src/gromacs/gmxlib/nonbonded/nb_kernel_sse2_double/nb_kernel_ElecRF_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 "gromacs/legacyheaders/types/simple.h"
  44 #include "gromacs/math/vec.h"
  45 #include "gromacs/legacyheaders/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_ElecRF_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_ElecRF_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     /* Avoid stupid compiler warnings */
 138     jnrA = jnrB = 0;
 139     j_coord_offsetA = 0;
 140     j_coord_offsetB = 0;
 141
 142     outeriter        = 0;
 143     inneriter        = 0;
 144
 145     /* Start outer loop over neighborlists */
 146     for(iidx=0; iidx<nri; iidx++)
 147     {
 148         /* Load shift vector for this list */
 149         i_shift_offset   = DIM*shiftidx[iidx];
 150
 151         /* Load limits for loop over neighbors */
 152         j_index_start    = jindex[iidx];
 153         j_index_end      = jindex[iidx+1];
 154
 155         /* Get outer coordinate index */
 156         inr              = iinr[iidx];
 157         i_coord_offset   = DIM*inr;
 158
 159         /* Load i particle coords and add shift vector */
 160         gmx_mm_load_shift_and_3rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
 161                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 162
 163         fix0             = _mm_setzero_pd();
 164         fiy0             = _mm_setzero_pd();
 165         fiz0             = _mm_setzero_pd();
 166         fix1             = _mm_setzero_pd();
 167         fiy1             = _mm_setzero_pd();
 168         fiz1             = _mm_setzero_pd();
 169         fix2             = _mm_setzero_pd();
 170         fiy2             = _mm_setzero_pd();
 171         fiz2             = _mm_setzero_pd();
 172
 173         /* Reset potential sums */
 174         velecsum         = _mm_setzero_pd();
 175         vvdwsum          = _mm_setzero_pd();
 176
 177         /* Start inner kernel loop */
 178         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 179         {
 180
 181             /* Get j neighbor index, and coordinate index */
 182             jnrA             = jjnr[jidx];
 183             jnrB             = jjnr[jidx+1];
 184             j_coord_offsetA  = DIM*jnrA;
 185             j_coord_offsetB  = DIM*jnrB;
 186
 187             /* load j atom coordinates */
 188             gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 189                                               &jx0,&jy0,&jz0);
 190
 191             /* Calculate displacement vector */
 192             dx00             = _mm_sub_pd(ix0,jx0);
 193             dy00             = _mm_sub_pd(iy0,jy0);
 194             dz00             = _mm_sub_pd(iz0,jz0);
 195             dx10             = _mm_sub_pd(ix1,jx0);
 196             dy10             = _mm_sub_pd(iy1,jy0);
 197             dz10             = _mm_sub_pd(iz1,jz0);
 198             dx20             = _mm_sub_pd(ix2,jx0);
 199             dy20             = _mm_sub_pd(iy2,jy0);
 200             dz20             = _mm_sub_pd(iz2,jz0);
 201
 202             /* Calculate squared distance and things based on it */
 203             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 204             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
 205             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
 206
 207             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 208             rinv10           = gmx_mm_invsqrt_pd(rsq10);
 209             rinv20           = gmx_mm_invsqrt_pd(rsq20);
 210
 211             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 212             rinvsq10         = _mm_mul_pd(rinv10,rinv10);
 213             rinvsq20         = _mm_mul_pd(rinv20,rinv20);
 214
 215             /* Load parameters for j particles */
 216             jq0              = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
 217             vdwjidx0A        = 2*vdwtype[jnrA+0];
 218             vdwjidx0B        = 2*vdwtype[jnrB+0];
 219
 220             fjx0             = _mm_setzero_pd();
 221             fjy0             = _mm_setzero_pd();
 222             fjz0             = _mm_setzero_pd();
 223
 224             /**************************
 225              * CALCULATE INTERACTIONS *
 226              **************************/
 227
 228             r00              = _mm_mul_pd(rsq00,rinv00);
 229
 230             /* Compute parameters for interactions between i and j atoms */
 231             qq00             = _mm_mul_pd(iq0,jq0);
 232             gmx_mm_load_2pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,
 233                                          vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
 234
 235             /* Calculate table index by multiplying r with table scale and truncate to integer */
 236             rt               = _mm_mul_pd(r00,vftabscale);
 237             vfitab           = _mm_cvttpd_epi32(rt);
 238             vfeps            = _mm_sub_pd(rt,_mm_cvtepi32_pd(vfitab));
 239             vfitab           = _mm_slli_epi32(vfitab,3);
 240
 241             /* REACTION-FIELD ELECTROSTATICS */
 242             velec            = _mm_mul_pd(qq00,_mm_sub_pd(_mm_add_pd(rinv00,_mm_mul_pd(krf,rsq00)),crf));
 243             felec            = _mm_mul_pd(qq00,_mm_sub_pd(_mm_mul_pd(rinv00,rinvsq00),krf2));
 244
 245             /* CUBIC SPLINE TABLE DISPERSION */
 246             Y                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
 247             F                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) );
 248             GMX_MM_TRANSPOSE2_PD(Y,F);
