[AMDGPU] Require waitcnt before barrier for all targets; adjust tests.

[msearles]

Summary

[AMDGPU] Require waitcnt before barrier for all targets; adjust tests.

[rampitec]

LGTM

[kzhuravl]

You can also combine gfx8/gfx9 s_waitcnt check into gcn check

[msearles]

Update test - combine GFX8/GFX9 s_waitcnt check into GCN check

[rampitec]

Missing the source itself ;)

[msearles]

Doh! Add lib/Target/AMDGPU/AMDGPUSubtarget.h back into the diff.

[arsenm]

Maybe this should be a full subtarget feature so it can be disabled

[msearles]

Removing content of this comment as it pertained to a different revision. Sorry for the noise.

[msearles]

Ug. Fix revision; added wrong set of diffs; correcting.

[msearles]

r304217

lib/Target/AMDGPU/AMDGPUTargetMachine.cpp

 static cl::opt<bool> EnableSIInsertWaitcntsPass(
   "enable-si-insert-waitcnts",
   cl::desc("Use new waitcnt insertion pass"),
-  cl::init(false));
+  cl::init(true));
 
 // Option to run late CFG structurizer
 static cl::opt<bool> LateCFGStructurize(

lib/Target/AMDGPU/DSInstructions.td

   [(set i32:$vdst,
    (node (DS1Addr1Offset i32:$addr, i16:$offset), i32:$data0))] > {
 
-  let mayLoad = 0;
+  let mayLoad = 1;
   let mayStore = 0;
   let isConvergent = 1;
 

 def DS_WRITE_SRC2_B32 : DS_1A<"ds_write_src2_b32">;
 def DS_WRITE_SRC2_B64 : DS_1A<"ds_write_src2_b64">;
 
-let Uses = [EXEC], mayLoad = 0, mayStore = 0, isConvergent = 1 in {
+let Uses = [EXEC], mayLoad = 1, mayStore = 0, isConvergent = 1 in {
 def DS_SWIZZLE_B32 : DS_1A_RET <"ds_swizzle_b32">;
 }
 

lib/Target/AMDGPU/SIInsertWaitcnts.cpp

   // NOTE: this could be improved with knowledge of all call sites or
   //   with knowledge of the called routines.
   if (MI.getOpcode() == AMDGPU::RETURN ||
-      MI.getOpcode() == AMDGPU::SI_RETURN_TO_EPILOG) {
+      MI.getOpcode() == AMDGPU::SI_RETURN_TO_EPILOG ||
+      MI.getOpcode() == AMDGPU::S_SETPC_B64_return) {
     for (enum InstCounterType T = VM_CNT; T < NUM_INST_CNTS;
          T = (enum InstCounterType)(T + 1)) {
       if (ScoreBrackets->getScoreUB(T) > ScoreBrackets->getScoreLB(T)) {

   // bracket and the destination operand scores.
   // TODO: Use the (TSFlags & SIInstrFlags::LGKM_CNT) property everywhere.
   if (TII->isDS(Inst) && (Inst.mayLoad() || Inst.mayStore())) {
-    if (TII->getNamedOperand(Inst, AMDGPU::OpName::gds)->getImm() != 0) {
+    if (TII->getNamedOperand(Inst, AMDGPU::OpName::gds) &&
+       TII->getNamedOperand(Inst, AMDGPU::OpName::gds)->getImm() != 0) {
       ScoreBrackets->updateByEvent(TII, TRI, MRI, GDS_ACCESS, Inst);
       ScoreBrackets->updateByEvent(TII, TRI, MRI, GDS_GPR_LOCK, Inst);
     } else {

              Inst.getOpcode() != AMDGPU::BUFFER_WBINVL1_VOL) {
     ScoreBrackets->updateByEvent(TII, TRI, MRI, VMEM_ACCESS, Inst);
     if ( // TODO: assumed yes -- target_info->MemWriteNeedsExpWait() &&
-        (Inst.mayStore() || AMDGPU::getAtomicNoRetOp(Inst.getOpcode()))) {
+        (Inst.mayStore() || AMDGPU::getAtomicNoRetOp(Inst.getOpcode()) != -1)) {
       ScoreBrackets->updateByEvent(TII, TRI, MRI, VMW_GPR_LOCK, Inst);
     }
   } else if (TII->isSMRD(Inst)) {

