?  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_ElecEw_VdwLJEw_GeomW3P1_sse2_double.c
blob73d23388cf48492d5b4ea1c2966b31c3704b7abc
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_ElecEw_VdwLJEw_GeomW3P1_VF_sse2_double
53 * Electrostatics interaction: Ewald
54 * VdW interaction: LJEwald
55 * Geometry: Water3-Particle
56 * Calculate force/pot: PotentialAndForce
57 */
58 void
59 nb_kernel_ElecEw_VdwLJEw_GeomW3P1_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 vdwjidx0A,vdwjidx0B;
88 __m128d jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
89 __m128d dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
90 __m128d dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
91 __m128d dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
92 __m128d velec,felec,velecsum,facel,crf,krf,krf2;
93 real *charge;
94 int nvdwtype;
95 __m128d rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
96 int *vdwtype;
97 real *vdwparam;
98 __m128d one_sixth = _mm_set1_pd(1.0/6.0);
99 __m128d one_twelfth = _mm_set1_pd(1.0/12.0);
100 __m128d c6grid_00;
101 __m128d c6grid_10;
102 __m128d c6grid_20;
103 __m128d ewclj,ewclj2,ewclj6,ewcljrsq,poly,exponent,f6A,f6B,sh_lj_ewald;
104 real *vdwgridparam;
105 __m128d one_half = _mm_set1_pd(0.5);
106 __m128d minus_one = _mm_set1_pd(-1.0);
107 __m128i ewitab;
108 __m128d ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
109 real *ewtab;
110 __m128d dummy_mask,cutoff_mask;
111 __m128d signbit = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
112 __m128d one = _mm_set1_pd(1.0);
113 __m128d two = _mm_set1_pd(2.0);
114 x = xx[0];
115 f = ff[0];
116
117 nri = nlist->nri;
118 iinr = nlist->iinr;
119 jindex = nlist->jindex;
120 jjnr = nlist->jjnr;
121 shiftidx = nlist->shift;
122 gid = nlist->gid;
123 shiftvec = fr->shift_vec[0];
124 fshift = fr->fshift[0];
125 facel = _mm_set1_pd(fr->epsfac);
126 charge = mdatoms->chargeA;
127 nvdwtype = fr->ntype;
128 vdwparam = fr->nbfp;
129 vdwtype = mdatoms->typeA;
130 vdwgridparam = fr->ljpme_c6grid;
131 sh_lj_ewald = _mm_set1_pd(fr->ic->sh_lj_ewald);
132 ewclj = _mm_set1_pd(fr->ewaldcoeff_lj);
133 ewclj2 = _mm_mul_pd(minus_one,_mm_mul_pd(ewclj,ewclj));
134
135 sh_ewald = _mm_set1_pd(fr->ic->sh_ewald);
136 ewtab = fr->ic->tabq_coul_FDV0;
137 ewtabscale = _mm_set1_pd(fr->ic->tabq_scale);
138 ewtabhalfspace = _mm_set1_pd(0.5/fr->ic->tabq_scale);
139
140 /* Setup water-specific parameters */
141 inr = nlist->iinr[0];
142 iq0 = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+0]));
143 iq1 = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+1]));
144 iq2 = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+2]));
145 vdwioffset0 = 2*nvdwtype*vdwtype[inr+0];
146
147 /* Avoid stupid compiler warnings */
148 jnrA = jnrB = 0;
149 j_coord_offsetA = 0;
150 j_coord_offsetB = 0;
151
152 outeriter = 0;
153 inneriter = 0;
154
155 /* Start outer loop over neighborlists */
156 for(iidx=0; iidx<nri; iidx++)
157 {
158 /* Load shift vector for this list */
159 i_shift_offset = DIM*shiftidx[iidx];
160
161 /* Load limits for loop over neighbors */
162 j_index_start = jindex[iidx];
163 j_index_end = jindex[iidx+1];
164
165 /* Get outer coordinate index */
166 inr = iinr[iidx];
167 i_coord_offset = DIM*inr;
168
169 /* Load i particle coords and add shift vector */
170 gmx_mm_load_shift_and_3rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
171 &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
172
173 fix0 = _mm_setzero_pd();
174 fiy0 = _mm_setzero_pd();
175 fiz0 = _mm_setzero_pd();
176 fix1 = _mm_setzero_pd();
177 fiy1 = _mm_setzero_pd();
178 fiz1 = _mm_setzero_pd();
179 fix2 = _mm_setzero_pd();
180 fiy2 = _mm_setzero_pd();
181 fiz2 = _mm_setzero_pd();
182
183 /* Reset potential sums */
184 velecsum = _mm_setzero_pd();
185 vvdwsum = _mm_setzero_pd();
186
187 /* Start inner kernel loop */
188 for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
189 {
190
191 /* Get j neighbor index, and coordinate index */
192 jnrA = jjnr[jidx];
193 jnrB = jjnr[jidx+1];
194 j_coord_offsetA = DIM*jnrA;
195 j_coord_offsetB = DIM*jnrB;
196
197 /* load j atom coordinates */
198 gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
199 &jx0,&jy0,&jz0);
200
201 /* Calculate displacement vector */
202 dx00 = _mm_sub_pd(ix0,jx0);
203 dy00 = _mm_sub_pd(iy0,jy0);
204 dz00 = _mm_sub_pd(iz0,jz0);
205 dx10 = _mm_sub_pd(ix1,jx0);
206 dy10 = _mm_sub_pd(iy1,jy0);
207 dz10 = _mm_sub_pd(iz1,jz0);
208 dx20 = _mm_sub_pd(ix2,jx0);
209 dy20 = _mm_sub_pd(iy2,jy0);
210 dz20 = _mm_sub_pd(iz2,jz0);
211
212 /* Calculate squared distance and things based on it */
213 rsq00 = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
214 rsq10 = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
215 rsq20 = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
