2 * This file is part of the GROMACS molecular simulation package.
4 * Copyright (c) 2012,2013,2014, 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.
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.
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.
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.
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.
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.
36 * Note: this file was generated by the GROMACS avx_256_double kernel generator.
42 #include "../nb_kernel.h"
43 #include "gromacs/legacyheaders/types/simple.h"
44 #include "gromacs/math/vec.h"
45 #include "gromacs/legacyheaders/nrnb.h"
47 #include "gromacs/simd/math_x86_avx_256_double.h"
48 #include "kernelutil_x86_avx_256_double.h"
51 * Gromacs nonbonded kernel: nb_kernel_ElecCoul_VdwLJ_GeomW3P1_VF_avx_256_double
52 * Electrostatics interaction: Coulomb
53 * VdW interaction: LennardJones
54 * Geometry: Water3-Particle
55 * Calculate force/pot: PotentialAndForce
58 nb_kernel_ElecCoul_VdwLJ_GeomW3P1_VF_avx_256_double
59 (t_nblist * gmx_restrict nlist,
60 rvec * gmx_restrict xx,
61 rvec * gmx_restrict ff,
62 t_forcerec * gmx_restrict fr,
63 t_mdatoms * gmx_restrict mdatoms,
64 nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
65 t_nrnb * gmx_restrict nrnb)
67 /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
68 * just 0 for non-waters.
69 * Suffixes A,B,C,D refer to j loop unrolling done with AVX, e.g. for the four different
70 * jnr indices corresponding to data put in the four positions in the SIMD register.
72 int i_shift_offset,i_coord_offset,outeriter,inneriter;
73 int j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
74 int jnrA,jnrB,jnrC,jnrD;
75 int jnrlistA,jnrlistB,jnrlistC,jnrlistD;
76 int jnrlistE,jnrlistF,jnrlistG,jnrlistH;
77 int j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
78 int *iinr,*jindex,*jjnr,*shiftidx,*gid;
80 real *shiftvec,*fshift,*x,*f;
81 real *fjptrA,*fjptrB,*fjptrC,*fjptrD;
83 __m256d tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
84 real * vdwioffsetptr0;
85 __m256d ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
86 real * vdwioffsetptr1;
87 __m256d ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
88 real * vdwioffsetptr2;
89 __m256d ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
90 int vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
91 __m256d jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
92 __m256d dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
93 __m256d dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
94 __m256d dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
95 __m256d velec,felec,velecsum,facel,crf,krf,krf2;
98 __m256d rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
101 __m256d one_sixth = _mm256_set1_pd(1.0/6.0);
102 __m256d one_twelfth = _mm256_set1_pd(1.0/12.0);
103 __m256d dummy_mask,cutoff_mask;
104 __m128 tmpmask0,tmpmask1;
105 __m256d signbit = _mm256_castsi256_pd( _mm256_set1_epi32(0x80000000) );
106 __m256d one = _mm256_set1_pd(1.0);
107 __m256d two = _mm256_set1_pd(2.0);
113 jindex = nlist->jindex;
115 shiftidx = nlist->shift;
117 shiftvec = fr->shift_vec[0];
118 fshift = fr->fshift[0];
119 facel = _mm256_set1_pd(fr->epsfac);
120 charge = mdatoms->chargeA;
121 nvdwtype = fr->ntype;
123 vdwtype = mdatoms->typeA;
125 /* Setup water-specific parameters */
126 inr = nlist->iinr[0];
127 iq0 = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+0]));
128 iq1 = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+1]));
129 iq2 = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+2]));
130 vdwioffsetptr0 = vdwparam+2*nvdwtype*vdwtype[inr+0];
132 /* Avoid stupid compiler warnings */
133 jnrA = jnrB = jnrC = jnrD = 0;
142 for(iidx=0;iidx<4*DIM;iidx++)
147 /* Start outer loop over neighborlists */
148 for(iidx=0; iidx<nri; iidx++)
