?  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_ElecCoul_VdwLJ_GeomW3P1_sse2_double.c
blob24cbc0fcb5fae78a003d4973acbc781923d57890
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_ElecCoul_VdwLJ_GeomW3P1_VF_sse2_double
53 * Electrostatics interaction: Coulomb
54 * VdW interaction: LennardJones
55 * Geometry: Water3-Particle
56 * Calculate force/pot: PotentialAndForce
57 */
58 void
59 nb_kernel_ElecCoul_VdwLJ_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 dummy_mask,cutoff_mask;
101 __m128d signbit = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
102 __m128d one = _mm_set1_pd(1.0);
103 __m128d two = _mm_set1_pd(2.0);
104 x = xx[0];
105 f = ff[0];
106
107 nri = nlist->nri;
108 iinr = nlist->iinr;
109 jindex = nlist->jindex;
110 jjnr = nlist->jjnr;
111 shiftidx = nlist->shift;
112 gid = nlist->gid;
113 shiftvec = fr->shift_vec[0];
114 fshift = fr->fshift[0];
115 facel = _mm_set1_pd(fr->epsfac);
116 charge = mdatoms->chargeA;
117 nvdwtype = fr->ntype;
118 vdwparam = fr->nbfp;
119 vdwtype = mdatoms->typeA;
120
121 /* Setup water-specific parameters */
122 inr = nlist->iinr[0];
123 iq0 = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+0]));
124 iq1 = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+1]));
125 iq2 = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+2]));
126 vdwioffset0 = 2*nvdwtype*vdwtype[inr+0];
127
128 /* Avoid stupid compiler warnings */
129 jnrA = jnrB = 0;
130 j_coord_offsetA = 0;
131 j_coord_offsetB = 0;
132
133 outeriter = 0;
134 inneriter = 0;
135
136 /* Start outer loop over neighborlists */
137 for(iidx=0; iidx<nri; iidx++)
138 {
139 /* Load shift vector for this list */
140 i_shift_offset = DIM*shiftidx[iidx];
141
142 /* Load limits for loop over neighbors */
143 j_index_start = jindex[iidx];
144 j_index_end = jindex[iidx+1];
145
146 /* Get outer coordinate index */
147 inr = iinr[iidx];
148 i_coord_offset = DIM*inr;
149
150 /* Load i particle coords and add shift vector */
151 gmx_mm_load_shift_and_3rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
152 &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
153
154 fix0 = _mm_setzero_pd();
155 fiy0 = _mm_setzero_pd();
156 fiz0 = _mm_setzero_pd();
157 fix1 = _mm_setzero_pd();
158 fiy1 = _mm_setzero_pd();
159 fiz1 = _mm_setzero_pd();
160 fix2 = _mm_setzero_pd();
161 fiy2 = _mm_setzero_pd();
162 fiz2 = _mm_setzero_pd();
163
164 /* Reset potential sums */
165 velecsum = _mm_setzero_pd();
166 vvdwsum = _mm_setzero_pd();
167
168 /* Start inner kernel loop */
169 for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
170 {
171
172 /* Get j neighbor index, and coordinate index */
173 jnrA = jjnr[jidx];
174 jnrB = jjnr[jidx+1];
175 j_coord_offsetA = DIM*jnrA;
176 j_coord_offsetB = DIM*jnrB;
177
178 /* load j atom coordinates */
179 gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
180 &jx0,&jy0,&jz0);
181
182 /* Calculate displacement vector */
183 dx00 = _mm_sub_pd(ix0,jx0);
184 dy00 = _mm_sub_pd(iy0,jy0);
185 dz00 = _mm_sub_pd(iz0,jz0);
186 dx10 = _mm_sub_pd(ix1,jx0);
187 dy10 = _mm_sub_pd(iy1,jy0);
188 dz10 = _mm_sub_pd(iz1,jz0);
189 dx20 = _mm_sub_pd(ix2,jx0);
190 dy20 = _mm_sub_pd(iy2,jy0);
191 dz20 = _mm_sub_pd(iz2,jz0);
192
193 /* Calculate squared distance and things based on it */
194 rsq00 = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
195 rsq10 = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
