src/gromacs/gmxlib/nonbonded/nb_kernel_sse2_double/nb_kernel_ElecEw_VdwLJEw_GeomW3P1_sse2_double.c

   1 /*
   2  * This file is part of the GROMACS molecular simulation package.
   3  *
   4  * Copyright (c) 2012,2013,2014, by the GROMACS development team, led by
   5  * Mark Abraham, David van der Spoel, Berk Hess, and Erik Lindahl,
   6  * and including many others, as listed in the AUTHORS file in the
   7  * top-level source directory and at http://www.gromacs.org.
   8  *
   9  * GROMACS is free software; you can redistribute it and/or
  10  * modify it under the terms of the GNU Lesser General Public License
  11  * as published by the Free Software Foundation; either version 2.1
  12  * of the License, or (at your option) any later version.
  13  *
  14  * GROMACS is distributed in the hope that it will be useful,
  15  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  16  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  17  * Lesser General Public License for more details.
  18  *
  19  * You should have received a copy of the GNU Lesser General Public
  20  * License along with GROMACS; if not, see
  21  * http://www.gnu.org/licenses, or write to the Free Software Foundation,
  22  * Inc., 51 Franklin Street, Fifth Floor, Boston, MA  02110-1301  USA.
  23  *
  24  * If you want to redistribute modifications to GROMACS, please
  25  * consider that scientific software is very special. Version
  26  * control is crucial - bugs must be traceable. We will be happy to
  27  * consider code for inclusion in the official distribution, but
  28  * derived work must not be called official GROMACS. Details are found
  29  * in the README & COPYING files - if they are missing, get the
  30  * official version at http://www.gromacs.org.
  31  *
  32  * To help us fund GROMACS development, we humbly ask that you cite
  33  * the research papers on the package. Check out http://www.gromacs.org.
  34  */
  35 /*
  36  * Note: this file was generated by the GROMACS sse2_double kernel generator.
  37  */
  38 #include "gmxpre.h"
  39
  40 #include "config.h"
  41
  42 #include <math.h>
  43
  44 #include "../nb_kernel.h"
  45 #include "gromacs/legacyheaders/types/simple.h"
  46 #include "gromacs/math/vec.h"
  47 #include "gromacs/legacyheaders/nrnb.h"
  48
  49 #include "gromacs/simd/math_x86_sse2_double.h"
  50 #include "kernelutil_x86_sse2_double.h"
  51
  52 /*
  53  * Gromacs nonbonded kernel:   nb_kernel_ElecEw_VdwLJEw_GeomW3P1_VF_sse2_double
  54  * Electrostatics interaction: Ewald
  55  * VdW interaction:            LJEwald
  56  * Geometry:                   Water3-Particle
  57  * Calculate force/pot:        PotentialAndForce
  58  */
  59 void
  60 nb_kernel_ElecEw_VdwLJEw_GeomW3P1_VF_sse2_double
  61                     (t_nblist                    * gmx_restrict       nlist,
  62                      rvec                        * gmx_restrict          xx,
  63                      rvec                        * gmx_restrict          ff,
  64                      t_forcerec                  * gmx_restrict          fr,
  65                      t_mdatoms                   * gmx_restrict     mdatoms,
  66                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
  67                      t_nrnb                      * gmx_restrict        nrnb)
  68 {
  69     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
  70      * just 0 for non-waters.
  71      * Suffixes A,B refer to j loop unrolling done with SSE double precision, e.g. for the two different
  72      * jnr indices corresponding to data put in the four positions in the SIMD register.
  73      */
  74     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
  75     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
  76     int              jnrA,jnrB;
  77     int              j_coord_offsetA,j_coord_offsetB;
  78     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
  79     real             rcutoff_scalar;
  80     real             *shiftvec,*fshift,*x,*f;
  81     __m128d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
  82     int              vdwioffset0;
  83     __m128d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
  84     int              vdwioffset1;
  85     __m128d          ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
  86     int              vdwioffset2;
  87     __m128d          ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
  88     int              vdwjidx0A,vdwjidx0B;
  89     __m128d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  90     __m128d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  91     __m128d          dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
  92     __m128d          dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
  93     __m128d          velec,felec,velecsum,facel,crf,krf,krf2;
  94     real             *charge;
  95     int              nvdwtype;
  96     __m128d          rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
  97     int              *vdwtype;
  98     real             *vdwparam;
  99     __m128d          one_sixth   = _mm_set1_pd(1.0/6.0);
 100     __m128d          one_twelfth = _mm_set1_pd(1.0/12.0);
 101     __m128d           c6grid_00;
 102     __m128d           c6grid_10;
 103     __m128d           c6grid_20;
 104     __m128d           ewclj,ewclj2,ewclj6,ewcljrsq,poly,exponent,f6A,f6B,sh_lj_ewald;
 105     real             *vdwgridparam;
 106     __m128d           one_half = _mm_set1_pd(0.5);
 107     __m128d           minus_one = _mm_set1_pd(-1.0);
 108     __m128i          ewitab;
 109     __m128d          ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
 110     real             *ewtab;
 111     __m128d          dummy_mask,cutoff_mask;
 112     __m128d          signbit   = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
 113     __m128d          one     = _mm_set1_pd(1.0);
 114     __m128d          two     = _mm_set1_pd(2.0);
 115     x                = xx[0];
 116     f                = ff[0];
 117
 118     nri              = nlist->nri;
 119     iinr             = nlist->iinr;
 120     jindex           = nlist->jindex;
 121     jjnr             = nlist->jjnr;
 122     shiftidx         = nlist->shift;
 123     gid              = nlist->gid;
 124     shiftvec         = fr->shift_vec[0];
 125     fshift           = fr->fshift[0];
 126     facel            = _mm_set1_pd(fr->epsfac);
 127     charge           = mdatoms->chargeA;
 128     nvdwtype         = fr->ntype;
 129     vdwparam         = fr->nbfp;
 130     vdwtype          = mdatoms->typeA;
 131     vdwgridparam     = fr->ljpme_c6grid;
 132     sh_lj_ewald      = _mm_set1_pd(fr->ic->sh_lj_ewald);
 133     ewclj            = _mm_set1_pd(fr->ewaldcoeff_lj);
 134     ewclj2           = _mm_mul_pd(minus_one,_mm_mul_pd(ewclj,ewclj));
 135
 136     sh_ewald         = _mm_set1_pd(fr->ic->sh_ewald);
 137     ewtab            = fr->ic->tabq_coul_FDV0;
 138     ewtabscale       = _mm_set1_pd(fr->ic->tabq_scale);
 139     ewtabhalfspace   = _mm_set1_pd(0.5/fr->ic->tabq_scale);
 140
 141     /* Setup water-specific parameters */
 142     inr              = nlist->iinr[0];
