llvm/test/CodeGen/AMDGPU/machine-cse-commute-target-flags.mir

# NOTE: Assertions have been autogenerated by utils/update_mir_test_checks.py
# RUN: llc -mtriple=amdgpu10.10-amd-amdhsa -run-pass=machine-cse -o - %s | FileCheck %s
# RUN: llc -mtriple=amdgpu11.00-amd-amdhsa -run-pass=machine-cse -o - %s | FileCheck %s
# RUN: llc -mtriple=amdgpu9.0a-amd-amdhsa -run-pass=machine-cse -o - %s | FileCheck %s

--- |
  define void @commute_instruction_subreg_target_flag() { ret void }
  define void @commute_target_flag_frame_index() { ret void }
  define void @commute_target_flag_global() { ret void }
  define void @commute_target_flag_global_offset() { ret void }
  define void @commute_target_flag_global_offset_mismatch() { ret void }
  define void @pk_mov_no_cse() { ret void }
  define void @pk_mov_cse_identical() { ret void }

  declare void @func()
  @gv = external addrspace(1) global i32

...

# Make sure the subreg index is cleared when commuting a register and immediate.

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

    ; CHECK-LABEL: name: commute_instruction_subreg_target_flag
    ; CHECK: liveins: $vgpr0_vgpr1
    ; CHECK-NEXT: {{  $}}
    ; CHECK-NEXT: [[COPY:%[0-9]+]]:vreg_64 = COPY $vgpr0_vgpr1
    ; CHECK-NEXT: [[V_ADD_U32_e64_:%[0-9]+]]:vgpr_32 = V_ADD_U32_e64 [[COPY]].sub1, 64, 0, implicit $exec
    ; CHECK-NEXT: S_ENDPGM 0, implicit [[V_ADD_U32_e64_]], implicit [[V_ADD_U32_e64_]]
    %0:vreg_64 = COPY $vgpr0_vgpr1
    %1:vgpr_32 = V_ADD_U32_e64 %0.sub1, 64, 0, implicit $exec
    %2:vgpr_32 = V_ADD_U32_e64 64, %0.sub1, 0, implicit $exec
    S_ENDPGM 0, implicit %1, implicit %2

...

# FIXME: Why doesn't this CSE?
---
name: commute_target_flag_frame_index
tracksRegLiveness: true
stack:
  - { id: 0, type: default, offset: 0, size: 4, alignment: 4 }
body:             |
  bb.0:
    liveins: $vgpr0_vgpr1

    ; CHECK-LABEL: name: commute_target_flag_frame_index
    ; CHECK: liveins: $vgpr0_vgpr1
    ; CHECK-NEXT: {{  $}}
    ; CHECK-NEXT: [[COPY:%[0-9]+]]:vreg_64 = COPY $vgpr0_vgpr1
    ; CHECK-NEXT: [[V_ADD_U32_e64_:%[0-9]+]]:vgpr_32 = V_ADD_U32_e64 %stack.0, [[COPY]].sub0, 0, implicit $exec
    ; CHECK-NEXT: S_ENDPGM 0, implicit [[V_ADD_U32_e64_]], implicit [[V_ADD_U32_e64_]]
    %0:vreg_64 = COPY $vgpr0_vgpr1
    %1:vgpr_32 = V_ADD_U32_e64 %0.sub0, %stack.0, 0, implicit $exec
    %2:vgpr_32 = V_ADD_U32_e64 %stack.0, %0.sub0, 0, implicit $exec
    S_ENDPGM 0, implicit %1, implicit %2

...

# FIXME: Handle commuting global variables
---
name: commute_target_flag_global
tracksRegLiveness: true
body:             |
  bb.0:
    liveins: $sgpr0_sgpr1

    ; CHECK-LABEL: name: commute_target_flag_global
    ; CHECK: liveins: $sgpr0_sgpr1
    ; CHECK-NEXT: {{  $}}
    ; CHECK-NEXT: [[COPY:%[0-9]+]]:sreg_64 = COPY $sgpr0_sgpr1
    ; CHECK-NEXT: [[S_ADD_U32_:%[0-9]+]]:sreg_32 = S_ADD_U32 [[COPY]].sub0, target-flags(amdgpu-rel32-lo) @func, implicit-def dead $scc
    ; CHECK-NEXT: S_ENDPGM 0, implicit [[S_ADD_U32_]], implicit [[S_ADD_U32_]]
    %0:sreg_64 = COPY $sgpr0_sgpr1
    %1:sreg_32 = S_ADD_U32 %0.sub0, target-flags(amdgpu-rel32-lo) @func, implicit-def dead $scc
    %2:sreg_32 = S_ADD_U32 target-flags(amdgpu-rel32-lo) @func, %0.sub0, implicit-def dead $scc
    S_ENDPGM 0, implicit %1, implicit %2

...

