#include "nbnxm_gpu_data_mgmt.h"
+#include "gromacs/gpu_utils/device_stream_manager.h"
+#include "gromacs/gpu_utils/gputraits.h"
+#include "gromacs/gpu_utils/pmalloc.h"
#include "gromacs/hardware/device_information.h"
#include "gromacs/mdtypes/interaction_const.h"
+#include "gromacs/mdtypes/simulation_workload.h"
#include "gromacs/nbnxm/gpu_common_utils.h"
#include "gromacs/nbnxm/gpu_data_mgmt.h"
+#include "gromacs/nbnxm/gridset.h"
+#include "gromacs/pbcutil/ishift.h"
#include "gromacs/timing/gpu_timing.h"
+#include "gromacs/pbcutil/ishift.h"
#include "gromacs/utility/cstringutil.h"
+#include "gromacs/utility/exceptions.h"
#include "gromacs/utility/fatalerror.h"
#include "nbnxm_gpu.h"
namespace Nbnxm
{
-void init_ewald_coulomb_force_table(const EwaldCorrectionTables& tables,
- NBParamGpu* nbp,
- const DeviceContext& deviceContext)
+static inline void issueClFlushInStream(const DeviceStream& deviceStream)
+{
+#if GMX_GPU_OPENCL
+ /* Based on the v1.2 section 5.13 of the OpenCL spec, a flush is needed
+ * in the stream after marking an event in it in order to be able to sync with
+ * the event from another stream.
+ */
+ cl_int cl_error = clFlush(deviceStream.stream());
+ if (cl_error != CL_SUCCESS)
+ {
+ GMX_THROW(gmx::InternalError("clFlush failed: " + ocl_get_error_string(cl_error)));
+ }
+#else
+ GMX_UNUSED_VALUE(deviceStream);
+#endif
+}
+
+static inline void init_ewald_coulomb_force_table(const EwaldCorrectionTables& tables,
+ NBParamGpu* nbp,
+ const DeviceContext& deviceContext)
{
if (nbp->coulomb_tab)
{
- destroyParamLookupTable(&nbp->coulomb_tab, nbp->coulomb_tab_texobj);
+ destroyParamLookupTable(&nbp->coulomb_tab, &nbp->coulomb_tab_texobj);
}
nbp->coulomb_tab_scale = tables.scale;
&nbp->coulomb_tab, &nbp->coulomb_tab_texobj, tables.tableF.data(), tables.tableF.size(), deviceContext);
}
-enum ElecType nbnxn_gpu_pick_ewald_kernel_type(const interaction_const_t& ic,
- const DeviceInformation gmx_unused& deviceInfo)
+static inline ElecType nbnxn_gpu_pick_ewald_kernel_type(const interaction_const_t& ic,
+ const DeviceInformation gmx_unused& deviceInfo)
{
bool bTwinCut = (ic.rcoulomb != ic.rvdw);
}
}
-void set_cutoff_parameters(NBParamGpu* nbp, const interaction_const_t* ic, const PairlistParams& listParams)
+static inline void set_cutoff_parameters(NBParamGpu* nbp,
+ const interaction_const_t& ic,
+ const PairlistParams& listParams)
{
- nbp->ewald_beta = ic->ewaldcoeff_q;
- nbp->sh_ewald = ic->sh_ewald;
- nbp->epsfac = ic->epsfac;
- nbp->two_k_rf = 2.0 * ic->reactionFieldCoefficient;
- nbp->c_rf = ic->reactionFieldShift;
- nbp->rvdw_sq = ic->rvdw * ic->rvdw;
- nbp->rcoulomb_sq = ic->rcoulomb * ic->rcoulomb;
+ nbp->ewald_beta = ic.ewaldcoeff_q;
+ nbp->sh_ewald = ic.sh_ewald;
+ nbp->epsfac = ic.epsfac;
+ nbp->two_k_rf = 2.0 * ic.reactionFieldCoefficient;
+ nbp->c_rf = ic.reactionFieldShift;
+ nbp->rvdw_sq = ic.rvdw * ic.rvdw;
+ nbp->rcoulomb_sq = ic.rcoulomb * ic.rcoulomb;
nbp->rlistOuter_sq = listParams.rlistOuter * listParams.rlistOuter;
nbp->rlistInner_sq = listParams.rlistInner * listParams.rlistInner;
nbp->useDynamicPruning = listParams.useDynamicPruning;
- nbp->sh_lj_ewald = ic->sh_lj_ewald;
- nbp->ewaldcoeff_lj = ic->ewaldcoeff_lj;
+ nbp->sh_lj_ewald = ic.sh_lj_ewald;
+ nbp->ewaldcoeff_lj = ic.ewaldcoeff_lj;
- nbp->rvdw_switch = ic->rvdw_switch;
- nbp->dispersion_shift = ic->dispersion_shift;
- nbp->repulsion_shift = ic->repulsion_shift;
- nbp->vdw_switch = ic->vdw_switch;
+ nbp->rvdw_switch = ic.rvdw_switch;
+ nbp->dispersion_shift = ic.dispersion_shift;
+ nbp->repulsion_shift = ic.repulsion_shift;
+ nbp->vdw_switch = ic.vdw_switch;
}
-void gpu_pme_loadbal_update_param(const nonbonded_verlet_t* nbv, const interaction_const_t* ic)
-{
- if (!nbv || !nbv->useGpu())
- {
- return;
- }
- NbnxmGpu* nb = nbv->gpu_nbv;
- NBParamGpu* nbp = nb->nbparam;
-
- set_cutoff_parameters(nbp, ic, nbv->pairlistSets().params());
-
