src/gromacs/gmxlib/nonbonded/nb_kernel_sse2_double/nb_kernel_ElecRFCut_VdwCSTab_GeomW3P1_sse2_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 sse2_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_sse2_double.h"
  48 #include "kernelutil_x86_sse2_double.h"
  49
  50 /*
  51  * Gromacs nonbonded kernel:   nb_kernel_ElecRFCut_VdwCSTab_GeomW3P1_VF_sse2_double
  52  * Electrostatics interaction: ReactionField
  53  * VdW interaction:            CubicSplineTable
  54  * Geometry:                   Water3-Particle
  55  * Calculate force/pot:        PotentialAndForce
  56  */
  57 void
  58 nb_kernel_ElecRFCut_VdwCSTab_GeomW3P1_VF_sse2_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              vdwjidx0A,vdwjidx0B;
  87     __m128d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  88     __m128d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  89     __m128d          dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
  90     __m128d          dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
  91     __m128d          velec,felec,velecsum,facel,crf,krf,krf2;
  92     real             *charge;
  93     int              nvdwtype;
  94     __m128d          rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
  95     int              *vdwtype;
  96     real             *vdwparam;
  97     __m128d          one_sixth   = _mm_set1_pd(1.0/6.0);
  98     __m128d          one_twelfth = _mm_set1_pd(1.0/12.0);
  99     __m128i          vfitab;
 100     __m128i          ifour       = _mm_set1_epi32(4);
 101     __m128d          rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF;
 102     real             *vftab;
 103     __m128d          dummy_mask,cutoff_mask;
 104     __m128d          signbit   = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
 105     __m128d          one     = _mm_set1_pd(1.0);
 106     __m128d          two     = _mm_set1_pd(2.0);
 107     x                = xx[0];
 108     f                = ff[0];
 109
 110     nri              = nlist->nri;
 111     iinr             = nlist->iinr;
 112     jindex           = nlist->jindex;
 113     jjnr             = nlist->jjnr;
 114     shiftidx         = nlist->shift;
 115     gid              = nlist->gid;
 116     shiftvec         = fr->shift_vec[0];
 117     fshift           = fr->fshift[0];
 118     facel            = _mm_set1_pd(fr->epsfac);
 119     charge           = mdatoms->chargeA;
 120     krf              = _mm_set1_pd(fr->ic->k_rf);
 121     krf2             = _mm_set1_pd(fr->ic->k_rf*2.0);
 122     crf              = _mm_set1_pd(fr->ic->c_rf);
 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     iq0              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+0]));
 133     iq1              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+1]));
 134     iq2              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+2]));
 135     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 136
 137     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 138     rcutoff_scalar   = fr->rcoulomb;
 139     rcutoff          = _mm_set1_pd(rcutoff_scalar);
 140     rcutoff2         = _mm_mul_pd(rcutoff,rcutoff);
 141
 142     /* Avoid stupid compiler warnings */
 143     jnrA = jnrB = 0;
 144     j_coord_offsetA = 0;
 145     j_coord_offsetB = 0;
 146
 147     outeriter        = 0;
 148     inneriter        = 0;
 149
 150     /* Start outer loop over neighborlists */
 151     for(iidx=0; iidx<nri; iidx++)
 152     {
 153         /* Load shift vector for this list */
 154         i_shift_offset   = DIM*shiftidx[iidx];
 155
 156         /* Load limits for loop over neighbors */
 157         j_index_start    = jindex[iidx];
 158         j_index_end      = jindex[iidx+1];
 159
 160         /* Get outer coordinate index */
 161         inr              = iinr[iidx];
 162         i_coord_offset   = DIM*inr;
 163
 164         /* Load i particle coords and add shift vector */
 165         gmx_mm_load_shift_and_3rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
 166                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 167
 168         fix0             = _mm_setzero_pd();
 169         fiy0             = _mm_setzero_pd();
 170         fiz0             = _mm_setzero_pd();
 171         fix1             = _mm_setzero_pd();
 172         fiy1             = _mm_setzero_pd();
 173         fiz1             = _mm_setzero_pd();
 174         fix2             = _mm_setzero_pd();
 175         fiy2             = _mm_setzero_pd();
 176         fiz2             = _mm_setzero_pd();
 177
 178         /* Reset potential sums */
 179         velecsum         = _mm_setzero_pd();
 180         vvdwsum          = _mm_setzero_pd();
 181
 182         /* Start inner kernel loop */
 183         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 184         {
 185
 186             /* Get j neighbor index, and coordinate index */
 187             jnrA             = jjnr[jidx];
 188             jnrB             = jjnr[jidx+1];
 189             j_coord_offsetA  = DIM*jnrA;
 190             j_coord_offsetB  = DIM*jnrB;
 191
 192             /* load j atom coordinates */
 193             gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 194                                               &jx0,&jy0,&jz0);
 195
 196             /* Calculate displacement vector */
 197             dx00             = _mm_sub_pd(ix0,jx0);
 198             dy00             = _mm_sub_pd(iy0,jy0);
 199             dz00             = _mm_sub_pd(iz0,jz0);
 200             dx10             = _mm_sub_pd(ix1,jx0);
 201             dy10             = _mm_sub_pd(iy1,jy0);
 202             dz10             = _mm_sub_pd(iz1,jz0);
 203             dx20             = _mm_sub_pd(ix2,jx0);
 204             dy20             = _mm_sub_pd(iy2,jy0);
 205             dz20             = _mm_sub_pd(iz2,jz0);
 206
 207             /* Calculate squared distance and things based on it */
 208             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 209             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
 210             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
 211
 212             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 213             rinv10           = gmx_mm_invsqrt_pd(rsq10);
 214             rinv20           = gmx_mm_invsqrt_pd(rsq20);
