src/gromacs/gmxlib/nonbonded/nb_kernel_sse4_1_single/nb_kernel_ElecEw_VdwCSTab_GeomW3P1_sse4_1_single.c

   1 /*
   2  * This file is part of the GROMACS molecular simulation package.
   3  *
   4  * Copyright (c) 2012,2013,2014,2015,2017, by the GROMACS development team, led by
   5  * Mark Abraham, David van der Spoel, Berk Hess, and Erik Lindahl,
   6  * and including many others, as listed in the AUTHORS file in the
   7  * top-level source directory and at http://www.gromacs.org.
   8  *
   9  * GROMACS is free software; you can redistribute it and/or
  10  * modify it under the terms of the GNU Lesser General Public License
  11  * as published by the Free Software Foundation; either version 2.1
  12  * of the License, or (at your option) any later version.
  13  *
  14  * GROMACS is distributed in the hope that it will be useful,
  15  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  16  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  17  * Lesser General Public License for more details.
  18  *
  19  * You should have received a copy of the GNU Lesser General Public
  20  * License along with GROMACS; if not, see
  21  * http://www.gnu.org/licenses, or write to the Free Software Foundation,
  22  * Inc., 51 Franklin Street, Fifth Floor, Boston, MA  02110-1301  USA.
  23  *
  24  * If you want to redistribute modifications to GROMACS, please
  25  * consider that scientific software is very special. Version
  26  * control is crucial - bugs must be traceable. We will be happy to
  27  * consider code for inclusion in the official distribution, but
  28  * derived work must not be called official GROMACS. Details are found
  29  * in the README & COPYING files - if they are missing, get the
  30  * official version at http://www.gromacs.org.
  31  *
  32  * To help us fund GROMACS development, we humbly ask that you cite
  33  * the research papers on the package. Check out http://www.gromacs.org.
  34  */
  35 /*
  36  * Note: this file was generated by the GROMACS sse4_1_single kernel generator.
  37  */
  38 #include "gmxpre.h"
  39
  40 #include "config.h"
  41
  42 #include <math.h>
  43
  44 #include "../nb_kernel.h"
  45 #include "gromacs/gmxlib/nrnb.h"
  46
  47 #include "kernelutil_x86_sse4_1_single.h"
  48
  49 /*
  50  * Gromacs nonbonded kernel:   nb_kernel_ElecEw_VdwCSTab_GeomW3P1_VF_sse4_1_single
  51  * Electrostatics interaction: Ewald
  52  * VdW interaction:            CubicSplineTable
  53  * Geometry:                   Water3-Particle
  54  * Calculate force/pot:        PotentialAndForce
  55  */
  56 void
  57 nb_kernel_ElecEw_VdwCSTab_GeomW3P1_VF_sse4_1_single
  58                     (t_nblist                    * gmx_restrict       nlist,
  59                      rvec                        * gmx_restrict          xx,
  60                      rvec                        * gmx_restrict          ff,
  61                      struct t_forcerec           * gmx_restrict          fr,
  62                      t_mdatoms                   * gmx_restrict     mdatoms,
  63                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
  64                      t_nrnb                      * gmx_restrict        nrnb)
  65 {
  66     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
  67      * just 0 for non-waters.
  68      * Suffixes A,B,C,D refer to j loop unrolling done with SSE, e.g. for the four different
  69      * jnr indices corresponding to data put in the four positions in the SIMD register.
  70      */
  71     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
  72     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
  73     int              jnrA,jnrB,jnrC,jnrD;
  74     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
  75     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
  76     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
  77     real             rcutoff_scalar;
  78     real             *shiftvec,*fshift,*x,*f;
  79     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
  80     real             scratch[4*DIM];
  81     __m128           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
  82     int              vdwioffset0;
  83     __m128           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
  84     int              vdwioffset1;
  85     __m128           ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
  86     int              vdwioffset2;
  87     __m128           ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
  88     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
  89     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  90     __m128           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  91     __m128           dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
  92     __m128           dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
  93     __m128           velec,felec,velecsum,facel,crf,krf,krf2;
  94     real             *charge;
  95     int              nvdwtype;
  96     __m128           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
  97     int              *vdwtype;
  98     real             *vdwparam;
  99     __m128           one_sixth   = _mm_set1_ps(1.0/6.0);
 100     __m128           one_twelfth = _mm_set1_ps(1.0/12.0);
 101     __m128i          vfitab;
 102     __m128i          ifour       = _mm_set1_epi32(4);
 103     __m128           rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF;
 104     real             *vftab;
 105     __m128i          ewitab;
 106     __m128           ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
 107     real             *ewtab;
 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->ic->epsfac);
 124     charge           = mdatoms->chargeA;
 125     nvdwtype         = fr->ntype;
 126     vdwparam         = fr->nbfp;
 127     vdwtype          = mdatoms->typeA;
 128
 129     vftab            = kernel_data->table_vdw->data;
 130     vftabscale       = _mm_set1_ps(kernel_data->table_vdw->scale);
 131
 132     sh_ewald         = _mm_set1_ps(fr->ic->sh_ewald);
 133     ewtab            = fr->ic->tabq_coul_FDV0;
 134     ewtabscale       = _mm_set1_ps(fr->ic->tabq_scale);
 135     ewtabhalfspace   = _mm_set1_ps(0.5/fr->ic->tabq_scale);
 136
 137     /* Setup water-specific parameters */
 138     inr              = nlist->iinr[0];
 139     iq0              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+0]));
 140     iq1              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+1]));
 141     iq2              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+2]));
 142     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 143
 144     /* Avoid stupid compiler warnings */
 145     jnrA = jnrB = jnrC = jnrD = 0;
 146     j_coord_offsetA = 0;
 147     j_coord_offsetB = 0;
 148     j_coord_offsetC = 0;
 149     j_coord_offsetD = 0;
 150
 151     outeriter        = 0;
 152     inneriter        = 0;
 153
 154     for(iidx=0;iidx<4*DIM;iidx++)
 155     {
 156         scratch[iidx] = 0.0;
 157     }
 158
 159     /* Start outer loop over neighborlists */
 160     for(iidx=0; iidx<nri; iidx++)
 161     {
 162         /* Load shift vector for this list */
 163         i_shift_offset   = DIM*shiftidx[iidx];
 164
 165         /* Load limits for loop over neighbors */
 166         j_index_start    = jindex[iidx];
 167         j_index_end      = jindex[iidx+1];
 168
 169         /* Get outer coordinate index */
 170         inr              = iinr[iidx];
 171         i_coord_offset   = DIM*inr;
 172
 173         /* Load i particle coords and add shift vector */
 174         gmx_mm_load_shift_and_3rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
 175                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 176
 177         fix0             = _mm_setzero_ps();
 178         fiy0             = _mm_setzero_ps();
 179         fiz0             = _mm_setzero_ps();
 180         fix1             = _mm_setzero_ps();
 181         fiy1             = _mm_setzero_ps();
 182         fiz1             = _mm_setzero_ps();
 183         fix2             = _mm_setzero_ps();
 184         fiy2             = _mm_setzero_ps();
 185         fiz2             = _mm_setzero_ps();
 186
 187         /* Reset potential sums */
 188         velecsum         = _mm_setzero_ps();
