! RUN: bbc -emit-hlfir -fcuda %s -o - | FileCheck %s

! Test CUDA Fortran procedures available in cudadevice module

attributes(global) subroutine devsub()
  implicit none
  integer :: ret
  real(4) :: af
  real(8) :: ad
  integer(4) :: ai
  integer(8) :: al
  integer(8) :: time

  call syncthreads()
  call syncwarp(1)
  call threadfence()
  call threadfence_block()
  call threadfence_system()
  ret = syncthreads_and(1)
  ret = syncthreads_count(1)
  ret = syncthreads_or(1)

  ai = atomicadd(ai, 1_4)
  al = atomicadd(al, 1_8)
  af = atomicadd(af, 1.0_4)
  ad = atomicadd(ad, 1.0_8)
  
  ai = atomicsub(ai, 1_4)
  al = atomicsub(al, 1_8)
  af = atomicsub(af, 1.0_4)
  ad = atomicsub(ad, 1.0_8)
  
  ai = atomicmax(ai, 1_4)
  al = atomicmax(al, 1_8)
  af = atomicmax(af, 1.0_4)
  ad = atomicmax(ad, 1.0_8)
  
  ai = atomicmin(ai, 1_4)
  al = atomicmin(al, 1_8)
  af = atomicmin(af, 1.0_4)
  ad = atomicmin(ad, 1.0_8)
  
  ai = atomicand(ai, 1_4)
  ai = atomicor(ai, 1_4)
  ai = atomicinc(ai, 1_4)
  ai = atomicdec(ai, 1_4)

  time = clock64()
end

! CHECK-LABEL: func.func @_QPdevsub() attributes {cuf.proc_attr = #cuf.cuda_proc<global>}
! CHECK: fir.call @llvm.nvvm.barrier0() fastmath<contract> : () -> ()
! CHECK: fir.call @llvm.nvvm.bar.warp.sync(%c1{{.*}}) fastmath<contract> : (i32) -> ()
! CHECK: fir.call @llvm.nvvm.membar.gl() fastmath<contract> : () -> ()
! CHECK: fir.call @llvm.nvvm.membar.cta() fastmath<contract> : () -> ()
! CHECK: fir.call @llvm.nvvm.membar.sys() fastmath<contract> : () -> ()
! CHECK: %{{.*}} = fir.call @llvm.nvvm.barrier0.and(%c1_i32_0) fastmath<contract> : (i32) -> i32
! CHECK: %{{.*}} = fir.call @llvm.nvvm.barrier0.popc(%c1_i32_1) fastmath<contract> : (i32) -> i32
! CHECK: %{{.*}} = fir.call @llvm.nvvm.barrier0.or(%c1_i32_2) fastmath<contract> : (i32) -> i32
! CHECK: %{{.*}} = llvm.atomicrmw add  %{{.*}}, %{{.*}} seq_cst : !llvm.ptr, i32
! CHECK: %{{.*}} = llvm.atomicrmw add  %{{.*}}, %{{.*}} seq_cst : !llvm.ptr, i64
! CHECK: %{{.*}} = llvm.atomicrmw fadd %{{.*}}, %{{.*}} seq_cst : !llvm.ptr, f32
! CHECK: %{{.*}} = llvm.atomicrmw fadd %{{.*}}, %{{.*}} seq_cst : !llvm.ptr, f64

! CHECK: %{{.*}} = llvm.atomicrmw sub  %{{.*}}, %{{.*}} seq_cst : !llvm.ptr, i32
! CHECK: %{{.*}} = llvm.atomicrmw sub  %{{.*}}, %{{.*}} seq_cst : !llvm.ptr, i64
! CHECK: %{{.*}} = llvm.atomicrmw fsub %{{.*}}, %{{.*}} seq_cst : !llvm.ptr, f32
! CHECK: %{{.*}} = llvm.atomicrmw fsub %{{.*}}, %{{.*}} seq_cst : !llvm.ptr, f64