 215
 216             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 217             rinvsq10         = _mm_mul_pd(rinv10,rinv10);
 218             rinvsq20         = _mm_mul_pd(rinv20,rinv20);
 219
 220             /* Load parameters for j particles */
 221             jq0              = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
 222             vdwjidx0A        = 2*vdwtype[jnrA+0];
 223             vdwjidx0B        = 2*vdwtype[jnrB+0];
 224
 225             fjx0             = _mm_setzero_pd();
 226             fjy0             = _mm_setzero_pd();
 227             fjz0             = _mm_setzero_pd();
 228
 229             /**************************
 230              * CALCULATE INTERACTIONS *
 231              **************************/
 232
 233             if (gmx_mm_any_lt(rsq00,rcutoff2))
 234             {
 235
 236             r00              = _mm_mul_pd(rsq00,rinv00);
 237
 238             /* Compute parameters for interactions between i and j atoms */
 239             qq00             = _mm_mul_pd(iq0,jq0);
 240             gmx_mm_load_2pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,
 241                                          vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
 242
 243             /* Calculate table index by multiplying r with table scale and truncate to integer */
 244             rt               = _mm_mul_pd(r00,vftabscale);
 245             vfitab           = _mm_cvttpd_epi32(rt);
 246             vfeps            = _mm_sub_pd(rt,_mm_cvtepi32_pd(vfitab));
 247             vfitab           = _mm_slli_epi32(vfitab,3);
 248
 249             /* REACTION-FIELD ELECTROSTATICS */
 250             velec            = _mm_mul_pd(qq00,_mm_sub_pd(_mm_add_pd(rinv00,_mm_mul_pd(krf,rsq00)),crf));
 251             felec            = _mm_mul_pd(qq00,_mm_sub_pd(_mm_mul_pd(rinv00,rinvsq00),krf2));
 252
 253             /* CUBIC SPLINE TABLE DISPERSION */
 254             Y                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
 255             F                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) );
 256             GMX_MM_TRANSPOSE2_PD(Y,F);
 257             G                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
 258             H                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) +2);
 259             GMX_MM_TRANSPOSE2_PD(G,H);
 260             Heps             = _mm_mul_pd(vfeps,H);
 261             Fp               = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
 262             VV               = _mm_add_pd(Y,_mm_mul_pd(vfeps,Fp));
 263             vvdw6            = _mm_mul_pd(c6_00,VV);
 264             FF               = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
 265             fvdw6            = _mm_mul_pd(c6_00,FF);
 266
 267             /* CUBIC SPLINE TABLE REPULSION */
 268             vfitab           = _mm_add_epi32(vfitab,ifour);
 269             Y                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
 270             F                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) );
 271             GMX_MM_TRANSPOSE2_PD(Y,F);
 272             G                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
 273             H                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) +2);
 274             GMX_MM_TRANSPOSE2_PD(G,H);
 275             Heps             = _mm_mul_pd(vfeps,H);
 276             Fp               = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
 277             VV               = _mm_add_pd(Y,_mm_mul_pd(vfeps,Fp));
 278             vvdw12           = _mm_mul_pd(c12_00,VV);
 279             FF               = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
 280             fvdw12           = _mm_mul_pd(c12_00,FF);
 281             vvdw             = _mm_add_pd(vvdw12,vvdw6);
 282             fvdw             = _mm_xor_pd(signbit,_mm_mul_pd(_mm_add_pd(fvdw6,fvdw12),_mm_mul_pd(vftabscale,rinv00)));
 283
 284             cutoff_mask      = _mm_cmplt_pd(rsq00,rcutoff2);
 285
 286             /* Update potential sum for this i atom from the interaction with this j atom. */
 287             velec            = _mm_and_pd(velec,cutoff_mask);
 288             velecsum         = _mm_add_pd(velecsum,velec);
 289             vvdw             = _mm_and_pd(vvdw,cutoff_mask);
 290             vvdwsum          = _mm_add_pd(vvdwsum,vvdw);
 291
 292             fscal            = _mm_add_pd(felec,fvdw);
 293
 294             fscal            = _mm_and_pd(fscal,cutoff_mask);
 295
 296             /* Calculate temporary vectorial force */
 297             tx               = _mm_mul_pd(fscal,dx00);
 298             ty               = _mm_mul_pd(fscal,dy00);
 299             tz               = _mm_mul_pd(fscal,dz00);
 300
 301             /* Update vectorial force */
 302             fix0             = _mm_add_pd(fix0,tx);
 303             fiy0             = _mm_add_pd(fiy0,ty);
 304             fiz0             = _mm_add_pd(fiz0,tz);
 305
 306             fjx0             = _mm_add_pd(fjx0,tx);
 307             fjy0             = _mm_add_pd(fjy0,ty);
 308             fjz0             = _mm_add_pd(fjz0,tz);
 309
 310             }
 311
 312             /**************************
 313              * CALCULATE INTERACTIONS *
 314              **************************/
 315
 316             if (gmx_mm_any_lt(rsq10,rcutoff2))
 317             {
 318
 319             /* Compute parameters for interactions between i and j atoms */
 320             qq10             = _mm_mul_pd(iq1,jq0);
 321
 322             /* REACTION-FIELD ELECTROSTATICS */
 323             velec            = _mm_mul_pd(qq10,_mm_sub_pd(_mm_add_pd(rinv10,_mm_mul_pd(krf,rsq10)),crf));
 324             felec            = _mm_mul_pd(qq10,_mm_sub_pd(_mm_mul_pd(rinv10,rinvsq10),krf2));
 325
 326             cutoff_mask      = _mm_cmplt_pd(rsq10,rcutoff2);
 327
 328             /* Update potential sum for this i atom from the interaction with this j atom. */
 329             velec            = _mm_and_pd(velec,cutoff_mask);
 330             velecsum         = _mm_add_pd(velecsum,velec);
 331
 332             fscal            = felec;
 333
 334             fscal            = _mm_and_pd(fscal,cutoff_mask);
 335
 336             /* Calculate temporary vectorial force */