150 /* Load shift vector for this list */
151 i_shift_offset = DIM*shiftidx[iidx];
153 /* Load limits for loop over neighbors */
154 j_index_start = jindex[iidx];
155 j_index_end = jindex[iidx+1];
157 /* Get outer coordinate index */
159 i_coord_offset = DIM*inr;
161 /* Load i particle coords and add shift vector */
162 gmx_mm256_load_shift_and_3rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
163 &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
165 fix0 = _mm256_setzero_pd();
166 fiy0 = _mm256_setzero_pd();
167 fiz0 = _mm256_setzero_pd();
168 fix1 = _mm256_setzero_pd();
169 fiy1 = _mm256_setzero_pd();
170 fiz1 = _mm256_setzero_pd();
171 fix2 = _mm256_setzero_pd();
172 fiy2 = _mm256_setzero_pd();
173 fiz2 = _mm256_setzero_pd();
175 /* Reset potential sums */
176 velecsum = _mm256_setzero_pd();
177 vvdwsum = _mm256_setzero_pd();
179 /* Start inner kernel loop */
180 for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
183 /* Get j neighbor index, and coordinate index */
188 j_coord_offsetA = DIM*jnrA;
189 j_coord_offsetB = DIM*jnrB;
190 j_coord_offsetC = DIM*jnrC;
191 j_coord_offsetD = DIM*jnrD;
193 /* load j atom coordinates */
194 gmx_mm256_load_1rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
195 x+j_coord_offsetC,x+j_coord_offsetD,
198 /* Calculate displacement vector */
199 dx00 = _mm256_sub_pd(ix0,jx0);
200 dy00 = _mm256_sub_pd(iy0,jy0);
201 dz00 = _mm256_sub_pd(iz0,jz0);
202 dx10 = _mm256_sub_pd(ix1,jx0);
203 dy10 = _mm256_sub_pd(iy1,jy0);
204 dz10 = _mm256_sub_pd(iz1,jz0);
205 dx20 = _mm256_sub_pd(ix2,jx0);
206 dy20 = _mm256_sub_pd(iy2,jy0);
207 dz20 = _mm256_sub_pd(iz2,jz0);
209 /* Calculate squared distance and things based on it */
210 rsq00 = gmx_mm256_calc_rsq_pd(dx00,dy00,dz00);
211 rsq10 = gmx_mm256_calc_rsq_pd(dx10,dy10,dz10);
212 rsq20 = gmx_mm256_calc_rsq_pd(dx20,dy20,dz20);
214 rinv00 = gmx_mm256_invsqrt_pd(rsq00);
215 rinv10 = gmx_mm256_invsqrt_pd(rsq10);
216 rinv20 = gmx_mm256_invsqrt_pd(rsq20);
218 rinvsq00 = _mm256_mul_pd(rinv00,rinv00);
219 rinvsq10 = _mm256_mul_pd(rinv10,rinv10);
220 rinvsq20 = _mm256_mul_pd(rinv20,rinv20);
222 /* Load parameters for j particles */
223 jq0 = gmx_mm256_load_4real_swizzle_pd(charge+jnrA+0,charge+jnrB+0,
224 charge+jnrC+0,charge+jnrD+0);
225 vdwjidx0A = 2*vdwtype[jnrA+0];
226 vdwjidx0B = 2*vdwtype[jnrB+0];
227 vdwjidx0C = 2*vdwtype[jnrC+0];
228 vdwjidx0D = 2*vdwtype[jnrD+0];
230 fjx0 = _mm256_setzero_pd();
231 fjy0 = _mm256_setzero_pd();
232 fjz0 = _mm256_setzero_pd();
234 /**************************
235 * CALCULATE INTERACTIONS *
236 **************************/
238 /* Compute parameters for interactions between i and j atoms */
239 qq00 = _mm256_mul_pd(iq0,jq0);
240 gmx_mm256_load_4pair_swizzle_pd(vdwioffsetptr0+vdwjidx0A,
241 vdwioffsetptr0+vdwjidx0B,
242 vdwioffsetptr0+vdwjidx0C,
243 vdwioffsetptr0+vdwjidx0D,
246 /* COULOMB ELECTROSTATICS */
247 velec = _mm256_mul_pd(qq00,rinv00);
248 felec = _mm256_mul_pd(velec,rinvsq00);
250 /* LENNARD-JONES DISPERSION/REPULSION */
252 rinvsix = _mm256_mul_pd(_mm256_mul_pd(rinvsq00,rinvsq00),rinvsq00);
253 vvdw6 = _mm256_mul_pd(c6_00,rinvsix);
254 vvdw12 = _mm256_mul_pd(c12_00,_mm256_mul_pd(rinvsix,rinvsix));
255 vvdw = _mm256_sub_pd( _mm256_mul_pd(vvdw12,one_twelfth) , _mm256_mul_pd(vvdw6,one_sixth) );
256 fvdw = _mm256_mul_pd(_mm256_sub_pd(vvdw12,vvdw6),rinvsq00);
258 /* Update potential sum for this i atom from the interaction with this j atom. */
259 velecsum = _mm256_add_pd(velecsum,velec);
260 vvdwsum = _mm256_add_pd(vvdwsum,vvdw);
262 fscal = _mm256_add_pd(felec,fvdw);
264 /* Calculate temporary vectorial force */
265 tx = _mm256_mul_pd(fscal,dx00);
266 ty = _mm256_mul_pd(fscal,dy00);
267 tz = _mm256_mul_pd(fscal,dz00);
269 /* Update vectorial force */
270 fix0 = _mm256_add_pd(fix0,tx);
271 fiy0 = _mm256_add_pd(fiy0,ty);
