2 * Note: this file was generated by the Gromacs avx_256_single kernel generator.
4 * This source code is part of
8 * Copyright (c) 2001-2012, The GROMACS Development Team
10 * Gromacs is a library for molecular simulation and trajectory analysis,
11 * written by Erik Lindahl, David van der Spoel, Berk Hess, and others - for
12 * a full list of developers and information, check out http://www.gromacs.org
14 * This program is free software; you can redistribute it and/or modify it under
15 * the terms of the GNU Lesser General Public License as published by the Free
16 * Software Foundation; either version 2 of the License, or (at your option) any
19 * To help fund GROMACS development, we humbly ask that you cite
20 * the papers people have written on it - you can find them on the website.
28 #include "../nb_kernel.h"
29 #include "types/simple.h"
33 #include "gmx_math_x86_avx_256_single.h"
34 #include "kernelutil_x86_avx_256_single.h"
37 * Gromacs nonbonded kernel: nb_kernel_ElecCoul_VdwNone_GeomW4P1_VF_avx_256_single
38 * Electrostatics interaction: Coulomb
39 * VdW interaction: None
40 * Geometry: Water4-Particle
41 * Calculate force/pot: PotentialAndForce
44 nb_kernel_ElecCoul_VdwNone_GeomW4P1_VF_avx_256_single
45 (t_nblist * gmx_restrict nlist,
46 rvec * gmx_restrict xx,
47 rvec * gmx_restrict ff,
48 t_forcerec * gmx_restrict fr,
49 t_mdatoms * gmx_restrict mdatoms,
50 nb_kernel_data_t * gmx_restrict kernel_data,
51 t_nrnb * gmx_restrict nrnb)
53 /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
54 * just 0 for non-waters.
55 * Suffixes A,B,C,D,E,F,G,H refer to j loop unrolling done with AVX, e.g. for the eight different
56 * jnr indices corresponding to data put in the four positions in the SIMD register.
58 int i_shift_offset,i_coord_offset,outeriter,inneriter;
59 int j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
60 int jnrA,jnrB,jnrC,jnrD;
61 int jnrE,jnrF,jnrG,jnrH;
62 int jnrlistA,jnrlistB,jnrlistC,jnrlistD;
63 int jnrlistE,jnrlistF,jnrlistG,jnrlistH;
64 int j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
65 int j_coord_offsetE,j_coord_offsetF,j_coord_offsetG,j_coord_offsetH;
66 int *iinr,*jindex,*jjnr,*shiftidx,*gid;
68 real *shiftvec,*fshift,*x,*f;
69 real *fjptrA,*fjptrB,*fjptrC,*fjptrD,*fjptrE,*fjptrF,*fjptrG,*fjptrH;
71 __m256 tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
72 real * vdwioffsetptr1;
73 __m256 ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
74 real * vdwioffsetptr2;
75 __m256 ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
76 real * vdwioffsetptr3;
77 __m256 ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
78 int vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D,vdwjidx0E,vdwjidx0F,vdwjidx0G,vdwjidx0H;
79 __m256 jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
80 __m256 dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
81 __m256 dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
82 __m256 dx30,dy30,dz30,rsq30,rinv30,rinvsq30,r30,qq30,c6_30,c12_30;
83 __m256 velec,felec,velecsum,facel,crf,krf,krf2;
85 __m256 dummy_mask,cutoff_mask;
86 __m256 signbit = _mm256_castsi256_ps( _mm256_set1_epi32(0x80000000) );
87 __m256 one = _mm256_set1_ps(1.0);
88 __m256 two = _mm256_set1_ps(2.0);
94 jindex = nlist->jindex;
96 shiftidx = nlist->shift;
98 shiftvec = fr->shift_vec[0];
99 fshift = fr->fshift[0];
100 facel = _mm256_set1_ps(fr->epsfac);
101 charge = mdatoms->chargeA;
103 /* Setup water-specific parameters */
104 inr = nlist->iinr[0];
105 iq1 = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+1]));
106 iq2 = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+2]));
107 iq3 = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+3]));
109 /* Avoid stupid compiler warnings */
