?  search:
summary | log | graphiclog1 | graphiclog2 | commit | commitdiff | tree | refs | edit | fork
blame | history | raw | HEAD
Removed simple.h from nb_kernel_sse2_XX
[gromacs.git] / src / gromacs / gmxlib / nonbonded / nb_kernel_sse2_double / nb_kernel_ElecEwSh_VdwLJEwSh_GeomW4P1_sse2_double.c
blob7ef30dca0b872f58b31593b3ed2a66c9c87f99e9
1 /*
2 * This file is part of the GROMACS molecular simulation package.
3 *
4 * Copyright (c) 2012,2013,2014,2015, 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.
8 *
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.
13 *
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.
18 *
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.
23 *
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.
31 *
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.
34 */
35 /*
36 * Note: this file was generated by the GROMACS sse2_double kernel generator.
37 */
38 #include "gmxpre.h"
39
40 #include "config.h"
41
42 #include <math.h>
43
44 #include "../nb_kernel.h"
45 #include "gromacs/math/vec.h"
46 #include "gromacs/legacyheaders/nrnb.h"
47
48 #include "gromacs/simd/math_x86_sse2_double.h"
49 #include "kernelutil_x86_sse2_double.h"
50
51 /*
52 * Gromacs nonbonded kernel: nb_kernel_ElecEwSh_VdwLJEwSh_GeomW4P1_VF_sse2_double
53 * Electrostatics interaction: Ewald
54 * VdW interaction: LJEwald
55 * Geometry: Water4-Particle
56 * Calculate force/pot: PotentialAndForce
57 */
58 void
59 nb_kernel_ElecEwSh_VdwLJEwSh_GeomW4P1_VF_sse2_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)
67 {
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 SSE double precision, e.g. for the two different
71 * jnr indices corresponding to data put in the four positions in the SIMD register.
72 */
73 int i_shift_offset,i_coord_offset,outeriter,inneriter;
74 int j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
75 int jnrA,jnrB;
76 int j_coord_offsetA,j_coord_offsetB;
77 int *iinr,*jindex,*jjnr,*shiftidx,*gid;
78 real rcutoff_scalar;
79 real *shiftvec,*fshift,*x,*f;
80 __m128d tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
81 int vdwioffset0;
82 __m128d ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
83 int vdwioffset1;
84 __m128d ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
85 int vdwioffset2;
86 __m128d ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
87 int vdwioffset3;
88 __m128d ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
89 int vdwjidx0A,vdwjidx0B;
90 __m128d jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
91 __m128d dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
92 __m128d dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
93 __m128d dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
94 __m128d dx30,dy30,dz30,rsq30,rinv30,rinvsq30,r30,qq30,c6_30,c12_30;
95 __m128d velec,felec,velecsum,facel,crf,krf,krf2;
96 real *charge;
97 int nvdwtype;
98 __m128d rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
99 int *vdwtype;
100 real *vdwparam;
101 __m128d one_sixth = _mm_set1_pd(1.0/6.0);
102 __m128d one_twelfth = _mm_set1_pd(1.0/12.0);
103 __m128d c6grid_00;
104 __m128d c6grid_10;
105 __m128d c6grid_20;
106 __m128d c6grid_30;
107 __m128d ewclj,ewclj2,ewclj6,ewcljrsq,poly,exponent,f6A,f6B,sh_lj_ewald;
108 real *vdwgridparam;
109 __m128d one_half = _mm_set1_pd(0.5);
110 __m128d minus_one = _mm_set1_pd(-1.0);
111 __m128i ewitab;
112 __m128d ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
113 real *ewtab;
114 __m128d dummy_mask,cutoff_mask;
115 __m128d signbit = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
116 __m128d one = _mm_set1_pd(1.0);
117 __m128d two = _mm_set1_pd(2.0);
118 x = xx[0];
119 f = ff[0];
120
121 nri = nlist->nri;
122 iinr = nlist->iinr;
123 jindex = nlist->jindex;
124 jjnr = nlist->jjnr;
125 shiftidx = nlist->shift;
126 gid = nlist->gid;
127 shiftvec = fr->shift_vec[0];
128 fshift = fr->fshift[0];
129 facel = _mm_set1_pd(fr->epsfac);
130 charge = mdatoms->chargeA;
131 nvdwtype = fr->ntype;
132 vdwparam = fr->nbfp;
133 vdwtype = mdatoms->typeA;
134 vdwgridparam = fr->ljpme_c6grid;
135 sh_lj_ewald = _mm_set1_pd(fr->ic->sh_lj_ewald);
136 ewclj = _mm_set1_pd(fr->ewaldcoeff_lj);
137 ewclj2 = _mm_mul_pd(minus_one,_mm_mul_pd(ewclj,ewclj));
138
139 sh_ewald = _mm_set1_pd(fr->ic->sh_ewald);
140 ewtab = fr->ic->tabq_coul_FDV0;
141 ewtabscale = _mm_set1_pd(fr->ic->tabq_scale);
142 ewtabhalfspace = _mm_set1_pd(0.5/fr->ic->tabq_scale);
143
144 /* Setup water-specific parameters */
145 inr = nlist->iinr[0];
146 iq1 = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+1]));
147 iq2 = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+2]));
148 iq3 = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+3]));
149 vdwioffset0 = 2*nvdwtype*vdwtype[inr+0];
150
151 /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
152 rcutoff_scalar = fr->rcoulomb;
153 rcutoff = _mm_set1_pd(rcutoff_scalar);
154 rcutoff2 = _mm_mul_pd(rcutoff,rcutoff);
155
156 sh_vdw_invrcut6 = _mm_set1_pd(fr->ic->sh_invrc6);
157 rvdw = _mm_set1_pd(fr->rvdw);
158
159 /* Avoid stupid compiler warnings */
160 jnrA = jnrB = 0;
161 j_coord_offsetA = 0;
162 j_coord_offsetB = 0;
163
164 outeriter = 0;
165 inneriter = 0;
166
167 /* Start outer loop over neighborlists */
168 for(iidx=0; iidx<nri; iidx++)
169 {
170 /* Load shift vector for this list */
171 i_shift_offset = DIM*shiftidx[iidx];
172
173 /* Load limits for loop over neighbors */
174 j_index_start = jindex[iidx];
175 j_index_end = jindex[iidx+1];
176
177 /* Get outer coordinate index */
178 inr = iinr[iidx];
179 i_coord_offset = DIM*inr;
180
181 /* Load i particle coords and add shift vector */
182 gmx_mm_load_shift_and_4rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
183 &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
184
185 fix0 = _mm_setzero_pd();
186 fiy0 = _mm_setzero_pd();
187 fiz0 = _mm_setzero_pd();
188 fix1 = _mm_setzero_pd();
189 fiy1 = _mm_setzero_pd();
190 fiz1 = _mm_setzero_pd();
191 fix2 = _mm_setzero_pd();
192 fiy2 = _mm_setzero_pd();
