 %val0, ptr %out, align 4
   ret void

 ; GFX11-WGP:       ; %bb.0: ; %entry
 ; GFX11-WGP-NEXT:    s_load_b128 s[0:3], s[0:1], 0x0
 ; GFX11-WGP-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-WGP-NEXT:    v_dual_mov_b32 v0, s0 :: v_dual_mov_b32 v1, s1
 ; GFX11-WGP-NEXT:    v_dual_mov_b32 v2, s2 :: v_dual_mov_b32 v3, s3
 ; GFX11-WGP-NEXT:    flat_atomic_cmpswap_b32 v2, v[0:1], v[2:3] offset:16 glc
 ; GFX11-WGP-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX11-WGP-NEXT:    flat_store_b32 v[0:1], v2
-; GFX11-WGP-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-WGP-NEXT:    s_endpgm
 ;
 ; GFX11-CU-LABEL: flat_wavefront_monotonic_acquire_ret_cmpxchg:
 ; GFX11-CU:       ; %bb.0: ; %entry
 ; GFX11-CU-NEXT:    s_load_b128 s[0:3], s[0:1], 0x0
 ; GFX11-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v0, s0 :: v_dual_mov_b32 v1, s1
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v2, s2 :: v_dual_mov_b32 v3, s3
 ; GFX11-CU-NEXT:    flat_atomic_cmpswap_b32 v2, v[0:1], v[2:3] offset:16 glc
 ; GFX11-CU-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX11-CU-NEXT:    flat_store_b32 v[0:1], v2
-; GFX11-CU-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-CU-NEXT:    s_endpgm
     ptr %out, i32 %in, i32 %old) {
 entry:
   %gep = getelementptr i32, ptr %out, i32 4
   %val = cmpxchg volatile ptr %gep, i32 %old, i32 %in syncscope("wavefront") monotonic acquire
   %val0 = extractvalue { i32, i1 } %val, 0
   store i32 %val0, ptr %out, align 4
   ret void

 ; GFX11-WGP:       ; %bb.0: ; %entry
 ; GFX11-WGP-NEXT:    s_load_b128 s[0:3], s[0:1], 0x0
 ; GFX11-WGP-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-WGP-NEXT:    v_dual_mov_b32 v0, s0 :: v_dual_mov_b32 v1, s1
 ; GFX11-WGP-NEXT:    v_dual_mov_b32 v2, s2 :: v_dual_mov_b32 v3, s3
 ; GFX11-WGP-NEXT:    flat_atomic_cmpswap_b32 v2, v[0:1], v[2:3] offset:16 glc
 ; GFX11-WGP-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX11-WGP-NEXT:    flat_store_b32 v[0:1], v2
-; GFX11-WGP-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-WGP-NEXT:    s_endpgm
 ;
 ; GFX11-CU-LABEL: flat_wavefront_acquire_acquire_ret_cmpxchg:
 ; GFX11-CU:       ; %bb.0: ; %entry
 ; GFX11-CU-NEXT:    s_load_b128 s[0:3], s[0:1], 0x0
 ; GFX11-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v0, s0 :: v_dual_mov_b32 v1, s1
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v2, s2 :: v_dual_mov_b32 v3, s3
 ; GFX11-CU-NEXT:    flat_atomic_cmpswap_b32 v2, v[0:1], v[2:3] offset:16 glc
 ; GFX11-CU-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX11-CU-NEXT:    flat_store_b32 v[0:1], v2
-; GFX11-CU-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-CU-NEXT:    s_endpgm
     ptr %out, i32 %in, i32 %old) {
 entry:
   %gep = getelementptr i32, ptr %out, i32 4
   %val = cmpxchg volatile ptr %gep, i32 %old, i32 %in syncscope("wavefront") acquire acquire
   %val0 = extractvalue { i32, i1 } %val, 0
   store i32 %val0, ptr %out, align 4
   ret void

 ; GFX11-WGP:       ; %bb.0: ; %entry
 ; GFX11-WGP-NEXT:    s_load_b128 s[0:3], s[0:1], 0x0
 ; GFX11-WGP-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-WGP-NEXT:    v_dual_mov_b32 v0, s0 :: v_dual_mov_b32 v1, s1
 ; GFX11-WGP-NEXT:    v_dual_mov_b32 v2, s2 :: v_dual_mov_b32 v3, s3
 ; GFX11-WGP-NEXT:    flat_atomic_cmpswap_b32 v2, v[0:1], v[2:3] offset:16 glc
 ; GFX11-WGP-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX11-WGP-NEXT:    flat_store_b32 v[0:1], v2
-; GFX11-WGP-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-WGP-NEXT:    s_endpgm
 ;
 ; GFX11-CU-LABEL: flat_wavefront_release_acquire_ret_cmpxchg:
 ; GFX11-CU:       ; %bb.0: ; %entry
 ; GFX11-CU-NEXT:    s_load_b128 s[0:3], s[0:1], 0x0
 ; GFX11-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v0, s0 :: v_dual_mov_b32 v1, s1
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v2, s2 :: v_dual_mov_b32 v3, s3
 ; GFX11-CU-NEXT:    flat_atomic_cmpswap_b32 v2, v[0:1], v[2:3] offset:16 glc
 ; GFX11-CU-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX11-CU-NEXT:    flat_store_b32 v[0:1], v2
-; GFX11-CU-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-CU-NEXT:    s_endpgm
     ptr %out, i32 %in, i32 %old) {
 entry:
   %gep = getelementptr i32, ptr %out, i32 4
   %val = cmpxchg volatile ptr %gep, i32 %old, i32 %in syncscope("wavefront") release acquire
   %val0 = extractvalue { i32, i1 } %val, 0
   store i32 %val0, ptr %out, align 4
   ret void

 ; GFX11-WGP:       ; %bb.0: ; %entry
 ; GFX11-WGP-NEXT:    s_load_b128 s[0:3], s[0:1], 0x0
 ; GFX11-WGP-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-WGP-NEXT:    v_dual_mov_b32 v0, s0 :: v_dual_mov_b32 v1, s1
 ; GFX11-WGP-NEXT:    v_dual_mov_b32 v2, s2 :: v_dual_mov_b32 v3, s3
 ; GFX11-WGP-NEXT:    flat_atomic_cmpswap_b32 v2, v[0:1], v[2:3] offset:16 glc
 ; GFX11-WGP-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX11-WGP-NEXT:    flat_store_b32 v[0:1], v2
-; GFX11-WGP-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-WGP-NEXT:    s_endpgm
 ;
 ; GFX11-CU-LABEL: flat_wavefront_acq_rel_acquire_ret_cmpxchg:
 ; GFX11-CU:       ; %bb.0: ; %entry
 ; GFX11-CU-NEXT:    s_load_b128 s[0:3], s[0:1], 0x0
 ; GFX11-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v0, s0 :: v_dual_mov_b32 v1, s1
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v2, s2 :: v_dual_mov_b32 v3, s3
 ; GFX11-CU-NEXT:    flat_atomic_cmpswap_b32 v2, v[0:1], v[2:3] offset:16 glc
 ; GFX11-CU-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX11-CU-NEXT:    flat_store_b32 v[0:1], v2
-; GFX11-CU-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-CU-NEXT:    s_endpgm
     ptr %out, i32 %in, i32 %old) {
 entry:
   %gep = getelementptr i32, ptr %out, i32 4
   %val = cmpxchg volatile ptr %gep, i32 %old, i32 %in syncscope("wavefront") acq_rel acquire
   %val0 = extractvalue { i32, i1 } %val, 0
   store i32 %val0, ptr %out, align 4
   ret void

 ; GFX11-WGP:       ; %bb.0: ; %entry
 ; GFX11-WGP-NEXT:    s_load_b128 s[0:3], s[0:1], 0x0
 ; GFX11-WGP-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-WGP-NEXT:    v_dual_mov_b32 v0, s0 :: v_dual_mov_b32 v1, s1
 ; GFX11-WGP-NEXT:    v_dual_mov_b32 v2, s2 :: v_dual_mov_b32 v3, s3
 ; GFX11-WGP-NEXT:    flat_atomic_cmpswap_b32 v2, v[0:1], v[2:3] offset:16 glc
 ; GFX11-WGP-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX11-WGP-NEXT:    flat_store_b32 v[0:1], v2
-; GFX11-WGP-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-WGP-NEXT:    s_endpgm
 ;
 ; GFX11-CU-LABEL: flat_wavefront_seq_cst_acquire_ret_cmpxchg:
 ; GFX11-CU:       ; %bb.0: ; %entry
 ; GFX11-CU-NEXT:    s_load_b128 s[0:3], s[0:1], 0x0
 ; GFX11-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v0, s0 :: v_dual_mov_b32 v1, s1
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v2, s2 :: v_dual_mov_b32 v3, s3
 ; GFX11-CU-NEXT:    flat_atomic_cmpswap_b32 v2, v[0:1], v[2:3] offset:16 glc
 ; GFX11-CU-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX11-CU-NEXT:    flat_store_b32 v[0:1], v2
-; GFX11-CU-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-CU-NEXT:    s_endpgm
     ptr %out, i32 %in, i32 %old) {
 entry:
   %gep = getelementptr i32, ptr %out, i32 4
   %val = cmpxchg volatile ptr %gep, i32 %old, i32 %in syncscope("wavefront") seq_cst acquire
   %val0 = extractvalue { i32, i1 } %val, 0
   store i32 %val0, ptr %out, align 4
   ret void

 ; GFX11-WGP:       ; %bb.0: ; %entry
 ; GFX11-WGP-NEXT:    s_load_b128 s[0:3], s[0:1], 0x0
 ; GFX11-WGP-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-WGP-NEXT:    v_dual_mov_b32 v0, s0 :: v_dual_mov_b32 v1, s1
 ; GFX11-WGP-NEXT:    v_dual_mov_b32 v2, s2 :: v_dual_mov_b32 v3, s3
 ; GFX11-WGP-NEXT:    flat_atomic_cmpswap_b32 v2, v[0:1], v[2:3] offset:16 glc
 ; GFX11-WGP-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX11-WGP-NEXT:    flat_store_b32 v[0:1], v2
-; GFX11-WGP-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-WGP-NEXT:    s_endpgm
 ;
 ; GFX11-CU-LABEL: flat_wavefront_monotonic_seq_cst_ret_cmpxchg:
 ; GFX11-CU:       ; %bb.0: ; %entry
 ; GFX11-CU-NEXT:    s_load_b128 s[0:3], s[0:1], 0x0
 ; GFX11-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v0, s0 :: v_dual_mov_b32 v1, s1
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v2, s2 :: v_dual_mov_b32 v3, s3
 ; GFX11-CU-NEXT:    flat_atomic_cmpswap_b32 v2, v[0:1], v[2:3] offset:16 glc
 ; GFX11-CU-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX11-CU-NEXT:    flat_store_b32 v[0:1], v2
-; GFX11-CU-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-CU-NEXT:    s_endpgm
     ptr %out, i32 %in, i32 %old) {
 entry:
   %gep = getelementptr i32, ptr %out, i32 4
   %val = cmpxchg volatile ptr %gep, i32 %old, i32 %in syncscope("wavefront") monotonic seq_cst
   %val0 = extractvalue { i32, i1 } %val, 0
   store i32 %val0, ptr %out, align 4
   ret void

 ; GFX11-WGP:       ; %bb.0: ; %entry
 ; GFX11-WGP-NEXT:    s_load_b128 s[0:3], s[0:1], 0x0
 ; GFX11-WGP-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-WGP-NEXT:    v_dual_mov_b32 v0, s0 :: v_dual_mov_b32 v1, s1
 ; GFX11-WGP-NEXT:    v_dual_mov_b32 v2, s2 :: v_dual_mov_b32 v3, s3
 ; GFX11-WGP-NEXT:    flat_atomic_cmpswap_b32 v2, v[0:1], v[2:3] offset:16 glc
 ; GFX11-WGP-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX11-WGP-NEXT:    flat_store_b32 v[0:1], v2
-; GFX11-WGP-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-WGP-NEXT:    s_endpgm
 ;
 ; GFX11-CU-LABEL: flat_wavefront_acquire_seq_cst_ret_cmpxchg:
 ; GFX11-CU:       ; %bb.0: ; %entry
 ; GFX11-CU-NEXT:    s_load_b128 s[0:3], s[0:1], 0x0
 ; GFX11-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v0, s0 :: v_dual_mov_b32 v1, s1
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v2, s2 :: v_dual_mov_b32 v3, s3
 ; GFX11-CU-NEXT:    flat_atomic_cmpswap_b32 v2, v[0:1], v[2:3] offset:16 glc
 ; GFX11-CU-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX11-CU-NEXT:    flat_store_b32 v[0:1], v2
-; GFX11-CU-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-CU-NEXT:    s_endpgm
     ptr %out, i32 %in, i32 %old) {
 entry:
   %gep = getelementptr i32, ptr %out, i32 4
   %val = cmpxchg volatile ptr %gep, i32 %old, i32 %in syncscope("wavefront") acquire seq_cst
   %val0 = extractvalue { i32, i1 } %val, 0
   store i32 %val0, ptr %out, align 4
   ret void

 ; GFX11-WGP:       ; %bb.0: ; %entry
 ; GFX11-WGP-NEXT:    s_load_b128 s[0:3], s[0:1], 0x0
 ; GFX11-WGP-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-WGP-NEXT:    v_dual_mov_b32 v0, s0 :: v_dual_mov_b32 v1, s1
 ; GFX11-WGP-NEXT:    v_dual_mov_b32 v2, s2 :: v_dual_mov_b32 v3, s3
 ; GFX11-WGP-NEXT:    flat_atomic_cmpswap_b32 v2, v[0:1], v[2:3] offset:16 glc
 ; GFX11-WGP-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX11-WGP-NEXT:    flat_store_b32 v[0:1], v2
-; GFX11-WGP-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-WGP-NEXT:    s_endpgm
 ;
 ; GFX11-CU-LABEL: flat_wavefront_release_seq_cst_ret_cmpxchg:
 ; GFX11-CU:       ; %bb.0: ; %entry
 ; GFX11-CU-NEXT:    s_load_b128 s[0:3], s[0:1], 0x0
 ; GFX11-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v0, s0 :: v_dual_mov_b32 v1, s1
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v2, s2 :: v_dual_mov_b32 v3, s3
 ; GFX11-CU-NEXT:    flat_atomic_cmpswap_b32 v2, v[0:1], v[2:3] offset:16 glc
 ; GFX11-CU-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX11-CU-NEXT:    flat_store_b32 v[0:1], v2
-; GFX11-CU-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-CU-NEXT:    s_endpgm
     ptr %out, i32 %in, i32 %old) {
 entry:
   %gep = getelementptr i32, ptr %out, i32 4
   %val = cmpxchg volatile ptr %gep, i32 %old, i32 %in syncscope("wavefront") release seq_cst
   %val0 = extractvalue { i32, i1 } %val, 0
   store i32 %val0, ptr %out, align 4
   ret void

 ; GFX11-WGP:       ; %bb.0: ; %entry
 ; GFX11-WGP-NEXT:    s_load_b128 s[0:3], s[0:1], 0x0
 ; GFX11-WGP-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-WGP-NEXT:    v_dual_mov_b32 v0, s0 :: v_dual_mov_b32 v1, s1
 ; GFX11-WGP-NEXT:    v_dual_mov_b32 v2, s2 :: v_dual_mov_b32 v3, s3
 ; GFX11-WGP-NEXT:    flat_atomic_cmpswap_b32 v2, v[0:1], v[2:3] offset:16 glc
 ; GFX11-WGP-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX11-WGP-NEXT:    flat_store_b32 v[0:1], v2
-; GFX11-WGP-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-WGP-NEXT:    s_endpgm
 ;
 ; GFX11-CU-LABEL: flat_wavefront_acq_rel_seq_cst_ret_cmpxchg:
 ; GFX11-CU:       ; %bb.0: ; %entry
 ; GFX11-CU-NEXT:    s_load_b128 s[0:3], s[0:1], 0x0
 ; GFX11-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v0, s0 :: v_dual_mov_b32 v1, s1
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v2, s2 :: v_dual_mov_b32 v3, s3
 ; GFX11-CU-NEXT:    flat_atomic_cmpswap_b32 v2, v[0:1], v[2:3] offset:16 glc
 ; GFX11-CU-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX11-CU-NEXT:    flat_store_b32 v[0:1], v2
-; GFX11-CU-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-CU-NEXT:    s_endpgm
     ptr %out, i32 %in, i32 %old) {
 entry:
   %gep = getelementptr i32, ptr %out, i32 4
   %val = cmpxchg volatile ptr %gep, i32 %old, i32 %in syncscope("wavefront") acq_rel seq_cst
   %val0 = extractvalue { i32, i1 } %val, 0
   store i32 %val0, ptr %out, align 4
   ret void

 ; GFX11-WGP:       ; %bb.0: ; %entry
 ; GFX11-WGP-NEXT:    s_load_b128 s[0:3], s[0:1], 0x0
 ; GFX11-WGP-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-WGP-NEXT:    v_dual_mov_b32 v0, s0 :: v_dual_mov_b32 v1, s1
 ; GFX11-WGP-NEXT:    v_dual_mov_b32 v2, s2 :: v_dual_mov_b32 v3, s3
 ; GFX11-WGP-NEXT:    flat_atomic_cmpswap_b32 v2, v[0:1], v[2:3] offset:16 glc
 ; GFX11-WGP-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX11-WGP-NEXT:    flat_store_b32 v[0:1], v2
-; GFX11-WGP-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-WGP-NEXT:    s_endpgm
 ;
 ; GFX11-CU-LABEL: flat_wavefront_seq_cst_seq_cst_ret_cmpxchg:
 ; GFX11-CU:       ; %bb.0: ; %entry
 ; GFX11-CU-NEXT:    s_load_b128 s[0:3], s[0:1], 0x0
 ; GFX11-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v0, s0 :: v_dual_mov_b32 v1, s1
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v2, s2 :: v_dual_mov_b32 v3, s3
 ; GFX11-CU-NEXT:    flat_atomic_cmpswap_b32 v2, v[0:1], v[2:3] offset:16 glc
 ; GFX11-CU-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX11-CU-NEXT:    flat_store_b32 v[0:1], v2
-; GFX11-CU-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-CU-NEXT:    s_endpgm
     ptr %out, i32 %in, i32 %old) {
 entry:
   %gep = getelementptr i32, ptr %out, i32 4
   %val = cmpxchg volatile ptr %gep, i32 %old, i32 %in syncscope("wavefront") seq_cst seq_cst
   %val0 = extractvalue { i32, i1 } %val, 0
   store i32 %val0, ptr %out, align 4
   ret void

 ; GFX11-WGP:       ; %bb.0: ; %entry
 ; GFX11-WGP-NEXT:    s_load_b128 s[0:3], s[0:1], 0x0
 ; GFX11-WGP-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-WGP-NEXT:    v_dual_mov_b32 v0, s0 :: v_dual_mov_b32 v1, s1
 ; GFX11-WGP-NEXT:    flat_load_b32 v2, v[0:1]
 ; GFX11-WGP-NEXT:    v_dual_mov_b32 v0, s2 :: v_dual_mov_b32 v1, s3
 ; GFX11-WGP-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX11-WGP-NEXT:    flat_store_b32 v[0:1], v2
-; GFX11-WGP-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-WGP-NEXT:    s_endpgm
 ;
 ; GFX11-CU-LABEL: flat_wavefront_one_as_unordered_load:
 ; GFX11-CU:       ; %bb.0: ; %entry
 ; GFX11-CU-NEXT:    s_load_b128 s[0:3], s[0:1], 0x0
 ; GFX11-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v0, s0 :: v_dual_mov_b32 v1, s1
 ; GFX11-CU-NEXT:    flat_load_b32 v2, v[0:1]
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v0, s2 :: v_dual_mov_b32 v1, s3
 ; GFX11-CU-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX11-CU-NEXT:    flat_store_b32 v[0:1], v2
-; GFX11-CU-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-CU-NEXT:    s_endpgm
     ptr %in, ptr %out) {
 entry:
   %val = load atomic i32, ptr %in syncscope("wavefront-one-as") unordered, align 4
   store i32 %val, ptr %out
   ret void
 }
 

 ; GFX11-WGP:       ; %bb.0: ; %entry
 ; GFX11-WGP-NEXT:    s_load_b128 s[0:3], s[0:1], 0x0
 ; GFX11-WGP-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-WGP-NEXT:    v_dual_mov_b32 v0, s0 :: v_dual_mov_b32 v1, s1
 ; GFX11-WGP-NEXT:    flat_load_b32 v2, v[0:1]
 ; GFX11-WGP-NEXT:    v_dual_mov_b32 v0, s2 :: v_dual_mov_b32 v1, s3
 ; GFX11-WGP-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX11-WGP-NEXT:    flat_store_b32 v[0:1], v2
-; GFX11-WGP-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-WGP-NEXT:    s_endpgm
 ;
 ; GFX11-CU-LABEL: flat_wavefront_one_as_monotonic_load:
 ; GFX11-CU:       ; %bb.0: ; %entry
 ; GFX11-CU-NEXT:    s_load_b128 s[0:3], s[0:1], 0x0
 ; GFX11-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v0, s0 :: v_dual_mov_b32 v1, s1
 ; GFX11-CU-NEXT:    flat_load_b32 v2, v[0:1]
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v0, s2 :: v_dual_mov_b32 v1, s3
 ; GFX11-CU-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX11-CU-NEXT:    flat_store_b32 v[0:1], v2
-; GFX11-CU-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-CU-NEXT:    s_endpgm
     ptr %in, ptr %out) {
 entry:
   %val = load atomic i32, ptr %in syncscope("wavefront-one-as") monotonic, align 4
   store i32 %val, ptr %out
   ret void
 }
 

 ; GFX11-WGP:       ; %bb.0: ; %entry
 ; GFX11-WGP-NEXT:    s_load_b128 s[0:3], s[0:1], 0x0
 ; GFX11-WGP-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-WGP-NEXT:    v_dual_mov_b32 v0, s0 :: v_dual_mov_b32 v1, s1
 ; GFX11-WGP-NEXT:    flat_load_b32 v2, v[0:1]
 ; GFX11-WGP-NEXT:    v_dual_mov_b32 v0, s2 :: v_dual_mov_b32 v1, s3
 ; GFX11-WGP-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX11-WGP-NEXT:    flat_store_b32 v[0:1], v2
-; GFX11-WGP-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-WGP-NEXT:    s_endpgm
 ;
 ; GFX11-CU-LABEL: flat_wavefront_one_as_acquire_load:
 ; GFX11-CU:       ; %bb.0: ; %entry
 ; GFX11-CU-NEXT:    s_load_b128 s[0:3], s[0:1], 0x0
 ; GFX11-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v0, s0 :: v_dual_mov_b32 v1, s1
 ; GFX11-CU-NEXT:    flat_load_b32 v2, v[0:1]
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v0, s2 :: v_dual_mov_b32 v1, s3
 ; GFX11-CU-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX11-CU-NEXT:    flat_store_b32 v[0:1], v2
-; GFX11-CU-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-CU-NEXT:    s_endpgm
     ptr %in, ptr %out) {
 entry:
   %val = load atomic i32, ptr %in syncscope("wavefront-one-as") acquire, align 4
   store i32 %val, ptr %out
   ret void
 }
 

 ; GFX11-WGP:       ; %bb.0: ; %entry
 ; GFX11-WGP-NEXT:    s_load_b128 s[0:3], s[0:1], 0x0
 ; GFX11-WGP-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-WGP-NEXT:    v_dual_mov_b32 v0, s0 :: v_dual_mov_b32 v1, s1
 ; GFX11-WGP-NEXT:    flat_load_b32 v2, v[0:1]
 ; GFX11-WGP-NEXT:    v_dual_mov_b32 v0, s2 :: v_dual_mov_b32 v1, s3
 ; GFX11-WGP-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX11-WGP-NEXT:    flat_store_b32 v[0:1], v2
-; GFX11-WGP-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-WGP-NEXT:    s_endpgm
 ;
 ; GFX11-CU-LABEL: flat_wavefront_one_as_seq_cst_load:
 ; GFX11-CU:       ; %bb.0: ; %entry
 ; GFX11-CU-NEXT:    s_load_b128 s[0:3], s[0:1], 0x0
 ; GFX11-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v0, s0 :: v_dual_mov_b32 v1, s1
 ; GFX11-CU-NEXT:    flat_load_b32 v2, v[0:1]
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v0, s2 :: v_dual_mov_b32 v1, s3
 ; GFX11-CU-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX11-CU-NEXT:    flat_store_b32 v[0:1], v2
-; GFX11-CU-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-CU-NEXT:    s_endpgm
     ptr %in, ptr %out) {
 entry:
   %val = load atomic i32, ptr %in syncscope("wavefront-one-as") seq_cst, align 4
   store i32 %val, ptr %out
   ret void
 }
 

 ; GFX11-WGP:       ; %bb.0: ; %entry
 ; GFX11-WGP-NEXT:    s_clause 0x1
 ; GFX11-WGP-NEXT:    s_load_b64 s[2:3], s[0:1], 0x8
 ; GFX11-WGP-NEXT:    s_load_b32 s0, s[0:1], 0x0
 ; GFX11-WGP-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-WGP-NEXT:    v_dual_mov_b32 v0, s2 :: v_dual_mov_b32 v1, s3
 ; GFX11-WGP-NEXT:    v_mov_b32_e32 v2, s0
 ; GFX11-WGP-NEXT:    flat_store_b32 v[0:1], v2
-; GFX11-WGP-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-WGP-NEXT:    s_endpgm
 ;
 ; GFX11-CU-LABEL: flat_wavefront_one_as_unordered_store:
 ; GFX11-CU:       ; %bb.0: ; %entry
 ; GFX11-CU-NEXT:    s_clause 0x1
 ; GFX11-CU-NEXT:    s_load_b64 s[2:3], s[0:1], 0x8
 ; GFX11-CU-NEXT:    s_load_b32 s0, s[0:1], 0x0
 ; GFX11-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v0, s2 :: v_dual_mov_b32 v1, s3
 ; GFX11-CU-NEXT:    v_mov_b32_e32 v2, s0
 ; GFX11-CU-NEXT:    flat_store_b32 v[0:1], v2
-; GFX11-CU-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-CU-NEXT:    s_endpgm
     i32 %in, ptr %out) {
 entry:
   store atomic i32 %in, ptr %out syncscope("wavefront-one-as") unordered, align 4
   ret void
 }
 
 define amdgpu_kernel void @flat_wavefront_one_as_monotonic_store(

 ; GFX11-WGP:       ; %bb.0: ; %entry
 ; GFX11-WGP-NEXT:    s_clause 0x1
 ; GFX11-WGP-NEXT:    s_load_b64 s[2:3], s[0:1], 0x8
 ; GFX11-WGP-NEXT:    s_load_b32 s0, s[0:1], 0x0
 ; GFX11-WGP-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-WGP-NEXT:    v_dual_mov_b32 v0, s2 :: v_dual_mov_b32 v1, s3
 ; GFX11-WGP-NEXT:    v_mov_b32_e32 v2, s0
 ; GFX11-WGP-NEXT:    flat_store_b32 v[0:1], v2
-; GFX11-WGP-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-WGP-NEXT:    s_endpgm
 ;
 ; GFX11-CU-LABEL: flat_wavefront_one_as_monotonic_store:
 ; GFX11-CU:       ; %bb.0: ; %entry
 ; GFX11-CU-NEXT:    s_clause 0x1
 ; GFX11-CU-NEXT:    s_load_b64 s[2:3], s[0:1], 0x8
 ; GFX11-CU-NEXT:    s_load_b32 s0, s[0:1], 0x0
 ; GFX11-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v0, s2 :: v_dual_mov_b32 v1, s3
 ; GFX11-CU-NEXT:    v_mov_b32_e32 v2, s0
 ; GFX11-CU-NEXT:    flat_store_b32 v[0:1], v2
-; GFX11-CU-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-CU-NEXT:    s_endpgm
     i32 %in, ptr %out) {
 entry:
   store atomic i32 %in, ptr %out syncscope("wavefront-one-as") monotonic, align 4
   ret void
 }
 
 define amdgpu_kernel void @flat_wavefront_one_as_release_store(

 ; GFX11-WGP:       ; %bb.0: ; %entry
 ; GFX11-WGP-NEXT:    s_clause 0x1
 ; GFX11-WGP-NEXT:    s_load_b64 s[2:3], s[0:1], 0x8
 ; GFX11-WGP-NEXT:    s_load_b32 s0, s[0:1], 0x0
 ; GFX11-WGP-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-WGP-NEXT:    v_dual_mov_b32 v0, s2 :: v_dual_mov_b32 v1, s3
 ; GFX11-WGP-NEXT:    v_mov_b32_e32 v2, s0
 ; GFX11-WGP-NEXT:    flat_store_b32 v[0:1], v2
-; GFX11-WGP-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-WGP-NEXT:    s_endpgm
 ;
 ; GFX11-CU-LABEL: flat_wavefront_one_as_release_store:
 ; GFX11-CU:       ; %bb.0: ; %entry
 ; GFX11-CU-NEXT:    s_clause 0x1
 ; GFX11-CU-NEXT:    s_load_b64 s[2:3], s[0:1], 0x8
 ; GFX11-CU-NEXT:    s_load_b32 s0, s[0:1], 0x0
 ; GFX11-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v0, s2 :: v_dual_mov_b32 v1, s3
 ; GFX11-CU-NEXT:    v_mov_b32_e32 v2, s0
 ; GFX11-CU-NEXT:    flat_store_b32 v[0:1], v2
-; GFX11-CU-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-CU-NEXT:    s_endpgm
     i32 %in, ptr %out) {
 entry:
   store atomic i32 %in, ptr %out syncscope("wavefront-one-as") release, align 4
   ret void
 }
 
 define amdgpu_kernel void @flat_wavefront_one_as_seq_cst_store(

 ; GFX11-WGP:       ; %bb.0: ; %entry
 ; GFX11-WGP-NEXT:    s_clause 0x1
 ; GFX11-WGP-NEXT:    s_load_b64 s[2:3], s[0:1], 0x8
 ; GFX11-WGP-NEXT:    s_load_b32 s0, s[0:1], 0x0
 ; GFX11-WGP-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-WGP-NEXT:    v_dual_mov_b32 v0, s2 :: v_dual_mov_b32 v1, s3
 ; GFX11-WGP-NEXT:    v_mov_b32_e32 v2, s0
 ; GFX11-WGP-NEXT:    flat_store_b32 v[0:1], v2
-; GFX11-WGP-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-WGP-NEXT:    s_endpgm
 ;
 ; GFX11-CU-LABEL: flat_wavefront_one_as_seq_cst_store:
 ; GFX11-CU:       ; %bb.0: ; %entry
 ; GFX11-CU-NEXT:    s_clause 0x1
 ; GFX11-CU-NEXT:    s_load_b64 s[2:3], s[0:1], 0x8
 ; GFX11-CU-NEXT:    s_load_b32 s0, s[0:1], 0x0
 ; GFX11-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v0, s2 :: v_dual_mov_b32 v1, s3
 ; GFX11-CU-NEXT:    v_mov_b32_e32 v2, s0
 ; GFX11-CU-NEXT:    flat_store_b32 v[0:1], v2
-; GFX11-CU-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-CU-NEXT:    s_endpgm
     i32 %in, ptr %out) {
 entry:
   store atomic i32 %in, ptr %out syncscope("wavefront-one-as") seq_cst, align 4
   ret void
 }
 
 define amdgpu_kernel void @flat_wavefront_one_as_monotonic_atomicrmw(

 ; GFX11-WGP:       ; %bb.0: ; %entry
 ; GFX11-WGP-NEXT:    s_clause 0x1
 ; GFX11-WGP-NEXT:    s_load_b64 s[2:3], s[0:1], 0x0
 ; GFX11-WGP-NEXT:    s_load_b32 s0, s[0:1], 0x8
 ; GFX11-WGP-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-WGP-NEXT:    v_dual_mov_b32 v0, s2 :: v_dual_mov_b32 v1, s3
 ; GFX11-WGP-NEXT:    v_mov_b32_e32 v2, s0
 ; GFX11-WGP-NEXT:    flat_atomic_swap_b32 v[0:1], v2
-; GFX11-WGP-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-WGP-NEXT:    s_endpgm
 ;
 ; GFX11-CU-LABEL: flat_wavefront_one_as_monotonic_atomicrmw:
 ; GFX11-CU:       ; %bb.0: ; %entry
 ; GFX11-CU-NEXT:    s_clause 0x1
 ; GFX11-CU-NEXT:    s_load_b64 s[2:3], s[0:1], 0x0
 ; GFX11-CU-NEXT:    s_load_b32 s0, s[0:1], 0x8
 ; GFX11-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v0, s2 :: v_dual_mov_b32 v1, s3
 ; GFX11-CU-NEXT:    v_mov_b32_e32 v2, s0
 ; GFX11-CU-NEXT:    flat_atomic_swap_b32 v[0:1], v2
-; GFX11-CU-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-CU-NEXT:    s_endpgm
     ptr %out, i32 %in) {
 entry:
   %val = atomicrmw volatile xchg ptr %out, i32 %in syncscope("wavefront-one-as") monotonic
   ret void
 }
 
 define amdgpu_kernel void @flat_wavefront_one_as_acquire_atomicrmw(

 ; GFX11-WGP:       ; %bb.0: ; %entry
 ; GFX11-WGP-NEXT:    s_clause 0x1
 ; GFX11-WGP-NEXT:    s_load_b64 s[2:3], s[0:1], 0x0
 ; GFX11-WGP-NEXT:    s_load_b32 s0, s[0:1], 0x8
 ; GFX11-WGP-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-WGP-NEXT:    v_dual_mov_b32 v0, s2 :: v_dual_mov_b32 v1, s3
 ; GFX11-WGP-NEXT:    v_mov_b32_e32 v2, s0
 ; GFX11-WGP-NEXT:    flat_atomic_swap_b32 v[0:1], v2
-; GFX11-WGP-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-WGP-NEXT:    s_endpgm
 ;
 ; GFX11-CU-LABEL: flat_wavefront_one_as_acquire_atomicrmw:
 ; GFX11-CU:       ; %bb.0: ; %entry
 ; GFX11-CU-NEXT:    s_clause 0x1
 ; GFX11-CU-NEXT:    s_load_b64 s[2:3], s[0:1], 0x0
 ; GFX11-CU-NEXT:    s_load_b32 s0, s[0:1], 0x8
 ; GFX11-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v0, s2 :: v_dual_mov_b32 v1, s3
 ; GFX11-CU-NEXT:    v_mov_b32_e32 v2, s0
 ; GFX11-CU-NEXT:    flat_atomic_swap_b32 v[0:1], v2
-; GFX11-CU-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-CU-NEXT:    s_endpgm
     ptr %out, i32 %in) {
 entry:
   %val = atomicrmw volatile xchg ptr %out, i32 %in syncscope("wavefront-one-as") acquire
   ret void
 }
 
 define amdgpu_kernel void @flat_wavefront_one_as_release_atomicrmw(

 ; GFX11-WGP:       ; %bb.0: ; %entry
 ; GFX11-WGP-NEXT:    s_clause 0x1
 ; GFX11-WGP-NEXT:    s_load_b64 s[2:3], s[0:1], 0x0
 ; GFX11-WGP-NEXT:    s_load_b32 s0, s[0:1], 0x8
 ; GFX11-WGP-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-WGP-NEXT:    v_dual_mov_b32 v0, s2 :: v_dual_mov_b32 v1, s3
 ; GFX11-WGP-NEXT:    v_mov_b32_e32 v2, s0
 ; GFX11-WGP-NEXT:    flat_atomic_swap_b32 v[0:1], v2
-; GFX11-WGP-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-WGP-NEXT:    s_endpgm
 ;
 ; GFX11-CU-LABEL: flat_wavefront_one_as_release_atomicrmw:
 ; GFX11-CU:       ; %bb.0: ; %entry
 ; GFX11-CU-NEXT:    s_clause 0x1
 ; GFX11-CU-NEXT:    s_load_b64 s[2:3], s[0:1], 0x0
 ; GFX11-CU-NEXT:    s_load_b32 s0, s[0:1], 0x8
 ; GFX11-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v0, s2 :: v_dual_mov_b32 v1, s3
 ; GFX11-CU-NEXT:    v_mov_b32_e32 v2, s0
 ; GFX11-CU-NEXT:    flat_atomic_swap_b32 v[0:1], v2
-; GFX11-CU-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-CU-NEXT:    s_endpgm
     ptr %out, i32 %in) {
 entry:
   %val = atomicrmw volatile xchg ptr %out, i32 %in syncscope("wavefront-one-as") release
   ret void
 }
 
 define amdgpu_kernel void @flat_wavefront_one_as_acq_rel_atomicrmw(

 ; GFX11-WGP:       ; %bb.0: ; %entry
 ; GFX11-WGP-NEXT:    s_clause 0x1
 ; GFX11-WGP-NEXT:    s_load_b64 s[2:3], s[0:1], 0x0
 ; GFX11-WGP-NEXT:    s_load_b32 s0, s[0:1], 0x8
 ; GFX11-WGP-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-WGP-NEXT:    v_dual_mov_b32 v0, s2 :: v_dual_mov_b32 v1, s3
 ; GFX11-WGP-NEXT:    v_mov_b32_e32 v2, s0
 ; GFX11-WGP-NEXT:    flat_atomic_swap_b32 v[0:1], v2
-; GFX11-WGP-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-WGP-NEXT:    s_endpgm
 ;
 ; GFX11-CU-LABEL: flat_wavefront_one_as_acq_rel_atomicrmw:
 ; GFX11-CU:       ; %bb.0: ; %entry
 ; GFX11-CU-NEXT:    s_clause 0x1
 ; GFX11-CU-NEXT:    s_load_b64 s[2:3], s[0:1], 0x0
 ; GFX11-CU-NEXT:    s_load_b32 s0, s[0:1], 0x8
 ; GFX11-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v0, s2 :: v_dual_mov_b32 v1, s3
 ; GFX11-CU-NEXT:    v_mov_b32_e32 v2, s0
 ; GFX11-CU-NEXT:    flat_atomic_swap_b32 v[0:1], v2
-; GFX11-CU-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-CU-NEXT:    s_endpgm
     ptr %out, i32 %in) {
 entry:
   %val = atomicrmw volatile xchg ptr %out, i32 %in syncscope("wavefront-one-as") acq_rel
   ret void
 }
 
 define amdgpu_kernel void @flat_wavefront_one_as_seq_cst_atomicrmw(

 ; GFX11-WGP:       ; %bb.0: ; %entry
 ; GFX11-WGP-NEXT:    s_clause 0x1
 ; GFX11-WGP-NEXT:    s_load_b64 s[2:3], s[0:1], 0x0
 ; GFX11-WGP-NEXT:    s_load_b32 s0, s[0:1], 0x8
 ; GFX11-WGP-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-WGP-NEXT:    v_dual_mov_b32 v0, s2 :: v_dual_mov_b32 v1, s3
 ; GFX11-WGP-NEXT:    v_mov_b32_e32 v2, s0
 ; GFX11-WGP-NEXT:    flat_atomic_swap_b32 v[0:1], v2
-; GFX11-WGP-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-WGP-NEXT:    s_endpgm
 ;
 ; GFX11-CU-LABEL: flat_wavefront_one_as_seq_cst_atomicrmw:
 ; GFX11-CU:       ; %bb.0: ; %entry
 ; GFX11-CU-NEXT:    s_clause 0x1
 ; GFX11-CU-NEXT:    s_load_b64 s[2:3], s[0:1], 0x0
 ; GFX11-CU-NEXT:    s_load_b32 s0, s[0:1], 0x8
 ; GFX11-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v0, s2 :: v_dual_mov_b32 v1, s3
 ; GFX11-CU-NEXT:    v_mov_b32_e32 v2, s0
 ; GFX11-CU-NEXT:    flat_atomic_swap_b32 v[0:1], v2
-; GFX11-CU-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-CU-NEXT:    s_endpgm
     ptr %out, i32 %in) {
 entry:
   %val = atomicrmw volatile xchg ptr %out, i32 %in syncscope("wavefront-one-as") seq_cst
   ret void
 }
 
 define amdgpu_kernel void @flat_wavefront_one_as_acquire_ret_atomicrmw(

 ; GFX11-WGP-NEXT:    s_load_b64 s[2:3], s[0:1], 0x0
 ; GFX11-WGP-NEXT:    s_load_b32 s0, s[0:1], 0x8
 ; GFX11-WGP-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-WGP-NEXT:    v_dual_mov_b32 v0, s2 :: v_dual_mov_b32 v1, s3
 ; GFX11-WGP-NEXT:    v_mov_b32_e32 v2, s0
 ; GFX11-WGP-NEXT:    flat_atomic_swap_b32 v2, v[0:1], v2 glc
 ; GFX11-WGP-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX11-WGP-NEXT:    flat_store_b32 v[0:1], v2
-; GFX11-WGP-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-WGP-NEXT:    s_endpgm
 ;
 ; GFX11-CU-LABEL: flat_wavefront_one_as_acquire_ret_atomicrmw:
 ; GFX11-CU:       ; %bb.0: ; %entry
 ; GFX11-CU-NEXT:    s_clause 0x1
 ; GFX11-CU-NEXT:    s_load_b64 s[2:3], s[0:1], 0x0
 ; GFX11-CU-NEXT:    s_load_b32 s0, s[0:1], 0x8
 ; GFX11-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v0, s2 :: v_dual_mov_b32 v1, s3
 ; GFX11-CU-NEXT:    v_mov_b32_e32 v2, s0
 ; GFX11-CU-NEXT:    flat_atomic_swap_b32 v2, v[0:1], v2 glc
 ; GFX11-CU-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX11-CU-NEXT:    flat_store_b32 v[0:1], v2
-; GFX11-CU-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-CU-NEXT:    s_endpgm
     ptr %out, i32 %in) {
 entry:
   %val = atomicrmw volatile xchg ptr %out, i32 %in syncscope("wavefront-one-as") acquire
   store i32 %val, ptr %out, align 4
   ret void
 }
 