216
217 rinv00 = gmx_mm_invsqrt_pd(rsq00);
218 rinv10 = gmx_mm_invsqrt_pd(rsq10);
219 rinv20 = gmx_mm_invsqrt_pd(rsq20);
220
221 rinvsq00 = _mm_mul_pd(rinv00,rinv00);
222 rinvsq10 = _mm_mul_pd(rinv10,rinv10);
223 rinvsq20 = _mm_mul_pd(rinv20,rinv20);
224
225 /* Load parameters for j particles */
226 jq0 = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
227 vdwjidx0A = 2*vdwtype[jnrA+0];
228 vdwjidx0B = 2*vdwtype[jnrB+0];
229
230 fjx0 = _mm_setzero_pd();
231 fjy0 = _mm_setzero_pd();
232 fjz0 = _mm_setzero_pd();
233
234 /**************************
235 * CALCULATE INTERACTIONS *
236 **************************/
237
238 r00 = _mm_mul_pd(rsq00,rinv00);
239
240 /* Compute parameters for interactions between i and j atoms */
241 qq00 = _mm_mul_pd(iq0,jq0);
242 gmx_mm_load_2pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,
243 vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
244
245 c6grid_00 = gmx_mm_load_2real_swizzle_pd(vdwgridparam+vdwioffset0+vdwjidx0A,
246 vdwgridparam+vdwioffset0+vdwjidx0B);
247
248 /* EWALD ELECTROSTATICS */
249
250 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
251 ewrt = _mm_mul_pd(r00,ewtabscale);
252 ewitab = _mm_cvttpd_epi32(ewrt);
253 eweps = _mm_sub_pd(ewrt,_mm_cvtepi32_pd(ewitab));
254 ewitab = _mm_slli_epi32(ewitab,2);
255 ewtabF = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,0) );
256 ewtabD = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,1) );
257 GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
258 ewtabV = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,0) +2);
259 ewtabFn = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,1) +2);
260 GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
261 felec = _mm_add_pd(ewtabF,_mm_mul_pd(eweps,ewtabD));
262 velec = _mm_sub_pd(ewtabV,_mm_mul_pd(_mm_mul_pd(ewtabhalfspace,eweps),_mm_add_pd(ewtabF,felec)));
263 velec = _mm_mul_pd(qq00,_mm_sub_pd(rinv00,velec));
264 felec = _mm_mul_pd(_mm_mul_pd(qq00,rinv00),_mm_sub_pd(rinvsq00,felec));
265
266 /* Analytical LJ-PME */
267 rinvsix = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
268 ewcljrsq = _mm_mul_pd(ewclj2,rsq00);
269 ewclj6 = _mm_mul_pd(ewclj2,_mm_mul_pd(ewclj2,ewclj2));
270 exponent = gmx_simd_exp_d(ewcljrsq);
271 /* poly = exp(-(beta*r)^2) * (1 + (beta*r)^2 + (beta*r)^4 /2) */
272 poly = _mm_mul_pd(exponent,_mm_add_pd(_mm_sub_pd(one,ewcljrsq),_mm_mul_pd(_mm_mul_pd(ewcljrsq,ewcljrsq),one_half)));
273 /* vvdw6 = [C6 - C6grid * (1-poly)]/r6 */
274 vvdw6 = _mm_mul_pd(_mm_sub_pd(c6_00,_mm_mul_pd(c6grid_00,_mm_sub_pd(one,poly))),rinvsix);
275 vvdw12 = _mm_mul_pd(c12_00,_mm_mul_pd(rinvsix,rinvsix));
276 vvdw = _mm_sub_pd(_mm_mul_pd(vvdw12,one_twelfth),_mm_mul_pd(vvdw6,one_sixth));
277 /* fvdw = vvdw12/r - (vvdw6/r + (C6grid * exponent * beta^6)/r) */
278 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);
279
280 /* Update potential sum for this i atom from the interaction with this j atom. */
281 velecsum = _mm_add_pd(velecsum,velec);
282 vvdwsum = _mm_add_pd(vvdwsum,vvdw);
283
284 fscal = _mm_add_pd(felec,fvdw);
285
286 /* Calculate temporary vectorial force */
287 tx = _mm_mul_pd(fscal,dx00);
288 ty = _mm_mul_pd(fscal,dy00);
289 tz = _mm_mul_pd(fscal,dz00);
290
291 /* Update vectorial force */
292 fix0 = _mm_add_pd(fix0,tx);
293 fiy0 = _mm_add_pd(fiy0,ty);
294 fiz0 = _mm_add_pd(fiz0,tz);
295
296 fjx0 = _mm_add_pd(fjx0,tx);
297 fjy0 = _mm_add_pd(fjy0,ty);
298 fjz0 = _mm_add_pd(fjz0,tz);
299
300 /**************************
301 * CALCULATE INTERACTIONS *
302 **************************/
303
304 r10 = _mm_mul_pd(rsq10,rinv10);
305
306 /* Compute parameters for interactions between i and j atoms */
307 qq10 = _mm_mul_pd(iq1,jq0);
308
309 /* EWALD ELECTROSTATICS */
310
311 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
312 ewrt = _mm_mul_pd(r10,ewtabscale);
313 ewitab = _mm_cvttpd_epi32(ewrt);
314 eweps = _mm_sub_pd(ewrt,_mm_cvtepi32_pd(ewitab));
315 ewitab = _mm_slli_epi32(ewitab,2);
316 ewtabF = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,0) );
317 ewtabD = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,1) );
318 GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
319 ewtabV = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,0) +2);
320 ewtabFn = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,1) +2);
321 GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
322 felec = _mm_add_pd(ewtabF,_mm_mul_pd(eweps,ewtabD));
323 velec = _mm_sub_pd(ewtabV,_mm_mul_pd(_mm_mul_pd(ewtabhalfspace,eweps),_mm_add_pd(ewtabF,felec)));
324 velec = _mm_mul_pd(qq10,_mm_sub_pd(rinv10,velec));
325 felec = _mm_mul_pd(_mm_mul_pd(qq10,rinv10),_mm_sub_pd(rinvsq10,felec));
326
327 /* Update potential sum for this i atom from the interaction with this j atom. */
