src/gromacs/gmxlib/nonbonded/nb_kernel_avx_128_fma_single/nb_kernel_ElecCSTab_VdwCSTab_GeomW3P1_avx_128_fma_single.c

   1 /*
   2  * This file is part of the GROMACS molecular simulation package.
   3  *
   4  * Copyright (c) 2012,2013,2014,2015,2017, by the GROMACS development team, led by
   5  * Mark Abraham, David van der Spoel, Berk Hess, and Erik Lindahl,
   6  * and including many others, as listed in the AUTHORS file in the
   7  * top-level source directory and at http://www.gromacs.org.
   8  *
   9  * GROMACS is free software; you can redistribute it and/or
  10  * modify it under the terms of the GNU Lesser General Public License
  11  * as published by the Free Software Foundation; either version 2.1
  12  * of the License, or (at your option) any later version.
  13  *
  14  * GROMACS is distributed in the hope that it will be useful,
  15  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  16  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  17  * Lesser General Public License for more details.
  18  *
  19  * You should have received a copy of the GNU Lesser General Public
  20  * License along with GROMACS; if not, see
  21  * http://www.gnu.org/licenses, or write to the Free Software Foundation,
  22  * Inc., 51 Franklin Street, Fifth Floor, Boston, MA  02110-1301  USA.
  23  *
  24  * If you want to redistribute modifications to GROMACS, please
  25  * consider that scientific software is very special. Version
  26  * control is crucial - bugs must be traceable. We will be happy to
  27  * consider code for inclusion in the official distribution, but
  28  * derived work must not be called official GROMACS. Details are found
  29  * in the README & COPYING files - if they are missing, get the
  30  * official version at http://www.gromacs.org.
  31  *
  32  * To help us fund GROMACS development, we humbly ask that you cite
  33  * the research papers on the package. Check out http://www.gromacs.org.
  34  */
  35 /*
  36  * Note: this file was generated by the GROMACS avx_128_fma_single kernel generator.
  37  */
  38 #include "gmxpre.h"
  39
  40 #include "config.h"
  41
  42 #include <math.h>
  43
  44 #include "../nb_kernel.h"
  45 #include "gromacs/gmxlib/nrnb.h"
  46
  47 #include "kernelutil_x86_avx_128_fma_single.h"
  48
  49 /*
  50  * Gromacs nonbonded kernel:   nb_kernel_ElecCSTab_VdwCSTab_GeomW3P1_VF_avx_128_fma_single
  51  * Electrostatics interaction: CubicSplineTable
  52  * VdW interaction:            CubicSplineTable
  53  * Geometry:                   Water3-Particle
  54  * Calculate force/pot:        PotentialAndForce
  55  */
  56 void
  57 nb_kernel_ElecCSTab_VdwCSTab_GeomW3P1_VF_avx_128_fma_single
  58                     (t_nblist                    * gmx_restrict       nlist,
  59                      rvec                        * gmx_restrict          xx,
  60                      rvec                        * gmx_restrict          ff,
  61                      struct t_forcerec           * gmx_restrict          fr,
  62                      t_mdatoms                   * gmx_restrict     mdatoms,
  63                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
  64                      t_nrnb                      * gmx_restrict        nrnb)
  65 {
  66     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
  67      * just 0 for non-waters.
  68      * Suffixes A,B,C,D refer to j loop unrolling done with AVX_128, e.g. for the four different
  69      * jnr indices corresponding to data put in the four positions in the SIMD register.
  70      */
  71     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
  72     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
  73     int              jnrA,jnrB,jnrC,jnrD;
  74     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
  75     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
  76     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
  77     real             rcutoff_scalar;
  78     real             *shiftvec,*fshift,*x,*f;
  79     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
  80     real             scratch[4*DIM];
  81     __m128           fscal,rcutoff,rcutoff2,jidxall;
  82     int              vdwioffset0;
  83     __m128           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
  84     int              vdwioffset1;
  85     __m128           ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
  86     int              vdwioffset2;
  87     __m128           ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
  88     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
  89     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  90     __m128           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  91     __m128           dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
  92     __m128           dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
  93     __m128           velec,felec,velecsum,facel,crf,krf,krf2;
  94     real             *charge;
  95     int              nvdwtype;
  96     __m128           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
  97     int              *vdwtype;
  98     real             *vdwparam;
  99     __m128           one_sixth   = _mm_set1_ps(1.0/6.0);
 100     __m128           one_twelfth = _mm_set1_ps(1.0/12.0);
 101     __m128i          vfitab;
 102     __m128i          ifour       = _mm_set1_epi32(4);
 103     __m128           rt,vfeps,twovfeps,vftabscale,Y,F,G,H,Fp,VV,FF;
 104     real             *vftab;
 105     __m128           dummy_mask,cutoff_mask;
 106     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
 107     __m128           one     = _mm_set1_ps(1.0);
 108     __m128           two     = _mm_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            = _mm_set1_ps(fr->ic->epsfac);
 121     charge           = mdatoms->chargeA;
 122     nvdwtype         = fr->ntype;
 123     vdwparam         = fr->nbfp;
 124     vdwtype          = mdatoms->typeA;
 125
 126     vftab            = kernel_data->table_elec_vdw->data;
 127     vftabscale       = _mm_set1_ps(kernel_data->table_elec_vdw->scale);
 128
 129     /* Setup water-specific parameters */
 130     inr              = nlist->iinr[0];
 131     iq0              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+0]));
 132     iq1              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+1]));
 133     iq2              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+2]));
 134     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 135
 136     /* Avoid stupid compiler warnings */
 137     jnrA = jnrB = jnrC = jnrD = 0;
 138     j_coord_offsetA = 0;
 139     j_coord_offsetB = 0;
 140     j_coord_offsetC = 0;
 141     j_coord_offsetD = 0;
 142
 143     outeriter        = 0;
 144     inneriter        = 0;
 145
 146     for(iidx=0;iidx<4*DIM;iidx++)
 147     {
 148         scratch[iidx] = 0.0;
 149     }
 150
 151     /* Start outer loop over neighborlists */
 152     for(iidx=0; iidx<nri; iidx++)
 153     {
 154         /* Load shift vector for this list */
 155         i_shift_offset   = DIM*shiftidx[iidx];
 156
 157         /* Load limits for loop over neighbors */
 158         j_index_start    = jindex[iidx];
 159         j_index_end      = jindex[iidx+1];
 160
 161         /* Get outer coordinate index */
 162         inr              = iinr[iidx];
 163         i_coord_offset   = DIM*inr;
 164
 165         /* Load i particle coords and add shift vector */
 166         gmx_mm_load_shift_and_3rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
 167                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 168
 169         fix0             = _mm_setzero_ps();
 170         fiy0             = _mm_setzero_ps();
 171         fiz0             = _mm_setzero_ps();
 172         fix1             = _mm_setzero_ps();
 173         fiy1             = _mm_setzero_ps();
 174         fiz1             = _mm_setzero_ps();
 175         fix2             = _mm_setzero_ps();
 176         fiy2             = _mm_setzero_ps();
 177         fiz2             = _mm_setzero_ps();
 178
 179         /* Reset potential sums */
 180         velecsum         = _mm_setzero_ps();
 181         vvdwsum          = _mm_setzero_ps();
 182
 183         /* Start inner kernel loop */
