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 "config.h"
  39
  40 #include <math.h>
  41
  42 #include "../nb_kernel.h"
  43 #include "types/simple.h"
  44 #include "gromacs/math/vec.h"
  45 #include "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_ElecRF_VdwCSTab_GeomW4P1_VF_sse4_1_double
  52  * Electrostatics interaction: ReactionField
  53  * VdW interaction:            CubicSplineTable
  54  * Geometry:                   Water4-Particle
  55  * Calculate force/pot:        PotentialAndForce
  56  */
  57 void
  58 nb_kernel_ElecRF_VdwCSTab_GeomW4P1_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              vdwioffset3;
  87     __m128d          ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
  88     int              vdwjidx0A,vdwjidx0B;
  89     __m128d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  90     __m128d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  91     __m128d          dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
  92     __m128d          dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
  93     __m128d          dx30,dy30,dz30,rsq30,rinv30,rinvsq30,r30,qq30,c6_30,c12_30;
  94     __m128d          velec,felec,velecsum,facel,crf,krf,krf2;
  95     real             *charge;
  96     int              nvdwtype;
  97     __m128d          rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
  98     int              *vdwtype;
  99     real             *vdwparam;
 100     __m128d          one_sixth   = _mm_set1_pd(1.0/6.0);
 101     __m128d          one_twelfth = _mm_set1_pd(1.0/12.0);
 102     __m128i          vfitab;
 103     __m128i          ifour       = _mm_set1_epi32(4);
 104     __m128d          rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF;
 105     real             *vftab;
 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     krf              = _mm_set1_pd(fr->ic->k_rf);
 124     krf2             = _mm_set1_pd(fr->ic->k_rf*2.0);
 125     crf              = _mm_set1_pd(fr->ic->c_rf);
 126     nvdwtype         = fr->ntype;
 127     vdwparam         = fr->nbfp;
 128     vdwtype          = mdatoms->typeA;
 129
 130     vftab            = kernel_data->table_vdw->data;
 131     vftabscale       = _mm_set1_pd(kernel_data->table_vdw->scale);
 132
 133     /* Setup water-specific parameters */
 134     inr              = nlist->iinr[0];
 135     iq1              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+1]));
 136     iq2              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+2]));
 137     iq3              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+3]));
 138     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 139
 140     /* Avoid stupid compiler warnings */
 141     jnrA = jnrB = 0;
 142     j_coord_offsetA = 0;
 143     j_coord_offsetB = 0;
 144
 145     outeriter        = 0;
 146     inneriter        = 0;
 147
 148     /* Start outer loop over neighborlists */
 149     for(iidx=0; iidx<nri; iidx++)
 150     {
 151         /* Load shift vector for this list */
 152         i_shift_offset   = DIM*shiftidx[iidx];
 153
 154         /* Load limits for loop over neighbors */
 155         j_index_start    = jindex[iidx];
 156         j_index_end      = jindex[iidx+1];
 157
 158         /* Get outer coordinate index */
 159         inr              = iinr[iidx];
 160         i_coord_offset   = DIM*inr;
 161
 162         /* Load i particle coords and add shift vector */
 163         gmx_mm_load_shift_and_4rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
 164                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
 165
 166         fix0             = _mm_setzero_pd();
 167         fiy0             = _mm_setzero_pd();
 168         fiz0             = _mm_setzero_pd();
 169         fix1             = _mm_setzero_pd();
 170         fiy1             = _mm_setzero_pd();
 171         fiz1             = _mm_setzero_pd();
 172         fix2             = _mm_setzero_pd();
 173         fiy2             = _mm_setzero_pd();
 174         fiz2             = _mm_setzero_pd();
 175         fix3             = _mm_setzero_pd();
 176         fiy3             = _mm_setzero_pd();
 177         fiz3             = _mm_setzero_pd();
 178
 179         /* Reset potential sums */
 180         velecsum         = _mm_setzero_pd();
 181         vvdwsum          = _mm_setzero_pd();
 182
 183         /* Start inner kernel loop */
 184         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 185         {
 186
 187             /* Get j neighbor index, and coordinate index */
 188             jnrA             = jjnr[jidx];
 189             jnrB             = jjnr[jidx+1];
 190             j_coord_offsetA  = DIM*jnrA;
 191             j_coord_offsetB  = DIM*jnrB;
 192
 193             /* load j atom coordinates */
 194             gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 195                                               &jx0,&jy0,&jz0);
 196
 197             /* Calculate displacement vector */
 198             dx00             = _mm_sub_pd(ix0,jx0);
 199             dy00             = _mm_sub_pd(iy0,jy0);
 200             dz00             = _mm_sub_pd(iz0,jz0);
 201             dx10             = _mm_sub_pd(ix1,jx0);
 202             dy10             = _mm_sub_pd(iy1,jy0);
 203             dz10             = _mm_sub_pd(iz1,jz0);
 204             dx20             = _mm_sub_pd(ix2,jx0);
 205             dy20             = _mm_sub_pd(iy2,jy0);
 206             dz20             = _mm_sub_pd(iz2,jz0);
 207             dx30             = _mm_sub_pd(ix3,jx0);
 208             dy30             = _mm_sub_pd(iy3,jy0);
 209             dz30             = _mm_sub_pd(iz3,jz0);
 210
 211             /* Calculate squared distance and things based on it */
 212             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 213             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
 214             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
 215             rsq30            = gmx_mm_calc_rsq_pd(dx30,dy30,dz30);
 216
 217             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 218             rinv10           = gmx_mm_invsqrt_pd(rsq10);
 219             rinv20           = gmx_mm_invsqrt_pd(rsq20);
 220             rinv30           = gmx_mm_invsqrt_pd(rsq30);
 221
 222             rinvsq10         = _mm_mul_pd(rinv10,rinv10);
 223             rinvsq20         = _mm_mul_pd(rinv20,rinv20);
 224             rinvsq30         = _mm_mul_pd(rinv30,rinv30);
 225
 226             /* Load parameters for j particles */
 227             jq0              = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
 228             vdwjidx0A        = 2*vdwtype[jnrA+0];
 229             vdwjidx0B        = 2*vdwtype[jnrB+0];
 230
 231             fjx0             = _mm_setzero_pd();
