2 * This file is part of the GROMACS molecular simulation package.
4 * Copyright (c) 2012,2013, 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 "types/simple.h"
49 #include "kernelutil_sparc64_hpc_ace_double.h"
52 * Gromacs nonbonded kernel: nb_kernel_ElecEwSw_VdwLJSw_GeomP1P1_VF_sparc64_hpc_ace_double
53 * Electrostatics interaction: Ewald
54 * VdW interaction: LennardJones
55 * Geometry: Particle-Particle
56 * Calculate force/pot: PotentialAndForce
59 nb_kernel_ElecEwSw_VdwLJSw_GeomP1P1_VF_sparc64_hpc_ace_double
60 (t_nblist * gmx_restrict nlist,
61 rvec * gmx_restrict xx,
62 rvec * gmx_restrict ff,
63 t_forcerec * gmx_restrict fr,
64 t_mdatoms * gmx_restrict mdatoms,
65 nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
66 t_nrnb * gmx_restrict nrnb)
68 /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
69 * just 0 for non-waters.
70 * Suffixes A,B refer to j loop unrolling done with double precision SIMD, e.g. for the two different
71 * jnr indices corresponding to data put in the four positions in the SIMD register.
73 int i_shift_offset,i_coord_offset,outeriter,inneriter;
74 int j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
76 int j_coord_offsetA,j_coord_offsetB;
77 int *iinr,*jindex,*jjnr,*shiftidx,*gid;
79 real *shiftvec,*fshift,*x,*f;
80 _fjsp_v2r8 tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
82 _fjsp_v2r8 ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
83 int vdwjidx0A,vdwjidx0B;
84 _fjsp_v2r8 jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
85 _fjsp_v2r8 dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
86 _fjsp_v2r8 velec,felec,velecsum,facel,crf,krf,krf2;
89 _fjsp_v2r8 rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
92 _fjsp_v2r8 one_sixth = gmx_fjsp_set1_v2r8(1.0/6.0);
93 _fjsp_v2r8 one_twelfth = gmx_fjsp_set1_v2r8(1.0/12.0);
94 _fjsp_v2r8 ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
96 _fjsp_v2r8 rswitch,swV3,swV4,swV5,swF2,swF3,swF4,d,d2,sw,dsw;
97 real rswitch_scalar,d_scalar;
99 _fjsp_v2r8 dummy_mask,cutoff_mask;
100 _fjsp_v2r8 one = gmx_fjsp_set1_v2r8(1.0);
101 _fjsp_v2r8 two = gmx_fjsp_set1_v2r8(2.0);
102 union { _fjsp_v2r8 simd; long long int i[2]; } vfconv,gbconv,ewconv;
109 jindex = nlist->jindex;
111 shiftidx = nlist->shift;
113 shiftvec = fr->shift_vec[0];
114 fshift = fr->fshift[0];
115 facel = gmx_fjsp_set1_v2r8(fr->epsfac);
116 charge = mdatoms->chargeA;
117 nvdwtype = fr->ntype;
119 vdwtype = mdatoms->typeA;
121 sh_ewald = gmx_fjsp_set1_v2r8(fr->ic->sh_ewald);
122 ewtab = fr->ic->tabq_coul_FDV0;
123 ewtabscale = gmx_fjsp_set1_v2r8(fr->ic->tabq_scale);
124 ewtabhalfspace = gmx_fjsp_set1_v2r8(0.5/fr->ic->tabq_scale);
126 /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
127 rcutoff_scalar = fr->rcoulomb;
128 rcutoff = gmx_fjsp_set1_v2r8(rcutoff_scalar);
129 rcutoff2 = _fjsp_mul_v2r8(rcutoff,rcutoff);
131 rswitch_scalar = fr->rcoulomb_switch;
132 rswitch = gmx_fjsp_set1_v2r8(rswitch_scalar);
133 /* Setup switch parameters */
134 d_scalar = rcutoff_scalar-rswitch_scalar;
135 d = gmx_fjsp_set1_v2r8(d_scalar);
