src/gromacs/gmxlib/nonbonded/nb_kernel_avx_256_single/nb_kernel_ElecRFCut_VdwNone_GeomW3P1_avx_256_single.c

   1 /*
   2  * This file is part of the GROMACS molecular simulation package.
   3  *
   4  * Copyright (c) 2012,2013,2014, by the GROMACS development team, led by
   5  * Mark Abraham, David van der Spoel, Berk Hess, and Erik Lindahl,
   6  * and including many others, as listed in the AUTHORS file in the
   7  * top-level source directory and at http://www.gromacs.org.
   8  *
   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.
  13  *
  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.
  18  *
  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.
  23  *
  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.
  31  *
  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.
  34  */
  35 /*
  36  * Note: this file was generated by the GROMACS avx_256_single kernel generator.
  37  */
  38 #include "config.h"
  39
  40 #include <math.h>
  41
  42 #include "../nb_kernel.h"
  43 #include "types/simple.h"
  44 #include "gromacs/math/vec.h"
  45 #include "nrnb.h"
  46
  47 #include "gromacs/simd/math_x86_avx_256_single.h"
  48 #include "kernelutil_x86_avx_256_single.h"
  49
  50 /*
  51  * Gromacs nonbonded kernel:   nb_kernel_ElecRFCut_VdwNone_GeomW3P1_VF_avx_256_single
  52  * Electrostatics interaction: ReactionField
  53  * VdW interaction:            None
  54  * Geometry:                   Water3-Particle
  55  * Calculate force/pot:        PotentialAndForce
  56  */
  57 void
  58 nb_kernel_ElecRFCut_VdwNone_GeomW3P1_VF_avx_256_single
  59                     (t_nblist                    * gmx_restrict       nlist,
  60                      rvec                        * gmx_restrict          xx,
  61                      rvec                        * gmx_restrict          ff,
  62                      t_forcerec                  * gmx_restrict          fr,
  63                      t_mdatoms                   * gmx_restrict     mdatoms,
  64                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
  65                      t_nrnb                      * gmx_restrict        nrnb)
  66 {
  67     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
  68      * just 0 for non-waters.
  69      * Suffixes A,B,C,D,E,F,G,H refer to j loop unrolling done with AVX, e.g. for the eight different
  70      * jnr indices corresponding to data put in the four positions in the SIMD register.
  71      */
  72     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
  73     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
  74     int              jnrA,jnrB,jnrC,jnrD;
  75     int              jnrE,jnrF,jnrG,jnrH;
  76     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
  77     int              jnrlistE,jnrlistF,jnrlistG,jnrlistH;
  78     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
  79     int              j_coord_offsetE,j_coord_offsetF,j_coord_offsetG,j_coord_offsetH;
  80     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
  81     real             rcutoff_scalar;
  82     real             *shiftvec,*fshift,*x,*f;
  83     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD,*fjptrE,*fjptrF,*fjptrG,*fjptrH;
  84     real             scratch[4*DIM];
  85     __m256           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
  86     real *           vdwioffsetptr0;
  87     __m256           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
  88     real *           vdwioffsetptr1;
  89     __m256           ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
  90     real *           vdwioffsetptr2;
  91     __m256           ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
  92     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D,vdwjidx0E,vdwjidx0F,vdwjidx0G,vdwjidx0H;
  93     __m256           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  94     __m256           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  95     __m256           dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
  96     __m256           dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
  97     __m256           velec,felec,velecsum,facel,crf,krf,krf2;
  98     real             *charge;
  99     __m256           dummy_mask,cutoff_mask;
 100     __m256           signbit = _mm256_castsi256_ps( _mm256_set1_epi32(0x80000000) );
 101     __m256           one     = _mm256_set1_ps(1.0);
 102     __m256           two     = _mm256_set1_ps(2.0);
 103     x                = xx[0];
 104     f                = ff[0];
 105
 106     nri              = nlist->nri;
 107     iinr             = nlist->iinr;
 108     jindex           = nlist->jindex;
 109     jjnr             = nlist->jjnr;
 110     shiftidx         = nlist->shift;
 111     gid              = nlist->gid;
 112     shiftvec         = fr->shift_vec[0];
 113     fshift           = fr->fshift[0];
 114     facel            = _mm256_set1_ps(fr->epsfac);
 115     charge           = mdatoms->chargeA;
 116     krf              = _mm256_set1_ps(fr->ic->k_rf);
 117     krf2             = _mm256_set1_ps(fr->ic->k_rf*2.0);
 118     crf              = _mm256_set1_ps(fr->ic->c_rf);
 119
 120     /* Setup water-specific parameters */
 121     inr              = nlist->iinr[0];
 122     iq0              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+0]));
 123     iq1              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+1]));
 124     iq2              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+2]));
 125
 126     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 127     rcutoff_scalar   = fr->rcoulomb;
 128     rcutoff          = _mm256_set1_ps(rcutoff_scalar);
 129     rcutoff2         = _mm256_mul_ps(rcutoff,rcutoff);
 130
 131     /* Avoid stupid compiler warnings */
 132     jnrA = jnrB = jnrC = jnrD = jnrE = jnrF = jnrG = jnrH = 0;
 133     j_coord_offsetA = 0;
 134     j_coord_offsetB = 0;
 135     j_coord_offsetC = 0;
 136     j_coord_offsetD = 0;
 137     j_coord_offsetE = 0;
 138     j_coord_offsetF = 0;
 139     j_coord_offsetG = 0;
 140     j_coord_offsetH = 0;
 141
 142     outeriter        = 0;
 143     inneriter        = 0;
 144
 145     for(iidx=0;iidx<4*DIM;iidx++)
 146     {
 147         scratch[iidx] = 0.0;
 148     }
 149
 150     /* Start outer loop over neighborlists */
 151     for(iidx=0; iidx<nri; iidx++)
 152     {
 153         /* Load shift vector for this list */
 154         i_shift_offset   = DIM*shiftidx[iidx];
 155
 156         /* Load limits for loop over neighbors */
 157         j_index_start    = jindex[iidx];
 158         j_index_end      = jindex[iidx+1];
 159
 160         /* Get outer coordinate index */
