2 * This file is part of the GROMACS molecular simulation package.
4 * Copyright (c) 2012,2013, 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.
44 #include "../nb_kernel.h"
45 #include "types/simple.h"
49 #include "gromacs/simd/math_x86_avx_256_double.h"
50 #include "kernelutil_x86_avx_256_double.h"
53 * Gromacs nonbonded kernel: nb_kernel_ElecCoul_VdwLJ_GeomW3P1_VF_avx_256_double
54 * Electrostatics interaction: Coulomb
55 * VdW interaction: LennardJones
56 * Geometry: Water3-Particle
57 * Calculate force/pot: PotentialAndForce
60 nb_kernel_ElecCoul_VdwLJ_GeomW3P1_VF_avx_256_double
61 (t_nblist * gmx_restrict nlist,
62 rvec * gmx_restrict xx,
63 rvec * gmx_restrict ff,
64 t_forcerec * gmx_restrict fr,
65 t_mdatoms * gmx_restrict mdatoms,
66 nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
67 t_nrnb * gmx_restrict nrnb)
69 /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
70 * just 0 for non-waters.
71 * Suffixes A,B,C,D refer to j loop unrolling done with AVX, e.g. for the four different
72 * jnr indices corresponding to data put in the four positions in the SIMD register.
74 int i_shift_offset,i_coord_offset,outeriter,inneriter;
75 int j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
76 int jnrA,jnrB,jnrC,jnrD;
77 int jnrlistA,jnrlistB,jnrlistC,jnrlistD;
78 int jnrlistE,jnrlistF,jnrlistG,jnrlistH;
79 int j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
80 int *iinr,*jindex,*jjnr,*shiftidx,*gid;
82 real *shiftvec,*fshift,*x,*f;
83 real *fjptrA,*fjptrB,*fjptrC,*fjptrD;
85 __m256d tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
86 real * vdwioffsetptr0;
87 __m256d ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
88 real * vdwioffsetptr1;
89 __m256d ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
90 real * vdwioffsetptr2;
91 __m256d ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
92 int vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
93 __m256d jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
94 __m256d dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
95 __m256d dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
96 __m256d dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
97 __m256d velec,felec,velecsum,facel,crf,krf,krf2;
100 __m256d rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
103 __m256d one_sixth = _mm256_set1_pd(1.0/6.0);
104 __m256d one_twelfth = _mm256_set1_pd(1.0/12.0);
105 __m256d dummy_mask,cutoff_mask;
106 __m128 tmpmask0,tmpmask1;
107 __m256d signbit = _mm256_castsi256_pd( _mm256_set1_epi32(0x80000000) );
108 __m256d one = _mm256_set1_pd(1.0);
109 __m256d two = _mm256_set1_pd(2.0);
115 jindex = nlist->jindex;
117 shiftidx = nlist->shift;
119 shiftvec = fr->shift_vec[0];
120 fshift = fr->fshift[0];
121 facel = _mm256_set1_pd(fr->epsfac);
122 charge = mdatoms->chargeA;
123 nvdwtype = fr->ntype;
125 vdwtype = mdatoms->typeA;
127 /* Setup water-specific parameters */
128 inr = nlist->iinr[0];
129 iq0 = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+0]));
130 iq1 = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+1]));
131 iq2 = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+2]));
132 vdwioffsetptr0 = vdwparam+2*nvdwtype*vdwtype[inr+0];
134 /* Avoid stupid compiler warnings */
135 jnrA = jnrB = jnrC = jnrD = 0;
144 for(iidx=0;iidx<4*DIM;iidx++)
149 /* Start outer loop over neighborlists */
150 for(iidx=0; iidx<nri; iidx++)
152 /* Load shift vector for this list */
153 i_shift_offset = DIM*shiftidx[iidx];
155 /* Load limits for loop over neighbors */
156 j_index_start = jindex[iidx];
157 j_index_end = jindex[iidx+1];
159 /* Get outer coordinate index */
161 i_coord_offset = DIM*inr;
163 /* Load i particle coords and add shift vector */
164 gmx_mm256_load_shift_and_3rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