196 rsq20 = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
197
198 rinv00 = gmx_mm_invsqrt_pd(rsq00);
199 rinv10 = gmx_mm_invsqrt_pd(rsq10);
200 rinv20 = gmx_mm_invsqrt_pd(rsq20);
201
202 rinvsq00 = _mm_mul_pd(rinv00,rinv00);
203 rinvsq10 = _mm_mul_pd(rinv10,rinv10);
204 rinvsq20 = _mm_mul_pd(rinv20,rinv20);
205
206 /* Load parameters for j particles */
207 jq0 = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
208 vdwjidx0A = 2*vdwtype[jnrA+0];
209 vdwjidx0B = 2*vdwtype[jnrB+0];
210
211 fjx0 = _mm_setzero_pd();
212 fjy0 = _mm_setzero_pd();
213 fjz0 = _mm_setzero_pd();
214
215 /**************************
216 * CALCULATE INTERACTIONS *
217 **************************/
218
219 /* Compute parameters for interactions between i and j atoms */
220 qq00 = _mm_mul_pd(iq0,jq0);
221 gmx_mm_load_2pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,
222 vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
223
224 /* COULOMB ELECTROSTATICS */
225 velec = _mm_mul_pd(qq00,rinv00);
226 felec = _mm_mul_pd(velec,rinvsq00);
227
228 /* LENNARD-JONES DISPERSION/REPULSION */
229
230 rinvsix = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
231 vvdw6 = _mm_mul_pd(c6_00,rinvsix);
232 vvdw12 = _mm_mul_pd(c12_00,_mm_mul_pd(rinvsix,rinvsix));
233 vvdw = _mm_sub_pd( _mm_mul_pd(vvdw12,one_twelfth) , _mm_mul_pd(vvdw6,one_sixth) );
234 fvdw = _mm_mul_pd(_mm_sub_pd(vvdw12,vvdw6),rinvsq00);
235
236 /* Update potential sum for this i atom from the interaction with this j atom. */
237 velecsum = _mm_add_pd(velecsum,velec);
238 vvdwsum = _mm_add_pd(vvdwsum,vvdw);
239
240 fscal = _mm_add_pd(felec,fvdw);
241
242 /* Calculate temporary vectorial force */
243 tx = _mm_mul_pd(fscal,dx00);
244 ty = _mm_mul_pd(fscal,dy00);
245 tz = _mm_mul_pd(fscal,dz00);
246
247 /* Update vectorial force */
248 fix0 = _mm_add_pd(fix0,tx);
249 fiy0 = _mm_add_pd(fiy0,ty);
250 fiz0 = _mm_add_pd(fiz0,tz);
251
252 fjx0 = _mm_add_pd(fjx0,tx);
253 fjy0 = _mm_add_pd(fjy0,ty);
254 fjz0 = _mm_add_pd(fjz0,tz);
255
256 /**************************
257 * CALCULATE INTERACTIONS *
258 **************************/
259
260 /* Compute parameters for interactions between i and j atoms */
261 qq10 = _mm_mul_pd(iq1,jq0);
262
263 /* COULOMB ELECTROSTATICS */
264 velec = _mm_mul_pd(qq10,rinv10);
265 felec = _mm_mul_pd(velec,rinvsq10);
266
267 /* Update potential sum for this i atom from the interaction with this j atom. */
268 velecsum = _mm_add_pd(velecsum,velec);
269
270 fscal = felec;
271
272 /* Calculate temporary vectorial force */
273 tx = _mm_mul_pd(fscal,dx10);
274 ty = _mm_mul_pd(fscal,dy10);
275 tz = _mm_mul_pd(fscal,dz10);
276
277 /* Update vectorial force */
278 fix1 = _mm_add_pd(fix1,tx);
279 fiy1 = _mm_add_pd(fiy1,ty);
280 fiz1 = _mm_add_pd(fiz1,tz);
281
282 fjx0 = _mm_add_pd(fjx0,tx);
283 fjy0 = _mm_add_pd(fjy0,ty);
284 fjz0 = _mm_add_pd(fjz0,tz);
285
286 /**************************
287 * CALCULATE INTERACTIONS *
288 **************************/
289
290 /* Compute parameters for interactions between i and j atoms */
291 qq20 = _mm_mul_pd(iq2,jq0);
292
293 /* COULOMB ELECTROSTATICS */
294 velec = _mm_mul_pd(qq20,rinv20);
295 felec = _mm_mul_pd(velec,rinvsq20);
296
297 /* Update potential sum for this i atom from the interaction with this j atom. */
298 velecsum = _mm_add_pd(velecsum,velec);
299
300 fscal = felec;
301
