src/gromacs/gmxlib/nonbonded/nb_kernel_sse2_double/nb_kernel_ElecEw_VdwCSTab_GeomW3P1_sse2_double.cpp

   1 /*
   2  * This file is part of the GROMACS molecular simulation package.
   3  *
   4  * Copyright (c) 2012,2013,2014,2015,2017,2018, 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 sse2_double 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_sse2_double.h"
  48
  49 /*
  50  * Gromacs nonbonded kernel:   nb_kernel_ElecEw_VdwCSTab_GeomW3P1_VF_sse2_double
  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_sse2_double
  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 refer to j loop unrolling done with SSE double precision, e.g. for the two 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;
  74     int              j_coord_offsetA,j_coord_offsetB;
  75     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
  76     real             rcutoff_scalar;
  77     real             *shiftvec,*fshift,*x,*f;
  78     __m128d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
  79     int              vdwioffset0;
  80     __m128d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
  81     int              vdwioffset1;
  82     __m128d          ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
  83     int              vdwioffset2;
  84     __m128d          ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
  85     int              vdwjidx0A,vdwjidx0B;
  86     __m128d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  87     __m128d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  88     __m128d          dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
  89     __m128d          dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
  90     __m128d          velec,felec,velecsum,facel,crf,krf,krf2;
  91     real             *charge;
  92     int              nvdwtype;
  93     __m128d          rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
  94     int              *vdwtype;
  95     real             *vdwparam;
  96     __m128d          one_sixth   = _mm_set1_pd(1.0/6.0);
  97     __m128d          one_twelfth = _mm_set1_pd(1.0/12.0);
  98     __m128i          vfitab;
  99     __m128i          ifour       = _mm_set1_epi32(4);
 100     __m128d          rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF;
 101     real             *vftab;
 102     __m128i          ewitab;
 103     __m128d          ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
 104     real             *ewtab;
 105     __m128d          dummy_mask,cutoff_mask;
 106     __m128d          signbit   = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
 107     __m128d          one     = _mm_set1_pd(1.0);
 108     __m128d          two     = _mm_set1_pd(2.0);
 109     x                = xx[0];
 110     f                = ff[0];
 111
 112     nri              = nlist->nri;
 113     iinr             = nlist->iinr;
 114     jindex           = nlist->jindex;
 115     jjnr             = nlist->jjnr;
 116     shiftidx         = nlist->shift;
 117     gid              = nlist->gid;
 118     shiftvec         = fr->shift_vec[0];
 119     fshift           = fr->fshift[0];
 120     facel            = _mm_set1_pd(fr->ic->epsfac);
 121     charge           = mdatoms->chargeA;
 122     nvdwtype         = fr->ntype;
 123     vdwparam         = fr->nbfp;
 124     vdwtype          = mdatoms->typeA;
 125
 126     vftab            = kernel_data->table_vdw->data;
 127     vftabscale       = _mm_set1_pd(kernel_data->table_vdw->scale);
 128
 129     sh_ewald         = _mm_set1_pd(fr->ic->sh_ewald);
 130     ewtab            = fr->ic->tabq_coul_FDV0;
 131     ewtabscale       = _mm_set1_pd(fr->ic->tabq_scale);
 132     ewtabhalfspace   = _mm_set1_pd(0.5/fr->ic->tabq_scale);
 133
 134     /* Setup water-specific parameters */
 135     inr              = nlist->iinr[0];
 136     iq0              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+0]));
 137     iq1              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+1]));
 138     iq2              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+2]));
 139     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 140
 141     /* Avoid stupid compiler warnings */
 142     jnrA = jnrB = 0;
 143     j_coord_offsetA = 0;
 144     j_coord_offsetB = 0;
 145
 146     outeriter        = 0;
 147     inneriter        = 0;
 148
 149     /* Start outer loop over neighborlists */
 150     for(iidx=0; iidx<nri; iidx++)
 151     {
 152         /* Load shift vector for this list */
 153         i_shift_offset   = DIM*shiftidx[iidx];
 154
 155         /* Load limits for loop over neighbors */
 156         j_index_start    = jindex[iidx];
 157         j_index_end      = jindex[iidx+1];
 158
 159         /* Get outer coordinate index */
 160         inr              = iinr[iidx];
 161         i_coord_offset   = DIM*inr;
 162
 163         /* Load i particle coords and add shift vector */
 164         gmx_mm_load_shift_and_3rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
 165                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 166
 167         fix0             = _mm_setzero_pd();
 168         fiy0             = _mm_setzero_pd();
 169         fiz0             = _mm_setzero_pd();
 170         fix1             = _mm_setzero_pd();
 171         fiy1             = _mm_setzero_pd();
 172         fiz1             = _mm_setzero_pd();
 173         fix2             = _mm_setzero_pd();
 174         fiy2             = _mm_setzero_pd();
 175         fiz2             = _mm_setzero_pd();
 176
 177         /* Reset potential sums */
 178         velecsum         = _mm_setzero_pd();
 179         vvdwsum          = _mm_setzero_pd();
 180
 181         /* Start inner kernel loop */
 182         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 183         {
 184
 185             /* Get j neighbor index, and coordinate index */
 186             jnrA             = jjnr[jidx];
 187             jnrB             = jjnr[jidx+1];
 188             j_coord_offsetA  = DIM*jnrA;
 189             j_coord_offsetB  = DIM*jnrB;
 190
 191             /* load j atom coordinates */
 192             gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 193                                               &jx0,&jy0,&jz0);
 194
 195             /* Calculate displacement vector */
 196             dx00             = _mm_sub_pd(ix0,jx0);
 197             dy00             = _mm_sub_pd(iy0,jy0);
 198             dz00             = _mm_sub_pd(iz0,jz0);
 199             dx10             = _mm_sub_pd(ix1,jx0);
 200             dy10             = _mm_sub_pd(iy1,jy0);
 201             dz10             = _mm_sub_pd(iz1,jz0);
 202             dx20             = _mm_sub_pd(ix2,jx0);
 203             dy20             = _mm_sub_pd(iy2,jy0);
 204             dz20             = _mm_sub_pd(iz2,jz0);
 205
 206             /* Calculate squared distance and things based on it */
 207             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 208             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
 209             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
 210
 211             rinv00           = sse2_invsqrt_d(rsq00);
 212             rinv10           = sse2_invsqrt_d(rsq10);