 143     iq0              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+0]));
 144     iq1              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+1]));
 145     iq2              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+2]));
 146     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 147
 148     /* Avoid stupid compiler warnings */
 149     jnrA = jnrB = 0;
 150     j_coord_offsetA = 0;
 151     j_coord_offsetB = 0;
 152
 153     outeriter        = 0;
 154     inneriter        = 0;
 155
 156     /* Start outer loop over neighborlists */
 157     for(iidx=0; iidx<nri; iidx++)
 158     {
 159         /* Load shift vector for this list */
 160         i_shift_offset   = DIM*shiftidx[iidx];
 161
 162         /* Load limits for loop over neighbors */
 163         j_index_start    = jindex[iidx];
 164         j_index_end      = jindex[iidx+1];
 165
 166         /* Get outer coordinate index */
 167         inr              = iinr[iidx];
 168         i_coord_offset   = DIM*inr;
 169
 170         /* Load i particle coords and add shift vector */
 171         gmx_mm_load_shift_and_3rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
 172                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 173
 174         fix0             = _mm_setzero_pd();
 175         fiy0             = _mm_setzero_pd();
 176         fiz0             = _mm_setzero_pd();
 177         fix1             = _mm_setzero_pd();
 178         fiy1             = _mm_setzero_pd();
 179         fiz1             = _mm_setzero_pd();
 180         fix2             = _mm_setzero_pd();
 181         fiy2             = _mm_setzero_pd();
 182         fiz2             = _mm_setzero_pd();
 183
 184         /* Reset potential sums */
 185         velecsum         = _mm_setzero_pd();
 186         vvdwsum          = _mm_setzero_pd();
 187
 188         /* Start inner kernel loop */
 189         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 190         {
 191
 192             /* Get j neighbor index, and coordinate index */
 193             jnrA             = jjnr[jidx];
 194             jnrB             = jjnr[jidx+1];
 195             j_coord_offsetA  = DIM*jnrA;
 196             j_coord_offsetB  = DIM*jnrB;
 197
 198             /* load j atom coordinates */
 199             gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 200                                               &jx0,&jy0,&jz0);
 201
 202             /* Calculate displacement vector */
 203             dx00             = _mm_sub_pd(ix0,jx0);
 204             dy00             = _mm_sub_pd(iy0,jy0);
 205             dz00             = _mm_sub_pd(iz0,jz0);
 206             dx10             = _mm_sub_pd(ix1,jx0);
 207             dy10             = _mm_sub_pd(iy1,jy0);
 208             dz10             = _mm_sub_pd(iz1,jz0);
 209             dx20             = _mm_sub_pd(ix2,jx0);
 210             dy20             = _mm_sub_pd(iy2,jy0);
 211             dz20             = _mm_sub_pd(iz2,jz0);
 212
 213             /* Calculate squared distance and things based on it */
 214             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 215             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
 216             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
 217
 218             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 219             rinv10           = gmx_mm_invsqrt_pd(rsq10);
 220             rinv20           = gmx_mm_invsqrt_pd(rsq20);
 221
 222             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 223             rinvsq10         = _mm_mul_pd(rinv10,rinv10);
 224             rinvsq20         = _mm_mul_pd(rinv20,rinv20);
 225
 226             /* Load parameters for j particles */
 227             jq0              = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
 228             vdwjidx0A        = 2*vdwtype[jnrA+0];
 229             vdwjidx0B        = 2*vdwtype[jnrB+0];
 230
 231             fjx0             = _mm_setzero_pd();
 232             fjy0             = _mm_setzero_pd();
 233             fjz0             = _mm_setzero_pd();
 234
 235             /**************************
 236              * CALCULATE INTERACTIONS *
 237              **************************/
 238
 239             r00              = _mm_mul_pd(rsq00,rinv00);
 240
 241             /* Compute parameters for interactions between i and j atoms */
 242             qq00             = _mm_mul_pd(iq0,jq0);
 243             gmx_mm_load_2pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,
 244                                          vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
 245
 246             c6grid_00       = gmx_mm_load_2real_swizzle_pd(vdwgridparam+vdwioffset0+vdwjidx0A,
 247                                                                vdwgridparam+vdwioffset0+vdwjidx0B);
 248
 249             /* EWALD ELECTROSTATICS */
 250
 251             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 252             ewrt             = _mm_mul_pd(r00,ewtabscale);
 253             ewitab           = _mm_cvttpd_epi32(ewrt);
 254             eweps            = _mm_sub_pd(ewrt,_mm_cvtepi32_pd(ewitab));
 255             ewitab           = _mm_slli_epi32(ewitab,2);
 256             ewtabF           = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 257             ewtabD           = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,1) );
 258             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 259             ewtabV           = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,0) +2);
 260             ewtabFn          = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,1) +2);
 261             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 262             felec            = _mm_add_pd(ewtabF,_mm_mul_pd(eweps,ewtabD));
 263             velec            = _mm_sub_pd(ewtabV,_mm_mul_pd(_mm_mul_pd(ewtabhalfspace,eweps),_mm_add_pd(ewtabF,felec)));
 264             velec            = _mm_mul_pd(qq00,_mm_sub_pd(rinv00,velec));
 265             felec            = _mm_mul_pd(_mm_mul_pd(qq00,rinv00),_mm_sub_pd(rinvsq00,felec));
 266
 267             /* Analytical LJ-PME */
 268             rinvsix          = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 269             ewcljrsq         = _mm_mul_pd(ewclj2,rsq00);
 270             ewclj6           = _mm_mul_pd(ewclj2,_mm_mul_pd(ewclj2,ewclj2));
 271             exponent         = gmx_simd_exp_d(ewcljrsq);
 272             /* poly = exp(-(beta*r)^2) * (1 + (beta*r)^2 + (beta*r)^4 /2) */
 273             poly             = _mm_mul_pd(exponent,_mm_add_pd(_mm_sub_pd(one,ewcljrsq),_mm_mul_pd(_mm_mul_pd(ewcljrsq,ewcljrsq),one_half)));
 274             /* vvdw6 = [C6 - C6grid * (1-poly)]/r6 */
 275             vvdw6            = _mm_mul_pd(_mm_sub_pd(c6_00,_mm_mul_pd(c6grid_00,_mm_sub_pd(one,poly))),rinvsix);
 276             vvdw12           = _mm_mul_pd(c12_00,_mm_mul_pd(rinvsix,rinvsix));
 277             vvdw             = _mm_sub_pd(_mm_mul_pd(vvdw12,one_twelfth),_mm_mul_pd(vvdw6,one_sixth));
 278             /* fvdw = vvdw12/r - (vvdw6/r + (C6grid * exponent * beta^6)/r) */
