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 #include "config.h"
  39
  40 #include <math.h>
  41
  42 #include "../nb_kernel.h"
  43 #include "gromacs/legacyheaders/types/simple.h"
  44 #include "gromacs/math/vec.h"
  45 #include "gromacs/legacyheaders/nrnb.h"
  46
  47 #include "gromacs/simd/math_x86_sse2_double.h"
  48 #include "kernelutil_x86_sse2_double.h"
  49
  50 /*
  51  * Gromacs nonbonded kernel:   nb_kernel_ElecEw_VdwLJEw_GeomW4P1_VF_sse2_double
  52  * Electrostatics interaction: Ewald
  53  * VdW interaction:            LJEwald
  54  * Geometry:                   Water4-Particle
  55  * Calculate force/pot:        PotentialAndForce
  56  */
  57 void
  58 nb_kernel_ElecEw_VdwLJEw_GeomW4P1_VF_sse2_double
  59                     (t_nblist                    * gmx_restrict       nlist,
  60                      rvec                        * gmx_restrict          xx,
  61                      rvec                        * gmx_restrict          ff,
  62                      t_forcerec                  * gmx_restrict          fr,
  63                      t_mdatoms                   * gmx_restrict     mdatoms,
  64                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
  65                      t_nrnb                      * gmx_restrict        nrnb)
  66 {
  67     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
  68      * just 0 for non-waters.
  69      * Suffixes A,B refer to j loop unrolling done with SSE double precision, e.g. for the two different
  70      * jnr indices corresponding to data put in the four positions in the SIMD register.
  71      */
  72     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
  73     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
  74     int              jnrA,jnrB;
  75     int              j_coord_offsetA,j_coord_offsetB;
  76     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
  77     real             rcutoff_scalar;
  78     real             *shiftvec,*fshift,*x,*f;
  79     __m128d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
  80     int              vdwioffset0;
  81     __m128d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
  82     int              vdwioffset1;
  83     __m128d          ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
  84     int              vdwioffset2;
  85     __m128d          ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
  86     int              vdwioffset3;
  87     __m128d          ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
  88     int              vdwjidx0A,vdwjidx0B;
  89     __m128d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  90     __m128d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  91     __m128d          dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
  92     __m128d          dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
  93     __m128d          dx30,dy30,dz30,rsq30,rinv30,rinvsq30,r30,qq30,c6_30,c12_30;
  94     __m128d          velec,felec,velecsum,facel,crf,krf,krf2;
  95     real             *charge;
  96     int              nvdwtype;
  97     __m128d          rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
  98     int              *vdwtype;
  99     real             *vdwparam;
 100     __m128d          one_sixth   = _mm_set1_pd(1.0/6.0);
 101     __m128d          one_twelfth = _mm_set1_pd(1.0/12.0);
 102     __m128d           c6grid_00;
 103     __m128d           c6grid_10;
 104     __m128d           c6grid_20;
 105     __m128d           c6grid_30;
 106     __m128d           ewclj,ewclj2,ewclj6,ewcljrsq,poly,exponent,f6A,f6B,sh_lj_ewald;
 107     real             *vdwgridparam;
 108     __m128d           one_half = _mm_set1_pd(0.5);
 109     __m128d           minus_one = _mm_set1_pd(-1.0);
 110     __m128i          ewitab;
 111     __m128d          ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
 112     real             *ewtab;
 113     __m128d          dummy_mask,cutoff_mask;
 114     __m128d          signbit   = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
 115     __m128d          one     = _mm_set1_pd(1.0);
 116     __m128d          two     = _mm_set1_pd(2.0);
 117     x                = xx[0];
 118     f                = ff[0];
 119
 120     nri              = nlist->nri;
 121     iinr             = nlist->iinr;
 122     jindex           = nlist->jindex;
 123     jjnr             = nlist->jjnr;
 124     shiftidx         = nlist->shift;
 125     gid              = nlist->gid;
 126     shiftvec         = fr->shift_vec[0];
 127     fshift           = fr->fshift[0];
 128     facel            = _mm_set1_pd(fr->epsfac);
 129     charge           = mdatoms->chargeA;
 130     nvdwtype         = fr->ntype;
 131     vdwparam         = fr->nbfp;
 132     vdwtype          = mdatoms->typeA;
 133     vdwgridparam     = fr->ljpme_c6grid;
 134     sh_lj_ewald      = _mm_set1_pd(fr->ic->sh_lj_ewald);
 135     ewclj            = _mm_set1_pd(fr->ewaldcoeff_lj);
 136     ewclj2           = _mm_mul_pd(minus_one,_mm_mul_pd(ewclj,ewclj));
 137
 138     sh_ewald         = _mm_set1_pd(fr->ic->sh_ewald);
 139     ewtab            = fr->ic->tabq_coul_FDV0;
 140     ewtabscale       = _mm_set1_pd(fr->ic->tabq_scale);
 141     ewtabhalfspace   = _mm_set1_pd(0.5/fr->ic->tabq_scale);
 142
 143     /* Setup water-specific parameters */
 144     inr              = nlist->iinr[0];
 145     iq1              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+1]));
 146     iq2              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+2]));
 147     iq3              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+3]));
 148     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 149
 150     /* Avoid stupid compiler warnings */
 151     jnrA = jnrB = 0;
 152     j_coord_offsetA = 0;
 153     j_coord_offsetB = 0;
 154
 155     outeriter        = 0;
 156     inneriter        = 0;
 157
 158     /* Start outer loop over neighborlists */
 159     for(iidx=0; iidx<nri; iidx++)
 160     {
 161         /* Load shift vector for this list */
 162         i_shift_offset   = DIM*shiftidx[iidx];
 163
 164         /* Load limits for loop over neighbors */
 165         j_index_start    = jindex[iidx];
 166         j_index_end      = jindex[iidx+1];
 167
 168         /* Get outer coordinate index */
 169         inr              = iinr[iidx];
 170         i_coord_offset   = DIM*inr;
 171
 172         /* Load i particle coords and add shift vector */
 173         gmx_mm_load_shift_and_4rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
 174                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
 175
 176         fix0             = _mm_setzero_pd();
 177         fiy0             = _mm_setzero_pd();
 178         fiz0             = _mm_setzero_pd();
 179         fix1             = _mm_setzero_pd();
 180         fiy1             = _mm_setzero_pd();
 181         fiz1             = _mm_setzero_pd();
 182         fix2             = _mm_setzero_pd();
 183         fiy2             = _mm_setzero_pd();
 184         fiz2             = _mm_setzero_pd();
 185         fix3             = _mm_setzero_pd();
 186         fiy3             = _mm_setzero_pd();
 187         fiz3             = _mm_setzero_pd();
 188
 189         /* Reset potential sums */
 190         velecsum         = _mm_setzero_pd();
 191         vvdwsum          = _mm_setzero_pd();
 192
 193         /* Start inner kernel loop */
 194         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 195         {
 196
 197             /* Get j neighbor index, and coordinate index */
 198             jnrA             = jjnr[jidx];
