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, 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/legacyheaders/types/simple.h"
  46 #include "gromacs/math/vec.h"
  47 #include "gromacs/legacyheaders/nrnb.h"
  48
  49 #include "gromacs/simd/math_x86_avx_256_single.h"
  50 #include "kernelutil_x86_avx_256_single.h"
  51
  52 /*
  53  * Gromacs nonbonded kernel:   nb_kernel_ElecGB_VdwNone_GeomP1P1_VF_avx_256_single
  54  * Electrostatics interaction: GeneralizedBorn
  55  * VdW interaction:            None
  56  * Geometry:                   Particle-Particle
  57  * Calculate force/pot:        PotentialAndForce
  58  */
  59 void
  60 nb_kernel_ElecGB_VdwNone_GeomP1P1_VF_avx_256_single
  61                     (t_nblist                    * gmx_restrict       nlist,
  62                      rvec                        * gmx_restrict          xx,
  63                      rvec                        * gmx_restrict          ff,
  64                      t_forcerec                  * gmx_restrict          fr,
  65                      t_mdatoms                   * gmx_restrict     mdatoms,
  66                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
  67                      t_nrnb                      * gmx_restrict        nrnb)
  68 {
  69     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
  70      * just 0 for non-waters.
  71      * Suffixes A,B,C,D,E,F,G,H refer to j loop unrolling done with AVX, e.g. for the eight different
  72      * jnr indices corresponding to data put in the four positions in the SIMD register.
  73      */
  74     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
  75     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
  76     int              jnrA,jnrB,jnrC,jnrD;
  77     int              jnrE,jnrF,jnrG,jnrH;
  78     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
  79     int              jnrlistE,jnrlistF,jnrlistG,jnrlistH;
  80     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
  81     int              j_coord_offsetE,j_coord_offsetF,j_coord_offsetG,j_coord_offsetH;
  82     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
  83     real             rcutoff_scalar;
  84     real             *shiftvec,*fshift,*x,*f;
  85     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD,*fjptrE,*fjptrF,*fjptrG,*fjptrH;
  86     real             scratch[4*DIM];
  87     __m256           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
  88     real *           vdwioffsetptr0;
  89     __m256           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
  90     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D,vdwjidx0E,vdwjidx0F,vdwjidx0G,vdwjidx0H;
  91     __m256           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  92     __m256           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  93     __m256           velec,felec,velecsum,facel,crf,krf,krf2;
  94     real             *charge;
  95     __m256i          gbitab;
  96     __m128i          gbitab_lo,gbitab_hi;
  97     __m256           vgb,fgb,vgbsum,dvdasum,gbscale,gbtabscale,isaprod,gbqqfactor,gbinvepsdiff,gbeps,dvdatmp;
  98     __m256           minushalf = _mm256_set1_ps(-0.5);
  99     real             *invsqrta,*dvda,*gbtab;
 100     __m256i          vfitab;
 101     __m128i          vfitab_lo,vfitab_hi;
 102     __m128i          ifour       = _mm_set1_epi32(4);
 103     __m256           rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF;
 104     real             *vftab;
 105     __m256           dummy_mask,cutoff_mask;
 106     __m256           signbit = _mm256_castsi256_ps( _mm256_set1_epi32(0x80000000) );
 107     __m256           one     = _mm256_set1_ps(1.0);
 108     __m256           two     = _mm256_set1_ps(2.0);
 109     x                = xx[0];
 110     f                = ff[0];
 111
 112     nri              = nlist->nri;
 113     iinr             = nlist->iinr;
 114     jindex           = nlist->jindex;
 115     jjnr             = nlist->jjnr;
 116     shiftidx         = nlist->shift;
 117     gid              = nlist->gid;
 118     shiftvec         = fr->shift_vec[0];
 119     fshift           = fr->fshift[0];
 120     facel            = _mm256_set1_ps(fr->epsfac);
 121     charge           = mdatoms->chargeA;
 122
 123     invsqrta         = fr->invsqrta;
 124     dvda             = fr->dvda;
 125     gbtabscale       = _mm256_set1_ps(fr->gbtab.scale);
 126     gbtab            = fr->gbtab.data;
 127     gbinvepsdiff     = _mm256_set1_ps((1.0/fr->epsilon_r) - (1.0/fr->gb_epsilon_solvent));
 128
 129     /* Avoid stupid compiler warnings */
 130     jnrA = jnrB = jnrC = jnrD = jnrE = jnrF = jnrG = jnrH = 0;
 131     j_coord_offsetA = 0;
 132     j_coord_offsetB = 0;
 133     j_coord_offsetC = 0;
 134     j_coord_offsetD = 0;
 135     j_coord_offsetE = 0;
 136     j_coord_offsetF = 0;
 137     j_coord_offsetG = 0;
 138     j_coord_offsetH = 0;
 139
 140     outeriter        = 0;
 141     inneriter        = 0;
 142
 143     for(iidx=0;iidx<4*DIM;iidx++)
 144     {
 145         scratch[iidx] = 0.0;
 146     }
 147
 148     /* Start outer loop over neighborlists */
 149     for(iidx=0; iidx<nri; iidx++)
 150     {
 151         /* Load shift vector for this list */
 152         i_shift_offset   = DIM*shiftidx[iidx];
 153
 154         /* Load limits for loop over neighbors */
 155         j_index_start    = jindex[iidx];
 156         j_index_end      = jindex[iidx+1];
 157
 158         /* Get outer coordinate index */
 159         inr              = iinr[iidx];
 160         i_coord_offset   = DIM*inr;
 161
 162         /* Load i particle coords and add shift vector */
 163         gmx_mm256_load_shift_and_1rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 164
 165         fix0             = _mm256_setzero_ps();
 166         fiy0             = _mm256_setzero_ps();
 167         fiz0             = _mm256_setzero_ps();
 168
 169         /* Load parameters for i particles */
 170         iq0              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+0]));
 171         isai0            = _mm256_set1_ps(invsqrta[inr+0]);
 172
 173         /* Reset potential sums */
 174         velecsum         = _mm256_setzero_ps();
 175         vgbsum           = _mm256_setzero_ps();
 176         dvdasum          = _mm256_setzero_ps();