328 velecsum = _mm_add_pd(velecsum,velec);
329
330 fscal = felec;
331
332 /* Calculate temporary vectorial force */
333 tx = _mm_mul_pd(fscal,dx10);
334 ty = _mm_mul_pd(fscal,dy10);
335 tz = _mm_mul_pd(fscal,dz10);
336
337 /* Update vectorial force */
338 fix1 = _mm_add_pd(fix1,tx);
339 fiy1 = _mm_add_pd(fiy1,ty);
340 fiz1 = _mm_add_pd(fiz1,tz);
341
342 fjx0 = _mm_add_pd(fjx0,tx);
343 fjy0 = _mm_add_pd(fjy0,ty);
344 fjz0 = _mm_add_pd(fjz0,tz);
345
346 /**************************
347 * CALCULATE INTERACTIONS *
348 **************************/
349
350 r20 = _mm_mul_pd(rsq20,rinv20);
351
352 /* Compute parameters for interactions between i and j atoms */
353 qq20 = _mm_mul_pd(iq2,jq0);
354
355 /* EWALD ELECTROSTATICS */
356
357 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
358 ewrt = _mm_mul_pd(r20,ewtabscale);
359 ewitab = _mm_cvttpd_epi32(ewrt);
360 eweps = _mm_sub_pd(ewrt,_mm_cvtepi32_pd(ewitab));
361 ewitab = _mm_slli_epi32(ewitab,2);
362 ewtabF = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,0) );
363 ewtabD = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,1) );
364 GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
365 ewtabV = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,0) +2);
366 ewtabFn = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,1) +2);
367 GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
368 felec = _mm_add_pd(ewtabF,_mm_mul_pd(eweps,ewtabD));
369 velec = _mm_sub_pd(ewtabV,_mm_mul_pd(_mm_mul_pd(ewtabhalfspace,eweps),_mm_add_pd(ewtabF,felec)));
370 velec = _mm_mul_pd(qq20,_mm_sub_pd(rinv20,velec));
371 felec = _mm_mul_pd(_mm_mul_pd(qq20,rinv20),_mm_sub_pd(rinvsq20,felec));
372
373 /* Update potential sum for this i atom from the interaction with this j atom. */
374 velecsum = _mm_add_pd(velecsum,velec);
375
376 fscal = felec;
377
378 /* Calculate temporary vectorial force */
379 tx = _mm_mul_pd(fscal,dx20);
380 ty = _mm_mul_pd(fscal,dy20);
381 tz = _mm_mul_pd(fscal,dz20);
382
383 /* Update vectorial force */
384 fix2 = _mm_add_pd(fix2,tx);
385 fiy2 = _mm_add_pd(fiy2,ty);
386 fiz2 = _mm_add_pd(fiz2,tz);
387
388 fjx0 = _mm_add_pd(fjx0,tx);
389 fjy0 = _mm_add_pd(fjy0,ty);
390 fjz0 = _mm_add_pd(fjz0,tz);
391
392 gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,fjx0,fjy0,fjz0);
393
394 /* Inner loop uses 154 flops */
395 }
396
397 if(jidx<j_index_end)
398 {
399
400 jnrA = jjnr[jidx];
401 j_coord_offsetA = DIM*jnrA;
402
403 /* load j atom coordinates */
404 gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
405 &jx0,&jy0,&jz0);
406
407 /* Calculate displacement vector */
408 dx00 = _mm_sub_pd(ix0,jx0);
409 dy00 = _mm_sub_pd(iy0,jy0);
410 dz00 = _mm_sub_pd(iz0,jz0);
411 dx10 = _mm_sub_pd(ix1,jx0);
412 dy10 = _mm_sub_pd(iy1,jy0);
413 dz10 = _mm_sub_pd(iz1,jz0);
414 dx20 = _mm_sub_pd(ix2,jx0);
415 dy20 = _mm_sub_pd(iy2,jy0);
416 dz20 = _mm_sub_pd(iz2,jz0);
417
418 /* Calculate squared distance and things based on it */
419 rsq00 = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
420 rsq10 = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
421 rsq20 = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
422
423 rinv00 = gmx_mm_invsqrt_pd(rsq00);
424 rinv10 = gmx_mm_invsqrt_pd(rsq10);
425 rinv20 = gmx_mm_invsqrt_pd(rsq20);
426
427 rinvsq00 = _mm_mul_pd(rinv00,rinv00);
428 rinvsq10 = _mm_mul_pd(rinv10,rinv10);
429 rinvsq20 = _mm_mul_pd(rinv20,rinv20);
430
431 /* Load parameters for j particles */
432 jq0 = _mm_load_sd(charge+jnrA+0);
433 vdwjidx0A = 2*vdwtype[jnrA+0];
434
435 fjx0 = _mm_setzero_pd();
436 fjy0 = _mm_setzero_pd();
437 fjz0 = _mm_setzero_pd();
438
439 /**************************
440 * CALCULATE INTERACTIONS *
441 **************************/
442
443 r00 = _mm_mul_pd(rsq00,rinv00);
444
445 /* Compute parameters for interactions between i and j atoms */
446 qq00 = _mm_mul_pd(iq0,jq0);
447 gmx_mm_load_1pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,&c6_00,&c12_00);
448
449 c6grid_00 = gmx_mm_load_1real_pd(vdwgridparam+vdwioffset0+vdwjidx0A);
450
451 /* EWALD ELECTROSTATICS */
452
453 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
454 ewrt = _mm_mul_pd(r00,ewtabscale);
455 ewitab = _mm_cvttpd_epi32(ewrt);
456 eweps = _mm_sub_pd(ewrt,_mm_cvtepi32_pd(ewitab));
457 ewitab = _mm_slli_epi32(ewitab,2);
458 ewtabF = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,0) );
459 ewtabD = _mm_setzero_pd();
460 GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
461 ewtabV = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,0) +2);
462 ewtabFn = _mm_setzero_pd();
463 GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
464 felec = _mm_add_pd(ewtabF,_mm_mul_pd(eweps,ewtabD));
465 velec = _mm_sub_pd(ewtabV,_mm_mul_pd(_mm_mul_pd(ewtabhalfspace,eweps),_mm_add_pd(ewtabF,felec)));
