src/gromacs/gmxlib/nonbonded/nb_kernel_avx_256_single/nb_kernel_ElecRFCut_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_ElecRFCut_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_ElecRFCut_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     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 101     rcutoff_scalar   = fr->rcoulomb;
 102     rcutoff          = _mm256_set1_ps(rcutoff_scalar);
 103     rcutoff2         = _mm256_mul_ps(rcutoff,rcutoff);
 104
 105     /* Avoid stupid compiler warnings */
 106     jnrA = jnrB = jnrC = jnrD = jnrE = jnrF = jnrG = jnrH = 0;
 107     j_coord_offsetA = 0;
 108     j_coord_offsetB = 0;
 109     j_coord_offsetC = 0;
 110     j_coord_offsetD = 0;
 111     j_coord_offsetE = 0;
 112     j_coord_offsetF = 0;
 113     j_coord_offsetG = 0;
 114     j_coord_offsetH = 0;
 115
 116     outeriter        = 0;
 117     inneriter        = 0;
 118
 119     for(iidx=0;iidx<4*DIM;iidx++)
 120     {
 121         scratch[iidx] = 0.0;
 122     }
 123
 124     /* Start outer loop over neighborlists */
 125     for(iidx=0; iidx<nri; iidx++)
 126     {
 127         /* Load shift vector for this list */
 128         i_shift_offset   = DIM*shiftidx[iidx];
 129
 130         /* Load limits for loop over neighbors */
 131         j_index_start    = jindex[iidx];
 132         j_index_end      = jindex[iidx+1];
 133
 134         /* Get outer coordinate index */
 135         inr              = iinr[iidx];
 136         i_coord_offset   = DIM*inr;
 137
 138         /* Load i particle coords and add shift vector */
 139         gmx_mm256_load_shift_and_1rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 140
 141         fix0             = _mm256_setzero_ps();
 142         fiy0             = _mm256_setzero_ps();
 143         fiz0             = _mm256_setzero_ps();
 144
 145         /* Load parameters for i particles */
 146         iq0              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+0]));
 147
 148         /* Reset potential sums */
 149         velecsum         = _mm256_setzero_ps();
 150
 151         /* Start inner kernel loop */
 152         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+7]>=0; jidx+=8)
 153         {
 154
 155             /* Get j neighbor index, and coordinate index */
 156             jnrA             = jjnr[jidx];
 157             jnrB             = jjnr[jidx+1];
 158             jnrC             = jjnr[jidx+2];
 159             jnrD             = jjnr[jidx+3];
 160             jnrE             = jjnr[jidx+4];
 161             jnrF             = jjnr[jidx+5];
 162             jnrG             = jjnr[jidx+6];
 163             jnrH             = jjnr[jidx+7];
 164             j_coord_offsetA  = DIM*jnrA;
 165             j_coord_offsetB  = DIM*jnrB;
 166             j_coord_offsetC  = DIM*jnrC;
 167             j_coord_offsetD  = DIM*jnrD;
 168             j_coord_offsetE  = DIM*jnrE;
 169             j_coord_offsetF  = DIM*jnrF;
 170             j_coord_offsetG  = DIM*jnrG;
 171             j_coord_offsetH  = DIM*jnrH;
 172
 173             /* load j atom coordinates */
 174             gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 175                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 176                                                  x+j_coord_offsetE,x+j_coord_offsetF,
 177                                                  x+j_coord_offsetG,x+j_coord_offsetH,
 178                                                  &jx0,&jy0,&jz0);
 179
 180             /* Calculate displacement vector */
 181             dx00             = _mm256_sub_ps(ix0,jx0);
 182             dy00             = _mm256_sub_ps(iy0,jy0);
 183             dz00             = _mm256_sub_ps(iz0,jz0);
 184
 185             /* Calculate squared distance and things based on it */
 186             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
 187
 188             rinv00           = gmx_mm256_invsqrt_ps(rsq00);
 189
 190             rinvsq00         = _mm256_mul_ps(rinv00,rinv00);
 191
 192             /* Load parameters for j particles */
 193             jq0              = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 194                                                                  charge+jnrC+0,charge+jnrD+0,
 195                                                                  charge+jnrE+0,charge+jnrF+0,
 196                                                                  charge+jnrG+0,charge+jnrH+0);
 197
 198             /**************************
 199              * CALCULATE INTERACTIONS *
 200              **************************/
 201
 202             if (gmx_mm256_any_lt(rsq00,rcutoff2))
 203             {
 204
 205             /* Compute parameters for interactions between i and j atoms */
