src/gromacs/gmxlib/nonbonded/nb_kernel_sse2_double/nb_kernel_ElecEwSw_VdwLJSw_GeomW4P1_sse2_double.c

   1 /*
   2  * This file is part of the GROMACS molecular simulation package.
   3  *
   4  * Copyright (c) 2012,2013,2014, by the GROMACS development team, led by
   5  * Mark Abraham, David van der Spoel, Berk Hess, and Erik Lindahl,
   6  * and including many others, as listed in the AUTHORS file in the
   7  * top-level source directory and at http://www.gromacs.org.
   8  *
   9  * GROMACS is free software; you can redistribute it and/or
  10  * modify it under the terms of the GNU Lesser General Public License
  11  * as published by the Free Software Foundation; either version 2.1
  12  * of the License, or (at your option) any later version.
  13  *
  14  * GROMACS is distributed in the hope that it will be useful,
  15  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  16  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  17  * Lesser General Public License for more details.
  18  *
  19  * You should have received a copy of the GNU Lesser General Public
  20  * License along with GROMACS; if not, see
  21  * http://www.gnu.org/licenses, or write to the Free Software Foundation,
  22  * Inc., 51 Franklin Street, Fifth Floor, Boston, MA  02110-1301  USA.
  23  *
  24  * If you want to redistribute modifications to GROMACS, please
  25  * consider that scientific software is very special. Version
  26  * control is crucial - bugs must be traceable. We will be happy to
  27  * consider code for inclusion in the official distribution, but
  28  * derived work must not be called official GROMACS. Details are found
  29  * in the README & COPYING files - if they are missing, get the
  30  * official version at http://www.gromacs.org.
  31  *
  32  * To help us fund GROMACS development, we humbly ask that you cite
  33  * the research papers on the package. Check out http://www.gromacs.org.
  34  */
  35 /*
  36  * Note: this file was generated by the GROMACS sse2_double kernel generator.
  37  */
  38 #include "config.h"
  39
  40 #include <math.h>
  41
  42 #include "../nb_kernel.h"
  43 #include "types/simple.h"
  44 #include "gromacs/math/vec.h"
  45 #include "nrnb.h"
  46
  47 #include "gromacs/simd/math_x86_sse2_double.h"
  48 #include "kernelutil_x86_sse2_double.h"
  49
  50 /*
  51  * Gromacs nonbonded kernel:   nb_kernel_ElecEwSw_VdwLJSw_GeomW4P1_VF_sse2_double
  52  * Electrostatics interaction: Ewald
  53  * VdW interaction:            LennardJones
  54  * Geometry:                   Water4-Particle
  55  * Calculate force/pot:        PotentialAndForce
  56  */
  57 void
  58 nb_kernel_ElecEwSw_VdwLJSw_GeomW4P1_VF_sse2_double
  59                     (t_nblist                    * gmx_restrict       nlist,
  60                      rvec                        * gmx_restrict          xx,
  61                      rvec                        * gmx_restrict          ff,
  62                      t_forcerec                  * gmx_restrict          fr,
  63                      t_mdatoms                   * gmx_restrict     mdatoms,
  64                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
  65                      t_nrnb                      * gmx_restrict        nrnb)
  66 {
  67     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
  68      * just 0 for non-waters.
  69      * Suffixes A,B refer to j loop unrolling done with SSE double precision, e.g. for the two different
  70      * jnr indices corresponding to data put in the four positions in the SIMD register.
  71      */
  72     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
  73     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
  74     int              jnrA,jnrB;
  75     int              j_coord_offsetA,j_coord_offsetB;
  76     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
  77     real             rcutoff_scalar;
  78     real             *shiftvec,*fshift,*x,*f;
  79     __m128d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
  80     int              vdwioffset0;
  81     __m128d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
  82     int              vdwioffset1;
  83     __m128d          ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
  84     int              vdwioffset2;
  85     __m128d          ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
  86     int              vdwioffset3;
  87     __m128d          ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
  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          dx30,dy30,dz30,rsq30,rinv30,rinvsq30,r30,qq30,c6_30,c12_30;
  94     __m128d          velec,felec,velecsum,facel,crf,krf,krf2;
  95     real             *charge;
  96     int              nvdwtype;
  97     __m128d          rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
  98     int              *vdwtype;
  99     real             *vdwparam;
 100     __m128d          one_sixth   = _mm_set1_pd(1.0/6.0);
 101     __m128d          one_twelfth = _mm_set1_pd(1.0/12.0);
 102     __m128i          ewitab;
 103     __m128d          ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
 104     real             *ewtab;
 105     __m128d          rswitch,swV3,swV4,swV5,swF2,swF3,swF4,d,d2,sw,dsw;
 106     real             rswitch_scalar,d_scalar;
 107     __m128d          dummy_mask,cutoff_mask;
 108     __m128d          signbit   = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
 109     __m128d          one     = _mm_set1_pd(1.0);
 110     __m128d          two     = _mm_set1_pd(2.0);
 111     x                = xx[0];
 112     f                = ff[0];
 113
 114     nri              = nlist->nri;
 115     iinr             = nlist->iinr;
 116     jindex           = nlist->jindex;
 117     jjnr             = nlist->jjnr;
 118     shiftidx         = nlist->shift;
 119     gid              = nlist->gid;
 120     shiftvec         = fr->shift_vec[0];
 121     fshift           = fr->fshift[0];
 122     facel            = _mm_set1_pd(fr->epsfac);
 123     charge           = mdatoms->chargeA;
 124     nvdwtype         = fr->ntype;
 125     vdwparam         = fr->nbfp;
 126     vdwtype          = mdatoms->typeA;
 127
 128     sh_ewald         = _mm_set1_pd(fr->ic->sh_ewald);
 129     ewtab            = fr->ic->tabq_coul_FDV0;
 130     ewtabscale       = _mm_set1_pd(fr->ic->tabq_scale);
 131     ewtabhalfspace   = _mm_set1_pd(0.5/fr->ic->tabq_scale);
 132
 133     /* Setup water-specific parameters */
 134     inr              = nlist->iinr[0];
 135     iq1              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+1]));
 136     iq2              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+2]));
 137     iq3              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+3]));
 138     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 139
 140     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 141     rcutoff_scalar   = fr->rcoulomb;
 142     rcutoff          = _mm_set1_pd(rcutoff_scalar);
 143     rcutoff2         = _mm_mul_pd(rcutoff,rcutoff);
 144
 145     rswitch_scalar   = fr->rcoulomb_switch;
 146     rswitch          = _mm_set1_pd(rswitch_scalar);
 147     /* Setup switch parameters */
 148     d_scalar         = rcutoff_scalar-rswitch_scalar;
 149     d                = _mm_set1_pd(d_scalar);
 150     swV3             = _mm_set1_pd(-10.0/(d_scalar*d_scalar*d_scalar));
 151     swV4             = _mm_set1_pd( 15.0/(d_scalar*d_scalar*d_scalar*d_scalar));
 152     swV5             = _mm_set1_pd( -6.0/(d_scalar*d_scalar*d_scalar*d_scalar*d_scalar));
 153     swF2             = _mm_set1_pd(-30.0/(d_scalar*d_scalar*d_scalar));
 154     swF3             = _mm_set1_pd( 60.0/(d_scalar*d_scalar*d_scalar*d_scalar));
 155     swF4             = _mm_set1_pd(-30.0/(d_scalar*d_scalar*d_scalar*d_scalar*d_scalar));
 156
 157     /* Avoid stupid compiler warnings */
 158     jnrA = jnrB = 0;
 159     j_coord_offsetA = 0;
 160     j_coord_offsetB = 0;
 161
 162     outeriter        = 0;
 163     inneriter        = 0;
 164
 165     /* Start outer loop over neighborlists */
 166     for(iidx=0; iidx<nri; iidx++)
 167     {
 168         /* Load shift vector for this list */
 169         i_shift_offset   = DIM*shiftidx[iidx];
 170
 171         /* Load limits for loop over neighbors */
 172         j_index_start    = jindex[iidx];
 173         j_index_end      = jindex[iidx+1];
 174
 175         /* Get outer coordinate index */
 176         inr              = iinr[iidx];
 177         i_coord_offset   = DIM*inr;
 178
 179         /* Load i particle coords and add shift vector */
 180         gmx_mm_load_shift_and_4rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
 181                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
 182
 183         fix0             = _mm_setzero_pd();
 184         fiy0             = _mm_setzero_pd();
 185         fiz0             = _mm_setzero_pd();
 186         fix1             = _mm_setzero_pd();
 187         fiy1             = _mm_setzero_pd();
 188         fiz1             = _mm_setzero_pd();
 189         fix2             = _mm_setzero_pd();
 190         fiy2             = _mm_setzero_pd();
 191         fiz2             = _mm_setzero_pd();
 192         fix3             = _mm_setzero_pd();
 193         fiy3             = _mm_setzero_pd();
 194         fiz3             = _mm_setzero_pd();
 195
 196         /* Reset potential sums */
 197         velecsum         = _mm_setzero_pd();
 198         vvdwsum          = _mm_setzero_pd();
 199
 200         /* Start inner kernel loop */
 201         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 202         {
 203
 204             /* Get j neighbor index, and coordinate index */
 205             jnrA             = jjnr[jidx];
 206             jnrB             = jjnr[jidx+1];
 207             j_coord_offsetA  = DIM*jnrA;
 208             j_coord_offsetB  = DIM*jnrB;
 209
 210             /* load j atom coordinates */
 211             gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 212                                               &jx0,&jy0,&jz0);
 213
 214             /* Calculate displacement vector */