 184         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 185         {
 186
 187             /* Get j neighbor index, and coordinate index */
 188             jnrA             = jjnr[jidx];
 189             jnrB             = jjnr[jidx+1];
 190             jnrC             = jjnr[jidx+2];
 191             jnrD             = jjnr[jidx+3];
 192             j_coord_offsetA  = DIM*jnrA;
 193             j_coord_offsetB  = DIM*jnrB;
 194             j_coord_offsetC  = DIM*jnrC;
 195             j_coord_offsetD  = DIM*jnrD;
 196
 197             /* load j atom coordinates */
 198             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 199                                               x+j_coord_offsetC,x+j_coord_offsetD,
 200                                               &jx0,&jy0,&jz0);
 201
 202             /* Calculate displacement vector */
 203             dx00             = _mm_sub_ps(ix0,jx0);
 204             dy00             = _mm_sub_ps(iy0,jy0);
 205             dz00             = _mm_sub_ps(iz0,jz0);
 206             dx10             = _mm_sub_ps(ix1,jx0);
 207             dy10             = _mm_sub_ps(iy1,jy0);
 208             dz10             = _mm_sub_ps(iz1,jz0);
 209             dx20             = _mm_sub_ps(ix2,jx0);
 210             dy20             = _mm_sub_ps(iy2,jy0);
 211             dz20             = _mm_sub_ps(iz2,jz0);
 212
 213             /* Calculate squared distance and things based on it */
 214             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 215             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 216             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 217
 218             rinv00           = avx128fma_invsqrt_f(rsq00);
 219             rinv10           = avx128fma_invsqrt_f(rsq10);
 220             rinv20           = avx128fma_invsqrt_f(rsq20);
 221
 222             /* Load parameters for j particles */
 223             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 224                                                               charge+jnrC+0,charge+jnrD+0);
 225             vdwjidx0A        = 2*vdwtype[jnrA+0];
 226             vdwjidx0B        = 2*vdwtype[jnrB+0];
 227             vdwjidx0C        = 2*vdwtype[jnrC+0];
 228             vdwjidx0D        = 2*vdwtype[jnrD+0];
 229
 230             fjx0             = _mm_setzero_ps();
 231             fjy0             = _mm_setzero_ps();
 232             fjz0             = _mm_setzero_ps();
 233
 234             /**************************
 235              * CALCULATE INTERACTIONS *
 236              **************************/
 237
 238             r00              = _mm_mul_ps(rsq00,rinv00);
 239
 240             /* Compute parameters for interactions between i and j atoms */
 241             qq00             = _mm_mul_ps(iq0,jq0);
 242             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 243                                          vdwparam+vdwioffset0+vdwjidx0B,
 244                                          vdwparam+vdwioffset0+vdwjidx0C,
 245                                          vdwparam+vdwioffset0+vdwjidx0D,
 246                                          &c6_00,&c12_00);
 247
 248             /* Calculate table index by multiplying r with table scale and truncate to integer */
 249             rt               = _mm_mul_ps(r00,vftabscale);
 250             vfitab           = _mm_cvttps_epi32(rt);
 251 #ifdef __XOP__
 252             vfeps            = _mm_frcz_ps(rt);
 253 #else
 254             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
 255 #endif
 256             twovfeps         = _mm_add_ps(vfeps,vfeps);
 257             vfitab           = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
 258
 259             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 260             Y                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
 261             F                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
 262             G                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
 263             H                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
 264             _MM_TRANSPOSE4_PS(Y,F,G,H);
 265             Fp               = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
 266             VV               = _mm_macc_ps(vfeps,Fp,Y);
 267             velec            = _mm_mul_ps(qq00,VV);
 268             FF               = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
 269             felec            = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq00,FF),_mm_mul_ps(vftabscale,rinv00)));
 270
 271             /* CUBIC SPLINE TABLE DISPERSION */
 272             vfitab           = _mm_add_epi32(vfitab,ifour);
 273             Y                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
 274             F                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
 275             G                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
 276             H                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
 277             _MM_TRANSPOSE4_PS(Y,F,G,H);
 278             Fp               = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
 279             VV               = _mm_macc_ps(vfeps,Fp,Y);
 280             vvdw6            = _mm_mul_ps(c6_00,VV);
 281             FF               = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
 282             fvdw6            = _mm_mul_ps(c6_00,FF);
 283
 284             /* CUBIC SPLINE TABLE REPULSION */
 285             vfitab           = _mm_add_epi32(vfitab,ifour);
 286             Y                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
 287             F                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
 288             G                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
 289             H                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
 290             _MM_TRANSPOSE4_PS(Y,F,G,H);
 291             Fp               = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
 292             VV               = _mm_macc_ps(vfeps,Fp,Y);
 293             vvdw12           = _mm_mul_ps(c12_00,VV);
 294             FF               = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
 295             fvdw12           = _mm_mul_ps(c12_00,FF);
 296             vvdw             = _mm_add_ps(vvdw12,vvdw6);
 297             fvdw             = _mm_xor_ps(signbit,_mm_mul_ps(_mm_add_ps(fvdw6,fvdw12),_mm_mul_ps(vftabscale,rinv00)));
 298
 299             /* Update potential sum for this i atom from the interaction with this j atom. */
 300             velecsum         = _mm_add_ps(velecsum,velec);
 301             vvdwsum          = _mm_add_ps(vvdwsum,vvdw);
 302
 303             fscal            = _mm_add_ps(felec,fvdw);
 304
 305              /* Update vectorial force */
 306             fix0             = _mm_macc_ps(dx00,fscal,fix0);
 307             fiy0             = _mm_macc_ps(dy00,fscal,fiy0);
 308             fiz0             = _mm_macc_ps(dz00,fscal,fiz0);
 309
 310             fjx0             = _mm_macc_ps(dx00,fscal,fjx0);
 311             fjy0             = _mm_macc_ps(dy00,fscal,fjy0);
 312             fjz0             = _mm_macc_ps(dz00,fscal,fjz0);
 313
 314             /**************************
 315              * CALCULATE INTERACTIONS *
 316              **************************/
 317
 318             r10              = _mm_mul_ps(rsq10,rinv10);
 319
 320             /* Compute parameters for interactions between i and j atoms */
 321             qq10             = _mm_mul_ps(iq1,jq0);
 322
 323             /* Calculate table index by multiplying r with table scale and truncate to integer */
 324             rt               = _mm_mul_ps(r10,vftabscale);
 325             vfitab           = _mm_cvttps_epi32(rt);
 326 #ifdef __XOP__
 327             vfeps            = _mm_frcz_ps(rt);
 328 #else
 329             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
 330 #endif
 331             twovfeps         = _mm_add_ps(vfeps,vfeps);
 332             vfitab           = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
 333
 334             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 335             Y                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
