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_ElecRFCut_VdwCSTab_GeomW4P1_VF_avx_256_single
54 * Electrostatics interaction: ReactionField
55 * VdW interaction: CubicSplineTable
56 * Geometry: Water4-Particle
57 * Calculate force/pot: PotentialAndForce
60 nb_kernel_ElecRFCut_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 krf = _mm256_set1_ps(fr->ic->k_rf);
133 krf2 = _mm256_set1_ps(fr->ic->k_rf*2.0);
134 crf = _mm256_set1_ps(fr->ic->c_rf);
135 nvdwtype = fr->ntype;
137 vdwtype = mdatoms->typeA;
139 vftab = kernel_data->table_vdw->data;
140 vftabscale = _mm256_set1_ps(kernel_data->table_vdw->scale);
142 /* Setup water-specific parameters */
143 inr = nlist->iinr[0];
144 iq1 = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+1]));
145 iq2 = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+2]));
146 iq3 = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+3]));
147 vdwioffsetptr0 = vdwparam+2*nvdwtype*vdwtype[inr+0];
149 /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
150 rcutoff_scalar = fr->rcoulomb;
151 rcutoff = _mm256_set1_ps(rcutoff_scalar);
152 rcutoff2 = _mm256_mul_ps(rcutoff,rcutoff);
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_4rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
189 &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
191 fix0 = _mm256_setzero_ps();
192 fiy0 = _mm256_setzero_ps();
193 fiz0 = _mm256_setzero_ps();
194 fix1 = _mm256_setzero_ps();
195 fiy1 = _mm256_setzero_ps();
196 fiz1 = _mm256_setzero_ps();
197 fix2 = _mm256_setzero_ps();
198 fiy2 = _mm256_setzero_ps();
199 fiz2 = _mm256_setzero_ps();
200 fix3 = _mm256_setzero_ps();
201 fiy3 = _mm256_setzero_ps();
202 fiz3 = _mm256_setzero_ps();
204 /* Reset potential sums */
205 velecsum = _mm256_setzero_ps();
206 vvdwsum = _mm256_setzero_ps();
208 /* Start inner kernel loop */
209 for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+7]>=0; jidx+=8)
212 /* Get j neighbor index, and coordinate index */
221 j_coord_offsetA = DIM*jnrA;
222 j_coord_offsetB = DIM*jnrB;
223 j_coord_offsetC = DIM*jnrC;
224 j_coord_offsetD = DIM*jnrD;
225 j_coord_offsetE = DIM*jnrE;
226 j_coord_offsetF = DIM*jnrF;
227 j_coord_offsetG = DIM*jnrG;
228 j_coord_offsetH = DIM*jnrH;
230 /* load j atom coordinates */
231 gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
232 x+j_coord_offsetC,x+j_coord_offsetD,
233 x+j_coord_offsetE,x+j_coord_offsetF,
234 x+j_coord_offsetG,x+j_coord_offsetH,
237 /* Calculate displacement vector */
238 dx00 = _mm256_sub_ps(ix0,jx0);
239 dy00 = _mm256_sub_ps(iy0,jy0);
240 dz00 = _mm256_sub_ps(iz0,jz0);
241 dx10 = _mm256_sub_ps(ix1,jx0);
242 dy10 = _mm256_sub_ps(iy1,jy0);
243 dz10 = _mm256_sub_ps(iz1,jz0);
244 dx20 = _mm256_sub_ps(ix2,jx0);
245 dy20 = _mm256_sub_ps(iy2,jy0);
246 dz20 = _mm256_sub_ps(iz2,jz0);
247 dx30 = _mm256_sub_ps(ix3,jx0);
248 dy30 = _mm256_sub_ps(iy3,jy0);
249 dz30 = _mm256_sub_ps(iz3,jz0);
251 /* Calculate squared distance and things based on it */
252 rsq00 = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
253 rsq10 = gmx_mm256_calc_rsq_ps(dx10,dy10,dz10);
254 rsq20 = gmx_mm256_calc_rsq_ps(dx20,dy20,dz20);
255 rsq30 = gmx_mm256_calc_rsq_ps(dx30,dy30,dz30);
257 rinv00 = gmx_mm256_invsqrt_ps(rsq00);
258 rinv10 = gmx_mm256_invsqrt_ps(rsq10);
259 rinv20 = gmx_mm256_invsqrt_ps(rsq20);
260 rinv30 = gmx_mm256_invsqrt_ps(rsq30);
262 rinvsq10 = _mm256_mul_ps(rinv10,rinv10);
263 rinvsq20 = _mm256_mul_ps(rinv20,rinv20);
264 rinvsq30 = _mm256_mul_ps(rinv30,rinv30);
266 /* Load parameters for j particles */
267 jq0 = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
268 charge+jnrC+0,charge+jnrD+0,
269 charge+jnrE+0,charge+jnrF+0,
270 charge+jnrG+0,charge+jnrH+0);
271 vdwjidx0A = 2*vdwtype[jnrA+0];
272 vdwjidx0B = 2*vdwtype[jnrB+0];
273 vdwjidx0C = 2*vdwtype[jnrC+0];
274 vdwjidx0D = 2*vdwtype[jnrD+0];
275 vdwjidx0E = 2*vdwtype[jnrE+0];
276 vdwjidx0F = 2*vdwtype[jnrF+0];
277 vdwjidx0G = 2*vdwtype[jnrG+0];
278 vdwjidx0H = 2*vdwtype[jnrH+0];
280 fjx0 = _mm256_setzero_ps();
281 fjy0 = _mm256_setzero_ps();
282 fjz0 = _mm256_setzero_ps();
284 /**************************
285 * CALCULATE INTERACTIONS *
286 **************************/
288 r00 = _mm256_mul_ps(rsq00,rinv00);
290 /* Compute parameters for interactions between i and j atoms */
291 gmx_mm256_load_8pair_swizzle_ps(vdwioffsetptr0+vdwjidx0A,
292 vdwioffsetptr0+vdwjidx0B,
293 vdwioffsetptr0+vdwjidx0C,
294 vdwioffsetptr0+vdwjidx0D,
295 vdwioffsetptr0+vdwjidx0E,
296 vdwioffsetptr0+vdwjidx0F,
297 vdwioffsetptr0+vdwjidx0G,
298 vdwioffsetptr0+vdwjidx0H,
301 /* Calculate table index by multiplying r with table scale and truncate to integer */
302 rt = _mm256_mul_ps(r00,vftabscale);
303 vfitab = _mm256_cvttps_epi32(rt);
304 vfeps = _mm256_sub_ps(rt,_mm256_round_ps(rt, _MM_FROUND_FLOOR));
305 /* AVX1 does not support 256-bit integer operations, so now we go to 128-bit mode... */
306 vfitab_lo = _mm256_extractf128_si256(vfitab,0x0);
307 vfitab_hi = _mm256_extractf128_si256(vfitab,0x1);
308 vfitab_lo = _mm_slli_epi32(vfitab_lo,3);
309 vfitab_hi = _mm_slli_epi32(vfitab_hi,3);
311 /* CUBIC SPLINE TABLE DISPERSION */
312 Y = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
313 _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
314 F = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
315 _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
316 G = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
317 _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
318 H = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
319 _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
320 GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
321 Heps = _mm256_mul_ps(vfeps,H);
322 Fp = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
323 VV = _mm256_add_ps(Y,_mm256_mul_ps(vfeps,Fp));
324 vvdw6 = _mm256_mul_ps(c6_00,VV);
325 FF = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