 215             dx00             = _mm_sub_pd(ix0,jx0);
 216             dy00             = _mm_sub_pd(iy0,jy0);
 217             dz00             = _mm_sub_pd(iz0,jz0);
 218             dx10             = _mm_sub_pd(ix1,jx0);
 219             dy10             = _mm_sub_pd(iy1,jy0);
 220             dz10             = _mm_sub_pd(iz1,jz0);
 221             dx20             = _mm_sub_pd(ix2,jx0);
 222             dy20             = _mm_sub_pd(iy2,jy0);
 223             dz20             = _mm_sub_pd(iz2,jz0);
 224             dx30             = _mm_sub_pd(ix3,jx0);
 225             dy30             = _mm_sub_pd(iy3,jy0);
 226             dz30             = _mm_sub_pd(iz3,jz0);
 227
 228             /* Calculate squared distance and things based on it */
 229             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 230             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
 231             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
 232             rsq30            = gmx_mm_calc_rsq_pd(dx30,dy30,dz30);
 233
 234             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 235             rinv10           = gmx_mm_invsqrt_pd(rsq10);
 236             rinv20           = gmx_mm_invsqrt_pd(rsq20);
 237             rinv30           = gmx_mm_invsqrt_pd(rsq30);
 238
 239             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 240             rinvsq10         = _mm_mul_pd(rinv10,rinv10);
 241             rinvsq20         = _mm_mul_pd(rinv20,rinv20);
 242             rinvsq30         = _mm_mul_pd(rinv30,rinv30);
 243
 244             /* Load parameters for j particles */
 245             jq0              = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
 246             vdwjidx0A        = 2*vdwtype[jnrA+0];
 247             vdwjidx0B        = 2*vdwtype[jnrB+0];
 248
 249             fjx0             = _mm_setzero_pd();
 250             fjy0             = _mm_setzero_pd();
 251             fjz0             = _mm_setzero_pd();
 252
 253             /**************************
 254              * CALCULATE INTERACTIONS *
 255              **************************/
 256
 257             if (gmx_mm_any_lt(rsq00,rcutoff2))
 258             {
 259
 260             r00              = _mm_mul_pd(rsq00,rinv00);
 261
 262             /* Compute parameters for interactions between i and j atoms */
 263             gmx_mm_load_2pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,
 264                                          vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
 265
 266             /* LENNARD-JONES DISPERSION/REPULSION */
 267
 268             rinvsix          = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 269             vvdw6            = _mm_mul_pd(c6_00,rinvsix);
 270             vvdw12           = _mm_mul_pd(c12_00,_mm_mul_pd(rinvsix,rinvsix));
 271             vvdw             = _mm_sub_pd( _mm_mul_pd(vvdw12,one_twelfth) , _mm_mul_pd(vvdw6,one_sixth) );
 272             fvdw             = _mm_mul_pd(_mm_sub_pd(vvdw12,vvdw6),rinvsq00);
 273
 274             d                = _mm_sub_pd(r00,rswitch);
 275             d                = _mm_max_pd(d,_mm_setzero_pd());
 276             d2               = _mm_mul_pd(d,d);
 277             sw               = _mm_add_pd(one,_mm_mul_pd(d2,_mm_mul_pd(d,_mm_add_pd(swV3,_mm_mul_pd(d,_mm_add_pd(swV4,_mm_mul_pd(d,swV5)))))));
 278
 279             dsw              = _mm_mul_pd(d2,_mm_add_pd(swF2,_mm_mul_pd(d,_mm_add_pd(swF3,_mm_mul_pd(d,swF4)))));
 280
 281             /* Evaluate switch function */
 282             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
 283             fvdw             = _mm_sub_pd( _mm_mul_pd(fvdw,sw) , _mm_mul_pd(rinv00,_mm_mul_pd(vvdw,dsw)) );
 284             vvdw             = _mm_mul_pd(vvdw,sw);
 285             cutoff_mask      = _mm_cmplt_pd(rsq00,rcutoff2);
 286
 287             /* Update potential sum for this i atom from the interaction with this j atom. */
 288             vvdw             = _mm_and_pd(vvdw,cutoff_mask);
 289             vvdwsum          = _mm_add_pd(vvdwsum,vvdw);
 290
 291             fscal            = fvdw;
 292
 293             fscal            = _mm_and_pd(fscal,cutoff_mask);
 294
 295             /* Calculate temporary vectorial force */
 296             tx               = _mm_mul_pd(fscal,dx00);
 297             ty               = _mm_mul_pd(fscal,dy00);
 298             tz               = _mm_mul_pd(fscal,dz00);
 299
 300             /* Update vectorial force */
 301             fix0             = _mm_add_pd(fix0,tx);
 302             fiy0             = _mm_add_pd(fiy0,ty);
 303             fiz0             = _mm_add_pd(fiz0,tz);
 304
 305             fjx0             = _mm_add_pd(fjx0,tx);
 306             fjy0             = _mm_add_pd(fjy0,ty);
 307             fjz0             = _mm_add_pd(fjz0,tz);
 308
 309             }
 310
 311             /**************************
 312              * CALCULATE INTERACTIONS *
 313              **************************/
 314
 315             if (gmx_mm_any_lt(rsq10,rcutoff2))
 316             {
 317
 318             r10              = _mm_mul_pd(rsq10,rinv10);
 319
 320             /* Compute parameters for interactions between i and j atoms */
 321             qq10             = _mm_mul_pd(iq1,jq0);
 322
 323             /* EWALD ELECTROSTATICS */
 324
 325             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 326             ewrt             = _mm_mul_pd(r10,ewtabscale);
 327             ewitab           = _mm_cvttpd_epi32(ewrt);
 328             eweps            = _mm_sub_pd(ewrt,_mm_cvtepi32_pd(ewitab));
 329             ewitab           = _mm_slli_epi32(ewitab,2);
 330             ewtabF           = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 331             ewtabD           = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,1) );
 332             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 333             ewtabV           = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,0) +2);
 334             ewtabFn          = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,1) +2);
 335             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 336             felec            = _mm_add_pd(ewtabF,_mm_mul_pd(eweps,ewtabD));
 337             velec            = _mm_sub_pd(ewtabV,_mm_mul_pd(_mm_mul_pd(ewtabhalfspace,eweps),_mm_add_pd(ewtabF,felec)));
 338             velec            = _mm_mul_pd(qq10,_mm_sub_pd(rinv10,velec));
 339             felec            = _mm_mul_pd(_mm_mul_pd(qq10,rinv10),_mm_sub_pd(rinvsq10,felec));
 340
 341             d                = _mm_sub_pd(r10,rswitch);
 342             d                = _mm_max_pd(d,_mm_setzero_pd());
 343             d2               = _mm_mul_pd(d,d);
 344             sw               = _mm_add_pd(one,_mm_mul_pd(d2,_mm_mul_pd(d,_mm_add_pd(swV3,_mm_mul_pd(d,_mm_add_pd(swV4,_mm_mul_pd(d,swV5)))))));
 345
 346             dsw              = _mm_mul_pd(d2,_mm_add_pd(swF2,_mm_mul_pd(d,_mm_add_pd(swF3,_mm_mul_pd(d,swF4)))));
 347
 348             /* Evaluate switch function */
 349             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
 350             felec            = _mm_sub_pd( _mm_mul_pd(felec,sw) , _mm_mul_pd(rinv10,_mm_mul_pd(velec,dsw)) );
 351             velec            = _mm_mul_pd(velec,sw);
 352             cutoff_mask      = _mm_cmplt_pd(rsq10,rcutoff2);
 353
 354             /* Update potential sum for this i atom from the interaction with this j atom. */
 355             velec            = _mm_and_pd(velec,cutoff_mask);
 356             velecsum         = _mm_add_pd(velecsum,velec);
 357
 358             fscal            = felec;
 359
 360             fscal            = _mm_and_pd(fscal,cutoff_mask);
 361
 362             /* Calculate temporary vectorial force */
 363             tx               = _mm_mul_pd(fscal,dx10);
 364             ty               = _mm_mul_pd(fscal,dy10);
 365             tz               = _mm_mul_pd(fscal,dz10);
 366
 367             /* Update vectorial force */
 368             fix1             = _mm_add_pd(fix1,tx);
 369             fiy1             = _mm_add_pd(fiy1,ty);
 370             fiz1             = _mm_add_pd(fiz1,tz);
 371
 372             fjx0             = _mm_add_pd(fjx0,tx);
 373             fjy0             = _mm_add_pd(fjy0,ty);
 374             fjz0             = _mm_add_pd(fjz0,tz);
 375
 376             }
 377
 378             /**************************
 379              * CALCULATE INTERACTIONS *
 380              **************************/
 381
 382             if (gmx_mm_any_lt(rsq20,rcutoff2))
 383             {
 384
 385             r20              = _mm_mul_pd(rsq20,rinv20);
 386
 387             /* Compute parameters for interactions between i and j atoms */
 388             qq20             = _mm_mul_pd(iq2,jq0);
 389
 390             /* EWALD ELECTROSTATICS */
 391
 392             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 393             ewrt             = _mm_mul_pd(r20,ewtabscale);
 394             ewitab           = _mm_cvttpd_epi32(ewrt);
 395             eweps            = _mm_sub_pd(ewrt,_mm_cvtepi32_pd(ewitab));
 396             ewitab           = _mm_slli_epi32(ewitab,2);
 397             ewtabF           = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 398             ewtabD           = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,1) );
 399             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 400             ewtabV           = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,0) +2);
 401             ewtabFn          = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,1) +2);
 402             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 403             felec            = _mm_add_pd(ewtabF,_mm_mul_pd(eweps,ewtabD));
 404             velec            = _mm_sub_pd(ewtabV,_mm_mul_pd(_mm_mul_pd(ewtabhalfspace,eweps),_mm_add_pd(ewtabF,felec)));
 405             velec            = _mm_mul_pd(qq20,_mm_sub_pd(rinv20,velec));
 406             felec            = _mm_mul_pd(_mm_mul_pd(qq20,rinv20),_mm_sub_pd(rinvsq20,felec));
