src/gromacs/gmxlib/nonbonded/nb_kernel_sse2_double/nb_kernel_ElecEw_VdwLJEw_GeomW4P1_sse2_double.c

   1 /*
   2  * This file is part of the GROMACS molecular simulation package.
   3  *
   4  * Copyright (c) 2012,2013,2014, by the GROMACS development team, led by
   5  * Mark Abraham, David van der Spoel, Berk Hess, and Erik Lindahl,
   6  * and including many others, as listed in the AUTHORS file in the
   7  * top-level source directory and at http://www.gromacs.org.
   8  *
   9  * GROMACS is free software; you can redistribute it and/or
  10  * modify it under the terms of the GNU Lesser General Public License
  11  * as published by the Free Software Foundation; either version 2.1
  12  * of the License, or (at your option) any later version.
  13  *
  14  * GROMACS is distributed in the hope that it will be useful,
  15  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  16  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  17  * Lesser General Public License for more details.
  18  *
  19  * You should have received a copy of the GNU Lesser General Public
  20  * License along with GROMACS; if not, see
  21  * http://www.gnu.org/licenses, or write to the Free Software Foundation,
  22  * Inc., 51 Franklin Street, Fifth Floor, Boston, MA  02110-1301  USA.
  23  *
  24  * If you want to redistribute modifications to GROMACS, please
  25  * consider that scientific software is very special. Version
  26  * control is crucial - bugs must be traceable. We will be happy to
  27  * consider code for inclusion in the official distribution, but
  28  * derived work must not be called official GROMACS. Details are found
  29  * in the README & COPYING files - if they are missing, get the
  30  * official version at http://www.gromacs.org.
  31  *
  32  * To help us fund GROMACS development, we humbly ask that you cite
  33  * the research papers on the package. Check out http://www.gromacs.org.
  34  */
  35 /*
  36  * Note: this file was generated by the GROMACS sse2_double kernel generator.
  37  */
  38 #ifdef HAVE_CONFIG_H
  39 #include <config.h>
  40 #endif
  41
  42 #include <math.h>
  43
  44 #include "../nb_kernel.h"
  45 #include "types/simple.h"
  46 #include "gromacs/math/vec.h"
  47 #include "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_ElecEw_VdwLJEw_GeomW4P1_VF_sse2_double
  54  * Electrostatics interaction: Ewald
  55  * VdW interaction:            LJEwald
  56  * Geometry:                   Water4-Particle
  57  * Calculate force/pot:        PotentialAndForce
  58  */
  59 void
  60 nb_kernel_ElecEw_VdwLJEw_GeomW4P1_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              vdwioffset3;
  89     __m128d          ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
  90     int              vdwjidx0A,vdwjidx0B;
  91     __m128d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  92     __m128d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  93     __m128d          dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
  94     __m128d          dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
  95     __m128d          dx30,dy30,dz30,rsq30,rinv30,rinvsq30,r30,qq30,c6_30,c12_30;
  96     __m128d          velec,felec,velecsum,facel,crf,krf,krf2;
  97     real             *charge;
  98     int              nvdwtype;
  99     __m128d          rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 100     int              *vdwtype;
 101     real             *vdwparam;
 102     __m128d          one_sixth   = _mm_set1_pd(1.0/6.0);
 103     __m128d          one_twelfth = _mm_set1_pd(1.0/12.0);
 104     __m128d           c6grid_00;
 105     __m128d           c6grid_10;
 106     __m128d           c6grid_20;
 107     __m128d           c6grid_30;
 108     __m128d           ewclj,ewclj2,ewclj6,ewcljrsq,poly,exponent,f6A,f6B,sh_lj_ewald;
 109     real             *vdwgridparam;
 110     __m128d           one_half = _mm_set1_pd(0.5);
 111     __m128d           minus_one = _mm_set1_pd(-1.0);
 112     __m128i          ewitab;
 113     __m128d          ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
 114     real             *ewtab;
 115     __m128d          dummy_mask,cutoff_mask;
 116     __m128d          signbit   = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
 117     __m128d          one     = _mm_set1_pd(1.0);
 118     __m128d          two     = _mm_set1_pd(2.0);
 119     x                = xx[0];
 120     f                = ff[0];
 121
 122     nri              = nlist->nri;
 123     iinr             = nlist->iinr;
 124     jindex           = nlist->jindex;
 125     jjnr             = nlist->jjnr;
 126     shiftidx         = nlist->shift;
 127     gid              = nlist->gid;
 128     shiftvec         = fr->shift_vec[0];
 129     fshift           = fr->fshift[0];
 130     facel            = _mm_set1_pd(fr->epsfac);
 131     charge           = mdatoms->chargeA;
 132     nvdwtype         = fr->ntype;
 133     vdwparam         = fr->nbfp;
 134     vdwtype          = mdatoms->typeA;
 135     vdwgridparam     = fr->ljpme_c6grid;
 136     sh_lj_ewald      = _mm_set1_pd(fr->ic->sh_lj_ewald);
 137     ewclj            = _mm_set1_pd(fr->ewaldcoeff_lj);
 138     ewclj2           = _mm_mul_pd(minus_one,_mm_mul_pd(ewclj,ewclj));
 139
 140     sh_ewald         = _mm_set1_pd(fr->ic->sh_ewald);
 141     ewtab            = fr->ic->tabq_coul_FDV0;
 142     ewtabscale       = _mm_set1_pd(fr->ic->tabq_scale);
 143     ewtabhalfspace   = _mm_set1_pd(0.5/fr->ic->tabq_scale);
 144
 145     /* Setup water-specific parameters */
 146     inr              = nlist->iinr[0];
 147     iq1              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+1]));
 148     iq2              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+2]));
 149     iq3              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+3]));
 150     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 151
 152     /* Avoid stupid compiler warnings */
 153     jnrA = jnrB = 0;
 154     j_coord_offsetA = 0;
 155     j_coord_offsetB = 0;
 156
 157     outeriter        = 0;
 158     inneriter        = 0;
 159
 160     /* Start outer loop over neighborlists */
 161     for(iidx=0; iidx<nri; iidx++)
 162     {
 163         /* Load shift vector for this list */
 164         i_shift_offset   = DIM*shiftidx[iidx];
 165
 166         /* Load limits for loop over neighbors */
 167         j_index_start    = jindex[iidx];
 168         j_index_end      = jindex[iidx+1];
 169
 170         /* Get outer coordinate index */
 171         inr              = iinr[iidx];
 172         i_coord_offset   = DIM*inr;
 173
 174         /* Load i particle coords and add shift vector */
 175         gmx_mm_load_shift_and_4rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
 176                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
 177
 178         fix0             = _mm_setzero_pd();
 179         fiy0             = _mm_setzero_pd();
 180         fiz0             = _mm_setzero_pd();
 181         fix1             = _mm_setzero_pd();
 182         fiy1             = _mm_setzero_pd();
 183         fiz1             = _mm_setzero_pd();
 184         fix2             = _mm_setzero_pd();
 185         fiy2             = _mm_setzero_pd();
 186         fiz2             = _mm_setzero_pd();
 187         fix3             = _mm_setzero_pd();
 188         fiy3             = _mm_setzero_pd();
 189         fiz3             = _mm_setzero_pd();
 190
 191         /* Reset potential sums */
 192         velecsum         = _mm_setzero_pd();
 193         vvdwsum          = _mm_setzero_pd();
 194
 195         /* Start inner kernel loop */
