src/gromacs/gmxlib/nonbonded/nb_kernel_avx_256_single/nb_kernel_ElecRFCut_VdwLJSh_GeomW3P1_avx_256_single.c

   1 /*
   2  * Note: this file was generated by the Gromacs avx_256_single kernel generator.
   3  *
   4  *                This source code is part of
   5  *
   6  *                 G   R   O   M   A   C   S
   7  *
   8  * Copyright (c) 2001-2012, The GROMACS Development Team
   9  *
  10  * Gromacs is a library for molecular simulation and trajectory analysis,
  11  * written by Erik Lindahl, David van der Spoel, Berk Hess, and others - for
  12  * a full list of developers and information, check out http://www.gromacs.org
  13  *
  14  * This program is free software; you can redistribute it and/or modify it under
  15  * the terms of the GNU Lesser General Public License as published by the Free
  16  * Software Foundation; either version 2 of the License, or (at your option) any
  17  * later version.
  18  *
  19  * To help fund GROMACS development, we humbly ask that you cite
  20  * the papers people have written on it - you can find them on the website.
  21  */
  22 #ifdef HAVE_CONFIG_H
  23 #include <config.h>
  24 #endif
  25
  26 #include <math.h>
  27
  28 #include "../nb_kernel.h"
  29 #include "types/simple.h"
  30 #include "vec.h"
  31 #include "nrnb.h"
  32
  33 #include "gmx_math_x86_avx_256_single.h"
  34 #include "kernelutil_x86_avx_256_single.h"
  35
  36 /*
  37  * Gromacs nonbonded kernel:   nb_kernel_ElecRFCut_VdwLJSh_GeomW3P1_VF_avx_256_single
  38  * Electrostatics interaction: ReactionField
  39  * VdW interaction:            LennardJones
  40  * Geometry:                   Water3-Particle
  41  * Calculate force/pot:        PotentialAndForce
  42  */
  43 void
  44 nb_kernel_ElecRFCut_VdwLJSh_GeomW3P1_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     real *           vdwioffsetptr1;
  75     __m256           ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
  76     real *           vdwioffsetptr2;
  77     __m256           ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
  78     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D,vdwjidx0E,vdwjidx0F,vdwjidx0G,vdwjidx0H;
  79     __m256           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  80     __m256           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  81     __m256           dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
  82     __m256           dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
  83     __m256           velec,felec,velecsum,facel,crf,krf,krf2;
  84     real             *charge;
  85     int              nvdwtype;
  86     __m256           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
  87     int              *vdwtype;
  88     real             *vdwparam;
  89     __m256           one_sixth   = _mm256_set1_ps(1.0/6.0);
  90     __m256           one_twelfth = _mm256_set1_ps(1.0/12.0);
  91     __m256           dummy_mask,cutoff_mask;
  92     __m256           signbit = _mm256_castsi256_ps( _mm256_set1_epi32(0x80000000) );
  93     __m256           one     = _mm256_set1_ps(1.0);
  94     __m256           two     = _mm256_set1_ps(2.0);
  95     x                = xx[0];
  96     f                = ff[0];
  97
  98     nri              = nlist->nri;
  99     iinr             = nlist->iinr;
 100     jindex           = nlist->jindex;
 101     jjnr             = nlist->jjnr;
 102     shiftidx         = nlist->shift;
 103     gid              = nlist->gid;
 104     shiftvec         = fr->shift_vec[0];
 105     fshift           = fr->fshift[0];
 106     facel            = _mm256_set1_ps(fr->epsfac);
 107     charge           = mdatoms->chargeA;
 108     krf              = _mm256_set1_ps(fr->ic->k_rf);
 109     krf2             = _mm256_set1_ps(fr->ic->k_rf*2.0);
 110     crf              = _mm256_set1_ps(fr->ic->c_rf);
 111     nvdwtype         = fr->ntype;
 112     vdwparam         = fr->nbfp;
 113     vdwtype          = mdatoms->typeA;
 114
 115     /* Setup water-specific parameters */
 116     inr              = nlist->iinr[0];
 117     iq0              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+0]));
 118     iq1              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+1]));
 119     iq2              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+2]));
 120     vdwioffsetptr0   = vdwparam+2*nvdwtype*vdwtype[inr+0];
 121
 122     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 123     rcutoff_scalar   = fr->rcoulomb;
 124     rcutoff          = _mm256_set1_ps(rcutoff_scalar);
 125     rcutoff2         = _mm256_mul_ps(rcutoff,rcutoff);
 126
 127     sh_vdw_invrcut6  = _mm256_set1_ps(fr->ic->sh_invrc6);
 128     rvdw             = _mm256_set1_ps(fr->rvdw);
 129
 130     /* Avoid stupid compiler warnings */
 131     jnrA = jnrB = jnrC = jnrD = jnrE = jnrF = jnrG = jnrH = 0;
 132     j_coord_offsetA = 0;
 133     j_coord_offsetB = 0;
 134     j_coord_offsetC = 0;
 135     j_coord_offsetD = 0;
 136     j_coord_offsetE = 0;
 137     j_coord_offsetF = 0;
 138     j_coord_offsetG = 0;
 139     j_coord_offsetH = 0;
 140
 141     outeriter        = 0;
 142     inneriter        = 0;
 143
 144     for(iidx=0;iidx<4*DIM;iidx++)
 145     {
 146         scratch[iidx] = 0.0;
 147     }
 148
 149     /* Start outer loop over neighborlists */
 150     for(iidx=0; iidx<nri; iidx++)
 151     {
 152         /* Load shift vector for this list */
 153         i_shift_offset   = DIM*shiftidx[iidx];
 154
 155         /* Load limits for loop over neighbors */
 156         j_index_start    = jindex[iidx];
 157         j_index_end      = jindex[iidx+1];
 158
 159         /* Get outer coordinate index */
 160         inr              = iinr[iidx];
 161         i_coord_offset   = DIM*inr;
 162
 163         /* Load i particle coords and add shift vector */
 164         gmx_mm256_load_shift_and_3rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
 165                                                     &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 166
 167         fix0             = _mm256_setzero_ps();
 168         fiy0             = _mm256_setzero_ps();
 169         fiz0             = _mm256_setzero_ps();
 170         fix1             = _mm256_setzero_ps();
 171         fiy1             = _mm256_setzero_ps();
 172         fiz1             = _mm256_setzero_ps();
 173         fix2             = _mm256_setzero_ps();
 174         fiy2             = _mm256_setzero_ps();
 175         fiz2             = _mm256_setzero_ps();
 176
 177         /* Reset potential sums */
 178         velecsum         = _mm256_setzero_ps();
 179         vvdwsum          = _mm256_setzero_ps();
 180
 181         /* Start inner kernel loop */
 182         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+7]>=0; jidx+=8)
 183         {
 184
 185             /* Get j neighbor index, and coordinate index */
 186             jnrA             = jjnr[jidx];
 187             jnrB             = jjnr[jidx+1];
 188             jnrC             = jjnr[jidx+2];
 189             jnrD             = jjnr[jidx+3];
 190             jnrE             = jjnr[jidx+4];
 191             jnrF             = jjnr[jidx+5];
 192             jnrG             = jjnr[jidx+6];
 193             jnrH             = jjnr[jidx+7];
 194             j_coord_offsetA  = DIM*jnrA;
 195             j_coord_offsetB  = DIM*jnrB;
 196             j_coord_offsetC  = DIM*jnrC;
 197             j_coord_offsetD  = DIM*jnrD;
 198             j_coord_offsetE  = DIM*jnrE;
