src/gromacs/gmxlib/nonbonded/nb_kernel_avx_256_single/nb_kernel_ElecCoul_VdwNone_GeomP1P1_avx_256_single.c

   1 /*
   2  * This file is part of the GROMACS molecular simulation package.
   3  *
   4  * Copyright (c) 2012,2013, by the GROMACS development team, led by
   5  * Mark Abraham, David van der Spoel, Berk Hess, and Erik Lindahl,
   6  * and including many others, as listed in the AUTHORS file in the
   7  * top-level source directory and at http://www.gromacs.org.
   8  *
   9  * GROMACS is free software; you can redistribute it and/or
  10  * modify it under the terms of the GNU Lesser General Public License
  11  * as published by the Free Software Foundation; either version 2.1
  12  * of the License, or (at your option) any later version.
  13  *
  14  * GROMACS is distributed in the hope that it will be useful,
  15  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  16  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  17  * Lesser General Public License for more details.
  18  *
  19  * You should have received a copy of the GNU Lesser General Public
  20  * License along with GROMACS; if not, see
  21  * http://www.gnu.org/licenses, or write to the Free Software Foundation,
  22  * Inc., 51 Franklin Street, Fifth Floor, Boston, MA  02110-1301  USA.
  23  *
  24  * If you want to redistribute modifications to GROMACS, please
  25  * consider that scientific software is very special. Version
  26  * control is crucial - bugs must be traceable. We will be happy to
  27  * consider code for inclusion in the official distribution, but
  28  * derived work must not be called official GROMACS. Details are found
  29  * in the README & COPYING files - if they are missing, get the
  30  * official version at http://www.gromacs.org.
  31  *
  32  * To help us fund GROMACS development, we humbly ask that you cite
  33  * the research papers on the package. Check out http://www.gromacs.org.
  34  */
  35 /*
  36  * Note: this file was generated by the GROMACS avx_256_single kernel generator.
  37  */
  38 #ifdef HAVE_CONFIG_H
  39 #include <config.h>
  40 #endif
  41
  42 #include <math.h>
  43
  44 #include "../nb_kernel.h"
  45 #include "types/simple.h"
  46 #include "vec.h"
  47 #include "nrnb.h"
  48
  49 #include "gromacs/simd/math_x86_avx_256_single.h"
  50 #include "kernelutil_x86_avx_256_single.h"
  51
  52 /*
  53  * Gromacs nonbonded kernel:   nb_kernel_ElecCoul_VdwNone_GeomP1P1_VF_avx_256_single
  54  * Electrostatics interaction: Coulomb
  55  * VdW interaction:            None
  56  * Geometry:                   Particle-Particle
  57  * Calculate force/pot:        PotentialAndForce
  58  */
  59 void
  60 nb_kernel_ElecCoul_VdwNone_GeomP1P1_VF_avx_256_single
  61                     (t_nblist                    * gmx_restrict       nlist,
  62                      rvec                        * gmx_restrict          xx,
  63                      rvec                        * gmx_restrict          ff,
  64                      t_forcerec                  * gmx_restrict          fr,
  65                      t_mdatoms                   * gmx_restrict     mdatoms,
  66                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
  67                      t_nrnb                      * gmx_restrict        nrnb)
  68 {
  69     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
  70      * just 0 for non-waters.
  71      * Suffixes A,B,C,D,E,F,G,H refer to j loop unrolling done with AVX, e.g. for the eight different
  72      * jnr indices corresponding to data put in the four positions in the SIMD register.