 196         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 197         {
 198
 199             /* Get j neighbor index, and coordinate index */
 200             jnrA             = jjnr[jidx];
 201             jnrB             = jjnr[jidx+1];
 202             j_coord_offsetA  = DIM*jnrA;
 203             j_coord_offsetB  = DIM*jnrB;
 204
 205             /* load j atom coordinates */
 206             gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 207                                               &jx0,&jy0,&jz0);
 208
 209             /* Calculate displacement vector */
 210             dx00             = _mm_sub_pd(ix0,jx0);
 211             dy00             = _mm_sub_pd(iy0,jy0);
 212             dz00             = _mm_sub_pd(iz0,jz0);
 213             dx10             = _mm_sub_pd(ix1,jx0);
 214             dy10             = _mm_sub_pd(iy1,jy0);
 215             dz10             = _mm_sub_pd(iz1,jz0);
 216             dx20             = _mm_sub_pd(ix2,jx0);
 217             dy20             = _mm_sub_pd(iy2,jy0);
 218             dz20             = _mm_sub_pd(iz2,jz0);
 219             dx30             = _mm_sub_pd(ix3,jx0);
 220             dy30             = _mm_sub_pd(iy3,jy0);
 221             dz30             = _mm_sub_pd(iz3,jz0);
 222
 223             /* Calculate squared distance and things based on it */
 224             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 225             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
 226             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
 227             rsq30            = gmx_mm_calc_rsq_pd(dx30,dy30,dz30);
 228
 229             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 230             rinv10           = gmx_mm_invsqrt_pd(rsq10);
 231             rinv20           = gmx_mm_invsqrt_pd(rsq20);
 232             rinv30           = gmx_mm_invsqrt_pd(rsq30);
 233
 234             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 235             rinvsq10         = _mm_mul_pd(rinv10,rinv10);
 236             rinvsq20         = _mm_mul_pd(rinv20,rinv20);
 237             rinvsq30         = _mm_mul_pd(rinv30,rinv30);
 238
 239             /* Load parameters for j particles */
 240             jq0              = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
 241             vdwjidx0A        = 2*vdwtype[jnrA+0];
 242             vdwjidx0B        = 2*vdwtype[jnrB+0];
 243
 244             fjx0             = _mm_setzero_pd();
 245             fjy0             = _mm_setzero_pd();
 246             fjz0             = _mm_setzero_pd();
 247
 248             /**************************
 249              * CALCULATE INTERACTIONS *
 250              **************************/
 251
 252             r00              = _mm_mul_pd(rsq00,rinv00);
 253
 254             /* Compute parameters for interactions between i and j atoms */
 255             gmx_mm_load_2pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,
 256                                          vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
 257
 258             c6grid_00       = gmx_mm_load_2real_swizzle_pd(vdwgridparam+vdwioffset0+vdwjidx0A,
 259                                                                vdwgridparam+vdwioffset0+vdwjidx0B);
 260
 261             /* Analytical LJ-PME */
 262             rinvsix          = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 263             ewcljrsq         = _mm_mul_pd(ewclj2,rsq00);
 264             ewclj6           = _mm_mul_pd(ewclj2,_mm_mul_pd(ewclj2,ewclj2));
 265             exponent         = gmx_simd_exp_d(ewcljrsq);
 266             /* poly = exp(-(beta*r)^2) * (1 + (beta*r)^2 + (beta*r)^4 /2) */
 267             poly             = _mm_mul_pd(exponent,_mm_add_pd(_mm_sub_pd(one,ewcljrsq),_mm_mul_pd(_mm_mul_pd(ewcljrsq,ewcljrsq),one_half)));
 268             /* vvdw6 = [C6 - C6grid * (1-poly)]/r6 */
 269             vvdw6            = _mm_mul_pd(_mm_sub_pd(c6_00,_mm_mul_pd(c6grid_00,_mm_sub_pd(one,poly))),rinvsix);
 270             vvdw12           = _mm_mul_pd(c12_00,_mm_mul_pd(rinvsix,rinvsix));
 271             vvdw             = _mm_sub_pd(_mm_mul_pd(vvdw12,one_twelfth),_mm_mul_pd(vvdw6,one_sixth));
 272             /* fvdw = vvdw12/r - (vvdw6/r + (C6grid * exponent * beta^6)/r) */
 273             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);
 274
 275             /* Update potential sum for this i atom from the interaction with this j atom. */
 276             vvdwsum          = _mm_add_pd(vvdwsum,vvdw);
 277
 278             fscal            = fvdw;
 279
 280             /* Calculate temporary vectorial force */
 281             tx               = _mm_mul_pd(fscal,dx00);
 282             ty               = _mm_mul_pd(fscal,dy00);
 283             tz               = _mm_mul_pd(fscal,dz00);
 284
 285             /* Update vectorial force */
 286             fix0             = _mm_add_pd(fix0,tx);
 287             fiy0             = _mm_add_pd(fiy0,ty);
 288             fiz0             = _mm_add_pd(fiz0,tz);
 289
 290             fjx0             = _mm_add_pd(fjx0,tx);
 291             fjy0             = _mm_add_pd(fjy0,ty);
 292             fjz0             = _mm_add_pd(fjz0,tz);
 293
 294             /**************************
 295              * CALCULATE INTERACTIONS *
 296              **************************/
 297
 298             r10              = _mm_mul_pd(rsq10,rinv10);
 299
 300             /* Compute parameters for interactions between i and j atoms */
 301             qq10             = _mm_mul_pd(iq1,jq0);
 302
 303             /* EWALD ELECTROSTATICS */
 304
 305             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 306             ewrt             = _mm_mul_pd(r10,ewtabscale);
 307             ewitab           = _mm_cvttpd_epi32(ewrt);
 308             eweps            = _mm_sub_pd(ewrt,_mm_cvtepi32_pd(ewitab));
 309             ewitab           = _mm_slli_epi32(ewitab,2);
 310             ewtabF           = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 311             ewtabD           = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,1) );
 312             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 313             ewtabV           = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,0) +2);
 314             ewtabFn          = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,1) +2);
 315             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 316             felec            = _mm_add_pd(ewtabF,_mm_mul_pd(eweps,ewtabD));
 317             velec            = _mm_sub_pd(ewtabV,_mm_mul_pd(_mm_mul_pd(ewtabhalfspace,eweps),_mm_add_pd(ewtabF,felec)));
 318             velec            = _mm_mul_pd(qq10,_mm_sub_pd(rinv10,velec));
 319             felec            = _mm_mul_pd(_mm_mul_pd(qq10,rinv10),_mm_sub_pd(rinvsq10,felec));
 320
 321             /* Update potential sum for this i atom from the interaction with this j atom. */
 322             velecsum         = _mm_add_pd(velecsum,velec);
 323
 324             fscal            = felec;
 325
 326             /* Calculate temporary vectorial force */
 327             tx               = _mm_mul_pd(fscal,dx10);
 328             ty               = _mm_mul_pd(fscal,dy10);
 329             tz               = _mm_mul_pd(fscal,dz10);
 330
 331             /* Update vectorial force */
 332             fix1             = _mm_add_pd(fix1,tx);
 333             fiy1             = _mm_add_pd(fiy1,ty);
 334             fiz1             = _mm_add_pd(fiz1,tz);
 335
 336             fjx0             = _mm_add_pd(fjx0,tx);
 337             fjy0             = _mm_add_pd(fjy0,ty);