 189         vvdwsum          = _mm_setzero_ps();
 190
 191         /* Start inner kernel loop */
 192         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 193         {
 194
 195             /* Get j neighbor index, and coordinate index */
 196             jnrA             = jjnr[jidx];
 197             jnrB             = jjnr[jidx+1];
 198             jnrC             = jjnr[jidx+2];
 199             jnrD             = jjnr[jidx+3];
 200             j_coord_offsetA  = DIM*jnrA;
 201             j_coord_offsetB  = DIM*jnrB;
 202             j_coord_offsetC  = DIM*jnrC;
 203             j_coord_offsetD  = DIM*jnrD;
 204
 205             /* load j atom coordinates */
 206             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 207                                               x+j_coord_offsetC,x+j_coord_offsetD,
 208                                               &jx0,&jy0,&jz0);
 209
 210             /* Calculate displacement vector */
 211             dx00             = _mm_sub_ps(ix0,jx0);
 212             dy00             = _mm_sub_ps(iy0,jy0);
 213             dz00             = _mm_sub_ps(iz0,jz0);
 214             dx10             = _mm_sub_ps(ix1,jx0);
 215             dy10             = _mm_sub_ps(iy1,jy0);
 216             dz10             = _mm_sub_ps(iz1,jz0);
 217             dx20             = _mm_sub_ps(ix2,jx0);
 218             dy20             = _mm_sub_ps(iy2,jy0);
 219             dz20             = _mm_sub_ps(iz2,jz0);
 220
 221             /* Calculate squared distance and things based on it */
 222             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 223             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 224             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 225
 226             rinv00           = sse41_invsqrt_f(rsq00);
 227             rinv10           = sse41_invsqrt_f(rsq10);
 228             rinv20           = sse41_invsqrt_f(rsq20);
 229
 230             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 231             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 232             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 233
 234             /* Load parameters for j particles */
 235             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 236                                                               charge+jnrC+0,charge+jnrD+0);
 237             vdwjidx0A        = 2*vdwtype[jnrA+0];
 238             vdwjidx0B        = 2*vdwtype[jnrB+0];
 239             vdwjidx0C        = 2*vdwtype[jnrC+0];
 240             vdwjidx0D        = 2*vdwtype[jnrD+0];
 241
 242             fjx0             = _mm_setzero_ps();
 243             fjy0             = _mm_setzero_ps();
 244             fjz0             = _mm_setzero_ps();
 245
 246             /**************************
 247              * CALCULATE INTERACTIONS *
 248              **************************/
 249
 250             r00              = _mm_mul_ps(rsq00,rinv00);
 251
 252             /* Compute parameters for interactions between i and j atoms */
 253             qq00             = _mm_mul_ps(iq0,jq0);
 254             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 255                                          vdwparam+vdwioffset0+vdwjidx0B,
 256                                          vdwparam+vdwioffset0+vdwjidx0C,
 257                                          vdwparam+vdwioffset0+vdwjidx0D,
 258                                          &c6_00,&c12_00);
 259
 260             /* Calculate table index by multiplying r with table scale and truncate to integer */
 261             rt               = _mm_mul_ps(r00,vftabscale);
 262             vfitab           = _mm_cvttps_epi32(rt);
 263             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
 264             vfitab           = _mm_slli_epi32(vfitab,3);
 265
 266             /* EWALD ELECTROSTATICS */
 267
 268             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 269             ewrt             = _mm_mul_ps(r00,ewtabscale);
 270             ewitab           = _mm_cvttps_epi32(ewrt);
 271             eweps            = _mm_sub_ps(ewrt,_mm_round_ps(ewrt, _MM_FROUND_FLOOR));
 272             ewitab           = _mm_slli_epi32(ewitab,2);
 273             ewtabF           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 274             ewtabD           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
 275             ewtabV           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
 276             ewtabFn          = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
 277             _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
 278             felec            = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
 279             velec            = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
 280             velec            = _mm_mul_ps(qq00,_mm_sub_ps(rinv00,velec));
 281             felec            = _mm_mul_ps(_mm_mul_ps(qq00,rinv00),_mm_sub_ps(rinvsq00,felec));
 282
 283             /* CUBIC SPLINE TABLE DISPERSION */
 284             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 285             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 286             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 287             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 288             _MM_TRANSPOSE4_PS(Y,F,G,H);
 289             Heps             = _mm_mul_ps(vfeps,H);
 290             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 291             VV               = _mm_add_ps(Y,_mm_mul_ps(vfeps,Fp));
 292             vvdw6            = _mm_mul_ps(c6_00,VV);
 293             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 294             fvdw6            = _mm_mul_ps(c6_00,FF);
 295
 296             /* CUBIC SPLINE TABLE REPULSION */
 297             vfitab           = _mm_add_epi32(vfitab,ifour);
 298             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 299             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 300             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 301             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 302             _MM_TRANSPOSE4_PS(Y,F,G,H);
 303             Heps             = _mm_mul_ps(vfeps,H);
 304             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 305             VV               = _mm_add_ps(Y,_mm_mul_ps(vfeps,Fp));
 306             vvdw12           = _mm_mul_ps(c12_00,VV);
 307             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 308             fvdw12           = _mm_mul_ps(c12_00,FF);
 309             vvdw             = _mm_add_ps(vvdw12,vvdw6);
 310             fvdw             = _mm_xor_ps(signbit,_mm_mul_ps(_mm_add_ps(fvdw6,fvdw12),_mm_mul_ps(vftabscale,rinv00)));
 311
 312             /* Update potential sum for this i atom from the interaction with this j atom. */
 313             velecsum         = _mm_add_ps(velecsum,velec);
 314             vvdwsum          = _mm_add_ps(vvdwsum,vvdw);
 315
 316             fscal            = _mm_add_ps(felec,fvdw);
 317
 318             /* Calculate temporary vectorial force */
 319             tx               = _mm_mul_ps(fscal,dx00);
 320             ty               = _mm_mul_ps(fscal,dy00);
 321             tz               = _mm_mul_ps(fscal,dz00);
 322
 323             /* Update vectorial force */
 324             fix0             = _mm_add_ps(fix0,tx);
 325             fiy0             = _mm_add_ps(fiy0,ty);
 326             fiz0             = _mm_add_ps(fiz0,tz);
 327
 328             fjx0             = _mm_add_ps(fjx0,tx);
 329             fjy0             = _mm_add_ps(fjy0,ty);
 330             fjz0             = _mm_add_ps(fjz0,tz);
 331
 332             /**************************
 333              * CALCULATE INTERACTIONS *
 334              **************************/
 335
 336             r10              = _mm_mul_ps(rsq10,rinv10);
 337
 338             /* Compute parameters for interactions between i and j atoms */
 339             qq10             = _mm_mul_ps(iq1,jq0);
 340
 341             /* EWALD ELECTROSTATICS */
 342
 343             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 344             ewrt             = _mm_mul_ps(r10,ewtabscale);
