src/gmxlib/nonbonded/nb_kernel_avx_256_single/nb_kernel_ElecCoul_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_ElecCoul_VdwNone_GeomP1P1_VF_avx_256_single
  38  * Electrostatics interaction: Coulomb
  39  * VdW interaction:            None
  40  * Geometry:                   Particle-Particle
  41  * Calculate force/pot:        PotentialAndForce
  42  */
  43 void
  44 nb_kernel_ElecCoul_VdwNone_GeomP1P1_VF_avx_256_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
  97     /* Avoid stupid compiler warnings */
  98     jnrA = jnrB = jnrC = jnrD = jnrE = jnrF = jnrG = jnrH = 0;
  99     j_coord_offsetA = 0;
 100     j_coord_offsetB = 0;
 101     j_coord_offsetC = 0;
 102     j_coord_offsetD = 0;
 103     j_coord_offsetE = 0;
 104     j_coord_offsetF = 0;
 105     j_coord_offsetG = 0;
 106     j_coord_offsetH = 0;
 107
 108     outeriter        = 0;
 109     inneriter        = 0;
 110
 111     for(iidx=0;iidx<4*DIM;iidx++)
 112     {
 113         scratch[iidx] = 0.0;
 114     }
 115
 116     /* Start outer loop over neighborlists */
 117     for(iidx=0; iidx<nri; iidx++)
 118     {
 119         /* Load shift vector for this list */
 120         i_shift_offset   = DIM*shiftidx[iidx];
 121
 122         /* Load limits for loop over neighbors */
 123         j_index_start    = jindex[iidx];
 124         j_index_end      = jindex[iidx+1];
 125
 126         /* Get outer coordinate index */
 127         inr              = iinr[iidx];
 128         i_coord_offset   = DIM*inr;
 129
 130         /* Load i particle coords and add shift vector */
 131         gmx_mm256_load_shift_and_1rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 132
 133         fix0             = _mm256_setzero_ps();
 134         fiy0             = _mm256_setzero_ps();
 135         fiz0             = _mm256_setzero_ps();
 136
 137         /* Load parameters for i particles */
 138         iq0              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+0]));
 139
 140         /* Reset potential sums */
 141         velecsum         = _mm256_setzero_ps();
 142
 143         /* Start inner kernel loop */
 144         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+7]>=0; jidx+=8)
 145         {
 146
 147             /* Get j neighbor index, and coordinate index */
 148             jnrA             = jjnr[jidx];
 149             jnrB             = jjnr[jidx+1];
 150             jnrC             = jjnr[jidx+2];
 151             jnrD             = jjnr[jidx+3];
 152             jnrE             = jjnr[jidx+4];
 153             jnrF             = jjnr[jidx+5];
 154             jnrG             = jjnr[jidx+6];
 155             jnrH             = jjnr[jidx+7];
 156             j_coord_offsetA  = DIM*jnrA;
 157             j_coord_offsetB  = DIM*jnrB;
 158             j_coord_offsetC  = DIM*jnrC;
 159             j_coord_offsetD  = DIM*jnrD;
 160             j_coord_offsetE  = DIM*jnrE;
 161             j_coord_offsetF  = DIM*jnrF;
 162             j_coord_offsetG  = DIM*jnrG;
 163             j_coord_offsetH  = DIM*jnrH;
 164
 165             /* load j atom coordinates */
 166             gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 167                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 168                                                  x+j_coord_offsetE,x+j_coord_offsetF,
 169                                                  x+j_coord_offsetG,x+j_coord_offsetH,
 170                                                  &jx0,&jy0,&jz0);
 171
 172             /* Calculate displacement vector */
 173             dx00             = _mm256_sub_ps(ix0,jx0);
 174             dy00             = _mm256_sub_ps(iy0,jy0);
 175             dz00             = _mm256_sub_ps(iz0,jz0);
 176
 177             /* Calculate squared distance and things based on it */
 178             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
 179
 180             rinv00           = gmx_mm256_invsqrt_ps(rsq00);
 181
 182             rinvsq00         = _mm256_mul_ps(rinv00,rinv00);
 183
 184             /* Load parameters for j particles */
 185             jq0              = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 186                                                                  charge+jnrC+0,charge+jnrD+0,
 187                                                                  charge+jnrE+0,charge+jnrF+0,
 188                                                                  charge+jnrG+0,charge+jnrH+0);
 189
