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 "gmxpre.h"
  39
  40 #include "config.h"
  41
  42 #include <math.h>
  43
  44 #include "../nb_kernel.h"
  45 #include "gromacs/legacyheaders/types/simple.h"
  46 #include "gromacs/math/vec.h"
  47 #include "gromacs/legacyheaders/nrnb.h"
  48
  49 #include "gromacs/simd/math_x86_sse2_double.h"
  50 #include "kernelutil_x86_sse2_double.h"
  51
  52 /*
  53  * Gromacs nonbonded kernel:   nb_kernel_ElecEwSh_VdwLJEwSh_GeomW3P1_VF_sse2_double
  54  * Electrostatics interaction: Ewald
  55  * VdW interaction:            LJEwald
  56  * Geometry:                   Water3-Particle
  57  * Calculate force/pot:        PotentialAndForce
  58  */
  59 void
  60 nb_kernel_ElecEwSh_VdwLJEwSh_GeomW3P1_VF_sse2_double
  61                     (t_nblist                    * gmx_restrict       nlist,
  62                      rvec                        * gmx_restrict          xx,
  63                      rvec                        * gmx_restrict          ff,
  64                      t_forcerec                  * gmx_restrict          fr,
  65                      t_mdatoms                   * gmx_restrict     mdatoms,
  66                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
  67                      t_nrnb                      * gmx_restrict        nrnb)
  68 {
  69     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
  70      * just 0 for non-waters.
  71      * Suffixes A,B refer to j loop unrolling done with SSE double precision, e.g. for the two different
  72      * jnr indices corresponding to data put in the four positions in the SIMD register.
  73      */
  74     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
  75     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
  76     int              jnrA,jnrB;
  77     int              j_coord_offsetA,j_coord_offsetB;
  78     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
  79     real             rcutoff_scalar;
  80     real             *shiftvec,*fshift,*x,*f;
  81     __m128d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
  82     int              vdwioffset0;
  83     __m128d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
  84     int              vdwioffset1;
  85     __m128d          ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
  86     int              vdwioffset2;
  87     __m128d          ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
  88     int              vdwjidx0A,vdwjidx0B;
  89     __m128d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  90     __m128d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  91     __m128d          dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
  92     __m128d          dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
  93     __m128d          velec,felec,velecsum,facel,crf,krf,krf2;
  94     real             *charge;
  95     int              nvdwtype;
  96     __m128d          rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
  97     int              *vdwtype;
  98     real             *vdwparam;
  99     __m128d          one_sixth   = _mm_set1_pd(1.0/6.0);
 100     __m128d          one_twelfth = _mm_set1_pd(1.0/12.0);
 101     __m128d           c6grid_00;
 102     __m128d           c6grid_10;
 103     __m128d           c6grid_20;
 104     __m128d           ewclj,ewclj2,ewclj6,ewcljrsq,poly,exponent,f6A,f6B,sh_lj_ewald;
 105     real             *vdwgridparam;
 106     __m128d           one_half = _mm_set1_pd(0.5);
 107     __m128d           minus_one = _mm_set1_pd(-1.0);
 108     __m128i          ewitab;
 109     __m128d          ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
 110     real             *ewtab;
 111     __m128d          dummy_mask,cutoff_mask;
 112     __m128d          signbit   = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
 113     __m128d          one     = _mm_set1_pd(1.0);
 114     __m128d          two     = _mm_set1_pd(2.0);
 115     x                = xx[0];
 116     f                = ff[0];
 117
 118     nri              = nlist->nri;
 119     iinr             = nlist->iinr;
 120     jindex           = nlist->jindex;
 121     jjnr             = nlist->jjnr;
 122     shiftidx         = nlist->shift;
 123     gid              = nlist->gid;
 124     shiftvec         = fr->shift_vec[0];
 125     fshift           = fr->fshift[0];
 126     facel            = _mm_set1_pd(fr->epsfac);
 127     charge           = mdatoms->chargeA;
 128     nvdwtype         = fr->ntype;
 129     vdwparam         = fr->nbfp;
 130     vdwtype          = mdatoms->typeA;
 131     vdwgridparam     = fr->ljpme_c6grid;
 132     sh_lj_ewald      = _mm_set1_pd(fr->ic->sh_lj_ewald);
 133     ewclj            = _mm_set1_pd(fr->ewaldcoeff_lj);
 134     ewclj2           = _mm_mul_pd(minus_one,_mm_mul_pd(ewclj,ewclj));
 135
 136     sh_ewald         = _mm_set1_pd(fr->ic->sh_ewald);
 137     ewtab            = fr->ic->tabq_coul_FDV0;
 138     ewtabscale       = _mm_set1_pd(fr->ic->tabq_scale);
 139     ewtabhalfspace   = _mm_set1_pd(0.5/fr->ic->tabq_scale);
 140
 141     /* Setup water-specific parameters */
 142     inr              = nlist->iinr[0];
 143     iq0              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+0]));
 144     iq1              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+1]));
 145     iq2              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+2]));
 146     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 147
 148     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 149     rcutoff_scalar   = fr->rcoulomb;
 150     rcutoff          = _mm_set1_pd(rcutoff_scalar);
 151     rcutoff2         = _mm_mul_pd(rcutoff,rcutoff);
 152
 153     sh_vdw_invrcut6  = _mm_set1_pd(fr->ic->sh_invrc6);
 154     rvdw             = _mm_set1_pd(fr->rvdw);
 155
 156     /* Avoid stupid compiler warnings */
 157     jnrA = jnrB = 0;
 158     j_coord_offsetA = 0;
 159     j_coord_offsetB = 0;
 160
 161     outeriter        = 0;
 162     inneriter        = 0;
 163
 164     /* Start outer loop over neighborlists */
 165     for(iidx=0; iidx<nri; iidx++)
 166     {
 167         /* Load shift vector for this list */
 168         i_shift_offset   = DIM*shiftidx[iidx];
 169
 170         /* Load limits for loop over neighbors */
 171         j_index_start    = jindex[iidx];
 172         j_index_end      = jindex[iidx+1];
 173
 174         /* Get outer coordinate index */
 175         inr              = iinr[iidx];
 176         i_coord_offset   = DIM*inr;
 177
 178         /* Load i particle coords and add shift vector */
 179         gmx_mm_load_shift_and_3rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
 180                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 181
 182         fix0             = _mm_setzero_pd();
 183         fiy0             = _mm_setzero_pd();
 184         fiz0             = _mm_setzero_pd();
 185         fix1             = _mm_setzero_pd();
 186         fiy1             = _mm_setzero_pd();
 187         fiz1             = _mm_setzero_pd();
 188         fix2             = _mm_setzero_pd();
 189         fiy2             = _mm_setzero_pd();
 190         fiz2             = _mm_setzero_pd();
 191
 192         /* Reset potential sums */
 193         velecsum         = _mm_setzero_pd();
 194         vvdwsum          = _mm_setzero_pd();
 195
 196         /* Start inner kernel loop */
 197         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 198         {
 199