 336             F                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
 337             G                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
 338             H                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
 339             _MM_TRANSPOSE4_PS(Y,F,G,H);
 340             Fp               = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
 341             VV               = _mm_macc_ps(vfeps,Fp,Y);
 342             velec            = _mm_mul_ps(qq10,VV);
 343             FF               = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
 344             felec            = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq10,FF),_mm_mul_ps(vftabscale,rinv10)));
 345
 346             /* Update potential sum for this i atom from the interaction with this j atom. */
 347             velecsum         = _mm_add_ps(velecsum,velec);
 348
 349             fscal            = felec;
 350
 351              /* Update vectorial force */
 352             fix1             = _mm_macc_ps(dx10,fscal,fix1);
 353             fiy1             = _mm_macc_ps(dy10,fscal,fiy1);
 354             fiz1             = _mm_macc_ps(dz10,fscal,fiz1);
 355
 356             fjx0             = _mm_macc_ps(dx10,fscal,fjx0);
 357             fjy0             = _mm_macc_ps(dy10,fscal,fjy0);
 358             fjz0             = _mm_macc_ps(dz10,fscal,fjz0);
 359
 360             /**************************
 361              * CALCULATE INTERACTIONS *
 362              **************************/
 363
 364             r20              = _mm_mul_ps(rsq20,rinv20);
 365
 366             /* Compute parameters for interactions between i and j atoms */
 367             qq20             = _mm_mul_ps(iq2,jq0);
 368
 369             /* Calculate table index by multiplying r with table scale and truncate to integer */
 370             rt               = _mm_mul_ps(r20,vftabscale);
 371             vfitab           = _mm_cvttps_epi32(rt);
 372 #ifdef __XOP__
 373             vfeps            = _mm_frcz_ps(rt);
 374 #else
 375             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
 376 #endif
 377             twovfeps         = _mm_add_ps(vfeps,vfeps);
 378             vfitab           = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
 379
 380             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 381             Y                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
 382             F                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
 383             G                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
 384             H                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
 385             _MM_TRANSPOSE4_PS(Y,F,G,H);
 386             Fp               = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
 387             VV               = _mm_macc_ps(vfeps,Fp,Y);
 388             velec            = _mm_mul_ps(qq20,VV);
 389             FF               = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
 390             felec            = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq20,FF),_mm_mul_ps(vftabscale,rinv20)));
 391
 392             /* Update potential sum for this i atom from the interaction with this j atom. */
 393             velecsum         = _mm_add_ps(velecsum,velec);
 394
 395             fscal            = felec;
 396
 397              /* Update vectorial force */
 398             fix2             = _mm_macc_ps(dx20,fscal,fix2);
 399             fiy2             = _mm_macc_ps(dy20,fscal,fiy2);
 400             fiz2             = _mm_macc_ps(dz20,fscal,fiz2);
 401
 402             fjx0             = _mm_macc_ps(dx20,fscal,fjx0);
 403             fjy0             = _mm_macc_ps(dy20,fscal,fjy0);
 404             fjz0             = _mm_macc_ps(dz20,fscal,fjz0);
 405
 406             fjptrA             = f+j_coord_offsetA;
 407             fjptrB             = f+j_coord_offsetB;
 408             fjptrC             = f+j_coord_offsetC;
 409             fjptrD             = f+j_coord_offsetD;
 410
 411             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 412
 413             /* Inner loop uses 168 flops */
 414         }
 415
 416         if(jidx<j_index_end)
 417         {
 418
 419             /* Get j neighbor index, and coordinate index */
 420             jnrlistA         = jjnr[jidx];
 421             jnrlistB         = jjnr[jidx+1];
 422             jnrlistC         = jjnr[jidx+2];
 423             jnrlistD         = jjnr[jidx+3];
 424             /* Sign of each element will be negative for non-real atoms.
 425              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 426              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 427              */
 428             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 429             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 430             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 431             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 432             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 433             j_coord_offsetA  = DIM*jnrA;
 434             j_coord_offsetB  = DIM*jnrB;
 435             j_coord_offsetC  = DIM*jnrC;
 436             j_coord_offsetD  = DIM*jnrD;
 437
 438             /* load j atom coordinates */
 439             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 440                                               x+j_coord_offsetC,x+j_coord_offsetD,
 441                                               &jx0,&jy0,&jz0);
 442
 443             /* Calculate displacement vector */
 444             dx00             = _mm_sub_ps(ix0,jx0);
 445             dy00             = _mm_sub_ps(iy0,jy0);
 446             dz00             = _mm_sub_ps(iz0,jz0);
 447             dx10             = _mm_sub_ps(ix1,jx0);
 448             dy10             = _mm_sub_ps(iy1,jy0);
 449             dz10             = _mm_sub_ps(iz1,jz0);
 450             dx20             = _mm_sub_ps(ix2,jx0);
 451             dy20             = _mm_sub_ps(iy2,jy0);
 452             dz20             = _mm_sub_ps(iz2,jz0);
 453
 454             /* Calculate squared distance and things based on it */
 455             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 456             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 457             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 458
 459             rinv00           = avx128fma_invsqrt_f(rsq00);
 460             rinv10           = avx128fma_invsqrt_f(rsq10);
 461             rinv20           = avx128fma_invsqrt_f(rsq20);
 462
 463             /* Load parameters for j particles */
 464             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 465                                                               charge+jnrC+0,charge+jnrD+0);
 466             vdwjidx0A        = 2*vdwtype[jnrA+0];
 467             vdwjidx0B        = 2*vdwtype[jnrB+0];
 468             vdwjidx0C        = 2*vdwtype[jnrC+0];
 469             vdwjidx0D        = 2*vdwtype[jnrD+0];
 470
 471             fjx0             = _mm_setzero_ps();
 472             fjy0             = _mm_setzero_ps();
 473             fjz0             = _mm_setzero_ps();
 474
 475             /**************************
 476              * CALCULATE INTERACTIONS *
 477              **************************/
 478
 479             r00              = _mm_mul_ps(rsq00,rinv00);
 480             r00              = _mm_andnot_ps(dummy_mask,r00);
 481
 482             /* Compute parameters for interactions between i and j atoms */
 483             qq00             = _mm_mul_ps(iq0,jq0);
 484             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 485                                          vdwparam+vdwioffset0+vdwjidx0B,
 486                                          vdwparam+vdwioffset0+vdwjidx0C,
 487                                          vdwparam+vdwioffset0+vdwjidx0D,
 488                                          &c6_00,&c12_00);
 489
 490             /* Calculate table index by multiplying r with table scale and truncate to integer */
 491             rt               = _mm_mul_ps(r00,vftabscale);