 213             rinv20           = sse2_invsqrt_d(rsq20);
 214
 215             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 216             rinvsq10         = _mm_mul_pd(rinv10,rinv10);
 217             rinvsq20         = _mm_mul_pd(rinv20,rinv20);
 218
 219             /* Load parameters for j particles */
 220             jq0              = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
 221             vdwjidx0A        = 2*vdwtype[jnrA+0];
 222             vdwjidx0B        = 2*vdwtype[jnrB+0];
 223
 224             fjx0             = _mm_setzero_pd();
 225             fjy0             = _mm_setzero_pd();
 226             fjz0             = _mm_setzero_pd();
 227
 228             /**************************
 229              * CALCULATE INTERACTIONS *
 230              **************************/
 231
 232             r00              = _mm_mul_pd(rsq00,rinv00);
 233
 234             /* Compute parameters for interactions between i and j atoms */
 235             qq00             = _mm_mul_pd(iq0,jq0);
 236             gmx_mm_load_2pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,
 237                                          vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
 238
 239             /* Calculate table index by multiplying r with table scale and truncate to integer */
 240             rt               = _mm_mul_pd(r00,vftabscale);
 241             vfitab           = _mm_cvttpd_epi32(rt);
 242             vfeps            = _mm_sub_pd(rt,_mm_cvtepi32_pd(vfitab));
 243             vfitab           = _mm_slli_epi32(vfitab,3);
 244
 245             /* EWALD ELECTROSTATICS */
 246
 247             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 248             ewrt             = _mm_mul_pd(r00,ewtabscale);
 249             ewitab           = _mm_cvttpd_epi32(ewrt);
 250             eweps            = _mm_sub_pd(ewrt,_mm_cvtepi32_pd(ewitab));
 251             ewitab           = _mm_slli_epi32(ewitab,2);
 252             ewtabF           = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 253             ewtabD           = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,1) );
 254             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 255             ewtabV           = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,0) +2);
 256             ewtabFn          = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,1) +2);
 257             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 258             felec            = _mm_add_pd(ewtabF,_mm_mul_pd(eweps,ewtabD));
 259             velec            = _mm_sub_pd(ewtabV,_mm_mul_pd(_mm_mul_pd(ewtabhalfspace,eweps),_mm_add_pd(ewtabF,felec)));
 260             velec            = _mm_mul_pd(qq00,_mm_sub_pd(rinv00,velec));
 261             felec            = _mm_mul_pd(_mm_mul_pd(qq00,rinv00),_mm_sub_pd(rinvsq00,felec));
 262
 263             /* CUBIC SPLINE TABLE DISPERSION */
 264             Y                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
 265             F                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) );
 266             GMX_MM_TRANSPOSE2_PD(Y,F);
 267             G                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
 268             H                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) +2);
 269             GMX_MM_TRANSPOSE2_PD(G,H);
 270             Heps             = _mm_mul_pd(vfeps,H);
 271             Fp               = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
 272             VV               = _mm_add_pd(Y,_mm_mul_pd(vfeps,Fp));
 273             vvdw6            = _mm_mul_pd(c6_00,VV);
 274             FF               = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
 275             fvdw6            = _mm_mul_pd(c6_00,FF);
 276
 277             /* CUBIC SPLINE TABLE REPULSION */
 278             vfitab           = _mm_add_epi32(vfitab,ifour);
 279             Y                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
 280             F                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) );
 281             GMX_MM_TRANSPOSE2_PD(Y,F);
 282             G                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
 283             H                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) +2);
 284             GMX_MM_TRANSPOSE2_PD(G,H);
 285             Heps             = _mm_mul_pd(vfeps,H);
 286             Fp               = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
 287             VV               = _mm_add_pd(Y,_mm_mul_pd(vfeps,Fp));
 288             vvdw12           = _mm_mul_pd(c12_00,VV);
 289             FF               = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
 290             fvdw12           = _mm_mul_pd(c12_00,FF);
 291             vvdw             = _mm_add_pd(vvdw12,vvdw6);
 292             fvdw             = _mm_xor_pd(signbit,_mm_mul_pd(_mm_add_pd(fvdw6,fvdw12),_mm_mul_pd(vftabscale,rinv00)));
 293
 294             /* Update potential sum for this i atom from the interaction with this j atom. */
 295             velecsum         = _mm_add_pd(velecsum,velec);
 296             vvdwsum          = _mm_add_pd(vvdwsum,vvdw);
 297
 298             fscal            = _mm_add_pd(felec,fvdw);
 299
 300             /* Calculate temporary vectorial force */
 301             tx               = _mm_mul_pd(fscal,dx00);
 302             ty               = _mm_mul_pd(fscal,dy00);
 303             tz               = _mm_mul_pd(fscal,dz00);
 304
 305             /* Update vectorial force */
 306             fix0             = _mm_add_pd(fix0,tx);
 307             fiy0             = _mm_add_pd(fiy0,ty);
 308             fiz0             = _mm_add_pd(fiz0,tz);
 309
 310             fjx0             = _mm_add_pd(fjx0,tx);
 311             fjy0             = _mm_add_pd(fjy0,ty);
 312             fjz0             = _mm_add_pd(fjz0,tz);
 313
 314             /**************************
 315              * CALCULATE INTERACTIONS *
 316              **************************/
 317
 318             r10              = _mm_mul_pd(rsq10,rinv10);
 319
 320             /* Compute parameters for interactions between i and j atoms */
 321             qq10             = _mm_mul_pd(iq1,jq0);
 322
 323             /* EWALD ELECTROSTATICS */
 324
 325             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 326             ewrt             = _mm_mul_pd(r10,ewtabscale);
 327             ewitab           = _mm_cvttpd_epi32(ewrt);
 328             eweps            = _mm_sub_pd(ewrt,_mm_cvtepi32_pd(ewitab));
 329             ewitab           = _mm_slli_epi32(ewitab,2);
 330             ewtabF           = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 331             ewtabD           = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,1) );
 332             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 333             ewtabV           = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,0) +2);
 334             ewtabFn          = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,1) +2);
 335             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 336             felec            = _mm_add_pd(ewtabF,_mm_mul_pd(eweps,ewtabD));