 338             fjz0             = _mm_add_pd(fjz0,tz);
 339
 340             /**************************
 341              * CALCULATE INTERACTIONS *
 342              **************************/
 343
 344             r20              = _mm_mul_pd(rsq20,rinv20);
 345
 346             /* Compute parameters for interactions between i and j atoms */
 347             qq20             = _mm_mul_pd(iq2,jq0);
 348
 349             /* EWALD ELECTROSTATICS */
 350
 351             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 352             ewrt             = _mm_mul_pd(r20,ewtabscale);
 353             ewitab           = _mm_cvttpd_epi32(ewrt);
 354             eweps            = _mm_sub_pd(ewrt,_mm_cvtepi32_pd(ewitab));
 355             ewitab           = _mm_slli_epi32(ewitab,2);
 356             ewtabF           = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 357             ewtabD           = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,1) );
 358             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 359             ewtabV           = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,0) +2);
 360             ewtabFn          = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,1) +2);
 361             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 362             felec            = _mm_add_pd(ewtabF,_mm_mul_pd(eweps,ewtabD));
 363             velec            = _mm_sub_pd(ewtabV,_mm_mul_pd(_mm_mul_pd(ewtabhalfspace,eweps),_mm_add_pd(ewtabF,felec)));
 364             velec            = _mm_mul_pd(qq20,_mm_sub_pd(rinv20,velec));
 365             felec            = _mm_mul_pd(_mm_mul_pd(qq20,rinv20),_mm_sub_pd(rinvsq20,felec));
 366
 367             /* Update potential sum for this i atom from the interaction with this j atom. */
 368             velecsum         = _mm_add_pd(velecsum,velec);
 369
 370             fscal            = felec;
 371
 372             /* Calculate temporary vectorial force */
 373             tx               = _mm_mul_pd(fscal,dx20);
 374             ty               = _mm_mul_pd(fscal,dy20);
 375             tz               = _mm_mul_pd(fscal,dz20);
 376
 377             /* Update vectorial force */
 378             fix2             = _mm_add_pd(fix2,tx);
 379             fiy2             = _mm_add_pd(fiy2,ty);
 380             fiz2             = _mm_add_pd(fiz2,tz);
 381
 382             fjx0             = _mm_add_pd(fjx0,tx);
 383             fjy0             = _mm_add_pd(fjy0,ty);
 384             fjz0             = _mm_add_pd(fjz0,tz);
 385
 386             /**************************
 387              * CALCULATE INTERACTIONS *
 388              **************************/
 389
 390             r30              = _mm_mul_pd(rsq30,rinv30);
 391
 392             /* Compute parameters for interactions between i and j atoms */
 393             qq30             = _mm_mul_pd(iq3,jq0);
 394
 395             /* EWALD ELECTROSTATICS */
 396
 397             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 398             ewrt             = _mm_mul_pd(r30,ewtabscale);
 399             ewitab           = _mm_cvttpd_epi32(ewrt);
 400             eweps            = _mm_sub_pd(ewrt,_mm_cvtepi32_pd(ewitab));
 401             ewitab           = _mm_slli_epi32(ewitab,2);
 402             ewtabF           = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 403             ewtabD           = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,1) );
 404             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 405             ewtabV           = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,0) +2);
 406             ewtabFn          = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,1) +2);
 407             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 408             felec            = _mm_add_pd(ewtabF,_mm_mul_pd(eweps,ewtabD));
 409             velec            = _mm_sub_pd(ewtabV,_mm_mul_pd(_mm_mul_pd(ewtabhalfspace,eweps),_mm_add_pd(ewtabF,felec)));
 410             velec            = _mm_mul_pd(qq30,_mm_sub_pd(rinv30,velec));
 411             felec            = _mm_mul_pd(_mm_mul_pd(qq30,rinv30),_mm_sub_pd(rinvsq30,felec));
 412
 413             /* Update potential sum for this i atom from the interaction with this j atom. */
 414             velecsum         = _mm_add_pd(velecsum,velec);
 415
 416             fscal            = felec;
 417
 418             /* Calculate temporary vectorial force */
 419             tx               = _mm_mul_pd(fscal,dx30);
 420             ty               = _mm_mul_pd(fscal,dy30);
 421             tz               = _mm_mul_pd(fscal,dz30);
 422
 423             /* Update vectorial force */
 424             fix3             = _mm_add_pd(fix3,tx);
 425             fiy3             = _mm_add_pd(fiy3,ty);
 426             fiz3             = _mm_add_pd(fiz3,tz);
 427
 428             fjx0             = _mm_add_pd(fjx0,tx);
 429             fjy0             = _mm_add_pd(fjy0,ty);
 430             fjz0             = _mm_add_pd(fjz0,tz);
 431
 432             gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,fjx0,fjy0,fjz0);
 433
 434             /* Inner loop uses 177 flops */
 435         }
 436
 437         if(jidx<j_index_end)
 438         {
 439
 440             jnrA             = jjnr[jidx];
 441             j_coord_offsetA  = DIM*jnrA;
 442
 443             /* load j atom coordinates */
 444             gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
 445                                               &jx0,&jy0,&jz0);
 446
 447             /* Calculate displacement vector */
 448             dx00             = _mm_sub_pd(ix0,jx0);
 449             dy00             = _mm_sub_pd(iy0,jy0);
 450             dz00             = _mm_sub_pd(iz0,jz0);
 451             dx10             = _mm_sub_pd(ix1,jx0);
 452             dy10             = _mm_sub_pd(iy1,jy0);
 453             dz10             = _mm_sub_pd(iz1,jz0);
 454             dx20             = _mm_sub_pd(ix2,jx0);
 455             dy20             = _mm_sub_pd(iy2,jy0);
 456             dz20             = _mm_sub_pd(iz2,jz0);
 457             dx30             = _mm_sub_pd(ix3,jx0);
 458             dy30             = _mm_sub_pd(iy3,jy0);
 459             dz30             = _mm_sub_pd(iz3,jz0);
 460
 461             /* Calculate squared distance and things based on it */
 462             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 463             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
 464             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
 465             rsq30            = gmx_mm_calc_rsq_pd(dx30,dy30,dz30);
 466
 467             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 468             rinv10           = gmx_mm_invsqrt_pd(rsq10);
 469             rinv20           = gmx_mm_invsqrt_pd(rsq20);
 470             rinv30           = gmx_mm_invsqrt_pd(rsq30);
 471
 472             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 473             rinvsq10         = _mm_mul_pd(rinv10,rinv10);
 474             rinvsq20         = _mm_mul_pd(rinv20,rinv20);
 475             rinvsq30         = _mm_mul_pd(rinv30,rinv30);
 476
 477             /* Load parameters for j particles */
 478             jq0              = _mm_load_sd(charge+jnrA+0);
 479             vdwjidx0A        = 2*vdwtype[jnrA+0];
 480
 481             fjx0             = _mm_setzero_pd();
 482             fjy0             = _mm_setzero_pd();
 483             fjz0             = _mm_setzero_pd();
 484
 485             /**************************
 486              * CALCULATE INTERACTIONS *