   MRI = &MF.getRegInfo();
   MLI = &getAnalysis<MachineLoopInfo>();
   IV = AMDGPU::IsaInfo::getIsaVersion(ST->getFeatureBits());
+  const SIMachineFunctionInfo *MFI = MF.getInfo<SIMachineFunctionInfo>();
   AMDGPUASI = ST->getAMDGPUAS();
 
   HardwareLimits.VmcntMax = AMDGPU::getVmcntBitMask(IV);

     }
   }
 
+  if (!MFI->isEntryFunction()) {
+    // Wait for any outstanding memory operations that the input registers may
+    // depend on. We can't track them and it's better to to the wait after the
+    // costly call sequence.
+
+    // TODO: Could insert earlier and schedule more liberally with operations
+    // that only use caller preserved registers.
+    MachineBasicBlock &EntryBB = MF.front();
+    BuildMI(EntryBB, EntryBB.getFirstNonPHI(), DebugLoc(), TII->get(AMDGPU::S_WAITCNT))
+      .addImm(0);
+
+    Modified = true;
+  }
+
   return Modified;
 }

test/CodeGen/AMDGPU/basic-branch.ll

 ; GCN: s_cbranch_vccnz [[END:BB[0-9]+_[0-9]+]]
 
 ; GCN: buffer_store_dword
-; GCNOPT-NEXT: s_waitcnt vmcnt(0) expcnt(0)
-; TODO: This waitcnt can be eliminated
 
 ; GCN: {{^}}[[END]]:
 ; GCN: s_endpgm

test/CodeGen/AMDGPU/branch-condition-and.ll

 
 ; GCN-NEXT: BB{{[0-9]+_[0-9]+}}: ; %bb4
 ; GCN: ds_write_b32
-; GCN: s_waitcnt
 
-; GCN-NEXT: [[BB5]]
+; GCN: [[BB5]]
 ; GCN: s_or_b64 exec, exec
 ; GCN-NEXT: s_endpgm
 ; GCN-NEXT: .Lfunc_end

test/CodeGen/AMDGPU/branch-relaxation.ll

 ; GCN-NEXT: [[BB2]]: ; %bb2
 ; GCN: v_mov_b32_e32 [[BB2_K:v[0-9]+]], 17
 ; GCN: buffer_store_dword [[BB2_K]]
-; GCN: s_waitcnt vmcnt(0)
 
 ; GCN-NEXT: [[LONG_JUMP1:BB[0-9]+_[0-9]+]]: ; %bb2
 ; GCN-NEXT: s_getpc_b64 vcc

 
 ; GCN-NEXT: ; BB#2: ; %if_uniform
 ; GCN: buffer_store_dword
-; GCN: s_waitcnt vmcnt(0)
 
 ; GCN-NEXT: [[ENDIF]]: ; %endif
 ; GCN-NEXT: s_or_b64 exec, exec, [[MASK]]

test/CodeGen/AMDGPU/control-flow-fastregalloc.ll

 
 ; GCN: s_mov_b64 exec, s{{\[}}[[ANDEXEC_LO]]:[[ANDEXEC_HI]]{{\]}}
 
-; GCN: s_waitcnt vmcnt(0) expcnt(0)
 ; GCN: mask branch [[ENDIF:BB[0-9]+_[0-9]+]]
 
 ; GCN: {{^}}BB{{[0-9]+}}_1: ; %if
 ; GCN: s_mov_b32 m0, -1
 ; GCN: ds_read_b32 [[LOAD1:v[0-9]+]]
+; GCN: s_waitcnt lgkmcnt(0)
 ; GCN: buffer_load_dword [[RELOAD_LOAD0:v[0-9]+]], off, s[0:3], s7 offset:[[LOAD0_OFFSET]] ; 4-byte Folded Reload
-; GCN: s_waitcnt vmcnt(0)
 
 ; Spill val register
 ; GCN: v_add_i32_e32 [[VAL:v[0-9]+]], vcc, [[LOAD1]], [[RELOAD_LOAD0]]
 ; GCN: buffer_store_dword [[VAL]], off, s[0:3], s7 offset:[[VAL_OFFSET:[0-9]+]] ; 4-byte Folded Spill
-; GCN: s_waitcnt vmcnt(0)
 
