2 * This file is part of the GROMACS molecular simulation package.
4 * Copyright (c) 2012,2013,2014,2015,2017, 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.
36 * Note: this file was generated by the GROMACS sparc64_hpc_ace_double kernel generator.
44 #include "../nb_kernel.h"
45 #include "gromacs/gmxlib/nrnb.h"
47 #include "kernelutil_sparc64_hpc_ace_double.h"
50 * Gromacs nonbonded kernel: nb_kernel_ElecGB_VdwCSTab_GeomP1P1_VF_sparc64_hpc_ace_double
51 * Electrostatics interaction: GeneralizedBorn
52 * VdW interaction: CubicSplineTable
53 * Geometry: Particle-Particle
54 * Calculate force/pot: PotentialAndForce
57 nb_kernel_ElecGB_VdwCSTab_GeomP1P1_VF_sparc64_hpc_ace_double
58 (t_nblist * gmx_restrict nlist,
59 rvec * gmx_restrict xx,
60 rvec * gmx_restrict ff,
61 struct t_forcerec * gmx_restrict fr,
62 t_mdatoms * gmx_restrict mdatoms,
63 nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
64 t_nrnb * gmx_restrict nrnb)
66 /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
67 * just 0 for non-waters.
68 * Suffixes A,B refer to j loop unrolling done with double precision SIMD, e.g. for the two different
69 * jnr indices corresponding to data put in the four positions in the SIMD register.
71 int i_shift_offset,i_coord_offset,outeriter,inneriter;
72 int j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
74 int j_coord_offsetA,j_coord_offsetB;
75 int *iinr,*jindex,*jjnr,*shiftidx,*gid;
77 real *shiftvec,*fshift,*x,*f;
78 _fjsp_v2r8 tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
80 _fjsp_v2r8 ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
81 int vdwjidx0A,vdwjidx0B;
82 _fjsp_v2r8 jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
83 _fjsp_v2r8 dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
84 _fjsp_v2r8 velec,felec,velecsum,facel,crf,krf,krf2;
86 _fjsp_v2r8 vgb,fgb,vgbsum,dvdasum,gbscale,gbtabscale,isaprod,gbqqfactor,gbinvepsdiff,dvdaj,gbeps,twogbeps,dvdatmp;
87 _fjsp_v2r8 minushalf = gmx_fjsp_set1_v2r8(-0.5);
88 real *invsqrta,*dvda,*gbtab;
90 _fjsp_v2r8 rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
93 _fjsp_v2r8 one_sixth = gmx_fjsp_set1_v2r8(1.0/6.0);
94 _fjsp_v2r8 one_twelfth = gmx_fjsp_set1_v2r8(1.0/12.0);
95 _fjsp_v2r8 rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF,twovfeps;
98 _fjsp_v2r8 dummy_mask,cutoff_mask;
99 _fjsp_v2r8 one = gmx_fjsp_set1_v2r8(1.0);
100 _fjsp_v2r8 two = gmx_fjsp_set1_v2r8(2.0);
101 union { _fjsp_v2r8 simd; long long int i[2]; } vfconv,gbconv,ewconv;
108 jindex = nlist->jindex;
110 shiftidx = nlist->shift;
112 shiftvec = fr->shift_vec[0];
113 fshift = fr->fshift[0];
114 facel = gmx_fjsp_set1_v2r8(fr->ic->epsfac);
115 charge = mdatoms->chargeA;
116 nvdwtype = fr->ntype;
118 vdwtype = mdatoms->typeA;
120 vftab = kernel_data->table_vdw->data;
121 vftabscale = gmx_fjsp_set1_v2r8(kernel_data->table_vdw->scale);
123 invsqrta = fr->invsqrta;
125 gbtabscale = gmx_fjsp_set1_v2r8(fr->gbtab->scale);
126 gbtab = fr->gbtab->data;
127 gbinvepsdiff = gmx_fjsp_set1_v2r8((1.0/fr->ic->epsilon_r) - (1.0/fr->gb_epsilon_solvent));
129 /* Avoid stupid compiler warnings */
137 /* Start outer loop over neighborlists */
138 for(iidx=0; iidx<nri; iidx++)
140 /* Load shift vector for this list */
141 i_shift_offset = DIM*shiftidx[iidx];
143 /* Load limits for loop over neighbors */
144 j_index_start = jindex[iidx];
145 j_index_end = jindex[iidx+1];
147 /* Get outer coordinate index */
149 i_coord_offset = DIM*inr;
151 /* Load i particle coords and add shift vector */
152 gmx_fjsp_load_shift_and_1rvec_broadcast_v2r8(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
