src/gmxlib/nonbonded/nb_kernel_avx_256_single/nb_kernel_ElecRF_VdwNone_GeomP1P1_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_ElecRF_VdwNone_GeomP1P1_VF_avx_256_single
  38  * Electrostatics interaction: ReactionField
  39  * VdW interaction:            None
  40  * Geometry:                   Particle-Particle
  41  * Calculate force/pot:        PotentialAndForce
  42  */
  43 void
  44 nb_kernel_ElecRF_VdwNone_GeomP1P1_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 *           vdwioffsetptr0;
  73     __m256           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
  74     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D,vdwjidx0E,vdwjidx0F,vdwjidx0G,vdwjidx0H;
  75     __m256           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  76     __m256           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  77     __m256           velec,felec,velecsum,facel,crf,krf,krf2;
  78     real             *charge;
  79     __m256           dummy_mask,cutoff_mask;
  80     __m256           signbit = _mm256_castsi256_ps( _mm256_set1_epi32(0x80000000) );
  81     __m256           one     = _mm256_set1_ps(1.0);
  82     __m256           two     = _mm256_set1_ps(2.0);
  83     x                = xx[0];
  84     f                = ff[0];
  85
  86     nri              = nlist->nri;
  87     iinr             = nlist->iinr;
  88     jindex           = nlist->jindex;
  89     jjnr             = nlist->jjnr;
  90     shiftidx         = nlist->shift;
  91     gid              = nlist->gid;
  92     shiftvec         = fr->shift_vec[0];
  93     fshift           = fr->fshift[0];
  94     facel            = _mm256_set1_ps(fr->epsfac);
  95     charge           = mdatoms->chargeA;
  96     krf              = _mm256_set1_ps(fr->ic->k_rf);
  97     krf2             = _mm256_set1_ps(fr->ic->k_rf*2.0);
  98     crf              = _mm256_set1_ps(fr->ic->c_rf);
  99
 100     /* Avoid stupid compiler warnings */
 101     jnrA = jnrB = jnrC = jnrD = jnrE = jnrF = jnrG = jnrH = 0;
 102     j_coord_offsetA = 0;
 103     j_coord_offsetB = 0;
 104     j_coord_offsetC = 0;
 105     j_coord_offsetD = 0;
 106     j_coord_offsetE = 0;
 107     j_coord_offsetF = 0;
 108     j_coord_offsetG = 0;
 109     j_coord_offsetH = 0;
 110
 111     outeriter        = 0;
 112     inneriter        = 0;
 113
 114     for(iidx=0;iidx<4*DIM;iidx++)
 115     {
 116         scratch[iidx] = 0.0;
 117     }
 118
 119     /* Start outer loop over neighborlists */
 120     for(iidx=0; iidx<nri; iidx++)
 121     {
 122         /* Load shift vector for this list */
 123         i_shift_offset   = DIM*shiftidx[iidx];
 124
 125         /* Load limits for loop over neighbors */
 126         j_index_start    = jindex[iidx];
 127         j_index_end      = jindex[iidx+1];
 128
 129         /* Get outer coordinate index */
 130         inr              = iinr[iidx];
 131         i_coord_offset   = DIM*inr;
 132
 133         /* Load i particle coords and add shift vector */
 134         gmx_mm256_load_shift_and_1rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 135
 136         fix0             = _mm256_setzero_ps();
 137         fiy0             = _mm256_setzero_ps();
 138         fiz0             = _mm256_setzero_ps();
 139
 140         /* Load parameters for i particles */
 141         iq0              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+0]));
 142
 143         /* Reset potential sums */
 144         velecsum         = _mm256_setzero_ps();
 145
 146         /* Start inner kernel loop */
 147         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+7]>=0; jidx+=8)
 148         {
 149
 150             /* Get j neighbor index, and coordinate index */
 151             jnrA             = jjnr[jidx];
 152             jnrB             = jjnr[jidx+1];
 153             jnrC             = jjnr[jidx+2];
 154             jnrD             = jjnr[jidx+3];
 155             jnrE             = jjnr[jidx+4];
 156             jnrF             = jjnr[jidx+5];
 157             jnrG             = jjnr[jidx+6];
 158             jnrH             = jjnr[jidx+7];
 159             j_coord_offsetA  = DIM*jnrA;
 160             j_coord_offsetB  = DIM*jnrB;
