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_ElecRF_VdwNone_GeomW4P1_VF_avx_256_single
38 * Electrostatics interaction: ReactionField
39 * VdW interaction: None
40 * Geometry: Water4-Particle
41 * Calculate force/pot: PotentialAndForce
44 nb_kernel_ElecRF_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;
102 krf = _mm256_set1_ps(fr->ic->k_rf);
103 krf2 = _mm256_set1_ps(fr->ic->k_rf*2.0);
104 crf = _mm256_set1_ps(fr->ic->c_rf);
106 /* Setup water-specific parameters */
107 inr = nlist->iinr[0];
108 iq1 = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+1]));
109 iq2 = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+2]));
110 iq3 = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+3]));
112 /* Avoid stupid compiler warnings */
113 jnrA = jnrB = jnrC = jnrD = jnrE = jnrF = jnrG = jnrH = 0;
126 for(iidx=0;iidx<4*DIM;iidx++)
131 /* Start outer loop over neighborlists */
132 for(iidx=0; iidx<nri; iidx++)
134 /* Load shift vector for this list */
135 i_shift_offset = DIM*shiftidx[iidx];
137 /* Load limits for loop over neighbors */
138 j_index_start = jindex[iidx];
139 j_index_end = jindex[iidx+1];
141 /* Get outer coordinate index */
143 i_coord_offset = DIM*inr;
145 /* Load i particle coords and add shift vector */
146 gmx_mm256_load_shift_and_3rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset+DIM,
147 &ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
149 fix1 = _mm256_setzero_ps();
150 fiy1 = _mm256_setzero_ps();
151 fiz1 = _mm256_setzero_ps();
152 fix2 = _mm256_setzero_ps();
153 fiy2 = _mm256_setzero_ps();
154 fiz2 = _mm256_setzero_ps();
155 fix3 = _mm256_setzero_ps();
156 fiy3 = _mm256_setzero_ps();
157 fiz3 = _mm256_setzero_ps();
159 /* Reset potential sums */
160 velecsum = _mm256_setzero_ps();
162 /* Start inner kernel loop */
163 for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+7]>=0; jidx+=8)
166 /* Get j neighbor index, and coordinate index */
175 j_coord_offsetA = DIM*jnrA;
176 j_coord_offsetB = DIM*jnrB;
177 j_coord_offsetC = DIM*jnrC;
178 j_coord_offsetD = DIM*jnrD;
179 j_coord_offsetE = DIM*jnrE;
180 j_coord_offsetF = DIM*jnrF;
181 j_coord_offsetG = DIM*jnrG;
182 j_coord_offsetH = DIM*jnrH;
184 /* load j atom coordinates */
185 gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
186 x+j_coord_offsetC,x+j_coord_offsetD,
187 x+j_coord_offsetE,x+j_coord_offsetF,
188 x+j_coord_offsetG,x+j_coord_offsetH,
191 /* Calculate displacement vector */
192 dx10 = _mm256_sub_ps(ix1,jx0);
193 dy10 = _mm256_sub_ps(iy1,jy0);
194 dz10 = _mm256_sub_ps(iz1,jz0);
195 dx20 = _mm256_sub_ps(ix2,jx0);
196 dy20 = _mm256_sub_ps(iy2,jy0);
197 dz20 = _mm256_sub_ps(iz2,jz0);
198 dx30 = _mm256_sub_ps(ix3,jx0);
199 dy30 = _mm256_sub_ps(iy3,jy0);
200 dz30 = _mm256_sub_ps(iz3,jz0);
202 /* Calculate squared distance and things based on it */
203 rsq10 = gmx_mm256_calc_rsq_ps(dx10,dy10,dz10);
204 rsq20 = gmx_mm256_calc_rsq_ps(dx20,dy20,dz20);
205 rsq30 = gmx_mm256_calc_rsq_ps(dx30,dy30,dz30);
207 rinv10 = gmx_mm256_invsqrt_ps(rsq10);
208 rinv20 = gmx_mm256_invsqrt_ps(rsq20);
209 rinv30 = gmx_mm256_invsqrt_ps(rsq30);
211 rinvsq10 = _mm256_mul_ps(rinv10,rinv10);
212 rinvsq20 = _mm256_mul_ps(rinv20,rinv20);
213 rinvsq30 = _mm256_mul_ps(rinv30,rinv30);
215 /* Load parameters for j particles */
216 jq0 = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
217 charge+jnrC+0,charge+jnrD+0,
218 charge+jnrE+0,charge+jnrF+0,
219 charge+jnrG+0,charge+jnrH+0);
221 fjx0 = _mm256_setzero_ps();
222 fjy0 = _mm256_setzero_ps();
223 fjz0 = _mm256_setzero_ps();
225 /**************************
