llvm/test/CodeGen/AMDGPU/optimize-exec-mask-pre-ra-loop-phi.mir

# NOTE: Assertions have been autogenerated by utils/update_mir_test_checks.py
# RUN: llc -mtriple=amdgcn-amd-amdhsa -mcpu=gfx906 -verify-machineinstrs -run-pass=si-optimize-exec-masking-pre-ra -o - %s | FileCheck %s


# Cannot fold this without moving the def of %7 after the and.
---
name:            no_fold_andn2_select_condition_live_out_phi
tracksRegLiveness: true
body:             |
  ; CHECK-LABEL: name: no_fold_andn2_select_condition_live_out_phi
  ; CHECK: bb.0:
  ; CHECK-NEXT:   successors: %bb.2(0x80000000)
  ; CHECK-NEXT: {{  $}}
  ; CHECK-NEXT:   [[S_MOV_B64_:%[0-9]+]]:sreg_64_xexec = S_MOV_B64 -1
  ; CHECK-NEXT:   undef [[V_MOV_B32_e32_:%[0-9]+]].sub0:vreg_64 = V_MOV_B32_e32 0, implicit $exec
  ; CHECK-NEXT:   S_BRANCH %bb.2
  ; CHECK-NEXT: {{  $}}
  ; CHECK-NEXT: bb.1:
  ; CHECK-NEXT:   S_ENDPGM 0
  ; CHECK-NEXT: {{  $}}
  ; CHECK-NEXT: bb.2:
  ; CHECK-NEXT:   successors: %bb.1(0x40000000), %bb.2(0x40000000)
  ; CHECK-NEXT: {{  $}}
  ; CHECK-NEXT:   [[V_CNDMASK_B32_e64_:%[0-9]+]]:vgpr_32 = V_CNDMASK_B32_e64 0, 0, 0, 1, [[S_MOV_B64_]], implicit $exec
  ; CHECK-NEXT:   V_CMP_NE_U32_e32 1, [[V_CNDMASK_B32_e64_]], implicit-def $vcc, implicit $exec
  ; CHECK-NEXT:   [[V_MOV_B32_e32_:%[0-9]+]].sub1:vreg_64 = COPY [[V_MOV_B32_e32_]].sub0
  ; CHECK-NEXT:   DS_WRITE_B64_gfx9 undef %3:vgpr_32, [[V_MOV_B32_e32_]], 0, 0, implicit $exec :: (store (s64), addrspace 3)
  ; CHECK-NEXT:   ATOMIC_FENCE 4, 2
  ; CHECK-NEXT:   [[S_MOV_B64_:%[0-9]+]]:sreg_64_xexec = S_MOV_B64 0
  ; CHECK-NEXT:   $vcc = S_AND_B64 $exec, $vcc, implicit-def dead $scc
  ; CHECK-NEXT:   S_CBRANCH_VCCNZ %bb.1, implicit $vcc
  ; CHECK-NEXT:   S_BRANCH %bb.2
  bb.0:
    successors: %bb.2

    %7:sreg_64_xexec = S_MOV_B64 -1
    undef %5.sub0:vreg_64 = V_MOV_B32_e32 0, implicit $exec
    S_BRANCH %bb.2

  bb.1:
    S_ENDPGM 0

  bb.2:
    successors: %bb.1, %bb.2

    %4:vgpr_32 = V_CNDMASK_B32_e64 0, 0, 0, 1, %7, implicit $exec
    V_CMP_NE_U32_e32 1, %4, implicit-def $vcc, implicit $exec
    %5.sub1:vreg_64 = COPY %5.sub0
    DS_WRITE_B64_gfx9 undef %6:vgpr_32, %5, 0, 0, implicit $exec :: (store (s64), addrspace 3)
    ATOMIC_FENCE 4, 2
    %7:sreg_64_xexec = S_MOV_B64 0
    $vcc = S_AND_B64 $exec, killed $vcc, implicit-def dead $scc
    S_CBRANCH_VCCNZ %bb.1, implicit killed $vcc
    S_BRANCH %bb.2

...