136 swV3 = gmx_fjsp_set1_v2r8(-10.0/(d_scalar*d_scalar*d_scalar));
137 swV4 = gmx_fjsp_set1_v2r8( 15.0/(d_scalar*d_scalar*d_scalar*d_scalar));
138 swV5 = gmx_fjsp_set1_v2r8( -6.0/(d_scalar*d_scalar*d_scalar*d_scalar*d_scalar));
139 swF2 = gmx_fjsp_set1_v2r8(-30.0/(d_scalar*d_scalar*d_scalar));
140 swF3 = gmx_fjsp_set1_v2r8( 60.0/(d_scalar*d_scalar*d_scalar*d_scalar));
141 swF4 = gmx_fjsp_set1_v2r8(-30.0/(d_scalar*d_scalar*d_scalar*d_scalar*d_scalar));
143 /* Avoid stupid compiler warnings */
151 /* Start outer loop over neighborlists */
152 for(iidx=0; iidx<nri; iidx++)
154 /* Load shift vector for this list */
155 i_shift_offset = DIM*shiftidx[iidx];
157 /* Load limits for loop over neighbors */
158 j_index_start = jindex[iidx];
159 j_index_end = jindex[iidx+1];
161 /* Get outer coordinate index */
163 i_coord_offset = DIM*inr;
165 /* Load i particle coords and add shift vector */
166 gmx_fjsp_load_shift_and_1rvec_broadcast_v2r8(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
168 fix0 = _fjsp_setzero_v2r8();
169 fiy0 = _fjsp_setzero_v2r8();
170 fiz0 = _fjsp_setzero_v2r8();
172 /* Load parameters for i particles */
173 iq0 = _fjsp_mul_v2r8(facel,gmx_fjsp_load1_v2r8(charge+inr+0));
174 vdwioffset0 = 2*nvdwtype*vdwtype[inr+0];
176 /* Reset potential sums */
177 velecsum = _fjsp_setzero_v2r8();
178 vvdwsum = _fjsp_setzero_v2r8();
180 /* Start inner kernel loop */
181 for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
184 /* Get j neighbor index, and coordinate index */
187 j_coord_offsetA = DIM*jnrA;
188 j_coord_offsetB = DIM*jnrB;
190 /* load j atom coordinates */
191 gmx_fjsp_load_1rvec_2ptr_swizzle_v2r8(x+j_coord_offsetA,x+j_coord_offsetB,
194 /* Calculate displacement vector */
195 dx00 = _fjsp_sub_v2r8(ix0,jx0);
196 dy00 = _fjsp_sub_v2r8(iy0,jy0);
197 dz00 = _fjsp_sub_v2r8(iz0,jz0);
199 /* Calculate squared distance and things based on it */
200 rsq00 = gmx_fjsp_calc_rsq_v2r8(dx00,dy00,dz00);
202 rinv00 = gmx_fjsp_invsqrt_v2r8(rsq00);
204 rinvsq00 = _fjsp_mul_v2r8(rinv00,rinv00);
206 /* Load parameters for j particles */
207 jq0 = gmx_fjsp_load_2real_swizzle_v2r8(charge+jnrA+0,charge+jnrB+0);
208 vdwjidx0A = 2*vdwtype[jnrA+0];
209 vdwjidx0B = 2*vdwtype[jnrB+0];
211 /**************************
212 * CALCULATE INTERACTIONS *
213 **************************/
215 if (gmx_fjsp_any_lt_v2r8(rsq00,rcutoff2))
218 r00 = _fjsp_mul_v2r8(rsq00,rinv00);
220 /* Compute parameters for interactions between i and j atoms */
221 qq00 = _fjsp_mul_v2r8(iq0,jq0);
222 gmx_fjsp_load_2pair_swizzle_v2r8(vdwparam+vdwioffset0+vdwjidx0A,
223 vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
225 /* EWALD ELECTROSTATICS */
227 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
228 ewrt = _fjsp_mul_v2r8(r00,ewtabscale);
229 itab_tmp = _fjsp_dtox_v2r8(ewrt);
230 eweps = _fjsp_sub_v2r8(ewrt,_fjsp_xtod_v2r8(itab_tmp));
231 _fjsp_store_v2r8(&ewconv.simd,itab_tmp);
233 ewtabF = _fjsp_load_v2r8( ewtab + 4*ewconv.i[0] );
234 ewtabD = _fjsp_load_v2r8( ewtab + 4*ewconv.i[1] );