 199             jnrB             = jjnr[jidx+1];
 200             j_coord_offsetA  = DIM*jnrA;
 201             j_coord_offsetB  = DIM*jnrB;
 202
 203             /* load j atom coordinates */
 204             gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 205                                               &jx0,&jy0,&jz0);
 206
 207             /* Calculate displacement vector */
 208             dx00             = _mm_sub_pd(ix0,jx0);
 209             dy00             = _mm_sub_pd(iy0,jy0);
 210             dz00             = _mm_sub_pd(iz0,jz0);
 211             dx10             = _mm_sub_pd(ix1,jx0);
 212             dy10             = _mm_sub_pd(iy1,jy0);
 213             dz10             = _mm_sub_pd(iz1,jz0);
 214             dx20             = _mm_sub_pd(ix2,jx0);
 215             dy20             = _mm_sub_pd(iy2,jy0);
 216             dz20             = _mm_sub_pd(iz2,jz0);
 217             dx30             = _mm_sub_pd(ix3,jx0);
 218             dy30             = _mm_sub_pd(iy3,jy0);
 219             dz30             = _mm_sub_pd(iz3,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             rsq30            = gmx_mm_calc_rsq_pd(dx30,dy30,dz30);
 226
 227             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 228             rinv10           = gmx_mm_invsqrt_pd(rsq10);
 229             rinv20           = gmx_mm_invsqrt_pd(rsq20);
 230             rinv30           = gmx_mm_invsqrt_pd(rsq30);
 231
 232             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 233             rinvsq10         = _mm_mul_pd(rinv10,rinv10);
 234             rinvsq20         = _mm_mul_pd(rinv20,rinv20);
 235             rinvsq30         = _mm_mul_pd(rinv30,rinv30);
 236
 237             /* Load parameters for j particles */
 238             jq0              = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
 239             vdwjidx0A        = 2*vdwtype[jnrA+0];
 240             vdwjidx0B        = 2*vdwtype[jnrB+0];
 241
 242             fjx0             = _mm_setzero_pd();
 243             fjy0             = _mm_setzero_pd();
 244             fjz0             = _mm_setzero_pd();
 245
 246             /**************************
 247              * CALCULATE INTERACTIONS *
 248              **************************/
 249
 250             r00              = _mm_mul_pd(rsq00,rinv00);
 251
 252             /* Compute parameters for interactions between i and j atoms */
 253             gmx_mm_load_2pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,
 254                                          vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
 255
 256             c6grid_00       = gmx_mm_load_2real_swizzle_pd(vdwgridparam+vdwioffset0+vdwjidx0A,
 257                                                                vdwgridparam+vdwioffset0+vdwjidx0B);
 258
 259             /* Analytical LJ-PME */
 260             rinvsix          = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 261             ewcljrsq         = _mm_mul_pd(ewclj2,rsq00);
 262             ewclj6           = _mm_mul_pd(ewclj2,_mm_mul_pd(ewclj2,ewclj2));
 263             exponent         = gmx_simd_exp_d(ewcljrsq);
 264             /* poly = exp(-(beta*r)^2) * (1 + (beta*r)^2 + (beta*r)^4 /2) */
 265             poly             = _mm_mul_pd(exponent,_mm_add_pd(_mm_sub_pd(one,ewcljrsq),_mm_mul_pd(_mm_mul_pd(ewcljrsq,ewcljrsq),one_half)));
 266             /* vvdw6 = [C6 - C6grid * (1-poly)]/r6 */
 267             vvdw6            = _mm_mul_pd(_mm_sub_pd(c6_00,_mm_mul_pd(c6grid_00,_mm_sub_pd(one,poly))),rinvsix);
 268             vvdw12           = _mm_mul_pd(c12_00,_mm_mul_pd(rinvsix,rinvsix));
 269             vvdw             = _mm_sub_pd(_mm_mul_pd(vvdw12,one_twelfth),_mm_mul_pd(vvdw6,one_sixth));
 270             /* fvdw = vvdw12/r - (vvdw6/r + (C6grid * exponent * beta^6)/r) */
 271             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);
 272
 273             /* Update potential sum for this i atom from the interaction with this j atom. */
 274             vvdwsum          = _mm_add_pd(vvdwsum,vvdw);
 275
 276             fscal            = fvdw;
 277
 278             /* Calculate temporary vectorial force */
 279             tx               = _mm_mul_pd(fscal,dx00);
 280             ty               = _mm_mul_pd(fscal,dy00);
 281             tz               = _mm_mul_pd(fscal,dz00);
 282
 283             /* Update vectorial force */
 284             fix0             = _mm_add_pd(fix0,tx);
 285             fiy0             = _mm_add_pd(fiy0,ty);
 286             fiz0             = _mm_add_pd(fiz0,tz);
 287
 288             fjx0             = _mm_add_pd(fjx0,tx);
 289             fjy0             = _mm_add_pd(fjy0,ty);
 290             fjz0             = _mm_add_pd(fjz0,tz);
 291
 292             /**************************
 293              * CALCULATE INTERACTIONS *
 294              **************************/
 295
 296             r10              = _mm_mul_pd(rsq10,rinv10);
 297
 298             /* Compute parameters for interactions between i and j atoms */
 299             qq10             = _mm_mul_pd(iq1,jq0);
 300
 301             /* EWALD ELECTROSTATICS */
 302
 303             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 304             ewrt             = _mm_mul_pd(r10,ewtabscale);
 305             ewitab           = _mm_cvttpd_epi32(ewrt);
 306             eweps            = _mm_sub_pd(ewrt,_mm_cvtepi32_pd(ewitab));
 307             ewitab           = _mm_slli_epi32(ewitab,2);
 308             ewtabF           = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 309             ewtabD           = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,1) );
 310             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 311             ewtabV           = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,0) +2);
 312             ewtabFn          = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,1) +2);
 313             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 314             felec            = _mm_add_pd(ewtabF,_mm_mul_pd(eweps,ewtabD));
 315             velec            = _mm_sub_pd(ewtabV,_mm_mul_pd(_mm_mul_pd(ewtabhalfspace,eweps),_mm_add_pd(ewtabF,felec)));
 316             velec            = _mm_mul_pd(qq10,_mm_sub_pd(rinv10,velec));
 317             felec            = _mm_mul_pd(_mm_mul_pd(qq10,rinv10),_mm_sub_pd(rinvsq10,felec));
 318
 319             /* Update potential sum for this i atom from the interaction with this j atom. */
 320             velecsum         = _mm_add_pd(velecsum,velec);
 321
 322             fscal            = felec;
 323
 324             /* Calculate temporary vectorial force */
 325             tx               = _mm_mul_pd(fscal,dx10);
 326             ty               = _mm_mul_pd(fscal,dy10);
 327             tz               = _mm_mul_pd(fscal,dz10);
 328
 329             /* Update vectorial force */
 330             fix1             = _mm_add_pd(fix1,tx);
 331             fiy1             = _mm_add_pd(fiy1,ty);
 332             fiz1             = _mm_add_pd(fiz1,tz);
 333
 334             fjx0             = _mm_add_pd(fjx0,tx);
 335             fjy0             = _mm_add_pd(fjy0,ty);
 336             fjz0             = _mm_add_pd(fjz0,tz);
 337
 338             /**************************
 339              * CALCULATE INTERACTIONS *
 340              **************************/
 341
 342             r20              = _mm_mul_pd(rsq20,rinv20);
 343
 344             /* Compute parameters for interactions between i and j atoms */
 345             qq20             = _mm_mul_pd(iq2,jq0);
 346
 347             /* EWALD ELECTROSTATICS */