- nbp->elecType = nbnxn_gpu_pick_ewald_kernel_type(*ic, nb->deviceContext_->deviceInfo());
-
- GMX_RELEASE_ASSERT(ic->coulombEwaldTables, "Need valid Coulomb Ewald correction tables");
- init_ewald_coulomb_force_table(*ic->coulombEwaldTables, nbp, *nb->deviceContext_);
-}
-
-void init_plist(gpu_plist* pl)
+static inline void init_plist(gpu_plist* pl)
{
/* initialize to nullptr pointers to data that is not allocated here and will
need reallocation in nbnxn_gpu_init_pairlist */
pl->rollingPruningPart = 0;
}
-void init_timings(gmx_wallclock_gpu_nbnxn_t* t)
+static inline void init_timings(gmx_wallclock_gpu_nbnxn_t* t)
{
t->nb_h2d_t = 0.0;
t->nb_d2h_t = 0.0;
t->dynamicPruneTime.t = 0.0;
}
+/*! \brief Initialize \p atomdata first time; it only gets filled at pair-search. */
+static inline void initAtomdataFirst(NBAtomDataGpu* atomdata,
+ int numTypes,
+ const DeviceContext& deviceContext,
+ const DeviceStream& localStream)
+{
+ atomdata->numTypes = numTypes;
+ allocateDeviceBuffer(&atomdata->shiftVec, gmx::c_numShiftVectors, deviceContext);
+ atomdata->shiftVecUploaded = false;
+
+ allocateDeviceBuffer(&atomdata->fShift, gmx::c_numShiftVectors, deviceContext);
+ allocateDeviceBuffer(&atomdata->eLJ, 1, deviceContext);
+ allocateDeviceBuffer(&atomdata->eElec, 1, deviceContext);
+
+ clearDeviceBufferAsync(&atomdata->fShift, 0, gmx::c_numShiftVectors, localStream);
+ clearDeviceBufferAsync(&atomdata->eElec, 0, 1, localStream);
+ clearDeviceBufferAsync(&atomdata->eLJ, 0, 1, localStream);
+
+ /* initialize to nullptr pointers to data that is not allocated here and will
+ need reallocation in later */
+ atomdata->xq = nullptr;
+ atomdata->f = nullptr;
+
+ /* size -1 indicates that the respective array hasn't been initialized yet */
+ atomdata->numAtoms = -1;
+ atomdata->numAtomsAlloc = -1;
+}
+
+static inline VdwType nbnxmGpuPickVdwKernelType(const interaction_const_t& ic,
+ LJCombinationRule ljCombinationRule)
+{
+ if (ic.vdwtype == VanDerWaalsType::Cut)
+ {
+ switch (ic.vdw_modifier)
+ {
+ case InteractionModifiers::None:
+ case InteractionModifiers::PotShift:
+ switch (ljCombinationRule)
+ {
+ case LJCombinationRule::None: return VdwType::Cut;
+ case LJCombinationRule::Geometric: return VdwType::CutCombGeom;
+ case LJCombinationRule::LorentzBerthelot: return VdwType::CutCombLB;
+ default:
+ GMX_THROW(gmx::InconsistentInputError(gmx::formatString(
+ "The requested LJ combination rule %s is not implemented in "
+ "the GPU accelerated kernels!",
+ enumValueToString(ljCombinationRule))));
+ }
+ case InteractionModifiers::ForceSwitch: return VdwType::FSwitch;
+ case InteractionModifiers::PotSwitch: return VdwType::PSwitch;
+ default:
+ GMX_THROW(gmx::InconsistentInputError(
+ gmx::formatString("The requested VdW interaction modifier %s is not "
+ "implemented in the GPU accelerated kernels!",
+ enumValueToString(ic.vdw_modifier))));
+ }
+ }
+ else if (ic.vdwtype == VanDerWaalsType::Pme)
+ {
+ if (ic.ljpme_comb_rule == LongRangeVdW::Geom)
+ {
+ GMX_RELEASE_ASSERT(
+ ljCombinationRule == LJCombinationRule::Geometric,
+ "Combination rules for long- and short-range interactions should match.");
+ return VdwType::EwaldGeom;
+ }
+ else
+ {
+ GMX_RELEASE_ASSERT(
+ ljCombinationRule == LJCombinationRule::LorentzBerthelot,
+ "Combination rules for long- and short-range interactions should match.");
+ return VdwType::EwaldLB;
+ }
+ }
+ else
+ {
+ GMX_THROW(gmx::InconsistentInputError(gmx::formatString(
+ "The requested VdW type %s is not implemented in the GPU accelerated kernels!",
+ enumValueToString(ic.vdwtype))));
+ }
+}
+
+static inline ElecType nbnxmGpuPickElectrostaticsKernelType(const interaction_const_t& ic,
+ const DeviceInformation& deviceInfo)
+{
+ if (ic.eeltype == CoulombInteractionType::Cut)
+ {
+ return ElecType::Cut;
+ }
+ else if (EEL_RF(ic.eeltype))
+ {
+ return ElecType::RF;
+ }
+ else if ((EEL_PME(ic.eeltype) || ic.eeltype == CoulombInteractionType::Ewald))
+ {