 161             j_coord_offsetC  = DIM*jnrC;
 162             j_coord_offsetD  = DIM*jnrD;
 163             j_coord_offsetE  = DIM*jnrE;
 164             j_coord_offsetF  = DIM*jnrF;
 165             j_coord_offsetG  = DIM*jnrG;
 166             j_coord_offsetH  = DIM*jnrH;
 167
 168             /* load j atom coordinates */
 169             gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 170                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 171                                                  x+j_coord_offsetE,x+j_coord_offsetF,
 172                                                  x+j_coord_offsetG,x+j_coord_offsetH,
 173                                                  &jx0,&jy0,&jz0);
 174
 175             /* Calculate displacement vector */
 176             dx00             = _mm256_sub_ps(ix0,jx0);
 177             dy00             = _mm256_sub_ps(iy0,jy0);
 178             dz00             = _mm256_sub_ps(iz0,jz0);
 179
 180             /* Calculate squared distance and things based on it */
 181             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
 182
 183             rinv00           = gmx_mm256_invsqrt_ps(rsq00);
 184
 185             rinvsq00         = _mm256_mul_ps(rinv00,rinv00);
 186
 187             /* Load parameters for j particles */
 188             jq0              = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 189                                                                  charge+jnrC+0,charge+jnrD+0,
 190                                                                  charge+jnrE+0,charge+jnrF+0,
 191                                                                  charge+jnrG+0,charge+jnrH+0);
 192
 193             /**************************
 194              * CALCULATE INTERACTIONS *
 195              **************************/
 196
 197             /* Compute parameters for interactions between i and j atoms */
 198             qq00             = _mm256_mul_ps(iq0,jq0);
 199
 200             /* REACTION-FIELD ELECTROSTATICS */
 201             velec            = _mm256_mul_ps(qq00,_mm256_sub_ps(_mm256_add_ps(rinv00,_mm256_mul_ps(krf,rsq00)),crf));
 202             felec            = _mm256_mul_ps(qq00,_mm256_sub_ps(_mm256_mul_ps(rinv00,rinvsq00),krf2));
 203
 204             /* Update potential sum for this i atom from the interaction with this j atom. */
 205             velecsum         = _mm256_add_ps(velecsum,velec);
 206
 207             fscal            = felec;
 208
 209             /* Calculate temporary vectorial force */
 210             tx               = _mm256_mul_ps(fscal,dx00);
 211             ty               = _mm256_mul_ps(fscal,dy00);
 212             tz               = _mm256_mul_ps(fscal,dz00);
 213
 214             /* Update vectorial force */
 215             fix0             = _mm256_add_ps(fix0,tx);
 216             fiy0             = _mm256_add_ps(fiy0,ty);
 217             fiz0             = _mm256_add_ps(fiz0,tz);
 218
 219             fjptrA             = f+j_coord_offsetA;
 220             fjptrB             = f+j_coord_offsetB;
 221             fjptrC             = f+j_coord_offsetC;
 222             fjptrD             = f+j_coord_offsetD;
 223             fjptrE             = f+j_coord_offsetE;
 224             fjptrF             = f+j_coord_offsetF;
 225             fjptrG             = f+j_coord_offsetG;
 226             fjptrH             = f+j_coord_offsetH;
 227             gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,tx,ty,tz);
 228
 229             /* Inner loop uses 32 flops */
 230         }
 231
 232         if(jidx<j_index_end)
 233         {
 234
 235             /* Get j neighbor index, and coordinate index */
 236             jnrlistA         = jjnr[jidx];
 237             jnrlistB         = jjnr[jidx+1];
 238             jnrlistC         = jjnr[jidx+2];
 239             jnrlistD         = jjnr[jidx+3];
 240             jnrlistE         = jjnr[jidx+4];
 241             jnrlistF         = jjnr[jidx+5];
 242             jnrlistG         = jjnr[jidx+6];
 243             jnrlistH         = jjnr[jidx+7];
 244             /* Sign of each element will be negative for non-real atoms.
