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

   1 /*
   2  * This file is part of the GROMACS molecular simulation package.
   3  *
   4  * Copyright (c) 2012,2013, 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 #ifdef HAVE_CONFIG_H
  39 #include <config.h>
  40 #endif
  41
  42 #include <math.h>
  43
  44 #include "../nb_kernel.h"
  45 #include "types/simple.h"
  46 #include "vec.h"
  47 #include "nrnb.h"
  48
  49 #include "gromacs/simd/math_x86_sse2_double.h"
  50 #include "kernelutil_x86_sse2_double.h"
  51
  52 /*
  53  * Gromacs nonbonded kernel:   nb_kernel_ElecRFCut_VdwCSTab_GeomW4P1_VF_sse2_double
  54  * Electrostatics interaction: ReactionField
  55  * VdW interaction:            CubicSplineTable
  56  * Geometry:                   Water4-Particle
  57  * Calculate force/pot:        PotentialAndForce
  58  */
  59 void
  60 nb_kernel_ElecRFCut_VdwCSTab_GeomW4P1_VF_sse2_double
  61                     (t_nblist                    * gmx_restrict       nlist,
  62                      rvec                        * gmx_restrict          xx,
  63                      rvec                        * gmx_restrict          ff,
  64                      t_forcerec                  * gmx_restrict          fr,
  65                      t_mdatoms                   * gmx_restrict     mdatoms,
  66                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
  67                      t_nrnb                      * gmx_restrict        nrnb)
  68 {
  69     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
  70      * just 0 for non-waters.
  71      * Suffixes A,B refer to j loop unrolling done with SSE double precision, e.g. for the two different
  72      * jnr indices corresponding to data put in the four positions in the SIMD register.
  73      */
  74     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
  75     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
  76     int              jnrA,jnrB;
  77     int              j_coord_offsetA,j_coord_offsetB;
  78     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
  79     real             rcutoff_scalar;
  80     real             *shiftvec,*fshift,*x,*f;
  81     __m128d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
  82     int              vdwioffset0;
  83     __m128d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
  84     int              vdwioffset1;
  85     __m128d          ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
  86     int              vdwioffset2;
  87     __m128d          ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
  88     int              vdwioffset3;
  89     __m128d          ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
  90     int              vdwjidx0A,vdwjidx0B;
  91     __m128d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  92     __m128d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  93     __m128d          dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
  94     __m128d          dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
  95     __m128d          dx30,dy30,dz30,rsq30,rinv30,rinvsq30,r30,qq30,c6_30,c12_30;
  96     __m128d          velec,felec,velecsum,facel,crf,krf,krf2;
  97     real             *charge;
  98     int              nvdwtype;
  99     __m128d          rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 100     int              *vdwtype;
 101     real             *vdwparam;
 102     __m128d          one_sixth   = _mm_set1_pd(1.0/6.0);
 103     __m128d          one_twelfth = _mm_set1_pd(1.0/12.0);
 104     __m128i          vfitab;
 105     __m128i          ifour       = _mm_set1_epi32(4);
 106     __m128d          rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF;
 107     real             *vftab;
 108     __m128d          dummy_mask,cutoff_mask;
 109     __m128d          signbit   = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
 110     __m128d          one     = _mm_set1_pd(1.0);
 111     __m128d          two     = _mm_set1_pd(2.0);
 112     x                = xx[0];
 113     f                = ff[0];
 114
 115     nri              = nlist->nri;
 116     iinr             = nlist->iinr;
 117     jindex           = nlist->jindex;
 118     jjnr             = nlist->jjnr;
 119     shiftidx         = nlist->shift;
 120     gid              = nlist->gid;
 121     shiftvec         = fr->shift_vec[0];
 122     fshift           = fr->fshift[0];
 123     facel            = _mm_set1_pd(fr->epsfac);
 124     charge           = mdatoms->chargeA;
 125     krf              = _mm_set1_pd(fr->ic->k_rf);
 126     krf2             = _mm_set1_pd(fr->ic->k_rf*2.0);
 127     crf              = _mm_set1_pd(fr->ic->c_rf);
 128     nvdwtype         = fr->ntype;
 129     vdwparam         = fr->nbfp;
 130     vdwtype          = mdatoms->typeA;
 131
 132     vftab            = kernel_data->table_vdw->data;
 133     vftabscale       = _mm_set1_pd(kernel_data->table_vdw->scale);
 134
 135     /* Setup water-specific parameters */
 136     inr              = nlist->iinr[0];
 137     iq1              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+1]));
 138     iq2              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+2]));
 139     iq3              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+3]));
 140     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 141
 142     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 143     rcutoff_scalar   = fr->rcoulomb;
 144     rcutoff          = _mm_set1_pd(rcutoff_scalar);
 145     rcutoff2         = _mm_mul_pd(rcutoff,rcutoff);
 146
 147     /* Avoid stupid compiler warnings */
 148     jnrA = jnrB = 0;
 149     j_coord_offsetA = 0;
 150     j_coord_offsetB = 0;
 151
 152     outeriter        = 0;
 153     inneriter        = 0;
 154
 155     /* Start outer loop over neighborlists */
 156     for(iidx=0; iidx<nri; iidx++)
 157     {
 158         /* Load shift vector for this list */
 159         i_shift_offset   = DIM*shiftidx[iidx];
 160
 161         /* Load limits for loop over neighbors */
 162         j_index_start    = jindex[iidx];
 163         j_index_end      = jindex[iidx+1];
 164
 165         /* Get outer coordinate index */
 166         inr              = iinr[iidx];
 167         i_coord_offset   = DIM*inr;
 168
 169         /* Load i particle coords and add shift vector */
 170         gmx_mm_load_shift_and_4rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
 171                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
 172
 173         fix0             = _mm_setzero_pd();
 174         fiy0             = _mm_setzero_pd();
 175         fiz0             = _mm_setzero_pd();
 176         fix1             = _mm_setzero_pd();
 177         fiy1             = _mm_setzero_pd();
 178         fiz1             = _mm_setzero_pd();
 179         fix2             = _mm_setzero_pd();
 180         fiy2             = _mm_setzero_pd();
 181         fiz2             = _mm_setzero_pd();
 182         fix3             = _mm_setzero_pd();
 183         fiy3             = _mm_setzero_pd();
 184         fiz3             = _mm_setzero_pd();
 185
 186         /* Reset potential sums */
 187         velecsum         = _mm_setzero_pd();
 188         vvdwsum          = _mm_setzero_pd();
 189
 190         /* Start inner kernel loop */
 191         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 192         {
