src/gromacs/gmxlib/nonbonded/nb_kernel_sse4_1_double/nb_kernel_ElecRF_VdwCSTab_GeomW4P1_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_ElecRF_VdwCSTab_GeomW4P1_VF_sse4_1_double
  54  * Electrostatics interaction: ReactionField
  55  * VdW interaction:            CubicSplineTable
  56  * Geometry:                   Water4-Particle
  57  * Calculate force/pot:        PotentialAndForce
  58  */
  59 void
  60 nb_kernel_ElecRF_VdwCSTab_GeomW4P1_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              vdwioffset3;
  89     __m128d          ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
  90     int              vdwjidx0A,vdwjidx0B;
  91     __m128d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  92     __m128d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  93     __m128d          dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
  94     __m128d          dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
  95     __m128d          dx30,dy30,dz30,rsq30,rinv30,rinvsq30,r30,qq30,c6_30,c12_30;
  96     __m128d          velec,felec,velecsum,facel,crf,krf,krf2;
  97     real             *charge;
  98     int              nvdwtype;
  99     __m128d          rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 100     int              *vdwtype;
 101     real             *vdwparam;
 102     __m128d          one_sixth   = _mm_set1_pd(1.0/6.0);
 103     __m128d          one_twelfth = _mm_set1_pd(1.0/12.0);
 104     __m128i          vfitab;
 105     __m128i          ifour       = _mm_set1_epi32(4);
 106     __m128d          rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF;
 107     real             *vftab;
 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     krf              = _mm_set1_pd(fr->ic->k_rf);
 126     krf2             = _mm_set1_pd(fr->ic->k_rf*2.0);
 127     crf              = _mm_set1_pd(fr->ic->c_rf);
 128     nvdwtype         = fr->ntype;
 129     vdwparam         = fr->nbfp;
 130     vdwtype          = mdatoms->typeA;
 131
 132     vftab            = kernel_data->table_vdw->data;
 133     vftabscale       = _mm_set1_pd(kernel_data->table_vdw->scale);
 134
 135     /* Setup water-specific parameters */
 136     inr              = nlist->iinr[0];
 137     iq1              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+1]));
 138     iq2              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+2]));
 139     iq3              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+3]));
 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_4rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
 166                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
 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         fix3             = _mm_setzero_pd();
 178         fiy3             = _mm_setzero_pd();
 179         fiz3             = _mm_setzero_pd();
 180
 181         /* Reset potential sums */
 182         velecsum         = _mm_setzero_pd();
 183         vvdwsum          = _mm_setzero_pd();
 184
 185         /* Start inner kernel loop */
 186         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 187         {
 188
 189             /* Get j neighbor index, and coordinate index */
 190             jnrA             = jjnr[jidx];
 191             jnrB             = jjnr[jidx+1];
 192             j_coord_offsetA  = DIM*jnrA;
 193             j_coord_offsetB  = DIM*jnrB;
 194
 195             /* load j atom coordinates */
 196             gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 197                                               &jx0,&jy0,&jz0);
 198
 199             /* Calculate displacement vector */
 200             dx00             = _mm_sub_pd(ix0,jx0);
 201             dy00             = _mm_sub_pd(iy0,jy0);
 202             dz00             = _mm_sub_pd(iz0,jz0);
 203             dx10             = _mm_sub_pd(ix1,jx0);
 204             dy10             = _mm_sub_pd(iy1,jy0);
 205             dz10             = _mm_sub_pd(iz1,jz0);
 206             dx20             = _mm_sub_pd(ix2,jx0);
 207             dy20             = _mm_sub_pd(iy2,jy0);
 208             dz20             = _mm_sub_pd(iz2,jz0);
 209             dx30             = _mm_sub_pd(ix3,jx0);
 210             dy30             = _mm_sub_pd(iy3,jy0);
 211             dz30             = _mm_sub_pd(iz3,jz0);
 212
 213             /* Calculate squared distance and things based on it */
 214             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 215             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
 216             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
 217             rsq30            = gmx_mm_calc_rsq_pd(dx30,dy30,dz30);
 218
 219             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 220             rinv10           = gmx_mm_invsqrt_pd(rsq10);
 221             rinv20           = gmx_mm_invsqrt_pd(rsq20);
 222             rinv30           = gmx_mm_invsqrt_pd(rsq30);
 223
 224             rinvsq10         = _mm_mul_pd(rinv10,rinv10);
 225             rinvsq20         = _mm_mul_pd(rinv20,rinv20);
 226             rinvsq30         = _mm_mul_pd(rinv30,rinv30);
 227
 228             /* Load parameters for j particles */
 229             jq0              = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
 230             vdwjidx0A        = 2*vdwtype[jnrA+0];
 231             vdwjidx0B        = 2*vdwtype[jnrB+0];
 232
 233             fjx0             = _mm_setzero_pd();
 234             fjy0             = _mm_setzero_pd();
 235             fjz0             = _mm_setzero_pd();
 236
 237             /**************************
 238              * CALCULATE INTERACTIONS *
 239              **************************/
 240
 241             r00              = _mm_mul_pd(rsq00,rinv00);
 242
 243             /* Compute parameters for interactions between i and j atoms */
 244             gmx_mm_load_2pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,
 245                                          vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
 246
 247             /* Calculate table index by multiplying r with table scale and truncate to integer */
 248             rt               = _mm_mul_pd(r00,vftabscale);
 249             vfitab           = _mm_cvttpd_epi32(rt);
 250             vfeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
 251             vfitab           = _mm_slli_epi32(vfitab,3);
 252
 253             /* CUBIC SPLINE TABLE DISPERSION */