 337             velec            = _mm_sub_pd(ewtabV,_mm_mul_pd(_mm_mul_pd(ewtabhalfspace,eweps),_mm_add_pd(ewtabF,felec)));
 338             velec            = _mm_mul_pd(qq10,_mm_sub_pd(rinv10,velec));
 339             felec            = _mm_mul_pd(_mm_mul_pd(qq10,rinv10),_mm_sub_pd(rinvsq10,felec));
 340
 341             /* Update potential sum for this i atom from the interaction with this j atom. */
 342             velecsum         = _mm_add_pd(velecsum,velec);
 343
 344             fscal            = felec;
 345
 346             /* Calculate temporary vectorial force */
 347             tx               = _mm_mul_pd(fscal,dx10);
 348             ty               = _mm_mul_pd(fscal,dy10);
 349             tz               = _mm_mul_pd(fscal,dz10);
 350
 351             /* Update vectorial force */
 352             fix1             = _mm_add_pd(fix1,tx);
 353             fiy1             = _mm_add_pd(fiy1,ty);
 354             fiz1             = _mm_add_pd(fiz1,tz);
 355
 356             fjx0             = _mm_add_pd(fjx0,tx);
 357             fjy0             = _mm_add_pd(fjy0,ty);
 358             fjz0             = _mm_add_pd(fjz0,tz);
 359
 360             /**************************
 361              * CALCULATE INTERACTIONS *
 362              **************************/
 363
 364             r20              = _mm_mul_pd(rsq20,rinv20);
 365
 366             /* Compute parameters for interactions between i and j atoms */
 367             qq20             = _mm_mul_pd(iq2,jq0);
 368
 369             /* EWALD ELECTROSTATICS */
 370
 371             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 372             ewrt             = _mm_mul_pd(r20,ewtabscale);
 373             ewitab           = _mm_cvttpd_epi32(ewrt);
 374             eweps            = _mm_sub_pd(ewrt,_mm_cvtepi32_pd(ewitab));
 375             ewitab           = _mm_slli_epi32(ewitab,2);
 376             ewtabF           = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 377             ewtabD           = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,1) );
 378             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 379             ewtabV           = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,0) +2);
 380             ewtabFn          = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,1) +2);
 381             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 382             felec            = _mm_add_pd(ewtabF,_mm_mul_pd(eweps,ewtabD));
 383             velec            = _mm_sub_pd(ewtabV,_mm_mul_pd(_mm_mul_pd(ewtabhalfspace,eweps),_mm_add_pd(ewtabF,felec)));
 384             velec            = _mm_mul_pd(qq20,_mm_sub_pd(rinv20,velec));
 385             felec            = _mm_mul_pd(_mm_mul_pd(qq20,rinv20),_mm_sub_pd(rinvsq20,felec));
 386
 387             /* Update potential sum for this i atom from the interaction with this j atom. */
 388             velecsum         = _mm_add_pd(velecsum,velec);
 389
 390             fscal            = felec;
 391
 392             /* Calculate temporary vectorial force */
 393             tx               = _mm_mul_pd(fscal,dx20);
 394             ty               = _mm_mul_pd(fscal,dy20);
 395             tz               = _mm_mul_pd(fscal,dz20);
 396
 397             /* Update vectorial force */
 398             fix2             = _mm_add_pd(fix2,tx);
 399             fiy2             = _mm_add_pd(fiy2,ty);
 400             fiz2             = _mm_add_pd(fiz2,tz);
 401
 402             fjx0             = _mm_add_pd(fjx0,tx);
 403             fjy0             = _mm_add_pd(fjy0,ty);
 404             fjz0             = _mm_add_pd(fjz0,tz);
 405
 406             gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,fjx0,fjy0,fjz0);
 407
 408             /* Inner loop uses 160 flops */
 409         }
 410
 411         if(jidx<j_index_end)
 412         {
 413
 414             jnrA             = jjnr[jidx];
 415             j_coord_offsetA  = DIM*jnrA;
 416
 417             /* load j atom coordinates */
 418             gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
 419                                               &jx0,&jy0,&jz0);
 420
 421             /* Calculate displacement vector */
 422             dx00             = _mm_sub_pd(ix0,jx0);
 423             dy00             = _mm_sub_pd(iy0,jy0);
 424             dz00             = _mm_sub_pd(iz0,jz0);
 425             dx10             = _mm_sub_pd(ix1,jx0);
 426             dy10             = _mm_sub_pd(iy1,jy0);
 427             dz10             = _mm_sub_pd(iz1,jz0);
 428             dx20             = _mm_sub_pd(ix2,jx0);
 429             dy20             = _mm_sub_pd(iy2,jy0);
 430             dz20             = _mm_sub_pd(iz2,jz0);
 431
 432             /* Calculate squared distance and things based on it */
 433             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 434             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
 435             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
 436
 437             rinv00           = sse2_invsqrt_d(rsq00);
 438             rinv10           = sse2_invsqrt_d(rsq10);
 439             rinv20           = sse2_invsqrt_d(rsq20);
 440
 441             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 442             rinvsq10         = _mm_mul_pd(rinv10,rinv10);
 443             rinvsq20         = _mm_mul_pd(rinv20,rinv20);
 444
 445             /* Load parameters for j particles */
 446             jq0              = _mm_load_sd(charge+jnrA+0);
 447             vdwjidx0A        = 2*vdwtype[jnrA+0];
 448
 449             fjx0             = _mm_setzero_pd();
 450             fjy0             = _mm_setzero_pd();
 451             fjz0             = _mm_setzero_pd();
 452
 453             /**************************
 454              * CALCULATE INTERACTIONS *
 455              **************************/
 456
 457             r00              = _mm_mul_pd(rsq00,rinv00);
 458
 459             /* Compute parameters for interactions between i and j atoms */
 460             qq00             = _mm_mul_pd(iq0,jq0);
 461             gmx_mm_load_1pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,&c6_00,&c12_00);
 462
 463             /* Calculate table index by multiplying r with table scale and truncate to integer */
 464             rt               = _mm_mul_pd(r00,vftabscale);
 465             vfitab           = _mm_cvttpd_epi32(rt);
 466             vfeps            = _mm_sub_pd(rt,_mm_cvtepi32_pd(vfitab));
 467             vfitab           = _mm_slli_epi32(vfitab,3);
 468
 469             /* EWALD ELECTROSTATICS */
 470
 471             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 472             ewrt             = _mm_mul_pd(r00,ewtabscale);
 473             ewitab           = _mm_cvttpd_epi32(ewrt);
 474             eweps            = _mm_sub_pd(ewrt,_mm_cvtepi32_pd(ewitab));
 475             ewitab           = _mm_slli_epi32(ewitab,2);
 476             ewtabF           = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 477             ewtabD           = _mm_setzero_pd();
 478             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 479             ewtabV           = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,0) +2);
