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) 2012,2013,2014,2015,2017, 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/gmxlib/nrnb.h"
  46
  47 #include "kernelutil_x86_sse2_double.h"
  48
  49 /*
  50  * Gromacs nonbonded kernel:   nb_kernel_ElecEw_VdwLJEw_GeomW3P1_VF_sse2_double
  51  * Electrostatics interaction: Ewald
  52  * VdW interaction:            LJEwald
  53  * Geometry:                   Water3-Particle
  54  * Calculate force/pot:        PotentialAndForce
  55  */
  56 void
  57 nb_kernel_ElecEw_VdwLJEw_GeomW3P1_VF_sse2_double
  58                     (t_nblist                    * gmx_restrict       nlist,
  59                      rvec                        * gmx_restrict          xx,
  60                      rvec                        * gmx_restrict          ff,
  61                      struct t_forcerec           * gmx_restrict          fr,
  62                      t_mdatoms                   * gmx_restrict     mdatoms,
  63                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
  64                      t_nrnb                      * gmx_restrict        nrnb)
  65 {
  66     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
  67      * just 0 for non-waters.
  68      * Suffixes A,B refer to j loop unrolling done with SSE double precision, e.g. for the two different
  69      * jnr indices corresponding to data put in the four positions in the SIMD register.
  70      */
  71     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
  72     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
  73     int              jnrA,jnrB;
  74     int              j_coord_offsetA,j_coord_offsetB;
  75     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
  76     real             rcutoff_scalar;
  77     real             *shiftvec,*fshift,*x,*f;
  78     __m128d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
  79     int              vdwioffset0;
  80     __m128d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
  81     int              vdwioffset1;
  82     __m128d          ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
  83     int              vdwioffset2;
  84     __m128d          ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
  85     int              vdwjidx0A,vdwjidx0B;
  86     __m128d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  87     __m128d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  88     __m128d          dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
  89     __m128d          dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
  90     __m128d          velec,felec,velecsum,facel,crf,krf,krf2;
  91     real             *charge;
  92     int              nvdwtype;
  93     __m128d          rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
  94     int              *vdwtype;
  95     real             *vdwparam;
  96     __m128d          one_sixth   = _mm_set1_pd(1.0/6.0);
  97     __m128d          one_twelfth = _mm_set1_pd(1.0/12.0);
  98     __m128d           c6grid_00;
  99     __m128d           c6grid_10;
 100     __m128d           c6grid_20;
 101     __m128d           ewclj,ewclj2,ewclj6,ewcljrsq,poly,exponent,f6A,f6B,sh_lj_ewald;
 102     real             *vdwgridparam;
 103     __m128d           one_half = _mm_set1_pd(0.5);
 104     __m128d           minus_one = _mm_set1_pd(-1.0);
 105     __m128i          ewitab;
 106     __m128d          ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
 107     real             *ewtab;
 108     __m128d          dummy_mask,cutoff_mask;
 109     __m128d          signbit   = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
 110     __m128d          one     = _mm_set1_pd(1.0);
 111     __m128d          two     = _mm_set1_pd(2.0);
 112     x                = xx[0];
 113     f                = ff[0];
 114
 115     nri              = nlist->nri;
 116     iinr             = nlist->iinr;
 117     jindex           = nlist->jindex;
 118     jjnr             = nlist->jjnr;
 119     shiftidx         = nlist->shift;
 120     gid              = nlist->gid;
 121     shiftvec         = fr->shift_vec[0];
 122     fshift           = fr->fshift[0];
 123     facel            = _mm_set1_pd(fr->ic->epsfac);
 124     charge           = mdatoms->chargeA;
 125     nvdwtype         = fr->ntype;
 126     vdwparam         = fr->nbfp;
 127     vdwtype          = mdatoms->typeA;
 128     vdwgridparam     = fr->ljpme_c6grid;
 129     sh_lj_ewald      = _mm_set1_pd(fr->ic->sh_lj_ewald);
 130     ewclj            = _mm_set1_pd(fr->ic->ewaldcoeff_lj);
 131     ewclj2           = _mm_mul_pd(minus_one,_mm_mul_pd(ewclj,ewclj));
 132
 133     sh_ewald         = _mm_set1_pd(fr->ic->sh_ewald);
 134     ewtab            = fr->ic->tabq_coul_FDV0;
 135     ewtabscale       = _mm_set1_pd(fr->ic->tabq_scale);
 136     ewtabhalfspace   = _mm_set1_pd(0.5/fr->ic->tabq_scale);
 137
