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 "config.h"
  39
  40 #include <math.h>
  41
  42 #include "../nb_kernel.h"
  43 #include "gromacs/legacyheaders/types/simple.h"
  44 #include "gromacs/math/vec.h"
  45 #include "gromacs/legacyheaders/nrnb.h"
  46
  47 #include "gromacs/simd/math_x86_sse4_1_double.h"
  48 #include "kernelutil_x86_sse4_1_double.h"
  49
  50 /*
  51  * Gromacs nonbonded kernel:   nb_kernel_ElecEw_VdwCSTab_GeomW3P1_VF_sse4_1_double
  52  * Electrostatics interaction: Ewald
  53  * VdW interaction:            CubicSplineTable
  54  * Geometry:                   Water3-Particle
  55  * Calculate force/pot:        PotentialAndForce
  56  */
  57 void
  58 nb_kernel_ElecEw_VdwCSTab_GeomW3P1_VF_sse4_1_double
  59                     (t_nblist                    * gmx_restrict       nlist,
  60                      rvec                        * gmx_restrict          xx,
  61                      rvec                        * gmx_restrict          ff,
  62                      t_forcerec                  * gmx_restrict          fr,
  63                      t_mdatoms                   * gmx_restrict     mdatoms,
  64                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
  65                      t_nrnb                      * gmx_restrict        nrnb)
  66 {
  67     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
  68      * just 0 for non-waters.
  69      * Suffixes A,B refer to j loop unrolling done with SSE double precision, e.g. for the two different
  70      * jnr indices corresponding to data put in the four positions in the SIMD register.
  71      */
  72     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
  73     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
  74     int              jnrA,jnrB;
  75     int              j_coord_offsetA,j_coord_offsetB;
  76     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
  77     real             rcutoff_scalar;
  78     real             *shiftvec,*fshift,*x,*f;
  79     __m128d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
  80     int              vdwioffset0;
  81     __m128d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
  82     int              vdwioffset1;
  83     __m128d          ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
  84     int              vdwioffset2;
  85     __m128d          ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
  86     int              vdwjidx0A,vdwjidx0B;
  87     __m128d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  88     __m128d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  89     __m128d          dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
  90     __m128d          dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
  91     __m128d          velec,felec,velecsum,facel,crf,krf,krf2;
  92     real             *charge;
  93     int              nvdwtype;
  94     __m128d          rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
  95     int              *vdwtype;
  96     real             *vdwparam;
  97     __m128d          one_sixth   = _mm_set1_pd(1.0/6.0);
  98     __m128d          one_twelfth = _mm_set1_pd(1.0/12.0);
  99     __m128i          vfitab;
 100     __m128i          ifour       = _mm_set1_epi32(4);
 101     __m128d          rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF;
 102     real             *vftab;
 103     __m128i          ewitab;
 104     __m128d          ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
 105     real             *ewtab;
 106     __m128d          dummy_mask,cutoff_mask;
 107     __m128d          signbit   = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
 108     __m128d          one     = _mm_set1_pd(1.0);
 109     __m128d          two     = _mm_set1_pd(2.0);
 110     x                = xx[0];
 111     f                = ff[0];
 112
 113     nri              = nlist->nri;
 114     iinr             = nlist->iinr;
 115     jindex           = nlist->jindex;
 116     jjnr             = nlist->jjnr;
 117     shiftidx         = nlist->shift;
 118     gid              = nlist->gid;
 119     shiftvec         = fr->shift_vec[0];
 120     fshift           = fr->fshift[0];
 121     facel            = _mm_set1_pd(fr->epsfac);
 122     charge           = mdatoms->chargeA;
 123     nvdwtype         = fr->ntype;
 124     vdwparam         = fr->nbfp;
 125     vdwtype          = mdatoms->typeA;
 126
 127     vftab            = kernel_data->table_vdw->data;
 128     vftabscale       = _mm_set1_pd(kernel_data->table_vdw->scale);
 129
 130     sh_ewald         = _mm_set1_pd(fr->ic->sh_ewald);
 131     ewtab            = fr->ic->tabq_coul_FDV0;
 132     ewtabscale       = _mm_set1_pd(fr->ic->tabq_scale);
 133     ewtabhalfspace   = _mm_set1_pd(0.5/fr->ic->tabq_scale);
 134
 135     /* Setup water-specific parameters */
 136     inr              = nlist->iinr[0];
 137     iq0              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+0]));
 138     iq1              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+1]));
 139     iq2              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+2]));
 140     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 141
 142     /* Avoid stupid compiler warnings */
 143     jnrA = jnrB = 0;
 144     j_coord_offsetA = 0;
 145     j_coord_offsetB = 0;
 146
 147     outeriter        = 0;
 148     inneriter        = 0;
 149
 150     /* Start outer loop over neighborlists */
 151     for(iidx=0; iidx<nri; iidx++)
 152     {
 153         /* Load shift vector for this list */
 154         i_shift_offset   = DIM*shiftidx[iidx];
 155
 156         /* Load limits for loop over neighbors */
 157         j_index_start    = jindex[iidx];
 158         j_index_end      = jindex[iidx+1];
 159
 160         /* Get outer coordinate index */
 161         inr              = iinr[iidx];
 162         i_coord_offset   = DIM*inr;
 163
 164         /* Load i particle coords and add shift vector */
 165         gmx_mm_load_shift_and_3rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
 166                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 167
 168         fix0             = _mm_setzero_pd();
 169         fiy0             = _mm_setzero_pd();
 170         fiz0             = _mm_setzero_pd();
 171         fix1             = _mm_setzero_pd();
 172         fiy1             = _mm_setzero_pd();
 173         fiz1             = _mm_setzero_pd();
 174         fix2             = _mm_setzero_pd();
 175         fiy2             = _mm_setzero_pd();
 176         fiz2             = _mm_setzero_pd();
 177
 178         /* Reset potential sums */
 179         velecsum         = _mm_setzero_pd();
 180         vvdwsum          = _mm_setzero_pd();
 181
 182         /* Start inner kernel loop */
 183         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 184         {
 185
 186             /* Get j neighbor index, and coordinate index */
 187             jnrA             = jjnr[jidx];
 188             jnrB             = jjnr[jidx+1];
 189             j_coord_offsetA  = DIM*jnrA;
 190             j_coord_offsetB  = DIM*jnrB;
 191
 192             /* load j atom coordinates */
 193             gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 194                                               &jx0,&jy0,&jz0);
 195
 196             /* Calculate displacement vector */