 407
 408             d                = _mm_sub_pd(r20,rswitch);
 409             d                = _mm_max_pd(d,_mm_setzero_pd());
 410             d2               = _mm_mul_pd(d,d);
 411             sw               = _mm_add_pd(one,_mm_mul_pd(d2,_mm_mul_pd(d,_mm_add_pd(swV3,_mm_mul_pd(d,_mm_add_pd(swV4,_mm_mul_pd(d,swV5)))))));
 412
 413             dsw              = _mm_mul_pd(d2,_mm_add_pd(swF2,_mm_mul_pd(d,_mm_add_pd(swF3,_mm_mul_pd(d,swF4)))));
 414
 415             /* Evaluate switch function */
 416             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
 417             felec            = _mm_sub_pd( _mm_mul_pd(felec,sw) , _mm_mul_pd(rinv20,_mm_mul_pd(velec,dsw)) );
 418             velec            = _mm_mul_pd(velec,sw);
 419             cutoff_mask      = _mm_cmplt_pd(rsq20,rcutoff2);
 420
 421             /* Update potential sum for this i atom from the interaction with this j atom. */
 422             velec            = _mm_and_pd(velec,cutoff_mask);
 423             velecsum         = _mm_add_pd(velecsum,velec);
 424
 425             fscal            = felec;
 426
 427             fscal            = _mm_and_pd(fscal,cutoff_mask);
 428
 429             /* Calculate temporary vectorial force */
 430             tx               = _mm_mul_pd(fscal,dx20);
 431             ty               = _mm_mul_pd(fscal,dy20);
 432             tz               = _mm_mul_pd(fscal,dz20);
 433
 434             /* Update vectorial force */
 435             fix2             = _mm_add_pd(fix2,tx);
 436             fiy2             = _mm_add_pd(fiy2,ty);
 437             fiz2             = _mm_add_pd(fiz2,tz);
 438
 439             fjx0             = _mm_add_pd(fjx0,tx);
 440             fjy0             = _mm_add_pd(fjy0,ty);
 441             fjz0             = _mm_add_pd(fjz0,tz);
 442
 443             }
 444
 445             /**************************
 446              * CALCULATE INTERACTIONS *
 447              **************************/
 448
 449             if (gmx_mm_any_lt(rsq30,rcutoff2))
 450             {
 451
 452             r30              = _mm_mul_pd(rsq30,rinv30);
 453
 454             /* Compute parameters for interactions between i and j atoms */
 455             qq30             = _mm_mul_pd(iq3,jq0);
 456
 457             /* EWALD ELECTROSTATICS */
 458
 459             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 460             ewrt             = _mm_mul_pd(r30,ewtabscale);
 461             ewitab           = _mm_cvttpd_epi32(ewrt);
 462             eweps            = _mm_sub_pd(ewrt,_mm_cvtepi32_pd(ewitab));
 463             ewitab           = _mm_slli_epi32(ewitab,2);
 464             ewtabF           = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 465             ewtabD           = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,1) );
 466             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 467             ewtabV           = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,0) +2);
 468             ewtabFn          = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,1) +2);
 469             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 470             felec            = _mm_add_pd(ewtabF,_mm_mul_pd(eweps,ewtabD));
 471             velec            = _mm_sub_pd(ewtabV,_mm_mul_pd(_mm_mul_pd(ewtabhalfspace,eweps),_mm_add_pd(ewtabF,felec)));
 472             velec            = _mm_mul_pd(qq30,_mm_sub_pd(rinv30,velec));
 473             felec            = _mm_mul_pd(_mm_mul_pd(qq30,rinv30),_mm_sub_pd(rinvsq30,felec));
 474
 475             d                = _mm_sub_pd(r30,rswitch);
 476             d                = _mm_max_pd(d,_mm_setzero_pd());
 477             d2               = _mm_mul_pd(d,d);
 478             sw               = _mm_add_pd(one,_mm_mul_pd(d2,_mm_mul_pd(d,_mm_add_pd(swV3,_mm_mul_pd(d,_mm_add_pd(swV4,_mm_mul_pd(d,swV5)))))));
 479
 480             dsw              = _mm_mul_pd(d2,_mm_add_pd(swF2,_mm_mul_pd(d,_mm_add_pd(swF3,_mm_mul_pd(d,swF4)))));
 481
 482             /* Evaluate switch function */
 483             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
 484             felec            = _mm_sub_pd( _mm_mul_pd(felec,sw) , _mm_mul_pd(rinv30,_mm_mul_pd(velec,dsw)) );
 485             velec            = _mm_mul_pd(velec,sw);
 486             cutoff_mask      = _mm_cmplt_pd(rsq30,rcutoff2);
 487
 488             /* Update potential sum for this i atom from the interaction with this j atom. */
 489             velec            = _mm_and_pd(velec,cutoff_mask);
 490             velecsum         = _mm_add_pd(velecsum,velec);
 491
 492             fscal            = felec;
 493
 494             fscal            = _mm_and_pd(fscal,cutoff_mask);
 495
 496             /* Calculate temporary vectorial force */
 497             tx               = _mm_mul_pd(fscal,dx30);
 498             ty               = _mm_mul_pd(fscal,dy30);
 499             tz               = _mm_mul_pd(fscal,dz30);
 500
 501             /* Update vectorial force */
 502             fix3             = _mm_add_pd(fix3,tx);
 503             fiy3             = _mm_add_pd(fiy3,ty);
 504             fiz3             = _mm_add_pd(fiz3,tz);
 505
 506             fjx0             = _mm_add_pd(fjx0,tx);
 507             fjy0             = _mm_add_pd(fjy0,ty);
 508             fjz0             = _mm_add_pd(fjz0,tz);
 509
 510             }
 511
 512             gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,fjx0,fjy0,fjz0);
 513
 514             /* Inner loop uses 257 flops */
 515         }
 516
 517         if(jidx<j_index_end)
 518         {
 519
 520             jnrA             = jjnr[jidx];
 521             j_coord_offsetA  = DIM*jnrA;
 522
 523             /* load j atom coordinates */
 524             gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
 525                                               &jx0,&jy0,&jz0);
 526
 527             /* Calculate displacement vector */
 528             dx00             = _mm_sub_pd(ix0,jx0);
 529             dy00             = _mm_sub_pd(iy0,jy0);
 530             dz00             = _mm_sub_pd(iz0,jz0);
 531             dx10             = _mm_sub_pd(ix1,jx0);
 532             dy10             = _mm_sub_pd(iy1,jy0);
 533             dz10             = _mm_sub_pd(iz1,jz0);
 534             dx20             = _mm_sub_pd(ix2,jx0);
 535             dy20             = _mm_sub_pd(iy2,jy0);
 536             dz20             = _mm_sub_pd(iz2,jz0);
 537             dx30             = _mm_sub_pd(ix3,jx0);
 538             dy30             = _mm_sub_pd(iy3,jy0);
 539             dz30             = _mm_sub_pd(iz3,jz0);
 540
 541             /* Calculate squared distance and things based on it */
 542             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 543             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
 544             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
 545             rsq30            = gmx_mm_calc_rsq_pd(dx30,dy30,dz30);
 546
 547             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 548             rinv10           = gmx_mm_invsqrt_pd(rsq10);
 549             rinv20           = gmx_mm_invsqrt_pd(rsq20);
 550             rinv30           = gmx_mm_invsqrt_pd(rsq30);
 551
 552             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 553             rinvsq10         = _mm_mul_pd(rinv10,rinv10);
 554             rinvsq20         = _mm_mul_pd(rinv20,rinv20);
 555             rinvsq30         = _mm_mul_pd(rinv30,rinv30);
 556
 557             /* Load parameters for j particles */
 558             jq0              = _mm_load_sd(charge+jnrA+0);
 559             vdwjidx0A        = 2*vdwtype[jnrA+0];
 560
 561             fjx0             = _mm_setzero_pd();
 562             fjy0             = _mm_setzero_pd();
 563             fjz0             = _mm_setzero_pd();
 564
 565             /**************************
 566              * CALCULATE INTERACTIONS *
 567              **************************/
 568
 569             if (gmx_mm_any_lt(rsq00,rcutoff2))
 570             {
 571
 572             r00              = _mm_mul_pd(rsq00,rinv00);
 573
 574             /* Compute parameters for interactions between i and j atoms */
 575             gmx_mm_load_1pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,&c6_00,&c12_00);
 576
 577             /* LENNARD-JONES DISPERSION/REPULSION */
 578
 579             rinvsix          = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 580             vvdw6            = _mm_mul_pd(c6_00,rinvsix);
 581             vvdw12           = _mm_mul_pd(c12_00,_mm_mul_pd(rinvsix,rinvsix));
 582             vvdw             = _mm_sub_pd( _mm_mul_pd(vvdw12,one_twelfth) , _mm_mul_pd(vvdw6,one_sixth) );
 583             fvdw             = _mm_mul_pd(_mm_sub_pd(vvdw12,vvdw6),rinvsq00);
 584
 585             d                = _mm_sub_pd(r00,rswitch);
 586             d                = _mm_max_pd(d,_mm_setzero_pd());
 587             d2               = _mm_mul_pd(d,d);
 588             sw               = _mm_add_pd(one,_mm_mul_pd(d2,_mm_mul_pd(d,_mm_add_pd(swV3,_mm_mul_pd(d,_mm_add_pd(swV4,_mm_mul_pd(d,swV5)))))));
 589
 590             dsw              = _mm_mul_pd(d2,_mm_add_pd(swF2,_mm_mul_pd(d,_mm_add_pd(swF3,_mm_mul_pd(d,swF4)))));
 591
 592             /* Evaluate switch function */
 593             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
 594             fvdw             = _mm_sub_pd( _mm_mul_pd(fvdw,sw) , _mm_mul_pd(rinv00,_mm_mul_pd(vvdw,dsw)) );
 595             vvdw             = _mm_mul_pd(vvdw,sw);
 596             cutoff_mask      = _mm_cmplt_pd(rsq00,rcutoff2);
 597
 598             /* Update potential sum for this i atom from the interaction with this j atom. */
 599             vvdw             = _mm_and_pd(vvdw,cutoff_mask);
 600             vvdw             = _mm_unpacklo_pd(vvdw,_mm_setzero_pd());
 601             vvdwsum          = _mm_add_pd(vvdwsum,vvdw);
 602
 603             fscal            = fvdw;
 604
 605             fscal            = _mm_and_pd(fscal,cutoff_mask);
 606
