2 * This file is part of the GROMACS molecular simulation package.
4 * Copyright (c) 2012,2013,2014, by the GROMACS development team, led by
5 * Mark Abraham, David van der Spoel, Berk Hess, and Erik Lindahl,
6 * and including many others, as listed in the AUTHORS file in the
7 * top-level source directory and at http://www.gromacs.org.
9 * GROMACS is free software; you can redistribute it and/or
10 * modify it under the terms of the GNU Lesser General Public License
11 * as published by the Free Software Foundation; either version 2.1
12 * of the License, or (at your option) any later version.
14 * GROMACS is distributed in the hope that it will be useful,
15 * but WITHOUT ANY WARRANTY; without even the implied warranty of
16 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
17 * Lesser General Public License for more details.
19 * You should have received a copy of the GNU Lesser General Public
20 * License along with GROMACS; if not, see
21 * http://www.gnu.org/licenses, or write to the Free Software Foundation,
22 * Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA.
24 * If you want to redistribute modifications to GROMACS, please
25 * consider that scientific software is very special. Version
26 * control is crucial - bugs must be traceable. We will be happy to
27 * consider code for inclusion in the official distribution, but
28 * derived work must not be called official GROMACS. Details are found
29 * in the README & COPYING files - if they are missing, get the
30 * official version at http://www.gromacs.org.
32 * To help us fund GROMACS development, we humbly ask that you cite
33 * the research papers on the package. Check out http://www.gromacs.org.
36 * Note: this file was generated by the GROMACS avx_256_single kernel generator.
44 #include "../nb_kernel.h"
45 #include "gromacs/legacyheaders/types/simple.h"
46 #include "gromacs/math/vec.h"
47 #include "gromacs/legacyheaders/nrnb.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_ElecEwSh_VdwLJEwSh_GeomP1P1_VF_avx_256_single
54 * Electrostatics interaction: Ewald
55 * VdW interaction: LJEwald
56 * Geometry: Particle-Particle
57 * Calculate force/pot: PotentialAndForce
60 nb_kernel_ElecEwSh_VdwLJEwSh_GeomP1P1_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 * vdwioffsetptr0;
89 real * vdwgridioffsetptr0;
90 __m256 ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
91 int vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D,vdwjidx0E,vdwjidx0F,vdwjidx0G,vdwjidx0H;
92 __m256 jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
93 __m256 dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
94 __m256 velec,felec,velecsum,facel,crf,krf,krf2;
97 __m256 rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
100 __m256 one_sixth = _mm256_set1_ps(1.0/6.0);
101 __m256 one_twelfth = _mm256_set1_ps(1.0/12.0);
104 __m256 ewclj,ewclj2,ewclj6,ewcljrsq,poly,exponent,f6A,f6B,sh_lj_ewald;
105 __m256 one_half = _mm256_set1_ps(0.5);
106 __m256 minus_one = _mm256_set1_ps(-1.0);
108 __m128i ewitab_lo,ewitab_hi;
109 __m256 ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
110 __m256 beta,beta2,beta3,zeta2,pmecorrF,pmecorrV,rinv3;
112 __m256 dummy_mask,cutoff_mask;
113 __m256 signbit = _mm256_castsi256_ps( _mm256_set1_epi32(0x80000000) );
114 __m256 one = _mm256_set1_ps(1.0);
115 __m256 two = _mm256_set1_ps(2.0);
121 jindex = nlist->jindex;
123 shiftidx = nlist->shift;
125 shiftvec = fr->shift_vec[0];
126 fshift = fr->fshift[0];
127 facel = _mm256_set1_ps(fr->epsfac);
128 charge = mdatoms->chargeA;
129 nvdwtype = fr->ntype;
131 vdwtype = mdatoms->typeA;
132 vdwgridparam = fr->ljpme_c6grid;
133 sh_lj_ewald = _mm256_set1_ps(fr->ic->sh_lj_ewald);
134 ewclj = _mm256_set1_ps(fr->ewaldcoeff_lj);
135 ewclj2 = _mm256_mul_ps(minus_one,_mm256_mul_ps(ewclj,ewclj));
137 sh_ewald = _mm256_set1_ps(fr->ic->sh_ewald);
138 beta = _mm256_set1_ps(fr->ic->ewaldcoeff_q);
139 beta2 = _mm256_mul_ps(beta,beta);
140 beta3 = _mm256_mul_ps(beta,beta2);
142 ewtab = fr->ic->tabq_coul_FDV0;