 190             /**************************
 191              * CALCULATE INTERACTIONS *
 192              **************************/
 193
 194             /* Compute parameters for interactions between i and j atoms */
 195             qq00             = _mm256_mul_ps(iq0,jq0);
 196
 197             /* COULOMB ELECTROSTATICS */
 198             velec            = _mm256_mul_ps(qq00,rinv00);
 199             felec            = _mm256_mul_ps(velec,rinvsq00);
 200
 201             /* Update potential sum for this i atom from the interaction with this j atom. */
 202             velecsum         = _mm256_add_ps(velecsum,velec);
 203
 204             fscal            = felec;
 205
 206             /* Calculate temporary vectorial force */
 207             tx               = _mm256_mul_ps(fscal,dx00);
 208             ty               = _mm256_mul_ps(fscal,dy00);
 209             tz               = _mm256_mul_ps(fscal,dz00);
 210
 211             /* Update vectorial force */
 212             fix0             = _mm256_add_ps(fix0,tx);
 213             fiy0             = _mm256_add_ps(fiy0,ty);
 214             fiz0             = _mm256_add_ps(fiz0,tz);
 215
 216             fjptrA             = f+j_coord_offsetA;
 217             fjptrB             = f+j_coord_offsetB;
 218             fjptrC             = f+j_coord_offsetC;
 219             fjptrD             = f+j_coord_offsetD;
 220             fjptrE             = f+j_coord_offsetE;
 221             fjptrF             = f+j_coord_offsetF;
 222             fjptrG             = f+j_coord_offsetG;
 223             fjptrH             = f+j_coord_offsetH;
 224             gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,tx,ty,tz);
 225
 226             /* Inner loop uses 27 flops */
 227         }
 228
 229         if(jidx<j_index_end)
 230         {
 231
 232             /* Get j neighbor index, and coordinate index */
 233             jnrlistA         = jjnr[jidx];
 234             jnrlistB         = jjnr[jidx+1];
 235             jnrlistC         = jjnr[jidx+2];
 236             jnrlistD         = jjnr[jidx+3];
 237             jnrlistE         = jjnr[jidx+4];
 238             jnrlistF         = jjnr[jidx+5];
 239             jnrlistG         = jjnr[jidx+6];
 240             jnrlistH         = jjnr[jidx+7];
 241             /* Sign of each element will be negative for non-real atoms.
 242              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 243              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 244              */
 245             dummy_mask = gmx_mm256_set_m128(gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx+4)),_mm_setzero_si128())),
 246                                             gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128())));
 247
 248             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 249             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 250             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 251             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 252             jnrE       = (jnrlistE>=0) ? jnrlistE : 0;
 253             jnrF       = (jnrlistF>=0) ? jnrlistF : 0;
 254             jnrG       = (jnrlistG>=0) ? jnrlistG : 0;
 255             jnrH       = (jnrlistH>=0) ? jnrlistH : 0;
 256             j_coord_offsetA  = DIM*jnrA;
 257             j_coord_offsetB  = DIM*jnrB;
 258             j_coord_offsetC  = DIM*jnrC;
 259             j_coord_offsetD  = DIM*jnrD;
 260             j_coord_offsetE  = DIM*jnrE;
 261             j_coord_offsetF  = DIM*jnrF;
 262             j_coord_offsetG  = DIM*jnrG;
 263             j_coord_offsetH  = DIM*jnrH;
 264
 265             /* load j atom coordinates */
 266             gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 267                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 268                                                  x+j_coord_offsetE,x+j_coord_offsetF,
 269                                                  x+j_coord_offsetG,x+j_coord_offsetH,
 270                                                  &jx0,&jy0,&jz0);
 271
 272             /* Calculate displacement vector */
 273             dx00             = _mm256_sub_ps(ix0,jx0);
 274             dy00             = _mm256_sub_ps(iy0,jy0);
 275             dz00             = _mm256_sub_ps(iz0,jz0);
 276
 277             /* Calculate squared distance and things based on it */
 278             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