 232             fjy0             = _mm_setzero_pd();
 233             fjz0             = _mm_setzero_pd();
 234
 235             /**************************
 236              * CALCULATE INTERACTIONS *
 237              **************************/
 238
 239             r00              = _mm_mul_pd(rsq00,rinv00);
 240
 241             /* Compute parameters for interactions between i and j atoms */
 242             gmx_mm_load_2pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,
 243                                          vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
 244
 245             /* Calculate table index by multiplying r with table scale and truncate to integer */
 246             rt               = _mm_mul_pd(r00,vftabscale);
 247             vfitab           = _mm_cvttpd_epi32(rt);
 248             vfeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
 249             vfitab           = _mm_slli_epi32(vfitab,3);
 250
 251             /* CUBIC SPLINE TABLE DISPERSION */
 252             Y                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
 253             F                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) );
 254             GMX_MM_TRANSPOSE2_PD(Y,F);
 255             G                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
 256             H                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) +2);
 257             GMX_MM_TRANSPOSE2_PD(G,H);
 258             Heps             = _mm_mul_pd(vfeps,H);
 259             Fp               = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
 260             VV               = _mm_add_pd(Y,_mm_mul_pd(vfeps,Fp));
 261             vvdw6            = _mm_mul_pd(c6_00,VV);
 262             FF               = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
 263             fvdw6            = _mm_mul_pd(c6_00,FF);
 264
 265             /* CUBIC SPLINE TABLE REPULSION */
 266             vfitab           = _mm_add_epi32(vfitab,ifour);
 267             Y                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
 268             F                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) );
 269             GMX_MM_TRANSPOSE2_PD(Y,F);
 270             G                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
 271             H                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) +2);
 272             GMX_MM_TRANSPOSE2_PD(G,H);
 273             Heps             = _mm_mul_pd(vfeps,H);
 274             Fp               = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
 275             VV               = _mm_add_pd(Y,_mm_mul_pd(vfeps,Fp));
 276             vvdw12           = _mm_mul_pd(c12_00,VV);
 277             FF               = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
 278             fvdw12           = _mm_mul_pd(c12_00,FF);
 279             vvdw             = _mm_add_pd(vvdw12,vvdw6);
 280             fvdw             = _mm_xor_pd(signbit,_mm_mul_pd(_mm_add_pd(fvdw6,fvdw12),_mm_mul_pd(vftabscale,rinv00)));
 281
 282             /* Update potential sum for this i atom from the interaction with this j atom. */
 283             vvdwsum          = _mm_add_pd(vvdwsum,vvdw);
 284
 285             fscal            = fvdw;
 286
 287             /* Calculate temporary vectorial force */
 288             tx               = _mm_mul_pd(fscal,dx00);
 289             ty               = _mm_mul_pd(fscal,dy00);
 290             tz               = _mm_mul_pd(fscal,dz00);
 291
 292             /* Update vectorial force */
 293             fix0             = _mm_add_pd(fix0,tx);
 294             fiy0             = _mm_add_pd(fiy0,ty);
 295             fiz0             = _mm_add_pd(fiz0,tz);
 296
 297             fjx0             = _mm_add_pd(fjx0,tx);
 298             fjy0             = _mm_add_pd(fjy0,ty);
 299             fjz0             = _mm_add_pd(fjz0,tz);
 300
 301             /**************************
 302              * CALCULATE INTERACTIONS *
 303              **************************/
 304
 305             /* Compute parameters for interactions between i and j atoms */
 306             qq10             = _mm_mul_pd(iq1,jq0);
 307
 308             /* REACTION-FIELD ELECTROSTATICS */
 309             velec            = _mm_mul_pd(qq10,_mm_sub_pd(_mm_add_pd(rinv10,_mm_mul_pd(krf,rsq10)),crf));
 310             felec            = _mm_mul_pd(qq10,_mm_sub_pd(_mm_mul_pd(rinv10,rinvsq10),krf2));
 311
 312             /* Update potential sum for this i atom from the interaction with this j atom. */
 313             velecsum         = _mm_add_pd(velecsum,velec);
 314
 315             fscal            = felec;
 316
 317             /* Calculate temporary vectorial force */
 318             tx               = _mm_mul_pd(fscal,dx10);
 319             ty               = _mm_mul_pd(fscal,dy10);
 320             tz               = _mm_mul_pd(fscal,dz10);
 321
 322             /* Update vectorial force */
 323             fix1             = _mm_add_pd(fix1,tx);
 324             fiy1             = _mm_add_pd(fiy1,ty);
 325             fiz1             = _mm_add_pd(fiz1,tz);
 326
 327             fjx0             = _mm_add_pd(fjx0,tx);
 328             fjy0             = _mm_add_pd(fjy0,ty);
 329             fjz0             = _mm_add_pd(fjz0,tz);
 330
 331             /**************************
 332              * CALCULATE INTERACTIONS *
 333              **************************/
 334
 335             /* Compute parameters for interactions between i and j atoms */
 336             qq20             = _mm_mul_pd(iq2,jq0);
 337
 338             /* REACTION-FIELD ELECTROSTATICS */
 339             velec            = _mm_mul_pd(qq20,_mm_sub_pd(_mm_add_pd(rinv20,_mm_mul_pd(krf,rsq20)),crf));
 340             felec            = _mm_mul_pd(qq20,_mm_sub_pd(_mm_mul_pd(rinv20,rinvsq20),krf2));
 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,dx20);
 349             ty               = _mm_mul_pd(fscal,dy20);
 350             tz               = _mm_mul_pd(fscal,dz20);
 351
 352             /* Update vectorial force */
 353             fix2             = _mm_add_pd(fix2,tx);
 354             fiy2             = _mm_add_pd(fiy2,ty);
 355             fiz2             = _mm_add_pd(fiz2,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             /* Compute parameters for interactions between i and j atoms */
 366             qq30             = _mm_mul_pd(iq3,jq0);
 367
 368             /* REACTION-FIELD ELECTROSTATICS */
 369             velec            = _mm_mul_pd(qq30,_mm_sub_pd(_mm_add_pd(rinv30,_mm_mul_pd(krf,rsq30)),crf));
 370             felec            = _mm_mul_pd(qq30,_mm_sub_pd(_mm_mul_pd(rinv30,rinvsq30),krf2));
 371
 372             /* Update potential sum for this i atom from the interaction with this j atom. */
