src/gromacs/gmxlib/nonbonded/nb_kernel_sse4_1_double/nb_kernel_ElecCoul_VdwCSTab_GeomW4P1_sse4_1_double.c

   1 /*
   2  * This file is part of the GROMACS molecular simulation package.
   3  *
   4  * Copyright (c) 2012,2013,2014, by the GROMACS development team, led by
   5  * Mark Abraham, David van der Spoel, Berk Hess, and Erik Lindahl,
   6  * and including many others, as listed in the AUTHORS file in the
   7  * top-level source directory and at http://www.gromacs.org.
   8  *
   9  * GROMACS is free software; you can redistribute it and/or
  10  * modify it under the terms of the GNU Lesser General Public License
  11  * as published by the Free Software Foundation; either version 2.1
  12  * of the License, or (at your option) any later version.
  13  *
  14  * GROMACS is distributed in the hope that it will be useful,
  15  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  16  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  17  * Lesser General Public License for more details.
  18  *
  19  * You should have received a copy of the GNU Lesser General Public
  20  * License along with GROMACS; if not, see
  21  * http://www.gnu.org/licenses, or write to the Free Software Foundation,
  22  * Inc., 51 Franklin Street, Fifth Floor, Boston, MA  02110-1301  USA.
  23  *
  24  * If you want to redistribute modifications to GROMACS, please
  25  * consider that scientific software is very special. Version
  26  * control is crucial - bugs must be traceable. We will be happy to
  27  * consider code for inclusion in the official distribution, but
  28  * derived work must not be called official GROMACS. Details are found
  29  * in the README & COPYING files - if they are missing, get the
  30  * official version at http://www.gromacs.org.
  31  *
  32  * To help us fund GROMACS development, we humbly ask that you cite
  33  * the research papers on the package. Check out http://www.gromacs.org.
  34  */
  35 /*
  36  * Note: this file was generated by the GROMACS sse4_1_double kernel generator.
  37  */
  38 #include "config.h"
  39
  40 #include <math.h>
  41
  42 #include "../nb_kernel.h"
  43 #include "types/simple.h"
  44 #include "gromacs/math/vec.h"
  45 #include "nrnb.h"
  46
  47 #include "gromacs/simd/math_x86_sse4_1_double.h"
  48 #include "kernelutil_x86_sse4_1_double.h"
  49
  50 /*
  51  * Gromacs nonbonded kernel:   nb_kernel_ElecCoul_VdwCSTab_GeomW4P1_VF_sse4_1_double
  52  * Electrostatics interaction: Coulomb
  53  * VdW interaction:            CubicSplineTable
  54  * Geometry:                   Water4-Particle
  55  * Calculate force/pot:        PotentialAndForce
  56  */
  57 void
  58 nb_kernel_ElecCoul_VdwCSTab_GeomW4P1_VF_sse4_1_double
  59                     (t_nblist                    * gmx_restrict       nlist,
  60                      rvec                        * gmx_restrict          xx,
  61                      rvec                        * gmx_restrict          ff,
  62                      t_forcerec                  * gmx_restrict          fr,
  63                      t_mdatoms                   * gmx_restrict     mdatoms,
  64                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
  65                      t_nrnb                      * gmx_restrict        nrnb)
  66 {
  67     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
  68      * just 0 for non-waters.
  69      * Suffixes A,B refer to j loop unrolling done with SSE double precision, e.g. for the two different
  70      * jnr indices corresponding to data put in the four positions in the SIMD register.
  71      */
  72     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
  73     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
  74     int              jnrA,jnrB;
  75     int              j_coord_offsetA,j_coord_offsetB;
  76     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
  77     real             rcutoff_scalar;
  78     real             *shiftvec,*fshift,*x,*f;
  79     __m128d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
  80     int              vdwioffset0;
  81     __m128d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
  82     int              vdwioffset1;
  83     __m128d          ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
  84     int              vdwioffset2;
  85     __m128d          ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
  86     int              vdwioffset3;
  87     __m128d          ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
  88     int              vdwjidx0A,vdwjidx0B;
  89     __m128d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  90     __m128d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  91     __m128d          dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
  92     __m128d          dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
  93     __m128d          dx30,dy30,dz30,rsq30,rinv30,rinvsq30,r30,qq30,c6_30,c12_30;
  94     __m128d          velec,felec,velecsum,facel,crf,krf,krf2;
  95     real             *charge;
  96     int              nvdwtype;
  97     __m128d          rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
  98     int              *vdwtype;
  99     real             *vdwparam;
 100     __m128d          one_sixth   = _mm_set1_pd(1.0/6.0);
 101     __m128d          one_twelfth = _mm_set1_pd(1.0/12.0);
 102     __m128i          vfitab;
 103     __m128i          ifour       = _mm_set1_epi32(4);
 104     __m128d          rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF;
 105     real             *vftab;
 106     __m128d          dummy_mask,cutoff_mask;
 107     __m128d          signbit   = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
 108     __m128d          one     = _mm_set1_pd(1.0);
 109     __m128d          two     = _mm_set1_pd(2.0);
 110     x                = xx[0];
 111     f                = ff[0];
 112
 113     nri              = nlist->nri;
 114     iinr             = nlist->iinr;
 115     jindex           = nlist->jindex;
 116     jjnr             = nlist->jjnr;
 117     shiftidx         = nlist->shift;
 118     gid              = nlist->gid;
 119     shiftvec         = fr->shift_vec[0];
 120     fshift           = fr->fshift[0];
 121     facel            = _mm_set1_pd(fr->epsfac);
 122     charge           = mdatoms->chargeA;
 123     nvdwtype         = fr->ntype;
 124     vdwparam         = fr->nbfp;
 125     vdwtype          = mdatoms->typeA;
 126
 127     vftab            = kernel_data->table_vdw->data;
 128     vftabscale       = _mm_set1_pd(kernel_data->table_vdw->scale);
 129
 130     /* Setup water-specific parameters */
 131     inr              = nlist->iinr[0];
