src/gromacs/gmxlib/nonbonded/nb_kernel_sse4_1_double/nb_kernel_ElecEw_VdwCSTab_GeomW3P1_sse4_1_double.c

   1 /*
   2  * This file is part of the GROMACS molecular simulation package.
   3  *
   4  * Copyright (c) 2012,2013,2014, by the GROMACS development team, led by
   5  * Mark Abraham, David van der Spoel, Berk Hess, and Erik Lindahl,
   6  * and including many others, as listed in the AUTHORS file in the
   7  * top-level source directory and at http://www.gromacs.org.
   8  *
   9  * GROMACS is free software; you can redistribute it and/or
  10  * modify it under the terms of the GNU Lesser General Public License
  11  * as published by the Free Software Foundation; either version 2.1
  12  * of the License, or (at your option) any later version.
  13  *
  14  * GROMACS is distributed in the hope that it will be useful,
  15  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  16  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  17  * Lesser General Public License for more details.
  18  *
  19  * You should have received a copy of the GNU Lesser General Public
  20  * License along with GROMACS; if not, see
  21  * http://www.gnu.org/licenses, or write to the Free Software Foundation,
  22  * Inc., 51 Franklin Street, Fifth Floor, Boston, MA  02110-1301  USA.
  23  *
  24  * If you want to redistribute modifications to GROMACS, please
  25  * consider that scientific software is very special. Version
  26  * control is crucial - bugs must be traceable. We will be happy to
  27  * consider code for inclusion in the official distribution, but
  28  * derived work must not be called official GROMACS. Details are found
  29  * in the README & COPYING files - if they are missing, get the
  30  * official version at http://www.gromacs.org.
  31  *
  32  * To help us fund GROMACS development, we humbly ask that you cite
  33  * the research papers on the package. Check out http://www.gromacs.org.
  34  */
  35 /*
  36  * Note: this file was generated by the GROMACS sse4_1_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/legacyheaders/types/simple.h"
  46 #include "gromacs/math/vec.h"
  47 #include "gromacs/legacyheaders/nrnb.h"
  48
  49 #include "gromacs/simd/math_x86_sse4_1_double.h"
  50 #include "kernelutil_x86_sse4_1_double.h"
  51
  52 /*
  53  * Gromacs nonbonded kernel:   nb_kernel_ElecEw_VdwCSTab_GeomW3P1_VF_sse4_1_double
  54  * Electrostatics interaction: Ewald
  55  * VdW interaction:            CubicSplineTable
  56  * Geometry:                   Water3-Particle
  57  * Calculate force/pot:        PotentialAndForce
  58  */
  59 void
  60 nb_kernel_ElecEw_VdwCSTab_GeomW3P1_VF_sse4_1_double
  61                     (t_nblist                    * gmx_restrict       nlist,
  62                      rvec                        * gmx_restrict          xx,
  63                      rvec                        * gmx_restrict          ff,
  64                      t_forcerec                  * gmx_restrict          fr,
  65                      t_mdatoms                   * gmx_restrict     mdatoms,
  66                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
  67                      t_nrnb                      * gmx_restrict        nrnb)
  68 {
  69     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
  70      * just 0 for non-waters.
  71      * Suffixes A,B refer to j loop unrolling done with SSE double precision, e.g. for the two different
  72      * jnr indices corresponding to data put in the four positions in the SIMD register.
  73      */
  74     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
  75     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
  76     int              jnrA,jnrB;
  77     int              j_coord_offsetA,j_coord_offsetB;
  78     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
  79     real             rcutoff_scalar;
  80     real             *shiftvec,*fshift,*x,*f;
  81     __m128d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
  82     int              vdwioffset0;
  83     __m128d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
  84     int              vdwioffset1;
  85     __m128d          ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
  86     int              vdwioffset2;
  87     __m128d          ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
  88     int              vdwjidx0A,vdwjidx0B;
  89     __m128d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  90     __m128d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  91     __m128d          dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
  92     __m128d          dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
  93     __m128d          velec,felec,velecsum,facel,crf,krf,krf2;
  94     real             *charge;
  95     int              nvdwtype;
  96     __m128d          rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
  97     int              *vdwtype;
  98     real             *vdwparam;
  99     __m128d          one_sixth   = _mm_set1_pd(1.0/6.0);
 100     __m128d          one_twelfth = _mm_set1_pd(1.0/12.0);
 101     __m128i          vfitab;
 102     __m128i          ifour       = _mm_set1_epi32(4);
 103     __m128d          rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF;
 104     real             *vftab;
 105     __m128i          ewitab;
 106     __m128d          ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
 107     real             *ewtab;
 108     __m128d          dummy_mask,cutoff_mask;
 109     __m128d          signbit   = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
 110     __m128d          one     = _mm_set1_pd(1.0);
 111     __m128d          two     = _mm_set1_pd(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_pd(fr->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_pd(kernel_data->table_vdw->scale);
 131
 132     sh_ewald         = _mm_set1_pd(fr->ic->sh_ewald);
 133     ewtab            = fr->ic->tabq_coul_FDV0;
 134     ewtabscale       = _mm_set1_pd(fr->ic->tabq_scale);
 135     ewtabhalfspace   = _mm_set1_pd(0.5/fr->ic->tabq_scale);
 136
 137     /* Setup water-specific parameters */
 138     inr              = nlist->iinr[0];
 139     iq0              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+0]));
 140     iq1              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+1]));
 141     iq2              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+2]));
 142     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 143
 144     /* Avoid stupid compiler warnings */
 145     jnrA = jnrB = 0;
 146     j_coord_offsetA = 0;
 147     j_coord_offsetB = 0;
 148
 149     outeriter        = 0;
 150     inneriter        = 0;
 151
 152     /* Start outer loop over neighborlists */
 153     for(iidx=0; iidx<nri; iidx++)
 154     {
 155         /* Load shift vector for this list */
 156         i_shift_offset   = DIM*shiftidx[iidx];
 157
 158         /* Load limits for loop over neighbors */
 159         j_index_start    = jindex[iidx];
 160         j_index_end      = jindex[iidx+1];
 161
 162         /* Get outer coordinate index */
 163         inr              = iinr[iidx];
 164         i_coord_offset   = DIM*inr;
 165
 166         /* Load i particle coords and add shift vector */
 167         gmx_mm_load_shift_and_3rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
 168                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 169
 170         fix0             = _mm_setzero_pd();
 171         fiy0             = _mm_setzero_pd();
