llvm/test/CodeGen/AMDGPU/coalesce-identity-copies-undef-subregs.mir

# NOTE: Assertions have been autogenerated by utils/update_mir_test_checks.py
# RUN: llc -mtriple=amdgcn-amd-amdhsa -mcpu=gfx906 -verify-coalescing -verify-machineinstrs -amdgpu-enable-rewrite-partial-reg-uses=false -start-before=register-coalescer -stop-after=machine-scheduler -o - %s | FileCheck %s

# Tests that break due to the handling of partially undef registers
# when whole register identity copies are erased.

# Make sure there is no verifier error after
# RenameIndepependentSubregs processes this. The coalescer would
# remove the identity copy in %bb.1, and leave behind a dummy interval
# across bb.1 with no corresponding value anywhere in the function.

---
name:            identity_copy_undef_subrange
tracksRegLiveness: true
body:             |
  ; CHECK-LABEL: name: identity_copy_undef_subrange
  ; CHECK: bb.0:
  ; CHECK-NEXT:   successors: %bb.1(0x80000000)
  ; CHECK-NEXT:   liveins: $vgpr0
  ; CHECK-NEXT: {{  $}}
  ; CHECK-NEXT:   undef [[COPY:%[0-9]+]].sub1:vreg_64 = COPY $vgpr0
  ; CHECK-NEXT: {{  $}}
  ; CHECK-NEXT: bb.1:
  ; CHECK-NEXT:   successors: %bb.2(0x40000000), %bb.1(0x40000000)
  ; CHECK-NEXT: {{  $}}
  ; CHECK-NEXT:   S_CBRANCH_EXECNZ %bb.1, implicit $exec
  ; CHECK-NEXT:   S_BRANCH %bb.2
  ; CHECK-NEXT: {{  $}}
  ; CHECK-NEXT: bb.2:
  ; CHECK-NEXT:   successors: %bb.1(0x80000000)
  ; CHECK-NEXT: {{  $}}
  ; CHECK-NEXT:   undef [[COPY:%[0-9]+]].sub1:vreg_64 = nofpexcept V_CEIL_F32_e32 [[COPY]].sub1, implicit $mode, implicit $exec
  ; CHECK-NEXT:   S_BRANCH %bb.1
  bb.0:
    liveins: $vgpr0

    undef %0.sub1:vreg_64 = COPY killed $vgpr0

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

    %0:vreg_64 = COPY %0
    S_CBRANCH_EXECNZ %bb.1, implicit $exec
    S_BRANCH %bb.2

  bb.2:
    undef %0.sub1:vreg_64 = nofpexcept V_CEIL_F32_e32 killed %0.sub1, implicit $mode, implicit $exec
    S_BRANCH %bb.1

...

# Test another use of the register inside the block.
---
name:            identity_copy_undef_subrange_other_uses0
tracksRegLiveness: true
body:             |
  ; CHECK-LABEL: name: identity_copy_undef_subrange_other_uses0
  ; CHECK: bb.0:
  ; CHECK-NEXT:   successors: %bb.1(0x80000000)
  ; CHECK-NEXT:   liveins: $vgpr0
  ; CHECK-NEXT: {{  $}}
  ; CHECK-NEXT:   undef [[COPY:%[0-9]+]].sub1:vreg_64 = COPY $vgpr0
  ; CHECK-NEXT: {{  $}}
  ; CHECK-NEXT: bb.1:
  ; CHECK-NEXT:   successors: %bb.2(0x40000000), %bb.1(0x40000000)
  ; CHECK-NEXT: {{  $}}
  ; CHECK-NEXT:   S_NOP 0, implicit undef [[COPY]].sub0
  ; CHECK-NEXT:   S_NOP 0, implicit undef [[COPY]].sub0
  ; CHECK-NEXT:   S_CBRANCH_EXECNZ %bb.1, implicit $exec
  ; CHECK-NEXT:   S_BRANCH %bb.2
  ; CHECK-NEXT: {{  $}}
  ; CHECK-NEXT: bb.2:
  ; CHECK-NEXT:   successors: %bb.1(0x80000000)
  ; CHECK-NEXT: {{  $}}
  ; CHECK-NEXT:   undef [[COPY:%[0-9]+]].sub1:vreg_64 = nofpexcept V_CEIL_F32_e32 [[COPY]].sub1, implicit $mode, implicit $exec
  ; CHECK-NEXT:   S_BRANCH %bb.1
  bb.0:
    liveins: $vgpr0
    undef %0.sub1:vreg_64 = COPY killed $vgpr0

