src/gromacs/gmxlib/nonbonded/nb_kernel_sse4_1_double/nb_kernel_ElecRF_VdwCSTab_GeomW3P1_sse4_1_double.c

   1 /*
   2  * This file is part of the GROMACS molecular simulation package.
   3  *
   4  * Copyright (c) 2012,2013,2014, by the GROMACS development team, led by
   5  * Mark Abraham, David van der Spoel, Berk Hess, and Erik Lindahl,
   6  * and including many others, as listed in the AUTHORS file in the
   7  * top-level source directory and at http://www.gromacs.org.
   8  *
   9  * GROMACS is free software; you can redistribute it and/or
  10  * modify it under the terms of the GNU Lesser General Public License
  11  * as published by the Free Software Foundation; either version 2.1
  12  * of the License, or (at your option) any later version.
  13  *
  14  * GROMACS is distributed in the hope that it will be useful,
  15  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  16  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  17  * Lesser General Public License for more details.
  18  *
  19  * You should have received a copy of the GNU Lesser General Public
  20  * License along with GROMACS; if not, see
  21  * http://www.gnu.org/licenses, or write to the Free Software Foundation,
  22  * Inc., 51 Franklin Street, Fifth Floor, Boston, MA  02110-1301  USA.
  23  *
  24  * If you want to redistribute modifications to GROMACS, please
  25  * consider that scientific software is very special. Version
  26  * control is crucial - bugs must be traceable. We will be happy to
  27  * consider code for inclusion in the official distribution, but
  28  * derived work must not be called official GROMACS. Details are found
  29  * in the README & COPYING files - if they are missing, get the
  30  * official version at http://www.gromacs.org.
  31  *
  32  * To help us fund GROMACS development, we humbly ask that you cite
  33  * the research papers on the package. Check out http://www.gromacs.org.
  34  */
  35 /*
  36  * Note: this file was generated by the GROMACS sse4_1_double kernel generator.
  37  */
  38 #include "gmxpre.h"
  39
  40 #include "config.h"
  41
  42 #include <math.h>
  43
  44 #include "../nb_kernel.h"
  45 #include "gromacs/legacyheaders/types/simple.h"
  46 #include "gromacs/math/vec.h"
  47 #include "gromacs/legacyheaders/nrnb.h"
  48
  49 #include "gromacs/simd/math_x86_sse4_1_double.h"
  50 #include "kernelutil_x86_sse4_1_double.h"
  51
  52 /*
  53  * Gromacs nonbonded kernel:   nb_kernel_ElecRF_VdwCSTab_GeomW3P1_VF_sse4_1_double
  54  * Electrostatics interaction: ReactionField
  55  * VdW interaction:            CubicSplineTable
  56  * Geometry:                   Water3-Particle
  57  * Calculate force/pot:        PotentialAndForce
  58  */
  59 void
  60 nb_kernel_ElecRF_VdwCSTab_GeomW3P1_VF_sse4_1_double
  61                     (t_nblist                    * gmx_restrict       nlist,
  62                      rvec                        * gmx_restrict          xx,
  63                      rvec                        * gmx_restrict          ff,
  64                      t_forcerec                  * gmx_restrict          fr,
  65                      t_mdatoms                   * gmx_restrict     mdatoms,
  66                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
  67                      t_nrnb                      * gmx_restrict        nrnb)
  68 {
  69     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
  70      * just 0 for non-waters.
  71      * Suffixes A,B refer to j loop unrolling done with SSE double precision, e.g. for the two different
  72      * jnr indices corresponding to data put in the four positions in the SIMD register.
  73      */
  74     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
  75     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
  76     int              jnrA,jnrB;
  77     int              j_coord_offsetA,j_coord_offsetB;
  78     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
  79     real             rcutoff_scalar;
  80     real             *shiftvec,*fshift,*x,*f;
  81     __m128d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
  82     int              vdwioffset0;
  83     __m128d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
  84     int              vdwioffset1;
  85     __m128d          ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
  86     int              vdwioffset2;
  87     __m128d          ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
  88     int              vdwjidx0A,vdwjidx0B;
  89     __m128d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  90     __m128d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  91     __m128d          dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
  92     __m128d          dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
  93     __m128d          velec,felec,velecsum,facel,crf,krf,krf2;
  94     real             *charge;
  95     int              nvdwtype;
  96     __m128d          rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
  97     int              *vdwtype;
  98     real             *vdwparam;
  99     __m128d          one_sixth   = _mm_set1_pd(1.0/6.0);
 100     __m128d          one_twelfth = _mm_set1_pd(1.0/12.0);
 101     __m128i          vfitab;
 102     __m128i          ifour       = _mm_set1_epi32(4);
 103     __m128d          rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF;
 104     real             *vftab;
 105     __m128d          dummy_mask,cutoff_mask;
 106     __m128d          signbit   = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
 107     __m128d          one     = _mm_set1_pd(1.0);
 108     __m128d          two     = _mm_set1_pd(2.0);
 109     x                = xx[0];
 110     f                = ff[0];
 111
 112     nri              = nlist->nri;
 113     iinr             = nlist->iinr;
 114     jindex           = nlist->jindex;
 115     jjnr             = nlist->jjnr;
 116     shiftidx         = nlist->shift;
 117     gid              = nlist->gid;
 118     shiftvec         = fr->shift_vec[0];
 119     fshift           = fr->fshift[0];
 120     facel            = _mm_set1_pd(fr->epsfac);
 121     charge           = mdatoms->chargeA;
 122     krf              = _mm_set1_pd(fr->ic->k_rf);
