?  search:
summary | log | graphiclog1 | graphiclog2 | commit | commitdiff | tree | refs | edit | fork
blame | history | raw | HEAD
Removed simple.h from nb_kernel_sse4_1_XX
[gromacs.git] / src / gromacs / gmxlib / nonbonded / nb_kernel_sse4_1_double / nb_kernel_ElecCoul_VdwCSTab_GeomW4P1_sse4_1_double.c
blob5381836544ba5533cd397625ba9e3409183975bf
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 sse4_1_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_sse4_1_double.h"
49 #include "kernelutil_x86_sse4_1_double.h"
50
51 /*
52 * Gromacs nonbonded kernel: nb_kernel_ElecCoul_VdwCSTab_GeomW4P1_VF_sse4_1_double
53 * Electrostatics interaction: Coulomb
54 * VdW interaction: CubicSplineTable
55 * Geometry: Water4-Particle
56 * Calculate force/pot: PotentialAndForce
57 */
58 void
59 nb_kernel_ElecCoul_VdwCSTab_GeomW4P1_VF_sse4_1_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 __m128i vfitab;
104 __m128i ifour = _mm_set1_epi32(4);
105 __m128d rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF;
106 real *vftab;
107 __m128d dummy_mask,cutoff_mask;
108 __m128d signbit = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
109 __m128d one = _mm_set1_pd(1.0);
110 __m128d two = _mm_set1_pd(2.0);
111 x = xx[0];
112 f = ff[0];
113
114 nri = nlist->nri;
115 iinr = nlist->iinr;
116 jindex = nlist->jindex;
117 jjnr = nlist->jjnr;
118 shiftidx = nlist->shift;
119 gid = nlist->gid;
120 shiftvec = fr->shift_vec[0];
121 fshift = fr->fshift[0];
122 facel = _mm_set1_pd(fr->epsfac);
123 charge = mdatoms->chargeA;
124 nvdwtype = fr->ntype;
125 vdwparam = fr->nbfp;
126 vdwtype = mdatoms->typeA;
127
128 vftab = kernel_data->table_vdw->data;
129 vftabscale = _mm_set1_pd(kernel_data->table_vdw->scale);
130
131 /* Setup water-specific parameters */
132 inr = nlist->iinr[0];
133 iq1 = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+1]));
134 iq2 = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+2]));
135 iq3 = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+3]));
136 vdwioffset0 = 2*nvdwtype*vdwtype[inr+0];
137
138 /* Avoid stupid compiler warnings */
139 jnrA = jnrB = 0;
140 j_coord_offsetA = 0;
141 j_coord_offsetB = 0;
142
143 outeriter = 0;
144 inneriter = 0;
145
146 /* Start outer loop over neighborlists */
147 for(iidx=0; iidx<nri; iidx++)
148 {
149 /* Load shift vector for this list */
150 i_shift_offset = DIM*shiftidx[iidx];
151
152 /* Load limits for loop over neighbors */
153 j_index_start = jindex[iidx];
154 j_index_end = jindex[iidx+1];
155
156 /* Get outer coordinate index */
157 inr = iinr[iidx];
158 i_coord_offset = DIM*inr;
159
160 /* Load i particle coords and add shift vector */
161 gmx_mm_load_shift_and_4rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
162 &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
163
164 fix0 = _mm_setzero_pd();
165 fiy0 = _mm_setzero_pd();
166 fiz0 = _mm_setzero_pd();
167 fix1 = _mm_setzero_pd();
168 fiy1 = _mm_setzero_pd();
169 fiz1 = _mm_setzero_pd();
170 fix2 = _mm_setzero_pd();
171 fiy2 = _mm_setzero_pd();
172 fiz2 = _mm_setzero_pd();
173 fix3 = _mm_setzero_pd();
174 fiy3 = _mm_setzero_pd();
175 fiz3 = _mm_setzero_pd();
176
177 /* Reset potential sums */
178 velecsum = _mm_setzero_pd();
179 vvdwsum = _mm_setzero_pd();
180
181 /* Start inner kernel loop */
182 for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
183 {
184
185 /* Get j neighbor index, and coordinate index */
186 jnrA = jjnr[jidx];
187 jnrB = jjnr[jidx+1];
188 j_coord_offsetA = DIM*jnrA;
189 j_coord_offsetB = DIM*jnrB;
190
191 /* load j atom coordinates */
192 gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
193 &jx0,&jy0,&jz0);
194
195 /* Calculate displacement vector */
196 dx00 = _mm_sub_pd(ix0,jx0);
197 dy00 = _mm_sub_pd(iy0,jy0);
198 dz00 = _mm_sub_pd(iz0,jz0);
199 dx10 = _mm_sub_pd(ix1,jx0);
200 dy10 = _mm_sub_pd(iy1,jy0);
201 dz10 = _mm_sub_pd(iz1,jz0);
202 dx20 = _mm_sub_pd(ix2,jx0);
203 dy20 = _mm_sub_pd(iy2,jy0);
204 dz20 = _mm_sub_pd(iz2,jz0);
205 dx30 = _mm_sub_pd(ix3,jx0);
206 dy30 = _mm_sub_pd(iy3,jy0);
207 dz30 = _mm_sub_pd(iz3,jz0);
208
209 /* Calculate squared distance and things based on it */
210 rsq00 = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
211 rsq10 = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
212 rsq20 = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
