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_double kernel generator.
44 #include "../nb_kernel.h"
45 #include "types/simple.h"
49 #include "gromacs/simd/math_x86_avx_256_double.h"
50 #include "kernelutil_x86_avx_256_double.h"
53 * Gromacs nonbonded kernel: nb_kernel_ElecRFCut_VdwCSTab_GeomW4P1_VF_avx_256_double
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_double
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 refer to j loop unrolling done with AVX, e.g. for the four 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 jnrlistA,jnrlistB,jnrlistC,jnrlistD;
78 int jnrlistE,jnrlistF,jnrlistG,jnrlistH;
79 int j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
80 int *iinr,*jindex,*jjnr,*shiftidx,*gid;
82 real *shiftvec,*fshift,*x,*f;
83 real *fjptrA,*fjptrB,*fjptrC,*fjptrD;
85 __m256d tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
86 real * vdwioffsetptr0;
87 __m256d ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
88 real * vdwioffsetptr1;
89 __m256d ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
90 real * vdwioffsetptr2;
91 __m256d ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
92 real * vdwioffsetptr3;
93 __m256d ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
94 int vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
95 __m256d jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
96 __m256d dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
97 __m256d dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
98 __m256d dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
99 __m256d dx30,dy30,dz30,rsq30,rinv30,rinvsq30,r30,qq30,c6_30,c12_30;
100 __m256d velec,felec,velecsum,facel,crf,krf,krf2;
103 __m256d rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
106 __m256d one_sixth = _mm256_set1_pd(1.0/6.0);
107 __m256d one_twelfth = _mm256_set1_pd(1.0/12.0);
109 __m128i ifour = _mm_set1_epi32(4);
110 __m256d rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF;
112 __m256d dummy_mask,cutoff_mask;
113 __m128 tmpmask0,tmpmask1;
114 __m256d signbit = _mm256_castsi256_pd( _mm256_set1_epi32(0x80000000) );
115 __m256d one = _mm256_set1_pd(1.0);
116 __m256d two = _mm256_set1_pd(2.0);
122 jindex = nlist->jindex;
124 shiftidx = nlist->shift;
126 shiftvec = fr->shift_vec[0];
127 fshift = fr->fshift[0];
128 facel = _mm256_set1_pd(fr->epsfac);
129 charge = mdatoms->chargeA;
130 krf = _mm256_set1_pd(fr->ic->k_rf);
131 krf2 = _mm256_set1_pd(fr->ic->k_rf*2.0);
132 crf = _mm256_set1_pd(fr->ic->c_rf);
133 nvdwtype = fr->ntype;
135 vdwtype = mdatoms->typeA;
137 vftab = kernel_data->table_vdw->data;
138 vftabscale = _mm256_set1_pd(kernel_data->table_vdw->scale);
140 /* Setup water-specific parameters */
141 inr = nlist->iinr[0];
142 iq1 = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+1]));
143 iq2 = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+2]));
144 iq3 = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+3]));
145 vdwioffsetptr0 = vdwparam+2*nvdwtype*vdwtype[inr+0];
147 /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
148 rcutoff_scalar = fr->rcoulomb;
149 rcutoff = _mm256_set1_pd(rcutoff_scalar);
150 rcutoff2 = _mm256_mul_pd(rcutoff,rcutoff);
152 /* Avoid stupid compiler warnings */
153 jnrA = jnrB = jnrC = jnrD = 0;
162 for(iidx=0;iidx<4*DIM;iidx++)
167 /* Start outer loop over neighborlists */
168 for(iidx=0; iidx<nri; iidx++)
170 /* Load shift vector for this list */
171 i_shift_offset = DIM*shiftidx[iidx];
173 /* Load limits for loop over neighbors */
174 j_index_start = jindex[iidx];
175 j_index_end = jindex[iidx+1];
177 /* Get outer coordinate index */
179 i_coord_offset = DIM*inr;
181 /* Load i particle coords and add shift vector */
182 gmx_mm256_load_shift_and_4rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
183 &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
185 fix0 = _mm256_setzero_pd();
186 fiy0 = _mm256_setzero_pd();
187 fiz0 = _mm256_setzero_pd();
188 fix1 = _mm256_setzero_pd();
189 fiy1 = _mm256_setzero_pd();
190 fiz1 = _mm256_setzero_pd();
191 fix2 = _mm256_setzero_pd();
192 fiy2 = _mm256_setzero_pd();
193 fiz2 = _mm256_setzero_pd();
194 fix3 = _mm256_setzero_pd();
195 fiy3 = _mm256_setzero_pd();
196 fiz3 = _mm256_setzero_pd();
198 /* Reset potential sums */
199 velecsum = _mm256_setzero_pd();
200 vvdwsum = _mm256_setzero_pd();
202 /* Start inner kernel loop */
203 for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
206 /* Get j neighbor index, and coordinate index */
211 j_coord_offsetA = DIM*jnrA;
212 j_coord_offsetB = DIM*jnrB;
213 j_coord_offsetC = DIM*jnrC;
214 j_coord_offsetD = DIM*jnrD;
216 /* load j atom coordinates */
217 gmx_mm256_load_1rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
218 x+j_coord_offsetC,x+j_coord_offsetD,
221 /* Calculate displacement vector */
222 dx00 = _mm256_sub_pd(ix0,jx0);
223 dy00 = _mm256_sub_pd(iy0,jy0);
224 dz00 = _mm256_sub_pd(iz0,jz0);
225 dx10 = _mm256_sub_pd(ix1,jx0);
226 dy10 = _mm256_sub_pd(iy1,jy0);
227 dz10 = _mm256_sub_pd(iz1,jz0);
228 dx20 = _mm256_sub_pd(ix2,jx0);
229 dy20 = _mm256_sub_pd(iy2,jy0);
230 dz20 = _mm256_sub_pd(iz2,jz0);
231 dx30 = _mm256_sub_pd(ix3,jx0);
232 dy30 = _mm256_sub_pd(iy3,jy0);
233 dz30 = _mm256_sub_pd(iz3,jz0);
235 /* Calculate squared distance and things based on it */
236 rsq00 = gmx_mm256_calc_rsq_pd(dx00,dy00,dz00);
237 rsq10 = gmx_mm256_calc_rsq_pd(dx10,dy10,dz10);
238 rsq20 = gmx_mm256_calc_rsq_pd(dx20,dy20,dz20);
239 rsq30 = gmx_mm256_calc_rsq_pd(dx30,dy30,dz30);
241 rinv00 = gmx_mm256_invsqrt_pd(rsq00);
242 rinv10 = gmx_mm256_invsqrt_pd(rsq10);
243 rinv20 = gmx_mm256_invsqrt_pd(rsq20);
244 rinv30 = gmx_mm256_invsqrt_pd(rsq30);
246 rinvsq10 = _mm256_mul_pd(rinv10,rinv10);
247 rinvsq20 = _mm256_mul_pd(rinv20,rinv20);
248 rinvsq30 = _mm256_mul_pd(rinv30,rinv30);
