?  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_single / nb_kernel_ElecCoul_VdwNone_GeomW3P1_sse2_single.c
blob34c3f5c3c5c4a79c1adaa086ce50749743e477c7
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_single 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_single.h"
49 #include "kernelutil_x86_sse2_single.h"
50
51 /*
52 * Gromacs nonbonded kernel: nb_kernel_ElecCoul_VdwNone_GeomW3P1_VF_sse2_single
53 * Electrostatics interaction: Coulomb
54 * VdW interaction: None
55 * Geometry: Water3-Particle
56 * Calculate force/pot: PotentialAndForce
57 */
58 void
59 nb_kernel_ElecCoul_VdwNone_GeomW3P1_VF_sse2_single
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,C,D refer to j loop unrolling done with SSE, e.g. for the four 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,jnrC,jnrD;
76 int jnrlistA,jnrlistB,jnrlistC,jnrlistD;
77 int j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
78 int *iinr,*jindex,*jjnr,*shiftidx,*gid;
79 real rcutoff_scalar;
80 real *shiftvec,*fshift,*x,*f;
81 real *fjptrA,*fjptrB,*fjptrC,*fjptrD;
82 real scratch[4*DIM];
83 __m128 tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
84 int vdwioffset0;
85 __m128 ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
86 int vdwioffset1;
87 __m128 ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
88 int vdwioffset2;
89 __m128 ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
90 int vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
91 __m128 jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
92 __m128 dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
93 __m128 dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
94 __m128 dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
95 __m128 velec,felec,velecsum,facel,crf,krf,krf2;
96 real *charge;
97 __m128 dummy_mask,cutoff_mask;
98 __m128 signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
99 __m128 one = _mm_set1_ps(1.0);
100 __m128 two = _mm_set1_ps(2.0);
101 x = xx[0];
102 f = ff[0];
103
104 nri = nlist->nri;
105 iinr = nlist->iinr;
106 jindex = nlist->jindex;
107 jjnr = nlist->jjnr;
108 shiftidx = nlist->shift;
109 gid = nlist->gid;
110 shiftvec = fr->shift_vec[0];
111 fshift = fr->fshift[0];
112 facel = _mm_set1_ps(fr->epsfac);
113 charge = mdatoms->chargeA;
114
115 /* Setup water-specific parameters */
116 inr = nlist->iinr[0];
117 iq0 = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+0]));
118 iq1 = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+1]));
119 iq2 = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+2]));
120
121 /* Avoid stupid compiler warnings */
122 jnrA = jnrB = jnrC = jnrD = 0;
123 j_coord_offsetA = 0;
124 j_coord_offsetB = 0;
125 j_coord_offsetC = 0;
126 j_coord_offsetD = 0;
127
128 outeriter = 0;
129 inneriter = 0;
130
131 for(iidx=0;iidx<4*DIM;iidx++)
132 {
133 scratch[iidx] = 0.0;
134 }
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_ps(shiftvec+i_shift_offset,x+i_coord_offset,
152 &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
153
154 fix0 = _mm_setzero_ps();
155 fiy0 = _mm_setzero_ps();
156 fiz0 = _mm_setzero_ps();
157 fix1 = _mm_setzero_ps();
158 fiy1 = _mm_setzero_ps();
159 fiz1 = _mm_setzero_ps();
160 fix2 = _mm_setzero_ps();
161 fiy2 = _mm_setzero_ps();
162 fiz2 = _mm_setzero_ps();
163
164 /* Reset potential sums */
165 velecsum = _mm_setzero_ps();
166
167 /* Start inner kernel loop */
168 for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
169 {
170
171 /* Get j neighbor index, and coordinate index */
172 jnrA = jjnr[jidx];
173 jnrB = jjnr[jidx+1];
174 jnrC = jjnr[jidx+2];
