src/gromacs/gmxlib/nonbonded/nb_kernel_sse4_1_double/nb_kernel_ElecCSTab_VdwCSTab_GeomW3P1_sse4_1_double.c

   1 /*
   2  * This file is part of the GROMACS molecular simulation package.
   3  *
   4  * Copyright (c) 2012,2013, by the GROMACS development team, led by
   5  * Mark Abraham, David van der Spoel, Berk Hess, and Erik Lindahl,
   6  * and including many others, as listed in the AUTHORS file in the
   7  * top-level source directory and at http://www.gromacs.org.
   8  *
   9  * GROMACS is free software; you can redistribute it and/or
  10  * modify it under the terms of the GNU Lesser General Public License
  11  * as published by the Free Software Foundation; either version 2.1
  12  * of the License, or (at your option) any later version.
  13  *
  14  * GROMACS is distributed in the hope that it will be useful,
  15  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  16  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  17  * Lesser General Public License for more details.
  18  *
  19  * You should have received a copy of the GNU Lesser General Public
  20  * License along with GROMACS; if not, see
  21  * http://www.gnu.org/licenses, or write to the Free Software Foundation,
  22  * Inc., 51 Franklin Street, Fifth Floor, Boston, MA  02110-1301  USA.
  23  *
  24  * If you want to redistribute modifications to GROMACS, please
  25  * consider that scientific software is very special. Version
  26  * control is crucial - bugs must be traceable. We will be happy to
  27  * consider code for inclusion in the official distribution, but
  28  * derived work must not be called official GROMACS. Details are found
  29  * in the README & COPYING files - if they are missing, get the
  30  * official version at http://www.gromacs.org.
  31  *
  32  * To help us fund GROMACS development, we humbly ask that you cite
  33  * the research papers on the package. Check out http://www.gromacs.org.
  34  */
  35 /*
  36  * Note: this file was generated by the GROMACS sse4_1_double kernel generator.
  37  */
  38 #ifdef HAVE_CONFIG_H
  39 #include <config.h>
  40 #endif
  41
  42 #include <math.h>
  43
  44 #include "../nb_kernel.h"
  45 #include "types/simple.h"
  46 #include "vec.h"
  47 #include "nrnb.h"
  48
  49 #include "gromacs/simd/math_x86_sse4_1_double.h"
  50 #include "kernelutil_x86_sse4_1_double.h"
  51
  52 /*
  53  * Gromacs nonbonded kernel:   nb_kernel_ElecCSTab_VdwCSTab_GeomW3P1_VF_sse4_1_double
  54  * Electrostatics interaction: CubicSplineTable
  55  * VdW interaction:            CubicSplineTable
  56  * Geometry:                   Water3-Particle
  57  * Calculate force/pot:        PotentialAndForce
  58  */
  59 void
  60 nb_kernel_ElecCSTab_VdwCSTab_GeomW3P1_VF_sse4_1_double
  61                     (t_nblist                    * gmx_restrict       nlist,
  62                      rvec                        * gmx_restrict          xx,
  63                      rvec                        * gmx_restrict          ff,
  64                      t_forcerec                  * gmx_restrict          fr,
  65                      t_mdatoms                   * gmx_restrict     mdatoms,
  66                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
  67                      t_nrnb                      * gmx_restrict        nrnb)
  68 {
  69     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
  70      * just 0 for non-waters.
  71      * Suffixes A,B refer to j loop unrolling done with SSE double precision, e.g. for the two different
  72      * jnr indices corresponding to data put in the four positions in the SIMD register.
  73      */
  74     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
  75     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
  76     int              jnrA,jnrB;
  77     int              j_coord_offsetA,j_coord_offsetB;
  78     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
  79     real             rcutoff_scalar;
  80     real             *shiftvec,*fshift,*x,*f;
  81     __m128d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
  82     int              vdwioffset0;
  83     __m128d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
  84     int              vdwioffset1;
  85     __m128d          ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
  86     int              vdwioffset2;
  87     __m128d          ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
  88     int              vdwjidx0A,vdwjidx0B;
  89     __m128d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  90     __m128d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  91     __m128d          dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
  92     __m128d          dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
  93     __m128d          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->epsfac);
 121     charge           = mdatoms->chargeA;
 122     nvdwtype         = fr->ntype;
 123     vdwparam         = fr->nbfp;
 124     vdwtype          = mdatoms->typeA;
 125
 126     vftab            = kernel_data->table_elec_vdw->data;
 127     vftabscale       = _mm_set1_pd(kernel_data->table_elec_vdw->scale);
 128
 129     /* Setup water-specific parameters */
 130     inr              = nlist->iinr[0];
 131     iq0              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+0]));
 132     iq1              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+1]));
 133     iq2              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+2]));
 134     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_3rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
 160                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 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
 172         /* Reset potential sums */
 173         velecsum         = _mm_setzero_pd();
 174         vvdwsum          = _mm_setzero_pd();
 175
 176         /* Start inner kernel loop */
 177         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 178         {
 179
 180             /* Get j neighbor index, and coordinate index */
 181             jnrA             = jjnr[jidx];
 182             jnrB             = jjnr[jidx+1];
 183             j_coord_offsetA  = DIM*jnrA;
 184             j_coord_offsetB  = DIM*jnrB;
 185
 186             /* load j atom coordinates */
 187             gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 188                                               &jx0,&jy0,&jz0);
 189
 190             /* Calculate displacement vector */
 191             dx00             = _mm_sub_pd(ix0,jx0);
 192             dy00             = _mm_sub_pd(iy0,jy0);
 193             dz00             = _mm_sub_pd(iz0,jz0);
 194             dx10             = _mm_sub_pd(ix1,jx0);
 195             dy10             = _mm_sub_pd(iy1,jy0);
 196             dz10             = _mm_sub_pd(iz1,jz0);
 197             dx20             = _mm_sub_pd(ix2,jx0);
 198             dy20             = _mm_sub_pd(iy2,jy0);
 199             dz20             = _mm_sub_pd(iz2,jz0);
 200
 201             /* Calculate squared distance and things based on it */
 202             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 203             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
 204             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
 205
 206             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 207             rinv10           = gmx_mm_invsqrt_pd(rsq10);