272 fiz0 = _mm256_add_pd(fiz0,tz);
274 fjx0 = _mm256_add_pd(fjx0,tx);
275 fjy0 = _mm256_add_pd(fjy0,ty);
276 fjz0 = _mm256_add_pd(fjz0,tz);
278 /**************************
279 * CALCULATE INTERACTIONS *
280 **************************/
282 /* Compute parameters for interactions between i and j atoms */
283 qq10 = _mm256_mul_pd(iq1,jq0);
285 /* COULOMB ELECTROSTATICS */
286 velec = _mm256_mul_pd(qq10,rinv10);
287 felec = _mm256_mul_pd(velec,rinvsq10);
289 /* Update potential sum for this i atom from the interaction with this j atom. */
290 velecsum = _mm256_add_pd(velecsum,velec);
294 /* Calculate temporary vectorial force */
295 tx = _mm256_mul_pd(fscal,dx10);
296 ty = _mm256_mul_pd(fscal,dy10);
297 tz = _mm256_mul_pd(fscal,dz10);
299 /* Update vectorial force */
300 fix1 = _mm256_add_pd(fix1,tx);
301 fiy1 = _mm256_add_pd(fiy1,ty);
302 fiz1 = _mm256_add_pd(fiz1,tz);
304 fjx0 = _mm256_add_pd(fjx0,tx);
305 fjy0 = _mm256_add_pd(fjy0,ty);
306 fjz0 = _mm256_add_pd(fjz0,tz);
308 /**************************
309 * CALCULATE INTERACTIONS *
310 **************************/
312 /* Compute parameters for interactions between i and j atoms */
313 qq20 = _mm256_mul_pd(iq2,jq0);
315 /* COULOMB ELECTROSTATICS */
316 velec = _mm256_mul_pd(qq20,rinv20);
317 felec = _mm256_mul_pd(velec,rinvsq20);
319 /* Update potential sum for this i atom from the interaction with this j atom. */
320 velecsum = _mm256_add_pd(velecsum,velec);
324 /* Calculate temporary vectorial force */
325 tx = _mm256_mul_pd(fscal,dx20);
326 ty = _mm256_mul_pd(fscal,dy20);
327 tz = _mm256_mul_pd(fscal,dz20);
329 /* Update vectorial force */
330 fix2 = _mm256_add_pd(fix2,tx);
331 fiy2 = _mm256_add_pd(fiy2,ty);
332 fiz2 = _mm256_add_pd(fiz2,tz);
334 fjx0 = _mm256_add_pd(fjx0,tx);
335 fjy0 = _mm256_add_pd(fjy0,ty);
336 fjz0 = _mm256_add_pd(fjz0,tz);
338 fjptrA = f+j_coord_offsetA;
339 fjptrB = f+j_coord_offsetB;
340 fjptrC = f+j_coord_offsetC;
341 fjptrD = f+j_coord_offsetD;
343 gmx_mm256_decrement_1rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
345 /* Inner loop uses 96 flops */
351 /* Get j neighbor index, and coordinate index */
352 jnrlistA = jjnr[jidx];
353 jnrlistB = jjnr[jidx+1];
354 jnrlistC = jjnr[jidx+2];
355 jnrlistD = jjnr[jidx+3];
356 /* Sign of each element will be negative for non-real atoms.
357 * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
358 * so use it as val = _mm_andnot_pd(mask,val) to clear dummy entries.
360 tmpmask0 = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
362 tmpmask1 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(3,3,2,2));
363 tmpmask0 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(1,1,0,0));
364 dummy_mask = _mm256_castps_pd(gmx_mm256_set_m128(tmpmask1,tmpmask0));
366 jnrA = (jnrlistA>=0) ? jnrlistA : 0;
367 jnrB = (jnrlistB>=0) ? jnrlistB : 0;
368 jnrC = (jnrlistC>=0) ? jnrlistC : 0;
369 jnrD = (jnrlistD>=0) ? jnrlistD : 0;
370 j_coord_offsetA = DIM*jnrA;
371 j_coord_offsetB = DIM*jnrB;
372 j_coord_offsetC = DIM*jnrC;
373 j_coord_offsetD = DIM*jnrD;
375 /* load j atom coordinates */
376 gmx_mm256_load_1rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
377 x+j_coord_offsetC,x+j_coord_offsetD,
380 /* Calculate displacement vector */
381 dx00 = _mm256_sub_pd(ix0,jx0);
382 dy00 = _mm256_sub_pd(iy0,jy0);
383 dz00 = _mm256_sub_pd(iz0,jz0);
384 dx10 = _mm256_sub_pd(ix1,jx0);
385 dy10 = _mm256_sub_pd(iy1,jy0);
386 dz10 = _mm256_sub_pd(iz1,jz0);
387 dx20 = _mm256_sub_pd(ix2,jx0);
388 dy20 = _mm256_sub_pd(iy2,jy0);
389 dz20 = _mm256_sub_pd(iz2,jz0);
391 /* Calculate squared distance and things based on it */
392 rsq00 = gmx_mm256_calc_rsq_pd(dx00,dy00,dz00);
393 rsq10 = gmx_mm256_calc_rsq_pd(dx10,dy10,dz10);
394 rsq20 = gmx_mm256_calc_rsq_pd(dx20,dy20,dz20);
396 rinv00 = gmx_mm256_invsqrt_pd(rsq00);
397 rinv10 = gmx_mm256_invsqrt_pd(rsq10);
398 rinv20 = gmx_mm256_invsqrt_pd(rsq20);
