?  search:
summary | log | graphiclog1 | graphiclog2 | commit | commitdiff | tree | refs | edit | fork
blame | history | raw | HEAD
Double precision SSE2 kernels
[gromacs.git] / src / gmxlib / nonbonded / nb_kernel_sse2_double / nb_kernel_ElecRFCut_VdwNone_GeomW4P1_sse2_double.c
blob7198549eae17b3b612709e82fb2a3f972e81e8e9
1 /*
2 * Note: this file was generated by the Gromacs sse2_double kernel generator.
3 *
4 * This source code is part of
5 *
6 * G R O M A C S
7 *
8 * Copyright (c) 2001-2012, The GROMACS Development Team
9 *
10 * Gromacs is a library for molecular simulation and trajectory analysis,
11 * written by Erik Lindahl, David van der Spoel, Berk Hess, and others - for
12 * a full list of developers and information, check out http://www.gromacs.org
13 *
14 * This program is free software; you can redistribute it and/or modify it under
15 * the terms of the GNU Lesser General Public License as published by the Free
16 * Software Foundation; either version 2 of the License, or (at your option) any
17 * later version.
18 *
19 * To help fund GROMACS development, we humbly ask that you cite
20 * the papers people have written on it - you can find them on the website.
21 */
22 #ifdef HAVE_CONFIG_H
23 #include <config.h>
24 #endif
25
26 #include <math.h>
27
28 #include "../nb_kernel.h"
29 #include "types/simple.h"
30 #include "vec.h"
31 #include "nrnb.h"
32
33 #include "gmx_math_x86_sse2_double.h"
34 #include "kernelutil_x86_sse2_double.h"
35
36 /*
37 * Gromacs nonbonded kernel: nb_kernel_ElecRFCut_VdwNone_GeomW4P1_VF_sse2_double
38 * Electrostatics interaction: ReactionField
39 * VdW interaction: None
40 * Geometry: Water4-Particle
41 * Calculate force/pot: PotentialAndForce
42 */
43 void
44 nb_kernel_ElecRFCut_VdwNone_GeomW4P1_VF_sse2_double
45 (t_nblist * gmx_restrict nlist,
46 rvec * gmx_restrict xx,
47 rvec * gmx_restrict ff,
48 t_forcerec * gmx_restrict fr,
49 t_mdatoms * gmx_restrict mdatoms,
50 nb_kernel_data_t * gmx_restrict kernel_data,
51 t_nrnb * gmx_restrict nrnb)
52 {
53 /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
54 * just 0 for non-waters.
55 * Suffixes A,B refer to j loop unrolling done with SSE double precision, e.g. for the two different
56 * jnr indices corresponding to data put in the four positions in the SIMD register.
57 */
58 int i_shift_offset,i_coord_offset,outeriter,inneriter;
59 int j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
60 int jnrA,jnrB;
61 int j_coord_offsetA,j_coord_offsetB;
62 int *iinr,*jindex,*jjnr,*shiftidx,*gid;
63 real rcutoff_scalar;
64 real *shiftvec,*fshift,*x,*f;
65 __m128d tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
66 int vdwioffset1;
67 __m128d ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
68 int vdwioffset2;
69 __m128d ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
70 int vdwioffset3;
71 __m128d ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
72 int vdwjidx0A,vdwjidx0B;
73 __m128d jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
74 __m128d dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
75 __m128d dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
76 __m128d dx30,dy30,dz30,rsq30,rinv30,rinvsq30,r30,qq30,c6_30,c12_30;
77 __m128d velec,felec,velecsum,facel,crf,krf,krf2;
78 real *charge;
79 __m128d dummy_mask,cutoff_mask;
80 __m128d signbit = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
81 __m128d one = _mm_set1_pd(1.0);
82 __m128d two = _mm_set1_pd(2.0);
83 x = xx[0];
84 f = ff[0];
85
86 nri = nlist->nri;
87 iinr = nlist->iinr;
88 jindex = nlist->jindex;
89 jjnr = nlist->jjnr;
90 shiftidx = nlist->shift;
91 gid = nlist->gid;
92 shiftvec = fr->shift_vec[0];