  73      */
  74     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
  75     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
  76     int              jnrA,jnrB,jnrC,jnrD;
  77     int              jnrE,jnrF,jnrG,jnrH;
  78     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
  79     int              jnrlistE,jnrlistF,jnrlistG,jnrlistH;
  80     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
  81     int              j_coord_offsetE,j_coord_offsetF,j_coord_offsetG,j_coord_offsetH;
  82     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
  83     real             rcutoff_scalar;
  84     real             *shiftvec,*fshift,*x,*f;
  85     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD,*fjptrE,*fjptrF,*fjptrG,*fjptrH;
  86     real             scratch[4*DIM];
  87     __m256           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
  88     real *           vdwioffsetptr0;
  89     __m256           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
  90     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D,vdwjidx0E,vdwjidx0F,vdwjidx0G,vdwjidx0H;
  91     __m256           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  92     __m256           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  93     __m256           velec,felec,velecsum,facel,crf,krf,krf2;
  94     real             *charge;
  95     __m256           dummy_mask,cutoff_mask;
  96     __m256           signbit = _mm256_castsi256_ps( _mm256_set1_epi32(0x80000000) );
  97     __m256           one     = _mm256_set1_ps(1.0);
  98     __m256           two     = _mm256_set1_ps(2.0);
  99     x                = xx[0];
 100     f                = ff[0];
 101
 102     nri              = nlist->nri;
 103     iinr             = nlist->iinr;
 104     jindex           = nlist->jindex;
 105     jjnr             = nlist->jjnr;
 106     shiftidx         = nlist->shift;
 107     gid              = nlist->gid;
 108     shiftvec         = fr->shift_vec[0];
 109     fshift           = fr->fshift[0];
 110     facel            = _mm256_set1_ps(fr->epsfac);
 111     charge           = mdatoms->chargeA;
 112
 113     /* Avoid stupid compiler warnings */
 114     jnrA = jnrB = jnrC = jnrD = jnrE = jnrF = jnrG = jnrH = 0;
 115     j_coord_offsetA = 0;
 116     j_coord_offsetB = 0;
 117     j_coord_offsetC = 0;
 118     j_coord_offsetD = 0;
 119     j_coord_offsetE = 0;
 120     j_coord_offsetF = 0;
 121     j_coord_offsetG = 0;
 122     j_coord_offsetH = 0;
 123
 124     outeriter        = 0;
 125     inneriter        = 0;
 126
 127     for(iidx=0;iidx<4*DIM;iidx++)
 128     {
 129         scratch[iidx] = 0.0;
 130     }
 131
 132     /* Start outer loop over neighborlists */
 133     for(iidx=0; iidx<nri; iidx++)
 134     {
 135         /* Load shift vector for this list */
 136         i_shift_offset   = DIM*shiftidx[iidx];
 137
 138         /* Load limits for loop over neighbors */
 139         j_index_start    = jindex[iidx];
 140         j_index_end      = jindex[iidx+1];
 141
 142         /* Get outer coordinate index */
 143         inr              = iinr[iidx];
 144         i_coord_offset   = DIM*inr;
 145
 146         /* Load i particle coords and add shift vector */
 147         gmx_mm256_load_shift_and_1rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 148
 149         fix0             = _mm256_setzero_ps();
 150         fiy0             = _mm256_setzero_ps();
 151         fiz0             = _mm256_setzero_ps();
 152
 153         /* Load parameters for i particles */
 154         iq0              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+0]));
 155
 156         /* Reset potential sums */
 157         velecsum         = _mm256_setzero_ps();
 158
 159         /* Start inner kernel loop */
 160         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+7]>=0; jidx+=8)
 161         {
 162
 163             /* Get j neighbor index, and coordinate index */
 164             jnrA             = jjnr[jidx];
 165             jnrB             = jjnr[jidx+1];
 166             jnrC             = jjnr[jidx+2];
 167             jnrD             = jjnr[jidx+3];
 168             jnrE             = jjnr[jidx+4];
 169             jnrF             = jjnr[jidx+5];
 170             jnrG             = jjnr[jidx+6];
 171             jnrH             = jjnr[jidx+7];
 172             j_coord_offsetA  = DIM*jnrA;
 173             j_coord_offsetB  = DIM*jnrB;
 174             j_coord_offsetC  = DIM*jnrC;
 175             j_coord_offsetD  = DIM*jnrD;
 176             j_coord_offsetE  = DIM*jnrE;
 177             j_coord_offsetF  = DIM*jnrF;
 178             j_coord_offsetG  = DIM*jnrG;
 179             j_coord_offsetH  = DIM*jnrH;
 180
 181             /* load j atom coordinates */
