src/gromacs/gmxlib/nonbonded/nb_kernel_sse2_double/nb_kernel_ElecRFCut_VdwCSTab_GeomW4P1_sse2_double.cpp

   1 /*
   2  * This file is part of the GROMACS molecular simulation package.
   3  *
   4  * Copyright (c) 2012,2013,2014,2015,2017,2018, 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 sse2_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/gmxlib/nrnb.h"
  46
  47 #include "kernelutil_x86_sse2_double.h"
  48
  49 /*
  50  * Gromacs nonbonded kernel:   nb_kernel_ElecRFCut_VdwCSTab_GeomW4P1_VF_sse2_double
  51  * Electrostatics interaction: ReactionField
  52  * VdW interaction:            CubicSplineTable
  53  * Geometry:                   Water4-Particle
  54  * Calculate force/pot:        PotentialAndForce
  55  */
  56 void
  57 nb_kernel_ElecRFCut_VdwCSTab_GeomW4P1_VF_sse2_double
  58                     (t_nblist                    * gmx_restrict       nlist,
  59                      rvec                        * gmx_restrict          xx,
  60                      rvec                        * gmx_restrict          ff,
  61                      struct t_forcerec           * gmx_restrict          fr,
  62                      t_mdatoms                   * gmx_restrict     mdatoms,
  63                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
  64                      t_nrnb                      * gmx_restrict        nrnb)
  65 {
  66     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
  67      * just 0 for non-waters.
  68      * Suffixes A,B refer to j loop unrolling done with SSE double precision, e.g. for the two different
  69      * jnr indices corresponding to data put in the four positions in the SIMD register.
  70      */
  71     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
  72     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
  73     int              jnrA,jnrB;
  74     int              j_coord_offsetA,j_coord_offsetB;
  75     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
  76     real             rcutoff_scalar;
  77     real             *shiftvec,*fshift,*x,*f;
  78     __m128d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
  79     int              vdwioffset0;
  80     __m128d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
  81     int              vdwioffset1;
  82     __m128d          ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
  83     int              vdwioffset2;
  84     __m128d          ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
  85     int              vdwioffset3;
  86     __m128d          ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
  87     int              vdwjidx0A,vdwjidx0B;
  88     __m128d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  89     __m128d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  90     __m128d          dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
  91     __m128d          dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
  92     __m128d          dx30,dy30,dz30,rsq30,rinv30,rinvsq30,r30,qq30,c6_30,c12_30;
  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->ic->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     iq1              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+1]));
 135     iq2              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+2]));
 136     iq3              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+3]));
 137     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 138
 139     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 140     rcutoff_scalar   = fr->ic->rcoulomb;
 141     rcutoff          = _mm_set1_pd(rcutoff_scalar);
 142     rcutoff2         = _mm_mul_pd(rcutoff,rcutoff);
 143
 144     /* Avoid stupid compiler warnings */
 145     jnrA = jnrB = 0;
 146     j_coord_offsetA = 0;
 147     j_coord_offsetB = 0;
 148
 149     outeriter        = 0;
 150     inneriter        = 0;
 151
 152     /* Start outer loop over neighborlists */
 153     for(iidx=0; iidx<nri; iidx++)
 154     {
 155         /* Load shift vector for this list */
 156         i_shift_offset   = DIM*shiftidx[iidx];
 157
 158         /* Load limits for loop over neighbors */
 159         j_index_start    = jindex[iidx];
 160         j_index_end      = jindex[iidx+1];
 161
 162         /* Get outer coordinate index */
 163         inr              = iinr[iidx];
 164         i_coord_offset   = DIM*inr;
 165
 166         /* Load i particle coords and add shift vector */
 167         gmx_mm_load_shift_and_4rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
 168                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
 169
 170         fix0             = _mm_setzero_pd();
 171         fiy0             = _mm_setzero_pd();
 172         fiz0             = _mm_setzero_pd();
 173         fix1             = _mm_setzero_pd();
 174         fiy1             = _mm_setzero_pd();
 175         fiz1             = _mm_setzero_pd();
 176         fix2             = _mm_setzero_pd();
 177         fiy2             = _mm_setzero_pd();
 178         fiz2             = _mm_setzero_pd();
 179         fix3             = _mm_setzero_pd();
 180         fiy3             = _mm_setzero_pd();
 181         fiz3             = _mm_setzero_pd();
 182
 183         /* Reset potential sums */
 184         velecsum         = _mm_setzero_pd();
 185         vvdwsum          = _mm_setzero_pd();
 186
 187         /* Start inner kernel loop */
 188         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 189         {
 190
 191             /* Get j neighbor index, and coordinate index */
 192             jnrA             = jjnr[jidx];
 193             jnrB             = jjnr[jidx+1];
 194             j_coord_offsetA  = DIM*jnrA;
 195             j_coord_offsetB  = DIM*jnrB;
 196
 197             /* load j atom coordinates */
 198             gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 199                                               &jx0,&jy0,&jz0);
 200
 201             /* Calculate displacement vector */
 202             dx00             = _mm_sub_pd(ix0,jx0);
 203             dy00             = _mm_sub_pd(iy0,jy0);
 204             dz00             = _mm_sub_pd(iz0,jz0);
 205             dx10             = _mm_sub_pd(ix1,jx0);
 206             dy10             = _mm_sub_pd(iy1,jy0);
 207             dz10             = _mm_sub_pd(iz1,jz0);
 208             dx20             = _mm_sub_pd(ix2,jx0);
 209             dy20             = _mm_sub_pd(iy2,jy0);
 210             dz20             = _mm_sub_pd(iz2,jz0);
 211             dx30             = _mm_sub_pd(ix3,jx0);
 212             dy30             = _mm_sub_pd(iy3,jy0);
 213             dz30             = _mm_sub_pd(iz3,jz0);
 214
 215             /* Calculate squared distance and things based on it */
 216             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 217             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
 218             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
 219             rsq30            = gmx_mm_calc_rsq_pd(dx30,dy30,dz30);
 220
 221             rinv00           = sse2_invsqrt_d(rsq00);