93 fshift = fr->fshift[0];
94 facel = _mm_set1_pd(fr->epsfac);
95 charge = mdatoms->chargeA;
96 krf = _mm_set1_pd(fr->ic->k_rf);
97 krf2 = _mm_set1_pd(fr->ic->k_rf*2.0);
98 crf = _mm_set1_pd(fr->ic->c_rf);
99
100 /* Setup water-specific parameters */
101 inr = nlist->iinr[0];
102 iq1 = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+1]));
103 iq2 = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+2]));
104 iq3 = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+3]));
105
106 /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
107 rcutoff_scalar = fr->rcoulomb;
108 rcutoff = _mm_set1_pd(rcutoff_scalar);
109 rcutoff2 = _mm_mul_pd(rcutoff,rcutoff);
110
111 /* Avoid stupid compiler warnings */
112 jnrA = jnrB = 0;
113 j_coord_offsetA = 0;
114 j_coord_offsetB = 0;
115
116 outeriter = 0;
117 inneriter = 0;
118
119 /* Start outer loop over neighborlists */
120 for(iidx=0; iidx<nri; iidx++)
121 {
122 /* Load shift vector for this list */
123 i_shift_offset = DIM*shiftidx[iidx];
124
125 /* Load limits for loop over neighbors */
126 j_index_start = jindex[iidx];
127 j_index_end = jindex[iidx+1];
128
129 /* Get outer coordinate index */
130 inr = iinr[iidx];
131 i_coord_offset = DIM*inr;
132
133 /* Load i particle coords and add shift vector */
134 gmx_mm_load_shift_and_3rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset+DIM,
135 &ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
136
137 fix1 = _mm_setzero_pd();
138 fiy1 = _mm_setzero_pd();
139 fiz1 = _mm_setzero_pd();
140 fix2 = _mm_setzero_pd();
141 fiy2 = _mm_setzero_pd();
142 fiz2 = _mm_setzero_pd();
143 fix3 = _mm_setzero_pd();
144 fiy3 = _mm_setzero_pd();
145 fiz3 = _mm_setzero_pd();
146
147 /* Reset potential sums */
148 velecsum = _mm_setzero_pd();
149
150 /* Start inner kernel loop */
151 for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
152 {
153
154 /* Get j neighbor index, and coordinate index */
155 jnrA = jjnr[jidx];
156 jnrB = jjnr[jidx+1];
157 j_coord_offsetA = DIM*jnrA;
158 j_coord_offsetB = DIM*jnrB;
159
160 /* load j atom coordinates */
161 gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
162 &jx0,&jy0,&jz0);
163
164 /* Calculate displacement vector */
165 dx10 = _mm_sub_pd(ix1,jx0);
166 dy10 = _mm_sub_pd(iy1,jy0);
167 dz10 = _mm_sub_pd(iz1,jz0);
168 dx20 = _mm_sub_pd(ix2,jx0);
169 dy20 = _mm_sub_pd(iy2,jy0);
170 dz20 = _mm_sub_pd(iz2,jz0);
171 dx30 = _mm_sub_pd(ix3,jx0);
172 dy30 = _mm_sub_pd(iy3,jy0);
173 dz30 = _mm_sub_pd(iz3,jz0);
174
175 /* Calculate squared distance and things based on it */
176 rsq10 = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
177 rsq20 = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
178 rsq30 = gmx_mm_calc_rsq_pd(dx30,dy30,dz30);
179
180 rinv10 = gmx_mm_invsqrt_pd(rsq10);
181 rinv20 = gmx_mm_invsqrt_pd(rsq20);
182 rinv30 = gmx_mm_invsqrt_pd(rsq30);
183
184 rinvsq10 = _mm_mul_pd(rinv10,rinv10);
185 rinvsq20 = _mm_mul_pd(rinv20,rinv20);
186 rinvsq30 = _mm_mul_pd(rinv30,rinv30);
187
188 /* Load parameters for j particles */
189 jq0 = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
190
191 fjx0 = _mm_setzero_pd();
192 fjy0 = _mm_setzero_pd();
193 fjz0 = _mm_setzero_pd();
194
195 /**************************
196 * CALCULATE INTERACTIONS *
197 **************************/
198
199 if (gmx_mm_any_lt(rsq10,rcutoff2))
200 {
201
202 /* Compute parameters for interactions between i and j atoms */
203 qq10 = _mm_mul_pd(iq1,jq0);
204
205 /* REACTION-FIELD ELECTROSTATICS */
206 velec = _mm_mul_pd(qq10,_mm_sub_pd(_mm_add_pd(rinv10,_mm_mul_pd(krf,rsq10)),crf));
207 felec = _mm_mul_pd(qq10,_mm_sub_pd(_mm_mul_pd(rinv10,rinvsq10),krf2));
208