! CHECK: %{{.*}} = llvm.atomicrmw max  %{{.*}}, %{{.*}} seq_cst : !llvm.ptr, i32
! CHECK: %{{.*}} = llvm.atomicrmw max  %{{.*}}, %{{.*}} seq_cst : !llvm.ptr, i64
! CHECK: %{{.*}} = llvm.atomicrmw fmax %{{.*}}, %{{.*}} seq_cst : !llvm.ptr, f32
! CHECK: %{{.*}} = llvm.atomicrmw fmax %{{.*}}, %{{.*}} seq_cst : !llvm.ptr, f64

! CHECK: %{{.*}} = llvm.atomicrmw min  %{{.*}}, %{{.*}} seq_cst : !llvm.ptr, i32
! CHECK: %{{.*}} = llvm.atomicrmw min  %{{.*}}, %{{.*}} seq_cst : !llvm.ptr, i64
! CHECK: %{{.*}} = llvm.atomicrmw fmin %{{.*}}, %{{.*}} seq_cst : !llvm.ptr, f32
! CHECK: %{{.*}} = llvm.atomicrmw fmin %{{.*}}, %{{.*}} seq_cst : !llvm.ptr, f64

! CHECK: %{{.*}} = llvm.atomicrmw _and  %{{.*}}, %{{.*}} seq_cst : !llvm.ptr, i32
! CHECK: %{{.*}} = llvm.atomicrmw _or  %{{.*}}, %{{.*}} seq_cst : !llvm.ptr, i32
! CHECK: %{{.*}} = llvm.atomicrmw uinc_wrap  %{{.*}}, %{{.*}} seq_cst : !llvm.ptr, i32
! CHECK: %{{.*}} = llvm.atomicrmw udec_wrap  %{{.*}}, %{{.*}} seq_cst : !llvm.ptr, i32

! CHECK: fir.call @llvm.nvvm.read.ptx.sreg.clock64()

subroutine host1()
  integer, device :: a(32)
  integer, device :: ret
  integer :: i, j

block; use cudadevice
  !$cuf kernel do(1) <<<*,32>>>
  do i = 1, 32
    a(i) = a(i) * 2.0
    call syncthreads()
    a(i) = a(i) + a(j) - 34.0

    call syncwarp(1)
    ret = syncthreads_and(1)
    ret = syncthreads_count(1)
    ret = syncthreads_or(1)
  end do
end block
end 

! CHECK-LABEL: func.func @_QPhost1()
! CHECK: cuf.kernel
! CHECK: fir.call @llvm.nvvm.barrier0() fastmath<contract> : () -> ()
! CHECK: fir.call @llvm.nvvm.bar.warp.sync(%c1{{.*}}) fastmath<contract> : (i32) -> ()
! CHECK: fir.call @llvm.nvvm.barrier0.and(%c1{{.*}}) fastmath<contract> : (i32) -> i32
! CHECK: fir.call @llvm.nvvm.barrier0.popc(%c1{{.*}}) fastmath<contract> : (i32) -> i32
! CHECK: fir.call @llvm.nvvm.barrier0.or(%c1{{.*}}) fastmath<contract> : (i32) -> i32

attributes(device) subroutine testMatch()
  integer :: a, ipred, mask, v32
  integer(8) :: v64
  real(4) :: r4
  real(8) :: r8
  a = match_all_sync(mask, v32, ipred)
  a = match_all_sync(mask, v64, ipred)
  a = match_all_sync(mask, r4, ipred)
  a = match_all_sync(mask, r8, ipred)
end subroutine

! CHECK-LABEL: func.func @_QPtestmatch()
! CHECK: fir.call @llvm.nvvm.match.all.sync.i32p
! CHECK: fir.call @llvm.nvvm.match.all.sync.i64p
! CHECK: fir.convert %{{.*}} : (f32) -> i32
! CHECK: fir.call @llvm.nvvm.match.all.sync.i32p
! CHECK: fir.convert %{{.*}} : (f64) -> i64
! CHECK: fir.call @llvm.nvvm.match.all.sync.i64p

attributes(device) subroutine testMatchAny()
  integer :: a, mask, v32
  integer(8) :: v64
  real(4) :: r4
  real(8) :: r8
  a = match_any_sync(mask, v32)
  a = match_any_sync(mask, v64)
  a = match_any_sync(mask, r4)
  a = match_any_sync(mask, r8)
end subroutine