193 fiz2 = _mm_setzero_pd();
194 fix3 = _mm_setzero_pd();
195 fiy3 = _mm_setzero_pd();
196 fiz3 = _mm_setzero_pd();
197
198 /* Reset potential sums */
199 velecsum = _mm_setzero_pd();
200 vvdwsum = _mm_setzero_pd();
201
202 /* Start inner kernel loop */
203 for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
204 {
205
206 /* Get j neighbor index, and coordinate index */
207 jnrA = jjnr[jidx];
208 jnrB = jjnr[jidx+1];
209 j_coord_offsetA = DIM*jnrA;
210 j_coord_offsetB = DIM*jnrB;
211
212 /* load j atom coordinates */
213 gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
214 &jx0,&jy0,&jz0);
215
216 /* Calculate displacement vector */
217 dx00 = _mm_sub_pd(ix0,jx0);
218 dy00 = _mm_sub_pd(iy0,jy0);
219 dz00 = _mm_sub_pd(iz0,jz0);
220 dx10 = _mm_sub_pd(ix1,jx0);
221 dy10 = _mm_sub_pd(iy1,jy0);
222 dz10 = _mm_sub_pd(iz1,jz0);
223 dx20 = _mm_sub_pd(ix2,jx0);
224 dy20 = _mm_sub_pd(iy2,jy0);
225 dz20 = _mm_sub_pd(iz2,jz0);
226 dx30 = _mm_sub_pd(ix3,jx0);
227 dy30 = _mm_sub_pd(iy3,jy0);
228 dz30 = _mm_sub_pd(iz3,jz0);
229
230 /* Calculate squared distance and things based on it */
231 rsq00 = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
232 rsq10 = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
233 rsq20 = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
234 rsq30 = gmx_mm_calc_rsq_pd(dx30,dy30,dz30);
235
236 rinv00 = gmx_mm_invsqrt_pd(rsq00);
237 rinv10 = gmx_mm_invsqrt_pd(rsq10);
238 rinv20 = gmx_mm_invsqrt_pd(rsq20);
239 rinv30 = gmx_mm_invsqrt_pd(rsq30);
240
241 rinvsq00 = _mm_mul_pd(rinv00,rinv00);
242 rinvsq10 = _mm_mul_pd(rinv10,rinv10);
243 rinvsq20 = _mm_mul_pd(rinv20,rinv20);
244 rinvsq30 = _mm_mul_pd(rinv30,rinv30);
245
246 /* Load parameters for j particles */
247 jq0 = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
248 vdwjidx0A = 2*vdwtype[jnrA+0];
249 vdwjidx0B = 2*vdwtype[jnrB+0];
250
251 fjx0 = _mm_setzero_pd();
252 fjy0 = _mm_setzero_pd();
253 fjz0 = _mm_setzero_pd();
254
255 /**************************
256 * CALCULATE INTERACTIONS *
257 **************************/
258
259 if (gmx_mm_any_lt(rsq00,rcutoff2))
260 {
261
262 r00 = _mm_mul_pd(rsq00,rinv00);
263
264 /* Compute parameters for interactions between i and j atoms */
265 gmx_mm_load_2pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,
266 vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
267
268 c6grid_00 = gmx_mm_load_2real_swizzle_pd(vdwgridparam+vdwioffset0+vdwjidx0A,
269 vdwgridparam+vdwioffset0+vdwjidx0B);
270
271 /* Analytical LJ-PME */
272 rinvsix = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
273 ewcljrsq = _mm_mul_pd(ewclj2,rsq00);
274 ewclj6 = _mm_mul_pd(ewclj2,_mm_mul_pd(ewclj2,ewclj2));
275 exponent = gmx_simd_exp_d(ewcljrsq);
276 /* poly = exp(-(beta*r)^2) * (1 + (beta*r)^2 + (beta*r)^4 /2) */
277 poly = _mm_mul_pd(exponent,_mm_add_pd(_mm_sub_pd(one,ewcljrsq),_mm_mul_pd(_mm_mul_pd(ewcljrsq,ewcljrsq),one_half)));
278 /* vvdw6 = [C6 - C6grid * (1-poly)]/r6 */
279 vvdw6 = _mm_mul_pd(_mm_sub_pd(c6_00,_mm_mul_pd(c6grid_00,_mm_sub_pd(one,poly))),rinvsix);
280 vvdw12 = _mm_mul_pd(c12_00,_mm_mul_pd(rinvsix,rinvsix));
281 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),
282 _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));
283 /* fvdw = vvdw12/r - (vvdw6/r + (C6grid * exponent * beta^6)/r) */
284 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);
285
286 cutoff_mask = _mm_cmplt_pd(rsq00,rcutoff2);
287
288 /* Update potential sum for this i atom from the interaction with this j atom. */
289 vvdw = _mm_and_pd(vvdw,cutoff_mask);
290 vvdwsum = _mm_add_pd(vvdwsum,vvdw);
291
292 fscal = fvdw;
293
294 fscal = _mm_and_pd(fscal,cutoff_mask);
295
296 /* Calculate temporary vectorial force */
297 tx = _mm_mul_pd(fscal,dx00);
298 ty = _mm_mul_pd(fscal,dy00);
299 tz = _mm_mul_pd(fscal,dz00);
300
301 /* Update vectorial force */
302 fix0 = _mm_add_pd(fix0,tx);
303 fiy0 = _mm_add_pd(fiy0,ty);
304 fiz0 = _mm_add_pd(fiz0,tz);
305
306 fjx0 = _mm_add_pd(fjx0,tx);
307 fjy0 = _mm_add_pd(fjy0,ty);
308 fjz0 = _mm_add_pd(fjz0,tz);
309
310 }
311
312 /**************************
313 * CALCULATE INTERACTIONS *
314 **************************/
315
316 if (gmx_mm_any_lt(rsq10,rcutoff2))
317 {
318
319 r10 = _mm_mul_pd(rsq10,rinv10);
320
321 /* Compute parameters for interactions between i and j atoms */
322 qq10 = _mm_mul_pd(iq1,jq0);
323
324 /* EWALD ELECTROSTATICS */
325
326 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
327 ewrt = _mm_mul_pd(r10,ewtabscale);
328 ewitab = _mm_cvttpd_epi32(ewrt);
329 eweps = _mm_sub_pd(ewrt,_mm_cvtepi32_pd(ewitab));
330 ewitab = _mm_slli_epi32(ewitab,2);
331 ewtabF = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,0) );
332 ewtabD = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,1) );
333 GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
334 ewtabV = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,0) +2);
335 ewtabFn = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,1) +2);
336 GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
337 felec = _mm_add_pd(ewtabF,_mm_mul_pd(eweps,ewtabD));
338 velec = _mm_sub_pd(ewtabV,_mm_mul_pd(_mm_mul_pd(ewtabhalfspace,eweps),_mm_add_pd(ewtabF,felec)));
339 velec = _mm_mul_pd(qq10,_mm_sub_pd(_mm_sub_pd(rinv10,sh_ewald),velec));
340 felec = _mm_mul_pd(_mm_mul_pd(qq10,rinv10),_mm_sub_pd(rinvsq10,felec));
341
342 cutoff_mask = _mm_cmplt_pd(rsq10,rcutoff2);
343
344 /* Update potential sum for this i atom from the interaction with this j atom. */
345 velec = _mm_and_pd(velec,cutoff_mask);
346 velecsum = _mm_add_pd(velecsum,velec);
347
348 fscal = felec;
349
350 fscal = _mm_and_pd(fscal,cutoff_mask);
351
352 /* Calculate temporary vectorial force */
353 tx = _mm_mul_pd(fscal,dx10);
354 ty = _mm_mul_pd(fscal,dy10);
355 tz = _mm_mul_pd(fscal,dz10);
356
357 /* Update vectorial force */
358 fix1 = _mm_add_pd(fix1,tx);
359 fiy1 = _mm_add_pd(fiy1,ty);
360 fiz1 = _mm_add_pd(fiz1,tz);
361
362 fjx0 = _mm_add_pd(fjx0,tx);
363 fjy0 = _mm_add_pd(fjy0,ty);
364 fjz0 = _mm_add_pd(fjz0,tz);
365
366 }
367
