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, 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 "config.h"
  39
  40 #include <math.h>
  41
  42 #include "../nb_kernel.h"
  43 #include "gromacs/legacyheaders/types/simple.h"
  44 #include "gromacs/math/vec.h"
  45 #include "gromacs/legacyheaders/nrnb.h"
  46
  47 #include "gromacs/simd/math_x86_avx_128_fma_single.h"
  48 #include "kernelutil_x86_avx_128_fma_single.h"
  49
  50 /*
  51  * Gromacs nonbonded kernel:   nb_kernel_ElecCSTab_VdwCSTab_GeomW3P1_VF_avx_128_fma_single
  52  * Electrostatics interaction: CubicSplineTable
  53  * VdW interaction:            CubicSplineTable
  54  * Geometry:                   Water3-Particle
  55  * Calculate force/pot:        PotentialAndForce
  56  */
  57 void
  58 nb_kernel_ElecCSTab_VdwCSTab_GeomW3P1_VF_avx_128_fma_single
  59                     (t_nblist                    * gmx_restrict       nlist,
  60                      rvec                        * gmx_restrict          xx,
  61                      rvec                        * gmx_restrict          ff,
  62                      t_forcerec                  * gmx_restrict          fr,
  63                      t_mdatoms                   * gmx_restrict     mdatoms,
  64                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
  65                      t_nrnb                      * gmx_restrict        nrnb)
  66 {
  67     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
  68      * just 0 for non-waters.
  69      * Suffixes A,B,C,D refer to j loop unrolling done with AVX_128, e.g. for the four different
  70      * jnr indices corresponding to data put in the four positions in the SIMD register.
  71      */
  72     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
  73     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
  74     int              jnrA,jnrB,jnrC,jnrD;
  75     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
  76     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
  77     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
  78     real             rcutoff_scalar;
  79     real             *shiftvec,*fshift,*x,*f;
  80     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
  81     real             scratch[4*DIM];
  82     __m128           fscal,rcutoff,rcutoff2,jidxall;
  83     int              vdwioffset0;
  84     __m128           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
  85     int              vdwioffset1;
  86     __m128           ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
  87     int              vdwioffset2;
  88     __m128           ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
  89     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
  90     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  91     __m128           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  92     __m128           dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
  93     __m128           dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
  94     __m128           velec,felec,velecsum,facel,crf,krf,krf2;
  95     real             *charge;
  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     vftab            = kernel_data->table_elec_vdw->data;
 128     vftabscale       = _mm_set1_ps(kernel_data->table_elec_vdw->scale);
 129
 130     /* Setup water-specific parameters */
 131     inr              = nlist->iinr[0];
 132     iq0              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+0]));
 133     iq1              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+1]));
 134     iq2              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+2]));
 135     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 136
 137     /* Avoid stupid compiler warnings */
 138     jnrA = jnrB = jnrC = jnrD = 0;
 139     j_coord_offsetA = 0;
 140     j_coord_offsetB = 0;
 141     j_coord_offsetC = 0;
 142     j_coord_offsetD = 0;
 143
 144     outeriter        = 0;
 145     inneriter        = 0;
 146
 147     for(iidx=0;iidx<4*DIM;iidx++)
 148     {
 149         scratch[iidx] = 0.0;
 150     }
 151
 152     /* Start outer loop over neighborlists */
 153     for(iidx=0; iidx<nri; iidx++)
 154     {
 155         /* Load shift vector for this list */
 156         i_shift_offset   = DIM*shiftidx[iidx];
 157
 158         /* Load limits for loop over neighbors */
 159         j_index_start    = jindex[iidx];
 160         j_index_end      = jindex[iidx+1];
 161
 162         /* Get outer coordinate index */
 163         inr              = iinr[iidx];
 164         i_coord_offset   = DIM*inr;
 165
 166         /* Load i particle coords and add shift vector */
 167         gmx_mm_load_shift_and_3rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
 168                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 169
 170         fix0             = _mm_setzero_ps();
 171         fiy0             = _mm_setzero_ps();
 172         fiz0             = _mm_setzero_ps();
 173         fix1             = _mm_setzero_ps();
 174         fiy1             = _mm_setzero_ps();
 175         fiz1             = _mm_setzero_ps();
 176         fix2             = _mm_setzero_ps();
 177         fiy2             = _mm_setzero_ps();
 178         fiz2             = _mm_setzero_ps();
 179
 180         /* Reset potential sums */
 181         velecsum         = _mm_setzero_ps();
 182         vvdwsum          = _mm_setzero_ps();
 183
 184         /* Start inner kernel loop */
 185         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 186         {
 187
 188             /* Get j neighbor index, and coordinate index */
 189             jnrA             = jjnr[jidx];
 190             jnrB             = jjnr[jidx+1];
 191             jnrC             = jjnr[jidx+2];
 192             jnrD             = jjnr[jidx+3];
 193             j_coord_offsetA  = DIM*jnrA;
 194             j_coord_offsetB  = DIM*jnrB;
 195             j_coord_offsetC  = DIM*jnrC;
 196             j_coord_offsetD  = DIM*jnrD;
 197
 198             /* load j atom coordinates */
 199             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 200                                               x+j_coord_offsetC,x+j_coord_offsetD,
 201                                               &jx0,&jy0,&jz0);
 202
 203             /* Calculate displacement vector */
 204             dx00             = _mm_sub_ps(ix0,jx0);
 205             dy00             = _mm_sub_ps(iy0,jy0);
 206             dz00             = _mm_sub_ps(iz0,jz0);
 207             dx10             = _mm_sub_ps(ix1,jx0);
 208             dy10             = _mm_sub_ps(iy1,jy0);
 209             dz10             = _mm_sub_ps(iz1,jz0);
 210             dx20             = _mm_sub_ps(ix2,jx0);
 211             dy20             = _mm_sub_ps(iy2,jy0);
 212             dz20             = _mm_sub_ps(iz2,jz0);
 213
 214             /* Calculate squared distance and things based on it */
 215             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 216             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 217             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 218
 219             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 220             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 221             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 222
 223             /* Load parameters for j particles */
 224             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 225                                                               charge+jnrC+0,charge+jnrD+0);
 226             vdwjidx0A        = 2*vdwtype[jnrA+0];
 227             vdwjidx0B        = 2*vdwtype[jnrB+0];
 228             vdwjidx0C        = 2*vdwtype[jnrC+0];
 229             vdwjidx0D        = 2*vdwtype[jnrD+0];
 230
 231             fjx0             = _mm_setzero_ps();
 232             fjy0             = _mm_setzero_ps();
