PmeGpuKernelParams Struct Reference

#include <gromacs/ewald/pme_gpu_types.h>

Collaboration diagram for PmeGpuKernelParams:

Description

A single structure encompassing all the PME data used in GPU kernels on device. To extend the list with platform-specific parameters, this can be inherited by the GPU framework-specific structure.

Public Attributes
struct PmeGpuConstParams	constants
	Constant data that is set once.
struct PmeGpuGridParams	grid
	Data dependent on the grid size/cutoff.
struct PmeGpuAtomParams	atoms
	Data dependent on the DD and local atoms.
struct PmeGpuDynamicParams	current
	Data that possibly changes for every new PME computation. This should be kept up-to-date by calling pme_gpu_prepare_computation(...) before launching spreading.
char	usePipeline
	Whether pipelining with PP communications is active char rather than bool to avoid problem with OpenCL compiler.
int	pipelineAtomStart
	Start atom for this stage of pipeline.
int	pipelineAtomEnd
	End atom for this stage of pipeline.
int	ppRanksInfoSize
	PpRanksSendFInfo struct total size.
DeviceBuffer< PpRanksSendFInfo >	ppRanksInfo
	PpRanksSendFInfo struct containing each PP rank forces buffer offsets.
DeviceBuffer< unsigned int >	lastProcessedBlockPerPpRank
	atomic counter used for tracking last processed block in pme gather kernel for each PP rank
DeviceBuffer< uint64_t >	forcesReadyNvshmemFlags
	sync object used for NVSHMEM pme-pp force comm
uint64_t	forcesReadyNvshmemFlagsCounter
	sync object counter used for nvshmem pme-pp force comm
int	isVirialStep
	whether this is virial step
int	useNvshmem
	whether to use NVSHMEM for GPU comm
DeviceTexture	fractShiftsTableTexture
	Texture object for accessing grid.d_fractShiftsTable.
DeviceTexture	gridlineIndicesTableTexture
	Texture object for accessing grid.d_gridlineIndicesTable.

---

The documentation for this struct was generated from the following file:

• src/gromacs/ewald/pme_gpu_types.h