src/gromacs/gmxlib/nonbonded/nb_kernel_sse2_double/nb_kernel_ElecEwSh_VdwLJEwSh_GeomW3P1_sse2_double.c

   1 /*
   2  * This file is part of the GROMACS molecular simulation package.
   3  *
   4  * Copyright (c) 2012,2013,2014, by the GROMACS development team, led by
   5  * Mark Abraham, David van der Spoel, Berk Hess, and Erik Lindahl,
   6  * and including many others, as listed in the AUTHORS file in the
   7  * top-level source directory and at http://www.gromacs.org.
   8  *
   9  * GROMACS is free software; you can redistribute it and/or
  10  * modify it under the terms of the GNU Lesser General Public License
  11  * as published by the Free Software Foundation; either version 2.1
  12  * of the License, or (at your option) any later version.
  13  *
  14  * GROMACS is distributed in the hope that it will be useful,
  15  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  16  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  17  * Lesser General Public License for more details.
  18  *
  19  * You should have received a copy of the GNU Lesser General Public
  20  * License along with GROMACS; if not, see
  21  * http://www.gnu.org/licenses, or write to the Free Software Foundation,
  22  * Inc., 51 Franklin Street, Fifth Floor, Boston, MA  02110-1301  USA.
  23  *
  24  * If you want to redistribute modifications to GROMACS, please
  25  * consider that scientific software is very special. Version
  26  * control is crucial - bugs must be traceable. We will be happy to
  27  * consider code for inclusion in the official distribution, but
  28  * derived work must not be called official GROMACS. Details are found
  29  * in the README & COPYING files - if they are missing, get the
  30  * official version at http://www.gromacs.org.
  31  *
  32  * To help us fund GROMACS development, we humbly ask that you cite
  33  * the research papers on the package. Check out http://www.gromacs.org.
  34  */
  35 /*
  36  * Note: this file was generated by the GROMACS sse2_double kernel generator.
  37  */
  38 #include "config.h"
  39
  40 #include <math.h>
  41
  42 #include "../nb_kernel.h"
  43 #include "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_ElecEwSh_VdwLJEwSh_GeomW3P1_VF_sse2_double
  52  * Electrostatics interaction: Ewald
  53  * VdW interaction:            LJEwald
  54  * Geometry:                   Water3-Particle
  55  * Calculate force/pot:        PotentialAndForce
  56  */
  57 void
  58 nb_kernel_ElecEwSh_VdwLJEwSh_GeomW3P1_VF_sse2_double
  59                     (t_nblist                    * gmx_restrict       nlist,
  60                      rvec                        * gmx_restrict          xx,
  61                      rvec                        * gmx_restrict          ff,
  62                      t_forcerec                  * gmx_restrict          fr,
  63                      t_mdatoms                   * gmx_restrict     mdatoms,
  64                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
  65                      t_nrnb                      * gmx_restrict        nrnb)
  66 {
  67     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
  68      * just 0 for non-waters.
  69      * Suffixes A,B refer to j loop unrolling done with SSE double precision, e.g. for the two different
  70      * jnr indices corresponding to data put in the four positions in the SIMD register.
  71      */
  72     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
  73     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
  74     int              jnrA,jnrB;
  75     int              j_coord_offsetA,j_coord_offsetB;
  76     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
  77     real             rcutoff_scalar;
  78     real             *shiftvec,*fshift,*x,*f;
  79     __m128d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
  80     int              vdwioffset0;
  81     __m128d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
  82     int              vdwioffset1;
  83     __m128d          ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
  84     int              vdwioffset2;
  85     __m128d          ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
  86     int              vdwjidx0A,vdwjidx0B;
  87     __m128d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  88     __m128d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  89     __m128d          dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
  90     __m128d          dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
  91     __m128d          velec,felec,velecsum,facel,crf,krf,krf2;
  92     real             *charge;
  93     int              nvdwtype;
  94     __m128d          rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
  95     int              *vdwtype;
  96     real             *vdwparam;
  97     __m128d          one_sixth   = _mm_set1_pd(1.0/6.0);
  98     __m128d          one_twelfth = _mm_set1_pd(1.0/12.0);
  99     __m128d           c6grid_00;
 100     __m128d           c6grid_10;
 101     __m128d           c6grid_20;
 102     __m128d           ewclj,ewclj2,ewclj6,ewcljrsq,poly,exponent,f6A,f6B,sh_lj_ewald;
 103     real             *vdwgridparam;
 104     __m128d           one_half = _mm_set1_pd(0.5);
 105     __m128d           minus_one = _mm_set1_pd(-1.0);
 106     __m128i          ewitab;
 107     __m128d          ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
 108     real             *ewtab;
 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     vdwgridparam     = fr->ljpme_c6grid;
 130     sh_lj_ewald      = _mm_set1_pd(fr->ic->sh_lj_ewald);
 131     ewclj            = _mm_set1_pd(fr->ewaldcoeff_lj);
 132     ewclj2           = _mm_mul_pd(minus_one,_mm_mul_pd(ewclj,ewclj));
 133
 134     sh_ewald         = _mm_set1_pd(fr->ic->sh_ewald);
 135     ewtab            = fr->ic->tabq_coul_FDV0;
 136     ewtabscale       = _mm_set1_pd(fr->ic->tabq_scale);
 137     ewtabhalfspace   = _mm_set1_pd(0.5/fr->ic->tabq_scale);
 138
 139     /* Setup water-specific parameters */
 140     inr              = nlist->iinr[0];
 141     iq0              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+0]));
 142     iq1              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+1]));
 143     iq2              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+2]));
 144     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 145
 146     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 147     rcutoff_scalar   = fr->rcoulomb;
 148     rcutoff          = _mm_set1_pd(rcutoff_scalar);
 149     rcutoff2         = _mm_mul_pd(rcutoff,rcutoff);
 150
 151     sh_vdw_invrcut6  = _mm_set1_pd(fr->ic->sh_invrc6);
 152     rvdw             = _mm_set1_pd(fr->rvdw);
 153
 154     /* Avoid stupid compiler warnings */
 155     jnrA = jnrB = 0;
 156     j_coord_offsetA = 0;
 157     j_coord_offsetB = 0;
 158
 159     outeriter        = 0;
 160     inneriter        = 0;
 161
 162     /* Start outer loop over neighborlists */
 163     for(iidx=0; iidx<nri; iidx++)
 164     {
 165         /* Load shift vector for this list */
 166         i_shift_offset   = DIM*shiftidx[iidx];
 167
 168         /* Load limits for loop over neighbors */
 169         j_index_start    = jindex[iidx];
 170         j_index_end      = jindex[iidx+1];
 171
 172         /* Get outer coordinate index */
 173         inr              = iinr[iidx];
 174         i_coord_offset   = DIM*inr;
 175
 176         /* Load i particle coords and add shift vector */
 177         gmx_mm_load_shift_and_3rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
 178                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 179