---
name: commute_target_flag_global_offset
tracksRegLiveness: true
body:             |
  bb.0:
    liveins: $sgpr0_sgpr1

    ; CHECK-LABEL: name: commute_target_flag_global_offset
    ; CHECK: liveins: $sgpr0_sgpr1
    ; CHECK-NEXT: {{  $}}
    ; CHECK-NEXT: [[COPY:%[0-9]+]]:sreg_64 = COPY $sgpr0_sgpr1
    ; CHECK-NEXT: [[S_ADD_U32_:%[0-9]+]]:sreg_32 = S_ADD_U32 [[COPY]].sub0, target-flags(amdgpu-rel32-lo) @gv + 4, implicit-def dead $scc
    ; CHECK-NEXT: S_ENDPGM 0, implicit [[S_ADD_U32_]], implicit [[S_ADD_U32_]]
    %0:sreg_64 = COPY $sgpr0_sgpr1
    %1:sreg_32 = S_ADD_U32 %0.sub0, target-flags(amdgpu-rel32-lo) @gv + 4, implicit-def dead $scc
    %2:sreg_32 = S_ADD_U32 target-flags(amdgpu-rel32-lo) @gv + 4, %0.sub0, implicit-def dead $scc
    S_ENDPGM 0, implicit %1, implicit %2

...

---
name: commute_target_flag_global_offset_mismatch
tracksRegLiveness: true
body:             |
  bb.0:
    liveins: $sgpr0_sgpr1

    ; CHECK-LABEL: name: commute_target_flag_global_offset_mismatch
    ; CHECK: liveins: $sgpr0_sgpr1
    ; CHECK-NEXT: {{  $}}
    ; CHECK-NEXT: [[COPY:%[0-9]+]]:sreg_64 = COPY $sgpr0_sgpr1
    ; CHECK-NEXT: [[S_ADD_U32_:%[0-9]+]]:sreg_32 = S_ADD_U32 [[COPY]].sub0, target-flags(amdgpu-rel32-lo) @gv + 4, implicit-def dead $scc
    ; CHECK-NEXT: [[S_ADD_U32_1:%[0-9]+]]:sreg_32 = S_ADD_U32 target-flags(amdgpu-rel32-lo) @gv + 8, [[COPY]].sub0, implicit-def dead $scc
    ; CHECK-NEXT: S_ENDPGM 0, implicit [[S_ADD_U32_]], implicit [[S_ADD_U32_1]]
    %0:sreg_64 = COPY $sgpr0_sgpr1
    %1:sreg_32 = S_ADD_U32 %0.sub0, target-flags(amdgpu-rel32-lo) @gv + 4, implicit-def dead $scc
    %2:sreg_32 = S_ADD_U32 target-flags(amdgpu-rel32-lo) @gv + 8, %0.sub0, implicit-def dead $scc
    S_ENDPGM 0, implicit %1, implicit %2

...

---
name:            pk_mov_no_cse
tracksRegLiveness: true
body: |
  bb.0:
    liveins: $vgpr0_vgpr1, $vgpr2_vgpr3, $vgpr4_vgpr5

