src/gromacs/gmxlib/nonbonded/nb_kernel_sse4_1_double/nb_kernel_ElecCSTab_VdwLJ_GeomW4P1_sse4_1_double.c

   1 /*
   2  * This file is part of the GROMACS molecular simulation package.
   3  *
   4  * Copyright (c) 2012,2013,2014,2015,2017, by the GROMACS development team, led by
   5  * Mark Abraham, David van der Spoel, Berk Hess, and Erik Lindahl,
   6  * and including many others, as listed in the AUTHORS file in the
   7  * top-level source directory and at http://www.gromacs.org.
   8  *
   9  * GROMACS is free software; you can redistribute it and/or
  10  * modify it under the terms of the GNU Lesser General Public License
  11  * as published by the Free Software Foundation; either version 2.1
  12  * of the License, or (at your option) any later version.
  13  *
  14  * GROMACS is distributed in the hope that it will be useful,
  15  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  16  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  17  * Lesser General Public License for more details.
  18  *
  19  * You should have received a copy of the GNU Lesser General Public
  20  * License along with GROMACS; if not, see
  21  * http://www.gnu.org/licenses, or write to the Free Software Foundation,
  22  * Inc., 51 Franklin Street, Fifth Floor, Boston, MA  02110-1301  USA.
  23  *
  24  * If you want to redistribute modifications to GROMACS, please
  25  * consider that scientific software is very special. Version
  26  * control is crucial - bugs must be traceable. We will be happy to
  27  * consider code for inclusion in the official distribution, but
  28  * derived work must not be called official GROMACS. Details are found
  29  * in the README & COPYING files - if they are missing, get the
  30  * official version at http://www.gromacs.org.
  31  *
  32  * To help us fund GROMACS development, we humbly ask that you cite
  33  * the research papers on the package. Check out http://www.gromacs.org.
  34  */
  35 /*
  36  * Note: this file was generated by the GROMACS sse4_1_double kernel generator.
  37  */
  38 #include "gmxpre.h"
  39
  40 #include "config.h"
  41
  42 #include <math.h>
  43
  44 #include "../nb_kernel.h"
  45 #include "gromacs/gmxlib/nrnb.h"
  46
  47 #include "kernelutil_x86_sse4_1_double.h"
  48
  49 /*
  50  * Gromacs nonbonded kernel:   nb_kernel_ElecCSTab_VdwLJ_GeomW4P1_VF_sse4_1_double
  51  * Electrostatics interaction: CubicSplineTable
  52  * VdW interaction:            LennardJones
  53  * Geometry:                   Water4-Particle
  54  * Calculate force/pot:        PotentialAndForce
  55  */
  56 void
  57 nb_kernel_ElecCSTab_VdwLJ_GeomW4P1_VF_sse4_1_double
  58                     (t_nblist                    * gmx_restrict       nlist,
  59                      rvec                        * gmx_restrict          xx,
  60                      rvec                        * gmx_restrict          ff,
  61                      struct t_forcerec           * gmx_restrict          fr,
  62                      t_mdatoms                   * gmx_restrict     mdatoms,
  63                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
  64                      t_nrnb                      * gmx_restrict        nrnb)
  65 {
  66     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
  67      * just 0 for non-waters.
  68      * Suffixes A,B refer to j loop unrolling done with SSE double precision, e.g. for the two different
  69      * jnr indices corresponding to data put in the four positions in the SIMD register.
  70      */
  71     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
  72     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
  73     int              jnrA,jnrB;
  74     int              j_coord_offsetA,j_coord_offsetB;
  75     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
  76     real             rcutoff_scalar;
  77     real             *shiftvec,*fshift,*x,*f;
  78     __m128d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
  79     int              vdwioffset0;
  80     __m128d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
  81     int              vdwioffset1;
  82     __m128d          ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
  83     int              vdwioffset2;
  84     __m128d          ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
  85     int              vdwioffset3;
  86     __m128d          ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
  87     int              vdwjidx0A,vdwjidx0B;
  88     __m128d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  89     __m128d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  90     __m128d          dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
  91     __m128d          dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
  92     __m128d          dx30,dy30,dz30,rsq30,rinv30,rinvsq30,r30,qq30,c6_30,c12_30;
  93     __m128d          velec,felec,velecsum,facel,crf,krf,krf2;
  94     real             *charge;
  95     int              nvdwtype;
  96     __m128d          rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
  97     int              *vdwtype;
  98     real             *vdwparam;
  99     __m128d          one_sixth   = _mm_set1_pd(1.0/6.0);
 100     __m128d          one_twelfth = _mm_set1_pd(1.0/12.0);
 101     __m128i          vfitab;
 102     __m128i          ifour       = _mm_set1_epi32(4);
 103     __m128d          rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF;
 104     real             *vftab;
 105     __m128d          dummy_mask,cutoff_mask;
 106     __m128d          signbit   = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
 107     __m128d          one     = _mm_set1_pd(1.0);
 108     __m128d          two     = _mm_set1_pd(2.0);
 109     x                = xx[0];
 110     f                = ff[0];
 111
 112     nri              = nlist->nri;
 113     iinr             = nlist->iinr;
 114     jindex           = nlist->jindex;
 115     jjnr             = nlist->jjnr;
 116     shiftidx         = nlist->shift;
 117     gid              = nlist->gid;
 118     shiftvec         = fr->shift_vec[0];
 119     fshift           = fr->fshift[0];
 120     facel            = _mm_set1_pd(fr->ic->epsfac);
 121     charge           = mdatoms->chargeA;
 122     nvdwtype         = fr->ntype;
 123     vdwparam         = fr->nbfp;
 124     vdwtype          = mdatoms->typeA;
 125
 126     vftab            = kernel_data->table_elec->data;
 127     vftabscale       = _mm_set1_pd(kernel_data->table_elec->scale);
 128
 129     /* Setup water-specific parameters */
 130     inr              = nlist->iinr[0];
 131     iq1              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+1]));
 132     iq2              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+2]));
 133     iq3              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+3]));
 134     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 135
 136     /* Avoid stupid compiler warnings */
 137     jnrA = jnrB = 0;
 138     j_coord_offsetA = 0;
 139     j_coord_offsetB = 0;
 140
 141     outeriter        = 0;
 142     inneriter        = 0;
 143
 144     /* Start outer loop over neighborlists */
 145     for(iidx=0; iidx<nri; iidx++)
 146     {
 147         /* Load shift vector for this list */
 148         i_shift_offset   = DIM*shiftidx[iidx];
 149
 150         /* Load limits for loop over neighbors */
 151         j_index_start    = jindex[iidx];
 152         j_index_end      = jindex[iidx+1];
 153
 154         /* Get outer coordinate index */
 155         inr              = iinr[iidx];
