| 1 | // RUN: %libomptarget-compile-generic |
| 2 | // RUN: env HSA_XNACK=1 \ |
| 3 | // RUN: %libomptarget-run-generic | %fcheck-generic |
| 4 | // XFAIL: nvptx64-nvidia-cuda |
| 5 | // XFAIL: nvptx64-nvidia-cuda-LTO |
| 6 | |
| 7 | // REQUIRES: unified_shared_memory |
| 8 | |
| 9 | #include <omp.h> |
| 10 | #include <stdio.h> |
| 11 | |
| 12 | // End of definitions copied from OpenMP RTL. |
| 13 | // --------------------------------------------------------------------------- |
| 14 | |
| 15 | #pragma omp requires unified_shared_memory |
| 16 | |
| 17 | #define N 1024 |
| 18 | |
| 19 | void init(int A[], int B[], int C[]) { |
| 20 | for (int i = 0; i < N; ++i) { |
| 21 | A[i] = 0; |
| 22 | B[i] = 1; |
| 23 | C[i] = i; |
| 24 | } |
| 25 | } |
| 26 | |
| 27 | int main(int argc, char *argv[]) { |
| 28 | const int device = omp_get_default_device(); |
| 29 | |
| 30 | // CHECK: Initial device: [[INITIAL_DEVICE:[0-9]+]] |
| 31 | printf(format: "Initial device: %d\n" , omp_get_initial_device()); |
| 32 | // CHECK: Num devices: [[INITIAL_DEVICE]] |
| 33 | printf(format: "Num devices: %d\n" , omp_get_num_devices()); |
| 34 | |
| 35 | // |
| 36 | // Target alloc & target memcpy |
| 37 | // |
| 38 | int A[N], B[N], C[N]; |
| 39 | |
| 40 | // Init |
| 41 | init(A, B, C); |
| 42 | |
| 43 | int *pA, *pB, *pC; |
| 44 | |
| 45 | // map ptrs |
| 46 | pA = &A[0]; |
| 47 | pB = &B[0]; |
| 48 | pC = &C[0]; |
| 49 | |
| 50 | int *d_A = (int *)omp_target_alloc(N * sizeof(int), device); |
| 51 | int *d_B = (int *)omp_target_alloc(N * sizeof(int), device); |
| 52 | int *d_C = (int *)omp_target_alloc(N * sizeof(int), device); |
| 53 | |
| 54 | // CHECK: omp_target_alloc succeeded |
| 55 | printf(format: "omp_target_alloc %s\n" , d_A && d_B && d_C ? "succeeded" : "failed" ); |
| 56 | |
| 57 | omp_target_memcpy(d_B, pB, N * sizeof(int), 0, 0, device, |
| 58 | omp_get_initial_device()); |
| 59 | omp_target_memcpy(d_C, pC, N * sizeof(int), 0, 0, device, |
| 60 | omp_get_initial_device()); |
| 61 | |
| 62 | #pragma omp target is_device_ptr(d_A, d_B, d_C) device(device) |
| 63 | { |
| 64 | #pragma omp parallel for schedule(static, 1) |
| 65 | for (int i = 0; i < N; i++) { |
| 66 | d_A[i] = d_B[i] + d_C[i] + 1; |
| 67 | } |
| 68 | } |
| 69 | |
| 70 | omp_target_memcpy(pA, d_A, N * sizeof(int), 0, 0, omp_get_initial_device(), |
| 71 | device); |
| 72 | |
| 73 | // CHECK: Test omp_target_memcpy: Succeeded |
| 74 | int fail = 0; |
| 75 | for (int i = 0; i < N; ++i) { |
| 76 | if (A[i] != i + 2) |
| 77 | fail++; |
| 78 | } |
| 79 | if (fail) { |
| 80 | printf(format: "Test omp_target_memcpy: Failed\n" ); |
| 81 | } else { |
| 82 | printf(format: "Test omp_target_memcpy: Succeeded\n" ); |
| 83 | } |
| 84 | |
| 85 | // |
| 86 | // target_is_present and target_associate/disassociate_ptr |
| 87 | // |
| 88 | init(A, B, C); |
| 89 | |
| 90 | // CHECK: B is not present, associating it... |
| 91 | // CHECK: omp_target_associate_ptr B succeeded |
| 92 | if (!omp_target_is_present(B, device)) { |