 161         inr              = iinr[iidx];
 162         i_coord_offset   = DIM*inr;
 163
 164         /* Load i particle coords and add shift vector */
 165         gmx_mm256_load_shift_and_3rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
 166                                                     &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 167
 168         fix0             = _mm256_setzero_ps();
 169         fiy0             = _mm256_setzero_ps();
 170         fiz0             = _mm256_setzero_ps();
 171         fix1             = _mm256_setzero_ps();
 172         fiy1             = _mm256_setzero_ps();
 173         fiz1             = _mm256_setzero_ps();
 174         fix2             = _mm256_setzero_ps();
 175         fiy2             = _mm256_setzero_ps();
 176         fiz2             = _mm256_setzero_ps();
 177
 178         /* Reset potential sums */
 179         velecsum         = _mm256_setzero_ps();
 180
 181         /* Start inner kernel loop */
 182         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+7]>=0; jidx+=8)
 183         {
 184
 185             /* Get j neighbor index, and coordinate index */
 186             jnrA             = jjnr[jidx];
 187             jnrB             = jjnr[jidx+1];
 188             jnrC             = jjnr[jidx+2];
 189             jnrD             = jjnr[jidx+3];
 190             jnrE             = jjnr[jidx+4];
 191             jnrF             = jjnr[jidx+5];
 192             jnrG             = jjnr[jidx+6];
 193             jnrH             = jjnr[jidx+7];
 194             j_coord_offsetA  = DIM*jnrA;
 195             j_coord_offsetB  = DIM*jnrB;
 196             j_coord_offsetC  = DIM*jnrC;
 197             j_coord_offsetD  = DIM*jnrD;
 198             j_coord_offsetE  = DIM*jnrE;
 199             j_coord_offsetF  = DIM*jnrF;
 200             j_coord_offsetG  = DIM*jnrG;
 201             j_coord_offsetH  = DIM*jnrH;
 202
 203             /* load j atom coordinates */
 204             gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 205                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 206                                                  x+j_coord_offsetE,x+j_coord_offsetF,
 207                                                  x+j_coord_offsetG,x+j_coord_offsetH,
 208                                                  &jx0,&jy0,&jz0);
 209
 210             /* Calculate displacement vector */
 211             dx00             = _mm256_sub_ps(ix0,jx0);
 212             dy00             = _mm256_sub_ps(iy0,jy0);
 213             dz00             = _mm256_sub_ps(iz0,jz0);
 214             dx10             = _mm256_sub_ps(ix1,jx0);
 215             dy10             = _mm256_sub_ps(iy1,jy0);
 216             dz10             = _mm256_sub_ps(iz1,jz0);
 217             dx20             = _mm256_sub_ps(ix2,jx0);
 218             dy20             = _mm256_sub_ps(iy2,jy0);
 219             dz20             = _mm256_sub_ps(iz2,jz0);
 220
 221             /* Calculate squared distance and things based on it */
 222             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
 223             rsq10            = gmx_mm256_calc_rsq_ps(dx10,dy10,dz10);
 224             rsq20            = gmx_mm256_calc_rsq_ps(dx20,dy20,dz20);
 225
 226             rinv00           = gmx_mm256_invsqrt_ps(rsq00);
 227             rinv10           = gmx_mm256_invsqrt_ps(rsq10);
 228             rinv20           = gmx_mm256_invsqrt_ps(rsq20);
 229
 230             rinvsq00         = _mm256_mul_ps(rinv00,rinv00);
 231             rinvsq10         = _mm256_mul_ps(rinv10,rinv10);
 232             rinvsq20         = _mm256_mul_ps(rinv20,rinv20);
 233
 234             /* Load parameters for j particles */
 235             jq0              = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 236                                                                  charge+jnrC+0,charge+jnrD+0,
 237                                                                  charge+jnrE+0,charge+jnrF+0,
 238                                                                  charge+jnrG+0,charge+jnrH+0);
 239
 240             fjx0             = _mm256_setzero_ps();
 241             fjy0             = _mm256_setzero_ps();
 242             fjz0             = _mm256_setzero_ps();
 243
 244             /**************************
 245              * CALCULATE INTERACTIONS *
 246              **************************/
 247
 248             if (gmx_mm256_any_lt(rsq00,rcutoff2))
 249             {
 250
 251             /* Compute parameters for interactions between i and j atoms */
 252             qq00             = _mm256_mul_ps(iq0,jq0);
 253
 254             /* REACTION-FIELD ELECTROSTATICS */
 255             velec            = _mm256_mul_ps(qq00,_mm256_sub_ps(_mm256_add_ps(rinv00,_mm256_mul_ps(krf,rsq00)),crf));
 256             felec            = _mm256_mul_ps(qq00,_mm256_sub_ps(_mm256_mul_ps(rinv00,rinvsq00),krf2));
 257
 258             cutoff_mask      = _mm256_cmp_ps(rsq00,rcutoff2,_CMP_LT_OQ);
 259
 260             /* Update potential sum for this i atom from the interaction with this j atom. */
 261             velec            = _mm256_and_ps(velec,cutoff_mask);
 262             velecsum         = _mm256_add_ps(velecsum,velec);
 263
 264             fscal            = felec;
 265
 266             fscal            = _mm256_and_ps(fscal,cutoff_mask);
 267
 268             /* Calculate temporary vectorial force */
 269             tx               = _mm256_mul_ps(fscal,dx00);
 270             ty               = _mm256_mul_ps(fscal,dy00);
 271             tz               = _mm256_mul_ps(fscal,dz00);
 272
 273             /* Update vectorial force */
 274             fix0             = _mm256_add_ps(fix0,tx);
 275             fiy0             = _mm256_add_ps(fiy0,ty);
 276             fiz0             = _mm256_add_ps(fiz0,tz);
 277
 278             fjx0             = _mm256_add_ps(fjx0,tx);
 279             fjy0             = _mm256_add_ps(fjy0,ty);
 280             fjz0             = _mm256_add_ps(fjz0,tz);
 281
 282             }
 283
 284             /**************************
 285              * CALCULATE INTERACTIONS *
 286              **************************/
 287
 288             if (gmx_mm256_any_lt(rsq10,rcutoff2))
 289             {
 290
 291             /* Compute parameters for interactions between i and j atoms */
 292             qq10             = _mm256_mul_ps(iq1,jq0);
 293
 294             /* REACTION-FIELD ELECTROSTATICS */
 295             velec            = _mm256_mul_ps(qq10,_mm256_sub_ps(_mm256_add_ps(rinv10,_mm256_mul_ps(krf,rsq10)),crf));
 296             felec            = _mm256_mul_ps(qq10,_mm256_sub_ps(_mm256_mul_ps(rinv10,rinvsq10),krf2));