 249             G                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
 250             H                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) +2);
 251             GMX_MM_TRANSPOSE2_PD(G,H);
 252             Heps             = _mm_mul_pd(vfeps,H);
 253             Fp               = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
 254             VV               = _mm_add_pd(Y,_mm_mul_pd(vfeps,Fp));
 255             vvdw6            = _mm_mul_pd(c6_00,VV);
 256             FF               = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
 257             fvdw6            = _mm_mul_pd(c6_00,FF);
 258
 259             /* CUBIC SPLINE TABLE REPULSION */
 260             vfitab           = _mm_add_epi32(vfitab,ifour);
 261             Y                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
 262             F                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) );
 263             GMX_MM_TRANSPOSE2_PD(Y,F);
 264             G                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
 265             H                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) +2);
 266             GMX_MM_TRANSPOSE2_PD(G,H);
 267             Heps             = _mm_mul_pd(vfeps,H);
 268             Fp               = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
 269             VV               = _mm_add_pd(Y,_mm_mul_pd(vfeps,Fp));
 270             vvdw12           = _mm_mul_pd(c12_00,VV);
 271             FF               = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
 272             fvdw12           = _mm_mul_pd(c12_00,FF);
 273             vvdw             = _mm_add_pd(vvdw12,vvdw6);
 274             fvdw             = _mm_xor_pd(signbit,_mm_mul_pd(_mm_add_pd(fvdw6,fvdw12),_mm_mul_pd(vftabscale,rinv00)));
 275
 276             /* Update potential sum for this i atom from the interaction with this j atom. */
 277             velecsum         = _mm_add_pd(velecsum,velec);
 278             vvdwsum          = _mm_add_pd(vvdwsum,vvdw);
 279
 280             fscal            = _mm_add_pd(felec,fvdw);
 281
 282             /* Calculate temporary vectorial force */
 283             tx               = _mm_mul_pd(fscal,dx00);
 284             ty               = _mm_mul_pd(fscal,dy00);
 285             tz               = _mm_mul_pd(fscal,dz00);
 286
 287             /* Update vectorial force */
 288             fix0             = _mm_add_pd(fix0,tx);
 289             fiy0             = _mm_add_pd(fiy0,ty);
 290             fiz0             = _mm_add_pd(fiz0,tz);
 291
 292             fjx0             = _mm_add_pd(fjx0,tx);
 293             fjy0             = _mm_add_pd(fjy0,ty);
 294             fjz0             = _mm_add_pd(fjz0,tz);
 295
 296             /**************************
 297              * CALCULATE INTERACTIONS *
 298              **************************/
 299
 300             /* Compute parameters for interactions between i and j atoms */
 301             qq10             = _mm_mul_pd(iq1,jq0);
 302
 303             /* REACTION-FIELD ELECTROSTATICS */
 304             velec            = _mm_mul_pd(qq10,_mm_sub_pd(_mm_add_pd(rinv10,_mm_mul_pd(krf,rsq10)),crf));
 305             felec            = _mm_mul_pd(qq10,_mm_sub_pd(_mm_mul_pd(rinv10,rinvsq10),krf2));
 306
 307             /* Update potential sum for this i atom from the interaction with this j atom. */
 308             velecsum         = _mm_add_pd(velecsum,velec);
 309
 310             fscal            = felec;
 311
 312             /* Calculate temporary vectorial force */
 313             tx               = _mm_mul_pd(fscal,dx10);
 314             ty               = _mm_mul_pd(fscal,dy10);
 315             tz               = _mm_mul_pd(fscal,dz10);
 316
 317             /* Update vectorial force */
 318             fix1             = _mm_add_pd(fix1,tx);
 319             fiy1             = _mm_add_pd(fiy1,ty);
 320             fiz1             = _mm_add_pd(fiz1,tz);
 321
 322             fjx0             = _mm_add_pd(fjx0,tx);
 323             fjy0             = _mm_add_pd(fjy0,ty);
 324             fjz0             = _mm_add_pd(fjz0,tz);
 325
 326             /**************************
 327              * CALCULATE INTERACTIONS *
 328              **************************/
 329
 330             /* Compute parameters for interactions between i and j atoms */
 331             qq20             = _mm_mul_pd(iq2,jq0);
 332
 333             /* REACTION-FIELD ELECTROSTATICS */
 334             velec            = _mm_mul_pd(qq20,_mm_sub_pd(_mm_add_pd(rinv20,_mm_mul_pd(krf,rsq20)),crf));
 335             felec            = _mm_mul_pd(qq20,_mm_sub_pd(_mm_mul_pd(rinv20,rinvsq20),krf2));
 336
 337             /* Update potential sum for this i atom from the interaction with this j atom. */
 338             velecsum         = _mm_add_pd(velecsum,velec);
 339
 340             fscal            = felec;
 341
 342             /* Calculate temporary vectorial force */
 343             tx               = _mm_mul_pd(fscal,dx20);
 344             ty               = _mm_mul_pd(fscal,dy20);
 345             tz               = _mm_mul_pd(fscal,dz20);
 346
 347             /* Update vectorial force */
 348             fix2             = _mm_add_pd(fix2,tx);
 349             fiy2             = _mm_add_pd(fiy2,ty);
 350             fiz2             = _mm_add_pd(fiz2,tz);
 351
 352             fjx0             = _mm_add_pd(fjx0,tx);
 353             fjy0             = _mm_add_pd(fjy0,ty);
 354             fjz0             = _mm_add_pd(fjz0,tz);
 355
 356             gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,fjx0,fjy0,fjz0);
 357
 358             /* Inner loop uses 134 flops */