 208             rinv20           = gmx_mm_invsqrt_pd(rsq20);
 209
 210             /* Load parameters for j particles */
 211             jq0              = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
 212             vdwjidx0A        = 2*vdwtype[jnrA+0];
 213             vdwjidx0B        = 2*vdwtype[jnrB+0];
 214
 215             fjx0             = _mm_setzero_pd();
 216             fjy0             = _mm_setzero_pd();
 217             fjz0             = _mm_setzero_pd();
 218
 219             /**************************
 220              * CALCULATE INTERACTIONS *
 221              **************************/
 222
 223             r00              = _mm_mul_pd(rsq00,rinv00);
 224
 225             /* Compute parameters for interactions between i and j atoms */
 226             qq00             = _mm_mul_pd(iq0,jq0);
 227             gmx_mm_load_2pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,
 228                                          vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
 229
 230             /* Calculate table index by multiplying r with table scale and truncate to integer */
 231             rt               = _mm_mul_pd(r00,vftabscale);
 232             vfitab           = _mm_cvttpd_epi32(rt);
 233             vfeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
 234             vfitab           = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
 235
 236             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 237             Y                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
 238             F                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) );
 239             GMX_MM_TRANSPOSE2_PD(Y,F);
 240             G                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
 241             H                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) +2);
 242             GMX_MM_TRANSPOSE2_PD(G,H);
 243             Heps             = _mm_mul_pd(vfeps,H);
 244             Fp               = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
 245             VV               = _mm_add_pd(Y,_mm_mul_pd(vfeps,Fp));
 246             velec            = _mm_mul_pd(qq00,VV);
 247             FF               = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
 248             felec            = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq00,FF),_mm_mul_pd(vftabscale,rinv00)));
 249
 250             /* CUBIC SPLINE TABLE DISPERSION */
 251             vfitab           = _mm_add_epi32(vfitab,ifour);
 252             Y                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
 253             F                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) );
 254             GMX_MM_TRANSPOSE2_PD(Y,F);
 255             G                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
 256             H                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) +2);
 257             GMX_MM_TRANSPOSE2_PD(G,H);
 258             Heps             = _mm_mul_pd(vfeps,H);
 259             Fp               = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
 260             VV               = _mm_add_pd(Y,_mm_mul_pd(vfeps,Fp));
 261             vvdw6            = _mm_mul_pd(c6_00,VV);
 262             FF               = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
 263             fvdw6            = _mm_mul_pd(c6_00,FF);
 264
 265             /* CUBIC SPLINE TABLE REPULSION */
 266             vfitab           = _mm_add_epi32(vfitab,ifour);
 267             Y                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
 268             F                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) );
 269             GMX_MM_TRANSPOSE2_PD(Y,F);
 270             G                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
 271             H                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) +2);
 272             GMX_MM_TRANSPOSE2_PD(G,H);
 273             Heps             = _mm_mul_pd(vfeps,H);
 274             Fp               = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
 275             VV               = _mm_add_pd(Y,_mm_mul_pd(vfeps,Fp));
 276             vvdw12           = _mm_mul_pd(c12_00,VV);
 277             FF               = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
 278             fvdw12           = _mm_mul_pd(c12_00,FF);
 279             vvdw             = _mm_add_pd(vvdw12,vvdw6);
 280             fvdw             = _mm_xor_pd(signbit,_mm_mul_pd(_mm_add_pd(fvdw6,fvdw12),_mm_mul_pd(vftabscale,rinv00)));
 281
 282             /* Update potential sum for this i atom from the interaction with this j atom. */
 283             velecsum         = _mm_add_pd(velecsum,velec);
 284             vvdwsum          = _mm_add_pd(vvdwsum,vvdw);
 285
 286             fscal            = _mm_add_pd(felec,fvdw);
 287
 288             /* Calculate temporary vectorial force */
 289             tx               = _mm_mul_pd(fscal,dx00);
 290             ty               = _mm_mul_pd(fscal,dy00);
 291             tz               = _mm_mul_pd(fscal,dz00);
 292
 293             /* Update vectorial force */
 294             fix0             = _mm_add_pd(fix0,tx);
 295             fiy0             = _mm_add_pd(fiy0,ty);
 296             fiz0             = _mm_add_pd(fiz0,tz);
 297
 298             fjx0             = _mm_add_pd(fjx0,tx);
 299             fjy0             = _mm_add_pd(fjy0,ty);
 300             fjz0             = _mm_add_pd(fjz0,tz);
 301
 302             /**************************
 303              * CALCULATE INTERACTIONS *
 304              **************************/
 305
 306             r10              = _mm_mul_pd(rsq10,rinv10);
 307
 308             /* Compute parameters for interactions between i and j atoms */
 309             qq10             = _mm_mul_pd(iq1,jq0);
 310
 311             /* Calculate table index by multiplying r with table scale and truncate to integer */
 312             rt               = _mm_mul_pd(r10,vftabscale);
 313             vfitab           = _mm_cvttpd_epi32(rt);
 314             vfeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
 315             vfitab           = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
 316
 317             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 318             Y                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
 319             F                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) );
 320             GMX_MM_TRANSPOSE2_PD(Y,F);
 321             G                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
 322             H                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) +2);
 323             GMX_MM_TRANSPOSE2_PD(G,H);
 324             Heps             = _mm_mul_pd(vfeps,H);
 325             Fp               = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
 326             VV               = _mm_add_pd(Y,_mm_mul_pd(vfeps,Fp));
 327             velec            = _mm_mul_pd(qq10,VV);
 328             FF               = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
 329             felec            = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq10,FF),_mm_mul_pd(vftabscale,rinv10)));
 330
 331             /* Update potential sum for this i atom from the interaction with this j atom. */
 332             velecsum         = _mm_add_pd(velecsum,velec);
 333
 334             fscal            = felec;