175 jnrD = jjnr[jidx+3];
176 j_coord_offsetA = DIM*jnrA;
177 j_coord_offsetB = DIM*jnrB;
178 j_coord_offsetC = DIM*jnrC;
179 j_coord_offsetD = DIM*jnrD;
180
181 /* load j atom coordinates */
182 gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
183 x+j_coord_offsetC,x+j_coord_offsetD,
184 &jx0,&jy0,&jz0);
185
186 /* Calculate displacement vector */
187 dx00 = _mm_sub_ps(ix0,jx0);
188 dy00 = _mm_sub_ps(iy0,jy0);
189 dz00 = _mm_sub_ps(iz0,jz0);
190 dx10 = _mm_sub_ps(ix1,jx0);
191 dy10 = _mm_sub_ps(iy1,jy0);
192 dz10 = _mm_sub_ps(iz1,jz0);
193 dx20 = _mm_sub_ps(ix2,jx0);
194 dy20 = _mm_sub_ps(iy2,jy0);
195 dz20 = _mm_sub_ps(iz2,jz0);
196
197 /* Calculate squared distance and things based on it */
198 rsq00 = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
199 rsq10 = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
200 rsq20 = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
201
202 rinv00 = gmx_mm_invsqrt_ps(rsq00);
203 rinv10 = gmx_mm_invsqrt_ps(rsq10);
204 rinv20 = gmx_mm_invsqrt_ps(rsq20);
205
206 rinvsq00 = _mm_mul_ps(rinv00,rinv00);
207 rinvsq10 = _mm_mul_ps(rinv10,rinv10);
208 rinvsq20 = _mm_mul_ps(rinv20,rinv20);
209
210 /* Load parameters for j particles */
211 jq0 = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
212 charge+jnrC+0,charge+jnrD+0);
213
214 fjx0 = _mm_setzero_ps();
215 fjy0 = _mm_setzero_ps();
216 fjz0 = _mm_setzero_ps();
217
218 /**************************
219 * CALCULATE INTERACTIONS *
220 **************************/
221
222 /* Compute parameters for interactions between i and j atoms */
223 qq00 = _mm_mul_ps(iq0,jq0);
224
225 /* COULOMB ELECTROSTATICS */
226 velec = _mm_mul_ps(qq00,rinv00);
227 felec = _mm_mul_ps(velec,rinvsq00);
228
229 /* Update potential sum for this i atom from the interaction with this j atom. */
230 velecsum = _mm_add_ps(velecsum,velec);
231
232 fscal = felec;
233
234 /* Calculate temporary vectorial force */
235 tx = _mm_mul_ps(fscal,dx00);
236 ty = _mm_mul_ps(fscal,dy00);
237 tz = _mm_mul_ps(fscal,dz00);
238
239 /* Update vectorial force */
240 fix0 = _mm_add_ps(fix0,tx);
241 fiy0 = _mm_add_ps(fiy0,ty);
242 fiz0 = _mm_add_ps(fiz0,tz);
243
244 fjx0 = _mm_add_ps(fjx0,tx);
245 fjy0 = _mm_add_ps(fjy0,ty);
246 fjz0 = _mm_add_ps(fjz0,tz);
247
248 /**************************
249 * CALCULATE INTERACTIONS *
250 **************************/
251
252 /* Compute parameters for interactions between i and j atoms */
253 qq10 = _mm_mul_ps(iq1,jq0);
254
255 /* COULOMB ELECTROSTATICS */
256 velec = _mm_mul_ps(qq10,rinv10);
257 felec = _mm_mul_ps(velec,rinvsq10);
258
259 /* Update potential sum for this i atom from the interaction with this j atom. */
260 velecsum = _mm_add_ps(velecsum,velec);
261
262 fscal = felec;
263
264 /* Calculate temporary vectorial force */
265 tx = _mm_mul_ps(fscal,dx10);
266 ty = _mm_mul_ps(fscal,dy10);
267 tz = _mm_mul_ps(fscal,dz10);
268
269 /* Update vectorial force */
270 fix1 = _mm_add_ps(fix1,tx);
271 fiy1 = _mm_add_ps(fiy1,ty);
272 fiz1 = _mm_add_ps(fiz1,tz);
273
274 fjx0 = _mm_add_ps(fjx0,tx);
275 fjy0 = _mm_add_ps(fjy0,ty);
276 fjz0 = _mm_add_ps(fjz0,tz);
277
278 /**************************
279 * CALCULATE INTERACTIONS *
280 **************************/
281
282 /* Compute parameters for interactions between i and j atoms */
283 qq20 = _mm_mul_ps(iq2,jq0);
284
285 /* COULOMB ELECTROSTATICS */
286 velec = _mm_mul_ps(qq20,rinv20);
287 felec = _mm_mul_ps(velec,rinvsq20);
288
289 /* Update potential sum for this i atom from the interaction with this j atom. */
290 velecsum = _mm_add_ps(velecsum,velec);
291