 279
 280             rinv00           = gmx_mm256_invsqrt_ps(rsq00);
 281
 282             rinvsq00         = _mm256_mul_ps(rinv00,rinv00);
 283
 284             /* Load parameters for j particles */
 285             jq0              = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 286                                                                  charge+jnrC+0,charge+jnrD+0,
 287                                                                  charge+jnrE+0,charge+jnrF+0,
 288                                                                  charge+jnrG+0,charge+jnrH+0);
 289
 290             /**************************
 291              * CALCULATE INTERACTIONS *
 292              **************************/
 293
 294             /* Compute parameters for interactions between i and j atoms */
 295             qq00             = _mm256_mul_ps(iq0,jq0);
 296
 297             /* COULOMB ELECTROSTATICS */
 298             velec            = _mm256_mul_ps(qq00,rinv00);
 299             felec            = _mm256_mul_ps(velec,rinvsq00);
 300
 301             /* Update potential sum for this i atom from the interaction with this j atom. */
 302             velec            = _mm256_andnot_ps(dummy_mask,velec);
 303             velecsum         = _mm256_add_ps(velecsum,velec);
 304
 305             fscal            = felec;
 306
 307             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
 308
 309             /* Calculate temporary vectorial force */
 310             tx               = _mm256_mul_ps(fscal,dx00);
 311             ty               = _mm256_mul_ps(fscal,dy00);
 312             tz               = _mm256_mul_ps(fscal,dz00);
 313
 314             /* Update vectorial force */
 315             fix0             = _mm256_add_ps(fix0,tx);
 316             fiy0             = _mm256_add_ps(fiy0,ty);
 317             fiz0             = _mm256_add_ps(fiz0,tz);
 318
 319             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 320             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 321             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 322             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 323             fjptrE             = (jnrlistE>=0) ? f+j_coord_offsetE : scratch;
 324             fjptrF             = (jnrlistF>=0) ? f+j_coord_offsetF : scratch;
 325             fjptrG             = (jnrlistG>=0) ? f+j_coord_offsetG : scratch;
 326             fjptrH             = (jnrlistH>=0) ? f+j_coord_offsetH : scratch;
 327             gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,tx,ty,tz);
 328
 329             /* Inner loop uses 27 flops */
 330         }
 331
 332         /* End of innermost loop */
 333
 334         gmx_mm256_update_iforce_1atom_swizzle_ps(fix0,fiy0,fiz0,
 335                                                  f+i_coord_offset,fshift+i_shift_offset);
 336
 337         ggid                        = gid[iidx];
 338         /* Update potential energies */
 339         gmx_mm256_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid);
 340
 341         /* Increment number of inner iterations */
 342         inneriter                  += j_index_end - j_index_start;
 343
 344         /* Outer loop uses 8 flops */
 345     }
 346
 347     /* Increment number of outer iterations */
 348     outeriter        += nri;
 349
 350     /* Update outer/inner flops */
 351
 352     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VF,outeriter*8 + inneriter*27);
 353 }
 354 /*
 355  * Gromacs nonbonded kernel:   nb_kernel_ElecCoul_VdwNone_GeomP1P1_F_avx_256_single
 356  * Electrostatics interaction: Coulomb
 357  * VdW interaction:            None
 358  * Geometry:                   Particle-Particle
 359  * Calculate force/pot:        Force
 360  */
 361 void
 362 nb_kernel_ElecCoul_VdwNone_GeomP1P1_F_avx_256_single
 363                     (t_nblist * gmx_restrict                nlist,
 364                      rvec * gmx_restrict                    xx,
 365                      rvec * gmx_restrict                    ff,
 366                      t_forcerec * gmx_restrict              fr,
 367                      t_mdatoms * gmx_restrict               mdatoms,
 368                      nb_kernel_data_t * gmx_restrict        kernel_data,
 369                      t_nrnb * gmx_restrict                  nrnb)
 370 {
 371     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 372      * just 0 for non-waters.