235 GMX_FJSP_TRANSPOSE2_V2R8(ewtabF,ewtabD);
236 ewtabV = _fjsp_loadl_v2r8(_fjsp_setzero_v2r8(), ewtab + 4*ewconv.i[0] +2);
237 ewtabFn = _fjsp_loadl_v2r8(_fjsp_setzero_v2r8(), ewtab + 4*ewconv.i[1] +2);
238 GMX_FJSP_TRANSPOSE2_V2R8(ewtabV,ewtabFn);
239 felec = _fjsp_madd_v2r8(eweps,ewtabD,ewtabF);
240 velec = _fjsp_nmsub_v2r8(_fjsp_mul_v2r8(ewtabhalfspace,eweps) ,_fjsp_add_v2r8(ewtabF,felec), ewtabV);
241 velec = _fjsp_mul_v2r8(qq00,_fjsp_sub_v2r8(rinv00,velec));
242 felec = _fjsp_mul_v2r8(_fjsp_mul_v2r8(qq00,rinv00),_fjsp_sub_v2r8(rinvsq00,felec));
244 /* LENNARD-JONES DISPERSION/REPULSION */
246 rinvsix = _fjsp_mul_v2r8(_fjsp_mul_v2r8(rinvsq00,rinvsq00),rinvsq00);
247 vvdw6 = _fjsp_mul_v2r8(c6_00,rinvsix);
248 vvdw12 = _fjsp_mul_v2r8(c12_00,_fjsp_mul_v2r8(rinvsix,rinvsix));
249 vvdw = _fjsp_msub_v2r8( vvdw12,one_twelfth, _fjsp_mul_v2r8(vvdw6,one_sixth) );
250 fvdw = _fjsp_mul_v2r8(_fjsp_sub_v2r8(vvdw12,vvdw6),rinvsq00);
252 d = _fjsp_sub_v2r8(r00,rswitch);
253 d = _fjsp_max_v2r8(d,_fjsp_setzero_v2r8());
254 d2 = _fjsp_mul_v2r8(d,d);
255 sw = _fjsp_add_v2r8(one,_fjsp_mul_v2r8(d2,_fjsp_mul_v2r8(d,_fjsp_madd_v2r8(d,_fjsp_madd_v2r8(d,swV5,swV4),swV3))));
257 dsw = _fjsp_mul_v2r8(d2,_fjsp_madd_v2r8(d,_fjsp_madd_v2r8(d,swF4,swF3),swF2));
259 /* Evaluate switch function */
260 /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
261 felec = _fjsp_msub_v2r8( felec,sw , _fjsp_mul_v2r8(rinv00,_fjsp_mul_v2r8(velec,dsw)) );
262 fvdw = _fjsp_msub_v2r8( fvdw,sw , _fjsp_mul_v2r8(rinv00,_fjsp_mul_v2r8(vvdw,dsw)) );
263 velec = _fjsp_mul_v2r8(velec,sw);
264 vvdw = _fjsp_mul_v2r8(vvdw,sw);
265 cutoff_mask = _fjsp_cmplt_v2r8(rsq00,rcutoff2);
267 /* Update potential sum for this i atom from the interaction with this j atom. */
268 velec = _fjsp_and_v2r8(velec,cutoff_mask);
269 velecsum = _fjsp_add_v2r8(velecsum,velec);
270 vvdw = _fjsp_and_v2r8(vvdw,cutoff_mask);
271 vvdwsum = _fjsp_add_v2r8(vvdwsum,vvdw);
273 fscal = _fjsp_add_v2r8(felec,fvdw);
275 fscal = _fjsp_and_v2r8(fscal,cutoff_mask);
277 /* Update vectorial force */
278 fix0 = _fjsp_madd_v2r8(dx00,fscal,fix0);
279 fiy0 = _fjsp_madd_v2r8(dy00,fscal,fiy0);
280 fiz0 = _fjsp_madd_v2r8(dz00,fscal,fiz0);
282 gmx_fjsp_decrement_fma_1rvec_2ptr_swizzle_v2r8(f+j_coord_offsetA,f+j_coord_offsetB,fscal,dx00,dy00,dz00);
286 /* Inner loop uses 86 flops */
293 j_coord_offsetA = DIM*jnrA;
295 /* load j atom coordinates */
296 gmx_fjsp_load_1rvec_1ptr_swizzle_v2r8(x+j_coord_offsetA,
299 /* Calculate displacement vector */
300 dx00 = _fjsp_sub_v2r8(ix0,jx0);
301 dy00 = _fjsp_sub_v2r8(iy0,jy0);
302 dz00 = _fjsp_sub_v2r8(iz0,jz0);
304 /* Calculate squared distance and things based on it */
305 rsq00 = gmx_fjsp_calc_rsq_v2r8(dx00,dy00,dz00);
307 rinv00 = gmx_fjsp_invsqrt_v2r8(rsq00);
309 rinvsq00 = _fjsp_mul_v2r8(rinv00,rinv00);
311 /* Load parameters for j particles */
312 jq0 = _fjsp_loadl_v2r8(_fjsp_setzero_v2r8(),charge+jnrA+0);
313 vdwjidx0A = 2*vdwtype[jnrA+0];
315 /**************************
316 * CALCULATE INTERACTIONS *
317 **************************/