209 cutoff_mask = _mm_cmplt_pd(rsq10,rcutoff2);
210
211 /* Update potential sum for this i atom from the interaction with this j atom. */
212 velec = _mm_and_pd(velec,cutoff_mask);
213 velecsum = _mm_add_pd(velecsum,velec);
214
215 fscal = felec;
216
217 fscal = _mm_and_pd(fscal,cutoff_mask);
218
219 /* Calculate temporary vectorial force */
220 tx = _mm_mul_pd(fscal,dx10);
221 ty = _mm_mul_pd(fscal,dy10);
222 tz = _mm_mul_pd(fscal,dz10);
223
224 /* Update vectorial force */
225 fix1 = _mm_add_pd(fix1,tx);
226 fiy1 = _mm_add_pd(fiy1,ty);
227 fiz1 = _mm_add_pd(fiz1,tz);
228
229 fjx0 = _mm_add_pd(fjx0,tx);
230 fjy0 = _mm_add_pd(fjy0,ty);
231 fjz0 = _mm_add_pd(fjz0,tz);
232
233 }
234
235 /**************************
236 * CALCULATE INTERACTIONS *
237 **************************/
238
239 if (gmx_mm_any_lt(rsq20,rcutoff2))
240 {
241
242 /* Compute parameters for interactions between i and j atoms */
243 qq20 = _mm_mul_pd(iq2,jq0);
244
245 /* REACTION-FIELD ELECTROSTATICS */
246 velec = _mm_mul_pd(qq20,_mm_sub_pd(_mm_add_pd(rinv20,_mm_mul_pd(krf,rsq20)),crf));
247 felec = _mm_mul_pd(qq20,_mm_sub_pd(_mm_mul_pd(rinv20,rinvsq20),krf2));
248
249 cutoff_mask = _mm_cmplt_pd(rsq20,rcutoff2);
250
251 /* Update potential sum for this i atom from the interaction with this j atom. */
252 velec = _mm_and_pd(velec,cutoff_mask);
253 velecsum = _mm_add_pd(velecsum,velec);
254
255 fscal = felec;
256
257 fscal = _mm_and_pd(fscal,cutoff_mask);
258
259 /* Calculate temporary vectorial force */
260 tx = _mm_mul_pd(fscal,dx20);
261 ty = _mm_mul_pd(fscal,dy20);
262 tz = _mm_mul_pd(fscal,dz20);
263
264 /* Update vectorial force */
265 fix2 = _mm_add_pd(fix2,tx);
266 fiy2 = _mm_add_pd(fiy2,ty);
267 fiz2 = _mm_add_pd(fiz2,tz);
268
269 fjx0 = _mm_add_pd(fjx0,tx);
270 fjy0 = _mm_add_pd(fjy0,ty);
271 fjz0 = _mm_add_pd(fjz0,tz);
272
273 }
274
275 /**************************
276 * CALCULATE INTERACTIONS *
277 **************************/
278
279 if (gmx_mm_any_lt(rsq30,rcutoff2))
280 {
281
282 /* Compute parameters for interactions between i and j atoms */
283 qq30 = _mm_mul_pd(iq3,jq0);
284
285 /* REACTION-FIELD ELECTROSTATICS */
286 velec = _mm_mul_pd(qq30,_mm_sub_pd(_mm_add_pd(rinv30,_mm_mul_pd(krf,rsq30)),crf));
287 felec = _mm_mul_pd(qq30,_mm_sub_pd(_mm_mul_pd(rinv30,rinvsq30),krf2));
288
289 cutoff_mask = _mm_cmplt_pd(rsq30,rcutoff2);
290
291 /* Update potential sum for this i atom from the interaction with this j atom. */
292 velec = _mm_and_pd(velec,cutoff_mask);
293 velecsum = _mm_add_pd(velecsum,velec);
294
295 fscal = felec;
296
297 fscal = _mm_and_pd(fscal,cutoff_mask);
298
299 /* Calculate temporary vectorial force */
300 tx = _mm_mul_pd(fscal,dx30);
301 ty = _mm_mul_pd(fscal,dy30);
302 tz = _mm_mul_pd(fscal,dz30);
303
304 /* Update vectorial force */
305 fix3 = _mm_add_pd(fix3,tx);
306 fiy3 = _mm_add_pd(fiy3,ty);
307 fiz3 = _mm_add_pd(fiz3,tz);
308
309 fjx0 = _mm_add_pd(fjx0,tx);
310 fjy0 = _mm_add_pd(fjy0,ty);
311 fjz0 = _mm_add_pd(fjz0,tz);
312
313 }
314
315 gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,fjx0,fjy0,fjz0);
316
317 /* Inner loop uses 111 flops */
318 }
319
320 if(jidx<j_index_end)
321 {
322
323 jnrA = jjnr[jidx];
324 j_coord_offsetA = DIM*jnrA;
325
326 /* load j atom coordinates */
327 gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
328 &jx0,&jy0,&jz0);
329
330 /* Calculate displacement vector */
331 dx10 = _mm_sub_pd(ix1,jx0);
332 dy10 = _mm_sub_pd(iy1,jy0);
333 dz10 = _mm_sub_pd(iz1,jz0);
334 dx20 = _mm_sub_pd(ix2,jx0);
335 dy20 = _mm_sub_pd(iy2,jy0);
336 dz20 = _mm_sub_pd(iz2,jz0);
337 dx30 = _mm_sub_pd(ix3,jx0);
338 dy30 = _mm_sub_pd(iy3,jy0);
339 dz30 = _mm_sub_pd(iz3,jz0);