 197             dx00             = _mm_sub_pd(ix0,jx0);
 198             dy00             = _mm_sub_pd(iy0,jy0);
 199             dz00             = _mm_sub_pd(iz0,jz0);
 200             dx10             = _mm_sub_pd(ix1,jx0);
 201             dy10             = _mm_sub_pd(iy1,jy0);
 202             dz10             = _mm_sub_pd(iz1,jz0);
 203             dx20             = _mm_sub_pd(ix2,jx0);
 204             dy20             = _mm_sub_pd(iy2,jy0);
 205             dz20             = _mm_sub_pd(iz2,jz0);
 206
 207             /* Calculate squared distance and things based on it */
 208             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 209             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
 210             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
 211
 212             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 213             rinv10           = gmx_mm_invsqrt_pd(rsq10);
 214             rinv20           = gmx_mm_invsqrt_pd(rsq20);
 215
 216             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 217             rinvsq10         = _mm_mul_pd(rinv10,rinv10);
 218             rinvsq20         = _mm_mul_pd(rinv20,rinv20);
 219
 220             /* Load parameters for j particles */
 221             jq0              = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
 222             vdwjidx0A        = 2*vdwtype[jnrA+0];
 223             vdwjidx0B        = 2*vdwtype[jnrB+0];
 224
 225             fjx0             = _mm_setzero_pd();
 226             fjy0             = _mm_setzero_pd();
 227             fjz0             = _mm_setzero_pd();
 228
 229             /**************************
 230              * CALCULATE INTERACTIONS *
 231              **************************/
 232
 233             r00              = _mm_mul_pd(rsq00,rinv00);
 234
 235             /* Compute parameters for interactions between i and j atoms */
 236             qq00             = _mm_mul_pd(iq0,jq0);
 237             gmx_mm_load_2pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,
 238                                          vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
 239
 240             /* Calculate table index by multiplying r with table scale and truncate to integer */
 241             rt               = _mm_mul_pd(r00,vftabscale);
 242             vfitab           = _mm_cvttpd_epi32(rt);
 243             vfeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
 244             vfitab           = _mm_slli_epi32(vfitab,3);
 245
 246             /* EWALD ELECTROSTATICS */
 247
 248             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 249             ewrt             = _mm_mul_pd(r00,ewtabscale);
 250             ewitab           = _mm_cvttpd_epi32(ewrt);
 251             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 252             ewitab           = _mm_slli_epi32(ewitab,2);
 253             ewtabF           = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 254             ewtabD           = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,1) );
 255             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 256             ewtabV           = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,0) +2);
 257             ewtabFn          = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,1) +2);
 258             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 259             felec            = _mm_add_pd(ewtabF,_mm_mul_pd(eweps,ewtabD));
 260             velec            = _mm_sub_pd(ewtabV,_mm_mul_pd(_mm_mul_pd(ewtabhalfspace,eweps),_mm_add_pd(ewtabF,felec)));
 261             velec            = _mm_mul_pd(qq00,_mm_sub_pd(rinv00,velec));
 262             felec            = _mm_mul_pd(_mm_mul_pd(qq00,rinv00),_mm_sub_pd(rinvsq00,felec));
 263
 264             /* CUBIC SPLINE TABLE DISPERSION */
 265             Y                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
 266             F                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) );
 267             GMX_MM_TRANSPOSE2_PD(Y,F);
 268             G                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
 269             H                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) +2);
 270             GMX_MM_TRANSPOSE2_PD(G,H);
 271             Heps             = _mm_mul_pd(vfeps,H);
 272             Fp               = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
 273             VV               = _mm_add_pd(Y,_mm_mul_pd(vfeps,Fp));
 274             vvdw6            = _mm_mul_pd(c6_00,VV);
 275             FF               = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
 276             fvdw6            = _mm_mul_pd(c6_00,FF);
 277
 278             /* CUBIC SPLINE TABLE REPULSION */
 279             vfitab           = _mm_add_epi32(vfitab,ifour);
 280             Y                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
 281             F                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) );
 282             GMX_MM_TRANSPOSE2_PD(Y,F);
 283             G                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
 284             H                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) +2);
 285             GMX_MM_TRANSPOSE2_PD(G,H);
 286             Heps             = _mm_mul_pd(vfeps,H);
 287             Fp               = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
 288             VV               = _mm_add_pd(Y,_mm_mul_pd(vfeps,Fp));
 289             vvdw12           = _mm_mul_pd(c12_00,VV);
 290             FF               = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
 291             fvdw12           = _mm_mul_pd(c12_00,FF);
 292             vvdw             = _mm_add_pd(vvdw12,vvdw6);
 293             fvdw             = _mm_xor_pd(signbit,_mm_mul_pd(_mm_add_pd(fvdw6,fvdw12),_mm_mul_pd(vftabscale,rinv00)));
 294
 295             /* Update potential sum for this i atom from the interaction with this j atom. */
 296             velecsum         = _mm_add_pd(velecsum,velec);
 297             vvdwsum          = _mm_add_pd(vvdwsum,vvdw);
 298
 299             fscal            = _mm_add_pd(felec,fvdw);
 300
 301             /* Calculate temporary vectorial force */
 302             tx               = _mm_mul_pd(fscal,dx00);
 303             ty               = _mm_mul_pd(fscal,dy00);
 304             tz               = _mm_mul_pd(fscal,dz00);
 305
 306             /* Update vectorial force */
 307             fix0             = _mm_add_pd(fix0,tx);
 308             fiy0             = _mm_add_pd(fiy0,ty);
 309             fiz0             = _mm_add_pd(fiz0,tz);
 310
 311             fjx0             = _mm_add_pd(fjx0,tx);
 312             fjy0             = _mm_add_pd(fjy0,ty);
 313             fjz0             = _mm_add_pd(fjz0,tz);
 314
 315             /**************************
 316              * CALCULATE INTERACTIONS *
 317              **************************/
 318
 319             r10              = _mm_mul_pd(rsq10,rinv10);
 320
 321             /* Compute parameters for interactions between i and j atoms */
 322             qq10             = _mm_mul_pd(iq1,jq0);
 323
 324             /* EWALD ELECTROSTATICS */
 325
 326             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 327             ewrt             = _mm_mul_pd(r10,ewtabscale);