 123     krf2             = _mm_set1_pd(fr->ic->k_rf*2.0);
 124     crf              = _mm_set1_pd(fr->ic->c_rf);
 125     nvdwtype         = fr->ntype;
 126     vdwparam         = fr->nbfp;
 127     vdwtype          = mdatoms->typeA;
 128
 129     vftab            = kernel_data->table_vdw->data;
 130     vftabscale       = _mm_set1_pd(kernel_data->table_vdw->scale);
 131
 132     /* Setup water-specific parameters */
 133     inr              = nlist->iinr[0];
 134     iq0              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+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     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 138
 139     /* Avoid stupid compiler warnings */
 140     jnrA = jnrB = 0;
 141     j_coord_offsetA = 0;
 142     j_coord_offsetB = 0;
 143
 144     outeriter        = 0;
 145     inneriter        = 0;
 146
 147     /* Start outer loop over neighborlists */
 148     for(iidx=0; iidx<nri; iidx++)
 149     {
 150         /* Load shift vector for this list */
 151         i_shift_offset   = DIM*shiftidx[iidx];
 152
 153         /* Load limits for loop over neighbors */
 154         j_index_start    = jindex[iidx];
 155         j_index_end      = jindex[iidx+1];
 156
 157         /* Get outer coordinate index */
 158         inr              = iinr[iidx];
 159         i_coord_offset   = DIM*inr;
 160
 161         /* Load i particle coords and add shift vector */
 162         gmx_mm_load_shift_and_3rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
 163                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 164
 165         fix0             = _mm_setzero_pd();
 166         fiy0             = _mm_setzero_pd();
 167         fiz0             = _mm_setzero_pd();
 168         fix1             = _mm_setzero_pd();
 169         fiy1             = _mm_setzero_pd();
 170         fiz1             = _mm_setzero_pd();
 171         fix2             = _mm_setzero_pd();
 172         fiy2             = _mm_setzero_pd();
 173         fiz2             = _mm_setzero_pd();
 174
 175         /* Reset potential sums */
 176         velecsum         = _mm_setzero_pd();
 177         vvdwsum          = _mm_setzero_pd();
 178
 179         /* Start inner kernel loop */
 180         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 181         {
 182
 183             /* Get j neighbor index, and coordinate index */
 184             jnrA             = jjnr[jidx];
 185             jnrB             = jjnr[jidx+1];
 186             j_coord_offsetA  = DIM*jnrA;
 187             j_coord_offsetB  = DIM*jnrB;
 188
 189             /* load j atom coordinates */
 190             gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 191                                               &jx0,&jy0,&jz0);
 192
 193             /* Calculate displacement vector */
 194             dx00             = _mm_sub_pd(ix0,jx0);
 195             dy00             = _mm_sub_pd(iy0,jy0);
 196             dz00             = _mm_sub_pd(iz0,jz0);
 197             dx10             = _mm_sub_pd(ix1,jx0);
 198             dy10             = _mm_sub_pd(iy1,jy0);
 199             dz10             = _mm_sub_pd(iz1,jz0);
 200             dx20             = _mm_sub_pd(ix2,jx0);
 201             dy20             = _mm_sub_pd(iy2,jy0);
 202             dz20             = _mm_sub_pd(iz2,jz0);
 203
 204             /* Calculate squared distance and things based on it */
 205             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 206             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
 207             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
 208
 209             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 210             rinv10           = gmx_mm_invsqrt_pd(rsq10);
 211             rinv20           = gmx_mm_invsqrt_pd(rsq20);
 212
 213             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 214             rinvsq10         = _mm_mul_pd(rinv10,rinv10);
 215             rinvsq20         = _mm_mul_pd(rinv20,rinv20);
 216
 217             /* Load parameters for j particles */
 218             jq0              = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
 219             vdwjidx0A        = 2*vdwtype[jnrA+0];
 220             vdwjidx0B        = 2*vdwtype[jnrB+0];
 221
 222             fjx0             = _mm_setzero_pd();
 223             fjy0             = _mm_setzero_pd();
 224             fjz0             = _mm_setzero_pd();
 225
 226             /**************************
 227              * CALCULATE INTERACTIONS *
 228              **************************/
 229
 230             r00              = _mm_mul_pd(rsq00,rinv00);
 231
 232             /* Compute parameters for interactions between i and j atoms */
 233             qq00             = _mm_mul_pd(iq0,jq0);
 234             gmx_mm_load_2pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,
 235                                          vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
 236
 237             /* Calculate table index by multiplying r with table scale and truncate to integer */
 238             rt               = _mm_mul_pd(r00,vftabscale);
 239             vfitab           = _mm_cvttpd_epi32(rt);
 240             vfeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
 241             vfitab           = _mm_slli_epi32(vfitab,3);
 242
 243             /* REACTION-FIELD ELECTROSTATICS */
 244             velec            = _mm_mul_pd(qq00,_mm_sub_pd(_mm_add_pd(rinv00,_mm_mul_pd(krf,rsq00)),crf));
 245             felec            = _mm_mul_pd(qq00,_mm_sub_pd(_mm_mul_pd(rinv00,rinvsq00),krf2));
 246
 247             /* CUBIC SPLINE TABLE DISPERSION */
 248             Y                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
 249             F                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) );
 250             GMX_MM_TRANSPOSE2_PD(Y,F);
 251             G                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
 252             H                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) +2);
 253             GMX_MM_TRANSPOSE2_PD(G,H);
 254             Heps             = _mm_mul_pd(vfeps,H);
 255             Fp               = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