 177
 178         /* Start inner kernel loop */
 179         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+7]>=0; jidx+=8)
 180         {
 181
 182             /* Get j neighbor index, and coordinate index */
 183             jnrA             = jjnr[jidx];
 184             jnrB             = jjnr[jidx+1];
 185             jnrC             = jjnr[jidx+2];
 186             jnrD             = jjnr[jidx+3];
 187             jnrE             = jjnr[jidx+4];
 188             jnrF             = jjnr[jidx+5];
 189             jnrG             = jjnr[jidx+6];
 190             jnrH             = jjnr[jidx+7];
 191             j_coord_offsetA  = DIM*jnrA;
 192             j_coord_offsetB  = DIM*jnrB;
 193             j_coord_offsetC  = DIM*jnrC;
 194             j_coord_offsetD  = DIM*jnrD;
 195             j_coord_offsetE  = DIM*jnrE;
 196             j_coord_offsetF  = DIM*jnrF;
 197             j_coord_offsetG  = DIM*jnrG;
 198             j_coord_offsetH  = DIM*jnrH;
 199
 200             /* load j atom coordinates */
 201             gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 202                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 203                                                  x+j_coord_offsetE,x+j_coord_offsetF,
 204                                                  x+j_coord_offsetG,x+j_coord_offsetH,
 205                                                  &jx0,&jy0,&jz0);
 206
 207             /* Calculate displacement vector */
 208             dx00             = _mm256_sub_ps(ix0,jx0);
 209             dy00             = _mm256_sub_ps(iy0,jy0);
 210             dz00             = _mm256_sub_ps(iz0,jz0);
 211
 212             /* Calculate squared distance and things based on it */
 213             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
 214
 215             rinv00           = gmx_mm256_invsqrt_ps(rsq00);
 216
 217             /* Load parameters for j particles */
 218             jq0              = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 219                                                                  charge+jnrC+0,charge+jnrD+0,
 220                                                                  charge+jnrE+0,charge+jnrF+0,
 221                                                                  charge+jnrG+0,charge+jnrH+0);
 222             isaj0            = gmx_mm256_load_8real_swizzle_ps(invsqrta+jnrA+0,invsqrta+jnrB+0,
 223                                                                  invsqrta+jnrC+0,invsqrta+jnrD+0,
 224                                                                  invsqrta+jnrE+0,invsqrta+jnrF+0,
 225                                                                  invsqrta+jnrG+0,invsqrta+jnrH+0);
 226
 227             /**************************
 228              * CALCULATE INTERACTIONS *
 229              **************************/
 230
 231             r00              = _mm256_mul_ps(rsq00,rinv00);
 232
 233             /* Compute parameters for interactions between i and j atoms */
 234             qq00             = _mm256_mul_ps(iq0,jq0);
 235
 236             /* GENERALIZED BORN AND COULOMB ELECTROSTATICS */
 237             isaprod          = _mm256_mul_ps(isai0,isaj0);
 238             gbqqfactor       = _mm256_xor_ps(signbit,_mm256_mul_ps(qq00,_mm256_mul_ps(isaprod,gbinvepsdiff)));
 239             gbscale          = _mm256_mul_ps(isaprod,gbtabscale);
 240
 241             /* Calculate generalized born table index - this is a separate table from the normal one,
 242              * but we use the same procedure by multiplying r with scale and truncating to integer.
 243              */
 244             rt               = _mm256_mul_ps(r00,gbscale);
 245             gbitab           = _mm256_cvttps_epi32(rt);
 246             gbeps            = _mm256_sub_ps(rt,_mm256_round_ps(rt, _MM_FROUND_FLOOR));
 247             /*         AVX1 does not support 256-bit integer operations, so now we go to 128-bit mode... */
 248             gbitab_lo        = _mm256_extractf128_si256(gbitab,0x0);
 249             gbitab_hi        = _mm256_extractf128_si256(gbitab,0x1);
 250             gbitab_lo        = _mm_slli_epi32(gbitab_lo,2);
 251             gbitab_hi        = _mm_slli_epi32(gbitab_hi,2);
 252             Y                = gmx_mm256_set_m128(_mm_load_ps(gbtab + _mm_extract_epi32(gbitab_hi,0)),
 253                                                   _mm_load_ps(gbtab + _mm_extract_epi32(gbitab_lo,0)));
 254             F                = gmx_mm256_set_m128(_mm_load_ps(gbtab + _mm_extract_epi32(gbitab_hi,1)),
 255                                                   _mm_load_ps(gbtab + _mm_extract_epi32(gbitab_lo,1)));
 256             G                = gmx_mm256_set_m128(_mm_load_ps(gbtab + _mm_extract_epi32(gbitab_hi,2)),
 257                                                   _mm_load_ps(gbtab + _mm_extract_epi32(gbitab_lo,2)));
 258             H                = gmx_mm256_set_m128(_mm_load_ps(gbtab + _mm_extract_epi32(gbitab_hi,3)),
 259                                                   _mm_load_ps(gbtab + _mm_extract_epi32(gbitab_lo,3)));
 260             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
 261             Heps             = _mm256_mul_ps(gbeps,H);
 262             Fp               = _mm256_add_ps(F,_mm256_mul_ps(gbeps,_mm256_add_ps(G,Heps)));
 263             VV               = _mm256_add_ps(Y,_mm256_mul_ps(gbeps,Fp));
 264             vgb              = _mm256_mul_ps(gbqqfactor,VV);
 265
 266             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(gbeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
 267             fgb              = _mm256_mul_ps(gbqqfactor,_mm256_mul_ps(FF,gbscale));
 268             dvdatmp          = _mm256_mul_ps(minushalf,_mm256_add_ps(vgb,_mm256_mul_ps(fgb,r00)));
 269             dvdasum          = _mm256_add_ps(dvdasum,dvdatmp);
 270             fjptrA           = dvda+jnrA;
 271             fjptrB           = dvda+jnrB;
 272             fjptrC           = dvda+jnrC;
 273             fjptrD           = dvda+jnrD;
 274             fjptrE           = dvda+jnrE;
 275             fjptrF           = dvda+jnrF;
 276             fjptrG           = dvda+jnrG;
 277             fjptrH           = dvda+jnrH;
 278             gmx_mm256_increment_8real_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,