 345             ewitab           = _mm_cvttps_epi32(ewrt);
 346             eweps            = _mm_sub_ps(ewrt,_mm_round_ps(ewrt, _MM_FROUND_FLOOR));
 347             ewitab           = _mm_slli_epi32(ewitab,2);
 348             ewtabF           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 349             ewtabD           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
 350             ewtabV           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
 351             ewtabFn          = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
 352             _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
 353             felec            = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
 354             velec            = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
 355             velec            = _mm_mul_ps(qq10,_mm_sub_ps(rinv10,velec));
 356             felec            = _mm_mul_ps(_mm_mul_ps(qq10,rinv10),_mm_sub_ps(rinvsq10,felec));
 357
 358             /* Update potential sum for this i atom from the interaction with this j atom. */
 359             velecsum         = _mm_add_ps(velecsum,velec);
 360
 361             fscal            = felec;
 362
 363             /* Calculate temporary vectorial force */
 364             tx               = _mm_mul_ps(fscal,dx10);
 365             ty               = _mm_mul_ps(fscal,dy10);
 366             tz               = _mm_mul_ps(fscal,dz10);
 367
 368             /* Update vectorial force */
 369             fix1             = _mm_add_ps(fix1,tx);
 370             fiy1             = _mm_add_ps(fiy1,ty);
 371             fiz1             = _mm_add_ps(fiz1,tz);
 372
 373             fjx0             = _mm_add_ps(fjx0,tx);
 374             fjy0             = _mm_add_ps(fjy0,ty);
 375             fjz0             = _mm_add_ps(fjz0,tz);
 376
 377             /**************************
 378              * CALCULATE INTERACTIONS *
 379              **************************/
 380
 381             r20              = _mm_mul_ps(rsq20,rinv20);
 382
 383             /* Compute parameters for interactions between i and j atoms */
 384             qq20             = _mm_mul_ps(iq2,jq0);
 385
 386             /* EWALD ELECTROSTATICS */
 387
 388             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 389             ewrt             = _mm_mul_ps(r20,ewtabscale);
 390             ewitab           = _mm_cvttps_epi32(ewrt);
 391             eweps            = _mm_sub_ps(ewrt,_mm_round_ps(ewrt, _MM_FROUND_FLOOR));
 392             ewitab           = _mm_slli_epi32(ewitab,2);
 393             ewtabF           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 394             ewtabD           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
 395             ewtabV           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
 396             ewtabFn          = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
 397             _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
 398             felec            = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
 399             velec            = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
 400             velec            = _mm_mul_ps(qq20,_mm_sub_ps(rinv20,velec));
 401             felec            = _mm_mul_ps(_mm_mul_ps(qq20,rinv20),_mm_sub_ps(rinvsq20,felec));
 402
 403             /* Update potential sum for this i atom from the interaction with this j atom. */
 404             velecsum         = _mm_add_ps(velecsum,velec);
 405
 406             fscal            = felec;
 407
 408             /* Calculate temporary vectorial force */
 409             tx               = _mm_mul_ps(fscal,dx20);
 410             ty               = _mm_mul_ps(fscal,dy20);
 411             tz               = _mm_mul_ps(fscal,dz20);
 412
 413             /* Update vectorial force */
 414             fix2             = _mm_add_ps(fix2,tx);
 415             fiy2             = _mm_add_ps(fiy2,ty);
 416             fiz2             = _mm_add_ps(fiz2,tz);
 417
 418             fjx0             = _mm_add_ps(fjx0,tx);
 419             fjy0             = _mm_add_ps(fjy0,ty);
 420             fjz0             = _mm_add_ps(fjz0,tz);
 421
 422             fjptrA             = f+j_coord_offsetA;
 423             fjptrB             = f+j_coord_offsetB;
 424             fjptrC             = f+j_coord_offsetC;
 425             fjptrD             = f+j_coord_offsetD;
 426
 427             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 428
 429             /* Inner loop uses 157 flops */
 430         }
 431
 432         if(jidx<j_index_end)
 433         {
 434
 435             /* Get j neighbor index, and coordinate index */
 436             jnrlistA         = jjnr[jidx];
 437             jnrlistB         = jjnr[jidx+1];
 438             jnrlistC         = jjnr[jidx+2];
 439             jnrlistD         = jjnr[jidx+3];
 440             /* Sign of each element will be negative for non-real atoms.
 441              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 442              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 443              */
 444             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 445             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 446             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 447             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 448             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 449             j_coord_offsetA  = DIM*jnrA;
 450             j_coord_offsetB  = DIM*jnrB;
 451             j_coord_offsetC  = DIM*jnrC;
 452             j_coord_offsetD  = DIM*jnrD;
 453
 454             /* load j atom coordinates */
 455             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 456                                               x+j_coord_offsetC,x+j_coord_offsetD,
 457                                               &jx0,&jy0,&jz0);
 458
 459             /* Calculate displacement vector */
 460             dx00             = _mm_sub_ps(ix0,jx0);
 461             dy00             = _mm_sub_ps(iy0,jy0);
 462             dz00             = _mm_sub_ps(iz0,jz0);
 463             dx10             = _mm_sub_ps(ix1,jx0);
 464             dy10             = _mm_sub_ps(iy1,jy0);
 465             dz10             = _mm_sub_ps(iz1,jz0);
 466             dx20             = _mm_sub_ps(ix2,jx0);
 467             dy20             = _mm_sub_ps(iy2,jy0);
 468             dz20             = _mm_sub_ps(iz2,jz0);
 469
 470             /* Calculate squared distance and things based on it */
 471             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 472             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 473             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 474
 475             rinv00           = sse41_invsqrt_f(rsq00);
 476             rinv10           = sse41_invsqrt_f(rsq10);
 477             rinv20           = sse41_invsqrt_f(rsq20);
 478
 479             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 480             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 481             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 482
 483             /* Load parameters for j particles */
 484             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 485                                                               charge+jnrC+0,charge+jnrD+0);
 486             vdwjidx0A        = 2*vdwtype[jnrA+0];
 487             vdwjidx0B        = 2*vdwtype[jnrB+0];
 488             vdwjidx0C        = 2*vdwtype[jnrC+0];
 489             vdwjidx0D        = 2*vdwtype[jnrD+0];
 490
 491             fjx0             = _mm_setzero_ps();
 492             fjy0             = _mm_setzero_ps();
 493             fjz0             = _mm_setzero_ps();
 494
 495             /**************************
 496              * CALCULATE INTERACTIONS *
 497              **************************/
 498
 499             r00              = _mm_mul_ps(rsq00,rinv00);
 500             r00              = _mm_andnot_ps(dummy_mask,r00);
 501
 502             /* Compute parameters for interactions between i and j atoms */
 503             qq00             = _mm_mul_ps(iq0,jq0);
 504             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 505                                          vdwparam+vdwioffset0+vdwjidx0B,