368 /**************************
369 * CALCULATE INTERACTIONS *
370 **************************/
371
372 if (gmx_mm_any_lt(rsq20,rcutoff2))
373 {
374
375 r20 = _mm_mul_pd(rsq20,rinv20);
376
377 /* Compute parameters for interactions between i and j atoms */
378 qq20 = _mm_mul_pd(iq2,jq0);
379
380 /* EWALD ELECTROSTATICS */
381
382 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
383 ewrt = _mm_mul_pd(r20,ewtabscale);
384 ewitab = _mm_cvttpd_epi32(ewrt);
385 eweps = _mm_sub_pd(ewrt,_mm_cvtepi32_pd(ewitab));
386 ewitab = _mm_slli_epi32(ewitab,2);
387 ewtabF = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,0) );
388 ewtabD = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,1) );
389 GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
390 ewtabV = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,0) +2);
391 ewtabFn = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,1) +2);
392 GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
393 felec = _mm_add_pd(ewtabF,_mm_mul_pd(eweps,ewtabD));
394 velec = _mm_sub_pd(ewtabV,_mm_mul_pd(_mm_mul_pd(ewtabhalfspace,eweps),_mm_add_pd(ewtabF,felec)));
395 velec = _mm_mul_pd(qq20,_mm_sub_pd(_mm_sub_pd(rinv20,sh_ewald),velec));
396 felec = _mm_mul_pd(_mm_mul_pd(qq20,rinv20),_mm_sub_pd(rinvsq20,felec));
397
398 cutoff_mask = _mm_cmplt_pd(rsq20,rcutoff2);
399
400 /* Update potential sum for this i atom from the interaction with this j atom. */
401 velec = _mm_and_pd(velec,cutoff_mask);
402 velecsum = _mm_add_pd(velecsum,velec);
403
404 fscal = felec;
405
406 fscal = _mm_and_pd(fscal,cutoff_mask);
407
408 /* Calculate temporary vectorial force */
409 tx = _mm_mul_pd(fscal,dx20);
410 ty = _mm_mul_pd(fscal,dy20);
411 tz = _mm_mul_pd(fscal,dz20);
412
413 /* Update vectorial force */
414 fix2 = _mm_add_pd(fix2,tx);
415 fiy2 = _mm_add_pd(fiy2,ty);
416 fiz2 = _mm_add_pd(fiz2,tz);
417
418 fjx0 = _mm_add_pd(fjx0,tx);
419 fjy0 = _mm_add_pd(fjy0,ty);
420 fjz0 = _mm_add_pd(fjz0,tz);
421
422 }
423
424 /**************************
425 * CALCULATE INTERACTIONS *
426 **************************/
427
428 if (gmx_mm_any_lt(rsq30,rcutoff2))
429 {
430
431 r30 = _mm_mul_pd(rsq30,rinv30);
432
433 /* Compute parameters for interactions between i and j atoms */
434 qq30 = _mm_mul_pd(iq3,jq0);
435
436 /* EWALD ELECTROSTATICS */
437
438 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
439 ewrt = _mm_mul_pd(r30,ewtabscale);
440 ewitab = _mm_cvttpd_epi32(ewrt);
441 eweps = _mm_sub_pd(ewrt,_mm_cvtepi32_pd(ewitab));
442 ewitab = _mm_slli_epi32(ewitab,2);
443 ewtabF = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,0) );
444 ewtabD = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,1) );
445 GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
446 ewtabV = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,0) +2);
447 ewtabFn = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,1) +2);
448 GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
449 felec = _mm_add_pd(ewtabF,_mm_mul_pd(eweps,ewtabD));
450 velec = _mm_sub_pd(ewtabV,_mm_mul_pd(_mm_mul_pd(ewtabhalfspace,eweps),_mm_add_pd(ewtabF,felec)));
451 velec = _mm_mul_pd(qq30,_mm_sub_pd(_mm_sub_pd(rinv30,sh_ewald),velec));
452 felec = _mm_mul_pd(_mm_mul_pd(qq30,rinv30),_mm_sub_pd(rinvsq30,felec));
453
454 cutoff_mask = _mm_cmplt_pd(rsq30,rcutoff2);
455
456 /* Update potential sum for this i atom from the interaction with this j atom. */
457 velec = _mm_and_pd(velec,cutoff_mask);
458 velecsum = _mm_add_pd(velecsum,velec);
459
460 fscal = felec;
461
462 fscal = _mm_and_pd(fscal,cutoff_mask);
463
464 /* Calculate temporary vectorial force */
465 tx = _mm_mul_pd(fscal,dx30);
466 ty = _mm_mul_pd(fscal,dy30);
467 tz = _mm_mul_pd(fscal,dz30);
468
469 /* Update vectorial force */
470 fix3 = _mm_add_pd(fix3,tx);
471 fiy3 = _mm_add_pd(fiy3,ty);
472 fiz3 = _mm_add_pd(fiz3,tz);
473
474 fjx0 = _mm_add_pd(fjx0,tx);
475 fjy0 = _mm_add_pd(fjy0,ty);
476 fjz0 = _mm_add_pd(fjz0,tz);
477
478 }
479
480 gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,fjx0,fjy0,fjz0);
481
482 /* Inner loop uses 203 flops */
483 }
484
485 if(jidx<j_index_end)
486 {
487
488 jnrA = jjnr[jidx];
489 j_coord_offsetA = DIM*jnrA;
490
491 /* load j atom coordinates */
492 gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
493 &jx0,&jy0,&jz0);
494
495 /* Calculate displacement vector */
496 dx00 = _mm_sub_pd(ix0,jx0);
497 dy00 = _mm_sub_pd(iy0,jy0);
498 dz00 = _mm_sub_pd(iz0,jz0);
499 dx10 = _mm_sub_pd(ix1,jx0);
500 dy10 = _mm_sub_pd(iy1,jy0);
501 dz10 = _mm_sub_pd(iz1,jz0);
502 dx20 = _mm_sub_pd(ix2,jx0);
503 dy20 = _mm_sub_pd(iy2,jy0);
504 dz20 = _mm_sub_pd(iz2,jz0);
505 dx30 = _mm_sub_pd(ix3,jx0);
506 dy30 = _mm_sub_pd(iy3,jy0);
507 dz30 = _mm_sub_pd(iz3,jz0);
508
509 /* Calculate squared distance and things based on it */
510 rsq00 = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
511 rsq10 = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
512 rsq20 = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
513 rsq30 = gmx_mm_calc_rsq_pd(dx30,dy30,dz30);
514
515 rinv00 = gmx_mm_invsqrt_pd(rsq00);
516 rinv10 = gmx_mm_invsqrt_pd(rsq10);
517 rinv20 = gmx_mm_invsqrt_pd(rsq20);
518 rinv30 = gmx_mm_invsqrt_pd(rsq30);
519
520 rinvsq00 = _mm_mul_pd(rinv00,rinv00);
521 rinvsq10 = _mm_mul_pd(rinv10,rinv10);
522 rinvsq20 = _mm_mul_pd(rinv20,rinv20);
523 rinvsq30 = _mm_mul_pd(rinv30,rinv30);
524
525 /* Load parameters for j particles */
526 jq0 = _mm_load_sd(charge+jnrA+0);
527 vdwjidx0A = 2*vdwtype[jnrA+0];
528
529 fjx0 = _mm_setzero_pd();
530 fjy0 = _mm_setzero_pd();
531 fjz0 = _mm_setzero_pd();
532
533 /**************************
534 * CALCULATE INTERACTIONS *
535 **************************/
536
537 if (gmx_mm_any_lt(rsq00,rcutoff2))
538 {
539
540 r00 = _mm_mul_pd(rsq00,rinv00);
541
542 /* Compute parameters for interactions between i and j atoms */
543 gmx_mm_load_1pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,&c6_00,&c12_00);
544
545 c6grid_00 = gmx_mm_load_1real_pd(vdwgridparam+vdwioffset0+vdwjidx0A);
546
547 /* Analytical LJ-PME */