 200             /* Get j neighbor index, and coordinate index */
 201             jnrA             = jjnr[jidx];
 202             jnrB             = jjnr[jidx+1];
 203             j_coord_offsetA  = DIM*jnrA;
 204             j_coord_offsetB  = DIM*jnrB;
 205
 206             /* load j atom coordinates */
 207             gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 208                                               &jx0,&jy0,&jz0);
 209
 210             /* Calculate displacement vector */
 211             dx00             = _mm_sub_pd(ix0,jx0);
 212             dy00             = _mm_sub_pd(iy0,jy0);
 213             dz00             = _mm_sub_pd(iz0,jz0);
 214             dx10             = _mm_sub_pd(ix1,jx0);
 215             dy10             = _mm_sub_pd(iy1,jy0);
 216             dz10             = _mm_sub_pd(iz1,jz0);
 217             dx20             = _mm_sub_pd(ix2,jx0);
 218             dy20             = _mm_sub_pd(iy2,jy0);
 219             dz20             = _mm_sub_pd(iz2,jz0);
 220
 221             /* Calculate squared distance and things based on it */
 222             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 223             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
 224             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
 225
 226             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 227             rinv10           = gmx_mm_invsqrt_pd(rsq10);
 228             rinv20           = gmx_mm_invsqrt_pd(rsq20);
 229
 230             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 231             rinvsq10         = _mm_mul_pd(rinv10,rinv10);
 232             rinvsq20         = _mm_mul_pd(rinv20,rinv20);
 233
 234             /* Load parameters for j particles */
 235             jq0              = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
 236             vdwjidx0A        = 2*vdwtype[jnrA+0];
 237             vdwjidx0B        = 2*vdwtype[jnrB+0];
 238
 239             fjx0             = _mm_setzero_pd();
 240             fjy0             = _mm_setzero_pd();
 241             fjz0             = _mm_setzero_pd();
 242
 243             /**************************
 244              * CALCULATE INTERACTIONS *
 245              **************************/
 246
 247             if (gmx_mm_any_lt(rsq00,rcutoff2))
 248             {
 249
 250             r00              = _mm_mul_pd(rsq00,rinv00);
 251
 252             /* Compute parameters for interactions between i and j atoms */
 253             qq00             = _mm_mul_pd(iq0,jq0);
 254             gmx_mm_load_2pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,
 255                                          vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
 256
 257             c6grid_00       = gmx_mm_load_2real_swizzle_pd(vdwgridparam+vdwioffset0+vdwjidx0A,
 258                                                                vdwgridparam+vdwioffset0+vdwjidx0B);
 259
 260             /* EWALD ELECTROSTATICS */
 261
 262             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 263             ewrt             = _mm_mul_pd(r00,ewtabscale);
 264             ewitab           = _mm_cvttpd_epi32(ewrt);
 265             eweps            = _mm_sub_pd(ewrt,_mm_cvtepi32_pd(ewitab));
 266             ewitab           = _mm_slli_epi32(ewitab,2);
 267             ewtabF           = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 268             ewtabD           = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,1) );
 269             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 270             ewtabV           = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,0) +2);
 271             ewtabFn          = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,1) +2);
 272             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 273             felec            = _mm_add_pd(ewtabF,_mm_mul_pd(eweps,ewtabD));
 274             velec            = _mm_sub_pd(ewtabV,_mm_mul_pd(_mm_mul_pd(ewtabhalfspace,eweps),_mm_add_pd(ewtabF,felec)));
 275             velec            = _mm_mul_pd(qq00,_mm_sub_pd(_mm_sub_pd(rinv00,sh_ewald),velec));
 276             felec            = _mm_mul_pd(_mm_mul_pd(qq00,rinv00),_mm_sub_pd(rinvsq00,felec));
 277
 278             /* Analytical LJ-PME */
 279             rinvsix          = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 280             ewcljrsq         = _mm_mul_pd(ewclj2,rsq00);
 281             ewclj6           = _mm_mul_pd(ewclj2,_mm_mul_pd(ewclj2,ewclj2));
 282             exponent         = gmx_simd_exp_d(ewcljrsq);
 283             /* poly = exp(-(beta*r)^2) * (1 + (beta*r)^2 + (beta*r)^4 /2) */
 284             poly             = _mm_mul_pd(exponent,_mm_add_pd(_mm_sub_pd(one,ewcljrsq),_mm_mul_pd(_mm_mul_pd(ewcljrsq,ewcljrsq),one_half)));
 285             /* vvdw6 = [C6 - C6grid * (1-poly)]/r6 */
 286             vvdw6            = _mm_mul_pd(_mm_sub_pd(c6_00,_mm_mul_pd(c6grid_00,_mm_sub_pd(one,poly))),rinvsix);
 287             vvdw12           = _mm_mul_pd(c12_00,_mm_mul_pd(rinvsix,rinvsix));
 288             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),
 289                                _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));
 290             /* fvdw = vvdw12/r - (vvdw6/r + (C6grid * exponent * beta^6)/r) */
 291             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);
 292
 293             cutoff_mask      = _mm_cmplt_pd(rsq00,rcutoff2);
 294
 295             /* Update potential sum for this i atom from the interaction with this j atom. */
 296             velec            = _mm_and_pd(velec,cutoff_mask);
 297             velecsum         = _mm_add_pd(velecsum,velec);
 298             vvdw             = _mm_and_pd(vvdw,cutoff_mask);
 299             vvdwsum          = _mm_add_pd(vvdwsum,vvdw);
 300
 301             fscal            = _mm_add_pd(felec,fvdw);
 302
 303             fscal            = _mm_and_pd(fscal,cutoff_mask);
 304
 305             /* Calculate temporary vectorial force */
 306             tx               = _mm_mul_pd(fscal,dx00);
 307             ty               = _mm_mul_pd(fscal,dy00);
 308             tz               = _mm_mul_pd(fscal,dz00);
 309
 310             /* Update vectorial force */
 311             fix0             = _mm_add_pd(fix0,tx);
 312             fiy0             = _mm_add_pd(fiy0,ty);
 313             fiz0             = _mm_add_pd(fiz0,tz);
 314
 315             fjx0             = _mm_add_pd(fjx0,tx);
 316             fjy0             = _mm_add_pd(fjy0,ty);
 317             fjz0             = _mm_add_pd(fjz0,tz);
 318
 319             }
 320
 321             /**************************
 322              * CALCULATE INTERACTIONS *
 323              **************************/
 324
 325             if (gmx_mm_any_lt(rsq10,rcutoff2))
 326             {
 327
 328             r10              = _mm_mul_pd(rsq10,rinv10);
 329
 330             /* Compute parameters for interactions between i and j atoms */
 331             qq10             = _mm_mul_pd(iq1,jq0);
 332
 333             /* EWALD ELECTROSTATICS */
 334
 335             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 336             ewrt             = _mm_mul_pd(r10,ewtabscale);
 337             ewitab           = _mm_cvttpd_epi32(ewrt);
 338             eweps            = _mm_sub_pd(ewrt,_mm_cvtepi32_pd(ewitab));
 339             ewitab           = _mm_slli_epi32(ewitab,2);