165 &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
167 fix0 = _mm256_setzero_pd();
168 fiy0 = _mm256_setzero_pd();
169 fiz0 = _mm256_setzero_pd();
170 fix1 = _mm256_setzero_pd();
171 fiy1 = _mm256_setzero_pd();
172 fiz1 = _mm256_setzero_pd();
173 fix2 = _mm256_setzero_pd();
174 fiy2 = _mm256_setzero_pd();
175 fiz2 = _mm256_setzero_pd();
177 /* Reset potential sums */
178 velecsum = _mm256_setzero_pd();
179 vvdwsum = _mm256_setzero_pd();
181 /* Start inner kernel loop */
182 for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
185 /* Get j neighbor index, and coordinate index */
190 j_coord_offsetA = DIM*jnrA;
191 j_coord_offsetB = DIM*jnrB;
192 j_coord_offsetC = DIM*jnrC;
193 j_coord_offsetD = DIM*jnrD;
195 /* load j atom coordinates */
196 gmx_mm256_load_1rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
197 x+j_coord_offsetC,x+j_coord_offsetD,
200 /* Calculate displacement vector */
201 dx00 = _mm256_sub_pd(ix0,jx0);
202 dy00 = _mm256_sub_pd(iy0,jy0);
203 dz00 = _mm256_sub_pd(iz0,jz0);
204 dx10 = _mm256_sub_pd(ix1,jx0);
205 dy10 = _mm256_sub_pd(iy1,jy0);
206 dz10 = _mm256_sub_pd(iz1,jz0);
207 dx20 = _mm256_sub_pd(ix2,jx0);
208 dy20 = _mm256_sub_pd(iy2,jy0);
209 dz20 = _mm256_sub_pd(iz2,jz0);
211 /* Calculate squared distance and things based on it */
212 rsq00 = gmx_mm256_calc_rsq_pd(dx00,dy00,dz00);
213 rsq10 = gmx_mm256_calc_rsq_pd(dx10,dy10,dz10);
214 rsq20 = gmx_mm256_calc_rsq_pd(dx20,dy20,dz20);
216 rinv00 = gmx_mm256_invsqrt_pd(rsq00);
217 rinv10 = gmx_mm256_invsqrt_pd(rsq10);
218 rinv20 = gmx_mm256_invsqrt_pd(rsq20);
220 rinvsq00 = _mm256_mul_pd(rinv00,rinv00);
221 rinvsq10 = _mm256_mul_pd(rinv10,rinv10);
222 rinvsq20 = _mm256_mul_pd(rinv20,rinv20);
224 /* Load parameters for j particles */
225 jq0 = gmx_mm256_load_4real_swizzle_pd(charge+jnrA+0,charge+jnrB+0,
226 charge+jnrC+0,charge+jnrD+0);
227 vdwjidx0A = 2*vdwtype[jnrA+0];
228 vdwjidx0B = 2*vdwtype[jnrB+0];
229 vdwjidx0C = 2*vdwtype[jnrC+0];
230 vdwjidx0D = 2*vdwtype[jnrD+0];
232 fjx0 = _mm256_setzero_pd();
233 fjy0 = _mm256_setzero_pd();
234 fjz0 = _mm256_setzero_pd();
236 /**************************
237 * CALCULATE INTERACTIONS *
238 **************************/
240 /* Compute parameters for interactions between i and j atoms */
241 qq00 = _mm256_mul_pd(iq0,jq0);
242 gmx_mm256_load_4pair_swizzle_pd(vdwioffsetptr0+vdwjidx0A,
243 vdwioffsetptr0+vdwjidx0B,
244 vdwioffsetptr0+vdwjidx0C,
245 vdwioffsetptr0+vdwjidx0D,
248 /* COULOMB ELECTROSTATICS */
249 velec = _mm256_mul_pd(qq00,rinv00);
250 felec = _mm256_mul_pd(velec,rinvsq00);
252 /* LENNARD-JONES DISPERSION/REPULSION */
254 rinvsix = _mm256_mul_pd(_mm256_mul_pd(rinvsq00,rinvsq00),rinvsq00);
255 vvdw6 = _mm256_mul_pd(c6_00,rinvsix);
256 vvdw12 = _mm256_mul_pd(c12_00,_mm256_mul_pd(rinvsix,rinvsix));
257 vvdw = _mm256_sub_pd( _mm256_mul_pd(vvdw12,one_twelfth) , _mm256_mul_pd(vvdw6,one_sixth) );
258 fvdw = _mm256_mul_pd(_mm256_sub_pd(vvdw12,vvdw6),rinvsq00);
260 /* Update potential sum for this i atom from the interaction with this j atom. */
261 velecsum = _mm256_add_pd(velecsum,velec);
262 vvdwsum = _mm256_add_pd(vvdwsum,vvdw);
264 fscal = _mm256_add_pd(felec,fvdw);
266 /* Calculate temporary vectorial force */
267 tx = _mm256_mul_pd(fscal,dx00);
268 ty = _mm256_mul_pd(fscal,dy00);
269 tz = _mm256_mul_pd(fscal,dz00);
271 /* Update vectorial force */
272 fix0 = _mm256_add_pd(fix0,tx);
273 fiy0 = _mm256_add_pd(fiy0,ty);
274 fiz0 = _mm256_add_pd(fiz0,tz);
276 fjx0 = _mm256_add_pd(fjx0,tx);
277 fjy0 = _mm256_add_pd(fjy0,ty);
278 fjz0 = _mm256_add_pd(fjz0,tz);
280 /**************************
281 * CALCULATE INTERACTIONS *
282 **************************/
284 /* Compute parameters for interactions between i and j atoms */
285 qq10 = _mm256_mul_pd(iq1,jq0);
287 /* COULOMB ELECTROSTATICS */
