src/gromacs/gmxlib/nonbonded/nb_kernel_avx_256_single/nb_kernel_ElecRFCut_VdwNone_GeomW4P1_avx_256_single.c

   1 /*
   2  * Note: this file was generated by the Gromacs avx_256_single kernel generator.
   3  *
   4  *                This source code is part of
   5  *
   6  *                 G   R   O   M   A   C   S
   7  *
   8  * Copyright (c) 2001-2012, The GROMACS Development Team
   9  *
  10  * Gromacs is a library for molecular simulation and trajectory analysis,
  11  * written by Erik Lindahl, David van der Spoel, Berk Hess, and others - for
  12  * a full list of developers and information, check out http://www.gromacs.org
  13  *
  14  * This program is free software; you can redistribute it and/or modify it under
  15  * the terms of the GNU Lesser General Public License as published by the Free
  16  * Software Foundation; either version 2 of the License, or (at your option) any
  17  * later version.
  18  *
  19  * To help fund GROMACS development, we humbly ask that you cite
  20  * the papers people have written on it - you can find them on the website.
  21  */
  22 #ifdef HAVE_CONFIG_H
  23 #include <config.h>
  24 #endif
  25
  26 #include <math.h>
  27
  28 #include "../nb_kernel.h"
  29 #include "types/simple.h"
  30 #include "vec.h"
  31 #include "nrnb.h"
  32
  33 #include "gmx_math_x86_avx_256_single.h"
  34 #include "kernelutil_x86_avx_256_single.h"
  35
  36 /*
  37  * Gromacs nonbonded kernel:   nb_kernel_ElecRFCut_VdwNone_GeomW4P1_VF_avx_256_single
  38  * Electrostatics interaction: ReactionField
  39  * VdW interaction:            None
  40  * Geometry:                   Water4-Particle
  41  * Calculate force/pot:        PotentialAndForce
  42  */
  43 void
  44 nb_kernel_ElecRFCut_VdwNone_GeomW4P1_VF_avx_256_single
  45                     (t_nblist * gmx_restrict                nlist,
  46                      rvec * gmx_restrict                    xx,
  47                      rvec * gmx_restrict                    ff,
  48                      t_forcerec * gmx_restrict              fr,
  49                      t_mdatoms * gmx_restrict               mdatoms,
  50                      nb_kernel_data_t * gmx_restrict        kernel_data,
  51                      t_nrnb * gmx_restrict                  nrnb)
  52 {
  53     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
  54      * just 0 for non-waters.
  55      * Suffixes A,B,C,D,E,F,G,H refer to j loop unrolling done with AVX, e.g. for the eight different
  56      * jnr indices corresponding to data put in the four positions in the SIMD register.
  57      */
  58     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
  59     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
  60     int              jnrA,jnrB,jnrC,jnrD;
  61     int              jnrE,jnrF,jnrG,jnrH;
  62     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
  63     int              jnrlistE,jnrlistF,jnrlistG,jnrlistH;
  64     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
  65     int              j_coord_offsetE,j_coord_offsetF,j_coord_offsetG,j_coord_offsetH;
  66     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
  67     real             rcutoff_scalar;
  68     real             *shiftvec,*fshift,*x,*f;
  69     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD,*fjptrE,*fjptrF,*fjptrG,*fjptrH;
  70     real             scratch[4*DIM];
  71     __m256           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
  72     real *           vdwioffsetptr1;
  73     __m256           ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
  74     real *           vdwioffsetptr2;
  75     __m256           ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
  76     real *           vdwioffsetptr3;
  77     __m256           ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
  78     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D,vdwjidx0E,vdwjidx0F,vdwjidx0G,vdwjidx0H;
  79     __m256           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  80     __m256           dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
  81     __m256           dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
  82     __m256           dx30,dy30,dz30,rsq30,rinv30,rinvsq30,r30,qq30,c6_30,c12_30;
  83     __m256           velec,felec,velecsum,facel,crf,krf,krf2;
  84     real             *charge;
  85     __m256           dummy_mask,cutoff_mask;
  86     __m256           signbit = _mm256_castsi256_ps( _mm256_set1_epi32(0x80000000) );
  87     __m256           one     = _mm256_set1_ps(1.0);
  88     __m256           two     = _mm256_set1_ps(2.0);
  89     x                = xx[0];
  90     f                = ff[0];
  91
  92     nri              = nlist->nri;
  93     iinr             = nlist->iinr;
  94     jindex           = nlist->jindex;
  95     jjnr             = nlist->jjnr;
  96     shiftidx         = nlist->shift;
  97     gid              = nlist->gid;
  98     shiftvec         = fr->shift_vec[0];
  99     fshift           = fr->fshift[0];
 100     facel            = _mm256_set1_ps(fr->epsfac);
 101     charge           = mdatoms->chargeA;
 102     krf              = _mm256_set1_ps(fr->ic->k_rf);
 103     krf2             = _mm256_set1_ps(fr->ic->k_rf*2.0);
 104     crf              = _mm256_set1_ps(fr->ic->c_rf);
 105
 106     /* Setup water-specific parameters */
 107     inr              = nlist->iinr[0];
 108     iq1              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+1]));
 109     iq2              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+2]));
 110     iq3              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+3]));
 111
 112     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 113     rcutoff_scalar   = fr->rcoulomb;
 114     rcutoff          = _mm256_set1_ps(rcutoff_scalar);
 115     rcutoff2         = _mm256_mul_ps(rcutoff,rcutoff);
 116
