2 * This file is part of the GROMACS molecular simulation package.
4 * Copyright (c) 2012,2013, by the GROMACS development team, led by
5 * Mark Abraham, David van der Spoel, Berk Hess, and Erik Lindahl,
6 * and including many others, as listed in the AUTHORS file in the
7 * top-level source directory and at http://www.gromacs.org.
9 * GROMACS is free software; you can redistribute it and/or
10 * modify it under the terms of the GNU Lesser General Public License
11 * as published by the Free Software Foundation; either version 2.1
12 * of the License, or (at your option) any later version.
14 * GROMACS is distributed in the hope that it will be useful,
15 * but WITHOUT ANY WARRANTY; without even the implied warranty of
16 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
17 * Lesser General Public License for more details.
19 * You should have received a copy of the GNU Lesser General Public
20 * License along with GROMACS; if not, see
21 * http://www.gnu.org/licenses, or write to the Free Software Foundation,
22 * Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA.
24 * If you want to redistribute modifications to GROMACS, please
25 * consider that scientific software is very special. Version
26 * control is crucial - bugs must be traceable. We will be happy to
27 * consider code for inclusion in the official distribution, but
28 * derived work must not be called official GROMACS. Details are found
29 * in the README & COPYING files - if they are missing, get the
30 * official version at http://www.gromacs.org.
32 * To help us fund GROMACS development, we humbly ask that you cite
33 * the research papers on the package. Check out http://www.gromacs.org.
36 * Note: this file was generated by the GROMACS avx_256_single kernel generator.
44 #include "../nb_kernel.h"
45 #include "types/simple.h"
49 #include "gromacs/simd/math_x86_avx_256_single.h"
50 #include "kernelutil_x86_avx_256_single.h"
53 * Gromacs nonbonded kernel: nb_kernel_ElecCoul_VdwCSTab_GeomW4P1_VF_avx_256_single
54 * Electrostatics interaction: Coulomb
55 * VdW interaction: CubicSplineTable
56 * Geometry: Water4-Particle
57 * Calculate force/pot: PotentialAndForce
60 nb_kernel_ElecCoul_VdwCSTab_GeomW4P1_VF_avx_256_single
61 (t_nblist * gmx_restrict nlist,
62 rvec * gmx_restrict xx,
63 rvec * gmx_restrict ff,
64 t_forcerec * gmx_restrict fr,
65 t_mdatoms * gmx_restrict mdatoms,
66 nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
67 t_nrnb * gmx_restrict nrnb)
69 /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
70 * just 0 for non-waters.
71 * Suffixes A,B,C,D,E,F,G,H refer to j loop unrolling done with AVX, e.g. for the eight different
72 * jnr indices corresponding to data put in the four positions in the SIMD register.
74 int i_shift_offset,i_coord_offset,outeriter,inneriter;
75 int j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
76 int jnrA,jnrB,jnrC,jnrD;
77 int jnrE,jnrF,jnrG,jnrH;
78 int jnrlistA,jnrlistB,jnrlistC,jnrlistD;
79 int jnrlistE,jnrlistF,jnrlistG,jnrlistH;
80 int j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
81 int j_coord_offsetE,j_coord_offsetF,j_coord_offsetG,j_coord_offsetH;
82 int *iinr,*jindex,*jjnr,*shiftidx,*gid;
84 real *shiftvec,*fshift,*x,*f;
85 real *fjptrA,*fjptrB,*fjptrC,*fjptrD,*fjptrE,*fjptrF,*fjptrG,*fjptrH;
87 __m256 tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
88 real * vdwioffsetptr0;
89 __m256 ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
90 real * vdwioffsetptr1;
91 __m256 ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
92 real * vdwioffsetptr2;
93 __m256 ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
94 real * vdwioffsetptr3;
95 __m256 ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
96 int vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D,vdwjidx0E,vdwjidx0F,vdwjidx0G,vdwjidx0H;
97 __m256 jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
98 __m256 dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
99 __m256 dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
100 __m256 dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
101 __m256 dx30,dy30,dz30,rsq30,rinv30,rinvsq30,r30,qq30,c6_30,c12_30;
102 __m256 velec,felec,velecsum,facel,crf,krf,krf2;
105 __m256 rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
108 __m256 one_sixth = _mm256_set1_ps(1.0/6.0);
109 __m256 one_twelfth = _mm256_set1_ps(1.0/12.0);
111 __m128i vfitab_lo,vfitab_hi;
112 __m128i ifour = _mm_set1_epi32(4);
113 __m256 rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF;
115 __m256 dummy_mask,cutoff_mask;
116 __m256 signbit = _mm256_castsi256_ps( _mm256_set1_epi32(0x80000000) );
117 __m256 one = _mm256_set1_ps(1.0);
118 __m256 two = _mm256_set1_ps(2.0);
124 jindex = nlist->jindex;
126 shiftidx = nlist->shift;
128 shiftvec = fr->shift_vec[0];
129 fshift = fr->fshift[0];
130 facel = _mm256_set1_ps(fr->epsfac);
131 charge = mdatoms->chargeA;
132 nvdwtype = fr->ntype;
134 vdwtype = mdatoms->typeA;
136 vftab = kernel_data->table_vdw->data;
137 vftabscale = _mm256_set1_ps(kernel_data->table_vdw->scale);
139 /* Setup water-specific parameters */
140 inr = nlist->iinr[0];
141 iq1 = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+1]));
142 iq2 = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+2]));
143 iq3 = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+3]));
144 vdwioffsetptr0 = vdwparam+2*nvdwtype*vdwtype[inr+0];
146 /* Avoid stupid compiler warnings */
147 jnrA = jnrB = jnrC = jnrD = jnrE = jnrF = jnrG = jnrH = 0;
160 for(iidx=0;iidx<4*DIM;iidx++)
165 /* Start outer loop over neighborlists */
166 for(iidx=0; iidx<nri; iidx++)
168 /* Load shift vector for this list */
169 i_shift_offset = DIM*shiftidx[iidx];
171 /* Load limits for loop over neighbors */
172 j_index_start = jindex[iidx];
173 j_index_end = jindex[iidx+1];
175 /* Get outer coordinate index */
177 i_coord_offset = DIM*inr;
179 /* Load i particle coords and add shift vector */
180 gmx_mm256_load_shift_and_4rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
181 &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