288 velec = _mm256_mul_pd(qq10,rinv10);
289 felec = _mm256_mul_pd(velec,rinvsq10);
291 /* Update potential sum for this i atom from the interaction with this j atom. */
292 velecsum = _mm256_add_pd(velecsum,velec);
296 /* Calculate temporary vectorial force */
297 tx = _mm256_mul_pd(fscal,dx10);
298 ty = _mm256_mul_pd(fscal,dy10);
299 tz = _mm256_mul_pd(fscal,dz10);
301 /* Update vectorial force */
302 fix1 = _mm256_add_pd(fix1,tx);
303 fiy1 = _mm256_add_pd(fiy1,ty);
304 fiz1 = _mm256_add_pd(fiz1,tz);
306 fjx0 = _mm256_add_pd(fjx0,tx);
307 fjy0 = _mm256_add_pd(fjy0,ty);
308 fjz0 = _mm256_add_pd(fjz0,tz);
310 /**************************
311 * CALCULATE INTERACTIONS *
312 **************************/
314 /* Compute parameters for interactions between i and j atoms */
315 qq20 = _mm256_mul_pd(iq2,jq0);
317 /* COULOMB ELECTROSTATICS */
318 velec = _mm256_mul_pd(qq20,rinv20);
319 felec = _mm256_mul_pd(velec,rinvsq20);
321 /* Update potential sum for this i atom from the interaction with this j atom. */
322 velecsum = _mm256_add_pd(velecsum,velec);
326 /* Calculate temporary vectorial force */
327 tx = _mm256_mul_pd(fscal,dx20);
328 ty = _mm256_mul_pd(fscal,dy20);
329 tz = _mm256_mul_pd(fscal,dz20);
331 /* Update vectorial force */
332 fix2 = _mm256_add_pd(fix2,tx);
333 fiy2 = _mm256_add_pd(fiy2,ty);
334 fiz2 = _mm256_add_pd(fiz2,tz);
336 fjx0 = _mm256_add_pd(fjx0,tx);
337 fjy0 = _mm256_add_pd(fjy0,ty);
338 fjz0 = _mm256_add_pd(fjz0,tz);
340 fjptrA = f+j_coord_offsetA;
341 fjptrB = f+j_coord_offsetB;
342 fjptrC = f+j_coord_offsetC;
343 fjptrD = f+j_coord_offsetD;
345 gmx_mm256_decrement_1rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
347 /* Inner loop uses 96 flops */
353 /* Get j neighbor index, and coordinate index */
354 jnrlistA = jjnr[jidx];
355 jnrlistB = jjnr[jidx+1];
356 jnrlistC = jjnr[jidx+2];
357 jnrlistD = jjnr[jidx+3];
358 /* Sign of each element will be negative for non-real atoms.
359 * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
360 * so use it as val = _mm_andnot_pd(mask,val) to clear dummy entries.
362 tmpmask0 = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
364 tmpmask1 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(3,3,2,2));
365 tmpmask0 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(1,1,0,0));
366 dummy_mask = _mm256_castps_pd(gmx_mm256_set_m128(tmpmask1,tmpmask0));
368 jnrA = (jnrlistA>=0) ? jnrlistA : 0;
369 jnrB = (jnrlistB>=0) ? jnrlistB : 0;
370 jnrC = (jnrlistC>=0) ? jnrlistC : 0;
371 jnrD = (jnrlistD>=0) ? jnrlistD : 0;
372 j_coord_offsetA = DIM*jnrA;
373 j_coord_offsetB = DIM*jnrB;
374 j_coord_offsetC = DIM*jnrC;
375 j_coord_offsetD = DIM*jnrD;
377 /* load j atom coordinates */
378 gmx_mm256_load_1rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
379 x+j_coord_offsetC,x+j_coord_offsetD,
382 /* Calculate displacement vector */
383 dx00 = _mm256_sub_pd(ix0,jx0);
384 dy00 = _mm256_sub_pd(iy0,jy0);
385 dz00 = _mm256_sub_pd(iz0,jz0);
386 dx10 = _mm256_sub_pd(ix1,jx0);
387 dy10 = _mm256_sub_pd(iy1,jy0);
388 dz10 = _mm256_sub_pd(iz1,jz0);
389 dx20 = _mm256_sub_pd(ix2,jx0);
390 dy20 = _mm256_sub_pd(iy2,jy0);
391 dz20 = _mm256_sub_pd(iz2,jz0);
393 /* Calculate squared distance and things based on it */
394 rsq00 = gmx_mm256_calc_rsq_pd(dx00,dy00,dz00);
395 rsq10 = gmx_mm256_calc_rsq_pd(dx10,dy10,dz10);
396 rsq20 = gmx_mm256_calc_rsq_pd(dx20,dy20,dz20);
398 rinv00 = gmx_mm256_invsqrt_pd(rsq00);
399 rinv10 = gmx_mm256_invsqrt_pd(rsq10);
400 rinv20 = gmx_mm256_invsqrt_pd(rsq20);
402 rinvsq00 = _mm256_mul_pd(rinv00,rinv00);
403 rinvsq10 = _mm256_mul_pd(rinv10,rinv10);
404 rinvsq20 = _mm256_mul_pd(rinv20,rinv20);
406 /* Load parameters for j particles */
407 jq0 = gmx_mm256_load_4real_swizzle_pd(charge+jnrA+0,charge+jnrB+0,
408 charge+jnrC+0,charge+jnrD+0);