 335
 336             /* Calculate temporary vectorial force */
 337             tx               = _mm_mul_pd(fscal,dx10);
 338             ty               = _mm_mul_pd(fscal,dy10);
 339             tz               = _mm_mul_pd(fscal,dz10);
 340
 341             /* Update vectorial force */
 342             fix1             = _mm_add_pd(fix1,tx);
 343             fiy1             = _mm_add_pd(fiy1,ty);
 344             fiz1             = _mm_add_pd(fiz1,tz);
 345
 346             fjx0             = _mm_add_pd(fjx0,tx);
 347             fjy0             = _mm_add_pd(fjy0,ty);
 348             fjz0             = _mm_add_pd(fjz0,tz);
 349
 350             /**************************
 351              * CALCULATE INTERACTIONS *
 352              **************************/
 353
 354             r20              = _mm_mul_pd(rsq20,rinv20);
 355
 356             /* Compute parameters for interactions between i and j atoms */
 357             qq20             = _mm_mul_pd(iq2,jq0);
 358
 359             /* Calculate table index by multiplying r with table scale and truncate to integer */
 360             rt               = _mm_mul_pd(r20,vftabscale);
 361             vfitab           = _mm_cvttpd_epi32(rt);
 362             vfeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
 363             vfitab           = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
 364
 365             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 366             Y                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
 367             F                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) );
 368             GMX_MM_TRANSPOSE2_PD(Y,F);
 369             G                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
 370             H                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) +2);
 371             GMX_MM_TRANSPOSE2_PD(G,H);
 372             Heps             = _mm_mul_pd(vfeps,H);
 373             Fp               = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
 374             VV               = _mm_add_pd(Y,_mm_mul_pd(vfeps,Fp));
 375             velec            = _mm_mul_pd(qq20,VV);
 376             FF               = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
 377             felec            = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq20,FF),_mm_mul_pd(vftabscale,rinv20)));
 378
 379             /* Update potential sum for this i atom from the interaction with this j atom. */
 380             velecsum         = _mm_add_pd(velecsum,velec);
 381
 382             fscal            = felec;
 383
 384             /* Calculate temporary vectorial force */
 385             tx               = _mm_mul_pd(fscal,dx20);
 386             ty               = _mm_mul_pd(fscal,dy20);
 387             tz               = _mm_mul_pd(fscal,dz20);
 388
 389             /* Update vectorial force */
 390             fix2             = _mm_add_pd(fix2,tx);
 391             fiy2             = _mm_add_pd(fiy2,ty);
 392             fiz2             = _mm_add_pd(fiz2,tz);
 393
 394             fjx0             = _mm_add_pd(fjx0,tx);
 395             fjy0             = _mm_add_pd(fjy0,ty);
 396             fjz0             = _mm_add_pd(fjz0,tz);
 397
 398             gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,fjx0,fjy0,fjz0);
 399
 400             /* Inner loop uses 162 flops */
 401         }
 402
 403         if(jidx<j_index_end)
 404         {
 405
 406             jnrA             = jjnr[jidx];
 407             j_coord_offsetA  = DIM*jnrA;
 408
 409             /* load j atom coordinates */
 410             gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
 411                                               &jx0,&jy0,&jz0);
 412
 413             /* Calculate displacement vector */
 414             dx00             = _mm_sub_pd(ix0,jx0);
 415             dy00             = _mm_sub_pd(iy0,jy0);
 416             dz00             = _mm_sub_pd(iz0,jz0);
 417             dx10             = _mm_sub_pd(ix1,jx0);
 418             dy10             = _mm_sub_pd(iy1,jy0);
 419             dz10             = _mm_sub_pd(iz1,jz0);
 420             dx20             = _mm_sub_pd(ix2,jx0);
 421             dy20             = _mm_sub_pd(iy2,jy0);
 422             dz20             = _mm_sub_pd(iz2,jz0);
 423
 424             /* Calculate squared distance and things based on it */
 425             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 426             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
 427             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
 428
 429             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 430             rinv10           = gmx_mm_invsqrt_pd(rsq10);
 431             rinv20           = gmx_mm_invsqrt_pd(rsq20);
 432
 433             /* Load parameters for j particles */
 434             jq0              = _mm_load_sd(charge+jnrA+0);
 435             vdwjidx0A        = 2*vdwtype[jnrA+0];
 436
 437             fjx0             = _mm_setzero_pd();
 438             fjy0             = _mm_setzero_pd();
 439             fjz0             = _mm_setzero_pd();
 440
 441             /**************************
 442              * CALCULATE INTERACTIONS *
 443              **************************/
 444
 445             r00              = _mm_mul_pd(rsq00,rinv00);
 446
 447             /* Compute parameters for interactions between i and j atoms */
 448             qq00             = _mm_mul_pd(iq0,jq0);
 449             gmx_mm_load_1pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,&c6_00,&c12_00);
 450
 451             /* Calculate table index by multiplying r with table scale and truncate to integer */
 452             rt               = _mm_mul_pd(r00,vftabscale);
 453             vfitab           = _mm_cvttpd_epi32(rt);
 454             vfeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
 455             vfitab           = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
 456
 457             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 458             Y                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
 459             F                = _mm_setzero_pd();
 460             GMX_MM_TRANSPOSE2_PD(Y,F);
 461             G                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
 462             H                = _mm_setzero_pd();
 463             GMX_MM_TRANSPOSE2_PD(G,H);
 464             Heps             = _mm_mul_pd(vfeps,H);
 465             Fp               = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
 466             VV               = _mm_add_pd(Y,_mm_mul_pd(vfeps,Fp));
 467             velec            = _mm_mul_pd(qq00,VV);
 468             FF               = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
 469             felec            = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq00,FF),_mm_mul_pd(vftabscale,rinv00)));
 470
 471             /* CUBIC SPLINE TABLE DISPERSION */
 472             vfitab           = _mm_add_epi32(vfitab,ifour);
 473             Y                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
 474             F                = _mm_setzero_pd();
 475             GMX_MM_TRANSPOSE2_PD(Y,F);
 476             G                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
 477             H                = _mm_setzero_pd();
 478             GMX_MM_TRANSPOSE2_PD(G,H);
 479             Heps             = _mm_mul_pd(vfeps,H);