 193
 194             /* Get j neighbor index, and coordinate index */
 195             jnrA             = jjnr[jidx];
 196             jnrB             = jjnr[jidx+1];
 197             j_coord_offsetA  = DIM*jnrA;
 198             j_coord_offsetB  = DIM*jnrB;
 199
 200             /* load j atom coordinates */
 201             gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 202                                               &jx0,&jy0,&jz0);
 203
 204             /* Calculate displacement vector */
 205             dx00             = _mm_sub_pd(ix0,jx0);
 206             dy00             = _mm_sub_pd(iy0,jy0);
 207             dz00             = _mm_sub_pd(iz0,jz0);
 208             dx10             = _mm_sub_pd(ix1,jx0);
 209             dy10             = _mm_sub_pd(iy1,jy0);
 210             dz10             = _mm_sub_pd(iz1,jz0);
 211             dx20             = _mm_sub_pd(ix2,jx0);
 212             dy20             = _mm_sub_pd(iy2,jy0);
 213             dz20             = _mm_sub_pd(iz2,jz0);
 214             dx30             = _mm_sub_pd(ix3,jx0);
 215             dy30             = _mm_sub_pd(iy3,jy0);
 216             dz30             = _mm_sub_pd(iz3,jz0);
 217
 218             /* Calculate squared distance and things based on it */
 219             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 220             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
 221             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
 222             rsq30            = gmx_mm_calc_rsq_pd(dx30,dy30,dz30);
 223
 224             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 225             rinv10           = gmx_mm_invsqrt_pd(rsq10);
 226             rinv20           = gmx_mm_invsqrt_pd(rsq20);
 227             rinv30           = gmx_mm_invsqrt_pd(rsq30);
 228
 229             rinvsq10         = _mm_mul_pd(rinv10,rinv10);
 230             rinvsq20         = _mm_mul_pd(rinv20,rinv20);
 231             rinvsq30         = _mm_mul_pd(rinv30,rinv30);
 232
 233             /* Load parameters for j particles */
 234             jq0              = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
 235             vdwjidx0A        = 2*vdwtype[jnrA+0];
 236             vdwjidx0B        = 2*vdwtype[jnrB+0];
 237
 238             fjx0             = _mm_setzero_pd();
 239             fjy0             = _mm_setzero_pd();
 240             fjz0             = _mm_setzero_pd();
 241
 242             /**************************
 243              * CALCULATE INTERACTIONS *
 244              **************************/
 245
 246             r00              = _mm_mul_pd(rsq00,rinv00);
 247
 248             /* Compute parameters for interactions between i and j atoms */
 249             gmx_mm_load_2pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,
 250                                          vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
 251
 252             /* Calculate table index by multiplying r with table scale and truncate to integer */
 253             rt               = _mm_mul_pd(r00,vftabscale);
 254             vfitab           = _mm_cvttpd_epi32(rt);
 255             vfeps            = _mm_sub_pd(rt,_mm_cvtepi32_pd(vfitab));
 256             vfitab           = _mm_slli_epi32(vfitab,3);
 257
 258             /* CUBIC SPLINE TABLE DISPERSION */
 259             Y                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
 260             F                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) );
 261             GMX_MM_TRANSPOSE2_PD(Y,F);
 262             G                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
 263             H                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) +2);
 264             GMX_MM_TRANSPOSE2_PD(G,H);
 265             Heps             = _mm_mul_pd(vfeps,H);
 266             Fp               = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
 267             VV               = _mm_add_pd(Y,_mm_mul_pd(vfeps,Fp));
 268             vvdw6            = _mm_mul_pd(c6_00,VV);
 269             FF               = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
 270             fvdw6            = _mm_mul_pd(c6_00,FF);
 271
 272             /* CUBIC SPLINE TABLE REPULSION */
 273             vfitab           = _mm_add_epi32(vfitab,ifour);
 274             Y                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
 275             F                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) );
 276             GMX_MM_TRANSPOSE2_PD(Y,F);
 277             G                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
 278             H                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) +2);
 279             GMX_MM_TRANSPOSE2_PD(G,H);
 280             Heps             = _mm_mul_pd(vfeps,H);
 281             Fp               = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
 282             VV               = _mm_add_pd(Y,_mm_mul_pd(vfeps,Fp));
 283             vvdw12           = _mm_mul_pd(c12_00,VV);
 284             FF               = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
 285             fvdw12           = _mm_mul_pd(c12_00,FF);
 286             vvdw             = _mm_add_pd(vvdw12,vvdw6);
 287             fvdw             = _mm_xor_pd(signbit,_mm_mul_pd(_mm_add_pd(fvdw6,fvdw12),_mm_mul_pd(vftabscale,rinv00)));
 288
 289             /* Update potential sum for this i atom from the interaction with this j atom. */
 290             vvdwsum          = _mm_add_pd(vvdwsum,vvdw);
 291
 292             fscal            = fvdw;
 293
 294             /* Calculate temporary vectorial force */
 295             tx               = _mm_mul_pd(fscal,dx00);
 296             ty               = _mm_mul_pd(fscal,dy00);
 297             tz               = _mm_mul_pd(fscal,dz00);
 298
 299             /* Update vectorial force */
 300             fix0             = _mm_add_pd(fix0,tx);
 301             fiy0             = _mm_add_pd(fiy0,ty);
 302             fiz0             = _mm_add_pd(fiz0,tz);
 303
 304             fjx0             = _mm_add_pd(fjx0,tx);
 305             fjy0             = _mm_add_pd(fjy0,ty);
 306             fjz0             = _mm_add_pd(fjz0,tz);
 307
 308             /**************************
 309              * CALCULATE INTERACTIONS *
 310              **************************/
 311
 312             if (gmx_mm_any_lt(rsq10,rcutoff2))
 313             {
 314
 315             /* Compute parameters for interactions between i and j atoms */
 316             qq10             = _mm_mul_pd(iq1,jq0);
 317
 318             /* REACTION-FIELD ELECTROSTATICS */
 319             velec            = _mm_mul_pd(qq10,_mm_sub_pd(_mm_add_pd(rinv10,_mm_mul_pd(krf,rsq10)),crf));
 320             felec            = _mm_mul_pd(qq10,_mm_sub_pd(_mm_mul_pd(rinv10,rinvsq10),krf2));
 321
 322             cutoff_mask      = _mm_cmplt_pd(rsq10,rcutoff2);
 323
 324             /* Update potential sum for this i atom from the interaction with this j atom. */
 325             velec            = _mm_and_pd(velec,cutoff_mask);
 326             velecsum         = _mm_add_pd(velecsum,velec);
 327
 328             fscal            = felec;
 329
 330             fscal            = _mm_and_pd(fscal,cutoff_mask);
 331