213 rsq30 = gmx_mm_calc_rsq_pd(dx30,dy30,dz30);
214
215 rinv00 = gmx_mm_invsqrt_pd(rsq00);
216 rinv10 = gmx_mm_invsqrt_pd(rsq10);
217 rinv20 = gmx_mm_invsqrt_pd(rsq20);
218 rinv30 = gmx_mm_invsqrt_pd(rsq30);
219
220 rinvsq10 = _mm_mul_pd(rinv10,rinv10);
221 rinvsq20 = _mm_mul_pd(rinv20,rinv20);
222 rinvsq30 = _mm_mul_pd(rinv30,rinv30);
223
224 /* Load parameters for j particles */
225 jq0 = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
226 vdwjidx0A = 2*vdwtype[jnrA+0];
227 vdwjidx0B = 2*vdwtype[jnrB+0];
228
229 fjx0 = _mm_setzero_pd();
230 fjy0 = _mm_setzero_pd();
231 fjz0 = _mm_setzero_pd();
232
233 /**************************
234 * CALCULATE INTERACTIONS *
235 **************************/
236
237 r00 = _mm_mul_pd(rsq00,rinv00);
238
239 /* Compute parameters for interactions between i and j atoms */
240 gmx_mm_load_2pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,
241 vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
242
243 /* Calculate table index by multiplying r with table scale and truncate to integer */
244 rt = _mm_mul_pd(r00,vftabscale);
245 vfitab = _mm_cvttpd_epi32(rt);
246 vfeps = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
247 vfitab = _mm_slli_epi32(vfitab,3);
248
249 /* CUBIC SPLINE TABLE DISPERSION */
250 Y = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
251 F = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) );
252 GMX_MM_TRANSPOSE2_PD(Y,F);
253 G = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
254 H = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) +2);
255 GMX_MM_TRANSPOSE2_PD(G,H);
256 Heps = _mm_mul_pd(vfeps,H);
257 Fp = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
258 VV = _mm_add_pd(Y,_mm_mul_pd(vfeps,Fp));
259 vvdw6 = _mm_mul_pd(c6_00,VV);
260 FF = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
261 fvdw6 = _mm_mul_pd(c6_00,FF);
262
263 /* CUBIC SPLINE TABLE REPULSION */
264 vfitab = _mm_add_epi32(vfitab,ifour);
265 Y = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
266 F = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) );
267 GMX_MM_TRANSPOSE2_PD(Y,F);
268 G = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
269 H = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) +2);
270 GMX_MM_TRANSPOSE2_PD(G,H);
271 Heps = _mm_mul_pd(vfeps,H);
272 Fp = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
273 VV = _mm_add_pd(Y,_mm_mul_pd(vfeps,Fp));
274 vvdw12 = _mm_mul_pd(c12_00,VV);
275 FF = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
276 fvdw12 = _mm_mul_pd(c12_00,FF);
277 vvdw = _mm_add_pd(vvdw12,vvdw6);
278 fvdw = _mm_xor_pd(signbit,_mm_mul_pd(_mm_add_pd(fvdw6,fvdw12),_mm_mul_pd(vftabscale,rinv00)));
279
280 /* Update potential sum for this i atom from the interaction with this j atom. */
281 vvdwsum = _mm_add_pd(vvdwsum,vvdw);
282
283 fscal = fvdw;
284
285 /* Calculate temporary vectorial force */
286 tx = _mm_mul_pd(fscal,dx00);
287 ty = _mm_mul_pd(fscal,dy00);
288 tz = _mm_mul_pd(fscal,dz00);
289
290 /* Update vectorial force */
291 fix0 = _mm_add_pd(fix0,tx);
292 fiy0 = _mm_add_pd(fiy0,ty);
293 fiz0 = _mm_add_pd(fiz0,tz);
294
295 fjx0 = _mm_add_pd(fjx0,tx);
296 fjy0 = _mm_add_pd(fjy0,ty);
297 fjz0 = _mm_add_pd(fjz0,tz);
298
299 /**************************
300 * CALCULATE INTERACTIONS *
301 **************************/
302
303 /* Compute parameters for interactions between i and j atoms */
304 qq10 = _mm_mul_pd(iq1,jq0);
305
306 /* COULOMB ELECTROSTATICS */
307 velec = _mm_mul_pd(qq10,rinv10);
308 felec = _mm_mul_pd(velec,rinvsq10);
309
310 /* Update potential sum for this i atom from the interaction with this j atom. */
311 velecsum = _mm_add_pd(velecsum,velec);
312
313 fscal = felec;
314
315 /* Calculate temporary vectorial force */
316 tx = _mm_mul_pd(fscal,dx10);
317 ty = _mm_mul_pd(fscal,dy10);
318 tz = _mm_mul_pd(fscal,dz10);
319
320 /* Update vectorial force */
321 fix1 = _mm_add_pd(fix1,tx);
322 fiy1 = _mm_add_pd(fiy1,ty);
323 fiz1 = _mm_add_pd(fiz1,tz);
324
325 fjx0 = _mm_add_pd(fjx0,tx);
326 fjy0 = _mm_add_pd(fjy0,ty);
327 fjz0 = _mm_add_pd(fjz0,tz);
328
329 /**************************
330 * CALCULATE INTERACTIONS *
331 **************************/
332
333 /* Compute parameters for interactions between i and j atoms */
334 qq20 = _mm_mul_pd(iq2,jq0);
335
336 /* COULOMB ELECTROSTATICS */
337 velec = _mm_mul_pd(qq20,rinv20);