250 /* Load parameters for j particles */
251 jq0 = gmx_mm256_load_4real_swizzle_pd(charge+jnrA+0,charge+jnrB+0,
252 charge+jnrC+0,charge+jnrD+0);
253 vdwjidx0A = 2*vdwtype[jnrA+0];
254 vdwjidx0B = 2*vdwtype[jnrB+0];
255 vdwjidx0C = 2*vdwtype[jnrC+0];
256 vdwjidx0D = 2*vdwtype[jnrD+0];
258 fjx0 = _mm256_setzero_pd();
259 fjy0 = _mm256_setzero_pd();
260 fjz0 = _mm256_setzero_pd();
262 /**************************
263 * CALCULATE INTERACTIONS *
264 **************************/
266 r00 = _mm256_mul_pd(rsq00,rinv00);
268 /* Compute parameters for interactions between i and j atoms */
269 gmx_mm256_load_4pair_swizzle_pd(vdwioffsetptr0+vdwjidx0A,
270 vdwioffsetptr0+vdwjidx0B,
271 vdwioffsetptr0+vdwjidx0C,
272 vdwioffsetptr0+vdwjidx0D,
275 /* Calculate table index by multiplying r with table scale and truncate to integer */
276 rt = _mm256_mul_pd(r00,vftabscale);
277 vfitab = _mm256_cvttpd_epi32(rt);
278 vfeps = _mm256_sub_pd(rt,_mm256_round_pd(rt, _MM_FROUND_FLOOR));
279 vfitab = _mm_slli_epi32(vfitab,3);
281 /* CUBIC SPLINE TABLE DISPERSION */
282 Y = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
283 F = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
284 G = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
285 H = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
286 GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
287 Heps = _mm256_mul_pd(vfeps,H);
288 Fp = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
289 VV = _mm256_add_pd(Y,_mm256_mul_pd(vfeps,Fp));
290 vvdw6 = _mm256_mul_pd(c6_00,VV);
291 FF = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
292 fvdw6 = _mm256_mul_pd(c6_00,FF);
294 /* CUBIC SPLINE TABLE REPULSION */
295 vfitab = _mm_add_epi32(vfitab,ifour);
296 Y = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
297 F = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
298 G = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
299 H = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
300 GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
301 Heps = _mm256_mul_pd(vfeps,H);
302 Fp = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
303 VV = _mm256_add_pd(Y,_mm256_mul_pd(vfeps,Fp));
304 vvdw12 = _mm256_mul_pd(c12_00,VV);
305 FF = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
306 fvdw12 = _mm256_mul_pd(c12_00,FF);
307 vvdw = _mm256_add_pd(vvdw12,vvdw6);
308 fvdw = _mm256_xor_pd(signbit,_mm256_mul_pd(_mm256_add_pd(fvdw6,fvdw12),_mm256_mul_pd(vftabscale,rinv00)));
310 /* Update potential sum for this i atom from the interaction with this j atom. */
311 vvdwsum = _mm256_add_pd(vvdwsum,vvdw);
315 /* Calculate temporary vectorial force */
316 tx = _mm256_mul_pd(fscal,dx00);
317 ty = _mm256_mul_pd(fscal,dy00);
318 tz = _mm256_mul_pd(fscal,dz00);
320 /* Update vectorial force */
321 fix0 = _mm256_add_pd(fix0,tx);
322 fiy0 = _mm256_add_pd(fiy0,ty);
323 fiz0 = _mm256_add_pd(fiz0,tz);
325 fjx0 = _mm256_add_pd(fjx0,tx);
326 fjy0 = _mm256_add_pd(fjy0,ty);
327 fjz0 = _mm256_add_pd(fjz0,tz);
329 /**************************
330 * CALCULATE INTERACTIONS *
331 **************************/
333 if (gmx_mm256_any_lt(rsq10,rcutoff2))
336 /* Compute parameters for interactions between i and j atoms */
337 qq10 = _mm256_mul_pd(iq1,jq0);
339 /* REACTION-FIELD ELECTROSTATICS */
340 velec = _mm256_mul_pd(qq10,_mm256_sub_pd(_mm256_add_pd(rinv10,_mm256_mul_pd(krf,rsq10)),crf));
341 felec = _mm256_mul_pd(qq10,_mm256_sub_pd(_mm256_mul_pd(rinv10,rinvsq10),krf2));
343 cutoff_mask = _mm256_cmp_pd(rsq10,rcutoff2,_CMP_LT_OQ);
345 /* Update potential sum for this i atom from the interaction with this j atom. */
346 velec = _mm256_and_pd(velec,cutoff_mask);
347 velecsum = _mm256_add_pd(velecsum,velec);
351 fscal = _mm256_and_pd(fscal,cutoff_mask);
353 /* Calculate temporary vectorial force */
354 tx = _mm256_mul_pd(fscal,dx10);
355 ty = _mm256_mul_pd(fscal,dy10);
356 tz = _mm256_mul_pd(fscal,dz10);
358 /* Update vectorial force */
359 fix1 = _mm256_add_pd(fix1,tx);
360 fiy1 = _mm256_add_pd(fiy1,ty);
361 fiz1 = _mm256_add_pd(fiz1,tz);
363 fjx0 = _mm256_add_pd(fjx0,tx);
364 fjy0 = _mm256_add_pd(fjy0,ty);
365 fjz0 = _mm256_add_pd(fjz0,tz);
369 /**************************
370 * CALCULATE INTERACTIONS *
371 **************************/
373 if (gmx_mm256_any_lt(rsq20,rcutoff2))
376 /* Compute parameters for interactions between i and j atoms */
377 qq20 = _mm256_mul_pd(iq2,jq0);
379 /* REACTION-FIELD ELECTROSTATICS */
380 velec = _mm256_mul_pd(qq20,_mm256_sub_pd(_mm256_add_pd(rinv20,_mm256_mul_pd(krf,rsq20)),crf));
381 felec = _mm256_mul_pd(qq20,_mm256_sub_pd(_mm256_mul_pd(rinv20,rinvsq20),krf2));
383 cutoff_mask = _mm256_cmp_pd(rsq20,rcutoff2,_CMP_LT_OQ);
385 /* Update potential sum for this i atom from the interaction with this j atom. */
386 velec = _mm256_and_pd(velec,cutoff_mask);
387 velecsum = _mm256_add_pd(velecsum,velec);
391 fscal = _mm256_and_pd(fscal,cutoff_mask);
393 /* Calculate temporary vectorial force */
394 tx = _mm256_mul_pd(fscal,dx20);
395 ty = _mm256_mul_pd(fscal,dy20);
396 tz = _mm256_mul_pd(fscal,dz20);
398 /* Update vectorial force */
399 fix2 = _mm256_add_pd(fix2,tx);
400 fiy2 = _mm256_add_pd(fiy2,ty);
401 fiz2 = _mm256_add_pd(fiz2,tz);
403 fjx0 = _mm256_add_pd(fjx0,tx);
404 fjy0 = _mm256_add_pd(fjy0,ty);
405 fjz0 = _mm256_add_pd(fjz0,tz);
409 /**************************
410 * CALCULATE INTERACTIONS *
411 **************************/
413 if (gmx_mm256_any_lt(rsq30,rcutoff2))
416 /* Compute parameters for interactions between i and j atoms */
417 qq30 = _mm256_mul_pd(iq3,jq0);
419 /* REACTION-FIELD ELECTROSTATICS */
420 velec = _mm256_mul_pd(qq30,_mm256_sub_pd(_mm256_add_pd(rinv30,_mm256_mul_pd(krf,rsq30)),crf));
421 felec = _mm256_mul_pd(qq30,_mm256_sub_pd(_mm256_mul_pd(rinv30,rinvsq30),krf2));
