llvm/test/CodeGen/AMDGPU/statepoint-insert-waitcnts.mir

# NOTE: Assertions have been autogenerated by utils/update_mir_test_checks.py UTC_ARGS: --version 6
# RUN: llc -mtriple=amdgpu9.00-amd-amdhsa -run-pass=si-insert-waitcnts -o - %s | FileCheck %s

# Make sure the waitcnt pass doesn't crash on statepoint
# pseudoinstructions, and handles the wait for the callee operand
# correctly.

---
name:            test_wait_statepoint_callee
tracksRegLiveness: true
frameInfo:
  stackSize:       16
  maxAlignment:    4
  adjustsStack:    true
  hasCalls:        true
  maxCallFrameSize: 0
  isCalleeSavedInfoValid: true
stack:
  - { id: 0, offset: 4, size: 4, alignment: 4 }
  - { id: 1, type: spill-slot, size: 4, alignment: 4 }
  - { id: 2, type: spill-slot, size: 4, alignment: 4, stack-id: sgpr-spill }
machineFunctionInfo:
  hasSpilledSGPRs: true
  scratchRSrcReg:  '$sgpr0_sgpr1_sgpr2_sgpr3'
  frameOffsetReg:  '$sgpr33'
  stackPtrOffsetReg: '$sgpr32'
  spillPhysVGPRs:
    - '$vgpr40'
  wwmReservedRegs:
    - '$vgpr40'
  scavengeFI:      '%stack.0'
body:             |
  bb.0:
    liveins: $sgpr12, $sgpr13, $sgpr14, $sgpr15, $sgpr30, $sgpr31, $vgpr31, $sgpr4_sgpr5, $sgpr6_sgpr7, $sgpr8_sgpr9, $sgpr10_sgpr11, $sgpr12_sgpr13

    ; CHECK-LABEL: name: test_wait_statepoint_callee
    ; CHECK: liveins: $sgpr12, $sgpr13, $sgpr14, $sgpr15, $sgpr30, $sgpr31, $vgpr31, $sgpr4_sgpr5, $sgpr6_sgpr7, $sgpr8_sgpr9, $sgpr10_sgpr11, $sgpr12_sgpr13
    ; CHECK-NEXT: {{  $}}
    ; CHECK-NEXT: S_WAITCNT .Vmcnt_0_Expcnt_0_Lgkmcnt_0
    ; CHECK-NEXT: $sgpr16 = S_MOV_B32 $sgpr33
    ; CHECK-NEXT: $sgpr33 = S_MOV_B32 $sgpr32
    ; CHECK-NEXT: $sgpr18_sgpr19 = S_OR_SAVEEXEC_B64 -1, implicit-def $exec, implicit-def dead $scc, implicit $exec
    ; CHECK-NEXT: BUFFER_STORE_DWORD_OFFSET $vgpr40, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr33, 0, 0, 0, implicit $exec :: (store (s32) into %stack.1, addrspace 5)
    ; CHECK-NEXT: $exec = S_MOV_B64 killed $sgpr18_sgpr19
    ; CHECK-NEXT: $vgpr40 = V_WRITELANE_B32 killed $sgpr16, 2, undef $vgpr40
    ; CHECK-NEXT: $vgpr40 = V_WRITELANE_B32 $sgpr30, 0, $vgpr40
    ; CHECK-NEXT: $sgpr32 = frame-setup S_ADDK_I32 $sgpr32, 1024, implicit-def dead $scc
    ; CHECK-NEXT: $vgpr40 = V_WRITELANE_B32 $sgpr31, 1, $vgpr40
    ; CHECK-NEXT: $sgpr14_sgpr15 = S_LOAD_DWORDX2_IMM $sgpr12_sgpr13, 0, 0
    ; CHECK-NEXT: S_WAITCNT .Lgkmcnt_0
    ; CHECK-NEXT: STATEPOINT 2882400015, 0, 11, renamable $sgpr14_sgpr15, 0, killed $sgpr4_sgpr5, killed $sgpr6_sgpr7, killed $sgpr8_sgpr9, killed $sgpr10_sgpr11, killed $sgpr12, killed $sgpr13, killed $sgpr14, killed $sgpr15, killed $vgpr31, $sgpr0_sgpr1_sgpr2_sgpr3, 2, 0, 2, 0, 2, 1, 2, 0, 2, 0, 2, 0, 2, 0, csr_amdgpu
    $sgpr16 = S_MOV_B32 $sgpr33
    $sgpr33 = S_MOV_B32 $sgpr32
    $sgpr18_sgpr19 = S_OR_SAVEEXEC_B64 -1, implicit-def $exec, implicit-def dead $scc, implicit $exec
    BUFFER_STORE_DWORD_OFFSET $vgpr40, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr33, 0, 0, 0, implicit $exec :: (store (s32) into %stack.1, addrspace 5)
    $exec = S_MOV_B64 killed $sgpr18_sgpr19
    $vgpr40 = V_WRITELANE_B32 killed $sgpr16, 2, undef $vgpr40
    $vgpr40 = V_WRITELANE_B32 $sgpr30, 0, $vgpr40
    $sgpr32 = frame-setup S_ADDK_I32 $sgpr32, 1024, implicit-def dead $scc
    $vgpr40 = V_WRITELANE_B32 $sgpr31, 1, $vgpr40
    $sgpr14_sgpr15 = S_LOAD_DWORDX2_IMM $sgpr12_sgpr13, 0, 0
    STATEPOINT 2882400015, 0, 11, renamable $sgpr14_sgpr15, 0, killed $sgpr4_sgpr5, killed $sgpr6_sgpr7, killed $sgpr8_sgpr9, killed $sgpr10_sgpr11, killed $sgpr12, killed $sgpr13, killed $sgpr14, killed $sgpr15, killed $vgpr31, $sgpr0_sgpr1_sgpr2_sgpr3, 2, 0, 2, 0, 2, 1, 2, 0, 2, 0, 2, 0, 2, 0, csr_amdgpu

...