400 rinvsq00 = _mm256_mul_pd(rinv00,rinv00);
401 rinvsq10 = _mm256_mul_pd(rinv10,rinv10);
402 rinvsq20 = _mm256_mul_pd(rinv20,rinv20);
404 /* Load parameters for j particles */
405 jq0 = gmx_mm256_load_4real_swizzle_pd(charge+jnrA+0,charge+jnrB+0,
406 charge+jnrC+0,charge+jnrD+0);
407 vdwjidx0A = 2*vdwtype[jnrA+0];
408 vdwjidx0B = 2*vdwtype[jnrB+0];
409 vdwjidx0C = 2*vdwtype[jnrC+0];
410 vdwjidx0D = 2*vdwtype[jnrD+0];
412 fjx0 = _mm256_setzero_pd();
413 fjy0 = _mm256_setzero_pd();
414 fjz0 = _mm256_setzero_pd();
416 /**************************
417 * CALCULATE INTERACTIONS *
418 **************************/
420 /* Compute parameters for interactions between i and j atoms */
421 qq00 = _mm256_mul_pd(iq0,jq0);
422 gmx_mm256_load_4pair_swizzle_pd(vdwioffsetptr0+vdwjidx0A,
423 vdwioffsetptr0+vdwjidx0B,
424 vdwioffsetptr0+vdwjidx0C,
425 vdwioffsetptr0+vdwjidx0D,
428 /* COULOMB ELECTROSTATICS */
429 velec = _mm256_mul_pd(qq00,rinv00);
430 felec = _mm256_mul_pd(velec,rinvsq00);
432 /* LENNARD-JONES DISPERSION/REPULSION */
434 rinvsix = _mm256_mul_pd(_mm256_mul_pd(rinvsq00,rinvsq00),rinvsq00);
435 vvdw6 = _mm256_mul_pd(c6_00,rinvsix);
436 vvdw12 = _mm256_mul_pd(c12_00,_mm256_mul_pd(rinvsix,rinvsix));
437 vvdw = _mm256_sub_pd( _mm256_mul_pd(vvdw12,one_twelfth) , _mm256_mul_pd(vvdw6,one_sixth) );
438 fvdw = _mm256_mul_pd(_mm256_sub_pd(vvdw12,vvdw6),rinvsq00);
440 /* Update potential sum for this i atom from the interaction with this j atom. */
441 velec = _mm256_andnot_pd(dummy_mask,velec);
442 velecsum = _mm256_add_pd(velecsum,velec);
443 vvdw = _mm256_andnot_pd(dummy_mask,vvdw);
444 vvdwsum = _mm256_add_pd(vvdwsum,vvdw);
446 fscal = _mm256_add_pd(felec,fvdw);
448 fscal = _mm256_andnot_pd(dummy_mask,fscal);
450 /* Calculate temporary vectorial force */
451 tx = _mm256_mul_pd(fscal,dx00);
452 ty = _mm256_mul_pd(fscal,dy00);
453 tz = _mm256_mul_pd(fscal,dz00);
455 /* Update vectorial force */
456 fix0 = _mm256_add_pd(fix0,tx);
457 fiy0 = _mm256_add_pd(fiy0,ty);
458 fiz0 = _mm256_add_pd(fiz0,tz);
460 fjx0 = _mm256_add_pd(fjx0,tx);
461 fjy0 = _mm256_add_pd(fjy0,ty);
462 fjz0 = _mm256_add_pd(fjz0,tz);
464 /**************************
465 * CALCULATE INTERACTIONS *
466 **************************/
468 /* Compute parameters for interactions between i and j atoms */
469 qq10 = _mm256_mul_pd(iq1,jq0);
471 /* COULOMB ELECTROSTATICS */
472 velec = _mm256_mul_pd(qq10,rinv10);
473 felec = _mm256_mul_pd(velec,rinvsq10);
475 /* Update potential sum for this i atom from the interaction with this j atom. */
476 velec = _mm256_andnot_pd(dummy_mask,velec);
477 velecsum = _mm256_add_pd(velecsum,velec);
481 fscal = _mm256_andnot_pd(dummy_mask,fscal);
483 /* Calculate temporary vectorial force */
484 tx = _mm256_mul_pd(fscal,dx10);
485 ty = _mm256_mul_pd(fscal,dy10);
486 tz = _mm256_mul_pd(fscal,dz10);
488 /* Update vectorial force */
489 fix1 = _mm256_add_pd(fix1,tx);
490 fiy1 = _mm256_add_pd(fiy1,ty);
491 fiz1 = _mm256_add_pd(fiz1,tz);
493 fjx0 = _mm256_add_pd(fjx0,tx);
494 fjy0 = _mm256_add_pd(fjy0,ty);
495 fjz0 = _mm256_add_pd(fjz0,tz);
497 /**************************
498 * CALCULATE INTERACTIONS *
499 **************************/
501 /* Compute parameters for interactions between i and j atoms */
502 qq20 = _mm256_mul_pd(iq2,jq0);
504 /* COULOMB ELECTROSTATICS */
505 velec = _mm256_mul_pd(qq20,rinv20);
506 felec = _mm256_mul_pd(velec,rinvsq20);
508 /* Update potential sum for this i atom from the interaction with this j atom. */
509 velec = _mm256_andnot_pd(dummy_mask,velec);
510 velecsum = _mm256_add_pd(velecsum,velec);
514 fscal = _mm256_andnot_pd(dummy_mask,fscal);
516 /* Calculate temporary vectorial force */
517 tx = _mm256_mul_pd(fscal,dx20);
518 ty = _mm256_mul_pd(fscal,dy20);
519 tz = _mm256_mul_pd(fscal,dz20);
521 /* Update vectorial force */
522 fix2 = _mm256_add_pd(fix2,tx);