110 jnrA = jnrB = jnrC = jnrD = jnrE = jnrF = jnrG = jnrH = 0;
123 for(iidx=0;iidx<4*DIM;iidx++)
128 /* Start outer loop over neighborlists */
129 for(iidx=0; iidx<nri; iidx++)
131 /* Load shift vector for this list */
132 i_shift_offset = DIM*shiftidx[iidx];
134 /* Load limits for loop over neighbors */
135 j_index_start = jindex[iidx];
136 j_index_end = jindex[iidx+1];
138 /* Get outer coordinate index */
140 i_coord_offset = DIM*inr;
142 /* Load i particle coords and add shift vector */
143 gmx_mm256_load_shift_and_3rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset+DIM,
144 &ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
146 fix1 = _mm256_setzero_ps();
147 fiy1 = _mm256_setzero_ps();
148 fiz1 = _mm256_setzero_ps();
149 fix2 = _mm256_setzero_ps();
150 fiy2 = _mm256_setzero_ps();
151 fiz2 = _mm256_setzero_ps();
152 fix3 = _mm256_setzero_ps();
153 fiy3 = _mm256_setzero_ps();
154 fiz3 = _mm256_setzero_ps();
156 /* Reset potential sums */
157 velecsum = _mm256_setzero_ps();
159 /* Start inner kernel loop */
160 for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+7]>=0; jidx+=8)
163 /* Get j neighbor index, and coordinate index */
172 j_coord_offsetA = DIM*jnrA;
173 j_coord_offsetB = DIM*jnrB;
174 j_coord_offsetC = DIM*jnrC;
175 j_coord_offsetD = DIM*jnrD;
176 j_coord_offsetE = DIM*jnrE;
177 j_coord_offsetF = DIM*jnrF;
178 j_coord_offsetG = DIM*jnrG;
179 j_coord_offsetH = DIM*jnrH;
181 /* load j atom coordinates */
182 gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
183 x+j_coord_offsetC,x+j_coord_offsetD,
184 x+j_coord_offsetE,x+j_coord_offsetF,
185 x+j_coord_offsetG,x+j_coord_offsetH,
188 /* Calculate displacement vector */
189 dx10 = _mm256_sub_ps(ix1,jx0);
190 dy10 = _mm256_sub_ps(iy1,jy0);
191 dz10 = _mm256_sub_ps(iz1,jz0);
192 dx20 = _mm256_sub_ps(ix2,jx0);
193 dy20 = _mm256_sub_ps(iy2,jy0);
194 dz20 = _mm256_sub_ps(iz2,jz0);
195 dx30 = _mm256_sub_ps(ix3,jx0);
196 dy30 = _mm256_sub_ps(iy3,jy0);
197 dz30 = _mm256_sub_ps(iz3,jz0);
199 /* Calculate squared distance and things based on it */
200 rsq10 = gmx_mm256_calc_rsq_ps(dx10,dy10,dz10);
201 rsq20 = gmx_mm256_calc_rsq_ps(dx20,dy20,dz20);
202 rsq30 = gmx_mm256_calc_rsq_ps(dx30,dy30,dz30);
204 rinv10 = gmx_mm256_invsqrt_ps(rsq10);
205 rinv20 = gmx_mm256_invsqrt_ps(rsq20);
206 rinv30 = gmx_mm256_invsqrt_ps(rsq30);
208 rinvsq10 = _mm256_mul_ps(rinv10,rinv10);
209 rinvsq20 = _mm256_mul_ps(rinv20,rinv20);
210 rinvsq30 = _mm256_mul_ps(rinv30,rinv30);
212 /* Load parameters for j particles */
213 jq0 = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
214 charge+jnrC+0,charge+jnrD+0,
215 charge+jnrE+0,charge+jnrF+0,
216 charge+jnrG+0,charge+jnrH+0);
218 fjx0 = _mm256_setzero_ps();
219 fjy0 = _mm256_setzero_ps();
220 fjz0 = _mm256_setzero_ps();
222 /**************************
223 * CALCULATE INTERACTIONS *
224 **************************/
226 /* Compute parameters for interactions between i and j atoms */
227 qq10 = _mm256_mul_ps(iq1,jq0);
229 /* COULOMB ELECTROSTATICS */
230 velec = _mm256_mul_ps(qq10,rinv10);
231 felec = _mm256_mul_ps(velec,rinvsq10);
233 /* Update potential sum for this i atom from the interaction with this j atom. */
234 velecsum = _mm256_add_ps(velecsum,velec);
238 /* Calculate temporary vectorial force */
239 tx = _mm256_mul_ps(fscal,dx10);
240 ty = _mm256_mul_ps(fscal,dy10);
241 tz = _mm256_mul_ps(fscal,dz10);
243 /* Update vectorial force */
244 fix1 = _mm256_add_ps(fix1,tx);
245 fiy1 = _mm256_add_ps(fiy1,ty);
246 fiz1 = _mm256_add_ps(fiz1,tz);
248 fjx0 = _mm256_add_ps(fjx0,tx);
249 fjy0 = _mm256_add_ps(fjy0,ty);
250 fjz0 = _mm256_add_ps(fjz0,tz);
252 /**************************
253 * CALCULATE INTERACTIONS *
254 **************************/
256 /* Compute parameters for interactions between i and j atoms */