 332             /* Calculate temporary vectorial force */
 333             tx               = _mm_mul_pd(fscal,dx10);
 334             ty               = _mm_mul_pd(fscal,dy10);
 335             tz               = _mm_mul_pd(fscal,dz10);
 336
 337             /* Update vectorial force */
 338             fix1             = _mm_add_pd(fix1,tx);
 339             fiy1             = _mm_add_pd(fiy1,ty);
 340             fiz1             = _mm_add_pd(fiz1,tz);
 341
 342             fjx0             = _mm_add_pd(fjx0,tx);
 343             fjy0             = _mm_add_pd(fjy0,ty);
 344             fjz0             = _mm_add_pd(fjz0,tz);
 345
 346             }
 347
 348             /**************************
 349              * CALCULATE INTERACTIONS *
 350              **************************/
 351
 352             if (gmx_mm_any_lt(rsq20,rcutoff2))
 353             {
 354
 355             /* Compute parameters for interactions between i and j atoms */
 356             qq20             = _mm_mul_pd(iq2,jq0);
 357
 358             /* REACTION-FIELD ELECTROSTATICS */
 359             velec            = _mm_mul_pd(qq20,_mm_sub_pd(_mm_add_pd(rinv20,_mm_mul_pd(krf,rsq20)),crf));
 360             felec            = _mm_mul_pd(qq20,_mm_sub_pd(_mm_mul_pd(rinv20,rinvsq20),krf2));
 361
 362             cutoff_mask      = _mm_cmplt_pd(rsq20,rcutoff2);
 363
 364             /* Update potential sum for this i atom from the interaction with this j atom. */
 365             velec            = _mm_and_pd(velec,cutoff_mask);
 366             velecsum         = _mm_add_pd(velecsum,velec);
 367
 368             fscal            = felec;
 369
 370             fscal            = _mm_and_pd(fscal,cutoff_mask);
 371
 372             /* Calculate temporary vectorial force */
 373             tx               = _mm_mul_pd(fscal,dx20);
 374             ty               = _mm_mul_pd(fscal,dy20);
 375             tz               = _mm_mul_pd(fscal,dz20);
 376
 377             /* Update vectorial force */
 378             fix2             = _mm_add_pd(fix2,tx);
 379             fiy2             = _mm_add_pd(fiy2,ty);
 380             fiz2             = _mm_add_pd(fiz2,tz);
 381
 382             fjx0             = _mm_add_pd(fjx0,tx);
 383             fjy0             = _mm_add_pd(fjy0,ty);
 384             fjz0             = _mm_add_pd(fjz0,tz);
 385
 386             }
 387
 388             /**************************
 389              * CALCULATE INTERACTIONS *
 390              **************************/
 391
 392             if (gmx_mm_any_lt(rsq30,rcutoff2))
 393             {
 394
 395             /* Compute parameters for interactions between i and j atoms */
 396             qq30             = _mm_mul_pd(iq3,jq0);
 397
 398             /* REACTION-FIELD ELECTROSTATICS */
 399             velec            = _mm_mul_pd(qq30,_mm_sub_pd(_mm_add_pd(rinv30,_mm_mul_pd(krf,rsq30)),crf));
 400             felec            = _mm_mul_pd(qq30,_mm_sub_pd(_mm_mul_pd(rinv30,rinvsq30),krf2));
 401
 402             cutoff_mask      = _mm_cmplt_pd(rsq30,rcutoff2);
 403
 404             /* Update potential sum for this i atom from the interaction with this j atom. */
 405             velec            = _mm_and_pd(velec,cutoff_mask);
 406             velecsum         = _mm_add_pd(velecsum,velec);
 407
 408             fscal            = felec;
 409
 410             fscal            = _mm_and_pd(fscal,cutoff_mask);
 411
 412             /* Calculate temporary vectorial force */
 413             tx               = _mm_mul_pd(fscal,dx30);
 414             ty               = _mm_mul_pd(fscal,dy30);
 415             tz               = _mm_mul_pd(fscal,dz30);
 416
 417             /* Update vectorial force */
 418             fix3             = _mm_add_pd(fix3,tx);
 419             fiy3             = _mm_add_pd(fiy3,ty);
 420             fiz3             = _mm_add_pd(fiz3,tz);
 421
 422             fjx0             = _mm_add_pd(fjx0,tx);
 423             fjy0             = _mm_add_pd(fjy0,ty);
 424             fjz0             = _mm_add_pd(fjz0,tz);
 425
 426             }
 427
 428             gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,fjx0,fjy0,fjz0);
 429
 430             /* Inner loop uses 167 flops */
 431         }
 432
 433         if(jidx<j_index_end)
 434         {
 435
 436             jnrA             = jjnr[jidx];
 437             j_coord_offsetA  = DIM*jnrA;
 438
 439             /* load j atom coordinates */
 440             gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
 441                                               &jx0,&jy0,&jz0);
 442
 443             /* Calculate displacement vector */
 444             dx00             = _mm_sub_pd(ix0,jx0);
 445             dy00             = _mm_sub_pd(iy0,jy0);
 446             dz00             = _mm_sub_pd(iz0,jz0);
 447             dx10             = _mm_sub_pd(ix1,jx0);
 448             dy10             = _mm_sub_pd(iy1,jy0);
 449             dz10             = _mm_sub_pd(iz1,jz0);
 450             dx20             = _mm_sub_pd(ix2,jx0);
 451             dy20             = _mm_sub_pd(iy2,jy0);
 452             dz20             = _mm_sub_pd(iz2,jz0);
 453             dx30             = _mm_sub_pd(ix3,jx0);
 454             dy30             = _mm_sub_pd(iy3,jy0);
 455             dz30             = _mm_sub_pd(iz3,jz0);
 456
 457             /* Calculate squared distance and things based on it */
 458             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 459             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
 460             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
 461             rsq30            = gmx_mm_calc_rsq_pd(dx30,dy30,dz30);
 462
 463             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 464             rinv10           = gmx_mm_invsqrt_pd(rsq10);
 465             rinv20           = gmx_mm_invsqrt_pd(rsq20);
 466             rinv30           = gmx_mm_invsqrt_pd(rsq30);
 467
 468             rinvsq10         = _mm_mul_pd(rinv10,rinv10);
 469             rinvsq20         = _mm_mul_pd(rinv20,rinv20);
 470             rinvsq30         = _mm_mul_pd(rinv30,rinv30);
 471
 472             /* Load parameters for j particles */
 473             jq0              = _mm_load_sd(charge+jnrA+0);
 474             vdwjidx0A        = 2*vdwtype[jnrA+0];
 475
 476             fjx0             = _mm_setzero_pd();
 477             fjy0             = _mm_setzero_pd();
 478             fjz0             = _mm_setzero_pd();
 479
 480             /**************************
 481              * CALCULATE INTERACTIONS *
 482              **************************/
 483
 484             r00              = _mm_mul_pd(rsq00,rinv00);
 485
 486             /* Compute parameters for interactions between i and j atoms */
 487             gmx_mm_load_1pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,&c6_00,&c12_00);
 488
 489             /* Calculate table index by multiplying r with table scale and truncate to integer */
 490             rt               = _mm_mul_pd(r00,vftabscale);
 491             vfitab           = _mm_cvttpd_epi32(rt);
 492             vfeps            = _mm_sub_pd(rt,_mm_cvtepi32_pd(vfitab));
 493             vfitab           = _mm_slli_epi32(vfitab,3);
 494
 495             /* CUBIC SPLINE TABLE DISPERSION */
 496             Y                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