 180         fix0             = _mm_setzero_pd();
 181         fiy0             = _mm_setzero_pd();
 182         fiz0             = _mm_setzero_pd();
 183         fix1             = _mm_setzero_pd();
 184         fiy1             = _mm_setzero_pd();
 185         fiz1             = _mm_setzero_pd();
 186         fix2             = _mm_setzero_pd();
 187         fiy2             = _mm_setzero_pd();
 188         fiz2             = _mm_setzero_pd();
 189
 190         /* Reset potential sums */
 191         velecsum         = _mm_setzero_pd();
 192         vvdwsum          = _mm_setzero_pd();
 193
 194         /* Start inner kernel loop */
 195         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 196         {
 197
 198             /* Get j neighbor index, and coordinate index */
 199             jnrA             = jjnr[jidx];
 200             jnrB             = jjnr[jidx+1];
 201             j_coord_offsetA  = DIM*jnrA;
 202             j_coord_offsetB  = DIM*jnrB;
 203
 204             /* load j atom coordinates */
 205             gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 206                                               &jx0,&jy0,&jz0);
 207
 208             /* Calculate displacement vector */
 209             dx00             = _mm_sub_pd(ix0,jx0);
 210             dy00             = _mm_sub_pd(iy0,jy0);
 211             dz00             = _mm_sub_pd(iz0,jz0);
 212             dx10             = _mm_sub_pd(ix1,jx0);
 213             dy10             = _mm_sub_pd(iy1,jy0);
 214             dz10             = _mm_sub_pd(iz1,jz0);
 215             dx20             = _mm_sub_pd(ix2,jx0);
 216             dy20             = _mm_sub_pd(iy2,jy0);
 217             dz20             = _mm_sub_pd(iz2,jz0);
 218
 219             /* Calculate squared distance and things based on it */
 220             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 221             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
 222             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
 223
 224             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 225             rinv10           = gmx_mm_invsqrt_pd(rsq10);
 226             rinv20           = gmx_mm_invsqrt_pd(rsq20);
 227
 228             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 229             rinvsq10         = _mm_mul_pd(rinv10,rinv10);
 230             rinvsq20         = _mm_mul_pd(rinv20,rinv20);
 231
 232             /* Load parameters for j particles */
 233             jq0              = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
 234             vdwjidx0A        = 2*vdwtype[jnrA+0];
 235             vdwjidx0B        = 2*vdwtype[jnrB+0];
 236
 237             fjx0             = _mm_setzero_pd();
 238             fjy0             = _mm_setzero_pd();
 239             fjz0             = _mm_setzero_pd();
 240
 241             /**************************
 242              * CALCULATE INTERACTIONS *
 243              **************************/
 244
 245             if (gmx_mm_any_lt(rsq00,rcutoff2))
 246             {
 247
 248             r00              = _mm_mul_pd(rsq00,rinv00);
 249
 250             /* Compute parameters for interactions between i and j atoms */
 251             qq00             = _mm_mul_pd(iq0,jq0);
 252             gmx_mm_load_2pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,
 253                                          vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
 254
 255             c6grid_00       = gmx_mm_load_2real_swizzle_pd(vdwgridparam+vdwioffset0+vdwjidx0A,
 256                                                                vdwgridparam+vdwioffset0+vdwjidx0B);
 257
 258             /* EWALD ELECTROSTATICS */
 259
 260             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 261             ewrt             = _mm_mul_pd(r00,ewtabscale);
 262             ewitab           = _mm_cvttpd_epi32(ewrt);
 263             eweps            = _mm_sub_pd(ewrt,_mm_cvtepi32_pd(ewitab));
 264             ewitab           = _mm_slli_epi32(ewitab,2);
 265             ewtabF           = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 266             ewtabD           = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,1) );
 267             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 268             ewtabV           = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,0) +2);
 269             ewtabFn          = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,1) +2);
 270             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 271             felec            = _mm_add_pd(ewtabF,_mm_mul_pd(eweps,ewtabD));
 272             velec            = _mm_sub_pd(ewtabV,_mm_mul_pd(_mm_mul_pd(ewtabhalfspace,eweps),_mm_add_pd(ewtabF,felec)));
 273             velec            = _mm_mul_pd(qq00,_mm_sub_pd(_mm_sub_pd(rinv00,sh_ewald),velec));
 274             felec            = _mm_mul_pd(_mm_mul_pd(qq00,rinv00),_mm_sub_pd(rinvsq00,felec));
 275
 276             /* Analytical LJ-PME */
 277             rinvsix          = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 278             ewcljrsq         = _mm_mul_pd(ewclj2,rsq00);
 279             ewclj6           = _mm_mul_pd(ewclj2,_mm_mul_pd(ewclj2,ewclj2));
 280             exponent         = gmx_simd_exp_d(ewcljrsq);
 281             /* poly = exp(-(beta*r)^2) * (1 + (beta*r)^2 + (beta*r)^4 /2) */
 282             poly             = _mm_mul_pd(exponent,_mm_add_pd(_mm_sub_pd(one,ewcljrsq),_mm_mul_pd(_mm_mul_pd(ewcljrsq,ewcljrsq),one_half)));
 283             /* vvdw6 = [C6 - C6grid * (1-poly)]/r6 */
 284             vvdw6            = _mm_mul_pd(_mm_sub_pd(c6_00,_mm_mul_pd(c6grid_00,_mm_sub_pd(one,poly))),rinvsix);
 285             vvdw12           = _mm_mul_pd(c12_00,_mm_mul_pd(rinvsix,rinvsix));
 286             vvdw             = _mm_sub_pd(_mm_mul_pd( _mm_sub_pd(vvdw12 , _mm_mul_pd(c12_00,_mm_mul_pd(sh_vdw_invrcut6,sh_vdw_invrcut6))),one_twelfth),
 287                                _mm_mul_pd( _mm_sub_pd(vvdw6,_mm_add_pd(_mm_mul_pd(c6_00,sh_vdw_invrcut6),_mm_mul_pd(c6grid_00,sh_lj_ewald))),one_sixth));
 288             /* fvdw = vvdw12/r - (vvdw6/r + (C6grid * exponent * beta^6)/r) */
 289             fvdw             = _mm_mul_pd(_mm_sub_pd(vvdw12,_mm_sub_pd(vvdw6,_mm_mul_pd(_mm_mul_pd(c6grid_00,one_sixth),_mm_mul_pd(exponent,ewclj6)))),rinvsq00);
 290
 291             cutoff_mask      = _mm_cmplt_pd(rsq00,rcutoff2);
 292
 293             /* Update potential sum for this i atom from the interaction with this j atom. */
 294             velec            = _mm_and_pd(velec,cutoff_mask);
 295             velecsum         = _mm_add_pd(velecsum,velec);
 296             vvdw             = _mm_and_pd(vvdw,cutoff_mask);
 297             vvdwsum          = _mm_add_pd(vvdwsum,vvdw);
 298
 299             fscal            = _mm_add_pd(felec,fvdw);
 300
 301             fscal            = _mm_and_pd(fscal,cutoff_mask);
 302
 303             /* Calculate temporary vectorial force */
 304             tx               = _mm_mul_pd(fscal,dx00);
 305             ty               = _mm_mul_pd(fscal,dy00);
 306             tz               = _mm_mul_pd(fscal,dz00);
 307
 308             /* Update vectorial force */
 309             fix0             = _mm_add_pd(fix0,tx);
 310             fiy0             = _mm_add_pd(fiy0,ty);
 311             fiz0             = _mm_add_pd(fiz0,tz);
 312
 313             fjx0             = _mm_add_pd(fjx0,tx);
 314             fjy0             = _mm_add_pd(fjy0,ty);
 315             fjz0             = _mm_add_pd(fjz0,tz);
 316
 317             }
 318
 319             /**************************
 320              * CALCULATE INTERACTIONS *