548 rinvsix = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
549 ewcljrsq = _mm_mul_pd(ewclj2,rsq00);
550 ewclj6 = _mm_mul_pd(ewclj2,_mm_mul_pd(ewclj2,ewclj2));
551 exponent = gmx_simd_exp_d(ewcljrsq);
552 /* poly = exp(-(beta*r)^2) * (1 + (beta*r)^2 + (beta*r)^4 /2) */
553 poly = _mm_mul_pd(exponent,_mm_add_pd(_mm_sub_pd(one,ewcljrsq),_mm_mul_pd(_mm_mul_pd(ewcljrsq,ewcljrsq),one_half)));
554 /* vvdw6 = [C6 - C6grid * (1-poly)]/r6 */
555 vvdw6 = _mm_mul_pd(_mm_sub_pd(c6_00,_mm_mul_pd(c6grid_00,_mm_sub_pd(one,poly))),rinvsix);
556 vvdw12 = _mm_mul_pd(c12_00,_mm_mul_pd(rinvsix,rinvsix));
557 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),
558 _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));
559 /* fvdw = vvdw12/r - (vvdw6/r + (C6grid * exponent * beta^6)/r) */
560 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);
561
562 cutoff_mask = _mm_cmplt_pd(rsq00,rcutoff2);
563
564 /* Update potential sum for this i atom from the interaction with this j atom. */
565 vvdw = _mm_and_pd(vvdw,cutoff_mask);
566 vvdw = _mm_unpacklo_pd(vvdw,_mm_setzero_pd());
567 vvdwsum = _mm_add_pd(vvdwsum,vvdw);
568
569 fscal = fvdw;
570
571 fscal = _mm_and_pd(fscal,cutoff_mask);
572
573 fscal = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
574
575 /* Calculate temporary vectorial force */
576 tx = _mm_mul_pd(fscal,dx00);
577 ty = _mm_mul_pd(fscal,dy00);
578 tz = _mm_mul_pd(fscal,dz00);
579
580 /* Update vectorial force */
581 fix0 = _mm_add_pd(fix0,tx);
582 fiy0 = _mm_add_pd(fiy0,ty);
583 fiz0 = _mm_add_pd(fiz0,tz);
584
585 fjx0 = _mm_add_pd(fjx0,tx);
586 fjy0 = _mm_add_pd(fjy0,ty);
587 fjz0 = _mm_add_pd(fjz0,tz);
588
589 }
590
591 /**************************
592 * CALCULATE INTERACTIONS *
593 **************************/
594
595 if (gmx_mm_any_lt(rsq10,rcutoff2))
596 {
597
598 r10 = _mm_mul_pd(rsq10,rinv10);
599
600 /* Compute parameters for interactions between i and j atoms */
601 qq10 = _mm_mul_pd(iq1,jq0);
602
603 /* EWALD ELECTROSTATICS */
604
605 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
606 ewrt = _mm_mul_pd(r10,ewtabscale);
607 ewitab = _mm_cvttpd_epi32(ewrt);
608 eweps = _mm_sub_pd(ewrt,_mm_cvtepi32_pd(ewitab));
609 ewitab = _mm_slli_epi32(ewitab,2);
610 ewtabF = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,0) );
611 ewtabD = _mm_setzero_pd();
612 GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
613 ewtabV = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,0) +2);
614 ewtabFn = _mm_setzero_pd();
615 GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
616 felec = _mm_add_pd(ewtabF,_mm_mul_pd(eweps,ewtabD));
617 velec = _mm_sub_pd(ewtabV,_mm_mul_pd(_mm_mul_pd(ewtabhalfspace,eweps),_mm_add_pd(ewtabF,felec)));
618 velec = _mm_mul_pd(qq10,_mm_sub_pd(_mm_sub_pd(rinv10,sh_ewald),velec));
619 felec = _mm_mul_pd(_mm_mul_pd(qq10,rinv10),_mm_sub_pd(rinvsq10,felec));
620
621 cutoff_mask = _mm_cmplt_pd(rsq10,rcutoff2);
622
623 /* Update potential sum for this i atom from the interaction with this j atom. */
624 velec = _mm_and_pd(velec,cutoff_mask);
625 velec = _mm_unpacklo_pd(velec,_mm_setzero_pd());
626 velecsum = _mm_add_pd(velecsum,velec);
627
628 fscal = felec;
629
630 fscal = _mm_and_pd(fscal,cutoff_mask);
631
632 fscal = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
633
634 /* Calculate temporary vectorial force */
635 tx = _mm_mul_pd(fscal,dx10);
636 ty = _mm_mul_pd(fscal,dy10);
637 tz = _mm_mul_pd(fscal,dz10);
638
639 /* Update vectorial force */
640 fix1 = _mm_add_pd(fix1,tx);
641 fiy1 = _mm_add_pd(fiy1,ty);
642 fiz1 = _mm_add_pd(fiz1,tz);
643
644 fjx0 = _mm_add_pd(fjx0,tx);
645 fjy0 = _mm_add_pd(fjy0,ty);
646 fjz0 = _mm_add_pd(fjz0,tz);
647
648 }
649
650 /**************************
651 * CALCULATE INTERACTIONS *
652 **************************/
653
654 if (gmx_mm_any_lt(rsq20,rcutoff2))
655 {
656
657 r20 = _mm_mul_pd(rsq20,rinv20);
658
659 /* Compute parameters for interactions between i and j atoms */
660 qq20 = _mm_mul_pd(iq2,jq0);
661
662 /* EWALD ELECTROSTATICS */
663
664 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
665 ewrt = _mm_mul_pd(r20,ewtabscale);
666 ewitab = _mm_cvttpd_epi32(ewrt);
667 eweps = _mm_sub_pd(ewrt,_mm_cvtepi32_pd(ewitab));
668 ewitab = _mm_slli_epi32(ewitab,2);
669 ewtabF = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,0) );
670 ewtabD = _mm_setzero_pd();
671 GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
672 ewtabV = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,0) +2);
673 ewtabFn = _mm_setzero_pd();
674 GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
675 felec = _mm_add_pd(ewtabF,_mm_mul_pd(eweps,ewtabD));
676 velec = _mm_sub_pd(ewtabV,_mm_mul_pd(_mm_mul_pd(ewtabhalfspace,eweps),_mm_add_pd(ewtabF,felec)));
677 velec = _mm_mul_pd(qq20,_mm_sub_pd(_mm_sub_pd(rinv20,sh_ewald),velec));
678 felec = _mm_mul_pd(_mm_mul_pd(qq20,rinv20),_mm_sub_pd(rinvsq20,felec));
679
680 cutoff_mask = _mm_cmplt_pd(rsq20,rcutoff2);
681
682 /* Update potential sum for this i atom from the interaction with this j atom. */
683 velec = _mm_and_pd(velec,cutoff_mask);
684 velec = _mm_unpacklo_pd(velec,_mm_setzero_pd());
685 velecsum = _mm_add_pd(velecsum,velec);
686
687 fscal = felec;
688
689 fscal = _mm_and_pd(fscal,cutoff_mask);
690
691 fscal = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
692
693 /* Calculate temporary vectorial force */
694 tx = _mm_mul_pd(fscal,dx20);
695 ty = _mm_mul_pd(fscal,dy20);
696 tz = _mm_mul_pd(fscal,dz20);
697
698 /* Update vectorial force */
699 fix2 = _mm_add_pd(fix2,tx);
700 fiy2 = _mm_add_pd(fiy2,ty);
701 fiz2 = _mm_add_pd(fiz2,tz);
702
703 fjx0 = _mm_add_pd(fjx0,tx);
704 fjy0 = _mm_add_pd(fjy0,ty);
705 fjz0 = _mm_add_pd(fjz0,tz);
706
707 }
708
709 /**************************
710 * CALCULATE INTERACTIONS *
711 **************************/
712
713 if (gmx_mm_any_lt(rsq30,rcutoff2))
714 {
715
716 r30 = _mm_mul_pd(rsq30,rinv30);
717
718 /* Compute parameters for interactions between i and j atoms */
719 qq30 = _mm_mul_pd(iq3,jq0);
720
721 /* EWALD ELECTROSTATICS */