226 * CALCULATE INTERACTIONS *
227 **************************/
229 /* Compute parameters for interactions between i and j atoms */
230 qq10 = _mm256_mul_ps(iq1,jq0);
232 /* REACTION-FIELD ELECTROSTATICS */
233 velec = _mm256_mul_ps(qq10,_mm256_sub_ps(_mm256_add_ps(rinv10,_mm256_mul_ps(krf,rsq10)),crf));
234 felec = _mm256_mul_ps(qq10,_mm256_sub_ps(_mm256_mul_ps(rinv10,rinvsq10),krf2));
236 /* Update potential sum for this i atom from the interaction with this j atom. */
237 velecsum = _mm256_add_ps(velecsum,velec);
241 /* Calculate temporary vectorial force */
242 tx = _mm256_mul_ps(fscal,dx10);
243 ty = _mm256_mul_ps(fscal,dy10);
244 tz = _mm256_mul_ps(fscal,dz10);
246 /* Update vectorial force */
247 fix1 = _mm256_add_ps(fix1,tx);
248 fiy1 = _mm256_add_ps(fiy1,ty);
249 fiz1 = _mm256_add_ps(fiz1,tz);
251 fjx0 = _mm256_add_ps(fjx0,tx);
252 fjy0 = _mm256_add_ps(fjy0,ty);
253 fjz0 = _mm256_add_ps(fjz0,tz);
255 /**************************
256 * CALCULATE INTERACTIONS *
257 **************************/
259 /* Compute parameters for interactions between i and j atoms */
260 qq20 = _mm256_mul_ps(iq2,jq0);
262 /* REACTION-FIELD ELECTROSTATICS */
263 velec = _mm256_mul_ps(qq20,_mm256_sub_ps(_mm256_add_ps(rinv20,_mm256_mul_ps(krf,rsq20)),crf));
264 felec = _mm256_mul_ps(qq20,_mm256_sub_ps(_mm256_mul_ps(rinv20,rinvsq20),krf2));
266 /* Update potential sum for this i atom from the interaction with this j atom. */
267 velecsum = _mm256_add_ps(velecsum,velec);
271 /* Calculate temporary vectorial force */
272 tx = _mm256_mul_ps(fscal,dx20);
273 ty = _mm256_mul_ps(fscal,dy20);
274 tz = _mm256_mul_ps(fscal,dz20);
276 /* Update vectorial force */
277 fix2 = _mm256_add_ps(fix2,tx);
278 fiy2 = _mm256_add_ps(fiy2,ty);
279 fiz2 = _mm256_add_ps(fiz2,tz);
281 fjx0 = _mm256_add_ps(fjx0,tx);
282 fjy0 = _mm256_add_ps(fjy0,ty);
283 fjz0 = _mm256_add_ps(fjz0,tz);
285 /**************************
286 * CALCULATE INTERACTIONS *
287 **************************/
289 /* Compute parameters for interactions between i and j atoms */
290 qq30 = _mm256_mul_ps(iq3,jq0);
292 /* REACTION-FIELD ELECTROSTATICS */
293 velec = _mm256_mul_ps(qq30,_mm256_sub_ps(_mm256_add_ps(rinv30,_mm256_mul_ps(krf,rsq30)),crf));
294 felec = _mm256_mul_ps(qq30,_mm256_sub_ps(_mm256_mul_ps(rinv30,rinvsq30),krf2));
296 /* Update potential sum for this i atom from the interaction with this j atom. */
297 velecsum = _mm256_add_ps(velecsum,velec);
301 /* Calculate temporary vectorial force */
302 tx = _mm256_mul_ps(fscal,dx30);
303 ty = _mm256_mul_ps(fscal,dy30);
304 tz = _mm256_mul_ps(fscal,dz30);
306 /* Update vectorial force */
307 fix3 = _mm256_add_ps(fix3,tx);
308 fiy3 = _mm256_add_ps(fiy3,ty);
309 fiz3 = _mm256_add_ps(fiz3,tz);
311 fjx0 = _mm256_add_ps(fjx0,tx);
312 fjy0 = _mm256_add_ps(fjy0,ty);
313 fjz0 = _mm256_add_ps(fjz0,tz);
315 fjptrA = f+j_coord_offsetA;
316 fjptrB = f+j_coord_offsetB;
317 fjptrC = f+j_coord_offsetC;
318 fjptrD = f+j_coord_offsetD;
319 fjptrE = f+j_coord_offsetE;
320 fjptrF = f+j_coord_offsetF;
321 fjptrG = f+j_coord_offsetG;
322 fjptrH = f+j_coord_offsetH;
324 gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,fjx0,fjy0,fjz0);
326 /* Inner loop uses 99 flops */
332 /* Get j neighbor index, and coordinate index */
333 jnrlistA = jjnr[jidx];
334 jnrlistB = jjnr[jidx+1];
335 jnrlistC = jjnr[jidx+2];
336 jnrlistD = jjnr[jidx+3];
337 jnrlistE = jjnr[jidx+4];
338 jnrlistF = jjnr[jidx+5];
339 jnrlistG = jjnr[jidx+6];
340 jnrlistH = jjnr[jidx+7];
341 /* Sign of each element will be negative for non-real atoms.