 256             VV               = _mm_add_pd(Y,_mm_mul_pd(vfeps,Fp));
 257             vvdw6            = _mm_mul_pd(c6_00,VV);
 258             FF               = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
 259             fvdw6            = _mm_mul_pd(c6_00,FF);
 260
 261             /* CUBIC SPLINE TABLE REPULSION */
 262             vfitab           = _mm_add_epi32(vfitab,ifour);
 263             Y                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
 264             F                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) );
 265             GMX_MM_TRANSPOSE2_PD(Y,F);
 266             G                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
 267             H                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) +2);
 268             GMX_MM_TRANSPOSE2_PD(G,H);
 269             Heps             = _mm_mul_pd(vfeps,H);
 270             Fp               = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
 271             VV               = _mm_add_pd(Y,_mm_mul_pd(vfeps,Fp));
 272             vvdw12           = _mm_mul_pd(c12_00,VV);
 273             FF               = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
 274             fvdw12           = _mm_mul_pd(c12_00,FF);
 275             vvdw             = _mm_add_pd(vvdw12,vvdw6);
 276             fvdw             = _mm_xor_pd(signbit,_mm_mul_pd(_mm_add_pd(fvdw6,fvdw12),_mm_mul_pd(vftabscale,rinv00)));
 277
 278             /* Update potential sum for this i atom from the interaction with this j atom. */
 279             velecsum         = _mm_add_pd(velecsum,velec);
 280             vvdwsum          = _mm_add_pd(vvdwsum,vvdw);
 281
 282             fscal            = _mm_add_pd(felec,fvdw);
 283
 284             /* Calculate temporary vectorial force */
 285             tx               = _mm_mul_pd(fscal,dx00);
 286             ty               = _mm_mul_pd(fscal,dy00);
 287             tz               = _mm_mul_pd(fscal,dz00);
 288
 289             /* Update vectorial force */
 290             fix0             = _mm_add_pd(fix0,tx);
 291             fiy0             = _mm_add_pd(fiy0,ty);
 292             fiz0             = _mm_add_pd(fiz0,tz);
 293
 294             fjx0             = _mm_add_pd(fjx0,tx);
 295             fjy0             = _mm_add_pd(fjy0,ty);
 296             fjz0             = _mm_add_pd(fjz0,tz);
 297
 298             /**************************
 299              * CALCULATE INTERACTIONS *
 300              **************************/
 301
 302             /* Compute parameters for interactions between i and j atoms */
 303             qq10             = _mm_mul_pd(iq1,jq0);
 304
 305             /* REACTION-FIELD ELECTROSTATICS */
 306             velec            = _mm_mul_pd(qq10,_mm_sub_pd(_mm_add_pd(rinv10,_mm_mul_pd(krf,rsq10)),crf));
 307             felec            = _mm_mul_pd(qq10,_mm_sub_pd(_mm_mul_pd(rinv10,rinvsq10),krf2));
 308
 309             /* Update potential sum for this i atom from the interaction with this j atom. */
 310             velecsum         = _mm_add_pd(velecsum,velec);
 311
 312             fscal            = felec;
 313
 314             /* Calculate temporary vectorial force */
 315             tx               = _mm_mul_pd(fscal,dx10);
 316             ty               = _mm_mul_pd(fscal,dy10);
 317             tz               = _mm_mul_pd(fscal,dz10);
 318
 319             /* Update vectorial force */
 320             fix1             = _mm_add_pd(fix1,tx);
 321             fiy1             = _mm_add_pd(fiy1,ty);
 322             fiz1             = _mm_add_pd(fiz1,tz);
 323
 324             fjx0             = _mm_add_pd(fjx0,tx);
 325             fjy0             = _mm_add_pd(fjy0,ty);
 326             fjz0             = _mm_add_pd(fjz0,tz);
 327
 328             /**************************
 329              * CALCULATE INTERACTIONS *
 330              **************************/
 331
 332             /* Compute parameters for interactions between i and j atoms */
 333             qq20             = _mm_mul_pd(iq2,jq0);
 334
 335             /* REACTION-FIELD ELECTROSTATICS */
 336             velec            = _mm_mul_pd(qq20,_mm_sub_pd(_mm_add_pd(rinv20,_mm_mul_pd(krf,rsq20)),crf));
 337             felec            = _mm_mul_pd(qq20,_mm_sub_pd(_mm_mul_pd(rinv20,rinvsq20),krf2));
 338
 339             /* Update potential sum for this i atom from the interaction with this j atom. */
 340             velecsum         = _mm_add_pd(velecsum,velec);
 341
 342             fscal            = felec;
 343
 344             /* Calculate temporary vectorial force */
 345             tx               = _mm_mul_pd(fscal,dx20);
 346             ty               = _mm_mul_pd(fscal,dy20);
 347             tz               = _mm_mul_pd(fscal,dz20);
 348
 349             /* Update vectorial force */
 350             fix2             = _mm_add_pd(fix2,tx);
 351             fiy2             = _mm_add_pd(fiy2,ty);
 352             fiz2             = _mm_add_pd(fiz2,tz);
 353
 354             fjx0             = _mm_add_pd(fjx0,tx);
 355             fjy0             = _mm_add_pd(fjy0,ty);
 356             fjz0             = _mm_add_pd(fjz0,tz);
 357
 358             gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,fjx0,fjy0,fjz0);
 359
 360             /* Inner loop uses 134 flops */
 361         }
 362
 363         if(jidx<j_index_end)
 364         {
 365
 366             jnrA             = jjnr[jidx];
 367             j_coord_offsetA  = DIM*jnrA;
 368
 369             /* load j atom coordinates */
 370             gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
 371                                               &jx0,&jy0,&jz0);
 372
 373             /* Calculate displacement vector */
 374             dx00             = _mm_sub_pd(ix0,jx0);
 375             dy00             = _mm_sub_pd(iy0,jy0);
 376             dz00             = _mm_sub_pd(iz0,jz0);
 377             dx10             = _mm_sub_pd(ix1,jx0);
 378             dy10             = _mm_sub_pd(iy1,jy0);
 379             dz10             = _mm_sub_pd(iz1,jz0);
 380             dx20             = _mm_sub_pd(ix2,jx0);
 381             dy20             = _mm_sub_pd(iy2,jy0);
 382             dz20             = _mm_sub_pd(iz2,jz0);
