src/gromacs/gmxlib/nonbonded/nb_kernel_avx_256_single/nb_kernel_ElecGB_VdwNone_GeomP1P1_avx_256_single.c

   1 /*
   2  * This file is part of the GROMACS molecular simulation package.
   3  *
   4  * Copyright (c) 2012,2013,2014,2015,2017, by the GROMACS development team, led by
   5  * Mark Abraham, David van der Spoel, Berk Hess, and Erik Lindahl,
   6  * and including many others, as listed in the AUTHORS file in the
   7  * top-level source directory and at http://www.gromacs.org.
   8  *
   9  * GROMACS is free software; you can redistribute it and/or
  10  * modify it under the terms of the GNU Lesser General Public License
  11  * as published by the Free Software Foundation; either version 2.1
  12  * of the License, or (at your option) any later version.
  13  *
  14  * GROMACS is distributed in the hope that it will be useful,
  15  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  16  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  17  * Lesser General Public License for more details.
  18  *
  19  * You should have received a copy of the GNU Lesser General Public
  20  * License along with GROMACS; if not, see
  21  * http://www.gnu.org/licenses, or write to the Free Software Foundation,
  22  * Inc., 51 Franklin Street, Fifth Floor, Boston, MA  02110-1301  USA.
  23  *
  24  * If you want to redistribute modifications to GROMACS, please
  25  * consider that scientific software is very special. Version
  26  * control is crucial - bugs must be traceable. We will be happy to
  27  * consider code for inclusion in the official distribution, but
  28  * derived work must not be called official GROMACS. Details are found
  29  * in the README & COPYING files - if they are missing, get the
  30  * official version at http://www.gromacs.org.
  31  *
  32  * To help us fund GROMACS development, we humbly ask that you cite
  33  * the research papers on the package. Check out http://www.gromacs.org.
  34  */
  35 /*
  36  * Note: this file was generated by the GROMACS avx_256_single kernel generator.
  37  */
  38 #include "gmxpre.h"
  39
  40 #include "config.h"
  41
  42 #include <math.h>
  43
  44 #include "../nb_kernel.h"
  45 #include "gromacs/gmxlib/nrnb.h"
  46
  47 #include "kernelutil_x86_avx_256_single.h"
  48
  49 /*
  50  * Gromacs nonbonded kernel:   nb_kernel_ElecGB_VdwNone_GeomP1P1_VF_avx_256_single
  51  * Electrostatics interaction: GeneralizedBorn
  52  * VdW interaction:            None
  53  * Geometry:                   Particle-Particle
  54  * Calculate force/pot:        PotentialAndForce
  55  */
  56 void
  57 nb_kernel_ElecGB_VdwNone_GeomP1P1_VF_avx_256_single
  58                     (t_nblist                    * gmx_restrict       nlist,
  59                      rvec                        * gmx_restrict          xx,
  60                      rvec                        * gmx_restrict          ff,
  61                      struct t_forcerec           * gmx_restrict          fr,
  62                      t_mdatoms                   * gmx_restrict     mdatoms,
  63                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
  64                      t_nrnb                      * gmx_restrict        nrnb)
  65 {
  66     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
  67      * just 0 for non-waters.
  68      * Suffixes A,B,C,D,E,F,G,H refer to j loop unrolling done with AVX, e.g. for the eight different
  69      * jnr indices corresponding to data put in the four positions in the SIMD register.
  70      */
  71     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
  72     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
  73     int              jnrA,jnrB,jnrC,jnrD;
  74     int              jnrE,jnrF,jnrG,jnrH;
  75     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
  76     int              jnrlistE,jnrlistF,jnrlistG,jnrlistH;
  77     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
  78     int              j_coord_offsetE,j_coord_offsetF,j_coord_offsetG,j_coord_offsetH;
  79     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
  80     real             rcutoff_scalar;
  81     real             *shiftvec,*fshift,*x,*f;
  82     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD,*fjptrE,*fjptrF,*fjptrG,*fjptrH;
  83     real             scratch[4*DIM];
  84     __m256           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
  85     real *           vdwioffsetptr0;
  86     __m256           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
  87     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D,vdwjidx0E,vdwjidx0F,vdwjidx0G,vdwjidx0H;
  88     __m256           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  89     __m256           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  90     __m256           velec,felec,velecsum,facel,crf,krf,krf2;
  91     real             *charge;
  92     __m256i          gbitab;
  93     __m128i          gbitab_lo,gbitab_hi;
  94     __m256           vgb,fgb,vgbsum,dvdasum,gbscale,gbtabscale,isaprod,gbqqfactor,gbinvepsdiff,gbeps,dvdatmp;
  95     __m256           minushalf = _mm256_set1_ps(-0.5);
  96     real             *invsqrta,*dvda,*gbtab;
  97     __m256i          vfitab;
  98     __m128i          vfitab_lo,vfitab_hi;
  99     __m128i          ifour       = _mm_set1_epi32(4);
 100     __m256           rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF;
 101     real             *vftab;
 102     __m256           dummy_mask,cutoff_mask;
 103     __m256           signbit = _mm256_castsi256_ps( _mm256_set1_epi32(0x80000000) );
 104     __m256           one     = _mm256_set1_ps(1.0);
 105     __m256           two     = _mm256_set1_ps(2.0);
 106     x                = xx[0];
 107     f                = ff[0];
 108
 109     nri              = nlist->nri;
 110     iinr             = nlist->iinr;
 111     jindex           = nlist->jindex;
 112     jjnr             = nlist->jjnr;