154 fix0 = _fjsp_setzero_v2r8();
155 fiy0 = _fjsp_setzero_v2r8();
156 fiz0 = _fjsp_setzero_v2r8();
158 /* Load parameters for i particles */
159 iq0 = _fjsp_mul_v2r8(facel,gmx_fjsp_load1_v2r8(charge+inr+0));
160 isai0 = gmx_fjsp_load1_v2r8(invsqrta+inr+0);
161 vdwioffset0 = 2*nvdwtype*vdwtype[inr+0];
163 /* Reset potential sums */
164 velecsum = _fjsp_setzero_v2r8();
165 vgbsum = _fjsp_setzero_v2r8();
166 vvdwsum = _fjsp_setzero_v2r8();
167 dvdasum = _fjsp_setzero_v2r8();
169 /* Start inner kernel loop */
170 for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
173 /* Get j neighbor index, and coordinate index */
176 j_coord_offsetA = DIM*jnrA;
177 j_coord_offsetB = DIM*jnrB;
179 /* load j atom coordinates */
180 gmx_fjsp_load_1rvec_2ptr_swizzle_v2r8(x+j_coord_offsetA,x+j_coord_offsetB,
183 /* Calculate displacement vector */
184 dx00 = _fjsp_sub_v2r8(ix0,jx0);
185 dy00 = _fjsp_sub_v2r8(iy0,jy0);
186 dz00 = _fjsp_sub_v2r8(iz0,jz0);
188 /* Calculate squared distance and things based on it */
189 rsq00 = gmx_fjsp_calc_rsq_v2r8(dx00,dy00,dz00);
191 rinv00 = gmx_fjsp_invsqrt_v2r8(rsq00);
193 /* Load parameters for j particles */
194 jq0 = gmx_fjsp_load_2real_swizzle_v2r8(charge+jnrA+0,charge+jnrB+0);
195 isaj0 = gmx_fjsp_load_2real_swizzle_v2r8(invsqrta+jnrA+0,invsqrta+jnrB+0);
196 vdwjidx0A = 2*vdwtype[jnrA+0];
197 vdwjidx0B = 2*vdwtype[jnrB+0];
199 /**************************
200 * CALCULATE INTERACTIONS *
201 **************************/
203 r00 = _fjsp_mul_v2r8(rsq00,rinv00);
205 /* Compute parameters for interactions between i and j atoms */
206 qq00 = _fjsp_mul_v2r8(iq0,jq0);
207 gmx_fjsp_load_2pair_swizzle_v2r8(vdwparam+vdwioffset0+vdwjidx0A,
208 vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
210 /* Calculate table index by multiplying r with table scale and truncate to integer */
211 rt = _fjsp_mul_v2r8(r00,vftabscale);
212 itab_tmp = _fjsp_dtox_v2r8(rt);
213 vfeps = _fjsp_sub_v2r8(rt, _fjsp_xtod_v2r8(itab_tmp));
214 twovfeps = _fjsp_add_v2r8(vfeps,vfeps);
215 _fjsp_store_v2r8(&vfconv.simd,itab_tmp);
220 /* GENERALIZED BORN AND COULOMB ELECTROSTATICS */
221 isaprod = _fjsp_mul_v2r8(isai0,isaj0);
222 gbqqfactor = _fjsp_neg_v2r8(_fjsp_mul_v2r8(qq00,_fjsp_mul_v2r8(isaprod,gbinvepsdiff)));
223 gbscale = _fjsp_mul_v2r8(isaprod,gbtabscale);
225 /* Calculate generalized born table index - this is a separate table from the normal one,
226 * but we use the same procedure by multiplying r with scale and truncating to integer.
228 rt = _fjsp_mul_v2r8(r00,gbscale);
229 itab_tmp = _fjsp_dtox_v2r8(rt);
230 gbeps = _fjsp_sub_v2r8(rt,_fjsp_xtod_v2r8(itab_tmp));
231 _fjsp_store_v2r8(&gbconv.simd,itab_tmp);
233 Y = _fjsp_load_v2r8( gbtab + 4*gbconv.i[0] );
234 F = _fjsp_load_v2r8( gbtab + 4*gbconv.i[1] );
235 GMX_FJSP_TRANSPOSE2_V2R8(Y,F);
236 G = _fjsp_load_v2r8( gbtab + 4*gbconv.i[0] +2);
237 H = _fjsp_load_v2r8( gbtab + 4*gbconv.i[1] +2);
238 GMX_FJSP_TRANSPOSE2_V2R8(G,H);
239 Fp = _fjsp_madd_v2r8(gbeps,_fjsp_madd_v2r8(gbeps,H,G),F);
240 VV = _fjsp_madd_v2r8(gbeps,Fp,Y);
241 vgb = _fjsp_mul_v2r8(gbqqfactor,VV);
243 twogbeps = _fjsp_add_v2r8(gbeps,gbeps);
244 FF = _fjsp_madd_v2r8(_fjsp_madd_v2r8(twogbeps,H,G),gbeps,Fp);
245 fgb = _fjsp_mul_v2r8(gbqqfactor,_fjsp_mul_v2r8(FF,gbscale));
246 dvdatmp = _fjsp_mul_v2r8(minushalf,_fjsp_madd_v2r8(fgb,r00,vgb));
247 dvdasum = _fjsp_add_v2r8(dvdasum,dvdatmp);
248 gmx_fjsp_increment_2real_swizzle_v2r8(dvda+jnrA,dvda+jnrB,_fjsp_mul_v2r8(dvdatmp,_fjsp_mul_v2r8(isaj0,isaj0)));
249 velec = _fjsp_mul_v2r8(qq00,rinv00);
250 felec = _fjsp_mul_v2r8(_fjsp_msub_v2r8(velec,rinv00,fgb),rinv00);
