src/gromacs/gmxlib/nonbonded/nb_kernel_sse4_1_double/nb_kernel_ElecEw_VdwLJEw_GeomW3P1_sse4_1_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 sse4_1_double kernel generator.
  37  */
  38 #include "config.h"
  39
  40 #include <math.h>
  41
  42 #include "../nb_kernel.h"
  43 #include "types/simple.h"
  44 #include "gromacs/math/vec.h"
  45 #include "nrnb.h"
  46
  47 #include "gromacs/simd/math_x86_sse4_1_double.h"
  48 #include "kernelutil_x86_sse4_1_double.h"
  49
  50 /*
  51  * Gromacs nonbonded kernel:   nb_kernel_ElecEw_VdwLJEw_GeomW3P1_VF_sse4_1_double
  52  * Electrostatics interaction: Ewald
  53  * VdW interaction:            LJEwald
  54  * Geometry:                   Water3-Particle
  55  * Calculate force/pot:        PotentialAndForce
  56  */
  57 void
  58 nb_kernel_ElecEw_VdwLJEw_GeomW3P1_VF_sse4_1_double
  59                     (t_nblist                    * gmx_restrict       nlist,
  60                      rvec                        * gmx_restrict          xx,
  61                      rvec                        * gmx_restrict          ff,
  62                      t_forcerec                  * gmx_restrict          fr,
  63                      t_mdatoms                   * gmx_restrict     mdatoms,
  64                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
  65                      t_nrnb                      * gmx_restrict        nrnb)
  66 {
  67     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
  68      * just 0 for non-waters.
  69      * Suffixes A,B refer to j loop unrolling done with SSE double precision, e.g. for the two different
  70      * jnr indices corresponding to data put in the four positions in the SIMD register.
  71      */
  72     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
  73     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
  74     int              jnrA,jnrB;
  75     int              j_coord_offsetA,j_coord_offsetB;
  76     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
  77     real             rcutoff_scalar;
  78     real             *shiftvec,*fshift,*x,*f;
  79     __m128d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
  80     int              vdwioffset0;
  81     __m128d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
  82     int              vdwioffset1;
  83     __m128d          ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
  84     int              vdwioffset2;
  85     __m128d          ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
  86     int              vdwjidx0A,vdwjidx0B;
  87     __m128d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  88     __m128d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  89     __m128d          dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
  90     __m128d          dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
  91     __m128d          velec,felec,velecsum,facel,crf,krf,krf2;
  92     real             *charge;
  93     int              nvdwtype;
  94     __m128d          rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
  95     int              *vdwtype;
  96     real             *vdwparam;
  97     __m128d          one_sixth   = _mm_set1_pd(1.0/6.0);
  98     __m128d          one_twelfth = _mm_set1_pd(1.0/12.0);
  99     __m128d           c6grid_00;
 100     __m128d           c6grid_10;
 101     __m128d           c6grid_20;
 102     __m128d           ewclj,ewclj2,ewclj6,ewcljrsq,poly,exponent,f6A,f6B,sh_lj_ewald;
 103     real             *vdwgridparam;
 104     __m128d           one_half = _mm_set1_pd(0.5);
 105     __m128d           minus_one = _mm_set1_pd(-1.0);
 106     __m128i          ewitab;
 107     __m128d          ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
 108     real             *ewtab;
 109     __m128d          dummy_mask,cutoff_mask;
 110     __m128d          signbit   = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
 111     __m128d          one     = _mm_set1_pd(1.0);
 112     __m128d          two     = _mm_set1_pd(2.0);
 113     x                = xx[0];
 114     f                = ff[0];
 115
 116     nri              = nlist->nri;
 117     iinr             = nlist->iinr;
 118     jindex           = nlist->jindex;
 119     jjnr             = nlist->jjnr;
 120     shiftidx         = nlist->shift;
 121     gid              = nlist->gid;
 122     shiftvec         = fr->shift_vec[0];
 123     fshift           = fr->fshift[0];
 124     facel            = _mm_set1_pd(fr->epsfac);
 125     charge           = mdatoms->chargeA;
 126     nvdwtype         = fr->ntype;
 127     vdwparam         = fr->nbfp;
 128     vdwtype          = mdatoms->typeA;
 129     vdwgridparam     = fr->ljpme_c6grid;
 130     sh_lj_ewald      = _mm_set1_pd(fr->ic->sh_lj_ewald);
 131     ewclj            = _mm_set1_pd(fr->ewaldcoeff_lj);
 132     ewclj2           = _mm_mul_pd(minus_one,_mm_mul_pd(ewclj,ewclj));
 133
 134     sh_ewald         = _mm_set1_pd(fr->ic->sh_ewald);
 135     ewtab            = fr->ic->tabq_coul_FDV0;
 136     ewtabscale       = _mm_set1_pd(fr->ic->tabq_scale);
 137     ewtabhalfspace   = _mm_set1_pd(0.5/fr->ic->tabq_scale);
 138
 139     /* Setup water-specific parameters */
 140     inr              = nlist->iinr[0];
 141     iq0              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+0]));
 142     iq1              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+1]));
 143     iq2              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+2]));
 144     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 145
 146     /* Avoid stupid compiler warnings */
 147     jnrA = jnrB = 0;
 148     j_coord_offsetA = 0;
 149     j_coord_offsetB = 0;
 150
 151     outeriter        = 0;
 152     inneriter        = 0;
 153
 154     /* Start outer loop over neighborlists */
 155     for(iidx=0; iidx<nri; iidx++)
 156     {
 157         /* Load shift vector for this list */
 158         i_shift_offset   = DIM*shiftidx[iidx];
 159
 160         /* Load limits for loop over neighbors */
 161         j_index_start    = jindex[iidx];
 162         j_index_end      = jindex[iidx+1];
 163
 164         /* Get outer coordinate index */
 165         inr              = iinr[iidx];
 166         i_coord_offset   = DIM*inr;
 167
 168         /* Load i particle coords and add shift vector */
 169         gmx_mm_load_shift_and_3rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
 170                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 171
 172         fix0             = _mm_setzero_pd();
 173         fiy0             = _mm_setzero_pd();
 174         fiz0             = _mm_setzero_pd();
 175         fix1             = _mm_setzero_pd();
 176         fiy1             = _mm_setzero_pd();
 177         fiz1             = _mm_setzero_pd();
 178         fix2             = _mm_setzero_pd();
 179         fiy2             = _mm_setzero_pd();
 180         fiz2             = _mm_setzero_pd();