423 cutoff_mask = _mm256_cmp_pd(rsq30,rcutoff2,_CMP_LT_OQ);
425 /* Update potential sum for this i atom from the interaction with this j atom. */
426 velec = _mm256_and_pd(velec,cutoff_mask);
427 velecsum = _mm256_add_pd(velecsum,velec);
431 fscal = _mm256_and_pd(fscal,cutoff_mask);
433 /* Calculate temporary vectorial force */
434 tx = _mm256_mul_pd(fscal,dx30);
435 ty = _mm256_mul_pd(fscal,dy30);
436 tz = _mm256_mul_pd(fscal,dz30);
438 /* Update vectorial force */
439 fix3 = _mm256_add_pd(fix3,tx);
440 fiy3 = _mm256_add_pd(fiy3,ty);
441 fiz3 = _mm256_add_pd(fiz3,tz);
443 fjx0 = _mm256_add_pd(fjx0,tx);
444 fjy0 = _mm256_add_pd(fjy0,ty);
445 fjz0 = _mm256_add_pd(fjz0,tz);
449 fjptrA = f+j_coord_offsetA;
450 fjptrB = f+j_coord_offsetB;
451 fjptrC = f+j_coord_offsetC;
452 fjptrD = f+j_coord_offsetD;
454 gmx_mm256_decrement_1rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
456 /* Inner loop uses 167 flops */
462 /* Get j neighbor index, and coordinate index */
463 jnrlistA = jjnr[jidx];
464 jnrlistB = jjnr[jidx+1];
465 jnrlistC = jjnr[jidx+2];
466 jnrlistD = jjnr[jidx+3];
467 /* Sign of each element will be negative for non-real atoms.
468 * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
469 * so use it as val = _mm_andnot_pd(mask,val) to clear dummy entries.
471 tmpmask0 = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
473 tmpmask1 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(3,3,2,2));
474 tmpmask0 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(1,1,0,0));
475 dummy_mask = _mm256_castps_pd(gmx_mm256_set_m128(tmpmask1,tmpmask0));
477 jnrA = (jnrlistA>=0) ? jnrlistA : 0;
478 jnrB = (jnrlistB>=0) ? jnrlistB : 0;
479 jnrC = (jnrlistC>=0) ? jnrlistC : 0;
480 jnrD = (jnrlistD>=0) ? jnrlistD : 0;
481 j_coord_offsetA = DIM*jnrA;
482 j_coord_offsetB = DIM*jnrB;
483 j_coord_offsetC = DIM*jnrC;
484 j_coord_offsetD = DIM*jnrD;
486 /* load j atom coordinates */
487 gmx_mm256_load_1rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
488 x+j_coord_offsetC,x+j_coord_offsetD,
491 /* Calculate displacement vector */
492 dx00 = _mm256_sub_pd(ix0,jx0);
493 dy00 = _mm256_sub_pd(iy0,jy0);
494 dz00 = _mm256_sub_pd(iz0,jz0);
495 dx10 = _mm256_sub_pd(ix1,jx0);
496 dy10 = _mm256_sub_pd(iy1,jy0);
497 dz10 = _mm256_sub_pd(iz1,jz0);
498 dx20 = _mm256_sub_pd(ix2,jx0);
499 dy20 = _mm256_sub_pd(iy2,jy0);
500 dz20 = _mm256_sub_pd(iz2,jz0);
501 dx30 = _mm256_sub_pd(ix3,jx0);
502 dy30 = _mm256_sub_pd(iy3,jy0);
503 dz30 = _mm256_sub_pd(iz3,jz0);
505 /* Calculate squared distance and things based on it */
506 rsq00 = gmx_mm256_calc_rsq_pd(dx00,dy00,dz00);
507 rsq10 = gmx_mm256_calc_rsq_pd(dx10,dy10,dz10);
508 rsq20 = gmx_mm256_calc_rsq_pd(dx20,dy20,dz20);
509 rsq30 = gmx_mm256_calc_rsq_pd(dx30,dy30,dz30);
511 rinv00 = gmx_mm256_invsqrt_pd(rsq00);
512 rinv10 = gmx_mm256_invsqrt_pd(rsq10);
513 rinv20 = gmx_mm256_invsqrt_pd(rsq20);
514 rinv30 = gmx_mm256_invsqrt_pd(rsq30);
516 rinvsq10 = _mm256_mul_pd(rinv10,rinv10);
517 rinvsq20 = _mm256_mul_pd(rinv20,rinv20);
518 rinvsq30 = _mm256_mul_pd(rinv30,rinv30);
520 /* Load parameters for j particles */
521 jq0 = gmx_mm256_load_4real_swizzle_pd(charge+jnrA+0,charge+jnrB+0,
522 charge+jnrC+0,charge+jnrD+0);
523 vdwjidx0A = 2*vdwtype[jnrA+0];
524 vdwjidx0B = 2*vdwtype[jnrB+0];
525 vdwjidx0C = 2*vdwtype[jnrC+0];
526 vdwjidx0D = 2*vdwtype[jnrD+0];
528 fjx0 = _mm256_setzero_pd();
529 fjy0 = _mm256_setzero_pd();
530 fjz0 = _mm256_setzero_pd();
532 /**************************
533 * CALCULATE INTERACTIONS *
534 **************************/
536 r00 = _mm256_mul_pd(rsq00,rinv00);
537 r00 = _mm256_andnot_pd(dummy_mask,r00);
539 /* Compute parameters for interactions between i and j atoms */
540 gmx_mm256_load_4pair_swizzle_pd(vdwioffsetptr0+vdwjidx0A,
541 vdwioffsetptr0+vdwjidx0B,
542 vdwioffsetptr0+vdwjidx0C,
543 vdwioffsetptr0+vdwjidx0D,
546 /* Calculate table index by multiplying r with table scale and truncate to integer */
547 rt = _mm256_mul_pd(r00,vftabscale);
548 vfitab = _mm256_cvttpd_epi32(rt);
549 vfeps = _mm256_sub_pd(rt,_mm256_round_pd(rt, _MM_FROUND_FLOOR));
550 vfitab = _mm_slli_epi32(vfitab,3);
552 /* CUBIC SPLINE TABLE DISPERSION */
553 Y = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
554 F = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
555 G = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
556 H = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
557 GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
558 Heps = _mm256_mul_pd(vfeps,H);
559 Fp = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
560 VV = _mm256_add_pd(Y,_mm256_mul_pd(vfeps,Fp));
561 vvdw6 = _mm256_mul_pd(c6_00,VV);
562 FF = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
563 fvdw6 = _mm256_mul_pd(c6_00,FF);
565 /* CUBIC SPLINE TABLE REPULSION */
566 vfitab = _mm_add_epi32(vfitab,ifour);
567 Y = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
568 F = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
569 G = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
570 H = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
571 GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
572 Heps = _mm256_mul_pd(vfeps,H);
573 Fp = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
574 VV = _mm256_add_pd(Y,_mm256_mul_pd(vfeps,Fp));
575 vvdw12 = _mm256_mul_pd(c12_00,VV);
576 FF = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
577 fvdw12 = _mm256_mul_pd(c12_00,FF);
578 vvdw = _mm256_add_pd(vvdw12,vvdw6);
579 fvdw = _mm256_xor_pd(signbit,_mm256_mul_pd(_mm256_add_pd(fvdw6,fvdw12),_mm256_mul_pd(vftabscale,rinv00)));
581 /* Update potential sum for this i atom from the interaction with this j atom. */
582 vvdw = _mm256_andnot_pd(dummy_mask,vvdw);
583 vvdwsum = _mm256_add_pd(vvdwsum,vvdw);
587 fscal = _mm256_andnot_pd(dummy_mask,fscal);