 138     /* Setup water-specific parameters */
 139     inr              = nlist->iinr[0];
 140     iq0              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+0]));
 141     iq1              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+1]));
 142     iq2              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+2]));
 143     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 144
 145     /* Avoid stupid compiler warnings */
 146     jnrA = jnrB = 0;
 147     j_coord_offsetA = 0;
 148     j_coord_offsetB = 0;
 149
 150     outeriter        = 0;
 151     inneriter        = 0;
 152
 153     /* Start outer loop over neighborlists */
 154     for(iidx=0; iidx<nri; iidx++)
 155     {
 156         /* Load shift vector for this list */
 157         i_shift_offset   = DIM*shiftidx[iidx];
 158
 159         /* Load limits for loop over neighbors */
 160         j_index_start    = jindex[iidx];
 161         j_index_end      = jindex[iidx+1];
 162
 163         /* Get outer coordinate index */
 164         inr              = iinr[iidx];
 165         i_coord_offset   = DIM*inr;
 166
 167         /* Load i particle coords and add shift vector */
 168         gmx_mm_load_shift_and_3rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
 169                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 170
 171         fix0             = _mm_setzero_pd();
 172         fiy0             = _mm_setzero_pd();
 173         fiz0             = _mm_setzero_pd();
 174         fix1             = _mm_setzero_pd();
 175         fiy1             = _mm_setzero_pd();
 176         fiz1             = _mm_setzero_pd();
 177         fix2             = _mm_setzero_pd();
 178         fiy2             = _mm_setzero_pd();
 179         fiz2             = _mm_setzero_pd();
 180
 181         /* Reset potential sums */
 182         velecsum         = _mm_setzero_pd();
 183         vvdwsum          = _mm_setzero_pd();
 184
 185         /* Start inner kernel loop */
 186         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 187         {
 188
 189             /* Get j neighbor index, and coordinate index */
 190             jnrA             = jjnr[jidx];
 191             jnrB             = jjnr[jidx+1];
 192             j_coord_offsetA  = DIM*jnrA;
 193             j_coord_offsetB  = DIM*jnrB;
 194
 195             /* load j atom coordinates */
 196             gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 197                                               &jx0,&jy0,&jz0);
 198
 199             /* Calculate displacement vector */
 200             dx00             = _mm_sub_pd(ix0,jx0);
 201             dy00             = _mm_sub_pd(iy0,jy0);
 202             dz00             = _mm_sub_pd(iz0,jz0);
 203             dx10             = _mm_sub_pd(ix1,jx0);
 204             dy10             = _mm_sub_pd(iy1,jy0);
 205             dz10             = _mm_sub_pd(iz1,jz0);
 206             dx20             = _mm_sub_pd(ix2,jx0);
 207             dy20             = _mm_sub_pd(iy2,jy0);
 208             dz20             = _mm_sub_pd(iz2,jz0);
 209
 210             /* Calculate squared distance and things based on it */
 211             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 212             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
 213             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
 214
 215             rinv00           = sse2_invsqrt_d(rsq00);
 216             rinv10           = sse2_invsqrt_d(rsq10);
 217             rinv20           = sse2_invsqrt_d(rsq20);
 218
 219             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 220             rinvsq10         = _mm_mul_pd(rinv10,rinv10);
 221             rinvsq20         = _mm_mul_pd(rinv20,rinv20);
 222
 223             /* Load parameters for j particles */
 224             jq0              = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
 225             vdwjidx0A        = 2*vdwtype[jnrA+0];
 226             vdwjidx0B        = 2*vdwtype[jnrB+0];
 227
 228             fjx0             = _mm_setzero_pd();
 229             fjy0             = _mm_setzero_pd();
 230             fjz0             = _mm_setzero_pd();
 231
 232             /**************************
 233              * CALCULATE INTERACTIONS *
 234              **************************/
 235
 236             r00              = _mm_mul_pd(rsq00,rinv00);
 237
 238             /* Compute parameters for interactions between i and j atoms */
 239             qq00             = _mm_mul_pd(iq0,jq0);
 240             gmx_mm_load_2pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,
 241                                          vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
 242
 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_cvtepi32_pd(ewitab));
 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         = sse2_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_cvtepi32_pd(ewitab));