 337             tx               = _mm_mul_pd(fscal,dx10);
 338             ty               = _mm_mul_pd(fscal,dy10);
 339             tz               = _mm_mul_pd(fscal,dz10);
 340
 341             /* Update vectorial force */
 342             fix1             = _mm_add_pd(fix1,tx);
 343             fiy1             = _mm_add_pd(fiy1,ty);
 344             fiz1             = _mm_add_pd(fiz1,tz);
 345
 346             fjx0             = _mm_add_pd(fjx0,tx);
 347             fjy0             = _mm_add_pd(fjy0,ty);
 348             fjz0             = _mm_add_pd(fjz0,tz);
 349
 350             }
 351
 352             /**************************
 353              * CALCULATE INTERACTIONS *
 354              **************************/
 355
 356             if (gmx_mm_any_lt(rsq20,rcutoff2))
 357             {
 358
 359             /* Compute parameters for interactions between i and j atoms */
 360             qq20             = _mm_mul_pd(iq2,jq0);
 361
 362             /* REACTION-FIELD ELECTROSTATICS */
 363             velec            = _mm_mul_pd(qq20,_mm_sub_pd(_mm_add_pd(rinv20,_mm_mul_pd(krf,rsq20)),crf));
 364             felec            = _mm_mul_pd(qq20,_mm_sub_pd(_mm_mul_pd(rinv20,rinvsq20),krf2));
 365
 366             cutoff_mask      = _mm_cmplt_pd(rsq20,rcutoff2);
 367
 368             /* Update potential sum for this i atom from the interaction with this j atom. */
 369             velec            = _mm_and_pd(velec,cutoff_mask);
 370             velecsum         = _mm_add_pd(velecsum,velec);
 371
 372             fscal            = felec;
 373
 374             fscal            = _mm_and_pd(fscal,cutoff_mask);
 375
 376             /* Calculate temporary vectorial force */
 377             tx               = _mm_mul_pd(fscal,dx20);
 378             ty               = _mm_mul_pd(fscal,dy20);
 379             tz               = _mm_mul_pd(fscal,dz20);
 380
 381             /* Update vectorial force */
 382             fix2             = _mm_add_pd(fix2,tx);
 383             fiy2             = _mm_add_pd(fiy2,ty);
 384             fiz2             = _mm_add_pd(fiz2,tz);
 385
 386             fjx0             = _mm_add_pd(fjx0,tx);
 387             fjy0             = _mm_add_pd(fjy0,ty);
 388             fjz0             = _mm_add_pd(fjz0,tz);
 389
 390             }
 391
 392             gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,fjx0,fjy0,fjz0);
 393
 394             /* Inner loop uses 147 flops */
 395         }
 396
 397         if(jidx<j_index_end)
 398         {
 399
 400             jnrA             = jjnr[jidx];
 401             j_coord_offsetA  = DIM*jnrA;
 402
 403             /* load j atom coordinates */
 404             gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
 405                                               &jx0,&jy0,&jz0);
 406
 407             /* Calculate displacement vector */
 408             dx00             = _mm_sub_pd(ix0,jx0);
 409             dy00             = _mm_sub_pd(iy0,jy0);
 410             dz00             = _mm_sub_pd(iz0,jz0);
 411             dx10             = _mm_sub_pd(ix1,jx0);
 412             dy10             = _mm_sub_pd(iy1,jy0);
 413             dz10             = _mm_sub_pd(iz1,jz0);
 414             dx20             = _mm_sub_pd(ix2,jx0);
 415             dy20             = _mm_sub_pd(iy2,jy0);
 416             dz20             = _mm_sub_pd(iz2,jz0);
 417
 418             /* Calculate squared distance and things based on it */
 419             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 420             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
 421             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
 422
 423             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 424             rinv10           = gmx_mm_invsqrt_pd(rsq10);
 425             rinv20           = gmx_mm_invsqrt_pd(rsq20);
 426
 427             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 428             rinvsq10         = _mm_mul_pd(rinv10,rinv10);
 429             rinvsq20         = _mm_mul_pd(rinv20,rinv20);
 430
 431             /* Load parameters for j particles */
 432             jq0              = _mm_load_sd(charge+jnrA+0);
 433             vdwjidx0A        = 2*vdwtype[jnrA+0];
 434
 435             fjx0             = _mm_setzero_pd();
 436             fjy0             = _mm_setzero_pd();
 437             fjz0             = _mm_setzero_pd();
 438
 439             /**************************
 440              * CALCULATE INTERACTIONS *
 441              **************************/
 442
 443             if (gmx_mm_any_lt(rsq00,rcutoff2))
 444             {
 445
 446             r00              = _mm_mul_pd(rsq00,rinv00);
 447
 448             /* Compute parameters for interactions between i and j atoms */
 449             qq00             = _mm_mul_pd(iq0,jq0);
 450             gmx_mm_load_1pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,&c6_00,&c12_00);
 451
 452             /* Calculate table index by multiplying r with table scale and truncate to integer */
 453             rt               = _mm_mul_pd(r00,vftabscale);
 454             vfitab           = _mm_cvttpd_epi32(rt);
 455             vfeps            = _mm_sub_pd(rt,_mm_cvtepi32_pd(vfitab));
 456             vfitab           = _mm_slli_epi32(vfitab,3);
 457
 458             /* REACTION-FIELD ELECTROSTATICS */
 459             velec            = _mm_mul_pd(qq00,_mm_sub_pd(_mm_add_pd(rinv00,_mm_mul_pd(krf,rsq00)),crf));
 460             felec            = _mm_mul_pd(qq00,_mm_sub_pd(_mm_mul_pd(rinv00,rinvsq00),krf2));
 461
 462             /* CUBIC SPLINE TABLE DISPERSION */
 463             Y                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
 464             F                = _mm_setzero_pd();
 465             GMX_MM_TRANSPOSE2_PD(Y,F);
 466             G                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
 467             H                = _mm_setzero_pd();
 468             GMX_MM_TRANSPOSE2_PD(G,H);
 469             Heps             = _mm_mul_pd(vfeps,H);
 470             Fp               = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
 471             VV               = _mm_add_pd(Y,_mm_mul_pd(vfeps,Fp));
 472             vvdw6            = _mm_mul_pd(c6_00,VV);
 473             FF               = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