 487              **************************/
 488
 489             r00              = _mm_mul_pd(rsq00,rinv00);
 490
 491             /* Compute parameters for interactions between i and j atoms */
 492             gmx_mm_load_1pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,&c6_00,&c12_00);
 493
 494             c6grid_00       = gmx_mm_load_1real_pd(vdwgridparam+vdwioffset0+vdwjidx0A);
 495
 496             /* Analytical LJ-PME */
 497             rinvsix          = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 498             ewcljrsq         = _mm_mul_pd(ewclj2,rsq00);
 499             ewclj6           = _mm_mul_pd(ewclj2,_mm_mul_pd(ewclj2,ewclj2));
 500             exponent         = gmx_simd_exp_d(ewcljrsq);
 501             /* poly = exp(-(beta*r)^2) * (1 + (beta*r)^2 + (beta*r)^4 /2) */
 502             poly             = _mm_mul_pd(exponent,_mm_add_pd(_mm_sub_pd(one,ewcljrsq),_mm_mul_pd(_mm_mul_pd(ewcljrsq,ewcljrsq),one_half)));
 503             /* vvdw6 = [C6 - C6grid * (1-poly)]/r6 */
 504             vvdw6            = _mm_mul_pd(_mm_sub_pd(c6_00,_mm_mul_pd(c6grid_00,_mm_sub_pd(one,poly))),rinvsix);
 505             vvdw12           = _mm_mul_pd(c12_00,_mm_mul_pd(rinvsix,rinvsix));
 506             vvdw             = _mm_sub_pd(_mm_mul_pd(vvdw12,one_twelfth),_mm_mul_pd(vvdw6,one_sixth));
 507             /* fvdw = vvdw12/r - (vvdw6/r + (C6grid * exponent * beta^6)/r) */
 508             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);
 509
 510             /* Update potential sum for this i atom from the interaction with this j atom. */
 511             vvdw             = _mm_unpacklo_pd(vvdw,_mm_setzero_pd());
 512             vvdwsum          = _mm_add_pd(vvdwsum,vvdw);
 513
 514             fscal            = fvdw;
 515
 516             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 517
 518             /* Calculate temporary vectorial force */
 519             tx               = _mm_mul_pd(fscal,dx00);
 520             ty               = _mm_mul_pd(fscal,dy00);
 521             tz               = _mm_mul_pd(fscal,dz00);
 522
 523             /* Update vectorial force */
 524             fix0             = _mm_add_pd(fix0,tx);
 525             fiy0             = _mm_add_pd(fiy0,ty);
 526             fiz0             = _mm_add_pd(fiz0,tz);
 527
 528             fjx0             = _mm_add_pd(fjx0,tx);
 529             fjy0             = _mm_add_pd(fjy0,ty);
 530             fjz0             = _mm_add_pd(fjz0,tz);
 531
 532             /**************************
 533              * CALCULATE INTERACTIONS *
 534              **************************/
 535
 536             r10              = _mm_mul_pd(rsq10,rinv10);
 537
 538             /* Compute parameters for interactions between i and j atoms */
 539             qq10             = _mm_mul_pd(iq1,jq0);
 540
 541             /* EWALD ELECTROSTATICS */
 542
 543             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 544             ewrt             = _mm_mul_pd(r10,ewtabscale);
 545             ewitab           = _mm_cvttpd_epi32(ewrt);
 546             eweps            = _mm_sub_pd(ewrt,_mm_cvtepi32_pd(ewitab));
 547             ewitab           = _mm_slli_epi32(ewitab,2);
 548             ewtabF           = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 549             ewtabD           = _mm_setzero_pd();
 550             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 551             ewtabV           = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,0) +2);
 552             ewtabFn          = _mm_setzero_pd();
 553             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 554             felec            = _mm_add_pd(ewtabF,_mm_mul_pd(eweps,ewtabD));
 555             velec            = _mm_sub_pd(ewtabV,_mm_mul_pd(_mm_mul_pd(ewtabhalfspace,eweps),_mm_add_pd(ewtabF,felec)));
 556             velec            = _mm_mul_pd(qq10,_mm_sub_pd(rinv10,velec));
 557             felec            = _mm_mul_pd(_mm_mul_pd(qq10,rinv10),_mm_sub_pd(rinvsq10,felec));
 558
 559             /* Update potential sum for this i atom from the interaction with this j atom. */
 560             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 561             velecsum         = _mm_add_pd(velecsum,velec);
 562
 563             fscal            = felec;
 564
 565             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 566
 567             /* Calculate temporary vectorial force */
 568             tx               = _mm_mul_pd(fscal,dx10);
 569             ty               = _mm_mul_pd(fscal,dy10);
 570             tz               = _mm_mul_pd(fscal,dz10);
 571
 572             /* Update vectorial force */
 573             fix1             = _mm_add_pd(fix1,tx);
 574             fiy1             = _mm_add_pd(fiy1,ty);
 575             fiz1             = _mm_add_pd(fiz1,tz);
 576
 577             fjx0             = _mm_add_pd(fjx0,tx);
 578             fjy0             = _mm_add_pd(fjy0,ty);
 579             fjz0             = _mm_add_pd(fjz0,tz);
 580
 581             /**************************
 582              * CALCULATE INTERACTIONS *
 583              **************************/
 584
 585             r20              = _mm_mul_pd(rsq20,rinv20);
 586
 587             /* Compute parameters for interactions between i and j atoms */
 588             qq20             = _mm_mul_pd(iq2,jq0);
 589
 590             /* EWALD ELECTROSTATICS */
 591
 592             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 593             ewrt             = _mm_mul_pd(r20,ewtabscale);
 594             ewitab           = _mm_cvttpd_epi32(ewrt);
 595             eweps            = _mm_sub_pd(ewrt,_mm_cvtepi32_pd(ewitab));
 596             ewitab           = _mm_slli_epi32(ewitab,2);
 597             ewtabF           = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 598             ewtabD           = _mm_setzero_pd();
 599             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 600             ewtabV           = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,0) +2);
 601             ewtabFn          = _mm_setzero_pd();
 602             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 603             felec            = _mm_add_pd(ewtabF,_mm_mul_pd(eweps,ewtabD));
 604             velec            = _mm_sub_pd(ewtabV,_mm_mul_pd(_mm_mul_pd(ewtabhalfspace,eweps),_mm_add_pd(ewtabF,felec)));
 605             velec            = _mm_mul_pd(qq20,_mm_sub_pd(rinv20,velec));
 606             felec            = _mm_mul_pd(_mm_mul_pd(qq20,rinv20),_mm_sub_pd(rinvsq20,felec));
 607
 608             /* Update potential sum for this i atom from the interaction with this j atom. */
 609             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 610             velecsum         = _mm_add_pd(velecsum,velec);
 611
 612             fscal            = felec;
 613
 614             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 615
 616             /* Calculate temporary vectorial force */
 617             tx               = _mm_mul_pd(fscal,dx20);
 618             ty               = _mm_mul_pd(fscal,dy20);
 619             tz               = _mm_mul_pd(fscal,dz20);
 620
 621             /* Update vectorial force */
 622             fix2             = _mm_add_pd(fix2,tx);
 623             fiy2             = _mm_add_pd(fiy2,ty);
 624             fiz2             = _mm_add_pd(fiz2,tz);
 625
 626             fjx0             = _mm_add_pd(fjx0,tx);
 627             fjy0             = _mm_add_pd(fjy0,ty);
 628             fjz0             = _mm_add_pd(fjz0,tz);
 629
 630             /**************************