 359         }
 360
 361         if(jidx<j_index_end)
 362         {
 363
 364             jnrA             = jjnr[jidx];
 365             j_coord_offsetA  = DIM*jnrA;
 366
 367             /* load j atom coordinates */
 368             gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
 369                                               &jx0,&jy0,&jz0);
 370
 371             /* Calculate displacement vector */
 372             dx00             = _mm_sub_pd(ix0,jx0);
 373             dy00             = _mm_sub_pd(iy0,jy0);
 374             dz00             = _mm_sub_pd(iz0,jz0);
 375             dx10             = _mm_sub_pd(ix1,jx0);
 376             dy10             = _mm_sub_pd(iy1,jy0);
 377             dz10             = _mm_sub_pd(iz1,jz0);
 378             dx20             = _mm_sub_pd(ix2,jx0);
 379             dy20             = _mm_sub_pd(iy2,jy0);
 380             dz20             = _mm_sub_pd(iz2,jz0);
 381
 382             /* Calculate squared distance and things based on it */
 383             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 384             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
 385             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
 386
 387             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 388             rinv10           = gmx_mm_invsqrt_pd(rsq10);
 389             rinv20           = gmx_mm_invsqrt_pd(rsq20);
 390
 391             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 392             rinvsq10         = _mm_mul_pd(rinv10,rinv10);
 393             rinvsq20         = _mm_mul_pd(rinv20,rinv20);
 394
 395             /* Load parameters for j particles */
 396             jq0              = _mm_load_sd(charge+jnrA+0);
 397             vdwjidx0A        = 2*vdwtype[jnrA+0];
 398
 399             fjx0             = _mm_setzero_pd();
 400             fjy0             = _mm_setzero_pd();
 401             fjz0             = _mm_setzero_pd();
 402
 403             /**************************
 404              * CALCULATE INTERACTIONS *
 405              **************************/
 406
 407             r00              = _mm_mul_pd(rsq00,rinv00);
 408
 409             /* Compute parameters for interactions between i and j atoms */
 410             qq00             = _mm_mul_pd(iq0,jq0);
 411             gmx_mm_load_1pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,&c6_00,&c12_00);
 412
 413             /* Calculate table index by multiplying r with table scale and truncate to integer */
 414             rt               = _mm_mul_pd(r00,vftabscale);
 415             vfitab           = _mm_cvttpd_epi32(rt);
 416             vfeps            = _mm_sub_pd(rt,_mm_cvtepi32_pd(vfitab));
 417             vfitab           = _mm_slli_epi32(vfitab,3);
 418
 419             /* REACTION-FIELD ELECTROSTATICS */
 420             velec            = _mm_mul_pd(qq00,_mm_sub_pd(_mm_add_pd(rinv00,_mm_mul_pd(krf,rsq00)),crf));
 421             felec            = _mm_mul_pd(qq00,_mm_sub_pd(_mm_mul_pd(rinv00,rinvsq00),krf2));
 422
 423             /* CUBIC SPLINE TABLE DISPERSION */
 424             Y                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
 425             F                = _mm_setzero_pd();
 426             GMX_MM_TRANSPOSE2_PD(Y,F);
 427             G                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
 428             H                = _mm_setzero_pd();
 429             GMX_MM_TRANSPOSE2_PD(G,H);
 430             Heps             = _mm_mul_pd(vfeps,H);
 431             Fp               = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
 432             VV               = _mm_add_pd(Y,_mm_mul_pd(vfeps,Fp));
 433             vvdw6            = _mm_mul_pd(c6_00,VV);
 434             FF               = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
 435             fvdw6            = _mm_mul_pd(c6_00,FF);
 436
 437             /* CUBIC SPLINE TABLE REPULSION */
 438             vfitab           = _mm_add_epi32(vfitab,ifour);
 439             Y                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
 440             F                = _mm_setzero_pd();
 441             GMX_MM_TRANSPOSE2_PD(Y,F);
 442             G                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
 443             H                = _mm_setzero_pd();
 444             GMX_MM_TRANSPOSE2_PD(G,H);
 445             Heps             = _mm_mul_pd(vfeps,H);
 446             Fp               = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
 447             VV               = _mm_add_pd(Y,_mm_mul_pd(vfeps,Fp));
 448             vvdw12           = _mm_mul_pd(c12_00,VV);
 449             FF               = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
 450             fvdw12           = _mm_mul_pd(c12_00,FF);
 451             vvdw             = _mm_add_pd(vvdw12,vvdw6);
 452             fvdw             = _mm_xor_pd(signbit,_mm_mul_pd(_mm_add_pd(fvdw6,fvdw12),_mm_mul_pd(vftabscale,rinv00)));
 453
 454             /* Update potential sum for this i atom from the interaction with this j atom. */
 455             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 456             velecsum         = _mm_add_pd(velecsum,velec);
 457             vvdw             = _mm_unpacklo_pd(vvdw,_mm_setzero_pd());
 458             vvdwsum          = _mm_add_pd(vvdwsum,vvdw);
 459
 460             fscal            = _mm_add_pd(felec,fvdw);
 461
 462             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 463
 464             /* Calculate temporary vectorial force */
 465             tx               = _mm_mul_pd(fscal,dx00);
 466             ty               = _mm_mul_pd(fscal,dy00);