338 felec = _mm_mul_pd(velec,rinvsq20);
339
340 /* Update potential sum for this i atom from the interaction with this j atom. */
341 velecsum = _mm_add_pd(velecsum,velec);
342
343 fscal = felec;
344
345 /* Calculate temporary vectorial force */
346 tx = _mm_mul_pd(fscal,dx20);
347 ty = _mm_mul_pd(fscal,dy20);
348 tz = _mm_mul_pd(fscal,dz20);
349
350 /* Update vectorial force */
351 fix2 = _mm_add_pd(fix2,tx);
352 fiy2 = _mm_add_pd(fiy2,ty);
353 fiz2 = _mm_add_pd(fiz2,tz);
354
355 fjx0 = _mm_add_pd(fjx0,tx);
356 fjy0 = _mm_add_pd(fjy0,ty);
357 fjz0 = _mm_add_pd(fjz0,tz);
358
359 /**************************
360 * CALCULATE INTERACTIONS *
361 **************************/
362
363 /* Compute parameters for interactions between i and j atoms */
364 qq30 = _mm_mul_pd(iq3,jq0);
365
366 /* COULOMB ELECTROSTATICS */
367 velec = _mm_mul_pd(qq30,rinv30);
368 felec = _mm_mul_pd(velec,rinvsq30);
369
370 /* Update potential sum for this i atom from the interaction with this j atom. */
371 velecsum = _mm_add_pd(velecsum,velec);
372
373 fscal = felec;
374
375 /* Calculate temporary vectorial force */
376 tx = _mm_mul_pd(fscal,dx30);
377 ty = _mm_mul_pd(fscal,dy30);
378 tz = _mm_mul_pd(fscal,dz30);
379
380 /* Update vectorial force */
381 fix3 = _mm_add_pd(fix3,tx);
382 fiy3 = _mm_add_pd(fiy3,ty);
383 fiz3 = _mm_add_pd(fiz3,tz);
384
385 fjx0 = _mm_add_pd(fjx0,tx);
386 fjy0 = _mm_add_pd(fjy0,ty);
387 fjz0 = _mm_add_pd(fjz0,tz);
388
389 gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,fjx0,fjy0,fjz0);
390
391 /* Inner loop uses 143 flops */
392 }
393
394 if(jidx<j_index_end)
395 {
396
397 jnrA = jjnr[jidx];
398 j_coord_offsetA = DIM*jnrA;
399
400 /* load j atom coordinates */
401 gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
402 &jx0,&jy0,&jz0);
403
404 /* Calculate displacement vector */
405 dx00 = _mm_sub_pd(ix0,jx0);
406 dy00 = _mm_sub_pd(iy0,jy0);
407 dz00 = _mm_sub_pd(iz0,jz0);
408 dx10 = _mm_sub_pd(ix1,jx0);
409 dy10 = _mm_sub_pd(iy1,jy0);
410 dz10 = _mm_sub_pd(iz1,jz0);
411 dx20 = _mm_sub_pd(ix2,jx0);
412 dy20 = _mm_sub_pd(iy2,jy0);
413 dz20 = _mm_sub_pd(iz2,jz0);
414 dx30 = _mm_sub_pd(ix3,jx0);
415 dy30 = _mm_sub_pd(iy3,jy0);
416 dz30 = _mm_sub_pd(iz3,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 rsq30 = gmx_mm_calc_rsq_pd(dx30,dy30,dz30);
423
424 rinv00 = gmx_mm_invsqrt_pd(rsq00);
425 rinv10 = gmx_mm_invsqrt_pd(rsq10);
426 rinv20 = gmx_mm_invsqrt_pd(rsq20);
427 rinv30 = gmx_mm_invsqrt_pd(rsq30);
428
429 rinvsq10 = _mm_mul_pd(rinv10,rinv10);
430 rinvsq20 = _mm_mul_pd(rinv20,rinv20);
431 rinvsq30 = _mm_mul_pd(rinv30,rinv30);
432
433 /* Load parameters for j particles */
434 jq0 = _mm_load_sd(charge+jnrA+0);
435 vdwjidx0A = 2*vdwtype[jnrA+0];
436
437 fjx0 = _mm_setzero_pd();
438 fjy0 = _mm_setzero_pd();
439 fjz0 = _mm_setzero_pd();
440
441 /**************************
442 * CALCULATE INTERACTIONS *
443 **************************/
444
445 r00 = _mm_mul_pd(rsq00,rinv00);
446
447 /* Compute parameters for interactions between i and j atoms */
448 gmx_mm_load_1pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,&c6_00,&c12_00);
449
450 /* Calculate table index by multiplying r with table scale and truncate to integer */
451 rt = _mm_mul_pd(r00,vftabscale);
452 vfitab = _mm_cvttpd_epi32(rt);
453 vfeps = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
454 vfitab = _mm_slli_epi32(vfitab,3);
455
456 /* CUBIC SPLINE TABLE DISPERSION */
457 Y = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
458 F = _mm_setzero_pd();
459 GMX_MM_TRANSPOSE2_PD(Y,F);
460 G = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
461 H = _mm_setzero_pd();
462 GMX_MM_TRANSPOSE2_PD(G,H);
463 Heps = _mm_mul_pd(vfeps,H);
464 Fp = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
465 VV = _mm_add_pd(Y,_mm_mul_pd(vfeps,Fp));
466 vvdw6 = _mm_mul_pd(c6_00,VV);
467 FF = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
468 fvdw6 = _mm_mul_pd(c6_00,FF);
469
470 /* CUBIC SPLINE TABLE REPULSION */
471 vfitab = _mm_add_epi32(vfitab,ifour);
472 Y = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
473 F = _mm_setzero_pd();
474 GMX_MM_TRANSPOSE2_PD(Y,F);
475 G = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
476 H = _mm_setzero_pd();
477 GMX_MM_TRANSPOSE2_PD(G,H);
478 Heps = _mm_mul_pd(vfeps,H);
479 Fp = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