326 fvdw6 = _mm256_mul_ps(c6_00,FF);
328 /* CUBIC SPLINE TABLE REPULSION */
329 vfitab_lo = _mm_add_epi32(vfitab_lo,ifour);
330 vfitab_hi = _mm_add_epi32(vfitab_hi,ifour);
331 Y = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
332 _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
333 F = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
334 _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
335 G = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
336 _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
337 H = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
338 _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
339 GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
340 Heps = _mm256_mul_ps(vfeps,H);
341 Fp = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
342 VV = _mm256_add_ps(Y,_mm256_mul_ps(vfeps,Fp));
343 vvdw12 = _mm256_mul_ps(c12_00,VV);
344 FF = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
345 fvdw12 = _mm256_mul_ps(c12_00,FF);
346 vvdw = _mm256_add_ps(vvdw12,vvdw6);
347 fvdw = _mm256_xor_ps(signbit,_mm256_mul_ps(_mm256_add_ps(fvdw6,fvdw12),_mm256_mul_ps(vftabscale,rinv00)));
349 /* Update potential sum for this i atom from the interaction with this j atom. */
350 vvdwsum = _mm256_add_ps(vvdwsum,vvdw);
354 /* Calculate temporary vectorial force */
355 tx = _mm256_mul_ps(fscal,dx00);
356 ty = _mm256_mul_ps(fscal,dy00);
357 tz = _mm256_mul_ps(fscal,dz00);
359 /* Update vectorial force */
360 fix0 = _mm256_add_ps(fix0,tx);
361 fiy0 = _mm256_add_ps(fiy0,ty);
362 fiz0 = _mm256_add_ps(fiz0,tz);
364 fjx0 = _mm256_add_ps(fjx0,tx);
365 fjy0 = _mm256_add_ps(fjy0,ty);
366 fjz0 = _mm256_add_ps(fjz0,tz);
368 /**************************
369 * CALCULATE INTERACTIONS *
370 **************************/
372 if (gmx_mm256_any_lt(rsq10,rcutoff2))
375 /* Compute parameters for interactions between i and j atoms */
376 qq10 = _mm256_mul_ps(iq1,jq0);
378 /* REACTION-FIELD ELECTROSTATICS */
379 velec = _mm256_mul_ps(qq10,_mm256_sub_ps(_mm256_add_ps(rinv10,_mm256_mul_ps(krf,rsq10)),crf));
380 felec = _mm256_mul_ps(qq10,_mm256_sub_ps(_mm256_mul_ps(rinv10,rinvsq10),krf2));
382 cutoff_mask = _mm256_cmp_ps(rsq10,rcutoff2,_CMP_LT_OQ);
384 /* Update potential sum for this i atom from the interaction with this j atom. */
385 velec = _mm256_and_ps(velec,cutoff_mask);
386 velecsum = _mm256_add_ps(velecsum,velec);
390 fscal = _mm256_and_ps(fscal,cutoff_mask);
392 /* Calculate temporary vectorial force */
393 tx = _mm256_mul_ps(fscal,dx10);
394 ty = _mm256_mul_ps(fscal,dy10);
395 tz = _mm256_mul_ps(fscal,dz10);
397 /* Update vectorial force */
398 fix1 = _mm256_add_ps(fix1,tx);
399 fiy1 = _mm256_add_ps(fiy1,ty);
400 fiz1 = _mm256_add_ps(fiz1,tz);
402 fjx0 = _mm256_add_ps(fjx0,tx);
403 fjy0 = _mm256_add_ps(fjy0,ty);
404 fjz0 = _mm256_add_ps(fjz0,tz);
408 /**************************
409 * CALCULATE INTERACTIONS *
410 **************************/
412 if (gmx_mm256_any_lt(rsq20,rcutoff2))
415 /* Compute parameters for interactions between i and j atoms */
416 qq20 = _mm256_mul_ps(iq2,jq0);
418 /* REACTION-FIELD ELECTROSTATICS */
419 velec = _mm256_mul_ps(qq20,_mm256_sub_ps(_mm256_add_ps(rinv20,_mm256_mul_ps(krf,rsq20)),crf));
420 felec = _mm256_mul_ps(qq20,_mm256_sub_ps(_mm256_mul_ps(rinv20,rinvsq20),krf2));
422 cutoff_mask = _mm256_cmp_ps(rsq20,rcutoff2,_CMP_LT_OQ);
424 /* Update potential sum for this i atom from the interaction with this j atom. */
425 velec = _mm256_and_ps(velec,cutoff_mask);
426 velecsum = _mm256_add_ps(velecsum,velec);
430 fscal = _mm256_and_ps(fscal,cutoff_mask);
432 /* Calculate temporary vectorial force */
433 tx = _mm256_mul_ps(fscal,dx20);
434 ty = _mm256_mul_ps(fscal,dy20);
435 tz = _mm256_mul_ps(fscal,dz20);
437 /* Update vectorial force */
438 fix2 = _mm256_add_ps(fix2,tx);
439 fiy2 = _mm256_add_ps(fiy2,ty);
440 fiz2 = _mm256_add_ps(fiz2,tz);
442 fjx0 = _mm256_add_ps(fjx0,tx);
443 fjy0 = _mm256_add_ps(fjy0,ty);
444 fjz0 = _mm256_add_ps(fjz0,tz);
448 /**************************
449 * CALCULATE INTERACTIONS *
450 **************************/
452 if (gmx_mm256_any_lt(rsq30,rcutoff2))
455 /* Compute parameters for interactions between i and j atoms */
456 qq30 = _mm256_mul_ps(iq3,jq0);
458 /* REACTION-FIELD ELECTROSTATICS */
459 velec = _mm256_mul_ps(qq30,_mm256_sub_ps(_mm256_add_ps(rinv30,_mm256_mul_ps(krf,rsq30)),crf));
460 felec = _mm256_mul_ps(qq30,_mm256_sub_ps(_mm256_mul_ps(rinv30,rinvsq30),krf2));
462 cutoff_mask = _mm256_cmp_ps(rsq30,rcutoff2,_CMP_LT_OQ);
464 /* Update potential sum for this i atom from the interaction with this j atom. */
465 velec = _mm256_and_ps(velec,cutoff_mask);
466 velecsum = _mm256_add_ps(velecsum,velec);
470 fscal = _mm256_and_ps(fscal,cutoff_mask);
472 /* Calculate temporary vectorial force */
473 tx = _mm256_mul_ps(fscal,dx30);
474 ty = _mm256_mul_ps(fscal,dy30);
475 tz = _mm256_mul_ps(fscal,dz30);
477 /* Update vectorial force */
478 fix3 = _mm256_add_ps(fix3,tx);
479 fiy3 = _mm256_add_ps(fiy3,ty);
480 fiz3 = _mm256_add_ps(fiz3,tz);
482 fjx0 = _mm256_add_ps(fjx0,tx);
483 fjy0 = _mm256_add_ps(fjy0,ty);
484 fjz0 = _mm256_add_ps(fjz0,tz);
488 fjptrA = f+j_coord_offsetA;
489 fjptrB = f+j_coord_offsetB;
490 fjptrC = f+j_coord_offsetC;
491 fjptrD = f+j_coord_offsetD;
492 fjptrE = f+j_coord_offsetE;
493 fjptrF = f+j_coord_offsetF;
494 fjptrG = f+j_coord_offsetG;
495 fjptrH = f+j_coord_offsetH;
497 gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,fjx0,fjy0,fjz0);
499 /* Inner loop uses 167 flops */
505 /* Get j neighbor index, and coordinate index */
506 jnrlistA = jjnr[jidx];
507 jnrlistB = jjnr[jidx+1];
508 jnrlistC = jjnr[jidx+2];
509 jnrlistD = jjnr[jidx+3];
510 jnrlistE = jjnr[jidx+4];
511 jnrlistF = jjnr[jidx+5];
512 jnrlistG = jjnr[jidx+6];
513 jnrlistH = jjnr[jidx+7];
514 /* Sign of each element will be negative for non-real atoms.