 117     /* Avoid stupid compiler warnings */
 118     jnrA = jnrB = jnrC = jnrD = jnrE = jnrF = jnrG = jnrH = 0;
 119     j_coord_offsetA = 0;
 120     j_coord_offsetB = 0;
 121     j_coord_offsetC = 0;
 122     j_coord_offsetD = 0;
 123     j_coord_offsetE = 0;
 124     j_coord_offsetF = 0;
 125     j_coord_offsetG = 0;
 126     j_coord_offsetH = 0;
 127
 128     outeriter        = 0;
 129     inneriter        = 0;
 130
 131     for(iidx=0;iidx<4*DIM;iidx++)
 132     {
 133         scratch[iidx] = 0.0;
 134     }
 135
 136     /* Start outer loop over neighborlists */
 137     for(iidx=0; iidx<nri; iidx++)
 138     {
 139         /* Load shift vector for this list */
 140         i_shift_offset   = DIM*shiftidx[iidx];
 141
 142         /* Load limits for loop over neighbors */
 143         j_index_start    = jindex[iidx];
 144         j_index_end      = jindex[iidx+1];
 145
 146         /* Get outer coordinate index */
 147         inr              = iinr[iidx];
 148         i_coord_offset   = DIM*inr;
 149
 150         /* Load i particle coords and add shift vector */
 151         gmx_mm256_load_shift_and_3rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset+DIM,
 152                                                     &ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
 153
 154         fix1             = _mm256_setzero_ps();
 155         fiy1             = _mm256_setzero_ps();
 156         fiz1             = _mm256_setzero_ps();
 157         fix2             = _mm256_setzero_ps();
 158         fiy2             = _mm256_setzero_ps();
 159         fiz2             = _mm256_setzero_ps();
 160         fix3             = _mm256_setzero_ps();
 161         fiy3             = _mm256_setzero_ps();
 162         fiz3             = _mm256_setzero_ps();
 163
 164         /* Reset potential sums */
 165         velecsum         = _mm256_setzero_ps();
 166
 167         /* Start inner kernel loop */
 168         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+7]>=0; jidx+=8)
 169         {
 170
 171             /* Get j neighbor index, and coordinate index */
 172             jnrA             = jjnr[jidx];
 173             jnrB             = jjnr[jidx+1];
 174             jnrC             = jjnr[jidx+2];
 175             jnrD             = jjnr[jidx+3];
 176             jnrE             = jjnr[jidx+4];
 177             jnrF             = jjnr[jidx+5];
 178             jnrG             = jjnr[jidx+6];
 179             jnrH             = jjnr[jidx+7];
 180             j_coord_offsetA  = DIM*jnrA;
 181             j_coord_offsetB  = DIM*jnrB;
 182             j_coord_offsetC  = DIM*jnrC;
 183             j_coord_offsetD  = DIM*jnrD;
 184             j_coord_offsetE  = DIM*jnrE;
 185             j_coord_offsetF  = DIM*jnrF;
 186             j_coord_offsetG  = DIM*jnrG;
 187             j_coord_offsetH  = DIM*jnrH;
 188
 189             /* load j atom coordinates */
 190             gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 191                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 192                                                  x+j_coord_offsetE,x+j_coord_offsetF,
 193                                                  x+j_coord_offsetG,x+j_coord_offsetH,
 194                                                  &jx0,&jy0,&jz0);
 195
 196             /* Calculate displacement vector */
 197             dx10             = _mm256_sub_ps(ix1,jx0);
 198             dy10             = _mm256_sub_ps(iy1,jy0);
 199             dz10             = _mm256_sub_ps(iz1,jz0);
 200             dx20             = _mm256_sub_ps(ix2,jx0);
 201             dy20             = _mm256_sub_ps(iy2,jy0);
 202             dz20             = _mm256_sub_ps(iz2,jz0);
 203             dx30             = _mm256_sub_ps(ix3,jx0);
 204             dy30             = _mm256_sub_ps(iy3,jy0);
 205             dz30             = _mm256_sub_ps(iz3,jz0);
 206
 207             /* Calculate squared distance and things based on it */
 208             rsq10            = gmx_mm256_calc_rsq_ps(dx10,dy10,dz10);
 209             rsq20            = gmx_mm256_calc_rsq_ps(dx20,dy20,dz20);
 210             rsq30            = gmx_mm256_calc_rsq_ps(dx30,dy30,dz30);
 211
 212             rinv10           = gmx_mm256_invsqrt_ps(rsq10);
 213             rinv20           = gmx_mm256_invsqrt_ps(rsq20);
 214             rinv30           = gmx_mm256_invsqrt_ps(rsq30);
 215
 216             rinvsq10         = _mm256_mul_ps(rinv10,rinv10);
 217             rinvsq20         = _mm256_mul_ps(rinv20,rinv20);
 218             rinvsq30         = _mm256_mul_ps(rinv30,rinv30);
 219
 220             /* Load parameters for j particles */
 221             jq0              = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 222                                                                  charge+jnrC+0,charge+jnrD+0,
 223                                                                  charge+jnrE+0,charge+jnrF+0,
 224                                                                  charge+jnrG+0,charge+jnrH+0);
 225
 226             fjx0             = _mm256_setzero_ps();
 227             fjy0             = _mm256_setzero_ps();
 228             fjz0             = _mm256_setzero_ps();
 229
 230             /**************************
 231              * CALCULATE INTERACTIONS *
 232              **************************/
 233
 234             if (gmx_mm256_any_lt(rsq10,rcutoff2))
 235             {
 236
 237             /* Compute parameters for interactions between i and j atoms */
 238             qq10             = _mm256_mul_ps(iq1,jq0);
 239
 240             /* REACTION-FIELD ELECTROSTATICS */
 241             velec            = _mm256_mul_ps(qq10,_mm256_sub_ps(_mm256_add_ps(rinv10,_mm256_mul_ps(krf,rsq10)),crf));
 242             felec            = _mm256_mul_ps(qq10,_mm256_sub_ps(_mm256_mul_ps(rinv10,rinvsq10),krf2));
 243