 172         fiz0             = _mm_setzero_pd();
 173         fix1             = _mm_setzero_pd();
 174         fiy1             = _mm_setzero_pd();
 175         fiz1             = _mm_setzero_pd();
 176         fix2             = _mm_setzero_pd();
 177         fiy2             = _mm_setzero_pd();
 178         fiz2             = _mm_setzero_pd();
 179
 180         /* Reset potential sums */
 181         velecsum         = _mm_setzero_pd();
 182         vvdwsum          = _mm_setzero_pd();
 183
 184         /* Start inner kernel loop */
 185         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 186         {
 187
 188             /* Get j neighbor index, and coordinate index */
 189             jnrA             = jjnr[jidx];
 190             jnrB             = jjnr[jidx+1];
 191             j_coord_offsetA  = DIM*jnrA;
 192             j_coord_offsetB  = DIM*jnrB;
 193
 194             /* load j atom coordinates */
 195             gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 196                                               &jx0,&jy0,&jz0);
 197
 198             /* Calculate displacement vector */
 199             dx00             = _mm_sub_pd(ix0,jx0);
 200             dy00             = _mm_sub_pd(iy0,jy0);
 201             dz00             = _mm_sub_pd(iz0,jz0);
 202             dx10             = _mm_sub_pd(ix1,jx0);
 203             dy10             = _mm_sub_pd(iy1,jy0);
 204             dz10             = _mm_sub_pd(iz1,jz0);
 205             dx20             = _mm_sub_pd(ix2,jx0);
 206             dy20             = _mm_sub_pd(iy2,jy0);
 207             dz20             = _mm_sub_pd(iz2,jz0);
 208
 209             /* Calculate squared distance and things based on it */
 210             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 211             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
 212             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
 213
 214             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 215             rinv10           = gmx_mm_invsqrt_pd(rsq10);
 216             rinv20           = gmx_mm_invsqrt_pd(rsq20);
 217
 218             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 219             rinvsq10         = _mm_mul_pd(rinv10,rinv10);
 220             rinvsq20         = _mm_mul_pd(rinv20,rinv20);
 221
 222             /* Load parameters for j particles */
 223             jq0              = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
 224             vdwjidx0A        = 2*vdwtype[jnrA+0];
 225             vdwjidx0B        = 2*vdwtype[jnrB+0];
 226
 227             fjx0             = _mm_setzero_pd();
 228             fjy0             = _mm_setzero_pd();
 229             fjz0             = _mm_setzero_pd();
 230
 231             /**************************
 232              * CALCULATE INTERACTIONS *
 233              **************************/
 234
 235             r00              = _mm_mul_pd(rsq00,rinv00);
 236
 237             /* Compute parameters for interactions between i and j atoms */
 238             qq00             = _mm_mul_pd(iq0,jq0);
 239             gmx_mm_load_2pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,
 240                                          vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
 241
 242             /* Calculate table index by multiplying r with table scale and truncate to integer */
 243             rt               = _mm_mul_pd(r00,vftabscale);
 244             vfitab           = _mm_cvttpd_epi32(rt);
 245             vfeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
 246             vfitab           = _mm_slli_epi32(vfitab,3);
 247
 248             /* EWALD ELECTROSTATICS */
 249
 250             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 251             ewrt             = _mm_mul_pd(r00,ewtabscale);
 252             ewitab           = _mm_cvttpd_epi32(ewrt);
 253             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 254             ewitab           = _mm_slli_epi32(ewitab,2);
 255             ewtabF           = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 256             ewtabD           = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,1) );
 257             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 258             ewtabV           = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,0) +2);
 259             ewtabFn          = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,1) +2);
 260             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 261             felec            = _mm_add_pd(ewtabF,_mm_mul_pd(eweps,ewtabD));
 262             velec            = _mm_sub_pd(ewtabV,_mm_mul_pd(_mm_mul_pd(ewtabhalfspace,eweps),_mm_add_pd(ewtabF,felec)));
 263             velec            = _mm_mul_pd(qq00,_mm_sub_pd(rinv00,velec));
 264             felec            = _mm_mul_pd(_mm_mul_pd(qq00,rinv00),_mm_sub_pd(rinvsq00,felec));
 265
 266             /* CUBIC SPLINE TABLE DISPERSION */
 267             Y                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
 268             F                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) );
 269             GMX_MM_TRANSPOSE2_PD(Y,F);
 270             G                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
 271             H                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) +2);
 272             GMX_MM_TRANSPOSE2_PD(G,H);
 273             Heps             = _mm_mul_pd(vfeps,H);
 274             Fp               = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
 275             VV               = _mm_add_pd(Y,_mm_mul_pd(vfeps,Fp));
 276             vvdw6            = _mm_mul_pd(c6_00,VV);
 277             FF               = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
 278             fvdw6            = _mm_mul_pd(c6_00,FF);
 279
 280             /* CUBIC SPLINE TABLE REPULSION */
 281             vfitab           = _mm_add_epi32(vfitab,ifour);
 282             Y                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
 283             F                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) );
 284             GMX_MM_TRANSPOSE2_PD(Y,F);
 285             G                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
 286             H                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) +2);
 287             GMX_MM_TRANSPOSE2_PD(G,H);
 288             Heps             = _mm_mul_pd(vfeps,H);
 289             Fp               = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
 290             VV               = _mm_add_pd(Y,_mm_mul_pd(vfeps,Fp));
 291             vvdw12           = _mm_mul_pd(c12_00,VV);
 292             FF               = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
 293             fvdw12           = _mm_mul_pd(c12_00,FF);
 294             vvdw             = _mm_add_pd(vvdw12,vvdw6);
 295             fvdw             = _mm_xor_pd(signbit,_mm_mul_pd(_mm_add_pd(fvdw6,fvdw12),_mm_mul_pd(vftabscale,rinv00)));
 296
 297             /* Update potential sum for this i atom from the interaction with this j atom. */
 298             velecsum         = _mm_add_pd(velecsum,velec);
 299             vvdwsum          = _mm_add_pd(vvdwsum,vvdw);
 300
 301             fscal            = _mm_add_pd(felec,fvdw);
 302
 303             /* Calculate temporary vectorial force */
 304             tx               = _mm_mul_pd(fscal,dx00);
 305             ty               = _mm_mul_pd(fscal,dy00);
 306             tz               = _mm_mul_pd(fscal,dz00);