 182             gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 183                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 184                                                  x+j_coord_offsetE,x+j_coord_offsetF,
 185                                                  x+j_coord_offsetG,x+j_coord_offsetH,
 186                                                  &jx0,&jy0,&jz0);
 187
 188             /* Calculate displacement vector */
 189             dx00             = _mm256_sub_ps(ix0,jx0);
 190             dy00             = _mm256_sub_ps(iy0,jy0);
 191             dz00             = _mm256_sub_ps(iz0,jz0);
 192
 193             /* Calculate squared distance and things based on it */
 194             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
 195
 196             rinv00           = gmx_mm256_invsqrt_ps(rsq00);
 197
 198             rinvsq00         = _mm256_mul_ps(rinv00,rinv00);
 199
 200             /* Load parameters for j particles */
 201             jq0              = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 202                                                                  charge+jnrC+0,charge+jnrD+0,
 203                                                                  charge+jnrE+0,charge+jnrF+0,
 204                                                                  charge+jnrG+0,charge+jnrH+0);
 205
 206             /**************************
 207              * CALCULATE INTERACTIONS *
 208              **************************/
 209
 210             /* Compute parameters for interactions between i and j atoms */
 211             qq00             = _mm256_mul_ps(iq0,jq0);
 212
 213             /* COULOMB ELECTROSTATICS */
 214             velec            = _mm256_mul_ps(qq00,rinv00);
 215             felec            = _mm256_mul_ps(velec,rinvsq00);
 216
 217             /* Update potential sum for this i atom from the interaction with this j atom. */
 218             velecsum         = _mm256_add_ps(velecsum,velec);
 219
 220             fscal            = felec;
 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             /* Inner loop uses 27 flops */
 243         }
 244
 245         if(jidx<j_index_end)
 246         {
 247
 248             /* Get j neighbor index, and coordinate index */
 249             jnrlistA         = jjnr[jidx];
 250             jnrlistB         = jjnr[jidx+1];
 251             jnrlistC         = jjnr[jidx+2];
 252             jnrlistD         = jjnr[jidx+3];
 253             jnrlistE         = jjnr[jidx+4];
 254             jnrlistF         = jjnr[jidx+5];
 255             jnrlistG         = jjnr[jidx+6];
 256             jnrlistH         = jjnr[jidx+7];
 257             /* Sign of each element will be negative for non-real atoms.
 258              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 259              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 260              */
 261             dummy_mask = gmx_mm256_set_m128(gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx+4)),_mm_setzero_si128())),
 262                                             gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128())));
 263
 264             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 265             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 266             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 267             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 268             jnrE       = (jnrlistE>=0) ? jnrlistE : 0;
 269             jnrF       = (jnrlistF>=0) ? jnrlistF : 0;
 270             jnrG       = (jnrlistG>=0) ? jnrlistG : 0;
 271             jnrH       = (jnrlistH>=0) ? jnrlistH : 0;
 272             j_coord_offsetA  = DIM*jnrA;
 273             j_coord_offsetB  = DIM*jnrB;
 274             j_coord_offsetC  = DIM*jnrC;
 275             j_coord_offsetD  = DIM*jnrD;
 276             j_coord_offsetE  = DIM*jnrE;
 277             j_coord_offsetF  = DIM*jnrF;
 278             j_coord_offsetG  = DIM*jnrG;
 279             j_coord_offsetH  = DIM*jnrH;
 280
 281             /* load j atom coordinates */
 282             gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 283                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 284                                                  x+j_coord_offsetE,x+j_coord_offsetF,
 285                                                  x+j_coord_offsetG,x+j_coord_offsetH,
 286                                                  &jx0,&jy0,&jz0);
 287
 288             /* Calculate displacement vector */
 289             dx00             = _mm256_sub_ps(ix0,jx0);
 290             dy00             = _mm256_sub_ps(iy0,jy0);
 291             dz00             = _mm256_sub_ps(iz0,jz0);
 292
 293             /* Calculate squared distance and things based on it */