 254             Y                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
 255             F                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) );
 256             GMX_MM_TRANSPOSE2_PD(Y,F);
 257             G                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
 258             H                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) +2);
 259             GMX_MM_TRANSPOSE2_PD(G,H);
 260             Heps             = _mm_mul_pd(vfeps,H);
 261             Fp               = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
 262             VV               = _mm_add_pd(Y,_mm_mul_pd(vfeps,Fp));
 263             vvdw6            = _mm_mul_pd(c6_00,VV);
 264             FF               = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
 265             fvdw6            = _mm_mul_pd(c6_00,FF);
 266
 267             /* CUBIC SPLINE TABLE REPULSION */
 268             vfitab           = _mm_add_epi32(vfitab,ifour);
 269             Y                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
 270             F                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) );
 271             GMX_MM_TRANSPOSE2_PD(Y,F);
 272             G                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
 273             H                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) +2);
 274             GMX_MM_TRANSPOSE2_PD(G,H);
 275             Heps             = _mm_mul_pd(vfeps,H);
 276             Fp               = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
 277             VV               = _mm_add_pd(Y,_mm_mul_pd(vfeps,Fp));
 278             vvdw12           = _mm_mul_pd(c12_00,VV);
 279             FF               = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
 280             fvdw12           = _mm_mul_pd(c12_00,FF);
 281             vvdw             = _mm_add_pd(vvdw12,vvdw6);
 282             fvdw             = _mm_xor_pd(signbit,_mm_mul_pd(_mm_add_pd(fvdw6,fvdw12),_mm_mul_pd(vftabscale,rinv00)));
 283
 284             /* Update potential sum for this i atom from the interaction with this j atom. */
 285             vvdwsum          = _mm_add_pd(vvdwsum,vvdw);
 286
 287             fscal            = fvdw;
 288
 289             /* Calculate temporary vectorial force */
 290             tx               = _mm_mul_pd(fscal,dx00);
 291             ty               = _mm_mul_pd(fscal,dy00);
 292             tz               = _mm_mul_pd(fscal,dz00);
 293
 294             /* Update vectorial force */
 295             fix0             = _mm_add_pd(fix0,tx);
 296             fiy0             = _mm_add_pd(fiy0,ty);
 297             fiz0             = _mm_add_pd(fiz0,tz);
 298
 299             fjx0             = _mm_add_pd(fjx0,tx);
 300             fjy0             = _mm_add_pd(fjy0,ty);
 301             fjz0             = _mm_add_pd(fjz0,tz);
 302
 303             /**************************
 304              * CALCULATE INTERACTIONS *
 305              **************************/
 306
 307             /* Compute parameters for interactions between i and j atoms */
 308             qq10             = _mm_mul_pd(iq1,jq0);
 309
 310             /* REACTION-FIELD ELECTROSTATICS */
 311             velec            = _mm_mul_pd(qq10,_mm_sub_pd(_mm_add_pd(rinv10,_mm_mul_pd(krf,rsq10)),crf));
 312             felec            = _mm_mul_pd(qq10,_mm_sub_pd(_mm_mul_pd(rinv10,rinvsq10),krf2));
 313
 314             /* Update potential sum for this i atom from the interaction with this j atom. */
 315             velecsum         = _mm_add_pd(velecsum,velec);
 316
 317             fscal            = felec;
 318
 319             /* Calculate temporary vectorial force */
 320             tx               = _mm_mul_pd(fscal,dx10);
 321             ty               = _mm_mul_pd(fscal,dy10);
 322             tz               = _mm_mul_pd(fscal,dz10);
 323
 324             /* Update vectorial force */
 325             fix1             = _mm_add_pd(fix1,tx);
 326             fiy1             = _mm_add_pd(fiy1,ty);
 327             fiz1             = _mm_add_pd(fiz1,tz);
 328
 329             fjx0             = _mm_add_pd(fjx0,tx);
 330             fjy0             = _mm_add_pd(fjy0,ty);
 331             fjz0             = _mm_add_pd(fjz0,tz);
 332
 333             /**************************
 334              * CALCULATE INTERACTIONS *
 335              **************************/
 336
 337             /* Compute parameters for interactions between i and j atoms */
 338             qq20             = _mm_mul_pd(iq2,jq0);
 339
 340             /* REACTION-FIELD ELECTROSTATICS */
 341             velec            = _mm_mul_pd(qq20,_mm_sub_pd(_mm_add_pd(rinv20,_mm_mul_pd(krf,rsq20)),crf));
 342             felec            = _mm_mul_pd(qq20,_mm_sub_pd(_mm_mul_pd(rinv20,rinvsq20),krf2));
 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,dx20);
 351             ty               = _mm_mul_pd(fscal,dy20);
 352             tz               = _mm_mul_pd(fscal,dz20);
 353
 354             /* Update vectorial force */
 355             fix2             = _mm_add_pd(fix2,tx);
 356             fiy2             = _mm_add_pd(fiy2,ty);
 357             fiz2             = _mm_add_pd(fiz2,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             /* Compute parameters for interactions between i and j atoms */
 368             qq30             = _mm_mul_pd(iq3,jq0);
 369
 370             /* REACTION-FIELD ELECTROSTATICS */
 371             velec            = _mm_mul_pd(qq30,_mm_sub_pd(_mm_add_pd(rinv30,_mm_mul_pd(krf,rsq30)),crf));
 372             felec            = _mm_mul_pd(qq30,_mm_sub_pd(_mm_mul_pd(rinv30,rinvsq30),krf2));
 373
 374             /* Update potential sum for this i atom from the interaction with this j atom. */
 375             velecsum         = _mm_add_pd(velecsum,velec);
 376
 377             fscal            = felec;
 378
 379             /* Calculate temporary vectorial force */
 380             tx               = _mm_mul_pd(fscal,dx30);
 381             ty               = _mm_mul_pd(fscal,dy30);
 382             tz               = _mm_mul_pd(fscal,dz30);
 383
 384             /* Update vectorial force */
 385             fix3             = _mm_add_pd(fix3,tx);
 386             fiy3             = _mm_add_pd(fiy3,ty);
 387             fiz3             = _mm_add_pd(fiz3,tz);
 388
 389             fjx0             = _mm_add_pd(fjx0,tx);
 390             fjy0             = _mm_add_pd(fjy0,ty);