143 ewtabscale = _mm256_set1_ps(fr->ic->tabq_scale);
144 ewtabhalfspace = _mm256_set1_ps(0.5/fr->ic->tabq_scale);
146 /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
147 rcutoff_scalar = fr->rcoulomb;
148 rcutoff = _mm256_set1_ps(rcutoff_scalar);
149 rcutoff2 = _mm256_mul_ps(rcutoff,rcutoff);
151 sh_vdw_invrcut6 = _mm256_set1_ps(fr->ic->sh_invrc6);
152 rvdw = _mm256_set1_ps(fr->rvdw);
154 /* Avoid stupid compiler warnings */
155 jnrA = jnrB = jnrC = jnrD = jnrE = jnrF = jnrG = jnrH = 0;
168 for(iidx=0;iidx<4*DIM;iidx++)
173 /* Start outer loop over neighborlists */
174 for(iidx=0; iidx<nri; iidx++)
176 /* Load shift vector for this list */
177 i_shift_offset = DIM*shiftidx[iidx];
179 /* Load limits for loop over neighbors */
180 j_index_start = jindex[iidx];
181 j_index_end = jindex[iidx+1];
183 /* Get outer coordinate index */
185 i_coord_offset = DIM*inr;
187 /* Load i particle coords and add shift vector */
188 gmx_mm256_load_shift_and_1rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
190 fix0 = _mm256_setzero_ps();
191 fiy0 = _mm256_setzero_ps();
192 fiz0 = _mm256_setzero_ps();
194 /* Load parameters for i particles */
195 iq0 = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+0]));
196 vdwioffsetptr0 = vdwparam+2*nvdwtype*vdwtype[inr+0];
197 vdwgridioffsetptr0 = vdwgridparam+2*nvdwtype*vdwtype[inr+0];
199 /* Reset potential sums */
200 velecsum = _mm256_setzero_ps();
201 vvdwsum = _mm256_setzero_ps();
203 /* Start inner kernel loop */
204 for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+7]>=0; jidx+=8)
207 /* Get j neighbor index, and coordinate index */
216 j_coord_offsetA = DIM*jnrA;
217 j_coord_offsetB = DIM*jnrB;
218 j_coord_offsetC = DIM*jnrC;
219 j_coord_offsetD = DIM*jnrD;
220 j_coord_offsetE = DIM*jnrE;
221 j_coord_offsetF = DIM*jnrF;
222 j_coord_offsetG = DIM*jnrG;
223 j_coord_offsetH = DIM*jnrH;
225 /* load j atom coordinates */
226 gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
227 x+j_coord_offsetC,x+j_coord_offsetD,
228 x+j_coord_offsetE,x+j_coord_offsetF,
229 x+j_coord_offsetG,x+j_coord_offsetH,
232 /* Calculate displacement vector */
233 dx00 = _mm256_sub_ps(ix0,jx0);
234 dy00 = _mm256_sub_ps(iy0,jy0);
235 dz00 = _mm256_sub_ps(iz0,jz0);
237 /* Calculate squared distance and things based on it */
238 rsq00 = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
240 rinv00 = gmx_mm256_invsqrt_ps(rsq00);
242 rinvsq00 = _mm256_mul_ps(rinv00,rinv00);
244 /* Load parameters for j particles */
245 jq0 = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
246 charge+jnrC+0,charge+jnrD+0,
247 charge+jnrE+0,charge+jnrF+0,
248 charge+jnrG+0,charge+jnrH+0);
249 vdwjidx0A = 2*vdwtype[jnrA+0];
250 vdwjidx0B = 2*vdwtype[jnrB+0];
251 vdwjidx0C = 2*vdwtype[jnrC+0];
252 vdwjidx0D = 2*vdwtype[jnrD+0];
253 vdwjidx0E = 2*vdwtype[jnrE+0];
254 vdwjidx0F = 2*vdwtype[jnrF+0];
255 vdwjidx0G = 2*vdwtype[jnrG+0];
256 vdwjidx0H = 2*vdwtype[jnrH+0];
258 /**************************
259 * CALCULATE INTERACTIONS *
260 **************************/
262 if (gmx_mm256_any_lt(rsq00,rcutoff2))
265 r00 = _mm256_mul_ps(rsq00,rinv00);
267 /* Compute parameters for interactions between i and j atoms */
268 qq00 = _mm256_mul_ps(iq0,jq0);
269 gmx_mm256_load_8pair_swizzle_ps(vdwioffsetptr0+vdwjidx0A,
270 vdwioffsetptr0+vdwjidx0B,
271 vdwioffsetptr0+vdwjidx0C,
272 vdwioffsetptr0+vdwjidx0D,
273 vdwioffsetptr0+vdwjidx0E,
274 vdwioffsetptr0+vdwjidx0F,
275 vdwioffsetptr0+vdwjidx0G,
276 vdwioffsetptr0+vdwjidx0H,
279 c6grid_00 = gmx_mm256_load_8real_swizzle_ps(vdwgridioffsetptr0+vdwjidx0A,
280 vdwgridioffsetptr0+vdwjidx0B,
281 vdwgridioffsetptr0+vdwjidx0C,
282 vdwgridioffsetptr0+vdwjidx0D,
283 vdwgridioffsetptr0+vdwjidx0E,
284 vdwgridioffsetptr0+vdwjidx0F,
285 vdwgridioffsetptr0+vdwjidx0G,
286 vdwgridioffsetptr0+vdwjidx0H);
288 /* EWALD ELECTROSTATICS */
290 /* Analytical PME correction */
291 zeta2 = _mm256_mul_ps(beta2,rsq00);
292 rinv3 = _mm256_mul_ps(rinvsq00,rinv00);
293 pmecorrF = gmx_mm256_pmecorrF_ps(zeta2);