466 velec = _mm_mul_pd(qq00,_mm_sub_pd(rinv00,velec));
467 felec = _mm_mul_pd(_mm_mul_pd(qq00,rinv00),_mm_sub_pd(rinvsq00,felec));
468
469 /* Analytical LJ-PME */
470 rinvsix = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
471 ewcljrsq = _mm_mul_pd(ewclj2,rsq00);
472 ewclj6 = _mm_mul_pd(ewclj2,_mm_mul_pd(ewclj2,ewclj2));
473 exponent = gmx_simd_exp_d(ewcljrsq);
474 /* poly = exp(-(beta*r)^2) * (1 + (beta*r)^2 + (beta*r)^4 /2) */
475 poly = _mm_mul_pd(exponent,_mm_add_pd(_mm_sub_pd(one,ewcljrsq),_mm_mul_pd(_mm_mul_pd(ewcljrsq,ewcljrsq),one_half)));
476 /* vvdw6 = [C6 - C6grid * (1-poly)]/r6 */
477 vvdw6 = _mm_mul_pd(_mm_sub_pd(c6_00,_mm_mul_pd(c6grid_00,_mm_sub_pd(one,poly))),rinvsix);
478 vvdw12 = _mm_mul_pd(c12_00,_mm_mul_pd(rinvsix,rinvsix));
479 vvdw = _mm_sub_pd(_mm_mul_pd(vvdw12,one_twelfth),_mm_mul_pd(vvdw6,one_sixth));
480 /* fvdw = vvdw12/r - (vvdw6/r + (C6grid * exponent * beta^6)/r) */
481 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);
482
483 /* Update potential sum for this i atom from the interaction with this j atom. */
484 velec = _mm_unpacklo_pd(velec,_mm_setzero_pd());
485 velecsum = _mm_add_pd(velecsum,velec);
486 vvdw = _mm_unpacklo_pd(vvdw,_mm_setzero_pd());
487 vvdwsum = _mm_add_pd(vvdwsum,vvdw);
488
489 fscal = _mm_add_pd(felec,fvdw);
490
491 fscal = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
492
493 /* Calculate temporary vectorial force */
494 tx = _mm_mul_pd(fscal,dx00);
495 ty = _mm_mul_pd(fscal,dy00);
496 tz = _mm_mul_pd(fscal,dz00);
497
498 /* Update vectorial force */
499 fix0 = _mm_add_pd(fix0,tx);
500 fiy0 = _mm_add_pd(fiy0,ty);
501 fiz0 = _mm_add_pd(fiz0,tz);
502
503 fjx0 = _mm_add_pd(fjx0,tx);
504 fjy0 = _mm_add_pd(fjy0,ty);
505 fjz0 = _mm_add_pd(fjz0,tz);
506
507 /**************************
508 * CALCULATE INTERACTIONS *
509 **************************/
510
511 r10 = _mm_mul_pd(rsq10,rinv10);
512
513 /* Compute parameters for interactions between i and j atoms */
514 qq10 = _mm_mul_pd(iq1,jq0);
515
516 /* EWALD ELECTROSTATICS */
517
518 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
519 ewrt = _mm_mul_pd(r10,ewtabscale);
520 ewitab = _mm_cvttpd_epi32(ewrt);
521 eweps = _mm_sub_pd(ewrt,_mm_cvtepi32_pd(ewitab));
522 ewitab = _mm_slli_epi32(ewitab,2);
523 ewtabF = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,0) );
524 ewtabD = _mm_setzero_pd();
525 GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
526 ewtabV = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,0) +2);
527 ewtabFn = _mm_setzero_pd();
528 GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
529 felec = _mm_add_pd(ewtabF,_mm_mul_pd(eweps,ewtabD));
530 velec = _mm_sub_pd(ewtabV,_mm_mul_pd(_mm_mul_pd(ewtabhalfspace,eweps),_mm_add_pd(ewtabF,felec)));
531 velec = _mm_mul_pd(qq10,_mm_sub_pd(rinv10,velec));
532 felec = _mm_mul_pd(_mm_mul_pd(qq10,rinv10),_mm_sub_pd(rinvsq10,felec));
533
534 /* Update potential sum for this i atom from the interaction with this j atom. */
535 velec = _mm_unpacklo_pd(velec,_mm_setzero_pd());
536 velecsum = _mm_add_pd(velecsum,velec);
537
538 fscal = felec;
539
540 fscal = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
541
542 /* Calculate temporary vectorial force */
543 tx = _mm_mul_pd(fscal,dx10);
544 ty = _mm_mul_pd(fscal,dy10);
545 tz = _mm_mul_pd(fscal,dz10);
546
547 /* Update vectorial force */
548 fix1 = _mm_add_pd(fix1,tx);
549 fiy1 = _mm_add_pd(fiy1,ty);
550 fiz1 = _mm_add_pd(fiz1,tz);
551
552 fjx0 = _mm_add_pd(fjx0,tx);
553 fjy0 = _mm_add_pd(fjy0,ty);
554 fjz0 = _mm_add_pd(fjz0,tz);
555
556 /**************************
557 * CALCULATE INTERACTIONS *
558 **************************/
559
560 r20 = _mm_mul_pd(rsq20,rinv20);
561
562 /* Compute parameters for interactions between i and j atoms */
563 qq20 = _mm_mul_pd(iq2,jq0);
564
565 /* EWALD ELECTROSTATICS */
566
567 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
568 ewrt = _mm_mul_pd(r20,ewtabscale);
569 ewitab = _mm_cvttpd_epi32(ewrt);
570 eweps = _mm_sub_pd(ewrt,_mm_cvtepi32_pd(ewitab));
571 ewitab = _mm_slli_epi32(ewitab,2);
572 ewtabF = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,0) );
573 ewtabD = _mm_setzero_pd();
574 GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
575 ewtabV = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,0) +2);
576 ewtabFn = _mm_setzero_pd();
577 GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
578 felec = _mm_add_pd(ewtabF,_mm_mul_pd(eweps,ewtabD));
579 velec = _mm_sub_pd(ewtabV,_mm_mul_pd(_mm_mul_pd(ewtabhalfspace,eweps),_mm_add_pd(ewtabF,felec)));
580 velec = _mm_mul_pd(qq20,_mm_sub_pd(rinv20,velec));
581 felec = _mm_mul_pd(_mm_mul_pd(qq20,rinv20),_mm_sub_pd(rinvsq20,felec));
582
583 /* Update potential sum for this i atom from the interaction with this j atom. */
