src/gromacs/gmxlib/nonbonded/nb_kernel_avx_256_double/nb_kernel_ElecCoul_VdwNone_GeomP1P1_avx_256_double.c

   1 /*
   2  * Note: this file was generated by the Gromacs avx_256_double 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_double.h"
  34 #include "kernelutil_x86_avx_256_double.h"
  35
  36 /*
  37  * Gromacs nonbonded kernel:   nb_kernel_ElecCoul_VdwNone_GeomP1P1_VF_avx_256_double
  38  * Electrostatics interaction: Coulomb
  39  * VdW interaction:            None
  40  * Geometry:                   Particle-Particle
  41  * Calculate force/pot:        PotentialAndForce
  42  */
  43 void
  44 nb_kernel_ElecCoul_VdwNone_GeomP1P1_VF_avx_256_double
  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 refer to j loop unrolling done with AVX, e.g. for the four 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              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
  62     int              jnrlistE,jnrlistF,jnrlistG,jnrlistH;
  63     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
  64     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
  65     real             rcutoff_scalar;
  66     real             *shiftvec,*fshift,*x,*f;
  67     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
  68     real             scratch[4*DIM];
  69     __m256d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
  70     real *           vdwioffsetptr0;
  71     __m256d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
  72     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
  73     __m256d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  74     __m256d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  75     __m256d          velec,felec,velecsum,facel,crf,krf,krf2;
  76     real             *charge;
  77     __m256d          dummy_mask,cutoff_mask;
  78     __m128           tmpmask0,tmpmask1;
  79     __m256d          signbit = _mm256_castsi256_pd( _mm256_set1_epi32(0x80000000) );
  80     __m256d          one     = _mm256_set1_pd(1.0);
  81     __m256d          two     = _mm256_set1_pd(2.0);
  82     x                = xx[0];
  83     f                = ff[0];
  84
  85     nri              = nlist->nri;
  86     iinr             = nlist->iinr;
  87     jindex           = nlist->jindex;
  88     jjnr             = nlist->jjnr;
  89     shiftidx         = nlist->shift;
  90     gid              = nlist->gid;
  91     shiftvec         = fr->shift_vec[0];
  92     fshift           = fr->fshift[0];
  93     facel            = _mm256_set1_pd(fr->epsfac);
  94     charge           = mdatoms->chargeA;
  95
  96     /* Avoid stupid compiler warnings */
  97     jnrA = jnrB = jnrC = jnrD = 0;
  98     j_coord_offsetA = 0;
  99     j_coord_offsetB = 0;
 100     j_coord_offsetC = 0;
 101     j_coord_offsetD = 0;
 102
 103     outeriter        = 0;
 104     inneriter        = 0;
 105
 106     for(iidx=0;iidx<4*DIM;iidx++)
 107     {
 108         scratch[iidx] = 0.0;
 109     }
 110
 111     /* Start outer loop over neighborlists */
 112     for(iidx=0; iidx<nri; iidx++)
 113     {
 114         /* Load shift vector for this list */
 115         i_shift_offset   = DIM*shiftidx[iidx];
 116
 117         /* Load limits for loop over neighbors */
 118         j_index_start    = jindex[iidx];
 119         j_index_end      = jindex[iidx+1];
 120
 121         /* Get outer coordinate index */
 122         inr              = iinr[iidx];
 123         i_coord_offset   = DIM*inr;
 124
 125         /* Load i particle coords and add shift vector */
 126         gmx_mm256_load_shift_and_1rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 127
 128         fix0             = _mm256_setzero_pd();
 129         fiy0             = _mm256_setzero_pd();
 130         fiz0             = _mm256_setzero_pd();
 131
 132         /* Load parameters for i particles */
 133         iq0              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+0]));
 134
 135         /* Reset potential sums */
 136         velecsum         = _mm256_setzero_pd();
 137
 138         /* Start inner kernel loop */
 139         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 140         {
 141
 142             /* Get j neighbor index, and coordinate index */
 143             jnrA             = jjnr[jidx];
 144             jnrB             = jjnr[jidx+1];
 145             jnrC             = jjnr[jidx+2];
 146             jnrD             = jjnr[jidx+3];
 147             j_coord_offsetA  = DIM*jnrA;
