test/CodeGen/AMDGPU/waitcnt-buffer-inv.mir

# RUN: llc -mtriple=amdgpu9.42-amd-amdhsa -verify-machineinstrs -run-pass=si-insert-waitcnts -o - %s | FileCheck --check-prefix=GFX94 %s
# RUN: llc -mtriple=amdgpu9.50-amd-amdhsa -verify-machineinstrs -run-pass=si-insert-waitcnts -o - %s | FileCheck --check-prefix=GFX94 %s

---
name:            buffer_inv_advances_vmcnt
tracksRegLiveness: true
machineFunctionInfo:
  isEntryFunction: true
body:             |
  bb.0:
    liveins: $exec, $vgpr0_vgpr1

    ; GFX94-LABEL: name: buffer_inv_advances_vmcnt
    ; GFX94: $vgpr2 = GLOBAL_LOAD_DWORD $vgpr0_vgpr1, 0, 0, implicit $exec
    ; GFX94-NEXT: $vgpr3 = GLOBAL_LOAD_DWORD $vgpr0_vgpr1, 0, 0, implicit $exec
    ; GFX94-NEXT: BUFFER_INV 0, implicit $exec
    ; GFX94-NEXT: $vgpr4 = GLOBAL_LOAD_DWORD $vgpr0_vgpr1, 0, 0, implicit $exec
    ; GFX94-NEXT: S_WAITCNT .Vmcnt_2
    ; GFX94-NEXT: $vgpr5 = V_MOV_B32_e32 $vgpr3, implicit $exec
    $vgpr2 = GLOBAL_LOAD_DWORD $vgpr0_vgpr1, 0, 0, implicit $exec
    $vgpr3 = GLOBAL_LOAD_DWORD $vgpr0_vgpr1, 0, 0, implicit $exec
    BUFFER_INV 0, implicit $exec
    $vgpr4 = GLOBAL_LOAD_DWORD $vgpr0_vgpr1, 0, 0, implicit $exec
    $vgpr5 = V_MOV_B32_e32 $vgpr3, implicit $exec
    S_ENDPGM 0
...

---
name:            buffer_inv_only_does_not_wait_at_return
tracksRegLiveness: true
machineFunctionInfo:
  isEntryFunction: false
body:             |
  bb.0:
    liveins: $sgpr30_sgpr31

    ; GFX94-LABEL: name: buffer_inv_only_does_not_wait_at_return
    ; GFX94: S_WAITCNT
    ; GFX94-NEXT: BUFFER_INV 0, implicit $exec
    ; GFX94-NOT: S_WAITCNT
    ; GFX94-NEXT: S_SETPC_B64_return $sgpr30_sgpr31
    BUFFER_INV 0, implicit $exec
    S_WAITCNT_soft 0
    S_SETPC_B64_return $sgpr30_sgpr31
...