515 * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
516 * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
518 dummy_mask = gmx_mm256_set_m128(gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx+4)),_mm_setzero_si128())),
519 gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128())));
521 jnrA = (jnrlistA>=0) ? jnrlistA : 0;
522 jnrB = (jnrlistB>=0) ? jnrlistB : 0;
523 jnrC = (jnrlistC>=0) ? jnrlistC : 0;
524 jnrD = (jnrlistD>=0) ? jnrlistD : 0;
525 jnrE = (jnrlistE>=0) ? jnrlistE : 0;
526 jnrF = (jnrlistF>=0) ? jnrlistF : 0;
527 jnrG = (jnrlistG>=0) ? jnrlistG : 0;
528 jnrH = (jnrlistH>=0) ? jnrlistH : 0;
529 j_coord_offsetA = DIM*jnrA;
530 j_coord_offsetB = DIM*jnrB;
531 j_coord_offsetC = DIM*jnrC;
532 j_coord_offsetD = DIM*jnrD;
533 j_coord_offsetE = DIM*jnrE;
534 j_coord_offsetF = DIM*jnrF;
535 j_coord_offsetG = DIM*jnrG;
536 j_coord_offsetH = DIM*jnrH;
538 /* load j atom coordinates */
539 gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
540 x+j_coord_offsetC,x+j_coord_offsetD,
541 x+j_coord_offsetE,x+j_coord_offsetF,
542 x+j_coord_offsetG,x+j_coord_offsetH,
545 /* Calculate displacement vector */
546 dx00 = _mm256_sub_ps(ix0,jx0);
547 dy00 = _mm256_sub_ps(iy0,jy0);
548 dz00 = _mm256_sub_ps(iz0,jz0);
549 dx10 = _mm256_sub_ps(ix1,jx0);
550 dy10 = _mm256_sub_ps(iy1,jy0);
551 dz10 = _mm256_sub_ps(iz1,jz0);
552 dx20 = _mm256_sub_ps(ix2,jx0);
553 dy20 = _mm256_sub_ps(iy2,jy0);
554 dz20 = _mm256_sub_ps(iz2,jz0);
555 dx30 = _mm256_sub_ps(ix3,jx0);
556 dy30 = _mm256_sub_ps(iy3,jy0);
557 dz30 = _mm256_sub_ps(iz3,jz0);
559 /* Calculate squared distance and things based on it */
560 rsq00 = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
561 rsq10 = gmx_mm256_calc_rsq_ps(dx10,dy10,dz10);
562 rsq20 = gmx_mm256_calc_rsq_ps(dx20,dy20,dz20);
563 rsq30 = gmx_mm256_calc_rsq_ps(dx30,dy30,dz30);
565 rinv00 = gmx_mm256_invsqrt_ps(rsq00);
566 rinv10 = gmx_mm256_invsqrt_ps(rsq10);
567 rinv20 = gmx_mm256_invsqrt_ps(rsq20);
568 rinv30 = gmx_mm256_invsqrt_ps(rsq30);
570 rinvsq10 = _mm256_mul_ps(rinv10,rinv10);
571 rinvsq20 = _mm256_mul_ps(rinv20,rinv20);
572 rinvsq30 = _mm256_mul_ps(rinv30,rinv30);
574 /* Load parameters for j particles */
575 jq0 = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
576 charge+jnrC+0,charge+jnrD+0,
577 charge+jnrE+0,charge+jnrF+0,
578 charge+jnrG+0,charge+jnrH+0);
579 vdwjidx0A = 2*vdwtype[jnrA+0];
580 vdwjidx0B = 2*vdwtype[jnrB+0];
581 vdwjidx0C = 2*vdwtype[jnrC+0];
582 vdwjidx0D = 2*vdwtype[jnrD+0];
583 vdwjidx0E = 2*vdwtype[jnrE+0];
584 vdwjidx0F = 2*vdwtype[jnrF+0];
585 vdwjidx0G = 2*vdwtype[jnrG+0];
586 vdwjidx0H = 2*vdwtype[jnrH+0];
588 fjx0 = _mm256_setzero_ps();
589 fjy0 = _mm256_setzero_ps();
590 fjz0 = _mm256_setzero_ps();
592 /**************************
593 * CALCULATE INTERACTIONS *
594 **************************/
596 r00 = _mm256_mul_ps(rsq00,rinv00);
597 r00 = _mm256_andnot_ps(dummy_mask,r00);
599 /* Compute parameters for interactions between i and j atoms */
600 gmx_mm256_load_8pair_swizzle_ps(vdwioffsetptr0+vdwjidx0A,
601 vdwioffsetptr0+vdwjidx0B,
602 vdwioffsetptr0+vdwjidx0C,
603 vdwioffsetptr0+vdwjidx0D,
604 vdwioffsetptr0+vdwjidx0E,
605 vdwioffsetptr0+vdwjidx0F,
606 vdwioffsetptr0+vdwjidx0G,
607 vdwioffsetptr0+vdwjidx0H,
610 /* Calculate table index by multiplying r with table scale and truncate to integer */
611 rt = _mm256_mul_ps(r00,vftabscale);
612 vfitab = _mm256_cvttps_epi32(rt);
613 vfeps = _mm256_sub_ps(rt,_mm256_round_ps(rt, _MM_FROUND_FLOOR));
614 /* AVX1 does not support 256-bit integer operations, so now we go to 128-bit mode... */
615 vfitab_lo = _mm256_extractf128_si256(vfitab,0x0);
616 vfitab_hi = _mm256_extractf128_si256(vfitab,0x1);
617 vfitab_lo = _mm_slli_epi32(vfitab_lo,3);
618 vfitab_hi = _mm_slli_epi32(vfitab_hi,3);
620 /* CUBIC SPLINE TABLE DISPERSION */
621 Y = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
622 _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
623 F = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
624 _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
625 G = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
626 _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
627 H = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
628 _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
629 GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
630 Heps = _mm256_mul_ps(vfeps,H);
631 Fp = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
632 VV = _mm256_add_ps(Y,_mm256_mul_ps(vfeps,Fp));
633 vvdw6 = _mm256_mul_ps(c6_00,VV);
634 FF = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
635 fvdw6 = _mm256_mul_ps(c6_00,FF);
637 /* CUBIC SPLINE TABLE REPULSION */
638 vfitab_lo = _mm_add_epi32(vfitab_lo,ifour);
639 vfitab_hi = _mm_add_epi32(vfitab_hi,ifour);
640 Y = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
641 _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
642 F = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
643 _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
644 G = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
645 _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
646 H = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
647 _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
648 GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
649 Heps = _mm256_mul_ps(vfeps,H);
650 Fp = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
651 VV = _mm256_add_ps(Y,_mm256_mul_ps(vfeps,Fp));
652 vvdw12 = _mm256_mul_ps(c12_00,VV);
653 FF = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
654 fvdw12 = _mm256_mul_ps(c12_00,FF);
655 vvdw = _mm256_add_ps(vvdw12,vvdw6);
656 fvdw = _mm256_xor_ps(signbit,_mm256_mul_ps(_mm256_add_ps(fvdw6,fvdw12),_mm256_mul_ps(vftabscale,rinv00)));
658 /* Update potential sum for this i atom from the interaction with this j atom. */
659 vvdw = _mm256_andnot_ps(dummy_mask,vvdw);
660 vvdwsum = _mm256_add_ps(vvdwsum,vvdw);
664 fscal = _mm256_andnot_ps(dummy_mask,fscal);