 321              **************************/
 322
 323             if (gmx_mm_any_lt(rsq10,rcutoff2))
 324             {
 325
 326             r10              = _mm_mul_pd(rsq10,rinv10);
 327
 328             /* Compute parameters for interactions between i and j atoms */
 329             qq10             = _mm_mul_pd(iq1,jq0);
 330
 331             /* EWALD ELECTROSTATICS */
 332
 333             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 334             ewrt             = _mm_mul_pd(r10,ewtabscale);
 335             ewitab           = _mm_cvttpd_epi32(ewrt);
 336             eweps            = _mm_sub_pd(ewrt,_mm_cvtepi32_pd(ewitab));
 337             ewitab           = _mm_slli_epi32(ewitab,2);
 338             ewtabF           = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 339             ewtabD           = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,1) );
 340             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 341             ewtabV           = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,0) +2);
 342             ewtabFn          = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,1) +2);
 343             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 344             felec            = _mm_add_pd(ewtabF,_mm_mul_pd(eweps,ewtabD));
 345             velec            = _mm_sub_pd(ewtabV,_mm_mul_pd(_mm_mul_pd(ewtabhalfspace,eweps),_mm_add_pd(ewtabF,felec)));
 346             velec            = _mm_mul_pd(qq10,_mm_sub_pd(_mm_sub_pd(rinv10,sh_ewald),velec));
 347             felec            = _mm_mul_pd(_mm_mul_pd(qq10,rinv10),_mm_sub_pd(rinvsq10,felec));
 348
 349             cutoff_mask      = _mm_cmplt_pd(rsq10,rcutoff2);
 350
 351             /* Update potential sum for this i atom from the interaction with this j atom. */
 352             velec            = _mm_and_pd(velec,cutoff_mask);
 353             velecsum         = _mm_add_pd(velecsum,velec);
 354
 355             fscal            = felec;
 356
 357             fscal            = _mm_and_pd(fscal,cutoff_mask);
 358
 359             /* Calculate temporary vectorial force */
 360             tx               = _mm_mul_pd(fscal,dx10);
 361             ty               = _mm_mul_pd(fscal,dy10);
 362             tz               = _mm_mul_pd(fscal,dz10);
 363
 364             /* Update vectorial force */
 365             fix1             = _mm_add_pd(fix1,tx);
 366             fiy1             = _mm_add_pd(fiy1,ty);
 367             fiz1             = _mm_add_pd(fiz1,tz);
 368
 369             fjx0             = _mm_add_pd(fjx0,tx);
 370             fjy0             = _mm_add_pd(fjy0,ty);
 371             fjz0             = _mm_add_pd(fjz0,tz);
 372
 373             }
 374
 375             /**************************
 376              * CALCULATE INTERACTIONS *
 377              **************************/
 378
 379             if (gmx_mm_any_lt(rsq20,rcutoff2))
 380             {
 381
 382             r20              = _mm_mul_pd(rsq20,rinv20);
 383
 384             /* Compute parameters for interactions between i and j atoms */
 385             qq20             = _mm_mul_pd(iq2,jq0);
 386
 387             /* EWALD ELECTROSTATICS */
 388
 389             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 390             ewrt             = _mm_mul_pd(r20,ewtabscale);
 391             ewitab           = _mm_cvttpd_epi32(ewrt);
 392             eweps            = _mm_sub_pd(ewrt,_mm_cvtepi32_pd(ewitab));
 393             ewitab           = _mm_slli_epi32(ewitab,2);
 394             ewtabF           = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 395             ewtabD           = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,1) );
 396             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 397             ewtabV           = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,0) +2);
 398             ewtabFn          = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,1) +2);
 399             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 400             felec            = _mm_add_pd(ewtabF,_mm_mul_pd(eweps,ewtabD));
 401             velec            = _mm_sub_pd(ewtabV,_mm_mul_pd(_mm_mul_pd(ewtabhalfspace,eweps),_mm_add_pd(ewtabF,felec)));
 402             velec            = _mm_mul_pd(qq20,_mm_sub_pd(_mm_sub_pd(rinv20,sh_ewald),velec));
 403             felec            = _mm_mul_pd(_mm_mul_pd(qq20,rinv20),_mm_sub_pd(rinvsq20,felec));
 404
 405             cutoff_mask      = _mm_cmplt_pd(rsq20,rcutoff2);
 406
 407             /* Update potential sum for this i atom from the interaction with this j atom. */
 408             velec            = _mm_and_pd(velec,cutoff_mask);
 409             velecsum         = _mm_add_pd(velecsum,velec);
 410
 411             fscal            = felec;
 412
 413             fscal            = _mm_and_pd(fscal,cutoff_mask);
 414
 415             /* Calculate temporary vectorial force */
 416             tx               = _mm_mul_pd(fscal,dx20);
 417             ty               = _mm_mul_pd(fscal,dy20);
 418             tz               = _mm_mul_pd(fscal,dz20);
 419
 420             /* Update vectorial force */
 421             fix2             = _mm_add_pd(fix2,tx);
 422             fiy2             = _mm_add_pd(fiy2,ty);
 423             fiz2             = _mm_add_pd(fiz2,tz);
 424
 425             fjx0             = _mm_add_pd(fjx0,tx);
 426             fjy0             = _mm_add_pd(fjy0,ty);
 427             fjz0             = _mm_add_pd(fjz0,tz);
 428
 429             }
 430
 431             gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,fjx0,fjy0,fjz0);
 432
 433             /* Inner loop uses 177 flops */
 434         }
 435
 436         if(jidx<j_index_end)
 437         {
 438
 439             jnrA             = jjnr[jidx];
 440             j_coord_offsetA  = DIM*jnrA;
 441
 442             /* load j atom coordinates */
 443             gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
 444                                               &jx0,&jy0,&jz0);
 445
 446             /* Calculate displacement vector */
 447             dx00             = _mm_sub_pd(ix0,jx0);
 448             dy00             = _mm_sub_pd(iy0,jy0);
 449             dz00             = _mm_sub_pd(iz0,jz0);
 450             dx10             = _mm_sub_pd(ix1,jx0);
 451             dy10             = _mm_sub_pd(iy1,jy0);
 452             dz10             = _mm_sub_pd(iz1,jz0);
 453             dx20             = _mm_sub_pd(ix2,jx0);
 454             dy20             = _mm_sub_pd(iy2,jy0);
 455             dz20             = _mm_sub_pd(iz2,jz0);
 456
 457             /* Calculate squared distance and things based on it */
 458             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 459             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
 460             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
 461
 462             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 463             rinv10           = gmx_mm_invsqrt_pd(rsq10);
 464             rinv20           = gmx_mm_invsqrt_pd(rsq20);
 465
 466             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 467             rinvsq10         = _mm_mul_pd(rinv10,rinv10);
 468             rinvsq20         = _mm_mul_pd(rinv20,rinv20);
 469
 470             /* Load parameters for j particles */
 471             jq0              = _mm_load_sd(charge+jnrA+0);
 472             vdwjidx0A        = 2*vdwtype[jnrA+0];
 473
 474             fjx0             = _mm_setzero_pd();
 475             fjy0             = _mm_setzero_pd();
 476             fjz0             = _mm_setzero_pd();
 477