 373             velecsum         = _mm_add_pd(velecsum,velec);
 374
 375             fscal            = felec;
 376
 377             /* Calculate temporary vectorial force */
 378             tx               = _mm_mul_pd(fscal,dx30);
 379             ty               = _mm_mul_pd(fscal,dy30);
 380             tz               = _mm_mul_pd(fscal,dz30);
 381
 382             /* Update vectorial force */
 383             fix3             = _mm_add_pd(fix3,tx);
 384             fiy3             = _mm_add_pd(fiy3,ty);
 385             fiz3             = _mm_add_pd(fiz3,tz);
 386
 387             fjx0             = _mm_add_pd(fjx0,tx);
 388             fjy0             = _mm_add_pd(fjy0,ty);
 389             fjz0             = _mm_add_pd(fjz0,tz);
 390
 391             gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,fjx0,fjy0,fjz0);
 392
 393             /* Inner loop uses 155 flops */
 394         }
 395
 396         if(jidx<j_index_end)
 397         {
 398
 399             jnrA             = jjnr[jidx];
 400             j_coord_offsetA  = DIM*jnrA;
 401
 402             /* load j atom coordinates */
 403             gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
 404                                               &jx0,&jy0,&jz0);
 405
 406             /* Calculate displacement vector */
 407             dx00             = _mm_sub_pd(ix0,jx0);
 408             dy00             = _mm_sub_pd(iy0,jy0);
 409             dz00             = _mm_sub_pd(iz0,jz0);
 410             dx10             = _mm_sub_pd(ix1,jx0);
 411             dy10             = _mm_sub_pd(iy1,jy0);
 412             dz10             = _mm_sub_pd(iz1,jz0);
 413             dx20             = _mm_sub_pd(ix2,jx0);
 414             dy20             = _mm_sub_pd(iy2,jy0);
 415             dz20             = _mm_sub_pd(iz2,jz0);
 416             dx30             = _mm_sub_pd(ix3,jx0);
 417             dy30             = _mm_sub_pd(iy3,jy0);
 418             dz30             = _mm_sub_pd(iz3,jz0);
 419
 420             /* Calculate squared distance and things based on it */
 421             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 422             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
 423             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
 424             rsq30            = gmx_mm_calc_rsq_pd(dx30,dy30,dz30);
 425
 426             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 427             rinv10           = gmx_mm_invsqrt_pd(rsq10);
 428             rinv20           = gmx_mm_invsqrt_pd(rsq20);
 429             rinv30           = gmx_mm_invsqrt_pd(rsq30);
 430
 431             rinvsq10         = _mm_mul_pd(rinv10,rinv10);
 432             rinvsq20         = _mm_mul_pd(rinv20,rinv20);
 433             rinvsq30         = _mm_mul_pd(rinv30,rinv30);
 434
 435             /* Load parameters for j particles */
 436             jq0              = _mm_load_sd(charge+jnrA+0);
 437             vdwjidx0A        = 2*vdwtype[jnrA+0];
 438
 439             fjx0             = _mm_setzero_pd();
 440             fjy0             = _mm_setzero_pd();
 441             fjz0             = _mm_setzero_pd();
 442
 443             /**************************
 444              * CALCULATE INTERACTIONS *
 445              **************************/
 446
 447             r00              = _mm_mul_pd(rsq00,rinv00);
 448
 449             /* Compute parameters for interactions between i and j atoms */
 450             gmx_mm_load_1pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,&c6_00,&c12_00);
 451
 452             /* Calculate table index by multiplying r with table scale and truncate to integer */
 453             rt               = _mm_mul_pd(r00,vftabscale);
 454             vfitab           = _mm_cvttpd_epi32(rt);
 455             vfeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
 456             vfitab           = _mm_slli_epi32(vfitab,3);
 457
 458             /* CUBIC SPLINE TABLE DISPERSION */
 459             Y                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
 460             F                = _mm_setzero_pd();
 461             GMX_MM_TRANSPOSE2_PD(Y,F);
 462             G                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
 463             H                = _mm_setzero_pd();
 464             GMX_MM_TRANSPOSE2_PD(G,H);
 465             Heps             = _mm_mul_pd(vfeps,H);
 466             Fp               = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
 467             VV               = _mm_add_pd(Y,_mm_mul_pd(vfeps,Fp));
 468             vvdw6            = _mm_mul_pd(c6_00,VV);
 469             FF               = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
 470             fvdw6            = _mm_mul_pd(c6_00,FF);
 471
 472             /* CUBIC SPLINE TABLE REPULSION */
 473             vfitab           = _mm_add_epi32(vfitab,ifour);
 474             Y                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
 475             F                = _mm_setzero_pd();
 476             GMX_MM_TRANSPOSE2_PD(Y,F);
 477             G                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
 478             H                = _mm_setzero_pd();
 479             GMX_MM_TRANSPOSE2_PD(G,H);
 480             Heps             = _mm_mul_pd(vfeps,H);
 481             Fp               = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
 482             VV               = _mm_add_pd(Y,_mm_mul_pd(vfeps,Fp));
 483             vvdw12           = _mm_mul_pd(c12_00,VV);
 484             FF               = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
 485             fvdw12           = _mm_mul_pd(c12_00,FF);
 486             vvdw             = _mm_add_pd(vvdw12,vvdw6);
 487             fvdw             = _mm_xor_pd(signbit,_mm_mul_pd(_mm_add_pd(fvdw6,fvdw12),_mm_mul_pd(vftabscale,rinv00)));
 488
 489             /* Update potential sum for this i atom from the interaction with this j atom. */
 490             vvdw             = _mm_unpacklo_pd(vvdw,_mm_setzero_pd());
 491             vvdwsum          = _mm_add_pd(vvdwsum,vvdw);
 492
 493             fscal            = fvdw;
 494
 495             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 496
 497             /* Calculate temporary vectorial force */
 498             tx               = _mm_mul_pd(fscal,dx00);
 499             ty               = _mm_mul_pd(fscal,dy00);
 500             tz               = _mm_mul_pd(fscal,dz00);
 501
 502             /* Update vectorial force */
 503             fix0             = _mm_add_pd(fix0,tx);
 504             fiy0             = _mm_add_pd(fiy0,ty);
 505             fiz0             = _mm_add_pd(fiz0,tz);
 506
 507             fjx0             = _mm_add_pd(fjx0,tx);
 508             fjy0             = _mm_add_pd(fjy0,ty);