666 /* Calculate temporary vectorial force */
667 tx = _mm256_mul_ps(fscal,dx00);
668 ty = _mm256_mul_ps(fscal,dy00);
669 tz = _mm256_mul_ps(fscal,dz00);
671 /* Update vectorial force */
672 fix0 = _mm256_add_ps(fix0,tx);
673 fiy0 = _mm256_add_ps(fiy0,ty);
674 fiz0 = _mm256_add_ps(fiz0,tz);
676 fjx0 = _mm256_add_ps(fjx0,tx);
677 fjy0 = _mm256_add_ps(fjy0,ty);
678 fjz0 = _mm256_add_ps(fjz0,tz);
680 /**************************
681 * CALCULATE INTERACTIONS *
682 **************************/
684 if (gmx_mm256_any_lt(rsq10,rcutoff2))
687 /* Compute parameters for interactions between i and j atoms */
688 qq10 = _mm256_mul_ps(iq1,jq0);
690 /* REACTION-FIELD ELECTROSTATICS */
691 velec = _mm256_mul_ps(qq10,_mm256_sub_ps(_mm256_add_ps(rinv10,_mm256_mul_ps(krf,rsq10)),crf));
692 felec = _mm256_mul_ps(qq10,_mm256_sub_ps(_mm256_mul_ps(rinv10,rinvsq10),krf2));
694 cutoff_mask = _mm256_cmp_ps(rsq10,rcutoff2,_CMP_LT_OQ);
696 /* Update potential sum for this i atom from the interaction with this j atom. */
697 velec = _mm256_and_ps(velec,cutoff_mask);
698 velec = _mm256_andnot_ps(dummy_mask,velec);
699 velecsum = _mm256_add_ps(velecsum,velec);
703 fscal = _mm256_and_ps(fscal,cutoff_mask);
705 fscal = _mm256_andnot_ps(dummy_mask,fscal);
707 /* Calculate temporary vectorial force */
708 tx = _mm256_mul_ps(fscal,dx10);
709 ty = _mm256_mul_ps(fscal,dy10);
710 tz = _mm256_mul_ps(fscal,dz10);
712 /* Update vectorial force */
713 fix1 = _mm256_add_ps(fix1,tx);
714 fiy1 = _mm256_add_ps(fiy1,ty);
715 fiz1 = _mm256_add_ps(fiz1,tz);
717 fjx0 = _mm256_add_ps(fjx0,tx);
718 fjy0 = _mm256_add_ps(fjy0,ty);
719 fjz0 = _mm256_add_ps(fjz0,tz);
723 /**************************
724 * CALCULATE INTERACTIONS *
725 **************************/
727 if (gmx_mm256_any_lt(rsq20,rcutoff2))
730 /* Compute parameters for interactions between i and j atoms */
731 qq20 = _mm256_mul_ps(iq2,jq0);
733 /* REACTION-FIELD ELECTROSTATICS */
734 velec = _mm256_mul_ps(qq20,_mm256_sub_ps(_mm256_add_ps(rinv20,_mm256_mul_ps(krf,rsq20)),crf));
735 felec = _mm256_mul_ps(qq20,_mm256_sub_ps(_mm256_mul_ps(rinv20,rinvsq20),krf2));
737 cutoff_mask = _mm256_cmp_ps(rsq20,rcutoff2,_CMP_LT_OQ);
739 /* Update potential sum for this i atom from the interaction with this j atom. */
740 velec = _mm256_and_ps(velec,cutoff_mask);
741 velec = _mm256_andnot_ps(dummy_mask,velec);
742 velecsum = _mm256_add_ps(velecsum,velec);
746 fscal = _mm256_and_ps(fscal,cutoff_mask);
748 fscal = _mm256_andnot_ps(dummy_mask,fscal);
750 /* Calculate temporary vectorial force */
751 tx = _mm256_mul_ps(fscal,dx20);
752 ty = _mm256_mul_ps(fscal,dy20);
753 tz = _mm256_mul_ps(fscal,dz20);
755 /* Update vectorial force */
756 fix2 = _mm256_add_ps(fix2,tx);
757 fiy2 = _mm256_add_ps(fiy2,ty);
758 fiz2 = _mm256_add_ps(fiz2,tz);
760 fjx0 = _mm256_add_ps(fjx0,tx);
761 fjy0 = _mm256_add_ps(fjy0,ty);
762 fjz0 = _mm256_add_ps(fjz0,tz);
766 /**************************
767 * CALCULATE INTERACTIONS *
768 **************************/
770 if (gmx_mm256_any_lt(rsq30,rcutoff2))
773 /* Compute parameters for interactions between i and j atoms */
774 qq30 = _mm256_mul_ps(iq3,jq0);
776 /* REACTION-FIELD ELECTROSTATICS */
777 velec = _mm256_mul_ps(qq30,_mm256_sub_ps(_mm256_add_ps(rinv30,_mm256_mul_ps(krf,rsq30)),crf));
778 felec = _mm256_mul_ps(qq30,_mm256_sub_ps(_mm256_mul_ps(rinv30,rinvsq30),krf2));
780 cutoff_mask = _mm256_cmp_ps(rsq30,rcutoff2,_CMP_LT_OQ);
782 /* Update potential sum for this i atom from the interaction with this j atom. */
783 velec = _mm256_and_ps(velec,cutoff_mask);
784 velec = _mm256_andnot_ps(dummy_mask,velec);
785 velecsum = _mm256_add_ps(velecsum,velec);
789 fscal = _mm256_and_ps(fscal,cutoff_mask);
791 fscal = _mm256_andnot_ps(dummy_mask,fscal);
793 /* Calculate temporary vectorial force */
794 tx = _mm256_mul_ps(fscal,dx30);
795 ty = _mm256_mul_ps(fscal,dy30);
796 tz = _mm256_mul_ps(fscal,dz30);
798 /* Update vectorial force */
799 fix3 = _mm256_add_ps(fix3,tx);
800 fiy3 = _mm256_add_ps(fiy3,ty);
801 fiz3 = _mm256_add_ps(fiz3,tz);
803 fjx0 = _mm256_add_ps(fjx0,tx);
804 fjy0 = _mm256_add_ps(fjy0,ty);
805 fjz0 = _mm256_add_ps(fjz0,tz);
809 fjptrA = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
810 fjptrB = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
811 fjptrC = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
812 fjptrD = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
813 fjptrE = (jnrlistE>=0) ? f+j_coord_offsetE : scratch;
814 fjptrF = (jnrlistF>=0) ? f+j_coord_offsetF : scratch;
815 fjptrG = (jnrlistG>=0) ? f+j_coord_offsetG : scratch;
816 fjptrH = (jnrlistH>=0) ? f+j_coord_offsetH : scratch;
818 gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,fjx0,fjy0,fjz0);
820 /* Inner loop uses 168 flops */
823 /* End of innermost loop */
825 gmx_mm256_update_iforce_4atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
826 f+i_coord_offset,fshift+i_shift_offset);
829 /* Update potential energies */
830 gmx_mm256_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid);
831 gmx_mm256_update_1pot_ps(vvdwsum,kernel_data->energygrp_vdw+ggid);
833 /* Increment number of inner iterations */
834 inneriter += j_index_end - j_index_start;
836 /* Outer loop uses 26 flops */
839 /* Increment number of outer iterations */
842 /* Update outer/inner flops */
844 inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W4_VF,outeriter*26 + inneriter*168);
847 * Gromacs nonbonded kernel: nb_kernel_ElecRFCut_VdwCSTab_GeomW4P1_F_avx_256_single
848 * Electrostatics interaction: ReactionField
849 * VdW interaction: CubicSplineTable
850 * Geometry: Water4-Particle
851 * Calculate force/pot: Force
854 nb_kernel_ElecRFCut_VdwCSTab_GeomW4P1_F_avx_256_single
855 (t_nblist * gmx_restrict nlist,
856 rvec * gmx_restrict xx,
857 rvec * gmx_restrict ff,
858 t_forcerec * gmx_restrict fr,
859 t_mdatoms * gmx_restrict mdatoms,
860 nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
861 t_nrnb * gmx_restrict nrnb)
863 /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
864 * just 0 for non-waters.
