#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/pbcutil/ishift.h"
#include "gromacs/timing/gpu_timing.h"
#include "gromacs/utility/cstringutil.h"
#include "gromacs/utility/exceptions.h"
return haveGpuShortRangeWork(*nb, gpuAtomToInteractionLocality(aLocality));
}
+/*! \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 = gpuAtomToInteractionLocality(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 */
+ NBAtomData* 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))
+ {
+ /* 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;
+ }
+
+ /* local/nonlocal offset and length used for xq and f */
+ auto atomsRange = getGpuAtomRange(adat, atomLocality);
+
+ /* beginning of timed D2H section */
+ if (bDoTime)
+ {
+ timers->xf[atomLocality].nb_d2h.openTimingRegion(deviceStream);
+ }
+
+ /* With DD the local D2H transfer can only start after the non-local
+ has been launched. */
+ if (iloc == InteractionLocality::Local && nb->bNonLocalStreamDoneMarked)
+ {
+ nb->nonlocal_done.enqueueWaitEvent(deviceStream);
+ nb->bNonLocalStreamDoneMarked = false;
+ }
+
+ /* DtoH f */
+ 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);
+
+ /* 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)
+ {
+ nb->nonlocal_done.markEvent(deviceStream);
+ nb->bNonLocalStreamDoneMarked = true;
+ }
+
+ /* only transfer energies in the local stream */
+ if (iloc == InteractionLocality::Local)
+ {
+ /* DtoH fshift when virial is needed */
+ if (stepWork.computeVirial)
+ {
+ 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,
+ SHIFTS,
+ deviceStream,
+ GpuApiCallBehavior::Async,
+ bDoTime ? timers->xf[atomLocality].nb_d2h.fetchNextEvent() : nullptr);
+ }
+
+ /* DtoH energies */
+ if (stepWork.computeEnergy)
+ {
+ 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);
+ }
+ }
+
+ if (bDoTime)
+ {
+ timers->xf[atomLocality].nb_d2h.closeTimingRegion(deviceStream);
+ }
+}
+
void nbnxnInsertNonlocalGpuDependency(NbnxmGpu* nb, const InteractionLocality interactionLocality)
{
const DeviceStream& deviceStream = *nb->deviceStreams[interactionLocality];
biod.pnpi.spb.ru / alexxy / gromacs.git / blobdiff