 480             ewtabFn          = _mm_setzero_pd();
 481             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 482             felec            = _mm_add_pd(ewtabF,_mm_mul_pd(eweps,ewtabD));
 483             velec            = _mm_sub_pd(ewtabV,_mm_mul_pd(_mm_mul_pd(ewtabhalfspace,eweps),_mm_add_pd(ewtabF,felec)));
 484             velec            = _mm_mul_pd(qq00,_mm_sub_pd(rinv00,velec));
 485             felec            = _mm_mul_pd(_mm_mul_pd(qq00,rinv00),_mm_sub_pd(rinvsq00,felec));
 486
 487             /* CUBIC SPLINE TABLE DISPERSION */
 488             Y                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
 489             F                = _mm_setzero_pd();
 490             GMX_MM_TRANSPOSE2_PD(Y,F);
 491             G                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
 492             H                = _mm_setzero_pd();
 493             GMX_MM_TRANSPOSE2_PD(G,H);
 494             Heps             = _mm_mul_pd(vfeps,H);
 495             Fp               = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
 496             VV               = _mm_add_pd(Y,_mm_mul_pd(vfeps,Fp));
 497             vvdw6            = _mm_mul_pd(c6_00,VV);
 498             FF               = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
 499             fvdw6            = _mm_mul_pd(c6_00,FF);
 500
 501             /* CUBIC SPLINE TABLE REPULSION */
 502             vfitab           = _mm_add_epi32(vfitab,ifour);
 503             Y                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
 504             F                = _mm_setzero_pd();
 505             GMX_MM_TRANSPOSE2_PD(Y,F);
 506             G                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
 507             H                = _mm_setzero_pd();
 508             GMX_MM_TRANSPOSE2_PD(G,H);
 509             Heps             = _mm_mul_pd(vfeps,H);
 510             Fp               = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
 511             VV               = _mm_add_pd(Y,_mm_mul_pd(vfeps,Fp));
 512             vvdw12           = _mm_mul_pd(c12_00,VV);
 513             FF               = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
 514             fvdw12           = _mm_mul_pd(c12_00,FF);
 515             vvdw             = _mm_add_pd(vvdw12,vvdw6);
 516             fvdw             = _mm_xor_pd(signbit,_mm_mul_pd(_mm_add_pd(fvdw6,fvdw12),_mm_mul_pd(vftabscale,rinv00)));
 517
 518             /* Update potential sum for this i atom from the interaction with this j atom. */
 519             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 520             velecsum         = _mm_add_pd(velecsum,velec);
 521             vvdw             = _mm_unpacklo_pd(vvdw,_mm_setzero_pd());
 522             vvdwsum          = _mm_add_pd(vvdwsum,vvdw);
 523
 524             fscal            = _mm_add_pd(felec,fvdw);
 525
 526             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 527
 528             /* Calculate temporary vectorial force */
 529             tx               = _mm_mul_pd(fscal,dx00);
 530             ty               = _mm_mul_pd(fscal,dy00);
 531             tz               = _mm_mul_pd(fscal,dz00);
 532
 533             /* Update vectorial force */
 534             fix0             = _mm_add_pd(fix0,tx);
 535             fiy0             = _mm_add_pd(fiy0,ty);
 536             fiz0             = _mm_add_pd(fiz0,tz);
 537
 538             fjx0             = _mm_add_pd(fjx0,tx);
 539             fjy0             = _mm_add_pd(fjy0,ty);
 540             fjz0             = _mm_add_pd(fjz0,tz);
 541
 542             /**************************
 543              * CALCULATE INTERACTIONS *
 544              **************************/
 545
 546             r10              = _mm_mul_pd(rsq10,rinv10);
 547
 548             /* Compute parameters for interactions between i and j atoms */
 549             qq10             = _mm_mul_pd(iq1,jq0);
 550
 551             /* EWALD ELECTROSTATICS */
 552
 553             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 554             ewrt             = _mm_mul_pd(r10,ewtabscale);
 555             ewitab           = _mm_cvttpd_epi32(ewrt);
 556             eweps            = _mm_sub_pd(ewrt,_mm_cvtepi32_pd(ewitab));
 557             ewitab           = _mm_slli_epi32(ewitab,2);
 558             ewtabF           = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 559             ewtabD           = _mm_setzero_pd();
 560             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 561             ewtabV           = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,0) +2);
 562             ewtabFn          = _mm_setzero_pd();
 563             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 564             felec            = _mm_add_pd(ewtabF,_mm_mul_pd(eweps,ewtabD));
 565             velec            = _mm_sub_pd(ewtabV,_mm_mul_pd(_mm_mul_pd(ewtabhalfspace,eweps),_mm_add_pd(ewtabF,felec)));
 566             velec            = _mm_mul_pd(qq10,_mm_sub_pd(rinv10,velec));
 567             felec            = _mm_mul_pd(_mm_mul_pd(qq10,rinv10),_mm_sub_pd(rinvsq10,felec));
 568
 569             /* Update potential sum for this i atom from the interaction with this j atom. */
 570             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 571             velecsum         = _mm_add_pd(velecsum,velec);
 572
 573             fscal            = felec;
 574
 575             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 576
 577             /* Calculate temporary vectorial force */
 578             tx               = _mm_mul_pd(fscal,dx10);
 579             ty               = _mm_mul_pd(fscal,dy10);
 580             tz               = _mm_mul_pd(fscal,dz10);
 581
 582             /* Update vectorial force */
 583             fix1             = _mm_add_pd(fix1,tx);
 584             fiy1             = _mm_add_pd(fiy1,ty);
 585             fiz1             = _mm_add_pd(fiz1,tz);
 586
 587             fjx0             = _mm_add_pd(fjx0,tx);
 588             fjy0             = _mm_add_pd(fjy0,ty);
 589             fjz0             = _mm_add_pd(fjz0,tz);
 590
 591             /**************************
 592              * CALCULATE INTERACTIONS *
 593              **************************/
 594
 595             r20              = _mm_mul_pd(rsq20,rinv20);
 596
 597             /* Compute parameters for interactions between i and j atoms */
 598             qq20             = _mm_mul_pd(iq2,jq0);
 599
 600             /* EWALD ELECTROSTATICS */
 601
 602             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 603             ewrt             = _mm_mul_pd(r20,ewtabscale);
 604             ewitab           = _mm_cvttpd_epi32(ewrt);
 605             eweps            = _mm_sub_pd(ewrt,_mm_cvtepi32_pd(ewitab));
 606             ewitab           = _mm_slli_epi32(ewitab,2);
 607             ewtabF           = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 608             ewtabD           = _mm_setzero_pd();
 609             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 610             ewtabV           = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,0) +2);
 611             ewtabFn          = _mm_setzero_pd();