523 fiy2 = _mm256_add_pd(fiy2,ty);
524 fiz2 = _mm256_add_pd(fiz2,tz);
526 fjx0 = _mm256_add_pd(fjx0,tx);
527 fjy0 = _mm256_add_pd(fjy0,ty);
528 fjz0 = _mm256_add_pd(fjz0,tz);
530 fjptrA = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
531 fjptrB = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
532 fjptrC = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
533 fjptrD = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
535 gmx_mm256_decrement_1rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
537 /* Inner loop uses 96 flops */
540 /* End of innermost loop */
542 gmx_mm256_update_iforce_3atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
543 f+i_coord_offset,fshift+i_shift_offset);
546 /* Update potential energies */
547 gmx_mm256_update_1pot_pd(velecsum,kernel_data->energygrp_elec+ggid);
548 gmx_mm256_update_1pot_pd(vvdwsum,kernel_data->energygrp_vdw+ggid);
550 /* Increment number of inner iterations */
551 inneriter += j_index_end - j_index_start;
553 /* Outer loop uses 20 flops */
556 /* Increment number of outer iterations */
559 /* Update outer/inner flops */
561 inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3_VF,outeriter*20 + inneriter*96);
564 * Gromacs nonbonded kernel: nb_kernel_ElecCoul_VdwLJ_GeomW3P1_F_avx_256_double
565 * Electrostatics interaction: Coulomb
566 * VdW interaction: LennardJones
567 * Geometry: Water3-Particle
568 * Calculate force/pot: Force
571 nb_kernel_ElecCoul_VdwLJ_GeomW3P1_F_avx_256_double
572 (t_nblist * gmx_restrict nlist,
573 rvec * gmx_restrict xx,
574 rvec * gmx_restrict ff,
575 t_forcerec * gmx_restrict fr,
576 t_mdatoms * gmx_restrict mdatoms,
577 nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
578 t_nrnb * gmx_restrict nrnb)
580 /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
581 * just 0 for non-waters.
582 * Suffixes A,B,C,D refer to j loop unrolling done with AVX, e.g. for the four different
583 * jnr indices corresponding to data put in the four positions in the SIMD register.
585 int i_shift_offset,i_coord_offset,outeriter,inneriter;
586 int j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
587 int jnrA,jnrB,jnrC,jnrD;
588 int jnrlistA,jnrlistB,jnrlistC,jnrlistD;
589 int jnrlistE,jnrlistF,jnrlistG,jnrlistH;
590 int j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
591 int *iinr,*jindex,*jjnr,*shiftidx,*gid;
593 real *shiftvec,*fshift,*x,*f;
594 real *fjptrA,*fjptrB,*fjptrC,*fjptrD;
596 __m256d tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
597 real * vdwioffsetptr0;
598 __m256d ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
599 real * vdwioffsetptr1;
600 __m256d ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
601 real * vdwioffsetptr2;
602 __m256d ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
603 int vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
604 __m256d jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
605 __m256d dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
606 __m256d dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
607 __m256d dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
608 __m256d velec,felec,velecsum,facel,crf,krf,krf2;
611 __m256d rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
614 __m256d one_sixth = _mm256_set1_pd(1.0/6.0);
615 __m256d one_twelfth = _mm256_set1_pd(1.0/12.0);
616 __m256d dummy_mask,cutoff_mask;
617 __m128 tmpmask0,tmpmask1;
618 __m256d signbit = _mm256_castsi256_pd( _mm256_set1_epi32(0x80000000) );
619 __m256d one = _mm256_set1_pd(1.0);
620 __m256d two = _mm256_set1_pd(2.0);
626 jindex = nlist->jindex;
628 shiftidx = nlist->shift;
630 shiftvec = fr->shift_vec[0];
631 fshift = fr->fshift[0];
632 facel = _mm256_set1_pd(fr->epsfac);
633 charge = mdatoms->chargeA;
634 nvdwtype = fr->ntype;
636 vdwtype = mdatoms->typeA;
638 /* Setup water-specific parameters */
639 inr = nlist->iinr[0];