 506                                          vdwparam+vdwioffset0+vdwjidx0C,
 507                                          vdwparam+vdwioffset0+vdwjidx0D,
 508                                          &c6_00,&c12_00);
 509
 510             /* Calculate table index by multiplying r with table scale and truncate to integer */
 511             rt               = _mm_mul_ps(r00,vftabscale);
 512             vfitab           = _mm_cvttps_epi32(rt);
 513             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
 514             vfitab           = _mm_slli_epi32(vfitab,3);
 515
 516             /* EWALD ELECTROSTATICS */
 517
 518             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 519             ewrt             = _mm_mul_ps(r00,ewtabscale);
 520             ewitab           = _mm_cvttps_epi32(ewrt);
 521             eweps            = _mm_sub_ps(ewrt,_mm_round_ps(ewrt, _MM_FROUND_FLOOR));
 522             ewitab           = _mm_slli_epi32(ewitab,2);
 523             ewtabF           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 524             ewtabD           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
 525             ewtabV           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
 526             ewtabFn          = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
 527             _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
 528             felec            = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
 529             velec            = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
 530             velec            = _mm_mul_ps(qq00,_mm_sub_ps(rinv00,velec));
 531             felec            = _mm_mul_ps(_mm_mul_ps(qq00,rinv00),_mm_sub_ps(rinvsq00,felec));
 532
 533             /* CUBIC SPLINE TABLE DISPERSION */
 534             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 535             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 536             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 537             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 538             _MM_TRANSPOSE4_PS(Y,F,G,H);
 539             Heps             = _mm_mul_ps(vfeps,H);
 540             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 541             VV               = _mm_add_ps(Y,_mm_mul_ps(vfeps,Fp));
 542             vvdw6            = _mm_mul_ps(c6_00,VV);
 543             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 544             fvdw6            = _mm_mul_ps(c6_00,FF);
 545
 546             /* CUBIC SPLINE TABLE REPULSION */
 547             vfitab           = _mm_add_epi32(vfitab,ifour);
 548             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 549             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 550             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 551             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 552             _MM_TRANSPOSE4_PS(Y,F,G,H);
 553             Heps             = _mm_mul_ps(vfeps,H);
 554             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 555             VV               = _mm_add_ps(Y,_mm_mul_ps(vfeps,Fp));
 556             vvdw12           = _mm_mul_ps(c12_00,VV);
 557             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 558             fvdw12           = _mm_mul_ps(c12_00,FF);
 559             vvdw             = _mm_add_ps(vvdw12,vvdw6);
 560             fvdw             = _mm_xor_ps(signbit,_mm_mul_ps(_mm_add_ps(fvdw6,fvdw12),_mm_mul_ps(vftabscale,rinv00)));
 561
 562             /* Update potential sum for this i atom from the interaction with this j atom. */
 563             velec            = _mm_andnot_ps(dummy_mask,velec);
 564             velecsum         = _mm_add_ps(velecsum,velec);
 565             vvdw             = _mm_andnot_ps(dummy_mask,vvdw);
 566             vvdwsum          = _mm_add_ps(vvdwsum,vvdw);
 567
 568             fscal            = _mm_add_ps(felec,fvdw);
 569
 570             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 571
 572             /* Calculate temporary vectorial force */
 573             tx               = _mm_mul_ps(fscal,dx00);
 574             ty               = _mm_mul_ps(fscal,dy00);
 575             tz               = _mm_mul_ps(fscal,dz00);
 576
 577             /* Update vectorial force */
 578             fix0             = _mm_add_ps(fix0,tx);
 579             fiy0             = _mm_add_ps(fiy0,ty);
 580             fiz0             = _mm_add_ps(fiz0,tz);
 581
 582             fjx0             = _mm_add_ps(fjx0,tx);
 583             fjy0             = _mm_add_ps(fjy0,ty);
 584             fjz0             = _mm_add_ps(fjz0,tz);
 585
 586             /**************************
 587              * CALCULATE INTERACTIONS *
 588              **************************/
 589
 590             r10              = _mm_mul_ps(rsq10,rinv10);
 591             r10              = _mm_andnot_ps(dummy_mask,r10);
 592
 593             /* Compute parameters for interactions between i and j atoms */
 594             qq10             = _mm_mul_ps(iq1,jq0);
 595
 596             /* EWALD ELECTROSTATICS */
 597
 598             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 599             ewrt             = _mm_mul_ps(r10,ewtabscale);
 600             ewitab           = _mm_cvttps_epi32(ewrt);
 601             eweps            = _mm_sub_ps(ewrt,_mm_round_ps(ewrt, _MM_FROUND_FLOOR));
 602             ewitab           = _mm_slli_epi32(ewitab,2);
 603             ewtabF           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 604             ewtabD           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
 605             ewtabV           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
 606             ewtabFn          = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
 607             _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
 608             felec            = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
 609             velec            = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
 610             velec            = _mm_mul_ps(qq10,_mm_sub_ps(rinv10,velec));
 611             felec            = _mm_mul_ps(_mm_mul_ps(qq10,rinv10),_mm_sub_ps(rinvsq10,felec));
 612
 613             /* Update potential sum for this i atom from the interaction with this j atom. */
 614             velec            = _mm_andnot_ps(dummy_mask,velec);
 615             velecsum         = _mm_add_ps(velecsum,velec);
 616
 617             fscal            = felec;
 618
 619             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 620
 621             /* Calculate temporary vectorial force */
 622             tx               = _mm_mul_ps(fscal,dx10);
 623             ty               = _mm_mul_ps(fscal,dy10);
 624             tz               = _mm_mul_ps(fscal,dz10);
 625
 626             /* Update vectorial force */
 627             fix1             = _mm_add_ps(fix1,tx);
 628             fiy1             = _mm_add_ps(fiy1,ty);
 629             fiz1             = _mm_add_ps(fiz1,tz);
 630
 631             fjx0             = _mm_add_ps(fjx0,tx);
 632             fjy0             = _mm_add_ps(fjy0,ty);
 633             fjz0             = _mm_add_ps(fjz0,tz);
 634
 635             /**************************
 636              * CALCULATE INTERACTIONS *
 637              **************************/
 638
 639             r20              = _mm_mul_ps(rsq20,rinv20);
 640             r20              = _mm_andnot_ps(dummy_mask,r20);
 641
 642             /* Compute parameters for interactions between i and j atoms */
 643             qq20             = _mm_mul_ps(iq2,jq0);
 644
 645             /* EWALD ELECTROSTATICS */
 646
 647             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 648             ewrt             = _mm_mul_ps(r20,ewtabscale);
 649             ewitab           = _mm_cvttps_epi32(ewrt);
 650             eweps            = _mm_sub_ps(ewrt,_mm_round_ps(ewrt, _MM_FROUND_FLOOR));
 651             ewitab           = _mm_slli_epi32(ewitab,2);
 652             ewtabF           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 653             ewtabD           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