409 vdwjidx0A = 2*vdwtype[jnrA+0];
410 vdwjidx0B = 2*vdwtype[jnrB+0];
411 vdwjidx0C = 2*vdwtype[jnrC+0];
412 vdwjidx0D = 2*vdwtype[jnrD+0];
414 fjx0 = _mm256_setzero_pd();
415 fjy0 = _mm256_setzero_pd();
416 fjz0 = _mm256_setzero_pd();
418 /**************************
419 * CALCULATE INTERACTIONS *
420 **************************/
422 /* Compute parameters for interactions between i and j atoms */
423 qq00 = _mm256_mul_pd(iq0,jq0);
424 gmx_mm256_load_4pair_swizzle_pd(vdwioffsetptr0+vdwjidx0A,
425 vdwioffsetptr0+vdwjidx0B,
426 vdwioffsetptr0+vdwjidx0C,
427 vdwioffsetptr0+vdwjidx0D,
430 /* COULOMB ELECTROSTATICS */
431 velec = _mm256_mul_pd(qq00,rinv00);
432 felec = _mm256_mul_pd(velec,rinvsq00);
434 /* LENNARD-JONES DISPERSION/REPULSION */
436 rinvsix = _mm256_mul_pd(_mm256_mul_pd(rinvsq00,rinvsq00),rinvsq00);
437 vvdw6 = _mm256_mul_pd(c6_00,rinvsix);
438 vvdw12 = _mm256_mul_pd(c12_00,_mm256_mul_pd(rinvsix,rinvsix));
439 vvdw = _mm256_sub_pd( _mm256_mul_pd(vvdw12,one_twelfth) , _mm256_mul_pd(vvdw6,one_sixth) );
440 fvdw = _mm256_mul_pd(_mm256_sub_pd(vvdw12,vvdw6),rinvsq00);
442 /* Update potential sum for this i atom from the interaction with this j atom. */
443 velec = _mm256_andnot_pd(dummy_mask,velec);
444 velecsum = _mm256_add_pd(velecsum,velec);
445 vvdw = _mm256_andnot_pd(dummy_mask,vvdw);
446 vvdwsum = _mm256_add_pd(vvdwsum,vvdw);
448 fscal = _mm256_add_pd(felec,fvdw);
450 fscal = _mm256_andnot_pd(dummy_mask,fscal);
452 /* Calculate temporary vectorial force */
453 tx = _mm256_mul_pd(fscal,dx00);
454 ty = _mm256_mul_pd(fscal,dy00);
455 tz = _mm256_mul_pd(fscal,dz00);
457 /* Update vectorial force */
458 fix0 = _mm256_add_pd(fix0,tx);
459 fiy0 = _mm256_add_pd(fiy0,ty);
460 fiz0 = _mm256_add_pd(fiz0,tz);
462 fjx0 = _mm256_add_pd(fjx0,tx);
463 fjy0 = _mm256_add_pd(fjy0,ty);
464 fjz0 = _mm256_add_pd(fjz0,tz);
466 /**************************
467 * CALCULATE INTERACTIONS *
468 **************************/
470 /* Compute parameters for interactions between i and j atoms */
471 qq10 = _mm256_mul_pd(iq1,jq0);
473 /* COULOMB ELECTROSTATICS */
474 velec = _mm256_mul_pd(qq10,rinv10);
475 felec = _mm256_mul_pd(velec,rinvsq10);
477 /* Update potential sum for this i atom from the interaction with this j atom. */
478 velec = _mm256_andnot_pd(dummy_mask,velec);
479 velecsum = _mm256_add_pd(velecsum,velec);
483 fscal = _mm256_andnot_pd(dummy_mask,fscal);
485 /* Calculate temporary vectorial force */
486 tx = _mm256_mul_pd(fscal,dx10);
487 ty = _mm256_mul_pd(fscal,dy10);
488 tz = _mm256_mul_pd(fscal,dz10);
490 /* Update vectorial force */
491 fix1 = _mm256_add_pd(fix1,tx);
492 fiy1 = _mm256_add_pd(fiy1,ty);
493 fiz1 = _mm256_add_pd(fiz1,tz);
495 fjx0 = _mm256_add_pd(fjx0,tx);
496 fjy0 = _mm256_add_pd(fjy0,ty);
497 fjz0 = _mm256_add_pd(fjz0,tz);
499 /**************************
500 * CALCULATE INTERACTIONS *
501 **************************/
503 /* Compute parameters for interactions between i and j atoms */
504 qq20 = _mm256_mul_pd(iq2,jq0);
506 /* COULOMB ELECTROSTATICS */
507 velec = _mm256_mul_pd(qq20,rinv20);
508 felec = _mm256_mul_pd(velec,rinvsq20);
510 /* Update potential sum for this i atom from the interaction with this j atom. */
511 velec = _mm256_andnot_pd(dummy_mask,velec);
512 velecsum = _mm256_add_pd(velecsum,velec);
516 fscal = _mm256_andnot_pd(dummy_mask,fscal);
518 /* Calculate temporary vectorial force */
519 tx = _mm256_mul_pd(fscal,dx20);
520 ty = _mm256_mul_pd(fscal,dy20);
521 tz = _mm256_mul_pd(fscal,dz20);
523 /* Update vectorial force */
524 fix2 = _mm256_add_pd(fix2,tx);
525 fiy2 = _mm256_add_pd(fiy2,ty);
526 fiz2 = _mm256_add_pd(fiz2,tz);
528 fjx0 = _mm256_add_pd(fjx0,tx);
529 fjy0 = _mm256_add_pd(fjy0,ty);
530 fjz0 = _mm256_add_pd(fjz0,tz);
532 fjptrA = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
