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_ElecEwSh_VdwLJEwSh_GeomP1P1_VF_sse2_double
54 * Electrostatics interaction: Ewald
55 * VdW interaction: LJEwald
56 * Geometry: Particle-Particle
57 * Calculate force/pot: PotentialAndForce
60 nb_kernel_ElecEwSh_VdwLJEwSh_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);
96 __m128d ewclj,ewclj2,ewclj6,ewcljrsq,poly,exponent,f6A,f6B,sh_lj_ewald;
98 __m128d one_half = _mm_set1_pd(0.5);
99 __m128d minus_one = _mm_set1_pd(-1.0);
101 __m128d ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
103 __m128d dummy_mask,cutoff_mask;
104 __m128d signbit = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
105 __m128d one = _mm_set1_pd(1.0);
106 __m128d two = _mm_set1_pd(2.0);
112 jindex = nlist->jindex;
114 shiftidx = nlist->shift;
116 shiftvec = fr->shift_vec[0];
117 fshift = fr->fshift[0];
118 facel = _mm_set1_pd(fr->epsfac);
119 charge = mdatoms->chargeA;
120 nvdwtype = fr->ntype;
122 vdwtype = mdatoms->typeA;
123 vdwgridparam = fr->ljpme_c6grid;
124 sh_lj_ewald = _mm_set1_pd(fr->ic->sh_lj_ewald);
125 ewclj = _mm_set1_pd(fr->ewaldcoeff_lj);
126 ewclj2 = _mm_mul_pd(minus_one,_mm_mul_pd(ewclj,ewclj));
128 sh_ewald = _mm_set1_pd(fr->ic->sh_ewald);
129 ewtab = fr->ic->tabq_coul_FDV0;
130 ewtabscale = _mm_set1_pd(fr->ic->tabq_scale);
131 ewtabhalfspace = _mm_set1_pd(0.5/fr->ic->tabq_scale);
133 /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
134 rcutoff_scalar = fr->rcoulomb;
135 rcutoff = _mm_set1_pd(rcutoff_scalar);
136 rcutoff2 = _mm_mul_pd(rcutoff,rcutoff);
138 sh_vdw_invrcut6 = _mm_set1_pd(fr->ic->sh_invrc6);
139 rvdw = _mm_set1_pd(fr->rvdw);
141 /* Avoid stupid compiler warnings */
149 /* Start outer loop over neighborlists */
150 for(iidx=0; iidx<nri; iidx++)
152 /* Load shift vector for this list */
153 i_shift_offset = DIM*shiftidx[iidx];
155 /* Load limits for loop over neighbors */
156 j_index_start = jindex[iidx];
157 j_index_end = jindex[iidx+1];
159 /* Get outer coordinate index */
161 i_coord_offset = DIM*inr;
163 /* Load i particle coords and add shift vector */
164 gmx_mm_load_shift_and_1rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
166 fix0 = _mm_setzero_pd();
167 fiy0 = _mm_setzero_pd();
168 fiz0 = _mm_setzero_pd();
170 /* Load parameters for i particles */
171 iq0 = _mm_mul_pd(facel,_mm_load1_pd(charge+inr+0));
172 vdwioffset0 = 2*nvdwtype*vdwtype[inr+0];
174 /* Reset potential sums */
175 velecsum = _mm_setzero_pd();
176 vvdwsum = _mm_setzero_pd();
178 /* Start inner kernel loop */
179 for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
182 /* Get j neighbor index, and coordinate index */
185 j_coord_offsetA = DIM*jnrA;
186 j_coord_offsetB = DIM*jnrB;
188 /* load j atom coordinates */
189 gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
192 /* Calculate displacement vector */
193 dx00 = _mm_sub_pd(ix0,jx0);
194 dy00 = _mm_sub_pd(iy0,jy0);
195 dz00 = _mm_sub_pd(iz0,jz0);
197 /* Calculate squared distance and things based on it */