 279                                                  _mm256_mul_ps(dvdatmp,_mm256_mul_ps(isaj0,isaj0)));
 280             velec            = _mm256_mul_ps(qq00,rinv00);
 281             felec            = _mm256_mul_ps(_mm256_sub_ps(_mm256_mul_ps(velec,rinv00),fgb),rinv00);
 282
 283             /* Update potential sum for this i atom from the interaction with this j atom. */
 284             velecsum         = _mm256_add_ps(velecsum,velec);
 285             vgbsum           = _mm256_add_ps(vgbsum,vgb);
 286
 287             fscal            = felec;
 288
 289             /* Calculate temporary vectorial force */
 290             tx               = _mm256_mul_ps(fscal,dx00);
 291             ty               = _mm256_mul_ps(fscal,dy00);
 292             tz               = _mm256_mul_ps(fscal,dz00);
 293
 294             /* Update vectorial force */
 295             fix0             = _mm256_add_ps(fix0,tx);
 296             fiy0             = _mm256_add_ps(fiy0,ty);
 297             fiz0             = _mm256_add_ps(fiz0,tz);
 298
 299             fjptrA             = f+j_coord_offsetA;
 300             fjptrB             = f+j_coord_offsetB;
 301             fjptrC             = f+j_coord_offsetC;
 302             fjptrD             = f+j_coord_offsetD;
 303             fjptrE             = f+j_coord_offsetE;
 304             fjptrF             = f+j_coord_offsetF;
 305             fjptrG             = f+j_coord_offsetG;
 306             fjptrH             = f+j_coord_offsetH;
 307             gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,tx,ty,tz);
 308
 309             /* Inner loop uses 57 flops */
 310         }
 311
 312         if(jidx<j_index_end)
 313         {
 314
 315             /* Get j neighbor index, and coordinate index */
 316             jnrlistA         = jjnr[jidx];
 317             jnrlistB         = jjnr[jidx+1];
 318             jnrlistC         = jjnr[jidx+2];
 319             jnrlistD         = jjnr[jidx+3];
 320             jnrlistE         = jjnr[jidx+4];
 321             jnrlistF         = jjnr[jidx+5];
 322             jnrlistG         = jjnr[jidx+6];
 323             jnrlistH         = jjnr[jidx+7];
 324             /* Sign of each element will be negative for non-real atoms.
 325              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 326              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 327              */
 328             dummy_mask = gmx_mm256_set_m128(gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx+4)),_mm_setzero_si128())),
 329                                             gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128())));
 330
 331             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 332             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 333             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 334             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 335             jnrE       = (jnrlistE>=0) ? jnrlistE : 0;
 336             jnrF       = (jnrlistF>=0) ? jnrlistF : 0;
 337             jnrG       = (jnrlistG>=0) ? jnrlistG : 0;
 338             jnrH       = (jnrlistH>=0) ? jnrlistH : 0;
 339             j_coord_offsetA  = DIM*jnrA;
 340             j_coord_offsetB  = DIM*jnrB;
 341             j_coord_offsetC  = DIM*jnrC;
 342             j_coord_offsetD  = DIM*jnrD;
 343             j_coord_offsetE  = DIM*jnrE;
 344             j_coord_offsetF  = DIM*jnrF;
 345             j_coord_offsetG  = DIM*jnrG;
 346             j_coord_offsetH  = DIM*jnrH;
 347
 348             /* load j atom coordinates */
 349             gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 350                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 351                                                  x+j_coord_offsetE,x+j_coord_offsetF,
 352                                                  x+j_coord_offsetG,x+j_coord_offsetH,
 353                                                  &jx0,&jy0,&jz0);
 354
 355             /* Calculate displacement vector */
 356             dx00             = _mm256_sub_ps(ix0,jx0);
 357             dy00             = _mm256_sub_ps(iy0,jy0);
 358             dz00             = _mm256_sub_ps(iz0,jz0);
 359
 360             /* Calculate squared distance and things based on it */
 361             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
 362
 363             rinv00           = gmx_mm256_invsqrt_ps(rsq00);
 364
 365             /* Load parameters for j particles */
 366             jq0              = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 367                                                                  charge+jnrC+0,charge+jnrD+0,
 368                                                                  charge+jnrE+0,charge+jnrF+0,
 369                                                                  charge+jnrG+0,charge+jnrH+0);
 370             isaj0            = gmx_mm256_load_8real_swizzle_ps(invsqrta+jnrA+0,invsqrta+jnrB+0,
 371                                                                  invsqrta+jnrC+0,invsqrta+jnrD+0,
 372                                                                  invsqrta+jnrE+0,invsqrta+jnrF+0,
 373                                                                  invsqrta+jnrG+0,invsqrta+jnrH+0);
 374
 375             /**************************
 376              * CALCULATE INTERACTIONS *
 377              **************************/
 378
 379             r00              = _mm256_mul_ps(rsq00,rinv00);
 380             r00              = _mm256_andnot_ps(dummy_mask,r00);
 381
 382             /* Compute parameters for interactions between i and j atoms */
 383             qq00             = _mm256_mul_ps(iq0,jq0);
 384
 385             /* GENERALIZED BORN AND COULOMB ELECTROSTATICS */
 386             isaprod          = _mm256_mul_ps(isai0,isaj0);
 387             gbqqfactor       = _mm256_xor_ps(signbit,_mm256_mul_ps(qq00,_mm256_mul_ps(isaprod,gbinvepsdiff)));
 388             gbscale          = _mm256_mul_ps(isaprod,gbtabscale);
 389
 390             /* Calculate generalized born table index - this is a separate table from the normal one,
 391              * but we use the same procedure by multiplying r with scale and truncating to integer.