+ return nbnxn_gpu_pick_ewald_kernel_type(ic, deviceInfo);
+ }
+ else
+ {
+ /* Shouldn't happen, as this is checked when choosing Verlet-scheme */
+ GMX_THROW(gmx::InconsistentInputError(
+ gmx::formatString("The requested electrostatics type %s is not implemented in "
+ "the GPU accelerated kernels!",
+ enumValueToString(ic.eeltype))));
+ }
+}
+
+/*! \brief Initialize the nonbonded parameter data structure. */
+static inline void initNbparam(NBParamGpu* nbp,
+ const interaction_const_t& ic,
+ const PairlistParams& listParams,
+ const nbnxn_atomdata_t::Params& nbatParams,
+ const DeviceContext& deviceContext)
+{
+ const int numTypes = nbatParams.numTypes;
+
+ set_cutoff_parameters(nbp, ic, listParams);
+
+ nbp->vdwType = nbnxmGpuPickVdwKernelType(ic, nbatParams.ljCombinationRule);
+ nbp->elecType = nbnxmGpuPickElectrostaticsKernelType(ic, deviceContext.deviceInfo());
+
+ if (ic.vdwtype == VanDerWaalsType::Pme)
+ {
+ if (ic.ljpme_comb_rule == LongRangeVdW::Geom)
+ {
+ GMX_ASSERT(nbatParams.ljCombinationRule == LJCombinationRule::Geometric,
+ "Combination rule mismatch!");
+ }
+ else
+ {
+ GMX_ASSERT(nbatParams.ljCombinationRule == LJCombinationRule::LorentzBerthelot,
+ "Combination rule mismatch!");
+ }
+ }
+
+ /* generate table for PME */
+ if (nbp->elecType == ElecType::EwaldTab || nbp->elecType == ElecType::EwaldTabTwin)
+ {
+ GMX_RELEASE_ASSERT(ic.coulombEwaldTables, "Need valid Coulomb Ewald correction tables");
+ init_ewald_coulomb_force_table(*ic.coulombEwaldTables, nbp, deviceContext);
+ }
+
+ /* set up LJ parameter lookup table */
+ if (!useLjCombRule(nbp->vdwType))
+ {
+ static_assert(sizeof(decltype(nbp->nbfp)) == 2 * sizeof(decltype(*nbatParams.nbfp.data())),
+ "Mismatch in the size of host / device data types");
+ initParamLookupTable(&nbp->nbfp,
+ &nbp->nbfp_texobj,
+ reinterpret_cast<const Float2*>(nbatParams.nbfp.data()),
+ numTypes * numTypes,
+ deviceContext);
+ }
+
+ /* set up LJ-PME parameter lookup table */
+ if (ic.vdwtype == VanDerWaalsType::Pme)
+ {
+ static_assert(sizeof(decltype(nbp->nbfp_comb))
+ == 2 * sizeof(decltype(*nbatParams.nbfp_comb.data())),
+ "Mismatch in the size of host / device data types");
+ initParamLookupTable(&nbp->nbfp_comb,
+ &nbp->nbfp_comb_texobj,
+ reinterpret_cast<const Float2*>(nbatParams.nbfp_comb.data()),
+ numTypes,
+ deviceContext);
+ }
+}
+
+NbnxmGpu* gpu_init(const gmx::DeviceStreamManager& deviceStreamManager,
+ const interaction_const_t* ic,
+ const PairlistParams& listParams,
+ const nbnxn_atomdata_t* nbat,
+ const bool bLocalAndNonlocal)
+{
+ auto* nb = new NbnxmGpu();
+ nb->deviceContext_ = &deviceStreamManager.context();
+ nb->atdat = new NBAtomDataGpu;
+ nb->nbparam = new NBParamGpu;
+ nb->plist[InteractionLocality::Local] = new Nbnxm::gpu_plist;
+ if (bLocalAndNonlocal)
+ {
+ nb->plist[InteractionLocality::NonLocal] = new Nbnxm::gpu_plist;
+ }
+
+ nb->bUseTwoStreams = bLocalAndNonlocal;
+
+ nb->timers = new Nbnxm::GpuTimers();
+ snew(nb->timings, 1);
+
+ nb->bDoTime = decideGpuTimingsUsage();
+
+ if (nb->bDoTime)
+ {
+ init_timings(nb->timings);
+ }
+
+ /* init nbst */
+ pmalloc(reinterpret_cast<void**>(&nb->nbst.eLJ), sizeof(*nb->nbst.eLJ));
+ pmalloc(reinterpret_cast<void**>(&nb->nbst.eElec), sizeof(*nb->nbst.eElec));
+ pmalloc(reinterpret_cast<void**>(&nb->nbst.fShift), gmx::c_numShiftVectors * sizeof(*nb->nbst.fShift));
+
+ init_plist(nb->plist[InteractionLocality::Local]);
+
+ /* local/non-local GPU streams */
+ GMX_RELEASE_ASSERT(deviceStreamManager.streamIsValid(gmx::DeviceStreamType::NonBondedLocal),
+ "Local non-bonded stream should be initialized to use GPU for non-bonded.");
+ const DeviceStream& localStream = deviceStreamManager.stream(gmx::DeviceStreamType::NonBondedLocal);
+ nb->deviceStreams[InteractionLocality::Local] = &localStream;
+ // In general, it's not strictly necessary to use 2 streams for SYCL, since they are
+ // out-of-order. But for the time being, it will be less disruptive to keep them.
