2 * This file is part of the GROMACS molecular simulation package.
4 * Copyright (c) 2012,2013,2014, 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 sse2_double kernel generator.
44 #include "../nb_kernel.h"
45 #include "gromacs/legacyheaders/types/simple.h"
46 #include "gromacs/math/vec.h"
47 #include "gromacs/legacyheaders/nrnb.h"
49 #include "gromacs/simd/math_x86_sse2_double.h"
50 #include "kernelutil_x86_sse2_double.h"
53 * Gromacs nonbonded kernel: nb_kernel_ElecRFCut_VdwLJSw_GeomP1P1_VF_sse2_double
54 * Electrostatics interaction: ReactionField
55 * VdW interaction: LennardJones
56 * Geometry: Particle-Particle
57 * Calculate force/pot: PotentialAndForce
60 nb_kernel_ElecRFCut_VdwLJSw_GeomP1P1_VF_sse2_double
61 (t_nblist * gmx_restrict nlist,
62 rvec * gmx_restrict xx,
63 rvec * gmx_restrict ff,
64 t_forcerec * gmx_restrict fr,
65 t_mdatoms * gmx_restrict mdatoms,
66 nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
67 t_nrnb * gmx_restrict nrnb)
69 /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
70 * just 0 for non-waters.
71 * Suffixes A,B refer to j loop unrolling done with SSE double precision, e.g. for the two different
72 * jnr indices corresponding to data put in the four positions in the SIMD register.
74 int i_shift_offset,i_coord_offset,outeriter,inneriter;
75 int j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
77 int j_coord_offsetA,j_coord_offsetB;
78 int *iinr,*jindex,*jjnr,*shiftidx,*gid;
80 real *shiftvec,*fshift,*x,*f;
81 __m128d tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
83 __m128d ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
84 int vdwjidx0A,vdwjidx0B;
85 __m128d jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
86 __m128d dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
87 __m128d velec,felec,velecsum,facel,crf,krf,krf2;
90 __m128d rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
93 __m128d one_sixth = _mm_set1_pd(1.0/6.0);
94 __m128d one_twelfth = _mm_set1_pd(1.0/12.0);
95 __m128d rswitch,swV3,swV4,swV5,swF2,swF3,swF4,d,d2,sw,dsw;
96 real rswitch_scalar,d_scalar;
97 __m128d dummy_mask,cutoff_mask;
98 __m128d signbit = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
99 __m128d one = _mm_set1_pd(1.0);
100 __m128d two = _mm_set1_pd(2.0);
106 jindex = nlist->jindex;
108 shiftidx = nlist->shift;
110 shiftvec = fr->shift_vec[0];
111 fshift = fr->fshift[0];
112 facel = _mm_set1_pd(fr->epsfac);
113 charge = mdatoms->chargeA;
114 krf = _mm_set1_pd(fr->ic->k_rf);
115 krf2 = _mm_set1_pd(fr->ic->k_rf*2.0);
116 crf = _mm_set1_pd(fr->ic->c_rf);
117 nvdwtype = fr->ntype;
119 vdwtype = mdatoms->typeA;
121 /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
122 rcutoff_scalar = fr->rcoulomb;
123 rcutoff = _mm_set1_pd(rcutoff_scalar);
124 rcutoff2 = _mm_mul_pd(rcutoff,rcutoff);
126 rswitch_scalar = fr->rvdw_switch;
127 rswitch = _mm_set1_pd(rswitch_scalar);