 156         i_coord_offset   = DIM*inr;
 157
 158         /* Load i particle coords and add shift vector */
 159         gmx_mm_load_shift_and_4rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
 160                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
 161
 162         fix0             = _mm_setzero_pd();
 163         fiy0             = _mm_setzero_pd();
 164         fiz0             = _mm_setzero_pd();
 165         fix1             = _mm_setzero_pd();
 166         fiy1             = _mm_setzero_pd();
 167         fiz1             = _mm_setzero_pd();
 168         fix2             = _mm_setzero_pd();
 169         fiy2             = _mm_setzero_pd();
 170         fiz2             = _mm_setzero_pd();
 171         fix3             = _mm_setzero_pd();
 172         fiy3             = _mm_setzero_pd();
 173         fiz3             = _mm_setzero_pd();
 174
 175         /* Reset potential sums */
 176         velecsum         = _mm_setzero_pd();
 177         vvdwsum          = _mm_setzero_pd();
 178
 179         /* Start inner kernel loop */
 180         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 181         {
 182
 183             /* Get j neighbor index, and coordinate index */
 184             jnrA             = jjnr[jidx];
 185             jnrB             = jjnr[jidx+1];
 186             j_coord_offsetA  = DIM*jnrA;
 187             j_coord_offsetB  = DIM*jnrB;
 188
 189             /* load j atom coordinates */
 190             gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 191                                               &jx0,&jy0,&jz0);
 192
 193             /* Calculate displacement vector */
 194             dx00             = _mm_sub_pd(ix0,jx0);
 195             dy00             = _mm_sub_pd(iy0,jy0);
 196             dz00             = _mm_sub_pd(iz0,jz0);
 197             dx10             = _mm_sub_pd(ix1,jx0);
 198             dy10             = _mm_sub_pd(iy1,jy0);
 199             dz10             = _mm_sub_pd(iz1,jz0);
 200             dx20             = _mm_sub_pd(ix2,jx0);
 201             dy20             = _mm_sub_pd(iy2,jy0);
 202             dz20             = _mm_sub_pd(iz2,jz0);
 203             dx30             = _mm_sub_pd(ix3,jx0);
 204             dy30             = _mm_sub_pd(iy3,jy0);
 205             dz30             = _mm_sub_pd(iz3,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             rsq30            = gmx_mm_calc_rsq_pd(dx30,dy30,dz30);
 212
 213             rinv10           = sse41_invsqrt_d(rsq10);
 214             rinv20           = sse41_invsqrt_d(rsq20);
 215             rinv30           = sse41_invsqrt_d(rsq30);
 216
 217             rinvsq00         = sse41_inv_d(rsq00);
 218
 219             /* Load parameters for j particles */
 220             jq0              = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
 221             vdwjidx0A        = 2*vdwtype[jnrA+0];
 222             vdwjidx0B        = 2*vdwtype[jnrB+0];
 223
 224             fjx0             = _mm_setzero_pd();
 225             fjy0             = _mm_setzero_pd();
 226             fjz0             = _mm_setzero_pd();
 227
 228             /**************************
 229              * CALCULATE INTERACTIONS *
 230              **************************/
 231
 232             /* Compute parameters for interactions between i and j atoms */
 233             gmx_mm_load_2pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,
 234                                          vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
 235
 236             /* LENNARD-JONES DISPERSION/REPULSION */
 237
 238             rinvsix          = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 239             vvdw6            = _mm_mul_pd(c6_00,rinvsix);
 240             vvdw12           = _mm_mul_pd(c12_00,_mm_mul_pd(rinvsix,rinvsix));
 241             vvdw             = _mm_sub_pd( _mm_mul_pd(vvdw12,one_twelfth) , _mm_mul_pd(vvdw6,one_sixth) );
 242             fvdw             = _mm_mul_pd(_mm_sub_pd(vvdw12,vvdw6),rinvsq00);
 243
 244             /* Update potential sum for this i atom from the interaction with this j atom. */
 245             vvdwsum          = _mm_add_pd(vvdwsum,vvdw);
 246
 247             fscal            = fvdw;
 248
 249             /* Calculate temporary vectorial force */
 250             tx               = _mm_mul_pd(fscal,dx00);
 251             ty               = _mm_mul_pd(fscal,dy00);
 252             tz               = _mm_mul_pd(fscal,dz00);
 253
 254             /* Update vectorial force */
 255             fix0             = _mm_add_pd(fix0,tx);
 256             fiy0             = _mm_add_pd(fiy0,ty);
 257             fiz0             = _mm_add_pd(fiz0,tz);
 258
 259             fjx0             = _mm_add_pd(fjx0,tx);
 260             fjy0             = _mm_add_pd(fjy0,ty);
 261             fjz0             = _mm_add_pd(fjz0,tz);
 262
 263             /**************************
 264              * CALCULATE INTERACTIONS *
 265              **************************/
 266
 267             r10              = _mm_mul_pd(rsq10,rinv10);
 268
 269             /* Compute parameters for interactions between i and j atoms */
 270             qq10             = _mm_mul_pd(iq1,jq0);
 271
 272             /* Calculate table index by multiplying r with table scale and truncate to integer */
 273             rt               = _mm_mul_pd(r10,vftabscale);
 274             vfitab           = _mm_cvttpd_epi32(rt);
 275             vfeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
 276             vfitab           = _mm_slli_epi32(vfitab,2);
 277
 278             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 279             Y                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
 280             F                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) );
 281             GMX_MM_TRANSPOSE2_PD(Y,F);
 282             G                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
 283             H                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) +2);
 284             GMX_MM_TRANSPOSE2_PD(G,H);
 285             Heps             = _mm_mul_pd(vfeps,H);
 286             Fp               = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
 287             VV               = _mm_add_pd(Y,_mm_mul_pd(vfeps,Fp));
 288             velec            = _mm_mul_pd(qq10,VV);
 289             FF               = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
 290             felec            = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq10,FF),_mm_mul_pd(vftabscale,rinv10)));
 291
 292             /* Update potential sum for this i atom from the interaction with this j atom. */
 293             velecsum         = _mm_add_pd(velecsum,velec);
 294
 295             fscal            = felec;
 296
 297             /* Calculate temporary vectorial force */
 298             tx               = _mm_mul_pd(fscal,dx10);
 299             ty               = _mm_mul_pd(fscal,dy10);
 300             tz               = _mm_mul_pd(fscal,dz10);
 301
 302             /* Update vectorial force */
 303             fix1             = _mm_add_pd(fix1,tx);
 304             fiy1             = _mm_add_pd(fiy1,ty);
 305             fiz1             = _mm_add_pd(fiz1,tz);
 306
 307             fjx0             = _mm_add_pd(fjx0,tx);
 308             fjy0             = _mm_add_pd(fjy0,ty);
 309             fjz0             = _mm_add_pd(fjz0,tz);
 310
 311             /**************************
 312              * CALCULATE INTERACTIONS *
 313              **************************/
