2 * This file is part of the GROMACS molecular simulation package.
4 * Copyright (c) 1991-2000, University of Groningen, The Netherlands.
5 * Copyright (c) 2001-2004, The GROMACS development team.
6 * Copyright (c) 2013-2016,2017,2018,2019,2020, by the GROMACS development team, led by
7 * Mark Abraham, David van der Spoel, Berk Hess, and Erik Lindahl,
8 * and including many others, as listed in the AUTHORS file in the
9 * top-level source directory and at http://www.gromacs.org.
11 * GROMACS is free software; you can redistribute it and/or
12 * modify it under the terms of the GNU Lesser General Public License
13 * as published by the Free Software Foundation; either version 2.1
14 * of the License, or (at your option) any later version.
16 * GROMACS is distributed in the hope that it will be useful,
17 * but WITHOUT ANY WARRANTY; without even the implied warranty of
18 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
19 * Lesser General Public License for more details.
21 * You should have received a copy of the GNU Lesser General Public
22 * License along with GROMACS; if not, see
23 * http://www.gnu.org/licenses, or write to the Free Software Foundation,
24 * Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA.
26 * If you want to redistribute modifications to GROMACS, please
27 * consider that scientific software is very special. Version
28 * control is crucial - bugs must be traceable. We will be happy to
29 * consider code for inclusion in the official distribution, but
30 * derived work must not be called official GROMACS. Details are found
31 * in the README & COPYING files - if they are missing, get the
32 * official version at http://www.gromacs.org.
34 * To help us fund GROMACS development, we humbly ask that you cite
35 * the research papers on the package. Check out http://www.gromacs.org.
39 * \brief Implements PME GPU spline calculation and charge spreading in CUDA.
40 * TODO: consider always pre-sorting particles (as in DD case).
42 * \author Aleksei Iupinov <a.yupinov@gmail.com>
49 #include "gromacs/gpu_utils/cuda_kernel_utils.cuh"
50 #include "gromacs/gpu_utils/typecasts.cuh"
53 #include "pme_calculate_splines.cuh"
54 #include "pme_gpu_utils.h"
58 * Charge spreading onto the grid.
59 * This corresponds to the CPU function spread_coefficients_bsplines_thread().
60 * Optional second stage of the spline_and_spread_kernel.
62 * \tparam[in] order PME interpolation order.
63 * \tparam[in] wrapX Whether the grid overlap in dimension X should be wrapped.
64 * \tparam[in] wrapY Whether the grid overlap in dimension Y should be wrapped.
65 * \tparam[in] useOrderThreads Whether we should use order threads per atom (order*order used if false).
67 * \param[in] kernelParams Input PME CUDA data in constant memory.
68 * \param[in] atomIndexOffset Starting atom index for the execution block w.r.t. global memory.
69 * \param[in] atomCharge Atom charge/coefficient of atom processed by thread.
70 * \param[in] sm_gridlineIndices Atom gridline indices in the shared memory.
71 * \param[in] sm_theta Atom spline values in the shared memory.
73 template<const int order, const bool wrapX, const bool wrapY, const bool useOrderThreads>
74 __device__ __forceinline__ void spread_charges(const PmeGpuCudaKernelParams kernelParams,
76 const float* atomCharge,
77 const int* __restrict__ sm_gridlineIndices,
78 const float* __restrict__ sm_theta)
80 /* Global memory pointer to the output grid */
81 float* __restrict__ gm_grid = kernelParams.grid.d_realGrid;
84 const int atomsPerWarp = useOrderThreads ? c_pmeSpreadGatherAtomsPerWarp4ThPerAtom
85 : c_pmeSpreadGatherAtomsPerWarp;