 297
 298             cutoff_mask      = _mm256_cmp_ps(rsq10,rcutoff2,_CMP_LT_OQ);
 299
 300             /* Update potential sum for this i atom from the interaction with this j atom. */
 301             velec            = _mm256_and_ps(velec,cutoff_mask);
 302             velecsum         = _mm256_add_ps(velecsum,velec);
 303
 304             fscal            = felec;
 305
 306             fscal            = _mm256_and_ps(fscal,cutoff_mask);
 307
 308             /* Calculate temporary vectorial force */
 309             tx               = _mm256_mul_ps(fscal,dx10);
 310             ty               = _mm256_mul_ps(fscal,dy10);
 311             tz               = _mm256_mul_ps(fscal,dz10);
 312
 313             /* Update vectorial force */
 314             fix1             = _mm256_add_ps(fix1,tx);
 315             fiy1             = _mm256_add_ps(fiy1,ty);
 316             fiz1             = _mm256_add_ps(fiz1,tz);
 317
 318             fjx0             = _mm256_add_ps(fjx0,tx);
 319             fjy0             = _mm256_add_ps(fjy0,ty);
 320             fjz0             = _mm256_add_ps(fjz0,tz);
 321
 322             }
 323
 324             /**************************
 325              * CALCULATE INTERACTIONS *
 326              **************************/
 327
 328             if (gmx_mm256_any_lt(rsq20,rcutoff2))
 329             {
 330
 331             /* Compute parameters for interactions between i and j atoms */
 332             qq20             = _mm256_mul_ps(iq2,jq0);
 333
 334             /* REACTION-FIELD ELECTROSTATICS */
 335             velec            = _mm256_mul_ps(qq20,_mm256_sub_ps(_mm256_add_ps(rinv20,_mm256_mul_ps(krf,rsq20)),crf));
 336             felec            = _mm256_mul_ps(qq20,_mm256_sub_ps(_mm256_mul_ps(rinv20,rinvsq20),krf2));
 337
 338             cutoff_mask      = _mm256_cmp_ps(rsq20,rcutoff2,_CMP_LT_OQ);
 339
 340             /* Update potential sum for this i atom from the interaction with this j atom. */
 341             velec            = _mm256_and_ps(velec,cutoff_mask);
 342             velecsum         = _mm256_add_ps(velecsum,velec);
 343
 344             fscal            = felec;
 345
 346             fscal            = _mm256_and_ps(fscal,cutoff_mask);
 347
 348             /* Calculate temporary vectorial force */
 349             tx               = _mm256_mul_ps(fscal,dx20);
 350             ty               = _mm256_mul_ps(fscal,dy20);
 351             tz               = _mm256_mul_ps(fscal,dz20);
 352
 353             /* Update vectorial force */
 354             fix2             = _mm256_add_ps(fix2,tx);
 355             fiy2             = _mm256_add_ps(fiy2,ty);
 356             fiz2             = _mm256_add_ps(fiz2,tz);
 357
 358             fjx0             = _mm256_add_ps(fjx0,tx);
 359             fjy0             = _mm256_add_ps(fjy0,ty);
 360             fjz0             = _mm256_add_ps(fjz0,tz);
 361
 362             }
 363
 364             fjptrA             = f+j_coord_offsetA;
 365             fjptrB             = f+j_coord_offsetB;
 366             fjptrC             = f+j_coord_offsetC;
 367             fjptrD             = f+j_coord_offsetD;
 368             fjptrE             = f+j_coord_offsetE;
 369             fjptrF             = f+j_coord_offsetF;
 370             fjptrG             = f+j_coord_offsetG;
 371             fjptrH             = f+j_coord_offsetH;
 372
 373             gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,fjx0,fjy0,fjz0);
 374
 375             /* Inner loop uses 111 flops */
 376         }
 377
 378         if(jidx<j_index_end)
 379         {
 380
 381             /* Get j neighbor index, and coordinate index */
 382             jnrlistA         = jjnr[jidx];
 383             jnrlistB         = jjnr[jidx+1];
 384             jnrlistC         = jjnr[jidx+2];
 385             jnrlistD         = jjnr[jidx+3];
 386             jnrlistE         = jjnr[jidx+4];
 387             jnrlistF         = jjnr[jidx+5];
 388             jnrlistG         = jjnr[jidx+6];
 389             jnrlistH         = jjnr[jidx+7];
 390             /* Sign of each element will be negative for non-real atoms.
 391              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 392              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 393              */
 394             dummy_mask = gmx_mm256_set_m128(gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx+4)),_mm_setzero_si128())),
 395                                             gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128())));
 396
 397             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 398             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 399             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 400             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 401             jnrE       = (jnrlistE>=0) ? jnrlistE : 0;
 402             jnrF       = (jnrlistF>=0) ? jnrlistF : 0;
 403             jnrG       = (jnrlistG>=0) ? jnrlistG : 0;
 404             jnrH       = (jnrlistH>=0) ? jnrlistH : 0;
 405             j_coord_offsetA  = DIM*jnrA;
 406             j_coord_offsetB  = DIM*jnrB;
 407             j_coord_offsetC  = DIM*jnrC;
 408             j_coord_offsetD  = DIM*jnrD;
 409             j_coord_offsetE  = DIM*jnrE;
 410             j_coord_offsetF  = DIM*jnrF;
 411             j_coord_offsetG  = DIM*jnrG;
 412             j_coord_offsetH  = DIM*jnrH;
 413
 414             /* load j atom coordinates */
 415             gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 416                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 417                                                  x+j_coord_offsetE,x+j_coord_offsetF,
 418                                                  x+j_coord_offsetG,x+j_coord_offsetH,
 419                                                  &jx0,&jy0,&jz0);
 420
 421             /* Calculate displacement vector */
 422             dx00             = _mm256_sub_ps(ix0,jx0);
 423             dy00             = _mm256_sub_ps(iy0,jy0);
 424             dz00             = _mm256_sub_ps(iz0,jz0);
 425             dx10             = _mm256_sub_ps(ix1,jx0);
 426             dy10             = _mm256_sub_ps(iy1,jy0);
 427             dz10             = _mm256_sub_ps(iz1,jz0);
 428             dx20             = _mm256_sub_ps(ix2,jx0);
 429             dy20             = _mm256_sub_ps(iy2,jy0);
 430             dz20             = _mm256_sub_ps(iz2,jz0);
 431
 432             /* Calculate squared distance and things based on it */
 433             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