 328             ewitab           = _mm_cvttpd_epi32(ewrt);
 329             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 330             ewitab           = _mm_slli_epi32(ewitab,2);
 331             ewtabF           = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 332             ewtabD           = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,1) );
 333             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 334             ewtabV           = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,0) +2);
 335             ewtabFn          = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,1) +2);
 336             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 337             felec            = _mm_add_pd(ewtabF,_mm_mul_pd(eweps,ewtabD));
 338             velec            = _mm_sub_pd(ewtabV,_mm_mul_pd(_mm_mul_pd(ewtabhalfspace,eweps),_mm_add_pd(ewtabF,felec)));
 339             velec            = _mm_mul_pd(qq10,_mm_sub_pd(rinv10,velec));
 340             felec            = _mm_mul_pd(_mm_mul_pd(qq10,rinv10),_mm_sub_pd(rinvsq10,felec));
 341
 342             /* Update potential sum for this i atom from the interaction with this j atom. */
 343             velecsum         = _mm_add_pd(velecsum,velec);
 344
 345             fscal            = felec;
 346
 347             /* Calculate temporary vectorial force */
 348             tx               = _mm_mul_pd(fscal,dx10);
 349             ty               = _mm_mul_pd(fscal,dy10);
 350             tz               = _mm_mul_pd(fscal,dz10);
 351
 352             /* Update vectorial force */
 353             fix1             = _mm_add_pd(fix1,tx);
 354             fiy1             = _mm_add_pd(fiy1,ty);
 355             fiz1             = _mm_add_pd(fiz1,tz);
 356
 357             fjx0             = _mm_add_pd(fjx0,tx);
 358             fjy0             = _mm_add_pd(fjy0,ty);
 359             fjz0             = _mm_add_pd(fjz0,tz);
 360
 361             /**************************
 362              * CALCULATE INTERACTIONS *
 363              **************************/
 364
 365             r20              = _mm_mul_pd(rsq20,rinv20);
 366
 367             /* Compute parameters for interactions between i and j atoms */
 368             qq20             = _mm_mul_pd(iq2,jq0);
 369
 370             /* EWALD ELECTROSTATICS */
 371
 372             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 373             ewrt             = _mm_mul_pd(r20,ewtabscale);
 374             ewitab           = _mm_cvttpd_epi32(ewrt);
 375             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 376             ewitab           = _mm_slli_epi32(ewitab,2);
 377             ewtabF           = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 378             ewtabD           = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,1) );
 379             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 380             ewtabV           = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,0) +2);
 381             ewtabFn          = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,1) +2);
 382             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 383             felec            = _mm_add_pd(ewtabF,_mm_mul_pd(eweps,ewtabD));
 384             velec            = _mm_sub_pd(ewtabV,_mm_mul_pd(_mm_mul_pd(ewtabhalfspace,eweps),_mm_add_pd(ewtabF,felec)));
 385             velec            = _mm_mul_pd(qq20,_mm_sub_pd(rinv20,velec));
 386             felec            = _mm_mul_pd(_mm_mul_pd(qq20,rinv20),_mm_sub_pd(rinvsq20,felec));
 387
 388             /* Update potential sum for this i atom from the interaction with this j atom. */
 389             velecsum         = _mm_add_pd(velecsum,velec);
 390
 391             fscal            = felec;
 392
 393             /* Calculate temporary vectorial force */
 394             tx               = _mm_mul_pd(fscal,dx20);
 395             ty               = _mm_mul_pd(fscal,dy20);
 396             tz               = _mm_mul_pd(fscal,dz20);
 397
 398             /* Update vectorial force */
 399             fix2             = _mm_add_pd(fix2,tx);
 400             fiy2             = _mm_add_pd(fiy2,ty);
 401             fiz2             = _mm_add_pd(fiz2,tz);
 402
 403             fjx0             = _mm_add_pd(fjx0,tx);
 404             fjy0             = _mm_add_pd(fjy0,ty);
 405             fjz0             = _mm_add_pd(fjz0,tz);
 406
 407             gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,fjx0,fjy0,fjz0);
 408
 409             /* Inner loop uses 160 flops */
 410         }
 411
 412         if(jidx<j_index_end)
 413         {
 414
 415             jnrA             = jjnr[jidx];
 416             j_coord_offsetA  = DIM*jnrA;
 417
 418             /* load j atom coordinates */
 419             gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
 420                                               &jx0,&jy0,&jz0);
 421
 422             /* Calculate displacement vector */
 423             dx00             = _mm_sub_pd(ix0,jx0);
 424             dy00             = _mm_sub_pd(iy0,jy0);
 425             dz00             = _mm_sub_pd(iz0,jz0);
 426             dx10             = _mm_sub_pd(ix1,jx0);
 427             dy10             = _mm_sub_pd(iy1,jy0);
 428             dz10             = _mm_sub_pd(iz1,jz0);
 429             dx20             = _mm_sub_pd(ix2,jx0);
 430             dy20             = _mm_sub_pd(iy2,jy0);
 431             dz20             = _mm_sub_pd(iz2,jz0);
 432
 433             /* Calculate squared distance and things based on it */
 434             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 435             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
 436             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
 437
 438             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 439             rinv10           = gmx_mm_invsqrt_pd(rsq10);
 440             rinv20           = gmx_mm_invsqrt_pd(rsq20);
 441
 442             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 443             rinvsq10         = _mm_mul_pd(rinv10,rinv10);
 444             rinvsq20         = _mm_mul_pd(rinv20,rinv20);
 445
 446             /* Load parameters for j particles */
 447             jq0              = _mm_load_sd(charge+jnrA+0);
 448             vdwjidx0A        = 2*vdwtype[jnrA+0];
 449
 450             fjx0             = _mm_setzero_pd();
 451             fjy0             = _mm_setzero_pd();
 452             fjz0             = _mm_setzero_pd();
 453
 454             /**************************
 455              * CALCULATE INTERACTIONS *
 456              **************************/
 457
 458             r00              = _mm_mul_pd(rsq00,rinv00);
 459
 460             /* Compute parameters for interactions between i and j atoms */
 461             qq00             = _mm_mul_pd(iq0,jq0);
 462             gmx_mm_load_1pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,&c6_00,&c12_00);
 463
 464             /* Calculate table index by multiplying r with table scale and truncate to integer */
 465             rt               = _mm_mul_pd(r00,vftabscale);
 466             vfitab           = _mm_cvttpd_epi32(rt);
 467             vfeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
 468             vfitab           = _mm_slli_epi32(vfitab,3);
 469
 470             /* EWALD ELECTROSTATICS */