257 qq20 = _mm256_mul_ps(iq2,jq0);
259 /* COULOMB ELECTROSTATICS */
260 velec = _mm256_mul_ps(qq20,rinv20);
261 felec = _mm256_mul_ps(velec,rinvsq20);
263 /* Update potential sum for this i atom from the interaction with this j atom. */
264 velecsum = _mm256_add_ps(velecsum,velec);
268 /* Calculate temporary vectorial force */
269 tx = _mm256_mul_ps(fscal,dx20);
270 ty = _mm256_mul_ps(fscal,dy20);
271 tz = _mm256_mul_ps(fscal,dz20);
273 /* Update vectorial force */
274 fix2 = _mm256_add_ps(fix2,tx);
275 fiy2 = _mm256_add_ps(fiy2,ty);
276 fiz2 = _mm256_add_ps(fiz2,tz);
278 fjx0 = _mm256_add_ps(fjx0,tx);
279 fjy0 = _mm256_add_ps(fjy0,ty);
280 fjz0 = _mm256_add_ps(fjz0,tz);
282 /**************************
283 * CALCULATE INTERACTIONS *
284 **************************/
286 /* Compute parameters for interactions between i and j atoms */
287 qq30 = _mm256_mul_ps(iq3,jq0);
289 /* COULOMB ELECTROSTATICS */
290 velec = _mm256_mul_ps(qq30,rinv30);
291 felec = _mm256_mul_ps(velec,rinvsq30);
293 /* Update potential sum for this i atom from the interaction with this j atom. */
294 velecsum = _mm256_add_ps(velecsum,velec);
298 /* Calculate temporary vectorial force */
299 tx = _mm256_mul_ps(fscal,dx30);
300 ty = _mm256_mul_ps(fscal,dy30);
301 tz = _mm256_mul_ps(fscal,dz30);
303 /* Update vectorial force */
304 fix3 = _mm256_add_ps(fix3,tx);
305 fiy3 = _mm256_add_ps(fiy3,ty);
306 fiz3 = _mm256_add_ps(fiz3,tz);
308 fjx0 = _mm256_add_ps(fjx0,tx);
309 fjy0 = _mm256_add_ps(fjy0,ty);
310 fjz0 = _mm256_add_ps(fjz0,tz);
312 fjptrA = f+j_coord_offsetA;
313 fjptrB = f+j_coord_offsetB;
314 fjptrC = f+j_coord_offsetC;
315 fjptrD = f+j_coord_offsetD;
316 fjptrE = f+j_coord_offsetE;
317 fjptrF = f+j_coord_offsetF;
318 fjptrG = f+j_coord_offsetG;
319 fjptrH = f+j_coord_offsetH;
321 gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,fjx0,fjy0,fjz0);
323 /* Inner loop uses 84 flops */
329 /* Get j neighbor index, and coordinate index */
330 jnrlistA = jjnr[jidx];
331 jnrlistB = jjnr[jidx+1];
332 jnrlistC = jjnr[jidx+2];
333 jnrlistD = jjnr[jidx+3];
334 jnrlistE = jjnr[jidx+4];
335 jnrlistF = jjnr[jidx+5];
336 jnrlistG = jjnr[jidx+6];
337 jnrlistH = jjnr[jidx+7];
338 /* Sign of each element will be negative for non-real atoms.
339 * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
340 * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
342 dummy_mask = gmx_mm256_set_m128(gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx+4)),_mm_setzero_si128())),
343 gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128())));
345 jnrA = (jnrlistA>=0) ? jnrlistA : 0;
346 jnrB = (jnrlistB>=0) ? jnrlistB : 0;
347 jnrC = (jnrlistC>=0) ? jnrlistC : 0;
348 jnrD = (jnrlistD>=0) ? jnrlistD : 0;
349 jnrE = (jnrlistE>=0) ? jnrlistE : 0;
350 jnrF = (jnrlistF>=0) ? jnrlistF : 0;
351 jnrG = (jnrlistG>=0) ? jnrlistG : 0;
352 jnrH = (jnrlistH>=0) ? jnrlistH : 0;
353 j_coord_offsetA = DIM*jnrA;
354 j_coord_offsetB = DIM*jnrB;
355 j_coord_offsetC = DIM*jnrC;
356 j_coord_offsetD = DIM*jnrD;
357 j_coord_offsetE = DIM*jnrE;
358 j_coord_offsetF = DIM*jnrF;
359 j_coord_offsetG = DIM*jnrG;
360 j_coord_offsetH = DIM*jnrH;
362 /* load j atom coordinates */
363 gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
364 x+j_coord_offsetC,x+j_coord_offsetD,
365 x+j_coord_offsetE,x+j_coord_offsetF,
366 x+j_coord_offsetG,x+j_coord_offsetH,
369 /* Calculate displacement vector */
370 dx10 = _mm256_sub_ps(ix1,jx0);
371 dy10 = _mm256_sub_ps(iy1,jy0);
372 dz10 = _mm256_sub_ps(iz1,jz0);
373 dx20 = _mm256_sub_ps(ix2,jx0);
374 dy20 = _mm256_sub_ps(iy2,jy0);
375 dz20 = _mm256_sub_ps(iz2,jz0);
376 dx30 = _mm256_sub_ps(ix3,jx0);
377 dy30 = _mm256_sub_ps(iy3,jy0);
378 dz30 = _mm256_sub_ps(iz3,jz0);
380 /* Calculate squared distance and things based on it */