584 velec = _mm_unpacklo_pd(velec,_mm_setzero_pd());
585 velecsum = _mm_add_pd(velecsum,velec);
586
587 fscal = felec;
588
589 fscal = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
590
591 /* Calculate temporary vectorial force */
592 tx = _mm_mul_pd(fscal,dx20);
593 ty = _mm_mul_pd(fscal,dy20);
594 tz = _mm_mul_pd(fscal,dz20);
595
596 /* Update vectorial force */
597 fix2 = _mm_add_pd(fix2,tx);
598 fiy2 = _mm_add_pd(fiy2,ty);
599 fiz2 = _mm_add_pd(fiz2,tz);
600
601 fjx0 = _mm_add_pd(fjx0,tx);
602 fjy0 = _mm_add_pd(fjy0,ty);
603 fjz0 = _mm_add_pd(fjz0,tz);
604
605 gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,fjx0,fjy0,fjz0);
606
607 /* Inner loop uses 154 flops */
608 }
609
610 /* End of innermost loop */
611
612 gmx_mm_update_iforce_3atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
613 f+i_coord_offset,fshift+i_shift_offset);
614
615 ggid = gid[iidx];
616 /* Update potential energies */
617 gmx_mm_update_1pot_pd(velecsum,kernel_data->energygrp_elec+ggid);
618 gmx_mm_update_1pot_pd(vvdwsum,kernel_data->energygrp_vdw+ggid);
619
620 /* Increment number of inner iterations */
621 inneriter += j_index_end - j_index_start;
622
623 /* Outer loop uses 20 flops */
624 }
625
626 /* Increment number of outer iterations */
627 outeriter += nri;
628
629 /* Update outer/inner flops */
630
631 inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3_VF,outeriter*20 + inneriter*154);
632 }
633 /*
634 * Gromacs nonbonded kernel: nb_kernel_ElecEw_VdwLJEw_GeomW3P1_F_sse2_double
635 * Electrostatics interaction: Ewald
636 * VdW interaction: LJEwald
637 * Geometry: Water3-Particle
638 * Calculate force/pot: Force
639 */
640 void
641 nb_kernel_ElecEw_VdwLJEw_GeomW3P1_F_sse2_double
642 (t_nblist * gmx_restrict nlist,
643 rvec * gmx_restrict xx,
644 rvec * gmx_restrict ff,
645 t_forcerec * gmx_restrict fr,
646 t_mdatoms * gmx_restrict mdatoms,
647 nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
648 t_nrnb * gmx_restrict nrnb)
649 {
650 /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
651 * just 0 for non-waters.
652 * Suffixes A,B refer to j loop unrolling done with SSE double precision, e.g. for the two different
653 * jnr indices corresponding to data put in the four positions in the SIMD register.
654 */
655 int i_shift_offset,i_coord_offset,outeriter,inneriter;
656 int j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
657 int jnrA,jnrB;
658 int j_coord_offsetA,j_coord_offsetB;
659 int *iinr,*jindex,*jjnr,*shiftidx,*gid;
660 real rcutoff_scalar;
661 real *shiftvec,*fshift,*x,*f;
662 __m128d tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
663 int vdwioffset0;
664 __m128d ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
665 int vdwioffset1;
666 __m128d ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
667 int vdwioffset2;
668 __m128d ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
669 int vdwjidx0A,vdwjidx0B;
670 __m128d jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
671 __m128d dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
672 __m128d dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
673 __m128d dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
674 __m128d velec,felec,velecsum,facel,crf,krf,krf2;
675 real *charge;
676 int nvdwtype;
677 __m128d rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
678 int *vdwtype;
679 real *vdwparam;
680 __m128d one_sixth = _mm_set1_pd(1.0/6.0);
681 __m128d one_twelfth = _mm_set1_pd(1.0/12.0);
682 __m128d c6grid_00;
683 __m128d c6grid_10;
684 __m128d c6grid_20;
685 __m128d ewclj,ewclj2,ewclj6,ewcljrsq,poly,exponent,f6A,f6B,sh_lj_ewald;
686 real *vdwgridparam;
687 __m128d one_half = _mm_set1_pd(0.5);
688 __m128d minus_one = _mm_set1_pd(-1.0);
689 __m128i ewitab;
690 __m128d ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
691 real *ewtab;
692 __m128d dummy_mask,cutoff_mask;
693 __m128d signbit = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
694 __m128d one = _mm_set1_pd(1.0);
695 __m128d two = _mm_set1_pd(2.0);
696 x = xx[0];
697 f = ff[0];
698
699 nri = nlist->nri;
700 iinr = nlist->iinr;
701 jindex = nlist->jindex;
702 jjnr = nlist->jjnr;
703 shiftidx = nlist->shift;
704 gid = nlist->gid;
705 shiftvec = fr->shift_vec[0];
706 fshift = fr->fshift[0];
707 facel = _mm_set1_pd(fr->epsfac);
708 charge = mdatoms->chargeA;
709 nvdwtype = fr->ntype;
710 vdwparam = fr->nbfp;
711 vdwtype = mdatoms->typeA;
712 vdwgridparam = fr->ljpme_c6grid;
713 sh_lj_ewald = _mm_set1_pd(fr->ic->sh_lj_ewald);
714 ewclj = _mm_set1_pd(fr->ewaldcoeff_lj);
715 ewclj2 = _mm_mul_pd(minus_one,_mm_mul_pd(ewclj,ewclj));
716
717 sh_ewald = _mm_set1_pd(fr->ic->sh_ewald);
718 ewtab = fr->ic->tabq_coul_F;
719 ewtabscale = _mm_set1_pd(fr->ic->tabq_scale);
720 ewtabhalfspace = _mm_set1_pd(0.5/fr->ic->tabq_scale);