 631              * CALCULATE INTERACTIONS *
 632              **************************/
 633
 634             r30              = _mm_mul_pd(rsq30,rinv30);
 635
 636             /* Compute parameters for interactions between i and j atoms */
 637             qq30             = _mm_mul_pd(iq3,jq0);
 638
 639             /* EWALD ELECTROSTATICS */
 640
 641             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 642             ewrt             = _mm_mul_pd(r30,ewtabscale);
 643             ewitab           = _mm_cvttpd_epi32(ewrt);
 644             eweps            = _mm_sub_pd(ewrt,_mm_cvtepi32_pd(ewitab));
 645             ewitab           = _mm_slli_epi32(ewitab,2);
 646             ewtabF           = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 647             ewtabD           = _mm_setzero_pd();
 648             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 649             ewtabV           = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,0) +2);
 650             ewtabFn          = _mm_setzero_pd();
 651             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 652             felec            = _mm_add_pd(ewtabF,_mm_mul_pd(eweps,ewtabD));
 653             velec            = _mm_sub_pd(ewtabV,_mm_mul_pd(_mm_mul_pd(ewtabhalfspace,eweps),_mm_add_pd(ewtabF,felec)));
 654             velec            = _mm_mul_pd(qq30,_mm_sub_pd(rinv30,velec));
 655             felec            = _mm_mul_pd(_mm_mul_pd(qq30,rinv30),_mm_sub_pd(rinvsq30,felec));
 656
 657             /* Update potential sum for this i atom from the interaction with this j atom. */
 658             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 659             velecsum         = _mm_add_pd(velecsum,velec);
 660
 661             fscal            = felec;
 662
 663             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 664
 665             /* Calculate temporary vectorial force */
 666             tx               = _mm_mul_pd(fscal,dx30);
 667             ty               = _mm_mul_pd(fscal,dy30);
 668             tz               = _mm_mul_pd(fscal,dz30);
 669
 670             /* Update vectorial force */
 671             fix3             = _mm_add_pd(fix3,tx);
 672             fiy3             = _mm_add_pd(fiy3,ty);
 673             fiz3             = _mm_add_pd(fiz3,tz);
 674
 675             fjx0             = _mm_add_pd(fjx0,tx);
 676             fjy0             = _mm_add_pd(fjy0,ty);
 677             fjz0             = _mm_add_pd(fjz0,tz);
 678
 679             gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,fjx0,fjy0,fjz0);
 680
 681             /* Inner loop uses 177 flops */
 682         }
 683
 684         /* End of innermost loop */
 685
 686         gmx_mm_update_iforce_4atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
 687                                               f+i_coord_offset,fshift+i_shift_offset);
 688
 689         ggid                        = gid[iidx];
 690         /* Update potential energies */
 691         gmx_mm_update_1pot_pd(velecsum,kernel_data->energygrp_elec+ggid);
 692         gmx_mm_update_1pot_pd(vvdwsum,kernel_data->energygrp_vdw+ggid);
 693
 694         /* Increment number of inner iterations */
 695         inneriter                  += j_index_end - j_index_start;
 696
 697         /* Outer loop uses 26 flops */
 698     }
 699
 700     /* Increment number of outer iterations */
 701     outeriter        += nri;
 702
 703     /* Update outer/inner flops */
 704
 705     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W4_VF,outeriter*26 + inneriter*177);
 706 }
 707 /*
 708  * Gromacs nonbonded kernel:   nb_kernel_ElecEw_VdwLJEw_GeomW4P1_F_sse2_double
 709  * Electrostatics interaction: Ewald
 710  * VdW interaction:            LJEwald
 711  * Geometry:                   Water4-Particle
 712  * Calculate force/pot:        Force
 713  */
 714 void
 715 nb_kernel_ElecEw_VdwLJEw_GeomW4P1_F_sse2_double
 716                     (t_nblist                    * gmx_restrict       nlist,
 717                      rvec                        * gmx_restrict          xx,
 718                      rvec                        * gmx_restrict          ff,
 719                      t_forcerec                  * gmx_restrict          fr,
 720                      t_mdatoms                   * gmx_restrict     mdatoms,
 721                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 722                      t_nrnb                      * gmx_restrict        nrnb)
 723 {
 724     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 725      * just 0 for non-waters.
 726      * Suffixes A,B refer to j loop unrolling done with SSE double precision, e.g. for the two different
 727      * jnr indices corresponding to data put in the four positions in the SIMD register.
 728      */
 729     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 730     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 731     int              jnrA,jnrB;
 732     int              j_coord_offsetA,j_coord_offsetB;
 733     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 734     real             rcutoff_scalar;
 735     real             *shiftvec,*fshift,*x,*f;
 736     __m128d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 737     int              vdwioffset0;
 738     __m128d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 739     int              vdwioffset1;
 740     __m128d          ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
 741     int              vdwioffset2;
 742     __m128d          ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
 743     int              vdwioffset3;
 744     __m128d          ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
 745     int              vdwjidx0A,vdwjidx0B;
 746     __m128d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 747     __m128d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 748     __m128d          dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
 749     __m128d          dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
 750     __m128d          dx30,dy30,dz30,rsq30,rinv30,rinvsq30,r30,qq30,c6_30,c12_30;
 751     __m128d          velec,felec,velecsum,facel,crf,krf,krf2;
 752     real             *charge;
 753     int              nvdwtype;
 754     __m128d          rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 755     int              *vdwtype;
 756     real             *vdwparam;
 757     __m128d          one_sixth   = _mm_set1_pd(1.0/6.0);
 758     __m128d          one_twelfth = _mm_set1_pd(1.0/12.0);
 759     __m128d           c6grid_00;
 760     __m128d           c6grid_10;
 761     __m128d           c6grid_20;
 762     __m128d           c6grid_30;
 763     __m128d           ewclj,ewclj2,ewclj6,ewcljrsq,poly,exponent,f6A,f6B,sh_lj_ewald;
 764     real             *vdwgridparam;
 765     __m128d           one_half = _mm_set1_pd(0.5);
 766     __m128d           minus_one = _mm_set1_pd(-1.0);
 767     __m128i          ewitab;
 768     __m128d          ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
 769     real             *ewtab;
 770     __m128d          dummy_mask,cutoff_mask;
 771     __m128d          signbit   = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
 772     __m128d          one     = _mm_set1_pd(1.0);
 773     __m128d          two     = _mm_set1_pd(2.0);
 774     x                = xx[0];
 775     f                = ff[0];
 776
 777     nri              = nlist->nri;
 778     iinr             = nlist->iinr;
 779     jindex           = nlist->jindex;
 780     jjnr             = nlist->jjnr;