 181
 182         /* Reset potential sums */
 183         velecsum         = _mm_setzero_pd();
 184         vvdwsum          = _mm_setzero_pd();
 185
 186         /* Start inner kernel loop */
 187         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 188         {
 189
 190             /* Get j neighbor index, and coordinate index */
 191             jnrA             = jjnr[jidx];
 192             jnrB             = jjnr[jidx+1];
 193             j_coord_offsetA  = DIM*jnrA;
 194             j_coord_offsetB  = DIM*jnrB;
 195
 196             /* load j atom coordinates */
 197             gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 198                                               &jx0,&jy0,&jz0);
 199
 200             /* Calculate displacement vector */
 201             dx00             = _mm_sub_pd(ix0,jx0);
 202             dy00             = _mm_sub_pd(iy0,jy0);
 203             dz00             = _mm_sub_pd(iz0,jz0);
 204             dx10             = _mm_sub_pd(ix1,jx0);
 205             dy10             = _mm_sub_pd(iy1,jy0);
 206             dz10             = _mm_sub_pd(iz1,jz0);
 207             dx20             = _mm_sub_pd(ix2,jx0);
 208             dy20             = _mm_sub_pd(iy2,jy0);
 209             dz20             = _mm_sub_pd(iz2,jz0);
 210
 211             /* Calculate squared distance and things based on it */
 212             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 213             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
 214             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
 215
 216             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 217             rinv10           = gmx_mm_invsqrt_pd(rsq10);
 218             rinv20           = gmx_mm_invsqrt_pd(rsq20);
 219
 220             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 221             rinvsq10         = _mm_mul_pd(rinv10,rinv10);
 222             rinvsq20         = _mm_mul_pd(rinv20,rinv20);
 223
 224             /* Load parameters for j particles */
 225             jq0              = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
 226             vdwjidx0A        = 2*vdwtype[jnrA+0];
 227             vdwjidx0B        = 2*vdwtype[jnrB+0];
 228
 229             fjx0             = _mm_setzero_pd();
 230             fjy0             = _mm_setzero_pd();
 231             fjz0             = _mm_setzero_pd();
 232
 233             /**************************
 234              * CALCULATE INTERACTIONS *
 235              **************************/
 236
 237             r00              = _mm_mul_pd(rsq00,rinv00);
 238
 239             /* Compute parameters for interactions between i and j atoms */
 240             qq00             = _mm_mul_pd(iq0,jq0);
 241             gmx_mm_load_2pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,
 242                                          vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
 243             c6grid_00       = gmx_mm_load_2real_swizzle_pd(vdwgridparam+vdwioffset0+vdwjidx0A,
 244                                                                vdwgridparam+vdwioffset0+vdwjidx0B);
 245
 246             /* EWALD ELECTROSTATICS */
 247
 248             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 249             ewrt             = _mm_mul_pd(r00,ewtabscale);
 250             ewitab           = _mm_cvttpd_epi32(ewrt);
 251             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 252             ewitab           = _mm_slli_epi32(ewitab,2);
 253             ewtabF           = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 254             ewtabD           = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,1) );
 255             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 256             ewtabV           = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,0) +2);
 257             ewtabFn          = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,1) +2);
 258             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 259             felec            = _mm_add_pd(ewtabF,_mm_mul_pd(eweps,ewtabD));
 260             velec            = _mm_sub_pd(ewtabV,_mm_mul_pd(_mm_mul_pd(ewtabhalfspace,eweps),_mm_add_pd(ewtabF,felec)));
 261             velec            = _mm_mul_pd(qq00,_mm_sub_pd(rinv00,velec));
 262             felec            = _mm_mul_pd(_mm_mul_pd(qq00,rinv00),_mm_sub_pd(rinvsq00,felec));
 263
 264             /* Analytical LJ-PME */
 265             rinvsix          = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 266             ewcljrsq         = _mm_mul_pd(ewclj2,rsq00);
 267             ewclj6           = _mm_mul_pd(ewclj2,_mm_mul_pd(ewclj2,ewclj2));
 268             exponent         = gmx_simd_exp_d(ewcljrsq);
 269             /* poly = exp(-(beta*r)^2) * (1 + (beta*r)^2 + (beta*r)^4 /2) */
 270             poly             = _mm_mul_pd(exponent,_mm_add_pd(_mm_sub_pd(one,ewcljrsq),_mm_mul_pd(_mm_mul_pd(ewcljrsq,ewcljrsq),one_half)));
 271             /* vvdw6 = [C6 - C6grid * (1-poly)]/r6 */
 272             vvdw6            = _mm_mul_pd(_mm_sub_pd(c6_00,_mm_mul_pd(c6grid_00,_mm_sub_pd(one,poly))),rinvsix);
 273             vvdw12           = _mm_mul_pd(c12_00,_mm_mul_pd(rinvsix,rinvsix));
 274             vvdw             = _mm_sub_pd(_mm_mul_pd(vvdw12,one_twelfth),_mm_mul_pd(vvdw6,one_sixth));
 275             /* fvdw = vvdw12/r - (vvdw6/r + (C6grid * exponent * beta^6)/r) */
 276             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);
 277
 278             /* Update potential sum for this i atom from the interaction with this j atom. */
 279             velecsum         = _mm_add_pd(velecsum,velec);
 280             vvdwsum          = _mm_add_pd(vvdwsum,vvdw);
 281
 282             fscal            = _mm_add_pd(felec,fvdw);
 283
 284             /* Calculate temporary vectorial force */
 285             tx               = _mm_mul_pd(fscal,dx00);
 286             ty               = _mm_mul_pd(fscal,dy00);
 287             tz               = _mm_mul_pd(fscal,dz00);
 288
 289             /* Update vectorial force */
 290             fix0             = _mm_add_pd(fix0,tx);
 291             fiy0             = _mm_add_pd(fiy0,ty);
 292             fiz0             = _mm_add_pd(fiz0,tz);
 293
 294             fjx0             = _mm_add_pd(fjx0,tx);
 295             fjy0             = _mm_add_pd(fjy0,ty);
 296             fjz0             = _mm_add_pd(fjz0,tz);
 297
 298             /**************************
 299              * CALCULATE INTERACTIONS *
 300              **************************/
 301
 302             r10              = _mm_mul_pd(rsq10,rinv10);
 303
 304             /* Compute parameters for interactions between i and j atoms */
 305             qq10             = _mm_mul_pd(iq1,jq0);
 306
 307             /* EWALD ELECTROSTATICS */
 308
 309             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 310             ewrt             = _mm_mul_pd(r10,ewtabscale);
