diff --git a/flang/module/cudadevice.f90 b/flang/module/cudadevice.f90
index d815b30923cfc..65c703df0a905 100644
--- a/flang/module/cudadevice.f90
+++ b/flang/module/cudadevice.f90
@@ -399,6 +399,13 @@ attributes(device) subroutine sincospi(x, y, z) bind(c,name='__nv_sincospi')
     end function
   end interface
 
+  interface __fdividef
+    attributes(device) real function __fdividef(r,d) bind(c, name='__nv_fast_fdividef')
+      !dir$ ignore_tkr (d) r, (d) d
+      real, value :: r,d
+    end function
+  end interface
+
   interface __sinf
     attributes(device) real function __sinf(r) bind(c, name='__nv_sinf')
       !dir$ ignore_tkr (d) r
diff --git a/flang/test/Lower/CUDA/cuda-libdevice.cuf b/flang/test/Lower/CUDA/cuda-libdevice.cuf
index ae5fdd3b50678..7bb2ab065316a 100644
--- a/flang/test/Lower/CUDA/cuda-libdevice.cuf
+++ b/flang/test/Lower/CUDA/cuda-libdevice.cuf
@@ -245,6 +245,15 @@ end subroutine
 ! CHECK: %{{.*}} = fir.call @__nv_float2uint_ru(%{{.*}}) proc_attrs<bind_c> fastmath<contract> : (f32) -> i32
 ! CHECK: %{{.*}} = fir.call @__nv_float2uint_rz(%{{.*}}) proc_attrs<bind_c> fastmath<contract> : (f32) -> i32
 
+attributes(global) subroutine test_fdividef()
+  real :: res
+  real :: r
+  res = __fdividef(r, r)
+end subroutine
+
+! CHECK-LABEL: _QPtest_fdividef
+! CHECK: %{{.*}} = fir.call @__nv_fast_fdividef(%{{.*}}, %{{.*}}) proc_attrs<bind_c> fastmath<contract> : (f32, f32) -> f32
+
 attributes(global) subroutine test_double_as_longlong()
   integer(8) :: res
   real(8) :: r