 206             qq00             = _mm256_mul_ps(iq0,jq0);
 207
 208             /* REACTION-FIELD ELECTROSTATICS */
 209             velec            = _mm256_mul_ps(qq00,_mm256_sub_ps(_mm256_add_ps(rinv00,_mm256_mul_ps(krf,rsq00)),crf));
 210             felec            = _mm256_mul_ps(qq00,_mm256_sub_ps(_mm256_mul_ps(rinv00,rinvsq00),krf2));
 211
 212             cutoff_mask      = _mm256_cmp_ps(rsq00,rcutoff2,_CMP_LT_OQ);
 213
 214             /* Update potential sum for this i atom from the interaction with this j atom. */
 215             velec            = _mm256_and_ps(velec,cutoff_mask);
 216             velecsum         = _mm256_add_ps(velecsum,velec);
 217
 218             fscal            = felec;
 219
 220             fscal            = _mm256_and_ps(fscal,cutoff_mask);
 221
 222             /* Calculate temporary vectorial force */
 223             tx               = _mm256_mul_ps(fscal,dx00);
 224             ty               = _mm256_mul_ps(fscal,dy00);
 225             tz               = _mm256_mul_ps(fscal,dz00);
 226
 227             /* Update vectorial force */
 228             fix0             = _mm256_add_ps(fix0,tx);
 229             fiy0             = _mm256_add_ps(fiy0,ty);
 230             fiz0             = _mm256_add_ps(fiz0,tz);
 231
 232             fjptrA             = f+j_coord_offsetA;
 233             fjptrB             = f+j_coord_offsetB;
 234             fjptrC             = f+j_coord_offsetC;
 235             fjptrD             = f+j_coord_offsetD;
 236             fjptrE             = f+j_coord_offsetE;
 237             fjptrF             = f+j_coord_offsetF;
 238             fjptrG             = f+j_coord_offsetG;
 239             fjptrH             = f+j_coord_offsetH;
 240             gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,tx,ty,tz);
 241
 242             }
 243
 244             /* Inner loop uses 36 flops */
 245         }
 246
 247         if(jidx<j_index_end)
 248         {
 249
 250             /* Get j neighbor index, and coordinate index */
 251             jnrlistA         = jjnr[jidx];
 252             jnrlistB         = jjnr[jidx+1];
 253             jnrlistC         = jjnr[jidx+2];
 254             jnrlistD         = jjnr[jidx+3];
 255             jnrlistE         = jjnr[jidx+4];
 256             jnrlistF         = jjnr[jidx+5];
 257             jnrlistG         = jjnr[jidx+6];
 258             jnrlistH         = jjnr[jidx+7];
 259             /* Sign of each element will be negative for non-real atoms.
 260              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 261              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 262              */
 263             dummy_mask = gmx_mm256_set_m128(gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx+4)),_mm_setzero_si128())),
 264                                             gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128())));
 265
 266             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 267             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 268             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 269             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 270             jnrE       = (jnrlistE>=0) ? jnrlistE : 0;
 271             jnrF       = (jnrlistF>=0) ? jnrlistF : 0;
 272             jnrG       = (jnrlistG>=0) ? jnrlistG : 0;
 273             jnrH       = (jnrlistH>=0) ? jnrlistH : 0;
 274             j_coord_offsetA  = DIM*jnrA;
 275             j_coord_offsetB  = DIM*jnrB;
 276             j_coord_offsetC  = DIM*jnrC;
 277             j_coord_offsetD  = DIM*jnrD;
 278             j_coord_offsetE  = DIM*jnrE;
 279             j_coord_offsetF  = DIM*jnrF;
 280             j_coord_offsetG  = DIM*jnrG;
 281             j_coord_offsetH  = DIM*jnrH;
 282
 283             /* load j atom coordinates */
 284             gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 285                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 286                                                  x+j_coord_offsetE,x+j_coord_offsetF,
 287                                                  x+j_coord_offsetG,x+j_coord_offsetH,
 288                                                  &jx0,&jy0,&jz0);
 289
 290             /* Calculate displacement vector */
 291             dx00             = _mm256_sub_ps(ix0,jx0);
 292             dy00             = _mm256_sub_ps(iy0,jy0);
 293             dz00             = _mm256_sub_ps(iz0,jz0);
 294
 295             /* Calculate squared distance and things based on it */
 296             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
 297
 298             rinv00           = gmx_mm256_invsqrt_ps(rsq00);
 299
 300             rinvsq00         = _mm256_mul_ps(rinv00,rinv00);