381 rsq10 = gmx_mm256_calc_rsq_ps(dx10,dy10,dz10);
382 rsq20 = gmx_mm256_calc_rsq_ps(dx20,dy20,dz20);
383 rsq30 = gmx_mm256_calc_rsq_ps(dx30,dy30,dz30);
385 rinv10 = gmx_mm256_invsqrt_ps(rsq10);
386 rinv20 = gmx_mm256_invsqrt_ps(rsq20);
387 rinv30 = gmx_mm256_invsqrt_ps(rsq30);
389 rinvsq10 = _mm256_mul_ps(rinv10,rinv10);
390 rinvsq20 = _mm256_mul_ps(rinv20,rinv20);
391 rinvsq30 = _mm256_mul_ps(rinv30,rinv30);
393 /* Load parameters for j particles */
394 jq0 = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
395 charge+jnrC+0,charge+jnrD+0,
396 charge+jnrE+0,charge+jnrF+0,
397 charge+jnrG+0,charge+jnrH+0);
399 fjx0 = _mm256_setzero_ps();
400 fjy0 = _mm256_setzero_ps();
401 fjz0 = _mm256_setzero_ps();
403 /**************************
404 * CALCULATE INTERACTIONS *
405 **************************/
407 /* Compute parameters for interactions between i and j atoms */
408 qq10 = _mm256_mul_ps(iq1,jq0);
410 /* COULOMB ELECTROSTATICS */
411 velec = _mm256_mul_ps(qq10,rinv10);
412 felec = _mm256_mul_ps(velec,rinvsq10);
414 /* Update potential sum for this i atom from the interaction with this j atom. */
415 velec = _mm256_andnot_ps(dummy_mask,velec);
416 velecsum = _mm256_add_ps(velecsum,velec);
420 fscal = _mm256_andnot_ps(dummy_mask,fscal);
422 /* Calculate temporary vectorial force */
423 tx = _mm256_mul_ps(fscal,dx10);
424 ty = _mm256_mul_ps(fscal,dy10);
425 tz = _mm256_mul_ps(fscal,dz10);
427 /* Update vectorial force */
428 fix1 = _mm256_add_ps(fix1,tx);
429 fiy1 = _mm256_add_ps(fiy1,ty);
430 fiz1 = _mm256_add_ps(fiz1,tz);
432 fjx0 = _mm256_add_ps(fjx0,tx);
433 fjy0 = _mm256_add_ps(fjy0,ty);
434 fjz0 = _mm256_add_ps(fjz0,tz);
436 /**************************
437 * CALCULATE INTERACTIONS *
438 **************************/
440 /* Compute parameters for interactions between i and j atoms */
441 qq20 = _mm256_mul_ps(iq2,jq0);
443 /* COULOMB ELECTROSTATICS */
444 velec = _mm256_mul_ps(qq20,rinv20);
445 felec = _mm256_mul_ps(velec,rinvsq20);
447 /* Update potential sum for this i atom from the interaction with this j atom. */
448 velec = _mm256_andnot_ps(dummy_mask,velec);
449 velecsum = _mm256_add_ps(velecsum,velec);
453 fscal = _mm256_andnot_ps(dummy_mask,fscal);
455 /* Calculate temporary vectorial force */
456 tx = _mm256_mul_ps(fscal,dx20);
457 ty = _mm256_mul_ps(fscal,dy20);
458 tz = _mm256_mul_ps(fscal,dz20);
460 /* Update vectorial force */
461 fix2 = _mm256_add_ps(fix2,tx);
462 fiy2 = _mm256_add_ps(fiy2,ty);
463 fiz2 = _mm256_add_ps(fiz2,tz);
465 fjx0 = _mm256_add_ps(fjx0,tx);
466 fjy0 = _mm256_add_ps(fjy0,ty);
467 fjz0 = _mm256_add_ps(fjz0,tz);
469 /**************************
470 * CALCULATE INTERACTIONS *
471 **************************/
473 /* Compute parameters for interactions between i and j atoms */
474 qq30 = _mm256_mul_ps(iq3,jq0);
476 /* COULOMB ELECTROSTATICS */
477 velec = _mm256_mul_ps(qq30,rinv30);
478 felec = _mm256_mul_ps(velec,rinvsq30);
480 /* Update potential sum for this i atom from the interaction with this j atom. */
481 velec = _mm256_andnot_ps(dummy_mask,velec);
482 velecsum = _mm256_add_ps(velecsum,velec);
486 fscal = _mm256_andnot_ps(dummy_mask,fscal);
488 /* Calculate temporary vectorial force */
489 tx = _mm256_mul_ps(fscal,dx30);
490 ty = _mm256_mul_ps(fscal,dy30);
491 tz = _mm256_mul_ps(fscal,dz30);
493 /* Update vectorial force */
494 fix3 = _mm256_add_ps(fix3,tx);
495 fiy3 = _mm256_add_ps(fiy3,ty);
496 fiz3 = _mm256_add_ps(fiz3,tz);
498 fjx0 = _mm256_add_ps(fjx0,tx);
499 fjy0 = _mm256_add_ps(fjy0,ty);
500 fjz0 = _mm256_add_ps(fjz0,tz);
502 fjptrA = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
503 fjptrB = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
504 fjptrC = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