 132     iq1              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+1]));
 133     iq2              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+2]));
 134     iq3              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+3]));
 135     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 136
 137     /* Avoid stupid compiler warnings */
 138     jnrA = jnrB = 0;
 139     j_coord_offsetA = 0;
 140     j_coord_offsetB = 0;
 141
 142     outeriter        = 0;
 143     inneriter        = 0;
 144
 145     /* Start outer loop over neighborlists */
 146     for(iidx=0; iidx<nri; iidx++)
 147     {
 148         /* Load shift vector for this list */
 149         i_shift_offset   = DIM*shiftidx[iidx];
 150
 151         /* Load limits for loop over neighbors */
 152         j_index_start    = jindex[iidx];
 153         j_index_end      = jindex[iidx+1];
 154
 155         /* Get outer coordinate index */
 156         inr              = iinr[iidx];
 157         i_coord_offset   = DIM*inr;
 158
 159         /* Load i particle coords and add shift vector */
 160         gmx_mm_load_shift_and_4rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
 161                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
 162
 163         fix0             = _mm_setzero_pd();
 164         fiy0             = _mm_setzero_pd();
 165         fiz0             = _mm_setzero_pd();
 166         fix1             = _mm_setzero_pd();
 167         fiy1             = _mm_setzero_pd();
 168         fiz1             = _mm_setzero_pd();
 169         fix2             = _mm_setzero_pd();
 170         fiy2             = _mm_setzero_pd();
 171         fiz2             = _mm_setzero_pd();
 172         fix3             = _mm_setzero_pd();
 173         fiy3             = _mm_setzero_pd();
 174         fiz3             = _mm_setzero_pd();
 175
 176         /* Reset potential sums */
 177         velecsum         = _mm_setzero_pd();
 178         vvdwsum          = _mm_setzero_pd();
 179
 180         /* Start inner kernel loop */
 181         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 182         {
 183
 184             /* Get j neighbor index, and coordinate index */
 185             jnrA             = jjnr[jidx];
 186             jnrB             = jjnr[jidx+1];
 187             j_coord_offsetA  = DIM*jnrA;
 188             j_coord_offsetB  = DIM*jnrB;
 189
 190             /* load j atom coordinates */
 191             gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 192                                               &jx0,&jy0,&jz0);
 193
 194             /* Calculate displacement vector */
 195             dx00             = _mm_sub_pd(ix0,jx0);
 196             dy00             = _mm_sub_pd(iy0,jy0);
 197             dz00             = _mm_sub_pd(iz0,jz0);
 198             dx10             = _mm_sub_pd(ix1,jx0);
 199             dy10             = _mm_sub_pd(iy1,jy0);
 200             dz10             = _mm_sub_pd(iz1,jz0);
 201             dx20             = _mm_sub_pd(ix2,jx0);
 202             dy20             = _mm_sub_pd(iy2,jy0);
 203             dz20             = _mm_sub_pd(iz2,jz0);
 204             dx30             = _mm_sub_pd(ix3,jx0);
 205             dy30             = _mm_sub_pd(iy3,jy0);
 206             dz30             = _mm_sub_pd(iz3,jz0);
 207
 208             /* Calculate squared distance and things based on it */
 209             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 210             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
 211             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
 212             rsq30            = gmx_mm_calc_rsq_pd(dx30,dy30,dz30);
 213
 214             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 215             rinv10           = gmx_mm_invsqrt_pd(rsq10);
 216             rinv20           = gmx_mm_invsqrt_pd(rsq20);
 217             rinv30           = gmx_mm_invsqrt_pd(rsq30);
 218
 219             rinvsq10         = _mm_mul_pd(rinv10,rinv10);
 220             rinvsq20         = _mm_mul_pd(rinv20,rinv20);
 221             rinvsq30         = _mm_mul_pd(rinv30,rinv30);
 222
 223             /* Load parameters for j particles */
 224             jq0              = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
 225             vdwjidx0A        = 2*vdwtype[jnrA+0];
 226             vdwjidx0B        = 2*vdwtype[jnrB+0];
 227
 228             fjx0             = _mm_setzero_pd();
 229             fjy0             = _mm_setzero_pd();
 230             fjz0             = _mm_setzero_pd();
 231
 232             /**************************
 233              * CALCULATE INTERACTIONS *
 234              **************************/
 235
 236             r00              = _mm_mul_pd(rsq00,rinv00);
 237
 238             /* Compute parameters for interactions between i and j atoms */
 239             gmx_mm_load_2pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,
 240                                          vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
 241
 242             /* Calculate table index by multiplying r with table scale and truncate to integer */
 243             rt               = _mm_mul_pd(r00,vftabscale);
 244             vfitab           = _mm_cvttpd_epi32(rt);
 245             vfeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
 246             vfitab           = _mm_slli_epi32(vfitab,3);
 247
 248             /* CUBIC SPLINE TABLE DISPERSION */
 249             Y                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
 250             F                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) );
 251             GMX_MM_TRANSPOSE2_PD(Y,F);
 252             G                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
 253             H                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) +2);
 254             GMX_MM_TRANSPOSE2_PD(G,H);
 255             Heps             = _mm_mul_pd(vfeps,H);
 256             Fp               = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
 257             VV               = _mm_add_pd(Y,_mm_mul_pd(vfeps,Fp));
 258             vvdw6            = _mm_mul_pd(c6_00,VV);
 259             FF               = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
 260             fvdw6            = _mm_mul_pd(c6_00,FF);
 261
 262             /* CUBIC SPLINE TABLE REPULSION */
 263             vfitab           = _mm_add_epi32(vfitab,ifour);
 264             Y                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
 265             F                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) );
 266             GMX_MM_TRANSPOSE2_PD(Y,F);