 612             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 613             felec            = _mm_add_pd(ewtabF,_mm_mul_pd(eweps,ewtabD));
 614             velec            = _mm_sub_pd(ewtabV,_mm_mul_pd(_mm_mul_pd(ewtabhalfspace,eweps),_mm_add_pd(ewtabF,felec)));
 615             velec            = _mm_mul_pd(qq20,_mm_sub_pd(rinv20,velec));
 616             felec            = _mm_mul_pd(_mm_mul_pd(qq20,rinv20),_mm_sub_pd(rinvsq20,felec));
 617
 618             /* Update potential sum for this i atom from the interaction with this j atom. */
 619             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 620             velecsum         = _mm_add_pd(velecsum,velec);
 621
 622             fscal            = felec;
 623
 624             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 625
 626             /* Calculate temporary vectorial force */
 627             tx               = _mm_mul_pd(fscal,dx20);
 628             ty               = _mm_mul_pd(fscal,dy20);
 629             tz               = _mm_mul_pd(fscal,dz20);
 630
 631             /* Update vectorial force */
 632             fix2             = _mm_add_pd(fix2,tx);
 633             fiy2             = _mm_add_pd(fiy2,ty);
 634             fiz2             = _mm_add_pd(fiz2,tz);
 635
 636             fjx0             = _mm_add_pd(fjx0,tx);
 637             fjy0             = _mm_add_pd(fjy0,ty);
 638             fjz0             = _mm_add_pd(fjz0,tz);
 639
 640             gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,fjx0,fjy0,fjz0);
 641
 642             /* Inner loop uses 160 flops */
 643         }
 644
 645         /* End of innermost loop */
 646
 647         gmx_mm_update_iforce_3atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
 648                                               f+i_coord_offset,fshift+i_shift_offset);
 649
 650         ggid                        = gid[iidx];
 651         /* Update potential energies */
 652         gmx_mm_update_1pot_pd(velecsum,kernel_data->energygrp_elec+ggid);
 653         gmx_mm_update_1pot_pd(vvdwsum,kernel_data->energygrp_vdw+ggid);
 654
 655         /* Increment number of inner iterations */
 656         inneriter                  += j_index_end - j_index_start;
 657
 658         /* Outer loop uses 20 flops */
 659     }
 660
 661     /* Increment number of outer iterations */
 662     outeriter        += nri;
 663
 664     /* Update outer/inner flops */
 665
 666     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3_VF,outeriter*20 + inneriter*160);
 667 }
 668 /*
 669  * Gromacs nonbonded kernel:   nb_kernel_ElecEw_VdwCSTab_GeomW3P1_F_sse2_double
 670  * Electrostatics interaction: Ewald
 671  * VdW interaction:            CubicSplineTable
 672  * Geometry:                   Water3-Particle
 673  * Calculate force/pot:        Force
 674  */
 675 void
 676 nb_kernel_ElecEw_VdwCSTab_GeomW3P1_F_sse2_double
 677                     (t_nblist                    * gmx_restrict       nlist,
 678                      rvec                        * gmx_restrict          xx,
 679                      rvec                        * gmx_restrict          ff,
 680                      struct t_forcerec           * gmx_restrict          fr,
 681                      t_mdatoms                   * gmx_restrict     mdatoms,
 682                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 683                      t_nrnb                      * gmx_restrict        nrnb)
 684 {
 685     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 686      * just 0 for non-waters.
 687      * Suffixes A,B refer to j loop unrolling done with SSE double precision, e.g. for the two different
 688      * jnr indices corresponding to data put in the four positions in the SIMD register.
 689      */
 690     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 691     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 692     int              jnrA,jnrB;
 693     int              j_coord_offsetA,j_coord_offsetB;
 694     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 695     real             rcutoff_scalar;
 696     real             *shiftvec,*fshift,*x,*f;
 697     __m128d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 698     int              vdwioffset0;
 699     __m128d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 700     int              vdwioffset1;
 701     __m128d          ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
 702     int              vdwioffset2;
 703     __m128d          ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
 704     int              vdwjidx0A,vdwjidx0B;
 705     __m128d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 706     __m128d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 707     __m128d          dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
 708     __m128d          dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
 709     __m128d          velec,felec,velecsum,facel,crf,krf,krf2;
 710     real             *charge;
 711     int              nvdwtype;
 712     __m128d          rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 713     int              *vdwtype;
 714     real             *vdwparam;
 715     __m128d          one_sixth   = _mm_set1_pd(1.0/6.0);
 716     __m128d          one_twelfth = _mm_set1_pd(1.0/12.0);
 717     __m128i          vfitab;
 718     __m128i          ifour       = _mm_set1_epi32(4);
 719     __m128d          rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF;
 720     real             *vftab;
 721     __m128i          ewitab;
 722     __m128d          ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
 723     real             *ewtab;
 724     __m128d          dummy_mask,cutoff_mask;
 725     __m128d          signbit   = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
 726     __m128d          one     = _mm_set1_pd(1.0);
 727     __m128d          two     = _mm_set1_pd(2.0);
 728     x                = xx[0];
 729     f                = ff[0];
 730
 731     nri              = nlist->nri;
 732     iinr             = nlist->iinr;
 733     jindex           = nlist->jindex;
 734     jjnr             = nlist->jjnr;
 735     shiftidx         = nlist->shift;
 736     gid              = nlist->gid;
 737     shiftvec         = fr->shift_vec[0];
 738     fshift           = fr->fshift[0];
 739     facel            = _mm_set1_pd(fr->ic->epsfac);
 740     charge           = mdatoms->chargeA;
 741     nvdwtype         = fr->ntype;
 742     vdwparam         = fr->nbfp;
 743     vdwtype          = mdatoms->typeA;
 744
 745     vftab            = kernel_data->table_vdw->data;
 746     vftabscale       = _mm_set1_pd(kernel_data->table_vdw->scale);
 747
 748     sh_ewald         = _mm_set1_pd(fr->ic->sh_ewald);
 749     ewtab            = fr->ic->tabq_coul_F;
 750     ewtabscale       = _mm_set1_pd(fr->ic->tabq_scale);
 751     ewtabhalfspace   = _mm_set1_pd(0.5/fr->ic->tabq_scale);
 752
 753     /* Setup water-specific parameters */
 754     inr              = nlist->iinr[0];
 755     iq0              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+0]));
 756     iq1              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+1]));
 757     iq2              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+2]));