 148             j_coord_offsetB  = DIM*jnrB;
 149             j_coord_offsetC  = DIM*jnrC;
 150             j_coord_offsetD  = DIM*jnrD;
 151
 152             /* load j atom coordinates */
 153             gmx_mm256_load_1rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 154                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 155                                                  &jx0,&jy0,&jz0);
 156
 157             /* Calculate displacement vector */
 158             dx00             = _mm256_sub_pd(ix0,jx0);
 159             dy00             = _mm256_sub_pd(iy0,jy0);
 160             dz00             = _mm256_sub_pd(iz0,jz0);
 161
 162             /* Calculate squared distance and things based on it */
 163             rsq00            = gmx_mm256_calc_rsq_pd(dx00,dy00,dz00);
 164
 165             rinv00           = gmx_mm256_invsqrt_pd(rsq00);
 166
 167             rinvsq00         = _mm256_mul_pd(rinv00,rinv00);
 168
 169             /* Load parameters for j particles */
 170             jq0              = gmx_mm256_load_4real_swizzle_pd(charge+jnrA+0,charge+jnrB+0,
 171                                                                  charge+jnrC+0,charge+jnrD+0);
 172
 173             /**************************
 174              * CALCULATE INTERACTIONS *
 175              **************************/
 176
 177             /* Compute parameters for interactions between i and j atoms */
 178             qq00             = _mm256_mul_pd(iq0,jq0);
 179
 180             /* COULOMB ELECTROSTATICS */
 181             velec            = _mm256_mul_pd(qq00,rinv00);
 182             felec            = _mm256_mul_pd(velec,rinvsq00);
 183
 184             /* Update potential sum for this i atom from the interaction with this j atom. */
 185             velecsum         = _mm256_add_pd(velecsum,velec);
 186
 187             fscal            = felec;
 188
 189             /* Calculate temporary vectorial force */
 190             tx               = _mm256_mul_pd(fscal,dx00);
 191             ty               = _mm256_mul_pd(fscal,dy00);
 192             tz               = _mm256_mul_pd(fscal,dz00);
 193
 194             /* Update vectorial force */
 195             fix0             = _mm256_add_pd(fix0,tx);
 196             fiy0             = _mm256_add_pd(fiy0,ty);
 197             fiz0             = _mm256_add_pd(fiz0,tz);
 198
 199             fjptrA             = f+j_coord_offsetA;
 200             fjptrB             = f+j_coord_offsetB;
 201             fjptrC             = f+j_coord_offsetC;
 202             fjptrD             = f+j_coord_offsetD;
 203             gmx_mm256_decrement_1rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 204
 205             /* Inner loop uses 27 flops */
 206         }
 207
 208         if(jidx<j_index_end)
 209         {
 210
 211             /* Get j neighbor index, and coordinate index */
 212             jnrlistA         = jjnr[jidx];
 213             jnrlistB         = jjnr[jidx+1];
 214             jnrlistC         = jjnr[jidx+2];
 215             jnrlistD         = jjnr[jidx+3];
 216             /* Sign of each element will be negative for non-real atoms.
 217              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 218              * so use it as val = _mm_andnot_pd(mask,val) to clear dummy entries.
 219              */
 220             tmpmask0 = gmx_mm_castsi128_pd(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 221
 222             tmpmask1 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(3,3,2,2));
 223             tmpmask0 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(1,1,0,0));
 224             dummy_mask = _mm256_castps_pd(gmx_mm256_set_m128(tmpmask1,tmpmask0));
 225
 226             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 227             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 228             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 229             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 230             j_coord_offsetA  = DIM*jnrA;
 231             j_coord_offsetB  = DIM*jnrB;
 232             j_coord_offsetC  = DIM*jnrC;
 233             j_coord_offsetD  = DIM*jnrD;
 234
 235             /* load j atom coordinates */
 236             gmx_mm256_load_1rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 237                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 238                                                  &jx0,&jy0,&jz0);
 239
 240             /* Calculate displacement vector */