252 /* CUBIC SPLINE TABLE DISPERSION */
253 Y = _fjsp_load_v2r8( vftab + vfconv.i[0] );
254 F = _fjsp_load_v2r8( vftab + vfconv.i[1] );
255 GMX_FJSP_TRANSPOSE2_V2R8(Y,F);
256 G = _fjsp_load_v2r8( vftab + vfconv.i[0] + 2 );
257 H = _fjsp_load_v2r8( vftab + vfconv.i[1] + 2 );
258 GMX_FJSP_TRANSPOSE2_V2R8(G,H);
259 Fp = _fjsp_madd_v2r8(vfeps,_fjsp_madd_v2r8(H,vfeps,G),F);
260 VV = _fjsp_madd_v2r8(vfeps,Fp,Y);
261 vvdw6 = _fjsp_mul_v2r8(c6_00,VV);
262 FF = _fjsp_madd_v2r8(vfeps,_fjsp_madd_v2r8(twovfeps,H,G),Fp);
263 fvdw6 = _fjsp_mul_v2r8(c6_00,FF);
265 /* CUBIC SPLINE TABLE REPULSION */
266 Y = _fjsp_load_v2r8( vftab + vfconv.i[0] + 4 );
267 F = _fjsp_load_v2r8( vftab + vfconv.i[1] + 4 );
268 GMX_FJSP_TRANSPOSE2_V2R8(Y,F);
269 G = _fjsp_load_v2r8( vftab + vfconv.i[0] + 6 );
270 H = _fjsp_load_v2r8( vftab + vfconv.i[1] + 6 );
271 GMX_FJSP_TRANSPOSE2_V2R8(G,H);
272 Fp = _fjsp_madd_v2r8(vfeps,_fjsp_madd_v2r8(H,vfeps,G),F);
273 VV = _fjsp_madd_v2r8(vfeps,Fp,Y);
274 vvdw12 = _fjsp_mul_v2r8(c12_00,VV);
275 FF = _fjsp_madd_v2r8(vfeps,_fjsp_madd_v2r8(twovfeps,H,G),Fp);
276 fvdw12 = _fjsp_mul_v2r8(c12_00,FF);
277 vvdw = _fjsp_add_v2r8(vvdw12,vvdw6);
278 fvdw = _fjsp_neg_v2r8(_fjsp_mul_v2r8(_fjsp_add_v2r8(fvdw6,fvdw12),_fjsp_mul_v2r8(vftabscale,rinv00)));
280 /* Update potential sum for this i atom from the interaction with this j atom. */
281 velecsum = _fjsp_add_v2r8(velecsum,velec);
282 vgbsum = _fjsp_add_v2r8(vgbsum,vgb);
283 vvdwsum = _fjsp_add_v2r8(vvdwsum,vvdw);
285 fscal = _fjsp_add_v2r8(felec,fvdw);
287 /* Update vectorial force */
288 fix0 = _fjsp_madd_v2r8(dx00,fscal,fix0);
289 fiy0 = _fjsp_madd_v2r8(dy00,fscal,fiy0);
290 fiz0 = _fjsp_madd_v2r8(dz00,fscal,fiz0);
292 gmx_fjsp_decrement_fma_1rvec_2ptr_swizzle_v2r8(f+j_coord_offsetA,f+j_coord_offsetB,fscal,dx00,dy00,dz00);
294 /* Inner loop uses 95 flops */
301 j_coord_offsetA = DIM*jnrA;
303 /* load j atom coordinates */
304 gmx_fjsp_load_1rvec_1ptr_swizzle_v2r8(x+j_coord_offsetA,
307 /* Calculate displacement vector */
308 dx00 = _fjsp_sub_v2r8(ix0,jx0);
309 dy00 = _fjsp_sub_v2r8(iy0,jy0);
310 dz00 = _fjsp_sub_v2r8(iz0,jz0);
312 /* Calculate squared distance and things based on it */
313 rsq00 = gmx_fjsp_calc_rsq_v2r8(dx00,dy00,dz00);
315 rinv00 = gmx_fjsp_invsqrt_v2r8(rsq00);
317 /* Load parameters for j particles */
318 jq0 = _fjsp_loadl_v2r8(_fjsp_setzero_v2r8(),charge+jnrA+0);
319 isaj0 = _fjsp_loadl_v2r8(_fjsp_setzero_v2r8(),invsqrta+jnrA+0);
320 vdwjidx0A = 2*vdwtype[jnrA+0];
322 /**************************
323 * CALCULATE INTERACTIONS *
324 **************************/
326 r00 = _fjsp_mul_v2r8(rsq00,rinv00);
328 /* Compute parameters for interactions between i and j atoms */
329 qq00 = _fjsp_mul_v2r8(iq0,jq0);
330 gmx_fjsp_load_2pair_swizzle_v2r8(vdwparam+vdwioffset0+vdwjidx0A,
331 vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
333 /* Calculate table index by multiplying r with table scale and truncate to integer */
334 rt = _fjsp_mul_v2r8(r00,vftabscale);
335 itab_tmp = _fjsp_dtox_v2r8(rt);
336 vfeps = _fjsp_sub_v2r8(rt, _fjsp_xtod_v2r8(itab_tmp));
337 twovfeps = _fjsp_add_v2r8(vfeps,vfeps);
338 _fjsp_store_v2r8(&vfconv.simd,itab_tmp);
343 /* GENERALIZED BORN AND COULOMB ELECTROSTATICS */
344 isaprod = _fjsp_mul_v2r8(isai0,isaj0);
345 gbqqfactor = _fjsp_neg_v2r8(_fjsp_mul_v2r8(qq00,_fjsp_mul_v2r8(isaprod,gbinvepsdiff)));
346 gbscale = _fjsp_mul_v2r8(isaprod,gbtabscale);
348 /* Calculate generalized born table index - this is a separate table from the normal one,
349 * but we use the same procedure by multiplying r with scale and truncating to integer.
