cwpearson/astaroth
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

MPI: corners are now transferred by default because 1) with those Astaroth works with any symmetric stencil and 2) corners have very small impact on performance. Also disabled resetting the device s.t. one could potentially assign many subgrids to one GPU (f.ex. in AMR)

Browse Source
This commit is contained in:
jpekkila
2020-08-19 12:05:20 +03:00
parent 7f7b0b89ea
commit a85b8b8cd1
1 changed files with 2 additions and 2 deletions
Download Patch File Download Diff File Expand all files Collapse all files

4
src/core/device.cc
View File

@@ -14,7 +14,7 @@
#define MPI_DECOMPOSITION_AXES (3) #define MPI_DECOMPOSITION_AXES (3)
#define MPI_COMPUTE_ENABLED (1) #define MPI_COMPUTE_ENABLED (1)
#define MPI_COMM_ENABLED (1) #define MPI_COMM_ENABLED (1)
#define MPI_INCL_CORNERS (0) #define MPI_INCL_CORNERS (1)
#define MPI_USE_PINNED (0) // Do inter-node comm with pinned memory #define MPI_USE_PINNED (0) // Do inter-node comm with pinned memory
#define MPI_USE_CUDA_DRIVER_PINNING (0) // Pin with cuPointerSetAttribute, otherwise cudaMallocHost #define MPI_USE_CUDA_DRIVER_PINNING (0) // Pin with cuPointerSetAttribute, otherwise cudaMallocHost
@@ -114,7 +114,7 @@ AcResult
acDeviceCreate(const int id, const AcMeshInfo device_config, Device* device_handle) acDeviceCreate(const int id, const AcMeshInfo device_config, Device* device_handle)
{ {
cudaSetDevice(id); cudaSetDevice(id);
cudaDeviceReset(); // Would be good for safety, but messes stuff up if we want to emulate // cudaDeviceReset(); // Would be good for safety, but messes stuff up if we want to emulate
// multiple devices with a single GPU // multiple devices with a single GPU
// Create Device // Create Device