 ; GFX11-WGP-NEXT:    s_load_b64 s[2:3], s[0:1], 0x0
 ; GFX11-WGP-NEXT:    s_load_b32 s0, s[0:1], 0x8
 ; GFX11-WGP-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-WGP-NEXT:    v_dual_mov_b32 v0, s2 :: v_dual_mov_b32 v1, s3
 ; GFX11-WGP-NEXT:    v_mov_b32_e32 v2, s0
 ; GFX11-WGP-NEXT:    flat_atomic_swap_b32 v2, v[0:1], v2 glc
 ; GFX11-WGP-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX11-WGP-NEXT:    flat_store_b32 v[0:1], v2
-; GFX11-WGP-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-WGP-NEXT:    s_endpgm
 ;
 ; GFX11-CU-LABEL: flat_wavefront_one_as_acq_rel_ret_atomicrmw:
 ; GFX11-CU:       ; %bb.0: ; %entry
 ; GFX11-CU-NEXT:    s_clause 0x1
 ; GFX11-CU-NEXT:    s_load_b64 s[2:3], s[0:1], 0x0
 ; GFX11-CU-NEXT:    s_load_b32 s0, s[0:1], 0x8
 ; GFX11-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v0, s2 :: v_dual_mov_b32 v1, s3
 ; GFX11-CU-NEXT:    v_mov_b32_e32 v2, s0
 ; GFX11-CU-NEXT:    flat_atomic_swap_b32 v2, v[0:1], v2 glc
 ; GFX11-CU-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX11-CU-NEXT:    flat_store_b32 v[0:1], v2
-; GFX11-CU-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-CU-NEXT:    s_endpgm
     ptr %out, i32 %in) {
 entry:
   %val = atomicrmw volatile xchg ptr %out, i32 %in syncscope("wavefront-one-as") acq_rel
   store i32 %val, ptr %out, align 4
   ret void
 }
 

 ; GFX11-WGP-NEXT:    s_load_b64 s[2:3], s[0:1], 0x0
 ; GFX11-WGP-NEXT:    s_load_b32 s0, s[0:1], 0x8
 ; GFX11-WGP-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-WGP-NEXT:    v_dual_mov_b32 v0, s2 :: v_dual_mov_b32 v1, s3
 ; GFX11-WGP-NEXT:    v_mov_b32_e32 v2, s0
 ; GFX11-WGP-NEXT:    flat_atomic_swap_b32 v2, v[0:1], v2 glc
 ; GFX11-WGP-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX11-WGP-NEXT:    flat_store_b32 v[0:1], v2
-; GFX11-WGP-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-WGP-NEXT:    s_endpgm
 ;
 ; GFX11-CU-LABEL: flat_wavefront_one_as_seq_cst_ret_atomicrmw:
 ; GFX11-CU:       ; %bb.0: ; %entry
 ; GFX11-CU-NEXT:    s_clause 0x1
 ; GFX11-CU-NEXT:    s_load_b64 s[2:3], s[0:1], 0x0
 ; GFX11-CU-NEXT:    s_load_b32 s0, s[0:1], 0x8
 ; GFX11-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v0, s2 :: v_dual_mov_b32 v1, s3
 ; GFX11-CU-NEXT:    v_mov_b32_e32 v2, s0
 ; GFX11-CU-NEXT:    flat_atomic_swap_b32 v2, v[0:1], v2 glc
 ; GFX11-CU-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX11-CU-NEXT:    flat_store_b32 v[0:1], v2
-; GFX11-CU-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-CU-NEXT:    s_endpgm
     ptr %out, i32 %in) {
 entry:
   %val = atomicrmw volatile xchg ptr %out, i32 %in syncscope("wavefront-one-as") seq_cst
   store i32 %val, ptr %out, align 4
   ret void
 }
 

 ;
 ; GFX11-WGP-LABEL: flat_wavefront_one_as_monotonic_monotonic_cmpxchg:
 ; GFX11-WGP:       ; %bb.0: ; %entry
 ; GFX11-WGP-NEXT:    s_load_b128 s[0:3], s[0:1], 0x0
 ; GFX11-WGP-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-WGP-NEXT:    v_dual_mov_b32 v0, s0 :: v_dual_mov_b32 v1, s1
 ; GFX11-WGP-NEXT:    v_dual_mov_b32 v2, s2 :: v_dual_mov_b32 v3, s3
 ; GFX11-WGP-NEXT:    flat_atomic_cmpswap_b32 v[0:1], v[2:3] offset:16
-; GFX11-WGP-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-WGP-NEXT:    s_endpgm
 ;
 ; GFX11-CU-LABEL: flat_wavefront_one_as_monotonic_monotonic_cmpxchg:
 ; GFX11-CU:       ; %bb.0: ; %entry
 ; GFX11-CU-NEXT:    s_load_b128 s[0:3], s[0:1], 0x0
 ; GFX11-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v0, s0 :: v_dual_mov_b32 v1, s1
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v2, s2 :: v_dual_mov_b32 v3, s3
 ; GFX11-CU-NEXT:    flat_atomic_cmpswap_b32 v[0:1], v[2:3] offset:16
-; GFX11-CU-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-CU-NEXT:    s_endpgm
     ptr %out, i32 %in, i32 %old) {
 entry:
   %gep = getelementptr i32, ptr %out, i32 4
   %val = cmpxchg volatile ptr %gep, i32 %old, i32 %in syncscope("wavefront-one-as") monotonic monotonic
   ret void
 }
 

 ;
 ; GFX11-WGP-LABEL: flat_wavefront_one_as_acquire_monotonic_cmpxchg:
 ; GFX11-WGP:       ; %bb.0: ; %entry
 ; GFX11-WGP-NEXT:    s_load_b128 s[0:3], s[0:1], 0x0
 ; GFX11-WGP-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-WGP-NEXT:    v_dual_mov_b32 v0, s0 :: v_dual_mov_b32 v1, s1
 ; GFX11-WGP-NEXT:    v_dual_mov_b32 v2, s2 :: v_dual_mov_b32 v3, s3
 ; GFX11-WGP-NEXT:    flat_atomic_cmpswap_b32 v[0:1], v[2:3] offset:16
-; GFX11-WGP-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-WGP-NEXT:    s_endpgm
 ;
 ; GFX11-CU-LABEL: flat_wavefront_one_as_acquire_monotonic_cmpxchg:
 ; GFX11-CU:       ; %bb.0: ; %entry
 ; GFX11-CU-NEXT:    s_load_b128 s[0:3], s[0:1], 0x0
 ; GFX11-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v0, s0 :: v_dual_mov_b32 v1, s1
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v2, s2 :: v_dual_mov_b32 v3, s3
 ; GFX11-CU-NEXT:    flat_atomic_cmpswap_b32 v[0:1], v[2:3] offset:16
-; GFX11-CU-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-CU-NEXT:    s_endpgm
     ptr %out, i32 %in, i32 %old) {
 entry:
   %gep = getelementptr i32, ptr %out, i32 4
   %val = cmpxchg volatile ptr %gep, i32 %old, i32 %in syncscope("wavefront-one-as") acquire monotonic
   ret void
 }
 

 ;
 ; GFX11-WGP-LABEL: flat_wavefront_one_as_release_monotonic_cmpxchg:
 ; GFX11-WGP:       ; %bb.0: ; %entry
 ; GFX11-WGP-NEXT:    s_load_b128 s[0:3], s[0:1], 0x0
 ; GFX11-WGP-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-WGP-NEXT:    v_dual_mov_b32 v0, s0 :: v_dual_mov_b32 v1, s1
 ; GFX11-WGP-NEXT:    v_dual_mov_b32 v2, s2 :: v_dual_mov_b32 v3, s3
 ; GFX11-WGP-NEXT:    flat_atomic_cmpswap_b32 v[0:1], v[2:3] offset:16
-; GFX11-WGP-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-WGP-NEXT:    s_endpgm
 ;
 ; GFX11-CU-LABEL: flat_wavefront_one_as_release_monotonic_cmpxchg:
 ; GFX11-CU:       ; %bb.0: ; %entry
 ; GFX11-CU-NEXT:    s_load_b128 s[0:3], s[0:1], 0x0
 ; GFX11-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v0, s0 :: v_dual_mov_b32 v1, s1
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v2, s2 :: v_dual_mov_b32 v3, s3
 ; GFX11-CU-NEXT:    flat_atomic_cmpswap_b32 v[0:1], v[2:3] offset:16
-; GFX11-CU-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-CU-NEXT:    s_endpgm
     ptr %out, i32 %in, i32 %old) {
 entry:
   %gep = getelementptr i32, ptr %out, i32 4
   %val = cmpxchg volatile ptr %gep, i32 %old, i32 %in syncscope("wavefront-one-as") release monotonic
   ret void
 }
 

 ;
 ; GFX11-WGP-LABEL: flat_wavefront_one_as_acq_rel_monotonic_cmpxchg:
 ; GFX11-WGP:       ; %bb.0: ; %entry
 ; GFX11-WGP-NEXT:    s_load_b128 s[0:3], s[0:1], 0x0
 ; GFX11-WGP-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-WGP-NEXT:    v_dual_mov_b32 v0, s0 :: v_dual_mov_b32 v1, s1
 ; GFX11-WGP-NEXT:    v_dual_mov_b32 v2, s2 :: v_dual_mov_b32 v3, s3
 ; GFX11-WGP-NEXT:    flat_atomic_cmpswap_b32 v[0:1], v[2:3] offset:16
-; GFX11-WGP-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-WGP-NEXT:    s_endpgm
 ;
 ; GFX11-CU-LABEL: flat_wavefront_one_as_acq_rel_monotonic_cmpxchg:
 ; GFX11-CU:       ; %bb.0: ; %entry
 ; GFX11-CU-NEXT:    s_load_b128 s[0:3], s[0:1], 0x0
 ; GFX11-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v0, s0 :: v_dual_mov_b32 v1, s1
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v2, s2 :: v_dual_mov_b32 v3, s3
 ; GFX11-CU-NEXT:    flat_atomic_cmpswap_b32 v[0:1], v[2:3] offset:16
-; GFX11-CU-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-CU-NEXT:    s_endpgm
     ptr %out, i32 %in, i32 %old) {
 entry:
   %gep = getelementptr i32, ptr %out, i32 4
   %val = cmpxchg volatile ptr %gep, i32 %old, i32 %in syncscope("wavefront-one-as") acq_rel monotonic
   ret void
 }
 

 ;
 ; GFX11-WGP-LABEL: flat_wavefront_one_as_seq_cst_monotonic_cmpxchg:
 ; GFX11-WGP:       ; %bb.0: ; %entry
 ; GFX11-WGP-NEXT:    s_load_b128 s[0:3], s[0:1], 0x0
 ; GFX11-WGP-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-WGP-NEXT:    v_dual_mov_b32 v0, s0 :: v_dual_mov_b32 v1, s1
 ; GFX11-WGP-NEXT:    v_dual_mov_b32 v2, s2 :: v_dual_mov_b32 v3, s3
 ; GFX11-WGP-NEXT:    flat_atomic_cmpswap_b32 v[0:1], v[2:3] offset:16
-; GFX11-WGP-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-WGP-NEXT:    s_endpgm
 ;
 ; GFX11-CU-LABEL: flat_wavefront_one_as_seq_cst_monotonic_cmpxchg:
 ; GFX11-CU:       ; %bb.0: ; %entry
 ; GFX11-CU-NEXT:    s_load_b128 s[0:3], s[0:1], 0x0
 ; GFX11-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v0, s0 :: v_dual_mov_b32 v1, s1
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v2, s2 :: v_dual_mov_b32 v3, s3
 ; GFX11-CU-NEXT:    flat_atomic_cmpswap_b32 v[0:1], v[2:3] offset:16
-; GFX11-CU-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-CU-NEXT:    s_endpgm
     ptr %out, i32 %in, i32 %old) {
 entry:
   %gep = getelementptr i32, ptr %out, i32 4
   %val = cmpxchg volatile ptr %gep, i32 %old, i32 %in syncscope("wavefront-one-as") seq_cst monotonic
   ret void
 }
 

 ;
 ; GFX11-WGP-LABEL: flat_wavefront_one_as_monotonic_acquire_cmpxchg:
 ; GFX11-WGP:       ; %bb.0: ; %entry
 ; GFX11-WGP-NEXT:    s_load_b128 s[0:3], s[0:1], 0x0
 ; GFX11-WGP-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-WGP-NEXT:    v_dual_mov_b32 v0, s0 :: v_dual_mov_b32 v1, s1
 ; GFX11-WGP-NEXT:    v_dual_mov_b32 v2, s2 :: v_dual_mov_b32 v3, s3
 ; GFX11-WGP-NEXT:    flat_atomic_cmpswap_b32 v[0:1], v[2:3] offset:16
-; GFX11-WGP-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-WGP-NEXT:    s_endpgm
 ;
 ; GFX11-CU-LABEL: flat_wavefront_one_as_monotonic_acquire_cmpxchg:
 ; GFX11-CU:       ; %bb.0: ; %entry
 ; GFX11-CU-NEXT:    s_load_b128 s[0:3], s[0:1], 0x0
 ; GFX11-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v0, s0 :: v_dual_mov_b32 v1, s1
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v2, s2 :: v_dual_mov_b32 v3, s3
 ; GFX11-CU-NEXT:    flat_atomic_cmpswap_b32 v[0:1], v[2:3] offset:16
-; GFX11-CU-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-CU-NEXT:    s_endpgm
     ptr %out, i32 %in, i32 %old) {
 entry:
   %gep = getelementptr i32, ptr %out, i32 4
   %val = cmpxchg volatile ptr %gep, i32 %old, i32 %in syncscope("wavefront-one-as") monotonic acquire
   ret void
 }
 

 ;
 ; GFX11-WGP-LABEL: flat_wavefront_one_as_acquire_acquire_cmpxchg:
 ; GFX11-WGP:       ; %bb.0: ; %entry
 ; GFX11-WGP-NEXT:    s_load_b128 s[0:3], s[0:1], 0x0
 ; GFX11-WGP-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-WGP-NEXT:    v_dual_mov_b32 v0, s0 :: v_dual_mov_b32 v1, s1
 ; GFX11-WGP-NEXT:    v_dual_mov_b32 v2, s2 :: v_dual_mov_b32 v3, s3
 ; GFX11-WGP-NEXT:    flat_atomic_cmpswap_b32 v[0:1], v[2:3] offset:16
-; GFX11-WGP-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-WGP-NEXT:    s_endpgm
 ;
 ; GFX11-CU-LABEL: flat_wavefront_one_as_acquire_acquire_cmpxchg:
 ; GFX11-CU:       ; %bb.0: ; %entry
 ; GFX11-CU-NEXT:    s_load_b128 s[0:3], s[0:1], 0x0
 ; GFX11-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v0, s0 :: v_dual_mov_b32 v1, s1
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v2, s2 :: v_dual_mov_b32 v3, s3
 ; GFX11-CU-NEXT:    flat_atomic_cmpswap_b32 v[0:1], v[2:3] offset:16
-; GFX11-CU-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-CU-NEXT:    s_endpgm
     ptr %out, i32 %in, i32 %old) {
 entry:
   %gep = getelementptr i32, ptr %out, i32 4
   %val = cmpxchg volatile ptr %gep, i32 %old, i32 %in syncscope("wavefront-one-as") acquire acquire
   ret void
 }
 

 ;
 ; GFX11-WGP-LABEL: flat_wavefront_one_as_release_acquire_cmpxchg:
 ; GFX11-WGP:       ; %bb.0: ; %entry
 ; GFX11-WGP-NEXT:    s_load_b128 s[0:3], s[0:1], 0x0
 ; GFX11-WGP-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-WGP-NEXT:    v_dual_mov_b32 v0, s0 :: v_dual_mov_b32 v1, s1
 ; GFX11-WGP-NEXT:    v_dual_mov_b32 v2, s2 :: v_dual_mov_b32 v3, s3
 ; GFX11-WGP-NEXT:    flat_atomic_cmpswap_b32 v[0:1], v[2:3] offset:16
-; GFX11-WGP-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-WGP-NEXT:    s_endpgm
 ;
 ; GFX11-CU-LABEL: flat_wavefront_one_as_release_acquire_cmpxchg:
 ; GFX11-CU:       ; %bb.0: ; %entry
 ; GFX11-CU-NEXT:    s_load_b128 s[0:3], s[0:1], 0x0
 ; GFX11-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v0, s0 :: v_dual_mov_b32 v1, s1
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v2, s2 :: v_dual_mov_b32 v3, s3
 ; GFX11-CU-NEXT:    flat_atomic_cmpswap_b32 v[0:1], v[2:3] offset:16
-; GFX11-CU-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-CU-NEXT:    s_endpgm
     ptr %out, i32 %in, i32 %old) {
 entry:
   %gep = getelementptr i32, ptr %out, i32 4
   %val = cmpxchg volatile ptr %gep, i32 %old, i32 %in syncscope("wavefront-one-as") release acquire
   ret void
 }
 

 ;
 ; GFX11-WGP-LABEL: flat_wavefront_one_as_acq_rel_acquire_cmpxchg:
 ; GFX11-WGP:       ; %bb.0: ; %entry
 ; GFX11-WGP-NEXT:    s_load_b128 s[0:3], s[0:1], 0x0
 ; GFX11-WGP-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-WGP-NEXT:    v_dual_mov_b32 v0, s0 :: v_dual_mov_b32 v1, s1
 ; GFX11-WGP-NEXT:    v_dual_mov_b32 v2, s2 :: v_dual_mov_b32 v3, s3
 ; GFX11-WGP-NEXT:    flat_atomic_cmpswap_b32 v[0:1], v[2:3] offset:16
-; GFX11-WGP-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-WGP-NEXT:    s_endpgm
 ;
 ; GFX11-CU-LABEL: flat_wavefront_one_as_acq_rel_acquire_cmpxchg:
 ; GFX11-CU:       ; %bb.0: ; %entry
 ; GFX11-CU-NEXT:    s_load_b128 s[0:3], s[0:1], 0x0
 ; GFX11-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v0, s0 :: v_dual_mov_b32 v1, s1
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v2, s2 :: v_dual_mov_b32 v3, s3
 ; GFX11-CU-NEXT:    flat_atomic_cmpswap_b32 v[0:1], v[2:3] offset:16
-; GFX11-CU-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-CU-NEXT:    s_endpgm
     ptr %out, i32 %in, i32 %old) {
 entry:
   %gep = getelementptr i32, ptr %out, i32 4
   %val = cmpxchg volatile ptr %gep, i32 %old, i32 %in syncscope("wavefront-one-as") acq_rel acquire
   ret void
 }
 

 ;
 ; GFX11-WGP-LABEL: flat_wavefront_one_as_seq_cst_acquire_cmpxchg:
 ; GFX11-WGP:       ; %bb.0: ; %entry
 ; GFX11-WGP-NEXT:    s_load_b128 s[0:3], s[0:1], 0x0
 ; GFX11-WGP-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-WGP-NEXT:    v_dual_mov_b32 v0, s0 :: v_dual_mov_b32 v1, s1
 ; GFX11-WGP-NEXT:    v_dual_mov_b32 v2, s2 :: v_dual_mov_b32 v3, s3
 ; GFX11-WGP-NEXT:    flat_atomic_cmpswap_b32 v[0:1], v[2:3] offset:16
-; GFX11-WGP-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-WGP-NEXT:    s_endpgm
 ;
 ; GFX11-CU-LABEL: flat_wavefront_one_as_seq_cst_acquire_cmpxchg:
 ; GFX11-CU:       ; %bb.0: ; %entry
 ; GFX11-CU-NEXT:    s_load_b128 s[0:3], s[0:1], 0x0
 ; GFX11-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v0, s0 :: v_dual_mov_b32 v1, s1
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v2, s2 :: v_dual_mov_b32 v3, s3
 ; GFX11-CU-NEXT:    flat_atomic_cmpswap_b32 v[0:1], v[2:3] offset:16
-; GFX11-CU-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-CU-NEXT:    s_endpgm
     ptr %out, i32 %in, i32 %old) {
 entry:
   %gep = getelementptr i32, ptr %out, i32 4
   %val = cmpxchg volatile ptr %gep, i32 %old, i32 %in syncscope("wavefront-one-as") seq_cst acquire
   ret void
 }
 

 ;
 ; GFX11-WGP-LABEL: flat_wavefront_one_as_monotonic_seq_cst_cmpxchg:
 ; GFX11-WGP:       ; %bb.0: ; %entry
 ; GFX11-WGP-NEXT:    s_load_b128 s[0:3], s[0:1], 0x0
 ; GFX11-WGP-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-WGP-NEXT:    v_dual_mov_b32 v0, s0 :: v_dual_mov_b32 v1, s1
 ; GFX11-WGP-NEXT:    v_dual_mov_b32 v2, s2 :: v_dual_mov_b32 v3, s3
 ; GFX11-WGP-NEXT:    flat_atomic_cmpswap_b32 v[0:1], v[2:3] offset:16
-; GFX11-WGP-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-WGP-NEXT:    s_endpgm
 ;
 ; GFX11-CU-LABEL: flat_wavefront_one_as_monotonic_seq_cst_cmpxchg:
 ; GFX11-CU:       ; %bb.0: ; %entry
 ; GFX11-CU-NEXT:    s_load_b128 s[0:3], s[0:1], 0x0
 ; GFX11-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v0, s0 :: v_dual_mov_b32 v1, s1
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v2, s2 :: v_dual_mov_b32 v3, s3
 ; GFX11-CU-NEXT:    flat_atomic_cmpswap_b32 v[0:1], v[2:3] offset:16
-; GFX11-CU-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-CU-NEXT:    s_endpgm
     ptr %out, i32 %in, i32 %old) {
 entry:
   %gep = getelementptr i32, ptr %out, i32 4
   %val = cmpxchg volatile ptr %gep, i32 %old, i32 %in syncscope("wavefront-one-as") monotonic seq_cst
   ret void
 }
 

 ;
 ; GFX11-WGP-LABEL: flat_wavefront_one_as_acquire_seq_cst_cmpxchg:
 ; GFX11-WGP:       ; %bb.0: ; %entry
 ; GFX11-WGP-NEXT:    s_load_b128 s[0:3], s[0:1], 0x0
 ; GFX11-WGP-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-WGP-NEXT:    v_dual_mov_b32 v0, s0 :: v_dual_mov_b32 v1, s1
 ; GFX11-WGP-NEXT:    v_dual_mov_b32 v2, s2 :: v_dual_mov_b32 v3, s3
 ; GFX11-WGP-NEXT:    flat_atomic_cmpswap_b32 v[0:1], v[2:3] offset:16
-; GFX11-WGP-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-WGP-NEXT:    s_endpgm
 ;
 ; GFX11-CU-LABEL: flat_wavefront_one_as_acquire_seq_cst_cmpxchg:
 ; GFX11-CU:       ; %bb.0: ; %entry
 ; GFX11-CU-NEXT:    s_load_b128 s[0:3], s[0:1], 0x0
 ; GFX11-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v0, s0 :: v_dual_mov_b32 v1, s1
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v2, s2 :: v_dual_mov_b32 v3, s3
 ; GFX11-CU-NEXT:    flat_atomic_cmpswap_b32 v[0:1], v[2:3] offset:16
-; GFX11-CU-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-CU-NEXT:    s_endpgm
     ptr %out, i32 %in, i32 %old) {
 entry:
   %gep = getelementptr i32, ptr %out, i32 4
   %val = cmpxchg volatile ptr %gep, i32 %old, i32 %in syncscope("wavefront-one-as") acquire seq_cst
   ret void
 }
 

 ;
 ; GFX11-WGP-LABEL: flat_wavefront_one_as_release_seq_cst_cmpxchg:
 ; GFX11-WGP:       ; %bb.0: ; %entry
 ; GFX11-WGP-NEXT:    s_load_b128 s[0:3], s[0:1], 0x0
 ; GFX11-WGP-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-WGP-NEXT:    v_dual_mov_b32 v0, s0 :: v_dual_mov_b32 v1, s1
 ; GFX11-WGP-NEXT:    v_dual_mov_b32 v2, s2 :: v_dual_mov_b32 v3, s3
 ; GFX11-WGP-NEXT:    flat_atomic_cmpswap_b32 v[0:1], v[2:3] offset:16
-; GFX11-WGP-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-WGP-NEXT:    s_endpgm
 ;
 ; GFX11-CU-LABEL: flat_wavefront_one_as_release_seq_cst_cmpxchg:
 ; GFX11-CU:       ; %bb.0: ; %entry
 ; GFX11-CU-NEXT:    s_load_b128 s[0:3], s[0:1], 0x0
 ; GFX11-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v0, s0 :: v_dual_mov_b32 v1, s1
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v2, s2 :: v_dual_mov_b32 v3, s3
 ; GFX11-CU-NEXT:    flat_atomic_cmpswap_b32 v[0:1], v[2:3] offset:16
-; GFX11-CU-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-CU-NEXT:    s_endpgm
     ptr %out, i32 %in, i32 %old) {
 entry:
   %gep = getelementptr i32, ptr %out, i32 4
   %val = cmpxchg volatile ptr %gep, i32 %old, i32 %in syncscope("wavefront-one-as") release seq_cst
   ret void
 }
 

 ;
 ; GFX11-WGP-LABEL: flat_wavefront_one_as_acq_rel_seq_cst_cmpxchg:
 ; GFX11-WGP:       ; %bb.0: ; %entry
 ; GFX11-WGP-NEXT:    s_load_b128 s[0:3], s[0:1], 0x0
 ; GFX11-WGP-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-WGP-NEXT:    v_dual_mov_b32 v0, s0 :: v_dual_mov_b32 v1, s1
 ; GFX11-WGP-NEXT:    v_dual_mov_b32 v2, s2 :: v_dual_mov_b32 v3, s3
 ; GFX11-WGP-NEXT:    flat_atomic_cmpswap_b32 v[0:1], v[2:3] offset:16
-; GFX11-WGP-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-WGP-NEXT:    s_endpgm
 ;
 ; GFX11-CU-LABEL: flat_wavefront_one_as_acq_rel_seq_cst_cmpxchg:
 ; GFX11-CU:       ; %bb.0: ; %entry
 ; GFX11-CU-NEXT:    s_load_b128 s[0:3], s[0:1], 0x0
 ; GFX11-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v0, s0 :: v_dual_mov_b32 v1, s1
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v2, s2 :: v_dual_mov_b32 v3, s3
 ; GFX11-CU-NEXT:    flat_atomic_cmpswap_b32 v[0:1], v[2:3] offset:16
-; GFX11-CU-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-CU-NEXT:    s_endpgm
     ptr %out, i32 %in, i32 %old) {
 entry:
   %gep = getelementptr i32, ptr %out, i32 4
   %val = cmpxchg volatile ptr %gep, i32 %old, i32 %in syncscope("wavefront-one-as") acq_rel seq_cst
   ret void
 }
 

 ;
 ; GFX11-WGP-LABEL: flat_wavefront_one_as_seq_cst_seq_cst_cmpxchg:
 ; GFX11-WGP:       ; %bb.0: ; %entry
 ; GFX11-WGP-NEXT:    s_load_b128 s[0:3], s[0:1], 0x0
 ; GFX11-WGP-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-WGP-NEXT:    v_dual_mov_b32 v0, s0 :: v_dual_mov_b32 v1, s1
 ; GFX11-WGP-NEXT:    v_dual_mov_b32 v2, s2 :: v_dual_mov_b32 v3, s3
 ; GFX11-WGP-NEXT:    flat_atomic_cmpswap_b32 v[0:1], v[2:3] offset:16
-; GFX11-WGP-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-WGP-NEXT:    s_endpgm
 ;
 ; GFX11-CU-LABEL: flat_wavefront_one_as_seq_cst_seq_cst_cmpxchg:
 ; GFX11-CU:       ; %bb.0: ; %entry
 ; GFX11-CU-NEXT:    s_load_b128 s[0:3], s[0:1], 0x0
 ; GFX11-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v0, s0 :: v_dual_mov_b32 v1, s1
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v2, s2 :: v_dual_mov_b32 v3, s3
 ; GFX11-CU-NEXT:    flat_atomic_cmpswap_b32 v[0:1], v[2:3] offset:16
-; GFX11-CU-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-CU-NEXT:    s_endpgm
     ptr %out, i32 %in, i32 %old) {
 entry:
   %gep = getelementptr i32, ptr %out, i32 4
   %val = cmpxchg volatile ptr %gep, i32 %old, i32 %in syncscope("wavefront-one-as") seq_cst seq_cst
   ret void
 }
 

 ; GFX11-WGP:       ; %bb.0: ; %entry
 ; GFX11-WGP-NEXT:    s_load_b128 s[0:3], s[0:1], 0x0
 ; GFX11-WGP-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-WGP-NEXT:    v_dual_mov_b32 v0, s0 :: v_dual_mov_b32 v1, s1
 ; GFX11-WGP-NEXT:    v_dual_mov_b32 v2, s2 :: v_dual_mov_b32 v3, s3
 ; GFX11-WGP-NEXT:    flat_atomic_cmpswap_b32 v2, v[0:1], v[2:3] offset:16 glc
 ; GFX11-WGP-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX11-WGP-NEXT:    flat_store_b32 v[0:1], v2
-; GFX11-WGP-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-WGP-NEXT:    s_endpgm
 ;
 ; GFX11-CU-LABEL: flat_wavefront_one_as_monotonic_monotonic_ret_cmpxchg:
 ; GFX11-CU:       ; %bb.0: ; %entry
 ; GFX11-CU-NEXT:    s_load_b128 s[0:3], s[0:1], 0x0
 ; GFX11-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v0, s0 :: v_dual_mov_b32 v1, s1
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v2, s2 :: v_dual_mov_b32 v3, s3
 ; GFX11-CU-NEXT:    flat_atomic_cmpswap_b32 v2, v[0:1], v[2:3] offset:16 glc
 ; GFX11-CU-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX11-CU-NEXT:    flat_store_b32 v[0:1], v2
-; GFX11-CU-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-CU-NEXT:    s_endpgm
     ptr %out, i32 %in, i32 %old) {
 entry:
   %gep = getelementptr i32, ptr %out, i32 4
   %val = cmpxchg volatile ptr %gep, i32 %old, i32 %in syncscope("wavefront-one-as") monotonic monotonic
   %val0 = extractvalue { i32, i1 } %val, 0
   store i32 %val0, ptr %out, align 4
   ret void

 ; GFX11-WGP:       ; %bb.0: ; %entry
 ; GFX11-WGP-NEXT:    s_load_b128 s[0:3], s[0:1], 0x0
 ; GFX11-WGP-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-WGP-NEXT:    v_dual_mov_b32 v0, s0 :: v_dual_mov_b32 v1, s1
 ; GFX11-WGP-NEXT:    v_dual_mov_b32 v2, s2 :: v_dual_mov_b32 v3, s3
 ; GFX11-WGP-NEXT:    flat_atomic_cmpswap_b32 v2, v[0:1], v[2:3] offset:16 glc
 ; GFX11-WGP-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX11-WGP-NEXT:    flat_store_b32 v[0:1], v2
-; GFX11-WGP-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-WGP-NEXT:    s_endpgm
 ;
 ; GFX11-CU-LABEL: flat_wavefront_one_as_acquire_monotonic_ret_cmpxchg:
 ; GFX11-CU:       ; %bb.0: ; %entry
 ; GFX11-CU-NEXT:    s_load_b128 s[0:3], s[0:1], 0x0
 ; GFX11-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v0, s0 :: v_dual_mov_b32 v1, s1
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v2, s2 :: v_dual_mov_b32 v3, s3
 ; GFX11-CU-NEXT:    flat_atomic_cmpswap_b32 v2, v[0:1], v[2:3] offset:16 glc
 ; GFX11-CU-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX11-CU-NEXT:    flat_store_b32 v[0:1], v2
-; GFX11-CU-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-CU-NEXT:    s_endpgm
     ptr %out, i32 %in, i32 %old) {
 entry:
   %gep = getelementptr i32, ptr %out, i32 4
   %val = cmpxchg volatile ptr %gep, i32 %old, i32 %in syncscope("wavefront-one-as") acquire monotonic
   %val0 = extractvalue { i32, i1 } %val, 0
   store i32 %val0, ptr %out, align 4
   ret void