 781     shiftidx         = nlist->shift;
 782     gid              = nlist->gid;
 783     shiftvec         = fr->shift_vec[0];
 784     fshift           = fr->fshift[0];
 785     facel            = _mm_set1_pd(fr->epsfac);
 786     charge           = mdatoms->chargeA;
 787     nvdwtype         = fr->ntype;
 788     vdwparam         = fr->nbfp;
 789     vdwtype          = mdatoms->typeA;
 790     vdwgridparam     = fr->ljpme_c6grid;
 791     sh_lj_ewald      = _mm_set1_pd(fr->ic->sh_lj_ewald);
 792     ewclj            = _mm_set1_pd(fr->ewaldcoeff_lj);
 793     ewclj2           = _mm_mul_pd(minus_one,_mm_mul_pd(ewclj,ewclj));
 794
 795     sh_ewald         = _mm_set1_pd(fr->ic->sh_ewald);
 796     ewtab            = fr->ic->tabq_coul_F;
 797     ewtabscale       = _mm_set1_pd(fr->ic->tabq_scale);
 798     ewtabhalfspace   = _mm_set1_pd(0.5/fr->ic->tabq_scale);
 799
 800     /* Setup water-specific parameters */
 801     inr              = nlist->iinr[0];
 802     iq1              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+1]));
 803     iq2              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+2]));
 804     iq3              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+3]));
 805     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 806
 807     /* Avoid stupid compiler warnings */
 808     jnrA = jnrB = 0;
 809     j_coord_offsetA = 0;
 810     j_coord_offsetB = 0;
 811
 812     outeriter        = 0;
 813     inneriter        = 0;
 814
 815     /* Start outer loop over neighborlists */
 816     for(iidx=0; iidx<nri; iidx++)
 817     {
 818         /* Load shift vector for this list */
 819         i_shift_offset   = DIM*shiftidx[iidx];
 820
 821         /* Load limits for loop over neighbors */
 822         j_index_start    = jindex[iidx];
 823         j_index_end      = jindex[iidx+1];
 824
 825         /* Get outer coordinate index */
 826         inr              = iinr[iidx];
 827         i_coord_offset   = DIM*inr;
 828
 829         /* Load i particle coords and add shift vector */
 830         gmx_mm_load_shift_and_4rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
 831                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
 832
 833         fix0             = _mm_setzero_pd();
 834         fiy0             = _mm_setzero_pd();
 835         fiz0             = _mm_setzero_pd();
 836         fix1             = _mm_setzero_pd();
 837         fiy1             = _mm_setzero_pd();
 838         fiz1             = _mm_setzero_pd();
 839         fix2             = _mm_setzero_pd();
 840         fiy2             = _mm_setzero_pd();
 841         fiz2             = _mm_setzero_pd();
 842         fix3             = _mm_setzero_pd();
 843         fiy3             = _mm_setzero_pd();
 844         fiz3             = _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             dx30             = _mm_sub_pd(ix3,jx0);
 871             dy30             = _mm_sub_pd(iy3,jy0);
 872             dz30             = _mm_sub_pd(iz3,jz0);
 873
 874             /* Calculate squared distance and things based on it */
 875             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 876             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
 877             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
 878             rsq30            = gmx_mm_calc_rsq_pd(dx30,dy30,dz30);
 879
 880             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 881             rinv10           = gmx_mm_invsqrt_pd(rsq10);
 882             rinv20           = gmx_mm_invsqrt_pd(rsq20);
 883             rinv30           = gmx_mm_invsqrt_pd(rsq30);
 884
 885             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 886             rinvsq10         = _mm_mul_pd(rinv10,rinv10);
 887             rinvsq20         = _mm_mul_pd(rinv20,rinv20);
 888             rinvsq30         = _mm_mul_pd(rinv30,rinv30);
 889
 890             /* Load parameters for j particles */
 891             jq0              = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
 892             vdwjidx0A        = 2*vdwtype[jnrA+0];
 893             vdwjidx0B        = 2*vdwtype[jnrB+0];
 894
 895             fjx0             = _mm_setzero_pd();
 896             fjy0             = _mm_setzero_pd();
 897             fjz0             = _mm_setzero_pd();
 898
 899             /**************************
 900              * CALCULATE INTERACTIONS *
 901              **************************/
 902
 903             r00              = _mm_mul_pd(rsq00,rinv00);
 904
 905             /* Compute parameters for interactions between i and j atoms */
 906             gmx_mm_load_2pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,
 907                                          vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
 908
 909             c6grid_00       = gmx_mm_load_2real_swizzle_pd(vdwgridparam+vdwioffset0+vdwjidx0A,
 910                                                                vdwgridparam+vdwioffset0+vdwjidx0B);
 911
 912             /* Analytical LJ-PME */
 913             rinvsix          = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 914             ewcljrsq         = _mm_mul_pd(ewclj2,rsq00);
 915             ewclj6           = _mm_mul_pd(ewclj2,_mm_mul_pd(ewclj2,ewclj2));
 916             exponent         = gmx_simd_exp_d(ewcljrsq);
 917             /* poly = exp(-(beta*r)^2) * (1 + (beta*r)^2 + (beta*r)^4 /2) */
 918             poly             = _mm_mul_pd(exponent,_mm_add_pd(_mm_sub_pd(one,ewcljrsq),_mm_mul_pd(_mm_mul_pd(ewcljrsq,ewcljrsq),one_half)));
 919             /* f6A = 6 * C6grid * (1 - poly) */
 920             f6A              = _mm_mul_pd(c6grid_00,_mm_sub_pd(one,poly));
 921             /* f6B = C6grid * exponent * beta^6 */
 922             f6B              = _mm_mul_pd(_mm_mul_pd(c6grid_00,one_sixth),_mm_mul_pd(exponent,ewclj6));
 923             /* fvdw = 12*C12/r13 - ((6*C6 - f6A)/r6 + f6B)/r */
 924             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);
 925
 926             fscal            = fvdw;
 927
 928             /* Calculate temporary vectorial force */
 929             tx               = _mm_mul_pd(fscal,dx00);
 930             ty               = _mm_mul_pd(fscal,dy00);
 931             tz               = _mm_mul_pd(fscal,dz00);
 932
 933             /* Update vectorial force */
 934             fix0             = _mm_add_pd(fix0,tx);
 935             fiy0             = _mm_add_pd(fiy0,ty);
 936             fiz0             = _mm_add_pd(fiz0,tz);
 937
 938             fjx0             = _mm_add_pd(fjx0,tx);
 939             fjy0             = _mm_add_pd(fjy0,ty);
 940             fjz0             = _mm_add_pd(fjz0,tz);
 941
 942             /**************************
 943              * CALCULATE INTERACTIONS *
 944              **************************/
 945
 946             r10              = _mm_mul_pd(rsq10,rinv10);
 947
 948             /* Compute parameters for interactions between i and j atoms */
 949             qq10             = _mm_mul_pd(iq1,jq0);
 950
 951             /* EWALD ELECTROSTATICS */
 952
 953             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 954             ewrt             = _mm_mul_pd(r10,ewtabscale);
 955             ewitab           = _mm_cvttpd_epi32(ewrt);
 956             eweps            = _mm_sub_pd(ewrt,_mm_cvtepi32_pd(ewitab));
 957             gmx_mm_load_2pair_swizzle_pd(ewtab+gmx_mm_extract_epi32(ewitab,0),ewtab+gmx_mm_extract_epi32(ewitab,1),
 958                                          &ewtabF,&ewtabFn);
 959             felec            = _mm_add_pd(_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF),_mm_mul_pd(eweps,ewtabFn));