 348
 349             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 350             ewrt             = _mm_mul_pd(r20,ewtabscale);
 351             ewitab           = _mm_cvttpd_epi32(ewrt);
 352             eweps            = _mm_sub_pd(ewrt,_mm_cvtepi32_pd(ewitab));
 353             ewitab           = _mm_slli_epi32(ewitab,2);
 354             ewtabF           = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 355             ewtabD           = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,1) );
 356             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 357             ewtabV           = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,0) +2);
 358             ewtabFn          = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,1) +2);
 359             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 360             felec            = _mm_add_pd(ewtabF,_mm_mul_pd(eweps,ewtabD));
 361             velec            = _mm_sub_pd(ewtabV,_mm_mul_pd(_mm_mul_pd(ewtabhalfspace,eweps),_mm_add_pd(ewtabF,felec)));
 362             velec            = _mm_mul_pd(qq20,_mm_sub_pd(rinv20,velec));
 363             felec            = _mm_mul_pd(_mm_mul_pd(qq20,rinv20),_mm_sub_pd(rinvsq20,felec));
 364
 365             /* Update potential sum for this i atom from the interaction with this j atom. */
 366             velecsum         = _mm_add_pd(velecsum,velec);
 367
 368             fscal            = felec;
 369
 370             /* Calculate temporary vectorial force */
 371             tx               = _mm_mul_pd(fscal,dx20);
 372             ty               = _mm_mul_pd(fscal,dy20);
 373             tz               = _mm_mul_pd(fscal,dz20);
 374
 375             /* Update vectorial force */
 376             fix2             = _mm_add_pd(fix2,tx);
 377             fiy2             = _mm_add_pd(fiy2,ty);
 378             fiz2             = _mm_add_pd(fiz2,tz);
 379
 380             fjx0             = _mm_add_pd(fjx0,tx);
 381             fjy0             = _mm_add_pd(fjy0,ty);
 382             fjz0             = _mm_add_pd(fjz0,tz);
 383
 384             /**************************
 385              * CALCULATE INTERACTIONS *
 386              **************************/
 387
 388             r30              = _mm_mul_pd(rsq30,rinv30);
 389
 390             /* Compute parameters for interactions between i and j atoms */
 391             qq30             = _mm_mul_pd(iq3,jq0);
 392
 393             /* EWALD ELECTROSTATICS */
 394
 395             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 396             ewrt             = _mm_mul_pd(r30,ewtabscale);
 397             ewitab           = _mm_cvttpd_epi32(ewrt);
 398             eweps            = _mm_sub_pd(ewrt,_mm_cvtepi32_pd(ewitab));
 399             ewitab           = _mm_slli_epi32(ewitab,2);
 400             ewtabF           = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 401             ewtabD           = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,1) );
 402             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 403             ewtabV           = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,0) +2);
 404             ewtabFn          = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,1) +2);
 405             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 406             felec            = _mm_add_pd(ewtabF,_mm_mul_pd(eweps,ewtabD));
 407             velec            = _mm_sub_pd(ewtabV,_mm_mul_pd(_mm_mul_pd(ewtabhalfspace,eweps),_mm_add_pd(ewtabF,felec)));
 408             velec            = _mm_mul_pd(qq30,_mm_sub_pd(rinv30,velec));
 409             felec            = _mm_mul_pd(_mm_mul_pd(qq30,rinv30),_mm_sub_pd(rinvsq30,felec));
 410
 411             /* Update potential sum for this i atom from the interaction with this j atom. */
 412             velecsum         = _mm_add_pd(velecsum,velec);
 413
 414             fscal            = felec;
 415
 416             /* Calculate temporary vectorial force */
 417             tx               = _mm_mul_pd(fscal,dx30);
 418             ty               = _mm_mul_pd(fscal,dy30);
 419             tz               = _mm_mul_pd(fscal,dz30);
 420
 421             /* Update vectorial force */
 422             fix3             = _mm_add_pd(fix3,tx);
 423             fiy3             = _mm_add_pd(fiy3,ty);
 424             fiz3             = _mm_add_pd(fiz3,tz);
 425
 426             fjx0             = _mm_add_pd(fjx0,tx);
 427             fjy0             = _mm_add_pd(fjy0,ty);
 428             fjz0             = _mm_add_pd(fjz0,tz);
 429
 430             gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,fjx0,fjy0,fjz0);
 431
 432             /* Inner loop uses 177 flops */
 433         }
 434
 435         if(jidx<j_index_end)
 436         {
 437
 438             jnrA             = jjnr[jidx];
 439             j_coord_offsetA  = DIM*jnrA;
 440
 441             /* load j atom coordinates */
 442             gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
 443                                               &jx0,&jy0,&jz0);
 444
 445             /* Calculate displacement vector */
 446             dx00             = _mm_sub_pd(ix0,jx0);
 447             dy00             = _mm_sub_pd(iy0,jy0);
 448             dz00             = _mm_sub_pd(iz0,jz0);
 449             dx10             = _mm_sub_pd(ix1,jx0);
 450             dy10             = _mm_sub_pd(iy1,jy0);
 451             dz10             = _mm_sub_pd(iz1,jz0);
 452             dx20             = _mm_sub_pd(ix2,jx0);
 453             dy20             = _mm_sub_pd(iy2,jy0);
 454             dz20             = _mm_sub_pd(iz2,jz0);
 455             dx30             = _mm_sub_pd(ix3,jx0);
 456             dy30             = _mm_sub_pd(iy3,jy0);
 457             dz30             = _mm_sub_pd(iz3,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             rsq30            = gmx_mm_calc_rsq_pd(dx30,dy30,dz30);
 464
 465             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 466             rinv10           = gmx_mm_invsqrt_pd(rsq10);
 467             rinv20           = gmx_mm_invsqrt_pd(rsq20);
 468             rinv30           = gmx_mm_invsqrt_pd(rsq30);
 469
 470             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 471             rinvsq10         = _mm_mul_pd(rinv10,rinv10);
 472             rinvsq20         = _mm_mul_pd(rinv20,rinv20);
 473             rinvsq30         = _mm_mul_pd(rinv30,rinv30);
 474
 475             /* Load parameters for j particles */
 476             jq0              = _mm_load_sd(charge+jnrA+0);
 477             vdwjidx0A        = 2*vdwtype[jnrA+0];
 478
 479             fjx0             = _mm_setzero_pd();
 480             fjy0             = _mm_setzero_pd();
 481             fjz0             = _mm_setzero_pd();
 482
 483             /**************************
 484              * CALCULATE INTERACTIONS *
 485              **************************/
 486
 487             r00              = _mm_mul_pd(rsq00,rinv00);
 488
 489             /* Compute parameters for interactions between i and j atoms */
 490             gmx_mm_load_1pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,&c6_00,&c12_00);
 491
 492             c6grid_00       = gmx_mm_load_1real_pd(vdwgridparam+vdwioffset0+vdwjidx0A);
 493
 494             /* Analytical LJ-PME */
 495             rinvsix          = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 496             ewcljrsq         = _mm_mul_pd(ewclj2,rsq00);
 497             ewclj6           = _mm_mul_pd(ewclj2,_mm_mul_pd(ewclj2,ewclj2));
 498             exponent         = gmx_simd_exp_d(ewcljrsq);