 314
 315             r20              = _mm_mul_pd(rsq20,rinv20);
 316
 317             /* Compute parameters for interactions between i and j atoms */
 318             qq20             = _mm_mul_pd(iq2,jq0);
 319
 320             /* Calculate table index by multiplying r with table scale and truncate to integer */
 321             rt               = _mm_mul_pd(r20,vftabscale);
 322             vfitab           = _mm_cvttpd_epi32(rt);
 323             vfeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
 324             vfitab           = _mm_slli_epi32(vfitab,2);
 325
 326             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 327             Y                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
 328             F                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) );
 329             GMX_MM_TRANSPOSE2_PD(Y,F);
 330             G                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
 331             H                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) +2);
 332             GMX_MM_TRANSPOSE2_PD(G,H);
 333             Heps             = _mm_mul_pd(vfeps,H);
 334             Fp               = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
 335             VV               = _mm_add_pd(Y,_mm_mul_pd(vfeps,Fp));
 336             velec            = _mm_mul_pd(qq20,VV);
 337             FF               = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
 338             felec            = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq20,FF),_mm_mul_pd(vftabscale,rinv20)));
 339
 340             /* Update potential sum for this i atom from the interaction with this j atom. */
 341             velecsum         = _mm_add_pd(velecsum,velec);
 342
 343             fscal            = felec;
 344
 345             /* Calculate temporary vectorial force */
 346             tx               = _mm_mul_pd(fscal,dx20);
 347             ty               = _mm_mul_pd(fscal,dy20);
 348             tz               = _mm_mul_pd(fscal,dz20);
 349
 350             /* Update vectorial force */
 351             fix2             = _mm_add_pd(fix2,tx);
 352             fiy2             = _mm_add_pd(fiy2,ty);
 353             fiz2             = _mm_add_pd(fiz2,tz);
 354
 355             fjx0             = _mm_add_pd(fjx0,tx);
 356             fjy0             = _mm_add_pd(fjy0,ty);
 357             fjz0             = _mm_add_pd(fjz0,tz);
 358
 359             /**************************
 360              * CALCULATE INTERACTIONS *
 361              **************************/
 362
 363             r30              = _mm_mul_pd(rsq30,rinv30);
 364
 365             /* Compute parameters for interactions between i and j atoms */
 366             qq30             = _mm_mul_pd(iq3,jq0);
 367
 368             /* Calculate table index by multiplying r with table scale and truncate to integer */
 369             rt               = _mm_mul_pd(r30,vftabscale);
 370             vfitab           = _mm_cvttpd_epi32(rt);
 371             vfeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
 372             vfitab           = _mm_slli_epi32(vfitab,2);
 373
 374             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 375             Y                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
 376             F                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) );
 377             GMX_MM_TRANSPOSE2_PD(Y,F);
 378             G                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
 379             H                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) +2);
 380             GMX_MM_TRANSPOSE2_PD(G,H);
 381             Heps             = _mm_mul_pd(vfeps,H);
 382             Fp               = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
 383             VV               = _mm_add_pd(Y,_mm_mul_pd(vfeps,Fp));
 384             velec            = _mm_mul_pd(qq30,VV);
 385             FF               = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
 386             felec            = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq30,FF),_mm_mul_pd(vftabscale,rinv30)));
 387
 388             /* Update potential sum for this i atom from the interaction with this j atom. */
 389             velecsum         = _mm_add_pd(velecsum,velec);
 390
 391             fscal            = felec;
 392
 393             /* Calculate temporary vectorial force */
 394             tx               = _mm_mul_pd(fscal,dx30);
 395             ty               = _mm_mul_pd(fscal,dy30);
 396             tz               = _mm_mul_pd(fscal,dz30);
 397
 398             /* Update vectorial force */
 399             fix3             = _mm_add_pd(fix3,tx);
 400             fiy3             = _mm_add_pd(fiy3,ty);
 401             fiz3             = _mm_add_pd(fiz3,tz);
 402
 403             fjx0             = _mm_add_pd(fjx0,tx);
 404             fjy0             = _mm_add_pd(fjy0,ty);
 405             fjz0             = _mm_add_pd(fjz0,tz);
 406
 407             gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,fjx0,fjy0,fjz0);
 408
 409             /* Inner loop uses 164 flops */
 410         }
 411
 412         if(jidx<j_index_end)
 413         {
 414
 415             jnrA             = jjnr[jidx];
 416             j_coord_offsetA  = DIM*jnrA;
 417
 418             /* load j atom coordinates */
 419             gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
 420                                               &jx0,&jy0,&jz0);
 421
 422             /* Calculate displacement vector */
 423             dx00             = _mm_sub_pd(ix0,jx0);
 424             dy00             = _mm_sub_pd(iy0,jy0);
 425             dz00             = _mm_sub_pd(iz0,jz0);
 426             dx10             = _mm_sub_pd(ix1,jx0);
 427             dy10             = _mm_sub_pd(iy1,jy0);
 428             dz10             = _mm_sub_pd(iz1,jz0);
 429             dx20             = _mm_sub_pd(ix2,jx0);
 430             dy20             = _mm_sub_pd(iy2,jy0);
 431             dz20             = _mm_sub_pd(iz2,jz0);
 432             dx30             = _mm_sub_pd(ix3,jx0);
 433             dy30             = _mm_sub_pd(iy3,jy0);
 434             dz30             = _mm_sub_pd(iz3,jz0);
 435
 436             /* Calculate squared distance and things based on it */
 437             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 438             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
 439             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
 440             rsq30            = gmx_mm_calc_rsq_pd(dx30,dy30,dz30);
 441
 442             rinv10           = sse41_invsqrt_d(rsq10);
 443             rinv20           = sse41_invsqrt_d(rsq20);
 444             rinv30           = sse41_invsqrt_d(rsq30);
 445
 446             rinvsq00         = sse41_inv_d(rsq00);
 447
 448             /* Load parameters for j particles */
 449             jq0              = _mm_load_sd(charge+jnrA+0);
 450             vdwjidx0A        = 2*vdwtype[jnrA+0];
 451
 452             fjx0             = _mm_setzero_pd();
 453             fjy0             = _mm_setzero_pd();
 454             fjz0             = _mm_setzero_pd();
 455
 456             /**************************
 457              * CALCULATE INTERACTIONS *
 458              **************************/
 459
 460             /* Compute parameters for interactions between i and j atoms */
 461             gmx_mm_load_1pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,&c6_00,&c12_00);
 462
