src/gromacs/gmxlib/nonbonded/nb_kernel_sse4_1_double/nb_kernel_ElecEwSw_VdwLJSw_GeomW4P1_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_ElecEwSw_VdwLJSw_GeomW4P1_VF_sse4_1_double
  54  * Electrostatics interaction: Ewald
  55  * VdW interaction:            LennardJones
  56  * Geometry:                   Water4-Particle
  57  * Calculate force/pot:        PotentialAndForce
  58  */
  59 void
  60 nb_kernel_ElecEwSw_VdwLJSw_GeomW4P1_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              vdwioffset3;
  89     __m128d          ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
  90     int              vdwjidx0A,vdwjidx0B;
  91     __m128d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  92     __m128d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  93     __m128d          dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
  94     __m128d          dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
  95     __m128d          dx30,dy30,dz30,rsq30,rinv30,rinvsq30,r30,qq30,c6_30,c12_30;
  96     __m128d          velec,felec,velecsum,facel,crf,krf,krf2;
  97     real             *charge;
  98     int              nvdwtype;
  99     __m128d          rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 100     int              *vdwtype;
 101     real             *vdwparam;
 102     __m128d          one_sixth   = _mm_set1_pd(1.0/6.0);
 103     __m128d          one_twelfth = _mm_set1_pd(1.0/12.0);
 104     __m128i          ewitab;
 105     __m128d          ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
 106     real             *ewtab;
 107     __m128d          rswitch,swV3,swV4,swV5,swF2,swF3,swF4,d,d2,sw,dsw;
 108     real             rswitch_scalar,d_scalar;
 109     __m128d          dummy_mask,cutoff_mask;
 110     __m128d          signbit   = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
 111     __m128d          one     = _mm_set1_pd(1.0);
 112     __m128d          two     = _mm_set1_pd(2.0);
 113     x                = xx[0];
 114     f                = ff[0];
 115
 116     nri              = nlist->nri;
 117     iinr             = nlist->iinr;
 118     jindex           = nlist->jindex;
 119     jjnr             = nlist->jjnr;
 120     shiftidx         = nlist->shift;
 121     gid              = nlist->gid;
 122     shiftvec         = fr->shift_vec[0];
 123     fshift           = fr->fshift[0];
 124     facel            = _mm_set1_pd(fr->epsfac);
 125     charge           = mdatoms->chargeA;
 126     nvdwtype         = fr->ntype;
 127     vdwparam         = fr->nbfp;
 128     vdwtype          = mdatoms->typeA;
 129
 130     sh_ewald         = _mm_set1_pd(fr->ic->sh_ewald);
 131     ewtab            = fr->ic->tabq_coul_FDV0;
 132     ewtabscale       = _mm_set1_pd(fr->ic->tabq_scale);
 133     ewtabhalfspace   = _mm_set1_pd(0.5/fr->ic->tabq_scale);
 134
 135     /* Setup water-specific parameters */
 136     inr              = nlist->iinr[0];
 137     iq1              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+1]));
 138     iq2              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+2]));
 139     iq3              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+3]));
 140     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 141
 142     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 143     rcutoff_scalar   = fr->rcoulomb;
 144     rcutoff          = _mm_set1_pd(rcutoff_scalar);
 145     rcutoff2         = _mm_mul_pd(rcutoff,rcutoff);
 146
 147     rswitch_scalar   = fr->rcoulomb_switch;
 148     rswitch          = _mm_set1_pd(rswitch_scalar);
 149     /* Setup switch parameters */
 150     d_scalar         = rcutoff_scalar-rswitch_scalar;
 151     d                = _mm_set1_pd(d_scalar);
 152     swV3             = _mm_set1_pd(-10.0/(d_scalar*d_scalar*d_scalar));
 153     swV4             = _mm_set1_pd( 15.0/(d_scalar*d_scalar*d_scalar*d_scalar));
 154     swV5             = _mm_set1_pd( -6.0/(d_scalar*d_scalar*d_scalar*d_scalar*d_scalar));
 155     swF2             = _mm_set1_pd(-30.0/(d_scalar*d_scalar*d_scalar));
 156     swF3             = _mm_set1_pd( 60.0/(d_scalar*d_scalar*d_scalar*d_scalar));
 157     swF4             = _mm_set1_pd(-30.0/(d_scalar*d_scalar*d_scalar*d_scalar*d_scalar));
 158
 159     /* Avoid stupid compiler warnings */
 160     jnrA = jnrB = 0;
 161     j_coord_offsetA = 0;
 162     j_coord_offsetB = 0;
 163
 164     outeriter        = 0;
 165     inneriter        = 0;
 166
 167     /* Start outer loop over neighborlists */
 168     for(iidx=0; iidx<nri; iidx++)
 169     {
 170         /* Load shift vector for this list */
 171         i_shift_offset   = DIM*shiftidx[iidx];
 172
 173         /* Load limits for loop over neighbors */
 174         j_index_start    = jindex[iidx];
 175         j_index_end      = jindex[iidx+1];
 176
 177         /* Get outer coordinate index */
 178         inr              = iinr[iidx];
 179         i_coord_offset   = DIM*inr;
 180
 181         /* Load i particle coords and add shift vector */
 182         gmx_mm_load_shift_and_4rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
 183                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
 184
 185         fix0             = _mm_setzero_pd();
 186         fiy0             = _mm_setzero_pd();
 187         fiz0             = _mm_setzero_pd();
 188         fix1             = _mm_setzero_pd();
 189         fiy1             = _mm_setzero_pd();
 190         fiz1             = _mm_setzero_pd();
 191         fix2             = _mm_setzero_pd();
 192         fiy2             = _mm_setzero_pd();
 193         fiz2             = _mm_setzero_pd();
 194         fix3             = _mm_setzero_pd();
 195         fiy3             = _mm_setzero_pd();
 196         fiz3             = _mm_setzero_pd();
 197
 198         /* Reset potential sums */
 199         velecsum         = _mm_setzero_pd();
 200         vvdwsum          = _mm_setzero_pd();
 201
 202         /* Start inner kernel loop */
 203         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 204         {
 205
 206             /* Get j neighbor index, and coordinate index */
 207             jnrA             = jjnr[jidx];
 208             jnrB             = jjnr[jidx+1];
 209             j_coord_offsetA  = DIM*jnrA;
 210             j_coord_offsetB  = DIM*jnrB;
 211
 212             /* load j atom coordinates */
 213             gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 214                                               &jx0,&jy0,&jz0);
 215
 216             /* Calculate displacement vector */
 217             dx00             = _mm_sub_pd(ix0,jx0);
 218             dy00             = _mm_sub_pd(iy0,jy0);
 219             dz00             = _mm_sub_pd(iz0,jz0);
 220             dx10             = _mm_sub_pd(ix1,jx0);
 221             dy10             = _mm_sub_pd(iy1,jy0);
 222             dz10             = _mm_sub_pd(iz1,jz0);
 223             dx20             = _mm_sub_pd(ix2,jx0);
 224             dy20             = _mm_sub_pd(iy2,jy0);
 225             dz20             = _mm_sub_pd(iz2,jz0);
 226             dx30             = _mm_sub_pd(ix3,jx0);
 227             dy30             = _mm_sub_pd(iy3,jy0);
 228             dz30             = _mm_sub_pd(iz3,jz0);
 229
 230             /* Calculate squared distance and things based on it */
 231             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 232             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
 233             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
 234             rsq30            = gmx_mm_calc_rsq_pd(dx30,dy30,dz30);
 235
 236             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 237             rinv10           = gmx_mm_invsqrt_pd(rsq10);
 238             rinv20           = gmx_mm_invsqrt_pd(rsq20);
 239             rinv30           = gmx_mm_invsqrt_pd(rsq30);
 240
 241             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 242             rinvsq10         = _mm_mul_pd(rinv10,rinv10);
 243             rinvsq20         = _mm_mul_pd(rinv20,rinv20);
 244             rinvsq30         = _mm_mul_pd(rinv30,rinv30);
 245
 246             /* Load parameters for j particles */
 247             jq0              = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
 248             vdwjidx0A        = 2*vdwtype[jnrA+0];
 249             vdwjidx0B        = 2*vdwtype[jnrB+0];
 250
 251             fjx0             = _mm_setzero_pd();
 252             fjy0             = _mm_setzero_pd();
 253             fjz0             = _mm_setzero_pd();
 254
 255             /**************************
 256              * CALCULATE INTERACTIONS *
 257              **************************/
 258
 259             if (gmx_mm_any_lt(rsq00,rcutoff2))
 260             {
 261
 262             r00              = _mm_mul_pd(rsq00,rinv00);
 263
 264             /* Compute parameters for interactions between i and j atoms */
 265             gmx_mm_load_2pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,
 266                                          vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
 267
 268             /* LENNARD-JONES DISPERSION/REPULSION */
 269
 270             rinvsix          = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 271             vvdw6            = _mm_mul_pd(c6_00,rinvsix);
 272             vvdw12           = _mm_mul_pd(c12_00,_mm_mul_pd(rinvsix,rinvsix));
 273             vvdw             = _mm_sub_pd( _mm_mul_pd(vvdw12,one_twelfth) , _mm_mul_pd(vvdw6,one_sixth) );
 274             fvdw             = _mm_mul_pd(_mm_sub_pd(vvdw12,vvdw6),rinvsq00);
 275
 276             d                = _mm_sub_pd(r00,rswitch);
 277             d                = _mm_max_pd(d,_mm_setzero_pd());
 278             d2               = _mm_mul_pd(d,d);
 279             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)))))));
 280
 281             dsw              = _mm_mul_pd(d2,_mm_add_pd(swF2,_mm_mul_pd(d,_mm_add_pd(swF3,_mm_mul_pd(d,swF4)))));