 499             /* poly = exp(-(beta*r)^2) * (1 + (beta*r)^2 + (beta*r)^4 /2) */
 500             poly             = _mm_mul_pd(exponent,_mm_add_pd(_mm_sub_pd(one,ewcljrsq),_mm_mul_pd(_mm_mul_pd(ewcljrsq,ewcljrsq),one_half)));
 501             /* vvdw6 = [C6 - C6grid * (1-poly)]/r6 */
 502             vvdw6            = _mm_mul_pd(_mm_sub_pd(c6_00,_mm_mul_pd(c6grid_00,_mm_sub_pd(one,poly))),rinvsix);
 503             vvdw12           = _mm_mul_pd(c12_00,_mm_mul_pd(rinvsix,rinvsix));
 504             vvdw             = _mm_sub_pd(_mm_mul_pd(vvdw12,one_twelfth),_mm_mul_pd(vvdw6,one_sixth));
 505             /* fvdw = vvdw12/r - (vvdw6/r + (C6grid * exponent * beta^6)/r) */
 506             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);
 507
 508             /* Update potential sum for this i atom from the interaction with this j atom. */
 509             vvdw             = _mm_unpacklo_pd(vvdw,_mm_setzero_pd());
 510             vvdwsum          = _mm_add_pd(vvdwsum,vvdw);
 511
 512             fscal            = fvdw;
 513
 514             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 515
 516             /* Calculate temporary vectorial force */
 517             tx               = _mm_mul_pd(fscal,dx00);
 518             ty               = _mm_mul_pd(fscal,dy00);
 519             tz               = _mm_mul_pd(fscal,dz00);
 520
 521             /* Update vectorial force */
 522             fix0             = _mm_add_pd(fix0,tx);
 523             fiy0             = _mm_add_pd(fiy0,ty);
 524             fiz0             = _mm_add_pd(fiz0,tz);
 525
 526             fjx0             = _mm_add_pd(fjx0,tx);
 527             fjy0             = _mm_add_pd(fjy0,ty);
 528             fjz0             = _mm_add_pd(fjz0,tz);
 529
 530             /**************************
 531              * CALCULATE INTERACTIONS *
 532              **************************/
 533
 534             r10              = _mm_mul_pd(rsq10,rinv10);
 535
 536             /* Compute parameters for interactions between i and j atoms */
 537             qq10             = _mm_mul_pd(iq1,jq0);
 538
 539             /* EWALD ELECTROSTATICS */
 540
 541             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 542             ewrt             = _mm_mul_pd(r10,ewtabscale);
 543             ewitab           = _mm_cvttpd_epi32(ewrt);
 544             eweps            = _mm_sub_pd(ewrt,_mm_cvtepi32_pd(ewitab));
 545             ewitab           = _mm_slli_epi32(ewitab,2);
 546             ewtabF           = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 547             ewtabD           = _mm_setzero_pd();
 548             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 549             ewtabV           = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,0) +2);
 550             ewtabFn          = _mm_setzero_pd();
 551             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 552             felec            = _mm_add_pd(ewtabF,_mm_mul_pd(eweps,ewtabD));
 553             velec            = _mm_sub_pd(ewtabV,_mm_mul_pd(_mm_mul_pd(ewtabhalfspace,eweps),_mm_add_pd(ewtabF,felec)));
 554             velec            = _mm_mul_pd(qq10,_mm_sub_pd(rinv10,velec));
 555             felec            = _mm_mul_pd(_mm_mul_pd(qq10,rinv10),_mm_sub_pd(rinvsq10,felec));
 556
 557             /* Update potential sum for this i atom from the interaction with this j atom. */
 558             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 559             velecsum         = _mm_add_pd(velecsum,velec);
 560
 561             fscal            = felec;
 562
 563             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 564
 565             /* Calculate temporary vectorial force */
 566             tx               = _mm_mul_pd(fscal,dx10);
 567             ty               = _mm_mul_pd(fscal,dy10);
 568             tz               = _mm_mul_pd(fscal,dz10);
 569
 570             /* Update vectorial force */
 571             fix1             = _mm_add_pd(fix1,tx);
 572             fiy1             = _mm_add_pd(fiy1,ty);
 573             fiz1             = _mm_add_pd(fiz1,tz);
 574
 575             fjx0             = _mm_add_pd(fjx0,tx);
 576             fjy0             = _mm_add_pd(fjy0,ty);
 577             fjz0             = _mm_add_pd(fjz0,tz);
 578
 579             /**************************
 580              * CALCULATE INTERACTIONS *
 581              **************************/
 582
 583             r20              = _mm_mul_pd(rsq20,rinv20);
 584
 585             /* Compute parameters for interactions between i and j atoms */
 586             qq20             = _mm_mul_pd(iq2,jq0);
 587
 588             /* EWALD ELECTROSTATICS */
 589
 590             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 591             ewrt             = _mm_mul_pd(r20,ewtabscale);
 592             ewitab           = _mm_cvttpd_epi32(ewrt);
 593             eweps            = _mm_sub_pd(ewrt,_mm_cvtepi32_pd(ewitab));
 594             ewitab           = _mm_slli_epi32(ewitab,2);
 595             ewtabF           = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 596             ewtabD           = _mm_setzero_pd();
 597             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 598             ewtabV           = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,0) +2);
 599             ewtabFn          = _mm_setzero_pd();
 600             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 601             felec            = _mm_add_pd(ewtabF,_mm_mul_pd(eweps,ewtabD));
 602             velec            = _mm_sub_pd(ewtabV,_mm_mul_pd(_mm_mul_pd(ewtabhalfspace,eweps),_mm_add_pd(ewtabF,felec)));
 603             velec            = _mm_mul_pd(qq20,_mm_sub_pd(rinv20,velec));
 604             felec            = _mm_mul_pd(_mm_mul_pd(qq20,rinv20),_mm_sub_pd(rinvsq20,felec));
 605
 606             /* Update potential sum for this i atom from the interaction with this j atom. */
 607             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 608             velecsum         = _mm_add_pd(velecsum,velec);
 609
 610             fscal            = felec;
 611
 612             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 613
 614             /* Calculate temporary vectorial force */
 615             tx               = _mm_mul_pd(fscal,dx20);
 616             ty               = _mm_mul_pd(fscal,dy20);
 617             tz               = _mm_mul_pd(fscal,dz20);
 618
 619             /* Update vectorial force */
 620             fix2             = _mm_add_pd(fix2,tx);
 621             fiy2             = _mm_add_pd(fiy2,ty);
 622             fiz2             = _mm_add_pd(fiz2,tz);
 623
 624             fjx0             = _mm_add_pd(fjx0,tx);
 625             fjy0             = _mm_add_pd(fjy0,ty);
 626             fjz0             = _mm_add_pd(fjz0,tz);
 627
 628             /**************************
 629              * CALCULATE INTERACTIONS *
 630              **************************/
 631
 632             r30              = _mm_mul_pd(rsq30,rinv30);
 633
 634             /* Compute parameters for interactions between i and j atoms */
 635             qq30             = _mm_mul_pd(iq3,jq0);
 636
 637             /* EWALD ELECTROSTATICS */
 638
 639             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 640             ewrt             = _mm_mul_pd(r30,ewtabscale);
 641             ewitab           = _mm_cvttpd_epi32(ewrt);
 642             eweps            = _mm_sub_pd(ewrt,_mm_cvtepi32_pd(ewitab));
 643             ewitab           = _mm_slli_epi32(ewitab,2);
