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 "gmxpre.h"
  39
  40 #include "config.h"
  41
  42 #include <math.h>
  43
  44 #include "../nb_kernel.h"
  45 #include "gromacs/legacyheaders/types/simple.h"
  46 #include "gromacs/math/vec.h"
  47 #include "gromacs/legacyheaders/nrnb.h"
  48
  49 #include "gromacs/simd/math_x86_avx_128_fma_single.h"
  50 #include "kernelutil_x86_avx_128_fma_single.h"
  51
  52 /*
  53  * Gromacs nonbonded kernel:   nb_kernel_ElecCSTab_VdwCSTab_GeomW3P1_VF_avx_128_fma_single
  54  * Electrostatics interaction: CubicSplineTable
  55  * VdW interaction:            CubicSplineTable
  56  * Geometry:                   Water3-Particle
  57  * Calculate force/pot:        PotentialAndForce
  58  */
  59 void
  60 nb_kernel_ElecCSTab_VdwCSTab_GeomW3P1_VF_avx_128_fma_single
  61                     (t_nblist                    * gmx_restrict       nlist,
  62                      rvec                        * gmx_restrict          xx,
  63                      rvec                        * gmx_restrict          ff,
  64                      t_forcerec                  * gmx_restrict          fr,
  65                      t_mdatoms                   * gmx_restrict     mdatoms,
  66                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
  67                      t_nrnb                      * gmx_restrict        nrnb)
  68 {
  69     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
  70      * just 0 for non-waters.
  71      * Suffixes A,B,C,D refer to j loop unrolling done with AVX_128, e.g. for the four different
  72      * jnr indices corresponding to data put in the four positions in the SIMD register.
  73      */
  74     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
  75     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
  76     int              jnrA,jnrB,jnrC,jnrD;
  77     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
  78     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
  79     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
  80     real             rcutoff_scalar;
  81     real             *shiftvec,*fshift,*x,*f;
  82     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
  83     real             scratch[4*DIM];
  84     __m128           fscal,rcutoff,rcutoff2,jidxall;
  85     int              vdwioffset0;
  86     __m128           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
  87     int              vdwioffset1;
  88     __m128           ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
  89     int              vdwioffset2;
  90     __m128           ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
  91     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
  92     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  93     __m128           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  94     __m128           dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
  95     __m128           dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
  96     __m128           velec,felec,velecsum,facel,crf,krf,krf2;
  97     real             *charge;
  98     int              nvdwtype;
  99     __m128           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 100     int              *vdwtype;
 101     real             *vdwparam;
 102     __m128           one_sixth   = _mm_set1_ps(1.0/6.0);
 103     __m128           one_twelfth = _mm_set1_ps(1.0/12.0);
 104     __m128i          vfitab;
 105     __m128i          ifour       = _mm_set1_epi32(4);
 106     __m128           rt,vfeps,twovfeps,vftabscale,Y,F,G,H,Fp,VV,FF;
 107     real             *vftab;
 108     __m128           dummy_mask,cutoff_mask;
 109     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
 110     __m128           one     = _mm_set1_ps(1.0);
 111     __m128           two     = _mm_set1_ps(2.0);
 112     x                = xx[0];
 113     f                = ff[0];
 114
 115     nri              = nlist->nri;
 116     iinr             = nlist->iinr;
 117     jindex           = nlist->jindex;
 118     jjnr             = nlist->jjnr;
 119     shiftidx         = nlist->shift;
 120     gid              = nlist->gid;
 121     shiftvec         = fr->shift_vec[0];
 122     fshift           = fr->fshift[0];
 123     facel            = _mm_set1_ps(fr->epsfac);
 124     charge           = mdatoms->chargeA;
 125     nvdwtype         = fr->ntype;
 126     vdwparam         = fr->nbfp;
 127     vdwtype          = mdatoms->typeA;
 128
 129     vftab            = kernel_data->table_elec_vdw->data;
 130     vftabscale       = _mm_set1_ps(kernel_data->table_elec_vdw->scale);
 131
 132     /* Setup water-specific parameters */
 133     inr              = nlist->iinr[0];
 134     iq0              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+0]));
 135     iq1              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+1]));
 136     iq2              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+2]));
 137     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 138
 139     /* Avoid stupid compiler warnings */
 140     jnrA = jnrB = jnrC = jnrD = 0;
 141     j_coord_offsetA = 0;
 142     j_coord_offsetB = 0;
 143     j_coord_offsetC = 0;
 144     j_coord_offsetD = 0;
 145
 146     outeriter        = 0;
 147     inneriter        = 0;
 148
 149     for(iidx=0;iidx<4*DIM;iidx++)
 150     {
 151         scratch[iidx] = 0.0;
 152     }
 153
 154     /* Start outer loop over neighborlists */
 155     for(iidx=0; iidx<nri; iidx++)
 156     {
 157         /* Load shift vector for this list */
 158         i_shift_offset   = DIM*shiftidx[iidx];
 159
 160         /* Load limits for loop over neighbors */
 161         j_index_start    = jindex[iidx];
 162         j_index_end      = jindex[iidx+1];
 163
 164         /* Get outer coordinate index */
 165         inr              = iinr[iidx];
 166         i_coord_offset   = DIM*inr;
 167
 168         /* Load i particle coords and add shift vector */
 169         gmx_mm_load_shift_and_3rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
 170                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 171
 172         fix0             = _mm_setzero_ps();
 173         fiy0             = _mm_setzero_ps();
 174         fiz0             = _mm_setzero_ps();
 175         fix1             = _mm_setzero_ps();
 176         fiy1             = _mm_setzero_ps();
 177         fiz1             = _mm_setzero_ps();
 178         fix2             = _mm_setzero_ps();
 179         fiy2             = _mm_setzero_ps();
 180         fiz2             = _mm_setzero_ps();
 181
 182         /* Reset potential sums */
 183         velecsum         = _mm_setzero_ps();
 184         vvdwsum          = _mm_setzero_ps();
 185
 186         /* Start inner kernel loop */
 187         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 188         {
 189
 190             /* Get j neighbor index, and coordinate index */
 191             jnrA             = jjnr[jidx];
 192             jnrB             = jjnr[jidx+1];
 193             jnrC             = jjnr[jidx+2];
 194             jnrD             = jjnr[jidx+3];
 195             j_coord_offsetA  = DIM*jnrA;
 196             j_coord_offsetB  = DIM*jnrB;
 197             j_coord_offsetC  = DIM*jnrC;
 198             j_coord_offsetD  = DIM*jnrD;
 199
 200             /* load j atom coordinates */
 201             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 202                                               x+j_coord_offsetC,x+j_coord_offsetD,
 203                                               &jx0,&jy0,&jz0);
 204
 205             /* Calculate displacement vector */
 206             dx00             = _mm_sub_ps(ix0,jx0);
 207             dy00             = _mm_sub_ps(iy0,jy0);
 208             dz00             = _mm_sub_ps(iz0,jz0);
 209             dx10             = _mm_sub_ps(ix1,jx0);
 210             dy10             = _mm_sub_ps(iy1,jy0);
 211             dz10             = _mm_sub_ps(iz1,jz0);
 212             dx20             = _mm_sub_ps(ix2,jx0);
 213             dy20             = _mm_sub_ps(iy2,jy0);