 392              */
 393             rt               = _mm256_mul_ps(r00,gbscale);
 394             gbitab           = _mm256_cvttps_epi32(rt);
 395             gbeps            = _mm256_sub_ps(rt,_mm256_round_ps(rt, _MM_FROUND_FLOOR));
 396             /*         AVX1 does not support 256-bit integer operations, so now we go to 128-bit mode... */
 397             gbitab_lo        = _mm256_extractf128_si256(gbitab,0x0);
 398             gbitab_hi        = _mm256_extractf128_si256(gbitab,0x1);
 399             gbitab_lo        = _mm_slli_epi32(gbitab_lo,2);
 400             gbitab_hi        = _mm_slli_epi32(gbitab_hi,2);
 401             Y                = gmx_mm256_set_m128(_mm_load_ps(gbtab + _mm_extract_epi32(gbitab_hi,0)),
 402                                                   _mm_load_ps(gbtab + _mm_extract_epi32(gbitab_lo,0)));
 403             F                = gmx_mm256_set_m128(_mm_load_ps(gbtab + _mm_extract_epi32(gbitab_hi,1)),
 404                                                   _mm_load_ps(gbtab + _mm_extract_epi32(gbitab_lo,1)));
 405             G                = gmx_mm256_set_m128(_mm_load_ps(gbtab + _mm_extract_epi32(gbitab_hi,2)),
 406                                                   _mm_load_ps(gbtab + _mm_extract_epi32(gbitab_lo,2)));
 407             H                = gmx_mm256_set_m128(_mm_load_ps(gbtab + _mm_extract_epi32(gbitab_hi,3)),
 408                                                   _mm_load_ps(gbtab + _mm_extract_epi32(gbitab_lo,3)));
 409             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
 410             Heps             = _mm256_mul_ps(gbeps,H);
 411             Fp               = _mm256_add_ps(F,_mm256_mul_ps(gbeps,_mm256_add_ps(G,Heps)));
 412             VV               = _mm256_add_ps(Y,_mm256_mul_ps(gbeps,Fp));
 413             vgb              = _mm256_mul_ps(gbqqfactor,VV);
 414
 415             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(gbeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
 416             fgb              = _mm256_mul_ps(gbqqfactor,_mm256_mul_ps(FF,gbscale));
 417             dvdatmp          = _mm256_mul_ps(minushalf,_mm256_add_ps(vgb,_mm256_mul_ps(fgb,r00)));
 418             dvdatmp          = _mm256_andnot_ps(dummy_mask,dvdatmp);
 419             dvdasum          = _mm256_add_ps(dvdasum,dvdatmp);
 420             /* 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. */
 421             fjptrA             = (jnrlistA>=0) ? dvda+jnrA : scratch;
 422             fjptrB             = (jnrlistB>=0) ? dvda+jnrB : scratch;
 423             fjptrC             = (jnrlistC>=0) ? dvda+jnrC : scratch;
 424             fjptrD             = (jnrlistD>=0) ? dvda+jnrD : scratch;
 425             fjptrE             = (jnrlistE>=0) ? dvda+jnrE : scratch;
 426             fjptrF             = (jnrlistF>=0) ? dvda+jnrF : scratch;
 427             fjptrG             = (jnrlistG>=0) ? dvda+jnrG : scratch;
 428             fjptrH             = (jnrlistH>=0) ? dvda+jnrH : scratch;
 429             gmx_mm256_increment_8real_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,
 430                                                  _mm256_mul_ps(dvdatmp,_mm256_mul_ps(isaj0,isaj0)));
 431             velec            = _mm256_mul_ps(qq00,rinv00);
 432             felec            = _mm256_mul_ps(_mm256_sub_ps(_mm256_mul_ps(velec,rinv00),fgb),rinv00);
 433
 434             /* Update potential sum for this i atom from the interaction with this j atom. */
 435             velec            = _mm256_andnot_ps(dummy_mask,velec);
 436             velecsum         = _mm256_add_ps(velecsum,velec);
 437             vgb              = _mm256_andnot_ps(dummy_mask,vgb);
 438             vgbsum           = _mm256_add_ps(vgbsum,vgb);
 439
 440             fscal            = felec;
 441
 442             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
 443
 444             /* Calculate temporary vectorial force */
 445             tx               = _mm256_mul_ps(fscal,dx00);
 446             ty               = _mm256_mul_ps(fscal,dy00);
 447             tz               = _mm256_mul_ps(fscal,dz00);
 448
 449             /* Update vectorial force */
 450             fix0             = _mm256_add_ps(fix0,tx);
 451             fiy0             = _mm256_add_ps(fiy0,ty);
 452             fiz0             = _mm256_add_ps(fiz0,tz);
 453
 454             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 455             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 456             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 457             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 458             fjptrE             = (jnrlistE>=0) ? f+j_coord_offsetE : scratch;
 459             fjptrF             = (jnrlistF>=0) ? f+j_coord_offsetF : scratch;
 460             fjptrG             = (jnrlistG>=0) ? f+j_coord_offsetG : scratch;
 461             fjptrH             = (jnrlistH>=0) ? f+j_coord_offsetH : scratch;
 462             gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,tx,ty,tz);
 463
 464             /* Inner loop uses 58 flops */
 465         }
 466
 467         /* End of innermost loop */
 468
 469         gmx_mm256_update_iforce_1atom_swizzle_ps(fix0,fiy0,fiz0,
 470                                                  f+i_coord_offset,fshift+i_shift_offset);
 471
 472         ggid                        = gid[iidx];
 473         /* Update potential energies */
 474         gmx_mm256_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid);
 475         gmx_mm256_update_1pot_ps(vgbsum,kernel_data->energygrp_polarization+ggid);
 476         dvdasum = _mm256_mul_ps(dvdasum, _mm256_mul_ps(isai0,isai0));
 477         gmx_mm256_update_1pot_ps(dvdasum,dvda+inr);
 478
 479         /* Increment number of inner iterations */
 480         inneriter                  += j_index_end - j_index_start;
 481
 482         /* Outer loop uses 9 flops */
 483     }
 484
 485     /* Increment number of outer iterations */
 486     outeriter        += nri;
 487
 488     /* Update outer/inner flops */
 489
 490     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VF,outeriter*9 + inneriter*58);
 491 }
 492 /*
 493  * Gromacs nonbonded kernel:   nb_kernel_ElecGB_VdwNone_GeomP1P1_F_avx_256_single
 494  * Electrostatics interaction: GeneralizedBorn
 495  * VdW interaction:            None
 496  * Geometry:                   Particle-Particle
 497  * Calculate force/pot:        Force
 498  */
 499 void
 500 nb_kernel_ElecGB_VdwNone_GeomP1P1_F_avx_256_single
 501                     (t_nblist                    * gmx_restrict       nlist,
 502                      rvec                        * gmx_restrict          xx,
 503                      rvec                        * gmx_restrict          ff,
 504                      t_forcerec                  * gmx_restrict          fr,
 505                      t_mdatoms                   * gmx_restrict     mdatoms,
 506                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 507                      t_nrnb                      * gmx_restrict        nrnb)
 508 {
 509     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 510      * just 0 for non-waters.