342 * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
343 * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
345 dummy_mask = gmx_mm256_set_m128(gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx+4)),_mm_setzero_si128())),
346 gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128())));
348 jnrA = (jnrlistA>=0) ? jnrlistA : 0;
349 jnrB = (jnrlistB>=0) ? jnrlistB : 0;
350 jnrC = (jnrlistC>=0) ? jnrlistC : 0;
351 jnrD = (jnrlistD>=0) ? jnrlistD : 0;
352 jnrE = (jnrlistE>=0) ? jnrlistE : 0;
353 jnrF = (jnrlistF>=0) ? jnrlistF : 0;
354 jnrG = (jnrlistG>=0) ? jnrlistG : 0;
355 jnrH = (jnrlistH>=0) ? jnrlistH : 0;
356 j_coord_offsetA = DIM*jnrA;
357 j_coord_offsetB = DIM*jnrB;
358 j_coord_offsetC = DIM*jnrC;
359 j_coord_offsetD = DIM*jnrD;
360 j_coord_offsetE = DIM*jnrE;
361 j_coord_offsetF = DIM*jnrF;
362 j_coord_offsetG = DIM*jnrG;
363 j_coord_offsetH = DIM*jnrH;
365 /* load j atom coordinates */
366 gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
367 x+j_coord_offsetC,x+j_coord_offsetD,
368 x+j_coord_offsetE,x+j_coord_offsetF,
369 x+j_coord_offsetG,x+j_coord_offsetH,
372 /* Calculate displacement vector */
373 dx10 = _mm256_sub_ps(ix1,jx0);
374 dy10 = _mm256_sub_ps(iy1,jy0);
375 dz10 = _mm256_sub_ps(iz1,jz0);
376 dx20 = _mm256_sub_ps(ix2,jx0);
377 dy20 = _mm256_sub_ps(iy2,jy0);
378 dz20 = _mm256_sub_ps(iz2,jz0);
379 dx30 = _mm256_sub_ps(ix3,jx0);
380 dy30 = _mm256_sub_ps(iy3,jy0);
381 dz30 = _mm256_sub_ps(iz3,jz0);
383 /* Calculate squared distance and things based on it */
384 rsq10 = gmx_mm256_calc_rsq_ps(dx10,dy10,dz10);
385 rsq20 = gmx_mm256_calc_rsq_ps(dx20,dy20,dz20);
386 rsq30 = gmx_mm256_calc_rsq_ps(dx30,dy30,dz30);
388 rinv10 = gmx_mm256_invsqrt_ps(rsq10);
389 rinv20 = gmx_mm256_invsqrt_ps(rsq20);
390 rinv30 = gmx_mm256_invsqrt_ps(rsq30);
392 rinvsq10 = _mm256_mul_ps(rinv10,rinv10);
393 rinvsq20 = _mm256_mul_ps(rinv20,rinv20);
394 rinvsq30 = _mm256_mul_ps(rinv30,rinv30);
396 /* Load parameters for j particles */
397 jq0 = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
398 charge+jnrC+0,charge+jnrD+0,
399 charge+jnrE+0,charge+jnrF+0,
400 charge+jnrG+0,charge+jnrH+0);
402 fjx0 = _mm256_setzero_ps();
403 fjy0 = _mm256_setzero_ps();
404 fjz0 = _mm256_setzero_ps();
406 /**************************
407 * CALCULATE INTERACTIONS *
408 **************************/
410 /* Compute parameters for interactions between i and j atoms */
411 qq10 = _mm256_mul_ps(iq1,jq0);
413 /* REACTION-FIELD ELECTROSTATICS */
414 velec = _mm256_mul_ps(qq10,_mm256_sub_ps(_mm256_add_ps(rinv10,_mm256_mul_ps(krf,rsq10)),crf));
415 felec = _mm256_mul_ps(qq10,_mm256_sub_ps(_mm256_mul_ps(rinv10,rinvsq10),krf2));
417 /* Update potential sum for this i atom from the interaction with this j atom. */
418 velec = _mm256_andnot_ps(dummy_mask,velec);
419 velecsum = _mm256_add_ps(velecsum,velec);
423 fscal = _mm256_andnot_ps(dummy_mask,fscal);
425 /* Calculate temporary vectorial force */
426 tx = _mm256_mul_ps(fscal,dx10);
427 ty = _mm256_mul_ps(fscal,dy10);
428 tz = _mm256_mul_ps(fscal,dz10);
430 /* Update vectorial force */
431 fix1 = _mm256_add_ps(fix1,tx);
432 fiy1 = _mm256_add_ps(fiy1,ty);
433 fiz1 = _mm256_add_ps(fiz1,tz);
435 fjx0 = _mm256_add_ps(fjx0,tx);
