Make sure that CUDA instances are destroyed prior to destroying streams

aprokop · Apr 1, 2024 · b19cb1c · b19cb1c
1 parent b4f6a16
commit b19cb1c
Show file tree

Hide file tree

Showing 2 changed files with 22 additions and 15 deletions.
diff --git a/benchmarks/execution_space_instances/execution_space_instances_driver.cpp b/benchmarks/execution_space_instances/execution_space_instances_driver.cpp
@@ -59,6 +59,9 @@ class InstanceManager<Kokkos::Cuda>
 
   ~InstanceManager()
   {
+    // Destroy CUDA instances prior to destroying streams
+    _instances.clear();
+
     for (auto &stream : _streams)
       cudaStreamDestroy(stream);
   }

diff --git a/examples/access_traits/example_cuda_access_traits.cpp b/examples/access_traits/example_cuda_access_traits.cpp
@@ -77,21 +77,25 @@ int main(int argc, char *argv[])
   cudaStreamCreate(&stream);
   cudaMemcpyAsync(d_a, a.data(), sizeof(a), cudaMemcpyHostToDevice, stream);
 
-  Kokkos::Cuda cuda{stream};
-  ArborX::BVH<Kokkos::CudaSpace> bvh{cuda, PointCloud{d_a, d_a, d_a, N}};
-
-  Kokkos::View<int *, Kokkos::CudaSpace> indices("Example::indices", 0);
-  Kokkos::View<int *, Kokkos::CudaSpace> offset("Example::offset", 0);
-  ArborX::query(bvh, cuda, Spheres{d_a, d_a, d_a, d_a, N}, indices, offset);
-
-  Kokkos::parallel_for(
-      "Example::print_indices", Kokkos::RangePolicy<Kokkos::Cuda>(cuda, 0, N),
-      KOKKOS_LAMBDA(int i) {
-        for (int j = offset(i); j < offset(i + 1); ++j)
-        {
-          printf("%i %i\n", i, indices(j));
-        }
-      });
+  {
+    // Place the code using CUDA instance in a block to make sure the instance
+    // is destroyed prior to destroying the stream.
+    Kokkos::Cuda cuda{stream};
+    ArborX::BVH<Kokkos::CudaSpace> bvh{cuda, PointCloud{d_a, d_a, d_a, N}};
+
+    Kokkos::View<int *, Kokkos::CudaSpace> indices("Example::indices", 0);
+    Kokkos::View<int *, Kokkos::CudaSpace> offset("Example::offset", 0);
+    ArborX::query(bvh, cuda, Spheres{d_a, d_a, d_a, d_a, N}, indices, offset);
+
+    Kokkos::parallel_for(
+        "Example::print_indices", Kokkos::RangePolicy<Kokkos::Cuda>(cuda, 0, N),
+        KOKKOS_LAMBDA(int i) {
+          for (int j = offset(i); j < offset(i + 1); ++j)
+          {
+            printf("%i %i\n", i, indices(j));
+          }
+        });
+  }
 
   cudaStreamDestroy(stream);