llvm-project/llvm/test/CodeGen/AMDGPU/clamp-omod-special-case.mir

# RUN: llc -march=amdgcn -verify-machineinstrs -run-pass si-fold-operands  %s -o - | FileCheck -check-prefix=GCN %s
---
# GCN-LABEL: name: v_max_self_clamp_not_set_f32
# GCN: %20:vgpr_32 = V_ADD_F32_e64 0, killed %17, 0, 1065353216, 0, 0, implicit $exec
# GCN-NEXT: %21:vgpr_32 = V_MAX_F32_e64 0, killed %20, 0, killed %20, 0, 0, implicit $exec

name:            v_max_self_clamp_not_set_f32
tracksRegLiveness: true
registers:
  - { id: 0, class: sgpr_64 }
  - { id: 1, class: sreg_32_xm0 }
  - { id: 2, class: sgpr_32 }
  - { id: 3, class: vgpr_32 }
  - { id: 4, class: sreg_64_xexec }
  - { id: 5, class: sreg_64_xexec }
  - { id: 6, class: sreg_32 }
  - { id: 7, class: sreg_32 }
  - { id: 8, class: sreg_32_xm0 }
  - { id: 9, class: sreg_64 }
  - { id: 10, class: sreg_32_xm0 }
  - { id: 11, class: sreg_32_xm0 }
  - { id: 12, class: sgpr_64 }
  - { id: 13, class: sgpr_128 }
  - { id: 14, class: sreg_32_xm0 }
  - { id: 15, class: sreg_64 }
  - { id: 16, class: sgpr_128 }
  - { id: 17, class: vgpr_32 }
  - { id: 18, class: vreg_64 }
  - { id: 19, class: vgpr_32 }
  - { id: 20, class: vgpr_32 }
  - { id: 21, class: vgpr_32 }
  - { id: 22, class: vgpr_32 }
  - { id: 23, class: vreg_64 }
  - { id: 24, class: vgpr_32 }
  - { id: 25, class: vreg_64 }
  - { id: 26, class: vreg_64 }
liveins:
  - { reg: '$sgpr0_sgpr1', virtual-reg: '%0' }
  - { reg: '$vgpr0', virtual-reg: '%3' }
body:             |
  bb.0:
    liveins: $sgpr0_sgpr1, $vgpr0

    %3 = COPY $vgpr0
    %0 = COPY $sgpr0_sgpr1
    %4 = S_LOAD_DWORDX2_IMM %0, 9, 0, 0:: (non-temporal dereferenceable invariant load 8 from `i64 addrspace(4)* undef`)
    %5 = S_LOAD_DWORDX2_IMM %0, 11, 0, 0 :: (non-temporal dereferenceable invariant load 8 from `i64 addrspace(4)* undef`)
    %24 = V_ASHRREV_I32_e32 31, %3, implicit $exec
    %25 = REG_SEQUENCE %3, 1, %24, 2
    %10 = S_MOV_B32 61440
    %11 = S_MOV_B32 0
    %12 = REG_SEQUENCE killed %11, 1, killed %10, 2
    %13 = REG_SEQUENCE killed %5, 17, %12, 18
    %14 = S_MOV_B32 2
    %26 = V_LSHL_B64 killed %25, 2, implicit $exec
    %16 = REG_SEQUENCE killed %4, 17, %12, 18
    %18 = COPY %26
    %17 = BUFFER_LOAD_DWORD_ADDR64 %26, killed %13, 0, 0, 0, 0, 0, 0, implicit $exec
    %20 = V_ADD_F32_e64 0, killed %17, 0, 1065353216, 0, 0, implicit $exec
    %21 = V_MAX_F32_e64 0, killed %20, 0, killed %20, 0, 0, implicit $exec
    BUFFER_STORE_DWORD_ADDR64 killed %21, %26, killed %16, 0, 0, 0, 0, 0, 0, implicit $exec
    S_ENDPGM 0