183 fix0 = _mm256_setzero_ps();
184 fiy0 = _mm256_setzero_ps();
185 fiz0 = _mm256_setzero_ps();
186 fix1 = _mm256_setzero_ps();
187 fiy1 = _mm256_setzero_ps();
188 fiz1 = _mm256_setzero_ps();
189 fix2 = _mm256_setzero_ps();
190 fiy2 = _mm256_setzero_ps();
191 fiz2 = _mm256_setzero_ps();
192 fix3 = _mm256_setzero_ps();
193 fiy3 = _mm256_setzero_ps();
194 fiz3 = _mm256_setzero_ps();
196 /* Reset potential sums */
197 velecsum = _mm256_setzero_ps();
198 vvdwsum = _mm256_setzero_ps();
200 /* Start inner kernel loop */
201 for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+7]>=0; jidx+=8)
204 /* Get j neighbor index, and coordinate index */
213 j_coord_offsetA = DIM*jnrA;
214 j_coord_offsetB = DIM*jnrB;
215 j_coord_offsetC = DIM*jnrC;
216 j_coord_offsetD = DIM*jnrD;
217 j_coord_offsetE = DIM*jnrE;
218 j_coord_offsetF = DIM*jnrF;
219 j_coord_offsetG = DIM*jnrG;
220 j_coord_offsetH = DIM*jnrH;
222 /* load j atom coordinates */
223 gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
224 x+j_coord_offsetC,x+j_coord_offsetD,
225 x+j_coord_offsetE,x+j_coord_offsetF,
226 x+j_coord_offsetG,x+j_coord_offsetH,
229 /* Calculate displacement vector */
230 dx00 = _mm256_sub_ps(ix0,jx0);
231 dy00 = _mm256_sub_ps(iy0,jy0);
232 dz00 = _mm256_sub_ps(iz0,jz0);
233 dx10 = _mm256_sub_ps(ix1,jx0);
234 dy10 = _mm256_sub_ps(iy1,jy0);
235 dz10 = _mm256_sub_ps(iz1,jz0);
236 dx20 = _mm256_sub_ps(ix2,jx0);
237 dy20 = _mm256_sub_ps(iy2,jy0);
238 dz20 = _mm256_sub_ps(iz2,jz0);
239 dx30 = _mm256_sub_ps(ix3,jx0);
240 dy30 = _mm256_sub_ps(iy3,jy0);
241 dz30 = _mm256_sub_ps(iz3,jz0);
243 /* Calculate squared distance and things based on it */
244 rsq00 = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
245 rsq10 = gmx_mm256_calc_rsq_ps(dx10,dy10,dz10);
246 rsq20 = gmx_mm256_calc_rsq_ps(dx20,dy20,dz20);
247 rsq30 = gmx_mm256_calc_rsq_ps(dx30,dy30,dz30);
249 rinv00 = gmx_mm256_invsqrt_ps(rsq00);
250 rinv10 = gmx_mm256_invsqrt_ps(rsq10);
251 rinv20 = gmx_mm256_invsqrt_ps(rsq20);
252 rinv30 = gmx_mm256_invsqrt_ps(rsq30);
254 rinvsq10 = _mm256_mul_ps(rinv10,rinv10);
255 rinvsq20 = _mm256_mul_ps(rinv20,rinv20);
256 rinvsq30 = _mm256_mul_ps(rinv30,rinv30);
258 /* Load parameters for j particles */
259 jq0 = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
260 charge+jnrC+0,charge+jnrD+0,
261 charge+jnrE+0,charge+jnrF+0,
262 charge+jnrG+0,charge+jnrH+0);
263 vdwjidx0A = 2*vdwtype[jnrA+0];
264 vdwjidx0B = 2*vdwtype[jnrB+0];
265 vdwjidx0C = 2*vdwtype[jnrC+0];
266 vdwjidx0D = 2*vdwtype[jnrD+0];
267 vdwjidx0E = 2*vdwtype[jnrE+0];
268 vdwjidx0F = 2*vdwtype[jnrF+0];
269 vdwjidx0G = 2*vdwtype[jnrG+0];
270 vdwjidx0H = 2*vdwtype[jnrH+0];
272 fjx0 = _mm256_setzero_ps();
273 fjy0 = _mm256_setzero_ps();
274 fjz0 = _mm256_setzero_ps();
276 /**************************
277 * CALCULATE INTERACTIONS *
278 **************************/
280 r00 = _mm256_mul_ps(rsq00,rinv00);
282 /* Compute parameters for interactions between i and j atoms */
283 gmx_mm256_load_8pair_swizzle_ps(vdwioffsetptr0+vdwjidx0A,
284 vdwioffsetptr0+vdwjidx0B,
285 vdwioffsetptr0+vdwjidx0C,
286 vdwioffsetptr0+vdwjidx0D,
287 vdwioffsetptr0+vdwjidx0E,
288 vdwioffsetptr0+vdwjidx0F,
289 vdwioffsetptr0+vdwjidx0G,
290 vdwioffsetptr0+vdwjidx0H,
293 /* Calculate table index by multiplying r with table scale and truncate to integer */
294 rt = _mm256_mul_ps(r00,vftabscale);
295 vfitab = _mm256_cvttps_epi32(rt);
296 vfeps = _mm256_sub_ps(rt,_mm256_round_ps(rt, _MM_FROUND_FLOOR));
297 /* AVX1 does not support 256-bit integer operations, so now we go to 128-bit mode... */
298 vfitab_lo = _mm256_extractf128_si256(vfitab,0x0);
299 vfitab_hi = _mm256_extractf128_si256(vfitab,0x1);
300 vfitab_lo = _mm_slli_epi32(vfitab_lo,3);
301 vfitab_hi = _mm_slli_epi32(vfitab_hi,3);
303 /* CUBIC SPLINE TABLE DISPERSION */
304 Y = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
305 _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
306 F = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
307 _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
308 G = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
309 _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
310 H = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
311 _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
312 GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
313 Heps = _mm256_mul_ps(vfeps,H);
314 Fp = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
315 VV = _mm256_add_ps(Y,_mm256_mul_ps(vfeps,Fp));
316 vvdw6 = _mm256_mul_ps(c6_00,VV);
317 FF = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
318 fvdw6 = _mm256_mul_ps(c6_00,FF);
320 /* CUBIC SPLINE TABLE REPULSION */
321 vfitab_lo = _mm_add_epi32(vfitab_lo,ifour);
322 vfitab_hi = _mm_add_epi32(vfitab_hi,ifour);
323 Y = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
324 _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
325 F = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
326 _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
327 G = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
328 _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
329 H = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
330 _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
331 GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
332 Heps = _mm256_mul_ps(vfeps,H);
333 Fp = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
334 VV = _mm256_add_ps(Y,_mm256_mul_ps(vfeps,Fp));
335 vvdw12 = _mm256_mul_ps(c12_00,VV);
336 FF = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
337 fvdw12 = _mm256_mul_ps(c12_00,FF);
338 vvdw = _mm256_add_ps(vvdw12,vvdw6);
339 fvdw = _mm256_xor_ps(signbit,_mm256_mul_ps(_mm256_add_ps(fvdw6,fvdw12),_mm256_mul_ps(vftabscale,rinv00)));
341 /* Update potential sum for this i atom from the interaction with this j atom. */
342 vvdwsum = _mm256_add_ps(vvdwsum,vvdw);
346 /* Calculate temporary vectorial force */
347 tx = _mm256_mul_ps(fscal,dx00);
348 ty = _mm256_mul_ps(fscal,dy00);
349 tz = _mm256_mul_ps(fscal,dz00);
351 /* Update vectorial force */
352 fix0 = _mm256_add_ps(fix0,tx);