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

    %0:vreg_64 = COPY %0
    S_NOP 0, implicit %0.sub0
    S_NOP 0, implicit %0.sub0
    S_CBRANCH_EXECNZ %bb.1, implicit $exec
    S_BRANCH %bb.2

  bb.2:
    undef %0.sub1:vreg_64 = nofpexcept V_CEIL_F32_e32 killed %0.sub1, implicit $mode, implicit $exec
    S_BRANCH %bb.1

...

---
name: second_identity_copy
tracksRegLiveness: true
body:             |
  ; CHECK-LABEL: name: second_identity_copy
  ; CHECK: bb.0:
  ; CHECK-NEXT:   successors: %bb.1(0x80000000)
  ; CHECK-NEXT:   liveins: $vgpr0
  ; CHECK-NEXT: {{  $}}
  ; CHECK-NEXT:   undef [[COPY:%[0-9]+]].sub1:vreg_64 = COPY $vgpr0
  ; CHECK-NEXT: {{  $}}
  ; CHECK-NEXT: bb.1:
  ; CHECK-NEXT:   successors: %bb.2(0x40000000), %bb.1(0x40000000)
  ; CHECK-NEXT: {{  $}}
  ; CHECK-NEXT:   S_CBRANCH_EXECNZ %bb.1, implicit $exec
  ; CHECK-NEXT:   S_BRANCH %bb.2
  ; CHECK-NEXT: {{  $}}
  ; CHECK-NEXT: bb.2:
  ; CHECK-NEXT:   successors: %bb.1(0x80000000)
  ; CHECK-NEXT: {{  $}}
  ; CHECK-NEXT:   undef [[COPY:%[0-9]+]].sub1:vreg_64 = nofpexcept V_MUL_F32_e32 0, [[COPY]].sub1, implicit $mode, implicit $exec
  ; CHECK-NEXT:   S_BRANCH %bb.1
  bb.0:
    liveins: $vgpr0

    undef %0.sub1:vreg_64 = COPY killed $vgpr0

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

    %0:vreg_64 = COPY killed %0
    %0:vreg_64 = COPY %0
    S_CBRANCH_EXECNZ %bb.1, implicit $exec
    S_BRANCH %bb.2

  bb.2:
    undef %0.sub1:vreg_64 = nofpexcept V_MUL_F32_e32 0, killed %0.sub1, implicit $mode, implicit $exec
    S_BRANCH %bb.1

...

---
name: second_identity_copy_undef_lane_outblock
tracksRegLiveness: true
body:             |
  ; CHECK-LABEL: name: second_identity_copy_undef_lane_outblock
  ; CHECK: bb.0:
  ; CHECK-NEXT:   successors: %bb.1(0x80000000)
  ; CHECK-NEXT:   liveins: $vgpr0
  ; CHECK-NEXT: {{  $}}
  ; CHECK-NEXT:   undef [[COPY:%[0-9]+]].sub1:vreg_64 = COPY $vgpr0
  ; CHECK-NEXT: {{  $}}
  ; CHECK-NEXT: bb.1:
  ; CHECK-NEXT:   successors: %bb.2(0x40000000), %bb.1(0x40000000)
  ; CHECK-NEXT: {{  $}}
  ; CHECK-NEXT:   S_CBRANCH_EXECNZ %bb.1, implicit $exec
  ; CHECK-NEXT:   S_BRANCH %bb.2
  ; CHECK-NEXT: {{  $}}
  ; CHECK-NEXT: bb.2:
  ; CHECK-NEXT:   successors: %bb.1(0x80000000)
  ; CHECK-NEXT: {{  $}}
  ; CHECK-NEXT:   [[COPY:%[0-9]+]].sub1:vreg_64 = nofpexcept V_MUL_F32_e32 0, [[COPY]].sub1, implicit $mode, implicit $exec
  ; CHECK-NEXT:   S_BRANCH %bb.1
  bb.0:
    liveins: $vgpr0