 478             /**************************
 479              * CALCULATE INTERACTIONS *
 480              **************************/
 481
 482             if (gmx_mm_any_lt(rsq00,rcutoff2))
 483             {
 484
 485             r00              = _mm_mul_pd(rsq00,rinv00);
 486
 487             /* Compute parameters for interactions between i and j atoms */
 488             qq00             = _mm_mul_pd(iq0,jq0);
 489             gmx_mm_load_1pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,&c6_00,&c12_00);
 490
 491             c6grid_00       = gmx_mm_load_1real_pd(vdwgridparam+vdwioffset0+vdwjidx0A);
 492
 493             /* EWALD ELECTROSTATICS */
 494
 495             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 496             ewrt             = _mm_mul_pd(r00,ewtabscale);
 497             ewitab           = _mm_cvttpd_epi32(ewrt);
 498             eweps            = _mm_sub_pd(ewrt,_mm_cvtepi32_pd(ewitab));
 499             ewitab           = _mm_slli_epi32(ewitab,2);
 500             ewtabF           = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 501             ewtabD           = _mm_setzero_pd();
 502             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 503             ewtabV           = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,0) +2);
 504             ewtabFn          = _mm_setzero_pd();
 505             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 506             felec            = _mm_add_pd(ewtabF,_mm_mul_pd(eweps,ewtabD));
 507             velec            = _mm_sub_pd(ewtabV,_mm_mul_pd(_mm_mul_pd(ewtabhalfspace,eweps),_mm_add_pd(ewtabF,felec)));
 508             velec            = _mm_mul_pd(qq00,_mm_sub_pd(_mm_sub_pd(rinv00,sh_ewald),velec));
 509             felec            = _mm_mul_pd(_mm_mul_pd(qq00,rinv00),_mm_sub_pd(rinvsq00,felec));
 510
 511             /* Analytical LJ-PME */
 512             rinvsix          = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 513             ewcljrsq         = _mm_mul_pd(ewclj2,rsq00);
 514             ewclj6           = _mm_mul_pd(ewclj2,_mm_mul_pd(ewclj2,ewclj2));
 515             exponent         = gmx_simd_exp_d(ewcljrsq);
 516             /* poly = exp(-(beta*r)^2) * (1 + (beta*r)^2 + (beta*r)^4 /2) */
 517             poly             = _mm_mul_pd(exponent,_mm_add_pd(_mm_sub_pd(one,ewcljrsq),_mm_mul_pd(_mm_mul_pd(ewcljrsq,ewcljrsq),one_half)));
 518             /* vvdw6 = [C6 - C6grid * (1-poly)]/r6 */
 519             vvdw6            = _mm_mul_pd(_mm_sub_pd(c6_00,_mm_mul_pd(c6grid_00,_mm_sub_pd(one,poly))),rinvsix);
 520             vvdw12           = _mm_mul_pd(c12_00,_mm_mul_pd(rinvsix,rinvsix));
 521             vvdw             = _mm_sub_pd(_mm_mul_pd( _mm_sub_pd(vvdw12 , _mm_mul_pd(c12_00,_mm_mul_pd(sh_vdw_invrcut6,sh_vdw_invrcut6))),one_twelfth),
 522                                _mm_mul_pd( _mm_sub_pd(vvdw6,_mm_add_pd(_mm_mul_pd(c6_00,sh_vdw_invrcut6),_mm_mul_pd(c6grid_00,sh_lj_ewald))),one_sixth));
 523             /* fvdw = vvdw12/r - (vvdw6/r + (C6grid * exponent * beta^6)/r) */
 524             fvdw             = _mm_mul_pd(_mm_sub_pd(vvdw12,_mm_sub_pd(vvdw6,_mm_mul_pd(_mm_mul_pd(c6grid_00,one_sixth),_mm_mul_pd(exponent,ewclj6)))),rinvsq00);
 525
 526             cutoff_mask      = _mm_cmplt_pd(rsq00,rcutoff2);
 527
 528             /* Update potential sum for this i atom from the interaction with this j atom. */
 529             velec            = _mm_and_pd(velec,cutoff_mask);
 530             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 531             velecsum         = _mm_add_pd(velecsum,velec);
 532             vvdw             = _mm_and_pd(vvdw,cutoff_mask);
 533             vvdw             = _mm_unpacklo_pd(vvdw,_mm_setzero_pd());
 534             vvdwsum          = _mm_add_pd(vvdwsum,vvdw);
 535
 536             fscal            = _mm_add_pd(felec,fvdw);
 537
 538             fscal            = _mm_and_pd(fscal,cutoff_mask);
 539
 540             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 541
 542             /* Calculate temporary vectorial force */
 543             tx               = _mm_mul_pd(fscal,dx00);
 544             ty               = _mm_mul_pd(fscal,dy00);
 545             tz               = _mm_mul_pd(fscal,dz00);
 546
 547             /* Update vectorial force */
 548             fix0             = _mm_add_pd(fix0,tx);
 549             fiy0             = _mm_add_pd(fiy0,ty);
 550             fiz0             = _mm_add_pd(fiz0,tz);
 551
 552             fjx0             = _mm_add_pd(fjx0,tx);
 553             fjy0             = _mm_add_pd(fjy0,ty);
 554             fjz0             = _mm_add_pd(fjz0,tz);
 555
 556             }
 557
 558             /**************************
 559              * CALCULATE INTERACTIONS *
 560              **************************/
 561
 562             if (gmx_mm_any_lt(rsq10,rcutoff2))
 563             {
 564
 565             r10              = _mm_mul_pd(rsq10,rinv10);
 566
 567             /* Compute parameters for interactions between i and j atoms */
 568             qq10             = _mm_mul_pd(iq1,jq0);
 569
 570             /* EWALD ELECTROSTATICS */
 571
 572             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 573             ewrt             = _mm_mul_pd(r10,ewtabscale);
 574             ewitab           = _mm_cvttpd_epi32(ewrt);
 575             eweps            = _mm_sub_pd(ewrt,_mm_cvtepi32_pd(ewitab));
 576             ewitab           = _mm_slli_epi32(ewitab,2);
 577             ewtabF           = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 578             ewtabD           = _mm_setzero_pd();
 579             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 580             ewtabV           = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,0) +2);
 581             ewtabFn          = _mm_setzero_pd();
 582             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 583             felec            = _mm_add_pd(ewtabF,_mm_mul_pd(eweps,ewtabD));
 584             velec            = _mm_sub_pd(ewtabV,_mm_mul_pd(_mm_mul_pd(ewtabhalfspace,eweps),_mm_add_pd(ewtabF,felec)));
 585             velec            = _mm_mul_pd(qq10,_mm_sub_pd(_mm_sub_pd(rinv10,sh_ewald),velec));
 586             felec            = _mm_mul_pd(_mm_mul_pd(qq10,rinv10),_mm_sub_pd(rinvsq10,felec));
 587
 588             cutoff_mask      = _mm_cmplt_pd(rsq10,rcutoff2);
 589
 590             /* Update potential sum for this i atom from the interaction with this j atom. */
 591             velec            = _mm_and_pd(velec,cutoff_mask);
 592             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 593             velecsum         = _mm_add_pd(velecsum,velec);
 594
 595             fscal            = felec;
 596
 597             fscal            = _mm_and_pd(fscal,cutoff_mask);
 598
 599             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 600
 601             /* Calculate temporary vectorial force */
 602             tx               = _mm_mul_pd(fscal,dx10);
 603             ty               = _mm_mul_pd(fscal,dy10);
 604             tz               = _mm_mul_pd(fscal,dz10);
 605
 606             /* Update vectorial force */
 607             fix1             = _mm_add_pd(fix1,tx);
 608             fiy1             = _mm_add_pd(fiy1,ty);
 609             fiz1             = _mm_add_pd(fiz1,tz);
 610
 611             fjx0             = _mm_add_pd(fjx0,tx);
 612             fjy0             = _mm_add_pd(fjy0,ty);
 613             fjz0             = _mm_add_pd(fjz0,tz);
 614
 615             }
 616
 617             /**************************
 618              * CALCULATE INTERACTIONS *
 619              **************************/