 199             j_coord_offsetF  = DIM*jnrF;
 200             j_coord_offsetG  = DIM*jnrG;
 201             j_coord_offsetH  = DIM*jnrH;
 202
 203             /* load j atom coordinates */
 204             gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 205                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 206                                                  x+j_coord_offsetE,x+j_coord_offsetF,
 207                                                  x+j_coord_offsetG,x+j_coord_offsetH,
 208                                                  &jx0,&jy0,&jz0);
 209
 210             /* Calculate displacement vector */
 211             dx00             = _mm256_sub_ps(ix0,jx0);
 212             dy00             = _mm256_sub_ps(iy0,jy0);
 213             dz00             = _mm256_sub_ps(iz0,jz0);
 214             dx10             = _mm256_sub_ps(ix1,jx0);
 215             dy10             = _mm256_sub_ps(iy1,jy0);
 216             dz10             = _mm256_sub_ps(iz1,jz0);
 217             dx20             = _mm256_sub_ps(ix2,jx0);
 218             dy20             = _mm256_sub_ps(iy2,jy0);
 219             dz20             = _mm256_sub_ps(iz2,jz0);
 220
 221             /* Calculate squared distance and things based on it */
 222             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
 223             rsq10            = gmx_mm256_calc_rsq_ps(dx10,dy10,dz10);
 224             rsq20            = gmx_mm256_calc_rsq_ps(dx20,dy20,dz20);
 225
 226             rinv00           = gmx_mm256_invsqrt_ps(rsq00);
 227             rinv10           = gmx_mm256_invsqrt_ps(rsq10);
 228             rinv20           = gmx_mm256_invsqrt_ps(rsq20);
 229
 230             rinvsq00         = _mm256_mul_ps(rinv00,rinv00);
 231             rinvsq10         = _mm256_mul_ps(rinv10,rinv10);
 232             rinvsq20         = _mm256_mul_ps(rinv20,rinv20);
 233
 234             /* Load parameters for j particles */
 235             jq0              = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 236                                                                  charge+jnrC+0,charge+jnrD+0,
 237                                                                  charge+jnrE+0,charge+jnrF+0,
 238                                                                  charge+jnrG+0,charge+jnrH+0);
 239             vdwjidx0A        = 2*vdwtype[jnrA+0];
 240             vdwjidx0B        = 2*vdwtype[jnrB+0];
 241             vdwjidx0C        = 2*vdwtype[jnrC+0];
 242             vdwjidx0D        = 2*vdwtype[jnrD+0];
 243             vdwjidx0E        = 2*vdwtype[jnrE+0];
 244             vdwjidx0F        = 2*vdwtype[jnrF+0];
 245             vdwjidx0G        = 2*vdwtype[jnrG+0];
 246             vdwjidx0H        = 2*vdwtype[jnrH+0];
 247
 248             fjx0             = _mm256_setzero_ps();
 249             fjy0             = _mm256_setzero_ps();
 250             fjz0             = _mm256_setzero_ps();
 251
 252             /**************************
 253              * CALCULATE INTERACTIONS *
 254              **************************/
 255
 256             if (gmx_mm256_any_lt(rsq00,rcutoff2))
 257             {
 258
 259             /* Compute parameters for interactions between i and j atoms */
 260             qq00             = _mm256_mul_ps(iq0,jq0);
 261             gmx_mm256_load_8pair_swizzle_ps(vdwioffsetptr0+vdwjidx0A,
 262                                             vdwioffsetptr0+vdwjidx0B,
 263                                             vdwioffsetptr0+vdwjidx0C,
 264                                             vdwioffsetptr0+vdwjidx0D,
 265                                             vdwioffsetptr0+vdwjidx0E,
 266                                             vdwioffsetptr0+vdwjidx0F,
 267                                             vdwioffsetptr0+vdwjidx0G,
 268                                             vdwioffsetptr0+vdwjidx0H,
 269                                             &c6_00,&c12_00);
 270
 271             /* REACTION-FIELD ELECTROSTATICS */
 272             velec            = _mm256_mul_ps(qq00,_mm256_sub_ps(_mm256_add_ps(rinv00,_mm256_mul_ps(krf,rsq00)),crf));
 273             felec            = _mm256_mul_ps(qq00,_mm256_sub_ps(_mm256_mul_ps(rinv00,rinvsq00),krf2));
 274
 275             /* LENNARD-JONES DISPERSION/REPULSION */
 276
 277             rinvsix          = _mm256_mul_ps(_mm256_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 278             vvdw6            = _mm256_mul_ps(c6_00,rinvsix);
 279             vvdw12           = _mm256_mul_ps(c12_00,_mm256_mul_ps(rinvsix,rinvsix));
 280             vvdw             = _mm256_sub_ps(_mm256_mul_ps( _mm256_sub_ps(vvdw12 , _mm256_mul_ps(c12_00,_mm256_mul_ps(sh_vdw_invrcut6,sh_vdw_invrcut6))), one_twelfth) ,
 281                                           _mm256_mul_ps( _mm256_sub_ps(vvdw6,_mm256_mul_ps(c6_00,sh_vdw_invrcut6)),one_sixth));
 282             fvdw             = _mm256_mul_ps(_mm256_sub_ps(vvdw12,vvdw6),rinvsq00);
 283
 284             cutoff_mask      = _mm256_cmp_ps(rsq00,rcutoff2,_CMP_LT_OQ);
 285
 286             /* Update potential sum for this i atom from the interaction with this j atom. */
 287             velec            = _mm256_and_ps(velec,cutoff_mask);
 288             velecsum         = _mm256_add_ps(velecsum,velec);
 289             vvdw             = _mm256_and_ps(vvdw,cutoff_mask);
 290             vvdwsum          = _mm256_add_ps(vvdwsum,vvdw);
 291
 292             fscal            = _mm256_add_ps(felec,fvdw);
 293
 294             fscal            = _mm256_and_ps(fscal,cutoff_mask);
 295
 296             /* Calculate temporary vectorial force */
 297             tx               = _mm256_mul_ps(fscal,dx00);
 298             ty               = _mm256_mul_ps(fscal,dy00);
 299             tz               = _mm256_mul_ps(fscal,dz00);
 300
 301             /* Update vectorial force */
 302             fix0             = _mm256_add_ps(fix0,tx);
 303             fiy0             = _mm256_add_ps(fiy0,ty);
 304             fiz0             = _mm256_add_ps(fiz0,tz);
 305
 306             fjx0             = _mm256_add_ps(fjx0,tx);
 307             fjy0             = _mm256_add_ps(fjy0,ty);
 308             fjz0             = _mm256_add_ps(fjz0,tz);
 309
 310             }
 311
 312             /**************************
 313              * CALCULATE INTERACTIONS *
 314              **************************/
 315
 316             if (gmx_mm256_any_lt(rsq10,rcutoff2))
 317             {
 318
 319             /* Compute parameters for interactions between i and j atoms */
 320             qq10             = _mm256_mul_ps(iq1,jq0);
 321
 322             /* REACTION-FIELD ELECTROSTATICS */
 323             velec            = _mm256_mul_ps(qq10,_mm256_sub_ps(_mm256_add_ps(rinv10,_mm256_mul_ps(krf,rsq10)),crf));
 324             felec            = _mm256_mul_ps(qq10,_mm256_sub_ps(_mm256_mul_ps(rinv10,rinvsq10),krf2));
 325
 326             cutoff_mask      = _mm256_cmp_ps(rsq10,rcutoff2,_CMP_LT_OQ);
 327
 328             /* Update potential sum for this i atom from the interaction with this j atom. */
 329             velec            = _mm256_and_ps(velec,cutoff_mask);
 330             velecsum         = _mm256_add_ps(velecsum,velec);
 331
 332             fscal            = felec;
 333
 334             fscal            = _mm256_and_ps(fscal,cutoff_mask);
 335
 336             /* Calculate temporary vectorial force */
 337             tx               = _mm256_mul_ps(fscal,dx10);