 ; GFX11-WGP:       ; %bb.0: ; %entry
 ; GFX11-WGP-NEXT:    s_load_b128 s[0:3], s[0:1], 0x0
 ; GFX11-WGP-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-WGP-NEXT:    v_dual_mov_b32 v0, s0 :: v_dual_mov_b32 v1, s1
 ; GFX11-WGP-NEXT:    v_dual_mov_b32 v2, s2 :: v_dual_mov_b32 v3, s3
 ; GFX11-WGP-NEXT:    flat_atomic_cmpswap_b32 v2, v[0:1], v[2:3] offset:16 glc
 ; GFX11-WGP-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX11-WGP-NEXT:    flat_store_b32 v[0:1], v2
-; GFX11-WGP-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-WGP-NEXT:    s_endpgm
 ;
 ; GFX11-CU-LABEL: flat_wavefront_one_as_acq_rel_monotonic_ret_cmpxchg:
 ; GFX11-CU:       ; %bb.0: ; %entry
 ; GFX11-CU-NEXT:    s_load_b128 s[0:3], s[0:1], 0x0
 ; GFX11-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v0, s0 :: v_dual_mov_b32 v1, s1
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v2, s2 :: v_dual_mov_b32 v3, s3
 ; GFX11-CU-NEXT:    flat_atomic_cmpswap_b32 v2, v[0:1], v[2:3] offset:16 glc
 ; GFX11-CU-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX11-CU-NEXT:    flat_store_b32 v[0:1], v2
-; GFX11-CU-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-CU-NEXT:    s_endpgm
     ptr %out, i32 %in, i32 %old) {
 entry:
   %gep = getelementptr i32, ptr %out, i32 4
   %val = cmpxchg volatile ptr %gep, i32 %old, i32 %in syncscope("wavefront-one-as") acq_rel monotonic
   %val0 = extractvalue { i32, i1 } %val, 0
   store i32 %val0, ptr %out, align 4
   ret void

 ; GFX11-WGP:       ; %bb.0: ; %entry
 ; GFX11-WGP-NEXT:    s_load_b128 s[0:3], s[0:1], 0x0
 ; GFX11-WGP-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-WGP-NEXT:    v_dual_mov_b32 v0, s0 :: v_dual_mov_b32 v1, s1
 ; GFX11-WGP-NEXT:    v_dual_mov_b32 v2, s2 :: v_dual_mov_b32 v3, s3
 ; GFX11-WGP-NEXT:    flat_atomic_cmpswap_b32 v2, v[0:1], v[2:3] offset:16 glc
 ; GFX11-WGP-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX11-WGP-NEXT:    flat_store_b32 v[0:1], v2
-; GFX11-WGP-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-WGP-NEXT:    s_endpgm
 ;
 ; GFX11-CU-LABEL: flat_wavefront_one_as_seq_cst_monotonic_ret_cmpxchg:
 ; GFX11-CU:       ; %bb.0: ; %entry
 ; GFX11-CU-NEXT:    s_load_b128 s[0:3], s[0:1], 0x0
 ; GFX11-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v0, s0 :: v_dual_mov_b32 v1, s1
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v2, s2 :: v_dual_mov_b32 v3, s3
 ; GFX11-CU-NEXT:    flat_atomic_cmpswap_b32 v2, v[0:1], v[2:3] offset:16 glc
 ; GFX11-CU-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX11-CU-NEXT:    flat_store_b32 v[0:1], v2
-; GFX11-CU-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-CU-NEXT:    s_endpgm
     ptr %out, i32 %in, i32 %old) {
 entry:
   %gep = getelementptr i32, ptr %out, i32 4
   %val = cmpxchg volatile ptr %gep, i32 %old, i32 %in syncscope("wavefront-one-as") seq_cst monotonic
   %val0 = extractvalue { i32, i1 } %val, 0
   store i32 %val0, ptr %out, align 4
   ret void

 ; GFX11-WGP:       ; %bb.0: ; %entry
 ; GFX11-WGP-NEXT:    s_load_b128 s[0:3], s[0:1], 0x0
 ; GFX11-WGP-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-WGP-NEXT:    v_dual_mov_b32 v0, s0 :: v_dual_mov_b32 v1, s1
 ; GFX11-WGP-NEXT:    v_dual_mov_b32 v2, s2 :: v_dual_mov_b32 v3, s3
 ; GFX11-WGP-NEXT:    flat_atomic_cmpswap_b32 v2, v[0:1], v[2:3] offset:16 glc
 ; GFX11-WGP-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX11-WGP-NEXT:    flat_store_b32 v[0:1], v2
-; GFX11-WGP-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-WGP-NEXT:    s_endpgm
 ;
 ; GFX11-CU-LABEL: flat_wavefront_one_as_monotonic_acquire_ret_cmpxchg:
 ; GFX11-CU:       ; %bb.0: ; %entry
 ; GFX11-CU-NEXT:    s_load_b128 s[0:3], s[0:1], 0x0
 ; GFX11-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v0, s0 :: v_dual_mov_b32 v1, s1
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v2, s2 :: v_dual_mov_b32 v3, s3
 ; GFX11-CU-NEXT:    flat_atomic_cmpswap_b32 v2, v[0:1], v[2:3] offset:16 glc
 ; GFX11-CU-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX11-CU-NEXT:    flat_store_b32 v[0:1], v2
-; GFX11-CU-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-CU-NEXT:    s_endpgm
     ptr %out, i32 %in, i32 %old) {
 entry:
   %gep = getelementptr i32, ptr %out, i32 4
   %val = cmpxchg volatile ptr %gep, i32 %old, i32 %in syncscope("wavefront-one-as") monotonic acquire
   %val0 = extractvalue { i32, i1 } %val, 0
   store i32 %val0, ptr %out, align 4
   ret void

 ; GFX11-WGP:       ; %bb.0: ; %entry
 ; GFX11-WGP-NEXT:    s_load_b128 s[0:3], s[0:1], 0x0
 ; GFX11-WGP-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-WGP-NEXT:    v_dual_mov_b32 v0, s0 :: v_dual_mov_b32 v1, s1
 ; GFX11-WGP-NEXT:    v_dual_mov_b32 v2, s2 :: v_dual_mov_b32 v3, s3
 ; GFX11-WGP-NEXT:    flat_atomic_cmpswap_b32 v2, v[0:1], v[2:3] offset:16 glc
 ; GFX11-WGP-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX11-WGP-NEXT:    flat_store_b32 v[0:1], v2
-; GFX11-WGP-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-WGP-NEXT:    s_endpgm
 ;
 ; GFX11-CU-LABEL: flat_wavefront_one_as_acquire_acquire_ret_cmpxchg:
 ; GFX11-CU:       ; %bb.0: ; %entry
 ; GFX11-CU-NEXT:    s_load_b128 s[0:3], s[0:1], 0x0
 ; GFX11-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v0, s0 :: v_dual_mov_b32 v1, s1
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v2, s2 :: v_dual_mov_b32 v3, s3
 ; GFX11-CU-NEXT:    flat_atomic_cmpswap_b32 v2, v[0:1], v[2:3] offset:16 glc
 ; GFX11-CU-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX11-CU-NEXT:    flat_store_b32 v[0:1], v2
-; GFX11-CU-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-CU-NEXT:    s_endpgm
     ptr %out, i32 %in, i32 %old) {
 entry:
   %gep = getelementptr i32, ptr %out, i32 4
   %val = cmpxchg volatile ptr %gep, i32 %old, i32 %in syncscope("wavefront-one-as") acquire acquire
   %val0 = extractvalue { i32, i1 } %val, 0
   store i32 %val0, ptr %out, align 4
   ret void

 ; GFX11-WGP:       ; %bb.0: ; %entry
 ; GFX11-WGP-NEXT:    s_load_b128 s[0:3], s[0:1], 0x0
 ; GFX11-WGP-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-WGP-NEXT:    v_dual_mov_b32 v0, s0 :: v_dual_mov_b32 v1, s1
 ; GFX11-WGP-NEXT:    v_dual_mov_b32 v2, s2 :: v_dual_mov_b32 v3, s3
 ; GFX11-WGP-NEXT:    flat_atomic_cmpswap_b32 v2, v[0:1], v[2:3] offset:16 glc
 ; GFX11-WGP-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX11-WGP-NEXT:    flat_store_b32 v[0:1], v2
-; GFX11-WGP-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-WGP-NEXT:    s_endpgm
 ;
 ; GFX11-CU-LABEL: flat_wavefront_one_as_release_acquire_ret_cmpxchg:
 ; GFX11-CU:       ; %bb.0: ; %entry
 ; GFX11-CU-NEXT:    s_load_b128 s[0:3], s[0:1], 0x0
 ; GFX11-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v0, s0 :: v_dual_mov_b32 v1, s1
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v2, s2 :: v_dual_mov_b32 v3, s3
 ; GFX11-CU-NEXT:    flat_atomic_cmpswap_b32 v2, v[0:1], v[2:3] offset:16 glc
 ; GFX11-CU-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX11-CU-NEXT:    flat_store_b32 v[0:1], v2
-; GFX11-CU-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-CU-NEXT:    s_endpgm
     ptr %out, i32 %in, i32 %old) {
 entry:
   %gep = getelementptr i32, ptr %out, i32 4
   %val = cmpxchg volatile ptr %gep, i32 %old, i32 %in syncscope("wavefront-one-as") release acquire
   %val0 = extractvalue { i32, i1 } %val, 0
   store i32 %val0, ptr %out, align 4
   ret void

 ; GFX11-WGP:       ; %bb.0: ; %entry
 ; GFX11-WGP-NEXT:    s_load_b128 s[0:3], s[0:1], 0x0
 ; GFX11-WGP-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-WGP-NEXT:    v_dual_mov_b32 v0, s0 :: v_dual_mov_b32 v1, s1
 ; GFX11-WGP-NEXT:    v_dual_mov_b32 v2, s2 :: v_dual_mov_b32 v3, s3
 ; GFX11-WGP-NEXT:    flat_atomic_cmpswap_b32 v2, v[0:1], v[2:3] offset:16 glc
 ; GFX11-WGP-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX11-WGP-NEXT:    flat_store_b32 v[0:1], v2
-; GFX11-WGP-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-WGP-NEXT:    s_endpgm
 ;
 ; GFX11-CU-LABEL: flat_wavefront_one_as_acq_rel_acquire_ret_cmpxchg:
 ; GFX11-CU:       ; %bb.0: ; %entry
 ; GFX11-CU-NEXT:    s_load_b128 s[0:3], s[0:1], 0x0
 ; GFX11-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v0, s0 :: v_dual_mov_b32 v1, s1
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v2, s2 :: v_dual_mov_b32 v3, s3
 ; GFX11-CU-NEXT:    flat_atomic_cmpswap_b32 v2, v[0:1], v[2:3] offset:16 glc
 ; GFX11-CU-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX11-CU-NEXT:    flat_store_b32 v[0:1], v2
-; GFX11-CU-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-CU-NEXT:    s_endpgm
     ptr %out, i32 %in, i32 %old) {
 entry:
   %gep = getelementptr i32, ptr %out, i32 4
   %val = cmpxchg volatile ptr %gep, i32 %old, i32 %in syncscope("wavefront-one-as") acq_rel acquire
   %val0 = extractvalue { i32, i1 } %val, 0
   store i32 %val0, ptr %out, align 4
   ret void

 ; GFX11-WGP:       ; %bb.0: ; %entry
 ; GFX11-WGP-NEXT:    s_load_b128 s[0:3], s[0:1], 0x0
 ; GFX11-WGP-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-WGP-NEXT:    v_dual_mov_b32 v0, s0 :: v_dual_mov_b32 v1, s1
 ; GFX11-WGP-NEXT:    v_dual_mov_b32 v2, s2 :: v_dual_mov_b32 v3, s3
 ; GFX11-WGP-NEXT:    flat_atomic_cmpswap_b32 v2, v[0:1], v[2:3] offset:16 glc
 ; GFX11-WGP-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX11-WGP-NEXT:    flat_store_b32 v[0:1], v2
-; GFX11-WGP-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-WGP-NEXT:    s_endpgm
 ;
 ; GFX11-CU-LABEL: flat_wavefront_one_as_seq_cst_acquire_ret_cmpxchg:
 ; GFX11-CU:       ; %bb.0: ; %entry
 ; GFX11-CU-NEXT:    s_load_b128 s[0:3], s[0:1], 0x0
 ; GFX11-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v0, s0 :: v_dual_mov_b32 v1, s1
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v2, s2 :: v_dual_mov_b32 v3, s3
 ; GFX11-CU-NEXT:    flat_atomic_cmpswap_b32 v2, v[0:1], v[2:3] offset:16 glc
 ; GFX11-CU-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX11-CU-NEXT:    flat_store_b32 v[0:1], v2
-; GFX11-CU-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-CU-NEXT:    s_endpgm
     ptr %out, i32 %in, i32 %old) {
 entry:
   %gep = getelementptr i32, ptr %out, i32 4
   %val = cmpxchg volatile ptr %gep, i32 %old, i32 %in syncscope("wavefront-one-as") seq_cst acquire
   %val0 = extractvalue { i32, i1 } %val, 0
   store i32 %val0, ptr %out, align 4
   ret void

 ; GFX11-WGP:       ; %bb.0: ; %entry
 ; GFX11-WGP-NEXT:    s_load_b128 s[0:3], s[0:1], 0x0
 ; GFX11-WGP-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-WGP-NEXT:    v_dual_mov_b32 v0, s0 :: v_dual_mov_b32 v1, s1
 ; GFX11-WGP-NEXT:    v_dual_mov_b32 v2, s2 :: v_dual_mov_b32 v3, s3
 ; GFX11-WGP-NEXT:    flat_atomic_cmpswap_b32 v2, v[0:1], v[2:3] offset:16 glc
 ; GFX11-WGP-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX11-WGP-NEXT:    flat_store_b32 v[0:1], v2
-; GFX11-WGP-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-WGP-NEXT:    s_endpgm
 ;
 ; GFX11-CU-LABEL: flat_wavefront_one_as_monotonic_seq_cst_ret_cmpxchg:
 ; GFX11-CU:       ; %bb.0: ; %entry
 ; GFX11-CU-NEXT:    s_load_b128 s[0:3], s[0:1], 0x0
 ; GFX11-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v0, s0 :: v_dual_mov_b32 v1, s1
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v2, s2 :: v_dual_mov_b32 v3, s3
 ; GFX11-CU-NEXT:    flat_atomic_cmpswap_b32 v2, v[0:1], v[2:3] offset:16 glc
 ; GFX11-CU-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX11-CU-NEXT:    flat_store_b32 v[0:1], v2
-; GFX11-CU-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-CU-NEXT:    s_endpgm
    ptr %out, i32 %in, i32 %old) {
 entry:
   %gep = getelementptr i32, ptr %out, i32 4
   %val = cmpxchg volatile ptr %gep, i32 %old, i32 %in syncscope("wavefront-one-as") monotonic seq_cst
   %val0 = extractvalue { i32, i1 } %val, 0
   store i32 %val0, ptr %out, align 4
   ret void

 ; GFX11-WGP:       ; %bb.0: ; %entry
 ; GFX11-WGP-NEXT:    s_load_b128 s[0:3], s[0:1], 0x0
 ; GFX11-WGP-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-WGP-NEXT:    v_dual_mov_b32 v0, s0 :: v_dual_mov_b32 v1, s1
 ; GFX11-WGP-NEXT:    v_dual_mov_b32 v2, s2 :: v_dual_mov_b32 v3, s3
 ; GFX11-WGP-NEXT:    flat_atomic_cmpswap_b32 v2, v[0:1], v[2:3] offset:16 glc
 ; GFX11-WGP-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX11-WGP-NEXT:    flat_store_b32 v[0:1], v2
-; GFX11-WGP-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-WGP-NEXT:    s_endpgm
 ;
 ; GFX11-CU-LABEL: flat_wavefront_one_as_acquire_seq_cst_ret_cmpxchg:
 ; GFX11-CU:       ; %bb.0: ; %entry
 ; GFX11-CU-NEXT:    s_load_b128 s[0:3], s[0:1], 0x0
 ; GFX11-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v0, s0 :: v_dual_mov_b32 v1, s1
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v2, s2 :: v_dual_mov_b32 v3, s3
 ; GFX11-CU-NEXT:    flat_atomic_cmpswap_b32 v2, v[0:1], v[2:3] offset:16 glc
 ; GFX11-CU-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX11-CU-NEXT:    flat_store_b32 v[0:1], v2
-; GFX11-CU-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-CU-NEXT:    s_endpgm
    ptr %out, i32 %in, i32 %old) {
 entry:
   %gep = getelementptr i32, ptr %out, i32 4
   %val = cmpxchg volatile ptr %gep, i32 %old, i32 %in syncscope("wavefront-one-as") acquire seq_cst
   %val0 = extractvalue { i32, i1 } %val, 0
   store i32 %val0, ptr %out, align 4
   ret void

 ; GFX11-WGP:       ; %bb.0: ; %entry
 ; GFX11-WGP-NEXT:    s_load_b128 s[0:3], s[0:1], 0x0
 ; GFX11-WGP-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-WGP-NEXT:    v_dual_mov_b32 v0, s0 :: v_dual_mov_b32 v1, s1
 ; GFX11-WGP-NEXT:    v_dual_mov_b32 v2, s2 :: v_dual_mov_b32 v3, s3
 ; GFX11-WGP-NEXT:    flat_atomic_cmpswap_b32 v2, v[0:1], v[2:3] offset:16 glc
 ; GFX11-WGP-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX11-WGP-NEXT:    flat_store_b32 v[0:1], v2
-; GFX11-WGP-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-WGP-NEXT:    s_endpgm
 ;
 ; GFX11-CU-LABEL: flat_wavefront_one_as_release_seq_cst_ret_cmpxchg:
 ; GFX11-CU:       ; %bb.0: ; %entry
 ; GFX11-CU-NEXT:    s_load_b128 s[0:3], s[0:1], 0x0
 ; GFX11-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v0, s0 :: v_dual_mov_b32 v1, s1
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v2, s2 :: v_dual_mov_b32 v3, s3
 ; GFX11-CU-NEXT:    flat_atomic_cmpswap_b32 v2, v[0:1], v[2:3] offset:16 glc
 ; GFX11-CU-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX11-CU-NEXT:    flat_store_b32 v[0:1], v2
-; GFX11-CU-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-CU-NEXT:    s_endpgm
    ptr %out, i32 %in, i32 %old) {
 entry:
   %gep = getelementptr i32, ptr %out, i32 4
   %val = cmpxchg volatile ptr %gep, i32 %old, i32 %in syncscope("wavefront-one-as") release seq_cst
   %val0 = extractvalue { i32, i1 } %val, 0
   store i32 %val0, ptr %out, align 4
   ret void

 ; GFX11-WGP:       ; %bb.0: ; %entry
 ; GFX11-WGP-NEXT:    s_load_b128 s[0:3], s[0:1], 0x0
 ; GFX11-WGP-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-WGP-NEXT:    v_dual_mov_b32 v0, s0 :: v_dual_mov_b32 v1, s1
 ; GFX11-WGP-NEXT:    v_dual_mov_b32 v2, s2 :: v_dual_mov_b32 v3, s3
 ; GFX11-WGP-NEXT:    flat_atomic_cmpswap_b32 v2, v[0:1], v[2:3] offset:16 glc
 ; GFX11-WGP-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX11-WGP-NEXT:    flat_store_b32 v[0:1], v2
-; GFX11-WGP-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-WGP-NEXT:    s_endpgm
 ;
 ; GFX11-CU-LABEL: flat_wavefront_one_as_acq_relc_seq_cst_ret_cmpxchg:
 ; GFX11-CU:       ; %bb.0: ; %entry
 ; GFX11-CU-NEXT:    s_load_b128 s[0:3], s[0:1], 0x0
 ; GFX11-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v0, s0 :: v_dual_mov_b32 v1, s1
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v2, s2 :: v_dual_mov_b32 v3, s3
 ; GFX11-CU-NEXT:    flat_atomic_cmpswap_b32 v2, v[0:1], v[2:3] offset:16 glc
 ; GFX11-CU-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX11-CU-NEXT:    flat_store_b32 v[0:1], v2
-; GFX11-CU-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-CU-NEXT:    s_endpgm
    ptr %out, i32 %in, i32 %old) {
 entry:
   %gep = getelementptr i32, ptr %out, i32 4
   %val = cmpxchg volatile ptr %gep, i32 %old, i32 %in syncscope("wavefront-one-as") acq_rel seq_cst
   %val0 = extractvalue { i32, i1 } %val, 0
   store i32 %val0, ptr %out, align 4
   ret void

 ; GFX11-WGP:       ; %bb.0: ; %entry
 ; GFX11-WGP-NEXT:    s_load_b128 s[0:3], s[0:1], 0x0
 ; GFX11-WGP-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-WGP-NEXT:    v_dual_mov_b32 v0, s0 :: v_dual_mov_b32 v1, s1
 ; GFX11-WGP-NEXT:    v_dual_mov_b32 v2, s2 :: v_dual_mov_b32 v3, s3
 ; GFX11-WGP-NEXT:    flat_atomic_cmpswap_b32 v2, v[0:1], v[2:3] offset:16 glc
 ; GFX11-WGP-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX11-WGP-NEXT:    flat_store_b32 v[0:1], v2
-; GFX11-WGP-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-WGP-NEXT:    s_endpgm
 ;
 ; GFX11-CU-LABEL: flat_wavefront_one_as_seq_cst_seq_cst_ret_cmpxchg:
 ; GFX11-CU:       ; %bb.0: ; %entry
 ; GFX11-CU-NEXT:    s_load_b128 s[0:3], s[0:1], 0x0
 ; GFX11-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v0, s0 :: v_dual_mov_b32 v1, s1
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v2, s2 :: v_dual_mov_b32 v3, s3
 ; GFX11-CU-NEXT:    flat_atomic_cmpswap_b32 v2, v[0:1], v[2:3] offset:16 glc
 ; GFX11-CU-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX11-CU-NEXT:    flat_store_b32 v[0:1], v2
-; GFX11-CU-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-CU-NEXT:    s_endpgm
     ptr %out, i32 %in, i32 %old) {
 entry:
   %gep = getelementptr i32, ptr %out, i32 4
   %val = cmpxchg volatile ptr %gep, i32 %old, i32 %in syncscope("wavefront-one-as") seq_cst seq_cst
   %val0 = extractvalue { i32, i1 } %val, 0
   store i32 %val0, ptr %out, align 4
   ret void
 }
 

llvm/test/CodeGen/AMDGPU/memory-legalizer-flat-workgroup.ll

 ; GFX11-WGP:       ; %bb.0: ; %entry
 ; GFX11-WGP-NEXT:    s_load_b128 s[0:3], s[0:1], 0x0
 ; GFX11-WGP-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-WGP-NEXT:    v_dual_mov_b32 v0, s0 :: v_dual_mov_b32 v1, s1
 ; GFX11-WGP-NEXT:    flat_load_b32 v2, v[0:1]
 ; GFX11-WGP-NEXT:    v_dual_mov_b32 v0, s2 :: v_dual_mov_b32 v1, s3
 ; GFX11-WGP-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX11-WGP-NEXT:    flat_store_b32 v[0:1], v2
-; GFX11-WGP-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-WGP-NEXT:    s_endpgm
 ;
 ; GFX11-CU-LABEL: flat_workgroup_unordered_load:
 ; GFX11-CU:       ; %bb.0: ; %entry
 ; GFX11-CU-NEXT:    s_load_b128 s[0:3], s[0:1], 0x0
 ; GFX11-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v0, s0 :: v_dual_mov_b32 v1, s1
 ; GFX11-CU-NEXT:    flat_load_b32 v2, v[0:1]
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v0, s2 :: v_dual_mov_b32 v1, s3
 ; GFX11-CU-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX11-CU-NEXT:    flat_store_b32 v[0:1], v2
-; GFX11-CU-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-CU-NEXT:    s_endpgm
     ptr %in, ptr %out) {
 entry:
   %val = load atomic i32, ptr %in syncscope("workgroup") unordered, align 4
   store i32 %val, ptr %out
   ret void
 }
 

 ; GFX11-WGP:       ; %bb.0: ; %entry
 ; GFX11-WGP-NEXT:    s_load_b128 s[0:3], s[0:1], 0x0
 ; GFX11-WGP-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-WGP-NEXT:    v_dual_mov_b32 v0, s0 :: v_dual_mov_b32 v1, s1
 ; GFX11-WGP-NEXT:    flat_load_b32 v2, v[0:1] glc
 ; GFX11-WGP-NEXT:    v_dual_mov_b32 v0, s2 :: v_dual_mov_b32 v1, s3
 ; GFX11-WGP-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX11-WGP-NEXT:    flat_store_b32 v[0:1], v2
-; GFX11-WGP-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-WGP-NEXT:    s_endpgm
 ;
 ; GFX11-CU-LABEL: flat_workgroup_monotonic_load:
 ; GFX11-CU:       ; %bb.0: ; %entry
 ; GFX11-CU-NEXT:    s_load_b128 s[0:3], s[0:1], 0x0
 ; GFX11-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v0, s0 :: v_dual_mov_b32 v1, s1
 ; GFX11-CU-NEXT:    flat_load_b32 v2, v[0:1]
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v0, s2 :: v_dual_mov_b32 v1, s3
 ; GFX11-CU-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX11-CU-NEXT:    flat_store_b32 v[0:1], v2
-; GFX11-CU-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-CU-NEXT:    s_endpgm
     ptr %in, ptr %out) {
 entry:
   %val = load atomic i32, ptr %in syncscope("workgroup") monotonic, align 4
   store i32 %val, ptr %out
   ret void
 }
 

 ; GFX11-WGP-NEXT:    s_load_b128 s[0:3], s[0:1], 0x0
 ; GFX11-WGP-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-WGP-NEXT:    v_dual_mov_b32 v0, s0 :: v_dual_mov_b32 v1, s1
 ; GFX11-WGP-NEXT:    flat_load_b32 v2, v[0:1] glc
 ; GFX11-WGP-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX11-WGP-NEXT:    buffer_gl0_inv
 ; GFX11-WGP-NEXT:    v_dual_mov_b32 v0, s2 :: v_dual_mov_b32 v1, s3
 ; GFX11-WGP-NEXT:    flat_store_b32 v[0:1], v2
-; GFX11-WGP-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-WGP-NEXT:    s_endpgm
 ;
 ; GFX11-CU-LABEL: flat_workgroup_acquire_load:
 ; GFX11-CU:       ; %bb.0: ; %entry
 ; GFX11-CU-NEXT:    s_load_b128 s[0:3], s[0:1], 0x0
 ; GFX11-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v0, s0 :: v_dual_mov_b32 v1, s1
 ; GFX11-CU-NEXT:    flat_load_b32 v2, v[0:1]
 ; GFX11-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v0, s2 :: v_dual_mov_b32 v1, s3
 ; GFX11-CU-NEXT:    s_waitcnt vmcnt(0)
 ; GFX11-CU-NEXT:    flat_store_b32 v[0:1], v2
-; GFX11-CU-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-CU-NEXT:    s_endpgm
     ptr %in, ptr %out) {
 entry:
   %val = load atomic i32, ptr %in syncscope("workgroup") acquire, align 4
   store i32 %val, ptr %out
   ret void
 }
 

 ; GFX11-WGP-NEXT:    v_dual_mov_b32 v0, s0 :: v_dual_mov_b32 v1, s1
 ; GFX11-WGP-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX11-WGP-NEXT:    s_waitcnt_vscnt null, 0x0
 ; GFX11-WGP-NEXT:    flat_load_b32 v2, v[0:1] glc
 ; GFX11-WGP-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX11-WGP-NEXT:    buffer_gl0_inv
 ; GFX11-WGP-NEXT:    v_dual_mov_b32 v0, s2 :: v_dual_mov_b32 v1, s3
 ; GFX11-WGP-NEXT:    flat_store_b32 v[0:1], v2
-; GFX11-WGP-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-WGP-NEXT:    s_endpgm
 ;
 ; GFX11-CU-LABEL: flat_workgroup_seq_cst_load:
 ; GFX11-CU:       ; %bb.0: ; %entry
 ; GFX11-CU-NEXT:    s_load_b128 s[0:3], s[0:1], 0x0
 ; GFX11-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v0, s0 :: v_dual_mov_b32 v1, s1
 ; GFX11-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-CU-NEXT:    flat_load_b32 v2, v[0:1]
 ; GFX11-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v0, s2 :: v_dual_mov_b32 v1, s3
 ; GFX11-CU-NEXT:    s_waitcnt vmcnt(0)
 ; GFX11-CU-NEXT:    flat_store_b32 v[0:1], v2
-; GFX11-CU-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-CU-NEXT:    s_endpgm
     ptr %in, ptr %out) {
 entry:
   %val = load atomic i32, ptr %in syncscope("workgroup") seq_cst, align 4
   store i32 %val, ptr %out
   ret void
 }
 

 ; GFX11-WGP:       ; %bb.0: ; %entry
 ; GFX11-WGP-NEXT:    s_clause 0x1
 ; GFX11-WGP-NEXT:    s_load_b64 s[2:3], s[0:1], 0x8
 ; GFX11-WGP-NEXT:    s_load_b32 s0, s[0:1], 0x0
 ; GFX11-WGP-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-WGP-NEXT:    v_dual_mov_b32 v0, s2 :: v_dual_mov_b32 v1, s3
 ; GFX11-WGP-NEXT:    v_mov_b32_e32 v2, s0
 ; GFX11-WGP-NEXT:    flat_store_b32 v[0:1], v2
-; GFX11-WGP-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-WGP-NEXT:    s_endpgm
 ;
 ; GFX11-CU-LABEL: flat_workgroup_unordered_store:
 ; GFX11-CU:       ; %bb.0: ; %entry
 ; GFX11-CU-NEXT:    s_clause 0x1
 ; GFX11-CU-NEXT:    s_load_b64 s[2:3], s[0:1], 0x8
 ; GFX11-CU-NEXT:    s_load_b32 s0, s[0:1], 0x0
 ; GFX11-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v0, s2 :: v_dual_mov_b32 v1, s3
 ; GFX11-CU-NEXT:    v_mov_b32_e32 v2, s0
 ; GFX11-CU-NEXT:    flat_store_b32 v[0:1], v2
-; GFX11-CU-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-CU-NEXT:    s_endpgm
     i32 %in, ptr %out) {
 entry:
   store atomic i32 %in, ptr %out syncscope("workgroup") unordered, align 4
   ret void
 }
 
 define amdgpu_kernel void @flat_workgroup_monotonic_store(

 ; GFX11-WGP:       ; %bb.0: ; %entry
 ; GFX11-WGP-NEXT:    s_clause 0x1
 ; GFX11-WGP-NEXT:    s_load_b64 s[2:3], s[0:1], 0x8
 ; GFX11-WGP-NEXT:    s_load_b32 s0, s[0:1], 0x0
 ; GFX11-WGP-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-WGP-NEXT:    v_dual_mov_b32 v0, s2 :: v_dual_mov_b32 v1, s3
 ; GFX11-WGP-NEXT:    v_mov_b32_e32 v2, s0
 ; GFX11-WGP-NEXT:    flat_store_b32 v[0:1], v2
-; GFX11-WGP-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-WGP-NEXT:    s_endpgm
 ;
 ; GFX11-CU-LABEL: flat_workgroup_monotonic_store:
 ; GFX11-CU:       ; %bb.0: ; %entry
 ; GFX11-CU-NEXT:    s_clause 0x1
 ; GFX11-CU-NEXT:    s_load_b64 s[2:3], s[0:1], 0x8
 ; GFX11-CU-NEXT:    s_load_b32 s0, s[0:1], 0x0
 ; GFX11-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v0, s2 :: v_dual_mov_b32 v1, s3
 ; GFX11-CU-NEXT:    v_mov_b32_e32 v2, s0
 ; GFX11-CU-NEXT:    flat_store_b32 v[0:1], v2
-; GFX11-CU-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-CU-NEXT:    s_endpgm
     i32 %in, ptr %out) {
 entry:
   store atomic i32 %in, ptr %out syncscope("workgroup") monotonic, align 4
   ret void
 }
 
 define amdgpu_kernel void @flat_workgroup_release_store(

 ; GFX11-WGP-NEXT:    s_load_b64 s[2:3], s[0:1], 0x8
 ; GFX11-WGP-NEXT:    s_load_b32 s0, s[0:1], 0x0
 ; GFX11-WGP-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-WGP-NEXT:    v_dual_mov_b32 v0, s2 :: v_dual_mov_b32 v1, s3
 ; GFX11-WGP-NEXT:    v_mov_b32_e32 v2, s0
 ; GFX11-WGP-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX11-WGP-NEXT:    s_waitcnt_vscnt null, 0x0
 ; GFX11-WGP-NEXT:    flat_store_b32 v[0:1], v2
-; GFX11-WGP-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-WGP-NEXT:    s_endpgm
 ;
 ; GFX11-CU-LABEL: flat_workgroup_release_store:
 ; GFX11-CU:       ; %bb.0: ; %entry
 ; GFX11-CU-NEXT:    s_clause 0x1
 ; GFX11-CU-NEXT:    s_load_b64 s[2:3], s[0:1], 0x8
 ; GFX11-CU-NEXT:    s_load_b32 s0, s[0:1], 0x0
 ; GFX11-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v0, s2 :: v_dual_mov_b32 v1, s3
 ; GFX11-CU-NEXT:    v_mov_b32_e32 v2, s0
 ; GFX11-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-CU-NEXT:    flat_store_b32 v[0:1], v2
-; GFX11-CU-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-CU-NEXT:    s_endpgm
     i32 %in, ptr %out) {
 entry:
   store atomic i32 %in, ptr %out syncscope("workgroup") release, align 4
   ret void
 }
 
 define amdgpu_kernel void @flat_workgroup_seq_cst_store(

 ; GFX11-WGP-NEXT:    s_load_b64 s[2:3], s[0:1], 0x8
 ; GFX11-WGP-NEXT:    s_load_b32 s0, s[0:1], 0x0
 ; GFX11-WGP-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-WGP-NEXT:    v_dual_mov_b32 v0, s2 :: v_dual_mov_b32 v1, s3
 ; GFX11-WGP-NEXT:    v_mov_b32_e32 v2, s0
 ; GFX11-WGP-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX11-WGP-NEXT:    s_waitcnt_vscnt null, 0x0
 ; GFX11-WGP-NEXT:    flat_store_b32 v[0:1], v2
-; GFX11-WGP-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-WGP-NEXT:    s_endpgm
 ;
 ; GFX11-CU-LABEL: flat_workgroup_seq_cst_store:
 ; GFX11-CU:       ; %bb.0: ; %entry
 ; GFX11-CU-NEXT:    s_clause 0x1
 ; GFX11-CU-NEXT:    s_load_b64 s[2:3], s[0:1], 0x8
 ; GFX11-CU-NEXT:    s_load_b32 s0, s[0:1], 0x0
 ; GFX11-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v0, s2 :: v_dual_mov_b32 v1, s3
 ; GFX11-CU-NEXT:    v_mov_b32_e32 v2, s0
 ; GFX11-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-CU-NEXT:    flat_store_b32 v[0:1], v2
-; GFX11-CU-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-CU-NEXT:    s_endpgm
     i32 %in, ptr %out) {
 entry:
   store atomic i32 %in, ptr %out syncscope("workgroup") seq_cst, align 4
   ret void
 }
 