353 fiy0 = _mm256_add_ps(fiy0,ty);
354 fiz0 = _mm256_add_ps(fiz0,tz);
356 fjx0 = _mm256_add_ps(fjx0,tx);
357 fjy0 = _mm256_add_ps(fjy0,ty);
358 fjz0 = _mm256_add_ps(fjz0,tz);
360 /**************************
361 * CALCULATE INTERACTIONS *
362 **************************/
364 /* Compute parameters for interactions between i and j atoms */
365 qq10 = _mm256_mul_ps(iq1,jq0);
367 /* COULOMB ELECTROSTATICS */
368 velec = _mm256_mul_ps(qq10,rinv10);
369 felec = _mm256_mul_ps(velec,rinvsq10);
371 /* Update potential sum for this i atom from the interaction with this j atom. */
372 velecsum = _mm256_add_ps(velecsum,velec);
376 /* Calculate temporary vectorial force */
377 tx = _mm256_mul_ps(fscal,dx10);
378 ty = _mm256_mul_ps(fscal,dy10);
379 tz = _mm256_mul_ps(fscal,dz10);
381 /* Update vectorial force */
382 fix1 = _mm256_add_ps(fix1,tx);
383 fiy1 = _mm256_add_ps(fiy1,ty);
384 fiz1 = _mm256_add_ps(fiz1,tz);
386 fjx0 = _mm256_add_ps(fjx0,tx);
387 fjy0 = _mm256_add_ps(fjy0,ty);
388 fjz0 = _mm256_add_ps(fjz0,tz);
390 /**************************
391 * CALCULATE INTERACTIONS *
392 **************************/
394 /* Compute parameters for interactions between i and j atoms */
395 qq20 = _mm256_mul_ps(iq2,jq0);
397 /* COULOMB ELECTROSTATICS */
398 velec = _mm256_mul_ps(qq20,rinv20);
399 felec = _mm256_mul_ps(velec,rinvsq20);
401 /* Update potential sum for this i atom from the interaction with this j atom. */
402 velecsum = _mm256_add_ps(velecsum,velec);
406 /* Calculate temporary vectorial force */
407 tx = _mm256_mul_ps(fscal,dx20);
408 ty = _mm256_mul_ps(fscal,dy20);
409 tz = _mm256_mul_ps(fscal,dz20);
411 /* Update vectorial force */
412 fix2 = _mm256_add_ps(fix2,tx);
413 fiy2 = _mm256_add_ps(fiy2,ty);
414 fiz2 = _mm256_add_ps(fiz2,tz);
416 fjx0 = _mm256_add_ps(fjx0,tx);
417 fjy0 = _mm256_add_ps(fjy0,ty);
418 fjz0 = _mm256_add_ps(fjz0,tz);
420 /**************************
421 * CALCULATE INTERACTIONS *
422 **************************/
424 /* Compute parameters for interactions between i and j atoms */
425 qq30 = _mm256_mul_ps(iq3,jq0);
427 /* COULOMB ELECTROSTATICS */
428 velec = _mm256_mul_ps(qq30,rinv30);
429 felec = _mm256_mul_ps(velec,rinvsq30);
431 /* Update potential sum for this i atom from the interaction with this j atom. */
432 velecsum = _mm256_add_ps(velecsum,velec);
436 /* Calculate temporary vectorial force */
437 tx = _mm256_mul_ps(fscal,dx30);
438 ty = _mm256_mul_ps(fscal,dy30);
439 tz = _mm256_mul_ps(fscal,dz30);
441 /* Update vectorial force */
442 fix3 = _mm256_add_ps(fix3,tx);
443 fiy3 = _mm256_add_ps(fiy3,ty);
444 fiz3 = _mm256_add_ps(fiz3,tz);
446 fjx0 = _mm256_add_ps(fjx0,tx);
447 fjy0 = _mm256_add_ps(fjy0,ty);
448 fjz0 = _mm256_add_ps(fjz0,tz);
450 fjptrA = f+j_coord_offsetA;
451 fjptrB = f+j_coord_offsetB;
452 fjptrC = f+j_coord_offsetC;
453 fjptrD = f+j_coord_offsetD;
454 fjptrE = f+j_coord_offsetE;
455 fjptrF = f+j_coord_offsetF;
456 fjptrG = f+j_coord_offsetG;
457 fjptrH = f+j_coord_offsetH;
459 gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,fjx0,fjy0,fjz0);
461 /* Inner loop uses 140 flops */
467 /* Get j neighbor index, and coordinate index */
468 jnrlistA = jjnr[jidx];
469 jnrlistB = jjnr[jidx+1];
470 jnrlistC = jjnr[jidx+2];
471 jnrlistD = jjnr[jidx+3];
472 jnrlistE = jjnr[jidx+4];
473 jnrlistF = jjnr[jidx+5];
474 jnrlistG = jjnr[jidx+6];
475 jnrlistH = jjnr[jidx+7];
476 /* Sign of each element will be negative for non-real atoms.
477 * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
478 * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
480 dummy_mask = gmx_mm256_set_m128(gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx+4)),_mm_setzero_si128())),
481 gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128())));
483 jnrA = (jnrlistA>=0) ? jnrlistA : 0;
484 jnrB = (jnrlistB>=0) ? jnrlistB : 0;
485 jnrC = (jnrlistC>=0) ? jnrlistC : 0;
486 jnrD = (jnrlistD>=0) ? jnrlistD : 0;
487 jnrE = (jnrlistE>=0) ? jnrlistE : 0;
488 jnrF = (jnrlistF>=0) ? jnrlistF : 0;
489 jnrG = (jnrlistG>=0) ? jnrlistG : 0;
490 jnrH = (jnrlistH>=0) ? jnrlistH : 0;
491 j_coord_offsetA = DIM*jnrA;
492 j_coord_offsetB = DIM*jnrB;
493 j_coord_offsetC = DIM*jnrC;
494 j_coord_offsetD = DIM*jnrD;
495 j_coord_offsetE = DIM*jnrE;
496 j_coord_offsetF = DIM*jnrF;
497 j_coord_offsetG = DIM*jnrG;
498 j_coord_offsetH = DIM*jnrH;
500 /* load j atom coordinates */
501 gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
502 x+j_coord_offsetC,x+j_coord_offsetD,
503 x+j_coord_offsetE,x+j_coord_offsetF,
504 x+j_coord_offsetG,x+j_coord_offsetH,
507 /* Calculate displacement vector */
508 dx00 = _mm256_sub_ps(ix0,jx0);
509 dy00 = _mm256_sub_ps(iy0,jy0);
510 dz00 = _mm256_sub_ps(iz0,jz0);
511 dx10 = _mm256_sub_ps(ix1,jx0);
512 dy10 = _mm256_sub_ps(iy1,jy0);
513 dz10 = _mm256_sub_ps(iz1,jz0);
514 dx20 = _mm256_sub_ps(ix2,jx0);
515 dy20 = _mm256_sub_ps(iy2,jy0);
516 dz20 = _mm256_sub_ps(iz2,jz0);
517 dx30 = _mm256_sub_ps(ix3,jx0);
518 dy30 = _mm256_sub_ps(iy3,jy0);
519 dz30 = _mm256_sub_ps(iz3,jz0);
521 /* Calculate squared distance and things based on it */
522 rsq00 = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
523 rsq10 = gmx_mm256_calc_rsq_ps(dx10,dy10,dz10);
524 rsq20 = gmx_mm256_calc_rsq_ps(dx20,dy20,dz20);
525 rsq30 = gmx_mm256_calc_rsq_ps(dx30,dy30,dz30);
527 rinv00 = gmx_mm256_invsqrt_ps(rsq00);
528 rinv10 = gmx_mm256_invsqrt_ps(rsq10);
529 rinv20 = gmx_mm256_invsqrt_ps(rsq20);
530 rinv30 = gmx_mm256_invsqrt_ps(rsq30);
532 rinvsq10 = _mm256_mul_ps(rinv10,rinv10);
533 rinvsq20 = _mm256_mul_ps(rinv20,rinv20);
534 rinvsq30 = _mm256_mul_ps(rinv30,rinv30);
536 /* Load parameters for j particles */
537 jq0 = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
538 charge+jnrC+0,charge+jnrD+0,
539 charge+jnrE+0,charge+jnrF+0,
540 charge+jnrG+0,charge+jnrH+0);
541 vdwjidx0A = 2*vdwtype[jnrA+0];
542 vdwjidx0B = 2*vdwtype[jnrB+0];
543 vdwjidx0C = 2*vdwtype[jnrC+0];
544 vdwjidx0D = 2*vdwtype[jnrD+0];
545 vdwjidx0E = 2*vdwtype[jnrE+0];
546 vdwjidx0F = 2*vdwtype[jnrF+0];
547 vdwjidx0G = 2*vdwtype[jnrG+0];
548 vdwjidx0H = 2*vdwtype[jnrH+0];