    ; CHECK-LABEL: name: pk_mov_no_cse
    ; CHECK: liveins: $vgpr0_vgpr1, $vgpr2_vgpr3, $vgpr4_vgpr5
    ; CHECK-NEXT: {{  $}}
    ; CHECK-NEXT: [[COPY:%[0-9]+]]:vreg_64_align2 = COPY $vgpr0_vgpr1
    ; CHECK-NEXT: [[COPY1:%[0-9]+]]:vreg_64_align2 = COPY $vgpr2_vgpr3
    ; CHECK-NEXT: [[COPY2:%[0-9]+]]:vreg_64_align2 = COPY $vgpr4_vgpr5
    ; CHECK-NEXT: [[V_PK_MOV_B32_:%[0-9]+]]:vreg_64_align2 = V_PK_MOV_B32 12, [[COPY]], 8, [[COPY1]], 0, 0, 0, 0, 0, implicit $exec
    ; CHECK-NEXT: [[V_PK_MOV_B32_1:%[0-9]+]]:vreg_64_align2 = V_PK_MOV_B32 8, [[COPY1]], 12, [[COPY]], 0, 0, 0, 0, 0, implicit $exec
    ; CHECK-NEXT: GLOBAL_STORE_DWORDX2 [[COPY2]], [[V_PK_MOV_B32_]], 0, 0, implicit $exec :: (store (s64), addrspace 1)
    ; CHECK-NEXT: GLOBAL_STORE_DWORDX2 [[COPY2]], [[V_PK_MOV_B32_1]], 8, 0, implicit $exec :: (store (s64), addrspace 1)
    ; CHECK-NEXT: SI_RETURN
    %0:vreg_64_align2 = COPY $vgpr0_vgpr1
    %1:vreg_64_align2 = COPY $vgpr2_vgpr3
    %2:vreg_64_align2 = COPY $vgpr4_vgpr5
    %4:vreg_64_align2 = V_PK_MOV_B32 12, %0, 8, %1, 0, 0, 0, 0, 0, implicit $exec
    %5:vreg_64_align2 = V_PK_MOV_B32 8, %1, 12, %0, 0, 0, 0, 0, 0, implicit $exec
    GLOBAL_STORE_DWORDX2 %2, %4, 0, 0, implicit $exec :: (store (s64), addrspace 1)
    GLOBAL_STORE_DWORDX2 %2, %5, 8, 0, implicit $exec :: (store (s64), addrspace 1)
    SI_RETURN

...

---
name:            pk_mov_cse_identical
tracksRegLiveness: true
body: |
  bb.0:
    liveins: $vgpr0_vgpr1, $vgpr2_vgpr3, $vgpr4_vgpr5

    ; CHECK-LABEL: name: pk_mov_cse_identical
    ; CHECK: liveins: $vgpr0_vgpr1, $vgpr2_vgpr3, $vgpr4_vgpr5
    ; CHECK-NEXT: {{  $}}
    ; CHECK-NEXT: [[COPY:%[0-9]+]]:vreg_64_align2 = COPY $vgpr0_vgpr1
    ; CHECK-NEXT: [[COPY1:%[0-9]+]]:vreg_64_align2 = COPY $vgpr2_vgpr3
    ; CHECK-NEXT: [[COPY2:%[0-9]+]]:vreg_64_align2 = COPY $vgpr4_vgpr5
    ; CHECK-NEXT: [[V_PK_MOV_B32_:%[0-9]+]]:vreg_64_align2 = V_PK_MOV_B32 12, [[COPY]], 8, [[COPY1]], 0, 0, 0, 0, 0, implicit $exec
    ; CHECK-NEXT: GLOBAL_STORE_DWORDX2 [[COPY2]], [[V_PK_MOV_B32_]], 0, 0, implicit $exec :: (store (s64), addrspace 1)
    ; CHECK-NEXT: GLOBAL_STORE_DWORDX2 [[COPY2]], [[V_PK_MOV_B32_]], 8, 0, implicit $exec :: (store (s64), addrspace 1)
    ; CHECK-NEXT: SI_RETURN
    %0:vreg_64_align2 = COPY $vgpr0_vgpr1
    %1:vreg_64_align2 = COPY $vgpr2_vgpr3
    %2:vreg_64_align2 = COPY $vgpr4_vgpr5
    %4:vreg_64_align2 = V_PK_MOV_B32 12, %0, 8, %1, 0, 0, 0, 0, 0, implicit $exec
    %5:vreg_64_align2 = V_PK_MOV_B32 12, %0, 8, %1, 0, 0, 0, 0, 0, implicit $exec
    GLOBAL_STORE_DWORDX2 %2, %4, 0, 0, implicit $exec :: (store (s64), addrspace 1)
    GLOBAL_STORE_DWORDX2 %2, %5, 8, 0, implicit $exec :: (store (s64), addrspace 1)
    SI_RETURN

...