Skip to content

Commit

Permalink
use dynamic cast in cuda/hip getters from executors
Browse files Browse the repository at this point in the history
  • Loading branch information
@youwuyou
youwuyou committed Jun 6, 2024
1 parent d992285 commit 03d72ab
Show file tree
Hide file tree
Showing 2 changed files with 123 additions and 59 deletions.
166 changes: 115 additions & 51 deletions core/c_api.cpp
View file
Original file line number Diff line number Diff line change
Expand Up @@ -121,56 +121,120 @@ size_t ginkgo_executor_cpu_get_num_threads_per_core(gko_executor exec_st_ptr)

//---------------------------- GPU -----------------------------
// Get the number of multiprocessor of this executor.
// size_t ginkgo_executor_gpu_get_num_multiprocessor(gko_executor exec_st_ptr)
// {
// return exec_st_ptr->shared_ptr->get_num_multiprocessor();
// }
size_t ginkgo_executor_gpu_get_num_multiprocessor(gko_executor exec_st_ptr)
{
if (auto cuda_exec = std::dynamic_pointer_cast<gko::CudaExecutor>(
exec_st_ptr->shared_ptr)) {
return cuda_exec->get_num_multiprocessor();
} else if (auto hip_exec = std::dynamic_pointer_cast<gko::HipExecutor>(
exec_st_ptr->shared_ptr)) {
return hip_exec->get_num_multiprocessor();
} else {
return 0;
}
}

// // Get the device id of the device associated to this executor.
// size_t ginkgo_executor_gpu_get_device_id(gko_executor exec_st_ptr)
// {
// return exec_st_ptr->shared_ptr->get_device_id();
// }
// Get the device id of the device associated to this executor.
size_t ginkgo_executor_gpu_get_device_id(gko_executor exec_st_ptr)
{
if (auto cuda_exec = std::dynamic_pointer_cast<gko::CudaExecutor>(
exec_st_ptr->shared_ptr)) {
return cuda_exec->get_device_id();
} else if (auto hip_exec = std::dynamic_pointer_cast<gko::HipExecutor>(
exec_st_ptr->shared_ptr)) {
return hip_exec->get_device_id();
} else {
return 0;
}
}

// // Get the number of warps per SM of this executor.
// size_t ginkgo_executor_gpu_get_num_warps_per_sm(gko_executor exec_st_ptr)
// {
// return exec_st_ptr->shared_ptr->get_num_warps_per_sm();
// }
// Get the number of warps per SM of this executor.
size_t ginkgo_executor_gpu_get_num_warps_per_sm(gko_executor exec_st_ptr)
{
// return exec_st_ptr->shared_ptr->get_num_warps_per_sm();
if (auto cuda_exec = std::dynamic_pointer_cast<gko::CudaExecutor>(
exec_st_ptr->shared_ptr)) {
return cuda_exec->get_device_id();
} else if (auto hip_exec = std::dynamic_pointer_cast<gko::HipExecutor>(
exec_st_ptr->shared_ptr)) {
return hip_exec->get_device_id();
} else {
return 0;
}
}

// // Get the number of warps of this executor.
// size_t ginkgo_executor_gpu_get_num_warps(gko_executor exec_st_ptr)
// {
// return exec_st_ptr->shared_ptr->get_num_warps();
// }
// Get the number of warps of this executor.
size_t ginkgo_executor_gpu_get_num_warps(gko_executor exec_st_ptr)
{
if (auto cuda_exec = std::dynamic_pointer_cast<gko::CudaExecutor>(
exec_st_ptr->shared_ptr)) {
return cuda_exec->get_num_warps();
} else if (auto hip_exec = std::dynamic_pointer_cast<gko::HipExecutor>(
exec_st_ptr->shared_ptr)) {
return hip_exec->get_num_warps();
} else {
return 0;
}
}

// // Get the warp size of this executor.
// size_t ginkgo_executor_gpu_get_warp_size(gko_executor exec_st_ptr)
// {
// return exec_st_ptr->shared_ptr->get_warp_size();
// }
// Get the warp size of this executor.
size_t ginkgo_executor_gpu_get_warp_size(gko_executor exec_st_ptr)
{
if (auto cuda_exec = std::dynamic_pointer_cast<gko::CudaExecutor>(
exec_st_ptr->shared_ptr)) {
return cuda_exec->get_warp_size();
} else if (auto hip_exec = std::dynamic_pointer_cast<gko::HipExecutor>(
exec_st_ptr->shared_ptr)) {
return hip_exec->get_warp_size();
} else {
return 0;
}
}

// // Get the major version of compute capability.
// size_t ginkgo_executor_gpu_get_major_version(gko_executor exec_st_ptr)
// {
// return exec_st_ptr->shared_ptr->get_major_version();
// }
// Get the major version of compute capability.
size_t ginkgo_executor_gpu_get_major_version(gko_executor exec_st_ptr)
{
if (auto cuda_exec = std::dynamic_pointer_cast<gko::CudaExecutor>(
exec_st_ptr->shared_ptr)) {
return cuda_exec->get_major_version();
} else if (auto hip_exec = std::dynamic_pointer_cast<gko::HipExecutor>(
exec_st_ptr->shared_ptr)) {
return hip_exec->get_major_version();
} else {
return 0;
}
}