 654             ewtabV           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
 655             ewtabFn          = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
 656             _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
 657             felec            = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
 658             velec            = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
 659             velec            = _mm_mul_ps(qq20,_mm_sub_ps(rinv20,velec));
 660             felec            = _mm_mul_ps(_mm_mul_ps(qq20,rinv20),_mm_sub_ps(rinvsq20,felec));
 661
 662             /* Update potential sum for this i atom from the interaction with this j atom. */
 663             velec            = _mm_andnot_ps(dummy_mask,velec);
 664             velecsum         = _mm_add_ps(velecsum,velec);
 665
 666             fscal            = felec;
 667
 668             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 669
 670             /* Calculate temporary vectorial force */
 671             tx               = _mm_mul_ps(fscal,dx20);
 672             ty               = _mm_mul_ps(fscal,dy20);
 673             tz               = _mm_mul_ps(fscal,dz20);
 674
 675             /* Update vectorial force */
 676             fix2             = _mm_add_ps(fix2,tx);
 677             fiy2             = _mm_add_ps(fiy2,ty);
 678             fiz2             = _mm_add_ps(fiz2,tz);
 679
 680             fjx0             = _mm_add_ps(fjx0,tx);
 681             fjy0             = _mm_add_ps(fjy0,ty);
 682             fjz0             = _mm_add_ps(fjz0,tz);
 683
 684             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 685             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 686             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 687             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 688
 689             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 690
 691             /* Inner loop uses 160 flops */
 692         }
 693
 694         /* End of innermost loop */
 695
 696         gmx_mm_update_iforce_3atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
 697                                               f+i_coord_offset,fshift+i_shift_offset);
 698
 699         ggid                        = gid[iidx];
 700         /* Update potential energies */
 701         gmx_mm_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid);
 702         gmx_mm_update_1pot_ps(vvdwsum,kernel_data->energygrp_vdw+ggid);
 703
 704         /* Increment number of inner iterations */
 705         inneriter                  += j_index_end - j_index_start;
 706
 707         /* Outer loop uses 20 flops */
 708     }
 709
 710     /* Increment number of outer iterations */
 711     outeriter        += nri;
 712
 713     /* Update outer/inner flops */
 714
 715     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3_VF,outeriter*20 + inneriter*160);
 716 }
 717 /*
 718  * Gromacs nonbonded kernel:   nb_kernel_ElecEw_VdwCSTab_GeomW3P1_F_sse4_1_single
 719  * Electrostatics interaction: Ewald
 720  * VdW interaction:            CubicSplineTable
 721  * Geometry:                   Water3-Particle
 722  * Calculate force/pot:        Force
 723  */
 724 void
 725 nb_kernel_ElecEw_VdwCSTab_GeomW3P1_F_sse4_1_single
 726                     (t_nblist                    * gmx_restrict       nlist,
 727                      rvec                        * gmx_restrict          xx,
 728                      rvec                        * gmx_restrict          ff,
 729                      struct t_forcerec           * gmx_restrict          fr,
 730                      t_mdatoms                   * gmx_restrict     mdatoms,
 731                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 732                      t_nrnb                      * gmx_restrict        nrnb)
 733 {
 734     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 735      * just 0 for non-waters.
 736      * Suffixes A,B,C,D refer to j loop unrolling done with SSE, e.g. for the four different
 737      * jnr indices corresponding to data put in the four positions in the SIMD register.
 738      */
 739     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 740     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 741     int              jnrA,jnrB,jnrC,jnrD;
 742     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
 743     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 744     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 745     real             rcutoff_scalar;
 746     real             *shiftvec,*fshift,*x,*f;
 747     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
 748     real             scratch[4*DIM];
 749     __m128           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 750     int              vdwioffset0;
 751     __m128           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 752     int              vdwioffset1;
 753     __m128           ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
 754     int              vdwioffset2;
 755     __m128           ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
 756     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
 757     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 758     __m128           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 759     __m128           dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
 760     __m128           dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
 761     __m128           velec,felec,velecsum,facel,crf,krf,krf2;
 762     real             *charge;
 763     int              nvdwtype;
 764     __m128           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 765     int              *vdwtype;
 766     real             *vdwparam;
 767     __m128           one_sixth   = _mm_set1_ps(1.0/6.0);
 768     __m128           one_twelfth = _mm_set1_ps(1.0/12.0);
 769     __m128i          vfitab;
 770     __m128i          ifour       = _mm_set1_epi32(4);
 771     __m128           rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF;
 772     real             *vftab;
 773     __m128i          ewitab;
 774     __m128           ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
 775     real             *ewtab;
 776     __m128           dummy_mask,cutoff_mask;
 777     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
 778     __m128           one     = _mm_set1_ps(1.0);
 779     __m128           two     = _mm_set1_ps(2.0);
 780     x                = xx[0];
 781     f                = ff[0];
 782
 783     nri              = nlist->nri;
 784     iinr             = nlist->iinr;
 785     jindex           = nlist->jindex;
 786     jjnr             = nlist->jjnr;
 787     shiftidx         = nlist->shift;
 788     gid              = nlist->gid;
 789     shiftvec         = fr->shift_vec[0];
 790     fshift           = fr->fshift[0];
 791     facel            = _mm_set1_ps(fr->ic->epsfac);
 792     charge           = mdatoms->chargeA;
 793     nvdwtype         = fr->ntype;
 794     vdwparam         = fr->nbfp;
 795     vdwtype          = mdatoms->typeA;
 796
 797     vftab            = kernel_data->table_vdw->data;
 798     vftabscale       = _mm_set1_ps(kernel_data->table_vdw->scale);
 799
 800     sh_ewald         = _mm_set1_ps(fr->ic->sh_ewald);
 801     ewtab            = fr->ic->tabq_coul_F;
 802     ewtabscale       = _mm_set1_ps(fr->ic->tabq_scale);
 803     ewtabhalfspace   = _mm_set1_ps(0.5/fr->ic->tabq_scale);
 804
 805     /* Setup water-specific parameters */
 806     inr              = nlist->iinr[0];
 807     iq0              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+0]));
 808     iq1              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+1]));
 809     iq2              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+2]));
 810     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 811
 812     /* Avoid stupid compiler warnings */
 813     jnrA = jnrB = jnrC = jnrD = 0;
 814     j_coord_offsetA = 0;
 815     j_coord_offsetB = 0;
 816     j_coord_offsetC = 0;
 817     j_coord_offsetD = 0;
 818
 819     outeriter        = 0;
 820     inneriter        = 0;
 821
 822     for(iidx=0;iidx<4*DIM;iidx++)
 823     {