 492             vfitab           = _mm_cvttps_epi32(rt);
 493 #ifdef __XOP__
 494             vfeps            = _mm_frcz_ps(rt);
 495 #else
 496             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
 497 #endif
 498             twovfeps         = _mm_add_ps(vfeps,vfeps);
 499             vfitab           = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
 500
 501             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 502             Y                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
 503             F                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
 504             G                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
 505             H                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
 506             _MM_TRANSPOSE4_PS(Y,F,G,H);
 507             Fp               = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
 508             VV               = _mm_macc_ps(vfeps,Fp,Y);
 509             velec            = _mm_mul_ps(qq00,VV);
 510             FF               = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
 511             felec            = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq00,FF),_mm_mul_ps(vftabscale,rinv00)));
 512
 513             /* CUBIC SPLINE TABLE DISPERSION */
 514             vfitab           = _mm_add_epi32(vfitab,ifour);
 515             Y                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
 516             F                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
 517             G                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
 518             H                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
 519             _MM_TRANSPOSE4_PS(Y,F,G,H);
 520             Fp               = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
 521             VV               = _mm_macc_ps(vfeps,Fp,Y);
 522             vvdw6            = _mm_mul_ps(c6_00,VV);
 523             FF               = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
 524             fvdw6            = _mm_mul_ps(c6_00,FF);
 525
 526             /* CUBIC SPLINE TABLE REPULSION */
 527             vfitab           = _mm_add_epi32(vfitab,ifour);
 528             Y                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
 529             F                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
 530             G                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
 531             H                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
 532             _MM_TRANSPOSE4_PS(Y,F,G,H);
 533             Fp               = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
 534             VV               = _mm_macc_ps(vfeps,Fp,Y);
 535             vvdw12           = _mm_mul_ps(c12_00,VV);
 536             FF               = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
 537             fvdw12           = _mm_mul_ps(c12_00,FF);
 538             vvdw             = _mm_add_ps(vvdw12,vvdw6);
 539             fvdw             = _mm_xor_ps(signbit,_mm_mul_ps(_mm_add_ps(fvdw6,fvdw12),_mm_mul_ps(vftabscale,rinv00)));
 540
 541             /* Update potential sum for this i atom from the interaction with this j atom. */
 542             velec            = _mm_andnot_ps(dummy_mask,velec);
 543             velecsum         = _mm_add_ps(velecsum,velec);
 544             vvdw             = _mm_andnot_ps(dummy_mask,vvdw);
 545             vvdwsum          = _mm_add_ps(vvdwsum,vvdw);
 546
 547             fscal            = _mm_add_ps(felec,fvdw);
 548
 549             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 550
 551              /* Update vectorial force */
 552             fix0             = _mm_macc_ps(dx00,fscal,fix0);
 553             fiy0             = _mm_macc_ps(dy00,fscal,fiy0);
 554             fiz0             = _mm_macc_ps(dz00,fscal,fiz0);
 555
 556             fjx0             = _mm_macc_ps(dx00,fscal,fjx0);
 557             fjy0             = _mm_macc_ps(dy00,fscal,fjy0);
 558             fjz0             = _mm_macc_ps(dz00,fscal,fjz0);
 559
 560             /**************************
 561              * CALCULATE INTERACTIONS *
 562              **************************/
 563
 564             r10              = _mm_mul_ps(rsq10,rinv10);
 565             r10              = _mm_andnot_ps(dummy_mask,r10);
 566
 567             /* Compute parameters for interactions between i and j atoms */
 568             qq10             = _mm_mul_ps(iq1,jq0);
 569
 570             /* Calculate table index by multiplying r with table scale and truncate to integer */
 571             rt               = _mm_mul_ps(r10,vftabscale);
 572             vfitab           = _mm_cvttps_epi32(rt);
 573 #ifdef __XOP__
 574             vfeps            = _mm_frcz_ps(rt);
 575 #else
 576             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
 577 #endif
 578             twovfeps         = _mm_add_ps(vfeps,vfeps);
 579             vfitab           = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
 580
 581             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 582             Y                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
 583             F                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
 584             G                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
 585             H                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
 586             _MM_TRANSPOSE4_PS(Y,F,G,H);
 587             Fp               = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
 588             VV               = _mm_macc_ps(vfeps,Fp,Y);
 589             velec            = _mm_mul_ps(qq10,VV);
 590             FF               = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
 591             felec            = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq10,FF),_mm_mul_ps(vftabscale,rinv10)));
 592
 593             /* Update potential sum for this i atom from the interaction with this j atom. */
 594             velec            = _mm_andnot_ps(dummy_mask,velec);
 595             velecsum         = _mm_add_ps(velecsum,velec);
 596
 597             fscal            = felec;
 598
 599             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 600
 601              /* Update vectorial force */
 602             fix1             = _mm_macc_ps(dx10,fscal,fix1);
 603             fiy1             = _mm_macc_ps(dy10,fscal,fiy1);
 604             fiz1             = _mm_macc_ps(dz10,fscal,fiz1);
 605
 606             fjx0             = _mm_macc_ps(dx10,fscal,fjx0);
 607             fjy0             = _mm_macc_ps(dy10,fscal,fjy0);
 608             fjz0             = _mm_macc_ps(dz10,fscal,fjz0);
 609
 610             /**************************
 611              * CALCULATE INTERACTIONS *
 612              **************************/
 613
 614             r20              = _mm_mul_ps(rsq20,rinv20);
 615             r20              = _mm_andnot_ps(dummy_mask,r20);
 616
 617             /* Compute parameters for interactions between i and j atoms */
 618             qq20             = _mm_mul_ps(iq2,jq0);
 619
 620             /* Calculate table index by multiplying r with table scale and truncate to integer */
 621             rt               = _mm_mul_ps(r20,vftabscale);
 622             vfitab           = _mm_cvttps_epi32(rt);
 623 #ifdef __XOP__
 624             vfeps            = _mm_frcz_ps(rt);
 625 #else
 626             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
 627 #endif
 628             twovfeps         = _mm_add_ps(vfeps,vfeps);
 629             vfitab           = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
 630
 631             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 632             Y                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
 633             F                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
 634             G                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
 635             H                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
 636             _MM_TRANSPOSE4_PS(Y,F,G,H);
 637             Fp               = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