351 rt = _fjsp_mul_v2r8(r00,gbscale);
352 itab_tmp = _fjsp_dtox_v2r8(rt);
353 gbeps = _fjsp_sub_v2r8(rt,_fjsp_xtod_v2r8(itab_tmp));
354 _fjsp_store_v2r8(&gbconv.simd,itab_tmp);
356 Y = _fjsp_load_v2r8( gbtab + 4*gbconv.i[0] );
357 F = _fjsp_setzero_v2r8();
358 GMX_FJSP_TRANSPOSE2_V2R8(Y,F);
359 G = _fjsp_load_v2r8( gbtab + 4*gbconv.i[0] +2);
360 H = _fjsp_setzero_v2r8();
361 GMX_FJSP_TRANSPOSE2_V2R8(G,H);
362 Fp = _fjsp_madd_v2r8(gbeps,_fjsp_madd_v2r8(gbeps,H,G),F);
363 VV = _fjsp_madd_v2r8(gbeps,Fp,Y);
364 vgb = _fjsp_mul_v2r8(gbqqfactor,VV);
366 twogbeps = _fjsp_add_v2r8(gbeps,gbeps);
367 FF = _fjsp_madd_v2r8(_fjsp_madd_v2r8(twogbeps,H,G),gbeps,Fp);
368 fgb = _fjsp_mul_v2r8(gbqqfactor,_fjsp_mul_v2r8(FF,gbscale));
369 dvdatmp = _fjsp_mul_v2r8(minushalf,_fjsp_madd_v2r8(fgb,r00,vgb));
370 dvdasum = _fjsp_add_v2r8(dvdasum,dvdatmp);
371 gmx_fjsp_increment_1real_v2r8(dvda+jnrA,_fjsp_mul_v2r8(dvdatmp,_fjsp_mul_v2r8(isaj0,isaj0)));
372 velec = _fjsp_mul_v2r8(qq00,rinv00);
373 felec = _fjsp_mul_v2r8(_fjsp_msub_v2r8(velec,rinv00,fgb),rinv00);
375 /* CUBIC SPLINE TABLE DISPERSION */
376 Y = _fjsp_load_v2r8( vftab + vfconv.i[0] );
377 F = _fjsp_setzero_v2r8();
378 GMX_FJSP_TRANSPOSE2_V2R8(Y,F);
379 G = _fjsp_load_v2r8( vftab + vfconv.i[0] + 2 );
380 H = _fjsp_setzero_v2r8();
381 GMX_FJSP_TRANSPOSE2_V2R8(G,H);
382 Fp = _fjsp_madd_v2r8(vfeps,_fjsp_madd_v2r8(H,vfeps,G),F);
383 VV = _fjsp_madd_v2r8(vfeps,Fp,Y);
384 vvdw6 = _fjsp_mul_v2r8(c6_00,VV);
385 FF = _fjsp_madd_v2r8(vfeps,_fjsp_madd_v2r8(twovfeps,H,G),Fp);
386 fvdw6 = _fjsp_mul_v2r8(c6_00,FF);
388 /* CUBIC SPLINE TABLE REPULSION */
389 Y = _fjsp_load_v2r8( vftab + vfconv.i[0] + 4 );
390 F = _fjsp_setzero_v2r8();
391 GMX_FJSP_TRANSPOSE2_V2R8(Y,F);
392 G = _fjsp_load_v2r8( vftab + vfconv.i[0] + 6 );
393 H = _fjsp_setzero_v2r8();
394 GMX_FJSP_TRANSPOSE2_V2R8(G,H);
395 Fp = _fjsp_madd_v2r8(vfeps,_fjsp_madd_v2r8(H,vfeps,G),F);
396 VV = _fjsp_madd_v2r8(vfeps,Fp,Y);
397 vvdw12 = _fjsp_mul_v2r8(c12_00,VV);
398 FF = _fjsp_madd_v2r8(vfeps,_fjsp_madd_v2r8(twovfeps,H,G),Fp);
399 fvdw12 = _fjsp_mul_v2r8(c12_00,FF);
400 vvdw = _fjsp_add_v2r8(vvdw12,vvdw6);
401 fvdw = _fjsp_neg_v2r8(_fjsp_mul_v2r8(_fjsp_add_v2r8(fvdw6,fvdw12),_fjsp_mul_v2r8(vftabscale,rinv00)));
403 /* Update potential sum for this i atom from the interaction with this j atom. */
404 velec = _fjsp_unpacklo_v2r8(velec,_fjsp_setzero_v2r8());
405 velecsum = _fjsp_add_v2r8(velecsum,velec);
406 vgb = _fjsp_unpacklo_v2r8(vgb,_fjsp_setzero_v2r8());
407 vgbsum = _fjsp_add_v2r8(vgbsum,vgb);
408 vvdw = _fjsp_unpacklo_v2r8(vvdw,_fjsp_setzero_v2r8());
409 vvdwsum = _fjsp_add_v2r8(vvdwsum,vvdw);
411 fscal = _fjsp_add_v2r8(felec,fvdw);
413 fscal = _fjsp_unpacklo_v2r8(fscal,_fjsp_setzero_v2r8());
415 /* Update vectorial force */
416 fix0 = _fjsp_madd_v2r8(dx00,fscal,fix0);
417 fiy0 = _fjsp_madd_v2r8(dy00,fscal,fiy0);
418 fiz0 = _fjsp_madd_v2r8(dz00,fscal,fiz0);
420 gmx_fjsp_decrement_fma_1rvec_1ptr_swizzle_v2r8(f+j_coord_offsetA,fscal,dx00,dy00,dz00);
422 /* Inner loop uses 95 flops */
425 /* End of innermost loop */
427 gmx_fjsp_update_iforce_1atom_swizzle_v2r8(fix0,fiy0,fiz0,
428 f+i_coord_offset,fshift+i_shift_offset);
431 /* Update potential energies */
432 gmx_fjsp_update_1pot_v2r8(velecsum,kernel_data->energygrp_elec+ggid);
433 gmx_fjsp_update_1pot_v2r8(vgbsum,kernel_data->energygrp_polarization+ggid);
434 gmx_fjsp_update_1pot_v2r8(vvdwsum,kernel_data->energygrp_vdw+ggid);