 434             rsq10            = gmx_mm256_calc_rsq_ps(dx10,dy10,dz10);
 435             rsq20            = gmx_mm256_calc_rsq_ps(dx20,dy20,dz20);
 436
 437             rinv00           = gmx_mm256_invsqrt_ps(rsq00);
 438             rinv10           = gmx_mm256_invsqrt_ps(rsq10);
 439             rinv20           = gmx_mm256_invsqrt_ps(rsq20);
 440
 441             rinvsq00         = _mm256_mul_ps(rinv00,rinv00);
 442             rinvsq10         = _mm256_mul_ps(rinv10,rinv10);
 443             rinvsq20         = _mm256_mul_ps(rinv20,rinv20);
 444
 445             /* Load parameters for j particles */
 446             jq0              = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 447                                                                  charge+jnrC+0,charge+jnrD+0,
 448                                                                  charge+jnrE+0,charge+jnrF+0,
 449                                                                  charge+jnrG+0,charge+jnrH+0);
 450
 451             fjx0             = _mm256_setzero_ps();
 452             fjy0             = _mm256_setzero_ps();
 453             fjz0             = _mm256_setzero_ps();
 454
 455             /**************************
 456              * CALCULATE INTERACTIONS *
 457              **************************/
 458
 459             if (gmx_mm256_any_lt(rsq00,rcutoff2))
 460             {
 461
 462             /* Compute parameters for interactions between i and j atoms */
 463             qq00             = _mm256_mul_ps(iq0,jq0);
 464
 465             /* REACTION-FIELD ELECTROSTATICS */
 466             velec            = _mm256_mul_ps(qq00,_mm256_sub_ps(_mm256_add_ps(rinv00,_mm256_mul_ps(krf,rsq00)),crf));
 467             felec            = _mm256_mul_ps(qq00,_mm256_sub_ps(_mm256_mul_ps(rinv00,rinvsq00),krf2));
 468
 469             cutoff_mask      = _mm256_cmp_ps(rsq00,rcutoff2,_CMP_LT_OQ);
 470
 471             /* Update potential sum for this i atom from the interaction with this j atom. */
 472             velec            = _mm256_and_ps(velec,cutoff_mask);
 473             velec            = _mm256_andnot_ps(dummy_mask,velec);
 474             velecsum         = _mm256_add_ps(velecsum,velec);
 475
 476             fscal            = felec;
 477
 478             fscal            = _mm256_and_ps(fscal,cutoff_mask);
 479
 480             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
 481
 482             /* Calculate temporary vectorial force */
 483             tx               = _mm256_mul_ps(fscal,dx00);
 484             ty               = _mm256_mul_ps(fscal,dy00);
 485             tz               = _mm256_mul_ps(fscal,dz00);
 486
 487             /* Update vectorial force */
 488             fix0             = _mm256_add_ps(fix0,tx);
 489             fiy0             = _mm256_add_ps(fiy0,ty);
 490             fiz0             = _mm256_add_ps(fiz0,tz);
 491
 492             fjx0             = _mm256_add_ps(fjx0,tx);
 493             fjy0             = _mm256_add_ps(fjy0,ty);
 494             fjz0             = _mm256_add_ps(fjz0,tz);
 495
 496             }
 497
 498             /**************************
 499              * CALCULATE INTERACTIONS *
 500              **************************/
 501
 502             if (gmx_mm256_any_lt(rsq10,rcutoff2))
 503             {
 504
 505             /* Compute parameters for interactions between i and j atoms */
 506             qq10             = _mm256_mul_ps(iq1,jq0);
 507
 508             /* REACTION-FIELD ELECTROSTATICS */
 509             velec            = _mm256_mul_ps(qq10,_mm256_sub_ps(_mm256_add_ps(rinv10,_mm256_mul_ps(krf,rsq10)),crf));
 510             felec            = _mm256_mul_ps(qq10,_mm256_sub_ps(_mm256_mul_ps(rinv10,rinvsq10),krf2));
 511
 512             cutoff_mask      = _mm256_cmp_ps(rsq10,rcutoff2,_CMP_LT_OQ);
 513
 514             /* Update potential sum for this i atom from the interaction with this j atom. */
 515             velec            = _mm256_and_ps(velec,cutoff_mask);
 516             velec            = _mm256_andnot_ps(dummy_mask,velec);
 517             velecsum         = _mm256_add_ps(velecsum,velec);
 518
 519             fscal            = felec;
 520
 521             fscal            = _mm256_and_ps(fscal,cutoff_mask);
 522
 523             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
 524
 525             /* Calculate temporary vectorial force */
 526             tx               = _mm256_mul_ps(fscal,dx10);
 527             ty               = _mm256_mul_ps(fscal,dy10);
 528             tz               = _mm256_mul_ps(fscal,dz10);
 529
 530             /* Update vectorial force */
 531             fix1             = _mm256_add_ps(fix1,tx);
 532             fiy1             = _mm256_add_ps(fiy1,ty);
 533             fiz1             = _mm256_add_ps(fiz1,tz);
 534
 535             fjx0             = _mm256_add_ps(fjx0,tx);
 536             fjy0             = _mm256_add_ps(fjy0,ty);
 537             fjz0             = _mm256_add_ps(fjz0,tz);
 538
 539             }
 540
 541             /**************************
 542              * CALCULATE INTERACTIONS *
 543              **************************/
 544
 545             if (gmx_mm256_any_lt(rsq20,rcutoff2))
 546             {
 547
 548             /* Compute parameters for interactions between i and j atoms */
 549             qq20             = _mm256_mul_ps(iq2,jq0);
 550
 551             /* REACTION-FIELD ELECTROSTATICS */
 552             velec            = _mm256_mul_ps(qq20,_mm256_sub_ps(_mm256_add_ps(rinv20,_mm256_mul_ps(krf,rsq20)),crf));
 553             felec            = _mm256_mul_ps(qq20,_mm256_sub_ps(_mm256_mul_ps(rinv20,rinvsq20),krf2));
 554
 555             cutoff_mask      = _mm256_cmp_ps(rsq20,rcutoff2,_CMP_LT_OQ);
 556
 557             /* Update potential sum for this i atom from the interaction with this j atom. */
 558             velec            = _mm256_and_ps(velec,cutoff_mask);
 559             velec            = _mm256_andnot_ps(dummy_mask,velec);
 560             velecsum         = _mm256_add_ps(velecsum,velec);
 561
 562             fscal            = felec;
 563
 564             fscal            = _mm256_and_ps(fscal,cutoff_mask);
 565
 566             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
 567
 568             /* Calculate temporary vectorial force */
 569             tx               = _mm256_mul_ps(fscal,dx20);
 570             ty               = _mm256_mul_ps(fscal,dy20);
 571             tz               = _mm256_mul_ps(fscal,dz20);
 572
 573             /* Update vectorial force */
 574             fix2             = _mm256_add_ps(fix2,tx);