 233             fjz0             = _mm_setzero_ps();
 234
 235             /**************************
 236              * CALCULATE INTERACTIONS *
 237              **************************/
 238
 239             r00              = _mm_mul_ps(rsq00,rinv00);
 240
 241             /* Compute parameters for interactions between i and j atoms */
 242             qq00             = _mm_mul_ps(iq0,jq0);
 243             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 244                                          vdwparam+vdwioffset0+vdwjidx0B,
 245                                          vdwparam+vdwioffset0+vdwjidx0C,
 246                                          vdwparam+vdwioffset0+vdwjidx0D,
 247                                          &c6_00,&c12_00);
 248
 249             /* Calculate table index by multiplying r with table scale and truncate to integer */
 250             rt               = _mm_mul_ps(r00,vftabscale);
 251             vfitab           = _mm_cvttps_epi32(rt);
 252 #ifdef __XOP__
 253             vfeps            = _mm_frcz_ps(rt);
 254 #else
 255             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
 256 #endif
 257             twovfeps         = _mm_add_ps(vfeps,vfeps);
 258             vfitab           = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
 259
 260             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 261             Y                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
 262             F                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
 263             G                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
 264             H                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
 265             _MM_TRANSPOSE4_PS(Y,F,G,H);
 266             Fp               = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
 267             VV               = _mm_macc_ps(vfeps,Fp,Y);
 268             velec            = _mm_mul_ps(qq00,VV);
 269             FF               = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
 270             felec            = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq00,FF),_mm_mul_ps(vftabscale,rinv00)));
 271
 272             /* CUBIC SPLINE TABLE DISPERSION */
 273             vfitab           = _mm_add_epi32(vfitab,ifour);
 274             Y                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
 275             F                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
 276             G                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
 277             H                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
 278             _MM_TRANSPOSE4_PS(Y,F,G,H);
 279             Fp               = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
 280             VV               = _mm_macc_ps(vfeps,Fp,Y);
 281             vvdw6            = _mm_mul_ps(c6_00,VV);
 282             FF               = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
 283             fvdw6            = _mm_mul_ps(c6_00,FF);
 284
 285             /* CUBIC SPLINE TABLE REPULSION */
 286             vfitab           = _mm_add_epi32(vfitab,ifour);
 287             Y                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
 288             F                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
 289             G                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
 290             H                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
 291             _MM_TRANSPOSE4_PS(Y,F,G,H);
 292             Fp               = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
 293             VV               = _mm_macc_ps(vfeps,Fp,Y);
 294             vvdw12           = _mm_mul_ps(c12_00,VV);
 295             FF               = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
 296             fvdw12           = _mm_mul_ps(c12_00,FF);
 297             vvdw             = _mm_add_ps(vvdw12,vvdw6);
 298             fvdw             = _mm_xor_ps(signbit,_mm_mul_ps(_mm_add_ps(fvdw6,fvdw12),_mm_mul_ps(vftabscale,rinv00)));
 299
 300             /* Update potential sum for this i atom from the interaction with this j atom. */
 301             velecsum         = _mm_add_ps(velecsum,velec);
 302             vvdwsum          = _mm_add_ps(vvdwsum,vvdw);
 303
 304             fscal            = _mm_add_ps(felec,fvdw);
 305
 306              /* Update vectorial force */
 307             fix0             = _mm_macc_ps(dx00,fscal,fix0);
 308             fiy0             = _mm_macc_ps(dy00,fscal,fiy0);
 309             fiz0             = _mm_macc_ps(dz00,fscal,fiz0);
 310
 311             fjx0             = _mm_macc_ps(dx00,fscal,fjx0);
 312             fjy0             = _mm_macc_ps(dy00,fscal,fjy0);
 313             fjz0             = _mm_macc_ps(dz00,fscal,fjz0);
 314
 315             /**************************
 316              * CALCULATE INTERACTIONS *
 317              **************************/
 318
 319             r10              = _mm_mul_ps(rsq10,rinv10);
 320
 321             /* Compute parameters for interactions between i and j atoms */
 322             qq10             = _mm_mul_ps(iq1,jq0);
 323
 324             /* Calculate table index by multiplying r with table scale and truncate to integer */
 325             rt               = _mm_mul_ps(r10,vftabscale);
 326             vfitab           = _mm_cvttps_epi32(rt);
 327 #ifdef __XOP__
 328             vfeps            = _mm_frcz_ps(rt);
 329 #else
 330             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
 331 #endif
 332             twovfeps         = _mm_add_ps(vfeps,vfeps);
 333             vfitab           = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
 334
 335             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 336             Y                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
 337             F                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
 338             G                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
 339             H                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
 340             _MM_TRANSPOSE4_PS(Y,F,G,H);
 341             Fp               = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
 342             VV               = _mm_macc_ps(vfeps,Fp,Y);
 343             velec            = _mm_mul_ps(qq10,VV);
 344             FF               = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
 345             felec            = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq10,FF),_mm_mul_ps(vftabscale,rinv10)));
 346
 347             /* Update potential sum for this i atom from the interaction with this j atom. */
 348             velecsum         = _mm_add_ps(velecsum,velec);
 349
 350             fscal            = felec;
 351
 352              /* Update vectorial force */
 353             fix1             = _mm_macc_ps(dx10,fscal,fix1);
 354             fiy1             = _mm_macc_ps(dy10,fscal,fiy1);
 355             fiz1             = _mm_macc_ps(dz10,fscal,fiz1);
 356
 357             fjx0             = _mm_macc_ps(dx10,fscal,fjx0);
 358             fjy0             = _mm_macc_ps(dy10,fscal,fjy0);
 359             fjz0             = _mm_macc_ps(dz10,fscal,fjz0);
 360
 361             /**************************
 362              * CALCULATE INTERACTIONS *
 363              **************************/
 364
 365             r20              = _mm_mul_ps(rsq20,rinv20);
 366
 367             /* Compute parameters for interactions between i and j atoms */
 368             qq20             = _mm_mul_ps(iq2,jq0);
 369
 370             /* Calculate table index by multiplying r with table scale and truncate to integer */
 371             rt               = _mm_mul_ps(r20,vftabscale);
 372             vfitab           = _mm_cvttps_epi32(rt);
 373 #ifdef __XOP__