302 /* Calculate temporary vectorial force */
303 tx = _mm_mul_pd(fscal,dx20);
304 ty = _mm_mul_pd(fscal,dy20);
305 tz = _mm_mul_pd(fscal,dz20);
306
307 /* Update vectorial force */
308 fix2 = _mm_add_pd(fix2,tx);
309 fiy2 = _mm_add_pd(fiy2,ty);
310 fiz2 = _mm_add_pd(fiz2,tz);
311
312 fjx0 = _mm_add_pd(fjx0,tx);
313 fjy0 = _mm_add_pd(fjy0,ty);
314 fjz0 = _mm_add_pd(fjz0,tz);
315
316 gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,fjx0,fjy0,fjz0);
317
318 /* Inner loop uses 99 flops */
319 }
320
321 if(jidx<j_index_end)
322 {
323
324 jnrA = jjnr[jidx];
325 j_coord_offsetA = DIM*jnrA;
326
327 /* load j atom coordinates */
328 gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
329 &jx0,&jy0,&jz0);
330
331 /* Calculate displacement vector */
332 dx00 = _mm_sub_pd(ix0,jx0);
333 dy00 = _mm_sub_pd(iy0,jy0);
334 dz00 = _mm_sub_pd(iz0,jz0);
335 dx10 = _mm_sub_pd(ix1,jx0);
336 dy10 = _mm_sub_pd(iy1,jy0);
337 dz10 = _mm_sub_pd(iz1,jz0);
338 dx20 = _mm_sub_pd(ix2,jx0);
339 dy20 = _mm_sub_pd(iy2,jy0);
340 dz20 = _mm_sub_pd(iz2,jz0);
341
342 /* Calculate squared distance and things based on it */
343 rsq00 = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
344 rsq10 = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
345 rsq20 = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
346
347 rinv00 = gmx_mm_invsqrt_pd(rsq00);
348 rinv10 = gmx_mm_invsqrt_pd(rsq10);
349 rinv20 = gmx_mm_invsqrt_pd(rsq20);
350
351 rinvsq00 = _mm_mul_pd(rinv00,rinv00);
352 rinvsq10 = _mm_mul_pd(rinv10,rinv10);
353 rinvsq20 = _mm_mul_pd(rinv20,rinv20);
354
355 /* Load parameters for j particles */
356 jq0 = _mm_load_sd(charge+jnrA+0);
357 vdwjidx0A = 2*vdwtype[jnrA+0];
358
359 fjx0 = _mm_setzero_pd();
360 fjy0 = _mm_setzero_pd();
361 fjz0 = _mm_setzero_pd();
362
363 /**************************
364 * CALCULATE INTERACTIONS *
365 **************************/
366
367 /* Compute parameters for interactions between i and j atoms */
368 qq00 = _mm_mul_pd(iq0,jq0);
369 gmx_mm_load_1pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,&c6_00,&c12_00);
370
371 /* COULOMB ELECTROSTATICS */
372 velec = _mm_mul_pd(qq00,rinv00);
373 felec = _mm_mul_pd(velec,rinvsq00);
374
375 /* LENNARD-JONES DISPERSION/REPULSION */
376
377 rinvsix = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
378 vvdw6 = _mm_mul_pd(c6_00,rinvsix);
379 vvdw12 = _mm_mul_pd(c12_00,_mm_mul_pd(rinvsix,rinvsix));
380 vvdw = _mm_sub_pd( _mm_mul_pd(vvdw12,one_twelfth) , _mm_mul_pd(vvdw6,one_sixth) );
381 fvdw = _mm_mul_pd(_mm_sub_pd(vvdw12,vvdw6),rinvsq00);
382
383 /* Update potential sum for this i atom from the interaction with this j atom. */
384 velec = _mm_unpacklo_pd(velec,_mm_setzero_pd());
385 velecsum = _mm_add_pd(velecsum,velec);
386 vvdw = _mm_unpacklo_pd(vvdw,_mm_setzero_pd());
387 vvdwsum = _mm_add_pd(vvdwsum,vvdw);
388
389 fscal = _mm_add_pd(felec,fvdw);
390
391 fscal = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
392
393 /* Calculate temporary vectorial force */
394 tx = _mm_mul_pd(fscal,dx00);
395 ty = _mm_mul_pd(fscal,dy00);
396 tz = _mm_mul_pd(fscal,dz00);
397
398 /* Update vectorial force */
399 fix0 = _mm_add_pd(fix0,tx);
400 fiy0 = _mm_add_pd(fiy0,ty);
401 fiz0 = _mm_add_pd(fiz0,tz);
402
403 fjx0 = _mm_add_pd(fjx0,tx);
404 fjy0 = _mm_add_pd(fjy0,ty);
405 fjz0 = _mm_add_pd(fjz0,tz);
406
407 /**************************
408 * CALCULATE INTERACTIONS *