 241             dx00             = _mm256_sub_pd(ix0,jx0);
 242             dy00             = _mm256_sub_pd(iy0,jy0);
 243             dz00             = _mm256_sub_pd(iz0,jz0);
 244
 245             /* Calculate squared distance and things based on it */
 246             rsq00            = gmx_mm256_calc_rsq_pd(dx00,dy00,dz00);
 247
 248             rinv00           = gmx_mm256_invsqrt_pd(rsq00);
 249
 250             rinvsq00         = _mm256_mul_pd(rinv00,rinv00);
 251
 252             /* Load parameters for j particles */
 253             jq0              = gmx_mm256_load_4real_swizzle_pd(charge+jnrA+0,charge+jnrB+0,
 254                                                                  charge+jnrC+0,charge+jnrD+0);
 255
 256             /**************************
 257              * CALCULATE INTERACTIONS *
 258              **************************/
 259
 260             /* Compute parameters for interactions between i and j atoms */
 261             qq00             = _mm256_mul_pd(iq0,jq0);
 262
 263             /* COULOMB ELECTROSTATICS */
 264             velec            = _mm256_mul_pd(qq00,rinv00);
 265             felec            = _mm256_mul_pd(velec,rinvsq00);
 266
 267             /* Update potential sum for this i atom from the interaction with this j atom. */
 268             velec            = _mm256_andnot_pd(dummy_mask,velec);
 269             velecsum         = _mm256_add_pd(velecsum,velec);
 270
 271             fscal            = felec;
 272
 273             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
 274
 275             /* Calculate temporary vectorial force */
 276             tx               = _mm256_mul_pd(fscal,dx00);
 277             ty               = _mm256_mul_pd(fscal,dy00);
 278             tz               = _mm256_mul_pd(fscal,dz00);
 279
 280             /* Update vectorial force */
 281             fix0             = _mm256_add_pd(fix0,tx);
 282             fiy0             = _mm256_add_pd(fiy0,ty);
 283             fiz0             = _mm256_add_pd(fiz0,tz);
 284
 285             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 286             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 287             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 288             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 289             gmx_mm256_decrement_1rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 290
 291             /* Inner loop uses 27 flops */
 292         }
 293
 294         /* End of innermost loop */
 295
 296         gmx_mm256_update_iforce_1atom_swizzle_pd(fix0,fiy0,fiz0,
 297                                                  f+i_coord_offset,fshift+i_shift_offset);
 298
 299         ggid                        = gid[iidx];
 300         /* Update potential energies */
 301         gmx_mm256_update_1pot_pd(velecsum,kernel_data->energygrp_elec+ggid);
 302
 303         /* Increment number of inner iterations */
 304         inneriter                  += j_index_end - j_index_start;
 305
 306         /* Outer loop uses 8 flops */
 307     }
 308
 309     /* Increment number of outer iterations */
 310     outeriter        += nri;
 311
 312     /* Update outer/inner flops */
 313
 314     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VF,outeriter*8 + inneriter*27);
 315 }
 316 /*
 317  * Gromacs nonbonded kernel:   nb_kernel_ElecCoul_VdwNone_GeomP1P1_F_avx_256_double
 318  * Electrostatics interaction: Coulomb
 319  * VdW interaction:            None
 320  * Geometry:                   Particle-Particle
 321  * Calculate force/pot:        Force
 322  */
 323 void
 324 nb_kernel_ElecCoul_VdwNone_GeomP1P1_F_avx_256_double
 325                     (t_nblist * gmx_restrict                nlist,
 326                      rvec * gmx_restrict                    xx,
 327                      rvec * gmx_restrict                    ff,
 328                      t_forcerec * gmx_restrict              fr,
 329                      t_mdatoms * gmx_restrict               mdatoms,
 330                      nb_kernel_data_t * gmx_restrict        kernel_data,
 331                      t_nrnb * gmx_restrict                  nrnb)
 332 {
 333     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 334      * just 0 for non-waters.
 335      * Suffixes A,B,C,D refer to j loop unrolling done with AVX, e.g. for the four different
 336      * jnr indices corresponding to data put in the four positions in the SIMD register.