533 fjptrB = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
534 fjptrC = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
535 fjptrD = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
537 gmx_mm256_decrement_1rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
539 /* Inner loop uses 96 flops */
542 /* End of innermost loop */
544 gmx_mm256_update_iforce_3atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
545 f+i_coord_offset,fshift+i_shift_offset);
548 /* Update potential energies */
549 gmx_mm256_update_1pot_pd(velecsum,kernel_data->energygrp_elec+ggid);
550 gmx_mm256_update_1pot_pd(vvdwsum,kernel_data->energygrp_vdw+ggid);
552 /* Increment number of inner iterations */
553 inneriter += j_index_end - j_index_start;
555 /* Outer loop uses 20 flops */
558 /* Increment number of outer iterations */
561 /* Update outer/inner flops */
563 inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3_VF,outeriter*20 + inneriter*96);
566 * Gromacs nonbonded kernel: nb_kernel_ElecCoul_VdwLJ_GeomW3P1_F_avx_256_double
567 * Electrostatics interaction: Coulomb
568 * VdW interaction: LennardJones
569 * Geometry: Water3-Particle
570 * Calculate force/pot: Force
573 nb_kernel_ElecCoul_VdwLJ_GeomW3P1_F_avx_256_double
574 (t_nblist * gmx_restrict nlist,
575 rvec * gmx_restrict xx,
576 rvec * gmx_restrict ff,
577 t_forcerec * gmx_restrict fr,
578 t_mdatoms * gmx_restrict mdatoms,
579 nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
580 t_nrnb * gmx_restrict nrnb)
582 /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
583 * just 0 for non-waters.
584 * Suffixes A,B,C,D refer to j loop unrolling done with AVX, e.g. for the four different
585 * jnr indices corresponding to data put in the four positions in the SIMD register.
587 int i_shift_offset,i_coord_offset,outeriter,inneriter;
588 int j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
589 int jnrA,jnrB,jnrC,jnrD;
590 int jnrlistA,jnrlistB,jnrlistC,jnrlistD;
591 int jnrlistE,jnrlistF,jnrlistG,jnrlistH;
592 int j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
593 int *iinr,*jindex,*jjnr,*shiftidx,*gid;
595 real *shiftvec,*fshift,*x,*f;
596 real *fjptrA,*fjptrB,*fjptrC,*fjptrD;
598 __m256d tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
599 real * vdwioffsetptr0;
600 __m256d ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
601 real * vdwioffsetptr1;
602 __m256d ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
603 real * vdwioffsetptr2;
604 __m256d ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
605 int vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
606 __m256d jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
607 __m256d dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
608 __m256d dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
609 __m256d dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
610 __m256d velec,felec,velecsum,facel,crf,krf,krf2;
613 __m256d rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
616 __m256d one_sixth = _mm256_set1_pd(1.0/6.0);
617 __m256d one_twelfth = _mm256_set1_pd(1.0/12.0);
618 __m256d dummy_mask,cutoff_mask;
619 __m128 tmpmask0,tmpmask1;
620 __m256d signbit = _mm256_castsi256_pd( _mm256_set1_epi32(0x80000000) );
621 __m256d one = _mm256_set1_pd(1.0);
622 __m256d two = _mm256_set1_pd(2.0);
628 jindex = nlist->jindex;
630 shiftidx = nlist->shift;
632 shiftvec = fr->shift_vec[0];
633 fshift = fr->fshift[0];
634 facel = _mm256_set1_pd(fr->epsfac);
635 charge = mdatoms->chargeA;
636 nvdwtype = fr->ntype;
638 vdwtype = mdatoms->typeA;
640 /* Setup water-specific parameters */
641 inr = nlist->iinr[0];
642 iq0 = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+0]));
643 iq1 = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+1]));