 383
 384             /* Calculate squared distance and things based on it */
 385             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 386             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
 387             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
 388
 389             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 390             rinv10           = gmx_mm_invsqrt_pd(rsq10);
 391             rinv20           = gmx_mm_invsqrt_pd(rsq20);
 392
 393             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 394             rinvsq10         = _mm_mul_pd(rinv10,rinv10);
 395             rinvsq20         = _mm_mul_pd(rinv20,rinv20);
 396
 397             /* Load parameters for j particles */
 398             jq0              = _mm_load_sd(charge+jnrA+0);
 399             vdwjidx0A        = 2*vdwtype[jnrA+0];
 400
 401             fjx0             = _mm_setzero_pd();
 402             fjy0             = _mm_setzero_pd();
 403             fjz0             = _mm_setzero_pd();
 404
 405             /**************************
 406              * CALCULATE INTERACTIONS *
 407              **************************/
 408
 409             r00              = _mm_mul_pd(rsq00,rinv00);
 410
 411             /* Compute parameters for interactions between i and j atoms */
 412             qq00             = _mm_mul_pd(iq0,jq0);
 413             gmx_mm_load_1pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,&c6_00,&c12_00);
 414
 415             /* Calculate table index by multiplying r with table scale and truncate to integer */
 416             rt               = _mm_mul_pd(r00,vftabscale);
 417             vfitab           = _mm_cvttpd_epi32(rt);
 418             vfeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
 419             vfitab           = _mm_slli_epi32(vfitab,3);
 420
 421             /* REACTION-FIELD ELECTROSTATICS */
 422             velec            = _mm_mul_pd(qq00,_mm_sub_pd(_mm_add_pd(rinv00,_mm_mul_pd(krf,rsq00)),crf));
 423             felec            = _mm_mul_pd(qq00,_mm_sub_pd(_mm_mul_pd(rinv00,rinvsq00),krf2));
 424
 425             /* CUBIC SPLINE TABLE DISPERSION */
 426             Y                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
 427             F                = _mm_setzero_pd();
 428             GMX_MM_TRANSPOSE2_PD(Y,F);
 429             G                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
 430             H                = _mm_setzero_pd();
 431             GMX_MM_TRANSPOSE2_PD(G,H);
 432             Heps             = _mm_mul_pd(vfeps,H);
 433             Fp               = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
 434             VV               = _mm_add_pd(Y,_mm_mul_pd(vfeps,Fp));
 435             vvdw6            = _mm_mul_pd(c6_00,VV);
 436             FF               = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
 437             fvdw6            = _mm_mul_pd(c6_00,FF);
 438
 439             /* CUBIC SPLINE TABLE REPULSION */
 440             vfitab           = _mm_add_epi32(vfitab,ifour);
 441             Y                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
 442             F                = _mm_setzero_pd();
 443             GMX_MM_TRANSPOSE2_PD(Y,F);
 444             G                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
 445             H                = _mm_setzero_pd();
 446             GMX_MM_TRANSPOSE2_PD(G,H);
 447             Heps             = _mm_mul_pd(vfeps,H);
 448             Fp               = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
 449             VV               = _mm_add_pd(Y,_mm_mul_pd(vfeps,Fp));
 450             vvdw12           = _mm_mul_pd(c12_00,VV);
 451             FF               = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
 452             fvdw12           = _mm_mul_pd(c12_00,FF);
 453             vvdw             = _mm_add_pd(vvdw12,vvdw6);
 454             fvdw             = _mm_xor_pd(signbit,_mm_mul_pd(_mm_add_pd(fvdw6,fvdw12),_mm_mul_pd(vftabscale,rinv00)));
 455
 456             /* Update potential sum for this i atom from the interaction with this j atom. */
 457             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 458             velecsum         = _mm_add_pd(velecsum,velec);
 459             vvdw             = _mm_unpacklo_pd(vvdw,_mm_setzero_pd());
 460             vvdwsum          = _mm_add_pd(vvdwsum,vvdw);
 461
 462             fscal            = _mm_add_pd(felec,fvdw);
 463
 464             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 465
 466             /* Calculate temporary vectorial force */
 467             tx               = _mm_mul_pd(fscal,dx00);
 468             ty               = _mm_mul_pd(fscal,dy00);
 469             tz               = _mm_mul_pd(fscal,dz00);
 470
 471             /* Update vectorial force */
 472             fix0             = _mm_add_pd(fix0,tx);
 473             fiy0             = _mm_add_pd(fiy0,ty);
 474             fiz0             = _mm_add_pd(fiz0,tz);
 475
 476             fjx0             = _mm_add_pd(fjx0,tx);
 477             fjy0             = _mm_add_pd(fjy0,ty);
 478             fjz0             = _mm_add_pd(fjz0,tz);
 479
 480             /**************************
 481              * CALCULATE INTERACTIONS *
 482              **************************/
 483
 484             /* Compute parameters for interactions between i and j atoms */
 485             qq10             = _mm_mul_pd(iq1,jq0);
 486
 487             /* REACTION-FIELD ELECTROSTATICS */
 488             velec            = _mm_mul_pd(qq10,_mm_sub_pd(_mm_add_pd(rinv10,_mm_mul_pd(krf,rsq10)),crf));
 489             felec            = _mm_mul_pd(qq10,_mm_sub_pd(_mm_mul_pd(rinv10,rinvsq10),krf2));
 490
 491             /* Update potential sum for this i atom from the interaction with this j atom. */
 492             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 493             velecsum         = _mm_add_pd(velecsum,velec);
 494
 495             fscal            = felec;
 496
 497             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 498
 499             /* Calculate temporary vectorial force */
 500             tx               = _mm_mul_pd(fscal,dx10);