292 fscal = felec;
293
294 /* Calculate temporary vectorial force */
295 tx = _mm_mul_ps(fscal,dx20);
296 ty = _mm_mul_ps(fscal,dy20);
297 tz = _mm_mul_ps(fscal,dz20);
298
299 /* Update vectorial force */
300 fix2 = _mm_add_ps(fix2,tx);
301 fiy2 = _mm_add_ps(fiy2,ty);
302 fiz2 = _mm_add_ps(fiz2,tz);
303
304 fjx0 = _mm_add_ps(fjx0,tx);
305 fjy0 = _mm_add_ps(fjy0,ty);
306 fjz0 = _mm_add_ps(fjz0,tz);
307
308 fjptrA = f+j_coord_offsetA;
309 fjptrB = f+j_coord_offsetB;
310 fjptrC = f+j_coord_offsetC;
311 fjptrD = f+j_coord_offsetD;
312
313 gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
314
315 /* Inner loop uses 84 flops */
316 }
317
318 if(jidx<j_index_end)
319 {
320
321 /* Get j neighbor index, and coordinate index */
322 jnrlistA = jjnr[jidx];
323 jnrlistB = jjnr[jidx+1];
324 jnrlistC = jjnr[jidx+2];
325 jnrlistD = jjnr[jidx+3];
326 /* Sign of each element will be negative for non-real atoms.
327 * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
328 * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
329 */
330 dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
331 jnrA = (jnrlistA>=0) ? jnrlistA : 0;
332 jnrB = (jnrlistB>=0) ? jnrlistB : 0;
333 jnrC = (jnrlistC>=0) ? jnrlistC : 0;
334 jnrD = (jnrlistD>=0) ? jnrlistD : 0;
335 j_coord_offsetA = DIM*jnrA;
336 j_coord_offsetB = DIM*jnrB;
337 j_coord_offsetC = DIM*jnrC;
338 j_coord_offsetD = DIM*jnrD;
339
340 /* load j atom coordinates */
341 gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
342 x+j_coord_offsetC,x+j_coord_offsetD,
343 &jx0,&jy0,&jz0);
344
345 /* Calculate displacement vector */
346 dx00 = _mm_sub_ps(ix0,jx0);
347 dy00 = _mm_sub_ps(iy0,jy0);
348 dz00 = _mm_sub_ps(iz0,jz0);
349 dx10 = _mm_sub_ps(ix1,jx0);
350 dy10 = _mm_sub_ps(iy1,jy0);
351 dz10 = _mm_sub_ps(iz1,jz0);
352 dx20 = _mm_sub_ps(ix2,jx0);
353 dy20 = _mm_sub_ps(iy2,jy0);
354 dz20 = _mm_sub_ps(iz2,jz0);
355
356 /* Calculate squared distance and things based on it */
357 rsq00 = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
358 rsq10 = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
359 rsq20 = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
360
361 rinv00 = gmx_mm_invsqrt_ps(rsq00);
362 rinv10 = gmx_mm_invsqrt_ps(rsq10);
363 rinv20 = gmx_mm_invsqrt_ps(rsq20);
364
365 rinvsq00 = _mm_mul_ps(rinv00,rinv00);
366 rinvsq10 = _mm_mul_ps(rinv10,rinv10);
367 rinvsq20 = _mm_mul_ps(rinv20,rinv20);
368
369 /* Load parameters for j particles */
370 jq0 = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
371 charge+jnrC+0,charge+jnrD+0);
372
373 fjx0 = _mm_setzero_ps();
374 fjy0 = _mm_setzero_ps();
375 fjz0 = _mm_setzero_ps();
376
377 /**************************
378 * CALCULATE INTERACTIONS *
379 **************************/
380
381 /* Compute parameters for interactions between i and j atoms */
382 qq00 = _mm_mul_ps(iq0,jq0);
383
384 /* COULOMB ELECTROSTATICS */
385 velec = _mm_mul_ps(qq00,rinv00);
386 felec = _mm_mul_ps(velec,rinvsq00);
387
388 /* Update potential sum for this i atom from the interaction with this j atom. */
389 velec = _mm_andnot_ps(dummy_mask,velec);
390 velecsum = _mm_add_ps(velecsum,velec);
391
392 fscal = felec;
393
394 fscal = _mm_andnot_ps(dummy_mask,fscal);
395
396 /* Calculate temporary vectorial force */
397 tx = _mm_mul_ps(fscal,dx00);
398 ty = _mm_mul_ps(fscal,dy00);
399 tz = _mm_mul_ps(fscal,dz00);
400
401 /* Update vectorial force */
402 fix0 = _mm_add_ps(fix0,tx);
403 fiy0 = _mm_add_ps(fiy0,ty);
404 fiz0 = _mm_add_ps(fiz0,tz);
405
406 fjx0 = _mm_add_ps(fjx0,tx);
407 fjy0 = _mm_add_ps(fjy0,ty);
408 fjz0 = _mm_add_ps(fjz0,tz);
409