 307
 308             /* Update vectorial force */
 309             fix0             = _mm_add_pd(fix0,tx);
 310             fiy0             = _mm_add_pd(fiy0,ty);
 311             fiz0             = _mm_add_pd(fiz0,tz);
 312
 313             fjx0             = _mm_add_pd(fjx0,tx);
 314             fjy0             = _mm_add_pd(fjy0,ty);
 315             fjz0             = _mm_add_pd(fjz0,tz);
 316
 317             /**************************
 318              * CALCULATE INTERACTIONS *
 319              **************************/
 320
 321             r10              = _mm_mul_pd(rsq10,rinv10);
 322
 323             /* Compute parameters for interactions between i and j atoms */
 324             qq10             = _mm_mul_pd(iq1,jq0);
 325
 326             /* EWALD ELECTROSTATICS */
 327
 328             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 329             ewrt             = _mm_mul_pd(r10,ewtabscale);
 330             ewitab           = _mm_cvttpd_epi32(ewrt);
 331             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 332             ewitab           = _mm_slli_epi32(ewitab,2);
 333             ewtabF           = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 334             ewtabD           = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,1) );
 335             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 336             ewtabV           = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,0) +2);
 337             ewtabFn          = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,1) +2);
 338             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 339             felec            = _mm_add_pd(ewtabF,_mm_mul_pd(eweps,ewtabD));
 340             velec            = _mm_sub_pd(ewtabV,_mm_mul_pd(_mm_mul_pd(ewtabhalfspace,eweps),_mm_add_pd(ewtabF,felec)));
 341             velec            = _mm_mul_pd(qq10,_mm_sub_pd(rinv10,velec));
 342             felec            = _mm_mul_pd(_mm_mul_pd(qq10,rinv10),_mm_sub_pd(rinvsq10,felec));
 343
 344             /* Update potential sum for this i atom from the interaction with this j atom. */
 345             velecsum         = _mm_add_pd(velecsum,velec);
 346
 347             fscal            = felec;
 348
 349             /* Calculate temporary vectorial force */
 350             tx               = _mm_mul_pd(fscal,dx10);
 351             ty               = _mm_mul_pd(fscal,dy10);
 352             tz               = _mm_mul_pd(fscal,dz10);
 353
 354             /* Update vectorial force */
 355             fix1             = _mm_add_pd(fix1,tx);
 356             fiy1             = _mm_add_pd(fiy1,ty);
 357             fiz1             = _mm_add_pd(fiz1,tz);
 358
 359             fjx0             = _mm_add_pd(fjx0,tx);
 360             fjy0             = _mm_add_pd(fjy0,ty);
 361             fjz0             = _mm_add_pd(fjz0,tz);
 362
 363             /**************************
 364              * CALCULATE INTERACTIONS *
 365              **************************/
 366
 367             r20              = _mm_mul_pd(rsq20,rinv20);
 368
 369             /* Compute parameters for interactions between i and j atoms */
 370             qq20             = _mm_mul_pd(iq2,jq0);
 371
 372             /* EWALD ELECTROSTATICS */
 373
 374             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 375             ewrt             = _mm_mul_pd(r20,ewtabscale);
 376             ewitab           = _mm_cvttpd_epi32(ewrt);
 377             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 378             ewitab           = _mm_slli_epi32(ewitab,2);
 379             ewtabF           = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 380             ewtabD           = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,1) );
 381             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 382             ewtabV           = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,0) +2);
 383             ewtabFn          = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,1) +2);
 384             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 385             felec            = _mm_add_pd(ewtabF,_mm_mul_pd(eweps,ewtabD));
 386             velec            = _mm_sub_pd(ewtabV,_mm_mul_pd(_mm_mul_pd(ewtabhalfspace,eweps),_mm_add_pd(ewtabF,felec)));
 387             velec            = _mm_mul_pd(qq20,_mm_sub_pd(rinv20,velec));
 388             felec            = _mm_mul_pd(_mm_mul_pd(qq20,rinv20),_mm_sub_pd(rinvsq20,felec));
 389
 390             /* Update potential sum for this i atom from the interaction with this j atom. */
 391             velecsum         = _mm_add_pd(velecsum,velec);
 392
 393             fscal            = felec;
 394
 395             /* Calculate temporary vectorial force */
 396             tx               = _mm_mul_pd(fscal,dx20);
 397             ty               = _mm_mul_pd(fscal,dy20);
 398             tz               = _mm_mul_pd(fscal,dz20);
 399
 400             /* Update vectorial force */
 401             fix2             = _mm_add_pd(fix2,tx);
 402             fiy2             = _mm_add_pd(fiy2,ty);
 403             fiz2             = _mm_add_pd(fiz2,tz);
 404
 405             fjx0             = _mm_add_pd(fjx0,tx);
 406             fjy0             = _mm_add_pd(fjy0,ty);
 407             fjz0             = _mm_add_pd(fjz0,tz);
 408
 409             gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,fjx0,fjy0,fjz0);
 410
 411             /* Inner loop uses 160 flops */
 412         }
 413
 414         if(jidx<j_index_end)
 415         {
 416
 417             jnrA             = jjnr[jidx];
 418             j_coord_offsetA  = DIM*jnrA;
 419
 420             /* load j atom coordinates */
 421             gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
 422                                               &jx0,&jy0,&jz0);
 423
 424             /* Calculate displacement vector */
 425             dx00             = _mm_sub_pd(ix0,jx0);
 426             dy00             = _mm_sub_pd(iy0,jy0);
 427             dz00             = _mm_sub_pd(iz0,jz0);
 428             dx10             = _mm_sub_pd(ix1,jx0);
 429             dy10             = _mm_sub_pd(iy1,jy0);
 430             dz10             = _mm_sub_pd(iz1,jz0);
 431             dx20             = _mm_sub_pd(ix2,jx0);
 432             dy20             = _mm_sub_pd(iy2,jy0);
 433             dz20             = _mm_sub_pd(iz2,jz0);
 434
 435             /* Calculate squared distance and things based on it */
 436             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 437             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
 438             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
 439
 440             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 441             rinv10           = gmx_mm_invsqrt_pd(rsq10);
 442             rinv20           = gmx_mm_invsqrt_pd(rsq20);
 443
 444             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 445             rinvsq10         = _mm_mul_pd(rinv10,rinv10);
 446             rinvsq20         = _mm_mul_pd(rinv20,rinv20);
 447
 448             /* Load parameters for j particles */
 449             jq0              = _mm_load_sd(charge+jnrA+0);
 450             vdwjidx0A        = 2*vdwtype[jnrA+0];
 451
 452             fjx0             = _mm_setzero_pd();
 453             fjy0             = _mm_setzero_pd();
 454             fjz0             = _mm_setzero_pd();
 455
 456             /**************************
