src/gromacs/gmxlib/nonbonded/nb_kernel_avx_128_fma_single/nb_kernel_ElecGB_VdwLJ_GeomP1P1_avx_128_fma_single.c

   1 /*
   2  * This file is part of the GROMACS molecular simulation package.
   3  *
   4  * Copyright (c) 2012,2013, 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_128_fma_single kernel generator.
  37  */
  38 #ifdef HAVE_CONFIG_H
  39 #include <config.h>
  40 #endif
  41
  42 #include <math.h>
  43
  44 #include "../nb_kernel.h"
  45 #include "types/simple.h"
  46 #include "vec.h"
  47 #include "nrnb.h"
  48
  49 #include "gromacs/simd/math_x86_avx_128_fma_single.h"
  50 #include "kernelutil_x86_avx_128_fma_single.h"
  51
  52 /*
  53  * Gromacs nonbonded kernel:   nb_kernel_ElecGB_VdwLJ_GeomP1P1_VF_avx_128_fma_single
  54  * Electrostatics interaction: GeneralizedBorn
  55  * VdW interaction:            LennardJones
  56  * Geometry:                   Particle-Particle
  57  * Calculate force/pot:        PotentialAndForce
  58  */
  59 void
  60 nb_kernel_ElecGB_VdwLJ_GeomP1P1_VF_avx_128_fma_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 refer to j loop unrolling done with AVX_128, e.g. for the four 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              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
  78     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
  79     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
  80     real             rcutoff_scalar;
  81     real             *shiftvec,*fshift,*x,*f;
  82     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
  83     real             scratch[4*DIM];
  84     __m128           fscal,rcutoff,rcutoff2,jidxall;
  85     int              vdwioffset0;
  86     __m128           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
  87     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
  88     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  89     __m128           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  90     __m128           velec,felec,velecsum,facel,crf,krf,krf2;
  91     real             *charge;
  92     __m128i          gbitab;
  93     __m128           vgb,fgb,vgbsum,dvdasum,gbscale,gbtabscale,isaprod,gbqqfactor,gbinvepsdiff,gbeps,twogbeps,dvdatmp;
  94     __m128           minushalf = _mm_set1_ps(-0.5);
  95     real             *invsqrta,*dvda,*gbtab;
  96     int              nvdwtype;
  97     __m128           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
  98     int              *vdwtype;
  99     real             *vdwparam;
 100     __m128           one_sixth   = _mm_set1_ps(1.0/6.0);
 101     __m128           one_twelfth = _mm_set1_ps(1.0/12.0);
 102     __m128i          vfitab;
 103     __m128i          ifour       = _mm_set1_epi32(4);
 104     __m128           rt,vfeps,twovfeps,vftabscale,Y,F,G,H,Fp,VV,FF;
 105     real             *vftab;
 106     __m128           dummy_mask,cutoff_mask;
 107     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
 108     __m128           one     = _mm_set1_ps(1.0);
 109     __m128           two     = _mm_set1_ps(2.0);
 110     x                = xx[0];
 111     f                = ff[0];
 112
 113     nri              = nlist->nri;
 114     iinr             = nlist->iinr;
 115     jindex           = nlist->jindex;
 116     jjnr             = nlist->jjnr;
 117     shiftidx         = nlist->shift;
 118     gid              = nlist->gid;
 119     shiftvec         = fr->shift_vec[0];
 120     fshift           = fr->fshift[0];
 121     facel            = _mm_set1_ps(fr->epsfac);
 122     charge           = mdatoms->chargeA;
 123     nvdwtype         = fr->ntype;
 124     vdwparam         = fr->nbfp;
 125     vdwtype          = mdatoms->typeA;
 126
 127     invsqrta         = fr->invsqrta;
 128     dvda             = fr->dvda;
 129     gbtabscale       = _mm_set1_ps(fr->gbtab.scale);
 130     gbtab            = fr->gbtab.data;
 131     gbinvepsdiff     = _mm_set1_ps((1.0/fr->epsilon_r) - (1.0/fr->gb_epsilon_solvent));
 132
 133     /* Avoid stupid compiler warnings */
 134     jnrA = jnrB = jnrC = jnrD = 0;
 135     j_coord_offsetA = 0;
 136     j_coord_offsetB = 0;
 137     j_coord_offsetC = 0;
 138     j_coord_offsetD = 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_mm_load_shift_and_1rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 164
 165         fix0             = _mm_setzero_ps();
 166         fiy0             = _mm_setzero_ps();
 167         fiz0             = _mm_setzero_ps();
 168
 169         /* Load parameters for i particles */
 170         iq0              = _mm_mul_ps(facel,_mm_load1_ps(charge+inr+0));
 171         isai0            = _mm_load1_ps(invsqrta+inr+0);
 172         vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 173
 174         /* Reset potential sums */
 175         velecsum         = _mm_setzero_ps();
 176         vgbsum           = _mm_setzero_ps();
 177         vvdwsum          = _mm_setzero_ps();
 178         dvdasum          = _mm_setzero_ps();
 179
 180         /* Start inner kernel loop */
 181         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 182         {
 183
 184             /* Get j neighbor index, and coordinate index */
 185             jnrA             = jjnr[jidx];