 960             felec            = _mm_mul_pd(_mm_mul_pd(qq10,rinv10),_mm_sub_pd(rinvsq10,felec));
 961
 962             fscal            = felec;
 963
 964             /* Calculate temporary vectorial force */
 965             tx               = _mm_mul_pd(fscal,dx10);
 966             ty               = _mm_mul_pd(fscal,dy10);
 967             tz               = _mm_mul_pd(fscal,dz10);
 968
 969             /* Update vectorial force */
 970             fix1             = _mm_add_pd(fix1,tx);
 971             fiy1             = _mm_add_pd(fiy1,ty);
 972             fiz1             = _mm_add_pd(fiz1,tz);
 973
 974             fjx0             = _mm_add_pd(fjx0,tx);
 975             fjy0             = _mm_add_pd(fjy0,ty);
 976             fjz0             = _mm_add_pd(fjz0,tz);
 977
 978             /**************************
 979              * CALCULATE INTERACTIONS *
 980              **************************/
 981
 982             r20              = _mm_mul_pd(rsq20,rinv20);
 983
 984             /* Compute parameters for interactions between i and j atoms */
 985             qq20             = _mm_mul_pd(iq2,jq0);
 986
 987             /* EWALD ELECTROSTATICS */
 988
 989             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 990             ewrt             = _mm_mul_pd(r20,ewtabscale);
 991             ewitab           = _mm_cvttpd_epi32(ewrt);
 992             eweps            = _mm_sub_pd(ewrt,_mm_cvtepi32_pd(ewitab));
 993             gmx_mm_load_2pair_swizzle_pd(ewtab+gmx_mm_extract_epi32(ewitab,0),ewtab+gmx_mm_extract_epi32(ewitab,1),
 994                                          &ewtabF,&ewtabFn);
 995             felec            = _mm_add_pd(_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF),_mm_mul_pd(eweps,ewtabFn));
 996             felec            = _mm_mul_pd(_mm_mul_pd(qq20,rinv20),_mm_sub_pd(rinvsq20,felec));
 997
 998             fscal            = felec;
 999
1000             /* Calculate temporary vectorial force */
1001             tx               = _mm_mul_pd(fscal,dx20);
1002             ty               = _mm_mul_pd(fscal,dy20);
1003             tz               = _mm_mul_pd(fscal,dz20);
1004
1005             /* Update vectorial force */
1006             fix2             = _mm_add_pd(fix2,tx);
1007             fiy2             = _mm_add_pd(fiy2,ty);
1008             fiz2             = _mm_add_pd(fiz2,tz);
1009
1010             fjx0             = _mm_add_pd(fjx0,tx);
1011             fjy0             = _mm_add_pd(fjy0,ty);
1012             fjz0             = _mm_add_pd(fjz0,tz);
1013
1014             /**************************
1015              * CALCULATE INTERACTIONS *
1016              **************************/
1017
1018             r30              = _mm_mul_pd(rsq30,rinv30);
1019
1020             /* Compute parameters for interactions between i and j atoms */
1021             qq30             = _mm_mul_pd(iq3,jq0);
1022
1023             /* EWALD ELECTROSTATICS */
1024
1025             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1026             ewrt             = _mm_mul_pd(r30,ewtabscale);
1027             ewitab           = _mm_cvttpd_epi32(ewrt);
1028             eweps            = _mm_sub_pd(ewrt,_mm_cvtepi32_pd(ewitab));
1029             gmx_mm_load_2pair_swizzle_pd(ewtab+gmx_mm_extract_epi32(ewitab,0),ewtab+gmx_mm_extract_epi32(ewitab,1),
1030                                          &ewtabF,&ewtabFn);
1031             felec            = _mm_add_pd(_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF),_mm_mul_pd(eweps,ewtabFn));
1032             felec            = _mm_mul_pd(_mm_mul_pd(qq30,rinv30),_mm_sub_pd(rinvsq30,felec));
1033
1034             fscal            = felec;
1035
1036             /* Calculate temporary vectorial force */
1037             tx               = _mm_mul_pd(fscal,dx30);
1038             ty               = _mm_mul_pd(fscal,dy30);
1039             tz               = _mm_mul_pd(fscal,dz30);
1040
1041             /* Update vectorial force */
1042             fix3             = _mm_add_pd(fix3,tx);
1043             fiy3             = _mm_add_pd(fiy3,ty);
1044             fiz3             = _mm_add_pd(fiz3,tz);
1045
1046             fjx0             = _mm_add_pd(fjx0,tx);
1047             fjy0             = _mm_add_pd(fjy0,ty);
1048             fjz0             = _mm_add_pd(fjz0,tz);
1049
1050             gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,fjx0,fjy0,fjz0);
1051
1052             /* Inner loop uses 157 flops */
1053         }
1054
1055         if(jidx<j_index_end)
1056         {
1057
1058             jnrA             = jjnr[jidx];
1059             j_coord_offsetA  = DIM*jnrA;
1060
1061             /* load j atom coordinates */
1062             gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
1063                                               &jx0,&jy0,&jz0);
1064
1065             /* Calculate displacement vector */
1066             dx00             = _mm_sub_pd(ix0,jx0);
1067             dy00             = _mm_sub_pd(iy0,jy0);
1068             dz00             = _mm_sub_pd(iz0,jz0);
1069             dx10             = _mm_sub_pd(ix1,jx0);
1070             dy10             = _mm_sub_pd(iy1,jy0);
1071             dz10             = _mm_sub_pd(iz1,jz0);
1072             dx20             = _mm_sub_pd(ix2,jx0);
1073             dy20             = _mm_sub_pd(iy2,jy0);
1074             dz20             = _mm_sub_pd(iz2,jz0);
1075             dx30             = _mm_sub_pd(ix3,jx0);
1076             dy30             = _mm_sub_pd(iy3,jy0);
1077             dz30             = _mm_sub_pd(iz3,jz0);
1078
1079             /* Calculate squared distance and things based on it */
1080             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
1081             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
1082             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
1083             rsq30            = gmx_mm_calc_rsq_pd(dx30,dy30,dz30);
1084
1085             rinv00           = gmx_mm_invsqrt_pd(rsq00);
1086             rinv10           = gmx_mm_invsqrt_pd(rsq10);
1087             rinv20           = gmx_mm_invsqrt_pd(rsq20);
1088             rinv30           = gmx_mm_invsqrt_pd(rsq30);
1089
1090             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
1091             rinvsq10         = _mm_mul_pd(rinv10,rinv10);
1092             rinvsq20         = _mm_mul_pd(rinv20,rinv20);
1093             rinvsq30         = _mm_mul_pd(rinv30,rinv30);
1094
1095             /* Load parameters for j particles */
1096             jq0              = _mm_load_sd(charge+jnrA+0);
1097             vdwjidx0A        = 2*vdwtype[jnrA+0];
1098
1099             fjx0             = _mm_setzero_pd();
1100             fjy0             = _mm_setzero_pd();
1101             fjz0             = _mm_setzero_pd();
1102
1103             /**************************
1104              * CALCULATE INTERACTIONS *
1105              **************************/
1106
1107             r00              = _mm_mul_pd(rsq00,rinv00);
1108
1109             /* Compute parameters for interactions between i and j atoms */
1110             gmx_mm_load_1pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,&c6_00,&c12_00);
1111
1112             c6grid_00       = gmx_mm_load_1real_pd(vdwgridparam+vdwioffset0+vdwjidx0A);
1113
1114             /* Analytical LJ-PME */
1115             rinvsix          = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
1116             ewcljrsq         = _mm_mul_pd(ewclj2,rsq00);
1117             ewclj6           = _mm_mul_pd(ewclj2,_mm_mul_pd(ewclj2,ewclj2));