 391             fjz0             = _mm_add_pd(fjz0,tz);
 392
 393             gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,fjx0,fjy0,fjz0);
 394
 395             /* Inner loop uses 155 flops */
 396         }
 397
 398         if(jidx<j_index_end)
 399         {
 400
 401             jnrA             = jjnr[jidx];
 402             j_coord_offsetA  = DIM*jnrA;
 403
 404             /* load j atom coordinates */
 405             gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
 406                                               &jx0,&jy0,&jz0);
 407
 408             /* Calculate displacement vector */
 409             dx00             = _mm_sub_pd(ix0,jx0);
 410             dy00             = _mm_sub_pd(iy0,jy0);
 411             dz00             = _mm_sub_pd(iz0,jz0);
 412             dx10             = _mm_sub_pd(ix1,jx0);
 413             dy10             = _mm_sub_pd(iy1,jy0);
 414             dz10             = _mm_sub_pd(iz1,jz0);
 415             dx20             = _mm_sub_pd(ix2,jx0);
 416             dy20             = _mm_sub_pd(iy2,jy0);
 417             dz20             = _mm_sub_pd(iz2,jz0);
 418             dx30             = _mm_sub_pd(ix3,jx0);
 419             dy30             = _mm_sub_pd(iy3,jy0);
 420             dz30             = _mm_sub_pd(iz3,jz0);
 421
 422             /* Calculate squared distance and things based on it */
 423             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 424             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
 425             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
 426             rsq30            = gmx_mm_calc_rsq_pd(dx30,dy30,dz30);
 427
 428             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 429             rinv10           = gmx_mm_invsqrt_pd(rsq10);
 430             rinv20           = gmx_mm_invsqrt_pd(rsq20);
 431             rinv30           = gmx_mm_invsqrt_pd(rsq30);
 432
 433             rinvsq10         = _mm_mul_pd(rinv10,rinv10);
 434             rinvsq20         = _mm_mul_pd(rinv20,rinv20);
 435             rinvsq30         = _mm_mul_pd(rinv30,rinv30);
 436
 437             /* Load parameters for j particles */
 438             jq0              = _mm_load_sd(charge+jnrA+0);
 439             vdwjidx0A        = 2*vdwtype[jnrA+0];
 440
 441             fjx0             = _mm_setzero_pd();
 442             fjy0             = _mm_setzero_pd();
 443             fjz0             = _mm_setzero_pd();
 444
 445             /**************************
 446              * CALCULATE INTERACTIONS *
 447              **************************/
 448
 449             r00              = _mm_mul_pd(rsq00,rinv00);
 450
 451             /* Compute parameters for interactions between i and j atoms */
 452             gmx_mm_load_1pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,&c6_00,&c12_00);
 453
 454             /* Calculate table index by multiplying r with table scale and truncate to integer */
 455             rt               = _mm_mul_pd(r00,vftabscale);
 456             vfitab           = _mm_cvttpd_epi32(rt);
 457             vfeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
 458             vfitab           = _mm_slli_epi32(vfitab,3);
 459
 460             /* CUBIC SPLINE TABLE DISPERSION */
 461             Y                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
 462             F                = _mm_setzero_pd();
 463             GMX_MM_TRANSPOSE2_PD(Y,F);
 464             G                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
 465             H                = _mm_setzero_pd();
 466             GMX_MM_TRANSPOSE2_PD(G,H);
 467             Heps             = _mm_mul_pd(vfeps,H);
 468             Fp               = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
 469             VV               = _mm_add_pd(Y,_mm_mul_pd(vfeps,Fp));
 470             vvdw6            = _mm_mul_pd(c6_00,VV);
 471             FF               = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
 472             fvdw6            = _mm_mul_pd(c6_00,FF);
 473
 474             /* CUBIC SPLINE TABLE REPULSION */
 475             vfitab           = _mm_add_epi32(vfitab,ifour);
 476             Y                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
 477             F                = _mm_setzero_pd();
 478             GMX_MM_TRANSPOSE2_PD(Y,F);
 479             G                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
 480             H                = _mm_setzero_pd();
 481             GMX_MM_TRANSPOSE2_PD(G,H);
 482             Heps             = _mm_mul_pd(vfeps,H);
 483             Fp               = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
 484             VV               = _mm_add_pd(Y,_mm_mul_pd(vfeps,Fp));
 485             vvdw12           = _mm_mul_pd(c12_00,VV);
 486             FF               = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
 487             fvdw12           = _mm_mul_pd(c12_00,FF);
 488             vvdw             = _mm_add_pd(vvdw12,vvdw6);
 489             fvdw             = _mm_xor_pd(signbit,_mm_mul_pd(_mm_add_pd(fvdw6,fvdw12),_mm_mul_pd(vftabscale,rinv00)));
 490
 491             /* Update potential sum for this i atom from the interaction with this j atom. */
 492             vvdw             = _mm_unpacklo_pd(vvdw,_mm_setzero_pd());
 493             vvdwsum          = _mm_add_pd(vvdwsum,vvdw);
 494
 495             fscal            = fvdw;
 496
 497             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 498
 499             /* Calculate temporary vectorial force */
 500             tx               = _mm_mul_pd(fscal,dx00);
 501             ty               = _mm_mul_pd(fscal,dy00);
 502             tz               = _mm_mul_pd(fscal,dz00);
 503
 504             /* Update vectorial force */
 505             fix0             = _mm_add_pd(fix0,tx);
 506             fiy0             = _mm_add_pd(fiy0,ty);
 507             fiz0             = _mm_add_pd(fiz0,tz);
 508
 509             fjx0             = _mm_add_pd(fjx0,tx);
 510             fjy0             = _mm_add_pd(fjy0,ty);
 511             fjz0             = _mm_add_pd(fjz0,tz);
 512
 513             /**************************
 514              * CALCULATE INTERACTIONS *
 515              **************************/
 516
 517             /* Compute parameters for interactions between i and j atoms */
 518             qq10             = _mm_mul_pd(iq1,jq0);
 519