 497             F                = _mm_setzero_pd();
 498             GMX_MM_TRANSPOSE2_PD(Y,F);
 499             G                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
 500             H                = _mm_setzero_pd();
 501             GMX_MM_TRANSPOSE2_PD(G,H);
 502             Heps             = _mm_mul_pd(vfeps,H);
 503             Fp               = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
 504             VV               = _mm_add_pd(Y,_mm_mul_pd(vfeps,Fp));
 505             vvdw6            = _mm_mul_pd(c6_00,VV);
 506             FF               = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
 507             fvdw6            = _mm_mul_pd(c6_00,FF);
 508
 509             /* CUBIC SPLINE TABLE REPULSION */
 510             vfitab           = _mm_add_epi32(vfitab,ifour);
 511             Y                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
 512             F                = _mm_setzero_pd();
 513             GMX_MM_TRANSPOSE2_PD(Y,F);
 514             G                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
 515             H                = _mm_setzero_pd();
 516             GMX_MM_TRANSPOSE2_PD(G,H);
 517             Heps             = _mm_mul_pd(vfeps,H);
 518             Fp               = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
 519             VV               = _mm_add_pd(Y,_mm_mul_pd(vfeps,Fp));
 520             vvdw12           = _mm_mul_pd(c12_00,VV);
 521             FF               = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
 522             fvdw12           = _mm_mul_pd(c12_00,FF);
 523             vvdw             = _mm_add_pd(vvdw12,vvdw6);
 524             fvdw             = _mm_xor_pd(signbit,_mm_mul_pd(_mm_add_pd(fvdw6,fvdw12),_mm_mul_pd(vftabscale,rinv00)));
 525
 526             /* Update potential sum for this i atom from the interaction with this j atom. */
 527             vvdw             = _mm_unpacklo_pd(vvdw,_mm_setzero_pd());
 528             vvdwsum          = _mm_add_pd(vvdwsum,vvdw);
 529
 530             fscal            = fvdw;
 531
 532             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 533
 534             /* Calculate temporary vectorial force */
 535             tx               = _mm_mul_pd(fscal,dx00);
 536             ty               = _mm_mul_pd(fscal,dy00);
 537             tz               = _mm_mul_pd(fscal,dz00);
 538
 539             /* Update vectorial force */
 540             fix0             = _mm_add_pd(fix0,tx);
 541             fiy0             = _mm_add_pd(fiy0,ty);
 542             fiz0             = _mm_add_pd(fiz0,tz);
 543
 544             fjx0             = _mm_add_pd(fjx0,tx);
 545             fjy0             = _mm_add_pd(fjy0,ty);
 546             fjz0             = _mm_add_pd(fjz0,tz);
 547
 548             /**************************
 549              * CALCULATE INTERACTIONS *
 550              **************************/
 551
 552             if (gmx_mm_any_lt(rsq10,rcutoff2))
 553             {
 554
 555             /* Compute parameters for interactions between i and j atoms */
 556             qq10             = _mm_mul_pd(iq1,jq0);
 557
 558             /* REACTION-FIELD ELECTROSTATICS */
 559             velec            = _mm_mul_pd(qq10,_mm_sub_pd(_mm_add_pd(rinv10,_mm_mul_pd(krf,rsq10)),crf));
 560             felec            = _mm_mul_pd(qq10,_mm_sub_pd(_mm_mul_pd(rinv10,rinvsq10),krf2));
 561
 562             cutoff_mask      = _mm_cmplt_pd(rsq10,rcutoff2);
 563
 564             /* Update potential sum for this i atom from the interaction with this j atom. */
 565             velec            = _mm_and_pd(velec,cutoff_mask);
 566             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 567             velecsum         = _mm_add_pd(velecsum,velec);
 568
 569             fscal            = felec;
 570
 571             fscal            = _mm_and_pd(fscal,cutoff_mask);
 572
 573             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 574
 575             /* Calculate temporary vectorial force */
 576             tx               = _mm_mul_pd(fscal,dx10);
 577             ty               = _mm_mul_pd(fscal,dy10);
 578             tz               = _mm_mul_pd(fscal,dz10);
 579
 580             /* Update vectorial force */
 581             fix1             = _mm_add_pd(fix1,tx);
 582             fiy1             = _mm_add_pd(fiy1,ty);
 583             fiz1             = _mm_add_pd(fiz1,tz);
 584
 585             fjx0             = _mm_add_pd(fjx0,tx);
 586             fjy0             = _mm_add_pd(fjy0,ty);
 587             fjz0             = _mm_add_pd(fjz0,tz);
 588
 589             }
 590
 591             /**************************
 592              * CALCULATE INTERACTIONS *
 593              **************************/
 594
 595             if (gmx_mm_any_lt(rsq20,rcutoff2))
 596             {
 597
 598             /* Compute parameters for interactions between i and j atoms */
 599             qq20             = _mm_mul_pd(iq2,jq0);
 600
 601             /* REACTION-FIELD ELECTROSTATICS */
 602             velec            = _mm_mul_pd(qq20,_mm_sub_pd(_mm_add_pd(rinv20,_mm_mul_pd(krf,rsq20)),crf));
 603             felec            = _mm_mul_pd(qq20,_mm_sub_pd(_mm_mul_pd(rinv20,rinvsq20),krf2));
 604
 605             cutoff_mask      = _mm_cmplt_pd(rsq20,rcutoff2);
 606
 607             /* Update potential sum for this i atom from the interaction with this j atom. */
 608             velec            = _mm_and_pd(velec,cutoff_mask);
 609             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 610             velecsum         = _mm_add_pd(velecsum,velec);
 611
 612             fscal            = felec;
 613
 614             fscal            = _mm_and_pd(fscal,cutoff_mask);
 615
 616             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 617
 618             /* Calculate temporary vectorial force */
 619             tx               = _mm_mul_pd(fscal,dx20);
 620             ty               = _mm_mul_pd(fscal,dy20);
 621             tz               = _mm_mul_pd(fscal,dz20);
 622
 623             /* Update vectorial force */
 624             fix2             = _mm_add_pd(fix2,tx);
 625             fiy2             = _mm_add_pd(fiy2,ty);
 626             fiz2             = _mm_add_pd(fiz2,tz);
 627
 628             fjx0             = _mm_add_pd(fjx0,tx);
 629             fjy0             = _mm_add_pd(fjy0,ty);
 630             fjz0             = _mm_add_pd(fjz0,tz);
 631
 632             }
 633
 634             /**************************
 635              * CALCULATE INTERACTIONS *
 636              **************************/
 637
 638             if (gmx_mm_any_lt(rsq30,rcutoff2))
 639             {
 640
 641             /* Compute parameters for interactions between i and j atoms */
 642             qq30             = _mm_mul_pd(iq3,jq0);
 643
 644             /* REACTION-FIELD ELECTROSTATICS */
 645             velec            = _mm_mul_pd(qq30,_mm_sub_pd(_mm_add_pd(rinv30,_mm_mul_pd(krf,rsq30)),crf));
 646             felec            = _mm_mul_pd(qq30,_mm_sub_pd(_mm_mul_pd(rinv30,rinvsq30),krf2));
 647
 648             cutoff_mask      = _mm_cmplt_pd(rsq30,rcutoff2);
 649
 650             /* Update potential sum for this i atom from the interaction with this j atom. */