 620
 621             if (gmx_mm_any_lt(rsq20,rcutoff2))
 622             {
 623
 624             r20              = _mm_mul_pd(rsq20,rinv20);
 625
 626             /* Compute parameters for interactions between i and j atoms */
 627             qq20             = _mm_mul_pd(iq2,jq0);
 628
 629             /* EWALD ELECTROSTATICS */
 630
 631             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 632             ewrt             = _mm_mul_pd(r20,ewtabscale);
 633             ewitab           = _mm_cvttpd_epi32(ewrt);
 634             eweps            = _mm_sub_pd(ewrt,_mm_cvtepi32_pd(ewitab));
 635             ewitab           = _mm_slli_epi32(ewitab,2);
 636             ewtabF           = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 637             ewtabD           = _mm_setzero_pd();
 638             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 639             ewtabV           = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,0) +2);
 640             ewtabFn          = _mm_setzero_pd();
 641             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 642             felec            = _mm_add_pd(ewtabF,_mm_mul_pd(eweps,ewtabD));
 643             velec            = _mm_sub_pd(ewtabV,_mm_mul_pd(_mm_mul_pd(ewtabhalfspace,eweps),_mm_add_pd(ewtabF,felec)));
 644             velec            = _mm_mul_pd(qq20,_mm_sub_pd(_mm_sub_pd(rinv20,sh_ewald),velec));
 645             felec            = _mm_mul_pd(_mm_mul_pd(qq20,rinv20),_mm_sub_pd(rinvsq20,felec));
 646
 647             cutoff_mask      = _mm_cmplt_pd(rsq20,rcutoff2);
 648
 649             /* Update potential sum for this i atom from the interaction with this j atom. */
 650             velec            = _mm_and_pd(velec,cutoff_mask);
 651             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 652             velecsum         = _mm_add_pd(velecsum,velec);
 653
 654             fscal            = felec;
 655
 656             fscal            = _mm_and_pd(fscal,cutoff_mask);
 657
 658             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 659
 660             /* Calculate temporary vectorial force */
 661             tx               = _mm_mul_pd(fscal,dx20);
 662             ty               = _mm_mul_pd(fscal,dy20);
 663             tz               = _mm_mul_pd(fscal,dz20);
 664
 665             /* Update vectorial force */
 666             fix2             = _mm_add_pd(fix2,tx);
 667             fiy2             = _mm_add_pd(fiy2,ty);
 668             fiz2             = _mm_add_pd(fiz2,tz);
 669
 670             fjx0             = _mm_add_pd(fjx0,tx);
 671             fjy0             = _mm_add_pd(fjy0,ty);
 672             fjz0             = _mm_add_pd(fjz0,tz);
 673
 674             }
 675
 676             gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,fjx0,fjy0,fjz0);
 677
 678             /* Inner loop uses 177 flops */
 679         }
 680
 681         /* End of innermost loop */
 682
 683         gmx_mm_update_iforce_3atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
 684                                               f+i_coord_offset,fshift+i_shift_offset);
 685
 686         ggid                        = gid[iidx];
 687         /* Update potential energies */
 688         gmx_mm_update_1pot_pd(velecsum,kernel_data->energygrp_elec+ggid);
 689         gmx_mm_update_1pot_pd(vvdwsum,kernel_data->energygrp_vdw+ggid);
 690
 691         /* Increment number of inner iterations */
 692         inneriter                  += j_index_end - j_index_start;
 693
 694         /* Outer loop uses 20 flops */
 695     }
 696
 697     /* Increment number of outer iterations */
 698     outeriter        += nri;
 699
 700     /* Update outer/inner flops */
 701
 702     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3_VF,outeriter*20 + inneriter*177);
 703 }
 704 /*
 705  * Gromacs nonbonded kernel:   nb_kernel_ElecEwSh_VdwLJEwSh_GeomW3P1_F_sse2_double
 706  * Electrostatics interaction: Ewald
 707  * VdW interaction:            LJEwald
 708  * Geometry:                   Water3-Particle
 709  * Calculate force/pot:        Force
 710  */
 711 void
 712 nb_kernel_ElecEwSh_VdwLJEwSh_GeomW3P1_F_sse2_double
 713                     (t_nblist                    * gmx_restrict       nlist,
 714                      rvec                        * gmx_restrict          xx,
 715                      rvec                        * gmx_restrict          ff,
 716                      t_forcerec                  * gmx_restrict          fr,
 717                      t_mdatoms                   * gmx_restrict     mdatoms,
 718                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 719                      t_nrnb                      * gmx_restrict        nrnb)
 720 {
 721     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 722      * just 0 for non-waters.
 723      * Suffixes A,B refer to j loop unrolling done with SSE double precision, e.g. for the two different
 724      * jnr indices corresponding to data put in the four positions in the SIMD register.
 725      */
 726     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 727     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 728     int              jnrA,jnrB;
 729     int              j_coord_offsetA,j_coord_offsetB;
 730     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 731     real             rcutoff_scalar;
 732     real             *shiftvec,*fshift,*x,*f;
 733     __m128d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 734     int              vdwioffset0;
 735     __m128d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 736     int              vdwioffset1;
 737     __m128d          ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
 738     int              vdwioffset2;
 739     __m128d          ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
 740     int              vdwjidx0A,vdwjidx0B;
 741     __m128d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 742     __m128d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 743     __m128d          dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
 744     __m128d          dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
 745     __m128d          velec,felec,velecsum,facel,crf,krf,krf2;
 746     real             *charge;
 747     int              nvdwtype;
 748     __m128d          rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 749     int              *vdwtype;
 750     real             *vdwparam;
 751     __m128d          one_sixth   = _mm_set1_pd(1.0/6.0);
 752     __m128d          one_twelfth = _mm_set1_pd(1.0/12.0);
 753     __m128d           c6grid_00;
 754     __m128d           c6grid_10;
 755     __m128d           c6grid_20;
 756     __m128d           ewclj,ewclj2,ewclj6,ewcljrsq,poly,exponent,f6A,f6B,sh_lj_ewald;
 757     real             *vdwgridparam;
 758     __m128d           one_half = _mm_set1_pd(0.5);
 759     __m128d           minus_one = _mm_set1_pd(-1.0);
 760     __m128i          ewitab;
 761     __m128d          ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
 762     real             *ewtab;
 763     __m128d          dummy_mask,cutoff_mask;
 764     __m128d          signbit   = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
 765     __m128d          one     = _mm_set1_pd(1.0);
 766     __m128d          two     = _mm_set1_pd(2.0);
 767     x                = xx[0];
 768     f                = ff[0];
 769
 770     nri              = nlist->nri;
 771     iinr             = nlist->iinr;
 772     jindex           = nlist->jindex;
 773     jjnr             = nlist->jjnr;
 774     shiftidx         = nlist->shift;