+ if (nb->bUseTwoStreams)
+ {
+ init_plist(nb->plist[InteractionLocality::NonLocal]);
+
+ GMX_RELEASE_ASSERT(deviceStreamManager.streamIsValid(gmx::DeviceStreamType::NonBondedNonLocal),
+ "Non-local non-bonded stream should be initialized to use GPU for "
+ "non-bonded with domain decomposition.");
+ nb->deviceStreams[InteractionLocality::NonLocal] =
+ &deviceStreamManager.stream(gmx::DeviceStreamType::NonBondedNonLocal);
+ }
+
+ const nbnxn_atomdata_t::Params& nbatParams = nbat->params();
+ const DeviceContext& deviceContext = *nb->deviceContext_;
+
+ initNbparam(nb->nbparam, *ic, listParams, nbatParams, deviceContext);
+ initAtomdataFirst(nb->atdat, nbatParams.numTypes, deviceContext, localStream);
+
+ gpu_init_platform_specific(nb);
+
+ if (debug)
+ {
+ fprintf(debug, "Initialized NBNXM GPU data structures.\n");
+ }
+
+ return nb;
+}
+
+void gpu_pme_loadbal_update_param(const nonbonded_verlet_t* nbv, const interaction_const_t& ic)
+{
+ if (!nbv || !nbv->useGpu())
+ {
+ return;
+ }
+ NbnxmGpu* nb = nbv->gpu_nbv;
+ NBParamGpu* nbp = nb->nbparam;
+
+ set_cutoff_parameters(nbp, ic, nbv->pairlistSets().params());
+
+ nbp->elecType = nbnxn_gpu_pick_ewald_kernel_type(ic, nb->deviceContext_->deviceInfo());
+
+ GMX_RELEASE_ASSERT(ic.coulombEwaldTables, "Need valid Coulomb Ewald correction tables");
+ init_ewald_coulomb_force_table(*ic.coulombEwaldTables, nbp, *nb->deviceContext_);
+}
+
+void gpu_upload_shiftvec(NbnxmGpu* nb, const nbnxn_atomdata_t* nbatom)
+{
+ NBAtomDataGpu* adat = nb->atdat;
+ const DeviceStream& localStream = *nb->deviceStreams[InteractionLocality::Local];
+
+ /* only if we have a dynamic box */
+ if (nbatom->bDynamicBox || !adat->shiftVecUploaded)
+ {
+ copyToDeviceBuffer(&adat->shiftVec,
+ gmx::asGenericFloat3Pointer(nbatom->shift_vec),
+ 0,
+ gmx::c_numShiftVectors,
+ localStream,
+ GpuApiCallBehavior::Async,
+ nullptr);
+ adat->shiftVecUploaded = true;
+ }
+}
+
//! This function is documented in the header file
void gpu_init_pairlist(NbnxmGpu* nb, const NbnxnPairlistGpu* h_plist, const InteractionLocality iloc)
{
d_plist->haveFreshList = true;
}
+void gpu_init_atomdata(NbnxmGpu* nb, const nbnxn_atomdata_t* nbat)
+{
+ bool bDoTime = nb->bDoTime;
+ Nbnxm::GpuTimers* timers = bDoTime ? nb->timers : nullptr;
+ NBAtomDataGpu* atdat = nb->atdat;
+ const DeviceContext& deviceContext = *nb->deviceContext_;
+ const DeviceStream& localStream = *nb->deviceStreams[InteractionLocality::Local];
+
+ int numAtoms = nbat->numAtoms();
+ bool realloced = false;
+
+ if (bDoTime)
+ {
+ /* time async copy */
+ timers->atdat.openTimingRegion(localStream);
+ }
+
+ /* need to reallocate if we have to copy more atoms than the amount of space
+ available and only allocate if we haven't initialized yet, i.e atdat->natoms == -1 */
+ if (numAtoms > atdat->numAtomsAlloc)
+ {
+ int numAlloc = over_alloc_small(numAtoms);
+
+ /* free up first if the arrays have already been initialized */
+ if (atdat->numAtomsAlloc != -1)
+ {
+ freeDeviceBuffer(&atdat->f);
+ freeDeviceBuffer(&atdat->xq);
+ if (useLjCombRule(nb->nbparam->vdwType))
+ {
+ freeDeviceBuffer(&atdat->ljComb);
+ }
+ else
+ {
+ freeDeviceBuffer(&atdat->atomTypes);
+ }
+ }
+
+
+ allocateDeviceBuffer(&atdat->f, numAlloc, deviceContext);
+ allocateDeviceBuffer(&atdat->xq, numAlloc, deviceContext);
+
+ if (useLjCombRule(nb->nbparam->vdwType))
+ {
+ // Two Lennard-Jones parameters per atom
+ allocateDeviceBuffer(&atdat->ljComb, numAlloc, deviceContext);
+ }
+ else
+ {
+ allocateDeviceBuffer(&atdat->atomTypes, numAlloc, deviceContext);
+ }
+
+ atdat->numAtomsAlloc = numAlloc;
+ realloced = true;
+ }
+
+ atdat->numAtoms = numAtoms;
+ atdat->numAtomsLocal = nbat->natoms_local;
+
+ /* need to clear GPU f output if realloc happened */
+ if (realloced)
+ {
+ clearDeviceBufferAsync(&atdat->f, 0, atdat->numAtomsAlloc, localStream);
+ }
+
+ if (useLjCombRule(nb->nbparam->vdwType))
+ {
+ static_assert(
+ sizeof(Float2) == 2 * sizeof(*nbat->params().lj_comb.data()),
+ "Size of a pair of LJ parameters elements should be equal to the size of Float2.");
+ copyToDeviceBuffer(&atdat->ljComb,
+ reinterpret_cast<const Float2*>(nbat->params().lj_comb.data()),
+ 0,
+ numAtoms,
+ localStream,
+ GpuApiCallBehavior::Async,
+ bDoTime ? timers->atdat.fetchNextEvent() : nullptr);
+ }
+ else
+ {
+ static_assert(sizeof(int) == sizeof(*nbat->params().type.data()),
+ "Sizes of host- and device-side atom types should be the same.");