 340             ewtabF           = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 341             ewtabD           = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,1) );
 342             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 343             ewtabV           = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,0) +2);
 344             ewtabFn          = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,1) +2);
 345             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 346             felec            = _mm_add_pd(ewtabF,_mm_mul_pd(eweps,ewtabD));
 347             velec            = _mm_sub_pd(ewtabV,_mm_mul_pd(_mm_mul_pd(ewtabhalfspace,eweps),_mm_add_pd(ewtabF,felec)));
 348             velec            = _mm_mul_pd(qq10,_mm_sub_pd(_mm_sub_pd(rinv10,sh_ewald),velec));
 349             felec            = _mm_mul_pd(_mm_mul_pd(qq10,rinv10),_mm_sub_pd(rinvsq10,felec));
 350
 351             cutoff_mask      = _mm_cmplt_pd(rsq10,rcutoff2);
 352
 353             /* Update potential sum for this i atom from the interaction with this j atom. */
 354             velec            = _mm_and_pd(velec,cutoff_mask);
 355             velecsum         = _mm_add_pd(velecsum,velec);
 356
 357             fscal            = felec;
 358
 359             fscal            = _mm_and_pd(fscal,cutoff_mask);
 360
 361             /* Calculate temporary vectorial force */
 362             tx               = _mm_mul_pd(fscal,dx10);
 363             ty               = _mm_mul_pd(fscal,dy10);
 364             tz               = _mm_mul_pd(fscal,dz10);
 365
 366             /* Update vectorial force */
 367             fix1             = _mm_add_pd(fix1,tx);
 368             fiy1             = _mm_add_pd(fiy1,ty);
 369             fiz1             = _mm_add_pd(fiz1,tz);
 370
 371             fjx0             = _mm_add_pd(fjx0,tx);
 372             fjy0             = _mm_add_pd(fjy0,ty);
 373             fjz0             = _mm_add_pd(fjz0,tz);
 374
 375             }
 376
 377             /**************************
 378              * CALCULATE INTERACTIONS *
 379              **************************/
 380
 381             if (gmx_mm_any_lt(rsq20,rcutoff2))
 382             {
 383
 384             r20              = _mm_mul_pd(rsq20,rinv20);
 385
 386             /* Compute parameters for interactions between i and j atoms */
 387             qq20             = _mm_mul_pd(iq2,jq0);
 388
 389             /* EWALD ELECTROSTATICS */
 390
 391             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 392             ewrt             = _mm_mul_pd(r20,ewtabscale);
 393             ewitab           = _mm_cvttpd_epi32(ewrt);
 394             eweps            = _mm_sub_pd(ewrt,_mm_cvtepi32_pd(ewitab));
 395             ewitab           = _mm_slli_epi32(ewitab,2);
 396             ewtabF           = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 397             ewtabD           = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,1) );
 398             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 399             ewtabV           = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,0) +2);
 400             ewtabFn          = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,1) +2);
 401             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 402             felec            = _mm_add_pd(ewtabF,_mm_mul_pd(eweps,ewtabD));
 403             velec            = _mm_sub_pd(ewtabV,_mm_mul_pd(_mm_mul_pd(ewtabhalfspace,eweps),_mm_add_pd(ewtabF,felec)));
 404             velec            = _mm_mul_pd(qq20,_mm_sub_pd(_mm_sub_pd(rinv20,sh_ewald),velec));
 405             felec            = _mm_mul_pd(_mm_mul_pd(qq20,rinv20),_mm_sub_pd(rinvsq20,felec));
 406
 407             cutoff_mask      = _mm_cmplt_pd(rsq20,rcutoff2);
 408
 409             /* Update potential sum for this i atom from the interaction with this j atom. */
 410             velec            = _mm_and_pd(velec,cutoff_mask);
 411             velecsum         = _mm_add_pd(velecsum,velec);
 412
 413             fscal            = felec;
 414
 415             fscal            = _mm_and_pd(fscal,cutoff_mask);
 416
 417             /* Calculate temporary vectorial force */
 418             tx               = _mm_mul_pd(fscal,dx20);
 419             ty               = _mm_mul_pd(fscal,dy20);
 420             tz               = _mm_mul_pd(fscal,dz20);
 421
 422             /* Update vectorial force */
 423             fix2             = _mm_add_pd(fix2,tx);
 424             fiy2             = _mm_add_pd(fiy2,ty);
 425             fiz2             = _mm_add_pd(fiz2,tz);
 426
 427             fjx0             = _mm_add_pd(fjx0,tx);
 428             fjy0             = _mm_add_pd(fjy0,ty);
 429             fjz0             = _mm_add_pd(fjz0,tz);
 430
 431             }
 432
 433             gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,fjx0,fjy0,fjz0);
 434
 435             /* Inner loop uses 177 flops */
 436         }
 437
 438         if(jidx<j_index_end)
 439         {
 440
 441             jnrA             = jjnr[jidx];
 442             j_coord_offsetA  = DIM*jnrA;
 443
 444             /* load j atom coordinates */
 445             gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
 446                                               &jx0,&jy0,&jz0);
 447
 448             /* Calculate displacement vector */
 449             dx00             = _mm_sub_pd(ix0,jx0);
 450             dy00             = _mm_sub_pd(iy0,jy0);
 451             dz00             = _mm_sub_pd(iz0,jz0);
 452             dx10             = _mm_sub_pd(ix1,jx0);
 453             dy10             = _mm_sub_pd(iy1,jy0);
 454             dz10             = _mm_sub_pd(iz1,jz0);
 455             dx20             = _mm_sub_pd(ix2,jx0);
 456             dy20             = _mm_sub_pd(iy2,jy0);
 457             dz20             = _mm_sub_pd(iz2,jz0);
 458
 459             /* Calculate squared distance and things based on it */
 460             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 461             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
 462             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
 463
 464             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 465             rinv10           = gmx_mm_invsqrt_pd(rsq10);
 466             rinv20           = gmx_mm_invsqrt_pd(rsq20);
 467
 468             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 469             rinvsq10         = _mm_mul_pd(rinv10,rinv10);
 470             rinvsq20         = _mm_mul_pd(rinv20,rinv20);
 471
 472             /* Load parameters for j particles */
 473             jq0              = _mm_load_sd(charge+jnrA+0);
 474             vdwjidx0A        = 2*vdwtype[jnrA+0];
 475
 476             fjx0             = _mm_setzero_pd();
 477             fjy0             = _mm_setzero_pd();
 478             fjz0             = _mm_setzero_pd();
 479
 480             /**************************
 481              * CALCULATE INTERACTIONS *
 482              **************************/
 483
 484             if (gmx_mm_any_lt(rsq00,rcutoff2))
 485             {
 486
 487             r00              = _mm_mul_pd(rsq00,rinv00);
 488
 489             /* Compute parameters for interactions between i and j atoms */
 490             qq00             = _mm_mul_pd(iq0,jq0);
 491             gmx_mm_load_1pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,&c6_00,&c12_00);
 492
 493             c6grid_00       = gmx_mm_load_1real_pd(vdwgridparam+vdwioffset0+vdwjidx0A);