 244             cutoff_mask      = _mm256_cmp_ps(rsq10,rcutoff2,_CMP_LT_OQ);
 245
 246             /* Update potential sum for this i atom from the interaction with this j atom. */
 247             velec            = _mm256_and_ps(velec,cutoff_mask);
 248             velecsum         = _mm256_add_ps(velecsum,velec);
 249
 250             fscal            = felec;
 251
 252             fscal            = _mm256_and_ps(fscal,cutoff_mask);
 253
 254             /* Calculate temporary vectorial force */
 255             tx               = _mm256_mul_ps(fscal,dx10);
 256             ty               = _mm256_mul_ps(fscal,dy10);
 257             tz               = _mm256_mul_ps(fscal,dz10);
 258
 259             /* Update vectorial force */
 260             fix1             = _mm256_add_ps(fix1,tx);
 261             fiy1             = _mm256_add_ps(fiy1,ty);
 262             fiz1             = _mm256_add_ps(fiz1,tz);
 263
 264             fjx0             = _mm256_add_ps(fjx0,tx);
 265             fjy0             = _mm256_add_ps(fjy0,ty);
 266             fjz0             = _mm256_add_ps(fjz0,tz);
 267
 268             }
 269
 270             /**************************
 271              * CALCULATE INTERACTIONS *
 272              **************************/
 273
 274             if (gmx_mm256_any_lt(rsq20,rcutoff2))
 275             {
 276
 277             /* Compute parameters for interactions between i and j atoms */
 278             qq20             = _mm256_mul_ps(iq2,jq0);
 279
 280             /* REACTION-FIELD ELECTROSTATICS */
 281             velec            = _mm256_mul_ps(qq20,_mm256_sub_ps(_mm256_add_ps(rinv20,_mm256_mul_ps(krf,rsq20)),crf));
 282             felec            = _mm256_mul_ps(qq20,_mm256_sub_ps(_mm256_mul_ps(rinv20,rinvsq20),krf2));
 283
 284             cutoff_mask      = _mm256_cmp_ps(rsq20,rcutoff2,_CMP_LT_OQ);
 285
 286             /* Update potential sum for this i atom from the interaction with this j atom. */
 287             velec            = _mm256_and_ps(velec,cutoff_mask);
 288             velecsum         = _mm256_add_ps(velecsum,velec);
 289
 290             fscal            = felec;
 291
 292             fscal            = _mm256_and_ps(fscal,cutoff_mask);
 293
 294             /* Calculate temporary vectorial force */
 295             tx               = _mm256_mul_ps(fscal,dx20);
 296             ty               = _mm256_mul_ps(fscal,dy20);
 297             tz               = _mm256_mul_ps(fscal,dz20);
 298
 299             /* Update vectorial force */
 300             fix2             = _mm256_add_ps(fix2,tx);
 301             fiy2             = _mm256_add_ps(fiy2,ty);
 302             fiz2             = _mm256_add_ps(fiz2,tz);
 303
 304             fjx0             = _mm256_add_ps(fjx0,tx);
 305             fjy0             = _mm256_add_ps(fjy0,ty);
 306             fjz0             = _mm256_add_ps(fjz0,tz);
 307
 308             }
 309
 310             /**************************
 311              * CALCULATE INTERACTIONS *
 312              **************************/
 313
 314             if (gmx_mm256_any_lt(rsq30,rcutoff2))
 315             {
 316
 317             /* Compute parameters for interactions between i and j atoms */
 318             qq30             = _mm256_mul_ps(iq3,jq0);
 319
 320             /* REACTION-FIELD ELECTROSTATICS */
 321             velec            = _mm256_mul_ps(qq30,_mm256_sub_ps(_mm256_add_ps(rinv30,_mm256_mul_ps(krf,rsq30)),crf));
 322             felec            = _mm256_mul_ps(qq30,_mm256_sub_ps(_mm256_mul_ps(rinv30,rinvsq30),krf2));
 323
 324             cutoff_mask      = _mm256_cmp_ps(rsq30,rcutoff2,_CMP_LT_OQ);
 325
 326             /* Update potential sum for this i atom from the interaction with this j atom. */
 327             velec            = _mm256_and_ps(velec,cutoff_mask);
 328             velecsum         = _mm256_add_ps(velecsum,velec);
 329
 330             fscal            = felec;
 331
 332             fscal            = _mm256_and_ps(fscal,cutoff_mask);
 333
 334             /* Calculate temporary vectorial force */
 335             tx               = _mm256_mul_ps(fscal,dx30);
 336             ty               = _mm256_mul_ps(fscal,dy30);
 337             tz               = _mm256_mul_ps(fscal,dz30);
 338
 339             /* Update vectorial force */
 340             fix3             = _mm256_add_ps(fix3,tx);
 341             fiy3             = _mm256_add_ps(fiy3,ty);
 342             fiz3             = _mm256_add_ps(fiz3,tz);
 343
 344             fjx0             = _mm256_add_ps(fjx0,tx);
 345             fjy0             = _mm256_add_ps(fjy0,ty);
 346             fjz0             = _mm256_add_ps(fjz0,tz);
 347
 348             }
 349
 350             fjptrA             = f+j_coord_offsetA;
 351             fjptrB             = f+j_coord_offsetB;
 352             fjptrC             = f+j_coord_offsetC;
 353             fjptrD             = f+j_coord_offsetD;
 354             fjptrE             = f+j_coord_offsetE;
 355             fjptrF             = f+j_coord_offsetF;
 356             fjptrG             = f+j_coord_offsetG;
 357             fjptrH             = f+j_coord_offsetH;
 358
 359             gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,fjx0,fjy0,fjz0);
 360
 361             /* Inner loop uses 111 flops */
 362         }
 363
 364         if(jidx<j_index_end)
 365         {
 366
 367             /* Get j neighbor index, and coordinate index */
 368             jnrlistA         = jjnr[jidx];
 369             jnrlistB         = jjnr[jidx+1];
 370             jnrlistC         = jjnr[jidx+2];
 371             jnrlistD         = jjnr[jidx+3];
 372             jnrlistE         = jjnr[jidx+4];
 373             jnrlistF         = jjnr[jidx+5];
 374             jnrlistG         = jjnr[jidx+6];
 375             jnrlistH         = jjnr[jidx+7];
 376             /* Sign of each element will be negative for non-real atoms.