 301
 302             /* Load parameters for j particles */
 303             jq0              = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 304                                                                  charge+jnrC+0,charge+jnrD+0,
 305                                                                  charge+jnrE+0,charge+jnrF+0,
 306                                                                  charge+jnrG+0,charge+jnrH+0);
 307
 308             /**************************
 309              * CALCULATE INTERACTIONS *
 310              **************************/
 311
 312             if (gmx_mm256_any_lt(rsq00,rcutoff2))
 313             {
 314
 315             /* Compute parameters for interactions between i and j atoms */
 316             qq00             = _mm256_mul_ps(iq0,jq0);
 317
 318             /* REACTION-FIELD ELECTROSTATICS */
 319             velec            = _mm256_mul_ps(qq00,_mm256_sub_ps(_mm256_add_ps(rinv00,_mm256_mul_ps(krf,rsq00)),crf));
 320             felec            = _mm256_mul_ps(qq00,_mm256_sub_ps(_mm256_mul_ps(rinv00,rinvsq00),krf2));
 321
 322             cutoff_mask      = _mm256_cmp_ps(rsq00,rcutoff2,_CMP_LT_OQ);
 323
 324             /* Update potential sum for this i atom from the interaction with this j atom. */
 325             velec            = _mm256_and_ps(velec,cutoff_mask);
 326             velec            = _mm256_andnot_ps(dummy_mask,velec);
 327             velecsum         = _mm256_add_ps(velecsum,velec);
 328
 329             fscal            = felec;
 330
 331             fscal            = _mm256_and_ps(fscal,cutoff_mask);
 332
 333             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
 334
 335             /* Calculate temporary vectorial force */
 336             tx               = _mm256_mul_ps(fscal,dx00);
 337             ty               = _mm256_mul_ps(fscal,dy00);
 338             tz               = _mm256_mul_ps(fscal,dz00);
 339
 340             /* Update vectorial force */
 341             fix0             = _mm256_add_ps(fix0,tx);
 342             fiy0             = _mm256_add_ps(fiy0,ty);
 343             fiz0             = _mm256_add_ps(fiz0,tz);
 344
 345             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 346             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 347             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 348             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 349             fjptrE             = (jnrlistE>=0) ? f+j_coord_offsetE : scratch;
 350             fjptrF             = (jnrlistF>=0) ? f+j_coord_offsetF : scratch;
 351             fjptrG             = (jnrlistG>=0) ? f+j_coord_offsetG : scratch;
 352             fjptrH             = (jnrlistH>=0) ? f+j_coord_offsetH : scratch;
 353             gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,tx,ty,tz);
 354
 355             }
 356
 357             /* Inner loop uses 36 flops */
 358         }
 359
 360         /* End of innermost loop */
 361
 362         gmx_mm256_update_iforce_1atom_swizzle_ps(fix0,fiy0,fiz0,
 363                                                  f+i_coord_offset,fshift+i_shift_offset);
 364
 365         ggid                        = gid[iidx];
 366         /* Update potential energies */
 367         gmx_mm256_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid);
 368
 369         /* Increment number of inner iterations */
 370         inneriter                  += j_index_end - j_index_start;
 371
 372         /* Outer loop uses 8 flops */
 373     }
 374
 375     /* Increment number of outer iterations */
 376     outeriter        += nri;
 377
 378     /* Update outer/inner flops */
 379
 380     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VF,outeriter*8 + inneriter*36);
 381 }
 382 /*
 383  * Gromacs nonbonded kernel:   nb_kernel_ElecRFCut_VdwNone_GeomP1P1_F_avx_256_single
 384  * Electrostatics interaction: ReactionField
 385  * VdW interaction:            None
 386  * Geometry:                   Particle-Particle
 387  * Calculate force/pot:        Force
 388  */
 389 void
 390 nb_kernel_ElecRFCut_VdwNone_GeomP1P1_F_avx_256_single
 391                     (t_nblist * gmx_restrict                nlist,
 392                      rvec * gmx_restrict                    xx,
 393                      rvec * gmx_restrict                    ff,
 394                      t_forcerec * gmx_restrict              fr,
 395                      t_mdatoms * gmx_restrict               mdatoms,
 396                      nb_kernel_data_t * gmx_restrict        kernel_data,
 397                      t_nrnb * gmx_restrict                  nrnb)
 398 {
 399     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 400      * just 0 for non-waters.