 474             fvdw6            = _mm_mul_pd(c6_00,FF);
 475
 476             /* CUBIC SPLINE TABLE REPULSION */
 477             vfitab           = _mm_add_epi32(vfitab,ifour);
 478             Y                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
 479             F                = _mm_setzero_pd();
 480             GMX_MM_TRANSPOSE2_PD(Y,F);
 481             G                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
 482             H                = _mm_setzero_pd();
 483             GMX_MM_TRANSPOSE2_PD(G,H);
 484             Heps             = _mm_mul_pd(vfeps,H);
 485             Fp               = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
 486             VV               = _mm_add_pd(Y,_mm_mul_pd(vfeps,Fp));
 487             vvdw12           = _mm_mul_pd(c12_00,VV);
 488             FF               = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
 489             fvdw12           = _mm_mul_pd(c12_00,FF);
 490             vvdw             = _mm_add_pd(vvdw12,vvdw6);
 491             fvdw             = _mm_xor_pd(signbit,_mm_mul_pd(_mm_add_pd(fvdw6,fvdw12),_mm_mul_pd(vftabscale,rinv00)));
 492
 493             cutoff_mask      = _mm_cmplt_pd(rsq00,rcutoff2);
 494
 495             /* Update potential sum for this i atom from the interaction with this j atom. */
 496             velec            = _mm_and_pd(velec,cutoff_mask);
 497             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 498             velecsum         = _mm_add_pd(velecsum,velec);
 499             vvdw             = _mm_and_pd(vvdw,cutoff_mask);
 500             vvdw             = _mm_unpacklo_pd(vvdw,_mm_setzero_pd());
 501             vvdwsum          = _mm_add_pd(vvdwsum,vvdw);
 502
 503             fscal            = _mm_add_pd(felec,fvdw);
 504
 505             fscal            = _mm_and_pd(fscal,cutoff_mask);
 506
 507             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 508
 509             /* Calculate temporary vectorial force */
 510             tx               = _mm_mul_pd(fscal,dx00);
 511             ty               = _mm_mul_pd(fscal,dy00);
 512             tz               = _mm_mul_pd(fscal,dz00);
 513
 514             /* Update vectorial force */
 515             fix0             = _mm_add_pd(fix0,tx);
 516             fiy0             = _mm_add_pd(fiy0,ty);
 517             fiz0             = _mm_add_pd(fiz0,tz);
 518
 519             fjx0             = _mm_add_pd(fjx0,tx);
 520             fjy0             = _mm_add_pd(fjy0,ty);
 521             fjz0             = _mm_add_pd(fjz0,tz);
 522
 523             }
 524
 525             /**************************
 526              * CALCULATE INTERACTIONS *
 527              **************************/
 528
 529             if (gmx_mm_any_lt(rsq10,rcutoff2))
 530             {
 531
 532             /* Compute parameters for interactions between i and j atoms */
 533             qq10             = _mm_mul_pd(iq1,jq0);
 534
 535             /* REACTION-FIELD ELECTROSTATICS */
 536             velec            = _mm_mul_pd(qq10,_mm_sub_pd(_mm_add_pd(rinv10,_mm_mul_pd(krf,rsq10)),crf));
 537             felec            = _mm_mul_pd(qq10,_mm_sub_pd(_mm_mul_pd(rinv10,rinvsq10),krf2));
 538
 539             cutoff_mask      = _mm_cmplt_pd(rsq10,rcutoff2);
 540
 541             /* Update potential sum for this i atom from the interaction with this j atom. */
 542             velec            = _mm_and_pd(velec,cutoff_mask);
 543             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 544             velecsum         = _mm_add_pd(velecsum,velec);
 545
 546             fscal            = felec;
 547
 548             fscal            = _mm_and_pd(fscal,cutoff_mask);
 549
 550             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 551
 552             /* Calculate temporary vectorial force */
 553             tx               = _mm_mul_pd(fscal,dx10);
 554             ty               = _mm_mul_pd(fscal,dy10);
 555             tz               = _mm_mul_pd(fscal,dz10);
 556
 557             /* Update vectorial force */
 558             fix1             = _mm_add_pd(fix1,tx);
 559             fiy1             = _mm_add_pd(fiy1,ty);
 560             fiz1             = _mm_add_pd(fiz1,tz);
 561
 562             fjx0             = _mm_add_pd(fjx0,tx);
 563             fjy0             = _mm_add_pd(fjy0,ty);
 564             fjz0             = _mm_add_pd(fjz0,tz);
 565
 566             }
 567
 568             /**************************
 569              * CALCULATE INTERACTIONS *
 570              **************************/
 571
 572             if (gmx_mm_any_lt(rsq20,rcutoff2))
 573             {
 574
 575             /* Compute parameters for interactions between i and j atoms */
 576             qq20             = _mm_mul_pd(iq2,jq0);
 577
 578             /* REACTION-FIELD ELECTROSTATICS */
 579             velec            = _mm_mul_pd(qq20,_mm_sub_pd(_mm_add_pd(rinv20,_mm_mul_pd(krf,rsq20)),crf));
 580             felec            = _mm_mul_pd(qq20,_mm_sub_pd(_mm_mul_pd(rinv20,rinvsq20),krf2));
 581
 582             cutoff_mask      = _mm_cmplt_pd(rsq20,rcutoff2);
 583
 584             /* Update potential sum for this i atom from the interaction with this j atom. */
 585             velec            = _mm_and_pd(velec,cutoff_mask);
 586             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 587             velecsum         = _mm_add_pd(velecsum,velec);
 588
 589             fscal            = felec;
 590
 591             fscal            = _mm_and_pd(fscal,cutoff_mask);
 592
 593             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 594
 595             /* Calculate temporary vectorial force */
 596             tx               = _mm_mul_pd(fscal,dx20);
 597             ty               = _mm_mul_pd(fscal,dy20);
 598             tz               = _mm_mul_pd(fscal,dz20);
 599
 600             /* Update vectorial force */
 601             fix2             = _mm_add_pd(fix2,tx);
 602             fiy2             = _mm_add_pd(fiy2,ty);
 603             fiz2             = _mm_add_pd(fiz2,tz);
 604
 605             fjx0             = _mm_add_pd(fjx0,tx);
 606             fjy0             = _mm_add_pd(fjy0,ty);
 607             fjz0             = _mm_add_pd(fjz0,tz);