319 if (gmx_fjsp_any_lt_v2r8(rsq00,rcutoff2))
322 r00 = _fjsp_mul_v2r8(rsq00,rinv00);
324 /* Compute parameters for interactions between i and j atoms */
325 qq00 = _fjsp_mul_v2r8(iq0,jq0);
326 gmx_fjsp_load_1pair_swizzle_v2r8(vdwparam+vdwioffset0+vdwjidx0A,&c6_00,&c12_00);
328 /* EWALD ELECTROSTATICS */
330 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
331 ewrt = _fjsp_mul_v2r8(r00,ewtabscale);
332 itab_tmp = _fjsp_dtox_v2r8(ewrt);
333 eweps = _fjsp_sub_v2r8(ewrt,_fjsp_xtod_v2r8(itab_tmp));
334 _fjsp_store_v2r8(&ewconv.simd,itab_tmp);
336 ewtabF = _fjsp_load_v2r8( ewtab + 4*ewconv.i[0] );
337 ewtabD = _fjsp_setzero_v2r8();
338 GMX_FJSP_TRANSPOSE2_V2R8(ewtabF,ewtabD);
339 ewtabV = _fjsp_loadl_v2r8(_fjsp_setzero_v2r8(), ewtab + 4*ewconv.i[0] +2);
340 ewtabFn = _fjsp_setzero_v2r8();
341 GMX_FJSP_TRANSPOSE2_V2R8(ewtabV,ewtabFn);
342 felec = _fjsp_madd_v2r8(eweps,ewtabD,ewtabF);
343 velec = _fjsp_nmsub_v2r8(_fjsp_mul_v2r8(ewtabhalfspace,eweps) ,_fjsp_add_v2r8(ewtabF,felec), ewtabV);
344 velec = _fjsp_mul_v2r8(qq00,_fjsp_sub_v2r8(rinv00,velec));
345 felec = _fjsp_mul_v2r8(_fjsp_mul_v2r8(qq00,rinv00),_fjsp_sub_v2r8(rinvsq00,felec));
347 /* LENNARD-JONES DISPERSION/REPULSION */
349 rinvsix = _fjsp_mul_v2r8(_fjsp_mul_v2r8(rinvsq00,rinvsq00),rinvsq00);
350 vvdw6 = _fjsp_mul_v2r8(c6_00,rinvsix);
351 vvdw12 = _fjsp_mul_v2r8(c12_00,_fjsp_mul_v2r8(rinvsix,rinvsix));
352 vvdw = _fjsp_msub_v2r8( vvdw12,one_twelfth, _fjsp_mul_v2r8(vvdw6,one_sixth) );
353 fvdw = _fjsp_mul_v2r8(_fjsp_sub_v2r8(vvdw12,vvdw6),rinvsq00);
355 d = _fjsp_sub_v2r8(r00,rswitch);
356 d = _fjsp_max_v2r8(d,_fjsp_setzero_v2r8());
357 d2 = _fjsp_mul_v2r8(d,d);
358 sw = _fjsp_add_v2r8(one,_fjsp_mul_v2r8(d2,_fjsp_mul_v2r8(d,_fjsp_madd_v2r8(d,_fjsp_madd_v2r8(d,swV5,swV4),swV3))));
360 dsw = _fjsp_mul_v2r8(d2,_fjsp_madd_v2r8(d,_fjsp_madd_v2r8(d,swF4,swF3),swF2));
362 /* Evaluate switch function */
363 /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
364 felec = _fjsp_msub_v2r8( felec,sw , _fjsp_mul_v2r8(rinv00,_fjsp_mul_v2r8(velec,dsw)) );
365 fvdw = _fjsp_msub_v2r8( fvdw,sw , _fjsp_mul_v2r8(rinv00,_fjsp_mul_v2r8(vvdw,dsw)) );
366 velec = _fjsp_mul_v2r8(velec,sw);
367 vvdw = _fjsp_mul_v2r8(vvdw,sw);
368 cutoff_mask = _fjsp_cmplt_v2r8(rsq00,rcutoff2);
370 /* Update potential sum for this i atom from the interaction with this j atom. */
371 velec = _fjsp_and_v2r8(velec,cutoff_mask);
372 velec = _fjsp_unpacklo_v2r8(velec,_fjsp_setzero_v2r8());
373 velecsum = _fjsp_add_v2r8(velecsum,velec);
374 vvdw = _fjsp_and_v2r8(vvdw,cutoff_mask);
375 vvdw = _fjsp_unpacklo_v2r8(vvdw,_fjsp_setzero_v2r8());
376 vvdwsum = _fjsp_add_v2r8(vvdwsum,vvdw);
378 fscal = _fjsp_add_v2r8(felec,fvdw);
380 fscal = _fjsp_and_v2r8(fscal,cutoff_mask);
382 fscal = _fjsp_unpacklo_v2r8(fscal,_fjsp_setzero_v2r8());
384 /* Update vectorial force */
385 fix0 = _fjsp_madd_v2r8(dx00,fscal,fix0);
386 fiy0 = _fjsp_madd_v2r8(dy00,fscal,fiy0);
387 fiz0 = _fjsp_madd_v2r8(dz00,fscal,fiz0);
389 gmx_fjsp_decrement_fma_1rvec_1ptr_swizzle_v2r8(f+j_coord_offsetA,fscal,dx00,dy00,dz00);