128 /* Setup switch parameters */
129 d_scalar = rcutoff_scalar-rswitch_scalar;
130 d = _mm_set1_pd(d_scalar);
131 swV3 = _mm_set1_pd(-10.0/(d_scalar*d_scalar*d_scalar));
132 swV4 = _mm_set1_pd( 15.0/(d_scalar*d_scalar*d_scalar*d_scalar));
133 swV5 = _mm_set1_pd( -6.0/(d_scalar*d_scalar*d_scalar*d_scalar*d_scalar));
134 swF2 = _mm_set1_pd(-30.0/(d_scalar*d_scalar*d_scalar));
135 swF3 = _mm_set1_pd( 60.0/(d_scalar*d_scalar*d_scalar*d_scalar));
136 swF4 = _mm_set1_pd(-30.0/(d_scalar*d_scalar*d_scalar*d_scalar*d_scalar));
138 /* Avoid stupid compiler warnings */
146 /* Start outer loop over neighborlists */
147 for(iidx=0; iidx<nri; iidx++)
149 /* Load shift vector for this list */
150 i_shift_offset = DIM*shiftidx[iidx];
152 /* Load limits for loop over neighbors */
153 j_index_start = jindex[iidx];
154 j_index_end = jindex[iidx+1];
156 /* Get outer coordinate index */
158 i_coord_offset = DIM*inr;
160 /* Load i particle coords and add shift vector */
161 gmx_mm_load_shift_and_1rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
163 fix0 = _mm_setzero_pd();
164 fiy0 = _mm_setzero_pd();
165 fiz0 = _mm_setzero_pd();
167 /* Load parameters for i particles */
168 iq0 = _mm_mul_pd(facel,_mm_load1_pd(charge+inr+0));
169 vdwioffset0 = 2*nvdwtype*vdwtype[inr+0];
171 /* Reset potential sums */
172 velecsum = _mm_setzero_pd();
173 vvdwsum = _mm_setzero_pd();
175 /* Start inner kernel loop */
176 for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
179 /* Get j neighbor index, and coordinate index */
182 j_coord_offsetA = DIM*jnrA;
183 j_coord_offsetB = DIM*jnrB;
185 /* load j atom coordinates */
186 gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
189 /* Calculate displacement vector */
190 dx00 = _mm_sub_pd(ix0,jx0);
191 dy00 = _mm_sub_pd(iy0,jy0);
192 dz00 = _mm_sub_pd(iz0,jz0);
194 /* Calculate squared distance and things based on it */
195 rsq00 = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
197 rinv00 = gmx_mm_invsqrt_pd(rsq00);
199 rinvsq00 = _mm_mul_pd(rinv00,rinv00);
201 /* Load parameters for j particles */
202 jq0 = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
203 vdwjidx0A = 2*vdwtype[jnrA+0];
204 vdwjidx0B = 2*vdwtype[jnrB+0];
206 /**************************
207 * CALCULATE INTERACTIONS *
208 **************************/
210 if (gmx_mm_any_lt(rsq00,rcutoff2))
213 r00 = _mm_mul_pd(rsq00,rinv00);
215 /* Compute parameters for interactions between i and j atoms */
216 qq00 = _mm_mul_pd(iq0,jq0);
217 gmx_mm_load_2pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,
218 vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
220 /* REACTION-FIELD ELECTROSTATICS */
221 velec = _mm_mul_pd(qq00,_mm_sub_pd(_mm_add_pd(rinv00,_mm_mul_pd(krf,rsq00)),crf));
222 felec = _mm_mul_pd(qq00,_mm_sub_pd(_mm_mul_pd(rinv00,rinvsq00),krf2));
224 /* LENNARD-JONES DISPERSION/REPULSION */