 214             dz20             = _mm_sub_ps(iz2,jz0);
 215
 216             /* Calculate squared distance and things based on it */
 217             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 218             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 219             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 220
 221             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 222             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 223             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 224
 225             /* Load parameters for j particles */
 226             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 227                                                               charge+jnrC+0,charge+jnrD+0);
 228             vdwjidx0A        = 2*vdwtype[jnrA+0];
 229             vdwjidx0B        = 2*vdwtype[jnrB+0];
 230             vdwjidx0C        = 2*vdwtype[jnrC+0];
 231             vdwjidx0D        = 2*vdwtype[jnrD+0];
 232
 233             fjx0             = _mm_setzero_ps();
 234             fjy0             = _mm_setzero_ps();
 235             fjz0             = _mm_setzero_ps();
 236
 237             /**************************
 238              * CALCULATE INTERACTIONS *
 239              **************************/
 240
 241             r00              = _mm_mul_ps(rsq00,rinv00);
 242
 243             /* Compute parameters for interactions between i and j atoms */
 244             qq00             = _mm_mul_ps(iq0,jq0);
 245             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 246                                          vdwparam+vdwioffset0+vdwjidx0B,
 247                                          vdwparam+vdwioffset0+vdwjidx0C,
 248                                          vdwparam+vdwioffset0+vdwjidx0D,
 249                                          &c6_00,&c12_00);
 250
 251             /* Calculate table index by multiplying r with table scale and truncate to integer */
 252             rt               = _mm_mul_ps(r00,vftabscale);
 253             vfitab           = _mm_cvttps_epi32(rt);
 254 #ifdef __XOP__
 255             vfeps            = _mm_frcz_ps(rt);
 256 #else
 257             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
 258 #endif
 259             twovfeps         = _mm_add_ps(vfeps,vfeps);
 260             vfitab           = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
 261
 262             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 263             Y                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
 264             F                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
 265             G                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
 266             H                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
 267             _MM_TRANSPOSE4_PS(Y,F,G,H);
 268             Fp               = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
 269             VV               = _mm_macc_ps(vfeps,Fp,Y);
 270             velec            = _mm_mul_ps(qq00,VV);
 271             FF               = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
 272             felec            = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq00,FF),_mm_mul_ps(vftabscale,rinv00)));
 273
 274             /* CUBIC SPLINE TABLE DISPERSION */
 275             vfitab           = _mm_add_epi32(vfitab,ifour);
 276             Y                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
 277             F                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
 278             G                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
 279             H                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
 280             _MM_TRANSPOSE4_PS(Y,F,G,H);
 281             Fp               = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
 282             VV               = _mm_macc_ps(vfeps,Fp,Y);
 283             vvdw6            = _mm_mul_ps(c6_00,VV);
 284             FF               = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
 285             fvdw6            = _mm_mul_ps(c6_00,FF);
 286
 287             /* CUBIC SPLINE TABLE REPULSION */
 288             vfitab           = _mm_add_epi32(vfitab,ifour);
 289             Y                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
 290             F                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
 291             G                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
 292             H                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
 293             _MM_TRANSPOSE4_PS(Y,F,G,H);
 294             Fp               = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
 295             VV               = _mm_macc_ps(vfeps,Fp,Y);
 296             vvdw12           = _mm_mul_ps(c12_00,VV);
 297             FF               = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
 298             fvdw12           = _mm_mul_ps(c12_00,FF);
 299             vvdw             = _mm_add_ps(vvdw12,vvdw6);
 300             fvdw             = _mm_xor_ps(signbit,_mm_mul_ps(_mm_add_ps(fvdw6,fvdw12),_mm_mul_ps(vftabscale,rinv00)));
 301
 302             /* Update potential sum for this i atom from the interaction with this j atom. */
 303             velecsum         = _mm_add_ps(velecsum,velec);
 304             vvdwsum          = _mm_add_ps(vvdwsum,vvdw);
 305
 306             fscal            = _mm_add_ps(felec,fvdw);
 307
 308              /* Update vectorial force */
 309             fix0             = _mm_macc_ps(dx00,fscal,fix0);
 310             fiy0             = _mm_macc_ps(dy00,fscal,fiy0);
 311             fiz0             = _mm_macc_ps(dz00,fscal,fiz0);
 312
 313             fjx0             = _mm_macc_ps(dx00,fscal,fjx0);
 314             fjy0             = _mm_macc_ps(dy00,fscal,fjy0);
 315             fjz0             = _mm_macc_ps(dz00,fscal,fjz0);
 316
 317             /**************************
 318              * CALCULATE INTERACTIONS *
 319              **************************/
 320
 321             r10              = _mm_mul_ps(rsq10,rinv10);
 322
 323             /* Compute parameters for interactions between i and j atoms */
 324             qq10             = _mm_mul_ps(iq1,jq0);
 325
 326             /* Calculate table index by multiplying r with table scale and truncate to integer */
 327             rt               = _mm_mul_ps(r10,vftabscale);
 328             vfitab           = _mm_cvttps_epi32(rt);
 329 #ifdef __XOP__
 330             vfeps            = _mm_frcz_ps(rt);
 331 #else
 332             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
 333 #endif
 334             twovfeps         = _mm_add_ps(vfeps,vfeps);
 335             vfitab           = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
 336
 337             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 338             Y                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
 339             F                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
 340             G                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
 341             H                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
 342             _MM_TRANSPOSE4_PS(Y,F,G,H);
 343             Fp               = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
 344             VV               = _mm_macc_ps(vfeps,Fp,Y);
 345             velec            = _mm_mul_ps(qq10,VV);
 346             FF               = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
 347             felec            = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq10,FF),_mm_mul_ps(vftabscale,rinv10)));
 348
 349             /* Update potential sum for this i atom from the interaction with this j atom. */
 350             velecsum         = _mm_add_ps(velecsum,velec);
 351
 352             fscal            = felec;
 353
 354              /* Update vectorial force */
 355             fix1             = _mm_macc_ps(dx10,fscal,fix1);