 480             Fp               = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
 481             VV               = _mm_add_pd(Y,_mm_mul_pd(vfeps,Fp));
 482             vvdw6            = _mm_mul_pd(c6_00,VV);
 483             FF               = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
 484             fvdw6            = _mm_mul_pd(c6_00,FF);
 485
 486             /* CUBIC SPLINE TABLE REPULSION */
 487             vfitab           = _mm_add_epi32(vfitab,ifour);
 488             Y                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
 489             F                = _mm_setzero_pd();
 490             GMX_MM_TRANSPOSE2_PD(Y,F);
 491             G                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
 492             H                = _mm_setzero_pd();
 493             GMX_MM_TRANSPOSE2_PD(G,H);
 494             Heps             = _mm_mul_pd(vfeps,H);
 495             Fp               = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
 496             VV               = _mm_add_pd(Y,_mm_mul_pd(vfeps,Fp));
 497             vvdw12           = _mm_mul_pd(c12_00,VV);
 498             FF               = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
 499             fvdw12           = _mm_mul_pd(c12_00,FF);
 500             vvdw             = _mm_add_pd(vvdw12,vvdw6);
 501             fvdw             = _mm_xor_pd(signbit,_mm_mul_pd(_mm_add_pd(fvdw6,fvdw12),_mm_mul_pd(vftabscale,rinv00)));
 502
 503             /* Update potential sum for this i atom from the interaction with this j atom. */
 504             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 505             velecsum         = _mm_add_pd(velecsum,velec);
 506             vvdw             = _mm_unpacklo_pd(vvdw,_mm_setzero_pd());
 507             vvdwsum          = _mm_add_pd(vvdwsum,vvdw);
 508
 509             fscal            = _mm_add_pd(felec,fvdw);
 510
 511             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 512
 513             /* Calculate temporary vectorial force */
 514             tx               = _mm_mul_pd(fscal,dx00);
 515             ty               = _mm_mul_pd(fscal,dy00);
 516             tz               = _mm_mul_pd(fscal,dz00);
 517
 518             /* Update vectorial force */
 519             fix0             = _mm_add_pd(fix0,tx);
 520             fiy0             = _mm_add_pd(fiy0,ty);
 521             fiz0             = _mm_add_pd(fiz0,tz);
 522
 523             fjx0             = _mm_add_pd(fjx0,tx);
 524             fjy0             = _mm_add_pd(fjy0,ty);
 525             fjz0             = _mm_add_pd(fjz0,tz);
 526
 527             /**************************
 528              * CALCULATE INTERACTIONS *
 529              **************************/
 530
 531             r10              = _mm_mul_pd(rsq10,rinv10);
 532
 533             /* Compute parameters for interactions between i and j atoms */
 534             qq10             = _mm_mul_pd(iq1,jq0);
 535
 536             /* Calculate table index by multiplying r with table scale and truncate to integer */
 537             rt               = _mm_mul_pd(r10,vftabscale);
 538             vfitab           = _mm_cvttpd_epi32(rt);
 539             vfeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
 540             vfitab           = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
 541
 542             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 543             Y                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
 544             F                = _mm_setzero_pd();
 545             GMX_MM_TRANSPOSE2_PD(Y,F);
 546             G                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
 547             H                = _mm_setzero_pd();
 548             GMX_MM_TRANSPOSE2_PD(G,H);
 549             Heps             = _mm_mul_pd(vfeps,H);
 550             Fp               = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
 551             VV               = _mm_add_pd(Y,_mm_mul_pd(vfeps,Fp));
 552             velec            = _mm_mul_pd(qq10,VV);
 553             FF               = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
 554             felec            = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq10,FF),_mm_mul_pd(vftabscale,rinv10)));
 555
 556             /* Update potential sum for this i atom from the interaction with this j atom. */
 557             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 558             velecsum         = _mm_add_pd(velecsum,velec);
 559
 560             fscal            = felec;
 561
 562             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 563
 564             /* Calculate temporary vectorial force */
 565             tx               = _mm_mul_pd(fscal,dx10);
 566             ty               = _mm_mul_pd(fscal,dy10);
 567             tz               = _mm_mul_pd(fscal,dz10);
 568
 569             /* Update vectorial force */
 570             fix1             = _mm_add_pd(fix1,tx);
 571             fiy1             = _mm_add_pd(fiy1,ty);
 572             fiz1             = _mm_add_pd(fiz1,tz);
 573
 574             fjx0             = _mm_add_pd(fjx0,tx);
 575             fjy0             = _mm_add_pd(fjy0,ty);
 576             fjz0             = _mm_add_pd(fjz0,tz);
 577
 578             /**************************
 579              * CALCULATE INTERACTIONS *
 580              **************************/
 581
 582             r20              = _mm_mul_pd(rsq20,rinv20);
 583
 584             /* Compute parameters for interactions between i and j atoms */
 585             qq20             = _mm_mul_pd(iq2,jq0);
 586
 587             /* Calculate table index by multiplying r with table scale and truncate to integer */
 588             rt               = _mm_mul_pd(r20,vftabscale);
 589             vfitab           = _mm_cvttpd_epi32(rt);
 590             vfeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
 591             vfitab           = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
 592
 593             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 594             Y                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
 595             F                = _mm_setzero_pd();
 596             GMX_MM_TRANSPOSE2_PD(Y,F);
 597             G                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
 598             H                = _mm_setzero_pd();
 599             GMX_MM_TRANSPOSE2_PD(G,H);
 600             Heps             = _mm_mul_pd(vfeps,H);
 601             Fp               = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
 602             VV               = _mm_add_pd(Y,_mm_mul_pd(vfeps,Fp));
 603             velec            = _mm_mul_pd(qq20,VV);
 604             FF               = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
 605             felec            = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq20,FF),_mm_mul_pd(vftabscale,rinv20)));
 606
 607             /* Update potential sum for this i atom from the interaction with this j atom. */
 608             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 609             velecsum         = _mm_add_pd(velecsum,velec);
 610
 611             fscal            = felec;
 612
 613             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 614
 615             /* Calculate temporary vectorial force */
 616             tx               = _mm_mul_pd(fscal,dx20);