589 /* Calculate temporary vectorial force */
590 tx = _mm256_mul_pd(fscal,dx00);
591 ty = _mm256_mul_pd(fscal,dy00);
592 tz = _mm256_mul_pd(fscal,dz00);
594 /* Update vectorial force */
595 fix0 = _mm256_add_pd(fix0,tx);
596 fiy0 = _mm256_add_pd(fiy0,ty);
597 fiz0 = _mm256_add_pd(fiz0,tz);
599 fjx0 = _mm256_add_pd(fjx0,tx);
600 fjy0 = _mm256_add_pd(fjy0,ty);
601 fjz0 = _mm256_add_pd(fjz0,tz);
603 /**************************
604 * CALCULATE INTERACTIONS *
605 **************************/
607 if (gmx_mm256_any_lt(rsq10,rcutoff2))
610 /* Compute parameters for interactions between i and j atoms */
611 qq10 = _mm256_mul_pd(iq1,jq0);
613 /* REACTION-FIELD ELECTROSTATICS */
614 velec = _mm256_mul_pd(qq10,_mm256_sub_pd(_mm256_add_pd(rinv10,_mm256_mul_pd(krf,rsq10)),crf));
615 felec = _mm256_mul_pd(qq10,_mm256_sub_pd(_mm256_mul_pd(rinv10,rinvsq10),krf2));
617 cutoff_mask = _mm256_cmp_pd(rsq10,rcutoff2,_CMP_LT_OQ);
619 /* Update potential sum for this i atom from the interaction with this j atom. */
620 velec = _mm256_and_pd(velec,cutoff_mask);
621 velec = _mm256_andnot_pd(dummy_mask,velec);
622 velecsum = _mm256_add_pd(velecsum,velec);
626 fscal = _mm256_and_pd(fscal,cutoff_mask);
628 fscal = _mm256_andnot_pd(dummy_mask,fscal);
630 /* Calculate temporary vectorial force */
631 tx = _mm256_mul_pd(fscal,dx10);
632 ty = _mm256_mul_pd(fscal,dy10);
633 tz = _mm256_mul_pd(fscal,dz10);
635 /* Update vectorial force */
636 fix1 = _mm256_add_pd(fix1,tx);
637 fiy1 = _mm256_add_pd(fiy1,ty);
638 fiz1 = _mm256_add_pd(fiz1,tz);
640 fjx0 = _mm256_add_pd(fjx0,tx);
641 fjy0 = _mm256_add_pd(fjy0,ty);
642 fjz0 = _mm256_add_pd(fjz0,tz);
646 /**************************
647 * CALCULATE INTERACTIONS *
648 **************************/
650 if (gmx_mm256_any_lt(rsq20,rcutoff2))
653 /* Compute parameters for interactions between i and j atoms */
654 qq20 = _mm256_mul_pd(iq2,jq0);
656 /* REACTION-FIELD ELECTROSTATICS */
657 velec = _mm256_mul_pd(qq20,_mm256_sub_pd(_mm256_add_pd(rinv20,_mm256_mul_pd(krf,rsq20)),crf));
658 felec = _mm256_mul_pd(qq20,_mm256_sub_pd(_mm256_mul_pd(rinv20,rinvsq20),krf2));
660 cutoff_mask = _mm256_cmp_pd(rsq20,rcutoff2,_CMP_LT_OQ);
662 /* Update potential sum for this i atom from the interaction with this j atom. */
663 velec = _mm256_and_pd(velec,cutoff_mask);
664 velec = _mm256_andnot_pd(dummy_mask,velec);
665 velecsum = _mm256_add_pd(velecsum,velec);
669 fscal = _mm256_and_pd(fscal,cutoff_mask);
671 fscal = _mm256_andnot_pd(dummy_mask,fscal);
673 /* Calculate temporary vectorial force */
674 tx = _mm256_mul_pd(fscal,dx20);
675 ty = _mm256_mul_pd(fscal,dy20);
676 tz = _mm256_mul_pd(fscal,dz20);
678 /* Update vectorial force */
679 fix2 = _mm256_add_pd(fix2,tx);
680 fiy2 = _mm256_add_pd(fiy2,ty);
681 fiz2 = _mm256_add_pd(fiz2,tz);
683 fjx0 = _mm256_add_pd(fjx0,tx);
684 fjy0 = _mm256_add_pd(fjy0,ty);
685 fjz0 = _mm256_add_pd(fjz0,tz);
689 /**************************
690 * CALCULATE INTERACTIONS *
691 **************************/
693 if (gmx_mm256_any_lt(rsq30,rcutoff2))
696 /* Compute parameters for interactions between i and j atoms */
697 qq30 = _mm256_mul_pd(iq3,jq0);
699 /* REACTION-FIELD ELECTROSTATICS */
700 velec = _mm256_mul_pd(qq30,_mm256_sub_pd(_mm256_add_pd(rinv30,_mm256_mul_pd(krf,rsq30)),crf));
701 felec = _mm256_mul_pd(qq30,_mm256_sub_pd(_mm256_mul_pd(rinv30,rinvsq30),krf2));
703 cutoff_mask = _mm256_cmp_pd(rsq30,rcutoff2,_CMP_LT_OQ);
705 /* Update potential sum for this i atom from the interaction with this j atom. */
706 velec = _mm256_and_pd(velec,cutoff_mask);
707 velec = _mm256_andnot_pd(dummy_mask,velec);
708 velecsum = _mm256_add_pd(velecsum,velec);
712 fscal = _mm256_and_pd(fscal,cutoff_mask);
714 fscal = _mm256_andnot_pd(dummy_mask,fscal);
716 /* Calculate temporary vectorial force */
717 tx = _mm256_mul_pd(fscal,dx30);
718 ty = _mm256_mul_pd(fscal,dy30);
719 tz = _mm256_mul_pd(fscal,dz30);
721 /* Update vectorial force */
722 fix3 = _mm256_add_pd(fix3,tx);
723 fiy3 = _mm256_add_pd(fiy3,ty);
724 fiz3 = _mm256_add_pd(fiz3,tz);
726 fjx0 = _mm256_add_pd(fjx0,tx);
727 fjy0 = _mm256_add_pd(fjy0,ty);
728 fjz0 = _mm256_add_pd(fjz0,tz);
732 fjptrA = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
733 fjptrB = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
734 fjptrC = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
735 fjptrD = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
737 gmx_mm256_decrement_1rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
739 /* Inner loop uses 168 flops */
742 /* End of innermost loop */
744 gmx_mm256_update_iforce_4atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
745 f+i_coord_offset,fshift+i_shift_offset);
748 /* Update potential energies */
749 gmx_mm256_update_1pot_pd(velecsum,kernel_data->energygrp_elec+ggid);
750 gmx_mm256_update_1pot_pd(vvdwsum,kernel_data->energygrp_vdw+ggid);
752 /* Increment number of inner iterations */
753 inneriter += j_index_end - j_index_start;
755 /* Outer loop uses 26 flops */
758 /* Increment number of outer iterations */
761 /* Update outer/inner flops */
763 inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W4_VF,outeriter*26 + inneriter*168);
766 * Gromacs nonbonded kernel: nb_kernel_ElecRFCut_VdwCSTab_GeomW4P1_F_avx_256_double
767 * Electrostatics interaction: ReactionField
768 * VdW interaction: CubicSplineTable
769 * Geometry: Water4-Particle
770 * Calculate force/pot: Force
773 nb_kernel_ElecRFCut_VdwCSTab_GeomW4P1_F_avx_256_double
774 (t_nblist * gmx_restrict nlist,
775 rvec * gmx_restrict xx,
776 rvec * gmx_restrict ff,
777 t_forcerec * gmx_restrict fr,
778 t_mdatoms * gmx_restrict mdatoms,
779 nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
780 t_nrnb * gmx_restrict nrnb)
782 /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
783 * just 0 for non-waters.