 267             G                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
 268             H                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) +2);
 269             GMX_MM_TRANSPOSE2_PD(G,H);
 270             Heps             = _mm_mul_pd(vfeps,H);
 271             Fp               = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
 272             VV               = _mm_add_pd(Y,_mm_mul_pd(vfeps,Fp));
 273             vvdw12           = _mm_mul_pd(c12_00,VV);
 274             FF               = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
 275             fvdw12           = _mm_mul_pd(c12_00,FF);
 276             vvdw             = _mm_add_pd(vvdw12,vvdw6);
 277             fvdw             = _mm_xor_pd(signbit,_mm_mul_pd(_mm_add_pd(fvdw6,fvdw12),_mm_mul_pd(vftabscale,rinv00)));
 278
 279             /* Update potential sum for this i atom from the interaction with this j atom. */
 280             vvdwsum          = _mm_add_pd(vvdwsum,vvdw);
 281
 282             fscal            = 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             /* COULOMB ELECTROSTATICS */
 306             velec            = _mm_mul_pd(qq10,rinv10);
 307             felec            = _mm_mul_pd(velec,rinvsq10);
 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             /* COULOMB ELECTROSTATICS */
 336             velec            = _mm_mul_pd(qq20,rinv20);
 337             felec            = _mm_mul_pd(velec,rinvsq20);
 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             /**************************
 359              * CALCULATE INTERACTIONS *
 360              **************************/
 361
 362             /* Compute parameters for interactions between i and j atoms */
 363             qq30             = _mm_mul_pd(iq3,jq0);
 364
 365             /* COULOMB ELECTROSTATICS */
 366             velec            = _mm_mul_pd(qq30,rinv30);
 367             felec            = _mm_mul_pd(velec,rinvsq30);
 368
 369             /* Update potential sum for this i atom from the interaction with this j atom. */
 370             velecsum         = _mm_add_pd(velecsum,velec);
 371
 372             fscal            = felec;
 373
 374             /* Calculate temporary vectorial force */
 375             tx               = _mm_mul_pd(fscal,dx30);
 376             ty               = _mm_mul_pd(fscal,dy30);
 377             tz               = _mm_mul_pd(fscal,dz30);
 378
 379             /* Update vectorial force */
 380             fix3             = _mm_add_pd(fix3,tx);
 381             fiy3             = _mm_add_pd(fiy3,ty);
 382             fiz3             = _mm_add_pd(fiz3,tz);
 383
 384             fjx0             = _mm_add_pd(fjx0,tx);
 385             fjy0             = _mm_add_pd(fjy0,ty);
 386             fjz0             = _mm_add_pd(fjz0,tz);
 387
 388             gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,fjx0,fjy0,fjz0);
 389
 390             /* Inner loop uses 143 flops */
 391         }
 392
 393         if(jidx<j_index_end)
 394         {
 395
 396             jnrA             = jjnr[jidx];
 397             j_coord_offsetA  = DIM*jnrA;
 398
 399             /* load j atom coordinates */
 400             gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
 401                                               &jx0,&jy0,&jz0);
 402
 403             /* Calculate displacement vector */
 404             dx00             = _mm_sub_pd(ix0,jx0);
 405             dy00             = _mm_sub_pd(iy0,jy0);
 406             dz00             = _mm_sub_pd(iz0,jz0);
 407             dx10             = _mm_sub_pd(ix1,jx0);
 408             dy10             = _mm_sub_pd(iy1,jy0);
 409             dz10             = _mm_sub_pd(iz1,jz0);
 410             dx20             = _mm_sub_pd(ix2,jx0);
 411             dy20             = _mm_sub_pd(iy2,jy0);
 412             dz20             = _mm_sub_pd(iz2,jz0);
 413             dx30             = _mm_sub_pd(ix3,jx0);
 414             dy30             = _mm_sub_pd(iy3,jy0);
 415             dz30             = _mm_sub_pd(iz3,jz0);
 416
 417             /* Calculate squared distance and things based on it */
 418             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 419             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
 420             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
 421             rsq30            = gmx_mm_calc_rsq_pd(dx30,dy30,dz30);
 422
 423             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 424             rinv10           = gmx_mm_invsqrt_pd(rsq10);
 425             rinv20           = gmx_mm_invsqrt_pd(rsq20);
 426             rinv30           = gmx_mm_invsqrt_pd(rsq30);
 427
 428             rinvsq10         = _mm_mul_pd(rinv10,rinv10);
 429             rinvsq20         = _mm_mul_pd(rinv20,rinv20);
 430             rinvsq30         = _mm_mul_pd(rinv30,rinv30);
 431
 432             /* Load parameters for j particles */
 433             jq0              = _mm_load_sd(charge+jnrA+0);
 434             vdwjidx0A        = 2*vdwtype[jnrA+0];
 435
 436             fjx0             = _mm_setzero_pd();
 437             fjy0             = _mm_setzero_pd();
 438             fjz0             = _mm_setzero_pd();
 439
 440             /**************************
 441              * CALCULATE INTERACTIONS *
 442              **************************/
 443
 444             r00              = _mm_mul_pd(rsq00,rinv00);
 445
 446             /* Compute parameters for interactions between i and j atoms */
 447             gmx_mm_load_1pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,&c6_00,&c12_00);
 448
 449             /* Calculate table index by multiplying r with table scale and truncate to integer */
 450             rt               = _mm_mul_pd(r00,vftabscale);
 451             vfitab           = _mm_cvttpd_epi32(rt);
 452             vfeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
 453             vfitab           = _mm_slli_epi32(vfitab,3);
 454
 455             /* CUBIC SPLINE TABLE DISPERSION */
 456             Y                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
 457             F                = _mm_setzero_pd();
 458             GMX_MM_TRANSPOSE2_PD(Y,F);
 459             G                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
 460             H                = _mm_setzero_pd();
 461             GMX_MM_TRANSPOSE2_PD(G,H);
 462             Heps             = _mm_mul_pd(vfeps,H);
 463             Fp               = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
 464             VV               = _mm_add_pd(Y,_mm_mul_pd(vfeps,Fp));