436 fjy0 = _mm256_add_ps(fjy0,ty);
437 fjz0 = _mm256_add_ps(fjz0,tz);
439 /**************************
440 * CALCULATE INTERACTIONS *
441 **************************/
443 /* Compute parameters for interactions between i and j atoms */
444 qq20 = _mm256_mul_ps(iq2,jq0);
446 /* REACTION-FIELD ELECTROSTATICS */
447 velec = _mm256_mul_ps(qq20,_mm256_sub_ps(_mm256_add_ps(rinv20,_mm256_mul_ps(krf,rsq20)),crf));
448 felec = _mm256_mul_ps(qq20,_mm256_sub_ps(_mm256_mul_ps(rinv20,rinvsq20),krf2));
450 /* Update potential sum for this i atom from the interaction with this j atom. */
451 velec = _mm256_andnot_ps(dummy_mask,velec);
452 velecsum = _mm256_add_ps(velecsum,velec);
456 fscal = _mm256_andnot_ps(dummy_mask,fscal);
458 /* Calculate temporary vectorial force */
459 tx = _mm256_mul_ps(fscal,dx20);
460 ty = _mm256_mul_ps(fscal,dy20);
461 tz = _mm256_mul_ps(fscal,dz20);
463 /* Update vectorial force */
464 fix2 = _mm256_add_ps(fix2,tx);
465 fiy2 = _mm256_add_ps(fiy2,ty);
466 fiz2 = _mm256_add_ps(fiz2,tz);
468 fjx0 = _mm256_add_ps(fjx0,tx);
469 fjy0 = _mm256_add_ps(fjy0,ty);
470 fjz0 = _mm256_add_ps(fjz0,tz);
472 /**************************
473 * CALCULATE INTERACTIONS *
474 **************************/
476 /* Compute parameters for interactions between i and j atoms */
477 qq30 = _mm256_mul_ps(iq3,jq0);
479 /* REACTION-FIELD ELECTROSTATICS */
480 velec = _mm256_mul_ps(qq30,_mm256_sub_ps(_mm256_add_ps(rinv30,_mm256_mul_ps(krf,rsq30)),crf));
481 felec = _mm256_mul_ps(qq30,_mm256_sub_ps(_mm256_mul_ps(rinv30,rinvsq30),krf2));
483 /* Update potential sum for this i atom from the interaction with this j atom. */
484 velec = _mm256_andnot_ps(dummy_mask,velec);
485 velecsum = _mm256_add_ps(velecsum,velec);
489 fscal = _mm256_andnot_ps(dummy_mask,fscal);
491 /* Calculate temporary vectorial force */
492 tx = _mm256_mul_ps(fscal,dx30);
493 ty = _mm256_mul_ps(fscal,dy30);
494 tz = _mm256_mul_ps(fscal,dz30);
496 /* Update vectorial force */
497 fix3 = _mm256_add_ps(fix3,tx);
498 fiy3 = _mm256_add_ps(fiy3,ty);
499 fiz3 = _mm256_add_ps(fiz3,tz);
501 fjx0 = _mm256_add_ps(fjx0,tx);
502 fjy0 = _mm256_add_ps(fjy0,ty);
503 fjz0 = _mm256_add_ps(fjz0,tz);
505 fjptrA = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
506 fjptrB = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
507 fjptrC = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
508 fjptrD = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
509 fjptrE = (jnrlistE>=0) ? f+j_coord_offsetE : scratch;
510 fjptrF = (jnrlistF>=0) ? f+j_coord_offsetF : scratch;
511 fjptrG = (jnrlistG>=0) ? f+j_coord_offsetG : scratch;
512 fjptrH = (jnrlistH>=0) ? f+j_coord_offsetH : scratch;
514 gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,fjx0,fjy0,fjz0);
516 /* Inner loop uses 99 flops */
519 /* End of innermost loop */
521 gmx_mm256_update_iforce_3atom_swizzle_ps(fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
522 f+i_coord_offset+DIM,fshift+i_shift_offset);
525 /* Update potential energies */
526 gmx_mm256_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid);
528 /* Increment number of inner iterations */
529 inneriter += j_index_end - j_index_start;
531 /* Outer loop uses 19 flops */
534 /* Increment number of outer iterations */