550 fjx0 = _mm256_setzero_ps();
551 fjy0 = _mm256_setzero_ps();
552 fjz0 = _mm256_setzero_ps();
554 /**************************
555 * CALCULATE INTERACTIONS *
556 **************************/
558 r00 = _mm256_mul_ps(rsq00,rinv00);
559 r00 = _mm256_andnot_ps(dummy_mask,r00);
561 /* Compute parameters for interactions between i and j atoms */
562 gmx_mm256_load_8pair_swizzle_ps(vdwioffsetptr0+vdwjidx0A,
563 vdwioffsetptr0+vdwjidx0B,
564 vdwioffsetptr0+vdwjidx0C,
565 vdwioffsetptr0+vdwjidx0D,
566 vdwioffsetptr0+vdwjidx0E,
567 vdwioffsetptr0+vdwjidx0F,
568 vdwioffsetptr0+vdwjidx0G,
569 vdwioffsetptr0+vdwjidx0H,
572 /* Calculate table index by multiplying r with table scale and truncate to integer */
573 rt = _mm256_mul_ps(r00,vftabscale);
574 vfitab = _mm256_cvttps_epi32(rt);
575 vfeps = _mm256_sub_ps(rt,_mm256_round_ps(rt, _MM_FROUND_FLOOR));
576 /* AVX1 does not support 256-bit integer operations, so now we go to 128-bit mode... */
577 vfitab_lo = _mm256_extractf128_si256(vfitab,0x0);
578 vfitab_hi = _mm256_extractf128_si256(vfitab,0x1);
579 vfitab_lo = _mm_slli_epi32(vfitab_lo,3);
580 vfitab_hi = _mm_slli_epi32(vfitab_hi,3);
582 /* CUBIC SPLINE TABLE DISPERSION */
583 Y = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
584 _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
585 F = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
586 _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
587 G = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
588 _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
589 H = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
590 _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
591 GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
592 Heps = _mm256_mul_ps(vfeps,H);
593 Fp = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
594 VV = _mm256_add_ps(Y,_mm256_mul_ps(vfeps,Fp));
595 vvdw6 = _mm256_mul_ps(c6_00,VV);
596 FF = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
597 fvdw6 = _mm256_mul_ps(c6_00,FF);
599 /* CUBIC SPLINE TABLE REPULSION */
600 vfitab_lo = _mm_add_epi32(vfitab_lo,ifour);
601 vfitab_hi = _mm_add_epi32(vfitab_hi,ifour);
602 Y = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
603 _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
604 F = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
605 _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
606 G = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
607 _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
608 H = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
609 _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
610 GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
611 Heps = _mm256_mul_ps(vfeps,H);
612 Fp = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
613 VV = _mm256_add_ps(Y,_mm256_mul_ps(vfeps,Fp));
614 vvdw12 = _mm256_mul_ps(c12_00,VV);
615 FF = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
616 fvdw12 = _mm256_mul_ps(c12_00,FF);
617 vvdw = _mm256_add_ps(vvdw12,vvdw6);
618 fvdw = _mm256_xor_ps(signbit,_mm256_mul_ps(_mm256_add_ps(fvdw6,fvdw12),_mm256_mul_ps(vftabscale,rinv00)));
620 /* Update potential sum for this i atom from the interaction with this j atom. */
621 vvdw = _mm256_andnot_ps(dummy_mask,vvdw);
622 vvdwsum = _mm256_add_ps(vvdwsum,vvdw);
626 fscal = _mm256_andnot_ps(dummy_mask,fscal);
628 /* Calculate temporary vectorial force */
629 tx = _mm256_mul_ps(fscal,dx00);
630 ty = _mm256_mul_ps(fscal,dy00);
631 tz = _mm256_mul_ps(fscal,dz00);
633 /* Update vectorial force */
634 fix0 = _mm256_add_ps(fix0,tx);
635 fiy0 = _mm256_add_ps(fiy0,ty);
636 fiz0 = _mm256_add_ps(fiz0,tz);
638 fjx0 = _mm256_add_ps(fjx0,tx);
639 fjy0 = _mm256_add_ps(fjy0,ty);
640 fjz0 = _mm256_add_ps(fjz0,tz);
642 /**************************
643 * CALCULATE INTERACTIONS *
644 **************************/
646 /* Compute parameters for interactions between i and j atoms */
647 qq10 = _mm256_mul_ps(iq1,jq0);
649 /* COULOMB ELECTROSTATICS */
650 velec = _mm256_mul_ps(qq10,rinv10);
651 felec = _mm256_mul_ps(velec,rinvsq10);
653 /* Update potential sum for this i atom from the interaction with this j atom. */
654 velec = _mm256_andnot_ps(dummy_mask,velec);
655 velecsum = _mm256_add_ps(velecsum,velec);
659 fscal = _mm256_andnot_ps(dummy_mask,fscal);
661 /* Calculate temporary vectorial force */
662 tx = _mm256_mul_ps(fscal,dx10);
663 ty = _mm256_mul_ps(fscal,dy10);
664 tz = _mm256_mul_ps(fscal,dz10);
666 /* Update vectorial force */
667 fix1 = _mm256_add_ps(fix1,tx);
668 fiy1 = _mm256_add_ps(fiy1,ty);
669 fiz1 = _mm256_add_ps(fiz1,tz);
671 fjx0 = _mm256_add_ps(fjx0,tx);
672 fjy0 = _mm256_add_ps(fjy0,ty);
673 fjz0 = _mm256_add_ps(fjz0,tz);
675 /**************************
676 * CALCULATE INTERACTIONS *
677 **************************/
679 /* Compute parameters for interactions between i and j atoms */
680 qq20 = _mm256_mul_ps(iq2,jq0);
682 /* COULOMB ELECTROSTATICS */
683 velec = _mm256_mul_ps(qq20,rinv20);
684 felec = _mm256_mul_ps(velec,rinvsq20);
686 /* Update potential sum for this i atom from the interaction with this j atom. */
687 velec = _mm256_andnot_ps(dummy_mask,velec);
688 velecsum = _mm256_add_ps(velecsum,velec);
692 fscal = _mm256_andnot_ps(dummy_mask,fscal);
694 /* Calculate temporary vectorial force */
695 tx = _mm256_mul_ps(fscal,dx20);
696 ty = _mm256_mul_ps(fscal,dy20);
697 tz = _mm256_mul_ps(fscal,dz20);
699 /* Update vectorial force */
700 fix2 = _mm256_add_ps(fix2,tx);
701 fiy2 = _mm256_add_ps(fiy2,ty);
702 fiz2 = _mm256_add_ps(fiz2,tz);
704 fjx0 = _mm256_add_ps(fjx0,tx);
705 fjy0 = _mm256_add_ps(fjy0,ty);
706 fjz0 = _mm256_add_ps(fjz0,tz);
708 /**************************
709 * CALCULATE INTERACTIONS *
710 **************************/
712 /* Compute parameters for interactions between i and j atoms */
713 qq30 = _mm256_mul_ps(iq3,jq0);
715 /* COULOMB ELECTROSTATICS */
716 velec = _mm256_mul_ps(qq30,rinv30);
717 felec = _mm256_mul_ps(velec,rinvsq30);
719 /* Update potential sum for this i atom from the interaction with this j atom. */
720 velec = _mm256_andnot_ps(dummy_mask,velec);