 520             /* REACTION-FIELD ELECTROSTATICS */
 521             velec            = _mm_mul_pd(qq10,_mm_sub_pd(_mm_add_pd(rinv10,_mm_mul_pd(krf,rsq10)),crf));
 522             felec            = _mm_mul_pd(qq10,_mm_sub_pd(_mm_mul_pd(rinv10,rinvsq10),krf2));
 523
 524             /* Update potential sum for this i atom from the interaction with this j atom. */
 525             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 526             velecsum         = _mm_add_pd(velecsum,velec);
 527
 528             fscal            = felec;
 529
 530             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 531
 532             /* Calculate temporary vectorial force */
 533             tx               = _mm_mul_pd(fscal,dx10);
 534             ty               = _mm_mul_pd(fscal,dy10);
 535             tz               = _mm_mul_pd(fscal,dz10);
 536
 537             /* Update vectorial force */
 538             fix1             = _mm_add_pd(fix1,tx);
 539             fiy1             = _mm_add_pd(fiy1,ty);
 540             fiz1             = _mm_add_pd(fiz1,tz);
 541
 542             fjx0             = _mm_add_pd(fjx0,tx);
 543             fjy0             = _mm_add_pd(fjy0,ty);
 544             fjz0             = _mm_add_pd(fjz0,tz);
 545
 546             /**************************
 547              * CALCULATE INTERACTIONS *
 548              **************************/
 549
 550             /* Compute parameters for interactions between i and j atoms */
 551             qq20             = _mm_mul_pd(iq2,jq0);
 552
 553             /* REACTION-FIELD ELECTROSTATICS */
 554             velec            = _mm_mul_pd(qq20,_mm_sub_pd(_mm_add_pd(rinv20,_mm_mul_pd(krf,rsq20)),crf));
 555             felec            = _mm_mul_pd(qq20,_mm_sub_pd(_mm_mul_pd(rinv20,rinvsq20),krf2));
 556
 557             /* Update potential sum for this i atom from the interaction with this j atom. */
 558             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 559             velecsum         = _mm_add_pd(velecsum,velec);
 560
 561             fscal            = felec;
 562
 563             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 564
 565             /* Calculate temporary vectorial force */
 566             tx               = _mm_mul_pd(fscal,dx20);
 567             ty               = _mm_mul_pd(fscal,dy20);
 568             tz               = _mm_mul_pd(fscal,dz20);
 569
 570             /* Update vectorial force */
 571             fix2             = _mm_add_pd(fix2,tx);
 572             fiy2             = _mm_add_pd(fiy2,ty);
 573             fiz2             = _mm_add_pd(fiz2,tz);
 574
 575             fjx0             = _mm_add_pd(fjx0,tx);
 576             fjy0             = _mm_add_pd(fjy0,ty);
 577             fjz0             = _mm_add_pd(fjz0,tz);
 578
 579             /**************************
 580              * CALCULATE INTERACTIONS *
 581              **************************/
 582
 583             /* Compute parameters for interactions between i and j atoms */
 584             qq30             = _mm_mul_pd(iq3,jq0);
 585
 586             /* REACTION-FIELD ELECTROSTATICS */
 587             velec            = _mm_mul_pd(qq30,_mm_sub_pd(_mm_add_pd(rinv30,_mm_mul_pd(krf,rsq30)),crf));
 588             felec            = _mm_mul_pd(qq30,_mm_sub_pd(_mm_mul_pd(rinv30,rinvsq30),krf2));
 589
 590             /* Update potential sum for this i atom from the interaction with this j atom. */
 591             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 592             velecsum         = _mm_add_pd(velecsum,velec);
 593
 594             fscal            = felec;
 595
 596             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 597
 598             /* Calculate temporary vectorial force */
 599             tx               = _mm_mul_pd(fscal,dx30);
 600             ty               = _mm_mul_pd(fscal,dy30);
 601             tz               = _mm_mul_pd(fscal,dz30);
 602
 603             /* Update vectorial force */
 604             fix3             = _mm_add_pd(fix3,tx);
 605             fiy3             = _mm_add_pd(fiy3,ty);
 606             fiz3             = _mm_add_pd(fiz3,tz);
 607
 608             fjx0             = _mm_add_pd(fjx0,tx);
 609             fjy0             = _mm_add_pd(fjy0,ty);
 610             fjz0             = _mm_add_pd(fjz0,tz);
 611
 612             gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,fjx0,fjy0,fjz0);
 613
 614             /* Inner loop uses 155 flops */
 615         }
 616
 617         /* End of innermost loop */
 618
 619         gmx_mm_update_iforce_4atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
 620                                               f+i_coord_offset,fshift+i_shift_offset);
 621
 622         ggid                        = gid[iidx];
 623         /* Update potential energies */
 624         gmx_mm_update_1pot_pd(velecsum,kernel_data->energygrp_elec+ggid);
 625         gmx_mm_update_1pot_pd(vvdwsum,kernel_data->energygrp_vdw+ggid);
 626
 627         /* Increment number of inner iterations */
 628         inneriter                  += j_index_end - j_index_start;
 629
 630         /* Outer loop uses 26 flops */
 631     }
 632
 633     /* Increment number of outer iterations */
 634     outeriter        += nri;
 635
 636     /* Update outer/inner flops */
 637
 638     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W4_VF,outeriter*26 + inneriter*155);
 639 }
 640 /*
 641  * Gromacs nonbonded kernel:   nb_kernel_ElecRF_VdwCSTab_GeomW4P1_F_sse4_1_double
 642  * Electrostatics interaction: ReactionField
 643  * VdW interaction:            CubicSplineTable
 644  * Geometry:                   Water4-Particle
 645  * Calculate force/pot:        Force
 646  */
 647 void
 648 nb_kernel_ElecRF_VdwCSTab_GeomW4P1_F_sse4_1_double
 649                     (t_nblist                    * gmx_restrict       nlist,
 650                      rvec                        * gmx_restrict          xx,
 651                      rvec                        * gmx_restrict          ff,
 652                      t_forcerec                  * gmx_restrict          fr,
 653                      t_mdatoms                   * gmx_restrict     mdatoms,
 654                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 655                      t_nrnb                      * gmx_restrict        nrnb)
 656 {
 657     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 658      * just 0 for non-waters.