 824         scratch[iidx] = 0.0;
 825     }
 826
 827     /* Start outer loop over neighborlists */
 828     for(iidx=0; iidx<nri; iidx++)
 829     {
 830         /* Load shift vector for this list */
 831         i_shift_offset   = DIM*shiftidx[iidx];
 832
 833         /* Load limits for loop over neighbors */
 834         j_index_start    = jindex[iidx];
 835         j_index_end      = jindex[iidx+1];
 836
 837         /* Get outer coordinate index */
 838         inr              = iinr[iidx];
 839         i_coord_offset   = DIM*inr;
 840
 841         /* Load i particle coords and add shift vector */
 842         gmx_mm_load_shift_and_3rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
 843                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 844
 845         fix0             = _mm_setzero_ps();
 846         fiy0             = _mm_setzero_ps();
 847         fiz0             = _mm_setzero_ps();
 848         fix1             = _mm_setzero_ps();
 849         fiy1             = _mm_setzero_ps();
 850         fiz1             = _mm_setzero_ps();
 851         fix2             = _mm_setzero_ps();
 852         fiy2             = _mm_setzero_ps();
 853         fiz2             = _mm_setzero_ps();
 854
 855         /* Start inner kernel loop */
 856         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 857         {
 858
 859             /* Get j neighbor index, and coordinate index */
 860             jnrA             = jjnr[jidx];
 861             jnrB             = jjnr[jidx+1];
 862             jnrC             = jjnr[jidx+2];
 863             jnrD             = jjnr[jidx+3];
 864             j_coord_offsetA  = DIM*jnrA;
 865             j_coord_offsetB  = DIM*jnrB;
 866             j_coord_offsetC  = DIM*jnrC;
 867             j_coord_offsetD  = DIM*jnrD;
 868
 869             /* load j atom coordinates */
 870             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 871                                               x+j_coord_offsetC,x+j_coord_offsetD,
 872                                               &jx0,&jy0,&jz0);
 873
 874             /* Calculate displacement vector */
 875             dx00             = _mm_sub_ps(ix0,jx0);
 876             dy00             = _mm_sub_ps(iy0,jy0);
 877             dz00             = _mm_sub_ps(iz0,jz0);
 878             dx10             = _mm_sub_ps(ix1,jx0);
 879             dy10             = _mm_sub_ps(iy1,jy0);
 880             dz10             = _mm_sub_ps(iz1,jz0);
 881             dx20             = _mm_sub_ps(ix2,jx0);
 882             dy20             = _mm_sub_ps(iy2,jy0);
 883             dz20             = _mm_sub_ps(iz2,jz0);
 884
 885             /* Calculate squared distance and things based on it */
 886             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 887             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 888             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 889
 890             rinv00           = sse41_invsqrt_f(rsq00);
 891             rinv10           = sse41_invsqrt_f(rsq10);
 892             rinv20           = sse41_invsqrt_f(rsq20);
 893
 894             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 895             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 896             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 897
 898             /* Load parameters for j particles */
 899             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 900                                                               charge+jnrC+0,charge+jnrD+0);
 901             vdwjidx0A        = 2*vdwtype[jnrA+0];
 902             vdwjidx0B        = 2*vdwtype[jnrB+0];
 903             vdwjidx0C        = 2*vdwtype[jnrC+0];
 904             vdwjidx0D        = 2*vdwtype[jnrD+0];
 905
 906             fjx0             = _mm_setzero_ps();
 907             fjy0             = _mm_setzero_ps();
 908             fjz0             = _mm_setzero_ps();
 909
 910             /**************************
 911              * CALCULATE INTERACTIONS *
 912              **************************/
 913
 914             r00              = _mm_mul_ps(rsq00,rinv00);
 915
 916             /* Compute parameters for interactions between i and j atoms */
 917             qq00             = _mm_mul_ps(iq0,jq0);
 918             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 919                                          vdwparam+vdwioffset0+vdwjidx0B,
 920                                          vdwparam+vdwioffset0+vdwjidx0C,
 921                                          vdwparam+vdwioffset0+vdwjidx0D,
 922                                          &c6_00,&c12_00);
 923
 924             /* Calculate table index by multiplying r with table scale and truncate to integer */
 925             rt               = _mm_mul_ps(r00,vftabscale);
 926             vfitab           = _mm_cvttps_epi32(rt);
 927             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
 928             vfitab           = _mm_slli_epi32(vfitab,3);
 929
 930             /* EWALD ELECTROSTATICS */
 931
 932             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 933             ewrt             = _mm_mul_ps(r00,ewtabscale);
 934             ewitab           = _mm_cvttps_epi32(ewrt);
 935             eweps            = _mm_sub_ps(ewrt,_mm_round_ps(ewrt, _MM_FROUND_FLOOR));
 936             gmx_mm_load_4pair_swizzle_ps(ewtab + gmx_mm_extract_epi32(ewitab,0),ewtab + gmx_mm_extract_epi32(ewitab,1),
 937                                          ewtab + gmx_mm_extract_epi32(ewitab,2),ewtab + gmx_mm_extract_epi32(ewitab,3),
 938                                          &ewtabF,&ewtabFn);
 939             felec            = _mm_add_ps(_mm_mul_ps( _mm_sub_ps(one,eweps),ewtabF),_mm_mul_ps(eweps,ewtabFn));
 940             felec            = _mm_mul_ps(_mm_mul_ps(qq00,rinv00),_mm_sub_ps(rinvsq00,felec));
 941
 942             /* CUBIC SPLINE TABLE DISPERSION */
 943             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 944             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 945             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 946             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 947             _MM_TRANSPOSE4_PS(Y,F,G,H);
 948             Heps             = _mm_mul_ps(vfeps,H);
 949             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 950             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 951             fvdw6            = _mm_mul_ps(c6_00,FF);
 952
 953             /* CUBIC SPLINE TABLE REPULSION */
 954             vfitab           = _mm_add_epi32(vfitab,ifour);
 955             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 956             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 957             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 958             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 959             _MM_TRANSPOSE4_PS(Y,F,G,H);
 960             Heps             = _mm_mul_ps(vfeps,H);
 961             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 962             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 963             fvdw12           = _mm_mul_ps(c12_00,FF);
 964             fvdw             = _mm_xor_ps(signbit,_mm_mul_ps(_mm_add_ps(fvdw6,fvdw12),_mm_mul_ps(vftabscale,rinv00)));
 965
 966             fscal            = _mm_add_ps(felec,fvdw);
 967
 968             /* Calculate temporary vectorial force */
 969             tx               = _mm_mul_ps(fscal,dx00);
 970             ty               = _mm_mul_ps(fscal,dy00);
 971             tz               = _mm_mul_ps(fscal,dz00);
 972
 973             /* Update vectorial force */
 974             fix0             = _mm_add_ps(fix0,tx);
 975             fiy0             = _mm_add_ps(fiy0,ty);
 976             fiz0             = _mm_add_ps(fiz0,tz);
 977
 978             fjx0             = _mm_add_ps(fjx0,tx);
 979             fjy0             = _mm_add_ps(fjy0,ty);
 980             fjz0             = _mm_add_ps(fjz0,tz);
 981
 982             /**************************
 983              * CALCULATE INTERACTIONS *
 984              **************************/
 985
 986             r10              = _mm_mul_ps(rsq10,rinv10);
 987
 988             /* Compute parameters for interactions between i and j atoms */