 644             ewtabF           = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 645             ewtabD           = _mm_setzero_pd();
 646             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 647             ewtabV           = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,0) +2);
 648             ewtabFn          = _mm_setzero_pd();
 649             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 650             felec            = _mm_add_pd(ewtabF,_mm_mul_pd(eweps,ewtabD));
 651             velec            = _mm_sub_pd(ewtabV,_mm_mul_pd(_mm_mul_pd(ewtabhalfspace,eweps),_mm_add_pd(ewtabF,felec)));
 652             velec            = _mm_mul_pd(qq30,_mm_sub_pd(rinv30,velec));
 653             felec            = _mm_mul_pd(_mm_mul_pd(qq30,rinv30),_mm_sub_pd(rinvsq30,felec));
 654
 655             /* Update potential sum for this i atom from the interaction with this j atom. */
 656             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 657             velecsum         = _mm_add_pd(velecsum,velec);
 658
 659             fscal            = felec;
 660
 661             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 662
 663             /* Calculate temporary vectorial force */
 664             tx               = _mm_mul_pd(fscal,dx30);
 665             ty               = _mm_mul_pd(fscal,dy30);
 666             tz               = _mm_mul_pd(fscal,dz30);
 667
 668             /* Update vectorial force */
 669             fix3             = _mm_add_pd(fix3,tx);
 670             fiy3             = _mm_add_pd(fiy3,ty);
 671             fiz3             = _mm_add_pd(fiz3,tz);
 672
 673             fjx0             = _mm_add_pd(fjx0,tx);
 674             fjy0             = _mm_add_pd(fjy0,ty);
 675             fjz0             = _mm_add_pd(fjz0,tz);
 676
 677             gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,fjx0,fjy0,fjz0);
 678
 679             /* Inner loop uses 177 flops */
 680         }
 681
 682         /* End of innermost loop */
 683
 684         gmx_mm_update_iforce_4atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
 685                                               f+i_coord_offset,fshift+i_shift_offset);
 686
 687         ggid                        = gid[iidx];
 688         /* Update potential energies */
 689         gmx_mm_update_1pot_pd(velecsum,kernel_data->energygrp_elec+ggid);
 690         gmx_mm_update_1pot_pd(vvdwsum,kernel_data->energygrp_vdw+ggid);
 691
 692         /* Increment number of inner iterations */
 693         inneriter                  += j_index_end - j_index_start;
 694
 695         /* Outer loop uses 26 flops */
 696     }
 697
 698     /* Increment number of outer iterations */
 699     outeriter        += nri;
 700
 701     /* Update outer/inner flops */
 702
 703     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W4_VF,outeriter*26 + inneriter*177);
 704 }
 705 /*
 706  * Gromacs nonbonded kernel:   nb_kernel_ElecEw_VdwLJEw_GeomW4P1_F_sse2_double
 707  * Electrostatics interaction: Ewald
 708  * VdW interaction:            LJEwald
 709  * Geometry:                   Water4-Particle
 710  * Calculate force/pot:        Force
 711  */
 712 void
 713 nb_kernel_ElecEw_VdwLJEw_GeomW4P1_F_sse2_double
 714                     (t_nblist                    * gmx_restrict       nlist,
 715                      rvec                        * gmx_restrict          xx,
 716                      rvec                        * gmx_restrict          ff,
 717                      t_forcerec                  * gmx_restrict          fr,
 718                      t_mdatoms                   * gmx_restrict     mdatoms,
 719                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 720                      t_nrnb                      * gmx_restrict        nrnb)
 721 {
 722     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 723      * just 0 for non-waters.
 724      * Suffixes A,B refer to j loop unrolling done with SSE double precision, e.g. for the two different
 725      * jnr indices corresponding to data put in the four positions in the SIMD register.
 726      */
 727     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 728     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 729     int              jnrA,jnrB;
 730     int              j_coord_offsetA,j_coord_offsetB;
 731     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 732     real             rcutoff_scalar;
 733     real             *shiftvec,*fshift,*x,*f;
 734     __m128d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 735     int              vdwioffset0;
 736     __m128d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 737     int              vdwioffset1;
 738     __m128d          ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
 739     int              vdwioffset2;
 740     __m128d          ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
 741     int              vdwioffset3;
 742     __m128d          ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
 743     int              vdwjidx0A,vdwjidx0B;
 744     __m128d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 745     __m128d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 746     __m128d          dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
 747     __m128d          dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
 748     __m128d          dx30,dy30,dz30,rsq30,rinv30,rinvsq30,r30,qq30,c6_30,c12_30;
 749     __m128d          velec,felec,velecsum,facel,crf,krf,krf2;
 750     real             *charge;
 751     int              nvdwtype;
 752     __m128d          rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 753     int              *vdwtype;
 754     real             *vdwparam;
 755     __m128d          one_sixth   = _mm_set1_pd(1.0/6.0);
 756     __m128d          one_twelfth = _mm_set1_pd(1.0/12.0);
 757     __m128d           c6grid_00;
 758     __m128d           c6grid_10;
 759     __m128d           c6grid_20;
 760     __m128d           c6grid_30;
 761     __m128d           ewclj,ewclj2,ewclj6,ewcljrsq,poly,exponent,f6A,f6B,sh_lj_ewald;
 762     real             *vdwgridparam;
 763     __m128d           one_half = _mm_set1_pd(0.5);
 764     __m128d           minus_one = _mm_set1_pd(-1.0);
 765     __m128i          ewitab;
 766     __m128d          ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
 767     real             *ewtab;
 768     __m128d          dummy_mask,cutoff_mask;
 769     __m128d          signbit   = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
 770     __m128d          one     = _mm_set1_pd(1.0);
 771     __m128d          two     = _mm_set1_pd(2.0);
 772     x                = xx[0];
 773     f                = ff[0];
 774
 775     nri              = nlist->nri;
 776     iinr             = nlist->iinr;
 777     jindex           = nlist->jindex;
 778     jjnr             = nlist->jjnr;
 779     shiftidx         = nlist->shift;
 780     gid              = nlist->gid;
 781     shiftvec         = fr->shift_vec[0];
 782     fshift           = fr->fshift[0];
 783     facel            = _mm_set1_pd(fr->epsfac);
 784     charge           = mdatoms->chargeA;
 785     nvdwtype         = fr->ntype;
 786     vdwparam         = fr->nbfp;
 787     vdwtype          = mdatoms->typeA;
 788     vdwgridparam     = fr->ljpme_c6grid;
 789     sh_lj_ewald      = _mm_set1_pd(fr->ic->sh_lj_ewald);
 790     ewclj            = _mm_set1_pd(fr->ewaldcoeff_lj);
 791     ewclj2           = _mm_mul_pd(minus_one,_mm_mul_pd(ewclj,ewclj));
 792
 793     sh_ewald         = _mm_set1_pd(fr->ic->sh_ewald);
 794     ewtab            = fr->ic->tabq_coul_F;
