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 "types/simple.h"
44 #include "gromacs/math/vec.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_VdwNone_GeomW4P1_VF_avx_256_double
52 * Electrostatics interaction: Coulomb
53 * VdW interaction: None
54 * Geometry: Water4-Particle
55 * Calculate force/pot: PotentialAndForce
58 nb_kernel_ElecCoul_VdwNone_GeomW4P1_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 * vdwioffsetptr1;
85 __m256d ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
86 real * vdwioffsetptr2;
87 __m256d ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
88 real * vdwioffsetptr3;
89 __m256d ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
90 int vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
91 __m256d jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
92 __m256d dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
93 __m256d dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
94 __m256d dx30,dy30,dz30,rsq30,rinv30,rinvsq30,r30,qq30,c6_30,c12_30;
95 __m256d velec,felec,velecsum,facel,crf,krf,krf2;
97 __m256d dummy_mask,cutoff_mask;
98 __m128 tmpmask0,tmpmask1;
99 __m256d signbit = _mm256_castsi256_pd( _mm256_set1_epi32(0x80000000) );
100 __m256d one = _mm256_set1_pd(1.0);
101 __m256d two = _mm256_set1_pd(2.0);
107 jindex = nlist->jindex;
109 shiftidx = nlist->shift;
111 shiftvec = fr->shift_vec[0];
112 fshift = fr->fshift[0];
113 facel = _mm256_set1_pd(fr->epsfac);
114 charge = mdatoms->chargeA;
116 /* Setup water-specific parameters */
117 inr = nlist->iinr[0];
118 iq1 = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+1]));
119 iq2 = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+2]));
120 iq3 = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+3]));
122 /* Avoid stupid compiler warnings */
123 jnrA = jnrB = jnrC = jnrD = 0;
132 for(iidx=0;iidx<4*DIM;iidx++)
137 /* Start outer loop over neighborlists */
138 for(iidx=0; iidx<nri; iidx++)
140 /* Load shift vector for this list */
141 i_shift_offset = DIM*shiftidx[iidx];
143 /* Load limits for loop over neighbors */
144 j_index_start = jindex[iidx];
145 j_index_end = jindex[iidx+1];
147 /* Get outer coordinate index */
149 i_coord_offset = DIM*inr;
151 /* Load i particle coords and add shift vector */
152 gmx_mm256_load_shift_and_3rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset+DIM,
153 &ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
155 fix1 = _mm256_setzero_pd();
156 fiy1 = _mm256_setzero_pd();
157 fiz1 = _mm256_setzero_pd();
158 fix2 = _mm256_setzero_pd();
159 fiy2 = _mm256_setzero_pd();
160 fiz2 = _mm256_setzero_pd();
161 fix3 = _mm256_setzero_pd();
162 fiy3 = _mm256_setzero_pd();
163 fiz3 = _mm256_setzero_pd();
165 /* Reset potential sums */
166 velecsum = _mm256_setzero_pd();
168 /* Start inner kernel loop */
169 for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
172 /* Get j neighbor index, and coordinate index */
177 j_coord_offsetA = DIM*jnrA;
178 j_coord_offsetB = DIM*jnrB;
179 j_coord_offsetC = DIM*jnrC;
180 j_coord_offsetD = DIM*jnrD;
182 /* load j atom coordinates */
183 gmx_mm256_load_1rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
184 x+j_coord_offsetC,x+j_coord_offsetD,
187 /* Calculate displacement vector */
188 dx10 = _mm256_sub_pd(ix1,jx0);
189 dy10 = _mm256_sub_pd(iy1,jy0);
190 dz10 = _mm256_sub_pd(iz1,jz0);
191 dx20 = _mm256_sub_pd(ix2,jx0);
192 dy20 = _mm256_sub_pd(iy2,jy0);
193 dz20 = _mm256_sub_pd(iz2,jz0);
194 dx30 = _mm256_sub_pd(ix3,jx0);
195 dy30 = _mm256_sub_pd(iy3,jy0);
196 dz30 = _mm256_sub_pd(iz3,jz0);
198 /* Calculate squared distance and things based on it */