 401      * Suffixes A,B,C,D,E,F,G,H refer to j loop unrolling done with AVX, e.g. for the eight different
 402      * jnr indices corresponding to data put in the four positions in the SIMD register.
 403      */
 404     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 405     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 406     int              jnrA,jnrB,jnrC,jnrD;
 407     int              jnrE,jnrF,jnrG,jnrH;
 408     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
 409     int              jnrlistE,jnrlistF,jnrlistG,jnrlistH;
 410     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 411     int              j_coord_offsetE,j_coord_offsetF,j_coord_offsetG,j_coord_offsetH;
 412     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 413     real             rcutoff_scalar;
 414     real             *shiftvec,*fshift,*x,*f;
 415     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD,*fjptrE,*fjptrF,*fjptrG,*fjptrH;
 416     real             scratch[4*DIM];
 417     __m256           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 418     real *           vdwioffsetptr0;
 419     __m256           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 420     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D,vdwjidx0E,vdwjidx0F,vdwjidx0G,vdwjidx0H;
 421     __m256           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 422     __m256           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 423     __m256           velec,felec,velecsum,facel,crf,krf,krf2;
 424     real             *charge;
 425     __m256           dummy_mask,cutoff_mask;
 426     __m256           signbit = _mm256_castsi256_ps( _mm256_set1_epi32(0x80000000) );
 427     __m256           one     = _mm256_set1_ps(1.0);
 428     __m256           two     = _mm256_set1_ps(2.0);
 429     x                = xx[0];
 430     f                = ff[0];
 431
 432     nri              = nlist->nri;
 433     iinr             = nlist->iinr;
 434     jindex           = nlist->jindex;
 435     jjnr             = nlist->jjnr;
 436     shiftidx         = nlist->shift;
 437     gid              = nlist->gid;
 438     shiftvec         = fr->shift_vec[0];
 439     fshift           = fr->fshift[0];
 440     facel            = _mm256_set1_ps(fr->epsfac);
 441     charge           = mdatoms->chargeA;
 442     krf              = _mm256_set1_ps(fr->ic->k_rf);
 443     krf2             = _mm256_set1_ps(fr->ic->k_rf*2.0);
 444     crf              = _mm256_set1_ps(fr->ic->c_rf);
 445
 446     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 447     rcutoff_scalar   = fr->rcoulomb;
 448     rcutoff          = _mm256_set1_ps(rcutoff_scalar);
 449     rcutoff2         = _mm256_mul_ps(rcutoff,rcutoff);
 450
 451     /* Avoid stupid compiler warnings */
 452     jnrA = jnrB = jnrC = jnrD = jnrE = jnrF = jnrG = jnrH = 0;
 453     j_coord_offsetA = 0;
 454     j_coord_offsetB = 0;
 455     j_coord_offsetC = 0;
 456     j_coord_offsetD = 0;
 457     j_coord_offsetE = 0;
 458     j_coord_offsetF = 0;
 459     j_coord_offsetG = 0;
 460     j_coord_offsetH = 0;
 461
 462     outeriter        = 0;
 463     inneriter        = 0;
 464
 465     for(iidx=0;iidx<4*DIM;iidx++)
 466     {
 467         scratch[iidx] = 0.0;
 468     }
 469
 470     /* Start outer loop over neighborlists */
 471     for(iidx=0; iidx<nri; iidx++)
 472     {
 473         /* Load shift vector for this list */
 474         i_shift_offset   = DIM*shiftidx[iidx];
 475
 476         /* Load limits for loop over neighbors */
 477         j_index_start    = jindex[iidx];
 478         j_index_end      = jindex[iidx+1];
 479
 480         /* Get outer coordinate index */
 481         inr              = iinr[iidx];
 482         i_coord_offset   = DIM*inr;
 483
 484         /* Load i particle coords and add shift vector */
 485         gmx_mm256_load_shift_and_1rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 486
 487         fix0             = _mm256_setzero_ps();
 488         fiy0             = _mm256_setzero_ps();
 489         fiz0             = _mm256_setzero_ps();
 490
 491         /* Load parameters for i particles */
 492         iq0              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+0]));
 493
 494         /* Start inner kernel loop */
 495         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+7]>=0; jidx+=8)
 496         {
 497
 498             /* Get j neighbor index, and coordinate index */
 499             jnrA             = jjnr[jidx];
 500             jnrB             = jjnr[jidx+1];