 374             vfeps            = _mm_frcz_ps(rt);
 375 #else
 376             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
 377 #endif
 378             twovfeps         = _mm_add_ps(vfeps,vfeps);
 379             vfitab           = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
 380
 381             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 382             Y                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
 383             F                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
 384             G                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
 385             H                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
 386             _MM_TRANSPOSE4_PS(Y,F,G,H);
 387             Fp               = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
 388             VV               = _mm_macc_ps(vfeps,Fp,Y);
 389             velec            = _mm_mul_ps(qq20,VV);
 390             FF               = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
 391             felec            = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq20,FF),_mm_mul_ps(vftabscale,rinv20)));
 392
 393             /* Update potential sum for this i atom from the interaction with this j atom. */
 394             velecsum         = _mm_add_ps(velecsum,velec);
 395
 396             fscal            = felec;
 397
 398              /* Update vectorial force */
 399             fix2             = _mm_macc_ps(dx20,fscal,fix2);
 400             fiy2             = _mm_macc_ps(dy20,fscal,fiy2);
 401             fiz2             = _mm_macc_ps(dz20,fscal,fiz2);
 402
 403             fjx0             = _mm_macc_ps(dx20,fscal,fjx0);
 404             fjy0             = _mm_macc_ps(dy20,fscal,fjy0);
 405             fjz0             = _mm_macc_ps(dz20,fscal,fjz0);
 406
 407             fjptrA             = f+j_coord_offsetA;
 408             fjptrB             = f+j_coord_offsetB;
 409             fjptrC             = f+j_coord_offsetC;
 410             fjptrD             = f+j_coord_offsetD;
 411
 412             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 413
 414             /* Inner loop uses 168 flops */
 415         }
 416
 417         if(jidx<j_index_end)
 418         {
 419
 420             /* Get j neighbor index, and coordinate index */
 421             jnrlistA         = jjnr[jidx];
 422             jnrlistB         = jjnr[jidx+1];
 423             jnrlistC         = jjnr[jidx+2];
 424             jnrlistD         = jjnr[jidx+3];
 425             /* Sign of each element will be negative for non-real atoms.
 426              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 427              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 428              */
 429             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 430             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 431             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 432             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 433             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 434             j_coord_offsetA  = DIM*jnrA;
 435             j_coord_offsetB  = DIM*jnrB;
 436             j_coord_offsetC  = DIM*jnrC;
 437             j_coord_offsetD  = DIM*jnrD;
 438
 439             /* load j atom coordinates */
 440             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 441                                               x+j_coord_offsetC,x+j_coord_offsetD,
 442                                               &jx0,&jy0,&jz0);
 443
 444             /* Calculate displacement vector */
 445             dx00             = _mm_sub_ps(ix0,jx0);
 446             dy00             = _mm_sub_ps(iy0,jy0);
 447             dz00             = _mm_sub_ps(iz0,jz0);
 448             dx10             = _mm_sub_ps(ix1,jx0);
 449             dy10             = _mm_sub_ps(iy1,jy0);
 450             dz10             = _mm_sub_ps(iz1,jz0);
 451             dx20             = _mm_sub_ps(ix2,jx0);
 452             dy20             = _mm_sub_ps(iy2,jy0);
 453             dz20             = _mm_sub_ps(iz2,jz0);
 454
 455             /* Calculate squared distance and things based on it */
 456             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 457             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 458             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 459
 460             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 461             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 462             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 463
 464             /* Load parameters for j particles */
 465             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 466                                                               charge+jnrC+0,charge+jnrD+0);
 467             vdwjidx0A        = 2*vdwtype[jnrA+0];
 468             vdwjidx0B        = 2*vdwtype[jnrB+0];
 469             vdwjidx0C        = 2*vdwtype[jnrC+0];
 470             vdwjidx0D        = 2*vdwtype[jnrD+0];
 471
 472             fjx0             = _mm_setzero_ps();
 473             fjy0             = _mm_setzero_ps();
 474             fjz0             = _mm_setzero_ps();
 475
 476             /**************************
 477              * CALCULATE INTERACTIONS *
 478              **************************/
 479
 480             r00              = _mm_mul_ps(rsq00,rinv00);
 481             r00              = _mm_andnot_ps(dummy_mask,r00);
 482
 483             /* Compute parameters for interactions between i and j atoms */
 484             qq00             = _mm_mul_ps(iq0,jq0);
 485             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 486                                          vdwparam+vdwioffset0+vdwjidx0B,
 487                                          vdwparam+vdwioffset0+vdwjidx0C,
 488                                          vdwparam+vdwioffset0+vdwjidx0D,
 489                                          &c6_00,&c12_00);
 490
 491             /* Calculate table index by multiplying r with table scale and truncate to integer */
 492             rt               = _mm_mul_ps(r00,vftabscale);
 493             vfitab           = _mm_cvttps_epi32(rt);
 494 #ifdef __XOP__
 495             vfeps            = _mm_frcz_ps(rt);
 496 #else
 497             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
 498 #endif
 499             twovfeps         = _mm_add_ps(vfeps,vfeps);
 500             vfitab           = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
 501
 502             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 503             Y                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
 504             F                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
 505             G                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
 506             H                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
 507             _MM_TRANSPOSE4_PS(Y,F,G,H);
 508             Fp               = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
 509             VV               = _mm_macc_ps(vfeps,Fp,Y);
 510             velec            = _mm_mul_ps(qq00,VV);
 511             FF               = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
 512             felec            = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq00,FF),_mm_mul_ps(vftabscale,rinv00)));
 513
 514             /* CUBIC SPLINE TABLE DISPERSION */
 515             vfitab           = _mm_add_epi32(vfitab,ifour);
 516             Y                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
 517             F                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
 518             G                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
 519             H                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