 245              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 246              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 247              */
 248             dummy_mask = gmx_mm256_set_m128(gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx+4)),_mm_setzero_si128())),
 249                                             gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128())));
 250
 251             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 252             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 253             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 254             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 255             jnrE       = (jnrlistE>=0) ? jnrlistE : 0;
 256             jnrF       = (jnrlistF>=0) ? jnrlistF : 0;
 257             jnrG       = (jnrlistG>=0) ? jnrlistG : 0;
 258             jnrH       = (jnrlistH>=0) ? jnrlistH : 0;
 259             j_coord_offsetA  = DIM*jnrA;
 260             j_coord_offsetB  = DIM*jnrB;
 261             j_coord_offsetC  = DIM*jnrC;
 262             j_coord_offsetD  = DIM*jnrD;
 263             j_coord_offsetE  = DIM*jnrE;
 264             j_coord_offsetF  = DIM*jnrF;
 265             j_coord_offsetG  = DIM*jnrG;
 266             j_coord_offsetH  = DIM*jnrH;
 267
 268             /* load j atom coordinates */
 269             gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 270                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 271                                                  x+j_coord_offsetE,x+j_coord_offsetF,
 272                                                  x+j_coord_offsetG,x+j_coord_offsetH,
 273                                                  &jx0,&jy0,&jz0);
 274
 275             /* Calculate displacement vector */
 276             dx00             = _mm256_sub_ps(ix0,jx0);
 277             dy00             = _mm256_sub_ps(iy0,jy0);
 278             dz00             = _mm256_sub_ps(iz0,jz0);
 279
 280             /* Calculate squared distance and things based on it */
 281             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
 282
 283             rinv00           = gmx_mm256_invsqrt_ps(rsq00);
 284
 285             rinvsq00         = _mm256_mul_ps(rinv00,rinv00);
 286
 287             /* Load parameters for j particles */
 288             jq0              = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 289                                                                  charge+jnrC+0,charge+jnrD+0,
 290                                                                  charge+jnrE+0,charge+jnrF+0,
 291                                                                  charge+jnrG+0,charge+jnrH+0);
 292
 293             /**************************
 294              * CALCULATE INTERACTIONS *
 295              **************************/
 296
 297             /* Compute parameters for interactions between i and j atoms */
 298             qq00             = _mm256_mul_ps(iq0,jq0);
 299
 300             /* REACTION-FIELD ELECTROSTATICS */
 301             velec            = _mm256_mul_ps(qq00,_mm256_sub_ps(_mm256_add_ps(rinv00,_mm256_mul_ps(krf,rsq00)),crf));
 302             felec            = _mm256_mul_ps(qq00,_mm256_sub_ps(_mm256_mul_ps(rinv00,rinvsq00),krf2));
 303
 304             /* Update potential sum for this i atom from the interaction with this j atom. */
 305             velec            = _mm256_andnot_ps(dummy_mask,velec);
 306             velecsum         = _mm256_add_ps(velecsum,velec);
 307
 308             fscal            = felec;
 309
 310             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
 311
 312             /* Calculate temporary vectorial force */
 313             tx               = _mm256_mul_ps(fscal,dx00);
 314             ty               = _mm256_mul_ps(fscal,dy00);
 315             tz               = _mm256_mul_ps(fscal,dz00);
 316
 317             /* Update vectorial force */
 318             fix0             = _mm256_add_ps(fix0,tx);
 319             fiy0             = _mm256_add_ps(fiy0,ty);
 320             fiz0             = _mm256_add_ps(fiz0,tz);
 321
 322             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 323             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 324             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 325             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 326             fjptrE             = (jnrlistE>=0) ? f+j_coord_offsetE : scratch;
 327             fjptrF             = (jnrlistF>=0) ? f+j_coord_offsetF : scratch;
 328             fjptrG             = (jnrlistG>=0) ? f+j_coord_offsetG : scratch;
 329             fjptrH             = (jnrlistH>=0) ? f+j_coord_offsetH : scratch;
 330             gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,tx,ty,tz);
 331
 332             /* Inner loop uses 32 flops */
 333         }
 334
 335         /* End of innermost loop */
 336
 337         gmx_mm256_update_iforce_1atom_swizzle_ps(fix0,fiy0,fiz0,
 338                                                  f+i_coord_offset,fshift+i_shift_offset);
 339
 340         ggid                        = gid[iidx];
 341         /* Update potential energies */
 342         gmx_mm256_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid);
 343
 344         /* Increment number of inner iterations */
 345         inneriter                  += j_index_end - j_index_start;
 346
 347         /* Outer loop uses 8 flops */
 348     }
 349
 350     /* Increment number of outer iterations */
 351     outeriter        += nri;
 352
 353     /* Update outer/inner flops */
 354
 355     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VF,outeriter*8 + inneriter*32);
 356 }
 357 /*
 358  * Gromacs nonbonded kernel:   nb_kernel_ElecRF_VdwNone_GeomP1P1_F_avx_256_single
 359  * Electrostatics interaction: ReactionField
 360  * VdW interaction:            None
 361  * Geometry:                   Particle-Particle
 362  * Calculate force/pot:        Force
 363  */
 364 void
 365 nb_kernel_ElecRF_VdwNone_GeomP1P1_F_avx_256_single
 366                     (t_nblist * gmx_restrict                nlist,
 367                      rvec * gmx_restrict                    xx,
 368                      rvec * gmx_restrict                    ff,
 369                      t_forcerec * gmx_restrict              fr,
 370                      t_mdatoms * gmx_restrict               mdatoms,
 371                      nb_kernel_data_t * gmx_restrict        kernel_data,
 372                      t_nrnb * gmx_restrict                  nrnb)
 373 {
 374     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 375      * just 0 for non-waters.