 775     gid              = nlist->gid;
 776     shiftvec         = fr->shift_vec[0];
 777     fshift           = fr->fshift[0];
 778     facel            = _mm_set1_pd(fr->epsfac);
 779     charge           = mdatoms->chargeA;
 780     nvdwtype         = fr->ntype;
 781     vdwparam         = fr->nbfp;
 782     vdwtype          = mdatoms->typeA;
 783     vdwgridparam     = fr->ljpme_c6grid;
 784     sh_lj_ewald      = _mm_set1_pd(fr->ic->sh_lj_ewald);
 785     ewclj            = _mm_set1_pd(fr->ewaldcoeff_lj);
 786     ewclj2           = _mm_mul_pd(minus_one,_mm_mul_pd(ewclj,ewclj));
 787
 788     sh_ewald         = _mm_set1_pd(fr->ic->sh_ewald);
 789     ewtab            = fr->ic->tabq_coul_F;
 790     ewtabscale       = _mm_set1_pd(fr->ic->tabq_scale);
 791     ewtabhalfspace   = _mm_set1_pd(0.5/fr->ic->tabq_scale);
 792
 793     /* Setup water-specific parameters */
 794     inr              = nlist->iinr[0];
 795     iq0              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+0]));
 796     iq1              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+1]));
 797     iq2              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+2]));
 798     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 799
 800     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 801     rcutoff_scalar   = fr->rcoulomb;
 802     rcutoff          = _mm_set1_pd(rcutoff_scalar);
 803     rcutoff2         = _mm_mul_pd(rcutoff,rcutoff);
 804
 805     sh_vdw_invrcut6  = _mm_set1_pd(fr->ic->sh_invrc6);
 806     rvdw             = _mm_set1_pd(fr->rvdw);
 807
 808     /* Avoid stupid compiler warnings */
 809     jnrA = jnrB = 0;
 810     j_coord_offsetA = 0;
 811     j_coord_offsetB = 0;
 812
 813     outeriter        = 0;
 814     inneriter        = 0;
 815
 816     /* Start outer loop over neighborlists */
 817     for(iidx=0; iidx<nri; iidx++)
 818     {
 819         /* Load shift vector for this list */
 820         i_shift_offset   = DIM*shiftidx[iidx];
 821
 822         /* Load limits for loop over neighbors */
 823         j_index_start    = jindex[iidx];
 824         j_index_end      = jindex[iidx+1];
 825
 826         /* Get outer coordinate index */
 827         inr              = iinr[iidx];
 828         i_coord_offset   = DIM*inr;
 829
 830         /* Load i particle coords and add shift vector */
 831         gmx_mm_load_shift_and_3rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
 832                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 833
 834         fix0             = _mm_setzero_pd();
 835         fiy0             = _mm_setzero_pd();
 836         fiz0             = _mm_setzero_pd();
 837         fix1             = _mm_setzero_pd();
 838         fiy1             = _mm_setzero_pd();
 839         fiz1             = _mm_setzero_pd();
 840         fix2             = _mm_setzero_pd();
 841         fiy2             = _mm_setzero_pd();
 842         fiz2             = _mm_setzero_pd();
 843
 844         /* Start inner kernel loop */
 845         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 846         {
 847
 848             /* Get j neighbor index, and coordinate index */
 849             jnrA             = jjnr[jidx];
 850             jnrB             = jjnr[jidx+1];
 851             j_coord_offsetA  = DIM*jnrA;
 852             j_coord_offsetB  = DIM*jnrB;
 853
 854             /* load j atom coordinates */
 855             gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 856                                               &jx0,&jy0,&jz0);
 857
 858             /* Calculate displacement vector */
 859             dx00             = _mm_sub_pd(ix0,jx0);
 860             dy00             = _mm_sub_pd(iy0,jy0);
 861             dz00             = _mm_sub_pd(iz0,jz0);
 862             dx10             = _mm_sub_pd(ix1,jx0);
 863             dy10             = _mm_sub_pd(iy1,jy0);
 864             dz10             = _mm_sub_pd(iz1,jz0);
 865             dx20             = _mm_sub_pd(ix2,jx0);
 866             dy20             = _mm_sub_pd(iy2,jy0);
 867             dz20             = _mm_sub_pd(iz2,jz0);
 868
 869             /* Calculate squared distance and things based on it */
 870             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 871             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
 872             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
 873
 874             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 875             rinv10           = gmx_mm_invsqrt_pd(rsq10);
 876             rinv20           = gmx_mm_invsqrt_pd(rsq20);
 877
 878             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 879             rinvsq10         = _mm_mul_pd(rinv10,rinv10);
 880             rinvsq20         = _mm_mul_pd(rinv20,rinv20);
 881
 882             /* Load parameters for j particles */
 883             jq0              = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
 884             vdwjidx0A        = 2*vdwtype[jnrA+0];
 885             vdwjidx0B        = 2*vdwtype[jnrB+0];
 886
 887             fjx0             = _mm_setzero_pd();
 888             fjy0             = _mm_setzero_pd();
 889             fjz0             = _mm_setzero_pd();
 890
 891             /**************************
 892              * CALCULATE INTERACTIONS *
 893              **************************/
 894
 895             if (gmx_mm_any_lt(rsq00,rcutoff2))
 896             {
 897
 898             r00              = _mm_mul_pd(rsq00,rinv00);
 899
 900             /* Compute parameters for interactions between i and j atoms */
 901             qq00             = _mm_mul_pd(iq0,jq0);
 902             gmx_mm_load_2pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,
 903                                          vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
 904
 905             c6grid_00       = gmx_mm_load_2real_swizzle_pd(vdwgridparam+vdwioffset0+vdwjidx0A,
 906                                                                vdwgridparam+vdwioffset0+vdwjidx0B);
 907
 908             /* EWALD ELECTROSTATICS */
 909
 910             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 911             ewrt             = _mm_mul_pd(r00,ewtabscale);
 912             ewitab           = _mm_cvttpd_epi32(ewrt);
 913             eweps            = _mm_sub_pd(ewrt,_mm_cvtepi32_pd(ewitab));
 914             gmx_mm_load_2pair_swizzle_pd(ewtab+gmx_mm_extract_epi32(ewitab,0),ewtab+gmx_mm_extract_epi32(ewitab,1),
 915                                          &ewtabF,&ewtabFn);
 916             felec            = _mm_add_pd(_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF),_mm_mul_pd(eweps,ewtabFn));
 917             felec            = _mm_mul_pd(_mm_mul_pd(qq00,rinv00),_mm_sub_pd(rinvsq00,felec));
 918
 919             /* Analytical LJ-PME */
 920             rinvsix          = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 921             ewcljrsq         = _mm_mul_pd(ewclj2,rsq00);
 922             ewclj6           = _mm_mul_pd(ewclj2,_mm_mul_pd(ewclj2,ewclj2));
 923             exponent         = gmx_simd_exp_d(ewcljrsq);
 924             /* poly = exp(-(beta*r)^2) * (1 + (beta*r)^2 + (beta*r)^4 /2) */
 925             poly             = _mm_mul_pd(exponent,_mm_add_pd(_mm_sub_pd(one,ewcljrsq),_mm_mul_pd(_mm_mul_pd(ewcljrsq,ewcljrsq),one_half)));
 926             /* f6A = 6 * C6grid * (1 - poly) */
 927             f6A              = _mm_mul_pd(c6grid_00,_mm_sub_pd(one,poly));
 928             /* f6B = C6grid * exponent * beta^6 */
 929             f6B              = _mm_mul_pd(_mm_mul_pd(c6grid_00,one_sixth),_mm_mul_pd(exponent,ewclj6));
 930             /* fvdw = 12*C12/r13 - ((6*C6 - f6A)/r6 + f6B)/r */