393 /* Inner loop uses 86 flops */
396 /* End of innermost loop */
398 gmx_fjsp_update_iforce_1atom_swizzle_v2r8(fix0,fiy0,fiz0,
399 f+i_coord_offset,fshift+i_shift_offset);
402 /* Update potential energies */
403 gmx_fjsp_update_1pot_v2r8(velecsum,kernel_data->energygrp_elec+ggid);
404 gmx_fjsp_update_1pot_v2r8(vvdwsum,kernel_data->energygrp_vdw+ggid);
406 /* Increment number of inner iterations */
407 inneriter += j_index_end - j_index_start;
409 /* Outer loop uses 9 flops */
412 /* Increment number of outer iterations */
415 /* Update outer/inner flops */
417 inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_VF,outeriter*9 + inneriter*86);
420 * Gromacs nonbonded kernel: nb_kernel_ElecEwSw_VdwLJSw_GeomP1P1_F_sparc64_hpc_ace_double
421 * Electrostatics interaction: Ewald
422 * VdW interaction: LennardJones
423 * Geometry: Particle-Particle
424 * Calculate force/pot: Force
427 nb_kernel_ElecEwSw_VdwLJSw_GeomP1P1_F_sparc64_hpc_ace_double
428 (t_nblist * gmx_restrict nlist,
429 rvec * gmx_restrict xx,
430 rvec * gmx_restrict ff,
431 t_forcerec * gmx_restrict fr,
432 t_mdatoms * gmx_restrict mdatoms,
433 nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
434 t_nrnb * gmx_restrict nrnb)
436 /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
437 * just 0 for non-waters.
438 * Suffixes A,B refer to j loop unrolling done with double precision SIMD, e.g. for the two different
439 * jnr indices corresponding to data put in the four positions in the SIMD register.
441 int i_shift_offset,i_coord_offset,outeriter,inneriter;
442 int j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
444 int j_coord_offsetA,j_coord_offsetB;
445 int *iinr,*jindex,*jjnr,*shiftidx,*gid;
447 real *shiftvec,*fshift,*x,*f;
448 _fjsp_v2r8 tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
450 _fjsp_v2r8 ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
451 int vdwjidx0A,vdwjidx0B;
452 _fjsp_v2r8 jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
453 _fjsp_v2r8 dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
454 _fjsp_v2r8 velec,felec,velecsum,facel,crf,krf,krf2;
457 _fjsp_v2r8 rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
460 _fjsp_v2r8 one_sixth = gmx_fjsp_set1_v2r8(1.0/6.0);
461 _fjsp_v2r8 one_twelfth = gmx_fjsp_set1_v2r8(1.0/12.0);
462 _fjsp_v2r8 ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
464 _fjsp_v2r8 rswitch,swV3,swV4,swV5,swF2,swF3,swF4,d,d2,sw,dsw;
465 real rswitch_scalar,d_scalar;
467 _fjsp_v2r8 dummy_mask,cutoff_mask;
468 _fjsp_v2r8 one = gmx_fjsp_set1_v2r8(1.0);
469 _fjsp_v2r8 two = gmx_fjsp_set1_v2r8(2.0);
470 union { _fjsp_v2r8 simd; long long int i[2]; } vfconv,gbconv,ewconv;
477 jindex = nlist->jindex;
479 shiftidx = nlist->shift;
481 shiftvec = fr->shift_vec[0];
482 fshift = fr->fshift[0];
483 facel = gmx_fjsp_set1_v2r8(fr->epsfac);
484 charge = mdatoms->chargeA;
485 nvdwtype = fr->ntype;
487 vdwtype = mdatoms->typeA;
489 sh_ewald = gmx_fjsp_set1_v2r8(fr->ic->sh_ewald);
490 ewtab = fr->ic->tabq_coul_FDV0;
491 ewtabscale = gmx_fjsp_set1_v2r8(fr->ic->tabq_scale);