 282
 283             /* Evaluate switch function */
 284             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
 285             fvdw             = _mm_sub_pd( _mm_mul_pd(fvdw,sw) , _mm_mul_pd(rinv00,_mm_mul_pd(vvdw,dsw)) );
 286             vvdw             = _mm_mul_pd(vvdw,sw);
 287             cutoff_mask      = _mm_cmplt_pd(rsq00,rcutoff2);
 288
 289             /* Update potential sum for this i atom from the interaction with this j atom. */
 290             vvdw             = _mm_and_pd(vvdw,cutoff_mask);
 291             vvdwsum          = _mm_add_pd(vvdwsum,vvdw);
 292
 293             fscal            = fvdw;
 294
 295             fscal            = _mm_and_pd(fscal,cutoff_mask);
 296
 297             /* Calculate temporary vectorial force */
 298             tx               = _mm_mul_pd(fscal,dx00);
 299             ty               = _mm_mul_pd(fscal,dy00);
 300             tz               = _mm_mul_pd(fscal,dz00);
 301
 302             /* Update vectorial force */
 303             fix0             = _mm_add_pd(fix0,tx);
 304             fiy0             = _mm_add_pd(fiy0,ty);
 305             fiz0             = _mm_add_pd(fiz0,tz);
 306
 307             fjx0             = _mm_add_pd(fjx0,tx);
 308             fjy0             = _mm_add_pd(fjy0,ty);
 309             fjz0             = _mm_add_pd(fjz0,tz);
 310
 311             }
 312
 313             /**************************
 314              * CALCULATE INTERACTIONS *
 315              **************************/
 316
 317             if (gmx_mm_any_lt(rsq10,rcutoff2))
 318             {
 319
 320             r10              = _mm_mul_pd(rsq10,rinv10);
 321
 322             /* Compute parameters for interactions between i and j atoms */
 323             qq10             = _mm_mul_pd(iq1,jq0);
 324
 325             /* EWALD ELECTROSTATICS */
 326
 327             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 328             ewrt             = _mm_mul_pd(r10,ewtabscale);
 329             ewitab           = _mm_cvttpd_epi32(ewrt);
 330             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 331             ewitab           = _mm_slli_epi32(ewitab,2);
 332             ewtabF           = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 333             ewtabD           = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,1) );
 334             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 335             ewtabV           = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,0) +2);
 336             ewtabFn          = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,1) +2);
 337             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 338             felec            = _mm_add_pd(ewtabF,_mm_mul_pd(eweps,ewtabD));
 339             velec            = _mm_sub_pd(ewtabV,_mm_mul_pd(_mm_mul_pd(ewtabhalfspace,eweps),_mm_add_pd(ewtabF,felec)));
 340             velec            = _mm_mul_pd(qq10,_mm_sub_pd(rinv10,velec));
 341             felec            = _mm_mul_pd(_mm_mul_pd(qq10,rinv10),_mm_sub_pd(rinvsq10,felec));
 342
 343             d                = _mm_sub_pd(r10,rswitch);
 344             d                = _mm_max_pd(d,_mm_setzero_pd());
 345             d2               = _mm_mul_pd(d,d);
 346             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)))))));
 347
 348             dsw              = _mm_mul_pd(d2,_mm_add_pd(swF2,_mm_mul_pd(d,_mm_add_pd(swF3,_mm_mul_pd(d,swF4)))));
 349
 350             /* Evaluate switch function */
 351             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
 352             felec            = _mm_sub_pd( _mm_mul_pd(felec,sw) , _mm_mul_pd(rinv10,_mm_mul_pd(velec,dsw)) );
 353             velec            = _mm_mul_pd(velec,sw);
 354             cutoff_mask      = _mm_cmplt_pd(rsq10,rcutoff2);
 355
 356             /* Update potential sum for this i atom from the interaction with this j atom. */
 357             velec            = _mm_and_pd(velec,cutoff_mask);
 358             velecsum         = _mm_add_pd(velecsum,velec);
 359
 360             fscal            = felec;
 361
 362             fscal            = _mm_and_pd(fscal,cutoff_mask);
 363
 364             /* Calculate temporary vectorial force */
 365             tx               = _mm_mul_pd(fscal,dx10);
 366             ty               = _mm_mul_pd(fscal,dy10);
 367             tz               = _mm_mul_pd(fscal,dz10);
 368
 369             /* Update vectorial force */
 370             fix1             = _mm_add_pd(fix1,tx);
 371             fiy1             = _mm_add_pd(fiy1,ty);
 372             fiz1             = _mm_add_pd(fiz1,tz);
 373
 374             fjx0             = _mm_add_pd(fjx0,tx);
 375             fjy0             = _mm_add_pd(fjy0,ty);
 376             fjz0             = _mm_add_pd(fjz0,tz);
 377
 378             }
 379
 380             /**************************
 381              * CALCULATE INTERACTIONS *
 382              **************************/
 383
 384             if (gmx_mm_any_lt(rsq20,rcutoff2))
 385             {
 386
 387             r20              = _mm_mul_pd(rsq20,rinv20);
 388
 389             /* Compute parameters for interactions between i and j atoms */
 390             qq20             = _mm_mul_pd(iq2,jq0);
 391
 392             /* EWALD ELECTROSTATICS */
 393
 394             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 395             ewrt             = _mm_mul_pd(r20,ewtabscale);
 396             ewitab           = _mm_cvttpd_epi32(ewrt);
 397             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 398             ewitab           = _mm_slli_epi32(ewitab,2);
 399             ewtabF           = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 400             ewtabD           = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,1) );
 401             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 402             ewtabV           = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,0) +2);
 403             ewtabFn          = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,1) +2);
 404             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 405             felec            = _mm_add_pd(ewtabF,_mm_mul_pd(eweps,ewtabD));
 406             velec            = _mm_sub_pd(ewtabV,_mm_mul_pd(_mm_mul_pd(ewtabhalfspace,eweps),_mm_add_pd(ewtabF,felec)));
 407             velec            = _mm_mul_pd(qq20,_mm_sub_pd(rinv20,velec));
 408             felec            = _mm_mul_pd(_mm_mul_pd(qq20,rinv20),_mm_sub_pd(rinvsq20,felec));
 409
 410             d                = _mm_sub_pd(r20,rswitch);
 411             d                = _mm_max_pd(d,_mm_setzero_pd());
 412             d2               = _mm_mul_pd(d,d);
 413             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)))))));
 414
 415             dsw              = _mm_mul_pd(d2,_mm_add_pd(swF2,_mm_mul_pd(d,_mm_add_pd(swF3,_mm_mul_pd(d,swF4)))));
 416
 417             /* Evaluate switch function */
 418             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
 419             felec            = _mm_sub_pd( _mm_mul_pd(felec,sw) , _mm_mul_pd(rinv20,_mm_mul_pd(velec,dsw)) );
 420             velec            = _mm_mul_pd(velec,sw);
 421             cutoff_mask      = _mm_cmplt_pd(rsq20,rcutoff2);
 422
 423             /* Update potential sum for this i atom from the interaction with this j atom. */
 424             velec            = _mm_and_pd(velec,cutoff_mask);
 425             velecsum         = _mm_add_pd(velecsum,velec);
 426
 427             fscal            = felec;
 428
 429             fscal            = _mm_and_pd(fscal,cutoff_mask);
 430
 431             /* Calculate temporary vectorial force */
 432             tx               = _mm_mul_pd(fscal,dx20);
 433             ty               = _mm_mul_pd(fscal,dy20);
 434             tz               = _mm_mul_pd(fscal,dz20);
 435
 436             /* Update vectorial force */
 437             fix2             = _mm_add_pd(fix2,tx);
 438             fiy2             = _mm_add_pd(fiy2,ty);
 439             fiz2             = _mm_add_pd(fiz2,tz);
 440
 441             fjx0             = _mm_add_pd(fjx0,tx);
 442             fjy0             = _mm_add_pd(fjy0,ty);
 443             fjz0             = _mm_add_pd(fjz0,tz);
 444
 445             }
 446
 447             /**************************
 448              * CALCULATE INTERACTIONS *
 449              **************************/
 450
 451             if (gmx_mm_any_lt(rsq30,rcutoff2))
 452             {
 453
 454             r30              = _mm_mul_pd(rsq30,rinv30);
 455
 456             /* Compute parameters for interactions between i and j atoms */
 457             qq30             = _mm_mul_pd(iq3,jq0);
 458
 459             /* EWALD ELECTROSTATICS */
 460
 461             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 462             ewrt             = _mm_mul_pd(r30,ewtabscale);
 463             ewitab           = _mm_cvttpd_epi32(ewrt);
 464             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 465             ewitab           = _mm_slli_epi32(ewitab,2);
 466             ewtabF           = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 467             ewtabD           = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,1) );
 468             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 469             ewtabV           = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,0) +2);
 470             ewtabFn          = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,1) +2);
 471             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 472             felec            = _mm_add_pd(ewtabF,_mm_mul_pd(eweps,ewtabD));
 473             velec            = _mm_sub_pd(ewtabV,_mm_mul_pd(_mm_mul_pd(ewtabhalfspace,eweps),_mm_add_pd(ewtabF,felec)));
 474             velec            = _mm_mul_pd(qq30,_mm_sub_pd(rinv30,velec));