 ; VMEM: [[ENDIF]]:
 ; Reload and restore exec mask
+; VGPR: s_waitcnt lgkmcnt(0)
 ; VGPR: v_readlane_b32 s[[S_RELOAD_SAVEEXEC_LO:[0-9]+]], [[SPILL_VGPR]], [[SAVEEXEC_LO_LANE]]
 ; VGPR: v_readlane_b32 s[[S_RELOAD_SAVEEXEC_HI:[0-9]+]], [[SPILL_VGPR]], [[SAVEEXEC_HI_LANE]]
 

 
 ; GCN: s_mov_b64 exec, s{{\[}}[[ANDEXEC_LO]]:[[ANDEXEC_HI]]{{\]}}
 
-; GCN: s_waitcnt vmcnt(0) expcnt(0)
 ; GCN-NEXT: ; mask branch [[END:BB[0-9]+_[0-9]+]]
 ; GCN-NEXT: s_cbranch_execz [[END]]
 

 ; GCN: v_cmp_ne_u32_e32 vcc,
 ; GCN: s_and_b64 vcc, exec, vcc
 ; GCN: buffer_store_dword [[VAL_LOOP]], off, s[0:3], s7 offset:[[VAL_SUB_OFFSET:[0-9]+]] ; 4-byte Folded Spill
-; GCN: s_waitcnt vmcnt(0) expcnt(0)
 ; GCN-NEXT: s_cbranch_vccnz [[LOOP]]
 
 

 ; VMEM: buffer_store_dword v[[V_SAVEEXEC_HI]], off, s[0:3], s7 offset:[[SAVEEXEC_HI_OFFSET:[0-9]+]] ; 4-byte Folded Spill
 
 ; GCN: s_mov_b64 exec, [[CMP0]]
-; GCN: s_waitcnt vmcnt(0) expcnt(0)
 
 ; FIXME: It makes no sense to put this skip here
 ; GCN-NEXT: ; mask branch [[FLOW:BB[0-9]+_[0-9]+]]

 
 ; GCN: buffer_store_dword [[FLOW_VAL]], off, s[0:3], s7 offset:[[RESULT_OFFSET:[0-9]+]] ; 4-byte Folded Spill
 ; GCN: s_xor_b64 exec, exec, s{{\[}}[[FLOW_S_RELOAD_SAVEEXEC_LO]]:[[FLOW_S_RELOAD_SAVEEXEC_HI]]{{\]}}
-; GCN-NEXT: s_waitcnt vmcnt(0) expcnt(0)
 ; GCN-NEXT: ; mask branch [[ENDIF:BB[0-9]+_[0-9]+]]
 ; GCN-NEXT: s_cbranch_execz [[ENDIF]]
 

 ; GCN: buffer_load_dword v[[LOAD0_RELOAD:[0-9]+]], off, s[0:3], s7 offset:4 ; 4-byte Folded Reload
 ; GCN: v_add_i32_e32 [[ADD:v[0-9]+]], vcc, v{{[0-9]+}}, v[[LOAD0_RELOAD]]
 ; GCN: buffer_store_dword [[ADD]], off, s[0:3], s7 offset:[[RESULT_OFFSET]] ; 4-byte Folded Spill
-; GCN: s_waitcnt vmcnt(0) expcnt(0)
 ; GCN-NEXT: s_branch [[ENDIF:BB[0-9]+_[0-9]+]]
 
 ; GCN: [[ELSE]]: ; %else
 ; GCN: buffer_load_dword v[[LOAD0_RELOAD:[0-9]+]], off, s[0:3], s7 offset:4 ; 4-byte Folded Reload
 ; GCN: v_subrev_i32_e32 [[SUB:v[0-9]+]], vcc, v{{[0-9]+}}, v[[LOAD0_RELOAD]]
 ; GCN: buffer_store_dword [[ADD]], off, s[0:3], s7 offset:[[FLOW_RESULT_OFFSET:[0-9]+]] ; 4-byte Folded Spill
-; GCN: s_waitcnt vmcnt(0) expcnt(0)
 ; GCN-NEXT: s_branch [[FLOW]]
 