 457              * CALCULATE INTERACTIONS *
 458              **************************/
 459
 460             r00              = _mm_mul_pd(rsq00,rinv00);
 461
 462             /* Compute parameters for interactions between i and j atoms */
 463             qq00             = _mm_mul_pd(iq0,jq0);
 464             gmx_mm_load_1pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,&c6_00,&c12_00);
 465
 466             /* Calculate table index by multiplying r with table scale and truncate to integer */
 467             rt               = _mm_mul_pd(r00,vftabscale);
 468             vfitab           = _mm_cvttpd_epi32(rt);
 469             vfeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
 470             vfitab           = _mm_slli_epi32(vfitab,3);
 471
 472             /* EWALD ELECTROSTATICS */
 473
 474             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 475             ewrt             = _mm_mul_pd(r00,ewtabscale);
 476             ewitab           = _mm_cvttpd_epi32(ewrt);
 477             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 478             ewitab           = _mm_slli_epi32(ewitab,2);
 479             ewtabF           = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 480             ewtabD           = _mm_setzero_pd();
 481             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 482             ewtabV           = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,0) +2);
 483             ewtabFn          = _mm_setzero_pd();
 484             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 485             felec            = _mm_add_pd(ewtabF,_mm_mul_pd(eweps,ewtabD));
 486             velec            = _mm_sub_pd(ewtabV,_mm_mul_pd(_mm_mul_pd(ewtabhalfspace,eweps),_mm_add_pd(ewtabF,felec)));
 487             velec            = _mm_mul_pd(qq00,_mm_sub_pd(rinv00,velec));
 488             felec            = _mm_mul_pd(_mm_mul_pd(qq00,rinv00),_mm_sub_pd(rinvsq00,felec));
 489
 490             /* CUBIC SPLINE TABLE DISPERSION */
 491             Y                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
 492             F                = _mm_setzero_pd();
 493             GMX_MM_TRANSPOSE2_PD(Y,F);
 494             G                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
 495             H                = _mm_setzero_pd();
 496             GMX_MM_TRANSPOSE2_PD(G,H);
 497             Heps             = _mm_mul_pd(vfeps,H);
 498             Fp               = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
 499             VV               = _mm_add_pd(Y,_mm_mul_pd(vfeps,Fp));
 500             vvdw6            = _mm_mul_pd(c6_00,VV);
 501             FF               = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
 502             fvdw6            = _mm_mul_pd(c6_00,FF);
 503
 504             /* CUBIC SPLINE TABLE REPULSION */
 505             vfitab           = _mm_add_epi32(vfitab,ifour);
 506             Y                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
 507             F                = _mm_setzero_pd();
 508             GMX_MM_TRANSPOSE2_PD(Y,F);
 509             G                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
 510             H                = _mm_setzero_pd();
 511             GMX_MM_TRANSPOSE2_PD(G,H);
 512             Heps             = _mm_mul_pd(vfeps,H);
 513             Fp               = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
 514             VV               = _mm_add_pd(Y,_mm_mul_pd(vfeps,Fp));
 515             vvdw12           = _mm_mul_pd(c12_00,VV);
 516             FF               = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
 517             fvdw12           = _mm_mul_pd(c12_00,FF);
 518             vvdw             = _mm_add_pd(vvdw12,vvdw6);
 519             fvdw             = _mm_xor_pd(signbit,_mm_mul_pd(_mm_add_pd(fvdw6,fvdw12),_mm_mul_pd(vftabscale,rinv00)));
 520
 521             /* Update potential sum for this i atom from the interaction with this j atom. */
 522             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 523             velecsum         = _mm_add_pd(velecsum,velec);
 524             vvdw             = _mm_unpacklo_pd(vvdw,_mm_setzero_pd());
 525             vvdwsum          = _mm_add_pd(vvdwsum,vvdw);
 526
 527             fscal            = _mm_add_pd(felec,fvdw);
 528
 529             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 530
 531             /* Calculate temporary vectorial force */
 532             tx               = _mm_mul_pd(fscal,dx00);
 533             ty               = _mm_mul_pd(fscal,dy00);
 534             tz               = _mm_mul_pd(fscal,dz00);
 535
 536             /* Update vectorial force */
 537             fix0             = _mm_add_pd(fix0,tx);
 538             fiy0             = _mm_add_pd(fiy0,ty);
 539             fiz0             = _mm_add_pd(fiz0,tz);
 540
 541             fjx0             = _mm_add_pd(fjx0,tx);
 542             fjy0             = _mm_add_pd(fjy0,ty);
 543             fjz0             = _mm_add_pd(fjz0,tz);
 544
 545             /**************************
 546              * CALCULATE INTERACTIONS *
 547              **************************/
 548
 549             r10              = _mm_mul_pd(rsq10,rinv10);
 550
 551             /* Compute parameters for interactions between i and j atoms */
 552             qq10             = _mm_mul_pd(iq1,jq0);
 553
 554             /* EWALD ELECTROSTATICS */
 555
 556             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 557             ewrt             = _mm_mul_pd(r10,ewtabscale);
 558             ewitab           = _mm_cvttpd_epi32(ewrt);
 559             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 560             ewitab           = _mm_slli_epi32(ewitab,2);
 561             ewtabF           = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 562             ewtabD           = _mm_setzero_pd();
 563             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 564             ewtabV           = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,0) +2);
 565             ewtabFn          = _mm_setzero_pd();
 566             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 567             felec            = _mm_add_pd(ewtabF,_mm_mul_pd(eweps,ewtabD));
 568             velec            = _mm_sub_pd(ewtabV,_mm_mul_pd(_mm_mul_pd(ewtabhalfspace,eweps),_mm_add_pd(ewtabF,felec)));
 569             velec            = _mm_mul_pd(qq10,_mm_sub_pd(rinv10,velec));
 570             felec            = _mm_mul_pd(_mm_mul_pd(qq10,rinv10),_mm_sub_pd(rinvsq10,felec));
 571
 572             /* Update potential sum for this i atom from the interaction with this j atom. */
 573             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 574             velecsum         = _mm_add_pd(velecsum,velec);
 575
 576             fscal            = felec;
 577
 578             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 579
 580             /* Calculate temporary vectorial force */
 581             tx               = _mm_mul_pd(fscal,dx10);
 582             ty               = _mm_mul_pd(fscal,dy10);
 583             tz               = _mm_mul_pd(fscal,dz10);
 584
 585             /* Update vectorial force */
 586             fix1             = _mm_add_pd(fix1,tx);
 587             fiy1             = _mm_add_pd(fiy1,ty);
 588             fiz1             = _mm_add_pd(fiz1,tz);
 589
 590             fjx0             = _mm_add_pd(fjx0,tx);
 591             fjy0             = _mm_add_pd(fjy0,ty);