198 rsq00 = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
200 rinv00 = gmx_mm_invsqrt_pd(rsq00);
202 rinvsq00 = _mm_mul_pd(rinv00,rinv00);
204 /* Load parameters for j particles */
205 jq0 = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
206 vdwjidx0A = 2*vdwtype[jnrA+0];
207 vdwjidx0B = 2*vdwtype[jnrB+0];
209 /**************************
210 * CALCULATE INTERACTIONS *
211 **************************/
213 if (gmx_mm_any_lt(rsq00,rcutoff2))
216 r00 = _mm_mul_pd(rsq00,rinv00);
218 /* Compute parameters for interactions between i and j atoms */
219 qq00 = _mm_mul_pd(iq0,jq0);
220 gmx_mm_load_2pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,
221 vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
223 c6grid_00 = gmx_mm_load_2real_swizzle_pd(vdwgridparam+vdwioffset0+vdwjidx0A,
224 vdwgridparam+vdwioffset0+vdwjidx0B);
226 /* EWALD ELECTROSTATICS */
228 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
229 ewrt = _mm_mul_pd(r00,ewtabscale);
230 ewitab = _mm_cvttpd_epi32(ewrt);
231 eweps = _mm_sub_pd(ewrt,_mm_cvtepi32_pd(ewitab));
232 ewitab = _mm_slli_epi32(ewitab,2);
233 ewtabF = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,0) );
234 ewtabD = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,1) );
235 GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
236 ewtabV = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,0) +2);
237 ewtabFn = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,1) +2);
238 GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
239 felec = _mm_add_pd(ewtabF,_mm_mul_pd(eweps,ewtabD));
240 velec = _mm_sub_pd(ewtabV,_mm_mul_pd(_mm_mul_pd(ewtabhalfspace,eweps),_mm_add_pd(ewtabF,felec)));
241 velec = _mm_mul_pd(qq00,_mm_sub_pd(_mm_sub_pd(rinv00,sh_ewald),velec));
242 felec = _mm_mul_pd(_mm_mul_pd(qq00,rinv00),_mm_sub_pd(rinvsq00,felec));
244 /* Analytical LJ-PME */
245 rinvsix = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
246 ewcljrsq = _mm_mul_pd(ewclj2,rsq00);
247 ewclj6 = _mm_mul_pd(ewclj2,_mm_mul_pd(ewclj2,ewclj2));
248 exponent = gmx_simd_exp_d(ewcljrsq);
249 /* poly = exp(-(beta*r)^2) * (1 + (beta*r)^2 + (beta*r)^4 /2) */
250 poly = _mm_mul_pd(exponent,_mm_add_pd(_mm_sub_pd(one,ewcljrsq),_mm_mul_pd(_mm_mul_pd(ewcljrsq,ewcljrsq),one_half)));
251 /* vvdw6 = [C6 - C6grid * (1-poly)]/r6 */
252 vvdw6 = _mm_mul_pd(_mm_sub_pd(c6_00,_mm_mul_pd(c6grid_00,_mm_sub_pd(one,poly))),rinvsix);
253 vvdw12 = _mm_mul_pd(c12_00,_mm_mul_pd(rinvsix,rinvsix));
254 vvdw = _mm_sub_pd(_mm_mul_pd( _mm_sub_pd(vvdw12 , _mm_mul_pd(c12_00,_mm_mul_pd(sh_vdw_invrcut6,sh_vdw_invrcut6))),one_twelfth),
255 _mm_mul_pd( _mm_sub_pd(vvdw6,_mm_add_pd(_mm_mul_pd(c6_00,sh_vdw_invrcut6),_mm_mul_pd(c6grid_00,sh_lj_ewald))),one_sixth));
256 /* fvdw = vvdw12/r - (vvdw6/r + (C6grid * exponent * beta^6)/r) */
257 fvdw = _mm_mul_pd(_mm_sub_pd(vvdw12,_mm_sub_pd(vvdw6,_mm_mul_pd(_mm_mul_pd(c6grid_00,one_sixth),_mm_mul_pd(exponent,ewclj6)))),rinvsq00);
259 cutoff_mask = _mm_cmplt_pd(rsq00,rcutoff2);
261 /* Update potential sum for this i atom from the interaction with this j atom. */