! CHECK-LABEL: func.func @_QPtestmatchany()
! CHECK: fir.call @llvm.nvvm.match.any.sync.i32p
! CHECK: fir.call @llvm.nvvm.match.any.sync.i64p
! CHECK: fir.convert %{{.*}} : (f32) -> i32
! CHECK: fir.call @llvm.nvvm.match.any.sync.i32p
! CHECK: fir.convert %{{.*}} : (f64) -> i64
! CHECK: fir.call @llvm.nvvm.match.any.sync.i64p

attributes(device) subroutine testAtomic(aa, n)
  integer :: aa(*)
  integer, intent(in) :: n
  integer :: a, istat, j, i
  real :: r
  istat = atomicexch(a,0)
  istat = atomicexch(r, 0.0)
  istat = atomicxor(a, j)
  istat = atomiccas(a, i, 14)
  do i = 1, n
    istat = atomicxor(aa, i)
    istat = atomiccas(aa, i, 14)
    istat = atomicexch(aa, 0)
  end do
end subroutine

! CHECK-LABEL: func.func @_QPtestatomic
! CHECK: llvm.atomicrmw xchg %{{.*}}, %c0{{.*}} seq_cst : !llvm.ptr, i32
! CHECK: llvm.atomicrmw xchg %{{.*}}, %{{.*}} seq_cst : !llvm.ptr, f32
! CHECK: llvm.atomicrmw _xor %{{.*}}, %{{.*}} seq_cst : !llvm.ptr, i32
! CHECK: %[[ADDR:.*]] = builtin.unrealized_conversion_cast %{{.*}}#0 : !fir.ref<i32> to !llvm.ptr
! CHECK: llvm.cmpxchg %[[ADDR]], %{{.*}}, %c14{{.*}} acq_rel monotonic : !llvm.ptr, i32
! CHECK: fir.do_loop
! CHECK: llvm.atomicrmw _xor %{{.*}}, %{{.*}} seq_cst : !llvm.ptr, i32
! CHECK: %[[ADDR:.*]] = builtin.unrealized_conversion_cast %{{.*}}#1 : !fir.ref<!fir.array<?xi32>> to !llvm.ptr 
! CHECK: llvm.cmpxchg %[[ADDR]], %{{.*}}, %c14{{.*}} acq_rel monotonic : !llvm.ptr, i32
! CHECK: llvm.atomicrmw xchg %{{.*}}, %c0{{.*}} seq_cst : !llvm.ptr, i32

attributes(device) subroutine testAtomic2()
  integer(8) :: a, i, istat
  istat = atomiccas(a, i, 14)
end subroutine

! CHECK-LABEL: func.func @_QPtestatomic2()
! CHECK: %[[VAL:.*]] = fir.convert %c14{{.*}} : (i32) -> i64
! CHECK: %[[ADDR:.*]] = builtin.unrealized_conversion_cast %{{.*}}#0 : !fir.ref<i64> to !llvm.ptr
! CHECK: llvm.cmpxchg %{{.*}}, %{{.*}}, %[[VAL]] acq_rel monotonic : !llvm.ptr, i64

attributes(device) subroutine testAtomic3()
  real :: a, i, istat
  istat = atomiccas(a, i, 14.0)
end subroutine

! CHECK-LABEL: func.func @_QPtestatomic3()
! CHECK: %[[BCAST1:.*]] = llvm.bitcast %{{.*}} : f32 to i32
! CHECK: %[[BCAST2:.*]] = llvm.bitcast %{{.*}} : f32 to i32
! CHECK: %[[CAST:.*]] = builtin.unrealized_conversion_cast %{{.*}}#0 : !fir.ref<f32> to !llvm.ptr
! CHECK: llvm.cmpxchg %[[CAST]], %[[BCAST1]], %[[BCAST2]] acq_rel monotonic : !llvm.ptr, i32

attributes(device) subroutine testAtomic4()
  real(8) :: a, i, istat
  istat = atomiccas(a, i, 14.0d0)
end subroutine