480 VV = _mm_add_pd(Y,_mm_mul_pd(vfeps,Fp));
481 vvdw12 = _mm_mul_pd(c12_00,VV);
482 FF = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
483 fvdw12 = _mm_mul_pd(c12_00,FF);
484 vvdw = _mm_add_pd(vvdw12,vvdw6);
485 fvdw = _mm_xor_pd(signbit,_mm_mul_pd(_mm_add_pd(fvdw6,fvdw12),_mm_mul_pd(vftabscale,rinv00)));
486
487 /* Update potential sum for this i atom from the interaction with this j atom. */
488 vvdw = _mm_unpacklo_pd(vvdw,_mm_setzero_pd());
489 vvdwsum = _mm_add_pd(vvdwsum,vvdw);
490
491 fscal = fvdw;
492
493 fscal = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
494
495 /* Calculate temporary vectorial force */
496 tx = _mm_mul_pd(fscal,dx00);
497 ty = _mm_mul_pd(fscal,dy00);
498 tz = _mm_mul_pd(fscal,dz00);
499
500 /* Update vectorial force */
501 fix0 = _mm_add_pd(fix0,tx);
502 fiy0 = _mm_add_pd(fiy0,ty);
503 fiz0 = _mm_add_pd(fiz0,tz);
504
505 fjx0 = _mm_add_pd(fjx0,tx);
506 fjy0 = _mm_add_pd(fjy0,ty);
507 fjz0 = _mm_add_pd(fjz0,tz);
508
509 /**************************
510 * CALCULATE INTERACTIONS *
511 **************************/
512
513 /* Compute parameters for interactions between i and j atoms */
514 qq10 = _mm_mul_pd(iq1,jq0);
515
516 /* COULOMB ELECTROSTATICS */
517 velec = _mm_mul_pd(qq10,rinv10);
518 felec = _mm_mul_pd(velec,rinvsq10);
519
520 /* Update potential sum for this i atom from the interaction with this j atom. */
521 velec = _mm_unpacklo_pd(velec,_mm_setzero_pd());
522 velecsum = _mm_add_pd(velecsum,velec);
523
524 fscal = felec;
525
526 fscal = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
527
528 /* Calculate temporary vectorial force */
529 tx = _mm_mul_pd(fscal,dx10);
530 ty = _mm_mul_pd(fscal,dy10);
531 tz = _mm_mul_pd(fscal,dz10);
532
533 /* Update vectorial force */
534 fix1 = _mm_add_pd(fix1,tx);
535 fiy1 = _mm_add_pd(fiy1,ty);
536 fiz1 = _mm_add_pd(fiz1,tz);
537
538 fjx0 = _mm_add_pd(fjx0,tx);
539 fjy0 = _mm_add_pd(fjy0,ty);
540 fjz0 = _mm_add_pd(fjz0,tz);
541
542 /**************************
543 * CALCULATE INTERACTIONS *
544 **************************/
545
546 /* Compute parameters for interactions between i and j atoms */
547 qq20 = _mm_mul_pd(iq2,jq0);
548
549 /* COULOMB ELECTROSTATICS */
550 velec = _mm_mul_pd(qq20,rinv20);
551 felec = _mm_mul_pd(velec,rinvsq20);
552
553 /* Update potential sum for this i atom from the interaction with this j atom. */
554 velec = _mm_unpacklo_pd(velec,_mm_setzero_pd());
555 velecsum = _mm_add_pd(velecsum,velec);
556
557 fscal = felec;
558
559 fscal = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
560
561 /* Calculate temporary vectorial force */
562 tx = _mm_mul_pd(fscal,dx20);
563 ty = _mm_mul_pd(fscal,dy20);
564 tz = _mm_mul_pd(fscal,dz20);
565
566 /* Update vectorial force */
567 fix2 = _mm_add_pd(fix2,tx);
568 fiy2 = _mm_add_pd(fiy2,ty);
569 fiz2 = _mm_add_pd(fiz2,tz);
570
571 fjx0 = _mm_add_pd(fjx0,tx);
572 fjy0 = _mm_add_pd(fjy0,ty);
573 fjz0 = _mm_add_pd(fjz0,tz);
574
575 /**************************
576 * CALCULATE INTERACTIONS *
577 **************************/
578
579 /* Compute parameters for interactions between i and j atoms */
580 qq30 = _mm_mul_pd(iq3,jq0);
581
582 /* COULOMB ELECTROSTATICS */
583 velec = _mm_mul_pd(qq30,rinv30);
584 felec = _mm_mul_pd(velec,rinvsq30);
585
586 /* Update potential sum for this i atom from the interaction with this j atom. */
587 velec = _mm_unpacklo_pd(velec,_mm_setzero_pd());
588 velecsum = _mm_add_pd(velecsum,velec);
589
590 fscal = felec;
591
592 fscal = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
593
594 /* Calculate temporary vectorial force */
595 tx = _mm_mul_pd(fscal,dx30);
596 ty = _mm_mul_pd(fscal,dy30);
597 tz = _mm_mul_pd(fscal,dz30);
598
599 /* Update vectorial force */
600 fix3 = _mm_add_pd(fix3,tx);
601 fiy3 = _mm_add_pd(fiy3,ty);
602 fiz3 = _mm_add_pd(fiz3,tz);
603
604 fjx0 = _mm_add_pd(fjx0,tx);
605 fjy0 = _mm_add_pd(fjy0,ty);
606 fjz0 = _mm_add_pd(fjz0,tz);
607
608 gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,fjx0,fjy0,fjz0);
609
610 /* Inner loop uses 143 flops */
611 }
612
613 /* End of innermost loop */
614
615 gmx_mm_update_iforce_4atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
616 f+i_coord_offset,fshift+i_shift_offset);
617
618 ggid = gid[iidx];
619 /* Update potential energies */
