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_single kernel generator.
44 #include "../nb_kernel.h"
45 #include "types/simple.h"
49 #include "gromacs/simd/math_x86_avx_256_single.h"
50 #include "kernelutil_x86_avx_256_single.h"
53 * Gromacs nonbonded kernel: nb_kernel_ElecRF_VdwNone_GeomW4P1_VF_avx_256_single
54 * Electrostatics interaction: ReactionField
55 * VdW interaction: None
56 * Geometry: Water4-Particle
57 * Calculate force/pot: PotentialAndForce
60 nb_kernel_ElecRF_VdwNone_GeomW4P1_VF_avx_256_single
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,E,F,G,H refer to j loop unrolling done with AVX, e.g. for the eight 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 jnrE,jnrF,jnrG,jnrH;
78 int jnrlistA,jnrlistB,jnrlistC,jnrlistD;
79 int jnrlistE,jnrlistF,jnrlistG,jnrlistH;
80 int j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
81 int j_coord_offsetE,j_coord_offsetF,j_coord_offsetG,j_coord_offsetH;
82 int *iinr,*jindex,*jjnr,*shiftidx,*gid;
84 real *shiftvec,*fshift,*x,*f;
85 real *fjptrA,*fjptrB,*fjptrC,*fjptrD,*fjptrE,*fjptrF,*fjptrG,*fjptrH;
87 __m256 tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
88 real * vdwioffsetptr1;
89 __m256 ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
90 real * vdwioffsetptr2;
91 __m256 ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
92 real * vdwioffsetptr3;
93 __m256 ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
94 int vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D,vdwjidx0E,vdwjidx0F,vdwjidx0G,vdwjidx0H;
95 __m256 jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
96 __m256 dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
97 __m256 dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
98 __m256 dx30,dy30,dz30,rsq30,rinv30,rinvsq30,r30,qq30,c6_30,c12_30;
99 __m256 velec,felec,velecsum,facel,crf,krf,krf2;
101 __m256 dummy_mask,cutoff_mask;
102 __m256 signbit = _mm256_castsi256_ps( _mm256_set1_epi32(0x80000000) );
103 __m256 one = _mm256_set1_ps(1.0);
104 __m256 two = _mm256_set1_ps(2.0);
110 jindex = nlist->jindex;
112 shiftidx = nlist->shift;
114 shiftvec = fr->shift_vec[0];
115 fshift = fr->fshift[0];
116 facel = _mm256_set1_ps(fr->epsfac);
117 charge = mdatoms->chargeA;
118 krf = _mm256_set1_ps(fr->ic->k_rf);
119 krf2 = _mm256_set1_ps(fr->ic->k_rf*2.0);
120 crf = _mm256_set1_ps(fr->ic->c_rf);
122 /* Setup water-specific parameters */
123 inr = nlist->iinr[0];
124 iq1 = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+1]));
125 iq2 = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+2]));
126 iq3 = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+3]));
128 /* Avoid stupid compiler warnings */
129 jnrA = jnrB = jnrC = jnrD = jnrE = jnrF = jnrG = jnrH = 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_ps(shiftvec+i_shift_offset,x+i_coord_offset+DIM,
163 &ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
165 fix1 = _mm256_setzero_ps();
166 fiy1 = _mm256_setzero_ps();
167 fiz1 = _mm256_setzero_ps();
168 fix2 = _mm256_setzero_ps();
169 fiy2 = _mm256_setzero_ps();
170 fiz2 = _mm256_setzero_ps();
171 fix3 = _mm256_setzero_ps();
172 fiy3 = _mm256_setzero_ps();
173 fiz3 = _mm256_setzero_ps();
175 /* Reset potential sums */
176 velecsum = _mm256_setzero_ps();
178 /* Start inner kernel loop */
179 for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+7]>=0; jidx+=8)
182 /* Get j neighbor index, and coordinate index */
191 j_coord_offsetA = DIM*jnrA;
192 j_coord_offsetB = DIM*jnrB;
193 j_coord_offsetC = DIM*jnrC;
194 j_coord_offsetD = DIM*jnrD;
195 j_coord_offsetE = DIM*jnrE;
196 j_coord_offsetF = DIM*jnrF;
197 j_coord_offsetG = DIM*jnrG;
198 j_coord_offsetH = DIM*jnrH;
200 /* load j atom coordinates */
201 gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
202 x+j_coord_offsetC,x+j_coord_offsetD,
203 x+j_coord_offsetE,x+j_coord_offsetF,
204 x+j_coord_offsetG,x+j_coord_offsetH,
207 /* Calculate displacement vector */