 define amdgpu_kernel void @flat_workgroup_monotonic_atomicrmw(

 ; GFX11-WGP:       ; %bb.0: ; %entry
 ; GFX11-WGP-NEXT:    s_clause 0x1
 ; GFX11-WGP-NEXT:    s_load_b64 s[2:3], s[0:1], 0x0
 ; GFX11-WGP-NEXT:    s_load_b32 s0, s[0:1], 0x8
 ; GFX11-WGP-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-WGP-NEXT:    v_dual_mov_b32 v0, s2 :: v_dual_mov_b32 v1, s3
 ; GFX11-WGP-NEXT:    v_mov_b32_e32 v2, s0
 ; GFX11-WGP-NEXT:    flat_atomic_swap_b32 v[0:1], v2
-; GFX11-WGP-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-WGP-NEXT:    s_endpgm
 ;
 ; GFX11-CU-LABEL: flat_workgroup_monotonic_atomicrmw:
 ; GFX11-CU:       ; %bb.0: ; %entry
 ; GFX11-CU-NEXT:    s_clause 0x1
 ; GFX11-CU-NEXT:    s_load_b64 s[2:3], s[0:1], 0x0
 ; GFX11-CU-NEXT:    s_load_b32 s0, s[0:1], 0x8
 ; GFX11-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v0, s2 :: v_dual_mov_b32 v1, s3
 ; GFX11-CU-NEXT:    v_mov_b32_e32 v2, s0
 ; GFX11-CU-NEXT:    flat_atomic_swap_b32 v[0:1], v2
-; GFX11-CU-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-CU-NEXT:    s_endpgm
     ptr %out, i32 %in) {
 entry:
   %val = atomicrmw volatile xchg ptr %out, i32 %in syncscope("workgroup") monotonic
   ret void
 }
 
 define amdgpu_kernel void @flat_workgroup_acquire_atomicrmw(

 ; GFX11-CU-NEXT:    s_clause 0x1
 ; GFX11-CU-NEXT:    s_load_b64 s[2:3], s[0:1], 0x0
 ; GFX11-CU-NEXT:    s_load_b32 s0, s[0:1], 0x8
 ; GFX11-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v0, s2 :: v_dual_mov_b32 v1, s3
 ; GFX11-CU-NEXT:    v_mov_b32_e32 v2, s0
 ; GFX11-CU-NEXT:    flat_atomic_swap_b32 v[0:1], v2
 ; GFX11-CU-NEXT:    s_waitcnt lgkmcnt(0)
-; GFX11-CU-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-CU-NEXT:    s_endpgm
     ptr %out, i32 %in) {
 entry:
   %val = atomicrmw volatile xchg ptr %out, i32 %in syncscope("workgroup") acquire
   ret void
 }
 
 define amdgpu_kernel void @flat_workgroup_release_atomicrmw(

 ; GFX11-WGP-NEXT:    s_load_b64 s[2:3], s[0:1], 0x0
 ; GFX11-WGP-NEXT:    s_load_b32 s0, s[0:1], 0x8
 ; GFX11-WGP-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-WGP-NEXT:    v_dual_mov_b32 v0, s2 :: v_dual_mov_b32 v1, s3
 ; GFX11-WGP-NEXT:    v_mov_b32_e32 v2, s0
 ; GFX11-WGP-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX11-WGP-NEXT:    s_waitcnt_vscnt null, 0x0
 ; GFX11-WGP-NEXT:    flat_atomic_swap_b32 v[0:1], v2
-; GFX11-WGP-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-WGP-NEXT:    s_endpgm
 ;
 ; GFX11-CU-LABEL: flat_workgroup_release_atomicrmw:
 ; GFX11-CU:       ; %bb.0: ; %entry
 ; GFX11-CU-NEXT:    s_clause 0x1
 ; GFX11-CU-NEXT:    s_load_b64 s[2:3], s[0:1], 0x0
 ; GFX11-CU-NEXT:    s_load_b32 s0, s[0:1], 0x8
 ; GFX11-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v0, s2 :: v_dual_mov_b32 v1, s3
 ; GFX11-CU-NEXT:    v_mov_b32_e32 v2, s0
 ; GFX11-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-CU-NEXT:    flat_atomic_swap_b32 v[0:1], v2
-; GFX11-CU-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-CU-NEXT:    s_endpgm
     ptr %out, i32 %in) {
 entry:
   %val = atomicrmw volatile xchg ptr %out, i32 %in syncscope("workgroup") release
   ret void
 }
 
 define amdgpu_kernel void @flat_workgroup_acq_rel_atomicrmw(

 ; GFX11-CU-NEXT:    s_load_b64 s[2:3], s[0:1], 0x0
 ; GFX11-CU-NEXT:    s_load_b32 s0, s[0:1], 0x8
 ; GFX11-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v0, s2 :: v_dual_mov_b32 v1, s3
 ; GFX11-CU-NEXT:    v_mov_b32_e32 v2, s0
 ; GFX11-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-CU-NEXT:    flat_atomic_swap_b32 v[0:1], v2
 ; GFX11-CU-NEXT:    s_waitcnt lgkmcnt(0)
-; GFX11-CU-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-CU-NEXT:    s_endpgm
     ptr %out, i32 %in) {
 entry:
   %val = atomicrmw volatile xchg ptr %out, i32 %in syncscope("workgroup") acq_rel
   ret void
 }
 
 define amdgpu_kernel void @flat_workgroup_seq_cst_atomicrmw(

 ; GFX11-CU-NEXT:    s_load_b64 s[2:3], s[0:1], 0x0
 ; GFX11-CU-NEXT:    s_load_b32 s0, s[0:1], 0x8
 ; GFX11-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v0, s2 :: v_dual_mov_b32 v1, s3
 ; GFX11-CU-NEXT:    v_mov_b32_e32 v2, s0
 ; GFX11-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-CU-NEXT:    flat_atomic_swap_b32 v[0:1], v2
 ; GFX11-CU-NEXT:    s_waitcnt lgkmcnt(0)
-; GFX11-CU-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-CU-NEXT:    s_endpgm
     ptr %out, i32 %in) {
 entry:
   %val = atomicrmw volatile xchg ptr %out, i32 %in syncscope("workgroup") seq_cst
   ret void
 }
 
 define amdgpu_kernel void @flat_workgroup_acquire_ret_atomicrmw(

 ; GFX11-WGP-NEXT:    s_load_b32 s0, s[0:1], 0x8
 ; GFX11-WGP-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-WGP-NEXT:    v_dual_mov_b32 v0, s2 :: v_dual_mov_b32 v1, s3
 ; GFX11-WGP-NEXT:    v_mov_b32_e32 v2, s0
 ; GFX11-WGP-NEXT:    flat_atomic_swap_b32 v2, v[0:1], v2 glc
 ; GFX11-WGP-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX11-WGP-NEXT:    buffer_gl0_inv
 ; GFX11-WGP-NEXT:    flat_store_b32 v[0:1], v2
-; GFX11-WGP-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-WGP-NEXT:    s_endpgm
 ;
 ; GFX11-CU-LABEL: flat_workgroup_acquire_ret_atomicrmw:
 ; GFX11-CU:       ; %bb.0: ; %entry
 ; GFX11-CU-NEXT:    s_clause 0x1
 ; GFX11-CU-NEXT:    s_load_b64 s[2:3], s[0:1], 0x0
 ; GFX11-CU-NEXT:    s_load_b32 s0, s[0:1], 0x8
 ; GFX11-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v0, s2 :: v_dual_mov_b32 v1, s3
 ; GFX11-CU-NEXT:    v_mov_b32_e32 v2, s0
 ; GFX11-CU-NEXT:    flat_atomic_swap_b32 v2, v[0:1], v2 glc
 ; GFX11-CU-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX11-CU-NEXT:    flat_store_b32 v[0:1], v2
-; GFX11-CU-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-CU-NEXT:    s_endpgm
     ptr %out, i32 %in) {
 entry:
   %val = atomicrmw volatile xchg ptr %out, i32 %in syncscope("workgroup") acquire
   store i32 %val, ptr %out, align 4
   ret void
 }
 

 ; GFX11-WGP-NEXT:    v_dual_mov_b32 v0, s2 :: v_dual_mov_b32 v1, s3
 ; GFX11-WGP-NEXT:    v_mov_b32_e32 v2, s0
 ; GFX11-WGP-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX11-WGP-NEXT:    s_waitcnt_vscnt null, 0x0
 ; GFX11-WGP-NEXT:    flat_atomic_swap_b32 v2, v[0:1], v2 glc
 ; GFX11-WGP-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX11-WGP-NEXT:    buffer_gl0_inv
 ; GFX11-WGP-NEXT:    flat_store_b32 v[0:1], v2
-; GFX11-WGP-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-WGP-NEXT:    s_endpgm
 ;
 ; GFX11-CU-LABEL: flat_workgroup_acq_rel_ret_atomicrmw:
 ; GFX11-CU:       ; %bb.0: ; %entry
 ; GFX11-CU-NEXT:    s_clause 0x1
 ; GFX11-CU-NEXT:    s_load_b64 s[2:3], s[0:1], 0x0
 ; GFX11-CU-NEXT:    s_load_b32 s0, s[0:1], 0x8
 ; GFX11-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v0, s2 :: v_dual_mov_b32 v1, s3
 ; GFX11-CU-NEXT:    v_mov_b32_e32 v2, s0
 ; GFX11-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-CU-NEXT:    flat_atomic_swap_b32 v2, v[0:1], v2 glc
 ; GFX11-CU-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX11-CU-NEXT:    flat_store_b32 v[0:1], v2
-; GFX11-CU-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-CU-NEXT:    s_endpgm
     ptr %out, i32 %in) {
 entry:
   %val = atomicrmw volatile xchg ptr %out, i32 %in syncscope("workgroup") acq_rel
   store i32 %val, ptr %out, align 4
   ret void
 }
 

 ; GFX11-WGP-NEXT:    v_dual_mov_b32 v0, s2 :: v_dual_mov_b32 v1, s3
 ; GFX11-WGP-NEXT:    v_mov_b32_e32 v2, s0
 ; GFX11-WGP-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX11-WGP-NEXT:    s_waitcnt_vscnt null, 0x0
 ; GFX11-WGP-NEXT:    flat_atomic_swap_b32 v2, v[0:1], v2 glc
 ; GFX11-WGP-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX11-WGP-NEXT:    buffer_gl0_inv
 ; GFX11-WGP-NEXT:    flat_store_b32 v[0:1], v2
-; GFX11-WGP-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-WGP-NEXT:    s_endpgm
 ;
 ; GFX11-CU-LABEL: flat_workgroup_seq_cst_ret_atomicrmw:
 ; GFX11-CU:       ; %bb.0: ; %entry
 ; GFX11-CU-NEXT:    s_clause 0x1
 ; GFX11-CU-NEXT:    s_load_b64 s[2:3], s[0:1], 0x0
 ; GFX11-CU-NEXT:    s_load_b32 s0, s[0:1], 0x8
 ; GFX11-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v0, s2 :: v_dual_mov_b32 v1, s3
 ; GFX11-CU-NEXT:    v_mov_b32_e32 v2, s0
 ; GFX11-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-CU-NEXT:    flat_atomic_swap_b32 v2, v[0:1], v2 glc
 ; GFX11-CU-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX11-CU-NEXT:    flat_store_b32 v[0:1], v2
-; GFX11-CU-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-CU-NEXT:    s_endpgm
     ptr %out, i32 %in) {
 entry:
   %val = atomicrmw volatile xchg ptr %out, i32 %in syncscope("workgroup") seq_cst
   store i32 %val, ptr %out, align 4
   ret void
 }
 

 ;
 ; GFX11-WGP-LABEL: flat_workgroup_monotonic_monotonic_cmpxchg:
 ; GFX11-WGP:       ; %bb.0: ; %entry
 ; GFX11-WGP-NEXT:    s_load_b128 s[0:3], s[0:1], 0x0
 ; GFX11-WGP-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-WGP-NEXT:    v_dual_mov_b32 v0, s0 :: v_dual_mov_b32 v1, s1
 ; GFX11-WGP-NEXT:    v_dual_mov_b32 v2, s2 :: v_dual_mov_b32 v3, s3
 ; GFX11-WGP-NEXT:    flat_atomic_cmpswap_b32 v[0:1], v[2:3] offset:16
-; GFX11-WGP-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-WGP-NEXT:    s_endpgm
 ;
 ; GFX11-CU-LABEL: flat_workgroup_monotonic_monotonic_cmpxchg:
 ; GFX11-CU:       ; %bb.0: ; %entry
 ; GFX11-CU-NEXT:    s_load_b128 s[0:3], s[0:1], 0x0
 ; GFX11-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v0, s0 :: v_dual_mov_b32 v1, s1
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v2, s2 :: v_dual_mov_b32 v3, s3
 ; GFX11-CU-NEXT:    flat_atomic_cmpswap_b32 v[0:1], v[2:3] offset:16
-; GFX11-CU-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-CU-NEXT:    s_endpgm
     ptr %out, i32 %in, i32 %old) {
 entry:
   %gep = getelementptr i32, ptr %out, i32 4
   %val = cmpxchg volatile ptr %gep, i32 %old, i32 %in syncscope("workgroup") monotonic monotonic
   ret void
 }
 

 ; GFX11-CU-LABEL: flat_workgroup_acquire_monotonic_cmpxchg:
 ; GFX11-CU:       ; %bb.0: ; %entry
 ; GFX11-CU-NEXT:    s_load_b128 s[0:3], s[0:1], 0x0
 ; GFX11-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v0, s0 :: v_dual_mov_b32 v1, s1
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v2, s2 :: v_dual_mov_b32 v3, s3
 ; GFX11-CU-NEXT:    flat_atomic_cmpswap_b32 v[0:1], v[2:3] offset:16
 ; GFX11-CU-NEXT:    s_waitcnt lgkmcnt(0)
-; GFX11-CU-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-CU-NEXT:    s_endpgm
     ptr %out, i32 %in, i32 %old) {
 entry:
   %gep = getelementptr i32, ptr %out, i32 4
   %val = cmpxchg volatile ptr %gep, i32 %old, i32 %in syncscope("workgroup") acquire monotonic
   ret void
 }
 

 ; GFX11-WGP:       ; %bb.0: ; %entry
 ; GFX11-WGP-NEXT:    s_load_b128 s[0:3], s[0:1], 0x0
 ; GFX11-WGP-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-WGP-NEXT:    v_dual_mov_b32 v0, s0 :: v_dual_mov_b32 v1, s1
 ; GFX11-WGP-NEXT:    v_dual_mov_b32 v2, s2 :: v_dual_mov_b32 v3, s3
 ; GFX11-WGP-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX11-WGP-NEXT:    s_waitcnt_vscnt null, 0x0
 ; GFX11-WGP-NEXT:    flat_atomic_cmpswap_b32 v[0:1], v[2:3] offset:16
-; GFX11-WGP-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-WGP-NEXT:    s_endpgm
 ;
 ; GFX11-CU-LABEL: flat_workgroup_release_monotonic_cmpxchg:
 ; GFX11-CU:       ; %bb.0: ; %entry
 ; GFX11-CU-NEXT:    s_load_b128 s[0:3], s[0:1], 0x0
 ; GFX11-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v0, s0 :: v_dual_mov_b32 v1, s1
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v2, s2 :: v_dual_mov_b32 v3, s3
 ; GFX11-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-CU-NEXT:    flat_atomic_cmpswap_b32 v[0:1], v[2:3] offset:16
-; GFX11-CU-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-CU-NEXT:    s_endpgm
     ptr %out, i32 %in, i32 %old) {
 entry:
   %gep = getelementptr i32, ptr %out, i32 4
   %val = cmpxchg volatile ptr %gep, i32 %old, i32 %in syncscope("workgroup") release monotonic
   ret void
 }
 

 ; GFX11-CU:       ; %bb.0: ; %entry
 ; GFX11-CU-NEXT:    s_load_b128 s[0:3], s[0:1], 0x0
 ; GFX11-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v0, s0 :: v_dual_mov_b32 v1, s1
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v2, s2 :: v_dual_mov_b32 v3, s3
 ; GFX11-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-CU-NEXT:    flat_atomic_cmpswap_b32 v[0:1], v[2:3] offset:16
 ; GFX11-CU-NEXT:    s_waitcnt lgkmcnt(0)
-; GFX11-CU-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-CU-NEXT:    s_endpgm
     ptr %out, i32 %in, i32 %old) {
 entry:
   %gep = getelementptr i32, ptr %out, i32 4
   %val = cmpxchg volatile ptr %gep, i32 %old, i32 %in syncscope("workgroup") acq_rel monotonic
   ret void
 }
 

 ; GFX11-CU:       ; %bb.0: ; %entry
 ; GFX11-CU-NEXT:    s_load_b128 s[0:3], s[0:1], 0x0
 ; GFX11-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v0, s0 :: v_dual_mov_b32 v1, s1
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v2, s2 :: v_dual_mov_b32 v3, s3
 ; GFX11-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-CU-NEXT:    flat_atomic_cmpswap_b32 v[0:1], v[2:3] offset:16
 ; GFX11-CU-NEXT:    s_waitcnt lgkmcnt(0)
-; GFX11-CU-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-CU-NEXT:    s_endpgm
     ptr %out, i32 %in, i32 %old) {
 entry:
   %gep = getelementptr i32, ptr %out, i32 4
   %val = cmpxchg volatile ptr %gep, i32 %old, i32 %in syncscope("workgroup") seq_cst monotonic
   ret void
 }
 

 ; GFX11-CU-LABEL: flat_workgroup_monotonic_acquire_cmpxchg:
 ; GFX11-CU:       ; %bb.0: ; %entry
 ; GFX11-CU-NEXT:    s_load_b128 s[0:3], s[0:1], 0x0
 ; GFX11-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v0, s0 :: v_dual_mov_b32 v1, s1
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v2, s2 :: v_dual_mov_b32 v3, s3
 ; GFX11-CU-NEXT:    flat_atomic_cmpswap_b32 v[0:1], v[2:3] offset:16
 ; GFX11-CU-NEXT:    s_waitcnt lgkmcnt(0)
-; GFX11-CU-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-CU-NEXT:    s_endpgm
     ptr %out, i32 %in, i32 %old) {
 entry:
   %gep = getelementptr i32, ptr %out, i32 4
   %val = cmpxchg volatile ptr %gep, i32 %old, i32 %in syncscope("workgroup") monotonic acquire
   ret void
 }
 

 ; GFX11-CU-LABEL: flat_workgroup_acquire_acquire_cmpxchg:
 ; GFX11-CU:       ; %bb.0: ; %entry
 ; GFX11-CU-NEXT:    s_load_b128 s[0:3], s[0:1], 0x0
 ; GFX11-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v0, s0 :: v_dual_mov_b32 v1, s1
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v2, s2 :: v_dual_mov_b32 v3, s3
 ; GFX11-CU-NEXT:    flat_atomic_cmpswap_b32 v[0:1], v[2:3] offset:16
 ; GFX11-CU-NEXT:    s_waitcnt lgkmcnt(0)
-; GFX11-CU-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-CU-NEXT:    s_endpgm
     ptr %out, i32 %in, i32 %old) {
 entry:
   %gep = getelementptr i32, ptr %out, i32 4
   %val = cmpxchg volatile ptr %gep, i32 %old, i32 %in syncscope("workgroup") acquire acquire
   ret void
 }
 

 ; GFX11-CU:       ; %bb.0: ; %entry
 ; GFX11-CU-NEXT:    s_load_b128 s[0:3], s[0:1], 0x0
 ; GFX11-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v0, s0 :: v_dual_mov_b32 v1, s1
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v2, s2 :: v_dual_mov_b32 v3, s3
 ; GFX11-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-CU-NEXT:    flat_atomic_cmpswap_b32 v[0:1], v[2:3] offset:16
 ; GFX11-CU-NEXT:    s_waitcnt lgkmcnt(0)
-; GFX11-CU-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-CU-NEXT:    s_endpgm
     ptr %out, i32 %in, i32 %old) {
 entry:
   %gep = getelementptr i32, ptr %out, i32 4
   %val = cmpxchg volatile ptr %gep, i32 %old, i32 %in syncscope("workgroup") release acquire
   ret void
 }
 

 ; GFX11-CU:       ; %bb.0: ; %entry
 ; GFX11-CU-NEXT:    s_load_b128 s[0:3], s[0:1], 0x0
 ; GFX11-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v0, s0 :: v_dual_mov_b32 v1, s1
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v2, s2 :: v_dual_mov_b32 v3, s3
 ; GFX11-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-CU-NEXT:    flat_atomic_cmpswap_b32 v[0:1], v[2:3] offset:16
 ; GFX11-CU-NEXT:    s_waitcnt lgkmcnt(0)
-; GFX11-CU-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-CU-NEXT:    s_endpgm
     ptr %out, i32 %in, i32 %old) {
 entry:
   %gep = getelementptr i32, ptr %out, i32 4
   %val = cmpxchg volatile ptr %gep, i32 %old, i32 %in syncscope("workgroup") acq_rel acquire
   ret void
 }
 

 ; GFX11-CU:       ; %bb.0: ; %entry
 ; GFX11-CU-NEXT:    s_load_b128 s[0:3], s[0:1], 0x0
 ; GFX11-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v0, s0 :: v_dual_mov_b32 v1, s1
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v2, s2 :: v_dual_mov_b32 v3, s3
 ; GFX11-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-CU-NEXT:    flat_atomic_cmpswap_b32 v[0:1], v[2:3] offset:16
 ; GFX11-CU-NEXT:    s_waitcnt lgkmcnt(0)
-; GFX11-CU-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-CU-NEXT:    s_endpgm
     ptr %out, i32 %in, i32 %old) {
 entry:
   %gep = getelementptr i32, ptr %out, i32 4
   %val = cmpxchg volatile ptr %gep, i32 %old, i32 %in syncscope("workgroup") seq_cst acquire
   ret void
 }
 

 ; GFX11-CU:       ; %bb.0: ; %entry
 ; GFX11-CU-NEXT:    s_load_b128 s[0:3], s[0:1], 0x0
 ; GFX11-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v0, s0 :: v_dual_mov_b32 v1, s1
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v2, s2 :: v_dual_mov_b32 v3, s3
 ; GFX11-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-CU-NEXT:    flat_atomic_cmpswap_b32 v[0:1], v[2:3] offset:16
 ; GFX11-CU-NEXT:    s_waitcnt lgkmcnt(0)
-; GFX11-CU-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-CU-NEXT:    s_endpgm
     ptr %out, i32 %in, i32 %old) {
 entry:
   %gep = getelementptr i32, ptr %out, i32 4
   %val = cmpxchg volatile ptr %gep, i32 %old, i32 %in syncscope("workgroup") seq_cst seq_cst
   ret void
 }
 

 ; GFX11-WGP:       ; %bb.0: ; %entry
 ; GFX11-WGP-NEXT:    s_load_b128 s[0:3], s[0:1], 0x0
 ; GFX11-WGP-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-WGP-NEXT:    v_dual_mov_b32 v0, s0 :: v_dual_mov_b32 v1, s1
 ; GFX11-WGP-NEXT:    v_dual_mov_b32 v2, s2 :: v_dual_mov_b32 v3, s3
 ; GFX11-WGP-NEXT:    flat_atomic_cmpswap_b32 v2, v[0:1], v[2:3] offset:16 glc
 ; GFX11-WGP-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX11-WGP-NEXT:    flat_store_b32 v[0:1], v2
-; GFX11-WGP-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-WGP-NEXT:    s_endpgm
 ;
 ; GFX11-CU-LABEL: flat_workgroup_monotonic_monotonic_ret_cmpxchg:
 ; GFX11-CU:       ; %bb.0: ; %entry
 ; GFX11-CU-NEXT:    s_load_b128 s[0:3], s[0:1], 0x0
 ; GFX11-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v0, s0 :: v_dual_mov_b32 v1, s1
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v2, s2 :: v_dual_mov_b32 v3, s3
 ; GFX11-CU-NEXT:    flat_atomic_cmpswap_b32 v2, v[0:1], v[2:3] offset:16 glc
 ; GFX11-CU-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX11-CU-NEXT:    flat_store_b32 v[0:1], v2
-; GFX11-CU-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-CU-NEXT:    s_endpgm
     ptr %out, i32 %in, i32 %old) {
 entry:
   %gep = getelementptr i32, ptr %out, i32 4
   %val = cmpxchg volatile ptr %gep, i32 %old, i32 %in syncscope("workgroup") monotonic monotonic
   %val0 = extractvalue { i32, i1 } %val, 0
   store i32 %val0, ptr %out, align 4
   ret void

 ; GFX11-WGP-NEXT:    s_load_b128 s[0:3], s[0:1], 0x0
 ; GFX11-WGP-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-WGP-NEXT:    v_dual_mov_b32 v0, s0 :: v_dual_mov_b32 v1, s1
 ; GFX11-WGP-NEXT:    v_dual_mov_b32 v2, s2 :: v_dual_mov_b32 v3, s3
 ; GFX11-WGP-NEXT:    flat_atomic_cmpswap_b32 v2, v[0:1], v[2:3] offset:16 glc
 ; GFX11-WGP-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX11-WGP-NEXT:    buffer_gl0_inv
 ; GFX11-WGP-NEXT:    flat_store_b32 v[0:1], v2
-; GFX11-WGP-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-WGP-NEXT:    s_endpgm
 ;
 ; GFX11-CU-LABEL: flat_workgroup_acquire_monotonic_ret_cmpxchg:
 ; GFX11-CU:       ; %bb.0: ; %entry
 ; GFX11-CU-NEXT:    s_load_b128 s[0:3], s[0:1], 0x0
 ; GFX11-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v0, s0 :: v_dual_mov_b32 v1, s1
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v2, s2 :: v_dual_mov_b32 v3, s3
 ; GFX11-CU-NEXT:    flat_atomic_cmpswap_b32 v2, v[0:1], v[2:3] offset:16 glc
 ; GFX11-CU-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX11-CU-NEXT:    flat_store_b32 v[0:1], v2
-; GFX11-CU-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-CU-NEXT:    s_endpgm
     ptr %out, i32 %in, i32 %old) {
 entry:
   %gep = getelementptr i32, ptr %out, i32 4
   %val = cmpxchg volatile ptr %gep, i32 %old, i32 %in syncscope("workgroup") acquire monotonic
   %val0 = extractvalue { i32, i1 } %val, 0
   store i32 %val0, ptr %out, align 4
   ret void

 ; GFX11-WGP-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-WGP-NEXT:    v_dual_mov_b32 v0, s0 :: v_dual_mov_b32 v1, s1
 ; GFX11-WGP-NEXT:    v_dual_mov_b32 v2, s2 :: v_dual_mov_b32 v3, s3
 ; GFX11-WGP-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX11-WGP-NEXT:    s_waitcnt_vscnt null, 0x0
 ; GFX11-WGP-NEXT:    flat_atomic_cmpswap_b32 v2, v[0:1], v[2:3] offset:16 glc
 ; GFX11-WGP-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX11-WGP-NEXT:    flat_store_b32 v[0:1], v2
-; GFX11-WGP-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-WGP-NEXT:    s_endpgm
 ;
 ; GFX11-CU-LABEL: flat_workgroup_release_monotonic_ret_cmpxchg:
 ; GFX11-CU:       ; %bb.0: ; %entry
 ; GFX11-CU-NEXT:    s_load_b128 s[0:3], s[0:1], 0x0
 ; GFX11-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v0, s0 :: v_dual_mov_b32 v1, s1
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v2, s2 :: v_dual_mov_b32 v3, s3
 ; GFX11-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-CU-NEXT:    flat_atomic_cmpswap_b32 v2, v[0:1], v[2:3] offset:16 glc
 ; GFX11-CU-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX11-CU-NEXT:    flat_store_b32 v[0:1], v2
-; GFX11-CU-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-CU-NEXT:    s_endpgm
     ptr %out, i32 %in, i32 %old) {
 entry:
   %gep = getelementptr i32, ptr %out, i32 4
   %val = cmpxchg volatile ptr %gep, i32 %old, i32 %in syncscope("workgroup") release monotonic
   %val0 = extractvalue { i32, i1 } %val, 0
   store i32 %val0, ptr %out, align 4
   ret void

 ; GFX11-WGP-NEXT:    v_dual_mov_b32 v0, s0 :: v_dual_mov_b32 v1, s1
 ; GFX11-WGP-NEXT:    v_dual_mov_b32 v2, s2 :: v_dual_mov_b32 v3, s3
 ; GFX11-WGP-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX11-WGP-NEXT:    s_waitcnt_vscnt null, 0x0
 ; GFX11-WGP-NEXT:    flat_atomic_cmpswap_b32 v2, v[0:1], v[2:3] offset:16 glc
 ; GFX11-WGP-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX11-WGP-NEXT:    buffer_gl0_inv
 ; GFX11-WGP-NEXT:    flat_store_b32 v[0:1], v2
-; GFX11-WGP-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-WGP-NEXT:    s_endpgm
 ;
 ; GFX11-CU-LABEL: flat_workgroup_acq_rel_monotonic_ret_cmpxchg:
 ; GFX11-CU:       ; %bb.0: ; %entry
 ; GFX11-CU-NEXT:    s_load_b128 s[0:3], s[0:1], 0x0
 ; GFX11-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v0, s0 :: v_dual_mov_b32 v1, s1
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v2, s2 :: v_dual_mov_b32 v3, s3
 ; GFX11-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-CU-NEXT:    flat_atomic_cmpswap_b32 v2, v[0:1], v[2:3] offset:16 glc
 ; GFX11-CU-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX11-CU-NEXT:    flat_store_b32 v[0:1], v2
-; GFX11-CU-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-CU-NEXT:    s_endpgm
     ptr %out, i32 %in, i32 %old) {
 entry:
   %gep = getelementptr i32, ptr %out, i32 4
   %val = cmpxchg volatile ptr %gep, i32 %old, i32 %in syncscope("workgroup") acq_rel monotonic
   %val0 = extractvalue { i32, i1 } %val, 0
   store i32 %val0, ptr %out, align 4
   ret void

 ; GFX11-WGP-NEXT:    v_dual_mov_b32 v0, s0 :: v_dual_mov_b32 v1, s1
 ; GFX11-WGP-NEXT:    v_dual_mov_b32 v2, s2 :: v_dual_mov_b32 v3, s3
 ; GFX11-WGP-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX11-WGP-NEXT:    s_waitcnt_vscnt null, 0x0
 ; GFX11-WGP-NEXT:    flat_atomic_cmpswap_b32 v2, v[0:1], v[2:3] offset:16 glc
 ; GFX11-WGP-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX11-WGP-NEXT:    buffer_gl0_inv
 ; GFX11-WGP-NEXT:    flat_store_b32 v[0:1], v2
-; GFX11-WGP-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-WGP-NEXT:    s_endpgm
 ;
 ; GFX11-CU-LABEL: flat_workgroup_seq_cst_monotonic_ret_cmpxchg:
 ; GFX11-CU:       ; %bb.0: ; %entry
 ; GFX11-CU-NEXT:    s_load_b128 s[0:3], s[0:1], 0x0
 ; GFX11-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v0, s0 :: v_dual_mov_b32 v1, s1
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v2, s2 :: v_dual_mov_b32 v3, s3
 ; GFX11-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-CU-NEXT:    flat_atomic_cmpswap_b32 v2, v[0:1], v[2:3] offset:16 glc
 ; GFX11-CU-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX11-CU-NEXT:    flat_store_b32 v[0:1], v2
-; GFX11-CU-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-CU-NEXT:    s_endpgm
     ptr %out, i32 %in, i32 %old) {
 entry:
   %gep = getelementptr i32, ptr %out, i32 4
   %val = cmpxchg volatile ptr %gep, i32 %old, i32 %in syncscope("workgroup") seq_cst monotonic
   %val0 = extractvalue { i32, i1 } %val, 0
   store i32 %val0, ptr %out, align 4
   ret void

 ; GFX11-WGP-NEXT:    s_load_b128 s[0:3], s[0:1], 0x0
 ; GFX11-WGP-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-WGP-NEXT:    v_dual_mov_b32 v0, s0 :: v_dual_mov_b32 v1, s1
 ; GFX11-WGP-NEXT:    v_dual_mov_b32 v2, s2 :: v_dual_mov_b32 v3, s3
 ; GFX11-WGP-NEXT:    flat_atomic_cmpswap_b32 v2, v[0:1], v[2:3] offset:16 glc
 ; GFX11-WGP-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX11-WGP-NEXT:    buffer_gl0_inv
 ; GFX11-WGP-NEXT:    flat_store_b32 v[0:1], v2
-; GFX11-WGP-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-WGP-NEXT:    s_endpgm
 ;
 ; GFX11-CU-LABEL: flat_workgroup_monotonic_acquire_ret_cmpxchg:
 ; GFX11-CU:       ; %bb.0: ; %entry
 ; GFX11-CU-NEXT:    s_load_b128 s[0:3], s[0:1], 0x0
 ; GFX11-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v0, s0 :: v_dual_mov_b32 v1, s1
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v2, s2 :: v_dual_mov_b32 v3, s3
 ; GFX11-CU-NEXT:    flat_atomic_cmpswap_b32 v2, v[0:1], v[2:3] offset:16 glc
 ; GFX11-CU-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX11-CU-NEXT:    flat_store_b32 v[0:1], v2
-; GFX11-CU-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-CU-NEXT:    s_endpgm
     ptr %out, i32 %in, i32 %old) {
 entry:
   %gep = getelementptr i32, ptr %out, i32 4
   %val = cmpxchg volatile ptr %gep, i32 %old, i32 %in syncscope("workgroup") monotonic acquire
   %val0 = extractvalue { i32, i1 } %val, 0
   store i32 %val0, ptr %out, align 4
   ret void

 ; GFX11-WGP-NEXT:    s_load_b128 s[0:3], s[0:1], 0x0
 ; GFX11-WGP-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-WGP-NEXT:    v_dual_mov_b32 v0, s0 :: v_dual_mov_b32 v1, s1
 ; GFX11-WGP-NEXT:    v_dual_mov_b32 v2, s2 :: v_dual_mov_b32 v3, s3
 ; GFX11-WGP-NEXT:    flat_atomic_cmpswap_b32 v2, v[0:1], v[2:3] offset:16 glc
 ; GFX11-WGP-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX11-WGP-NEXT:    buffer_gl0_inv
 ; GFX11-WGP-NEXT:    flat_store_b32 v[0:1], v2
-; GFX11-WGP-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-WGP-NEXT:    s_endpgm
 ;
 ; GFX11-CU-LABEL: flat_workgroup_acquire_acquire_ret_cmpxchg:
 ; GFX11-CU:       ; %bb.0: ; %entry
 ; GFX11-CU-NEXT:    s_load_b128 s[0:3], s[0:1], 0x0
 ; GFX11-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v0, s0 :: v_dual_mov_b32 v1, s1
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v2, s2 :: v_dual_mov_b32 v3, s3
 ; GFX11-CU-NEXT:    flat_atomic_cmpswap_b32 v2, v[0:1], v[2:3] offset:16 glc
 ; GFX11-CU-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX11-CU-NEXT:    flat_store_b32 v[0:1], v2
-; GFX11-CU-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-CU-NEXT:    s_endpgm
     ptr %out, i32 %in, i32 %old) {
 entry:
   %gep = getelementptr i32, ptr %out, i32 4
   %val = cmpxchg volatile ptr %gep, i32 %old, i32 %in syncscope("workgroup") acquire acquire
   %val0 = extractvalue { i32, i1 } %val, 0
   store i32 %val0, ptr %out, align 4
   ret void

 ; GFX11-WGP-NEXT:    v_dual_mov_b32 v0, s0 :: v_dual_mov_b32 v1, s1
 ; GFX11-WGP-NEXT:    v_dual_mov_b32 v2, s2 :: v_dual_mov_b32 v3, s3
 ; GFX11-WGP-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX11-WGP-NEXT:    s_waitcnt_vscnt null, 0x0
 ; GFX11-WGP-NEXT:    flat_atomic_cmpswap_b32 v2, v[0:1], v[2:3] offset:16 glc
 ; GFX11-WGP-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX11-WGP-NEXT:    buffer_gl0_inv
 ; GFX11-WGP-NEXT:    flat_store_b32 v[0:1], v2
-; GFX11-WGP-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-WGP-NEXT:    s_endpgm
 ;
 ; GFX11-CU-LABEL: flat_workgroup_release_acquire_ret_cmpxchg:
 ; GFX11-CU:       ; %bb.0: ; %entry
 ; GFX11-CU-NEXT:    s_load_b128 s[0:3], s[0:1], 0x0
 ; GFX11-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v0, s0 :: v_dual_mov_b32 v1, s1
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v2, s2 :: v_dual_mov_b32 v3, s3
 ; GFX11-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-CU-NEXT:    flat_atomic_cmpswap_b32 v2, v[0:1], v[2:3] offset:16 glc
 ; GFX11-CU-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX11-CU-NEXT:    flat_store_b32 v[0:1], v2
-; GFX11-CU-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-CU-NEXT:    s_endpgm
     ptr %out, i32 %in, i32 %old) {
 entry:
   %gep = getelementptr i32, ptr %out, i32 4
   %val = cmpxchg volatile ptr %gep, i32 %old, i32 %in syncscope("workgroup") release acquire
   %val0 = extractvalue { i32, i1 } %val, 0
   store i32 %val0, ptr %out, align 4
   ret void

