llvm/flang/test/Lower/CUDA/cuda-device-proc.cuf

! 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

  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)
end

! CHECK-LABEL: func.func @_QPdevsub() attributes {cuf.proc_attr = #cuf.cuda_proc<global>}
! CHECK: fir.call @__syncthreads()
! CHECK: fir.call @__syncwarp(%{{.*}}) proc_attrs<bind_c> fastmath<contract> : (!fir.ref<i32>) -> ()
! CHECK: fir.call @__threadfence()
! CHECK: fir.call @__threadfence_block()
! CHECK: fir.call @__threadfence_system()
! CHECK: %{{.*}} = fir.call @__syncthreads_and(%{{.*}}) proc_attrs<bind_c> fastmath<contract> : (!fir.ref<i32>) -> i32
! CHECK: %{{.*}} = fir.call @__syncthreads_count(%{{.*}}) proc_attrs<bind_c> fastmath<contract> : (!fir.ref<i32>) -> i32
! CHECK: %{{.*}} = fir.call @__syncthreads_or(%{{.*}}) proc_attrs<bind_c> fastmath<contract> : (!fir.ref<i32>) -> i32

! CHECK: func.func private @__syncthreads() attributes {cuf.proc_attr = #cuf.cuda_proc<device>, fir.bindc_name = "__syncthreads", fir.proc_attrs = #fir.proc_attrs<bind_c>}
! CHECK: func.func private @__syncwarp(!fir.ref<i32> {cuf.data_attr = #cuf.cuda<device>}) attributes {cuf.proc_attr = #cuf.cuda_proc<device>, fir.bindc_name = "__syncwarp", fir.proc_attrs = #fir.proc_attrs<bind_c>}
! CHECK: func.func private @__threadfence() attributes {cuf.proc_attr = #cuf.cuda_proc<device>, fir.bindc_name = "__threadfence", fir.proc_attrs = #fir.proc_attrs<bind_c>}
! CHECK: func.func private @__threadfence_block() attributes {cuf.proc_attr = #cuf.cuda_proc<device>, fir.bindc_name = "__threadfence_block", fir.proc_attrs = #fir.proc_attrs<bind_c>}
! CHECK: func.func private @__threadfence_system() attributes {cuf.proc_attr = #cuf.cuda_proc<device>, fir.bindc_name = "__threadfence_system", fir.proc_attrs = #fir.proc_attrs<bind_c>}
! CHECK: func.func private @__syncthreads_and(!fir.ref<i32> {cuf.data_attr = #cuf.cuda<device>}) -> i32 attributes {cuf.proc_attr = #cuf.cuda_proc<device>, fir.bindc_name = "__syncthreads_and", fir.proc_attrs = #fir.proc_attrs<bind_c>}
! CHECK: func.func private @__syncthreads_count(!fir.ref<i32> {cuf.data_attr = #cuf.cuda<device>}) -> i32 attributes {cuf.proc_attr = #cuf.cuda_proc<device>, fir.bindc_name = "__syncthreads_count", fir.proc_attrs = #fir.proc_attrs<bind_c>}
! CHECK: func.func private @__syncthreads_or(!fir.ref<i32> {cuf.data_attr = #cuf.cuda<device>}) -> i32 attributes {cuf.proc_attr = #cuf.cuda_proc<device>, fir.bindc_name = "__syncthreads_or", fir.proc_attrs = #fir.proc_attrs<bind_c>}