Home
last modified time | relevance | path

Searched refs:pmeGpu (Results 1 – 11 of 11) sorted by relevance

/dports/science/gromacs/gromacs-2021.4/src/gromacs/ewald/
H A Dpme_gpu_internal.cpp140 pmeGpu->common->ngrids == 1 || pmeGpu->common->ngrids == 2, in pme_gpu_alloc_energy_virial()
173 pmeGpu->common->ngrids == 1 || pmeGpu->common->ngrids == 2, in pme_gpu_realloc_and_copy_bspline_values()
225 pmeGpu->staging.h_forces.reserveWithPadding(pmeGpu->nAtomsAlloc); in pme_gpu_realloc_forces()
671 pmeGpu->common->ngrids == 1 || pmeGpu->common->ngrids == 2, in pme_gpu_reinit_grids()
887 delete pmeGpu; in pme_gpu_destroy()
1172 pmeGpu->common->ngrids == 1 || pmeGpu->common->ngrids == 2, in pme_gpu_spread()
1306 pmeGpu->common->ngrids == 1 || pmeGpu->common->ngrids == 2, in pme_gpu_solve()
1504 pmeGpu->common->ngrids == 1 || pmeGpu->common->ngrids == 2, in pme_gpu_gather()
1598 if (pmeGpu && pmeGpu->kernelParams) in pme_gpu_get_kernelparam_forces()
1610 GMX_ASSERT(pmeGpu && pmeGpu->kernelParams, in pme_gpu_set_kernelparam_coordinates()
[all …]
H A Dpme_gpu_timings.cpp53 bool pme_gpu_timings_enabled(const PmeGpu* pmeGpu) in pme_gpu_timings_enabled() argument
55 return pmeGpu->archSpecific->useTiming; in pme_gpu_timings_enabled()
60 if (pme_gpu_timings_enabled(pmeGpu)) in pme_gpu_start_timing()
64 … pmeGpu->archSpecific->timingEvents[PMEStageId].openTimingRegion(pmeGpu->archSpecific->pmeStream_); in pme_gpu_start_timing()
70 if (pme_gpu_timings_enabled(pmeGpu)) in pme_gpu_stop_timing()
74 …pmeGpu->archSpecific->timingEvents[PMEStageId].closeTimingRegion(pmeGpu->archSpecific->pmeStream_); in pme_gpu_stop_timing()
80 if (pme_gpu_timings_enabled(pmeGpu)) in pme_gpu_get_timings()
93 if (pme_gpu_timings_enabled(pmeGpu)) in pme_gpu_update_timings()
104 if (pme_gpu_timings_enabled(pmeGpu)) in pme_gpu_reinit_timings()
126 void pme_gpu_reset_timings(const PmeGpu* pmeGpu) in pme_gpu_reset_timings() argument
[all …]
H A Dpme_gpu_internal.h132 void pme_gpu_free_energy_virial(PmeGpu* pmeGpu);
163 void pme_gpu_realloc_forces(PmeGpu* pmeGpu);
170 void pme_gpu_free_forces(const PmeGpu* pmeGpu);
178 void pme_gpu_copy_input_forces(PmeGpu* pmeGpu);
185 void pme_gpu_copy_output_forces(PmeGpu* pmeGpu);
252 void pme_gpu_realloc_grids(PmeGpu* pmeGpu);
259 void pme_gpu_free_grids(const PmeGpu* pmeGpu);
267 void pme_gpu_clear_grids(const PmeGpu* pmeGpu);
426 return pmeGpu->settings; in pme_gpu_settings()
437 return pmeGpu->staging; in pme_gpu_staging()
[all …]
H A Dpme_gpu_timings.h57 void pme_gpu_start_timing(const PmeGpu* pmeGpu, size_t PMEStageId);
65 void pme_gpu_stop_timing(const PmeGpu* pmeGpu, size_t PMEStageId);
73 bool pme_gpu_timings_enabled(const PmeGpu* pmeGpu);
80 void pme_gpu_update_timings(const PmeGpu* pmeGpu);
87 void pme_gpu_reinit_timings(const PmeGpu* pmeGpu);
94 void pme_gpu_reset_timings(const PmeGpu* pmeGpu);
102 void pme_gpu_get_timings(const PmeGpu* pmeGpu, gmx_wallclock_gpu_pme_t* timings);
H A Dpme_gpu.cpp159 PmeGpu* pmeGpu = pme->gpu; in pme_gpu_prepare_computation() local