 607             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 608
 609             /* Calculate temporary vectorial force */
 610             tx               = _mm_mul_pd(fscal,dx00);
 611             ty               = _mm_mul_pd(fscal,dy00);
 612             tz               = _mm_mul_pd(fscal,dz00);
 613
 614             /* Update vectorial force */
 615             fix0             = _mm_add_pd(fix0,tx);
 616             fiy0             = _mm_add_pd(fiy0,ty);
 617             fiz0             = _mm_add_pd(fiz0,tz);
 618
 619             fjx0             = _mm_add_pd(fjx0,tx);
 620             fjy0             = _mm_add_pd(fjy0,ty);
 621             fjz0             = _mm_add_pd(fjz0,tz);
 622
 623             }
 624
 625             /**************************
 626              * CALCULATE INTERACTIONS *
 627              **************************/
 628
 629             if (gmx_mm_any_lt(rsq10,rcutoff2))
 630             {
 631
 632             r10              = _mm_mul_pd(rsq10,rinv10);
 633
 634             /* Compute parameters for interactions between i and j atoms */
 635             qq10             = _mm_mul_pd(iq1,jq0);
 636
 637             /* EWALD ELECTROSTATICS */
 638
 639             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 640             ewrt             = _mm_mul_pd(r10,ewtabscale);
 641             ewitab           = _mm_cvttpd_epi32(ewrt);
 642             eweps            = _mm_sub_pd(ewrt,_mm_cvtepi32_pd(ewitab));
 643             ewitab           = _mm_slli_epi32(ewitab,2);
 644             ewtabF           = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 645             ewtabD           = _mm_setzero_pd();
 646             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 647             ewtabV           = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,0) +2);
 648             ewtabFn          = _mm_setzero_pd();
 649             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 650             felec            = _mm_add_pd(ewtabF,_mm_mul_pd(eweps,ewtabD));
 651             velec            = _mm_sub_pd(ewtabV,_mm_mul_pd(_mm_mul_pd(ewtabhalfspace,eweps),_mm_add_pd(ewtabF,felec)));
 652             velec            = _mm_mul_pd(qq10,_mm_sub_pd(rinv10,velec));
 653             felec            = _mm_mul_pd(_mm_mul_pd(qq10,rinv10),_mm_sub_pd(rinvsq10,felec));
 654
 655             d                = _mm_sub_pd(r10,rswitch);
 656             d                = _mm_max_pd(d,_mm_setzero_pd());
 657             d2               = _mm_mul_pd(d,d);
 658             sw               = _mm_add_pd(one,_mm_mul_pd(d2,_mm_mul_pd(d,_mm_add_pd(swV3,_mm_mul_pd(d,_mm_add_pd(swV4,_mm_mul_pd(d,swV5)))))));
 659
 660             dsw              = _mm_mul_pd(d2,_mm_add_pd(swF2,_mm_mul_pd(d,_mm_add_pd(swF3,_mm_mul_pd(d,swF4)))));
 661
 662             /* Evaluate switch function */
 663             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
 664             felec            = _mm_sub_pd( _mm_mul_pd(felec,sw) , _mm_mul_pd(rinv10,_mm_mul_pd(velec,dsw)) );
 665             velec            = _mm_mul_pd(velec,sw);
 666             cutoff_mask      = _mm_cmplt_pd(rsq10,rcutoff2);
 667
 668             /* Update potential sum for this i atom from the interaction with this j atom. */
 669             velec            = _mm_and_pd(velec,cutoff_mask);
 670             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 671             velecsum         = _mm_add_pd(velecsum,velec);
 672
 673             fscal            = felec;
 674
 675             fscal            = _mm_and_pd(fscal,cutoff_mask);
 676
 677             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 678
 679             /* Calculate temporary vectorial force */
 680             tx               = _mm_mul_pd(fscal,dx10);
 681             ty               = _mm_mul_pd(fscal,dy10);
 682             tz               = _mm_mul_pd(fscal,dz10);
 683
 684             /* Update vectorial force */
 685             fix1             = _mm_add_pd(fix1,tx);
 686             fiy1             = _mm_add_pd(fiy1,ty);
 687             fiz1             = _mm_add_pd(fiz1,tz);
 688
 689             fjx0             = _mm_add_pd(fjx0,tx);
 690             fjy0             = _mm_add_pd(fjy0,ty);
 691             fjz0             = _mm_add_pd(fjz0,tz);
 692
 693             }
 694
 695             /**************************
 696              * CALCULATE INTERACTIONS *
 697              **************************/
 698
 699             if (gmx_mm_any_lt(rsq20,rcutoff2))
 700             {
 701
 702             r20              = _mm_mul_pd(rsq20,rinv20);
 703
 704             /* Compute parameters for interactions between i and j atoms */
 705             qq20             = _mm_mul_pd(iq2,jq0);
 706
 707             /* EWALD ELECTROSTATICS */
 708
 709             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 710             ewrt             = _mm_mul_pd(r20,ewtabscale);
 711             ewitab           = _mm_cvttpd_epi32(ewrt);
 712             eweps            = _mm_sub_pd(ewrt,_mm_cvtepi32_pd(ewitab));
 713             ewitab           = _mm_slli_epi32(ewitab,2);
 714             ewtabF           = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 715             ewtabD           = _mm_setzero_pd();
 716             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 717             ewtabV           = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,0) +2);
 718             ewtabFn          = _mm_setzero_pd();
 719             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 720             felec            = _mm_add_pd(ewtabF,_mm_mul_pd(eweps,ewtabD));
 721             velec            = _mm_sub_pd(ewtabV,_mm_mul_pd(_mm_mul_pd(ewtabhalfspace,eweps),_mm_add_pd(ewtabF,felec)));
 722             velec            = _mm_mul_pd(qq20,_mm_sub_pd(rinv20,velec));
 723             felec            = _mm_mul_pd(_mm_mul_pd(qq20,rinv20),_mm_sub_pd(rinvsq20,felec));
 724
 725             d                = _mm_sub_pd(r20,rswitch);
 726             d                = _mm_max_pd(d,_mm_setzero_pd());
 727             d2               = _mm_mul_pd(d,d);
 728             sw               = _mm_add_pd(one,_mm_mul_pd(d2,_mm_mul_pd(d,_mm_add_pd(swV3,_mm_mul_pd(d,_mm_add_pd(swV4,_mm_mul_pd(d,swV5)))))));
 729
 730             dsw              = _mm_mul_pd(d2,_mm_add_pd(swF2,_mm_mul_pd(d,_mm_add_pd(swF3,_mm_mul_pd(d,swF4)))));
 731
 732             /* Evaluate switch function */
 733             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
 734             felec            = _mm_sub_pd( _mm_mul_pd(felec,sw) , _mm_mul_pd(rinv20,_mm_mul_pd(velec,dsw)) );
 735             velec            = _mm_mul_pd(velec,sw);
 736             cutoff_mask      = _mm_cmplt_pd(rsq20,rcutoff2);
 737
 738             /* Update potential sum for this i atom from the interaction with this j atom. */
 739             velec            = _mm_and_pd(velec,cutoff_mask);
 740             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 741             velecsum         = _mm_add_pd(velecsum,velec);
 742
 743             fscal            = felec;
 744
 745             fscal            = _mm_and_pd(fscal,cutoff_mask);
 746
 747             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 748
 749             /* Calculate temporary vectorial force */
 750             tx               = _mm_mul_pd(fscal,dx20);
 751             ty               = _mm_mul_pd(fscal,dy20);
 752             tz               = _mm_mul_pd(fscal,dz20);
 753
 754             /* Update vectorial force */
 755             fix2             = _mm_add_pd(fix2,tx);
 756             fiy2             = _mm_add_pd(fiy2,ty);
 757             fiz2             = _mm_add_pd(fiz2,tz);
 758
 759             fjx0             = _mm_add_pd(fjx0,tx);
 760             fjy0             = _mm_add_pd(fjy0,ty);
 761             fjz0             = _mm_add_pd(fjz0,tz);
 762
 763             }
 764
 765             /**************************
 766              * CALCULATE INTERACTIONS *
 767              **************************/
 768
 769             if (gmx_mm_any_lt(rsq30,rcutoff2))
 770             {
 771
 772             r30              = _mm_mul_pd(rsq30,rinv30);
 773
 774             /* Compute parameters for interactions between i and j atoms */
 775             qq30             = _mm_mul_pd(iq3,jq0);
 776
 777             /* EWALD ELECTROSTATICS */
 778
 779             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 780             ewrt             = _mm_mul_pd(r30,ewtabscale);
 781             ewitab           = _mm_cvttpd_epi32(ewrt);
 782             eweps            = _mm_sub_pd(ewrt,_mm_cvtepi32_pd(ewitab));
 783             ewitab           = _mm_slli_epi32(ewitab,2);
 784             ewtabF           = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 785             ewtabD           = _mm_setzero_pd();
 786             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 787             ewtabV           = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,0) +2);
 788             ewtabFn          = _mm_setzero_pd();
 789             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 790             felec            = _mm_add_pd(ewtabF,_mm_mul_pd(eweps,ewtabD));
 791             velec            = _mm_sub_pd(ewtabV,_mm_mul_pd(_mm_mul_pd(ewtabhalfspace,eweps),_mm_add_pd(ewtabF,felec)));
 792             velec            = _mm_mul_pd(qq30,_mm_sub_pd(rinv30,velec));
 793             felec            = _mm_mul_pd(_mm_mul_pd(qq30,rinv30),_mm_sub_pd(rinvsq30,felec));
 794
 795             d                = _mm_sub_pd(r30,rswitch);
 796             d                = _mm_max_pd(d,_mm_setzero_pd());
 797             d2               = _mm_mul_pd(d,d);
 798             sw               = _mm_add_pd(one,_mm_mul_pd(d2,_mm_mul_pd(d,_mm_add_pd(swV3,_mm_mul_pd(d,_mm_add_pd(swV4,_mm_mul_pd(d,swV5)))))));
 799
 800             dsw              = _mm_mul_pd(d2,_mm_add_pd(swF2,_mm_mul_pd(d,_mm_add_pd(swF3,_mm_mul_pd(d,swF4)))));
 801
 802             /* Evaluate switch function */
 803             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