87 const int nx = kernelParams.grid.realGridSize[XX];
88 const int ny = kernelParams.grid.realGridSize[YY];
89 const int nz = kernelParams.grid.realGridSize[ZZ];
90 const int pny = kernelParams.grid.realGridSizePadded[YY];
91 const int pnz = kernelParams.grid.realGridSizePadded[ZZ];
93 const int offx = 0, offy = 0, offz = 0; // unused for now
95 const int atomIndexLocal = threadIdx.z;
96 const int atomIndexGlobal = atomIndexOffset + atomIndexLocal;
98 const int globalCheck = pme_gpu_check_atom_data_index(atomIndexGlobal, kernelParams.atoms.nAtoms);
99 const int chargeCheck = pme_gpu_check_atom_charge(*atomCharge);
100 if (chargeCheck & globalCheck)
102 // Spline Z coordinates
103 const int ithz = threadIdx.x;
105 const int ixBase = sm_gridlineIndices[atomIndexLocal * DIM + XX] - offx;
106 const int iyBase = sm_gridlineIndices[atomIndexLocal * DIM + YY] - offy;
107 int iz = sm_gridlineIndices[atomIndexLocal * DIM + ZZ] - offz + ithz;
112 /* Atom index w.r.t. warp - alternating 0 1 0 1 .. */
113 const int atomWarpIndex = atomIndexLocal % atomsPerWarp;
114 /* Warp index w.r.t. block - could probably be obtained easier? */
115 const int warpIndex = atomIndexLocal / atomsPerWarp;
117 const int splineIndexBase = getSplineParamIndexBase<order, atomsPerWarp>(warpIndex, atomWarpIndex);
118 const int splineIndexZ = getSplineParamIndex<order, atomsPerWarp>(splineIndexBase, ZZ, ithz);
119 const float thetaZ = sm_theta[splineIndexZ];
121 /* loop not used if order*order threads per atom */
122 const int ithyMin = useOrderThreads ? 0 : threadIdx.y;
123 const int ithyMax = useOrderThreads ? order : threadIdx.y + 1;
124 for (int ithy = ithyMin; ithy < ithyMax; ithy++)
126 int iy = iyBase + ithy;
127 if (wrapY & (iy >= ny))
132 const int splineIndexY = getSplineParamIndex<order, atomsPerWarp>(splineIndexBase, YY, ithy);
133 float thetaY = sm_theta[splineIndexY];
134 const float Val = thetaZ * thetaY * (*atomCharge);
135 assert(isfinite(Val));
136 const int offset = iy * pnz + iz;
139 for (int ithx = 0; (ithx < order); ithx++)
141 int ix = ixBase + ithx;
142 if (wrapX & (ix >= nx))
146 const int gridIndexGlobal = ix * pny * pnz + offset;
147 const int splineIndexX =
148 getSplineParamIndex<order, atomsPerWarp>(splineIndexBase, XX, ithx);
149 const float thetaX = sm_theta[splineIndexX];
150 assert(isfinite(thetaX));
151 assert(isfinite(gm_grid[gridIndexGlobal]));
152 atomicAdd(gm_grid + gridIndexGlobal, thetaX * Val);
159 * A spline computation and charge spreading kernel function.
161 * Two tuning parameters can be used for additional performance. For small systems and for debugging
162 * writeGlobal should be used removing the need to recalculate the theta values in the gather kernel.
163 * Similarly for useOrderThreads large systems order threads per atom gives higher performance than order*order threads
165 * \tparam[in] order PME interpolation order.
166 * \tparam[in] computeSplines A boolean which tells if the spline parameter and
167 * gridline indices' computation should be performed.
168 * \tparam[in] spreadCharges A boolean which tells if the charge spreading should be performed.
169 * \tparam[in] wrapX A boolean which tells if the grid overlap in dimension X should be wrapped.
170 * \tparam[in] wrapY A boolean which tells if the grid overlap in dimension Y should be wrapped.
171 * \tparam[in] writeGlobal A boolean which tells if the theta values and gridlines should be written to global memory.
172 * \tparam[in] useOrderThreads A boolean which tells if we should use order threads per atom (order*order used if false).
173 * \param[in] kernelParams Input PME CUDA data in constant memory.