640 iq0 = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+0]));
641 iq1 = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+1]));
642 iq2 = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+2]));
643 vdwioffsetptr0 = vdwparam+2*nvdwtype*vdwtype[inr+0];
645 /* Avoid stupid compiler warnings */
646 jnrA = jnrB = jnrC = jnrD = 0;
655 for(iidx=0;iidx<4*DIM;iidx++)
660 /* Start outer loop over neighborlists */
661 for(iidx=0; iidx<nri; iidx++)
663 /* Load shift vector for this list */
664 i_shift_offset = DIM*shiftidx[iidx];
666 /* Load limits for loop over neighbors */
667 j_index_start = jindex[iidx];
668 j_index_end = jindex[iidx+1];
670 /* Get outer coordinate index */
672 i_coord_offset = DIM*inr;
674 /* Load i particle coords and add shift vector */
675 gmx_mm256_load_shift_and_3rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
676 &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
678 fix0 = _mm256_setzero_pd();
679 fiy0 = _mm256_setzero_pd();
680 fiz0 = _mm256_setzero_pd();
681 fix1 = _mm256_setzero_pd();
682 fiy1 = _mm256_setzero_pd();
683 fiz1 = _mm256_setzero_pd();
684 fix2 = _mm256_setzero_pd();
685 fiy2 = _mm256_setzero_pd();
686 fiz2 = _mm256_setzero_pd();
688 /* Start inner kernel loop */
689 for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
692 /* Get j neighbor index, and coordinate index */
697 j_coord_offsetA = DIM*jnrA;
698 j_coord_offsetB = DIM*jnrB;
699 j_coord_offsetC = DIM*jnrC;
700 j_coord_offsetD = DIM*jnrD;
702 /* load j atom coordinates */
703 gmx_mm256_load_1rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
704 x+j_coord_offsetC,x+j_coord_offsetD,
707 /* Calculate displacement vector */
708 dx00 = _mm256_sub_pd(ix0,jx0);
709 dy00 = _mm256_sub_pd(iy0,jy0);
710 dz00 = _mm256_sub_pd(iz0,jz0);
711 dx10 = _mm256_sub_pd(ix1,jx0);
712 dy10 = _mm256_sub_pd(iy1,jy0);
713 dz10 = _mm256_sub_pd(iz1,jz0);
714 dx20 = _mm256_sub_pd(ix2,jx0);
715 dy20 = _mm256_sub_pd(iy2,jy0);
716 dz20 = _mm256_sub_pd(iz2,jz0);
718 /* Calculate squared distance and things based on it */
719 rsq00 = gmx_mm256_calc_rsq_pd(dx00,dy00,dz00);
720 rsq10 = gmx_mm256_calc_rsq_pd(dx10,dy10,dz10);
721 rsq20 = gmx_mm256_calc_rsq_pd(dx20,dy20,dz20);
723 rinv00 = gmx_mm256_invsqrt_pd(rsq00);
724 rinv10 = gmx_mm256_invsqrt_pd(rsq10);
725 rinv20 = gmx_mm256_invsqrt_pd(rsq20);
727 rinvsq00 = _mm256_mul_pd(rinv00,rinv00);
728 rinvsq10 = _mm256_mul_pd(rinv10,rinv10);
729 rinvsq20 = _mm256_mul_pd(rinv20,rinv20);
731 /* Load parameters for j particles */
732 jq0 = gmx_mm256_load_4real_swizzle_pd(charge+jnrA+0,charge+jnrB+0,
733 charge+jnrC+0,charge+jnrD+0);
734 vdwjidx0A = 2*vdwtype[jnrA+0];
735 vdwjidx0B = 2*vdwtype[jnrB+0];
736 vdwjidx0C = 2*vdwtype[jnrC+0];
737 vdwjidx0D = 2*vdwtype[jnrD+0];
739 fjx0 = _mm256_setzero_pd();
740 fjy0 = _mm256_setzero_pd();
741 fjz0 = _mm256_setzero_pd();
743 /**************************
744 * CALCULATE INTERACTIONS *
745 **************************/
747 /* Compute parameters for interactions between i and j atoms */
748 qq00 = _mm256_mul_pd(iq0,jq0);
749 gmx_mm256_load_4pair_swizzle_pd(vdwioffsetptr0+vdwjidx0A,
750 vdwioffsetptr0+vdwjidx0B,
751 vdwioffsetptr0+vdwjidx0C,
752 vdwioffsetptr0+vdwjidx0D,
755 /* COULOMB ELECTROSTATICS */
756 velec = _mm256_mul_pd(qq00,rinv00);
757 felec = _mm256_mul_pd(velec,rinvsq00);
759 /* LENNARD-JONES DISPERSION/REPULSION */
761 rinvsix = _mm256_mul_pd(_mm256_mul_pd(rinvsq00,rinvsq00),rinvsq00);
762 fvdw = _mm256_mul_pd(_mm256_sub_pd(_mm256_mul_pd(c12_00,rinvsix),c6_00),_mm256_mul_pd(rinvsix,rinvsq00));
764 fscal = _mm256_add_pd(felec,fvdw);
766 /* Calculate temporary vectorial force */
767 tx = _mm256_mul_pd(fscal,dx00);
768 ty = _mm256_mul_pd(fscal,dy00);
769 tz = _mm256_mul_pd(fscal,dz00);
771 /* Update vectorial force */
772 fix0 = _mm256_add_pd(fix0,tx);
773 fiy0 = _mm256_add_pd(fiy0,ty);