 592             fjz0             = _mm_add_pd(fjz0,tz);
 593
 594             /**************************
 595              * CALCULATE INTERACTIONS *
 596              **************************/
 597
 598             r20              = _mm_mul_pd(rsq20,rinv20);
 599
 600             /* Compute parameters for interactions between i and j atoms */
 601             qq20             = _mm_mul_pd(iq2,jq0);
 602
 603             /* EWALD ELECTROSTATICS */
 604
 605             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 606             ewrt             = _mm_mul_pd(r20,ewtabscale);
 607             ewitab           = _mm_cvttpd_epi32(ewrt);
 608             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 609             ewitab           = _mm_slli_epi32(ewitab,2);
 610             ewtabF           = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 611             ewtabD           = _mm_setzero_pd();
 612             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 613             ewtabV           = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,0) +2);
 614             ewtabFn          = _mm_setzero_pd();
 615             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 616             felec            = _mm_add_pd(ewtabF,_mm_mul_pd(eweps,ewtabD));
 617             velec            = _mm_sub_pd(ewtabV,_mm_mul_pd(_mm_mul_pd(ewtabhalfspace,eweps),_mm_add_pd(ewtabF,felec)));
 618             velec            = _mm_mul_pd(qq20,_mm_sub_pd(rinv20,velec));
 619             felec            = _mm_mul_pd(_mm_mul_pd(qq20,rinv20),_mm_sub_pd(rinvsq20,felec));
 620
 621             /* Update potential sum for this i atom from the interaction with this j atom. */
 622             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 623             velecsum         = _mm_add_pd(velecsum,velec);
 624
 625             fscal            = felec;
 626
 627             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 628
 629             /* Calculate temporary vectorial force */
 630             tx               = _mm_mul_pd(fscal,dx20);
 631             ty               = _mm_mul_pd(fscal,dy20);
 632             tz               = _mm_mul_pd(fscal,dz20);
 633
 634             /* Update vectorial force */
 635             fix2             = _mm_add_pd(fix2,tx);
 636             fiy2             = _mm_add_pd(fiy2,ty);
 637             fiz2             = _mm_add_pd(fiz2,tz);
 638
 639             fjx0             = _mm_add_pd(fjx0,tx);
 640             fjy0             = _mm_add_pd(fjy0,ty);
 641             fjz0             = _mm_add_pd(fjz0,tz);
 642
 643             gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,fjx0,fjy0,fjz0);
 644
 645             /* Inner loop uses 160 flops */
 646         }
 647
 648         /* End of innermost loop */
 649
 650         gmx_mm_update_iforce_3atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
 651                                               f+i_coord_offset,fshift+i_shift_offset);
 652
 653         ggid                        = gid[iidx];
 654         /* Update potential energies */
 655         gmx_mm_update_1pot_pd(velecsum,kernel_data->energygrp_elec+ggid);
 656         gmx_mm_update_1pot_pd(vvdwsum,kernel_data->energygrp_vdw+ggid);
 657
 658         /* Increment number of inner iterations */
 659         inneriter                  += j_index_end - j_index_start;
 660
 661         /* Outer loop uses 20 flops */
 662     }
 663
 664     /* Increment number of outer iterations */
 665     outeriter        += nri;
 666
 667     /* Update outer/inner flops */
 668
 669     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3_VF,outeriter*20 + inneriter*160);
 670 }
 671 /*
 672  * Gromacs nonbonded kernel:   nb_kernel_ElecEw_VdwCSTab_GeomW3P1_F_sse4_1_double
 673  * Electrostatics interaction: Ewald
 674  * VdW interaction:            CubicSplineTable
 675  * Geometry:                   Water3-Particle
 676  * Calculate force/pot:        Force
 677  */
 678 void
 679 nb_kernel_ElecEw_VdwCSTab_GeomW3P1_F_sse4_1_double
 680                     (t_nblist                    * gmx_restrict       nlist,
 681                      rvec                        * gmx_restrict          xx,
 682                      rvec                        * gmx_restrict          ff,
 683                      t_forcerec                  * gmx_restrict          fr,
 684                      t_mdatoms                   * gmx_restrict     mdatoms,
 685                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 686                      t_nrnb                      * gmx_restrict        nrnb)
 687 {
 688     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 689      * just 0 for non-waters.
 690      * Suffixes A,B refer to j loop unrolling done with SSE double precision, e.g. for the two different
 691      * jnr indices corresponding to data put in the four positions in the SIMD register.
 692      */
 693     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 694     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 695     int              jnrA,jnrB;
 696     int              j_coord_offsetA,j_coord_offsetB;
 697     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 698     real             rcutoff_scalar;
 699     real             *shiftvec,*fshift,*x,*f;
 700     __m128d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 701     int              vdwioffset0;
 702     __m128d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 703     int              vdwioffset1;
 704     __m128d          ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
 705     int              vdwioffset2;
 706     __m128d          ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
 707     int              vdwjidx0A,vdwjidx0B;
 708     __m128d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 709     __m128d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 710     __m128d          dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
 711     __m128d          dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
 712     __m128d          velec,felec,velecsum,facel,crf,krf,krf2;
 713     real             *charge;
 714     int              nvdwtype;
 715     __m128d          rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 716     int              *vdwtype;
 717     real             *vdwparam;
 718     __m128d          one_sixth   = _mm_set1_pd(1.0/6.0);
 719     __m128d          one_twelfth = _mm_set1_pd(1.0/12.0);
 720     __m128i          vfitab;
 721     __m128i          ifour       = _mm_set1_epi32(4);
 722     __m128d          rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF;
 723     real             *vftab;
 724     __m128i          ewitab;
 725     __m128d          ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
 726     real             *ewtab;
 727     __m128d          dummy_mask,cutoff_mask;
 728     __m128d          signbit   = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
 729     __m128d          one     = _mm_set1_pd(1.0);
 730     __m128d          two     = _mm_set1_pd(2.0);
 731     x                = xx[0];
 732     f                = ff[0];
 733
 734     nri              = nlist->nri;
 735     iinr             = nlist->iinr;
 736     jindex           = nlist->jindex;
 737     jjnr             = nlist->jjnr;
 738     shiftidx         = nlist->shift;
 739     gid              = nlist->gid;
 740     shiftvec         = fr->shift_vec[0];
 741     fshift           = fr->fshift[0];