 ; GCN: [[ENDIF]]:

test/CodeGen/AMDGPU/indirect-addressing-si.ll

 ; FIXME: The waitcnt for the argument load can go after the loop
 ; IDXMODE: s_set_gpr_idx_on 0, src0
 ; GCN: s_mov_b64 s{{\[[0-9]+:[0-9]+\]}}, exec
-; GCN: s_waitcnt lgkmcnt(0)
-
+; GCN: [[LOOPBB:BB[0-9]+_[0-9]+]]:
 ; GCN: v_readfirstlane_b32 [[READLANE:s[0-9]+]], v{{[0-9]+}}
 
 ; MOVREL: s_add_i32 m0, [[READLANE]], 0xfffffe0

 ; GCN-DAG: v_mov_b32_e32 [[VEC_ELT3:v[0-9]+]], 4{{$}}
 
 ; GCN: s_mov_b64 [[SAVEEXEC:s\[[0-9]+:[0-9]+\]]], exec
-; GCN: s_waitcnt lgkmcnt(0)
-
 ; GCN: [[LOOPBB:BB[0-9]+_[0-9]+]]:
 ; GCN: v_readfirstlane_b32 [[READLANE:s[0-9]+]]
 

 ; IDXMODE: s_set_gpr_idx_on 0, dst
 
 ; GCN: s_mov_b64 [[SAVEEXEC:s\[[0-9]+:[0-9]+\]]], exec
-; GCN: s_waitcnt lgkmcnt(0)
 
 ; The offset depends on the register that holds the first element of the vector.
 ; GCN: v_readfirstlane_b32 [[READLANE:s[0-9]+]]

 ; IDXMODE: s_set_gpr_idx_on 0, src0
 
 ; GCN: s_mov_b64 [[MASK:s\[[0-9]+:[0-9]+\]]], exec
-; GCN: s_waitcnt vmcnt(0)
 
 ; GCN: [[LOOP0:BB[0-9]+_[0-9]+]]:
+; GCN-NEXT: s_waitcnt vmcnt(0)
 ; GCN-NEXT: v_readfirstlane_b32 [[READLANE:s[0-9]+]], [[IDX0]]
 ; GCN: v_cmp_eq_u32_e32 vcc, [[READLANE]], [[IDX0]]
 

 ; IDXMODE: s_set_gpr_idx_on 0, dst
 
 ; GCN: [[LOOP0:BB[0-9]+_[0-9]+]]:
+; GCN-NEXT: s_waitcnt vmcnt(0)
 ; GCN-NEXT: v_readfirstlane_b32 [[READLANE:s[0-9]+]], [[IDX0]]
 ; GCN: v_cmp_eq_u32_e32 vcc, [[READLANE]], [[IDX0]]
 

test/CodeGen/AMDGPU/infinite-loop.ll

 ; SI-LABEL: {{^}}infinite_loop:
 ; SI: v_mov_b32_e32 [[REG:v[0-9]+]], 0x3e7
 ; SI: BB0_1:
+; SI: s_waitcnt lgkmcnt(0)
 ; SI: buffer_store_dword [[REG]]
-; SI: s_waitcnt vmcnt(0) expcnt(0)
 ; SI: s_branch BB0_1
 define amdgpu_kernel void @infinite_loop(i32 addrspace(1)* %out) {
 entry:

test/CodeGen/AMDGPU/llvm.amdgcn.buffer.store.format.ll

 ;
 ;CHECK-LABEL: {{^}}buffer_store_wait:
 ;CHECK: buffer_store_format_xyzw v[0:3], v4, s[0:3], 0 idxen
-;CHECK: s_waitcnt vmcnt(0) expcnt(0)
+;CHECK: s_waitcnt expcnt(0)
 ;CHECK: buffer_load_format_xyzw v[0:3], v5, s[0:3], 0 idxen
 ;CHECK: s_waitcnt vmcnt(0)
 ;CHECK: buffer_store_format_xyzw v[0:3], v6, s[0:3], 0 idxen

test/CodeGen/AMDGPU/llvm.amdgcn.buffer.store.ll

 ;
 ;CHECK-LABEL: {{^}}buffer_store_wait:
 ;CHECK: buffer_store_dwordx4 v[0:3], v4, s[0:3], 0 idxen
-;CHECK: s_waitcnt vmcnt(0) expcnt(0)
+;CHECK: s_waitcnt expcnt(0)
 ;CHECK: buffer_load_dwordx4 v[0:3], v5, s[0:3], 0 idxen
 ;CHECK: s_waitcnt vmcnt(0)
 ;CHECK: buffer_store_dwordx4 v[0:3], v6, s[0:3], 0 idxen