435 dvdasum = _fjsp_mul_v2r8(dvdasum, _fjsp_mul_v2r8(isai0,isai0));
436 gmx_fjsp_update_1pot_v2r8(dvdasum,dvda+inr);
438 /* Increment number of inner iterations */
439 inneriter += j_index_end - j_index_start;
441 /* Outer loop uses 10 flops */
444 /* Increment number of outer iterations */
447 /* Update outer/inner flops */
449 inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_VF,outeriter*10 + inneriter*95);
452 * Gromacs nonbonded kernel: nb_kernel_ElecGB_VdwCSTab_GeomP1P1_F_sparc64_hpc_ace_double
453 * Electrostatics interaction: GeneralizedBorn
454 * VdW interaction: CubicSplineTable
455 * Geometry: Particle-Particle
456 * Calculate force/pot: Force
459 nb_kernel_ElecGB_VdwCSTab_GeomP1P1_F_sparc64_hpc_ace_double
460 (t_nblist * gmx_restrict nlist,
461 rvec * gmx_restrict xx,
462 rvec * gmx_restrict ff,
463 struct t_forcerec * gmx_restrict fr,
464 t_mdatoms * gmx_restrict mdatoms,
465 nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
466 t_nrnb * gmx_restrict nrnb)
468 /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
469 * just 0 for non-waters.
470 * Suffixes A,B refer to j loop unrolling done with double precision SIMD, e.g. for the two different
471 * jnr indices corresponding to data put in the four positions in the SIMD register.
473 int i_shift_offset,i_coord_offset,outeriter,inneriter;
474 int j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
476 int j_coord_offsetA,j_coord_offsetB;
477 int *iinr,*jindex,*jjnr,*shiftidx,*gid;
479 real *shiftvec,*fshift,*x,*f;
480 _fjsp_v2r8 tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
482 _fjsp_v2r8 ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
483 int vdwjidx0A,vdwjidx0B;
484 _fjsp_v2r8 jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
485 _fjsp_v2r8 dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
486 _fjsp_v2r8 velec,felec,velecsum,facel,crf,krf,krf2;
488 _fjsp_v2r8 vgb,fgb,vgbsum,dvdasum,gbscale,gbtabscale,isaprod,gbqqfactor,gbinvepsdiff,dvdaj,gbeps,twogbeps,dvdatmp;
489 _fjsp_v2r8 minushalf = gmx_fjsp_set1_v2r8(-0.5);
490 real *invsqrta,*dvda,*gbtab;
492 _fjsp_v2r8 rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
495 _fjsp_v2r8 one_sixth = gmx_fjsp_set1_v2r8(1.0/6.0);
496 _fjsp_v2r8 one_twelfth = gmx_fjsp_set1_v2r8(1.0/12.0);
497 _fjsp_v2r8 rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF,twovfeps;
500 _fjsp_v2r8 dummy_mask,cutoff_mask;
501 _fjsp_v2r8 one = gmx_fjsp_set1_v2r8(1.0);
502 _fjsp_v2r8 two = gmx_fjsp_set1_v2r8(2.0);
503 union { _fjsp_v2r8 simd; long long int i[2]; } vfconv,gbconv,ewconv;
510 jindex = nlist->jindex;
512 shiftidx = nlist->shift;
514 shiftvec = fr->shift_vec[0];
515 fshift = fr->fshift[0];
516 facel = gmx_fjsp_set1_v2r8(fr->ic->epsfac);
517 charge = mdatoms->chargeA;
518 nvdwtype = fr->ntype;
520 vdwtype = mdatoms->typeA;
522 vftab = kernel_data->table_vdw->data;
523 vftabscale = gmx_fjsp_set1_v2r8(kernel_data->table_vdw->scale);
525 invsqrta = fr->invsqrta;
527 gbtabscale = gmx_fjsp_set1_v2r8(fr->gbtab->scale);
528 gbtab = fr->gbtab->data;
529 gbinvepsdiff = gmx_fjsp_set1_v2r8((1.0/fr->ic->epsilon_r) - (1.0/fr->gb_epsilon_solvent));
531 /* Avoid stupid compiler warnings */
539 /* Start outer loop over neighborlists */
540 for(iidx=0; iidx<nri; iidx++)
542 /* Load shift vector for this list */
543 i_shift_offset = DIM*shiftidx[iidx];
545 /* Load limits for loop over neighbors */
546 j_index_start = jindex[iidx];
547 j_index_end = jindex[iidx+1];