 742     facel            = _mm_set1_pd(fr->epsfac);
 743     charge           = mdatoms->chargeA;
 744     nvdwtype         = fr->ntype;
 745     vdwparam         = fr->nbfp;
 746     vdwtype          = mdatoms->typeA;
 747
 748     vftab            = kernel_data->table_vdw->data;
 749     vftabscale       = _mm_set1_pd(kernel_data->table_vdw->scale);
 750
 751     sh_ewald         = _mm_set1_pd(fr->ic->sh_ewald);
 752     ewtab            = fr->ic->tabq_coul_F;
 753     ewtabscale       = _mm_set1_pd(fr->ic->tabq_scale);
 754     ewtabhalfspace   = _mm_set1_pd(0.5/fr->ic->tabq_scale);
 755
 756     /* Setup water-specific parameters */
 757     inr              = nlist->iinr[0];
 758     iq0              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+0]));
 759     iq1              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+1]));
 760     iq2              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+2]));
 761     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 762
 763     /* Avoid stupid compiler warnings */
 764     jnrA = jnrB = 0;
 765     j_coord_offsetA = 0;
 766     j_coord_offsetB = 0;
 767
 768     outeriter        = 0;
 769     inneriter        = 0;
 770
 771     /* Start outer loop over neighborlists */
 772     for(iidx=0; iidx<nri; iidx++)
 773     {
 774         /* Load shift vector for this list */
 775         i_shift_offset   = DIM*shiftidx[iidx];
 776
 777         /* Load limits for loop over neighbors */
 778         j_index_start    = jindex[iidx];
 779         j_index_end      = jindex[iidx+1];
 780
 781         /* Get outer coordinate index */
 782         inr              = iinr[iidx];
 783         i_coord_offset   = DIM*inr;
 784
 785         /* Load i particle coords and add shift vector */
 786         gmx_mm_load_shift_and_3rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
 787                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 788
 789         fix0             = _mm_setzero_pd();
 790         fiy0             = _mm_setzero_pd();
 791         fiz0             = _mm_setzero_pd();
 792         fix1             = _mm_setzero_pd();
 793         fiy1             = _mm_setzero_pd();
 794         fiz1             = _mm_setzero_pd();
 795         fix2             = _mm_setzero_pd();
 796         fiy2             = _mm_setzero_pd();
 797         fiz2             = _mm_setzero_pd();
 798
 799         /* Start inner kernel loop */
 800         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 801         {
 802
 803             /* Get j neighbor index, and coordinate index */
 804             jnrA             = jjnr[jidx];
 805             jnrB             = jjnr[jidx+1];
 806             j_coord_offsetA  = DIM*jnrA;
 807             j_coord_offsetB  = DIM*jnrB;
 808
 809             /* load j atom coordinates */
 810             gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 811                                               &jx0,&jy0,&jz0);
 812
 813             /* Calculate displacement vector */
 814             dx00             = _mm_sub_pd(ix0,jx0);
 815             dy00             = _mm_sub_pd(iy0,jy0);
 816             dz00             = _mm_sub_pd(iz0,jz0);
 817             dx10             = _mm_sub_pd(ix1,jx0);
 818             dy10             = _mm_sub_pd(iy1,jy0);
 819             dz10             = _mm_sub_pd(iz1,jz0);
 820             dx20             = _mm_sub_pd(ix2,jx0);
 821             dy20             = _mm_sub_pd(iy2,jy0);
 822             dz20             = _mm_sub_pd(iz2,jz0);
 823
 824             /* Calculate squared distance and things based on it */
 825             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 826             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
 827             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
 828
 829             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 830             rinv10           = gmx_mm_invsqrt_pd(rsq10);
 831             rinv20           = gmx_mm_invsqrt_pd(rsq20);
 832
 833             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 834             rinvsq10         = _mm_mul_pd(rinv10,rinv10);
 835             rinvsq20         = _mm_mul_pd(rinv20,rinv20);
 836
 837             /* Load parameters for j particles */
 838             jq0              = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
 839             vdwjidx0A        = 2*vdwtype[jnrA+0];
 840             vdwjidx0B        = 2*vdwtype[jnrB+0];
 841
 842             fjx0             = _mm_setzero_pd();
 843             fjy0             = _mm_setzero_pd();
 844             fjz0             = _mm_setzero_pd();
 845
 846             /**************************
 847              * CALCULATE INTERACTIONS *
 848              **************************/
 849
 850             r00              = _mm_mul_pd(rsq00,rinv00);
 851
 852             /* Compute parameters for interactions between i and j atoms */
 853             qq00             = _mm_mul_pd(iq0,jq0);
 854             gmx_mm_load_2pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,
 855                                          vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
 856
 857             /* Calculate table index by multiplying r with table scale and truncate to integer */
 858             rt               = _mm_mul_pd(r00,vftabscale);
 859             vfitab           = _mm_cvttpd_epi32(rt);
 860             vfeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
 861             vfitab           = _mm_slli_epi32(vfitab,3);
 862
 863             /* EWALD ELECTROSTATICS */
 864
 865             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 866             ewrt             = _mm_mul_pd(r00,ewtabscale);
 867             ewitab           = _mm_cvttpd_epi32(ewrt);
 868             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 869             gmx_mm_load_2pair_swizzle_pd(ewtab+gmx_mm_extract_epi32(ewitab,0),ewtab+gmx_mm_extract_epi32(ewitab,1),
 870                                          &ewtabF,&ewtabFn);
 871             felec            = _mm_add_pd(_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF),_mm_mul_pd(eweps,ewtabFn));
 872             felec            = _mm_mul_pd(_mm_mul_pd(qq00,rinv00),_mm_sub_pd(rinvsq00,felec));
 873
 874             /* CUBIC SPLINE TABLE DISPERSION */
 875             Y                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
 876             F                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) );
 877             GMX_MM_TRANSPOSE2_PD(Y,F);
 878             G                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
 879             H                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) +2);
 880             GMX_MM_TRANSPOSE2_PD(G,H);
 881             Heps             = _mm_mul_pd(vfeps,H);
 882             Fp               = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
 883             FF               = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
 884             fvdw6            = _mm_mul_pd(c6_00,FF);
 885
 886             /* CUBIC SPLINE TABLE REPULSION */
 887             vfitab           = _mm_add_epi32(vfitab,ifour);
 888             Y                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
 889             F                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) );
 890             GMX_MM_TRANSPOSE2_PD(Y,F);
 891             G                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
 892             H                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) +2);