 377              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 378              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 379              */
 380             dummy_mask = gmx_mm256_set_m128(gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx+4)),_mm_setzero_si128())),
 381                                             gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128())));
 382
 383             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 384             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 385             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 386             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 387             jnrE       = (jnrlistE>=0) ? jnrlistE : 0;
 388             jnrF       = (jnrlistF>=0) ? jnrlistF : 0;
 389             jnrG       = (jnrlistG>=0) ? jnrlistG : 0;
 390             jnrH       = (jnrlistH>=0) ? jnrlistH : 0;
 391             j_coord_offsetA  = DIM*jnrA;
 392             j_coord_offsetB  = DIM*jnrB;
 393             j_coord_offsetC  = DIM*jnrC;
 394             j_coord_offsetD  = DIM*jnrD;
 395             j_coord_offsetE  = DIM*jnrE;
 396             j_coord_offsetF  = DIM*jnrF;
 397             j_coord_offsetG  = DIM*jnrG;
 398             j_coord_offsetH  = DIM*jnrH;
 399
 400             /* load j atom coordinates */
 401             gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 402                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 403                                                  x+j_coord_offsetE,x+j_coord_offsetF,
 404                                                  x+j_coord_offsetG,x+j_coord_offsetH,
 405                                                  &jx0,&jy0,&jz0);
 406
 407             /* Calculate displacement vector */
 408             dx10             = _mm256_sub_ps(ix1,jx0);
 409             dy10             = _mm256_sub_ps(iy1,jy0);
 410             dz10             = _mm256_sub_ps(iz1,jz0);
 411             dx20             = _mm256_sub_ps(ix2,jx0);
 412             dy20             = _mm256_sub_ps(iy2,jy0);
 413             dz20             = _mm256_sub_ps(iz2,jz0);
 414             dx30             = _mm256_sub_ps(ix3,jx0);
 415             dy30             = _mm256_sub_ps(iy3,jy0);
 416             dz30             = _mm256_sub_ps(iz3,jz0);
 417
 418             /* Calculate squared distance and things based on it */
 419             rsq10            = gmx_mm256_calc_rsq_ps(dx10,dy10,dz10);
 420             rsq20            = gmx_mm256_calc_rsq_ps(dx20,dy20,dz20);
 421             rsq30            = gmx_mm256_calc_rsq_ps(dx30,dy30,dz30);
 422
 423             rinv10           = gmx_mm256_invsqrt_ps(rsq10);
 424             rinv20           = gmx_mm256_invsqrt_ps(rsq20);
 425             rinv30           = gmx_mm256_invsqrt_ps(rsq30);
 426
 427             rinvsq10         = _mm256_mul_ps(rinv10,rinv10);
 428             rinvsq20         = _mm256_mul_ps(rinv20,rinv20);
 429             rinvsq30         = _mm256_mul_ps(rinv30,rinv30);
 430
 431             /* Load parameters for j particles */
 432             jq0              = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 433                                                                  charge+jnrC+0,charge+jnrD+0,
 434                                                                  charge+jnrE+0,charge+jnrF+0,
 435                                                                  charge+jnrG+0,charge+jnrH+0);
 436
 437             fjx0             = _mm256_setzero_ps();
 438             fjy0             = _mm256_setzero_ps();
 439             fjz0             = _mm256_setzero_ps();
 440
 441             /**************************
 442              * CALCULATE INTERACTIONS *
 443              **************************/
 444
 445             if (gmx_mm256_any_lt(rsq10,rcutoff2))
 446             {
 447
 448             /* Compute parameters for interactions between i and j atoms */
 449             qq10             = _mm256_mul_ps(iq1,jq0);
 450
 451             /* REACTION-FIELD ELECTROSTATICS */
 452             velec            = _mm256_mul_ps(qq10,_mm256_sub_ps(_mm256_add_ps(rinv10,_mm256_mul_ps(krf,rsq10)),crf));
 453             felec            = _mm256_mul_ps(qq10,_mm256_sub_ps(_mm256_mul_ps(rinv10,rinvsq10),krf2));
 454
 455             cutoff_mask      = _mm256_cmp_ps(rsq10,rcutoff2,_CMP_LT_OQ);
 456
 457             /* Update potential sum for this i atom from the interaction with this j atom. */
 458             velec            = _mm256_and_ps(velec,cutoff_mask);
 459             velec            = _mm256_andnot_ps(dummy_mask,velec);
 460             velecsum         = _mm256_add_ps(velecsum,velec);
 461
 462             fscal            = felec;
 463
 464             fscal            = _mm256_and_ps(fscal,cutoff_mask);
 465
 466             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
 467
 468             /* Calculate temporary vectorial force */
 469             tx               = _mm256_mul_ps(fscal,dx10);
 470             ty               = _mm256_mul_ps(fscal,dy10);
 471             tz               = _mm256_mul_ps(fscal,dz10);
 472
 473             /* Update vectorial force */
 474             fix1             = _mm256_add_ps(fix1,tx);
 475             fiy1             = _mm256_add_ps(fiy1,ty);
 476             fiz1             = _mm256_add_ps(fiz1,tz);
 477
 478             fjx0             = _mm256_add_ps(fjx0,tx);
 479             fjy0             = _mm256_add_ps(fjy0,ty);
 480             fjz0             = _mm256_add_ps(fjz0,tz);
 481
 482             }
 483
 484             /**************************
 485              * CALCULATE INTERACTIONS *
 486              **************************/
 487
 488             if (gmx_mm256_any_lt(rsq20,rcutoff2))
 489             {
 490
 491             /* Compute parameters for interactions between i and j atoms */
 492             qq20             = _mm256_mul_ps(iq2,jq0);
 493
 494             /* REACTION-FIELD ELECTROSTATICS */
 495             velec            = _mm256_mul_ps(qq20,_mm256_sub_ps(_mm256_add_ps(rinv20,_mm256_mul_ps(krf,rsq20)),crf));
 496             felec            = _mm256_mul_ps(qq20,_mm256_sub_ps(_mm256_mul_ps(rinv20,rinvsq20),krf2));
 497
 498             cutoff_mask      = _mm256_cmp_ps(rsq20,rcutoff2,_CMP_LT_OQ);
 499
 500             /* Update potential sum for this i atom from the interaction with this j atom. */
 501             velec            = _mm256_and_ps(velec,cutoff_mask);
 502             velec            = _mm256_andnot_ps(dummy_mask,velec);
 503             velecsum         = _mm256_add_ps(velecsum,velec);
 504
 505             fscal            = felec;
 506
 507             fscal            = _mm256_and_ps(fscal,cutoff_mask);
 508
 509             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
 510
 511             /* Calculate temporary vectorial force */
 512             tx               = _mm256_mul_ps(fscal,dx20);
 513             ty               = _mm256_mul_ps(fscal,dy20);
 514             tz               = _mm256_mul_ps(fscal,dz20);
 515
 516             /* Update vectorial force */