 338             ty               = _mm256_mul_ps(fscal,dy10);
 339             tz               = _mm256_mul_ps(fscal,dz10);
 340
 341             /* Update vectorial force */
 342             fix1             = _mm256_add_ps(fix1,tx);
 343             fiy1             = _mm256_add_ps(fiy1,ty);
 344             fiz1             = _mm256_add_ps(fiz1,tz);
 345
 346             fjx0             = _mm256_add_ps(fjx0,tx);
 347             fjy0             = _mm256_add_ps(fjy0,ty);
 348             fjz0             = _mm256_add_ps(fjz0,tz);
 349
 350             }
 351
 352             /**************************
 353              * CALCULATE INTERACTIONS *
 354              **************************/
 355
 356             if (gmx_mm256_any_lt(rsq20,rcutoff2))
 357             {
 358
 359             /* Compute parameters for interactions between i and j atoms */
 360             qq20             = _mm256_mul_ps(iq2,jq0);
 361
 362             /* REACTION-FIELD ELECTROSTATICS */
 363             velec            = _mm256_mul_ps(qq20,_mm256_sub_ps(_mm256_add_ps(rinv20,_mm256_mul_ps(krf,rsq20)),crf));
 364             felec            = _mm256_mul_ps(qq20,_mm256_sub_ps(_mm256_mul_ps(rinv20,rinvsq20),krf2));
 365
 366             cutoff_mask      = _mm256_cmp_ps(rsq20,rcutoff2,_CMP_LT_OQ);
 367
 368             /* Update potential sum for this i atom from the interaction with this j atom. */
 369             velec            = _mm256_and_ps(velec,cutoff_mask);
 370             velecsum         = _mm256_add_ps(velecsum,velec);
 371
 372             fscal            = felec;
 373
 374             fscal            = _mm256_and_ps(fscal,cutoff_mask);
 375
 376             /* Calculate temporary vectorial force */
 377             tx               = _mm256_mul_ps(fscal,dx20);
 378             ty               = _mm256_mul_ps(fscal,dy20);
 379             tz               = _mm256_mul_ps(fscal,dz20);
 380
 381             /* Update vectorial force */
 382             fix2             = _mm256_add_ps(fix2,tx);
 383             fiy2             = _mm256_add_ps(fiy2,ty);
 384             fiz2             = _mm256_add_ps(fiz2,tz);
 385
 386             fjx0             = _mm256_add_ps(fjx0,tx);
 387             fjy0             = _mm256_add_ps(fjy0,ty);
 388             fjz0             = _mm256_add_ps(fjz0,tz);
 389
 390             }
 391
 392             fjptrA             = f+j_coord_offsetA;
 393             fjptrB             = f+j_coord_offsetB;
 394             fjptrC             = f+j_coord_offsetC;
 395             fjptrD             = f+j_coord_offsetD;
 396             fjptrE             = f+j_coord_offsetE;
 397             fjptrF             = f+j_coord_offsetF;
 398             fjptrG             = f+j_coord_offsetG;
 399             fjptrH             = f+j_coord_offsetH;
 400
 401             gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,fjx0,fjy0,fjz0);
 402
 403             /* Inner loop uses 129 flops */
 404         }
 405
 406         if(jidx<j_index_end)
 407         {
 408
 409             /* Get j neighbor index, and coordinate index */
 410             jnrlistA         = jjnr[jidx];
 411             jnrlistB         = jjnr[jidx+1];
 412             jnrlistC         = jjnr[jidx+2];
 413             jnrlistD         = jjnr[jidx+3];
 414             jnrlistE         = jjnr[jidx+4];
 415             jnrlistF         = jjnr[jidx+5];
 416             jnrlistG         = jjnr[jidx+6];
 417             jnrlistH         = jjnr[jidx+7];
 418             /* Sign of each element will be negative for non-real atoms.
 419              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 420              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 421              */
 422             dummy_mask = gmx_mm256_set_m128(gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx+4)),_mm_setzero_si128())),
 423                                             gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128())));
 424
 425             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 426             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 427             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 428             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 429             jnrE       = (jnrlistE>=0) ? jnrlistE : 0;
 430             jnrF       = (jnrlistF>=0) ? jnrlistF : 0;
 431             jnrG       = (jnrlistG>=0) ? jnrlistG : 0;
 432             jnrH       = (jnrlistH>=0) ? jnrlistH : 0;
 433             j_coord_offsetA  = DIM*jnrA;
 434             j_coord_offsetB  = DIM*jnrB;
 435             j_coord_offsetC  = DIM*jnrC;
 436             j_coord_offsetD  = DIM*jnrD;
 437             j_coord_offsetE  = DIM*jnrE;
 438             j_coord_offsetF  = DIM*jnrF;
 439             j_coord_offsetG  = DIM*jnrG;
 440             j_coord_offsetH  = DIM*jnrH;
 441
 442             /* load j atom coordinates */
 443             gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 444                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 445                                                  x+j_coord_offsetE,x+j_coord_offsetF,
 446                                                  x+j_coord_offsetG,x+j_coord_offsetH,
 447                                                  &jx0,&jy0,&jz0);
 448
 449             /* Calculate displacement vector */
 450             dx00             = _mm256_sub_ps(ix0,jx0);
 451             dy00             = _mm256_sub_ps(iy0,jy0);
 452             dz00             = _mm256_sub_ps(iz0,jz0);
 453             dx10             = _mm256_sub_ps(ix1,jx0);
 454             dy10             = _mm256_sub_ps(iy1,jy0);
 455             dz10             = _mm256_sub_ps(iz1,jz0);
 456             dx20             = _mm256_sub_ps(ix2,jx0);
 457             dy20             = _mm256_sub_ps(iy2,jy0);
 458             dz20             = _mm256_sub_ps(iz2,jz0);
 459
 460             /* Calculate squared distance and things based on it */
 461             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
 462             rsq10            = gmx_mm256_calc_rsq_ps(dx10,dy10,dz10);
 463             rsq20            = gmx_mm256_calc_rsq_ps(dx20,dy20,dz20);
 464
 465             rinv00           = gmx_mm256_invsqrt_ps(rsq00);
 466             rinv10           = gmx_mm256_invsqrt_ps(rsq10);
 467             rinv20           = gmx_mm256_invsqrt_ps(rsq20);
 468
 469             rinvsq00         = _mm256_mul_ps(rinv00,rinv00);
 470             rinvsq10         = _mm256_mul_ps(rinv10,rinv10);
 471             rinvsq20         = _mm256_mul_ps(rinv20,rinv20);
 472
 473             /* Load parameters for j particles */
 474             jq0              = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 475                                                                  charge+jnrC+0,charge+jnrD+0,
 476                                                                  charge+jnrE+0,charge+jnrF+0,
 477                                                                  charge+jnrG+0,charge+jnrH+0);
 478             vdwjidx0A        = 2*vdwtype[jnrA+0];
 479             vdwjidx0B        = 2*vdwtype[jnrB+0];
 480             vdwjidx0C        = 2*vdwtype[jnrC+0];
 481             vdwjidx0D        = 2*vdwtype[jnrD+0];
 482             vdwjidx0E        = 2*vdwtype[jnrE+0];
 483             vdwjidx0F        = 2*vdwtype[jnrF+0];
 484             vdwjidx0G        = 2*vdwtype[jnrG+0];
 485             vdwjidx0H        = 2*vdwtype[jnrH+0];
 486