 494
 495             /* EWALD ELECTROSTATICS */
 496
 497             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 498             ewrt             = _mm_mul_pd(r00,ewtabscale);
 499             ewitab           = _mm_cvttpd_epi32(ewrt);
 500             eweps            = _mm_sub_pd(ewrt,_mm_cvtepi32_pd(ewitab));
 501             ewitab           = _mm_slli_epi32(ewitab,2);
 502             ewtabF           = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 503             ewtabD           = _mm_setzero_pd();
 504             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 505             ewtabV           = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,0) +2);
 506             ewtabFn          = _mm_setzero_pd();
 507             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 508             felec            = _mm_add_pd(ewtabF,_mm_mul_pd(eweps,ewtabD));
 509             velec            = _mm_sub_pd(ewtabV,_mm_mul_pd(_mm_mul_pd(ewtabhalfspace,eweps),_mm_add_pd(ewtabF,felec)));
 510             velec            = _mm_mul_pd(qq00,_mm_sub_pd(_mm_sub_pd(rinv00,sh_ewald),velec));
 511             felec            = _mm_mul_pd(_mm_mul_pd(qq00,rinv00),_mm_sub_pd(rinvsq00,felec));
 512
 513             /* Analytical LJ-PME */
 514             rinvsix          = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 515             ewcljrsq         = _mm_mul_pd(ewclj2,rsq00);
 516             ewclj6           = _mm_mul_pd(ewclj2,_mm_mul_pd(ewclj2,ewclj2));
 517             exponent         = gmx_simd_exp_d(ewcljrsq);
 518             /* poly = exp(-(beta*r)^2) * (1 + (beta*r)^2 + (beta*r)^4 /2) */
 519             poly             = _mm_mul_pd(exponent,_mm_add_pd(_mm_sub_pd(one,ewcljrsq),_mm_mul_pd(_mm_mul_pd(ewcljrsq,ewcljrsq),one_half)));
 520             /* vvdw6 = [C6 - C6grid * (1-poly)]/r6 */
 521             vvdw6            = _mm_mul_pd(_mm_sub_pd(c6_00,_mm_mul_pd(c6grid_00,_mm_sub_pd(one,poly))),rinvsix);
 522             vvdw12           = _mm_mul_pd(c12_00,_mm_mul_pd(rinvsix,rinvsix));
 523             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),
 524                                _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));
 525             /* fvdw = vvdw12/r - (vvdw6/r + (C6grid * exponent * beta^6)/r) */
 526             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);
 527
 528             cutoff_mask      = _mm_cmplt_pd(rsq00,rcutoff2);
 529
 530             /* Update potential sum for this i atom from the interaction with this j atom. */
 531             velec            = _mm_and_pd(velec,cutoff_mask);
 532             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 533             velecsum         = _mm_add_pd(velecsum,velec);
 534             vvdw             = _mm_and_pd(vvdw,cutoff_mask);
 535             vvdw             = _mm_unpacklo_pd(vvdw,_mm_setzero_pd());
 536             vvdwsum          = _mm_add_pd(vvdwsum,vvdw);
 537
 538             fscal            = _mm_add_pd(felec,fvdw);
 539
 540             fscal            = _mm_and_pd(fscal,cutoff_mask);
 541
 542             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 543
 544             /* Calculate temporary vectorial force */
 545             tx               = _mm_mul_pd(fscal,dx00);
 546             ty               = _mm_mul_pd(fscal,dy00);
 547             tz               = _mm_mul_pd(fscal,dz00);
 548
 549             /* Update vectorial force */
 550             fix0             = _mm_add_pd(fix0,tx);
 551             fiy0             = _mm_add_pd(fiy0,ty);
 552             fiz0             = _mm_add_pd(fiz0,tz);
 553
 554             fjx0             = _mm_add_pd(fjx0,tx);
 555             fjy0             = _mm_add_pd(fjy0,ty);
 556             fjz0             = _mm_add_pd(fjz0,tz);
 557
 558             }
 559
 560             /**************************
 561              * CALCULATE INTERACTIONS *
 562              **************************/
 563
 564             if (gmx_mm_any_lt(rsq10,rcutoff2))
 565             {
 566
 567             r10              = _mm_mul_pd(rsq10,rinv10);
 568
 569             /* Compute parameters for interactions between i and j atoms */
 570             qq10             = _mm_mul_pd(iq1,jq0);
 571
 572             /* EWALD ELECTROSTATICS */
 573
 574             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 575             ewrt             = _mm_mul_pd(r10,ewtabscale);
 576             ewitab           = _mm_cvttpd_epi32(ewrt);
 577             eweps            = _mm_sub_pd(ewrt,_mm_cvtepi32_pd(ewitab));
 578             ewitab           = _mm_slli_epi32(ewitab,2);
 579             ewtabF           = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 580             ewtabD           = _mm_setzero_pd();
 581             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 582             ewtabV           = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,0) +2);
 583             ewtabFn          = _mm_setzero_pd();
 584             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 585             felec            = _mm_add_pd(ewtabF,_mm_mul_pd(eweps,ewtabD));
 586             velec            = _mm_sub_pd(ewtabV,_mm_mul_pd(_mm_mul_pd(ewtabhalfspace,eweps),_mm_add_pd(ewtabF,felec)));
 587             velec            = _mm_mul_pd(qq10,_mm_sub_pd(_mm_sub_pd(rinv10,sh_ewald),velec));
 588             felec            = _mm_mul_pd(_mm_mul_pd(qq10,rinv10),_mm_sub_pd(rinvsq10,felec));
 589
 590             cutoff_mask      = _mm_cmplt_pd(rsq10,rcutoff2);
 591
 592             /* Update potential sum for this i atom from the interaction with this j atom. */
 593             velec            = _mm_and_pd(velec,cutoff_mask);
 594             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 595             velecsum         = _mm_add_pd(velecsum,velec);
 596
 597             fscal            = felec;
 598
 599             fscal            = _mm_and_pd(fscal,cutoff_mask);
 600
 601             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 602
 603             /* Calculate temporary vectorial force */
 604             tx               = _mm_mul_pd(fscal,dx10);
 605             ty               = _mm_mul_pd(fscal,dy10);
 606             tz               = _mm_mul_pd(fscal,dz10);
 607
 608             /* Update vectorial force */
 609             fix1             = _mm_add_pd(fix1,tx);
 610             fiy1             = _mm_add_pd(fiy1,ty);
 611             fiz1             = _mm_add_pd(fiz1,tz);
 612
 613             fjx0             = _mm_add_pd(fjx0,tx);
 614             fjy0             = _mm_add_pd(fjy0,ty);
 615             fjz0             = _mm_add_pd(fjz0,tz);
 616
 617             }
 618
 619             /**************************
 620              * CALCULATE INTERACTIONS *
 621              **************************/
 622
 623             if (gmx_mm_any_lt(rsq20,rcutoff2))
 624             {
 625
 626             r20              = _mm_mul_pd(rsq20,rinv20);
 627
 628             /* Compute parameters for interactions between i and j atoms */
 629             qq20             = _mm_mul_pd(iq2,jq0);
 630
 631             /* EWALD ELECTROSTATICS */
 632
 633             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 634             ewrt             = _mm_mul_pd(r20,ewtabscale);
 635             ewitab           = _mm_cvttpd_epi32(ewrt);
 636             eweps            = _mm_sub_pd(ewrt,_mm_cvtepi32_pd(ewitab));