 617             ty               = _mm_mul_pd(fscal,dy20);
 618             tz               = _mm_mul_pd(fscal,dz20);
 619
 620             /* Update vectorial force */
 621             fix2             = _mm_add_pd(fix2,tx);
 622             fiy2             = _mm_add_pd(fiy2,ty);
 623             fiz2             = _mm_add_pd(fiz2,tz);
 624
 625             fjx0             = _mm_add_pd(fjx0,tx);
 626             fjy0             = _mm_add_pd(fjy0,ty);
 627             fjz0             = _mm_add_pd(fjz0,tz);
 628
 629             gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,fjx0,fjy0,fjz0);
 630
 631             /* Inner loop uses 162 flops */
 632         }
 633
 634         /* End of innermost loop */
 635
 636         gmx_mm_update_iforce_3atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
 637                                               f+i_coord_offset,fshift+i_shift_offset);
 638
 639         ggid                        = gid[iidx];
 640         /* Update potential energies */
 641         gmx_mm_update_1pot_pd(velecsum,kernel_data->energygrp_elec+ggid);
 642         gmx_mm_update_1pot_pd(vvdwsum,kernel_data->energygrp_vdw+ggid);
 643
 644         /* Increment number of inner iterations */
 645         inneriter                  += j_index_end - j_index_start;
 646
 647         /* Outer loop uses 20 flops */
 648     }
 649
 650     /* Increment number of outer iterations */
 651     outeriter        += nri;
 652
 653     /* Update outer/inner flops */
 654
 655     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3_VF,outeriter*20 + inneriter*162);
 656 }
 657 /*
 658  * Gromacs nonbonded kernel:   nb_kernel_ElecCSTab_VdwCSTab_GeomW3P1_F_sse4_1_double
 659  * Electrostatics interaction: CubicSplineTable
 660  * VdW interaction:            CubicSplineTable
 661  * Geometry:                   Water3-Particle
 662  * Calculate force/pot:        Force
 663  */
 664 void
 665 nb_kernel_ElecCSTab_VdwCSTab_GeomW3P1_F_sse4_1_double
 666                     (t_nblist                    * gmx_restrict       nlist,
 667                      rvec                        * gmx_restrict          xx,
 668                      rvec                        * gmx_restrict          ff,
 669                      t_forcerec                  * gmx_restrict          fr,
 670                      t_mdatoms                   * gmx_restrict     mdatoms,
 671                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 672                      t_nrnb                      * gmx_restrict        nrnb)
 673 {
 674     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 675      * just 0 for non-waters.
 676      * Suffixes A,B refer to j loop unrolling done with SSE double precision, e.g. for the two different
 677      * jnr indices corresponding to data put in the four positions in the SIMD register.
 678      */
 679     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 680     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 681     int              jnrA,jnrB;
 682     int              j_coord_offsetA,j_coord_offsetB;
 683     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 684     real             rcutoff_scalar;
 685     real             *shiftvec,*fshift,*x,*f;
 686     __m128d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 687     int              vdwioffset0;
 688     __m128d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 689     int              vdwioffset1;
 690     __m128d          ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
 691     int              vdwioffset2;
 692     __m128d          ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
 693     int              vdwjidx0A,vdwjidx0B;
 694     __m128d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 695     __m128d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 696     __m128d          dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
 697     __m128d          dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
 698     __m128d          velec,felec,velecsum,facel,crf,krf,krf2;
 699     real             *charge;
 700     int              nvdwtype;
 701     __m128d          rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 702     int              *vdwtype;
 703     real             *vdwparam;
 704     __m128d          one_sixth   = _mm_set1_pd(1.0/6.0);
 705     __m128d          one_twelfth = _mm_set1_pd(1.0/12.0);
 706     __m128i          vfitab;
 707     __m128i          ifour       = _mm_set1_epi32(4);
 708     __m128d          rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF;
 709     real             *vftab;
 710     __m128d          dummy_mask,cutoff_mask;
 711     __m128d          signbit   = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
 712     __m128d          one     = _mm_set1_pd(1.0);
 713     __m128d          two     = _mm_set1_pd(2.0);
 714     x                = xx[0];
 715     f                = ff[0];
 716
 717     nri              = nlist->nri;
 718     iinr             = nlist->iinr;
 719     jindex           = nlist->jindex;
 720     jjnr             = nlist->jjnr;
 721     shiftidx         = nlist->shift;
 722     gid              = nlist->gid;
 723     shiftvec         = fr->shift_vec[0];
 724     fshift           = fr->fshift[0];
 725     facel            = _mm_set1_pd(fr->epsfac);
 726     charge           = mdatoms->chargeA;
 727     nvdwtype         = fr->ntype;
 728     vdwparam         = fr->nbfp;
 729     vdwtype          = mdatoms->typeA;
 730
 731     vftab            = kernel_data->table_elec_vdw->data;
 732     vftabscale       = _mm_set1_pd(kernel_data->table_elec_vdw->scale);
 733
 734     /* Setup water-specific parameters */
 735     inr              = nlist->iinr[0];
 736     iq0              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+0]));
 737     iq1              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+1]));
 738     iq2              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+2]));
 739     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 740
 741     /* Avoid stupid compiler warnings */
 742     jnrA = jnrB = 0;
 743     j_coord_offsetA = 0;
 744     j_coord_offsetB = 0;
 745
 746     outeriter        = 0;
 747     inneriter        = 0;
 748
 749     /* Start outer loop over neighborlists */
 750     for(iidx=0; iidx<nri; iidx++)
 751     {
 752         /* Load shift vector for this list */
 753         i_shift_offset   = DIM*shiftidx[iidx];
 754
 755         /* Load limits for loop over neighbors */
 756         j_index_start    = jindex[iidx];
 757         j_index_end      = jindex[iidx+1];
 758
 759         /* Get outer coordinate index */
 760         inr              = iinr[iidx];
 761         i_coord_offset   = DIM*inr;
 762
 763         /* Load i particle coords and add shift vector */
 764         gmx_mm_load_shift_and_3rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
 765                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 766
 767         fix0             = _mm_setzero_pd();
 768         fiy0             = _mm_setzero_pd();
 769         fiz0             = _mm_setzero_pd();
 770         fix1             = _mm_setzero_pd();
 771         fiy1             = _mm_setzero_pd();
 772         fiz1             = _mm_setzero_pd();
 773         fix2             = _mm_setzero_pd();
 774         fiy2             = _mm_setzero_pd();
 775         fiz2             = _mm_setzero_pd();
 776
 777         /* Start inner kernel loop */
