2 * This file is part of the GROMACS molecular simulation package.
4 * Copyright (c) 2018,2019,2020, 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.
39 * CUDA kernels for GPU versions of copy_rvec_to_nbat_real and add_nbat_f_to_f.
41 * \author Alan Gray <alang@nvidia.com>
42 * \author Jon Vincent <jvincent@nvidia.com>
45 #include "gromacs/gpu_utils/vectype_ops.cuh"
46 #include "gromacs/nbnxm/nbnxm.h"
48 /*! \brief CUDA kernel for transforming position coordinates from rvec to nbnxm layout.
51 * - improve/simplify/document use of cxy_na and na_round
52 * - rename kernel so naming matches with the other NBNXM kernels;
53 * - enable separate compilation unit
55 * \param[in] numColumns Extent of cell-level parallelism.
56 * \param[out] gm_xq Coordinates buffer in nbnxm layout.
57 * \param[in] setFillerCoords Whether to set the coordinates of the filler particles.
58 * \param[in] gm_x Coordinates buffer.
59 * \param[in] gm_atomIndex Atom index mapping.
60 * \param[in] gm_numAtoms Array of number of atoms.
61 * \param[in] gm_cellIndex Array of cell indices.
62 * \param[in] cellOffset Airst cell.
63 * \param[in] numAtomsPerCell Number of atoms per cell.
65 static __global__ void nbnxn_gpu_x_to_nbat_x_kernel(int numColumns,
66 float4* __restrict__ gm_xq,
68 const float3* __restrict__ gm_x,
69 const int* __restrict__ gm_atomIndex,
70 const int* __restrict__ gm_numAtoms,
71 const int* __restrict__ gm_cellIndex,
77 const float farAway = -1000000.0f;
79 // Map cell-level parallelism to y component of CUDA block index.
85 const int numAtoms = gm_numAtoms[cxy];
86 const int offset = (cellOffset + gm_cellIndex[cxy]) * numAtomsPerCell;
90 // TODO: This can be done more efficiently
91 numAtomsRounded = (gm_cellIndex[cxy + 1] - gm_cellIndex[cxy]) * numAtomsPerCell;
95 // We fill only the real particle locations.
96 // We assume the filling entries at the end have been
97 // properly set before during pair-list generation.
98 numAtomsRounded = numAtoms;
101 const int threadIndex = blockIdx.x * blockDim.x + threadIdx.x;
103 // Destination address where x should be stored in nbnxm layout. We use this cast here to
104 // save only x, y and z components, not touching the w (q) component, which is pre-defined.
105 float3* gm_xqDest = (float3*)&gm_xq[threadIndex + offset];
107 // Perform layout conversion of each element.
108 if (threadIndex < numAtomsRounded)
110 if (threadIndex < numAtoms)
112 *gm_xqDest = gm_x[gm_atomIndex[threadIndex + offset]];
116 *gm_xqDest = make_float3(farAway);
122 /*! \brief CUDA kernel to sum up the force components
124 * \tparam accumulateForce If the initial forces in \p gm_fTotal should be saved.
125 * \tparam addPmeForce Whether the PME force should be added to the total.
127 * \param[in] gm_fNB Non-bonded forces in nbnxm format.
128 * \param[in] gm_fPme PME forces.
129 * \param[in,out] gm_fTotal Force buffer to be reduced into.
130 * \param[in] cell Cell index mapping.
131 * \param[in] atomStart Start atom index.
132 * \param[in] numAtoms Number of atoms.
134 template<bool accumulateForce, bool addPmeForce>
135 static __global__ void nbnxn_gpu_add_nbat_f_to_f_kernel(const float3* __restrict__ gm_fNB,
136 const float3* __restrict__ gm_fPme,
138 const int* __restrict__ gm_cell,
143 /* map particle-level parallelism to 1D CUDA thread and block index */
144 const int threadIndex = blockIdx.x * blockDim.x + threadIdx.x;
146 /* perform addition for each particle*/
147 if (threadIndex < numAtoms)
150 const int i = gm_cell[atomStart + threadIndex];
151 float3* gm_fDest = &gm_fTotal[atomStart + threadIndex];
165 temp += gm_fPme[atomStart + threadIndex];