208 dx10 = _mm256_sub_ps(ix1,jx0);
209 dy10 = _mm256_sub_ps(iy1,jy0);
210 dz10 = _mm256_sub_ps(iz1,jz0);
211 dx20 = _mm256_sub_ps(ix2,jx0);
212 dy20 = _mm256_sub_ps(iy2,jy0);
213 dz20 = _mm256_sub_ps(iz2,jz0);
214 dx30 = _mm256_sub_ps(ix3,jx0);
215 dy30 = _mm256_sub_ps(iy3,jy0);
216 dz30 = _mm256_sub_ps(iz3,jz0);
218 /* Calculate squared distance and things based on it */
219 rsq10 = gmx_mm256_calc_rsq_ps(dx10,dy10,dz10);
220 rsq20 = gmx_mm256_calc_rsq_ps(dx20,dy20,dz20);
221 rsq30 = gmx_mm256_calc_rsq_ps(dx30,dy30,dz30);
223 rinv10 = gmx_mm256_invsqrt_ps(rsq10);
224 rinv20 = gmx_mm256_invsqrt_ps(rsq20);
225 rinv30 = gmx_mm256_invsqrt_ps(rsq30);
227 rinvsq10 = _mm256_mul_ps(rinv10,rinv10);
228 rinvsq20 = _mm256_mul_ps(rinv20,rinv20);
229 rinvsq30 = _mm256_mul_ps(rinv30,rinv30);
231 /* Load parameters for j particles */
232 jq0 = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
233 charge+jnrC+0,charge+jnrD+0,
234 charge+jnrE+0,charge+jnrF+0,
235 charge+jnrG+0,charge+jnrH+0);
237 fjx0 = _mm256_setzero_ps();
238 fjy0 = _mm256_setzero_ps();
239 fjz0 = _mm256_setzero_ps();
241 /**************************
242 * CALCULATE INTERACTIONS *
243 **************************/
245 /* Compute parameters for interactions between i and j atoms */
246 qq10 = _mm256_mul_ps(iq1,jq0);
248 /* REACTION-FIELD ELECTROSTATICS */
249 velec = _mm256_mul_ps(qq10,_mm256_sub_ps(_mm256_add_ps(rinv10,_mm256_mul_ps(krf,rsq10)),crf));
250 felec = _mm256_mul_ps(qq10,_mm256_sub_ps(_mm256_mul_ps(rinv10,rinvsq10),krf2));
252 /* Update potential sum for this i atom from the interaction with this j atom. */
253 velecsum = _mm256_add_ps(velecsum,velec);
257 /* Calculate temporary vectorial force */
258 tx = _mm256_mul_ps(fscal,dx10);
259 ty = _mm256_mul_ps(fscal,dy10);
260 tz = _mm256_mul_ps(fscal,dz10);
262 /* Update vectorial force */
263 fix1 = _mm256_add_ps(fix1,tx);
264 fiy1 = _mm256_add_ps(fiy1,ty);
265 fiz1 = _mm256_add_ps(fiz1,tz);
267 fjx0 = _mm256_add_ps(fjx0,tx);
268 fjy0 = _mm256_add_ps(fjy0,ty);
269 fjz0 = _mm256_add_ps(fjz0,tz);
271 /**************************
272 * CALCULATE INTERACTIONS *
273 **************************/
275 /* Compute parameters for interactions between i and j atoms */
276 qq20 = _mm256_mul_ps(iq2,jq0);
278 /* REACTION-FIELD ELECTROSTATICS */
279 velec = _mm256_mul_ps(qq20,_mm256_sub_ps(_mm256_add_ps(rinv20,_mm256_mul_ps(krf,rsq20)),crf));
280 felec = _mm256_mul_ps(qq20,_mm256_sub_ps(_mm256_mul_ps(rinv20,rinvsq20),krf2));
282 /* Update potential sum for this i atom from the interaction with this j atom. */
283 velecsum = _mm256_add_ps(velecsum,velec);
287 /* Calculate temporary vectorial force */
288 tx = _mm256_mul_ps(fscal,dx20);
289 ty = _mm256_mul_ps(fscal,dy20);
290 tz = _mm256_mul_ps(fscal,dz20);
292 /* Update vectorial force */
293 fix2 = _mm256_add_ps(fix2,tx);
294 fiy2 = _mm256_add_ps(fiy2,ty);
295 fiz2 = _mm256_add_ps(fiz2,tz);
297 fjx0 = _mm256_add_ps(fjx0,tx);
298 fjy0 = _mm256_add_ps(fjy0,ty);
299 fjz0 = _mm256_add_ps(fjz0,tz);
301 /**************************
302 * CALCULATE INTERACTIONS *
303 **************************/
305 /* Compute parameters for interactions between i and j atoms */
306 qq30 = _mm256_mul_ps(iq3,jq0);
308 /* REACTION-FIELD ELECTROSTATICS */
309 velec = _mm256_mul_ps(qq30,_mm256_sub_ps(_mm256_add_ps(rinv30,_mm256_mul_ps(krf,rsq30)),crf));
310 felec = _mm256_mul_ps(qq30,_mm256_sub_ps(_mm256_mul_ps(rinv30,rinvsq30),krf2));
312 /* Update potential sum for this i atom from the interaction with this j atom. */
313 velecsum = _mm256_add_ps(velecsum,velec);
317 /* Calculate temporary vectorial force */
318 tx = _mm256_mul_ps(fscal,dx30);
319 ty = _mm256_mul_ps(fscal,dy30);
320 tz = _mm256_mul_ps(fscal,dz30);
322 /* Update vectorial force */
323 fix3 = _mm256_add_ps(fix3,tx);
324 fiy3 = _mm256_add_ps(fiy3,ty);
325 fiz3 = _mm256_add_ps(fiz3,tz);
327 fjx0 = _mm256_add_ps(fjx0,tx);
328 fjy0 = _mm256_add_ps(fjy0,ty);
329 fjz0 = _mm256_add_ps(fjz0,tz);