 463             /* LENNARD-JONES DISPERSION/REPULSION */
 464
 465             rinvsix          = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 466             vvdw6            = _mm_mul_pd(c6_00,rinvsix);
 467             vvdw12           = _mm_mul_pd(c12_00,_mm_mul_pd(rinvsix,rinvsix));
 468             vvdw             = _mm_sub_pd( _mm_mul_pd(vvdw12,one_twelfth) , _mm_mul_pd(vvdw6,one_sixth) );
 469             fvdw             = _mm_mul_pd(_mm_sub_pd(vvdw12,vvdw6),rinvsq00);
 470
 471             /* Update potential sum for this i atom from the interaction with this j atom. */
 472             vvdw             = _mm_unpacklo_pd(vvdw,_mm_setzero_pd());
 473             vvdwsum          = _mm_add_pd(vvdwsum,vvdw);
 474
 475             fscal            = fvdw;
 476
 477             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 478
 479             /* Calculate temporary vectorial force */
 480             tx               = _mm_mul_pd(fscal,dx00);
 481             ty               = _mm_mul_pd(fscal,dy00);
 482             tz               = _mm_mul_pd(fscal,dz00);
 483
 484             /* Update vectorial force */
 485             fix0             = _mm_add_pd(fix0,tx);
 486             fiy0             = _mm_add_pd(fiy0,ty);
 487             fiz0             = _mm_add_pd(fiz0,tz);
 488
 489             fjx0             = _mm_add_pd(fjx0,tx);
 490             fjy0             = _mm_add_pd(fjy0,ty);
 491             fjz0             = _mm_add_pd(fjz0,tz);
 492
 493             /**************************
 494              * CALCULATE INTERACTIONS *
 495              **************************/
 496
 497             r10              = _mm_mul_pd(rsq10,rinv10);
 498
 499             /* Compute parameters for interactions between i and j atoms */
 500             qq10             = _mm_mul_pd(iq1,jq0);
 501
 502             /* Calculate table index by multiplying r with table scale and truncate to integer */
 503             rt               = _mm_mul_pd(r10,vftabscale);
 504             vfitab           = _mm_cvttpd_epi32(rt);
 505             vfeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
 506             vfitab           = _mm_slli_epi32(vfitab,2);
 507
 508             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 509             Y                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
 510             F                = _mm_setzero_pd();
 511             GMX_MM_TRANSPOSE2_PD(Y,F);
 512             G                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
 513             H                = _mm_setzero_pd();
 514             GMX_MM_TRANSPOSE2_PD(G,H);
 515             Heps             = _mm_mul_pd(vfeps,H);
 516             Fp               = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
 517             VV               = _mm_add_pd(Y,_mm_mul_pd(vfeps,Fp));
 518             velec            = _mm_mul_pd(qq10,VV);
 519             FF               = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
 520             felec            = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq10,FF),_mm_mul_pd(vftabscale,rinv10)));
 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,dx10);
 532             ty               = _mm_mul_pd(fscal,dy10);
 533             tz               = _mm_mul_pd(fscal,dz10);
 534
 535             /* Update vectorial force */
 536             fix1             = _mm_add_pd(fix1,tx);
 537             fiy1             = _mm_add_pd(fiy1,ty);
 538             fiz1             = _mm_add_pd(fiz1,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             /**************************
 545              * CALCULATE INTERACTIONS *
 546              **************************/
 547
 548             r20              = _mm_mul_pd(rsq20,rinv20);
 549
 550             /* Compute parameters for interactions between i and j atoms */
 551             qq20             = _mm_mul_pd(iq2,jq0);
 552
 553             /* Calculate table index by multiplying r with table scale and truncate to integer */
 554             rt               = _mm_mul_pd(r20,vftabscale);
 555             vfitab           = _mm_cvttpd_epi32(rt);
 556             vfeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
 557             vfitab           = _mm_slli_epi32(vfitab,2);
 558
 559             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 560             Y                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
 561             F                = _mm_setzero_pd();
 562             GMX_MM_TRANSPOSE2_PD(Y,F);
 563             G                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
 564             H                = _mm_setzero_pd();
 565             GMX_MM_TRANSPOSE2_PD(G,H);
 566             Heps             = _mm_mul_pd(vfeps,H);
 567             Fp               = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
 568             VV               = _mm_add_pd(Y,_mm_mul_pd(vfeps,Fp));
 569             velec            = _mm_mul_pd(qq20,VV);
 570             FF               = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
 571             felec            = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq20,FF),_mm_mul_pd(vftabscale,rinv20)));
 572
 573             /* Update potential sum for this i atom from the interaction with this j atom. */
 574             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 575             velecsum         = _mm_add_pd(velecsum,velec);
 576
 577             fscal            = felec;
 578
 579             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 580
 581             /* Calculate temporary vectorial force */
 582             tx               = _mm_mul_pd(fscal,dx20);
 583             ty               = _mm_mul_pd(fscal,dy20);
 584             tz               = _mm_mul_pd(fscal,dz20);
 585
 586             /* Update vectorial force */
 587             fix2             = _mm_add_pd(fix2,tx);
 588             fiy2             = _mm_add_pd(fiy2,ty);
 589             fiz2             = _mm_add_pd(fiz2,tz);
 590
 591             fjx0             = _mm_add_pd(fjx0,tx);
 592             fjy0             = _mm_add_pd(fjy0,ty);
 593             fjz0             = _mm_add_pd(fjz0,tz);
 594
 595             /**************************
 596              * CALCULATE INTERACTIONS *
 597              **************************/
 598
 599             r30              = _mm_mul_pd(rsq30,rinv30);
 600
 601             /* Compute parameters for interactions between i and j atoms */
 602             qq30             = _mm_mul_pd(iq3,jq0);
 603
 604             /* Calculate table index by multiplying r with table scale and truncate to integer */
 605             rt               = _mm_mul_pd(r30,vftabscale);
 606             vfitab           = _mm_cvttpd_epi32(rt);
 607             vfeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
 608             vfitab           = _mm_slli_epi32(vfitab,2);
 609
 610             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 611             Y                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
 612             F                = _mm_setzero_pd();
 613             GMX_MM_TRANSPOSE2_PD(Y,F);
 614             G                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
 615             H                = _mm_setzero_pd();
 616             GMX_MM_TRANSPOSE2_PD(G,H);
 617             Heps             = _mm_mul_pd(vfeps,H);
 618             Fp               = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
 619             VV               = _mm_add_pd(Y,_mm_mul_pd(vfeps,Fp));