 575             fiy2             = _mm256_add_ps(fiy2,ty);
 576             fiz2             = _mm256_add_ps(fiz2,tz);
 577
 578             fjx0             = _mm256_add_ps(fjx0,tx);
 579             fjy0             = _mm256_add_ps(fjy0,ty);
 580             fjz0             = _mm256_add_ps(fjz0,tz);
 581
 582             }
 583
 584             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 585             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 586             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 587             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 588             fjptrE             = (jnrlistE>=0) ? f+j_coord_offsetE : scratch;
 589             fjptrF             = (jnrlistF>=0) ? f+j_coord_offsetF : scratch;
 590             fjptrG             = (jnrlistG>=0) ? f+j_coord_offsetG : scratch;
 591             fjptrH             = (jnrlistH>=0) ? f+j_coord_offsetH : scratch;
 592
 593             gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,fjx0,fjy0,fjz0);
 594
 595             /* Inner loop uses 111 flops */
 596         }
 597
 598         /* End of innermost loop */
 599
 600         gmx_mm256_update_iforce_3atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
 601                                                  f+i_coord_offset,fshift+i_shift_offset);
 602
 603         ggid                        = gid[iidx];
 604         /* Update potential energies */
 605         gmx_mm256_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid);
 606
 607         /* Increment number of inner iterations */
 608         inneriter                  += j_index_end - j_index_start;
 609
 610         /* Outer loop uses 19 flops */
 611     }
 612
 613     /* Increment number of outer iterations */
 614     outeriter        += nri;
 615
 616     /* Update outer/inner flops */
 617
 618     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_W3_VF,outeriter*19 + inneriter*111);
 619 }
 620 /*
 621  * Gromacs nonbonded kernel:   nb_kernel_ElecRFCut_VdwNone_GeomW3P1_F_avx_256_single
 622  * Electrostatics interaction: ReactionField
 623  * VdW interaction:            None
 624  * Geometry:                   Water3-Particle
 625  * Calculate force/pot:        Force
 626  */
 627 void
 628 nb_kernel_ElecRFCut_VdwNone_GeomW3P1_F_avx_256_single
 629                     (t_nblist                    * gmx_restrict       nlist,
 630                      rvec                        * gmx_restrict          xx,
 631                      rvec                        * gmx_restrict          ff,
 632                      t_forcerec                  * gmx_restrict          fr,
 633                      t_mdatoms                   * gmx_restrict     mdatoms,
 634                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 635                      t_nrnb                      * gmx_restrict        nrnb)
 636 {
 637     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 638      * just 0 for non-waters.
 639      * Suffixes A,B,C,D,E,F,G,H refer to j loop unrolling done with AVX, e.g. for the eight different
 640      * jnr indices corresponding to data put in the four positions in the SIMD register.
 641      */
 642     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 643     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 644     int              jnrA,jnrB,jnrC,jnrD;
 645     int              jnrE,jnrF,jnrG,jnrH;
 646     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
 647     int              jnrlistE,jnrlistF,jnrlistG,jnrlistH;
 648     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 649     int              j_coord_offsetE,j_coord_offsetF,j_coord_offsetG,j_coord_offsetH;
 650     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 651     real             rcutoff_scalar;
 652     real             *shiftvec,*fshift,*x,*f;
 653     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD,*fjptrE,*fjptrF,*fjptrG,*fjptrH;
 654     real             scratch[4*DIM];
 655     __m256           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 656     real *           vdwioffsetptr0;
 657     __m256           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 658     real *           vdwioffsetptr1;
 659     __m256           ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
 660     real *           vdwioffsetptr2;
 661     __m256           ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
 662     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D,vdwjidx0E,vdwjidx0F,vdwjidx0G,vdwjidx0H;
 663     __m256           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 664     __m256           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 665     __m256           dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
 666     __m256           dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
 667     __m256           velec,felec,velecsum,facel,crf,krf,krf2;
 668     real             *charge;
 669     __m256           dummy_mask,cutoff_mask;
 670     __m256           signbit = _mm256_castsi256_ps( _mm256_set1_epi32(0x80000000) );
 671     __m256           one     = _mm256_set1_ps(1.0);
 672     __m256           two     = _mm256_set1_ps(2.0);
 673     x                = xx[0];
 674     f                = ff[0];
 675
 676     nri              = nlist->nri;
 677     iinr             = nlist->iinr;
 678     jindex           = nlist->jindex;
 679     jjnr             = nlist->jjnr;
 680     shiftidx         = nlist->shift;
 681     gid              = nlist->gid;
 682     shiftvec         = fr->shift_vec[0];
 683     fshift           = fr->fshift[0];
 684     facel            = _mm256_set1_ps(fr->epsfac);
 685     charge           = mdatoms->chargeA;
 686     krf              = _mm256_set1_ps(fr->ic->k_rf);
 687     krf2             = _mm256_set1_ps(fr->ic->k_rf*2.0);
 688     crf              = _mm256_set1_ps(fr->ic->c_rf);
 689
 690     /* Setup water-specific parameters */
 691     inr              = nlist->iinr[0];
 692     iq0              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+0]));
 693     iq1              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+1]));
 694     iq2              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+2]));
 695
 696     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 697     rcutoff_scalar   = fr->rcoulomb;
 698     rcutoff          = _mm256_set1_ps(rcutoff_scalar);
 699     rcutoff2         = _mm256_mul_ps(rcutoff,rcutoff);
 700
 701     /* Avoid stupid compiler warnings */
 702     jnrA = jnrB = jnrC = jnrD = jnrE = jnrF = jnrG = jnrH = 0;
 703     j_coord_offsetA = 0;
 704     j_coord_offsetB = 0;
 705     j_coord_offsetC = 0;
