2 * This file is part of the GROMACS molecular simulation package.
4 * Copyright (c) 2018,2019, by the GROMACS development team, led by
5 * Mark Abraham, David van der Spoel, Berk Hess, and Erik Lindahl,
6 * and including many others, as listed in the AUTHORS file in the
7 * top-level source directory and at http://www.gromacs.org.
9 * GROMACS is free software; you can redistribute it and/or
10 * modify it under the terms of the GNU Lesser General Public License
11 * as published by the Free Software Foundation; either version 2.1
12 * of the License, or (at your option) any later version.
14 * GROMACS is distributed in the hope that it will be useful,
15 * but WITHOUT ANY WARRANTY; without even the implied warranty of
16 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
17 * Lesser General Public License for more details.
19 * You should have received a copy of the GNU Lesser General Public
20 * License along with GROMACS; if not, see
21 * http://www.gnu.org/licenses, or write to the Free Software Foundation,
22 * Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA.
24 * If you want to redistribute modifications to GROMACS, please
25 * consider that scientific software is very special. Version
26 * control is crucial - bugs must be traceable. We will be happy to
27 * consider code for inclusion in the official distribution, but
28 * derived work must not be called official GROMACS. Details are found
29 * in the README & COPYING files - if they are missing, get the
30 * official version at http://www.gromacs.org.
32 * To help us fund GROMACS development, we humbly ask that you cite
33 * the research papers on the package. Check out http://www.gromacs.org.
37 * \brief Implements GPU bonded lists for CUDA
39 * \author Berk Hess <hess@kth.se>
40 * \author Szilárd Páll <pall.szilard@gmail.com>
41 * \author Mark Abraham <mark.j.abraham@gmail.com>
42 * \author Magnus Lundborg <lundborg.magnus@gmail.com>
44 * \ingroup module_listed_forces
49 #include "gpubonded_impl.h"
51 #include "gromacs/gpu_utils/cuda_arch_utils.cuh"
52 #include "gromacs/gpu_utils/cudautils.cuh"
53 #include "gromacs/gpu_utils/devicebuffer.h"
54 #include "gromacs/mdtypes/enerdata.h"
55 #include "gromacs/timing/wallcycle.h"
56 #include "gromacs/topology/forcefieldparameters.h"
63 // ---- GpuBonded::Impl
65 GpuBonded::Impl::Impl(const gmx_ffparams_t& ffparams, void* streamPtr, gmx_wallcycle* wcycle)
67 stream_ = *static_cast<CommandStream*>(streamPtr);
70 allocateDeviceBuffer(&d_forceParams_, ffparams.numTypes(), nullptr);
71 // This could be an async transfer (if the source is pinned), so
72 // long as it uses the same stream as the kernels and we are happy
73 // to consume additional pinned pages.
74 copyToDeviceBuffer(&d_forceParams_, ffparams.iparams.data(), 0, ffparams.numTypes(), stream_,
75 GpuApiCallBehavior::Sync, nullptr);
77 allocateDeviceBuffer(&d_vTot_, F_NRE, nullptr);
78 clearDeviceBufferAsync(&d_vTot_, 0, F_NRE, stream_);
80 for (int fType = 0; fType < F_NRE; fType++)
82 d_iLists_[fType].nr = 0;
83 d_iLists_[fType].iatoms = nullptr;
84 d_iLists_[fType].nalloc = 0;
87 kernelParams_.d_forceParams = d_forceParams_;
88 kernelParams_.d_xq = d_xq_;
89 kernelParams_.d_f = d_f_;
90 kernelParams_.d_fShift = d_fShift_;
91 kernelParams_.d_vTot = d_vTot_;
92 for (int i = 0; i < numFTypesOnGpu; i++)
94 kernelParams_.d_iatoms[i] = nullptr;
95 kernelParams_.fTypeRangeStart[i] = 0;
96 kernelParams_.fTypeRangeEnd[i] = -1;
100 GpuBonded::Impl::~Impl()
102 for (int fType : fTypesOnGpu)
104 if (d_iLists_[fType].iatoms)
106 freeDeviceBuffer(&d_iLists_[fType].iatoms);
107 d_iLists_[fType].iatoms = nullptr;
111 freeDeviceBuffer(&d_forceParams_);
112 freeDeviceBuffer(&d_vTot_);
115 //! Return whether function type \p fType in \p idef has perturbed interactions
116 static bool fTypeHasPerturbedEntries(const t_idef& idef, int fType)
118 GMX_ASSERT(idef.ilsort == ilsortNO_FE || idef.ilsort == ilsortFE_SORTED,
119 "Perturbed interations should be sorted here");
121 const t_ilist& ilist = idef.il[fType];
123 return (idef.ilsort != ilsortNO_FE && idef.numNonperturbedInteractions[fType] != ilist.nr);
126 //! Converts \p src with atom indices in state order to \p dest in nbnxn order
127 static void convertIlistToNbnxnOrder(const t_ilist& src,
128 HostInteractionList* dest,
129 int numAtomsPerInteraction,
130 ArrayRef<const int> nbnxnAtomOrder)
132 GMX_ASSERT(src.size() == 0 || !nbnxnAtomOrder.empty(), "We need the nbnxn atom order");
134 dest->iatoms.resize(src.size());
136 // TODO use OpenMP to parallelise this loop
137 for (int i = 0; i < src.size(); i += 1 + numAtomsPerInteraction)
139 dest->iatoms[i] = src.iatoms[i];
140 for (int a = 0; a < numAtomsPerInteraction; a++)
142 dest->iatoms[i + 1 + a] = nbnxnAtomOrder[src.iatoms[i + 1 + a]];
147 //! Returns \p input rounded up to the closest multiple of \p factor.