410 /**************************
411 * CALCULATE INTERACTIONS *
412 **************************/
413
414 /* Compute parameters for interactions between i and j atoms */
415 qq10 = _mm_mul_ps(iq1,jq0);
416
417 /* COULOMB ELECTROSTATICS */
418 velec = _mm_mul_ps(qq10,rinv10);
419 felec = _mm_mul_ps(velec,rinvsq10);
420
421 /* Update potential sum for this i atom from the interaction with this j atom. */
422 velec = _mm_andnot_ps(dummy_mask,velec);
423 velecsum = _mm_add_ps(velecsum,velec);
424
425 fscal = felec;
426
427 fscal = _mm_andnot_ps(dummy_mask,fscal);
428
429 /* Calculate temporary vectorial force */
430 tx = _mm_mul_ps(fscal,dx10);
431 ty = _mm_mul_ps(fscal,dy10);
432 tz = _mm_mul_ps(fscal,dz10);
433
434 /* Update vectorial force */
435 fix1 = _mm_add_ps(fix1,tx);
436 fiy1 = _mm_add_ps(fiy1,ty);
437 fiz1 = _mm_add_ps(fiz1,tz);
438
439 fjx0 = _mm_add_ps(fjx0,tx);
440 fjy0 = _mm_add_ps(fjy0,ty);
441 fjz0 = _mm_add_ps(fjz0,tz);
442
443 /**************************
444 * CALCULATE INTERACTIONS *
445 **************************/
446
447 /* Compute parameters for interactions between i and j atoms */
448 qq20 = _mm_mul_ps(iq2,jq0);
449
450 /* COULOMB ELECTROSTATICS */
451 velec = _mm_mul_ps(qq20,rinv20);
452 felec = _mm_mul_ps(velec,rinvsq20);
453
454 /* Update potential sum for this i atom from the interaction with this j atom. */
455 velec = _mm_andnot_ps(dummy_mask,velec);
456 velecsum = _mm_add_ps(velecsum,velec);
457
458 fscal = felec;
459
460 fscal = _mm_andnot_ps(dummy_mask,fscal);
461
462 /* Calculate temporary vectorial force */
463 tx = _mm_mul_ps(fscal,dx20);
464 ty = _mm_mul_ps(fscal,dy20);
465 tz = _mm_mul_ps(fscal,dz20);
466
467 /* Update vectorial force */
468 fix2 = _mm_add_ps(fix2,tx);
469 fiy2 = _mm_add_ps(fiy2,ty);
470 fiz2 = _mm_add_ps(fiz2,tz);
471
472 fjx0 = _mm_add_ps(fjx0,tx);
473 fjy0 = _mm_add_ps(fjy0,ty);
474 fjz0 = _mm_add_ps(fjz0,tz);
475
476 fjptrA = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
477 fjptrB = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
478 fjptrC = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
479 fjptrD = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
480
481 gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
482
483 /* Inner loop uses 84 flops */
484 }
485
486 /* End of innermost loop */
487
488 gmx_mm_update_iforce_3atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
489 f+i_coord_offset,fshift+i_shift_offset);
490
491 ggid = gid[iidx];
492 /* Update potential energies */
493 gmx_mm_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid);
494
495 /* Increment number of inner iterations */
496 inneriter += j_index_end - j_index_start;
497
498 /* Outer loop uses 19 flops */
499 }
500
501 /* Increment number of outer iterations */
502 outeriter += nri;
503
504 /* Update outer/inner flops */
505
506 inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_W3_VF,outeriter*19 + inneriter*84);
507 }
508 /*
509 * Gromacs nonbonded kernel: nb_kernel_ElecCoul_VdwNone_GeomW3P1_F_sse2_single
510 * Electrostatics interaction: Coulomb
511 * VdW interaction: None
512 * Geometry: Water3-Particle
513 * Calculate force/pot: Force
514 */
515 void
516 nb_kernel_ElecCoul_VdwNone_GeomW3P1_F_sse2_single
517 (t_nblist * gmx_restrict nlist,
518 rvec * gmx_restrict xx,
519 rvec * gmx_restrict ff,
520 t_forcerec * gmx_restrict fr,
521 t_mdatoms * gmx_restrict mdatoms,
522 nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
523 t_nrnb * gmx_restrict nrnb)
524 {
525 /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
526 * just 0 for non-waters.