 509             fjz0             = _mm_add_pd(fjz0,tz);
 510
 511             /**************************
 512              * CALCULATE INTERACTIONS *
 513              **************************/
 514
 515             /* Compute parameters for interactions between i and j atoms */
 516             qq10             = _mm_mul_pd(iq1,jq0);
 517
 518             /* REACTION-FIELD ELECTROSTATICS */
 519             velec            = _mm_mul_pd(qq10,_mm_sub_pd(_mm_add_pd(rinv10,_mm_mul_pd(krf,rsq10)),crf));
 520             felec            = _mm_mul_pd(qq10,_mm_sub_pd(_mm_mul_pd(rinv10,rinvsq10),krf2));
 521
 522             /* Update potential sum for this i atom from the interaction with this j atom. */
 523             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 524             velecsum         = _mm_add_pd(velecsum,velec);
 525
 526             fscal            = felec;
 527
 528             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 529
 530             /* Calculate temporary vectorial force */
 531             tx               = _mm_mul_pd(fscal,dx10);
 532             ty               = _mm_mul_pd(fscal,dy10);
 533             tz               = _mm_mul_pd(fscal,dz10);
 534
 535             /* Update vectorial force */
 536             fix1             = _mm_add_pd(fix1,tx);
 537             fiy1             = _mm_add_pd(fiy1,ty);
 538             fiz1             = _mm_add_pd(fiz1,tz);
 539
 540             fjx0             = _mm_add_pd(fjx0,tx);
 541             fjy0             = _mm_add_pd(fjy0,ty);
 542             fjz0             = _mm_add_pd(fjz0,tz);
 543
 544             /**************************
 545              * CALCULATE INTERACTIONS *
 546              **************************/
 547
 548             /* Compute parameters for interactions between i and j atoms */
 549             qq20             = _mm_mul_pd(iq2,jq0);
 550
 551             /* REACTION-FIELD ELECTROSTATICS */
 552             velec            = _mm_mul_pd(qq20,_mm_sub_pd(_mm_add_pd(rinv20,_mm_mul_pd(krf,rsq20)),crf));
 553             felec            = _mm_mul_pd(qq20,_mm_sub_pd(_mm_mul_pd(rinv20,rinvsq20),krf2));
 554
 555             /* Update potential sum for this i atom from the interaction with this j atom. */
 556             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 557             velecsum         = _mm_add_pd(velecsum,velec);
 558
 559             fscal            = felec;
 560
 561             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 562
 563             /* Calculate temporary vectorial force */
 564             tx               = _mm_mul_pd(fscal,dx20);
 565             ty               = _mm_mul_pd(fscal,dy20);
 566             tz               = _mm_mul_pd(fscal,dz20);
 567
 568             /* Update vectorial force */
 569             fix2             = _mm_add_pd(fix2,tx);
 570             fiy2             = _mm_add_pd(fiy2,ty);
 571             fiz2             = _mm_add_pd(fiz2,tz);
 572
 573             fjx0             = _mm_add_pd(fjx0,tx);
 574             fjy0             = _mm_add_pd(fjy0,ty);
 575             fjz0             = _mm_add_pd(fjz0,tz);
 576
 577             /**************************
 578              * CALCULATE INTERACTIONS *
 579              **************************/
 580
 581             /* Compute parameters for interactions between i and j atoms */
 582             qq30             = _mm_mul_pd(iq3,jq0);
 583
 584             /* REACTION-FIELD ELECTROSTATICS */
 585             velec            = _mm_mul_pd(qq30,_mm_sub_pd(_mm_add_pd(rinv30,_mm_mul_pd(krf,rsq30)),crf));
 586             felec            = _mm_mul_pd(qq30,_mm_sub_pd(_mm_mul_pd(rinv30,rinvsq30),krf2));
 587
 588             /* Update potential sum for this i atom from the interaction with this j atom. */
 589             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 590             velecsum         = _mm_add_pd(velecsum,velec);
 591
 592             fscal            = felec;
 593
 594             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 595
 596             /* Calculate temporary vectorial force */
 597             tx               = _mm_mul_pd(fscal,dx30);
 598             ty               = _mm_mul_pd(fscal,dy30);
 599             tz               = _mm_mul_pd(fscal,dz30);
 600
 601             /* Update vectorial force */
 602             fix3             = _mm_add_pd(fix3,tx);
 603             fiy3             = _mm_add_pd(fiy3,ty);
 604             fiz3             = _mm_add_pd(fiz3,tz);
 605
 606             fjx0             = _mm_add_pd(fjx0,tx);
 607             fjy0             = _mm_add_pd(fjy0,ty);
 608             fjz0             = _mm_add_pd(fjz0,tz);
 609
 610             gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,fjx0,fjy0,fjz0);
 611
 612             /* Inner loop uses 155 flops */
 613         }
 614
 615         /* End of innermost loop */
 616
 617         gmx_mm_update_iforce_4atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
 618                                               f+i_coord_offset,fshift+i_shift_offset);
 619
 620         ggid                        = gid[iidx];
 621         /* Update potential energies */
 622         gmx_mm_update_1pot_pd(velecsum,kernel_data->energygrp_elec+ggid);
 623         gmx_mm_update_1pot_pd(vvdwsum,kernel_data->energygrp_vdw+ggid);
 624
 625         /* Increment number of inner iterations */
 626         inneriter                  += j_index_end - j_index_start;
 627
 628         /* Outer loop uses 26 flops */
 629     }
 630
 631     /* Increment number of outer iterations */
 632     outeriter        += nri;
 633
 634     /* Update outer/inner flops */
 635
 636     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W4_VF,outeriter*26 + inneriter*155);
 637 }
 638 /*
 639  * Gromacs nonbonded kernel:   nb_kernel_ElecRF_VdwCSTab_GeomW4P1_F_sse4_1_double
 640  * Electrostatics interaction: ReactionField
 641  * VdW interaction:            CubicSplineTable
 642  * Geometry:                   Water4-Particle
 643  * Calculate force/pot:        Force
 644  */
 645 void
 646 nb_kernel_ElecRF_VdwCSTab_GeomW4P1_F_sse4_1_double
 647                     (t_nblist                    * gmx_restrict       nlist,
 648                      rvec                        * gmx_restrict          xx,
 649                      rvec                        * gmx_restrict          ff,
 650                      t_forcerec                  * gmx_restrict          fr,
 651                      t_mdatoms                   * gmx_restrict     mdatoms,
 652                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 653                      t_nrnb                      * gmx_restrict        nrnb)
 654 {
 655     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 656      * just 0 for non-waters.