 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_cvtepi32_pd(ewitab));
 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           = sse2_invsqrt_d(rsq00);
 422             rinv10           = sse2_invsqrt_d(rsq10);
 423             rinv20           = sse2_invsqrt_d(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_cvtepi32_pd(ewitab));
 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         = sse2_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_cvtepi32_pd(ewitab));
 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_cvtepi32_pd(ewitab));
 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_sse2_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_sse2_double
 640                     (t_nblist                    * gmx_restrict       nlist,
 641                      rvec                        * gmx_restrict          xx,
 642                      rvec                        * gmx_restrict          ff,
 643                      struct 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->ic->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->ic->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           = sse2_invsqrt_d(rsq00);
 794             rinv10           = sse2_invsqrt_d(rsq10);
 795             rinv20           = sse2_invsqrt_d(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
 821             c6grid_00       = gmx_mm_load_2real_swizzle_pd(vdwgridparam+vdwioffset0+vdwjidx0A,
 822                                                                vdwgridparam+vdwioffset0+vdwjidx0B);
 823
 824             /* EWALD ELECTROSTATICS */
 825
 826             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 827             ewrt             = _mm_mul_pd(r00,ewtabscale);
 828             ewitab           = _mm_cvttpd_epi32(ewrt);
 829             eweps            = _mm_sub_pd(ewrt,_mm_cvtepi32_pd(ewitab));
 830             gmx_mm_load_2pair_swizzle_pd(ewtab+gmx_mm_extract_epi32(ewitab,0),ewtab+gmx_mm_extract_epi32(ewitab,1),
 831                                          &ewtabF,&ewtabFn);
 832             felec            = _mm_add_pd(_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF),_mm_mul_pd(eweps,ewtabFn));
 833             felec            = _mm_mul_pd(_mm_mul_pd(qq00,rinv00),_mm_sub_pd(rinvsq00,felec));
 834
 835             /* Analytical LJ-PME */
 836             rinvsix          = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 837             ewcljrsq         = _mm_mul_pd(ewclj2,rsq00);
 838             ewclj6           = _mm_mul_pd(ewclj2,_mm_mul_pd(ewclj2,ewclj2));
 839             exponent         = sse2_exp_d(ewcljrsq);
 840             /* poly = exp(-(beta*r)^2) * (1 + (beta*r)^2 + (beta*r)^4 /2) */
 841             poly             = _mm_mul_pd(exponent,_mm_add_pd(_mm_sub_pd(one,ewcljrsq),_mm_mul_pd(_mm_mul_pd(ewcljrsq,ewcljrsq),one_half)));
 842             /* f6A = 6 * C6grid * (1 - poly) */
 843             f6A              = _mm_mul_pd(c6grid_00,_mm_sub_pd(one,poly));
 844             /* f6B = C6grid * exponent * beta^6 */
 845             f6B              = _mm_mul_pd(_mm_mul_pd(c6grid_00,one_sixth),_mm_mul_pd(exponent,ewclj6));
 846             /* fvdw = 12*C12/r13 - ((6*C6 - f6A)/r6 + f6B)/r */
 847             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);
 848
 849             fscal            = _mm_add_pd(felec,fvdw);
 850
 851             /* Calculate temporary vectorial force */
 852             tx               = _mm_mul_pd(fscal,dx00);
 853             ty               = _mm_mul_pd(fscal,dy00);
 854             tz               = _mm_mul_pd(fscal,dz00);
 855
 856             /* Update vectorial force */
 857             fix0             = _mm_add_pd(fix0,tx);
 858             fiy0             = _mm_add_pd(fiy0,ty);
 859             fiz0             = _mm_add_pd(fiz0,tz);
 860
 861             fjx0             = _mm_add_pd(fjx0,tx);
 862             fjy0             = _mm_add_pd(fjy0,ty);
 863             fjz0             = _mm_add_pd(fjz0,tz);
 864
 865             /**************************
 866              * CALCULATE INTERACTIONS *
 867              **************************/
 868
 869             r10              = _mm_mul_pd(rsq10,rinv10);
 870
 871             /* Compute parameters for interactions between i and j atoms */
 872             qq10             = _mm_mul_pd(iq1,jq0);
 873
 874             /* EWALD ELECTROSTATICS */
 875
 876             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 877             ewrt             = _mm_mul_pd(r10,ewtabscale);
 878             ewitab           = _mm_cvttpd_epi32(ewrt);
 879             eweps            = _mm_sub_pd(ewrt,_mm_cvtepi32_pd(ewitab));
 880             gmx_mm_load_2pair_swizzle_pd(ewtab+gmx_mm_extract_epi32(ewitab,0),ewtab+gmx_mm_extract_epi32(ewitab,1),