+ copyToDeviceBuffer(&atdat->atomTypes,
+ nbat->params().type.data(),
+ 0,
+ numAtoms,
+ localStream,
+ GpuApiCallBehavior::Async,
+ bDoTime ? timers->atdat.fetchNextEvent() : nullptr);
+ }
+
+ if (bDoTime)
+ {
+ timers->atdat.closeTimingRegion(localStream);
+ }
+
+ /* kick off the tasks enqueued above to ensure concurrency with the search */
+ issueClFlushInStream(localStream);
+}
+
+void gpu_clear_outputs(NbnxmGpu* nb, bool computeVirial)
+{
+ NBAtomDataGpu* adat = nb->atdat;
+ const DeviceStream& localStream = *nb->deviceStreams[InteractionLocality::Local];
+ // Clear forces
+ clearDeviceBufferAsync(&adat->f, 0, nb->atdat->numAtoms, localStream);
+ // Clear shift force array and energies if the outputs were used in the current step
+ if (computeVirial)
+ {
+ clearDeviceBufferAsync(&adat->fShift, 0, gmx::c_numShiftVectors, localStream);
+ clearDeviceBufferAsync(&adat->eLJ, 0, 1, localStream);
+ clearDeviceBufferAsync(&adat->eElec, 0, 1, localStream);
+ }
+ issueClFlushInStream(localStream);
+}
+
//! This function is documented in the header file
gmx_wallclock_gpu_nbnxn_t* gpu_get_timings(NbnxmGpu* nb)
{
|| (nb->nbparam->elecType == ElecType::EwaldAnaTwin));
}
-enum ElecType nbnxmGpuPickElectrostaticsKernelType(const interaction_const_t* ic,
- const DeviceInformation& deviceInfo)
+void setupGpuShortRangeWork(NbnxmGpu* nb,
+ const gmx::ListedForcesGpu* listedForcesGpu,
+ const gmx::InteractionLocality iLocality)
{
- if (ic->eeltype == CoulombInteractionType::Cut)
+ GMX_ASSERT(nb, "Need a valid nbnxn_gpu object");
+
+ // There is short-range work if the pair list for the provided
+ // interaction locality contains entries or if there is any
+ // bonded work (as this is not split into local/nonlocal).
+ nb->haveWork[iLocality] = ((nb->plist[iLocality]->nsci != 0)
+ || (listedForcesGpu != nullptr && listedForcesGpu->haveInteractions()));
+}
+
+bool haveGpuShortRangeWork(const NbnxmGpu* nb, const gmx::InteractionLocality interactionLocality)
+{
+ GMX_ASSERT(nb, "Need a valid nbnxn_gpu object");
+
+ return nb->haveWork[interactionLocality];
+}
+
+/*! \brief
+ * Launch asynchronously the download of nonbonded forces from the GPU
+ * (and energies/shift forces if required).
+ */
+void gpu_launch_cpyback(NbnxmGpu* nb,
+ struct nbnxn_atomdata_t* nbatom,
+ const gmx::StepWorkload& stepWork,
+ const AtomLocality atomLocality)
+{
+ GMX_ASSERT(nb, "Need a valid nbnxn_gpu object");
+
+ /* determine interaction locality from atom locality */
+ const InteractionLocality iloc = atomToInteractionLocality(atomLocality);
+ GMX_ASSERT(iloc == InteractionLocality::Local
+ || (iloc == InteractionLocality::NonLocal && nb->bNonLocalStreamDoneMarked == false),
+ "Non-local stream is indicating that the copy back event is enqueued at the "
+ "beginning of the copy back function.");
+
+ /* extract the data */
+ NBAtomDataGpu* adat = nb->atdat;
+ Nbnxm::GpuTimers* timers = nb->timers;
+ bool bDoTime = nb->bDoTime;
+ const DeviceStream& deviceStream = *nb->deviceStreams[iloc];
+
+ /* don't launch non-local copy-back if there was no non-local work to do */
+ if ((iloc == InteractionLocality::NonLocal) && !haveGpuShortRangeWork(nb, iloc))
{
- return ElecType::Cut;
+ /* TODO An alternative way to signal that non-local work is
+ complete is to use a clEnqueueMarker+clEnqueueBarrier
+ pair. However, the use of bNonLocalStreamDoneMarked has the
+ advantage of being local to the host, so probably minimizes
+ overhead. Curiously, for NVIDIA OpenCL with an empty-domain
+ test case, overall simulation performance was higher with
+ the API calls, but this has not been tested on AMD OpenCL,
+ so could be worth considering in future. */
+ nb->bNonLocalStreamDoneMarked = false;
+ return;
}
- else if (EEL_RF(ic->eeltype))
+
+ /* local/nonlocal offset and length used for xq and f */
+ auto atomsRange = getGpuAtomRange(adat, atomLocality);
+
+ /* beginning of timed D2H section */
+ if (bDoTime)
{
- return ElecType::RF;
+ timers->xf[atomLocality].nb_d2h.openTimingRegion(deviceStream);
}
- else if ((EEL_PME(ic->eeltype) || ic->eeltype == CoulombInteractionType::Ewald))
+
+ /* With DD the local D2H transfer can only start after the non-local
+ has been launched. */
+ if (iloc == InteractionLocality::Local && nb->bNonLocalStreamDoneMarked)
{
- return nbnxn_gpu_pick_ewald_kernel_type(*ic, deviceInfo);
+ nb->nonlocal_done.enqueueWaitEvent(deviceStream);
+ nb->bNonLocalStreamDoneMarked = false;
}
- else
+
+ /* DtoH f */