 186             jnrB             = jjnr[jidx+1];
 187             jnrC             = jjnr[jidx+2];
 188             jnrD             = jjnr[jidx+3];
 189             j_coord_offsetA  = DIM*jnrA;
 190             j_coord_offsetB  = DIM*jnrB;
 191             j_coord_offsetC  = DIM*jnrC;
 192             j_coord_offsetD  = DIM*jnrD;
 193
 194             /* load j atom coordinates */
 195             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 196                                               x+j_coord_offsetC,x+j_coord_offsetD,
 197                                               &jx0,&jy0,&jz0);
 198
 199             /* Calculate displacement vector */
 200             dx00             = _mm_sub_ps(ix0,jx0);
 201             dy00             = _mm_sub_ps(iy0,jy0);
 202             dz00             = _mm_sub_ps(iz0,jz0);
 203
 204             /* Calculate squared distance and things based on it */
 205             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 206
 207             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 208
 209             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 210
 211             /* Load parameters for j particles */
 212             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 213                                                               charge+jnrC+0,charge+jnrD+0);
 214             isaj0            = gmx_mm_load_4real_swizzle_ps(invsqrta+jnrA+0,invsqrta+jnrB+0,
 215                                                               invsqrta+jnrC+0,invsqrta+jnrD+0);
 216             vdwjidx0A        = 2*vdwtype[jnrA+0];
 217             vdwjidx0B        = 2*vdwtype[jnrB+0];
 218             vdwjidx0C        = 2*vdwtype[jnrC+0];
 219             vdwjidx0D        = 2*vdwtype[jnrD+0];
 220
 221             /**************************
 222              * CALCULATE INTERACTIONS *
 223              **************************/
 224
 225             r00              = _mm_mul_ps(rsq00,rinv00);
 226
 227             /* Compute parameters for interactions between i and j atoms */
 228             qq00             = _mm_mul_ps(iq0,jq0);
 229             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 230                                          vdwparam+vdwioffset0+vdwjidx0B,
 231                                          vdwparam+vdwioffset0+vdwjidx0C,
 232                                          vdwparam+vdwioffset0+vdwjidx0D,
 233                                          &c6_00,&c12_00);
 234
 235             /* GENERALIZED BORN AND COULOMB ELECTROSTATICS */
 236             isaprod          = _mm_mul_ps(isai0,isaj0);
 237             gbqqfactor       = _mm_xor_ps(signbit,_mm_mul_ps(qq00,_mm_mul_ps(isaprod,gbinvepsdiff)));
 238             gbscale          = _mm_mul_ps(isaprod,gbtabscale);
 239
 240             /* Calculate generalized born table index - this is a separate table from the normal one,
 241              * but we use the same procedure by multiplying r with scale and truncating to integer.
 242              */
 243             rt               = _mm_mul_ps(r00,gbscale);
 244             gbitab           = _mm_cvttps_epi32(rt);
 245 #ifdef __XOP__
 246             gbeps            = _mm_frcz_ps(rt);
 247 #else
 248             gbeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
 249 #endif
 250             gbitab           = _mm_slli_epi32(gbitab,2);
 251
 252             Y                = _mm_load_ps( gbtab + _mm_extract_epi32(gbitab,0) );
 253             F                = _mm_load_ps( gbtab + _mm_extract_epi32(gbitab,1) );
 254             G                = _mm_load_ps( gbtab + _mm_extract_epi32(gbitab,2) );
 255             H                = _mm_load_ps( gbtab + _mm_extract_epi32(gbitab,3) );
 256             _MM_TRANSPOSE4_PS(Y,F,G,H);
 257             Fp               = _mm_macc_ps(gbeps,_mm_macc_ps(gbeps,H,G),F);
 258             VV               = _mm_macc_ps(gbeps,Fp,Y);
 259             vgb              = _mm_mul_ps(gbqqfactor,VV);
 260
 261             twogbeps         = _mm_add_ps(gbeps,gbeps);
 262             FF               = _mm_macc_ps(_mm_macc_ps(twogbeps,H,G),gbeps,Fp);
 263             fgb              = _mm_mul_ps(gbqqfactor,_mm_mul_ps(FF,gbscale));
 264             dvdatmp          = _mm_mul_ps(minushalf,_mm_macc_ps(fgb,r00,vgb));
 265             dvdasum          = _mm_add_ps(dvdasum,dvdatmp);
 266             fjptrA           = dvda+jnrA;
 267             fjptrB           = dvda+jnrB;
 268             fjptrC           = dvda+jnrC;
 269             fjptrD           = dvda+jnrD;
 270             gmx_mm_increment_4real_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,_mm_mul_ps(dvdatmp,_mm_mul_ps(isaj0,isaj0)));
 271             velec            = _mm_mul_ps(qq00,rinv00);
 272             felec            = _mm_mul_ps(_mm_msub_ps(velec,rinv00,fgb),rinv00);
 273
 274             /* LENNARD-JONES DISPERSION/REPULSION */
 275
 276             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 277             vvdw6            = _mm_mul_ps(c6_00,rinvsix);
 278             vvdw12           = _mm_mul_ps(c12_00,_mm_mul_ps(rinvsix,rinvsix));