 311             ewitab           = _mm_cvttpd_epi32(ewrt);
 312             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 313             ewitab           = _mm_slli_epi32(ewitab,2);
 314             ewtabF           = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 315             ewtabD           = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,1) );
 316             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 317             ewtabV           = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,0) +2);
 318             ewtabFn          = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,1) +2);
 319             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 320             felec            = _mm_add_pd(ewtabF,_mm_mul_pd(eweps,ewtabD));
 321             velec            = _mm_sub_pd(ewtabV,_mm_mul_pd(_mm_mul_pd(ewtabhalfspace,eweps),_mm_add_pd(ewtabF,felec)));
 322             velec            = _mm_mul_pd(qq10,_mm_sub_pd(rinv10,velec));
 323             felec            = _mm_mul_pd(_mm_mul_pd(qq10,rinv10),_mm_sub_pd(rinvsq10,felec));
 324
 325             /* Update potential sum for this i atom from the interaction with this j atom. */
 326             velecsum         = _mm_add_pd(velecsum,velec);
 327
 328             fscal            = felec;
 329
 330             /* Calculate temporary vectorial force */
 331             tx               = _mm_mul_pd(fscal,dx10);
 332             ty               = _mm_mul_pd(fscal,dy10);
 333             tz               = _mm_mul_pd(fscal,dz10);
 334
 335             /* Update vectorial force */
 336             fix1             = _mm_add_pd(fix1,tx);
 337             fiy1             = _mm_add_pd(fiy1,ty);
 338             fiz1             = _mm_add_pd(fiz1,tz);
 339
 340             fjx0             = _mm_add_pd(fjx0,tx);
 341             fjy0             = _mm_add_pd(fjy0,ty);
 342             fjz0             = _mm_add_pd(fjz0,tz);
 343
 344             /**************************
 345              * CALCULATE INTERACTIONS *
 346              **************************/
 347
 348             r20              = _mm_mul_pd(rsq20,rinv20);
 349
 350             /* Compute parameters for interactions between i and j atoms */
 351             qq20             = _mm_mul_pd(iq2,jq0);
 352
 353             /* EWALD ELECTROSTATICS */
 354
 355             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 356             ewrt             = _mm_mul_pd(r20,ewtabscale);
 357             ewitab           = _mm_cvttpd_epi32(ewrt);
 358             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 359             ewitab           = _mm_slli_epi32(ewitab,2);
 360             ewtabF           = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 361             ewtabD           = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,1) );
 362             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 363             ewtabV           = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,0) +2);
 364             ewtabFn          = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,1) +2);
 365             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 366             felec            = _mm_add_pd(ewtabF,_mm_mul_pd(eweps,ewtabD));
 367             velec            = _mm_sub_pd(ewtabV,_mm_mul_pd(_mm_mul_pd(ewtabhalfspace,eweps),_mm_add_pd(ewtabF,felec)));
 368             velec            = _mm_mul_pd(qq20,_mm_sub_pd(rinv20,velec));
 369             felec            = _mm_mul_pd(_mm_mul_pd(qq20,rinv20),_mm_sub_pd(rinvsq20,felec));
 370
 371             /* Update potential sum for this i atom from the interaction with this j atom. */
 372             velecsum         = _mm_add_pd(velecsum,velec);
 373
 374             fscal            = felec;
 375
 376             /* Calculate temporary vectorial force */
 377             tx               = _mm_mul_pd(fscal,dx20);
 378             ty               = _mm_mul_pd(fscal,dy20);
 379             tz               = _mm_mul_pd(fscal,dz20);
 380
 381             /* Update vectorial force */
 382             fix2             = _mm_add_pd(fix2,tx);
 383             fiy2             = _mm_add_pd(fiy2,ty);
 384             fiz2             = _mm_add_pd(fiz2,tz);
 385
 386             fjx0             = _mm_add_pd(fjx0,tx);
 387             fjy0             = _mm_add_pd(fjy0,ty);
 388             fjz0             = _mm_add_pd(fjz0,tz);
 389
 390             gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,fjx0,fjy0,fjz0);
 391
 392             /* Inner loop uses 154 flops */
 393         }
 394
 395         if(jidx<j_index_end)
 396         {
 397
 398             jnrA             = jjnr[jidx];
 399             j_coord_offsetA  = DIM*jnrA;
 400
 401             /* load j atom coordinates */
 402             gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
 403                                               &jx0,&jy0,&jz0);
 404
 405             /* Calculate displacement vector */
 406             dx00             = _mm_sub_pd(ix0,jx0);
 407             dy00             = _mm_sub_pd(iy0,jy0);
 408             dz00             = _mm_sub_pd(iz0,jz0);
 409             dx10             = _mm_sub_pd(ix1,jx0);
 410             dy10             = _mm_sub_pd(iy1,jy0);
 411             dz10             = _mm_sub_pd(iz1,jz0);
 412             dx20             = _mm_sub_pd(ix2,jx0);
 413             dy20             = _mm_sub_pd(iy2,jy0);
 414             dz20             = _mm_sub_pd(iz2,jz0);
 415
 416             /* Calculate squared distance and things based on it */
 417             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 418             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
 419             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
 420
 421             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 422             rinv10           = gmx_mm_invsqrt_pd(rsq10);
 423             rinv20           = gmx_mm_invsqrt_pd(rsq20);
 424
 425             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 426             rinvsq10         = _mm_mul_pd(rinv10,rinv10);
 427             rinvsq20         = _mm_mul_pd(rinv20,rinv20);
 428
 429             /* Load parameters for j particles */
 430             jq0              = _mm_load_sd(charge+jnrA+0);
 431             vdwjidx0A        = 2*vdwtype[jnrA+0];
 432
 433             fjx0             = _mm_setzero_pd();
 434             fjy0             = _mm_setzero_pd();
 435             fjz0             = _mm_setzero_pd();
 436
 437             /**************************
 438              * CALCULATE INTERACTIONS *
 439              **************************/
 440
 441             r00              = _mm_mul_pd(rsq00,rinv00);
 442
 443             /* Compute parameters for interactions between i and j atoms */
 444             qq00             = _mm_mul_pd(iq0,jq0);
