?  search:
summary | log | graphiclog1 | graphiclog2 | commit | commitdiff | tree | refs | edit | fork
blame | history | raw | HEAD
Removed include simple.h from nb_kernel_avx_128_fma_XX
[gromacs.git] / src / gromacs / gmxlib / nonbonded / nb_kernel_avx_128_fma_double / nb_kernel_ElecCoul_VdwLJ_GeomW3P1_avx_128_fma_double.c
blob9e045a162fc70707acec80f51023ab389c0b2bc6
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 avx_128_fma_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_avx_128_fma_double.h"
49 #include "kernelutil_x86_avx_128_fma_double.h"
50
51 /*
52 * Gromacs nonbonded kernel: nb_kernel_ElecCoul_VdwLJ_GeomW3P1_VF_avx_128_fma_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_avx_128_fma_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_msub_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 /* Update vectorial force */
243 fix0 = _mm_macc_pd(dx00,fscal,fix0);
244 fiy0 = _mm_macc_pd(dy00,fscal,fiy0);
245 fiz0 = _mm_macc_pd(dz00,fscal,fiz0);
246
247 fjx0 = _mm_macc_pd(dx00,fscal,fjx0);
248 fjy0 = _mm_macc_pd(dy00,fscal,fjy0);
249 fjz0 = _mm_macc_pd(dz00,fscal,fjz0);
250
251 /**************************
252 * CALCULATE INTERACTIONS *
253 **************************/
254
255 /* Compute parameters for interactions between i and j atoms */
256 qq10 = _mm_mul_pd(iq1,jq0);
257
258 /* COULOMB ELECTROSTATICS */
259 velec = _mm_mul_pd(qq10,rinv10);
260 felec = _mm_mul_pd(velec,rinvsq10);
261
262 /* Update potential sum for this i atom from the interaction with this j atom. */
263 velecsum = _mm_add_pd(velecsum,velec);
264
265 fscal = felec;
266
267 /* Update vectorial force */
268 fix1 = _mm_macc_pd(dx10,fscal,fix1);
269 fiy1 = _mm_macc_pd(dy10,fscal,fiy1);
270 fiz1 = _mm_macc_pd(dz10,fscal,fiz1);
271
272 fjx0 = _mm_macc_pd(dx10,fscal,fjx0);
273 fjy0 = _mm_macc_pd(dy10,fscal,fjy0);
274 fjz0 = _mm_macc_pd(dz10,fscal,fjz0);
275
276 /**************************
277 * CALCULATE INTERACTIONS *
278 **************************/
279
280 /* Compute parameters for interactions between i and j atoms */
281 qq20 = _mm_mul_pd(iq2,jq0);
282
283 /* COULOMB ELECTROSTATICS */
284 velec = _mm_mul_pd(qq20,rinv20);
285 felec = _mm_mul_pd(velec,rinvsq20);
286
287 /* Update potential sum for this i atom from the interaction with this j atom. */
288 velecsum = _mm_add_pd(velecsum,velec);
289
290 fscal = felec;
291
292 /* Update vectorial force */
293 fix2 = _mm_macc_pd(dx20,fscal,fix2);
294 fiy2 = _mm_macc_pd(dy20,fscal,fiy2);
295 fiz2 = _mm_macc_pd(dz20,fscal,fiz2);
296
297 fjx0 = _mm_macc_pd(dx20,fscal,fjx0);
298 fjy0 = _mm_macc_pd(dy20,fscal,fjy0);
299 fjz0 = _mm_macc_pd(dz20,fscal,fjz0);
300
301 gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,fjx0,fjy0,fjz0);
302
303 /* Inner loop uses 108 flops */
304 }
305
306 if(jidx<j_index_end)
307 {
308
309 jnrA = jjnr[jidx];
310 j_coord_offsetA = DIM*jnrA;
311
312 /* load j atom coordinates */
313 gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
314 &jx0,&jy0,&jz0);
315
316 /* Calculate displacement vector */
317 dx00 = _mm_sub_pd(ix0,jx0);
318 dy00 = _mm_sub_pd(iy0,jy0);
319 dz00 = _mm_sub_pd(iz0,jz0);
320 dx10 = _mm_sub_pd(ix1,jx0);
321 dy10 = _mm_sub_pd(iy1,jy0);
322 dz10 = _mm_sub_pd(iz1,jz0);
323 dx20 = _mm_sub_pd(ix2,jx0);
324 dy20 = _mm_sub_pd(iy2,jy0);
325 dz20 = _mm_sub_pd(iz2,jz0);
326
327 /* Calculate squared distance and things based on it */