 706     j_coord_offsetD = 0;
 707     j_coord_offsetE = 0;
 708     j_coord_offsetF = 0;
 709     j_coord_offsetG = 0;
 710     j_coord_offsetH = 0;
 711
 712     outeriter        = 0;
 713     inneriter        = 0;
 714
 715     for(iidx=0;iidx<4*DIM;iidx++)
 716     {
 717         scratch[iidx] = 0.0;
 718     }
 719
 720     /* Start outer loop over neighborlists */
 721     for(iidx=0; iidx<nri; iidx++)
 722     {
 723         /* Load shift vector for this list */
 724         i_shift_offset   = DIM*shiftidx[iidx];
 725
 726         /* Load limits for loop over neighbors */
 727         j_index_start    = jindex[iidx];
 728         j_index_end      = jindex[iidx+1];
 729
 730         /* Get outer coordinate index */
 731         inr              = iinr[iidx];
 732         i_coord_offset   = DIM*inr;
 733
 734         /* Load i particle coords and add shift vector */
 735         gmx_mm256_load_shift_and_3rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
 736                                                     &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 737
 738         fix0             = _mm256_setzero_ps();
 739         fiy0             = _mm256_setzero_ps();
 740         fiz0             = _mm256_setzero_ps();
 741         fix1             = _mm256_setzero_ps();
 742         fiy1             = _mm256_setzero_ps();
 743         fiz1             = _mm256_setzero_ps();
 744         fix2             = _mm256_setzero_ps();
 745         fiy2             = _mm256_setzero_ps();
 746         fiz2             = _mm256_setzero_ps();
 747
 748         /* Start inner kernel loop */
 749         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+7]>=0; jidx+=8)
 750         {
 751
 752             /* Get j neighbor index, and coordinate index */
 753             jnrA             = jjnr[jidx];
 754             jnrB             = jjnr[jidx+1];
 755             jnrC             = jjnr[jidx+2];
 756             jnrD             = jjnr[jidx+3];
 757             jnrE             = jjnr[jidx+4];
 758             jnrF             = jjnr[jidx+5];
 759             jnrG             = jjnr[jidx+6];
 760             jnrH             = jjnr[jidx+7];
 761             j_coord_offsetA  = DIM*jnrA;
 762             j_coord_offsetB  = DIM*jnrB;
 763             j_coord_offsetC  = DIM*jnrC;
 764             j_coord_offsetD  = DIM*jnrD;
 765             j_coord_offsetE  = DIM*jnrE;
 766             j_coord_offsetF  = DIM*jnrF;
 767             j_coord_offsetG  = DIM*jnrG;
 768             j_coord_offsetH  = DIM*jnrH;
 769
 770             /* load j atom coordinates */
 771             gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 772                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 773                                                  x+j_coord_offsetE,x+j_coord_offsetF,
 774                                                  x+j_coord_offsetG,x+j_coord_offsetH,
 775                                                  &jx0,&jy0,&jz0);
 776
 777             /* Calculate displacement vector */
 778             dx00             = _mm256_sub_ps(ix0,jx0);
 779             dy00             = _mm256_sub_ps(iy0,jy0);
 780             dz00             = _mm256_sub_ps(iz0,jz0);
 781             dx10             = _mm256_sub_ps(ix1,jx0);
 782             dy10             = _mm256_sub_ps(iy1,jy0);
 783             dz10             = _mm256_sub_ps(iz1,jz0);
 784             dx20             = _mm256_sub_ps(ix2,jx0);
 785             dy20             = _mm256_sub_ps(iy2,jy0);
 786             dz20             = _mm256_sub_ps(iz2,jz0);
 787
 788             /* Calculate squared distance and things based on it */
 789             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
 790             rsq10            = gmx_mm256_calc_rsq_ps(dx10,dy10,dz10);
 791             rsq20            = gmx_mm256_calc_rsq_ps(dx20,dy20,dz20);
 792
 793             rinv00           = gmx_mm256_invsqrt_ps(rsq00);
 794             rinv10           = gmx_mm256_invsqrt_ps(rsq10);
 795             rinv20           = gmx_mm256_invsqrt_ps(rsq20);
 796
 797             rinvsq00         = _mm256_mul_ps(rinv00,rinv00);
 798             rinvsq10         = _mm256_mul_ps(rinv10,rinv10);
 799             rinvsq20         = _mm256_mul_ps(rinv20,rinv20);
 800
 801             /* Load parameters for j particles */
 802             jq0              = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 803                                                                  charge+jnrC+0,charge+jnrD+0,
 804                                                                  charge+jnrE+0,charge+jnrF+0,
 805                                                                  charge+jnrG+0,charge+jnrH+0);
 806
 807             fjx0             = _mm256_setzero_ps();
 808             fjy0             = _mm256_setzero_ps();
 809             fjz0             = _mm256_setzero_ps();
 810
 811             /**************************
 812              * CALCULATE INTERACTIONS *
 813              **************************/
 814
 815             if (gmx_mm256_any_lt(rsq00,rcutoff2))
 816             {
 817
 818             /* Compute parameters for interactions between i and j atoms */
 819             qq00             = _mm256_mul_ps(iq0,jq0);
 820
 821             /* REACTION-FIELD ELECTROSTATICS */
 822             felec            = _mm256_mul_ps(qq00,_mm256_sub_ps(_mm256_mul_ps(rinv00,rinvsq00),krf2));
 823
 824             cutoff_mask      = _mm256_cmp_ps(rsq00,rcutoff2,_CMP_LT_OQ);
 825
 826             fscal            = felec;
 827
 828             fscal            = _mm256_and_ps(fscal,cutoff_mask);
 829
 830             /* Calculate temporary vectorial force */
 831             tx               = _mm256_mul_ps(fscal,dx00);
 832             ty               = _mm256_mul_ps(fscal,dy00);
 833             tz               = _mm256_mul_ps(fscal,dz00);
 834
 835             /* Update vectorial force */
 836             fix0             = _mm256_add_ps(fix0,tx);
 837             fiy0             = _mm256_add_ps(fiy0,ty);
 838             fiz0             = _mm256_add_ps(fiz0,tz);
 839
 840             fjx0             = _mm256_add_ps(fjx0,tx);
 841             fjy0             = _mm256_add_ps(fjy0,ty);
 842             fjz0             = _mm256_add_ps(fjz0,tz);
 843
 844             }
 845
 846             /**************************
 847              * CALCULATE INTERACTIONS *
 848              **************************/
 849
 850             if (gmx_mm256_any_lt(rsq10,rcutoff2))
 851             {
 852
 853             /* Compute parameters for interactions between i and j atoms */
 854             qq10             = _mm256_mul_ps(iq1,jq0);