537 /* Update outer/inner flops */
539 inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_W4_VF,outeriter*19 + inneriter*99);
542 * Gromacs nonbonded kernel: nb_kernel_ElecRF_VdwNone_GeomW4P1_F_avx_256_single
543 * Electrostatics interaction: ReactionField
544 * VdW interaction: None
545 * Geometry: Water4-Particle
546 * Calculate force/pot: Force
549 nb_kernel_ElecRF_VdwNone_GeomW4P1_F_avx_256_single
550 (t_nblist * gmx_restrict nlist,
551 rvec * gmx_restrict xx,
552 rvec * gmx_restrict ff,
553 t_forcerec * gmx_restrict fr,
554 t_mdatoms * gmx_restrict mdatoms,
555 nb_kernel_data_t * gmx_restrict kernel_data,
556 t_nrnb * gmx_restrict nrnb)
558 /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
559 * just 0 for non-waters.
560 * Suffixes A,B,C,D,E,F,G,H refer to j loop unrolling done with AVX, e.g. for the eight different
561 * jnr indices corresponding to data put in the four positions in the SIMD register.
563 int i_shift_offset,i_coord_offset,outeriter,inneriter;
564 int j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
565 int jnrA,jnrB,jnrC,jnrD;
566 int jnrE,jnrF,jnrG,jnrH;
567 int jnrlistA,jnrlistB,jnrlistC,jnrlistD;
568 int jnrlistE,jnrlistF,jnrlistG,jnrlistH;
569 int j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
570 int j_coord_offsetE,j_coord_offsetF,j_coord_offsetG,j_coord_offsetH;
571 int *iinr,*jindex,*jjnr,*shiftidx,*gid;
573 real *shiftvec,*fshift,*x,*f;
574 real *fjptrA,*fjptrB,*fjptrC,*fjptrD,*fjptrE,*fjptrF,*fjptrG,*fjptrH;
576 __m256 tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
577 real * vdwioffsetptr1;
578 __m256 ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
579 real * vdwioffsetptr2;
580 __m256 ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
581 real * vdwioffsetptr3;
582 __m256 ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
583 int vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D,vdwjidx0E,vdwjidx0F,vdwjidx0G,vdwjidx0H;
584 __m256 jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
585 __m256 dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
586 __m256 dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
587 __m256 dx30,dy30,dz30,rsq30,rinv30,rinvsq30,r30,qq30,c6_30,c12_30;
588 __m256 velec,felec,velecsum,facel,crf,krf,krf2;
590 __m256 dummy_mask,cutoff_mask;
591 __m256 signbit = _mm256_castsi256_ps( _mm256_set1_epi32(0x80000000) );
592 __m256 one = _mm256_set1_ps(1.0);
593 __m256 two = _mm256_set1_ps(2.0);
599 jindex = nlist->jindex;
601 shiftidx = nlist->shift;
603 shiftvec = fr->shift_vec[0];
604 fshift = fr->fshift[0];
605 facel = _mm256_set1_ps(fr->epsfac);
606 charge = mdatoms->chargeA;
607 krf = _mm256_set1_ps(fr->ic->k_rf);
608 krf2 = _mm256_set1_ps(fr->ic->k_rf*2.0);
609 crf = _mm256_set1_ps(fr->ic->c_rf);
611 /* Setup water-specific parameters */
612 inr = nlist->iinr[0];
613 iq1 = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+1]));
614 iq2 = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+2]));
615 iq3 = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+3]));
617 /* Avoid stupid compiler warnings */
618 jnrA = jnrB = jnrC = jnrD = jnrE = jnrF = jnrG = jnrH = 0;
631 for(iidx=0;iidx<4*DIM;iidx++)
636 /* Start outer loop over neighborlists */