 501             ty               = _mm_mul_pd(fscal,dy10);
 502             tz               = _mm_mul_pd(fscal,dz10);
 503
 504             /* Update vectorial force */
 505             fix1             = _mm_add_pd(fix1,tx);
 506             fiy1             = _mm_add_pd(fiy1,ty);
 507             fiz1             = _mm_add_pd(fiz1,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             qq20             = _mm_mul_pd(iq2,jq0);
 519
 520             /* REACTION-FIELD ELECTROSTATICS */
 521             velec            = _mm_mul_pd(qq20,_mm_sub_pd(_mm_add_pd(rinv20,_mm_mul_pd(krf,rsq20)),crf));
 522             felec            = _mm_mul_pd(qq20,_mm_sub_pd(_mm_mul_pd(rinv20,rinvsq20),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,dx20);
 534             ty               = _mm_mul_pd(fscal,dy20);
 535             tz               = _mm_mul_pd(fscal,dz20);
 536
 537             /* Update vectorial force */
 538             fix2             = _mm_add_pd(fix2,tx);
 539             fiy2             = _mm_add_pd(fiy2,ty);
 540             fiz2             = _mm_add_pd(fiz2,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             gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,fjx0,fjy0,fjz0);
 547
 548             /* Inner loop uses 134 flops */
 549         }
 550
 551         /* End of innermost loop */
 552
 553         gmx_mm_update_iforce_3atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
 554                                               f+i_coord_offset,fshift+i_shift_offset);
 555
 556         ggid                        = gid[iidx];
 557         /* Update potential energies */
 558         gmx_mm_update_1pot_pd(velecsum,kernel_data->energygrp_elec+ggid);
 559         gmx_mm_update_1pot_pd(vvdwsum,kernel_data->energygrp_vdw+ggid);
 560
 561         /* Increment number of inner iterations */
 562         inneriter                  += j_index_end - j_index_start;
 563
 564         /* Outer loop uses 20 flops */
 565     }
 566
 567     /* Increment number of outer iterations */
 568     outeriter        += nri;
 569
 570     /* Update outer/inner flops */
 571
 572     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3_VF,outeriter*20 + inneriter*134);
 573 }
 574 /*
 575  * Gromacs nonbonded kernel:   nb_kernel_ElecRF_VdwCSTab_GeomW3P1_F_sse4_1_double
 576  * Electrostatics interaction: ReactionField
 577  * VdW interaction:            CubicSplineTable
 578  * Geometry:                   Water3-Particle
 579  * Calculate force/pot:        Force
 580  */
 581 void
 582 nb_kernel_ElecRF_VdwCSTab_GeomW3P1_F_sse4_1_double
 583                     (t_nblist                    * gmx_restrict       nlist,
 584                      rvec                        * gmx_restrict          xx,
 585                      rvec                        * gmx_restrict          ff,
 586                      t_forcerec                  * gmx_restrict          fr,
 587                      t_mdatoms                   * gmx_restrict     mdatoms,
 588                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 589                      t_nrnb                      * gmx_restrict        nrnb)
 590 {
 591     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 592      * just 0 for non-waters.
 593      * Suffixes A,B refer to j loop unrolling done with SSE double precision, e.g. for the two different
 594      * jnr indices corresponding to data put in the four positions in the SIMD register.
 595      */
 596     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 597     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 598     int              jnrA,jnrB;
 599     int              j_coord_offsetA,j_coord_offsetB;
 600     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 601     real             rcutoff_scalar;
 602     real             *shiftvec,*fshift,*x,*f;
 603     __m128d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 604     int              vdwioffset0;
 605     __m128d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 606     int              vdwioffset1;
 607     __m128d          ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
 608     int              vdwioffset2;
 609     __m128d          ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
 610     int              vdwjidx0A,vdwjidx0B;
 611     __m128d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 612     __m128d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 613     __m128d          dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
 614     __m128d          dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
 615     __m128d          velec,felec,velecsum,facel,crf,krf,krf2;
 616     real             *charge;
 617     int              nvdwtype;
 618     __m128d          rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 619     int              *vdwtype;
 620     real             *vdwparam;
 621     __m128d          one_sixth   = _mm_set1_pd(1.0/6.0);
 622     __m128d          one_twelfth = _mm_set1_pd(1.0/12.0);
 623     __m128i          vfitab;
 624     __m128i          ifour       = _mm_set1_epi32(4);
 625     __m128d          rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF;
 626     real             *vftab;
 627     __m128d          dummy_mask,cutoff_mask;
 628     __m128d          signbit   = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
 629     __m128d          one     = _mm_set1_pd(1.0);
 630     __m128d          two     = _mm_set1_pd(2.0);
 631     x                = xx[0];
 632     f                = ff[0];
 633
 634     nri              = nlist->nri;
 635     iinr             = nlist->iinr;
 636     jindex           = nlist->jindex;
 637     jjnr             = nlist->jjnr;
 638     shiftidx         = nlist->shift;
 639     gid              = nlist->gid;
 640     shiftvec         = fr->shift_vec[0];
 641     fshift           = fr->fshift[0];
 642     facel            = _mm_set1_pd(fr->epsfac);
 643     charge           = mdatoms->chargeA;
 644     krf              = _mm_set1_pd(fr->ic->k_rf);
 645     krf2             = _mm_set1_pd(fr->ic->k_rf*2.0);
 646     crf              = _mm_set1_pd(fr->ic->c_rf);
 647     nvdwtype         = fr->ntype;
 648     vdwparam         = fr->nbfp;