 855
 856             /* REACTION-FIELD ELECTROSTATICS */
 857             felec            = _mm256_mul_ps(qq10,_mm256_sub_ps(_mm256_mul_ps(rinv10,rinvsq10),krf2));
 858
 859             cutoff_mask      = _mm256_cmp_ps(rsq10,rcutoff2,_CMP_LT_OQ);
 860
 861             fscal            = felec;
 862
 863             fscal            = _mm256_and_ps(fscal,cutoff_mask);
 864
 865             /* Calculate temporary vectorial force */
 866             tx               = _mm256_mul_ps(fscal,dx10);
 867             ty               = _mm256_mul_ps(fscal,dy10);
 868             tz               = _mm256_mul_ps(fscal,dz10);
 869
 870             /* Update vectorial force */
 871             fix1             = _mm256_add_ps(fix1,tx);
 872             fiy1             = _mm256_add_ps(fiy1,ty);
 873             fiz1             = _mm256_add_ps(fiz1,tz);
 874
 875             fjx0             = _mm256_add_ps(fjx0,tx);
 876             fjy0             = _mm256_add_ps(fjy0,ty);
 877             fjz0             = _mm256_add_ps(fjz0,tz);
 878
 879             }
 880
 881             /**************************
 882              * CALCULATE INTERACTIONS *
 883              **************************/
 884
 885             if (gmx_mm256_any_lt(rsq20,rcutoff2))
 886             {
 887
 888             /* Compute parameters for interactions between i and j atoms */
 889             qq20             = _mm256_mul_ps(iq2,jq0);
 890
 891             /* REACTION-FIELD ELECTROSTATICS */
 892             felec            = _mm256_mul_ps(qq20,_mm256_sub_ps(_mm256_mul_ps(rinv20,rinvsq20),krf2));
 893
 894             cutoff_mask      = _mm256_cmp_ps(rsq20,rcutoff2,_CMP_LT_OQ);
 895
 896             fscal            = felec;
 897
 898             fscal            = _mm256_and_ps(fscal,cutoff_mask);
 899
 900             /* Calculate temporary vectorial force */
 901             tx               = _mm256_mul_ps(fscal,dx20);
 902             ty               = _mm256_mul_ps(fscal,dy20);
 903             tz               = _mm256_mul_ps(fscal,dz20);
 904
 905             /* Update vectorial force */
 906             fix2             = _mm256_add_ps(fix2,tx);
 907             fiy2             = _mm256_add_ps(fiy2,ty);
 908             fiz2             = _mm256_add_ps(fiz2,tz);
 909
 910             fjx0             = _mm256_add_ps(fjx0,tx);
 911             fjy0             = _mm256_add_ps(fjy0,ty);
 912             fjz0             = _mm256_add_ps(fjz0,tz);
 913
 914             }
 915
 916             fjptrA             = f+j_coord_offsetA;
 917             fjptrB             = f+j_coord_offsetB;
 918             fjptrC             = f+j_coord_offsetC;
 919             fjptrD             = f+j_coord_offsetD;
 920             fjptrE             = f+j_coord_offsetE;
 921             fjptrF             = f+j_coord_offsetF;
 922             fjptrG             = f+j_coord_offsetG;
 923             fjptrH             = f+j_coord_offsetH;
 924
 925             gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,fjx0,fjy0,fjz0);
 926
 927             /* Inner loop uses 93 flops */
 928         }
 929
 930         if(jidx<j_index_end)
 931         {
 932
 933             /* Get j neighbor index, and coordinate index */
 934             jnrlistA         = jjnr[jidx];
 935             jnrlistB         = jjnr[jidx+1];
 936             jnrlistC         = jjnr[jidx+2];
 937             jnrlistD         = jjnr[jidx+3];
 938             jnrlistE         = jjnr[jidx+4];
 939             jnrlistF         = jjnr[jidx+5];
 940             jnrlistG         = jjnr[jidx+6];
 941             jnrlistH         = jjnr[jidx+7];
 942             /* Sign of each element will be negative for non-real atoms.
 943              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 944              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 945              */
 946             dummy_mask = gmx_mm256_set_m128(gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx+4)),_mm_setzero_si128())),
 947                                             gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128())));
 948
 949             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 950             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 951             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 952             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 953             jnrE       = (jnrlistE>=0) ? jnrlistE : 0;
 954             jnrF       = (jnrlistF>=0) ? jnrlistF : 0;
 955             jnrG       = (jnrlistG>=0) ? jnrlistG : 0;
 956             jnrH       = (jnrlistH>=0) ? jnrlistH : 0;
 957             j_coord_offsetA  = DIM*jnrA;
 958             j_coord_offsetB  = DIM*jnrB;
 959             j_coord_offsetC  = DIM*jnrC;
 960             j_coord_offsetD  = DIM*jnrD;
 961             j_coord_offsetE  = DIM*jnrE;
 962             j_coord_offsetF  = DIM*jnrF;
 963             j_coord_offsetG  = DIM*jnrG;
 964             j_coord_offsetH  = DIM*jnrH;
 965
 966             /* load j atom coordinates */
 967             gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 968                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 969                                                  x+j_coord_offsetE,x+j_coord_offsetF,
 970                                                  x+j_coord_offsetG,x+j_coord_offsetH,
 971                                                  &jx0,&jy0,&jz0);
 972
 973             /* Calculate displacement vector */
 974             dx00             = _mm256_sub_ps(ix0,jx0);
 975             dy00             = _mm256_sub_ps(iy0,jy0);
 976             dz00             = _mm256_sub_ps(iz0,jz0);
 977             dx10             = _mm256_sub_ps(ix1,jx0);
 978             dy10             = _mm256_sub_ps(iy1,jy0);
 979             dz10             = _mm256_sub_ps(iz1,jz0);
 980             dx20             = _mm256_sub_ps(ix2,jx0);
 981             dy20             = _mm256_sub_ps(iy2,jy0);
 982             dz20             = _mm256_sub_ps(iz2,jz0);
 983
 984             /* Calculate squared distance and things based on it */
 985             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
 986             rsq10            = gmx_mm256_calc_rsq_ps(dx10,dy10,dz10);
 987             rsq20            = gmx_mm256_calc_rsq_ps(dx20,dy20,dz20);
 988
 989             rinv00           = gmx_mm256_invsqrt_ps(rsq00);
 990             rinv10           = gmx_mm256_invsqrt_ps(rsq10);
 991             rinv20           = gmx_mm256_invsqrt_ps(rsq20);
 992
 993             rinvsq00         = _mm256_mul_ps(rinv00,rinv00);
 994             rinvsq10         = _mm256_mul_ps(rinv10,rinv10);