 511      * Suffixes A,B,C,D,E,F,G,H refer to j loop unrolling done with AVX, e.g. for the eight different
 512      * jnr indices corresponding to data put in the four positions in the SIMD register.
 513      */
 514     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 515     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 516     int              jnrA,jnrB,jnrC,jnrD;
 517     int              jnrE,jnrF,jnrG,jnrH;
 518     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
 519     int              jnrlistE,jnrlistF,jnrlistG,jnrlistH;
 520     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 521     int              j_coord_offsetE,j_coord_offsetF,j_coord_offsetG,j_coord_offsetH;
 522     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 523     real             rcutoff_scalar;
 524     real             *shiftvec,*fshift,*x,*f;
 525     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD,*fjptrE,*fjptrF,*fjptrG,*fjptrH;
 526     real             scratch[4*DIM];
 527     __m256           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 528     real *           vdwioffsetptr0;
 529     __m256           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 530     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D,vdwjidx0E,vdwjidx0F,vdwjidx0G,vdwjidx0H;
 531     __m256           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 532     __m256           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 533     __m256           velec,felec,velecsum,facel,crf,krf,krf2;
 534     real             *charge;
 535     __m256i          gbitab;
 536     __m128i          gbitab_lo,gbitab_hi;
 537     __m256           vgb,fgb,vgbsum,dvdasum,gbscale,gbtabscale,isaprod,gbqqfactor,gbinvepsdiff,gbeps,dvdatmp;
 538     __m256           minushalf = _mm256_set1_ps(-0.5);
 539     real             *invsqrta,*dvda,*gbtab;
 540     __m256i          vfitab;
 541     __m128i          vfitab_lo,vfitab_hi;
 542     __m128i          ifour       = _mm_set1_epi32(4);
 543     __m256           rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF;
 544     real             *vftab;
 545     __m256           dummy_mask,cutoff_mask;
 546     __m256           signbit = _mm256_castsi256_ps( _mm256_set1_epi32(0x80000000) );
 547     __m256           one     = _mm256_set1_ps(1.0);
 548     __m256           two     = _mm256_set1_ps(2.0);
 549     x                = xx[0];
 550     f                = ff[0];
 551
 552     nri              = nlist->nri;
 553     iinr             = nlist->iinr;
 554     jindex           = nlist->jindex;
 555     jjnr             = nlist->jjnr;
 556     shiftidx         = nlist->shift;
 557     gid              = nlist->gid;
 558     shiftvec         = fr->shift_vec[0];
 559     fshift           = fr->fshift[0];
 560     facel            = _mm256_set1_ps(fr->epsfac);
 561     charge           = mdatoms->chargeA;
 562
 563     invsqrta         = fr->invsqrta;
 564     dvda             = fr->dvda;
 565     gbtabscale       = _mm256_set1_ps(fr->gbtab.scale);
 566     gbtab            = fr->gbtab.data;
 567     gbinvepsdiff     = _mm256_set1_ps((1.0/fr->epsilon_r) - (1.0/fr->gb_epsilon_solvent));
 568
 569     /* Avoid stupid compiler warnings */
 570     jnrA = jnrB = jnrC = jnrD = jnrE = jnrF = jnrG = jnrH = 0;
 571     j_coord_offsetA = 0;
 572     j_coord_offsetB = 0;
 573     j_coord_offsetC = 0;
 574     j_coord_offsetD = 0;
 575     j_coord_offsetE = 0;
 576     j_coord_offsetF = 0;
 577     j_coord_offsetG = 0;
 578     j_coord_offsetH = 0;
 579
 580     outeriter        = 0;
 581     inneriter        = 0;
 582
 583     for(iidx=0;iidx<4*DIM;iidx++)
 584     {
 585         scratch[iidx] = 0.0;
 586     }
 587
 588     /* Start outer loop over neighborlists */
 589     for(iidx=0; iidx<nri; iidx++)
 590     {
 591         /* Load shift vector for this list */
 592         i_shift_offset   = DIM*shiftidx[iidx];
 593
 594         /* Load limits for loop over neighbors */
 595         j_index_start    = jindex[iidx];
 596         j_index_end      = jindex[iidx+1];
 597
 598         /* Get outer coordinate index */
 599         inr              = iinr[iidx];
 600         i_coord_offset   = DIM*inr;
 601
 602         /* Load i particle coords and add shift vector */
 603         gmx_mm256_load_shift_and_1rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 604
 605         fix0             = _mm256_setzero_ps();
 606         fiy0             = _mm256_setzero_ps();
 607         fiz0             = _mm256_setzero_ps();
 608
 609         /* Load parameters for i particles */
 610         iq0              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+0]));
 611         isai0            = _mm256_set1_ps(invsqrta[inr+0]);
 612
 613         dvdasum          = _mm256_setzero_ps();
 614
 615         /* Start inner kernel loop */
 616         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+7]>=0; jidx+=8)
 617         {
 618
 619             /* Get j neighbor index, and coordinate index */
 620             jnrA             = jjnr[jidx];
 621             jnrB             = jjnr[jidx+1];
 622             jnrC             = jjnr[jidx+2];
 623             jnrD             = jjnr[jidx+3];
 624             jnrE             = jjnr[jidx+4];
 625             jnrF             = jjnr[jidx+5];
 626             jnrG             = jjnr[jidx+6];
 627             jnrH             = jjnr[jidx+7];
 628             j_coord_offsetA  = DIM*jnrA;
 629             j_coord_offsetB  = DIM*jnrB;
 630             j_coord_offsetC  = DIM*jnrC;
 631             j_coord_offsetD  = DIM*jnrD;
 632             j_coord_offsetE  = DIM*jnrE;
 633             j_coord_offsetF  = DIM*jnrF;
 634             j_coord_offsetG  = DIM*jnrG;