 651             velec            = _mm_and_pd(velec,cutoff_mask);
 652             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 653             velecsum         = _mm_add_pd(velecsum,velec);
 654
 655             fscal            = felec;
 656
 657             fscal            = _mm_and_pd(fscal,cutoff_mask);
 658
 659             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 660
 661             /* Calculate temporary vectorial force */
 662             tx               = _mm_mul_pd(fscal,dx30);
 663             ty               = _mm_mul_pd(fscal,dy30);
 664             tz               = _mm_mul_pd(fscal,dz30);
 665
 666             /* Update vectorial force */
 667             fix3             = _mm_add_pd(fix3,tx);
 668             fiy3             = _mm_add_pd(fiy3,ty);
 669             fiz3             = _mm_add_pd(fiz3,tz);
 670
 671             fjx0             = _mm_add_pd(fjx0,tx);
 672             fjy0             = _mm_add_pd(fjy0,ty);
 673             fjz0             = _mm_add_pd(fjz0,tz);
 674
 675             }
 676
 677             gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,fjx0,fjy0,fjz0);
 678
 679             /* Inner loop uses 167 flops */
 680         }
 681
 682         /* End of innermost loop */
 683
 684         gmx_mm_update_iforce_4atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
 685                                               f+i_coord_offset,fshift+i_shift_offset);
 686
 687         ggid                        = gid[iidx];
 688         /* Update potential energies */
 689         gmx_mm_update_1pot_pd(velecsum,kernel_data->energygrp_elec+ggid);
 690         gmx_mm_update_1pot_pd(vvdwsum,kernel_data->energygrp_vdw+ggid);
 691
 692         /* Increment number of inner iterations */
 693         inneriter                  += j_index_end - j_index_start;
 694
 695         /* Outer loop uses 26 flops */
 696     }
 697
 698     /* Increment number of outer iterations */
 699     outeriter        += nri;
 700
 701     /* Update outer/inner flops */
 702
 703     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W4_VF,outeriter*26 + inneriter*167);
 704 }
 705 /*
 706  * Gromacs nonbonded kernel:   nb_kernel_ElecRFCut_VdwCSTab_GeomW4P1_F_sse2_double
 707  * Electrostatics interaction: ReactionField
 708  * VdW interaction:            CubicSplineTable
 709  * Geometry:                   Water4-Particle
 710  * Calculate force/pot:        Force
 711  */
 712 void
 713 nb_kernel_ElecRFCut_VdwCSTab_GeomW4P1_F_sse2_double
 714                     (t_nblist                    * gmx_restrict       nlist,
 715                      rvec                        * gmx_restrict          xx,
 716                      rvec                        * gmx_restrict          ff,
 717                      t_forcerec                  * gmx_restrict          fr,
 718                      t_mdatoms                   * gmx_restrict     mdatoms,
 719                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 720                      t_nrnb                      * gmx_restrict        nrnb)
 721 {
 722     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 723      * just 0 for non-waters.
 724      * Suffixes A,B refer to j loop unrolling done with SSE double precision, e.g. for the two different
 725      * jnr indices corresponding to data put in the four positions in the SIMD register.
 726      */
 727     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 728     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 729     int              jnrA,jnrB;
 730     int              j_coord_offsetA,j_coord_offsetB;
 731     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 732     real             rcutoff_scalar;
 733     real             *shiftvec,*fshift,*x,*f;
 734     __m128d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 735     int              vdwioffset0;
 736     __m128d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 737     int              vdwioffset1;
 738     __m128d          ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
 739     int              vdwioffset2;
 740     __m128d          ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
 741     int              vdwioffset3;
 742     __m128d          ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
 743     int              vdwjidx0A,vdwjidx0B;
 744     __m128d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 745     __m128d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 746     __m128d          dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
 747     __m128d          dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
 748     __m128d          dx30,dy30,dz30,rsq30,rinv30,rinvsq30,r30,qq30,c6_30,c12_30;
 749     __m128d          velec,felec,velecsum,facel,crf,krf,krf2;
 750     real             *charge;
 751     int              nvdwtype;
 752     __m128d          rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 753     int              *vdwtype;
 754     real             *vdwparam;
 755     __m128d          one_sixth   = _mm_set1_pd(1.0/6.0);
 756     __m128d          one_twelfth = _mm_set1_pd(1.0/12.0);
 757     __m128i          vfitab;
 758     __m128i          ifour       = _mm_set1_epi32(4);
 759     __m128d          rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF;
 760     real             *vftab;
 761     __m128d          dummy_mask,cutoff_mask;
 762     __m128d          signbit   = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
 763     __m128d          one     = _mm_set1_pd(1.0);
 764     __m128d          two     = _mm_set1_pd(2.0);
 765     x                = xx[0];
 766     f                = ff[0];
 767
 768     nri              = nlist->nri;
 769     iinr             = nlist->iinr;
 770     jindex           = nlist->jindex;
 771     jjnr             = nlist->jjnr;
 772     shiftidx         = nlist->shift;
 773     gid              = nlist->gid;
 774     shiftvec         = fr->shift_vec[0];
 775     fshift           = fr->fshift[0];
 776     facel            = _mm_set1_pd(fr->epsfac);
 777     charge           = mdatoms->chargeA;
 778     krf              = _mm_set1_pd(fr->ic->k_rf);
 779     krf2             = _mm_set1_pd(fr->ic->k_rf*2.0);
 780     crf              = _mm_set1_pd(fr->ic->c_rf);
 781     nvdwtype         = fr->ntype;
 782     vdwparam         = fr->nbfp;
 783     vdwtype          = mdatoms->typeA;
 784
 785     vftab            = kernel_data->table_vdw->data;
 786     vftabscale       = _mm_set1_pd(kernel_data->table_vdw->scale);
 787
 788     /* Setup water-specific parameters */
 789     inr              = nlist->iinr[0];
 790     iq1              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+1]));
 791     iq2              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+2]));
 792     iq3              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+3]));
 793     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 794
 795     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 796     rcutoff_scalar   = fr->rcoulomb;
 797     rcutoff          = _mm_set1_pd(rcutoff_scalar);
 798     rcutoff2         = _mm_mul_pd(rcutoff,rcutoff);
 799
 800     /* Avoid stupid compiler warnings */
 801     jnrA = jnrB = 0;
 802     j_coord_offsetA = 0;
 803     j_coord_offsetB = 0;
 804
 805     outeriter        = 0;