 659      * Suffixes A,B refer to j loop unrolling done with SSE double precision, e.g. for the two different
 660      * jnr indices corresponding to data put in the four positions in the SIMD register.
 661      */
 662     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 663     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 664     int              jnrA,jnrB;
 665     int              j_coord_offsetA,j_coord_offsetB;
 666     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 667     real             rcutoff_scalar;
 668     real             *shiftvec,*fshift,*x,*f;
 669     __m128d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 670     int              vdwioffset0;
 671     __m128d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 672     int              vdwioffset1;
 673     __m128d          ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
 674     int              vdwioffset2;
 675     __m128d          ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
 676     int              vdwioffset3;
 677     __m128d          ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
 678     int              vdwjidx0A,vdwjidx0B;
 679     __m128d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 680     __m128d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 681     __m128d          dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
 682     __m128d          dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
 683     __m128d          dx30,dy30,dz30,rsq30,rinv30,rinvsq30,r30,qq30,c6_30,c12_30;
 684     __m128d          velec,felec,velecsum,facel,crf,krf,krf2;
 685     real             *charge;
 686     int              nvdwtype;
 687     __m128d          rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 688     int              *vdwtype;
 689     real             *vdwparam;
 690     __m128d          one_sixth   = _mm_set1_pd(1.0/6.0);
 691     __m128d          one_twelfth = _mm_set1_pd(1.0/12.0);
 692     __m128i          vfitab;
 693     __m128i          ifour       = _mm_set1_epi32(4);
 694     __m128d          rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF;
 695     real             *vftab;
 696     __m128d          dummy_mask,cutoff_mask;
 697     __m128d          signbit   = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
 698     __m128d          one     = _mm_set1_pd(1.0);
 699     __m128d          two     = _mm_set1_pd(2.0);
 700     x                = xx[0];
 701     f                = ff[0];
 702
 703     nri              = nlist->nri;
 704     iinr             = nlist->iinr;
 705     jindex           = nlist->jindex;
 706     jjnr             = nlist->jjnr;
 707     shiftidx         = nlist->shift;
 708     gid              = nlist->gid;
 709     shiftvec         = fr->shift_vec[0];
 710     fshift           = fr->fshift[0];
 711     facel            = _mm_set1_pd(fr->epsfac);
 712     charge           = mdatoms->chargeA;
 713     krf              = _mm_set1_pd(fr->ic->k_rf);
 714     krf2             = _mm_set1_pd(fr->ic->k_rf*2.0);
 715     crf              = _mm_set1_pd(fr->ic->c_rf);
 716     nvdwtype         = fr->ntype;
 717     vdwparam         = fr->nbfp;
 718     vdwtype          = mdatoms->typeA;
 719
 720     vftab            = kernel_data->table_vdw->data;
 721     vftabscale       = _mm_set1_pd(kernel_data->table_vdw->scale);
 722
 723     /* Setup water-specific parameters */
 724     inr              = nlist->iinr[0];
 725     iq1              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+1]));
 726     iq2              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+2]));
 727     iq3              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+3]));
 728     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 729
 730     /* Avoid stupid compiler warnings */
 731     jnrA = jnrB = 0;
 732     j_coord_offsetA = 0;
 733     j_coord_offsetB = 0;
 734
 735     outeriter        = 0;
 736     inneriter        = 0;
 737
 738     /* Start outer loop over neighborlists */
 739     for(iidx=0; iidx<nri; iidx++)
 740     {
 741         /* Load shift vector for this list */
 742         i_shift_offset   = DIM*shiftidx[iidx];
 743
 744         /* Load limits for loop over neighbors */
 745         j_index_start    = jindex[iidx];
 746         j_index_end      = jindex[iidx+1];
 747
 748         /* Get outer coordinate index */
 749         inr              = iinr[iidx];
 750         i_coord_offset   = DIM*inr;
 751
 752         /* Load i particle coords and add shift vector */
 753         gmx_mm_load_shift_and_4rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
 754                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
 755
 756         fix0             = _mm_setzero_pd();
 757         fiy0             = _mm_setzero_pd();
 758         fiz0             = _mm_setzero_pd();
 759         fix1             = _mm_setzero_pd();
 760         fiy1             = _mm_setzero_pd();
 761         fiz1             = _mm_setzero_pd();
 762         fix2             = _mm_setzero_pd();
 763         fiy2             = _mm_setzero_pd();
 764         fiz2             = _mm_setzero_pd();
 765         fix3             = _mm_setzero_pd();
 766         fiy3             = _mm_setzero_pd();
 767         fiz3             = _mm_setzero_pd();
 768
 769         /* Start inner kernel loop */
 770         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 771         {
 772
 773             /* Get j neighbor index, and coordinate index */
 774             jnrA             = jjnr[jidx];
 775             jnrB             = jjnr[jidx+1];
 776             j_coord_offsetA  = DIM*jnrA;
 777             j_coord_offsetB  = DIM*jnrB;
 778
 779             /* load j atom coordinates */
 780             gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 781                                               &jx0,&jy0,&jz0);
 782
 783             /* Calculate displacement vector */
 784             dx00             = _mm_sub_pd(ix0,jx0);
 785             dy00             = _mm_sub_pd(iy0,jy0);
 786             dz00             = _mm_sub_pd(iz0,jz0);