 113     shiftidx         = nlist->shift;
 114     gid              = nlist->gid;
 115     shiftvec         = fr->shift_vec[0];
 116     fshift           = fr->fshift[0];
 117     facel            = _mm256_set1_ps(fr->ic->epsfac);
 118     charge           = mdatoms->chargeA;
 119
 120     invsqrta         = fr->invsqrta;
 121     dvda             = fr->dvda;
 122     gbtabscale       = _mm256_set1_ps(fr->gbtab->scale);
 123     gbtab            = fr->gbtab->data;
 124     gbinvepsdiff     = _mm256_set1_ps((1.0/fr->ic->epsilon_r) - (1.0/fr->gb_epsilon_solvent));
 125
 126     /* Avoid stupid compiler warnings */
 127     jnrA = jnrB = jnrC = jnrD = jnrE = jnrF = jnrG = jnrH = 0;
 128     j_coord_offsetA = 0;
 129     j_coord_offsetB = 0;
 130     j_coord_offsetC = 0;
 131     j_coord_offsetD = 0;
 132     j_coord_offsetE = 0;
 133     j_coord_offsetF = 0;
 134     j_coord_offsetG = 0;
 135     j_coord_offsetH = 0;
 136
 137     outeriter        = 0;
 138     inneriter        = 0;
 139
 140     for(iidx=0;iidx<4*DIM;iidx++)
 141     {
 142         scratch[iidx] = 0.0;
 143     }
 144
 145     /* Start outer loop over neighborlists */
 146     for(iidx=0; iidx<nri; iidx++)
 147     {
 148         /* Load shift vector for this list */
 149         i_shift_offset   = DIM*shiftidx[iidx];
 150
 151         /* Load limits for loop over neighbors */
 152         j_index_start    = jindex[iidx];
 153         j_index_end      = jindex[iidx+1];
 154
 155         /* Get outer coordinate index */
 156         inr              = iinr[iidx];
 157         i_coord_offset   = DIM*inr;
 158
 159         /* Load i particle coords and add shift vector */
 160         gmx_mm256_load_shift_and_1rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 161
 162         fix0             = _mm256_setzero_ps();
 163         fiy0             = _mm256_setzero_ps();
 164         fiz0             = _mm256_setzero_ps();
 165
 166         /* Load parameters for i particles */
 167         iq0              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+0]));
 168         isai0            = _mm256_set1_ps(invsqrta[inr+0]);
 169
 170         /* Reset potential sums */
 171         velecsum         = _mm256_setzero_ps();
 172         vgbsum           = _mm256_setzero_ps();
 173         dvdasum          = _mm256_setzero_ps();
 174
 175         /* Start inner kernel loop */
 176         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+7]>=0; jidx+=8)
 177         {
 178
 179             /* Get j neighbor index, and coordinate index */
 180             jnrA             = jjnr[jidx];
 181             jnrB             = jjnr[jidx+1];
 182             jnrC             = jjnr[jidx+2];
 183             jnrD             = jjnr[jidx+3];
 184             jnrE             = jjnr[jidx+4];
 185             jnrF             = jjnr[jidx+5];
 186             jnrG             = jjnr[jidx+6];
 187             jnrH             = jjnr[jidx+7];
 188             j_coord_offsetA  = DIM*jnrA;
 189             j_coord_offsetB  = DIM*jnrB;
 190             j_coord_offsetC  = DIM*jnrC;
 191             j_coord_offsetD  = DIM*jnrD;
 192             j_coord_offsetE  = DIM*jnrE;
 193             j_coord_offsetF  = DIM*jnrF;
 194             j_coord_offsetG  = DIM*jnrG;
 195             j_coord_offsetH  = DIM*jnrH;
 196
 197             /* load j atom coordinates */
 198             gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 199                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 200                                                  x+j_coord_offsetE,x+j_coord_offsetF,
 201                                                  x+j_coord_offsetG,x+j_coord_offsetH,
 202                                                  &jx0,&jy0,&jz0);
 203
 204             /* Calculate displacement vector */
 205             dx00             = _mm256_sub_ps(ix0,jx0);
 206             dy00             = _mm256_sub_ps(iy0,jy0);
 207             dz00             = _mm256_sub_ps(iz0,jz0);
 208
 209             /* Calculate squared distance and things based on it */
 210             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
 211
 212             rinv00           = avx256_invsqrt_f(rsq00);
 213
 214             /* Load parameters for j particles */
 215             jq0              = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 216                                                                  charge+jnrC+0,charge+jnrD+0,
 217                                                                  charge+jnrE+0,charge+jnrF+0,
 218                                                                  charge+jnrG+0,charge+jnrH+0);
 219             isaj0            = gmx_mm256_load_8real_swizzle_ps(invsqrta+jnrA+0,invsqrta+jnrB+0,
 220                                                                  invsqrta+jnrC+0,invsqrta+jnrD+0,
 221                                                                  invsqrta+jnrE+0,invsqrta+jnrF+0,
 222                                                                  invsqrta+jnrG+0,invsqrta+jnrH+0);
 223
 224             /**************************
 225              * CALCULATE INTERACTIONS *
 226              **************************/
 227
 228             r00              = _mm256_mul_ps(rsq00,rinv00);
 229
 230             /* Compute parameters for interactions between i and j atoms */
 231             qq00             = _mm256_mul_ps(iq0,jq0);
 232
 233             /* GENERALIZED BORN AND COULOMB ELECTROSTATICS */
 234             isaprod          = _mm256_mul_ps(isai0,isaj0);
 235             gbqqfactor       = _mm256_xor_ps(signbit,_mm256_mul_ps(qq00,_mm256_mul_ps(isaprod,gbinvepsdiff)));
 236             gbscale          = _mm256_mul_ps(isaprod,gbtabscale);
 237
 238             /* Calculate generalized born table index - this is a separate table from the normal one,
 239              * but we use the same procedure by multiplying r with scale and truncating to integer.