409 **************************/
410
411 /* Compute parameters for interactions between i and j atoms */
412 qq10 = _mm_mul_pd(iq1,jq0);
413
414 /* COULOMB ELECTROSTATICS */
415 velec = _mm_mul_pd(qq10,rinv10);
416 felec = _mm_mul_pd(velec,rinvsq10);
417
418 /* Update potential sum for this i atom from the interaction with this j atom. */
419 velec = _mm_unpacklo_pd(velec,_mm_setzero_pd());
420 velecsum = _mm_add_pd(velecsum,velec);
421
422 fscal = felec;
423
424 fscal = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
425
426 /* Calculate temporary vectorial force */
427 tx = _mm_mul_pd(fscal,dx10);
428 ty = _mm_mul_pd(fscal,dy10);
429 tz = _mm_mul_pd(fscal,dz10);
430
431 /* Update vectorial force */
432 fix1 = _mm_add_pd(fix1,tx);
433 fiy1 = _mm_add_pd(fiy1,ty);
434 fiz1 = _mm_add_pd(fiz1,tz);
435
436 fjx0 = _mm_add_pd(fjx0,tx);
437 fjy0 = _mm_add_pd(fjy0,ty);
438 fjz0 = _mm_add_pd(fjz0,tz);
439
440 /**************************
441 * CALCULATE INTERACTIONS *
442 **************************/
443
444 /* Compute parameters for interactions between i and j atoms */
445 qq20 = _mm_mul_pd(iq2,jq0);
446
447 /* COULOMB ELECTROSTATICS */
448 velec = _mm_mul_pd(qq20,rinv20);
449 felec = _mm_mul_pd(velec,rinvsq20);
450
451 /* Update potential sum for this i atom from the interaction with this j atom. */
452 velec = _mm_unpacklo_pd(velec,_mm_setzero_pd());
453 velecsum = _mm_add_pd(velecsum,velec);
454
455 fscal = felec;
456
457 fscal = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
458
459 /* Calculate temporary vectorial force */
460 tx = _mm_mul_pd(fscal,dx20);
461 ty = _mm_mul_pd(fscal,dy20);
462 tz = _mm_mul_pd(fscal,dz20);
463
464 /* Update vectorial force */
465 fix2 = _mm_add_pd(fix2,tx);
466 fiy2 = _mm_add_pd(fiy2,ty);
467 fiz2 = _mm_add_pd(fiz2,tz);
468
469 fjx0 = _mm_add_pd(fjx0,tx);
470 fjy0 = _mm_add_pd(fjy0,ty);
471 fjz0 = _mm_add_pd(fjz0,tz);
472
473 gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,fjx0,fjy0,fjz0);
474
475 /* Inner loop uses 99 flops */
476 }
477
478 /* End of innermost loop */
479
480 gmx_mm_update_iforce_3atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
481 f+i_coord_offset,fshift+i_shift_offset);
482
483 ggid = gid[iidx];
484 /* Update potential energies */
485 gmx_mm_update_1pot_pd(velecsum,kernel_data->energygrp_elec+ggid);
486 gmx_mm_update_1pot_pd(vvdwsum,kernel_data->energygrp_vdw+ggid);
487
488 /* Increment number of inner iterations */
489 inneriter += j_index_end - j_index_start;
490
491 /* Outer loop uses 20 flops */
492 }
493
494 /* Increment number of outer iterations */
495 outeriter += nri;
496
497 /* Update outer/inner flops */
498
499 inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3_VF,outeriter*20 + inneriter*99);
500 }
501 /*
502 * Gromacs nonbonded kernel: nb_kernel_ElecCoul_VdwLJ_GeomW3P1_F_sse2_double
503 * Electrostatics interaction: Coulomb
504 * VdW interaction: LennardJones
505 * Geometry: Water3-Particle
506 * Calculate force/pot: Force
507 */
508 void
509 nb_kernel_ElecCoul_VdwLJ_GeomW3P1_F_sse2_double
510 (t_nblist * gmx_restrict nlist,
511 rvec * gmx_restrict xx,
512 rvec * gmx_restrict ff,
513 t_forcerec * gmx_restrict fr,
514 t_mdatoms * gmx_restrict mdatoms,
515 nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
516 t_nrnb * gmx_restrict nrnb)
517 {
518 /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
519 * just 0 for non-waters.