 989             qq10             = _mm_mul_ps(iq1,jq0);
 990
 991             /* EWALD ELECTROSTATICS */
 992
 993             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 994             ewrt             = _mm_mul_ps(r10,ewtabscale);
 995             ewitab           = _mm_cvttps_epi32(ewrt);
 996             eweps            = _mm_sub_ps(ewrt,_mm_round_ps(ewrt, _MM_FROUND_FLOOR));
 997             gmx_mm_load_4pair_swizzle_ps(ewtab + gmx_mm_extract_epi32(ewitab,0),ewtab + gmx_mm_extract_epi32(ewitab,1),
 998                                          ewtab + gmx_mm_extract_epi32(ewitab,2),ewtab + gmx_mm_extract_epi32(ewitab,3),
 999                                          &ewtabF,&ewtabFn);
1000             felec            = _mm_add_ps(_mm_mul_ps( _mm_sub_ps(one,eweps),ewtabF),_mm_mul_ps(eweps,ewtabFn));
1001             felec            = _mm_mul_ps(_mm_mul_ps(qq10,rinv10),_mm_sub_ps(rinvsq10,felec));
1002
1003             fscal            = felec;
1004
1005             /* Calculate temporary vectorial force */
1006             tx               = _mm_mul_ps(fscal,dx10);
1007             ty               = _mm_mul_ps(fscal,dy10);
1008             tz               = _mm_mul_ps(fscal,dz10);
1009
1010             /* Update vectorial force */
1011             fix1             = _mm_add_ps(fix1,tx);
1012             fiy1             = _mm_add_ps(fiy1,ty);
1013             fiz1             = _mm_add_ps(fiz1,tz);
1014
1015             fjx0             = _mm_add_ps(fjx0,tx);
1016             fjy0             = _mm_add_ps(fjy0,ty);
1017             fjz0             = _mm_add_ps(fjz0,tz);
1018
1019             /**************************
1020              * CALCULATE INTERACTIONS *
1021              **************************/
1022
1023             r20              = _mm_mul_ps(rsq20,rinv20);
1024
1025             /* Compute parameters for interactions between i and j atoms */
1026             qq20             = _mm_mul_ps(iq2,jq0);
1027
1028             /* EWALD ELECTROSTATICS */
1029
1030             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1031             ewrt             = _mm_mul_ps(r20,ewtabscale);
1032             ewitab           = _mm_cvttps_epi32(ewrt);
1033             eweps            = _mm_sub_ps(ewrt,_mm_round_ps(ewrt, _MM_FROUND_FLOOR));
1034             gmx_mm_load_4pair_swizzle_ps(ewtab + gmx_mm_extract_epi32(ewitab,0),ewtab + gmx_mm_extract_epi32(ewitab,1),
1035                                          ewtab + gmx_mm_extract_epi32(ewitab,2),ewtab + gmx_mm_extract_epi32(ewitab,3),
1036                                          &ewtabF,&ewtabFn);
1037             felec            = _mm_add_ps(_mm_mul_ps( _mm_sub_ps(one,eweps),ewtabF),_mm_mul_ps(eweps,ewtabFn));
1038             felec            = _mm_mul_ps(_mm_mul_ps(qq20,rinv20),_mm_sub_ps(rinvsq20,felec));
1039
1040             fscal            = felec;
1041
1042             /* Calculate temporary vectorial force */
1043             tx               = _mm_mul_ps(fscal,dx20);
1044             ty               = _mm_mul_ps(fscal,dy20);
1045             tz               = _mm_mul_ps(fscal,dz20);
1046
1047             /* Update vectorial force */
1048             fix2             = _mm_add_ps(fix2,tx);
1049             fiy2             = _mm_add_ps(fiy2,ty);
1050             fiz2             = _mm_add_ps(fiz2,tz);
1051
1052             fjx0             = _mm_add_ps(fjx0,tx);
1053             fjy0             = _mm_add_ps(fjy0,ty);
1054             fjz0             = _mm_add_ps(fjz0,tz);
1055
1056             fjptrA             = f+j_coord_offsetA;
1057             fjptrB             = f+j_coord_offsetB;
1058             fjptrC             = f+j_coord_offsetC;
1059             fjptrD             = f+j_coord_offsetD;
1060
1061             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
1062
1063             /* Inner loop uses 134 flops */
1064         }
1065
1066         if(jidx<j_index_end)
1067         {
1068
1069             /* Get j neighbor index, and coordinate index */
1070             jnrlistA         = jjnr[jidx];
1071             jnrlistB         = jjnr[jidx+1];
1072             jnrlistC         = jjnr[jidx+2];
1073             jnrlistD         = jjnr[jidx+3];
1074             /* Sign of each element will be negative for non-real atoms.
1075              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
1076              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
1077              */
1078             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
1079             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
1080             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
1081             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
1082             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
1083             j_coord_offsetA  = DIM*jnrA;
1084             j_coord_offsetB  = DIM*jnrB;
1085             j_coord_offsetC  = DIM*jnrC;
1086             j_coord_offsetD  = DIM*jnrD;
1087
1088             /* load j atom coordinates */
1089             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
1090                                               x+j_coord_offsetC,x+j_coord_offsetD,
1091                                               &jx0,&jy0,&jz0);
1092
1093             /* Calculate displacement vector */
1094             dx00             = _mm_sub_ps(ix0,jx0);
1095             dy00             = _mm_sub_ps(iy0,jy0);
1096             dz00             = _mm_sub_ps(iz0,jz0);
1097             dx10             = _mm_sub_ps(ix1,jx0);
1098             dy10             = _mm_sub_ps(iy1,jy0);
1099             dz10             = _mm_sub_ps(iz1,jz0);
1100             dx20             = _mm_sub_ps(ix2,jx0);
1101             dy20             = _mm_sub_ps(iy2,jy0);
1102             dz20             = _mm_sub_ps(iz2,jz0);
1103
1104             /* Calculate squared distance and things based on it */
1105             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
1106             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
1107             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
1108
1109             rinv00           = sse41_invsqrt_f(rsq00);
1110             rinv10           = sse41_invsqrt_f(rsq10);
1111             rinv20           = sse41_invsqrt_f(rsq20);
1112
1113             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
1114             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
1115             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
1116
1117             /* Load parameters for j particles */
1118             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
1119                                                               charge+jnrC+0,charge+jnrD+0);
1120             vdwjidx0A        = 2*vdwtype[jnrA+0];
1121             vdwjidx0B        = 2*vdwtype[jnrB+0];
1122             vdwjidx0C        = 2*vdwtype[jnrC+0];
1123             vdwjidx0D        = 2*vdwtype[jnrD+0];
1124
1125             fjx0             = _mm_setzero_ps();
1126             fjy0             = _mm_setzero_ps();
1127             fjz0             = _mm_setzero_ps();
1128
1129             /**************************
1130              * CALCULATE INTERACTIONS *
1131              **************************/
1132
1133             r00              = _mm_mul_ps(rsq00,rinv00);
1134             r00              = _mm_andnot_ps(dummy_mask,r00);
1135
1136             /* Compute parameters for interactions between i and j atoms */
1137             qq00             = _mm_mul_ps(iq0,jq0);
1138             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
1139                                          vdwparam+vdwioffset0+vdwjidx0B,
1140                                          vdwparam+vdwioffset0+vdwjidx0C,
1141                                          vdwparam+vdwioffset0+vdwjidx0D,
1142                                          &c6_00,&c12_00);
1143
1144             /* Calculate table index by multiplying r with table scale and truncate to integer */
1145             rt               = _mm_mul_ps(r00,vftabscale);
1146             vfitab           = _mm_cvttps_epi32(rt);
1147             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
1148             vfitab           = _mm_slli_epi32(vfitab,3);