328 rsq00 = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
329 rsq10 = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
330 rsq20 = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
331
332 rinv00 = gmx_mm_invsqrt_pd(rsq00);
333 rinv10 = gmx_mm_invsqrt_pd(rsq10);
334 rinv20 = gmx_mm_invsqrt_pd(rsq20);
335
336 rinvsq00 = _mm_mul_pd(rinv00,rinv00);
337 rinvsq10 = _mm_mul_pd(rinv10,rinv10);
338 rinvsq20 = _mm_mul_pd(rinv20,rinv20);
339
340 /* Load parameters for j particles */
341 jq0 = _mm_load_sd(charge+jnrA+0);
342 vdwjidx0A = 2*vdwtype[jnrA+0];
343
344 fjx0 = _mm_setzero_pd();
345 fjy0 = _mm_setzero_pd();
346 fjz0 = _mm_setzero_pd();
347
348 /**************************
349 * CALCULATE INTERACTIONS *
350 **************************/
351
352 /* Compute parameters for interactions between i and j atoms */
353 qq00 = _mm_mul_pd(iq0,jq0);
354 gmx_mm_load_1pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,&c6_00,&c12_00);
355
356 /* COULOMB ELECTROSTATICS */
357 velec = _mm_mul_pd(qq00,rinv00);
358 felec = _mm_mul_pd(velec,rinvsq00);
359
360 /* LENNARD-JONES DISPERSION/REPULSION */
361
362 rinvsix = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
363 vvdw6 = _mm_mul_pd(c6_00,rinvsix);
364 vvdw12 = _mm_mul_pd(c12_00,_mm_mul_pd(rinvsix,rinvsix));
365 vvdw = _mm_msub_pd( vvdw12,one_twelfth, _mm_mul_pd(vvdw6,one_sixth) );
366 fvdw = _mm_mul_pd(_mm_sub_pd(vvdw12,vvdw6),rinvsq00);
367
368 /* Update potential sum for this i atom from the interaction with this j atom. */
369 velec = _mm_unpacklo_pd(velec,_mm_setzero_pd());
370 velecsum = _mm_add_pd(velecsum,velec);
371 vvdw = _mm_unpacklo_pd(vvdw,_mm_setzero_pd());
372 vvdwsum = _mm_add_pd(vvdwsum,vvdw);
373
374 fscal = _mm_add_pd(felec,fvdw);
375
376 fscal = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
377
378 /* Update vectorial force */
379 fix0 = _mm_macc_pd(dx00,fscal,fix0);
380 fiy0 = _mm_macc_pd(dy00,fscal,fiy0);
381 fiz0 = _mm_macc_pd(dz00,fscal,fiz0);
382
383 fjx0 = _mm_macc_pd(dx00,fscal,fjx0);
384 fjy0 = _mm_macc_pd(dy00,fscal,fjy0);
385 fjz0 = _mm_macc_pd(dz00,fscal,fjz0);
386
387 /**************************
388 * CALCULATE INTERACTIONS *
389 **************************/
390
391 /* Compute parameters for interactions between i and j atoms */
392 qq10 = _mm_mul_pd(iq1,jq0);
393
394 /* COULOMB ELECTROSTATICS */
395 velec = _mm_mul_pd(qq10,rinv10);
396 felec = _mm_mul_pd(velec,rinvsq10);
397
398 /* Update potential sum for this i atom from the interaction with this j atom. */
399 velec = _mm_unpacklo_pd(velec,_mm_setzero_pd());
400 velecsum = _mm_add_pd(velecsum,velec);
401
402 fscal = felec;
403
404 fscal = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
405
406 /* Update vectorial force */
407 fix1 = _mm_macc_pd(dx10,fscal,fix1);
408 fiy1 = _mm_macc_pd(dy10,fscal,fiy1);
409 fiz1 = _mm_macc_pd(dz10,fscal,fiz1);
410
411 fjx0 = _mm_macc_pd(dx10,fscal,fjx0);
412 fjy0 = _mm_macc_pd(dy10,fscal,fjy0);
413 fjz0 = _mm_macc_pd(dz10,fscal,fjz0);
414
415 /**************************
416 * CALCULATE INTERACTIONS *
417 **************************/
418
419 /* Compute parameters for interactions between i and j atoms */
420 qq20 = _mm_mul_pd(iq2,jq0);
421
422 /* COULOMB ELECTROSTATICS */
423 velec = _mm_mul_pd(qq20,rinv20);
424 felec = _mm_mul_pd(velec,rinvsq20);
425
426 /* Update potential sum for this i atom from the interaction with this j atom. */