 806     inneriter        = 0;
 807
 808     /* Start outer loop over neighborlists */
 809     for(iidx=0; iidx<nri; iidx++)
 810     {
 811         /* Load shift vector for this list */
 812         i_shift_offset   = DIM*shiftidx[iidx];
 813
 814         /* Load limits for loop over neighbors */
 815         j_index_start    = jindex[iidx];
 816         j_index_end      = jindex[iidx+1];
 817
 818         /* Get outer coordinate index */
 819         inr              = iinr[iidx];
 820         i_coord_offset   = DIM*inr;
 821
 822         /* Load i particle coords and add shift vector */
 823         gmx_mm_load_shift_and_4rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
 824                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
 825
 826         fix0             = _mm_setzero_pd();
 827         fiy0             = _mm_setzero_pd();
 828         fiz0             = _mm_setzero_pd();
 829         fix1             = _mm_setzero_pd();
 830         fiy1             = _mm_setzero_pd();
 831         fiz1             = _mm_setzero_pd();
 832         fix2             = _mm_setzero_pd();
 833         fiy2             = _mm_setzero_pd();
 834         fiz2             = _mm_setzero_pd();
 835         fix3             = _mm_setzero_pd();
 836         fiy3             = _mm_setzero_pd();
 837         fiz3             = _mm_setzero_pd();
 838
 839         /* Start inner kernel loop */
 840         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 841         {
 842
 843             /* Get j neighbor index, and coordinate index */
 844             jnrA             = jjnr[jidx];
 845             jnrB             = jjnr[jidx+1];
 846             j_coord_offsetA  = DIM*jnrA;
 847             j_coord_offsetB  = DIM*jnrB;
 848
 849             /* load j atom coordinates */
 850             gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 851                                               &jx0,&jy0,&jz0);
 852
 853             /* Calculate displacement vector */
 854             dx00             = _mm_sub_pd(ix0,jx0);
 855             dy00             = _mm_sub_pd(iy0,jy0);
 856             dz00             = _mm_sub_pd(iz0,jz0);
 857             dx10             = _mm_sub_pd(ix1,jx0);
 858             dy10             = _mm_sub_pd(iy1,jy0);
 859             dz10             = _mm_sub_pd(iz1,jz0);
 860             dx20             = _mm_sub_pd(ix2,jx0);
 861             dy20             = _mm_sub_pd(iy2,jy0);
 862             dz20             = _mm_sub_pd(iz2,jz0);
 863             dx30             = _mm_sub_pd(ix3,jx0);
 864             dy30             = _mm_sub_pd(iy3,jy0);
 865             dz30             = _mm_sub_pd(iz3,jz0);
 866
 867             /* Calculate squared distance and things based on it */
 868             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 869             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
 870             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
 871             rsq30            = gmx_mm_calc_rsq_pd(dx30,dy30,dz30);
 872
 873             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 874             rinv10           = gmx_mm_invsqrt_pd(rsq10);
 875             rinv20           = gmx_mm_invsqrt_pd(rsq20);
 876             rinv30           = gmx_mm_invsqrt_pd(rsq30);
 877
 878             rinvsq10         = _mm_mul_pd(rinv10,rinv10);
 879             rinvsq20         = _mm_mul_pd(rinv20,rinv20);
 880             rinvsq30         = _mm_mul_pd(rinv30,rinv30);
 881
 882             /* Load parameters for j particles */
 883             jq0              = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
 884             vdwjidx0A        = 2*vdwtype[jnrA+0];
 885             vdwjidx0B        = 2*vdwtype[jnrB+0];
 886
 887             fjx0             = _mm_setzero_pd();
 888             fjy0             = _mm_setzero_pd();
 889             fjz0             = _mm_setzero_pd();
 890
 891             /**************************
 892              * CALCULATE INTERACTIONS *
 893              **************************/
 894
 895             r00              = _mm_mul_pd(rsq00,rinv00);
 896
 897             /* Compute parameters for interactions between i and j atoms */
 898             gmx_mm_load_2pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,
 899                                          vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
 900
 901             /* Calculate table index by multiplying r with table scale and truncate to integer */
 902             rt               = _mm_mul_pd(r00,vftabscale);
 903             vfitab           = _mm_cvttpd_epi32(rt);
 904             vfeps            = _mm_sub_pd(rt,_mm_cvtepi32_pd(vfitab));
 905             vfitab           = _mm_slli_epi32(vfitab,3);
 906
 907             /* CUBIC SPLINE TABLE DISPERSION */
 908             Y                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
 909             F                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) );
 910             GMX_MM_TRANSPOSE2_PD(Y,F);
 911             G                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
 912             H                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) +2);
 913             GMX_MM_TRANSPOSE2_PD(G,H);
 914             Heps             = _mm_mul_pd(vfeps,H);
 915             Fp               = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
 916             FF               = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
 917             fvdw6            = _mm_mul_pd(c6_00,FF);
 918
 919             /* CUBIC SPLINE TABLE REPULSION */
 920             vfitab           = _mm_add_epi32(vfitab,ifour);
 921             Y                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
 922             F                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) );
 923             GMX_MM_TRANSPOSE2_PD(Y,F);
 924             G                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
 925             H                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) +2);
 926             GMX_MM_TRANSPOSE2_PD(G,H);
 927             Heps             = _mm_mul_pd(vfeps,H);
 928             Fp               = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
 929             FF               = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
 930             fvdw12           = _mm_mul_pd(c12_00,FF);
 931             fvdw             = _mm_xor_pd(signbit,_mm_mul_pd(_mm_add_pd(fvdw6,fvdw12),_mm_mul_pd(vftabscale,rinv00)));
 932
 933             fscal            = fvdw;
 934
 935             /* Calculate temporary vectorial force */
 936             tx               = _mm_mul_pd(fscal,dx00);
 937             ty               = _mm_mul_pd(fscal,dy00);
 938             tz               = _mm_mul_pd(fscal,dz00);
 939
 940             /* Update vectorial force */
 941             fix0             = _mm_add_pd(fix0,tx);
 942             fiy0             = _mm_add_pd(fiy0,ty);
 943             fiz0             = _mm_add_pd(fiz0,tz);
 944
 945             fjx0             = _mm_add_pd(fjx0,tx);
 946             fjy0             = _mm_add_pd(fjy0,ty);
 947             fjz0             = _mm_add_pd(fjz0,tz);
 948
 949             /**************************
 950              * CALCULATE INTERACTIONS *
 951              **************************/