520 * Suffixes A,B refer to j loop unrolling done with SSE double precision, e.g. for the two different
521 * jnr indices corresponding to data put in the four positions in the SIMD register.
522 */
523 int i_shift_offset,i_coord_offset,outeriter,inneriter;
524 int j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
525 int jnrA,jnrB;
526 int j_coord_offsetA,j_coord_offsetB;
527 int *iinr,*jindex,*jjnr,*shiftidx,*gid;
528 real rcutoff_scalar;
529 real *shiftvec,*fshift,*x,*f;
530 __m128d tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
531 int vdwioffset0;
532 __m128d ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
533 int vdwioffset1;
534 __m128d ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
535 int vdwioffset2;
536 __m128d ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
537 int vdwjidx0A,vdwjidx0B;
538 __m128d jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
539 __m128d dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
540 __m128d dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
541 __m128d dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
542 __m128d velec,felec,velecsum,facel,crf,krf,krf2;
543 real *charge;
544 int nvdwtype;
545 __m128d rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
546 int *vdwtype;
547 real *vdwparam;
548 __m128d one_sixth = _mm_set1_pd(1.0/6.0);
549 __m128d one_twelfth = _mm_set1_pd(1.0/12.0);
550 __m128d dummy_mask,cutoff_mask;
551 __m128d signbit = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
552 __m128d one = _mm_set1_pd(1.0);
553 __m128d two = _mm_set1_pd(2.0);
554 x = xx[0];
555 f = ff[0];
556
557 nri = nlist->nri;
558 iinr = nlist->iinr;
559 jindex = nlist->jindex;
560 jjnr = nlist->jjnr;
561 shiftidx = nlist->shift;
562 gid = nlist->gid;
563 shiftvec = fr->shift_vec[0];
564 fshift = fr->fshift[0];
565 facel = _mm_set1_pd(fr->epsfac);
566 charge = mdatoms->chargeA;
567 nvdwtype = fr->ntype;
568 vdwparam = fr->nbfp;
569 vdwtype = mdatoms->typeA;
570
571 /* Setup water-specific parameters */
572 inr = nlist->iinr[0];
573 iq0 = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+0]));
574 iq1 = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+1]));
575 iq2 = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+2]));
576 vdwioffset0 = 2*nvdwtype*vdwtype[inr+0];
577
578 /* Avoid stupid compiler warnings */
579 jnrA = jnrB = 0;
580 j_coord_offsetA = 0;
581 j_coord_offsetB = 0;
582
583 outeriter = 0;
584 inneriter = 0;
585
586 /* Start outer loop over neighborlists */
587 for(iidx=0; iidx<nri; iidx++)
588 {
589 /* Load shift vector for this list */
590 i_shift_offset = DIM*shiftidx[iidx];
591
592 /* Load limits for loop over neighbors */
593 j_index_start = jindex[iidx];
594 j_index_end = jindex[iidx+1];
595
596 /* Get outer coordinate index */
597 inr = iinr[iidx];
598 i_coord_offset = DIM*inr;
599
600 /* Load i particle coords and add shift vector */
601 gmx_mm_load_shift_and_3rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
602 &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