262 velec = _mm_and_pd(velec,cutoff_mask);
263 velecsum = _mm_add_pd(velecsum,velec);
264 vvdw = _mm_and_pd(vvdw,cutoff_mask);
265 vvdwsum = _mm_add_pd(vvdwsum,vvdw);
267 fscal = _mm_add_pd(felec,fvdw);
269 fscal = _mm_and_pd(fscal,cutoff_mask);
271 /* Calculate temporary vectorial force */
272 tx = _mm_mul_pd(fscal,dx00);
273 ty = _mm_mul_pd(fscal,dy00);
274 tz = _mm_mul_pd(fscal,dz00);
276 /* Update vectorial force */
277 fix0 = _mm_add_pd(fix0,tx);
278 fiy0 = _mm_add_pd(fiy0,ty);
279 fiz0 = _mm_add_pd(fiz0,tz);
281 gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,tx,ty,tz);
285 /* Inner loop uses 82 flops */
292 j_coord_offsetA = DIM*jnrA;
294 /* load j atom coordinates */
295 gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
298 /* Calculate displacement vector */
299 dx00 = _mm_sub_pd(ix0,jx0);
300 dy00 = _mm_sub_pd(iy0,jy0);
301 dz00 = _mm_sub_pd(iz0,jz0);
303 /* Calculate squared distance and things based on it */
304 rsq00 = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
306 rinv00 = gmx_mm_invsqrt_pd(rsq00);
308 rinvsq00 = _mm_mul_pd(rinv00,rinv00);
310 /* Load parameters for j particles */
311 jq0 = _mm_load_sd(charge+jnrA+0);
312 vdwjidx0A = 2*vdwtype[jnrA+0];
314 /**************************
315 * CALCULATE INTERACTIONS *
316 **************************/
318 if (gmx_mm_any_lt(rsq00,rcutoff2))
321 r00 = _mm_mul_pd(rsq00,rinv00);
323 /* Compute parameters for interactions between i and j atoms */
324 qq00 = _mm_mul_pd(iq0,jq0);
325 gmx_mm_load_1pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,&c6_00,&c12_00);
327 c6grid_00 = gmx_mm_load_1real_pd(vdwgridparam+vdwioffset0+vdwjidx0A);
329 /* EWALD ELECTROSTATICS */
331 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
332 ewrt = _mm_mul_pd(r00,ewtabscale);
333 ewitab = _mm_cvttpd_epi32(ewrt);
334 eweps = _mm_sub_pd(ewrt,_mm_cvtepi32_pd(ewitab));
335 ewitab = _mm_slli_epi32(ewitab,2);
336 ewtabF = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,0) );
337 ewtabD = _mm_setzero_pd();
338 GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
339 ewtabV = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,0) +2);
340 ewtabFn = _mm_setzero_pd();
341 GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
342 felec = _mm_add_pd(ewtabF,_mm_mul_pd(eweps,ewtabD));
343 velec = _mm_sub_pd(ewtabV,_mm_mul_pd(_mm_mul_pd(ewtabhalfspace,eweps),_mm_add_pd(ewtabF,felec)));
344 velec = _mm_mul_pd(qq00,_mm_sub_pd(_mm_sub_pd(rinv00,sh_ewald),velec));
345 felec = _mm_mul_pd(_mm_mul_pd(qq00,rinv00),_mm_sub_pd(rinvsq00,felec));
347 /* Analytical LJ-PME */
348 rinvsix = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
349 ewcljrsq = _mm_mul_pd(ewclj2,rsq00);
350 ewclj6 = _mm_mul_pd(ewclj2,_mm_mul_pd(ewclj2,ewclj2));
351 exponent = gmx_simd_exp_d(ewcljrsq);
352 /* poly = exp(-(beta*r)^2) * (1 + (beta*r)^2 + (beta*r)^4 /2) */
353 poly = _mm_mul_pd(exponent,_mm_add_pd(_mm_sub_pd(one,ewcljrsq),_mm_mul_pd(_mm_mul_pd(ewcljrsq,ewcljrsq),one_half)));