 501             jnrC             = jjnr[jidx+2];
 502             jnrD             = jjnr[jidx+3];
 503             jnrE             = jjnr[jidx+4];
 504             jnrF             = jjnr[jidx+5];
 505             jnrG             = jjnr[jidx+6];
 506             jnrH             = jjnr[jidx+7];
 507             j_coord_offsetA  = DIM*jnrA;
 508             j_coord_offsetB  = DIM*jnrB;
 509             j_coord_offsetC  = DIM*jnrC;
 510             j_coord_offsetD  = DIM*jnrD;
 511             j_coord_offsetE  = DIM*jnrE;
 512             j_coord_offsetF  = DIM*jnrF;
 513             j_coord_offsetG  = DIM*jnrG;
 514             j_coord_offsetH  = DIM*jnrH;
 515
 516             /* load j atom coordinates */
 517             gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 518                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 519                                                  x+j_coord_offsetE,x+j_coord_offsetF,
 520                                                  x+j_coord_offsetG,x+j_coord_offsetH,
 521                                                  &jx0,&jy0,&jz0);
 522
 523             /* Calculate displacement vector */
 524             dx00             = _mm256_sub_ps(ix0,jx0);
 525             dy00             = _mm256_sub_ps(iy0,jy0);
 526             dz00             = _mm256_sub_ps(iz0,jz0);
 527
 528             /* Calculate squared distance and things based on it */
 529             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
 530
 531             rinv00           = gmx_mm256_invsqrt_ps(rsq00);
 532
 533             rinvsq00         = _mm256_mul_ps(rinv00,rinv00);
 534
 535             /* Load parameters for j particles */
 536             jq0              = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 537                                                                  charge+jnrC+0,charge+jnrD+0,
 538                                                                  charge+jnrE+0,charge+jnrF+0,
 539                                                                  charge+jnrG+0,charge+jnrH+0);
 540
 541             /**************************
 542              * CALCULATE INTERACTIONS *
 543              **************************/
 544
 545             if (gmx_mm256_any_lt(rsq00,rcutoff2))
 546             {
 547
 548             /* Compute parameters for interactions between i and j atoms */
 549             qq00             = _mm256_mul_ps(iq0,jq0);
 550
 551             /* REACTION-FIELD ELECTROSTATICS */
 552             felec            = _mm256_mul_ps(qq00,_mm256_sub_ps(_mm256_mul_ps(rinv00,rinvsq00),krf2));
 553
 554             cutoff_mask      = _mm256_cmp_ps(rsq00,rcutoff2,_CMP_LT_OQ);
 555
 556             fscal            = felec;
 557
 558             fscal            = _mm256_and_ps(fscal,cutoff_mask);
 559
 560             /* Calculate temporary vectorial force */
 561             tx               = _mm256_mul_ps(fscal,dx00);
 562             ty               = _mm256_mul_ps(fscal,dy00);
 563             tz               = _mm256_mul_ps(fscal,dz00);
 564
 565             /* Update vectorial force */
 566             fix0             = _mm256_add_ps(fix0,tx);
 567             fiy0             = _mm256_add_ps(fiy0,ty);
 568             fiz0             = _mm256_add_ps(fiz0,tz);
 569
 570             fjptrA             = f+j_coord_offsetA;
 571             fjptrB             = f+j_coord_offsetB;
 572             fjptrC             = f+j_coord_offsetC;
 573             fjptrD             = f+j_coord_offsetD;
 574             fjptrE             = f+j_coord_offsetE;
 575             fjptrF             = f+j_coord_offsetF;
 576             fjptrG             = f+j_coord_offsetG;
 577             fjptrH             = f+j_coord_offsetH;
 578             gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,tx,ty,tz);
 579
 580             }
 581
 582             /* Inner loop uses 30 flops */
 583         }
 584
 585         if(jidx<j_index_end)
 586         {
 587
 588             /* Get j neighbor index, and coordinate index */
 589             jnrlistA         = jjnr[jidx];
 590             jnrlistB         = jjnr[jidx+1];
 591             jnrlistC         = jjnr[jidx+2];
 592             jnrlistD         = jjnr[jidx+3];
 593             jnrlistE         = jjnr[jidx+4];
 594             jnrlistF         = jjnr[jidx+5];
 595             jnrlistG         = jjnr[jidx+6];
 596             jnrlistH         = jjnr[jidx+7];
 597             /* Sign of each element will be negative for non-real atoms.