644 iq2 = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+2]));
645 vdwioffsetptr0 = vdwparam+2*nvdwtype*vdwtype[inr+0];
647 /* Avoid stupid compiler warnings */
648 jnrA = jnrB = jnrC = jnrD = 0;
657 for(iidx=0;iidx<4*DIM;iidx++)
662 /* Start outer loop over neighborlists */
663 for(iidx=0; iidx<nri; iidx++)
665 /* Load shift vector for this list */
666 i_shift_offset = DIM*shiftidx[iidx];
668 /* Load limits for loop over neighbors */
669 j_index_start = jindex[iidx];
670 j_index_end = jindex[iidx+1];
672 /* Get outer coordinate index */
674 i_coord_offset = DIM*inr;
676 /* Load i particle coords and add shift vector */
677 gmx_mm256_load_shift_and_3rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
678 &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
680 fix0 = _mm256_setzero_pd();
681 fiy0 = _mm256_setzero_pd();
682 fiz0 = _mm256_setzero_pd();
683 fix1 = _mm256_setzero_pd();
684 fiy1 = _mm256_setzero_pd();
685 fiz1 = _mm256_setzero_pd();
686 fix2 = _mm256_setzero_pd();
687 fiy2 = _mm256_setzero_pd();
688 fiz2 = _mm256_setzero_pd();
690 /* Start inner kernel loop */
691 for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
694 /* Get j neighbor index, and coordinate index */
699 j_coord_offsetA = DIM*jnrA;
700 j_coord_offsetB = DIM*jnrB;
701 j_coord_offsetC = DIM*jnrC;
702 j_coord_offsetD = DIM*jnrD;
704 /* load j atom coordinates */
705 gmx_mm256_load_1rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
706 x+j_coord_offsetC,x+j_coord_offsetD,
709 /* Calculate displacement vector */
710 dx00 = _mm256_sub_pd(ix0,jx0);
711 dy00 = _mm256_sub_pd(iy0,jy0);
712 dz00 = _mm256_sub_pd(iz0,jz0);
713 dx10 = _mm256_sub_pd(ix1,jx0);
714 dy10 = _mm256_sub_pd(iy1,jy0);
715 dz10 = _mm256_sub_pd(iz1,jz0);
716 dx20 = _mm256_sub_pd(ix2,jx0);
717 dy20 = _mm256_sub_pd(iy2,jy0);
718 dz20 = _mm256_sub_pd(iz2,jz0);
720 /* Calculate squared distance and things based on it */
721 rsq00 = gmx_mm256_calc_rsq_pd(dx00,dy00,dz00);
722 rsq10 = gmx_mm256_calc_rsq_pd(dx10,dy10,dz10);
723 rsq20 = gmx_mm256_calc_rsq_pd(dx20,dy20,dz20);
725 rinv00 = gmx_mm256_invsqrt_pd(rsq00);
726 rinv10 = gmx_mm256_invsqrt_pd(rsq10);
727 rinv20 = gmx_mm256_invsqrt_pd(rsq20);
729 rinvsq00 = _mm256_mul_pd(rinv00,rinv00);
730 rinvsq10 = _mm256_mul_pd(rinv10,rinv10);
731 rinvsq20 = _mm256_mul_pd(rinv20,rinv20);
733 /* Load parameters for j particles */
734 jq0 = gmx_mm256_load_4real_swizzle_pd(charge+jnrA+0,charge+jnrB+0,
735 charge+jnrC+0,charge+jnrD+0);
736 vdwjidx0A = 2*vdwtype[jnrA+0];
737 vdwjidx0B = 2*vdwtype[jnrB+0];
738 vdwjidx0C = 2*vdwtype[jnrC+0];
739 vdwjidx0D = 2*vdwtype[jnrD+0];
741 fjx0 = _mm256_setzero_pd();
742 fjy0 = _mm256_setzero_pd();
743 fjz0 = _mm256_setzero_pd();
745 /**************************
746 * CALCULATE INTERACTIONS *
747 **************************/
749 /* Compute parameters for interactions between i and j atoms */
750 qq00 = _mm256_mul_pd(iq0,jq0);
751 gmx_mm256_load_4pair_swizzle_pd(vdwioffsetptr0+vdwjidx0A,
752 vdwioffsetptr0+vdwjidx0B,
753 vdwioffsetptr0+vdwjidx0C,
754 vdwioffsetptr0+vdwjidx0D,
757 /* COULOMB ELECTROSTATICS */
758 velec = _mm256_mul_pd(qq00,rinv00);
759 felec = _mm256_mul_pd(velec,rinvsq00);
761 /* LENNARD-JONES DISPERSION/REPULSION */
763 rinvsix = _mm256_mul_pd(_mm256_mul_pd(rinvsq00,rinvsq00),rinvsq00);
764 fvdw = _mm256_mul_pd(_mm256_sub_pd(_mm256_mul_pd(c12_00,rinvsix),c6_00),_mm256_mul_pd(rinvsix,rinvsq00));
766 fscal = _mm256_add_pd(felec,fvdw);
768 /* Calculate temporary vectorial force */
769 tx = _mm256_mul_pd(fscal,dx00);
770 ty = _mm256_mul_pd(fscal,dy00);
771 tz = _mm256_mul_pd(fscal,dz00);
773 /* Update vectorial force */
774 fix0 = _mm256_add_pd(fix0,tx);
775 fiy0 = _mm256_add_pd(fiy0,ty);
776 fiz0 = _mm256_add_pd(fiz0,tz);
778 fjx0 = _mm256_add_pd(fjx0,tx);
779 fjy0 = _mm256_add_pd(fjy0,ty);