 637             ewitab           = _mm_slli_epi32(ewitab,2);
 638             ewtabF           = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 639             ewtabD           = _mm_setzero_pd();
 640             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 641             ewtabV           = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,0) +2);
 642             ewtabFn          = _mm_setzero_pd();
 643             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 644             felec            = _mm_add_pd(ewtabF,_mm_mul_pd(eweps,ewtabD));
 645             velec            = _mm_sub_pd(ewtabV,_mm_mul_pd(_mm_mul_pd(ewtabhalfspace,eweps),_mm_add_pd(ewtabF,felec)));
 646             velec            = _mm_mul_pd(qq20,_mm_sub_pd(_mm_sub_pd(rinv20,sh_ewald),velec));
 647             felec            = _mm_mul_pd(_mm_mul_pd(qq20,rinv20),_mm_sub_pd(rinvsq20,felec));
 648
 649             cutoff_mask      = _mm_cmplt_pd(rsq20,rcutoff2);
 650
 651             /* Update potential sum for this i atom from the interaction with this j atom. */
 652             velec            = _mm_and_pd(velec,cutoff_mask);
 653             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 654             velecsum         = _mm_add_pd(velecsum,velec);
 655
 656             fscal            = felec;
 657
 658             fscal            = _mm_and_pd(fscal,cutoff_mask);
 659
 660             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 661
 662             /* Calculate temporary vectorial force */
 663             tx               = _mm_mul_pd(fscal,dx20);
 664             ty               = _mm_mul_pd(fscal,dy20);
 665             tz               = _mm_mul_pd(fscal,dz20);
 666
 667             /* Update vectorial force */
 668             fix2             = _mm_add_pd(fix2,tx);
 669             fiy2             = _mm_add_pd(fiy2,ty);
 670             fiz2             = _mm_add_pd(fiz2,tz);
 671
 672             fjx0             = _mm_add_pd(fjx0,tx);
 673             fjy0             = _mm_add_pd(fjy0,ty);
 674             fjz0             = _mm_add_pd(fjz0,tz);
 675
 676             }
 677
 678             gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,fjx0,fjy0,fjz0);
 679
 680             /* Inner loop uses 177 flops */
 681         }
 682
 683         /* End of innermost loop */
 684
 685         gmx_mm_update_iforce_3atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
 686                                               f+i_coord_offset,fshift+i_shift_offset);
 687
 688         ggid                        = gid[iidx];
 689         /* Update potential energies */
 690         gmx_mm_update_1pot_pd(velecsum,kernel_data->energygrp_elec+ggid);
 691         gmx_mm_update_1pot_pd(vvdwsum,kernel_data->energygrp_vdw+ggid);
 692
 693         /* Increment number of inner iterations */
 694         inneriter                  += j_index_end - j_index_start;
 695
 696         /* Outer loop uses 20 flops */
 697     }
 698
 699     /* Increment number of outer iterations */
 700     outeriter        += nri;
 701
 702     /* Update outer/inner flops */
 703
 704     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3_VF,outeriter*20 + inneriter*177);
 705 }
 706 /*
 707  * Gromacs nonbonded kernel:   nb_kernel_ElecEwSh_VdwLJEwSh_GeomW3P1_F_sse2_double
 708  * Electrostatics interaction: Ewald
 709  * VdW interaction:            LJEwald
 710  * Geometry:                   Water3-Particle
 711  * Calculate force/pot:        Force
 712  */
 713 void
 714 nb_kernel_ElecEwSh_VdwLJEwSh_GeomW3P1_F_sse2_double
 715                     (t_nblist                    * gmx_restrict       nlist,
 716                      rvec                        * gmx_restrict          xx,
 717                      rvec                        * gmx_restrict          ff,
 718                      t_forcerec                  * gmx_restrict          fr,
 719                      t_mdatoms                   * gmx_restrict     mdatoms,
 720                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 721                      t_nrnb                      * gmx_restrict        nrnb)
 722 {
 723     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 724      * just 0 for non-waters.
 725      * Suffixes A,B refer to j loop unrolling done with SSE double precision, e.g. for the two different
 726      * jnr indices corresponding to data put in the four positions in the SIMD register.
 727      */
 728     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 729     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 730     int              jnrA,jnrB;
 731     int              j_coord_offsetA,j_coord_offsetB;
 732     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 733     real             rcutoff_scalar;
 734     real             *shiftvec,*fshift,*x,*f;
 735     __m128d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 736     int              vdwioffset0;
 737     __m128d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 738     int              vdwioffset1;
 739     __m128d          ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
 740     int              vdwioffset2;
 741     __m128d          ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
 742     int              vdwjidx0A,vdwjidx0B;
 743     __m128d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 744     __m128d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 745     __m128d          dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
 746     __m128d          dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
 747     __m128d          velec,felec,velecsum,facel,crf,krf,krf2;
 748     real             *charge;
 749     int              nvdwtype;
 750     __m128d          rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 751     int              *vdwtype;
 752     real             *vdwparam;
 753     __m128d          one_sixth   = _mm_set1_pd(1.0/6.0);
 754     __m128d          one_twelfth = _mm_set1_pd(1.0/12.0);
 755     __m128d           c6grid_00;
 756     __m128d           c6grid_10;
 757     __m128d           c6grid_20;
 758     __m128d           ewclj,ewclj2,ewclj6,ewcljrsq,poly,exponent,f6A,f6B,sh_lj_ewald;
 759     real             *vdwgridparam;
 760     __m128d           one_half = _mm_set1_pd(0.5);
 761     __m128d           minus_one = _mm_set1_pd(-1.0);
 762     __m128i          ewitab;
 763     __m128d          ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
 764     real             *ewtab;
 765     __m128d          dummy_mask,cutoff_mask;
 766     __m128d          signbit   = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
 767     __m128d          one     = _mm_set1_pd(1.0);
 768     __m128d          two     = _mm_set1_pd(2.0);
 769     x                = xx[0];
 770     f                = ff[0];
 771
 772     nri              = nlist->nri;
 773     iinr             = nlist->iinr;
 774     jindex           = nlist->jindex;
 775     jjnr             = nlist->jjnr;
 776     shiftidx         = nlist->shift;
 777     gid              = nlist->gid;
 778     shiftvec         = fr->shift_vec[0];
 779     fshift           = fr->fshift[0];
 780     facel            = _mm_set1_pd(fr->epsfac);
 781     charge           = mdatoms->chargeA;
 782     nvdwtype         = fr->ntype;
 783     vdwparam         = fr->nbfp;
 784     vdwtype          = mdatoms->typeA;
 785     vdwgridparam     = fr->ljpme_c6grid;
 786     sh_lj_ewald      = _mm_set1_pd(fr->ic->sh_lj_ewald);