 240              */
 241             rt               = _mm256_mul_ps(r00,gbscale);
 242             gbitab           = _mm256_cvttps_epi32(rt);
 243             gbeps            = _mm256_sub_ps(rt,_mm256_round_ps(rt, _MM_FROUND_FLOOR));
 244             /*         AVX1 does not support 256-bit integer operations, so now we go to 128-bit mode... */
 245             gbitab_lo        = _mm256_extractf128_si256(gbitab,0x0);
 246             gbitab_hi        = _mm256_extractf128_si256(gbitab,0x1);
 247             gbitab_lo        = _mm_slli_epi32(gbitab_lo,2);
 248             gbitab_hi        = _mm_slli_epi32(gbitab_hi,2);
 249             Y                = gmx_mm256_set_m128(_mm_load_ps(gbtab + _mm_extract_epi32(gbitab_hi,0)),
 250                                                   _mm_load_ps(gbtab + _mm_extract_epi32(gbitab_lo,0)));
 251             F                = gmx_mm256_set_m128(_mm_load_ps(gbtab + _mm_extract_epi32(gbitab_hi,1)),
 252                                                   _mm_load_ps(gbtab + _mm_extract_epi32(gbitab_lo,1)));
 253             G                = gmx_mm256_set_m128(_mm_load_ps(gbtab + _mm_extract_epi32(gbitab_hi,2)),
 254                                                   _mm_load_ps(gbtab + _mm_extract_epi32(gbitab_lo,2)));
 255             H                = gmx_mm256_set_m128(_mm_load_ps(gbtab + _mm_extract_epi32(gbitab_hi,3)),
 256                                                   _mm_load_ps(gbtab + _mm_extract_epi32(gbitab_lo,3)));
 257             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
 258             Heps             = _mm256_mul_ps(gbeps,H);
 259             Fp               = _mm256_add_ps(F,_mm256_mul_ps(gbeps,_mm256_add_ps(G,Heps)));
 260             VV               = _mm256_add_ps(Y,_mm256_mul_ps(gbeps,Fp));
 261             vgb              = _mm256_mul_ps(gbqqfactor,VV);
 262
 263             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(gbeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
 264             fgb              = _mm256_mul_ps(gbqqfactor,_mm256_mul_ps(FF,gbscale));
 265             dvdatmp          = _mm256_mul_ps(minushalf,_mm256_add_ps(vgb,_mm256_mul_ps(fgb,r00)));
 266             dvdasum          = _mm256_add_ps(dvdasum,dvdatmp);
 267             fjptrA           = dvda+jnrA;
 268             fjptrB           = dvda+jnrB;
 269             fjptrC           = dvda+jnrC;
 270             fjptrD           = dvda+jnrD;
 271             fjptrE           = dvda+jnrE;
 272             fjptrF           = dvda+jnrF;
 273             fjptrG           = dvda+jnrG;
 274             fjptrH           = dvda+jnrH;
 275             gmx_mm256_increment_8real_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,
 276                                                  _mm256_mul_ps(dvdatmp,_mm256_mul_ps(isaj0,isaj0)));
 277             velec            = _mm256_mul_ps(qq00,rinv00);
 278             felec            = _mm256_mul_ps(_mm256_sub_ps(_mm256_mul_ps(velec,rinv00),fgb),rinv00);
 279
 280             /* Update potential sum for this i atom from the interaction with this j atom. */
 281             velecsum         = _mm256_add_ps(velecsum,velec);
 282             vgbsum           = _mm256_add_ps(vgbsum,vgb);
 283
 284             fscal            = felec;
 285
 286             /* Calculate temporary vectorial force */
 287             tx               = _mm256_mul_ps(fscal,dx00);
 288             ty               = _mm256_mul_ps(fscal,dy00);
 289             tz               = _mm256_mul_ps(fscal,dz00);
 290
 291             /* Update vectorial force */
 292             fix0             = _mm256_add_ps(fix0,tx);
 293             fiy0             = _mm256_add_ps(fiy0,ty);
 294             fiz0             = _mm256_add_ps(fiz0,tz);
 295
 296             fjptrA             = f+j_coord_offsetA;
 297             fjptrB             = f+j_coord_offsetB;
 298             fjptrC             = f+j_coord_offsetC;
 299             fjptrD             = f+j_coord_offsetD;
 300             fjptrE             = f+j_coord_offsetE;
 301             fjptrF             = f+j_coord_offsetF;
 302             fjptrG             = f+j_coord_offsetG;
 303             fjptrH             = f+j_coord_offsetH;
 304             gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,tx,ty,tz);
 305
 306             /* Inner loop uses 57 flops */
 307         }
 308
 309         if(jidx<j_index_end)
 310         {
 311
 312             /* Get j neighbor index, and coordinate index */
 313             jnrlistA         = jjnr[jidx];
 314             jnrlistB         = jjnr[jidx+1];
 315             jnrlistC         = jjnr[jidx+2];
 316             jnrlistD         = jjnr[jidx+3];
 317             jnrlistE         = jjnr[jidx+4];
 318             jnrlistF         = jjnr[jidx+5];
 319             jnrlistG         = jjnr[jidx+6];
 320             jnrlistH         = jjnr[jidx+7];
 321             /* Sign of each element will be negative for non-real atoms.
 322              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 323              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 324              */
 325             dummy_mask = gmx_mm256_set_m128(gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx+4)),_mm_setzero_si128())),
 326                                             gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128())));
 327
 328             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 329             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 330             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 331             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 332             jnrE       = (jnrlistE>=0) ? jnrlistE : 0;
 333             jnrF       = (jnrlistF>=0) ? jnrlistF : 0;
 334             jnrG       = (jnrlistG>=0) ? jnrlistG : 0;
 335             jnrH       = (jnrlistH>=0) ? jnrlistH : 0;
 336             j_coord_offsetA  = DIM*jnrA;
 337             j_coord_offsetB  = DIM*jnrB;
 338             j_coord_offsetC  = DIM*jnrC;
 339             j_coord_offsetD  = DIM*jnrD;
 340             j_coord_offsetE  = DIM*jnrE;
 341             j_coord_offsetF  = DIM*jnrF;
 342             j_coord_offsetG  = DIM*jnrG;
 343             j_coord_offsetH  = DIM*jnrH;
 344
 345             /* load j atom coordinates */
 346             gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 347                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 348                                                  x+j_coord_offsetE,x+j_coord_offsetF,
 349                                                  x+j_coord_offsetG,x+j_coord_offsetH,
 350                                                  &jx0,&jy0,&jz0);
 351
 352             /* Calculate displacement vector */
 353             dx00             = _mm256_sub_ps(ix0,jx0);
 354             dy00             = _mm256_sub_ps(iy0,jy0);
 355             dz00             = _mm256_sub_ps(iz0,jz0);
 356
 357             /* Calculate squared distance and things based on it */
 358             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
 359
 360             rinv00           = avx256_invsqrt_f(rsq00);
 361
 362             /* Load parameters for j particles */
 363             jq0              = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 364                                                                  charge+jnrC+0,charge+jnrD+0,
 365                                                                  charge+jnrE+0,charge+jnrF+0,
 366                                                                  charge+jnrG+0,charge+jnrH+0);
 367             isaj0            = gmx_mm256_load_8real_swizzle_ps(invsqrta+jnrA+0,invsqrta+jnrB+0,
 368                                                                  invsqrta+jnrC+0,invsqrta+jnrD+0,
 369                                                                  invsqrta+jnrE+0,invsqrta+jnrF+0,
 370                                                                  invsqrta+jnrG+0,invsqrta+jnrH+0);
 371
 372             /**************************
 373              * CALCULATE INTERACTIONS *
 374              **************************/
 375
 376             r00              = _mm256_mul_ps(rsq00,rinv00);
 377             r00              = _mm256_andnot_ps(dummy_mask,r00);
 378
 379             /* Compute parameters for interactions between i and j atoms */
 380             qq00             = _mm256_mul_ps(iq0,jq0);
 381
 382             /* GENERALIZED BORN AND COULOMB ELECTROSTATICS */
 383             isaprod          = _mm256_mul_ps(isai0,isaj0);
 384             gbqqfactor       = _mm256_xor_ps(signbit,_mm256_mul_ps(qq00,_mm256_mul_ps(isaprod,gbinvepsdiff)));
 385             gbscale          = _mm256_mul_ps(isaprod,gbtabscale);
 386
 387             /* Calculate generalized born table index - this is a separate table from the normal one,
 388              * but we use the same procedure by multiplying r with scale and truncating to integer.