 475             felec            = _mm_mul_pd(_mm_mul_pd(qq30,rinv30),_mm_sub_pd(rinvsq30,felec));
 476
 477             d                = _mm_sub_pd(r30,rswitch);
 478             d                = _mm_max_pd(d,_mm_setzero_pd());
 479             d2               = _mm_mul_pd(d,d);
 480             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)))))));
 481
 482             dsw              = _mm_mul_pd(d2,_mm_add_pd(swF2,_mm_mul_pd(d,_mm_add_pd(swF3,_mm_mul_pd(d,swF4)))));
 483
 484             /* Evaluate switch function */
 485             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
 486             felec            = _mm_sub_pd( _mm_mul_pd(felec,sw) , _mm_mul_pd(rinv30,_mm_mul_pd(velec,dsw)) );
 487             velec            = _mm_mul_pd(velec,sw);
 488             cutoff_mask      = _mm_cmplt_pd(rsq30,rcutoff2);
 489
 490             /* Update potential sum for this i atom from the interaction with this j atom. */
 491             velec            = _mm_and_pd(velec,cutoff_mask);
 492             velecsum         = _mm_add_pd(velecsum,velec);
 493
 494             fscal            = felec;
 495
 496             fscal            = _mm_and_pd(fscal,cutoff_mask);
 497
 498             /* Calculate temporary vectorial force */
 499             tx               = _mm_mul_pd(fscal,dx30);
 500             ty               = _mm_mul_pd(fscal,dy30);
 501             tz               = _mm_mul_pd(fscal,dz30);
 502
 503             /* Update vectorial force */
 504             fix3             = _mm_add_pd(fix3,tx);
 505             fiy3             = _mm_add_pd(fiy3,ty);
 506             fiz3             = _mm_add_pd(fiz3,tz);
 507
 508             fjx0             = _mm_add_pd(fjx0,tx);
 509             fjy0             = _mm_add_pd(fjy0,ty);
 510             fjz0             = _mm_add_pd(fjz0,tz);
 511
 512             }
 513
 514             gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,fjx0,fjy0,fjz0);
 515
 516             /* Inner loop uses 257 flops */
 517         }
 518
 519         if(jidx<j_index_end)
 520         {
 521
 522             jnrA             = jjnr[jidx];
 523             j_coord_offsetA  = DIM*jnrA;
 524
 525             /* load j atom coordinates */
 526             gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
 527                                               &jx0,&jy0,&jz0);
 528
 529             /* Calculate displacement vector */
 530             dx00             = _mm_sub_pd(ix0,jx0);
 531             dy00             = _mm_sub_pd(iy0,jy0);
 532             dz00             = _mm_sub_pd(iz0,jz0);
 533             dx10             = _mm_sub_pd(ix1,jx0);
 534             dy10             = _mm_sub_pd(iy1,jy0);
 535             dz10             = _mm_sub_pd(iz1,jz0);
 536             dx20             = _mm_sub_pd(ix2,jx0);
 537             dy20             = _mm_sub_pd(iy2,jy0);
 538             dz20             = _mm_sub_pd(iz2,jz0);
 539             dx30             = _mm_sub_pd(ix3,jx0);
 540             dy30             = _mm_sub_pd(iy3,jy0);
 541             dz30             = _mm_sub_pd(iz3,jz0);
 542
 543             /* Calculate squared distance and things based on it */
 544             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 545             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
 546             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
 547             rsq30            = gmx_mm_calc_rsq_pd(dx30,dy30,dz30);
 548
 549             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 550             rinv10           = gmx_mm_invsqrt_pd(rsq10);
 551             rinv20           = gmx_mm_invsqrt_pd(rsq20);
 552             rinv30           = gmx_mm_invsqrt_pd(rsq30);
 553
 554             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 555             rinvsq10         = _mm_mul_pd(rinv10,rinv10);
 556             rinvsq20         = _mm_mul_pd(rinv20,rinv20);
 557             rinvsq30         = _mm_mul_pd(rinv30,rinv30);
 558
 559             /* Load parameters for j particles */
 560             jq0              = _mm_load_sd(charge+jnrA+0);
 561             vdwjidx0A        = 2*vdwtype[jnrA+0];
 562
 563             fjx0             = _mm_setzero_pd();
 564             fjy0             = _mm_setzero_pd();
 565             fjz0             = _mm_setzero_pd();
 566
 567             /**************************
 568              * CALCULATE INTERACTIONS *
 569              **************************/
 570
 571             if (gmx_mm_any_lt(rsq00,rcutoff2))
 572             {
 573
 574             r00              = _mm_mul_pd(rsq00,rinv00);
 575
 576             /* Compute parameters for interactions between i and j atoms */
 577             gmx_mm_load_1pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,&c6_00,&c12_00);
 578
 579             /* LENNARD-JONES DISPERSION/REPULSION */
 580
 581             rinvsix          = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 582             vvdw6            = _mm_mul_pd(c6_00,rinvsix);
 583             vvdw12           = _mm_mul_pd(c12_00,_mm_mul_pd(rinvsix,rinvsix));
 584             vvdw             = _mm_sub_pd( _mm_mul_pd(vvdw12,one_twelfth) , _mm_mul_pd(vvdw6,one_sixth) );
 585             fvdw             = _mm_mul_pd(_mm_sub_pd(vvdw12,vvdw6),rinvsq00);
 586
 587             d                = _mm_sub_pd(r00,rswitch);
 588             d                = _mm_max_pd(d,_mm_setzero_pd());
 589             d2               = _mm_mul_pd(d,d);
 590             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)))))));
 591
 592             dsw              = _mm_mul_pd(d2,_mm_add_pd(swF2,_mm_mul_pd(d,_mm_add_pd(swF3,_mm_mul_pd(d,swF4)))));
 593
 594             /* Evaluate switch function */
 595             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
 596             fvdw             = _mm_sub_pd( _mm_mul_pd(fvdw,sw) , _mm_mul_pd(rinv00,_mm_mul_pd(vvdw,dsw)) );
 597             vvdw             = _mm_mul_pd(vvdw,sw);
 598             cutoff_mask      = _mm_cmplt_pd(rsq00,rcutoff2);
 599
 600             /* Update potential sum for this i atom from the interaction with this j atom. */
 601             vvdw             = _mm_and_pd(vvdw,cutoff_mask);
 602             vvdw             = _mm_unpacklo_pd(vvdw,_mm_setzero_pd());
 603             vvdwsum          = _mm_add_pd(vvdwsum,vvdw);
 604
 605             fscal            = fvdw;
 606
 607             fscal            = _mm_and_pd(fscal,cutoff_mask);
 608
 609             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 610
 611             /* Calculate temporary vectorial force */
 612             tx               = _mm_mul_pd(fscal,dx00);
 613             ty               = _mm_mul_pd(fscal,dy00);
 614             tz               = _mm_mul_pd(fscal,dz00);
 615
 616             /* Update vectorial force */
 617             fix0             = _mm_add_pd(fix0,tx);
 618             fiy0             = _mm_add_pd(fiy0,ty);
 619             fiz0             = _mm_add_pd(fiz0,tz);
 620
 621             fjx0             = _mm_add_pd(fjx0,tx);
 622             fjy0             = _mm_add_pd(fjy0,ty);
 623             fjz0             = _mm_add_pd(fjz0,tz);
 624
 625             }
 626
 627             /**************************
 628              * CALCULATE INTERACTIONS *
 629              **************************/
 630
 631             if (gmx_mm_any_lt(rsq10,rcutoff2))
 632             {
 633
 634             r10              = _mm_mul_pd(rsq10,rinv10);
 635
 636             /* Compute parameters for interactions between i and j atoms */
 637             qq10             = _mm_mul_pd(iq1,jq0);
 638
 639             /* EWALD ELECTROSTATICS */
 640
 641             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 642             ewrt             = _mm_mul_pd(r10,ewtabscale);
 643             ewitab           = _mm_cvttpd_epi32(ewrt);
 644             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 645             ewitab           = _mm_slli_epi32(ewitab,2);
 646             ewtabF           = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 647             ewtabD           = _mm_setzero_pd();
 648             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 649             ewtabV           = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,0) +2);
 650             ewtabFn          = _mm_setzero_pd();
 651             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 652             felec            = _mm_add_pd(ewtabF,_mm_mul_pd(eweps,ewtabD));
 653             velec            = _mm_sub_pd(ewtabV,_mm_mul_pd(_mm_mul_pd(ewtabhalfspace,eweps),_mm_add_pd(ewtabF,felec)));
 654             velec            = _mm_mul_pd(qq10,_mm_sub_pd(rinv10,velec));
 655             felec            = _mm_mul_pd(_mm_mul_pd(qq10,rinv10),_mm_sub_pd(rinvsq10,felec));
 656
 657             d                = _mm_sub_pd(r10,rswitch);
 658             d                = _mm_max_pd(d,_mm_setzero_pd());
 659             d2               = _mm_mul_pd(d,d);
 660             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)))))));
 661
 662             dsw              = _mm_mul_pd(d2,_mm_add_pd(swF2,_mm_mul_pd(d,_mm_add_pd(swF3,_mm_mul_pd(d,swF4)))));
 663
 664             /* Evaluate switch function */
 665             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
 666             felec            = _mm_sub_pd( _mm_mul_pd(felec,sw) , _mm_mul_pd(rinv10,_mm_mul_pd(velec,dsw)) );
 667             velec            = _mm_mul_pd(velec,sw);
 668             cutoff_mask      = _mm_cmplt_pd(rsq10,rcutoff2);
 669
 670             /* Update potential sum for this i atom from the interaction with this j atom. */
 671             velec            = _mm_and_pd(velec,cutoff_mask);
 672             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 673             velecsum         = _mm_add_pd(velecsum,velec);
 674
 675             fscal            = felec;
 676