! CHECK-LABEL: func.func @_QPtestatomic4()
! CHECK: %[[BCAST1:.*]] = llvm.bitcast %{{.*}} : f64 to i64
! CHECK: %[[BCAST2:.*]] = llvm.bitcast %{{.*}} : f64 to i64
! CHECK: %[[CAST:.*]] = builtin.unrealized_conversion_cast %{{.*}}#0 : !fir.ref<f64> to !llvm.ptr
! CHECK: %[[ATOMIC:.*]] = llvm.cmpxchg %[[CAST]], %[[BCAST1]], %[[BCAST2]] acq_rel monotonic : !llvm.ptr, i64
! CHECK: %[[RES:.*]] = llvm.extractvalue %[[ATOMIC]][1] : !llvm.struct<(i64, i1)> 

attributes(global) subroutine __ldXXi4(b)
  integer, device :: b(*)
  integer, device :: x(4)
  x(1:4) = __ldca(b(i:j))
  x = __ldcg(b(i:j))
  x = __ldcs(b(i:j))
  x(1:4) = __ldlu(b(i:j))
  x(1:4) = __ldcv(b(i:j))
end

! CHECK-LABEL: func.func @_QP__ldxxi4
! CHECK: fir.call @__ldca_i4x4_(%{{.*}}, %{{.*}}) fastmath<contract> : (!fir.ref<!fir.array<4xi32>>, !fir.ref<!fir.array<?xi32>>) -> ()
! CHECK: fir.call @__ldcg_i4x4_(%{{.*}}, %{{.*}}) fastmath<contract> : (!fir.ref<!fir.array<4xi32>>, !fir.ref<!fir.array<?xi32>>) -> ()
! CHECK: fir.call @__ldcs_i4x4_(%{{.*}}, %{{.*}}) fastmath<contract> : (!fir.ref<!fir.array<4xi32>>, !fir.ref<!fir.array<?xi32>>) -> ()
! CHECK: fir.call @__ldlu_i4x4_(%{{.*}}, %{{.*}}) fastmath<contract> : (!fir.ref<!fir.array<4xi32>>, !fir.ref<!fir.array<?xi32>>) -> ()
! CHECK: fir.call @__ldcv_i4x4_(%{{.*}}, %{{.*}}) fastmath<contract> : (!fir.ref<!fir.array<4xi32>>, !fir.ref<!fir.array<?xi32>>) -> ()

attributes(global) subroutine __ldXXi8(b)
  integer(8), device :: b(*)
  integer(8), device :: x(2)
  x(1:2) = __ldca(b(i:j))
  x = __ldcg(b(i:j))
  x = __ldcs(b(i:j))
  x(1:2) = __ldlu(b(i:j))
  x(1:2) = __ldcv(b(i:j))
end

! CHECK-LABEL: func.func @_QP__ldxxi8
! CHECK: fir.call @__ldca_i8x2_(%{{.*}}, %{{.*}}) fastmath<contract> : (!fir.ref<!fir.array<2xi64>>, !fir.ref<!fir.array<?xi64>>) -> ()
! CHECK: fir.call @__ldcg_i8x2_(%{{.*}}, %{{.*}}) fastmath<contract> : (!fir.ref<!fir.array<2xi64>>, !fir.ref<!fir.array<?xi64>>) -> ()
! CHECK: fir.call @__ldcs_i8x2_(%{{.*}}, %{{.*}}) fastmath<contract> : (!fir.ref<!fir.array<2xi64>>, !fir.ref<!fir.array<?xi64>>) -> ()
! CHECK: fir.call @__ldlu_i8x2_(%{{.*}}, %{{.*}}) fastmath<contract> : (!fir.ref<!fir.array<2xi64>>, !fir.ref<!fir.array<?xi64>>) -> ()
! CHECK: fir.call @__ldcv_i8x2_(%{{.*}}, %{{.*}}) fastmath<contract> : (!fir.ref<!fir.array<2xi64>>, !fir.ref<!fir.array<?xi64>>) -> ()

attributes(global) subroutine __ldXXr4(b)
  real, device :: b(*)
  real, device :: x(4)
  x(1:4) = __ldca(b(i:j))
  x = __ldcg(b(i:j))
  x = __ldcs(b(i:j))
  x(1:4) = __ldlu(b(i:j))
  x(1:4) = __ldcv(b(i:j))
end

