Skip to content

Fix: fix a deadlock bug in CUDA version of gint #5210

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Merged
merged 1 commit into from
Oct 10, 2024
Merged

Fix: fix a deadlock bug in CUDA version of gint #5210

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
1 commit
Select commit
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
8 changes: 2 additions & 6 deletions source/module_hamilt_lcao/module_gint/gint_force_gpu.cu
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -31,11 +31,7 @@ void gint_fvl_gpu(const hamilt::HContainer<double>* dm,
const Grid_Technique& gridt,
const UnitCell& ucell)
{
#ifdef __MPI
const int dev_id = base_device::information::set_device_by_rank();
#else
const int dev_id = 0;
#endif
checkCuda(cudaSetDevice(gridt.dev_id));
// checkCuda(cudaSetDeviceFlags(cudaDeviceScheduleBlockingSync));

const int nbzp = gridt.nbzp;
Expand Down Expand Up @@ -99,7 +95,7 @@ void gint_fvl_gpu(const hamilt::HContainer<double>* dm,
{
// 20240620 Note that it must be set again here because
// cuda's device is not safe in a multi-threaded environment.
checkCuda(cudaSetDevice(dev_id));
checkCuda(cudaSetDevice(gridt.dev_id));
const int sid = omp_get_thread_num();

int max_m = 0;
Expand Down
8 changes: 2 additions & 6 deletions source/module_hamilt_lcao/module_gint/gint_rho_gpu.cu
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -17,11 +17,7 @@ void gint_rho_gpu(const hamilt::HContainer<double>* dm,
const UnitCell& ucell,
double* rho)
{
#ifdef __MPI
const int dev_id = base_device::information::set_device_by_rank();
#else
const int dev_id = 0;
#endif
checkCuda(cudaSetDevice(gridt.dev_id));
// checkCuda(cudaSetDeviceFlags(cudaDeviceScheduleBlockingSync));

const int nbzp = gridt.nbzp;
Expand Down Expand Up @@ -80,7 +76,7 @@ void gint_rho_gpu(const hamilt::HContainer<double>* dm,
// 20240620 Note that it must be set again here because
// cuda's device is not safe in a multi-threaded environment.

checkCuda(cudaSetDevice(dev_id));
checkCuda(cudaSetDevice(gridt.dev_id));
// get stream id
const int sid = omp_get_thread_num();

Expand Down
8 changes: 2 additions & 6 deletions source/module_hamilt_lcao/module_gint/gint_vl_gpu.cu
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -29,11 +29,7 @@ void gint_vl_gpu(hamilt::HContainer<double>* hRGint,
double* pvpR,
const bool is_gamma_only)
{
#ifdef __MPI
const int dev_id = base_device::information::set_device_by_rank();
#else
const int dev_id = 0;
#endif
checkCuda(cudaSetDevice(gridt.dev_id));
// checkCuda(cudaSetDeviceFlags(cudaDeviceScheduleBlockingSync));
const int nbzp = gridt.nbzp;
const int num_streams = gridt.nstreams;
Expand Down Expand Up @@ -81,7 +77,7 @@ void gint_vl_gpu(hamilt::HContainer<double>* hRGint,
{
// 20240620 Note that it must be set again here because
// cuda's device is not safe in a multi-threaded environment.
checkCuda(cudaSetDevice(dev_id));
checkCuda(cudaSetDevice(gridt.dev_id));
const int sid = omp_get_thread_num();

int max_m = 0;
Expand Down
2 changes: 1 addition & 1 deletion source/module_hamilt_lcao/module_gint/grid_technique.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -562,7 +562,7 @@ int Grid_Technique::find_offset(const int id1, const int id2, const int iat1, co
void Grid_Technique::init_gpu_gint_variables(const UnitCell& ucell,
const int num_stream) {
#ifdef __MPI
base_device::information::set_device_by_rank();
dev_id = base_device::information::set_device_by_rank();
#endif
if (is_malloced) {
free_gpu_gint_variables(this->nat);
Expand Down
1 change: 1 addition & 0 deletions source/module_hamilt_lcao/module_gint/grid_technique.h
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -177,6 +177,7 @@ class Grid_Technique : public Grid_MeshBall {
double* rcut_g;
double*mcell_pos_g;

int dev_id = 0;
int nstreams = 4;
// streams[nstreams]
// TODO it needs to be implemented through configuration files
Expand Down
Loading