 279             vvdw             = _mm_msub_ps(vvdw12,one_twelfth,_mm_mul_ps(vvdw6,one_sixth));
 280             fvdw             = _mm_mul_ps(_mm_sub_ps(vvdw12,vvdw6),rinvsq00);
 281
 282             /* Update potential sum for this i atom from the interaction with this j atom. */
 283             velecsum         = _mm_add_ps(velecsum,velec);
 284             vgbsum           = _mm_add_ps(vgbsum,vgb);
 285             vvdwsum          = _mm_add_ps(vvdwsum,vvdw);
 286
 287             fscal            = _mm_add_ps(felec,fvdw);
 288
 289              /* Update vectorial force */
 290             fix0             = _mm_macc_ps(dx00,fscal,fix0);
 291             fiy0             = _mm_macc_ps(dy00,fscal,fiy0);
 292             fiz0             = _mm_macc_ps(dz00,fscal,fiz0);
 293
 294             fjptrA             = f+j_coord_offsetA;
 295             fjptrB             = f+j_coord_offsetB;
 296             fjptrC             = f+j_coord_offsetC;
 297             fjptrD             = f+j_coord_offsetD;
 298             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,
 299                                                    _mm_mul_ps(dx00,fscal),
 300                                                    _mm_mul_ps(dy00,fscal),
 301                                                    _mm_mul_ps(dz00,fscal));
 302
 303             /* Inner loop uses 74 flops */
 304         }
 305
 306         if(jidx<j_index_end)
 307         {
 308
 309             /* Get j neighbor index, and coordinate index */
 310             jnrlistA         = jjnr[jidx];
 311             jnrlistB         = jjnr[jidx+1];
 312             jnrlistC         = jjnr[jidx+2];
 313             jnrlistD         = jjnr[jidx+3];
 314             /* Sign of each element will be negative for non-real atoms.
 315              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 316              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 317              */
 318             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 319             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 320             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 321             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 322             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 323             j_coord_offsetA  = DIM*jnrA;
 324             j_coord_offsetB  = DIM*jnrB;
 325             j_coord_offsetC  = DIM*jnrC;
 326             j_coord_offsetD  = DIM*jnrD;
 327
 328             /* load j atom coordinates */
 329             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 330                                               x+j_coord_offsetC,x+j_coord_offsetD,
 331                                               &jx0,&jy0,&jz0);
 332
 333             /* Calculate displacement vector */
 334             dx00             = _mm_sub_ps(ix0,jx0);
 335             dy00             = _mm_sub_ps(iy0,jy0);
 336             dz00             = _mm_sub_ps(iz0,jz0);
 337
 338             /* Calculate squared distance and things based on it */
 339             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 340
 341             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 342
 343             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 344
 345             /* Load parameters for j particles */
 346             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 347                                                               charge+jnrC+0,charge+jnrD+0);
 348             isaj0            = gmx_mm_load_4real_swizzle_ps(invsqrta+jnrA+0,invsqrta+jnrB+0,
 349                                                               invsqrta+jnrC+0,invsqrta+jnrD+0);
 350             vdwjidx0A        = 2*vdwtype[jnrA+0];
 351             vdwjidx0B        = 2*vdwtype[jnrB+0];
 352             vdwjidx0C        = 2*vdwtype[jnrC+0];
 353             vdwjidx0D        = 2*vdwtype[jnrD+0];
 354
 355             /**************************
 356              * CALCULATE INTERACTIONS *
 357              **************************/
 358
 359             r00              = _mm_mul_ps(rsq00,rinv00);
 360             r00              = _mm_andnot_ps(dummy_mask,r00);
 361
 362             /* Compute parameters for interactions between i and j atoms */
 363             qq00             = _mm_mul_ps(iq0,jq0);
 364             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 365                                          vdwparam+vdwioffset0+vdwjidx0B,
 366                                          vdwparam+vdwioffset0+vdwjidx0C,
 367                                          vdwparam+vdwioffset0+vdwjidx0D,
 368                                          &c6_00,&c12_00);
 369
 370             /* GENERALIZED BORN AND COULOMB ELECTROSTATICS */
 371             isaprod          = _mm_mul_ps(isai0,isaj0);
 372             gbqqfactor       = _mm_xor_ps(signbit,_mm_mul_ps(qq00,_mm_mul_ps(isaprod,gbinvepsdiff)));
 373             gbscale          = _mm_mul_ps(isaprod,gbtabscale);
 374
 375             /* Calculate generalized born table index - this is a separate table from the normal one,
 376              * but we use the same procedure by multiplying r with scale and truncating to integer.