 445             gmx_mm_load_1pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,&c6_00,&c12_00);
 446
 447             c6grid_00       = gmx_mm_load_1real_pd(vdwgridparam+vdwioffset0+vdwjidx0A);
 448
 449             /* EWALD ELECTROSTATICS */
 450
 451             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 452             ewrt             = _mm_mul_pd(r00,ewtabscale);
 453             ewitab           = _mm_cvttpd_epi32(ewrt);
 454             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 455             ewitab           = _mm_slli_epi32(ewitab,2);
 456             ewtabF           = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 457             ewtabD           = _mm_setzero_pd();
 458             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 459             ewtabV           = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,0) +2);
 460             ewtabFn          = _mm_setzero_pd();
 461             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 462             felec            = _mm_add_pd(ewtabF,_mm_mul_pd(eweps,ewtabD));
 463             velec            = _mm_sub_pd(ewtabV,_mm_mul_pd(_mm_mul_pd(ewtabhalfspace,eweps),_mm_add_pd(ewtabF,felec)));
 464             velec            = _mm_mul_pd(qq00,_mm_sub_pd(rinv00,velec));
 465             felec            = _mm_mul_pd(_mm_mul_pd(qq00,rinv00),_mm_sub_pd(rinvsq00,felec));
 466
 467             /* Analytical LJ-PME */
 468             rinvsix          = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 469             ewcljrsq         = _mm_mul_pd(ewclj2,rsq00);
 470             ewclj6           = _mm_mul_pd(ewclj2,_mm_mul_pd(ewclj2,ewclj2));
 471             exponent         = gmx_simd_exp_d(ewcljrsq);
 472             /* poly = exp(-(beta*r)^2) * (1 + (beta*r)^2 + (beta*r)^4 /2) */
 473             poly             = _mm_mul_pd(exponent,_mm_add_pd(_mm_sub_pd(one,ewcljrsq),_mm_mul_pd(_mm_mul_pd(ewcljrsq,ewcljrsq),one_half)));
 474             /* vvdw6 = [C6 - C6grid * (1-poly)]/r6 */
 475             vvdw6            = _mm_mul_pd(_mm_sub_pd(c6_00,_mm_mul_pd(c6grid_00,_mm_sub_pd(one,poly))),rinvsix);
 476             vvdw12           = _mm_mul_pd(c12_00,_mm_mul_pd(rinvsix,rinvsix));
 477             vvdw             = _mm_sub_pd(_mm_mul_pd(vvdw12,one_twelfth),_mm_mul_pd(vvdw6,one_sixth));
 478             /* fvdw = vvdw12/r - (vvdw6/r + (C6grid * exponent * beta^6)/r) */
 479             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);
 480
 481             /* Update potential sum for this i atom from the interaction with this j atom. */
 482             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 483             velecsum         = _mm_add_pd(velecsum,velec);
 484             vvdw             = _mm_unpacklo_pd(vvdw,_mm_setzero_pd());
 485             vvdwsum          = _mm_add_pd(vvdwsum,vvdw);
 486
 487             fscal            = _mm_add_pd(felec,fvdw);
 488
 489             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 490
 491             /* Calculate temporary vectorial force */
 492             tx               = _mm_mul_pd(fscal,dx00);
 493             ty               = _mm_mul_pd(fscal,dy00);
 494             tz               = _mm_mul_pd(fscal,dz00);
 495
 496             /* Update vectorial force */
 497             fix0             = _mm_add_pd(fix0,tx);
 498             fiy0             = _mm_add_pd(fiy0,ty);
 499             fiz0             = _mm_add_pd(fiz0,tz);
 500
 501             fjx0             = _mm_add_pd(fjx0,tx);
 502             fjy0             = _mm_add_pd(fjy0,ty);
 503             fjz0             = _mm_add_pd(fjz0,tz);
 504
 505             /**************************
 506              * CALCULATE INTERACTIONS *
 507              **************************/
 508
 509             r10              = _mm_mul_pd(rsq10,rinv10);
 510
 511             /* Compute parameters for interactions between i and j atoms */
 512             qq10             = _mm_mul_pd(iq1,jq0);
 513
 514             /* EWALD ELECTROSTATICS */
 515
 516             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 517             ewrt             = _mm_mul_pd(r10,ewtabscale);
 518             ewitab           = _mm_cvttpd_epi32(ewrt);
 519             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 520             ewitab           = _mm_slli_epi32(ewitab,2);
 521             ewtabF           = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 522             ewtabD           = _mm_setzero_pd();
 523             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 524             ewtabV           = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,0) +2);
 525             ewtabFn          = _mm_setzero_pd();
 526             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 527             felec            = _mm_add_pd(ewtabF,_mm_mul_pd(eweps,ewtabD));
 528             velec            = _mm_sub_pd(ewtabV,_mm_mul_pd(_mm_mul_pd(ewtabhalfspace,eweps),_mm_add_pd(ewtabF,felec)));
 529             velec            = _mm_mul_pd(qq10,_mm_sub_pd(rinv10,velec));
 530             felec            = _mm_mul_pd(_mm_mul_pd(qq10,rinv10),_mm_sub_pd(rinvsq10,felec));
 531
 532             /* Update potential sum for this i atom from the interaction with this j atom. */
 533             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 534             velecsum         = _mm_add_pd(velecsum,velec);
 535
 536             fscal            = felec;
 537
 538             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 539
 540             /* Calculate temporary vectorial force */
 541             tx               = _mm_mul_pd(fscal,dx10);
 542             ty               = _mm_mul_pd(fscal,dy10);
 543             tz               = _mm_mul_pd(fscal,dz10);
 544
 545             /* Update vectorial force */
 546             fix1             = _mm_add_pd(fix1,tx);
 547             fiy1             = _mm_add_pd(fiy1,ty);
 548             fiz1             = _mm_add_pd(fiz1,tz);
 549
 550             fjx0             = _mm_add_pd(fjx0,tx);
 551             fjy0             = _mm_add_pd(fjy0,ty);
 552             fjz0             = _mm_add_pd(fjz0,tz);
 553
 554             /**************************
 555              * CALCULATE INTERACTIONS *
 556              **************************/
 557
 558             r20              = _mm_mul_pd(rsq20,rinv20);
 559
 560             /* Compute parameters for interactions between i and j atoms */
 561             qq20             = _mm_mul_pd(iq2,jq0);
 562
 563             /* EWALD ELECTROSTATICS */
 564
 565             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 566             ewrt             = _mm_mul_pd(r20,ewtabscale);
 567             ewitab           = _mm_cvttpd_epi32(ewrt);
 568             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 569             ewitab           = _mm_slli_epi32(ewitab,2);