 893             GMX_MM_TRANSPOSE2_PD(G,H);
 894             Heps             = _mm_mul_pd(vfeps,H);
 895             Fp               = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
 896             FF               = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
 897             fvdw12           = _mm_mul_pd(c12_00,FF);
 898             fvdw             = _mm_xor_pd(signbit,_mm_mul_pd(_mm_add_pd(fvdw6,fvdw12),_mm_mul_pd(vftabscale,rinv00)));
 899
 900             fscal            = _mm_add_pd(felec,fvdw);
 901
 902             /* Calculate temporary vectorial force */
 903             tx               = _mm_mul_pd(fscal,dx00);
 904             ty               = _mm_mul_pd(fscal,dy00);
 905             tz               = _mm_mul_pd(fscal,dz00);
 906
 907             /* Update vectorial force */
 908             fix0             = _mm_add_pd(fix0,tx);
 909             fiy0             = _mm_add_pd(fiy0,ty);
 910             fiz0             = _mm_add_pd(fiz0,tz);
 911
 912             fjx0             = _mm_add_pd(fjx0,tx);
 913             fjy0             = _mm_add_pd(fjy0,ty);
 914             fjz0             = _mm_add_pd(fjz0,tz);
 915
 916             /**************************
 917              * CALCULATE INTERACTIONS *
 918              **************************/
 919
 920             r10              = _mm_mul_pd(rsq10,rinv10);
 921
 922             /* Compute parameters for interactions between i and j atoms */
 923             qq10             = _mm_mul_pd(iq1,jq0);
 924
 925             /* EWALD ELECTROSTATICS */
 926
 927             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 928             ewrt             = _mm_mul_pd(r10,ewtabscale);
 929             ewitab           = _mm_cvttpd_epi32(ewrt);
 930             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 931             gmx_mm_load_2pair_swizzle_pd(ewtab+gmx_mm_extract_epi32(ewitab,0),ewtab+gmx_mm_extract_epi32(ewitab,1),
 932                                          &ewtabF,&ewtabFn);
 933             felec            = _mm_add_pd(_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF),_mm_mul_pd(eweps,ewtabFn));
 934             felec            = _mm_mul_pd(_mm_mul_pd(qq10,rinv10),_mm_sub_pd(rinvsq10,felec));
 935
 936             fscal            = felec;
 937
 938             /* Calculate temporary vectorial force */
 939             tx               = _mm_mul_pd(fscal,dx10);
 940             ty               = _mm_mul_pd(fscal,dy10);
 941             tz               = _mm_mul_pd(fscal,dz10);
 942
 943             /* Update vectorial force */
 944             fix1             = _mm_add_pd(fix1,tx);
 945             fiy1             = _mm_add_pd(fiy1,ty);
 946             fiz1             = _mm_add_pd(fiz1,tz);
 947
 948             fjx0             = _mm_add_pd(fjx0,tx);
 949             fjy0             = _mm_add_pd(fjy0,ty);
 950             fjz0             = _mm_add_pd(fjz0,tz);
 951
 952             /**************************
 953              * CALCULATE INTERACTIONS *
 954              **************************/
 955
 956             r20              = _mm_mul_pd(rsq20,rinv20);
 957
 958             /* Compute parameters for interactions between i and j atoms */
 959             qq20             = _mm_mul_pd(iq2,jq0);
 960
 961             /* EWALD ELECTROSTATICS */
 962
 963             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 964             ewrt             = _mm_mul_pd(r20,ewtabscale);
 965             ewitab           = _mm_cvttpd_epi32(ewrt);
 966             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 967             gmx_mm_load_2pair_swizzle_pd(ewtab+gmx_mm_extract_epi32(ewitab,0),ewtab+gmx_mm_extract_epi32(ewitab,1),
 968                                          &ewtabF,&ewtabFn);
 969             felec            = _mm_add_pd(_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF),_mm_mul_pd(eweps,ewtabFn));
 970             felec            = _mm_mul_pd(_mm_mul_pd(qq20,rinv20),_mm_sub_pd(rinvsq20,felec));
 971
 972             fscal            = felec;
 973
 974             /* Calculate temporary vectorial force */
 975             tx               = _mm_mul_pd(fscal,dx20);
 976             ty               = _mm_mul_pd(fscal,dy20);
 977             tz               = _mm_mul_pd(fscal,dz20);
 978
 979             /* Update vectorial force */
 980             fix2             = _mm_add_pd(fix2,tx);
 981             fiy2             = _mm_add_pd(fiy2,ty);
 982             fiz2             = _mm_add_pd(fiz2,tz);
 983
 984             fjx0             = _mm_add_pd(fjx0,tx);
 985             fjy0             = _mm_add_pd(fjy0,ty);
 986             fjz0             = _mm_add_pd(fjz0,tz);
 987
 988             gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,fjx0,fjy0,fjz0);
 989
 990             /* Inner loop uses 137 flops */
 991         }
 992
 993         if(jidx<j_index_end)
 994         {
 995
 996             jnrA             = jjnr[jidx];
 997             j_coord_offsetA  = DIM*jnrA;
 998
 999             /* load j atom coordinates */
1000             gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
1001                                               &jx0,&jy0,&jz0);
1002
1003             /* Calculate displacement vector */
1004             dx00             = _mm_sub_pd(ix0,jx0);
1005             dy00             = _mm_sub_pd(iy0,jy0);
1006             dz00             = _mm_sub_pd(iz0,jz0);
1007             dx10             = _mm_sub_pd(ix1,jx0);
1008             dy10             = _mm_sub_pd(iy1,jy0);
1009             dz10             = _mm_sub_pd(iz1,jz0);
1010             dx20             = _mm_sub_pd(ix2,jx0);
1011             dy20             = _mm_sub_pd(iy2,jy0);
1012             dz20             = _mm_sub_pd(iz2,jz0);
1013
1014             /* Calculate squared distance and things based on it */
1015             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
1016             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
1017             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
1018
1019             rinv00           = gmx_mm_invsqrt_pd(rsq00);
1020             rinv10           = gmx_mm_invsqrt_pd(rsq10);
1021             rinv20           = gmx_mm_invsqrt_pd(rsq20);
1022
1023             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
1024             rinvsq10         = _mm_mul_pd(rinv10,rinv10);
1025             rinvsq20         = _mm_mul_pd(rinv20,rinv20);
1026
1027             /* Load parameters for j particles */
1028             jq0              = _mm_load_sd(charge+jnrA+0);
1029             vdwjidx0A        = 2*vdwtype[jnrA+0];
1030
1031             fjx0             = _mm_setzero_pd();
1032             fjy0             = _mm_setzero_pd();
1033             fjz0             = _mm_setzero_pd();
1034
1035             /**************************
1036              * CALCULATE INTERACTIONS *
1037              **************************/
1038
1039             r00              = _mm_mul_pd(rsq00,rinv00);
1040
1041             /* Compute parameters for interactions between i and j atoms */
1042             qq00             = _mm_mul_pd(iq0,jq0);
1043             gmx_mm_load_1pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,&c6_00,&c12_00);
1044
1045             /* Calculate table index by multiplying r with table scale and truncate to integer */
1046             rt               = _mm_mul_pd(r00,vftabscale);
1047             vfitab           = _mm_cvttpd_epi32(rt);