 471
 472             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 473             ewrt             = _mm_mul_pd(r00,ewtabscale);
 474             ewitab           = _mm_cvttpd_epi32(ewrt);
 475             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 476             ewitab           = _mm_slli_epi32(ewitab,2);
 477             ewtabF           = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 478             ewtabD           = _mm_setzero_pd();
 479             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 480             ewtabV           = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,0) +2);
 481             ewtabFn          = _mm_setzero_pd();
 482             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 483             felec            = _mm_add_pd(ewtabF,_mm_mul_pd(eweps,ewtabD));
 484             velec            = _mm_sub_pd(ewtabV,_mm_mul_pd(_mm_mul_pd(ewtabhalfspace,eweps),_mm_add_pd(ewtabF,felec)));
 485             velec            = _mm_mul_pd(qq00,_mm_sub_pd(rinv00,velec));
 486             felec            = _mm_mul_pd(_mm_mul_pd(qq00,rinv00),_mm_sub_pd(rinvsq00,felec));
 487
 488             /* CUBIC SPLINE TABLE DISPERSION */
 489             Y                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
 490             F                = _mm_setzero_pd();
 491             GMX_MM_TRANSPOSE2_PD(Y,F);
 492             G                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
 493             H                = _mm_setzero_pd();
 494             GMX_MM_TRANSPOSE2_PD(G,H);
 495             Heps             = _mm_mul_pd(vfeps,H);
 496             Fp               = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
 497             VV               = _mm_add_pd(Y,_mm_mul_pd(vfeps,Fp));
 498             vvdw6            = _mm_mul_pd(c6_00,VV);
 499             FF               = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
 500             fvdw6            = _mm_mul_pd(c6_00,FF);
 501
 502             /* CUBIC SPLINE TABLE REPULSION */
 503             vfitab           = _mm_add_epi32(vfitab,ifour);
 504             Y                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
 505             F                = _mm_setzero_pd();
 506             GMX_MM_TRANSPOSE2_PD(Y,F);
 507             G                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
 508             H                = _mm_setzero_pd();
 509             GMX_MM_TRANSPOSE2_PD(G,H);
 510             Heps             = _mm_mul_pd(vfeps,H);
 511             Fp               = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
 512             VV               = _mm_add_pd(Y,_mm_mul_pd(vfeps,Fp));
 513             vvdw12           = _mm_mul_pd(c12_00,VV);
 514             FF               = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
 515             fvdw12           = _mm_mul_pd(c12_00,FF);
 516             vvdw             = _mm_add_pd(vvdw12,vvdw6);
 517             fvdw             = _mm_xor_pd(signbit,_mm_mul_pd(_mm_add_pd(fvdw6,fvdw12),_mm_mul_pd(vftabscale,rinv00)));
 518
 519             /* Update potential sum for this i atom from the interaction with this j atom. */
 520             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 521             velecsum         = _mm_add_pd(velecsum,velec);
 522             vvdw             = _mm_unpacklo_pd(vvdw,_mm_setzero_pd());
 523             vvdwsum          = _mm_add_pd(vvdwsum,vvdw);
 524
 525             fscal            = _mm_add_pd(felec,fvdw);
 526
 527             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 528
 529             /* Calculate temporary vectorial force */
 530             tx               = _mm_mul_pd(fscal,dx00);
 531             ty               = _mm_mul_pd(fscal,dy00);
 532             tz               = _mm_mul_pd(fscal,dz00);
 533
 534             /* Update vectorial force */
 535             fix0             = _mm_add_pd(fix0,tx);
 536             fiy0             = _mm_add_pd(fiy0,ty);
 537             fiz0             = _mm_add_pd(fiz0,tz);
 538
 539             fjx0             = _mm_add_pd(fjx0,tx);
 540             fjy0             = _mm_add_pd(fjy0,ty);
 541             fjz0             = _mm_add_pd(fjz0,tz);
 542
 543             /**************************
 544              * CALCULATE INTERACTIONS *
 545              **************************/
 546
 547             r10              = _mm_mul_pd(rsq10,rinv10);
 548
 549             /* Compute parameters for interactions between i and j atoms */
 550             qq10             = _mm_mul_pd(iq1,jq0);
 551
 552             /* EWALD ELECTROSTATICS */
 553
 554             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 555             ewrt             = _mm_mul_pd(r10,ewtabscale);
 556             ewitab           = _mm_cvttpd_epi32(ewrt);
 557             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 558             ewitab           = _mm_slli_epi32(ewitab,2);
 559             ewtabF           = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 560             ewtabD           = _mm_setzero_pd();
 561             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 562             ewtabV           = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,0) +2);
 563             ewtabFn          = _mm_setzero_pd();
 564             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 565             felec            = _mm_add_pd(ewtabF,_mm_mul_pd(eweps,ewtabD));
 566             velec            = _mm_sub_pd(ewtabV,_mm_mul_pd(_mm_mul_pd(ewtabhalfspace,eweps),_mm_add_pd(ewtabF,felec)));
 567             velec            = _mm_mul_pd(qq10,_mm_sub_pd(rinv10,velec));
 568             felec            = _mm_mul_pd(_mm_mul_pd(qq10,rinv10),_mm_sub_pd(rinvsq10,felec));
 569
 570             /* Update potential sum for this i atom from the interaction with this j atom. */
 571             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 572             velecsum         = _mm_add_pd(velecsum,velec);
 573
 574             fscal            = felec;
 575
 576             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 577
 578             /* Calculate temporary vectorial force */
 579             tx               = _mm_mul_pd(fscal,dx10);
 580             ty               = _mm_mul_pd(fscal,dy10);
 581             tz               = _mm_mul_pd(fscal,dz10);
 582
 583             /* Update vectorial force */
 584             fix1             = _mm_add_pd(fix1,tx);
 585             fiy1             = _mm_add_pd(fiy1,ty);
 586             fiz1             = _mm_add_pd(fiz1,tz);
 587
 588             fjx0             = _mm_add_pd(fjx0,tx);
 589             fjy0             = _mm_add_pd(fjy0,ty);
 590             fjz0             = _mm_add_pd(fjz0,tz);
 591
 592             /**************************
 593              * CALCULATE INTERACTIONS *
 594              **************************/
 595
 596             r20              = _mm_mul_pd(rsq20,rinv20);
 597
 598             /* Compute parameters for interactions between i and j atoms */
 599             qq20             = _mm_mul_pd(iq2,jq0);
 600
 601             /* EWALD ELECTROSTATICS */
 602
 603             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 604             ewrt             = _mm_mul_pd(r20,ewtabscale);
 605             ewitab           = _mm_cvttpd_epi32(ewrt);