 389              */
 390             rt               = _mm256_mul_ps(r00,gbscale);
 391             gbitab           = _mm256_cvttps_epi32(rt);
 392             gbeps            = _mm256_sub_ps(rt,_mm256_round_ps(rt, _MM_FROUND_FLOOR));
 393             /*         AVX1 does not support 256-bit integer operations, so now we go to 128-bit mode... */
 394             gbitab_lo        = _mm256_extractf128_si256(gbitab,0x0);
 395             gbitab_hi        = _mm256_extractf128_si256(gbitab,0x1);
 396             gbitab_lo        = _mm_slli_epi32(gbitab_lo,2);
 397             gbitab_hi        = _mm_slli_epi32(gbitab_hi,2);
 398             Y                = gmx_mm256_set_m128(_mm_load_ps(gbtab + _mm_extract_epi32(gbitab_hi,0)),
 399                                                   _mm_load_ps(gbtab + _mm_extract_epi32(gbitab_lo,0)));
 400             F                = gmx_mm256_set_m128(_mm_load_ps(gbtab + _mm_extract_epi32(gbitab_hi,1)),
 401                                                   _mm_load_ps(gbtab + _mm_extract_epi32(gbitab_lo,1)));
 402             G                = gmx_mm256_set_m128(_mm_load_ps(gbtab + _mm_extract_epi32(gbitab_hi,2)),
 403                                                   _mm_load_ps(gbtab + _mm_extract_epi32(gbitab_lo,2)));
 404             H                = gmx_mm256_set_m128(_mm_load_ps(gbtab + _mm_extract_epi32(gbitab_hi,3)),
 405                                                   _mm_load_ps(gbtab + _mm_extract_epi32(gbitab_lo,3)));
 406             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
 407             Heps             = _mm256_mul_ps(gbeps,H);
 408             Fp               = _mm256_add_ps(F,_mm256_mul_ps(gbeps,_mm256_add_ps(G,Heps)));
 409             VV               = _mm256_add_ps(Y,_mm256_mul_ps(gbeps,Fp));
 410             vgb              = _mm256_mul_ps(gbqqfactor,VV);
 411
 412             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(gbeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
 413             fgb              = _mm256_mul_ps(gbqqfactor,_mm256_mul_ps(FF,gbscale));
 414             dvdatmp          = _mm256_mul_ps(minushalf,_mm256_add_ps(vgb,_mm256_mul_ps(fgb,r00)));
 415             dvdatmp          = _mm256_andnot_ps(dummy_mask,dvdatmp);
 416             dvdasum          = _mm256_add_ps(dvdasum,dvdatmp);
 417             /* The pointers to scratch make sure that this code with compilers that take gmx_restrict seriously (e.g. icc 13) really can't screw things up. */
 418             fjptrA             = (jnrlistA>=0) ? dvda+jnrA : scratch;
 419             fjptrB             = (jnrlistB>=0) ? dvda+jnrB : scratch;
 420             fjptrC             = (jnrlistC>=0) ? dvda+jnrC : scratch;
 421             fjptrD             = (jnrlistD>=0) ? dvda+jnrD : scratch;
 422             fjptrE             = (jnrlistE>=0) ? dvda+jnrE : scratch;
 423             fjptrF             = (jnrlistF>=0) ? dvda+jnrF : scratch;
 424             fjptrG             = (jnrlistG>=0) ? dvda+jnrG : scratch;
 425             fjptrH             = (jnrlistH>=0) ? dvda+jnrH : scratch;
 426             gmx_mm256_increment_8real_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,
 427                                                  _mm256_mul_ps(dvdatmp,_mm256_mul_ps(isaj0,isaj0)));
 428             velec            = _mm256_mul_ps(qq00,rinv00);
 429             felec            = _mm256_mul_ps(_mm256_sub_ps(_mm256_mul_ps(velec,rinv00),fgb),rinv00);
 430
 431             /* Update potential sum for this i atom from the interaction with this j atom. */
 432             velec            = _mm256_andnot_ps(dummy_mask,velec);
 433             velecsum         = _mm256_add_ps(velecsum,velec);
 434             vgb              = _mm256_andnot_ps(dummy_mask,vgb);
 435             vgbsum           = _mm256_add_ps(vgbsum,vgb);
 436
 437             fscal            = felec;
 438
 439             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
 440
 441             /* Calculate temporary vectorial force */
 442             tx               = _mm256_mul_ps(fscal,dx00);
 443             ty               = _mm256_mul_ps(fscal,dy00);
 444             tz               = _mm256_mul_ps(fscal,dz00);
 445
 446             /* Update vectorial force */
 447             fix0             = _mm256_add_ps(fix0,tx);
 448             fiy0             = _mm256_add_ps(fiy0,ty);
 449             fiz0             = _mm256_add_ps(fiz0,tz);
 450
 451             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 452             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 453             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 454             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 455             fjptrE             = (jnrlistE>=0) ? f+j_coord_offsetE : scratch;
 456             fjptrF             = (jnrlistF>=0) ? f+j_coord_offsetF : scratch;
 457             fjptrG             = (jnrlistG>=0) ? f+j_coord_offsetG : scratch;
 458             fjptrH             = (jnrlistH>=0) ? f+j_coord_offsetH : scratch;
 459             gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,tx,ty,tz);
 460
 461             /* Inner loop uses 58 flops */
 462         }
 463
 464         /* End of innermost loop */