1118             exponent         = gmx_simd_exp_d(ewcljrsq);
1119             /* poly = exp(-(beta*r)^2) * (1 + (beta*r)^2 + (beta*r)^4 /2) */
1120             poly             = _mm_mul_pd(exponent,_mm_add_pd(_mm_sub_pd(one,ewcljrsq),_mm_mul_pd(_mm_mul_pd(ewcljrsq,ewcljrsq),one_half)));
1121             /* f6A = 6 * C6grid * (1 - poly) */
1122             f6A              = _mm_mul_pd(c6grid_00,_mm_sub_pd(one,poly));
1123             /* f6B = C6grid * exponent * beta^6 */
1124             f6B              = _mm_mul_pd(_mm_mul_pd(c6grid_00,one_sixth),_mm_mul_pd(exponent,ewclj6));
1125             /* fvdw = 12*C12/r13 - ((6*C6 - f6A)/r6 + f6B)/r */
1126             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);
1127
1128             fscal            = fvdw;
1129
1130             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1131
1132             /* Calculate temporary vectorial force */
1133             tx               = _mm_mul_pd(fscal,dx00);
1134             ty               = _mm_mul_pd(fscal,dy00);
1135             tz               = _mm_mul_pd(fscal,dz00);
1136
1137             /* Update vectorial force */
1138             fix0             = _mm_add_pd(fix0,tx);
1139             fiy0             = _mm_add_pd(fiy0,ty);
1140             fiz0             = _mm_add_pd(fiz0,tz);
1141
1142             fjx0             = _mm_add_pd(fjx0,tx);
1143             fjy0             = _mm_add_pd(fjy0,ty);
1144             fjz0             = _mm_add_pd(fjz0,tz);
1145
1146             /**************************
1147              * CALCULATE INTERACTIONS *
1148              **************************/
1149
1150             r10              = _mm_mul_pd(rsq10,rinv10);
1151
1152             /* Compute parameters for interactions between i and j atoms */
1153             qq10             = _mm_mul_pd(iq1,jq0);
1154
1155             /* EWALD ELECTROSTATICS */
1156
1157             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1158             ewrt             = _mm_mul_pd(r10,ewtabscale);
1159             ewitab           = _mm_cvttpd_epi32(ewrt);
1160             eweps            = _mm_sub_pd(ewrt,_mm_cvtepi32_pd(ewitab));
1161             gmx_mm_load_1pair_swizzle_pd(ewtab+gmx_mm_extract_epi32(ewitab,0),&ewtabF,&ewtabFn);
1162             felec            = _mm_add_pd(_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF),_mm_mul_pd(eweps,ewtabFn));
1163             felec            = _mm_mul_pd(_mm_mul_pd(qq10,rinv10),_mm_sub_pd(rinvsq10,felec));
1164
1165             fscal            = felec;
1166
1167             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1168
1169             /* Calculate temporary vectorial force */
1170             tx               = _mm_mul_pd(fscal,dx10);
1171             ty               = _mm_mul_pd(fscal,dy10);
1172             tz               = _mm_mul_pd(fscal,dz10);
1173
1174             /* Update vectorial force */
1175             fix1             = _mm_add_pd(fix1,tx);
1176             fiy1             = _mm_add_pd(fiy1,ty);
1177             fiz1             = _mm_add_pd(fiz1,tz);
1178
1179             fjx0             = _mm_add_pd(fjx0,tx);
1180             fjy0             = _mm_add_pd(fjy0,ty);
1181             fjz0             = _mm_add_pd(fjz0,tz);
1182
1183             /**************************
1184              * CALCULATE INTERACTIONS *
1185              **************************/
1186
1187             r20              = _mm_mul_pd(rsq20,rinv20);
1188
1189             /* Compute parameters for interactions between i and j atoms */
1190             qq20             = _mm_mul_pd(iq2,jq0);
1191
1192             /* EWALD ELECTROSTATICS */
1193
1194             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1195             ewrt             = _mm_mul_pd(r20,ewtabscale);
1196             ewitab           = _mm_cvttpd_epi32(ewrt);
1197             eweps            = _mm_sub_pd(ewrt,_mm_cvtepi32_pd(ewitab));
1198             gmx_mm_load_1pair_swizzle_pd(ewtab+gmx_mm_extract_epi32(ewitab,0),&ewtabF,&ewtabFn);
1199             felec            = _mm_add_pd(_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF),_mm_mul_pd(eweps,ewtabFn));
1200             felec            = _mm_mul_pd(_mm_mul_pd(qq20,rinv20),_mm_sub_pd(rinvsq20,felec));
1201
1202             fscal            = felec;
1203
1204             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1205
1206             /* Calculate temporary vectorial force */
1207             tx               = _mm_mul_pd(fscal,dx20);
1208             ty               = _mm_mul_pd(fscal,dy20);
1209             tz               = _mm_mul_pd(fscal,dz20);
1210
1211             /* Update vectorial force */
1212             fix2             = _mm_add_pd(fix2,tx);
1213             fiy2             = _mm_add_pd(fiy2,ty);
1214             fiz2             = _mm_add_pd(fiz2,tz);
1215
1216             fjx0             = _mm_add_pd(fjx0,tx);
1217             fjy0             = _mm_add_pd(fjy0,ty);
1218             fjz0             = _mm_add_pd(fjz0,tz);
1219
1220             /**************************
1221              * CALCULATE INTERACTIONS *
1222              **************************/
1223
1224             r30              = _mm_mul_pd(rsq30,rinv30);
1225
1226             /* Compute parameters for interactions between i and j atoms */
1227             qq30             = _mm_mul_pd(iq3,jq0);
1228
1229             /* EWALD ELECTROSTATICS */
1230
1231             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1232             ewrt             = _mm_mul_pd(r30,ewtabscale);
1233             ewitab           = _mm_cvttpd_epi32(ewrt);
1234             eweps            = _mm_sub_pd(ewrt,_mm_cvtepi32_pd(ewitab));
1235             gmx_mm_load_1pair_swizzle_pd(ewtab+gmx_mm_extract_epi32(ewitab,0),&ewtabF,&ewtabFn);
1236             felec            = _mm_add_pd(_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF),_mm_mul_pd(eweps,ewtabFn));
1237             felec            = _mm_mul_pd(_mm_mul_pd(qq30,rinv30),_mm_sub_pd(rinvsq30,felec));
1238
1239             fscal            = felec;
1240
1241             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1242
1243             /* Calculate temporary vectorial force */
1244             tx               = _mm_mul_pd(fscal,dx30);
1245             ty               = _mm_mul_pd(fscal,dy30);
1246             tz               = _mm_mul_pd(fscal,dz30);
1247
1248             /* Update vectorial force */
1249             fix3             = _mm_add_pd(fix3,tx);
1250             fiy3             = _mm_add_pd(fiy3,ty);
1251             fiz3             = _mm_add_pd(fiz3,tz);
1252
1253             fjx0             = _mm_add_pd(fjx0,tx);
1254             fjy0             = _mm_add_pd(fjy0,ty);
1255             fjz0             = _mm_add_pd(fjz0,tz);
1256
1257             gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,fjx0,fjy0,fjz0);
1258
1259             /* Inner loop uses 157 flops */
1260         }
1261
1262         /* End of innermost loop */
1263
1264         gmx_mm_update_iforce_4atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
1265                                               f+i_coord_offset,fshift+i_shift_offset);
1266
1267         /* Increment number of inner iterations */
1268         inneriter                  += j_index_end - j_index_start;
1269
1270         /* Outer loop uses 24 flops */
1271     }
1272
1273     /* Increment number of outer iterations */
1274     outeriter        += nri;
1275
1276     /* Update outer/inner flops */
1277
1278     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W4_F,outeriter*24 + inneriter*157);
1279 }