! CHECK-LABEL: func.func @_QP__ldxxr4
! CHECK: fir.call @__ldca_r4x4_(%{{.*}}, %{{.*}}) fastmath<contract> : (!fir.ref<!fir.array<4xf32>>, !fir.ref<!fir.array<?xf32>>) -> ()
! CHECK: fir.call @__ldcg_r4x4_(%{{.*}}, %{{.*}}) fastmath<contract> : (!fir.ref<!fir.array<4xf32>>, !fir.ref<!fir.array<?xf32>>) -> ()
! CHECK: fir.call @__ldcs_r4x4_(%{{.*}}, %{{.*}}) fastmath<contract> : (!fir.ref<!fir.array<4xf32>>, !fir.ref<!fir.array<?xf32>>) -> ()
! CHECK: fir.call @__ldlu_r4x4_(%{{.*}}, %{{.*}}) fastmath<contract> : (!fir.ref<!fir.array<4xf32>>, !fir.ref<!fir.array<?xf32>>) -> ()
! CHECK: fir.call @__ldcv_r4x4_(%{{.*}}, %{{.*}}) fastmath<contract> : (!fir.ref<!fir.array<4xf32>>, !fir.ref<!fir.array<?xf32>>) -> ()

attributes(global) subroutine __ldXXr2(b)
  real(2), device :: b(*)
  real(2), device :: x(2)
  x(1:2) = __ldca(b(i:j))
  x = __ldcg(b(i:j))
  x = __ldcs(b(i:j))
  x(1:2) = __ldlu(b(i:j))
  x(1:2) = __ldcv(b(i:j))
end

! CHECK-LABEL: func.func @_QP__ldxxr2
! CHECK: fir.call @__ldca_r2x2_(%{{.*}}, %{{.*}}) fastmath<contract> : (!fir.ref<!fir.array<2xf16>>, !fir.ref<!fir.array<?xf16>>) -> ()
! CHECK: fir.call @__ldcg_r2x2_(%{{.*}}, %{{.*}}) fastmath<contract> : (!fir.ref<!fir.array<2xf16>>, !fir.ref<!fir.array<?xf16>>) -> ()
! CHECK: fir.call @__ldcs_r2x2_(%{{.*}}, %{{.*}}) fastmath<contract> : (!fir.ref<!fir.array<2xf16>>, !fir.ref<!fir.array<?xf16>>) -> ()
! CHECK: fir.call @__ldlu_r2x2_(%{{.*}}, %{{.*}}) fastmath<contract> : (!fir.ref<!fir.array<2xf16>>, !fir.ref<!fir.array<?xf16>>) -> ()
! CHECK: fir.call @__ldcv_r2x2_(%{{.*}}, %{{.*}}) fastmath<contract> : (!fir.ref<!fir.array<2xf16>>, !fir.ref<!fir.array<?xf16>>) -> ()

attributes(global) subroutine __ldXXr8(b)
  real(8), device :: b(*)
  real(8), device :: x(2)
  x(1:2) = __ldca(b(i:j))
  x = __ldcg(b(i:j))
  x = __ldcs(b(i:j))
  x(1:2) = __ldlu(b(i:j))
  x(1:2) = __ldcv(b(i:j))
end

! CHECK-LABEL: func.func @_QP__ldxxr8
! CHECK: fir.call @__ldca_r8x2_(%{{.*}}, %{{.*}}) fastmath<contract> : (!fir.ref<!fir.array<2xf64>>, !fir.ref<!fir.array<?xf64>>) -> ()
! CHECK: fir.call @__ldcg_r8x2_(%{{.*}}, %{{.*}}) fastmath<contract> : (!fir.ref<!fir.array<2xf64>>, !fir.ref<!fir.array<?xf64>>) -> ()
! CHECK: fir.call @__ldcs_r8x2_(%{{.*}}, %{{.*}}) fastmath<contract> : (!fir.ref<!fir.array<2xf64>>, !fir.ref<!fir.array<?xf64>>) -> ()
! CHECK: fir.call @__ldlu_r8x2_(%{{.*}}, %{{.*}}) fastmath<contract> : (!fir.ref<!fir.array<2xf64>>, !fir.ref<!fir.array<?xf64>>) -> ()
! CHECK: fir.call @__ldcv_r8x2_(%{{.*}}, %{{.*}}) fastmath<contract> : (!fir.ref<!fir.array<2xf64>>, !fir.ref<!fir.array<?xf64>>) -> ()


