src/gmxlib/nonbonded/nb_kernel_avx_256_double/nb_kernel_ElecRFCut_VdwNone_GeomP1P1_avx_256_double.c

   1 /*
   2  * Note: this file was generated by the Gromacs avx_256_double kernel generator.
   3  *
   4  *                This source code is part of
   5  *
   6  *                 G   R   O   M   A   C   S
   7  *
   8  * Copyright (c) 2001-2012, The GROMACS Development Team
   9  *
  10  * Gromacs is a library for molecular simulation and trajectory analysis,
  11  * written by Erik Lindahl, David van der Spoel, Berk Hess, and others - for
  12  * a full list of developers and information, check out http://www.gromacs.org
  13  *
  14  * This program is free software; you can redistribute it and/or modify it under
  15  * the terms of the GNU Lesser General Public License as published by the Free
  16  * Software Foundation; either version 2 of the License, or (at your option) any
  17  * later version.
  18  *
  19  * To help fund GROMACS development, we humbly ask that you cite
  20  * the papers people have written on it - you can find them on the website.
  21  */
  22 #ifdef HAVE_CONFIG_H
  23 #include <config.h>
  24 #endif
  25
  26 #include <math.h>
  27
  28 #include "../nb_kernel.h"
  29 #include "types/simple.h"
  30 #include "vec.h"
  31 #include "nrnb.h"
  32
  33 #include "gmx_math_x86_avx_256_double.h"
  34 #include "kernelutil_x86_avx_256_double.h"
  35
  36 /*
  37  * Gromacs nonbonded kernel:   nb_kernel_ElecRFCut_VdwNone_GeomP1P1_VF_avx_256_double
  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_double
  45                     (t_nblist * gmx_restrict                nlist,
  46                      rvec * gmx_restrict                    xx,
  47                      rvec * gmx_restrict                    ff,
  48                      t_forcerec * gmx_restrict              fr,
  49                      t_mdatoms * gmx_restrict               mdatoms,
  50                      nb_kernel_data_t * gmx_restrict        kernel_data,
  51                      t_nrnb * gmx_restrict                  nrnb)
  52 {
  53     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
  54      * just 0 for non-waters.
  55      * Suffixes A,B,C,D refer to j loop unrolling done with AVX, e.g. for the four different
  56      * jnr indices corresponding to data put in the four positions in the SIMD register.
  57      */
  58     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
  59     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
  60     int              jnrA,jnrB,jnrC,jnrD;
  61     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
  62     int              jnrlistE,jnrlistF,jnrlistG,jnrlistH;
  63     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
  64     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
  65     real             rcutoff_scalar;
  66     real             *shiftvec,*fshift,*x,*f;
  67     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
  68     real             scratch[4*DIM];
  69     __m256d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
  70     real *           vdwioffsetptr0;
  71     __m256d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
  72     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
  73     __m256d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  74     __m256d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  75     __m256d          velec,felec,velecsum,facel,crf,krf,krf2;
  76     real             *charge;
  77     __m256d          dummy_mask,cutoff_mask;
  78     __m128           tmpmask0,tmpmask1;
  79     __m256d          signbit = _mm256_castsi256_pd( _mm256_set1_epi32(0x80000000) );
  80     __m256d          one     = _mm256_set1_pd(1.0);
  81     __m256d          two     = _mm256_set1_pd(2.0);
  82     x                = xx[0];
  83     f                = ff[0];
  84
  85     nri              = nlist->nri;
  86     iinr             = nlist->iinr;
  87     jindex           = nlist->jindex;
  88     jjnr             = nlist->jjnr;
  89     shiftidx         = nlist->shift;
  90     gid              = nlist->gid;
  91     shiftvec         = fr->shift_vec[0];
  92     fshift           = fr->fshift[0];
  93     facel            = _mm256_set1_pd(fr->epsfac);
  94     charge           = mdatoms->chargeA;
  95     krf              = _mm256_set1_pd(fr->ic->k_rf);
  96     krf2             = _mm256_set1_pd(fr->ic->k_rf*2.0);
  97     crf              = _mm256_set1_pd(fr->ic->c_rf);
  98
  99     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 100     rcutoff_scalar   = fr->rcoulomb;
 101     rcutoff          = _mm256_set1_pd(rcutoff_scalar);
 102     rcutoff2         = _mm256_mul_pd(rcutoff,rcutoff);
 103
 104     /* Avoid stupid compiler warnings */
 105     jnrA = jnrB = jnrC = jnrD = 0;
 106     j_coord_offsetA = 0;
 107     j_coord_offsetB = 0;
 108     j_coord_offsetC = 0;
 109     j_coord_offsetD = 0;
 110
 111     outeriter        = 0;
 112     inneriter        = 0;
 113
 114     for(iidx=0;iidx<4*DIM;iidx++)
 115     {
 116         scratch[iidx] = 0.0;
 117     }
 118