 677             fscal            = _mm_and_pd(fscal,cutoff_mask);
 678
 679             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 680
 681             /* Calculate temporary vectorial force */
 682             tx               = _mm_mul_pd(fscal,dx10);
 683             ty               = _mm_mul_pd(fscal,dy10);
 684             tz               = _mm_mul_pd(fscal,dz10);
 685
 686             /* Update vectorial force */
 687             fix1             = _mm_add_pd(fix1,tx);
 688             fiy1             = _mm_add_pd(fiy1,ty);
 689             fiz1             = _mm_add_pd(fiz1,tz);
 690
 691             fjx0             = _mm_add_pd(fjx0,tx);
 692             fjy0             = _mm_add_pd(fjy0,ty);
 693             fjz0             = _mm_add_pd(fjz0,tz);
 694
 695             }
 696
 697             /**************************
 698              * CALCULATE INTERACTIONS *
 699              **************************/
 700
 701             if (gmx_mm_any_lt(rsq20,rcutoff2))
 702             {
 703
 704             r20              = _mm_mul_pd(rsq20,rinv20);
 705
 706             /* Compute parameters for interactions between i and j atoms */
 707             qq20             = _mm_mul_pd(iq2,jq0);
 708
 709             /* EWALD ELECTROSTATICS */
 710
 711             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 712             ewrt             = _mm_mul_pd(r20,ewtabscale);
 713             ewitab           = _mm_cvttpd_epi32(ewrt);
 714             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 715             ewitab           = _mm_slli_epi32(ewitab,2);
 716             ewtabF           = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 717             ewtabD           = _mm_setzero_pd();
 718             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 719             ewtabV           = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,0) +2);
 720             ewtabFn          = _mm_setzero_pd();
 721             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 722             felec            = _mm_add_pd(ewtabF,_mm_mul_pd(eweps,ewtabD));
 723             velec            = _mm_sub_pd(ewtabV,_mm_mul_pd(_mm_mul_pd(ewtabhalfspace,eweps),_mm_add_pd(ewtabF,felec)));
 724             velec            = _mm_mul_pd(qq20,_mm_sub_pd(rinv20,velec));
 725             felec            = _mm_mul_pd(_mm_mul_pd(qq20,rinv20),_mm_sub_pd(rinvsq20,felec));
 726
 727             d                = _mm_sub_pd(r20,rswitch);
 728             d                = _mm_max_pd(d,_mm_setzero_pd());
 729             d2               = _mm_mul_pd(d,d);
 730             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)))))));
 731
 732             dsw              = _mm_mul_pd(d2,_mm_add_pd(swF2,_mm_mul_pd(d,_mm_add_pd(swF3,_mm_mul_pd(d,swF4)))));
 733
 734             /* Evaluate switch function */
 735             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
 736             felec            = _mm_sub_pd( _mm_mul_pd(felec,sw) , _mm_mul_pd(rinv20,_mm_mul_pd(velec,dsw)) );
 737             velec            = _mm_mul_pd(velec,sw);
 738             cutoff_mask      = _mm_cmplt_pd(rsq20,rcutoff2);
 739
 740             /* Update potential sum for this i atom from the interaction with this j atom. */
 741             velec            = _mm_and_pd(velec,cutoff_mask);
 742             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 743             velecsum         = _mm_add_pd(velecsum,velec);
 744
 745             fscal            = felec;
 746
 747             fscal            = _mm_and_pd(fscal,cutoff_mask);
 748
 749             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 750
 751             /* Calculate temporary vectorial force */
 752             tx               = _mm_mul_pd(fscal,dx20);
 753             ty               = _mm_mul_pd(fscal,dy20);
 754             tz               = _mm_mul_pd(fscal,dz20);
 755
 756             /* Update vectorial force */
 757             fix2             = _mm_add_pd(fix2,tx);
 758             fiy2             = _mm_add_pd(fiy2,ty);
 759             fiz2             = _mm_add_pd(fiz2,tz);
 760
 761             fjx0             = _mm_add_pd(fjx0,tx);
 762             fjy0             = _mm_add_pd(fjy0,ty);
 763             fjz0             = _mm_add_pd(fjz0,tz);
 764
 765             }
 766
 767             /**************************
 768              * CALCULATE INTERACTIONS *
 769              **************************/
 770
 771             if (gmx_mm_any_lt(rsq30,rcutoff2))
 772             {
 773
 774             r30              = _mm_mul_pd(rsq30,rinv30);
 775
 776             /* Compute parameters for interactions between i and j atoms */
 777             qq30             = _mm_mul_pd(iq3,jq0);
 778
 779             /* EWALD ELECTROSTATICS */
 780
 781             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 782             ewrt             = _mm_mul_pd(r30,ewtabscale);
 783             ewitab           = _mm_cvttpd_epi32(ewrt);
 784             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 785             ewitab           = _mm_slli_epi32(ewitab,2);
 786             ewtabF           = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 787             ewtabD           = _mm_setzero_pd();
 788             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 789             ewtabV           = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,0) +2);
 790             ewtabFn          = _mm_setzero_pd();
 791             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 792             felec            = _mm_add_pd(ewtabF,_mm_mul_pd(eweps,ewtabD));
 793             velec            = _mm_sub_pd(ewtabV,_mm_mul_pd(_mm_mul_pd(ewtabhalfspace,eweps),_mm_add_pd(ewtabF,felec)));
 794             velec            = _mm_mul_pd(qq30,_mm_sub_pd(rinv30,velec));
 795             felec            = _mm_mul_pd(_mm_mul_pd(qq30,rinv30),_mm_sub_pd(rinvsq30,felec));
 796
 797             d                = _mm_sub_pd(r30,rswitch);
 798             d                = _mm_max_pd(d,_mm_setzero_pd());
 799             d2               = _mm_mul_pd(d,d);
 800             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)))))));
 801
 802             dsw              = _mm_mul_pd(d2,_mm_add_pd(swF2,_mm_mul_pd(d,_mm_add_pd(swF3,_mm_mul_pd(d,swF4)))));
 803
 804             /* Evaluate switch function */
 805             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
 806             felec            = _mm_sub_pd( _mm_mul_pd(felec,sw) , _mm_mul_pd(rinv30,_mm_mul_pd(velec,dsw)) );
 807             velec            = _mm_mul_pd(velec,sw);
 808             cutoff_mask      = _mm_cmplt_pd(rsq30,rcutoff2);
 809
 810             /* Update potential sum for this i atom from the interaction with this j atom. */
 811             velec            = _mm_and_pd(velec,cutoff_mask);
 812             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 813             velecsum         = _mm_add_pd(velecsum,velec);
 814
 815             fscal            = felec;
 816
 817             fscal            = _mm_and_pd(fscal,cutoff_mask);
 818
 819             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 820
 821             /* Calculate temporary vectorial force */
 822             tx               = _mm_mul_pd(fscal,dx30);
 823             ty               = _mm_mul_pd(fscal,dy30);
 824             tz               = _mm_mul_pd(fscal,dz30);
 825
 826             /* Update vectorial force */
 827             fix3             = _mm_add_pd(fix3,tx);
 828             fiy3             = _mm_add_pd(fiy3,ty);
 829             fiz3             = _mm_add_pd(fiz3,tz);
 830
 831             fjx0             = _mm_add_pd(fjx0,tx);
 832             fjy0             = _mm_add_pd(fjy0,ty);
 833             fjz0             = _mm_add_pd(fjz0,tz);
 834
 835             }
 836
 837             gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,fjx0,fjy0,fjz0);
 838
 839             /* Inner loop uses 257 flops */
 840         }
 841
 842         /* End of innermost loop */
 843
 844         gmx_mm_update_iforce_4atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
 845                                               f+i_coord_offset,fshift+i_shift_offset);
 846
 847         ggid                        = gid[iidx];
 848         /* Update potential energies */
 849         gmx_mm_update_1pot_pd(velecsum,kernel_data->energygrp_elec+ggid);
 850         gmx_mm_update_1pot_pd(vvdwsum,kernel_data->energygrp_vdw+ggid);
 851
 852         /* Increment number of inner iterations */
 853         inneriter                  += j_index_end - j_index_start;
 854
 855         /* Outer loop uses 26 flops */
 856     }
 857
 858     /* Increment number of outer iterations */
 859     outeriter        += nri;
 860
 861     /* Update outer/inner flops */
 862
 863     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W4_VF,outeriter*26 + inneriter*257);
 864 }
 865 /*
 866  * Gromacs nonbonded kernel:   nb_kernel_ElecEwSw_VdwLJSw_GeomW4P1_F_sse4_1_double
 867  * Electrostatics interaction: Ewald
 868  * VdW interaction:            LennardJones
 869  * Geometry:                   Water4-Particle
 870  * Calculate force/pot:        Force
 871  */
 872 void
 873 nb_kernel_ElecEwSw_VdwLJSw_GeomW4P1_F_sse4_1_double
 874                     (t_nblist                    * gmx_restrict       nlist,
 875                      rvec                        * gmx_restrict          xx,
 876                      rvec                        * gmx_restrict          ff,
 877                      t_forcerec                  * gmx_restrict          fr,
 878                      t_mdatoms                   * gmx_restrict     mdatoms,
 879                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 880                      t_nrnb                      * gmx_restrict        nrnb)
 881 {
 882     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 883      * just 0 for non-waters.