199 rsq10 = gmx_mm256_calc_rsq_pd(dx10,dy10,dz10);
200 rsq20 = gmx_mm256_calc_rsq_pd(dx20,dy20,dz20);
201 rsq30 = gmx_mm256_calc_rsq_pd(dx30,dy30,dz30);
203 rinv10 = gmx_mm256_invsqrt_pd(rsq10);
204 rinv20 = gmx_mm256_invsqrt_pd(rsq20);
205 rinv30 = gmx_mm256_invsqrt_pd(rsq30);
207 rinvsq10 = _mm256_mul_pd(rinv10,rinv10);
208 rinvsq20 = _mm256_mul_pd(rinv20,rinv20);
209 rinvsq30 = _mm256_mul_pd(rinv30,rinv30);
211 /* Load parameters for j particles */
212 jq0 = gmx_mm256_load_4real_swizzle_pd(charge+jnrA+0,charge+jnrB+0,
213 charge+jnrC+0,charge+jnrD+0);
215 fjx0 = _mm256_setzero_pd();
216 fjy0 = _mm256_setzero_pd();
217 fjz0 = _mm256_setzero_pd();
219 /**************************
220 * CALCULATE INTERACTIONS *
221 **************************/
223 /* Compute parameters for interactions between i and j atoms */
224 qq10 = _mm256_mul_pd(iq1,jq0);
226 /* COULOMB ELECTROSTATICS */
227 velec = _mm256_mul_pd(qq10,rinv10);
228 felec = _mm256_mul_pd(velec,rinvsq10);
230 /* Update potential sum for this i atom from the interaction with this j atom. */
231 velecsum = _mm256_add_pd(velecsum,velec);
235 /* Calculate temporary vectorial force */
236 tx = _mm256_mul_pd(fscal,dx10);
237 ty = _mm256_mul_pd(fscal,dy10);
238 tz = _mm256_mul_pd(fscal,dz10);
240 /* Update vectorial force */
241 fix1 = _mm256_add_pd(fix1,tx);
242 fiy1 = _mm256_add_pd(fiy1,ty);
243 fiz1 = _mm256_add_pd(fiz1,tz);
245 fjx0 = _mm256_add_pd(fjx0,tx);
246 fjy0 = _mm256_add_pd(fjy0,ty);
247 fjz0 = _mm256_add_pd(fjz0,tz);
249 /**************************
250 * CALCULATE INTERACTIONS *
251 **************************/
253 /* Compute parameters for interactions between i and j atoms */
254 qq20 = _mm256_mul_pd(iq2,jq0);
256 /* COULOMB ELECTROSTATICS */
257 velec = _mm256_mul_pd(qq20,rinv20);
258 felec = _mm256_mul_pd(velec,rinvsq20);
260 /* Update potential sum for this i atom from the interaction with this j atom. */
261 velecsum = _mm256_add_pd(velecsum,velec);
265 /* Calculate temporary vectorial force */
266 tx = _mm256_mul_pd(fscal,dx20);
267 ty = _mm256_mul_pd(fscal,dy20);
268 tz = _mm256_mul_pd(fscal,dz20);
270 /* Update vectorial force */
271 fix2 = _mm256_add_pd(fix2,tx);
272 fiy2 = _mm256_add_pd(fiy2,ty);
273 fiz2 = _mm256_add_pd(fiz2,tz);
275 fjx0 = _mm256_add_pd(fjx0,tx);
276 fjy0 = _mm256_add_pd(fjy0,ty);
277 fjz0 = _mm256_add_pd(fjz0,tz);
279 /**************************
280 * CALCULATE INTERACTIONS *
281 **************************/
283 /* Compute parameters for interactions between i and j atoms */
284 qq30 = _mm256_mul_pd(iq3,jq0);
286 /* COULOMB ELECTROSTATICS */
287 velec = _mm256_mul_pd(qq30,rinv30);
288 felec = _mm256_mul_pd(velec,rinvsq30);
290 /* Update potential sum for this i atom from the interaction with this j atom. */
291 velecsum = _mm256_add_pd(velecsum,velec);
295 /* Calculate temporary vectorial force */
296 tx = _mm256_mul_pd(fscal,dx30);
297 ty = _mm256_mul_pd(fscal,dy30);
298 tz = _mm256_mul_pd(fscal,dz30);
300 /* Update vectorial force */
301 fix3 = _mm256_add_pd(fix3,tx);
302 fiy3 = _mm256_add_pd(fiy3,ty);
303 fiz3 = _mm256_add_pd(fiz3,tz);
305 fjx0 = _mm256_add_pd(fjx0,tx);
306 fjy0 = _mm256_add_pd(fjy0,ty);
307 fjz0 = _mm256_add_pd(fjz0,tz);
309 fjptrA = f+j_coord_offsetA;
310 fjptrB = f+j_coord_offsetB;
311 fjptrC = f+j_coord_offsetC;
312 fjptrD = f+j_coord_offsetD;
314 gmx_mm256_decrement_1rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
316 /* Inner loop uses 84 flops */
322 /* Get j neighbor index, and coordinate index */
323 jnrlistA = jjnr[jidx];
324 jnrlistB = jjnr[jidx+1];
325 jnrlistC = jjnr[jidx+2];
326 jnrlistD = jjnr[jidx+3];
327 /* Sign of each element will be negative for non-real atoms.