    undef %0.sub1:vreg_64 = COPY killed $vgpr0

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

    %0:vreg_64 = COPY killed %0
    %0:vreg_64 = COPY %0
    S_CBRANCH_EXECNZ %bb.1, implicit $exec
    S_BRANCH %bb.2

  bb.2:
    %0.sub1:vreg_64 = nofpexcept V_MUL_F32_e32 0, killed %0.sub1, implicit $mode, implicit $exec
    S_BRANCH %bb.1

...

# The same value number appears in multiple blocks
---
name: identity_copy_undef_subrange_null_vninfo_to_remove
tracksRegLiveness: true
body:             |
  ; CHECK-LABEL: name: identity_copy_undef_subrange_null_vninfo_to_remove
  ; CHECK: bb.0:
  ; CHECK-NEXT:   successors: %bb.1(0x80000000)
  ; CHECK-NEXT:   liveins: $vgpr0
  ; CHECK-NEXT: {{  $}}
  ; CHECK-NEXT:   undef [[COPY:%[0-9]+]].sub1:vreg_64 = COPY $vgpr0
  ; CHECK-NEXT: {{  $}}
  ; CHECK-NEXT: bb.1:
  ; CHECK-NEXT:   successors: %bb.3(0x40000000), %bb.2(0x40000000)
  ; CHECK-NEXT: {{  $}}
  ; CHECK-NEXT:   S_CBRANCH_EXECNZ %bb.3, implicit $exec
  ; CHECK-NEXT: {{  $}}
  ; CHECK-NEXT: bb.2:
  ; CHECK-NEXT:   successors: %bb.1(0x80000000)
  ; CHECK-NEXT: {{  $}}
  ; CHECK-NEXT:   S_NOP 0, implicit undef [[COPY]].sub0
  ; CHECK-NEXT:   undef [[COPY:%[0-9]+]].sub1:vreg_64 = nofpexcept V_CEIL_F32_e32 [[COPY]].sub1, implicit $mode, implicit $exec
  ; CHECK-NEXT:   S_BRANCH %bb.1
  ; CHECK-NEXT: {{  $}}
  ; CHECK-NEXT: bb.3:
  ; CHECK-NEXT:   S_NOP 0, implicit undef [[COPY]].sub0
  bb.0:
    liveins: $vgpr0

    undef %0.sub1:vreg_64 = COPY killed $vgpr0

  bb.1:
    %0:vreg_64 = COPY %0
    S_CBRANCH_EXECNZ %bb.3, implicit $exec

  bb.2:
    S_NOP 0, implicit %0.sub0
    undef %0.sub1:vreg_64 = nofpexcept V_CEIL_F32_e32 killed %0.sub1, implicit $mode, implicit $exec
    S_BRANCH %bb.1

  bb.3:
    S_NOP 0, implicit %0.sub0

...