 465
 466         gmx_mm256_update_iforce_1atom_swizzle_ps(fix0,fiy0,fiz0,
 467                                                  f+i_coord_offset,fshift+i_shift_offset);
 468
 469         ggid                        = gid[iidx];
 470         /* Update potential energies */
 471         gmx_mm256_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid);
 472         gmx_mm256_update_1pot_ps(vgbsum,kernel_data->energygrp_polarization+ggid);
 473         dvdasum = _mm256_mul_ps(dvdasum, _mm256_mul_ps(isai0,isai0));
 474         gmx_mm256_update_1pot_ps(dvdasum,dvda+inr);
 475
 476         /* Increment number of inner iterations */
 477         inneriter                  += j_index_end - j_index_start;
 478
 479         /* Outer loop uses 9 flops */
 480     }
 481
 482     /* Increment number of outer iterations */
 483     outeriter        += nri;
 484
 485     /* Update outer/inner flops */
 486
 487     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VF,outeriter*9 + inneriter*58);
 488 }
 489 /*
 490  * Gromacs nonbonded kernel:   nb_kernel_ElecGB_VdwNone_GeomP1P1_F_avx_256_single
 491  * Electrostatics interaction: GeneralizedBorn
 492  * VdW interaction:            None
 493  * Geometry:                   Particle-Particle
 494  * Calculate force/pot:        Force
 495  */
 496 void
 497 nb_kernel_ElecGB_VdwNone_GeomP1P1_F_avx_256_single
 498                     (t_nblist                    * gmx_restrict       nlist,
 499                      rvec                        * gmx_restrict          xx,
 500                      rvec                        * gmx_restrict          ff,
 501                      struct t_forcerec           * gmx_restrict          fr,
 502                      t_mdatoms                   * gmx_restrict     mdatoms,
 503                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 504                      t_nrnb                      * gmx_restrict        nrnb)
 505 {
 506     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 507      * just 0 for non-waters.
 508      * Suffixes A,B,C,D,E,F,G,H refer to j loop unrolling done with AVX, e.g. for the eight different
 509      * jnr indices corresponding to data put in the four positions in the SIMD register.
 510      */
 511     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 512     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 513     int              jnrA,jnrB,jnrC,jnrD;
 514     int              jnrE,jnrF,jnrG,jnrH;
 515     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
 516     int              jnrlistE,jnrlistF,jnrlistG,jnrlistH;
 517     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 518     int              j_coord_offsetE,j_coord_offsetF,j_coord_offsetG,j_coord_offsetH;
 519     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 520     real             rcutoff_scalar;
 521     real             *shiftvec,*fshift,*x,*f;
 522     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD,*fjptrE,*fjptrF,*fjptrG,*fjptrH;
 523     real             scratch[4*DIM];
 524     __m256           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 525     real *           vdwioffsetptr0;
 526     __m256           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 527     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D,vdwjidx0E,vdwjidx0F,vdwjidx0G,vdwjidx0H;
 528     __m256           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 529     __m256           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 530     __m256           velec,felec,velecsum,facel,crf,krf,krf2;
 531     real             *charge;
 532     __m256i          gbitab;
 533     __m128i          gbitab_lo,gbitab_hi;
 534     __m256           vgb,fgb,vgbsum,dvdasum,gbscale,gbtabscale,isaprod,gbqqfactor,gbinvepsdiff,gbeps,dvdatmp;
 535     __m256           minushalf = _mm256_set1_ps(-0.5);
 536     real             *invsqrta,*dvda,*gbtab;
 537     __m256i          vfitab;
 538     __m128i          vfitab_lo,vfitab_hi;
 539     __m128i          ifour       = _mm_set1_epi32(4);
 540     __m256           rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF;
 541     real             *vftab;
 542     __m256           dummy_mask,cutoff_mask;
 543     __m256           signbit = _mm256_castsi256_ps( _mm256_set1_epi32(0x80000000) );
 544     __m256           one     = _mm256_set1_ps(1.0);
 545     __m256           two     = _mm256_set1_ps(2.0);
 546     x                = xx[0];
 547     f                = ff[0];
 548
 549     nri              = nlist->nri;
 550     iinr             = nlist->iinr;
 551     jindex           = nlist->jindex;
 552     jjnr             = nlist->jjnr;
 553     shiftidx         = nlist->shift;
 554     gid              = nlist->gid;
 555     shiftvec         = fr->shift_vec[0];
 556     fshift           = fr->fshift[0];
 557     facel            = _mm256_set1_ps(fr->ic->epsfac);
 558     charge           = mdatoms->chargeA;
 559
 560     invsqrta         = fr->invsqrta;
 561     dvda             = fr->dvda;
 562     gbtabscale       = _mm256_set1_ps(fr->gbtab->scale);
 563     gbtab            = fr->gbtab->data;
 564     gbinvepsdiff     = _mm256_set1_ps((1.0/fr->ic->epsilon_r) - (1.0/fr->gb_epsilon_solvent));
 565