328 * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
329 * so use it as val = _mm_andnot_pd(mask,val) to clear dummy entries.
331 tmpmask0 = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
333 tmpmask1 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(3,3,2,2));
334 tmpmask0 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(1,1,0,0));
335 dummy_mask = _mm256_castps_pd(gmx_mm256_set_m128(tmpmask1,tmpmask0));
337 jnrA = (jnrlistA>=0) ? jnrlistA : 0;
338 jnrB = (jnrlistB>=0) ? jnrlistB : 0;
339 jnrC = (jnrlistC>=0) ? jnrlistC : 0;
340 jnrD = (jnrlistD>=0) ? jnrlistD : 0;
341 j_coord_offsetA = DIM*jnrA;
342 j_coord_offsetB = DIM*jnrB;
343 j_coord_offsetC = DIM*jnrC;
344 j_coord_offsetD = DIM*jnrD;
346 /* load j atom coordinates */
347 gmx_mm256_load_1rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
348 x+j_coord_offsetC,x+j_coord_offsetD,
351 /* Calculate displacement vector */
352 dx10 = _mm256_sub_pd(ix1,jx0);
353 dy10 = _mm256_sub_pd(iy1,jy0);
354 dz10 = _mm256_sub_pd(iz1,jz0);
355 dx20 = _mm256_sub_pd(ix2,jx0);
356 dy20 = _mm256_sub_pd(iy2,jy0);
357 dz20 = _mm256_sub_pd(iz2,jz0);
358 dx30 = _mm256_sub_pd(ix3,jx0);
359 dy30 = _mm256_sub_pd(iy3,jy0);
360 dz30 = _mm256_sub_pd(iz3,jz0);
362 /* Calculate squared distance and things based on it */
363 rsq10 = gmx_mm256_calc_rsq_pd(dx10,dy10,dz10);
364 rsq20 = gmx_mm256_calc_rsq_pd(dx20,dy20,dz20);
365 rsq30 = gmx_mm256_calc_rsq_pd(dx30,dy30,dz30);
367 rinv10 = gmx_mm256_invsqrt_pd(rsq10);
368 rinv20 = gmx_mm256_invsqrt_pd(rsq20);
369 rinv30 = gmx_mm256_invsqrt_pd(rsq30);
371 rinvsq10 = _mm256_mul_pd(rinv10,rinv10);
372 rinvsq20 = _mm256_mul_pd(rinv20,rinv20);
373 rinvsq30 = _mm256_mul_pd(rinv30,rinv30);
375 /* Load parameters for j particles */
376 jq0 = gmx_mm256_load_4real_swizzle_pd(charge+jnrA+0,charge+jnrB+0,
377 charge+jnrC+0,charge+jnrD+0);
379 fjx0 = _mm256_setzero_pd();
380 fjy0 = _mm256_setzero_pd();
381 fjz0 = _mm256_setzero_pd();
383 /**************************
384 * CALCULATE INTERACTIONS *
385 **************************/
387 /* Compute parameters for interactions between i and j atoms */
388 qq10 = _mm256_mul_pd(iq1,jq0);
390 /* COULOMB ELECTROSTATICS */
391 velec = _mm256_mul_pd(qq10,rinv10);
392 felec = _mm256_mul_pd(velec,rinvsq10);
394 /* Update potential sum for this i atom from the interaction with this j atom. */
395 velec = _mm256_andnot_pd(dummy_mask,velec);
396 velecsum = _mm256_add_pd(velecsum,velec);
400 fscal = _mm256_andnot_pd(dummy_mask,fscal);
402 /* Calculate temporary vectorial force */
403 tx = _mm256_mul_pd(fscal,dx10);
404 ty = _mm256_mul_pd(fscal,dy10);
405 tz = _mm256_mul_pd(fscal,dz10);
407 /* Update vectorial force */
408 fix1 = _mm256_add_pd(fix1,tx);
409 fiy1 = _mm256_add_pd(fiy1,ty);
410 fiz1 = _mm256_add_pd(fiz1,tz);
412 fjx0 = _mm256_add_pd(fjx0,tx);
413 fjy0 = _mm256_add_pd(fjy0,ty);
414 fjz0 = _mm256_add_pd(fjz0,tz);
416 /**************************
417 * CALCULATE INTERACTIONS *
418 **************************/
420 /* Compute parameters for interactions between i and j atoms */
421 qq20 = _mm256_mul_pd(iq2,jq0);
423 /* COULOMB ELECTROSTATICS */
424 velec = _mm256_mul_pd(qq20,rinv20);
425 felec = _mm256_mul_pd(velec,rinvsq20);