 337      */
 338     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 339     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 340     int              jnrA,jnrB,jnrC,jnrD;
 341     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
 342     int              jnrlistE,jnrlistF,jnrlistG,jnrlistH;
 343     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 344     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 345     real             rcutoff_scalar;
 346     real             *shiftvec,*fshift,*x,*f;
 347     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
 348     real             scratch[4*DIM];
 349     __m256d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 350     real *           vdwioffsetptr0;
 351     __m256d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 352     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
 353     __m256d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 354     __m256d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 355     __m256d          velec,felec,velecsum,facel,crf,krf,krf2;
 356     real             *charge;
 357     __m256d          dummy_mask,cutoff_mask;
 358     __m128           tmpmask0,tmpmask1;
 359     __m256d          signbit = _mm256_castsi256_pd( _mm256_set1_epi32(0x80000000) );
 360     __m256d          one     = _mm256_set1_pd(1.0);
 361     __m256d          two     = _mm256_set1_pd(2.0);
 362     x                = xx[0];
 363     f                = ff[0];
 364
 365     nri              = nlist->nri;
 366     iinr             = nlist->iinr;
 367     jindex           = nlist->jindex;
 368     jjnr             = nlist->jjnr;
 369     shiftidx         = nlist->shift;
 370     gid              = nlist->gid;
 371     shiftvec         = fr->shift_vec[0];
 372     fshift           = fr->fshift[0];
 373     facel            = _mm256_set1_pd(fr->epsfac);
 374     charge           = mdatoms->chargeA;
 375
 376     /* Avoid stupid compiler warnings */
 377     jnrA = jnrB = jnrC = jnrD = 0;
 378     j_coord_offsetA = 0;
 379     j_coord_offsetB = 0;
 380     j_coord_offsetC = 0;
 381     j_coord_offsetD = 0;
 382
 383     outeriter        = 0;
 384     inneriter        = 0;
 385
 386     for(iidx=0;iidx<4*DIM;iidx++)
 387     {
 388         scratch[iidx] = 0.0;
 389     }
 390
 391     /* Start outer loop over neighborlists */
 392     for(iidx=0; iidx<nri; iidx++)
 393     {
 394         /* Load shift vector for this list */
 395         i_shift_offset   = DIM*shiftidx[iidx];
 396
 397         /* Load limits for loop over neighbors */
 398         j_index_start    = jindex[iidx];
 399         j_index_end      = jindex[iidx+1];
 400
 401         /* Get outer coordinate index */
 402         inr              = iinr[iidx];
 403         i_coord_offset   = DIM*inr;
 404
 405         /* Load i particle coords and add shift vector */
 406         gmx_mm256_load_shift_and_1rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 407
 408         fix0             = _mm256_setzero_pd();
 409         fiy0             = _mm256_setzero_pd();
 410         fiz0             = _mm256_setzero_pd();
 411
 412         /* Load parameters for i particles */
 413         iq0              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+0]));
 414
 415         /* Start inner kernel loop */
 416         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 417         {
 418
 419             /* Get j neighbor index, and coordinate index */
 420             jnrA             = jjnr[jidx];
 421             jnrB             = jjnr[jidx+1];
 422             jnrC             = jjnr[jidx+2];
 423             jnrD             = jjnr[jidx+3];
 424             j_coord_offsetA  = DIM*jnrA;
 425             j_coord_offsetB  = DIM*jnrB;
 426             j_coord_offsetC  = DIM*jnrC;
 427             j_coord_offsetD  = DIM*jnrD;
 428
 429             /* load j atom coordinates */
 430             gmx_mm256_load_1rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 431                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 432                                                  &jx0,&jy0,&jz0);
 433
 434             /* Calculate displacement vector */
 435             dx00             = _mm256_sub_pd(ix0,jx0);
 436             dy00             = _mm256_sub_pd(iy0,jy0);
 437             dz00             = _mm256_sub_pd(iz0,jz0);
 438
 439             /* Calculate squared distance and things based on it */
 440             rsq00            = gmx_mm256_calc_rsq_pd(dx00,dy00,dz00);
 441
 442             rinv00           = gmx_mm256_invsqrt_pd(rsq00);
 443
 444             rinvsq00         = _mm256_mul_pd(rinv00,rinv00);
 445
 446             /* Load parameters for j particles */
 447             jq0              = gmx_mm256_load_4real_swizzle_pd(charge+jnrA+0,charge+jnrB+0,
 448                                                                  charge+jnrC+0,charge+jnrD+0);
 449
 450             /**************************
 451              * CALCULATE INTERACTIONS *
 452              **************************/
 453
 454             /* Compute parameters for interactions between i and j atoms */
 455             qq00             = _mm256_mul_pd(iq0,jq0);
 456
 457             /* COULOMB ELECTROSTATICS */
 458             velec            = _mm256_mul_pd(qq00,rinv00);
 459             felec            = _mm256_mul_pd(velec,rinvsq00);
 460
 461             fscal            = felec;
 462
 463             /* Calculate temporary vectorial force */
 464             tx               = _mm256_mul_pd(fscal,dx00);
 465             ty               = _mm256_mul_pd(fscal,dy00);
 466             tz               = _mm256_mul_pd(fscal,dz00);
 467
 468             /* Update vectorial force */
 469             fix0             = _mm256_add_pd(fix0,tx);
 470             fiy0             = _mm256_add_pd(fiy0,ty);
 471             fiz0             = _mm256_add_pd(fiz0,tz);
 472
 473             fjptrA             = f+j_coord_offsetA;
 474             fjptrB             = f+j_coord_offsetB;
 475             fjptrC             = f+j_coord_offsetC;
 476             fjptrD             = f+j_coord_offsetD;
 477             gmx_mm256_decrement_1rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 478
 479             /* Inner loop uses 26 flops */
 480         }
 481
 482         if(jidx<j_index_end)
 483         {
 484
 485             /* Get j neighbor index, and coordinate index */
 486             jnrlistA         = jjnr[jidx];
 487             jnrlistB         = jjnr[jidx+1];
 488             jnrlistC         = jjnr[jidx+2];
 489             jnrlistD         = jjnr[jidx+3];
 490             /* Sign of each element will be negative for non-real atoms.