 377              */
 378             rt               = _mm_mul_ps(r00,gbscale);
 379             gbitab           = _mm_cvttps_epi32(rt);
 380 #ifdef __XOP__
 381             gbeps            = _mm_frcz_ps(rt);
 382 #else
 383             gbeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
 384 #endif
 385             gbitab           = _mm_slli_epi32(gbitab,2);
 386
 387             Y                = _mm_load_ps( gbtab + _mm_extract_epi32(gbitab,0) );
 388             F                = _mm_load_ps( gbtab + _mm_extract_epi32(gbitab,1) );
 389             G                = _mm_load_ps( gbtab + _mm_extract_epi32(gbitab,2) );
 390             H                = _mm_load_ps( gbtab + _mm_extract_epi32(gbitab,3) );
 391             _MM_TRANSPOSE4_PS(Y,F,G,H);
 392             Fp               = _mm_macc_ps(gbeps,_mm_macc_ps(gbeps,H,G),F);
 393             VV               = _mm_macc_ps(gbeps,Fp,Y);
 394             vgb              = _mm_mul_ps(gbqqfactor,VV);
 395
 396             twogbeps         = _mm_add_ps(gbeps,gbeps);
 397             FF               = _mm_macc_ps(_mm_macc_ps(twogbeps,H,G),gbeps,Fp);
 398             fgb              = _mm_mul_ps(gbqqfactor,_mm_mul_ps(FF,gbscale));
 399             dvdatmp          = _mm_mul_ps(minushalf,_mm_macc_ps(fgb,r00,vgb));
 400             dvdatmp          = _mm_andnot_ps(dummy_mask,dvdatmp);
 401             dvdasum          = _mm_add_ps(dvdasum,dvdatmp);
 402             /* 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. */
 403             fjptrA             = (jnrlistA>=0) ? dvda+jnrA : scratch;
 404             fjptrB             = (jnrlistB>=0) ? dvda+jnrB : scratch;
 405             fjptrC             = (jnrlistC>=0) ? dvda+jnrC : scratch;
 406             fjptrD             = (jnrlistD>=0) ? dvda+jnrD : scratch;
 407             gmx_mm_increment_4real_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,_mm_mul_ps(dvdatmp,_mm_mul_ps(isaj0,isaj0)));
 408             velec            = _mm_mul_ps(qq00,rinv00);
 409             felec            = _mm_mul_ps(_mm_msub_ps(velec,rinv00,fgb),rinv00);
 410
 411             /* LENNARD-JONES DISPERSION/REPULSION */
 412
 413             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 414             vvdw6            = _mm_mul_ps(c6_00,rinvsix);
 415             vvdw12           = _mm_mul_ps(c12_00,_mm_mul_ps(rinvsix,rinvsix));
 416             vvdw             = _mm_msub_ps(vvdw12,one_twelfth,_mm_mul_ps(vvdw6,one_sixth));
 417             fvdw             = _mm_mul_ps(_mm_sub_ps(vvdw12,vvdw6),rinvsq00);
 418
 419             /* Update potential sum for this i atom from the interaction with this j atom. */
 420             velec            = _mm_andnot_ps(dummy_mask,velec);
 421             velecsum         = _mm_add_ps(velecsum,velec);
 422             vgb              = _mm_andnot_ps(dummy_mask,vgb);
 423             vgbsum           = _mm_add_ps(vgbsum,vgb);
 424             vvdw             = _mm_andnot_ps(dummy_mask,vvdw);
 425             vvdwsum          = _mm_add_ps(vvdwsum,vvdw);
 426
 427             fscal            = _mm_add_ps(felec,fvdw);
 428
 429             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 430
 431              /* Update vectorial force */
 432             fix0             = _mm_macc_ps(dx00,fscal,fix0);
 433             fiy0             = _mm_macc_ps(dy00,fscal,fiy0);
 434             fiz0             = _mm_macc_ps(dz00,fscal,fiz0);
 435
 436             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 437             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 438             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 439             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 440             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,
 441                                                    _mm_mul_ps(dx00,fscal),
 442                                                    _mm_mul_ps(dy00,fscal),
 443                                                    _mm_mul_ps(dz00,fscal));
 444
 445             /* Inner loop uses 75 flops */
 446         }
 447
 448         /* End of innermost loop */
 449
 450         gmx_mm_update_iforce_1atom_swizzle_ps(fix0,fiy0,fiz0,
 451                                               f+i_coord_offset,fshift+i_shift_offset);
 452
 453         ggid                        = gid[iidx];
 454         /* Update potential energies */
 455         gmx_mm_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid);
 456         gmx_mm_update_1pot_ps(vgbsum,kernel_data->energygrp_polarization+ggid);
 457         gmx_mm_update_1pot_ps(vvdwsum,kernel_data->energygrp_vdw+ggid);
 458         dvdasum = _mm_mul_ps(dvdasum, _mm_mul_ps(isai0,isai0));