 376      * Suffixes A,B,C,D,E,F,G,H refer to j loop unrolling done with AVX, e.g. for the eight different
 377      * jnr indices corresponding to data put in the four positions in the SIMD register.
 378      */
 379     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 380     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 381     int              jnrA,jnrB,jnrC,jnrD;
 382     int              jnrE,jnrF,jnrG,jnrH;
 383     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
 384     int              jnrlistE,jnrlistF,jnrlistG,jnrlistH;
 385     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 386     int              j_coord_offsetE,j_coord_offsetF,j_coord_offsetG,j_coord_offsetH;
 387     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 388     real             rcutoff_scalar;
 389     real             *shiftvec,*fshift,*x,*f;
 390     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD,*fjptrE,*fjptrF,*fjptrG,*fjptrH;
 391     real             scratch[4*DIM];
 392     __m256           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 393     real *           vdwioffsetptr0;
 394     __m256           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 395     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D,vdwjidx0E,vdwjidx0F,vdwjidx0G,vdwjidx0H;
 396     __m256           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 397     __m256           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 398     __m256           velec,felec,velecsum,facel,crf,krf,krf2;
 399     real             *charge;
 400     __m256           dummy_mask,cutoff_mask;
 401     __m256           signbit = _mm256_castsi256_ps( _mm256_set1_epi32(0x80000000) );
 402     __m256           one     = _mm256_set1_ps(1.0);
 403     __m256           two     = _mm256_set1_ps(2.0);
 404     x                = xx[0];
 405     f                = ff[0];
 406
 407     nri              = nlist->nri;
 408     iinr             = nlist->iinr;
 409     jindex           = nlist->jindex;
 410     jjnr             = nlist->jjnr;
 411     shiftidx         = nlist->shift;
 412     gid              = nlist->gid;
 413     shiftvec         = fr->shift_vec[0];
 414     fshift           = fr->fshift[0];
 415     facel            = _mm256_set1_ps(fr->epsfac);
 416     charge           = mdatoms->chargeA;
 417     krf              = _mm256_set1_ps(fr->ic->k_rf);
 418     krf2             = _mm256_set1_ps(fr->ic->k_rf*2.0);
 419     crf              = _mm256_set1_ps(fr->ic->c_rf);
 420
 421     /* Avoid stupid compiler warnings */
 422     jnrA = jnrB = jnrC = jnrD = jnrE = jnrF = jnrG = jnrH = 0;
 423     j_coord_offsetA = 0;
 424     j_coord_offsetB = 0;
 425     j_coord_offsetC = 0;
 426     j_coord_offsetD = 0;
 427     j_coord_offsetE = 0;
 428     j_coord_offsetF = 0;
 429     j_coord_offsetG = 0;
 430     j_coord_offsetH = 0;
 431
 432     outeriter        = 0;
 433     inneriter        = 0;
 434
 435     for(iidx=0;iidx<4*DIM;iidx++)
 436     {
 437         scratch[iidx] = 0.0;
 438     }
 439
 440     /* Start outer loop over neighborlists */
 441     for(iidx=0; iidx<nri; iidx++)
 442     {
 443         /* Load shift vector for this list */
 444         i_shift_offset   = DIM*shiftidx[iidx];
 445
 446         /* Load limits for loop over neighbors */
 447         j_index_start    = jindex[iidx];
 448         j_index_end      = jindex[iidx+1];
 449
 450         /* Get outer coordinate index */
 451         inr              = iinr[iidx];
 452         i_coord_offset   = DIM*inr;
 453
 454         /* Load i particle coords and add shift vector */
 455         gmx_mm256_load_shift_and_1rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 456
 457         fix0             = _mm256_setzero_ps();
 458         fiy0             = _mm256_setzero_ps();
 459         fiz0             = _mm256_setzero_ps();
 460
 461         /* Load parameters for i particles */
 462         iq0              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+0]));
 463
 464         /* Start inner kernel loop */
 465         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+7]>=0; jidx+=8)
 466         {
 467
 468             /* Get j neighbor index, and coordinate index */
 469             jnrA             = jjnr[jidx];
 470             jnrB             = jjnr[jidx+1];