 804             felec            = _mm_sub_pd( _mm_mul_pd(felec,sw) , _mm_mul_pd(rinv30,_mm_mul_pd(velec,dsw)) );
 805             velec            = _mm_mul_pd(velec,sw);
 806             cutoff_mask      = _mm_cmplt_pd(rsq30,rcutoff2);
 807
 808             /* Update potential sum for this i atom from the interaction with this j atom. */
 809             velec            = _mm_and_pd(velec,cutoff_mask);
 810             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 811             velecsum         = _mm_add_pd(velecsum,velec);
 812
 813             fscal            = felec;
 814
 815             fscal            = _mm_and_pd(fscal,cutoff_mask);
 816
 817             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 818
 819             /* Calculate temporary vectorial force */
 820             tx               = _mm_mul_pd(fscal,dx30);
 821             ty               = _mm_mul_pd(fscal,dy30);
 822             tz               = _mm_mul_pd(fscal,dz30);
 823
 824             /* Update vectorial force */
 825             fix3             = _mm_add_pd(fix3,tx);
 826             fiy3             = _mm_add_pd(fiy3,ty);
 827             fiz3             = _mm_add_pd(fiz3,tz);
 828
 829             fjx0             = _mm_add_pd(fjx0,tx);
 830             fjy0             = _mm_add_pd(fjy0,ty);
 831             fjz0             = _mm_add_pd(fjz0,tz);
 832
 833             }
 834
 835             gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,fjx0,fjy0,fjz0);
 836
 837             /* Inner loop uses 257 flops */
 838         }
 839
 840         /* End of innermost loop */
 841
 842         gmx_mm_update_iforce_4atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
 843                                               f+i_coord_offset,fshift+i_shift_offset);
 844
 845         ggid                        = gid[iidx];
 846         /* Update potential energies */
 847         gmx_mm_update_1pot_pd(velecsum,kernel_data->energygrp_elec+ggid);
 848         gmx_mm_update_1pot_pd(vvdwsum,kernel_data->energygrp_vdw+ggid);
 849
 850         /* Increment number of inner iterations */
 851         inneriter                  += j_index_end - j_index_start;
 852
 853         /* Outer loop uses 26 flops */
 854     }
 855
 856     /* Increment number of outer iterations */
 857     outeriter        += nri;
 858
 859     /* Update outer/inner flops */
 860
 861     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W4_VF,outeriter*26 + inneriter*257);
 862 }
 863 /*
 864  * Gromacs nonbonded kernel:   nb_kernel_ElecEwSw_VdwLJSw_GeomW4P1_F_sse2_double
 865  * Electrostatics interaction: Ewald
 866  * VdW interaction:            LennardJones
 867  * Geometry:                   Water4-Particle
 868  * Calculate force/pot:        Force
 869  */
 870 void
 871 nb_kernel_ElecEwSw_VdwLJSw_GeomW4P1_F_sse2_double
 872                     (t_nblist                    * gmx_restrict       nlist,
 873                      rvec                        * gmx_restrict          xx,
 874                      rvec                        * gmx_restrict          ff,
 875                      t_forcerec                  * gmx_restrict          fr,
 876                      t_mdatoms                   * gmx_restrict     mdatoms,
 877                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 878                      t_nrnb                      * gmx_restrict        nrnb)
 879 {
 880     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 881      * just 0 for non-waters.
 882      * Suffixes A,B refer to j loop unrolling done with SSE double precision, e.g. for the two different
 883      * jnr indices corresponding to data put in the four positions in the SIMD register.
 884      */
 885     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 886     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 887     int              jnrA,jnrB;
 888     int              j_coord_offsetA,j_coord_offsetB;
 889     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 890     real             rcutoff_scalar;
 891     real             *shiftvec,*fshift,*x,*f;
 892     __m128d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 893     int              vdwioffset0;
 894     __m128d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 895     int              vdwioffset1;
 896     __m128d          ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
 897     int              vdwioffset2;
 898     __m128d          ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
 899     int              vdwioffset3;
 900     __m128d          ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
 901     int              vdwjidx0A,vdwjidx0B;
 902     __m128d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 903     __m128d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 904     __m128d          dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
 905     __m128d          dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
 906     __m128d          dx30,dy30,dz30,rsq30,rinv30,rinvsq30,r30,qq30,c6_30,c12_30;
 907     __m128d          velec,felec,velecsum,facel,crf,krf,krf2;
 908     real             *charge;
 909     int              nvdwtype;
 910     __m128d          rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 911     int              *vdwtype;
 912     real             *vdwparam;
 913     __m128d          one_sixth   = _mm_set1_pd(1.0/6.0);
 914     __m128d          one_twelfth = _mm_set1_pd(1.0/12.0);
 915     __m128i          ewitab;
 916     __m128d          ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
 917     real             *ewtab;
 918     __m128d          rswitch,swV3,swV4,swV5,swF2,swF3,swF4,d,d2,sw,dsw;
 919     real             rswitch_scalar,d_scalar;
 920     __m128d          dummy_mask,cutoff_mask;
 921     __m128d          signbit   = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
 922     __m128d          one     = _mm_set1_pd(1.0);
 923     __m128d          two     = _mm_set1_pd(2.0);
 924     x                = xx[0];
 925     f                = ff[0];
 926
 927     nri              = nlist->nri;
 928     iinr             = nlist->iinr;
 929     jindex           = nlist->jindex;
 930     jjnr             = nlist->jjnr;
 931     shiftidx         = nlist->shift;
 932     gid              = nlist->gid;
 933     shiftvec         = fr->shift_vec[0];
 934     fshift           = fr->fshift[0];
 935     facel            = _mm_set1_pd(fr->epsfac);
 936     charge           = mdatoms->chargeA;
 937     nvdwtype         = fr->ntype;
 938     vdwparam         = fr->nbfp;
 939     vdwtype          = mdatoms->typeA;
 940
 941     sh_ewald         = _mm_set1_pd(fr->ic->sh_ewald);
 942     ewtab            = fr->ic->tabq_coul_FDV0;
 943     ewtabscale       = _mm_set1_pd(fr->ic->tabq_scale);
 944     ewtabhalfspace   = _mm_set1_pd(0.5/fr->ic->tabq_scale);
 945
 946     /* Setup water-specific parameters */
 947     inr              = nlist->iinr[0];
 948     iq1              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+1]));
 949     iq2              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+2]));
 950     iq3              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+3]));
 951     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 952
 953     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 954     rcutoff_scalar   = fr->rcoulomb;
 955     rcutoff          = _mm_set1_pd(rcutoff_scalar);
 956     rcutoff2         = _mm_mul_pd(rcutoff,rcutoff);
 957
 958     rswitch_scalar   = fr->rcoulomb_switch;
 959     rswitch          = _mm_set1_pd(rswitch_scalar);
 960     /* Setup switch parameters */
 961     d_scalar         = rcutoff_scalar-rswitch_scalar;
 962     d                = _mm_set1_pd(d_scalar);
 963     swV3             = _mm_set1_pd(-10.0/(d_scalar*d_scalar*d_scalar));
 964     swV4             = _mm_set1_pd( 15.0/(d_scalar*d_scalar*d_scalar*d_scalar));
 965     swV5             = _mm_set1_pd( -6.0/(d_scalar*d_scalar*d_scalar*d_scalar*d_scalar));
 966     swF2             = _mm_set1_pd(-30.0/(d_scalar*d_scalar*d_scalar));
 967     swF3             = _mm_set1_pd( 60.0/(d_scalar*d_scalar*d_scalar*d_scalar));
 968     swF4             = _mm_set1_pd(-30.0/(d_scalar*d_scalar*d_scalar*d_scalar*d_scalar));
 969
 970     /* Avoid stupid compiler warnings */
 971     jnrA = jnrB = 0;
 972     j_coord_offsetA = 0;
 973     j_coord_offsetB = 0;
 974
 975     outeriter        = 0;
 976     inneriter        = 0;
 977
 978     /* Start outer loop over neighborlists */
 979     for(iidx=0; iidx<nri; iidx++)
 980     {
 981         /* Load shift vector for this list */
 982         i_shift_offset   = DIM*shiftidx[iidx];
 983
 984         /* Load limits for loop over neighbors */
 985         j_index_start    = jindex[iidx];
 986         j_index_end      = jindex[iidx+1];
 987
 988         /* Get outer coordinate index */
 989         inr              = iinr[iidx];
 990         i_coord_offset   = DIM*inr;
 991
 992         /* Load i particle coords and add shift vector */
 993         gmx_mm_load_shift_and_4rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
 994                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
 995
 996         fix0             = _mm_setzero_pd();
 997         fiy0             = _mm_setzero_pd();
 998         fiz0             = _mm_setzero_pd();
 999         fix1             = _mm_setzero_pd();
1000         fiy1             = _mm_setzero_pd();
1001         fiz1             = _mm_setzero_pd();
1002         fix2             = _mm_setzero_pd();
1003         fiy2             = _mm_setzero_pd();
1004         fiz2             = _mm_setzero_pd();
1005         fix3             = _mm_setzero_pd();
1006         fiy3             = _mm_setzero_pd();
1007         fiz3             = _mm_setzero_pd();
1008
1009         /* Start inner kernel loop */
1010         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
1011         {
1012
1013             /* Get j neighbor index, and coordinate index */
1014             jnrA             = jjnr[jidx];
1015             jnrB             = jjnr[jidx+1];