 657      * Suffixes A,B refer to j loop unrolling done with SSE double precision, e.g. for the two different
 658      * jnr indices corresponding to data put in the four positions in the SIMD register.
 659      */
 660     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 661     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 662     int              jnrA,jnrB;
 663     int              j_coord_offsetA,j_coord_offsetB;
 664     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 665     real             rcutoff_scalar;
 666     real             *shiftvec,*fshift,*x,*f;
 667     __m128d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 668     int              vdwioffset0;
 669     __m128d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 670     int              vdwioffset1;
 671     __m128d          ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
 672     int              vdwioffset2;
 673     __m128d          ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
 674     int              vdwioffset3;
 675     __m128d          ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
 676     int              vdwjidx0A,vdwjidx0B;
 677     __m128d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 678     __m128d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 679     __m128d          dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
 680     __m128d          dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
 681     __m128d          dx30,dy30,dz30,rsq30,rinv30,rinvsq30,r30,qq30,c6_30,c12_30;
 682     __m128d          velec,felec,velecsum,facel,crf,krf,krf2;
 683     real             *charge;
 684     int              nvdwtype;
 685     __m128d          rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 686     int              *vdwtype;
 687     real             *vdwparam;
 688     __m128d          one_sixth   = _mm_set1_pd(1.0/6.0);
 689     __m128d          one_twelfth = _mm_set1_pd(1.0/12.0);
 690     __m128i          vfitab;
 691     __m128i          ifour       = _mm_set1_epi32(4);
 692     __m128d          rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF;
 693     real             *vftab;
 694     __m128d          dummy_mask,cutoff_mask;
 695     __m128d          signbit   = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
 696     __m128d          one     = _mm_set1_pd(1.0);
 697     __m128d          two     = _mm_set1_pd(2.0);
 698     x                = xx[0];
 699     f                = ff[0];
 700
 701     nri              = nlist->nri;
 702     iinr             = nlist->iinr;
 703     jindex           = nlist->jindex;
 704     jjnr             = nlist->jjnr;
 705     shiftidx         = nlist->shift;
 706     gid              = nlist->gid;
 707     shiftvec         = fr->shift_vec[0];
 708     fshift           = fr->fshift[0];
 709     facel            = _mm_set1_pd(fr->epsfac);
 710     charge           = mdatoms->chargeA;
 711     krf              = _mm_set1_pd(fr->ic->k_rf);
 712     krf2             = _mm_set1_pd(fr->ic->k_rf*2.0);
 713     crf              = _mm_set1_pd(fr->ic->c_rf);
 714     nvdwtype         = fr->ntype;
 715     vdwparam         = fr->nbfp;
 716     vdwtype          = mdatoms->typeA;
 717
 718     vftab            = kernel_data->table_vdw->data;
 719     vftabscale       = _mm_set1_pd(kernel_data->table_vdw->scale);
 720
 721     /* Setup water-specific parameters */
 722     inr              = nlist->iinr[0];
 723     iq1              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+1]));
 724     iq2              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+2]));
 725     iq3              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+3]));
 726     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 727
 728     /* Avoid stupid compiler warnings */
 729     jnrA = jnrB = 0;
 730     j_coord_offsetA = 0;
 731     j_coord_offsetB = 0;
 732
 733     outeriter        = 0;
 734     inneriter        = 0;
 735
 736     /* Start outer loop over neighborlists */
 737     for(iidx=0; iidx<nri; iidx++)
 738     {
 739         /* Load shift vector for this list */
 740         i_shift_offset   = DIM*shiftidx[iidx];
 741
 742         /* Load limits for loop over neighbors */
 743         j_index_start    = jindex[iidx];
 744         j_index_end      = jindex[iidx+1];
 745
 746         /* Get outer coordinate index */
 747         inr              = iinr[iidx];
 748         i_coord_offset   = DIM*inr;
 749
 750         /* Load i particle coords and add shift vector */
 751         gmx_mm_load_shift_and_4rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
 752                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
 753
 754         fix0             = _mm_setzero_pd();
 755         fiy0             = _mm_setzero_pd();
 756         fiz0             = _mm_setzero_pd();
 757         fix1             = _mm_setzero_pd();
 758         fiy1             = _mm_setzero_pd();
 759         fiz1             = _mm_setzero_pd();
 760         fix2             = _mm_setzero_pd();
 761         fiy2             = _mm_setzero_pd();
 762         fiz2             = _mm_setzero_pd();
 763         fix3             = _mm_setzero_pd();
 764         fiy3             = _mm_setzero_pd();
 765         fiz3             = _mm_setzero_pd();
 766
 767         /* Start inner kernel loop */
 768         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 769         {
 770
 771             /* Get j neighbor index, and coordinate index */
 772             jnrA             = jjnr[jidx];
 773             jnrB             = jjnr[jidx+1];
 774             j_coord_offsetA  = DIM*jnrA;
 775             j_coord_offsetB  = DIM*jnrB;
 776
 777             /* load j atom coordinates */
 778             gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 779                                               &jx0,&jy0,&jz0);
 780
 781             /* Calculate displacement vector */
 782             dx00             = _mm_sub_pd(ix0,jx0);
 783             dy00             = _mm_sub_pd(iy0,jy0);
 784             dz00             = _mm_sub_pd(iz0,jz0);