---
name: undef_copy_self_loop0
tracksRegLiveness: true
body:             |
  ; CHECK-LABEL: name: undef_copy_self_loop0
  ; CHECK: bb.0:
  ; CHECK-NEXT:   successors: %bb.1(0x80000000)
  ; CHECK-NEXT:   liveins: $vgpr0
  ; CHECK-NEXT: {{  $}}
  ; CHECK-NEXT:   undef [[COPY:%[0-9]+]].sub1:vreg_64 = COPY $vgpr0
  ; CHECK-NEXT: {{  $}}
  ; CHECK-NEXT: bb.1:
  ; CHECK-NEXT:   successors: %bb.1(0x40000000), %bb.2(0x40000000)
  ; CHECK-NEXT: {{  $}}
  ; CHECK-NEXT:   S_CBRANCH_EXECNZ %bb.1, implicit $exec
  ; CHECK-NEXT: {{  $}}
  ; CHECK-NEXT: bb.2:
  ; CHECK-NEXT:   S_NOP 0, implicit undef [[COPY]].sub0
  bb.0:
    liveins: $vgpr0

    undef %0.sub1:vreg_64 = COPY killed $vgpr0

  bb.1:
    %0:vreg_64 = COPY %0
    S_CBRANCH_EXECNZ %bb.1, implicit $exec

  bb.2:
    S_NOP 0, implicit %0.sub0

...

---
name: undef_copy_self_loop1
tracksRegLiveness: true
body:             |
  ; CHECK-LABEL: name: undef_copy_self_loop1
  ; CHECK: bb.0:
  ; CHECK-NEXT:   successors: %bb.1(0x80000000)
  ; CHECK-NEXT:   liveins: $vgpr0
  ; CHECK-NEXT: {{  $}}
  ; CHECK-NEXT:   undef [[COPY:%[0-9]+]].sub1:vreg_64 = COPY $vgpr0
  ; CHECK-NEXT: {{  $}}
  ; CHECK-NEXT: bb.1:
  ; CHECK-NEXT:   successors: %bb.1(0x40000000), %bb.2(0x40000000)
  ; CHECK-NEXT: {{  $}}
  ; CHECK-NEXT:   S_CBRANCH_EXECNZ %bb.1, implicit $exec
  ; CHECK-NEXT: {{  $}}
  ; CHECK-NEXT: bb.2:
  ; CHECK-NEXT:   S_NOP 0, implicit [[COPY]].sub1
  bb.0:
    liveins: $vgpr0

    undef %0.sub1:vreg_64 = COPY killed $vgpr0

  bb.1:
    %0:vreg_64 = COPY %0
    S_CBRANCH_EXECNZ %bb.1, implicit $exec

  bb.2:
    S_NOP 0, implicit %0.sub1

...

# The coalescing of the %0 = %2 COPY in %bb.2 needs to prune the dead
# phi range across %bb.1 after it is erased.
---
name: prune_subrange_phi_value_0
tracksRegLiveness: true
body:             |
  ; CHECK-LABEL: name: prune_subrange_phi_value_0
  ; CHECK: bb.0:
  ; CHECK-NEXT:   successors: %bb.1(0x80000000)
  ; CHECK-NEXT:   liveins: $vgpr0
  ; CHECK-NEXT: {{  $}}
  ; CHECK-NEXT:   dead undef [[COPY:%[0-9]+]].sub1:vreg_64 = COPY $vgpr0
  ; CHECK-NEXT: {{  $}}
  ; CHECK-NEXT: bb.1:
  ; CHECK-NEXT:   successors: %bb.2(0x40000000), %bb.1(0x40000000)
  ; CHECK-NEXT: {{  $}}
  ; CHECK-NEXT:   S_CBRANCH_EXECNZ %bb.1, implicit $exec
  ; CHECK-NEXT:   S_BRANCH %bb.2
  ; CHECK-NEXT: {{  $}}
  ; CHECK-NEXT: bb.2:
  ; CHECK-NEXT:   successors: %bb.1(0x80000000)
  ; CHECK-NEXT: {{  $}}
  ; CHECK-NEXT:   S_BRANCH %bb.1
  bb.0:
    liveins: $vgpr0

    undef %0.sub1:vreg_64 = COPY killed $vgpr0

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

    %1:vreg_64 = COPY killed %0
    %0:vreg_64 = COPY %1
    S_CBRANCH_EXECNZ %bb.1, implicit $exec
    S_BRANCH %bb.2

  bb.2:
    undef %2.sub1:vreg_64 = COPY %0.sub1
    %0:vreg_64 = COPY killed %2
    S_BRANCH %bb.1

...