 491              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 492              * so use it as val = _mm_andnot_pd(mask,val) to clear dummy entries.
 493              */
 494             tmpmask0 = gmx_mm_castsi128_pd(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 495
 496             tmpmask1 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(3,3,2,2));
 497             tmpmask0 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(1,1,0,0));
 498             dummy_mask = _mm256_castps_pd(gmx_mm256_set_m128(tmpmask1,tmpmask0));
 499
 500             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 501             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 502             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 503             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 504             j_coord_offsetA  = DIM*jnrA;
 505             j_coord_offsetB  = DIM*jnrB;
 506             j_coord_offsetC  = DIM*jnrC;
 507             j_coord_offsetD  = DIM*jnrD;
 508
 509             /* load j atom coordinates */
 510             gmx_mm256_load_1rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 511                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 512                                                  &jx0,&jy0,&jz0);
 513
 514             /* Calculate displacement vector */
 515             dx00             = _mm256_sub_pd(ix0,jx0);
 516             dy00             = _mm256_sub_pd(iy0,jy0);
 517             dz00             = _mm256_sub_pd(iz0,jz0);
 518
 519             /* Calculate squared distance and things based on it */
 520             rsq00            = gmx_mm256_calc_rsq_pd(dx00,dy00,dz00);
 521
 522             rinv00           = gmx_mm256_invsqrt_pd(rsq00);
 523
 524             rinvsq00         = _mm256_mul_pd(rinv00,rinv00);
 525
 526             /* Load parameters for j particles */
 527             jq0              = gmx_mm256_load_4real_swizzle_pd(charge+jnrA+0,charge+jnrB+0,
 528                                                                  charge+jnrC+0,charge+jnrD+0);
 529
 530             /**************************
 531              * CALCULATE INTERACTIONS *
 532              **************************/
 533
 534             /* Compute parameters for interactions between i and j atoms */
 535             qq00             = _mm256_mul_pd(iq0,jq0);
 536
 537             /* COULOMB ELECTROSTATICS */
 538             velec            = _mm256_mul_pd(qq00,rinv00);
 539             felec            = _mm256_mul_pd(velec,rinvsq00);
 540
 541             fscal            = felec;
 542
 543             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
 544
 545             /* Calculate temporary vectorial force */
 546             tx               = _mm256_mul_pd(fscal,dx00);
 547             ty               = _mm256_mul_pd(fscal,dy00);
 548             tz               = _mm256_mul_pd(fscal,dz00);
 549
 550             /* Update vectorial force */
 551             fix0             = _mm256_add_pd(fix0,tx);
 552             fiy0             = _mm256_add_pd(fiy0,ty);
 553             fiz0             = _mm256_add_pd(fiz0,tz);
 554
 555             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 556             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 557             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 558             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 559             gmx_mm256_decrement_1rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 560
 561             /* Inner loop uses 26 flops */
 562         }
 563
 564         /* End of innermost loop */
 565
 566         gmx_mm256_update_iforce_1atom_swizzle_pd(fix0,fiy0,fiz0,
 567                                                  f+i_coord_offset,fshift+i_shift_offset);
 568
 569         /* Increment number of inner iterations */
 570         inneriter                  += j_index_end - j_index_start;
 571
 572         /* Outer loop uses 7 flops */
 573     }
 574
 575     /* Increment number of outer iterations */
 576     outeriter        += nri;
 577
 578     /* Update outer/inner flops */
 579
 580     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_F,outeriter*7 + inneriter*26);
 581 }