 465             vvdw6            = _mm_mul_pd(c6_00,VV);
 466             FF               = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
 467             fvdw6            = _mm_mul_pd(c6_00,FF);
 468
 469             /* CUBIC SPLINE TABLE REPULSION */
 470             vfitab           = _mm_add_epi32(vfitab,ifour);
 471             Y                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
 472             F                = _mm_setzero_pd();
 473             GMX_MM_TRANSPOSE2_PD(Y,F);
 474             G                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
 475             H                = _mm_setzero_pd();
 476             GMX_MM_TRANSPOSE2_PD(G,H);
 477             Heps             = _mm_mul_pd(vfeps,H);
 478             Fp               = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
 479             VV               = _mm_add_pd(Y,_mm_mul_pd(vfeps,Fp));
 480             vvdw12           = _mm_mul_pd(c12_00,VV);
 481             FF               = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
 482             fvdw12           = _mm_mul_pd(c12_00,FF);
 483             vvdw             = _mm_add_pd(vvdw12,vvdw6);
 484             fvdw             = _mm_xor_pd(signbit,_mm_mul_pd(_mm_add_pd(fvdw6,fvdw12),_mm_mul_pd(vftabscale,rinv00)));
 485
 486             /* Update potential sum for this i atom from the interaction with this j atom. */
 487             vvdw             = _mm_unpacklo_pd(vvdw,_mm_setzero_pd());
 488             vvdwsum          = _mm_add_pd(vvdwsum,vvdw);
 489
 490             fscal            = fvdw;
 491
 492             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 493
 494             /* Calculate temporary vectorial force */
 495             tx               = _mm_mul_pd(fscal,dx00);
 496             ty               = _mm_mul_pd(fscal,dy00);
 497             tz               = _mm_mul_pd(fscal,dz00);
 498
 499             /* Update vectorial force */
 500             fix0             = _mm_add_pd(fix0,tx);
 501             fiy0             = _mm_add_pd(fiy0,ty);
 502             fiz0             = _mm_add_pd(fiz0,tz);
 503
 504             fjx0             = _mm_add_pd(fjx0,tx);
 505             fjy0             = _mm_add_pd(fjy0,ty);
 506             fjz0             = _mm_add_pd(fjz0,tz);
 507
 508             /**************************
 509              * CALCULATE INTERACTIONS *
 510              **************************/
 511
 512             /* Compute parameters for interactions between i and j atoms */
 513             qq10             = _mm_mul_pd(iq1,jq0);
 514
 515             /* COULOMB ELECTROSTATICS */
 516             velec            = _mm_mul_pd(qq10,rinv10);
 517             felec            = _mm_mul_pd(velec,rinvsq10);
 518
 519             /* Update potential sum for this i atom from the interaction with this j atom. */
 520             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 521             velecsum         = _mm_add_pd(velecsum,velec);
 522
 523             fscal            = felec;
 524
 525             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 526
 527             /* Calculate temporary vectorial force */
 528             tx               = _mm_mul_pd(fscal,dx10);
 529             ty               = _mm_mul_pd(fscal,dy10);
 530             tz               = _mm_mul_pd(fscal,dz10);
 531
 532             /* Update vectorial force */
 533             fix1             = _mm_add_pd(fix1,tx);
 534             fiy1             = _mm_add_pd(fiy1,ty);
 535             fiz1             = _mm_add_pd(fiz1,tz);
 536
 537             fjx0             = _mm_add_pd(fjx0,tx);
 538             fjy0             = _mm_add_pd(fjy0,ty);
 539             fjz0             = _mm_add_pd(fjz0,tz);
 540
 541             /**************************
 542              * CALCULATE INTERACTIONS *
 543              **************************/
 544
 545             /* Compute parameters for interactions between i and j atoms */
 546             qq20             = _mm_mul_pd(iq2,jq0);
 547
 548             /* COULOMB ELECTROSTATICS */
 549             velec            = _mm_mul_pd(qq20,rinv20);
 550             felec            = _mm_mul_pd(velec,rinvsq20);
 551
 552             /* Update potential sum for this i atom from the interaction with this j atom. */
 553             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 554             velecsum         = _mm_add_pd(velecsum,velec);
 555
 556             fscal            = felec;
 557
 558             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 559
 560             /* Calculate temporary vectorial force */
 561             tx               = _mm_mul_pd(fscal,dx20);
 562             ty               = _mm_mul_pd(fscal,dy20);
 563             tz               = _mm_mul_pd(fscal,dz20);
 564
 565             /* Update vectorial force */
 566             fix2             = _mm_add_pd(fix2,tx);
 567             fiy2             = _mm_add_pd(fiy2,ty);
 568             fiz2             = _mm_add_pd(fiz2,tz);
 569
 570             fjx0             = _mm_add_pd(fjx0,tx);
 571             fjy0             = _mm_add_pd(fjy0,ty);
 572             fjz0             = _mm_add_pd(fjz0,tz);
 573
 574             /**************************
 575              * CALCULATE INTERACTIONS *
 576              **************************/
 577
 578             /* Compute parameters for interactions between i and j atoms */
 579             qq30             = _mm_mul_pd(iq3,jq0);
 580
 581             /* COULOMB ELECTROSTATICS */
 582             velec            = _mm_mul_pd(qq30,rinv30);
 583             felec            = _mm_mul_pd(velec,rinvsq30);
 584
 585             /* Update potential sum for this i atom from the interaction with this j atom. */
 586             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 587             velecsum         = _mm_add_pd(velecsum,velec);
 588
 589             fscal            = felec;
 590
 591             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 592
 593             /* Calculate temporary vectorial force */
 594             tx               = _mm_mul_pd(fscal,dx30);
 595             ty               = _mm_mul_pd(fscal,dy30);
 596             tz               = _mm_mul_pd(fscal,dz30);
 597
 598             /* Update vectorial force */
 599             fix3             = _mm_add_pd(fix3,tx);
 600             fiy3             = _mm_add_pd(fiy3,ty);
 601             fiz3             = _mm_add_pd(fiz3,tz);
 602
 603             fjx0             = _mm_add_pd(fjx0,tx);
 604             fjy0             = _mm_add_pd(fjy0,ty);
 605             fjz0             = _mm_add_pd(fjz0,tz);
