| 1 | // RUN: %clang_cc1 -verify -fopenmp -triple x86_64-apple-darwin10.6.0 -fopenmp-targets=nvptx64-nvidia-cuda -emit-llvm-bc -o %t-host.bc %s |
|---|
| 2 | // RUN: %clang_cc1 -verify -fopenmp -triple nvptx64-nvidia-cuda -fopenmp-targets=nvptx64-nvidia-cuda -emit-llvm %s -fopenmp-is-device -fopenmp-host-ir-file-path %t-host.bc -o - | FileCheck %s |
|---|
| 3 | // expected-no-diagnostics |
|---|
| 4 | |
|---|
| 5 | #ifndef HEADER |
|---|
| 6 | #define HEADER |
|---|
| 7 | |
|---|
| 8 | #pragma omp declare target |
|---|
| 9 | typedef void **omp_allocator_handle_t; |
|---|
| 10 | extern const omp_allocator_handle_t omp_default_mem_alloc; |
|---|
| 11 | extern const omp_allocator_handle_t omp_large_cap_mem_alloc; |
|---|
| 12 | extern const omp_allocator_handle_t omp_const_mem_alloc; |
|---|
| 13 | extern const omp_allocator_handle_t omp_high_bw_mem_alloc; |
|---|
| 14 | extern const omp_allocator_handle_t omp_low_lat_mem_alloc; |
|---|
| 15 | extern const omp_allocator_handle_t omp_cgroup_mem_alloc; |
|---|
| 16 | extern const omp_allocator_handle_t omp_pteam_mem_alloc; |
|---|
| 17 | extern const omp_allocator_handle_t omp_thread_mem_alloc; |
|---|
| 18 | |
|---|
| 19 | // CHECK-DAG: @{{.+}}St1{{.+}}b{{.+}} = external global i32, |
|---|
| 20 | // CHECK-DAG: @a = global i32 0, |
|---|
| 21 | // CHECK-DAG: @b = addrspace(4) global i32 0, |
|---|
| 22 | // CHECK-DAG: @c = global i32 0, |
|---|
| 23 | // CHECK-DAG: @d = global %struct.St1 zeroinitializer, |
|---|
| 24 | // CHECK-DAG: @{{.+}}ns{{.+}}a{{.+}} = addrspace(3) global i32 0, |
|---|
| 25 | // CHECK-DAG: @{{.+}}main{{.+}}a{{.*}} = internal global i32 0, |
|---|
| 26 | // CHECK-DAG: @{{.+}}ST{{.+}}m{{.+}} = external global i32, |
|---|
| 27 | struct St{ |
|---|
| 28 | int a; |
|---|
| 29 | }; |
|---|
| 30 | |
|---|
| 31 | struct St1{ |
|---|
| 32 | int a; |
|---|
| 33 | static int b; |
|---|
| 34 | #pragma omp allocate(b) allocator(omp_default_mem_alloc) |
|---|
| 35 | } d; |
|---|
| 36 | |
|---|
| 37 | int a, b, c; |
|---|
| 38 | #pragma omp allocate(a) allocator(omp_large_cap_mem_alloc) |
|---|
| 39 | #pragma omp allocate(b) allocator(omp_const_mem_alloc) |
|---|
| 40 | #pragma omp allocate(d, c) allocator(omp_high_bw_mem_alloc) |
|---|
| 41 | |
|---|
| 42 | template <class T> |
|---|
| 43 | struct ST { |
|---|
| 44 | static T m; |
|---|
| 45 | #pragma omp allocate(m) allocator(omp_low_lat_mem_alloc) |
|---|
| 46 | }; |
|---|
| 47 | |
|---|
| 48 | template <class T> T foo() { |
|---|
| 49 | T v; |
|---|
| 50 | #pragma omp allocate(v) allocator(omp_cgroup_mem_alloc) |
|---|
| 51 | v = ST<T>::m; |
|---|
| 52 | return v; |
|---|
| 53 | } |
|---|
| 54 | |
|---|
| 55 | namespace ns{ |
|---|
| 56 | int a; |
|---|
| 57 | } |
|---|
| 58 | #pragma omp allocate(ns::a) allocator(omp_pteam_mem_alloc) |
|---|
| 59 | |
|---|
| 60 | // CHECK-LABEL: @main |
|---|
| 61 | int main () { |
|---|
| 62 | // CHECK: alloca double, |
|---|
| 63 | static int a; |
|---|
| 64 | #pragma omp allocate(a) allocator(omp_thread_mem_alloc) |
|---|
| 65 | a=2; |
|---|
| 66 | double b = 3; |
|---|
| 67 | #pragma omp allocate(b) allocator(omp_default_mem_alloc) |
|---|
| 68 | return (foo<int>()); |
|---|
| 69 | } |
|---|
| 70 | |
|---|
| 71 | // CHECK: define {{.*}}i32 @{{.+}}foo{{.+}}() |
|---|
| 72 | // CHECK: alloca i32, |
|---|
| 73 | |
|---|
| 74 | extern template int ST<int>::m; |
|---|
| 75 | #pragma omp end declare target |
|---|
| 76 | #endif |
|---|
| 77 | |
|---|