# It's OK to fold this, since the phi def is after the andn2 insert point.
---
name:            fold_andn2_select_condition_live_out_phi_reorder
tracksRegLiveness: true
body:             |
  ; CHECK-LABEL: name: fold_andn2_select_condition_live_out_phi_reorder
  ; CHECK: bb.0:
  ; CHECK-NEXT:   successors: %bb.2(0x80000000)
  ; CHECK-NEXT: {{  $}}
  ; CHECK-NEXT:   [[S_MOV_B64_:%[0-9]+]]:sreg_64_xexec = S_MOV_B64 -1
  ; CHECK-NEXT:   undef [[V_MOV_B32_e32_:%[0-9]+]].sub0:vreg_64 = V_MOV_B32_e32 0, implicit $exec
  ; CHECK-NEXT:   S_BRANCH %bb.2
  ; CHECK-NEXT: {{  $}}
  ; CHECK-NEXT: bb.1:
  ; CHECK-NEXT:   S_ENDPGM 0
  ; CHECK-NEXT: {{  $}}
  ; CHECK-NEXT: bb.2:
  ; CHECK-NEXT:   successors: %bb.1(0x40000000), %bb.2(0x40000000)
  ; CHECK-NEXT: {{  $}}
  ; CHECK-NEXT:   [[V_MOV_B32_e32_:%[0-9]+]].sub1:vreg_64 = COPY [[V_MOV_B32_e32_]].sub0
  ; CHECK-NEXT:   DS_WRITE_B64_gfx9 undef %3:vgpr_32, [[V_MOV_B32_e32_]], 0, 0, implicit $exec :: (store (s64), addrspace 3)
  ; CHECK-NEXT:   ATOMIC_FENCE 4, 2
  ; CHECK-NEXT:   $vcc = S_ANDN2_B64 $exec, [[S_MOV_B64_]], implicit-def dead $scc
  ; CHECK-NEXT:   [[S_MOV_B64_:%[0-9]+]]:sreg_64_xexec = S_MOV_B64 0
  ; CHECK-NEXT:   S_CBRANCH_VCCNZ %bb.1, implicit $vcc
  ; CHECK-NEXT:   S_BRANCH %bb.2
  bb.0:
    successors: %bb.2

    %7:sreg_64_xexec = S_MOV_B64 -1
    undef %5.sub0:vreg_64 = V_MOV_B32_e32 0, implicit $exec
    S_BRANCH %bb.2

  bb.1:
    S_ENDPGM 0

  bb.2:
    successors: %bb.1, %bb.2

    %4:vgpr_32 = V_CNDMASK_B32_e64 0, 0, 0, 1, %7, implicit $exec
    V_CMP_NE_U32_e32 1, %4, implicit-def $vcc, implicit $exec
    %5.sub1:vreg_64 = COPY %5.sub0
    DS_WRITE_B64_gfx9 undef %6:vgpr_32, %5, 0, 0, implicit $exec :: (store (s64), addrspace 3)
    ATOMIC_FENCE 4, 2
    $vcc = S_AND_B64 $exec, killed $vcc, implicit-def dead $scc
    %7:sreg_64_xexec = S_MOV_B64 0
    S_CBRANCH_VCCNZ %bb.1, implicit killed $vcc
    S_BRANCH %bb.2

...

---
name:            no_fold_andn2_select_condition_live_out_phi_physreg
tracksRegLiveness: true
body:             |
  ; CHECK-LABEL: name: no_fold_andn2_select_condition_live_out_phi_physreg
  ; CHECK: bb.0:
  ; CHECK-NEXT:   successors: %bb.2(0x80000000)
  ; CHECK-NEXT: {{  $}}
  ; CHECK-NEXT:   $sgpr4_sgpr5 = S_MOV_B64 -1
  ; CHECK-NEXT:   undef [[V_MOV_B32_e32_:%[0-9]+]].sub0:vreg_64 = V_MOV_B32_e32 0, implicit $exec
  ; CHECK-NEXT:   S_BRANCH %bb.2
  ; CHECK-NEXT: {{  $}}
  ; CHECK-NEXT: bb.1:
  ; CHECK-NEXT:   S_ENDPGM 0
  ; CHECK-NEXT: {{  $}}
  ; CHECK-NEXT: bb.2:
  ; CHECK-NEXT:   successors: %bb.1(0x40000000), %bb.2(0x40000000)
  ; CHECK-NEXT:   liveins: $sgpr4_sgpr5
  ; CHECK-NEXT: {{  $}}
  ; CHECK-NEXT:   [[V_CNDMASK_B32_e64_:%[0-9]+]]:vgpr_32 = V_CNDMASK_B32_e64 0, 0, 0, 1, $sgpr4_sgpr5, implicit $exec
  ; CHECK-NEXT:   V_CMP_NE_U32_e32 1, [[V_CNDMASK_B32_e64_]], implicit-def $vcc, implicit $exec
  ; CHECK-NEXT:   [[V_MOV_B32_e32_:%[0-9]+]].sub1:vreg_64 = COPY [[V_MOV_B32_e32_]].sub0
  ; CHECK-NEXT:   DS_WRITE_B64_gfx9 undef %2:vgpr_32, [[V_MOV_B32_e32_]], 0, 0, implicit $exec :: (store (s64), addrspace 3)
  ; CHECK-NEXT:   ATOMIC_FENCE 4, 2
  ; CHECK-NEXT:   $sgpr4_sgpr5 = S_MOV_B64 0
  ; CHECK-NEXT:   $vcc = S_AND_B64 $exec, $vcc, implicit-def dead $scc
  ; CHECK-NEXT:   S_CBRANCH_VCCNZ %bb.1, implicit $vcc
  ; CHECK-NEXT:   S_BRANCH %bb.2
  bb.0:
    successors: %bb.2