 795     ewtabscale       = _mm_set1_pd(fr->ic->tabq_scale);
 796     ewtabhalfspace   = _mm_set1_pd(0.5/fr->ic->tabq_scale);
 797
 798     /* Setup water-specific parameters */
 799     inr              = nlist->iinr[0];
 800     iq1              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+1]));
 801     iq2              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+2]));
 802     iq3              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+3]));
 803     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 804
 805     /* Avoid stupid compiler warnings */
 806     jnrA = jnrB = 0;
 807     j_coord_offsetA = 0;
 808     j_coord_offsetB = 0;
 809
 810     outeriter        = 0;
 811     inneriter        = 0;
 812
 813     /* Start outer loop over neighborlists */
 814     for(iidx=0; iidx<nri; iidx++)
 815     {
 816         /* Load shift vector for this list */
 817         i_shift_offset   = DIM*shiftidx[iidx];
 818
 819         /* Load limits for loop over neighbors */
 820         j_index_start    = jindex[iidx];
 821         j_index_end      = jindex[iidx+1];
 822
 823         /* Get outer coordinate index */
 824         inr              = iinr[iidx];
 825         i_coord_offset   = DIM*inr;
 826
 827         /* Load i particle coords and add shift vector */
 828         gmx_mm_load_shift_and_4rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
 829                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
 830
 831         fix0             = _mm_setzero_pd();
 832         fiy0             = _mm_setzero_pd();
 833         fiz0             = _mm_setzero_pd();
 834         fix1             = _mm_setzero_pd();
 835         fiy1             = _mm_setzero_pd();
 836         fiz1             = _mm_setzero_pd();
 837         fix2             = _mm_setzero_pd();
 838         fiy2             = _mm_setzero_pd();
 839         fiz2             = _mm_setzero_pd();
 840         fix3             = _mm_setzero_pd();
 841         fiy3             = _mm_setzero_pd();
 842         fiz3             = _mm_setzero_pd();
 843
 844         /* Start inner kernel loop */
 845         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 846         {
 847
 848             /* Get j neighbor index, and coordinate index */
 849             jnrA             = jjnr[jidx];
 850             jnrB             = jjnr[jidx+1];
 851             j_coord_offsetA  = DIM*jnrA;
 852             j_coord_offsetB  = DIM*jnrB;
 853
 854             /* load j atom coordinates */
 855             gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 856                                               &jx0,&jy0,&jz0);
 857
 858             /* Calculate displacement vector */
 859             dx00             = _mm_sub_pd(ix0,jx0);
 860             dy00             = _mm_sub_pd(iy0,jy0);
 861             dz00             = _mm_sub_pd(iz0,jz0);
 862             dx10             = _mm_sub_pd(ix1,jx0);
 863             dy10             = _mm_sub_pd(iy1,jy0);
 864             dz10             = _mm_sub_pd(iz1,jz0);
 865             dx20             = _mm_sub_pd(ix2,jx0);
 866             dy20             = _mm_sub_pd(iy2,jy0);
 867             dz20             = _mm_sub_pd(iz2,jz0);
 868             dx30             = _mm_sub_pd(ix3,jx0);
 869             dy30             = _mm_sub_pd(iy3,jy0);
 870             dz30             = _mm_sub_pd(iz3,jz0);
 871
 872             /* Calculate squared distance and things based on it */
 873             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 874             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
 875             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
 876             rsq30            = gmx_mm_calc_rsq_pd(dx30,dy30,dz30);
 877
 878             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 879             rinv10           = gmx_mm_invsqrt_pd(rsq10);
 880             rinv20           = gmx_mm_invsqrt_pd(rsq20);
 881             rinv30           = gmx_mm_invsqrt_pd(rsq30);
 882
 883             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 884             rinvsq10         = _mm_mul_pd(rinv10,rinv10);
 885             rinvsq20         = _mm_mul_pd(rinv20,rinv20);
 886             rinvsq30         = _mm_mul_pd(rinv30,rinv30);
 887
 888             /* Load parameters for j particles */
 889             jq0              = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
 890             vdwjidx0A        = 2*vdwtype[jnrA+0];
 891             vdwjidx0B        = 2*vdwtype[jnrB+0];
 892
 893             fjx0             = _mm_setzero_pd();
 894             fjy0             = _mm_setzero_pd();
 895             fjz0             = _mm_setzero_pd();
 896
 897             /**************************
 898              * CALCULATE INTERACTIONS *
 899              **************************/
 900
 901             r00              = _mm_mul_pd(rsq00,rinv00);
 902
 903             /* Compute parameters for interactions between i and j atoms */
 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             /* Analytical LJ-PME */
 911             rinvsix          = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 912             ewcljrsq         = _mm_mul_pd(ewclj2,rsq00);
 913             ewclj6           = _mm_mul_pd(ewclj2,_mm_mul_pd(ewclj2,ewclj2));
 914             exponent         = gmx_simd_exp_d(ewcljrsq);
 915             /* poly = exp(-(beta*r)^2) * (1 + (beta*r)^2 + (beta*r)^4 /2) */
 916             poly             = _mm_mul_pd(exponent,_mm_add_pd(_mm_sub_pd(one,ewcljrsq),_mm_mul_pd(_mm_mul_pd(ewcljrsq,ewcljrsq),one_half)));
 917             /* f6A = 6 * C6grid * (1 - poly) */
 918             f6A              = _mm_mul_pd(c6grid_00,_mm_sub_pd(one,poly));
 919             /* f6B = C6grid * exponent * beta^6 */
 920             f6B              = _mm_mul_pd(_mm_mul_pd(c6grid_00,one_sixth),_mm_mul_pd(exponent,ewclj6));
 921             /* fvdw = 12*C12/r13 - ((6*C6 - f6A)/r6 + f6B)/r */
 922             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);
 923
 924             fscal            = fvdw;
 925
 926             /* Calculate temporary vectorial force */
 927             tx               = _mm_mul_pd(fscal,dx00);
 928             ty               = _mm_mul_pd(fscal,dy00);
 929             tz               = _mm_mul_pd(fscal,dz00);
 930
 931             /* Update vectorial force */
 932             fix0             = _mm_add_pd(fix0,tx);
 933             fiy0             = _mm_add_pd(fiy0,ty);
 934             fiz0             = _mm_add_pd(fiz0,tz);
 935
 936             fjx0             = _mm_add_pd(fjx0,tx);
 937             fjy0             = _mm_add_pd(fjy0,ty);
 938             fjz0             = _mm_add_pd(fjz0,tz);
 939
 940             /**************************
 941              * CALCULATE INTERACTIONS *
 942              **************************/
 943
 944             r10              = _mm_mul_pd(rsq10,rinv10);
 945
 946             /* Compute parameters for interactions between i and j atoms */
 947             qq10             = _mm_mul_pd(iq1,jq0);
 948
 949             /* EWALD ELECTROSTATICS */
 950
 951             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 952             ewrt             = _mm_mul_pd(r10,ewtabscale);
 953             ewitab           = _mm_cvttpd_epi32(ewrt);
 954             eweps            = _mm_sub_pd(ewrt,_mm_cvtepi32_pd(ewitab));
 955             gmx_mm_load_2pair_swizzle_pd(ewtab+gmx_mm_extract_epi32(ewitab,0),ewtab+gmx_mm_extract_epi32(ewitab,1),
 956                                          &ewtabF,&ewtabFn);
 957             felec            = _mm_add_pd(_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF),_mm_mul_pd(eweps,ewtabFn));