 778         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 779         {
 780
 781             /* Get j neighbor index, and coordinate index */
 782             jnrA             = jjnr[jidx];
 783             jnrB             = jjnr[jidx+1];
 784             j_coord_offsetA  = DIM*jnrA;
 785             j_coord_offsetB  = DIM*jnrB;
 786
 787             /* load j atom coordinates */
 788             gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 789                                               &jx0,&jy0,&jz0);
 790
 791             /* Calculate displacement vector */
 792             dx00             = _mm_sub_pd(ix0,jx0);
 793             dy00             = _mm_sub_pd(iy0,jy0);
 794             dz00             = _mm_sub_pd(iz0,jz0);
 795             dx10             = _mm_sub_pd(ix1,jx0);
 796             dy10             = _mm_sub_pd(iy1,jy0);
 797             dz10             = _mm_sub_pd(iz1,jz0);
 798             dx20             = _mm_sub_pd(ix2,jx0);
 799             dy20             = _mm_sub_pd(iy2,jy0);
 800             dz20             = _mm_sub_pd(iz2,jz0);
 801
 802             /* Calculate squared distance and things based on it */
 803             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 804             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
 805             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
 806
 807             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 808             rinv10           = gmx_mm_invsqrt_pd(rsq10);
 809             rinv20           = gmx_mm_invsqrt_pd(rsq20);
 810
 811             /* Load parameters for j particles */
 812             jq0              = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
 813             vdwjidx0A        = 2*vdwtype[jnrA+0];
 814             vdwjidx0B        = 2*vdwtype[jnrB+0];
 815
 816             fjx0             = _mm_setzero_pd();
 817             fjy0             = _mm_setzero_pd();
 818             fjz0             = _mm_setzero_pd();
 819
 820             /**************************
 821              * CALCULATE INTERACTIONS *
 822              **************************/
 823
 824             r00              = _mm_mul_pd(rsq00,rinv00);
 825
 826             /* Compute parameters for interactions between i and j atoms */
 827             qq00             = _mm_mul_pd(iq0,jq0);
 828             gmx_mm_load_2pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,
 829                                          vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
 830
 831             /* Calculate table index by multiplying r with table scale and truncate to integer */
 832             rt               = _mm_mul_pd(r00,vftabscale);
 833             vfitab           = _mm_cvttpd_epi32(rt);
 834             vfeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
 835             vfitab           = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
 836
 837             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 838             Y                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
 839             F                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) );
 840             GMX_MM_TRANSPOSE2_PD(Y,F);
 841             G                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
 842             H                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) +2);
 843             GMX_MM_TRANSPOSE2_PD(G,H);
 844             Heps             = _mm_mul_pd(vfeps,H);
 845             Fp               = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
 846             FF               = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
 847             felec            = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq00,FF),_mm_mul_pd(vftabscale,rinv00)));
 848
 849             /* CUBIC SPLINE TABLE DISPERSION */
 850             vfitab           = _mm_add_epi32(vfitab,ifour);
 851             Y                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
 852             F                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) );
 853             GMX_MM_TRANSPOSE2_PD(Y,F);
 854             G                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
 855             H                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) +2);
 856             GMX_MM_TRANSPOSE2_PD(G,H);
 857             Heps             = _mm_mul_pd(vfeps,H);
 858             Fp               = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
 859             FF               = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
 860             fvdw6            = _mm_mul_pd(c6_00,FF);
 861
 862             /* CUBIC SPLINE TABLE REPULSION */
 863             vfitab           = _mm_add_epi32(vfitab,ifour);
 864             Y                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
 865             F                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) );
 866             GMX_MM_TRANSPOSE2_PD(Y,F);
 867             G                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
 868             H                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) +2);
 869             GMX_MM_TRANSPOSE2_PD(G,H);
 870             Heps             = _mm_mul_pd(vfeps,H);
 871             Fp               = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
 872             FF               = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
 873             fvdw12           = _mm_mul_pd(c12_00,FF);
 874             fvdw             = _mm_xor_pd(signbit,_mm_mul_pd(_mm_add_pd(fvdw6,fvdw12),_mm_mul_pd(vftabscale,rinv00)));
 875
 876             fscal            = _mm_add_pd(felec,fvdw);
 877
 878             /* Calculate temporary vectorial force */
 879             tx               = _mm_mul_pd(fscal,dx00);
 880             ty               = _mm_mul_pd(fscal,dy00);
 881             tz               = _mm_mul_pd(fscal,dz00);
 882
 883             /* Update vectorial force */
 884             fix0             = _mm_add_pd(fix0,tx);
 885             fiy0             = _mm_add_pd(fiy0,ty);
 886             fiz0             = _mm_add_pd(fiz0,tz);
 887
 888             fjx0             = _mm_add_pd(fjx0,tx);
 889             fjy0             = _mm_add_pd(fjy0,ty);
 890             fjz0             = _mm_add_pd(fjz0,tz);
 891
 892             /**************************
 893              * CALCULATE INTERACTIONS *
 894              **************************/
 895
 896             r10              = _mm_mul_pd(rsq10,rinv10);
 897
 898             /* Compute parameters for interactions between i and j atoms */
 899             qq10             = _mm_mul_pd(iq1,jq0);
 900
 901             /* Calculate table index by multiplying r with table scale and truncate to integer */
 902             rt               = _mm_mul_pd(r10,vftabscale);
 903             vfitab           = _mm_cvttpd_epi32(rt);
 904             vfeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
 905             vfitab           = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
 906
 907             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 908             Y                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
 909             F                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) );
 910             GMX_MM_TRANSPOSE2_PD(Y,F);
 911             G                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