 222             rinv10           = sse2_invsqrt_d(rsq10);
 223             rinv20           = sse2_invsqrt_d(rsq20);
 224             rinv30           = sse2_invsqrt_d(rsq30);
 225
 226             rinvsq10         = _mm_mul_pd(rinv10,rinv10);
 227             rinvsq20         = _mm_mul_pd(rinv20,rinv20);
 228             rinvsq30         = _mm_mul_pd(rinv30,rinv30);
 229
 230             /* Load parameters for j particles */
 231             jq0              = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
 232             vdwjidx0A        = 2*vdwtype[jnrA+0];
 233             vdwjidx0B        = 2*vdwtype[jnrB+0];
 234
 235             fjx0             = _mm_setzero_pd();
 236             fjy0             = _mm_setzero_pd();
 237             fjz0             = _mm_setzero_pd();
 238
 239             /**************************
 240              * CALCULATE INTERACTIONS *
 241              **************************/
 242
 243             r00              = _mm_mul_pd(rsq00,rinv00);
 244
 245             /* Compute parameters for interactions between i and j atoms */
 246             gmx_mm_load_2pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,
 247                                          vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
 248
 249             /* Calculate table index by multiplying r with table scale and truncate to integer */
 250             rt               = _mm_mul_pd(r00,vftabscale);
 251             vfitab           = _mm_cvttpd_epi32(rt);
 252             vfeps            = _mm_sub_pd(rt,_mm_cvtepi32_pd(vfitab));
 253             vfitab           = _mm_slli_epi32(vfitab,3);
 254
 255             /* CUBIC SPLINE TABLE DISPERSION */
 256             Y                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
 257             F                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) );
 258             GMX_MM_TRANSPOSE2_PD(Y,F);
 259             G                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
 260             H                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) +2);
 261             GMX_MM_TRANSPOSE2_PD(G,H);
 262             Heps             = _mm_mul_pd(vfeps,H);
 263             Fp               = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
 264             VV               = _mm_add_pd(Y,_mm_mul_pd(vfeps,Fp));
 265             vvdw6            = _mm_mul_pd(c6_00,VV);
 266             FF               = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
 267             fvdw6            = _mm_mul_pd(c6_00,FF);
 268
 269             /* CUBIC SPLINE TABLE REPULSION */
 270             vfitab           = _mm_add_epi32(vfitab,ifour);
 271             Y                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
 272             F                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) );
 273             GMX_MM_TRANSPOSE2_PD(Y,F);
 274             G                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
 275             H                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) +2);
 276             GMX_MM_TRANSPOSE2_PD(G,H);
 277             Heps             = _mm_mul_pd(vfeps,H);
 278             Fp               = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
 279             VV               = _mm_add_pd(Y,_mm_mul_pd(vfeps,Fp));
 280             vvdw12           = _mm_mul_pd(c12_00,VV);
 281             FF               = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
 282             fvdw12           = _mm_mul_pd(c12_00,FF);
 283             vvdw             = _mm_add_pd(vvdw12,vvdw6);
 284             fvdw             = _mm_xor_pd(signbit,_mm_mul_pd(_mm_add_pd(fvdw6,fvdw12),_mm_mul_pd(vftabscale,rinv00)));
 285
 286             /* Update potential sum for this i atom from the interaction with this j atom. */
 287             vvdwsum          = _mm_add_pd(vvdwsum,vvdw);
 288
 289             fscal            = fvdw;
 290
 291             /* Calculate temporary vectorial force */
 292             tx               = _mm_mul_pd(fscal,dx00);
 293             ty               = _mm_mul_pd(fscal,dy00);
 294             tz               = _mm_mul_pd(fscal,dz00);
 295
 296             /* Update vectorial force */
 297             fix0             = _mm_add_pd(fix0,tx);
 298             fiy0             = _mm_add_pd(fiy0,ty);
 299             fiz0             = _mm_add_pd(fiz0,tz);
 300
 301             fjx0             = _mm_add_pd(fjx0,tx);
 302             fjy0             = _mm_add_pd(fjy0,ty);
 303             fjz0             = _mm_add_pd(fjz0,tz);
 304
 305             /**************************
 306              * CALCULATE INTERACTIONS *
 307              **************************/
 308
 309             if (gmx_mm_any_lt(rsq10,rcutoff2))
 310             {
 311
 312             /* Compute parameters for interactions between i and j atoms */
 313             qq10             = _mm_mul_pd(iq1,jq0);
 314
 315             /* REACTION-FIELD ELECTROSTATICS */
 316             velec            = _mm_mul_pd(qq10,_mm_sub_pd(_mm_add_pd(rinv10,_mm_mul_pd(krf,rsq10)),crf));
 317             felec            = _mm_mul_pd(qq10,_mm_sub_pd(_mm_mul_pd(rinv10,rinvsq10),krf2));
 318
 319             cutoff_mask      = _mm_cmplt_pd(rsq10,rcutoff2);
 320
 321             /* Update potential sum for this i atom from the interaction with this j atom. */
 322             velec            = _mm_and_pd(velec,cutoff_mask);
 323             velecsum         = _mm_add_pd(velecsum,velec);
 324
 325             fscal            = felec;
 326
 327             fscal            = _mm_and_pd(fscal,cutoff_mask);
 328
 329             /* Calculate temporary vectorial force */
 330             tx               = _mm_mul_pd(fscal,dx10);
 331             ty               = _mm_mul_pd(fscal,dy10);
 332             tz               = _mm_mul_pd(fscal,dz10);
 333
 334             /* Update vectorial force */
 335             fix1             = _mm_add_pd(fix1,tx);
 336             fiy1             = _mm_add_pd(fiy1,ty);
 337             fiz1             = _mm_add_pd(fiz1,tz);
 338
 339             fjx0             = _mm_add_pd(fjx0,tx);
 340             fjy0             = _mm_add_pd(fjy0,ty);
 341             fjz0             = _mm_add_pd(fjz0,tz);
 342
 343             }
 344
 345             /**************************
 346              * CALCULATE INTERACTIONS *
 347              **************************/
 348
 349             if (gmx_mm_any_lt(rsq20,rcutoff2))
 350             {
 351
 352             /* Compute parameters for interactions between i and j atoms */
 353             qq20             = _mm_mul_pd(iq2,jq0);
 354
 355             /* REACTION-FIELD ELECTROSTATICS */
 356             velec            = _mm_mul_pd(qq20,_mm_sub_pd(_mm_add_pd(rinv20,_mm_mul_pd(krf,rsq20)),crf));
 357             felec            = _mm_mul_pd(qq20,_mm_sub_pd(_mm_mul_pd(rinv20,rinvsq20),krf2));
 358
 359             cutoff_mask      = _mm_cmplt_pd(rsq20,rcutoff2);
 360
 361             /* Update potential sum for this i atom from the interaction with this j atom. */