620 gmx_mm_update_1pot_pd(velecsum,kernel_data->energygrp_elec+ggid);
621 gmx_mm_update_1pot_pd(vvdwsum,kernel_data->energygrp_vdw+ggid);
622
623 /* Increment number of inner iterations */
624 inneriter += j_index_end - j_index_start;
625
626 /* Outer loop uses 26 flops */
627 }
628
629 /* Increment number of outer iterations */
630 outeriter += nri;
631
632 /* Update outer/inner flops */
633
634 inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W4_VF,outeriter*26 + inneriter*143);
635 }
636 /*
637 * Gromacs nonbonded kernel: nb_kernel_ElecCoul_VdwCSTab_GeomW4P1_F_sse4_1_double
638 * Electrostatics interaction: Coulomb
639 * VdW interaction: CubicSplineTable
640 * Geometry: Water4-Particle
641 * Calculate force/pot: Force
642 */
643 void
644 nb_kernel_ElecCoul_VdwCSTab_GeomW4P1_F_sse4_1_double
645 (t_nblist * gmx_restrict nlist,
646 rvec * gmx_restrict xx,
647 rvec * gmx_restrict ff,
648 t_forcerec * gmx_restrict fr,
649 t_mdatoms * gmx_restrict mdatoms,
650 nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
651 t_nrnb * gmx_restrict nrnb)
652 {
653 /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
654 * just 0 for non-waters.
655 * Suffixes A,B refer to j loop unrolling done with SSE double precision, e.g. for the two different
656 * jnr indices corresponding to data put in the four positions in the SIMD register.
657 */
658 int i_shift_offset,i_coord_offset,outeriter,inneriter;
659 int j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
660 int jnrA,jnrB;
661 int j_coord_offsetA,j_coord_offsetB;
662 int *iinr,*jindex,*jjnr,*shiftidx,*gid;
663 real rcutoff_scalar;
664 real *shiftvec,*fshift,*x,*f;
665 __m128d tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
666 int vdwioffset0;
667 __m128d ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
668 int vdwioffset1;
669 __m128d ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
670 int vdwioffset2;
671 __m128d ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
672 int vdwioffset3;
673 __m128d ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
674 int vdwjidx0A,vdwjidx0B;
675 __m128d jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
676 __m128d dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
677 __m128d dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
678 __m128d dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
679 __m128d dx30,dy30,dz30,rsq30,rinv30,rinvsq30,r30,qq30,c6_30,c12_30;
680 __m128d velec,felec,velecsum,facel,crf,krf,krf2;
681 real *charge;
682 int nvdwtype;
683 __m128d rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
684 int *vdwtype;
685 real *vdwparam;
686 __m128d one_sixth = _mm_set1_pd(1.0/6.0);
687 __m128d one_twelfth = _mm_set1_pd(1.0/12.0);
688 __m128i vfitab;
689 __m128i ifour = _mm_set1_epi32(4);
690 __m128d rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF;
691 real *vftab;
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
713 vftab = kernel_data->table_vdw->data;
714 vftabscale = _mm_set1_pd(kernel_data->table_vdw->scale);
715
716 /* Setup water-specific parameters */
717 inr = nlist->iinr[0];
718 iq1 = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+1]));
719 iq2 = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+2]));
720 iq3 = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+3]));
721 vdwioffset0 = 2*nvdwtype*vdwtype[inr+0];
722
723 /* Avoid stupid compiler warnings */
724 jnrA = jnrB = 0;
725 j_coord_offsetA = 0;
726 j_coord_offsetB = 0;
727
728 outeriter = 0;
729 inneriter = 0;
730
731 /* Start outer loop over neighborlists */
732 for(iidx=0; iidx<nri; iidx++)
733 {
734 /* Load shift vector for this list */
735 i_shift_offset = DIM*shiftidx[iidx];
736
737 /* Load limits for loop over neighbors */
738 j_index_start = jindex[iidx];
739 j_index_end = jindex[iidx+1];
740
741 /* Get outer coordinate index */
742 inr = iinr[iidx];
743 i_coord_offset = DIM*inr;
744
745 /* Load i particle coords and add shift vector */
746 gmx_mm_load_shift_and_4rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
747 &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
748
749 fix0 = _mm_setzero_pd();
750 fiy0 = _mm_setzero_pd();
751 fiz0 = _mm_setzero_pd();
752 fix1 = _mm_setzero_pd();
753 fiy1 = _mm_setzero_pd();
754 fiz1 = _mm_setzero_pd();
755 fix2 = _mm_setzero_pd();
756 fiy2 = _mm_setzero_pd();
757 fiz2 = _mm_setzero_pd();