 356             fiy1             = _mm_macc_ps(dy10,fscal,fiy1);
 357             fiz1             = _mm_macc_ps(dz10,fscal,fiz1);
 358
 359             fjx0             = _mm_macc_ps(dx10,fscal,fjx0);
 360             fjy0             = _mm_macc_ps(dy10,fscal,fjy0);
 361             fjz0             = _mm_macc_ps(dz10,fscal,fjz0);
 362
 363             /**************************
 364              * CALCULATE INTERACTIONS *
 365              **************************/
 366
 367             r20              = _mm_mul_ps(rsq20,rinv20);
 368
 369             /* Compute parameters for interactions between i and j atoms */
 370             qq20             = _mm_mul_ps(iq2,jq0);
 371
 372             /* Calculate table index by multiplying r with table scale and truncate to integer */
 373             rt               = _mm_mul_ps(r20,vftabscale);
 374             vfitab           = _mm_cvttps_epi32(rt);
 375 #ifdef __XOP__
 376             vfeps            = _mm_frcz_ps(rt);
 377 #else
 378             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
 379 #endif
 380             twovfeps         = _mm_add_ps(vfeps,vfeps);
 381             vfitab           = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
 382
 383             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 384             Y                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
 385             F                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
 386             G                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
 387             H                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
 388             _MM_TRANSPOSE4_PS(Y,F,G,H);
 389             Fp               = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
 390             VV               = _mm_macc_ps(vfeps,Fp,Y);
 391             velec            = _mm_mul_ps(qq20,VV);
 392             FF               = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
 393             felec            = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq20,FF),_mm_mul_ps(vftabscale,rinv20)));
 394
 395             /* Update potential sum for this i atom from the interaction with this j atom. */
 396             velecsum         = _mm_add_ps(velecsum,velec);
 397
 398             fscal            = felec;
 399
 400              /* Update vectorial force */
 401             fix2             = _mm_macc_ps(dx20,fscal,fix2);
 402             fiy2             = _mm_macc_ps(dy20,fscal,fiy2);
 403             fiz2             = _mm_macc_ps(dz20,fscal,fiz2);
 404
 405             fjx0             = _mm_macc_ps(dx20,fscal,fjx0);
 406             fjy0             = _mm_macc_ps(dy20,fscal,fjy0);
 407             fjz0             = _mm_macc_ps(dz20,fscal,fjz0);
 408
 409             fjptrA             = f+j_coord_offsetA;
 410             fjptrB             = f+j_coord_offsetB;
 411             fjptrC             = f+j_coord_offsetC;
 412             fjptrD             = f+j_coord_offsetD;
 413
 414             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 415
 416             /* Inner loop uses 168 flops */
 417         }
 418
 419         if(jidx<j_index_end)
 420         {
 421
 422             /* Get j neighbor index, and coordinate index */
 423             jnrlistA         = jjnr[jidx];
 424             jnrlistB         = jjnr[jidx+1];
 425             jnrlistC         = jjnr[jidx+2];
 426             jnrlistD         = jjnr[jidx+3];
 427             /* Sign of each element will be negative for non-real atoms.
 428              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 429              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 430              */
 431             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 432             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 433             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 434             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 435             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 436             j_coord_offsetA  = DIM*jnrA;
 437             j_coord_offsetB  = DIM*jnrB;
 438             j_coord_offsetC  = DIM*jnrC;
 439             j_coord_offsetD  = DIM*jnrD;
 440
 441             /* load j atom coordinates */
 442             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 443                                               x+j_coord_offsetC,x+j_coord_offsetD,
 444                                               &jx0,&jy0,&jz0);
 445
 446             /* Calculate displacement vector */
 447             dx00             = _mm_sub_ps(ix0,jx0);
 448             dy00             = _mm_sub_ps(iy0,jy0);
 449             dz00             = _mm_sub_ps(iz0,jz0);
 450             dx10             = _mm_sub_ps(ix1,jx0);
 451             dy10             = _mm_sub_ps(iy1,jy0);
 452             dz10             = _mm_sub_ps(iz1,jz0);
 453             dx20             = _mm_sub_ps(ix2,jx0);
 454             dy20             = _mm_sub_ps(iy2,jy0);
 455             dz20             = _mm_sub_ps(iz2,jz0);
 456
 457             /* Calculate squared distance and things based on it */
 458             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 459             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 460             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 461
 462             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 463             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 464             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 465
 466             /* Load parameters for j particles */
 467             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 468                                                               charge+jnrC+0,charge+jnrD+0);
 469             vdwjidx0A        = 2*vdwtype[jnrA+0];
 470             vdwjidx0B        = 2*vdwtype[jnrB+0];
 471             vdwjidx0C        = 2*vdwtype[jnrC+0];
 472             vdwjidx0D        = 2*vdwtype[jnrD+0];
 473
 474             fjx0             = _mm_setzero_ps();
 475             fjy0             = _mm_setzero_ps();
 476             fjz0             = _mm_setzero_ps();
 477
 478             /**************************
 479              * CALCULATE INTERACTIONS *
 480              **************************/
 481
 482             r00              = _mm_mul_ps(rsq00,rinv00);
 483             r00              = _mm_andnot_ps(dummy_mask,r00);
 484
 485             /* Compute parameters for interactions between i and j atoms */
 486             qq00             = _mm_mul_ps(iq0,jq0);
 487             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 488                                          vdwparam+vdwioffset0+vdwjidx0B,
 489                                          vdwparam+vdwioffset0+vdwjidx0C,
 490                                          vdwparam+vdwioffset0+vdwjidx0D,
 491                                          &c6_00,&c12_00);
 492
 493             /* Calculate table index by multiplying r with table scale and truncate to integer */
 494             rt               = _mm_mul_ps(r00,vftabscale);
 495             vfitab           = _mm_cvttps_epi32(rt);
 496 #ifdef __XOP__
 497             vfeps            = _mm_frcz_ps(rt);
 498 #else
 499             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
 500 #endif
 501             twovfeps         = _mm_add_ps(vfeps,vfeps);
 502             vfitab           = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
 503
 504             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 505             Y                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
 506             F                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
 507             G                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
 508             H                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
 509             _MM_TRANSPOSE4_PS(Y,F,G,H);
 510             Fp               = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