 294             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
 295
 296             rinv00           = gmx_mm256_invsqrt_ps(rsq00);
 297
 298             rinvsq00         = _mm256_mul_ps(rinv00,rinv00);
 299
 300             /* Load parameters for j particles */
 301             jq0              = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 302                                                                  charge+jnrC+0,charge+jnrD+0,
 303                                                                  charge+jnrE+0,charge+jnrF+0,
 304                                                                  charge+jnrG+0,charge+jnrH+0);
 305
 306             /**************************
 307              * CALCULATE INTERACTIONS *
 308              **************************/
 309
 310             /* Compute parameters for interactions between i and j atoms */
 311             qq00             = _mm256_mul_ps(iq0,jq0);
 312
 313             /* COULOMB ELECTROSTATICS */
 314             velec            = _mm256_mul_ps(qq00,rinv00);
 315             felec            = _mm256_mul_ps(velec,rinvsq00);
 316
 317             /* Update potential sum for this i atom from the interaction with this j atom. */
 318             velec            = _mm256_andnot_ps(dummy_mask,velec);
 319             velecsum         = _mm256_add_ps(velecsum,velec);
 320
 321             fscal            = felec;
 322
 323             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
 324
 325             /* Calculate temporary vectorial force */
 326             tx               = _mm256_mul_ps(fscal,dx00);
 327             ty               = _mm256_mul_ps(fscal,dy00);
 328             tz               = _mm256_mul_ps(fscal,dz00);
 329
 330             /* Update vectorial force */
 331             fix0             = _mm256_add_ps(fix0,tx);
 332             fiy0             = _mm256_add_ps(fiy0,ty);
 333             fiz0             = _mm256_add_ps(fiz0,tz);
 334
 335             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 336             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 337             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 338             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 339             fjptrE             = (jnrlistE>=0) ? f+j_coord_offsetE : scratch;
 340             fjptrF             = (jnrlistF>=0) ? f+j_coord_offsetF : scratch;
 341             fjptrG             = (jnrlistG>=0) ? f+j_coord_offsetG : scratch;
 342             fjptrH             = (jnrlistH>=0) ? f+j_coord_offsetH : scratch;
 343             gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,tx,ty,tz);
 344
 345             /* Inner loop uses 27 flops */
 346         }
 347
 348         /* End of innermost loop */
 349
 350         gmx_mm256_update_iforce_1atom_swizzle_ps(fix0,fiy0,fiz0,
 351                                                  f+i_coord_offset,fshift+i_shift_offset);
 352
 353         ggid                        = gid[iidx];
 354         /* Update potential energies */
 355         gmx_mm256_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid);
 356
 357         /* Increment number of inner iterations */
 358         inneriter                  += j_index_end - j_index_start;
 359
 360         /* Outer loop uses 8 flops */
 361     }
 362
 363     /* Increment number of outer iterations */
 364     outeriter        += nri;
 365
 366     /* Update outer/inner flops */
 367
 368     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VF,outeriter*8 + inneriter*27);
 369 }
 370 /*
 371  * Gromacs nonbonded kernel:   nb_kernel_ElecCoul_VdwNone_GeomP1P1_F_avx_256_single
 372  * Electrostatics interaction: Coulomb
 373  * VdW interaction:            None
 374  * Geometry:                   Particle-Particle
 375  * Calculate force/pot:        Force
 376  */
 377 void
 378 nb_kernel_ElecCoul_VdwNone_GeomP1P1_F_avx_256_single
 379                     (t_nblist                    * gmx_restrict       nlist,
 380                      rvec                        * gmx_restrict          xx,
 381                      rvec                        * gmx_restrict          ff,
 382                      t_forcerec                  * gmx_restrict          fr,
 383                      t_mdatoms                   * gmx_restrict     mdatoms,
 384                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 385                      t_nrnb                      * gmx_restrict        nrnb)
 386 {
 387     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 388      * just 0 for non-waters.
 389      * Suffixes A,B,C,D,E,F,G,H refer to j loop unrolling done with AVX, e.g. for the eight different
 390      * jnr indices corresponding to data put in the four positions in the SIMD register.
