1 /* -*- mode: c; tab-width: 4; indent-tabs-mode: nil; c-basic-offset: 4; c-file-style: "stroustrup"; -*-
4 * This source code is part of
8 * GROningen MAchine for Chemical Simulations
10 * Written by David van der Spoel, Erik Lindahl, Berk Hess, and others.
11 * Copyright (c) 1991-2000, University of Groningen, The Netherlands.
12 * Copyright (c) 2001-2012, The GROMACS development team,
13 * check out http://www.gromacs.org for more information.
15 * This program is free software; you can redistribute it and/or
16 * modify it under the terms of the GNU General Public License
17 * as published by the Free Software Foundation; either version 2
18 * of the License, or (at your option) any later version.
20 * If you want to redistribute modifications, please consider that
21 * scientific software is very special. Version control is crucial -
22 * bugs must be traceable. We will be happy to consider code for
23 * inclusion in the official distribution, but derived work must not
24 * be called official GROMACS. Details are found in the README & COPYING
25 * files - if they are missing, get the official version at www.gromacs.org.
27 * To help us fund GROMACS development, we humbly ask that you cite
28 * the papers on the package - you can find them in the top README file.
30 * For more info, check our website at http://www.gromacs.org
33 * Gallium Rubidium Oxygen Manganese Argon Carbon Silicon
36 /* Note that floating-point constants in CUDA code should be suffixed
37 * with f (e.g. 0.5f), to stop the compiler producing intermediate
38 * code that is in double precision.
41 #include "../../gmxlib/cuda_tools/vectype_ops.cuh"
43 #ifndef NBNXN_CUDA_KERNEL_UTILS_CUH
44 #define NBNXN_CUDA_KERNEL_UTILS_CUH
46 #define WARP_SIZE_POW2_EXPONENT (5)
47 #define CL_SIZE_POW2_EXPONENT (3) /* change this together with GPU_NS_CLUSTER_SIZE !*/
48 #define CL_SIZE_SQ (CL_SIZE * CL_SIZE)
49 #define FBUF_STRIDE (CL_SIZE_SQ)
51 /*! i-cluster interaction mask for a super-cluster with all NCL_PER_SUPERCL bits set */
52 const unsigned supercl_interaction_mask = ((1U << NCL_PER_SUPERCL) - 1U);
54 /*! Interpolate Ewald coulomb force using the table through the tex_nbfp texture.
55 * Original idea: OpenMM
57 static inline __device__
58 float interpolate_coulomb_force_r(float r, float scale)
60 float normalized = scale * r;
61 int index = (int) normalized;
62 float fract2 = normalized - index;
63 float fract1 = 1.0f - fract2;
65 return fract1 * tex1Dfetch(tex_coulomb_tab, index)
66 + fract2 * tex1Dfetch(tex_coulomb_tab, index + 1);
69 /*! Calculate analytical Ewald correction term. */
70 static inline __device__
71 float pmecorrF(float z2)
73 const float FN6 = -1.7357322914161492954e-8f;
74 const float FN5 = 1.4703624142580877519e-6f;
75 const float FN4 = -0.000053401640219807709149f;
76 const float FN3 = 0.0010054721316683106153f;
77 const float FN2 = -0.019278317264888380590f;
78 const float FN1 = 0.069670166153766424023f;
79 const float FN0 = -0.75225204789749321333f;
81 const float FD4 = 0.0011193462567257629232f;
82 const float FD3 = 0.014866955030185295499f;
83 const float FD2 = 0.11583842382862377919f;
84 const float FD1 = 0.50736591960530292870f;
85 const float FD0 = 1.0f;
88 float polyFN0,polyFN1,polyFD0,polyFD1;
92 polyFD0 = FD4*z4 + FD2;
93 polyFD1 = FD3*z4 + FD1;
94 polyFD0 = polyFD0*z4 + FD0;
95 polyFD0 = polyFD1*z2 + polyFD0;
97 polyFD0 = 1.0f/polyFD0;
99 polyFN0 = FN6*z4 + FN4;
100 polyFN1 = FN5*z4 + FN3;
101 polyFN0 = polyFN0*z4 + FN2;
102 polyFN1 = polyFN1*z4 + FN1;
103 polyFN0 = polyFN0*z4 + FN0;
104 polyFN0 = polyFN1*z2 + polyFN0;
106 return polyFN0*polyFD0;
109 /*! Final j-force reduction; this generic implementation works with
110 * arbitrary array sizes.
112 static inline __device__
113 void reduce_force_j_generic(float *f_buf, float3 *fout,
114 int tidxi, int tidxj, int aidx)
118 float3 f = make_float3(0.0f);
119 for (int j = tidxj * CL_SIZE; j < (tidxj + 1) * CL_SIZE; j++)
122 f.y += f_buf[ FBUF_STRIDE + j];
123 f.z += f_buf[2 * FBUF_STRIDE + j];
126 atomicAdd(&fout[aidx], f);
130 /*! Final j-force reduction; this implementation only with power of two
131 * array sizes and with sm >= 3.0
133 #if __CUDA_ARCH__ >= 300
134 static inline __device__
135 void reduce_force_j_warp_shfl(float3 f, float3 *fout,
141 for (i = 0; i < 3; i++)
143 f.x += __shfl_down(f.x, 1<<i);
144 f.y += __shfl_down(f.y, 1<<i);
145 f.z += __shfl_down(f.z, 1<<i);
148 /* Write the reduced j-force on one thread for each j */
151 atomicAdd(&fout[aidx], f);
156 /*! Final i-force reduction; this generic implementation works with
157 * arbitrary array sizes.
