gmx::NbnxmKernelBenchOptions Struct Reference

#include <gromacs/nbnxm/benchmark/bench_setup.h>

Description

The options for the kernel benchmarks.

Public Attributes
bool	useGpu = false
	Whether to use a GPU, currently GPUs are not supported.
int	numThreads = 1
	The number of OpenMP threads to use.
NbnxmBenchMarkKernels	nbnxmSimd = NbnxmBenchMarkKernels::SimdAuto
	The SIMD type for the kernel.
NbnxmBenchMarkCombRule	ljCombinationRule = NbnxmBenchMarkCombRule::RuleGeom
	The LJ combination rule.
NbnxmBenchMarkInteractionModifiers	interactionModifier = NbnxmBenchMarkInteractionModifiers::PotShift
	The Coulomb / VdW interaction modifier.
bool	useHalfLJOptimization = false
	Use i-cluster half-LJ optimization for clusters with <= half LJ.
real	pairlistCutoff = 1.0
	The pairlist and interaction cut-off.
real	ewaldcoeff_q = 0
	The Coulomb Ewald coefficient.
bool	computeVirialAndEnergy = false
	Whether to compute energies (shift forces for virial are always computed on CPU)
NbnxmBenchMarkCoulomb	coulombType = NbnxmBenchMarkCoulomb::Pme
	The Coulomb interaction function.
bool	useTabulatedEwaldCorr = false
	Whether to use tabulated PME grid correction instead of analytical, not applicable with simd=no.
bool	doAll = false
	Whether to run all combinations of Coulomb type, combination rule and SIMD.
int	numPreIterations = 1
	Number of iterations to run before running each kernel benchmark, currently always 1.
int	numIterations = 100
	The number of iterations for each kernel.
int	numWarmupIterations = 0
	The number of (untimed) iterations to run at startup to warm up the CPU.
bool	cyclesPerPair = false
	Print cycles/pair instead of pairs/cycle.
bool	reportTime = false
	Report in micro seconds instead of cycles.
std::string	outputFile
	Also report into a csv file.

---

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

• src/gromacs/nbnxm/benchmark/bench_setup.h