 391      */
 392     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 393     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 394     int              jnrA,jnrB,jnrC,jnrD;
 395     int              jnrE,jnrF,jnrG,jnrH;
 396     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
 397     int              jnrlistE,jnrlistF,jnrlistG,jnrlistH;
 398     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 399     int              j_coord_offsetE,j_coord_offsetF,j_coord_offsetG,j_coord_offsetH;
 400     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 401     real             rcutoff_scalar;
 402     real             *shiftvec,*fshift,*x,*f;
 403     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD,*fjptrE,*fjptrF,*fjptrG,*fjptrH;
 404     real             scratch[4*DIM];
 405     __m256           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 406     real *           vdwioffsetptr0;
 407     __m256           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 408     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D,vdwjidx0E,vdwjidx0F,vdwjidx0G,vdwjidx0H;
 409     __m256           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 410     __m256           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 411     __m256           velec,felec,velecsum,facel,crf,krf,krf2;
 412     real             *charge;
 413     __m256           dummy_mask,cutoff_mask;
 414     __m256           signbit = _mm256_castsi256_ps( _mm256_set1_epi32(0x80000000) );
 415     __m256           one     = _mm256_set1_ps(1.0);
 416     __m256           two     = _mm256_set1_ps(2.0);
 417     x                = xx[0];
 418     f                = ff[0];
 419
 420     nri              = nlist->nri;
 421     iinr             = nlist->iinr;
 422     jindex           = nlist->jindex;
 423     jjnr             = nlist->jjnr;
 424     shiftidx         = nlist->shift;
 425     gid              = nlist->gid;
 426     shiftvec         = fr->shift_vec[0];
 427     fshift           = fr->fshift[0];
 428     facel            = _mm256_set1_ps(fr->epsfac);
 429     charge           = mdatoms->chargeA;
 430
 431     /* Avoid stupid compiler warnings */
 432     jnrA = jnrB = jnrC = jnrD = jnrE = jnrF = jnrG = jnrH = 0;
 433     j_coord_offsetA = 0;
 434     j_coord_offsetB = 0;
 435     j_coord_offsetC = 0;
 436     j_coord_offsetD = 0;
 437     j_coord_offsetE = 0;
 438     j_coord_offsetF = 0;
 439     j_coord_offsetG = 0;
 440     j_coord_offsetH = 0;
 441
 442     outeriter        = 0;
 443     inneriter        = 0;
 444
 445     for(iidx=0;iidx<4*DIM;iidx++)
 446     {
 447         scratch[iidx] = 0.0;
 448     }
 449
 450     /* Start outer loop over neighborlists */
 451     for(iidx=0; iidx<nri; iidx++)
 452     {
 453         /* Load shift vector for this list */
 454         i_shift_offset   = DIM*shiftidx[iidx];
 455
 456         /* Load limits for loop over neighbors */
 457         j_index_start    = jindex[iidx];
 458         j_index_end      = jindex[iidx+1];
 459
 460         /* Get outer coordinate index */
 461         inr              = iinr[iidx];
 462         i_coord_offset   = DIM*inr;
 463
 464         /* Load i particle coords and add shift vector */
 465         gmx_mm256_load_shift_and_1rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 466
 467         fix0             = _mm256_setzero_ps();
 468         fiy0             = _mm256_setzero_ps();
 469         fiz0             = _mm256_setzero_ps();
 470
 471         /* Load parameters for i particles */
 472         iq0              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+0]));
 473
 474         /* Start inner kernel loop */
 475         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+7]>=0; jidx+=8)
 476         {
 477
 478             /* Get j neighbor index, and coordinate index */
 479             jnrA             = jjnr[jidx];
 480             jnrB             = jjnr[jidx+1];
 481             jnrC             = jjnr[jidx+2];
 482             jnrD             = jjnr[jidx+3];
 483             jnrE             = jjnr[jidx+4];
 484             jnrF             = jjnr[jidx+5];