 638             VV               = _mm_macc_ps(vfeps,Fp,Y);
 639             velec            = _mm_mul_ps(qq20,VV);
 640             FF               = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
 641             felec            = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq20,FF),_mm_mul_ps(vftabscale,rinv20)));
 642
 643             /* Update potential sum for this i atom from the interaction with this j atom. */
 644             velec            = _mm_andnot_ps(dummy_mask,velec);
 645             velecsum         = _mm_add_ps(velecsum,velec);
 646
 647             fscal            = felec;
 648
 649             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 650
 651              /* Update vectorial force */
 652             fix2             = _mm_macc_ps(dx20,fscal,fix2);
 653             fiy2             = _mm_macc_ps(dy20,fscal,fiy2);
 654             fiz2             = _mm_macc_ps(dz20,fscal,fiz2);
 655
 656             fjx0             = _mm_macc_ps(dx20,fscal,fjx0);
 657             fjy0             = _mm_macc_ps(dy20,fscal,fjy0);
 658             fjz0             = _mm_macc_ps(dz20,fscal,fjz0);
 659
 660             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 661             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 662             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 663             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 664
 665             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 666
 667             /* Inner loop uses 171 flops */
 668         }
 669
 670         /* End of innermost loop */
 671
 672         gmx_mm_update_iforce_3atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
 673                                               f+i_coord_offset,fshift+i_shift_offset);
 674
 675         ggid                        = gid[iidx];
 676         /* Update potential energies */
 677         gmx_mm_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid);
 678         gmx_mm_update_1pot_ps(vvdwsum,kernel_data->energygrp_vdw+ggid);
 679
 680         /* Increment number of inner iterations */
 681         inneriter                  += j_index_end - j_index_start;
 682
 683         /* Outer loop uses 20 flops */
 684     }
 685
 686     /* Increment number of outer iterations */
 687     outeriter        += nri;
 688
 689     /* Update outer/inner flops */
 690
 691     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3_VF,outeriter*20 + inneriter*171);
 692 }
 693 /*
 694  * Gromacs nonbonded kernel:   nb_kernel_ElecCSTab_VdwCSTab_GeomW3P1_F_avx_128_fma_single
 695  * Electrostatics interaction: CubicSplineTable
 696  * VdW interaction:            CubicSplineTable
 697  * Geometry:                   Water3-Particle
 698  * Calculate force/pot:        Force
 699  */
 700 void
 701 nb_kernel_ElecCSTab_VdwCSTab_GeomW3P1_F_avx_128_fma_single
 702                     (t_nblist                    * gmx_restrict       nlist,
 703                      rvec                        * gmx_restrict          xx,
 704                      rvec                        * gmx_restrict          ff,
 705                      struct t_forcerec           * gmx_restrict          fr,
 706                      t_mdatoms                   * gmx_restrict     mdatoms,
 707                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 708                      t_nrnb                      * gmx_restrict        nrnb)
 709 {
 710     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 711      * just 0 for non-waters.
 712      * Suffixes A,B,C,D refer to j loop unrolling done with AVX_128, e.g. for the four different
 713      * jnr indices corresponding to data put in the four positions in the SIMD register.
 714      */
 715     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 716     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 717     int              jnrA,jnrB,jnrC,jnrD;
 718     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
 719     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 720     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 721     real             rcutoff_scalar;
 722     real             *shiftvec,*fshift,*x,*f;
 723     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
 724     real             scratch[4*DIM];
 725     __m128           fscal,rcutoff,rcutoff2,jidxall;
 726     int              vdwioffset0;
 727     __m128           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 728     int              vdwioffset1;
 729     __m128           ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
 730     int              vdwioffset2;
 731     __m128           ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
 732     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
 733     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 734     __m128           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 735     __m128           dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
 736     __m128           dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
 737     __m128           velec,felec,velecsum,facel,crf,krf,krf2;
 738     real             *charge;
 739     int              nvdwtype;
 740     __m128           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 741     int              *vdwtype;
 742     real             *vdwparam;
 743     __m128           one_sixth   = _mm_set1_ps(1.0/6.0);
 744     __m128           one_twelfth = _mm_set1_ps(1.0/12.0);
 745     __m128i          vfitab;
 746     __m128i          ifour       = _mm_set1_epi32(4);
 747     __m128           rt,vfeps,twovfeps,vftabscale,Y,F,G,H,Fp,VV,FF;
 748     real             *vftab;
 749     __m128           dummy_mask,cutoff_mask;
 750     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
 751     __m128           one     = _mm_set1_ps(1.0);
 752     __m128           two     = _mm_set1_ps(2.0);
 753     x                = xx[0];
 754     f                = ff[0];
 755
 756     nri              = nlist->nri;
 757     iinr             = nlist->iinr;
 758     jindex           = nlist->jindex;
 759     jjnr             = nlist->jjnr;
 760     shiftidx         = nlist->shift;
 761     gid              = nlist->gid;
 762     shiftvec         = fr->shift_vec[0];
 763     fshift           = fr->fshift[0];
 764     facel            = _mm_set1_ps(fr->ic->epsfac);
 765     charge           = mdatoms->chargeA;
 766     nvdwtype         = fr->ntype;
 767     vdwparam         = fr->nbfp;
 768     vdwtype          = mdatoms->typeA;
 769
 770     vftab            = kernel_data->table_elec_vdw->data;
 771     vftabscale       = _mm_set1_ps(kernel_data->table_elec_vdw->scale);
 772
 773     /* Setup water-specific parameters */
 774     inr              = nlist->iinr[0];
 775     iq0              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+0]));
 776     iq1              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+1]));
 777     iq2              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+2]));
 778     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 779
 780     /* Avoid stupid compiler warnings */
 781     jnrA = jnrB = jnrC = jnrD = 0;
 782     j_coord_offsetA = 0;
 783     j_coord_offsetB = 0;
 784     j_coord_offsetC = 0;
 785     j_coord_offsetD = 0;
 786
 787     outeriter        = 0;
 788     inneriter        = 0;
 789
 790     for(iidx=0;iidx<4*DIM;iidx++)
 791     {
 792         scratch[iidx] = 0.0;
 793     }
 794
 795     /* Start outer loop over neighborlists */
 796     for(iidx=0; iidx<nri; iidx++)
 797     {
 798         /* Load shift vector for this list */
 799         i_shift_offset   = DIM*shiftidx[iidx];
 800
 801         /* Load limits for loop over neighbors */
 802         j_index_start    = jindex[iidx];
 803         j_index_end      = jindex[iidx+1];
 804
 805         /* Get outer coordinate index */
 806         inr              = iinr[iidx];
 807         i_coord_offset   = DIM*inr;
 808
 809         /* Load i particle coords and add shift vector */
 810         gmx_mm_load_shift_and_3rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
 811                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 812
 813         fix0             = _mm_setzero_ps();