492 ewtabhalfspace = gmx_fjsp_set1_v2r8(0.5/fr->ic->tabq_scale);
494 /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
495 rcutoff_scalar = fr->rcoulomb;
496 rcutoff = gmx_fjsp_set1_v2r8(rcutoff_scalar);
497 rcutoff2 = _fjsp_mul_v2r8(rcutoff,rcutoff);
499 rswitch_scalar = fr->rcoulomb_switch;
500 rswitch = gmx_fjsp_set1_v2r8(rswitch_scalar);
501 /* Setup switch parameters */
502 d_scalar = rcutoff_scalar-rswitch_scalar;
503 d = gmx_fjsp_set1_v2r8(d_scalar);
504 swV3 = gmx_fjsp_set1_v2r8(-10.0/(d_scalar*d_scalar*d_scalar));
505 swV4 = gmx_fjsp_set1_v2r8( 15.0/(d_scalar*d_scalar*d_scalar*d_scalar));
506 swV5 = gmx_fjsp_set1_v2r8( -6.0/(d_scalar*d_scalar*d_scalar*d_scalar*d_scalar));
507 swF2 = gmx_fjsp_set1_v2r8(-30.0/(d_scalar*d_scalar*d_scalar));
508 swF3 = gmx_fjsp_set1_v2r8( 60.0/(d_scalar*d_scalar*d_scalar*d_scalar));
509 swF4 = gmx_fjsp_set1_v2r8(-30.0/(d_scalar*d_scalar*d_scalar*d_scalar*d_scalar));
511 /* Avoid stupid compiler warnings */
519 /* Start outer loop over neighborlists */
520 for(iidx=0; iidx<nri; iidx++)
522 /* Load shift vector for this list */
523 i_shift_offset = DIM*shiftidx[iidx];
525 /* Load limits for loop over neighbors */
526 j_index_start = jindex[iidx];
527 j_index_end = jindex[iidx+1];
529 /* Get outer coordinate index */
531 i_coord_offset = DIM*inr;
533 /* Load i particle coords and add shift vector */
534 gmx_fjsp_load_shift_and_1rvec_broadcast_v2r8(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
536 fix0 = _fjsp_setzero_v2r8();
537 fiy0 = _fjsp_setzero_v2r8();
538 fiz0 = _fjsp_setzero_v2r8();
540 /* Load parameters for i particles */
541 iq0 = _fjsp_mul_v2r8(facel,gmx_fjsp_load1_v2r8(charge+inr+0));
542 vdwioffset0 = 2*nvdwtype*vdwtype[inr+0];
544 /* Start inner kernel loop */
545 for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
548 /* Get j neighbor index, and coordinate index */
551 j_coord_offsetA = DIM*jnrA;
552 j_coord_offsetB = DIM*jnrB;
554 /* load j atom coordinates */
555 gmx_fjsp_load_1rvec_2ptr_swizzle_v2r8(x+j_coord_offsetA,x+j_coord_offsetB,
558 /* Calculate displacement vector */
559 dx00 = _fjsp_sub_v2r8(ix0,jx0);
560 dy00 = _fjsp_sub_v2r8(iy0,jy0);
561 dz00 = _fjsp_sub_v2r8(iz0,jz0);
563 /* Calculate squared distance and things based on it */
564 rsq00 = gmx_fjsp_calc_rsq_v2r8(dx00,dy00,dz00);
566 rinv00 = gmx_fjsp_invsqrt_v2r8(rsq00);
568 rinvsq00 = _fjsp_mul_v2r8(rinv00,rinv00);
570 /* Load parameters for j particles */
571 jq0 = gmx_fjsp_load_2real_swizzle_v2r8(charge+jnrA+0,charge+jnrB+0);
572 vdwjidx0A = 2*vdwtype[jnrA+0];
573 vdwjidx0B = 2*vdwtype[jnrB+0];
575 /**************************
576 * CALCULATE INTERACTIONS *
577 **************************/
579 if (gmx_fjsp_any_lt_v2r8(rsq00,rcutoff2))
582 r00 = _fjsp_mul_v2r8(rsq00,rinv00);
584 /* Compute parameters for interactions between i and j atoms */
585 qq00 = _fjsp_mul_v2r8(iq0,jq0);
586 gmx_fjsp_load_2pair_swizzle_v2r8(vdwparam+vdwioffset0+vdwjidx0A,
587 vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
589 /* EWALD ELECTROSTATICS */