 649     vdwtype          = mdatoms->typeA;
 650
 651     vftab            = kernel_data->table_vdw->data;
 652     vftabscale       = _mm_set1_pd(kernel_data->table_vdw->scale);
 653
 654     /* Setup water-specific parameters */
 655     inr              = nlist->iinr[0];
 656     iq0              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+0]));
 657     iq1              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+1]));
 658     iq2              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+2]));
 659     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 660
 661     /* Avoid stupid compiler warnings */
 662     jnrA = jnrB = 0;
 663     j_coord_offsetA = 0;
 664     j_coord_offsetB = 0;
 665
 666     outeriter        = 0;
 667     inneriter        = 0;
 668
 669     /* Start outer loop over neighborlists */
 670     for(iidx=0; iidx<nri; iidx++)
 671     {
 672         /* Load shift vector for this list */
 673         i_shift_offset   = DIM*shiftidx[iidx];
 674
 675         /* Load limits for loop over neighbors */
 676         j_index_start    = jindex[iidx];
 677         j_index_end      = jindex[iidx+1];
 678
 679         /* Get outer coordinate index */
 680         inr              = iinr[iidx];
 681         i_coord_offset   = DIM*inr;
 682
 683         /* Load i particle coords and add shift vector */
 684         gmx_mm_load_shift_and_3rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
 685                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 686
 687         fix0             = _mm_setzero_pd();
 688         fiy0             = _mm_setzero_pd();
 689         fiz0             = _mm_setzero_pd();
 690         fix1             = _mm_setzero_pd();
 691         fiy1             = _mm_setzero_pd();
 692         fiz1             = _mm_setzero_pd();
 693         fix2             = _mm_setzero_pd();
 694         fiy2             = _mm_setzero_pd();
 695         fiz2             = _mm_setzero_pd();
 696
 697         /* Start inner kernel loop */
 698         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 699         {
 700
 701             /* Get j neighbor index, and coordinate index */
 702             jnrA             = jjnr[jidx];
 703             jnrB             = jjnr[jidx+1];
 704             j_coord_offsetA  = DIM*jnrA;
 705             j_coord_offsetB  = DIM*jnrB;
 706
 707             /* load j atom coordinates */
 708             gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 709                                               &jx0,&jy0,&jz0);
 710
 711             /* Calculate displacement vector */
 712             dx00             = _mm_sub_pd(ix0,jx0);
 713             dy00             = _mm_sub_pd(iy0,jy0);
 714             dz00             = _mm_sub_pd(iz0,jz0);
 715             dx10             = _mm_sub_pd(ix1,jx0);
 716             dy10             = _mm_sub_pd(iy1,jy0);
 717             dz10             = _mm_sub_pd(iz1,jz0);
 718             dx20             = _mm_sub_pd(ix2,jx0);
 719             dy20             = _mm_sub_pd(iy2,jy0);
 720             dz20             = _mm_sub_pd(iz2,jz0);
 721
 722             /* Calculate squared distance and things based on it */
 723             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 724             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
 725             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
 726
 727             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 728             rinv10           = gmx_mm_invsqrt_pd(rsq10);
 729             rinv20           = gmx_mm_invsqrt_pd(rsq20);
 730
 731             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 732             rinvsq10         = _mm_mul_pd(rinv10,rinv10);
 733             rinvsq20         = _mm_mul_pd(rinv20,rinv20);
 734
 735             /* Load parameters for j particles */
 736             jq0              = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
 737             vdwjidx0A        = 2*vdwtype[jnrA+0];
 738             vdwjidx0B        = 2*vdwtype[jnrB+0];
 739
 740             fjx0             = _mm_setzero_pd();
 741             fjy0             = _mm_setzero_pd();
 742             fjz0             = _mm_setzero_pd();
 743
 744             /**************************
 745              * CALCULATE INTERACTIONS *
 746              **************************/
 747
 748             r00              = _mm_mul_pd(rsq00,rinv00);
 749
 750             /* Compute parameters for interactions between i and j atoms */
 751             qq00             = _mm_mul_pd(iq0,jq0);
 752             gmx_mm_load_2pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,
 753                                          vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
 754
 755             /* Calculate table index by multiplying r with table scale and truncate to integer */
 756             rt               = _mm_mul_pd(r00,vftabscale);
 757             vfitab           = _mm_cvttpd_epi32(rt);
 758             vfeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
 759             vfitab           = _mm_slli_epi32(vfitab,3);
 760
 761             /* REACTION-FIELD ELECTROSTATICS */
 762             felec            = _mm_mul_pd(qq00,_mm_sub_pd(_mm_mul_pd(rinv00,rinvsq00),krf2));
 763
 764             /* CUBIC SPLINE TABLE DISPERSION */
 765             Y                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
 766             F                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) );
 767             GMX_MM_TRANSPOSE2_PD(Y,F);
 768             G                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
 769             H                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) +2);
 770             GMX_MM_TRANSPOSE2_PD(G,H);
 771             Heps             = _mm_mul_pd(vfeps,H);
 772             Fp               = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
 773             FF               = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
 774             fvdw6            = _mm_mul_pd(c6_00,FF);
 775
 776             /* CUBIC SPLINE TABLE REPULSION */
 777             vfitab           = _mm_add_epi32(vfitab,ifour);