 471             jnrC             = jjnr[jidx+2];
 472             jnrD             = jjnr[jidx+3];
 473             jnrE             = jjnr[jidx+4];
 474             jnrF             = jjnr[jidx+5];
 475             jnrG             = jjnr[jidx+6];
 476             jnrH             = jjnr[jidx+7];
 477             j_coord_offsetA  = DIM*jnrA;
 478             j_coord_offsetB  = DIM*jnrB;
 479             j_coord_offsetC  = DIM*jnrC;
 480             j_coord_offsetD  = DIM*jnrD;
 481             j_coord_offsetE  = DIM*jnrE;
 482             j_coord_offsetF  = DIM*jnrF;
 483             j_coord_offsetG  = DIM*jnrG;
 484             j_coord_offsetH  = DIM*jnrH;
 485
 486             /* load j atom coordinates */
 487             gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 488                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 489                                                  x+j_coord_offsetE,x+j_coord_offsetF,
 490                                                  x+j_coord_offsetG,x+j_coord_offsetH,
 491                                                  &jx0,&jy0,&jz0);
 492
 493             /* Calculate displacement vector */
 494             dx00             = _mm256_sub_ps(ix0,jx0);
 495             dy00             = _mm256_sub_ps(iy0,jy0);
 496             dz00             = _mm256_sub_ps(iz0,jz0);
 497
 498             /* Calculate squared distance and things based on it */
 499             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
 500
 501             rinv00           = gmx_mm256_invsqrt_ps(rsq00);
 502
 503             rinvsq00         = _mm256_mul_ps(rinv00,rinv00);
 504
 505             /* Load parameters for j particles */
 506             jq0              = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 507                                                                  charge+jnrC+0,charge+jnrD+0,
 508                                                                  charge+jnrE+0,charge+jnrF+0,
 509                                                                  charge+jnrG+0,charge+jnrH+0);
 510
 511             /**************************
 512              * CALCULATE INTERACTIONS *
 513              **************************/
 514
 515             /* Compute parameters for interactions between i and j atoms */
 516             qq00             = _mm256_mul_ps(iq0,jq0);
 517
 518             /* REACTION-FIELD ELECTROSTATICS */
 519             felec            = _mm256_mul_ps(qq00,_mm256_sub_ps(_mm256_mul_ps(rinv00,rinvsq00),krf2));
 520
 521             fscal            = felec;
 522
 523             /* Calculate temporary vectorial force */
 524             tx               = _mm256_mul_ps(fscal,dx00);
 525             ty               = _mm256_mul_ps(fscal,dy00);
 526             tz               = _mm256_mul_ps(fscal,dz00);
 527
 528             /* Update vectorial force */
 529             fix0             = _mm256_add_ps(fix0,tx);
 530             fiy0             = _mm256_add_ps(fiy0,ty);
 531             fiz0             = _mm256_add_ps(fiz0,tz);
 532
 533             fjptrA             = f+j_coord_offsetA;
 534             fjptrB             = f+j_coord_offsetB;
 535             fjptrC             = f+j_coord_offsetC;
 536             fjptrD             = f+j_coord_offsetD;
 537             fjptrE             = f+j_coord_offsetE;
 538             fjptrF             = f+j_coord_offsetF;
 539             fjptrG             = f+j_coord_offsetG;
 540             fjptrH             = f+j_coord_offsetH;
 541             gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,tx,ty,tz);
 542
 543             /* Inner loop uses 27 flops */
 544         }
 545
 546         if(jidx<j_index_end)
 547         {
 548
 549             /* Get j neighbor index, and coordinate index */
 550             jnrlistA         = jjnr[jidx];
 551             jnrlistB         = jjnr[jidx+1];
 552             jnrlistC         = jjnr[jidx+2];
 553             jnrlistD         = jjnr[jidx+3];
 554             jnrlistE         = jjnr[jidx+4];
 555             jnrlistF         = jjnr[jidx+5];
 556             jnrlistG         = jjnr[jidx+6];
 557             jnrlistH         = jjnr[jidx+7];
 558             /* Sign of each element will be negative for non-real atoms.