1149
1150             /* EWALD ELECTROSTATICS */
1151
1152             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1153             ewrt             = _mm_mul_ps(r00,ewtabscale);
1154             ewitab           = _mm_cvttps_epi32(ewrt);
1155             eweps            = _mm_sub_ps(ewrt,_mm_round_ps(ewrt, _MM_FROUND_FLOOR));
1156             gmx_mm_load_4pair_swizzle_ps(ewtab + gmx_mm_extract_epi32(ewitab,0),ewtab + gmx_mm_extract_epi32(ewitab,1),
1157                                          ewtab + gmx_mm_extract_epi32(ewitab,2),ewtab + gmx_mm_extract_epi32(ewitab,3),
1158                                          &ewtabF,&ewtabFn);
1159             felec            = _mm_add_ps(_mm_mul_ps( _mm_sub_ps(one,eweps),ewtabF),_mm_mul_ps(eweps,ewtabFn));
1160             felec            = _mm_mul_ps(_mm_mul_ps(qq00,rinv00),_mm_sub_ps(rinvsq00,felec));
1161
1162             /* CUBIC SPLINE TABLE DISPERSION */
1163             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
1164             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
1165             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
1166             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
1167             _MM_TRANSPOSE4_PS(Y,F,G,H);
1168             Heps             = _mm_mul_ps(vfeps,H);
1169             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
1170             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
1171             fvdw6            = _mm_mul_ps(c6_00,FF);
1172
1173             /* CUBIC SPLINE TABLE REPULSION */
1174             vfitab           = _mm_add_epi32(vfitab,ifour);
1175             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
1176             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
1177             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
1178             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
1179             _MM_TRANSPOSE4_PS(Y,F,G,H);
1180             Heps             = _mm_mul_ps(vfeps,H);
1181             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
1182             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
1183             fvdw12           = _mm_mul_ps(c12_00,FF);
1184             fvdw             = _mm_xor_ps(signbit,_mm_mul_ps(_mm_add_ps(fvdw6,fvdw12),_mm_mul_ps(vftabscale,rinv00)));
1185
1186             fscal            = _mm_add_ps(felec,fvdw);
1187
1188             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1189
1190             /* Calculate temporary vectorial force */
1191             tx               = _mm_mul_ps(fscal,dx00);
1192             ty               = _mm_mul_ps(fscal,dy00);
1193             tz               = _mm_mul_ps(fscal,dz00);
1194
1195             /* Update vectorial force */
1196             fix0             = _mm_add_ps(fix0,tx);
1197             fiy0             = _mm_add_ps(fiy0,ty);
1198             fiz0             = _mm_add_ps(fiz0,tz);
1199
1200             fjx0             = _mm_add_ps(fjx0,tx);
1201             fjy0             = _mm_add_ps(fjy0,ty);
1202             fjz0             = _mm_add_ps(fjz0,tz);
1203
1204             /**************************
1205              * CALCULATE INTERACTIONS *
1206              **************************/
1207
1208             r10              = _mm_mul_ps(rsq10,rinv10);
1209             r10              = _mm_andnot_ps(dummy_mask,r10);
1210
1211             /* Compute parameters for interactions between i and j atoms */
1212             qq10             = _mm_mul_ps(iq1,jq0);
1213
1214             /* EWALD ELECTROSTATICS */
1215
1216             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1217             ewrt             = _mm_mul_ps(r10,ewtabscale);
1218             ewitab           = _mm_cvttps_epi32(ewrt);
1219             eweps            = _mm_sub_ps(ewrt,_mm_round_ps(ewrt, _MM_FROUND_FLOOR));
1220             gmx_mm_load_4pair_swizzle_ps(ewtab + gmx_mm_extract_epi32(ewitab,0),ewtab + gmx_mm_extract_epi32(ewitab,1),
1221                                          ewtab + gmx_mm_extract_epi32(ewitab,2),ewtab + gmx_mm_extract_epi32(ewitab,3),
1222                                          &ewtabF,&ewtabFn);
1223             felec            = _mm_add_ps(_mm_mul_ps( _mm_sub_ps(one,eweps),ewtabF),_mm_mul_ps(eweps,ewtabFn));
1224             felec            = _mm_mul_ps(_mm_mul_ps(qq10,rinv10),_mm_sub_ps(rinvsq10,felec));
1225
1226             fscal            = felec;
1227
1228             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1229
1230             /* Calculate temporary vectorial force */
1231             tx               = _mm_mul_ps(fscal,dx10);
1232             ty               = _mm_mul_ps(fscal,dy10);
1233             tz               = _mm_mul_ps(fscal,dz10);
1234
1235             /* Update vectorial force */
1236             fix1             = _mm_add_ps(fix1,tx);
1237             fiy1             = _mm_add_ps(fiy1,ty);
1238             fiz1             = _mm_add_ps(fiz1,tz);
1239
1240             fjx0             = _mm_add_ps(fjx0,tx);
1241             fjy0             = _mm_add_ps(fjy0,ty);
1242             fjz0             = _mm_add_ps(fjz0,tz);
1243
1244             /**************************
1245              * CALCULATE INTERACTIONS *
1246              **************************/
1247
1248             r20              = _mm_mul_ps(rsq20,rinv20);
1249             r20              = _mm_andnot_ps(dummy_mask,r20);
1250
1251             /* Compute parameters for interactions between i and j atoms */
1252             qq20             = _mm_mul_ps(iq2,jq0);
1253
1254             /* EWALD ELECTROSTATICS */
1255
1256             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1257             ewrt             = _mm_mul_ps(r20,ewtabscale);
1258             ewitab           = _mm_cvttps_epi32(ewrt);
1259             eweps            = _mm_sub_ps(ewrt,_mm_round_ps(ewrt, _MM_FROUND_FLOOR));
1260             gmx_mm_load_4pair_swizzle_ps(ewtab + gmx_mm_extract_epi32(ewitab,0),ewtab + gmx_mm_extract_epi32(ewitab,1),
1261                                          ewtab + gmx_mm_extract_epi32(ewitab,2),ewtab + gmx_mm_extract_epi32(ewitab,3),
1262                                          &ewtabF,&ewtabFn);
1263             felec            = _mm_add_ps(_mm_mul_ps( _mm_sub_ps(one,eweps),ewtabF),_mm_mul_ps(eweps,ewtabFn));
1264             felec            = _mm_mul_ps(_mm_mul_ps(qq20,rinv20),_mm_sub_ps(rinvsq20,felec));
1265
1266             fscal            = felec;
1267
1268             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1269
1270             /* Calculate temporary vectorial force */
1271             tx               = _mm_mul_ps(fscal,dx20);
1272             ty               = _mm_mul_ps(fscal,dy20);
1273             tz               = _mm_mul_ps(fscal,dz20);
1274
1275             /* Update vectorial force */
1276             fix2             = _mm_add_ps(fix2,tx);
1277             fiy2             = _mm_add_ps(fiy2,ty);
1278             fiz2             = _mm_add_ps(fiz2,tz);
1279
1280             fjx0             = _mm_add_ps(fjx0,tx);
1281             fjy0             = _mm_add_ps(fjy0,ty);
1282             fjz0             = _mm_add_ps(fjz0,tz);
1283
1284             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
1285             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
1286             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
1287             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
1288
1289             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
1290
1291             /* Inner loop uses 137 flops */
1292         }
1293
1294         /* End of innermost loop */
1295
1296         gmx_mm_update_iforce_3atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
1297                                               f+i_coord_offset,fshift+i_shift_offset);
1298
1299         /* Increment number of inner iterations */
1300         inneriter                  += j_index_end - j_index_start;
1301
1302         /* Outer loop uses 18 flops */
1303     }
1304
1305     /* Increment number of outer iterations */
1306     outeriter        += nri;
1307
1308     /* Update outer/inner flops */
1309
1310     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3_F,outeriter*18 + inneriter*137);
1311 }