---
name: prune_subrange_phi_value_0_0
tracksRegLiveness: true
body:             |
  ; CHECK-LABEL: name: prune_subrange_phi_value_0_0
  ; CHECK: bb.0:
  ; CHECK-NEXT:   successors: %bb.1(0x80000000)
  ; CHECK-NEXT:   liveins: $vgpr0
  ; CHECK-NEXT: {{  $}}
  ; CHECK-NEXT:   dead undef [[COPY:%[0-9]+]].sub1:vreg_64 = COPY $vgpr0
  ; CHECK-NEXT: {{  $}}
  ; CHECK-NEXT: bb.1:
  ; CHECK-NEXT:   successors: %bb.2(0x40000000), %bb.1(0x40000000)
  ; CHECK-NEXT: {{  $}}
  ; CHECK-NEXT:   S_CBRANCH_EXECNZ %bb.1, implicit $exec
  ; CHECK-NEXT:   S_BRANCH %bb.2
  ; CHECK-NEXT: {{  $}}
  ; CHECK-NEXT: bb.2:
  ; CHECK-NEXT:   successors: %bb.1(0x80000000)
  ; CHECK-NEXT: {{  $}}
  ; CHECK-NEXT:   S_BRANCH %bb.1
  bb.0:
    liveins: $vgpr0

    undef %0.sub1:vreg_64 = COPY killed $vgpr0

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

    %1:vreg_64 = COPY killed %0
    %0:vreg_64 = COPY %1
    S_CBRANCH_EXECNZ %bb.1, implicit $exec
    S_BRANCH %bb.2

  bb.2:
    undef %0.sub1:vreg_64 = COPY %0.sub1
    S_BRANCH %bb.1

...

---
name: prune_subrange_phi_value_0_1
tracksRegLiveness: true
body:             |
  ; CHECK-LABEL: name: prune_subrange_phi_value_0_1
  ; CHECK: bb.0:
  ; CHECK-NEXT:   successors: %bb.1(0x80000000)
  ; CHECK-NEXT:   liveins: $vgpr0
  ; CHECK-NEXT: {{  $}}
  ; CHECK-NEXT:   dead undef [[COPY:%[0-9]+]].sub1:vreg_64 = COPY $vgpr0
  ; CHECK-NEXT: {{  $}}
  ; CHECK-NEXT: bb.1:
  ; CHECK-NEXT:   successors: %bb.2(0x40000000), %bb.1(0x40000000)
  ; CHECK-NEXT: {{  $}}
  ; CHECK-NEXT:   S_CBRANCH_EXECNZ %bb.1, implicit $exec
  ; CHECK-NEXT:   S_BRANCH %bb.2
  ; CHECK-NEXT: {{  $}}
  ; CHECK-NEXT: bb.2:
  ; CHECK-NEXT:   successors: %bb.1(0x80000000)
  ; CHECK-NEXT: {{  $}}
  ; CHECK-NEXT:   S_BRANCH %bb.1
  bb.0:
    liveins: $vgpr0

    undef %0.sub1:vreg_64 = COPY killed $vgpr0

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

    %1:vreg_64 = COPY killed %0
    %0:vreg_64 = COPY %1
    S_CBRANCH_EXECNZ %bb.1, implicit $exec
    S_BRANCH %bb.2

  bb.2:
    S_BRANCH %bb.1

...