 520             _MM_TRANSPOSE4_PS(Y,F,G,H);
 521             Fp               = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
 522             VV               = _mm_macc_ps(vfeps,Fp,Y);
 523             vvdw6            = _mm_mul_ps(c6_00,VV);
 524             FF               = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
 525             fvdw6            = _mm_mul_ps(c6_00,FF);
 526
 527             /* CUBIC SPLINE TABLE REPULSION */
 528             vfitab           = _mm_add_epi32(vfitab,ifour);
 529             Y                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
 530             F                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
 531             G                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
 532             H                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
 533             _MM_TRANSPOSE4_PS(Y,F,G,H);
 534             Fp               = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
 535             VV               = _mm_macc_ps(vfeps,Fp,Y);
 536             vvdw12           = _mm_mul_ps(c12_00,VV);
 537             FF               = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
 538             fvdw12           = _mm_mul_ps(c12_00,FF);
 539             vvdw             = _mm_add_ps(vvdw12,vvdw6);
 540             fvdw             = _mm_xor_ps(signbit,_mm_mul_ps(_mm_add_ps(fvdw6,fvdw12),_mm_mul_ps(vftabscale,rinv00)));
 541
 542             /* Update potential sum for this i atom from the interaction with this j atom. */
 543             velec            = _mm_andnot_ps(dummy_mask,velec);
 544             velecsum         = _mm_add_ps(velecsum,velec);
 545             vvdw             = _mm_andnot_ps(dummy_mask,vvdw);
 546             vvdwsum          = _mm_add_ps(vvdwsum,vvdw);
 547
 548             fscal            = _mm_add_ps(felec,fvdw);
 549
 550             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 551
 552              /* Update vectorial force */
 553             fix0             = _mm_macc_ps(dx00,fscal,fix0);
 554             fiy0             = _mm_macc_ps(dy00,fscal,fiy0);
 555             fiz0             = _mm_macc_ps(dz00,fscal,fiz0);
 556
 557             fjx0             = _mm_macc_ps(dx00,fscal,fjx0);
 558             fjy0             = _mm_macc_ps(dy00,fscal,fjy0);
 559             fjz0             = _mm_macc_ps(dz00,fscal,fjz0);
 560
 561             /**************************
 562              * CALCULATE INTERACTIONS *
 563              **************************/
 564
 565             r10              = _mm_mul_ps(rsq10,rinv10);
 566             r10              = _mm_andnot_ps(dummy_mask,r10);
 567
 568             /* Compute parameters for interactions between i and j atoms */
 569             qq10             = _mm_mul_ps(iq1,jq0);
 570
 571             /* Calculate table index by multiplying r with table scale and truncate to integer */
 572             rt               = _mm_mul_ps(r10,vftabscale);
 573             vfitab           = _mm_cvttps_epi32(rt);
 574 #ifdef __XOP__
 575             vfeps            = _mm_frcz_ps(rt);
 576 #else
 577             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
 578 #endif
 579             twovfeps         = _mm_add_ps(vfeps,vfeps);
 580             vfitab           = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
 581
 582             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 583             Y                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
 584             F                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
 585             G                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
 586             H                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
 587             _MM_TRANSPOSE4_PS(Y,F,G,H);
 588             Fp               = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
 589             VV               = _mm_macc_ps(vfeps,Fp,Y);
 590             velec            = _mm_mul_ps(qq10,VV);
 591             FF               = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
 592             felec            = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq10,FF),_mm_mul_ps(vftabscale,rinv10)));
 593
 594             /* Update potential sum for this i atom from the interaction with this j atom. */
 595             velec            = _mm_andnot_ps(dummy_mask,velec);
 596             velecsum         = _mm_add_ps(velecsum,velec);
 597
 598             fscal            = felec;
 599
 600             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 601
 602              /* Update vectorial force */
 603             fix1             = _mm_macc_ps(dx10,fscal,fix1);
 604             fiy1             = _mm_macc_ps(dy10,fscal,fiy1);
 605             fiz1             = _mm_macc_ps(dz10,fscal,fiz1);
 606
 607             fjx0             = _mm_macc_ps(dx10,fscal,fjx0);
 608             fjy0             = _mm_macc_ps(dy10,fscal,fjy0);
 609             fjz0             = _mm_macc_ps(dz10,fscal,fjz0);
 610
 611             /**************************
 612              * CALCULATE INTERACTIONS *
 613              **************************/
 614
 615             r20              = _mm_mul_ps(rsq20,rinv20);
 616             r20              = _mm_andnot_ps(dummy_mask,r20);
 617
 618             /* Compute parameters for interactions between i and j atoms */
 619             qq20             = _mm_mul_ps(iq2,jq0);
 620
 621             /* Calculate table index by multiplying r with table scale and truncate to integer */
 622             rt               = _mm_mul_ps(r20,vftabscale);
 623             vfitab           = _mm_cvttps_epi32(rt);
 624 #ifdef __XOP__
 625             vfeps            = _mm_frcz_ps(rt);
 626 #else
 627             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
 628 #endif
 629             twovfeps         = _mm_add_ps(vfeps,vfeps);
 630             vfitab           = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
 631
 632             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 633             Y                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
 634             F                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
 635             G                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
 636             H                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
 637             _MM_TRANSPOSE4_PS(Y,F,G,H);
 638             Fp               = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
 639             VV               = _mm_macc_ps(vfeps,Fp,Y);
 640             velec            = _mm_mul_ps(qq20,VV);
 641             FF               = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
 642             felec            = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq20,FF),_mm_mul_ps(vftabscale,rinv20)));
 643
 644             /* Update potential sum for this i atom from the interaction with this j atom. */
 645             velec            = _mm_andnot_ps(dummy_mask,velec);
 646             velecsum         = _mm_add_ps(velecsum,velec);
 647
 648             fscal            = felec;
 649
 650             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 651
 652              /* Update vectorial force */
 653             fix2             = _mm_macc_ps(dx20,fscal,fix2);
 654             fiy2             = _mm_macc_ps(dy20,fscal,fiy2);
 655             fiz2             = _mm_macc_ps(dz20,fscal,fiz2);
 656
 657             fjx0             = _mm_macc_ps(dx20,fscal,fjx0);
 658             fjy0             = _mm_macc_ps(dy20,fscal,fjy0);
 659             fjz0             = _mm_macc_ps(dz20,fscal,fjz0);
 660
 661             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 662             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 663             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 664             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 665