 608
 609             }
 610
 611             gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,fjx0,fjy0,fjz0);
 612
 613             /* Inner loop uses 147 flops */
 614         }
 615
 616         /* End of innermost loop */
 617
 618         gmx_mm_update_iforce_3atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
 619                                               f+i_coord_offset,fshift+i_shift_offset);
 620
 621         ggid                        = gid[iidx];
 622         /* Update potential energies */
 623         gmx_mm_update_1pot_pd(velecsum,kernel_data->energygrp_elec+ggid);
 624         gmx_mm_update_1pot_pd(vvdwsum,kernel_data->energygrp_vdw+ggid);
 625
 626         /* Increment number of inner iterations */
 627         inneriter                  += j_index_end - j_index_start;
 628
 629         /* Outer loop uses 20 flops */
 630     }
 631
 632     /* Increment number of outer iterations */
 633     outeriter        += nri;
 634
 635     /* Update outer/inner flops */
 636
 637     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3_VF,outeriter*20 + inneriter*147);
 638 }
 639 /*
 640  * Gromacs nonbonded kernel:   nb_kernel_ElecRFCut_VdwCSTab_GeomW3P1_F_sse2_double
 641  * Electrostatics interaction: ReactionField
 642  * VdW interaction:            CubicSplineTable
 643  * Geometry:                   Water3-Particle
 644  * Calculate force/pot:        Force
 645  */
 646 void
 647 nb_kernel_ElecRFCut_VdwCSTab_GeomW3P1_F_sse2_double
 648                     (t_nblist                    * gmx_restrict       nlist,
 649                      rvec                        * gmx_restrict          xx,
 650                      rvec                        * gmx_restrict          ff,
 651                      t_forcerec                  * gmx_restrict          fr,
 652                      t_mdatoms                   * gmx_restrict     mdatoms,
 653                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 654                      t_nrnb                      * gmx_restrict        nrnb)
 655 {
 656     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 657      * just 0 for non-waters.
 658      * Suffixes A,B refer to j loop unrolling done with SSE double precision, e.g. for the two different
 659      * jnr indices corresponding to data put in the four positions in the SIMD register.
 660      */
 661     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 662     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 663     int              jnrA,jnrB;
 664     int              j_coord_offsetA,j_coord_offsetB;
 665     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 666     real             rcutoff_scalar;
 667     real             *shiftvec,*fshift,*x,*f;
 668     __m128d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 669     int              vdwioffset0;
 670     __m128d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 671     int              vdwioffset1;
 672     __m128d          ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
 673     int              vdwioffset2;
 674     __m128d          ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
 675     int              vdwjidx0A,vdwjidx0B;
 676     __m128d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 677     __m128d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 678     __m128d          dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
 679     __m128d          dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
 680     __m128d          velec,felec,velecsum,facel,crf,krf,krf2;
 681     real             *charge;
 682     int              nvdwtype;
 683     __m128d          rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 684     int              *vdwtype;
 685     real             *vdwparam;
 686     __m128d          one_sixth   = _mm_set1_pd(1.0/6.0);
 687     __m128d          one_twelfth = _mm_set1_pd(1.0/12.0);
 688     __m128i          vfitab;
 689     __m128i          ifour       = _mm_set1_epi32(4);
 690     __m128d          rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF;
 691     real             *vftab;
 692     __m128d          dummy_mask,cutoff_mask;
 693     __m128d          signbit   = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
 694     __m128d          one     = _mm_set1_pd(1.0);
 695     __m128d          two     = _mm_set1_pd(2.0);
 696     x                = xx[0];
 697     f                = ff[0];
 698
 699     nri              = nlist->nri;
 700     iinr             = nlist->iinr;
 701     jindex           = nlist->jindex;
 702     jjnr             = nlist->jjnr;
 703     shiftidx         = nlist->shift;
 704     gid              = nlist->gid;
 705     shiftvec         = fr->shift_vec[0];
 706     fshift           = fr->fshift[0];
 707     facel            = _mm_set1_pd(fr->epsfac);
 708     charge           = mdatoms->chargeA;
 709     krf              = _mm_set1_pd(fr->ic->k_rf);
 710     krf2             = _mm_set1_pd(fr->ic->k_rf*2.0);
 711     crf              = _mm_set1_pd(fr->ic->c_rf);
 712     nvdwtype         = fr->ntype;
 713     vdwparam         = fr->nbfp;
 714     vdwtype          = mdatoms->typeA;
 715
 716     vftab            = kernel_data->table_vdw->data;
 717     vftabscale       = _mm_set1_pd(kernel_data->table_vdw->scale);
 718
 719     /* Setup water-specific parameters */
 720     inr              = nlist->iinr[0];
 721     iq0              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+0]));
 722     iq1              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+1]));
 723     iq2              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+2]));
 724     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 725
 726     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 727     rcutoff_scalar   = fr->rcoulomb;
 728     rcutoff          = _mm_set1_pd(rcutoff_scalar);
 729     rcutoff2         = _mm_mul_pd(rcutoff,rcutoff);
 730
 731     /* Avoid stupid compiler warnings */
 732     jnrA = jnrB = 0;
 733     j_coord_offsetA = 0;
 734     j_coord_offsetB = 0;
 735
 736     outeriter        = 0;
 737     inneriter        = 0;
 738
 739     /* Start outer loop over neighborlists */
 740     for(iidx=0; iidx<nri; iidx++)
 741     {
 742         /* Load shift vector for this list */
 743         i_shift_offset   = DIM*shiftidx[iidx];