1048             vfeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
1049             vfitab           = _mm_slli_epi32(vfitab,3);
1050
1051             /* EWALD ELECTROSTATICS */
1052
1053             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1054             ewrt             = _mm_mul_pd(r00,ewtabscale);
1055             ewitab           = _mm_cvttpd_epi32(ewrt);
1056             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
1057             gmx_mm_load_1pair_swizzle_pd(ewtab+gmx_mm_extract_epi32(ewitab,0),&ewtabF,&ewtabFn);
1058             felec            = _mm_add_pd(_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF),_mm_mul_pd(eweps,ewtabFn));
1059             felec            = _mm_mul_pd(_mm_mul_pd(qq00,rinv00),_mm_sub_pd(rinvsq00,felec));
1060
1061             /* CUBIC SPLINE TABLE DISPERSION */
1062             Y                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
1063             F                = _mm_setzero_pd();
1064             GMX_MM_TRANSPOSE2_PD(Y,F);
1065             G                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
1066             H                = _mm_setzero_pd();
1067             GMX_MM_TRANSPOSE2_PD(G,H);
1068             Heps             = _mm_mul_pd(vfeps,H);
1069             Fp               = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
1070             FF               = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
1071             fvdw6            = _mm_mul_pd(c6_00,FF);
1072
1073             /* CUBIC SPLINE TABLE REPULSION */
1074             vfitab           = _mm_add_epi32(vfitab,ifour);
1075             Y                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
1076             F                = _mm_setzero_pd();
1077             GMX_MM_TRANSPOSE2_PD(Y,F);
1078             G                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
1079             H                = _mm_setzero_pd();
1080             GMX_MM_TRANSPOSE2_PD(G,H);
1081             Heps             = _mm_mul_pd(vfeps,H);
1082             Fp               = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
1083             FF               = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
1084             fvdw12           = _mm_mul_pd(c12_00,FF);
1085             fvdw             = _mm_xor_pd(signbit,_mm_mul_pd(_mm_add_pd(fvdw6,fvdw12),_mm_mul_pd(vftabscale,rinv00)));
1086
1087             fscal            = _mm_add_pd(felec,fvdw);
1088
1089             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1090
1091             /* Calculate temporary vectorial force */
1092             tx               = _mm_mul_pd(fscal,dx00);
1093             ty               = _mm_mul_pd(fscal,dy00);
1094             tz               = _mm_mul_pd(fscal,dz00);
1095
1096             /* Update vectorial force */
1097             fix0             = _mm_add_pd(fix0,tx);
1098             fiy0             = _mm_add_pd(fiy0,ty);
1099             fiz0             = _mm_add_pd(fiz0,tz);
1100
1101             fjx0             = _mm_add_pd(fjx0,tx);
1102             fjy0             = _mm_add_pd(fjy0,ty);
1103             fjz0             = _mm_add_pd(fjz0,tz);
1104
1105             /**************************
1106              * CALCULATE INTERACTIONS *
1107              **************************/
1108
1109             r10              = _mm_mul_pd(rsq10,rinv10);
1110
1111             /* Compute parameters for interactions between i and j atoms */
1112             qq10             = _mm_mul_pd(iq1,jq0);
1113
1114             /* EWALD ELECTROSTATICS */
1115
1116             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1117             ewrt             = _mm_mul_pd(r10,ewtabscale);
1118             ewitab           = _mm_cvttpd_epi32(ewrt);
1119             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
1120             gmx_mm_load_1pair_swizzle_pd(ewtab+gmx_mm_extract_epi32(ewitab,0),&ewtabF,&ewtabFn);
1121             felec            = _mm_add_pd(_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF),_mm_mul_pd(eweps,ewtabFn));
1122             felec            = _mm_mul_pd(_mm_mul_pd(qq10,rinv10),_mm_sub_pd(rinvsq10,felec));
1123
1124             fscal            = felec;
1125
1126             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1127
1128             /* Calculate temporary vectorial force */
1129             tx               = _mm_mul_pd(fscal,dx10);
1130             ty               = _mm_mul_pd(fscal,dy10);
1131             tz               = _mm_mul_pd(fscal,dz10);
1132
1133             /* Update vectorial force */
1134             fix1             = _mm_add_pd(fix1,tx);
1135             fiy1             = _mm_add_pd(fiy1,ty);
1136             fiz1             = _mm_add_pd(fiz1,tz);
1137
1138             fjx0             = _mm_add_pd(fjx0,tx);
1139             fjy0             = _mm_add_pd(fjy0,ty);
1140             fjz0             = _mm_add_pd(fjz0,tz);
1141
1142             /**************************
1143              * CALCULATE INTERACTIONS *
1144              **************************/
1145
1146             r20              = _mm_mul_pd(rsq20,rinv20);
1147
1148             /* Compute parameters for interactions between i and j atoms */
1149             qq20             = _mm_mul_pd(iq2,jq0);
1150
1151             /* EWALD ELECTROSTATICS */
1152
1153             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1154             ewrt             = _mm_mul_pd(r20,ewtabscale);
1155             ewitab           = _mm_cvttpd_epi32(ewrt);
1156             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
1157             gmx_mm_load_1pair_swizzle_pd(ewtab+gmx_mm_extract_epi32(ewitab,0),&ewtabF,&ewtabFn);
1158             felec            = _mm_add_pd(_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF),_mm_mul_pd(eweps,ewtabFn));
1159             felec            = _mm_mul_pd(_mm_mul_pd(qq20,rinv20),_mm_sub_pd(rinvsq20,felec));
1160
1161             fscal            = felec;
1162
1163             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1164
1165             /* Calculate temporary vectorial force */
1166             tx               = _mm_mul_pd(fscal,dx20);
1167             ty               = _mm_mul_pd(fscal,dy20);
1168             tz               = _mm_mul_pd(fscal,dz20);
1169
1170             /* Update vectorial force */
1171             fix2             = _mm_add_pd(fix2,tx);
1172             fiy2             = _mm_add_pd(fiy2,ty);
1173             fiz2             = _mm_add_pd(fiz2,tz);
1174
1175             fjx0             = _mm_add_pd(fjx0,tx);
1176             fjy0             = _mm_add_pd(fjy0,ty);
1177             fjz0             = _mm_add_pd(fjz0,tz);
1178
1179             gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,fjx0,fjy0,fjz0);
1180
1181             /* Inner loop uses 137 flops */
1182         }
1183
1184         /* End of innermost loop */
1185
1186         gmx_mm_update_iforce_3atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
1187                                               f+i_coord_offset,fshift+i_shift_offset);
1188
1189         /* Increment number of inner iterations */
1190         inneriter                  += j_index_end - j_index_start;
1191
1192         /* Outer loop uses 18 flops */
1193     }
1194
1195     /* Increment number of outer iterations */
1196     outeriter        += nri;
1197
1198     /* Update outer/inner flops */
1199
1200     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3_F,outeriter*18 + inneriter*137);
1201 }