527 * Suffixes A,B,C,D refer to j loop unrolling done with SSE, e.g. for the four different
528 * jnr indices corresponding to data put in the four positions in the SIMD register.
529 */
530 int i_shift_offset,i_coord_offset,outeriter,inneriter;
531 int j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
532 int jnrA,jnrB,jnrC,jnrD;
533 int jnrlistA,jnrlistB,jnrlistC,jnrlistD;
534 int j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
535 int *iinr,*jindex,*jjnr,*shiftidx,*gid;
536 real rcutoff_scalar;
537 real *shiftvec,*fshift,*x,*f;
538 real *fjptrA,*fjptrB,*fjptrC,*fjptrD;
539 real scratch[4*DIM];
540 __m128 tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
541 int vdwioffset0;
542 __m128 ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
543 int vdwioffset1;
544 __m128 ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
545 int vdwioffset2;
546 __m128 ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
547 int vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
548 __m128 jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
549 __m128 dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
550 __m128 dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
551 __m128 dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
552 __m128 velec,felec,velecsum,facel,crf,krf,krf2;
553 real *charge;
554 __m128 dummy_mask,cutoff_mask;
555 __m128 signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
556 __m128 one = _mm_set1_ps(1.0);
557 __m128 two = _mm_set1_ps(2.0);
558 x = xx[0];
559 f = ff[0];
560
561 nri = nlist->nri;
562 iinr = nlist->iinr;
563 jindex = nlist->jindex;
564 jjnr = nlist->jjnr;
565 shiftidx = nlist->shift;
566 gid = nlist->gid;
567 shiftvec = fr->shift_vec[0];
568 fshift = fr->fshift[0];
569 facel = _mm_set1_ps(fr->epsfac);
570 charge = mdatoms->chargeA;
571
572 /* Setup water-specific parameters */
573 inr = nlist->iinr[0];
574 iq0 = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+0]));
575 iq1 = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+1]));
576 iq2 = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+2]));
577
578 /* Avoid stupid compiler warnings */
579 jnrA = jnrB = jnrC = jnrD = 0;
580 j_coord_offsetA = 0;
581 j_coord_offsetB = 0;
582 j_coord_offsetC = 0;
583 j_coord_offsetD = 0;
584
585 outeriter = 0;
586 inneriter = 0;
587
588 for(iidx=0;iidx<4*DIM;iidx++)
589 {
590 scratch[iidx] = 0.0;
591 }
592
593 /* Start outer loop over neighborlists */
594 for(iidx=0; iidx<nri; iidx++)
595 {
596 /* Load shift vector for this list */
597 i_shift_offset = DIM*shiftidx[iidx];
598
599 /* Load limits for loop over neighbors */
600 j_index_start = jindex[iidx];
601 j_index_end = jindex[iidx+1];
602
603 /* Get outer coordinate index */
604 inr = iinr[iidx];
605 i_coord_offset = DIM*inr;
606
607 /* Load i particle coords and add shift vector */
608 gmx_mm_load_shift_and_3rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
609 &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
610
611 fix0 = _mm_setzero_ps();
612 fiy0 = _mm_setzero_ps();
613 fiz0 = _mm_setzero_ps();
614 fix1 = _mm_setzero_ps();
615 fiy1 = _mm_setzero_ps();