 362             velec            = _mm_and_pd(velec,cutoff_mask);
 363             velecsum         = _mm_add_pd(velecsum,velec);
 364
 365             fscal            = felec;
 366
 367             fscal            = _mm_and_pd(fscal,cutoff_mask);
 368
 369             /* Calculate temporary vectorial force */
 370             tx               = _mm_mul_pd(fscal,dx20);
 371             ty               = _mm_mul_pd(fscal,dy20);
 372             tz               = _mm_mul_pd(fscal,dz20);
 373
 374             /* Update vectorial force */
 375             fix2             = _mm_add_pd(fix2,tx);
 376             fiy2             = _mm_add_pd(fiy2,ty);
 377             fiz2             = _mm_add_pd(fiz2,tz);
 378
 379             fjx0             = _mm_add_pd(fjx0,tx);
 380             fjy0             = _mm_add_pd(fjy0,ty);
 381             fjz0             = _mm_add_pd(fjz0,tz);
 382
 383             }
 384
 385             /**************************
 386              * CALCULATE INTERACTIONS *
 387              **************************/
 388
 389             if (gmx_mm_any_lt(rsq30,rcutoff2))
 390             {
 391
 392             /* Compute parameters for interactions between i and j atoms */
 393             qq30             = _mm_mul_pd(iq3,jq0);
 394
 395             /* REACTION-FIELD ELECTROSTATICS */
 396             velec            = _mm_mul_pd(qq30,_mm_sub_pd(_mm_add_pd(rinv30,_mm_mul_pd(krf,rsq30)),crf));
 397             felec            = _mm_mul_pd(qq30,_mm_sub_pd(_mm_mul_pd(rinv30,rinvsq30),krf2));
 398
 399             cutoff_mask      = _mm_cmplt_pd(rsq30,rcutoff2);
 400
 401             /* Update potential sum for this i atom from the interaction with this j atom. */
 402             velec            = _mm_and_pd(velec,cutoff_mask);
 403             velecsum         = _mm_add_pd(velecsum,velec);
 404
 405             fscal            = felec;
 406
 407             fscal            = _mm_and_pd(fscal,cutoff_mask);
 408
 409             /* Calculate temporary vectorial force */
 410             tx               = _mm_mul_pd(fscal,dx30);
 411             ty               = _mm_mul_pd(fscal,dy30);
 412             tz               = _mm_mul_pd(fscal,dz30);
 413
 414             /* Update vectorial force */
 415             fix3             = _mm_add_pd(fix3,tx);
 416             fiy3             = _mm_add_pd(fiy3,ty);
 417             fiz3             = _mm_add_pd(fiz3,tz);
 418
 419             fjx0             = _mm_add_pd(fjx0,tx);
 420             fjy0             = _mm_add_pd(fjy0,ty);
 421             fjz0             = _mm_add_pd(fjz0,tz);
 422
 423             }
 424
 425             gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,fjx0,fjy0,fjz0);
 426
 427             /* Inner loop uses 167 flops */
 428         }
 429
 430         if(jidx<j_index_end)
 431         {
 432
 433             jnrA             = jjnr[jidx];
 434             j_coord_offsetA  = DIM*jnrA;
 435
 436             /* load j atom coordinates */
 437             gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
 438                                               &jx0,&jy0,&jz0);
 439
 440             /* Calculate displacement vector */
 441             dx00             = _mm_sub_pd(ix0,jx0);
 442             dy00             = _mm_sub_pd(iy0,jy0);
 443             dz00             = _mm_sub_pd(iz0,jz0);
 444             dx10             = _mm_sub_pd(ix1,jx0);
 445             dy10             = _mm_sub_pd(iy1,jy0);
 446             dz10             = _mm_sub_pd(iz1,jz0);
 447             dx20             = _mm_sub_pd(ix2,jx0);
 448             dy20             = _mm_sub_pd(iy2,jy0);
 449             dz20             = _mm_sub_pd(iz2,jz0);
 450             dx30             = _mm_sub_pd(ix3,jx0);
 451             dy30             = _mm_sub_pd(iy3,jy0);
 452             dz30             = _mm_sub_pd(iz3,jz0);
 453
 454             /* Calculate squared distance and things based on it */
 455             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 456             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
 457             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
 458             rsq30            = gmx_mm_calc_rsq_pd(dx30,dy30,dz30);
 459
 460             rinv00           = sse2_invsqrt_d(rsq00);
 461             rinv10           = sse2_invsqrt_d(rsq10);
 462             rinv20           = sse2_invsqrt_d(rsq20);
 463             rinv30           = sse2_invsqrt_d(rsq30);
 464
 465             rinvsq10         = _mm_mul_pd(rinv10,rinv10);
 466             rinvsq20         = _mm_mul_pd(rinv20,rinv20);
 467             rinvsq30         = _mm_mul_pd(rinv30,rinv30);
 468
 469             /* Load parameters for j particles */
 470             jq0              = _mm_load_sd(charge+jnrA+0);
 471             vdwjidx0A        = 2*vdwtype[jnrA+0];
 472
 473             fjx0             = _mm_setzero_pd();
 474             fjy0             = _mm_setzero_pd();
 475             fjz0             = _mm_setzero_pd();
 476
 477             /**************************
 478              * CALCULATE INTERACTIONS *
 479              **************************/
 480
 481             r00              = _mm_mul_pd(rsq00,rinv00);
 482
 483             /* Compute parameters for interactions between i and j atoms */
 484             gmx_mm_load_1pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,&c6_00,&c12_00);
 485
 486             /* Calculate table index by multiplying r with table scale and truncate to integer */
 487             rt               = _mm_mul_pd(r00,vftabscale);
 488             vfitab           = _mm_cvttpd_epi32(rt);
 489             vfeps            = _mm_sub_pd(rt,_mm_cvtepi32_pd(vfitab));
 490             vfitab           = _mm_slli_epi32(vfitab,3);
 491
 492             /* CUBIC SPLINE TABLE DISPERSION */
 493             Y                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
 494             F                = _mm_setzero_pd();
 495             GMX_MM_TRANSPOSE2_PD(Y,F);
 496             G                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
 497             H                = _mm_setzero_pd();
 498             GMX_MM_TRANSPOSE2_PD(G,H);
 499             Heps             = _mm_mul_pd(vfeps,H);
 500             Fp               = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
 501             VV               = _mm_add_pd(Y,_mm_mul_pd(vfeps,Fp));
 502             vvdw6            = _mm_mul_pd(c6_00,VV);
 503             FF               = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
 504             fvdw6            = _mm_mul_pd(c6_00,FF);
 505
 506             /* CUBIC SPLINE TABLE REPULSION */
 507             vfitab           = _mm_add_epi32(vfitab,ifour);
 508             Y                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
 509             F                = _mm_setzero_pd();
 510             GMX_MM_TRANSPOSE2_PD(Y,F);
 511             G                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
 512             H                = _mm_setzero_pd();
 513             GMX_MM_TRANSPOSE2_PD(G,H);