 635             j_coord_offsetH  = DIM*jnrH;
 636
 637             /* load j atom coordinates */
 638             gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 639                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 640                                                  x+j_coord_offsetE,x+j_coord_offsetF,
 641                                                  x+j_coord_offsetG,x+j_coord_offsetH,
 642                                                  &jx0,&jy0,&jz0);
 643
 644             /* Calculate displacement vector */
 645             dx00             = _mm256_sub_ps(ix0,jx0);
 646             dy00             = _mm256_sub_ps(iy0,jy0);
 647             dz00             = _mm256_sub_ps(iz0,jz0);
 648
 649             /* Calculate squared distance and things based on it */
 650             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
 651
 652             rinv00           = gmx_mm256_invsqrt_ps(rsq00);
 653
 654             /* Load parameters for j particles */
 655             jq0              = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 656                                                                  charge+jnrC+0,charge+jnrD+0,
 657                                                                  charge+jnrE+0,charge+jnrF+0,
 658                                                                  charge+jnrG+0,charge+jnrH+0);
 659             isaj0            = gmx_mm256_load_8real_swizzle_ps(invsqrta+jnrA+0,invsqrta+jnrB+0,
 660                                                                  invsqrta+jnrC+0,invsqrta+jnrD+0,
 661                                                                  invsqrta+jnrE+0,invsqrta+jnrF+0,
 662                                                                  invsqrta+jnrG+0,invsqrta+jnrH+0);
 663
 664             /**************************
 665              * CALCULATE INTERACTIONS *
 666              **************************/
 667
 668             r00              = _mm256_mul_ps(rsq00,rinv00);
 669
 670             /* Compute parameters for interactions between i and j atoms */
 671             qq00             = _mm256_mul_ps(iq0,jq0);
 672
 673             /* GENERALIZED BORN AND COULOMB ELECTROSTATICS */
 674             isaprod          = _mm256_mul_ps(isai0,isaj0);
 675             gbqqfactor       = _mm256_xor_ps(signbit,_mm256_mul_ps(qq00,_mm256_mul_ps(isaprod,gbinvepsdiff)));
 676             gbscale          = _mm256_mul_ps(isaprod,gbtabscale);
 677
 678             /* Calculate generalized born table index - this is a separate table from the normal one,
 679              * but we use the same procedure by multiplying r with scale and truncating to integer.
 680              */
 681             rt               = _mm256_mul_ps(r00,gbscale);
 682             gbitab           = _mm256_cvttps_epi32(rt);
 683             gbeps            = _mm256_sub_ps(rt,_mm256_round_ps(rt, _MM_FROUND_FLOOR));
 684             /*         AVX1 does not support 256-bit integer operations, so now we go to 128-bit mode... */
 685             gbitab_lo        = _mm256_extractf128_si256(gbitab,0x0);
 686             gbitab_hi        = _mm256_extractf128_si256(gbitab,0x1);
 687             gbitab_lo        = _mm_slli_epi32(gbitab_lo,2);
 688             gbitab_hi        = _mm_slli_epi32(gbitab_hi,2);
 689             Y                = gmx_mm256_set_m128(_mm_load_ps(gbtab + _mm_extract_epi32(gbitab_hi,0)),
 690                                                   _mm_load_ps(gbtab + _mm_extract_epi32(gbitab_lo,0)));
 691             F                = gmx_mm256_set_m128(_mm_load_ps(gbtab + _mm_extract_epi32(gbitab_hi,1)),
 692                                                   _mm_load_ps(gbtab + _mm_extract_epi32(gbitab_lo,1)));
 693             G                = gmx_mm256_set_m128(_mm_load_ps(gbtab + _mm_extract_epi32(gbitab_hi,2)),
 694                                                   _mm_load_ps(gbtab + _mm_extract_epi32(gbitab_lo,2)));
 695             H                = gmx_mm256_set_m128(_mm_load_ps(gbtab + _mm_extract_epi32(gbitab_hi,3)),
 696                                                   _mm_load_ps(gbtab + _mm_extract_epi32(gbitab_lo,3)));
 697             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
 698             Heps             = _mm256_mul_ps(gbeps,H);
 699             Fp               = _mm256_add_ps(F,_mm256_mul_ps(gbeps,_mm256_add_ps(G,Heps)));
 700             VV               = _mm256_add_ps(Y,_mm256_mul_ps(gbeps,Fp));
 701             vgb              = _mm256_mul_ps(gbqqfactor,VV);
 702
 703             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(gbeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
 704             fgb              = _mm256_mul_ps(gbqqfactor,_mm256_mul_ps(FF,gbscale));
 705             dvdatmp          = _mm256_mul_ps(minushalf,_mm256_add_ps(vgb,_mm256_mul_ps(fgb,r00)));
 706             dvdasum          = _mm256_add_ps(dvdasum,dvdatmp);
 707             fjptrA           = dvda+jnrA;
 708             fjptrB           = dvda+jnrB;
 709             fjptrC           = dvda+jnrC;
 710             fjptrD           = dvda+jnrD;
 711             fjptrE           = dvda+jnrE;
 712             fjptrF           = dvda+jnrF;
 713             fjptrG           = dvda+jnrG;
 714             fjptrH           = dvda+jnrH;
 715             gmx_mm256_increment_8real_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,
 716                                                  _mm256_mul_ps(dvdatmp,_mm256_mul_ps(isaj0,isaj0)));
 717             velec            = _mm256_mul_ps(qq00,rinv00);
 718             felec            = _mm256_mul_ps(_mm256_sub_ps(_mm256_mul_ps(velec,rinv00),fgb),rinv00);
 719
 720             fscal            = felec;