505 fjptrD = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
506 fjptrE = (jnrlistE>=0) ? f+j_coord_offsetE : scratch;
507 fjptrF = (jnrlistF>=0) ? f+j_coord_offsetF : scratch;
508 fjptrG = (jnrlistG>=0) ? f+j_coord_offsetG : scratch;
509 fjptrH = (jnrlistH>=0) ? f+j_coord_offsetH : scratch;
511 gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,fjx0,fjy0,fjz0);
513 /* Inner loop uses 84 flops */
516 /* End of innermost loop */
518 gmx_mm256_update_iforce_3atom_swizzle_ps(fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
519 f+i_coord_offset+DIM,fshift+i_shift_offset);
522 /* Update potential energies */
523 gmx_mm256_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid);
525 /* Increment number of inner iterations */
526 inneriter += j_index_end - j_index_start;
528 /* Outer loop uses 19 flops */
531 /* Increment number of outer iterations */
534 /* Update outer/inner flops */
536 inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_W4_VF,outeriter*19 + inneriter*84);
539 * Gromacs nonbonded kernel: nb_kernel_ElecCoul_VdwNone_GeomW4P1_F_avx_256_single
540 * Electrostatics interaction: Coulomb
541 * VdW interaction: None
542 * Geometry: Water4-Particle
543 * Calculate force/pot: Force
546 nb_kernel_ElecCoul_VdwNone_GeomW4P1_F_avx_256_single
547 (t_nblist * gmx_restrict nlist,
548 rvec * gmx_restrict xx,
549 rvec * gmx_restrict ff,
550 t_forcerec * gmx_restrict fr,
551 t_mdatoms * gmx_restrict mdatoms,
552 nb_kernel_data_t * gmx_restrict kernel_data,
553 t_nrnb * gmx_restrict nrnb)
555 /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
556 * just 0 for non-waters.
557 * Suffixes A,B,C,D,E,F,G,H refer to j loop unrolling done with AVX, e.g. for the eight different
558 * jnr indices corresponding to data put in the four positions in the SIMD register.
560 int i_shift_offset,i_coord_offset,outeriter,inneriter;
561 int j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
562 int jnrA,jnrB,jnrC,jnrD;
563 int jnrE,jnrF,jnrG,jnrH;
564 int jnrlistA,jnrlistB,jnrlistC,jnrlistD;
565 int jnrlistE,jnrlistF,jnrlistG,jnrlistH;
566 int j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
567 int j_coord_offsetE,j_coord_offsetF,j_coord_offsetG,j_coord_offsetH;
568 int *iinr,*jindex,*jjnr,*shiftidx,*gid;
570 real *shiftvec,*fshift,*x,*f;
571 real *fjptrA,*fjptrB,*fjptrC,*fjptrD,*fjptrE,*fjptrF,*fjptrG,*fjptrH;
573 __m256 tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
574 real * vdwioffsetptr1;
575 __m256 ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
576 real * vdwioffsetptr2;
577 __m256 ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
578 real * vdwioffsetptr3;
579 __m256 ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
580 int vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D,vdwjidx0E,vdwjidx0F,vdwjidx0G,vdwjidx0H;
581 __m256 jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
582 __m256 dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
583 __m256 dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
584 __m256 dx30,dy30,dz30,rsq30,rinv30,rinvsq30,r30,qq30,c6_30,c12_30;
585 __m256 velec,felec,velecsum,facel,crf,krf,krf2;
587 __m256 dummy_mask,cutoff_mask;
588 __m256 signbit = _mm256_castsi256_ps( _mm256_set1_epi32(0x80000000) );
589 __m256 one = _mm256_set1_ps(1.0);
590 __m256 two = _mm256_set1_ps(2.0);
596 jindex = nlist->jindex;
598 shiftidx = nlist->shift;
600 shiftvec = fr->shift_vec[0];
601 fshift = fr->fshift[0];
602 facel = _mm256_set1_ps(fr->epsfac);
603 charge = mdatoms->chargeA;
605 /* Setup water-specific parameters */
606 inr = nlist->iinr[0];
607 iq1 = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+1]));
608 iq2 = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+2]));