 511             VV               = _mm_macc_ps(vfeps,Fp,Y);
 512             velec            = _mm_mul_ps(qq00,VV);
 513             FF               = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
 514             felec            = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq00,FF),_mm_mul_ps(vftabscale,rinv00)));
 515
 516             /* CUBIC SPLINE TABLE DISPERSION */
 517             vfitab           = _mm_add_epi32(vfitab,ifour);
 518             Y                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
 519             F                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
 520             G                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
 521             H                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
 522             _MM_TRANSPOSE4_PS(Y,F,G,H);
 523             Fp               = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
 524             VV               = _mm_macc_ps(vfeps,Fp,Y);
 525             vvdw6            = _mm_mul_ps(c6_00,VV);
 526             FF               = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
 527             fvdw6            = _mm_mul_ps(c6_00,FF);
 528
 529             /* CUBIC SPLINE TABLE REPULSION */
 530             vfitab           = _mm_add_epi32(vfitab,ifour);
 531             Y                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
 532             F                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
 533             G                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
 534             H                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
 535             _MM_TRANSPOSE4_PS(Y,F,G,H);
 536             Fp               = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
 537             VV               = _mm_macc_ps(vfeps,Fp,Y);
 538             vvdw12           = _mm_mul_ps(c12_00,VV);
 539             FF               = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
 540             fvdw12           = _mm_mul_ps(c12_00,FF);
 541             vvdw             = _mm_add_ps(vvdw12,vvdw6);
 542             fvdw             = _mm_xor_ps(signbit,_mm_mul_ps(_mm_add_ps(fvdw6,fvdw12),_mm_mul_ps(vftabscale,rinv00)));
 543
 544             /* Update potential sum for this i atom from the interaction with this j atom. */
 545             velec            = _mm_andnot_ps(dummy_mask,velec);
 546             velecsum         = _mm_add_ps(velecsum,velec);
 547             vvdw             = _mm_andnot_ps(dummy_mask,vvdw);
 548             vvdwsum          = _mm_add_ps(vvdwsum,vvdw);
 549
 550             fscal            = _mm_add_ps(felec,fvdw);
 551
 552             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 553
 554              /* Update vectorial force */
 555             fix0             = _mm_macc_ps(dx00,fscal,fix0);
 556             fiy0             = _mm_macc_ps(dy00,fscal,fiy0);
 557             fiz0             = _mm_macc_ps(dz00,fscal,fiz0);
 558
 559             fjx0             = _mm_macc_ps(dx00,fscal,fjx0);
 560             fjy0             = _mm_macc_ps(dy00,fscal,fjy0);
 561             fjz0             = _mm_macc_ps(dz00,fscal,fjz0);
 562
 563             /**************************
 564              * CALCULATE INTERACTIONS *
 565              **************************/
 566
 567             r10              = _mm_mul_ps(rsq10,rinv10);
 568             r10              = _mm_andnot_ps(dummy_mask,r10);
 569
 570             /* Compute parameters for interactions between i and j atoms */
 571             qq10             = _mm_mul_ps(iq1,jq0);
 572
 573             /* Calculate table index by multiplying r with table scale and truncate to integer */
 574             rt               = _mm_mul_ps(r10,vftabscale);
 575             vfitab           = _mm_cvttps_epi32(rt);
 576 #ifdef __XOP__
 577             vfeps            = _mm_frcz_ps(rt);
 578 #else
 579             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
 580 #endif
 581             twovfeps         = _mm_add_ps(vfeps,vfeps);
 582             vfitab           = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
 583
 584             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 585             Y                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
 586             F                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
 587             G                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
 588             H                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
 589             _MM_TRANSPOSE4_PS(Y,F,G,H);
 590             Fp               = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
 591             VV               = _mm_macc_ps(vfeps,Fp,Y);
 592             velec            = _mm_mul_ps(qq10,VV);
 593             FF               = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
 594             felec            = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq10,FF),_mm_mul_ps(vftabscale,rinv10)));
 595
 596             /* Update potential sum for this i atom from the interaction with this j atom. */
 597             velec            = _mm_andnot_ps(dummy_mask,velec);
 598             velecsum         = _mm_add_ps(velecsum,velec);
 599
 600             fscal            = felec;
 601
 602             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 603
 604              /* Update vectorial force */
 605             fix1             = _mm_macc_ps(dx10,fscal,fix1);
 606             fiy1             = _mm_macc_ps(dy10,fscal,fiy1);
 607             fiz1             = _mm_macc_ps(dz10,fscal,fiz1);
 608
 609             fjx0             = _mm_macc_ps(dx10,fscal,fjx0);
 610             fjy0             = _mm_macc_ps(dy10,fscal,fjy0);
 611             fjz0             = _mm_macc_ps(dz10,fscal,fjz0);
 612
 613             /**************************
 614              * CALCULATE INTERACTIONS *
 615              **************************/
 616
 617             r20              = _mm_mul_ps(rsq20,rinv20);
 618             r20              = _mm_andnot_ps(dummy_mask,r20);
 619
 620             /* Compute parameters for interactions between i and j atoms */
 621             qq20             = _mm_mul_ps(iq2,jq0);
 622
 623             /* Calculate table index by multiplying r with table scale and truncate to integer */
 624             rt               = _mm_mul_ps(r20,vftabscale);
 625             vfitab           = _mm_cvttps_epi32(rt);
 626 #ifdef __XOP__
 627             vfeps            = _mm_frcz_ps(rt);
 628 #else
 629             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
 630 #endif
 631             twovfeps         = _mm_add_ps(vfeps,vfeps);
 632             vfitab           = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
 633
 634             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 635             Y                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
 636             F                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
 637             G                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
 638             H                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
 639             _MM_TRANSPOSE4_PS(Y,F,G,H);
 640             Fp               = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
 641             VV               = _mm_macc_ps(vfeps,Fp,Y);
 642             velec            = _mm_mul_ps(qq20,VV);
 643             FF               = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
 644             felec            = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq20,FF),_mm_mul_ps(vftabscale,rinv20)));
 645
 646             /* Update potential sum for this i atom from the interaction with this j atom. */
 647             velec            = _mm_andnot_ps(dummy_mask,velec);
 648             velecsum         = _mm_add_ps(velecsum,velec);
 649
 650             fscal            = felec;
 651
 652             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 653