 119     /* Start outer loop over neighborlists */
 120     for(iidx=0; iidx<nri; iidx++)
 121     {
 122         /* Load shift vector for this list */
 123         i_shift_offset   = DIM*shiftidx[iidx];
 124
 125         /* Load limits for loop over neighbors */
 126         j_index_start    = jindex[iidx];
 127         j_index_end      = jindex[iidx+1];
 128
 129         /* Get outer coordinate index */
 130         inr              = iinr[iidx];
 131         i_coord_offset   = DIM*inr;
 132
 133         /* Load i particle coords and add shift vector */
 134         gmx_mm256_load_shift_and_1rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 135
 136         fix0             = _mm256_setzero_pd();
 137         fiy0             = _mm256_setzero_pd();
 138         fiz0             = _mm256_setzero_pd();
 139
 140         /* Load parameters for i particles */
 141         iq0              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+0]));
 142
 143         /* Reset potential sums */
 144         velecsum         = _mm256_setzero_pd();
 145
 146         /* Start inner kernel loop */
 147         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 148         {
 149
 150             /* Get j neighbor index, and coordinate index */
 151             jnrA             = jjnr[jidx];
 152             jnrB             = jjnr[jidx+1];
 153             jnrC             = jjnr[jidx+2];
 154             jnrD             = jjnr[jidx+3];
 155             j_coord_offsetA  = DIM*jnrA;
 156             j_coord_offsetB  = DIM*jnrB;
 157             j_coord_offsetC  = DIM*jnrC;
 158             j_coord_offsetD  = DIM*jnrD;
 159
 160             /* load j atom coordinates */
 161             gmx_mm256_load_1rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 162                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 163                                                  &jx0,&jy0,&jz0);
 164
 165             /* Calculate displacement vector */
 166             dx00             = _mm256_sub_pd(ix0,jx0);
 167             dy00             = _mm256_sub_pd(iy0,jy0);
 168             dz00             = _mm256_sub_pd(iz0,jz0);
 169
 170             /* Calculate squared distance and things based on it */
 171             rsq00            = gmx_mm256_calc_rsq_pd(dx00,dy00,dz00);
 172
 173             rinv00           = gmx_mm256_invsqrt_pd(rsq00);
 174
 175             rinvsq00         = _mm256_mul_pd(rinv00,rinv00);
 176
 177             /* Load parameters for j particles */
 178             jq0              = gmx_mm256_load_4real_swizzle_pd(charge+jnrA+0,charge+jnrB+0,
 179                                                                  charge+jnrC+0,charge+jnrD+0);
 180
 181             /**************************
 182              * CALCULATE INTERACTIONS *
 183              **************************/
 184
 185             if (gmx_mm256_any_lt(rsq00,rcutoff2))
 186             {
 187
 188             /* Compute parameters for interactions between i and j atoms */
 189             qq00             = _mm256_mul_pd(iq0,jq0);
 190
 191             /* REACTION-FIELD ELECTROSTATICS */
 192             velec            = _mm256_mul_pd(qq00,_mm256_sub_pd(_mm256_add_pd(rinv00,_mm256_mul_pd(krf,rsq00)),crf));
 193             felec            = _mm256_mul_pd(qq00,_mm256_sub_pd(_mm256_mul_pd(rinv00,rinvsq00),krf2));
 194
 195             cutoff_mask      = _mm256_cmp_pd(rsq00,rcutoff2,_CMP_LT_OQ);
 196
 197             /* Update potential sum for this i atom from the interaction with this j atom. */
 198             velec            = _mm256_and_pd(velec,cutoff_mask);
 199             velecsum         = _mm256_add_pd(velecsum,velec);
 200
 201             fscal            = felec;
 202
 203             fscal            = _mm256_and_pd(fscal,cutoff_mask);
 204
 205             /* Calculate temporary vectorial force */
 206             tx               = _mm256_mul_pd(fscal,dx00);
 207             ty               = _mm256_mul_pd(fscal,dy00);
 208             tz               = _mm256_mul_pd(fscal,dz00);
 209
 210             /* Update vectorial force */
 211             fix0             = _mm256_add_pd(fix0,tx);
 212             fiy0             = _mm256_add_pd(fiy0,ty);
 213             fiz0             = _mm256_add_pd(fiz0,tz);
 214
 215             fjptrA             = f+j_coord_offsetA;
 216             fjptrB             = f+j_coord_offsetB;
 217             fjptrC             = f+j_coord_offsetC;
 218             fjptrD             = f+j_coord_offsetD;
 219             gmx_mm256_decrement_1rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 220
 221             }
 222
 223             /* Inner loop uses 36 flops */
 224         }
 225
 226         if(jidx<j_index_end)
 227         {
 228
 229             /* Get j neighbor index, and coordinate index */
 230             jnrlistA         = jjnr[jidx];
 231             jnrlistB         = jjnr[jidx+1];
 232             jnrlistC         = jjnr[jidx+2];
 233             jnrlistD         = jjnr[jidx+3];
 234             /* Sign of each element will be negative for non-real atoms.