 514             Heps             = _mm_mul_pd(vfeps,H);
 515             Fp               = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
 516             VV               = _mm_add_pd(Y,_mm_mul_pd(vfeps,Fp));
 517             vvdw12           = _mm_mul_pd(c12_00,VV);
 518             FF               = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
 519             fvdw12           = _mm_mul_pd(c12_00,FF);
 520             vvdw             = _mm_add_pd(vvdw12,vvdw6);
 521             fvdw             = _mm_xor_pd(signbit,_mm_mul_pd(_mm_add_pd(fvdw6,fvdw12),_mm_mul_pd(vftabscale,rinv00)));
 522
 523             /* Update potential sum for this i atom from the interaction with this j atom. */
 524             vvdw             = _mm_unpacklo_pd(vvdw,_mm_setzero_pd());
 525             vvdwsum          = _mm_add_pd(vvdwsum,vvdw);
 526
 527             fscal            = fvdw;
 528
 529             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 530
 531             /* Calculate temporary vectorial force */
 532             tx               = _mm_mul_pd(fscal,dx00);
 533             ty               = _mm_mul_pd(fscal,dy00);
 534             tz               = _mm_mul_pd(fscal,dz00);
 535
 536             /* Update vectorial force */
 537             fix0             = _mm_add_pd(fix0,tx);
 538             fiy0             = _mm_add_pd(fiy0,ty);
 539             fiz0             = _mm_add_pd(fiz0,tz);
 540
 541             fjx0             = _mm_add_pd(fjx0,tx);
 542             fjy0             = _mm_add_pd(fjy0,ty);
 543             fjz0             = _mm_add_pd(fjz0,tz);
 544
 545             /**************************
 546              * CALCULATE INTERACTIONS *
 547              **************************/
 548
 549             if (gmx_mm_any_lt(rsq10,rcutoff2))
 550             {
 551
 552             /* Compute parameters for interactions between i and j atoms */
 553             qq10             = _mm_mul_pd(iq1,jq0);
 554
 555             /* REACTION-FIELD ELECTROSTATICS */
 556             velec            = _mm_mul_pd(qq10,_mm_sub_pd(_mm_add_pd(rinv10,_mm_mul_pd(krf,rsq10)),crf));
 557             felec            = _mm_mul_pd(qq10,_mm_sub_pd(_mm_mul_pd(rinv10,rinvsq10),krf2));
 558
 559             cutoff_mask      = _mm_cmplt_pd(rsq10,rcutoff2);
 560
 561             /* Update potential sum for this i atom from the interaction with this j atom. */
 562             velec            = _mm_and_pd(velec,cutoff_mask);
 563             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 564             velecsum         = _mm_add_pd(velecsum,velec);
 565
 566             fscal            = felec;
 567
 568             fscal            = _mm_and_pd(fscal,cutoff_mask);
 569
 570             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 571
 572             /* Calculate temporary vectorial force */
 573             tx               = _mm_mul_pd(fscal,dx10);
 574             ty               = _mm_mul_pd(fscal,dy10);
 575             tz               = _mm_mul_pd(fscal,dz10);
 576
 577             /* Update vectorial force */
 578             fix1             = _mm_add_pd(fix1,tx);
 579             fiy1             = _mm_add_pd(fiy1,ty);
 580             fiz1             = _mm_add_pd(fiz1,tz);
 581
 582             fjx0             = _mm_add_pd(fjx0,tx);
 583             fjy0             = _mm_add_pd(fjy0,ty);
 584             fjz0             = _mm_add_pd(fjz0,tz);
 585
 586             }
 587
 588             /**************************
 589              * CALCULATE INTERACTIONS *
 590              **************************/
 591
 592             if (gmx_mm_any_lt(rsq20,rcutoff2))
 593             {
 594
 595             /* Compute parameters for interactions between i and j atoms */
 596             qq20             = _mm_mul_pd(iq2,jq0);
 597
 598             /* REACTION-FIELD ELECTROSTATICS */
 599             velec            = _mm_mul_pd(qq20,_mm_sub_pd(_mm_add_pd(rinv20,_mm_mul_pd(krf,rsq20)),crf));
 600             felec            = _mm_mul_pd(qq20,_mm_sub_pd(_mm_mul_pd(rinv20,rinvsq20),krf2));
 601
 602             cutoff_mask      = _mm_cmplt_pd(rsq20,rcutoff2);
 603
 604             /* Update potential sum for this i atom from the interaction with this j atom. */
 605             velec            = _mm_and_pd(velec,cutoff_mask);
 606             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 607             velecsum         = _mm_add_pd(velecsum,velec);
 608
 609             fscal            = felec;
 610
 611             fscal            = _mm_and_pd(fscal,cutoff_mask);
 612
 613             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 614
 615             /* Calculate temporary vectorial force */
 616             tx               = _mm_mul_pd(fscal,dx20);
 617             ty               = _mm_mul_pd(fscal,dy20);
 618             tz               = _mm_mul_pd(fscal,dz20);
 619
 620             /* Update vectorial force */
 621             fix2             = _mm_add_pd(fix2,tx);
 622             fiy2             = _mm_add_pd(fiy2,ty);
 623             fiz2             = _mm_add_pd(fiz2,tz);
 624
 625             fjx0             = _mm_add_pd(fjx0,tx);
 626             fjy0             = _mm_add_pd(fjy0,ty);
 627             fjz0             = _mm_add_pd(fjz0,tz);
 628
 629             }
 630
 631             /**************************
 632              * CALCULATE INTERACTIONS *
 633              **************************/
 634
 635             if (gmx_mm_any_lt(rsq30,rcutoff2))
 636             {
 637
 638             /* Compute parameters for interactions between i and j atoms */
 639             qq30             = _mm_mul_pd(iq3,jq0);
 640
 641             /* REACTION-FIELD ELECTROSTATICS */
 642             velec            = _mm_mul_pd(qq30,_mm_sub_pd(_mm_add_pd(rinv30,_mm_mul_pd(krf,rsq30)),crf));
 643             felec            = _mm_mul_pd(qq30,_mm_sub_pd(_mm_mul_pd(rinv30,rinvsq30),krf2));
 644
 645             cutoff_mask      = _mm_cmplt_pd(rsq30,rcutoff2);
 646
 647             /* Update potential sum for this i atom from the interaction with this j atom. */
 648             velec            = _mm_and_pd(velec,cutoff_mask);
 649             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 650             velecsum         = _mm_add_pd(velecsum,velec);
 651
 652             fscal            = felec;
 653
 654             fscal            = _mm_and_pd(fscal,cutoff_mask);
 655
 656             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 657
 658             /* Calculate temporary vectorial force */
 659             tx               = _mm_mul_pd(fscal,dx30);
 660             ty               = _mm_mul_pd(fscal,dy30);
 661             tz               = _mm_mul_pd(fscal,dz30);
 662
 663             /* Update vectorial force */
 664             fix3             = _mm_add_pd(fix3,tx);
 665             fiy3             = _mm_add_pd(fiy3,ty);
 666             fiz3             = _mm_add_pd(fiz3,tz);
 667
 668             fjx0             = _mm_add_pd(fjx0,tx);
 669             fjy0             = _mm_add_pd(fjy0,ty);