 931             fvdw              = _mm_mul_pd(_mm_add_pd(_mm_mul_pd(_mm_sub_pd(_mm_mul_pd(c12_00,rinvsix),_mm_sub_pd(c6_00,f6A)),rinvsix),f6B),rinvsq00);
 932
 933             cutoff_mask      = _mm_cmplt_pd(rsq00,rcutoff2);
 934
 935             fscal            = _mm_add_pd(felec,fvdw);
 936
 937             fscal            = _mm_and_pd(fscal,cutoff_mask);
 938
 939             /* Calculate temporary vectorial force */
 940             tx               = _mm_mul_pd(fscal,dx00);
 941             ty               = _mm_mul_pd(fscal,dy00);
 942             tz               = _mm_mul_pd(fscal,dz00);
 943
 944             /* Update vectorial force */
 945             fix0             = _mm_add_pd(fix0,tx);
 946             fiy0             = _mm_add_pd(fiy0,ty);
 947             fiz0             = _mm_add_pd(fiz0,tz);
 948
 949             fjx0             = _mm_add_pd(fjx0,tx);
 950             fjy0             = _mm_add_pd(fjy0,ty);
 951             fjz0             = _mm_add_pd(fjz0,tz);
 952
 953             }
 954
 955             /**************************
 956              * CALCULATE INTERACTIONS *
 957              **************************/
 958
 959             if (gmx_mm_any_lt(rsq10,rcutoff2))
 960             {
 961
 962             r10              = _mm_mul_pd(rsq10,rinv10);
 963
 964             /* Compute parameters for interactions between i and j atoms */
 965             qq10             = _mm_mul_pd(iq1,jq0);
 966
 967             /* EWALD ELECTROSTATICS */
 968
 969             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 970             ewrt             = _mm_mul_pd(r10,ewtabscale);
 971             ewitab           = _mm_cvttpd_epi32(ewrt);
 972             eweps            = _mm_sub_pd(ewrt,_mm_cvtepi32_pd(ewitab));
 973             gmx_mm_load_2pair_swizzle_pd(ewtab+gmx_mm_extract_epi32(ewitab,0),ewtab+gmx_mm_extract_epi32(ewitab,1),
 974                                          &ewtabF,&ewtabFn);
 975             felec            = _mm_add_pd(_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF),_mm_mul_pd(eweps,ewtabFn));
 976             felec            = _mm_mul_pd(_mm_mul_pd(qq10,rinv10),_mm_sub_pd(rinvsq10,felec));
 977
 978             cutoff_mask      = _mm_cmplt_pd(rsq10,rcutoff2);
 979
 980             fscal            = felec;
 981
 982             fscal            = _mm_and_pd(fscal,cutoff_mask);
 983
 984             /* Calculate temporary vectorial force */
 985             tx               = _mm_mul_pd(fscal,dx10);
 986             ty               = _mm_mul_pd(fscal,dy10);
 987             tz               = _mm_mul_pd(fscal,dz10);
 988
 989             /* Update vectorial force */
 990             fix1             = _mm_add_pd(fix1,tx);
 991             fiy1             = _mm_add_pd(fiy1,ty);
 992             fiz1             = _mm_add_pd(fiz1,tz);
 993
 994             fjx0             = _mm_add_pd(fjx0,tx);
 995             fjy0             = _mm_add_pd(fjy0,ty);
 996             fjz0             = _mm_add_pd(fjz0,tz);
 997
 998             }
 999
1000             /**************************
1001              * CALCULATE INTERACTIONS *
1002              **************************/
1003
1004             if (gmx_mm_any_lt(rsq20,rcutoff2))
1005             {
1006
1007             r20              = _mm_mul_pd(rsq20,rinv20);
1008
1009             /* Compute parameters for interactions between i and j atoms */
1010             qq20             = _mm_mul_pd(iq2,jq0);
1011
1012             /* EWALD ELECTROSTATICS */
1013
1014             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1015             ewrt             = _mm_mul_pd(r20,ewtabscale);
1016             ewitab           = _mm_cvttpd_epi32(ewrt);
1017             eweps            = _mm_sub_pd(ewrt,_mm_cvtepi32_pd(ewitab));
1018             gmx_mm_load_2pair_swizzle_pd(ewtab+gmx_mm_extract_epi32(ewitab,0),ewtab+gmx_mm_extract_epi32(ewitab,1),
1019                                          &ewtabF,&ewtabFn);
1020             felec            = _mm_add_pd(_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF),_mm_mul_pd(eweps,ewtabFn));
1021             felec            = _mm_mul_pd(_mm_mul_pd(qq20,rinv20),_mm_sub_pd(rinvsq20,felec));
1022
1023             cutoff_mask      = _mm_cmplt_pd(rsq20,rcutoff2);
1024
1025             fscal            = felec;
1026
1027             fscal            = _mm_and_pd(fscal,cutoff_mask);
1028
1029             /* Calculate temporary vectorial force */
1030             tx               = _mm_mul_pd(fscal,dx20);
1031             ty               = _mm_mul_pd(fscal,dy20);
1032             tz               = _mm_mul_pd(fscal,dz20);
1033
1034             /* Update vectorial force */
1035             fix2             = _mm_add_pd(fix2,tx);
1036             fiy2             = _mm_add_pd(fiy2,ty);
1037             fiz2             = _mm_add_pd(fiz2,tz);
1038
1039             fjx0             = _mm_add_pd(fjx0,tx);
1040             fjy0             = _mm_add_pd(fjy0,ty);
1041             fjz0             = _mm_add_pd(fjz0,tz);
1042
1043             }
1044
1045             gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,fjx0,fjy0,fjz0);
1046
1047             /* Inner loop uses 143 flops */
1048         }
1049
1050         if(jidx<j_index_end)
1051         {
1052
1053             jnrA             = jjnr[jidx];
1054             j_coord_offsetA  = DIM*jnrA;
1055
1056             /* load j atom coordinates */
1057             gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
1058                                               &jx0,&jy0,&jz0);
1059
1060             /* Calculate displacement vector */
1061             dx00             = _mm_sub_pd(ix0,jx0);
1062             dy00             = _mm_sub_pd(iy0,jy0);
1063             dz00             = _mm_sub_pd(iz0,jz0);
1064             dx10             = _mm_sub_pd(ix1,jx0);
1065             dy10             = _mm_sub_pd(iy1,jy0);
1066             dz10             = _mm_sub_pd(iz1,jz0);
1067             dx20             = _mm_sub_pd(ix2,jx0);
1068             dy20             = _mm_sub_pd(iy2,jy0);
1069             dz20             = _mm_sub_pd(iz2,jz0);
1070
1071             /* Calculate squared distance and things based on it */
1072             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
1073             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
1074             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
1075
1076             rinv00           = gmx_mm_invsqrt_pd(rsq00);
1077             rinv10           = gmx_mm_invsqrt_pd(rsq10);
1078             rinv20           = gmx_mm_invsqrt_pd(rsq20);
1079
1080             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
1081             rinvsq10         = _mm_mul_pd(rinv10,rinv10);
1082             rinvsq20         = _mm_mul_pd(rinv20,rinv20);
1083
1084             /* Load parameters for j particles */
1085             jq0              = _mm_load_sd(charge+jnrA+0);
1086             vdwjidx0A        = 2*vdwtype[jnrA+0];
1087
1088             fjx0             = _mm_setzero_pd();
1089             fjy0             = _mm_setzero_pd();
1090             fjz0             = _mm_setzero_pd();
1091
1092             /**************************
1093              * CALCULATE INTERACTIONS *
1094              **************************/
1095
1096             if (gmx_mm_any_lt(rsq00,rcutoff2))
1097             {
1098
1099             r00              = _mm_mul_pd(rsq00,rinv00);
1100
1101             /* Compute parameters for interactions between i and j atoms */
1102             qq00             = _mm_mul_pd(iq0,jq0);
1103             gmx_mm_load_1pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,&c6_00,&c12_00);