 778             Y                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
 779             F                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) );
 780             GMX_MM_TRANSPOSE2_PD(Y,F);
 781             G                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
 782             H                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) +2);
 783             GMX_MM_TRANSPOSE2_PD(G,H);
 784             Heps             = _mm_mul_pd(vfeps,H);
 785             Fp               = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
 786             FF               = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
 787             fvdw12           = _mm_mul_pd(c12_00,FF);
 788             fvdw             = _mm_xor_pd(signbit,_mm_mul_pd(_mm_add_pd(fvdw6,fvdw12),_mm_mul_pd(vftabscale,rinv00)));
 789
 790             fscal            = _mm_add_pd(felec,fvdw);
 791
 792             /* Calculate temporary vectorial force */
 793             tx               = _mm_mul_pd(fscal,dx00);
 794             ty               = _mm_mul_pd(fscal,dy00);
 795             tz               = _mm_mul_pd(fscal,dz00);
 796
 797             /* Update vectorial force */
 798             fix0             = _mm_add_pd(fix0,tx);
 799             fiy0             = _mm_add_pd(fiy0,ty);
 800             fiz0             = _mm_add_pd(fiz0,tz);
 801
 802             fjx0             = _mm_add_pd(fjx0,tx);
 803             fjy0             = _mm_add_pd(fjy0,ty);
 804             fjz0             = _mm_add_pd(fjz0,tz);
 805
 806             /**************************
 807              * CALCULATE INTERACTIONS *
 808              **************************/
 809
 810             /* Compute parameters for interactions between i and j atoms */
 811             qq10             = _mm_mul_pd(iq1,jq0);
 812
 813             /* REACTION-FIELD ELECTROSTATICS */
 814             felec            = _mm_mul_pd(qq10,_mm_sub_pd(_mm_mul_pd(rinv10,rinvsq10),krf2));
 815
 816             fscal            = felec;
 817
 818             /* Calculate temporary vectorial force */
 819             tx               = _mm_mul_pd(fscal,dx10);
 820             ty               = _mm_mul_pd(fscal,dy10);
 821             tz               = _mm_mul_pd(fscal,dz10);
 822
 823             /* Update vectorial force */
 824             fix1             = _mm_add_pd(fix1,tx);
 825             fiy1             = _mm_add_pd(fiy1,ty);
 826             fiz1             = _mm_add_pd(fiz1,tz);
 827
 828             fjx0             = _mm_add_pd(fjx0,tx);
 829             fjy0             = _mm_add_pd(fjy0,ty);
 830             fjz0             = _mm_add_pd(fjz0,tz);
 831
 832             /**************************
 833              * CALCULATE INTERACTIONS *
 834              **************************/
 835
 836             /* Compute parameters for interactions between i and j atoms */
 837             qq20             = _mm_mul_pd(iq2,jq0);
 838
 839             /* REACTION-FIELD ELECTROSTATICS */
 840             felec            = _mm_mul_pd(qq20,_mm_sub_pd(_mm_mul_pd(rinv20,rinvsq20),krf2));
 841
 842             fscal            = felec;
 843
 844             /* Calculate temporary vectorial force */
 845             tx               = _mm_mul_pd(fscal,dx20);
 846             ty               = _mm_mul_pd(fscal,dy20);
 847             tz               = _mm_mul_pd(fscal,dz20);
 848
 849             /* Update vectorial force */
 850             fix2             = _mm_add_pd(fix2,tx);
 851             fiy2             = _mm_add_pd(fiy2,ty);
 852             fiz2             = _mm_add_pd(fiz2,tz);
 853
 854             fjx0             = _mm_add_pd(fjx0,tx);
 855             fjy0             = _mm_add_pd(fjy0,ty);
 856             fjz0             = _mm_add_pd(fjz0,tz);
 857
 858             gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,fjx0,fjy0,fjz0);
 859
 860             /* Inner loop uses 111 flops */
 861         }
 862
 863         if(jidx<j_index_end)
 864         {
 865
 866             jnrA             = jjnr[jidx];
 867             j_coord_offsetA  = DIM*jnrA;
 868
 869             /* load j atom coordinates */
 870             gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
 871                                               &jx0,&jy0,&jz0);
 872
 873             /* Calculate displacement vector */
 874             dx00             = _mm_sub_pd(ix0,jx0);
 875             dy00             = _mm_sub_pd(iy0,jy0);
 876             dz00             = _mm_sub_pd(iz0,jz0);
 877             dx10             = _mm_sub_pd(ix1,jx0);
 878             dy10             = _mm_sub_pd(iy1,jy0);
 879             dz10             = _mm_sub_pd(iz1,jz0);
 880             dx20             = _mm_sub_pd(ix2,jx0);
 881             dy20             = _mm_sub_pd(iy2,jy0);
 882             dz20             = _mm_sub_pd(iz2,jz0);
 883
 884             /* Calculate squared distance and things based on it */
 885             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 886             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
 887             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
 888
 889             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 890             rinv10           = gmx_mm_invsqrt_pd(rsq10);
 891             rinv20           = gmx_mm_invsqrt_pd(rsq20);
 892
 893             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 894             rinvsq10         = _mm_mul_pd(rinv10,rinv10);
 895             rinvsq20         = _mm_mul_pd(rinv20,rinv20);
 896
 897             /* Load parameters for j particles */
 898             jq0              = _mm_load_sd(charge+jnrA+0);
 899             vdwjidx0A        = 2*vdwtype[jnrA+0];
 900
 901             fjx0             = _mm_setzero_pd();
 902             fjy0             = _mm_setzero_pd();
 903             fjz0             = _mm_setzero_pd();
 904
 905             /**************************
 906              * CALCULATE INTERACTIONS *
 907              **************************/
 908
 909             r00              = _mm_mul_pd(rsq00,rinv00);
 910
 911             /* Compute parameters for interactions between i and j atoms */
 912             qq00             = _mm_mul_pd(iq0,jq0);