 467             tz               = _mm_mul_pd(fscal,dz00);
 468
 469             /* Update vectorial force */
 470             fix0             = _mm_add_pd(fix0,tx);
 471             fiy0             = _mm_add_pd(fiy0,ty);
 472             fiz0             = _mm_add_pd(fiz0,tz);
 473
 474             fjx0             = _mm_add_pd(fjx0,tx);
 475             fjy0             = _mm_add_pd(fjy0,ty);
 476             fjz0             = _mm_add_pd(fjz0,tz);
 477
 478             /**************************
 479              * CALCULATE INTERACTIONS *
 480              **************************/
 481
 482             /* Compute parameters for interactions between i and j atoms */
 483             qq10             = _mm_mul_pd(iq1,jq0);
 484
 485             /* REACTION-FIELD ELECTROSTATICS */
 486             velec            = _mm_mul_pd(qq10,_mm_sub_pd(_mm_add_pd(rinv10,_mm_mul_pd(krf,rsq10)),crf));
 487             felec            = _mm_mul_pd(qq10,_mm_sub_pd(_mm_mul_pd(rinv10,rinvsq10),krf2));
 488
 489             /* Update potential sum for this i atom from the interaction with this j atom. */
 490             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 491             velecsum         = _mm_add_pd(velecsum,velec);
 492
 493             fscal            = felec;
 494
 495             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 496
 497             /* Calculate temporary vectorial force */
 498             tx               = _mm_mul_pd(fscal,dx10);
 499             ty               = _mm_mul_pd(fscal,dy10);
 500             tz               = _mm_mul_pd(fscal,dz10);
 501
 502             /* Update vectorial force */
 503             fix1             = _mm_add_pd(fix1,tx);
 504             fiy1             = _mm_add_pd(fiy1,ty);
 505             fiz1             = _mm_add_pd(fiz1,tz);
 506
 507             fjx0             = _mm_add_pd(fjx0,tx);
 508             fjy0             = _mm_add_pd(fjy0,ty);
 509             fjz0             = _mm_add_pd(fjz0,tz);
 510
 511             /**************************
 512              * CALCULATE INTERACTIONS *
 513              **************************/
 514
 515             /* Compute parameters for interactions between i and j atoms */
 516             qq20             = _mm_mul_pd(iq2,jq0);
 517
 518             /* REACTION-FIELD ELECTROSTATICS */
 519             velec            = _mm_mul_pd(qq20,_mm_sub_pd(_mm_add_pd(rinv20,_mm_mul_pd(krf,rsq20)),crf));
 520             felec            = _mm_mul_pd(qq20,_mm_sub_pd(_mm_mul_pd(rinv20,rinvsq20),krf2));
 521
 522             /* Update potential sum for this i atom from the interaction with this j atom. */
 523             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 524             velecsum         = _mm_add_pd(velecsum,velec);
 525
 526             fscal            = felec;
 527
 528             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 529
 530             /* Calculate temporary vectorial force */
 531             tx               = _mm_mul_pd(fscal,dx20);
 532             ty               = _mm_mul_pd(fscal,dy20);
 533             tz               = _mm_mul_pd(fscal,dz20);
 534
 535             /* Update vectorial force */
 536             fix2             = _mm_add_pd(fix2,tx);
 537             fiy2             = _mm_add_pd(fiy2,ty);
 538             fiz2             = _mm_add_pd(fiz2,tz);
 539
 540             fjx0             = _mm_add_pd(fjx0,tx);
 541             fjy0             = _mm_add_pd(fjy0,ty);
 542             fjz0             = _mm_add_pd(fjz0,tz);
 543
 544             gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,fjx0,fjy0,fjz0);
 545
 546             /* Inner loop uses 134 flops */
 547         }
 548
 549         /* End of innermost loop */
 550
 551         gmx_mm_update_iforce_3atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
 552                                               f+i_coord_offset,fshift+i_shift_offset);
 553
 554         ggid                        = gid[iidx];
 555         /* Update potential energies */
 556         gmx_mm_update_1pot_pd(velecsum,kernel_data->energygrp_elec+ggid);
 557         gmx_mm_update_1pot_pd(vvdwsum,kernel_data->energygrp_vdw+ggid);
 558
 559         /* Increment number of inner iterations */
 560         inneriter                  += j_index_end - j_index_start;
 561
 562         /* Outer loop uses 20 flops */
 563     }
 564
 565     /* Increment number of outer iterations */
 566     outeriter        += nri;
 567
 568     /* Update outer/inner flops */
 569
 570     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3_VF,outeriter*20 + inneriter*134);
 571 }
 572 /*
 573  * Gromacs nonbonded kernel:   nb_kernel_ElecRF_VdwCSTab_GeomW3P1_F_sse2_double
 574  * Electrostatics interaction: ReactionField
 575  * VdW interaction:            CubicSplineTable
 576  * Geometry:                   Water3-Particle
 577  * Calculate force/pot:        Force
 578  */
 579 void
 580 nb_kernel_ElecRF_VdwCSTab_GeomW3P1_F_sse2_double
 581                     (t_nblist                    * gmx_restrict       nlist,
 582                      rvec                        * gmx_restrict          xx,
 583                      rvec                        * gmx_restrict          ff,
 584                      t_forcerec                  * gmx_restrict          fr,
 585                      t_mdatoms                   * gmx_restrict     mdatoms,
 586                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 587                      t_nrnb                      * gmx_restrict        nrnb)
 588 {
 589     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 590      * just 0 for non-waters.