 ; GFX11-WGP-NEXT:    v_dual_mov_b32 v0, s0 :: v_dual_mov_b32 v1, s1
 ; GFX11-WGP-NEXT:    v_dual_mov_b32 v2, s2 :: v_dual_mov_b32 v3, s3
 ; GFX11-WGP-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX11-WGP-NEXT:    s_waitcnt_vscnt null, 0x0
 ; GFX11-WGP-NEXT:    flat_atomic_cmpswap_b32 v2, v[0:1], v[2:3] offset:16 glc
 ; GFX11-WGP-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX11-WGP-NEXT:    buffer_gl0_inv
 ; GFX11-WGP-NEXT:    flat_store_b32 v[0:1], v2
-; GFX11-WGP-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-WGP-NEXT:    s_endpgm
 ;
 ; GFX11-CU-LABEL: flat_workgroup_acq_rel_acquire_ret_cmpxchg:
 ; GFX11-CU:       ; %bb.0: ; %entry
 ; GFX11-CU-NEXT:    s_load_b128 s[0:3], s[0:1], 0x0
 ; GFX11-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v0, s0 :: v_dual_mov_b32 v1, s1
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v2, s2 :: v_dual_mov_b32 v3, s3
 ; GFX11-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-CU-NEXT:    flat_atomic_cmpswap_b32 v2, v[0:1], v[2:3] offset:16 glc
 ; GFX11-CU-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX11-CU-NEXT:    flat_store_b32 v[0:1], v2
-; GFX11-CU-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-CU-NEXT:    s_endpgm
     ptr %out, i32 %in, i32 %old) {
 entry:
   %gep = getelementptr i32, ptr %out, i32 4
   %val = cmpxchg volatile ptr %gep, i32 %old, i32 %in syncscope("workgroup") acq_rel acquire
   %val0 = extractvalue { i32, i1 } %val, 0
   store i32 %val0, ptr %out, align 4
   ret void

 ; GFX11-WGP-NEXT:    v_dual_mov_b32 v0, s0 :: v_dual_mov_b32 v1, s1
 ; GFX11-WGP-NEXT:    v_dual_mov_b32 v2, s2 :: v_dual_mov_b32 v3, s3
 ; GFX11-WGP-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX11-WGP-NEXT:    s_waitcnt_vscnt null, 0x0
 ; GFX11-WGP-NEXT:    flat_atomic_cmpswap_b32 v2, v[0:1], v[2:3] offset:16 glc
 ; GFX11-WGP-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX11-WGP-NEXT:    buffer_gl0_inv
 ; GFX11-WGP-NEXT:    flat_store_b32 v[0:1], v2
-; GFX11-WGP-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-WGP-NEXT:    s_endpgm
 ;
 ; GFX11-CU-LABEL: flat_workgroup_seq_cst_acquire_ret_cmpxchg:
 ; GFX11-CU:       ; %bb.0: ; %entry
 ; GFX11-CU-NEXT:    s_load_b128 s[0:3], s[0:1], 0x0
 ; GFX11-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v0, s0 :: v_dual_mov_b32 v1, s1
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v2, s2 :: v_dual_mov_b32 v3, s3
 ; GFX11-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-CU-NEXT:    flat_atomic_cmpswap_b32 v2, v[0:1], v[2:3] offset:16 glc
 ; GFX11-CU-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX11-CU-NEXT:    flat_store_b32 v[0:1], v2
-; GFX11-CU-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-CU-NEXT:    s_endpgm
     ptr %out, i32 %in, i32 %old) {
 entry:
   %gep = getelementptr i32, ptr %out, i32 4
   %val = cmpxchg volatile ptr %gep, i32 %old, i32 %in syncscope("workgroup") seq_cst acquire
   %val0 = extractvalue { i32, i1 } %val, 0
   store i32 %val0, ptr %out, align 4
   ret void

 ; GFX11-WGP-NEXT:    v_dual_mov_b32 v0, s0 :: v_dual_mov_b32 v1, s1
 ; GFX11-WGP-NEXT:    v_dual_mov_b32 v2, s2 :: v_dual_mov_b32 v3, s3
 ; GFX11-WGP-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX11-WGP-NEXT:    s_waitcnt_vscnt null, 0x0
 ; GFX11-WGP-NEXT:    flat_atomic_cmpswap_b32 v2, v[0:1], v[2:3] offset:16 glc
 ; GFX11-WGP-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX11-WGP-NEXT:    buffer_gl0_inv
 ; GFX11-WGP-NEXT:    flat_store_b32 v[0:1], v2
-; GFX11-WGP-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-WGP-NEXT:    s_endpgm
 ;
 ; GFX11-CU-LABEL: flat_workgroup_monotonic_seq_cst_ret_cmpxchg:
 ; GFX11-CU:       ; %bb.0: ; %entry
 ; GFX11-CU-NEXT:    s_load_b128 s[0:3], s[0:1], 0x0
 ; GFX11-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v0, s0 :: v_dual_mov_b32 v1, s1
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v2, s2 :: v_dual_mov_b32 v3, s3
 ; GFX11-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-CU-NEXT:    flat_atomic_cmpswap_b32 v2, v[0:1], v[2:3] offset:16 glc
 ; GFX11-CU-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX11-CU-NEXT:    flat_store_b32 v[0:1], v2
-; GFX11-CU-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-CU-NEXT:    s_endpgm
     ptr %out, i32 %in, i32 %old) {
 entry:
   %gep = getelementptr i32, ptr %out, i32 4
   %val = cmpxchg volatile ptr %gep, i32 %old, i32 %in syncscope("workgroup") monotonic seq_cst
   %val0 = extractvalue { i32, i1 } %val, 0
   store i32 %val0, ptr %out, align 4
   ret void

 ; GFX11-WGP-NEXT:    v_dual_mov_b32 v0, s0 :: v_dual_mov_b32 v1, s1
 ; GFX11-WGP-NEXT:    v_dual_mov_b32 v2, s2 :: v_dual_mov_b32 v3, s3
 ; GFX11-WGP-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX11-WGP-NEXT:    s_waitcnt_vscnt null, 0x0
 ; GFX11-WGP-NEXT:    flat_atomic_cmpswap_b32 v2, v[0:1], v[2:3] offset:16 glc
 ; GFX11-WGP-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX11-WGP-NEXT:    buffer_gl0_inv
 ; GFX11-WGP-NEXT:    flat_store_b32 v[0:1], v2
-; GFX11-WGP-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-WGP-NEXT:    s_endpgm
 ;
 ; GFX11-CU-LABEL: flat_workgroup_acquire_seq_cst_ret_cmpxchg:
 ; GFX11-CU:       ; %bb.0: ; %entry
 ; GFX11-CU-NEXT:    s_load_b128 s[0:3], s[0:1], 0x0
 ; GFX11-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v0, s0 :: v_dual_mov_b32 v1, s1
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v2, s2 :: v_dual_mov_b32 v3, s3
 ; GFX11-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-CU-NEXT:    flat_atomic_cmpswap_b32 v2, v[0:1], v[2:3] offset:16 glc
 ; GFX11-CU-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX11-CU-NEXT:    flat_store_b32 v[0:1], v2
-; GFX11-CU-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-CU-NEXT:    s_endpgm
     ptr %out, i32 %in, i32 %old) {
 entry:
   %gep = getelementptr i32, ptr %out, i32 4
   %val = cmpxchg volatile ptr %gep, i32 %old, i32 %in syncscope("workgroup") acquire seq_cst
   %val0 = extractvalue { i32, i1 } %val, 0
   store i32 %val0, ptr %out, align 4
   ret void

 ; GFX11-WGP-NEXT:    v_dual_mov_b32 v0, s0 :: v_dual_mov_b32 v1, s1
 ; GFX11-WGP-NEXT:    v_dual_mov_b32 v2, s2 :: v_dual_mov_b32 v3, s3
 ; GFX11-WGP-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX11-WGP-NEXT:    s_waitcnt_vscnt null, 0x0
 ; GFX11-WGP-NEXT:    flat_atomic_cmpswap_b32 v2, v[0:1], v[2:3] offset:16 glc
 ; GFX11-WGP-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX11-WGP-NEXT:    buffer_gl0_inv
 ; GFX11-WGP-NEXT:    flat_store_b32 v[0:1], v2
-; GFX11-WGP-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-WGP-NEXT:    s_endpgm
 ;
 ; GFX11-CU-LABEL: flat_workgroup_release_seq_cst_ret_cmpxchg:
 ; GFX11-CU:       ; %bb.0: ; %entry
 ; GFX11-CU-NEXT:    s_load_b128 s[0:3], s[0:1], 0x0
 ; GFX11-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v0, s0 :: v_dual_mov_b32 v1, s1
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v2, s2 :: v_dual_mov_b32 v3, s3
 ; GFX11-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-CU-NEXT:    flat_atomic_cmpswap_b32 v2, v[0:1], v[2:3] offset:16 glc
 ; GFX11-CU-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX11-CU-NEXT:    flat_store_b32 v[0:1], v2
-; GFX11-CU-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-CU-NEXT:    s_endpgm
     ptr %out, i32 %in, i32 %old) {
 entry:
   %gep = getelementptr i32, ptr %out, i32 4
   %val = cmpxchg volatile ptr %gep, i32 %old, i32 %in syncscope("workgroup") release seq_cst
   %val0 = extractvalue { i32, i1 } %val, 0
   store i32 %val0, ptr %out, align 4
   ret void

 ; GFX11-WGP-NEXT:    v_dual_mov_b32 v0, s0 :: v_dual_mov_b32 v1, s1
 ; GFX11-WGP-NEXT:    v_dual_mov_b32 v2, s2 :: v_dual_mov_b32 v3, s3
 ; GFX11-WGP-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX11-WGP-NEXT:    s_waitcnt_vscnt null, 0x0
 ; GFX11-WGP-NEXT:    flat_atomic_cmpswap_b32 v2, v[0:1], v[2:3] offset:16 glc
 ; GFX11-WGP-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX11-WGP-NEXT:    buffer_gl0_inv
 ; GFX11-WGP-NEXT:    flat_store_b32 v[0:1], v2
-; GFX11-WGP-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-WGP-NEXT:    s_endpgm
 ;
 ; GFX11-CU-LABEL: flat_workgroup_acq_rel_seq_cst_ret_cmpxchg:
 ; GFX11-CU:       ; %bb.0: ; %entry
 ; GFX11-CU-NEXT:    s_load_b128 s[0:3], s[0:1], 0x0
 ; GFX11-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v0, s0 :: v_dual_mov_b32 v1, s1
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v2, s2 :: v_dual_mov_b32 v3, s3
 ; GFX11-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-CU-NEXT:    flat_atomic_cmpswap_b32 v2, v[0:1], v[2:3] offset:16 glc
 ; GFX11-CU-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX11-CU-NEXT:    flat_store_b32 v[0:1], v2
-; GFX11-CU-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-CU-NEXT:    s_endpgm
     ptr %out, i32 %in, i32 %old) {
 entry:
   %gep = getelementptr i32, ptr %out, i32 4
   %val = cmpxchg volatile ptr %gep, i32 %old, i32 %in syncscope("workgroup") acq_rel seq_cst
   %val0 = extractvalue { i32, i1 } %val, 0
   store i32 %val0, ptr %out, align 4
   ret void

 ; GFX11-WGP-NEXT:    v_dual_mov_b32 v0, s0 :: v_dual_mov_b32 v1, s1
 ; GFX11-WGP-NEXT:    v_dual_mov_b32 v2, s2 :: v_dual_mov_b32 v3, s3
 ; GFX11-WGP-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX11-WGP-NEXT:    s_waitcnt_vscnt null, 0x0
 ; GFX11-WGP-NEXT:    flat_atomic_cmpswap_b32 v2, v[0:1], v[2:3] offset:16 glc
 ; GFX11-WGP-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX11-WGP-NEXT:    buffer_gl0_inv
 ; GFX11-WGP-NEXT:    flat_store_b32 v[0:1], v2
-; GFX11-WGP-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-WGP-NEXT:    s_endpgm
 ;
 ; GFX11-CU-LABEL: flat_workgroup_seq_cst_seq_cst_ret_cmpxchg:
 ; GFX11-CU:       ; %bb.0: ; %entry
 ; GFX11-CU-NEXT:    s_load_b128 s[0:3], s[0:1], 0x0
 ; GFX11-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v0, s0 :: v_dual_mov_b32 v1, s1
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v2, s2 :: v_dual_mov_b32 v3, s3
 ; GFX11-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-CU-NEXT:    flat_atomic_cmpswap_b32 v2, v[0:1], v[2:3] offset:16 glc
 ; GFX11-CU-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX11-CU-NEXT:    flat_store_b32 v[0:1], v2
-; GFX11-CU-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-CU-NEXT:    s_endpgm
     ptr %out, i32 %in, i32 %old) {
 entry:
   %gep = getelementptr i32, ptr %out, i32 4
   %val = cmpxchg volatile ptr %gep, i32 %old, i32 %in syncscope("workgroup") seq_cst seq_cst
   %val0 = extractvalue { i32, i1 } %val, 0
   store i32 %val0, ptr %out, align 4
   ret void

 ; GFX11-WGP:       ; %bb.0: ; %entry
 ; GFX11-WGP-NEXT:    s_load_b128 s[0:3], s[0:1], 0x0
 ; GFX11-WGP-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-WGP-NEXT:    v_dual_mov_b32 v0, s0 :: v_dual_mov_b32 v1, s1
 ; GFX11-WGP-NEXT:    flat_load_b32 v2, v[0:1]
 ; GFX11-WGP-NEXT:    v_dual_mov_b32 v0, s2 :: v_dual_mov_b32 v1, s3
 ; GFX11-WGP-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX11-WGP-NEXT:    flat_store_b32 v[0:1], v2
-; GFX11-WGP-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-WGP-NEXT:    s_endpgm
 ;
 ; GFX11-CU-LABEL: flat_workgroup_one_as_unordered_load:
 ; GFX11-CU:       ; %bb.0: ; %entry
 ; GFX11-CU-NEXT:    s_load_b128 s[0:3], s[0:1], 0x0
 ; GFX11-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v0, s0 :: v_dual_mov_b32 v1, s1
 ; GFX11-CU-NEXT:    flat_load_b32 v2, v[0:1]
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v0, s2 :: v_dual_mov_b32 v1, s3
 ; GFX11-CU-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX11-CU-NEXT:    flat_store_b32 v[0:1], v2
-; GFX11-CU-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-CU-NEXT:    s_endpgm
     ptr %in, ptr %out) {
 entry:
   %val = load atomic i32, ptr %in syncscope("workgroup-one-as") unordered, align 4
   store i32 %val, ptr %out
   ret void
 }
 

 ; GFX11-WGP:       ; %bb.0: ; %entry
 ; GFX11-WGP-NEXT:    s_load_b128 s[0:3], s[0:1], 0x0
 ; GFX11-WGP-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-WGP-NEXT:    v_dual_mov_b32 v0, s0 :: v_dual_mov_b32 v1, s1
 ; GFX11-WGP-NEXT:    flat_load_b32 v2, v[0:1] glc
 ; GFX11-WGP-NEXT:    v_dual_mov_b32 v0, s2 :: v_dual_mov_b32 v1, s3
 ; GFX11-WGP-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX11-WGP-NEXT:    flat_store_b32 v[0:1], v2
-; GFX11-WGP-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-WGP-NEXT:    s_endpgm
 ;
 ; GFX11-CU-LABEL: flat_workgroup_one_as_monotonic_load:
 ; GFX11-CU:       ; %bb.0: ; %entry
 ; GFX11-CU-NEXT:    s_load_b128 s[0:3], s[0:1], 0x0
 ; GFX11-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v0, s0 :: v_dual_mov_b32 v1, s1
 ; GFX11-CU-NEXT:    flat_load_b32 v2, v[0:1]
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v0, s2 :: v_dual_mov_b32 v1, s3
 ; GFX11-CU-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX11-CU-NEXT:    flat_store_b32 v[0:1], v2
-; GFX11-CU-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-CU-NEXT:    s_endpgm
     ptr %in, ptr %out) {
 entry:
   %val = load atomic i32, ptr %in syncscope("workgroup-one-as") monotonic, align 4
   store i32 %val, ptr %out
   ret void
 }
 

 ; GFX11-WGP-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-WGP-NEXT:    v_dual_mov_b32 v0, s0 :: v_dual_mov_b32 v1, s1
 ; GFX11-WGP-NEXT:    flat_load_b32 v2, v[0:1] glc
 ; GFX11-WGP-NEXT:    s_waitcnt vmcnt(0)
 ; GFX11-WGP-NEXT:    buffer_gl0_inv
 ; GFX11-WGP-NEXT:    v_dual_mov_b32 v0, s2 :: v_dual_mov_b32 v1, s3
 ; GFX11-WGP-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-WGP-NEXT:    flat_store_b32 v[0:1], v2
-; GFX11-WGP-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-WGP-NEXT:    s_endpgm
 ;
 ; GFX11-CU-LABEL: flat_workgroup_one_as_acquire_load:
 ; GFX11-CU:       ; %bb.0: ; %entry
 ; GFX11-CU-NEXT:    s_load_b128 s[0:3], s[0:1], 0x0
 ; GFX11-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v0, s0 :: v_dual_mov_b32 v1, s1
 ; GFX11-CU-NEXT:    flat_load_b32 v2, v[0:1]
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v0, s2 :: v_dual_mov_b32 v1, s3
 ; GFX11-CU-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX11-CU-NEXT:    flat_store_b32 v[0:1], v2
-; GFX11-CU-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-CU-NEXT:    s_endpgm
     ptr %in, ptr %out) {
 entry:
   %val = load atomic i32, ptr %in syncscope("workgroup-one-as") acquire, align 4
   store i32 %val, ptr %out
   ret void
 }
 

 ; GFX11-WGP-NEXT:    s_waitcnt vmcnt(0)
 ; GFX11-WGP-NEXT:    s_waitcnt_vscnt null, 0x0
 ; GFX11-WGP-NEXT:    flat_load_b32 v2, v[0:1] glc
 ; GFX11-WGP-NEXT:    s_waitcnt vmcnt(0)
 ; GFX11-WGP-NEXT:    buffer_gl0_inv
 ; GFX11-WGP-NEXT:    v_dual_mov_b32 v0, s2 :: v_dual_mov_b32 v1, s3
 ; GFX11-WGP-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-WGP-NEXT:    flat_store_b32 v[0:1], v2
-; GFX11-WGP-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-WGP-NEXT:    s_endpgm
 ;
 ; GFX11-CU-LABEL: flat_workgroup_one_as_seq_cst_load:
 ; GFX11-CU:       ; %bb.0: ; %entry
 ; GFX11-CU-NEXT:    s_load_b128 s[0:3], s[0:1], 0x0
 ; GFX11-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v0, s0 :: v_dual_mov_b32 v1, s1
 ; GFX11-CU-NEXT:    flat_load_b32 v2, v[0:1]
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v0, s2 :: v_dual_mov_b32 v1, s3
 ; GFX11-CU-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX11-CU-NEXT:    flat_store_b32 v[0:1], v2
-; GFX11-CU-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-CU-NEXT:    s_endpgm
     ptr %in, ptr %out) {
 entry:
   %val = load atomic i32, ptr %in syncscope("workgroup-one-as") seq_cst, align 4
   store i32 %val, ptr %out
   ret void
 }
 

 ; GFX11-WGP:       ; %bb.0: ; %entry
 ; GFX11-WGP-NEXT:    s_clause 0x1
 ; GFX11-WGP-NEXT:    s_load_b64 s[2:3], s[0:1], 0x8
 ; GFX11-WGP-NEXT:    s_load_b32 s0, s[0:1], 0x0
 ; GFX11-WGP-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-WGP-NEXT:    v_dual_mov_b32 v0, s2 :: v_dual_mov_b32 v1, s3
 ; GFX11-WGP-NEXT:    v_mov_b32_e32 v2, s0
 ; GFX11-WGP-NEXT:    flat_store_b32 v[0:1], v2
-; GFX11-WGP-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-WGP-NEXT:    s_endpgm
 ;
 ; GFX11-CU-LABEL: flat_workgroup_one_as_unordered_store:
 ; GFX11-CU:       ; %bb.0: ; %entry
 ; GFX11-CU-NEXT:    s_clause 0x1
 ; GFX11-CU-NEXT:    s_load_b64 s[2:3], s[0:1], 0x8
 ; GFX11-CU-NEXT:    s_load_b32 s0, s[0:1], 0x0
 ; GFX11-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v0, s2 :: v_dual_mov_b32 v1, s3
 ; GFX11-CU-NEXT:    v_mov_b32_e32 v2, s0
 ; GFX11-CU-NEXT:    flat_store_b32 v[0:1], v2
-; GFX11-CU-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-CU-NEXT:    s_endpgm
     i32 %in, ptr %out) {
 entry:
   store atomic i32 %in, ptr %out syncscope("workgroup-one-as") unordered, align 4
   ret void
 }
 
 define amdgpu_kernel void @flat_workgroup_one_as_monotonic_store(

 ; GFX11-WGP:       ; %bb.0: ; %entry
 ; GFX11-WGP-NEXT:    s_clause 0x1
 ; GFX11-WGP-NEXT:    s_load_b64 s[2:3], s[0:1], 0x8
 ; GFX11-WGP-NEXT:    s_load_b32 s0, s[0:1], 0x0
 ; GFX11-WGP-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-WGP-NEXT:    v_dual_mov_b32 v0, s2 :: v_dual_mov_b32 v1, s3
 ; GFX11-WGP-NEXT:    v_mov_b32_e32 v2, s0
 ; GFX11-WGP-NEXT:    flat_store_b32 v[0:1], v2
-; GFX11-WGP-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-WGP-NEXT:    s_endpgm
 ;
 ; GFX11-CU-LABEL: flat_workgroup_one_as_monotonic_store:
 ; GFX11-CU:       ; %bb.0: ; %entry
 ; GFX11-CU-NEXT:    s_clause 0x1
 ; GFX11-CU-NEXT:    s_load_b64 s[2:3], s[0:1], 0x8
 ; GFX11-CU-NEXT:    s_load_b32 s0, s[0:1], 0x0
 ; GFX11-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v0, s2 :: v_dual_mov_b32 v1, s3
 ; GFX11-CU-NEXT:    v_mov_b32_e32 v2, s0
 ; GFX11-CU-NEXT:    flat_store_b32 v[0:1], v2
-; GFX11-CU-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-CU-NEXT:    s_endpgm
     i32 %in, ptr %out) {
 entry:
   store atomic i32 %in, ptr %out syncscope("workgroup-one-as") monotonic, align 4
   ret void
 }
 
 define amdgpu_kernel void @flat_workgroup_one_as_release_store(

 ; GFX11-WGP-NEXT:    s_load_b64 s[2:3], s[0:1], 0x8
 ; GFX11-WGP-NEXT:    s_load_b32 s0, s[0:1], 0x0
 ; GFX11-WGP-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-WGP-NEXT:    v_dual_mov_b32 v0, s2 :: v_dual_mov_b32 v1, s3
 ; GFX11-WGP-NEXT:    v_mov_b32_e32 v2, s0
 ; GFX11-WGP-NEXT:    s_waitcnt vmcnt(0)
 ; GFX11-WGP-NEXT:    s_waitcnt_vscnt null, 0x0
 ; GFX11-WGP-NEXT:    flat_store_b32 v[0:1], v2
-; GFX11-WGP-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-WGP-NEXT:    s_endpgm
 ;
 ; GFX11-CU-LABEL: flat_workgroup_one_as_release_store:
 ; GFX11-CU:       ; %bb.0: ; %entry
 ; GFX11-CU-NEXT:    s_clause 0x1
 ; GFX11-CU-NEXT:    s_load_b64 s[2:3], s[0:1], 0x8
 ; GFX11-CU-NEXT:    s_load_b32 s0, s[0:1], 0x0
 ; GFX11-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v0, s2 :: v_dual_mov_b32 v1, s3
 ; GFX11-CU-NEXT:    v_mov_b32_e32 v2, s0
 ; GFX11-CU-NEXT:    flat_store_b32 v[0:1], v2
-; GFX11-CU-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-CU-NEXT:    s_endpgm
     i32 %in, ptr %out) {
 entry:
   store atomic i32 %in, ptr %out syncscope("workgroup-one-as") release, align 4
   ret void
 }
 
 define amdgpu_kernel void @flat_workgroup_one_as_seq_cst_store(

 ; GFX11-WGP-NEXT:    s_load_b64 s[2:3], s[0:1], 0x8
 ; GFX11-WGP-NEXT:    s_load_b32 s0, s[0:1], 0x0
 ; GFX11-WGP-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-WGP-NEXT:    v_dual_mov_b32 v0, s2 :: v_dual_mov_b32 v1, s3
 ; GFX11-WGP-NEXT:    v_mov_b32_e32 v2, s0
 ; GFX11-WGP-NEXT:    s_waitcnt vmcnt(0)
 ; GFX11-WGP-NEXT:    s_waitcnt_vscnt null, 0x0
 ; GFX11-WGP-NEXT:    flat_store_b32 v[0:1], v2
-; GFX11-WGP-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-WGP-NEXT:    s_endpgm
 ;
 ; GFX11-CU-LABEL: flat_workgroup_one_as_seq_cst_store:
 ; GFX11-CU:       ; %bb.0: ; %entry
 ; GFX11-CU-NEXT:    s_clause 0x1
 ; GFX11-CU-NEXT:    s_load_b64 s[2:3], s[0:1], 0x8
 ; GFX11-CU-NEXT:    s_load_b32 s0, s[0:1], 0x0
 ; GFX11-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v0, s2 :: v_dual_mov_b32 v1, s3
 ; GFX11-CU-NEXT:    v_mov_b32_e32 v2, s0
 ; GFX11-CU-NEXT:    flat_store_b32 v[0:1], v2
-; GFX11-CU-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-CU-NEXT:    s_endpgm
     i32 %in, ptr %out) {
 entry:
   store atomic i32 %in, ptr %out syncscope("workgroup-one-as") seq_cst, align 4
   ret void
 }
 
 define amdgpu_kernel void @flat_workgroup_one_as_monotonic_atomicrmw(

 ; GFX11-WGP:       ; %bb.0: ; %entry
 ; GFX11-WGP-NEXT:    s_clause 0x1
 ; GFX11-WGP-NEXT:    s_load_b64 s[2:3], s[0:1], 0x0
 ; GFX11-WGP-NEXT:    s_load_b32 s0, s[0:1], 0x8
 ; GFX11-WGP-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-WGP-NEXT:    v_dual_mov_b32 v0, s2 :: v_dual_mov_b32 v1, s3
 ; GFX11-WGP-NEXT:    v_mov_b32_e32 v2, s0
 ; GFX11-WGP-NEXT:    flat_atomic_swap_b32 v[0:1], v2
-; GFX11-WGP-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-WGP-NEXT:    s_endpgm
 ;
 ; GFX11-CU-LABEL: flat_workgroup_one_as_monotonic_atomicrmw:
 ; GFX11-CU:       ; %bb.0: ; %entry
 ; GFX11-CU-NEXT:    s_clause 0x1
 ; GFX11-CU-NEXT:    s_load_b64 s[2:3], s[0:1], 0x0
 ; GFX11-CU-NEXT:    s_load_b32 s0, s[0:1], 0x8
 ; GFX11-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v0, s2 :: v_dual_mov_b32 v1, s3
 ; GFX11-CU-NEXT:    v_mov_b32_e32 v2, s0
 ; GFX11-CU-NEXT:    flat_atomic_swap_b32 v[0:1], v2
-; GFX11-CU-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-CU-NEXT:    s_endpgm
     ptr %out, i32 %in) {
 entry:
   %val = atomicrmw volatile xchg ptr %out, i32 %in syncscope("workgroup-one-as") monotonic
   ret void
 }
 
 define amdgpu_kernel void @flat_workgroup_one_as_acquire_atomicrmw(

 ; GFX11-CU:       ; %bb.0: ; %entry
 ; GFX11-CU-NEXT:    s_clause 0x1
 ; GFX11-CU-NEXT:    s_load_b64 s[2:3], s[0:1], 0x0
 ; GFX11-CU-NEXT:    s_load_b32 s0, s[0:1], 0x8
 ; GFX11-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v0, s2 :: v_dual_mov_b32 v1, s3
 ; GFX11-CU-NEXT:    v_mov_b32_e32 v2, s0
 ; GFX11-CU-NEXT:    flat_atomic_swap_b32 v[0:1], v2
-; GFX11-CU-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-CU-NEXT:    s_endpgm
     ptr %out, i32 %in) {
 entry:
   %val = atomicrmw volatile xchg ptr %out, i32 %in syncscope("workgroup-one-as") acquire
   ret void
 }
 
 define amdgpu_kernel void @flat_workgroup_one_as_release_atomicrmw(

 ; GFX11-WGP-NEXT:    s_load_b64 s[2:3], s[0:1], 0x0
 ; GFX11-WGP-NEXT:    s_load_b32 s0, s[0:1], 0x8
 ; GFX11-WGP-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-WGP-NEXT:    v_dual_mov_b32 v0, s2 :: v_dual_mov_b32 v1, s3
 ; GFX11-WGP-NEXT:    v_mov_b32_e32 v2, s0
 ; GFX11-WGP-NEXT:    s_waitcnt vmcnt(0)
 ; GFX11-WGP-NEXT:    s_waitcnt_vscnt null, 0x0
 ; GFX11-WGP-NEXT:    flat_atomic_swap_b32 v[0:1], v2
-; GFX11-WGP-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-WGP-NEXT:    s_endpgm
 ;
 ; GFX11-CU-LABEL: flat_workgroup_one_as_release_atomicrmw:
 ; GFX11-CU:       ; %bb.0: ; %entry
 ; GFX11-CU-NEXT:    s_clause 0x1
 ; GFX11-CU-NEXT:    s_load_b64 s[2:3], s[0:1], 0x0
 ; GFX11-CU-NEXT:    s_load_b32 s0, s[0:1], 0x8
 ; GFX11-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v0, s2 :: v_dual_mov_b32 v1, s3
 ; GFX11-CU-NEXT:    v_mov_b32_e32 v2, s0
 ; GFX11-CU-NEXT:    flat_atomic_swap_b32 v[0:1], v2
-; GFX11-CU-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-CU-NEXT:    s_endpgm
     ptr %out, i32 %in) {
 entry:
   %val = atomicrmw volatile xchg ptr %out, i32 %in syncscope("workgroup-one-as") release
   ret void
 }
 
 define amdgpu_kernel void @flat_workgroup_one_as_acq_rel_atomicrmw(

 ; GFX11-CU:       ; %bb.0: ; %entry
 ; GFX11-CU-NEXT:    s_clause 0x1
 ; GFX11-CU-NEXT:    s_load_b64 s[2:3], s[0:1], 0x0
 ; GFX11-CU-NEXT:    s_load_b32 s0, s[0:1], 0x8
 ; GFX11-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v0, s2 :: v_dual_mov_b32 v1, s3
 ; GFX11-CU-NEXT:    v_mov_b32_e32 v2, s0
 ; GFX11-CU-NEXT:    flat_atomic_swap_b32 v[0:1], v2
-; GFX11-CU-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-CU-NEXT:    s_endpgm
     ptr %out, i32 %in) {
 entry:
   %val = atomicrmw volatile xchg ptr %out, i32 %in syncscope("workgroup-one-as") acq_rel
   ret void
 }
 
 define amdgpu_kernel void @flat_workgroup_one_as_seq_cst_atomicrmw(

 ; GFX11-CU:       ; %bb.0: ; %entry
 ; GFX11-CU-NEXT:    s_clause 0x1
 ; GFX11-CU-NEXT:    s_load_b64 s[2:3], s[0:1], 0x0
 ; GFX11-CU-NEXT:    s_load_b32 s0, s[0:1], 0x8
 ; GFX11-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v0, s2 :: v_dual_mov_b32 v1, s3
 ; GFX11-CU-NEXT:    v_mov_b32_e32 v2, s0
 ; GFX11-CU-NEXT:    flat_atomic_swap_b32 v[0:1], v2
-; GFX11-CU-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-CU-NEXT:    s_endpgm
     ptr %out, i32 %in) {
 entry:
   %val = atomicrmw volatile xchg ptr %out, i32 %in syncscope("workgroup-one-as") seq_cst
   ret void
 }
 
 define amdgpu_kernel void @flat_workgroup_one_as_acquire_ret_atomicrmw(

 ; GFX11-WGP-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-WGP-NEXT:    v_dual_mov_b32 v0, s2 :: v_dual_mov_b32 v1, s3
 ; GFX11-WGP-NEXT:    v_mov_b32_e32 v2, s0
 ; GFX11-WGP-NEXT:    flat_atomic_swap_b32 v2, v[0:1], v2 glc
 ; GFX11-WGP-NEXT:    s_waitcnt vmcnt(0)
 ; GFX11-WGP-NEXT:    buffer_gl0_inv
 ; GFX11-WGP-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-WGP-NEXT:    flat_store_b32 v[0:1], v2
-; GFX11-WGP-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-WGP-NEXT:    s_endpgm
 ;
 ; GFX11-CU-LABEL: flat_workgroup_one_as_acquire_ret_atomicrmw:
 ; GFX11-CU:       ; %bb.0: ; %entry
 ; GFX11-CU-NEXT:    s_clause 0x1
 ; GFX11-CU-NEXT:    s_load_b64 s[2:3], s[0:1], 0x0
 ; GFX11-CU-NEXT:    s_load_b32 s0, s[0:1], 0x8
 ; GFX11-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v0, s2 :: v_dual_mov_b32 v1, s3
 ; GFX11-CU-NEXT:    v_mov_b32_e32 v2, s0
 ; GFX11-CU-NEXT:    flat_atomic_swap_b32 v2, v[0:1], v2 glc
 ; GFX11-CU-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX11-CU-NEXT:    flat_store_b32 v[0:1], v2
-; GFX11-CU-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-CU-NEXT:    s_endpgm
     ptr %out, i32 %in) {
 entry:
   %val = atomicrmw volatile xchg ptr %out, i32 %in syncscope("workgroup-one-as") acquire
   store i32 %val, ptr %out, align 4
   ret void
 }
 

 ; GFX11-WGP-NEXT:    v_mov_b32_e32 v2, s0
 ; GFX11-WGP-NEXT:    s_waitcnt vmcnt(0)
 ; GFX11-WGP-NEXT:    s_waitcnt_vscnt null, 0x0
 ; GFX11-WGP-NEXT:    flat_atomic_swap_b32 v2, v[0:1], v2 glc
 ; GFX11-WGP-NEXT:    s_waitcnt vmcnt(0)
 ; GFX11-WGP-NEXT:    buffer_gl0_inv
 ; GFX11-WGP-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-WGP-NEXT:    flat_store_b32 v[0:1], v2
-; GFX11-WGP-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-WGP-NEXT:    s_endpgm
 ;
 ; GFX11-CU-LABEL: flat_workgroup_one_as_acq_rel_ret_atomicrmw:
 ; GFX11-CU:       ; %bb.0: ; %entry
 ; GFX11-CU-NEXT:    s_clause 0x1
 ; GFX11-CU-NEXT:    s_load_b64 s[2:3], s[0:1], 0x0
 ; GFX11-CU-NEXT:    s_load_b32 s0, s[0:1], 0x8
 ; GFX11-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v0, s2 :: v_dual_mov_b32 v1, s3
 ; GFX11-CU-NEXT:    v_mov_b32_e32 v2, s0
 ; GFX11-CU-NEXT:    flat_atomic_swap_b32 v2, v[0:1], v2 glc
 ; GFX11-CU-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX11-CU-NEXT:    flat_store_b32 v[0:1], v2
-; GFX11-CU-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-CU-NEXT:    s_endpgm
     ptr %out, i32 %in) {
 entry:
   %val = atomicrmw volatile xchg ptr %out, i32 %in syncscope("workgroup-one-as") acq_rel
   store i32 %val, ptr %out, align 4
   ret void
 }
 