721
722 /* Setup water-specific parameters */
723 inr = nlist->iinr[0];
724 iq0 = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+0]));
725 iq1 = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+1]));
726 iq2 = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+2]));
727 vdwioffset0 = 2*nvdwtype*vdwtype[inr+0];
728
729 /* Avoid stupid compiler warnings */
730 jnrA = jnrB = 0;
731 j_coord_offsetA = 0;
732 j_coord_offsetB = 0;
733
734 outeriter = 0;
735 inneriter = 0;
736
737 /* Start outer loop over neighborlists */
738 for(iidx=0; iidx<nri; iidx++)
739 {
740 /* Load shift vector for this list */
741 i_shift_offset = DIM*shiftidx[iidx];
742
743 /* Load limits for loop over neighbors */
744 j_index_start = jindex[iidx];
745 j_index_end = jindex[iidx+1];
746
747 /* Get outer coordinate index */
748 inr = iinr[iidx];
749 i_coord_offset = DIM*inr;
750
751 /* Load i particle coords and add shift vector */
752 gmx_mm_load_shift_and_3rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
753 &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
754
755 fix0 = _mm_setzero_pd();
756 fiy0 = _mm_setzero_pd();
757 fiz0 = _mm_setzero_pd();
758 fix1 = _mm_setzero_pd();
759 fiy1 = _mm_setzero_pd();
760 fiz1 = _mm_setzero_pd();
761 fix2 = _mm_setzero_pd();
762 fiy2 = _mm_setzero_pd();
763 fiz2 = _mm_setzero_pd();
764
765 /* Start inner kernel loop */
766 for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
767 {
768
769 /* Get j neighbor index, and coordinate index */
770 jnrA = jjnr[jidx];
771 jnrB = jjnr[jidx+1];
772 j_coord_offsetA = DIM*jnrA;
773 j_coord_offsetB = DIM*jnrB;
774
775 /* load j atom coordinates */
776 gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
777 &jx0,&jy0,&jz0);
778
779 /* Calculate displacement vector */
780 dx00 = _mm_sub_pd(ix0,jx0);
781 dy00 = _mm_sub_pd(iy0,jy0);
782 dz00 = _mm_sub_pd(iz0,jz0);
783 dx10 = _mm_sub_pd(ix1,jx0);
784 dy10 = _mm_sub_pd(iy1,jy0);
785 dz10 = _mm_sub_pd(iz1,jz0);
786 dx20 = _mm_sub_pd(ix2,jx0);
787 dy20 = _mm_sub_pd(iy2,jy0);
788 dz20 = _mm_sub_pd(iz2,jz0);
789
790 /* Calculate squared distance and things based on it */
791 rsq00 = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
792 rsq10 = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
793 rsq20 = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
794
795 rinv00 = gmx_mm_invsqrt_pd(rsq00);
796 rinv10 = gmx_mm_invsqrt_pd(rsq10);
797 rinv20 = gmx_mm_invsqrt_pd(rsq20);
798
799 rinvsq00 = _mm_mul_pd(rinv00,rinv00);
800 rinvsq10 = _mm_mul_pd(rinv10,rinv10);
801 rinvsq20 = _mm_mul_pd(rinv20,rinv20);
802
803 /* Load parameters for j particles */
804 jq0 = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
805 vdwjidx0A = 2*vdwtype[jnrA+0];
806 vdwjidx0B = 2*vdwtype[jnrB+0];
807
808 fjx0 = _mm_setzero_pd();
809 fjy0 = _mm_setzero_pd();
810 fjz0 = _mm_setzero_pd();
811
812 /**************************
813 * CALCULATE INTERACTIONS *
814 **************************/
815
816 r00 = _mm_mul_pd(rsq00,rinv00);
817
818 /* Compute parameters for interactions between i and j atoms */
819 qq00 = _mm_mul_pd(iq0,jq0);
820 gmx_mm_load_2pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,
821 vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
822
823 c6grid_00 = gmx_mm_load_2real_swizzle_pd(vdwgridparam+vdwioffset0+vdwjidx0A,
824 vdwgridparam+vdwioffset0+vdwjidx0B);
825
826 /* EWALD ELECTROSTATICS */
827
828 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
829 ewrt = _mm_mul_pd(r00,ewtabscale);
830 ewitab = _mm_cvttpd_epi32(ewrt);
831 eweps = _mm_sub_pd(ewrt,_mm_cvtepi32_pd(ewitab));
832 gmx_mm_load_2pair_swizzle_pd(ewtab+gmx_mm_extract_epi32(ewitab,0),ewtab+gmx_mm_extract_epi32(ewitab,1),
833 &ewtabF,&ewtabFn);
834 felec = _mm_add_pd(_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF),_mm_mul_pd(eweps,ewtabFn));
835 felec = _mm_mul_pd(_mm_mul_pd(qq00,rinv00),_mm_sub_pd(rinvsq00,felec));
836
837 /* Analytical LJ-PME */
838 rinvsix = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
839 ewcljrsq = _mm_mul_pd(ewclj2,rsq00);
840 ewclj6 = _mm_mul_pd(ewclj2,_mm_mul_pd(ewclj2,ewclj2));
841 exponent = gmx_simd_exp_d(ewcljrsq);
842 /* poly = exp(-(beta*r)^2) * (1 + (beta*r)^2 + (beta*r)^4 /2) */
843 poly = _mm_mul_pd(exponent,_mm_add_pd(_mm_sub_pd(one,ewcljrsq),_mm_mul_pd(_mm_mul_pd(ewcljrsq,ewcljrsq),one_half)));
844 /* f6A = 6 * C6grid * (1 - poly) */
845 f6A = _mm_mul_pd(c6grid_00,_mm_sub_pd(one,poly));
846 /* f6B = C6grid * exponent * beta^6 */
847 f6B = _mm_mul_pd(_mm_mul_pd(c6grid_00,one_sixth),_mm_mul_pd(exponent,ewclj6));
848 /* fvdw = 12*C12/r13 - ((6*C6 - f6A)/r6 + f6B)/r */
849 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);
850
851 fscal = _mm_add_pd(felec,fvdw);
852
853 /* Calculate temporary vectorial force */
854 tx = _mm_mul_pd(fscal,dx00);