226 rinvsix = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
227 vvdw6 = _mm_mul_pd(c6_00,rinvsix);
228 vvdw12 = _mm_mul_pd(c12_00,_mm_mul_pd(rinvsix,rinvsix));
229 vvdw = _mm_sub_pd( _mm_mul_pd(vvdw12,one_twelfth) , _mm_mul_pd(vvdw6,one_sixth) );
230 fvdw = _mm_mul_pd(_mm_sub_pd(vvdw12,vvdw6),rinvsq00);
232 d = _mm_sub_pd(r00,rswitch);
233 d = _mm_max_pd(d,_mm_setzero_pd());
234 d2 = _mm_mul_pd(d,d);
235 sw = _mm_add_pd(one,_mm_mul_pd(d2,_mm_mul_pd(d,_mm_add_pd(swV3,_mm_mul_pd(d,_mm_add_pd(swV4,_mm_mul_pd(d,swV5)))))));
237 dsw = _mm_mul_pd(d2,_mm_add_pd(swF2,_mm_mul_pd(d,_mm_add_pd(swF3,_mm_mul_pd(d,swF4)))));
239 /* Evaluate switch function */
240 /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
241 fvdw = _mm_sub_pd( _mm_mul_pd(fvdw,sw) , _mm_mul_pd(rinv00,_mm_mul_pd(vvdw,dsw)) );
242 vvdw = _mm_mul_pd(vvdw,sw);
243 cutoff_mask = _mm_cmplt_pd(rsq00,rcutoff2);
245 /* Update potential sum for this i atom from the interaction with this j atom. */
246 velec = _mm_and_pd(velec,cutoff_mask);
247 velecsum = _mm_add_pd(velecsum,velec);
248 vvdw = _mm_and_pd(vvdw,cutoff_mask);
249 vvdwsum = _mm_add_pd(vvdwsum,vvdw);
251 fscal = _mm_add_pd(felec,fvdw);
253 fscal = _mm_and_pd(fscal,cutoff_mask);
255 /* Calculate temporary vectorial force */
256 tx = _mm_mul_pd(fscal,dx00);
257 ty = _mm_mul_pd(fscal,dy00);
258 tz = _mm_mul_pd(fscal,dz00);
260 /* Update vectorial force */
261 fix0 = _mm_add_pd(fix0,tx);
262 fiy0 = _mm_add_pd(fiy0,ty);
263 fiz0 = _mm_add_pd(fiz0,tz);
265 gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,tx,ty,tz);
269 /* Inner loop uses 70 flops */
276 j_coord_offsetA = DIM*jnrA;
278 /* load j atom coordinates */
279 gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
282 /* Calculate displacement vector */
283 dx00 = _mm_sub_pd(ix0,jx0);
284 dy00 = _mm_sub_pd(iy0,jy0);
285 dz00 = _mm_sub_pd(iz0,jz0);
287 /* Calculate squared distance and things based on it */
288 rsq00 = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
290 rinv00 = gmx_mm_invsqrt_pd(rsq00);
292 rinvsq00 = _mm_mul_pd(rinv00,rinv00);
294 /* Load parameters for j particles */
295 jq0 = _mm_load_sd(charge+jnrA+0);
296 vdwjidx0A = 2*vdwtype[jnrA+0];
298 /**************************
299 * CALCULATE INTERACTIONS *
300 **************************/
302 if (gmx_mm_any_lt(rsq00,rcutoff2))
305 r00 = _mm_mul_pd(rsq00,rinv00);
307 /* Compute parameters for interactions between i and j atoms */
308 qq00 = _mm_mul_pd(iq0,jq0);
309 gmx_mm_load_1pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,&c6_00,&c12_00);
311 /* REACTION-FIELD ELECTROSTATICS */
312 velec = _mm_mul_pd(qq00,_mm_sub_pd(_mm_add_pd(rinv00,_mm_mul_pd(krf,rsq00)),crf));
313 felec = _mm_mul_pd(qq00,_mm_sub_pd(_mm_mul_pd(rinv00,rinvsq00),krf2));
315 /* LENNARD-JONES DISPERSION/REPULSION */
317 rinvsix = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
318 vvdw6 = _mm_mul_pd(c6_00,rinvsix);