722
723 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
724 ewrt = _mm_mul_pd(r30,ewtabscale);
725 ewitab = _mm_cvttpd_epi32(ewrt);
726 eweps = _mm_sub_pd(ewrt,_mm_cvtepi32_pd(ewitab));
727 ewitab = _mm_slli_epi32(ewitab,2);
728 ewtabF = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,0) );
729 ewtabD = _mm_setzero_pd();
730 GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
731 ewtabV = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,0) +2);
732 ewtabFn = _mm_setzero_pd();
733 GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
734 felec = _mm_add_pd(ewtabF,_mm_mul_pd(eweps,ewtabD));
735 velec = _mm_sub_pd(ewtabV,_mm_mul_pd(_mm_mul_pd(ewtabhalfspace,eweps),_mm_add_pd(ewtabF,felec)));
736 velec = _mm_mul_pd(qq30,_mm_sub_pd(_mm_sub_pd(rinv30,sh_ewald),velec));
737 felec = _mm_mul_pd(_mm_mul_pd(qq30,rinv30),_mm_sub_pd(rinvsq30,felec));
738
739 cutoff_mask = _mm_cmplt_pd(rsq30,rcutoff2);
740
741 /* Update potential sum for this i atom from the interaction with this j atom. */
742 velec = _mm_and_pd(velec,cutoff_mask);
743 velec = _mm_unpacklo_pd(velec,_mm_setzero_pd());
744 velecsum = _mm_add_pd(velecsum,velec);
745
746 fscal = felec;
747
748 fscal = _mm_and_pd(fscal,cutoff_mask);
749
750 fscal = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
751
752 /* Calculate temporary vectorial force */
753 tx = _mm_mul_pd(fscal,dx30);
754 ty = _mm_mul_pd(fscal,dy30);
755 tz = _mm_mul_pd(fscal,dz30);
756
757 /* Update vectorial force */
758 fix3 = _mm_add_pd(fix3,tx);
759 fiy3 = _mm_add_pd(fiy3,ty);
760 fiz3 = _mm_add_pd(fiz3,tz);
761
762 fjx0 = _mm_add_pd(fjx0,tx);
763 fjy0 = _mm_add_pd(fjy0,ty);
764 fjz0 = _mm_add_pd(fjz0,tz);
765
766 }
767
768 gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,fjx0,fjy0,fjz0);
769
770 /* Inner loop uses 203 flops */
771 }
772
773 /* End of innermost loop */
774
775 gmx_mm_update_iforce_4atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
776 f+i_coord_offset,fshift+i_shift_offset);
777
778 ggid = gid[iidx];
779 /* Update potential energies */
780 gmx_mm_update_1pot_pd(velecsum,kernel_data->energygrp_elec+ggid);
781 gmx_mm_update_1pot_pd(vvdwsum,kernel_data->energygrp_vdw+ggid);
782
783 /* Increment number of inner iterations */
784 inneriter += j_index_end - j_index_start;
785
786 /* Outer loop uses 26 flops */
787 }
788
789 /* Increment number of outer iterations */
790 outeriter += nri;
791
792 /* Update outer/inner flops */
793
794 inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W4_VF,outeriter*26 + inneriter*203);
795 }
796 /*
797 * Gromacs nonbonded kernel: nb_kernel_ElecEwSh_VdwLJEwSh_GeomW4P1_F_sse2_double
798 * Electrostatics interaction: Ewald
799 * VdW interaction: LJEwald
800 * Geometry: Water4-Particle
801 * Calculate force/pot: Force
802 */
803 void
804 nb_kernel_ElecEwSh_VdwLJEwSh_GeomW4P1_F_sse2_double
805 (t_nblist * gmx_restrict nlist,
806 rvec * gmx_restrict xx,
807 rvec * gmx_restrict ff,
808 t_forcerec * gmx_restrict fr,
809 t_mdatoms * gmx_restrict mdatoms,
810 nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
811 t_nrnb * gmx_restrict nrnb)
812 {
813 /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
814 * just 0 for non-waters.
815 * Suffixes A,B refer to j loop unrolling done with SSE double precision, e.g. for the two different
816 * jnr indices corresponding to data put in the four positions in the SIMD register.
817 */
818 int i_shift_offset,i_coord_offset,outeriter,inneriter;
819 int j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
820 int jnrA,jnrB;
821 int j_coord_offsetA,j_coord_offsetB;
822 int *iinr,*jindex,*jjnr,*shiftidx,*gid;
823 real rcutoff_scalar;
824 real *shiftvec,*fshift,*x,*f;
825 __m128d tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
826 int vdwioffset0;
827 __m128d ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
828 int vdwioffset1;
829 __m128d ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
830 int vdwioffset2;
831 __m128d ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
832 int vdwioffset3;
833 __m128d ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
834 int vdwjidx0A,vdwjidx0B;
835 __m128d jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
836 __m128d dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
837 __m128d dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
838 __m128d dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
839 __m128d dx30,dy30,dz30,rsq30,rinv30,rinvsq30,r30,qq30,c6_30,c12_30;
840 __m128d velec,felec,velecsum,facel,crf,krf,krf2;
841 real *charge;
842 int nvdwtype;
843 __m128d rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
844 int *vdwtype;
845 real *vdwparam;
846 __m128d one_sixth = _mm_set1_pd(1.0/6.0);
847 __m128d one_twelfth = _mm_set1_pd(1.0/12.0);
848 __m128d c6grid_00;
849 __m128d c6grid_10;
850 __m128d c6grid_20;
851 __m128d c6grid_30;
852 __m128d ewclj,ewclj2,ewclj6,ewcljrsq,poly,exponent,f6A,f6B,sh_lj_ewald;
853 real *vdwgridparam;
854 __m128d one_half = _mm_set1_pd(0.5);
855 __m128d minus_one = _mm_set1_pd(-1.0);
856 __m128i ewitab;
857 __m128d ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
858 real *ewtab;
859 __m128d dummy_mask,cutoff_mask;
860 __m128d signbit = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
861 __m128d one = _mm_set1_pd(1.0);
862 __m128d two = _mm_set1_pd(2.0);
863 x = xx[0];
864 f = ff[0];
865
866 nri = nlist->nri;
867 iinr = nlist->iinr;
868 jindex = nlist->jindex;
869 jjnr = nlist->jjnr;
870 shiftidx = nlist->shift;
871 gid = nlist->gid;
872 shiftvec = fr->shift_vec[0];
873 fshift = fr->fshift[0];
874 facel = _mm_set1_pd(fr->epsfac);
875 charge = mdatoms->chargeA;
876 nvdwtype = fr->ntype;
877 vdwparam = fr->nbfp;
878 vdwtype = mdatoms->typeA;
879 vdwgridparam = fr->ljpme_c6grid;
880 sh_lj_ewald = _mm_set1_pd(fr->ic->sh_lj_ewald);
881 ewclj = _mm_set1_pd(fr->ewaldcoeff_lj);
882 ewclj2 = _mm_mul_pd(minus_one,_mm_mul_pd(ewclj,ewclj));
883
884 sh_ewald = _mm_set1_pd(fr->ic->sh_ewald);
885 ewtab = fr->ic->tabq_coul_F;