865 * Suffixes A,B,C,D,E,F,G,H refer to j loop unrolling done with AVX, e.g. for the eight different
866 * jnr indices corresponding to data put in the four positions in the SIMD register.
868 int i_shift_offset,i_coord_offset,outeriter,inneriter;
869 int j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
870 int jnrA,jnrB,jnrC,jnrD;
871 int jnrE,jnrF,jnrG,jnrH;
872 int jnrlistA,jnrlistB,jnrlistC,jnrlistD;
873 int jnrlistE,jnrlistF,jnrlistG,jnrlistH;
874 int j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
875 int j_coord_offsetE,j_coord_offsetF,j_coord_offsetG,j_coord_offsetH;
876 int *iinr,*jindex,*jjnr,*shiftidx,*gid;
878 real *shiftvec,*fshift,*x,*f;
879 real *fjptrA,*fjptrB,*fjptrC,*fjptrD,*fjptrE,*fjptrF,*fjptrG,*fjptrH;
881 __m256 tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
882 real * vdwioffsetptr0;
883 __m256 ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
884 real * vdwioffsetptr1;
885 __m256 ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
886 real * vdwioffsetptr2;
887 __m256 ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
888 real * vdwioffsetptr3;
889 __m256 ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
890 int vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D,vdwjidx0E,vdwjidx0F,vdwjidx0G,vdwjidx0H;
891 __m256 jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
892 __m256 dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
893 __m256 dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
894 __m256 dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
895 __m256 dx30,dy30,dz30,rsq30,rinv30,rinvsq30,r30,qq30,c6_30,c12_30;
896 __m256 velec,felec,velecsum,facel,crf,krf,krf2;
899 __m256 rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
902 __m256 one_sixth = _mm256_set1_ps(1.0/6.0);
903 __m256 one_twelfth = _mm256_set1_ps(1.0/12.0);
905 __m128i vfitab_lo,vfitab_hi;
906 __m128i ifour = _mm_set1_epi32(4);
907 __m256 rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF;
909 __m256 dummy_mask,cutoff_mask;
910 __m256 signbit = _mm256_castsi256_ps( _mm256_set1_epi32(0x80000000) );
911 __m256 one = _mm256_set1_ps(1.0);
912 __m256 two = _mm256_set1_ps(2.0);
918 jindex = nlist->jindex;
920 shiftidx = nlist->shift;
922 shiftvec = fr->shift_vec[0];
923 fshift = fr->fshift[0];
924 facel = _mm256_set1_ps(fr->epsfac);
925 charge = mdatoms->chargeA;
926 krf = _mm256_set1_ps(fr->ic->k_rf);
927 krf2 = _mm256_set1_ps(fr->ic->k_rf*2.0);
928 crf = _mm256_set1_ps(fr->ic->c_rf);
929 nvdwtype = fr->ntype;
931 vdwtype = mdatoms->typeA;
933 vftab = kernel_data->table_vdw->data;
934 vftabscale = _mm256_set1_ps(kernel_data->table_vdw->scale);
936 /* Setup water-specific parameters */
937 inr = nlist->iinr[0];
938 iq1 = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+1]));
939 iq2 = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+2]));
940 iq3 = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+3]));
941 vdwioffsetptr0 = vdwparam+2*nvdwtype*vdwtype[inr+0];
943 /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
944 rcutoff_scalar = fr->rcoulomb;
945 rcutoff = _mm256_set1_ps(rcutoff_scalar);
946 rcutoff2 = _mm256_mul_ps(rcutoff,rcutoff);
948 /* Avoid stupid compiler warnings */
949 jnrA = jnrB = jnrC = jnrD = jnrE = jnrF = jnrG = jnrH = 0;
962 for(iidx=0;iidx<4*DIM;iidx++)
967 /* Start outer loop over neighborlists */
968 for(iidx=0; iidx<nri; iidx++)
970 /* Load shift vector for this list */
971 i_shift_offset = DIM*shiftidx[iidx];
973 /* Load limits for loop over neighbors */
974 j_index_start = jindex[iidx];
975 j_index_end = jindex[iidx+1];
977 /* Get outer coordinate index */
979 i_coord_offset = DIM*inr;
981 /* Load i particle coords and add shift vector */
982 gmx_mm256_load_shift_and_4rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
983 &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
985 fix0 = _mm256_setzero_ps();
986 fiy0 = _mm256_setzero_ps();
987 fiz0 = _mm256_setzero_ps();
988 fix1 = _mm256_setzero_ps();
989 fiy1 = _mm256_setzero_ps();
990 fiz1 = _mm256_setzero_ps();
991 fix2 = _mm256_setzero_ps();
992 fiy2 = _mm256_setzero_ps();
993 fiz2 = _mm256_setzero_ps();
994 fix3 = _mm256_setzero_ps();
995 fiy3 = _mm256_setzero_ps();
996 fiz3 = _mm256_setzero_ps();
998 /* Start inner kernel loop */
999 for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+7]>=0; jidx+=8)
1002 /* Get j neighbor index, and coordinate index */
1004 jnrB = jjnr[jidx+1];
1005 jnrC = jjnr[jidx+2];
1006 jnrD = jjnr[jidx+3];
1007 jnrE = jjnr[jidx+4];
1008 jnrF = jjnr[jidx+5];
1009 jnrG = jjnr[jidx+6];
1010 jnrH = jjnr[jidx+7];
1011 j_coord_offsetA = DIM*jnrA;
1012 j_coord_offsetB = DIM*jnrB;
1013 j_coord_offsetC = DIM*jnrC;
1014 j_coord_offsetD = DIM*jnrD;
1015 j_coord_offsetE = DIM*jnrE;
1016 j_coord_offsetF = DIM*jnrF;
1017 j_coord_offsetG = DIM*jnrG;
1018 j_coord_offsetH = DIM*jnrH;
1020 /* load j atom coordinates */
1021 gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
1022 x+j_coord_offsetC,x+j_coord_offsetD,
1023 x+j_coord_offsetE,x+j_coord_offsetF,
1024 x+j_coord_offsetG,x+j_coord_offsetH,
1027 /* Calculate displacement vector */
1028 dx00 = _mm256_sub_ps(ix0,jx0);
1029 dy00 = _mm256_sub_ps(iy0,jy0);
1030 dz00 = _mm256_sub_ps(iz0,jz0);
1031 dx10 = _mm256_sub_ps(ix1,jx0);
1032 dy10 = _mm256_sub_ps(iy1,jy0);
1033 dz10 = _mm256_sub_ps(iz1,jz0);
1034 dx20 = _mm256_sub_ps(ix2,jx0);
1035 dy20 = _mm256_sub_ps(iy2,jy0);
1036 dz20 = _mm256_sub_ps(iz2,jz0);
1037 dx30 = _mm256_sub_ps(ix3,jx0);
1038 dy30 = _mm256_sub_ps(iy3,jy0);
1039 dz30 = _mm256_sub_ps(iz3,jz0);
1041 /* Calculate squared distance and things based on it */
1042 rsq00 = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
1043 rsq10 = gmx_mm256_calc_rsq_ps(dx10,dy10,dz10);
1044 rsq20 = gmx_mm256_calc_rsq_ps(dx20,dy20,dz20);
1045 rsq30 = gmx_mm256_calc_rsq_ps(dx30,dy30,dz30);
1047 rinv00 = gmx_mm256_invsqrt_ps(rsq00);
1048 rinv10 = gmx_mm256_invsqrt_ps(rsq10);
1049 rinv20 = gmx_mm256_invsqrt_ps(rsq20);
1050 rinv30 = gmx_mm256_invsqrt_ps(rsq30);
1052 rinvsq10 = _mm256_mul_ps(rinv10,rinv10);
1053 rinvsq20 = _mm256_mul_ps(rinv20,rinv20);
1054 rinvsq30 = _mm256_mul_ps(rinv30,rinv30);
1056 /* Load parameters for j particles */
1057 jq0 = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
1058 charge+jnrC+0,charge+jnrD+0,
1059 charge+jnrE+0,charge+jnrF+0,
1060 charge+jnrG+0,charge+jnrH+0);
1061 vdwjidx0A = 2*vdwtype[jnrA+0];
1062 vdwjidx0B = 2*vdwtype[jnrB+0];
1063 vdwjidx0C = 2*vdwtype[jnrC+0];
1064 vdwjidx0D = 2*vdwtype[jnrD+0];
1065 vdwjidx0E = 2*vdwtype[jnrE+0];
1066 vdwjidx0F = 2*vdwtype[jnrF+0];
1067 vdwjidx0G = 2*vdwtype[jnrG+0];
1068 vdwjidx0H = 2*vdwtype[jnrH+0];
1070 fjx0 = _mm256_setzero_ps();
1071 fjy0 = _mm256_setzero_ps();