 606             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 607             ewitab           = _mm_slli_epi32(ewitab,2);
 608             ewtabF           = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 609             ewtabD           = _mm_setzero_pd();
 610             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 611             ewtabV           = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,0) +2);
 612             ewtabFn          = _mm_setzero_pd();
 613             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 614             felec            = _mm_add_pd(ewtabF,_mm_mul_pd(eweps,ewtabD));
 615             velec            = _mm_sub_pd(ewtabV,_mm_mul_pd(_mm_mul_pd(ewtabhalfspace,eweps),_mm_add_pd(ewtabF,felec)));
 616             velec            = _mm_mul_pd(qq20,_mm_sub_pd(rinv20,velec));
 617             felec            = _mm_mul_pd(_mm_mul_pd(qq20,rinv20),_mm_sub_pd(rinvsq20,felec));
 618
 619             /* Update potential sum for this i atom from the interaction with this j atom. */
 620             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 621             velecsum         = _mm_add_pd(velecsum,velec);
 622
 623             fscal            = felec;
 624
 625             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 626
 627             /* Calculate temporary vectorial force */
 628             tx               = _mm_mul_pd(fscal,dx20);
 629             ty               = _mm_mul_pd(fscal,dy20);
 630             tz               = _mm_mul_pd(fscal,dz20);
 631
 632             /* Update vectorial force */
 633             fix2             = _mm_add_pd(fix2,tx);
 634             fiy2             = _mm_add_pd(fiy2,ty);
 635             fiz2             = _mm_add_pd(fiz2,tz);
 636
 637             fjx0             = _mm_add_pd(fjx0,tx);
 638             fjy0             = _mm_add_pd(fjy0,ty);
 639             fjz0             = _mm_add_pd(fjz0,tz);
 640
 641             gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,fjx0,fjy0,fjz0);
 642
 643             /* Inner loop uses 160 flops */
 644         }
 645
 646         /* End of innermost loop */
 647
 648         gmx_mm_update_iforce_3atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
 649                                               f+i_coord_offset,fshift+i_shift_offset);
 650
 651         ggid                        = gid[iidx];
 652         /* Update potential energies */
 653         gmx_mm_update_1pot_pd(velecsum,kernel_data->energygrp_elec+ggid);
 654         gmx_mm_update_1pot_pd(vvdwsum,kernel_data->energygrp_vdw+ggid);
 655
 656         /* Increment number of inner iterations */
 657         inneriter                  += j_index_end - j_index_start;
 658
 659         /* Outer loop uses 20 flops */
 660     }
 661
 662     /* Increment number of outer iterations */
 663     outeriter        += nri;
 664
 665     /* Update outer/inner flops */
 666
 667     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3_VF,outeriter*20 + inneriter*160);
 668 }
 669 /*
 670  * Gromacs nonbonded kernel:   nb_kernel_ElecEw_VdwCSTab_GeomW3P1_F_sse4_1_double
 671  * Electrostatics interaction: Ewald
 672  * VdW interaction:            CubicSplineTable
 673  * Geometry:                   Water3-Particle
 674  * Calculate force/pot:        Force
 675  */
 676 void
 677 nb_kernel_ElecEw_VdwCSTab_GeomW3P1_F_sse4_1_double
 678                     (t_nblist                    * gmx_restrict       nlist,
 679                      rvec                        * gmx_restrict          xx,
 680                      rvec                        * gmx_restrict          ff,
 681                      t_forcerec                  * gmx_restrict          fr,
 682                      t_mdatoms                   * gmx_restrict     mdatoms,
 683                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 684                      t_nrnb                      * gmx_restrict        nrnb)
 685 {
 686     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 687      * just 0 for non-waters.
 688      * Suffixes A,B refer to j loop unrolling done with SSE double precision, e.g. for the two different
 689      * jnr indices corresponding to data put in the four positions in the SIMD register.
 690      */
 691     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 692     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 693     int              jnrA,jnrB;
 694     int              j_coord_offsetA,j_coord_offsetB;
 695     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 696     real             rcutoff_scalar;
 697     real             *shiftvec,*fshift,*x,*f;
 698     __m128d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 699     int              vdwioffset0;
 700     __m128d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 701     int              vdwioffset1;
 702     __m128d          ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
 703     int              vdwioffset2;
 704     __m128d          ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
 705     int              vdwjidx0A,vdwjidx0B;
 706     __m128d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 707     __m128d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 708     __m128d          dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
 709     __m128d          dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
 710     __m128d          velec,felec,velecsum,facel,crf,krf,krf2;
 711     real             *charge;
 712     int              nvdwtype;
 713     __m128d          rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 714     int              *vdwtype;
 715     real             *vdwparam;
 716     __m128d          one_sixth   = _mm_set1_pd(1.0/6.0);
 717     __m128d          one_twelfth = _mm_set1_pd(1.0/12.0);
 718     __m128i          vfitab;
 719     __m128i          ifour       = _mm_set1_epi32(4);
 720     __m128d          rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF;
 721     real             *vftab;
 722     __m128i          ewitab;
 723     __m128d          ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
 724     real             *ewtab;
 725     __m128d          dummy_mask,cutoff_mask;
 726     __m128d          signbit   = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
 727     __m128d          one     = _mm_set1_pd(1.0);
 728     __m128d          two     = _mm_set1_pd(2.0);
 729     x                = xx[0];
 730     f                = ff[0];
 731
 732     nri              = nlist->nri;
 733     iinr             = nlist->iinr;
 734     jindex           = nlist->jindex;
 735     jjnr             = nlist->jjnr;
 736     shiftidx         = nlist->shift;
 737     gid              = nlist->gid;
 738     shiftvec         = fr->shift_vec[0];
 739     fshift           = fr->fshift[0];
 740     facel            = _mm_set1_pd(fr->epsfac);
 741     charge           = mdatoms->chargeA;
 742     nvdwtype         = fr->ntype;
 743     vdwparam         = fr->nbfp;
 744     vdwtype          = mdatoms->typeA;
 745
 746     vftab            = kernel_data->table_vdw->data;
 747     vftabscale       = _mm_set1_pd(kernel_data->table_vdw->scale);
 748
 749     sh_ewald         = _mm_set1_pd(fr->ic->sh_ewald);
 750     ewtab            = fr->ic->tabq_coul_F;
 751     ewtabscale       = _mm_set1_pd(fr->ic->tabq_scale);