340
341 /* Calculate squared distance and things based on it */
342 rsq10 = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
343 rsq20 = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
344 rsq30 = gmx_mm_calc_rsq_pd(dx30,dy30,dz30);
345
346 rinv10 = gmx_mm_invsqrt_pd(rsq10);
347 rinv20 = gmx_mm_invsqrt_pd(rsq20);
348 rinv30 = gmx_mm_invsqrt_pd(rsq30);
349
350 rinvsq10 = _mm_mul_pd(rinv10,rinv10);
351 rinvsq20 = _mm_mul_pd(rinv20,rinv20);
352 rinvsq30 = _mm_mul_pd(rinv30,rinv30);
353
354 /* Load parameters for j particles */
355 jq0 = _mm_load_sd(charge+jnrA+0);
356
357 fjx0 = _mm_setzero_pd();
358 fjy0 = _mm_setzero_pd();
359 fjz0 = _mm_setzero_pd();
360
361 /**************************
362 * CALCULATE INTERACTIONS *
363 **************************/
364
365 if (gmx_mm_any_lt(rsq10,rcutoff2))
366 {
367
368 /* Compute parameters for interactions between i and j atoms */
369 qq10 = _mm_mul_pd(iq1,jq0);
370
371 /* REACTION-FIELD ELECTROSTATICS */
372 velec = _mm_mul_pd(qq10,_mm_sub_pd(_mm_add_pd(rinv10,_mm_mul_pd(krf,rsq10)),crf));
373 felec = _mm_mul_pd(qq10,_mm_sub_pd(_mm_mul_pd(rinv10,rinvsq10),krf2));
374
375 cutoff_mask = _mm_cmplt_pd(rsq10,rcutoff2);
376
377 /* Update potential sum for this i atom from the interaction with this j atom. */
378 velec = _mm_and_pd(velec,cutoff_mask);
379 velec = _mm_unpacklo_pd(velec,_mm_setzero_pd());
380 velecsum = _mm_add_pd(velecsum,velec);
381
382 fscal = felec;
383
384 fscal = _mm_and_pd(fscal,cutoff_mask);
385
386 fscal = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
387
388 /* Calculate temporary vectorial force */
389 tx = _mm_mul_pd(fscal,dx10);
390 ty = _mm_mul_pd(fscal,dy10);
391 tz = _mm_mul_pd(fscal,dz10);
392
393 /* Update vectorial force */
394 fix1 = _mm_add_pd(fix1,tx);
395 fiy1 = _mm_add_pd(fiy1,ty);
396 fiz1 = _mm_add_pd(fiz1,tz);
397
398 fjx0 = _mm_add_pd(fjx0,tx);
399 fjy0 = _mm_add_pd(fjy0,ty);
400 fjz0 = _mm_add_pd(fjz0,tz);
401
402 }
403
404 /**************************
405 * CALCULATE INTERACTIONS *
406 **************************/
407
408 if (gmx_mm_any_lt(rsq20,rcutoff2))
409 {
410
411 /* Compute parameters for interactions between i and j atoms */
412 qq20 = _mm_mul_pd(iq2,jq0);
413
414 /* REACTION-FIELD ELECTROSTATICS */
415 velec = _mm_mul_pd(qq20,_mm_sub_pd(_mm_add_pd(rinv20,_mm_mul_pd(krf,rsq20)),crf));
416 felec = _mm_mul_pd(qq20,_mm_sub_pd(_mm_mul_pd(rinv20,rinvsq20),krf2));
417
418 cutoff_mask = _mm_cmplt_pd(rsq20,rcutoff2);
419
420 /* Update potential sum for this i atom from the interaction with this j atom. */
421 velec = _mm_and_pd(velec,cutoff_mask);
422 velec = _mm_unpacklo_pd(velec,_mm_setzero_pd());
423 velecsum = _mm_add_pd(velecsum,velec);
424
425 fscal = felec;
426
427 fscal = _mm_and_pd(fscal,cutoff_mask);
428
429 fscal = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
430
431 /* Calculate temporary vectorial force */
432 tx = _mm_mul_pd(fscal,dx20);
433 ty = _mm_mul_pd(fscal,dy20);
434 tz = _mm_mul_pd(fscal,dz20);
435
436 /* Update vectorial force */
437 fix2 = _mm_add_pd(fix2,tx);
438 fiy2 = _mm_add_pd(fiy2,ty);
439 fiz2 = _mm_add_pd(fiz2,tz);
440
441 fjx0 = _mm_add_pd(fjx0,tx);
442 fjy0 = _mm_add_pd(fjy0,ty);
443 fjz0 = _mm_add_pd(fjz0,tz);
444
445 }
446
447 /**************************
448 * CALCULATE INTERACTIONS *
449 **************************/
450
451 if (gmx_mm_any_lt(rsq30,rcutoff2))
452 {
453
454 /* Compute parameters for interactions between i and j atoms */
455 qq30 = _mm_mul_pd(iq3,jq0);
456
457 /* REACTION-FIELD ELECTROSTATICS */
458 velec = _mm_mul_pd(qq30,_mm_sub_pd(_mm_add_pd(rinv30,_mm_mul_pd(krf,rsq30)),crf));