427 /* Update potential sum for this i atom from the interaction with this j atom. */
428 velec = _mm256_andnot_pd(dummy_mask,velec);
429 velecsum = _mm256_add_pd(velecsum,velec);
433 fscal = _mm256_andnot_pd(dummy_mask,fscal);
435 /* Calculate temporary vectorial force */
436 tx = _mm256_mul_pd(fscal,dx20);
437 ty = _mm256_mul_pd(fscal,dy20);
438 tz = _mm256_mul_pd(fscal,dz20);
440 /* Update vectorial force */
441 fix2 = _mm256_add_pd(fix2,tx);
442 fiy2 = _mm256_add_pd(fiy2,ty);
443 fiz2 = _mm256_add_pd(fiz2,tz);
445 fjx0 = _mm256_add_pd(fjx0,tx);
446 fjy0 = _mm256_add_pd(fjy0,ty);
447 fjz0 = _mm256_add_pd(fjz0,tz);
449 /**************************
450 * CALCULATE INTERACTIONS *
451 **************************/
453 /* Compute parameters for interactions between i and j atoms */
454 qq30 = _mm256_mul_pd(iq3,jq0);
456 /* COULOMB ELECTROSTATICS */
457 velec = _mm256_mul_pd(qq30,rinv30);
458 felec = _mm256_mul_pd(velec,rinvsq30);
460 /* Update potential sum for this i atom from the interaction with this j atom. */
461 velec = _mm256_andnot_pd(dummy_mask,velec);
462 velecsum = _mm256_add_pd(velecsum,velec);
466 fscal = _mm256_andnot_pd(dummy_mask,fscal);
468 /* Calculate temporary vectorial force */
469 tx = _mm256_mul_pd(fscal,dx30);
470 ty = _mm256_mul_pd(fscal,dy30);
471 tz = _mm256_mul_pd(fscal,dz30);
473 /* Update vectorial force */
474 fix3 = _mm256_add_pd(fix3,tx);
475 fiy3 = _mm256_add_pd(fiy3,ty);
476 fiz3 = _mm256_add_pd(fiz3,tz);
478 fjx0 = _mm256_add_pd(fjx0,tx);
479 fjy0 = _mm256_add_pd(fjy0,ty);
480 fjz0 = _mm256_add_pd(fjz0,tz);
482 fjptrA = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
483 fjptrB = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
484 fjptrC = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
485 fjptrD = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
487 gmx_mm256_decrement_1rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
489 /* Inner loop uses 84 flops */
492 /* End of innermost loop */
494 gmx_mm256_update_iforce_3atom_swizzle_pd(fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
495 f+i_coord_offset+DIM,fshift+i_shift_offset);
498 /* Update potential energies */
499 gmx_mm256_update_1pot_pd(velecsum,kernel_data->energygrp_elec+ggid);
501 /* Increment number of inner iterations */
502 inneriter += j_index_end - j_index_start;
504 /* Outer loop uses 19 flops */
507 /* Increment number of outer iterations */
510 /* Update outer/inner flops */
512 inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_W4_VF,outeriter*19 + inneriter*84);
515 * Gromacs nonbonded kernel: nb_kernel_ElecCoul_VdwNone_GeomW4P1_F_avx_256_double
516 * Electrostatics interaction: Coulomb
517 * VdW interaction: None
518 * Geometry: Water4-Particle
519 * Calculate force/pot: Force
522 nb_kernel_ElecCoul_VdwNone_GeomW4P1_F_avx_256_double
523 (t_nblist * gmx_restrict nlist,
524 rvec * gmx_restrict xx,
525 rvec * gmx_restrict ff,
526 t_forcerec * gmx_restrict fr,
527 t_mdatoms * gmx_restrict mdatoms,
528 nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
529 t_nrnb * gmx_restrict nrnb)
531 /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
532 * just 0 for non-waters.