591 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
592 ewrt = _fjsp_mul_v2r8(r00,ewtabscale);
593 itab_tmp = _fjsp_dtox_v2r8(ewrt);
594 eweps = _fjsp_sub_v2r8(ewrt,_fjsp_xtod_v2r8(itab_tmp));
595 _fjsp_store_v2r8(&ewconv.simd,itab_tmp);
597 ewtabF = _fjsp_load_v2r8( ewtab + 4*ewconv.i[0] );
598 ewtabD = _fjsp_load_v2r8( ewtab + 4*ewconv.i[1] );
599 GMX_FJSP_TRANSPOSE2_V2R8(ewtabF,ewtabD);
600 ewtabV = _fjsp_loadl_v2r8(_fjsp_setzero_v2r8(), ewtab + 4*ewconv.i[0] +2);
601 ewtabFn = _fjsp_loadl_v2r8(_fjsp_setzero_v2r8(), ewtab + 4*ewconv.i[1] +2);
602 GMX_FJSP_TRANSPOSE2_V2R8(ewtabV,ewtabFn);
603 felec = _fjsp_madd_v2r8(eweps,ewtabD,ewtabF);
604 velec = _fjsp_nmsub_v2r8(_fjsp_mul_v2r8(ewtabhalfspace,eweps) ,_fjsp_add_v2r8(ewtabF,felec), ewtabV);
605 velec = _fjsp_mul_v2r8(qq00,_fjsp_sub_v2r8(rinv00,velec));
606 felec = _fjsp_mul_v2r8(_fjsp_mul_v2r8(qq00,rinv00),_fjsp_sub_v2r8(rinvsq00,felec));
608 /* LENNARD-JONES DISPERSION/REPULSION */
610 rinvsix = _fjsp_mul_v2r8(_fjsp_mul_v2r8(rinvsq00,rinvsq00),rinvsq00);
611 vvdw6 = _fjsp_mul_v2r8(c6_00,rinvsix);
612 vvdw12 = _fjsp_mul_v2r8(c12_00,_fjsp_mul_v2r8(rinvsix,rinvsix));
613 vvdw = _fjsp_msub_v2r8( vvdw12,one_twelfth, _fjsp_mul_v2r8(vvdw6,one_sixth) );
614 fvdw = _fjsp_mul_v2r8(_fjsp_sub_v2r8(vvdw12,vvdw6),rinvsq00);
616 d = _fjsp_sub_v2r8(r00,rswitch);
617 d = _fjsp_max_v2r8(d,_fjsp_setzero_v2r8());
618 d2 = _fjsp_mul_v2r8(d,d);
619 sw = _fjsp_add_v2r8(one,_fjsp_mul_v2r8(d2,_fjsp_mul_v2r8(d,_fjsp_madd_v2r8(d,_fjsp_madd_v2r8(d,swV5,swV4),swV3))));
621 dsw = _fjsp_mul_v2r8(d2,_fjsp_madd_v2r8(d,_fjsp_madd_v2r8(d,swF4,swF3),swF2));
623 /* Evaluate switch function */
624 /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
625 felec = _fjsp_msub_v2r8( felec,sw , _fjsp_mul_v2r8(rinv00,_fjsp_mul_v2r8(velec,dsw)) );
626 fvdw = _fjsp_msub_v2r8( fvdw,sw , _fjsp_mul_v2r8(rinv00,_fjsp_mul_v2r8(vvdw,dsw)) );
627 cutoff_mask = _fjsp_cmplt_v2r8(rsq00,rcutoff2);
629 fscal = _fjsp_add_v2r8(felec,fvdw);
631 fscal = _fjsp_and_v2r8(fscal,cutoff_mask);
633 /* Update vectorial force */
634 fix0 = _fjsp_madd_v2r8(dx00,fscal,fix0);
635 fiy0 = _fjsp_madd_v2r8(dy00,fscal,fiy0);
636 fiz0 = _fjsp_madd_v2r8(dz00,fscal,fiz0);
638 gmx_fjsp_decrement_fma_1rvec_2ptr_swizzle_v2r8(f+j_coord_offsetA,f+j_coord_offsetB,fscal,dx00,dy00,dz00);
642 /* Inner loop uses 80 flops */
649 j_coord_offsetA = DIM*jnrA;
651 /* load j atom coordinates */
652 gmx_fjsp_load_1rvec_1ptr_swizzle_v2r8(x+j_coord_offsetA,
655 /* Calculate displacement vector */
656 dx00 = _fjsp_sub_v2r8(ix0,jx0);
657 dy00 = _fjsp_sub_v2r8(iy0,jy0);
658 dz00 = _fjsp_sub_v2r8(iz0,jz0);
660 /* Calculate squared distance and things based on it */
661 rsq00 = gmx_fjsp_calc_rsq_v2r8(dx00,dy00,dz00);
663 rinv00 = gmx_fjsp_invsqrt_v2r8(rsq00);
665 rinvsq00 = _fjsp_mul_v2r8(rinv00,rinv00);
667 /* Load parameters for j particles */
668 jq0 = _fjsp_loadl_v2r8(_fjsp_setzero_v2r8(),charge+jnrA+0);
669 vdwjidx0A = 2*vdwtype[jnrA+0];
671 /**************************
672 * CALCULATE INTERACTIONS *
673 **************************/