758 fix3 = _mm_setzero_pd();
759 fiy3 = _mm_setzero_pd();
760 fiz3 = _mm_setzero_pd();
761
762 /* Start inner kernel loop */
763 for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
764 {
765
766 /* Get j neighbor index, and coordinate index */
767 jnrA = jjnr[jidx];
768 jnrB = jjnr[jidx+1];
769 j_coord_offsetA = DIM*jnrA;
770 j_coord_offsetB = DIM*jnrB;
771
772 /* load j atom coordinates */
773 gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
774 &jx0,&jy0,&jz0);
775
776 /* Calculate displacement vector */
777 dx00 = _mm_sub_pd(ix0,jx0);
778 dy00 = _mm_sub_pd(iy0,jy0);
779 dz00 = _mm_sub_pd(iz0,jz0);
780 dx10 = _mm_sub_pd(ix1,jx0);
781 dy10 = _mm_sub_pd(iy1,jy0);
782 dz10 = _mm_sub_pd(iz1,jz0);
783 dx20 = _mm_sub_pd(ix2,jx0);
784 dy20 = _mm_sub_pd(iy2,jy0);
785 dz20 = _mm_sub_pd(iz2,jz0);
786 dx30 = _mm_sub_pd(ix3,jx0);
787 dy30 = _mm_sub_pd(iy3,jy0);
788 dz30 = _mm_sub_pd(iz3,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 rsq30 = gmx_mm_calc_rsq_pd(dx30,dy30,dz30);
795
796 rinv00 = gmx_mm_invsqrt_pd(rsq00);
797 rinv10 = gmx_mm_invsqrt_pd(rsq10);
798 rinv20 = gmx_mm_invsqrt_pd(rsq20);
799 rinv30 = gmx_mm_invsqrt_pd(rsq30);
800
801 rinvsq10 = _mm_mul_pd(rinv10,rinv10);
802 rinvsq20 = _mm_mul_pd(rinv20,rinv20);
803 rinvsq30 = _mm_mul_pd(rinv30,rinv30);
804
805 /* Load parameters for j particles */
806 jq0 = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
807 vdwjidx0A = 2*vdwtype[jnrA+0];
808 vdwjidx0B = 2*vdwtype[jnrB+0];
809
810 fjx0 = _mm_setzero_pd();
811 fjy0 = _mm_setzero_pd();
812 fjz0 = _mm_setzero_pd();
813
814 /**************************
815 * CALCULATE INTERACTIONS *
816 **************************/
817
818 r00 = _mm_mul_pd(rsq00,rinv00);
819
820 /* Compute parameters for interactions between i and j atoms */
821 gmx_mm_load_2pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,
822 vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
823
824 /* Calculate table index by multiplying r with table scale and truncate to integer */
825 rt = _mm_mul_pd(r00,vftabscale);
826 vfitab = _mm_cvttpd_epi32(rt);
827 vfeps = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
828 vfitab = _mm_slli_epi32(vfitab,3);
829
830 /* CUBIC SPLINE TABLE DISPERSION */
831 Y = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
832 F = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) );
833 GMX_MM_TRANSPOSE2_PD(Y,F);
834 G = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
835 H = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) +2);
836 GMX_MM_TRANSPOSE2_PD(G,H);
837 Heps = _mm_mul_pd(vfeps,H);
838 Fp = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
839 FF = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
840 fvdw6 = _mm_mul_pd(c6_00,FF);
841
842 /* CUBIC SPLINE TABLE REPULSION */
843 vfitab = _mm_add_epi32(vfitab,ifour);
844 Y = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
845 F = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) );
846 GMX_MM_TRANSPOSE2_PD(Y,F);
847 G = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
848 H = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) +2);
849 GMX_MM_TRANSPOSE2_PD(G,H);
850 Heps = _mm_mul_pd(vfeps,H);
851 Fp = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
852 FF = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
853 fvdw12 = _mm_mul_pd(c12_00,FF);
854 fvdw = _mm_xor_pd(signbit,_mm_mul_pd(_mm_add_pd(fvdw6,fvdw12),_mm_mul_pd(vftabscale,rinv00)));
855
856 fscal = fvdw;
857
858 /* Calculate temporary vectorial force */
859 tx = _mm_mul_pd(fscal,dx00);
860 ty = _mm_mul_pd(fscal,dy00);
861 tz = _mm_mul_pd(fscal,dz00);
862
863 /* Update vectorial force */
864 fix0 = _mm_add_pd(fix0,tx);
865 fiy0 = _mm_add_pd(fiy0,ty);
866 fiz0 = _mm_add_pd(fiz0,tz);
867
868 fjx0 = _mm_add_pd(fjx0,tx);
869 fjy0 = _mm_add_pd(fjy0,ty);
870 fjz0 = _mm_add_pd(fjz0,tz);
871
872 /**************************
873 * CALCULATE INTERACTIONS *
874 **************************/
875
876 /* Compute parameters for interactions between i and j atoms */
877 qq10 = _mm_mul_pd(iq1,jq0);
878
879 /* COULOMB ELECTROSTATICS */
880 velec = _mm_mul_pd(qq10,rinv10);
881 felec = _mm_mul_pd(velec,rinvsq10);
882
883 fscal = felec;
884
885 /* Calculate temporary vectorial force */
886 tx = _mm_mul_pd(fscal,dx10);
887 ty = _mm_mul_pd(fscal,dy10);
888 tz = _mm_mul_pd(fscal,dz10);