294 felec = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
295 felec = _mm256_mul_ps(qq00,felec);
296 pmecorrV = gmx_mm256_pmecorrV_ps(zeta2);
297 pmecorrV = _mm256_mul_ps(pmecorrV,beta);
298 velec = _mm256_sub_ps(_mm256_sub_ps(rinv00,sh_ewald),pmecorrV);
299 velec = _mm256_mul_ps(qq00,velec);
301 /* Analytical LJ-PME */
302 rinvsix = _mm256_mul_ps(_mm256_mul_ps(rinvsq00,rinvsq00),rinvsq00);
303 ewcljrsq = _mm256_mul_ps(ewclj2,rsq00);
304 ewclj6 = _mm256_mul_ps(ewclj2,_mm256_mul_ps(ewclj2,ewclj2));
305 exponent = gmx_simd_exp_r(ewcljrsq);
306 /* poly = exp(-(beta*r)^2) * (1 + (beta*r)^2 + (beta*r)^4 /2) */
307 poly = _mm256_mul_ps(exponent,_mm256_add_ps(_mm256_sub_ps(one,ewcljrsq),_mm256_mul_ps(_mm256_mul_ps(ewcljrsq,ewcljrsq),one_half)));
308 /* vvdw6 = [C6 - C6grid * (1-poly)]/r6 */
309 vvdw6 = _mm256_mul_ps(_mm256_sub_ps(c6_00,_mm256_mul_ps(c6grid_00,_mm256_sub_ps(one,poly))),rinvsix);
310 vvdw12 = _mm256_mul_ps(c12_00,_mm256_mul_ps(rinvsix,rinvsix));
311 vvdw = _mm256_sub_ps(_mm256_mul_ps( _mm256_sub_ps(vvdw12 , _mm256_mul_ps(c12_00,_mm256_mul_ps(sh_vdw_invrcut6,sh_vdw_invrcut6))), one_twelfth) ,
312 _mm256_mul_ps( _mm256_sub_ps(vvdw6,_mm256_add_ps(_mm256_mul_ps(c6_00,sh_vdw_invrcut6),_mm256_mul_ps(c6grid_00,sh_lj_ewald))),one_sixth));
313 /* fvdw = vvdw12/r - (vvdw6/r + (C6grid * exponent * beta^6)/r) */
314 fvdw = _mm256_mul_ps(_mm256_sub_ps(vvdw12,_mm256_sub_ps(vvdw6,_mm256_mul_ps(_mm256_mul_ps(c6grid_00,one_sixth),_mm256_mul_ps(exponent,ewclj6)))),rinvsq00);
316 cutoff_mask = _mm256_cmp_ps(rsq00,rcutoff2,_CMP_LT_OQ);
318 /* Update potential sum for this i atom from the interaction with this j atom. */
319 velec = _mm256_and_ps(velec,cutoff_mask);
320 velecsum = _mm256_add_ps(velecsum,velec);
321 vvdw = _mm256_and_ps(vvdw,cutoff_mask);
322 vvdwsum = _mm256_add_ps(vvdwsum,vvdw);
324 fscal = _mm256_add_ps(felec,fvdw);
326 fscal = _mm256_and_ps(fscal,cutoff_mask);
328 /* Calculate temporary vectorial force */
329 tx = _mm256_mul_ps(fscal,dx00);
330 ty = _mm256_mul_ps(fscal,dy00);
331 tz = _mm256_mul_ps(fscal,dz00);
333 /* Update vectorial force */
334 fix0 = _mm256_add_ps(fix0,tx);
335 fiy0 = _mm256_add_ps(fiy0,ty);
336 fiz0 = _mm256_add_ps(fiz0,tz);
338 fjptrA = f+j_coord_offsetA;
339 fjptrB = f+j_coord_offsetB;
340 fjptrC = f+j_coord_offsetC;
341 fjptrD = f+j_coord_offsetD;
342 fjptrE = f+j_coord_offsetE;
343 fjptrF = f+j_coord_offsetF;
344 fjptrG = f+j_coord_offsetG;
345 fjptrH = f+j_coord_offsetH;
346 gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,tx,ty,tz);
350 /* Inner loop uses 145 flops */
356 /* Get j neighbor index, and coordinate index */
357 jnrlistA = jjnr[jidx];
358 jnrlistB = jjnr[jidx+1];
359 jnrlistC = jjnr[jidx+2];
360 jnrlistD = jjnr[jidx+3];
361 jnrlistE = jjnr[jidx+4];
362 jnrlistF = jjnr[jidx+5];
363 jnrlistG = jjnr[jidx+6];
364 jnrlistH = jjnr[jidx+7];
365 /* Sign of each element will be negative for non-real atoms.
366 * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
367 * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
369 dummy_mask = gmx_mm256_set_m128(gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx+4)),_mm_setzero_si128())),
370 gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128())));
372 jnrA = (jnrlistA>=0) ? jnrlistA : 0;
373 jnrB = (jnrlistB>=0) ? jnrlistB : 0;
374 jnrC = (jnrlistC>=0) ? jnrlistC : 0;
375 jnrD = (jnrlistD>=0) ? jnrlistD : 0;
376 jnrE = (jnrlistE>=0) ? jnrlistE : 0;
377 jnrF = (jnrlistF>=0) ? jnrlistF : 0;
378 jnrG = (jnrlistG>=0) ? jnrlistG : 0;
379 jnrH = (jnrlistH>=0) ? jnrlistH : 0;
380 j_coord_offsetA = DIM*jnrA;
381 j_coord_offsetB = DIM*jnrB;
382 j_coord_offsetC = DIM*jnrC;
383 j_coord_offsetD = DIM*jnrD;
384 j_coord_offsetE = DIM*jnrE;
385 j_coord_offsetF = DIM*jnrF;
386 j_coord_offsetG = DIM*jnrG;
387 j_coord_offsetH = DIM*jnrH;
389 /* load j atom coordinates */
390 gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
391 x+j_coord_offsetC,x+j_coord_offsetD,
392 x+j_coord_offsetE,x+j_coord_offsetF,
393 x+j_coord_offsetG,x+j_coord_offsetH,