 ; GFX11-WGP-NEXT:    v_mov_b32_e32 v2, s0
 ; GFX11-WGP-NEXT:    s_waitcnt vmcnt(0)
 ; GFX11-WGP-NEXT:    s_waitcnt_vscnt null, 0x0
 ; GFX11-WGP-NEXT:    flat_atomic_swap_b32 v2, v[0:1], v2 glc
 ; GFX11-WGP-NEXT:    s_waitcnt vmcnt(0)
 ; GFX11-WGP-NEXT:    buffer_gl0_inv
 ; GFX11-WGP-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-WGP-NEXT:    flat_store_b32 v[0:1], v2
-; GFX11-WGP-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-WGP-NEXT:    s_endpgm
 ;
 ; GFX11-CU-LABEL: flat_workgroup_one_as_seq_cst_ret_atomicrmw:
 ; GFX11-CU:       ; %bb.0: ; %entry
 ; GFX11-CU-NEXT:    s_clause 0x1
 ; GFX11-CU-NEXT:    s_load_b64 s[2:3], s[0:1], 0x0
 ; GFX11-CU-NEXT:    s_load_b32 s0, s[0:1], 0x8
 ; GFX11-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v0, s2 :: v_dual_mov_b32 v1, s3
 ; GFX11-CU-NEXT:    v_mov_b32_e32 v2, s0
 ; GFX11-CU-NEXT:    flat_atomic_swap_b32 v2, v[0:1], v2 glc
 ; GFX11-CU-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX11-CU-NEXT:    flat_store_b32 v[0:1], v2
-; GFX11-CU-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-CU-NEXT:    s_endpgm
     ptr %out, i32 %in) {
 entry:
   %val = atomicrmw volatile xchg ptr %out, i32 %in syncscope("workgroup-one-as") seq_cst
   store i32 %val, ptr %out, align 4
   ret void
 }
 

 ;
 ; GFX11-WGP-LABEL: flat_workgroup_one_as_monotonic_monotonic_cmpxchg:
 ; GFX11-WGP:       ; %bb.0: ; %entry
 ; GFX11-WGP-NEXT:    s_load_b128 s[0:3], s[0:1], 0x0
 ; GFX11-WGP-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-WGP-NEXT:    v_dual_mov_b32 v0, s0 :: v_dual_mov_b32 v1, s1
 ; GFX11-WGP-NEXT:    v_dual_mov_b32 v2, s2 :: v_dual_mov_b32 v3, s3
 ; GFX11-WGP-NEXT:    flat_atomic_cmpswap_b32 v[0:1], v[2:3] offset:16
-; GFX11-WGP-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-WGP-NEXT:    s_endpgm
 ;
 ; GFX11-CU-LABEL: flat_workgroup_one_as_monotonic_monotonic_cmpxchg:
 ; GFX11-CU:       ; %bb.0: ; %entry
 ; GFX11-CU-NEXT:    s_load_b128 s[0:3], s[0:1], 0x0
 ; GFX11-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v0, s0 :: v_dual_mov_b32 v1, s1
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v2, s2 :: v_dual_mov_b32 v3, s3
 ; GFX11-CU-NEXT:    flat_atomic_cmpswap_b32 v[0:1], v[2:3] offset:16
-; GFX11-CU-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-CU-NEXT:    s_endpgm
     ptr %out, i32 %in, i32 %old) {
 entry:
   %gep = getelementptr i32, ptr %out, i32 4
   %val = cmpxchg volatile ptr %gep, i32 %old, i32 %in syncscope("workgroup-one-as") monotonic monotonic
   ret void
 }
 

 ;
 ; GFX11-CU-LABEL: flat_workgroup_one_as_acquire_monotonic_cmpxchg:
 ; GFX11-CU:       ; %bb.0: ; %entry
 ; GFX11-CU-NEXT:    s_load_b128 s[0:3], s[0:1], 0x0
 ; GFX11-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v0, s0 :: v_dual_mov_b32 v1, s1
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v2, s2 :: v_dual_mov_b32 v3, s3
 ; GFX11-CU-NEXT:    flat_atomic_cmpswap_b32 v[0:1], v[2:3] offset:16
-; GFX11-CU-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-CU-NEXT:    s_endpgm
     ptr %out, i32 %in, i32 %old) {
 entry:
   %gep = getelementptr i32, ptr %out, i32 4
   %val = cmpxchg volatile ptr %gep, i32 %old, i32 %in syncscope("workgroup-one-as") acquire monotonic
   ret void
 }
 

 ; GFX11-WGP:       ; %bb.0: ; %entry
 ; GFX11-WGP-NEXT:    s_load_b128 s[0:3], s[0:1], 0x0
 ; GFX11-WGP-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-WGP-NEXT:    v_dual_mov_b32 v0, s0 :: v_dual_mov_b32 v1, s1
 ; GFX11-WGP-NEXT:    v_dual_mov_b32 v2, s2 :: v_dual_mov_b32 v3, s3
 ; GFX11-WGP-NEXT:    s_waitcnt vmcnt(0)
 ; GFX11-WGP-NEXT:    s_waitcnt_vscnt null, 0x0
 ; GFX11-WGP-NEXT:    flat_atomic_cmpswap_b32 v[0:1], v[2:3] offset:16
-; GFX11-WGP-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-WGP-NEXT:    s_endpgm
 ;
 ; GFX11-CU-LABEL: flat_workgroup_one_as_release_monotonic_cmpxchg:
 ; GFX11-CU:       ; %bb.0: ; %entry
 ; GFX11-CU-NEXT:    s_load_b128 s[0:3], s[0:1], 0x0
 ; GFX11-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v0, s0 :: v_dual_mov_b32 v1, s1
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v2, s2 :: v_dual_mov_b32 v3, s3
 ; GFX11-CU-NEXT:    flat_atomic_cmpswap_b32 v[0:1], v[2:3] offset:16
-; GFX11-CU-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-CU-NEXT:    s_endpgm
     ptr %out, i32 %in, i32 %old) {
 entry:
   %gep = getelementptr i32, ptr %out, i32 4
   %val = cmpxchg volatile ptr %gep, i32 %old, i32 %in syncscope("workgroup-one-as") release monotonic
   ret void
 }
 

 ;
 ; GFX11-CU-LABEL: flat_workgroup_one_as_acq_rel_monotonic_cmpxchg:
 ; GFX11-CU:       ; %bb.0: ; %entry
 ; GFX11-CU-NEXT:    s_load_b128 s[0:3], s[0:1], 0x0
 ; GFX11-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v0, s0 :: v_dual_mov_b32 v1, s1
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v2, s2 :: v_dual_mov_b32 v3, s3
 ; GFX11-CU-NEXT:    flat_atomic_cmpswap_b32 v[0:1], v[2:3] offset:16
-; GFX11-CU-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-CU-NEXT:    s_endpgm
     ptr %out, i32 %in, i32 %old) {
 entry:
   %gep = getelementptr i32, ptr %out, i32 4
   %val = cmpxchg volatile ptr %gep, i32 %old, i32 %in syncscope("workgroup-one-as") acq_rel monotonic
   ret void
 }
 

 ;
 ; GFX11-CU-LABEL: flat_workgroup_one_as_seq_cst_monotonic_cmpxchg:
 ; GFX11-CU:       ; %bb.0: ; %entry
 ; GFX11-CU-NEXT:    s_load_b128 s[0:3], s[0:1], 0x0
 ; GFX11-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v0, s0 :: v_dual_mov_b32 v1, s1
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v2, s2 :: v_dual_mov_b32 v3, s3
 ; GFX11-CU-NEXT:    flat_atomic_cmpswap_b32 v[0:1], v[2:3] offset:16
-; GFX11-CU-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-CU-NEXT:    s_endpgm
     ptr %out, i32 %in, i32 %old) {
 entry:
   %gep = getelementptr i32, ptr %out, i32 4
   %val = cmpxchg volatile ptr %gep, i32 %old, i32 %in syncscope("workgroup-one-as") seq_cst monotonic
   ret void
 }
 

 ;
 ; GFX11-CU-LABEL: flat_workgroup_one_as_monotonic_acquire_cmpxchg:
 ; GFX11-CU:       ; %bb.0: ; %entry
 ; GFX11-CU-NEXT:    s_load_b128 s[0:3], s[0:1], 0x0
 ; GFX11-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v0, s0 :: v_dual_mov_b32 v1, s1
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v2, s2 :: v_dual_mov_b32 v3, s3
 ; GFX11-CU-NEXT:    flat_atomic_cmpswap_b32 v[0:1], v[2:3] offset:16
-; GFX11-CU-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-CU-NEXT:    s_endpgm
     ptr %out, i32 %in, i32 %old) {
 entry:
   %gep = getelementptr i32, ptr %out, i32 4
   %val = cmpxchg volatile ptr %gep, i32 %old, i32 %in syncscope("workgroup-one-as") monotonic acquire
   ret void
 }
 

 ;
 ; GFX11-CU-LABEL: flat_workgroup_one_as_acquire_acquire_cmpxchg:
 ; GFX11-CU:       ; %bb.0: ; %entry
 ; GFX11-CU-NEXT:    s_load_b128 s[0:3], s[0:1], 0x0
 ; GFX11-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v0, s0 :: v_dual_mov_b32 v1, s1
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v2, s2 :: v_dual_mov_b32 v3, s3
 ; GFX11-CU-NEXT:    flat_atomic_cmpswap_b32 v[0:1], v[2:3] offset:16
-; GFX11-CU-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-CU-NEXT:    s_endpgm
     ptr %out, i32 %in, i32 %old) {
 entry:
   %gep = getelementptr i32, ptr %out, i32 4
   %val = cmpxchg volatile ptr %gep, i32 %old, i32 %in syncscope("workgroup-one-as") acquire acquire
   ret void
 }
 

 ;
 ; GFX11-CU-LABEL: flat_workgroup_one_as_release_acquire_cmpxchg:
 ; GFX11-CU:       ; %bb.0: ; %entry
 ; GFX11-CU-NEXT:    s_load_b128 s[0:3], s[0:1], 0x0
 ; GFX11-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v0, s0 :: v_dual_mov_b32 v1, s1
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v2, s2 :: v_dual_mov_b32 v3, s3
 ; GFX11-CU-NEXT:    flat_atomic_cmpswap_b32 v[0:1], v[2:3] offset:16
-; GFX11-CU-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-CU-NEXT:    s_endpgm
     ptr %out, i32 %in, i32 %old) {
 entry:
   %gep = getelementptr i32, ptr %out, i32 4
   %val = cmpxchg volatile ptr %gep, i32 %old, i32 %in syncscope("workgroup-one-as") release acquire
   ret void
 }
 

 ;
 ; GFX11-CU-LABEL: flat_workgroup_one_as_acq_rel_acquire_cmpxchg:
 ; GFX11-CU:       ; %bb.0: ; %entry
 ; GFX11-CU-NEXT:    s_load_b128 s[0:3], s[0:1], 0x0
 ; GFX11-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v0, s0 :: v_dual_mov_b32 v1, s1
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v2, s2 :: v_dual_mov_b32 v3, s3
 ; GFX11-CU-NEXT:    flat_atomic_cmpswap_b32 v[0:1], v[2:3] offset:16
-; GFX11-CU-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-CU-NEXT:    s_endpgm
     ptr %out, i32 %in, i32 %old) {
 entry:
   %gep = getelementptr i32, ptr %out, i32 4
   %val = cmpxchg volatile ptr %gep, i32 %old, i32 %in syncscope("workgroup-one-as") acq_rel acquire
   ret void
 }
 

 ;
 ; GFX11-CU-LABEL: flat_workgroup_one_as_seq_cst_acquire_cmpxchg:
 ; GFX11-CU:       ; %bb.0: ; %entry
 ; GFX11-CU-NEXT:    s_load_b128 s[0:3], s[0:1], 0x0
 ; GFX11-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v0, s0 :: v_dual_mov_b32 v1, s1
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v2, s2 :: v_dual_mov_b32 v3, s3
 ; GFX11-CU-NEXT:    flat_atomic_cmpswap_b32 v[0:1], v[2:3] offset:16
-; GFX11-CU-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-CU-NEXT:    s_endpgm
     ptr %out, i32 %in, i32 %old) {
 entry:
   %gep = getelementptr i32, ptr %out, i32 4
   %val = cmpxchg volatile ptr %gep, i32 %old, i32 %in syncscope("workgroup-one-as") seq_cst acquire
   ret void
 }
 

 ;
 ; GFX11-CU-LABEL: flat_workgroup_one_as_monotonic_seq_cst_cmpxchg:
 ; GFX11-CU:       ; %bb.0: ; %entry
 ; GFX11-CU-NEXT:    s_load_b128 s[0:3], s[0:1], 0x0
 ; GFX11-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v0, s0 :: v_dual_mov_b32 v1, s1
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v2, s2 :: v_dual_mov_b32 v3, s3
 ; GFX11-CU-NEXT:    flat_atomic_cmpswap_b32 v[0:1], v[2:3] offset:16
-; GFX11-CU-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-CU-NEXT:    s_endpgm
     ptr %out, i32 %in, i32 %old) {
 entry:
   %gep = getelementptr i32, ptr %out, i32 4
   %val = cmpxchg volatile ptr %gep, i32 %old, i32 %in syncscope("workgroup-one-as") monotonic seq_cst
   ret void
 }
 

 ;
 ; GFX11-CU-LABEL: flat_workgroup_one_as_acquire_seq_cst_cmpxchg:
 ; GFX11-CU:       ; %bb.0: ; %entry
 ; GFX11-CU-NEXT:    s_load_b128 s[0:3], s[0:1], 0x0
 ; GFX11-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v0, s0 :: v_dual_mov_b32 v1, s1
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v2, s2 :: v_dual_mov_b32 v3, s3
 ; GFX11-CU-NEXT:    flat_atomic_cmpswap_b32 v[0:1], v[2:3] offset:16
-; GFX11-CU-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-CU-NEXT:    s_endpgm
     ptr %out, i32 %in, i32 %old) {
 entry:
   %gep = getelementptr i32, ptr %out, i32 4
   %val = cmpxchg volatile ptr %gep, i32 %old, i32 %in syncscope("workgroup-one-as") acquire seq_cst
   ret void
 }
 

 ;
 ; GFX11-CU-LABEL: flat_workgroup_one_as_release_seq_cst_cmpxchg:
 ; GFX11-CU:       ; %bb.0: ; %entry
 ; GFX11-CU-NEXT:    s_load_b128 s[0:3], s[0:1], 0x0
 ; GFX11-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v0, s0 :: v_dual_mov_b32 v1, s1
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v2, s2 :: v_dual_mov_b32 v3, s3
 ; GFX11-CU-NEXT:    flat_atomic_cmpswap_b32 v[0:1], v[2:3] offset:16
-; GFX11-CU-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-CU-NEXT:    s_endpgm
     ptr %out, i32 %in, i32 %old) {
 entry:
   %gep = getelementptr i32, ptr %out, i32 4
   %val = cmpxchg volatile ptr %gep, i32 %old, i32 %in syncscope("workgroup-one-as") release seq_cst
   ret void
 }
 

 ;
 ; GFX11-CU-LABEL: flat_workgroup_one_as_acq_rel_seq_cst_cmpxchg:
 ; GFX11-CU:       ; %bb.0: ; %entry
 ; GFX11-CU-NEXT:    s_load_b128 s[0:3], s[0:1], 0x0
 ; GFX11-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v0, s0 :: v_dual_mov_b32 v1, s1
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v2, s2 :: v_dual_mov_b32 v3, s3
 ; GFX11-CU-NEXT:    flat_atomic_cmpswap_b32 v[0:1], v[2:3] offset:16
-; GFX11-CU-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-CU-NEXT:    s_endpgm
     ptr %out, i32 %in, i32 %old) {
 entry:
   %gep = getelementptr i32, ptr %out, i32 4
   %val = cmpxchg volatile ptr %gep, i32 %old, i32 %in syncscope("workgroup-one-as") acq_rel seq_cst
   ret void
 }
 

 ;
 ; GFX11-CU-LABEL: flat_workgroup_one_as_seq_cst_seq_cst_cmpxchg:
 ; GFX11-CU:       ; %bb.0: ; %entry
 ; GFX11-CU-NEXT:    s_load_b128 s[0:3], s[0:1], 0x0
 ; GFX11-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v0, s0 :: v_dual_mov_b32 v1, s1
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v2, s2 :: v_dual_mov_b32 v3, s3
 ; GFX11-CU-NEXT:    flat_atomic_cmpswap_b32 v[0:1], v[2:3] offset:16
-; GFX11-CU-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-CU-NEXT:    s_endpgm
     ptr %out, i32 %in, i32 %old) {
 entry:
   %gep = getelementptr i32, ptr %out, i32 4
   %val = cmpxchg volatile ptr %gep, i32 %old, i32 %in syncscope("workgroup-one-as") seq_cst seq_cst
   ret void
 }
 

 ; GFX11-WGP:       ; %bb.0: ; %entry
 ; GFX11-WGP-NEXT:    s_load_b128 s[0:3], s[0:1], 0x0
 ; GFX11-WGP-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-WGP-NEXT:    v_dual_mov_b32 v0, s0 :: v_dual_mov_b32 v1, s1
 ; GFX11-WGP-NEXT:    v_dual_mov_b32 v2, s2 :: v_dual_mov_b32 v3, s3
 ; GFX11-WGP-NEXT:    flat_atomic_cmpswap_b32 v2, v[0:1], v[2:3] offset:16 glc
 ; GFX11-WGP-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX11-WGP-NEXT:    flat_store_b32 v[0:1], v2
-; GFX11-WGP-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-WGP-NEXT:    s_endpgm
 ;
 ; GFX11-CU-LABEL: flat_workgroup_one_as_monotonicmonotonic_ret_cmpxchg:
 ; GFX11-CU:       ; %bb.0: ; %entry
 ; GFX11-CU-NEXT:    s_load_b128 s[0:3], s[0:1], 0x0
 ; GFX11-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v0, s0 :: v_dual_mov_b32 v1, s1
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v2, s2 :: v_dual_mov_b32 v3, s3
 ; GFX11-CU-NEXT:    flat_atomic_cmpswap_b32 v2, v[0:1], v[2:3] offset:16 glc
 ; GFX11-CU-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX11-CU-NEXT:    flat_store_b32 v[0:1], v2
-; GFX11-CU-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-CU-NEXT:    s_endpgm
     ptr %out, i32 %in, i32 %old) {
 entry:
   %gep = getelementptr i32, ptr %out, i32 4
   %val = cmpxchg volatile ptr %gep, i32 %old, i32 %in syncscope("workgroup-one-as") monotonic monotonic
   %val0 = extractvalue { i32, i1 } %val, 0
   store i32 %val0, ptr %out, align 4
   ret void

 ; GFX11-WGP-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-WGP-NEXT:    v_dual_mov_b32 v0, s0 :: v_dual_mov_b32 v1, s1
 ; GFX11-WGP-NEXT:    v_dual_mov_b32 v2, s2 :: v_dual_mov_b32 v3, s3
 ; GFX11-WGP-NEXT:    flat_atomic_cmpswap_b32 v2, v[0:1], v[2:3] offset:16 glc
 ; GFX11-WGP-NEXT:    s_waitcnt vmcnt(0)
 ; GFX11-WGP-NEXT:    buffer_gl0_inv
 ; GFX11-WGP-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-WGP-NEXT:    flat_store_b32 v[0:1], v2
-; GFX11-WGP-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-WGP-NEXT:    s_endpgm
 ;
 ; GFX11-CU-LABEL: flat_workgroup_one_as_acquire_monotonic_ret_cmpxchg:
 ; GFX11-CU:       ; %bb.0: ; %entry
 ; GFX11-CU-NEXT:    s_load_b128 s[0:3], s[0:1], 0x0
 ; GFX11-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v0, s0 :: v_dual_mov_b32 v1, s1
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v2, s2 :: v_dual_mov_b32 v3, s3
 ; GFX11-CU-NEXT:    flat_atomic_cmpswap_b32 v2, v[0:1], v[2:3] offset:16 glc
 ; GFX11-CU-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX11-CU-NEXT:    flat_store_b32 v[0:1], v2
-; GFX11-CU-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-CU-NEXT:    s_endpgm
     ptr %out, i32 %in, i32 %old) {
 entry:
   %gep = getelementptr i32, ptr %out, i32 4
   %val = cmpxchg volatile ptr %gep, i32 %old, i32 %in syncscope("workgroup-one-as") acquire monotonic
   %val0 = extractvalue { i32, i1 } %val, 0
   store i32 %val0, ptr %out, align 4
   ret void

 ; GFX11-WGP-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-WGP-NEXT:    v_dual_mov_b32 v0, s0 :: v_dual_mov_b32 v1, s1
 ; GFX11-WGP-NEXT:    v_dual_mov_b32 v2, s2 :: v_dual_mov_b32 v3, s3
 ; GFX11-WGP-NEXT:    s_waitcnt vmcnt(0)
 ; GFX11-WGP-NEXT:    s_waitcnt_vscnt null, 0x0
 ; GFX11-WGP-NEXT:    flat_atomic_cmpswap_b32 v2, v[0:1], v[2:3] offset:16 glc
 ; GFX11-WGP-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX11-WGP-NEXT:    flat_store_b32 v[0:1], v2
-; GFX11-WGP-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-WGP-NEXT:    s_endpgm
 ;
 ; GFX11-CU-LABEL: flat_workgroup_one_as_release_monotonic_ret_cmpxchg:
 ; GFX11-CU:       ; %bb.0: ; %entry
 ; GFX11-CU-NEXT:    s_load_b128 s[0:3], s[0:1], 0x0
 ; GFX11-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v0, s0 :: v_dual_mov_b32 v1, s1
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v2, s2 :: v_dual_mov_b32 v3, s3
 ; GFX11-CU-NEXT:    flat_atomic_cmpswap_b32 v2, v[0:1], v[2:3] offset:16 glc
 ; GFX11-CU-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX11-CU-NEXT:    flat_store_b32 v[0:1], v2
-; GFX11-CU-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-CU-NEXT:    s_endpgm
     ptr %out, i32 %in, i32 %old) {
 entry:
   %gep = getelementptr i32, ptr %out, i32 4
   %val = cmpxchg volatile ptr %gep, i32 %old, i32 %in syncscope("workgroup-one-as") release monotonic
   %val0 = extractvalue { i32, i1 } %val, 0
   store i32 %val0, ptr %out, align 4
   ret void

 ; GFX11-WGP-NEXT:    v_dual_mov_b32 v2, s2 :: v_dual_mov_b32 v3, s3
 ; GFX11-WGP-NEXT:    s_waitcnt vmcnt(0)
 ; GFX11-WGP-NEXT:    s_waitcnt_vscnt null, 0x0
 ; GFX11-WGP-NEXT:    flat_atomic_cmpswap_b32 v2, v[0:1], v[2:3] offset:16 glc
 ; GFX11-WGP-NEXT:    s_waitcnt vmcnt(0)
 ; GFX11-WGP-NEXT:    buffer_gl0_inv
 ; GFX11-WGP-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-WGP-NEXT:    flat_store_b32 v[0:1], v2
-; GFX11-WGP-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-WGP-NEXT:    s_endpgm
 ;
 ; GFX11-CU-LABEL: flat_workgroup_one_as_acq_rel_monotonic_ret_cmpxchg:
 ; GFX11-CU:       ; %bb.0: ; %entry
 ; GFX11-CU-NEXT:    s_load_b128 s[0:3], s[0:1], 0x0
 ; GFX11-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v0, s0 :: v_dual_mov_b32 v1, s1
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v2, s2 :: v_dual_mov_b32 v3, s3
 ; GFX11-CU-NEXT:    flat_atomic_cmpswap_b32 v2, v[0:1], v[2:3] offset:16 glc
 ; GFX11-CU-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX11-CU-NEXT:    flat_store_b32 v[0:1], v2
-; GFX11-CU-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-CU-NEXT:    s_endpgm
     ptr %out, i32 %in, i32 %old) {
 entry:
   %gep = getelementptr i32, ptr %out, i32 4
   %val = cmpxchg volatile ptr %gep, i32 %old, i32 %in syncscope("workgroup-one-as") acq_rel monotonic
   %val0 = extractvalue { i32, i1 } %val, 0
   store i32 %val0, ptr %out, align 4
   ret void

 ; GFX11-WGP-NEXT:    v_dual_mov_b32 v2, s2 :: v_dual_mov_b32 v3, s3
 ; GFX11-WGP-NEXT:    s_waitcnt vmcnt(0)
 ; GFX11-WGP-NEXT:    s_waitcnt_vscnt null, 0x0
 ; GFX11-WGP-NEXT:    flat_atomic_cmpswap_b32 v2, v[0:1], v[2:3] offset:16 glc
 ; GFX11-WGP-NEXT:    s_waitcnt vmcnt(0)
 ; GFX11-WGP-NEXT:    buffer_gl0_inv
 ; GFX11-WGP-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-WGP-NEXT:    flat_store_b32 v[0:1], v2
-; GFX11-WGP-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-WGP-NEXT:    s_endpgm
 ;
 ; GFX11-CU-LABEL: flat_workgroup_one_as_seq_cst_monotonic_ret_cmpxchg:
 ; GFX11-CU:       ; %bb.0: ; %entry
 ; GFX11-CU-NEXT:    s_load_b128 s[0:3], s[0:1], 0x0
 ; GFX11-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v0, s0 :: v_dual_mov_b32 v1, s1
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v2, s2 :: v_dual_mov_b32 v3, s3
 ; GFX11-CU-NEXT:    flat_atomic_cmpswap_b32 v2, v[0:1], v[2:3] offset:16 glc
 ; GFX11-CU-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX11-CU-NEXT:    flat_store_b32 v[0:1], v2
-; GFX11-CU-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-CU-NEXT:    s_endpgm
     ptr %out, i32 %in, i32 %old) {
 entry:
   %gep = getelementptr i32, ptr %out, i32 4
   %val = cmpxchg volatile ptr %gep, i32 %old, i32 %in syncscope("workgroup-one-as") seq_cst monotonic
   %val0 = extractvalue { i32, i1 } %val, 0
   store i32 %val0, ptr %out, align 4
   ret void

 ; GFX11-WGP-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-WGP-NEXT:    v_dual_mov_b32 v0, s0 :: v_dual_mov_b32 v1, s1
 ; GFX11-WGP-NEXT:    v_dual_mov_b32 v2, s2 :: v_dual_mov_b32 v3, s3
 ; GFX11-WGP-NEXT:    flat_atomic_cmpswap_b32 v2, v[0:1], v[2:3] offset:16 glc
 ; GFX11-WGP-NEXT:    s_waitcnt vmcnt(0)
 ; GFX11-WGP-NEXT:    buffer_gl0_inv
 ; GFX11-WGP-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-WGP-NEXT:    flat_store_b32 v[0:1], v2
-; GFX11-WGP-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-WGP-NEXT:    s_endpgm
 ;
 ; GFX11-CU-LABEL: flat_workgroup_one_as_monotonic_acquire_ret_cmpxchg:
 ; GFX11-CU:       ; %bb.0: ; %entry
 ; GFX11-CU-NEXT:    s_load_b128 s[0:3], s[0:1], 0x0
 ; GFX11-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v0, s0 :: v_dual_mov_b32 v1, s1
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v2, s2 :: v_dual_mov_b32 v3, s3
 ; GFX11-CU-NEXT:    flat_atomic_cmpswap_b32 v2, v[0:1], v[2:3] offset:16 glc
 ; GFX11-CU-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX11-CU-NEXT:    flat_store_b32 v[0:1], v2
-; GFX11-CU-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-CU-NEXT:    s_endpgm
     ptr %out, i32 %in, i32 %old) {
 entry:
   %gep = getelementptr i32, ptr %out, i32 4
   %val = cmpxchg volatile ptr %gep, i32 %old, i32 %in syncscope("workgroup-one-as") monotonic acquire
   %val0 = extractvalue { i32, i1 } %val, 0
   store i32 %val0, ptr %out, align 4
   ret void

 ; GFX11-WGP-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-WGP-NEXT:    v_dual_mov_b32 v0, s0 :: v_dual_mov_b32 v1, s1
 ; GFX11-WGP-NEXT:    v_dual_mov_b32 v2, s2 :: v_dual_mov_b32 v3, s3
 ; GFX11-WGP-NEXT:    flat_atomic_cmpswap_b32 v2, v[0:1], v[2:3] offset:16 glc
 ; GFX11-WGP-NEXT:    s_waitcnt vmcnt(0)
 ; GFX11-WGP-NEXT:    buffer_gl0_inv
 ; GFX11-WGP-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-WGP-NEXT:    flat_store_b32 v[0:1], v2
-; GFX11-WGP-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-WGP-NEXT:    s_endpgm
 ;
 ; GFX11-CU-LABEL: flat_workgroup_one_as_acquire_acquire_ret_cmpxchg:
 ; GFX11-CU:       ; %bb.0: ; %entry
 ; GFX11-CU-NEXT:    s_load_b128 s[0:3], s[0:1], 0x0
 ; GFX11-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v0, s0 :: v_dual_mov_b32 v1, s1
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v2, s2 :: v_dual_mov_b32 v3, s3
 ; GFX11-CU-NEXT:    flat_atomic_cmpswap_b32 v2, v[0:1], v[2:3] offset:16 glc
 ; GFX11-CU-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX11-CU-NEXT:    flat_store_b32 v[0:1], v2
-; GFX11-CU-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-CU-NEXT:    s_endpgm
     ptr %out, i32 %in, i32 %old) {
 entry:
   %gep = getelementptr i32, ptr %out, i32 4
   %val = cmpxchg volatile ptr %gep, i32 %old, i32 %in syncscope("workgroup-one-as") acquire acquire
   %val0 = extractvalue { i32, i1 } %val, 0
   store i32 %val0, ptr %out, align 4
   ret void

 ; GFX11-WGP-NEXT:    v_dual_mov_b32 v2, s2 :: v_dual_mov_b32 v3, s3
 ; GFX11-WGP-NEXT:    s_waitcnt vmcnt(0)
 ; GFX11-WGP-NEXT:    s_waitcnt_vscnt null, 0x0
 ; GFX11-WGP-NEXT:    flat_atomic_cmpswap_b32 v2, v[0:1], v[2:3] offset:16 glc
 ; GFX11-WGP-NEXT:    s_waitcnt vmcnt(0)
 ; GFX11-WGP-NEXT:    buffer_gl0_inv
 ; GFX11-WGP-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-WGP-NEXT:    flat_store_b32 v[0:1], v2
-; GFX11-WGP-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-WGP-NEXT:    s_endpgm
 ;
 ; GFX11-CU-LABEL: flat_workgroup_one_as_release_acquire_ret_cmpxchg:
 ; GFX11-CU:       ; %bb.0: ; %entry
 ; GFX11-CU-NEXT:    s_load_b128 s[0:3], s[0:1], 0x0
 ; GFX11-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v0, s0 :: v_dual_mov_b32 v1, s1
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v2, s2 :: v_dual_mov_b32 v3, s3
 ; GFX11-CU-NEXT:    flat_atomic_cmpswap_b32 v2, v[0:1], v[2:3] offset:16 glc
 ; GFX11-CU-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX11-CU-NEXT:    flat_store_b32 v[0:1], v2
-; GFX11-CU-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-CU-NEXT:    s_endpgm
     ptr %out, i32 %in, i32 %old) {
 entry:
   %gep = getelementptr i32, ptr %out, i32 4
   %val = cmpxchg volatile ptr %gep, i32 %old, i32 %in syncscope("workgroup-one-as") release acquire
   %val0 = extractvalue { i32, i1 } %val, 0
   store i32 %val0, ptr %out, align 4
   ret void

 ; GFX11-WGP-NEXT:    v_dual_mov_b32 v2, s2 :: v_dual_mov_b32 v3, s3
 ; GFX11-WGP-NEXT:    s_waitcnt vmcnt(0)
 ; GFX11-WGP-NEXT:    s_waitcnt_vscnt null, 0x0
 ; GFX11-WGP-NEXT:    flat_atomic_cmpswap_b32 v2, v[0:1], v[2:3] offset:16 glc
 ; GFX11-WGP-NEXT:    s_waitcnt vmcnt(0)
 ; GFX11-WGP-NEXT:    buffer_gl0_inv
 ; GFX11-WGP-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-WGP-NEXT:    flat_store_b32 v[0:1], v2
-; GFX11-WGP-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-WGP-NEXT:    s_endpgm
 ;
 ; GFX11-CU-LABEL: flat_workgroup_one_as_acq_rel_acquire_ret_cmpxchg:
 ; GFX11-CU:       ; %bb.0: ; %entry
 ; GFX11-CU-NEXT:    s_load_b128 s[0:3], s[0:1], 0x0
 ; GFX11-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v0, s0 :: v_dual_mov_b32 v1, s1
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v2, s2 :: v_dual_mov_b32 v3, s3
 ; GFX11-CU-NEXT:    flat_atomic_cmpswap_b32 v2, v[0:1], v[2:3] offset:16 glc
 ; GFX11-CU-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX11-CU-NEXT:    flat_store_b32 v[0:1], v2
-; GFX11-CU-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-CU-NEXT:    s_endpgm
     ptr %out, i32 %in, i32 %old) {
 entry:
   %gep = getelementptr i32, ptr %out, i32 4
   %val = cmpxchg volatile ptr %gep, i32 %old, i32 %in syncscope("workgroup-one-as") acq_rel acquire
   %val0 = extractvalue { i32, i1 } %val, 0
   store i32 %val0, ptr %out, align 4
   ret void

 ; GFX11-WGP-NEXT:    v_dual_mov_b32 v2, s2 :: v_dual_mov_b32 v3, s3
 ; GFX11-WGP-NEXT:    s_waitcnt vmcnt(0)
 ; GFX11-WGP-NEXT:    s_waitcnt_vscnt null, 0x0
 ; GFX11-WGP-NEXT:    flat_atomic_cmpswap_b32 v2, v[0:1], v[2:3] offset:16 glc
 ; GFX11-WGP-NEXT:    s_waitcnt vmcnt(0)
 ; GFX11-WGP-NEXT:    buffer_gl0_inv
 ; GFX11-WGP-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-WGP-NEXT:    flat_store_b32 v[0:1], v2
-; GFX11-WGP-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-WGP-NEXT:    s_endpgm
 ;
 ; GFX11-CU-LABEL: flat_workgroup_one_as_seq_cst_acquire_ret_cmpxchg:
 ; GFX11-CU:       ; %bb.0: ; %entry
 ; GFX11-CU-NEXT:    s_load_b128 s[0:3], s[0:1], 0x0
 ; GFX11-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v0, s0 :: v_dual_mov_b32 v1, s1
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v2, s2 :: v_dual_mov_b32 v3, s3
 ; GFX11-CU-NEXT:    flat_atomic_cmpswap_b32 v2, v[0:1], v[2:3] offset:16 glc
 ; GFX11-CU-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX11-CU-NEXT:    flat_store_b32 v[0:1], v2
-; GFX11-CU-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-CU-NEXT:    s_endpgm
     ptr %out, i32 %in, i32 %old) {
 entry:
   %gep = getelementptr i32, ptr %out, i32 4
   %val = cmpxchg volatile ptr %gep, i32 %old, i32 %in syncscope("workgroup-one-as") seq_cst acquire
   %val0 = extractvalue { i32, i1 } %val, 0
   store i32 %val0, ptr %out, align 4
   ret void

 ; GFX11-WGP-NEXT:    v_dual_mov_b32 v2, s2 :: v_dual_mov_b32 v3, s3
 ; GFX11-WGP-NEXT:    s_waitcnt vmcnt(0)
 ; GFX11-WGP-NEXT:    s_waitcnt_vscnt null, 0x0
 ; GFX11-WGP-NEXT:    flat_atomic_cmpswap_b32 v2, v[0:1], v[2:3] offset:16 glc
 ; GFX11-WGP-NEXT:    s_waitcnt vmcnt(0)
 ; GFX11-WGP-NEXT:    buffer_gl0_inv
 ; GFX11-WGP-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-WGP-NEXT:    flat_store_b32 v[0:1], v2
-; GFX11-WGP-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-WGP-NEXT:    s_endpgm
 ;
 ; GFX11-CU-LABEL: flat_workgroup_one_as_monotonic_seq_cst_ret_cmpxchg:
 ; GFX11-CU:       ; %bb.0: ; %entry
 ; GFX11-CU-NEXT:    s_load_b128 s[0:3], s[0:1], 0x0
 ; GFX11-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v0, s0 :: v_dual_mov_b32 v1, s1
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v2, s2 :: v_dual_mov_b32 v3, s3
 ; GFX11-CU-NEXT:    flat_atomic_cmpswap_b32 v2, v[0:1], v[2:3] offset:16 glc
 ; GFX11-CU-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX11-CU-NEXT:    flat_store_b32 v[0:1], v2
-; GFX11-CU-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-CU-NEXT:    s_endpgm
     ptr %out, i32 %in, i32 %old) {
 entry:
   %gep = getelementptr i32, ptr %out, i32 4
   %val = cmpxchg volatile ptr %gep, i32 %old, i32 %in syncscope("workgroup-one-as") monotonic seq_cst
   %val0 = extractvalue { i32, i1 } %val, 0
   store i32 %val0, ptr %out, align 4
   ret void