 995             rinvsq20         = _mm256_mul_ps(rinv20,rinv20);
 996
 997             /* Load parameters for j particles */
 998             jq0              = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 999                                                                  charge+jnrC+0,charge+jnrD+0,
1000                                                                  charge+jnrE+0,charge+jnrF+0,
1001                                                                  charge+jnrG+0,charge+jnrH+0);
1002
1003             fjx0             = _mm256_setzero_ps();
1004             fjy0             = _mm256_setzero_ps();
1005             fjz0             = _mm256_setzero_ps();
1006
1007             /**************************
1008              * CALCULATE INTERACTIONS *
1009              **************************/
1010
1011             if (gmx_mm256_any_lt(rsq00,rcutoff2))
1012             {
1013
1014             /* Compute parameters for interactions between i and j atoms */
1015             qq00             = _mm256_mul_ps(iq0,jq0);
1016
1017             /* REACTION-FIELD ELECTROSTATICS */
1018             felec            = _mm256_mul_ps(qq00,_mm256_sub_ps(_mm256_mul_ps(rinv00,rinvsq00),krf2));
1019
1020             cutoff_mask      = _mm256_cmp_ps(rsq00,rcutoff2,_CMP_LT_OQ);
1021
1022             fscal            = felec;
1023
1024             fscal            = _mm256_and_ps(fscal,cutoff_mask);
1025
1026             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
1027
1028             /* Calculate temporary vectorial force */
1029             tx               = _mm256_mul_ps(fscal,dx00);
1030             ty               = _mm256_mul_ps(fscal,dy00);
1031             tz               = _mm256_mul_ps(fscal,dz00);
1032
1033             /* Update vectorial force */
1034             fix0             = _mm256_add_ps(fix0,tx);
1035             fiy0             = _mm256_add_ps(fiy0,ty);
1036             fiz0             = _mm256_add_ps(fiz0,tz);
1037
1038             fjx0             = _mm256_add_ps(fjx0,tx);
1039             fjy0             = _mm256_add_ps(fjy0,ty);
1040             fjz0             = _mm256_add_ps(fjz0,tz);
1041
1042             }
1043
1044             /**************************
1045              * CALCULATE INTERACTIONS *
1046              **************************/
1047
1048             if (gmx_mm256_any_lt(rsq10,rcutoff2))
1049             {
1050
1051             /* Compute parameters for interactions between i and j atoms */
1052             qq10             = _mm256_mul_ps(iq1,jq0);
1053
1054             /* REACTION-FIELD ELECTROSTATICS */
1055             felec            = _mm256_mul_ps(qq10,_mm256_sub_ps(_mm256_mul_ps(rinv10,rinvsq10),krf2));
1056
1057             cutoff_mask      = _mm256_cmp_ps(rsq10,rcutoff2,_CMP_LT_OQ);
1058
1059             fscal            = felec;
1060
1061             fscal            = _mm256_and_ps(fscal,cutoff_mask);
1062
1063             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
1064
1065             /* Calculate temporary vectorial force */
1066             tx               = _mm256_mul_ps(fscal,dx10);
1067             ty               = _mm256_mul_ps(fscal,dy10);
1068             tz               = _mm256_mul_ps(fscal,dz10);
1069
1070             /* Update vectorial force */
1071             fix1             = _mm256_add_ps(fix1,tx);
1072             fiy1             = _mm256_add_ps(fiy1,ty);
1073             fiz1             = _mm256_add_ps(fiz1,tz);
1074
1075             fjx0             = _mm256_add_ps(fjx0,tx);
1076             fjy0             = _mm256_add_ps(fjy0,ty);
1077             fjz0             = _mm256_add_ps(fjz0,tz);
1078
1079             }
1080
1081             /**************************
1082              * CALCULATE INTERACTIONS *
1083              **************************/
1084
1085             if (gmx_mm256_any_lt(rsq20,rcutoff2))
1086             {
1087
1088             /* Compute parameters for interactions between i and j atoms */
1089             qq20             = _mm256_mul_ps(iq2,jq0);
1090
1091             /* REACTION-FIELD ELECTROSTATICS */
1092             felec            = _mm256_mul_ps(qq20,_mm256_sub_ps(_mm256_mul_ps(rinv20,rinvsq20),krf2));
1093
1094             cutoff_mask      = _mm256_cmp_ps(rsq20,rcutoff2,_CMP_LT_OQ);
1095
1096             fscal            = felec;
1097
1098             fscal            = _mm256_and_ps(fscal,cutoff_mask);
1099
1100             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
1101
1102             /* Calculate temporary vectorial force */
1103             tx               = _mm256_mul_ps(fscal,dx20);
1104             ty               = _mm256_mul_ps(fscal,dy20);
1105             tz               = _mm256_mul_ps(fscal,dz20);
1106
1107             /* Update vectorial force */
1108             fix2             = _mm256_add_ps(fix2,tx);
1109             fiy2             = _mm256_add_ps(fiy2,ty);
1110             fiz2             = _mm256_add_ps(fiz2,tz);
1111
1112             fjx0             = _mm256_add_ps(fjx0,tx);
1113             fjy0             = _mm256_add_ps(fjy0,ty);
1114             fjz0             = _mm256_add_ps(fjz0,tz);
1115
1116             }
1117
1118             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
1119             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
1120             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
1121             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
1122             fjptrE             = (jnrlistE>=0) ? f+j_coord_offsetE : scratch;
1123             fjptrF             = (jnrlistF>=0) ? f+j_coord_offsetF : scratch;
1124             fjptrG             = (jnrlistG>=0) ? f+j_coord_offsetG : scratch;
1125             fjptrH             = (jnrlistH>=0) ? f+j_coord_offsetH : scratch;
1126
1127             gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,fjx0,fjy0,fjz0);
1128
1129             /* Inner loop uses 93 flops */
1130         }
1131
1132         /* End of innermost loop */
1133
1134         gmx_mm256_update_iforce_3atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
1135                                                  f+i_coord_offset,fshift+i_shift_offset);
1136
1137         /* Increment number of inner iterations */
1138         inneriter                  += j_index_end - j_index_start;
1139
1140         /* Outer loop uses 18 flops */
1141     }
1142
1143     /* Increment number of outer iterations */
1144     outeriter        += nri;
1145
1146     /* Update outer/inner flops */
1147
1148     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_W3_F,outeriter*18 + inneriter*93);
1149 }