 785             dx10             = _mm_sub_pd(ix1,jx0);
 786             dy10             = _mm_sub_pd(iy1,jy0);
 787             dz10             = _mm_sub_pd(iz1,jz0);
 788             dx20             = _mm_sub_pd(ix2,jx0);
 789             dy20             = _mm_sub_pd(iy2,jy0);
 790             dz20             = _mm_sub_pd(iz2,jz0);
 791             dx30             = _mm_sub_pd(ix3,jx0);
 792             dy30             = _mm_sub_pd(iy3,jy0);
 793             dz30             = _mm_sub_pd(iz3,jz0);
 794
 795             /* Calculate squared distance and things based on it */
 796             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 797             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
 798             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
 799             rsq30            = gmx_mm_calc_rsq_pd(dx30,dy30,dz30);
 800
 801             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 802             rinv10           = gmx_mm_invsqrt_pd(rsq10);
 803             rinv20           = gmx_mm_invsqrt_pd(rsq20);
 804             rinv30           = gmx_mm_invsqrt_pd(rsq30);
 805
 806             rinvsq10         = _mm_mul_pd(rinv10,rinv10);
 807             rinvsq20         = _mm_mul_pd(rinv20,rinv20);
 808             rinvsq30         = _mm_mul_pd(rinv30,rinv30);
 809
 810             /* Load parameters for j particles */
 811             jq0              = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
 812             vdwjidx0A        = 2*vdwtype[jnrA+0];
 813             vdwjidx0B        = 2*vdwtype[jnrB+0];
 814
 815             fjx0             = _mm_setzero_pd();
 816             fjy0             = _mm_setzero_pd();
 817             fjz0             = _mm_setzero_pd();
 818
 819             /**************************
 820              * CALCULATE INTERACTIONS *
 821              **************************/
 822
 823             r00              = _mm_mul_pd(rsq00,rinv00);
 824
 825             /* Compute parameters for interactions between i and j atoms */
 826             gmx_mm_load_2pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,
 827                                          vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
 828
 829             /* Calculate table index by multiplying r with table scale and truncate to integer */
 830             rt               = _mm_mul_pd(r00,vftabscale);
 831             vfitab           = _mm_cvttpd_epi32(rt);
 832             vfeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
 833             vfitab           = _mm_slli_epi32(vfitab,3);
 834
 835             /* CUBIC SPLINE TABLE DISPERSION */
 836             Y                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
 837             F                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) );
 838             GMX_MM_TRANSPOSE2_PD(Y,F);
 839             G                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
 840             H                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) +2);
 841             GMX_MM_TRANSPOSE2_PD(G,H);
 842             Heps             = _mm_mul_pd(vfeps,H);
 843             Fp               = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
 844             FF               = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
 845             fvdw6            = _mm_mul_pd(c6_00,FF);
 846
 847             /* CUBIC SPLINE TABLE REPULSION */
 848             vfitab           = _mm_add_epi32(vfitab,ifour);
 849             Y                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
 850             F                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) );
 851             GMX_MM_TRANSPOSE2_PD(Y,F);
 852             G                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
 853             H                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) +2);
 854             GMX_MM_TRANSPOSE2_PD(G,H);
 855             Heps             = _mm_mul_pd(vfeps,H);
 856             Fp               = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
 857             FF               = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
 858             fvdw12           = _mm_mul_pd(c12_00,FF);
 859             fvdw             = _mm_xor_pd(signbit,_mm_mul_pd(_mm_add_pd(fvdw6,fvdw12),_mm_mul_pd(vftabscale,rinv00)));
 860
 861             fscal            = fvdw;
 862
 863             /* Calculate temporary vectorial force */
 864             tx               = _mm_mul_pd(fscal,dx00);
 865             ty               = _mm_mul_pd(fscal,dy00);
 866             tz               = _mm_mul_pd(fscal,dz00);
 867
 868             /* Update vectorial force */
 869             fix0             = _mm_add_pd(fix0,tx);
 870             fiy0             = _mm_add_pd(fiy0,ty);
 871             fiz0             = _mm_add_pd(fiz0,tz);
 872
 873             fjx0             = _mm_add_pd(fjx0,tx);
 874             fjy0             = _mm_add_pd(fjy0,ty);
 875             fjz0             = _mm_add_pd(fjz0,tz);
 876
 877             /**************************
 878              * CALCULATE INTERACTIONS *
 879              **************************/
 880
 881             /* Compute parameters for interactions between i and j atoms */
 882             qq10             = _mm_mul_pd(iq1,jq0);
 883
 884             /* REACTION-FIELD ELECTROSTATICS */
 885             felec            = _mm_mul_pd(qq10,_mm_sub_pd(_mm_mul_pd(rinv10,rinvsq10),krf2));
 886
 887             fscal            = felec;
 888
 889             /* Calculate temporary vectorial force */
 890             tx               = _mm_mul_pd(fscal,dx10);
 891             ty               = _mm_mul_pd(fscal,dy10);
 892             tz               = _mm_mul_pd(fscal,dz10);
 893
 894             /* Update vectorial force */
 895             fix1             = _mm_add_pd(fix1,tx);
 896             fiy1             = _mm_add_pd(fiy1,ty);
 897             fiz1             = _mm_add_pd(fiz1,tz);
 898
 899             fjx0             = _mm_add_pd(fjx0,tx);
 900             fjy0             = _mm_add_pd(fjy0,ty);