 620             velec            = _mm_mul_pd(qq30,VV);
 621             FF               = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
 622             felec            = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq30,FF),_mm_mul_pd(vftabscale,rinv30)));
 623
 624             /* Update potential sum for this i atom from the interaction with this j atom. */
 625             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 626             velecsum         = _mm_add_pd(velecsum,velec);
 627
 628             fscal            = felec;
 629
 630             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 631
 632             /* Calculate temporary vectorial force */
 633             tx               = _mm_mul_pd(fscal,dx30);
 634             ty               = _mm_mul_pd(fscal,dy30);
 635             tz               = _mm_mul_pd(fscal,dz30);
 636
 637             /* Update vectorial force */
 638             fix3             = _mm_add_pd(fix3,tx);
 639             fiy3             = _mm_add_pd(fiy3,ty);
 640             fiz3             = _mm_add_pd(fiz3,tz);
 641
 642             fjx0             = _mm_add_pd(fjx0,tx);
 643             fjy0             = _mm_add_pd(fjy0,ty);
 644             fjz0             = _mm_add_pd(fjz0,tz);
 645
 646             gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,fjx0,fjy0,fjz0);
 647
 648             /* Inner loop uses 164 flops */
 649         }
 650
 651         /* End of innermost loop */
 652
 653         gmx_mm_update_iforce_4atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
 654                                               f+i_coord_offset,fshift+i_shift_offset);
 655
 656         ggid                        = gid[iidx];
 657         /* Update potential energies */
 658         gmx_mm_update_1pot_pd(velecsum,kernel_data->energygrp_elec+ggid);
 659         gmx_mm_update_1pot_pd(vvdwsum,kernel_data->energygrp_vdw+ggid);
 660
 661         /* Increment number of inner iterations */
 662         inneriter                  += j_index_end - j_index_start;
 663
 664         /* Outer loop uses 26 flops */
 665     }
 666
 667     /* Increment number of outer iterations */
 668     outeriter        += nri;
 669
 670     /* Update outer/inner flops */
 671
 672     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W4_VF,outeriter*26 + inneriter*164);
 673 }
 674 /*
 675  * Gromacs nonbonded kernel:   nb_kernel_ElecCSTab_VdwLJ_GeomW4P1_F_sse4_1_double
 676  * Electrostatics interaction: CubicSplineTable
 677  * VdW interaction:            LennardJones
 678  * Geometry:                   Water4-Particle
 679  * Calculate force/pot:        Force
 680  */
 681 void
 682 nb_kernel_ElecCSTab_VdwLJ_GeomW4P1_F_sse4_1_double
 683                     (t_nblist                    * gmx_restrict       nlist,
 684                      rvec                        * gmx_restrict          xx,
 685                      rvec                        * gmx_restrict          ff,
 686                      struct t_forcerec           * gmx_restrict          fr,
 687                      t_mdatoms                   * gmx_restrict     mdatoms,
 688                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 689                      t_nrnb                      * gmx_restrict        nrnb)
 690 {
 691     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 692      * just 0 for non-waters.
 693      * Suffixes A,B refer to j loop unrolling done with SSE double precision, e.g. for the two different
 694      * jnr indices corresponding to data put in the four positions in the SIMD register.
 695      */
 696     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 697     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 698     int              jnrA,jnrB;
 699     int              j_coord_offsetA,j_coord_offsetB;
 700     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 701     real             rcutoff_scalar;
 702     real             *shiftvec,*fshift,*x,*f;
 703     __m128d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 704     int              vdwioffset0;
 705     __m128d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 706     int              vdwioffset1;
 707     __m128d          ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
 708     int              vdwioffset2;
 709     __m128d          ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
 710     int              vdwioffset3;
 711     __m128d          ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
 712     int              vdwjidx0A,vdwjidx0B;
 713     __m128d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 714     __m128d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 715     __m128d          dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
 716     __m128d          dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
 717     __m128d          dx30,dy30,dz30,rsq30,rinv30,rinvsq30,r30,qq30,c6_30,c12_30;
 718     __m128d          velec,felec,velecsum,facel,crf,krf,krf2;
 719     real             *charge;
 720     int              nvdwtype;
 721     __m128d          rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 722     int              *vdwtype;
 723     real             *vdwparam;
 724     __m128d          one_sixth   = _mm_set1_pd(1.0/6.0);
 725     __m128d          one_twelfth = _mm_set1_pd(1.0/12.0);
 726     __m128i          vfitab;
 727     __m128i          ifour       = _mm_set1_epi32(4);
 728     __m128d          rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF;
 729     real             *vftab;
 730     __m128d          dummy_mask,cutoff_mask;
 731     __m128d          signbit   = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
 732     __m128d          one     = _mm_set1_pd(1.0);
 733     __m128d          two     = _mm_set1_pd(2.0);
 734     x                = xx[0];
 735     f                = ff[0];
 736
 737     nri              = nlist->nri;
 738     iinr             = nlist->iinr;
 739     jindex           = nlist->jindex;
 740     jjnr             = nlist->jjnr;
 741     shiftidx         = nlist->shift;
 742     gid              = nlist->gid;
 743     shiftvec         = fr->shift_vec[0];
 744     fshift           = fr->fshift[0];
 745     facel            = _mm_set1_pd(fr->ic->epsfac);
 746     charge           = mdatoms->chargeA;
 747     nvdwtype         = fr->ntype;
 748     vdwparam         = fr->nbfp;
 749     vdwtype          = mdatoms->typeA;
 750
 751     vftab            = kernel_data->table_elec->data;
 752     vftabscale       = _mm_set1_pd(kernel_data->table_elec->scale);
 753
 754     /* Setup water-specific parameters */
 755     inr              = nlist->iinr[0];
 756     iq1              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+1]));
 757     iq2              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+2]));
 758     iq3              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+3]));
 759     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 760
 761     /* Avoid stupid compiler warnings */
 762     jnrA = jnrB = 0;
 763     j_coord_offsetA = 0;
 764     j_coord_offsetB = 0;
 765
 766     outeriter        = 0;
 767     inneriter        = 0;
 768
 769     /* Start outer loop over neighborlists */
 770     for(iidx=0; iidx<nri; iidx++)
 771     {
 772         /* Load shift vector for this list */
 773         i_shift_offset   = DIM*shiftidx[iidx];
 774
 775         /* Load limits for loop over neighbors */
 776         j_index_start    = jindex[iidx];
 777         j_index_end      = jindex[iidx+1];
 778
