1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
|
// SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
// SPDX-FileCopyrightText: Copyright Contributors to the Kokkos project
#include <Kokkos_Macros.hpp>
#ifdef KOKKOS_ENABLE_EXPERIMENTAL_CXX20_MODULES
import kokkos.core;
#else
#include <Kokkos_Core.hpp>
#endif
#include <TestCuda_Category.hpp>
#include <array>
namespace Test {
__global__ void offset(int* p) {
int idx = blockIdx.x * blockDim.x + threadIdx.x;
if (idx < 100) {
p[idx] += idx;
}
}
// Test whether allocations survive Kokkos initialize/finalize if done via Raw
// Cuda.
TEST(cuda, raw_cuda_interop) {
// Make sure that we use the same device for all allocations
Kokkos::initialize();
int* p;
KOKKOS_IMPL_CUDA_SAFE_CALL(cudaMalloc(&p, sizeof(int) * 100));
Kokkos::View<int*, Kokkos::MemoryTraits<Kokkos::Unmanaged>> v(p, 100);
Kokkos::deep_copy(v, 5);
Kokkos::finalize();
offset<<<100, 64>>>(p);
KOKKOS_IMPL_CUDA_SAFE_CALL(cudaDeviceSynchronize());
std::array<int, 100> h_p;
cudaMemcpy(h_p.data(), p, sizeof(int) * 100, cudaMemcpyDefault);
KOKKOS_IMPL_CUDA_SAFE_CALL(cudaDeviceSynchronize());
int64_t sum = 0;
int64_t sum_expect = 0;
for (int i = 0; i < 100; i++) {
sum += h_p[i];
sum_expect += 5 + i;
}
ASSERT_EQ(sum, sum_expect);
KOKKOS_IMPL_CUDA_SAFE_CALL(cudaFree(p));
}
} // namespace Test
|