 235              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 236              * so use it as val = _mm_andnot_pd(mask,val) to clear dummy entries.
 237              */
 238             tmpmask0 = gmx_mm_castsi128_pd(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 239
 240             tmpmask1 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(3,3,2,2));
 241             tmpmask0 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(1,1,0,0));
 242             dummy_mask = _mm256_castps_pd(gmx_mm256_set_m128(tmpmask1,tmpmask0));
 243
 244             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 245             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 246             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 247             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 248             j_coord_offsetA  = DIM*jnrA;
 249             j_coord_offsetB  = DIM*jnrB;
 250             j_coord_offsetC  = DIM*jnrC;
 251             j_coord_offsetD  = DIM*jnrD;
 252
 253             /* load j atom coordinates */
 254             gmx_mm256_load_1rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 255                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 256                                                  &jx0,&jy0,&jz0);
 257
 258             /* Calculate displacement vector */
 259             dx00             = _mm256_sub_pd(ix0,jx0);
 260             dy00             = _mm256_sub_pd(iy0,jy0);
 261             dz00             = _mm256_sub_pd(iz0,jz0);
 262
 263             /* Calculate squared distance and things based on it */
 264             rsq00            = gmx_mm256_calc_rsq_pd(dx00,dy00,dz00);
 265
 266             rinv00           = gmx_mm256_invsqrt_pd(rsq00);
 267
 268             rinvsq00         = _mm256_mul_pd(rinv00,rinv00);
 269
 270             /* Load parameters for j particles */
 271             jq0              = gmx_mm256_load_4real_swizzle_pd(charge+jnrA+0,charge+jnrB+0,
 272                                                                  charge+jnrC+0,charge+jnrD+0);
 273
 274             /**************************
 275              * CALCULATE INTERACTIONS *
 276              **************************/
 277
 278             if (gmx_mm256_any_lt(rsq00,rcutoff2))
 279             {
 280
 281             /* Compute parameters for interactions between i and j atoms */
 282             qq00             = _mm256_mul_pd(iq0,jq0);
 283
 284             /* REACTION-FIELD ELECTROSTATICS */
 285             velec            = _mm256_mul_pd(qq00,_mm256_sub_pd(_mm256_add_pd(rinv00,_mm256_mul_pd(krf,rsq00)),crf));
 286             felec            = _mm256_mul_pd(qq00,_mm256_sub_pd(_mm256_mul_pd(rinv00,rinvsq00),krf2));
 287
 288             cutoff_mask      = _mm256_cmp_pd(rsq00,rcutoff2,_CMP_LT_OQ);
 289
 290             /* Update potential sum for this i atom from the interaction with this j atom. */
 291             velec            = _mm256_and_pd(velec,cutoff_mask);
 292             velec            = _mm256_andnot_pd(dummy_mask,velec);
 293             velecsum         = _mm256_add_pd(velecsum,velec);
 294
 295             fscal            = felec;
 296
 297             fscal            = _mm256_and_pd(fscal,cutoff_mask);
 298
 299             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
 300
 301             /* Calculate temporary vectorial force */
 302             tx               = _mm256_mul_pd(fscal,dx00);
 303             ty               = _mm256_mul_pd(fscal,dy00);
 304             tz               = _mm256_mul_pd(fscal,dz00);
 305
 306             /* Update vectorial force */
 307             fix0             = _mm256_add_pd(fix0,tx);
 308             fiy0             = _mm256_add_pd(fiy0,ty);
 309             fiz0             = _mm256_add_pd(fiz0,tz);
 310
 311             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 312             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 313             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 314             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 315             gmx_mm256_decrement_1rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 316
 317             }
 318
 319             /* Inner loop uses 36 flops */
 320         }
 321
 322         /* End of innermost loop */
 323
 324         gmx_mm256_update_iforce_1atom_swizzle_pd(fix0,fiy0,fiz0,
 325                                                  f+i_coord_offset,fshift+i_shift_offset);
 326
 327         ggid                        = gid[iidx];