637 for(iidx=0; iidx<nri; iidx++)
639 /* Load shift vector for this list */
640 i_shift_offset = DIM*shiftidx[iidx];
642 /* Load limits for loop over neighbors */
643 j_index_start = jindex[iidx];
644 j_index_end = jindex[iidx+1];
646 /* Get outer coordinate index */
648 i_coord_offset = DIM*inr;
650 /* Load i particle coords and add shift vector */
651 gmx_mm256_load_shift_and_3rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset+DIM,
652 &ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
654 fix1 = _mm256_setzero_ps();
655 fiy1 = _mm256_setzero_ps();
656 fiz1 = _mm256_setzero_ps();
657 fix2 = _mm256_setzero_ps();
658 fiy2 = _mm256_setzero_ps();
659 fiz2 = _mm256_setzero_ps();
660 fix3 = _mm256_setzero_ps();
661 fiy3 = _mm256_setzero_ps();
662 fiz3 = _mm256_setzero_ps();
664 /* Start inner kernel loop */
665 for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+7]>=0; jidx+=8)
668 /* Get j neighbor index, and coordinate index */
677 j_coord_offsetA = DIM*jnrA;
678 j_coord_offsetB = DIM*jnrB;
679 j_coord_offsetC = DIM*jnrC;
680 j_coord_offsetD = DIM*jnrD;
681 j_coord_offsetE = DIM*jnrE;
682 j_coord_offsetF = DIM*jnrF;
683 j_coord_offsetG = DIM*jnrG;
684 j_coord_offsetH = DIM*jnrH;
686 /* load j atom coordinates */
687 gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
688 x+j_coord_offsetC,x+j_coord_offsetD,
689 x+j_coord_offsetE,x+j_coord_offsetF,
690 x+j_coord_offsetG,x+j_coord_offsetH,
693 /* Calculate displacement vector */
694 dx10 = _mm256_sub_ps(ix1,jx0);
695 dy10 = _mm256_sub_ps(iy1,jy0);
696 dz10 = _mm256_sub_ps(iz1,jz0);
697 dx20 = _mm256_sub_ps(ix2,jx0);
698 dy20 = _mm256_sub_ps(iy2,jy0);
699 dz20 = _mm256_sub_ps(iz2,jz0);
700 dx30 = _mm256_sub_ps(ix3,jx0);
701 dy30 = _mm256_sub_ps(iy3,jy0);
702 dz30 = _mm256_sub_ps(iz3,jz0);
704 /* Calculate squared distance and things based on it */
705 rsq10 = gmx_mm256_calc_rsq_ps(dx10,dy10,dz10);
706 rsq20 = gmx_mm256_calc_rsq_ps(dx20,dy20,dz20);
707 rsq30 = gmx_mm256_calc_rsq_ps(dx30,dy30,dz30);
709 rinv10 = gmx_mm256_invsqrt_ps(rsq10);
710 rinv20 = gmx_mm256_invsqrt_ps(rsq20);
711 rinv30 = gmx_mm256_invsqrt_ps(rsq30);
713 rinvsq10 = _mm256_mul_ps(rinv10,rinv10);
714 rinvsq20 = _mm256_mul_ps(rinv20,rinv20);
715 rinvsq30 = _mm256_mul_ps(rinv30,rinv30);
717 /* Load parameters for j particles */
718 jq0 = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
719 charge+jnrC+0,charge+jnrD+0,
720 charge+jnrE+0,charge+jnrF+0,
721 charge+jnrG+0,charge+jnrH+0);
723 fjx0 = _mm256_setzero_ps();
724 fjy0 = _mm256_setzero_ps();
725 fjz0 = _mm256_setzero_ps();
727 /**************************
728 * CALCULATE INTERACTIONS *
729 **************************/
731 /* Compute parameters for interactions between i and j atoms */
732 qq10 = _mm256_mul_ps(iq1,jq0);
734 /* REACTION-FIELD ELECTROSTATICS */
735 felec = _mm256_mul_ps(qq10,_mm256_sub_ps(_mm256_mul_ps(rinv10,rinvsq10),krf2));
739 /* Calculate temporary vectorial force */
740 tx = _mm256_mul_ps(fscal,dx10);
741 ty = _mm256_mul_ps(fscal,dy10);
742 tz = _mm256_mul_ps(fscal,dz10);
744 /* Update vectorial force */
745 fix1 = _mm256_add_ps(fix1,tx);
746 fiy1 = _mm256_add_ps(fiy1,ty);
747 fiz1 = _mm256_add_ps(fiz1,tz);
749 fjx0 = _mm256_add_ps(fjx0,tx);