 606
 607             gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,fjx0,fjy0,fjz0);
 608
 609             /* Inner loop uses 143 flops */
 610         }
 611
 612         /* End of innermost loop */
 613
 614         gmx_mm_update_iforce_4atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
 615                                               f+i_coord_offset,fshift+i_shift_offset);
 616
 617         ggid                        = gid[iidx];
 618         /* Update potential energies */
 619         gmx_mm_update_1pot_pd(velecsum,kernel_data->energygrp_elec+ggid);
 620         gmx_mm_update_1pot_pd(vvdwsum,kernel_data->energygrp_vdw+ggid);
 621
 622         /* Increment number of inner iterations */
 623         inneriter                  += j_index_end - j_index_start;
 624
 625         /* Outer loop uses 26 flops */
 626     }
 627
 628     /* Increment number of outer iterations */
 629     outeriter        += nri;
 630
 631     /* Update outer/inner flops */
 632
 633     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W4_VF,outeriter*26 + inneriter*143);
 634 }
 635 /*
 636  * Gromacs nonbonded kernel:   nb_kernel_ElecCoul_VdwCSTab_GeomW4P1_F_sse4_1_double
 637  * Electrostatics interaction: Coulomb
 638  * VdW interaction:            CubicSplineTable
 639  * Geometry:                   Water4-Particle
 640  * Calculate force/pot:        Force
 641  */
 642 void
 643 nb_kernel_ElecCoul_VdwCSTab_GeomW4P1_F_sse4_1_double
 644                     (t_nblist                    * gmx_restrict       nlist,
 645                      rvec                        * gmx_restrict          xx,
 646                      rvec                        * gmx_restrict          ff,
 647                      t_forcerec                  * gmx_restrict          fr,
 648                      t_mdatoms                   * gmx_restrict     mdatoms,
 649                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 650                      t_nrnb                      * gmx_restrict        nrnb)
 651 {
 652     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 653      * just 0 for non-waters.
 654      * Suffixes A,B refer to j loop unrolling done with SSE double precision, e.g. for the two different
 655      * jnr indices corresponding to data put in the four positions in the SIMD register.
 656      */
 657     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 658     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 659     int              jnrA,jnrB;
 660     int              j_coord_offsetA,j_coord_offsetB;
 661     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 662     real             rcutoff_scalar;
 663     real             *shiftvec,*fshift,*x,*f;
 664     __m128d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 665     int              vdwioffset0;
 666     __m128d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 667     int              vdwioffset1;
 668     __m128d          ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
 669     int              vdwioffset2;
 670     __m128d          ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
 671     int              vdwioffset3;
 672     __m128d          ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
 673     int              vdwjidx0A,vdwjidx0B;
 674     __m128d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 675     __m128d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 676     __m128d          dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
 677     __m128d          dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
 678     __m128d          dx30,dy30,dz30,rsq30,rinv30,rinvsq30,r30,qq30,c6_30,c12_30;
 679     __m128d          velec,felec,velecsum,facel,crf,krf,krf2;
 680     real             *charge;
 681     int              nvdwtype;
 682     __m128d          rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 683     int              *vdwtype;
 684     real             *vdwparam;
 685     __m128d          one_sixth   = _mm_set1_pd(1.0/6.0);
 686     __m128d          one_twelfth = _mm_set1_pd(1.0/12.0);
 687     __m128i          vfitab;
 688     __m128i          ifour       = _mm_set1_epi32(4);
 689     __m128d          rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF;
 690     real             *vftab;
 691     __m128d          dummy_mask,cutoff_mask;
 692     __m128d          signbit   = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
 693     __m128d          one     = _mm_set1_pd(1.0);
 694     __m128d          two     = _mm_set1_pd(2.0);
 695     x                = xx[0];
 696     f                = ff[0];
 697
 698     nri              = nlist->nri;
 699     iinr             = nlist->iinr;
 700     jindex           = nlist->jindex;
 701     jjnr             = nlist->jjnr;
 702     shiftidx         = nlist->shift;
 703     gid              = nlist->gid;
 704     shiftvec         = fr->shift_vec[0];
 705     fshift           = fr->fshift[0];
 706     facel            = _mm_set1_pd(fr->epsfac);
 707     charge           = mdatoms->chargeA;
 708     nvdwtype         = fr->ntype;
 709     vdwparam         = fr->nbfp;
 710     vdwtype          = mdatoms->typeA;
 711
 712     vftab            = kernel_data->table_vdw->data;
 713     vftabscale       = _mm_set1_pd(kernel_data->table_vdw->scale);
 714
 715     /* Setup water-specific parameters */
 716     inr              = nlist->iinr[0];
 717     iq1              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+1]));
 718     iq2              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+2]));
 719     iq3              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+3]));
 720     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 721
 722     /* Avoid stupid compiler warnings */
 723     jnrA = jnrB = 0;
 724     j_coord_offsetA = 0;
 725     j_coord_offsetB = 0;
 726
 727     outeriter        = 0;
 728     inneriter        = 0;
 729
 730     /* Start outer loop over neighborlists */
 731     for(iidx=0; iidx<nri; iidx++)
 732     {
 733         /* Load shift vector for this list */
 734         i_shift_offset   = DIM*shiftidx[iidx];
 735
 736         /* Load limits for loop over neighbors */
 737         j_index_start    = jindex[iidx];
 738         j_index_end      = jindex[iidx+1];
 739
 740         /* Get outer coordinate index */
 741         inr              = iinr[iidx];
 742         i_coord_offset   = DIM*inr;
 743
 744         /* Load i particle coords and add shift vector */
 745         gmx_mm_load_shift_and_4rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
 746                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