161 pmeGpu->settings.useGpuForceReduction = stepWork.useGpuPmeFReduction; in pme_gpu_prepare_computation()
168 shouldUpdateBox |= (pmeGpu->common->previousBox[i][j] != box[i][j]); in pme_gpu_prepare_computation()
169 pmeGpu->common->previousBox[i][j] = box[i][j]; in pme_gpu_prepare_computation()
177 pme_gpu_update_input_box(pmeGpu, box); in pme_gpu_prepare_computation()
181 if (!pme_gpu_settings(pmeGpu).performGPUSolve) in pme_gpu_prepare_computation()
185 pmeGpu->common->boxScaler->scaleBox(box, scaledBox); in pme_gpu_prepare_computation()
202 PmeGpu* pmeGpu = pme->gpu; in pme_gpu_launch_spread() local
204 GMX_ASSERT(pmeGpu->common->ngrids == 1 || (pmeGpu->common->ngrids == 2 && pme->bFEP_q), in pme_gpu_launch_spread()
225 PmeGpu* pmeGpu = pme->gpu; in pme_gpu_launch_complex_transforms() local
[all …]
H A Dpme_gpu_3dfft_ocl.cpp66 GpuParallel3dFft::GpuParallel3dFft(const PmeGpu* pmeGpu, const int gridIndex) in GpuParallel3dFft() argument
71 GMX_RELEASE_ASSERT(!pme_gpu_settings(pmeGpu).useDecomposition, in GpuParallel3dFft()
73 PmeGpuKernelParamsBase* kernelParamsPtr = pmeGpu->kernelParams.get(); in GpuParallel3dFft()
83 cl_context context = pmeGpu->archSpecific->deviceContext_.context(); in GpuParallel3dFft()
84 deviceStreams_.push_back(pmeGpu->archSpecific->pmeStream_.stream()); in GpuParallel3dFft()
87 const bool performOutOfPlaceFFT = pmeGpu->archSpecific->performOutOfPlaceFFT; in GpuParallel3dFft()
H A Dpme_gpu_3dfft.cu63 GpuParallel3dFft::GpuParallel3dFft(const PmeGpu* pmeGpu, const int gridIndex) in GpuParallel3dFft() argument
65 const PmeGpuCudaKernelParams* kernelParamsPtr = pmeGpu->kernelParams.get(); in GpuParallel3dFft()
74 GMX_RELEASE_ASSERT(!pme_gpu_settings(pmeGpu).useDecomposition, in GpuParallel3dFft()
107 cudaStream_t stream = pmeGpu->archSpecific->pmeStream_.stream(); in GpuParallel3dFft()
H A Dpme_gpu_3dfft.h78 GpuParallel3dFft(const PmeGpu* pmeGpu, int gridIndex);
H A Dpme.h150 PmeGpu* pmeGpu,
H A Dpme.cpp580 PmeGpu* pmeGpu, in gmx_pme_init() argument
801 pme->gpu = pmeGpu; /* Carrying over the single GPU structure */ in gmx_pme_init()
/dports/science/gromacs/gromacs-2021.4/src/gromacs/ewald/tests/
H A Dpmetestcommon.cpp534 static int pme_gpu_get_atoms_per_warp(const PmeGpu* pmeGpu) in pme_gpu_get_atoms_per_warp() argument
536 const int order = pmeGpu->common->pme_order; in pme_gpu_get_atoms_per_warp()
538 (pmeGpu->settings.threadsPerAtom == ThreadsPerAtom::Order ? order : order * order); in pme_gpu_get_atoms_per_warp()
539 return pmeGpu->programHandle_->warpSize() / threadsPerAtom; in pme_gpu_get_atoms_per_warp()
551 static void pme_gpu_transform_spline_atom_data(const PmeGpu* pmeGpu, in pme_gpu_transform_spline_atom_data() argument
565 const auto atomsPerWarp = pme_gpu_get_atoms_per_warp(pmeGpu); in pme_gpu_transform_spline_atom_data()
566 const auto pmeOrder = pmeGpu->common->pme_order; in pme_gpu_transform_spline_atom_data()
575 h_splineBuffer = pmeGpu->staging.h_theta; in pme_gpu_transform_spline_atom_data()
580 h_splineBuffer = pmeGpu->staging.h_dtheta; in pme_gpu_transform_spline_atom_data()