+ if (!stepWork.useGpuFBufferOps)
{
- /* Shouldn't happen, as this is checked when choosing Verlet-scheme */
- GMX_THROW(gmx::InconsistentInputError(
- gmx::formatString("The requested electrostatics type %s is not implemented in "
- "the GPU accelerated kernels!",
- enumValueToString(ic->eeltype))));
- }
-}
+ static_assert(
+ sizeof(*nbatom->out[0].f.data()) == sizeof(float),
+ "The host force buffer should be in single precision to match device data size.");
+ copyFromDeviceBuffer(reinterpret_cast<Float3*>(nbatom->out[0].f.data()) + atomsRange.begin(),
+ &adat->f,
+ atomsRange.begin(),
+ atomsRange.size(),
+ deviceStream,
+ GpuApiCallBehavior::Async,
+ bDoTime ? timers->xf[atomLocality].nb_d2h.fetchNextEvent() : nullptr);
+ issueClFlushInStream(deviceStream);
+ }
-enum VdwType nbnxmGpuPickVdwKernelType(const interaction_const_t* ic, LJCombinationRule ljCombinationRule)
-{
- if (ic->vdwtype == VanDerWaalsType::Cut)
+ /* After the non-local D2H is launched the nonlocal_done event can be
+ recorded which signals that the local D2H can proceed. This event is not
+ placed after the non-local kernel because we first need the non-local
+ data back first. */
+ if (iloc == InteractionLocality::NonLocal)
{
- switch (ic->vdw_modifier)
- {
- case InteractionModifiers::None:
- case InteractionModifiers::PotShift:
- switch (ljCombinationRule)
- {
- case LJCombinationRule::None: return VdwType::Cut;
- case LJCombinationRule::Geometric: return VdwType::CutCombGeom;
- case LJCombinationRule::LorentzBerthelot: return VdwType::CutCombLB;
- default:
- GMX_THROW(gmx::InconsistentInputError(gmx::formatString(
- "The requested LJ combination rule %s is not implemented in "
- "the GPU accelerated kernels!",
- enumValueToString(ljCombinationRule))));
- }
- case InteractionModifiers::ForceSwitch: return VdwType::FSwitch;
- case InteractionModifiers::PotSwitch: return VdwType::PSwitch;
- default:
- GMX_THROW(gmx::InconsistentInputError(
- gmx::formatString("The requested VdW interaction modifier %s is not "
- "implemented in the GPU accelerated kernels!",
- enumValueToString(ic->vdw_modifier))));
- }
+ nb->nonlocal_done.markEvent(deviceStream);
+ nb->bNonLocalStreamDoneMarked = true;
}
- else if (ic->vdwtype == VanDerWaalsType::Pme)
+
+ /* only transfer energies in the local stream */
+ if (iloc == InteractionLocality::Local)
{
- if (ic->ljpme_comb_rule == LongRangeVdW::Geom)
+ /* DtoH fshift when virial is needed */
+ if (stepWork.computeVirial)
{
- assert(ljCombinationRule == LJCombinationRule::Geometric);
- return VdwType::EwaldGeom;
+ static_assert(
+ sizeof(*nb->nbst.fShift) == sizeof(Float3),
+ "Sizes of host- and device-side shift vector elements should be the same.");
+ copyFromDeviceBuffer(nb->nbst.fShift,
+ &adat->fShift,
+ 0,
+ gmx::c_numShiftVectors,
+ deviceStream,
+ GpuApiCallBehavior::Async,
+ bDoTime ? timers->xf[atomLocality].nb_d2h.fetchNextEvent() : nullptr);
}
- else
+
+ /* DtoH energies */
+ if (stepWork.computeEnergy)
{
- assert(ljCombinationRule == LJCombinationRule::LorentzBerthelot);
- return VdwType::EwaldLB;
+ static_assert(sizeof(*nb->nbst.eLJ) == sizeof(float),
+ "Sizes of host- and device-side LJ energy terms should be the same.");
+ copyFromDeviceBuffer(nb->nbst.eLJ,
+ &adat->eLJ,
+ 0,
+ 1,
+ deviceStream,
+ GpuApiCallBehavior::Async,
+ bDoTime ? timers->xf[atomLocality].nb_d2h.fetchNextEvent() : nullptr);
+ static_assert(sizeof(*nb->nbst.eElec) == sizeof(float),
+ "Sizes of host- and device-side electrostatic energy terms should be the "
+ "same.");
+ copyFromDeviceBuffer(nb->nbst.eElec,
+ &adat->eElec,
+ 0,
+ 1,
+ deviceStream,
+ GpuApiCallBehavior::Async,
+ bDoTime ? timers->xf[atomLocality].nb_d2h.fetchNextEvent() : nullptr);
}
}
- else
+
+ if (bDoTime)
{
- GMX_THROW(gmx::InconsistentInputError(gmx::formatString(
- "The requested VdW type %s is not implemented in the GPU accelerated kernels!",
- enumValueToString(ic->vdwtype))));
+ timers->xf[atomLocality].nb_d2h.closeTimingRegion(deviceStream);
}
}
-void setupGpuShortRangeWork(NbnxmGpu* nb, const gmx::GpuBonded* gpuBonded, const gmx::InteractionLocality iLocality)
+void nbnxnInsertNonlocalGpuDependency(NbnxmGpu* nb, const InteractionLocality interactionLocality)
{
- GMX_ASSERT(nb, "Need a valid nbnxn_gpu object");
-
- // There is short-range work if the pair list for the provided
- // interaction locality contains entries or if there is any
- // bonded work (as this is not split into local/nonlocal).