533 * Suffixes A,B,C,D refer to j loop unrolling done with AVX, e.g. for the four different
534 * jnr indices corresponding to data put in the four positions in the SIMD register.
536 int i_shift_offset,i_coord_offset,outeriter,inneriter;
537 int j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
538 int jnrA,jnrB,jnrC,jnrD;
539 int jnrlistA,jnrlistB,jnrlistC,jnrlistD;
540 int jnrlistE,jnrlistF,jnrlistG,jnrlistH;
541 int j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
542 int *iinr,*jindex,*jjnr,*shiftidx,*gid;
544 real *shiftvec,*fshift,*x,*f;
545 real *fjptrA,*fjptrB,*fjptrC,*fjptrD;
547 __m256d tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
548 real * vdwioffsetptr1;
549 __m256d ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
550 real * vdwioffsetptr2;
551 __m256d ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
552 real * vdwioffsetptr3;
553 __m256d ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
554 int vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
555 __m256d jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
556 __m256d dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
557 __m256d dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
558 __m256d dx30,dy30,dz30,rsq30,rinv30,rinvsq30,r30,qq30,c6_30,c12_30;
559 __m256d velec,felec,velecsum,facel,crf,krf,krf2;
561 __m256d dummy_mask,cutoff_mask;
562 __m128 tmpmask0,tmpmask1;
563 __m256d signbit = _mm256_castsi256_pd( _mm256_set1_epi32(0x80000000) );
564 __m256d one = _mm256_set1_pd(1.0);
565 __m256d two = _mm256_set1_pd(2.0);
571 jindex = nlist->jindex;
573 shiftidx = nlist->shift;
575 shiftvec = fr->shift_vec[0];
576 fshift = fr->fshift[0];
577 facel = _mm256_set1_pd(fr->epsfac);
578 charge = mdatoms->chargeA;
580 /* Setup water-specific parameters */
581 inr = nlist->iinr[0];
582 iq1 = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+1]));
583 iq2 = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+2]));
584 iq3 = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+3]));
586 /* Avoid stupid compiler warnings */
587 jnrA = jnrB = jnrC = jnrD = 0;
596 for(iidx=0;iidx<4*DIM;iidx++)
601 /* Start outer loop over neighborlists */
602 for(iidx=0; iidx<nri; iidx++)
604 /* Load shift vector for this list */
605 i_shift_offset = DIM*shiftidx[iidx];
607 /* Load limits for loop over neighbors */
608 j_index_start = jindex[iidx];
609 j_index_end = jindex[iidx+1];
611 /* Get outer coordinate index */
613 i_coord_offset = DIM*inr;
615 /* Load i particle coords and add shift vector */
616 gmx_mm256_load_shift_and_3rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset+DIM,
617 &ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
619 fix1 = _mm256_setzero_pd();
620 fiy1 = _mm256_setzero_pd();
621 fiz1 = _mm256_setzero_pd();
622 fix2 = _mm256_setzero_pd();
623 fiy2 = _mm256_setzero_pd();
624 fiz2 = _mm256_setzero_pd();
625 fix3 = _mm256_setzero_pd();
626 fiy3 = _mm256_setzero_pd();
627 fiz3 = _mm256_setzero_pd();
629 /* Start inner kernel loop */
630 for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