 566     /* Avoid stupid compiler warnings */
 567     jnrA = jnrB = jnrC = jnrD = jnrE = jnrF = jnrG = jnrH = 0;
 568     j_coord_offsetA = 0;
 569     j_coord_offsetB = 0;
 570     j_coord_offsetC = 0;
 571     j_coord_offsetD = 0;
 572     j_coord_offsetE = 0;
 573     j_coord_offsetF = 0;
 574     j_coord_offsetG = 0;
 575     j_coord_offsetH = 0;
 576
 577     outeriter        = 0;
 578     inneriter        = 0;
 579
 580     for(iidx=0;iidx<4*DIM;iidx++)
 581     {
 582         scratch[iidx] = 0.0;
 583     }
 584
 585     /* Start outer loop over neighborlists */
 586     for(iidx=0; iidx<nri; iidx++)
 587     {
 588         /* Load shift vector for this list */
 589         i_shift_offset   = DIM*shiftidx[iidx];
 590
 591         /* Load limits for loop over neighbors */
 592         j_index_start    = jindex[iidx];
 593         j_index_end      = jindex[iidx+1];
 594
 595         /* Get outer coordinate index */
 596         inr              = iinr[iidx];
 597         i_coord_offset   = DIM*inr;
 598
 599         /* Load i particle coords and add shift vector */
 600         gmx_mm256_load_shift_and_1rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 601
 602         fix0             = _mm256_setzero_ps();
 603         fiy0             = _mm256_setzero_ps();
 604         fiz0             = _mm256_setzero_ps();
 605
 606         /* Load parameters for i particles */
 607         iq0              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+0]));
 608         isai0            = _mm256_set1_ps(invsqrta[inr+0]);
 609
 610         dvdasum          = _mm256_setzero_ps();
 611
 612         /* Start inner kernel loop */
 613         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+7]>=0; jidx+=8)
 614         {
 615
 616             /* Get j neighbor index, and coordinate index */
 617             jnrA             = jjnr[jidx];
 618             jnrB             = jjnr[jidx+1];
 619             jnrC             = jjnr[jidx+2];
 620             jnrD             = jjnr[jidx+3];
 621             jnrE             = jjnr[jidx+4];
 622             jnrF             = jjnr[jidx+5];
 623             jnrG             = jjnr[jidx+6];
 624             jnrH             = jjnr[jidx+7];
 625             j_coord_offsetA  = DIM*jnrA;
 626             j_coord_offsetB  = DIM*jnrB;
 627             j_coord_offsetC  = DIM*jnrC;
 628             j_coord_offsetD  = DIM*jnrD;
 629             j_coord_offsetE  = DIM*jnrE;
 630             j_coord_offsetF  = DIM*jnrF;
 631             j_coord_offsetG  = DIM*jnrG;
 632             j_coord_offsetH  = DIM*jnrH;
 633
 634             /* load j atom coordinates */
 635             gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 636                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 637                                                  x+j_coord_offsetE,x+j_coord_offsetF,
 638                                                  x+j_coord_offsetG,x+j_coord_offsetH,
 639                                                  &jx0,&jy0,&jz0);
 640
 641             /* Calculate displacement vector */
 642             dx00             = _mm256_sub_ps(ix0,jx0);
 643             dy00             = _mm256_sub_ps(iy0,jy0);
 644             dz00             = _mm256_sub_ps(iz0,jz0);
 645
 646             /* Calculate squared distance and things based on it */
 647             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
 648
 649             rinv00           = avx256_invsqrt_f(rsq00);
 650
 651             /* Load parameters for j particles */
 652             jq0              = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 653                                                                  charge+jnrC+0,charge+jnrD+0,
 654                                                                  charge+jnrE+0,charge+jnrF+0,
 655                                                                  charge+jnrG+0,charge+jnrH+0);
 656             isaj0            = gmx_mm256_load_8real_swizzle_ps(invsqrta+jnrA+0,invsqrta+jnrB+0,
 657                                                                  invsqrta+jnrC+0,invsqrta+jnrD+0,
 658                                                                  invsqrta+jnrE+0,invsqrta+jnrF+0,
 659                                                                  invsqrta+jnrG+0,invsqrta+jnrH+0);
 660
 661             /**************************
 662              * CALCULATE INTERACTIONS *
 663              **************************/
 664
 665             r00              = _mm256_mul_ps(rsq00,rinv00);
 666
 667             /* Compute parameters for interactions between i and j atoms */
 668             qq00             = _mm256_mul_ps(iq0,jq0);
 669
 670             /* GENERALIZED BORN AND COULOMB ELECTROSTATICS */
 671             isaprod          = _mm256_mul_ps(isai0,isaj0);
 672             gbqqfactor       = _mm256_xor_ps(signbit,_mm256_mul_ps(qq00,_mm256_mul_ps(isaprod,gbinvepsdiff)));
 673             gbscale          = _mm256_mul_ps(isaprod,gbtabscale);
 674
 675             /* Calculate generalized born table index - this is a separate table from the normal one,
 676              * but we use the same procedure by multiplying r with scale and truncating to integer.