 666             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 667
 668             /* Inner loop uses 171 flops */
 669         }
 670
 671         /* End of innermost loop */
 672
 673         gmx_mm_update_iforce_3atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
 674                                               f+i_coord_offset,fshift+i_shift_offset);
 675
 676         ggid                        = gid[iidx];
 677         /* Update potential energies */
 678         gmx_mm_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid);
 679         gmx_mm_update_1pot_ps(vvdwsum,kernel_data->energygrp_vdw+ggid);
 680
 681         /* Increment number of inner iterations */
 682         inneriter                  += j_index_end - j_index_start;
 683
 684         /* Outer loop uses 20 flops */
 685     }
 686
 687     /* Increment number of outer iterations */
 688     outeriter        += nri;
 689
 690     /* Update outer/inner flops */
 691
 692     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3_VF,outeriter*20 + inneriter*171);
 693 }
 694 /*
 695  * Gromacs nonbonded kernel:   nb_kernel_ElecCSTab_VdwCSTab_GeomW3P1_F_avx_128_fma_single
 696  * Electrostatics interaction: CubicSplineTable
 697  * VdW interaction:            CubicSplineTable
 698  * Geometry:                   Water3-Particle
 699  * Calculate force/pot:        Force
 700  */
 701 void
 702 nb_kernel_ElecCSTab_VdwCSTab_GeomW3P1_F_avx_128_fma_single
 703                     (t_nblist                    * gmx_restrict       nlist,
 704                      rvec                        * gmx_restrict          xx,
 705                      rvec                        * gmx_restrict          ff,
 706                      t_forcerec                  * gmx_restrict          fr,
 707                      t_mdatoms                   * gmx_restrict     mdatoms,
 708                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 709                      t_nrnb                      * gmx_restrict        nrnb)
 710 {
 711     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 712      * just 0 for non-waters.
 713      * Suffixes A,B,C,D refer to j loop unrolling done with AVX_128, e.g. for the four different
 714      * jnr indices corresponding to data put in the four positions in the SIMD register.
 715      */
 716     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 717     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 718     int              jnrA,jnrB,jnrC,jnrD;
 719     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
 720     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 721     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 722     real             rcutoff_scalar;
 723     real             *shiftvec,*fshift,*x,*f;
 724     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
 725     real             scratch[4*DIM];
 726     __m128           fscal,rcutoff,rcutoff2,jidxall;
 727     int              vdwioffset0;
 728     __m128           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 729     int              vdwioffset1;
 730     __m128           ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
 731     int              vdwioffset2;
 732     __m128           ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
 733     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
 734     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 735     __m128           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 736     __m128           dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
 737     __m128           dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
 738     __m128           velec,felec,velecsum,facel,crf,krf,krf2;
 739     real             *charge;
 740     int              nvdwtype;
 741     __m128           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 742     int              *vdwtype;
 743     real             *vdwparam;
 744     __m128           one_sixth   = _mm_set1_ps(1.0/6.0);
 745     __m128           one_twelfth = _mm_set1_ps(1.0/12.0);
 746     __m128i          vfitab;
 747     __m128i          ifour       = _mm_set1_epi32(4);
 748     __m128           rt,vfeps,twovfeps,vftabscale,Y,F,G,H,Fp,VV,FF;
 749     real             *vftab;
 750     __m128           dummy_mask,cutoff_mask;
 751     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
 752     __m128           one     = _mm_set1_ps(1.0);
 753     __m128           two     = _mm_set1_ps(2.0);
 754     x                = xx[0];
 755     f                = ff[0];
 756
 757     nri              = nlist->nri;
 758     iinr             = nlist->iinr;
 759     jindex           = nlist->jindex;
 760     jjnr             = nlist->jjnr;
 761     shiftidx         = nlist->shift;
 762     gid              = nlist->gid;
 763     shiftvec         = fr->shift_vec[0];
 764     fshift           = fr->fshift[0];
 765     facel            = _mm_set1_ps(fr->epsfac);
 766     charge           = mdatoms->chargeA;
 767     nvdwtype         = fr->ntype;
 768     vdwparam         = fr->nbfp;
 769     vdwtype          = mdatoms->typeA;
 770
 771     vftab            = kernel_data->table_elec_vdw->data;
 772     vftabscale       = _mm_set1_ps(kernel_data->table_elec_vdw->scale);
 773
 774     /* Setup water-specific parameters */
 775     inr              = nlist->iinr[0];
 776     iq0              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+0]));
 777     iq1              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+1]));
 778     iq2              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+2]));
 779     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 780
 781     /* Avoid stupid compiler warnings */
 782     jnrA = jnrB = jnrC = jnrD = 0;
 783     j_coord_offsetA = 0;
 784     j_coord_offsetB = 0;
 785     j_coord_offsetC = 0;
 786     j_coord_offsetD = 0;
 787
 788     outeriter        = 0;
 789     inneriter        = 0;
 790
 791     for(iidx=0;iidx<4*DIM;iidx++)
 792     {
 793         scratch[iidx] = 0.0;
 794     }
 795
 796     /* Start outer loop over neighborlists */
 797     for(iidx=0; iidx<nri; iidx++)
 798     {
 799         /* Load shift vector for this list */
 800         i_shift_offset   = DIM*shiftidx[iidx];
 801
 802         /* Load limits for loop over neighbors */
 803         j_index_start    = jindex[iidx];
 804         j_index_end      = jindex[iidx+1];
 805
 806         /* Get outer coordinate index */
 807         inr              = iinr[iidx];
 808         i_coord_offset   = DIM*inr;
 809
 810         /* Load i particle coords and add shift vector */
 811         gmx_mm_load_shift_and_3rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
 812                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 813
 814         fix0             = _mm_setzero_ps();
 815         fiy0             = _mm_setzero_ps();
 816         fiz0             = _mm_setzero_ps();
 817         fix1             = _mm_setzero_ps();
 818         fiy1             = _mm_setzero_ps();
 819         fiz1             = _mm_setzero_ps();
 820         fix2             = _mm_setzero_ps();
 821         fiy2             = _mm_setzero_ps();
 822         fiz2             = _mm_setzero_ps();
 823
 824         /* Start inner kernel loop */
 825         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 826         {
 827
 828             /* Get j neighbor index, and coordinate index */
 829             jnrA             = jjnr[jidx];
 830             jnrB             = jjnr[jidx+1];
 831             jnrC             = jjnr[jidx+2];
 832             jnrD             = jjnr[jidx+3];
 833             j_coord_offsetA  = DIM*jnrA;
 834             j_coord_offsetB  = DIM*jnrB;
 835             j_coord_offsetC  = DIM*jnrC;
 836             j_coord_offsetD  = DIM*jnrD;
 837
 838             /* load j atom coordinates */