 787     ewclj            = _mm_set1_pd(fr->ewaldcoeff_lj);
 788     ewclj2           = _mm_mul_pd(minus_one,_mm_mul_pd(ewclj,ewclj));
 789
 790     sh_ewald         = _mm_set1_pd(fr->ic->sh_ewald);
 791     ewtab            = fr->ic->tabq_coul_F;
 792     ewtabscale       = _mm_set1_pd(fr->ic->tabq_scale);
 793     ewtabhalfspace   = _mm_set1_pd(0.5/fr->ic->tabq_scale);
 794
 795     /* Setup water-specific parameters */
 796     inr              = nlist->iinr[0];
 797     iq0              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+0]));
 798     iq1              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+1]));
 799     iq2              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+2]));
 800     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 801
 802     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 803     rcutoff_scalar   = fr->rcoulomb;
 804     rcutoff          = _mm_set1_pd(rcutoff_scalar);
 805     rcutoff2         = _mm_mul_pd(rcutoff,rcutoff);
 806
 807     sh_vdw_invrcut6  = _mm_set1_pd(fr->ic->sh_invrc6);
 808     rvdw             = _mm_set1_pd(fr->rvdw);
 809
 810     /* Avoid stupid compiler warnings */
 811     jnrA = jnrB = 0;
 812     j_coord_offsetA = 0;
 813     j_coord_offsetB = 0;
 814
 815     outeriter        = 0;
 816     inneriter        = 0;
 817
 818     /* Start outer loop over neighborlists */
 819     for(iidx=0; iidx<nri; iidx++)
 820     {
 821         /* Load shift vector for this list */
 822         i_shift_offset   = DIM*shiftidx[iidx];
 823
 824         /* Load limits for loop over neighbors */
 825         j_index_start    = jindex[iidx];
 826         j_index_end      = jindex[iidx+1];
 827
 828         /* Get outer coordinate index */
 829         inr              = iinr[iidx];
 830         i_coord_offset   = DIM*inr;
 831
 832         /* Load i particle coords and add shift vector */
 833         gmx_mm_load_shift_and_3rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
 834                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 835
 836         fix0             = _mm_setzero_pd();
 837         fiy0             = _mm_setzero_pd();
 838         fiz0             = _mm_setzero_pd();
 839         fix1             = _mm_setzero_pd();
 840         fiy1             = _mm_setzero_pd();
 841         fiz1             = _mm_setzero_pd();
 842         fix2             = _mm_setzero_pd();
 843         fiy2             = _mm_setzero_pd();
 844         fiz2             = _mm_setzero_pd();
 845
 846         /* Start inner kernel loop */
 847         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 848         {
 849
 850             /* Get j neighbor index, and coordinate index */
 851             jnrA             = jjnr[jidx];
 852             jnrB             = jjnr[jidx+1];
 853             j_coord_offsetA  = DIM*jnrA;
 854             j_coord_offsetB  = DIM*jnrB;
 855
 856             /* load j atom coordinates */
 857             gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 858                                               &jx0,&jy0,&jz0);
 859
 860             /* Calculate displacement vector */
 861             dx00             = _mm_sub_pd(ix0,jx0);
 862             dy00             = _mm_sub_pd(iy0,jy0);
 863             dz00             = _mm_sub_pd(iz0,jz0);
 864             dx10             = _mm_sub_pd(ix1,jx0);
 865             dy10             = _mm_sub_pd(iy1,jy0);
 866             dz10             = _mm_sub_pd(iz1,jz0);
 867             dx20             = _mm_sub_pd(ix2,jx0);
 868             dy20             = _mm_sub_pd(iy2,jy0);
 869             dz20             = _mm_sub_pd(iz2,jz0);
 870
 871             /* Calculate squared distance and things based on it */
 872             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 873             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
 874             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
 875
 876             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 877             rinv10           = gmx_mm_invsqrt_pd(rsq10);
 878             rinv20           = gmx_mm_invsqrt_pd(rsq20);
 879
 880             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 881             rinvsq10         = _mm_mul_pd(rinv10,rinv10);
 882             rinvsq20         = _mm_mul_pd(rinv20,rinv20);
 883
 884             /* Load parameters for j particles */
 885             jq0              = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
 886             vdwjidx0A        = 2*vdwtype[jnrA+0];
 887             vdwjidx0B        = 2*vdwtype[jnrB+0];
 888
 889             fjx0             = _mm_setzero_pd();
 890             fjy0             = _mm_setzero_pd();
 891             fjz0             = _mm_setzero_pd();
 892
 893             /**************************
 894              * CALCULATE INTERACTIONS *
 895              **************************/
 896
 897             if (gmx_mm_any_lt(rsq00,rcutoff2))
 898             {
 899
 900             r00              = _mm_mul_pd(rsq00,rinv00);
 901
 902             /* Compute parameters for interactions between i and j atoms */
 903             qq00             = _mm_mul_pd(iq0,jq0);
 904             gmx_mm_load_2pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,
 905                                          vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
 906
 907             c6grid_00       = gmx_mm_load_2real_swizzle_pd(vdwgridparam+vdwioffset0+vdwjidx0A,
 908                                                                vdwgridparam+vdwioffset0+vdwjidx0B);
 909
 910             /* EWALD ELECTROSTATICS */
 911
 912             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 913             ewrt             = _mm_mul_pd(r00,ewtabscale);
 914             ewitab           = _mm_cvttpd_epi32(ewrt);
 915             eweps            = _mm_sub_pd(ewrt,_mm_cvtepi32_pd(ewitab));
 916             gmx_mm_load_2pair_swizzle_pd(ewtab+gmx_mm_extract_epi32(ewitab,0),ewtab+gmx_mm_extract_epi32(ewitab,1),
 917                                          &ewtabF,&ewtabFn);
 918             felec            = _mm_add_pd(_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF),_mm_mul_pd(eweps,ewtabFn));
 919             felec            = _mm_mul_pd(_mm_mul_pd(qq00,rinv00),_mm_sub_pd(rinvsq00,felec));
 920
 921             /* Analytical LJ-PME */
 922             rinvsix          = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 923             ewcljrsq         = _mm_mul_pd(ewclj2,rsq00);
 924             ewclj6           = _mm_mul_pd(ewclj2,_mm_mul_pd(ewclj2,ewclj2));
 925             exponent         = gmx_simd_exp_d(ewcljrsq);
 926             /* poly = exp(-(beta*r)^2) * (1 + (beta*r)^2 + (beta*r)^4 /2) */
 927             poly             = _mm_mul_pd(exponent,_mm_add_pd(_mm_sub_pd(one,ewcljrsq),_mm_mul_pd(_mm_mul_pd(ewcljrsq,ewcljrsq),one_half)));
 928             /* f6A = 6 * C6grid * (1 - poly) */
 929             f6A              = _mm_mul_pd(c6grid_00,_mm_sub_pd(one,poly));
 930             /* f6B = C6grid * exponent * beta^6 */
 931             f6B              = _mm_mul_pd(_mm_mul_pd(c6grid_00,one_sixth),_mm_mul_pd(exponent,ewclj6));
 932             /* fvdw = 12*C12/r13 - ((6*C6 - f6A)/r6 + f6B)/r */
 933             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);
 934
 935             cutoff_mask      = _mm_cmplt_pd(rsq00,rcutoff2);
 936
 937             fscal            = _mm_add_pd(felec,fvdw);