...
---
# GCN-LABEL: name: v_clamp_omod_already_set_f32
# GCN: %20:vgpr_32 = V_ADD_F32_e64 0, killed %17, 0, 1065353216, 0, 0, implicit $exec
# GCN: %21:vgpr_32 = V_MAX_F32_e64 0, killed %20, 0, killed %20, 1, 3, implicit $exec
name:            v_clamp_omod_already_set_f32
tracksRegLiveness: true
registers:
  - { id: 0, class: sgpr_64 }
  - { id: 1, class: sreg_32_xm0 }
  - { id: 2, class: sgpr_32 }
  - { id: 3, class: vgpr_32 }
  - { id: 4, class: sreg_64_xexec }
  - { id: 5, class: sreg_64_xexec }
  - { id: 6, class: sreg_32 }
  - { id: 7, class: sreg_32 }
  - { id: 8, class: sreg_32_xm0 }
  - { id: 9, class: sreg_64 }
  - { id: 10, class: sreg_32_xm0 }
  - { id: 11, class: sreg_32_xm0 }
  - { id: 12, class: sgpr_64 }
  - { id: 13, class: sgpr_128 }
  - { id: 14, class: sreg_32_xm0 }
  - { id: 15, class: sreg_64 }
  - { id: 16, class: sgpr_128 }
  - { id: 17, class: vgpr_32 }
  - { id: 18, class: vreg_64 }
  - { id: 19, class: vgpr_32 }
  - { id: 20, class: vgpr_32 }
  - { id: 21, class: vgpr_32 }
  - { id: 22, class: vgpr_32 }
  - { id: 23, class: vreg_64 }
  - { id: 24, class: vgpr_32 }
  - { id: 25, class: vreg_64 }
  - { id: 26, class: vreg_64 }
liveins:
  - { reg: '$sgpr0_sgpr1', virtual-reg: '%0' }
  - { reg: '$vgpr0', virtual-reg: '%3' }
body:             |
  bb.0:
    liveins: $sgpr0_sgpr1, $vgpr0

    %3 = COPY $vgpr0
    %0 = COPY $sgpr0_sgpr1
    %4 = S_LOAD_DWORDX2_IMM %0, 9, 0, 0 :: (non-temporal dereferenceable invariant load 8 from `i64 addrspace(4)* undef`)
    %5 = S_LOAD_DWORDX2_IMM %0, 11, 0, 0 :: (non-temporal dereferenceable invariant load 8 from `i64 addrspace(4)* undef`)
    %24 = V_ASHRREV_I32_e32 31, %3, implicit $exec
    %25 = REG_SEQUENCE %3, 1, %24, 2
    %10 = S_MOV_B32 61440
    %11 = S_MOV_B32 0
    %12 = REG_SEQUENCE killed %11, 1, killed %10, 2
    %13 = REG_SEQUENCE killed %5, 17, %12, 18
    %14 = S_MOV_B32 2
    %26 = V_LSHL_B64 killed %25, 2, implicit $exec
    %16 = REG_SEQUENCE killed %4, 17, %12, 18
    %18 = COPY %26
    %17 = BUFFER_LOAD_DWORD_ADDR64 %26, killed %13, 0, 0, 0, 0, 0, 0, implicit $exec
    %20 = V_ADD_F32_e64 0, killed %17, 0, 1065353216, 0, 0, implicit $exec
    %21 = V_MAX_F32_e64 0, killed %20, 0, killed %20, 1, 3, implicit $exec
    BUFFER_STORE_DWORD_ADDR64 killed %21, %26, killed %16, 0, 0, 0, 0, 0, 0, implicit $exec
    S_ENDPGM 0
...
---
# Don't fold a mul that looks like an omod if itself has omod set

# GCN-LABEL: name: v_omod_mul_omod_already_set_f32
# GCN: %20:vgpr_32 = V_ADD_F32_e64 0, killed %17, 0, 1065353216, 0, 0, implicit $exec
# GCN-NEXT: %21:vgpr_32 = V_MUL_F32_e64 0, killed %20, 0, 1056964608, 0, 3, implicit $exec
name:            v_omod_mul_omod_already_set_f32
tracksRegLiveness: true
registers:
  - { id: 0, class: sgpr_64 }
  - { id: 1, class: sreg_32_xm0 }
  - { id: 2, class: sgpr_32 }
  - { id: 3, class: vgpr_32 }
  - { id: 4, class: sreg_64_xexec }
  - { id: 5, class: sreg_64_xexec }
  - { id: 6, class: sreg_32 }
  - { id: 7, class: sreg_32 }
  - { id: 8, class: sreg_32_xm0 }
  - { id: 9, class: sreg_64 }
  - { id: 10, class: sreg_32_xm0 }
  - { id: 11, class: sreg_32_xm0 }
  - { id: 12, class: sgpr_64 }
  - { id: 13, class: sgpr_128 }
  - { id: 14, class: sreg_32_xm0 }
  - { id: 15, class: sreg_64 }
  - { id: 16, class: sgpr_128 }
  - { id: 17, class: vgpr_32 }
  - { id: 18, class: vreg_64 }
  - { id: 19, class: vgpr_32 }
  - { id: 20, class: vgpr_32 }
  - { id: 21, class: vgpr_32 }
  - { id: 22, class: vgpr_32 }
  - { id: 23, class: vreg_64 }
  - { id: 24, class: vgpr_32 }
  - { id: 25, class: vreg_64 }
  - { id: 26, class: vreg_64 }
liveins:
  - { reg: '$sgpr0_sgpr1', virtual-reg: '%0' }
  - { reg: '$vgpr0', virtual-reg: '%3' }
body:             |
  bb.0:
    liveins: $sgpr0_sgpr1, $vgpr0