616 fiz1 = _mm_setzero_ps();
617 fix2 = _mm_setzero_ps();
618 fiy2 = _mm_setzero_ps();
619 fiz2 = _mm_setzero_ps();
620
621 /* Start inner kernel loop */
622 for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
623 {
624
625 /* Get j neighbor index, and coordinate index */
626 jnrA = jjnr[jidx];
627 jnrB = jjnr[jidx+1];
628 jnrC = jjnr[jidx+2];
629 jnrD = jjnr[jidx+3];
630 j_coord_offsetA = DIM*jnrA;
631 j_coord_offsetB = DIM*jnrB;
632 j_coord_offsetC = DIM*jnrC;
633 j_coord_offsetD = DIM*jnrD;
634
635 /* load j atom coordinates */
636 gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
637 x+j_coord_offsetC,x+j_coord_offsetD,
638 &jx0,&jy0,&jz0);
639
640 /* Calculate displacement vector */
641 dx00 = _mm_sub_ps(ix0,jx0);
642 dy00 = _mm_sub_ps(iy0,jy0);
643 dz00 = _mm_sub_ps(iz0,jz0);
644 dx10 = _mm_sub_ps(ix1,jx0);
645 dy10 = _mm_sub_ps(iy1,jy0);
646 dz10 = _mm_sub_ps(iz1,jz0);
647 dx20 = _mm_sub_ps(ix2,jx0);
648 dy20 = _mm_sub_ps(iy2,jy0);
649 dz20 = _mm_sub_ps(iz2,jz0);
650
651 /* Calculate squared distance and things based on it */
652 rsq00 = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
653 rsq10 = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
654 rsq20 = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
655
656 rinv00 = gmx_mm_invsqrt_ps(rsq00);
657 rinv10 = gmx_mm_invsqrt_ps(rsq10);
658 rinv20 = gmx_mm_invsqrt_ps(rsq20);
659
660 rinvsq00 = _mm_mul_ps(rinv00,rinv00);
661 rinvsq10 = _mm_mul_ps(rinv10,rinv10);
662 rinvsq20 = _mm_mul_ps(rinv20,rinv20);
663
664 /* Load parameters for j particles */
665 jq0 = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
666 charge+jnrC+0,charge+jnrD+0);
667
668 fjx0 = _mm_setzero_ps();
669 fjy0 = _mm_setzero_ps();
670 fjz0 = _mm_setzero_ps();
671
672 /**************************
673 * CALCULATE INTERACTIONS *
674 **************************/
675
676 /* Compute parameters for interactions between i and j atoms */
677 qq00 = _mm_mul_ps(iq0,jq0);
678
679 /* COULOMB ELECTROSTATICS */
680 velec = _mm_mul_ps(qq00,rinv00);
681 felec = _mm_mul_ps(velec,rinvsq00);
682
683 fscal = felec;
684
685 /* Calculate temporary vectorial force */
686 tx = _mm_mul_ps(fscal,dx00);
687 ty = _mm_mul_ps(fscal,dy00);
688 tz = _mm_mul_ps(fscal,dz00);
689
690 /* Update vectorial force */
691 fix0 = _mm_add_ps(fix0,tx);
692 fiy0 = _mm_add_ps(fiy0,ty);
693 fiz0 = _mm_add_ps(fiz0,tz);
694
695 fjx0 = _mm_add_ps(fjx0,tx);
696 fjy0 = _mm_add_ps(fjy0,ty);
697 fjz0 = _mm_add_ps(fjz0,tz);
698
699 /**************************
700 * CALCULATE INTERACTIONS *
701 **************************/
702
703 /* Compute parameters for interactions between i and j atoms */
704 qq10 = _mm_mul_ps(iq1,jq0);
705
706 /* COULOMB ELECTROSTATICS */
707 velec = _mm_mul_ps(qq10,rinv10);
708 felec = _mm_mul_ps(velec,rinvsq10);
709
710 fscal = felec;
711
712 /* Calculate temporary vectorial force */
713 tx = _mm_mul_ps(fscal,dx10);
714 ty = _mm_mul_ps(fscal,dy10);
715 tz = _mm_mul_ps(fscal,dz10);
716
717 /* Update vectorial force */
718 fix1 = _mm_add_ps(fix1,tx);
719 fiy1 = _mm_add_ps(fiy1,ty);
720 fiz1 = _mm_add_ps(fiz1,tz);
721
722 fjx0 = _mm_add_ps(fjx0,tx);