 952
 953             if (gmx_mm_any_lt(rsq10,rcutoff2))
 954             {
 955
 956             /* Compute parameters for interactions between i and j atoms */
 957             qq10             = _mm_mul_pd(iq1,jq0);
 958
 959             /* REACTION-FIELD ELECTROSTATICS */
 960             felec            = _mm_mul_pd(qq10,_mm_sub_pd(_mm_mul_pd(rinv10,rinvsq10),krf2));
 961
 962             cutoff_mask      = _mm_cmplt_pd(rsq10,rcutoff2);
 963
 964             fscal            = felec;
 965
 966             fscal            = _mm_and_pd(fscal,cutoff_mask);
 967
 968             /* Calculate temporary vectorial force */
 969             tx               = _mm_mul_pd(fscal,dx10);
 970             ty               = _mm_mul_pd(fscal,dy10);
 971             tz               = _mm_mul_pd(fscal,dz10);
 972
 973             /* Update vectorial force */
 974             fix1             = _mm_add_pd(fix1,tx);
 975             fiy1             = _mm_add_pd(fiy1,ty);
 976             fiz1             = _mm_add_pd(fiz1,tz);
 977
 978             fjx0             = _mm_add_pd(fjx0,tx);
 979             fjy0             = _mm_add_pd(fjy0,ty);
 980             fjz0             = _mm_add_pd(fjz0,tz);
 981
 982             }
 983
 984             /**************************
 985              * CALCULATE INTERACTIONS *
 986              **************************/
 987
 988             if (gmx_mm_any_lt(rsq20,rcutoff2))
 989             {
 990
 991             /* Compute parameters for interactions between i and j atoms */
 992             qq20             = _mm_mul_pd(iq2,jq0);
 993
 994             /* REACTION-FIELD ELECTROSTATICS */
 995             felec            = _mm_mul_pd(qq20,_mm_sub_pd(_mm_mul_pd(rinv20,rinvsq20),krf2));
 996
 997             cutoff_mask      = _mm_cmplt_pd(rsq20,rcutoff2);
 998
 999             fscal            = felec;
1000
1001             fscal            = _mm_and_pd(fscal,cutoff_mask);
1002
1003             /* Calculate temporary vectorial force */
1004             tx               = _mm_mul_pd(fscal,dx20);
1005             ty               = _mm_mul_pd(fscal,dy20);
1006             tz               = _mm_mul_pd(fscal,dz20);
1007
1008             /* Update vectorial force */
1009             fix2             = _mm_add_pd(fix2,tx);
1010             fiy2             = _mm_add_pd(fiy2,ty);
1011             fiz2             = _mm_add_pd(fiz2,tz);
1012
1013             fjx0             = _mm_add_pd(fjx0,tx);
1014             fjy0             = _mm_add_pd(fjy0,ty);
1015             fjz0             = _mm_add_pd(fjz0,tz);
1016
1017             }
1018
1019             /**************************
1020              * CALCULATE INTERACTIONS *
1021              **************************/
1022
1023             if (gmx_mm_any_lt(rsq30,rcutoff2))
1024             {
1025
1026             /* Compute parameters for interactions between i and j atoms */
1027             qq30             = _mm_mul_pd(iq3,jq0);
1028
1029             /* REACTION-FIELD ELECTROSTATICS */
1030             felec            = _mm_mul_pd(qq30,_mm_sub_pd(_mm_mul_pd(rinv30,rinvsq30),krf2));
1031
1032             cutoff_mask      = _mm_cmplt_pd(rsq30,rcutoff2);
1033
1034             fscal            = felec;
1035
1036             fscal            = _mm_and_pd(fscal,cutoff_mask);
1037
1038             /* Calculate temporary vectorial force */
1039             tx               = _mm_mul_pd(fscal,dx30);
1040             ty               = _mm_mul_pd(fscal,dy30);
1041             tz               = _mm_mul_pd(fscal,dz30);
1042
1043             /* Update vectorial force */
1044             fix3             = _mm_add_pd(fix3,tx);
1045             fiy3             = _mm_add_pd(fiy3,ty);
1046             fiz3             = _mm_add_pd(fiz3,tz);
1047
1048             fjx0             = _mm_add_pd(fjx0,tx);
1049             fjy0             = _mm_add_pd(fjy0,ty);
1050             fjz0             = _mm_add_pd(fjz0,tz);
1051
1052             }
1053
1054             gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,fjx0,fjy0,fjz0);
1055
1056             /* Inner loop uses 141 flops */
1057         }
1058
1059         if(jidx<j_index_end)
1060         {
1061
1062             jnrA             = jjnr[jidx];
1063             j_coord_offsetA  = DIM*jnrA;
1064
1065             /* load j atom coordinates */
1066             gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
1067                                               &jx0,&jy0,&jz0);
1068
1069             /* Calculate displacement vector */
1070             dx00             = _mm_sub_pd(ix0,jx0);
1071             dy00             = _mm_sub_pd(iy0,jy0);
1072             dz00             = _mm_sub_pd(iz0,jz0);
1073             dx10             = _mm_sub_pd(ix1,jx0);
1074             dy10             = _mm_sub_pd(iy1,jy0);
1075             dz10             = _mm_sub_pd(iz1,jz0);
1076             dx20             = _mm_sub_pd(ix2,jx0);
1077             dy20             = _mm_sub_pd(iy2,jy0);
1078             dz20             = _mm_sub_pd(iz2,jz0);
1079             dx30             = _mm_sub_pd(ix3,jx0);
1080             dy30             = _mm_sub_pd(iy3,jy0);
1081             dz30             = _mm_sub_pd(iz3,jz0);
1082
1083             /* Calculate squared distance and things based on it */
1084             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
1085             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
1086             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
1087             rsq30            = gmx_mm_calc_rsq_pd(dx30,dy30,dz30);
1088
1089             rinv00           = gmx_mm_invsqrt_pd(rsq00);
1090             rinv10           = gmx_mm_invsqrt_pd(rsq10);
1091             rinv20           = gmx_mm_invsqrt_pd(rsq20);
1092             rinv30           = gmx_mm_invsqrt_pd(rsq30);
1093
1094             rinvsq10         = _mm_mul_pd(rinv10,rinv10);
1095             rinvsq20         = _mm_mul_pd(rinv20,rinv20);
1096             rinvsq30         = _mm_mul_pd(rinv30,rinv30);
1097
1098             /* Load parameters for j particles */
1099             jq0              = _mm_load_sd(charge+jnrA+0);
1100             vdwjidx0A        = 2*vdwtype[jnrA+0];
1101
1102             fjx0             = _mm_setzero_pd();
1103             fjy0             = _mm_setzero_pd();
1104             fjz0             = _mm_setzero_pd();
1105
1106             /**************************
1107              * CALCULATE INTERACTIONS *
1108              **************************/
1109
1110             r00              = _mm_mul_pd(rsq00,rinv00);
1111
1112             /* Compute parameters for interactions between i and j atoms */
1113             gmx_mm_load_1pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,&c6_00,&c12_00);
1114
1115             /* Calculate table index by multiplying r with table scale and truncate to integer */
1116             rt               = _mm_mul_pd(r00,vftabscale);
1117             vfitab           = _mm_cvttpd_epi32(rt);
1118             vfeps            = _mm_sub_pd(rt,_mm_cvtepi32_pd(vfitab));
1119             vfitab           = _mm_slli_epi32(vfitab,3);
1120
1121             /* CUBIC SPLINE TABLE DISPERSION */
1122             Y                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
1123             F                = _mm_setzero_pd();