 938
 939             fscal            = _mm_and_pd(fscal,cutoff_mask);
 940
 941             /* Calculate temporary vectorial force */
 942             tx               = _mm_mul_pd(fscal,dx00);
 943             ty               = _mm_mul_pd(fscal,dy00);
 944             tz               = _mm_mul_pd(fscal,dz00);
 945
 946             /* Update vectorial force */
 947             fix0             = _mm_add_pd(fix0,tx);
 948             fiy0             = _mm_add_pd(fiy0,ty);
 949             fiz0             = _mm_add_pd(fiz0,tz);
 950
 951             fjx0             = _mm_add_pd(fjx0,tx);
 952             fjy0             = _mm_add_pd(fjy0,ty);
 953             fjz0             = _mm_add_pd(fjz0,tz);
 954
 955             }
 956
 957             /**************************
 958              * CALCULATE INTERACTIONS *
 959              **************************/
 960
 961             if (gmx_mm_any_lt(rsq10,rcutoff2))
 962             {
 963
 964             r10              = _mm_mul_pd(rsq10,rinv10);
 965
 966             /* Compute parameters for interactions between i and j atoms */
 967             qq10             = _mm_mul_pd(iq1,jq0);
 968
 969             /* EWALD ELECTROSTATICS */
 970
 971             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 972             ewrt             = _mm_mul_pd(r10,ewtabscale);
 973             ewitab           = _mm_cvttpd_epi32(ewrt);
 974             eweps            = _mm_sub_pd(ewrt,_mm_cvtepi32_pd(ewitab));
 975             gmx_mm_load_2pair_swizzle_pd(ewtab+gmx_mm_extract_epi32(ewitab,0),ewtab+gmx_mm_extract_epi32(ewitab,1),
 976                                          &ewtabF,&ewtabFn);
 977             felec            = _mm_add_pd(_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF),_mm_mul_pd(eweps,ewtabFn));
 978             felec            = _mm_mul_pd(_mm_mul_pd(qq10,rinv10),_mm_sub_pd(rinvsq10,felec));
 979
 980             cutoff_mask      = _mm_cmplt_pd(rsq10,rcutoff2);
 981
 982             fscal            = felec;
 983
 984             fscal            = _mm_and_pd(fscal,cutoff_mask);
 985
 986             /* Calculate temporary vectorial force */
 987             tx               = _mm_mul_pd(fscal,dx10);
 988             ty               = _mm_mul_pd(fscal,dy10);
 989             tz               = _mm_mul_pd(fscal,dz10);
 990
 991             /* Update vectorial force */
 992             fix1             = _mm_add_pd(fix1,tx);
 993             fiy1             = _mm_add_pd(fiy1,ty);
 994             fiz1             = _mm_add_pd(fiz1,tz);
 995
 996             fjx0             = _mm_add_pd(fjx0,tx);
 997             fjy0             = _mm_add_pd(fjy0,ty);
 998             fjz0             = _mm_add_pd(fjz0,tz);
 999
1000             }
1001
1002             /**************************
1003              * CALCULATE INTERACTIONS *
1004              **************************/
1005
1006             if (gmx_mm_any_lt(rsq20,rcutoff2))
1007             {
1008
1009             r20              = _mm_mul_pd(rsq20,rinv20);
1010
1011             /* Compute parameters for interactions between i and j atoms */
1012             qq20             = _mm_mul_pd(iq2,jq0);
1013
1014             /* EWALD ELECTROSTATICS */
1015
1016             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1017             ewrt             = _mm_mul_pd(r20,ewtabscale);
1018             ewitab           = _mm_cvttpd_epi32(ewrt);
1019             eweps            = _mm_sub_pd(ewrt,_mm_cvtepi32_pd(ewitab));
1020             gmx_mm_load_2pair_swizzle_pd(ewtab+gmx_mm_extract_epi32(ewitab,0),ewtab+gmx_mm_extract_epi32(ewitab,1),
1021                                          &ewtabF,&ewtabFn);
1022             felec            = _mm_add_pd(_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF),_mm_mul_pd(eweps,ewtabFn));
1023             felec            = _mm_mul_pd(_mm_mul_pd(qq20,rinv20),_mm_sub_pd(rinvsq20,felec));
1024
1025             cutoff_mask      = _mm_cmplt_pd(rsq20,rcutoff2);
1026
1027             fscal            = felec;
1028
1029             fscal            = _mm_and_pd(fscal,cutoff_mask);
1030
1031             /* Calculate temporary vectorial force */
1032             tx               = _mm_mul_pd(fscal,dx20);
1033             ty               = _mm_mul_pd(fscal,dy20);
1034             tz               = _mm_mul_pd(fscal,dz20);
1035
1036             /* Update vectorial force */
1037             fix2             = _mm_add_pd(fix2,tx);
1038             fiy2             = _mm_add_pd(fiy2,ty);
1039             fiz2             = _mm_add_pd(fiz2,tz);
1040
1041             fjx0             = _mm_add_pd(fjx0,tx);
1042             fjy0             = _mm_add_pd(fjy0,ty);
1043             fjz0             = _mm_add_pd(fjz0,tz);
1044
1045             }
1046
1047             gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,fjx0,fjy0,fjz0);
1048
1049             /* Inner loop uses 143 flops */
1050         }
1051
1052         if(jidx<j_index_end)
1053         {
1054
1055             jnrA             = jjnr[jidx];
1056             j_coord_offsetA  = DIM*jnrA;
1057
1058             /* load j atom coordinates */
1059             gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
1060                                               &jx0,&jy0,&jz0);
1061
1062             /* Calculate displacement vector */
1063             dx00             = _mm_sub_pd(ix0,jx0);
1064             dy00             = _mm_sub_pd(iy0,jy0);
1065             dz00             = _mm_sub_pd(iz0,jz0);
1066             dx10             = _mm_sub_pd(ix1,jx0);
1067             dy10             = _mm_sub_pd(iy1,jy0);
1068             dz10             = _mm_sub_pd(iz1,jz0);
1069             dx20             = _mm_sub_pd(ix2,jx0);
1070             dy20             = _mm_sub_pd(iy2,jy0);
1071             dz20             = _mm_sub_pd(iz2,jz0);
1072
1073             /* Calculate squared distance and things based on it */
1074             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
1075             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
1076             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
1077
1078             rinv00           = gmx_mm_invsqrt_pd(rsq00);
1079             rinv10           = gmx_mm_invsqrt_pd(rsq10);
1080             rinv20           = gmx_mm_invsqrt_pd(rsq20);
1081
1082             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
1083             rinvsq10         = _mm_mul_pd(rinv10,rinv10);
1084             rinvsq20         = _mm_mul_pd(rinv20,rinv20);
1085
1086             /* Load parameters for j particles */
1087             jq0              = _mm_load_sd(charge+jnrA+0);
1088             vdwjidx0A        = 2*vdwtype[jnrA+0];
1089
1090             fjx0             = _mm_setzero_pd();
1091             fjy0             = _mm_setzero_pd();
1092             fjz0             = _mm_setzero_pd();
1093
1094             /**************************
1095              * CALCULATE INTERACTIONS *
1096              **************************/
1097
1098             if (gmx_mm_any_lt(rsq00,rcutoff2))
1099             {
1100
1101             r00              = _mm_mul_pd(rsq00,rinv00);
1102
1103             /* Compute parameters for interactions between i and j atoms */
1104             qq00             = _mm_mul_pd(iq0,jq0);
1105             gmx_mm_load_1pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,&c6_00,&c12_00);
1106
1107             c6grid_00       = gmx_mm_load_1real_pd(vdwgridparam+vdwioffset0+vdwjidx0A);