 517             fix2             = _mm256_add_ps(fix2,tx);
 518             fiy2             = _mm256_add_ps(fiy2,ty);
 519             fiz2             = _mm256_add_ps(fiz2,tz);
 520
 521             fjx0             = _mm256_add_ps(fjx0,tx);
 522             fjy0             = _mm256_add_ps(fjy0,ty);
 523             fjz0             = _mm256_add_ps(fjz0,tz);
 524
 525             }
 526
 527             /**************************
 528              * CALCULATE INTERACTIONS *
 529              **************************/
 530
 531             if (gmx_mm256_any_lt(rsq30,rcutoff2))
 532             {
 533
 534             /* Compute parameters for interactions between i and j atoms */
 535             qq30             = _mm256_mul_ps(iq3,jq0);
 536
 537             /* REACTION-FIELD ELECTROSTATICS */
 538             velec            = _mm256_mul_ps(qq30,_mm256_sub_ps(_mm256_add_ps(rinv30,_mm256_mul_ps(krf,rsq30)),crf));
 539             felec            = _mm256_mul_ps(qq30,_mm256_sub_ps(_mm256_mul_ps(rinv30,rinvsq30),krf2));
 540
 541             cutoff_mask      = _mm256_cmp_ps(rsq30,rcutoff2,_CMP_LT_OQ);
 542
 543             /* Update potential sum for this i atom from the interaction with this j atom. */
 544             velec            = _mm256_and_ps(velec,cutoff_mask);
 545             velec            = _mm256_andnot_ps(dummy_mask,velec);
 546             velecsum         = _mm256_add_ps(velecsum,velec);
 547
 548             fscal            = felec;
 549
 550             fscal            = _mm256_and_ps(fscal,cutoff_mask);
 551
 552             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
 553
 554             /* Calculate temporary vectorial force */
 555             tx               = _mm256_mul_ps(fscal,dx30);
 556             ty               = _mm256_mul_ps(fscal,dy30);
 557             tz               = _mm256_mul_ps(fscal,dz30);
 558
 559             /* Update vectorial force */
 560             fix3             = _mm256_add_ps(fix3,tx);
 561             fiy3             = _mm256_add_ps(fiy3,ty);
 562             fiz3             = _mm256_add_ps(fiz3,tz);
 563
 564             fjx0             = _mm256_add_ps(fjx0,tx);
 565             fjy0             = _mm256_add_ps(fjy0,ty);
 566             fjz0             = _mm256_add_ps(fjz0,tz);
 567
 568             }
 569
 570             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 571             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 572             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 573             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 574             fjptrE             = (jnrlistE>=0) ? f+j_coord_offsetE : scratch;
 575             fjptrF             = (jnrlistF>=0) ? f+j_coord_offsetF : scratch;
 576             fjptrG             = (jnrlistG>=0) ? f+j_coord_offsetG : scratch;
 577             fjptrH             = (jnrlistH>=0) ? f+j_coord_offsetH : scratch;
 578
 579             gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,fjx0,fjy0,fjz0);
 580
 581             /* Inner loop uses 111 flops */
 582         }
 583
 584         /* End of innermost loop */
 585
 586         gmx_mm256_update_iforce_3atom_swizzle_ps(fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
 587                                                  f+i_coord_offset+DIM,fshift+i_shift_offset);
 588
 589         ggid                        = gid[iidx];
 590         /* Update potential energies */
 591         gmx_mm256_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid);
 592
 593         /* Increment number of inner iterations */
 594         inneriter                  += j_index_end - j_index_start;
 595
 596         /* Outer loop uses 19 flops */
 597     }
 598
 599     /* Increment number of outer iterations */
 600     outeriter        += nri;
 601
 602     /* Update outer/inner flops */
 603
 604     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_W4_VF,outeriter*19 + inneriter*111);
 605 }
 606 /*
 607  * Gromacs nonbonded kernel:   nb_kernel_ElecRFCut_VdwNone_GeomW4P1_F_avx_256_single
 608  * Electrostatics interaction: ReactionField
 609  * VdW interaction:            None
 610  * Geometry:                   Water4-Particle
 611  * Calculate force/pot:        Force
 612  */
 613 void
 614 nb_kernel_ElecRFCut_VdwNone_GeomW4P1_F_avx_256_single
 615                     (t_nblist * gmx_restrict                nlist,
 616                      rvec * gmx_restrict                    xx,
 617                      rvec * gmx_restrict                    ff,
 618                      t_forcerec * gmx_restrict              fr,
 619                      t_mdatoms * gmx_restrict               mdatoms,
 620                      nb_kernel_data_t * gmx_restrict        kernel_data,
 621                      t_nrnb * gmx_restrict                  nrnb)
 622 {
 623     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 624      * just 0 for non-waters.
 625      * Suffixes A,B,C,D,E,F,G,H refer to j loop unrolling done with AVX, e.g. for the eight different
 626      * jnr indices corresponding to data put in the four positions in the SIMD register.
 627      */
 628     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 629     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 630     int              jnrA,jnrB,jnrC,jnrD;
 631     int              jnrE,jnrF,jnrG,jnrH;
 632     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
 633     int              jnrlistE,jnrlistF,jnrlistG,jnrlistH;
 634     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 635     int              j_coord_offsetE,j_coord_offsetF,j_coord_offsetG,j_coord_offsetH;
 636     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 637     real             rcutoff_scalar;
 638     real             *shiftvec,*fshift,*x,*f;
 639     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD,*fjptrE,*fjptrF,*fjptrG,*fjptrH;
 640     real             scratch[4*DIM];
 641     __m256           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 642     real *           vdwioffsetptr1;
 643     __m256           ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
 644     real *           vdwioffsetptr2;
 645     __m256           ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
 646     real *           vdwioffsetptr3;
 647     __m256           ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
 648     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D,vdwjidx0E,vdwjidx0F,vdwjidx0G,vdwjidx0H;
 649     __m256           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 650     __m256           dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
 651     __m256           dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
 652     __m256           dx30,dy30,dz30,rsq30,rinv30,rinvsq30,r30,qq30,c6_30,c12_30;
 653     __m256           velec,felec,velecsum,facel,crf,krf,krf2;
 654     real             *charge;
 655     __m256           dummy_mask,cutoff_mask;
 656     __m256           signbit = _mm256_castsi256_ps( _mm256_set1_epi32(0x80000000) );
 657     __m256           one     = _mm256_set1_ps(1.0);
 658     __m256           two     = _mm256_set1_ps(2.0);
 659     x                = xx[0];