780 fjz0 = _mm256_add_pd(fjz0,tz);
782 /**************************
783 * CALCULATE INTERACTIONS *
784 **************************/
786 /* Compute parameters for interactions between i and j atoms */
787 qq10 = _mm256_mul_pd(iq1,jq0);
789 /* COULOMB ELECTROSTATICS */
790 velec = _mm256_mul_pd(qq10,rinv10);
791 felec = _mm256_mul_pd(velec,rinvsq10);
795 /* Calculate temporary vectorial force */
796 tx = _mm256_mul_pd(fscal,dx10);
797 ty = _mm256_mul_pd(fscal,dy10);
798 tz = _mm256_mul_pd(fscal,dz10);
800 /* Update vectorial force */
801 fix1 = _mm256_add_pd(fix1,tx);
802 fiy1 = _mm256_add_pd(fiy1,ty);
803 fiz1 = _mm256_add_pd(fiz1,tz);
805 fjx0 = _mm256_add_pd(fjx0,tx);
806 fjy0 = _mm256_add_pd(fjy0,ty);
807 fjz0 = _mm256_add_pd(fjz0,tz);
809 /**************************
810 * CALCULATE INTERACTIONS *
811 **************************/
813 /* Compute parameters for interactions between i and j atoms */
814 qq20 = _mm256_mul_pd(iq2,jq0);
816 /* COULOMB ELECTROSTATICS */
817 velec = _mm256_mul_pd(qq20,rinv20);
818 felec = _mm256_mul_pd(velec,rinvsq20);
822 /* Calculate temporary vectorial force */
823 tx = _mm256_mul_pd(fscal,dx20);
824 ty = _mm256_mul_pd(fscal,dy20);
825 tz = _mm256_mul_pd(fscal,dz20);
827 /* Update vectorial force */
828 fix2 = _mm256_add_pd(fix2,tx);
829 fiy2 = _mm256_add_pd(fiy2,ty);
830 fiz2 = _mm256_add_pd(fiz2,tz);
832 fjx0 = _mm256_add_pd(fjx0,tx);
833 fjy0 = _mm256_add_pd(fjy0,ty);
834 fjz0 = _mm256_add_pd(fjz0,tz);
836 fjptrA = f+j_coord_offsetA;
837 fjptrB = f+j_coord_offsetB;
838 fjptrC = f+j_coord_offsetC;
839 fjptrD = f+j_coord_offsetD;
841 gmx_mm256_decrement_1rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
843 /* Inner loop uses 88 flops */
849 /* Get j neighbor index, and coordinate index */
850 jnrlistA = jjnr[jidx];
851 jnrlistB = jjnr[jidx+1];
852 jnrlistC = jjnr[jidx+2];
853 jnrlistD = jjnr[jidx+3];
854 /* Sign of each element will be negative for non-real atoms.
855 * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
856 * so use it as val = _mm_andnot_pd(mask,val) to clear dummy entries.
858 tmpmask0 = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
860 tmpmask1 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(3,3,2,2));
861 tmpmask0 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(1,1,0,0));
862 dummy_mask = _mm256_castps_pd(gmx_mm256_set_m128(tmpmask1,tmpmask0));
864 jnrA = (jnrlistA>=0) ? jnrlistA : 0;
865 jnrB = (jnrlistB>=0) ? jnrlistB : 0;
866 jnrC = (jnrlistC>=0) ? jnrlistC : 0;
867 jnrD = (jnrlistD>=0) ? jnrlistD : 0;
868 j_coord_offsetA = DIM*jnrA;
869 j_coord_offsetB = DIM*jnrB;
870 j_coord_offsetC = DIM*jnrC;
871 j_coord_offsetD = DIM*jnrD;
873 /* load j atom coordinates */
874 gmx_mm256_load_1rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
875 x+j_coord_offsetC,x+j_coord_offsetD,
878 /* Calculate displacement vector */
879 dx00 = _mm256_sub_pd(ix0,jx0);
880 dy00 = _mm256_sub_pd(iy0,jy0);
881 dz00 = _mm256_sub_pd(iz0,jz0);
882 dx10 = _mm256_sub_pd(ix1,jx0);
883 dy10 = _mm256_sub_pd(iy1,jy0);
884 dz10 = _mm256_sub_pd(iz1,jz0);
885 dx20 = _mm256_sub_pd(ix2,jx0);
886 dy20 = _mm256_sub_pd(iy2,jy0);
887 dz20 = _mm256_sub_pd(iz2,jz0);
889 /* Calculate squared distance and things based on it */
890 rsq00 = gmx_mm256_calc_rsq_pd(dx00,dy00,dz00);
891 rsq10 = gmx_mm256_calc_rsq_pd(dx10,dy10,dz10);
892 rsq20 = gmx_mm256_calc_rsq_pd(dx20,dy20,dz20);
894 rinv00 = gmx_mm256_invsqrt_pd(rsq00);
895 rinv10 = gmx_mm256_invsqrt_pd(rsq10);
896 rinv20 = gmx_mm256_invsqrt_pd(rsq20);
898 rinvsq00 = _mm256_mul_pd(rinv00,rinv00);
899 rinvsq10 = _mm256_mul_pd(rinv10,rinv10);
900 rinvsq20 = _mm256_mul_pd(rinv20,rinv20);
902 /* Load parameters for j particles */
903 jq0 = gmx_mm256_load_4real_swizzle_pd(charge+jnrA+0,charge+jnrB+0,
904 charge+jnrC+0,charge+jnrD+0);
