diff --git a/flang/test/Lower/CUDA/cuda-mod.cuf b/flang/test/Lower/CUDA/cuda-mod.cuf
index 2cc6439789a27..ae5bf63d2da49 100644
--- a/flang/test/Lower/CUDA/cuda-mod.cuf
+++ b/flang/test/Lower/CUDA/cuda-mod.cuf
@@ -4,6 +4,12 @@
 
 module cuf_mod
   real, device :: md
+
+contains
+  attributes(device) subroutine devsub()
+  end
 end module
 
 ! CHECK: fir.global @_QMcuf_modEmd {cuda_attr = #fir.cuda<device>} : f32
+
+! CHECK: func.func @_QMcuf_modPdevsub() attributes {fir.cuda_attr = #fir.cuda_proc<device>}
diff --git a/flang/test/Lower/CUDA/cuda-module-use.cuf b/flang/test/Lower/CUDA/cuda-module-use.cuf
index 43ab0f6ff8d68..f54083b026ee8 100644
--- a/flang/test/Lower/CUDA/cuda-module-use.cuf
+++ b/flang/test/Lower/CUDA/cuda-module-use.cuf
@@ -1,7 +1,7 @@
 ! RUN: bbc -emit-hlfir -fcuda %S/cuda-mod.cuf
 ! RUN: bbc -emit-hlfir -fcuda %s -o - | FileCheck %s
 
-! Test importing module with variable with CUDA attributes.
+! Test importing module containing variable and subroutine with CUDA attributes.
 
 subroutine sub1()
   use cuf_mod
@@ -12,4 +12,14 @@ end
 ! CHECK: %[[ADDR:.*]] = fir.address_of(@_QMcuf_modEmd) : !fir.ref<f32>
 ! CHECK: %{{.*}}:2 = hlfir.declare %[[ADDR]] {cuda_attr = #fir.cuda<device>, uniq_name = "_QMcuf_modEmd"} : (!fir.ref<f32>) -> (!fir.ref<f32>, !fir.ref<f32>)
 
-! CHECK: fir.global @_QMcuf_modEmd {cuda_attr = #fir.cuda<device>} : f32
+attributes(device) subroutine sub2()
+  use cuf_mod
+  call devsub()
+end
+
+! CHECK-LABEL: func.func @_QPsub2() attributes {fir.cuda_attr = #fir.cuda_proc<device>}
+! CHECK: fir.call @_QMcuf_modPdevsub()
+
+! CHECK-LABEL: fir.global @_QMcuf_modEmd {cuda_attr = #fir.cuda<device>} : f32
+
+! CHECK-LABEL: func.func private @_QMcuf_modPdevsub() attributes {fir.cuda_attr = #fir.cuda_proc<device>}