 747
 748         fix0             = _mm_setzero_pd();
 749         fiy0             = _mm_setzero_pd();
 750         fiz0             = _mm_setzero_pd();
 751         fix1             = _mm_setzero_pd();
 752         fiy1             = _mm_setzero_pd();
 753         fiz1             = _mm_setzero_pd();
 754         fix2             = _mm_setzero_pd();
 755         fiy2             = _mm_setzero_pd();
 756         fiz2             = _mm_setzero_pd();
 757         fix3             = _mm_setzero_pd();
 758         fiy3             = _mm_setzero_pd();
 759         fiz3             = _mm_setzero_pd();
 760
 761         /* Start inner kernel loop */
 762         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 763         {
 764
 765             /* Get j neighbor index, and coordinate index */
 766             jnrA             = jjnr[jidx];
 767             jnrB             = jjnr[jidx+1];
 768             j_coord_offsetA  = DIM*jnrA;
 769             j_coord_offsetB  = DIM*jnrB;
 770
 771             /* load j atom coordinates */
 772             gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 773                                               &jx0,&jy0,&jz0);
 774
 775             /* Calculate displacement vector */
 776             dx00             = _mm_sub_pd(ix0,jx0);
 777             dy00             = _mm_sub_pd(iy0,jy0);
 778             dz00             = _mm_sub_pd(iz0,jz0);
 779             dx10             = _mm_sub_pd(ix1,jx0);
 780             dy10             = _mm_sub_pd(iy1,jy0);
 781             dz10             = _mm_sub_pd(iz1,jz0);
 782             dx20             = _mm_sub_pd(ix2,jx0);
 783             dy20             = _mm_sub_pd(iy2,jy0);
 784             dz20             = _mm_sub_pd(iz2,jz0);
 785             dx30             = _mm_sub_pd(ix3,jx0);
 786             dy30             = _mm_sub_pd(iy3,jy0);
 787             dz30             = _mm_sub_pd(iz3,jz0);
 788
 789             /* Calculate squared distance and things based on it */
 790             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 791             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
 792             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
 793             rsq30            = gmx_mm_calc_rsq_pd(dx30,dy30,dz30);
 794
 795             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 796             rinv10           = gmx_mm_invsqrt_pd(rsq10);
 797             rinv20           = gmx_mm_invsqrt_pd(rsq20);
 798             rinv30           = gmx_mm_invsqrt_pd(rsq30);
 799
 800             rinvsq10         = _mm_mul_pd(rinv10,rinv10);
 801             rinvsq20         = _mm_mul_pd(rinv20,rinv20);
 802             rinvsq30         = _mm_mul_pd(rinv30,rinv30);
 803
 804             /* Load parameters for j particles */
 805             jq0              = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
 806             vdwjidx0A        = 2*vdwtype[jnrA+0];
 807             vdwjidx0B        = 2*vdwtype[jnrB+0];
 808
 809             fjx0             = _mm_setzero_pd();
 810             fjy0             = _mm_setzero_pd();
 811             fjz0             = _mm_setzero_pd();
 812
 813             /**************************
 814              * CALCULATE INTERACTIONS *
 815              **************************/
 816
 817             r00              = _mm_mul_pd(rsq00,rinv00);
 818
 819             /* Compute parameters for interactions between i and j atoms */
 820             gmx_mm_load_2pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,
 821                                          vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
 822
 823             /* Calculate table index by multiplying r with table scale and truncate to integer */
 824             rt               = _mm_mul_pd(r00,vftabscale);
 825             vfitab           = _mm_cvttpd_epi32(rt);
 826             vfeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
 827             vfitab           = _mm_slli_epi32(vfitab,3);
 828
 829             /* CUBIC SPLINE TABLE DISPERSION */
 830             Y                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
 831             F                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) );
 832             GMX_MM_TRANSPOSE2_PD(Y,F);
 833             G                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
 834             H                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) +2);
 835             GMX_MM_TRANSPOSE2_PD(G,H);
 836             Heps             = _mm_mul_pd(vfeps,H);
 837             Fp               = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
 838             FF               = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
 839             fvdw6            = _mm_mul_pd(c6_00,FF);
 840
 841             /* CUBIC SPLINE TABLE REPULSION */
 842             vfitab           = _mm_add_epi32(vfitab,ifour);
 843             Y                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
 844             F                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) );
 845             GMX_MM_TRANSPOSE2_PD(Y,F);
 846             G                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
 847             H                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) +2);
 848             GMX_MM_TRANSPOSE2_PD(G,H);
 849             Heps             = _mm_mul_pd(vfeps,H);
 850             Fp               = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
 851             FF               = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
 852             fvdw12           = _mm_mul_pd(c12_00,FF);
 853             fvdw             = _mm_xor_pd(signbit,_mm_mul_pd(_mm_add_pd(fvdw6,fvdw12),_mm_mul_pd(vftabscale,rinv00)));
 854
 855             fscal            = fvdw;
 856
 857             /* Calculate temporary vectorial force */
 858             tx               = _mm_mul_pd(fscal,dx00);
 859             ty               = _mm_mul_pd(fscal,dy00);
 860             tz               = _mm_mul_pd(fscal,dz00);
 861
 862             /* Update vectorial force */
 863             fix0             = _mm_add_pd(fix0,tx);
 864             fiy0             = _mm_add_pd(fiy0,ty);
 865             fiz0             = _mm_add_pd(fiz0,tz);
 866
 867             fjx0             = _mm_add_pd(fjx0,tx);
 868             fjy0             = _mm_add_pd(fjy0,ty);
 869             fjz0             = _mm_add_pd(fjz0,tz);
 870
 871             /**************************
 872              * CALCULATE INTERACTIONS *
 873              **************************/
 874