 570             ewtabF           = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 571             ewtabD           = _mm_setzero_pd();
 572             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 573             ewtabV           = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,0) +2);
 574             ewtabFn          = _mm_setzero_pd();
 575             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 576             felec            = _mm_add_pd(ewtabF,_mm_mul_pd(eweps,ewtabD));
 577             velec            = _mm_sub_pd(ewtabV,_mm_mul_pd(_mm_mul_pd(ewtabhalfspace,eweps),_mm_add_pd(ewtabF,felec)));
 578             velec            = _mm_mul_pd(qq20,_mm_sub_pd(rinv20,velec));
 579             felec            = _mm_mul_pd(_mm_mul_pd(qq20,rinv20),_mm_sub_pd(rinvsq20,felec));
 580
 581             /* Update potential sum for this i atom from the interaction with this j atom. */
 582             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 583             velecsum         = _mm_add_pd(velecsum,velec);
 584
 585             fscal            = felec;
 586
 587             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 588
 589             /* Calculate temporary vectorial force */
 590             tx               = _mm_mul_pd(fscal,dx20);
 591             ty               = _mm_mul_pd(fscal,dy20);
 592             tz               = _mm_mul_pd(fscal,dz20);
 593
 594             /* Update vectorial force */
 595             fix2             = _mm_add_pd(fix2,tx);
 596             fiy2             = _mm_add_pd(fiy2,ty);
 597             fiz2             = _mm_add_pd(fiz2,tz);
 598
 599             fjx0             = _mm_add_pd(fjx0,tx);
 600             fjy0             = _mm_add_pd(fjy0,ty);
 601             fjz0             = _mm_add_pd(fjz0,tz);
 602
 603             gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,fjx0,fjy0,fjz0);
 604
 605             /* Inner loop uses 154 flops */
 606         }
 607
 608         /* End of innermost loop */
 609
 610         gmx_mm_update_iforce_3atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
 611                                               f+i_coord_offset,fshift+i_shift_offset);
 612
 613         ggid                        = gid[iidx];
 614         /* Update potential energies */
 615         gmx_mm_update_1pot_pd(velecsum,kernel_data->energygrp_elec+ggid);
 616         gmx_mm_update_1pot_pd(vvdwsum,kernel_data->energygrp_vdw+ggid);
 617
 618         /* Increment number of inner iterations */
 619         inneriter                  += j_index_end - j_index_start;
 620
 621         /* Outer loop uses 20 flops */
 622     }
 623
 624     /* Increment number of outer iterations */
 625     outeriter        += nri;
 626
 627     /* Update outer/inner flops */
 628
 629     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3_VF,outeriter*20 + inneriter*154);
 630 }
 631 /*
 632  * Gromacs nonbonded kernel:   nb_kernel_ElecEw_VdwLJEw_GeomW3P1_F_sse4_1_double
 633  * Electrostatics interaction: Ewald
 634  * VdW interaction:            LJEwald
 635  * Geometry:                   Water3-Particle
 636  * Calculate force/pot:        Force
 637  */
 638 void
 639 nb_kernel_ElecEw_VdwLJEw_GeomW3P1_F_sse4_1_double
 640                     (t_nblist                    * gmx_restrict       nlist,
 641                      rvec                        * gmx_restrict          xx,
 642                      rvec                        * gmx_restrict          ff,
 643                      t_forcerec                  * gmx_restrict          fr,
 644                      t_mdatoms                   * gmx_restrict     mdatoms,
 645                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 646                      t_nrnb                      * gmx_restrict        nrnb)
 647 {
 648     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 649      * just 0 for non-waters.
 650      * Suffixes A,B refer to j loop unrolling done with SSE double precision, e.g. for the two different
 651      * jnr indices corresponding to data put in the four positions in the SIMD register.
 652      */
 653     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 654     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 655     int              jnrA,jnrB;
 656     int              j_coord_offsetA,j_coord_offsetB;
 657     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 658     real             rcutoff_scalar;
 659     real             *shiftvec,*fshift,*x,*f;
 660     __m128d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 661     int              vdwioffset0;
 662     __m128d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 663     int              vdwioffset1;
 664     __m128d          ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
 665     int              vdwioffset2;
 666     __m128d          ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
 667     int              vdwjidx0A,vdwjidx0B;
 668     __m128d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 669     __m128d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 670     __m128d          dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
 671     __m128d          dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
 672     __m128d          velec,felec,velecsum,facel,crf,krf,krf2;
 673     real             *charge;
 674     int              nvdwtype;
 675     __m128d          rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 676     int              *vdwtype;
 677     real             *vdwparam;
 678     __m128d          one_sixth   = _mm_set1_pd(1.0/6.0);
 679     __m128d          one_twelfth = _mm_set1_pd(1.0/12.0);
 680     __m128d           c6grid_00;
 681     __m128d           c6grid_10;
 682     __m128d           c6grid_20;
 683     __m128d           ewclj,ewclj2,ewclj6,ewcljrsq,poly,exponent,f6A,f6B,sh_lj_ewald;
 684     real             *vdwgridparam;
 685     __m128d           one_half = _mm_set1_pd(0.5);
 686     __m128d           minus_one = _mm_set1_pd(-1.0);
 687     __m128i          ewitab;
 688     __m128d          ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
 689     real             *ewtab;
 690     __m128d          dummy_mask,cutoff_mask;
 691     __m128d          signbit   = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
 692     __m128d          one     = _mm_set1_pd(1.0);
 693     __m128d          two     = _mm_set1_pd(2.0);
 694     x                = xx[0];
 695     f                = ff[0];
 696
 697     nri              = nlist->nri;
 698     iinr             = nlist->iinr;
 699     jindex           = nlist->jindex;
 700     jjnr             = nlist->jjnr;
 701     shiftidx         = nlist->shift;
 702     gid              = nlist->gid;
 703     shiftvec         = fr->shift_vec[0];
 704     fshift           = fr->fshift[0];
 705     facel            = _mm_set1_pd(fr->epsfac);
 706     charge           = mdatoms->chargeA;
