llvm/test/CodeGen/AMDGPU/schedule-barrier-fpmode.mir

# NOTE: Assertions have been autogenerated by utils/update_mir_test_checks.py
# RUN: llc -mtriple=amdgcn -mcpu=gfx1010 -run-pass=machine-scheduler -o - %s | FileCheck %s
# RUN: llc -mtriple=amdgcn -mcpu=gfx1100 -run-pass=machine-scheduler -o - %s | FileCheck %s
# Make sure FP mode is not a hard scheduling boundary

---
name: denorm_mode_not_barrier
tracksRegLiveness: true
body: |
  bb.0:
    liveins: $vgpr0_vgpr1

    ; CHECK-LABEL: name: denorm_mode_not_barrier
    ; CHECK: liveins: $vgpr0_vgpr1
    ; CHECK-NEXT: {{  $}}
    ; CHECK-NEXT: [[COPY:%[0-9]+]]:vreg_64 = COPY $vgpr0_vgpr1
    ; CHECK-NEXT: [[GLOBAL_LOAD_DWORD:%[0-9]+]]:vgpr_32 = GLOBAL_LOAD_DWORD [[COPY]], 0, 0, implicit $exec :: (load (s32))
    ; CHECK-NEXT: [[GLOBAL_LOAD_DWORD1:%[0-9]+]]:vgpr_32 = GLOBAL_LOAD_DWORD [[COPY]], 4, 0, implicit $exec :: (load (s32))
    ; CHECK-NEXT: [[V_ADD_U32_e32_:%[0-9]+]]:vgpr_32 = V_ADD_U32_e32 [[GLOBAL_LOAD_DWORD]], [[GLOBAL_LOAD_DWORD1]], implicit $exec
    ; CHECK-NEXT: S_DENORM_MODE 0, implicit-def $mode, implicit $mode
    ; CHECK-NEXT: S_ENDPGM 0, implicit [[V_ADD_U32_e32_]]
    %0:vreg_64 = COPY $vgpr0_vgpr1
    %1:vgpr_32 = GLOBAL_LOAD_DWORD %0, 0, 0, implicit $exec :: (load (s32))
    S_DENORM_MODE 0, implicit-def $mode, implicit $mode
    %2:vgpr_32 = GLOBAL_LOAD_DWORD %0, 4, 0, implicit $exec :: (load (s32))
    %3:vgpr_32 = V_ADD_U32_e32 %1, %2, implicit $exec
    S_ENDPGM 0, implicit %3
...

---
name: round_mode_not_barrier
tracksRegLiveness: true
body: |
  bb.0:
    liveins: $vgpr0_vgpr1

    ; CHECK-LABEL: name: round_mode_not_barrier
    ; CHECK: liveins: $vgpr0_vgpr1
    ; CHECK-NEXT: {{  $}}
    ; CHECK-NEXT: [[COPY:%[0-9]+]]:vreg_64 = COPY $vgpr0_vgpr1
    ; CHECK-NEXT: [[GLOBAL_LOAD_DWORD:%[0-9]+]]:vgpr_32 = GLOBAL_LOAD_DWORD [[COPY]], 0, 0, implicit $exec :: (load (s32))
    ; CHECK-NEXT: [[GLOBAL_LOAD_DWORD1:%[0-9]+]]:vgpr_32 = GLOBAL_LOAD_DWORD [[COPY]], 4, 0, implicit $exec :: (load (s32))
    ; CHECK-NEXT: [[V_ADD_U32_e32_:%[0-9]+]]:vgpr_32 = V_ADD_U32_e32 [[GLOBAL_LOAD_DWORD]], [[GLOBAL_LOAD_DWORD1]], implicit $exec
    ; CHECK-NEXT: S_ROUND_MODE 0, implicit-def $mode, implicit $mode
    ; CHECK-NEXT: S_ENDPGM 0, implicit [[V_ADD_U32_e32_]]
    %0:vreg_64 = COPY $vgpr0_vgpr1
    %1:vgpr_32 = GLOBAL_LOAD_DWORD %0, 0, 0, implicit $exec :: (load (s32))
    S_ROUND_MODE 0, implicit-def $mode, implicit $mode
    %2:vgpr_32 = GLOBAL_LOAD_DWORD %0, 4, 0, implicit $exec :: (load (s32))
    %3:vgpr_32 = V_ADD_U32_e32 %1, %2, implicit $exec
    S_ENDPGM 0, implicit %3
...