 487             fjx0             = _mm256_setzero_ps();
 488             fjy0             = _mm256_setzero_ps();
 489             fjz0             = _mm256_setzero_ps();
 490
 491             /**************************
 492              * CALCULATE INTERACTIONS *
 493              **************************/
 494
 495             if (gmx_mm256_any_lt(rsq00,rcutoff2))
 496             {
 497
 498             /* Compute parameters for interactions between i and j atoms */
 499             qq00             = _mm256_mul_ps(iq0,jq0);
 500             gmx_mm256_load_8pair_swizzle_ps(vdwioffsetptr0+vdwjidx0A,
 501                                             vdwioffsetptr0+vdwjidx0B,
 502                                             vdwioffsetptr0+vdwjidx0C,
 503                                             vdwioffsetptr0+vdwjidx0D,
 504                                             vdwioffsetptr0+vdwjidx0E,
 505                                             vdwioffsetptr0+vdwjidx0F,
 506                                             vdwioffsetptr0+vdwjidx0G,
 507                                             vdwioffsetptr0+vdwjidx0H,
 508                                             &c6_00,&c12_00);
 509
 510             /* REACTION-FIELD ELECTROSTATICS */
 511             velec            = _mm256_mul_ps(qq00,_mm256_sub_ps(_mm256_add_ps(rinv00,_mm256_mul_ps(krf,rsq00)),crf));
 512             felec            = _mm256_mul_ps(qq00,_mm256_sub_ps(_mm256_mul_ps(rinv00,rinvsq00),krf2));
 513
 514             /* LENNARD-JONES DISPERSION/REPULSION */
 515
 516             rinvsix          = _mm256_mul_ps(_mm256_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 517             vvdw6            = _mm256_mul_ps(c6_00,rinvsix);
 518             vvdw12           = _mm256_mul_ps(c12_00,_mm256_mul_ps(rinvsix,rinvsix));
 519             vvdw             = _mm256_sub_ps(_mm256_mul_ps( _mm256_sub_ps(vvdw12 , _mm256_mul_ps(c12_00,_mm256_mul_ps(sh_vdw_invrcut6,sh_vdw_invrcut6))), one_twelfth) ,
 520                                           _mm256_mul_ps( _mm256_sub_ps(vvdw6,_mm256_mul_ps(c6_00,sh_vdw_invrcut6)),one_sixth));
 521             fvdw             = _mm256_mul_ps(_mm256_sub_ps(vvdw12,vvdw6),rinvsq00);
 522
 523             cutoff_mask      = _mm256_cmp_ps(rsq00,rcutoff2,_CMP_LT_OQ);
 524
 525             /* Update potential sum for this i atom from the interaction with this j atom. */
 526             velec            = _mm256_and_ps(velec,cutoff_mask);
 527             velec            = _mm256_andnot_ps(dummy_mask,velec);
 528             velecsum         = _mm256_add_ps(velecsum,velec);
 529             vvdw             = _mm256_and_ps(vvdw,cutoff_mask);
 530             vvdw             = _mm256_andnot_ps(dummy_mask,vvdw);
 531             vvdwsum          = _mm256_add_ps(vvdwsum,vvdw);
 532
 533             fscal            = _mm256_add_ps(felec,fvdw);
 534
 535             fscal            = _mm256_and_ps(fscal,cutoff_mask);
 536
 537             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
 538
 539             /* Calculate temporary vectorial force */
 540             tx               = _mm256_mul_ps(fscal,dx00);
 541             ty               = _mm256_mul_ps(fscal,dy00);
 542             tz               = _mm256_mul_ps(fscal,dz00);
 543
 544             /* Update vectorial force */
 545             fix0             = _mm256_add_ps(fix0,tx);
 546             fiy0             = _mm256_add_ps(fiy0,ty);
 547             fiz0             = _mm256_add_ps(fiz0,tz);
 548
 549             fjx0             = _mm256_add_ps(fjx0,tx);
 550             fjy0             = _mm256_add_ps(fjy0,ty);
 551             fjz0             = _mm256_add_ps(fjz0,tz);
 552
 553             }
 554
 555             /**************************
 556              * CALCULATE INTERACTIONS *
 557              **************************/
 558
 559             if (gmx_mm256_any_lt(rsq10,rcutoff2))
 560             {
 561
 562             /* Compute parameters for interactions between i and j atoms */
 563             qq10             = _mm256_mul_ps(iq1,jq0);
 564
 565             /* REACTION-FIELD ELECTROSTATICS */
 566             velec            = _mm256_mul_ps(qq10,_mm256_sub_ps(_mm256_add_ps(rinv10,_mm256_mul_ps(krf,rsq10)),crf));
 567             felec            = _mm256_mul_ps(qq10,_mm256_sub_ps(_mm256_mul_ps(rinv10,rinvsq10),krf2));
 568
 569             cutoff_mask      = _mm256_cmp_ps(rsq10,rcutoff2,_CMP_LT_OQ);
 570
 571             /* Update potential sum for this i atom from the interaction with this j atom. */
 572             velec            = _mm256_and_ps(velec,cutoff_mask);
 573             velec            = _mm256_andnot_ps(dummy_mask,velec);
 574             velecsum         = _mm256_add_ps(velecsum,velec);
 575
 576             fscal            = felec;
 577
 578             fscal            = _mm256_and_ps(fscal,cutoff_mask);
 579
 580             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
 581
 582             /* Calculate temporary vectorial force */
 583             tx               = _mm256_mul_ps(fscal,dx10);
 584             ty               = _mm256_mul_ps(fscal,dy10);
 585             tz               = _mm256_mul_ps(fscal,dz10);
 586
 587             /* Update vectorial force */
 588             fix1             = _mm256_add_ps(fix1,tx);
 589             fiy1             = _mm256_add_ps(fiy1,ty);
 590             fiz1             = _mm256_add_ps(fiz1,tz);
 591
 592             fjx0             = _mm256_add_ps(fjx0,tx);
 593             fjy0             = _mm256_add_ps(fjy0,ty);
 594             fjz0             = _mm256_add_ps(fjz0,tz);
 595
 596             }
 597
 598             /**************************
 599              * CALCULATE INTERACTIONS *
 600              **************************/
 601
 602             if (gmx_mm256_any_lt(rsq20,rcutoff2))
 603             {
 604
 605             /* Compute parameters for interactions between i and j atoms */
 606             qq20             = _mm256_mul_ps(iq2,jq0);
 607
 608             /* REACTION-FIELD ELECTROSTATICS */
 609             velec            = _mm256_mul_ps(qq20,_mm256_sub_ps(_mm256_add_ps(rinv20,_mm256_mul_ps(krf,rsq20)),crf));
 610             felec            = _mm256_mul_ps(qq20,_mm256_sub_ps(_mm256_mul_ps(rinv20,rinvsq20),krf2));
 611
 612             cutoff_mask      = _mm256_cmp_ps(rsq20,rcutoff2,_CMP_LT_OQ);
 613
 614             /* Update potential sum for this i atom from the interaction with this j atom. */
 615             velec            = _mm256_and_ps(velec,cutoff_mask);
 616             velec            = _mm256_andnot_ps(dummy_mask,velec);
 617             velecsum         = _mm256_add_ps(velecsum,velec);
 618
 619             fscal            = felec;
 620
 621             fscal            = _mm256_and_ps(fscal,cutoff_mask);
 622
 623             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
 624
 625             /* Calculate temporary vectorial force */
 626             tx               = _mm256_mul_ps(fscal,dx20);
 627             ty               = _mm256_mul_ps(fscal,dy20);
 628             tz               = _mm256_mul_ps(fscal,dz20);
 629
 630             /* Update vectorial force */
 631             fix2             = _mm256_add_ps(fix2,tx);
 632             fiy2             = _mm256_add_ps(fiy2,ty);
 633             fiz2             = _mm256_add_ps(fiz2,tz);
 634
 635             fjx0             = _mm256_add_ps(fjx0,tx);
 636             fjy0             = _mm256_add_ps(fjy0,ty);