159 static inline __device__
160 void reduce_force_i_generic(float *f_buf, float3 *fout,
161 float3 *fshift_buf, bool bCalcFshift,
162 int tidxi, int tidxj, int aidx)
166 float3 f = make_float3(0.0f);
167 for (int j = tidxi; j < CL_SIZE_SQ; j += CL_SIZE)
170 f.y += f_buf[ FBUF_STRIDE + j];
171 f.z += f_buf[2 * FBUF_STRIDE + j];
174 atomicAdd(&fout[aidx], f);
183 /*! Final i-force reduction; this implementation works only with power of two
186 static inline __device__
187 void reduce_force_i_pow2(volatile float *f_buf, float3 *fout,
188 float3 *fshift_buf, bool bCalcFshift,
189 int tidxi, int tidxj, int aidx)
192 float3 f = make_float3(0.0f);
194 /* Reduce the initial CL_SIZE values for each i atom to half
195 * every step by using CL_SIZE * i threads.
196 * Can't just use i as loop variable because than nvcc refuses to unroll.
200 for (j = CL_SIZE_POW2_EXPONENT - 1; j > 0; j--)
205 f_buf[ tidxj * CL_SIZE + tidxi] += f_buf[ (tidxj + i) * CL_SIZE + tidxi];
206 f_buf[ FBUF_STRIDE + tidxj * CL_SIZE + tidxi] += f_buf[ FBUF_STRIDE + (tidxj + i) * CL_SIZE + tidxi];
207 f_buf[2 * FBUF_STRIDE + tidxj * CL_SIZE + tidxi] += f_buf[2 * FBUF_STRIDE + (tidxj + i) * CL_SIZE + tidxi];
212 /* i == 1, last reduction step, writing to global mem */
215 f.x = f_buf[ tidxj * CL_SIZE + tidxi] + f_buf[ (tidxj + i) * CL_SIZE + tidxi];
216 f.y = f_buf[ FBUF_STRIDE + tidxj * CL_SIZE + tidxi] + f_buf[ FBUF_STRIDE + (tidxj + i) * CL_SIZE + tidxi];
217 f.z = f_buf[2 * FBUF_STRIDE + tidxj * CL_SIZE + tidxi] + f_buf[2 * FBUF_STRIDE + (tidxj + i) * CL_SIZE + tidxi];
219 atomicAdd(&fout[aidx], f);
228 /*! Final i-force reduction wrapper; calls the generic or pow2 reduction depending
229 * on whether the size of the array to be reduced is power of two or not.
231 static inline __device__
232 void reduce_force_i(float *f_buf, float3 *f,
233 float3 *fshift_buf, bool bCalcFshift,
234 int tidxi, int tidxj, int ai)
236 if ((CL_SIZE & (CL_SIZE - 1)))
238 reduce_force_i_generic(f_buf, f, fshift_buf, bCalcFshift, tidxi, tidxj, ai);
242 reduce_force_i_pow2(f_buf, f, fshift_buf, bCalcFshift, tidxi, tidxj, ai);
246 /*! Final i-force reduction; this implementation works only with power of two
247 * array sizes and with sm >= 3.0
249 #if __CUDA_ARCH__ >= 300
250 static inline __device__
251 void reduce_force_i_warp_shfl(float3 fin, float3 *fout,
252 float3 *fshift_buf, bool bCalcFshift,
258 for (j = 0; j < 2; j++)
260 fin.x += __shfl_down(fin.x, CL_SIZE<<j);
261 fin.y += __shfl_down(fin.y, CL_SIZE<<j);
262 fin.z += __shfl_down(fin.z, CL_SIZE<<j);
265 /* The first thread in the warp writes the reduced force */
266 if (tidxj == 0 || tidxj == 4)
268 atomicAdd(&fout[aidx], fin);
272 fshift_buf->x += fin.x;
273 fshift_buf->y += fin.y;
274 fshift_buf->z += fin.z;
280 /*! Energy reduction; this implementation works only with power of two
283 static inline __device__
284 void reduce_energy_pow2(volatile float *buf,
285 float *e_lj, float *e_el,
293 /* Can't just use i as loop variable because than nvcc refuses to unroll. */
295 for (j = WARP_SIZE_POW2_EXPONENT - 1; j > 0; j--)
299 buf[ tidx] += buf[ tidx + i];
300 buf[FBUF_STRIDE + tidx] += buf[FBUF_STRIDE + tidx + i];
305 /* last reduction step, writing to global mem */
308 e1 = buf[ tidx] + buf[ tidx + i];
309 e2 = buf[FBUF_STRIDE + tidx] + buf[FBUF_STRIDE + tidx + i];
316 /*! Energy reduction; this implementation works only with power of two
317 * array sizes and with sm >= 3.0
319 #if __CUDA_ARCH__ >= 300
320 static inline __device__
321 void reduce_energy_warp_shfl(float E_lj, float E_el,
322 float *e_lj, float *e_el,
329 for (i = 0; i < 5; i++)
331 E_lj += __shfl_down(E_lj,sh);
332 E_el += __shfl_down(E_el,sh);
336 /* The first thread in the warp writes the reduced energies */
337 if (tidx == 0 || tidx == WARP_SIZE)
339 atomicAdd(e_lj,E_lj);
340 atomicAdd(e_el,E_el);
343 #endif /* __CUDA_ARCH__ */
345 #endif /* NBNXN_CUDA_KERNEL_UTILS_CUH */