 744
 745         /* Load limits for loop over neighbors */
 746         j_index_start    = jindex[iidx];
 747         j_index_end      = jindex[iidx+1];
 748
 749         /* Get outer coordinate index */
 750         inr              = iinr[iidx];
 751         i_coord_offset   = DIM*inr;
 752
 753         /* Load i particle coords and add shift vector */
 754         gmx_mm_load_shift_and_3rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
 755                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 756
 757         fix0             = _mm_setzero_pd();
 758         fiy0             = _mm_setzero_pd();
 759         fiz0             = _mm_setzero_pd();
 760         fix1             = _mm_setzero_pd();
 761         fiy1             = _mm_setzero_pd();
 762         fiz1             = _mm_setzero_pd();
 763         fix2             = _mm_setzero_pd();
 764         fiy2             = _mm_setzero_pd();
 765         fiz2             = _mm_setzero_pd();
 766
 767         /* Start inner kernel loop */
 768         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 769         {
 770
 771             /* Get j neighbor index, and coordinate index */
 772             jnrA             = jjnr[jidx];
 773             jnrB             = jjnr[jidx+1];
 774             j_coord_offsetA  = DIM*jnrA;
 775             j_coord_offsetB  = DIM*jnrB;
 776
 777             /* load j atom coordinates */
 778             gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 779                                               &jx0,&jy0,&jz0);
 780
 781             /* Calculate displacement vector */
 782             dx00             = _mm_sub_pd(ix0,jx0);
 783             dy00             = _mm_sub_pd(iy0,jy0);
 784             dz00             = _mm_sub_pd(iz0,jz0);
 785             dx10             = _mm_sub_pd(ix1,jx0);
 786             dy10             = _mm_sub_pd(iy1,jy0);
 787             dz10             = _mm_sub_pd(iz1,jz0);
 788             dx20             = _mm_sub_pd(ix2,jx0);
 789             dy20             = _mm_sub_pd(iy2,jy0);
 790             dz20             = _mm_sub_pd(iz2,jz0);
 791
 792             /* Calculate squared distance and things based on it */
 793             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 794             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
 795             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
 796
 797             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 798             rinv10           = gmx_mm_invsqrt_pd(rsq10);
 799             rinv20           = gmx_mm_invsqrt_pd(rsq20);
 800
 801             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 802             rinvsq10         = _mm_mul_pd(rinv10,rinv10);
 803             rinvsq20         = _mm_mul_pd(rinv20,rinv20);
 804
 805             /* Load parameters for j particles */
 806             jq0              = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
 807             vdwjidx0A        = 2*vdwtype[jnrA+0];
 808             vdwjidx0B        = 2*vdwtype[jnrB+0];
 809
 810             fjx0             = _mm_setzero_pd();
 811             fjy0             = _mm_setzero_pd();
 812             fjz0             = _mm_setzero_pd();
 813
 814             /**************************
 815              * CALCULATE INTERACTIONS *
 816              **************************/
 817
 818             if (gmx_mm_any_lt(rsq00,rcutoff2))
 819             {
 820
 821             r00              = _mm_mul_pd(rsq00,rinv00);
 822
 823             /* Compute parameters for interactions between i and j atoms */
 824             qq00             = _mm_mul_pd(iq0,jq0);
 825             gmx_mm_load_2pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,
 826                                          vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
 827
 828             /* Calculate table index by multiplying r with table scale and truncate to integer */
 829             rt               = _mm_mul_pd(r00,vftabscale);
 830             vfitab           = _mm_cvttpd_epi32(rt);
 831             vfeps            = _mm_sub_pd(rt,_mm_cvtepi32_pd(vfitab));
 832             vfitab           = _mm_slli_epi32(vfitab,3);
 833
 834             /* REACTION-FIELD ELECTROSTATICS */
 835             felec            = _mm_mul_pd(qq00,_mm_sub_pd(_mm_mul_pd(rinv00,rinvsq00),krf2));
 836
 837             /* CUBIC SPLINE TABLE DISPERSION */
 838             Y                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
 839             F                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) );
 840             GMX_MM_TRANSPOSE2_PD(Y,F);
 841             G                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
 842             H                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) +2);
 843             GMX_MM_TRANSPOSE2_PD(G,H);
 844             Heps             = _mm_mul_pd(vfeps,H);
 845             Fp               = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
 846             FF               = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
 847             fvdw6            = _mm_mul_pd(c6_00,FF);
 848
 849             /* CUBIC SPLINE TABLE REPULSION */
 850             vfitab           = _mm_add_epi32(vfitab,ifour);
 851             Y                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
 852             F                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) );
 853             GMX_MM_TRANSPOSE2_PD(Y,F);
 854             G                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
 855             H                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) +2);
 856             GMX_MM_TRANSPOSE2_PD(G,H);
 857             Heps             = _mm_mul_pd(vfeps,H);
 858             Fp               = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
 859             FF               = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
 860             fvdw12           = _mm_mul_pd(c12_00,FF);
 861             fvdw             = _mm_xor_pd(signbit,_mm_mul_pd(_mm_add_pd(fvdw6,fvdw12),_mm_mul_pd(vftabscale,rinv00)));
 862
 863             cutoff_mask      = _mm_cmplt_pd(rsq00,rcutoff2);
 864
 865             fscal            = _mm_add_pd(felec,fvdw);
 866