 839             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 840                                               x+j_coord_offsetC,x+j_coord_offsetD,
 841                                               &jx0,&jy0,&jz0);
 842
 843             /* Calculate displacement vector */
 844             dx00             = _mm_sub_ps(ix0,jx0);
 845             dy00             = _mm_sub_ps(iy0,jy0);
 846             dz00             = _mm_sub_ps(iz0,jz0);
 847             dx10             = _mm_sub_ps(ix1,jx0);
 848             dy10             = _mm_sub_ps(iy1,jy0);
 849             dz10             = _mm_sub_ps(iz1,jz0);
 850             dx20             = _mm_sub_ps(ix2,jx0);
 851             dy20             = _mm_sub_ps(iy2,jy0);
 852             dz20             = _mm_sub_ps(iz2,jz0);
 853
 854             /* Calculate squared distance and things based on it */
 855             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 856             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 857             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 858
 859             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 860             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 861             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 862
 863             /* Load parameters for j particles */
 864             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 865                                                               charge+jnrC+0,charge+jnrD+0);
 866             vdwjidx0A        = 2*vdwtype[jnrA+0];
 867             vdwjidx0B        = 2*vdwtype[jnrB+0];
 868             vdwjidx0C        = 2*vdwtype[jnrC+0];
 869             vdwjidx0D        = 2*vdwtype[jnrD+0];
 870
 871             fjx0             = _mm_setzero_ps();
 872             fjy0             = _mm_setzero_ps();
 873             fjz0             = _mm_setzero_ps();
 874
 875             /**************************
 876              * CALCULATE INTERACTIONS *
 877              **************************/
 878
 879             r00              = _mm_mul_ps(rsq00,rinv00);
 880
 881             /* Compute parameters for interactions between i and j atoms */
 882             qq00             = _mm_mul_ps(iq0,jq0);
 883             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 884                                          vdwparam+vdwioffset0+vdwjidx0B,
 885                                          vdwparam+vdwioffset0+vdwjidx0C,
 886                                          vdwparam+vdwioffset0+vdwjidx0D,
 887                                          &c6_00,&c12_00);
 888
 889             /* Calculate table index by multiplying r with table scale and truncate to integer */
 890             rt               = _mm_mul_ps(r00,vftabscale);
 891             vfitab           = _mm_cvttps_epi32(rt);
 892 #ifdef __XOP__
 893             vfeps            = _mm_frcz_ps(rt);
 894 #else
 895             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
 896 #endif
 897             twovfeps         = _mm_add_ps(vfeps,vfeps);
 898             vfitab           = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
 899
 900             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 901             Y                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
 902             F                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
 903             G                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
 904             H                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
 905             _MM_TRANSPOSE4_PS(Y,F,G,H);
 906             Fp               = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
 907             FF               = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
 908             felec            = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq00,FF),_mm_mul_ps(vftabscale,rinv00)));
 909
 910             /* CUBIC SPLINE TABLE DISPERSION */
 911             vfitab           = _mm_add_epi32(vfitab,ifour);
 912             Y                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
 913             F                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
 914             G                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
 915             H                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
 916             _MM_TRANSPOSE4_PS(Y,F,G,H);
 917             Fp               = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
 918             FF               = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
 919             fvdw6            = _mm_mul_ps(c6_00,FF);
 920
 921             /* CUBIC SPLINE TABLE REPULSION */
 922             vfitab           = _mm_add_epi32(vfitab,ifour);
 923             Y                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
 924             F                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
 925             G                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
 926             H                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
 927             _MM_TRANSPOSE4_PS(Y,F,G,H);
 928             Fp               = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
 929             FF               = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
 930             fvdw12           = _mm_mul_ps(c12_00,FF);
 931             fvdw             = _mm_xor_ps(signbit,_mm_mul_ps(_mm_add_ps(fvdw6,fvdw12),_mm_mul_ps(vftabscale,rinv00)));
 932
 933             fscal            = _mm_add_ps(felec,fvdw);
 934
 935              /* Update vectorial force */
 936             fix0             = _mm_macc_ps(dx00,fscal,fix0);
 937             fiy0             = _mm_macc_ps(dy00,fscal,fiy0);
 938             fiz0             = _mm_macc_ps(dz00,fscal,fiz0);
 939
 940             fjx0             = _mm_macc_ps(dx00,fscal,fjx0);
 941             fjy0             = _mm_macc_ps(dy00,fscal,fjy0);
 942             fjz0             = _mm_macc_ps(dz00,fscal,fjz0);
 943
 944             /**************************
 945              * CALCULATE INTERACTIONS *
 946              **************************/
 947
 948             r10              = _mm_mul_ps(rsq10,rinv10);
 949
 950             /* Compute parameters for interactions between i and j atoms */
 951             qq10             = _mm_mul_ps(iq1,jq0);
 952
 953             /* Calculate table index by multiplying r with table scale and truncate to integer */
 954             rt               = _mm_mul_ps(r10,vftabscale);
 955             vfitab           = _mm_cvttps_epi32(rt);
 956 #ifdef __XOP__
 957             vfeps            = _mm_frcz_ps(rt);
 958 #else
 959             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
 960 #endif
 961             twovfeps         = _mm_add_ps(vfeps,vfeps);
 962             vfitab           = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
 963
 964             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 965             Y                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
 966             F                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
 967             G                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
 968             H                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
 969             _MM_TRANSPOSE4_PS(Y,F,G,H);
 970             Fp               = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
 971             FF               = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
 972             felec            = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq10,FF),_mm_mul_ps(vftabscale,rinv10)));
 973
 974             fscal            = felec;
 975
 976              /* Update vectorial force */
 977             fix1             = _mm_macc_ps(dx10,fscal,fix1);