 779         /* Get outer coordinate index */
 780         inr              = iinr[iidx];
 781         i_coord_offset   = DIM*inr;
 782
 783         /* Load i particle coords and add shift vector */
 784         gmx_mm_load_shift_and_4rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
 785                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
 786
 787         fix0             = _mm_setzero_pd();
 788         fiy0             = _mm_setzero_pd();
 789         fiz0             = _mm_setzero_pd();
 790         fix1             = _mm_setzero_pd();
 791         fiy1             = _mm_setzero_pd();
 792         fiz1             = _mm_setzero_pd();
 793         fix2             = _mm_setzero_pd();
 794         fiy2             = _mm_setzero_pd();
 795         fiz2             = _mm_setzero_pd();
 796         fix3             = _mm_setzero_pd();
 797         fiy3             = _mm_setzero_pd();
 798         fiz3             = _mm_setzero_pd();
 799
 800         /* Start inner kernel loop */
 801         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 802         {
 803
 804             /* Get j neighbor index, and coordinate index */
 805             jnrA             = jjnr[jidx];
 806             jnrB             = jjnr[jidx+1];
 807             j_coord_offsetA  = DIM*jnrA;
 808             j_coord_offsetB  = DIM*jnrB;
 809
 810             /* load j atom coordinates */
 811             gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 812                                               &jx0,&jy0,&jz0);
 813
 814             /* Calculate displacement vector */
 815             dx00             = _mm_sub_pd(ix0,jx0);
 816             dy00             = _mm_sub_pd(iy0,jy0);
 817             dz00             = _mm_sub_pd(iz0,jz0);
 818             dx10             = _mm_sub_pd(ix1,jx0);
 819             dy10             = _mm_sub_pd(iy1,jy0);
 820             dz10             = _mm_sub_pd(iz1,jz0);
 821             dx20             = _mm_sub_pd(ix2,jx0);
 822             dy20             = _mm_sub_pd(iy2,jy0);
 823             dz20             = _mm_sub_pd(iz2,jz0);
 824             dx30             = _mm_sub_pd(ix3,jx0);
 825             dy30             = _mm_sub_pd(iy3,jy0);
 826             dz30             = _mm_sub_pd(iz3,jz0);
 827
 828             /* Calculate squared distance and things based on it */
 829             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 830             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
 831             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
 832             rsq30            = gmx_mm_calc_rsq_pd(dx30,dy30,dz30);
 833
 834             rinv10           = sse41_invsqrt_d(rsq10);
 835             rinv20           = sse41_invsqrt_d(rsq20);
 836             rinv30           = sse41_invsqrt_d(rsq30);
 837
 838             rinvsq00         = sse41_inv_d(rsq00);
 839
 840             /* Load parameters for j particles */
 841             jq0              = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
 842             vdwjidx0A        = 2*vdwtype[jnrA+0];
 843             vdwjidx0B        = 2*vdwtype[jnrB+0];
 844
 845             fjx0             = _mm_setzero_pd();
 846             fjy0             = _mm_setzero_pd();
 847             fjz0             = _mm_setzero_pd();
 848
 849             /**************************
 850              * CALCULATE INTERACTIONS *
 851              **************************/
 852
 853             /* Compute parameters for interactions between i and j atoms */
 854             gmx_mm_load_2pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,
 855                                          vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
 856
 857             /* LENNARD-JONES DISPERSION/REPULSION */
 858
 859             rinvsix          = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 860             fvdw             = _mm_mul_pd(_mm_sub_pd(_mm_mul_pd(c12_00,rinvsix),c6_00),_mm_mul_pd(rinvsix,rinvsq00));
 861
 862             fscal            = fvdw;
 863
 864             /* Calculate temporary vectorial force */
 865             tx               = _mm_mul_pd(fscal,dx00);
 866             ty               = _mm_mul_pd(fscal,dy00);
 867             tz               = _mm_mul_pd(fscal,dz00);
 868
 869             /* Update vectorial force */
 870             fix0             = _mm_add_pd(fix0,tx);
 871             fiy0             = _mm_add_pd(fiy0,ty);
 872             fiz0             = _mm_add_pd(fiz0,tz);
 873
 874             fjx0             = _mm_add_pd(fjx0,tx);
 875             fjy0             = _mm_add_pd(fjy0,ty);
 876             fjz0             = _mm_add_pd(fjz0,tz);
 877
 878             /**************************
 879              * CALCULATE INTERACTIONS *
 880              **************************/
 881
 882             r10              = _mm_mul_pd(rsq10,rinv10);
 883
 884             /* Compute parameters for interactions between i and j atoms */
 885             qq10             = _mm_mul_pd(iq1,jq0);
 886
 887             /* Calculate table index by multiplying r with table scale and truncate to integer */
 888             rt               = _mm_mul_pd(r10,vftabscale);
 889             vfitab           = _mm_cvttpd_epi32(rt);
 890             vfeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
 891             vfitab           = _mm_slli_epi32(vfitab,2);
 892
 893             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 894             Y                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
 895             F                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) );
 896             GMX_MM_TRANSPOSE2_PD(Y,F);
 897             G                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
 898             H                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) +2);
 899             GMX_MM_TRANSPOSE2_PD(G,H);
 900             Heps             = _mm_mul_pd(vfeps,H);
 901             Fp               = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
 902             FF               = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
 903             felec            = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq10,FF),_mm_mul_pd(vftabscale,rinv10)));
 904
 905             fscal            = felec;
 906
 907             /* Calculate temporary vectorial force */
 908             tx               = _mm_mul_pd(fscal,dx10);
 909             ty               = _mm_mul_pd(fscal,dy10);
 910             tz               = _mm_mul_pd(fscal,dz10);
 911
 912             /* Update vectorial force */
 913             fix1             = _mm_add_pd(fix1,tx);
 914             fiy1             = _mm_add_pd(fiy1,ty);
 915             fiz1             = _mm_add_pd(fiz1,tz);
 916
 917             fjx0             = _mm_add_pd(fjx0,tx);
 918             fjy0             = _mm_add_pd(fjy0,ty);
 919             fjz0             = _mm_add_pd(fjz0,tz);
 920
 921             /**************************
 922              * CALCULATE INTERACTIONS *
 923              **************************/
 924
 925             r20              = _mm_mul_pd(rsq20,rinv20);
 926
 927             /* Compute parameters for interactions between i and j atoms */
 928             qq20             = _mm_mul_pd(iq2,jq0);
 929
 930             /* Calculate table index by multiplying r with table scale and truncate to integer */
 931             rt               = _mm_mul_pd(r20,vftabscale);
 932             vfitab           = _mm_cvttpd_epi32(rt);