331 fjptrA = f+j_coord_offsetA;
332 fjptrB = f+j_coord_offsetB;
333 fjptrC = f+j_coord_offsetC;
334 fjptrD = f+j_coord_offsetD;
335 fjptrE = f+j_coord_offsetE;
336 fjptrF = f+j_coord_offsetF;
337 fjptrG = f+j_coord_offsetG;
338 fjptrH = f+j_coord_offsetH;
340 gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,fjx0,fjy0,fjz0);
342 /* Inner loop uses 99 flops */
348 /* Get j neighbor index, and coordinate index */
349 jnrlistA = jjnr[jidx];
350 jnrlistB = jjnr[jidx+1];
351 jnrlistC = jjnr[jidx+2];
352 jnrlistD = jjnr[jidx+3];
353 jnrlistE = jjnr[jidx+4];
354 jnrlistF = jjnr[jidx+5];
355 jnrlistG = jjnr[jidx+6];
356 jnrlistH = jjnr[jidx+7];
357 /* Sign of each element will be negative for non-real atoms.
358 * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
359 * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
361 dummy_mask = gmx_mm256_set_m128(gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx+4)),_mm_setzero_si128())),
362 gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128())));
364 jnrA = (jnrlistA>=0) ? jnrlistA : 0;
365 jnrB = (jnrlistB>=0) ? jnrlistB : 0;
366 jnrC = (jnrlistC>=0) ? jnrlistC : 0;
367 jnrD = (jnrlistD>=0) ? jnrlistD : 0;
368 jnrE = (jnrlistE>=0) ? jnrlistE : 0;
369 jnrF = (jnrlistF>=0) ? jnrlistF : 0;
370 jnrG = (jnrlistG>=0) ? jnrlistG : 0;
371 jnrH = (jnrlistH>=0) ? jnrlistH : 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;
376 j_coord_offsetE = DIM*jnrE;
377 j_coord_offsetF = DIM*jnrF;
378 j_coord_offsetG = DIM*jnrG;
379 j_coord_offsetH = DIM*jnrH;
381 /* load j atom coordinates */
382 gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
383 x+j_coord_offsetC,x+j_coord_offsetD,
384 x+j_coord_offsetE,x+j_coord_offsetF,
385 x+j_coord_offsetG,x+j_coord_offsetH,
388 /* Calculate displacement vector */
389 dx10 = _mm256_sub_ps(ix1,jx0);
390 dy10 = _mm256_sub_ps(iy1,jy0);
391 dz10 = _mm256_sub_ps(iz1,jz0);
392 dx20 = _mm256_sub_ps(ix2,jx0);
393 dy20 = _mm256_sub_ps(iy2,jy0);
394 dz20 = _mm256_sub_ps(iz2,jz0);
395 dx30 = _mm256_sub_ps(ix3,jx0);
396 dy30 = _mm256_sub_ps(iy3,jy0);
397 dz30 = _mm256_sub_ps(iz3,jz0);
399 /* Calculate squared distance and things based on it */
400 rsq10 = gmx_mm256_calc_rsq_ps(dx10,dy10,dz10);
401 rsq20 = gmx_mm256_calc_rsq_ps(dx20,dy20,dz20);
402 rsq30 = gmx_mm256_calc_rsq_ps(dx30,dy30,dz30);
404 rinv10 = gmx_mm256_invsqrt_ps(rsq10);
405 rinv20 = gmx_mm256_invsqrt_ps(rsq20);
406 rinv30 = gmx_mm256_invsqrt_ps(rsq30);
408 rinvsq10 = _mm256_mul_ps(rinv10,rinv10);
409 rinvsq20 = _mm256_mul_ps(rinv20,rinv20);
410 rinvsq30 = _mm256_mul_ps(rinv30,rinv30);
412 /* Load parameters for j particles */
413 jq0 = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
414 charge+jnrC+0,charge+jnrD+0,
415 charge+jnrE+0,charge+jnrF+0,
416 charge+jnrG+0,charge+jnrH+0);
418 fjx0 = _mm256_setzero_ps();
419 fjy0 = _mm256_setzero_ps();
420 fjz0 = _mm256_setzero_ps();
422 /**************************
423 * CALCULATE INTERACTIONS *
424 **************************/
426 /* Compute parameters for interactions between i and j atoms */
427 qq10 = _mm256_mul_ps(iq1,jq0);
429 /* REACTION-FIELD ELECTROSTATICS */
430 velec = _mm256_mul_ps(qq10,_mm256_sub_ps(_mm256_add_ps(rinv10,_mm256_mul_ps(krf,rsq10)),crf));
431 felec = _mm256_mul_ps(qq10,_mm256_sub_ps(_mm256_mul_ps(rinv10,rinvsq10),krf2));
433 /* Update potential sum for this i atom from the interaction with this j atom. */
434 velec = _mm256_andnot_ps(dummy_mask,velec);
435 velecsum = _mm256_add_ps(velecsum,velec);
439 fscal = _mm256_andnot_ps(dummy_mask,fscal);
441 /* Calculate temporary vectorial force */
442 tx = _mm256_mul_ps(fscal,dx10);
443 ty = _mm256_mul_ps(fscal,dy10);
444 tz = _mm256_mul_ps(fscal,dz10);
446 /* Update vectorial force */
447 fix1 = _mm256_add_ps(fix1,tx);
448 fiy1 = _mm256_add_ps(fiy1,ty);
449 fiz1 = _mm256_add_ps(fiz1,tz);
451 fjx0 = _mm256_add_ps(fjx0,tx);
452 fjy0 = _mm256_add_ps(fjy0,ty);
453 fjz0 = _mm256_add_ps(fjz0,tz);