 752     ewtabhalfspace   = _mm_set1_pd(0.5/fr->ic->tabq_scale);
 753
 754     /* Setup water-specific parameters */
 755     inr              = nlist->iinr[0];
 756     iq0              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+0]));
 757     iq1              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+1]));
 758     iq2              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+2]));
 759     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 760
 761     /* Avoid stupid compiler warnings */
 762     jnrA = jnrB = 0;
 763     j_coord_offsetA = 0;
 764     j_coord_offsetB = 0;
 765
 766     outeriter        = 0;
 767     inneriter        = 0;
 768
 769     /* Start outer loop over neighborlists */
 770     for(iidx=0; iidx<nri; iidx++)
 771     {
 772         /* Load shift vector for this list */
 773         i_shift_offset   = DIM*shiftidx[iidx];
 774
 775         /* Load limits for loop over neighbors */
 776         j_index_start    = jindex[iidx];
 777         j_index_end      = jindex[iidx+1];
 778
 779         /* Get outer coordinate index */
 780         inr              = iinr[iidx];
 781         i_coord_offset   = DIM*inr;
 782
 783         /* Load i particle coords and add shift vector */
 784         gmx_mm_load_shift_and_3rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
 785                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 786
 787         fix0             = _mm_setzero_pd();
 788         fiy0             = _mm_setzero_pd();
 789         fiz0             = _mm_setzero_pd();
 790         fix1             = _mm_setzero_pd();
 791         fiy1             = _mm_setzero_pd();
 792         fiz1             = _mm_setzero_pd();
 793         fix2             = _mm_setzero_pd();
 794         fiy2             = _mm_setzero_pd();
 795         fiz2             = _mm_setzero_pd();
 796
 797         /* Start inner kernel loop */
 798         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 799         {
 800
 801             /* Get j neighbor index, and coordinate index */
 802             jnrA             = jjnr[jidx];
 803             jnrB             = jjnr[jidx+1];
 804             j_coord_offsetA  = DIM*jnrA;
 805             j_coord_offsetB  = DIM*jnrB;
 806
 807             /* load j atom coordinates */
 808             gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 809                                               &jx0,&jy0,&jz0);
 810
 811             /* Calculate displacement vector */
 812             dx00             = _mm_sub_pd(ix0,jx0);
 813             dy00             = _mm_sub_pd(iy0,jy0);
 814             dz00             = _mm_sub_pd(iz0,jz0);
 815             dx10             = _mm_sub_pd(ix1,jx0);
 816             dy10             = _mm_sub_pd(iy1,jy0);
 817             dz10             = _mm_sub_pd(iz1,jz0);
 818             dx20             = _mm_sub_pd(ix2,jx0);
 819             dy20             = _mm_sub_pd(iy2,jy0);
 820             dz20             = _mm_sub_pd(iz2,jz0);
 821
 822             /* Calculate squared distance and things based on it */
 823             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 824             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
 825             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
 826
 827             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 828             rinv10           = gmx_mm_invsqrt_pd(rsq10);
 829             rinv20           = gmx_mm_invsqrt_pd(rsq20);
 830
 831             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 832             rinvsq10         = _mm_mul_pd(rinv10,rinv10);
 833             rinvsq20         = _mm_mul_pd(rinv20,rinv20);
 834
 835             /* Load parameters for j particles */
 836             jq0              = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
 837             vdwjidx0A        = 2*vdwtype[jnrA+0];
 838             vdwjidx0B        = 2*vdwtype[jnrB+0];
 839
 840             fjx0             = _mm_setzero_pd();
 841             fjy0             = _mm_setzero_pd();
 842             fjz0             = _mm_setzero_pd();
 843
 844             /**************************
 845              * CALCULATE INTERACTIONS *
 846              **************************/
 847
 848             r00              = _mm_mul_pd(rsq00,rinv00);
 849
 850             /* Compute parameters for interactions between i and j atoms */
 851             qq00             = _mm_mul_pd(iq0,jq0);
 852             gmx_mm_load_2pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,
 853                                          vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
 854
 855             /* Calculate table index by multiplying r with table scale and truncate to integer */
 856             rt               = _mm_mul_pd(r00,vftabscale);
 857             vfitab           = _mm_cvttpd_epi32(rt);
 858             vfeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
 859             vfitab           = _mm_slli_epi32(vfitab,3);
 860
 861             /* EWALD ELECTROSTATICS */
 862
 863             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 864             ewrt             = _mm_mul_pd(r00,ewtabscale);
 865             ewitab           = _mm_cvttpd_epi32(ewrt);
 866             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 867             gmx_mm_load_2pair_swizzle_pd(ewtab+gmx_mm_extract_epi32(ewitab,0),ewtab+gmx_mm_extract_epi32(ewitab,1),
 868                                          &ewtabF,&ewtabFn);
 869             felec            = _mm_add_pd(_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF),_mm_mul_pd(eweps,ewtabFn));
 870             felec            = _mm_mul_pd(_mm_mul_pd(qq00,rinv00),_mm_sub_pd(rinvsq00,felec));
 871
 872             /* CUBIC SPLINE TABLE DISPERSION */
 873             Y                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
 874             F                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) );
 875             GMX_MM_TRANSPOSE2_PD(Y,F);
 876             G                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
 877             H                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) +2);
 878             GMX_MM_TRANSPOSE2_PD(G,H);
 879             Heps             = _mm_mul_pd(vfeps,H);
 880             Fp               = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
 881             FF               = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
 882             fvdw6            = _mm_mul_pd(c6_00,FF);
 883
 884             /* CUBIC SPLINE TABLE REPULSION */
 885             vfitab           = _mm_add_epi32(vfitab,ifour);
 886             Y                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
 887             F                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) );
 888             GMX_MM_TRANSPOSE2_PD(Y,F);
 889             G                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
 890             H                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) +2);
 891             GMX_MM_TRANSPOSE2_PD(G,H);
 892             Heps             = _mm_mul_pd(vfeps,H);
 893             Fp               = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
 894             FF               = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