 654              /* Update vectorial force */
 655             fix2             = _mm_macc_ps(dx20,fscal,fix2);
 656             fiy2             = _mm_macc_ps(dy20,fscal,fiy2);
 657             fiz2             = _mm_macc_ps(dz20,fscal,fiz2);
 658
 659             fjx0             = _mm_macc_ps(dx20,fscal,fjx0);
 660             fjy0             = _mm_macc_ps(dy20,fscal,fjy0);
 661             fjz0             = _mm_macc_ps(dz20,fscal,fjz0);
 662
 663             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 664             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 665             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 666             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 667
 668             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 669
 670             /* Inner loop uses 171 flops */
 671         }
 672
 673         /* End of innermost loop */
 674
 675         gmx_mm_update_iforce_3atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
 676                                               f+i_coord_offset,fshift+i_shift_offset);
 677
 678         ggid                        = gid[iidx];
 679         /* Update potential energies */
 680         gmx_mm_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid);
 681         gmx_mm_update_1pot_ps(vvdwsum,kernel_data->energygrp_vdw+ggid);
 682
 683         /* Increment number of inner iterations */
 684         inneriter                  += j_index_end - j_index_start;
 685
 686         /* Outer loop uses 20 flops */
 687     }
 688
 689     /* Increment number of outer iterations */
 690     outeriter        += nri;
 691
 692     /* Update outer/inner flops */
 693
 694     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3_VF,outeriter*20 + inneriter*171);
 695 }
 696 /*
 697  * Gromacs nonbonded kernel:   nb_kernel_ElecCSTab_VdwCSTab_GeomW3P1_F_avx_128_fma_single
 698  * Electrostatics interaction: CubicSplineTable
 699  * VdW interaction:            CubicSplineTable
 700  * Geometry:                   Water3-Particle
 701  * Calculate force/pot:        Force
 702  */
 703 void
 704 nb_kernel_ElecCSTab_VdwCSTab_GeomW3P1_F_avx_128_fma_single
 705                     (t_nblist                    * gmx_restrict       nlist,
 706                      rvec                        * gmx_restrict          xx,
 707                      rvec                        * gmx_restrict          ff,
 708                      t_forcerec                  * gmx_restrict          fr,
 709                      t_mdatoms                   * gmx_restrict     mdatoms,
 710                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 711                      t_nrnb                      * gmx_restrict        nrnb)
 712 {
 713     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 714      * just 0 for non-waters.
 715      * Suffixes A,B,C,D refer to j loop unrolling done with AVX_128, e.g. for the four different
 716      * jnr indices corresponding to data put in the four positions in the SIMD register.
 717      */
 718     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 719     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 720     int              jnrA,jnrB,jnrC,jnrD;
 721     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
 722     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 723     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 724     real             rcutoff_scalar;
 725     real             *shiftvec,*fshift,*x,*f;
 726     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
 727     real             scratch[4*DIM];
 728     __m128           fscal,rcutoff,rcutoff2,jidxall;
 729     int              vdwioffset0;
 730     __m128           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 731     int              vdwioffset1;
 732     __m128           ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
 733     int              vdwioffset2;
 734     __m128           ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
 735     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
 736     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 737     __m128           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 738     __m128           dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
 739     __m128           dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
 740     __m128           velec,felec,velecsum,facel,crf,krf,krf2;
 741     real             *charge;
 742     int              nvdwtype;
 743     __m128           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 744     int              *vdwtype;
 745     real             *vdwparam;
 746     __m128           one_sixth   = _mm_set1_ps(1.0/6.0);
 747     __m128           one_twelfth = _mm_set1_ps(1.0/12.0);
 748     __m128i          vfitab;
 749     __m128i          ifour       = _mm_set1_epi32(4);
 750     __m128           rt,vfeps,twovfeps,vftabscale,Y,F,G,H,Fp,VV,FF;
 751     real             *vftab;
 752     __m128           dummy_mask,cutoff_mask;
 753     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
 754     __m128           one     = _mm_set1_ps(1.0);
 755     __m128           two     = _mm_set1_ps(2.0);
 756     x                = xx[0];
 757     f                = ff[0];
 758
 759     nri              = nlist->nri;
 760     iinr             = nlist->iinr;
 761     jindex           = nlist->jindex;
 762     jjnr             = nlist->jjnr;
 763     shiftidx         = nlist->shift;
 764     gid              = nlist->gid;
 765     shiftvec         = fr->shift_vec[0];
 766     fshift           = fr->fshift[0];
 767     facel            = _mm_set1_ps(fr->epsfac);
 768     charge           = mdatoms->chargeA;
 769     nvdwtype         = fr->ntype;
 770     vdwparam         = fr->nbfp;
 771     vdwtype          = mdatoms->typeA;
 772
 773     vftab            = kernel_data->table_elec_vdw->data;
 774     vftabscale       = _mm_set1_ps(kernel_data->table_elec_vdw->scale);
 775
 776     /* Setup water-specific parameters */
 777     inr              = nlist->iinr[0];
 778     iq0              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+0]));
 779     iq1              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+1]));
 780     iq2              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+2]));
 781     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 782
 783     /* Avoid stupid compiler warnings */
 784     jnrA = jnrB = jnrC = jnrD = 0;
 785     j_coord_offsetA = 0;
 786     j_coord_offsetB = 0;
 787     j_coord_offsetC = 0;
 788     j_coord_offsetD = 0;
 789
 790     outeriter        = 0;
 791     inneriter        = 0;
 792
 793     for(iidx=0;iidx<4*DIM;iidx++)
 794     {
 795         scratch[iidx] = 0.0;
 796     }
 797
 798     /* Start outer loop over neighborlists */
 799     for(iidx=0; iidx<nri; iidx++)
 800     {
 801         /* Load shift vector for this list */
 802         i_shift_offset   = DIM*shiftidx[iidx];
 803
 804         /* Load limits for loop over neighbors */
 805         j_index_start    = jindex[iidx];
 806         j_index_end      = jindex[iidx+1];
 807
 808         /* Get outer coordinate index */
 809         inr              = iinr[iidx];
 810         i_coord_offset   = DIM*inr;
 811
 812         /* Load i particle coords and add shift vector */
 813         gmx_mm_load_shift_and_3rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
 814                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 815
 816         fix0             = _mm_setzero_ps();
 817         fiy0             = _mm_setzero_ps();
 818         fiz0             = _mm_setzero_ps();
 819         fix1             = _mm_setzero_ps();
 820         fiy1             = _mm_setzero_ps();
 821         fiz1             = _mm_setzero_ps();
 822         fix2             = _mm_setzero_ps();
 823         fiy2             = _mm_setzero_ps();
 824         fiz2             = _mm_setzero_ps();
 825
 826         /* Start inner kernel loop */
 827         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 828         {
 829