455 /**************************
456 * CALCULATE INTERACTIONS *
457 **************************/
459 /* Compute parameters for interactions between i and j atoms */
460 qq20 = _mm256_mul_ps(iq2,jq0);
462 /* REACTION-FIELD ELECTROSTATICS */
463 velec = _mm256_mul_ps(qq20,_mm256_sub_ps(_mm256_add_ps(rinv20,_mm256_mul_ps(krf,rsq20)),crf));
464 felec = _mm256_mul_ps(qq20,_mm256_sub_ps(_mm256_mul_ps(rinv20,rinvsq20),krf2));
466 /* Update potential sum for this i atom from the interaction with this j atom. */
467 velec = _mm256_andnot_ps(dummy_mask,velec);
468 velecsum = _mm256_add_ps(velecsum,velec);
472 fscal = _mm256_andnot_ps(dummy_mask,fscal);
474 /* Calculate temporary vectorial force */
475 tx = _mm256_mul_ps(fscal,dx20);
476 ty = _mm256_mul_ps(fscal,dy20);
477 tz = _mm256_mul_ps(fscal,dz20);
479 /* Update vectorial force */
480 fix2 = _mm256_add_ps(fix2,tx);
481 fiy2 = _mm256_add_ps(fiy2,ty);
482 fiz2 = _mm256_add_ps(fiz2,tz);
484 fjx0 = _mm256_add_ps(fjx0,tx);
485 fjy0 = _mm256_add_ps(fjy0,ty);
486 fjz0 = _mm256_add_ps(fjz0,tz);
488 /**************************
489 * CALCULATE INTERACTIONS *
490 **************************/
492 /* Compute parameters for interactions between i and j atoms */
493 qq30 = _mm256_mul_ps(iq3,jq0);
495 /* REACTION-FIELD ELECTROSTATICS */
496 velec = _mm256_mul_ps(qq30,_mm256_sub_ps(_mm256_add_ps(rinv30,_mm256_mul_ps(krf,rsq30)),crf));
497 felec = _mm256_mul_ps(qq30,_mm256_sub_ps(_mm256_mul_ps(rinv30,rinvsq30),krf2));
499 /* Update potential sum for this i atom from the interaction with this j atom. */
500 velec = _mm256_andnot_ps(dummy_mask,velec);
501 velecsum = _mm256_add_ps(velecsum,velec);
505 fscal = _mm256_andnot_ps(dummy_mask,fscal);
507 /* Calculate temporary vectorial force */
508 tx = _mm256_mul_ps(fscal,dx30);
509 ty = _mm256_mul_ps(fscal,dy30);
510 tz = _mm256_mul_ps(fscal,dz30);
512 /* Update vectorial force */
513 fix3 = _mm256_add_ps(fix3,tx);
514 fiy3 = _mm256_add_ps(fiy3,ty);
515 fiz3 = _mm256_add_ps(fiz3,tz);
517 fjx0 = _mm256_add_ps(fjx0,tx);
518 fjy0 = _mm256_add_ps(fjy0,ty);
519 fjz0 = _mm256_add_ps(fjz0,tz);
521 fjptrA = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
522 fjptrB = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
523 fjptrC = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
524 fjptrD = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
525 fjptrE = (jnrlistE>=0) ? f+j_coord_offsetE : scratch;
526 fjptrF = (jnrlistF>=0) ? f+j_coord_offsetF : scratch;
527 fjptrG = (jnrlistG>=0) ? f+j_coord_offsetG : scratch;
528 fjptrH = (jnrlistH>=0) ? f+j_coord_offsetH : scratch;
530 gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,fjx0,fjy0,fjz0);
532 /* Inner loop uses 99 flops */
535 /* End of innermost loop */
537 gmx_mm256_update_iforce_3atom_swizzle_ps(fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
538 f+i_coord_offset+DIM,fshift+i_shift_offset);
541 /* Update potential energies */
542 gmx_mm256_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid);
544 /* Increment number of inner iterations */
545 inneriter += j_index_end - j_index_start;
547 /* Outer loop uses 19 flops */
550 /* Increment number of outer iterations */
553 /* Update outer/inner flops */
555 inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_W4_VF,outeriter*19 + inneriter*99);
558 * Gromacs nonbonded kernel: nb_kernel_ElecRF_VdwNone_GeomW4P1_F_avx_256_single
559 * Electrostatics interaction: ReactionField
560 * VdW interaction: None
561 * Geometry: Water4-Particle
562 * Calculate force/pot: Force
565 nb_kernel_ElecRF_VdwNone_GeomW4P1_F_avx_256_single
566 (t_nblist * gmx_restrict nlist,
567 rvec * gmx_restrict xx,
568 rvec * gmx_restrict ff,
569 t_forcerec * gmx_restrict fr,
570 t_mdatoms * gmx_restrict mdatoms,
571 nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
572 t_nrnb * gmx_restrict nrnb)
574 /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
575 * just 0 for non-waters.