    %3 = COPY $vgpr0
    %0 = COPY $sgpr0_sgpr1
    %4 = S_LOAD_DWORDX2_IMM %0, 9, 0, 0 :: (non-temporal dereferenceable invariant load 8 from `i64 addrspace(4)* undef`)
    %5 = S_LOAD_DWORDX2_IMM %0, 11, 0, 0 :: (non-temporal dereferenceable invariant load 8 from `i64 addrspace(4)* undef`)
    %24 = V_ASHRREV_I32_e32 31, %3, implicit $exec
    %25 = REG_SEQUENCE %3, 1, %24, 2
    %10 = S_MOV_B32 61440
    %11 = S_MOV_B32 0
    %12 = REG_SEQUENCE killed %11, 1, killed %10, 2
    %13 = REG_SEQUENCE killed %5, 17, %12, 18
    %14 = S_MOV_B32 2
    %26 = V_LSHL_B64 killed %25, 2, implicit $exec
    %16 = REG_SEQUENCE killed %4, 17, %12, 18
    %18 = COPY %26
    %17 = BUFFER_LOAD_DWORD_ADDR64 %26, killed %13, 0, 0, 0, 0, 0, 0, implicit $exec
    %20 = V_ADD_F32_e64 0, killed %17, 0, 1065353216, 0, 0, implicit $exec
    %21 = V_MUL_F32_e64 0, killed %20, 0, 1056964608, 0, 3, implicit $exec
    BUFFER_STORE_DWORD_ADDR64 killed %21, %26, killed %16, 0, 0, 0, 0, 0, 0, implicit $exec
    S_ENDPGM 0

...
---
# Don't fold a mul that looks like an omod if itself has clamp set
# This might be OK, but would require folding the clamp at the same time.
# GCN-LABEL: name: v_omod_mul_clamp_already_set_f32
# GCN: %20:vgpr_32 = V_ADD_F32_e64 0, killed %17, 0, 1065353216, 0, 0, implicit $exec
# GCN-NEXT: %21:vgpr_32 = V_MUL_F32_e64 0, killed %20, 0, 1056964608, 1, 0, implicit $exec

name:            v_omod_mul_clamp_already_set_f32
tracksRegLiveness: true
registers:
  - { id: 0, class: sgpr_64 }
  - { id: 1, class: sreg_32_xm0 }
  - { id: 2, class: sgpr_32 }
  - { id: 3, class: vgpr_32 }
  - { id: 4, class: sreg_64_xexec }
  - { id: 5, class: sreg_64_xexec }
  - { id: 6, class: sreg_32 }
  - { id: 7, class: sreg_32 }
  - { id: 8, class: sreg_32_xm0 }
  - { id: 9, class: sreg_64 }
  - { id: 10, class: sreg_32_xm0 }
  - { id: 11, class: sreg_32_xm0 }
  - { id: 12, class: sgpr_64 }
  - { id: 13, class: sgpr_128 }
  - { id: 14, class: sreg_32_xm0 }
  - { id: 15, class: sreg_64 }
  - { id: 16, class: sgpr_128 }
  - { id: 17, class: vgpr_32 }
  - { id: 18, class: vreg_64 }
  - { id: 19, class: vgpr_32 }
  - { id: 20, class: vgpr_32 }
  - { id: 21, class: vgpr_32 }
  - { id: 22, class: vgpr_32 }
  - { id: 23, class: vreg_64 }
  - { id: 24, class: vgpr_32 }
  - { id: 25, class: vreg_64 }
  - { id: 26, class: vreg_64 }
liveins:
  - { reg: '$sgpr0_sgpr1', virtual-reg: '%0' }
  - { reg: '$vgpr0', virtual-reg: '%3' }
body:             |
  bb.0:
    liveins: $sgpr0_sgpr1, $vgpr0