 279             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);
 280
 281             /* Update potential sum for this i atom from the interaction with this j atom. */
 282             velecsum         = _mm_add_pd(velecsum,velec);
 283             vvdwsum          = _mm_add_pd(vvdwsum,vvdw);
 284
 285             fscal            = _mm_add_pd(felec,fvdw);
 286
 287             /* Calculate temporary vectorial force */
 288             tx               = _mm_mul_pd(fscal,dx00);
 289             ty               = _mm_mul_pd(fscal,dy00);
 290             tz               = _mm_mul_pd(fscal,dz00);
 291
 292             /* Update vectorial force */
 293             fix0             = _mm_add_pd(fix0,tx);
 294             fiy0             = _mm_add_pd(fiy0,ty);
 295             fiz0             = _mm_add_pd(fiz0,tz);
 296
 297             fjx0             = _mm_add_pd(fjx0,tx);
 298             fjy0             = _mm_add_pd(fjy0,ty);
 299             fjz0             = _mm_add_pd(fjz0,tz);
 300
 301             /**************************
 302              * CALCULATE INTERACTIONS *
 303              **************************/
 304
 305             r10              = _mm_mul_pd(rsq10,rinv10);
 306
 307             /* Compute parameters for interactions between i and j atoms */
 308             qq10             = _mm_mul_pd(iq1,jq0);
 309
 310             /* EWALD ELECTROSTATICS */
 311
 312             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 313             ewrt             = _mm_mul_pd(r10,ewtabscale);
 314             ewitab           = _mm_cvttpd_epi32(ewrt);
 315             eweps            = _mm_sub_pd(ewrt,_mm_cvtepi32_pd(ewitab));
 316             ewitab           = _mm_slli_epi32(ewitab,2);
 317             ewtabF           = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 318             ewtabD           = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,1) );
 319             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 320             ewtabV           = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,0) +2);
 321             ewtabFn          = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,1) +2);
 322             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 323             felec            = _mm_add_pd(ewtabF,_mm_mul_pd(eweps,ewtabD));
 324             velec            = _mm_sub_pd(ewtabV,_mm_mul_pd(_mm_mul_pd(ewtabhalfspace,eweps),_mm_add_pd(ewtabF,felec)));
 325             velec            = _mm_mul_pd(qq10,_mm_sub_pd(rinv10,velec));
 326             felec            = _mm_mul_pd(_mm_mul_pd(qq10,rinv10),_mm_sub_pd(rinvsq10,felec));
 327
 328             /* Update potential sum for this i atom from the interaction with this j atom. */
 329             velecsum         = _mm_add_pd(velecsum,velec);
 330
 331             fscal            = felec;
 332
 333             /* Calculate temporary vectorial force */
 334             tx               = _mm_mul_pd(fscal,dx10);
 335             ty               = _mm_mul_pd(fscal,dy10);
 336             tz               = _mm_mul_pd(fscal,dz10);
 337
 338             /* Update vectorial force */
 339             fix1             = _mm_add_pd(fix1,tx);
 340             fiy1             = _mm_add_pd(fiy1,ty);
 341             fiz1             = _mm_add_pd(fiz1,tz);
 342
 343             fjx0             = _mm_add_pd(fjx0,tx);
 344             fjy0             = _mm_add_pd(fjy0,ty);
 345             fjz0             = _mm_add_pd(fjz0,tz);
 346
 347             /**************************
 348              * CALCULATE INTERACTIONS *
 349              **************************/
 350
 351             r20              = _mm_mul_pd(rsq20,rinv20);
 352
 353             /* Compute parameters for interactions between i and j atoms */
 354             qq20             = _mm_mul_pd(iq2,jq0);
 355
 356             /* EWALD ELECTROSTATICS */
 357
 358             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 359             ewrt             = _mm_mul_pd(r20,ewtabscale);
 360             ewitab           = _mm_cvttpd_epi32(ewrt);
 361             eweps            = _mm_sub_pd(ewrt,_mm_cvtepi32_pd(ewitab));
 362             ewitab           = _mm_slli_epi32(ewitab,2);
 363             ewtabF           = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 364             ewtabD           = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,1) );
 365             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 366             ewtabV           = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,0) +2);
 367             ewtabFn          = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,1) +2);
 368             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 369             felec            = _mm_add_pd(ewtabF,_mm_mul_pd(eweps,ewtabD));
 370             velec            = _mm_sub_pd(ewtabV,_mm_mul_pd(_mm_mul_pd(ewtabhalfspace,eweps),_mm_add_pd(ewtabF,felec)));
 371             velec            = _mm_mul_pd(qq20,_mm_sub_pd(rinv20,velec));
 372             felec            = _mm_mul_pd(_mm_mul_pd(qq20,rinv20),_mm_sub_pd(rinvsq20,felec));
 373
 374             /* Update potential sum for this i atom from the interaction with this j atom. */
 375             velecsum         = _mm_add_pd(velecsum,velec);
 376
 377             fscal            = felec;
 378
 379             /* Calculate temporary vectorial force */
 380             tx               = _mm_mul_pd(fscal,dx20);
 381             ty               = _mm_mul_pd(fscal,dy20);
 382             tz               = _mm_mul_pd(fscal,dz20);
 383
 384             /* Update vectorial force */
 385             fix2             = _mm_add_pd(fix2,tx);
 386             fiy2             = _mm_add_pd(fiy2,ty);
 387             fiz2             = _mm_add_pd(fiz2,tz);
 388
 389             fjx0             = _mm_add_pd(fjx0,tx);
 390             fjy0             = _mm_add_pd(fjy0,ty);
 391             fjz0             = _mm_add_pd(fjz0,tz);
 392
 393             gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,fjx0,fjy0,fjz0);
 394
 395             /* Inner loop uses 154 flops */
 396         }
 397
 398         if(jidx<j_index_end)
 399         {
 400
 401             jnrA             = jjnr[jidx];
 402             j_coord_offsetA  = DIM*jnrA;
 403
 404             /* load j atom coordinates */
 405             gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
 406                                               &jx0,&jy0,&jz0);
 407
 408             /* Calculate displacement vector */
 409             dx00             = _mm_sub_pd(ix0,jx0);
 410             dy00             = _mm_sub_pd(iy0,jy0);
 411             dz00             = _mm_sub_pd(iz0,jz0);
 412             dx10             = _mm_sub_pd(ix1,jx0);
 413             dy10             = _mm_sub_pd(iy1,jy0);
 414             dz10             = _mm_sub_pd(iz1,jz0);
 415             dx20             = _mm_sub_pd(ix2,jx0);
 416             dy20             = _mm_sub_pd(iy2,jy0);
 417             dz20             = _mm_sub_pd(iz2,jz0);
 418
 419             /* Calculate squared distance and things based on it */