 901             fjz0             = _mm_add_pd(fjz0,tz);
 902
 903             /**************************
 904              * CALCULATE INTERACTIONS *
 905              **************************/
 906
 907             /* Compute parameters for interactions between i and j atoms */
 908             qq20             = _mm_mul_pd(iq2,jq0);
 909
 910             /* REACTION-FIELD ELECTROSTATICS */
 911             felec            = _mm_mul_pd(qq20,_mm_sub_pd(_mm_mul_pd(rinv20,rinvsq20),krf2));
 912
 913             fscal            = felec;
 914
 915             /* Calculate temporary vectorial force */
 916             tx               = _mm_mul_pd(fscal,dx20);
 917             ty               = _mm_mul_pd(fscal,dy20);
 918             tz               = _mm_mul_pd(fscal,dz20);
 919
 920             /* Update vectorial force */
 921             fix2             = _mm_add_pd(fix2,tx);
 922             fiy2             = _mm_add_pd(fiy2,ty);
 923             fiz2             = _mm_add_pd(fiz2,tz);
 924
 925             fjx0             = _mm_add_pd(fjx0,tx);
 926             fjy0             = _mm_add_pd(fjy0,ty);
 927             fjz0             = _mm_add_pd(fjz0,tz);
 928
 929             /**************************
 930              * CALCULATE INTERACTIONS *
 931              **************************/
 932
 933             /* Compute parameters for interactions between i and j atoms */
 934             qq30             = _mm_mul_pd(iq3,jq0);
 935
 936             /* REACTION-FIELD ELECTROSTATICS */
 937             felec            = _mm_mul_pd(qq30,_mm_sub_pd(_mm_mul_pd(rinv30,rinvsq30),krf2));
 938
 939             fscal            = felec;
 940
 941             /* Calculate temporary vectorial force */
 942             tx               = _mm_mul_pd(fscal,dx30);
 943             ty               = _mm_mul_pd(fscal,dy30);
 944             tz               = _mm_mul_pd(fscal,dz30);
 945
 946             /* Update vectorial force */
 947             fix3             = _mm_add_pd(fix3,tx);
 948             fiy3             = _mm_add_pd(fiy3,ty);
 949             fiz3             = _mm_add_pd(fiz3,tz);
 950
 951             fjx0             = _mm_add_pd(fjx0,tx);
 952             fjy0             = _mm_add_pd(fjy0,ty);
 953             fjz0             = _mm_add_pd(fjz0,tz);
 954
 955             gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,fjx0,fjy0,fjz0);
 956
 957             /* Inner loop uses 132 flops */
 958         }
 959
 960         if(jidx<j_index_end)
 961         {
 962
 963             jnrA             = jjnr[jidx];
 964             j_coord_offsetA  = DIM*jnrA;
 965
 966             /* load j atom coordinates */
 967             gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
 968                                               &jx0,&jy0,&jz0);
 969
 970             /* Calculate displacement vector */
 971             dx00             = _mm_sub_pd(ix0,jx0);
 972             dy00             = _mm_sub_pd(iy0,jy0);
 973             dz00             = _mm_sub_pd(iz0,jz0);
 974             dx10             = _mm_sub_pd(ix1,jx0);
 975             dy10             = _mm_sub_pd(iy1,jy0);
 976             dz10             = _mm_sub_pd(iz1,jz0);
 977             dx20             = _mm_sub_pd(ix2,jx0);
 978             dy20             = _mm_sub_pd(iy2,jy0);
 979             dz20             = _mm_sub_pd(iz2,jz0);
 980             dx30             = _mm_sub_pd(ix3,jx0);
 981             dy30             = _mm_sub_pd(iy3,jy0);
 982             dz30             = _mm_sub_pd(iz3,jz0);
 983
 984             /* Calculate squared distance and things based on it */
 985             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 986             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
 987             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
 988             rsq30            = gmx_mm_calc_rsq_pd(dx30,dy30,dz30);
 989
 990             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 991             rinv10           = gmx_mm_invsqrt_pd(rsq10);
 992             rinv20           = gmx_mm_invsqrt_pd(rsq20);
 993             rinv30           = gmx_mm_invsqrt_pd(rsq30);
 994
 995             rinvsq10         = _mm_mul_pd(rinv10,rinv10);
 996             rinvsq20         = _mm_mul_pd(rinv20,rinv20);
 997             rinvsq30         = _mm_mul_pd(rinv30,rinv30);
 998
 999             /* Load parameters for j particles */
1000             jq0              = _mm_load_sd(charge+jnrA+0);
1001             vdwjidx0A        = 2*vdwtype[jnrA+0];
1002
1003             fjx0             = _mm_setzero_pd();
1004             fjy0             = _mm_setzero_pd();
1005             fjz0             = _mm_setzero_pd();
1006
1007             /**************************
1008              * CALCULATE INTERACTIONS *
1009              **************************/
1010
1011             r00              = _mm_mul_pd(rsq00,rinv00);
1012
1013             /* Compute parameters for interactions between i and j atoms */
1014             gmx_mm_load_1pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,&c6_00,&c12_00);
1015
1016             /* Calculate table index by multiplying r with table scale and truncate to integer */
1017             rt               = _mm_mul_pd(r00,vftabscale);
1018             vfitab           = _mm_cvttpd_epi32(rt);
1019             vfeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
1020             vfitab           = _mm_slli_epi32(vfitab,3);
1021
1022             /* CUBIC SPLINE TABLE DISPERSION */
1023             Y                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
1024             F                = _mm_setzero_pd();
1025             GMX_MM_TRANSPOSE2_PD(Y,F);
1026             G                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
1027             H                = _mm_setzero_pd();
1028             GMX_MM_TRANSPOSE2_PD(G,H);
1029             Heps             = _mm_mul_pd(vfeps,H);
1030             Fp               = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
1031             FF               = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