675 if (gmx_fjsp_any_lt_v2r8(rsq00,rcutoff2))
678 r00 = _fjsp_mul_v2r8(rsq00,rinv00);
680 /* Compute parameters for interactions between i and j atoms */
681 qq00 = _fjsp_mul_v2r8(iq0,jq0);
682 gmx_fjsp_load_1pair_swizzle_v2r8(vdwparam+vdwioffset0+vdwjidx0A,&c6_00,&c12_00);
684 /* EWALD ELECTROSTATICS */
686 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
687 ewrt = _fjsp_mul_v2r8(r00,ewtabscale);
688 itab_tmp = _fjsp_dtox_v2r8(ewrt);
689 eweps = _fjsp_sub_v2r8(ewrt,_fjsp_xtod_v2r8(itab_tmp));
690 _fjsp_store_v2r8(&ewconv.simd,itab_tmp);
692 ewtabF = _fjsp_load_v2r8( ewtab + 4*ewconv.i[0] );
693 ewtabD = _fjsp_setzero_v2r8();
694 GMX_FJSP_TRANSPOSE2_V2R8(ewtabF,ewtabD);
695 ewtabV = _fjsp_loadl_v2r8(_fjsp_setzero_v2r8(), ewtab + 4*ewconv.i[0] +2);
696 ewtabFn = _fjsp_setzero_v2r8();
697 GMX_FJSP_TRANSPOSE2_V2R8(ewtabV,ewtabFn);
698 felec = _fjsp_madd_v2r8(eweps,ewtabD,ewtabF);
699 velec = _fjsp_nmsub_v2r8(_fjsp_mul_v2r8(ewtabhalfspace,eweps) ,_fjsp_add_v2r8(ewtabF,felec), ewtabV);
700 velec = _fjsp_mul_v2r8(qq00,_fjsp_sub_v2r8(rinv00,velec));
701 felec = _fjsp_mul_v2r8(_fjsp_mul_v2r8(qq00,rinv00),_fjsp_sub_v2r8(rinvsq00,felec));
703 /* LENNARD-JONES DISPERSION/REPULSION */
705 rinvsix = _fjsp_mul_v2r8(_fjsp_mul_v2r8(rinvsq00,rinvsq00),rinvsq00);
706 vvdw6 = _fjsp_mul_v2r8(c6_00,rinvsix);
707 vvdw12 = _fjsp_mul_v2r8(c12_00,_fjsp_mul_v2r8(rinvsix,rinvsix));
708 vvdw = _fjsp_msub_v2r8( vvdw12,one_twelfth, _fjsp_mul_v2r8(vvdw6,one_sixth) );
709 fvdw = _fjsp_mul_v2r8(_fjsp_sub_v2r8(vvdw12,vvdw6),rinvsq00);
711 d = _fjsp_sub_v2r8(r00,rswitch);
712 d = _fjsp_max_v2r8(d,_fjsp_setzero_v2r8());
713 d2 = _fjsp_mul_v2r8(d,d);
714 sw = _fjsp_add_v2r8(one,_fjsp_mul_v2r8(d2,_fjsp_mul_v2r8(d,_fjsp_madd_v2r8(d,_fjsp_madd_v2r8(d,swV5,swV4),swV3))));
716 dsw = _fjsp_mul_v2r8(d2,_fjsp_madd_v2r8(d,_fjsp_madd_v2r8(d,swF4,swF3),swF2));
718 /* Evaluate switch function */
719 /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
720 felec = _fjsp_msub_v2r8( felec,sw , _fjsp_mul_v2r8(rinv00,_fjsp_mul_v2r8(velec,dsw)) );
721 fvdw = _fjsp_msub_v2r8( fvdw,sw , _fjsp_mul_v2r8(rinv00,_fjsp_mul_v2r8(vvdw,dsw)) );
722 cutoff_mask = _fjsp_cmplt_v2r8(rsq00,rcutoff2);
724 fscal = _fjsp_add_v2r8(felec,fvdw);
726 fscal = _fjsp_and_v2r8(fscal,cutoff_mask);
728 fscal = _fjsp_unpacklo_v2r8(fscal,_fjsp_setzero_v2r8());
730 /* Update vectorial force */
731 fix0 = _fjsp_madd_v2r8(dx00,fscal,fix0);
732 fiy0 = _fjsp_madd_v2r8(dy00,fscal,fiy0);
733 fiz0 = _fjsp_madd_v2r8(dz00,fscal,fiz0);
735 gmx_fjsp_decrement_fma_1rvec_1ptr_swizzle_v2r8(f+j_coord_offsetA,fscal,dx00,dy00,dz00);
739 /* Inner loop uses 80 flops */
742 /* End of innermost loop */
744 gmx_fjsp_update_iforce_1atom_swizzle_v2r8(fix0,fiy0,fiz0,
745 f+i_coord_offset,fshift+i_shift_offset);
747 /* Increment number of inner iterations */
748 inneriter += j_index_end - j_index_start;
750 /* Outer loop uses 7 flops */
753 /* Increment number of outer iterations */
756 /* Update outer/inner flops */
758 inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_F,outeriter*7 + inneriter*80);