1016             j_coord_offsetA  = DIM*jnrA;
1017             j_coord_offsetB  = DIM*jnrB;
1018
1019             /* load j atom coordinates */
1020             gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
1021                                               &jx0,&jy0,&jz0);
1022
1023             /* Calculate displacement vector */
1024             dx00             = _mm_sub_pd(ix0,jx0);
1025             dy00             = _mm_sub_pd(iy0,jy0);
1026             dz00             = _mm_sub_pd(iz0,jz0);
1027             dx10             = _mm_sub_pd(ix1,jx0);
1028             dy10             = _mm_sub_pd(iy1,jy0);
1029             dz10             = _mm_sub_pd(iz1,jz0);
1030             dx20             = _mm_sub_pd(ix2,jx0);
1031             dy20             = _mm_sub_pd(iy2,jy0);
1032             dz20             = _mm_sub_pd(iz2,jz0);
1033             dx30             = _mm_sub_pd(ix3,jx0);
1034             dy30             = _mm_sub_pd(iy3,jy0);
1035             dz30             = _mm_sub_pd(iz3,jz0);
1036
1037             /* Calculate squared distance and things based on it */
1038             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
1039             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
1040             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
1041             rsq30            = gmx_mm_calc_rsq_pd(dx30,dy30,dz30);
1042
1043             rinv00           = gmx_mm_invsqrt_pd(rsq00);
1044             rinv10           = gmx_mm_invsqrt_pd(rsq10);
1045             rinv20           = gmx_mm_invsqrt_pd(rsq20);
1046             rinv30           = gmx_mm_invsqrt_pd(rsq30);
1047
1048             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
1049             rinvsq10         = _mm_mul_pd(rinv10,rinv10);
1050             rinvsq20         = _mm_mul_pd(rinv20,rinv20);
1051             rinvsq30         = _mm_mul_pd(rinv30,rinv30);
1052
1053             /* Load parameters for j particles */
1054             jq0              = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
1055             vdwjidx0A        = 2*vdwtype[jnrA+0];
1056             vdwjidx0B        = 2*vdwtype[jnrB+0];
1057
1058             fjx0             = _mm_setzero_pd();
1059             fjy0             = _mm_setzero_pd();
1060             fjz0             = _mm_setzero_pd();
1061
1062             /**************************
1063              * CALCULATE INTERACTIONS *
1064              **************************/
1065
1066             if (gmx_mm_any_lt(rsq00,rcutoff2))
1067             {
1068
1069             r00              = _mm_mul_pd(rsq00,rinv00);
1070
1071             /* Compute parameters for interactions between i and j atoms */
1072             gmx_mm_load_2pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,
1073                                          vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
1074
1075             /* LENNARD-JONES DISPERSION/REPULSION */
1076
1077             rinvsix          = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
1078             vvdw6            = _mm_mul_pd(c6_00,rinvsix);
1079             vvdw12           = _mm_mul_pd(c12_00,_mm_mul_pd(rinvsix,rinvsix));
1080             vvdw             = _mm_sub_pd( _mm_mul_pd(vvdw12,one_twelfth) , _mm_mul_pd(vvdw6,one_sixth) );
1081             fvdw             = _mm_mul_pd(_mm_sub_pd(vvdw12,vvdw6),rinvsq00);
1082
1083             d                = _mm_sub_pd(r00,rswitch);
1084             d                = _mm_max_pd(d,_mm_setzero_pd());
1085             d2               = _mm_mul_pd(d,d);
1086             sw               = _mm_add_pd(one,_mm_mul_pd(d2,_mm_mul_pd(d,_mm_add_pd(swV3,_mm_mul_pd(d,_mm_add_pd(swV4,_mm_mul_pd(d,swV5)))))));
1087
1088             dsw              = _mm_mul_pd(d2,_mm_add_pd(swF2,_mm_mul_pd(d,_mm_add_pd(swF3,_mm_mul_pd(d,swF4)))));
1089
1090             /* Evaluate switch function */
1091             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
1092             fvdw             = _mm_sub_pd( _mm_mul_pd(fvdw,sw) , _mm_mul_pd(rinv00,_mm_mul_pd(vvdw,dsw)) );
1093             cutoff_mask      = _mm_cmplt_pd(rsq00,rcutoff2);
1094
1095             fscal            = fvdw;
1096
1097             fscal            = _mm_and_pd(fscal,cutoff_mask);
1098
1099             /* Calculate temporary vectorial force */
1100             tx               = _mm_mul_pd(fscal,dx00);
1101             ty               = _mm_mul_pd(fscal,dy00);
1102             tz               = _mm_mul_pd(fscal,dz00);
1103
1104             /* Update vectorial force */
1105             fix0             = _mm_add_pd(fix0,tx);
1106             fiy0             = _mm_add_pd(fiy0,ty);
1107             fiz0             = _mm_add_pd(fiz0,tz);
1108
1109             fjx0             = _mm_add_pd(fjx0,tx);
1110             fjy0             = _mm_add_pd(fjy0,ty);
1111             fjz0             = _mm_add_pd(fjz0,tz);
1112
1113             }
1114
1115             /**************************
1116              * CALCULATE INTERACTIONS *
1117              **************************/
1118
1119             if (gmx_mm_any_lt(rsq10,rcutoff2))
1120             {
1121
1122             r10              = _mm_mul_pd(rsq10,rinv10);
1123
1124             /* Compute parameters for interactions between i and j atoms */
1125             qq10             = _mm_mul_pd(iq1,jq0);
1126
1127             /* EWALD ELECTROSTATICS */
1128
1129             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1130             ewrt             = _mm_mul_pd(r10,ewtabscale);
1131             ewitab           = _mm_cvttpd_epi32(ewrt);
1132             eweps            = _mm_sub_pd(ewrt,_mm_cvtepi32_pd(ewitab));
1133             ewitab           = _mm_slli_epi32(ewitab,2);
1134             ewtabF           = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,0) );
1135             ewtabD           = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,1) );
1136             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
1137             ewtabV           = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,0) +2);
1138             ewtabFn          = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,1) +2);
1139             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
1140             felec            = _mm_add_pd(ewtabF,_mm_mul_pd(eweps,ewtabD));
1141             velec            = _mm_sub_pd(ewtabV,_mm_mul_pd(_mm_mul_pd(ewtabhalfspace,eweps),_mm_add_pd(ewtabF,felec)));
1142             velec            = _mm_mul_pd(qq10,_mm_sub_pd(rinv10,velec));
1143             felec            = _mm_mul_pd(_mm_mul_pd(qq10,rinv10),_mm_sub_pd(rinvsq10,felec));
1144
1145             d                = _mm_sub_pd(r10,rswitch);
1146             d                = _mm_max_pd(d,_mm_setzero_pd());
1147             d2               = _mm_mul_pd(d,d);
1148             sw               = _mm_add_pd(one,_mm_mul_pd(d2,_mm_mul_pd(d,_mm_add_pd(swV3,_mm_mul_pd(d,_mm_add_pd(swV4,_mm_mul_pd(d,swV5)))))));
1149
1150             dsw              = _mm_mul_pd(d2,_mm_add_pd(swF2,_mm_mul_pd(d,_mm_add_pd(swF3,_mm_mul_pd(d,swF4)))));
1151
1152             /* Evaluate switch function */
1153             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
1154             felec            = _mm_sub_pd( _mm_mul_pd(felec,sw) , _mm_mul_pd(rinv10,_mm_mul_pd(velec,dsw)) );
1155             cutoff_mask      = _mm_cmplt_pd(rsq10,rcutoff2);
1156
1157             fscal            = felec;
1158
1159             fscal            = _mm_and_pd(fscal,cutoff_mask);
1160
1161             /* Calculate temporary vectorial force */
1162             tx               = _mm_mul_pd(fscal,dx10);
1163             ty               = _mm_mul_pd(fscal,dy10);
1164             tz               = _mm_mul_pd(fscal,dz10);
1165
1166             /* Update vectorial force */
1167             fix1             = _mm_add_pd(fix1,tx);
1168             fiy1             = _mm_add_pd(fiy1,ty);
1169             fiz1             = _mm_add_pd(fiz1,tz);
1170
1171             fjx0             = _mm_add_pd(fjx0,tx);
1172             fjy0             = _mm_add_pd(fjy0,ty);
1173             fjz0             = _mm_add_pd(fjz0,tz);
1174
1175             }
1176
1177             /**************************
1178              * CALCULATE INTERACTIONS *
1179              **************************/
1180
1181             if (gmx_mm_any_lt(rsq20,rcutoff2))
1182             {
1183
1184             r20              = _mm_mul_pd(rsq20,rinv20);
1185
1186             /* Compute parameters for interactions between i and j atoms */
1187             qq20             = _mm_mul_pd(iq2,jq0);
1188
1189             /* EWALD ELECTROSTATICS */
1190
1191             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1192             ewrt             = _mm_mul_pd(r20,ewtabscale);
1193             ewitab           = _mm_cvttpd_epi32(ewrt);
1194             eweps            = _mm_sub_pd(ewrt,_mm_cvtepi32_pd(ewitab));
1195             ewitab           = _mm_slli_epi32(ewitab,2);
1196             ewtabF           = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,0) );
1197             ewtabD           = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,1) );
1198             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
1199             ewtabV           = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,0) +2);
1200             ewtabFn          = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,1) +2);
1201             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
1202             felec            = _mm_add_pd(ewtabF,_mm_mul_pd(eweps,ewtabD));
1203             velec            = _mm_sub_pd(ewtabV,_mm_mul_pd(_mm_mul_pd(ewtabhalfspace,eweps),_mm_add_pd(ewtabF,felec)));
1204             velec            = _mm_mul_pd(qq20,_mm_sub_pd(rinv20,velec));
1205             felec            = _mm_mul_pd(_mm_mul_pd(qq20,rinv20),_mm_sub_pd(rinvsq20,felec));
1206
1207             d                = _mm_sub_pd(r20,rswitch);
1208             d                = _mm_max_pd(d,_mm_setzero_pd());
1209             d2               = _mm_mul_pd(d,d);