 895             fvdw12           = _mm_mul_pd(c12_00,FF);
 896             fvdw             = _mm_xor_pd(signbit,_mm_mul_pd(_mm_add_pd(fvdw6,fvdw12),_mm_mul_pd(vftabscale,rinv00)));
 897
 898             fscal            = _mm_add_pd(felec,fvdw);
 899
 900             /* Calculate temporary vectorial force */
 901             tx               = _mm_mul_pd(fscal,dx00);
 902             ty               = _mm_mul_pd(fscal,dy00);
 903             tz               = _mm_mul_pd(fscal,dz00);
 904
 905             /* Update vectorial force */
 906             fix0             = _mm_add_pd(fix0,tx);
 907             fiy0             = _mm_add_pd(fiy0,ty);
 908             fiz0             = _mm_add_pd(fiz0,tz);
 909
 910             fjx0             = _mm_add_pd(fjx0,tx);
 911             fjy0             = _mm_add_pd(fjy0,ty);
 912             fjz0             = _mm_add_pd(fjz0,tz);
 913
 914             /**************************
 915              * CALCULATE INTERACTIONS *
 916              **************************/
 917
 918             r10              = _mm_mul_pd(rsq10,rinv10);
 919
 920             /* Compute parameters for interactions between i and j atoms */
 921             qq10             = _mm_mul_pd(iq1,jq0);
 922
 923             /* EWALD ELECTROSTATICS */
 924
 925             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 926             ewrt             = _mm_mul_pd(r10,ewtabscale);
 927             ewitab           = _mm_cvttpd_epi32(ewrt);
 928             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 929             gmx_mm_load_2pair_swizzle_pd(ewtab+gmx_mm_extract_epi32(ewitab,0),ewtab+gmx_mm_extract_epi32(ewitab,1),
 930                                          &ewtabF,&ewtabFn);
 931             felec            = _mm_add_pd(_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF),_mm_mul_pd(eweps,ewtabFn));
 932             felec            = _mm_mul_pd(_mm_mul_pd(qq10,rinv10),_mm_sub_pd(rinvsq10,felec));
 933
 934             fscal            = felec;
 935
 936             /* Calculate temporary vectorial force */
 937             tx               = _mm_mul_pd(fscal,dx10);
 938             ty               = _mm_mul_pd(fscal,dy10);
 939             tz               = _mm_mul_pd(fscal,dz10);
 940
 941             /* Update vectorial force */
 942             fix1             = _mm_add_pd(fix1,tx);
 943             fiy1             = _mm_add_pd(fiy1,ty);
 944             fiz1             = _mm_add_pd(fiz1,tz);
 945
 946             fjx0             = _mm_add_pd(fjx0,tx);
 947             fjy0             = _mm_add_pd(fjy0,ty);
 948             fjz0             = _mm_add_pd(fjz0,tz);
 949
 950             /**************************
 951              * CALCULATE INTERACTIONS *
 952              **************************/
 953
 954             r20              = _mm_mul_pd(rsq20,rinv20);
 955
 956             /* Compute parameters for interactions between i and j atoms */
 957             qq20             = _mm_mul_pd(iq2,jq0);
 958
 959             /* EWALD ELECTROSTATICS */
 960
 961             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 962             ewrt             = _mm_mul_pd(r20,ewtabscale);
 963             ewitab           = _mm_cvttpd_epi32(ewrt);
 964             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 965             gmx_mm_load_2pair_swizzle_pd(ewtab+gmx_mm_extract_epi32(ewitab,0),ewtab+gmx_mm_extract_epi32(ewitab,1),
 966                                          &ewtabF,&ewtabFn);
 967             felec            = _mm_add_pd(_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF),_mm_mul_pd(eweps,ewtabFn));
 968             felec            = _mm_mul_pd(_mm_mul_pd(qq20,rinv20),_mm_sub_pd(rinvsq20,felec));
 969
 970             fscal            = felec;
 971
 972             /* Calculate temporary vectorial force */
 973             tx               = _mm_mul_pd(fscal,dx20);
 974             ty               = _mm_mul_pd(fscal,dy20);
 975             tz               = _mm_mul_pd(fscal,dz20);
 976
 977             /* Update vectorial force */
 978             fix2             = _mm_add_pd(fix2,tx);
 979             fiy2             = _mm_add_pd(fiy2,ty);
 980             fiz2             = _mm_add_pd(fiz2,tz);
 981
 982             fjx0             = _mm_add_pd(fjx0,tx);
 983             fjy0             = _mm_add_pd(fjy0,ty);
 984             fjz0             = _mm_add_pd(fjz0,tz);
 985
 986             gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,fjx0,fjy0,fjz0);
 987
 988             /* Inner loop uses 137 flops */
 989         }
 990
 991         if(jidx<j_index_end)
 992         {
 993
 994             jnrA             = jjnr[jidx];
 995             j_coord_offsetA  = DIM*jnrA;
 996
 997             /* load j atom coordinates */
 998             gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
 999                                               &jx0,&jy0,&jz0);
1000
1001             /* Calculate displacement vector */
1002             dx00             = _mm_sub_pd(ix0,jx0);
1003             dy00             = _mm_sub_pd(iy0,jy0);
1004             dz00             = _mm_sub_pd(iz0,jz0);
1005             dx10             = _mm_sub_pd(ix1,jx0);
1006             dy10             = _mm_sub_pd(iy1,jy0);
1007             dz10             = _mm_sub_pd(iz1,jz0);
1008             dx20             = _mm_sub_pd(ix2,jx0);
1009             dy20             = _mm_sub_pd(iy2,jy0);
1010             dz20             = _mm_sub_pd(iz2,jz0);
1011
1012             /* Calculate squared distance and things based on it */
1013             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
1014             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
1015             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
1016
1017             rinv00           = gmx_mm_invsqrt_pd(rsq00);
1018             rinv10           = gmx_mm_invsqrt_pd(rsq10);
1019             rinv20           = gmx_mm_invsqrt_pd(rsq20);
1020
1021             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
1022             rinvsq10         = _mm_mul_pd(rinv10,rinv10);
1023             rinvsq20         = _mm_mul_pd(rinv20,rinv20);
1024
1025             /* Load parameters for j particles */
1026             jq0              = _mm_load_sd(charge+jnrA+0);
1027             vdwjidx0A        = 2*vdwtype[jnrA+0];
1028
1029             fjx0             = _mm_setzero_pd();
1030             fjy0             = _mm_setzero_pd();
1031             fjz0             = _mm_setzero_pd();
1032
1033             /**************************
1034              * CALCULATE INTERACTIONS *
1035              **************************/
1036
1037             r00              = _mm_mul_pd(rsq00,rinv00);
1038
1039             /* Compute parameters for interactions between i and j atoms */
1040             qq00             = _mm_mul_pd(iq0,jq0);
1041             gmx_mm_load_1pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,&c6_00,&c12_00);
1042
1043             /* Calculate table index by multiplying r with table scale and truncate to integer */
1044             rt               = _mm_mul_pd(r00,vftabscale);
1045             vfitab           = _mm_cvttpd_epi32(rt);
1046             vfeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
1047             vfitab           = _mm_slli_epi32(vfitab,3);
1048
1049             /* EWALD ELECTROSTATICS */