609 iq3 = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+3]));
611 /* Avoid stupid compiler warnings */
612 jnrA = jnrB = jnrC = jnrD = jnrE = jnrF = jnrG = jnrH = 0;
625 for(iidx=0;iidx<4*DIM;iidx++)
630 /* Start outer loop over neighborlists */
631 for(iidx=0; iidx<nri; iidx++)
633 /* Load shift vector for this list */
634 i_shift_offset = DIM*shiftidx[iidx];
636 /* Load limits for loop over neighbors */
637 j_index_start = jindex[iidx];
638 j_index_end = jindex[iidx+1];
640 /* Get outer coordinate index */
642 i_coord_offset = DIM*inr;
644 /* Load i particle coords and add shift vector */
645 gmx_mm256_load_shift_and_3rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset+DIM,
646 &ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
648 fix1 = _mm256_setzero_ps();
649 fiy1 = _mm256_setzero_ps();
650 fiz1 = _mm256_setzero_ps();
651 fix2 = _mm256_setzero_ps();
652 fiy2 = _mm256_setzero_ps();
653 fiz2 = _mm256_setzero_ps();
654 fix3 = _mm256_setzero_ps();
655 fiy3 = _mm256_setzero_ps();
656 fiz3 = _mm256_setzero_ps();
658 /* Start inner kernel loop */
659 for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+7]>=0; jidx+=8)
662 /* Get j neighbor index, and coordinate index */
671 j_coord_offsetA = DIM*jnrA;
672 j_coord_offsetB = DIM*jnrB;
673 j_coord_offsetC = DIM*jnrC;
674 j_coord_offsetD = DIM*jnrD;
675 j_coord_offsetE = DIM*jnrE;
676 j_coord_offsetF = DIM*jnrF;
677 j_coord_offsetG = DIM*jnrG;
678 j_coord_offsetH = DIM*jnrH;
680 /* load j atom coordinates */
681 gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
682 x+j_coord_offsetC,x+j_coord_offsetD,
683 x+j_coord_offsetE,x+j_coord_offsetF,
684 x+j_coord_offsetG,x+j_coord_offsetH,
687 /* Calculate displacement vector */
688 dx10 = _mm256_sub_ps(ix1,jx0);
689 dy10 = _mm256_sub_ps(iy1,jy0);
690 dz10 = _mm256_sub_ps(iz1,jz0);
691 dx20 = _mm256_sub_ps(ix2,jx0);
692 dy20 = _mm256_sub_ps(iy2,jy0);
693 dz20 = _mm256_sub_ps(iz2,jz0);
694 dx30 = _mm256_sub_ps(ix3,jx0);
695 dy30 = _mm256_sub_ps(iy3,jy0);
696 dz30 = _mm256_sub_ps(iz3,jz0);
698 /* Calculate squared distance and things based on it */
699 rsq10 = gmx_mm256_calc_rsq_ps(dx10,dy10,dz10);
700 rsq20 = gmx_mm256_calc_rsq_ps(dx20,dy20,dz20);
701 rsq30 = gmx_mm256_calc_rsq_ps(dx30,dy30,dz30);
703 rinv10 = gmx_mm256_invsqrt_ps(rsq10);
704 rinv20 = gmx_mm256_invsqrt_ps(rsq20);
705 rinv30 = gmx_mm256_invsqrt_ps(rsq30);
707 rinvsq10 = _mm256_mul_ps(rinv10,rinv10);
708 rinvsq20 = _mm256_mul_ps(rinv20,rinv20);
709 rinvsq30 = _mm256_mul_ps(rinv30,rinv30);
711 /* Load parameters for j particles */
712 jq0 = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
713 charge+jnrC+0,charge+jnrD+0,
714 charge+jnrE+0,charge+jnrF+0,
715 charge+jnrG+0,charge+jnrH+0);
717 fjx0 = _mm256_setzero_ps();
718 fjy0 = _mm256_setzero_ps();
719 fjz0 = _mm256_setzero_ps();
721 /**************************
722 * CALCULATE INTERACTIONS *
723 **************************/
725 /* Compute parameters for interactions between i and j atoms */
726 qq10 = _mm256_mul_ps(iq1,jq0);
728 /* COULOMB ELECTROSTATICS */
729 velec = _mm256_mul_ps(qq10,rinv10);
730 felec = _mm256_mul_ps(velec,rinvsq10);
734 /* Calculate temporary vectorial force */
735 tx = _mm256_mul_ps(fscal,dx10);
736 ty = _mm256_mul_ps(fscal,dy10);
737 tz = _mm256_mul_ps(fscal,dz10);
739 /* Update vectorial force */
740 fix1 = _mm256_add_ps(fix1,tx);
741 fiy1 = _mm256_add_ps(fiy1,ty);
742 fiz1 = _mm256_add_ps(fiz1,tz);
744 fjx0 = _mm256_add_ps(fjx0,tx);
745 fjy0 = _mm256_add_ps(fjy0,ty);
746 fjz0 = _mm256_add_ps(fjz0,tz);
748 /**************************
749 * CALCULATE INTERACTIONS *
750 **************************/
752 /* Compute parameters for interactions between i and j atoms */
753 qq20 = _mm256_mul_ps(iq2,jq0);
755 /* COULOMB ELECTROSTATICS */
756 velec = _mm256_mul_ps(qq20,rinv20);