175 template<const int order, const bool computeSplines, const bool spreadCharges, const bool wrapX, const bool wrapY, const bool writeGlobal, const bool useOrderThreads>
176 __launch_bounds__(c_spreadMaxThreadsPerBlock) CLANG_DISABLE_OPTIMIZATION_ATTRIBUTE __global__
177 void pme_spline_and_spread_kernel(const PmeGpuCudaKernelParams kernelParams)
179 const int atomsPerBlock =
180 useOrderThreads ? c_spreadMaxThreadsPerBlock / c_pmeSpreadGatherThreadsPerAtom4ThPerAtom
181 : c_spreadMaxThreadsPerBlock / c_pmeSpreadGatherThreadsPerAtom;
182 // Gridline indices, ivec
183 __shared__ int sm_gridlineIndices[atomsPerBlock * DIM];
185 __shared__ float sm_theta[atomsPerBlock * DIM * order];
188 const int atomsPerWarp = useOrderThreads ? c_pmeSpreadGatherAtomsPerWarp4ThPerAtom
189 : c_pmeSpreadGatherAtomsPerWarp;
194 const int blockIndex = blockIdx.y * gridDim.x + blockIdx.x;
195 const int atomIndexOffset = blockIndex * atomsPerBlock;
197 /* Thread index w.r.t. block */
198 const int threadLocalId =
199 (threadIdx.z * (blockDim.x * blockDim.y)) + (threadIdx.y * blockDim.x) + threadIdx.x;
200 /* Warp index w.r.t. block - could probably be obtained easier? */
201 const int warpIndex = threadLocalId / warp_size;
203 /* Atom index w.r.t. warp */
204 const int atomWarpIndex = threadIdx.z % atomsPerWarp;
205 /* Atom index w.r.t. block/shared memory */
206 const int atomIndexLocal = warpIndex * atomsPerWarp + atomWarpIndex;
207 /* Atom index w.r.t. global memory */
208 const int atomIndexGlobal = atomIndexOffset + atomIndexLocal;
210 /* Early return for fully empty blocks at the end
211 * (should only happen for billions of input atoms)
213 if (atomIndexOffset >= kernelParams.atoms.nAtoms)
217 /* Charges, required for both spline and spread */
218 if (c_useAtomDataPrefetch)
220 __shared__ float sm_coefficients[atomsPerBlock];
221 pme_gpu_stage_atom_data<float, atomsPerBlock, 1>(kernelParams, sm_coefficients,
222 kernelParams.atoms.d_coefficients);
224 atomCharge = sm_coefficients[atomIndexLocal];
228 atomCharge = kernelParams.atoms.d_coefficients[atomIndexGlobal];
233 const float3* __restrict__ gm_coordinates = asFloat3(kernelParams.atoms.d_coordinates);
234 if (c_useAtomDataPrefetch)
237 __shared__ float3 sm_coordinates[atomsPerBlock];
239 /* Staging coordinates */
240 pme_gpu_stage_atom_data<float3, atomsPerBlock, 1>(kernelParams, sm_coordinates, gm_coordinates);
242 atomX = sm_coordinates[atomIndexLocal];
246 atomX = gm_coordinates[atomIndexGlobal];
248 calculate_splines<order, atomsPerBlock, atomsPerWarp, false, writeGlobal>(
249 kernelParams, atomIndexOffset, atomX, atomCharge, sm_theta, &dtheta, sm_gridlineIndices);
254 /* Staging the data for spread
255 * (the data is assumed to be in GPU global memory with proper layout already,
256 * as in after running the spline kernel)
258 /* Spline data - only thetas (dthetas will only be needed in gather) */
259 pme_gpu_stage_atom_data<float, atomsPerBlock, DIM * order>(kernelParams, sm_theta,
260 kernelParams.atoms.d_theta);
261 /* Gridline indices */
262 pme_gpu_stage_atom_data<int, atomsPerBlock, DIM>(kernelParams, sm_gridlineIndices,
263 kernelParams.atoms.d_gridlineIndices);
271 spread_charges<order, wrapX, wrapY, useOrderThreads>(
272 kernelParams, atomIndexOffset, &atomCharge, sm_gridlineIndices, sm_theta);
276 //! Kernel instantiations
277 template __global__ void pme_spline_and_spread_kernel<4, true, true, true, true, true, true>(const PmeGpuCudaKernelParams);
278 template __global__ void
279 pme_spline_and_spread_kernel<4, true, false, true, true, true, true>(const PmeGpuCudaKernelParams);
280 template __global__ void
281 pme_spline_and_spread_kernel<4, false, true, true, true, true, true>(const PmeGpuCudaKernelParams);
283 template __global__ void
284 pme_spline_and_spread_kernel<4, true, true, true, true, false, true>(const PmeGpuCudaKernelParams);
286 template __global__ void
287 pme_spline_and_spread_kernel<4, true, true, true, true, true, false>(const PmeGpuCudaKernelParams);
288 template __global__ void
289 pme_spline_and_spread_kernel<4, true, false, true, true, true, false>(const PmeGpuCudaKernelParams);
290 template __global__ void
291 pme_spline_and_spread_kernel<4, false, true, true, true, true, false>(const PmeGpuCudaKernelParams);
293 template __global__ void
294 pme_spline_and_spread_kernel<4, true, true, true, true, false, false>(const PmeGpuCudaKernelParams);