1210             sw               = _mm_add_pd(one,_mm_mul_pd(d2,_mm_mul_pd(d,_mm_add_pd(swV3,_mm_mul_pd(d,_mm_add_pd(swV4,_mm_mul_pd(d,swV5)))))));
1211
1212             dsw              = _mm_mul_pd(d2,_mm_add_pd(swF2,_mm_mul_pd(d,_mm_add_pd(swF3,_mm_mul_pd(d,swF4)))));
1213
1214             /* Evaluate switch function */
1215             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
1216             felec            = _mm_sub_pd( _mm_mul_pd(felec,sw) , _mm_mul_pd(rinv20,_mm_mul_pd(velec,dsw)) );
1217             cutoff_mask      = _mm_cmplt_pd(rsq20,rcutoff2);
1218
1219             fscal            = felec;
1220
1221             fscal            = _mm_and_pd(fscal,cutoff_mask);
1222
1223             /* Calculate temporary vectorial force */
1224             tx               = _mm_mul_pd(fscal,dx20);
1225             ty               = _mm_mul_pd(fscal,dy20);
1226             tz               = _mm_mul_pd(fscal,dz20);
1227
1228             /* Update vectorial force */
1229             fix2             = _mm_add_pd(fix2,tx);
1230             fiy2             = _mm_add_pd(fiy2,ty);
1231             fiz2             = _mm_add_pd(fiz2,tz);
1232
1233             fjx0             = _mm_add_pd(fjx0,tx);
1234             fjy0             = _mm_add_pd(fjy0,ty);
1235             fjz0             = _mm_add_pd(fjz0,tz);
1236
1237             }
1238
1239             /**************************
1240              * CALCULATE INTERACTIONS *
1241              **************************/
1242
1243             if (gmx_mm_any_lt(rsq30,rcutoff2))
1244             {
1245
1246             r30              = _mm_mul_pd(rsq30,rinv30);
1247
1248             /* Compute parameters for interactions between i and j atoms */
1249             qq30             = _mm_mul_pd(iq3,jq0);
1250
1251             /* EWALD ELECTROSTATICS */
1252
1253             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1254             ewrt             = _mm_mul_pd(r30,ewtabscale);
1255             ewitab           = _mm_cvttpd_epi32(ewrt);
1256             eweps            = _mm_sub_pd(ewrt,_mm_cvtepi32_pd(ewitab));
1257             ewitab           = _mm_slli_epi32(ewitab,2);
1258             ewtabF           = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,0) );
1259             ewtabD           = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,1) );
1260             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
1261             ewtabV           = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,0) +2);
1262             ewtabFn          = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,1) +2);
1263             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
1264             felec            = _mm_add_pd(ewtabF,_mm_mul_pd(eweps,ewtabD));
1265             velec            = _mm_sub_pd(ewtabV,_mm_mul_pd(_mm_mul_pd(ewtabhalfspace,eweps),_mm_add_pd(ewtabF,felec)));
1266             velec            = _mm_mul_pd(qq30,_mm_sub_pd(rinv30,velec));
1267             felec            = _mm_mul_pd(_mm_mul_pd(qq30,rinv30),_mm_sub_pd(rinvsq30,felec));
1268
1269             d                = _mm_sub_pd(r30,rswitch);
1270             d                = _mm_max_pd(d,_mm_setzero_pd());
1271             d2               = _mm_mul_pd(d,d);
1272             sw               = _mm_add_pd(one,_mm_mul_pd(d2,_mm_mul_pd(d,_mm_add_pd(swV3,_mm_mul_pd(d,_mm_add_pd(swV4,_mm_mul_pd(d,swV5)))))));
1273
1274             dsw              = _mm_mul_pd(d2,_mm_add_pd(swF2,_mm_mul_pd(d,_mm_add_pd(swF3,_mm_mul_pd(d,swF4)))));
1275
1276             /* Evaluate switch function */
1277             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
1278             felec            = _mm_sub_pd( _mm_mul_pd(felec,sw) , _mm_mul_pd(rinv30,_mm_mul_pd(velec,dsw)) );
1279             cutoff_mask      = _mm_cmplt_pd(rsq30,rcutoff2);
1280
1281             fscal            = felec;
1282
1283             fscal            = _mm_and_pd(fscal,cutoff_mask);
1284
1285             /* Calculate temporary vectorial force */
1286             tx               = _mm_mul_pd(fscal,dx30);
1287             ty               = _mm_mul_pd(fscal,dy30);
1288             tz               = _mm_mul_pd(fscal,dz30);
1289
1290             /* Update vectorial force */
1291             fix3             = _mm_add_pd(fix3,tx);
1292             fiy3             = _mm_add_pd(fiy3,ty);
1293             fiz3             = _mm_add_pd(fiz3,tz);
1294
1295             fjx0             = _mm_add_pd(fjx0,tx);
1296             fjy0             = _mm_add_pd(fjy0,ty);
1297             fjz0             = _mm_add_pd(fjz0,tz);
1298
1299             }
1300
1301             gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,fjx0,fjy0,fjz0);
1302
1303             /* Inner loop uses 245 flops */
1304         }
1305
1306         if(jidx<j_index_end)
1307         {
1308
1309             jnrA             = jjnr[jidx];
1310             j_coord_offsetA  = DIM*jnrA;
1311
1312             /* load j atom coordinates */
1313             gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
1314                                               &jx0,&jy0,&jz0);
1315
1316             /* Calculate displacement vector */
1317             dx00             = _mm_sub_pd(ix0,jx0);
1318             dy00             = _mm_sub_pd(iy0,jy0);
1319             dz00             = _mm_sub_pd(iz0,jz0);
1320             dx10             = _mm_sub_pd(ix1,jx0);
1321             dy10             = _mm_sub_pd(iy1,jy0);
1322             dz10             = _mm_sub_pd(iz1,jz0);
1323             dx20             = _mm_sub_pd(ix2,jx0);
1324             dy20             = _mm_sub_pd(iy2,jy0);
1325             dz20             = _mm_sub_pd(iz2,jz0);
1326             dx30             = _mm_sub_pd(ix3,jx0);
1327             dy30             = _mm_sub_pd(iy3,jy0);
1328             dz30             = _mm_sub_pd(iz3,jz0);
1329
1330             /* Calculate squared distance and things based on it */
1331             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
1332             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
1333             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
1334             rsq30            = gmx_mm_calc_rsq_pd(dx30,dy30,dz30);
1335
1336             rinv00           = gmx_mm_invsqrt_pd(rsq00);
1337             rinv10           = gmx_mm_invsqrt_pd(rsq10);
1338             rinv20           = gmx_mm_invsqrt_pd(rsq20);
1339             rinv30           = gmx_mm_invsqrt_pd(rsq30);
1340
1341             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
1342             rinvsq10         = _mm_mul_pd(rinv10,rinv10);
1343             rinvsq20         = _mm_mul_pd(rinv20,rinv20);
1344             rinvsq30         = _mm_mul_pd(rinv30,rinv30);
1345
1346             /* Load parameters for j particles */
1347             jq0              = _mm_load_sd(charge+jnrA+0);
1348             vdwjidx0A        = 2*vdwtype[jnrA+0];
1349
1350             fjx0             = _mm_setzero_pd();
1351             fjy0             = _mm_setzero_pd();
1352             fjz0             = _mm_setzero_pd();
1353
1354             /**************************
1355              * CALCULATE INTERACTIONS *
1356              **************************/
1357
1358             if (gmx_mm_any_lt(rsq00,rcutoff2))
1359             {
1360
1361             r00              = _mm_mul_pd(rsq00,rinv00);
1362
1363             /* Compute parameters for interactions between i and j atoms */
1364             gmx_mm_load_1pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,&c6_00,&c12_00);
1365
1366             /* LENNARD-JONES DISPERSION/REPULSION */
1367
1368             rinvsix          = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
1369             vvdw6            = _mm_mul_pd(c6_00,rinvsix);
1370             vvdw12           = _mm_mul_pd(c12_00,_mm_mul_pd(rinvsix,rinvsix));
1371             vvdw             = _mm_sub_pd( _mm_mul_pd(vvdw12,one_twelfth) , _mm_mul_pd(vvdw6,one_sixth) );
1372             fvdw             = _mm_mul_pd(_mm_sub_pd(vvdw12,vvdw6),rinvsq00);
1373
1374             d                = _mm_sub_pd(r00,rswitch);
1375             d                = _mm_max_pd(d,_mm_setzero_pd());
1376             d2               = _mm_mul_pd(d,d);
1377             sw               = _mm_add_pd(one,_mm_mul_pd(d2,_mm_mul_pd(d,_mm_add_pd(swV3,_mm_mul_pd(d,_mm_add_pd(swV4,_mm_mul_pd(d,swV5)))))));
1378
1379             dsw              = _mm_mul_pd(d2,_mm_add_pd(swF2,_mm_mul_pd(d,_mm_add_pd(swF3,_mm_mul_pd(d,swF4)))));
1380
1381             /* Evaluate switch function */
1382             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
1383             fvdw             = _mm_sub_pd( _mm_mul_pd(fvdw,sw) , _mm_mul_pd(rinv00,_mm_mul_pd(vvdw,dsw)) );
1384             cutoff_mask      = _mm_cmplt_pd(rsq00,rcutoff2);
1385
1386             fscal            = fvdw;
1387
1388             fscal            = _mm_and_pd(fscal,cutoff_mask);
1389
1390             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1391
1392             /* Calculate temporary vectorial force */
1393             tx               = _mm_mul_pd(fscal,dx00);
1394             ty               = _mm_mul_pd(fscal,dy00);
1395             tz               = _mm_mul_pd(fscal,dz00);
1396
1397             /* Update vectorial force */
1398             fix0             = _mm_add_pd(fix0,tx);
1399             fiy0             = _mm_add_pd(fiy0,ty);
1400             fiz0             = _mm_add_pd(fiz0,tz);
1401
1402             fjx0             = _mm_add_pd(fjx0,tx);
1403             fjy0             = _mm_add_pd(fjy0,ty);
1404             fjz0             = _mm_add_pd(fjz0,tz);
1405
1406             }
1407
1408             /**************************
1409              * CALCULATE INTERACTIONS *
1410              **************************/
1411
1412             if (gmx_mm_any_lt(rsq10,rcutoff2))
1413             {
1414
1415             r10              = _mm_mul_pd(rsq10,rinv10);
1416
1417             /* Compute parameters for interactions between i and j atoms */
1418             qq10             = _mm_mul_pd(iq1,jq0);
1419
1420             /* EWALD ELECTROSTATICS */
1421