1032             fvdw6            = _mm_mul_pd(c6_00,FF);
1033
1034             /* CUBIC SPLINE TABLE REPULSION */
1035             vfitab           = _mm_add_epi32(vfitab,ifour);
1036             Y                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
1037             F                = _mm_setzero_pd();
1038             GMX_MM_TRANSPOSE2_PD(Y,F);
1039             G                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
1040             H                = _mm_setzero_pd();
1041             GMX_MM_TRANSPOSE2_PD(G,H);
1042             Heps             = _mm_mul_pd(vfeps,H);
1043             Fp               = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
1044             FF               = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
1045             fvdw12           = _mm_mul_pd(c12_00,FF);
1046             fvdw             = _mm_xor_pd(signbit,_mm_mul_pd(_mm_add_pd(fvdw6,fvdw12),_mm_mul_pd(vftabscale,rinv00)));
1047
1048             fscal            = fvdw;
1049
1050             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1051
1052             /* Calculate temporary vectorial force */
1053             tx               = _mm_mul_pd(fscal,dx00);
1054             ty               = _mm_mul_pd(fscal,dy00);
1055             tz               = _mm_mul_pd(fscal,dz00);
1056
1057             /* Update vectorial force */
1058             fix0             = _mm_add_pd(fix0,tx);
1059             fiy0             = _mm_add_pd(fiy0,ty);
1060             fiz0             = _mm_add_pd(fiz0,tz);
1061
1062             fjx0             = _mm_add_pd(fjx0,tx);
1063             fjy0             = _mm_add_pd(fjy0,ty);
1064             fjz0             = _mm_add_pd(fjz0,tz);
1065
1066             /**************************
1067              * CALCULATE INTERACTIONS *
1068              **************************/
1069
1070             /* Compute parameters for interactions between i and j atoms */
1071             qq10             = _mm_mul_pd(iq1,jq0);
1072
1073             /* REACTION-FIELD ELECTROSTATICS */
1074             felec            = _mm_mul_pd(qq10,_mm_sub_pd(_mm_mul_pd(rinv10,rinvsq10),krf2));
1075
1076             fscal            = felec;
1077
1078             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1079
1080             /* Calculate temporary vectorial force */
1081             tx               = _mm_mul_pd(fscal,dx10);
1082             ty               = _mm_mul_pd(fscal,dy10);
1083             tz               = _mm_mul_pd(fscal,dz10);
1084
1085             /* Update vectorial force */
1086             fix1             = _mm_add_pd(fix1,tx);
1087             fiy1             = _mm_add_pd(fiy1,ty);
1088             fiz1             = _mm_add_pd(fiz1,tz);
1089
1090             fjx0             = _mm_add_pd(fjx0,tx);
1091             fjy0             = _mm_add_pd(fjy0,ty);
1092             fjz0             = _mm_add_pd(fjz0,tz);
1093
1094             /**************************
1095              * CALCULATE INTERACTIONS *
1096              **************************/
1097
1098             /* Compute parameters for interactions between i and j atoms */
1099             qq20             = _mm_mul_pd(iq2,jq0);
1100
1101             /* REACTION-FIELD ELECTROSTATICS */
1102             felec            = _mm_mul_pd(qq20,_mm_sub_pd(_mm_mul_pd(rinv20,rinvsq20),krf2));
1103
1104             fscal            = felec;
1105
1106             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1107
1108             /* Calculate temporary vectorial force */
1109             tx               = _mm_mul_pd(fscal,dx20);
1110             ty               = _mm_mul_pd(fscal,dy20);
1111             tz               = _mm_mul_pd(fscal,dz20);
1112
1113             /* Update vectorial force */
1114             fix2             = _mm_add_pd(fix2,tx);
1115             fiy2             = _mm_add_pd(fiy2,ty);
1116             fiz2             = _mm_add_pd(fiz2,tz);
1117
1118             fjx0             = _mm_add_pd(fjx0,tx);
1119             fjy0             = _mm_add_pd(fjy0,ty);
1120             fjz0             = _mm_add_pd(fjz0,tz);
1121
1122             /**************************
1123              * CALCULATE INTERACTIONS *
1124              **************************/
1125
1126             /* Compute parameters for interactions between i and j atoms */
1127             qq30             = _mm_mul_pd(iq3,jq0);
1128
1129             /* REACTION-FIELD ELECTROSTATICS */
1130             felec            = _mm_mul_pd(qq30,_mm_sub_pd(_mm_mul_pd(rinv30,rinvsq30),krf2));
1131
1132             fscal            = felec;
1133
1134             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1135
1136             /* Calculate temporary vectorial force */
1137             tx               = _mm_mul_pd(fscal,dx30);
1138             ty               = _mm_mul_pd(fscal,dy30);
1139             tz               = _mm_mul_pd(fscal,dz30);
1140
1141             /* Update vectorial force */
1142             fix3             = _mm_add_pd(fix3,tx);
1143             fiy3             = _mm_add_pd(fiy3,ty);
1144             fiz3             = _mm_add_pd(fiz3,tz);
1145
1146             fjx0             = _mm_add_pd(fjx0,tx);
1147             fjy0             = _mm_add_pd(fjy0,ty);
1148             fjz0             = _mm_add_pd(fjz0,tz);
1149
1150             gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,fjx0,fjy0,fjz0);
1151
1152             /* Inner loop uses 132 flops */
1153         }
1154
1155         /* End of innermost loop */
1156
1157         gmx_mm_update_iforce_4atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
1158                                               f+i_coord_offset,fshift+i_shift_offset);
1159
1160         /* Increment number of inner iterations */
1161         inneriter                  += j_index_end - j_index_start;
1162
1163         /* Outer loop uses 24 flops */
1164     }
1165
1166     /* Increment number of outer iterations */
1167     outeriter        += nri;
1168
1169     /* Update outer/inner flops */
1170
1171     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W4_F,outeriter*24 + inneriter*132);
1172 }