427 velec = _mm_unpacklo_pd(velec,_mm_setzero_pd());
428 velecsum = _mm_add_pd(velecsum,velec);
429
430 fscal = felec;
431
432 fscal = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
433
434 /* Update vectorial force */
435 fix2 = _mm_macc_pd(dx20,fscal,fix2);
436 fiy2 = _mm_macc_pd(dy20,fscal,fiy2);
437 fiz2 = _mm_macc_pd(dz20,fscal,fiz2);
438
439 fjx0 = _mm_macc_pd(dx20,fscal,fjx0);
440 fjy0 = _mm_macc_pd(dy20,fscal,fjy0);
441 fjz0 = _mm_macc_pd(dz20,fscal,fjz0);
442
443 gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,fjx0,fjy0,fjz0);
444
445 /* Inner loop uses 108 flops */
446 }
447
448 /* End of innermost loop */
449
450 gmx_mm_update_iforce_3atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
451 f+i_coord_offset,fshift+i_shift_offset);
452
453 ggid = gid[iidx];
454 /* Update potential energies */
455 gmx_mm_update_1pot_pd(velecsum,kernel_data->energygrp_elec+ggid);
456 gmx_mm_update_1pot_pd(vvdwsum,kernel_data->energygrp_vdw+ggid);
457
458 /* Increment number of inner iterations */
459 inneriter += j_index_end - j_index_start;
460
461 /* Outer loop uses 20 flops */
462 }
463
464 /* Increment number of outer iterations */
465 outeriter += nri;
466
467 /* Update outer/inner flops */
468
469 inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3_VF,outeriter*20 + inneriter*108);
470 }
471 /*
472 * Gromacs nonbonded kernel: nb_kernel_ElecCoul_VdwLJ_GeomW3P1_F_avx_128_fma_double
473 * Electrostatics interaction: Coulomb
474 * VdW interaction: LennardJones
475 * Geometry: Water3-Particle
476 * Calculate force/pot: Force
477 */
478 void
479 nb_kernel_ElecCoul_VdwLJ_GeomW3P1_F_avx_128_fma_double
480 (t_nblist * gmx_restrict nlist,
481 rvec * gmx_restrict xx,
482 rvec * gmx_restrict ff,
483 t_forcerec * gmx_restrict fr,
484 t_mdatoms * gmx_restrict mdatoms,
485 nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
486 t_nrnb * gmx_restrict nrnb)
487 {
488 /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
489 * just 0 for non-waters.
490 * Suffixes A,B refer to j loop unrolling done with SSE double precision, e.g. for the two different
491 * jnr indices corresponding to data put in the four positions in the SIMD register.
492 */
493 int i_shift_offset,i_coord_offset,outeriter,inneriter;
494 int j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
495 int jnrA,jnrB;
496 int j_coord_offsetA,j_coord_offsetB;
497 int *iinr,*jindex,*jjnr,*shiftidx,*gid;
498 real rcutoff_scalar;
499 real *shiftvec,*fshift,*x,*f;
500 __m128d tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
501 int vdwioffset0;
502 __m128d ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
503 int vdwioffset1;
504 __m128d ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
505 int vdwioffset2;
506 __m128d ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
507 int vdwjidx0A,vdwjidx0B;
508 __m128d jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
509 __m128d dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
510 __m128d dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
511 __m128d dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
512 __m128d velec,felec,velecsum,facel,crf,krf,krf2;
513 real *charge;
514 int nvdwtype;
515 __m128d rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
516 int *vdwtype;
517 real *vdwparam;
518 __m128d one_sixth = _mm_set1_pd(1.0/6.0);
519 __m128d one_twelfth = _mm_set1_pd(1.0/12.0);