633 /* Get j neighbor index, and coordinate index */
638 j_coord_offsetA = DIM*jnrA;
639 j_coord_offsetB = DIM*jnrB;
640 j_coord_offsetC = DIM*jnrC;
641 j_coord_offsetD = DIM*jnrD;
643 /* load j atom coordinates */
644 gmx_mm256_load_1rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
645 x+j_coord_offsetC,x+j_coord_offsetD,
648 /* Calculate displacement vector */
649 dx10 = _mm256_sub_pd(ix1,jx0);
650 dy10 = _mm256_sub_pd(iy1,jy0);
651 dz10 = _mm256_sub_pd(iz1,jz0);
652 dx20 = _mm256_sub_pd(ix2,jx0);
653 dy20 = _mm256_sub_pd(iy2,jy0);
654 dz20 = _mm256_sub_pd(iz2,jz0);
655 dx30 = _mm256_sub_pd(ix3,jx0);
656 dy30 = _mm256_sub_pd(iy3,jy0);
657 dz30 = _mm256_sub_pd(iz3,jz0);
659 /* Calculate squared distance and things based on it */
660 rsq10 = gmx_mm256_calc_rsq_pd(dx10,dy10,dz10);
661 rsq20 = gmx_mm256_calc_rsq_pd(dx20,dy20,dz20);
662 rsq30 = gmx_mm256_calc_rsq_pd(dx30,dy30,dz30);
664 rinv10 = gmx_mm256_invsqrt_pd(rsq10);
665 rinv20 = gmx_mm256_invsqrt_pd(rsq20);
666 rinv30 = gmx_mm256_invsqrt_pd(rsq30);
668 rinvsq10 = _mm256_mul_pd(rinv10,rinv10);
669 rinvsq20 = _mm256_mul_pd(rinv20,rinv20);
670 rinvsq30 = _mm256_mul_pd(rinv30,rinv30);
672 /* Load parameters for j particles */
673 jq0 = gmx_mm256_load_4real_swizzle_pd(charge+jnrA+0,charge+jnrB+0,
674 charge+jnrC+0,charge+jnrD+0);
676 fjx0 = _mm256_setzero_pd();
677 fjy0 = _mm256_setzero_pd();
678 fjz0 = _mm256_setzero_pd();
680 /**************************
681 * CALCULATE INTERACTIONS *
682 **************************/
684 /* Compute parameters for interactions between i and j atoms */
685 qq10 = _mm256_mul_pd(iq1,jq0);
687 /* COULOMB ELECTROSTATICS */
688 velec = _mm256_mul_pd(qq10,rinv10);
689 felec = _mm256_mul_pd(velec,rinvsq10);
693 /* Calculate temporary vectorial force */
694 tx = _mm256_mul_pd(fscal,dx10);
695 ty = _mm256_mul_pd(fscal,dy10);
696 tz = _mm256_mul_pd(fscal,dz10);
698 /* Update vectorial force */
699 fix1 = _mm256_add_pd(fix1,tx);
700 fiy1 = _mm256_add_pd(fiy1,ty);
701 fiz1 = _mm256_add_pd(fiz1,tz);
703 fjx0 = _mm256_add_pd(fjx0,tx);
704 fjy0 = _mm256_add_pd(fjy0,ty);
705 fjz0 = _mm256_add_pd(fjz0,tz);
707 /**************************
708 * CALCULATE INTERACTIONS *
709 **************************/
711 /* Compute parameters for interactions between i and j atoms */
712 qq20 = _mm256_mul_pd(iq2,jq0);
714 /* COULOMB ELECTROSTATICS */
715 velec = _mm256_mul_pd(qq20,rinv20);
716 felec = _mm256_mul_pd(velec,rinvsq20);
720 /* Calculate temporary vectorial force */
721 tx = _mm256_mul_pd(fscal,dx20);
722 ty = _mm256_mul_pd(fscal,dy20);
723 tz = _mm256_mul_pd(fscal,dz20);
725 /* Update vectorial force */
726 fix2 = _mm256_add_pd(fix2,tx);
727 fiy2 = _mm256_add_pd(fiy2,ty);
728 fiz2 = _mm256_add_pd(fiz2,tz);
730 fjx0 = _mm256_add_pd(fjx0,tx);
731 fjy0 = _mm256_add_pd(fjy0,ty);
732 fjz0 = _mm256_add_pd(fjz0,tz);
734 /**************************
735 * CALCULATE INTERACTIONS *
736 **************************/
738 /* Compute parameters for interactions between i and j atoms */
739 qq30 = _mm256_mul_pd(iq3,jq0);