889
890 /* Update vectorial force */
891 fix1 = _mm_add_pd(fix1,tx);
892 fiy1 = _mm_add_pd(fiy1,ty);
893 fiz1 = _mm_add_pd(fiz1,tz);
894
895 fjx0 = _mm_add_pd(fjx0,tx);
896 fjy0 = _mm_add_pd(fjy0,ty);
897 fjz0 = _mm_add_pd(fjz0,tz);
898
899 /**************************
900 * CALCULATE INTERACTIONS *
901 **************************/
902
903 /* Compute parameters for interactions between i and j atoms */
904 qq20 = _mm_mul_pd(iq2,jq0);
905
906 /* COULOMB ELECTROSTATICS */
907 velec = _mm_mul_pd(qq20,rinv20);
908 felec = _mm_mul_pd(velec,rinvsq20);
909
910 fscal = felec;
911
912 /* Calculate temporary vectorial force */
913 tx = _mm_mul_pd(fscal,dx20);
914 ty = _mm_mul_pd(fscal,dy20);
915 tz = _mm_mul_pd(fscal,dz20);
916
917 /* Update vectorial force */
918 fix2 = _mm_add_pd(fix2,tx);
919 fiy2 = _mm_add_pd(fiy2,ty);
920 fiz2 = _mm_add_pd(fiz2,tz);
921
922 fjx0 = _mm_add_pd(fjx0,tx);
923 fjy0 = _mm_add_pd(fjy0,ty);
924 fjz0 = _mm_add_pd(fjz0,tz);
925
926 /**************************
927 * CALCULATE INTERACTIONS *
928 **************************/
929
930 /* Compute parameters for interactions between i and j atoms */
931 qq30 = _mm_mul_pd(iq3,jq0);
932
933 /* COULOMB ELECTROSTATICS */
934 velec = _mm_mul_pd(qq30,rinv30);
935 felec = _mm_mul_pd(velec,rinvsq30);
936
937 fscal = felec;
938
939 /* Calculate temporary vectorial force */
940 tx = _mm_mul_pd(fscal,dx30);
941 ty = _mm_mul_pd(fscal,dy30);
942 tz = _mm_mul_pd(fscal,dz30);
943
944 /* Update vectorial force */
945 fix3 = _mm_add_pd(fix3,tx);
946 fiy3 = _mm_add_pd(fiy3,ty);
947 fiz3 = _mm_add_pd(fiz3,tz);
948
949 fjx0 = _mm_add_pd(fjx0,tx);
950 fjy0 = _mm_add_pd(fjy0,ty);
951 fjz0 = _mm_add_pd(fjz0,tz);
952
953 gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,fjx0,fjy0,fjz0);
954
955 /* Inner loop uses 132 flops */
956 }
957
958 if(jidx<j_index_end)
959 {
960
961 jnrA = jjnr[jidx];
962 j_coord_offsetA = DIM*jnrA;
963
964 /* load j atom coordinates */
965 gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
966 &jx0,&jy0,&jz0);
967
968 /* Calculate displacement vector */
969 dx00 = _mm_sub_pd(ix0,jx0);
970 dy00 = _mm_sub_pd(iy0,jy0);
971 dz00 = _mm_sub_pd(iz0,jz0);
972 dx10 = _mm_sub_pd(ix1,jx0);
973 dy10 = _mm_sub_pd(iy1,jy0);
974 dz10 = _mm_sub_pd(iz1,jz0);
975 dx20 = _mm_sub_pd(ix2,jx0);
976 dy20 = _mm_sub_pd(iy2,jy0);
977 dz20 = _mm_sub_pd(iz2,jz0);
978 dx30 = _mm_sub_pd(ix3,jx0);
979 dy30 = _mm_sub_pd(iy3,jy0);
980 dz30 = _mm_sub_pd(iz3,jz0);
981
982 /* Calculate squared distance and things based on it */
983 rsq00 = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
984 rsq10 = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
985 rsq20 = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
986 rsq30 = gmx_mm_calc_rsq_pd(dx30,dy30,dz30);
987
988 rinv00 = gmx_mm_invsqrt_pd(rsq00);
989 rinv10 = gmx_mm_invsqrt_pd(rsq10);
990 rinv20 = gmx_mm_invsqrt_pd(rsq20);
991 rinv30 = gmx_mm_invsqrt_pd(rsq30);
992
993 rinvsq10 = _mm_mul_pd(rinv10,rinv10);
994 rinvsq20 = _mm_mul_pd(rinv20,rinv20);
995 rinvsq30 = _mm_mul_pd(rinv30,rinv30);
996
997 /* Load parameters for j particles */
998 jq0 = _mm_load_sd(charge+jnrA+0);
999 vdwjidx0A = 2*vdwtype[jnrA+0];
1000
1001 fjx0 = _mm_setzero_pd();
1002 fjy0 = _mm_setzero_pd();
1003 fjz0 = _mm_setzero_pd();
1004
1005 /**************************
1006 * CALCULATE INTERACTIONS *
1007 **************************/
1008
1009 r00 = _mm_mul_pd(rsq00,rinv00);
1010
1011 /* Compute parameters for interactions between i and j atoms */
1012 gmx_mm_load_1pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,&c6_00,&c12_00);
1013
1014 /* Calculate table index by multiplying r with table scale and truncate to integer */
1015 rt = _mm_mul_pd(r00,vftabscale);
1016 vfitab = _mm_cvttpd_epi32(rt);
1017 vfeps = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
1018 vfitab = _mm_slli_epi32(vfitab,3);
1019
1020 /* CUBIC SPLINE TABLE DISPERSION */
1021 Y = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
1022 F = _mm_setzero_pd();
1023 GMX_MM_TRANSPOSE2_PD(Y,F);
1024 G = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
1025 H = _mm_setzero_pd();
1026 GMX_MM_TRANSPOSE2_PD(G,H);
1027 Heps = _mm_mul_pd(vfeps,H);
1028 Fp = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
1029 FF = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
1030 fvdw6 = _mm_mul_pd(c6_00,FF);