 591      * Suffixes A,B refer to j loop unrolling done with SSE double precision, e.g. for the two different
 592      * jnr indices corresponding to data put in the four positions in the SIMD register.
 593      */
 594     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 595     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 596     int              jnrA,jnrB;
 597     int              j_coord_offsetA,j_coord_offsetB;
 598     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 599     real             rcutoff_scalar;
 600     real             *shiftvec,*fshift,*x,*f;
 601     __m128d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 602     int              vdwioffset0;
 603     __m128d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 604     int              vdwioffset1;
 605     __m128d          ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
 606     int              vdwioffset2;
 607     __m128d          ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
 608     int              vdwjidx0A,vdwjidx0B;
 609     __m128d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 610     __m128d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 611     __m128d          dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
 612     __m128d          dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
 613     __m128d          velec,felec,velecsum,facel,crf,krf,krf2;
 614     real             *charge;
 615     int              nvdwtype;
 616     __m128d          rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 617     int              *vdwtype;
 618     real             *vdwparam;
 619     __m128d          one_sixth   = _mm_set1_pd(1.0/6.0);
 620     __m128d          one_twelfth = _mm_set1_pd(1.0/12.0);
 621     __m128i          vfitab;
 622     __m128i          ifour       = _mm_set1_epi32(4);
 623     __m128d          rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF;
 624     real             *vftab;
 625     __m128d          dummy_mask,cutoff_mask;
 626     __m128d          signbit   = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
 627     __m128d          one     = _mm_set1_pd(1.0);
 628     __m128d          two     = _mm_set1_pd(2.0);
 629     x                = xx[0];
 630     f                = ff[0];
 631
 632     nri              = nlist->nri;
 633     iinr             = nlist->iinr;
 634     jindex           = nlist->jindex;
 635     jjnr             = nlist->jjnr;
 636     shiftidx         = nlist->shift;
 637     gid              = nlist->gid;
 638     shiftvec         = fr->shift_vec[0];
 639     fshift           = fr->fshift[0];
 640     facel            = _mm_set1_pd(fr->epsfac);
 641     charge           = mdatoms->chargeA;
 642     krf              = _mm_set1_pd(fr->ic->k_rf);
 643     krf2             = _mm_set1_pd(fr->ic->k_rf*2.0);
 644     crf              = _mm_set1_pd(fr->ic->c_rf);
 645     nvdwtype         = fr->ntype;
 646     vdwparam         = fr->nbfp;
 647     vdwtype          = mdatoms->typeA;
 648
 649     vftab            = kernel_data->table_vdw->data;
 650     vftabscale       = _mm_set1_pd(kernel_data->table_vdw->scale);
 651
 652     /* Setup water-specific parameters */
 653     inr              = nlist->iinr[0];
 654     iq0              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+0]));
 655     iq1              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+1]));
 656     iq2              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+2]));
 657     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 658
 659     /* Avoid stupid compiler warnings */
 660     jnrA = jnrB = 0;
 661     j_coord_offsetA = 0;
 662     j_coord_offsetB = 0;
 663
 664     outeriter        = 0;
 665     inneriter        = 0;
 666
 667     /* Start outer loop over neighborlists */
 668     for(iidx=0; iidx<nri; iidx++)
 669     {
 670         /* Load shift vector for this list */
 671         i_shift_offset   = DIM*shiftidx[iidx];
 672
 673         /* Load limits for loop over neighbors */
 674         j_index_start    = jindex[iidx];
 675         j_index_end      = jindex[iidx+1];
 676
 677         /* Get outer coordinate index */
 678         inr              = iinr[iidx];
 679         i_coord_offset   = DIM*inr;
 680
 681         /* Load i particle coords and add shift vector */
 682         gmx_mm_load_shift_and_3rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
 683                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 684
 685         fix0             = _mm_setzero_pd();
 686         fiy0             = _mm_setzero_pd();
 687         fiz0             = _mm_setzero_pd();
 688         fix1             = _mm_setzero_pd();
 689         fiy1             = _mm_setzero_pd();
 690         fiz1             = _mm_setzero_pd();
 691         fix2             = _mm_setzero_pd();
 692         fiy2             = _mm_setzero_pd();
 693         fiz2             = _mm_setzero_pd();
 694
 695         /* Start inner kernel loop */
 696         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 697         {
 698
 699             /* Get j neighbor index, and coordinate index */
 700             jnrA             = jjnr[jidx];
 701             jnrB             = jjnr[jidx+1];
 702             j_coord_offsetA  = DIM*jnrA;
 703             j_coord_offsetB  = DIM*jnrB;
 704
 705             /* load j atom coordinates */