750 fjy0 = _mm256_add_ps(fjy0,ty);
751 fjz0 = _mm256_add_ps(fjz0,tz);
753 /**************************
754 * CALCULATE INTERACTIONS *
755 **************************/
757 /* Compute parameters for interactions between i and j atoms */
758 qq20 = _mm256_mul_ps(iq2,jq0);
760 /* REACTION-FIELD ELECTROSTATICS */
761 felec = _mm256_mul_ps(qq20,_mm256_sub_ps(_mm256_mul_ps(rinv20,rinvsq20),krf2));
765 /* Calculate temporary vectorial force */
766 tx = _mm256_mul_ps(fscal,dx20);
767 ty = _mm256_mul_ps(fscal,dy20);
768 tz = _mm256_mul_ps(fscal,dz20);
770 /* Update vectorial force */
771 fix2 = _mm256_add_ps(fix2,tx);
772 fiy2 = _mm256_add_ps(fiy2,ty);
773 fiz2 = _mm256_add_ps(fiz2,tz);
775 fjx0 = _mm256_add_ps(fjx0,tx);
776 fjy0 = _mm256_add_ps(fjy0,ty);
777 fjz0 = _mm256_add_ps(fjz0,tz);
779 /**************************
780 * CALCULATE INTERACTIONS *
781 **************************/
783 /* Compute parameters for interactions between i and j atoms */
784 qq30 = _mm256_mul_ps(iq3,jq0);
786 /* REACTION-FIELD ELECTROSTATICS */
787 felec = _mm256_mul_ps(qq30,_mm256_sub_ps(_mm256_mul_ps(rinv30,rinvsq30),krf2));
791 /* Calculate temporary vectorial force */
792 tx = _mm256_mul_ps(fscal,dx30);
793 ty = _mm256_mul_ps(fscal,dy30);
794 tz = _mm256_mul_ps(fscal,dz30);
796 /* Update vectorial force */
797 fix3 = _mm256_add_ps(fix3,tx);
798 fiy3 = _mm256_add_ps(fiy3,ty);
799 fiz3 = _mm256_add_ps(fiz3,tz);
801 fjx0 = _mm256_add_ps(fjx0,tx);
802 fjy0 = _mm256_add_ps(fjy0,ty);
803 fjz0 = _mm256_add_ps(fjz0,tz);
805 fjptrA = f+j_coord_offsetA;
806 fjptrB = f+j_coord_offsetB;
807 fjptrC = f+j_coord_offsetC;
808 fjptrD = f+j_coord_offsetD;
809 fjptrE = f+j_coord_offsetE;
810 fjptrF = f+j_coord_offsetF;
811 fjptrG = f+j_coord_offsetG;
812 fjptrH = f+j_coord_offsetH;
814 gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,fjx0,fjy0,fjz0);
816 /* Inner loop uses 84 flops */
822 /* Get j neighbor index, and coordinate index */
823 jnrlistA = jjnr[jidx];
824 jnrlistB = jjnr[jidx+1];
825 jnrlistC = jjnr[jidx+2];
826 jnrlistD = jjnr[jidx+3];
827 jnrlistE = jjnr[jidx+4];
828 jnrlistF = jjnr[jidx+5];
829 jnrlistG = jjnr[jidx+6];
830 jnrlistH = jjnr[jidx+7];
831 /* Sign of each element will be negative for non-real atoms.
832 * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
833 * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
835 dummy_mask = gmx_mm256_set_m128(gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx+4)),_mm_setzero_si128())),
836 gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128())));
838 jnrA = (jnrlistA>=0) ? jnrlistA : 0;
839 jnrB = (jnrlistB>=0) ? jnrlistB : 0;
840 jnrC = (jnrlistC>=0) ? jnrlistC : 0;
841 jnrD = (jnrlistD>=0) ? jnrlistD : 0;
842 jnrE = (jnrlistE>=0) ? jnrlistE : 0;
843 jnrF = (jnrlistF>=0) ? jnrlistF : 0;
844 jnrG = (jnrlistG>=0) ? jnrlistG : 0;
845 jnrH = (jnrlistH>=0) ? jnrlistH : 0;
846 j_coord_offsetA = DIM*jnrA;
847 j_coord_offsetB = DIM*jnrB;
848 j_coord_offsetC = DIM*jnrC;
849 j_coord_offsetD = DIM*jnrD;
850 j_coord_offsetE = DIM*jnrE;
851 j_coord_offsetF = DIM*jnrF;
852 j_coord_offsetG = DIM*jnrG;
853 j_coord_offsetH = DIM*jnrH;
855 /* load j atom coordinates */
856 gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