576 * Suffixes A,B,C,D,E,F,G,H refer to j loop unrolling done with AVX, e.g. for the eight different
577 * jnr indices corresponding to data put in the four positions in the SIMD register.
579 int i_shift_offset,i_coord_offset,outeriter,inneriter;
580 int j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
581 int jnrA,jnrB,jnrC,jnrD;
582 int jnrE,jnrF,jnrG,jnrH;
583 int jnrlistA,jnrlistB,jnrlistC,jnrlistD;
584 int jnrlistE,jnrlistF,jnrlistG,jnrlistH;
585 int j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
586 int j_coord_offsetE,j_coord_offsetF,j_coord_offsetG,j_coord_offsetH;
587 int *iinr,*jindex,*jjnr,*shiftidx,*gid;
589 real *shiftvec,*fshift,*x,*f;
590 real *fjptrA,*fjptrB,*fjptrC,*fjptrD,*fjptrE,*fjptrF,*fjptrG,*fjptrH;
592 __m256 tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
593 real * vdwioffsetptr1;
594 __m256 ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
595 real * vdwioffsetptr2;
596 __m256 ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
597 real * vdwioffsetptr3;
598 __m256 ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
599 int vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D,vdwjidx0E,vdwjidx0F,vdwjidx0G,vdwjidx0H;
600 __m256 jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
601 __m256 dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
602 __m256 dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
603 __m256 dx30,dy30,dz30,rsq30,rinv30,rinvsq30,r30,qq30,c6_30,c12_30;
604 __m256 velec,felec,velecsum,facel,crf,krf,krf2;
606 __m256 dummy_mask,cutoff_mask;
607 __m256 signbit = _mm256_castsi256_ps( _mm256_set1_epi32(0x80000000) );
608 __m256 one = _mm256_set1_ps(1.0);
609 __m256 two = _mm256_set1_ps(2.0);
615 jindex = nlist->jindex;
617 shiftidx = nlist->shift;
619 shiftvec = fr->shift_vec[0];
620 fshift = fr->fshift[0];
621 facel = _mm256_set1_ps(fr->epsfac);
622 charge = mdatoms->chargeA;
623 krf = _mm256_set1_ps(fr->ic->k_rf);
624 krf2 = _mm256_set1_ps(fr->ic->k_rf*2.0);
625 crf = _mm256_set1_ps(fr->ic->c_rf);
627 /* Setup water-specific parameters */
628 inr = nlist->iinr[0];
629 iq1 = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+1]));
630 iq2 = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+2]));
631 iq3 = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+3]));
633 /* Avoid stupid compiler warnings */
634 jnrA = jnrB = jnrC = jnrD = jnrE = jnrF = jnrG = jnrH = 0;
647 for(iidx=0;iidx<4*DIM;iidx++)
652 /* Start outer loop over neighborlists */
653 for(iidx=0; iidx<nri; iidx++)
655 /* Load shift vector for this list */
656 i_shift_offset = DIM*shiftidx[iidx];
658 /* Load limits for loop over neighbors */
659 j_index_start = jindex[iidx];
660 j_index_end = jindex[iidx+1];
662 /* Get outer coordinate index */
664 i_coord_offset = DIM*inr;
666 /* Load i particle coords and add shift vector */
667 gmx_mm256_load_shift_and_3rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset+DIM,
668 &ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
670 fix1 = _mm256_setzero_ps();
671 fiy1 = _mm256_setzero_ps();
672 fiz1 = _mm256_setzero_ps();
673 fix2 = _mm256_setzero_ps();
674 fiy2 = _mm256_setzero_ps();
675 fiz2 = _mm256_setzero_ps();
676 fix3 = _mm256_setzero_ps();
677 fiy3 = _mm256_setzero_ps();
678 fiz3 = _mm256_setzero_ps();
680 /* Start inner kernel loop */
681 for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+7]>=0; jidx+=8)