1072 fjz0 = _mm256_setzero_ps();
1074 /**************************
1075 * CALCULATE INTERACTIONS *
1076 **************************/
1078 r00 = _mm256_mul_ps(rsq00,rinv00);
1080 /* Compute parameters for interactions between i and j atoms */
1081 gmx_mm256_load_8pair_swizzle_ps(vdwioffsetptr0+vdwjidx0A,
1082 vdwioffsetptr0+vdwjidx0B,
1083 vdwioffsetptr0+vdwjidx0C,
1084 vdwioffsetptr0+vdwjidx0D,
1085 vdwioffsetptr0+vdwjidx0E,
1086 vdwioffsetptr0+vdwjidx0F,
1087 vdwioffsetptr0+vdwjidx0G,
1088 vdwioffsetptr0+vdwjidx0H,
1091 /* Calculate table index by multiplying r with table scale and truncate to integer */
1092 rt = _mm256_mul_ps(r00,vftabscale);
1093 vfitab = _mm256_cvttps_epi32(rt);
1094 vfeps = _mm256_sub_ps(rt,_mm256_round_ps(rt, _MM_FROUND_FLOOR));
1095 /* AVX1 does not support 256-bit integer operations, so now we go to 128-bit mode... */
1096 vfitab_lo = _mm256_extractf128_si256(vfitab,0x0);
1097 vfitab_hi = _mm256_extractf128_si256(vfitab,0x1);
1098 vfitab_lo = _mm_slli_epi32(vfitab_lo,3);
1099 vfitab_hi = _mm_slli_epi32(vfitab_hi,3);
1101 /* CUBIC SPLINE TABLE DISPERSION */
1102 Y = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
1103 _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
1104 F = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
1105 _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
1106 G = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
1107 _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
1108 H = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
1109 _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
1110 GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
1111 Heps = _mm256_mul_ps(vfeps,H);
1112 Fp = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
1113 FF = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
1114 fvdw6 = _mm256_mul_ps(c6_00,FF);
1116 /* CUBIC SPLINE TABLE REPULSION */
1117 vfitab_lo = _mm_add_epi32(vfitab_lo,ifour);
1118 vfitab_hi = _mm_add_epi32(vfitab_hi,ifour);
1119 Y = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
1120 _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
1121 F = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
1122 _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
1123 G = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
1124 _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
1125 H = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
1126 _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
1127 GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
1128 Heps = _mm256_mul_ps(vfeps,H);
1129 Fp = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
1130 FF = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
1131 fvdw12 = _mm256_mul_ps(c12_00,FF);
1132 fvdw = _mm256_xor_ps(signbit,_mm256_mul_ps(_mm256_add_ps(fvdw6,fvdw12),_mm256_mul_ps(vftabscale,rinv00)));
1136 /* Calculate temporary vectorial force */
1137 tx = _mm256_mul_ps(fscal,dx00);
1138 ty = _mm256_mul_ps(fscal,dy00);
1139 tz = _mm256_mul_ps(fscal,dz00);
1141 /* Update vectorial force */
1142 fix0 = _mm256_add_ps(fix0,tx);
1143 fiy0 = _mm256_add_ps(fiy0,ty);
1144 fiz0 = _mm256_add_ps(fiz0,tz);
1146 fjx0 = _mm256_add_ps(fjx0,tx);
1147 fjy0 = _mm256_add_ps(fjy0,ty);
1148 fjz0 = _mm256_add_ps(fjz0,tz);
1150 /**************************
1151 * CALCULATE INTERACTIONS *
1152 **************************/
1154 if (gmx_mm256_any_lt(rsq10,rcutoff2))
1157 /* Compute parameters for interactions between i and j atoms */
1158 qq10 = _mm256_mul_ps(iq1,jq0);
1160 /* REACTION-FIELD ELECTROSTATICS */
1161 felec = _mm256_mul_ps(qq10,_mm256_sub_ps(_mm256_mul_ps(rinv10,rinvsq10),krf2));
1163 cutoff_mask = _mm256_cmp_ps(rsq10,rcutoff2,_CMP_LT_OQ);
1167 fscal = _mm256_and_ps(fscal,cutoff_mask);
1169 /* Calculate temporary vectorial force */
1170 tx = _mm256_mul_ps(fscal,dx10);
1171 ty = _mm256_mul_ps(fscal,dy10);
1172 tz = _mm256_mul_ps(fscal,dz10);
1174 /* Update vectorial force */
1175 fix1 = _mm256_add_ps(fix1,tx);
1176 fiy1 = _mm256_add_ps(fiy1,ty);
1177 fiz1 = _mm256_add_ps(fiz1,tz);
1179 fjx0 = _mm256_add_ps(fjx0,tx);
1180 fjy0 = _mm256_add_ps(fjy0,ty);
1181 fjz0 = _mm256_add_ps(fjz0,tz);
1185 /**************************
1186 * CALCULATE INTERACTIONS *
1187 **************************/
1189 if (gmx_mm256_any_lt(rsq20,rcutoff2))
1192 /* Compute parameters for interactions between i and j atoms */
1193 qq20 = _mm256_mul_ps(iq2,jq0);
1195 /* REACTION-FIELD ELECTROSTATICS */
1196 felec = _mm256_mul_ps(qq20,_mm256_sub_ps(_mm256_mul_ps(rinv20,rinvsq20),krf2));
1198 cutoff_mask = _mm256_cmp_ps(rsq20,rcutoff2,_CMP_LT_OQ);
1202 fscal = _mm256_and_ps(fscal,cutoff_mask);
1204 /* Calculate temporary vectorial force */
1205 tx = _mm256_mul_ps(fscal,dx20);
1206 ty = _mm256_mul_ps(fscal,dy20);
1207 tz = _mm256_mul_ps(fscal,dz20);
1209 /* Update vectorial force */
1210 fix2 = _mm256_add_ps(fix2,tx);
1211 fiy2 = _mm256_add_ps(fiy2,ty);
1212 fiz2 = _mm256_add_ps(fiz2,tz);
1214 fjx0 = _mm256_add_ps(fjx0,tx);
1215 fjy0 = _mm256_add_ps(fjy0,ty);
1216 fjz0 = _mm256_add_ps(fjz0,tz);
1220 /**************************
1221 * CALCULATE INTERACTIONS *
1222 **************************/
1224 if (gmx_mm256_any_lt(rsq30,rcutoff2))
1227 /* Compute parameters for interactions between i and j atoms */
1228 qq30 = _mm256_mul_ps(iq3,jq0);
1230 /* REACTION-FIELD ELECTROSTATICS */
1231 felec = _mm256_mul_ps(qq30,_mm256_sub_ps(_mm256_mul_ps(rinv30,rinvsq30),krf2));
1233 cutoff_mask = _mm256_cmp_ps(rsq30,rcutoff2,_CMP_LT_OQ);
1237 fscal = _mm256_and_ps(fscal,cutoff_mask);
1239 /* Calculate temporary vectorial force */
1240 tx = _mm256_mul_ps(fscal,dx30);
1241 ty = _mm256_mul_ps(fscal,dy30);
1242 tz = _mm256_mul_ps(fscal,dz30);
1244 /* Update vectorial force */
1245 fix3 = _mm256_add_ps(fix3,tx);
1246 fiy3 = _mm256_add_ps(fiy3,ty);
1247 fiz3 = _mm256_add_ps(fiz3,tz);
1249 fjx0 = _mm256_add_ps(fjx0,tx);
1250 fjy0 = _mm256_add_ps(fjy0,ty);
1251 fjz0 = _mm256_add_ps(fjz0,tz);
1255 fjptrA = f+j_coord_offsetA;
1256 fjptrB = f+j_coord_offsetB;
1257 fjptrC = f+j_coord_offsetC;
1258 fjptrD = f+j_coord_offsetD;
1259 fjptrE = f+j_coord_offsetE;
1260 fjptrF = f+j_coord_offsetF;
1261 fjptrG = f+j_coord_offsetG;
1262 fjptrH = f+j_coord_offsetH;
1264 gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,fjx0,fjy0,fjz0);
1266 /* Inner loop uses 141 flops */
1269 if(jidx<j_index_end)
1272 /* Get j neighbor index, and coordinate index */
1273 jnrlistA = jjnr[jidx];
1274 jnrlistB = jjnr[jidx+1];
1275 jnrlistC = jjnr[jidx+2];
1276 jnrlistD = jjnr[jidx+3];
1277 jnrlistE = jjnr[jidx+4];
1278 jnrlistF = jjnr[jidx+5];
1279 jnrlistG = jjnr[jidx+6];
1280 jnrlistH = jjnr[jidx+7];
1281 /* Sign of each element will be negative for non-real atoms.