148 static inline int roundUpToFactor(const int input, const int factor)
150 GMX_ASSERT(factor > 0, "The factor to round up to must be > 0.");
152 int remainder = input % factor;
158 return (input + (factor - remainder));
161 // TODO Consider whether this function should be a factory method that
162 // makes an object that is the only one capable of the device
163 // operations needed for the lifetime of an interaction list. This
164 // would be harder to misuse than GpuBonded, and exchange the problem
165 // of naming this method for the problem of what to name the
166 // BondedDeviceInteractionListHandler type.
168 /*! Divides bonded interactions over threads and GPU.
169 * The bonded interactions are assigned by interaction type to GPU threads. The intereaction
170 * types are assigned in blocks sized as <warp_size>. The beginning and end (thread index) of each
171 * interaction type are stored in kernelParams_. Pointers to the relevant data structures on the
172 * GPU are also stored in kernelParams_.
174 void GpuBonded::Impl::updateInteractionListsAndDeviceBuffers(ArrayRef<const int> nbnxnAtomOrder,
180 // TODO wallcycle sub start
181 haveInteractions_ = false;
182 int fTypesCounter = 0;
184 for (int fType : fTypesOnGpu)
186 auto& iList = iLists_[fType];
188 /* Perturbation is not implemented in the GPU bonded kernels.
189 * But instead of doing all interactions on the CPU, we can
190 * still easily handle the types that have no perturbed
191 * interactions on the GPU. */
192 if (idef.il[fType].nr > 0 && !fTypeHasPerturbedEntries(idef, fType))
194 haveInteractions_ = true;
196 convertIlistToNbnxnOrder(idef.il[fType], &iList, NRAL(fType), nbnxnAtomOrder);
200 iList.iatoms.clear();
203 // Update the device if necessary. This can leave some
204 // allocations on the device when the host size decreases to
205 // zero, which is OK, since we deallocate everything at the
207 if (iList.size() > 0)
209 t_ilist& d_iList = d_iLists_[fType];
211 reallocateDeviceBuffer(&d_iList.iatoms, iList.size(), &d_iList.nr, &d_iList.nalloc, nullptr);
213 copyToDeviceBuffer(&d_iList.iatoms, iList.iatoms.data(), 0, iList.size(), stream_,
214 GpuApiCallBehavior::Async, nullptr);
216 kernelParams_.fTypesOnGpu[fTypesCounter] = fType;
217 kernelParams_.numFTypeIAtoms[fTypesCounter] = iList.size();
218 int numBonds = iList.size() / (interaction_function[fType].nratoms + 1);
219 kernelParams_.numFTypeBonds[fTypesCounter] = numBonds;
220 kernelParams_.d_iatoms[fTypesCounter] = d_iLists_[fType].iatoms;
221 if (fTypesCounter == 0)
223 kernelParams_.fTypeRangeStart[fTypesCounter] = 0;
227 kernelParams_.fTypeRangeStart[fTypesCounter] =
228 kernelParams_.fTypeRangeEnd[fTypesCounter - 1] + 1;
230 kernelParams_.fTypeRangeEnd[fTypesCounter] = kernelParams_.fTypeRangeStart[fTypesCounter]
231 + roundUpToFactor(numBonds, warp_size) - 1;
233 GMX_ASSERT(numBonds > 0
234 || kernelParams_.fTypeRangeEnd[fTypesCounter]
235 <= kernelParams_.fTypeRangeStart[fTypesCounter],
236 "Invalid GPU listed forces setup. numBonds must be > 0 if there are threads "
237 "allocated to do work on that interaction function type.");
