#include "config/config.hpp"
#include "utils.hpp"
#include <cstddef>
#include <vector>

Include dependency graph for init.hpp:

This graph shows which files directly or indirectly include this file:

Classes
struct	EspressoGpuDevice
	Struct to hold information relevant to ESPResSo about GPUs. More...

Functions
void	cuda_init ()
	Initializes the CUDA stream.

int	cuda_get_n_gpus ()
	Get the number of CUDA devices.

int	cuda_check_gpu_compute_capability (int dev)
	Check that a given GPU has compute capability.

void	cuda_get_gpu_name (int dev, char *name)
	Get the name of a CUDA device.

void	cuda_set_device (int dev)
	Choose a device for future CUDA computations.

int	cuda_get_device ()
	Get the current CUDA device.

int	cuda_test_device_access ()
	Test if actual CUDA device works.

void	cuda_check_device ()
	Check that a device is available, that its compute capability is sufficient for ESPResSo, and that data can be written to and read from it.

std::vector< EspressoGpuDevice >	cuda_gather_gpus ()
	Gather unique list of CUDA devices on all nodes.

EspressoGpuDevice	cuda_get_device_props (int dev)
	Get properties of a CUDA device.

void	cuda_on_program_start ()
	Called on program start.

Function Documentation

◆ cuda_check_device()

void cuda_check_device ( )

Check that a device is available, that its compute capability is sufficient for ESPResSo, and that data can be written to and read from it.

Otherwise, throw an exception.

Definition at line 130 of file init_cuda.cu.

References cuda_check_gpu_compute_capability(), cuda_get_device(), cuda_get_n_gpus(), cuda_test_device_access(), and ES_OK.

◆ cuda_check_gpu_compute_capability()

int cuda_check_gpu_compute_capability ( int dev )

Check that a given GPU has compute capability.

The minimal compute capability required by ESPResSo is computeCapabilityMinMajor . computeCapabilityMinMinor .

Parameters

dev	CUDA device number

Returns: ES_OK if the GPU meets the requirements, else ES_ERROR.

Definition at line 50 of file init_cuda.cu.

References computeCapabilityMinMajor, computeCapabilityMinMinor, CUDA_CHECK, ES_ERROR, and ES_OK.

Referenced by cuda_check_device().

◆ cuda_gather_gpus()

std::vector< EspressoGpuDevice > cuda_gather_gpus ( )

Gather unique list of CUDA devices on all nodes.

Returns: vector of device properties.

Gather unique list of CUDA devices on all nodes.

It relies on MPI_Get_processor_name() to get a unique identifier of the physical node, as opposed to the logical rank of which there can be more than one per node.

Definition at line 52 of file init.cpp.

References communicator, cuda_get_device_props(), cuda_get_n_gpus(), invoke_skip_cuda_exceptions(), Communicator::size, and this_node.

Referenced by ScriptInterface::System::CudaInitHandle::do_call_method().