 637             fjz0             = _mm256_add_ps(fjz0,tz);
 638
 639             }
 640
 641             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 642             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 643             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 644             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 645             fjptrE             = (jnrlistE>=0) ? f+j_coord_offsetE : scratch;
 646             fjptrF             = (jnrlistF>=0) ? f+j_coord_offsetF : scratch;
 647             fjptrG             = (jnrlistG>=0) ? f+j_coord_offsetG : scratch;
 648             fjptrH             = (jnrlistH>=0) ? f+j_coord_offsetH : scratch;
 649
 650             gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,fjx0,fjy0,fjz0);
 651
 652             /* Inner loop uses 129 flops */
 653         }
 654
 655         /* End of innermost loop */
 656
 657         gmx_mm256_update_iforce_3atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
 658                                                  f+i_coord_offset,fshift+i_shift_offset);
 659
 660         ggid                        = gid[iidx];
 661         /* Update potential energies */
 662         gmx_mm256_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid);
 663         gmx_mm256_update_1pot_ps(vvdwsum,kernel_data->energygrp_vdw+ggid);
 664
 665         /* Increment number of inner iterations */
 666         inneriter                  += j_index_end - j_index_start;
 667
 668         /* Outer loop uses 20 flops */
 669     }
 670
 671     /* Increment number of outer iterations */
 672     outeriter        += nri;
 673
 674     /* Update outer/inner flops */
 675
 676     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3_VF,outeriter*20 + inneriter*129);
 677 }
 678 /*
 679  * Gromacs nonbonded kernel:   nb_kernel_ElecRFCut_VdwLJSh_GeomW3P1_F_avx_256_single
 680  * Electrostatics interaction: ReactionField
 681  * VdW interaction:            LennardJones
 682  * Geometry:                   Water3-Particle
 683  * Calculate force/pot:        Force
 684  */
 685 void
 686 nb_kernel_ElecRFCut_VdwLJSh_GeomW3P1_F_avx_256_single
 687                     (t_nblist * gmx_restrict                nlist,
 688                      rvec * gmx_restrict                    xx,
 689                      rvec * gmx_restrict                    ff,
 690                      t_forcerec * gmx_restrict              fr,
 691                      t_mdatoms * gmx_restrict               mdatoms,
 692                      nb_kernel_data_t * gmx_restrict        kernel_data,
 693                      t_nrnb * gmx_restrict                  nrnb)
 694 {
 695     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 696      * just 0 for non-waters.
 697      * Suffixes A,B,C,D,E,F,G,H refer to j loop unrolling done with AVX, e.g. for the eight different
 698      * jnr indices corresponding to data put in the four positions in the SIMD register.
 699      */
 700     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 701     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 702     int              jnrA,jnrB,jnrC,jnrD;
 703     int              jnrE,jnrF,jnrG,jnrH;
 704     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
 705     int              jnrlistE,jnrlistF,jnrlistG,jnrlistH;
 706     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 707     int              j_coord_offsetE,j_coord_offsetF,j_coord_offsetG,j_coord_offsetH;
 708     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 709     real             rcutoff_scalar;
 710     real             *shiftvec,*fshift,*x,*f;
 711     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD,*fjptrE,*fjptrF,*fjptrG,*fjptrH;
 712     real             scratch[4*DIM];
 713     __m256           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 714     real *           vdwioffsetptr0;
 715     __m256           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 716     real *           vdwioffsetptr1;
 717     __m256           ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
 718     real *           vdwioffsetptr2;
 719     __m256           ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
 720     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D,vdwjidx0E,vdwjidx0F,vdwjidx0G,vdwjidx0H;
 721     __m256           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 722     __m256           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 723     __m256           dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
 724     __m256           dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
 725     __m256           velec,felec,velecsum,facel,crf,krf,krf2;
 726     real             *charge;
 727     int              nvdwtype;
 728     __m256           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 729     int              *vdwtype;
 730     real             *vdwparam;
 731     __m256           one_sixth   = _mm256_set1_ps(1.0/6.0);
 732     __m256           one_twelfth = _mm256_set1_ps(1.0/12.0);
 733     __m256           dummy_mask,cutoff_mask;
 734     __m256           signbit = _mm256_castsi256_ps( _mm256_set1_epi32(0x80000000) );
 735     __m256           one     = _mm256_set1_ps(1.0);
 736     __m256           two     = _mm256_set1_ps(2.0);
 737     x                = xx[0];
 738     f                = ff[0];
 739
 740     nri              = nlist->nri;
 741     iinr             = nlist->iinr;
 742     jindex           = nlist->jindex;
 743     jjnr             = nlist->jjnr;
 744     shiftidx         = nlist->shift;
 745     gid              = nlist->gid;
 746     shiftvec         = fr->shift_vec[0];
 747     fshift           = fr->fshift[0];
 748     facel            = _mm256_set1_ps(fr->epsfac);
 749     charge           = mdatoms->chargeA;
 750     krf              = _mm256_set1_ps(fr->ic->k_rf);
 751     krf2             = _mm256_set1_ps(fr->ic->k_rf*2.0);
 752     crf              = _mm256_set1_ps(fr->ic->c_rf);
 753     nvdwtype         = fr->ntype;
 754     vdwparam         = fr->nbfp;
 755     vdwtype          = mdatoms->typeA;
 756
 757     /* Setup water-specific parameters */
 758     inr              = nlist->iinr[0];
 759     iq0              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+0]));
 760     iq1              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+1]));
 761     iq2              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+2]));
 762     vdwioffsetptr0   = vdwparam+2*nvdwtype*vdwtype[inr+0];
 763
 764     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 765     rcutoff_scalar   = fr->rcoulomb;
 766     rcutoff          = _mm256_set1_ps(rcutoff_scalar);
 767     rcutoff2         = _mm256_mul_ps(rcutoff,rcutoff);
 768
 769     sh_vdw_invrcut6  = _mm256_set1_ps(fr->ic->sh_invrc6);
 770     rvdw             = _mm256_set1_ps(fr->rvdw);
 771
 772     /* Avoid stupid compiler warnings */
 773     jnrA = jnrB = jnrC = jnrD = jnrE = jnrF = jnrG = jnrH = 0;
 774     j_coord_offsetA = 0;
 775     j_coord_offsetB = 0;
 776     j_coord_offsetC = 0;
 777     j_coord_offsetD = 0;
 778     j_coord_offsetE = 0;
 779     j_coord_offsetF = 0;
 780     j_coord_offsetG = 0;
 781     j_coord_offsetH = 0;
 782
 783     outeriter        = 0;
 784     inneriter        = 0;
 785
 786     for(iidx=0;iidx<4*DIM;iidx++)
 787     {
 788         scratch[iidx] = 0.0;
 789     }
 790
 791     /* Start outer loop over neighborlists */
 792     for(iidx=0; iidx<nri; iidx++)