459 felec = _mm_mul_pd(qq30,_mm_sub_pd(_mm_mul_pd(rinv30,rinvsq30),krf2));
460
461 cutoff_mask = _mm_cmplt_pd(rsq30,rcutoff2);
462
463 /* Update potential sum for this i atom from the interaction with this j atom. */
464 velec = _mm_and_pd(velec,cutoff_mask);
465 velec = _mm_unpacklo_pd(velec,_mm_setzero_pd());
466 velecsum = _mm_add_pd(velecsum,velec);
467
468 fscal = felec;
469
470 fscal = _mm_and_pd(fscal,cutoff_mask);
471
472 fscal = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
473
474 /* Calculate temporary vectorial force */
475 tx = _mm_mul_pd(fscal,dx30);
476 ty = _mm_mul_pd(fscal,dy30);
477 tz = _mm_mul_pd(fscal,dz30);
478
479 /* Update vectorial force */
480 fix3 = _mm_add_pd(fix3,tx);
481 fiy3 = _mm_add_pd(fiy3,ty);
482 fiz3 = _mm_add_pd(fiz3,tz);
483
484 fjx0 = _mm_add_pd(fjx0,tx);
485 fjy0 = _mm_add_pd(fjy0,ty);
486 fjz0 = _mm_add_pd(fjz0,tz);
487
488 }
489
490 gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,fjx0,fjy0,fjz0);
491
492 /* Inner loop uses 111 flops */
493 }
494
495 /* End of innermost loop */
496
497 gmx_mm_update_iforce_3atom_swizzle_pd(fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
498 f+i_coord_offset+DIM,fshift+i_shift_offset);
499
500 ggid = gid[iidx];
501 /* Update potential energies */
502 gmx_mm_update_1pot_pd(velecsum,kernel_data->energygrp_elec+ggid);
503
504 /* Increment number of inner iterations */
505 inneriter += j_index_end - j_index_start;
506
507 /* Outer loop uses 19 flops */
508 }
509
510 /* Increment number of outer iterations */
511 outeriter += nri;
512
513 /* Update outer/inner flops */
514
515 inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_W4_VF,outeriter*19 + inneriter*111);
516 }
517 /*
518 * Gromacs nonbonded kernel: nb_kernel_ElecRFCut_VdwNone_GeomW4P1_F_sse2_double
519 * Electrostatics interaction: ReactionField
520 * VdW interaction: None
521 * Geometry: Water4-Particle
522 * Calculate force/pot: Force
523 */
524 void
525 nb_kernel_ElecRFCut_VdwNone_GeomW4P1_F_sse2_double
526 (t_nblist * gmx_restrict nlist,
527 rvec * gmx_restrict xx,
528 rvec * gmx_restrict ff,
529 t_forcerec * gmx_restrict fr,
530 t_mdatoms * gmx_restrict mdatoms,
531 nb_kernel_data_t * gmx_restrict kernel_data,
532 t_nrnb * gmx_restrict nrnb)
533 {
534 /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
535 * just 0 for non-waters.
536 * Suffixes A,B refer to j loop unrolling done with SSE double precision, e.g. for the two different
537 * jnr indices corresponding to data put in the four positions in the SIMD register.
538 */
539 int i_shift_offset,i_coord_offset,outeriter,inneriter;
540 int j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
541 int jnrA,jnrB;
542 int j_coord_offsetA,j_coord_offsetB;
543 int *iinr,*jindex,*jjnr,*shiftidx,*gid;
544 real rcutoff_scalar;
545 real *shiftvec,*fshift,*x,*f;
546 __m128d tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
547 int vdwioffset1;
548 __m128d ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
549 int vdwioffset2;
550 __m128d ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
551 int vdwioffset3;
552 __m128d ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
553 int vdwjidx0A,vdwjidx0B;
554 __m128d jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
555 __m128d dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
556 __m128d dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
557 __m128d dx30,dy30,dz30,rsq30,rinv30,rinvsq30,r30,qq30,c6_30,c12_30;
558 __m128d velec,felec,velecsum,facel,crf,krf,krf2;
559 real *charge;
560 __m128d dummy_mask,cutoff_mask;
561 __m128d signbit = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
562 __m128d one = _mm_set1_pd(1.0);