319 vvdw12 = _mm_mul_pd(c12_00,_mm_mul_pd(rinvsix,rinvsix));
320 vvdw = _mm_sub_pd( _mm_mul_pd(vvdw12,one_twelfth) , _mm_mul_pd(vvdw6,one_sixth) );
321 fvdw = _mm_mul_pd(_mm_sub_pd(vvdw12,vvdw6),rinvsq00);
323 d = _mm_sub_pd(r00,rswitch);
324 d = _mm_max_pd(d,_mm_setzero_pd());
325 d2 = _mm_mul_pd(d,d);
326 sw = _mm_add_pd(one,_mm_mul_pd(d2,_mm_mul_pd(d,_mm_add_pd(swV3,_mm_mul_pd(d,_mm_add_pd(swV4,_mm_mul_pd(d,swV5)))))));
328 dsw = _mm_mul_pd(d2,_mm_add_pd(swF2,_mm_mul_pd(d,_mm_add_pd(swF3,_mm_mul_pd(d,swF4)))));
330 /* Evaluate switch function */
331 /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
332 fvdw = _mm_sub_pd( _mm_mul_pd(fvdw,sw) , _mm_mul_pd(rinv00,_mm_mul_pd(vvdw,dsw)) );
333 vvdw = _mm_mul_pd(vvdw,sw);
334 cutoff_mask = _mm_cmplt_pd(rsq00,rcutoff2);
336 /* Update potential sum for this i atom from the interaction with this j atom. */
337 velec = _mm_and_pd(velec,cutoff_mask);
338 velec = _mm_unpacklo_pd(velec,_mm_setzero_pd());
339 velecsum = _mm_add_pd(velecsum,velec);
340 vvdw = _mm_and_pd(vvdw,cutoff_mask);
341 vvdw = _mm_unpacklo_pd(vvdw,_mm_setzero_pd());
342 vvdwsum = _mm_add_pd(vvdwsum,vvdw);
344 fscal = _mm_add_pd(felec,fvdw);
346 fscal = _mm_and_pd(fscal,cutoff_mask);
348 fscal = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
350 /* Calculate temporary vectorial force */
351 tx = _mm_mul_pd(fscal,dx00);
352 ty = _mm_mul_pd(fscal,dy00);
353 tz = _mm_mul_pd(fscal,dz00);
355 /* Update vectorial force */
356 fix0 = _mm_add_pd(fix0,tx);
357 fiy0 = _mm_add_pd(fiy0,ty);
358 fiz0 = _mm_add_pd(fiz0,tz);
360 gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,tx,ty,tz);
364 /* Inner loop uses 70 flops */
367 /* End of innermost loop */
369 gmx_mm_update_iforce_1atom_swizzle_pd(fix0,fiy0,fiz0,
370 f+i_coord_offset,fshift+i_shift_offset);
373 /* Update potential energies */
374 gmx_mm_update_1pot_pd(velecsum,kernel_data->energygrp_elec+ggid);
375 gmx_mm_update_1pot_pd(vvdwsum,kernel_data->energygrp_vdw+ggid);
377 /* Increment number of inner iterations */
378 inneriter += j_index_end - j_index_start;
380 /* Outer loop uses 9 flops */
383 /* Increment number of outer iterations */
386 /* Update outer/inner flops */
388 inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_VF,outeriter*9 + inneriter*70);
391 * Gromacs nonbonded kernel: nb_kernel_ElecRFCut_VdwLJSw_GeomP1P1_F_sse2_double
392 * Electrostatics interaction: ReactionField
393 * VdW interaction: LennardJones
394 * Geometry: Particle-Particle
395 * Calculate force/pot: Force
398 nb_kernel_ElecRFCut_VdwLJSw_GeomP1P1_F_sse2_double
399 (t_nblist * gmx_restrict nlist,
400 rvec * gmx_restrict xx,
401 rvec * gmx_restrict ff,
402 t_forcerec * gmx_restrict fr,
403 t_mdatoms * gmx_restrict mdatoms,
404 nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
405 t_nrnb * gmx_restrict nrnb)
407 /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
408 * just 0 for non-waters.