723 fjy0 = _mm_add_ps(fjy0,ty);
724 fjz0 = _mm_add_ps(fjz0,tz);
725
726 /**************************
727 * CALCULATE INTERACTIONS *
728 **************************/
729
730 /* Compute parameters for interactions between i and j atoms */
731 qq20 = _mm_mul_ps(iq2,jq0);
732
733 /* COULOMB ELECTROSTATICS */
734 velec = _mm_mul_ps(qq20,rinv20);
735 felec = _mm_mul_ps(velec,rinvsq20);
736
737 fscal = felec;
738
739 /* Calculate temporary vectorial force */
740 tx = _mm_mul_ps(fscal,dx20);
741 ty = _mm_mul_ps(fscal,dy20);
742 tz = _mm_mul_ps(fscal,dz20);
743
744 /* Update vectorial force */
745 fix2 = _mm_add_ps(fix2,tx);
746 fiy2 = _mm_add_ps(fiy2,ty);
747 fiz2 = _mm_add_ps(fiz2,tz);
748
749 fjx0 = _mm_add_ps(fjx0,tx);
750 fjy0 = _mm_add_ps(fjy0,ty);
751 fjz0 = _mm_add_ps(fjz0,tz);
752
753 fjptrA = f+j_coord_offsetA;
754 fjptrB = f+j_coord_offsetB;
755 fjptrC = f+j_coord_offsetC;
756 fjptrD = f+j_coord_offsetD;
757
758 gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
759
760 /* Inner loop uses 81 flops */
761 }
762
763 if(jidx<j_index_end)
764 {
765
766 /* Get j neighbor index, and coordinate index */
767 jnrlistA = jjnr[jidx];
768 jnrlistB = jjnr[jidx+1];
769 jnrlistC = jjnr[jidx+2];
770 jnrlistD = jjnr[jidx+3];
771 /* Sign of each element will be negative for non-real atoms.
772 * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
773 * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
774 */
775 dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
776 jnrA = (jnrlistA>=0) ? jnrlistA : 0;
777 jnrB = (jnrlistB>=0) ? jnrlistB : 0;
778 jnrC = (jnrlistC>=0) ? jnrlistC : 0;
779 jnrD = (jnrlistD>=0) ? jnrlistD : 0;
780 j_coord_offsetA = DIM*jnrA;
781 j_coord_offsetB = DIM*jnrB;
782 j_coord_offsetC = DIM*jnrC;
783 j_coord_offsetD = DIM*jnrD;
784
785 /* load j atom coordinates */
786 gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
787 x+j_coord_offsetC,x+j_coord_offsetD,
788 &jx0,&jy0,&jz0);
789
790 /* Calculate displacement vector */
791 dx00 = _mm_sub_ps(ix0,jx0);
792 dy00 = _mm_sub_ps(iy0,jy0);
793 dz00 = _mm_sub_ps(iz0,jz0);
794 dx10 = _mm_sub_ps(ix1,jx0);
795 dy10 = _mm_sub_ps(iy1,jy0);
796 dz10 = _mm_sub_ps(iz1,jz0);
797 dx20 = _mm_sub_ps(ix2,jx0);
798 dy20 = _mm_sub_ps(iy2,jy0);
799 dz20 = _mm_sub_ps(iz2,jz0);
800
801 /* Calculate squared distance and things based on it */
802 rsq00 = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
803 rsq10 = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
804 rsq20 = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
805
806 rinv00 = gmx_mm_invsqrt_ps(rsq00);
807 rinv10 = gmx_mm_invsqrt_ps(rsq10);
808 rinv20 = gmx_mm_invsqrt_ps(rsq20);
809
810 rinvsq00 = _mm_mul_ps(rinv00,rinv00);
811 rinvsq10 = _mm_mul_ps(rinv10,rinv10);
812 rinvsq20 = _mm_mul_ps(rinv20,rinv20);
813
814 /* Load parameters for j particles */
815 jq0 = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
816 charge+jnrC+0,charge+jnrD+0);
817
818 fjx0 = _mm_setzero_ps();
819 fjy0 = _mm_setzero_ps();
820 fjz0 = _mm_setzero_ps();
821
822 /**************************
823 * CALCULATE INTERACTIONS *
824 **************************/
825