563 __m128d two = _mm_set1_pd(2.0);
564 x = xx[0];
565 f = ff[0];
566
567 nri = nlist->nri;
568 iinr = nlist->iinr;
569 jindex = nlist->jindex;
570 jjnr = nlist->jjnr;
571 shiftidx = nlist->shift;
572 gid = nlist->gid;
573 shiftvec = fr->shift_vec[0];
574 fshift = fr->fshift[0];
575 facel = _mm_set1_pd(fr->epsfac);
576 charge = mdatoms->chargeA;
577 krf = _mm_set1_pd(fr->ic->k_rf);
578 krf2 = _mm_set1_pd(fr->ic->k_rf*2.0);
579 crf = _mm_set1_pd(fr->ic->c_rf);
580
581 /* Setup water-specific parameters */
582 inr = nlist->iinr[0];
583 iq1 = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+1]));
584 iq2 = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+2]));
585 iq3 = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+3]));
586
587 /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
588 rcutoff_scalar = fr->rcoulomb;
589 rcutoff = _mm_set1_pd(rcutoff_scalar);
590 rcutoff2 = _mm_mul_pd(rcutoff,rcutoff);
591
592 /* Avoid stupid compiler warnings */
593 jnrA = jnrB = 0;
594 j_coord_offsetA = 0;
595 j_coord_offsetB = 0;
596
597 outeriter = 0;
598 inneriter = 0;
599
600 /* Start outer loop over neighborlists */
601 for(iidx=0; iidx<nri; iidx++)
602 {
603 /* Load shift vector for this list */
604 i_shift_offset = DIM*shiftidx[iidx];
605
606 /* Load limits for loop over neighbors */
607 j_index_start = jindex[iidx];
608 j_index_end = jindex[iidx+1];
609
610 /* Get outer coordinate index */
611 inr = iinr[iidx];
612 i_coord_offset = DIM*inr;
613
614 /* Load i particle coords and add shift vector */
615 gmx_mm_load_shift_and_3rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset+DIM,
616 &ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
617
618 fix1 = _mm_setzero_pd();
619 fiy1 = _mm_setzero_pd();
620 fiz1 = _mm_setzero_pd();
621 fix2 = _mm_setzero_pd();
622 fiy2 = _mm_setzero_pd();
623 fiz2 = _mm_setzero_pd();
624 fix3 = _mm_setzero_pd();
625 fiy3 = _mm_setzero_pd();
626 fiz3 = _mm_setzero_pd();
627
628 /* Start inner kernel loop */
629 for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
630 {
631
632 /* Get j neighbor index, and coordinate index */
633 jnrA = jjnr[jidx];
634 jnrB = jjnr[jidx+1];
635 j_coord_offsetA = DIM*jnrA;
636 j_coord_offsetB = DIM*jnrB;
637
638 /* load j atom coordinates */
639 gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
640 &jx0,&jy0,&jz0);
641
642 /* Calculate displacement vector */
643 dx10 = _mm_sub_pd(ix1,jx0);
644 dy10 = _mm_sub_pd(iy1,jy0);
645 dz10 = _mm_sub_pd(iz1,jz0);
646 dx20 = _mm_sub_pd(ix2,jx0);
647 dy20 = _mm_sub_pd(iy2,jy0);
648 dz20 = _mm_sub_pd(iz2,jz0);
649 dx30 = _mm_sub_pd(ix3,jx0);
650 dy30 = _mm_sub_pd(iy3,jy0);
651 dz30 = _mm_sub_pd(iz3,jz0);
652
653 /* Calculate squared distance and things based on it */
654 rsq10 = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
655 rsq20 = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
656 rsq30 = gmx_mm_calc_rsq_pd(dx30,dy30,dz30);
657
658 rinv10 = gmx_mm_invsqrt_pd(rsq10);
659 rinv20 = gmx_mm_invsqrt_pd(rsq20);
660 rinv30 = gmx_mm_invsqrt_pd(rsq30);
661
662 rinvsq10 = _mm_mul_pd(rinv10,rinv10);
663 rinvsq20 = _mm_mul_pd(rinv20,rinv20);
664 rinvsq30 = _mm_mul_pd(rinv30,rinv30);
665
666 /* Load parameters for j particles */
667 jq0 = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
668
669 fjx0 = _mm_setzero_pd();
670 fjy0 = _mm_setzero_pd();
671 fjz0 = _mm_setzero_pd();
672
673 /**************************
674 * CALCULATE INTERACTIONS *
675 **************************/
676
677 if (gmx_mm_any_lt(rsq10,rcutoff2))
678 {
679
680 /* Compute parameters for interactions between i and j atoms */
681 qq10 = _mm_mul_pd(iq1,jq0);
682
683 /* REACTION-FIELD ELECTROSTATICS */