409 * Suffixes A,B refer to j loop unrolling done with SSE double precision, e.g. for the two different
410 * jnr indices corresponding to data put in the four positions in the SIMD register.
412 int i_shift_offset,i_coord_offset,outeriter,inneriter;
413 int j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
415 int j_coord_offsetA,j_coord_offsetB;
416 int *iinr,*jindex,*jjnr,*shiftidx,*gid;
418 real *shiftvec,*fshift,*x,*f;
419 __m128d tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
421 __m128d ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
422 int vdwjidx0A,vdwjidx0B;
423 __m128d jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
424 __m128d dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
425 __m128d velec,felec,velecsum,facel,crf,krf,krf2;
428 __m128d rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
431 __m128d one_sixth = _mm_set1_pd(1.0/6.0);
432 __m128d one_twelfth = _mm_set1_pd(1.0/12.0);
433 __m128d rswitch,swV3,swV4,swV5,swF2,swF3,swF4,d,d2,sw,dsw;
434 real rswitch_scalar,d_scalar;
435 __m128d dummy_mask,cutoff_mask;
436 __m128d signbit = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
437 __m128d one = _mm_set1_pd(1.0);
438 __m128d two = _mm_set1_pd(2.0);
444 jindex = nlist->jindex;
446 shiftidx = nlist->shift;
448 shiftvec = fr->shift_vec[0];
449 fshift = fr->fshift[0];
450 facel = _mm_set1_pd(fr->epsfac);
451 charge = mdatoms->chargeA;
452 krf = _mm_set1_pd(fr->ic->k_rf);
453 krf2 = _mm_set1_pd(fr->ic->k_rf*2.0);
454 crf = _mm_set1_pd(fr->ic->c_rf);
455 nvdwtype = fr->ntype;
457 vdwtype = mdatoms->typeA;
459 /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
460 rcutoff_scalar = fr->rcoulomb;
461 rcutoff = _mm_set1_pd(rcutoff_scalar);
462 rcutoff2 = _mm_mul_pd(rcutoff,rcutoff);
464 rswitch_scalar = fr->rvdw_switch;
465 rswitch = _mm_set1_pd(rswitch_scalar);
466 /* Setup switch parameters */
467 d_scalar = rcutoff_scalar-rswitch_scalar;
468 d = _mm_set1_pd(d_scalar);
469 swV3 = _mm_set1_pd(-10.0/(d_scalar*d_scalar*d_scalar));
470 swV4 = _mm_set1_pd( 15.0/(d_scalar*d_scalar*d_scalar*d_scalar));
471 swV5 = _mm_set1_pd( -6.0/(d_scalar*d_scalar*d_scalar*d_scalar*d_scalar));
472 swF2 = _mm_set1_pd(-30.0/(d_scalar*d_scalar*d_scalar));
473 swF3 = _mm_set1_pd( 60.0/(d_scalar*d_scalar*d_scalar*d_scalar));
474 swF4 = _mm_set1_pd(-30.0/(d_scalar*d_scalar*d_scalar*d_scalar*d_scalar));
476 /* Avoid stupid compiler warnings */
484 /* Start outer loop over neighborlists */
485 for(iidx=0; iidx<nri; iidx++)
487 /* Load shift vector for this list */
488 i_shift_offset = DIM*shiftidx[iidx];
490 /* Load limits for loop over neighbors */
491 j_index_start = jindex[iidx];
492 j_index_end = jindex[iidx+1];
494 /* Get outer coordinate index */
496 i_coord_offset = DIM*inr;
498 /* Load i particle coords and add shift vector */
499 gmx_mm_load_shift_and_1rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
501 fix0 = _mm_setzero_pd();