396 /* Calculate displacement vector */
397 dx00 = _mm256_sub_ps(ix0,jx0);
398 dy00 = _mm256_sub_ps(iy0,jy0);
399 dz00 = _mm256_sub_ps(iz0,jz0);
401 /* Calculate squared distance and things based on it */
402 rsq00 = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
404 rinv00 = gmx_mm256_invsqrt_ps(rsq00);
406 rinvsq00 = _mm256_mul_ps(rinv00,rinv00);
408 /* Load parameters for j particles */
409 jq0 = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
410 charge+jnrC+0,charge+jnrD+0,
411 charge+jnrE+0,charge+jnrF+0,
412 charge+jnrG+0,charge+jnrH+0);
413 vdwjidx0A = 2*vdwtype[jnrA+0];
414 vdwjidx0B = 2*vdwtype[jnrB+0];
415 vdwjidx0C = 2*vdwtype[jnrC+0];
416 vdwjidx0D = 2*vdwtype[jnrD+0];
417 vdwjidx0E = 2*vdwtype[jnrE+0];
418 vdwjidx0F = 2*vdwtype[jnrF+0];
419 vdwjidx0G = 2*vdwtype[jnrG+0];
420 vdwjidx0H = 2*vdwtype[jnrH+0];
422 /**************************
423 * CALCULATE INTERACTIONS *
424 **************************/
426 if (gmx_mm256_any_lt(rsq00,rcutoff2))
429 r00 = _mm256_mul_ps(rsq00,rinv00);
430 r00 = _mm256_andnot_ps(dummy_mask,r00);
432 /* Compute parameters for interactions between i and j atoms */
433 qq00 = _mm256_mul_ps(iq0,jq0);
434 gmx_mm256_load_8pair_swizzle_ps(vdwioffsetptr0+vdwjidx0A,
435 vdwioffsetptr0+vdwjidx0B,
436 vdwioffsetptr0+vdwjidx0C,
437 vdwioffsetptr0+vdwjidx0D,
438 vdwioffsetptr0+vdwjidx0E,
439 vdwioffsetptr0+vdwjidx0F,
440 vdwioffsetptr0+vdwjidx0G,
441 vdwioffsetptr0+vdwjidx0H,
444 c6grid_00 = gmx_mm256_load_8real_swizzle_ps(vdwgridioffsetptr0+vdwjidx0A,
445 vdwgridioffsetptr0+vdwjidx0B,
446 vdwgridioffsetptr0+vdwjidx0C,
447 vdwgridioffsetptr0+vdwjidx0D,
448 vdwgridioffsetptr0+vdwjidx0E,
449 vdwgridioffsetptr0+vdwjidx0F,
450 vdwgridioffsetptr0+vdwjidx0G,
451 vdwgridioffsetptr0+vdwjidx0H);
453 /* EWALD ELECTROSTATICS */
455 /* Analytical PME correction */
456 zeta2 = _mm256_mul_ps(beta2,rsq00);
457 rinv3 = _mm256_mul_ps(rinvsq00,rinv00);
458 pmecorrF = gmx_mm256_pmecorrF_ps(zeta2);
459 felec = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
460 felec = _mm256_mul_ps(qq00,felec);
461 pmecorrV = gmx_mm256_pmecorrV_ps(zeta2);
462 pmecorrV = _mm256_mul_ps(pmecorrV,beta);
463 velec = _mm256_sub_ps(_mm256_sub_ps(rinv00,sh_ewald),pmecorrV);
464 velec = _mm256_mul_ps(qq00,velec);
466 /* Analytical LJ-PME */
467 rinvsix = _mm256_mul_ps(_mm256_mul_ps(rinvsq00,rinvsq00),rinvsq00);
468 ewcljrsq = _mm256_mul_ps(ewclj2,rsq00);
469 ewclj6 = _mm256_mul_ps(ewclj2,_mm256_mul_ps(ewclj2,ewclj2));
470 exponent = gmx_simd_exp_r(ewcljrsq);
471 /* poly = exp(-(beta*r)^2) * (1 + (beta*r)^2 + (beta*r)^4 /2) */
472 poly = _mm256_mul_ps(exponent,_mm256_add_ps(_mm256_sub_ps(one,ewcljrsq),_mm256_mul_ps(_mm256_mul_ps(ewcljrsq,ewcljrsq),one_half)));
473 /* vvdw6 = [C6 - C6grid * (1-poly)]/r6 */
474 vvdw6 = _mm256_mul_ps(_mm256_sub_ps(c6_00,_mm256_mul_ps(c6grid_00,_mm256_sub_ps(one,poly))),rinvsix);
475 vvdw12 = _mm256_mul_ps(c12_00,_mm256_mul_ps(rinvsix,rinvsix));
476 vvdw = _mm256_sub_ps(_mm256_mul_ps( _mm256_sub_ps(vvdw12 , _mm256_mul_ps(c12_00,_mm256_mul_ps(sh_vdw_invrcut6,sh_vdw_invrcut6))), one_twelfth) ,
477 _mm256_mul_ps( _mm256_sub_ps(vvdw6,_mm256_add_ps(_mm256_mul_ps(c6_00,sh_vdw_invrcut6),_mm256_mul_ps(c6grid_00,sh_lj_ewald))),one_sixth));
478 /* fvdw = vvdw12/r - (vvdw6/r + (C6grid * exponent * beta^6)/r) */
479 fvdw = _mm256_mul_ps(_mm256_sub_ps(vvdw12,_mm256_sub_ps(vvdw6,_mm256_mul_ps(_mm256_mul_ps(c6grid_00,one_sixth),_mm256_mul_ps(exponent,ewclj6)))),rinvsq00);
481 cutoff_mask = _mm256_cmp_ps(rsq00,rcutoff2,_CMP_LT_OQ);
483 /* Update potential sum for this i atom from the interaction with this j atom. */
484 velec = _mm256_and_ps(velec,cutoff_mask);
485 velec = _mm256_andnot_ps(dummy_mask,velec);
486 velecsum = _mm256_add_ps(velecsum,velec);
487 vvdw = _mm256_and_ps(vvdw,cutoff_mask);
488 vvdw = _mm256_andnot_ps(dummy_mask,vvdw);
489 vvdwsum = _mm256_add_ps(vvdwsum,vvdw);
491 fscal = _mm256_add_ps(felec,fvdw);
493 fscal = _mm256_and_ps(fscal,cutoff_mask);
495 fscal = _mm256_andnot_ps(dummy_mask,fscal);