 670             fjz0             = _mm_add_pd(fjz0,tz);
 671
 672             }
 673
 674             gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,fjx0,fjy0,fjz0);
 675
 676             /* Inner loop uses 167 flops */
 677         }
 678
 679         /* End of innermost loop */
 680
 681         gmx_mm_update_iforce_4atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
 682                                               f+i_coord_offset,fshift+i_shift_offset);
 683
 684         ggid                        = gid[iidx];
 685         /* Update potential energies */
 686         gmx_mm_update_1pot_pd(velecsum,kernel_data->energygrp_elec+ggid);
 687         gmx_mm_update_1pot_pd(vvdwsum,kernel_data->energygrp_vdw+ggid);
 688
 689         /* Increment number of inner iterations */
 690         inneriter                  += j_index_end - j_index_start;
 691
 692         /* Outer loop uses 26 flops */
 693     }
 694
 695     /* Increment number of outer iterations */
 696     outeriter        += nri;
 697
 698     /* Update outer/inner flops */
 699
 700     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W4_VF,outeriter*26 + inneriter*167);
 701 }
 702 /*
 703  * Gromacs nonbonded kernel:   nb_kernel_ElecRFCut_VdwCSTab_GeomW4P1_F_sse2_double
 704  * Electrostatics interaction: ReactionField
 705  * VdW interaction:            CubicSplineTable
 706  * Geometry:                   Water4-Particle
 707  * Calculate force/pot:        Force
 708  */
 709 void
 710 nb_kernel_ElecRFCut_VdwCSTab_GeomW4P1_F_sse2_double
 711                     (t_nblist                    * gmx_restrict       nlist,
 712                      rvec                        * gmx_restrict          xx,
 713                      rvec                        * gmx_restrict          ff,
 714                      struct t_forcerec           * gmx_restrict          fr,
 715                      t_mdatoms                   * gmx_restrict     mdatoms,
 716                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 717                      t_nrnb                      * gmx_restrict        nrnb)
 718 {
 719     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 720      * just 0 for non-waters.
 721      * Suffixes A,B refer to j loop unrolling done with SSE double precision, e.g. for the two different
 722      * jnr indices corresponding to data put in the four positions in the SIMD register.
 723      */
 724     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 725     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 726     int              jnrA,jnrB;
 727     int              j_coord_offsetA,j_coord_offsetB;
 728     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 729     real             rcutoff_scalar;
 730     real             *shiftvec,*fshift,*x,*f;
 731     __m128d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 732     int              vdwioffset0;
 733     __m128d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 734     int              vdwioffset1;
 735     __m128d          ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
 736     int              vdwioffset2;
 737     __m128d          ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
 738     int              vdwioffset3;
 739     __m128d          ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
 740     int              vdwjidx0A,vdwjidx0B;
 741     __m128d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 742     __m128d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 743     __m128d          dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
 744     __m128d          dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
 745     __m128d          dx30,dy30,dz30,rsq30,rinv30,rinvsq30,r30,qq30,c6_30,c12_30;
 746     __m128d          velec,felec,velecsum,facel,crf,krf,krf2;
 747     real             *charge;
 748     int              nvdwtype;
 749     __m128d          rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 750     int              *vdwtype;
 751     real             *vdwparam;
 752     __m128d          one_sixth   = _mm_set1_pd(1.0/6.0);
 753     __m128d          one_twelfth = _mm_set1_pd(1.0/12.0);
 754     __m128i          vfitab;
 755     __m128i          ifour       = _mm_set1_epi32(4);
 756     __m128d          rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF;
 757     real             *vftab;
 758     __m128d          dummy_mask,cutoff_mask;
 759     __m128d          signbit   = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
 760     __m128d          one     = _mm_set1_pd(1.0);
 761     __m128d          two     = _mm_set1_pd(2.0);
 762     x                = xx[0];
 763     f                = ff[0];
 764
 765     nri              = nlist->nri;
 766     iinr             = nlist->iinr;
 767     jindex           = nlist->jindex;
 768     jjnr             = nlist->jjnr;
 769     shiftidx         = nlist->shift;
 770     gid              = nlist->gid;
 771     shiftvec         = fr->shift_vec[0];
 772     fshift           = fr->fshift[0];
 773     facel            = _mm_set1_pd(fr->ic->epsfac);
 774     charge           = mdatoms->chargeA;
 775     krf              = _mm_set1_pd(fr->ic->k_rf);
 776     krf2             = _mm_set1_pd(fr->ic->k_rf*2.0);
 777     crf              = _mm_set1_pd(fr->ic->c_rf);
 778     nvdwtype         = fr->ntype;
 779     vdwparam         = fr->nbfp;
 780     vdwtype          = mdatoms->typeA;
 781
 782     vftab            = kernel_data->table_vdw->data;
 783     vftabscale       = _mm_set1_pd(kernel_data->table_vdw->scale);
 784
 785     /* Setup water-specific parameters */
 786     inr              = nlist->iinr[0];
 787     iq1              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+1]));
 788     iq2              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+2]));
 789     iq3              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+3]));
 790     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 791
 792     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 793     rcutoff_scalar   = fr->ic->rcoulomb;
 794     rcutoff          = _mm_set1_pd(rcutoff_scalar);
 795     rcutoff2         = _mm_mul_pd(rcutoff,rcutoff);
 796
 797     /* Avoid stupid compiler warnings */
 798     jnrA = jnrB = 0;
 799     j_coord_offsetA = 0;
 800     j_coord_offsetB = 0;
 801
 802     outeriter        = 0;
 803     inneriter        = 0;
 804
 805     /* Start outer loop over neighborlists */
 806     for(iidx=0; iidx<nri; iidx++)
 807     {
 808         /* Load shift vector for this list */
 809         i_shift_offset   = DIM*shiftidx[iidx];
 810
 811         /* Load limits for loop over neighbors */
 812         j_index_start    = jindex[iidx];
 813         j_index_end      = jindex[iidx+1];
 814
 815         /* Get outer coordinate index */
 816         inr              = iinr[iidx];
 817         i_coord_offset   = DIM*inr;
 818
 819         /* Load i particle coords and add shift vector */
 820         gmx_mm_load_shift_and_4rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