855 ty = _mm_mul_pd(fscal,dy00);
856 tz = _mm_mul_pd(fscal,dz00);
857
858 /* Update vectorial force */
859 fix0 = _mm_add_pd(fix0,tx);
860 fiy0 = _mm_add_pd(fiy0,ty);
861 fiz0 = _mm_add_pd(fiz0,tz);
862
863 fjx0 = _mm_add_pd(fjx0,tx);
864 fjy0 = _mm_add_pd(fjy0,ty);
865 fjz0 = _mm_add_pd(fjz0,tz);
866
867 /**************************
868 * CALCULATE INTERACTIONS *
869 **************************/
870
871 r10 = _mm_mul_pd(rsq10,rinv10);
872
873 /* Compute parameters for interactions between i and j atoms */
874 qq10 = _mm_mul_pd(iq1,jq0);
875
876 /* EWALD ELECTROSTATICS */
877
878 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
879 ewrt = _mm_mul_pd(r10,ewtabscale);
880 ewitab = _mm_cvttpd_epi32(ewrt);
881 eweps = _mm_sub_pd(ewrt,_mm_cvtepi32_pd(ewitab));
882 gmx_mm_load_2pair_swizzle_pd(ewtab+gmx_mm_extract_epi32(ewitab,0),ewtab+gmx_mm_extract_epi32(ewitab,1),
883 &ewtabF,&ewtabFn);
884 felec = _mm_add_pd(_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF),_mm_mul_pd(eweps,ewtabFn));
885 felec = _mm_mul_pd(_mm_mul_pd(qq10,rinv10),_mm_sub_pd(rinvsq10,felec));
886
887 fscal = felec;
888
889 /* Calculate temporary vectorial force */
890 tx = _mm_mul_pd(fscal,dx10);
891 ty = _mm_mul_pd(fscal,dy10);
892 tz = _mm_mul_pd(fscal,dz10);
893
894 /* Update vectorial force */
895 fix1 = _mm_add_pd(fix1,tx);
896 fiy1 = _mm_add_pd(fiy1,ty);
897 fiz1 = _mm_add_pd(fiz1,tz);
898
899 fjx0 = _mm_add_pd(fjx0,tx);
900 fjy0 = _mm_add_pd(fjy0,ty);
901 fjz0 = _mm_add_pd(fjz0,tz);
902
903 /**************************
904 * CALCULATE INTERACTIONS *
905 **************************/
906
907 r20 = _mm_mul_pd(rsq20,rinv20);
908
909 /* Compute parameters for interactions between i and j atoms */
910 qq20 = _mm_mul_pd(iq2,jq0);
911
912 /* EWALD ELECTROSTATICS */
913
914 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
915 ewrt = _mm_mul_pd(r20,ewtabscale);
916 ewitab = _mm_cvttpd_epi32(ewrt);
917 eweps = _mm_sub_pd(ewrt,_mm_cvtepi32_pd(ewitab));
918 gmx_mm_load_2pair_swizzle_pd(ewtab+gmx_mm_extract_epi32(ewitab,0),ewtab+gmx_mm_extract_epi32(ewitab,1),
919 &ewtabF,&ewtabFn);
920 felec = _mm_add_pd(_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF),_mm_mul_pd(eweps,ewtabFn));
921 felec = _mm_mul_pd(_mm_mul_pd(qq20,rinv20),_mm_sub_pd(rinvsq20,felec));
922
923 fscal = felec;
924
925 /* Calculate temporary vectorial force */
926 tx = _mm_mul_pd(fscal,dx20);
927 ty = _mm_mul_pd(fscal,dy20);
928 tz = _mm_mul_pd(fscal,dz20);
929
930 /* Update vectorial force */
931 fix2 = _mm_add_pd(fix2,tx);
932 fiy2 = _mm_add_pd(fiy2,ty);
933 fiz2 = _mm_add_pd(fiz2,tz);
934
935 fjx0 = _mm_add_pd(fjx0,tx);
936 fjy0 = _mm_add_pd(fjy0,ty);
937 fjz0 = _mm_add_pd(fjz0,tz);
938
939 gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,fjx0,fjy0,fjz0);
940
941 /* Inner loop uses 134 flops */
942 }
943
944 if(jidx<j_index_end)
945 {
946
947 jnrA = jjnr[jidx];
948 j_coord_offsetA = DIM*jnrA;
949
950 /* load j atom coordinates */
951 gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
952 &jx0,&jy0,&jz0);
953
954 /* Calculate displacement vector */
955 dx00 = _mm_sub_pd(ix0,jx0);
956 dy00 = _mm_sub_pd(iy0,jy0);
957 dz00 = _mm_sub_pd(iz0,jz0);
958 dx10 = _mm_sub_pd(ix1,jx0);
959 dy10 = _mm_sub_pd(iy1,jy0);
960 dz10 = _mm_sub_pd(iz1,jz0);
961 dx20 = _mm_sub_pd(ix2,jx0);
962 dy20 = _mm_sub_pd(iy2,jy0);
963 dz20 = _mm_sub_pd(iz2,jz0);
964
965 /* Calculate squared distance and things based on it */
966 rsq00 = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
967 rsq10 = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
968 rsq20 = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
969
970 rinv00 = gmx_mm_invsqrt_pd(rsq00);
971 rinv10 = gmx_mm_invsqrt_pd(rsq10);
972 rinv20 = gmx_mm_invsqrt_pd(rsq20);
973
974 rinvsq00 = _mm_mul_pd(rinv00,rinv00);
975 rinvsq10 = _mm_mul_pd(rinv10,rinv10);
976 rinvsq20 = _mm_mul_pd(rinv20,rinv20);
977
978 /* Load parameters for j particles */
979 jq0 = _mm_load_sd(charge+jnrA+0);
980 vdwjidx0A = 2*vdwtype[jnrA+0];
981
982 fjx0 = _mm_setzero_pd();
983 fjy0 = _mm_setzero_pd();
984 fjz0 = _mm_setzero_pd();
985
986 /**************************
987 * CALCULATE INTERACTIONS *
988 **************************/
989
990 r00 = _mm_mul_pd(rsq00,rinv00);
991
992 /* Compute parameters for interactions between i and j atoms */
993 qq00 = _mm_mul_pd(iq0,jq0);
994 gmx_mm_load_1pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,&c6_00,&c12_00);
995
996 c6grid_00 = gmx_mm_load_1real_pd(vdwgridparam+vdwioffset0+vdwjidx0A);
997
998 /* EWALD ELECTROSTATICS */
999
1000 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1001 ewrt = _mm_mul_pd(r00,ewtabscale);
1002 ewitab = _mm_cvttpd_epi32(ewrt);
1003 eweps = _mm_sub_pd(ewrt,_mm_cvtepi32_pd(ewitab));