549 /* Get outer coordinate index */
551 i_coord_offset = DIM*inr;
553 /* Load i particle coords and add shift vector */
554 gmx_fjsp_load_shift_and_1rvec_broadcast_v2r8(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
556 fix0 = _fjsp_setzero_v2r8();
557 fiy0 = _fjsp_setzero_v2r8();
558 fiz0 = _fjsp_setzero_v2r8();
560 /* Load parameters for i particles */
561 iq0 = _fjsp_mul_v2r8(facel,gmx_fjsp_load1_v2r8(charge+inr+0));
562 isai0 = gmx_fjsp_load1_v2r8(invsqrta+inr+0);
563 vdwioffset0 = 2*nvdwtype*vdwtype[inr+0];
565 dvdasum = _fjsp_setzero_v2r8();
567 /* Start inner kernel loop */
568 for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
571 /* Get j neighbor index, and coordinate index */
574 j_coord_offsetA = DIM*jnrA;
575 j_coord_offsetB = DIM*jnrB;
577 /* load j atom coordinates */
578 gmx_fjsp_load_1rvec_2ptr_swizzle_v2r8(x+j_coord_offsetA,x+j_coord_offsetB,
581 /* Calculate displacement vector */
582 dx00 = _fjsp_sub_v2r8(ix0,jx0);
583 dy00 = _fjsp_sub_v2r8(iy0,jy0);
584 dz00 = _fjsp_sub_v2r8(iz0,jz0);
586 /* Calculate squared distance and things based on it */
587 rsq00 = gmx_fjsp_calc_rsq_v2r8(dx00,dy00,dz00);
589 rinv00 = gmx_fjsp_invsqrt_v2r8(rsq00);
591 /* Load parameters for j particles */
592 jq0 = gmx_fjsp_load_2real_swizzle_v2r8(charge+jnrA+0,charge+jnrB+0);
593 isaj0 = gmx_fjsp_load_2real_swizzle_v2r8(invsqrta+jnrA+0,invsqrta+jnrB+0);
594 vdwjidx0A = 2*vdwtype[jnrA+0];
595 vdwjidx0B = 2*vdwtype[jnrB+0];
597 /**************************
598 * CALCULATE INTERACTIONS *
599 **************************/
601 r00 = _fjsp_mul_v2r8(rsq00,rinv00);
603 /* Compute parameters for interactions between i and j atoms */
604 qq00 = _fjsp_mul_v2r8(iq0,jq0);
605 gmx_fjsp_load_2pair_swizzle_v2r8(vdwparam+vdwioffset0+vdwjidx0A,
606 vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
608 /* Calculate table index by multiplying r with table scale and truncate to integer */
609 rt = _fjsp_mul_v2r8(r00,vftabscale);
610 itab_tmp = _fjsp_dtox_v2r8(rt);
611 vfeps = _fjsp_sub_v2r8(rt, _fjsp_xtod_v2r8(itab_tmp));
612 twovfeps = _fjsp_add_v2r8(vfeps,vfeps);
613 _fjsp_store_v2r8(&vfconv.simd,itab_tmp);
618 /* GENERALIZED BORN AND COULOMB ELECTROSTATICS */
619 isaprod = _fjsp_mul_v2r8(isai0,isaj0);
620 gbqqfactor = _fjsp_neg_v2r8(_fjsp_mul_v2r8(qq00,_fjsp_mul_v2r8(isaprod,gbinvepsdiff)));
621 gbscale = _fjsp_mul_v2r8(isaprod,gbtabscale);
623 /* Calculate generalized born table index - this is a separate table from the normal one,
624 * but we use the same procedure by multiplying r with scale and truncating to integer.
626 rt = _fjsp_mul_v2r8(r00,gbscale);
627 itab_tmp = _fjsp_dtox_v2r8(rt);
628 gbeps = _fjsp_sub_v2r8(rt,_fjsp_xtod_v2r8(itab_tmp));
629 _fjsp_store_v2r8(&gbconv.simd,itab_tmp);
631 Y = _fjsp_load_v2r8( gbtab + 4*gbconv.i[0] );
632 F = _fjsp_load_v2r8( gbtab + 4*gbconv.i[1] );
633 GMX_FJSP_TRANSPOSE2_V2R8(Y,F);
634 G = _fjsp_load_v2r8( gbtab + 4*gbconv.i[0] +2);
635 H = _fjsp_load_v2r8( gbtab + 4*gbconv.i[1] +2);
636 GMX_FJSP_TRANSPOSE2_V2R8(G,H);
637 Fp = _fjsp_madd_v2r8(gbeps,_fjsp_madd_v2r8(gbeps,H,G),F);
638 VV = _fjsp_madd_v2r8(gbeps,Fp,Y);
639 vgb = _fjsp_mul_v2r8(gbqqfactor,VV);
641 twogbeps = _fjsp_add_v2r8(gbeps,gbeps);
642 FF = _fjsp_madd_v2r8(_fjsp_madd_v2r8(twogbeps,H,G),gbeps,Fp);
643 fgb = _fjsp_mul_v2r8(gbqqfactor,_fjsp_mul_v2r8(FF,gbscale));
644 dvdatmp = _fjsp_mul_v2r8(minushalf,_fjsp_madd_v2r8(fgb,r00,vgb));