 958             felec            = _mm_mul_pd(_mm_mul_pd(qq10,rinv10),_mm_sub_pd(rinvsq10,felec));
 959
 960             fscal            = felec;
 961
 962             /* Calculate temporary vectorial force */
 963             tx               = _mm_mul_pd(fscal,dx10);
 964             ty               = _mm_mul_pd(fscal,dy10);
 965             tz               = _mm_mul_pd(fscal,dz10);
 966
 967             /* Update vectorial force */
 968             fix1             = _mm_add_pd(fix1,tx);
 969             fiy1             = _mm_add_pd(fiy1,ty);
 970             fiz1             = _mm_add_pd(fiz1,tz);
 971
 972             fjx0             = _mm_add_pd(fjx0,tx);
 973             fjy0             = _mm_add_pd(fjy0,ty);
 974             fjz0             = _mm_add_pd(fjz0,tz);
 975
 976             /**************************
 977              * CALCULATE INTERACTIONS *
 978              **************************/
 979
 980             r20              = _mm_mul_pd(rsq20,rinv20);
 981
 982             /* Compute parameters for interactions between i and j atoms */
 983             qq20             = _mm_mul_pd(iq2,jq0);
 984
 985             /* EWALD ELECTROSTATICS */
 986
 987             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 988             ewrt             = _mm_mul_pd(r20,ewtabscale);
 989             ewitab           = _mm_cvttpd_epi32(ewrt);
 990             eweps            = _mm_sub_pd(ewrt,_mm_cvtepi32_pd(ewitab));
 991             gmx_mm_load_2pair_swizzle_pd(ewtab+gmx_mm_extract_epi32(ewitab,0),ewtab+gmx_mm_extract_epi32(ewitab,1),
 992                                          &ewtabF,&ewtabFn);
 993             felec            = _mm_add_pd(_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF),_mm_mul_pd(eweps,ewtabFn));
 994             felec            = _mm_mul_pd(_mm_mul_pd(qq20,rinv20),_mm_sub_pd(rinvsq20,felec));
 995
 996             fscal            = felec;
 997
 998             /* Calculate temporary vectorial force */
 999             tx               = _mm_mul_pd(fscal,dx20);
1000             ty               = _mm_mul_pd(fscal,dy20);
1001             tz               = _mm_mul_pd(fscal,dz20);
1002
1003             /* Update vectorial force */
1004             fix2             = _mm_add_pd(fix2,tx);
1005             fiy2             = _mm_add_pd(fiy2,ty);
1006             fiz2             = _mm_add_pd(fiz2,tz);
1007
1008             fjx0             = _mm_add_pd(fjx0,tx);
1009             fjy0             = _mm_add_pd(fjy0,ty);
1010             fjz0             = _mm_add_pd(fjz0,tz);
1011
1012             /**************************
1013              * CALCULATE INTERACTIONS *
1014              **************************/
1015
1016             r30              = _mm_mul_pd(rsq30,rinv30);
1017
1018             /* Compute parameters for interactions between i and j atoms */
1019             qq30             = _mm_mul_pd(iq3,jq0);
1020
1021             /* EWALD ELECTROSTATICS */
1022
1023             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1024             ewrt             = _mm_mul_pd(r30,ewtabscale);
1025             ewitab           = _mm_cvttpd_epi32(ewrt);
1026             eweps            = _mm_sub_pd(ewrt,_mm_cvtepi32_pd(ewitab));
1027             gmx_mm_load_2pair_swizzle_pd(ewtab+gmx_mm_extract_epi32(ewitab,0),ewtab+gmx_mm_extract_epi32(ewitab,1),
1028                                          &ewtabF,&ewtabFn);
1029             felec            = _mm_add_pd(_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF),_mm_mul_pd(eweps,ewtabFn));
1030             felec            = _mm_mul_pd(_mm_mul_pd(qq30,rinv30),_mm_sub_pd(rinvsq30,felec));
1031
1032             fscal            = felec;
1033
1034             /* Calculate temporary vectorial force */
1035             tx               = _mm_mul_pd(fscal,dx30);
1036             ty               = _mm_mul_pd(fscal,dy30);
1037             tz               = _mm_mul_pd(fscal,dz30);
1038
1039             /* Update vectorial force */
1040             fix3             = _mm_add_pd(fix3,tx);
1041             fiy3             = _mm_add_pd(fiy3,ty);
1042             fiz3             = _mm_add_pd(fiz3,tz);
1043
1044             fjx0             = _mm_add_pd(fjx0,tx);
1045             fjy0             = _mm_add_pd(fjy0,ty);
1046             fjz0             = _mm_add_pd(fjz0,tz);
1047
1048             gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,fjx0,fjy0,fjz0);
1049
1050             /* Inner loop uses 157 flops */
1051         }
1052
1053         if(jidx<j_index_end)
1054         {
1055
1056             jnrA             = jjnr[jidx];
1057             j_coord_offsetA  = DIM*jnrA;
1058
1059             /* load j atom coordinates */
1060             gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
1061                                               &jx0,&jy0,&jz0);
1062
1063             /* Calculate displacement vector */
1064             dx00             = _mm_sub_pd(ix0,jx0);
1065             dy00             = _mm_sub_pd(iy0,jy0);
1066             dz00             = _mm_sub_pd(iz0,jz0);
1067             dx10             = _mm_sub_pd(ix1,jx0);
1068             dy10             = _mm_sub_pd(iy1,jy0);
1069             dz10             = _mm_sub_pd(iz1,jz0);
1070             dx20             = _mm_sub_pd(ix2,jx0);
1071             dy20             = _mm_sub_pd(iy2,jy0);
1072             dz20             = _mm_sub_pd(iz2,jz0);
1073             dx30             = _mm_sub_pd(ix3,jx0);
1074             dy30             = _mm_sub_pd(iy3,jy0);
1075             dz30             = _mm_sub_pd(iz3,jz0);
1076
1077             /* Calculate squared distance and things based on it */
1078             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
1079             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
1080             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
1081             rsq30            = gmx_mm_calc_rsq_pd(dx30,dy30,dz30);
1082
1083             rinv00           = gmx_mm_invsqrt_pd(rsq00);
1084             rinv10           = gmx_mm_invsqrt_pd(rsq10);
1085             rinv20           = gmx_mm_invsqrt_pd(rsq20);
1086             rinv30           = gmx_mm_invsqrt_pd(rsq30);
1087
1088             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
1089             rinvsq10         = _mm_mul_pd(rinv10,rinv10);
1090             rinvsq20         = _mm_mul_pd(rinv20,rinv20);
1091             rinvsq30         = _mm_mul_pd(rinv30,rinv30);
1092
1093             /* Load parameters for j particles */
1094             jq0              = _mm_load_sd(charge+jnrA+0);
1095             vdwjidx0A        = 2*vdwtype[jnrA+0];
1096
1097             fjx0             = _mm_setzero_pd();
1098             fjy0             = _mm_setzero_pd();
1099             fjz0             = _mm_setzero_pd();
1100
1101             /**************************
1102              * CALCULATE INTERACTIONS *
1103              **************************/
1104
1105             r00              = _mm_mul_pd(rsq00,rinv00);
1106
1107             /* Compute parameters for interactions between i and j atoms */
1108             gmx_mm_load_1pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,&c6_00,&c12_00);
1109
1110             c6grid_00       = gmx_mm_load_1real_pd(vdwgridparam+vdwioffset0+vdwjidx0A);
1111
1112             /* Analytical LJ-PME */
1113             rinvsix          = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
1114             ewcljrsq         = _mm_mul_pd(ewclj2,rsq00);
1115             ewclj6           = _mm_mul_pd(ewclj2,_mm_mul_pd(ewclj2,ewclj2));