757 felec = _mm256_mul_ps(velec,rinvsq20);
761 /* Calculate temporary vectorial force */
762 tx = _mm256_mul_ps(fscal,dx20);
763 ty = _mm256_mul_ps(fscal,dy20);
764 tz = _mm256_mul_ps(fscal,dz20);
766 /* Update vectorial force */
767 fix2 = _mm256_add_ps(fix2,tx);
768 fiy2 = _mm256_add_ps(fiy2,ty);
769 fiz2 = _mm256_add_ps(fiz2,tz);
771 fjx0 = _mm256_add_ps(fjx0,tx);
772 fjy0 = _mm256_add_ps(fjy0,ty);
773 fjz0 = _mm256_add_ps(fjz0,tz);
775 /**************************
776 * CALCULATE INTERACTIONS *
777 **************************/
779 /* Compute parameters for interactions between i and j atoms */
780 qq30 = _mm256_mul_ps(iq3,jq0);
782 /* COULOMB ELECTROSTATICS */
783 velec = _mm256_mul_ps(qq30,rinv30);
784 felec = _mm256_mul_ps(velec,rinvsq30);
788 /* Calculate temporary vectorial force */
789 tx = _mm256_mul_ps(fscal,dx30);
790 ty = _mm256_mul_ps(fscal,dy30);
791 tz = _mm256_mul_ps(fscal,dz30);
793 /* Update vectorial force */
794 fix3 = _mm256_add_ps(fix3,tx);
795 fiy3 = _mm256_add_ps(fiy3,ty);
796 fiz3 = _mm256_add_ps(fiz3,tz);
798 fjx0 = _mm256_add_ps(fjx0,tx);
799 fjy0 = _mm256_add_ps(fjy0,ty);
800 fjz0 = _mm256_add_ps(fjz0,tz);
802 fjptrA = f+j_coord_offsetA;
803 fjptrB = f+j_coord_offsetB;
804 fjptrC = f+j_coord_offsetC;
805 fjptrD = f+j_coord_offsetD;
806 fjptrE = f+j_coord_offsetE;
807 fjptrF = f+j_coord_offsetF;
808 fjptrG = f+j_coord_offsetG;
809 fjptrH = f+j_coord_offsetH;
811 gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,fjx0,fjy0,fjz0);
813 /* Inner loop uses 81 flops */
819 /* Get j neighbor index, and coordinate index */
820 jnrlistA = jjnr[jidx];
821 jnrlistB = jjnr[jidx+1];
822 jnrlistC = jjnr[jidx+2];
823 jnrlistD = jjnr[jidx+3];
824 jnrlistE = jjnr[jidx+4];
825 jnrlistF = jjnr[jidx+5];
826 jnrlistG = jjnr[jidx+6];
827 jnrlistH = jjnr[jidx+7];
828 /* Sign of each element will be negative for non-real atoms.
829 * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
830 * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
832 dummy_mask = gmx_mm256_set_m128(gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx+4)),_mm_setzero_si128())),
833 gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128())));
835 jnrA = (jnrlistA>=0) ? jnrlistA : 0;
836 jnrB = (jnrlistB>=0) ? jnrlistB : 0;
837 jnrC = (jnrlistC>=0) ? jnrlistC : 0;
838 jnrD = (jnrlistD>=0) ? jnrlistD : 0;
839 jnrE = (jnrlistE>=0) ? jnrlistE : 0;
840 jnrF = (jnrlistF>=0) ? jnrlistF : 0;
841 jnrG = (jnrlistG>=0) ? jnrlistG : 0;
842 jnrH = (jnrlistH>=0) ? jnrlistH : 0;
843 j_coord_offsetA = DIM*jnrA;
844 j_coord_offsetB = DIM*jnrB;
845 j_coord_offsetC = DIM*jnrC;
846 j_coord_offsetD = DIM*jnrD;
847 j_coord_offsetE = DIM*jnrE;
848 j_coord_offsetF = DIM*jnrF;
849 j_coord_offsetG = DIM*jnrG;
850 j_coord_offsetH = DIM*jnrH;
852 /* load j atom coordinates */
853 gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
854 x+j_coord_offsetC,x+j_coord_offsetD,
855 x+j_coord_offsetE,x+j_coord_offsetF,
856 x+j_coord_offsetG,x+j_coord_offsetH,
859 /* Calculate displacement vector */
860 dx10 = _mm256_sub_ps(ix1,jx0);
861 dy10 = _mm256_sub_ps(iy1,jy0);
862 dz10 = _mm256_sub_ps(iz1,jz0);
863 dx20 = _mm256_sub_ps(ix2,jx0);
864 dy20 = _mm256_sub_ps(iy2,jy0);
865 dz20 = _mm256_sub_ps(iz2,jz0);
866 dx30 = _mm256_sub_ps(ix3,jx0);
867 dy30 = _mm256_sub_ps(iy3,jy0);
868 dz30 = _mm256_sub_ps(iz3,jz0);
870 /* Calculate squared distance and things based on it */
871 rsq10 = gmx_mm256_calc_rsq_ps(dx10,dy10,dz10);
872 rsq20 = gmx_mm256_calc_rsq_ps(dx20,dy20,dz20);
873 rsq30 = gmx_mm256_calc_rsq_ps(dx30,dy30,dz30);
875 rinv10 = gmx_mm256_invsqrt_ps(rsq10);
876 rinv20 = gmx_mm256_invsqrt_ps(rsq20);
877 rinv30 = gmx_mm256_invsqrt_ps(rsq30);
879 rinvsq10 = _mm256_mul_ps(rinv10,rinv10);