 884      * Suffixes A,B refer to j loop unrolling done with SSE double precision, e.g. for the two different
 885      * jnr indices corresponding to data put in the four positions in the SIMD register.
 886      */
 887     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 888     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 889     int              jnrA,jnrB;
 890     int              j_coord_offsetA,j_coord_offsetB;
 891     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 892     real             rcutoff_scalar;
 893     real             *shiftvec,*fshift,*x,*f;
 894     __m128d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 895     int              vdwioffset0;
 896     __m128d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 897     int              vdwioffset1;
 898     __m128d          ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
 899     int              vdwioffset2;
 900     __m128d          ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
 901     int              vdwioffset3;
 902     __m128d          ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
 903     int              vdwjidx0A,vdwjidx0B;
 904     __m128d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 905     __m128d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 906     __m128d          dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
 907     __m128d          dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
 908     __m128d          dx30,dy30,dz30,rsq30,rinv30,rinvsq30,r30,qq30,c6_30,c12_30;
 909     __m128d          velec,felec,velecsum,facel,crf,krf,krf2;
 910     real             *charge;
 911     int              nvdwtype;
 912     __m128d          rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 913     int              *vdwtype;
 914     real             *vdwparam;
 915     __m128d          one_sixth   = _mm_set1_pd(1.0/6.0);
 916     __m128d          one_twelfth = _mm_set1_pd(1.0/12.0);
 917     __m128i          ewitab;
 918     __m128d          ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
 919     real             *ewtab;
 920     __m128d          rswitch,swV3,swV4,swV5,swF2,swF3,swF4,d,d2,sw,dsw;
 921     real             rswitch_scalar,d_scalar;
 922     __m128d          dummy_mask,cutoff_mask;
 923     __m128d          signbit   = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
 924     __m128d          one     = _mm_set1_pd(1.0);
 925     __m128d          two     = _mm_set1_pd(2.0);
 926     x                = xx[0];
 927     f                = ff[0];
 928
 929     nri              = nlist->nri;
 930     iinr             = nlist->iinr;
 931     jindex           = nlist->jindex;
 932     jjnr             = nlist->jjnr;
 933     shiftidx         = nlist->shift;
 934     gid              = nlist->gid;
 935     shiftvec         = fr->shift_vec[0];
 936     fshift           = fr->fshift[0];
 937     facel            = _mm_set1_pd(fr->epsfac);
 938     charge           = mdatoms->chargeA;
 939     nvdwtype         = fr->ntype;
 940     vdwparam         = fr->nbfp;
 941     vdwtype          = mdatoms->typeA;
 942
 943     sh_ewald         = _mm_set1_pd(fr->ic->sh_ewald);
 944     ewtab            = fr->ic->tabq_coul_FDV0;
 945     ewtabscale       = _mm_set1_pd(fr->ic->tabq_scale);
 946     ewtabhalfspace   = _mm_set1_pd(0.5/fr->ic->tabq_scale);
 947
 948     /* Setup water-specific parameters */
 949     inr              = nlist->iinr[0];
 950     iq1              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+1]));
 951     iq2              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+2]));
 952     iq3              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+3]));
 953     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 954
 955     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 956     rcutoff_scalar   = fr->rcoulomb;
 957     rcutoff          = _mm_set1_pd(rcutoff_scalar);
 958     rcutoff2         = _mm_mul_pd(rcutoff,rcutoff);
 959
 960     rswitch_scalar   = fr->rcoulomb_switch;
 961     rswitch          = _mm_set1_pd(rswitch_scalar);
 962     /* Setup switch parameters */
 963     d_scalar         = rcutoff_scalar-rswitch_scalar;
 964     d                = _mm_set1_pd(d_scalar);
 965     swV3             = _mm_set1_pd(-10.0/(d_scalar*d_scalar*d_scalar));
 966     swV4             = _mm_set1_pd( 15.0/(d_scalar*d_scalar*d_scalar*d_scalar));
 967     swV5             = _mm_set1_pd( -6.0/(d_scalar*d_scalar*d_scalar*d_scalar*d_scalar));
 968     swF2             = _mm_set1_pd(-30.0/(d_scalar*d_scalar*d_scalar));
 969     swF3             = _mm_set1_pd( 60.0/(d_scalar*d_scalar*d_scalar*d_scalar));
 970     swF4             = _mm_set1_pd(-30.0/(d_scalar*d_scalar*d_scalar*d_scalar*d_scalar));
 971
 972     /* Avoid stupid compiler warnings */
 973     jnrA = jnrB = 0;
 974     j_coord_offsetA = 0;
 975     j_coord_offsetB = 0;
 976
 977     outeriter        = 0;
 978     inneriter        = 0;
 979
 980     /* Start outer loop over neighborlists */
 981     for(iidx=0; iidx<nri; iidx++)
 982     {
 983         /* Load shift vector for this list */
 984         i_shift_offset   = DIM*shiftidx[iidx];
 985
 986         /* Load limits for loop over neighbors */
 987         j_index_start    = jindex[iidx];
 988         j_index_end      = jindex[iidx+1];
 989
 990         /* Get outer coordinate index */
 991         inr              = iinr[iidx];
 992         i_coord_offset   = DIM*inr;
 993
 994         /* Load i particle coords and add shift vector */
 995         gmx_mm_load_shift_and_4rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
 996                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
 997
 998         fix0             = _mm_setzero_pd();
 999         fiy0             = _mm_setzero_pd();
1000         fiz0             = _mm_setzero_pd();
1001         fix1             = _mm_setzero_pd();
1002         fiy1             = _mm_setzero_pd();
1003         fiz1             = _mm_setzero_pd();
1004         fix2             = _mm_setzero_pd();
1005         fiy2             = _mm_setzero_pd();
1006         fiz2             = _mm_setzero_pd();
1007         fix3             = _mm_setzero_pd();
1008         fiy3             = _mm_setzero_pd();
1009         fiz3             = _mm_setzero_pd();
1010
1011         /* Start inner kernel loop */
1012         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
1013         {
1014
1015             /* Get j neighbor index, and coordinate index */
1016             jnrA             = jjnr[jidx];
1017             jnrB             = jjnr[jidx+1];
1018             j_coord_offsetA  = DIM*jnrA;
1019             j_coord_offsetB  = DIM*jnrB;
1020
1021             /* load j atom coordinates */
1022             gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
1023                                               &jx0,&jy0,&jz0);
1024
1025             /* Calculate displacement vector */
1026             dx00             = _mm_sub_pd(ix0,jx0);
1027             dy00             = _mm_sub_pd(iy0,jy0);
1028             dz00             = _mm_sub_pd(iz0,jz0);
1029             dx10             = _mm_sub_pd(ix1,jx0);
1030             dy10             = _mm_sub_pd(iy1,jy0);
1031             dz10             = _mm_sub_pd(iz1,jz0);
1032             dx20             = _mm_sub_pd(ix2,jx0);
1033             dy20             = _mm_sub_pd(iy2,jy0);
1034             dz20             = _mm_sub_pd(iz2,jz0);
1035             dx30             = _mm_sub_pd(ix3,jx0);
1036             dy30             = _mm_sub_pd(iy3,jy0);
1037             dz30             = _mm_sub_pd(iz3,jz0);
1038
1039             /* Calculate squared distance and things based on it */
1040             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
1041             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
1042             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
1043             rsq30            = gmx_mm_calc_rsq_pd(dx30,dy30,dz30);
1044
1045             rinv00           = gmx_mm_invsqrt_pd(rsq00);
1046             rinv10           = gmx_mm_invsqrt_pd(rsq10);
1047             rinv20           = gmx_mm_invsqrt_pd(rsq20);
1048             rinv30           = gmx_mm_invsqrt_pd(rsq30);
1049
1050             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
1051             rinvsq10         = _mm_mul_pd(rinv10,rinv10);
1052             rinvsq20         = _mm_mul_pd(rinv20,rinv20);
1053             rinvsq30         = _mm_mul_pd(rinv30,rinv30);
1054
1055             /* Load parameters for j particles */
1056             jq0              = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
1057             vdwjidx0A        = 2*vdwtype[jnrA+0];
1058             vdwjidx0B        = 2*vdwtype[jnrB+0];
1059
1060             fjx0             = _mm_setzero_pd();
1061             fjy0             = _mm_setzero_pd();
1062             fjz0             = _mm_setzero_pd();
1063
1064             /**************************
1065              * CALCULATE INTERACTIONS *
1066              **************************/
1067