test/CodeGen/AMDGPU/llvm.amdgcn.ds.swizzle.ll

 
 ; FUNC-LABEL: {{^}}ds_swizzle:
 ; CHECK: ds_swizzle_b32 v{{[0-9]+}}, v{{[0-9]+}} offset:100
-; CHECK: s_waitcnt lgkmcnt
 define amdgpu_kernel void @ds_swizzle(i32 addrspace(1)* %out, i32 %src) nounwind {
   %swizzle = call i32 @llvm.amdgcn.ds.swizzle(i32 %src, i32 100) #0
   store i32 %swizzle, i32 addrspace(1)* %out, align 4

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

 ;
 ; GCN-LABEL: {{^}}image_store_wait:
 ; GCN: image_store v[0:3], v4, s[0:7] dmask:0xf unorm
-; GCN: s_waitcnt vmcnt(0) expcnt(0)
+; GCN: s_waitcnt expcnt(0)
 ; GCN: image_load v[0:3], v4, s[8:15] dmask:0xf unorm
 ; GCN: s_waitcnt vmcnt(0)
 ; GCN: image_store v[0:3], v4, s[16:23] dmask:0xf unorm

test/CodeGen/AMDGPU/llvm.amdgcn.s.dcache.inv.ll

 ; GCN: s_waitcnt lgkmcnt(0) ; encoding
 define amdgpu_kernel void @test_s_dcache_inv_insert_wait() #0 {
   call void @llvm.amdgcn.s.dcache.inv()
-  call void @llvm.amdgcn.s.waitcnt(i32 0)
+  call void @llvm.amdgcn.s.waitcnt(i32 127)
   br label %end
 
 end:

test/CodeGen/AMDGPU/llvm.amdgcn.s.dcache.inv.vol.ll

 ; GCN: s_waitcnt lgkmcnt(0) ; encoding
 define amdgpu_kernel void @test_s_dcache_inv_vol_insert_wait() #0 {
   call void @llvm.amdgcn.s.dcache.inv.vol()
-  call void @llvm.amdgcn.s.waitcnt(i32 0)
+  call void @llvm.amdgcn.s.waitcnt(i32 127)
   br label %end
 
 end:

test/CodeGen/AMDGPU/llvm.amdgcn.s.dcache.wb.ll

 ; VI: s_waitcnt lgkmcnt(0) ; encoding
 define amdgpu_kernel void @test_s_dcache_wb_insert_wait() #0 {
   call void @llvm.amdgcn.s.dcache.wb()
-  call void @llvm.amdgcn.s.waitcnt(i32 0)
+  call void @llvm.amdgcn.s.waitcnt(i32 127)
   br label %end
 
 end:

test/CodeGen/AMDGPU/llvm.amdgcn.s.dcache.wb.vol.ll

 ; VI: s_waitcnt lgkmcnt(0) ; encoding
 define amdgpu_kernel void @test_s_dcache_wb_vol_insert_wait() #0 {
   call void @llvm.amdgcn.s.dcache.wb.vol()
-  call void @llvm.amdgcn.s.waitcnt(i32 0)
+  call void @llvm.amdgcn.s.waitcnt(i32 127)
   br label %end
 
 end:

test/CodeGen/AMDGPU/llvm.amdgcn.s.waitcnt.ll

 ;
 ; CHECK-LABEL: {{^}}test2:
 ; CHECK: image_load
-; CHECK-NOT: s_waitcnt vmcnt(0){{$}}
-; CHECK: s_waitcnt
+; CHECK-NEXT: s_waitcnt
+; CHECK: s_waitcnt vmcnt(0){{$}}
 ; CHECK-NEXT: image_store
 define amdgpu_ps void @test2(<8 x i32> inreg %rsrc, i32 %c) {
   %t = call <4 x float> @llvm.amdgcn.image.load.v4f32.i32.v8i32(i32 %c, <8 x i32> %rsrc, i32 15, i1 0, i1 0, i1 0, i1 0)

test/CodeGen/AMDGPU/multi-divergent-exit-region.ll

 
 ; GCN: {{^BB[0-9]+_[0-9]+}}: ; %UnifiedReturnBlock
 ; GCN-NEXT: s_or_b64 exec, exec
+; GCN-NEXT: s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GCN-NEXT: ; return
 
 define amdgpu_ps float @uniform_branch_to_multi_divergent_region_exit_ret_ret_return_value(i32 inreg %sgpr, i32 %vgpr) #0 {

test/CodeGen/AMDGPU/ret_jump.ll

 
 ; GCN-NEXT:  ; %unreachable.bb
 ; GCN: ds_write_b32
-; GCN: s_waitcnt
 ; GCN: ; divergent unreachable
 