497 /* Calculate temporary vectorial force */
498 tx = _mm256_mul_ps(fscal,dx00);
499 ty = _mm256_mul_ps(fscal,dy00);
500 tz = _mm256_mul_ps(fscal,dz00);
502 /* Update vectorial force */
503 fix0 = _mm256_add_ps(fix0,tx);
504 fiy0 = _mm256_add_ps(fiy0,ty);
505 fiz0 = _mm256_add_ps(fiz0,tz);
507 fjptrA = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
508 fjptrB = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
509 fjptrC = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
510 fjptrD = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
511 fjptrE = (jnrlistE>=0) ? f+j_coord_offsetE : scratch;
512 fjptrF = (jnrlistF>=0) ? f+j_coord_offsetF : scratch;
513 fjptrG = (jnrlistG>=0) ? f+j_coord_offsetG : scratch;
514 fjptrH = (jnrlistH>=0) ? f+j_coord_offsetH : scratch;
515 gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,tx,ty,tz);
519 /* Inner loop uses 146 flops */
522 /* End of innermost loop */
524 gmx_mm256_update_iforce_1atom_swizzle_ps(fix0,fiy0,fiz0,
525 f+i_coord_offset,fshift+i_shift_offset);
528 /* Update potential energies */
529 gmx_mm256_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid);
530 gmx_mm256_update_1pot_ps(vvdwsum,kernel_data->energygrp_vdw+ggid);
532 /* Increment number of inner iterations */
533 inneriter += j_index_end - j_index_start;
535 /* Outer loop uses 9 flops */
538 /* Increment number of outer iterations */
541 /* Update outer/inner flops */
543 inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_VF,outeriter*9 + inneriter*146);
546 * Gromacs nonbonded kernel: nb_kernel_ElecEwSh_VdwLJEwSh_GeomP1P1_F_avx_256_single
547 * Electrostatics interaction: Ewald
548 * VdW interaction: LJEwald
549 * Geometry: Particle-Particle
550 * Calculate force/pot: Force
553 nb_kernel_ElecEwSh_VdwLJEwSh_GeomP1P1_F_avx_256_single
554 (t_nblist * gmx_restrict nlist,
555 rvec * gmx_restrict xx,
556 rvec * gmx_restrict ff,
557 t_forcerec * gmx_restrict fr,
558 t_mdatoms * gmx_restrict mdatoms,
559 nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
560 t_nrnb * gmx_restrict nrnb)
562 /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
563 * just 0 for non-waters.
564 * Suffixes A,B,C,D,E,F,G,H refer to j loop unrolling done with AVX, e.g. for the eight different
565 * jnr indices corresponding to data put in the four positions in the SIMD register.
567 int i_shift_offset,i_coord_offset,outeriter,inneriter;
568 int j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
569 int jnrA,jnrB,jnrC,jnrD;
570 int jnrE,jnrF,jnrG,jnrH;
571 int jnrlistA,jnrlistB,jnrlistC,jnrlistD;
572 int jnrlistE,jnrlistF,jnrlistG,jnrlistH;
573 int j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
574 int j_coord_offsetE,j_coord_offsetF,j_coord_offsetG,j_coord_offsetH;
575 int *iinr,*jindex,*jjnr,*shiftidx,*gid;
577 real *shiftvec,*fshift,*x,*f;
578 real *fjptrA,*fjptrB,*fjptrC,*fjptrD,*fjptrE,*fjptrF,*fjptrG,*fjptrH;
580 __m256 tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
581 real * vdwioffsetptr0;
582 real * vdwgridioffsetptr0;
583 __m256 ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
584 int vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D,vdwjidx0E,vdwjidx0F,vdwjidx0G,vdwjidx0H;
585 __m256 jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
586 __m256 dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
587 __m256 velec,felec,velecsum,facel,crf,krf,krf2;
590 __m256 rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
593 __m256 one_sixth = _mm256_set1_ps(1.0/6.0);
594 __m256 one_twelfth = _mm256_set1_ps(1.0/12.0);
597 __m256 ewclj,ewclj2,ewclj6,ewcljrsq,poly,exponent,f6A,f6B,sh_lj_ewald;
598 __m256 one_half = _mm256_set1_ps(0.5);
599 __m256 minus_one = _mm256_set1_ps(-1.0);
601 __m128i ewitab_lo,ewitab_hi;
602 __m256 ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
603 __m256 beta,beta2,beta3,zeta2,pmecorrF,pmecorrV,rinv3;
605 __m256 dummy_mask,cutoff_mask;
606 __m256 signbit = _mm256_castsi256_ps( _mm256_set1_epi32(0x80000000) );
607 __m256 one = _mm256_set1_ps(1.0);