 706             gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 707                                               &jx0,&jy0,&jz0);
 708
 709             /* Calculate displacement vector */
 710             dx00             = _mm_sub_pd(ix0,jx0);
 711             dy00             = _mm_sub_pd(iy0,jy0);
 712             dz00             = _mm_sub_pd(iz0,jz0);
 713             dx10             = _mm_sub_pd(ix1,jx0);
 714             dy10             = _mm_sub_pd(iy1,jy0);
 715             dz10             = _mm_sub_pd(iz1,jz0);
 716             dx20             = _mm_sub_pd(ix2,jx0);
 717             dy20             = _mm_sub_pd(iy2,jy0);
 718             dz20             = _mm_sub_pd(iz2,jz0);
 719
 720             /* Calculate squared distance and things based on it */
 721             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 722             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
 723             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
 724
 725             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 726             rinv10           = gmx_mm_invsqrt_pd(rsq10);
 727             rinv20           = gmx_mm_invsqrt_pd(rsq20);
 728
 729             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 730             rinvsq10         = _mm_mul_pd(rinv10,rinv10);
 731             rinvsq20         = _mm_mul_pd(rinv20,rinv20);
 732
 733             /* Load parameters for j particles */
 734             jq0              = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
 735             vdwjidx0A        = 2*vdwtype[jnrA+0];
 736             vdwjidx0B        = 2*vdwtype[jnrB+0];
 737
 738             fjx0             = _mm_setzero_pd();
 739             fjy0             = _mm_setzero_pd();
 740             fjz0             = _mm_setzero_pd();
 741
 742             /**************************
 743              * CALCULATE INTERACTIONS *
 744              **************************/
 745
 746             r00              = _mm_mul_pd(rsq00,rinv00);
 747
 748             /* Compute parameters for interactions between i and j atoms */
 749             qq00             = _mm_mul_pd(iq0,jq0);
 750             gmx_mm_load_2pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,
 751                                          vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
 752
 753             /* Calculate table index by multiplying r with table scale and truncate to integer */
 754             rt               = _mm_mul_pd(r00,vftabscale);
 755             vfitab           = _mm_cvttpd_epi32(rt);
 756             vfeps            = _mm_sub_pd(rt,_mm_cvtepi32_pd(vfitab));
 757             vfitab           = _mm_slli_epi32(vfitab,3);
 758
 759             /* REACTION-FIELD ELECTROSTATICS */
 760             felec            = _mm_mul_pd(qq00,_mm_sub_pd(_mm_mul_pd(rinv00,rinvsq00),krf2));
 761
 762             /* CUBIC SPLINE TABLE DISPERSION */
 763             Y                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
 764             F                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) );
 765             GMX_MM_TRANSPOSE2_PD(Y,F);
 766             G                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
 767             H                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) +2);
 768             GMX_MM_TRANSPOSE2_PD(G,H);
 769             Heps             = _mm_mul_pd(vfeps,H);
 770             Fp               = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
 771             FF               = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
 772             fvdw6            = _mm_mul_pd(c6_00,FF);
 773
 774             /* CUBIC SPLINE TABLE REPULSION */
 775             vfitab           = _mm_add_epi32(vfitab,ifour);
 776             Y                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
 777             F                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) );
 778             GMX_MM_TRANSPOSE2_PD(Y,F);
 779             G                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
 780             H                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) +2);
 781             GMX_MM_TRANSPOSE2_PD(G,H);
 782             Heps             = _mm_mul_pd(vfeps,H);
 783             Fp               = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
 784             FF               = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
 785             fvdw12           = _mm_mul_pd(c12_00,FF);
 786             fvdw             = _mm_xor_pd(signbit,_mm_mul_pd(_mm_add_pd(fvdw6,fvdw12),_mm_mul_pd(vftabscale,rinv00)));
 787
 788             fscal            = _mm_add_pd(felec,fvdw);
 789
 790             /* Calculate temporary vectorial force */
 791             tx               = _mm_mul_pd(fscal,dx00);
 792             ty               = _mm_mul_pd(fscal,dy00);
 793             tz               = _mm_mul_pd(fscal,dz00);
 794
 795             /* Update vectorial force */
 796             fix0             = _mm_add_pd(fix0,tx);
 797             fiy0             = _mm_add_pd(fiy0,ty);
 798             fiz0             = _mm_add_pd(fiz0,tz);
 799
 800             fjx0             = _mm_add_pd(fjx0,tx);
 801             fjy0             = _mm_add_pd(fjy0,ty);
 802             fjz0             = _mm_add_pd(fjz0,tz);
 803
 804             /**************************
 805              * CALCULATE INTERACTIONS *