 707     nvdwtype         = fr->ntype;
 708     vdwparam         = fr->nbfp;
 709     vdwtype          = mdatoms->typeA;
 710     vdwgridparam     = fr->ljpme_c6grid;
 711     sh_lj_ewald      = _mm_set1_pd(fr->ic->sh_lj_ewald);
 712     ewclj            = _mm_set1_pd(fr->ewaldcoeff_lj);
 713     ewclj2           = _mm_mul_pd(minus_one,_mm_mul_pd(ewclj,ewclj));
 714
 715     sh_ewald         = _mm_set1_pd(fr->ic->sh_ewald);
 716     ewtab            = fr->ic->tabq_coul_F;
 717     ewtabscale       = _mm_set1_pd(fr->ic->tabq_scale);
 718     ewtabhalfspace   = _mm_set1_pd(0.5/fr->ic->tabq_scale);
 719
 720     /* Setup water-specific parameters */
 721     inr              = nlist->iinr[0];
 722     iq0              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+0]));
 723     iq1              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+1]));
 724     iq2              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+2]));
 725     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 726
 727     /* Avoid stupid compiler warnings */
 728     jnrA = jnrB = 0;
 729     j_coord_offsetA = 0;
 730     j_coord_offsetB = 0;
 731
 732     outeriter        = 0;
 733     inneriter        = 0;
 734
 735     /* Start outer loop over neighborlists */
 736     for(iidx=0; iidx<nri; iidx++)
 737     {
 738         /* Load shift vector for this list */
 739         i_shift_offset   = DIM*shiftidx[iidx];
 740
 741         /* Load limits for loop over neighbors */
 742         j_index_start    = jindex[iidx];
 743         j_index_end      = jindex[iidx+1];
 744
 745         /* Get outer coordinate index */
 746         inr              = iinr[iidx];
 747         i_coord_offset   = DIM*inr;
 748
 749         /* Load i particle coords and add shift vector */
 750         gmx_mm_load_shift_and_3rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
 751                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 752
 753         fix0             = _mm_setzero_pd();
 754         fiy0             = _mm_setzero_pd();
 755         fiz0             = _mm_setzero_pd();
 756         fix1             = _mm_setzero_pd();
 757         fiy1             = _mm_setzero_pd();
 758         fiz1             = _mm_setzero_pd();
 759         fix2             = _mm_setzero_pd();
 760         fiy2             = _mm_setzero_pd();
 761         fiz2             = _mm_setzero_pd();
 762
 763         /* Start inner kernel loop */
 764         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 765         {
 766
 767             /* Get j neighbor index, and coordinate index */
 768             jnrA             = jjnr[jidx];
 769             jnrB             = jjnr[jidx+1];
 770             j_coord_offsetA  = DIM*jnrA;
 771             j_coord_offsetB  = DIM*jnrB;
 772
 773             /* load j atom coordinates */
 774             gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 775                                               &jx0,&jy0,&jz0);
 776
 777             /* Calculate displacement vector */
 778             dx00             = _mm_sub_pd(ix0,jx0);
 779             dy00             = _mm_sub_pd(iy0,jy0);
 780             dz00             = _mm_sub_pd(iz0,jz0);
 781             dx10             = _mm_sub_pd(ix1,jx0);
 782             dy10             = _mm_sub_pd(iy1,jy0);
 783             dz10             = _mm_sub_pd(iz1,jz0);
 784             dx20             = _mm_sub_pd(ix2,jx0);
 785             dy20             = _mm_sub_pd(iy2,jy0);
 786             dz20             = _mm_sub_pd(iz2,jz0);
 787
 788             /* Calculate squared distance and things based on it */
 789             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 790             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
 791             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
 792
 793             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 794             rinv10           = gmx_mm_invsqrt_pd(rsq10);
 795             rinv20           = gmx_mm_invsqrt_pd(rsq20);
 796
 797             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 798             rinvsq10         = _mm_mul_pd(rinv10,rinv10);
 799             rinvsq20         = _mm_mul_pd(rinv20,rinv20);
 800
 801             /* Load parameters for j particles */
 802             jq0              = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
 803             vdwjidx0A        = 2*vdwtype[jnrA+0];
 804             vdwjidx0B        = 2*vdwtype[jnrB+0];
 805
 806             fjx0             = _mm_setzero_pd();
 807             fjy0             = _mm_setzero_pd();
 808             fjz0             = _mm_setzero_pd();
 809
 810             /**************************
 811              * CALCULATE INTERACTIONS *
 812              **************************/
 813
 814             r00              = _mm_mul_pd(rsq00,rinv00);
 815
 816             /* Compute parameters for interactions between i and j atoms */
 817             qq00             = _mm_mul_pd(iq0,jq0);
 818             gmx_mm_load_2pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,
 819                                          vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
 820             c6grid_00       = gmx_mm_load_2real_swizzle_pd(vdwgridparam+vdwioffset0+vdwjidx0A,
 821                                                                vdwgridparam+vdwioffset0+vdwjidx0B);
 822
 823             /* EWALD ELECTROSTATICS */
 824
 825             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 826             ewrt             = _mm_mul_pd(r00,ewtabscale);
 827             ewitab           = _mm_cvttpd_epi32(ewrt);
 828             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 829             gmx_mm_load_2pair_swizzle_pd(ewtab+gmx_mm_extract_epi32(ewitab,0),ewtab+gmx_mm_extract_epi32(ewitab,1),
 830                                          &ewtabF,&ewtabFn);
 831             felec            = _mm_add_pd(_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF),_mm_mul_pd(eweps,ewtabFn));
 832             felec            = _mm_mul_pd(_mm_mul_pd(qq00,rinv00),_mm_sub_pd(rinvsq00,felec));
 833
 834             /* Analytical LJ-PME */
 835             rinvsix          = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 836             ewcljrsq         = _mm_mul_pd(ewclj2,rsq00);
 837             ewclj6           = _mm_mul_pd(ewclj2,_mm_mul_pd(ewclj2,ewclj2));
 838             exponent         = gmx_simd_exp_d(ewcljrsq);
 839             /* poly = exp(-(beta*r)^2) * (1 + (beta*r)^2 + (beta*r)^4 /2) */
 840             poly             = _mm_mul_pd(exponent,_mm_add_pd(_mm_sub_pd(one,ewcljrsq),_mm_mul_pd(_mm_mul_pd(ewcljrsq,ewcljrsq),one_half)));
 841             /* f6A = 6 * C6grid * (1 - poly) */
 842             f6A              = _mm_mul_pd(c6grid_00,_mm_sub_pd(one,poly));
 843             /* f6B = C6grid * exponent * beta^6 */
 844             f6B              = _mm_mul_pd(_mm_mul_pd(c6grid_00,one_sixth),_mm_mul_pd(exponent,ewclj6));