 660     f                = ff[0];
 661
 662     nri              = nlist->nri;
 663     iinr             = nlist->iinr;
 664     jindex           = nlist->jindex;
 665     jjnr             = nlist->jjnr;
 666     shiftidx         = nlist->shift;
 667     gid              = nlist->gid;
 668     shiftvec         = fr->shift_vec[0];
 669     fshift           = fr->fshift[0];
 670     facel            = _mm256_set1_ps(fr->epsfac);
 671     charge           = mdatoms->chargeA;
 672     krf              = _mm256_set1_ps(fr->ic->k_rf);
 673     krf2             = _mm256_set1_ps(fr->ic->k_rf*2.0);
 674     crf              = _mm256_set1_ps(fr->ic->c_rf);
 675
 676     /* Setup water-specific parameters */
 677     inr              = nlist->iinr[0];
 678     iq1              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+1]));
 679     iq2              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+2]));
 680     iq3              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+3]));
 681
 682     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 683     rcutoff_scalar   = fr->rcoulomb;
 684     rcutoff          = _mm256_set1_ps(rcutoff_scalar);
 685     rcutoff2         = _mm256_mul_ps(rcutoff,rcutoff);
 686
 687     /* Avoid stupid compiler warnings */
 688     jnrA = jnrB = jnrC = jnrD = jnrE = jnrF = jnrG = jnrH = 0;
 689     j_coord_offsetA = 0;
 690     j_coord_offsetB = 0;
 691     j_coord_offsetC = 0;
 692     j_coord_offsetD = 0;
 693     j_coord_offsetE = 0;
 694     j_coord_offsetF = 0;
 695     j_coord_offsetG = 0;
 696     j_coord_offsetH = 0;
 697
 698     outeriter        = 0;
 699     inneriter        = 0;
 700
 701     for(iidx=0;iidx<4*DIM;iidx++)
 702     {
 703         scratch[iidx] = 0.0;
 704     }
 705
 706     /* Start outer loop over neighborlists */
 707     for(iidx=0; iidx<nri; iidx++)
 708     {
 709         /* Load shift vector for this list */
 710         i_shift_offset   = DIM*shiftidx[iidx];
 711
 712         /* Load limits for loop over neighbors */
 713         j_index_start    = jindex[iidx];
 714         j_index_end      = jindex[iidx+1];
 715
 716         /* Get outer coordinate index */
 717         inr              = iinr[iidx];
 718         i_coord_offset   = DIM*inr;
 719
 720         /* Load i particle coords and add shift vector */
 721         gmx_mm256_load_shift_and_3rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset+DIM,
 722                                                     &ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
 723
 724         fix1             = _mm256_setzero_ps();
 725         fiy1             = _mm256_setzero_ps();
 726         fiz1             = _mm256_setzero_ps();
 727         fix2             = _mm256_setzero_ps();
 728         fiy2             = _mm256_setzero_ps();
 729         fiz2             = _mm256_setzero_ps();
 730         fix3             = _mm256_setzero_ps();
 731         fiy3             = _mm256_setzero_ps();
 732         fiz3             = _mm256_setzero_ps();
 733
 734         /* Start inner kernel loop */
 735         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+7]>=0; jidx+=8)
 736         {
 737
 738             /* Get j neighbor index, and coordinate index */
 739             jnrA             = jjnr[jidx];
 740             jnrB             = jjnr[jidx+1];
 741             jnrC             = jjnr[jidx+2];
 742             jnrD             = jjnr[jidx+3];
 743             jnrE             = jjnr[jidx+4];
 744             jnrF             = jjnr[jidx+5];
 745             jnrG             = jjnr[jidx+6];
 746             jnrH             = jjnr[jidx+7];
 747             j_coord_offsetA  = DIM*jnrA;
 748             j_coord_offsetB  = DIM*jnrB;
 749             j_coord_offsetC  = DIM*jnrC;
 750             j_coord_offsetD  = DIM*jnrD;
 751             j_coord_offsetE  = DIM*jnrE;
 752             j_coord_offsetF  = DIM*jnrF;
 753             j_coord_offsetG  = DIM*jnrG;
 754             j_coord_offsetH  = DIM*jnrH;
 755
 756             /* load j atom coordinates */
 757             gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 758                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 759                                                  x+j_coord_offsetE,x+j_coord_offsetF,
 760                                                  x+j_coord_offsetG,x+j_coord_offsetH,
 761                                                  &jx0,&jy0,&jz0);
 762
 763             /* Calculate displacement vector */
 764             dx10             = _mm256_sub_ps(ix1,jx0);
 765             dy10             = _mm256_sub_ps(iy1,jy0);
 766             dz10             = _mm256_sub_ps(iz1,jz0);
 767             dx20             = _mm256_sub_ps(ix2,jx0);
 768             dy20             = _mm256_sub_ps(iy2,jy0);
 769             dz20             = _mm256_sub_ps(iz2,jz0);
 770             dx30             = _mm256_sub_ps(ix3,jx0);
 771             dy30             = _mm256_sub_ps(iy3,jy0);
 772             dz30             = _mm256_sub_ps(iz3,jz0);
 773
 774             /* Calculate squared distance and things based on it */
 775             rsq10            = gmx_mm256_calc_rsq_ps(dx10,dy10,dz10);
 776             rsq20            = gmx_mm256_calc_rsq_ps(dx20,dy20,dz20);
 777             rsq30            = gmx_mm256_calc_rsq_ps(dx30,dy30,dz30);
 778
 779             rinv10           = gmx_mm256_invsqrt_ps(rsq10);
 780             rinv20           = gmx_mm256_invsqrt_ps(rsq20);
 781             rinv30           = gmx_mm256_invsqrt_ps(rsq30);
 782
 783             rinvsq10         = _mm256_mul_ps(rinv10,rinv10);
 784             rinvsq20         = _mm256_mul_ps(rinv20,rinv20);
 785             rinvsq30         = _mm256_mul_ps(rinv30,rinv30);
 786
 787             /* Load parameters for j particles */
 788             jq0              = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 789                                                                  charge+jnrC+0,charge+jnrD+0,
 790                                                                  charge+jnrE+0,charge+jnrF+0,
 791                                                                  charge+jnrG+0,charge+jnrH+0);
 792
 793             fjx0             = _mm256_setzero_ps();
 794             fjy0             = _mm256_setzero_ps();
 795             fjz0             = _mm256_setzero_ps();
 796
 797             /**************************
 798              * CALCULATE INTERACTIONS *
 799              **************************/
 800
 801             if (gmx_mm256_any_lt(rsq10,rcutoff2))
 802             {
 803
 804             /* Compute parameters for interactions between i and j atoms */
 805             qq10             = _mm256_mul_ps(iq1,jq0);
 806
 807             /* REACTION-FIELD ELECTROSTATICS */
 808             felec            = _mm256_mul_ps(qq10,_mm256_sub_ps(_mm256_mul_ps(rinv10,rinvsq10),krf2));
 809
 810             cutoff_mask      = _mm256_cmp_ps(rsq10,rcutoff2,_CMP_LT_OQ);
 811
 812             fscal            = felec;
 813
 814             fscal            = _mm256_and_ps(fscal,cutoff_mask);
 815
 816             /* Calculate temporary vectorial force */