684 felec = _mm_mul_pd(qq10,_mm_sub_pd(_mm_mul_pd(rinv10,rinvsq10),krf2));
685
686 cutoff_mask = _mm_cmplt_pd(rsq10,rcutoff2);
687
688 fscal = felec;
689
690 fscal = _mm_and_pd(fscal,cutoff_mask);
691
692 /* Calculate temporary vectorial force */
693 tx = _mm_mul_pd(fscal,dx10);
694 ty = _mm_mul_pd(fscal,dy10);
695 tz = _mm_mul_pd(fscal,dz10);
696
697 /* Update vectorial force */
698 fix1 = _mm_add_pd(fix1,tx);
699 fiy1 = _mm_add_pd(fiy1,ty);
700 fiz1 = _mm_add_pd(fiz1,tz);
701
702 fjx0 = _mm_add_pd(fjx0,tx);
703 fjy0 = _mm_add_pd(fjy0,ty);
704 fjz0 = _mm_add_pd(fjz0,tz);
705
706 }
707
708 /**************************
709 * CALCULATE INTERACTIONS *
710 **************************/
711
712 if (gmx_mm_any_lt(rsq20,rcutoff2))
713 {
714
715 /* Compute parameters for interactions between i and j atoms */
716 qq20 = _mm_mul_pd(iq2,jq0);
717
718 /* REACTION-FIELD ELECTROSTATICS */
719 felec = _mm_mul_pd(qq20,_mm_sub_pd(_mm_mul_pd(rinv20,rinvsq20),krf2));
720
721 cutoff_mask = _mm_cmplt_pd(rsq20,rcutoff2);
722
723 fscal = felec;
724
725 fscal = _mm_and_pd(fscal,cutoff_mask);
726
727 /* Calculate temporary vectorial force */
728 tx = _mm_mul_pd(fscal,dx20);
729 ty = _mm_mul_pd(fscal,dy20);
730 tz = _mm_mul_pd(fscal,dz20);
731
732 /* Update vectorial force */
733 fix2 = _mm_add_pd(fix2,tx);
734 fiy2 = _mm_add_pd(fiy2,ty);
735 fiz2 = _mm_add_pd(fiz2,tz);
736
737 fjx0 = _mm_add_pd(fjx0,tx);
738 fjy0 = _mm_add_pd(fjy0,ty);
739 fjz0 = _mm_add_pd(fjz0,tz);
740
741 }
742
743 /**************************
744 * CALCULATE INTERACTIONS *
745 **************************/
746
747 if (gmx_mm_any_lt(rsq30,rcutoff2))
748 {
749
750 /* Compute parameters for interactions between i and j atoms */
751 qq30 = _mm_mul_pd(iq3,jq0);
752
753 /* REACTION-FIELD ELECTROSTATICS */
754 felec = _mm_mul_pd(qq30,_mm_sub_pd(_mm_mul_pd(rinv30,rinvsq30),krf2));
755
756 cutoff_mask = _mm_cmplt_pd(rsq30,rcutoff2);
757
758 fscal = felec;
759
760 fscal = _mm_and_pd(fscal,cutoff_mask);
761
762 /* Calculate temporary vectorial force */
763 tx = _mm_mul_pd(fscal,dx30);
764 ty = _mm_mul_pd(fscal,dy30);
765 tz = _mm_mul_pd(fscal,dz30);
766
767 /* Update vectorial force */
768 fix3 = _mm_add_pd(fix3,tx);
769 fiy3 = _mm_add_pd(fiy3,ty);
770 fiz3 = _mm_add_pd(fiz3,tz);
771
772 fjx0 = _mm_add_pd(fjx0,tx);
773 fjy0 = _mm_add_pd(fjy0,ty);
774 fjz0 = _mm_add_pd(fjz0,tz);
775
776 }
777
778 gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,fjx0,fjy0,fjz0);
779
780 /* Inner loop uses 93 flops */
781 }
782
783 if(jidx<j_index_end)
784 {
785
786 jnrA = jjnr[jidx];
787 j_coord_offsetA = DIM*jnrA;
788
789 /* load j atom coordinates */
790 gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
791 &jx0,&jy0,&jz0);
792
793 /* Calculate displacement vector */
794 dx10 = _mm_sub_pd(ix1,jx0);
795 dy10 = _mm_sub_pd(iy1,jy0);
796 dz10 = _mm_sub_pd(iz1,jz0);
797 dx20 = _mm_sub_pd(ix2,jx0);
798 dy20 = _mm_sub_pd(iy2,jy0);
799 dz20 = _mm_sub_pd(iz2,jz0);
800 dx30 = _mm_sub_pd(ix3,jx0);
801 dy30 = _mm_sub_pd(iy3,jy0);
802 dz30 = _mm_sub_pd(iz3,jz0);
803
804 /* Calculate squared distance and things based on it */
805 rsq10 = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
806 rsq20 = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
807 rsq30 = gmx_mm_calc_rsq_pd(dx30,dy30,dz30);
808
809 rinv10 = gmx_mm_invsqrt_pd(rsq10);
810 rinv20 = gmx_mm_invsqrt_pd(rsq20);
811 rinv30 = gmx_mm_invsqrt_pd(rsq30);
812
813 rinvsq10 = _mm_mul_pd(rinv10,rinv10);
814 rinvsq20 = _mm_mul_pd(rinv20,rinv20);
815 rinvsq30 = _mm_mul_pd(rinv30,rinv30);
816
817 /* Load parameters for j particles */
818 jq0 = _mm_load_sd(charge+jnrA+0);
819
820 fjx0 = _mm_setzero_pd();