| 93 | printf(format: "B is not present, associating it...\n" ); |
| 94 | int rc = omp_target_associate_ptr(B, d_B, N * sizeof(int), 0, device); |
| 95 | printf(format: "omp_target_associate_ptr B %s\n" , !rc ? "succeeded" : "failed" ); |
| 96 | } |
| 97 | |
| 98 | // CHECK: C is not present, associating it... |
| 99 | // CHECK: omp_target_associate_ptr C succeeded |
| 100 | if (!omp_target_is_present(C, device)) { |
| 101 | printf(format: "C is not present, associating it...\n" ); |
| 102 | int rc = omp_target_associate_ptr(C, d_C, N * sizeof(int), 0, device); |
| 103 | printf(format: "omp_target_associate_ptr C %s\n" , !rc ? "succeeded" : "failed" ); |
| 104 | } |
| 105 | |
| 106 | // CHECK: Inside target data: A is not present |
| 107 | // CHECK: Inside target data: B is present |
| 108 | // CHECK: Inside target data: C is present |
| 109 | #pragma omp target data map(from : B, C) device(device) |
| 110 | { |
| 111 | printf(format: "Inside target data: A is%s present\n" , |
| 112 | omp_target_is_present(A, device) ? "" : " not" ); |
| 113 | printf(format: "Inside target data: B is%s present\n" , |
| 114 | omp_target_is_present(B, device) ? "" : " not" ); |
| 115 | printf(format: "Inside target data: C is%s present\n" , |
| 116 | omp_target_is_present(C, device) ? "" : " not" ); |
| 117 | |
| 118 | #pragma omp target map(from : A) device(device) |
| 119 | { |
| 120 | #pragma omp parallel for schedule(static, 1) |
| 121 | for (int i = 0; i < N; i++) |
| 122 | A[i] = B[i] + C[i] + 1; |
| 123 | } |
| 124 | } |
| 125 | |
| 126 | // CHECK: B is present, disassociating it... |
| 127 | // CHECK: omp_target_disassociate_ptr B succeeded |
| 128 | // CHECK: C is present, disassociating it... |
| 129 | // CHECK: omp_target_disassociate_ptr C succeeded |
| 130 | if (omp_target_is_present(B, device)) { |
| 131 | printf(format: "B is present, disassociating it...\n" ); |
| 132 | int rc = omp_target_disassociate_ptr(B, device); |
| 133 | printf(format: "omp_target_disassociate_ptr B %s\n" , !rc ? "succeeded" : "failed" ); |
| 134 | } |
| 135 | if (omp_target_is_present(C, device)) { |
| 136 | printf(format: "C is present, disassociating it...\n" ); |
| 137 | int rc = omp_target_disassociate_ptr(C, device); |
| 138 | printf(format: "omp_target_disassociate_ptr C %s\n" , !rc ? "succeeded" : "failed" ); |
| 139 | } |
| 140 | |
| 141 | // CHECK: Test omp_target_associate_ptr: Succeeded |
| 142 | fail = 0; |
| 143 | for (int i = 0; i < N; ++i) { |
| 144 | if (A[i] != i + 2) |
| 145 | fail++; |
| 146 | } |
| 147 | if (fail) { |
| 148 | printf(format: "Test omp_target_associate_ptr: Failed\n" ); |
| 149 | } else { |
| 150 | printf(format: "Test omp_target_associate_ptr: Succeeded\n" ); |
| 151 | } |
| 152 | |
| 153 | omp_target_free(d_A, device); |
| 154 | omp_target_free(d_B, device); |
| 155 | omp_target_free(d_C, device); |
| 156 | |
| 157 | printf(format: "Done!\n" ); |
| 158 | |
| 159 | return 0; |
| 160 | } |
| 161 | |