 373      * Suffixes A,B,C,D,E,F,G,H refer to j loop unrolling done with AVX, e.g. for the eight different
 374      * jnr indices corresponding to data put in the four positions in the SIMD register.
 375      */
 376     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 377     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 378     int              jnrA,jnrB,jnrC,jnrD;
 379     int              jnrE,jnrF,jnrG,jnrH;
 380     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
 381     int              jnrlistE,jnrlistF,jnrlistG,jnrlistH;
 382     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 383     int              j_coord_offsetE,j_coord_offsetF,j_coord_offsetG,j_coord_offsetH;
 384     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 385     real             rcutoff_scalar;
 386     real             *shiftvec,*fshift,*x,*f;
 387     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD,*fjptrE,*fjptrF,*fjptrG,*fjptrH;
 388     real             scratch[4*DIM];
 389     __m256           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 390     real *           vdwioffsetptr0;
 391     __m256           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 392     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D,vdwjidx0E,vdwjidx0F,vdwjidx0G,vdwjidx0H;
 393     __m256           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 394     __m256           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 395     __m256           velec,felec,velecsum,facel,crf,krf,krf2;
 396     real             *charge;
 397     __m256           dummy_mask,cutoff_mask;
 398     __m256           signbit = _mm256_castsi256_ps( _mm256_set1_epi32(0x80000000) );
 399     __m256           one     = _mm256_set1_ps(1.0);
 400     __m256           two     = _mm256_set1_ps(2.0);
 401     x                = xx[0];
 402     f                = ff[0];
 403
 404     nri              = nlist->nri;
 405     iinr             = nlist->iinr;
 406     jindex           = nlist->jindex;
 407     jjnr             = nlist->jjnr;
 408     shiftidx         = nlist->shift;
 409     gid              = nlist->gid;
 410     shiftvec         = fr->shift_vec[0];
 411     fshift           = fr->fshift[0];
 412     facel            = _mm256_set1_ps(fr->epsfac);
 413     charge           = mdatoms->chargeA;
 414
 415     /* Avoid stupid compiler warnings */
 416     jnrA = jnrB = jnrC = jnrD = jnrE = jnrF = jnrG = jnrH = 0;
 417     j_coord_offsetA = 0;
 418     j_coord_offsetB = 0;
 419     j_coord_offsetC = 0;
 420     j_coord_offsetD = 0;
 421     j_coord_offsetE = 0;
 422     j_coord_offsetF = 0;
 423     j_coord_offsetG = 0;
 424     j_coord_offsetH = 0;
 425
 426     outeriter        = 0;
 427     inneriter        = 0;
 428
 429     for(iidx=0;iidx<4*DIM;iidx++)
 430     {
 431         scratch[iidx] = 0.0;
 432     }
 433
 434     /* Start outer loop over neighborlists */
 435     for(iidx=0; iidx<nri; iidx++)
 436     {
 437         /* Load shift vector for this list */
 438         i_shift_offset   = DIM*shiftidx[iidx];
 439
 440         /* Load limits for loop over neighbors */
 441         j_index_start    = jindex[iidx];
 442         j_index_end      = jindex[iidx+1];
 443
 444         /* Get outer coordinate index */
 445         inr              = iinr[iidx];
 446         i_coord_offset   = DIM*inr;
 447
 448         /* Load i particle coords and add shift vector */
 449         gmx_mm256_load_shift_and_1rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 450
 451         fix0             = _mm256_setzero_ps();
 452         fiy0             = _mm256_setzero_ps();
 453         fiz0             = _mm256_setzero_ps();
 454
 455         /* Load parameters for i particles */
 456         iq0              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+0]));
 457
 458         /* Start inner kernel loop */
 459         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+7]>=0; jidx+=8)
 460         {
 461
 462             /* Get j neighbor index, and coordinate index */
 463             jnrA             = jjnr[jidx];
 464             jnrB             = jjnr[jidx+1];
 465             jnrC             = jjnr[jidx+2];
 466             jnrD             = jjnr[jidx+3];