 420             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 421             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
 422             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
 423
 424             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 425             rinv10           = gmx_mm_invsqrt_pd(rsq10);
 426             rinv20           = gmx_mm_invsqrt_pd(rsq20);
 427
 428             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 429             rinvsq10         = _mm_mul_pd(rinv10,rinv10);
 430             rinvsq20         = _mm_mul_pd(rinv20,rinv20);
 431
 432             /* Load parameters for j particles */
 433             jq0              = _mm_load_sd(charge+jnrA+0);
 434             vdwjidx0A        = 2*vdwtype[jnrA+0];
 435
 436             fjx0             = _mm_setzero_pd();
 437             fjy0             = _mm_setzero_pd();
 438             fjz0             = _mm_setzero_pd();
 439
 440             /**************************
 441              * CALCULATE INTERACTIONS *
 442              **************************/
 443
 444             r00              = _mm_mul_pd(rsq00,rinv00);
 445
 446             /* Compute parameters for interactions between i and j atoms */
 447             qq00             = _mm_mul_pd(iq0,jq0);
 448             gmx_mm_load_1pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,&c6_00,&c12_00);
 449
 450             c6grid_00       = gmx_mm_load_1real_pd(vdwgridparam+vdwioffset0+vdwjidx0A);
 451
 452             /* EWALD ELECTROSTATICS */
 453
 454             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 455             ewrt             = _mm_mul_pd(r00,ewtabscale);
 456             ewitab           = _mm_cvttpd_epi32(ewrt);
 457             eweps            = _mm_sub_pd(ewrt,_mm_cvtepi32_pd(ewitab));
 458             ewitab           = _mm_slli_epi32(ewitab,2);
 459             ewtabF           = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 460             ewtabD           = _mm_setzero_pd();
 461             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 462             ewtabV           = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,0) +2);
 463             ewtabFn          = _mm_setzero_pd();
 464             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 465             felec            = _mm_add_pd(ewtabF,_mm_mul_pd(eweps,ewtabD));
 466             velec            = _mm_sub_pd(ewtabV,_mm_mul_pd(_mm_mul_pd(ewtabhalfspace,eweps),_mm_add_pd(ewtabF,felec)));
 467             velec            = _mm_mul_pd(qq00,_mm_sub_pd(rinv00,velec));
 468             felec            = _mm_mul_pd(_mm_mul_pd(qq00,rinv00),_mm_sub_pd(rinvsq00,felec));
 469
 470             /* Analytical LJ-PME */
 471             rinvsix          = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 472             ewcljrsq         = _mm_mul_pd(ewclj2,rsq00);
 473             ewclj6           = _mm_mul_pd(ewclj2,_mm_mul_pd(ewclj2,ewclj2));
 474             exponent         = gmx_simd_exp_d(ewcljrsq);
 475             /* poly = exp(-(beta*r)^2) * (1 + (beta*r)^2 + (beta*r)^4 /2) */
 476             poly             = _mm_mul_pd(exponent,_mm_add_pd(_mm_sub_pd(one,ewcljrsq),_mm_mul_pd(_mm_mul_pd(ewcljrsq,ewcljrsq),one_half)));
 477             /* vvdw6 = [C6 - C6grid * (1-poly)]/r6 */
 478             vvdw6            = _mm_mul_pd(_mm_sub_pd(c6_00,_mm_mul_pd(c6grid_00,_mm_sub_pd(one,poly))),rinvsix);
 479             vvdw12           = _mm_mul_pd(c12_00,_mm_mul_pd(rinvsix,rinvsix));
 480             vvdw             = _mm_sub_pd(_mm_mul_pd(vvdw12,one_twelfth),_mm_mul_pd(vvdw6,one_sixth));
 481             /* fvdw = vvdw12/r - (vvdw6/r + (C6grid * exponent * beta^6)/r) */
 482             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);
 483
 484             /* Update potential sum for this i atom from the interaction with this j atom. */
 485             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 486             velecsum         = _mm_add_pd(velecsum,velec);
 487             vvdw             = _mm_unpacklo_pd(vvdw,_mm_setzero_pd());
 488             vvdwsum          = _mm_add_pd(vvdwsum,vvdw);
 489
 490             fscal            = _mm_add_pd(felec,fvdw);
 491
 492             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 493
 494             /* Calculate temporary vectorial force */
 495             tx               = _mm_mul_pd(fscal,dx00);
 496             ty               = _mm_mul_pd(fscal,dy00);
 497             tz               = _mm_mul_pd(fscal,dz00);
 498
 499             /* Update vectorial force */
 500             fix0             = _mm_add_pd(fix0,tx);
 501             fiy0             = _mm_add_pd(fiy0,ty);
 502             fiz0             = _mm_add_pd(fiz0,tz);
 503
 504             fjx0             = _mm_add_pd(fjx0,tx);
 505             fjy0             = _mm_add_pd(fjy0,ty);
 506             fjz0             = _mm_add_pd(fjz0,tz);
 507
 508             /**************************
 509              * CALCULATE INTERACTIONS *
 510              **************************/
 511
 512             r10              = _mm_mul_pd(rsq10,rinv10);
 513
 514             /* Compute parameters for interactions between i and j atoms */
 515             qq10             = _mm_mul_pd(iq1,jq0);
 516
 517             /* EWALD ELECTROSTATICS */
 518
 519             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 520             ewrt             = _mm_mul_pd(r10,ewtabscale);
 521             ewitab           = _mm_cvttpd_epi32(ewrt);
 522             eweps            = _mm_sub_pd(ewrt,_mm_cvtepi32_pd(ewitab));
 523             ewitab           = _mm_slli_epi32(ewitab,2);
 524             ewtabF           = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 525             ewtabD           = _mm_setzero_pd();
 526             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 527             ewtabV           = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,0) +2);
 528             ewtabFn          = _mm_setzero_pd();
 529             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 530             felec            = _mm_add_pd(ewtabF,_mm_mul_pd(eweps,ewtabD));
 531             velec            = _mm_sub_pd(ewtabV,_mm_mul_pd(_mm_mul_pd(ewtabhalfspace,eweps),_mm_add_pd(ewtabF,felec)));
 532             velec            = _mm_mul_pd(qq10,_mm_sub_pd(rinv10,velec));
 533             felec            = _mm_mul_pd(_mm_mul_pd(qq10,rinv10),_mm_sub_pd(rinvsq10,felec));
 534
 535             /* Update potential sum for this i atom from the interaction with this j atom. */
 536             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 537             velecsum         = _mm_add_pd(velecsum,velec);
 538
 539             fscal            = felec;
 540
 541             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 542
 543             /* Calculate temporary vectorial force */
 544             tx               = _mm_mul_pd(fscal,dx10);
 545             ty               = _mm_mul_pd(fscal,dy10);
 546             tz               = _mm_mul_pd(fscal,dz10);
 547
 548             /* Update vectorial force */
 549             fix1             = _mm_add_pd(fix1,tx);