 328         /* Update potential energies */
 329         gmx_mm256_update_1pot_pd(velecsum,kernel_data->energygrp_elec+ggid);
 330
 331         /* Increment number of inner iterations */
 332         inneriter                  += j_index_end - j_index_start;
 333
 334         /* Outer loop uses 8 flops */
 335     }
 336
 337     /* Increment number of outer iterations */
 338     outeriter        += nri;
 339
 340     /* Update outer/inner flops */
 341
 342     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VF,outeriter*8 + inneriter*36);
 343 }
 344 /*
 345  * Gromacs nonbonded kernel:   nb_kernel_ElecRFCut_VdwNone_GeomP1P1_F_avx_256_double
 346  * Electrostatics interaction: ReactionField
 347  * VdW interaction:            None
 348  * Geometry:                   Particle-Particle
 349  * Calculate force/pot:        Force
 350  */
 351 void
 352 nb_kernel_ElecRFCut_VdwNone_GeomP1P1_F_avx_256_double
 353                     (t_nblist * gmx_restrict                nlist,
 354                      rvec * gmx_restrict                    xx,
 355                      rvec * gmx_restrict                    ff,
 356                      t_forcerec * gmx_restrict              fr,
 357                      t_mdatoms * gmx_restrict               mdatoms,
 358                      nb_kernel_data_t * gmx_restrict        kernel_data,
 359                      t_nrnb * gmx_restrict                  nrnb)
 360 {
 361     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 362      * just 0 for non-waters.
 363      * Suffixes A,B,C,D refer to j loop unrolling done with AVX, e.g. for the four different
 364      * jnr indices corresponding to data put in the four positions in the SIMD register.
 365      */
 366     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 367     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 368     int              jnrA,jnrB,jnrC,jnrD;
 369     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
 370     int              jnrlistE,jnrlistF,jnrlistG,jnrlistH;
 371     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 372     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 373     real             rcutoff_scalar;
 374     real             *shiftvec,*fshift,*x,*f;
 375     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
 376     real             scratch[4*DIM];
 377     __m256d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 378     real *           vdwioffsetptr0;
 379     __m256d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 380     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
 381     __m256d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 382     __m256d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 383     __m256d          velec,felec,velecsum,facel,crf,krf,krf2;
 384     real             *charge;
 385     __m256d          dummy_mask,cutoff_mask;
 386     __m128           tmpmask0,tmpmask1;
 387     __m256d          signbit = _mm256_castsi256_pd( _mm256_set1_epi32(0x80000000) );
 388     __m256d          one     = _mm256_set1_pd(1.0);
 389     __m256d          two     = _mm256_set1_pd(2.0);
 390     x                = xx[0];
 391     f                = ff[0];
 392
 393     nri              = nlist->nri;
 394     iinr             = nlist->iinr;
 395     jindex           = nlist->jindex;
 396     jjnr             = nlist->jjnr;
 397     shiftidx         = nlist->shift;
 398     gid              = nlist->gid;
 399     shiftvec         = fr->shift_vec[0];
 400     fshift           = fr->fshift[0];
 401     facel            = _mm256_set1_pd(fr->epsfac);
 402     charge           = mdatoms->chargeA;
 403     krf              = _mm256_set1_pd(fr->ic->k_rf);
 404     krf2             = _mm256_set1_pd(fr->ic->k_rf*2.0);
 405     crf              = _mm256_set1_pd(fr->ic->c_rf);
 406
 407     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 408     rcutoff_scalar   = fr->rcoulomb;
 409     rcutoff          = _mm256_set1_pd(rcutoff_scalar);
 410     rcutoff2         = _mm256_mul_pd(rcutoff,rcutoff);
 411
 412     /* Avoid stupid compiler warnings */
 413     jnrA = jnrB = jnrC = jnrD = 0;
 414     j_coord_offsetA = 0;
 415     j_coord_offsetB = 0;
 416     j_coord_offsetC = 0;
 417     j_coord_offsetD = 0;
 418
 419     outeriter        = 0;
 420     inneriter        = 0;
 421
 422     for(iidx=0;iidx<4*DIM;iidx++)
 423     {
 424         scratch[iidx] = 0.0;
 425     }
 426
 427     /* Start outer loop over neighborlists */
 428     for(iidx=0; iidx<nri; iidx++)
 429     {