886 ewtabscale = _mm_set1_pd(fr->ic->tabq_scale);
887 ewtabhalfspace = _mm_set1_pd(0.5/fr->ic->tabq_scale);
888
889 /* Setup water-specific parameters */
890 inr = nlist->iinr[0];
891 iq1 = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+1]));
892 iq2 = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+2]));
893 iq3 = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+3]));
894 vdwioffset0 = 2*nvdwtype*vdwtype[inr+0];
895
896 /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
897 rcutoff_scalar = fr->rcoulomb;
898 rcutoff = _mm_set1_pd(rcutoff_scalar);
899 rcutoff2 = _mm_mul_pd(rcutoff,rcutoff);
900
901 sh_vdw_invrcut6 = _mm_set1_pd(fr->ic->sh_invrc6);
902 rvdw = _mm_set1_pd(fr->rvdw);
903
904 /* Avoid stupid compiler warnings */
905 jnrA = jnrB = 0;
906 j_coord_offsetA = 0;
907 j_coord_offsetB = 0;
908
909 outeriter = 0;
910 inneriter = 0;
911
912 /* Start outer loop over neighborlists */
913 for(iidx=0; iidx<nri; iidx++)
914 {
915 /* Load shift vector for this list */
916 i_shift_offset = DIM*shiftidx[iidx];
917
918 /* Load limits for loop over neighbors */
919 j_index_start = jindex[iidx];
920 j_index_end = jindex[iidx+1];
921
922 /* Get outer coordinate index */
923 inr = iinr[iidx];
924 i_coord_offset = DIM*inr;
925
926 /* Load i particle coords and add shift vector */
927 gmx_mm_load_shift_and_4rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
928 &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
929
930 fix0 = _mm_setzero_pd();
931 fiy0 = _mm_setzero_pd();
932 fiz0 = _mm_setzero_pd();
933 fix1 = _mm_setzero_pd();
934 fiy1 = _mm_setzero_pd();
935 fiz1 = _mm_setzero_pd();
936 fix2 = _mm_setzero_pd();
937 fiy2 = _mm_setzero_pd();
938 fiz2 = _mm_setzero_pd();
939 fix3 = _mm_setzero_pd();
940 fiy3 = _mm_setzero_pd();
941 fiz3 = _mm_setzero_pd();
942
943 /* Start inner kernel loop */
944 for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
945 {
946
947 /* Get j neighbor index, and coordinate index */
948 jnrA = jjnr[jidx];
949 jnrB = jjnr[jidx+1];
950 j_coord_offsetA = DIM*jnrA;
951 j_coord_offsetB = DIM*jnrB;
952
953 /* load j atom coordinates */
954 gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
955 &jx0,&jy0,&jz0);
956
957 /* Calculate displacement vector */
958 dx00 = _mm_sub_pd(ix0,jx0);
959 dy00 = _mm_sub_pd(iy0,jy0);
960 dz00 = _mm_sub_pd(iz0,jz0);
961 dx10 = _mm_sub_pd(ix1,jx0);
962 dy10 = _mm_sub_pd(iy1,jy0);
963 dz10 = _mm_sub_pd(iz1,jz0);
964 dx20 = _mm_sub_pd(ix2,jx0);
965 dy20 = _mm_sub_pd(iy2,jy0);
966 dz20 = _mm_sub_pd(iz2,jz0);
967 dx30 = _mm_sub_pd(ix3,jx0);
968 dy30 = _mm_sub_pd(iy3,jy0);
969 dz30 = _mm_sub_pd(iz3,jz0);
970
971 /* Calculate squared distance and things based on it */
972 rsq00 = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
973 rsq10 = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
974 rsq20 = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
975 rsq30 = gmx_mm_calc_rsq_pd(dx30,dy30,dz30);
976
977 rinv00 = gmx_mm_invsqrt_pd(rsq00);
978 rinv10 = gmx_mm_invsqrt_pd(rsq10);
979 rinv20 = gmx_mm_invsqrt_pd(rsq20);
980 rinv30 = gmx_mm_invsqrt_pd(rsq30);
981
982 rinvsq00 = _mm_mul_pd(rinv00,rinv00);
983 rinvsq10 = _mm_mul_pd(rinv10,rinv10);
984 rinvsq20 = _mm_mul_pd(rinv20,rinv20);
985 rinvsq30 = _mm_mul_pd(rinv30,rinv30);
986
987 /* Load parameters for j particles */
988 jq0 = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
989 vdwjidx0A = 2*vdwtype[jnrA+0];
990 vdwjidx0B = 2*vdwtype[jnrB+0];
991
992 fjx0 = _mm_setzero_pd();
993 fjy0 = _mm_setzero_pd();
994 fjz0 = _mm_setzero_pd();
995
996 /**************************
997 * CALCULATE INTERACTIONS *
998 **************************/
999
1000 if (gmx_mm_any_lt(rsq00,rcutoff2))
1001 {
1002
1003 r00 = _mm_mul_pd(rsq00,rinv00);
1004
1005 /* Compute parameters for interactions between i and j atoms */
1006 gmx_mm_load_2pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,
1007 vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
1008
1009 c6grid_00 = gmx_mm_load_2real_swizzle_pd(vdwgridparam+vdwioffset0+vdwjidx0A,
1010 vdwgridparam+vdwioffset0+vdwjidx0B);
1011
1012 /* Analytical LJ-PME */
1013 rinvsix = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
1014 ewcljrsq = _mm_mul_pd(ewclj2,rsq00);
1015 ewclj6 = _mm_mul_pd(ewclj2,_mm_mul_pd(ewclj2,ewclj2));
1016 exponent = gmx_simd_exp_d(ewcljrsq);
1017 /* poly = exp(-(beta*r)^2) * (1 + (beta*r)^2 + (beta*r)^4 /2) */
1018 poly = _mm_mul_pd(exponent,_mm_add_pd(_mm_sub_pd(one,ewcljrsq),_mm_mul_pd(_mm_mul_pd(ewcljrsq,ewcljrsq),one_half)));
1019 /* f6A = 6 * C6grid * (1 - poly) */
1020 f6A = _mm_mul_pd(c6grid_00,_mm_sub_pd(one,poly));
1021 /* f6B = C6grid * exponent * beta^6 */
1022 f6B = _mm_mul_pd(_mm_mul_pd(c6grid_00,one_sixth),_mm_mul_pd(exponent,ewclj6));
1023 /* fvdw = 12*C12/r13 - ((6*C6 - f6A)/r6 + f6B)/r */
1024 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);
1025
1026 cutoff_mask = _mm_cmplt_pd(rsq00,rcutoff2);
1027
1028 fscal = fvdw;
1029
1030 fscal = _mm_and_pd(fscal,cutoff_mask);
1031
1032 /* Calculate temporary vectorial force */
1033 tx = _mm_mul_pd(fscal,dx00);
1034 ty = _mm_mul_pd(fscal,dy00);
1035 tz = _mm_mul_pd(fscal,dz00);
1036
1037 /* Update vectorial force */
1038 fix0 = _mm_add_pd(fix0,tx);
1039 fiy0 = _mm_add_pd(fiy0,ty);
1040 fiz0 = _mm_add_pd(fiz0,tz);
1041
1042 fjx0 = _mm_add_pd(fjx0,tx);
1043 fjy0 = _mm_add_pd(fjy0,ty);
1044 fjz0 = _mm_add_pd(fjz0,tz);
1045
1046 }
1047
1048 /**************************
1049 * CALCULATE INTERACTIONS *
1050 **************************/
1051
1052 if (gmx_mm_any_lt(rsq10,rcutoff2))
1053 {
1054
1055 r10 = _mm_mul_pd(rsq10,rinv10);
1056
1057 /* Compute parameters for interactions between i and j atoms */
1058 qq10 = _mm_mul_pd(iq1,jq0);
1059
1060 /* EWALD ELECTROSTATICS */
1061
1062 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1063 ewrt = _mm_mul_pd(r10,ewtabscale);
1064 ewitab = _mm_cvttpd_epi32(ewrt);
1065 eweps = _mm_sub_pd(ewrt,_mm_cvtepi32_pd(ewitab));
1066 gmx_mm_load_2pair_swizzle_pd(ewtab+gmx_mm_extract_epi32(ewitab,0),ewtab+gmx_mm_extract_epi32(ewitab,1),