    $sgpr4_sgpr5 = S_MOV_B64 -1
    undef %5.sub0:vreg_64 = V_MOV_B32_e32 0, implicit $exec
    S_BRANCH %bb.2

  bb.1:
    S_ENDPGM 0

  bb.2:
    successors: %bb.1, %bb.2
    liveins: $sgpr4_sgpr5

    %4:vgpr_32 = V_CNDMASK_B32_e64 0, 0, 0, 1, $sgpr4_sgpr5, implicit $exec
    V_CMP_NE_U32_e32 1, %4, implicit-def $vcc, implicit $exec
    %5.sub1:vreg_64 = COPY %5.sub0
    DS_WRITE_B64_gfx9 undef %6:vgpr_32, %5, 0, 0, implicit $exec :: (store (s64), addrspace 3)
    ATOMIC_FENCE 4, 2
    $sgpr4_sgpr5 = S_MOV_B64 0
    $vcc = S_AND_B64 $exec, killed $vcc, implicit-def dead $scc
    S_CBRANCH_VCCNZ %bb.1, implicit killed $vcc
    S_BRANCH %bb.2

...

---
name:            fold_andn2_select_condition_live_out_phi_physreg_reorder
tracksRegLiveness: true
body:             |
  ; CHECK-LABEL: name: fold_andn2_select_condition_live_out_phi_physreg_reorder
  ; CHECK: bb.0:
  ; CHECK-NEXT:   successors: %bb.2(0x80000000)
  ; CHECK-NEXT: {{  $}}
  ; CHECK-NEXT:   $sgpr4_sgpr5 = S_MOV_B64 -1
  ; CHECK-NEXT:   undef [[V_MOV_B32_e32_:%[0-9]+]].sub0:vreg_64 = V_MOV_B32_e32 0, implicit $exec
  ; CHECK-NEXT:   S_BRANCH %bb.2
  ; CHECK-NEXT: {{  $}}
  ; CHECK-NEXT: bb.1:
  ; CHECK-NEXT:   S_ENDPGM 0
  ; CHECK-NEXT: {{  $}}
  ; CHECK-NEXT: bb.2:
  ; CHECK-NEXT:   successors: %bb.1(0x40000000), %bb.2(0x40000000)
  ; CHECK-NEXT:   liveins: $sgpr4_sgpr5
  ; CHECK-NEXT: {{  $}}
  ; CHECK-NEXT:   [[V_MOV_B32_e32_:%[0-9]+]].sub1:vreg_64 = COPY [[V_MOV_B32_e32_]].sub0
  ; CHECK-NEXT:   DS_WRITE_B64_gfx9 undef %2:vgpr_32, [[V_MOV_B32_e32_]], 0, 0, implicit $exec :: (store (s64), addrspace 3)
  ; CHECK-NEXT:   ATOMIC_FENCE 4, 2
  ; CHECK-NEXT:   $vcc = S_ANDN2_B64 $exec, $sgpr4_sgpr5, implicit-def dead $scc
  ; CHECK-NEXT:   $sgpr4_sgpr5 = S_MOV_B64 0
  ; CHECK-NEXT:   S_CBRANCH_VCCNZ %bb.1, implicit $vcc
  ; CHECK-NEXT:   S_BRANCH %bb.2
  bb.0:
    successors: %bb.2

    $sgpr4_sgpr5 = S_MOV_B64 -1
    undef %5.sub0:vreg_64 = V_MOV_B32_e32 0, implicit $exec
    S_BRANCH %bb.2

  bb.1:
    S_ENDPGM 0

  bb.2:
    successors: %bb.1, %bb.2
    liveins: $sgpr4_sgpr5

    %4:vgpr_32 = V_CNDMASK_B32_e64 0, 0, 0, 1, $sgpr4_sgpr5, implicit $exec
    V_CMP_NE_U32_e32 1, %4, implicit-def $vcc, implicit $exec
    %5.sub1:vreg_64 = COPY %5.sub0
    DS_WRITE_B64_gfx9 undef %6:vgpr_32, %5, 0, 0, implicit $exec :: (store (s64), addrspace 3)
    ATOMIC_FENCE 4, 2
    $vcc = S_AND_B64 $exec, killed $vcc, implicit-def dead $scc
    $sgpr4_sgpr5 = S_MOV_B64 0
    S_CBRANCH_VCCNZ %bb.1, implicit killed $vcc
    S_BRANCH %bb.2

...