- nb->haveWork[iLocality] = ((nb->plist[iLocality]->nsci != 0)
- || (gpuBonded != nullptr && gpuBonded->haveInteractions()));
-}
-
-bool haveGpuShortRangeWork(const NbnxmGpu* nb, const gmx::AtomLocality aLocality)
-{
- GMX_ASSERT(nb, "Need a valid nbnxn_gpu object");
+ const DeviceStream& deviceStream = *nb->deviceStreams[interactionLocality];
- return haveGpuShortRangeWork(*nb, gpuAtomToInteractionLocality(aLocality));
+ /* When we get here all misc operations issued in the local stream as well as
+ the local xq H2D are done,
+ so we record that in the local stream and wait for it in the nonlocal one.
+ This wait needs to precede any PP tasks, bonded or nonbonded, that may
+ compute on interactions between local and nonlocal atoms.
+ */
+ if (nb->bUseTwoStreams)
+ {
+ if (interactionLocality == InteractionLocality::Local)
+ {
+ nb->misc_ops_and_local_H2D_done.markEvent(deviceStream);
+ issueClFlushInStream(deviceStream);
+ }
+ else
+ {
+ nb->misc_ops_and_local_H2D_done.enqueueWaitEvent(deviceStream);
+ }
+ }
}
/*! \brief Launch asynchronously the xq buffer host to device copy. */
{
GMX_ASSERT(nb, "Need a valid nbnxn_gpu object");
- const InteractionLocality iloc = gpuAtomToInteractionLocality(atomLocality);
+ const InteractionLocality iloc = atomToInteractionLocality(atomLocality);
- NBAtomData* adat = nb->atdat;
+ NBAtomDataGpu* adat = nb->atdat;
gpu_plist* plist = nb->plist[iloc];
Nbnxm::GpuTimers* timers = nb->timers;
const DeviceStream& deviceStream = *nb->deviceStreams[iloc];
we always call the local local x+q copy (and the rest of the local
work in nbnxn_gpu_launch_kernel().
*/
- if ((iloc == InteractionLocality::NonLocal) && !haveGpuShortRangeWork(*nb, iloc))
+ if ((iloc == InteractionLocality::NonLocal) && !haveGpuShortRangeWork(nb, iloc))
{
plist->haveFreshList = false;
nbnxnInsertNonlocalGpuDependency(nb, iloc);
}
+
+/* Initialization for X buffer operations on GPU. */
+void nbnxn_gpu_init_x_to_nbat_x(const Nbnxm::GridSet& gridSet, NbnxmGpu* gpu_nbv)
+{
+ const DeviceStream& localStream = *gpu_nbv->deviceStreams[InteractionLocality::Local];
+ const bool bDoTime = gpu_nbv->bDoTime;
+ const int maxNumColumns = gridSet.numColumnsMax();
+
+ reallocateDeviceBuffer(&gpu_nbv->cxy_na,
+ maxNumColumns * gridSet.grids().size(),
+ &gpu_nbv->ncxy_na,
+ &gpu_nbv->ncxy_na_alloc,
+ *gpu_nbv->deviceContext_);
+ reallocateDeviceBuffer(&gpu_nbv->cxy_ind,
+ maxNumColumns * gridSet.grids().size(),
+ &gpu_nbv->ncxy_ind,
+ &gpu_nbv->ncxy_ind_alloc,
+ *gpu_nbv->deviceContext_);
+
+ for (unsigned int g = 0; g < gridSet.grids().size(); g++)
+ {
+ const Nbnxm::Grid& grid = gridSet.grids()[g];
+
+ const int numColumns = grid.numColumns();
+ const int* atomIndices = gridSet.atomIndices().data();
+ const int atomIndicesSize = gridSet.atomIndices().size();
+ const int* cxy_na = grid.cxy_na().data();
+ const int* cxy_ind = grid.cxy_ind().data();
+
+ auto* timerH2D = bDoTime ? &gpu_nbv->timers->xf[AtomLocality::Local].nb_h2d : nullptr;
+
+ reallocateDeviceBuffer(&gpu_nbv->atomIndices,
+ atomIndicesSize,
+ &gpu_nbv->atomIndicesSize,
+ &gpu_nbv->atomIndicesSize_alloc,
+ *gpu_nbv->deviceContext_);
+
+ if (atomIndicesSize > 0)
+ {
+ if (bDoTime)
+ {
+ timerH2D->openTimingRegion(localStream);
+ }
+
+ copyToDeviceBuffer(&gpu_nbv->atomIndices,
+ atomIndices,
+ 0,
+ atomIndicesSize,
+ localStream,
+ GpuApiCallBehavior::Async,
+ bDoTime ? timerH2D->fetchNextEvent() : nullptr);
+
+ if (bDoTime)
+ {
+ timerH2D->closeTimingRegion(localStream);
+ }
+ }
+
+ if (numColumns > 0)
+ {
+ if (bDoTime)
+ {
+ timerH2D->openTimingRegion(localStream);
+ }
+
+ copyToDeviceBuffer(&gpu_nbv->cxy_na,
+ cxy_na,
+ maxNumColumns * g,
+ numColumns,
+ localStream,
+ GpuApiCallBehavior::Async,
+ bDoTime ? timerH2D->fetchNextEvent() : nullptr);
+
+ if (bDoTime)
+ {
+ timerH2D->closeTimingRegion(localStream);
+ }
+
+ if (bDoTime)
+ {
+ timerH2D->openTimingRegion(localStream);
+ }
+
+ copyToDeviceBuffer(&gpu_nbv->cxy_ind,
+ cxy_ind,
+ maxNumColumns * g,
+ numColumns,
+ localStream,
+ GpuApiCallBehavior::Async,
+ bDoTime ? timerH2D->fetchNextEvent() : nullptr);
+
+ if (bDoTime)
+ {
+ timerH2D->closeTimingRegion(localStream);
+ }
+ }
+ }
+
+ // The above data is transferred on the local stream but is a
+ // dependency of the nonlocal stream (specifically the nonlocal X
+ // buf ops kernel). We therefore set a dependency to ensure
+ // that the nonlocal stream waits on the local stream here.