721 velecsum = _mm256_add_ps(velecsum,velec);
725 fscal = _mm256_andnot_ps(dummy_mask,fscal);
727 /* Calculate temporary vectorial force */
728 tx = _mm256_mul_ps(fscal,dx30);
729 ty = _mm256_mul_ps(fscal,dy30);
730 tz = _mm256_mul_ps(fscal,dz30);
732 /* Update vectorial force */
733 fix3 = _mm256_add_ps(fix3,tx);
734 fiy3 = _mm256_add_ps(fiy3,ty);
735 fiz3 = _mm256_add_ps(fiz3,tz);
737 fjx0 = _mm256_add_ps(fjx0,tx);
738 fjy0 = _mm256_add_ps(fjy0,ty);
739 fjz0 = _mm256_add_ps(fjz0,tz);
741 fjptrA = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
742 fjptrB = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
743 fjptrC = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
744 fjptrD = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
745 fjptrE = (jnrlistE>=0) ? f+j_coord_offsetE : scratch;
746 fjptrF = (jnrlistF>=0) ? f+j_coord_offsetF : scratch;
747 fjptrG = (jnrlistG>=0) ? f+j_coord_offsetG : scratch;
748 fjptrH = (jnrlistH>=0) ? f+j_coord_offsetH : scratch;
750 gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,fjx0,fjy0,fjz0);
752 /* Inner loop uses 141 flops */
755 /* End of innermost loop */
757 gmx_mm256_update_iforce_4atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
758 f+i_coord_offset,fshift+i_shift_offset);
761 /* Update potential energies */
762 gmx_mm256_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid);
763 gmx_mm256_update_1pot_ps(vvdwsum,kernel_data->energygrp_vdw+ggid);
765 /* Increment number of inner iterations */
766 inneriter += j_index_end - j_index_start;
768 /* Outer loop uses 26 flops */
771 /* Increment number of outer iterations */
774 /* Update outer/inner flops */
776 inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W4_VF,outeriter*26 + inneriter*141);
779 * Gromacs nonbonded kernel: nb_kernel_ElecCoul_VdwCSTab_GeomW4P1_F_avx_256_single
780 * Electrostatics interaction: Coulomb
781 * VdW interaction: CubicSplineTable
782 * Geometry: Water4-Particle
783 * Calculate force/pot: Force
786 nb_kernel_ElecCoul_VdwCSTab_GeomW4P1_F_avx_256_single
787 (t_nblist * gmx_restrict nlist,
788 rvec * gmx_restrict xx,
789 rvec * gmx_restrict ff,
790 t_forcerec * gmx_restrict fr,
791 t_mdatoms * gmx_restrict mdatoms,
792 nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
793 t_nrnb * gmx_restrict nrnb)
795 /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
796 * just 0 for non-waters.
797 * Suffixes A,B,C,D,E,F,G,H refer to j loop unrolling done with AVX, e.g. for the eight different
798 * jnr indices corresponding to data put in the four positions in the SIMD register.
800 int i_shift_offset,i_coord_offset,outeriter,inneriter;
801 int j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
802 int jnrA,jnrB,jnrC,jnrD;
803 int jnrE,jnrF,jnrG,jnrH;
804 int jnrlistA,jnrlistB,jnrlistC,jnrlistD;
805 int jnrlistE,jnrlistF,jnrlistG,jnrlistH;
806 int j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
807 int j_coord_offsetE,j_coord_offsetF,j_coord_offsetG,j_coord_offsetH;
808 int *iinr,*jindex,*jjnr,*shiftidx,*gid;
810 real *shiftvec,*fshift,*x,*f;
811 real *fjptrA,*fjptrB,*fjptrC,*fjptrD,*fjptrE,*fjptrF,*fjptrG,*fjptrH;
813 __m256 tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
814 real * vdwioffsetptr0;
815 __m256 ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
816 real * vdwioffsetptr1;
817 __m256 ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
818 real * vdwioffsetptr2;
819 __m256 ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
820 real * vdwioffsetptr3;
821 __m256 ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
822 int vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D,vdwjidx0E,vdwjidx0F,vdwjidx0G,vdwjidx0H;
823 __m256 jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
824 __m256 dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
825 __m256 dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
826 __m256 dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
827 __m256 dx30,dy30,dz30,rsq30,rinv30,rinvsq30,r30,qq30,c6_30,c12_30;
828 __m256 velec,felec,velecsum,facel,crf,krf,krf2;
831 __m256 rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
834 __m256 one_sixth = _mm256_set1_ps(1.0/6.0);
835 __m256 one_twelfth = _mm256_set1_ps(1.0/12.0);
837 __m128i vfitab_lo,vfitab_hi;
838 __m128i ifour = _mm_set1_epi32(4);
839 __m256 rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF;
841 __m256 dummy_mask,cutoff_mask;
842 __m256 signbit = _mm256_castsi256_ps( _mm256_set1_epi32(0x80000000) );
843 __m256 one = _mm256_set1_ps(1.0);
844 __m256 two = _mm256_set1_ps(2.0);
850 jindex = nlist->jindex;
852 shiftidx = nlist->shift;
854 shiftvec = fr->shift_vec[0];
855 fshift = fr->fshift[0];
856 facel = _mm256_set1_ps(fr->epsfac);
857 charge = mdatoms->chargeA;
858 nvdwtype = fr->ntype;
860 vdwtype = mdatoms->typeA;
862 vftab = kernel_data->table_vdw->data;
863 vftabscale = _mm256_set1_ps(kernel_data->table_vdw->scale);
865 /* Setup water-specific parameters */
866 inr = nlist->iinr[0];
867 iq1 = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+1]));
868 iq2 = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+2]));
869 iq3 = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+3]));
870 vdwioffsetptr0 = vdwparam+2*nvdwtype*vdwtype[inr+0];
872 /* Avoid stupid compiler warnings */
873 jnrA = jnrB = jnrC = jnrD = jnrE = jnrF = jnrG = jnrH = 0;
886 for(iidx=0;iidx<4*DIM;iidx++)
891 /* Start outer loop over neighborlists */
892 for(iidx=0; iidx<nri; iidx++)
894 /* Load shift vector for this list */
895 i_shift_offset = DIM*shiftidx[iidx];
897 /* Load limits for loop over neighbors */
898 j_index_start = jindex[iidx];
899 j_index_end = jindex[iidx+1];
901 /* Get outer coordinate index */
903 i_coord_offset = DIM*inr;
905 /* Load i particle coords and add shift vector */
906 gmx_mm256_load_shift_and_4rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
907 &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
909 fix0 = _mm256_setzero_ps();
910 fiy0 = _mm256_setzero_ps();
911 fiz0 = _mm256_setzero_ps();
912 fix1 = _mm256_setzero_ps();
913 fiy1 = _mm256_setzero_ps();