1031
1032 /* CUBIC SPLINE TABLE REPULSION */
1033 vfitab = _mm_add_epi32(vfitab,ifour);
1034 Y = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
1035 F = _mm_setzero_pd();
1036 GMX_MM_TRANSPOSE2_PD(Y,F);
1037 G = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
1038 H = _mm_setzero_pd();
1039 GMX_MM_TRANSPOSE2_PD(G,H);
1040 Heps = _mm_mul_pd(vfeps,H);
1041 Fp = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
1042 FF = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
1043 fvdw12 = _mm_mul_pd(c12_00,FF);
1044 fvdw = _mm_xor_pd(signbit,_mm_mul_pd(_mm_add_pd(fvdw6,fvdw12),_mm_mul_pd(vftabscale,rinv00)));
1045
1046 fscal = fvdw;
1047
1048 fscal = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1049
1050 /* Calculate temporary vectorial force */
1051 tx = _mm_mul_pd(fscal,dx00);
1052 ty = _mm_mul_pd(fscal,dy00);
1053 tz = _mm_mul_pd(fscal,dz00);
1054
1055 /* Update vectorial force */
1056 fix0 = _mm_add_pd(fix0,tx);
1057 fiy0 = _mm_add_pd(fiy0,ty);
1058 fiz0 = _mm_add_pd(fiz0,tz);
1059
1060 fjx0 = _mm_add_pd(fjx0,tx);
1061 fjy0 = _mm_add_pd(fjy0,ty);
1062 fjz0 = _mm_add_pd(fjz0,tz);
1063
1064 /**************************
1065 * CALCULATE INTERACTIONS *
1066 **************************/
1067
1068 /* Compute parameters for interactions between i and j atoms */
1069 qq10 = _mm_mul_pd(iq1,jq0);
1070
1071 /* COULOMB ELECTROSTATICS */
1072 velec = _mm_mul_pd(qq10,rinv10);
1073 felec = _mm_mul_pd(velec,rinvsq10);
1074
1075 fscal = felec;
1076
1077 fscal = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1078
1079 /* Calculate temporary vectorial force */
1080 tx = _mm_mul_pd(fscal,dx10);
1081 ty = _mm_mul_pd(fscal,dy10);
1082 tz = _mm_mul_pd(fscal,dz10);
1083
1084 /* Update vectorial force */
1085 fix1 = _mm_add_pd(fix1,tx);
1086 fiy1 = _mm_add_pd(fiy1,ty);
1087 fiz1 = _mm_add_pd(fiz1,tz);
1088
1089 fjx0 = _mm_add_pd(fjx0,tx);
1090 fjy0 = _mm_add_pd(fjy0,ty);
1091 fjz0 = _mm_add_pd(fjz0,tz);
1092
1093 /**************************
1094 * CALCULATE INTERACTIONS *
1095 **************************/
1096
1097 /* Compute parameters for interactions between i and j atoms */
1098 qq20 = _mm_mul_pd(iq2,jq0);
1099
1100 /* COULOMB ELECTROSTATICS */
1101 velec = _mm_mul_pd(qq20,rinv20);
1102 felec = _mm_mul_pd(velec,rinvsq20);
1103
1104 fscal = felec;
1105
1106 fscal = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1107
1108 /* Calculate temporary vectorial force */
1109 tx = _mm_mul_pd(fscal,dx20);
1110 ty = _mm_mul_pd(fscal,dy20);
1111 tz = _mm_mul_pd(fscal,dz20);
1112
1113 /* Update vectorial force */
1114 fix2 = _mm_add_pd(fix2,tx);
1115 fiy2 = _mm_add_pd(fiy2,ty);
1116 fiz2 = _mm_add_pd(fiz2,tz);
1117
1118 fjx0 = _mm_add_pd(fjx0,tx);
1119 fjy0 = _mm_add_pd(fjy0,ty);
1120 fjz0 = _mm_add_pd(fjz0,tz);
1121
1122 /**************************
1123 * CALCULATE INTERACTIONS *
1124 **************************/
1125
1126 /* Compute parameters for interactions between i and j atoms */
1127 qq30 = _mm_mul_pd(iq3,jq0);
1128
1129 /* COULOMB ELECTROSTATICS */
1130 velec = _mm_mul_pd(qq30,rinv30);
1131 felec = _mm_mul_pd(velec,rinvsq30);
1132
1133 fscal = felec;
1134
1135 fscal = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1136
1137 /* Calculate temporary vectorial force */
1138 tx = _mm_mul_pd(fscal,dx30);
1139 ty = _mm_mul_pd(fscal,dy30);
1140 tz = _mm_mul_pd(fscal,dz30);
1141
1142 /* Update vectorial force */
1143 fix3 = _mm_add_pd(fix3,tx);
1144 fiy3 = _mm_add_pd(fiy3,ty);
1145 fiz3 = _mm_add_pd(fiz3,tz);
1146
1147 fjx0 = _mm_add_pd(fjx0,tx);
1148 fjy0 = _mm_add_pd(fjy0,ty);
1149 fjz0 = _mm_add_pd(fjz0,tz);
1150
1151 gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,fjx0,fjy0,fjz0);
1152
1153 /* Inner loop uses 132 flops */
1154 }
1155
1156 /* End of innermost loop */
1157
1158 gmx_mm_update_iforce_4atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
1159 f+i_coord_offset,fshift+i_shift_offset);
1160
1161 /* Increment number of inner iterations */
1162 inneriter += j_index_end - j_index_start;
1163
1164 /* Outer loop uses 24 flops */
1165 }
1166
1167 /* Increment number of outer iterations */
1168 outeriter += nri;
1169
1170 /* Update outer/inner flops */
1171
1172 inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W4_F,outeriter*24 + inneriter*132);
1173 }
The ultimate molecular dynamics simulation package