 875             /* Compute parameters for interactions between i and j atoms */
 876             qq10             = _mm_mul_pd(iq1,jq0);
 877
 878             /* COULOMB ELECTROSTATICS */
 879             velec            = _mm_mul_pd(qq10,rinv10);
 880             felec            = _mm_mul_pd(velec,rinvsq10);
 881
 882             fscal            = felec;
 883
 884             /* Calculate temporary vectorial force */
 885             tx               = _mm_mul_pd(fscal,dx10);
 886             ty               = _mm_mul_pd(fscal,dy10);
 887             tz               = _mm_mul_pd(fscal,dz10);
 888
 889             /* Update vectorial force */
 890             fix1             = _mm_add_pd(fix1,tx);
 891             fiy1             = _mm_add_pd(fiy1,ty);
 892             fiz1             = _mm_add_pd(fiz1,tz);
 893
 894             fjx0             = _mm_add_pd(fjx0,tx);
 895             fjy0             = _mm_add_pd(fjy0,ty);
 896             fjz0             = _mm_add_pd(fjz0,tz);
 897
 898             /**************************
 899              * CALCULATE INTERACTIONS *
 900              **************************/
 901
 902             /* Compute parameters for interactions between i and j atoms */
 903             qq20             = _mm_mul_pd(iq2,jq0);
 904
 905             /* COULOMB ELECTROSTATICS */
 906             velec            = _mm_mul_pd(qq20,rinv20);
 907             felec            = _mm_mul_pd(velec,rinvsq20);
 908
 909             fscal            = felec;
 910
 911             /* Calculate temporary vectorial force */
 912             tx               = _mm_mul_pd(fscal,dx20);
 913             ty               = _mm_mul_pd(fscal,dy20);
 914             tz               = _mm_mul_pd(fscal,dz20);
 915
 916             /* Update vectorial force */
 917             fix2             = _mm_add_pd(fix2,tx);
 918             fiy2             = _mm_add_pd(fiy2,ty);
 919             fiz2             = _mm_add_pd(fiz2,tz);
 920
 921             fjx0             = _mm_add_pd(fjx0,tx);
 922             fjy0             = _mm_add_pd(fjy0,ty);
 923             fjz0             = _mm_add_pd(fjz0,tz);
 924
 925             /**************************
 926              * CALCULATE INTERACTIONS *
 927              **************************/
 928
 929             /* Compute parameters for interactions between i and j atoms */
 930             qq30             = _mm_mul_pd(iq3,jq0);
 931
 932             /* COULOMB ELECTROSTATICS */
 933             velec            = _mm_mul_pd(qq30,rinv30);
 934             felec            = _mm_mul_pd(velec,rinvsq30);
 935
 936             fscal            = felec;
 937
 938             /* Calculate temporary vectorial force */
 939             tx               = _mm_mul_pd(fscal,dx30);
 940             ty               = _mm_mul_pd(fscal,dy30);
 941             tz               = _mm_mul_pd(fscal,dz30);
 942
 943             /* Update vectorial force */
 944             fix3             = _mm_add_pd(fix3,tx);
 945             fiy3             = _mm_add_pd(fiy3,ty);
 946             fiz3             = _mm_add_pd(fiz3,tz);
 947
 948             fjx0             = _mm_add_pd(fjx0,tx);
 949             fjy0             = _mm_add_pd(fjy0,ty);
 950             fjz0             = _mm_add_pd(fjz0,tz);
 951
 952             gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,fjx0,fjy0,fjz0);
 953
 954             /* Inner loop uses 132 flops */
 955         }
 956
 957         if(jidx<j_index_end)
 958         {
 959
 960             jnrA             = jjnr[jidx];
 961             j_coord_offsetA  = DIM*jnrA;
 962
 963             /* load j atom coordinates */
 964             gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
 965                                               &jx0,&jy0,&jz0);
 966
 967             /* Calculate displacement vector */
 968             dx00             = _mm_sub_pd(ix0,jx0);
 969             dy00             = _mm_sub_pd(iy0,jy0);
 970             dz00             = _mm_sub_pd(iz0,jz0);
 971             dx10             = _mm_sub_pd(ix1,jx0);
 972             dy10             = _mm_sub_pd(iy1,jy0);
 973             dz10             = _mm_sub_pd(iz1,jz0);
 974             dx20             = _mm_sub_pd(ix2,jx0);
 975             dy20             = _mm_sub_pd(iy2,jy0);
 976             dz20             = _mm_sub_pd(iz2,jz0);
 977             dx30             = _mm_sub_pd(ix3,jx0);
 978             dy30             = _mm_sub_pd(iy3,jy0);
 979             dz30             = _mm_sub_pd(iz3,jz0);
 980
 981             /* Calculate squared distance and things based on it */
 982             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 983             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
 984             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
 985             rsq30            = gmx_mm_calc_rsq_pd(dx30,dy30,dz30);
 986
 987             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 988             rinv10           = gmx_mm_invsqrt_pd(rsq10);
 989             rinv20           = gmx_mm_invsqrt_pd(rsq20);
 990             rinv30           = gmx_mm_invsqrt_pd(rsq30);
 991
 992             rinvsq10         = _mm_mul_pd(rinv10,rinv10);
 993             rinvsq20         = _mm_mul_pd(rinv20,rinv20);
 994             rinvsq30         = _mm_mul_pd(rinv30,rinv30);
 995
 996             /* Load parameters for j particles */
 997             jq0              = _mm_load_sd(charge+jnrA+0);
 998             vdwjidx0A        = 2*vdwtype[jnrA+0];
 999
1000             fjx0             = _mm_setzero_pd();
1001             fjy0             = _mm_setzero_pd();
1002             fjz0             = _mm_setzero_pd();
1003
1004             /**************************
1005              * CALCULATE INTERACTIONS *
1006              **************************/
1007
1008             r00              = _mm_mul_pd(rsq00,rinv00);
1009
1010             /* Compute parameters for interactions between i and j atoms */
1011             gmx_mm_load_1pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,&c6_00,&c12_00);
1012
1013             /* Calculate table index by multiplying r with table scale and truncate to integer */
1014             rt               = _mm_mul_pd(r00,vftabscale);
1015             vfitab           = _mm_cvttpd_epi32(rt);
1016             vfeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
1017             vfitab           = _mm_slli_epi32(vfitab,3);
1018
1019             /* CUBIC SPLINE TABLE DISPERSION */
1020             Y                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
1021             F                = _mm_setzero_pd();
1022             GMX_MM_TRANSPOSE2_PD(Y,F);