 817             tx               = _mm256_mul_ps(fscal,dx10);
 818             ty               = _mm256_mul_ps(fscal,dy10);
 819             tz               = _mm256_mul_ps(fscal,dz10);
 820
 821             /* Update vectorial force */
 822             fix1             = _mm256_add_ps(fix1,tx);
 823             fiy1             = _mm256_add_ps(fiy1,ty);
 824             fiz1             = _mm256_add_ps(fiz1,tz);
 825
 826             fjx0             = _mm256_add_ps(fjx0,tx);
 827             fjy0             = _mm256_add_ps(fjy0,ty);
 828             fjz0             = _mm256_add_ps(fjz0,tz);
 829
 830             }
 831
 832             /**************************
 833              * CALCULATE INTERACTIONS *
 834              **************************/
 835
 836             if (gmx_mm256_any_lt(rsq20,rcutoff2))
 837             {
 838
 839             /* Compute parameters for interactions between i and j atoms */
 840             qq20             = _mm256_mul_ps(iq2,jq0);
 841
 842             /* REACTION-FIELD ELECTROSTATICS */
 843             felec            = _mm256_mul_ps(qq20,_mm256_sub_ps(_mm256_mul_ps(rinv20,rinvsq20),krf2));
 844
 845             cutoff_mask      = _mm256_cmp_ps(rsq20,rcutoff2,_CMP_LT_OQ);
 846
 847             fscal            = felec;
 848
 849             fscal            = _mm256_and_ps(fscal,cutoff_mask);
 850
 851             /* Calculate temporary vectorial force */
 852             tx               = _mm256_mul_ps(fscal,dx20);
 853             ty               = _mm256_mul_ps(fscal,dy20);
 854             tz               = _mm256_mul_ps(fscal,dz20);
 855
 856             /* Update vectorial force */
 857             fix2             = _mm256_add_ps(fix2,tx);
 858             fiy2             = _mm256_add_ps(fiy2,ty);
 859             fiz2             = _mm256_add_ps(fiz2,tz);
 860
 861             fjx0             = _mm256_add_ps(fjx0,tx);
 862             fjy0             = _mm256_add_ps(fjy0,ty);
 863             fjz0             = _mm256_add_ps(fjz0,tz);
 864
 865             }
 866
 867             /**************************
 868              * CALCULATE INTERACTIONS *
 869              **************************/
 870
 871             if (gmx_mm256_any_lt(rsq30,rcutoff2))
 872             {
 873
 874             /* Compute parameters for interactions between i and j atoms */
 875             qq30             = _mm256_mul_ps(iq3,jq0);
 876
 877             /* REACTION-FIELD ELECTROSTATICS */
 878             felec            = _mm256_mul_ps(qq30,_mm256_sub_ps(_mm256_mul_ps(rinv30,rinvsq30),krf2));
 879
 880             cutoff_mask      = _mm256_cmp_ps(rsq30,rcutoff2,_CMP_LT_OQ);
 881
 882             fscal            = felec;
 883
 884             fscal            = _mm256_and_ps(fscal,cutoff_mask);
 885
 886             /* Calculate temporary vectorial force */
 887             tx               = _mm256_mul_ps(fscal,dx30);
 888             ty               = _mm256_mul_ps(fscal,dy30);
 889             tz               = _mm256_mul_ps(fscal,dz30);
 890
 891             /* Update vectorial force */
 892             fix3             = _mm256_add_ps(fix3,tx);
 893             fiy3             = _mm256_add_ps(fiy3,ty);
 894             fiz3             = _mm256_add_ps(fiz3,tz);
 895
 896             fjx0             = _mm256_add_ps(fjx0,tx);
 897             fjy0             = _mm256_add_ps(fjy0,ty);
 898             fjz0             = _mm256_add_ps(fjz0,tz);
 899
 900             }
 901
 902             fjptrA             = f+j_coord_offsetA;
 903             fjptrB             = f+j_coord_offsetB;
 904             fjptrC             = f+j_coord_offsetC;
 905             fjptrD             = f+j_coord_offsetD;
 906             fjptrE             = f+j_coord_offsetE;
 907             fjptrF             = f+j_coord_offsetF;
 908             fjptrG             = f+j_coord_offsetG;
 909             fjptrH             = f+j_coord_offsetH;
 910
 911             gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,fjx0,fjy0,fjz0);
 912
 913             /* Inner loop uses 93 flops */
 914         }
 915
 916         if(jidx<j_index_end)
 917         {
 918
 919             /* Get j neighbor index, and coordinate index */
 920             jnrlistA         = jjnr[jidx];
 921             jnrlistB         = jjnr[jidx+1];
 922             jnrlistC         = jjnr[jidx+2];
 923             jnrlistD         = jjnr[jidx+3];
 924             jnrlistE         = jjnr[jidx+4];
 925             jnrlistF         = jjnr[jidx+5];
 926             jnrlistG         = jjnr[jidx+6];
 927             jnrlistH         = jjnr[jidx+7];
 928             /* Sign of each element will be negative for non-real atoms.
 929              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 930              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 931              */
 932             dummy_mask = gmx_mm256_set_m128(gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx+4)),_mm_setzero_si128())),
 933                                             gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128())));
 934
 935             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 936             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 937             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 938             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 939             jnrE       = (jnrlistE>=0) ? jnrlistE : 0;
 940             jnrF       = (jnrlistF>=0) ? jnrlistF : 0;
 941             jnrG       = (jnrlistG>=0) ? jnrlistG : 0;
 942             jnrH       = (jnrlistH>=0) ? jnrlistH : 0;
 943             j_coord_offsetA  = DIM*jnrA;
 944             j_coord_offsetB  = DIM*jnrB;
 945             j_coord_offsetC  = DIM*jnrC;
 946             j_coord_offsetD  = DIM*jnrD;
 947             j_coord_offsetE  = DIM*jnrE;
 948             j_coord_offsetF  = DIM*jnrF;
 949             j_coord_offsetG  = DIM*jnrG;
 950             j_coord_offsetH  = DIM*jnrH;
 951
 952             /* load j atom coordinates */
 953             gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 954                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 955                                                  x+j_coord_offsetE,x+j_coord_offsetF,
 956                                                  x+j_coord_offsetG,x+j_coord_offsetH,
 957                                                  &jx0,&jy0,&jz0);
 958
 959             /* Calculate displacement vector */
 960             dx10             = _mm256_sub_ps(ix1,jx0);
 961             dy10             = _mm256_sub_ps(iy1,jy0);
 962             dz10             = _mm256_sub_ps(iz1,jz0);
 963             dx20             = _mm256_sub_ps(ix2,jx0);
 964             dy20             = _mm256_sub_ps(iy2,jy0);
 965             dz20             = _mm256_sub_ps(iz2,jz0);
 966             dx30             = _mm256_sub_ps(ix3,jx0);
 967             dy30             = _mm256_sub_ps(iy3,jy0);
 968             dz30             = _mm256_sub_ps(iz3,jz0);
 969
 970             /* Calculate squared distance and things based on it */
 971             rsq10            = gmx_mm256_calc_rsq_ps(dx10,dy10,dz10);