1108
1109             /* EWALD ELECTROSTATICS */
1110
1111             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1112             ewrt             = _mm_mul_pd(r00,ewtabscale);
1113             ewitab           = _mm_cvttpd_epi32(ewrt);
1114             eweps            = _mm_sub_pd(ewrt,_mm_cvtepi32_pd(ewitab));
1115             gmx_mm_load_1pair_swizzle_pd(ewtab+gmx_mm_extract_epi32(ewitab,0),&ewtabF,&ewtabFn);
1116             felec            = _mm_add_pd(_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF),_mm_mul_pd(eweps,ewtabFn));
1117             felec            = _mm_mul_pd(_mm_mul_pd(qq00,rinv00),_mm_sub_pd(rinvsq00,felec));
1118
1119             /* Analytical LJ-PME */
1120             rinvsix          = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
1121             ewcljrsq         = _mm_mul_pd(ewclj2,rsq00);
1122             ewclj6           = _mm_mul_pd(ewclj2,_mm_mul_pd(ewclj2,ewclj2));
1123             exponent         = gmx_simd_exp_d(ewcljrsq);
1124             /* poly = exp(-(beta*r)^2) * (1 + (beta*r)^2 + (beta*r)^4 /2) */
1125             poly             = _mm_mul_pd(exponent,_mm_add_pd(_mm_sub_pd(one,ewcljrsq),_mm_mul_pd(_mm_mul_pd(ewcljrsq,ewcljrsq),one_half)));
1126             /* f6A = 6 * C6grid * (1 - poly) */
1127             f6A              = _mm_mul_pd(c6grid_00,_mm_sub_pd(one,poly));
1128             /* f6B = C6grid * exponent * beta^6 */
1129             f6B              = _mm_mul_pd(_mm_mul_pd(c6grid_00,one_sixth),_mm_mul_pd(exponent,ewclj6));
1130             /* fvdw = 12*C12/r13 - ((6*C6 - f6A)/r6 + f6B)/r */
1131             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);
1132
1133             cutoff_mask      = _mm_cmplt_pd(rsq00,rcutoff2);
1134
1135             fscal            = _mm_add_pd(felec,fvdw);
1136
1137             fscal            = _mm_and_pd(fscal,cutoff_mask);
1138
1139             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1140
1141             /* Calculate temporary vectorial force */
1142             tx               = _mm_mul_pd(fscal,dx00);
1143             ty               = _mm_mul_pd(fscal,dy00);
1144             tz               = _mm_mul_pd(fscal,dz00);
1145
1146             /* Update vectorial force */
1147             fix0             = _mm_add_pd(fix0,tx);
1148             fiy0             = _mm_add_pd(fiy0,ty);
1149             fiz0             = _mm_add_pd(fiz0,tz);
1150
1151             fjx0             = _mm_add_pd(fjx0,tx);
1152             fjy0             = _mm_add_pd(fjy0,ty);
1153             fjz0             = _mm_add_pd(fjz0,tz);
1154
1155             }
1156
1157             /**************************
1158              * CALCULATE INTERACTIONS *
1159              **************************/
1160
1161             if (gmx_mm_any_lt(rsq10,rcutoff2))
1162             {
1163
1164             r10              = _mm_mul_pd(rsq10,rinv10);
1165
1166             /* Compute parameters for interactions between i and j atoms */
1167             qq10             = _mm_mul_pd(iq1,jq0);
1168
1169             /* EWALD ELECTROSTATICS */
1170
1171             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1172             ewrt             = _mm_mul_pd(r10,ewtabscale);
1173             ewitab           = _mm_cvttpd_epi32(ewrt);
1174             eweps            = _mm_sub_pd(ewrt,_mm_cvtepi32_pd(ewitab));
1175             gmx_mm_load_1pair_swizzle_pd(ewtab+gmx_mm_extract_epi32(ewitab,0),&ewtabF,&ewtabFn);
1176             felec            = _mm_add_pd(_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF),_mm_mul_pd(eweps,ewtabFn));
1177             felec            = _mm_mul_pd(_mm_mul_pd(qq10,rinv10),_mm_sub_pd(rinvsq10,felec));
1178
1179             cutoff_mask      = _mm_cmplt_pd(rsq10,rcutoff2);
1180
1181             fscal            = felec;
1182
1183             fscal            = _mm_and_pd(fscal,cutoff_mask);
1184
1185             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1186
1187             /* Calculate temporary vectorial force */
1188             tx               = _mm_mul_pd(fscal,dx10);
1189             ty               = _mm_mul_pd(fscal,dy10);
1190             tz               = _mm_mul_pd(fscal,dz10);
1191
1192             /* Update vectorial force */
1193             fix1             = _mm_add_pd(fix1,tx);
1194             fiy1             = _mm_add_pd(fiy1,ty);
1195             fiz1             = _mm_add_pd(fiz1,tz);
1196
1197             fjx0             = _mm_add_pd(fjx0,tx);
1198             fjy0             = _mm_add_pd(fjy0,ty);
1199             fjz0             = _mm_add_pd(fjz0,tz);
1200
1201             }
1202
1203             /**************************
1204              * CALCULATE INTERACTIONS *
1205              **************************/
1206
1207             if (gmx_mm_any_lt(rsq20,rcutoff2))
1208             {
1209
1210             r20              = _mm_mul_pd(rsq20,rinv20);
1211
1212             /* Compute parameters for interactions between i and j atoms */
1213             qq20             = _mm_mul_pd(iq2,jq0);
1214
1215             /* EWALD ELECTROSTATICS */
1216
1217             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1218             ewrt             = _mm_mul_pd(r20,ewtabscale);
1219             ewitab           = _mm_cvttpd_epi32(ewrt);
1220             eweps            = _mm_sub_pd(ewrt,_mm_cvtepi32_pd(ewitab));
1221             gmx_mm_load_1pair_swizzle_pd(ewtab+gmx_mm_extract_epi32(ewitab,0),&ewtabF,&ewtabFn);
1222             felec            = _mm_add_pd(_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF),_mm_mul_pd(eweps,ewtabFn));
1223             felec            = _mm_mul_pd(_mm_mul_pd(qq20,rinv20),_mm_sub_pd(rinvsq20,felec));
1224
1225             cutoff_mask      = _mm_cmplt_pd(rsq20,rcutoff2);
1226
1227             fscal            = felec;
1228
1229             fscal            = _mm_and_pd(fscal,cutoff_mask);
1230
1231             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1232
1233             /* Calculate temporary vectorial force */
1234             tx               = _mm_mul_pd(fscal,dx20);
1235             ty               = _mm_mul_pd(fscal,dy20);
1236             tz               = _mm_mul_pd(fscal,dz20);
1237
1238             /* Update vectorial force */
1239             fix2             = _mm_add_pd(fix2,tx);
1240             fiy2             = _mm_add_pd(fiy2,ty);
1241             fiz2             = _mm_add_pd(fiz2,tz);
1242
1243             fjx0             = _mm_add_pd(fjx0,tx);
1244             fjy0             = _mm_add_pd(fjy0,ty);
1245             fjz0             = _mm_add_pd(fjz0,tz);
1246
1247             }
1248
1249             gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,fjx0,fjy0,fjz0);
1250
1251             /* Inner loop uses 143 flops */
1252         }
1253
1254         /* End of innermost loop */
1255
1256         gmx_mm_update_iforce_3atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
1257                                               f+i_coord_offset,fshift+i_shift_offset);
1258
1259         /* Increment number of inner iterations */
1260         inneriter                  += j_index_end - j_index_start;
1261
1262         /* Outer loop uses 18 flops */
1263     }
1264
1265     /* Increment number of outer iterations */
1266     outeriter        += nri;
1267
1268     /* Update outer/inner flops */
1269
1270     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3_F,outeriter*18 + inneriter*143);
1271 }