 787             dx10             = _mm_sub_pd(ix1,jx0);
 788             dy10             = _mm_sub_pd(iy1,jy0);
 789             dz10             = _mm_sub_pd(iz1,jz0);
 790             dx20             = _mm_sub_pd(ix2,jx0);
 791             dy20             = _mm_sub_pd(iy2,jy0);
 792             dz20             = _mm_sub_pd(iz2,jz0);
 793             dx30             = _mm_sub_pd(ix3,jx0);
 794             dy30             = _mm_sub_pd(iy3,jy0);
 795             dz30             = _mm_sub_pd(iz3,jz0);
 796
 797             /* Calculate squared distance and things based on it */
 798             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 799             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
 800             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
 801             rsq30            = gmx_mm_calc_rsq_pd(dx30,dy30,dz30);
 802
 803             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 804             rinv10           = gmx_mm_invsqrt_pd(rsq10);
 805             rinv20           = gmx_mm_invsqrt_pd(rsq20);
 806             rinv30           = gmx_mm_invsqrt_pd(rsq30);
 807
 808             rinvsq10         = _mm_mul_pd(rinv10,rinv10);
 809             rinvsq20         = _mm_mul_pd(rinv20,rinv20);
 810             rinvsq30         = _mm_mul_pd(rinv30,rinv30);
 811
 812             /* Load parameters for j particles */
 813             jq0              = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
 814             vdwjidx0A        = 2*vdwtype[jnrA+0];
 815             vdwjidx0B        = 2*vdwtype[jnrB+0];
 816
 817             fjx0             = _mm_setzero_pd();
 818             fjy0             = _mm_setzero_pd();
 819             fjz0             = _mm_setzero_pd();
 820
 821             /**************************
 822              * CALCULATE INTERACTIONS *
 823              **************************/
 824
 825             r00              = _mm_mul_pd(rsq00,rinv00);
 826
 827             /* Compute parameters for interactions between i and j atoms */
 828             gmx_mm_load_2pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,
 829                                          vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
 830
 831             /* Calculate table index by multiplying r with table scale and truncate to integer */
 832             rt               = _mm_mul_pd(r00,vftabscale);
 833             vfitab           = _mm_cvttpd_epi32(rt);
 834             vfeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
 835             vfitab           = _mm_slli_epi32(vfitab,3);
 836
 837             /* CUBIC SPLINE TABLE DISPERSION */
 838             Y                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
 839             F                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) );
 840             GMX_MM_TRANSPOSE2_PD(Y,F);
 841             G                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
 842             H                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) +2);
 843             GMX_MM_TRANSPOSE2_PD(G,H);
 844             Heps             = _mm_mul_pd(vfeps,H);
 845             Fp               = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
 846             FF               = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
 847             fvdw6            = _mm_mul_pd(c6_00,FF);
 848
 849             /* CUBIC SPLINE TABLE REPULSION */
 850             vfitab           = _mm_add_epi32(vfitab,ifour);
 851             Y                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
 852             F                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) );
 853             GMX_MM_TRANSPOSE2_PD(Y,F);
 854             G                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
 855             H                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) +2);
 856             GMX_MM_TRANSPOSE2_PD(G,H);
 857             Heps             = _mm_mul_pd(vfeps,H);
 858             Fp               = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
 859             FF               = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
 860             fvdw12           = _mm_mul_pd(c12_00,FF);
 861             fvdw             = _mm_xor_pd(signbit,_mm_mul_pd(_mm_add_pd(fvdw6,fvdw12),_mm_mul_pd(vftabscale,rinv00)));
 862
 863             fscal            = fvdw;
 864
 865             /* Calculate temporary vectorial force */
 866             tx               = _mm_mul_pd(fscal,dx00);
 867             ty               = _mm_mul_pd(fscal,dy00);
 868             tz               = _mm_mul_pd(fscal,dz00);
 869
 870             /* Update vectorial force */
 871             fix0             = _mm_add_pd(fix0,tx);
 872             fiy0             = _mm_add_pd(fiy0,ty);
 873             fiz0             = _mm_add_pd(fiz0,tz);
 874
 875             fjx0             = _mm_add_pd(fjx0,tx);
 876             fjy0             = _mm_add_pd(fjy0,ty);
 877             fjz0             = _mm_add_pd(fjz0,tz);
 878
 879             /**************************
 880              * CALCULATE INTERACTIONS *
 881              **************************/
 882
 883             /* Compute parameters for interactions between i and j atoms */
 884             qq10             = _mm_mul_pd(iq1,jq0);
 885
 886             /* REACTION-FIELD ELECTROSTATICS */
 887             felec            = _mm_mul_pd(qq10,_mm_sub_pd(_mm_mul_pd(rinv10,rinvsq10),krf2));
 888
 889             fscal            = felec;
 890
 891             /* Calculate temporary vectorial force */
 892             tx               = _mm_mul_pd(fscal,dx10);
 893             ty               = _mm_mul_pd(fscal,dy10);
 894             tz               = _mm_mul_pd(fscal,dz10);
 895
 896             /* Update vectorial force */
 897             fix1             = _mm_add_pd(fix1,tx);
 898             fiy1             = _mm_add_pd(fiy1,ty);
 899             fiz1             = _mm_add_pd(fiz1,tz);
 900
 901             fjx0             = _mm_add_pd(fjx0,tx);
 902             fjy0             = _mm_add_pd(fjy0,ty);