914 fiz1 = _mm256_setzero_ps();
915 fix2 = _mm256_setzero_ps();
916 fiy2 = _mm256_setzero_ps();
917 fiz2 = _mm256_setzero_ps();
918 fix3 = _mm256_setzero_ps();
919 fiy3 = _mm256_setzero_ps();
920 fiz3 = _mm256_setzero_ps();
922 /* Start inner kernel loop */
923 for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+7]>=0; jidx+=8)
926 /* Get j neighbor index, and coordinate index */
935 j_coord_offsetA = DIM*jnrA;
936 j_coord_offsetB = DIM*jnrB;
937 j_coord_offsetC = DIM*jnrC;
938 j_coord_offsetD = DIM*jnrD;
939 j_coord_offsetE = DIM*jnrE;
940 j_coord_offsetF = DIM*jnrF;
941 j_coord_offsetG = DIM*jnrG;
942 j_coord_offsetH = DIM*jnrH;
944 /* load j atom coordinates */
945 gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
946 x+j_coord_offsetC,x+j_coord_offsetD,
947 x+j_coord_offsetE,x+j_coord_offsetF,
948 x+j_coord_offsetG,x+j_coord_offsetH,
951 /* Calculate displacement vector */
952 dx00 = _mm256_sub_ps(ix0,jx0);
953 dy00 = _mm256_sub_ps(iy0,jy0);
954 dz00 = _mm256_sub_ps(iz0,jz0);
955 dx10 = _mm256_sub_ps(ix1,jx0);
956 dy10 = _mm256_sub_ps(iy1,jy0);
957 dz10 = _mm256_sub_ps(iz1,jz0);
958 dx20 = _mm256_sub_ps(ix2,jx0);
959 dy20 = _mm256_sub_ps(iy2,jy0);
960 dz20 = _mm256_sub_ps(iz2,jz0);
961 dx30 = _mm256_sub_ps(ix3,jx0);
962 dy30 = _mm256_sub_ps(iy3,jy0);
963 dz30 = _mm256_sub_ps(iz3,jz0);
965 /* Calculate squared distance and things based on it */
966 rsq00 = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
967 rsq10 = gmx_mm256_calc_rsq_ps(dx10,dy10,dz10);
968 rsq20 = gmx_mm256_calc_rsq_ps(dx20,dy20,dz20);
969 rsq30 = gmx_mm256_calc_rsq_ps(dx30,dy30,dz30);
971 rinv00 = gmx_mm256_invsqrt_ps(rsq00);
972 rinv10 = gmx_mm256_invsqrt_ps(rsq10);
973 rinv20 = gmx_mm256_invsqrt_ps(rsq20);
974 rinv30 = gmx_mm256_invsqrt_ps(rsq30);
976 rinvsq10 = _mm256_mul_ps(rinv10,rinv10);
977 rinvsq20 = _mm256_mul_ps(rinv20,rinv20);
978 rinvsq30 = _mm256_mul_ps(rinv30,rinv30);
980 /* Load parameters for j particles */
981 jq0 = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
982 charge+jnrC+0,charge+jnrD+0,
983 charge+jnrE+0,charge+jnrF+0,
984 charge+jnrG+0,charge+jnrH+0);
985 vdwjidx0A = 2*vdwtype[jnrA+0];
986 vdwjidx0B = 2*vdwtype[jnrB+0];
987 vdwjidx0C = 2*vdwtype[jnrC+0];
988 vdwjidx0D = 2*vdwtype[jnrD+0];
989 vdwjidx0E = 2*vdwtype[jnrE+0];
990 vdwjidx0F = 2*vdwtype[jnrF+0];
991 vdwjidx0G = 2*vdwtype[jnrG+0];
992 vdwjidx0H = 2*vdwtype[jnrH+0];
994 fjx0 = _mm256_setzero_ps();
995 fjy0 = _mm256_setzero_ps();
996 fjz0 = _mm256_setzero_ps();
998 /**************************
999 * CALCULATE INTERACTIONS *
1000 **************************/
1002 r00 = _mm256_mul_ps(rsq00,rinv00);
1004 /* Compute parameters for interactions between i and j atoms */
1005 gmx_mm256_load_8pair_swizzle_ps(vdwioffsetptr0+vdwjidx0A,
1006 vdwioffsetptr0+vdwjidx0B,
1007 vdwioffsetptr0+vdwjidx0C,
1008 vdwioffsetptr0+vdwjidx0D,
1009 vdwioffsetptr0+vdwjidx0E,
1010 vdwioffsetptr0+vdwjidx0F,
1011 vdwioffsetptr0+vdwjidx0G,
1012 vdwioffsetptr0+vdwjidx0H,
1015 /* Calculate table index by multiplying r with table scale and truncate to integer */
1016 rt = _mm256_mul_ps(r00,vftabscale);
1017 vfitab = _mm256_cvttps_epi32(rt);
1018 vfeps = _mm256_sub_ps(rt,_mm256_round_ps(rt, _MM_FROUND_FLOOR));
1019 /* AVX1 does not support 256-bit integer operations, so now we go to 128-bit mode... */
1020 vfitab_lo = _mm256_extractf128_si256(vfitab,0x0);
1021 vfitab_hi = _mm256_extractf128_si256(vfitab,0x1);
1022 vfitab_lo = _mm_slli_epi32(vfitab_lo,3);
1023 vfitab_hi = _mm_slli_epi32(vfitab_hi,3);
1025 /* CUBIC SPLINE TABLE DISPERSION */
1026 Y = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
1027 _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
1028 F = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
1029 _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
1030 G = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
1031 _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
1032 H = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
1033 _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
1034 GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
1035 Heps = _mm256_mul_ps(vfeps,H);
1036 Fp = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
1037 FF = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
1038 fvdw6 = _mm256_mul_ps(c6_00,FF);
1040 /* CUBIC SPLINE TABLE REPULSION */
1041 vfitab_lo = _mm_add_epi32(vfitab_lo,ifour);
1042 vfitab_hi = _mm_add_epi32(vfitab_hi,ifour);
1043 Y = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
1044 _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
1045 F = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
1046 _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
1047 G = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
1048 _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
1049 H = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
1050 _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
1051 GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
1052 Heps = _mm256_mul_ps(vfeps,H);
1053 Fp = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
1054 FF = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
1055 fvdw12 = _mm256_mul_ps(c12_00,FF);
1056 fvdw = _mm256_xor_ps(signbit,_mm256_mul_ps(_mm256_add_ps(fvdw6,fvdw12),_mm256_mul_ps(vftabscale,rinv00)));
1060 /* Calculate temporary vectorial force */
1061 tx = _mm256_mul_ps(fscal,dx00);
1062 ty = _mm256_mul_ps(fscal,dy00);
1063 tz = _mm256_mul_ps(fscal,dz00);
1065 /* Update vectorial force */
1066 fix0 = _mm256_add_ps(fix0,tx);
1067 fiy0 = _mm256_add_ps(fiy0,ty);
1068 fiz0 = _mm256_add_ps(fiz0,tz);
1070 fjx0 = _mm256_add_ps(fjx0,tx);
1071 fjy0 = _mm256_add_ps(fjy0,ty);
1072 fjz0 = _mm256_add_ps(fjz0,tz);
1074 /**************************
1075 * CALCULATE INTERACTIONS *
1076 **************************/
1078 /* Compute parameters for interactions between i and j atoms */
1079 qq10 = _mm256_mul_ps(iq1,jq0);
1081 /* COULOMB ELECTROSTATICS */
1082 velec = _mm256_mul_ps(qq10,rinv10);
1083 felec = _mm256_mul_ps(velec,rinvsq10);
1087 /* Calculate temporary vectorial force */