 978             fiy1             = _mm_macc_ps(dy10,fscal,fiy1);
 979             fiz1             = _mm_macc_ps(dz10,fscal,fiz1);
 980
 981             fjx0             = _mm_macc_ps(dx10,fscal,fjx0);
 982             fjy0             = _mm_macc_ps(dy10,fscal,fjy0);
 983             fjz0             = _mm_macc_ps(dz10,fscal,fjz0);
 984
 985             /**************************
 986              * CALCULATE INTERACTIONS *
 987              **************************/
 988
 989             r20              = _mm_mul_ps(rsq20,rinv20);
 990
 991             /* Compute parameters for interactions between i and j atoms */
 992             qq20             = _mm_mul_ps(iq2,jq0);
 993
 994             /* Calculate table index by multiplying r with table scale and truncate to integer */
 995             rt               = _mm_mul_ps(r20,vftabscale);
 996             vfitab           = _mm_cvttps_epi32(rt);
 997 #ifdef __XOP__
 998             vfeps            = _mm_frcz_ps(rt);
 999 #else
1000             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
1001 #endif
1002             twovfeps         = _mm_add_ps(vfeps,vfeps);
1003             vfitab           = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
1004
1005             /* CUBIC SPLINE TABLE ELECTROSTATICS */
1006             Y                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
1007             F                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
1008             G                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
1009             H                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
1010             _MM_TRANSPOSE4_PS(Y,F,G,H);
1011             Fp               = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
1012             FF               = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
1013             felec            = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq20,FF),_mm_mul_ps(vftabscale,rinv20)));
1014
1015             fscal            = felec;
1016
1017              /* Update vectorial force */
1018             fix2             = _mm_macc_ps(dx20,fscal,fix2);
1019             fiy2             = _mm_macc_ps(dy20,fscal,fiy2);
1020             fiz2             = _mm_macc_ps(dz20,fscal,fiz2);
1021
1022             fjx0             = _mm_macc_ps(dx20,fscal,fjx0);
1023             fjy0             = _mm_macc_ps(dy20,fscal,fjy0);
1024             fjz0             = _mm_macc_ps(dz20,fscal,fjz0);
1025
1026             fjptrA             = f+j_coord_offsetA;
1027             fjptrB             = f+j_coord_offsetB;
1028             fjptrC             = f+j_coord_offsetC;
1029             fjptrD             = f+j_coord_offsetD;
1030
1031             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
1032
1033             /* Inner loop uses 148 flops */
1034         }
1035
1036         if(jidx<j_index_end)
1037         {
1038
1039             /* Get j neighbor index, and coordinate index */
1040             jnrlistA         = jjnr[jidx];
1041             jnrlistB         = jjnr[jidx+1];
1042             jnrlistC         = jjnr[jidx+2];
1043             jnrlistD         = jjnr[jidx+3];
1044             /* Sign of each element will be negative for non-real atoms.
1045              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
1046              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
1047              */
1048             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
1049             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
1050             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
1051             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
1052             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
1053             j_coord_offsetA  = DIM*jnrA;
1054             j_coord_offsetB  = DIM*jnrB;
1055             j_coord_offsetC  = DIM*jnrC;
1056             j_coord_offsetD  = DIM*jnrD;
1057
1058             /* load j atom coordinates */
1059             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
1060                                               x+j_coord_offsetC,x+j_coord_offsetD,
1061                                               &jx0,&jy0,&jz0);
1062
1063             /* Calculate displacement vector */
1064             dx00             = _mm_sub_ps(ix0,jx0);
1065             dy00             = _mm_sub_ps(iy0,jy0);
1066             dz00             = _mm_sub_ps(iz0,jz0);
1067             dx10             = _mm_sub_ps(ix1,jx0);
1068             dy10             = _mm_sub_ps(iy1,jy0);
1069             dz10             = _mm_sub_ps(iz1,jz0);
1070             dx20             = _mm_sub_ps(ix2,jx0);
1071             dy20             = _mm_sub_ps(iy2,jy0);
1072             dz20             = _mm_sub_ps(iz2,jz0);
1073
1074             /* Calculate squared distance and things based on it */
1075             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
1076             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
1077             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
1078
1079             rinv00           = gmx_mm_invsqrt_ps(rsq00);
1080             rinv10           = gmx_mm_invsqrt_ps(rsq10);
1081             rinv20           = gmx_mm_invsqrt_ps(rsq20);
1082
1083             /* Load parameters for j particles */
1084             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
1085                                                               charge+jnrC+0,charge+jnrD+0);
1086             vdwjidx0A        = 2*vdwtype[jnrA+0];
1087             vdwjidx0B        = 2*vdwtype[jnrB+0];
1088             vdwjidx0C        = 2*vdwtype[jnrC+0];
1089             vdwjidx0D        = 2*vdwtype[jnrD+0];
1090
1091             fjx0             = _mm_setzero_ps();
1092             fjy0             = _mm_setzero_ps();
1093             fjz0             = _mm_setzero_ps();
1094
1095             /**************************
1096              * CALCULATE INTERACTIONS *
1097              **************************/
1098
1099             r00              = _mm_mul_ps(rsq00,rinv00);
1100             r00              = _mm_andnot_ps(dummy_mask,r00);
1101
1102             /* Compute parameters for interactions between i and j atoms */
1103             qq00             = _mm_mul_ps(iq0,jq0);
1104             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
1105                                          vdwparam+vdwioffset0+vdwjidx0B,
1106                                          vdwparam+vdwioffset0+vdwjidx0C,
1107                                          vdwparam+vdwioffset0+vdwjidx0D,
1108                                          &c6_00,&c12_00);
1109
1110             /* Calculate table index by multiplying r with table scale and truncate to integer */
1111             rt               = _mm_mul_ps(r00,vftabscale);
1112             vfitab           = _mm_cvttps_epi32(rt);
1113 #ifdef __XOP__
1114             vfeps            = _mm_frcz_ps(rt);
1115 #else
1116             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
1117 #endif
1118             twovfeps         = _mm_add_ps(vfeps,vfeps);
1119             vfitab           = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
1120
1121             /* CUBIC SPLINE TABLE ELECTROSTATICS */
1122             Y                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
1123             F                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
1124             G                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
1125             H                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
1126             _MM_TRANSPOSE4_PS(Y,F,G,H);
1127             Fp               = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
1128             FF               = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
1129             felec            = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq00,FF),_mm_mul_ps(vftabscale,rinv00)));