502 fiy0 = _mm_setzero_pd();
503 fiz0 = _mm_setzero_pd();
505 /* Load parameters for i particles */
506 iq0 = _mm_mul_pd(facel,_mm_load1_pd(charge+inr+0));
507 vdwioffset0 = 2*nvdwtype*vdwtype[inr+0];
509 /* Start inner kernel loop */
510 for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
513 /* Get j neighbor index, and coordinate index */
516 j_coord_offsetA = DIM*jnrA;
517 j_coord_offsetB = DIM*jnrB;
519 /* load j atom coordinates */
520 gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
523 /* Calculate displacement vector */
524 dx00 = _mm_sub_pd(ix0,jx0);
525 dy00 = _mm_sub_pd(iy0,jy0);
526 dz00 = _mm_sub_pd(iz0,jz0);
528 /* Calculate squared distance and things based on it */
529 rsq00 = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
531 rinv00 = gmx_mm_invsqrt_pd(rsq00);
533 rinvsq00 = _mm_mul_pd(rinv00,rinv00);
535 /* Load parameters for j particles */
536 jq0 = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
537 vdwjidx0A = 2*vdwtype[jnrA+0];
538 vdwjidx0B = 2*vdwtype[jnrB+0];
540 /**************************
541 * CALCULATE INTERACTIONS *
542 **************************/
544 if (gmx_mm_any_lt(rsq00,rcutoff2))
547 r00 = _mm_mul_pd(rsq00,rinv00);
549 /* Compute parameters for interactions between i and j atoms */
550 qq00 = _mm_mul_pd(iq0,jq0);
551 gmx_mm_load_2pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,
552 vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
554 /* REACTION-FIELD ELECTROSTATICS */
555 felec = _mm_mul_pd(qq00,_mm_sub_pd(_mm_mul_pd(rinv00,rinvsq00),krf2));
557 /* LENNARD-JONES DISPERSION/REPULSION */
559 rinvsix = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
560 vvdw6 = _mm_mul_pd(c6_00,rinvsix);
561 vvdw12 = _mm_mul_pd(c12_00,_mm_mul_pd(rinvsix,rinvsix));
562 vvdw = _mm_sub_pd( _mm_mul_pd(vvdw12,one_twelfth) , _mm_mul_pd(vvdw6,one_sixth) );
563 fvdw = _mm_mul_pd(_mm_sub_pd(vvdw12,vvdw6),rinvsq00);
565 d = _mm_sub_pd(r00,rswitch);
566 d = _mm_max_pd(d,_mm_setzero_pd());
567 d2 = _mm_mul_pd(d,d);
568 sw = _mm_add_pd(one,_mm_mul_pd(d2,_mm_mul_pd(d,_mm_add_pd(swV3,_mm_mul_pd(d,_mm_add_pd(swV4,_mm_mul_pd(d,swV5)))))));
570 dsw = _mm_mul_pd(d2,_mm_add_pd(swF2,_mm_mul_pd(d,_mm_add_pd(swF3,_mm_mul_pd(d,swF4)))));
572 /* Evaluate switch function */
573 /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
574 fvdw = _mm_sub_pd( _mm_mul_pd(fvdw,sw) , _mm_mul_pd(rinv00,_mm_mul_pd(vvdw,dsw)) );
575 cutoff_mask = _mm_cmplt_pd(rsq00,rcutoff2);
577 fscal = _mm_add_pd(felec,fvdw);
579 fscal = _mm_and_pd(fscal,cutoff_mask);
581 /* Calculate temporary vectorial force */
582 tx = _mm_mul_pd(fscal,dx00);
583 ty = _mm_mul_pd(fscal,dy00);
584 tz = _mm_mul_pd(fscal,dz00);
586 /* Update vectorial force */
587 fix0 = _mm_add_pd(fix0,tx);
588 fiy0 = _mm_add_pd(fiy0,ty);
589 fiz0 = _mm_add_pd(fiz0,tz);
591 gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,tx,ty,tz);
595 /* Inner loop uses 61 flops */