1282 * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
1283 * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
1285 dummy_mask = gmx_mm256_set_m128(gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx+4)),_mm_setzero_si128())),
1286 gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128())));
1288 jnrA = (jnrlistA>=0) ? jnrlistA : 0;
1289 jnrB = (jnrlistB>=0) ? jnrlistB : 0;
1290 jnrC = (jnrlistC>=0) ? jnrlistC : 0;
1291 jnrD = (jnrlistD>=0) ? jnrlistD : 0;
1292 jnrE = (jnrlistE>=0) ? jnrlistE : 0;
1293 jnrF = (jnrlistF>=0) ? jnrlistF : 0;
1294 jnrG = (jnrlistG>=0) ? jnrlistG : 0;
1295 jnrH = (jnrlistH>=0) ? jnrlistH : 0;
1296 j_coord_offsetA = DIM*jnrA;
1297 j_coord_offsetB = DIM*jnrB;
1298 j_coord_offsetC = DIM*jnrC;
1299 j_coord_offsetD = DIM*jnrD;
1300 j_coord_offsetE = DIM*jnrE;
1301 j_coord_offsetF = DIM*jnrF;
1302 j_coord_offsetG = DIM*jnrG;
1303 j_coord_offsetH = DIM*jnrH;
1305 /* load j atom coordinates */
1306 gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
1307 x+j_coord_offsetC,x+j_coord_offsetD,
1308 x+j_coord_offsetE,x+j_coord_offsetF,
1309 x+j_coord_offsetG,x+j_coord_offsetH,
1312 /* Calculate displacement vector */
1313 dx00 = _mm256_sub_ps(ix0,jx0);
1314 dy00 = _mm256_sub_ps(iy0,jy0);
1315 dz00 = _mm256_sub_ps(iz0,jz0);
1316 dx10 = _mm256_sub_ps(ix1,jx0);
1317 dy10 = _mm256_sub_ps(iy1,jy0);
1318 dz10 = _mm256_sub_ps(iz1,jz0);
1319 dx20 = _mm256_sub_ps(ix2,jx0);
1320 dy20 = _mm256_sub_ps(iy2,jy0);
1321 dz20 = _mm256_sub_ps(iz2,jz0);
1322 dx30 = _mm256_sub_ps(ix3,jx0);
1323 dy30 = _mm256_sub_ps(iy3,jy0);
1324 dz30 = _mm256_sub_ps(iz3,jz0);
1326 /* Calculate squared distance and things based on it */
1327 rsq00 = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
1328 rsq10 = gmx_mm256_calc_rsq_ps(dx10,dy10,dz10);
1329 rsq20 = gmx_mm256_calc_rsq_ps(dx20,dy20,dz20);
1330 rsq30 = gmx_mm256_calc_rsq_ps(dx30,dy30,dz30);
1332 rinv00 = gmx_mm256_invsqrt_ps(rsq00);
1333 rinv10 = gmx_mm256_invsqrt_ps(rsq10);
1334 rinv20 = gmx_mm256_invsqrt_ps(rsq20);
1335 rinv30 = gmx_mm256_invsqrt_ps(rsq30);
1337 rinvsq10 = _mm256_mul_ps(rinv10,rinv10);
1338 rinvsq20 = _mm256_mul_ps(rinv20,rinv20);
1339 rinvsq30 = _mm256_mul_ps(rinv30,rinv30);
1341 /* Load parameters for j particles */
1342 jq0 = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
1343 charge+jnrC+0,charge+jnrD+0,
1344 charge+jnrE+0,charge+jnrF+0,
1345 charge+jnrG+0,charge+jnrH+0);
1346 vdwjidx0A = 2*vdwtype[jnrA+0];
1347 vdwjidx0B = 2*vdwtype[jnrB+0];
1348 vdwjidx0C = 2*vdwtype[jnrC+0];
1349 vdwjidx0D = 2*vdwtype[jnrD+0];
1350 vdwjidx0E = 2*vdwtype[jnrE+0];
1351 vdwjidx0F = 2*vdwtype[jnrF+0];
1352 vdwjidx0G = 2*vdwtype[jnrG+0];
1353 vdwjidx0H = 2*vdwtype[jnrH+0];
1355 fjx0 = _mm256_setzero_ps();
1356 fjy0 = _mm256_setzero_ps();
1357 fjz0 = _mm256_setzero_ps();
1359 /**************************
1360 * CALCULATE INTERACTIONS *
1361 **************************/
1363 r00 = _mm256_mul_ps(rsq00,rinv00);
1364 r00 = _mm256_andnot_ps(dummy_mask,r00);
1366 /* Compute parameters for interactions between i and j atoms */
1367 gmx_mm256_load_8pair_swizzle_ps(vdwioffsetptr0+vdwjidx0A,
1368 vdwioffsetptr0+vdwjidx0B,
1369 vdwioffsetptr0+vdwjidx0C,
1370 vdwioffsetptr0+vdwjidx0D,
1371 vdwioffsetptr0+vdwjidx0E,
1372 vdwioffsetptr0+vdwjidx0F,
1373 vdwioffsetptr0+vdwjidx0G,
1374 vdwioffsetptr0+vdwjidx0H,
1377 /* Calculate table index by multiplying r with table scale and truncate to integer */
1378 rt = _mm256_mul_ps(r00,vftabscale);
1379 vfitab = _mm256_cvttps_epi32(rt);
1380 vfeps = _mm256_sub_ps(rt,_mm256_round_ps(rt, _MM_FROUND_FLOOR));
1381 /* AVX1 does not support 256-bit integer operations, so now we go to 128-bit mode... */
1382 vfitab_lo = _mm256_extractf128_si256(vfitab,0x0);
1383 vfitab_hi = _mm256_extractf128_si256(vfitab,0x1);
1384 vfitab_lo = _mm_slli_epi32(vfitab_lo,3);
1385 vfitab_hi = _mm_slli_epi32(vfitab_hi,3);
1387 /* CUBIC SPLINE TABLE DISPERSION */
1388 Y = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
1389 _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
1390 F = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
1391 _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
1392 G = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
1393 _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
1394 H = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
1395 _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
1396 GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
1397 Heps = _mm256_mul_ps(vfeps,H);
1398 Fp = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
1399 FF = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
1400 fvdw6 = _mm256_mul_ps(c6_00,FF);
1402 /* CUBIC SPLINE TABLE REPULSION */
1403 vfitab_lo = _mm_add_epi32(vfitab_lo,ifour);
1404 vfitab_hi = _mm_add_epi32(vfitab_hi,ifour);
1405 Y = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
1406 _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
1407 F = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
1408 _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