 721
 722             /* Calculate temporary vectorial force */
 723             tx               = _mm256_mul_ps(fscal,dx00);
 724             ty               = _mm256_mul_ps(fscal,dy00);
 725             tz               = _mm256_mul_ps(fscal,dz00);
 726
 727             /* Update vectorial force */
 728             fix0             = _mm256_add_ps(fix0,tx);
 729             fiy0             = _mm256_add_ps(fiy0,ty);
 730             fiz0             = _mm256_add_ps(fiz0,tz);
 731
 732             fjptrA             = f+j_coord_offsetA;
 733             fjptrB             = f+j_coord_offsetB;
 734             fjptrC             = f+j_coord_offsetC;
 735             fjptrD             = f+j_coord_offsetD;
 736             fjptrE             = f+j_coord_offsetE;
 737             fjptrF             = f+j_coord_offsetF;
 738             fjptrG             = f+j_coord_offsetG;
 739             fjptrH             = f+j_coord_offsetH;
 740             gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,tx,ty,tz);
 741
 742             /* Inner loop uses 55 flops */
 743         }
 744
 745         if(jidx<j_index_end)
 746         {
 747
 748             /* Get j neighbor index, and coordinate index */
 749             jnrlistA         = jjnr[jidx];
 750             jnrlistB         = jjnr[jidx+1];
 751             jnrlistC         = jjnr[jidx+2];
 752             jnrlistD         = jjnr[jidx+3];
 753             jnrlistE         = jjnr[jidx+4];
 754             jnrlistF         = jjnr[jidx+5];
 755             jnrlistG         = jjnr[jidx+6];
 756             jnrlistH         = jjnr[jidx+7];
 757             /* Sign of each element will be negative for non-real atoms.
 758              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 759              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 760              */
 761             dummy_mask = gmx_mm256_set_m128(gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx+4)),_mm_setzero_si128())),
 762                                             gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128())));
 763
 764             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 765             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 766             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 767             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 768             jnrE       = (jnrlistE>=0) ? jnrlistE : 0;
 769             jnrF       = (jnrlistF>=0) ? jnrlistF : 0;
 770             jnrG       = (jnrlistG>=0) ? jnrlistG : 0;
 771             jnrH       = (jnrlistH>=0) ? jnrlistH : 0;
 772             j_coord_offsetA  = DIM*jnrA;
 773             j_coord_offsetB  = DIM*jnrB;
 774             j_coord_offsetC  = DIM*jnrC;
 775             j_coord_offsetD  = DIM*jnrD;
 776             j_coord_offsetE  = DIM*jnrE;
 777             j_coord_offsetF  = DIM*jnrF;
 778             j_coord_offsetG  = DIM*jnrG;
 779             j_coord_offsetH  = DIM*jnrH;
 780
 781             /* load j atom coordinates */
 782             gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 783                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 784                                                  x+j_coord_offsetE,x+j_coord_offsetF,
 785                                                  x+j_coord_offsetG,x+j_coord_offsetH,
 786                                                  &jx0,&jy0,&jz0);
 787
 788             /* Calculate displacement vector */
 789             dx00             = _mm256_sub_ps(ix0,jx0);
 790             dy00             = _mm256_sub_ps(iy0,jy0);
 791             dz00             = _mm256_sub_ps(iz0,jz0);
 792
 793             /* Calculate squared distance and things based on it */
 794             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
 795
 796             rinv00           = gmx_mm256_invsqrt_ps(rsq00);
 797
 798             /* Load parameters for j particles */
 799             jq0              = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 800                                                                  charge+jnrC+0,charge+jnrD+0,
 801                                                                  charge+jnrE+0,charge+jnrF+0,
 802                                                                  charge+jnrG+0,charge+jnrH+0);
 803             isaj0            = gmx_mm256_load_8real_swizzle_ps(invsqrta+jnrA+0,invsqrta+jnrB+0,
 804                                                                  invsqrta+jnrC+0,invsqrta+jnrD+0,
 805                                                                  invsqrta+jnrE+0,invsqrta+jnrF+0,
 806                                                                  invsqrta+jnrG+0,invsqrta+jnrH+0);
 807
 808             /**************************
 809              * CALCULATE INTERACTIONS *
 810              **************************/
 811
 812             r00              = _mm256_mul_ps(rsq00,rinv00);
 813             r00              = _mm256_andnot_ps(dummy_mask,r00);
 814
 815             /* Compute parameters for interactions between i and j atoms */
 816             qq00             = _mm256_mul_ps(iq0,jq0);
 817
 818             /* GENERALIZED BORN AND COULOMB ELECTROSTATICS */
 819             isaprod          = _mm256_mul_ps(isai0,isaj0);
 820             gbqqfactor       = _mm256_xor_ps(signbit,_mm256_mul_ps(qq00,_mm256_mul_ps(isaprod,gbinvepsdiff)));
 821             gbscale          = _mm256_mul_ps(isaprod,gbtabscale);
 822
 823             /* Calculate generalized born table index - this is a separate table from the normal one,
 824              * but we use the same procedure by multiplying r with scale and truncating to integer.