 806              **************************/
 807
 808             /* Compute parameters for interactions between i and j atoms */
 809             qq10             = _mm_mul_pd(iq1,jq0);
 810
 811             /* REACTION-FIELD ELECTROSTATICS */
 812             felec            = _mm_mul_pd(qq10,_mm_sub_pd(_mm_mul_pd(rinv10,rinvsq10),krf2));
 813
 814             fscal            = felec;
 815
 816             /* Calculate temporary vectorial force */
 817             tx               = _mm_mul_pd(fscal,dx10);
 818             ty               = _mm_mul_pd(fscal,dy10);
 819             tz               = _mm_mul_pd(fscal,dz10);
 820
 821             /* Update vectorial force */
 822             fix1             = _mm_add_pd(fix1,tx);
 823             fiy1             = _mm_add_pd(fiy1,ty);
 824             fiz1             = _mm_add_pd(fiz1,tz);
 825
 826             fjx0             = _mm_add_pd(fjx0,tx);
 827             fjy0             = _mm_add_pd(fjy0,ty);
 828             fjz0             = _mm_add_pd(fjz0,tz);
 829
 830             /**************************
 831              * CALCULATE INTERACTIONS *
 832              **************************/
 833
 834             /* Compute parameters for interactions between i and j atoms */
 835             qq20             = _mm_mul_pd(iq2,jq0);
 836
 837             /* REACTION-FIELD ELECTROSTATICS */
 838             felec            = _mm_mul_pd(qq20,_mm_sub_pd(_mm_mul_pd(rinv20,rinvsq20),krf2));
 839
 840             fscal            = felec;
 841
 842             /* Calculate temporary vectorial force */
 843             tx               = _mm_mul_pd(fscal,dx20);
 844             ty               = _mm_mul_pd(fscal,dy20);
 845             tz               = _mm_mul_pd(fscal,dz20);
 846
 847             /* Update vectorial force */
 848             fix2             = _mm_add_pd(fix2,tx);
 849             fiy2             = _mm_add_pd(fiy2,ty);
 850             fiz2             = _mm_add_pd(fiz2,tz);
 851
 852             fjx0             = _mm_add_pd(fjx0,tx);
 853             fjy0             = _mm_add_pd(fjy0,ty);
 854             fjz0             = _mm_add_pd(fjz0,tz);
 855
 856             gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,fjx0,fjy0,fjz0);
 857
 858             /* Inner loop uses 111 flops */
 859         }
 860
 861         if(jidx<j_index_end)
 862         {
 863
 864             jnrA             = jjnr[jidx];
 865             j_coord_offsetA  = DIM*jnrA;
 866
 867             /* load j atom coordinates */
 868             gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
 869                                               &jx0,&jy0,&jz0);
 870
 871             /* Calculate displacement vector */
 872             dx00             = _mm_sub_pd(ix0,jx0);
 873             dy00             = _mm_sub_pd(iy0,jy0);
 874             dz00             = _mm_sub_pd(iz0,jz0);
 875             dx10             = _mm_sub_pd(ix1,jx0);
 876             dy10             = _mm_sub_pd(iy1,jy0);
 877             dz10             = _mm_sub_pd(iz1,jz0);
 878             dx20             = _mm_sub_pd(ix2,jx0);
 879             dy20             = _mm_sub_pd(iy2,jy0);
 880             dz20             = _mm_sub_pd(iz2,jz0);
 881
 882             /* Calculate squared distance and things based on it */
 883             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 884             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
 885             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
 886
 887             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 888             rinv10           = gmx_mm_invsqrt_pd(rsq10);
 889             rinv20           = gmx_mm_invsqrt_pd(rsq20);
 890
 891             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 892             rinvsq10         = _mm_mul_pd(rinv10,rinv10);
 893             rinvsq20         = _mm_mul_pd(rinv20,rinv20);
 894
 895             /* Load parameters for j particles */
 896             jq0              = _mm_load_sd(charge+jnrA+0);
 897             vdwjidx0A        = 2*vdwtype[jnrA+0];
 898
 899             fjx0             = _mm_setzero_pd();
 900             fjy0             = _mm_setzero_pd();
 901             fjz0             = _mm_setzero_pd();
 902
 903             /**************************
 904              * CALCULATE INTERACTIONS *
 905              **************************/
 906
 907             r00              = _mm_mul_pd(rsq00,rinv00);
 908
 909             /* Compute parameters for interactions between i and j atoms */
 910             qq00             = _mm_mul_pd(iq0,jq0);
 911             gmx_mm_load_1pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,&c6_00,&c12_00);
 912
 913             /* Calculate table index by multiplying r with table scale and truncate to integer */
 914             rt               = _mm_mul_pd(r00,vftabscale);
 915             vfitab           = _mm_cvttpd_epi32(rt);
 916             vfeps            = _mm_sub_pd(rt,_mm_cvtepi32_pd(vfitab));
 917             vfitab           = _mm_slli_epi32(vfitab,3);
 918
 919             /* REACTION-FIELD ELECTROSTATICS */
 920             felec            = _mm_mul_pd(qq00,_mm_sub_pd(_mm_mul_pd(rinv00,rinvsq00),krf2));
 921
 922             /* CUBIC SPLINE TABLE DISPERSION */
 923             Y                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
 924             F                = _mm_setzero_pd();