784 * Suffixes A,B,C,D refer to j loop unrolling done with AVX, e.g. for the four different
785 * jnr indices corresponding to data put in the four positions in the SIMD register.
787 int i_shift_offset,i_coord_offset,outeriter,inneriter;
788 int j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
789 int jnrA,jnrB,jnrC,jnrD;
790 int jnrlistA,jnrlistB,jnrlistC,jnrlistD;
791 int jnrlistE,jnrlistF,jnrlistG,jnrlistH;
792 int j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
793 int *iinr,*jindex,*jjnr,*shiftidx,*gid;
795 real *shiftvec,*fshift,*x,*f;
796 real *fjptrA,*fjptrB,*fjptrC,*fjptrD;
798 __m256d tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
799 real * vdwioffsetptr0;
800 __m256d ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
801 real * vdwioffsetptr1;
802 __m256d ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
803 real * vdwioffsetptr2;
804 __m256d ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
805 real * vdwioffsetptr3;
806 __m256d ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
807 int vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
808 __m256d jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
809 __m256d dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
810 __m256d dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
811 __m256d dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
812 __m256d dx30,dy30,dz30,rsq30,rinv30,rinvsq30,r30,qq30,c6_30,c12_30;
813 __m256d velec,felec,velecsum,facel,crf,krf,krf2;
816 __m256d rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
819 __m256d one_sixth = _mm256_set1_pd(1.0/6.0);
820 __m256d one_twelfth = _mm256_set1_pd(1.0/12.0);
822 __m128i ifour = _mm_set1_epi32(4);
823 __m256d rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF;
825 __m256d dummy_mask,cutoff_mask;
826 __m128 tmpmask0,tmpmask1;
827 __m256d signbit = _mm256_castsi256_pd( _mm256_set1_epi32(0x80000000) );
828 __m256d one = _mm256_set1_pd(1.0);
829 __m256d two = _mm256_set1_pd(2.0);
835 jindex = nlist->jindex;
837 shiftidx = nlist->shift;
839 shiftvec = fr->shift_vec[0];
840 fshift = fr->fshift[0];
841 facel = _mm256_set1_pd(fr->epsfac);
842 charge = mdatoms->chargeA;
843 krf = _mm256_set1_pd(fr->ic->k_rf);
844 krf2 = _mm256_set1_pd(fr->ic->k_rf*2.0);
845 crf = _mm256_set1_pd(fr->ic->c_rf);
846 nvdwtype = fr->ntype;
848 vdwtype = mdatoms->typeA;
850 vftab = kernel_data->table_vdw->data;
851 vftabscale = _mm256_set1_pd(kernel_data->table_vdw->scale);
853 /* Setup water-specific parameters */
854 inr = nlist->iinr[0];
855 iq1 = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+1]));
856 iq2 = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+2]));
857 iq3 = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+3]));
858 vdwioffsetptr0 = vdwparam+2*nvdwtype*vdwtype[inr+0];
860 /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
861 rcutoff_scalar = fr->rcoulomb;
862 rcutoff = _mm256_set1_pd(rcutoff_scalar);
863 rcutoff2 = _mm256_mul_pd(rcutoff,rcutoff);
865 /* Avoid stupid compiler warnings */
866 jnrA = jnrB = jnrC = jnrD = 0;
875 for(iidx=0;iidx<4*DIM;iidx++)
880 /* Start outer loop over neighborlists */
881 for(iidx=0; iidx<nri; iidx++)
883 /* Load shift vector for this list */
884 i_shift_offset = DIM*shiftidx[iidx];
886 /* Load limits for loop over neighbors */
887 j_index_start = jindex[iidx];
888 j_index_end = jindex[iidx+1];
890 /* Get outer coordinate index */
892 i_coord_offset = DIM*inr;
894 /* Load i particle coords and add shift vector */
895 gmx_mm256_load_shift_and_4rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
896 &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
898 fix0 = _mm256_setzero_pd();
899 fiy0 = _mm256_setzero_pd();
900 fiz0 = _mm256_setzero_pd();
901 fix1 = _mm256_setzero_pd();
902 fiy1 = _mm256_setzero_pd();
903 fiz1 = _mm256_setzero_pd();
904 fix2 = _mm256_setzero_pd();
905 fiy2 = _mm256_setzero_pd();
906 fiz2 = _mm256_setzero_pd();
907 fix3 = _mm256_setzero_pd();
908 fiy3 = _mm256_setzero_pd();
909 fiz3 = _mm256_setzero_pd();
911 /* Start inner kernel loop */
912 for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
915 /* Get j neighbor index, and coordinate index */
920 j_coord_offsetA = DIM*jnrA;
921 j_coord_offsetB = DIM*jnrB;
922 j_coord_offsetC = DIM*jnrC;
923 j_coord_offsetD = DIM*jnrD;
925 /* load j atom coordinates */
926 gmx_mm256_load_1rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
927 x+j_coord_offsetC,x+j_coord_offsetD,
930 /* Calculate displacement vector */
931 dx00 = _mm256_sub_pd(ix0,jx0);
932 dy00 = _mm256_sub_pd(iy0,jy0);
933 dz00 = _mm256_sub_pd(iz0,jz0);
934 dx10 = _mm256_sub_pd(ix1,jx0);
935 dy10 = _mm256_sub_pd(iy1,jy0);
936 dz10 = _mm256_sub_pd(iz1,jz0);
937 dx20 = _mm256_sub_pd(ix2,jx0);
938 dy20 = _mm256_sub_pd(iy2,jy0);
939 dz20 = _mm256_sub_pd(iz2,jz0);
940 dx30 = _mm256_sub_pd(ix3,jx0);
941 dy30 = _mm256_sub_pd(iy3,jy0);
942 dz30 = _mm256_sub_pd(iz3,jz0);
