pme_gpu_program_impl_ocl.cpp File Reference

#include "gmxpre.h"
#include "gromacs/gpu_utils/gmxopencl.h"
#include "gromacs/gpu_utils/ocl_compiler.h"
#include "gromacs/utility/stringutil.h"
#include "pme_gpu_constants.h"
#include "pme_gpu_internal.h"
#include "pme_gpu_program_impl.h"
#include "pme_gpu_types_host.h"
#include "pme_grid.h"

Include dependency graph for pme_gpu_program_impl_ocl.cpp:

Description

Implements PmeGpuProgramImpl, which stores permanent PME GPU context-derived data, such as (compiled) kernel handles.

Author

Aleksei Iupinov a.yup.nosp@m.inov.nosp@m.@gmai.nosp@m.l.co.nosp@m.m

Functions
static void	checkRequiredWarpSize (cl_kernel kernel, const char *kernelName, const DeviceInformation &deviceInfo)
	Ensure that spread/gather kernels have been compiled to a suitable warp size. More...

Function Documentation

static void checkRequiredWarpSize ( cl_kernel  kernel, const char *  kernelName, const DeviceInformation &  deviceInfo  )

static

Ensure that spread/gather kernels have been compiled to a suitable warp size.

On Intel the exec width/warp is decided at compile-time and can be smaller than the minimum order^2 required in spread/gather ATM which we need to check for.

Due to the one thread per atom and order=4 implementation constraints, order^2 threads should execute without synchronization needed.