 430         /* Load shift vector for this list */
 431         i_shift_offset   = DIM*shiftidx[iidx];
 432
 433         /* Load limits for loop over neighbors */
 434         j_index_start    = jindex[iidx];
 435         j_index_end      = jindex[iidx+1];
 436
 437         /* Get outer coordinate index */
 438         inr              = iinr[iidx];
 439         i_coord_offset   = DIM*inr;
 440
 441         /* Load i particle coords and add shift vector */
 442         gmx_mm256_load_shift_and_1rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 443
 444         fix0             = _mm256_setzero_pd();
 445         fiy0             = _mm256_setzero_pd();
 446         fiz0             = _mm256_setzero_pd();
 447
 448         /* Load parameters for i particles */
 449         iq0              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+0]));
 450
 451         /* Start inner kernel loop */
 452         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 453         {
 454
 455             /* Get j neighbor index, and coordinate index */
 456             jnrA             = jjnr[jidx];
 457             jnrB             = jjnr[jidx+1];
 458             jnrC             = jjnr[jidx+2];
 459             jnrD             = jjnr[jidx+3];
 460             j_coord_offsetA  = DIM*jnrA;
 461             j_coord_offsetB  = DIM*jnrB;
 462             j_coord_offsetC  = DIM*jnrC;
 463             j_coord_offsetD  = DIM*jnrD;
 464
 465             /* load j atom coordinates */
 466             gmx_mm256_load_1rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 467                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 468                                                  &jx0,&jy0,&jz0);
 469
 470             /* Calculate displacement vector */
 471             dx00             = _mm256_sub_pd(ix0,jx0);
 472             dy00             = _mm256_sub_pd(iy0,jy0);
 473             dz00             = _mm256_sub_pd(iz0,jz0);
 474
 475             /* Calculate squared distance and things based on it */
 476             rsq00            = gmx_mm256_calc_rsq_pd(dx00,dy00,dz00);
 477
 478             rinv00           = gmx_mm256_invsqrt_pd(rsq00);
 479
 480             rinvsq00         = _mm256_mul_pd(rinv00,rinv00);
 481
 482             /* Load parameters for j particles */
 483             jq0              = gmx_mm256_load_4real_swizzle_pd(charge+jnrA+0,charge+jnrB+0,
 484                                                                  charge+jnrC+0,charge+jnrD+0);
 485
 486             /**************************
 487              * CALCULATE INTERACTIONS *
 488              **************************/
 489
 490             if (gmx_mm256_any_lt(rsq00,rcutoff2))
 491             {
 492
 493             /* Compute parameters for interactions between i and j atoms */
 494             qq00             = _mm256_mul_pd(iq0,jq0);
 495
 496             /* REACTION-FIELD ELECTROSTATICS */
 497             felec            = _mm256_mul_pd(qq00,_mm256_sub_pd(_mm256_mul_pd(rinv00,rinvsq00),krf2));
 498
 499             cutoff_mask      = _mm256_cmp_pd(rsq00,rcutoff2,_CMP_LT_OQ);
 500
 501             fscal            = felec;
 502
 503             fscal            = _mm256_and_pd(fscal,cutoff_mask);
 504
 505             /* Calculate temporary vectorial force */
 506             tx               = _mm256_mul_pd(fscal,dx00);
 507             ty               = _mm256_mul_pd(fscal,dy00);
 508             tz               = _mm256_mul_pd(fscal,dz00);
 509
 510             /* Update vectorial force */
 511             fix0             = _mm256_add_pd(fix0,tx);
 512             fiy0             = _mm256_add_pd(fiy0,ty);
 513             fiz0             = _mm256_add_pd(fiz0,tz);
 514
 515             fjptrA             = f+j_coord_offsetA;
 516             fjptrB             = f+j_coord_offsetB;
 517             fjptrC             = f+j_coord_offsetC;
 518             fjptrD             = f+j_coord_offsetD;
 519             gmx_mm256_decrement_1rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 520
 521             }
 522
 523             /* Inner loop uses 30 flops */
 524         }
 525
 526         if(jidx<j_index_end)
 527         {
 528
 529             /* Get j neighbor index, and coordinate index */
 530             jnrlistA         = jjnr[jidx];
 531             jnrlistB         = jjnr[jidx+1];
 532             jnrlistC         = jjnr[jidx+2];
 533             jnrlistD         = jjnr[jidx+3];
 534             /* Sign of each element will be negative for non-real atoms.