 825              */
 826             rt               = _mm256_mul_ps(r00,gbscale);
 827             gbitab           = _mm256_cvttps_epi32(rt);
 828             gbeps            = _mm256_sub_ps(rt,_mm256_round_ps(rt, _MM_FROUND_FLOOR));
 829             /*         AVX1 does not support 256-bit integer operations, so now we go to 128-bit mode... */
 830             gbitab_lo        = _mm256_extractf128_si256(gbitab,0x0);
 831             gbitab_hi        = _mm256_extractf128_si256(gbitab,0x1);
 832             gbitab_lo        = _mm_slli_epi32(gbitab_lo,2);
 833             gbitab_hi        = _mm_slli_epi32(gbitab_hi,2);
 834             Y                = gmx_mm256_set_m128(_mm_load_ps(gbtab + _mm_extract_epi32(gbitab_hi,0)),
 835                                                   _mm_load_ps(gbtab + _mm_extract_epi32(gbitab_lo,0)));
 836             F                = gmx_mm256_set_m128(_mm_load_ps(gbtab + _mm_extract_epi32(gbitab_hi,1)),
 837                                                   _mm_load_ps(gbtab + _mm_extract_epi32(gbitab_lo,1)));
 838             G                = gmx_mm256_set_m128(_mm_load_ps(gbtab + _mm_extract_epi32(gbitab_hi,2)),
 839                                                   _mm_load_ps(gbtab + _mm_extract_epi32(gbitab_lo,2)));
 840             H                = gmx_mm256_set_m128(_mm_load_ps(gbtab + _mm_extract_epi32(gbitab_hi,3)),
 841                                                   _mm_load_ps(gbtab + _mm_extract_epi32(gbitab_lo,3)));
 842             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
 843             Heps             = _mm256_mul_ps(gbeps,H);
 844             Fp               = _mm256_add_ps(F,_mm256_mul_ps(gbeps,_mm256_add_ps(G,Heps)));
 845             VV               = _mm256_add_ps(Y,_mm256_mul_ps(gbeps,Fp));
 846             vgb              = _mm256_mul_ps(gbqqfactor,VV);
 847
 848             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(gbeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
 849             fgb              = _mm256_mul_ps(gbqqfactor,_mm256_mul_ps(FF,gbscale));
 850             dvdatmp          = _mm256_mul_ps(minushalf,_mm256_add_ps(vgb,_mm256_mul_ps(fgb,r00)));
 851             dvdatmp          = _mm256_andnot_ps(dummy_mask,dvdatmp);
 852             dvdasum          = _mm256_add_ps(dvdasum,dvdatmp);
 853             /* 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. */
 854             fjptrA             = (jnrlistA>=0) ? dvda+jnrA : scratch;
 855             fjptrB             = (jnrlistB>=0) ? dvda+jnrB : scratch;
 856             fjptrC             = (jnrlistC>=0) ? dvda+jnrC : scratch;
 857             fjptrD             = (jnrlistD>=0) ? dvda+jnrD : scratch;
 858             fjptrE             = (jnrlistE>=0) ? dvda+jnrE : scratch;
 859             fjptrF             = (jnrlistF>=0) ? dvda+jnrF : scratch;
 860             fjptrG             = (jnrlistG>=0) ? dvda+jnrG : scratch;
 861             fjptrH             = (jnrlistH>=0) ? dvda+jnrH : scratch;
 862             gmx_mm256_increment_8real_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,
 863                                                  _mm256_mul_ps(dvdatmp,_mm256_mul_ps(isaj0,isaj0)));
 864             velec            = _mm256_mul_ps(qq00,rinv00);
 865             felec            = _mm256_mul_ps(_mm256_sub_ps(_mm256_mul_ps(velec,rinv00),fgb),rinv00);
 866
 867             fscal            = felec;
 868
 869             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
 870
 871             /* Calculate temporary vectorial force */
 872             tx               = _mm256_mul_ps(fscal,dx00);
 873             ty               = _mm256_mul_ps(fscal,dy00);
 874             tz               = _mm256_mul_ps(fscal,dz00);
 875
 876             /* Update vectorial force */
 877             fix0             = _mm256_add_ps(fix0,tx);
 878             fiy0             = _mm256_add_ps(fiy0,ty);
 879             fiz0             = _mm256_add_ps(fiz0,tz);
 880
 881             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 882             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 883             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 884             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 885             fjptrE             = (jnrlistE>=0) ? f+j_coord_offsetE : scratch;
 886             fjptrF             = (jnrlistF>=0) ? f+j_coord_offsetF : scratch;
 887             fjptrG             = (jnrlistG>=0) ? f+j_coord_offsetG : scratch;
 888             fjptrH             = (jnrlistH>=0) ? f+j_coord_offsetH : scratch;
 889             gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,tx,ty,tz);
 890
 891             /* Inner loop uses 56 flops */
 892         }
 893
 894         /* End of innermost loop */
 895
 896         gmx_mm256_update_iforce_1atom_swizzle_ps(fix0,fiy0,fiz0,
 897                                                  f+i_coord_offset,fshift+i_shift_offset);
 898
 899         dvdasum = _mm256_mul_ps(dvdasum, _mm256_mul_ps(isai0,isai0));
 900         gmx_mm256_update_1pot_ps(dvdasum,dvda+inr);
 901
 902         /* Increment number of inner iterations */
 903         inneriter                  += j_index_end - j_index_start;
 904
 905         /* Outer loop uses 7 flops */
 906     }
 907
 908     /* Increment number of outer iterations */
 909     outeriter        += nri;
 910
 911     /* Update outer/inner flops */
 912
 913     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_F,outeriter*7 + inneriter*56);
 914 }