 972             rsq20            = gmx_mm256_calc_rsq_ps(dx20,dy20,dz20);
 973             rsq30            = gmx_mm256_calc_rsq_ps(dx30,dy30,dz30);
 974
 975             rinv10           = gmx_mm256_invsqrt_ps(rsq10);
 976             rinv20           = gmx_mm256_invsqrt_ps(rsq20);
 977             rinv30           = gmx_mm256_invsqrt_ps(rsq30);
 978
 979             rinvsq10         = _mm256_mul_ps(rinv10,rinv10);
 980             rinvsq20         = _mm256_mul_ps(rinv20,rinv20);
 981             rinvsq30         = _mm256_mul_ps(rinv30,rinv30);
 982
 983             /* Load parameters for j particles */
 984             jq0              = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 985                                                                  charge+jnrC+0,charge+jnrD+0,
 986                                                                  charge+jnrE+0,charge+jnrF+0,
 987                                                                  charge+jnrG+0,charge+jnrH+0);
 988
 989             fjx0             = _mm256_setzero_ps();
 990             fjy0             = _mm256_setzero_ps();
 991             fjz0             = _mm256_setzero_ps();
 992
 993             /**************************
 994              * CALCULATE INTERACTIONS *
 995              **************************/
 996
 997             if (gmx_mm256_any_lt(rsq10,rcutoff2))
 998             {
 999
1000             /* Compute parameters for interactions between i and j atoms */
1001             qq10             = _mm256_mul_ps(iq1,jq0);
1002
1003             /* REACTION-FIELD ELECTROSTATICS */
1004             felec            = _mm256_mul_ps(qq10,_mm256_sub_ps(_mm256_mul_ps(rinv10,rinvsq10),krf2));
1005
1006             cutoff_mask      = _mm256_cmp_ps(rsq10,rcutoff2,_CMP_LT_OQ);
1007
1008             fscal            = felec;
1009
1010             fscal            = _mm256_and_ps(fscal,cutoff_mask);
1011
1012             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
1013
1014             /* Calculate temporary vectorial force */
1015             tx               = _mm256_mul_ps(fscal,dx10);
1016             ty               = _mm256_mul_ps(fscal,dy10);
1017             tz               = _mm256_mul_ps(fscal,dz10);
1018
1019             /* Update vectorial force */
1020             fix1             = _mm256_add_ps(fix1,tx);
1021             fiy1             = _mm256_add_ps(fiy1,ty);
1022             fiz1             = _mm256_add_ps(fiz1,tz);
1023
1024             fjx0             = _mm256_add_ps(fjx0,tx);
1025             fjy0             = _mm256_add_ps(fjy0,ty);
1026             fjz0             = _mm256_add_ps(fjz0,tz);
1027
1028             }
1029
1030             /**************************
1031              * CALCULATE INTERACTIONS *
1032              **************************/
1033
1034             if (gmx_mm256_any_lt(rsq20,rcutoff2))
1035             {
1036
1037             /* Compute parameters for interactions between i and j atoms */
1038             qq20             = _mm256_mul_ps(iq2,jq0);
1039
1040             /* REACTION-FIELD ELECTROSTATICS */
1041             felec            = _mm256_mul_ps(qq20,_mm256_sub_ps(_mm256_mul_ps(rinv20,rinvsq20),krf2));
1042
1043             cutoff_mask      = _mm256_cmp_ps(rsq20,rcutoff2,_CMP_LT_OQ);
1044
1045             fscal            = felec;
1046
1047             fscal            = _mm256_and_ps(fscal,cutoff_mask);
1048
1049             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
1050
1051             /* Calculate temporary vectorial force */
1052             tx               = _mm256_mul_ps(fscal,dx20);
1053             ty               = _mm256_mul_ps(fscal,dy20);
1054             tz               = _mm256_mul_ps(fscal,dz20);
1055
1056             /* Update vectorial force */
1057             fix2             = _mm256_add_ps(fix2,tx);
1058             fiy2             = _mm256_add_ps(fiy2,ty);
1059             fiz2             = _mm256_add_ps(fiz2,tz);
1060
1061             fjx0             = _mm256_add_ps(fjx0,tx);
1062             fjy0             = _mm256_add_ps(fjy0,ty);
1063             fjz0             = _mm256_add_ps(fjz0,tz);
1064
1065             }
1066
1067             /**************************
1068              * CALCULATE INTERACTIONS *
1069              **************************/
1070
1071             if (gmx_mm256_any_lt(rsq30,rcutoff2))
1072             {
1073
1074             /* Compute parameters for interactions between i and j atoms */
1075             qq30             = _mm256_mul_ps(iq3,jq0);
1076
1077             /* REACTION-FIELD ELECTROSTATICS */
1078             felec            = _mm256_mul_ps(qq30,_mm256_sub_ps(_mm256_mul_ps(rinv30,rinvsq30),krf2));
1079
1080             cutoff_mask      = _mm256_cmp_ps(rsq30,rcutoff2,_CMP_LT_OQ);
1081
1082             fscal            = felec;
1083
1084             fscal            = _mm256_and_ps(fscal,cutoff_mask);
1085
1086             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
1087
1088             /* Calculate temporary vectorial force */
1089             tx               = _mm256_mul_ps(fscal,dx30);
1090             ty               = _mm256_mul_ps(fscal,dy30);
1091             tz               = _mm256_mul_ps(fscal,dz30);
1092
1093             /* Update vectorial force */
1094             fix3             = _mm256_add_ps(fix3,tx);
1095             fiy3             = _mm256_add_ps(fiy3,ty);
1096             fiz3             = _mm256_add_ps(fiz3,tz);
1097
1098             fjx0             = _mm256_add_ps(fjx0,tx);
1099             fjy0             = _mm256_add_ps(fjy0,ty);
1100             fjz0             = _mm256_add_ps(fjz0,tz);
1101
1102             }
1103
1104             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
1105             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
1106             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
1107             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
1108             fjptrE             = (jnrlistE>=0) ? f+j_coord_offsetE : scratch;
1109             fjptrF             = (jnrlistF>=0) ? f+j_coord_offsetF : scratch;
1110             fjptrG             = (jnrlistG>=0) ? f+j_coord_offsetG : scratch;
1111             fjptrH             = (jnrlistH>=0) ? f+j_coord_offsetH : scratch;
1112
1113             gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,fjx0,fjy0,fjz0);
1114
1115             /* Inner loop uses 93 flops */
1116         }
1117
1118         /* End of innermost loop */
1119
1120         gmx_mm256_update_iforce_3atom_swizzle_ps(fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
1121                                                  f+i_coord_offset+DIM,fshift+i_shift_offset);
1122
1123         /* Increment number of inner iterations */
1124         inneriter                  += j_index_end - j_index_start;
1125
1126         /* Outer loop uses 18 flops */
1127     }
1128
1129     /* Increment number of outer iterations */
1130     outeriter        += nri;
1131
1132     /* Update outer/inner flops */
1133
1134     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_W4_F,outeriter*18 + inneriter*93);
1135 }