354 /* vvdw6 = [C6 - C6grid * (1-poly)]/r6 */
355 vvdw6 = _mm_mul_pd(_mm_sub_pd(c6_00,_mm_mul_pd(c6grid_00,_mm_sub_pd(one,poly))),rinvsix);
356 vvdw12 = _mm_mul_pd(c12_00,_mm_mul_pd(rinvsix,rinvsix));
357 vvdw = _mm_sub_pd(_mm_mul_pd( _mm_sub_pd(vvdw12 , _mm_mul_pd(c12_00,_mm_mul_pd(sh_vdw_invrcut6,sh_vdw_invrcut6))),one_twelfth),
358 _mm_mul_pd( _mm_sub_pd(vvdw6,_mm_add_pd(_mm_mul_pd(c6_00,sh_vdw_invrcut6),_mm_mul_pd(c6grid_00,sh_lj_ewald))),one_sixth));
359 /* fvdw = vvdw12/r - (vvdw6/r + (C6grid * exponent * beta^6)/r) */
360 fvdw = _mm_mul_pd(_mm_sub_pd(vvdw12,_mm_sub_pd(vvdw6,_mm_mul_pd(_mm_mul_pd(c6grid_00,one_sixth),_mm_mul_pd(exponent,ewclj6)))),rinvsq00);
362 cutoff_mask = _mm_cmplt_pd(rsq00,rcutoff2);
364 /* Update potential sum for this i atom from the interaction with this j atom. */
365 velec = _mm_and_pd(velec,cutoff_mask);
366 velec = _mm_unpacklo_pd(velec,_mm_setzero_pd());
367 velecsum = _mm_add_pd(velecsum,velec);
368 vvdw = _mm_and_pd(vvdw,cutoff_mask);
369 vvdw = _mm_unpacklo_pd(vvdw,_mm_setzero_pd());
370 vvdwsum = _mm_add_pd(vvdwsum,vvdw);
372 fscal = _mm_add_pd(felec,fvdw);
374 fscal = _mm_and_pd(fscal,cutoff_mask);
376 fscal = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
378 /* Calculate temporary vectorial force */
379 tx = _mm_mul_pd(fscal,dx00);
380 ty = _mm_mul_pd(fscal,dy00);
381 tz = _mm_mul_pd(fscal,dz00);
383 /* Update vectorial force */
384 fix0 = _mm_add_pd(fix0,tx);
385 fiy0 = _mm_add_pd(fiy0,ty);
386 fiz0 = _mm_add_pd(fiz0,tz);
388 gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,tx,ty,tz);
392 /* Inner loop uses 82 flops */
395 /* End of innermost loop */
397 gmx_mm_update_iforce_1atom_swizzle_pd(fix0,fiy0,fiz0,
398 f+i_coord_offset,fshift+i_shift_offset);
401 /* Update potential energies */
402 gmx_mm_update_1pot_pd(velecsum,kernel_data->energygrp_elec+ggid);
403 gmx_mm_update_1pot_pd(vvdwsum,kernel_data->energygrp_vdw+ggid);
405 /* Increment number of inner iterations */
406 inneriter += j_index_end - j_index_start;
408 /* Outer loop uses 9 flops */
411 /* Increment number of outer iterations */
414 /* Update outer/inner flops */
416 inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_VF,outeriter*9 + inneriter*82);
419 * Gromacs nonbonded kernel: nb_kernel_ElecEwSh_VdwLJEwSh_GeomP1P1_F_sse2_double
420 * Electrostatics interaction: Ewald
421 * VdW interaction: LJEwald
422 * Geometry: Particle-Particle
423 * Calculate force/pot: Force
426 nb_kernel_ElecEwSh_VdwLJEwSh_GeomP1P1_F_sse2_double
427 (t_nblist * gmx_restrict nlist,
428 rvec * gmx_restrict xx,
429 rvec * gmx_restrict ff,
430 t_forcerec * gmx_restrict fr,
431 t_mdatoms * gmx_restrict mdatoms,
432 nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
433 t_nrnb * gmx_restrict nrnb)
435 /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
436 * just 0 for non-waters.
437 * Suffixes A,B refer to j loop unrolling done with SSE double precision, e.g. for the two different
438 * jnr indices corresponding to data put in the four positions in the SIMD register.