 814         fiy0             = _mm_setzero_ps();
 815         fiz0             = _mm_setzero_ps();
 816         fix1             = _mm_setzero_ps();
 817         fiy1             = _mm_setzero_ps();
 818         fiz1             = _mm_setzero_ps();
 819         fix2             = _mm_setzero_ps();
 820         fiy2             = _mm_setzero_ps();
 821         fiz2             = _mm_setzero_ps();
 822
 823         /* Start inner kernel loop */
 824         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 825         {
 826
 827             /* Get j neighbor index, and coordinate index */
 828             jnrA             = jjnr[jidx];
 829             jnrB             = jjnr[jidx+1];
 830             jnrC             = jjnr[jidx+2];
 831             jnrD             = jjnr[jidx+3];
 832             j_coord_offsetA  = DIM*jnrA;
 833             j_coord_offsetB  = DIM*jnrB;
 834             j_coord_offsetC  = DIM*jnrC;
 835             j_coord_offsetD  = DIM*jnrD;
 836
 837             /* load j atom coordinates */
 838             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 839                                               x+j_coord_offsetC,x+j_coord_offsetD,
 840                                               &jx0,&jy0,&jz0);
 841
 842             /* Calculate displacement vector */
 843             dx00             = _mm_sub_ps(ix0,jx0);
 844             dy00             = _mm_sub_ps(iy0,jy0);
 845             dz00             = _mm_sub_ps(iz0,jz0);
 846             dx10             = _mm_sub_ps(ix1,jx0);
 847             dy10             = _mm_sub_ps(iy1,jy0);
 848             dz10             = _mm_sub_ps(iz1,jz0);
 849             dx20             = _mm_sub_ps(ix2,jx0);
 850             dy20             = _mm_sub_ps(iy2,jy0);
 851             dz20             = _mm_sub_ps(iz2,jz0);
 852
 853             /* Calculate squared distance and things based on it */
 854             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 855             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 856             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 857
 858             rinv00           = avx128fma_invsqrt_f(rsq00);
 859             rinv10           = avx128fma_invsqrt_f(rsq10);
 860             rinv20           = avx128fma_invsqrt_f(rsq20);
 861
 862             /* Load parameters for j particles */
 863             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 864                                                               charge+jnrC+0,charge+jnrD+0);
 865             vdwjidx0A        = 2*vdwtype[jnrA+0];
 866             vdwjidx0B        = 2*vdwtype[jnrB+0];
 867             vdwjidx0C        = 2*vdwtype[jnrC+0];
 868             vdwjidx0D        = 2*vdwtype[jnrD+0];
 869
 870             fjx0             = _mm_setzero_ps();
 871             fjy0             = _mm_setzero_ps();
 872             fjz0             = _mm_setzero_ps();
 873
 874             /**************************
 875              * CALCULATE INTERACTIONS *
 876              **************************/
 877
 878             r00              = _mm_mul_ps(rsq00,rinv00);
 879
 880             /* Compute parameters for interactions between i and j atoms */
 881             qq00             = _mm_mul_ps(iq0,jq0);
 882             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 883                                          vdwparam+vdwioffset0+vdwjidx0B,
 884                                          vdwparam+vdwioffset0+vdwjidx0C,
 885                                          vdwparam+vdwioffset0+vdwjidx0D,
 886                                          &c6_00,&c12_00);
 887
 888             /* Calculate table index by multiplying r with table scale and truncate to integer */
 889             rt               = _mm_mul_ps(r00,vftabscale);
 890             vfitab           = _mm_cvttps_epi32(rt);
 891 #ifdef __XOP__
 892             vfeps            = _mm_frcz_ps(rt);
 893 #else
 894             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
 895 #endif
 896             twovfeps         = _mm_add_ps(vfeps,vfeps);
 897             vfitab           = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
 898
 899             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 900             Y                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
 901             F                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
 902             G                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
 903             H                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
 904             _MM_TRANSPOSE4_PS(Y,F,G,H);
 905             Fp               = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
 906             FF               = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
 907             felec            = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq00,FF),_mm_mul_ps(vftabscale,rinv00)));
 908
 909             /* CUBIC SPLINE TABLE DISPERSION */
 910             vfitab           = _mm_add_epi32(vfitab,ifour);
 911             Y                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
 912             F                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
 913             G                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
 914             H                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
 915             _MM_TRANSPOSE4_PS(Y,F,G,H);
 916             Fp               = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
 917             FF               = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
 918             fvdw6            = _mm_mul_ps(c6_00,FF);
 919
 920             /* CUBIC SPLINE TABLE REPULSION */
 921             vfitab           = _mm_add_epi32(vfitab,ifour);
 922             Y                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
 923             F                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
 924             G                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
 925             H                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
 926             _MM_TRANSPOSE4_PS(Y,F,G,H);
 927             Fp               = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
 928             FF               = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
 929             fvdw12           = _mm_mul_ps(c12_00,FF);
 930             fvdw             = _mm_xor_ps(signbit,_mm_mul_ps(_mm_add_ps(fvdw6,fvdw12),_mm_mul_ps(vftabscale,rinv00)));
 931
 932             fscal            = _mm_add_ps(felec,fvdw);
 933
 934              /* Update vectorial force */
 935             fix0             = _mm_macc_ps(dx00,fscal,fix0);
 936             fiy0             = _mm_macc_ps(dy00,fscal,fiy0);
 937             fiz0             = _mm_macc_ps(dz00,fscal,fiz0);
 938
 939             fjx0             = _mm_macc_ps(dx00,fscal,fjx0);
 940             fjy0             = _mm_macc_ps(dy00,fscal,fjy0);
 941             fjz0             = _mm_macc_ps(dz00,fscal,fjz0);
 942
 943             /**************************
 944              * CALCULATE INTERACTIONS *
 945              **************************/
 946
 947             r10              = _mm_mul_ps(rsq10,rinv10);
 948
 949             /* Compute parameters for interactions between i and j atoms */
 950             qq10             = _mm_mul_ps(iq1,jq0);
 951
 952             /* Calculate table index by multiplying r with table scale and truncate to integer */
 953             rt               = _mm_mul_ps(r10,vftabscale);
 954             vfitab           = _mm_cvttps_epi32(rt);
 955 #ifdef __XOP__
 956             vfeps            = _mm_frcz_ps(rt);
 957 #else
 958             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
 959 #endif
 960             twovfeps         = _mm_add_ps(vfeps,vfeps);
 961             vfitab           = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
 962
 963             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 964             Y                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
 965             F                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