1068             if (gmx_mm_any_lt(rsq00,rcutoff2))
1069             {
1070
1071             r00              = _mm_mul_pd(rsq00,rinv00);
1072
1073             /* Compute parameters for interactions between i and j atoms */
1074             gmx_mm_load_2pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,
1075                                          vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
1076
1077             /* LENNARD-JONES DISPERSION/REPULSION */
1078
1079             rinvsix          = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
1080             vvdw6            = _mm_mul_pd(c6_00,rinvsix);
1081             vvdw12           = _mm_mul_pd(c12_00,_mm_mul_pd(rinvsix,rinvsix));
1082             vvdw             = _mm_sub_pd( _mm_mul_pd(vvdw12,one_twelfth) , _mm_mul_pd(vvdw6,one_sixth) );
1083             fvdw             = _mm_mul_pd(_mm_sub_pd(vvdw12,vvdw6),rinvsq00);
1084
1085             d                = _mm_sub_pd(r00,rswitch);
1086             d                = _mm_max_pd(d,_mm_setzero_pd());
1087             d2               = _mm_mul_pd(d,d);
1088             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)))))));
1089
1090             dsw              = _mm_mul_pd(d2,_mm_add_pd(swF2,_mm_mul_pd(d,_mm_add_pd(swF3,_mm_mul_pd(d,swF4)))));
1091
1092             /* Evaluate switch function */
1093             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
1094             fvdw             = _mm_sub_pd( _mm_mul_pd(fvdw,sw) , _mm_mul_pd(rinv00,_mm_mul_pd(vvdw,dsw)) );
1095             cutoff_mask      = _mm_cmplt_pd(rsq00,rcutoff2);
1096
1097             fscal            = fvdw;
1098
1099             fscal            = _mm_and_pd(fscal,cutoff_mask);
1100
1101             /* Calculate temporary vectorial force */
1102             tx               = _mm_mul_pd(fscal,dx00);
1103             ty               = _mm_mul_pd(fscal,dy00);
1104             tz               = _mm_mul_pd(fscal,dz00);
1105
1106             /* Update vectorial force */
1107             fix0             = _mm_add_pd(fix0,tx);
1108             fiy0             = _mm_add_pd(fiy0,ty);
1109             fiz0             = _mm_add_pd(fiz0,tz);
1110
1111             fjx0             = _mm_add_pd(fjx0,tx);
1112             fjy0             = _mm_add_pd(fjy0,ty);
1113             fjz0             = _mm_add_pd(fjz0,tz);
1114
1115             }
1116
1117             /**************************
1118              * CALCULATE INTERACTIONS *
1119              **************************/
1120
1121             if (gmx_mm_any_lt(rsq10,rcutoff2))
1122             {
1123
1124             r10              = _mm_mul_pd(rsq10,rinv10);
1125
1126             /* Compute parameters for interactions between i and j atoms */
1127             qq10             = _mm_mul_pd(iq1,jq0);
1128
1129             /* EWALD ELECTROSTATICS */
1130
1131             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1132             ewrt             = _mm_mul_pd(r10,ewtabscale);
1133             ewitab           = _mm_cvttpd_epi32(ewrt);
1134             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
1135             ewitab           = _mm_slli_epi32(ewitab,2);
1136             ewtabF           = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,0) );
1137             ewtabD           = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,1) );
1138             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
1139             ewtabV           = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,0) +2);
1140             ewtabFn          = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,1) +2);
1141             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
1142             felec            = _mm_add_pd(ewtabF,_mm_mul_pd(eweps,ewtabD));
1143             velec            = _mm_sub_pd(ewtabV,_mm_mul_pd(_mm_mul_pd(ewtabhalfspace,eweps),_mm_add_pd(ewtabF,felec)));
1144             velec            = _mm_mul_pd(qq10,_mm_sub_pd(rinv10,velec));
1145             felec            = _mm_mul_pd(_mm_mul_pd(qq10,rinv10),_mm_sub_pd(rinvsq10,felec));
1146
1147             d                = _mm_sub_pd(r10,rswitch);
1148             d                = _mm_max_pd(d,_mm_setzero_pd());
1149             d2               = _mm_mul_pd(d,d);
1150             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)))))));
1151
1152             dsw              = _mm_mul_pd(d2,_mm_add_pd(swF2,_mm_mul_pd(d,_mm_add_pd(swF3,_mm_mul_pd(d,swF4)))));
1153
1154             /* Evaluate switch function */
1155             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
1156             felec            = _mm_sub_pd( _mm_mul_pd(felec,sw) , _mm_mul_pd(rinv10,_mm_mul_pd(velec,dsw)) );
1157             cutoff_mask      = _mm_cmplt_pd(rsq10,rcutoff2);
1158
1159             fscal            = felec;
1160
1161             fscal            = _mm_and_pd(fscal,cutoff_mask);
1162
1163             /* Calculate temporary vectorial force */
1164             tx               = _mm_mul_pd(fscal,dx10);
1165             ty               = _mm_mul_pd(fscal,dy10);
1166             tz               = _mm_mul_pd(fscal,dz10);
1167
1168             /* Update vectorial force */
1169             fix1             = _mm_add_pd(fix1,tx);
1170             fiy1             = _mm_add_pd(fiy1,ty);
1171             fiz1             = _mm_add_pd(fiz1,tz);
1172
1173             fjx0             = _mm_add_pd(fjx0,tx);
1174             fjy0             = _mm_add_pd(fjy0,ty);
1175             fjz0             = _mm_add_pd(fjz0,tz);
1176
1177             }
1178
1179             /**************************
1180              * CALCULATE INTERACTIONS *
1181              **************************/
1182
1183             if (gmx_mm_any_lt(rsq20,rcutoff2))
1184             {
1185
1186             r20              = _mm_mul_pd(rsq20,rinv20);
1187
1188             /* Compute parameters for interactions between i and j atoms */
1189             qq20             = _mm_mul_pd(iq2,jq0);
1190
1191             /* EWALD ELECTROSTATICS */
1192
1193             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1194             ewrt             = _mm_mul_pd(r20,ewtabscale);
1195             ewitab           = _mm_cvttpd_epi32(ewrt);
1196             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
1197             ewitab           = _mm_slli_epi32(ewitab,2);
1198             ewtabF           = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,0) );
1199             ewtabD           = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,1) );
1200             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
1201             ewtabV           = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,0) +2);
1202             ewtabFn          = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,1) +2);
1203             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
1204             felec            = _mm_add_pd(ewtabF,_mm_mul_pd(eweps,ewtabD));
1205             velec            = _mm_sub_pd(ewtabV,_mm_mul_pd(_mm_mul_pd(ewtabhalfspace,eweps),_mm_add_pd(ewtabF,felec)));
1206             velec            = _mm_mul_pd(qq20,_mm_sub_pd(rinv20,velec));
1207             felec            = _mm_mul_pd(_mm_mul_pd(qq20,rinv20),_mm_sub_pd(rinvsq20,felec));
1208
1209             d                = _mm_sub_pd(r20,rswitch);
1210             d                = _mm_max_pd(d,_mm_setzero_pd());
1211             d2               = _mm_mul_pd(d,d);
1212             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)))))));
1213
1214             dsw              = _mm_mul_pd(d2,_mm_add_pd(swF2,_mm_mul_pd(d,_mm_add_pd(swF3,_mm_mul_pd(d,swF4)))));
1215
1216             /* Evaluate switch function */
1217             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
1218             felec            = _mm_sub_pd( _mm_mul_pd(felec,sw) , _mm_mul_pd(rinv20,_mm_mul_pd(velec,dsw)) );
1219             cutoff_mask      = _mm_cmplt_pd(rsq20,rcutoff2);
1220
1221             fscal            = felec;
1222
1223             fscal            = _mm_and_pd(fscal,cutoff_mask);
1224
1225             /* Calculate temporary vectorial force */
1226             tx               = _mm_mul_pd(fscal,dx20);
1227             ty               = _mm_mul_pd(fscal,dy20);
1228             tz               = _mm_mul_pd(fscal,dz20);
1229
1230             /* Update vectorial force */
1231             fix2             = _mm_add_pd(fix2,tx);
1232             fiy2             = _mm_add_pd(fiy2,ty);
1233             fiz2             = _mm_add_pd(fiz2,tz);
1234
1235             fjx0             = _mm_add_pd(fjx0,tx);
1236             fjy0             = _mm_add_pd(fjy0,ty);
1237             fjz0             = _mm_add_pd(fjz0,tz);
1238
1239             }
1240
1241             /**************************
1242              * CALCULATE INTERACTIONS *
1243              **************************/
1244
1245             if (gmx_mm_any_lt(rsq30,rcutoff2))
1246             {
1247
1248             r30              = _mm_mul_pd(rsq30,rinv30);
1249
1250             /* Compute parameters for interactions between i and j atoms */