1409 G = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
1410 _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
1411 H = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
1412 _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
1413 GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
1414 Heps = _mm256_mul_ps(vfeps,H);
1415 Fp = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
1416 FF = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
1417 fvdw12 = _mm256_mul_ps(c12_00,FF);
1418 fvdw = _mm256_xor_ps(signbit,_mm256_mul_ps(_mm256_add_ps(fvdw6,fvdw12),_mm256_mul_ps(vftabscale,rinv00)));
1422 fscal = _mm256_andnot_ps(dummy_mask,fscal);
1424 /* Calculate temporary vectorial force */
1425 tx = _mm256_mul_ps(fscal,dx00);
1426 ty = _mm256_mul_ps(fscal,dy00);
1427 tz = _mm256_mul_ps(fscal,dz00);
1429 /* Update vectorial force */
1430 fix0 = _mm256_add_ps(fix0,tx);
1431 fiy0 = _mm256_add_ps(fiy0,ty);
1432 fiz0 = _mm256_add_ps(fiz0,tz);
1434 fjx0 = _mm256_add_ps(fjx0,tx);
1435 fjy0 = _mm256_add_ps(fjy0,ty);
1436 fjz0 = _mm256_add_ps(fjz0,tz);
1438 /**************************
1439 * CALCULATE INTERACTIONS *
1440 **************************/
1442 if (gmx_mm256_any_lt(rsq10,rcutoff2))
1445 /* Compute parameters for interactions between i and j atoms */
1446 qq10 = _mm256_mul_ps(iq1,jq0);
1448 /* REACTION-FIELD ELECTROSTATICS */
1449 felec = _mm256_mul_ps(qq10,_mm256_sub_ps(_mm256_mul_ps(rinv10,rinvsq10),krf2));
1451 cutoff_mask = _mm256_cmp_ps(rsq10,rcutoff2,_CMP_LT_OQ);
1455 fscal = _mm256_and_ps(fscal,cutoff_mask);
1457 fscal = _mm256_andnot_ps(dummy_mask,fscal);
1459 /* Calculate temporary vectorial force */
1460 tx = _mm256_mul_ps(fscal,dx10);
1461 ty = _mm256_mul_ps(fscal,dy10);
1462 tz = _mm256_mul_ps(fscal,dz10);
1464 /* Update vectorial force */
1465 fix1 = _mm256_add_ps(fix1,tx);
1466 fiy1 = _mm256_add_ps(fiy1,ty);
1467 fiz1 = _mm256_add_ps(fiz1,tz);
1469 fjx0 = _mm256_add_ps(fjx0,tx);
1470 fjy0 = _mm256_add_ps(fjy0,ty);
1471 fjz0 = _mm256_add_ps(fjz0,tz);
1475 /**************************
1476 * CALCULATE INTERACTIONS *
1477 **************************/
1479 if (gmx_mm256_any_lt(rsq20,rcutoff2))
1482 /* Compute parameters for interactions between i and j atoms */
1483 qq20 = _mm256_mul_ps(iq2,jq0);
1485 /* REACTION-FIELD ELECTROSTATICS */
1486 felec = _mm256_mul_ps(qq20,_mm256_sub_ps(_mm256_mul_ps(rinv20,rinvsq20),krf2));
1488 cutoff_mask = _mm256_cmp_ps(rsq20,rcutoff2,_CMP_LT_OQ);
1492 fscal = _mm256_and_ps(fscal,cutoff_mask);
1494 fscal = _mm256_andnot_ps(dummy_mask,fscal);
1496 /* Calculate temporary vectorial force */
1497 tx = _mm256_mul_ps(fscal,dx20);
1498 ty = _mm256_mul_ps(fscal,dy20);
1499 tz = _mm256_mul_ps(fscal,dz20);
1501 /* Update vectorial force */
1502 fix2 = _mm256_add_ps(fix2,tx);
1503 fiy2 = _mm256_add_ps(fiy2,ty);
1504 fiz2 = _mm256_add_ps(fiz2,tz);
1506 fjx0 = _mm256_add_ps(fjx0,tx);
1507 fjy0 = _mm256_add_ps(fjy0,ty);
1508 fjz0 = _mm256_add_ps(fjz0,tz);
1512 /**************************
1513 * CALCULATE INTERACTIONS *
1514 **************************/
1516 if (gmx_mm256_any_lt(rsq30,rcutoff2))
1519 /* Compute parameters for interactions between i and j atoms */
1520 qq30 = _mm256_mul_ps(iq3,jq0);
1522 /* REACTION-FIELD ELECTROSTATICS */
1523 felec = _mm256_mul_ps(qq30,_mm256_sub_ps(_mm256_mul_ps(rinv30,rinvsq30),krf2));
1525 cutoff_mask = _mm256_cmp_ps(rsq30,rcutoff2,_CMP_LT_OQ);
1529 fscal = _mm256_and_ps(fscal,cutoff_mask);
1531 fscal = _mm256_andnot_ps(dummy_mask,fscal);
1533 /* Calculate temporary vectorial force */
1534 tx = _mm256_mul_ps(fscal,dx30);
1535 ty = _mm256_mul_ps(fscal,dy30);
1536 tz = _mm256_mul_ps(fscal,dz30);
1538 /* Update vectorial force */
1539 fix3 = _mm256_add_ps(fix3,tx);
1540 fiy3 = _mm256_add_ps(fiy3,ty);
1541 fiz3 = _mm256_add_ps(fiz3,tz);
1543 fjx0 = _mm256_add_ps(fjx0,tx);
1544 fjy0 = _mm256_add_ps(fjy0,ty);
1545 fjz0 = _mm256_add_ps(fjz0,tz);
1549 fjptrA = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
1550 fjptrB = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
1551 fjptrC = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
1552 fjptrD = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
1553 fjptrE = (jnrlistE>=0) ? f+j_coord_offsetE : scratch;
1554 fjptrF = (jnrlistF>=0) ? f+j_coord_offsetF : scratch;
1555 fjptrG = (jnrlistG>=0) ? f+j_coord_offsetG : scratch;
1556 fjptrH = (jnrlistH>=0) ? f+j_coord_offsetH : scratch;
1558 gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,fjx0,fjy0,fjz0);
1560 /* Inner loop uses 142 flops */
1563 /* End of innermost loop */
1565 gmx_mm256_update_iforce_4atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
1566 f+i_coord_offset,fshift+i_shift_offset);
1568 /* Increment number of inner iterations */
1569 inneriter += j_index_end - j_index_start;
1571 /* Outer loop uses 24 flops */
1574 /* Increment number of outer iterations */
1577 /* Update outer/inner flops */
1579 inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W4_F,outeriter*24 + inneriter*142);