gmx::GpuConfigurationCapabilities Struct Reference

#include <gromacs/gpu_utils/capabilities.h>

Description

Collection of GPU capabilities for this configuration.

Static Public Attributes
static constexpr bool	Nonbonded = GMX_GPU
	Whether this configuration supports running NBNXM kernels on the device.
static constexpr bool	NonbondedFE = Nonbonded && (GMX_GPU_CUDA || GMX_GPU_HIP)
	Whether this configuration supports running NB free energy kernels on the device.
static constexpr bool	BufferOps = GMX_GPU && !GMX_GPU_OPENCL
	Whether this configuration supports running buffer operation kernels on the device.
static constexpr bool	Pme = GMX_GPU
	Whether this configuration supports running PME kernels on the device.
static constexpr bool	StreamQuery = GMX_GPU_CUDA || GMX_GPU_HIP
	Whether this configuration supports querying device stream for completion of all submitted tasks.
static constexpr bool	PmeParamLookupTable = GMX_GPU_CUDA || GMX_GPU_HIP
	Whether this configuration supports using the param lookup table.
static constexpr bool	PmePipelining = GMX_GPU && !GMX_GPU_OPENCL
	Whether this configuration supports PME pipelining.
static constexpr bool	PmeSupportsThreadsPerAtomOrder = GMX_GPU && !GMX_GPU_OPENCL
	Whether this configuration supports PME ThreadsPerAtomOrder.
static constexpr bool	PmeDynamicMaxGridSize = GMX_GPU_CUDA || GMX_GPU_HIP
	Whether this configuration supports PME max gridsize setting.
static constexpr bool	PmeSolveNeedsAtLeastFourWarps = GMX_GPU && !GMX_GPU_OPENCL
	Whether this configuration supports PME solve kernels with less than 4 warps.
static constexpr bool	Fft
	Whether this configuration supports running FFT kernels on the device. More...
static constexpr bool	Bonded = GMX_GPU && !GMX_GPU_OPENCL
	Whether this configuration supports running bonded kernels on the device.
static constexpr bool	UpdateLeapfrog = GMX_GPU && !GMX_GPU_OPENCL
	Whether this configuration supports running update with leapfrog on the device.
static constexpr bool	UpdateSD = false
	Whether this configuration supports running update with stochastic dynamics on the device.
static constexpr bool	Constraints = GMX_GPU && !GMX_GPU_OPENCL
	Whether this configuration supports running the constraint kernels (LINCS+SETTLE) on the device.
static constexpr bool	ThreadMpiDirectComm = GMX_GPU_CUDA || GMX_GPU_HIP
	Whether this configuration supports running the direct GPU communication path with thread-MPI.
static constexpr bool	LibraryMpiDirectComm = GMX_GPU && !GMX_GPU_OPENCL
	Whether this configuration supports running the direct GPU communication path with library-MPI.
static constexpr bool	MpiDirectComm
	Whether this configuration supports running the direct GPU communication path for the current build type. More...
static constexpr bool	DisableEventCounting = GMX_GPU_CUDA || GMX_GPU_HIP
	Whether we disable event counting.
static constexpr bool	PpPmeDirectComm = MpiDirectComm
	This flag enables the direct GPU communication path for PP-PME communication.
static constexpr bool	HaloExchangeDirectComm = MpiDirectComm
	This flag enables the direct GPU communication path for domain decomposition halo exchange.
static constexpr bool	GpuGraph = GMX_HAVE_GPU_GRAPH_SUPPORT
	Whether this configuration supports running gpu graphs.
static constexpr bool	PmeDecomposition
	Whether this configuration supports running gpu pme decomposition. More...
static constexpr bool	TwoDPmeDecomposition = PmeDecomposition && GMX_GPU_CUDA
static constexpr bool	Threefry = GMX_GPU && !GMX_GPU_OPENCL
	Whether we support running the threefry random number generator on the device.

Member Data Documentation

constexpr bool gmx::GpuConfigurationCapabilities::Fft

static

Initial value:

= GMX_GPU
                                && (GMX_GPU_FFT_MKL || GMX_GPU_FFT_ROCFFT || GMX_GPU_FFT_BBFFT || GMX_GPU_FFT_ONEMATH
                                    || GMX_GPU_FFT_VKFFT || GMX_GPU_FFT_CUFFT || GMX_GPU_FFT_CLFFT)

Whether this configuration supports running FFT kernels on the device.

constexpr bool gmx::GpuConfigurationCapabilities::MpiDirectComm

static

Initial value:

=
            (GMX_THREAD_MPI && ThreadMpiDirectComm) || (GMX_LIB_MPI && LibraryMpiDirectComm)

Whether this configuration supports running the direct GPU communication path for the current build type.

constexpr bool gmx::GpuConfigurationCapabilities::PmeDecomposition

static

Initial value:

= (GMX_GPU_CUDA && (GMX_USE_Heffte || GMX_USE_cuFFTMp))
                                             || (GMX_GPU_SYCL && GMX_USE_Heffte)
                                             || (GMX_GPU_HIP && GMX_USE_Heffte)

Whether this configuration supports running gpu pme decomposition.

---

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

• src/gromacs/gpu_utils/capabilities.h