+ // This call records an event in the local stream:
+ nbnxnInsertNonlocalGpuDependency(gpu_nbv, Nbnxm::InteractionLocality::Local);
+ // ...and this call instructs the nonlocal stream to wait on that event:
+ nbnxnInsertNonlocalGpuDependency(gpu_nbv, Nbnxm::InteractionLocality::NonLocal);
+}
+
+//! This function is documented in the header file
+void gpu_free(NbnxmGpu* nb)
+{
+ if (nb == nullptr)
+ {
+ return;
+ }
+
+ gpu_free_platform_specific(nb);
+
+ delete nb->timers;
+ sfree(nb->timings);
+
+ NBAtomDataGpu* atdat = nb->atdat;
+ NBParamGpu* nbparam = nb->nbparam;
+
+ /* Free atdat */
+ freeDeviceBuffer(&(nb->atdat->xq));
+ freeDeviceBuffer(&(nb->atdat->f));
+ freeDeviceBuffer(&(nb->atdat->eLJ));
+ freeDeviceBuffer(&(nb->atdat->eElec));
+ freeDeviceBuffer(&(nb->atdat->fShift));
+ freeDeviceBuffer(&(nb->atdat->shiftVec));
+ if (useLjCombRule(nb->nbparam->vdwType))
+ {
+ freeDeviceBuffer(&atdat->ljComb);
+ }
+ else
+ {
+ freeDeviceBuffer(&atdat->atomTypes);
+ }
+
+ /* Free nbparam */
+ if (nbparam->elecType == ElecType::EwaldTab || nbparam->elecType == ElecType::EwaldTabTwin)
+ {
+ destroyParamLookupTable(&nbparam->coulomb_tab, &nbparam->coulomb_tab_texobj);
+ }
+
+ if (!useLjCombRule(nb->nbparam->vdwType))
+ {
+ destroyParamLookupTable(&nbparam->nbfp, &nbparam->nbfp_texobj);
+ }
+
+ if (nbparam->vdwType == VdwType::EwaldGeom || nbparam->vdwType == VdwType::EwaldLB)
+ {
+ destroyParamLookupTable(&nbparam->nbfp_comb, &nbparam->nbfp_comb_texobj);
+ }
+
+ /* Free plist */
+ auto* plist = nb->plist[InteractionLocality::Local];
+ freeDeviceBuffer(&plist->sci);
+ freeDeviceBuffer(&plist->cj4);
+ freeDeviceBuffer(&plist->imask);
+ freeDeviceBuffer(&plist->excl);
+ delete plist;
+ if (nb->bUseTwoStreams)
+ {
+ auto* plist_nl = nb->plist[InteractionLocality::NonLocal];
+ freeDeviceBuffer(&plist_nl->sci);
+ freeDeviceBuffer(&plist_nl->cj4);
+ freeDeviceBuffer(&plist_nl->imask);
+ freeDeviceBuffer(&plist_nl->excl);
+ delete plist_nl;
+ }
+
+ /* Free nbst */
+ pfree(nb->nbst.eLJ);
+ nb->nbst.eLJ = nullptr;
+
+ pfree(nb->nbst.eElec);
+ nb->nbst.eElec = nullptr;
+
+ pfree(nb->nbst.fShift);
+ nb->nbst.fShift = nullptr;
+
+ delete atdat;
+ delete nbparam;
+ delete nb;
+
+ if (debug)
+ {
+ fprintf(debug, "Cleaned up NBNXM GPU data structures.\n");
+ }
+}
+
+DeviceBuffer<gmx::RVec> gpu_get_f(NbnxmGpu* nb)
+{
+ GMX_ASSERT(nb != nullptr, "nb pointer must be valid");
+
+ return nb->atdat->f;
+}
+
} // namespace Nbnxm
biod.pnpi.spb.ru / alexxy / gromacs.git / blobdiff