520 __m128d dummy_mask,cutoff_mask;
521 __m128d signbit = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
522 __m128d one = _mm_set1_pd(1.0);
523 __m128d two = _mm_set1_pd(2.0);
524 x = xx[0];
525 f = ff[0];
526
527 nri = nlist->nri;
528 iinr = nlist->iinr;
529 jindex = nlist->jindex;
530 jjnr = nlist->jjnr;
531 shiftidx = nlist->shift;
532 gid = nlist->gid;
533 shiftvec = fr->shift_vec[0];
534 fshift = fr->fshift[0];
535 facel = _mm_set1_pd(fr->epsfac);
536 charge = mdatoms->chargeA;
537 nvdwtype = fr->ntype;
538 vdwparam = fr->nbfp;
539 vdwtype = mdatoms->typeA;
540
541 /* Setup water-specific parameters */
542 inr = nlist->iinr[0];
543 iq0 = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+0]));
544 iq1 = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+1]));
545 iq2 = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+2]));
546 vdwioffset0 = 2*nvdwtype*vdwtype[inr+0];
547
548 /* Avoid stupid compiler warnings */
549 jnrA = jnrB = 0;
550 j_coord_offsetA = 0;
551 j_coord_offsetB = 0;
552
553 outeriter = 0;
554 inneriter = 0;
555
556 /* Start outer loop over neighborlists */
557 for(iidx=0; iidx<nri; iidx++)
558 {
559 /* Load shift vector for this list */
560 i_shift_offset = DIM*shiftidx[iidx];
561
562 /* Load limits for loop over neighbors */
563 j_index_start = jindex[iidx];
564 j_index_end = jindex[iidx+1];
565
566 /* Get outer coordinate index */
567 inr = iinr[iidx];
568 i_coord_offset = DIM*inr;
569
570 /* Load i particle coords and add shift vector */
571 gmx_mm_load_shift_and_3rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
572 &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
573
574 fix0 = _mm_setzero_pd();
575 fiy0 = _mm_setzero_pd();
576 fiz0 = _mm_setzero_pd();
577 fix1 = _mm_setzero_pd();
578 fiy1 = _mm_setzero_pd();
579 fiz1 = _mm_setzero_pd();
580 fix2 = _mm_setzero_pd();
581 fiy2 = _mm_setzero_pd();
582 fiz2 = _mm_setzero_pd();
583
584 /* Start inner kernel loop */
585 for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
586 {
587
588 /* Get j neighbor index, and coordinate index */
589 jnrA = jjnr[jidx];
590 jnrB = jjnr[jidx+1];
591 j_coord_offsetA = DIM*jnrA;
592 j_coord_offsetB = DIM*jnrB;
593
594 /* load j atom coordinates */
595 gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
596 &jx0,&jy0,&jz0);
597
598 /* Calculate displacement vector */
599 dx00 = _mm_sub_pd(ix0,jx0);
600 dy00 = _mm_sub_pd(iy0,jy0);
601 dz00 = _mm_sub_pd(iz0,jz0);
602 dx10 = _mm_sub_pd(ix1,jx0);
603 dy10 = _mm_sub_pd(iy1,jy0);
604 dz10 = _mm_sub_pd(iz1,jz0);
605 dx20 = _mm_sub_pd(ix2,jx0);
606 dy20 = _mm_sub_pd(iy2,jy0);
607 dz20 = _mm_sub_pd(iz2,jz0);
608
609 /* Calculate squared distance and things based on it */
610 rsq00 = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
611 rsq10 = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
612 rsq20 = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
613
614 rinv00 = gmx_mm_invsqrt_pd(rsq00);
615 rinv10 = gmx_mm_invsqrt_pd(rsq10);
616 rinv20 = gmx_mm_invsqrt_pd(rsq20);
617
618 rinvsq00 = _mm_mul_pd(rinv00,rinv00);
619 rinvsq10 = _mm_mul_pd(rinv10,rinv10);
620 rinvsq20 = _mm_mul_pd(rinv20,rinv20);
621
622 /* Load parameters for j particles */
623 jq0 = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
624 vdwjidx0A = 2*vdwtype[jnrA+0];
625 vdwjidx0B = 2*vdwtype[jnrB+0];
626
627 fjx0 = _mm_setzero_pd();
628 fjy0 = _mm_setzero_pd();
