third_party/llvm-7.0/llvm/test/CodeGen/AMDGPU/undefined-physreg-sgpr-spill.mir

# RUN: llc -mtriple=amdgcn--amdhsa -mcpu=gfx900 -run-pass si-optimize-exec-masking -verify-machineinstrs -o -  %s | FileCheck %s
--- |
  define amdgpu_kernel void @undefined_physreg_sgpr_spill() #0 {
    unreachable
  }

  define amdgpu_kernel void @undefined_physreg_sgpr_spill_reorder() #0 {
    unreachable
  }

  attributes #0 = { nounwind "amdgpu-num-sgpr"="16" }

...
---

# copy + s_and_b64 was turned into saveexec, deleting the copy,
# leaving a spill of the undefined register.

# CHECK-LABEL: name: undefined_physreg_sgpr_spill
# CHECK: $sgpr0_sgpr1 = COPY $exec, implicit-def $exec
# CHECK-NEXT: SI_SPILL_S64_SAVE $sgpr0_sgpr1,
# CHECK-NEXT: $sgpr2_sgpr3 = S_AND_B64 killed $sgpr0_sgpr1, killed $vcc, implicit-def dead $scc
# CHECK: $exec = COPY killed $sgpr2_sgpr3
name:            undefined_physreg_sgpr_spill
alignment:       0
exposesReturnsTwice: false
legalized:       false
regBankSelected: false
selected:        false
tracksRegLiveness: true
registers:
liveins:
  - { reg: '$vgpr0', virtual-reg: '' }
  - { reg: '$sgpr4_sgpr5', virtual-reg: '' }
stack:
  - { id: 0, name: '', type: spill-slot, offset: 0, size: 8, alignment: 4,
      stack-id: 1, callee-saved-register: '', callee-saved-restored: true,
      debug-info-variable: '', debug-info-expression: '',
      debug-info-location: '' }
constants:
body:             |
  bb.0:
    successors: %bb.1, %bb.2
    liveins: $vgpr0, $sgpr4_sgpr5

    $vgpr1_vgpr2 = COPY killed $sgpr4_sgpr5, implicit $exec
    $vgpr1 = GLOBAL_LOAD_UBYTE killed $vgpr1_vgpr2, 0, 0, 0, implicit $exec :: (non-temporal dereferenceable invariant load 1 from `i1 addrspace(2)* undef`)
    $vcc = V_CMP_NE_U32_e64 0, $vgpr0, implicit $exec
    $sgpr0_sgpr1 = V_CMP_EQ_U32_e64 1, killed $vgpr1, implicit $exec
    $vgpr1 = V_CNDMASK_B32_e64 0, -1, killed $sgpr0_sgpr1, implicit $exec
    $sgpr0_sgpr1 = COPY $exec, implicit-def $exec
    SI_SPILL_S64_SAVE $sgpr0_sgpr1, %stack.0, implicit $exec, implicit $sgpr8_sgpr9_sgpr10_sgpr11, implicit $sgpr13, implicit-def dead $m0 :: (store 8 into %stack.0, align 4, addrspace 5)
    $sgpr2_sgpr3 = S_AND_B64 killed $sgpr0_sgpr1, killed $vcc, implicit-def dead $scc
    $exec = S_MOV_B64_term killed $sgpr2_sgpr3
    SI_MASK_BRANCH %bb.2, implicit $exec
    S_BRANCH %bb.1

  bb.1:
    successors: %bb.3(0x80000000)
    liveins: $vgpr0, $vgpr1

    $sgpr2_sgpr3 = S_MOV_B64 0
    $vgpr2 = V_MOV_B32_e32 0, implicit $exec
    $sgpr4_sgpr5 = IMPLICIT_DEF
    S_BRANCH %bb.3

  bb.2:
    successors:

    $sgpr0_sgpr1 = SI_SPILL_S64_RESTORE %stack.0, implicit $exec, implicit $sgpr8_sgpr9_sgpr10_sgpr11, implicit $sgpr13, implicit-def dead $m0 :: (load 8 from %stack.0, align 4, addrspace 5)
    $exec = S_OR_B64 $exec, killed $sgpr0_sgpr1, implicit-def $scc

  bb.3:
    liveins: $vgpr0, $vgpr1, $vgpr2, $sgpr2_sgpr3, $sgpr4_sgpr5

    $vcc = COPY $vgpr1
    S_ENDPGM

...
---

# Move spill to after future save instruction
# CHECK-LABEL: {{^}}name: undefined_physreg_sgpr_spill_reorder
# CHECK: $sgpr0_sgpr1 = COPY $exec, implicit-def $exec
# CHECK: $sgpr2_sgpr3 = S_AND_B64 $sgpr0_sgpr1, killed $vcc, implicit-def dead $scc
# CHECK: SI_SPILL_S64_SAVE killed $sgpr0_sgpr1, %stack.0, implicit $exec, implicit $sgpr8_sgpr9_sgpr10_sgpr11, implicit $sgpr13, implicit-def dead $m0 :: (store 8 into %stack.0, align 4, addrspace 5)
# CHECK: $exec = COPY killed $sgpr2_sgpr3
name:            undefined_physreg_sgpr_spill_reorder
alignment:       0
exposesReturnsTwice: false
legalized:       false
regBankSelected: false
selected:        false
tracksRegLiveness: true
registers:
liveins:
  - { reg: '$vgpr0', virtual-reg: '' }
  - { reg: '$sgpr4_sgpr5', virtual-reg: '' }
stack:
  - { id: 0, name: '', type: spill-slot, offset: 0, size: 8, alignment: 4,
      stack-id: 1, callee-saved-register: '', callee-saved-restored: true,
      debug-info-variable: '', debug-info-expression: '',
      debug-info-location: '' }
constants:
body:             |
  bb.0:
    successors: %bb.1, %bb.2
    liveins: $vgpr0, $sgpr4_sgpr5

    $vgpr1_vgpr2 = COPY killed $sgpr4_sgpr5, implicit $exec
    $vgpr1 = GLOBAL_LOAD_UBYTE killed $vgpr1_vgpr2, 0, 0, 0, implicit $exec :: (non-temporal dereferenceable invariant load 1 from `i1 addrspace(2)* undef`)
    $vcc = V_CMP_NE_U32_e64 0, $vgpr0, implicit $exec
    $sgpr0_sgpr1 = V_CMP_EQ_U32_e64 1, killed $vgpr1, implicit $exec
    $vgpr1 = V_CNDMASK_B32_e64 0, -1, killed $sgpr0_sgpr1, implicit $exec
    $sgpr0_sgpr1 = COPY $exec, implicit-def $exec
    $sgpr2_sgpr3 = S_AND_B64 $sgpr0_sgpr1, killed $vcc, implicit-def dead $scc
    SI_SPILL_S64_SAVE killed $sgpr0_sgpr1, %stack.0, implicit $exec, implicit $sgpr8_sgpr9_sgpr10_sgpr11, implicit $sgpr13, implicit-def dead $m0 :: (store 8 into %stack.0, align 4, addrspace 5)
    $exec = S_MOV_B64_term killed $sgpr2_sgpr3
    SI_MASK_BRANCH %bb.2, implicit $exec
    S_BRANCH %bb.1

  bb.1:
    successors: %bb.3(0x80000000)
    liveins: $vgpr0, $vgpr1

    $sgpr2_sgpr3 = S_MOV_B64 0
    $vgpr2 = V_MOV_B32_e32 0, implicit $exec
    $sgpr4_sgpr5 = IMPLICIT_DEF
    S_BRANCH %bb.3

  bb.2:
    successors:

    $sgpr0_sgpr1 = SI_SPILL_S64_RESTORE %stack.0, implicit $exec, implicit $sgpr8_sgpr9_sgpr10_sgpr11, implicit $sgpr13, implicit-def dead $m0 :: (load 8 from %stack.0, align 4, addrspace 5)
    $exec = S_OR_B64 $exec, killed $sgpr0_sgpr1, implicit-def $scc

  bb.3:
    liveins: $vgpr0, $vgpr1, $vgpr2, $sgpr2_sgpr3, $sgpr4_sgpr5

    $vcc = COPY $vgpr1
    S_ENDPGM

...