1067 &ewtabF,&ewtabFn);
1068 felec = _mm_add_pd(_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF),_mm_mul_pd(eweps,ewtabFn));
1069 felec = _mm_mul_pd(_mm_mul_pd(qq10,rinv10),_mm_sub_pd(rinvsq10,felec));
1070
1071 cutoff_mask = _mm_cmplt_pd(rsq10,rcutoff2);
1072
1073 fscal = felec;
1074
1075 fscal = _mm_and_pd(fscal,cutoff_mask);
1076
1077 /* Calculate temporary vectorial force */
1078 tx = _mm_mul_pd(fscal,dx10);
1079 ty = _mm_mul_pd(fscal,dy10);
1080 tz = _mm_mul_pd(fscal,dz10);
1081
1082 /* Update vectorial force */
1083 fix1 = _mm_add_pd(fix1,tx);
1084 fiy1 = _mm_add_pd(fiy1,ty);
1085 fiz1 = _mm_add_pd(fiz1,tz);
1086
1087 fjx0 = _mm_add_pd(fjx0,tx);
1088 fjy0 = _mm_add_pd(fjy0,ty);
1089 fjz0 = _mm_add_pd(fjz0,tz);
1090
1091 }
1092
1093 /**************************
1094 * CALCULATE INTERACTIONS *
1095 **************************/
1096
1097 if (gmx_mm_any_lt(rsq20,rcutoff2))
1098 {
1099
1100 r20 = _mm_mul_pd(rsq20,rinv20);
1101
1102 /* Compute parameters for interactions between i and j atoms */
1103 qq20 = _mm_mul_pd(iq2,jq0);
1104
1105 /* EWALD ELECTROSTATICS */
1106
1107 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1108 ewrt = _mm_mul_pd(r20,ewtabscale);
1109 ewitab = _mm_cvttpd_epi32(ewrt);
1110 eweps = _mm_sub_pd(ewrt,_mm_cvtepi32_pd(ewitab));
1111 gmx_mm_load_2pair_swizzle_pd(ewtab+gmx_mm_extract_epi32(ewitab,0),ewtab+gmx_mm_extract_epi32(ewitab,1),
1112 &ewtabF,&ewtabFn);
1113 felec = _mm_add_pd(_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF),_mm_mul_pd(eweps,ewtabFn));
1114 felec = _mm_mul_pd(_mm_mul_pd(qq20,rinv20),_mm_sub_pd(rinvsq20,felec));
1115
1116 cutoff_mask = _mm_cmplt_pd(rsq20,rcutoff2);
1117
1118 fscal = felec;
1119
1120 fscal = _mm_and_pd(fscal,cutoff_mask);
1121
1122 /* Calculate temporary vectorial force */
1123 tx = _mm_mul_pd(fscal,dx20);
1124 ty = _mm_mul_pd(fscal,dy20);
1125 tz = _mm_mul_pd(fscal,dz20);
1126
1127 /* Update vectorial force */
1128 fix2 = _mm_add_pd(fix2,tx);
1129 fiy2 = _mm_add_pd(fiy2,ty);
1130 fiz2 = _mm_add_pd(fiz2,tz);
1131
1132 fjx0 = _mm_add_pd(fjx0,tx);
1133 fjy0 = _mm_add_pd(fjy0,ty);
1134 fjz0 = _mm_add_pd(fjz0,tz);
1135
1136 }
1137
1138 /**************************
1139 * CALCULATE INTERACTIONS *
1140 **************************/
1141
1142 if (gmx_mm_any_lt(rsq30,rcutoff2))
1143 {
1144
1145 r30 = _mm_mul_pd(rsq30,rinv30);
1146
1147 /* Compute parameters for interactions between i and j atoms */
1148 qq30 = _mm_mul_pd(iq3,jq0);
1149
1150 /* EWALD ELECTROSTATICS */
1151
1152 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1153 ewrt = _mm_mul_pd(r30,ewtabscale);
1154 ewitab = _mm_cvttpd_epi32(ewrt);
1155 eweps = _mm_sub_pd(ewrt,_mm_cvtepi32_pd(ewitab));
1156 gmx_mm_load_2pair_swizzle_pd(ewtab+gmx_mm_extract_epi32(ewitab,0),ewtab+gmx_mm_extract_epi32(ewitab,1),
1157 &ewtabF,&ewtabFn);
1158 felec = _mm_add_pd(_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF),_mm_mul_pd(eweps,ewtabFn));
1159 felec = _mm_mul_pd(_mm_mul_pd(qq30,rinv30),_mm_sub_pd(rinvsq30,felec));
1160
1161 cutoff_mask = _mm_cmplt_pd(rsq30,rcutoff2);
1162
1163 fscal = felec;
1164
1165 fscal = _mm_and_pd(fscal,cutoff_mask);
1166
1167 /* Calculate temporary vectorial force */
1168 tx = _mm_mul_pd(fscal,dx30);
1169 ty = _mm_mul_pd(fscal,dy30);
1170 tz = _mm_mul_pd(fscal,dz30);
1171
1172 /* Update vectorial force */
1173 fix3 = _mm_add_pd(fix3,tx);
1174 fiy3 = _mm_add_pd(fiy3,ty);
1175 fiz3 = _mm_add_pd(fiz3,tz);
1176
1177 fjx0 = _mm_add_pd(fjx0,tx);
1178 fjy0 = _mm_add_pd(fjy0,ty);
1179 fjz0 = _mm_add_pd(fjz0,tz);
1180
1181 }
1182
1183 gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,fjx0,fjy0,fjz0);
1184
1185 /* Inner loop uses 169 flops */
1186 }
1187
1188 if(jidx<j_index_end)
1189 {
1190
1191 jnrA = jjnr[jidx];
1192 j_coord_offsetA = DIM*jnrA;
1193
1194 /* load j atom coordinates */
1195 gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
1196 &jx0,&jy0,&jz0);
1197
1198 /* Calculate displacement vector */
1199 dx00 = _mm_sub_pd(ix0,jx0);
1200 dy00 = _mm_sub_pd(iy0,jy0);
1201 dz00 = _mm_sub_pd(iz0,jz0);
1202 dx10 = _mm_sub_pd(ix1,jx0);
1203 dy10 = _mm_sub_pd(iy1,jy0);
1204 dz10 = _mm_sub_pd(iz1,jz0);
1205 dx20 = _mm_sub_pd(ix2,jx0);
1206 dy20 = _mm_sub_pd(iy2,jy0);
1207 dz20 = _mm_sub_pd(iz2,jz0);
1208 dx30 = _mm_sub_pd(ix3,jx0);
1209 dy30 = _mm_sub_pd(iy3,jy0);
1210 dz30 = _mm_sub_pd(iz3,jz0);
1211
1212 /* Calculate squared distance and things based on it */
1213 rsq00 = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
1214 rsq10 = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
1215 rsq20 = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
1216 rsq30 = gmx_mm_calc_rsq_pd(dx30,dy30,dz30);
1217
1218 rinv00 = gmx_mm_invsqrt_pd(rsq00);
1219 rinv10 = gmx_mm_invsqrt_pd(rsq10);
1220 rinv20 = gmx_mm_invsqrt_pd(rsq20);
1221 rinv30 = gmx_mm_invsqrt_pd(rsq30);
1222
1223 rinvsq00 = _mm_mul_pd(rinv00,rinv00);
1224 rinvsq10 = _mm_mul_pd(rinv10,rinv10);
1225 rinvsq20 = _mm_mul_pd(rinv20,rinv20);
1226 rinvsq30 = _mm_mul_pd(rinv30,rinv30);
1227
1228 /* Load parameters for j particles */
1229 jq0 = _mm_load_sd(charge+jnrA+0);
1230 vdwjidx0A = 2*vdwtype[jnrA+0];
1231
1232 fjx0 = _mm_setzero_pd();
1233 fjy0 = _mm_setzero_pd();
1234 fjz0 = _mm_setzero_pd();
1235
1236 /**************************
1237 * CALCULATE INTERACTIONS *
1238 **************************/
1239
1240 if (gmx_mm_any_lt(rsq00,rcutoff2))
1241 {
1242
1243 r00 = _mm_mul_pd(rsq00,rinv00);
1244
1245 /* Compute parameters for interactions between i and j atoms */
1246 gmx_mm_load_1pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,&c6_00,&c12_00);
1247
1248 c6grid_00 = gmx_mm_load_1real_pd(vdwgridparam+vdwioffset0+vdwjidx0A);
1249
1250 /* Analytical LJ-PME */
1251 rinvsix = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
1252 ewcljrsq = _mm_mul_pd(ewclj2,rsq00);
1253 ewclj6 = _mm_mul_pd(ewclj2,_mm_mul_pd(ewclj2,ewclj2));
1254 exponent = gmx_simd_exp_d(ewcljrsq);
1255 /* poly = exp(-(beta*r)^2) * (1 + (beta*r)^2 + (beta*r)^4 /2) */