 845             /* fvdw = 12*C12/r13 - ((6*C6 - f6A)/r6 + f6B)/r */
 846             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);
 847
 848             fscal            = _mm_add_pd(felec,fvdw);
 849
 850             /* Calculate temporary vectorial force */
 851             tx               = _mm_mul_pd(fscal,dx00);
 852             ty               = _mm_mul_pd(fscal,dy00);
 853             tz               = _mm_mul_pd(fscal,dz00);
 854
 855             /* Update vectorial force */
 856             fix0             = _mm_add_pd(fix0,tx);
 857             fiy0             = _mm_add_pd(fiy0,ty);
 858             fiz0             = _mm_add_pd(fiz0,tz);
 859
 860             fjx0             = _mm_add_pd(fjx0,tx);
 861             fjy0             = _mm_add_pd(fjy0,ty);
 862             fjz0             = _mm_add_pd(fjz0,tz);
 863
 864             /**************************
 865              * CALCULATE INTERACTIONS *
 866              **************************/
 867
 868             r10              = _mm_mul_pd(rsq10,rinv10);
 869
 870             /* Compute parameters for interactions between i and j atoms */
 871             qq10             = _mm_mul_pd(iq1,jq0);
 872
 873             /* EWALD ELECTROSTATICS */
 874
 875             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 876             ewrt             = _mm_mul_pd(r10,ewtabscale);
 877             ewitab           = _mm_cvttpd_epi32(ewrt);
 878             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 879             gmx_mm_load_2pair_swizzle_pd(ewtab+gmx_mm_extract_epi32(ewitab,0),ewtab+gmx_mm_extract_epi32(ewitab,1),
 880                                          &ewtabF,&ewtabFn);
 881             felec            = _mm_add_pd(_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF),_mm_mul_pd(eweps,ewtabFn));
 882             felec            = _mm_mul_pd(_mm_mul_pd(qq10,rinv10),_mm_sub_pd(rinvsq10,felec));
 883
 884             fscal            = felec;
 885
 886             /* Calculate temporary vectorial force */
 887             tx               = _mm_mul_pd(fscal,dx10);
 888             ty               = _mm_mul_pd(fscal,dy10);
 889             tz               = _mm_mul_pd(fscal,dz10);
 890
 891             /* Update vectorial force */
 892             fix1             = _mm_add_pd(fix1,tx);
 893             fiy1             = _mm_add_pd(fiy1,ty);
 894             fiz1             = _mm_add_pd(fiz1,tz);
 895
 896             fjx0             = _mm_add_pd(fjx0,tx);
 897             fjy0             = _mm_add_pd(fjy0,ty);
 898             fjz0             = _mm_add_pd(fjz0,tz);
 899
 900             /**************************
 901              * CALCULATE INTERACTIONS *
 902              **************************/
 903
 904             r20              = _mm_mul_pd(rsq20,rinv20);
 905
 906             /* Compute parameters for interactions between i and j atoms */
 907             qq20             = _mm_mul_pd(iq2,jq0);
 908
 909             /* EWALD ELECTROSTATICS */
 910
 911             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 912             ewrt             = _mm_mul_pd(r20,ewtabscale);
 913             ewitab           = _mm_cvttpd_epi32(ewrt);
 914             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 915             gmx_mm_load_2pair_swizzle_pd(ewtab+gmx_mm_extract_epi32(ewitab,0),ewtab+gmx_mm_extract_epi32(ewitab,1),
 916                                          &ewtabF,&ewtabFn);
 917             felec            = _mm_add_pd(_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF),_mm_mul_pd(eweps,ewtabFn));
 918             felec            = _mm_mul_pd(_mm_mul_pd(qq20,rinv20),_mm_sub_pd(rinvsq20,felec));
 919
 920             fscal            = felec;
 921
 922             /* Calculate temporary vectorial force */
 923             tx               = _mm_mul_pd(fscal,dx20);
 924             ty               = _mm_mul_pd(fscal,dy20);
 925             tz               = _mm_mul_pd(fscal,dz20);
 926
 927             /* Update vectorial force */
 928             fix2             = _mm_add_pd(fix2,tx);
 929             fiy2             = _mm_add_pd(fiy2,ty);
 930             fiz2             = _mm_add_pd(fiz2,tz);
 931
 932             fjx0             = _mm_add_pd(fjx0,tx);
 933             fjy0             = _mm_add_pd(fjy0,ty);
 934             fjz0             = _mm_add_pd(fjz0,tz);
 935
 936             gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,fjx0,fjy0,fjz0);
 937
 938             /* Inner loop uses 134 flops */
 939         }
 940
 941         if(jidx<j_index_end)
 942         {
 943
 944             jnrA             = jjnr[jidx];
 945             j_coord_offsetA  = DIM*jnrA;
 946
 947             /* load j atom coordinates */
 948             gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
 949                                               &jx0,&jy0,&jz0);
 950
 951             /* Calculate displacement vector */
 952             dx00             = _mm_sub_pd(ix0,jx0);
 953             dy00             = _mm_sub_pd(iy0,jy0);
 954             dz00             = _mm_sub_pd(iz0,jz0);
 955             dx10             = _mm_sub_pd(ix1,jx0);
 956             dy10             = _mm_sub_pd(iy1,jy0);
 957             dz10             = _mm_sub_pd(iz1,jz0);
 958             dx20             = _mm_sub_pd(ix2,jx0);
 959             dy20             = _mm_sub_pd(iy2,jy0);
 960             dz20             = _mm_sub_pd(iz2,jz0);
 961
 962             /* Calculate squared distance and things based on it */
 963             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 964             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
 965             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
 966
 967             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 968             rinv10           = gmx_mm_invsqrt_pd(rsq10);
 969             rinv20           = gmx_mm_invsqrt_pd(rsq20);
 970
 971             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 972             rinvsq10         = _mm_mul_pd(rinv10,rinv10);
 973             rinvsq20         = _mm_mul_pd(rinv20,rinv20);
 974
 975             /* Load parameters for j particles */
 976             jq0              = _mm_load_sd(charge+jnrA+0);
 977             vdwjidx0A        = 2*vdwtype[jnrA+0];
 978
 979             fjx0             = _mm_setzero_pd();
 980             fjy0             = _mm_setzero_pd();
 981             fjz0             = _mm_setzero_pd();
 982
 983             /**************************
 984              * CALCULATE INTERACTIONS *
 985              **************************/
 986
 987             r00              = _mm_mul_pd(rsq00,rinv00);
 988
 989             /* Compute parameters for interactions between i and j atoms */
 990             qq00             = _mm_mul_pd(iq0,jq0);
 991             gmx_mm_load_1pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,&c6_00,&c12_00);