774 fiz0 = _mm256_add_pd(fiz0,tz);
776 fjx0 = _mm256_add_pd(fjx0,tx);
777 fjy0 = _mm256_add_pd(fjy0,ty);
778 fjz0 = _mm256_add_pd(fjz0,tz);
780 /**************************
781 * CALCULATE INTERACTIONS *
782 **************************/
784 /* Compute parameters for interactions between i and j atoms */
785 qq10 = _mm256_mul_pd(iq1,jq0);
787 /* COULOMB ELECTROSTATICS */
788 velec = _mm256_mul_pd(qq10,rinv10);
789 felec = _mm256_mul_pd(velec,rinvsq10);
793 /* Calculate temporary vectorial force */
794 tx = _mm256_mul_pd(fscal,dx10);
795 ty = _mm256_mul_pd(fscal,dy10);
796 tz = _mm256_mul_pd(fscal,dz10);
798 /* Update vectorial force */
799 fix1 = _mm256_add_pd(fix1,tx);
800 fiy1 = _mm256_add_pd(fiy1,ty);
801 fiz1 = _mm256_add_pd(fiz1,tz);
803 fjx0 = _mm256_add_pd(fjx0,tx);
804 fjy0 = _mm256_add_pd(fjy0,ty);
805 fjz0 = _mm256_add_pd(fjz0,tz);
807 /**************************
808 * CALCULATE INTERACTIONS *
809 **************************/
811 /* Compute parameters for interactions between i and j atoms */
812 qq20 = _mm256_mul_pd(iq2,jq0);
814 /* COULOMB ELECTROSTATICS */
815 velec = _mm256_mul_pd(qq20,rinv20);
816 felec = _mm256_mul_pd(velec,rinvsq20);
820 /* Calculate temporary vectorial force */
821 tx = _mm256_mul_pd(fscal,dx20);
822 ty = _mm256_mul_pd(fscal,dy20);
823 tz = _mm256_mul_pd(fscal,dz20);
825 /* Update vectorial force */
826 fix2 = _mm256_add_pd(fix2,tx);
827 fiy2 = _mm256_add_pd(fiy2,ty);
828 fiz2 = _mm256_add_pd(fiz2,tz);
830 fjx0 = _mm256_add_pd(fjx0,tx);
831 fjy0 = _mm256_add_pd(fjy0,ty);
832 fjz0 = _mm256_add_pd(fjz0,tz);
834 fjptrA = f+j_coord_offsetA;
835 fjptrB = f+j_coord_offsetB;
836 fjptrC = f+j_coord_offsetC;
837 fjptrD = f+j_coord_offsetD;
839 gmx_mm256_decrement_1rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
841 /* Inner loop uses 88 flops */
847 /* Get j neighbor index, and coordinate index */
848 jnrlistA = jjnr[jidx];
849 jnrlistB = jjnr[jidx+1];
850 jnrlistC = jjnr[jidx+2];
851 jnrlistD = jjnr[jidx+3];
852 /* Sign of each element will be negative for non-real atoms.
853 * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
854 * so use it as val = _mm_andnot_pd(mask,val) to clear dummy entries.
856 tmpmask0 = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
858 tmpmask1 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(3,3,2,2));
859 tmpmask0 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(1,1,0,0));
860 dummy_mask = _mm256_castps_pd(gmx_mm256_set_m128(tmpmask1,tmpmask0));
862 jnrA = (jnrlistA>=0) ? jnrlistA : 0;
863 jnrB = (jnrlistB>=0) ? jnrlistB : 0;
864 jnrC = (jnrlistC>=0) ? jnrlistC : 0;
865 jnrD = (jnrlistD>=0) ? jnrlistD : 0;
866 j_coord_offsetA = DIM*jnrA;
867 j_coord_offsetB = DIM*jnrB;
868 j_coord_offsetC = DIM*jnrC;
869 j_coord_offsetD = DIM*jnrD;
871 /* load j atom coordinates */
872 gmx_mm256_load_1rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
873 x+j_coord_offsetC,x+j_coord_offsetD,
876 /* Calculate displacement vector */
877 dx00 = _mm256_sub_pd(ix0,jx0);
878 dy00 = _mm256_sub_pd(iy0,jy0);
879 dz00 = _mm256_sub_pd(iz0,jz0);
880 dx10 = _mm256_sub_pd(ix1,jx0);
881 dy10 = _mm256_sub_pd(iy1,jy0);
882 dz10 = _mm256_sub_pd(iz1,jz0);
883 dx20 = _mm256_sub_pd(ix2,jx0);
884 dy20 = _mm256_sub_pd(iy2,jy0);
885 dz20 = _mm256_sub_pd(iz2,jz0);
887 /* Calculate squared distance and things based on it */
888 rsq00 = gmx_mm256_calc_rsq_pd(dx00,dy00,dz00);
889 rsq10 = gmx_mm256_calc_rsq_pd(dx10,dy10,dz10);
890 rsq20 = gmx_mm256_calc_rsq_pd(dx20,dy20,dz20);
892 rinv00 = gmx_mm256_invsqrt_pd(rsq00);
893 rinv10 = gmx_mm256_invsqrt_pd(rsq10);
894 rinv20 = gmx_mm256_invsqrt_pd(rsq20);
896 rinvsq00 = _mm256_mul_pd(rinv00,rinv00);
897 rinvsq10 = _mm256_mul_pd(rinv10,rinv10);
898 rinvsq20 = _mm256_mul_pd(rinv20,rinv20);
900 /* Load parameters for j particles */
901 jq0 = gmx_mm256_load_4real_swizzle_pd(charge+jnrA+0,charge+jnrB+0,