 933             vfeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
 934             vfitab           = _mm_slli_epi32(vfitab,2);
 935
 936             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 937             Y                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
 938             F                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) );
 939             GMX_MM_TRANSPOSE2_PD(Y,F);
 940             G                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
 941             H                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) +2);
 942             GMX_MM_TRANSPOSE2_PD(G,H);
 943             Heps             = _mm_mul_pd(vfeps,H);
 944             Fp               = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
 945             FF               = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
 946             felec            = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq20,FF),_mm_mul_pd(vftabscale,rinv20)));
 947
 948             fscal            = felec;
 949
 950             /* Calculate temporary vectorial force */
 951             tx               = _mm_mul_pd(fscal,dx20);
 952             ty               = _mm_mul_pd(fscal,dy20);
 953             tz               = _mm_mul_pd(fscal,dz20);
 954
 955             /* Update vectorial force */
 956             fix2             = _mm_add_pd(fix2,tx);
 957             fiy2             = _mm_add_pd(fiy2,ty);
 958             fiz2             = _mm_add_pd(fiz2,tz);
 959
 960             fjx0             = _mm_add_pd(fjx0,tx);
 961             fjy0             = _mm_add_pd(fjy0,ty);
 962             fjz0             = _mm_add_pd(fjz0,tz);
 963
 964             /**************************
 965              * CALCULATE INTERACTIONS *
 966              **************************/
 967
 968             r30              = _mm_mul_pd(rsq30,rinv30);
 969
 970             /* Compute parameters for interactions between i and j atoms */
 971             qq30             = _mm_mul_pd(iq3,jq0);
 972
 973             /* Calculate table index by multiplying r with table scale and truncate to integer */
 974             rt               = _mm_mul_pd(r30,vftabscale);
 975             vfitab           = _mm_cvttpd_epi32(rt);
 976             vfeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
 977             vfitab           = _mm_slli_epi32(vfitab,2);
 978
 979             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 980             Y                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
 981             F                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) );
 982             GMX_MM_TRANSPOSE2_PD(Y,F);
 983             G                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
 984             H                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) +2);
 985             GMX_MM_TRANSPOSE2_PD(G,H);
 986             Heps             = _mm_mul_pd(vfeps,H);
 987             Fp               = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
 988             FF               = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
 989             felec            = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq30,FF),_mm_mul_pd(vftabscale,rinv30)));
 990
 991             fscal            = felec;
 992
 993             /* Calculate temporary vectorial force */
 994             tx               = _mm_mul_pd(fscal,dx30);
 995             ty               = _mm_mul_pd(fscal,dy30);
 996             tz               = _mm_mul_pd(fscal,dz30);
 997
 998             /* Update vectorial force */
 999             fix3             = _mm_add_pd(fix3,tx);
1000             fiy3             = _mm_add_pd(fiy3,ty);
1001             fiz3             = _mm_add_pd(fiz3,tz);
1002
1003             fjx0             = _mm_add_pd(fjx0,tx);
1004             fjy0             = _mm_add_pd(fjy0,ty);
1005             fjz0             = _mm_add_pd(fjz0,tz);
1006
1007             gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,fjx0,fjy0,fjz0);
1008
1009             /* Inner loop uses 147 flops */
1010         }
1011
1012         if(jidx<j_index_end)
1013         {
1014
1015             jnrA             = jjnr[jidx];
1016             j_coord_offsetA  = DIM*jnrA;
1017
1018             /* load j atom coordinates */
1019             gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
1020                                               &jx0,&jy0,&jz0);
1021
1022             /* Calculate displacement vector */
1023             dx00             = _mm_sub_pd(ix0,jx0);
1024             dy00             = _mm_sub_pd(iy0,jy0);
1025             dz00             = _mm_sub_pd(iz0,jz0);
1026             dx10             = _mm_sub_pd(ix1,jx0);
1027             dy10             = _mm_sub_pd(iy1,jy0);
1028             dz10             = _mm_sub_pd(iz1,jz0);
1029             dx20             = _mm_sub_pd(ix2,jx0);
1030             dy20             = _mm_sub_pd(iy2,jy0);
1031             dz20             = _mm_sub_pd(iz2,jz0);
1032             dx30             = _mm_sub_pd(ix3,jx0);
1033             dy30             = _mm_sub_pd(iy3,jy0);
1034             dz30             = _mm_sub_pd(iz3,jz0);
1035
1036             /* Calculate squared distance and things based on it */
1037             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
1038             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
1039             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
1040             rsq30            = gmx_mm_calc_rsq_pd(dx30,dy30,dz30);
1041
1042             rinv10           = sse41_invsqrt_d(rsq10);
1043             rinv20           = sse41_invsqrt_d(rsq20);
1044             rinv30           = sse41_invsqrt_d(rsq30);
1045
1046             rinvsq00         = sse41_inv_d(rsq00);
1047
1048             /* Load parameters for j particles */
1049             jq0              = _mm_load_sd(charge+jnrA+0);
1050             vdwjidx0A        = 2*vdwtype[jnrA+0];
1051
1052             fjx0             = _mm_setzero_pd();
1053             fjy0             = _mm_setzero_pd();
1054             fjz0             = _mm_setzero_pd();
1055
1056             /**************************
1057              * CALCULATE INTERACTIONS *
1058              **************************/
1059
1060             /* Compute parameters for interactions between i and j atoms */
1061             gmx_mm_load_1pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,&c6_00,&c12_00);
1062
1063             /* LENNARD-JONES DISPERSION/REPULSION */
1064
1065             rinvsix          = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
1066             fvdw             = _mm_mul_pd(_mm_sub_pd(_mm_mul_pd(c12_00,rinvsix),c6_00),_mm_mul_pd(rinvsix,rinvsq00));
1067
1068             fscal            = fvdw;
1069
1070             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1071
1072             /* Calculate temporary vectorial force */
1073             tx               = _mm_mul_pd(fscal,dx00);
1074             ty               = _mm_mul_pd(fscal,dy00);
1075             tz               = _mm_mul_pd(fscal,dz00);
1076
1077             /* Update vectorial force */
1078             fix0             = _mm_add_pd(fix0,tx);
1079             fiy0             = _mm_add_pd(fiy0,ty);
1080             fiz0             = _mm_add_pd(fiz0,tz);
1081
1082             fjx0             = _mm_add_pd(fjx0,tx);
1083             fjy0             = _mm_add_pd(fjy0,ty);
1084             fjz0             = _mm_add_pd(fjz0,tz);
1085
1086             /**************************