857 x+j_coord_offsetC,x+j_coord_offsetD,
858 x+j_coord_offsetE,x+j_coord_offsetF,
859 x+j_coord_offsetG,x+j_coord_offsetH,
862 /* Calculate displacement vector */
863 dx10 = _mm256_sub_ps(ix1,jx0);
864 dy10 = _mm256_sub_ps(iy1,jy0);
865 dz10 = _mm256_sub_ps(iz1,jz0);
866 dx20 = _mm256_sub_ps(ix2,jx0);
867 dy20 = _mm256_sub_ps(iy2,jy0);
868 dz20 = _mm256_sub_ps(iz2,jz0);
869 dx30 = _mm256_sub_ps(ix3,jx0);
870 dy30 = _mm256_sub_ps(iy3,jy0);
871 dz30 = _mm256_sub_ps(iz3,jz0);
873 /* Calculate squared distance and things based on it */
874 rsq10 = gmx_mm256_calc_rsq_ps(dx10,dy10,dz10);
875 rsq20 = gmx_mm256_calc_rsq_ps(dx20,dy20,dz20);
876 rsq30 = gmx_mm256_calc_rsq_ps(dx30,dy30,dz30);
878 rinv10 = gmx_mm256_invsqrt_ps(rsq10);
879 rinv20 = gmx_mm256_invsqrt_ps(rsq20);
880 rinv30 = gmx_mm256_invsqrt_ps(rsq30);
882 rinvsq10 = _mm256_mul_ps(rinv10,rinv10);
883 rinvsq20 = _mm256_mul_ps(rinv20,rinv20);
884 rinvsq30 = _mm256_mul_ps(rinv30,rinv30);
886 /* Load parameters for j particles */
887 jq0 = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
888 charge+jnrC+0,charge+jnrD+0,
889 charge+jnrE+0,charge+jnrF+0,
890 charge+jnrG+0,charge+jnrH+0);
892 fjx0 = _mm256_setzero_ps();
893 fjy0 = _mm256_setzero_ps();
894 fjz0 = _mm256_setzero_ps();
896 /**************************
897 * CALCULATE INTERACTIONS *
898 **************************/
900 /* Compute parameters for interactions between i and j atoms */
901 qq10 = _mm256_mul_ps(iq1,jq0);
903 /* REACTION-FIELD ELECTROSTATICS */
904 felec = _mm256_mul_ps(qq10,_mm256_sub_ps(_mm256_mul_ps(rinv10,rinvsq10),krf2));
908 fscal = _mm256_andnot_ps(dummy_mask,fscal);
910 /* Calculate temporary vectorial force */
911 tx = _mm256_mul_ps(fscal,dx10);
912 ty = _mm256_mul_ps(fscal,dy10);
913 tz = _mm256_mul_ps(fscal,dz10);
915 /* Update vectorial force */
916 fix1 = _mm256_add_ps(fix1,tx);
917 fiy1 = _mm256_add_ps(fiy1,ty);
918 fiz1 = _mm256_add_ps(fiz1,tz);
920 fjx0 = _mm256_add_ps(fjx0,tx);
921 fjy0 = _mm256_add_ps(fjy0,ty);
922 fjz0 = _mm256_add_ps(fjz0,tz);
924 /**************************
925 * CALCULATE INTERACTIONS *
926 **************************/
928 /* Compute parameters for interactions between i and j atoms */
929 qq20 = _mm256_mul_ps(iq2,jq0);
931 /* REACTION-FIELD ELECTROSTATICS */
932 felec = _mm256_mul_ps(qq20,_mm256_sub_ps(_mm256_mul_ps(rinv20,rinvsq20),krf2));
936 fscal = _mm256_andnot_ps(dummy_mask,fscal);
938 /* Calculate temporary vectorial force */
939 tx = _mm256_mul_ps(fscal,dx20);
940 ty = _mm256_mul_ps(fscal,dy20);
941 tz = _mm256_mul_ps(fscal,dz20);
943 /* Update vectorial force */
944 fix2 = _mm256_add_ps(fix2,tx);
945 fiy2 = _mm256_add_ps(fiy2,ty);
946 fiz2 = _mm256_add_ps(fiz2,tz);
948 fjx0 = _mm256_add_ps(fjx0,tx);
949 fjy0 = _mm256_add_ps(fjy0,ty);
950 fjz0 = _mm256_add_ps(fjz0,tz);
952 /**************************
953 * CALCULATE INTERACTIONS *
954 **************************/
956 /* Compute parameters for interactions between i and j atoms */
957 qq30 = _mm256_mul_ps(iq3,jq0);
959 /* REACTION-FIELD ELECTROSTATICS */
960 felec = _mm256_mul_ps(qq30,_mm256_sub_ps(_mm256_mul_ps(rinv30,rinvsq30),krf2));
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 84 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*84);