 ; GCN: ; %ret.bb

 
 ; GCN: ; %UnifiedReturnBlock
 ; GCN-NEXT: s_or_b64 exec, exec
+; GCN-NEXT: s_waitcnt
 ; GCN-NEXT: ; return
 ; GCN-NEXT: .Lfunc_end
 define amdgpu_ps <{ i32, i32, i32, i32, i32, i32, i32, i32, i32, float, float, float, float, float, float, float, float, float, float, float, float, float, float }> @uniform_br_nontrivial_ret_divergent_br_nontrivial_unreachable([9 x <16 x i8>] addrspace(2)* byval %arg, [17 x <16 x i8>] addrspace(2)* byval %arg1, [17 x <8 x i32>] addrspace(2)* byval %arg2, i32 addrspace(2)* byval %arg3, float inreg %arg4, i32 inreg %arg5, <2 x i32> %arg6, <2 x i32> %arg7, <2 x i32> %arg8, <3 x i32> %arg9, <2 x i32> %arg10, <2 x i32> %arg11, <2 x i32> %arg12, float %arg13, float %arg14, float %arg15, float %arg16, float %arg17, i32 inreg %arg18, i32 %arg19, float %arg20, i32 %arg21) #0 {

test/CodeGen/AMDGPU/si-lower-control-flow-unreachable-block.ll

 ; GCN-NEXT: BB{{[0-9]+_[0-9]+}}: ; %unreachable
 ; GCN: ds_write_b32
 ; GCN: ; divergent unreachable
-; GCN: s_waitcnt
 
 ; GCN-NEXT: [[RET]]: ; %UnifiedReturnBlock
 ; GCN-NEXT: s_or_b64 exec, exec

 ; GCN-NEXT: {{^BB[0-9]+_[0-9]+}}: ; %unreachable
 ; GCN: ds_write_b32
 ; GCN: ; divergent unreachable
-; GCN: s_waitcnt
 
 ; GCN: [[RETURN]]:
 ; GCN-NEXT: s_or_b64 exec, exec

 
 ; GCN: [[UNREACHABLE]]:
 ; GCN: ds_write_b32
-; GCN: s_waitcnt
 define amdgpu_kernel void @uniform_lower_control_flow_unreachable_terminator(i32 %arg0) #0 {
 bb:
   %tmp63 = icmp eq i32 %arg0, 32

test/CodeGen/AMDGPU/smrd-vccz-bug.ll

 ; GCN-FUNC: {{^}}vccz_workaround:
 ; GCN: s_load_dword s{{[0-9]+}}, s[{{[0-9]+:[0-9]+}}], 0x0
 ; GCN: v_cmp_neq_f32_e64 vcc, s{{[0-9]+}}, 0{{$}}
-; GCN: s_waitcnt lgkmcnt(0)
+; VCCZ-BUG: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
 ; VCCZ-BUG: s_mov_b64 vcc, vcc
 ; NOVCCZ-BUG-NOT: s_mov_b64 vcc, vcc
 ; GCN: s_cbranch_vccnz [[EXIT:[0-9A-Za-z_]+]]

test/CodeGen/AMDGPU/spill-m0.ll

 ; TOVMEM-DAG: s_mov_b32 [[M0_COPY:s[0-9]+]], m0
 ; TOVMEM-DAG: v_mov_b32_e32 [[SPILL_VREG:v[0-9]+]], [[M0_COPY]]
 ; TOVMEM: buffer_store_dword [[SPILL_VREG]], off, s{{\[[0-9]+:[0-9]+\]}}, s{{[0-9]+}} offset:4 ; 4-byte Folded Spill
-; TOVMEM: s_waitcnt vmcnt(0)
 