684 /* Get j neighbor index, and coordinate index */
693 j_coord_offsetA = DIM*jnrA;
694 j_coord_offsetB = DIM*jnrB;
695 j_coord_offsetC = DIM*jnrC;
696 j_coord_offsetD = DIM*jnrD;
697 j_coord_offsetE = DIM*jnrE;
698 j_coord_offsetF = DIM*jnrF;
699 j_coord_offsetG = DIM*jnrG;
700 j_coord_offsetH = DIM*jnrH;
702 /* load j atom coordinates */
703 gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
704 x+j_coord_offsetC,x+j_coord_offsetD,
705 x+j_coord_offsetE,x+j_coord_offsetF,
706 x+j_coord_offsetG,x+j_coord_offsetH,
709 /* Calculate displacement vector */
710 dx10 = _mm256_sub_ps(ix1,jx0);
711 dy10 = _mm256_sub_ps(iy1,jy0);
712 dz10 = _mm256_sub_ps(iz1,jz0);
713 dx20 = _mm256_sub_ps(ix2,jx0);
714 dy20 = _mm256_sub_ps(iy2,jy0);
715 dz20 = _mm256_sub_ps(iz2,jz0);
716 dx30 = _mm256_sub_ps(ix3,jx0);
717 dy30 = _mm256_sub_ps(iy3,jy0);
718 dz30 = _mm256_sub_ps(iz3,jz0);
720 /* Calculate squared distance and things based on it */
721 rsq10 = gmx_mm256_calc_rsq_ps(dx10,dy10,dz10);
722 rsq20 = gmx_mm256_calc_rsq_ps(dx20,dy20,dz20);
723 rsq30 = gmx_mm256_calc_rsq_ps(dx30,dy30,dz30);
725 rinv10 = gmx_mm256_invsqrt_ps(rsq10);
726 rinv20 = gmx_mm256_invsqrt_ps(rsq20);
727 rinv30 = gmx_mm256_invsqrt_ps(rsq30);
729 rinvsq10 = _mm256_mul_ps(rinv10,rinv10);
730 rinvsq20 = _mm256_mul_ps(rinv20,rinv20);
731 rinvsq30 = _mm256_mul_ps(rinv30,rinv30);
733 /* Load parameters for j particles */
734 jq0 = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
735 charge+jnrC+0,charge+jnrD+0,
736 charge+jnrE+0,charge+jnrF+0,
737 charge+jnrG+0,charge+jnrH+0);
739 fjx0 = _mm256_setzero_ps();
740 fjy0 = _mm256_setzero_ps();
741 fjz0 = _mm256_setzero_ps();
743 /**************************
744 * CALCULATE INTERACTIONS *
745 **************************/
747 /* Compute parameters for interactions between i and j atoms */
748 qq10 = _mm256_mul_ps(iq1,jq0);
750 /* REACTION-FIELD ELECTROSTATICS */
751 felec = _mm256_mul_ps(qq10,_mm256_sub_ps(_mm256_mul_ps(rinv10,rinvsq10),krf2));
755 /* Calculate temporary vectorial force */
756 tx = _mm256_mul_ps(fscal,dx10);
757 ty = _mm256_mul_ps(fscal,dy10);
758 tz = _mm256_mul_ps(fscal,dz10);
760 /* Update vectorial force */
761 fix1 = _mm256_add_ps(fix1,tx);
762 fiy1 = _mm256_add_ps(fiy1,ty);
763 fiz1 = _mm256_add_ps(fiz1,tz);
765 fjx0 = _mm256_add_ps(fjx0,tx);
766 fjy0 = _mm256_add_ps(fjy0,ty);
767 fjz0 = _mm256_add_ps(fjz0,tz);
769 /**************************
770 * CALCULATE INTERACTIONS *
771 **************************/
773 /* Compute parameters for interactions between i and j atoms */
774 qq20 = _mm256_mul_ps(iq2,jq0);
776 /* REACTION-FIELD ELECTROSTATICS */
777 felec = _mm256_mul_ps(qq20,_mm256_sub_ps(_mm256_mul_ps(rinv20,rinvsq20),krf2));
781 /* Calculate temporary vectorial force */
782 tx = _mm256_mul_ps(fscal,dx20);
783 ty = _mm256_mul_ps(fscal,dy20);
784 tz = _mm256_mul_ps(fscal,dz20);
786 /* Update vectorial force */
787 fix2 = _mm256_add_ps(fix2,tx);
788 fiy2 = _mm256_add_ps(fiy2,ty);
789 fiz2 = _mm256_add_ps(fiz2,tz);
791 fjx0 = _mm256_add_ps(fjx0,tx);
792 fjy0 = _mm256_add_ps(fjy0,ty);
793 fjz0 = _mm256_add_ps(fjz0,tz);
795 /**************************
796 * CALCULATE INTERACTIONS *
797 **************************/
799 /* Compute parameters for interactions between i and j atoms */
800 qq30 = _mm256_mul_ps(iq3,jq0);
802 /* REACTION-FIELD ELECTROSTATICS */
803 felec = _mm256_mul_ps(qq30,_mm256_sub_ps(_mm256_mul_ps(rinv30,rinvsq30),krf2));
