1 // Make sure that __global__ functions are emitted along with correct
2 // annotations and are added to @llvm.used to prevent their elimination.
3 // REQUIRES: nvptx-registered-target
4 //
5 // RUN: %clang_cc1 %s -triple nvptx-unknown-unknown -fcuda-is-device -emit-llvm -o - | FileCheck %s
6 
7 #include "Inputs/cuda.h"
8 
9 // CHECK-LABEL: define void @device_function
10 extern "C"
11 __device__ void device_function() {}
12 
13 // CHECK-LABEL: define void @global_function
14 extern "C"
15 __global__ void global_function() {
16   // CHECK: call void @device_function
17   device_function();
18 }
19 
20 // Make sure host-instantiated kernels are preserved on device side.
21 template <typename T> __global__ void templated_kernel(T param) {}
22 // CHECK-DAG: define void @_Z16templated_kernelIiEvT_(
23 
24 namespace {
25 __global__ void anonymous_ns_kernel() {}
26 // CHECK-DAG: define void @_ZN12_GLOBAL__N_119anonymous_ns_kernelEv(
27 }
28 
29 void host_function() {
30   templated_kernel<<<0, 0>>>(0);
31   anonymous_ns_kernel<<<0,0>>>();
32 }
33 
34 // CHECK: !{{[0-9]+}} = !{void ()* @global_function, !"kernel", i32 1}
35 // CHECK: !{{[0-9]+}} = !{void (i32)* @_Z16templated_kernelIiEvT_, !"kernel", i32 1}
36