 ; TOSMEM-DAG: s_mov_b32 [[M0_COPY:s[0-9]+]], m0
 ; TOSMEM: s_add_u32 m0, s3, 0x100{{$}}
 ; TOSMEM-NOT: [[M0_COPY]]
 ; TOSMEM: s_buffer_store_dword [[M0_COPY]], s{{\[}}[[LO]]:[[HI]]], m0 ; 4-byte Folded Spill
-; TOSMEM: s_waitcnt lgkmcnt(0)
 
 ; GCN: s_cbranch_scc1 [[ENDIF:BB[0-9]+_[0-9]+]]
 

test/CodeGen/AMDGPU/valu-i1.ll

 ; SI: v_cmp_lt_i32_e32 vcc, 0,
 ; SI-NEXT: s_and_saveexec_b64 [[SAVE1:s\[[0-9]+:[0-9]+\]]], vcc
 ; SI-NEXT: s_xor_b64 [[SAVE2:s\[[0-9]+:[0-9]+\]]], exec, [[SAVE1]]
-; SI-NEXT: s_waitcnt lgkmcnt(0)
 ; SI-NEXT: ; mask branch [[FLOW_BB:BB[0-9]+_[0-9]+]]
 ; SI-NEXT: s_cbranch_execz [[FLOW_BB]]
 

 
 ; SI-NEXT: BB{{[0-9]+_[0-9]+}}:
 ; SI: buffer_store_dword
-; SI-NEXT: s_waitcnt
 
 ; SI-NEXT: {{^}}[[EXIT]]:
 ; SI: s_or_b64 exec, exec, [[BR_SREG]]

 
 ; SI-NEXT: BB{{[0-9]+_[0-9]+}}:
 ; SI: buffer_store_dword
-; SI-NEXT: s_waitcnt
 
 ; SI-NEXT: {{^}}[[EXIT]]:
 ; SI: s_or_b64 exec, exec, [[BR_SREG]]

 
 ; SI-NEXT: {{^BB[0-9]+_[0-9]+}}: ; %exit
 ; SI: ds_write_b32
-; SI: s_waitcnt
 
 ; SI-NEXT: {{^}}[[FLOW]]:
 ; SI-NEXT: s_or_saveexec_b64

 ; SI-NEXT: ; mask branch [[UNIFIED_RETURN:BB[0-9]+_[0-9]+]]
 
 ; SI-NEXT: {{^BB[0-9]+_[0-9]+}}: ; %then
-; SI: buffer_store_dword
-; SI-NEXT: s_waitcnt
+; SI: s_waitcnt
+; SI-NEXT: buffer_store_dword
 
 ; SI-NEXT: {{^}}[[UNIFIED_RETURN]]: ; %UnifiedReturnBlock
 ; SI: s_or_b64 exec, exec

test/CodeGen/AMDGPU/waitcnt-permute.mir

+# RUN: llc -mtriple=amdgcn -mcpu=fiji -verify-machineinstrs -run-pass si-insert-waits -o - %s | FileCheck %s
+
+--- |
+  define float @waitcnt-permute(i32 %x, i32 %y) {
+  entry:
+    %0 = call i32 @llvm.amdgcn.ds.bpermute(i32 %x, i32 %y)
+    %1 = bitcast i32 %0 to float
+    %2 = fadd float 1.000000e+00, %1
+    ret float %2
+  }
+
+  declare i32 @llvm.amdgcn.ds.bpermute(i32, i32)
+
+...
+---
+# CHECK-LABEL: name: waitcnt-permute{{$}}
+# CHECK: DS_BPERMUTE_B32
+# CHECK-NEXT: S_WAITCNT 127
+
+name:            waitcnt-permute
+liveins:
+  - { reg: '%vgpr0' }
+  - { reg: '%vgpr1' }
+  - { reg: '%sgpr30_sgpr31' }
+body:             |
+  bb.0:
+    liveins: %vgpr0, %vgpr1, %sgpr30_sgpr31
+
+    %vgpr0 = DS_BPERMUTE_B32 killed %vgpr0, killed %vgpr1, 0, implicit %exec
+    %vgpr0 = V_ADD_F32_e32 1065353216, killed %vgpr0, implicit %exec
+    S_SETPC_B64_return killed %sgpr30_sgpr31, implicit killed %vgpr0
+
+...