1422             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1423             ewrt             = _mm_mul_pd(r10,ewtabscale);
1424             ewitab           = _mm_cvttpd_epi32(ewrt);
1425             eweps            = _mm_sub_pd(ewrt,_mm_cvtepi32_pd(ewitab));
1426             ewitab           = _mm_slli_epi32(ewitab,2);
1427             ewtabF           = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,0) );
1428             ewtabD           = _mm_setzero_pd();
1429             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
1430             ewtabV           = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,0) +2);
1431             ewtabFn          = _mm_setzero_pd();
1432             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
1433             felec            = _mm_add_pd(ewtabF,_mm_mul_pd(eweps,ewtabD));
1434             velec            = _mm_sub_pd(ewtabV,_mm_mul_pd(_mm_mul_pd(ewtabhalfspace,eweps),_mm_add_pd(ewtabF,felec)));
1435             velec            = _mm_mul_pd(qq10,_mm_sub_pd(rinv10,velec));
1436             felec            = _mm_mul_pd(_mm_mul_pd(qq10,rinv10),_mm_sub_pd(rinvsq10,felec));
1437
1438             d                = _mm_sub_pd(r10,rswitch);
1439             d                = _mm_max_pd(d,_mm_setzero_pd());
1440             d2               = _mm_mul_pd(d,d);
1441             sw               = _mm_add_pd(one,_mm_mul_pd(d2,_mm_mul_pd(d,_mm_add_pd(swV3,_mm_mul_pd(d,_mm_add_pd(swV4,_mm_mul_pd(d,swV5)))))));
1442
1443             dsw              = _mm_mul_pd(d2,_mm_add_pd(swF2,_mm_mul_pd(d,_mm_add_pd(swF3,_mm_mul_pd(d,swF4)))));
1444
1445             /* Evaluate switch function */
1446             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
1447             felec            = _mm_sub_pd( _mm_mul_pd(felec,sw) , _mm_mul_pd(rinv10,_mm_mul_pd(velec,dsw)) );
1448             cutoff_mask      = _mm_cmplt_pd(rsq10,rcutoff2);
1449
1450             fscal            = felec;
1451
1452             fscal            = _mm_and_pd(fscal,cutoff_mask);
1453
1454             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1455
1456             /* Calculate temporary vectorial force */
1457             tx               = _mm_mul_pd(fscal,dx10);
1458             ty               = _mm_mul_pd(fscal,dy10);
1459             tz               = _mm_mul_pd(fscal,dz10);
1460
1461             /* Update vectorial force */
1462             fix1             = _mm_add_pd(fix1,tx);
1463             fiy1             = _mm_add_pd(fiy1,ty);
1464             fiz1             = _mm_add_pd(fiz1,tz);
1465
1466             fjx0             = _mm_add_pd(fjx0,tx);
1467             fjy0             = _mm_add_pd(fjy0,ty);
1468             fjz0             = _mm_add_pd(fjz0,tz);
1469
1470             }
1471
1472             /**************************
1473              * CALCULATE INTERACTIONS *
1474              **************************/
1475
1476             if (gmx_mm_any_lt(rsq20,rcutoff2))
1477             {
1478
1479             r20              = _mm_mul_pd(rsq20,rinv20);
1480
1481             /* Compute parameters for interactions between i and j atoms */
1482             qq20             = _mm_mul_pd(iq2,jq0);
1483
1484             /* EWALD ELECTROSTATICS */
1485
1486             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1487             ewrt             = _mm_mul_pd(r20,ewtabscale);
1488             ewitab           = _mm_cvttpd_epi32(ewrt);
1489             eweps            = _mm_sub_pd(ewrt,_mm_cvtepi32_pd(ewitab));
1490             ewitab           = _mm_slli_epi32(ewitab,2);
1491             ewtabF           = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,0) );
1492             ewtabD           = _mm_setzero_pd();
1493             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
1494             ewtabV           = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,0) +2);
1495             ewtabFn          = _mm_setzero_pd();
1496             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
1497             felec            = _mm_add_pd(ewtabF,_mm_mul_pd(eweps,ewtabD));
1498             velec            = _mm_sub_pd(ewtabV,_mm_mul_pd(_mm_mul_pd(ewtabhalfspace,eweps),_mm_add_pd(ewtabF,felec)));
1499             velec            = _mm_mul_pd(qq20,_mm_sub_pd(rinv20,velec));
1500             felec            = _mm_mul_pd(_mm_mul_pd(qq20,rinv20),_mm_sub_pd(rinvsq20,felec));
1501
1502             d                = _mm_sub_pd(r20,rswitch);
1503             d                = _mm_max_pd(d,_mm_setzero_pd());
1504             d2               = _mm_mul_pd(d,d);
1505             sw               = _mm_add_pd(one,_mm_mul_pd(d2,_mm_mul_pd(d,_mm_add_pd(swV3,_mm_mul_pd(d,_mm_add_pd(swV4,_mm_mul_pd(d,swV5)))))));
1506
1507             dsw              = _mm_mul_pd(d2,_mm_add_pd(swF2,_mm_mul_pd(d,_mm_add_pd(swF3,_mm_mul_pd(d,swF4)))));
1508
1509             /* Evaluate switch function */
1510             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
1511             felec            = _mm_sub_pd( _mm_mul_pd(felec,sw) , _mm_mul_pd(rinv20,_mm_mul_pd(velec,dsw)) );
1512             cutoff_mask      = _mm_cmplt_pd(rsq20,rcutoff2);
1513
1514             fscal            = felec;
1515
1516             fscal            = _mm_and_pd(fscal,cutoff_mask);
1517
1518             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1519
1520             /* Calculate temporary vectorial force */
1521             tx               = _mm_mul_pd(fscal,dx20);
1522             ty               = _mm_mul_pd(fscal,dy20);
1523             tz               = _mm_mul_pd(fscal,dz20);
1524
1525             /* Update vectorial force */
1526             fix2             = _mm_add_pd(fix2,tx);
1527             fiy2             = _mm_add_pd(fiy2,ty);
1528             fiz2             = _mm_add_pd(fiz2,tz);
1529
1530             fjx0             = _mm_add_pd(fjx0,tx);
1531             fjy0             = _mm_add_pd(fjy0,ty);
1532             fjz0             = _mm_add_pd(fjz0,tz);
1533
1534             }
1535
1536             /**************************
1537              * CALCULATE INTERACTIONS *
1538              **************************/
1539
1540             if (gmx_mm_any_lt(rsq30,rcutoff2))
1541             {
1542
1543             r30              = _mm_mul_pd(rsq30,rinv30);
1544
1545             /* Compute parameters for interactions between i and j atoms */
1546             qq30             = _mm_mul_pd(iq3,jq0);
1547
1548             /* EWALD ELECTROSTATICS */
1549
1550             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1551             ewrt             = _mm_mul_pd(r30,ewtabscale);
1552             ewitab           = _mm_cvttpd_epi32(ewrt);
1553             eweps            = _mm_sub_pd(ewrt,_mm_cvtepi32_pd(ewitab));
1554             ewitab           = _mm_slli_epi32(ewitab,2);
1555             ewtabF           = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,0) );
1556             ewtabD           = _mm_setzero_pd();
1557             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
1558             ewtabV           = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,0) +2);
1559             ewtabFn          = _mm_setzero_pd();
1560             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
1561             felec            = _mm_add_pd(ewtabF,_mm_mul_pd(eweps,ewtabD));
1562             velec            = _mm_sub_pd(ewtabV,_mm_mul_pd(_mm_mul_pd(ewtabhalfspace,eweps),_mm_add_pd(ewtabF,felec)));
1563             velec            = _mm_mul_pd(qq30,_mm_sub_pd(rinv30,velec));
1564             felec            = _mm_mul_pd(_mm_mul_pd(qq30,rinv30),_mm_sub_pd(rinvsq30,felec));
1565
1566             d                = _mm_sub_pd(r30,rswitch);
1567             d                = _mm_max_pd(d,_mm_setzero_pd());
1568             d2               = _mm_mul_pd(d,d);
1569             sw               = _mm_add_pd(one,_mm_mul_pd(d2,_mm_mul_pd(d,_mm_add_pd(swV3,_mm_mul_pd(d,_mm_add_pd(swV4,_mm_mul_pd(d,swV5)))))));
1570
1571             dsw              = _mm_mul_pd(d2,_mm_add_pd(swF2,_mm_mul_pd(d,_mm_add_pd(swF3,_mm_mul_pd(d,swF4)))));
1572
1573             /* Evaluate switch function */
1574             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
1575             felec            = _mm_sub_pd( _mm_mul_pd(felec,sw) , _mm_mul_pd(rinv30,_mm_mul_pd(velec,dsw)) );
1576             cutoff_mask      = _mm_cmplt_pd(rsq30,rcutoff2);
1577
1578             fscal            = felec;
1579
1580             fscal            = _mm_and_pd(fscal,cutoff_mask);
1581
1582             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1583
1584             /* Calculate temporary vectorial force */
1585             tx               = _mm_mul_pd(fscal,dx30);
1586             ty               = _mm_mul_pd(fscal,dy30);
1587             tz               = _mm_mul_pd(fscal,dz30);
1588
1589             /* Update vectorial force */
1590             fix3             = _mm_add_pd(fix3,tx);
1591             fiy3             = _mm_add_pd(fiy3,ty);
1592             fiz3             = _mm_add_pd(fiz3,tz);
1593
1594             fjx0             = _mm_add_pd(fjx0,tx);
1595             fjy0             = _mm_add_pd(fjy0,ty);
1596             fjz0             = _mm_add_pd(fjz0,tz);
1597
1598             }
1599
1600             gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,fjx0,fjy0,fjz0);
1601
1602             /* Inner loop uses 245 flops */
1603         }
1604
1605         /* End of innermost loop */
1606
1607         gmx_mm_update_iforce_4atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
1608                                               f+i_coord_offset,fshift+i_shift_offset);
1609
1610         /* Increment number of inner iterations */
1611         inneriter                  += j_index_end - j_index_start;
1612
1613         /* Outer loop uses 24 flops */
1614     }
1615
1616     /* Increment number of outer iterations */
1617     outeriter        += nri;
1618
1619     /* Update outer/inner flops */
1620
1621     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W4_F,outeriter*24 + inneriter*245);
1622 }