 459         gmx_mm_update_1pot_ps(dvdasum,dvda+inr);
 460
 461         /* Increment number of inner iterations */
 462         inneriter                  += j_index_end - j_index_start;
 463
 464         /* Outer loop uses 10 flops */
 465     }
 466
 467     /* Increment number of outer iterations */
 468     outeriter        += nri;
 469
 470     /* Update outer/inner flops */
 471
 472     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_VF,outeriter*10 + inneriter*75);
 473 }
 474 /*
 475  * Gromacs nonbonded kernel:   nb_kernel_ElecGB_VdwLJ_GeomP1P1_F_avx_128_fma_single
 476  * Electrostatics interaction: GeneralizedBorn
 477  * VdW interaction:            LennardJones
 478  * Geometry:                   Particle-Particle
 479  * Calculate force/pot:        Force
 480  */
 481 void
 482 nb_kernel_ElecGB_VdwLJ_GeomP1P1_F_avx_128_fma_single
 483                     (t_nblist                    * gmx_restrict       nlist,
 484                      rvec                        * gmx_restrict          xx,
 485                      rvec                        * gmx_restrict          ff,
 486                      t_forcerec                  * gmx_restrict          fr,
 487                      t_mdatoms                   * gmx_restrict     mdatoms,
 488                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 489                      t_nrnb                      * gmx_restrict        nrnb)
 490 {
 491     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 492      * just 0 for non-waters.
 493      * Suffixes A,B,C,D refer to j loop unrolling done with AVX_128, e.g. for the four different
 494      * jnr indices corresponding to data put in the four positions in the SIMD register.
 495      */
 496     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 497     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 498     int              jnrA,jnrB,jnrC,jnrD;
 499     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
 500     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 501     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 502     real             rcutoff_scalar;
 503     real             *shiftvec,*fshift,*x,*f;
 504     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
 505     real             scratch[4*DIM];
 506     __m128           fscal,rcutoff,rcutoff2,jidxall;
 507     int              vdwioffset0;
 508     __m128           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 509     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
 510     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 511     __m128           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 512     __m128           velec,felec,velecsum,facel,crf,krf,krf2;
 513     real             *charge;
 514     __m128i          gbitab;
 515     __m128           vgb,fgb,vgbsum,dvdasum,gbscale,gbtabscale,isaprod,gbqqfactor,gbinvepsdiff,gbeps,twogbeps,dvdatmp;
 516     __m128           minushalf = _mm_set1_ps(-0.5);
 517     real             *invsqrta,*dvda,*gbtab;
 518     int              nvdwtype;
 519     __m128           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 520     int              *vdwtype;
 521     real             *vdwparam;
 522     __m128           one_sixth   = _mm_set1_ps(1.0/6.0);
 523     __m128           one_twelfth = _mm_set1_ps(1.0/12.0);
 524     __m128i          vfitab;
 525     __m128i          ifour       = _mm_set1_epi32(4);
 526     __m128           rt,vfeps,twovfeps,vftabscale,Y,F,G,H,Fp,VV,FF;
 527     real             *vftab;
 528     __m128           dummy_mask,cutoff_mask;
 529     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
 530     __m128           one     = _mm_set1_ps(1.0);
 531     __m128           two     = _mm_set1_ps(2.0);
 532     x                = xx[0];
 533     f                = ff[0];
 534
 535     nri              = nlist->nri;
 536     iinr             = nlist->iinr;
 537     jindex           = nlist->jindex;
 538     jjnr             = nlist->jjnr;
 539     shiftidx         = nlist->shift;
 540     gid              = nlist->gid;
 541     shiftvec         = fr->shift_vec[0];
 542     fshift           = fr->fshift[0];
 543     facel            = _mm_set1_ps(fr->epsfac);
 544     charge           = mdatoms->chargeA;
 545     nvdwtype         = fr->ntype;
 546     vdwparam         = fr->nbfp;
 547     vdwtype          = mdatoms->typeA;
 548
 549     invsqrta         = fr->invsqrta;
 550     dvda             = fr->dvda;
 551     gbtabscale       = _mm_set1_ps(fr->gbtab.scale);
 552     gbtab            = fr->gbtab.data;
 553     gbinvepsdiff     = _mm_set1_ps((1.0/fr->epsilon_r) - (1.0/fr->gb_epsilon_solvent));
 554
 555     /* Avoid stupid compiler warnings */