944 /* Calculate squared distance and things based on it */
945 rsq00 = gmx_mm256_calc_rsq_pd(dx00,dy00,dz00);
946 rsq10 = gmx_mm256_calc_rsq_pd(dx10,dy10,dz10);
947 rsq20 = gmx_mm256_calc_rsq_pd(dx20,dy20,dz20);
948 rsq30 = gmx_mm256_calc_rsq_pd(dx30,dy30,dz30);
950 rinv00 = gmx_mm256_invsqrt_pd(rsq00);
951 rinv10 = gmx_mm256_invsqrt_pd(rsq10);
952 rinv20 = gmx_mm256_invsqrt_pd(rsq20);
953 rinv30 = gmx_mm256_invsqrt_pd(rsq30);
955 rinvsq10 = _mm256_mul_pd(rinv10,rinv10);
956 rinvsq20 = _mm256_mul_pd(rinv20,rinv20);
957 rinvsq30 = _mm256_mul_pd(rinv30,rinv30);
959 /* Load parameters for j particles */
960 jq0 = gmx_mm256_load_4real_swizzle_pd(charge+jnrA+0,charge+jnrB+0,
961 charge+jnrC+0,charge+jnrD+0);
962 vdwjidx0A = 2*vdwtype[jnrA+0];
963 vdwjidx0B = 2*vdwtype[jnrB+0];
964 vdwjidx0C = 2*vdwtype[jnrC+0];
965 vdwjidx0D = 2*vdwtype[jnrD+0];
967 fjx0 = _mm256_setzero_pd();
968 fjy0 = _mm256_setzero_pd();
969 fjz0 = _mm256_setzero_pd();
971 /**************************
972 * CALCULATE INTERACTIONS *
973 **************************/
975 r00 = _mm256_mul_pd(rsq00,rinv00);
977 /* Compute parameters for interactions between i and j atoms */
978 gmx_mm256_load_4pair_swizzle_pd(vdwioffsetptr0+vdwjidx0A,
979 vdwioffsetptr0+vdwjidx0B,
980 vdwioffsetptr0+vdwjidx0C,
981 vdwioffsetptr0+vdwjidx0D,
984 /* Calculate table index by multiplying r with table scale and truncate to integer */
985 rt = _mm256_mul_pd(r00,vftabscale);
986 vfitab = _mm256_cvttpd_epi32(rt);
987 vfeps = _mm256_sub_pd(rt,_mm256_round_pd(rt, _MM_FROUND_FLOOR));
988 vfitab = _mm_slli_epi32(vfitab,3);
990 /* CUBIC SPLINE TABLE DISPERSION */
991 Y = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
992 F = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
993 G = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
994 H = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
995 GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
996 Heps = _mm256_mul_pd(vfeps,H);
997 Fp = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
998 FF = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
999 fvdw6 = _mm256_mul_pd(c6_00,FF);
1001 /* CUBIC SPLINE TABLE REPULSION */
1002 vfitab = _mm_add_epi32(vfitab,ifour);
1003 Y = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
1004 F = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
1005 G = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
1006 H = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
1007 GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
1008 Heps = _mm256_mul_pd(vfeps,H);
1009 Fp = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
1010 FF = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
1011 fvdw12 = _mm256_mul_pd(c12_00,FF);
1012 fvdw = _mm256_xor_pd(signbit,_mm256_mul_pd(_mm256_add_pd(fvdw6,fvdw12),_mm256_mul_pd(vftabscale,rinv00)));
1016 /* Calculate temporary vectorial force */
1017 tx = _mm256_mul_pd(fscal,dx00);
1018 ty = _mm256_mul_pd(fscal,dy00);
1019 tz = _mm256_mul_pd(fscal,dz00);
1021 /* Update vectorial force */
1022 fix0 = _mm256_add_pd(fix0,tx);
1023 fiy0 = _mm256_add_pd(fiy0,ty);
1024 fiz0 = _mm256_add_pd(fiz0,tz);
1026 fjx0 = _mm256_add_pd(fjx0,tx);
1027 fjy0 = _mm256_add_pd(fjy0,ty);
1028 fjz0 = _mm256_add_pd(fjz0,tz);
1030 /**************************
1031 * CALCULATE INTERACTIONS *
1032 **************************/
1034 if (gmx_mm256_any_lt(rsq10,rcutoff2))
1037 /* Compute parameters for interactions between i and j atoms */
1038 qq10 = _mm256_mul_pd(iq1,jq0);
1040 /* REACTION-FIELD ELECTROSTATICS */
1041 felec = _mm256_mul_pd(qq10,_mm256_sub_pd(_mm256_mul_pd(rinv10,rinvsq10),krf2));
1043 cutoff_mask = _mm256_cmp_pd(rsq10,rcutoff2,_CMP_LT_OQ);
1047 fscal = _mm256_and_pd(fscal,cutoff_mask);
1049 /* Calculate temporary vectorial force */
1050 tx = _mm256_mul_pd(fscal,dx10);
1051 ty = _mm256_mul_pd(fscal,dy10);
1052 tz = _mm256_mul_pd(fscal,dz10);
1054 /* Update vectorial force */
1055 fix1 = _mm256_add_pd(fix1,tx);
1056 fiy1 = _mm256_add_pd(fiy1,ty);
1057 fiz1 = _mm256_add_pd(fiz1,tz);
1059 fjx0 = _mm256_add_pd(fjx0,tx);
1060 fjy0 = _mm256_add_pd(fjy0,ty);
1061 fjz0 = _mm256_add_pd(fjz0,tz);
1065 /**************************
1066 * CALCULATE INTERACTIONS *
1067 **************************/
1069 if (gmx_mm256_any_lt(rsq20,rcutoff2))
1072 /* Compute parameters for interactions between i and j atoms */
1073 qq20 = _mm256_mul_pd(iq2,jq0);
1075 /* REACTION-FIELD ELECTROSTATICS */
1076 felec = _mm256_mul_pd(qq20,_mm256_sub_pd(_mm256_mul_pd(rinv20,rinvsq20),krf2));
1078 cutoff_mask = _mm256_cmp_pd(rsq20,rcutoff2,_CMP_LT_OQ);
1082 fscal = _mm256_and_pd(fscal,cutoff_mask);
1084 /* Calculate temporary vectorial force */
1085 tx = _mm256_mul_pd(fscal,dx20);
1086 ty = _mm256_mul_pd(fscal,dy20);
1087 tz = _mm256_mul_pd(fscal,dz20);
1089 /* Update vectorial force */
1090 fix2 = _mm256_add_pd(fix2,tx);
1091 fiy2 = _mm256_add_pd(fiy2,ty);
1092 fiz2 = _mm256_add_pd(fiz2,tz);
1094 fjx0 = _mm256_add_pd(fjx0,tx);