 992
 993             c6grid_00       = gmx_mm_load_1real_pd(vdwgridparam+vdwioffset0+vdwjidx0A);
 994
 995             /* EWALD ELECTROSTATICS */
 996
 997             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 998             ewrt             = _mm_mul_pd(r00,ewtabscale);
 999             ewitab           = _mm_cvttpd_epi32(ewrt);
1000             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
1001             gmx_mm_load_1pair_swizzle_pd(ewtab+gmx_mm_extract_epi32(ewitab,0),&ewtabF,&ewtabFn);
1002             felec            = _mm_add_pd(_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF),_mm_mul_pd(eweps,ewtabFn));
1003             felec            = _mm_mul_pd(_mm_mul_pd(qq00,rinv00),_mm_sub_pd(rinvsq00,felec));
1004
1005             /* Analytical LJ-PME */
1006             rinvsix          = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
1007             ewcljrsq         = _mm_mul_pd(ewclj2,rsq00);
1008             ewclj6           = _mm_mul_pd(ewclj2,_mm_mul_pd(ewclj2,ewclj2));
1009             exponent         = gmx_simd_exp_d(ewcljrsq);
1010             /* poly = exp(-(beta*r)^2) * (1 + (beta*r)^2 + (beta*r)^4 /2) */
1011             poly             = _mm_mul_pd(exponent,_mm_add_pd(_mm_sub_pd(one,ewcljrsq),_mm_mul_pd(_mm_mul_pd(ewcljrsq,ewcljrsq),one_half)));
1012             /* f6A = 6 * C6grid * (1 - poly) */
1013             f6A              = _mm_mul_pd(c6grid_00,_mm_sub_pd(one,poly));
1014             /* f6B = C6grid * exponent * beta^6 */
1015             f6B              = _mm_mul_pd(_mm_mul_pd(c6grid_00,one_sixth),_mm_mul_pd(exponent,ewclj6));
1016             /* fvdw = 12*C12/r13 - ((6*C6 - f6A)/r6 + f6B)/r */
1017             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);
1018
1019             fscal            = _mm_add_pd(felec,fvdw);
1020
1021             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1022
1023             /* Calculate temporary vectorial force */
1024             tx               = _mm_mul_pd(fscal,dx00);
1025             ty               = _mm_mul_pd(fscal,dy00);
1026             tz               = _mm_mul_pd(fscal,dz00);
1027
1028             /* Update vectorial force */
1029             fix0             = _mm_add_pd(fix0,tx);
1030             fiy0             = _mm_add_pd(fiy0,ty);
1031             fiz0             = _mm_add_pd(fiz0,tz);
1032
1033             fjx0             = _mm_add_pd(fjx0,tx);
1034             fjy0             = _mm_add_pd(fjy0,ty);
1035             fjz0             = _mm_add_pd(fjz0,tz);
1036
1037             /**************************
1038              * CALCULATE INTERACTIONS *
1039              **************************/
1040
1041             r10              = _mm_mul_pd(rsq10,rinv10);
1042
1043             /* Compute parameters for interactions between i and j atoms */
1044             qq10             = _mm_mul_pd(iq1,jq0);
1045
1046             /* EWALD ELECTROSTATICS */
1047
1048             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1049             ewrt             = _mm_mul_pd(r10,ewtabscale);
1050             ewitab           = _mm_cvttpd_epi32(ewrt);
1051             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
1052             gmx_mm_load_1pair_swizzle_pd(ewtab+gmx_mm_extract_epi32(ewitab,0),&ewtabF,&ewtabFn);
1053             felec            = _mm_add_pd(_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF),_mm_mul_pd(eweps,ewtabFn));
1054             felec            = _mm_mul_pd(_mm_mul_pd(qq10,rinv10),_mm_sub_pd(rinvsq10,felec));
1055
1056             fscal            = felec;
1057
1058             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1059
1060             /* Calculate temporary vectorial force */
1061             tx               = _mm_mul_pd(fscal,dx10);
1062             ty               = _mm_mul_pd(fscal,dy10);
1063             tz               = _mm_mul_pd(fscal,dz10);
1064
1065             /* Update vectorial force */
1066             fix1             = _mm_add_pd(fix1,tx);
1067             fiy1             = _mm_add_pd(fiy1,ty);
1068             fiz1             = _mm_add_pd(fiz1,tz);
1069
1070             fjx0             = _mm_add_pd(fjx0,tx);
1071             fjy0             = _mm_add_pd(fjy0,ty);
1072             fjz0             = _mm_add_pd(fjz0,tz);
1073
1074             /**************************
1075              * CALCULATE INTERACTIONS *
1076              **************************/
1077
1078             r20              = _mm_mul_pd(rsq20,rinv20);
1079
1080             /* Compute parameters for interactions between i and j atoms */
1081             qq20             = _mm_mul_pd(iq2,jq0);
1082
1083             /* EWALD ELECTROSTATICS */
1084
1085             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1086             ewrt             = _mm_mul_pd(r20,ewtabscale);
1087             ewitab           = _mm_cvttpd_epi32(ewrt);
1088             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
1089             gmx_mm_load_1pair_swizzle_pd(ewtab+gmx_mm_extract_epi32(ewitab,0),&ewtabF,&ewtabFn);
1090             felec            = _mm_add_pd(_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF),_mm_mul_pd(eweps,ewtabFn));
1091             felec            = _mm_mul_pd(_mm_mul_pd(qq20,rinv20),_mm_sub_pd(rinvsq20,felec));
1092
1093             fscal            = felec;
1094
1095             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1096
1097             /* Calculate temporary vectorial force */
1098             tx               = _mm_mul_pd(fscal,dx20);
1099             ty               = _mm_mul_pd(fscal,dy20);
1100             tz               = _mm_mul_pd(fscal,dz20);
1101
1102             /* Update vectorial force */
1103             fix2             = _mm_add_pd(fix2,tx);
1104             fiy2             = _mm_add_pd(fiy2,ty);
1105             fiz2             = _mm_add_pd(fiz2,tz);
1106
1107             fjx0             = _mm_add_pd(fjx0,tx);
1108             fjy0             = _mm_add_pd(fjy0,ty);
1109             fjz0             = _mm_add_pd(fjz0,tz);
1110
1111             gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,fjx0,fjy0,fjz0);
1112
1113             /* Inner loop uses 134 flops */
1114         }
1115
1116         /* End of innermost loop */
1117
1118         gmx_mm_update_iforce_3atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
1119                                               f+i_coord_offset,fshift+i_shift_offset);
1120
1121         /* Increment number of inner iterations */
1122         inneriter                  += j_index_end - j_index_start;
1123
1124         /* Outer loop uses 18 flops */
1125     }
1126
1127     /* Increment number of outer iterations */
1128     outeriter        += nri;
1129
1130     /* Update outer/inner flops */
1131
1132     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3_F,outeriter*18 + inneriter*134);
1133 }