603
604 fix0 = _mm_setzero_pd();
605 fiy0 = _mm_setzero_pd();
606 fiz0 = _mm_setzero_pd();
607 fix1 = _mm_setzero_pd();
608 fiy1 = _mm_setzero_pd();
609 fiz1 = _mm_setzero_pd();
610 fix2 = _mm_setzero_pd();
611 fiy2 = _mm_setzero_pd();
612 fiz2 = _mm_setzero_pd();
613
614 /* Start inner kernel loop */
615 for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
616 {
617
618 /* Get j neighbor index, and coordinate index */
619 jnrA = jjnr[jidx];
620 jnrB = jjnr[jidx+1];
621 j_coord_offsetA = DIM*jnrA;
622 j_coord_offsetB = DIM*jnrB;
623
624 /* load j atom coordinates */
625 gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
626 &jx0,&jy0,&jz0);
627
628 /* Calculate displacement vector */
629 dx00 = _mm_sub_pd(ix0,jx0);
630 dy00 = _mm_sub_pd(iy0,jy0);
631 dz00 = _mm_sub_pd(iz0,jz0);
632 dx10 = _mm_sub_pd(ix1,jx0);
633 dy10 = _mm_sub_pd(iy1,jy0);
634 dz10 = _mm_sub_pd(iz1,jz0);
635 dx20 = _mm_sub_pd(ix2,jx0);
636 dy20 = _mm_sub_pd(iy2,jy0);
637 dz20 = _mm_sub_pd(iz2,jz0);
638
639 /* Calculate squared distance and things based on it */
640 rsq00 = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
641 rsq10 = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
642 rsq20 = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
643
644 rinv00 = gmx_mm_invsqrt_pd(rsq00);
645 rinv10 = gmx_mm_invsqrt_pd(rsq10);
646 rinv20 = gmx_mm_invsqrt_pd(rsq20);
647
648 rinvsq00 = _mm_mul_pd(rinv00,rinv00);
649 rinvsq10 = _mm_mul_pd(rinv10,rinv10);
650 rinvsq20 = _mm_mul_pd(rinv20,rinv20);
651
652 /* Load parameters for j particles */
653 jq0 = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
654 vdwjidx0A = 2*vdwtype[jnrA+0];
655 vdwjidx0B = 2*vdwtype[jnrB+0];
656
657 fjx0 = _mm_setzero_pd();
658 fjy0 = _mm_setzero_pd();
659 fjz0 = _mm_setzero_pd();
660
661 /**************************
662 * CALCULATE INTERACTIONS *
663 **************************/
664
665 /* Compute parameters for interactions between i and j atoms */
666 qq00 = _mm_mul_pd(iq0,jq0);
667 gmx_mm_load_2pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,
668 vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
669
670 /* COULOMB ELECTROSTATICS */
671 velec = _mm_mul_pd(qq00,rinv00);
672 felec = _mm_mul_pd(velec,rinvsq00);
673
674 /* LENNARD-JONES DISPERSION/REPULSION */
675
676 rinvsix = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
677 fvdw = _mm_mul_pd(_mm_sub_pd(_mm_mul_pd(c12_00,rinvsix),c6_00),_mm_mul_pd(rinvsix,rinvsq00));
678
679 fscal = _mm_add_pd(felec,fvdw);
680
681 /* Calculate temporary vectorial force */
682 tx = _mm_mul_pd(fscal,dx00);
683 ty = _mm_mul_pd(fscal,dy00);
684 tz = _mm_mul_pd(fscal,dz00);
685
686 /* Update vectorial force */
687 fix0 = _mm_add_pd(fix0,tx);
688 fiy0 = _mm_add_pd(fiy0,ty);
689 fiz0 = _mm_add_pd(fiz0,tz);
690
691 fjx0 = _mm_add_pd(fjx0,tx);
692 fjy0 = _mm_add_pd(fjy0,ty);
693 fjz0 = _mm_add_pd(fjz0,tz);
694
695 /**************************
696 * CALCULATE INTERACTIONS *
697 **************************/
698
699 /* Compute parameters for interactions between i and j atoms */