 793     {
 794         /* Load shift vector for this list */
 795         i_shift_offset   = DIM*shiftidx[iidx];
 796
 797         /* Load limits for loop over neighbors */
 798         j_index_start    = jindex[iidx];
 799         j_index_end      = jindex[iidx+1];
 800
 801         /* Get outer coordinate index */
 802         inr              = iinr[iidx];
 803         i_coord_offset   = DIM*inr;
 804
 805         /* Load i particle coords and add shift vector */
 806         gmx_mm256_load_shift_and_3rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
 807                                                     &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 808
 809         fix0             = _mm256_setzero_ps();
 810         fiy0             = _mm256_setzero_ps();
 811         fiz0             = _mm256_setzero_ps();
 812         fix1             = _mm256_setzero_ps();
 813         fiy1             = _mm256_setzero_ps();
 814         fiz1             = _mm256_setzero_ps();
 815         fix2             = _mm256_setzero_ps();
 816         fiy2             = _mm256_setzero_ps();
 817         fiz2             = _mm256_setzero_ps();
 818
 819         /* Start inner kernel loop */
 820         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+7]>=0; jidx+=8)
 821         {
 822
 823             /* Get j neighbor index, and coordinate index */
 824             jnrA             = jjnr[jidx];
 825             jnrB             = jjnr[jidx+1];
 826             jnrC             = jjnr[jidx+2];
 827             jnrD             = jjnr[jidx+3];
 828             jnrE             = jjnr[jidx+4];
 829             jnrF             = jjnr[jidx+5];
 830             jnrG             = jjnr[jidx+6];
 831             jnrH             = jjnr[jidx+7];
 832             j_coord_offsetA  = DIM*jnrA;
 833             j_coord_offsetB  = DIM*jnrB;
 834             j_coord_offsetC  = DIM*jnrC;
 835             j_coord_offsetD  = DIM*jnrD;
 836             j_coord_offsetE  = DIM*jnrE;
 837             j_coord_offsetF  = DIM*jnrF;
 838             j_coord_offsetG  = DIM*jnrG;
 839             j_coord_offsetH  = DIM*jnrH;
 840
 841             /* load j atom coordinates */
 842             gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 843                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 844                                                  x+j_coord_offsetE,x+j_coord_offsetF,
 845                                                  x+j_coord_offsetG,x+j_coord_offsetH,
 846                                                  &jx0,&jy0,&jz0);
 847
 848             /* Calculate displacement vector */
 849             dx00             = _mm256_sub_ps(ix0,jx0);
 850             dy00             = _mm256_sub_ps(iy0,jy0);
 851             dz00             = _mm256_sub_ps(iz0,jz0);
 852             dx10             = _mm256_sub_ps(ix1,jx0);
 853             dy10             = _mm256_sub_ps(iy1,jy0);
 854             dz10             = _mm256_sub_ps(iz1,jz0);
 855             dx20             = _mm256_sub_ps(ix2,jx0);
 856             dy20             = _mm256_sub_ps(iy2,jy0);
 857             dz20             = _mm256_sub_ps(iz2,jz0);
 858
 859             /* Calculate squared distance and things based on it */
 860             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
 861             rsq10            = gmx_mm256_calc_rsq_ps(dx10,dy10,dz10);
 862             rsq20            = gmx_mm256_calc_rsq_ps(dx20,dy20,dz20);
 863
 864             rinv00           = gmx_mm256_invsqrt_ps(rsq00);
 865             rinv10           = gmx_mm256_invsqrt_ps(rsq10);
 866             rinv20           = gmx_mm256_invsqrt_ps(rsq20);
 867
 868             rinvsq00         = _mm256_mul_ps(rinv00,rinv00);
 869             rinvsq10         = _mm256_mul_ps(rinv10,rinv10);
 870             rinvsq20         = _mm256_mul_ps(rinv20,rinv20);
 871
 872             /* Load parameters for j particles */
 873             jq0              = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 874                                                                  charge+jnrC+0,charge+jnrD+0,
 875                                                                  charge+jnrE+0,charge+jnrF+0,
 876                                                                  charge+jnrG+0,charge+jnrH+0);
 877             vdwjidx0A        = 2*vdwtype[jnrA+0];
 878             vdwjidx0B        = 2*vdwtype[jnrB+0];
 879             vdwjidx0C        = 2*vdwtype[jnrC+0];
 880             vdwjidx0D        = 2*vdwtype[jnrD+0];
 881             vdwjidx0E        = 2*vdwtype[jnrE+0];
 882             vdwjidx0F        = 2*vdwtype[jnrF+0];
 883             vdwjidx0G        = 2*vdwtype[jnrG+0];
 884             vdwjidx0H        = 2*vdwtype[jnrH+0];
 885
 886             fjx0             = _mm256_setzero_ps();
 887             fjy0             = _mm256_setzero_ps();
 888             fjz0             = _mm256_setzero_ps();
 889
 890             /**************************
 891              * CALCULATE INTERACTIONS *
 892              **************************/
 893
 894             if (gmx_mm256_any_lt(rsq00,rcutoff2))
 895             {
 896
 897             /* Compute parameters for interactions between i and j atoms */
 898             qq00             = _mm256_mul_ps(iq0,jq0);
 899             gmx_mm256_load_8pair_swizzle_ps(vdwioffsetptr0+vdwjidx0A,
 900                                             vdwioffsetptr0+vdwjidx0B,
 901                                             vdwioffsetptr0+vdwjidx0C,
 902                                             vdwioffsetptr0+vdwjidx0D,
 903                                             vdwioffsetptr0+vdwjidx0E,
 904                                             vdwioffsetptr0+vdwjidx0F,
 905                                             vdwioffsetptr0+vdwjidx0G,
 906                                             vdwioffsetptr0+vdwjidx0H,
 907                                             &c6_00,&c12_00);
 908
 909             /* REACTION-FIELD ELECTROSTATICS */
 910             felec            = _mm256_mul_ps(qq00,_mm256_sub_ps(_mm256_mul_ps(rinv00,rinvsq00),krf2));
 911
 912             /* LENNARD-JONES DISPERSION/REPULSION */
 913
 914             rinvsix          = _mm256_mul_ps(_mm256_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 915             fvdw             = _mm256_mul_ps(_mm256_sub_ps(_mm256_mul_ps(c12_00,rinvsix),c6_00),_mm256_mul_ps(rinvsix,rinvsq00));
 916
 917             cutoff_mask      = _mm256_cmp_ps(rsq00,rcutoff2,_CMP_LT_OQ);
 918
 919             fscal            = _mm256_add_ps(felec,fvdw);
 920
 921             fscal            = _mm256_and_ps(fscal,cutoff_mask);
 922
 923             /* Calculate temporary vectorial force */
 924             tx               = _mm256_mul_ps(fscal,dx00);
 925             ty               = _mm256_mul_ps(fscal,dy00);
 926             tz               = _mm256_mul_ps(fscal,dz00);
 927
 928             /* Update vectorial force */
 929             fix0             = _mm256_add_ps(fix0,tx);
 930             fiy0             = _mm256_add_ps(fiy0,ty);
 931             fiz0             = _mm256_add_ps(fiz0,tz);
 932
 933             fjx0             = _mm256_add_ps(fjx0,tx);
 934             fjy0             = _mm256_add_ps(fjy0,ty);
 935             fjz0             = _mm256_add_ps(fjz0,tz);
 936
 937             }
 938
 939             /**************************
 940              * CALCULATE INTERACTIONS *
 941              **************************/