 830             /* Get j neighbor index, and coordinate index */
 831             jnrA             = jjnr[jidx];
 832             jnrB             = jjnr[jidx+1];
 833             jnrC             = jjnr[jidx+2];
 834             jnrD             = jjnr[jidx+3];
 835             j_coord_offsetA  = DIM*jnrA;
 836             j_coord_offsetB  = DIM*jnrB;
 837             j_coord_offsetC  = DIM*jnrC;
 838             j_coord_offsetD  = DIM*jnrD;
 839
 840             /* load j atom coordinates */
 841             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 842                                               x+j_coord_offsetC,x+j_coord_offsetD,
 843                                               &jx0,&jy0,&jz0);
 844
 845             /* Calculate displacement vector */
 846             dx00             = _mm_sub_ps(ix0,jx0);
 847             dy00             = _mm_sub_ps(iy0,jy0);
 848             dz00             = _mm_sub_ps(iz0,jz0);
 849             dx10             = _mm_sub_ps(ix1,jx0);
 850             dy10             = _mm_sub_ps(iy1,jy0);
 851             dz10             = _mm_sub_ps(iz1,jz0);
 852             dx20             = _mm_sub_ps(ix2,jx0);
 853             dy20             = _mm_sub_ps(iy2,jy0);
 854             dz20             = _mm_sub_ps(iz2,jz0);
 855
 856             /* Calculate squared distance and things based on it */
 857             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 858             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 859             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 860
 861             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 862             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 863             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 864
 865             /* Load parameters for j particles */
 866             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 867                                                               charge+jnrC+0,charge+jnrD+0);
 868             vdwjidx0A        = 2*vdwtype[jnrA+0];
 869             vdwjidx0B        = 2*vdwtype[jnrB+0];
 870             vdwjidx0C        = 2*vdwtype[jnrC+0];
 871             vdwjidx0D        = 2*vdwtype[jnrD+0];
 872
 873             fjx0             = _mm_setzero_ps();
 874             fjy0             = _mm_setzero_ps();
 875             fjz0             = _mm_setzero_ps();
 876
 877             /**************************
 878              * CALCULATE INTERACTIONS *
 879              **************************/
 880
 881             r00              = _mm_mul_ps(rsq00,rinv00);
 882
 883             /* Compute parameters for interactions between i and j atoms */
 884             qq00             = _mm_mul_ps(iq0,jq0);
 885             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 886                                          vdwparam+vdwioffset0+vdwjidx0B,
 887                                          vdwparam+vdwioffset0+vdwjidx0C,
 888                                          vdwparam+vdwioffset0+vdwjidx0D,
 889                                          &c6_00,&c12_00);
 890
 891             /* Calculate table index by multiplying r with table scale and truncate to integer */
 892             rt               = _mm_mul_ps(r00,vftabscale);
 893             vfitab           = _mm_cvttps_epi32(rt);
 894 #ifdef __XOP__
 895             vfeps            = _mm_frcz_ps(rt);
 896 #else
 897             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
 898 #endif
 899             twovfeps         = _mm_add_ps(vfeps,vfeps);
 900             vfitab           = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
 901
 902             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 903             Y                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
 904             F                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
 905             G                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
 906             H                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
 907             _MM_TRANSPOSE4_PS(Y,F,G,H);
 908             Fp               = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
 909             FF               = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
 910             felec            = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq00,FF),_mm_mul_ps(vftabscale,rinv00)));
 911
 912             /* CUBIC SPLINE TABLE DISPERSION */
 913             vfitab           = _mm_add_epi32(vfitab,ifour);
 914             Y                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
 915             F                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
 916             G                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
 917             H                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
 918             _MM_TRANSPOSE4_PS(Y,F,G,H);
 919             Fp               = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
 920             FF               = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
 921             fvdw6            = _mm_mul_ps(c6_00,FF);
 922
 923             /* CUBIC SPLINE TABLE REPULSION */
 924             vfitab           = _mm_add_epi32(vfitab,ifour);
 925             Y                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
 926             F                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
 927             G                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
 928             H                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
 929             _MM_TRANSPOSE4_PS(Y,F,G,H);
 930             Fp               = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
 931             FF               = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
 932             fvdw12           = _mm_mul_ps(c12_00,FF);
 933             fvdw             = _mm_xor_ps(signbit,_mm_mul_ps(_mm_add_ps(fvdw6,fvdw12),_mm_mul_ps(vftabscale,rinv00)));
 934
 935             fscal            = _mm_add_ps(felec,fvdw);
 936
 937              /* Update vectorial force */
 938             fix0             = _mm_macc_ps(dx00,fscal,fix0);
 939             fiy0             = _mm_macc_ps(dy00,fscal,fiy0);
 940             fiz0             = _mm_macc_ps(dz00,fscal,fiz0);
 941
 942             fjx0             = _mm_macc_ps(dx00,fscal,fjx0);
 943             fjy0             = _mm_macc_ps(dy00,fscal,fjy0);
 944             fjz0             = _mm_macc_ps(dz00,fscal,fjz0);
 945
 946             /**************************
 947              * CALCULATE INTERACTIONS *
 948              **************************/
 949
 950             r10              = _mm_mul_ps(rsq10,rinv10);
 951
 952             /* Compute parameters for interactions between i and j atoms */
 953             qq10             = _mm_mul_ps(iq1,jq0);
 954
 955             /* Calculate table index by multiplying r with table scale and truncate to integer */
 956             rt               = _mm_mul_ps(r10,vftabscale);
 957             vfitab           = _mm_cvttps_epi32(rt);
 958 #ifdef __XOP__
 959             vfeps            = _mm_frcz_ps(rt);
 960 #else
 961             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
 962 #endif
 963             twovfeps         = _mm_add_ps(vfeps,vfeps);
 964             vfitab           = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
 965
 966             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 967             Y                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
 968             F                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
 969             G                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
 970             H                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
 971             _MM_TRANSPOSE4_PS(Y,F,G,H);
 972             Fp               = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
 973             FF               = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