 550             fiy1             = _mm_add_pd(fiy1,ty);
 551             fiz1             = _mm_add_pd(fiz1,tz);
 552
 553             fjx0             = _mm_add_pd(fjx0,tx);
 554             fjy0             = _mm_add_pd(fjy0,ty);
 555             fjz0             = _mm_add_pd(fjz0,tz);
 556
 557             /**************************
 558              * CALCULATE INTERACTIONS *
 559              **************************/
 560
 561             r20              = _mm_mul_pd(rsq20,rinv20);
 562
 563             /* Compute parameters for interactions between i and j atoms */
 564             qq20             = _mm_mul_pd(iq2,jq0);
 565
 566             /* EWALD ELECTROSTATICS */
 567
 568             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 569             ewrt             = _mm_mul_pd(r20,ewtabscale);
 570             ewitab           = _mm_cvttpd_epi32(ewrt);
 571             eweps            = _mm_sub_pd(ewrt,_mm_cvtepi32_pd(ewitab));
 572             ewitab           = _mm_slli_epi32(ewitab,2);
 573             ewtabF           = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 574             ewtabD           = _mm_setzero_pd();
 575             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 576             ewtabV           = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,0) +2);
 577             ewtabFn          = _mm_setzero_pd();
 578             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 579             felec            = _mm_add_pd(ewtabF,_mm_mul_pd(eweps,ewtabD));
 580             velec            = _mm_sub_pd(ewtabV,_mm_mul_pd(_mm_mul_pd(ewtabhalfspace,eweps),_mm_add_pd(ewtabF,felec)));
 581             velec            = _mm_mul_pd(qq20,_mm_sub_pd(rinv20,velec));
 582             felec            = _mm_mul_pd(_mm_mul_pd(qq20,rinv20),_mm_sub_pd(rinvsq20,felec));
 583
 584             /* Update potential sum for this i atom from the interaction with this j atom. */
 585             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 586             velecsum         = _mm_add_pd(velecsum,velec);
 587
 588             fscal            = felec;
 589
 590             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 591
 592             /* Calculate temporary vectorial force */
 593             tx               = _mm_mul_pd(fscal,dx20);
 594             ty               = _mm_mul_pd(fscal,dy20);
 595             tz               = _mm_mul_pd(fscal,dz20);
 596
 597             /* Update vectorial force */
 598             fix2             = _mm_add_pd(fix2,tx);
 599             fiy2             = _mm_add_pd(fiy2,ty);
 600             fiz2             = _mm_add_pd(fiz2,tz);
 601
 602             fjx0             = _mm_add_pd(fjx0,tx);
 603             fjy0             = _mm_add_pd(fjy0,ty);
 604             fjz0             = _mm_add_pd(fjz0,tz);
 605
 606             gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,fjx0,fjy0,fjz0);
 607
 608             /* Inner loop uses 154 flops */
 609         }
 610
 611         /* End of innermost loop */
 612
 613         gmx_mm_update_iforce_3atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
 614                                               f+i_coord_offset,fshift+i_shift_offset);
 615
 616         ggid                        = gid[iidx];
 617         /* Update potential energies */
 618         gmx_mm_update_1pot_pd(velecsum,kernel_data->energygrp_elec+ggid);
 619         gmx_mm_update_1pot_pd(vvdwsum,kernel_data->energygrp_vdw+ggid);
 620
 621         /* Increment number of inner iterations */
 622         inneriter                  += j_index_end - j_index_start;
 623
 624         /* Outer loop uses 20 flops */
 625     }
 626
 627     /* Increment number of outer iterations */
 628     outeriter        += nri;
 629
 630     /* Update outer/inner flops */
 631
 632     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3_VF,outeriter*20 + inneriter*154);
 633 }
 634 /*
 635  * Gromacs nonbonded kernel:   nb_kernel_ElecEw_VdwLJEw_GeomW3P1_F_sse2_double
 636  * Electrostatics interaction: Ewald
 637  * VdW interaction:            LJEwald
 638  * Geometry:                   Water3-Particle
 639  * Calculate force/pot:        Force
 640  */
 641 void
 642 nb_kernel_ElecEw_VdwLJEw_GeomW3P1_F_sse2_double
 643                     (t_nblist                    * gmx_restrict       nlist,
 644                      rvec                        * gmx_restrict          xx,
 645                      rvec                        * gmx_restrict          ff,
 646                      t_forcerec                  * gmx_restrict          fr,
 647                      t_mdatoms                   * gmx_restrict     mdatoms,
 648                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 649                      t_nrnb                      * gmx_restrict        nrnb)
 650 {
 651     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 652      * just 0 for non-waters.
 653      * Suffixes A,B refer to j loop unrolling done with SSE double precision, e.g. for the two different
 654      * jnr indices corresponding to data put in the four positions in the SIMD register.
 655      */
 656     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 657     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 658     int              jnrA,jnrB;
 659     int              j_coord_offsetA,j_coord_offsetB;
 660     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 661     real             rcutoff_scalar;
 662     real             *shiftvec,*fshift,*x,*f;
 663     __m128d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 664     int              vdwioffset0;
 665     __m128d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 666     int              vdwioffset1;
 667     __m128d          ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
 668     int              vdwioffset2;
 669     __m128d          ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
 670     int              vdwjidx0A,vdwjidx0B;
 671     __m128d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 672     __m128d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 673     __m128d          dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
 674     __m128d          dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
 675     __m128d          velec,felec,velecsum,facel,crf,krf,krf2;
 676     real             *charge;
 677     int              nvdwtype;
 678     __m128d          rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 679     int              *vdwtype;
 680     real             *vdwparam;
 681     __m128d          one_sixth   = _mm_set1_pd(1.0/6.0);
 682     __m128d          one_twelfth = _mm_set1_pd(1.0/12.0);
 683     __m128d           c6grid_00;
 684     __m128d           c6grid_10;
 685     __m128d           c6grid_20;
 686     __m128d           ewclj,ewclj2,ewclj6,ewcljrsq,poly,exponent,f6A,f6B,sh_lj_ewald;
 687     real             *vdwgridparam;
 688     __m128d           one_half = _mm_set1_pd(0.5);
 689     __m128d           minus_one = _mm_set1_pd(-1.0);
 690     __m128i          ewitab;
 691     __m128d          ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
 692     real             *ewtab;