 912             H                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) +2);
 913             GMX_MM_TRANSPOSE2_PD(G,H);
 914             Heps             = _mm_mul_pd(vfeps,H);
 915             Fp               = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
 916             FF               = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
 917             felec            = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq10,FF),_mm_mul_pd(vftabscale,rinv10)));
 918
 919             fscal            = felec;
 920
 921             /* Calculate temporary vectorial force */
 922             tx               = _mm_mul_pd(fscal,dx10);
 923             ty               = _mm_mul_pd(fscal,dy10);
 924             tz               = _mm_mul_pd(fscal,dz10);
 925
 926             /* Update vectorial force */
 927             fix1             = _mm_add_pd(fix1,tx);
 928             fiy1             = _mm_add_pd(fiy1,ty);
 929             fiz1             = _mm_add_pd(fiz1,tz);
 930
 931             fjx0             = _mm_add_pd(fjx0,tx);
 932             fjy0             = _mm_add_pd(fjy0,ty);
 933             fjz0             = _mm_add_pd(fjz0,tz);
 934
 935             /**************************
 936              * CALCULATE INTERACTIONS *
 937              **************************/
 938
 939             r20              = _mm_mul_pd(rsq20,rinv20);
 940
 941             /* Compute parameters for interactions between i and j atoms */
 942             qq20             = _mm_mul_pd(iq2,jq0);
 943
 944             /* Calculate table index by multiplying r with table scale and truncate to integer */
 945             rt               = _mm_mul_pd(r20,vftabscale);
 946             vfitab           = _mm_cvttpd_epi32(rt);
 947             vfeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
 948             vfitab           = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
 949
 950             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 951             Y                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
 952             F                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) );
 953             GMX_MM_TRANSPOSE2_PD(Y,F);
 954             G                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
 955             H                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) +2);
 956             GMX_MM_TRANSPOSE2_PD(G,H);
 957             Heps             = _mm_mul_pd(vfeps,H);
 958             Fp               = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
 959             FF               = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
 960             felec            = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq20,FF),_mm_mul_pd(vftabscale,rinv20)));
 961
 962             fscal            = felec;
 963
 964             /* Calculate temporary vectorial force */
 965             tx               = _mm_mul_pd(fscal,dx20);
 966             ty               = _mm_mul_pd(fscal,dy20);
 967             tz               = _mm_mul_pd(fscal,dz20);
 968
 969             /* Update vectorial force */
 970             fix2             = _mm_add_pd(fix2,tx);
 971             fiy2             = _mm_add_pd(fiy2,ty);
 972             fiz2             = _mm_add_pd(fiz2,tz);
 973
 974             fjx0             = _mm_add_pd(fjx0,tx);
 975             fjy0             = _mm_add_pd(fjy0,ty);
 976             fjz0             = _mm_add_pd(fjz0,tz);
 977
 978             gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,fjx0,fjy0,fjz0);
 979
 980             /* Inner loop uses 142 flops */
 981         }
 982
 983         if(jidx<j_index_end)
 984         {
 985
 986             jnrA             = jjnr[jidx];
 987             j_coord_offsetA  = DIM*jnrA;
 988
 989             /* load j atom coordinates */
 990             gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
 991                                               &jx0,&jy0,&jz0);
 992
 993             /* Calculate displacement vector */
 994             dx00             = _mm_sub_pd(ix0,jx0);
 995             dy00             = _mm_sub_pd(iy0,jy0);
 996             dz00             = _mm_sub_pd(iz0,jz0);
 997             dx10             = _mm_sub_pd(ix1,jx0);
 998             dy10             = _mm_sub_pd(iy1,jy0);
 999             dz10             = _mm_sub_pd(iz1,jz0);
1000             dx20             = _mm_sub_pd(ix2,jx0);
1001             dy20             = _mm_sub_pd(iy2,jy0);
1002             dz20             = _mm_sub_pd(iz2,jz0);
1003
1004             /* Calculate squared distance and things based on it */
1005             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
1006             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
1007             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
1008
1009             rinv00           = gmx_mm_invsqrt_pd(rsq00);
1010             rinv10           = gmx_mm_invsqrt_pd(rsq10);
1011             rinv20           = gmx_mm_invsqrt_pd(rsq20);
1012
1013             /* Load parameters for j particles */
1014             jq0              = _mm_load_sd(charge+jnrA+0);
1015             vdwjidx0A        = 2*vdwtype[jnrA+0];
1016
1017             fjx0             = _mm_setzero_pd();
1018             fjy0             = _mm_setzero_pd();
1019             fjz0             = _mm_setzero_pd();
1020
1021             /**************************
1022              * CALCULATE INTERACTIONS *
1023              **************************/
1024
1025             r00              = _mm_mul_pd(rsq00,rinv00);
1026
1027             /* Compute parameters for interactions between i and j atoms */
1028             qq00             = _mm_mul_pd(iq0,jq0);
1029             gmx_mm_load_1pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,&c6_00,&c12_00);
1030
1031             /* Calculate table index by multiplying r with table scale and truncate to integer */
1032             rt               = _mm_mul_pd(r00,vftabscale);
1033             vfitab           = _mm_cvttpd_epi32(rt);
1034             vfeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
1035             vfitab           = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
1036
1037             /* CUBIC SPLINE TABLE ELECTROSTATICS */
1038             Y                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
1039             F                = _mm_setzero_pd();
1040             GMX_MM_TRANSPOSE2_PD(Y,F);
1041             G                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
1042             H                = _mm_setzero_pd();
1043             GMX_MM_TRANSPOSE2_PD(G,H);
1044             Heps             = _mm_mul_pd(vfeps,H);
1045             Fp               = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
1046             FF               = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
1047             felec            = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq00,FF),_mm_mul_pd(vftabscale,rinv00)));
1048
1049             /* CUBIC SPLINE TABLE DISPERSION */
1050             vfitab           = _mm_add_epi32(vfitab,ifour);
1051             Y                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
1052             F                = _mm_setzero_pd();
1053             GMX_MM_TRANSPOSE2_PD(Y,F);
1054             G                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