905 vdwjidx0A = 2*vdwtype[jnrA+0];
906 vdwjidx0B = 2*vdwtype[jnrB+0];
907 vdwjidx0C = 2*vdwtype[jnrC+0];
908 vdwjidx0D = 2*vdwtype[jnrD+0];
910 fjx0 = _mm256_setzero_pd();
911 fjy0 = _mm256_setzero_pd();
912 fjz0 = _mm256_setzero_pd();
914 /**************************
915 * CALCULATE INTERACTIONS *
916 **************************/
918 /* Compute parameters for interactions between i and j atoms */
919 qq00 = _mm256_mul_pd(iq0,jq0);
920 gmx_mm256_load_4pair_swizzle_pd(vdwioffsetptr0+vdwjidx0A,
921 vdwioffsetptr0+vdwjidx0B,
922 vdwioffsetptr0+vdwjidx0C,
923 vdwioffsetptr0+vdwjidx0D,
926 /* COULOMB ELECTROSTATICS */
927 velec = _mm256_mul_pd(qq00,rinv00);
928 felec = _mm256_mul_pd(velec,rinvsq00);
930 /* LENNARD-JONES DISPERSION/REPULSION */
932 rinvsix = _mm256_mul_pd(_mm256_mul_pd(rinvsq00,rinvsq00),rinvsq00);
933 fvdw = _mm256_mul_pd(_mm256_sub_pd(_mm256_mul_pd(c12_00,rinvsix),c6_00),_mm256_mul_pd(rinvsix,rinvsq00));
935 fscal = _mm256_add_pd(felec,fvdw);
937 fscal = _mm256_andnot_pd(dummy_mask,fscal);
939 /* Calculate temporary vectorial force */
940 tx = _mm256_mul_pd(fscal,dx00);
941 ty = _mm256_mul_pd(fscal,dy00);
942 tz = _mm256_mul_pd(fscal,dz00);
944 /* Update vectorial force */
945 fix0 = _mm256_add_pd(fix0,tx);
946 fiy0 = _mm256_add_pd(fiy0,ty);
947 fiz0 = _mm256_add_pd(fiz0,tz);
949 fjx0 = _mm256_add_pd(fjx0,tx);
950 fjy0 = _mm256_add_pd(fjy0,ty);
951 fjz0 = _mm256_add_pd(fjz0,tz);
953 /**************************
954 * CALCULATE INTERACTIONS *
955 **************************/
957 /* Compute parameters for interactions between i and j atoms */
958 qq10 = _mm256_mul_pd(iq1,jq0);
960 /* COULOMB ELECTROSTATICS */
961 velec = _mm256_mul_pd(qq10,rinv10);
962 felec = _mm256_mul_pd(velec,rinvsq10);
966 fscal = _mm256_andnot_pd(dummy_mask,fscal);
968 /* Calculate temporary vectorial force */
969 tx = _mm256_mul_pd(fscal,dx10);
970 ty = _mm256_mul_pd(fscal,dy10);
971 tz = _mm256_mul_pd(fscal,dz10);
973 /* Update vectorial force */
974 fix1 = _mm256_add_pd(fix1,tx);
975 fiy1 = _mm256_add_pd(fiy1,ty);
976 fiz1 = _mm256_add_pd(fiz1,tz);
978 fjx0 = _mm256_add_pd(fjx0,tx);
979 fjy0 = _mm256_add_pd(fjy0,ty);
980 fjz0 = _mm256_add_pd(fjz0,tz);
982 /**************************
983 * CALCULATE INTERACTIONS *
984 **************************/
986 /* Compute parameters for interactions between i and j atoms */
987 qq20 = _mm256_mul_pd(iq2,jq0);
989 /* COULOMB ELECTROSTATICS */
990 velec = _mm256_mul_pd(qq20,rinv20);
991 felec = _mm256_mul_pd(velec,rinvsq20);
995 fscal = _mm256_andnot_pd(dummy_mask,fscal);
997 /* Calculate temporary vectorial force */
998 tx = _mm256_mul_pd(fscal,dx20);
999 ty = _mm256_mul_pd(fscal,dy20);
1000 tz = _mm256_mul_pd(fscal,dz20);
1002 /* Update vectorial force */
1003 fix2 = _mm256_add_pd(fix2,tx);
1004 fiy2 = _mm256_add_pd(fiy2,ty);
1005 fiz2 = _mm256_add_pd(fiz2,tz);
1007 fjx0 = _mm256_add_pd(fjx0,tx);
1008 fjy0 = _mm256_add_pd(fjy0,ty);
1009 fjz0 = _mm256_add_pd(fjz0,tz);
1011 fjptrA = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
1012 fjptrB = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
1013 fjptrC = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
1014 fjptrD = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
1016 gmx_mm256_decrement_1rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
1018 /* Inner loop uses 88 flops */
1021 /* End of innermost loop */
1023 gmx_mm256_update_iforce_3atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
1024 f+i_coord_offset,fshift+i_shift_offset);
1026 /* Increment number of inner iterations */
1027 inneriter += j_index_end - j_index_start;
1029 /* Outer loop uses 18 flops */
1032 /* Increment number of outer iterations */
1035 /* Update outer/inner flops */
1037 inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3_F,outeriter*18 + inneriter*88);