 693     __m128d          dummy_mask,cutoff_mask;
 694     __m128d          signbit   = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
 695     __m128d          one     = _mm_set1_pd(1.0);
 696     __m128d          two     = _mm_set1_pd(2.0);
 697     x                = xx[0];
 698     f                = ff[0];
 699
 700     nri              = nlist->nri;
 701     iinr             = nlist->iinr;
 702     jindex           = nlist->jindex;
 703     jjnr             = nlist->jjnr;
 704     shiftidx         = nlist->shift;
 705     gid              = nlist->gid;
 706     shiftvec         = fr->shift_vec[0];
 707     fshift           = fr->fshift[0];
 708     facel            = _mm_set1_pd(fr->epsfac);
 709     charge           = mdatoms->chargeA;
 710     nvdwtype         = fr->ntype;
 711     vdwparam         = fr->nbfp;
 712     vdwtype          = mdatoms->typeA;
 713     vdwgridparam     = fr->ljpme_c6grid;
 714     sh_lj_ewald      = _mm_set1_pd(fr->ic->sh_lj_ewald);
 715     ewclj            = _mm_set1_pd(fr->ewaldcoeff_lj);
 716     ewclj2           = _mm_mul_pd(minus_one,_mm_mul_pd(ewclj,ewclj));
 717
 718     sh_ewald         = _mm_set1_pd(fr->ic->sh_ewald);
 719     ewtab            = fr->ic->tabq_coul_F;
 720     ewtabscale       = _mm_set1_pd(fr->ic->tabq_scale);
 721     ewtabhalfspace   = _mm_set1_pd(0.5/fr->ic->tabq_scale);
 722
 723     /* Setup water-specific parameters */
 724     inr              = nlist->iinr[0];
 725     iq0              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+0]));
 726     iq1              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+1]));
 727     iq2              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+2]));
 728     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 729
 730     /* Avoid stupid compiler warnings */
 731     jnrA = jnrB = 0;
 732     j_coord_offsetA = 0;
 733     j_coord_offsetB = 0;
 734
 735     outeriter        = 0;
 736     inneriter        = 0;
 737
 738     /* Start outer loop over neighborlists */
 739     for(iidx=0; iidx<nri; iidx++)
 740     {
 741         /* Load shift vector for this list */
 742         i_shift_offset   = DIM*shiftidx[iidx];
 743
 744         /* Load limits for loop over neighbors */
 745         j_index_start    = jindex[iidx];
 746         j_index_end      = jindex[iidx+1];
 747
 748         /* Get outer coordinate index */
 749         inr              = iinr[iidx];
 750         i_coord_offset   = DIM*inr;
 751
 752         /* Load i particle coords and add shift vector */
 753         gmx_mm_load_shift_and_3rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
 754                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 755
 756         fix0             = _mm_setzero_pd();
 757         fiy0             = _mm_setzero_pd();
 758         fiz0             = _mm_setzero_pd();
 759         fix1             = _mm_setzero_pd();
 760         fiy1             = _mm_setzero_pd();
 761         fiz1             = _mm_setzero_pd();
 762         fix2             = _mm_setzero_pd();
 763         fiy2             = _mm_setzero_pd();
 764         fiz2             = _mm_setzero_pd();
 765
 766         /* Start inner kernel loop */
 767         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 768         {
 769
 770             /* Get j neighbor index, and coordinate index */
 771             jnrA             = jjnr[jidx];
 772             jnrB             = jjnr[jidx+1];
 773             j_coord_offsetA  = DIM*jnrA;
 774             j_coord_offsetB  = DIM*jnrB;
 775
 776             /* load j atom coordinates */
 777             gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 778                                               &jx0,&jy0,&jz0);
 779
 780             /* Calculate displacement vector */
 781             dx00             = _mm_sub_pd(ix0,jx0);
 782             dy00             = _mm_sub_pd(iy0,jy0);
 783             dz00             = _mm_sub_pd(iz0,jz0);
 784             dx10             = _mm_sub_pd(ix1,jx0);
 785             dy10             = _mm_sub_pd(iy1,jy0);
 786             dz10             = _mm_sub_pd(iz1,jz0);
 787             dx20             = _mm_sub_pd(ix2,jx0);
 788             dy20             = _mm_sub_pd(iy2,jy0);
 789             dz20             = _mm_sub_pd(iz2,jz0);
 790
 791             /* Calculate squared distance and things based on it */
 792             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 793             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
 794             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
 795
 796             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 797             rinv10           = gmx_mm_invsqrt_pd(rsq10);
 798             rinv20           = gmx_mm_invsqrt_pd(rsq20);
 799
 800             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 801             rinvsq10         = _mm_mul_pd(rinv10,rinv10);
 802             rinvsq20         = _mm_mul_pd(rinv20,rinv20);
 803
 804             /* Load parameters for j particles */
 805             jq0              = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
 806             vdwjidx0A        = 2*vdwtype[jnrA+0];
 807             vdwjidx0B        = 2*vdwtype[jnrB+0];
 808
 809             fjx0             = _mm_setzero_pd();
 810             fjy0             = _mm_setzero_pd();
 811             fjz0             = _mm_setzero_pd();
 812
 813             /**************************
 814              * CALCULATE INTERACTIONS *
 815              **************************/
 816
 817             r00              = _mm_mul_pd(rsq00,rinv00);
 818
 819             /* Compute parameters for interactions between i and j atoms */
 820             qq00             = _mm_mul_pd(iq0,jq0);
 821             gmx_mm_load_2pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,
 822                                          vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
 823
 824             c6grid_00       = gmx_mm_load_2real_swizzle_pd(vdwgridparam+vdwioffset0+vdwjidx0A,
 825                                                                vdwgridparam+vdwioffset0+vdwjidx0B);
 826
 827             /* EWALD ELECTROSTATICS */
 828
 829             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 830             ewrt             = _mm_mul_pd(r00,ewtabscale);
 831             ewitab           = _mm_cvttpd_epi32(ewrt);
 832             eweps            = _mm_sub_pd(ewrt,_mm_cvtepi32_pd(ewitab));
 833             gmx_mm_load_2pair_swizzle_pd(ewtab+gmx_mm_extract_epi32(ewitab,0),ewtab+gmx_mm_extract_epi32(ewitab,1),
 834                                          &ewtabF,&ewtabFn);
 835             felec            = _mm_add_pd(_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF),_mm_mul_pd(eweps,ewtabFn));
 836             felec            = _mm_mul_pd(_mm_mul_pd(qq00,rinv00),_mm_sub_pd(rinvsq00,felec));
 837
 838             /* Analytical LJ-PME */
 839             rinvsix          = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 840             ewcljrsq         = _mm_mul_pd(ewclj2,rsq00);
 841             ewclj6           = _mm_mul_pd(ewclj2,_mm_mul_pd(ewclj2,ewclj2));