 903             fjz0             = _mm_add_pd(fjz0,tz);
 904
 905             /**************************
 906              * CALCULATE INTERACTIONS *
 907              **************************/
 908
 909             /* Compute parameters for interactions between i and j atoms */
 910             qq20             = _mm_mul_pd(iq2,jq0);
 911
 912             /* REACTION-FIELD ELECTROSTATICS */
 913             felec            = _mm_mul_pd(qq20,_mm_sub_pd(_mm_mul_pd(rinv20,rinvsq20),krf2));
 914
 915             fscal            = felec;
 916
 917             /* Calculate temporary vectorial force */
 918             tx               = _mm_mul_pd(fscal,dx20);
 919             ty               = _mm_mul_pd(fscal,dy20);
 920             tz               = _mm_mul_pd(fscal,dz20);
 921
 922             /* Update vectorial force */
 923             fix2             = _mm_add_pd(fix2,tx);
 924             fiy2             = _mm_add_pd(fiy2,ty);
 925             fiz2             = _mm_add_pd(fiz2,tz);
 926
 927             fjx0             = _mm_add_pd(fjx0,tx);
 928             fjy0             = _mm_add_pd(fjy0,ty);
 929             fjz0             = _mm_add_pd(fjz0,tz);
 930
 931             /**************************
 932              * CALCULATE INTERACTIONS *
 933              **************************/
 934
 935             /* Compute parameters for interactions between i and j atoms */
 936             qq30             = _mm_mul_pd(iq3,jq0);
 937
 938             /* REACTION-FIELD ELECTROSTATICS */
 939             felec            = _mm_mul_pd(qq30,_mm_sub_pd(_mm_mul_pd(rinv30,rinvsq30),krf2));
 940
 941             fscal            = felec;
 942
 943             /* Calculate temporary vectorial force */
 944             tx               = _mm_mul_pd(fscal,dx30);
 945             ty               = _mm_mul_pd(fscal,dy30);
 946             tz               = _mm_mul_pd(fscal,dz30);
 947
 948             /* Update vectorial force */
 949             fix3             = _mm_add_pd(fix3,tx);
 950             fiy3             = _mm_add_pd(fiy3,ty);
 951             fiz3             = _mm_add_pd(fiz3,tz);
 952
 953             fjx0             = _mm_add_pd(fjx0,tx);
 954             fjy0             = _mm_add_pd(fjy0,ty);
 955             fjz0             = _mm_add_pd(fjz0,tz);
 956
 957             gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,fjx0,fjy0,fjz0);
 958
 959             /* Inner loop uses 132 flops */
 960         }
 961
 962         if(jidx<j_index_end)
 963         {
 964
 965             jnrA             = jjnr[jidx];
 966             j_coord_offsetA  = DIM*jnrA;
 967
 968             /* load j atom coordinates */
 969             gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
 970                                               &jx0,&jy0,&jz0);
 971
 972             /* Calculate displacement vector */
 973             dx00             = _mm_sub_pd(ix0,jx0);
 974             dy00             = _mm_sub_pd(iy0,jy0);
 975             dz00             = _mm_sub_pd(iz0,jz0);
 976             dx10             = _mm_sub_pd(ix1,jx0);
 977             dy10             = _mm_sub_pd(iy1,jy0);
 978             dz10             = _mm_sub_pd(iz1,jz0);
 979             dx20             = _mm_sub_pd(ix2,jx0);
 980             dy20             = _mm_sub_pd(iy2,jy0);
 981             dz20             = _mm_sub_pd(iz2,jz0);
 982             dx30             = _mm_sub_pd(ix3,jx0);
 983             dy30             = _mm_sub_pd(iy3,jy0);
 984             dz30             = _mm_sub_pd(iz3,jz0);
 985
 986             /* Calculate squared distance and things based on it */
 987             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 988             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
 989             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
 990             rsq30            = gmx_mm_calc_rsq_pd(dx30,dy30,dz30);
 991
 992             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 993             rinv10           = gmx_mm_invsqrt_pd(rsq10);
 994             rinv20           = gmx_mm_invsqrt_pd(rsq20);
 995             rinv30           = gmx_mm_invsqrt_pd(rsq30);
 996
 997             rinvsq10         = _mm_mul_pd(rinv10,rinv10);
 998             rinvsq20         = _mm_mul_pd(rinv20,rinv20);
 999             rinvsq30         = _mm_mul_pd(rinv30,rinv30);
1000
1001             /* Load parameters for j particles */
1002             jq0              = _mm_load_sd(charge+jnrA+0);
1003             vdwjidx0A        = 2*vdwtype[jnrA+0];
1004
1005             fjx0             = _mm_setzero_pd();
1006             fjy0             = _mm_setzero_pd();
1007             fjz0             = _mm_setzero_pd();
1008
1009             /**************************
1010              * CALCULATE INTERACTIONS *
1011              **************************/
1012
1013             r00              = _mm_mul_pd(rsq00,rinv00);
1014
1015             /* Compute parameters for interactions between i and j atoms */
1016             gmx_mm_load_1pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,&c6_00,&c12_00);
1017
1018             /* Calculate table index by multiplying r with table scale and truncate to integer */
1019             rt               = _mm_mul_pd(r00,vftabscale);
1020             vfitab           = _mm_cvttpd_epi32(rt);
1021             vfeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
1022             vfitab           = _mm_slli_epi32(vfitab,3);
1023
1024             /* CUBIC SPLINE TABLE DISPERSION */
1025             Y                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
1026             F                = _mm_setzero_pd();
1027             GMX_MM_TRANSPOSE2_PD(Y,F);
1028             G                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
1029             H                = _mm_setzero_pd();
1030             GMX_MM_TRANSPOSE2_PD(G,H);
1031             Heps             = _mm_mul_pd(vfeps,H);
1032             Fp               = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
1033             FF               = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