! CHECK-DAG: func.func private @__ldca_i4x4_(!fir.ref<!fir.array<4xi32>>, !fir.ref<!fir.array<4xi32>>)
! CHECK-DAG: func.func private @__ldcg_i4x4_(!fir.ref<!fir.array<4xi32>>, !fir.ref<!fir.array<4xi32>>)
! CHECK-DAG: func.func private @__ldcs_i4x4_(!fir.ref<!fir.array<4xi32>>, !fir.ref<!fir.array<4xi32>>)
! CHECK-DAG: func.func private @__ldlu_i4x4_(!fir.ref<!fir.array<4xi32>>, !fir.ref<!fir.array<4xi32>>)
! CHECK-DAG: func.func private @__ldcv_i4x4_(!fir.ref<!fir.array<4xi32>>, !fir.ref<!fir.array<4xi32>>)
! CHECK-DAG: func.func private @__ldca_i8x2_(!fir.ref<!fir.array<2xi64>>, !fir.ref<!fir.array<2xi64>>)
! CHECK-DAG: func.func private @__ldcg_i8x2_(!fir.ref<!fir.array<2xi64>>, !fir.ref<!fir.array<2xi64>>)
! CHECK-DAG: func.func private @__ldcs_i8x2_(!fir.ref<!fir.array<2xi64>>, !fir.ref<!fir.array<2xi64>>)
! CHECK-DAG: func.func private @__ldlu_i8x2_(!fir.ref<!fir.array<2xi64>>, !fir.ref<!fir.array<2xi64>>)
! CHECK-DAG: func.func private @__ldcv_i8x2_(!fir.ref<!fir.array<2xi64>>, !fir.ref<!fir.array<2xi64>>)
! CHECK-DAG: func.func private @__ldca_r4x4_(!fir.ref<!fir.array<4xf32>>, !fir.ref<!fir.array<4xf32>>)
! CHECK-DAG: func.func private @__ldcg_r4x4_(!fir.ref<!fir.array<4xf32>>, !fir.ref<!fir.array<4xf32>>)
! CHECK-DAG: func.func private @__ldcs_r4x4_(!fir.ref<!fir.array<4xf32>>, !fir.ref<!fir.array<4xf32>>)
! CHECK-DAG: func.func private @__ldlu_r4x4_(!fir.ref<!fir.array<4xf32>>, !fir.ref<!fir.array<4xf32>>)
! CHECK-DAG: func.func private @__ldcv_r4x4_(!fir.ref<!fir.array<4xf32>>, !fir.ref<!fir.array<4xf32>>)
! CHECK-DAG: func.func private @__ldca_r2x2_(!fir.ref<!fir.array<2xf16>>, !fir.ref<!fir.array<2xf16>>)
! CHECK-DAG: func.func private @__ldcg_r2x2_(!fir.ref<!fir.array<2xf16>>, !fir.ref<!fir.array<2xf16>>)
! CHECK-DAG: func.func private @__ldcs_r2x2_(!fir.ref<!fir.array<2xf16>>, !fir.ref<!fir.array<2xf16>>)
! CHECK-DAG: func.func private @__ldlu_r2x2_(!fir.ref<!fir.array<2xf16>>, !fir.ref<!fir.array<2xf16>>)
! CHECK-DAG: func.func private @__ldcv_r2x2_(!fir.ref<!fir.array<2xf16>>, !fir.ref<!fir.array<2xf16>>)
! CHECK-DAG: func.func private @__ldca_r8x2_(!fir.ref<!fir.array<2xf64>>, !fir.ref<!fir.array<2xf64>>)
! CHECK-DAG: func.func private @__ldcg_r8x2_(!fir.ref<!fir.array<2xf64>>, !fir.ref<!fir.array<2xf64>>)
! CHECK-DAG: func.func private @__ldcs_r8x2_(!fir.ref<!fir.array<2xf64>>, !fir.ref<!fir.array<2xf64>>)
! CHECK-DAG: func.func private @__ldlu_r8x2_(!fir.ref<!fir.array<2xf64>>, !fir.ref<!fir.array<2xf64>>)
! CHECK-DAG: func.func private @__ldcv_r8x2_(!fir.ref<!fir.array<2xf64>>, !fir.ref<!fir.array<2xf64>>)