 758     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 759
 760     /* Avoid stupid compiler warnings */
 761     jnrA = jnrB = 0;
 762     j_coord_offsetA = 0;
 763     j_coord_offsetB = 0;
 764
 765     outeriter        = 0;
 766     inneriter        = 0;
 767
 768     /* Start outer loop over neighborlists */
 769     for(iidx=0; iidx<nri; iidx++)
 770     {
 771         /* Load shift vector for this list */
 772         i_shift_offset   = DIM*shiftidx[iidx];
 773
 774         /* Load limits for loop over neighbors */
 775         j_index_start    = jindex[iidx];
 776         j_index_end      = jindex[iidx+1];
 777
 778         /* Get outer coordinate index */
 779         inr              = iinr[iidx];
 780         i_coord_offset   = DIM*inr;
 781
 782         /* Load i particle coords and add shift vector */
 783         gmx_mm_load_shift_and_3rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
 784                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 785
 786         fix0             = _mm_setzero_pd();
 787         fiy0             = _mm_setzero_pd();
 788         fiz0             = _mm_setzero_pd();
 789         fix1             = _mm_setzero_pd();
 790         fiy1             = _mm_setzero_pd();
 791         fiz1             = _mm_setzero_pd();
 792         fix2             = _mm_setzero_pd();
 793         fiy2             = _mm_setzero_pd();
 794         fiz2             = _mm_setzero_pd();
 795
 796         /* Start inner kernel loop */
 797         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 798         {
 799
 800             /* Get j neighbor index, and coordinate index */
 801             jnrA             = jjnr[jidx];
 802             jnrB             = jjnr[jidx+1];
 803             j_coord_offsetA  = DIM*jnrA;
 804             j_coord_offsetB  = DIM*jnrB;
 805
 806             /* load j atom coordinates */
 807             gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 808                                               &jx0,&jy0,&jz0);
 809
 810             /* Calculate displacement vector */
 811             dx00             = _mm_sub_pd(ix0,jx0);
 812             dy00             = _mm_sub_pd(iy0,jy0);
 813             dz00             = _mm_sub_pd(iz0,jz0);
 814             dx10             = _mm_sub_pd(ix1,jx0);
 815             dy10             = _mm_sub_pd(iy1,jy0);
 816             dz10             = _mm_sub_pd(iz1,jz0);
 817             dx20             = _mm_sub_pd(ix2,jx0);
 818             dy20             = _mm_sub_pd(iy2,jy0);
 819             dz20             = _mm_sub_pd(iz2,jz0);
 820
 821             /* Calculate squared distance and things based on it */
 822             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 823             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
 824             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
 825
 826             rinv00           = sse2_invsqrt_d(rsq00);
 827             rinv10           = sse2_invsqrt_d(rsq10);
 828             rinv20           = sse2_invsqrt_d(rsq20);
 829
 830             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 831             rinvsq10         = _mm_mul_pd(rinv10,rinv10);
 832             rinvsq20         = _mm_mul_pd(rinv20,rinv20);
 833
 834             /* Load parameters for j particles */
 835             jq0              = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
 836             vdwjidx0A        = 2*vdwtype[jnrA+0];
 837             vdwjidx0B        = 2*vdwtype[jnrB+0];
 838
 839             fjx0             = _mm_setzero_pd();
 840             fjy0             = _mm_setzero_pd();
 841             fjz0             = _mm_setzero_pd();
 842
 843             /**************************
 844              * CALCULATE INTERACTIONS *
 845              **************************/
 846
 847             r00              = _mm_mul_pd(rsq00,rinv00);
 848
 849             /* Compute parameters for interactions between i and j atoms */
 850             qq00             = _mm_mul_pd(iq0,jq0);
 851             gmx_mm_load_2pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,
 852                                          vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
 853
 854             /* Calculate table index by multiplying r with table scale and truncate to integer */
 855             rt               = _mm_mul_pd(r00,vftabscale);
 856             vfitab           = _mm_cvttpd_epi32(rt);
 857             vfeps            = _mm_sub_pd(rt,_mm_cvtepi32_pd(vfitab));
 858             vfitab           = _mm_slli_epi32(vfitab,3);
 859
 860             /* EWALD ELECTROSTATICS */
 861
 862             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 863             ewrt             = _mm_mul_pd(r00,ewtabscale);
 864             ewitab           = _mm_cvttpd_epi32(ewrt);
 865             eweps            = _mm_sub_pd(ewrt,_mm_cvtepi32_pd(ewitab));
 866             gmx_mm_load_2pair_swizzle_pd(ewtab+gmx_mm_extract_epi32(ewitab,0),ewtab+gmx_mm_extract_epi32(ewitab,1),
 867                                          &ewtabF,&ewtabFn);
 868             felec            = _mm_add_pd(_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF),_mm_mul_pd(eweps,ewtabFn));
 869             felec            = _mm_mul_pd(_mm_mul_pd(qq00,rinv00),_mm_sub_pd(rinvsq00,felec));
 870
 871             /* CUBIC SPLINE TABLE DISPERSION */
 872             Y                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
 873             F                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) );
 874             GMX_MM_TRANSPOSE2_PD(Y,F);
 875             G                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
 876             H                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) +2);
 877             GMX_MM_TRANSPOSE2_PD(G,H);
 878             Heps             = _mm_mul_pd(vfeps,H);
 879             Fp               = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
 880             FF               = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
 881             fvdw6            = _mm_mul_pd(c6_00,FF);
 882
 883             /* CUBIC SPLINE TABLE REPULSION */
 884             vfitab           = _mm_add_epi32(vfitab,ifour);
 885             Y                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
 886             F                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) );
 887             GMX_MM_TRANSPOSE2_PD(Y,F);
 888             G                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
 889             H                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) +2);
 890             GMX_MM_TRANSPOSE2_PD(G,H);
 891             Heps             = _mm_mul_pd(vfeps,H);
 892             Fp               = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
 893             FF               = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
 894             fvdw12           = _mm_mul_pd(c12_00,FF);
 895             fvdw             = _mm_xor_pd(signbit,_mm_mul_pd(_mm_add_pd(fvdw6,fvdw12),_mm_mul_pd(vftabscale,rinv00)));
 896
 897             fscal            = _mm_add_pd(felec,fvdw);
 898
 899             /* Calculate temporary vectorial force */