608 __m256 two = _mm256_set1_ps(2.0);
614 jindex = nlist->jindex;
616 shiftidx = nlist->shift;
618 shiftvec = fr->shift_vec[0];
619 fshift = fr->fshift[0];
620 facel = _mm256_set1_ps(fr->epsfac);
621 charge = mdatoms->chargeA;
622 nvdwtype = fr->ntype;
624 vdwtype = mdatoms->typeA;
625 vdwgridparam = fr->ljpme_c6grid;
626 sh_lj_ewald = _mm256_set1_ps(fr->ic->sh_lj_ewald);
627 ewclj = _mm256_set1_ps(fr->ewaldcoeff_lj);
628 ewclj2 = _mm256_mul_ps(minus_one,_mm256_mul_ps(ewclj,ewclj));
630 sh_ewald = _mm256_set1_ps(fr->ic->sh_ewald);
631 beta = _mm256_set1_ps(fr->ic->ewaldcoeff_q);
632 beta2 = _mm256_mul_ps(beta,beta);
633 beta3 = _mm256_mul_ps(beta,beta2);
635 ewtab = fr->ic->tabq_coul_F;
636 ewtabscale = _mm256_set1_ps(fr->ic->tabq_scale);
637 ewtabhalfspace = _mm256_set1_ps(0.5/fr->ic->tabq_scale);
639 /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
640 rcutoff_scalar = fr->rcoulomb;
641 rcutoff = _mm256_set1_ps(rcutoff_scalar);
642 rcutoff2 = _mm256_mul_ps(rcutoff,rcutoff);
644 sh_vdw_invrcut6 = _mm256_set1_ps(fr->ic->sh_invrc6);
645 rvdw = _mm256_set1_ps(fr->rvdw);
647 /* Avoid stupid compiler warnings */
648 jnrA = jnrB = jnrC = jnrD = jnrE = jnrF = jnrG = jnrH = 0;
661 for(iidx=0;iidx<4*DIM;iidx++)
666 /* Start outer loop over neighborlists */
667 for(iidx=0; iidx<nri; iidx++)
669 /* Load shift vector for this list */
670 i_shift_offset = DIM*shiftidx[iidx];
672 /* Load limits for loop over neighbors */
673 j_index_start = jindex[iidx];
674 j_index_end = jindex[iidx+1];
676 /* Get outer coordinate index */
678 i_coord_offset = DIM*inr;
680 /* Load i particle coords and add shift vector */
681 gmx_mm256_load_shift_and_1rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
683 fix0 = _mm256_setzero_ps();
684 fiy0 = _mm256_setzero_ps();
685 fiz0 = _mm256_setzero_ps();
687 /* Load parameters for i particles */
688 iq0 = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+0]));
689 vdwioffsetptr0 = vdwparam+2*nvdwtype*vdwtype[inr+0];
690 vdwgridioffsetptr0 = vdwgridparam+2*nvdwtype*vdwtype[inr+0];
692 /* Start inner kernel loop */
693 for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+7]>=0; jidx+=8)
696 /* Get j neighbor index, and coordinate index */
705 j_coord_offsetA = DIM*jnrA;
706 j_coord_offsetB = DIM*jnrB;
707 j_coord_offsetC = DIM*jnrC;
708 j_coord_offsetD = DIM*jnrD;
709 j_coord_offsetE = DIM*jnrE;
710 j_coord_offsetF = DIM*jnrF;
711 j_coord_offsetG = DIM*jnrG;
712 j_coord_offsetH = DIM*jnrH;
714 /* load j atom coordinates */
715 gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
716 x+j_coord_offsetC,x+j_coord_offsetD,
717 x+j_coord_offsetE,x+j_coord_offsetF,
718 x+j_coord_offsetG,x+j_coord_offsetH,
721 /* Calculate displacement vector */
722 dx00 = _mm256_sub_ps(ix0,jx0);
723 dy00 = _mm256_sub_ps(iy0,jy0);
724 dz00 = _mm256_sub_ps(iz0,jz0);
726 /* Calculate squared distance and things based on it */
727 rsq00 = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
729 rinv00 = gmx_mm256_invsqrt_ps(rsq00);
731 rinvsq00 = _mm256_mul_ps(rinv00,rinv00);
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);
738 vdwjidx0A = 2*vdwtype[jnrA+0];
739 vdwjidx0B = 2*vdwtype[jnrB+0];
740 vdwjidx0C = 2*vdwtype[jnrC+0];
741 vdwjidx0D = 2*vdwtype[jnrD+0];
742 vdwjidx0E = 2*vdwtype[jnrE+0];
743 vdwjidx0F = 2*vdwtype[jnrF+0];
744 vdwjidx0G = 2*vdwtype[jnrG+0];
745 vdwjidx0H = 2*vdwtype[jnrH+0];
747 /**************************
748 * CALCULATE INTERACTIONS *
749 **************************/
751 if (gmx_mm256_any_lt(rsq00,rcutoff2))
754 r00 = _mm256_mul_ps(rsq00,rinv00);
756 /* Compute parameters for interactions between i and j atoms */
757 qq00 = _mm256_mul_ps(iq0,jq0);
758 gmx_mm256_load_8pair_swizzle_ps(vdwioffsetptr0+vdwjidx0A,
759 vdwioffsetptr0+vdwjidx0B,
760 vdwioffsetptr0+vdwjidx0C,
761 vdwioffsetptr0+vdwjidx0D,
762 vdwioffsetptr0+vdwjidx0E,
763 vdwioffsetptr0+vdwjidx0F,
764 vdwioffsetptr0+vdwjidx0G,
765 vdwioffsetptr0+vdwjidx0H,
768 c6grid_00 = gmx_mm256_load_8real_swizzle_ps(vdwgridioffsetptr0+vdwjidx0A,