---
name: denorm_mode_mode_def_use
tracksRegLiveness: true
body: |
  bb.0:
    liveins: $vgpr0_vgpr1

    ; CHECK-LABEL: name: denorm_mode_mode_def_use
    ; CHECK: liveins: $vgpr0_vgpr1
    ; CHECK-NEXT: {{  $}}
    ; CHECK-NEXT: [[COPY:%[0-9]+]]:vreg_64 = COPY $vgpr0_vgpr1
    ; CHECK-NEXT: [[GLOBAL_LOAD_DWORD:%[0-9]+]]:vgpr_32 = GLOBAL_LOAD_DWORD [[COPY]], 0, 0, implicit $exec :: (load (s32))
    ; CHECK-NEXT: dead %3:vgpr_32 = GLOBAL_LOAD_DWORD [[COPY]], 4, 0, implicit $exec :: (load (s32))
    ; CHECK-NEXT: S_DENORM_MODE 0, implicit-def $mode, implicit $mode
    ; CHECK-NEXT: [[V_ADD_F32_e32_:%[0-9]+]]:vgpr_32 = V_ADD_F32_e32 0, [[GLOBAL_LOAD_DWORD]], implicit $mode, implicit $exec
    ; CHECK-NEXT: [[V_ADD_U32_e32_:%[0-9]+]]:vgpr_32 = V_ADD_U32_e32 [[GLOBAL_LOAD_DWORD]], [[V_ADD_F32_e32_]], implicit $exec
    ; CHECK-NEXT: S_ENDPGM 0, implicit [[V_ADD_F32_e32_]], implicit [[V_ADD_U32_e32_]]
    %0:vreg_64 = COPY $vgpr0_vgpr1
    %1:vgpr_32 = GLOBAL_LOAD_DWORD %0, 0, 0, implicit $exec :: (load (s32))
    S_DENORM_MODE 0, implicit-def $mode, implicit $mode
    %2:vgpr_32 = V_ADD_F32_e32 0, %1, implicit $mode, implicit $exec
    %3:vgpr_32 = GLOBAL_LOAD_DWORD %0, 4, 0, implicit $exec :: (load (s32))
    %4:vgpr_32 = V_ADD_U32_e32 %1, %2, implicit $exec
    S_ENDPGM 0, implicit %2, implicit %4
...

---
name: round_mode_mode_def_use
tracksRegLiveness: true
body: |
  bb.0:
    liveins: $vgpr0_vgpr1

    ; CHECK-LABEL: name: round_mode_mode_def_use
    ; CHECK: liveins: $vgpr0_vgpr1
    ; CHECK-NEXT: {{  $}}
    ; CHECK-NEXT: [[COPY:%[0-9]+]]:vreg_64 = COPY $vgpr0_vgpr1
    ; CHECK-NEXT: [[GLOBAL_LOAD_DWORD:%[0-9]+]]:vgpr_32 = GLOBAL_LOAD_DWORD [[COPY]], 0, 0, implicit $exec :: (load (s32))
    ; CHECK-NEXT: dead %3:vgpr_32 = GLOBAL_LOAD_DWORD [[COPY]], 4, 0, implicit $exec :: (load (s32))
    ; CHECK-NEXT: S_ROUND_MODE 0, implicit-def $mode, implicit $mode
    ; CHECK-NEXT: [[V_ADD_F32_e32_:%[0-9]+]]:vgpr_32 = V_ADD_F32_e32 0, [[GLOBAL_LOAD_DWORD]], implicit $mode, implicit $exec
    ; CHECK-NEXT: [[V_ADD_U32_e32_:%[0-9]+]]:vgpr_32 = V_ADD_U32_e32 [[GLOBAL_LOAD_DWORD]], [[V_ADD_F32_e32_]], implicit $exec
    ; CHECK-NEXT: S_ENDPGM 0, implicit [[V_ADD_F32_e32_]], implicit [[V_ADD_U32_e32_]]
    %0:vreg_64 = COPY $vgpr0_vgpr1
    %1:vgpr_32 = GLOBAL_LOAD_DWORD %0, 0, 0, implicit $exec :: (load (s32))
    S_ROUND_MODE 0, implicit-def $mode, implicit $mode
    %2:vgpr_32 = V_ADD_F32_e32 0, %1, implicit $mode, implicit $exec
    %3:vgpr_32 = GLOBAL_LOAD_DWORD %0, 4, 0, implicit $exec :: (load (s32))
    %4:vgpr_32 = V_ADD_U32_e32 %1, %2, implicit $exec
    S_ENDPGM 0, implicit %2, implicit %4
...