 556     jnrA = jnrB = jnrC = jnrD = 0;
 557     j_coord_offsetA = 0;
 558     j_coord_offsetB = 0;
 559     j_coord_offsetC = 0;
 560     j_coord_offsetD = 0;
 561
 562     outeriter        = 0;
 563     inneriter        = 0;
 564
 565     for(iidx=0;iidx<4*DIM;iidx++)
 566     {
 567         scratch[iidx] = 0.0;
 568     }
 569
 570     /* Start outer loop over neighborlists */
 571     for(iidx=0; iidx<nri; iidx++)
 572     {
 573         /* Load shift vector for this list */
 574         i_shift_offset   = DIM*shiftidx[iidx];
 575
 576         /* Load limits for loop over neighbors */
 577         j_index_start    = jindex[iidx];
 578         j_index_end      = jindex[iidx+1];
 579
 580         /* Get outer coordinate index */
 581         inr              = iinr[iidx];
 582         i_coord_offset   = DIM*inr;
 583
 584         /* Load i particle coords and add shift vector */
 585         gmx_mm_load_shift_and_1rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 586
 587         fix0             = _mm_setzero_ps();
 588         fiy0             = _mm_setzero_ps();
 589         fiz0             = _mm_setzero_ps();
 590
 591         /* Load parameters for i particles */
 592         iq0              = _mm_mul_ps(facel,_mm_load1_ps(charge+inr+0));
 593         isai0            = _mm_load1_ps(invsqrta+inr+0);
 594         vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 595
 596         dvdasum          = _mm_setzero_ps();
 597
 598         /* Start inner kernel loop */
 599         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 600         {
 601
 602             /* Get j neighbor index, and coordinate index */
 603             jnrA             = jjnr[jidx];
 604             jnrB             = jjnr[jidx+1];
 605             jnrC             = jjnr[jidx+2];
 606             jnrD             = jjnr[jidx+3];
 607             j_coord_offsetA  = DIM*jnrA;
 608             j_coord_offsetB  = DIM*jnrB;
 609             j_coord_offsetC  = DIM*jnrC;
 610             j_coord_offsetD  = DIM*jnrD;
 611
 612             /* load j atom coordinates */
 613             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 614                                               x+j_coord_offsetC,x+j_coord_offsetD,
 615                                               &jx0,&jy0,&jz0);
 616
 617             /* Calculate displacement vector */
 618             dx00             = _mm_sub_ps(ix0,jx0);
 619             dy00             = _mm_sub_ps(iy0,jy0);
 620             dz00             = _mm_sub_ps(iz0,jz0);
 621
 622             /* Calculate squared distance and things based on it */
 623             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 624
 625             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 626
 627             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 628
 629             /* Load parameters for j particles */
 630             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 631                                                               charge+jnrC+0,charge+jnrD+0);
 632             isaj0            = gmx_mm_load_4real_swizzle_ps(invsqrta+jnrA+0,invsqrta+jnrB+0,
 633                                                               invsqrta+jnrC+0,invsqrta+jnrD+0);
 634             vdwjidx0A        = 2*vdwtype[jnrA+0];
 635             vdwjidx0B        = 2*vdwtype[jnrB+0];
 636             vdwjidx0C        = 2*vdwtype[jnrC+0];
 637             vdwjidx0D        = 2*vdwtype[jnrD+0];
 638
 639             /**************************
 640              * CALCULATE INTERACTIONS *
 641              **************************/
 642
 643             r00              = _mm_mul_ps(rsq00,rinv00);
 644
 645             /* Compute parameters for interactions between i and j atoms */
 646             qq00             = _mm_mul_ps(iq0,jq0);
 647             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 648                                          vdwparam+vdwioffset0+vdwjidx0B,
 649                                          vdwparam+vdwioffset0+vdwjidx0C,
 650                                          vdwparam+vdwioffset0+vdwjidx0D,
 651                                          &c6_00,&c12_00);
 652
 653             /* GENERALIZED BORN AND COULOMB ELECTROSTATICS */
 654             isaprod          = _mm_mul_ps(isai0,isaj0);
 655             gbqqfactor       = _mm_xor_ps(signbit,_mm_mul_ps(qq00,_mm_mul_ps(isaprod,gbinvepsdiff)));
 656             gbscale          = _mm_mul_ps(isaprod,gbtabscale);
 657
 658             /* Calculate generalized born table index - this is a separate table from the normal one,
 659              * but we use the same procedure by multiplying r with scale and truncating to integer.