 966             G                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
 967             H                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
 968             _MM_TRANSPOSE4_PS(Y,F,G,H);
 969             Fp               = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
 970             FF               = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
 971             felec            = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq10,FF),_mm_mul_ps(vftabscale,rinv10)));
 972
 973             fscal            = felec;
 974
 975              /* Update vectorial force */
 976             fix1             = _mm_macc_ps(dx10,fscal,fix1);
 977             fiy1             = _mm_macc_ps(dy10,fscal,fiy1);
 978             fiz1             = _mm_macc_ps(dz10,fscal,fiz1);
 979
 980             fjx0             = _mm_macc_ps(dx10,fscal,fjx0);
 981             fjy0             = _mm_macc_ps(dy10,fscal,fjy0);
 982             fjz0             = _mm_macc_ps(dz10,fscal,fjz0);
 983
 984             /**************************
 985              * CALCULATE INTERACTIONS *
 986              **************************/
 987
 988             r20              = _mm_mul_ps(rsq20,rinv20);
 989
 990             /* Compute parameters for interactions between i and j atoms */
 991             qq20             = _mm_mul_ps(iq2,jq0);
 992
 993             /* Calculate table index by multiplying r with table scale and truncate to integer */
 994             rt               = _mm_mul_ps(r20,vftabscale);
 995             vfitab           = _mm_cvttps_epi32(rt);
 996 #ifdef __XOP__
 997             vfeps            = _mm_frcz_ps(rt);
 998 #else
 999             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
1000 #endif
1001             twovfeps         = _mm_add_ps(vfeps,vfeps);
1002             vfitab           = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
1003
1004             /* CUBIC SPLINE TABLE ELECTROSTATICS */
1005             Y                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
1006             F                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
1007             G                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
1008             H                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
1009             _MM_TRANSPOSE4_PS(Y,F,G,H);
1010             Fp               = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
1011             FF               = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
1012             felec            = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq20,FF),_mm_mul_ps(vftabscale,rinv20)));
1013
1014             fscal            = felec;
1015
1016              /* Update vectorial force */
1017             fix2             = _mm_macc_ps(dx20,fscal,fix2);
1018             fiy2             = _mm_macc_ps(dy20,fscal,fiy2);
1019             fiz2             = _mm_macc_ps(dz20,fscal,fiz2);
1020
1021             fjx0             = _mm_macc_ps(dx20,fscal,fjx0);
1022             fjy0             = _mm_macc_ps(dy20,fscal,fjy0);
1023             fjz0             = _mm_macc_ps(dz20,fscal,fjz0);
1024
1025             fjptrA             = f+j_coord_offsetA;
1026             fjptrB             = f+j_coord_offsetB;
1027             fjptrC             = f+j_coord_offsetC;
1028             fjptrD             = f+j_coord_offsetD;
1029
1030             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
1031
1032             /* Inner loop uses 148 flops */
1033         }
1034
1035         if(jidx<j_index_end)
1036         {
1037
1038             /* Get j neighbor index, and coordinate index */
1039             jnrlistA         = jjnr[jidx];
1040             jnrlistB         = jjnr[jidx+1];
1041             jnrlistC         = jjnr[jidx+2];
1042             jnrlistD         = jjnr[jidx+3];
1043             /* Sign of each element will be negative for non-real atoms.
1044              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
1045              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
1046              */
1047             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
1048             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
1049             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
1050             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
1051             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
1052             j_coord_offsetA  = DIM*jnrA;
1053             j_coord_offsetB  = DIM*jnrB;
1054             j_coord_offsetC  = DIM*jnrC;
1055             j_coord_offsetD  = DIM*jnrD;
1056
1057             /* load j atom coordinates */
1058             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
1059                                               x+j_coord_offsetC,x+j_coord_offsetD,
1060                                               &jx0,&jy0,&jz0);
1061
1062             /* Calculate displacement vector */
1063             dx00             = _mm_sub_ps(ix0,jx0);
1064             dy00             = _mm_sub_ps(iy0,jy0);
1065             dz00             = _mm_sub_ps(iz0,jz0);
1066             dx10             = _mm_sub_ps(ix1,jx0);
1067             dy10             = _mm_sub_ps(iy1,jy0);
1068             dz10             = _mm_sub_ps(iz1,jz0);
1069             dx20             = _mm_sub_ps(ix2,jx0);
1070             dy20             = _mm_sub_ps(iy2,jy0);
1071             dz20             = _mm_sub_ps(iz2,jz0);
1072
1073             /* Calculate squared distance and things based on it */
1074             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
1075             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
1076             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
1077
1078             rinv00           = avx128fma_invsqrt_f(rsq00);
1079             rinv10           = avx128fma_invsqrt_f(rsq10);
1080             rinv20           = avx128fma_invsqrt_f(rsq20);
1081
1082             /* Load parameters for j particles */
1083             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
1084                                                               charge+jnrC+0,charge+jnrD+0);
1085             vdwjidx0A        = 2*vdwtype[jnrA+0];
1086             vdwjidx0B        = 2*vdwtype[jnrB+0];
1087             vdwjidx0C        = 2*vdwtype[jnrC+0];
1088             vdwjidx0D        = 2*vdwtype[jnrD+0];
1089
1090             fjx0             = _mm_setzero_ps();
1091             fjy0             = _mm_setzero_ps();
1092             fjz0             = _mm_setzero_ps();
1093
1094             /**************************
1095              * CALCULATE INTERACTIONS *
1096              **************************/
1097
1098             r00              = _mm_mul_ps(rsq00,rinv00);
1099             r00              = _mm_andnot_ps(dummy_mask,r00);
1100
1101             /* Compute parameters for interactions between i and j atoms */
1102             qq00             = _mm_mul_ps(iq0,jq0);
1103             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
1104                                          vdwparam+vdwioffset0+vdwjidx0B,
1105                                          vdwparam+vdwioffset0+vdwjidx0C,
1106                                          vdwparam+vdwioffset0+vdwjidx0D,
1107                                          &c6_00,&c12_00);
1108
1109             /* Calculate table index by multiplying r with table scale and truncate to integer */
1110             rt               = _mm_mul_ps(r00,vftabscale);
1111             vfitab           = _mm_cvttps_epi32(rt);
1112 #ifdef __XOP__
1113             vfeps            = _mm_frcz_ps(rt);
1114 #else
1115             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
1116 #endif
1117             twovfeps         = _mm_add_ps(vfeps,vfeps);
1118             vfitab           = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
1119
1120             /* CUBIC SPLINE TABLE ELECTROSTATICS */
1121             Y                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
1122             F                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
1123             G                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
1124             H                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