741 /* COULOMB ELECTROSTATICS */
742 velec = _mm256_mul_pd(qq30,rinv30);
743 felec = _mm256_mul_pd(velec,rinvsq30);
747 /* Calculate temporary vectorial force */
748 tx = _mm256_mul_pd(fscal,dx30);
749 ty = _mm256_mul_pd(fscal,dy30);
750 tz = _mm256_mul_pd(fscal,dz30);
752 /* Update vectorial force */
753 fix3 = _mm256_add_pd(fix3,tx);
754 fiy3 = _mm256_add_pd(fiy3,ty);
755 fiz3 = _mm256_add_pd(fiz3,tz);
757 fjx0 = _mm256_add_pd(fjx0,tx);
758 fjy0 = _mm256_add_pd(fjy0,ty);
759 fjz0 = _mm256_add_pd(fjz0,tz);
761 fjptrA = f+j_coord_offsetA;
762 fjptrB = f+j_coord_offsetB;
763 fjptrC = f+j_coord_offsetC;
764 fjptrD = f+j_coord_offsetD;
766 gmx_mm256_decrement_1rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
768 /* Inner loop uses 81 flops */
774 /* Get j neighbor index, and coordinate index */
775 jnrlistA = jjnr[jidx];
776 jnrlistB = jjnr[jidx+1];
777 jnrlistC = jjnr[jidx+2];
778 jnrlistD = jjnr[jidx+3];
779 /* Sign of each element will be negative for non-real atoms.
780 * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
781 * so use it as val = _mm_andnot_pd(mask,val) to clear dummy entries.
783 tmpmask0 = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
785 tmpmask1 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(3,3,2,2));
786 tmpmask0 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(1,1,0,0));
787 dummy_mask = _mm256_castps_pd(gmx_mm256_set_m128(tmpmask1,tmpmask0));
789 jnrA = (jnrlistA>=0) ? jnrlistA : 0;
790 jnrB = (jnrlistB>=0) ? jnrlistB : 0;
791 jnrC = (jnrlistC>=0) ? jnrlistC : 0;
792 jnrD = (jnrlistD>=0) ? jnrlistD : 0;
793 j_coord_offsetA = DIM*jnrA;
794 j_coord_offsetB = DIM*jnrB;
795 j_coord_offsetC = DIM*jnrC;
796 j_coord_offsetD = DIM*jnrD;
798 /* load j atom coordinates */
799 gmx_mm256_load_1rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
800 x+j_coord_offsetC,x+j_coord_offsetD,
803 /* Calculate displacement vector */
804 dx10 = _mm256_sub_pd(ix1,jx0);
805 dy10 = _mm256_sub_pd(iy1,jy0);
806 dz10 = _mm256_sub_pd(iz1,jz0);
807 dx20 = _mm256_sub_pd(ix2,jx0);
808 dy20 = _mm256_sub_pd(iy2,jy0);
809 dz20 = _mm256_sub_pd(iz2,jz0);
810 dx30 = _mm256_sub_pd(ix3,jx0);
811 dy30 = _mm256_sub_pd(iy3,jy0);
812 dz30 = _mm256_sub_pd(iz3,jz0);
814 /* Calculate squared distance and things based on it */
815 rsq10 = gmx_mm256_calc_rsq_pd(dx10,dy10,dz10);
816 rsq20 = gmx_mm256_calc_rsq_pd(dx20,dy20,dz20);
817 rsq30 = gmx_mm256_calc_rsq_pd(dx30,dy30,dz30);
819 rinv10 = gmx_mm256_invsqrt_pd(rsq10);
820 rinv20 = gmx_mm256_invsqrt_pd(rsq20);
821 rinv30 = gmx_mm256_invsqrt_pd(rsq30);
823 rinvsq10 = _mm256_mul_pd(rinv10,rinv10);
824 rinvsq20 = _mm256_mul_pd(rinv20,rinv20);
825 rinvsq30 = _mm256_mul_pd(rinv30,rinv30);
827 /* Load parameters for j particles */
828 jq0 = gmx_mm256_load_4real_swizzle_pd(charge+jnrA+0,charge+jnrB+0,
829 charge+jnrC+0,charge+jnrD+0);
831 fjx0 = _mm256_setzero_pd();
832 fjy0 = _mm256_setzero_pd();
833 fjz0 = _mm256_setzero_pd();
835 /**************************
836 * CALCULATE INTERACTIONS *
837 **************************/