 842             exponent         = gmx_simd_exp_d(ewcljrsq);
 843             /* poly = exp(-(beta*r)^2) * (1 + (beta*r)^2 + (beta*r)^4 /2) */
 844             poly             = _mm_mul_pd(exponent,_mm_add_pd(_mm_sub_pd(one,ewcljrsq),_mm_mul_pd(_mm_mul_pd(ewcljrsq,ewcljrsq),one_half)));
 845             /* f6A = 6 * C6grid * (1 - poly) */
 846             f6A              = _mm_mul_pd(c6grid_00,_mm_sub_pd(one,poly));
 847             /* f6B = C6grid * exponent * beta^6 */
 848             f6B              = _mm_mul_pd(_mm_mul_pd(c6grid_00,one_sixth),_mm_mul_pd(exponent,ewclj6));
 849             /* fvdw = 12*C12/r13 - ((6*C6 - f6A)/r6 + f6B)/r */
 850             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);
 851
 852             fscal            = _mm_add_pd(felec,fvdw);
 853
 854             /* Calculate temporary vectorial force */
 855             tx               = _mm_mul_pd(fscal,dx00);
 856             ty               = _mm_mul_pd(fscal,dy00);
 857             tz               = _mm_mul_pd(fscal,dz00);
 858
 859             /* Update vectorial force */
 860             fix0             = _mm_add_pd(fix0,tx);
 861             fiy0             = _mm_add_pd(fiy0,ty);
 862             fiz0             = _mm_add_pd(fiz0,tz);
 863
 864             fjx0             = _mm_add_pd(fjx0,tx);
 865             fjy0             = _mm_add_pd(fjy0,ty);
 866             fjz0             = _mm_add_pd(fjz0,tz);
 867
 868             /**************************
 869              * CALCULATE INTERACTIONS *
 870              **************************/
 871
 872             r10              = _mm_mul_pd(rsq10,rinv10);
 873
 874             /* Compute parameters for interactions between i and j atoms */
 875             qq10             = _mm_mul_pd(iq1,jq0);
 876
 877             /* EWALD ELECTROSTATICS */
 878
 879             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 880             ewrt             = _mm_mul_pd(r10,ewtabscale);
 881             ewitab           = _mm_cvttpd_epi32(ewrt);
 882             eweps            = _mm_sub_pd(ewrt,_mm_cvtepi32_pd(ewitab));
 883             gmx_mm_load_2pair_swizzle_pd(ewtab+gmx_mm_extract_epi32(ewitab,0),ewtab+gmx_mm_extract_epi32(ewitab,1),
 884                                          &ewtabF,&ewtabFn);
 885             felec            = _mm_add_pd(_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF),_mm_mul_pd(eweps,ewtabFn));
 886             felec            = _mm_mul_pd(_mm_mul_pd(qq10,rinv10),_mm_sub_pd(rinvsq10,felec));
 887
 888             fscal            = felec;
 889
 890             /* Calculate temporary vectorial force */
 891             tx               = _mm_mul_pd(fscal,dx10);
 892             ty               = _mm_mul_pd(fscal,dy10);
 893             tz               = _mm_mul_pd(fscal,dz10);
 894
 895             /* Update vectorial force */
 896             fix1             = _mm_add_pd(fix1,tx);
 897             fiy1             = _mm_add_pd(fiy1,ty);
 898             fiz1             = _mm_add_pd(fiz1,tz);
 899
 900             fjx0             = _mm_add_pd(fjx0,tx);
 901             fjy0             = _mm_add_pd(fjy0,ty);
 902             fjz0             = _mm_add_pd(fjz0,tz);
 903
 904             /**************************
 905              * CALCULATE INTERACTIONS *
 906              **************************/
 907
 908             r20              = _mm_mul_pd(rsq20,rinv20);
 909
 910             /* Compute parameters for interactions between i and j atoms */
 911             qq20             = _mm_mul_pd(iq2,jq0);
 912
 913             /* EWALD ELECTROSTATICS */
 914
 915             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 916             ewrt             = _mm_mul_pd(r20,ewtabscale);
 917             ewitab           = _mm_cvttpd_epi32(ewrt);
 918             eweps            = _mm_sub_pd(ewrt,_mm_cvtepi32_pd(ewitab));
 919             gmx_mm_load_2pair_swizzle_pd(ewtab+gmx_mm_extract_epi32(ewitab,0),ewtab+gmx_mm_extract_epi32(ewitab,1),
 920                                          &ewtabF,&ewtabFn);
 921             felec            = _mm_add_pd(_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF),_mm_mul_pd(eweps,ewtabFn));
 922             felec            = _mm_mul_pd(_mm_mul_pd(qq20,rinv20),_mm_sub_pd(rinvsq20,felec));
 923
 924             fscal            = felec;
 925
 926             /* Calculate temporary vectorial force */
 927             tx               = _mm_mul_pd(fscal,dx20);
 928             ty               = _mm_mul_pd(fscal,dy20);
 929             tz               = _mm_mul_pd(fscal,dz20);
 930
 931             /* Update vectorial force */
 932             fix2             = _mm_add_pd(fix2,tx);
 933             fiy2             = _mm_add_pd(fiy2,ty);
 934             fiz2             = _mm_add_pd(fiz2,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             gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,fjx0,fjy0,fjz0);
 941
 942             /* Inner loop uses 134 flops */
 943         }
 944
 945         if(jidx<j_index_end)
 946         {
 947
 948             jnrA             = jjnr[jidx];
 949             j_coord_offsetA  = DIM*jnrA;
 950
 951             /* load j atom coordinates */
 952             gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
 953                                               &jx0,&jy0,&jz0);
 954
 955             /* Calculate displacement vector */
 956             dx00             = _mm_sub_pd(ix0,jx0);
 957             dy00             = _mm_sub_pd(iy0,jy0);
 958             dz00             = _mm_sub_pd(iz0,jz0);
 959             dx10             = _mm_sub_pd(ix1,jx0);
 960             dy10             = _mm_sub_pd(iy1,jy0);
 961             dz10             = _mm_sub_pd(iz1,jz0);
 962             dx20             = _mm_sub_pd(ix2,jx0);
 963             dy20             = _mm_sub_pd(iy2,jy0);
 964             dz20             = _mm_sub_pd(iz2,jz0);
 965
 966             /* Calculate squared distance and things based on it */
 967             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 968             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
 969             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
 970
 971             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 972             rinv10           = gmx_mm_invsqrt_pd(rsq10);
 973             rinv20           = gmx_mm_invsqrt_pd(rsq20);
 974
 975             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 976             rinvsq10         = _mm_mul_pd(rinv10,rinv10);
 977             rinvsq20         = _mm_mul_pd(rinv20,rinv20);
 978
 979             /* Load parameters for j particles */
 980             jq0              = _mm_load_sd(charge+jnrA+0);
 981             vdwjidx0A        = 2*vdwtype[jnrA+0];
 982
 983             fjx0             = _mm_setzero_pd();
 984             fjy0             = _mm_setzero_pd();
 985             fjz0             = _mm_setzero_pd();
 986
 987             /**************************
 988              * CALCULATE INTERACTIONS *
 989              **************************/
 990
 991             r00              = _mm_mul_pd(rsq00,rinv00);