 942
 943             if (gmx_mm256_any_lt(rsq10,rcutoff2))
 944             {
 945
 946             /* Compute parameters for interactions between i and j atoms */
 947             qq10             = _mm256_mul_ps(iq1,jq0);
 948
 949             /* REACTION-FIELD ELECTROSTATICS */
 950             felec            = _mm256_mul_ps(qq10,_mm256_sub_ps(_mm256_mul_ps(rinv10,rinvsq10),krf2));
 951
 952             cutoff_mask      = _mm256_cmp_ps(rsq10,rcutoff2,_CMP_LT_OQ);
 953
 954             fscal            = felec;
 955
 956             fscal            = _mm256_and_ps(fscal,cutoff_mask);
 957
 958             /* Calculate temporary vectorial force */
 959             tx               = _mm256_mul_ps(fscal,dx10);
 960             ty               = _mm256_mul_ps(fscal,dy10);
 961             tz               = _mm256_mul_ps(fscal,dz10);
 962
 963             /* Update vectorial force */
 964             fix1             = _mm256_add_ps(fix1,tx);
 965             fiy1             = _mm256_add_ps(fiy1,ty);
 966             fiz1             = _mm256_add_ps(fiz1,tz);
 967
 968             fjx0             = _mm256_add_ps(fjx0,tx);
 969             fjy0             = _mm256_add_ps(fjy0,ty);
 970             fjz0             = _mm256_add_ps(fjz0,tz);
 971
 972             }
 973
 974             /**************************
 975              * CALCULATE INTERACTIONS *
 976              **************************/
 977
 978             if (gmx_mm256_any_lt(rsq20,rcutoff2))
 979             {
 980
 981             /* Compute parameters for interactions between i and j atoms */
 982             qq20             = _mm256_mul_ps(iq2,jq0);
 983
 984             /* REACTION-FIELD ELECTROSTATICS */
 985             felec            = _mm256_mul_ps(qq20,_mm256_sub_ps(_mm256_mul_ps(rinv20,rinvsq20),krf2));
 986
 987             cutoff_mask      = _mm256_cmp_ps(rsq20,rcutoff2,_CMP_LT_OQ);
 988
 989             fscal            = felec;
 990
 991             fscal            = _mm256_and_ps(fscal,cutoff_mask);
 992
 993             /* Calculate temporary vectorial force */
 994             tx               = _mm256_mul_ps(fscal,dx20);
 995             ty               = _mm256_mul_ps(fscal,dy20);
 996             tz               = _mm256_mul_ps(fscal,dz20);
 997
 998             /* Update vectorial force */
 999             fix2             = _mm256_add_ps(fix2,tx);
1000             fiy2             = _mm256_add_ps(fiy2,ty);
1001             fiz2             = _mm256_add_ps(fiz2,tz);
1002
1003             fjx0             = _mm256_add_ps(fjx0,tx);
1004             fjy0             = _mm256_add_ps(fjy0,ty);
1005             fjz0             = _mm256_add_ps(fjz0,tz);
1006
1007             }
1008
1009             fjptrA             = f+j_coord_offsetA;
1010             fjptrB             = f+j_coord_offsetB;
1011             fjptrC             = f+j_coord_offsetC;
1012             fjptrD             = f+j_coord_offsetD;
1013             fjptrE             = f+j_coord_offsetE;
1014             fjptrF             = f+j_coord_offsetF;
1015             fjptrG             = f+j_coord_offsetG;
1016             fjptrH             = f+j_coord_offsetH;
1017
1018             gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,fjx0,fjy0,fjz0);
1019
1020             /* Inner loop uses 100 flops */
1021         }
1022
1023         if(jidx<j_index_end)
1024         {
1025
1026             /* Get j neighbor index, and coordinate index */
1027             jnrlistA         = jjnr[jidx];
1028             jnrlistB         = jjnr[jidx+1];
1029             jnrlistC         = jjnr[jidx+2];
1030             jnrlistD         = jjnr[jidx+3];
1031             jnrlistE         = jjnr[jidx+4];
1032             jnrlistF         = jjnr[jidx+5];
1033             jnrlistG         = jjnr[jidx+6];
1034             jnrlistH         = jjnr[jidx+7];
1035             /* Sign of each element will be negative for non-real atoms.
1036              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
1037              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
1038              */
1039             dummy_mask = gmx_mm256_set_m128(gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx+4)),_mm_setzero_si128())),
1040                                             gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128())));
1041
1042             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
1043             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
1044             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
1045             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
1046             jnrE       = (jnrlistE>=0) ? jnrlistE : 0;
1047             jnrF       = (jnrlistF>=0) ? jnrlistF : 0;
1048             jnrG       = (jnrlistG>=0) ? jnrlistG : 0;
1049             jnrH       = (jnrlistH>=0) ? jnrlistH : 0;
1050             j_coord_offsetA  = DIM*jnrA;
1051             j_coord_offsetB  = DIM*jnrB;
1052             j_coord_offsetC  = DIM*jnrC;
1053             j_coord_offsetD  = DIM*jnrD;
1054             j_coord_offsetE  = DIM*jnrE;
1055             j_coord_offsetF  = DIM*jnrF;
1056             j_coord_offsetG  = DIM*jnrG;
1057             j_coord_offsetH  = DIM*jnrH;
1058
1059             /* load j atom coordinates */
1060             gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
1061                                                  x+j_coord_offsetC,x+j_coord_offsetD,
1062                                                  x+j_coord_offsetE,x+j_coord_offsetF,
1063                                                  x+j_coord_offsetG,x+j_coord_offsetH,
1064                                                  &jx0,&jy0,&jz0);
1065
1066             /* Calculate displacement vector */
1067             dx00             = _mm256_sub_ps(ix0,jx0);
1068             dy00             = _mm256_sub_ps(iy0,jy0);
1069             dz00             = _mm256_sub_ps(iz0,jz0);
1070             dx10             = _mm256_sub_ps(ix1,jx0);
1071             dy10             = _mm256_sub_ps(iy1,jy0);
1072             dz10             = _mm256_sub_ps(iz1,jz0);
1073             dx20             = _mm256_sub_ps(ix2,jx0);
1074             dy20             = _mm256_sub_ps(iy2,jy0);
1075             dz20             = _mm256_sub_ps(iz2,jz0);
1076
1077             /* Calculate squared distance and things based on it */
1078             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
1079             rsq10            = gmx_mm256_calc_rsq_ps(dx10,dy10,dz10);
1080             rsq20            = gmx_mm256_calc_rsq_ps(dx20,dy20,dz20);
1081
1082             rinv00           = gmx_mm256_invsqrt_ps(rsq00);
1083             rinv10           = gmx_mm256_invsqrt_ps(rsq10);
1084             rinv20           = gmx_mm256_invsqrt_ps(rsq20);
1085
1086             rinvsq00         = _mm256_mul_ps(rinv00,rinv00);
1087             rinvsq10         = _mm256_mul_ps(rinv10,rinv10);
1088             rinvsq20         = _mm256_mul_ps(rinv20,rinv20);
1089
1090             /* Load parameters for j particles */
1091             jq0              = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
1092                                                                  charge+jnrC+0,charge+jnrD+0,
1093                                                                  charge+jnrE+0,charge+jnrF+0,
1094                                                                  charge+jnrG+0,charge+jnrH+0);
1095             vdwjidx0A        = 2*vdwtype[jnrA+0];