 867             fscal            = _mm_and_pd(fscal,cutoff_mask);
 868
 869             /* Calculate temporary vectorial force */
 870             tx               = _mm_mul_pd(fscal,dx00);
 871             ty               = _mm_mul_pd(fscal,dy00);
 872             tz               = _mm_mul_pd(fscal,dz00);
 873
 874             /* Update vectorial force */
 875             fix0             = _mm_add_pd(fix0,tx);
 876             fiy0             = _mm_add_pd(fiy0,ty);
 877             fiz0             = _mm_add_pd(fiz0,tz);
 878
 879             fjx0             = _mm_add_pd(fjx0,tx);
 880             fjy0             = _mm_add_pd(fjy0,ty);
 881             fjz0             = _mm_add_pd(fjz0,tz);
 882
 883             }
 884
 885             /**************************
 886              * CALCULATE INTERACTIONS *
 887              **************************/
 888
 889             if (gmx_mm_any_lt(rsq10,rcutoff2))
 890             {
 891
 892             /* Compute parameters for interactions between i and j atoms */
 893             qq10             = _mm_mul_pd(iq1,jq0);
 894
 895             /* REACTION-FIELD ELECTROSTATICS */
 896             felec            = _mm_mul_pd(qq10,_mm_sub_pd(_mm_mul_pd(rinv10,rinvsq10),krf2));
 897
 898             cutoff_mask      = _mm_cmplt_pd(rsq10,rcutoff2);
 899
 900             fscal            = felec;
 901
 902             fscal            = _mm_and_pd(fscal,cutoff_mask);
 903
 904             /* Calculate temporary vectorial force */
 905             tx               = _mm_mul_pd(fscal,dx10);
 906             ty               = _mm_mul_pd(fscal,dy10);
 907             tz               = _mm_mul_pd(fscal,dz10);
 908
 909             /* Update vectorial force */
 910             fix1             = _mm_add_pd(fix1,tx);
 911             fiy1             = _mm_add_pd(fiy1,ty);
 912             fiz1             = _mm_add_pd(fiz1,tz);
 913
 914             fjx0             = _mm_add_pd(fjx0,tx);
 915             fjy0             = _mm_add_pd(fjy0,ty);
 916             fjz0             = _mm_add_pd(fjz0,tz);
 917
 918             }
 919
 920             /**************************
 921              * CALCULATE INTERACTIONS *
 922              **************************/
 923
 924             if (gmx_mm_any_lt(rsq20,rcutoff2))
 925             {
 926
 927             /* Compute parameters for interactions between i and j atoms */
 928             qq20             = _mm_mul_pd(iq2,jq0);
 929
 930             /* REACTION-FIELD ELECTROSTATICS */
 931             felec            = _mm_mul_pd(qq20,_mm_sub_pd(_mm_mul_pd(rinv20,rinvsq20),krf2));
 932
 933             cutoff_mask      = _mm_cmplt_pd(rsq20,rcutoff2);
 934
 935             fscal            = felec;
 936
 937             fscal            = _mm_and_pd(fscal,cutoff_mask);
 938
 939             /* Calculate temporary vectorial force */
 940             tx               = _mm_mul_pd(fscal,dx20);
 941             ty               = _mm_mul_pd(fscal,dy20);
 942             tz               = _mm_mul_pd(fscal,dz20);
 943
 944             /* Update vectorial force */
 945             fix2             = _mm_add_pd(fix2,tx);
 946             fiy2             = _mm_add_pd(fiy2,ty);
 947             fiz2             = _mm_add_pd(fiz2,tz);
 948
 949             fjx0             = _mm_add_pd(fjx0,tx);
 950             fjy0             = _mm_add_pd(fjy0,ty);
 951             fjz0             = _mm_add_pd(fjz0,tz);
 952
 953             }
 954
 955             gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,fjx0,fjy0,fjz0);
 956
 957             /* Inner loop uses 120 flops */
 958         }
 959
 960         if(jidx<j_index_end)
 961         {
 962
 963             jnrA             = jjnr[jidx];
 964             j_coord_offsetA  = DIM*jnrA;
 965
 966             /* load j atom coordinates */
 967             gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
 968                                               &jx0,&jy0,&jz0);
 969
 970             /* Calculate displacement vector */
 971             dx00             = _mm_sub_pd(ix0,jx0);
 972             dy00             = _mm_sub_pd(iy0,jy0);
 973             dz00             = _mm_sub_pd(iz0,jz0);
 974             dx10             = _mm_sub_pd(ix1,jx0);
 975             dy10             = _mm_sub_pd(iy1,jy0);
 976             dz10             = _mm_sub_pd(iz1,jz0);
 977             dx20             = _mm_sub_pd(ix2,jx0);
 978             dy20             = _mm_sub_pd(iy2,jy0);
 979             dz20             = _mm_sub_pd(iz2,jz0);
 980
 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
 986             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 987             rinv10           = gmx_mm_invsqrt_pd(rsq10);
 988             rinv20           = gmx_mm_invsqrt_pd(rsq20);
 989
 990             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 991             rinvsq10         = _mm_mul_pd(rinv10,rinv10);
 992             rinvsq20         = _mm_mul_pd(rinv20,rinv20);
 993
 994             /* Load parameters for j particles */
 995             jq0              = _mm_load_sd(charge+jnrA+0);
 996             vdwjidx0A        = 2*vdwtype[jnrA+0];
 997
 998             fjx0             = _mm_setzero_pd();
 999             fjy0             = _mm_setzero_pd();
1000             fjz0             = _mm_setzero_pd();
1001
1002             /**************************
1003              * CALCULATE INTERACTIONS *
1004              **************************/
1005
1006             if (gmx_mm_any_lt(rsq00,rcutoff2))
1007             {
1008
1009             r00              = _mm_mul_pd(rsq00,rinv00);
1010
1011             /* Compute parameters for interactions between i and j atoms */
1012             qq00             = _mm_mul_pd(iq0,jq0);
1013             gmx_mm_load_1pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,&c6_00,&c12_00);
1014
1015             /* Calculate table index by multiplying r with table scale and truncate to integer */
1016             rt               = _mm_mul_pd(r00,vftabscale);
1017             vfitab           = _mm_cvttpd_epi32(rt);
1018             vfeps            = _mm_sub_pd(rt,_mm_cvtepi32_pd(vfitab));
1019             vfitab           = _mm_slli_epi32(vfitab,3);
1020
1021             /* REACTION-FIELD ELECTROSTATICS */
1022             felec            = _mm_mul_pd(qq00,_mm_sub_pd(_mm_mul_pd(rinv00,rinvsq00),krf2));