 ; GFX11-WGP-NEXT:    v_dual_mov_b32 v2, s2 :: v_dual_mov_b32 v3, s3
 ; GFX11-WGP-NEXT:    s_waitcnt vmcnt(0)
 ; GFX11-WGP-NEXT:    s_waitcnt_vscnt null, 0x0
 ; GFX11-WGP-NEXT:    flat_atomic_cmpswap_b32 v2, v[0:1], v[2:3] offset:16 glc
 ; GFX11-WGP-NEXT:    s_waitcnt vmcnt(0)
 ; GFX11-WGP-NEXT:    buffer_gl0_inv
 ; GFX11-WGP-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-WGP-NEXT:    flat_store_b32 v[0:1], v2
-; GFX11-WGP-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-WGP-NEXT:    s_endpgm
 ;
 ; GFX11-CU-LABEL: flat_workgroup_one_as_acquire_seq_cst_ret_cmpxchg:
 ; GFX11-CU:       ; %bb.0: ; %entry
 ; GFX11-CU-NEXT:    s_load_b128 s[0:3], s[0:1], 0x0
 ; GFX11-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v0, s0 :: v_dual_mov_b32 v1, s1
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v2, s2 :: v_dual_mov_b32 v3, s3
 ; GFX11-CU-NEXT:    flat_atomic_cmpswap_b32 v2, v[0:1], v[2:3] offset:16 glc
 ; GFX11-CU-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX11-CU-NEXT:    flat_store_b32 v[0:1], v2
-; GFX11-CU-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-CU-NEXT:    s_endpgm
     ptr %out, i32 %in, i32 %old) {
 entry:
   %gep = getelementptr i32, ptr %out, i32 4
   %val = cmpxchg volatile ptr %gep, i32 %old, i32 %in syncscope("workgroup-one-as") acquire seq_cst
   %val0 = extractvalue { i32, i1 } %val, 0
   store i32 %val0, ptr %out, align 4
   ret void

 ; GFX11-WGP-NEXT:    v_dual_mov_b32 v2, s2 :: v_dual_mov_b32 v3, s3
 ; GFX11-WGP-NEXT:    s_waitcnt vmcnt(0)
 ; GFX11-WGP-NEXT:    s_waitcnt_vscnt null, 0x0
 ; GFX11-WGP-NEXT:    flat_atomic_cmpswap_b32 v2, v[0:1], v[2:3] offset:16 glc
 ; GFX11-WGP-NEXT:    s_waitcnt vmcnt(0)
 ; GFX11-WGP-NEXT:    buffer_gl0_inv
 ; GFX11-WGP-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-WGP-NEXT:    flat_store_b32 v[0:1], v2
-; GFX11-WGP-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-WGP-NEXT:    s_endpgm
 ;
 ; GFX11-CU-LABEL: flat_workgroup_one_as_release_seq_cst_ret_cmpxchg:
 ; GFX11-CU:       ; %bb.0: ; %entry
 ; GFX11-CU-NEXT:    s_load_b128 s[0:3], s[0:1], 0x0
 ; GFX11-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v0, s0 :: v_dual_mov_b32 v1, s1
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v2, s2 :: v_dual_mov_b32 v3, s3
 ; GFX11-CU-NEXT:    flat_atomic_cmpswap_b32 v2, v[0:1], v[2:3] offset:16 glc
 ; GFX11-CU-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX11-CU-NEXT:    flat_store_b32 v[0:1], v2
-; GFX11-CU-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-CU-NEXT:    s_endpgm
     ptr %out, i32 %in, i32 %old) {
 entry:
   %gep = getelementptr i32, ptr %out, i32 4
   %val = cmpxchg volatile ptr %gep, i32 %old, i32 %in syncscope("workgroup-one-as") release seq_cst
   %val0 = extractvalue { i32, i1 } %val, 0
   store i32 %val0, ptr %out, align 4
   ret void

 ; GFX11-WGP-NEXT:    v_dual_mov_b32 v2, s2 :: v_dual_mov_b32 v3, s3
 ; GFX11-WGP-NEXT:    s_waitcnt vmcnt(0)
 ; GFX11-WGP-NEXT:    s_waitcnt_vscnt null, 0x0
 ; GFX11-WGP-NEXT:    flat_atomic_cmpswap_b32 v2, v[0:1], v[2:3] offset:16 glc
 ; GFX11-WGP-NEXT:    s_waitcnt vmcnt(0)
 ; GFX11-WGP-NEXT:    buffer_gl0_inv
 ; GFX11-WGP-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-WGP-NEXT:    flat_store_b32 v[0:1], v2
-; GFX11-WGP-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-WGP-NEXT:    s_endpgm
 ;
 ; GFX11-CU-LABEL: flat_workgroup_one_as_acq_rel_seq_cst_ret_cmpxchg:
 ; GFX11-CU:       ; %bb.0: ; %entry
 ; GFX11-CU-NEXT:    s_load_b128 s[0:3], s[0:1], 0x0
 ; GFX11-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v0, s0 :: v_dual_mov_b32 v1, s1
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v2, s2 :: v_dual_mov_b32 v3, s3
 ; GFX11-CU-NEXT:    flat_atomic_cmpswap_b32 v2, v[0:1], v[2:3] offset:16 glc
 ; GFX11-CU-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX11-CU-NEXT:    flat_store_b32 v[0:1], v2
-; GFX11-CU-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-CU-NEXT:    s_endpgm
     ptr %out, i32 %in, i32 %old) {
 entry:
   %gep = getelementptr i32, ptr %out, i32 4
   %val = cmpxchg volatile ptr %gep, i32 %old, i32 %in syncscope("workgroup-one-as") acq_rel seq_cst
   %val0 = extractvalue { i32, i1 } %val, 0
   store i32 %val0, ptr %out, align 4
   ret void

 ; GFX11-WGP-NEXT:    v_dual_mov_b32 v2, s2 :: v_dual_mov_b32 v3, s3
 ; GFX11-WGP-NEXT:    s_waitcnt vmcnt(0)
 ; GFX11-WGP-NEXT:    s_waitcnt_vscnt null, 0x0
 ; GFX11-WGP-NEXT:    flat_atomic_cmpswap_b32 v2, v[0:1], v[2:3] offset:16 glc
 ; GFX11-WGP-NEXT:    s_waitcnt vmcnt(0)
 ; GFX11-WGP-NEXT:    buffer_gl0_inv
 ; GFX11-WGP-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-WGP-NEXT:    flat_store_b32 v[0:1], v2
-; GFX11-WGP-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-WGP-NEXT:    s_endpgm
 ;
 ; GFX11-CU-LABEL: flat_workgroup_one_as_seq_cst_seq_cst_ret_cmpxchg:
 ; GFX11-CU:       ; %bb.0: ; %entry
 ; GFX11-CU-NEXT:    s_load_b128 s[0:3], s[0:1], 0x0
 ; GFX11-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v0, s0 :: v_dual_mov_b32 v1, s1
 ; GFX11-CU-NEXT:    v_dual_mov_b32 v2, s2 :: v_dual_mov_b32 v3, s3
 ; GFX11-CU-NEXT:    flat_atomic_cmpswap_b32 v2, v[0:1], v[2:3] offset:16 glc
 ; GFX11-CU-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX11-CU-NEXT:    flat_store_b32 v[0:1], v2
-; GFX11-CU-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-CU-NEXT:    s_endpgm
     ptr %out, i32 %in, i32 %old) {
 entry:
   %gep = getelementptr i32, ptr %out, i32 4
   %val = cmpxchg volatile ptr %gep, i32 %old, i32 %in syncscope("workgroup-one-as") seq_cst seq_cst
   %val0 = extractvalue { i32, i1 } %val, 0
   store i32 %val0, ptr %out, align 4
   ret void
 }
 

llvm/test/CodeGen/AMDGPU/memory-legalizer-private-nontemporal.ll

 ; GFX11-WGP-NEXT:    s_clause 0x1
 ; GFX11-WGP-NEXT:    s_load_b64 s[2:3], s[0:1], 0x0
 ; GFX11-WGP-NEXT:    s_load_b32 s0, s[0:1], 0x8
 ; GFX11-WGP-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-WGP-NEXT:    s_load_b32 s1, s[2:3], 0x0
 ; GFX11-WGP-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-WGP-NEXT:    v_mov_b32_e32 v0, s1
 ; GFX11-WGP-NEXT:    scratch_store_b32 off, v0, s0 glc slc dlc
-; GFX11-WGP-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-WGP-NEXT:    s_endpgm
 ;
 ; GFX11-CU-LABEL: private_nontemporal_store_0:
 ; GFX11-CU:       ; %bb.0: ; %entry
 ; GFX11-CU-NEXT:    s_clause 0x1
 ; GFX11-CU-NEXT:    s_load_b64 s[2:3], s[0:1], 0x0
 ; GFX11-CU-NEXT:    s_load_b32 s0, s[0:1], 0x8
 ; GFX11-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-CU-NEXT:    s_load_b32 s1, s[2:3], 0x0
 ; GFX11-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-CU-NEXT:    v_mov_b32_e32 v0, s1
 ; GFX11-CU-NEXT:    scratch_store_b32 off, v0, s0 glc slc dlc
-; GFX11-CU-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-CU-NEXT:    s_endpgm
     ptr addrspace(1) %in, ptr addrspace(5) %out) {
 entry:
   %val = load i32, ptr addrspace(1) %in, align 4
   store i32 %val, ptr addrspace(5) %out, !nontemporal !0
   ret void
 }
 

 ; GFX11-WGP-NEXT:    s_load_b64 s[2:3], s[0:1], 0x0
 ; GFX11-WGP-NEXT:    s_load_b32 s0, s[0:1], 0x8
 ; GFX11-WGP-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-WGP-NEXT:    s_load_b32 s1, s[2:3], 0x0
 ; GFX11-WGP-NEXT:    v_lshl_add_u32 v0, v0, 2, s0
 ; GFX11-WGP-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-WGP-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX11-WGP-NEXT:    scratch_store_b32 v0, v1, off glc slc dlc
-; GFX11-WGP-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-WGP-NEXT:    s_endpgm
 ;
 ; GFX11-CU-LABEL: private_nontemporal_store_1:
 ; GFX11-CU:       ; %bb.0: ; %entry
 ; GFX11-CU-NEXT:    s_clause 0x1
 ; GFX11-CU-NEXT:    s_load_b64 s[2:3], s[0:1], 0x0
 ; GFX11-CU-NEXT:    s_load_b32 s0, s[0:1], 0x8
 ; GFX11-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-CU-NEXT:    s_load_b32 s1, s[2:3], 0x0
 ; GFX11-CU-NEXT:    v_lshl_add_u32 v0, v0, 2, s0
 ; GFX11-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-CU-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX11-CU-NEXT:    scratch_store_b32 v0, v1, off glc slc dlc
-; GFX11-CU-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-CU-NEXT:    s_endpgm
     ptr addrspace(1) %in, ptr addrspace(5) %out) {
 entry:
   %tid = call i32 @llvm.amdgcn.workitem.id.x()
   %val = load i32, ptr addrspace(1) %in, align 4
   %out.gep = getelementptr inbounds i32, ptr addrspace(5) %out, i32 %tid
   store i32 %val, ptr addrspace(5) %out.gep, !nontemporal !0
   ret void
 }
 
 !0 = !{i32 1}
 declare i32 @llvm.amdgcn.workitem.id.x()

llvm/test/CodeGen/AMDGPU/memory-legalizer-private-volatile.ll

 ; GFX11-WGP-NEXT:    s_load_b64 s[2:3], s[0:1], 0x0
 ; GFX11-WGP-NEXT:    s_load_b32 s0, s[0:1], 0x8
 ; GFX11-WGP-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-WGP-NEXT:    s_load_b32 s1, s[2:3], 0x0
 ; GFX11-WGP-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-WGP-NEXT:    v_mov_b32_e32 v0, s1
 ; GFX11-WGP-NEXT:    scratch_store_b32 off, v0, s0 dlc
 ; GFX11-WGP-NEXT:    s_waitcnt_vscnt null, 0x0
-; GFX11-WGP-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-WGP-NEXT:    s_endpgm
 ;
 ; GFX11-CU-LABEL: private_volatile_store_0:
 ; GFX11-CU:       ; %bb.0: ; %entry
 ; GFX11-CU-NEXT:    s_clause 0x1
 ; GFX11-CU-NEXT:    s_load_b64 s[2:3], s[0:1], 0x0
 ; GFX11-CU-NEXT:    s_load_b32 s0, s[0:1], 0x8
 ; GFX11-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-CU-NEXT:    s_load_b32 s1, s[2:3], 0x0
 ; GFX11-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-CU-NEXT:    v_mov_b32_e32 v0, s1
 ; GFX11-CU-NEXT:    scratch_store_b32 off, v0, s0 dlc
 ; GFX11-CU-NEXT:    s_waitcnt_vscnt null, 0x0
-; GFX11-CU-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-CU-NEXT:    s_endpgm
     ptr addrspace(1) %in, ptr addrspace(5) %out) {
 entry:
   %val = load i32, ptr addrspace(1) %in, align 4
   store volatile i32 %val, ptr addrspace(5) %out
   ret void
 }
 

 ; GFX11-WGP-NEXT:    s_load_b32 s0, s[0:1], 0x8
 ; GFX11-WGP-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-WGP-NEXT:    s_load_b32 s1, s[2:3], 0x0
 ; GFX11-WGP-NEXT:    v_lshl_add_u32 v0, v0, 2, s0
 ; GFX11-WGP-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-WGP-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX11-WGP-NEXT:    scratch_store_b32 v0, v1, off dlc
 ; GFX11-WGP-NEXT:    s_waitcnt_vscnt null, 0x0
-; GFX11-WGP-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-WGP-NEXT:    s_endpgm
 ;
 ; GFX11-CU-LABEL: private_volatile_store_1:
 ; GFX11-CU:       ; %bb.0: ; %entry
 ; GFX11-CU-NEXT:    s_clause 0x1
 ; GFX11-CU-NEXT:    s_load_b64 s[2:3], s[0:1], 0x0
 ; GFX11-CU-NEXT:    s_load_b32 s0, s[0:1], 0x8
 ; GFX11-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-CU-NEXT:    s_load_b32 s1, s[2:3], 0x0
 ; GFX11-CU-NEXT:    v_lshl_add_u32 v0, v0, 2, s0
 ; GFX11-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-CU-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX11-CU-NEXT:    scratch_store_b32 v0, v1, off dlc
 ; GFX11-CU-NEXT:    s_waitcnt_vscnt null, 0x0
-; GFX11-CU-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-CU-NEXT:    s_endpgm
     ptr addrspace(1) %in, ptr addrspace(5) %out) {
 entry:
   %tid = call i32 @llvm.amdgcn.workitem.id.x()
   %val = load i32, ptr addrspace(1) %in, align 4
   %out.gep = getelementptr inbounds i32, ptr addrspace(5) %out, i32 %tid
   store volatile i32 %val, ptr addrspace(5) %out.gep
   ret void
 }
 
 declare i32 @llvm.amdgcn.workitem.id.x()

llvm/test/CodeGen/AMDGPU/offset-split-flat.ll

 ; GFX11-LABEL: flat_inst_salu_offset_1:
 ; GFX11:       ; %bb.0:
 ; GFX11-NEXT:    s_load_b64 s[0:1], s[0:1], 0x24
 ; GFX11-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-NEXT:    v_dual_mov_b32 v0, s0 :: v_dual_mov_b32 v1, s1
 ; GFX11-NEXT:    flat_load_u8 v0, v[0:1] offset:1 glc dlc
 ; GFX11-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX11-NEXT:    flat_store_b8 v[0:1], v0
-; GFX11-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-NEXT:    s_endpgm
   %gep = getelementptr i8, ptr %p, i64 1
   %load = load volatile i8, ptr %gep, align 1
   store i8 %load, ptr undef
   ret void
 }
 
 define amdgpu_kernel void @flat_inst_salu_offset_11bit_max(ptr %p) {

 ; GFX11-LABEL: flat_inst_salu_offset_11bit_max:
 ; GFX11:       ; %bb.0:
 ; GFX11-NEXT:    s_load_b64 s[0:1], s[0:1], 0x24
 ; GFX11-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-NEXT:    v_dual_mov_b32 v0, s0 :: v_dual_mov_b32 v1, s1
 ; GFX11-NEXT:    flat_load_u8 v0, v[0:1] offset:2047 glc dlc
 ; GFX11-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX11-NEXT:    flat_store_b8 v[0:1], v0
-; GFX11-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-NEXT:    s_endpgm
   %gep = getelementptr i8, ptr %p, i64 2047
   %load = load volatile i8, ptr %gep, align 1
   store i8 %load, ptr undef
   ret void
 }
 
 define amdgpu_kernel void @flat_inst_salu_offset_12bit_max(ptr %p) {

 ; GFX11-LABEL: flat_inst_salu_offset_12bit_max:
 ; GFX11:       ; %bb.0:
 ; GFX11-NEXT:    s_load_b64 s[0:1], s[0:1], 0x24
 ; GFX11-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-NEXT:    v_dual_mov_b32 v0, s0 :: v_dual_mov_b32 v1, s1
 ; GFX11-NEXT:    flat_load_u8 v0, v[0:1] offset:4095 glc dlc
 ; GFX11-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX11-NEXT:    flat_store_b8 v[0:1], v0
-; GFX11-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-NEXT:    s_endpgm
   %gep = getelementptr i8, ptr %p, i64 4095
   %load = load volatile i8, ptr %gep, align 1
   store i8 %load, ptr undef
   ret void
 }
 
 define amdgpu_kernel void @flat_inst_salu_offset_13bit_max(ptr %p) {

 ; GFX11-SDAG-NEXT:    s_load_b64 s[0:1], s[0:1], 0x24
 ; GFX11-SDAG-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-SDAG-NEXT:    v_add_co_u32 v0, s0, 0x1000, s0
 ; GFX11-SDAG-NEXT:    s_delay_alu instid0(VALU_DEP_1)
 ; GFX11-SDAG-NEXT:    v_add_co_ci_u32_e64 v1, null, 0, s1, s0
 ; GFX11-SDAG-NEXT:    flat_load_u8 v0, v[0:1] offset:4095 glc dlc
 ; GFX11-SDAG-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX11-SDAG-NEXT:    flat_store_b8 v[0:1], v0
-; GFX11-SDAG-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-SDAG-NEXT:    s_endpgm
 ;
 ; GFX9-GISEL-LABEL: flat_inst_salu_offset_13bit_max:
 ; GFX9-GISEL:       ; %bb.0:
 ; GFX9-GISEL-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x24
 ; GFX9-GISEL-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX9-GISEL-NEXT:    s_add_u32 s0, s0, 0x1fff
 ; GFX9-GISEL-NEXT:    s_addc_u32 s1, s1, 0

 ; GFX11-GISEL-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-GISEL-NEXT:    s_add_u32 s0, s0, 0x1fff
 ; GFX11-GISEL-NEXT:    s_addc_u32 s1, s1, 0
 ; GFX11-GISEL-NEXT:    s_delay_alu instid0(SALU_CYCLE_1)
 ; GFX11-GISEL-NEXT:    v_dual_mov_b32 v0, s0 :: v_dual_mov_b32 v1, s1
 ; GFX11-GISEL-NEXT:    flat_load_u8 v0, v[0:1] glc dlc
 ; GFX11-GISEL-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX11-GISEL-NEXT:    flat_store_b8 v[0:1], v0
-; GFX11-GISEL-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-GISEL-NEXT:    s_endpgm
   %gep = getelementptr i8, ptr %p, i64 8191
   %load = load volatile i8, ptr %gep, align 1
   store i8 %load, ptr undef
   ret void
 }
 
 define amdgpu_kernel void @flat_inst_salu_offset_neg_11bit_max(ptr %p) {

 ; GFX11-SDAG-NEXT:    s_load_b64 s[0:1], s[0:1], 0x24
 ; GFX11-SDAG-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-SDAG-NEXT:    v_add_co_u32 v0, s0, 0xfffff800, s0
 ; GFX11-SDAG-NEXT:    s_delay_alu instid0(VALU_DEP_1)
 ; GFX11-SDAG-NEXT:    v_add_co_ci_u32_e64 v1, null, -1, s1, s0
 ; GFX11-SDAG-NEXT:    flat_load_u8 v0, v[0:1] glc dlc
 ; GFX11-SDAG-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX11-SDAG-NEXT:    flat_store_b8 v[0:1], v0
-; GFX11-SDAG-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-SDAG-NEXT:    s_endpgm
 ;
 ; GFX9-GISEL-LABEL: flat_inst_salu_offset_neg_11bit_max:
 ; GFX9-GISEL:       ; %bb.0:
 ; GFX9-GISEL-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x24
 ; GFX9-GISEL-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX9-GISEL-NEXT:    s_add_u32 s0, s0, 0xfffff800
 ; GFX9-GISEL-NEXT:    s_addc_u32 s1, s1, -1

 ; GFX11-GISEL-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-GISEL-NEXT:    s_add_u32 s0, s0, 0xfffff800
 ; GFX11-GISEL-NEXT:    s_addc_u32 s1, s1, -1
 ; GFX11-GISEL-NEXT:    s_delay_alu instid0(SALU_CYCLE_1)
 ; GFX11-GISEL-NEXT:    v_dual_mov_b32 v0, s0 :: v_dual_mov_b32 v1, s1
 ; GFX11-GISEL-NEXT:    flat_load_u8 v0, v[0:1] glc dlc
 ; GFX11-GISEL-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX11-GISEL-NEXT:    flat_store_b8 v[0:1], v0
-; GFX11-GISEL-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-GISEL-NEXT:    s_endpgm
   %gep = getelementptr i8, ptr %p, i64 -2048
   %load = load volatile i8, ptr %gep, align 1
   store i8 %load, ptr undef
   ret void
 }
 
 define amdgpu_kernel void @flat_inst_salu_offset_neg_12bit_max(ptr %p) {

 ; GFX11-SDAG-NEXT:    s_load_b64 s[0:1], s[0:1], 0x24
 ; GFX11-SDAG-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-SDAG-NEXT:    v_add_co_u32 v0, s0, 0xfffff000, s0
 ; GFX11-SDAG-NEXT:    s_delay_alu instid0(VALU_DEP_1)
 ; GFX11-SDAG-NEXT:    v_add_co_ci_u32_e64 v1, null, -1, s1, s0
 ; GFX11-SDAG-NEXT:    flat_load_u8 v0, v[0:1] glc dlc
 ; GFX11-SDAG-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX11-SDAG-NEXT:    flat_store_b8 v[0:1], v0
-; GFX11-SDAG-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-SDAG-NEXT:    s_endpgm
 ;
 ; GFX9-GISEL-LABEL: flat_inst_salu_offset_neg_12bit_max:
 ; GFX9-GISEL:       ; %bb.0:
 ; GFX9-GISEL-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x24
 ; GFX9-GISEL-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX9-GISEL-NEXT:    s_add_u32 s0, s0, 0xfffff000
 ; GFX9-GISEL-NEXT:    s_addc_u32 s1, s1, -1

 ; GFX11-GISEL-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-GISEL-NEXT:    s_add_u32 s0, s0, 0xfffff000
 ; GFX11-GISEL-NEXT:    s_addc_u32 s1, s1, -1
 ; GFX11-GISEL-NEXT:    s_delay_alu instid0(SALU_CYCLE_1)
 ; GFX11-GISEL-NEXT:    v_dual_mov_b32 v0, s0 :: v_dual_mov_b32 v1, s1
 ; GFX11-GISEL-NEXT:    flat_load_u8 v0, v[0:1] glc dlc
 ; GFX11-GISEL-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX11-GISEL-NEXT:    flat_store_b8 v[0:1], v0
-; GFX11-GISEL-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-GISEL-NEXT:    s_endpgm
   %gep = getelementptr i8, ptr %p, i64 -4096
   %load = load volatile i8, ptr %gep, align 1
   store i8 %load, ptr undef
   ret void
 }
 
 define amdgpu_kernel void @flat_inst_salu_offset_neg_13bit_max(ptr %p) {

 ; GFX11-SDAG-NEXT:    s_load_b64 s[0:1], s[0:1], 0x24
 ; GFX11-SDAG-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-SDAG-NEXT:    v_add_co_u32 v0, s0, 0xffffe000, s0
 ; GFX11-SDAG-NEXT:    s_delay_alu instid0(VALU_DEP_1)
 ; GFX11-SDAG-NEXT:    v_add_co_ci_u32_e64 v1, null, -1, s1, s0
 ; GFX11-SDAG-NEXT:    flat_load_u8 v0, v[0:1] glc dlc
 ; GFX11-SDAG-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX11-SDAG-NEXT:    flat_store_b8 v[0:1], v0
-; GFX11-SDAG-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-SDAG-NEXT:    s_endpgm
 ;
 ; GFX9-GISEL-LABEL: flat_inst_salu_offset_neg_13bit_max:
 ; GFX9-GISEL:       ; %bb.0:
 ; GFX9-GISEL-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x24
 ; GFX9-GISEL-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX9-GISEL-NEXT:    s_add_u32 s0, s0, 0xffffe000
 ; GFX9-GISEL-NEXT:    s_addc_u32 s1, s1, -1

 ; GFX11-GISEL-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-GISEL-NEXT:    s_add_u32 s0, s0, 0xffffe000
 ; GFX11-GISEL-NEXT:    s_addc_u32 s1, s1, -1
 ; GFX11-GISEL-NEXT:    s_delay_alu instid0(SALU_CYCLE_1)
 ; GFX11-GISEL-NEXT:    v_dual_mov_b32 v0, s0 :: v_dual_mov_b32 v1, s1
 ; GFX11-GISEL-NEXT:    flat_load_u8 v0, v[0:1] glc dlc
 ; GFX11-GISEL-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX11-GISEL-NEXT:    flat_store_b8 v[0:1], v0
-; GFX11-GISEL-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-GISEL-NEXT:    s_endpgm
   %gep = getelementptr i8, ptr %p, i64 -8192
   %load = load volatile i8, ptr %gep, align 1
   store i8 %load, ptr undef
   ret void
 }
 
 define amdgpu_kernel void @flat_inst_salu_offset_2x_11bit_max(ptr %p) {

 ; GFX11-LABEL: flat_inst_salu_offset_2x_11bit_max:
 ; GFX11:       ; %bb.0:
 ; GFX11-NEXT:    s_load_b64 s[0:1], s[0:1], 0x24
 ; GFX11-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-NEXT:    v_dual_mov_b32 v0, s0 :: v_dual_mov_b32 v1, s1
 ; GFX11-NEXT:    flat_load_u8 v0, v[0:1] offset:4095 glc dlc
 ; GFX11-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX11-NEXT:    flat_store_b8 v[0:1], v0
-; GFX11-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-NEXT:    s_endpgm
   %gep = getelementptr i8, ptr %p, i64 4095
   %load = load volatile i8, ptr %gep, align 1
   store i8 %load, ptr undef
   ret void
 }
 
 define amdgpu_kernel void @flat_inst_salu_offset_2x_12bit_max(ptr %p) {

 ; GFX11-SDAG-NEXT:    s_load_b64 s[0:1], s[0:1], 0x24
 ; GFX11-SDAG-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-SDAG-NEXT:    v_add_co_u32 v0, s0, 0x1000, s0
 ; GFX11-SDAG-NEXT:    s_delay_alu instid0(VALU_DEP_1)
 ; GFX11-SDAG-NEXT:    v_add_co_ci_u32_e64 v1, null, 0, s1, s0
 ; GFX11-SDAG-NEXT:    flat_load_u8 v0, v[0:1] offset:4095 glc dlc
 ; GFX11-SDAG-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX11-SDAG-NEXT:    flat_store_b8 v[0:1], v0
-; GFX11-SDAG-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-SDAG-NEXT:    s_endpgm
 ;
 ; GFX9-GISEL-LABEL: flat_inst_salu_offset_2x_12bit_max:
 ; GFX9-GISEL:       ; %bb.0:
 ; GFX9-GISEL-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x24
 ; GFX9-GISEL-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX9-GISEL-NEXT:    s_add_u32 s0, s0, 0x1fff
 ; GFX9-GISEL-NEXT:    s_addc_u32 s1, s1, 0

 ; GFX11-GISEL-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-GISEL-NEXT:    s_add_u32 s0, s0, 0x1fff
 ; GFX11-GISEL-NEXT:    s_addc_u32 s1, s1, 0
 ; GFX11-GISEL-NEXT:    s_delay_alu instid0(SALU_CYCLE_1)
 ; GFX11-GISEL-NEXT:    v_dual_mov_b32 v0, s0 :: v_dual_mov_b32 v1, s1
 ; GFX11-GISEL-NEXT:    flat_load_u8 v0, v[0:1] glc dlc
 ; GFX11-GISEL-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX11-GISEL-NEXT:    flat_store_b8 v[0:1], v0
-; GFX11-GISEL-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-GISEL-NEXT:    s_endpgm
   %gep = getelementptr i8, ptr %p, i64 8191
   %load = load volatile i8, ptr %gep, align 1
   store i8 %load, ptr undef
   ret void
 }
 
 define amdgpu_kernel void @flat_inst_salu_offset_2x_13bit_max(ptr %p) {

 ; GFX11-SDAG-NEXT:    s_load_b64 s[0:1], s[0:1], 0x24
 ; GFX11-SDAG-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-SDAG-NEXT:    v_add_co_u32 v0, s0, 0x3000, s0
 ; GFX11-SDAG-NEXT:    s_delay_alu instid0(VALU_DEP_1)
 ; GFX11-SDAG-NEXT:    v_add_co_ci_u32_e64 v1, null, 0, s1, s0
 ; GFX11-SDAG-NEXT:    flat_load_u8 v0, v[0:1] offset:4095 glc dlc
 ; GFX11-SDAG-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX11-SDAG-NEXT:    flat_store_b8 v[0:1], v0
-; GFX11-SDAG-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-SDAG-NEXT:    s_endpgm
 ;
 ; GFX9-GISEL-LABEL: flat_inst_salu_offset_2x_13bit_max:
 ; GFX9-GISEL:       ; %bb.0:
 ; GFX9-GISEL-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x24
 ; GFX9-GISEL-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX9-GISEL-NEXT:    s_add_u32 s0, s0, 0x3fff
 ; GFX9-GISEL-NEXT:    s_addc_u32 s1, s1, 0

 ; GFX11-GISEL-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-GISEL-NEXT:    s_add_u32 s0, s0, 0x3fff
 ; GFX11-GISEL-NEXT:    s_addc_u32 s1, s1, 0
 ; GFX11-GISEL-NEXT:    s_delay_alu instid0(SALU_CYCLE_1)
 ; GFX11-GISEL-NEXT:    v_dual_mov_b32 v0, s0 :: v_dual_mov_b32 v1, s1
 ; GFX11-GISEL-NEXT:    flat_load_u8 v0, v[0:1] glc dlc
 ; GFX11-GISEL-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX11-GISEL-NEXT:    flat_store_b8 v[0:1], v0
-; GFX11-GISEL-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-GISEL-NEXT:    s_endpgm
   %gep = getelementptr i8, ptr %p, i64 16383
   %load = load volatile i8, ptr %gep, align 1
   store i8 %load, ptr undef
   ret void
 }
 
 define amdgpu_kernel void @flat_inst_salu_offset_2x_neg_11bit_max(ptr %p) {

 ; GFX11-SDAG-NEXT:    s_load_b64 s[0:1], s[0:1], 0x24
 ; GFX11-SDAG-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-SDAG-NEXT:    v_add_co_u32 v0, s0, 0xfffff000, s0
 ; GFX11-SDAG-NEXT:    s_delay_alu instid0(VALU_DEP_1)
 ; GFX11-SDAG-NEXT:    v_add_co_ci_u32_e64 v1, null, -1, s1, s0
 ; GFX11-SDAG-NEXT:    flat_load_u8 v0, v[0:1] glc dlc
 ; GFX11-SDAG-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX11-SDAG-NEXT:    flat_store_b8 v[0:1], v0
-; GFX11-SDAG-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-SDAG-NEXT:    s_endpgm
 ;
 ; GFX9-GISEL-LABEL: flat_inst_salu_offset_2x_neg_11bit_max:
 ; GFX9-GISEL:       ; %bb.0:
 ; GFX9-GISEL-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x24
 ; GFX9-GISEL-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX9-GISEL-NEXT:    s_add_u32 s0, s0, 0xfffff000
 ; GFX9-GISEL-NEXT:    s_addc_u32 s1, s1, -1

 ; GFX11-GISEL-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-GISEL-NEXT:    s_add_u32 s0, s0, 0xfffff000
 ; GFX11-GISEL-NEXT:    s_addc_u32 s1, s1, -1
 ; GFX11-GISEL-NEXT:    s_delay_alu instid0(SALU_CYCLE_1)
 ; GFX11-GISEL-NEXT:    v_dual_mov_b32 v0, s0 :: v_dual_mov_b32 v1, s1
 ; GFX11-GISEL-NEXT:    flat_load_u8 v0, v[0:1] glc dlc
 ; GFX11-GISEL-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX11-GISEL-NEXT:    flat_store_b8 v[0:1], v0
-; GFX11-GISEL-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-GISEL-NEXT:    s_endpgm
   %gep = getelementptr i8, ptr %p, i64 -4096
   %load = load volatile i8, ptr %gep, align 1
   store i8 %load, ptr undef
   ret void
 }
 
 define amdgpu_kernel void @flat_inst_salu_offset_2x_neg_12bit_max(ptr %p) {

 ; GFX11-SDAG-NEXT:    s_load_b64 s[0:1], s[0:1], 0x24
 ; GFX11-SDAG-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-SDAG-NEXT:    v_add_co_u32 v0, s0, 0xffffe000, s0
 ; GFX11-SDAG-NEXT:    s_delay_alu instid0(VALU_DEP_1)
 ; GFX11-SDAG-NEXT:    v_add_co_ci_u32_e64 v1, null, -1, s1, s0
 ; GFX11-SDAG-NEXT:    flat_load_u8 v0, v[0:1] glc dlc
 ; GFX11-SDAG-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX11-SDAG-NEXT:    flat_store_b8 v[0:1], v0
-; GFX11-SDAG-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-SDAG-NEXT:    s_endpgm
 ;
 ; GFX9-GISEL-LABEL: flat_inst_salu_offset_2x_neg_12bit_max:
 ; GFX9-GISEL:       ; %bb.0:
 ; GFX9-GISEL-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x24
 ; GFX9-GISEL-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX9-GISEL-NEXT:    s_add_u32 s0, s0, 0xffffe000
 ; GFX9-GISEL-NEXT:    s_addc_u32 s1, s1, -1

 ; GFX11-GISEL-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-GISEL-NEXT:    s_add_u32 s0, s0, 0xffffe000
 ; GFX11-GISEL-NEXT:    s_addc_u32 s1, s1, -1
 ; GFX11-GISEL-NEXT:    s_delay_alu instid0(SALU_CYCLE_1)
 ; GFX11-GISEL-NEXT:    v_dual_mov_b32 v0, s0 :: v_dual_mov_b32 v1, s1
 ; GFX11-GISEL-NEXT:    flat_load_u8 v0, v[0:1] glc dlc
 ; GFX11-GISEL-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX11-GISEL-NEXT:    flat_store_b8 v[0:1], v0
-; GFX11-GISEL-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-GISEL-NEXT:    s_endpgm
   %gep = getelementptr i8, ptr %p, i64 -8192
   %load = load volatile i8, ptr %gep, align 1
   store i8 %load, ptr undef
   ret void
 }
 
 define amdgpu_kernel void @flat_inst_salu_offset_2x_neg_13bit_max(ptr %p) {

 ; GFX11-SDAG-NEXT:    s_load_b64 s[0:1], s[0:1], 0x24
 ; GFX11-SDAG-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-SDAG-NEXT:    v_add_co_u32 v0, s0, 0xffffc000, s0
 ; GFX11-SDAG-NEXT:    s_delay_alu instid0(VALU_DEP_1)
 ; GFX11-SDAG-NEXT:    v_add_co_ci_u32_e64 v1, null, -1, s1, s0
 ; GFX11-SDAG-NEXT:    flat_load_u8 v0, v[0:1] glc dlc
 ; GFX11-SDAG-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX11-SDAG-NEXT:    flat_store_b8 v[0:1], v0
-; GFX11-SDAG-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-SDAG-NEXT:    s_endpgm
 ;
 ; GFX9-GISEL-LABEL: flat_inst_salu_offset_2x_neg_13bit_max:
 ; GFX9-GISEL:       ; %bb.0:
 ; GFX9-GISEL-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x24
 ; GFX9-GISEL-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX9-GISEL-NEXT:    s_add_u32 s0, s0, 0xffffc000
 ; GFX9-GISEL-NEXT:    s_addc_u32 s1, s1, -1