839 /* Compute parameters for interactions between i and j atoms */
840 qq10 = _mm256_mul_pd(iq1,jq0);
842 /* COULOMB ELECTROSTATICS */
843 velec = _mm256_mul_pd(qq10,rinv10);
844 felec = _mm256_mul_pd(velec,rinvsq10);
848 fscal = _mm256_andnot_pd(dummy_mask,fscal);
850 /* Calculate temporary vectorial force */
851 tx = _mm256_mul_pd(fscal,dx10);
852 ty = _mm256_mul_pd(fscal,dy10);
853 tz = _mm256_mul_pd(fscal,dz10);
855 /* Update vectorial force */
856 fix1 = _mm256_add_pd(fix1,tx);
857 fiy1 = _mm256_add_pd(fiy1,ty);
858 fiz1 = _mm256_add_pd(fiz1,tz);
860 fjx0 = _mm256_add_pd(fjx0,tx);
861 fjy0 = _mm256_add_pd(fjy0,ty);
862 fjz0 = _mm256_add_pd(fjz0,tz);
864 /**************************
865 * CALCULATE INTERACTIONS *
866 **************************/
868 /* Compute parameters for interactions between i and j atoms */
869 qq20 = _mm256_mul_pd(iq2,jq0);
871 /* COULOMB ELECTROSTATICS */
872 velec = _mm256_mul_pd(qq20,rinv20);
873 felec = _mm256_mul_pd(velec,rinvsq20);
877 fscal = _mm256_andnot_pd(dummy_mask,fscal);
879 /* Calculate temporary vectorial force */
880 tx = _mm256_mul_pd(fscal,dx20);
881 ty = _mm256_mul_pd(fscal,dy20);
882 tz = _mm256_mul_pd(fscal,dz20);
884 /* Update vectorial force */
885 fix2 = _mm256_add_pd(fix2,tx);
886 fiy2 = _mm256_add_pd(fiy2,ty);
887 fiz2 = _mm256_add_pd(fiz2,tz);
889 fjx0 = _mm256_add_pd(fjx0,tx);
890 fjy0 = _mm256_add_pd(fjy0,ty);
891 fjz0 = _mm256_add_pd(fjz0,tz);
893 /**************************
894 * CALCULATE INTERACTIONS *
895 **************************/
897 /* Compute parameters for interactions between i and j atoms */
898 qq30 = _mm256_mul_pd(iq3,jq0);
900 /* COULOMB ELECTROSTATICS */
901 velec = _mm256_mul_pd(qq30,rinv30);
902 felec = _mm256_mul_pd(velec,rinvsq30);
906 fscal = _mm256_andnot_pd(dummy_mask,fscal);
908 /* Calculate temporary vectorial force */
909 tx = _mm256_mul_pd(fscal,dx30);
910 ty = _mm256_mul_pd(fscal,dy30);
911 tz = _mm256_mul_pd(fscal,dz30);
913 /* Update vectorial force */
914 fix3 = _mm256_add_pd(fix3,tx);
915 fiy3 = _mm256_add_pd(fiy3,ty);
916 fiz3 = _mm256_add_pd(fiz3,tz);
918 fjx0 = _mm256_add_pd(fjx0,tx);
919 fjy0 = _mm256_add_pd(fjy0,ty);
920 fjz0 = _mm256_add_pd(fjz0,tz);
922 fjptrA = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
923 fjptrB = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
924 fjptrC = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
925 fjptrD = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
927 gmx_mm256_decrement_1rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
929 /* Inner loop uses 81 flops */
932 /* End of innermost loop */
934 gmx_mm256_update_iforce_3atom_swizzle_pd(fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
935 f+i_coord_offset+DIM,fshift+i_shift_offset);
937 /* Increment number of inner iterations */
938 inneriter += j_index_end - j_index_start;
940 /* Outer loop uses 18 flops */
943 /* Increment number of outer iterations */
946 /* Update outer/inner flops */
948 inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_W4_F,outeriter*18 + inneriter*81);