// // Get the minor version of compute capability.
// size_t ginkgo_executor_gpu_get_minor_version(gko_executor exec_st_ptr)
// {
// return exec_st_ptr->shared_ptr->get_minor_version();
// }
// Get the minor version of compute capability.
size_t ginkgo_executor_gpu_get_minor_version(gko_executor exec_st_ptr)
{
if (auto cuda_exec = std::dynamic_pointer_cast<gko::CudaExecutor>(
exec_st_ptr->shared_ptr)) {
return cuda_exec->get_minor_version();
} else if (auto hip_exec = std::dynamic_pointer_cast<gko::HipExecutor>(
exec_st_ptr->shared_ptr)) {
return hip_exec->get_minor_version();
} else {
return 0;
}
}

// // Get the closest NUMA node.
// size_t ginkgo_executor_gpu_get_closest_numa(gko_executor exec_st_ptr)
// {
// return exec_st_ptr->shared_ptr->get_closest_numa();
// }
// Get the closest NUMA node.
size_t ginkgo_executor_gpu_get_closest_numa(gko_executor exec_st_ptr)
{
if (auto cuda_exec = std::dynamic_pointer_cast<gko::CudaExecutor>(
exec_st_ptr->shared_ptr)) {
return cuda_exec->get_closest_numa();
} else if (auto hip_exec = std::dynamic_pointer_cast<gko::HipExecutor>(
exec_st_ptr->shared_ptr)) {
return hip_exec->get_closest_numa();
} else {
return 0;
}
}

//
// int* ginkgo_executor_get_closest_pus(gko_executor exec_st_ptr){
// return exec_st_ptr->shared_ptr->get_closest_pus();
// // return exec_st_ptr->shared_ptr->get_closest_pus();
// }

// CUDA
Expand Down Expand Up @@ -201,17 +265,17 @@ size_t ginkgo_executor_hip_get_num_devices()
return gko::HipExecutor::get_num_devices();
}

// DPCPP/SYCL
gko_executor ginkgo_executor_dpcpp_create(size_t device_id)
{
return new gko_executor_st{gko::DpcppExecutor::create(
device_id, gko::ReferenceExecutor::create())};
}
// // DPCPP/SYCL
// gko_executor ginkgo_executor_dpcpp_create(size_t device_id)
// {
// return new gko_executor_st{gko::DpcppExecutor::create(
// device_id, gko::ReferenceExecutor::create())};
// }

size_t ginkgo_executor_dpcpp_get_num_devices()
{
return gko::DpcppExecutor::get_num_devices("gpu");
}
// size_t ginkgo_executor_dpcpp_get_num_devices()
// {
// return gko::DpcppExecutor::get_num_devices("gpu");
// }

/* ----------------------------------------------------------------------
* Library functions for creating arrays and array operations in GINKGO
Expand Down
16 changes: 8 additions & 8 deletions include/ginkgo/c_api.h
View file
Original file line number Diff line number Diff line change
Expand Up @@ -513,14 +513,14 @@ bool ginkgo_executor_memory_accessible(gko_executor exec_st_ptr,
void ginkgo_executor_synchronize(gko_executor exec_st_ptr);
size_t ginkgo_executor_cpu_get_num_cores(gko_executor exec_st_ptr);
size_t ginkgo_executor_cpu_get_num_threads_per_core(gko_executor exec_st_ptr);
// size_t ginkgo_executor_gpu_get_num_multiprocessor(gko_executor exec_st_ptr);
// size_t ginkgo_executor_gpu_get_device_id(gko_executor exec_st_ptr);
// size_t ginkgo_executor_gpu_get_num_warps_per_sm(gko_executor exec_st_ptr);
// size_t ginkgo_executor_gpu_get_num_warps(gko_executor exec_st_ptr);
// size_t ginkgo_executor_gpu_get_warp_size(gko_executor exec_st_ptr);
// size_t ginkgo_executor_gpu_get_major_version(gko_executor exec_st_ptr);
// size_t ginkgo_executor_gpu_get_minor_version(gko_executor exec_st_ptr);
// size_t ginkgo_executor_gpu_get_closest_numa(gko_executor exec_st_ptr);
size_t ginkgo_executor_gpu_get_num_multiprocessor(gko_executor exec_st_ptr);
size_t ginkgo_executor_gpu_get_device_id(gko_executor exec_st_ptr);
size_t ginkgo_executor_gpu_get_num_warps_per_sm(gko_executor exec_st_ptr);
size_t ginkgo_executor_gpu_get_num_warps(gko_executor exec_st_ptr);
size_t ginkgo_executor_gpu_get_warp_size(gko_executor exec_st_ptr);
size_t ginkgo_executor_gpu_get_major_version(gko_executor exec_st_ptr);
size_t ginkgo_executor_gpu_get_minor_version(gko_executor exec_st_ptr);
size_t ginkgo_executor_gpu_get_closest_numa(gko_executor exec_st_ptr);


/* ----------------------------------------------------------------------
Expand Down

0 comments on commit 03d72ab

Please sign in to comment.