880 rinvsq20 = _mm256_mul_ps(rinv20,rinv20);
881 rinvsq30 = _mm256_mul_ps(rinv30,rinv30);
883 /* Load parameters for j particles */
884 jq0 = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
885 charge+jnrC+0,charge+jnrD+0,
886 charge+jnrE+0,charge+jnrF+0,
887 charge+jnrG+0,charge+jnrH+0);
889 fjx0 = _mm256_setzero_ps();
890 fjy0 = _mm256_setzero_ps();
891 fjz0 = _mm256_setzero_ps();
893 /**************************
894 * CALCULATE INTERACTIONS *
895 **************************/
897 /* Compute parameters for interactions between i and j atoms */
898 qq10 = _mm256_mul_ps(iq1,jq0);
900 /* COULOMB ELECTROSTATICS */
901 velec = _mm256_mul_ps(qq10,rinv10);
902 felec = _mm256_mul_ps(velec,rinvsq10);
906 fscal = _mm256_andnot_ps(dummy_mask,fscal);
908 /* Calculate temporary vectorial force */
909 tx = _mm256_mul_ps(fscal,dx10);
910 ty = _mm256_mul_ps(fscal,dy10);
911 tz = _mm256_mul_ps(fscal,dz10);
913 /* Update vectorial force */
914 fix1 = _mm256_add_ps(fix1,tx);
915 fiy1 = _mm256_add_ps(fiy1,ty);
916 fiz1 = _mm256_add_ps(fiz1,tz);
918 fjx0 = _mm256_add_ps(fjx0,tx);
919 fjy0 = _mm256_add_ps(fjy0,ty);
920 fjz0 = _mm256_add_ps(fjz0,tz);
922 /**************************
923 * CALCULATE INTERACTIONS *
924 **************************/
926 /* Compute parameters for interactions between i and j atoms */
927 qq20 = _mm256_mul_ps(iq2,jq0);
929 /* COULOMB ELECTROSTATICS */
930 velec = _mm256_mul_ps(qq20,rinv20);
931 felec = _mm256_mul_ps(velec,rinvsq20);
935 fscal = _mm256_andnot_ps(dummy_mask,fscal);
937 /* Calculate temporary vectorial force */
938 tx = _mm256_mul_ps(fscal,dx20);
939 ty = _mm256_mul_ps(fscal,dy20);
940 tz = _mm256_mul_ps(fscal,dz20);
942 /* Update vectorial force */
943 fix2 = _mm256_add_ps(fix2,tx);
944 fiy2 = _mm256_add_ps(fiy2,ty);
945 fiz2 = _mm256_add_ps(fiz2,tz);
947 fjx0 = _mm256_add_ps(fjx0,tx);
948 fjy0 = _mm256_add_ps(fjy0,ty);
949 fjz0 = _mm256_add_ps(fjz0,tz);
951 /**************************
952 * CALCULATE INTERACTIONS *
953 **************************/
955 /* Compute parameters for interactions between i and j atoms */
956 qq30 = _mm256_mul_ps(iq3,jq0);
958 /* COULOMB ELECTROSTATICS */
959 velec = _mm256_mul_ps(qq30,rinv30);
960 felec = _mm256_mul_ps(velec,rinvsq30);
964 fscal = _mm256_andnot_ps(dummy_mask,fscal);
966 /* Calculate temporary vectorial force */
967 tx = _mm256_mul_ps(fscal,dx30);
968 ty = _mm256_mul_ps(fscal,dy30);
969 tz = _mm256_mul_ps(fscal,dz30);
971 /* Update vectorial force */
972 fix3 = _mm256_add_ps(fix3,tx);
973 fiy3 = _mm256_add_ps(fiy3,ty);
974 fiz3 = _mm256_add_ps(fiz3,tz);
976 fjx0 = _mm256_add_ps(fjx0,tx);
977 fjy0 = _mm256_add_ps(fjy0,ty);
978 fjz0 = _mm256_add_ps(fjz0,tz);
980 fjptrA = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
981 fjptrB = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
982 fjptrC = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
983 fjptrD = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
984 fjptrE = (jnrlistE>=0) ? f+j_coord_offsetE : scratch;
985 fjptrF = (jnrlistF>=0) ? f+j_coord_offsetF : scratch;
986 fjptrG = (jnrlistG>=0) ? f+j_coord_offsetG : scratch;
987 fjptrH = (jnrlistH>=0) ? f+j_coord_offsetH : scratch;
989 gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,fjx0,fjy0,fjz0);
991 /* Inner loop uses 81 flops */
994 /* End of innermost loop */
996 gmx_mm256_update_iforce_3atom_swizzle_ps(fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
997 f+i_coord_offset+DIM,fshift+i_shift_offset);
999 /* Increment number of inner iterations */
1000 inneriter += j_index_end - j_index_start;
1002 /* Outer loop uses 18 flops */
1005 /* Increment number of outer iterations */
1008 /* Update outer/inner flops */
1010 inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_W4_F,outeriter*18 + inneriter*81);