 660              */
 661             rt               = _mm_mul_ps(r00,gbscale);
 662             gbitab           = _mm_cvttps_epi32(rt);
 663 #ifdef __XOP__
 664             gbeps            = _mm_frcz_ps(rt);
 665 #else
 666             gbeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
 667 #endif
 668             gbitab           = _mm_slli_epi32(gbitab,2);
 669
 670             Y                = _mm_load_ps( gbtab + _mm_extract_epi32(gbitab,0) );
 671             F                = _mm_load_ps( gbtab + _mm_extract_epi32(gbitab,1) );
 672             G                = _mm_load_ps( gbtab + _mm_extract_epi32(gbitab,2) );
 673             H                = _mm_load_ps( gbtab + _mm_extract_epi32(gbitab,3) );
 674             _MM_TRANSPOSE4_PS(Y,F,G,H);
 675             Fp               = _mm_macc_ps(gbeps,_mm_macc_ps(gbeps,H,G),F);
 676             VV               = _mm_macc_ps(gbeps,Fp,Y);
 677             vgb              = _mm_mul_ps(gbqqfactor,VV);
 678
 679             twogbeps         = _mm_add_ps(gbeps,gbeps);
 680             FF               = _mm_macc_ps(_mm_macc_ps(twogbeps,H,G),gbeps,Fp);
 681             fgb              = _mm_mul_ps(gbqqfactor,_mm_mul_ps(FF,gbscale));
 682             dvdatmp          = _mm_mul_ps(minushalf,_mm_macc_ps(fgb,r00,vgb));
 683             dvdasum          = _mm_add_ps(dvdasum,dvdatmp);
 684             fjptrA           = dvda+jnrA;
 685             fjptrB           = dvda+jnrB;
 686             fjptrC           = dvda+jnrC;
 687             fjptrD           = dvda+jnrD;
 688             gmx_mm_increment_4real_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,_mm_mul_ps(dvdatmp,_mm_mul_ps(isaj0,isaj0)));
 689             velec            = _mm_mul_ps(qq00,rinv00);
 690             felec            = _mm_mul_ps(_mm_msub_ps(velec,rinv00,fgb),rinv00);
 691
 692             /* LENNARD-JONES DISPERSION/REPULSION */
 693
 694             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 695             fvdw             = _mm_mul_ps(_mm_msub_ps(c12_00,rinvsix,c6_00),_mm_mul_ps(rinvsix,rinvsq00));
 696
 697             fscal            = _mm_add_ps(felec,fvdw);
 698
 699              /* Update vectorial force */
 700             fix0             = _mm_macc_ps(dx00,fscal,fix0);
 701             fiy0             = _mm_macc_ps(dy00,fscal,fiy0);
 702             fiz0             = _mm_macc_ps(dz00,fscal,fiz0);
 703
 704             fjptrA             = f+j_coord_offsetA;
 705             fjptrB             = f+j_coord_offsetB;
 706             fjptrC             = f+j_coord_offsetC;
 707             fjptrD             = f+j_coord_offsetD;
 708             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,
 709                                                    _mm_mul_ps(dx00,fscal),
 710                                                    _mm_mul_ps(dy00,fscal),
 711                                                    _mm_mul_ps(dz00,fscal));
 712
 713             /* Inner loop uses 67 flops */
 714         }
 715
 716         if(jidx<j_index_end)
 717         {
 718
 719             /* Get j neighbor index, and coordinate index */
 720             jnrlistA         = jjnr[jidx];
 721             jnrlistB         = jjnr[jidx+1];
 722             jnrlistC         = jjnr[jidx+2];
 723             jnrlistD         = jjnr[jidx+3];
 724             /* Sign of each element will be negative for non-real atoms.
 725              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 726              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 727              */
 728             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 729             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 730             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 731             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 732             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 733             j_coord_offsetA  = DIM*jnrA;
 734             j_coord_offsetB  = DIM*jnrB;
 735             j_coord_offsetC  = DIM*jnrC;
 736             j_coord_offsetD  = DIM*jnrD;
 737
 738             /* load j atom coordinates */
 739             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 740                                               x+j_coord_offsetC,x+j_coord_offsetD,
 741                                               &jx0,&jy0,&jz0);
 742
 743             /* Calculate displacement vector */
 744             dx00             = _mm_sub_ps(ix0,jx0);
 745             dy00             = _mm_sub_ps(iy0,jy0);
 746             dz00             = _mm_sub_ps(iz0,jz0);
 747
 748             /* Calculate squared distance and things based on it */
 749             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 750
 751             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 752
 753             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 754
 755             /* Load parameters for j particles */
 756             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 757                                                               charge+jnrC+0,charge+jnrD+0);
 758             isaj0            = gmx_mm_load_4real_swizzle_ps(invsqrta+jnrA+0,invsqrta+jnrB+0,
 759                                                               invsqrta+jnrC+0,invsqrta+jnrD+0);
 760             vdwjidx0A        = 2*vdwtype[jnrA+0];
 761             vdwjidx0B        = 2*vdwtype[jnrB+0];
 762             vdwjidx0C        = 2*vdwtype[jnrC+0];
 763             vdwjidx0D        = 2*vdwtype[jnrD+0];
 764
 765             /**************************
 766              * CALCULATE INTERACTIONS *
 767              **************************/
 768
 769             r00              = _mm_mul_ps(rsq00,rinv00);
 770             r00              = _mm_andnot_ps(dummy_mask,r00);
 771
 772             /* Compute parameters for interactions between i and j atoms */
 773             qq00             = _mm_mul_ps(iq0,jq0);
 774             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 775                                          vdwparam+vdwioffset0+vdwjidx0B,
 776                                          vdwparam+vdwioffset0+vdwjidx0C,
 777                                          vdwparam+vdwioffset0+vdwjidx0D,
 778                                          &c6_00,&c12_00);
 779
 780             /* GENERALIZED BORN AND COULOMB ELECTROSTATICS */
 781             isaprod          = _mm_mul_ps(isai0,isaj0);
 782             gbqqfactor       = _mm_xor_ps(signbit,_mm_mul_ps(qq00,_mm_mul_ps(isaprod,gbinvepsdiff)));
 783             gbscale          = _mm_mul_ps(isaprod,gbtabscale);
 784
 785             /* Calculate generalized born table index - this is a separate table from the normal one,
 786              * but we use the same procedure by multiplying r with scale and truncating to integer.