238 GMX_ASSERT(kernelParams_.fTypeRangeStart[fTypesCounter] % warp_size == 0
239 && (kernelParams_.fTypeRangeEnd[fTypesCounter] + 1) % warp_size == 0,
240 "The bonded interactions must be assigned to the GPU in blocks of warp size.");
245 d_xq_ = static_cast<float4*>(d_xqPtr);
246 d_f_ = static_cast<fvec*>(d_fPtr);
247 d_fShift_ = static_cast<fvec*>(d_fShiftPtr);
249 kernelParams_.d_xq = d_xq_;
250 kernelParams_.d_f = d_f_;
251 kernelParams_.d_fShift = d_fShift_;
252 kernelParams_.d_forceParams = d_forceParams_;
253 kernelParams_.d_vTot = d_vTot_;
255 // TODO wallcycle sub stop
258 bool GpuBonded::Impl::haveInteractions() const
260 return haveInteractions_;
263 void GpuBonded::Impl::launchEnergyTransfer()
265 GMX_ASSERT(haveInteractions_,
266 "No GPU bonded interactions, so no energies will be computed, so transfer should "
269 wallcycle_sub_start_nocount(wcycle_, ewcsLAUNCH_GPU_BONDED);
270 // TODO add conditional on whether there has been any compute (and make sure host buffer doesn't contain garbage)
271 float* h_vTot = vTot_.data();
272 copyFromDeviceBuffer(h_vTot, &d_vTot_, 0, F_NRE, stream_, GpuApiCallBehavior::Async, nullptr);
273 wallcycle_sub_stop(wcycle_, ewcsLAUNCH_GPU_BONDED);
276 void GpuBonded::Impl::waitAccumulateEnergyTerms(gmx_enerdata_t* enerd)
278 GMX_ASSERT(haveInteractions_,
279 "No GPU bonded interactions, so no energies will be computed or transferred, so "
280 "accumulation should not occur");
282 wallcycle_start(wcycle_, ewcWAIT_GPU_BONDED);
283 cudaError_t stat = cudaStreamSynchronize(stream_);
284 CU_RET_ERR(stat, "D2H transfer of bonded energies failed");
285 wallcycle_stop(wcycle_, ewcWAIT_GPU_BONDED);
287 for (int fType : fTypesOnGpu)
289 if (fType != F_LJ14 && fType != F_COUL14)
291 enerd->term[fType] += vTot_[fType];
295 // Note: We do not support energy groups here
296 gmx_grppairener_t* grppener = &enerd->grpp;
297 GMX_RELEASE_ASSERT(grppener->nener == 1, "No energy group support for bondeds on the GPU");
298 grppener->ener[egLJ14][0] += vTot_[F_LJ14];
299 grppener->ener[egCOUL14][0] += vTot_[F_COUL14];
302 void GpuBonded::Impl::clearEnergies()
304 wallcycle_start_nocount(wcycle_, ewcLAUNCH_GPU);
305 wallcycle_sub_start_nocount(wcycle_, ewcsLAUNCH_GPU_BONDED);
306 clearDeviceBufferAsync(&d_vTot_, 0, F_NRE, stream_);
307 wallcycle_sub_stop(wcycle_, ewcsLAUNCH_GPU_BONDED);
308 wallcycle_stop(wcycle_, ewcLAUNCH_GPU);
313 GpuBonded::GpuBonded(const gmx_ffparams_t& ffparams, void* streamPtr, gmx_wallcycle* wcycle) :
314 impl_(new Impl(ffparams, streamPtr, wcycle))
318 GpuBonded::~GpuBonded() = default;
320 void GpuBonded::updateInteractionListsAndDeviceBuffers(ArrayRef<const int> nbnxnAtomOrder,
326 impl_->updateInteractionListsAndDeviceBuffers(nbnxnAtomOrder, idef, d_xq, d_f, d_fShift);
329 bool GpuBonded::haveInteractions() const
331 return impl_->haveInteractions();
334 void GpuBonded::launchEnergyTransfer()
336 impl_->launchEnergyTransfer();
339 void GpuBonded::waitAccumulateEnergyTerms(gmx_enerdata_t* enerd)
341 impl_->waitAccumulateEnergyTerms(enerd);
344 void GpuBonded::clearEnergies()
346 impl_->clearEnergies();