440 int i_shift_offset,i_coord_offset,outeriter,inneriter;
441 int j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
443 int j_coord_offsetA,j_coord_offsetB;
444 int *iinr,*jindex,*jjnr,*shiftidx,*gid;
446 real *shiftvec,*fshift,*x,*f;
447 __m128d tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
449 __m128d ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
450 int vdwjidx0A,vdwjidx0B;
451 __m128d jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
452 __m128d dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
453 __m128d velec,felec,velecsum,facel,crf,krf,krf2;
456 __m128d rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
459 __m128d one_sixth = _mm_set1_pd(1.0/6.0);
460 __m128d one_twelfth = _mm_set1_pd(1.0/12.0);
462 __m128d ewclj,ewclj2,ewclj6,ewcljrsq,poly,exponent,f6A,f6B,sh_lj_ewald;
464 __m128d one_half = _mm_set1_pd(0.5);
465 __m128d minus_one = _mm_set1_pd(-1.0);
467 __m128d ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
469 __m128d dummy_mask,cutoff_mask;
470 __m128d signbit = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
471 __m128d one = _mm_set1_pd(1.0);
472 __m128d two = _mm_set1_pd(2.0);
478 jindex = nlist->jindex;
480 shiftidx = nlist->shift;
482 shiftvec = fr->shift_vec[0];
483 fshift = fr->fshift[0];
484 facel = _mm_set1_pd(fr->epsfac);
485 charge = mdatoms->chargeA;
486 nvdwtype = fr->ntype;
488 vdwtype = mdatoms->typeA;
489 vdwgridparam = fr->ljpme_c6grid;
490 sh_lj_ewald = _mm_set1_pd(fr->ic->sh_lj_ewald);
491 ewclj = _mm_set1_pd(fr->ewaldcoeff_lj);
492 ewclj2 = _mm_mul_pd(minus_one,_mm_mul_pd(ewclj,ewclj));
494 sh_ewald = _mm_set1_pd(fr->ic->sh_ewald);
495 ewtab = fr->ic->tabq_coul_F;
496 ewtabscale = _mm_set1_pd(fr->ic->tabq_scale);
497 ewtabhalfspace = _mm_set1_pd(0.5/fr->ic->tabq_scale);
499 /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
500 rcutoff_scalar = fr->rcoulomb;
501 rcutoff = _mm_set1_pd(rcutoff_scalar);
502 rcutoff2 = _mm_mul_pd(rcutoff,rcutoff);
504 sh_vdw_invrcut6 = _mm_set1_pd(fr->ic->sh_invrc6);
505 rvdw = _mm_set1_pd(fr->rvdw);
507 /* Avoid stupid compiler warnings */
515 /* Start outer loop over neighborlists */
516 for(iidx=0; iidx<nri; iidx++)
518 /* Load shift vector for this list */
519 i_shift_offset = DIM*shiftidx[iidx];
521 /* Load limits for loop over neighbors */
522 j_index_start = jindex[iidx];
523 j_index_end = jindex[iidx+1];
525 /* Get outer coordinate index */
527 i_coord_offset = DIM*inr;
529 /* Load i particle coords and add shift vector */
530 gmx_mm_load_shift_and_1rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
532 fix0 = _mm_setzero_pd();
533 fiy0 = _mm_setzero_pd();
534 fiz0 = _mm_setzero_pd();
536 /* Load parameters for i particles */
537 iq0 = _mm_mul_pd(facel,_mm_load1_pd(charge+inr+0));
538 vdwioffset0 = 2*nvdwtype*vdwtype[inr+0];
540 /* Start inner kernel loop */
541 for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
544 /* Get j neighbor index, and coordinate index */
547 j_coord_offsetA = DIM*jnrA;
548 j_coord_offsetB = DIM*jnrB;
550 /* load j atom coordinates */
551 gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