769 vdwgridioffsetptr0+vdwjidx0B,
770 vdwgridioffsetptr0+vdwjidx0C,
771 vdwgridioffsetptr0+vdwjidx0D,
772 vdwgridioffsetptr0+vdwjidx0E,
773 vdwgridioffsetptr0+vdwjidx0F,
774 vdwgridioffsetptr0+vdwjidx0G,
775 vdwgridioffsetptr0+vdwjidx0H);
777 /* EWALD ELECTROSTATICS */
779 /* Analytical PME correction */
780 zeta2 = _mm256_mul_ps(beta2,rsq00);
781 rinv3 = _mm256_mul_ps(rinvsq00,rinv00);
782 pmecorrF = gmx_mm256_pmecorrF_ps(zeta2);
783 felec = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
784 felec = _mm256_mul_ps(qq00,felec);
786 /* Analytical LJ-PME */
787 rinvsix = _mm256_mul_ps(_mm256_mul_ps(rinvsq00,rinvsq00),rinvsq00);
788 ewcljrsq = _mm256_mul_ps(ewclj2,rsq00);
789 ewclj6 = _mm256_mul_ps(ewclj2,_mm256_mul_ps(ewclj2,ewclj2));
790 exponent = gmx_simd_exp_r(ewcljrsq);
791 /* poly = exp(-(beta*r)^2) * (1 + (beta*r)^2 + (beta*r)^4 /2) */
792 poly = _mm256_mul_ps(exponent,_mm256_add_ps(_mm256_sub_ps(one,ewcljrsq),_mm256_mul_ps(_mm256_mul_ps(ewcljrsq,ewcljrsq),one_half)));
793 /* f6A = 6 * C6grid * (1 - poly) */
794 f6A = _mm256_mul_ps(c6grid_00,_mm256_sub_ps(one,poly));
795 /* f6B = C6grid * exponent * beta^6 */
796 f6B = _mm256_mul_ps(_mm256_mul_ps(c6grid_00,one_sixth),_mm256_mul_ps(exponent,ewclj6));
797 /* fvdw = 12*C12/r13 - ((6*C6 - f6A)/r6 + f6B)/r */
798 fvdw = _mm256_mul_ps(_mm256_add_ps(_mm256_mul_ps(_mm256_sub_ps(_mm256_mul_ps(c12_00,rinvsix),_mm256_sub_ps(c6_00,f6A)),rinvsix),f6B),rinvsq00);
800 cutoff_mask = _mm256_cmp_ps(rsq00,rcutoff2,_CMP_LT_OQ);
802 fscal = _mm256_add_ps(felec,fvdw);
804 fscal = _mm256_and_ps(fscal,cutoff_mask);
806 /* Calculate temporary vectorial force */
807 tx = _mm256_mul_ps(fscal,dx00);
808 ty = _mm256_mul_ps(fscal,dy00);
809 tz = _mm256_mul_ps(fscal,dz00);
811 /* Update vectorial force */
812 fix0 = _mm256_add_ps(fix0,tx);
813 fiy0 = _mm256_add_ps(fiy0,ty);
814 fiz0 = _mm256_add_ps(fiz0,tz);
816 fjptrA = f+j_coord_offsetA;
817 fjptrB = f+j_coord_offsetB;
818 fjptrC = f+j_coord_offsetC;
819 fjptrD = f+j_coord_offsetD;
820 fjptrE = f+j_coord_offsetE;
821 fjptrF = f+j_coord_offsetF;
822 fjptrG = f+j_coord_offsetG;
823 fjptrH = f+j_coord_offsetH;
824 gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,tx,ty,tz);
828 /* Inner loop uses 82 flops */
834 /* Get j neighbor index, and coordinate index */
835 jnrlistA = jjnr[jidx];
836 jnrlistB = jjnr[jidx+1];
837 jnrlistC = jjnr[jidx+2];
838 jnrlistD = jjnr[jidx+3];
839 jnrlistE = jjnr[jidx+4];
840 jnrlistF = jjnr[jidx+5];
841 jnrlistG = jjnr[jidx+6];
842 jnrlistH = jjnr[jidx+7];
843 /* Sign of each element will be negative for non-real atoms.
844 * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
845 * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
847 dummy_mask = gmx_mm256_set_m128(gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx+4)),_mm_setzero_si128())),
848 gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128())));
850 jnrA = (jnrlistA>=0) ? jnrlistA : 0;
851 jnrB = (jnrlistB>=0) ? jnrlistB : 0;
852 jnrC = (jnrlistC>=0) ? jnrlistC : 0;
853 jnrD = (jnrlistD>=0) ? jnrlistD : 0;
854 jnrE = (jnrlistE>=0) ? jnrlistE : 0;
855 jnrF = (jnrlistF>=0) ? jnrlistF : 0;
856 jnrG = (jnrlistG>=0) ? jnrlistG : 0;
857 jnrH = (jnrlistH>=0) ? jnrlistH : 0;
858 j_coord_offsetA = DIM*jnrA;
859 j_coord_offsetB = DIM*jnrB;
860 j_coord_offsetC = DIM*jnrC;
861 j_coord_offsetD = DIM*jnrD;
862 j_coord_offsetE = DIM*jnrE;
863 j_coord_offsetF = DIM*jnrF;
864 j_coord_offsetG = DIM*jnrG;
865 j_coord_offsetH = DIM*jnrH;
867 /* load j atom coordinates */
868 gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
869 x+j_coord_offsetC,x+j_coord_offsetD,
870 x+j_coord_offsetE,x+j_coord_offsetF,
871 x+j_coord_offsetG,x+j_coord_offsetH,
874 /* Calculate displacement vector */
875 dx00 = _mm256_sub_ps(ix0,jx0);
876 dy00 = _mm256_sub_ps(iy0,jy0);
877 dz00 = _mm256_sub_ps(iz0,jz0);
879 /* Calculate squared distance and things based on it */
880 rsq00 = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
882 rinv00 = gmx_mm256_invsqrt_ps(rsq00);