 467             jnrE             = jjnr[jidx+4];
 468             jnrF             = jjnr[jidx+5];
 469             jnrG             = jjnr[jidx+6];
 470             jnrH             = jjnr[jidx+7];
 471             j_coord_offsetA  = DIM*jnrA;
 472             j_coord_offsetB  = DIM*jnrB;
 473             j_coord_offsetC  = DIM*jnrC;
 474             j_coord_offsetD  = DIM*jnrD;
 475             j_coord_offsetE  = DIM*jnrE;
 476             j_coord_offsetF  = DIM*jnrF;
 477             j_coord_offsetG  = DIM*jnrG;
 478             j_coord_offsetH  = DIM*jnrH;
 479
 480             /* load j atom coordinates */
 481             gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 482                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 483                                                  x+j_coord_offsetE,x+j_coord_offsetF,
 484                                                  x+j_coord_offsetG,x+j_coord_offsetH,
 485                                                  &jx0,&jy0,&jz0);
 486
 487             /* Calculate displacement vector */
 488             dx00             = _mm256_sub_ps(ix0,jx0);
 489             dy00             = _mm256_sub_ps(iy0,jy0);
 490             dz00             = _mm256_sub_ps(iz0,jz0);
 491
 492             /* Calculate squared distance and things based on it */
 493             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
 494
 495             rinv00           = gmx_mm256_invsqrt_ps(rsq00);
 496
 497             rinvsq00         = _mm256_mul_ps(rinv00,rinv00);
 498
 499             /* Load parameters for j particles */
 500             jq0              = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 501                                                                  charge+jnrC+0,charge+jnrD+0,
 502                                                                  charge+jnrE+0,charge+jnrF+0,
 503                                                                  charge+jnrG+0,charge+jnrH+0);
 504
 505             /**************************
 506              * CALCULATE INTERACTIONS *
 507              **************************/
 508
 509             /* Compute parameters for interactions between i and j atoms */
 510             qq00             = _mm256_mul_ps(iq0,jq0);
 511
 512             /* COULOMB ELECTROSTATICS */
 513             velec            = _mm256_mul_ps(qq00,rinv00);
 514             felec            = _mm256_mul_ps(velec,rinvsq00);
 515
 516             fscal            = felec;
 517
 518             /* Calculate temporary vectorial force */
 519             tx               = _mm256_mul_ps(fscal,dx00);
 520             ty               = _mm256_mul_ps(fscal,dy00);
 521             tz               = _mm256_mul_ps(fscal,dz00);
 522
 523             /* Update vectorial force */
 524             fix0             = _mm256_add_ps(fix0,tx);
 525             fiy0             = _mm256_add_ps(fiy0,ty);
 526             fiz0             = _mm256_add_ps(fiz0,tz);
 527
 528             fjptrA             = f+j_coord_offsetA;
 529             fjptrB             = f+j_coord_offsetB;
 530             fjptrC             = f+j_coord_offsetC;
 531             fjptrD             = f+j_coord_offsetD;
 532             fjptrE             = f+j_coord_offsetE;
 533             fjptrF             = f+j_coord_offsetF;
 534             fjptrG             = f+j_coord_offsetG;
 535             fjptrH             = f+j_coord_offsetH;
 536             gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,tx,ty,tz);
 537
 538             /* Inner loop uses 26 flops */
 539         }
 540
 541         if(jidx<j_index_end)
 542         {
 543
 544             /* Get j neighbor index, and coordinate index */
 545             jnrlistA         = jjnr[jidx];
 546             jnrlistB         = jjnr[jidx+1];
 547             jnrlistC         = jjnr[jidx+2];
 548             jnrlistD         = jjnr[jidx+3];
 549             jnrlistE         = jjnr[jidx+4];
 550             jnrlistF         = jjnr[jidx+5];
 551             jnrlistG         = jjnr[jidx+6];
 552             jnrlistH         = jjnr[jidx+7];
 553             /* Sign of each element will be negative for non-real atoms.