902 charge+jnrC+0,charge+jnrD+0);
903 vdwjidx0A = 2*vdwtype[jnrA+0];
904 vdwjidx0B = 2*vdwtype[jnrB+0];
905 vdwjidx0C = 2*vdwtype[jnrC+0];
906 vdwjidx0D = 2*vdwtype[jnrD+0];
908 fjx0 = _mm256_setzero_pd();
909 fjy0 = _mm256_setzero_pd();
910 fjz0 = _mm256_setzero_pd();
912 /**************************
913 * CALCULATE INTERACTIONS *
914 **************************/
916 /* Compute parameters for interactions between i and j atoms */
917 qq00 = _mm256_mul_pd(iq0,jq0);
918 gmx_mm256_load_4pair_swizzle_pd(vdwioffsetptr0+vdwjidx0A,
919 vdwioffsetptr0+vdwjidx0B,
920 vdwioffsetptr0+vdwjidx0C,
921 vdwioffsetptr0+vdwjidx0D,
924 /* COULOMB ELECTROSTATICS */
925 velec = _mm256_mul_pd(qq00,rinv00);
926 felec = _mm256_mul_pd(velec,rinvsq00);
928 /* LENNARD-JONES DISPERSION/REPULSION */
930 rinvsix = _mm256_mul_pd(_mm256_mul_pd(rinvsq00,rinvsq00),rinvsq00);
931 fvdw = _mm256_mul_pd(_mm256_sub_pd(_mm256_mul_pd(c12_00,rinvsix),c6_00),_mm256_mul_pd(rinvsix,rinvsq00));
933 fscal = _mm256_add_pd(felec,fvdw);
935 fscal = _mm256_andnot_pd(dummy_mask,fscal);
937 /* Calculate temporary vectorial force */
938 tx = _mm256_mul_pd(fscal,dx00);
939 ty = _mm256_mul_pd(fscal,dy00);
940 tz = _mm256_mul_pd(fscal,dz00);
942 /* Update vectorial force */
943 fix0 = _mm256_add_pd(fix0,tx);
944 fiy0 = _mm256_add_pd(fiy0,ty);
945 fiz0 = _mm256_add_pd(fiz0,tz);
947 fjx0 = _mm256_add_pd(fjx0,tx);
948 fjy0 = _mm256_add_pd(fjy0,ty);
949 fjz0 = _mm256_add_pd(fjz0,tz);
951 /**************************
952 * CALCULATE INTERACTIONS *
953 **************************/
955 /* Compute parameters for interactions between i and j atoms */
956 qq10 = _mm256_mul_pd(iq1,jq0);
958 /* COULOMB ELECTROSTATICS */
959 velec = _mm256_mul_pd(qq10,rinv10);
960 felec = _mm256_mul_pd(velec,rinvsq10);
964 fscal = _mm256_andnot_pd(dummy_mask,fscal);
966 /* Calculate temporary vectorial force */
967 tx = _mm256_mul_pd(fscal,dx10);
968 ty = _mm256_mul_pd(fscal,dy10);
969 tz = _mm256_mul_pd(fscal,dz10);
971 /* Update vectorial force */
972 fix1 = _mm256_add_pd(fix1,tx);
973 fiy1 = _mm256_add_pd(fiy1,ty);
974 fiz1 = _mm256_add_pd(fiz1,tz);
976 fjx0 = _mm256_add_pd(fjx0,tx);
977 fjy0 = _mm256_add_pd(fjy0,ty);
978 fjz0 = _mm256_add_pd(fjz0,tz);
980 /**************************
981 * CALCULATE INTERACTIONS *
982 **************************/
984 /* Compute parameters for interactions between i and j atoms */
985 qq20 = _mm256_mul_pd(iq2,jq0);
987 /* COULOMB ELECTROSTATICS */
988 velec = _mm256_mul_pd(qq20,rinv20);
989 felec = _mm256_mul_pd(velec,rinvsq20);
993 fscal = _mm256_andnot_pd(dummy_mask,fscal);
995 /* Calculate temporary vectorial force */
996 tx = _mm256_mul_pd(fscal,dx20);
997 ty = _mm256_mul_pd(fscal,dy20);
998 tz = _mm256_mul_pd(fscal,dz20);
1000 /* Update vectorial force */
1001 fix2 = _mm256_add_pd(fix2,tx);
1002 fiy2 = _mm256_add_pd(fiy2,ty);
1003 fiz2 = _mm256_add_pd(fiz2,tz);
1005 fjx0 = _mm256_add_pd(fjx0,tx);
1006 fjy0 = _mm256_add_pd(fjy0,ty);
1007 fjz0 = _mm256_add_pd(fjz0,tz);
1009 fjptrA = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
1010 fjptrB = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
1011 fjptrC = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
1012 fjptrD = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
1014 gmx_mm256_decrement_1rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
1016 /* Inner loop uses 88 flops */
1019 /* End of innermost loop */
1021 gmx_mm256_update_iforce_3atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
1022 f+i_coord_offset,fshift+i_shift_offset);
1024 /* Increment number of inner iterations */
1025 inneriter += j_index_end - j_index_start;
1027 /* Outer loop uses 18 flops */
1030 /* Increment number of outer iterations */
1033 /* Update outer/inner flops */
1035 inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3_F,outeriter*18 + inneriter*88);