 787              */
 788             rt               = _mm_mul_ps(r00,gbscale);
 789             gbitab           = _mm_cvttps_epi32(rt);
 790 #ifdef __XOP__
 791             gbeps            = _mm_frcz_ps(rt);
 792 #else
 793             gbeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
 794 #endif
 795             gbitab           = _mm_slli_epi32(gbitab,2);
 796
 797             Y                = _mm_load_ps( gbtab + _mm_extract_epi32(gbitab,0) );
 798             F                = _mm_load_ps( gbtab + _mm_extract_epi32(gbitab,1) );
 799             G                = _mm_load_ps( gbtab + _mm_extract_epi32(gbitab,2) );
 800             H                = _mm_load_ps( gbtab + _mm_extract_epi32(gbitab,3) );
 801             _MM_TRANSPOSE4_PS(Y,F,G,H);
 802             Fp               = _mm_macc_ps(gbeps,_mm_macc_ps(gbeps,H,G),F);
 803             VV               = _mm_macc_ps(gbeps,Fp,Y);
 804             vgb              = _mm_mul_ps(gbqqfactor,VV);
 805
 806             twogbeps         = _mm_add_ps(gbeps,gbeps);
 807             FF               = _mm_macc_ps(_mm_macc_ps(twogbeps,H,G),gbeps,Fp);
 808             fgb              = _mm_mul_ps(gbqqfactor,_mm_mul_ps(FF,gbscale));
 809             dvdatmp          = _mm_mul_ps(minushalf,_mm_macc_ps(fgb,r00,vgb));
 810             dvdatmp          = _mm_andnot_ps(dummy_mask,dvdatmp);
 811             dvdasum          = _mm_add_ps(dvdasum,dvdatmp);
 812             /* 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. */
 813             fjptrA             = (jnrlistA>=0) ? dvda+jnrA : scratch;
 814             fjptrB             = (jnrlistB>=0) ? dvda+jnrB : scratch;
 815             fjptrC             = (jnrlistC>=0) ? dvda+jnrC : scratch;
 816             fjptrD             = (jnrlistD>=0) ? dvda+jnrD : scratch;
 817             gmx_mm_increment_4real_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,_mm_mul_ps(dvdatmp,_mm_mul_ps(isaj0,isaj0)));
 818             velec            = _mm_mul_ps(qq00,rinv00);
 819             felec            = _mm_mul_ps(_mm_msub_ps(velec,rinv00,fgb),rinv00);
 820
 821             /* LENNARD-JONES DISPERSION/REPULSION */
 822
 823             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 824             fvdw             = _mm_mul_ps(_mm_msub_ps(c12_00,rinvsix,c6_00),_mm_mul_ps(rinvsix,rinvsq00));
 825
 826             fscal            = _mm_add_ps(felec,fvdw);
 827
 828             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 829
 830              /* Update vectorial force */
 831             fix0             = _mm_macc_ps(dx00,fscal,fix0);
 832             fiy0             = _mm_macc_ps(dy00,fscal,fiy0);
 833             fiz0             = _mm_macc_ps(dz00,fscal,fiz0);
 834
 835             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 836             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 837             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 838             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 839             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,
 840                                                    _mm_mul_ps(dx00,fscal),
 841                                                    _mm_mul_ps(dy00,fscal),
 842                                                    _mm_mul_ps(dz00,fscal));
 843
 844             /* Inner loop uses 68 flops */
 845         }
 846
 847         /* End of innermost loop */
 848
 849         gmx_mm_update_iforce_1atom_swizzle_ps(fix0,fiy0,fiz0,
 850                                               f+i_coord_offset,fshift+i_shift_offset);
 851
 852         dvdasum = _mm_mul_ps(dvdasum, _mm_mul_ps(isai0,isai0));
 853         gmx_mm_update_1pot_ps(dvdasum,dvda+inr);
 854
 855         /* Increment number of inner iterations */
 856         inneriter                  += j_index_end - j_index_start;
 857
 858         /* Outer loop uses 7 flops */
 859     }
 860
 861     /* Increment number of outer iterations */
 862     outeriter        += nri;
 863
 864     /* Update outer/inner flops */
 865
 866     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_F,outeriter*7 + inneriter*68);
 867 }