1116             exponent         = gmx_simd_exp_d(ewcljrsq);
1117             /* poly = exp(-(beta*r)^2) * (1 + (beta*r)^2 + (beta*r)^4 /2) */
1118             poly             = _mm_mul_pd(exponent,_mm_add_pd(_mm_sub_pd(one,ewcljrsq),_mm_mul_pd(_mm_mul_pd(ewcljrsq,ewcljrsq),one_half)));
1119             /* f6A = 6 * C6grid * (1 - poly) */
1120             f6A              = _mm_mul_pd(c6grid_00,_mm_sub_pd(one,poly));
1121             /* f6B = C6grid * exponent * beta^6 */
1122             f6B              = _mm_mul_pd(_mm_mul_pd(c6grid_00,one_sixth),_mm_mul_pd(exponent,ewclj6));
1123             /* fvdw = 12*C12/r13 - ((6*C6 - f6A)/r6 + f6B)/r */
1124             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);
1125
1126             fscal            = fvdw;
1127
1128             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1129
1130             /* Calculate temporary vectorial force */
1131             tx               = _mm_mul_pd(fscal,dx00);
1132             ty               = _mm_mul_pd(fscal,dy00);
1133             tz               = _mm_mul_pd(fscal,dz00);
1134
1135             /* Update vectorial force */
1136             fix0             = _mm_add_pd(fix0,tx);
1137             fiy0             = _mm_add_pd(fiy0,ty);
1138             fiz0             = _mm_add_pd(fiz0,tz);
1139
1140             fjx0             = _mm_add_pd(fjx0,tx);
1141             fjy0             = _mm_add_pd(fjy0,ty);
1142             fjz0             = _mm_add_pd(fjz0,tz);
1143
1144             /**************************
1145              * CALCULATE INTERACTIONS *
1146              **************************/
1147
1148             r10              = _mm_mul_pd(rsq10,rinv10);
1149
1150             /* Compute parameters for interactions between i and j atoms */
1151             qq10             = _mm_mul_pd(iq1,jq0);
1152
1153             /* EWALD ELECTROSTATICS */
1154
1155             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1156             ewrt             = _mm_mul_pd(r10,ewtabscale);
1157             ewitab           = _mm_cvttpd_epi32(ewrt);
1158             eweps            = _mm_sub_pd(ewrt,_mm_cvtepi32_pd(ewitab));
1159             gmx_mm_load_1pair_swizzle_pd(ewtab+gmx_mm_extract_epi32(ewitab,0),&ewtabF,&ewtabFn);
1160             felec            = _mm_add_pd(_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF),_mm_mul_pd(eweps,ewtabFn));
1161             felec            = _mm_mul_pd(_mm_mul_pd(qq10,rinv10),_mm_sub_pd(rinvsq10,felec));
1162
1163             fscal            = felec;
1164
1165             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1166
1167             /* Calculate temporary vectorial force */
1168             tx               = _mm_mul_pd(fscal,dx10);
1169             ty               = _mm_mul_pd(fscal,dy10);
1170             tz               = _mm_mul_pd(fscal,dz10);
1171
1172             /* Update vectorial force */
1173             fix1             = _mm_add_pd(fix1,tx);
1174             fiy1             = _mm_add_pd(fiy1,ty);
1175             fiz1             = _mm_add_pd(fiz1,tz);
1176
1177             fjx0             = _mm_add_pd(fjx0,tx);
1178             fjy0             = _mm_add_pd(fjy0,ty);
1179             fjz0             = _mm_add_pd(fjz0,tz);
1180
1181             /**************************
1182              * CALCULATE INTERACTIONS *
1183              **************************/
1184
1185             r20              = _mm_mul_pd(rsq20,rinv20);
1186
1187             /* Compute parameters for interactions between i and j atoms */
1188             qq20             = _mm_mul_pd(iq2,jq0);
1189
1190             /* EWALD ELECTROSTATICS */
1191
1192             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1193             ewrt             = _mm_mul_pd(r20,ewtabscale);
1194             ewitab           = _mm_cvttpd_epi32(ewrt);
1195             eweps            = _mm_sub_pd(ewrt,_mm_cvtepi32_pd(ewitab));
1196             gmx_mm_load_1pair_swizzle_pd(ewtab+gmx_mm_extract_epi32(ewitab,0),&ewtabF,&ewtabFn);
1197             felec            = _mm_add_pd(_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF),_mm_mul_pd(eweps,ewtabFn));
1198             felec            = _mm_mul_pd(_mm_mul_pd(qq20,rinv20),_mm_sub_pd(rinvsq20,felec));
1199
1200             fscal            = felec;
1201
1202             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1203
1204             /* Calculate temporary vectorial force */
1205             tx               = _mm_mul_pd(fscal,dx20);
1206             ty               = _mm_mul_pd(fscal,dy20);
1207             tz               = _mm_mul_pd(fscal,dz20);
1208
1209             /* Update vectorial force */
1210             fix2             = _mm_add_pd(fix2,tx);
1211             fiy2             = _mm_add_pd(fiy2,ty);
1212             fiz2             = _mm_add_pd(fiz2,tz);
1213
1214             fjx0             = _mm_add_pd(fjx0,tx);
1215             fjy0             = _mm_add_pd(fjy0,ty);
1216             fjz0             = _mm_add_pd(fjz0,tz);
1217
1218             /**************************
1219              * CALCULATE INTERACTIONS *
1220              **************************/
1221
1222             r30              = _mm_mul_pd(rsq30,rinv30);
1223
1224             /* Compute parameters for interactions between i and j atoms */
1225             qq30             = _mm_mul_pd(iq3,jq0);
1226
1227             /* EWALD ELECTROSTATICS */
1228
1229             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1230             ewrt             = _mm_mul_pd(r30,ewtabscale);
1231             ewitab           = _mm_cvttpd_epi32(ewrt);
1232             eweps            = _mm_sub_pd(ewrt,_mm_cvtepi32_pd(ewitab));
1233             gmx_mm_load_1pair_swizzle_pd(ewtab+gmx_mm_extract_epi32(ewitab,0),&ewtabF,&ewtabFn);
1234             felec            = _mm_add_pd(_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF),_mm_mul_pd(eweps,ewtabFn));
1235             felec            = _mm_mul_pd(_mm_mul_pd(qq30,rinv30),_mm_sub_pd(rinvsq30,felec));
1236
1237             fscal            = felec;
1238
1239             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1240
1241             /* Calculate temporary vectorial force */
1242             tx               = _mm_mul_pd(fscal,dx30);
1243             ty               = _mm_mul_pd(fscal,dy30);
1244             tz               = _mm_mul_pd(fscal,dz30);
1245
1246             /* Update vectorial force */
1247             fix3             = _mm_add_pd(fix3,tx);
1248             fiy3             = _mm_add_pd(fiy3,ty);
1249             fiz3             = _mm_add_pd(fiz3,tz);
1250
1251             fjx0             = _mm_add_pd(fjx0,tx);
1252             fjy0             = _mm_add_pd(fjy0,ty);
1253             fjz0             = _mm_add_pd(fjz0,tz);
1254
1255             gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,fjx0,fjy0,fjz0);
1256
1257             /* Inner loop uses 157 flops */
1258         }
1259
1260         /* End of innermost loop */
1261
1262         gmx_mm_update_iforce_4atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
1263                                               f+i_coord_offset,fshift+i_shift_offset);
1264
1265         /* Increment number of inner iterations */
1266         inneriter                  += j_index_end - j_index_start;
1267
1268         /* Outer loop uses 24 flops */
1269     }
1270
1271     /* Increment number of outer iterations */
1272     outeriter        += nri;
1273
1274     /* Update outer/inner flops */
1275
1276     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W4_F,outeriter*24 + inneriter*157);
1277 }