1125             _MM_TRANSPOSE4_PS(Y,F,G,H);
1126             Fp               = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
1127             FF               = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
1128             felec            = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq00,FF),_mm_mul_ps(vftabscale,rinv00)));
1129
1130             /* CUBIC SPLINE TABLE DISPERSION */
1131             vfitab           = _mm_add_epi32(vfitab,ifour);
1132             Y                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
1133             F                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
1134             G                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
1135             H                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
1136             _MM_TRANSPOSE4_PS(Y,F,G,H);
1137             Fp               = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
1138             FF               = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
1139             fvdw6            = _mm_mul_ps(c6_00,FF);
1140
1141             /* CUBIC SPLINE TABLE REPULSION */
1142             vfitab           = _mm_add_epi32(vfitab,ifour);
1143             Y                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
1144             F                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
1145             G                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
1146             H                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
1147             _MM_TRANSPOSE4_PS(Y,F,G,H);
1148             Fp               = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
1149             FF               = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
1150             fvdw12           = _mm_mul_ps(c12_00,FF);
1151             fvdw             = _mm_xor_ps(signbit,_mm_mul_ps(_mm_add_ps(fvdw6,fvdw12),_mm_mul_ps(vftabscale,rinv00)));
1152
1153             fscal            = _mm_add_ps(felec,fvdw);
1154
1155             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1156
1157              /* Update vectorial force */
1158             fix0             = _mm_macc_ps(dx00,fscal,fix0);
1159             fiy0             = _mm_macc_ps(dy00,fscal,fiy0);
1160             fiz0             = _mm_macc_ps(dz00,fscal,fiz0);
1161
1162             fjx0             = _mm_macc_ps(dx00,fscal,fjx0);
1163             fjy0             = _mm_macc_ps(dy00,fscal,fjy0);
1164             fjz0             = _mm_macc_ps(dz00,fscal,fjz0);
1165
1166             /**************************
1167              * CALCULATE INTERACTIONS *
1168              **************************/
1169
1170             r10              = _mm_mul_ps(rsq10,rinv10);
1171             r10              = _mm_andnot_ps(dummy_mask,r10);
1172
1173             /* Compute parameters for interactions between i and j atoms */
1174             qq10             = _mm_mul_ps(iq1,jq0);
1175
1176             /* Calculate table index by multiplying r with table scale and truncate to integer */
1177             rt               = _mm_mul_ps(r10,vftabscale);
1178             vfitab           = _mm_cvttps_epi32(rt);
1179 #ifdef __XOP__
1180             vfeps            = _mm_frcz_ps(rt);
1181 #else
1182             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
1183 #endif
1184             twovfeps         = _mm_add_ps(vfeps,vfeps);
1185             vfitab           = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
1186
1187             /* CUBIC SPLINE TABLE ELECTROSTATICS */
1188             Y                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
1189             F                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
1190             G                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
1191             H                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
1192             _MM_TRANSPOSE4_PS(Y,F,G,H);
1193             Fp               = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
1194             FF               = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
1195             felec            = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq10,FF),_mm_mul_ps(vftabscale,rinv10)));
1196
1197             fscal            = felec;
1198
1199             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1200
1201              /* Update vectorial force */
1202             fix1             = _mm_macc_ps(dx10,fscal,fix1);
1203             fiy1             = _mm_macc_ps(dy10,fscal,fiy1);
1204             fiz1             = _mm_macc_ps(dz10,fscal,fiz1);
1205
1206             fjx0             = _mm_macc_ps(dx10,fscal,fjx0);
1207             fjy0             = _mm_macc_ps(dy10,fscal,fjy0);
1208             fjz0             = _mm_macc_ps(dz10,fscal,fjz0);
1209
1210             /**************************
1211              * CALCULATE INTERACTIONS *
1212              **************************/
1213
1214             r20              = _mm_mul_ps(rsq20,rinv20);
1215             r20              = _mm_andnot_ps(dummy_mask,r20);
1216
1217             /* Compute parameters for interactions between i and j atoms */
1218             qq20             = _mm_mul_ps(iq2,jq0);
1219
1220             /* Calculate table index by multiplying r with table scale and truncate to integer */
1221             rt               = _mm_mul_ps(r20,vftabscale);
1222             vfitab           = _mm_cvttps_epi32(rt);
1223 #ifdef __XOP__
1224             vfeps            = _mm_frcz_ps(rt);
1225 #else
1226             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
1227 #endif
1228             twovfeps         = _mm_add_ps(vfeps,vfeps);
1229             vfitab           = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
1230
1231             /* CUBIC SPLINE TABLE ELECTROSTATICS */
1232             Y                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
1233             F                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
1234             G                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
1235             H                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
1236             _MM_TRANSPOSE4_PS(Y,F,G,H);
1237             Fp               = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
1238             FF               = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
1239             felec            = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq20,FF),_mm_mul_ps(vftabscale,rinv20)));
1240
1241             fscal            = felec;
1242
1243             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1244
1245              /* Update vectorial force */
1246             fix2             = _mm_macc_ps(dx20,fscal,fix2);
1247             fiy2             = _mm_macc_ps(dy20,fscal,fiy2);
1248             fiz2             = _mm_macc_ps(dz20,fscal,fiz2);
1249
1250             fjx0             = _mm_macc_ps(dx20,fscal,fjx0);
1251             fjy0             = _mm_macc_ps(dy20,fscal,fjy0);
1252             fjz0             = _mm_macc_ps(dz20,fscal,fjz0);
1253
1254             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
1255             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
1256             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
1257             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
1258
1259             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
1260
1261             /* Inner loop uses 151 flops */
1262         }
1263
1264         /* End of innermost loop */
1265
1266         gmx_mm_update_iforce_3atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
1267                                               f+i_coord_offset,fshift+i_shift_offset);
1268
1269         /* Increment number of inner iterations */
1270         inneriter                  += j_index_end - j_index_start;
1271
1272         /* Outer loop uses 18 flops */
1273     }
1274
1275     /* Increment number of outer iterations */
1276     outeriter        += nri;
1277
1278     /* Update outer/inner flops */
1279
1280     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3_F,outeriter*18 + inneriter*151);
1281 }