700 qq10 = _mm_mul_pd(iq1,jq0);
701
702 /* COULOMB ELECTROSTATICS */
703 velec = _mm_mul_pd(qq10,rinv10);
704 felec = _mm_mul_pd(velec,rinvsq10);
705
706 fscal = felec;
707
708 /* Calculate temporary vectorial force */
709 tx = _mm_mul_pd(fscal,dx10);
710 ty = _mm_mul_pd(fscal,dy10);
711 tz = _mm_mul_pd(fscal,dz10);
712
713 /* Update vectorial force */
714 fix1 = _mm_add_pd(fix1,tx);
715 fiy1 = _mm_add_pd(fiy1,ty);
716 fiz1 = _mm_add_pd(fiz1,tz);
717
718 fjx0 = _mm_add_pd(fjx0,tx);
719 fjy0 = _mm_add_pd(fjy0,ty);
720 fjz0 = _mm_add_pd(fjz0,tz);
721
722 /**************************
723 * CALCULATE INTERACTIONS *
724 **************************/
725
726 /* Compute parameters for interactions between i and j atoms */
727 qq20 = _mm_mul_pd(iq2,jq0);
728
729 /* COULOMB ELECTROSTATICS */
730 velec = _mm_mul_pd(qq20,rinv20);
731 felec = _mm_mul_pd(velec,rinvsq20);
732
733 fscal = felec;
734
735 /* Calculate temporary vectorial force */
736 tx = _mm_mul_pd(fscal,dx20);
737 ty = _mm_mul_pd(fscal,dy20);
738 tz = _mm_mul_pd(fscal,dz20);
739
740 /* Update vectorial force */
741 fix2 = _mm_add_pd(fix2,tx);
742 fiy2 = _mm_add_pd(fiy2,ty);
743 fiz2 = _mm_add_pd(fiz2,tz);
744
745 fjx0 = _mm_add_pd(fjx0,tx);
746 fjy0 = _mm_add_pd(fjy0,ty);
747 fjz0 = _mm_add_pd(fjz0,tz);
748
749 gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,fjx0,fjy0,fjz0);
750
751 /* Inner loop uses 91 flops */
752 }
753
754 if(jidx<j_index_end)
755 {
756
757 jnrA = jjnr[jidx];
758 j_coord_offsetA = DIM*jnrA;
759
760 /* load j atom coordinates */
761 gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
762 &jx0,&jy0,&jz0);
763
764 /* Calculate displacement vector */
765 dx00 = _mm_sub_pd(ix0,jx0);
766 dy00 = _mm_sub_pd(iy0,jy0);
767 dz00 = _mm_sub_pd(iz0,jz0);
768 dx10 = _mm_sub_pd(ix1,jx0);
769 dy10 = _mm_sub_pd(iy1,jy0);
770 dz10 = _mm_sub_pd(iz1,jz0);
771 dx20 = _mm_sub_pd(ix2,jx0);
772 dy20 = _mm_sub_pd(iy2,jy0);
773 dz20 = _mm_sub_pd(iz2,jz0);
774
775 /* Calculate squared distance and things based on it */
776 rsq00 = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
777 rsq10 = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
778 rsq20 = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
779
780 rinv00 = gmx_mm_invsqrt_pd(rsq00);
781 rinv10 = gmx_mm_invsqrt_pd(rsq10);
782 rinv20 = gmx_mm_invsqrt_pd(rsq20);
783
784 rinvsq00 = _mm_mul_pd(rinv00,rinv00);
785 rinvsq10 = _mm_mul_pd(rinv10,rinv10);
786 rinvsq20 = _mm_mul_pd(rinv20,rinv20);
787
788 /* Load parameters for j particles */
789 jq0 = _mm_load_sd(charge+jnrA+0);
790 vdwjidx0A = 2*vdwtype[jnrA+0];
791
792 fjx0 = _mm_setzero_pd();
793 fjy0 = _mm_setzero_pd();
794 fjz0 = _mm_setzero_pd();
795
796 /**************************
797 * CALCULATE INTERACTIONS *
798 **************************/
799
800 /* Compute parameters for interactions between i and j atoms */
801 qq00 = _mm_mul_pd(iq0,jq0);
802 gmx_mm_load_1pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,&c6_00,&c12_00);
803
804 /* COULOMB ELECTROSTATICS */
805 velec = _mm_mul_pd(qq00,rinv00);