1004 gmx_mm_load_1pair_swizzle_pd(ewtab+gmx_mm_extract_epi32(ewitab,0),&ewtabF,&ewtabFn);
1005 felec = _mm_add_pd(_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF),_mm_mul_pd(eweps,ewtabFn));
1006 felec = _mm_mul_pd(_mm_mul_pd(qq00,rinv00),_mm_sub_pd(rinvsq00,felec));
1007
1008 /* Analytical LJ-PME */
1009 rinvsix = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
1010 ewcljrsq = _mm_mul_pd(ewclj2,rsq00);
1011 ewclj6 = _mm_mul_pd(ewclj2,_mm_mul_pd(ewclj2,ewclj2));
1012 exponent = gmx_simd_exp_d(ewcljrsq);
1013 /* poly = exp(-(beta*r)^2) * (1 + (beta*r)^2 + (beta*r)^4 /2) */
1014 poly = _mm_mul_pd(exponent,_mm_add_pd(_mm_sub_pd(one,ewcljrsq),_mm_mul_pd(_mm_mul_pd(ewcljrsq,ewcljrsq),one_half)));
1015 /* f6A = 6 * C6grid * (1 - poly) */
1016 f6A = _mm_mul_pd(c6grid_00,_mm_sub_pd(one,poly));
1017 /* f6B = C6grid * exponent * beta^6 */
1018 f6B = _mm_mul_pd(_mm_mul_pd(c6grid_00,one_sixth),_mm_mul_pd(exponent,ewclj6));
1019 /* fvdw = 12*C12/r13 - ((6*C6 - f6A)/r6 + f6B)/r */
1020 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);
1021
1022 fscal = _mm_add_pd(felec,fvdw);
1023
1024 fscal = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1025
1026 /* Calculate temporary vectorial force */
1027 tx = _mm_mul_pd(fscal,dx00);
1028 ty = _mm_mul_pd(fscal,dy00);
1029 tz = _mm_mul_pd(fscal,dz00);
1030
1031 /* Update vectorial force */
1032 fix0 = _mm_add_pd(fix0,tx);
1033 fiy0 = _mm_add_pd(fiy0,ty);
1034 fiz0 = _mm_add_pd(fiz0,tz);
1035
1036 fjx0 = _mm_add_pd(fjx0,tx);
1037 fjy0 = _mm_add_pd(fjy0,ty);
1038 fjz0 = _mm_add_pd(fjz0,tz);
1039
1040 /**************************
1041 * CALCULATE INTERACTIONS *
1042 **************************/
1043
1044 r10 = _mm_mul_pd(rsq10,rinv10);
1045
1046 /* Compute parameters for interactions between i and j atoms */
1047 qq10 = _mm_mul_pd(iq1,jq0);
1048
1049 /* EWALD ELECTROSTATICS */
1050
1051 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1052 ewrt = _mm_mul_pd(r10,ewtabscale);
1053 ewitab = _mm_cvttpd_epi32(ewrt);
1054 eweps = _mm_sub_pd(ewrt,_mm_cvtepi32_pd(ewitab));
1055 gmx_mm_load_1pair_swizzle_pd(ewtab+gmx_mm_extract_epi32(ewitab,0),&ewtabF,&ewtabFn);
1056 felec = _mm_add_pd(_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF),_mm_mul_pd(eweps,ewtabFn));
1057 felec = _mm_mul_pd(_mm_mul_pd(qq10,rinv10),_mm_sub_pd(rinvsq10,felec));
1058
1059 fscal = felec;
1060
1061 fscal = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1062
1063 /* Calculate temporary vectorial force */
1064 tx = _mm_mul_pd(fscal,dx10);
1065 ty = _mm_mul_pd(fscal,dy10);
1066 tz = _mm_mul_pd(fscal,dz10);
1067
1068 /* Update vectorial force */
1069 fix1 = _mm_add_pd(fix1,tx);
1070 fiy1 = _mm_add_pd(fiy1,ty);
1071 fiz1 = _mm_add_pd(fiz1,tz);
1072
1073 fjx0 = _mm_add_pd(fjx0,tx);
1074 fjy0 = _mm_add_pd(fjy0,ty);
1075 fjz0 = _mm_add_pd(fjz0,tz);
1076
1077 /**************************
1078 * CALCULATE INTERACTIONS *
1079 **************************/
1080
1081 r20 = _mm_mul_pd(rsq20,rinv20);
1082
1083 /* Compute parameters for interactions between i and j atoms */
1084 qq20 = _mm_mul_pd(iq2,jq0);
1085
1086 /* EWALD ELECTROSTATICS */
1087
1088 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1089 ewrt = _mm_mul_pd(r20,ewtabscale);
1090 ewitab = _mm_cvttpd_epi32(ewrt);
1091 eweps = _mm_sub_pd(ewrt,_mm_cvtepi32_pd(ewitab));
1092 gmx_mm_load_1pair_swizzle_pd(ewtab+gmx_mm_extract_epi32(ewitab,0),&ewtabF,&ewtabFn);
1093 felec = _mm_add_pd(_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF),_mm_mul_pd(eweps,ewtabFn));
1094 felec = _mm_mul_pd(_mm_mul_pd(qq20,rinv20),_mm_sub_pd(rinvsq20,felec));
1095
1096 fscal = felec;
1097
1098 fscal = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1099
1100 /* Calculate temporary vectorial force */
1101 tx = _mm_mul_pd(fscal,dx20);
1102 ty = _mm_mul_pd(fscal,dy20);
1103 tz = _mm_mul_pd(fscal,dz20);
1104
1105 /* Update vectorial force */
1106 fix2 = _mm_add_pd(fix2,tx);
1107 fiy2 = _mm_add_pd(fiy2,ty);
1108 fiz2 = _mm_add_pd(fiz2,tz);
1109
1110 fjx0 = _mm_add_pd(fjx0,tx);
1111 fjy0 = _mm_add_pd(fjy0,ty);
1112 fjz0 = _mm_add_pd(fjz0,tz);
1113
1114 gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,fjx0,fjy0,fjz0);
1115
1116 /* Inner loop uses 134 flops */
1117 }
1118
1119 /* End of innermost loop */
1120
1121 gmx_mm_update_iforce_3atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
1122 f+i_coord_offset,fshift+i_shift_offset);
1123
1124 /* Increment number of inner iterations */
1125 inneriter += j_index_end - j_index_start;
1126
1127 /* Outer loop uses 18 flops */
1128 }
1129
1130 /* Increment number of outer iterations */
1131 outeriter += nri;
1132
1133 /* Update outer/inner flops */
1134
1135 inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3_F,outeriter*18 + inneriter*134);
1136 }
The ultimate molecular dynamics simulation package