 900             tx               = _mm_mul_pd(fscal,dx00);
 901             ty               = _mm_mul_pd(fscal,dy00);
 902             tz               = _mm_mul_pd(fscal,dz00);
 903
 904             /* Update vectorial force */
 905             fix0             = _mm_add_pd(fix0,tx);
 906             fiy0             = _mm_add_pd(fiy0,ty);
 907             fiz0             = _mm_add_pd(fiz0,tz);
 908
 909             fjx0             = _mm_add_pd(fjx0,tx);
 910             fjy0             = _mm_add_pd(fjy0,ty);
 911             fjz0             = _mm_add_pd(fjz0,tz);
 912
 913             /**************************
 914              * CALCULATE INTERACTIONS *
 915              **************************/
 916
 917             r10              = _mm_mul_pd(rsq10,rinv10);
 918
 919             /* Compute parameters for interactions between i and j atoms */
 920             qq10             = _mm_mul_pd(iq1,jq0);
 921
 922             /* EWALD ELECTROSTATICS */
 923
 924             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 925             ewrt             = _mm_mul_pd(r10,ewtabscale);
 926             ewitab           = _mm_cvttpd_epi32(ewrt);
 927             eweps            = _mm_sub_pd(ewrt,_mm_cvtepi32_pd(ewitab));
 928             gmx_mm_load_2pair_swizzle_pd(ewtab+gmx_mm_extract_epi32(ewitab,0),ewtab+gmx_mm_extract_epi32(ewitab,1),
 929                                          &ewtabF,&ewtabFn);
 930             felec            = _mm_add_pd(_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF),_mm_mul_pd(eweps,ewtabFn));
 931             felec            = _mm_mul_pd(_mm_mul_pd(qq10,rinv10),_mm_sub_pd(rinvsq10,felec));
 932
 933             fscal            = felec;
 934
 935             /* Calculate temporary vectorial force */
 936             tx               = _mm_mul_pd(fscal,dx10);
 937             ty               = _mm_mul_pd(fscal,dy10);
 938             tz               = _mm_mul_pd(fscal,dz10);
 939
 940             /* Update vectorial force */
 941             fix1             = _mm_add_pd(fix1,tx);
 942             fiy1             = _mm_add_pd(fiy1,ty);
 943             fiz1             = _mm_add_pd(fiz1,tz);
 944
 945             fjx0             = _mm_add_pd(fjx0,tx);
 946             fjy0             = _mm_add_pd(fjy0,ty);
 947             fjz0             = _mm_add_pd(fjz0,tz);
 948
 949             /**************************
 950              * CALCULATE INTERACTIONS *
 951              **************************/
 952
 953             r20              = _mm_mul_pd(rsq20,rinv20);
 954
 955             /* Compute parameters for interactions between i and j atoms */
 956             qq20             = _mm_mul_pd(iq2,jq0);
 957
 958             /* EWALD ELECTROSTATICS */
 959
 960             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 961             ewrt             = _mm_mul_pd(r20,ewtabscale);
 962             ewitab           = _mm_cvttpd_epi32(ewrt);
 963             eweps            = _mm_sub_pd(ewrt,_mm_cvtepi32_pd(ewitab));
 964             gmx_mm_load_2pair_swizzle_pd(ewtab+gmx_mm_extract_epi32(ewitab,0),ewtab+gmx_mm_extract_epi32(ewitab,1),
 965                                          &ewtabF,&ewtabFn);
 966             felec            = _mm_add_pd(_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF),_mm_mul_pd(eweps,ewtabFn));
 967             felec            = _mm_mul_pd(_mm_mul_pd(qq20,rinv20),_mm_sub_pd(rinvsq20,felec));
 968
 969             fscal            = felec;
 970
 971             /* Calculate temporary vectorial force */
 972             tx               = _mm_mul_pd(fscal,dx20);
 973             ty               = _mm_mul_pd(fscal,dy20);
 974             tz               = _mm_mul_pd(fscal,dz20);
 975
 976             /* Update vectorial force */
 977             fix2             = _mm_add_pd(fix2,tx);
 978             fiy2             = _mm_add_pd(fiy2,ty);
 979             fiz2             = _mm_add_pd(fiz2,tz);
 980
 981             fjx0             = _mm_add_pd(fjx0,tx);
 982             fjy0             = _mm_add_pd(fjy0,ty);
 983             fjz0             = _mm_add_pd(fjz0,tz);
 984
 985             gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,fjx0,fjy0,fjz0);
 986
 987             /* Inner loop uses 137 flops */
 988         }
 989
 990         if(jidx<j_index_end)
 991         {
 992
 993             jnrA             = jjnr[jidx];
 994             j_coord_offsetA  = DIM*jnrA;
 995
 996             /* load j atom coordinates */
 997             gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
 998                                               &jx0,&jy0,&jz0);
 999
1000             /* Calculate displacement vector */
1001             dx00             = _mm_sub_pd(ix0,jx0);
1002             dy00             = _mm_sub_pd(iy0,jy0);
1003             dz00             = _mm_sub_pd(iz0,jz0);
1004             dx10             = _mm_sub_pd(ix1,jx0);
1005             dy10             = _mm_sub_pd(iy1,jy0);
1006             dz10             = _mm_sub_pd(iz1,jz0);
1007             dx20             = _mm_sub_pd(ix2,jx0);
1008             dy20             = _mm_sub_pd(iy2,jy0);
1009             dz20             = _mm_sub_pd(iz2,jz0);
1010
1011             /* Calculate squared distance and things based on it */
1012             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
1013             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
1014             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
1015
1016             rinv00           = sse2_invsqrt_d(rsq00);
1017             rinv10           = sse2_invsqrt_d(rsq10);
1018             rinv20           = sse2_invsqrt_d(rsq20);
1019
1020             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
1021             rinvsq10         = _mm_mul_pd(rinv10,rinv10);
1022             rinvsq20         = _mm_mul_pd(rinv20,rinv20);
1023
1024             /* Load parameters for j particles */
1025             jq0              = _mm_load_sd(charge+jnrA+0);
1026             vdwjidx0A        = 2*vdwtype[jnrA+0];
1027
1028             fjx0             = _mm_setzero_pd();
1029             fjy0             = _mm_setzero_pd();
1030             fjz0             = _mm_setzero_pd();
1031
1032             /**************************
1033              * CALCULATE INTERACTIONS *
1034              **************************/
1035
1036             r00              = _mm_mul_pd(rsq00,rinv00);
1037
1038             /* Compute parameters for interactions between i and j atoms */
1039             qq00             = _mm_mul_pd(iq0,jq0);
1040             gmx_mm_load_1pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,&c6_00,&c12_00);
1041
1042             /* Calculate table index by multiplying r with table scale and truncate to integer */
1043             rt               = _mm_mul_pd(r00,vftabscale);
1044             vfitab           = _mm_cvttpd_epi32(rt);
1045             vfeps            = _mm_sub_pd(rt,_mm_cvtepi32_pd(vfitab));
1046             vfitab           = _mm_slli_epi32(vfitab,3);
1047
1048             /* EWALD ELECTROSTATICS */
1049
1050             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1051             ewrt             = _mm_mul_pd(r00,ewtabscale);