 913             gmx_mm_load_1pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,&c6_00,&c12_00);
 914
 915             /* Calculate table index by multiplying r with table scale and truncate to integer */
 916             rt               = _mm_mul_pd(r00,vftabscale);
 917             vfitab           = _mm_cvttpd_epi32(rt);
 918             vfeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
 919             vfitab           = _mm_slli_epi32(vfitab,3);
 920
 921             /* REACTION-FIELD ELECTROSTATICS */
 922             felec            = _mm_mul_pd(qq00,_mm_sub_pd(_mm_mul_pd(rinv00,rinvsq00),krf2));
 923
 924             /* CUBIC SPLINE TABLE DISPERSION */
 925             Y                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
 926             F                = _mm_setzero_pd();
 927             GMX_MM_TRANSPOSE2_PD(Y,F);
 928             G                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
 929             H                = _mm_setzero_pd();
 930             GMX_MM_TRANSPOSE2_PD(G,H);
 931             Heps             = _mm_mul_pd(vfeps,H);
 932             Fp               = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
 933             FF               = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
 934             fvdw6            = _mm_mul_pd(c6_00,FF);
 935
 936             /* CUBIC SPLINE TABLE REPULSION */
 937             vfitab           = _mm_add_epi32(vfitab,ifour);
 938             Y                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
 939             F                = _mm_setzero_pd();
 940             GMX_MM_TRANSPOSE2_PD(Y,F);
 941             G                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
 942             H                = _mm_setzero_pd();
 943             GMX_MM_TRANSPOSE2_PD(G,H);
 944             Heps             = _mm_mul_pd(vfeps,H);
 945             Fp               = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
 946             FF               = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
 947             fvdw12           = _mm_mul_pd(c12_00,FF);
 948             fvdw             = _mm_xor_pd(signbit,_mm_mul_pd(_mm_add_pd(fvdw6,fvdw12),_mm_mul_pd(vftabscale,rinv00)));
 949
 950             fscal            = _mm_add_pd(felec,fvdw);
 951
 952             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 953
 954             /* Calculate temporary vectorial force */
 955             tx               = _mm_mul_pd(fscal,dx00);
 956             ty               = _mm_mul_pd(fscal,dy00);
 957             tz               = _mm_mul_pd(fscal,dz00);
 958
 959             /* Update vectorial force */
 960             fix0             = _mm_add_pd(fix0,tx);
 961             fiy0             = _mm_add_pd(fiy0,ty);
 962             fiz0             = _mm_add_pd(fiz0,tz);
 963
 964             fjx0             = _mm_add_pd(fjx0,tx);
 965             fjy0             = _mm_add_pd(fjy0,ty);
 966             fjz0             = _mm_add_pd(fjz0,tz);
 967
 968             /**************************
 969              * CALCULATE INTERACTIONS *
 970              **************************/
 971
 972             /* Compute parameters for interactions between i and j atoms */
 973             qq10             = _mm_mul_pd(iq1,jq0);
 974
 975             /* REACTION-FIELD ELECTROSTATICS */
 976             felec            = _mm_mul_pd(qq10,_mm_sub_pd(_mm_mul_pd(rinv10,rinvsq10),krf2));
 977
 978             fscal            = felec;
 979
 980             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 981
 982             /* Calculate temporary vectorial force */
 983             tx               = _mm_mul_pd(fscal,dx10);
 984             ty               = _mm_mul_pd(fscal,dy10);
 985             tz               = _mm_mul_pd(fscal,dz10);
 986
 987             /* Update vectorial force */
 988             fix1             = _mm_add_pd(fix1,tx);
 989             fiy1             = _mm_add_pd(fiy1,ty);
 990             fiz1             = _mm_add_pd(fiz1,tz);
 991
 992             fjx0             = _mm_add_pd(fjx0,tx);
 993             fjy0             = _mm_add_pd(fjy0,ty);
 994             fjz0             = _mm_add_pd(fjz0,tz);
 995
 996             /**************************
 997              * CALCULATE INTERACTIONS *
 998              **************************/
 999
1000             /* Compute parameters for interactions between i and j atoms */
1001             qq20             = _mm_mul_pd(iq2,jq0);
1002
1003             /* REACTION-FIELD ELECTROSTATICS */
1004             felec            = _mm_mul_pd(qq20,_mm_sub_pd(_mm_mul_pd(rinv20,rinvsq20),krf2));
1005
1006             fscal            = felec;
1007
1008             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1009
1010             /* Calculate temporary vectorial force */
1011             tx               = _mm_mul_pd(fscal,dx20);
1012             ty               = _mm_mul_pd(fscal,dy20);
1013             tz               = _mm_mul_pd(fscal,dz20);
1014
1015             /* Update vectorial force */
1016             fix2             = _mm_add_pd(fix2,tx);
1017             fiy2             = _mm_add_pd(fiy2,ty);
1018             fiz2             = _mm_add_pd(fiz2,tz);
1019
1020             fjx0             = _mm_add_pd(fjx0,tx);
1021             fjy0             = _mm_add_pd(fjy0,ty);
1022             fjz0             = _mm_add_pd(fjz0,tz);
1023
1024             gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,fjx0,fjy0,fjz0);
1025
1026             /* Inner loop uses 111 flops */
1027         }
1028
1029         /* End of innermost loop */
1030
1031         gmx_mm_update_iforce_3atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
1032                                               f+i_coord_offset,fshift+i_shift_offset);
1033
1034         /* Increment number of inner iterations */
1035         inneriter                  += j_index_end - j_index_start;
1036
1037         /* Outer loop uses 18 flops */
1038     }
1039
1040     /* Increment number of outer iterations */
1041     outeriter        += nri;
1042
1043     /* Update outer/inner flops */
1044
1045     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3_F,outeriter*18 + inneriter*111);
1046 }