602 j_coord_offsetA = DIM*jnrA;
604 /* load j atom coordinates */
605 gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
608 /* Calculate displacement vector */
609 dx00 = _mm_sub_pd(ix0,jx0);
610 dy00 = _mm_sub_pd(iy0,jy0);
611 dz00 = _mm_sub_pd(iz0,jz0);
613 /* Calculate squared distance and things based on it */
614 rsq00 = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
616 rinv00 = gmx_mm_invsqrt_pd(rsq00);
618 rinvsq00 = _mm_mul_pd(rinv00,rinv00);
620 /* Load parameters for j particles */
621 jq0 = _mm_load_sd(charge+jnrA+0);
622 vdwjidx0A = 2*vdwtype[jnrA+0];
624 /**************************
625 * CALCULATE INTERACTIONS *
626 **************************/
628 if (gmx_mm_any_lt(rsq00,rcutoff2))
631 r00 = _mm_mul_pd(rsq00,rinv00);
633 /* Compute parameters for interactions between i and j atoms */
634 qq00 = _mm_mul_pd(iq0,jq0);
635 gmx_mm_load_1pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,&c6_00,&c12_00);
637 /* REACTION-FIELD ELECTROSTATICS */
638 felec = _mm_mul_pd(qq00,_mm_sub_pd(_mm_mul_pd(rinv00,rinvsq00),krf2));
640 /* LENNARD-JONES DISPERSION/REPULSION */
642 rinvsix = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
643 vvdw6 = _mm_mul_pd(c6_00,rinvsix);
644 vvdw12 = _mm_mul_pd(c12_00,_mm_mul_pd(rinvsix,rinvsix));
645 vvdw = _mm_sub_pd( _mm_mul_pd(vvdw12,one_twelfth) , _mm_mul_pd(vvdw6,one_sixth) );
646 fvdw = _mm_mul_pd(_mm_sub_pd(vvdw12,vvdw6),rinvsq00);
648 d = _mm_sub_pd(r00,rswitch);
649 d = _mm_max_pd(d,_mm_setzero_pd());
650 d2 = _mm_mul_pd(d,d);
651 sw = _mm_add_pd(one,_mm_mul_pd(d2,_mm_mul_pd(d,_mm_add_pd(swV3,_mm_mul_pd(d,_mm_add_pd(swV4,_mm_mul_pd(d,swV5)))))));
653 dsw = _mm_mul_pd(d2,_mm_add_pd(swF2,_mm_mul_pd(d,_mm_add_pd(swF3,_mm_mul_pd(d,swF4)))));
655 /* Evaluate switch function */
656 /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
657 fvdw = _mm_sub_pd( _mm_mul_pd(fvdw,sw) , _mm_mul_pd(rinv00,_mm_mul_pd(vvdw,dsw)) );
658 cutoff_mask = _mm_cmplt_pd(rsq00,rcutoff2);
660 fscal = _mm_add_pd(felec,fvdw);
662 fscal = _mm_and_pd(fscal,cutoff_mask);
664 fscal = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
666 /* Calculate temporary vectorial force */
667 tx = _mm_mul_pd(fscal,dx00);
668 ty = _mm_mul_pd(fscal,dy00);
669 tz = _mm_mul_pd(fscal,dz00);
671 /* Update vectorial force */
672 fix0 = _mm_add_pd(fix0,tx);
673 fiy0 = _mm_add_pd(fiy0,ty);
674 fiz0 = _mm_add_pd(fiz0,tz);
676 gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,tx,ty,tz);
680 /* Inner loop uses 61 flops */
683 /* End of innermost loop */
685 gmx_mm_update_iforce_1atom_swizzle_pd(fix0,fiy0,fiz0,
686 f+i_coord_offset,fshift+i_shift_offset);
688 /* Increment number of inner iterations */
689 inneriter += j_index_end - j_index_start;
691 /* Outer loop uses 7 flops */
694 /* Increment number of outer iterations */
697 /* Update outer/inner flops */
699 inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_F,outeriter*7 + inneriter*61);