554 /* Calculate displacement vector */
555 dx00 = _mm_sub_pd(ix0,jx0);
556 dy00 = _mm_sub_pd(iy0,jy0);
557 dz00 = _mm_sub_pd(iz0,jz0);
559 /* Calculate squared distance and things based on it */
560 rsq00 = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
562 rinv00 = gmx_mm_invsqrt_pd(rsq00);
564 rinvsq00 = _mm_mul_pd(rinv00,rinv00);
566 /* Load parameters for j particles */
567 jq0 = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
568 vdwjidx0A = 2*vdwtype[jnrA+0];
569 vdwjidx0B = 2*vdwtype[jnrB+0];
571 /**************************
572 * CALCULATE INTERACTIONS *
573 **************************/
575 if (gmx_mm_any_lt(rsq00,rcutoff2))
578 r00 = _mm_mul_pd(rsq00,rinv00);
580 /* Compute parameters for interactions between i and j atoms */
581 qq00 = _mm_mul_pd(iq0,jq0);
582 gmx_mm_load_2pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,
583 vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
585 c6grid_00 = gmx_mm_load_2real_swizzle_pd(vdwgridparam+vdwioffset0+vdwjidx0A,
586 vdwgridparam+vdwioffset0+vdwjidx0B);
588 /* EWALD ELECTROSTATICS */
590 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
591 ewrt = _mm_mul_pd(r00,ewtabscale);
592 ewitab = _mm_cvttpd_epi32(ewrt);
593 eweps = _mm_sub_pd(ewrt,_mm_cvtepi32_pd(ewitab));
594 gmx_mm_load_2pair_swizzle_pd(ewtab+gmx_mm_extract_epi32(ewitab,0),ewtab+gmx_mm_extract_epi32(ewitab,1),
596 felec = _mm_add_pd(_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF),_mm_mul_pd(eweps,ewtabFn));
597 felec = _mm_mul_pd(_mm_mul_pd(qq00,rinv00),_mm_sub_pd(rinvsq00,felec));
599 /* Analytical LJ-PME */
600 rinvsix = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
601 ewcljrsq = _mm_mul_pd(ewclj2,rsq00);
602 ewclj6 = _mm_mul_pd(ewclj2,_mm_mul_pd(ewclj2,ewclj2));
603 exponent = gmx_simd_exp_d(ewcljrsq);
604 /* poly = exp(-(beta*r)^2) * (1 + (beta*r)^2 + (beta*r)^4 /2) */
605 poly = _mm_mul_pd(exponent,_mm_add_pd(_mm_sub_pd(one,ewcljrsq),_mm_mul_pd(_mm_mul_pd(ewcljrsq,ewcljrsq),one_half)));
606 /* f6A = 6 * C6grid * (1 - poly) */
607 f6A = _mm_mul_pd(c6grid_00,_mm_sub_pd(one,poly));
608 /* f6B = C6grid * exponent * beta^6 */
609 f6B = _mm_mul_pd(_mm_mul_pd(c6grid_00,one_sixth),_mm_mul_pd(exponent,ewclj6));
610 /* fvdw = 12*C12/r13 - ((6*C6 - f6A)/r6 + f6B)/r */
611 fvdw = _mm_mul_pd(_mm_add_pd(_mm_mul_pd(_mm_sub_pd(_mm_mul_pd(c12_00,rinvsix),_mm_sub_pd(c6_00,f6A)),rinvsix),f6B),rinvsq00);
613 cutoff_mask = _mm_cmplt_pd(rsq00,rcutoff2);
615 fscal = _mm_add_pd(felec,fvdw);
617 fscal = _mm_and_pd(fscal,cutoff_mask);
619 /* Calculate temporary vectorial force */
620 tx = _mm_mul_pd(fscal,dx00);
621 ty = _mm_mul_pd(fscal,dy00);
622 tz = _mm_mul_pd(fscal,dz00);
624 /* Update vectorial force */
625 fix0 = _mm_add_pd(fix0,tx);
626 fiy0 = _mm_add_pd(fiy0,ty);
627 fiz0 = _mm_add_pd(fiz0,tz);
629 gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,tx,ty,tz);
633 /* Inner loop uses 62 flops */
640 j_coord_offsetA = DIM*jnrA;
642 /* load j atom coordinates */
643 gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
646 /* Calculate displacement vector */
647 dx00 = _mm_sub_pd(ix0,jx0);
648 dy00 = _mm_sub_pd(iy0,jy0);