645 dvdasum = _fjsp_add_v2r8(dvdasum,dvdatmp);
646 gmx_fjsp_increment_2real_swizzle_v2r8(dvda+jnrA,dvda+jnrB,_fjsp_mul_v2r8(dvdatmp,_fjsp_mul_v2r8(isaj0,isaj0)));
647 velec = _fjsp_mul_v2r8(qq00,rinv00);
648 felec = _fjsp_mul_v2r8(_fjsp_msub_v2r8(velec,rinv00,fgb),rinv00);
650 /* CUBIC SPLINE TABLE DISPERSION */
651 Y = _fjsp_load_v2r8( vftab + vfconv.i[0] );
652 F = _fjsp_load_v2r8( vftab + vfconv.i[1] );
653 GMX_FJSP_TRANSPOSE2_V2R8(Y,F);
654 G = _fjsp_load_v2r8( vftab + vfconv.i[0] + 2 );
655 H = _fjsp_load_v2r8( vftab + vfconv.i[1] + 2 );
656 GMX_FJSP_TRANSPOSE2_V2R8(G,H);
657 Fp = _fjsp_madd_v2r8(vfeps,_fjsp_madd_v2r8(H,vfeps,G),F);
658 FF = _fjsp_madd_v2r8(vfeps,_fjsp_madd_v2r8(twovfeps,H,G),Fp);
659 fvdw6 = _fjsp_mul_v2r8(c6_00,FF);
661 /* CUBIC SPLINE TABLE REPULSION */
662 Y = _fjsp_load_v2r8( vftab + vfconv.i[0] + 4 );
663 F = _fjsp_load_v2r8( vftab + vfconv.i[1] + 4 );
664 GMX_FJSP_TRANSPOSE2_V2R8(Y,F);
665 G = _fjsp_load_v2r8( vftab + vfconv.i[0] + 6 );
666 H = _fjsp_load_v2r8( vftab + vfconv.i[1] + 6 );
667 GMX_FJSP_TRANSPOSE2_V2R8(G,H);
668 Fp = _fjsp_madd_v2r8(vfeps,_fjsp_madd_v2r8(H,vfeps,G),F);
669 FF = _fjsp_madd_v2r8(vfeps,_fjsp_madd_v2r8(twovfeps,H,G),Fp);
670 fvdw12 = _fjsp_mul_v2r8(c12_00,FF);
671 fvdw = _fjsp_neg_v2r8(_fjsp_mul_v2r8(_fjsp_add_v2r8(fvdw6,fvdw12),_fjsp_mul_v2r8(vftabscale,rinv00)));
673 fscal = _fjsp_add_v2r8(felec,fvdw);
675 /* Update vectorial force */
676 fix0 = _fjsp_madd_v2r8(dx00,fscal,fix0);
677 fiy0 = _fjsp_madd_v2r8(dy00,fscal,fiy0);
678 fiz0 = _fjsp_madd_v2r8(dz00,fscal,fiz0);
680 gmx_fjsp_decrement_fma_1rvec_2ptr_swizzle_v2r8(f+j_coord_offsetA,f+j_coord_offsetB,fscal,dx00,dy00,dz00);
682 /* Inner loop uses 85 flops */
689 j_coord_offsetA = DIM*jnrA;
691 /* load j atom coordinates */
692 gmx_fjsp_load_1rvec_1ptr_swizzle_v2r8(x+j_coord_offsetA,
695 /* Calculate displacement vector */
696 dx00 = _fjsp_sub_v2r8(ix0,jx0);
697 dy00 = _fjsp_sub_v2r8(iy0,jy0);
698 dz00 = _fjsp_sub_v2r8(iz0,jz0);
700 /* Calculate squared distance and things based on it */
701 rsq00 = gmx_fjsp_calc_rsq_v2r8(dx00,dy00,dz00);
703 rinv00 = gmx_fjsp_invsqrt_v2r8(rsq00);
705 /* Load parameters for j particles */
706 jq0 = _fjsp_loadl_v2r8(_fjsp_setzero_v2r8(),charge+jnrA+0);
707 isaj0 = _fjsp_loadl_v2r8(_fjsp_setzero_v2r8(),invsqrta+jnrA+0);
708 vdwjidx0A = 2*vdwtype[jnrA+0];
710 /**************************
711 * CALCULATE INTERACTIONS *
712 **************************/
714 r00 = _fjsp_mul_v2r8(rsq00,rinv00);
716 /* Compute parameters for interactions between i and j atoms */
717 qq00 = _fjsp_mul_v2r8(iq0,jq0);
718 gmx_fjsp_load_2pair_swizzle_v2r8(vdwparam+vdwioffset0+vdwjidx0A,
719 vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
721 /* Calculate table index by multiplying r with table scale and truncate to integer */
722 rt = _fjsp_mul_v2r8(r00,vftabscale);
723 itab_tmp = _fjsp_dtox_v2r8(rt);
724 vfeps = _fjsp_sub_v2r8(rt, _fjsp_xtod_v2r8(itab_tmp));
725 twovfeps = _fjsp_add_v2r8(vfeps,vfeps);
726 _fjsp_store_v2r8(&vfconv.simd,itab_tmp);
731 /* GENERALIZED BORN AND COULOMB ELECTROSTATICS */
732 isaprod = _fjsp_mul_v2r8(isai0,isaj0);
733 gbqqfactor = _fjsp_neg_v2r8(_fjsp_mul_v2r8(qq00,_fjsp_mul_v2r8(isaprod,gbinvepsdiff)));
734 gbscale = _fjsp_mul_v2r8(isaprod,gbtabscale);
736 /* Calculate generalized born table index - this is a separate table from the normal one,
737 * but we use the same procedure by multiplying r with scale and truncating to integer.