 598              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 599              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 600              */
 601             dummy_mask = gmx_mm256_set_m128(gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx+4)),_mm_setzero_si128())),
 602                                             gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128())));
 603
 604             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 605             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 606             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 607             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 608             jnrE       = (jnrlistE>=0) ? jnrlistE : 0;
 609             jnrF       = (jnrlistF>=0) ? jnrlistF : 0;
 610             jnrG       = (jnrlistG>=0) ? jnrlistG : 0;
 611             jnrH       = (jnrlistH>=0) ? jnrlistH : 0;
 612             j_coord_offsetA  = DIM*jnrA;
 613             j_coord_offsetB  = DIM*jnrB;
 614             j_coord_offsetC  = DIM*jnrC;
 615             j_coord_offsetD  = DIM*jnrD;
 616             j_coord_offsetE  = DIM*jnrE;
 617             j_coord_offsetF  = DIM*jnrF;
 618             j_coord_offsetG  = DIM*jnrG;
 619             j_coord_offsetH  = DIM*jnrH;
 620
 621             /* load j atom coordinates */
 622             gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 623                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 624                                                  x+j_coord_offsetE,x+j_coord_offsetF,
 625                                                  x+j_coord_offsetG,x+j_coord_offsetH,
 626                                                  &jx0,&jy0,&jz0);
 627
 628             /* Calculate displacement vector */
 629             dx00             = _mm256_sub_ps(ix0,jx0);
 630             dy00             = _mm256_sub_ps(iy0,jy0);
 631             dz00             = _mm256_sub_ps(iz0,jz0);
 632
 633             /* Calculate squared distance and things based on it */
 634             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
 635
 636             rinv00           = gmx_mm256_invsqrt_ps(rsq00);
 637
 638             rinvsq00         = _mm256_mul_ps(rinv00,rinv00);
 639
 640             /* Load parameters for j particles */
 641             jq0              = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 642                                                                  charge+jnrC+0,charge+jnrD+0,
 643                                                                  charge+jnrE+0,charge+jnrF+0,
 644                                                                  charge+jnrG+0,charge+jnrH+0);
 645
 646             /**************************
 647              * CALCULATE INTERACTIONS *
 648              **************************/
 649
 650             if (gmx_mm256_any_lt(rsq00,rcutoff2))
 651             {
 652
 653             /* Compute parameters for interactions between i and j atoms */
 654             qq00             = _mm256_mul_ps(iq0,jq0);
 655
 656             /* REACTION-FIELD ELECTROSTATICS */
 657             felec            = _mm256_mul_ps(qq00,_mm256_sub_ps(_mm256_mul_ps(rinv00,rinvsq00),krf2));
 658
 659             cutoff_mask      = _mm256_cmp_ps(rsq00,rcutoff2,_CMP_LT_OQ);
 660
 661             fscal            = felec;
 662
 663             fscal            = _mm256_and_ps(fscal,cutoff_mask);
 664
 665             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
 666
 667             /* Calculate temporary vectorial force */
 668             tx               = _mm256_mul_ps(fscal,dx00);
 669             ty               = _mm256_mul_ps(fscal,dy00);
 670             tz               = _mm256_mul_ps(fscal,dz00);
 671
 672             /* Update vectorial force */
 673             fix0             = _mm256_add_ps(fix0,tx);
 674             fiy0             = _mm256_add_ps(fiy0,ty);
 675             fiz0             = _mm256_add_ps(fiz0,tz);
 676
 677             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 678             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 679             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 680             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 681             fjptrE             = (jnrlistE>=0) ? f+j_coord_offsetE : scratch;
 682             fjptrF             = (jnrlistF>=0) ? f+j_coord_offsetF : scratch;
 683             fjptrG             = (jnrlistG>=0) ? f+j_coord_offsetG : scratch;
 684             fjptrH             = (jnrlistH>=0) ? f+j_coord_offsetH : scratch;
 685             gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,tx,ty,tz);
 686
 687             }
 688
 689             /* Inner loop uses 30 flops */
 690         }
 691
 692         /* End of innermost loop */
 693
 694         gmx_mm256_update_iforce_1atom_swizzle_ps(fix0,fiy0,fiz0,
 695                                                  f+i_coord_offset,fshift+i_shift_offset);
 696
 697         /* Increment number of inner iterations */
 698         inneriter                  += j_index_end - j_index_start;
 699
 700         /* Outer loop uses 7 flops */
 701     }
 702
 703     /* Increment number of outer iterations */
 704     outeriter        += nri;
 705
 706     /* Update outer/inner flops */
 707
 708     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_F,outeriter*7 + inneriter*30);
 709 }