649 dz00 = _mm_sub_pd(iz0,jz0);
651 /* Calculate squared distance and things based on it */
652 rsq00 = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
654 rinv00 = gmx_mm_invsqrt_pd(rsq00);
656 rinvsq00 = _mm_mul_pd(rinv00,rinv00);
658 /* Load parameters for j particles */
659 jq0 = _mm_load_sd(charge+jnrA+0);
660 vdwjidx0A = 2*vdwtype[jnrA+0];
662 /**************************
663 * CALCULATE INTERACTIONS *
664 **************************/
666 if (gmx_mm_any_lt(rsq00,rcutoff2))
669 r00 = _mm_mul_pd(rsq00,rinv00);
671 /* Compute parameters for interactions between i and j atoms */
672 qq00 = _mm_mul_pd(iq0,jq0);
673 gmx_mm_load_1pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,&c6_00,&c12_00);
675 c6grid_00 = gmx_mm_load_1real_pd(vdwgridparam+vdwioffset0+vdwjidx0A);
677 /* EWALD ELECTROSTATICS */
679 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
680 ewrt = _mm_mul_pd(r00,ewtabscale);
681 ewitab = _mm_cvttpd_epi32(ewrt);
682 eweps = _mm_sub_pd(ewrt,_mm_cvtepi32_pd(ewitab));
683 gmx_mm_load_1pair_swizzle_pd(ewtab+gmx_mm_extract_epi32(ewitab,0),&ewtabF,&ewtabFn);
684 felec = _mm_add_pd(_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF),_mm_mul_pd(eweps,ewtabFn));
685 felec = _mm_mul_pd(_mm_mul_pd(qq00,rinv00),_mm_sub_pd(rinvsq00,felec));
687 /* Analytical LJ-PME */
688 rinvsix = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
689 ewcljrsq = _mm_mul_pd(ewclj2,rsq00);
690 ewclj6 = _mm_mul_pd(ewclj2,_mm_mul_pd(ewclj2,ewclj2));
691 exponent = gmx_simd_exp_d(ewcljrsq);
692 /* poly = exp(-(beta*r)^2) * (1 + (beta*r)^2 + (beta*r)^4 /2) */
693 poly = _mm_mul_pd(exponent,_mm_add_pd(_mm_sub_pd(one,ewcljrsq),_mm_mul_pd(_mm_mul_pd(ewcljrsq,ewcljrsq),one_half)));
694 /* f6A = 6 * C6grid * (1 - poly) */
695 f6A = _mm_mul_pd(c6grid_00,_mm_sub_pd(one,poly));
696 /* f6B = C6grid * exponent * beta^6 */
697 f6B = _mm_mul_pd(_mm_mul_pd(c6grid_00,one_sixth),_mm_mul_pd(exponent,ewclj6));
698 /* fvdw = 12*C12/r13 - ((6*C6 - f6A)/r6 + f6B)/r */
699 fvdw = _mm_mul_pd(_mm_add_pd(_mm_mul_pd(_mm_sub_pd(_mm_mul_pd(c12_00,rinvsix),_mm_sub_pd(c6_00,f6A)),rinvsix),f6B),rinvsq00);
701 cutoff_mask = _mm_cmplt_pd(rsq00,rcutoff2);
703 fscal = _mm_add_pd(felec,fvdw);
705 fscal = _mm_and_pd(fscal,cutoff_mask);
707 fscal = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
709 /* Calculate temporary vectorial force */
710 tx = _mm_mul_pd(fscal,dx00);
711 ty = _mm_mul_pd(fscal,dy00);
712 tz = _mm_mul_pd(fscal,dz00);
714 /* Update vectorial force */
715 fix0 = _mm_add_pd(fix0,tx);
716 fiy0 = _mm_add_pd(fiy0,ty);
717 fiz0 = _mm_add_pd(fiz0,tz);
719 gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,tx,ty,tz);
723 /* Inner loop uses 62 flops */
726 /* End of innermost loop */
728 gmx_mm_update_iforce_1atom_swizzle_pd(fix0,fiy0,fiz0,
729 f+i_coord_offset,fshift+i_shift_offset);
731 /* Increment number of inner iterations */
732 inneriter += j_index_end - j_index_start;
734 /* Outer loop uses 7 flops */
737 /* Increment number of outer iterations */
740 /* Update outer/inner flops */
742 inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_F,outeriter*7 + inneriter*62);