 974             felec            = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq10,FF),_mm_mul_ps(vftabscale,rinv10)));
 975
 976             fscal            = felec;
 977
 978              /* Update vectorial force */
 979             fix1             = _mm_macc_ps(dx10,fscal,fix1);
 980             fiy1             = _mm_macc_ps(dy10,fscal,fiy1);
 981             fiz1             = _mm_macc_ps(dz10,fscal,fiz1);
 982
 983             fjx0             = _mm_macc_ps(dx10,fscal,fjx0);
 984             fjy0             = _mm_macc_ps(dy10,fscal,fjy0);
 985             fjz0             = _mm_macc_ps(dz10,fscal,fjz0);
 986
 987             /**************************
 988              * CALCULATE INTERACTIONS *
 989              **************************/
 990
 991             r20              = _mm_mul_ps(rsq20,rinv20);
 992
 993             /* Compute parameters for interactions between i and j atoms */
 994             qq20             = _mm_mul_ps(iq2,jq0);
 995
 996             /* Calculate table index by multiplying r with table scale and truncate to integer */
 997             rt               = _mm_mul_ps(r20,vftabscale);
 998             vfitab           = _mm_cvttps_epi32(rt);
 999 #ifdef __XOP__
1000             vfeps            = _mm_frcz_ps(rt);
1001 #else
1002             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
1003 #endif
1004             twovfeps         = _mm_add_ps(vfeps,vfeps);
1005             vfitab           = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
1006
1007             /* CUBIC SPLINE TABLE ELECTROSTATICS */
1008             Y                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
1009             F                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
1010             G                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
1011             H                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
1012             _MM_TRANSPOSE4_PS(Y,F,G,H);
1013             Fp               = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
1014             FF               = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
1015             felec            = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq20,FF),_mm_mul_ps(vftabscale,rinv20)));
1016
1017             fscal            = felec;
1018
1019              /* Update vectorial force */
1020             fix2             = _mm_macc_ps(dx20,fscal,fix2);
1021             fiy2             = _mm_macc_ps(dy20,fscal,fiy2);
1022             fiz2             = _mm_macc_ps(dz20,fscal,fiz2);
1023
1024             fjx0             = _mm_macc_ps(dx20,fscal,fjx0);
1025             fjy0             = _mm_macc_ps(dy20,fscal,fjy0);
1026             fjz0             = _mm_macc_ps(dz20,fscal,fjz0);
1027
1028             fjptrA             = f+j_coord_offsetA;
1029             fjptrB             = f+j_coord_offsetB;
1030             fjptrC             = f+j_coord_offsetC;
1031             fjptrD             = f+j_coord_offsetD;
1032
1033             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
1034
1035             /* Inner loop uses 148 flops */
1036         }
1037
1038         if(jidx<j_index_end)
1039         {
1040
1041             /* Get j neighbor index, and coordinate index */
1042             jnrlistA         = jjnr[jidx];
1043             jnrlistB         = jjnr[jidx+1];
1044             jnrlistC         = jjnr[jidx+2];
1045             jnrlistD         = jjnr[jidx+3];
1046             /* Sign of each element will be negative for non-real atoms.
1047              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
1048              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
1049              */
1050             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
1051             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
1052             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
1053             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
1054             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
1055             j_coord_offsetA  = DIM*jnrA;
1056             j_coord_offsetB  = DIM*jnrB;
1057             j_coord_offsetC  = DIM*jnrC;
1058             j_coord_offsetD  = DIM*jnrD;
1059
1060             /* load j atom coordinates */
1061             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
1062                                               x+j_coord_offsetC,x+j_coord_offsetD,
1063                                               &jx0,&jy0,&jz0);
1064
1065             /* Calculate displacement vector */
1066             dx00             = _mm_sub_ps(ix0,jx0);
1067             dy00             = _mm_sub_ps(iy0,jy0);
1068             dz00             = _mm_sub_ps(iz0,jz0);
1069             dx10             = _mm_sub_ps(ix1,jx0);
1070             dy10             = _mm_sub_ps(iy1,jy0);
1071             dz10             = _mm_sub_ps(iz1,jz0);
1072             dx20             = _mm_sub_ps(ix2,jx0);
1073             dy20             = _mm_sub_ps(iy2,jy0);
1074             dz20             = _mm_sub_ps(iz2,jz0);
1075
1076             /* Calculate squared distance and things based on it */
1077             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
1078             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
1079             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
1080
1081             rinv00           = gmx_mm_invsqrt_ps(rsq00);
1082             rinv10           = gmx_mm_invsqrt_ps(rsq10);
1083             rinv20           = gmx_mm_invsqrt_ps(rsq20);
1084
1085             /* Load parameters for j particles */
1086             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
1087                                                               charge+jnrC+0,charge+jnrD+0);
1088             vdwjidx0A        = 2*vdwtype[jnrA+0];
1089             vdwjidx0B        = 2*vdwtype[jnrB+0];
1090             vdwjidx0C        = 2*vdwtype[jnrC+0];
1091             vdwjidx0D        = 2*vdwtype[jnrD+0];
1092
1093             fjx0             = _mm_setzero_ps();
1094             fjy0             = _mm_setzero_ps();
1095             fjz0             = _mm_setzero_ps();
1096
1097             /**************************
1098              * CALCULATE INTERACTIONS *
1099              **************************/
1100
1101             r00              = _mm_mul_ps(rsq00,rinv00);
1102             r00              = _mm_andnot_ps(dummy_mask,r00);
1103
1104             /* Compute parameters for interactions between i and j atoms */
1105             qq00             = _mm_mul_ps(iq0,jq0);
1106             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
1107                                          vdwparam+vdwioffset0+vdwjidx0B,
1108                                          vdwparam+vdwioffset0+vdwjidx0C,
1109                                          vdwparam+vdwioffset0+vdwjidx0D,
1110                                          &c6_00,&c12_00);
1111
1112             /* Calculate table index by multiplying r with table scale and truncate to integer */
1113             rt               = _mm_mul_ps(r00,vftabscale);
1114             vfitab           = _mm_cvttps_epi32(rt);
1115 #ifdef __XOP__
1116             vfeps            = _mm_frcz_ps(rt);
1117 #else
1118             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
1119 #endif
1120             twovfeps         = _mm_add_ps(vfeps,vfeps);
1121             vfitab           = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
1122
1123             /* CUBIC SPLINE TABLE ELECTROSTATICS */
1124             Y                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
1125             F                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
1126             G                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
1127             H                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
1128             _MM_TRANSPOSE4_PS(Y,F,G,H);
1129             Fp               = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