 485             jnrG             = jjnr[jidx+6];
 486             jnrH             = jjnr[jidx+7];
 487             j_coord_offsetA  = DIM*jnrA;
 488             j_coord_offsetB  = DIM*jnrB;
 489             j_coord_offsetC  = DIM*jnrC;
 490             j_coord_offsetD  = DIM*jnrD;
 491             j_coord_offsetE  = DIM*jnrE;
 492             j_coord_offsetF  = DIM*jnrF;
 493             j_coord_offsetG  = DIM*jnrG;
 494             j_coord_offsetH  = DIM*jnrH;
 495
 496             /* load j atom coordinates */
 497             gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 498                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 499                                                  x+j_coord_offsetE,x+j_coord_offsetF,
 500                                                  x+j_coord_offsetG,x+j_coord_offsetH,
 501                                                  &jx0,&jy0,&jz0);
 502
 503             /* Calculate displacement vector */
 504             dx00             = _mm256_sub_ps(ix0,jx0);
 505             dy00             = _mm256_sub_ps(iy0,jy0);
 506             dz00             = _mm256_sub_ps(iz0,jz0);
 507
 508             /* Calculate squared distance and things based on it */
 509             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
 510
 511             rinv00           = gmx_mm256_invsqrt_ps(rsq00);
 512
 513             rinvsq00         = _mm256_mul_ps(rinv00,rinv00);
 514
 515             /* Load parameters for j particles */
 516             jq0              = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 517                                                                  charge+jnrC+0,charge+jnrD+0,
 518                                                                  charge+jnrE+0,charge+jnrF+0,
 519                                                                  charge+jnrG+0,charge+jnrH+0);
 520
 521             /**************************
 522              * CALCULATE INTERACTIONS *
 523              **************************/
 524
 525             /* Compute parameters for interactions between i and j atoms */
 526             qq00             = _mm256_mul_ps(iq0,jq0);
 527
 528             /* COULOMB ELECTROSTATICS */
 529             velec            = _mm256_mul_ps(qq00,rinv00);
 530             felec            = _mm256_mul_ps(velec,rinvsq00);
 531
 532             fscal            = felec;
 533
 534             /* Calculate temporary vectorial force */
 535             tx               = _mm256_mul_ps(fscal,dx00);
 536             ty               = _mm256_mul_ps(fscal,dy00);
 537             tz               = _mm256_mul_ps(fscal,dz00);
 538
 539             /* Update vectorial force */
 540             fix0             = _mm256_add_ps(fix0,tx);
 541             fiy0             = _mm256_add_ps(fiy0,ty);
 542             fiz0             = _mm256_add_ps(fiz0,tz);
 543
 544             fjptrA             = f+j_coord_offsetA;
 545             fjptrB             = f+j_coord_offsetB;
 546             fjptrC             = f+j_coord_offsetC;
 547             fjptrD             = f+j_coord_offsetD;
 548             fjptrE             = f+j_coord_offsetE;
 549             fjptrF             = f+j_coord_offsetF;
 550             fjptrG             = f+j_coord_offsetG;
 551             fjptrH             = f+j_coord_offsetH;
 552             gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,tx,ty,tz);
 553
 554             /* Inner loop uses 26 flops */
 555         }
 556
 557         if(jidx<j_index_end)
 558         {
 559
 560             /* Get j neighbor index, and coordinate index */
 561             jnrlistA         = jjnr[jidx];
 562             jnrlistB         = jjnr[jidx+1];
 563             jnrlistC         = jjnr[jidx+2];
 564             jnrlistD         = jjnr[jidx+3];
 565             jnrlistE         = jjnr[jidx+4];
 566             jnrlistF         = jjnr[jidx+5];
 567             jnrlistG         = jjnr[jidx+6];
 568             jnrlistH         = jjnr[jidx+7];
 569             /* Sign of each element will be negative for non-real atoms.