    %3 = COPY $vgpr0
    %0 = COPY $sgpr0_sgpr1
    %4 = S_LOAD_DWORDX2_IMM %0, 9, 0, 0 :: (non-temporal dereferenceable invariant load 8 from `i64 addrspace(4)* undef`)
    %5 = S_LOAD_DWORDX2_IMM %0, 11, 0, 0 :: (non-temporal dereferenceable invariant load 8 from `i64 addrspace(4)* undef`)
    %24 = V_ASHRREV_I32_e32 31, %3, implicit $exec
    %25 = REG_SEQUENCE %3, 1, %24, 2
    %10 = S_MOV_B32 61440
    %11 = S_MOV_B32 0
    %12 = REG_SEQUENCE killed %11, 1, killed %10, 2
    %13 = REG_SEQUENCE killed %5, 17, %12, 18
    %14 = S_MOV_B32 2
    %26 = V_LSHL_B64 killed %25, 2, implicit $exec
    %16 = REG_SEQUENCE killed %4, 17, %12, 18
    %18 = COPY %26
    %17 = BUFFER_LOAD_DWORD_ADDR64 %26, killed %13, 0, 0, 0, 0, 0, 0, implicit $exec
    %20 = V_ADD_F32_e64 0, killed %17, 0, 1065353216, 0, 0, implicit $exec
    %21 = V_MUL_F32_e64 0, killed %20, 0, 1056964608, 1, 0, implicit $exec
    BUFFER_STORE_DWORD_ADDR64 killed %21, %26, killed %16, 0, 0, 0, 0, 0, 0, implicit $exec
    S_ENDPGM 0

...













---
# Don't fold a mul that looks like an omod if itself has omod set

# GCN-LABEL: name: v_omod_add_omod_already_set_f32
# GCN: %20:vgpr_32 = V_ADD_F32_e64 0, killed %17, 0, 1065353216, 0, 0, implicit $exec
# GCN-NEXT: %21:vgpr_32 = V_ADD_F32_e64 0, killed %20, 0, killed %20, 0, 3, implicit $exec
name:            v_omod_add_omod_already_set_f32
tracksRegLiveness: true
registers:
  - { id: 0, class: sgpr_64 }
  - { id: 1, class: sreg_32_xm0 }
  - { id: 2, class: sgpr_32 }
  - { id: 3, class: vgpr_32 }
  - { id: 4, class: sreg_64_xexec }
  - { id: 5, class: sreg_64_xexec }
  - { id: 6, class: sreg_32 }
  - { id: 7, class: sreg_32 }
  - { id: 8, class: sreg_32_xm0 }
  - { id: 9, class: sreg_64 }
  - { id: 10, class: sreg_32_xm0 }
  - { id: 11, class: sreg_32_xm0 }
  - { id: 12, class: sgpr_64 }
  - { id: 13, class: sgpr_128 }
  - { id: 14, class: sreg_32_xm0 }
  - { id: 15, class: sreg_64 }
  - { id: 16, class: sgpr_128 }
  - { id: 17, class: vgpr_32 }
  - { id: 18, class: vreg_64 }
  - { id: 19, class: vgpr_32 }
  - { id: 20, class: vgpr_32 }
  - { id: 21, class: vgpr_32 }
  - { id: 22, class: vgpr_32 }
  - { id: 23, class: vreg_64 }
  - { id: 24, class: vgpr_32 }
  - { id: 25, class: vreg_64 }
  - { id: 26, class: vreg_64 }
liveins:
  - { reg: '$sgpr0_sgpr1', virtual-reg: '%0' }
  - { reg: '$vgpr0', virtual-reg: '%3' }
body:             |
  bb.0:
    liveins: $sgpr0_sgpr1, $vgpr0

    %3 = COPY $vgpr0
    %0 = COPY $sgpr0_sgpr1
    %4 = S_LOAD_DWORDX2_IMM %0, 9, 0, 0 :: (non-temporal dereferenceable invariant load 8 from `i64 addrspace(4)* undef`)
    %5 = S_LOAD_DWORDX2_IMM %0, 11, 0, 0 :: (non-temporal dereferenceable invariant load 8 from `i64 addrspace(4)* undef`)
    %24 = V_ASHRREV_I32_e32 31, %3, implicit $exec
    %25 = REG_SEQUENCE %3, 1, %24, 2
    %10 = S_MOV_B32 61440
    %11 = S_MOV_B32 0
    %12 = REG_SEQUENCE killed %11, 1, killed %10, 2
    %13 = REG_SEQUENCE killed %5, 17, %12, 18
    %14 = S_MOV_B32 2
    %26 = V_LSHL_B64 killed %25, 2, implicit $exec
    %16 = REG_SEQUENCE killed %4, 17, %12, 18
    %18 = COPY %26
    %17 = BUFFER_LOAD_DWORD_ADDR64 %26, killed %13, 0, 0, 0, 0, 0, 0, implicit $exec
    %20 = V_ADD_F32_e64 0, killed %17, 0, 1065353216, 0, 0, implicit $exec
    %21 = V_ADD_F32_e64 0, killed %20, 0, killed %20, 0, 3, implicit $exec
    BUFFER_STORE_DWORD_ADDR64 killed %21, %26, killed %16, 0, 0, 0, 0, 0, 0, implicit $exec
    S_ENDPGM 0