1096             vdwjidx0B        = 2*vdwtype[jnrB+0];
1097             vdwjidx0C        = 2*vdwtype[jnrC+0];
1098             vdwjidx0D        = 2*vdwtype[jnrD+0];
1099             vdwjidx0E        = 2*vdwtype[jnrE+0];
1100             vdwjidx0F        = 2*vdwtype[jnrF+0];
1101             vdwjidx0G        = 2*vdwtype[jnrG+0];
1102             vdwjidx0H        = 2*vdwtype[jnrH+0];
1103
1104             fjx0             = _mm256_setzero_ps();
1105             fjy0             = _mm256_setzero_ps();
1106             fjz0             = _mm256_setzero_ps();
1107
1108             /**************************
1109              * CALCULATE INTERACTIONS *
1110              **************************/
1111
1112             if (gmx_mm256_any_lt(rsq00,rcutoff2))
1113             {
1114
1115             /* Compute parameters for interactions between i and j atoms */
1116             qq00             = _mm256_mul_ps(iq0,jq0);
1117             gmx_mm256_load_8pair_swizzle_ps(vdwioffsetptr0+vdwjidx0A,
1118                                             vdwioffsetptr0+vdwjidx0B,
1119                                             vdwioffsetptr0+vdwjidx0C,
1120                                             vdwioffsetptr0+vdwjidx0D,
1121                                             vdwioffsetptr0+vdwjidx0E,
1122                                             vdwioffsetptr0+vdwjidx0F,
1123                                             vdwioffsetptr0+vdwjidx0G,
1124                                             vdwioffsetptr0+vdwjidx0H,
1125                                             &c6_00,&c12_00);
1126
1127             /* REACTION-FIELD ELECTROSTATICS */
1128             felec            = _mm256_mul_ps(qq00,_mm256_sub_ps(_mm256_mul_ps(rinv00,rinvsq00),krf2));
1129
1130             /* LENNARD-JONES DISPERSION/REPULSION */
1131
1132             rinvsix          = _mm256_mul_ps(_mm256_mul_ps(rinvsq00,rinvsq00),rinvsq00);
1133             fvdw             = _mm256_mul_ps(_mm256_sub_ps(_mm256_mul_ps(c12_00,rinvsix),c6_00),_mm256_mul_ps(rinvsix,rinvsq00));
1134
1135             cutoff_mask      = _mm256_cmp_ps(rsq00,rcutoff2,_CMP_LT_OQ);
1136
1137             fscal            = _mm256_add_ps(felec,fvdw);
1138
1139             fscal            = _mm256_and_ps(fscal,cutoff_mask);
1140
1141             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
1142
1143             /* Calculate temporary vectorial force */
1144             tx               = _mm256_mul_ps(fscal,dx00);
1145             ty               = _mm256_mul_ps(fscal,dy00);
1146             tz               = _mm256_mul_ps(fscal,dz00);
1147
1148             /* Update vectorial force */
1149             fix0             = _mm256_add_ps(fix0,tx);
1150             fiy0             = _mm256_add_ps(fiy0,ty);
1151             fiz0             = _mm256_add_ps(fiz0,tz);
1152
1153             fjx0             = _mm256_add_ps(fjx0,tx);
1154             fjy0             = _mm256_add_ps(fjy0,ty);
1155             fjz0             = _mm256_add_ps(fjz0,tz);
1156
1157             }
1158
1159             /**************************
1160              * CALCULATE INTERACTIONS *
1161              **************************/
1162
1163             if (gmx_mm256_any_lt(rsq10,rcutoff2))
1164             {
1165
1166             /* Compute parameters for interactions between i and j atoms */
1167             qq10             = _mm256_mul_ps(iq1,jq0);
1168
1169             /* REACTION-FIELD ELECTROSTATICS */
1170             felec            = _mm256_mul_ps(qq10,_mm256_sub_ps(_mm256_mul_ps(rinv10,rinvsq10),krf2));
1171
1172             cutoff_mask      = _mm256_cmp_ps(rsq10,rcutoff2,_CMP_LT_OQ);
1173
1174             fscal            = felec;
1175
1176             fscal            = _mm256_and_ps(fscal,cutoff_mask);
1177
1178             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
1179
1180             /* Calculate temporary vectorial force */
1181             tx               = _mm256_mul_ps(fscal,dx10);
1182             ty               = _mm256_mul_ps(fscal,dy10);
1183             tz               = _mm256_mul_ps(fscal,dz10);
1184
1185             /* Update vectorial force */
1186             fix1             = _mm256_add_ps(fix1,tx);
1187             fiy1             = _mm256_add_ps(fiy1,ty);
1188             fiz1             = _mm256_add_ps(fiz1,tz);
1189
1190             fjx0             = _mm256_add_ps(fjx0,tx);
1191             fjy0             = _mm256_add_ps(fjy0,ty);
1192             fjz0             = _mm256_add_ps(fjz0,tz);
1193
1194             }
1195
1196             /**************************
1197              * CALCULATE INTERACTIONS *
1198              **************************/
1199
1200             if (gmx_mm256_any_lt(rsq20,rcutoff2))
1201             {
1202
1203             /* Compute parameters for interactions between i and j atoms */
1204             qq20             = _mm256_mul_ps(iq2,jq0);
1205
1206             /* REACTION-FIELD ELECTROSTATICS */
1207             felec            = _mm256_mul_ps(qq20,_mm256_sub_ps(_mm256_mul_ps(rinv20,rinvsq20),krf2));
1208
1209             cutoff_mask      = _mm256_cmp_ps(rsq20,rcutoff2,_CMP_LT_OQ);
1210
1211             fscal            = felec;
1212
1213             fscal            = _mm256_and_ps(fscal,cutoff_mask);
1214
1215             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
1216
1217             /* Calculate temporary vectorial force */
1218             tx               = _mm256_mul_ps(fscal,dx20);
1219             ty               = _mm256_mul_ps(fscal,dy20);
1220             tz               = _mm256_mul_ps(fscal,dz20);
1221
1222             /* Update vectorial force */
1223             fix2             = _mm256_add_ps(fix2,tx);
1224             fiy2             = _mm256_add_ps(fiy2,ty);
1225             fiz2             = _mm256_add_ps(fiz2,tz);
1226
1227             fjx0             = _mm256_add_ps(fjx0,tx);
1228             fjy0             = _mm256_add_ps(fjy0,ty);
1229             fjz0             = _mm256_add_ps(fjz0,tz);
1230
1231             }
1232
1233             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
1234             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
1235             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
1236             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
1237             fjptrE             = (jnrlistE>=0) ? f+j_coord_offsetE : scratch;
1238             fjptrF             = (jnrlistF>=0) ? f+j_coord_offsetF : scratch;
1239             fjptrG             = (jnrlistG>=0) ? f+j_coord_offsetG : scratch;
1240             fjptrH             = (jnrlistH>=0) ? f+j_coord_offsetH : scratch;
1241
1242             gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,fjx0,fjy0,fjz0);
1243
1244             /* Inner loop uses 100 flops */
1245         }
1246
1247         /* End of innermost loop */
1248
1249         gmx_mm256_update_iforce_3atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
1250                                                  f+i_coord_offset,fshift+i_shift_offset);
1251
1252         /* Increment number of inner iterations */
1253         inneriter                  += j_index_end - j_index_start;
1254
1255         /* Outer loop uses 18 flops */
1256     }
1257
1258     /* Increment number of outer iterations */
1259     outeriter        += nri;
1260
1261     /* Update outer/inner flops */
1262
1263     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3_F,outeriter*18 + inneriter*100);
1264 }