 ; GFX11-GISEL-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-GISEL-NEXT:    s_add_u32 s0, s0, 0xffffc000
 ; GFX11-GISEL-NEXT:    s_addc_u32 s1, s1, -1
 ; GFX11-GISEL-NEXT:    s_delay_alu instid0(SALU_CYCLE_1)
 ; GFX11-GISEL-NEXT:    v_dual_mov_b32 v0, s0 :: v_dual_mov_b32 v1, s1
 ; GFX11-GISEL-NEXT:    flat_load_u8 v0, v[0:1] glc dlc
 ; GFX11-GISEL-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX11-GISEL-NEXT:    flat_store_b8 v[0:1], v0
-; GFX11-GISEL-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-GISEL-NEXT:    s_endpgm
   %gep = getelementptr i8, ptr %p, i64 -16384
   %load = load volatile i8, ptr %gep, align 1
   store i8 %load, ptr undef
   ret void
 }
 
 ; Fill 11-bit low-bits (1ull << 33) | 2047

 ; GFX11-SDAG-NEXT:    s_load_b64 s[0:1], s[0:1], 0x24
 ; GFX11-SDAG-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-SDAG-NEXT:    v_add_co_u32 v0, s0, 0, s0
 ; GFX11-SDAG-NEXT:    s_delay_alu instid0(VALU_DEP_1)
 ; GFX11-SDAG-NEXT:    v_add_co_ci_u32_e64 v1, null, 2, s1, s0
 ; GFX11-SDAG-NEXT:    flat_load_u8 v0, v[0:1] offset:2047 glc dlc
 ; GFX11-SDAG-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX11-SDAG-NEXT:    flat_store_b8 v[0:1], v0
-; GFX11-SDAG-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-SDAG-NEXT:    s_endpgm
 ;
 ; GFX9-GISEL-LABEL: flat_inst_salu_offset_64bit_11bit_split0:
 ; GFX9-GISEL:       ; %bb.0:
 ; GFX9-GISEL-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x24
 ; GFX9-GISEL-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX9-GISEL-NEXT:    s_add_u32 s0, s0, 0x7ff
 ; GFX9-GISEL-NEXT:    s_addc_u32 s1, s1, 2

 ; GFX11-GISEL-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-GISEL-NEXT:    s_add_u32 s0, s0, 0x7ff
 ; GFX11-GISEL-NEXT:    s_addc_u32 s1, s1, 2
 ; GFX11-GISEL-NEXT:    s_delay_alu instid0(SALU_CYCLE_1)
 ; GFX11-GISEL-NEXT:    v_dual_mov_b32 v0, s0 :: v_dual_mov_b32 v1, s1
 ; GFX11-GISEL-NEXT:    flat_load_u8 v0, v[0:1] glc dlc
 ; GFX11-GISEL-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX11-GISEL-NEXT:    flat_store_b8 v[0:1], v0
-; GFX11-GISEL-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-GISEL-NEXT:    s_endpgm
   %gep = getelementptr i8, ptr %p, i64 8589936639
   %load = load volatile i8, ptr %gep, align 1
   store i8 %load, ptr undef
   ret void
 }
 
 ; Fill 11-bit low-bits (1ull << 33) | 2048

 ; GFX11-SDAG-NEXT:    s_load_b64 s[0:1], s[0:1], 0x24
 ; GFX11-SDAG-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-SDAG-NEXT:    v_add_co_u32 v0, s0, 0, s0
 ; GFX11-SDAG-NEXT:    s_delay_alu instid0(VALU_DEP_1)
 ; GFX11-SDAG-NEXT:    v_add_co_ci_u32_e64 v1, null, 2, s1, s0
 ; GFX11-SDAG-NEXT:    flat_load_u8 v0, v[0:1] offset:2048 glc dlc
 ; GFX11-SDAG-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX11-SDAG-NEXT:    flat_store_b8 v[0:1], v0
-; GFX11-SDAG-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-SDAG-NEXT:    s_endpgm
 ;
 ; GFX9-GISEL-LABEL: flat_inst_salu_offset_64bit_11bit_split1:
 ; GFX9-GISEL:       ; %bb.0:
 ; GFX9-GISEL-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x24
 ; GFX9-GISEL-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX9-GISEL-NEXT:    s_add_u32 s0, s0, 0x800
 ; GFX9-GISEL-NEXT:    s_addc_u32 s1, s1, 2

 ; GFX11-GISEL-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-GISEL-NEXT:    s_add_u32 s0, s0, 0x800
 ; GFX11-GISEL-NEXT:    s_addc_u32 s1, s1, 2
 ; GFX11-GISEL-NEXT:    s_delay_alu instid0(SALU_CYCLE_1)
 ; GFX11-GISEL-NEXT:    v_dual_mov_b32 v0, s0 :: v_dual_mov_b32 v1, s1
 ; GFX11-GISEL-NEXT:    flat_load_u8 v0, v[0:1] glc dlc
 ; GFX11-GISEL-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX11-GISEL-NEXT:    flat_store_b8 v[0:1], v0
-; GFX11-GISEL-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-GISEL-NEXT:    s_endpgm
   %gep = getelementptr i8, ptr %p, i64 8589936640
   %load = load volatile i8, ptr %gep, align 1
   store i8 %load, ptr undef
   ret void
 }
 
 ; Fill 12-bit low-bits (1ull << 33) | 4095

 ; GFX11-SDAG-NEXT:    s_load_b64 s[0:1], s[0:1], 0x24
 ; GFX11-SDAG-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-SDAG-NEXT:    v_add_co_u32 v0, s0, 0, s0
 ; GFX11-SDAG-NEXT:    s_delay_alu instid0(VALU_DEP_1)
 ; GFX11-SDAG-NEXT:    v_add_co_ci_u32_e64 v1, null, 2, s1, s0
 ; GFX11-SDAG-NEXT:    flat_load_u8 v0, v[0:1] offset:4095 glc dlc
 ; GFX11-SDAG-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX11-SDAG-NEXT:    flat_store_b8 v[0:1], v0
-; GFX11-SDAG-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-SDAG-NEXT:    s_endpgm
 ;
 ; GFX9-GISEL-LABEL: flat_inst_salu_offset_64bit_12bit_split0:
 ; GFX9-GISEL:       ; %bb.0:
 ; GFX9-GISEL-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x24
 ; GFX9-GISEL-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX9-GISEL-NEXT:    s_add_u32 s0, s0, 0xfff
 ; GFX9-GISEL-NEXT:    s_addc_u32 s1, s1, 2

 ; GFX11-GISEL-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-GISEL-NEXT:    s_add_u32 s0, s0, 0xfff
 ; GFX11-GISEL-NEXT:    s_addc_u32 s1, s1, 2
 ; GFX11-GISEL-NEXT:    s_delay_alu instid0(SALU_CYCLE_1)
 ; GFX11-GISEL-NEXT:    v_dual_mov_b32 v0, s0 :: v_dual_mov_b32 v1, s1
 ; GFX11-GISEL-NEXT:    flat_load_u8 v0, v[0:1] glc dlc
 ; GFX11-GISEL-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX11-GISEL-NEXT:    flat_store_b8 v[0:1], v0
-; GFX11-GISEL-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-GISEL-NEXT:    s_endpgm
   %gep = getelementptr i8, ptr %p, i64 8589938687
   %load = load volatile i8, ptr %gep, align 1
   store i8 %load, ptr undef
   ret void
 }
 
 ; Fill 12-bit low-bits (1ull << 33) | 4096

 ; GFX11-SDAG-NEXT:    s_load_b64 s[0:1], s[0:1], 0x24
 ; GFX11-SDAG-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-SDAG-NEXT:    v_add_co_u32 v0, s0, 0x1000, s0
 ; GFX11-SDAG-NEXT:    s_delay_alu instid0(VALU_DEP_1)
 ; GFX11-SDAG-NEXT:    v_add_co_ci_u32_e64 v1, null, 2, s1, s0
 ; GFX11-SDAG-NEXT:    flat_load_u8 v0, v[0:1] glc dlc
 ; GFX11-SDAG-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX11-SDAG-NEXT:    flat_store_b8 v[0:1], v0
-; GFX11-SDAG-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-SDAG-NEXT:    s_endpgm
 ;
 ; GFX9-GISEL-LABEL: flat_inst_salu_offset_64bit_12bit_split1:
 ; GFX9-GISEL:       ; %bb.0:
 ; GFX9-GISEL-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x24
 ; GFX9-GISEL-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX9-GISEL-NEXT:    s_add_u32 s0, s0, 0x1000
 ; GFX9-GISEL-NEXT:    s_addc_u32 s1, s1, 2

 ; GFX11-GISEL-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-GISEL-NEXT:    s_add_u32 s0, s0, 0x1000
 ; GFX11-GISEL-NEXT:    s_addc_u32 s1, s1, 2
 ; GFX11-GISEL-NEXT:    s_delay_alu instid0(SALU_CYCLE_1)
 ; GFX11-GISEL-NEXT:    v_dual_mov_b32 v0, s0 :: v_dual_mov_b32 v1, s1
 ; GFX11-GISEL-NEXT:    flat_load_u8 v0, v[0:1] glc dlc
 ; GFX11-GISEL-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX11-GISEL-NEXT:    flat_store_b8 v[0:1], v0
-; GFX11-GISEL-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-GISEL-NEXT:    s_endpgm
   %gep = getelementptr i8, ptr %p, i64 8589938688
   %load = load volatile i8, ptr %gep, align 1
   store i8 %load, ptr undef
   ret void
 }
 
 ; Fill 13-bit low-bits (1ull << 33) | 8191

 ; GFX11-SDAG-NEXT:    s_load_b64 s[0:1], s[0:1], 0x24
 ; GFX11-SDAG-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-SDAG-NEXT:    v_add_co_u32 v0, s0, 0x1000, s0
 ; GFX11-SDAG-NEXT:    s_delay_alu instid0(VALU_DEP_1)
 ; GFX11-SDAG-NEXT:    v_add_co_ci_u32_e64 v1, null, 2, s1, s0
 ; GFX11-SDAG-NEXT:    flat_load_u8 v0, v[0:1] offset:4095 glc dlc
 ; GFX11-SDAG-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX11-SDAG-NEXT:    flat_store_b8 v[0:1], v0
-; GFX11-SDAG-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-SDAG-NEXT:    s_endpgm
 ;
 ; GFX9-GISEL-LABEL: flat_inst_salu_offset_64bit_13bit_split0:
 ; GFX9-GISEL:       ; %bb.0:
 ; GFX9-GISEL-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x24
 ; GFX9-GISEL-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX9-GISEL-NEXT:    s_add_u32 s0, s0, 0x1fff
 ; GFX9-GISEL-NEXT:    s_addc_u32 s1, s1, 2

 ; GFX11-GISEL-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-GISEL-NEXT:    s_add_u32 s0, s0, 0x1fff
 ; GFX11-GISEL-NEXT:    s_addc_u32 s1, s1, 2
 ; GFX11-GISEL-NEXT:    s_delay_alu instid0(SALU_CYCLE_1)
 ; GFX11-GISEL-NEXT:    v_dual_mov_b32 v0, s0 :: v_dual_mov_b32 v1, s1
 ; GFX11-GISEL-NEXT:    flat_load_u8 v0, v[0:1] glc dlc
 ; GFX11-GISEL-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX11-GISEL-NEXT:    flat_store_b8 v[0:1], v0
-; GFX11-GISEL-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-GISEL-NEXT:    s_endpgm
   %gep = getelementptr i8, ptr %p, i64 8589942783
   %load = load volatile i8, ptr %gep, align 1
   store i8 %load, ptr undef
   ret void
 }
 
 ; Fill 13-bit low-bits (1ull << 33) | 8192

 ; GFX11-SDAG-NEXT:    s_load_b64 s[0:1], s[0:1], 0x24
 ; GFX11-SDAG-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-SDAG-NEXT:    v_add_co_u32 v0, s0, 0x2000, s0
 ; GFX11-SDAG-NEXT:    s_delay_alu instid0(VALU_DEP_1)
 ; GFX11-SDAG-NEXT:    v_add_co_ci_u32_e64 v1, null, 2, s1, s0
 ; GFX11-SDAG-NEXT:    flat_load_u8 v0, v[0:1] glc dlc
 ; GFX11-SDAG-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX11-SDAG-NEXT:    flat_store_b8 v[0:1], v0
-; GFX11-SDAG-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-SDAG-NEXT:    s_endpgm
 ;
 ; GFX9-GISEL-LABEL: flat_inst_salu_offset_64bit_13bit_split1:
 ; GFX9-GISEL:       ; %bb.0:
 ; GFX9-GISEL-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x24
 ; GFX9-GISEL-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX9-GISEL-NEXT:    s_add_u32 s0, s0, 0x2000
 ; GFX9-GISEL-NEXT:    s_addc_u32 s1, s1, 2

 ; GFX11-GISEL-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-GISEL-NEXT:    s_add_u32 s0, s0, 0x2000
 ; GFX11-GISEL-NEXT:    s_addc_u32 s1, s1, 2
 ; GFX11-GISEL-NEXT:    s_delay_alu instid0(SALU_CYCLE_1)
 ; GFX11-GISEL-NEXT:    v_dual_mov_b32 v0, s0 :: v_dual_mov_b32 v1, s1
 ; GFX11-GISEL-NEXT:    flat_load_u8 v0, v[0:1] glc dlc
 ; GFX11-GISEL-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX11-GISEL-NEXT:    flat_store_b8 v[0:1], v0
-; GFX11-GISEL-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-GISEL-NEXT:    s_endpgm
   %gep = getelementptr i8, ptr %p, i64 8589942784
   %load = load volatile i8, ptr %gep, align 1
   store i8 %load, ptr undef
   ret void
 }
 
 ; Fill 11-bit low-bits, negative high bits (1ull << 63) | 2047

 ; GFX11-SDAG-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-SDAG-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX11-SDAG-NEXT:    v_add_co_u32 v0, vcc_lo, 0x7ff, s0
 ; GFX11-SDAG-NEXT:    s_delay_alu instid0(VALU_DEP_2)
 ; GFX11-SDAG-NEXT:    v_add_co_ci_u32_e32 v1, vcc_lo, 0x80000000, v1, vcc_lo
 ; GFX11-SDAG-NEXT:    flat_load_u8 v0, v[0:1] glc dlc
 ; GFX11-SDAG-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX11-SDAG-NEXT:    flat_store_b8 v[0:1], v0
-; GFX11-SDAG-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-SDAG-NEXT:    s_endpgm
 ;
 ; GFX9-GISEL-LABEL: flat_inst_salu_offset_64bit_11bit_neg_high_split0:
 ; GFX9-GISEL:       ; %bb.0:
 ; GFX9-GISEL-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x24
 ; GFX9-GISEL-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX9-GISEL-NEXT:    s_add_u32 s0, s0, 0x7ff
 ; GFX9-GISEL-NEXT:    s_addc_u32 s1, s1, 0x80000000

 ; GFX11-GISEL-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-GISEL-NEXT:    s_add_u32 s0, s0, 0x7ff
 ; GFX11-GISEL-NEXT:    s_addc_u32 s1, s1, 0x80000000
 ; GFX11-GISEL-NEXT:    s_delay_alu instid0(SALU_CYCLE_1)
 ; GFX11-GISEL-NEXT:    v_dual_mov_b32 v0, s0 :: v_dual_mov_b32 v1, s1
 ; GFX11-GISEL-NEXT:    flat_load_u8 v0, v[0:1] glc dlc
 ; GFX11-GISEL-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX11-GISEL-NEXT:    flat_store_b8 v[0:1], v0
-; GFX11-GISEL-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-GISEL-NEXT:    s_endpgm
   %gep = getelementptr i8, ptr %p, i64 -9223372036854773761
   %load = load volatile i8, ptr %gep, align 1
   store i8 %load, ptr undef
   ret void
 }
 
 ; Fill 11-bit low-bits, negative high bits (1ull << 63) | 2048

 ; GFX11-SDAG-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-SDAG-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX11-SDAG-NEXT:    v_add_co_u32 v0, vcc_lo, 0x800, s0
 ; GFX11-SDAG-NEXT:    s_delay_alu instid0(VALU_DEP_2)
 ; GFX11-SDAG-NEXT:    v_add_co_ci_u32_e32 v1, vcc_lo, 0x80000000, v1, vcc_lo
 ; GFX11-SDAG-NEXT:    flat_load_u8 v0, v[0:1] glc dlc
 ; GFX11-SDAG-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX11-SDAG-NEXT:    flat_store_b8 v[0:1], v0
-; GFX11-SDAG-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-SDAG-NEXT:    s_endpgm
 ;
 ; GFX9-GISEL-LABEL: flat_inst_salu_offset_64bit_11bit_neg_high_split1:
 ; GFX9-GISEL:       ; %bb.0:
 ; GFX9-GISEL-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x24
 ; GFX9-GISEL-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX9-GISEL-NEXT:    s_add_u32 s0, s0, 0x800
 ; GFX9-GISEL-NEXT:    s_addc_u32 s1, s1, 0x80000000

 ; GFX11-GISEL-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-GISEL-NEXT:    s_add_u32 s0, s0, 0x800
 ; GFX11-GISEL-NEXT:    s_addc_u32 s1, s1, 0x80000000
 ; GFX11-GISEL-NEXT:    s_delay_alu instid0(SALU_CYCLE_1)
 ; GFX11-GISEL-NEXT:    v_dual_mov_b32 v0, s0 :: v_dual_mov_b32 v1, s1
 ; GFX11-GISEL-NEXT:    flat_load_u8 v0, v[0:1] glc dlc
 ; GFX11-GISEL-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX11-GISEL-NEXT:    flat_store_b8 v[0:1], v0
-; GFX11-GISEL-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-GISEL-NEXT:    s_endpgm
   %gep = getelementptr i8, ptr %p, i64 -9223372036854773760
   %load = load volatile i8, ptr %gep, align 1
   store i8 %load, ptr undef
   ret void
 }
 
 ; Fill 12-bit low-bits, negative high bits (1ull << 63) | 4095

 ; GFX11-SDAG-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-SDAG-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX11-SDAG-NEXT:    v_add_co_u32 v0, vcc_lo, 0xfff, s0
 ; GFX11-SDAG-NEXT:    s_delay_alu instid0(VALU_DEP_2)
 ; GFX11-SDAG-NEXT:    v_add_co_ci_u32_e32 v1, vcc_lo, 0x80000000, v1, vcc_lo
 ; GFX11-SDAG-NEXT:    flat_load_u8 v0, v[0:1] glc dlc
 ; GFX11-SDAG-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX11-SDAG-NEXT:    flat_store_b8 v[0:1], v0
-; GFX11-SDAG-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-SDAG-NEXT:    s_endpgm
 ;
 ; GFX9-GISEL-LABEL: flat_inst_salu_offset_64bit_12bit_neg_high_split0:
 ; GFX9-GISEL:       ; %bb.0:
 ; GFX9-GISEL-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x24
 ; GFX9-GISEL-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX9-GISEL-NEXT:    s_add_u32 s0, s0, 0xfff
 ; GFX9-GISEL-NEXT:    s_addc_u32 s1, s1, 0x80000000

 ; GFX11-GISEL-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-GISEL-NEXT:    s_add_u32 s0, s0, 0xfff
 ; GFX11-GISEL-NEXT:    s_addc_u32 s1, s1, 0x80000000
 ; GFX11-GISEL-NEXT:    s_delay_alu instid0(SALU_CYCLE_1)
 ; GFX11-GISEL-NEXT:    v_dual_mov_b32 v0, s0 :: v_dual_mov_b32 v1, s1
 ; GFX11-GISEL-NEXT:    flat_load_u8 v0, v[0:1] glc dlc
 ; GFX11-GISEL-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX11-GISEL-NEXT:    flat_store_b8 v[0:1], v0
-; GFX11-GISEL-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-GISEL-NEXT:    s_endpgm
   %gep = getelementptr i8, ptr %p, i64 -9223372036854771713
   %load = load volatile i8, ptr %gep, align 1
   store i8 %load, ptr undef
   ret void
 }
 
 ; Fill 12-bit low-bits, negative high bits (1ull << 63) | 4096

 ; GFX11-SDAG-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-SDAG-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX11-SDAG-NEXT:    v_add_co_u32 v0, vcc_lo, 0x1000, s0
 ; GFX11-SDAG-NEXT:    s_delay_alu instid0(VALU_DEP_2)
 ; GFX11-SDAG-NEXT:    v_add_co_ci_u32_e32 v1, vcc_lo, 0x80000000, v1, vcc_lo
 ; GFX11-SDAG-NEXT:    flat_load_u8 v0, v[0:1] glc dlc
 ; GFX11-SDAG-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX11-SDAG-NEXT:    flat_store_b8 v[0:1], v0
-; GFX11-SDAG-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-SDAG-NEXT:    s_endpgm
 ;
 ; GFX9-GISEL-LABEL: flat_inst_salu_offset_64bit_12bit_neg_high_split1:
 ; GFX9-GISEL:       ; %bb.0:
 ; GFX9-GISEL-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x24
 ; GFX9-GISEL-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX9-GISEL-NEXT:    s_add_u32 s0, s0, 0x1000
 ; GFX9-GISEL-NEXT:    s_addc_u32 s1, s1, 0x80000000

 ; GFX11-GISEL-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-GISEL-NEXT:    s_add_u32 s0, s0, 0x1000
 ; GFX11-GISEL-NEXT:    s_addc_u32 s1, s1, 0x80000000
 ; GFX11-GISEL-NEXT:    s_delay_alu instid0(SALU_CYCLE_1)
 ; GFX11-GISEL-NEXT:    v_dual_mov_b32 v0, s0 :: v_dual_mov_b32 v1, s1
 ; GFX11-GISEL-NEXT:    flat_load_u8 v0, v[0:1] glc dlc
 ; GFX11-GISEL-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX11-GISEL-NEXT:    flat_store_b8 v[0:1], v0
-; GFX11-GISEL-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-GISEL-NEXT:    s_endpgm
   %gep = getelementptr i8, ptr %p, i64 -9223372036854771712
   %load = load volatile i8, ptr %gep, align 1
   store i8 %load, ptr undef
   ret void
 }
 
 ; Fill 13-bit low-bits, negative high bits (1ull << 63) | 8191

 ; GFX11-SDAG-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-SDAG-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX11-SDAG-NEXT:    v_add_co_u32 v0, vcc_lo, 0x1fff, s0
 ; GFX11-SDAG-NEXT:    s_delay_alu instid0(VALU_DEP_2)
 ; GFX11-SDAG-NEXT:    v_add_co_ci_u32_e32 v1, vcc_lo, 0x80000000, v1, vcc_lo
 ; GFX11-SDAG-NEXT:    flat_load_u8 v0, v[0:1] glc dlc
 ; GFX11-SDAG-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX11-SDAG-NEXT:    flat_store_b8 v[0:1], v0
-; GFX11-SDAG-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-SDAG-NEXT:    s_endpgm
 ;
 ; GFX9-GISEL-LABEL: flat_inst_salu_offset_64bit_13bit_neg_high_split0:
 ; GFX9-GISEL:       ; %bb.0:
 ; GFX9-GISEL-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x24
 ; GFX9-GISEL-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX9-GISEL-NEXT:    s_add_u32 s0, s0, 0x1fff
 ; GFX9-GISEL-NEXT:    s_addc_u32 s1, s1, 0x80000000

 ; GFX11-GISEL-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-GISEL-NEXT:    s_add_u32 s0, s0, 0x1fff
 ; GFX11-GISEL-NEXT:    s_addc_u32 s1, s1, 0x80000000
 ; GFX11-GISEL-NEXT:    s_delay_alu instid0(SALU_CYCLE_1)
 ; GFX11-GISEL-NEXT:    v_dual_mov_b32 v0, s0 :: v_dual_mov_b32 v1, s1
 ; GFX11-GISEL-NEXT:    flat_load_u8 v0, v[0:1] glc dlc
 ; GFX11-GISEL-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX11-GISEL-NEXT:    flat_store_b8 v[0:1], v0
-; GFX11-GISEL-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-GISEL-NEXT:    s_endpgm
   %gep = getelementptr i8, ptr %p, i64 -9223372036854767617
   %load = load volatile i8, ptr %gep, align 1
   store i8 %load, ptr undef
   ret void
 }
 
 ; Fill 13-bit low-bits, negative high bits (1ull << 63) | 8192

 ; GFX11-SDAG-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-SDAG-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX11-SDAG-NEXT:    v_add_co_u32 v0, vcc_lo, 0x2000, s0
 ; GFX11-SDAG-NEXT:    s_delay_alu instid0(VALU_DEP_2)
 ; GFX11-SDAG-NEXT:    v_add_co_ci_u32_e32 v1, vcc_lo, 0x80000000, v1, vcc_lo
 ; GFX11-SDAG-NEXT:    flat_load_u8 v0, v[0:1] glc dlc
 ; GFX11-SDAG-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX11-SDAG-NEXT:    flat_store_b8 v[0:1], v0
-; GFX11-SDAG-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-SDAG-NEXT:    s_endpgm
 ;
 ; GFX9-GISEL-LABEL: flat_inst_salu_offset_64bit_13bit_neg_high_split1:
 ; GFX9-GISEL:       ; %bb.0:
 ; GFX9-GISEL-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x24
 ; GFX9-GISEL-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX9-GISEL-NEXT:    s_add_u32 s0, s0, 0x2000
 ; GFX9-GISEL-NEXT:    s_addc_u32 s1, s1, 0x80000000

 ; GFX11-GISEL-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-GISEL-NEXT:    s_add_u32 s0, s0, 0x2000
 ; GFX11-GISEL-NEXT:    s_addc_u32 s1, s1, 0x80000000
 ; GFX11-GISEL-NEXT:    s_delay_alu instid0(SALU_CYCLE_1)
 ; GFX11-GISEL-NEXT:    v_dual_mov_b32 v0, s0 :: v_dual_mov_b32 v1, s1
 ; GFX11-GISEL-NEXT:    flat_load_u8 v0, v[0:1] glc dlc
 ; GFX11-GISEL-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX11-GISEL-NEXT:    flat_store_b8 v[0:1], v0
-; GFX11-GISEL-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-GISEL-NEXT:    s_endpgm
   %gep = getelementptr i8, ptr %p, i64 -9223372036854767616
   %load = load volatile i8, ptr %gep, align 1
   store i8 %load, ptr undef
   ret void
 }

llvm/test/CodeGen/AMDGPU/stack-pointer-offset-relative-frameindex.ll

 ; MUBUF11-NEXT:    s_movk_i32 s0, 0x4000
 ; MUBUF11-NEXT:    scratch_load_b64 v[0:1], off, s0 offset:4
 ; MUBUF11-NEXT:    s_mov_b32 s0, 0x41c64e6d
 ; MUBUF11-NEXT:    s_waitcnt vmcnt(0)
 ; MUBUF11-NEXT:    v_add_nc_u32_e32 v2, v1, v0
 ; MUBUF11-NEXT:    v_mad_u64_u32 v[0:1], null, v2, s0, 0x3039
 ; MUBUF11-NEXT:    scratch_store_b32 off, v0, s0
 ; MUBUF11-NEXT:  .LBB0_2: ; %shader_eval_surface.exit
-; MUBUF11-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; MUBUF11-NEXT:    s_endpgm
 ;
 ; FLATSCR11-LABEL: kernel_background_evaluate:
 ; FLATSCR11:       ; %bb.0: ; %entry
 ; FLATSCR11-NEXT:    s_load_b32 s2, s[0:1], 0x24
 ; FLATSCR11-NEXT:    v_mov_b32_e32 v1, 0x2000
 ; FLATSCR11-NEXT:    v_dual_mov_b32 v2, 0x4000 :: v_dual_mov_b32 v3, 0
 ; FLATSCR11-NEXT:    v_mov_b32_e32 v4, 0x400000

 ; FLATSCR11-NEXT:    s_movk_i32 s0, 0x4000
 ; FLATSCR11-NEXT:    scratch_load_b64 v[0:1], off, s0 offset:4
 ; FLATSCR11-NEXT:    s_mov_b32 s0, 0x41c64e6d
 ; FLATSCR11-NEXT:    s_waitcnt vmcnt(0)
 ; FLATSCR11-NEXT:    v_add_nc_u32_e32 v2, v1, v0
 ; FLATSCR11-NEXT:    v_mad_u64_u32 v[0:1], null, v2, s0, 0x3039
 ; FLATSCR11-NEXT:    scratch_store_b32 off, v0, s0
 ; FLATSCR11-NEXT:  .LBB0_2: ; %shader_eval_surface.exit
-; FLATSCR11-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; FLATSCR11-NEXT:    s_endpgm
 entry:
   %sd = alloca < 1339 x i32>, align 8192, addrspace(5)
   %state = alloca <4 x i32>, align 16, addrspace(5)
   %rslt = call i32 @svm_eval_nodes(ptr addrspace(5) %kg, ptr addrspace(5) %sd, ptr addrspace(5) %state, i32 0, i32 4194304)
   %cmp = icmp eq i32 %rslt, 0
   br i1 %cmp, label %shader_eval_surface.exit, label %if.then4.i
 

llvm/test/CodeGen/AMDGPU/waitcnt-preexisting-vscnt.mir

bb.0:
     ; GFX11-NEXT: S_WAITCNT 0
     ; GFX11-NEXT: S_WAITCNT_VSCNT undef $sgpr_null, 0
     ; GFX11-NEXT: GLOBAL_STORE_DWORD $vgpr0_vgpr1, $vgpr2, 0, 0, implicit $exec
     ; GFX11-NEXT: S_WAITCNT_VSCNT undef $sgpr_null, 0
     ; GFX11-NEXT: S_BARRIER
     ; GFX11-NEXT: $vgpr0 = FLAT_LOAD_DWORD $vgpr0_vgpr1, 0, 0, implicit $exec, implicit $flat_scr
     ; GFX11-NEXT: S_WAITCNT 7
     ; GFX11-NEXT: FLAT_STORE_DWORD $vgpr0_vgpr1, $vgpr0, 0, 0, implicit $exec, implicit $flat_scr
-    ; GFX11-NEXT: S_SENDMSG 3, implicit $exec, implicit $m0
     ; GFX11-NEXT: S_ENDPGM 0
     GLOBAL_STORE_DWORD $vgpr0_vgpr1, $vgpr2, 0, 0, implicit $exec
     S_WAITCNT_VSCNT undef $sgpr_null, 0
     S_BARRIER
     $vgpr0 = FLAT_LOAD_DWORD $vgpr0_vgpr1, 0, 0, implicit $exec, implicit $flat_scr
     FLAT_STORE_DWORD $vgpr0_vgpr1, $vgpr0, 0, 0, implicit $exec, implicit $flat_scr
     S_ENDPGM 0
 ...
bb.0:
     ; GFX11-NEXT: S_WAITCNT 0
     ; GFX11-NEXT: S_WAITCNT_VSCNT undef $sgpr_null, 0
     ; GFX11-NEXT: GLOBAL_STORE_DWORD $vgpr0_vgpr1, $vgpr2, 0, 0, implicit $exec
     ; GFX11-NEXT: S_WAITCNT_VSCNT undef $sgpr_null, 0
     ; GFX11-NEXT: S_BARRIER
     ; GFX11-NEXT: $vgpr0 = FLAT_LOAD_DWORD $vgpr0_vgpr1, 0, 0, implicit $exec, implicit $flat_scr
     ; GFX11-NEXT: S_WAITCNT 7
     ; GFX11-NEXT: FLAT_STORE_DWORD $vgpr0_vgpr1, $vgpr0, 0, 0, implicit $exec, implicit $flat_scr
-    ; GFX11-NEXT: S_SENDMSG 3, implicit $exec, implicit $m0
     ; GFX11-NEXT: S_ENDPGM 0
     GLOBAL_STORE_DWORD $vgpr0_vgpr1, $vgpr2, 0, 0, implicit $exec
     S_WAITCNT_VSCNT undef $sgpr_null, 1
     S_BARRIER
     $vgpr0 = FLAT_LOAD_DWORD $vgpr0_vgpr1, 0, 0, implicit $exec, implicit $flat_scr
     FLAT_STORE_DWORD $vgpr0_vgpr1, $vgpr0, 0, 0, implicit $exec, implicit $flat_scr
     S_ENDPGM 0
 ...
bb.0:
     ; GFX11-NEXT: S_WAITCNT_VSCNT undef $sgpr_null, 0
     ; GFX11-NEXT: GLOBAL_STORE_DWORD $vgpr0_vgpr1, $vgpr2, 0, 0, implicit $exec
     ; GFX11-NEXT: S_WAITCNT 112
     ; GFX11-NEXT: S_WAITCNT_VSCNT undef $sgpr_null, 0
     ; GFX11-NEXT: S_BARRIER
     ; GFX11-NEXT: $vgpr0 = FLAT_LOAD_DWORD $vgpr0_vgpr1, 0, 0, implicit $exec, implicit $flat_scr
     ; GFX11-NEXT: S_WAITCNT 7
     ; GFX11-NEXT: FLAT_STORE_DWORD $vgpr0_vgpr1, $vgpr0, 0, 0, implicit $exec, implicit $flat_scr
-    ; GFX11-NEXT: S_SENDMSG 3, implicit $exec, implicit $m0
     ; GFX11-NEXT: S_ENDPGM 0
     GLOBAL_STORE_DWORD $vgpr0_vgpr1, $vgpr2, 0, 0, implicit $exec
     S_WAITCNT 112
     S_WAITCNT_VSCNT undef $sgpr_null, 0
     S_BARRIER
     $vgpr0 = FLAT_LOAD_DWORD $vgpr0_vgpr1, 0, 0, implicit $exec, implicit $flat_scr
     FLAT_STORE_DWORD $vgpr0_vgpr1, $vgpr0, 0, 0, implicit $exec, implicit $flat_scr
     S_ENDPGM 0
bb.0:
     ; GFX11-NEXT: S_WAITCNT 0
     ; GFX11-NEXT: S_WAITCNT_VSCNT undef $sgpr_null, 0
     ; GFX11-NEXT: GLOBAL_STORE_DWORD $vgpr0_vgpr1, $vgpr2, 0, 0, implicit $exec
     ; GFX11-NEXT: S_WAITCNT_VSCNT undef $sgpr_null, 0
     ; GFX11-NEXT: S_BARRIER
     ; GFX11-NEXT: $vgpr0 = FLAT_LOAD_DWORD $vgpr0_vgpr1, 0, 0, implicit $exec, implicit $flat_scr
     ; GFX11-NEXT: S_WAITCNT 7
     ; GFX11-NEXT: FLAT_STORE_DWORD $vgpr0_vgpr1, $vgpr0, 0, 0, implicit $exec, implicit $flat_scr
-    ; GFX11-NEXT: S_SENDMSG 3, implicit $exec, implicit $m0
     ; GFX11-NEXT: S_ENDPGM 0
     GLOBAL_STORE_DWORD $vgpr0_vgpr1, $vgpr2, 0, 0, implicit $exec
     S_WAITCNT_VSCNT undef $sgpr_null, 0
     S_WAITCNT_VSCNT undef $sgpr_null, 1
     S_WAITCNT_VSCNT undef $sgpr_null, 2
     S_BARRIER
     $vgpr0 = FLAT_LOAD_DWORD $vgpr0_vgpr1, 0, 0, implicit $exec, implicit $flat_scr
     FLAT_STORE_DWORD $vgpr0_vgpr1, $vgpr0, 0, 0, implicit $exec, implicit $flat_scr
bb.0:
     ; GFX11-NEXT: S_WAITCNT_VSCNT undef $sgpr_null, 0
     ; GFX11-NEXT: GLOBAL_STORE_DWORD $vgpr0_vgpr1, $vgpr2, 0, 0, implicit $exec
     ; GFX11-NEXT: S_WAITCNT 0
     ; GFX11-NEXT: S_WAITCNT_VSCNT undef $sgpr_null, 0
     ; GFX11-NEXT: S_BARRIER
     ; GFX11-NEXT: $vgpr0 = FLAT_LOAD_DWORD $vgpr0_vgpr1, 0, 0, implicit $exec, implicit $flat_scr
     ; GFX11-NEXT: S_WAITCNT 7
     ; GFX11-NEXT: FLAT_STORE_DWORD $vgpr0_vgpr1, $vgpr0, 0, 0, implicit $exec, implicit $flat_scr
-    ; GFX11-NEXT: S_SENDMSG 3, implicit $exec, implicit $m0
     ; GFX11-NEXT: S_ENDPGM 0
     GLOBAL_STORE_DWORD $vgpr0_vgpr1, $vgpr2, 0, 0, implicit $exec
     S_WAITCNT 0
     S_WAITCNT_VSCNT undef $sgpr_null, 1
     S_WAITCNT 0
     S_WAITCNT_VSCNT undef $sgpr_null, 2
     S_WAITCNT 0
     S_BARRIER
     $vgpr0 = FLAT_LOAD_DWORD $vgpr0_vgpr1, 0, 0, implicit $exec, implicit $flat_scr
     FLAT_STORE_DWORD $vgpr0_vgpr1, $vgpr0, 0, 0, implicit $exec, implicit $flat_scr
     S_ENDPGM 0
 ...