 535              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 536              * so use it as val = _mm_andnot_pd(mask,val) to clear dummy entries.
 537              */
 538             tmpmask0 = gmx_mm_castsi128_pd(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 539
 540             tmpmask1 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(3,3,2,2));
 541             tmpmask0 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(1,1,0,0));
 542             dummy_mask = _mm256_castps_pd(gmx_mm256_set_m128(tmpmask1,tmpmask0));
 543
 544             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 545             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 546             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 547             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 548             j_coord_offsetA  = DIM*jnrA;
 549             j_coord_offsetB  = DIM*jnrB;
 550             j_coord_offsetC  = DIM*jnrC;
 551             j_coord_offsetD  = DIM*jnrD;
 552
 553             /* load j atom coordinates */
 554             gmx_mm256_load_1rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 555                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 556                                                  &jx0,&jy0,&jz0);
 557
 558             /* Calculate displacement vector */
 559             dx00             = _mm256_sub_pd(ix0,jx0);
 560             dy00             = _mm256_sub_pd(iy0,jy0);
 561             dz00             = _mm256_sub_pd(iz0,jz0);
 562
 563             /* Calculate squared distance and things based on it */
 564             rsq00            = gmx_mm256_calc_rsq_pd(dx00,dy00,dz00);
 565
 566             rinv00           = gmx_mm256_invsqrt_pd(rsq00);
 567
 568             rinvsq00         = _mm256_mul_pd(rinv00,rinv00);
 569
 570             /* Load parameters for j particles */
 571             jq0              = gmx_mm256_load_4real_swizzle_pd(charge+jnrA+0,charge+jnrB+0,
 572                                                                  charge+jnrC+0,charge+jnrD+0);
 573
 574             /**************************
 575              * CALCULATE INTERACTIONS *
 576              **************************/
 577
 578             if (gmx_mm256_any_lt(rsq00,rcutoff2))
 579             {
 580
 581             /* Compute parameters for interactions between i and j atoms */
 582             qq00             = _mm256_mul_pd(iq0,jq0);
 583
 584             /* REACTION-FIELD ELECTROSTATICS */
 585             felec            = _mm256_mul_pd(qq00,_mm256_sub_pd(_mm256_mul_pd(rinv00,rinvsq00),krf2));
 586
 587             cutoff_mask      = _mm256_cmp_pd(rsq00,rcutoff2,_CMP_LT_OQ);
 588
 589             fscal            = felec;
 590
 591             fscal            = _mm256_and_pd(fscal,cutoff_mask);
 592
 593             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
 594
 595             /* Calculate temporary vectorial force */
 596             tx               = _mm256_mul_pd(fscal,dx00);
 597             ty               = _mm256_mul_pd(fscal,dy00);
 598             tz               = _mm256_mul_pd(fscal,dz00);
 599
 600             /* Update vectorial force */
 601             fix0             = _mm256_add_pd(fix0,tx);
 602             fiy0             = _mm256_add_pd(fiy0,ty);
 603             fiz0             = _mm256_add_pd(fiz0,tz);
 604
 605             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 606             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 607             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 608             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 609             gmx_mm256_decrement_1rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 610
 611             }
 612
 613             /* Inner loop uses 30 flops */
 614         }
 615
 616         /* End of innermost loop */
 617
 618         gmx_mm256_update_iforce_1atom_swizzle_pd(fix0,fiy0,fiz0,
 619                                                  f+i_coord_offset,fshift+i_shift_offset);
 620
 621         /* Increment number of inner iterations */
 622         inneriter                  += j_index_end - j_index_start;
 623
 624         /* Outer loop uses 7 flops */
 625     }
 626
 627     /* Increment number of outer iterations */
 628     outeriter        += nri;
 629
 630     /* Update outer/inner flops */
 631
 632     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_F,outeriter*7 + inneriter*30);
 633 }