1034             fvdw6            = _mm_mul_pd(c6_00,FF);
1035
1036             /* CUBIC SPLINE TABLE REPULSION */
1037             vfitab           = _mm_add_epi32(vfitab,ifour);
1038             Y                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
1039             F                = _mm_setzero_pd();
1040             GMX_MM_TRANSPOSE2_PD(Y,F);
1041             G                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
1042             H                = _mm_setzero_pd();
1043             GMX_MM_TRANSPOSE2_PD(G,H);
1044             Heps             = _mm_mul_pd(vfeps,H);
1045             Fp               = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
1046             FF               = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
1047             fvdw12           = _mm_mul_pd(c12_00,FF);
1048             fvdw             = _mm_xor_pd(signbit,_mm_mul_pd(_mm_add_pd(fvdw6,fvdw12),_mm_mul_pd(vftabscale,rinv00)));
1049
1050             fscal            = fvdw;
1051
1052             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1053
1054             /* Calculate temporary vectorial force */
1055             tx               = _mm_mul_pd(fscal,dx00);
1056             ty               = _mm_mul_pd(fscal,dy00);
1057             tz               = _mm_mul_pd(fscal,dz00);
1058
1059             /* Update vectorial force */
1060             fix0             = _mm_add_pd(fix0,tx);
1061             fiy0             = _mm_add_pd(fiy0,ty);
1062             fiz0             = _mm_add_pd(fiz0,tz);
1063
1064             fjx0             = _mm_add_pd(fjx0,tx);
1065             fjy0             = _mm_add_pd(fjy0,ty);
1066             fjz0             = _mm_add_pd(fjz0,tz);
1067
1068             /**************************
1069              * CALCULATE INTERACTIONS *
1070              **************************/
1071
1072             /* Compute parameters for interactions between i and j atoms */
1073             qq10             = _mm_mul_pd(iq1,jq0);
1074
1075             /* REACTION-FIELD ELECTROSTATICS */
1076             felec            = _mm_mul_pd(qq10,_mm_sub_pd(_mm_mul_pd(rinv10,rinvsq10),krf2));
1077
1078             fscal            = felec;
1079
1080             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1081
1082             /* Calculate temporary vectorial force */
1083             tx               = _mm_mul_pd(fscal,dx10);
1084             ty               = _mm_mul_pd(fscal,dy10);
1085             tz               = _mm_mul_pd(fscal,dz10);
1086
1087             /* Update vectorial force */
1088             fix1             = _mm_add_pd(fix1,tx);
1089             fiy1             = _mm_add_pd(fiy1,ty);
1090             fiz1             = _mm_add_pd(fiz1,tz);
1091
1092             fjx0             = _mm_add_pd(fjx0,tx);
1093             fjy0             = _mm_add_pd(fjy0,ty);
1094             fjz0             = _mm_add_pd(fjz0,tz);
1095
1096             /**************************
1097              * CALCULATE INTERACTIONS *
1098              **************************/
1099
1100             /* Compute parameters for interactions between i and j atoms */
1101             qq20             = _mm_mul_pd(iq2,jq0);
1102
1103             /* REACTION-FIELD ELECTROSTATICS */
1104             felec            = _mm_mul_pd(qq20,_mm_sub_pd(_mm_mul_pd(rinv20,rinvsq20),krf2));
1105
1106             fscal            = felec;
1107
1108             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1109
1110             /* Calculate temporary vectorial force */
1111             tx               = _mm_mul_pd(fscal,dx20);
1112             ty               = _mm_mul_pd(fscal,dy20);
1113             tz               = _mm_mul_pd(fscal,dz20);
1114
1115             /* Update vectorial force */
1116             fix2             = _mm_add_pd(fix2,tx);
1117             fiy2             = _mm_add_pd(fiy2,ty);
1118             fiz2             = _mm_add_pd(fiz2,tz);
1119
1120             fjx0             = _mm_add_pd(fjx0,tx);
1121             fjy0             = _mm_add_pd(fjy0,ty);
1122             fjz0             = _mm_add_pd(fjz0,tz);
1123
1124             /**************************
1125              * CALCULATE INTERACTIONS *
1126              **************************/
1127
1128             /* Compute parameters for interactions between i and j atoms */
1129             qq30             = _mm_mul_pd(iq3,jq0);
1130
1131             /* REACTION-FIELD ELECTROSTATICS */
1132             felec            = _mm_mul_pd(qq30,_mm_sub_pd(_mm_mul_pd(rinv30,rinvsq30),krf2));
1133
1134             fscal            = felec;
1135
1136             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1137
1138             /* Calculate temporary vectorial force */
1139             tx               = _mm_mul_pd(fscal,dx30);
1140             ty               = _mm_mul_pd(fscal,dy30);
1141             tz               = _mm_mul_pd(fscal,dz30);
1142
1143             /* Update vectorial force */
1144             fix3             = _mm_add_pd(fix3,tx);
1145             fiy3             = _mm_add_pd(fiy3,ty);
1146             fiz3             = _mm_add_pd(fiz3,tz);
1147
1148             fjx0             = _mm_add_pd(fjx0,tx);
1149             fjy0             = _mm_add_pd(fjy0,ty);
1150             fjz0             = _mm_add_pd(fjz0,tz);
1151
1152             gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,fjx0,fjy0,fjz0);
1153
1154             /* Inner loop uses 132 flops */
1155         }
1156
1157         /* End of innermost loop */
1158
1159         gmx_mm_update_iforce_4atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
1160                                               f+i_coord_offset,fshift+i_shift_offset);
1161
1162         /* Increment number of inner iterations */
1163         inneriter                  += j_index_end - j_index_start;
1164
1165         /* Outer loop uses 24 flops */
1166     }
1167
1168     /* Increment number of outer iterations */
1169     outeriter        += nri;
1170
1171     /* Update outer/inner flops */
1172
1173     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W4_F,outeriter*24 + inneriter*132);
1174 }