1087              * CALCULATE INTERACTIONS *
1088              **************************/
1089
1090             r10              = _mm_mul_pd(rsq10,rinv10);
1091
1092             /* Compute parameters for interactions between i and j atoms */
1093             qq10             = _mm_mul_pd(iq1,jq0);
1094
1095             /* Calculate table index by multiplying r with table scale and truncate to integer */
1096             rt               = _mm_mul_pd(r10,vftabscale);
1097             vfitab           = _mm_cvttpd_epi32(rt);
1098             vfeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
1099             vfitab           = _mm_slli_epi32(vfitab,2);
1100
1101             /* CUBIC SPLINE TABLE ELECTROSTATICS */
1102             Y                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
1103             F                = _mm_setzero_pd();
1104             GMX_MM_TRANSPOSE2_PD(Y,F);
1105             G                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
1106             H                = _mm_setzero_pd();
1107             GMX_MM_TRANSPOSE2_PD(G,H);
1108             Heps             = _mm_mul_pd(vfeps,H);
1109             Fp               = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
1110             FF               = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
1111             felec            = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq10,FF),_mm_mul_pd(vftabscale,rinv10)));
1112
1113             fscal            = felec;
1114
1115             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1116
1117             /* Calculate temporary vectorial force */
1118             tx               = _mm_mul_pd(fscal,dx10);
1119             ty               = _mm_mul_pd(fscal,dy10);
1120             tz               = _mm_mul_pd(fscal,dz10);
1121
1122             /* Update vectorial force */
1123             fix1             = _mm_add_pd(fix1,tx);
1124             fiy1             = _mm_add_pd(fiy1,ty);
1125             fiz1             = _mm_add_pd(fiz1,tz);
1126
1127             fjx0             = _mm_add_pd(fjx0,tx);
1128             fjy0             = _mm_add_pd(fjy0,ty);
1129             fjz0             = _mm_add_pd(fjz0,tz);
1130
1131             /**************************
1132              * CALCULATE INTERACTIONS *
1133              **************************/
1134
1135             r20              = _mm_mul_pd(rsq20,rinv20);
1136
1137             /* Compute parameters for interactions between i and j atoms */
1138             qq20             = _mm_mul_pd(iq2,jq0);
1139
1140             /* Calculate table index by multiplying r with table scale and truncate to integer */
1141             rt               = _mm_mul_pd(r20,vftabscale);
1142             vfitab           = _mm_cvttpd_epi32(rt);
1143             vfeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
1144             vfitab           = _mm_slli_epi32(vfitab,2);
1145
1146             /* CUBIC SPLINE TABLE ELECTROSTATICS */
1147             Y                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
1148             F                = _mm_setzero_pd();
1149             GMX_MM_TRANSPOSE2_PD(Y,F);
1150             G                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
1151             H                = _mm_setzero_pd();
1152             GMX_MM_TRANSPOSE2_PD(G,H);
1153             Heps             = _mm_mul_pd(vfeps,H);
1154             Fp               = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
1155             FF               = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
1156             felec            = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq20,FF),_mm_mul_pd(vftabscale,rinv20)));
1157
1158             fscal            = felec;
1159
1160             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1161
1162             /* Calculate temporary vectorial force */
1163             tx               = _mm_mul_pd(fscal,dx20);
1164             ty               = _mm_mul_pd(fscal,dy20);
1165             tz               = _mm_mul_pd(fscal,dz20);
1166
1167             /* Update vectorial force */
1168             fix2             = _mm_add_pd(fix2,tx);
1169             fiy2             = _mm_add_pd(fiy2,ty);
1170             fiz2             = _mm_add_pd(fiz2,tz);
1171
1172             fjx0             = _mm_add_pd(fjx0,tx);
1173             fjy0             = _mm_add_pd(fjy0,ty);
1174             fjz0             = _mm_add_pd(fjz0,tz);
1175
1176             /**************************
1177              * CALCULATE INTERACTIONS *
1178              **************************/
1179
1180             r30              = _mm_mul_pd(rsq30,rinv30);
1181
1182             /* Compute parameters for interactions between i and j atoms */
1183             qq30             = _mm_mul_pd(iq3,jq0);
1184
1185             /* Calculate table index by multiplying r with table scale and truncate to integer */
1186             rt               = _mm_mul_pd(r30,vftabscale);
1187             vfitab           = _mm_cvttpd_epi32(rt);
1188             vfeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
1189             vfitab           = _mm_slli_epi32(vfitab,2);
1190
1191             /* CUBIC SPLINE TABLE ELECTROSTATICS */
1192             Y                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
1193             F                = _mm_setzero_pd();
1194             GMX_MM_TRANSPOSE2_PD(Y,F);
1195             G                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
1196             H                = _mm_setzero_pd();
1197             GMX_MM_TRANSPOSE2_PD(G,H);
1198             Heps             = _mm_mul_pd(vfeps,H);
1199             Fp               = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
1200             FF               = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
1201             felec            = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq30,FF),_mm_mul_pd(vftabscale,rinv30)));
1202
1203             fscal            = felec;
1204
1205             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1206
1207             /* Calculate temporary vectorial force */
1208             tx               = _mm_mul_pd(fscal,dx30);
1209             ty               = _mm_mul_pd(fscal,dy30);
1210             tz               = _mm_mul_pd(fscal,dz30);
1211
1212             /* Update vectorial force */
1213             fix3             = _mm_add_pd(fix3,tx);
1214             fiy3             = _mm_add_pd(fiy3,ty);
1215             fiz3             = _mm_add_pd(fiz3,tz);
1216
1217             fjx0             = _mm_add_pd(fjx0,tx);
1218             fjy0             = _mm_add_pd(fjy0,ty);
1219             fjz0             = _mm_add_pd(fjz0,tz);
1220
1221             gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,fjx0,fjy0,fjz0);
1222
1223             /* Inner loop uses 147 flops */
1224         }
1225
1226         /* End of innermost loop */
1227
1228         gmx_mm_update_iforce_4atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
1229                                               f+i_coord_offset,fshift+i_shift_offset);
1230
1231         /* Increment number of inner iterations */
1232         inneriter                  += j_index_end - j_index_start;
1233
1234         /* Outer loop uses 24 flops */
1235     }
1236
1237     /* Increment number of outer iterations */
1238     outeriter        += nri;
1239
1240     /* Update outer/inner flops */
1241
1242     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W4_F,outeriter*24 + inneriter*147);
1243 }