1095 fjy0 = _mm256_add_pd(fjy0,ty);
1096 fjz0 = _mm256_add_pd(fjz0,tz);
1100 /**************************
1101 * CALCULATE INTERACTIONS *
1102 **************************/
1104 if (gmx_mm256_any_lt(rsq30,rcutoff2))
1107 /* Compute parameters for interactions between i and j atoms */
1108 qq30 = _mm256_mul_pd(iq3,jq0);
1110 /* REACTION-FIELD ELECTROSTATICS */
1111 felec = _mm256_mul_pd(qq30,_mm256_sub_pd(_mm256_mul_pd(rinv30,rinvsq30),krf2));
1113 cutoff_mask = _mm256_cmp_pd(rsq30,rcutoff2,_CMP_LT_OQ);
1117 fscal = _mm256_and_pd(fscal,cutoff_mask);
1119 /* Calculate temporary vectorial force */
1120 tx = _mm256_mul_pd(fscal,dx30);
1121 ty = _mm256_mul_pd(fscal,dy30);
1122 tz = _mm256_mul_pd(fscal,dz30);
1124 /* Update vectorial force */
1125 fix3 = _mm256_add_pd(fix3,tx);
1126 fiy3 = _mm256_add_pd(fiy3,ty);
1127 fiz3 = _mm256_add_pd(fiz3,tz);
1129 fjx0 = _mm256_add_pd(fjx0,tx);
1130 fjy0 = _mm256_add_pd(fjy0,ty);
1131 fjz0 = _mm256_add_pd(fjz0,tz);
1135 fjptrA = f+j_coord_offsetA;
1136 fjptrB = f+j_coord_offsetB;
1137 fjptrC = f+j_coord_offsetC;
1138 fjptrD = f+j_coord_offsetD;
1140 gmx_mm256_decrement_1rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
1142 /* Inner loop uses 141 flops */
1145 if(jidx<j_index_end)
1148 /* Get j neighbor index, and coordinate index */
1149 jnrlistA = jjnr[jidx];
1150 jnrlistB = jjnr[jidx+1];
1151 jnrlistC = jjnr[jidx+2];
1152 jnrlistD = jjnr[jidx+3];
1153 /* Sign of each element will be negative for non-real atoms.
1154 * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
1155 * so use it as val = _mm_andnot_pd(mask,val) to clear dummy entries.
1157 tmpmask0 = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
1159 tmpmask1 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(3,3,2,2));
1160 tmpmask0 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(1,1,0,0));
1161 dummy_mask = _mm256_castps_pd(gmx_mm256_set_m128(tmpmask1,tmpmask0));
1163 jnrA = (jnrlistA>=0) ? jnrlistA : 0;
1164 jnrB = (jnrlistB>=0) ? jnrlistB : 0;
1165 jnrC = (jnrlistC>=0) ? jnrlistC : 0;
1166 jnrD = (jnrlistD>=0) ? jnrlistD : 0;
1167 j_coord_offsetA = DIM*jnrA;
1168 j_coord_offsetB = DIM*jnrB;
1169 j_coord_offsetC = DIM*jnrC;
1170 j_coord_offsetD = DIM*jnrD;
1172 /* load j atom coordinates */
1173 gmx_mm256_load_1rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
1174 x+j_coord_offsetC,x+j_coord_offsetD,
1177 /* Calculate displacement vector */
1178 dx00 = _mm256_sub_pd(ix0,jx0);
1179 dy00 = _mm256_sub_pd(iy0,jy0);
1180 dz00 = _mm256_sub_pd(iz0,jz0);
1181 dx10 = _mm256_sub_pd(ix1,jx0);
1182 dy10 = _mm256_sub_pd(iy1,jy0);
1183 dz10 = _mm256_sub_pd(iz1,jz0);
1184 dx20 = _mm256_sub_pd(ix2,jx0);
1185 dy20 = _mm256_sub_pd(iy2,jy0);
1186 dz20 = _mm256_sub_pd(iz2,jz0);
1187 dx30 = _mm256_sub_pd(ix3,jx0);
1188 dy30 = _mm256_sub_pd(iy3,jy0);
1189 dz30 = _mm256_sub_pd(iz3,jz0);
1191 /* Calculate squared distance and things based on it */
1192 rsq00 = gmx_mm256_calc_rsq_pd(dx00,dy00,dz00);
1193 rsq10 = gmx_mm256_calc_rsq_pd(dx10,dy10,dz10);
1194 rsq20 = gmx_mm256_calc_rsq_pd(dx20,dy20,dz20);
1195 rsq30 = gmx_mm256_calc_rsq_pd(dx30,dy30,dz30);
1197 rinv00 = gmx_mm256_invsqrt_pd(rsq00);
1198 rinv10 = gmx_mm256_invsqrt_pd(rsq10);
1199 rinv20 = gmx_mm256_invsqrt_pd(rsq20);
1200 rinv30 = gmx_mm256_invsqrt_pd(rsq30);
1202 rinvsq10 = _mm256_mul_pd(rinv10,rinv10);
1203 rinvsq20 = _mm256_mul_pd(rinv20,rinv20);
1204 rinvsq30 = _mm256_mul_pd(rinv30,rinv30);
1206 /* Load parameters for j particles */
1207 jq0 = gmx_mm256_load_4real_swizzle_pd(charge+jnrA+0,charge+jnrB+0,
1208 charge+jnrC+0,charge+jnrD+0);
1209 vdwjidx0A = 2*vdwtype[jnrA+0];
1210 vdwjidx0B = 2*vdwtype[jnrB+0];
1211 vdwjidx0C = 2*vdwtype[jnrC+0];
1212 vdwjidx0D = 2*vdwtype[jnrD+0];
1214 fjx0 = _mm256_setzero_pd();
1215 fjy0 = _mm256_setzero_pd();
1216 fjz0 = _mm256_setzero_pd();
1218 /**************************
1219 * CALCULATE INTERACTIONS *
1220 **************************/
1222 r00 = _mm256_mul_pd(rsq00,rinv00);
1223 r00 = _mm256_andnot_pd(dummy_mask,r00);
1225 /* Compute parameters for interactions between i and j atoms */
1226 gmx_mm256_load_4pair_swizzle_pd(vdwioffsetptr0+vdwjidx0A,
1227 vdwioffsetptr0+vdwjidx0B,
1228 vdwioffsetptr0+vdwjidx0C,
1229 vdwioffsetptr0+vdwjidx0D,
1232 /* Calculate table index by multiplying r with table scale and truncate to integer */
1233 rt = _mm256_mul_pd(r00,vftabscale);
1234 vfitab = _mm256_cvttpd_epi32(rt);
1235 vfeps = _mm256_sub_pd(rt,_mm256_round_pd(rt, _MM_FROUND_FLOOR));
1236 vfitab = _mm_slli_epi32(vfitab,3);
1238 /* CUBIC SPLINE TABLE DISPERSION */
1239 Y = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
1240 F = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
1241 G = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
1242 H = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
1243 GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
1244 Heps = _mm256_mul_pd(vfeps,H);
1245 Fp = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