807 /* Calculate temporary vectorial force */
808 tx = _mm256_mul_ps(fscal,dx30);
809 ty = _mm256_mul_ps(fscal,dy30);
810 tz = _mm256_mul_ps(fscal,dz30);
812 /* Update vectorial force */
813 fix3 = _mm256_add_ps(fix3,tx);
814 fiy3 = _mm256_add_ps(fiy3,ty);
815 fiz3 = _mm256_add_ps(fiz3,tz);
817 fjx0 = _mm256_add_ps(fjx0,tx);
818 fjy0 = _mm256_add_ps(fjy0,ty);
819 fjz0 = _mm256_add_ps(fjz0,tz);
821 fjptrA = f+j_coord_offsetA;
822 fjptrB = f+j_coord_offsetB;
823 fjptrC = f+j_coord_offsetC;
824 fjptrD = f+j_coord_offsetD;
825 fjptrE = f+j_coord_offsetE;
826 fjptrF = f+j_coord_offsetF;
827 fjptrG = f+j_coord_offsetG;
828 fjptrH = f+j_coord_offsetH;
830 gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,fjx0,fjy0,fjz0);
832 /* Inner loop uses 84 flops */
838 /* Get j neighbor index, and coordinate index */
839 jnrlistA = jjnr[jidx];
840 jnrlistB = jjnr[jidx+1];
841 jnrlistC = jjnr[jidx+2];
842 jnrlistD = jjnr[jidx+3];
843 jnrlistE = jjnr[jidx+4];
844 jnrlistF = jjnr[jidx+5];
845 jnrlistG = jjnr[jidx+6];
846 jnrlistH = jjnr[jidx+7];
847 /* Sign of each element will be negative for non-real atoms.
848 * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
849 * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
851 dummy_mask = gmx_mm256_set_m128(gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx+4)),_mm_setzero_si128())),
852 gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128())));
854 jnrA = (jnrlistA>=0) ? jnrlistA : 0;
855 jnrB = (jnrlistB>=0) ? jnrlistB : 0;
856 jnrC = (jnrlistC>=0) ? jnrlistC : 0;
857 jnrD = (jnrlistD>=0) ? jnrlistD : 0;
858 jnrE = (jnrlistE>=0) ? jnrlistE : 0;
859 jnrF = (jnrlistF>=0) ? jnrlistF : 0;
860 jnrG = (jnrlistG>=0) ? jnrlistG : 0;
861 jnrH = (jnrlistH>=0) ? jnrlistH : 0;
862 j_coord_offsetA = DIM*jnrA;
863 j_coord_offsetB = DIM*jnrB;
864 j_coord_offsetC = DIM*jnrC;
865 j_coord_offsetD = DIM*jnrD;
866 j_coord_offsetE = DIM*jnrE;
867 j_coord_offsetF = DIM*jnrF;
868 j_coord_offsetG = DIM*jnrG;
869 j_coord_offsetH = DIM*jnrH;
871 /* load j atom coordinates */
872 gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
873 x+j_coord_offsetC,x+j_coord_offsetD,
874 x+j_coord_offsetE,x+j_coord_offsetF,
875 x+j_coord_offsetG,x+j_coord_offsetH,
878 /* Calculate displacement vector */
879 dx10 = _mm256_sub_ps(ix1,jx0);
880 dy10 = _mm256_sub_ps(iy1,jy0);
881 dz10 = _mm256_sub_ps(iz1,jz0);
882 dx20 = _mm256_sub_ps(ix2,jx0);
883 dy20 = _mm256_sub_ps(iy2,jy0);
884 dz20 = _mm256_sub_ps(iz2,jz0);
885 dx30 = _mm256_sub_ps(ix3,jx0);
886 dy30 = _mm256_sub_ps(iy3,jy0);
887 dz30 = _mm256_sub_ps(iz3,jz0);
889 /* Calculate squared distance and things based on it */
890 rsq10 = gmx_mm256_calc_rsq_ps(dx10,dy10,dz10);
891 rsq20 = gmx_mm256_calc_rsq_ps(dx20,dy20,dz20);
892 rsq30 = gmx_mm256_calc_rsq_ps(dx30,dy30,dz30);
894 rinv10 = gmx_mm256_invsqrt_ps(rsq10);
895 rinv20 = gmx_mm256_invsqrt_ps(rsq20);
896 rinv30 = gmx_mm256_invsqrt_ps(rsq30);
898 rinvsq10 = _mm256_mul_ps(rinv10,rinv10);
899 rinvsq20 = _mm256_mul_ps(rinv20,rinv20);
900 rinvsq30 = _mm256_mul_ps(rinv30,rinv30);
902 /* Load parameters for j particles */
903 jq0 = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
904 charge+jnrC+0,charge+jnrD+0,
905 charge+jnrE+0,charge+jnrF+0,
906 charge+jnrG+0,charge+jnrH+0);
908 fjx0 = _mm256_setzero_ps();
909 fjy0 = _mm256_setzero_ps();
910 fjz0 = _mm256_setzero_ps();
912 /**************************