 554              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 555              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 556              */
 557             dummy_mask = gmx_mm256_set_m128(gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx+4)),_mm_setzero_si128())),
 558                                             gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128())));
 559
 560             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 561             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 562             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 563             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 564             jnrE       = (jnrlistE>=0) ? jnrlistE : 0;
 565             jnrF       = (jnrlistF>=0) ? jnrlistF : 0;
 566             jnrG       = (jnrlistG>=0) ? jnrlistG : 0;
 567             jnrH       = (jnrlistH>=0) ? jnrlistH : 0;
 568             j_coord_offsetA  = DIM*jnrA;
 569             j_coord_offsetB  = DIM*jnrB;
 570             j_coord_offsetC  = DIM*jnrC;
 571             j_coord_offsetD  = DIM*jnrD;
 572             j_coord_offsetE  = DIM*jnrE;
 573             j_coord_offsetF  = DIM*jnrF;
 574             j_coord_offsetG  = DIM*jnrG;
 575             j_coord_offsetH  = DIM*jnrH;
 576
 577             /* load j atom coordinates */
 578             gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 579                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 580                                                  x+j_coord_offsetE,x+j_coord_offsetF,
 581                                                  x+j_coord_offsetG,x+j_coord_offsetH,
 582                                                  &jx0,&jy0,&jz0);
 583
 584             /* Calculate displacement vector */
 585             dx00             = _mm256_sub_ps(ix0,jx0);
 586             dy00             = _mm256_sub_ps(iy0,jy0);
 587             dz00             = _mm256_sub_ps(iz0,jz0);
 588
 589             /* Calculate squared distance and things based on it */
 590             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
 591
 592             rinv00           = gmx_mm256_invsqrt_ps(rsq00);
 593
 594             rinvsq00         = _mm256_mul_ps(rinv00,rinv00);
 595
 596             /* Load parameters for j particles */
 597             jq0              = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 598                                                                  charge+jnrC+0,charge+jnrD+0,
 599                                                                  charge+jnrE+0,charge+jnrF+0,
 600                                                                  charge+jnrG+0,charge+jnrH+0);
 601
 602             /**************************
 603              * CALCULATE INTERACTIONS *
 604              **************************/
 605
 606             /* Compute parameters for interactions between i and j atoms */
 607             qq00             = _mm256_mul_ps(iq0,jq0);
 608
 609             /* COULOMB ELECTROSTATICS */
 610             velec            = _mm256_mul_ps(qq00,rinv00);
 611             felec            = _mm256_mul_ps(velec,rinvsq00);
 612
 613             fscal            = felec;
 614
 615             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
 616
 617             /* Calculate temporary vectorial force */
 618             tx               = _mm256_mul_ps(fscal,dx00);
 619             ty               = _mm256_mul_ps(fscal,dy00);
 620             tz               = _mm256_mul_ps(fscal,dz00);
 621
 622             /* Update vectorial force */
 623             fix0             = _mm256_add_ps(fix0,tx);
 624             fiy0             = _mm256_add_ps(fiy0,ty);
 625             fiz0             = _mm256_add_ps(fiz0,tz);
 626
 627             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 628             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 629             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 630             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 631             fjptrE             = (jnrlistE>=0) ? f+j_coord_offsetE : scratch;
 632             fjptrF             = (jnrlistF>=0) ? f+j_coord_offsetF : scratch;
 633             fjptrG             = (jnrlistG>=0) ? f+j_coord_offsetG : scratch;
 634             fjptrH             = (jnrlistH>=0) ? f+j_coord_offsetH : scratch;
 635             gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,tx,ty,tz);
 636
 637             /* Inner loop uses 26 flops */
 638         }
 639
 640         /* End of innermost loop */
 641
 642         gmx_mm256_update_iforce_1atom_swizzle_ps(fix0,fiy0,fiz0,
 643                                                  f+i_coord_offset,fshift+i_shift_offset);
 644
 645         /* Increment number of inner iterations */
 646         inneriter                  += j_index_end - j_index_start;
 647
 648         /* Outer loop uses 7 flops */
 649     }
 650
 651     /* Increment number of outer iterations */
 652     outeriter        += nri;
 653
 654     /* Update outer/inner flops */
 655
 656     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_F,outeriter*7 + inneriter*26);
 657 }