 677              */
 678             rt               = _mm256_mul_ps(r00,gbscale);
 679             gbitab           = _mm256_cvttps_epi32(rt);
 680             gbeps            = _mm256_sub_ps(rt,_mm256_round_ps(rt, _MM_FROUND_FLOOR));
 681             /*         AVX1 does not support 256-bit integer operations, so now we go to 128-bit mode... */
 682             gbitab_lo        = _mm256_extractf128_si256(gbitab,0x0);
 683             gbitab_hi        = _mm256_extractf128_si256(gbitab,0x1);
 684             gbitab_lo        = _mm_slli_epi32(gbitab_lo,2);
 685             gbitab_hi        = _mm_slli_epi32(gbitab_hi,2);
 686             Y                = gmx_mm256_set_m128(_mm_load_ps(gbtab + _mm_extract_epi32(gbitab_hi,0)),
 687                                                   _mm_load_ps(gbtab + _mm_extract_epi32(gbitab_lo,0)));
 688             F                = gmx_mm256_set_m128(_mm_load_ps(gbtab + _mm_extract_epi32(gbitab_hi,1)),
 689                                                   _mm_load_ps(gbtab + _mm_extract_epi32(gbitab_lo,1)));
 690             G                = gmx_mm256_set_m128(_mm_load_ps(gbtab + _mm_extract_epi32(gbitab_hi,2)),
 691                                                   _mm_load_ps(gbtab + _mm_extract_epi32(gbitab_lo,2)));
 692             H                = gmx_mm256_set_m128(_mm_load_ps(gbtab + _mm_extract_epi32(gbitab_hi,3)),
 693                                                   _mm_load_ps(gbtab + _mm_extract_epi32(gbitab_lo,3)));
 694             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
 695             Heps             = _mm256_mul_ps(gbeps,H);
 696             Fp               = _mm256_add_ps(F,_mm256_mul_ps(gbeps,_mm256_add_ps(G,Heps)));
 697             VV               = _mm256_add_ps(Y,_mm256_mul_ps(gbeps,Fp));
 698             vgb              = _mm256_mul_ps(gbqqfactor,VV);
 699
 700             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(gbeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
 701             fgb              = _mm256_mul_ps(gbqqfactor,_mm256_mul_ps(FF,gbscale));
 702             dvdatmp          = _mm256_mul_ps(minushalf,_mm256_add_ps(vgb,_mm256_mul_ps(fgb,r00)));
 703             dvdasum          = _mm256_add_ps(dvdasum,dvdatmp);
 704             fjptrA           = dvda+jnrA;
 705             fjptrB           = dvda+jnrB;
 706             fjptrC           = dvda+jnrC;
 707             fjptrD           = dvda+jnrD;
 708             fjptrE           = dvda+jnrE;
 709             fjptrF           = dvda+jnrF;
 710             fjptrG           = dvda+jnrG;
 711             fjptrH           = dvda+jnrH;
 712             gmx_mm256_increment_8real_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,
 713                                                  _mm256_mul_ps(dvdatmp,_mm256_mul_ps(isaj0,isaj0)));
 714             velec            = _mm256_mul_ps(qq00,rinv00);
 715             felec            = _mm256_mul_ps(_mm256_sub_ps(_mm256_mul_ps(velec,rinv00),fgb),rinv00);
 716
 717             fscal            = felec;
 718
 719             /* Calculate temporary vectorial force */
 720             tx               = _mm256_mul_ps(fscal,dx00);
 721             ty               = _mm256_mul_ps(fscal,dy00);
 722             tz               = _mm256_mul_ps(fscal,dz00);
 723
 724             /* Update vectorial force */
 725             fix0             = _mm256_add_ps(fix0,tx);
 726             fiy0             = _mm256_add_ps(fiy0,ty);
 727             fiz0             = _mm256_add_ps(fiz0,tz);
 728
 729             fjptrA             = f+j_coord_offsetA;
 730             fjptrB             = f+j_coord_offsetB;
 731             fjptrC             = f+j_coord_offsetC;
 732             fjptrD             = f+j_coord_offsetD;
 733             fjptrE             = f+j_coord_offsetE;
 734             fjptrF             = f+j_coord_offsetF;
 735             fjptrG             = f+j_coord_offsetG;
 736             fjptrH             = f+j_coord_offsetH;
 737             gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,tx,ty,tz);
 738
 739             /* Inner loop uses 55 flops */
 740         }
 741
 742         if(jidx<j_index_end)
 743         {
 744
 745             /* Get j neighbor index, and coordinate index */
 746             jnrlistA         = jjnr[jidx];
 747             jnrlistB         = jjnr[jidx+1];
 748             jnrlistC         = jjnr[jidx+2];
 749             jnrlistD         = jjnr[jidx+3];
 750             jnrlistE         = jjnr[jidx+4];
 751             jnrlistF         = jjnr[jidx+5];
 752             jnrlistG         = jjnr[jidx+6];
 753             jnrlistH         = jjnr[jidx+7];
 754             /* Sign of each element will be negative for non-real atoms.
 755              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 756              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 757              */
 758             dummy_mask = gmx_mm256_set_m128(gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx+4)),_mm_setzero_si128())),
 759                                             gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128())));
 760
 761             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 762             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 763             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 764             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 765             jnrE       = (jnrlistE>=0) ? jnrlistE : 0;
 766             jnrF       = (jnrlistF>=0) ? jnrlistF : 0;
 767             jnrG       = (jnrlistG>=0) ? jnrlistG : 0;
 768             jnrH       = (jnrlistH>=0) ? jnrlistH : 0;
 769             j_coord_offsetA  = DIM*jnrA;
 770             j_coord_offsetB  = DIM*jnrB;
 771             j_coord_offsetC  = DIM*jnrC;
 772             j_coord_offsetD  = DIM*jnrD;
 773             j_coord_offsetE  = DIM*jnrE;
 774             j_coord_offsetF  = DIM*jnrF;
 775             j_coord_offsetG  = DIM*jnrG;
 776             j_coord_offsetH  = DIM*jnrH;
 777
 778             /* load j atom coordinates */
 779             gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 780                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 781                                                  x+j_coord_offsetE,x+j_coord_offsetF,
 782                                                  x+j_coord_offsetG,x+j_coord_offsetH,
 783                                                  &jx0,&jy0,&jz0);
 784
 785             /* Calculate displacement vector */
 786             dx00             = _mm256_sub_ps(ix0,jx0);
 787             dy00             = _mm256_sub_ps(iy0,jy0);
 788             dz00             = _mm256_sub_ps(iz0,jz0);
 789
 790             /* Calculate squared distance and things based on it */
 791             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
 792
 793             rinv00           = avx256_invsqrt_f(rsq00);
 794
 795             /* Load parameters for j particles */
 796             jq0              = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 797                                                                  charge+jnrC+0,charge+jnrD+0,
 798                                                                  charge+jnrE+0,charge+jnrF+0,
 799                                                                  charge+jnrG+0,charge+jnrH+0);
 800             isaj0            = gmx_mm256_load_8real_swizzle_ps(invsqrta+jnrA+0,invsqrta+jnrB+0,
 801                                                                  invsqrta+jnrC+0,invsqrta+jnrD+0,
 802                                                                  invsqrta+jnrE+0,invsqrta+jnrF+0,
 803                                                                  invsqrta+jnrG+0,invsqrta+jnrH+0);
 804
 805             /**************************
 806              * CALCULATE INTERACTIONS *
 807              **************************/
 808
 809             r00              = _mm256_mul_ps(rsq00,rinv00);
 810             r00              = _mm256_andnot_ps(dummy_mask,r00);
 811
 812             /* Compute parameters for interactions between i and j atoms */
 813             qq00             = _mm256_mul_ps(iq0,jq0);
 814
 815             /* GENERALIZED BORN AND COULOMB ELECTROSTATICS */
 816             isaprod          = _mm256_mul_ps(isai0,isaj0);
 817             gbqqfactor       = _mm256_xor_ps(signbit,_mm256_mul_ps(qq00,_mm256_mul_ps(isaprod,gbinvepsdiff)));
 818             gbscale          = _mm256_mul_ps(isaprod,gbtabscale);
 819
 820             /* Calculate generalized born table index - this is a separate table from the normal one,
 821              * but we use the same procedure by multiplying r with scale and truncating to integer.