913 * CALCULATE INTERACTIONS *
914 **************************/
916 /* Compute parameters for interactions between i and j atoms */
917 qq10 = _mm256_mul_ps(iq1,jq0);
919 /* REACTION-FIELD ELECTROSTATICS */
920 felec = _mm256_mul_ps(qq10,_mm256_sub_ps(_mm256_mul_ps(rinv10,rinvsq10),krf2));
924 fscal = _mm256_andnot_ps(dummy_mask,fscal);
926 /* Calculate temporary vectorial force */
927 tx = _mm256_mul_ps(fscal,dx10);
928 ty = _mm256_mul_ps(fscal,dy10);
929 tz = _mm256_mul_ps(fscal,dz10);
931 /* Update vectorial force */
932 fix1 = _mm256_add_ps(fix1,tx);
933 fiy1 = _mm256_add_ps(fiy1,ty);
934 fiz1 = _mm256_add_ps(fiz1,tz);
936 fjx0 = _mm256_add_ps(fjx0,tx);
937 fjy0 = _mm256_add_ps(fjy0,ty);
938 fjz0 = _mm256_add_ps(fjz0,tz);
940 /**************************
941 * CALCULATE INTERACTIONS *
942 **************************/
944 /* Compute parameters for interactions between i and j atoms */
945 qq20 = _mm256_mul_ps(iq2,jq0);
947 /* REACTION-FIELD ELECTROSTATICS */
948 felec = _mm256_mul_ps(qq20,_mm256_sub_ps(_mm256_mul_ps(rinv20,rinvsq20),krf2));
952 fscal = _mm256_andnot_ps(dummy_mask,fscal);
954 /* Calculate temporary vectorial force */
955 tx = _mm256_mul_ps(fscal,dx20);
956 ty = _mm256_mul_ps(fscal,dy20);
957 tz = _mm256_mul_ps(fscal,dz20);
959 /* Update vectorial force */
960 fix2 = _mm256_add_ps(fix2,tx);
961 fiy2 = _mm256_add_ps(fiy2,ty);
962 fiz2 = _mm256_add_ps(fiz2,tz);
964 fjx0 = _mm256_add_ps(fjx0,tx);
965 fjy0 = _mm256_add_ps(fjy0,ty);
966 fjz0 = _mm256_add_ps(fjz0,tz);
968 /**************************
969 * CALCULATE INTERACTIONS *
970 **************************/
972 /* Compute parameters for interactions between i and j atoms */
973 qq30 = _mm256_mul_ps(iq3,jq0);
975 /* REACTION-FIELD ELECTROSTATICS */
976 felec = _mm256_mul_ps(qq30,_mm256_sub_ps(_mm256_mul_ps(rinv30,rinvsq30),krf2));
980 fscal = _mm256_andnot_ps(dummy_mask,fscal);
982 /* Calculate temporary vectorial force */
983 tx = _mm256_mul_ps(fscal,dx30);
984 ty = _mm256_mul_ps(fscal,dy30);
985 tz = _mm256_mul_ps(fscal,dz30);
987 /* Update vectorial force */
988 fix3 = _mm256_add_ps(fix3,tx);
989 fiy3 = _mm256_add_ps(fiy3,ty);
990 fiz3 = _mm256_add_ps(fiz3,tz);
992 fjx0 = _mm256_add_ps(fjx0,tx);
993 fjy0 = _mm256_add_ps(fjy0,ty);
994 fjz0 = _mm256_add_ps(fjz0,tz);
996 fjptrA = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
997 fjptrB = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
998 fjptrC = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
999 fjptrD = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
1000 fjptrE = (jnrlistE>=0) ? f+j_coord_offsetE : scratch;
1001 fjptrF = (jnrlistF>=0) ? f+j_coord_offsetF : scratch;
1002 fjptrG = (jnrlistG>=0) ? f+j_coord_offsetG : scratch;
1003 fjptrH = (jnrlistH>=0) ? f+j_coord_offsetH : scratch;
1005 gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,fjx0,fjy0,fjz0);
1007 /* Inner loop uses 84 flops */
1010 /* End of innermost loop */
1012 gmx_mm256_update_iforce_3atom_swizzle_ps(fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
1013 f+i_coord_offset+DIM,fshift+i_shift_offset);
1015 /* Increment number of inner iterations */
1016 inneriter += j_index_end - j_index_start;
1018 /* Outer loop uses 18 flops */
1021 /* Increment number of outer iterations */
1024 /* Update outer/inner flops */
1026 inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_W4_F,outeriter*18 + inneriter*84);