 559              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 560              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 561              */
 562             dummy_mask = gmx_mm256_set_m128(gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx+4)),_mm_setzero_si128())),
 563                                             gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128())));
 564
 565             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 566             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 567             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 568             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 569             jnrE       = (jnrlistE>=0) ? jnrlistE : 0;
 570             jnrF       = (jnrlistF>=0) ? jnrlistF : 0;
 571             jnrG       = (jnrlistG>=0) ? jnrlistG : 0;
 572             jnrH       = (jnrlistH>=0) ? jnrlistH : 0;
 573             j_coord_offsetA  = DIM*jnrA;
 574             j_coord_offsetB  = DIM*jnrB;
 575             j_coord_offsetC  = DIM*jnrC;
 576             j_coord_offsetD  = DIM*jnrD;
 577             j_coord_offsetE  = DIM*jnrE;
 578             j_coord_offsetF  = DIM*jnrF;
 579             j_coord_offsetG  = DIM*jnrG;
 580             j_coord_offsetH  = DIM*jnrH;
 581
 582             /* load j atom coordinates */
 583             gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 584                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 585                                                  x+j_coord_offsetE,x+j_coord_offsetF,
 586                                                  x+j_coord_offsetG,x+j_coord_offsetH,
 587                                                  &jx0,&jy0,&jz0);
 588
 589             /* Calculate displacement vector */
 590             dx00             = _mm256_sub_ps(ix0,jx0);
 591             dy00             = _mm256_sub_ps(iy0,jy0);
 592             dz00             = _mm256_sub_ps(iz0,jz0);
 593
 594             /* Calculate squared distance and things based on it */
 595             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
 596
 597             rinv00           = gmx_mm256_invsqrt_ps(rsq00);
 598
 599             rinvsq00         = _mm256_mul_ps(rinv00,rinv00);
 600
 601             /* Load parameters for j particles */
 602             jq0              = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 603                                                                  charge+jnrC+0,charge+jnrD+0,
 604                                                                  charge+jnrE+0,charge+jnrF+0,
 605                                                                  charge+jnrG+0,charge+jnrH+0);
 606
 607             /**************************
 608              * CALCULATE INTERACTIONS *
 609              **************************/
 610
 611             /* Compute parameters for interactions between i and j atoms */
 612             qq00             = _mm256_mul_ps(iq0,jq0);
 613
 614             /* REACTION-FIELD ELECTROSTATICS */
 615             felec            = _mm256_mul_ps(qq00,_mm256_sub_ps(_mm256_mul_ps(rinv00,rinvsq00),krf2));
 616
 617             fscal            = felec;
 618
 619             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
 620
 621             /* Calculate temporary vectorial force */
 622             tx               = _mm256_mul_ps(fscal,dx00);
 623             ty               = _mm256_mul_ps(fscal,dy00);
 624             tz               = _mm256_mul_ps(fscal,dz00);
 625
 626             /* Update vectorial force */
 627             fix0             = _mm256_add_ps(fix0,tx);
 628             fiy0             = _mm256_add_ps(fiy0,ty);
 629             fiz0             = _mm256_add_ps(fiz0,tz);
 630
 631             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 632             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 633             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 634             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 635             fjptrE             = (jnrlistE>=0) ? f+j_coord_offsetE : scratch;
 636             fjptrF             = (jnrlistF>=0) ? f+j_coord_offsetF : scratch;
 637             fjptrG             = (jnrlistG>=0) ? f+j_coord_offsetG : scratch;
 638             fjptrH             = (jnrlistH>=0) ? f+j_coord_offsetH : scratch;
 639             gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,tx,ty,tz);
 640
 641             /* Inner loop uses 27 flops */
 642         }
 643
 644         /* End of innermost loop */
 645
 646         gmx_mm256_update_iforce_1atom_swizzle_ps(fix0,fiy0,fiz0,
 647                                                  f+i_coord_offset,fshift+i_shift_offset);
 648
 649         /* Increment number of inner iterations */
 650         inneriter                  += j_index_end - j_index_start;
 651
 652         /* Outer loop uses 7 flops */
 653     }
 654
 655     /* Increment number of outer iterations */
 656     outeriter        += nri;
 657
 658     /* Update outer/inner flops */
 659
 660     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_F,outeriter*7 + inneriter*27);
 661 }