821 fjy0 = _mm_setzero_pd();
822 fjz0 = _mm_setzero_pd();
823
824 /**************************
825 * CALCULATE INTERACTIONS *
826 **************************/
827
828 if (gmx_mm_any_lt(rsq10,rcutoff2))
829 {
830
831 /* Compute parameters for interactions between i and j atoms */
832 qq10 = _mm_mul_pd(iq1,jq0);
833
834 /* REACTION-FIELD ELECTROSTATICS */
835 felec = _mm_mul_pd(qq10,_mm_sub_pd(_mm_mul_pd(rinv10,rinvsq10),krf2));
836
837 cutoff_mask = _mm_cmplt_pd(rsq10,rcutoff2);
838
839 fscal = felec;
840
841 fscal = _mm_and_pd(fscal,cutoff_mask);
842
843 fscal = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
844
845 /* Calculate temporary vectorial force */
846 tx = _mm_mul_pd(fscal,dx10);
847 ty = _mm_mul_pd(fscal,dy10);
848 tz = _mm_mul_pd(fscal,dz10);
849
850 /* Update vectorial force */
851 fix1 = _mm_add_pd(fix1,tx);
852 fiy1 = _mm_add_pd(fiy1,ty);
853 fiz1 = _mm_add_pd(fiz1,tz);
854
855 fjx0 = _mm_add_pd(fjx0,tx);
856 fjy0 = _mm_add_pd(fjy0,ty);
857 fjz0 = _mm_add_pd(fjz0,tz);
858
859 }
860
861 /**************************
862 * CALCULATE INTERACTIONS *
863 **************************/
864
865 if (gmx_mm_any_lt(rsq20,rcutoff2))
866 {
867
868 /* Compute parameters for interactions between i and j atoms */
869 qq20 = _mm_mul_pd(iq2,jq0);
870
871 /* REACTION-FIELD ELECTROSTATICS */
872 felec = _mm_mul_pd(qq20,_mm_sub_pd(_mm_mul_pd(rinv20,rinvsq20),krf2));
873
874 cutoff_mask = _mm_cmplt_pd(rsq20,rcutoff2);
875
876 fscal = felec;
877
878 fscal = _mm_and_pd(fscal,cutoff_mask);
879
880 fscal = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
881
882 /* Calculate temporary vectorial force */
883 tx = _mm_mul_pd(fscal,dx20);
884 ty = _mm_mul_pd(fscal,dy20);
885 tz = _mm_mul_pd(fscal,dz20);
886
887 /* Update vectorial force */
888 fix2 = _mm_add_pd(fix2,tx);
889 fiy2 = _mm_add_pd(fiy2,ty);
890 fiz2 = _mm_add_pd(fiz2,tz);
891
892 fjx0 = _mm_add_pd(fjx0,tx);
893 fjy0 = _mm_add_pd(fjy0,ty);
894 fjz0 = _mm_add_pd(fjz0,tz);
895
896 }
897
898 /**************************
899 * CALCULATE INTERACTIONS *
900 **************************/
901
902 if (gmx_mm_any_lt(rsq30,rcutoff2))
903 {
904
905 /* Compute parameters for interactions between i and j atoms */
906 qq30 = _mm_mul_pd(iq3,jq0);
907
908 /* REACTION-FIELD ELECTROSTATICS */
909 felec = _mm_mul_pd(qq30,_mm_sub_pd(_mm_mul_pd(rinv30,rinvsq30),krf2));
910
911 cutoff_mask = _mm_cmplt_pd(rsq30,rcutoff2);
912
913 fscal = felec;
914
915 fscal = _mm_and_pd(fscal,cutoff_mask);
916
917 fscal = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
918
919 /* Calculate temporary vectorial force */
920 tx = _mm_mul_pd(fscal,dx30);
921 ty = _mm_mul_pd(fscal,dy30);
922 tz = _mm_mul_pd(fscal,dz30);
923
924 /* Update vectorial force */
925 fix3 = _mm_add_pd(fix3,tx);
926 fiy3 = _mm_add_pd(fiy3,ty);
927 fiz3 = _mm_add_pd(fiz3,tz);
928
929 fjx0 = _mm_add_pd(fjx0,tx);
930 fjy0 = _mm_add_pd(fjy0,ty);
931 fjz0 = _mm_add_pd(fjz0,tz);
932
933 }
934
935 gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,fjx0,fjy0,fjz0);
936
937 /* Inner loop uses 93 flops */
938 }
939
940 /* End of innermost loop */
941
942 gmx_mm_update_iforce_3atom_swizzle_pd(fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
943 f+i_coord_offset+DIM,fshift+i_shift_offset);
944
945 /* Increment number of inner iterations */
946 inneriter += j_index_end - j_index_start;
947
948 /* Outer loop uses 18 flops */
949 }
950
951 /* Increment number of outer iterations */
952 outeriter += nri;
953
954 /* Update outer/inner flops */
955
956 inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_W4_F,outeriter*18 + inneriter*93);
957 }
The ultimate molecular dynamics simulation package