826 /* Compute parameters for interactions between i and j atoms */
827 qq00 = _mm_mul_ps(iq0,jq0);
828
829 /* COULOMB ELECTROSTATICS */
830 velec = _mm_mul_ps(qq00,rinv00);
831 felec = _mm_mul_ps(velec,rinvsq00);
832
833 fscal = felec;
834
835 fscal = _mm_andnot_ps(dummy_mask,fscal);
836
837 /* Calculate temporary vectorial force */
838 tx = _mm_mul_ps(fscal,dx00);
839 ty = _mm_mul_ps(fscal,dy00);
840 tz = _mm_mul_ps(fscal,dz00);
841
842 /* Update vectorial force */
843 fix0 = _mm_add_ps(fix0,tx);
844 fiy0 = _mm_add_ps(fiy0,ty);
845 fiz0 = _mm_add_ps(fiz0,tz);
846
847 fjx0 = _mm_add_ps(fjx0,tx);
848 fjy0 = _mm_add_ps(fjy0,ty);
849 fjz0 = _mm_add_ps(fjz0,tz);
850
851 /**************************
852 * CALCULATE INTERACTIONS *
853 **************************/
854
855 /* Compute parameters for interactions between i and j atoms */
856 qq10 = _mm_mul_ps(iq1,jq0);
857
858 /* COULOMB ELECTROSTATICS */
859 velec = _mm_mul_ps(qq10,rinv10);
860 felec = _mm_mul_ps(velec,rinvsq10);
861
862 fscal = felec;
863
864 fscal = _mm_andnot_ps(dummy_mask,fscal);
865
866 /* Calculate temporary vectorial force */
867 tx = _mm_mul_ps(fscal,dx10);
868 ty = _mm_mul_ps(fscal,dy10);
869 tz = _mm_mul_ps(fscal,dz10);
870
871 /* Update vectorial force */
872 fix1 = _mm_add_ps(fix1,tx);
873 fiy1 = _mm_add_ps(fiy1,ty);
874 fiz1 = _mm_add_ps(fiz1,tz);
875
876 fjx0 = _mm_add_ps(fjx0,tx);
877 fjy0 = _mm_add_ps(fjy0,ty);
878 fjz0 = _mm_add_ps(fjz0,tz);
879
880 /**************************
881 * CALCULATE INTERACTIONS *
882 **************************/
883
884 /* Compute parameters for interactions between i and j atoms */
885 qq20 = _mm_mul_ps(iq2,jq0);
886
887 /* COULOMB ELECTROSTATICS */
888 velec = _mm_mul_ps(qq20,rinv20);
889 felec = _mm_mul_ps(velec,rinvsq20);
890
891 fscal = felec;
892
893 fscal = _mm_andnot_ps(dummy_mask,fscal);
894
895 /* Calculate temporary vectorial force */
896 tx = _mm_mul_ps(fscal,dx20);
897 ty = _mm_mul_ps(fscal,dy20);
898 tz = _mm_mul_ps(fscal,dz20);
899
900 /* Update vectorial force */
901 fix2 = _mm_add_ps(fix2,tx);
902 fiy2 = _mm_add_ps(fiy2,ty);
903 fiz2 = _mm_add_ps(fiz2,tz);
904
905 fjx0 = _mm_add_ps(fjx0,tx);
906 fjy0 = _mm_add_ps(fjy0,ty);
907 fjz0 = _mm_add_ps(fjz0,tz);
908
909 fjptrA = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
910 fjptrB = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
911 fjptrC = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
912 fjptrD = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
913
914 gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
915
916 /* Inner loop uses 81 flops */
917 }
918
919 /* End of innermost loop */
920
921 gmx_mm_update_iforce_3atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
922 f+i_coord_offset,fshift+i_shift_offset);
923
924 /* Increment number of inner iterations */
925 inneriter += j_index_end - j_index_start;
926
927 /* Outer loop uses 18 flops */
928 }
929
930 /* Increment number of outer iterations */
931 outeriter += nri;
932
933 /* Update outer/inner flops */
934
935 inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_W3_F,outeriter*18 + inneriter*81);
936 }
The ultimate molecular dynamics simulation package