884 rinvsq00 = _mm256_mul_ps(rinv00,rinv00);
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);
891 vdwjidx0A = 2*vdwtype[jnrA+0];
892 vdwjidx0B = 2*vdwtype[jnrB+0];
893 vdwjidx0C = 2*vdwtype[jnrC+0];
894 vdwjidx0D = 2*vdwtype[jnrD+0];
895 vdwjidx0E = 2*vdwtype[jnrE+0];
896 vdwjidx0F = 2*vdwtype[jnrF+0];
897 vdwjidx0G = 2*vdwtype[jnrG+0];
898 vdwjidx0H = 2*vdwtype[jnrH+0];
900 /**************************
901 * CALCULATE INTERACTIONS *
902 **************************/
904 if (gmx_mm256_any_lt(rsq00,rcutoff2))
907 r00 = _mm256_mul_ps(rsq00,rinv00);
908 r00 = _mm256_andnot_ps(dummy_mask,r00);
910 /* Compute parameters for interactions between i and j atoms */
911 qq00 = _mm256_mul_ps(iq0,jq0);
912 gmx_mm256_load_8pair_swizzle_ps(vdwioffsetptr0+vdwjidx0A,
913 vdwioffsetptr0+vdwjidx0B,
914 vdwioffsetptr0+vdwjidx0C,
915 vdwioffsetptr0+vdwjidx0D,
916 vdwioffsetptr0+vdwjidx0E,
917 vdwioffsetptr0+vdwjidx0F,
918 vdwioffsetptr0+vdwjidx0G,
919 vdwioffsetptr0+vdwjidx0H,
922 c6grid_00 = gmx_mm256_load_8real_swizzle_ps(vdwgridioffsetptr0+vdwjidx0A,
923 vdwgridioffsetptr0+vdwjidx0B,
924 vdwgridioffsetptr0+vdwjidx0C,
925 vdwgridioffsetptr0+vdwjidx0D,
926 vdwgridioffsetptr0+vdwjidx0E,
927 vdwgridioffsetptr0+vdwjidx0F,
928 vdwgridioffsetptr0+vdwjidx0G,
929 vdwgridioffsetptr0+vdwjidx0H);
931 /* EWALD ELECTROSTATICS */
933 /* Analytical PME correction */
934 zeta2 = _mm256_mul_ps(beta2,rsq00);
935 rinv3 = _mm256_mul_ps(rinvsq00,rinv00);
936 pmecorrF = gmx_mm256_pmecorrF_ps(zeta2);
937 felec = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
938 felec = _mm256_mul_ps(qq00,felec);
940 /* Analytical LJ-PME */
941 rinvsix = _mm256_mul_ps(_mm256_mul_ps(rinvsq00,rinvsq00),rinvsq00);
942 ewcljrsq = _mm256_mul_ps(ewclj2,rsq00);
943 ewclj6 = _mm256_mul_ps(ewclj2,_mm256_mul_ps(ewclj2,ewclj2));
944 exponent = gmx_simd_exp_r(ewcljrsq);
945 /* poly = exp(-(beta*r)^2) * (1 + (beta*r)^2 + (beta*r)^4 /2) */
946 poly = _mm256_mul_ps(exponent,_mm256_add_ps(_mm256_sub_ps(one,ewcljrsq),_mm256_mul_ps(_mm256_mul_ps(ewcljrsq,ewcljrsq),one_half)));
947 /* f6A = 6 * C6grid * (1 - poly) */
948 f6A = _mm256_mul_ps(c6grid_00,_mm256_sub_ps(one,poly));
949 /* f6B = C6grid * exponent * beta^6 */
950 f6B = _mm256_mul_ps(_mm256_mul_ps(c6grid_00,one_sixth),_mm256_mul_ps(exponent,ewclj6));
951 /* fvdw = 12*C12/r13 - ((6*C6 - f6A)/r6 + f6B)/r */
952 fvdw = _mm256_mul_ps(_mm256_add_ps(_mm256_mul_ps(_mm256_sub_ps(_mm256_mul_ps(c12_00,rinvsix),_mm256_sub_ps(c6_00,f6A)),rinvsix),f6B),rinvsq00);
954 cutoff_mask = _mm256_cmp_ps(rsq00,rcutoff2,_CMP_LT_OQ);
956 fscal = _mm256_add_ps(felec,fvdw);
958 fscal = _mm256_and_ps(fscal,cutoff_mask);
960 fscal = _mm256_andnot_ps(dummy_mask,fscal);
962 /* Calculate temporary vectorial force */
963 tx = _mm256_mul_ps(fscal,dx00);
964 ty = _mm256_mul_ps(fscal,dy00);
965 tz = _mm256_mul_ps(fscal,dz00);
967 /* Update vectorial force */
968 fix0 = _mm256_add_ps(fix0,tx);
969 fiy0 = _mm256_add_ps(fiy0,ty);
970 fiz0 = _mm256_add_ps(fiz0,tz);
972 fjptrA = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
973 fjptrB = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
974 fjptrC = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
975 fjptrD = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
976 fjptrE = (jnrlistE>=0) ? f+j_coord_offsetE : scratch;
977 fjptrF = (jnrlistF>=0) ? f+j_coord_offsetF : scratch;
978 fjptrG = (jnrlistG>=0) ? f+j_coord_offsetG : scratch;
979 fjptrH = (jnrlistH>=0) ? f+j_coord_offsetH : scratch;
980 gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,tx,ty,tz);
984 /* Inner loop uses 83 flops */
987 /* End of innermost loop */
989 gmx_mm256_update_iforce_1atom_swizzle_ps(fix0,fiy0,fiz0,
990 f+i_coord_offset,fshift+i_shift_offset);
992 /* Increment number of inner iterations */
993 inneriter += j_index_end - j_index_start;
995 /* Outer loop uses 7 flops */
998 /* Increment number of outer iterations */
1001 /* Update outer/inner flops */
1003 inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_F,outeriter*7 + inneriter*83);