1246 FF = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
1247 fvdw6 = _mm256_mul_pd(c6_00,FF);
1249 /* CUBIC SPLINE TABLE REPULSION */
1250 vfitab = _mm_add_epi32(vfitab,ifour);
1251 Y = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
1252 F = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
1253 G = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
1254 H = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
1255 GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
1256 Heps = _mm256_mul_pd(vfeps,H);
1257 Fp = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
1258 FF = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
1259 fvdw12 = _mm256_mul_pd(c12_00,FF);
1260 fvdw = _mm256_xor_pd(signbit,_mm256_mul_pd(_mm256_add_pd(fvdw6,fvdw12),_mm256_mul_pd(vftabscale,rinv00)));
1264 fscal = _mm256_andnot_pd(dummy_mask,fscal);
1266 /* Calculate temporary vectorial force */
1267 tx = _mm256_mul_pd(fscal,dx00);
1268 ty = _mm256_mul_pd(fscal,dy00);
1269 tz = _mm256_mul_pd(fscal,dz00);
1271 /* Update vectorial force */
1272 fix0 = _mm256_add_pd(fix0,tx);
1273 fiy0 = _mm256_add_pd(fiy0,ty);
1274 fiz0 = _mm256_add_pd(fiz0,tz);
1276 fjx0 = _mm256_add_pd(fjx0,tx);
1277 fjy0 = _mm256_add_pd(fjy0,ty);
1278 fjz0 = _mm256_add_pd(fjz0,tz);
1280 /**************************
1281 * CALCULATE INTERACTIONS *
1282 **************************/
1284 if (gmx_mm256_any_lt(rsq10,rcutoff2))
1287 /* Compute parameters for interactions between i and j atoms */
1288 qq10 = _mm256_mul_pd(iq1,jq0);
1290 /* REACTION-FIELD ELECTROSTATICS */
1291 felec = _mm256_mul_pd(qq10,_mm256_sub_pd(_mm256_mul_pd(rinv10,rinvsq10),krf2));
1293 cutoff_mask = _mm256_cmp_pd(rsq10,rcutoff2,_CMP_LT_OQ);
1297 fscal = _mm256_and_pd(fscal,cutoff_mask);
1299 fscal = _mm256_andnot_pd(dummy_mask,fscal);
1301 /* Calculate temporary vectorial force */
1302 tx = _mm256_mul_pd(fscal,dx10);
1303 ty = _mm256_mul_pd(fscal,dy10);
1304 tz = _mm256_mul_pd(fscal,dz10);
1306 /* Update vectorial force */
1307 fix1 = _mm256_add_pd(fix1,tx);
1308 fiy1 = _mm256_add_pd(fiy1,ty);
1309 fiz1 = _mm256_add_pd(fiz1,tz);
1311 fjx0 = _mm256_add_pd(fjx0,tx);
1312 fjy0 = _mm256_add_pd(fjy0,ty);
1313 fjz0 = _mm256_add_pd(fjz0,tz);
1317 /**************************
1318 * CALCULATE INTERACTIONS *
1319 **************************/
1321 if (gmx_mm256_any_lt(rsq20,rcutoff2))
1324 /* Compute parameters for interactions between i and j atoms */
1325 qq20 = _mm256_mul_pd(iq2,jq0);
1327 /* REACTION-FIELD ELECTROSTATICS */
1328 felec = _mm256_mul_pd(qq20,_mm256_sub_pd(_mm256_mul_pd(rinv20,rinvsq20),krf2));
1330 cutoff_mask = _mm256_cmp_pd(rsq20,rcutoff2,_CMP_LT_OQ);
1334 fscal = _mm256_and_pd(fscal,cutoff_mask);
1336 fscal = _mm256_andnot_pd(dummy_mask,fscal);
1338 /* Calculate temporary vectorial force */
1339 tx = _mm256_mul_pd(fscal,dx20);
1340 ty = _mm256_mul_pd(fscal,dy20);
1341 tz = _mm256_mul_pd(fscal,dz20);
1343 /* Update vectorial force */
1344 fix2 = _mm256_add_pd(fix2,tx);
1345 fiy2 = _mm256_add_pd(fiy2,ty);
1346 fiz2 = _mm256_add_pd(fiz2,tz);
1348 fjx0 = _mm256_add_pd(fjx0,tx);
1349 fjy0 = _mm256_add_pd(fjy0,ty);
1350 fjz0 = _mm256_add_pd(fjz0,tz);
1354 /**************************
1355 * CALCULATE INTERACTIONS *
1356 **************************/
1358 if (gmx_mm256_any_lt(rsq30,rcutoff2))
1361 /* Compute parameters for interactions between i and j atoms */
1362 qq30 = _mm256_mul_pd(iq3,jq0);
1364 /* REACTION-FIELD ELECTROSTATICS */
1365 felec = _mm256_mul_pd(qq30,_mm256_sub_pd(_mm256_mul_pd(rinv30,rinvsq30),krf2));
1367 cutoff_mask = _mm256_cmp_pd(rsq30,rcutoff2,_CMP_LT_OQ);
1371 fscal = _mm256_and_pd(fscal,cutoff_mask);
1373 fscal = _mm256_andnot_pd(dummy_mask,fscal);
1375 /* Calculate temporary vectorial force */
1376 tx = _mm256_mul_pd(fscal,dx30);
1377 ty = _mm256_mul_pd(fscal,dy30);
1378 tz = _mm256_mul_pd(fscal,dz30);
1380 /* Update vectorial force */
1381 fix3 = _mm256_add_pd(fix3,tx);
1382 fiy3 = _mm256_add_pd(fiy3,ty);
1383 fiz3 = _mm256_add_pd(fiz3,tz);
1385 fjx0 = _mm256_add_pd(fjx0,tx);
1386 fjy0 = _mm256_add_pd(fjy0,ty);
1387 fjz0 = _mm256_add_pd(fjz0,tz);
1391 fjptrA = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
1392 fjptrB = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
1393 fjptrC = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
1394 fjptrD = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
1396 gmx_mm256_decrement_1rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
1398 /* Inner loop uses 142 flops */
1401 /* End of innermost loop */
1403 gmx_mm256_update_iforce_4atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
1404 f+i_coord_offset,fshift+i_shift_offset);
1406 /* Increment number of inner iterations */
1407 inneriter += j_index_end - j_index_start;
1409 /* Outer loop uses 24 flops */
1412 /* Increment number of outer iterations */
1415 /* Update outer/inner flops */
1417 inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W4_F,outeriter*24 + inneriter*142);