1251             qq30             = _mm_mul_pd(iq3,jq0);
1252
1253             /* EWALD ELECTROSTATICS */
1254
1255             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1256             ewrt             = _mm_mul_pd(r30,ewtabscale);
1257             ewitab           = _mm_cvttpd_epi32(ewrt);
1258             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
1259             ewitab           = _mm_slli_epi32(ewitab,2);
1260             ewtabF           = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,0) );
1261             ewtabD           = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,1) );
1262             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
1263             ewtabV           = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,0) +2);
1264             ewtabFn          = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,1) +2);
1265             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
1266             felec            = _mm_add_pd(ewtabF,_mm_mul_pd(eweps,ewtabD));
1267             velec            = _mm_sub_pd(ewtabV,_mm_mul_pd(_mm_mul_pd(ewtabhalfspace,eweps),_mm_add_pd(ewtabF,felec)));
1268             velec            = _mm_mul_pd(qq30,_mm_sub_pd(rinv30,velec));
1269             felec            = _mm_mul_pd(_mm_mul_pd(qq30,rinv30),_mm_sub_pd(rinvsq30,felec));
1270
1271             d                = _mm_sub_pd(r30,rswitch);
1272             d                = _mm_max_pd(d,_mm_setzero_pd());
1273             d2               = _mm_mul_pd(d,d);
1274             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)))))));
1275
1276             dsw              = _mm_mul_pd(d2,_mm_add_pd(swF2,_mm_mul_pd(d,_mm_add_pd(swF3,_mm_mul_pd(d,swF4)))));
1277
1278             /* Evaluate switch function */
1279             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
1280             felec            = _mm_sub_pd( _mm_mul_pd(felec,sw) , _mm_mul_pd(rinv30,_mm_mul_pd(velec,dsw)) );
1281             cutoff_mask      = _mm_cmplt_pd(rsq30,rcutoff2);
1282
1283             fscal            = felec;
1284
1285             fscal            = _mm_and_pd(fscal,cutoff_mask);
1286
1287             /* Calculate temporary vectorial force */
1288             tx               = _mm_mul_pd(fscal,dx30);
1289             ty               = _mm_mul_pd(fscal,dy30);
1290             tz               = _mm_mul_pd(fscal,dz30);
1291
1292             /* Update vectorial force */
1293             fix3             = _mm_add_pd(fix3,tx);
1294             fiy3             = _mm_add_pd(fiy3,ty);
1295             fiz3             = _mm_add_pd(fiz3,tz);
1296
1297             fjx0             = _mm_add_pd(fjx0,tx);
1298             fjy0             = _mm_add_pd(fjy0,ty);
1299             fjz0             = _mm_add_pd(fjz0,tz);
1300
1301             }
1302
1303             gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,fjx0,fjy0,fjz0);
1304
1305             /* Inner loop uses 245 flops */
1306         }
1307
1308         if(jidx<j_index_end)
1309         {
1310
1311             jnrA             = jjnr[jidx];
1312             j_coord_offsetA  = DIM*jnrA;
1313
1314             /* load j atom coordinates */
1315             gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
1316                                               &jx0,&jy0,&jz0);
1317
1318             /* Calculate displacement vector */
1319             dx00             = _mm_sub_pd(ix0,jx0);
1320             dy00             = _mm_sub_pd(iy0,jy0);
1321             dz00             = _mm_sub_pd(iz0,jz0);
1322             dx10             = _mm_sub_pd(ix1,jx0);
1323             dy10             = _mm_sub_pd(iy1,jy0);
1324             dz10             = _mm_sub_pd(iz1,jz0);
1325             dx20             = _mm_sub_pd(ix2,jx0);
1326             dy20             = _mm_sub_pd(iy2,jy0);
1327             dz20             = _mm_sub_pd(iz2,jz0);
1328             dx30             = _mm_sub_pd(ix3,jx0);
1329             dy30             = _mm_sub_pd(iy3,jy0);
1330             dz30             = _mm_sub_pd(iz3,jz0);
1331
1332             /* Calculate squared distance and things based on it */
1333             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
1334             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
1335             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
1336             rsq30            = gmx_mm_calc_rsq_pd(dx30,dy30,dz30);
1337
1338             rinv00           = gmx_mm_invsqrt_pd(rsq00);
1339             rinv10           = gmx_mm_invsqrt_pd(rsq10);
1340             rinv20           = gmx_mm_invsqrt_pd(rsq20);
1341             rinv30           = gmx_mm_invsqrt_pd(rsq30);
1342
1343             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
1344             rinvsq10         = _mm_mul_pd(rinv10,rinv10);
1345             rinvsq20         = _mm_mul_pd(rinv20,rinv20);
1346             rinvsq30         = _mm_mul_pd(rinv30,rinv30);
1347
1348             /* Load parameters for j particles */
1349             jq0              = _mm_load_sd(charge+jnrA+0);
1350             vdwjidx0A        = 2*vdwtype[jnrA+0];
1351
1352             fjx0             = _mm_setzero_pd();
1353             fjy0             = _mm_setzero_pd();
1354             fjz0             = _mm_setzero_pd();
1355
1356             /**************************
1357              * CALCULATE INTERACTIONS *
1358              **************************/
1359
1360             if (gmx_mm_any_lt(rsq00,rcutoff2))
1361             {
1362
1363             r00              = _mm_mul_pd(rsq00,rinv00);
1364
1365             /* Compute parameters for interactions between i and j atoms */
1366             gmx_mm_load_1pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,&c6_00,&c12_00);
1367
1368             /* LENNARD-JONES DISPERSION/REPULSION */
1369
1370             rinvsix          = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
1371             vvdw6            = _mm_mul_pd(c6_00,rinvsix);
1372             vvdw12           = _mm_mul_pd(c12_00,_mm_mul_pd(rinvsix,rinvsix));
1373             vvdw             = _mm_sub_pd( _mm_mul_pd(vvdw12,one_twelfth) , _mm_mul_pd(vvdw6,one_sixth) );
1374             fvdw             = _mm_mul_pd(_mm_sub_pd(vvdw12,vvdw6),rinvsq00);
1375
1376             d                = _mm_sub_pd(r00,rswitch);
1377             d                = _mm_max_pd(d,_mm_setzero_pd());
1378             d2               = _mm_mul_pd(d,d);
1379             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)))))));
1380
1381             dsw              = _mm_mul_pd(d2,_mm_add_pd(swF2,_mm_mul_pd(d,_mm_add_pd(swF3,_mm_mul_pd(d,swF4)))));
1382
1383             /* Evaluate switch function */
1384             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
1385             fvdw             = _mm_sub_pd( _mm_mul_pd(fvdw,sw) , _mm_mul_pd(rinv00,_mm_mul_pd(vvdw,dsw)) );
1386             cutoff_mask      = _mm_cmplt_pd(rsq00,rcutoff2);
1387
1388             fscal            = fvdw;
1389
1390             fscal            = _mm_and_pd(fscal,cutoff_mask);
1391
1392             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1393
1394             /* Calculate temporary vectorial force */
1395             tx               = _mm_mul_pd(fscal,dx00);
1396             ty               = _mm_mul_pd(fscal,dy00);
1397             tz               = _mm_mul_pd(fscal,dz00);
1398
1399             /* Update vectorial force */
1400             fix0             = _mm_add_pd(fix0,tx);
1401             fiy0             = _mm_add_pd(fiy0,ty);
1402             fiz0             = _mm_add_pd(fiz0,tz);
1403
1404             fjx0             = _mm_add_pd(fjx0,tx);
1405             fjy0             = _mm_add_pd(fjy0,ty);
1406             fjz0             = _mm_add_pd(fjz0,tz);
1407
1408             }
1409
1410             /**************************
1411              * CALCULATE INTERACTIONS *
1412              **************************/
1413
1414             if (gmx_mm_any_lt(rsq10,rcutoff2))
1415             {
1416
1417             r10              = _mm_mul_pd(rsq10,rinv10);
1418
1419             /* Compute parameters for interactions between i and j atoms */
1420             qq10             = _mm_mul_pd(iq1,jq0);
1421
1422             /* EWALD ELECTROSTATICS */
1423
1424             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1425             ewrt             = _mm_mul_pd(r10,ewtabscale);
1426             ewitab           = _mm_cvttpd_epi32(ewrt);
1427             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
1428             ewitab           = _mm_slli_epi32(ewitab,2);
1429             ewtabF           = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,0) );
1430             ewtabD           = _mm_setzero_pd();
1431             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
1432             ewtabV           = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,0) +2);
1433             ewtabFn          = _mm_setzero_pd();
1434             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