1023
1024             /* CUBIC SPLINE TABLE DISPERSION */
1025             Y                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
1026             F                = _mm_setzero_pd();
1027             GMX_MM_TRANSPOSE2_PD(Y,F);
1028             G                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
1029             H                = _mm_setzero_pd();
1030             GMX_MM_TRANSPOSE2_PD(G,H);
1031             Heps             = _mm_mul_pd(vfeps,H);
1032             Fp               = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
1033             FF               = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
1034             fvdw6            = _mm_mul_pd(c6_00,FF);
1035
1036             /* CUBIC SPLINE TABLE REPULSION */
1037             vfitab           = _mm_add_epi32(vfitab,ifour);
1038             Y                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
1039             F                = _mm_setzero_pd();
1040             GMX_MM_TRANSPOSE2_PD(Y,F);
1041             G                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
1042             H                = _mm_setzero_pd();
1043             GMX_MM_TRANSPOSE2_PD(G,H);
1044             Heps             = _mm_mul_pd(vfeps,H);
1045             Fp               = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
1046             FF               = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
1047             fvdw12           = _mm_mul_pd(c12_00,FF);
1048             fvdw             = _mm_xor_pd(signbit,_mm_mul_pd(_mm_add_pd(fvdw6,fvdw12),_mm_mul_pd(vftabscale,rinv00)));
1049
1050             cutoff_mask      = _mm_cmplt_pd(rsq00,rcutoff2);
1051
1052             fscal            = _mm_add_pd(felec,fvdw);
1053
1054             fscal            = _mm_and_pd(fscal,cutoff_mask);
1055
1056             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1057
1058             /* Calculate temporary vectorial force */
1059             tx               = _mm_mul_pd(fscal,dx00);
1060             ty               = _mm_mul_pd(fscal,dy00);
1061             tz               = _mm_mul_pd(fscal,dz00);
1062
1063             /* Update vectorial force */
1064             fix0             = _mm_add_pd(fix0,tx);
1065             fiy0             = _mm_add_pd(fiy0,ty);
1066             fiz0             = _mm_add_pd(fiz0,tz);
1067
1068             fjx0             = _mm_add_pd(fjx0,tx);
1069             fjy0             = _mm_add_pd(fjy0,ty);
1070             fjz0             = _mm_add_pd(fjz0,tz);
1071
1072             }
1073
1074             /**************************
1075              * CALCULATE INTERACTIONS *
1076              **************************/
1077
1078             if (gmx_mm_any_lt(rsq10,rcutoff2))
1079             {
1080
1081             /* Compute parameters for interactions between i and j atoms */
1082             qq10             = _mm_mul_pd(iq1,jq0);
1083
1084             /* REACTION-FIELD ELECTROSTATICS */
1085             felec            = _mm_mul_pd(qq10,_mm_sub_pd(_mm_mul_pd(rinv10,rinvsq10),krf2));
1086
1087             cutoff_mask      = _mm_cmplt_pd(rsq10,rcutoff2);
1088
1089             fscal            = felec;
1090
1091             fscal            = _mm_and_pd(fscal,cutoff_mask);
1092
1093             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1094
1095             /* Calculate temporary vectorial force */
1096             tx               = _mm_mul_pd(fscal,dx10);
1097             ty               = _mm_mul_pd(fscal,dy10);
1098             tz               = _mm_mul_pd(fscal,dz10);
1099
1100             /* Update vectorial force */
1101             fix1             = _mm_add_pd(fix1,tx);
1102             fiy1             = _mm_add_pd(fiy1,ty);
1103             fiz1             = _mm_add_pd(fiz1,tz);
1104
1105             fjx0             = _mm_add_pd(fjx0,tx);
1106             fjy0             = _mm_add_pd(fjy0,ty);
1107             fjz0             = _mm_add_pd(fjz0,tz);
1108
1109             }
1110
1111             /**************************
1112              * CALCULATE INTERACTIONS *
1113              **************************/
1114
1115             if (gmx_mm_any_lt(rsq20,rcutoff2))
1116             {
1117
1118             /* Compute parameters for interactions between i and j atoms */
1119             qq20             = _mm_mul_pd(iq2,jq0);
1120
1121             /* REACTION-FIELD ELECTROSTATICS */
1122             felec            = _mm_mul_pd(qq20,_mm_sub_pd(_mm_mul_pd(rinv20,rinvsq20),krf2));
1123
1124             cutoff_mask      = _mm_cmplt_pd(rsq20,rcutoff2);
1125
1126             fscal            = felec;
1127
1128             fscal            = _mm_and_pd(fscal,cutoff_mask);
1129
1130             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1131
1132             /* Calculate temporary vectorial force */
1133             tx               = _mm_mul_pd(fscal,dx20);
1134             ty               = _mm_mul_pd(fscal,dy20);
1135             tz               = _mm_mul_pd(fscal,dz20);
1136
1137             /* Update vectorial force */
1138             fix2             = _mm_add_pd(fix2,tx);
1139             fiy2             = _mm_add_pd(fiy2,ty);
1140             fiz2             = _mm_add_pd(fiz2,tz);
1141
1142             fjx0             = _mm_add_pd(fjx0,tx);
1143             fjy0             = _mm_add_pd(fjy0,ty);
1144             fjz0             = _mm_add_pd(fjz0,tz);
1145
1146             }
1147
1148             gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,fjx0,fjy0,fjz0);
1149
1150             /* Inner loop uses 120 flops */
1151         }
1152
1153         /* End of innermost loop */
1154
1155         gmx_mm_update_iforce_3atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
1156                                               f+i_coord_offset,fshift+i_shift_offset);
1157
1158         /* Increment number of inner iterations */
1159         inneriter                  += j_index_end - j_index_start;
1160
1161         /* Outer loop uses 18 flops */
1162     }
1163
1164     /* Increment number of outer iterations */
1165     outeriter        += nri;
1166
1167     /* Update outer/inner flops */
1168
1169     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3_F,outeriter*18 + inneriter*120);
1170 }