...
---
# Don't fold a mul that looks like an omod if itself has clamp set
# This might be OK, but would require folding the clamp at the same time.
# GCN-LABEL: name: v_omod_add_clamp_already_set_f32
# GCN: %20:vgpr_32 = V_ADD_F32_e64 0, killed %17, 0, 1065353216, 0, 0, implicit $exec
# GCN-NEXT: %21:vgpr_32 = V_ADD_F32_e64 0, killed %20, 0, killed %20, 1, 0, implicit $exec

name:            v_omod_add_clamp_already_set_f32
tracksRegLiveness: true
registers:
  - { id: 0, class: sgpr_64 }
  - { id: 1, class: sreg_32_xm0 }
  - { id: 2, class: sgpr_32 }
  - { id: 3, class: vgpr_32 }
  - { id: 4, class: sreg_64_xexec }
  - { id: 5, class: sreg_64_xexec }
  - { id: 6, class: sreg_32 }
  - { id: 7, class: sreg_32 }
  - { id: 8, class: sreg_32_xm0 }
  - { id: 9, class: sreg_64 }
  - { id: 10, class: sreg_32_xm0 }
  - { id: 11, class: sreg_32_xm0 }
  - { id: 12, class: sgpr_64 }
  - { id: 13, class: sgpr_128 }
  - { id: 14, class: sreg_32_xm0 }
  - { id: 15, class: sreg_64 }
  - { id: 16, class: sgpr_128 }
  - { id: 17, class: vgpr_32 }
  - { id: 18, class: vreg_64 }
  - { id: 19, class: vgpr_32 }
  - { id: 20, class: vgpr_32 }
  - { id: 21, class: vgpr_32 }
  - { id: 22, class: vgpr_32 }
  - { id: 23, class: vreg_64 }
  - { id: 24, class: vgpr_32 }
  - { id: 25, class: vreg_64 }
  - { id: 26, class: vreg_64 }
liveins:
  - { reg: '$sgpr0_sgpr1', virtual-reg: '%0' }
  - { reg: '$vgpr0', virtual-reg: '%3' }
body:             |
  bb.0:
    liveins: $sgpr0_sgpr1, $vgpr0

    %3 = COPY $vgpr0
    %0 = COPY $sgpr0_sgpr1
    %4 = S_LOAD_DWORDX2_IMM %0, 9, 0, 0 :: (non-temporal dereferenceable invariant load 8 from `i64 addrspace(4)* undef`)
    %5 = S_LOAD_DWORDX2_IMM %0, 11, 0, 0 :: (non-temporal dereferenceable invariant load 8 from `i64 addrspace(4)* undef`)
    %24 = V_ASHRREV_I32_e32 31, %3, implicit $exec
    %25 = REG_SEQUENCE %3, 1, %24, 2
    %10 = S_MOV_B32 61440
    %11 = S_MOV_B32 0
    %12 = REG_SEQUENCE killed %11, 1, killed %10, 2
    %13 = REG_SEQUENCE killed %5, 17, %12, 18
    %14 = S_MOV_B32 2
    %26 = V_LSHL_B64 killed %25, 2, implicit $exec
    %16 = REG_SEQUENCE killed %4, 17, %12, 18
    %18 = COPY %26
    %17 = BUFFER_LOAD_DWORD_ADDR64 %26, killed %13, 0, 0, 0, 0, 0, 0, implicit $exec
    %20 = V_ADD_F32_e64 0, killed %17, 0, 1065353216, 0, 0, implicit $exec
    %21 = V_ADD_F32_e64 0, killed %20, 0, killed %20, 1, 0, implicit $exec
    BUFFER_STORE_DWORD_ADDR64 killed %21, %26, killed %16, 0, 0, 0, 0, 0, 0, implicit $exec
    S_ENDPGM 0

...
---

# Pass used to crash with immediate second operand of max
name:            v_max_reg_imm_f32
tracksRegLiveness: true
registers:
  - { id: 0, class: vgpr_32 }
  - { id: 1, class: vgpr_32 }
body:             |
  bb.0:
    liveins: $vgpr0

    %0 = COPY $vgpr0
    %1 = V_MAX_F32_e64 0, killed %0, 0, 1056964608, 1, 0, implicit $exec

...