1052             ewitab           = _mm_cvttpd_epi32(ewrt);
1053             eweps            = _mm_sub_pd(ewrt,_mm_cvtepi32_pd(ewitab));
1054             gmx_mm_load_1pair_swizzle_pd(ewtab+gmx_mm_extract_epi32(ewitab,0),&ewtabF,&ewtabFn);
1055             felec            = _mm_add_pd(_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF),_mm_mul_pd(eweps,ewtabFn));
1056             felec            = _mm_mul_pd(_mm_mul_pd(qq00,rinv00),_mm_sub_pd(rinvsq00,felec));
1057
1058             /* CUBIC SPLINE TABLE DISPERSION */
1059             Y                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
1060             F                = _mm_setzero_pd();
1061             GMX_MM_TRANSPOSE2_PD(Y,F);
1062             G                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
1063             H                = _mm_setzero_pd();
1064             GMX_MM_TRANSPOSE2_PD(G,H);
1065             Heps             = _mm_mul_pd(vfeps,H);
1066             Fp               = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
1067             FF               = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
1068             fvdw6            = _mm_mul_pd(c6_00,FF);
1069
1070             /* CUBIC SPLINE TABLE REPULSION */
1071             vfitab           = _mm_add_epi32(vfitab,ifour);
1072             Y                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
1073             F                = _mm_setzero_pd();
1074             GMX_MM_TRANSPOSE2_PD(Y,F);
1075             G                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
1076             H                = _mm_setzero_pd();
1077             GMX_MM_TRANSPOSE2_PD(G,H);
1078             Heps             = _mm_mul_pd(vfeps,H);
1079             Fp               = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
1080             FF               = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
1081             fvdw12           = _mm_mul_pd(c12_00,FF);
1082             fvdw             = _mm_xor_pd(signbit,_mm_mul_pd(_mm_add_pd(fvdw6,fvdw12),_mm_mul_pd(vftabscale,rinv00)));
1083
1084             fscal            = _mm_add_pd(felec,fvdw);
1085
1086             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1087
1088             /* Calculate temporary vectorial force */
1089             tx               = _mm_mul_pd(fscal,dx00);
1090             ty               = _mm_mul_pd(fscal,dy00);
1091             tz               = _mm_mul_pd(fscal,dz00);
1092
1093             /* Update vectorial force */
1094             fix0             = _mm_add_pd(fix0,tx);
1095             fiy0             = _mm_add_pd(fiy0,ty);
1096             fiz0             = _mm_add_pd(fiz0,tz);
1097
1098             fjx0             = _mm_add_pd(fjx0,tx);
1099             fjy0             = _mm_add_pd(fjy0,ty);
1100             fjz0             = _mm_add_pd(fjz0,tz);
1101
1102             /**************************
1103              * CALCULATE INTERACTIONS *
1104              **************************/
1105
1106             r10              = _mm_mul_pd(rsq10,rinv10);
1107
1108             /* Compute parameters for interactions between i and j atoms */
1109             qq10             = _mm_mul_pd(iq1,jq0);
1110
1111             /* EWALD ELECTROSTATICS */
1112
1113             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1114             ewrt             = _mm_mul_pd(r10,ewtabscale);
1115             ewitab           = _mm_cvttpd_epi32(ewrt);
1116             eweps            = _mm_sub_pd(ewrt,_mm_cvtepi32_pd(ewitab));
1117             gmx_mm_load_1pair_swizzle_pd(ewtab+gmx_mm_extract_epi32(ewitab,0),&ewtabF,&ewtabFn);
1118             felec            = _mm_add_pd(_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF),_mm_mul_pd(eweps,ewtabFn));
1119             felec            = _mm_mul_pd(_mm_mul_pd(qq10,rinv10),_mm_sub_pd(rinvsq10,felec));
1120
1121             fscal            = felec;
1122
1123             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1124
1125             /* Calculate temporary vectorial force */
1126             tx               = _mm_mul_pd(fscal,dx10);
1127             ty               = _mm_mul_pd(fscal,dy10);
1128             tz               = _mm_mul_pd(fscal,dz10);
1129
1130             /* Update vectorial force */
1131             fix1             = _mm_add_pd(fix1,tx);
1132             fiy1             = _mm_add_pd(fiy1,ty);
1133             fiz1             = _mm_add_pd(fiz1,tz);
1134
1135             fjx0             = _mm_add_pd(fjx0,tx);
1136             fjy0             = _mm_add_pd(fjy0,ty);
1137             fjz0             = _mm_add_pd(fjz0,tz);
1138
1139             /**************************
1140              * CALCULATE INTERACTIONS *
1141              **************************/
1142
1143             r20              = _mm_mul_pd(rsq20,rinv20);
1144
1145             /* Compute parameters for interactions between i and j atoms */
1146             qq20             = _mm_mul_pd(iq2,jq0);
1147
1148             /* EWALD ELECTROSTATICS */
1149
1150             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1151             ewrt             = _mm_mul_pd(r20,ewtabscale);
1152             ewitab           = _mm_cvttpd_epi32(ewrt);
1153             eweps            = _mm_sub_pd(ewrt,_mm_cvtepi32_pd(ewitab));
1154             gmx_mm_load_1pair_swizzle_pd(ewtab+gmx_mm_extract_epi32(ewitab,0),&ewtabF,&ewtabFn);
1155             felec            = _mm_add_pd(_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF),_mm_mul_pd(eweps,ewtabFn));
1156             felec            = _mm_mul_pd(_mm_mul_pd(qq20,rinv20),_mm_sub_pd(rinvsq20,felec));
1157
1158             fscal            = felec;
1159
1160             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1161
1162             /* Calculate temporary vectorial force */
1163             tx               = _mm_mul_pd(fscal,dx20);
1164             ty               = _mm_mul_pd(fscal,dy20);
1165             tz               = _mm_mul_pd(fscal,dz20);
1166
1167             /* Update vectorial force */
1168             fix2             = _mm_add_pd(fix2,tx);
1169             fiy2             = _mm_add_pd(fiy2,ty);
1170             fiz2             = _mm_add_pd(fiz2,tz);
1171
1172             fjx0             = _mm_add_pd(fjx0,tx);
1173             fjy0             = _mm_add_pd(fjy0,ty);
1174             fjz0             = _mm_add_pd(fjz0,tz);
1175
1176             gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,fjx0,fjy0,fjz0);
1177
1178             /* Inner loop uses 137 flops */
1179         }
1180
1181         /* End of innermost loop */
1182
1183         gmx_mm_update_iforce_3atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
1184                                               f+i_coord_offset,fshift+i_shift_offset);
1185
1186         /* Increment number of inner iterations */
1187         inneriter                  += j_index_end - j_index_start;
1188
1189         /* Outer loop uses 18 flops */
1190     }
1191
1192     /* Increment number of outer iterations */
1193     outeriter        += nri;
1194
1195     /* Update outer/inner flops */
1196
1197     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3_F,outeriter*18 + inneriter*137);
1198 }