1088 tx = _mm256_mul_ps(fscal,dx10);
1089 ty = _mm256_mul_ps(fscal,dy10);
1090 tz = _mm256_mul_ps(fscal,dz10);
1092 /* Update vectorial force */
1093 fix1 = _mm256_add_ps(fix1,tx);
1094 fiy1 = _mm256_add_ps(fiy1,ty);
1095 fiz1 = _mm256_add_ps(fiz1,tz);
1097 fjx0 = _mm256_add_ps(fjx0,tx);
1098 fjy0 = _mm256_add_ps(fjy0,ty);
1099 fjz0 = _mm256_add_ps(fjz0,tz);
1101 /**************************
1102 * CALCULATE INTERACTIONS *
1103 **************************/
1105 /* Compute parameters for interactions between i and j atoms */
1106 qq20 = _mm256_mul_ps(iq2,jq0);
1108 /* COULOMB ELECTROSTATICS */
1109 velec = _mm256_mul_ps(qq20,rinv20);
1110 felec = _mm256_mul_ps(velec,rinvsq20);
1114 /* Calculate temporary vectorial force */
1115 tx = _mm256_mul_ps(fscal,dx20);
1116 ty = _mm256_mul_ps(fscal,dy20);
1117 tz = _mm256_mul_ps(fscal,dz20);
1119 /* Update vectorial force */
1120 fix2 = _mm256_add_ps(fix2,tx);
1121 fiy2 = _mm256_add_ps(fiy2,ty);
1122 fiz2 = _mm256_add_ps(fiz2,tz);
1124 fjx0 = _mm256_add_ps(fjx0,tx);
1125 fjy0 = _mm256_add_ps(fjy0,ty);
1126 fjz0 = _mm256_add_ps(fjz0,tz);
1128 /**************************
1129 * CALCULATE INTERACTIONS *
1130 **************************/
1132 /* Compute parameters for interactions between i and j atoms */
1133 qq30 = _mm256_mul_ps(iq3,jq0);
1135 /* COULOMB ELECTROSTATICS */
1136 velec = _mm256_mul_ps(qq30,rinv30);
1137 felec = _mm256_mul_ps(velec,rinvsq30);
1141 /* Calculate temporary vectorial force */
1142 tx = _mm256_mul_ps(fscal,dx30);
1143 ty = _mm256_mul_ps(fscal,dy30);
1144 tz = _mm256_mul_ps(fscal,dz30);
1146 /* Update vectorial force */
1147 fix3 = _mm256_add_ps(fix3,tx);
1148 fiy3 = _mm256_add_ps(fiy3,ty);
1149 fiz3 = _mm256_add_ps(fiz3,tz);
1151 fjx0 = _mm256_add_ps(fjx0,tx);
1152 fjy0 = _mm256_add_ps(fjy0,ty);
1153 fjz0 = _mm256_add_ps(fjz0,tz);
1155 fjptrA = f+j_coord_offsetA;
1156 fjptrB = f+j_coord_offsetB;
1157 fjptrC = f+j_coord_offsetC;
1158 fjptrD = f+j_coord_offsetD;
1159 fjptrE = f+j_coord_offsetE;
1160 fjptrF = f+j_coord_offsetF;
1161 fjptrG = f+j_coord_offsetG;
1162 fjptrH = f+j_coord_offsetH;
1164 gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,fjx0,fjy0,fjz0);
1166 /* Inner loop uses 129 flops */
1169 if(jidx<j_index_end)
1172 /* Get j neighbor index, and coordinate index */
1173 jnrlistA = jjnr[jidx];
1174 jnrlistB = jjnr[jidx+1];
1175 jnrlistC = jjnr[jidx+2];
1176 jnrlistD = jjnr[jidx+3];
1177 jnrlistE = jjnr[jidx+4];
1178 jnrlistF = jjnr[jidx+5];
1179 jnrlistG = jjnr[jidx+6];
1180 jnrlistH = jjnr[jidx+7];
1181 /* Sign of each element will be negative for non-real atoms.
1182 * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
1183 * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
1185 dummy_mask = gmx_mm256_set_m128(gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx+4)),_mm_setzero_si128())),
1186 gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128())));
1188 jnrA = (jnrlistA>=0) ? jnrlistA : 0;
1189 jnrB = (jnrlistB>=0) ? jnrlistB : 0;
1190 jnrC = (jnrlistC>=0) ? jnrlistC : 0;
1191 jnrD = (jnrlistD>=0) ? jnrlistD : 0;
1192 jnrE = (jnrlistE>=0) ? jnrlistE : 0;
1193 jnrF = (jnrlistF>=0) ? jnrlistF : 0;
1194 jnrG = (jnrlistG>=0) ? jnrlistG : 0;
1195 jnrH = (jnrlistH>=0) ? jnrlistH : 0;
1196 j_coord_offsetA = DIM*jnrA;
1197 j_coord_offsetB = DIM*jnrB;
1198 j_coord_offsetC = DIM*jnrC;
1199 j_coord_offsetD = DIM*jnrD;
1200 j_coord_offsetE = DIM*jnrE;
1201 j_coord_offsetF = DIM*jnrF;
1202 j_coord_offsetG = DIM*jnrG;
1203 j_coord_offsetH = DIM*jnrH;
1205 /* load j atom coordinates */
1206 gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
1207 x+j_coord_offsetC,x+j_coord_offsetD,
1208 x+j_coord_offsetE,x+j_coord_offsetF,
1209 x+j_coord_offsetG,x+j_coord_offsetH,
1212 /* Calculate displacement vector */
1213 dx00 = _mm256_sub_ps(ix0,jx0);
1214 dy00 = _mm256_sub_ps(iy0,jy0);
1215 dz00 = _mm256_sub_ps(iz0,jz0);
1216 dx10 = _mm256_sub_ps(ix1,jx0);
1217 dy10 = _mm256_sub_ps(iy1,jy0);
1218 dz10 = _mm256_sub_ps(iz1,jz0);
1219 dx20 = _mm256_sub_ps(ix2,jx0);
1220 dy20 = _mm256_sub_ps(iy2,jy0);
1221 dz20 = _mm256_sub_ps(iz2,jz0);
1222 dx30 = _mm256_sub_ps(ix3,jx0);
1223 dy30 = _mm256_sub_ps(iy3,jy0);
1224 dz30 = _mm256_sub_ps(iz3,jz0);
1226 /* Calculate squared distance and things based on it */
1227 rsq00 = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
1228 rsq10 = gmx_mm256_calc_rsq_ps(dx10,dy10,dz10);
1229 rsq20 = gmx_mm256_calc_rsq_ps(dx20,dy20,dz20);
1230 rsq30 = gmx_mm256_calc_rsq_ps(dx30,dy30,dz30);
1232 rinv00 = gmx_mm256_invsqrt_ps(rsq00);
1233 rinv10 = gmx_mm256_invsqrt_ps(rsq10);
1234 rinv20 = gmx_mm256_invsqrt_ps(rsq20);
1235 rinv30 = gmx_mm256_invsqrt_ps(rsq30);
1237 rinvsq10 = _mm256_mul_ps(rinv10,rinv10);
1238 rinvsq20 = _mm256_mul_ps(rinv20,rinv20);
1239 rinvsq30 = _mm256_mul_ps(rinv30,rinv30);
1241 /* Load parameters for j particles */
1242 jq0 = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
1243 charge+jnrC+0,charge+jnrD+0,
1244 charge+jnrE+0,charge+jnrF+0,
1245 charge+jnrG+0,charge+jnrH+0);
1246 vdwjidx0A = 2*vdwtype[jnrA+0];
1247 vdwjidx0B = 2*vdwtype[jnrB+0];
1248 vdwjidx0C = 2*vdwtype[jnrC+0];
1249 vdwjidx0D = 2*vdwtype[jnrD+0];
1250 vdwjidx0E = 2*vdwtype[jnrE+0];
1251 vdwjidx0F = 2*vdwtype[jnrF+0];
1252 vdwjidx0G = 2*vdwtype[jnrG+0];
1253 vdwjidx0H = 2*vdwtype[jnrH+0];
1255 fjx0 = _mm256_setzero_ps();
1256 fjy0 = _mm256_setzero_ps();
1257 fjz0 = _mm256_setzero_ps();
1259 /**************************
1260 * CALCULATE INTERACTIONS *
1261 **************************/
1263 r00 = _mm256_mul_ps(rsq00,rinv00);
1264 r00 = _mm256_andnot_ps(dummy_mask,r00);
1266 /* Compute parameters for interactions between i and j atoms */
1267 gmx_mm256_load_8pair_swizzle_ps(vdwioffsetptr0+vdwjidx0A,
1268 vdwioffsetptr0+vdwjidx0B,
1269 vdwioffsetptr0+vdwjidx0C,
1270 vdwioffsetptr0+vdwjidx0D,
1271 vdwioffsetptr0+vdwjidx0E,
1272 vdwioffsetptr0+vdwjidx0F,
1273 vdwioffsetptr0+vdwjidx0G,
1274 vdwioffsetptr0+vdwjidx0H,
1277 /* Calculate table index by multiplying r with table scale and truncate to integer */
1278 rt = _mm256_mul_ps(r00,vftabscale);