1130
1131             /* CUBIC SPLINE TABLE DISPERSION */
1132             vfitab           = _mm_add_epi32(vfitab,ifour);
1133             Y                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
1134             F                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
1135             G                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
1136             H                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
1137             _MM_TRANSPOSE4_PS(Y,F,G,H);
1138             Fp               = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
1139             FF               = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
1140             fvdw6            = _mm_mul_ps(c6_00,FF);
1141
1142             /* CUBIC SPLINE TABLE REPULSION */
1143             vfitab           = _mm_add_epi32(vfitab,ifour);
1144             Y                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
1145             F                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
1146             G                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
1147             H                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
1148             _MM_TRANSPOSE4_PS(Y,F,G,H);
1149             Fp               = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
1150             FF               = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
1151             fvdw12           = _mm_mul_ps(c12_00,FF);
1152             fvdw             = _mm_xor_ps(signbit,_mm_mul_ps(_mm_add_ps(fvdw6,fvdw12),_mm_mul_ps(vftabscale,rinv00)));
1153
1154             fscal            = _mm_add_ps(felec,fvdw);
1155
1156             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1157
1158              /* Update vectorial force */
1159             fix0             = _mm_macc_ps(dx00,fscal,fix0);
1160             fiy0             = _mm_macc_ps(dy00,fscal,fiy0);
1161             fiz0             = _mm_macc_ps(dz00,fscal,fiz0);
1162
1163             fjx0             = _mm_macc_ps(dx00,fscal,fjx0);
1164             fjy0             = _mm_macc_ps(dy00,fscal,fjy0);
1165             fjz0             = _mm_macc_ps(dz00,fscal,fjz0);
1166
1167             /**************************
1168              * CALCULATE INTERACTIONS *
1169              **************************/
1170
1171             r10              = _mm_mul_ps(rsq10,rinv10);
1172             r10              = _mm_andnot_ps(dummy_mask,r10);
1173
1174             /* Compute parameters for interactions between i and j atoms */
1175             qq10             = _mm_mul_ps(iq1,jq0);
1176
1177             /* Calculate table index by multiplying r with table scale and truncate to integer */
1178             rt               = _mm_mul_ps(r10,vftabscale);
1179             vfitab           = _mm_cvttps_epi32(rt);
1180 #ifdef __XOP__
1181             vfeps            = _mm_frcz_ps(rt);
1182 #else
1183             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
1184 #endif
1185             twovfeps         = _mm_add_ps(vfeps,vfeps);
1186             vfitab           = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
1187
1188             /* CUBIC SPLINE TABLE ELECTROSTATICS */
1189             Y                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
1190             F                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
1191             G                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
1192             H                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
1193             _MM_TRANSPOSE4_PS(Y,F,G,H);
1194             Fp               = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
1195             FF               = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
1196             felec            = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq10,FF),_mm_mul_ps(vftabscale,rinv10)));
1197
1198             fscal            = felec;
1199
1200             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1201
1202              /* Update vectorial force */
1203             fix1             = _mm_macc_ps(dx10,fscal,fix1);
1204             fiy1             = _mm_macc_ps(dy10,fscal,fiy1);
1205             fiz1             = _mm_macc_ps(dz10,fscal,fiz1);
1206
1207             fjx0             = _mm_macc_ps(dx10,fscal,fjx0);
1208             fjy0             = _mm_macc_ps(dy10,fscal,fjy0);
1209             fjz0             = _mm_macc_ps(dz10,fscal,fjz0);
1210
1211             /**************************
1212              * CALCULATE INTERACTIONS *
1213              **************************/
1214
1215             r20              = _mm_mul_ps(rsq20,rinv20);
1216             r20              = _mm_andnot_ps(dummy_mask,r20);
1217
1218             /* Compute parameters for interactions between i and j atoms */
1219             qq20             = _mm_mul_ps(iq2,jq0);
1220
1221             /* Calculate table index by multiplying r with table scale and truncate to integer */
1222             rt               = _mm_mul_ps(r20,vftabscale);
1223             vfitab           = _mm_cvttps_epi32(rt);
1224 #ifdef __XOP__
1225             vfeps            = _mm_frcz_ps(rt);
1226 #else
1227             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
1228 #endif
1229             twovfeps         = _mm_add_ps(vfeps,vfeps);
1230             vfitab           = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
1231
1232             /* CUBIC SPLINE TABLE ELECTROSTATICS */
1233             Y                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
1234             F                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
1235             G                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
1236             H                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
1237             _MM_TRANSPOSE4_PS(Y,F,G,H);
1238             Fp               = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
1239             FF               = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
1240             felec            = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq20,FF),_mm_mul_ps(vftabscale,rinv20)));
1241
1242             fscal            = felec;
1243
1244             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1245
1246              /* Update vectorial force */
1247             fix2             = _mm_macc_ps(dx20,fscal,fix2);
1248             fiy2             = _mm_macc_ps(dy20,fscal,fiy2);
1249             fiz2             = _mm_macc_ps(dz20,fscal,fiz2);
1250
1251             fjx0             = _mm_macc_ps(dx20,fscal,fjx0);
1252             fjy0             = _mm_macc_ps(dy20,fscal,fjy0);
1253             fjz0             = _mm_macc_ps(dz20,fscal,fjz0);
1254
1255             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
1256             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
1257             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
1258             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
1259
1260             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
1261
1262             /* Inner loop uses 151 flops */
1263         }
1264
1265         /* End of innermost loop */
1266
1267         gmx_mm_update_iforce_3atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
1268                                               f+i_coord_offset,fshift+i_shift_offset);
1269
1270         /* Increment number of inner iterations */
1271         inneriter                  += j_index_end - j_index_start;
1272
1273         /* Outer loop uses 18 flops */
1274     }
1275
1276     /* Increment number of outer iterations */
1277     outeriter        += nri;
1278
1279     /* Update outer/inner flops */
1280
1281     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3_F,outeriter*18 + inneriter*151);
1282 }