 821                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
 822
 823         fix0             = _mm_setzero_pd();
 824         fiy0             = _mm_setzero_pd();
 825         fiz0             = _mm_setzero_pd();
 826         fix1             = _mm_setzero_pd();
 827         fiy1             = _mm_setzero_pd();
 828         fiz1             = _mm_setzero_pd();
 829         fix2             = _mm_setzero_pd();
 830         fiy2             = _mm_setzero_pd();
 831         fiz2             = _mm_setzero_pd();
 832         fix3             = _mm_setzero_pd();
 833         fiy3             = _mm_setzero_pd();
 834         fiz3             = _mm_setzero_pd();
 835
 836         /* Start inner kernel loop */
 837         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 838         {
 839
 840             /* Get j neighbor index, and coordinate index */
 841             jnrA             = jjnr[jidx];
 842             jnrB             = jjnr[jidx+1];
 843             j_coord_offsetA  = DIM*jnrA;
 844             j_coord_offsetB  = DIM*jnrB;
 845
 846             /* load j atom coordinates */
 847             gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 848                                               &jx0,&jy0,&jz0);
 849
 850             /* Calculate displacement vector */
 851             dx00             = _mm_sub_pd(ix0,jx0);
 852             dy00             = _mm_sub_pd(iy0,jy0);
 853             dz00             = _mm_sub_pd(iz0,jz0);
 854             dx10             = _mm_sub_pd(ix1,jx0);
 855             dy10             = _mm_sub_pd(iy1,jy0);
 856             dz10             = _mm_sub_pd(iz1,jz0);
 857             dx20             = _mm_sub_pd(ix2,jx0);
 858             dy20             = _mm_sub_pd(iy2,jy0);
 859             dz20             = _mm_sub_pd(iz2,jz0);
 860             dx30             = _mm_sub_pd(ix3,jx0);
 861             dy30             = _mm_sub_pd(iy3,jy0);
 862             dz30             = _mm_sub_pd(iz3,jz0);
 863
 864             /* Calculate squared distance and things based on it */
 865             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 866             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
 867             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
 868             rsq30            = gmx_mm_calc_rsq_pd(dx30,dy30,dz30);
 869
 870             rinv00           = sse2_invsqrt_d(rsq00);
 871             rinv10           = sse2_invsqrt_d(rsq10);
 872             rinv20           = sse2_invsqrt_d(rsq20);
 873             rinv30           = sse2_invsqrt_d(rsq30);
 874
 875             rinvsq10         = _mm_mul_pd(rinv10,rinv10);
 876             rinvsq20         = _mm_mul_pd(rinv20,rinv20);
 877             rinvsq30         = _mm_mul_pd(rinv30,rinv30);
 878
 879             /* Load parameters for j particles */
 880             jq0              = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
 881             vdwjidx0A        = 2*vdwtype[jnrA+0];
 882             vdwjidx0B        = 2*vdwtype[jnrB+0];
 883
 884             fjx0             = _mm_setzero_pd();
 885             fjy0             = _mm_setzero_pd();
 886             fjz0             = _mm_setzero_pd();
 887
 888             /**************************
 889              * CALCULATE INTERACTIONS *
 890              **************************/
 891
 892             r00              = _mm_mul_pd(rsq00,rinv00);
 893
 894             /* Compute parameters for interactions between i and j atoms */
 895             gmx_mm_load_2pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,
 896                                          vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
 897
 898             /* Calculate table index by multiplying r with table scale and truncate to integer */
 899             rt               = _mm_mul_pd(r00,vftabscale);
 900             vfitab           = _mm_cvttpd_epi32(rt);
 901             vfeps            = _mm_sub_pd(rt,_mm_cvtepi32_pd(vfitab));
 902             vfitab           = _mm_slli_epi32(vfitab,3);
 903
 904             /* CUBIC SPLINE TABLE DISPERSION */
 905             Y                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
 906             F                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) );
 907             GMX_MM_TRANSPOSE2_PD(Y,F);
 908             G                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
 909             H                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) +2);
 910             GMX_MM_TRANSPOSE2_PD(G,H);
 911             Heps             = _mm_mul_pd(vfeps,H);
 912             Fp               = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
 913             FF               = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
 914             fvdw6            = _mm_mul_pd(c6_00,FF);
 915
 916             /* CUBIC SPLINE TABLE REPULSION */
 917             vfitab           = _mm_add_epi32(vfitab,ifour);
 918             Y                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
 919             F                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) );
 920             GMX_MM_TRANSPOSE2_PD(Y,F);
 921             G                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
 922             H                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) +2);
 923             GMX_MM_TRANSPOSE2_PD(G,H);
 924             Heps             = _mm_mul_pd(vfeps,H);
 925             Fp               = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
 926             FF               = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
 927             fvdw12           = _mm_mul_pd(c12_00,FF);
 928             fvdw             = _mm_xor_pd(signbit,_mm_mul_pd(_mm_add_pd(fvdw6,fvdw12),_mm_mul_pd(vftabscale,rinv00)));
 929
 930             fscal            = fvdw;
 931
 932             /* Calculate temporary vectorial force */
 933             tx               = _mm_mul_pd(fscal,dx00);
 934             ty               = _mm_mul_pd(fscal,dy00);
 935             tz               = _mm_mul_pd(fscal,dz00);
 936
 937             /* Update vectorial force */
 938             fix0             = _mm_add_pd(fix0,tx);
 939             fiy0             = _mm_add_pd(fiy0,ty);
 940             fiz0             = _mm_add_pd(fiz0,tz);
 941
 942             fjx0             = _mm_add_pd(fjx0,tx);
 943             fjy0             = _mm_add_pd(fjy0,ty);
 944             fjz0             = _mm_add_pd(fjz0,tz);
 945
 946             /**************************
 947              * CALCULATE INTERACTIONS *
 948              **************************/
 949
 950             if (gmx_mm_any_lt(rsq10,rcutoff2))
 951             {
 952
 953             /* Compute parameters for interactions between i and j atoms */
 954             qq10             = _mm_mul_pd(iq1,jq0);
 955
 956             /* REACTION-FIELD ELECTROSTATICS */
 957             felec            = _mm_mul_pd(qq10,_mm_sub_pd(_mm_mul_pd(rinv10,rinvsq10),krf2));
 958
 959             cutoff_mask      = _mm_cmplt_pd(rsq10,rcutoff2);
 960