1104
1105             c6grid_00       = gmx_mm_load_1real_pd(vdwgridparam+vdwioffset0+vdwjidx0A);
1106
1107             /* EWALD ELECTROSTATICS */
1108
1109             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1110             ewrt             = _mm_mul_pd(r00,ewtabscale);
1111             ewitab           = _mm_cvttpd_epi32(ewrt);
1112             eweps            = _mm_sub_pd(ewrt,_mm_cvtepi32_pd(ewitab));
1113             gmx_mm_load_1pair_swizzle_pd(ewtab+gmx_mm_extract_epi32(ewitab,0),&ewtabF,&ewtabFn);
1114             felec            = _mm_add_pd(_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF),_mm_mul_pd(eweps,ewtabFn));
1115             felec            = _mm_mul_pd(_mm_mul_pd(qq00,rinv00),_mm_sub_pd(rinvsq00,felec));
1116
1117             /* Analytical LJ-PME */
1118             rinvsix          = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
1119             ewcljrsq         = _mm_mul_pd(ewclj2,rsq00);
1120             ewclj6           = _mm_mul_pd(ewclj2,_mm_mul_pd(ewclj2,ewclj2));
1121             exponent         = gmx_simd_exp_d(ewcljrsq);
1122             /* poly = exp(-(beta*r)^2) * (1 + (beta*r)^2 + (beta*r)^4 /2) */
1123             poly             = _mm_mul_pd(exponent,_mm_add_pd(_mm_sub_pd(one,ewcljrsq),_mm_mul_pd(_mm_mul_pd(ewcljrsq,ewcljrsq),one_half)));
1124             /* f6A = 6 * C6grid * (1 - poly) */
1125             f6A              = _mm_mul_pd(c6grid_00,_mm_sub_pd(one,poly));
1126             /* f6B = C6grid * exponent * beta^6 */
1127             f6B              = _mm_mul_pd(_mm_mul_pd(c6grid_00,one_sixth),_mm_mul_pd(exponent,ewclj6));
1128             /* fvdw = 12*C12/r13 - ((6*C6 - f6A)/r6 + f6B)/r */
1129             fvdw              = _mm_mul_pd(_mm_add_pd(_mm_mul_pd(_mm_sub_pd(_mm_mul_pd(c12_00,rinvsix),_mm_sub_pd(c6_00,f6A)),rinvsix),f6B),rinvsq00);
1130
1131             cutoff_mask      = _mm_cmplt_pd(rsq00,rcutoff2);
1132
1133             fscal            = _mm_add_pd(felec,fvdw);
1134
1135             fscal            = _mm_and_pd(fscal,cutoff_mask);
1136
1137             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1138
1139             /* Calculate temporary vectorial force */
1140             tx               = _mm_mul_pd(fscal,dx00);
1141             ty               = _mm_mul_pd(fscal,dy00);
1142             tz               = _mm_mul_pd(fscal,dz00);
1143
1144             /* Update vectorial force */
1145             fix0             = _mm_add_pd(fix0,tx);
1146             fiy0             = _mm_add_pd(fiy0,ty);
1147             fiz0             = _mm_add_pd(fiz0,tz);
1148
1149             fjx0             = _mm_add_pd(fjx0,tx);
1150             fjy0             = _mm_add_pd(fjy0,ty);
1151             fjz0             = _mm_add_pd(fjz0,tz);
1152
1153             }
1154
1155             /**************************
1156              * CALCULATE INTERACTIONS *
1157              **************************/
1158
1159             if (gmx_mm_any_lt(rsq10,rcutoff2))
1160             {
1161
1162             r10              = _mm_mul_pd(rsq10,rinv10);
1163
1164             /* Compute parameters for interactions between i and j atoms */
1165             qq10             = _mm_mul_pd(iq1,jq0);
1166
1167             /* EWALD ELECTROSTATICS */
1168
1169             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1170             ewrt             = _mm_mul_pd(r10,ewtabscale);
1171             ewitab           = _mm_cvttpd_epi32(ewrt);
1172             eweps            = _mm_sub_pd(ewrt,_mm_cvtepi32_pd(ewitab));
1173             gmx_mm_load_1pair_swizzle_pd(ewtab+gmx_mm_extract_epi32(ewitab,0),&ewtabF,&ewtabFn);
1174             felec            = _mm_add_pd(_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF),_mm_mul_pd(eweps,ewtabFn));
1175             felec            = _mm_mul_pd(_mm_mul_pd(qq10,rinv10),_mm_sub_pd(rinvsq10,felec));
1176
1177             cutoff_mask      = _mm_cmplt_pd(rsq10,rcutoff2);
1178
1179             fscal            = felec;
1180
1181             fscal            = _mm_and_pd(fscal,cutoff_mask);
1182
1183             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1184
1185             /* Calculate temporary vectorial force */
1186             tx               = _mm_mul_pd(fscal,dx10);
1187             ty               = _mm_mul_pd(fscal,dy10);
1188             tz               = _mm_mul_pd(fscal,dz10);
1189
1190             /* Update vectorial force */
1191             fix1             = _mm_add_pd(fix1,tx);
1192             fiy1             = _mm_add_pd(fiy1,ty);
1193             fiz1             = _mm_add_pd(fiz1,tz);
1194
1195             fjx0             = _mm_add_pd(fjx0,tx);
1196             fjy0             = _mm_add_pd(fjy0,ty);
1197             fjz0             = _mm_add_pd(fjz0,tz);
1198
1199             }
1200
1201             /**************************
1202              * CALCULATE INTERACTIONS *
1203              **************************/
1204
1205             if (gmx_mm_any_lt(rsq20,rcutoff2))
1206             {
1207
1208             r20              = _mm_mul_pd(rsq20,rinv20);
1209
1210             /* Compute parameters for interactions between i and j atoms */
1211             qq20             = _mm_mul_pd(iq2,jq0);
1212
1213             /* EWALD ELECTROSTATICS */
1214
1215             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1216             ewrt             = _mm_mul_pd(r20,ewtabscale);
1217             ewitab           = _mm_cvttpd_epi32(ewrt);
1218             eweps            = _mm_sub_pd(ewrt,_mm_cvtepi32_pd(ewitab));
1219             gmx_mm_load_1pair_swizzle_pd(ewtab+gmx_mm_extract_epi32(ewitab,0),&ewtabF,&ewtabFn);
1220             felec            = _mm_add_pd(_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF),_mm_mul_pd(eweps,ewtabFn));
1221             felec            = _mm_mul_pd(_mm_mul_pd(qq20,rinv20),_mm_sub_pd(rinvsq20,felec));
1222
1223             cutoff_mask      = _mm_cmplt_pd(rsq20,rcutoff2);
1224
1225             fscal            = felec;
1226
1227             fscal            = _mm_and_pd(fscal,cutoff_mask);
1228
1229             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1230
1231             /* Calculate temporary vectorial force */
1232             tx               = _mm_mul_pd(fscal,dx20);
1233             ty               = _mm_mul_pd(fscal,dy20);
1234             tz               = _mm_mul_pd(fscal,dz20);
1235
1236             /* Update vectorial force */
1237             fix2             = _mm_add_pd(fix2,tx);
1238             fiy2             = _mm_add_pd(fiy2,ty);
1239             fiz2             = _mm_add_pd(fiz2,tz);
1240
1241             fjx0             = _mm_add_pd(fjx0,tx);
1242             fjy0             = _mm_add_pd(fjy0,ty);
1243             fjz0             = _mm_add_pd(fjz0,tz);
1244
1245             }
1246
1247             gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,fjx0,fjy0,fjz0);
1248
1249             /* Inner loop uses 143 flops */
1250         }
1251
1252         /* End of innermost loop */
1253
1254         gmx_mm_update_iforce_3atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
1255                                               f+i_coord_offset,fshift+i_shift_offset);
1256
1257         /* Increment number of inner iterations */
1258         inneriter                  += j_index_end - j_index_start;
1259
1260         /* Outer loop uses 18 flops */
1261     }
1262
1263     /* Increment number of outer iterations */
1264     outeriter        += nri;
1265
1266     /* Update outer/inner flops */
1267
1268     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3_F,outeriter*18 + inneriter*143);
1269 }