806 felec = _mm_mul_pd(velec,rinvsq00);
807
808 /* LENNARD-JONES DISPERSION/REPULSION */
809
810 rinvsix = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
811 fvdw = _mm_mul_pd(_mm_sub_pd(_mm_mul_pd(c12_00,rinvsix),c6_00),_mm_mul_pd(rinvsix,rinvsq00));
812
813 fscal = _mm_add_pd(felec,fvdw);
814
815 fscal = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
816
817 /* Calculate temporary vectorial force */
818 tx = _mm_mul_pd(fscal,dx00);
819 ty = _mm_mul_pd(fscal,dy00);
820 tz = _mm_mul_pd(fscal,dz00);
821
822 /* Update vectorial force */
823 fix0 = _mm_add_pd(fix0,tx);
824 fiy0 = _mm_add_pd(fiy0,ty);
825 fiz0 = _mm_add_pd(fiz0,tz);
826
827 fjx0 = _mm_add_pd(fjx0,tx);
828 fjy0 = _mm_add_pd(fjy0,ty);
829 fjz0 = _mm_add_pd(fjz0,tz);
830
831 /**************************
832 * CALCULATE INTERACTIONS *
833 **************************/
834
835 /* Compute parameters for interactions between i and j atoms */
836 qq10 = _mm_mul_pd(iq1,jq0);
837
838 /* COULOMB ELECTROSTATICS */
839 velec = _mm_mul_pd(qq10,rinv10);
840 felec = _mm_mul_pd(velec,rinvsq10);
841
842 fscal = felec;
843
844 fscal = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
845
846 /* Calculate temporary vectorial force */
847 tx = _mm_mul_pd(fscal,dx10);
848 ty = _mm_mul_pd(fscal,dy10);
849 tz = _mm_mul_pd(fscal,dz10);
850
851 /* Update vectorial force */
852 fix1 = _mm_add_pd(fix1,tx);
853 fiy1 = _mm_add_pd(fiy1,ty);
854 fiz1 = _mm_add_pd(fiz1,tz);
855
856 fjx0 = _mm_add_pd(fjx0,tx);
857 fjy0 = _mm_add_pd(fjy0,ty);
858 fjz0 = _mm_add_pd(fjz0,tz);
859
860 /**************************
861 * CALCULATE INTERACTIONS *
862 **************************/
863
864 /* Compute parameters for interactions between i and j atoms */
865 qq20 = _mm_mul_pd(iq2,jq0);
866
867 /* COULOMB ELECTROSTATICS */
868 velec = _mm_mul_pd(qq20,rinv20);
869 felec = _mm_mul_pd(velec,rinvsq20);
870
871 fscal = felec;
872
873 fscal = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
874
875 /* Calculate temporary vectorial force */
876 tx = _mm_mul_pd(fscal,dx20);
877 ty = _mm_mul_pd(fscal,dy20);
878 tz = _mm_mul_pd(fscal,dz20);
879
880 /* Update vectorial force */
881 fix2 = _mm_add_pd(fix2,tx);
882 fiy2 = _mm_add_pd(fiy2,ty);
883 fiz2 = _mm_add_pd(fiz2,tz);
884
885 fjx0 = _mm_add_pd(fjx0,tx);
886 fjy0 = _mm_add_pd(fjy0,ty);
887 fjz0 = _mm_add_pd(fjz0,tz);
888
889 gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,fjx0,fjy0,fjz0);
890
891 /* Inner loop uses 91 flops */
892 }
893
894 /* End of innermost loop */
895
896 gmx_mm_update_iforce_3atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
897 f+i_coord_offset,fshift+i_shift_offset);
898
899 /* Increment number of inner iterations */
900 inneriter += j_index_end - j_index_start;
901
902 /* Outer loop uses 18 flops */
903 }
904
905 /* Increment number of outer iterations */
906 outeriter += nri;
907
908 /* Update outer/inner flops */
909
910 inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3_F,outeriter*18 + inneriter*91);
911 }
The ultimate molecular dynamics simulation package