 961             fscal            = felec;
 962
 963             fscal            = _mm_and_pd(fscal,cutoff_mask);
 964
 965             /* Calculate temporary vectorial force */
 966             tx               = _mm_mul_pd(fscal,dx10);
 967             ty               = _mm_mul_pd(fscal,dy10);
 968             tz               = _mm_mul_pd(fscal,dz10);
 969
 970             /* Update vectorial force */
 971             fix1             = _mm_add_pd(fix1,tx);
 972             fiy1             = _mm_add_pd(fiy1,ty);
 973             fiz1             = _mm_add_pd(fiz1,tz);
 974
 975             fjx0             = _mm_add_pd(fjx0,tx);
 976             fjy0             = _mm_add_pd(fjy0,ty);
 977             fjz0             = _mm_add_pd(fjz0,tz);
 978
 979             }
 980
 981             /**************************
 982              * CALCULATE INTERACTIONS *
 983              **************************/
 984
 985             if (gmx_mm_any_lt(rsq20,rcutoff2))
 986             {
 987
 988             /* Compute parameters for interactions between i and j atoms */
 989             qq20             = _mm_mul_pd(iq2,jq0);
 990
 991             /* REACTION-FIELD ELECTROSTATICS */
 992             felec            = _mm_mul_pd(qq20,_mm_sub_pd(_mm_mul_pd(rinv20,rinvsq20),krf2));
 993
 994             cutoff_mask      = _mm_cmplt_pd(rsq20,rcutoff2);
 995
 996             fscal            = felec;
 997
 998             fscal            = _mm_and_pd(fscal,cutoff_mask);
 999
1000             /* Calculate temporary vectorial force */
1001             tx               = _mm_mul_pd(fscal,dx20);
1002             ty               = _mm_mul_pd(fscal,dy20);
1003             tz               = _mm_mul_pd(fscal,dz20);
1004
1005             /* Update vectorial force */
1006             fix2             = _mm_add_pd(fix2,tx);
1007             fiy2             = _mm_add_pd(fiy2,ty);
1008             fiz2             = _mm_add_pd(fiz2,tz);
1009
1010             fjx0             = _mm_add_pd(fjx0,tx);
1011             fjy0             = _mm_add_pd(fjy0,ty);
1012             fjz0             = _mm_add_pd(fjz0,tz);
1013
1014             }
1015
1016             /**************************
1017              * CALCULATE INTERACTIONS *
1018              **************************/
1019
1020             if (gmx_mm_any_lt(rsq30,rcutoff2))
1021             {
1022
1023             /* Compute parameters for interactions between i and j atoms */
1024             qq30             = _mm_mul_pd(iq3,jq0);
1025
1026             /* REACTION-FIELD ELECTROSTATICS */
1027             felec            = _mm_mul_pd(qq30,_mm_sub_pd(_mm_mul_pd(rinv30,rinvsq30),krf2));
1028
1029             cutoff_mask      = _mm_cmplt_pd(rsq30,rcutoff2);
1030
1031             fscal            = felec;
1032
1033             fscal            = _mm_and_pd(fscal,cutoff_mask);
1034
1035             /* Calculate temporary vectorial force */
1036             tx               = _mm_mul_pd(fscal,dx30);
1037             ty               = _mm_mul_pd(fscal,dy30);
1038             tz               = _mm_mul_pd(fscal,dz30);
1039
1040             /* Update vectorial force */
1041             fix3             = _mm_add_pd(fix3,tx);
1042             fiy3             = _mm_add_pd(fiy3,ty);
1043             fiz3             = _mm_add_pd(fiz3,tz);
1044
1045             fjx0             = _mm_add_pd(fjx0,tx);
1046             fjy0             = _mm_add_pd(fjy0,ty);
1047             fjz0             = _mm_add_pd(fjz0,tz);
1048
1049             }
1050
1051             gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,fjx0,fjy0,fjz0);
1052
1053             /* Inner loop uses 141 flops */
1054         }
1055
1056         if(jidx<j_index_end)
1057         {
1058
1059             jnrA             = jjnr[jidx];
1060             j_coord_offsetA  = DIM*jnrA;
1061
1062             /* load j atom coordinates */
1063             gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
1064                                               &jx0,&jy0,&jz0);
1065
1066             /* Calculate displacement vector */
1067             dx00             = _mm_sub_pd(ix0,jx0);
1068             dy00             = _mm_sub_pd(iy0,jy0);
1069             dz00             = _mm_sub_pd(iz0,jz0);
1070             dx10             = _mm_sub_pd(ix1,jx0);
1071             dy10             = _mm_sub_pd(iy1,jy0);
1072             dz10             = _mm_sub_pd(iz1,jz0);
1073             dx20             = _mm_sub_pd(ix2,jx0);
1074             dy20             = _mm_sub_pd(iy2,jy0);
1075             dz20             = _mm_sub_pd(iz2,jz0);
1076             dx30             = _mm_sub_pd(ix3,jx0);
1077             dy30             = _mm_sub_pd(iy3,jy0);
1078             dz30             = _mm_sub_pd(iz3,jz0);
1079
1080             /* Calculate squared distance and things based on it */
1081             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
1082             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
1083             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
1084             rsq30            = gmx_mm_calc_rsq_pd(dx30,dy30,dz30);
1085
1086             rinv00           = sse2_invsqrt_d(rsq00);
1087             rinv10           = sse2_invsqrt_d(rsq10);
1088             rinv20           = sse2_invsqrt_d(rsq20);
1089             rinv30           = sse2_invsqrt_d(rsq30);
1090
1091             rinvsq10         = _mm_mul_pd(rinv10,rinv10);
1092             rinvsq20         = _mm_mul_pd(rinv20,rinv20);
1093             rinvsq30         = _mm_mul_pd(rinv30,rinv30);
1094
1095             /* Load parameters for j particles */
1096             jq0              = _mm_load_sd(charge+jnrA+0);
1097             vdwjidx0A        = 2*vdwtype[jnrA+0];
1098
1099             fjx0             = _mm_setzero_pd();
1100             fjy0             = _mm_setzero_pd();
1101             fjz0             = _mm_setzero_pd();
1102
1103             /**************************
1104              * CALCULATE INTERACTIONS *
1105              **************************/
1106
1107             r00              = _mm_mul_pd(rsq00,rinv00);
1108
1109             /* Compute parameters for interactions between i and j atoms */
1110             gmx_mm_load_1pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,&c6_00,&c12_00);
1111
1112             /* Calculate table index by multiplying r with table scale and truncate to integer */
1113             rt               = _mm_mul_pd(r00,vftabscale);
1114             vfitab           = _mm_cvttpd_epi32(rt);
1115             vfeps            = _mm_sub_pd(rt,_mm_cvtepi32_pd(vfitab));
1116             vfitab           = _mm_slli_epi32(vfitab,3);
1117
1118             /* CUBIC SPLINE TABLE DISPERSION */
1119             Y                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
1120             F                = _mm_setzero_pd();
1121             GMX_MM_TRANSPOSE2_PD(Y,F);
1122             G                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
1123             H                = _mm_setzero_pd();
1124             GMX_MM_TRANSPOSE2_PD(G,H);