1130             FF               = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
1131             felec            = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq00,FF),_mm_mul_ps(vftabscale,rinv00)));
1132
1133             /* CUBIC SPLINE TABLE DISPERSION */
1134             vfitab           = _mm_add_epi32(vfitab,ifour);
1135             Y                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
1136             F                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
1137             G                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
1138             H                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
1139             _MM_TRANSPOSE4_PS(Y,F,G,H);
1140             Fp               = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
1141             FF               = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
1142             fvdw6            = _mm_mul_ps(c6_00,FF);
1143
1144             /* CUBIC SPLINE TABLE REPULSION */
1145             vfitab           = _mm_add_epi32(vfitab,ifour);
1146             Y                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
1147             F                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
1148             G                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
1149             H                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
1150             _MM_TRANSPOSE4_PS(Y,F,G,H);
1151             Fp               = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
1152             FF               = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
1153             fvdw12           = _mm_mul_ps(c12_00,FF);
1154             fvdw             = _mm_xor_ps(signbit,_mm_mul_ps(_mm_add_ps(fvdw6,fvdw12),_mm_mul_ps(vftabscale,rinv00)));
1155
1156             fscal            = _mm_add_ps(felec,fvdw);
1157
1158             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1159
1160              /* Update vectorial force */
1161             fix0             = _mm_macc_ps(dx00,fscal,fix0);
1162             fiy0             = _mm_macc_ps(dy00,fscal,fiy0);
1163             fiz0             = _mm_macc_ps(dz00,fscal,fiz0);
1164
1165             fjx0             = _mm_macc_ps(dx00,fscal,fjx0);
1166             fjy0             = _mm_macc_ps(dy00,fscal,fjy0);
1167             fjz0             = _mm_macc_ps(dz00,fscal,fjz0);
1168
1169             /**************************
1170              * CALCULATE INTERACTIONS *
1171              **************************/
1172
1173             r10              = _mm_mul_ps(rsq10,rinv10);
1174             r10              = _mm_andnot_ps(dummy_mask,r10);
1175
1176             /* Compute parameters for interactions between i and j atoms */
1177             qq10             = _mm_mul_ps(iq1,jq0);
1178
1179             /* Calculate table index by multiplying r with table scale and truncate to integer */
1180             rt               = _mm_mul_ps(r10,vftabscale);
1181             vfitab           = _mm_cvttps_epi32(rt);
1182 #ifdef __XOP__
1183             vfeps            = _mm_frcz_ps(rt);
1184 #else
1185             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
1186 #endif
1187             twovfeps         = _mm_add_ps(vfeps,vfeps);
1188             vfitab           = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
1189
1190             /* CUBIC SPLINE TABLE ELECTROSTATICS */
1191             Y                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
1192             F                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
1193             G                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
1194             H                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
1195             _MM_TRANSPOSE4_PS(Y,F,G,H);
1196             Fp               = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
1197             FF               = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
1198             felec            = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq10,FF),_mm_mul_ps(vftabscale,rinv10)));
1199
1200             fscal            = felec;
1201
1202             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1203
1204              /* Update vectorial force */
1205             fix1             = _mm_macc_ps(dx10,fscal,fix1);
1206             fiy1             = _mm_macc_ps(dy10,fscal,fiy1);
1207             fiz1             = _mm_macc_ps(dz10,fscal,fiz1);
1208
1209             fjx0             = _mm_macc_ps(dx10,fscal,fjx0);
1210             fjy0             = _mm_macc_ps(dy10,fscal,fjy0);
1211             fjz0             = _mm_macc_ps(dz10,fscal,fjz0);
1212
1213             /**************************
1214              * CALCULATE INTERACTIONS *
1215              **************************/
1216
1217             r20              = _mm_mul_ps(rsq20,rinv20);
1218             r20              = _mm_andnot_ps(dummy_mask,r20);
1219
1220             /* Compute parameters for interactions between i and j atoms */
1221             qq20             = _mm_mul_ps(iq2,jq0);
1222
1223             /* Calculate table index by multiplying r with table scale and truncate to integer */
1224             rt               = _mm_mul_ps(r20,vftabscale);
1225             vfitab           = _mm_cvttps_epi32(rt);
1226 #ifdef __XOP__
1227             vfeps            = _mm_frcz_ps(rt);
1228 #else
1229             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
1230 #endif
1231             twovfeps         = _mm_add_ps(vfeps,vfeps);
1232             vfitab           = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
1233
1234             /* CUBIC SPLINE TABLE ELECTROSTATICS */
1235             Y                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
1236             F                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
1237             G                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
1238             H                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
1239             _MM_TRANSPOSE4_PS(Y,F,G,H);
1240             Fp               = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
1241             FF               = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
1242             felec            = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq20,FF),_mm_mul_ps(vftabscale,rinv20)));
1243
1244             fscal            = felec;
1245
1246             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1247
1248              /* Update vectorial force */
1249             fix2             = _mm_macc_ps(dx20,fscal,fix2);
1250             fiy2             = _mm_macc_ps(dy20,fscal,fiy2);
1251             fiz2             = _mm_macc_ps(dz20,fscal,fiz2);
1252
1253             fjx0             = _mm_macc_ps(dx20,fscal,fjx0);
1254             fjy0             = _mm_macc_ps(dy20,fscal,fjy0);
1255             fjz0             = _mm_macc_ps(dz20,fscal,fjz0);
1256
1257             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
1258             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
1259             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
1260             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
1261
1262             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
1263
1264             /* Inner loop uses 151 flops */
1265         }
1266
1267         /* End of innermost loop */
1268
1269         gmx_mm_update_iforce_3atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
1270                                               f+i_coord_offset,fshift+i_shift_offset);
1271
1272         /* Increment number of inner iterations */
1273         inneriter                  += j_index_end - j_index_start;
1274
1275         /* Outer loop uses 18 flops */
1276     }
1277
1278     /* Increment number of outer iterations */
1279     outeriter        += nri;
1280
1281     /* Update outer/inner flops */
1282
1283     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3_F,outeriter*18 + inneriter*151);
1284 }