 925             GMX_MM_TRANSPOSE2_PD(Y,F);
 926             G                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
 927             H                = _mm_setzero_pd();
 928             GMX_MM_TRANSPOSE2_PD(G,H);
 929             Heps             = _mm_mul_pd(vfeps,H);
 930             Fp               = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
 931             FF               = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
 932             fvdw6            = _mm_mul_pd(c6_00,FF);
 933
 934             /* CUBIC SPLINE TABLE REPULSION */
 935             vfitab           = _mm_add_epi32(vfitab,ifour);
 936             Y                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
 937             F                = _mm_setzero_pd();
 938             GMX_MM_TRANSPOSE2_PD(Y,F);
 939             G                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
 940             H                = _mm_setzero_pd();
 941             GMX_MM_TRANSPOSE2_PD(G,H);
 942             Heps             = _mm_mul_pd(vfeps,H);
 943             Fp               = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
 944             FF               = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
 945             fvdw12           = _mm_mul_pd(c12_00,FF);
 946             fvdw             = _mm_xor_pd(signbit,_mm_mul_pd(_mm_add_pd(fvdw6,fvdw12),_mm_mul_pd(vftabscale,rinv00)));
 947
 948             fscal            = _mm_add_pd(felec,fvdw);
 949
 950             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 951
 952             /* Calculate temporary vectorial force */
 953             tx               = _mm_mul_pd(fscal,dx00);
 954             ty               = _mm_mul_pd(fscal,dy00);
 955             tz               = _mm_mul_pd(fscal,dz00);
 956
 957             /* Update vectorial force */
 958             fix0             = _mm_add_pd(fix0,tx);
 959             fiy0             = _mm_add_pd(fiy0,ty);
 960             fiz0             = _mm_add_pd(fiz0,tz);
 961
 962             fjx0             = _mm_add_pd(fjx0,tx);
 963             fjy0             = _mm_add_pd(fjy0,ty);
 964             fjz0             = _mm_add_pd(fjz0,tz);
 965
 966             /**************************
 967              * CALCULATE INTERACTIONS *
 968              **************************/
 969
 970             /* Compute parameters for interactions between i and j atoms */
 971             qq10             = _mm_mul_pd(iq1,jq0);
 972
 973             /* REACTION-FIELD ELECTROSTATICS */
 974             felec            = _mm_mul_pd(qq10,_mm_sub_pd(_mm_mul_pd(rinv10,rinvsq10),krf2));
 975
 976             fscal            = felec;
 977
 978             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 979
 980             /* Calculate temporary vectorial force */
 981             tx               = _mm_mul_pd(fscal,dx10);
 982             ty               = _mm_mul_pd(fscal,dy10);
 983             tz               = _mm_mul_pd(fscal,dz10);
 984
 985             /* Update vectorial force */
 986             fix1             = _mm_add_pd(fix1,tx);
 987             fiy1             = _mm_add_pd(fiy1,ty);
 988             fiz1             = _mm_add_pd(fiz1,tz);
 989
 990             fjx0             = _mm_add_pd(fjx0,tx);
 991             fjy0             = _mm_add_pd(fjy0,ty);
 992             fjz0             = _mm_add_pd(fjz0,tz);
 993
 994             /**************************
 995              * CALCULATE INTERACTIONS *
 996              **************************/
 997
 998             /* Compute parameters for interactions between i and j atoms */
 999             qq20             = _mm_mul_pd(iq2,jq0);
1000
1001             /* REACTION-FIELD ELECTROSTATICS */
1002             felec            = _mm_mul_pd(qq20,_mm_sub_pd(_mm_mul_pd(rinv20,rinvsq20),krf2));
1003
1004             fscal            = felec;
1005
1006             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1007
1008             /* Calculate temporary vectorial force */
1009             tx               = _mm_mul_pd(fscal,dx20);
1010             ty               = _mm_mul_pd(fscal,dy20);
1011             tz               = _mm_mul_pd(fscal,dz20);
1012
1013             /* Update vectorial force */
1014             fix2             = _mm_add_pd(fix2,tx);
1015             fiy2             = _mm_add_pd(fiy2,ty);
1016             fiz2             = _mm_add_pd(fiz2,tz);
1017
1018             fjx0             = _mm_add_pd(fjx0,tx);
1019             fjy0             = _mm_add_pd(fjy0,ty);
1020             fjz0             = _mm_add_pd(fjz0,tz);
1021
1022             gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,fjx0,fjy0,fjz0);
1023
1024             /* Inner loop uses 111 flops */
1025         }
1026
1027         /* End of innermost loop */
1028
1029         gmx_mm_update_iforce_3atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
1030                                               f+i_coord_offset,fshift+i_shift_offset);
1031
1032         /* Increment number of inner iterations */
1033         inneriter                  += j_index_end - j_index_start;
1034
1035         /* Outer loop uses 18 flops */
1036     }
1037
1038     /* Increment number of outer iterations */
1039     outeriter        += nri;
1040
1041     /* Update outer/inner flops */
1042
1043     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3_F,outeriter*18 + inneriter*111);
1044 }