1256 poly = _mm_mul_pd(exponent,_mm_add_pd(_mm_sub_pd(one,ewcljrsq),_mm_mul_pd(_mm_mul_pd(ewcljrsq,ewcljrsq),one_half)));
1257 /* f6A = 6 * C6grid * (1 - poly) */
1258 f6A = _mm_mul_pd(c6grid_00,_mm_sub_pd(one,poly));
1259 /* f6B = C6grid * exponent * beta^6 */
1260 f6B = _mm_mul_pd(_mm_mul_pd(c6grid_00,one_sixth),_mm_mul_pd(exponent,ewclj6));
1261 /* fvdw = 12*C12/r13 - ((6*C6 - f6A)/r6 + f6B)/r */
1262 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);
1263
1264 cutoff_mask = _mm_cmplt_pd(rsq00,rcutoff2);
1265
1266 fscal = fvdw;
1267
1268 fscal = _mm_and_pd(fscal,cutoff_mask);
1269
1270 fscal = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1271
1272 /* Calculate temporary vectorial force */
1273 tx = _mm_mul_pd(fscal,dx00);
1274 ty = _mm_mul_pd(fscal,dy00);
1275 tz = _mm_mul_pd(fscal,dz00);
1276
1277 /* Update vectorial force */
1278 fix0 = _mm_add_pd(fix0,tx);
1279 fiy0 = _mm_add_pd(fiy0,ty);
1280 fiz0 = _mm_add_pd(fiz0,tz);
1281
1282 fjx0 = _mm_add_pd(fjx0,tx);
1283 fjy0 = _mm_add_pd(fjy0,ty);
1284 fjz0 = _mm_add_pd(fjz0,tz);
1285
1286 }
1287
1288 /**************************
1289 * CALCULATE INTERACTIONS *
1290 **************************/
1291
1292 if (gmx_mm_any_lt(rsq10,rcutoff2))
1293 {
1294
1295 r10 = _mm_mul_pd(rsq10,rinv10);
1296
1297 /* Compute parameters for interactions between i and j atoms */
1298 qq10 = _mm_mul_pd(iq1,jq0);
1299
1300 /* EWALD ELECTROSTATICS */
1301
1302 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1303 ewrt = _mm_mul_pd(r10,ewtabscale);
1304 ewitab = _mm_cvttpd_epi32(ewrt);
1305 eweps = _mm_sub_pd(ewrt,_mm_cvtepi32_pd(ewitab));
1306 gmx_mm_load_1pair_swizzle_pd(ewtab+gmx_mm_extract_epi32(ewitab,0),&ewtabF,&ewtabFn);
1307 felec = _mm_add_pd(_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF),_mm_mul_pd(eweps,ewtabFn));
1308 felec = _mm_mul_pd(_mm_mul_pd(qq10,rinv10),_mm_sub_pd(rinvsq10,felec));
1309
1310 cutoff_mask = _mm_cmplt_pd(rsq10,rcutoff2);
1311
1312 fscal = felec;
1313
1314 fscal = _mm_and_pd(fscal,cutoff_mask);
1315
1316 fscal = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1317
1318 /* Calculate temporary vectorial force */
1319 tx = _mm_mul_pd(fscal,dx10);
1320 ty = _mm_mul_pd(fscal,dy10);
1321 tz = _mm_mul_pd(fscal,dz10);
1322
1323 /* Update vectorial force */
1324 fix1 = _mm_add_pd(fix1,tx);
1325 fiy1 = _mm_add_pd(fiy1,ty);
1326 fiz1 = _mm_add_pd(fiz1,tz);
1327
1328 fjx0 = _mm_add_pd(fjx0,tx);
1329 fjy0 = _mm_add_pd(fjy0,ty);
1330 fjz0 = _mm_add_pd(fjz0,tz);
1331
1332 }
1333
1334 /**************************
1335 * CALCULATE INTERACTIONS *
1336 **************************/
1337
1338 if (gmx_mm_any_lt(rsq20,rcutoff2))
1339 {
1340
1341 r20 = _mm_mul_pd(rsq20,rinv20);
1342
1343 /* Compute parameters for interactions between i and j atoms */
1344 qq20 = _mm_mul_pd(iq2,jq0);
1345
1346 /* EWALD ELECTROSTATICS */
1347
1348 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1349 ewrt = _mm_mul_pd(r20,ewtabscale);
1350 ewitab = _mm_cvttpd_epi32(ewrt);
1351 eweps = _mm_sub_pd(ewrt,_mm_cvtepi32_pd(ewitab));
1352 gmx_mm_load_1pair_swizzle_pd(ewtab+gmx_mm_extract_epi32(ewitab,0),&ewtabF,&ewtabFn);
1353 felec = _mm_add_pd(_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF),_mm_mul_pd(eweps,ewtabFn));
1354 felec = _mm_mul_pd(_mm_mul_pd(qq20,rinv20),_mm_sub_pd(rinvsq20,felec));
1355
1356 cutoff_mask = _mm_cmplt_pd(rsq20,rcutoff2);
1357
1358 fscal = felec;
1359
1360 fscal = _mm_and_pd(fscal,cutoff_mask);
1361
1362 fscal = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1363
1364 /* Calculate temporary vectorial force */
1365 tx = _mm_mul_pd(fscal,dx20);
1366 ty = _mm_mul_pd(fscal,dy20);
1367 tz = _mm_mul_pd(fscal,dz20);
1368
1369 /* Update vectorial force */
1370 fix2 = _mm_add_pd(fix2,tx);
1371 fiy2 = _mm_add_pd(fiy2,ty);
1372 fiz2 = _mm_add_pd(fiz2,tz);
1373
1374 fjx0 = _mm_add_pd(fjx0,tx);
1375 fjy0 = _mm_add_pd(fjy0,ty);
1376 fjz0 = _mm_add_pd(fjz0,tz);
1377
1378 }
1379
1380 /**************************
1381 * CALCULATE INTERACTIONS *
1382 **************************/
1383
1384 if (gmx_mm_any_lt(rsq30,rcutoff2))
1385 {
1386
1387 r30 = _mm_mul_pd(rsq30,rinv30);
1388
1389 /* Compute parameters for interactions between i and j atoms */
1390 qq30 = _mm_mul_pd(iq3,jq0);
1391
1392 /* EWALD ELECTROSTATICS */
1393
1394 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1395 ewrt = _mm_mul_pd(r30,ewtabscale);
1396 ewitab = _mm_cvttpd_epi32(ewrt);
1397 eweps = _mm_sub_pd(ewrt,_mm_cvtepi32_pd(ewitab));
1398 gmx_mm_load_1pair_swizzle_pd(ewtab+gmx_mm_extract_epi32(ewitab,0),&ewtabF,&ewtabFn);
1399 felec = _mm_add_pd(_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF),_mm_mul_pd(eweps,ewtabFn));
1400 felec = _mm_mul_pd(_mm_mul_pd(qq30,rinv30),_mm_sub_pd(rinvsq30,felec));
1401
1402 cutoff_mask = _mm_cmplt_pd(rsq30,rcutoff2);
1403
1404 fscal = felec;
1405
1406 fscal = _mm_and_pd(fscal,cutoff_mask);
1407
1408 fscal = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1409
1410 /* Calculate temporary vectorial force */
1411 tx = _mm_mul_pd(fscal,dx30);
1412 ty = _mm_mul_pd(fscal,dy30);
1413 tz = _mm_mul_pd(fscal,dz30);
1414
1415 /* Update vectorial force */
1416 fix3 = _mm_add_pd(fix3,tx);
1417 fiy3 = _mm_add_pd(fiy3,ty);
1418 fiz3 = _mm_add_pd(fiz3,tz);
1419
1420 fjx0 = _mm_add_pd(fjx0,tx);
1421 fjy0 = _mm_add_pd(fjy0,ty);
1422 fjz0 = _mm_add_pd(fjz0,tz);
1423
1424 }
1425
1426 gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,fjx0,fjy0,fjz0);
1427
1428 /* Inner loop uses 169 flops */
1429 }
1430
1431 /* End of innermost loop */
1432
1433 gmx_mm_update_iforce_4atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
1434 f+i_coord_offset,fshift+i_shift_offset);
1435
1436 /* Increment number of inner iterations */
1437 inneriter += j_index_end - j_index_start;
1438
1439 /* Outer loop uses 24 flops */
1440 }
1441
1442 /* Increment number of outer iterations */
1443 outeriter += nri;
1444
1445 /* Update outer/inner flops */
1446
1447 inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W4_F,outeriter*24 + inneriter*169);
1448 }
The ultimate molecular dynamics simulation package