629 fjz0 = _mm_setzero_pd();
630
631 /**************************
632 * CALCULATE INTERACTIONS *
633 **************************/
634
635 /* Compute parameters for interactions between i and j atoms */
636 qq00 = _mm_mul_pd(iq0,jq0);
637 gmx_mm_load_2pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,
638 vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
639
640 /* COULOMB ELECTROSTATICS */
641 velec = _mm_mul_pd(qq00,rinv00);
642 felec = _mm_mul_pd(velec,rinvsq00);
643
644 /* LENNARD-JONES DISPERSION/REPULSION */
645
646 rinvsix = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
647 fvdw = _mm_mul_pd(_mm_msub_pd(c12_00,rinvsix,c6_00),_mm_mul_pd(rinvsix,rinvsq00));
648
649 fscal = _mm_add_pd(felec,fvdw);
650
651 /* Update vectorial force */
652 fix0 = _mm_macc_pd(dx00,fscal,fix0);
653 fiy0 = _mm_macc_pd(dy00,fscal,fiy0);
654 fiz0 = _mm_macc_pd(dz00,fscal,fiz0);
655
656 fjx0 = _mm_macc_pd(dx00,fscal,fjx0);
657 fjy0 = _mm_macc_pd(dy00,fscal,fjy0);
658 fjz0 = _mm_macc_pd(dz00,fscal,fjz0);
659
660 /**************************
661 * CALCULATE INTERACTIONS *
662 **************************/
663
664 /* Compute parameters for interactions between i and j atoms */
665 qq10 = _mm_mul_pd(iq1,jq0);
666
667 /* COULOMB ELECTROSTATICS */
668 velec = _mm_mul_pd(qq10,rinv10);
669 felec = _mm_mul_pd(velec,rinvsq10);
670
671 fscal = felec;
672
673 /* Update vectorial force */
674 fix1 = _mm_macc_pd(dx10,fscal,fix1);
675 fiy1 = _mm_macc_pd(dy10,fscal,fiy1);
676 fiz1 = _mm_macc_pd(dz10,fscal,fiz1);
677
678 fjx0 = _mm_macc_pd(dx10,fscal,fjx0);
679 fjy0 = _mm_macc_pd(dy10,fscal,fjy0);
680 fjz0 = _mm_macc_pd(dz10,fscal,fjz0);
681
682 /**************************
683 * CALCULATE INTERACTIONS *
684 **************************/
685
686 /* Compute parameters for interactions between i and j atoms */
687 qq20 = _mm_mul_pd(iq2,jq0);
688
689 /* COULOMB ELECTROSTATICS */
690 velec = _mm_mul_pd(qq20,rinv20);
691 felec = _mm_mul_pd(velec,rinvsq20);
692
693 fscal = felec;
694
695 /* Update vectorial force */
696 fix2 = _mm_macc_pd(dx20,fscal,fix2);
697 fiy2 = _mm_macc_pd(dy20,fscal,fiy2);
698 fiz2 = _mm_macc_pd(dz20,fscal,fiz2);
699
700 fjx0 = _mm_macc_pd(dx20,fscal,fjx0);
701 fjy0 = _mm_macc_pd(dy20,fscal,fjy0);
702 fjz0 = _mm_macc_pd(dz20,fscal,fjz0);
703
704 gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,fjx0,fjy0,fjz0);
705
706 /* Inner loop uses 100 flops */
707 }
708
709 if(jidx<j_index_end)
710 {
711
712 jnrA = jjnr[jidx];
713 j_coord_offsetA = DIM*jnrA;
714
715 /* load j atom coordinates */
716 gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
717 &jx0,&jy0,&jz0);
718
719 /* Calculate displacement vector */
720 dx00 = _mm_sub_pd(ix0,jx0);
721 dy00 = _mm_sub_pd(iy0,jy0);
722 dz00 = _mm_sub_pd(iz0,jz0);
723 dx10 = _mm_sub_pd(ix1,jx0);
724 dy10 = _mm_sub_pd(iy1,jy0);
725 dz10 = _mm_sub_pd(iz1,jz0);
726 dx20 = _mm_sub_pd(ix2,jx0);
727 dy20 = _mm_sub_pd(iy2,jy0);
728 dz20 = _mm_sub_pd(iz2,jz0);
729
730 /* Calculate squared distance and things based on it */
731 rsq00 = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
732 rsq10 = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
733 rsq20 = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
734
735 rinv00 = gmx_mm_invsqrt_pd(rsq00);
736 rinv10 = gmx_mm_invsqrt_pd(rsq10);
737 rinv20 = gmx_mm_invsqrt_pd(rsq20);
738
739 rinvsq00 = _mm_mul_pd(rinv00,rinv00);