1279 vfitab = _mm256_cvttps_epi32(rt);
1280 vfeps = _mm256_sub_ps(rt,_mm256_round_ps(rt, _MM_FROUND_FLOOR));
1281 /* AVX1 does not support 256-bit integer operations, so now we go to 128-bit mode... */
1282 vfitab_lo = _mm256_extractf128_si256(vfitab,0x0);
1283 vfitab_hi = _mm256_extractf128_si256(vfitab,0x1);
1284 vfitab_lo = _mm_slli_epi32(vfitab_lo,3);
1285 vfitab_hi = _mm_slli_epi32(vfitab_hi,3);
1287 /* CUBIC SPLINE TABLE DISPERSION */
1288 Y = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
1289 _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
1290 F = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
1291 _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
1292 G = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
1293 _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
1294 H = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
1295 _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
1296 GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
1297 Heps = _mm256_mul_ps(vfeps,H);
1298 Fp = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
1299 FF = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
1300 fvdw6 = _mm256_mul_ps(c6_00,FF);
1302 /* CUBIC SPLINE TABLE REPULSION */
1303 vfitab_lo = _mm_add_epi32(vfitab_lo,ifour);
1304 vfitab_hi = _mm_add_epi32(vfitab_hi,ifour);
1305 Y = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
1306 _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
1307 F = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
1308 _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
1309 G = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
1310 _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
1311 H = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
1312 _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
1313 GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
1314 Heps = _mm256_mul_ps(vfeps,H);
1315 Fp = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
1316 FF = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
1317 fvdw12 = _mm256_mul_ps(c12_00,FF);
1318 fvdw = _mm256_xor_ps(signbit,_mm256_mul_ps(_mm256_add_ps(fvdw6,fvdw12),_mm256_mul_ps(vftabscale,rinv00)));
1322 fscal = _mm256_andnot_ps(dummy_mask,fscal);
1324 /* Calculate temporary vectorial force */
1325 tx = _mm256_mul_ps(fscal,dx00);
1326 ty = _mm256_mul_ps(fscal,dy00);
1327 tz = _mm256_mul_ps(fscal,dz00);
1329 /* Update vectorial force */
1330 fix0 = _mm256_add_ps(fix0,tx);
1331 fiy0 = _mm256_add_ps(fiy0,ty);
1332 fiz0 = _mm256_add_ps(fiz0,tz);
1334 fjx0 = _mm256_add_ps(fjx0,tx);
1335 fjy0 = _mm256_add_ps(fjy0,ty);
1336 fjz0 = _mm256_add_ps(fjz0,tz);
1338 /**************************
1339 * CALCULATE INTERACTIONS *
1340 **************************/
1342 /* Compute parameters for interactions between i and j atoms */
1343 qq10 = _mm256_mul_ps(iq1,jq0);
1345 /* COULOMB ELECTROSTATICS */
1346 velec = _mm256_mul_ps(qq10,rinv10);
1347 felec = _mm256_mul_ps(velec,rinvsq10);
1351 fscal = _mm256_andnot_ps(dummy_mask,fscal);
1353 /* Calculate temporary vectorial force */
1354 tx = _mm256_mul_ps(fscal,dx10);
1355 ty = _mm256_mul_ps(fscal,dy10);
1356 tz = _mm256_mul_ps(fscal,dz10);
1358 /* Update vectorial force */
1359 fix1 = _mm256_add_ps(fix1,tx);
1360 fiy1 = _mm256_add_ps(fiy1,ty);
1361 fiz1 = _mm256_add_ps(fiz1,tz);
1363 fjx0 = _mm256_add_ps(fjx0,tx);
1364 fjy0 = _mm256_add_ps(fjy0,ty);
1365 fjz0 = _mm256_add_ps(fjz0,tz);
1367 /**************************
1368 * CALCULATE INTERACTIONS *
1369 **************************/
1371 /* Compute parameters for interactions between i and j atoms */
1372 qq20 = _mm256_mul_ps(iq2,jq0);
1374 /* COULOMB ELECTROSTATICS */
1375 velec = _mm256_mul_ps(qq20,rinv20);
1376 felec = _mm256_mul_ps(velec,rinvsq20);
1380 fscal = _mm256_andnot_ps(dummy_mask,fscal);
1382 /* Calculate temporary vectorial force */
1383 tx = _mm256_mul_ps(fscal,dx20);
1384 ty = _mm256_mul_ps(fscal,dy20);
1385 tz = _mm256_mul_ps(fscal,dz20);
1387 /* Update vectorial force */
1388 fix2 = _mm256_add_ps(fix2,tx);
1389 fiy2 = _mm256_add_ps(fiy2,ty);
1390 fiz2 = _mm256_add_ps(fiz2,tz);
1392 fjx0 = _mm256_add_ps(fjx0,tx);
1393 fjy0 = _mm256_add_ps(fjy0,ty);
1394 fjz0 = _mm256_add_ps(fjz0,tz);
1396 /**************************
1397 * CALCULATE INTERACTIONS *
1398 **************************/
1400 /* Compute parameters for interactions between i and j atoms */
1401 qq30 = _mm256_mul_ps(iq3,jq0);
1403 /* COULOMB ELECTROSTATICS */
1404 velec = _mm256_mul_ps(qq30,rinv30);
1405 felec = _mm256_mul_ps(velec,rinvsq30);
1409 fscal = _mm256_andnot_ps(dummy_mask,fscal);
1411 /* Calculate temporary vectorial force */
1412 tx = _mm256_mul_ps(fscal,dx30);
1413 ty = _mm256_mul_ps(fscal,dy30);
1414 tz = _mm256_mul_ps(fscal,dz30);
1416 /* Update vectorial force */
1417 fix3 = _mm256_add_ps(fix3,tx);
1418 fiy3 = _mm256_add_ps(fiy3,ty);
1419 fiz3 = _mm256_add_ps(fiz3,tz);
1421 fjx0 = _mm256_add_ps(fjx0,tx);
1422 fjy0 = _mm256_add_ps(fjy0,ty);
1423 fjz0 = _mm256_add_ps(fjz0,tz);
1425 fjptrA = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
1426 fjptrB = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
1427 fjptrC = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
1428 fjptrD = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
1429 fjptrE = (jnrlistE>=0) ? f+j_coord_offsetE : scratch;
1430 fjptrF = (jnrlistF>=0) ? f+j_coord_offsetF : scratch;
1431 fjptrG = (jnrlistG>=0) ? f+j_coord_offsetG : scratch;
1432 fjptrH = (jnrlistH>=0) ? f+j_coord_offsetH : scratch;
1434 gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,fjx0,fjy0,fjz0);
1436 /* Inner loop uses 130 flops */
1439 /* End of innermost loop */
1441 gmx_mm256_update_iforce_4atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
1442 f+i_coord_offset,fshift+i_shift_offset);
1444 /* Increment number of inner iterations */
1445 inneriter += j_index_end - j_index_start;
1447 /* Outer loop uses 24 flops */
1450 /* Increment number of outer iterations */
1453 /* Update outer/inner flops */
1455 inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W4_F,outeriter*24 + inneriter*130);