1124             GMX_MM_TRANSPOSE2_PD(Y,F);
1125             G                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
1126             H                = _mm_setzero_pd();
1127             GMX_MM_TRANSPOSE2_PD(G,H);
1128             Heps             = _mm_mul_pd(vfeps,H);
1129             Fp               = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
1130             FF               = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
1131             fvdw6            = _mm_mul_pd(c6_00,FF);
1132
1133             /* CUBIC SPLINE TABLE REPULSION */
1134             vfitab           = _mm_add_epi32(vfitab,ifour);
1135             Y                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
1136             F                = _mm_setzero_pd();
1137             GMX_MM_TRANSPOSE2_PD(Y,F);
1138             G                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
1139             H                = _mm_setzero_pd();
1140             GMX_MM_TRANSPOSE2_PD(G,H);
1141             Heps             = _mm_mul_pd(vfeps,H);
1142             Fp               = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
1143             FF               = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
1144             fvdw12           = _mm_mul_pd(c12_00,FF);
1145             fvdw             = _mm_xor_pd(signbit,_mm_mul_pd(_mm_add_pd(fvdw6,fvdw12),_mm_mul_pd(vftabscale,rinv00)));
1146
1147             fscal            = fvdw;
1148
1149             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1150
1151             /* Calculate temporary vectorial force */
1152             tx               = _mm_mul_pd(fscal,dx00);
1153             ty               = _mm_mul_pd(fscal,dy00);
1154             tz               = _mm_mul_pd(fscal,dz00);
1155
1156             /* Update vectorial force */
1157             fix0             = _mm_add_pd(fix0,tx);
1158             fiy0             = _mm_add_pd(fiy0,ty);
1159             fiz0             = _mm_add_pd(fiz0,tz);
1160
1161             fjx0             = _mm_add_pd(fjx0,tx);
1162             fjy0             = _mm_add_pd(fjy0,ty);
1163             fjz0             = _mm_add_pd(fjz0,tz);
1164
1165             /**************************
1166              * CALCULATE INTERACTIONS *
1167              **************************/
1168
1169             if (gmx_mm_any_lt(rsq10,rcutoff2))
1170             {
1171
1172             /* Compute parameters for interactions between i and j atoms */
1173             qq10             = _mm_mul_pd(iq1,jq0);
1174
1175             /* REACTION-FIELD ELECTROSTATICS */
1176             felec            = _mm_mul_pd(qq10,_mm_sub_pd(_mm_mul_pd(rinv10,rinvsq10),krf2));
1177
1178             cutoff_mask      = _mm_cmplt_pd(rsq10,rcutoff2);
1179
1180             fscal            = felec;
1181
1182             fscal            = _mm_and_pd(fscal,cutoff_mask);
1183
1184             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1185
1186             /* Calculate temporary vectorial force */
1187             tx               = _mm_mul_pd(fscal,dx10);
1188             ty               = _mm_mul_pd(fscal,dy10);
1189             tz               = _mm_mul_pd(fscal,dz10);
1190
1191             /* Update vectorial force */
1192             fix1             = _mm_add_pd(fix1,tx);
1193             fiy1             = _mm_add_pd(fiy1,ty);
1194             fiz1             = _mm_add_pd(fiz1,tz);
1195
1196             fjx0             = _mm_add_pd(fjx0,tx);
1197             fjy0             = _mm_add_pd(fjy0,ty);
1198             fjz0             = _mm_add_pd(fjz0,tz);
1199
1200             }
1201
1202             /**************************
1203              * CALCULATE INTERACTIONS *
1204              **************************/
1205
1206             if (gmx_mm_any_lt(rsq20,rcutoff2))
1207             {
1208
1209             /* Compute parameters for interactions between i and j atoms */
1210             qq20             = _mm_mul_pd(iq2,jq0);
1211
1212             /* REACTION-FIELD ELECTROSTATICS */
1213             felec            = _mm_mul_pd(qq20,_mm_sub_pd(_mm_mul_pd(rinv20,rinvsq20),krf2));
1214
1215             cutoff_mask      = _mm_cmplt_pd(rsq20,rcutoff2);
1216
1217             fscal            = felec;
1218
1219             fscal            = _mm_and_pd(fscal,cutoff_mask);
1220
1221             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1222
1223             /* Calculate temporary vectorial force */
1224             tx               = _mm_mul_pd(fscal,dx20);
1225             ty               = _mm_mul_pd(fscal,dy20);
1226             tz               = _mm_mul_pd(fscal,dz20);
1227
1228             /* Update vectorial force */
1229             fix2             = _mm_add_pd(fix2,tx);
1230             fiy2             = _mm_add_pd(fiy2,ty);
1231             fiz2             = _mm_add_pd(fiz2,tz);
1232
1233             fjx0             = _mm_add_pd(fjx0,tx);
1234             fjy0             = _mm_add_pd(fjy0,ty);
1235             fjz0             = _mm_add_pd(fjz0,tz);
1236
1237             }
1238
1239             /**************************
1240              * CALCULATE INTERACTIONS *
1241              **************************/
1242
1243             if (gmx_mm_any_lt(rsq30,rcutoff2))
1244             {
1245
1246             /* Compute parameters for interactions between i and j atoms */
1247             qq30             = _mm_mul_pd(iq3,jq0);
1248
1249             /* REACTION-FIELD ELECTROSTATICS */
1250             felec            = _mm_mul_pd(qq30,_mm_sub_pd(_mm_mul_pd(rinv30,rinvsq30),krf2));
1251
1252             cutoff_mask      = _mm_cmplt_pd(rsq30,rcutoff2);
1253
1254             fscal            = felec;
1255
1256             fscal            = _mm_and_pd(fscal,cutoff_mask);
1257
1258             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1259
1260             /* Calculate temporary vectorial force */
1261             tx               = _mm_mul_pd(fscal,dx30);
1262             ty               = _mm_mul_pd(fscal,dy30);
1263             tz               = _mm_mul_pd(fscal,dz30);
1264
1265             /* Update vectorial force */
1266             fix3             = _mm_add_pd(fix3,tx);
1267             fiy3             = _mm_add_pd(fiy3,ty);
1268             fiz3             = _mm_add_pd(fiz3,tz);
1269
1270             fjx0             = _mm_add_pd(fjx0,tx);
1271             fjy0             = _mm_add_pd(fjy0,ty);
1272             fjz0             = _mm_add_pd(fjz0,tz);
1273
1274             }
1275
1276             gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,fjx0,fjy0,fjz0);
1277
1278             /* Inner loop uses 141 flops */
1279         }
1280
1281         /* End of innermost loop */
1282
1283         gmx_mm_update_iforce_4atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
1284                                               f+i_coord_offset,fshift+i_shift_offset);
1285
1286         /* Increment number of inner iterations */
1287         inneriter                  += j_index_end - j_index_start;
1288
1289         /* Outer loop uses 24 flops */
1290     }
1291
1292     /* Increment number of outer iterations */
1293     outeriter        += nri;
1294
1295     /* Update outer/inner flops */
1296
1297     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W4_F,outeriter*24 + inneriter*141);
1298 }