# Variant of testcase that asserts since there wasn't already an
# incoming segment at the erased copy, and no valid end point.
---
name:            prune_subrange_phi_value_1
tracksRegLiveness: true
body:             |
  ; CHECK-LABEL: name: prune_subrange_phi_value_1
  ; CHECK: bb.0:
  ; CHECK-NEXT:   successors: %bb.1(0x80000000)
  ; CHECK-NEXT:   liveins: $vgpr0
  ; CHECK-NEXT: {{  $}}
  ; CHECK-NEXT:   undef [[COPY:%[0-9]+]].sub1:vreg_64 = COPY $vgpr0
  ; CHECK-NEXT: {{  $}}
  ; CHECK-NEXT: bb.1:
  ; CHECK-NEXT:   successors: %bb.2(0x40000000), %bb.1(0x40000000)
  ; CHECK-NEXT: {{  $}}
  ; CHECK-NEXT:   S_CBRANCH_EXECNZ %bb.1, implicit $exec
  ; CHECK-NEXT:   S_BRANCH %bb.2
  ; CHECK-NEXT: {{  $}}
  ; CHECK-NEXT: bb.2:
  ; CHECK-NEXT:   successors: %bb.1(0x80000000)
  ; CHECK-NEXT: {{  $}}
  ; CHECK-NEXT:   undef [[COPY:%[0-9]+]].sub1:vreg_64 = nofpexcept V_CEIL_F32_e32 [[COPY]].sub1, implicit $mode, implicit $exec
  ; CHECK-NEXT:   S_BRANCH %bb.1
  bb.0:
    liveins: $vgpr0

    undef %0.sub1:vreg_64 = COPY killed $vgpr0

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

    %0:vreg_64 = COPY killed %0
    S_CBRANCH_EXECNZ %bb.1, implicit $exec
    S_BRANCH %bb.2

  bb.2:
    undef %1.sub1:vreg_64 = nofpexcept V_CEIL_F32_e32 killed %0.sub1, implicit $mode, implicit $exec
    %0:vreg_64 = COPY killed %1
    S_BRANCH %bb.1

...

---
name:            prune_subrange_phi_value_2
tracksRegLiveness: true
body:             |
  ; CHECK-LABEL: name: prune_subrange_phi_value_2
  ; CHECK: bb.0:
  ; CHECK-NEXT:   successors: %bb.1(0x80000000)
  ; CHECK-NEXT:   liveins: $vgpr0
  ; CHECK-NEXT: {{  $}}
  ; CHECK-NEXT:   undef [[COPY:%[0-9]+]].sub1:vreg_64 = COPY $vgpr0
  ; CHECK-NEXT: {{  $}}
  ; CHECK-NEXT: bb.1:
  ; CHECK-NEXT:   successors: %bb.2(0x40000000), %bb.1(0x40000000)
  ; CHECK-NEXT: {{  $}}
  ; CHECK-NEXT:   S_CBRANCH_EXECNZ %bb.1, implicit $exec
  ; CHECK-NEXT:   S_BRANCH %bb.2
  ; CHECK-NEXT: {{  $}}
  ; CHECK-NEXT: bb.2:
  ; CHECK-NEXT:   successors: %bb.1(0x80000000)
  ; CHECK-NEXT: {{  $}}
  ; CHECK-NEXT:   [[COPY:%[0-9]+]].sub1:vreg_64 = nofpexcept V_CEIL_F32_e32 [[COPY]].sub1, implicit $mode, implicit $exec
  ; CHECK-NEXT:   S_BRANCH %bb.1
  bb.0:
    liveins: $vgpr0

    undef %0.sub1:vreg_64 = COPY killed $vgpr0

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

    %0:vreg_64 = COPY killed %0
    S_CBRANCH_EXECNZ %bb.1, implicit $exec
    S_BRANCH %bb.2

  bb.2:
    %0.sub1:vreg_64 = nofpexcept V_CEIL_F32_e32 %0.sub1, implicit $mode, implicit $exec
    S_BRANCH %bb.1

...