1055             H                = _mm_setzero_pd();
1056             GMX_MM_TRANSPOSE2_PD(G,H);
1057             Heps             = _mm_mul_pd(vfeps,H);
1058             Fp               = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
1059             FF               = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
1060             fvdw6            = _mm_mul_pd(c6_00,FF);
1061
1062             /* CUBIC SPLINE TABLE REPULSION */
1063             vfitab           = _mm_add_epi32(vfitab,ifour);
1064             Y                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
1065             F                = _mm_setzero_pd();
1066             GMX_MM_TRANSPOSE2_PD(Y,F);
1067             G                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
1068             H                = _mm_setzero_pd();
1069             GMX_MM_TRANSPOSE2_PD(G,H);
1070             Heps             = _mm_mul_pd(vfeps,H);
1071             Fp               = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
1072             FF               = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
1073             fvdw12           = _mm_mul_pd(c12_00,FF);
1074             fvdw             = _mm_xor_pd(signbit,_mm_mul_pd(_mm_add_pd(fvdw6,fvdw12),_mm_mul_pd(vftabscale,rinv00)));
1075
1076             fscal            = _mm_add_pd(felec,fvdw);
1077
1078             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1079
1080             /* Calculate temporary vectorial force */
1081             tx               = _mm_mul_pd(fscal,dx00);
1082             ty               = _mm_mul_pd(fscal,dy00);
1083             tz               = _mm_mul_pd(fscal,dz00);
1084
1085             /* Update vectorial force */
1086             fix0             = _mm_add_pd(fix0,tx);
1087             fiy0             = _mm_add_pd(fiy0,ty);
1088             fiz0             = _mm_add_pd(fiz0,tz);
1089
1090             fjx0             = _mm_add_pd(fjx0,tx);
1091             fjy0             = _mm_add_pd(fjy0,ty);
1092             fjz0             = _mm_add_pd(fjz0,tz);
1093
1094             /**************************
1095              * CALCULATE INTERACTIONS *
1096              **************************/
1097
1098             r10              = _mm_mul_pd(rsq10,rinv10);
1099
1100             /* Compute parameters for interactions between i and j atoms */
1101             qq10             = _mm_mul_pd(iq1,jq0);
1102
1103             /* Calculate table index by multiplying r with table scale and truncate to integer */
1104             rt               = _mm_mul_pd(r10,vftabscale);
1105             vfitab           = _mm_cvttpd_epi32(rt);
1106             vfeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
1107             vfitab           = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
1108
1109             /* CUBIC SPLINE TABLE ELECTROSTATICS */
1110             Y                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
1111             F                = _mm_setzero_pd();
1112             GMX_MM_TRANSPOSE2_PD(Y,F);
1113             G                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
1114             H                = _mm_setzero_pd();
1115             GMX_MM_TRANSPOSE2_PD(G,H);
1116             Heps             = _mm_mul_pd(vfeps,H);
1117             Fp               = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
1118             FF               = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
1119             felec            = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq10,FF),_mm_mul_pd(vftabscale,rinv10)));
1120
1121             fscal            = felec;
1122
1123             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1124
1125             /* Calculate temporary vectorial force */
1126             tx               = _mm_mul_pd(fscal,dx10);
1127             ty               = _mm_mul_pd(fscal,dy10);
1128             tz               = _mm_mul_pd(fscal,dz10);
1129
1130             /* Update vectorial force */
1131             fix1             = _mm_add_pd(fix1,tx);
1132             fiy1             = _mm_add_pd(fiy1,ty);
1133             fiz1             = _mm_add_pd(fiz1,tz);
1134
1135             fjx0             = _mm_add_pd(fjx0,tx);
1136             fjy0             = _mm_add_pd(fjy0,ty);
1137             fjz0             = _mm_add_pd(fjz0,tz);
1138
1139             /**************************
1140              * CALCULATE INTERACTIONS *
1141              **************************/
1142
1143             r20              = _mm_mul_pd(rsq20,rinv20);
1144
1145             /* Compute parameters for interactions between i and j atoms */
1146             qq20             = _mm_mul_pd(iq2,jq0);
1147
1148             /* Calculate table index by multiplying r with table scale and truncate to integer */
1149             rt               = _mm_mul_pd(r20,vftabscale);
1150             vfitab           = _mm_cvttpd_epi32(rt);
1151             vfeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
1152             vfitab           = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
1153
1154             /* CUBIC SPLINE TABLE ELECTROSTATICS */
1155             Y                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
1156             F                = _mm_setzero_pd();
1157             GMX_MM_TRANSPOSE2_PD(Y,F);
1158             G                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
1159             H                = _mm_setzero_pd();
1160             GMX_MM_TRANSPOSE2_PD(G,H);
1161             Heps             = _mm_mul_pd(vfeps,H);
1162             Fp               = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
1163             FF               = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
1164             felec            = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq20,FF),_mm_mul_pd(vftabscale,rinv20)));
1165
1166             fscal            = felec;
1167
1168             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1169
1170             /* Calculate temporary vectorial force */
1171             tx               = _mm_mul_pd(fscal,dx20);
1172             ty               = _mm_mul_pd(fscal,dy20);
1173             tz               = _mm_mul_pd(fscal,dz20);
1174
1175             /* Update vectorial force */
1176             fix2             = _mm_add_pd(fix2,tx);
1177             fiy2             = _mm_add_pd(fiy2,ty);
1178             fiz2             = _mm_add_pd(fiz2,tz);
1179
1180             fjx0             = _mm_add_pd(fjx0,tx);
1181             fjy0             = _mm_add_pd(fjy0,ty);
1182             fjz0             = _mm_add_pd(fjz0,tz);
1183
1184             gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,fjx0,fjy0,fjz0);
1185
1186             /* Inner loop uses 142 flops */
1187         }
1188
1189         /* End of innermost loop */
1190
1191         gmx_mm_update_iforce_3atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
1192                                               f+i_coord_offset,fshift+i_shift_offset);
1193
1194         /* Increment number of inner iterations */
1195         inneriter                  += j_index_end - j_index_start;
1196
1197         /* Outer loop uses 18 flops */
1198     }
1199
1200     /* Increment number of outer iterations */
1201     outeriter        += nri;
1202
1203     /* Update outer/inner flops */
1204
1205     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3_F,outeriter*18 + inneriter*142);
1206 }