740 rinvsq10 = _mm_mul_pd(rinv10,rinv10);
741 rinvsq20 = _mm_mul_pd(rinv20,rinv20);
742
743 /* Load parameters for j particles */
744 jq0 = _mm_load_sd(charge+jnrA+0);
745 vdwjidx0A = 2*vdwtype[jnrA+0];
746
747 fjx0 = _mm_setzero_pd();
748 fjy0 = _mm_setzero_pd();
749 fjz0 = _mm_setzero_pd();
750
751 /**************************
752 * CALCULATE INTERACTIONS *
753 **************************/
754
755 /* Compute parameters for interactions between i and j atoms */
756 qq00 = _mm_mul_pd(iq0,jq0);
757 gmx_mm_load_1pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,&c6_00,&c12_00);
758
759 /* COULOMB ELECTROSTATICS */
760 velec = _mm_mul_pd(qq00,rinv00);
761 felec = _mm_mul_pd(velec,rinvsq00);
762
763 /* LENNARD-JONES DISPERSION/REPULSION */
764
765 rinvsix = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
766 fvdw = _mm_mul_pd(_mm_msub_pd(c12_00,rinvsix,c6_00),_mm_mul_pd(rinvsix,rinvsq00));
767
768 fscal = _mm_add_pd(felec,fvdw);
769
770 fscal = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
771
772 /* Update vectorial force */
773 fix0 = _mm_macc_pd(dx00,fscal,fix0);
774 fiy0 = _mm_macc_pd(dy00,fscal,fiy0);
775 fiz0 = _mm_macc_pd(dz00,fscal,fiz0);
776
777 fjx0 = _mm_macc_pd(dx00,fscal,fjx0);
778 fjy0 = _mm_macc_pd(dy00,fscal,fjy0);
779 fjz0 = _mm_macc_pd(dz00,fscal,fjz0);
780
781 /**************************
782 * CALCULATE INTERACTIONS *
783 **************************/
784
785 /* Compute parameters for interactions between i and j atoms */
786 qq10 = _mm_mul_pd(iq1,jq0);
787
788 /* COULOMB ELECTROSTATICS */
789 velec = _mm_mul_pd(qq10,rinv10);
790 felec = _mm_mul_pd(velec,rinvsq10);
791
792 fscal = felec;
793
794 fscal = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
795
796 /* Update vectorial force */
797 fix1 = _mm_macc_pd(dx10,fscal,fix1);
798 fiy1 = _mm_macc_pd(dy10,fscal,fiy1);
799 fiz1 = _mm_macc_pd(dz10,fscal,fiz1);
800
801 fjx0 = _mm_macc_pd(dx10,fscal,fjx0);
802 fjy0 = _mm_macc_pd(dy10,fscal,fjy0);
803 fjz0 = _mm_macc_pd(dz10,fscal,fjz0);
804
805 /**************************
806 * CALCULATE INTERACTIONS *
807 **************************/
808
809 /* Compute parameters for interactions between i and j atoms */
810 qq20 = _mm_mul_pd(iq2,jq0);
811
812 /* COULOMB ELECTROSTATICS */
813 velec = _mm_mul_pd(qq20,rinv20);
814 felec = _mm_mul_pd(velec,rinvsq20);
815
816 fscal = felec;
817
818 fscal = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
819
820 /* Update vectorial force */
821 fix2 = _mm_macc_pd(dx20,fscal,fix2);
822 fiy2 = _mm_macc_pd(dy20,fscal,fiy2);
823 fiz2 = _mm_macc_pd(dz20,fscal,fiz2);
824
825 fjx0 = _mm_macc_pd(dx20,fscal,fjx0);
826 fjy0 = _mm_macc_pd(dy20,fscal,fjy0);
827 fjz0 = _mm_macc_pd(dz20,fscal,fjz0);
828
829 gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,fjx0,fjy0,fjz0);
830
831 /* Inner loop uses 100 flops */
832 }
833
834 /* End of innermost loop */
835
836 gmx_mm_update_iforce_3atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
837 f+i_coord_offset,fshift+i_shift_offset);
838
839 /* Increment number of inner iterations */
840 inneriter += j_index_end - j_index_start;
841
842 /* Outer loop uses 18 flops */
843 }
844
845 /* Increment number of outer iterations */
846 outeriter += nri;
847
848 /* Update outer/inner flops */
849
850 inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3_F,outeriter*18 + inneriter*100);
851 }
The ultimate molecular dynamics simulation package