1023             G                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
1024             H                = _mm_setzero_pd();
1025             GMX_MM_TRANSPOSE2_PD(G,H);
1026             Heps             = _mm_mul_pd(vfeps,H);
1027             Fp               = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
1028             FF               = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
1029             fvdw6            = _mm_mul_pd(c6_00,FF);
1030
1031             /* CUBIC SPLINE TABLE REPULSION */
1032             vfitab           = _mm_add_epi32(vfitab,ifour);
1033             Y                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
1034             F                = _mm_setzero_pd();
1035             GMX_MM_TRANSPOSE2_PD(Y,F);
1036             G                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
1037             H                = _mm_setzero_pd();
1038             GMX_MM_TRANSPOSE2_PD(G,H);
1039             Heps             = _mm_mul_pd(vfeps,H);
1040             Fp               = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
1041             FF               = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
1042             fvdw12           = _mm_mul_pd(c12_00,FF);
1043             fvdw             = _mm_xor_pd(signbit,_mm_mul_pd(_mm_add_pd(fvdw6,fvdw12),_mm_mul_pd(vftabscale,rinv00)));
1044
1045             fscal            = fvdw;
1046
1047             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1048
1049             /* Calculate temporary vectorial force */
1050             tx               = _mm_mul_pd(fscal,dx00);
1051             ty               = _mm_mul_pd(fscal,dy00);
1052             tz               = _mm_mul_pd(fscal,dz00);
1053
1054             /* Update vectorial force */
1055             fix0             = _mm_add_pd(fix0,tx);
1056             fiy0             = _mm_add_pd(fiy0,ty);
1057             fiz0             = _mm_add_pd(fiz0,tz);
1058
1059             fjx0             = _mm_add_pd(fjx0,tx);
1060             fjy0             = _mm_add_pd(fjy0,ty);
1061             fjz0             = _mm_add_pd(fjz0,tz);
1062
1063             /**************************
1064              * CALCULATE INTERACTIONS *
1065              **************************/
1066
1067             /* Compute parameters for interactions between i and j atoms */
1068             qq10             = _mm_mul_pd(iq1,jq0);
1069
1070             /* COULOMB ELECTROSTATICS */
1071             velec            = _mm_mul_pd(qq10,rinv10);
1072             felec            = _mm_mul_pd(velec,rinvsq10);
1073
1074             fscal            = felec;
1075
1076             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1077
1078             /* Calculate temporary vectorial force */
1079             tx               = _mm_mul_pd(fscal,dx10);
1080             ty               = _mm_mul_pd(fscal,dy10);
1081             tz               = _mm_mul_pd(fscal,dz10);
1082
1083             /* Update vectorial force */
1084             fix1             = _mm_add_pd(fix1,tx);
1085             fiy1             = _mm_add_pd(fiy1,ty);
1086             fiz1             = _mm_add_pd(fiz1,tz);
1087
1088             fjx0             = _mm_add_pd(fjx0,tx);
1089             fjy0             = _mm_add_pd(fjy0,ty);
1090             fjz0             = _mm_add_pd(fjz0,tz);
1091
1092             /**************************
1093              * CALCULATE INTERACTIONS *
1094              **************************/
1095
1096             /* Compute parameters for interactions between i and j atoms */
1097             qq20             = _mm_mul_pd(iq2,jq0);
1098
1099             /* COULOMB ELECTROSTATICS */
1100             velec            = _mm_mul_pd(qq20,rinv20);
1101             felec            = _mm_mul_pd(velec,rinvsq20);
1102
1103             fscal            = felec;
1104
1105             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1106
1107             /* Calculate temporary vectorial force */
1108             tx               = _mm_mul_pd(fscal,dx20);
1109             ty               = _mm_mul_pd(fscal,dy20);
1110             tz               = _mm_mul_pd(fscal,dz20);
1111
1112             /* Update vectorial force */
1113             fix2             = _mm_add_pd(fix2,tx);
1114             fiy2             = _mm_add_pd(fiy2,ty);
1115             fiz2             = _mm_add_pd(fiz2,tz);
1116
1117             fjx0             = _mm_add_pd(fjx0,tx);
1118             fjy0             = _mm_add_pd(fjy0,ty);
1119             fjz0             = _mm_add_pd(fjz0,tz);
1120
1121             /**************************
1122              * CALCULATE INTERACTIONS *
1123              **************************/
1124
1125             /* Compute parameters for interactions between i and j atoms */
1126             qq30             = _mm_mul_pd(iq3,jq0);
1127
1128             /* COULOMB ELECTROSTATICS */
1129             velec            = _mm_mul_pd(qq30,rinv30);
1130             felec            = _mm_mul_pd(velec,rinvsq30);
1131
1132             fscal            = felec;
1133
1134             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1135
1136             /* Calculate temporary vectorial force */
1137             tx               = _mm_mul_pd(fscal,dx30);
1138             ty               = _mm_mul_pd(fscal,dy30);
1139             tz               = _mm_mul_pd(fscal,dz30);
1140
1141             /* Update vectorial force */
1142             fix3             = _mm_add_pd(fix3,tx);
1143             fiy3             = _mm_add_pd(fiy3,ty);
1144             fiz3             = _mm_add_pd(fiz3,tz);
1145
1146             fjx0             = _mm_add_pd(fjx0,tx);
1147             fjy0             = _mm_add_pd(fjy0,ty);
1148             fjz0             = _mm_add_pd(fjz0,tz);
1149
1150             gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,fjx0,fjy0,fjz0);
1151
1152             /* Inner loop uses 132 flops */
1153         }
1154
1155         /* End of innermost loop */
1156
1157         gmx_mm_update_iforce_4atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
1158                                               f+i_coord_offset,fshift+i_shift_offset);
1159
1160         /* Increment number of inner iterations */
1161         inneriter                  += j_index_end - j_index_start;
1162
1163         /* Outer loop uses 24 flops */
1164     }
1165
1166     /* Increment number of outer iterations */
1167     outeriter        += nri;
1168
1169     /* Update outer/inner flops */
1170
1171     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W4_F,outeriter*24 + inneriter*132);
1172 }