 822              */
 823             rt               = _mm256_mul_ps(r00,gbscale);
 824             gbitab           = _mm256_cvttps_epi32(rt);
 825             gbeps            = _mm256_sub_ps(rt,_mm256_round_ps(rt, _MM_FROUND_FLOOR));
 826             /*         AVX1 does not support 256-bit integer operations, so now we go to 128-bit mode... */
 827             gbitab_lo        = _mm256_extractf128_si256(gbitab,0x0);
 828             gbitab_hi        = _mm256_extractf128_si256(gbitab,0x1);
 829             gbitab_lo        = _mm_slli_epi32(gbitab_lo,2);
 830             gbitab_hi        = _mm_slli_epi32(gbitab_hi,2);
 831             Y                = gmx_mm256_set_m128(_mm_load_ps(gbtab + _mm_extract_epi32(gbitab_hi,0)),
 832                                                   _mm_load_ps(gbtab + _mm_extract_epi32(gbitab_lo,0)));
 833             F                = gmx_mm256_set_m128(_mm_load_ps(gbtab + _mm_extract_epi32(gbitab_hi,1)),
 834                                                   _mm_load_ps(gbtab + _mm_extract_epi32(gbitab_lo,1)));
 835             G                = gmx_mm256_set_m128(_mm_load_ps(gbtab + _mm_extract_epi32(gbitab_hi,2)),
 836                                                   _mm_load_ps(gbtab + _mm_extract_epi32(gbitab_lo,2)));
 837             H                = gmx_mm256_set_m128(_mm_load_ps(gbtab + _mm_extract_epi32(gbitab_hi,3)),
 838                                                   _mm_load_ps(gbtab + _mm_extract_epi32(gbitab_lo,3)));
 839             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
 840             Heps             = _mm256_mul_ps(gbeps,H);
 841             Fp               = _mm256_add_ps(F,_mm256_mul_ps(gbeps,_mm256_add_ps(G,Heps)));
 842             VV               = _mm256_add_ps(Y,_mm256_mul_ps(gbeps,Fp));
 843             vgb              = _mm256_mul_ps(gbqqfactor,VV);
 844
 845             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(gbeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
 846             fgb              = _mm256_mul_ps(gbqqfactor,_mm256_mul_ps(FF,gbscale));
 847             dvdatmp          = _mm256_mul_ps(minushalf,_mm256_add_ps(vgb,_mm256_mul_ps(fgb,r00)));
 848             dvdatmp          = _mm256_andnot_ps(dummy_mask,dvdatmp);
 849             dvdasum          = _mm256_add_ps(dvdasum,dvdatmp);
 850             /* The pointers to scratch make sure that this code with compilers that take gmx_restrict seriously (e.g. icc 13) really can't screw things up. */
 851             fjptrA             = (jnrlistA>=0) ? dvda+jnrA : scratch;
 852             fjptrB             = (jnrlistB>=0) ? dvda+jnrB : scratch;
 853             fjptrC             = (jnrlistC>=0) ? dvda+jnrC : scratch;
 854             fjptrD             = (jnrlistD>=0) ? dvda+jnrD : scratch;
 855             fjptrE             = (jnrlistE>=0) ? dvda+jnrE : scratch;
 856             fjptrF             = (jnrlistF>=0) ? dvda+jnrF : scratch;
 857             fjptrG             = (jnrlistG>=0) ? dvda+jnrG : scratch;
 858             fjptrH             = (jnrlistH>=0) ? dvda+jnrH : scratch;
 859             gmx_mm256_increment_8real_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,
 860                                                  _mm256_mul_ps(dvdatmp,_mm256_mul_ps(isaj0,isaj0)));
 861             velec            = _mm256_mul_ps(qq00,rinv00);
 862             felec            = _mm256_mul_ps(_mm256_sub_ps(_mm256_mul_ps(velec,rinv00),fgb),rinv00);
 863
 864             fscal            = felec;
 865
 866             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
 867
 868             /* Calculate temporary vectorial force */
 869             tx               = _mm256_mul_ps(fscal,dx00);
 870             ty               = _mm256_mul_ps(fscal,dy00);
 871             tz               = _mm256_mul_ps(fscal,dz00);
 872
 873             /* Update vectorial force */
 874             fix0             = _mm256_add_ps(fix0,tx);
 875             fiy0             = _mm256_add_ps(fiy0,ty);
 876             fiz0             = _mm256_add_ps(fiz0,tz);
 877
 878             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 879             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 880             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 881             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 882             fjptrE             = (jnrlistE>=0) ? f+j_coord_offsetE : scratch;
 883             fjptrF             = (jnrlistF>=0) ? f+j_coord_offsetF : scratch;
 884             fjptrG             = (jnrlistG>=0) ? f+j_coord_offsetG : scratch;
 885             fjptrH             = (jnrlistH>=0) ? f+j_coord_offsetH : scratch;
 886             gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,tx,ty,tz);
 887
 888             /* Inner loop uses 56 flops */
 889         }
 890
 891         /* End of innermost loop */
 892
 893         gmx_mm256_update_iforce_1atom_swizzle_ps(fix0,fiy0,fiz0,
 894                                                  f+i_coord_offset,fshift+i_shift_offset);
 895
 896         dvdasum = _mm256_mul_ps(dvdasum, _mm256_mul_ps(isai0,isai0));
 897         gmx_mm256_update_1pot_ps(dvdasum,dvda+inr);
 898
 899         /* Increment number of inner iterations */
 900         inneriter                  += j_index_end - j_index_start;
 901
 902         /* Outer loop uses 7 flops */
 903     }
 904
 905     /* Increment number of outer iterations */
 906     outeriter        += nri;
 907
 908     /* Update outer/inner flops */
 909
 910     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_F,outeriter*7 + inneriter*56);
 911 }