1050
1051             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1052             ewrt             = _mm_mul_pd(r00,ewtabscale);
1053             ewitab           = _mm_cvttpd_epi32(ewrt);
1054             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
1055             gmx_mm_load_1pair_swizzle_pd(ewtab+gmx_mm_extract_epi32(ewitab,0),&ewtabF,&ewtabFn);
1056             felec            = _mm_add_pd(_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF),_mm_mul_pd(eweps,ewtabFn));
1057             felec            = _mm_mul_pd(_mm_mul_pd(qq00,rinv00),_mm_sub_pd(rinvsq00,felec));
1058
1059             /* CUBIC SPLINE TABLE DISPERSION */
1060             Y                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
1061             F                = _mm_setzero_pd();
1062             GMX_MM_TRANSPOSE2_PD(Y,F);
1063             G                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
1064             H                = _mm_setzero_pd();
1065             GMX_MM_TRANSPOSE2_PD(G,H);
1066             Heps             = _mm_mul_pd(vfeps,H);
1067             Fp               = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
1068             FF               = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
1069             fvdw6            = _mm_mul_pd(c6_00,FF);
1070
1071             /* CUBIC SPLINE TABLE REPULSION */
1072             vfitab           = _mm_add_epi32(vfitab,ifour);
1073             Y                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
1074             F                = _mm_setzero_pd();
1075             GMX_MM_TRANSPOSE2_PD(Y,F);
1076             G                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
1077             H                = _mm_setzero_pd();
1078             GMX_MM_TRANSPOSE2_PD(G,H);
1079             Heps             = _mm_mul_pd(vfeps,H);
1080             Fp               = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
1081             FF               = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
1082             fvdw12           = _mm_mul_pd(c12_00,FF);
1083             fvdw             = _mm_xor_pd(signbit,_mm_mul_pd(_mm_add_pd(fvdw6,fvdw12),_mm_mul_pd(vftabscale,rinv00)));
1084
1085             fscal            = _mm_add_pd(felec,fvdw);
1086
1087             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1088
1089             /* Calculate temporary vectorial force */
1090             tx               = _mm_mul_pd(fscal,dx00);
1091             ty               = _mm_mul_pd(fscal,dy00);
1092             tz               = _mm_mul_pd(fscal,dz00);
1093
1094             /* Update vectorial force */
1095             fix0             = _mm_add_pd(fix0,tx);
1096             fiy0             = _mm_add_pd(fiy0,ty);
1097             fiz0             = _mm_add_pd(fiz0,tz);
1098
1099             fjx0             = _mm_add_pd(fjx0,tx);
1100             fjy0             = _mm_add_pd(fjy0,ty);
1101             fjz0             = _mm_add_pd(fjz0,tz);
1102
1103             /**************************
1104              * CALCULATE INTERACTIONS *
1105              **************************/
1106
1107             r10              = _mm_mul_pd(rsq10,rinv10);
1108
1109             /* Compute parameters for interactions between i and j atoms */
1110             qq10             = _mm_mul_pd(iq1,jq0);
1111
1112             /* EWALD ELECTROSTATICS */
1113
1114             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1115             ewrt             = _mm_mul_pd(r10,ewtabscale);
1116             ewitab           = _mm_cvttpd_epi32(ewrt);
1117             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
1118             gmx_mm_load_1pair_swizzle_pd(ewtab+gmx_mm_extract_epi32(ewitab,0),&ewtabF,&ewtabFn);
1119             felec            = _mm_add_pd(_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF),_mm_mul_pd(eweps,ewtabFn));
1120             felec            = _mm_mul_pd(_mm_mul_pd(qq10,rinv10),_mm_sub_pd(rinvsq10,felec));
1121
1122             fscal            = felec;
1123
1124             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1125
1126             /* Calculate temporary vectorial force */
1127             tx               = _mm_mul_pd(fscal,dx10);
1128             ty               = _mm_mul_pd(fscal,dy10);
1129             tz               = _mm_mul_pd(fscal,dz10);
1130
1131             /* Update vectorial force */
1132             fix1             = _mm_add_pd(fix1,tx);
1133             fiy1             = _mm_add_pd(fiy1,ty);
1134             fiz1             = _mm_add_pd(fiz1,tz);
1135
1136             fjx0             = _mm_add_pd(fjx0,tx);
1137             fjy0             = _mm_add_pd(fjy0,ty);
1138             fjz0             = _mm_add_pd(fjz0,tz);
1139
1140             /**************************
1141              * CALCULATE INTERACTIONS *
1142              **************************/
1143
1144             r20              = _mm_mul_pd(rsq20,rinv20);
1145
1146             /* Compute parameters for interactions between i and j atoms */
1147             qq20             = _mm_mul_pd(iq2,jq0);
1148
1149             /* EWALD ELECTROSTATICS */
1150
1151             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1152             ewrt             = _mm_mul_pd(r20,ewtabscale);
1153             ewitab           = _mm_cvttpd_epi32(ewrt);
1154             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
1155             gmx_mm_load_1pair_swizzle_pd(ewtab+gmx_mm_extract_epi32(ewitab,0),&ewtabF,&ewtabFn);
1156             felec            = _mm_add_pd(_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF),_mm_mul_pd(eweps,ewtabFn));
1157             felec            = _mm_mul_pd(_mm_mul_pd(qq20,rinv20),_mm_sub_pd(rinvsq20,felec));
1158
1159             fscal            = felec;
1160
1161             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1162
1163             /* Calculate temporary vectorial force */
1164             tx               = _mm_mul_pd(fscal,dx20);
1165             ty               = _mm_mul_pd(fscal,dy20);
1166             tz               = _mm_mul_pd(fscal,dz20);
1167
1168             /* Update vectorial force */
1169             fix2             = _mm_add_pd(fix2,tx);
1170             fiy2             = _mm_add_pd(fiy2,ty);
1171             fiz2             = _mm_add_pd(fiz2,tz);
1172
1173             fjx0             = _mm_add_pd(fjx0,tx);
1174             fjy0             = _mm_add_pd(fjy0,ty);
1175             fjz0             = _mm_add_pd(fjz0,tz);
1176
1177             gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,fjx0,fjy0,fjz0);
1178
1179             /* Inner loop uses 137 flops */
1180         }
1181
1182         /* End of innermost loop */
1183
1184         gmx_mm_update_iforce_3atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
1185                                               f+i_coord_offset,fshift+i_shift_offset);
1186
1187         /* Increment number of inner iterations */
1188         inneriter                  += j_index_end - j_index_start;
1189
1190         /* Outer loop uses 18 flops */
1191     }
1192
1193     /* Increment number of outer iterations */
1194     outeriter        += nri;
1195
1196     /* Update outer/inner flops */
1197
1198     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3_F,outeriter*18 + inneriter*137);
1199 }