1435             felec            = _mm_add_pd(ewtabF,_mm_mul_pd(eweps,ewtabD));
1436             velec            = _mm_sub_pd(ewtabV,_mm_mul_pd(_mm_mul_pd(ewtabhalfspace,eweps),_mm_add_pd(ewtabF,felec)));
1437             velec            = _mm_mul_pd(qq10,_mm_sub_pd(rinv10,velec));
1438             felec            = _mm_mul_pd(_mm_mul_pd(qq10,rinv10),_mm_sub_pd(rinvsq10,felec));
1439
1440             d                = _mm_sub_pd(r10,rswitch);
1441             d                = _mm_max_pd(d,_mm_setzero_pd());
1442             d2               = _mm_mul_pd(d,d);
1443             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)))))));
1444
1445             dsw              = _mm_mul_pd(d2,_mm_add_pd(swF2,_mm_mul_pd(d,_mm_add_pd(swF3,_mm_mul_pd(d,swF4)))));
1446
1447             /* Evaluate switch function */
1448             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
1449             felec            = _mm_sub_pd( _mm_mul_pd(felec,sw) , _mm_mul_pd(rinv10,_mm_mul_pd(velec,dsw)) );
1450             cutoff_mask      = _mm_cmplt_pd(rsq10,rcutoff2);
1451
1452             fscal            = felec;
1453
1454             fscal            = _mm_and_pd(fscal,cutoff_mask);
1455
1456             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1457
1458             /* Calculate temporary vectorial force */
1459             tx               = _mm_mul_pd(fscal,dx10);
1460             ty               = _mm_mul_pd(fscal,dy10);
1461             tz               = _mm_mul_pd(fscal,dz10);
1462
1463             /* Update vectorial force */
1464             fix1             = _mm_add_pd(fix1,tx);
1465             fiy1             = _mm_add_pd(fiy1,ty);
1466             fiz1             = _mm_add_pd(fiz1,tz);
1467
1468             fjx0             = _mm_add_pd(fjx0,tx);
1469             fjy0             = _mm_add_pd(fjy0,ty);
1470             fjz0             = _mm_add_pd(fjz0,tz);
1471
1472             }
1473
1474             /**************************
1475              * CALCULATE INTERACTIONS *
1476              **************************/
1477
1478             if (gmx_mm_any_lt(rsq20,rcutoff2))
1479             {
1480
1481             r20              = _mm_mul_pd(rsq20,rinv20);
1482
1483             /* Compute parameters for interactions between i and j atoms */
1484             qq20             = _mm_mul_pd(iq2,jq0);
1485
1486             /* EWALD ELECTROSTATICS */
1487
1488             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1489             ewrt             = _mm_mul_pd(r20,ewtabscale);
1490             ewitab           = _mm_cvttpd_epi32(ewrt);
1491             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
1492             ewitab           = _mm_slli_epi32(ewitab,2);
1493             ewtabF           = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,0) );
1494             ewtabD           = _mm_setzero_pd();
1495             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
1496             ewtabV           = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,0) +2);
1497             ewtabFn          = _mm_setzero_pd();
1498             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
1499             felec            = _mm_add_pd(ewtabF,_mm_mul_pd(eweps,ewtabD));
1500             velec            = _mm_sub_pd(ewtabV,_mm_mul_pd(_mm_mul_pd(ewtabhalfspace,eweps),_mm_add_pd(ewtabF,felec)));
1501             velec            = _mm_mul_pd(qq20,_mm_sub_pd(rinv20,velec));
1502             felec            = _mm_mul_pd(_mm_mul_pd(qq20,rinv20),_mm_sub_pd(rinvsq20,felec));
1503
1504             d                = _mm_sub_pd(r20,rswitch);
1505             d                = _mm_max_pd(d,_mm_setzero_pd());
1506             d2               = _mm_mul_pd(d,d);
1507             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)))))));
1508
1509             dsw              = _mm_mul_pd(d2,_mm_add_pd(swF2,_mm_mul_pd(d,_mm_add_pd(swF3,_mm_mul_pd(d,swF4)))));
1510
1511             /* Evaluate switch function */
1512             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
1513             felec            = _mm_sub_pd( _mm_mul_pd(felec,sw) , _mm_mul_pd(rinv20,_mm_mul_pd(velec,dsw)) );
1514             cutoff_mask      = _mm_cmplt_pd(rsq20,rcutoff2);
1515
1516             fscal            = felec;
1517
1518             fscal            = _mm_and_pd(fscal,cutoff_mask);
1519
1520             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1521
1522             /* Calculate temporary vectorial force */
1523             tx               = _mm_mul_pd(fscal,dx20);
1524             ty               = _mm_mul_pd(fscal,dy20);
1525             tz               = _mm_mul_pd(fscal,dz20);
1526
1527             /* Update vectorial force */
1528             fix2             = _mm_add_pd(fix2,tx);
1529             fiy2             = _mm_add_pd(fiy2,ty);
1530             fiz2             = _mm_add_pd(fiz2,tz);
1531
1532             fjx0             = _mm_add_pd(fjx0,tx);
1533             fjy0             = _mm_add_pd(fjy0,ty);
1534             fjz0             = _mm_add_pd(fjz0,tz);
1535
1536             }
1537
1538             /**************************
1539              * CALCULATE INTERACTIONS *
1540              **************************/
1541
1542             if (gmx_mm_any_lt(rsq30,rcutoff2))
1543             {
1544
1545             r30              = _mm_mul_pd(rsq30,rinv30);
1546
1547             /* Compute parameters for interactions between i and j atoms */
1548             qq30             = _mm_mul_pd(iq3,jq0);
1549
1550             /* EWALD ELECTROSTATICS */
1551
1552             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1553             ewrt             = _mm_mul_pd(r30,ewtabscale);
1554             ewitab           = _mm_cvttpd_epi32(ewrt);
1555             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
1556             ewitab           = _mm_slli_epi32(ewitab,2);
1557             ewtabF           = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,0) );
1558             ewtabD           = _mm_setzero_pd();
1559             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
1560             ewtabV           = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,0) +2);
1561             ewtabFn          = _mm_setzero_pd();
1562             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
1563             felec            = _mm_add_pd(ewtabF,_mm_mul_pd(eweps,ewtabD));
1564             velec            = _mm_sub_pd(ewtabV,_mm_mul_pd(_mm_mul_pd(ewtabhalfspace,eweps),_mm_add_pd(ewtabF,felec)));
1565             velec            = _mm_mul_pd(qq30,_mm_sub_pd(rinv30,velec));
1566             felec            = _mm_mul_pd(_mm_mul_pd(qq30,rinv30),_mm_sub_pd(rinvsq30,felec));
1567
1568             d                = _mm_sub_pd(r30,rswitch);
1569             d                = _mm_max_pd(d,_mm_setzero_pd());
1570             d2               = _mm_mul_pd(d,d);
1571             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)))))));
1572
1573             dsw              = _mm_mul_pd(d2,_mm_add_pd(swF2,_mm_mul_pd(d,_mm_add_pd(swF3,_mm_mul_pd(d,swF4)))));
1574
1575             /* Evaluate switch function */
1576             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
1577             felec            = _mm_sub_pd( _mm_mul_pd(felec,sw) , _mm_mul_pd(rinv30,_mm_mul_pd(velec,dsw)) );
1578             cutoff_mask      = _mm_cmplt_pd(rsq30,rcutoff2);
1579
1580             fscal            = felec;
1581
1582             fscal            = _mm_and_pd(fscal,cutoff_mask);
1583
1584             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1585
1586             /* Calculate temporary vectorial force */
1587             tx               = _mm_mul_pd(fscal,dx30);
1588             ty               = _mm_mul_pd(fscal,dy30);
1589             tz               = _mm_mul_pd(fscal,dz30);
1590
1591             /* Update vectorial force */
1592             fix3             = _mm_add_pd(fix3,tx);
1593             fiy3             = _mm_add_pd(fiy3,ty);
1594             fiz3             = _mm_add_pd(fiz3,tz);
1595
1596             fjx0             = _mm_add_pd(fjx0,tx);
1597             fjy0             = _mm_add_pd(fjy0,ty);
1598             fjz0             = _mm_add_pd(fjz0,tz);
1599
1600             }
1601
1602             gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,fjx0,fjy0,fjz0);
1603
1604             /* Inner loop uses 245 flops */
1605         }
1606
1607         /* End of innermost loop */
1608
1609         gmx_mm_update_iforce_4atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
1610                                               f+i_coord_offset,fshift+i_shift_offset);
1611
1612         /* Increment number of inner iterations */
1613         inneriter                  += j_index_end - j_index_start;
1614
1615         /* Outer loop uses 24 flops */
1616     }
1617
1618     /* Increment number of outer iterations */
1619     outeriter        += nri;
1620
1621     /* Update outer/inner flops */
1622
1623     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W4_F,outeriter*24 + inneriter*245);
1624 }