 570              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 571              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 572              */
 573             dummy_mask = gmx_mm256_set_m128(gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx+4)),_mm_setzero_si128())),
 574                                             gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128())));
 575
 576             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 577             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 578             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 579             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 580             jnrE       = (jnrlistE>=0) ? jnrlistE : 0;
 581             jnrF       = (jnrlistF>=0) ? jnrlistF : 0;
 582             jnrG       = (jnrlistG>=0) ? jnrlistG : 0;
 583             jnrH       = (jnrlistH>=0) ? jnrlistH : 0;
 584             j_coord_offsetA  = DIM*jnrA;
 585             j_coord_offsetB  = DIM*jnrB;
 586             j_coord_offsetC  = DIM*jnrC;
 587             j_coord_offsetD  = DIM*jnrD;
 588             j_coord_offsetE  = DIM*jnrE;
 589             j_coord_offsetF  = DIM*jnrF;
 590             j_coord_offsetG  = DIM*jnrG;
 591             j_coord_offsetH  = DIM*jnrH;
 592
 593             /* load j atom coordinates */
 594             gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 595                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 596                                                  x+j_coord_offsetE,x+j_coord_offsetF,
 597                                                  x+j_coord_offsetG,x+j_coord_offsetH,
 598                                                  &jx0,&jy0,&jz0);
 599
 600             /* Calculate displacement vector */
 601             dx00             = _mm256_sub_ps(ix0,jx0);
 602             dy00             = _mm256_sub_ps(iy0,jy0);
 603             dz00             = _mm256_sub_ps(iz0,jz0);
 604
 605             /* Calculate squared distance and things based on it */
 606             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
 607
 608             rinv00           = gmx_mm256_invsqrt_ps(rsq00);
 609
 610             rinvsq00         = _mm256_mul_ps(rinv00,rinv00);
 611
 612             /* Load parameters for j particles */
 613             jq0              = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 614                                                                  charge+jnrC+0,charge+jnrD+0,
 615                                                                  charge+jnrE+0,charge+jnrF+0,
 616                                                                  charge+jnrG+0,charge+jnrH+0);
 617
 618             /**************************
 619              * CALCULATE INTERACTIONS *
 620              **************************/
 621
 622             /* Compute parameters for interactions between i and j atoms */
 623             qq00             = _mm256_mul_ps(iq0,jq0);
 624
 625             /* COULOMB ELECTROSTATICS */
 626             velec            = _mm256_mul_ps(qq00,rinv00);
 627             felec            = _mm256_mul_ps(velec,rinvsq00);
 628
 629             fscal            = felec;
 630
 631             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
 632
 633             /* Calculate temporary vectorial force */
 634             tx               = _mm256_mul_ps(fscal,dx00);
 635             ty               = _mm256_mul_ps(fscal,dy00);
 636             tz               = _mm256_mul_ps(fscal,dz00);
 637
 638             /* Update vectorial force */
 639             fix0             = _mm256_add_ps(fix0,tx);
 640             fiy0             = _mm256_add_ps(fiy0,ty);
 641             fiz0             = _mm256_add_ps(fiz0,tz);
 642
 643             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 644             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 645             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 646             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 647             fjptrE             = (jnrlistE>=0) ? f+j_coord_offsetE : scratch;
 648             fjptrF             = (jnrlistF>=0) ? f+j_coord_offsetF : scratch;
 649             fjptrG             = (jnrlistG>=0) ? f+j_coord_offsetG : scratch;
 650             fjptrH             = (jnrlistH>=0) ? f+j_coord_offsetH : scratch;
 651             gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,tx,ty,tz);
 652
 653             /* Inner loop uses 26 flops */
 654         }
 655
 656         /* End of innermost loop */
 657
 658         gmx_mm256_update_iforce_1atom_swizzle_ps(fix0,fiy0,fiz0,
 659                                                  f+i_coord_offset,fshift+i_shift_offset);
 660
 661         /* Increment number of inner iterations */
 662         inneriter                  += j_index_end - j_index_start;
 663
 664         /* Outer loop uses 7 flops */
 665     }
 666
 667     /* Increment number of outer iterations */
 668     outeriter        += nri;
 669
 670     /* Update outer/inner flops */
 671
 672     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_F,outeriter*7 + inneriter*26);
 673 }