1125             Heps             = _mm_mul_pd(vfeps,H);
1126             Fp               = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
1127             FF               = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
1128             fvdw6            = _mm_mul_pd(c6_00,FF);
1129
1130             /* CUBIC SPLINE TABLE REPULSION */
1131             vfitab           = _mm_add_epi32(vfitab,ifour);
1132             Y                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
1133             F                = _mm_setzero_pd();
1134             GMX_MM_TRANSPOSE2_PD(Y,F);
1135             G                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
1136             H                = _mm_setzero_pd();
1137             GMX_MM_TRANSPOSE2_PD(G,H);
1138             Heps             = _mm_mul_pd(vfeps,H);
1139             Fp               = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
1140             FF               = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
1141             fvdw12           = _mm_mul_pd(c12_00,FF);
1142             fvdw             = _mm_xor_pd(signbit,_mm_mul_pd(_mm_add_pd(fvdw6,fvdw12),_mm_mul_pd(vftabscale,rinv00)));
1143
1144             fscal            = fvdw;
1145
1146             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1147
1148             /* Calculate temporary vectorial force */
1149             tx               = _mm_mul_pd(fscal,dx00);
1150             ty               = _mm_mul_pd(fscal,dy00);
1151             tz               = _mm_mul_pd(fscal,dz00);
1152
1153             /* Update vectorial force */
1154             fix0             = _mm_add_pd(fix0,tx);
1155             fiy0             = _mm_add_pd(fiy0,ty);
1156             fiz0             = _mm_add_pd(fiz0,tz);
1157
1158             fjx0             = _mm_add_pd(fjx0,tx);
1159             fjy0             = _mm_add_pd(fjy0,ty);
1160             fjz0             = _mm_add_pd(fjz0,tz);
1161
1162             /**************************
1163              * CALCULATE INTERACTIONS *
1164              **************************/
1165
1166             if (gmx_mm_any_lt(rsq10,rcutoff2))
1167             {
1168
1169             /* Compute parameters for interactions between i and j atoms */
1170             qq10             = _mm_mul_pd(iq1,jq0);
1171
1172             /* REACTION-FIELD ELECTROSTATICS */
1173             felec            = _mm_mul_pd(qq10,_mm_sub_pd(_mm_mul_pd(rinv10,rinvsq10),krf2));
1174
1175             cutoff_mask      = _mm_cmplt_pd(rsq10,rcutoff2);
1176
1177             fscal            = felec;
1178
1179             fscal            = _mm_and_pd(fscal,cutoff_mask);
1180
1181             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1182
1183             /* Calculate temporary vectorial force */
1184             tx               = _mm_mul_pd(fscal,dx10);
1185             ty               = _mm_mul_pd(fscal,dy10);
1186             tz               = _mm_mul_pd(fscal,dz10);
1187
1188             /* Update vectorial force */
1189             fix1             = _mm_add_pd(fix1,tx);
1190             fiy1             = _mm_add_pd(fiy1,ty);
1191             fiz1             = _mm_add_pd(fiz1,tz);
1192
1193             fjx0             = _mm_add_pd(fjx0,tx);
1194             fjy0             = _mm_add_pd(fjy0,ty);
1195             fjz0             = _mm_add_pd(fjz0,tz);
1196
1197             }
1198
1199             /**************************
1200              * CALCULATE INTERACTIONS *
1201              **************************/
1202
1203             if (gmx_mm_any_lt(rsq20,rcutoff2))
1204             {
1205
1206             /* Compute parameters for interactions between i and j atoms */
1207             qq20             = _mm_mul_pd(iq2,jq0);
1208
1209             /* REACTION-FIELD ELECTROSTATICS */
1210             felec            = _mm_mul_pd(qq20,_mm_sub_pd(_mm_mul_pd(rinv20,rinvsq20),krf2));
1211
1212             cutoff_mask      = _mm_cmplt_pd(rsq20,rcutoff2);
1213
1214             fscal            = felec;
1215
1216             fscal            = _mm_and_pd(fscal,cutoff_mask);
1217
1218             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1219
1220             /* Calculate temporary vectorial force */
1221             tx               = _mm_mul_pd(fscal,dx20);
1222             ty               = _mm_mul_pd(fscal,dy20);
1223             tz               = _mm_mul_pd(fscal,dz20);
1224
1225             /* Update vectorial force */
1226             fix2             = _mm_add_pd(fix2,tx);
1227             fiy2             = _mm_add_pd(fiy2,ty);
1228             fiz2             = _mm_add_pd(fiz2,tz);
1229
1230             fjx0             = _mm_add_pd(fjx0,tx);
1231             fjy0             = _mm_add_pd(fjy0,ty);
1232             fjz0             = _mm_add_pd(fjz0,tz);
1233
1234             }
1235
1236             /**************************
1237              * CALCULATE INTERACTIONS *
1238              **************************/
1239
1240             if (gmx_mm_any_lt(rsq30,rcutoff2))
1241             {
1242
1243             /* Compute parameters for interactions between i and j atoms */
1244             qq30             = _mm_mul_pd(iq3,jq0);
1245
1246             /* REACTION-FIELD ELECTROSTATICS */
1247             felec            = _mm_mul_pd(qq30,_mm_sub_pd(_mm_mul_pd(rinv30,rinvsq30),krf2));
1248
1249             cutoff_mask      = _mm_cmplt_pd(rsq30,rcutoff2);
1250
1251             fscal            = felec;
1252
1253             fscal            = _mm_and_pd(fscal,cutoff_mask);
1254
1255             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1256
1257             /* Calculate temporary vectorial force */
1258             tx               = _mm_mul_pd(fscal,dx30);
1259             ty               = _mm_mul_pd(fscal,dy30);
1260             tz               = _mm_mul_pd(fscal,dz30);
1261
1262             /* Update vectorial force */
1263             fix3             = _mm_add_pd(fix3,tx);
1264             fiy3             = _mm_add_pd(fiy3,ty);
1265             fiz3             = _mm_add_pd(fiz3,tz);
1266
1267             fjx0             = _mm_add_pd(fjx0,tx);
1268             fjy0             = _mm_add_pd(fjy0,ty);
1269             fjz0             = _mm_add_pd(fjz0,tz);
1270
1271             }
1272
1273             gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,fjx0,fjy0,fjz0);
1274
1275             /* Inner loop uses 141 flops */
1276         }
1277
1278         /* End of innermost loop */
1279
1280         gmx_mm_update_iforce_4atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
1281                                               f+i_coord_offset,fshift+i_shift_offset);
1282
1283         /* Increment number of inner iterations */
1284         inneriter                  += j_index_end - j_index_start;
1285
1286         /* Outer loop uses 24 flops */
1287     }
1288
1289     /* Increment number of outer iterations */
1290     outeriter        += nri;
1291
1292     /* Update outer/inner flops */
1293
1294     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W4_F,outeriter*24 + inneriter*141);
1295 }