 992
 993             /* Compute parameters for interactions between i and j atoms */
 994             qq00             = _mm_mul_pd(iq0,jq0);
 995             gmx_mm_load_1pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,&c6_00,&c12_00);
 996
 997             c6grid_00       = gmx_mm_load_1real_pd(vdwgridparam+vdwioffset0+vdwjidx0A);
 998
 999             /* EWALD ELECTROSTATICS */
1000
1001             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1002             ewrt             = _mm_mul_pd(r00,ewtabscale);
1003             ewitab           = _mm_cvttpd_epi32(ewrt);
1004             eweps            = _mm_sub_pd(ewrt,_mm_cvtepi32_pd(ewitab));
1005             gmx_mm_load_1pair_swizzle_pd(ewtab+gmx_mm_extract_epi32(ewitab,0),&ewtabF,&ewtabFn);
1006             felec            = _mm_add_pd(_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF),_mm_mul_pd(eweps,ewtabFn));
1007             felec            = _mm_mul_pd(_mm_mul_pd(qq00,rinv00),_mm_sub_pd(rinvsq00,felec));
1008
1009             /* Analytical LJ-PME */
1010             rinvsix          = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
1011             ewcljrsq         = _mm_mul_pd(ewclj2,rsq00);
1012             ewclj6           = _mm_mul_pd(ewclj2,_mm_mul_pd(ewclj2,ewclj2));
1013             exponent         = gmx_simd_exp_d(ewcljrsq);
1014             /* poly = exp(-(beta*r)^2) * (1 + (beta*r)^2 + (beta*r)^4 /2) */
1015             poly             = _mm_mul_pd(exponent,_mm_add_pd(_mm_sub_pd(one,ewcljrsq),_mm_mul_pd(_mm_mul_pd(ewcljrsq,ewcljrsq),one_half)));
1016             /* f6A = 6 * C6grid * (1 - poly) */
1017             f6A              = _mm_mul_pd(c6grid_00,_mm_sub_pd(one,poly));
1018             /* f6B = C6grid * exponent * beta^6 */
1019             f6B              = _mm_mul_pd(_mm_mul_pd(c6grid_00,one_sixth),_mm_mul_pd(exponent,ewclj6));
1020             /* fvdw = 12*C12/r13 - ((6*C6 - f6A)/r6 + f6B)/r */
1021             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);
1022
1023             fscal            = _mm_add_pd(felec,fvdw);
1024
1025             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1026
1027             /* Calculate temporary vectorial force */
1028             tx               = _mm_mul_pd(fscal,dx00);
1029             ty               = _mm_mul_pd(fscal,dy00);
1030             tz               = _mm_mul_pd(fscal,dz00);
1031
1032             /* Update vectorial force */
1033             fix0             = _mm_add_pd(fix0,tx);
1034             fiy0             = _mm_add_pd(fiy0,ty);
1035             fiz0             = _mm_add_pd(fiz0,tz);
1036
1037             fjx0             = _mm_add_pd(fjx0,tx);
1038             fjy0             = _mm_add_pd(fjy0,ty);
1039             fjz0             = _mm_add_pd(fjz0,tz);
1040
1041             /**************************
1042              * CALCULATE INTERACTIONS *
1043              **************************/
1044
1045             r10              = _mm_mul_pd(rsq10,rinv10);
1046
1047             /* Compute parameters for interactions between i and j atoms */
1048             qq10             = _mm_mul_pd(iq1,jq0);
1049
1050             /* EWALD ELECTROSTATICS */
1051
1052             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1053             ewrt             = _mm_mul_pd(r10,ewtabscale);
1054             ewitab           = _mm_cvttpd_epi32(ewrt);
1055             eweps            = _mm_sub_pd(ewrt,_mm_cvtepi32_pd(ewitab));
1056             gmx_mm_load_1pair_swizzle_pd(ewtab+gmx_mm_extract_epi32(ewitab,0),&ewtabF,&ewtabFn);
1057             felec            = _mm_add_pd(_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF),_mm_mul_pd(eweps,ewtabFn));
1058             felec            = _mm_mul_pd(_mm_mul_pd(qq10,rinv10),_mm_sub_pd(rinvsq10,felec));
1059
1060             fscal            = felec;
1061
1062             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1063
1064             /* Calculate temporary vectorial force */
1065             tx               = _mm_mul_pd(fscal,dx10);
1066             ty               = _mm_mul_pd(fscal,dy10);
1067             tz               = _mm_mul_pd(fscal,dz10);
1068
1069             /* Update vectorial force */
1070             fix1             = _mm_add_pd(fix1,tx);
1071             fiy1             = _mm_add_pd(fiy1,ty);
1072             fiz1             = _mm_add_pd(fiz1,tz);
1073
1074             fjx0             = _mm_add_pd(fjx0,tx);
1075             fjy0             = _mm_add_pd(fjy0,ty);
1076             fjz0             = _mm_add_pd(fjz0,tz);
1077
1078             /**************************
1079              * CALCULATE INTERACTIONS *
1080              **************************/
1081
1082             r20              = _mm_mul_pd(rsq20,rinv20);
1083
1084             /* Compute parameters for interactions between i and j atoms */
1085             qq20             = _mm_mul_pd(iq2,jq0);
1086
1087             /* EWALD ELECTROSTATICS */
1088
1089             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1090             ewrt             = _mm_mul_pd(r20,ewtabscale);
1091             ewitab           = _mm_cvttpd_epi32(ewrt);
1092             eweps            = _mm_sub_pd(ewrt,_mm_cvtepi32_pd(ewitab));
1093             gmx_mm_load_1pair_swizzle_pd(ewtab+gmx_mm_extract_epi32(ewitab,0),&ewtabF,&ewtabFn);
1094             felec            = _mm_add_pd(_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF),_mm_mul_pd(eweps,ewtabFn));
1095             felec            = _mm_mul_pd(_mm_mul_pd(qq20,rinv20),_mm_sub_pd(rinvsq20,felec));
1096
1097             fscal            = felec;
1098
1099             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1100
1101             /* Calculate temporary vectorial force */
1102             tx               = _mm_mul_pd(fscal,dx20);
1103             ty               = _mm_mul_pd(fscal,dy20);
1104             tz               = _mm_mul_pd(fscal,dz20);
1105
1106             /* Update vectorial force */
1107             fix2             = _mm_add_pd(fix2,tx);
1108             fiy2             = _mm_add_pd(fiy2,ty);
1109             fiz2             = _mm_add_pd(fiz2,tz);
1110
1111             fjx0             = _mm_add_pd(fjx0,tx);
1112             fjy0             = _mm_add_pd(fjy0,ty);
1113             fjz0             = _mm_add_pd(fjz0,tz);
1114
1115             gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,fjx0,fjy0,fjz0);
1116
1117             /* Inner loop uses 134 flops */
1118         }
1119
1120         /* End of innermost loop */
1121
1122         gmx_mm_update_iforce_3atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
1123                                               f+i_coord_offset,fshift+i_shift_offset);
1124
1125         /* Increment number of inner iterations */
1126         inneriter                  += j_index_end - j_index_start;
1127
1128         /* Outer loop uses 18 flops */
1129     }
1130
1131     /* Increment number of outer iterations */
1132     outeriter        += nri;
1133
1134     /* Update outer/inner flops */
1135
1136     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3_F,outeriter*18 + inneriter*134);
1137 }