739 rt = _fjsp_mul_v2r8(r00,gbscale);
740 itab_tmp = _fjsp_dtox_v2r8(rt);
741 gbeps = _fjsp_sub_v2r8(rt,_fjsp_xtod_v2r8(itab_tmp));
742 _fjsp_store_v2r8(&gbconv.simd,itab_tmp);
744 Y = _fjsp_load_v2r8( gbtab + 4*gbconv.i[0] );
745 F = _fjsp_setzero_v2r8();
746 GMX_FJSP_TRANSPOSE2_V2R8(Y,F);
747 G = _fjsp_load_v2r8( gbtab + 4*gbconv.i[0] +2);
748 H = _fjsp_setzero_v2r8();
749 GMX_FJSP_TRANSPOSE2_V2R8(G,H);
750 Fp = _fjsp_madd_v2r8(gbeps,_fjsp_madd_v2r8(gbeps,H,G),F);
751 VV = _fjsp_madd_v2r8(gbeps,Fp,Y);
752 vgb = _fjsp_mul_v2r8(gbqqfactor,VV);
754 twogbeps = _fjsp_add_v2r8(gbeps,gbeps);
755 FF = _fjsp_madd_v2r8(_fjsp_madd_v2r8(twogbeps,H,G),gbeps,Fp);
756 fgb = _fjsp_mul_v2r8(gbqqfactor,_fjsp_mul_v2r8(FF,gbscale));
757 dvdatmp = _fjsp_mul_v2r8(minushalf,_fjsp_madd_v2r8(fgb,r00,vgb));
758 dvdasum = _fjsp_add_v2r8(dvdasum,dvdatmp);
759 gmx_fjsp_increment_1real_v2r8(dvda+jnrA,_fjsp_mul_v2r8(dvdatmp,_fjsp_mul_v2r8(isaj0,isaj0)));
760 velec = _fjsp_mul_v2r8(qq00,rinv00);
761 felec = _fjsp_mul_v2r8(_fjsp_msub_v2r8(velec,rinv00,fgb),rinv00);
763 /* CUBIC SPLINE TABLE DISPERSION */
764 Y = _fjsp_load_v2r8( vftab + vfconv.i[0] );
765 F = _fjsp_setzero_v2r8();
766 GMX_FJSP_TRANSPOSE2_V2R8(Y,F);
767 G = _fjsp_load_v2r8( vftab + vfconv.i[0] + 2 );
768 H = _fjsp_setzero_v2r8();
769 GMX_FJSP_TRANSPOSE2_V2R8(G,H);
770 Fp = _fjsp_madd_v2r8(vfeps,_fjsp_madd_v2r8(H,vfeps,G),F);
771 FF = _fjsp_madd_v2r8(vfeps,_fjsp_madd_v2r8(twovfeps,H,G),Fp);
772 fvdw6 = _fjsp_mul_v2r8(c6_00,FF);
774 /* CUBIC SPLINE TABLE REPULSION */
775 Y = _fjsp_load_v2r8( vftab + vfconv.i[0] + 4 );
776 F = _fjsp_setzero_v2r8();
777 GMX_FJSP_TRANSPOSE2_V2R8(Y,F);
778 G = _fjsp_load_v2r8( vftab + vfconv.i[0] + 6 );
779 H = _fjsp_setzero_v2r8();
780 GMX_FJSP_TRANSPOSE2_V2R8(G,H);
781 Fp = _fjsp_madd_v2r8(vfeps,_fjsp_madd_v2r8(H,vfeps,G),F);
782 FF = _fjsp_madd_v2r8(vfeps,_fjsp_madd_v2r8(twovfeps,H,G),Fp);
783 fvdw12 = _fjsp_mul_v2r8(c12_00,FF);
784 fvdw = _fjsp_neg_v2r8(_fjsp_mul_v2r8(_fjsp_add_v2r8(fvdw6,fvdw12),_fjsp_mul_v2r8(vftabscale,rinv00)));
786 fscal = _fjsp_add_v2r8(felec,fvdw);
788 fscal = _fjsp_unpacklo_v2r8(fscal,_fjsp_setzero_v2r8());
790 /* Update vectorial force */
791 fix0 = _fjsp_madd_v2r8(dx00,fscal,fix0);
792 fiy0 = _fjsp_madd_v2r8(dy00,fscal,fiy0);
793 fiz0 = _fjsp_madd_v2r8(dz00,fscal,fiz0);
795 gmx_fjsp_decrement_fma_1rvec_1ptr_swizzle_v2r8(f+j_coord_offsetA,fscal,dx00,dy00,dz00);
797 /* Inner loop uses 85 flops */
800 /* End of innermost loop */
802 gmx_fjsp_update_iforce_1atom_swizzle_v2r8(fix0,fiy0,fiz0,
803 f+i_coord_offset,fshift+i_shift_offset);
805 dvdasum = _fjsp_mul_v2r8(dvdasum, _fjsp_mul_v2r8(isai0,isai0));
806 gmx_fjsp_update_1pot_v2r8(dvdasum,dvda+inr);
808 /* Increment number of inner iterations */
809 inneriter += j_index_end - j_index_start;
811 /* Outer loop uses 7 flops */
814 /* Increment number of outer iterations */
817 /* Update outer/inner flops */
819 inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_F,outeriter*7 + inneriter*85);