 881                                          &ewtabF,&ewtabFn);
 882             felec            = _mm_add_pd(_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF),_mm_mul_pd(eweps,ewtabFn));
 883             felec            = _mm_mul_pd(_mm_mul_pd(qq10,rinv10),_mm_sub_pd(rinvsq10,felec));
 884
 885             fscal            = felec;
 886
 887             /* Calculate temporary vectorial force */
 888             tx               = _mm_mul_pd(fscal,dx10);
 889             ty               = _mm_mul_pd(fscal,dy10);
 890             tz               = _mm_mul_pd(fscal,dz10);
 891
 892             /* Update vectorial force */
 893             fix1             = _mm_add_pd(fix1,tx);
 894             fiy1             = _mm_add_pd(fiy1,ty);
 895             fiz1             = _mm_add_pd(fiz1,tz);
 896
 897             fjx0             = _mm_add_pd(fjx0,tx);
 898             fjy0             = _mm_add_pd(fjy0,ty);
 899             fjz0             = _mm_add_pd(fjz0,tz);
 900
 901             /**************************
 902              * CALCULATE INTERACTIONS *
 903              **************************/
 904
 905             r20              = _mm_mul_pd(rsq20,rinv20);
 906
 907             /* Compute parameters for interactions between i and j atoms */
 908             qq20             = _mm_mul_pd(iq2,jq0);
 909
 910             /* EWALD ELECTROSTATICS */
 911
 912             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 913             ewrt             = _mm_mul_pd(r20,ewtabscale);
 914             ewitab           = _mm_cvttpd_epi32(ewrt);
 915             eweps            = _mm_sub_pd(ewrt,_mm_cvtepi32_pd(ewitab));
 916             gmx_mm_load_2pair_swizzle_pd(ewtab+gmx_mm_extract_epi32(ewitab,0),ewtab+gmx_mm_extract_epi32(ewitab,1),
 917                                          &ewtabF,&ewtabFn);
 918             felec            = _mm_add_pd(_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF),_mm_mul_pd(eweps,ewtabFn));
 919             felec            = _mm_mul_pd(_mm_mul_pd(qq20,rinv20),_mm_sub_pd(rinvsq20,felec));
 920
 921             fscal            = felec;
 922
 923             /* Calculate temporary vectorial force */
 924             tx               = _mm_mul_pd(fscal,dx20);
 925             ty               = _mm_mul_pd(fscal,dy20);
 926             tz               = _mm_mul_pd(fscal,dz20);
 927
 928             /* Update vectorial force */
 929             fix2             = _mm_add_pd(fix2,tx);
 930             fiy2             = _mm_add_pd(fiy2,ty);
 931             fiz2             = _mm_add_pd(fiz2,tz);
 932
 933             fjx0             = _mm_add_pd(fjx0,tx);
 934             fjy0             = _mm_add_pd(fjy0,ty);
 935             fjz0             = _mm_add_pd(fjz0,tz);
 936
 937             gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,fjx0,fjy0,fjz0);
 938
 939             /* Inner loop uses 134 flops */
 940         }
 941
 942         if(jidx<j_index_end)
 943         {
 944
 945             jnrA             = jjnr[jidx];
 946             j_coord_offsetA  = DIM*jnrA;
 947
 948             /* load j atom coordinates */
 949             gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
 950                                               &jx0,&jy0,&jz0);
 951
 952             /* Calculate displacement vector */
 953             dx00             = _mm_sub_pd(ix0,jx0);
 954             dy00             = _mm_sub_pd(iy0,jy0);
 955             dz00             = _mm_sub_pd(iz0,jz0);
 956             dx10             = _mm_sub_pd(ix1,jx0);
 957             dy10             = _mm_sub_pd(iy1,jy0);
 958             dz10             = _mm_sub_pd(iz1,jz0);
 959             dx20             = _mm_sub_pd(ix2,jx0);
 960             dy20             = _mm_sub_pd(iy2,jy0);
 961             dz20             = _mm_sub_pd(iz2,jz0);
 962
 963             /* Calculate squared distance and things based on it */
 964             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 965             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
 966             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
 967
 968             rinv00           = sse2_invsqrt_d(rsq00);
 969             rinv10           = sse2_invsqrt_d(rsq10);
 970             rinv20           = sse2_invsqrt_d(rsq20);
 971
 972             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 973             rinvsq10         = _mm_mul_pd(rinv10,rinv10);
 974             rinvsq20         = _mm_mul_pd(rinv20,rinv20);
 975
 976             /* Load parameters for j particles */
 977             jq0              = _mm_load_sd(charge+jnrA+0);
 978             vdwjidx0A        = 2*vdwtype[jnrA+0];
 979
 980             fjx0             = _mm_setzero_pd();
 981             fjy0             = _mm_setzero_pd();
 982             fjz0             = _mm_setzero_pd();
 983
 984             /**************************
 985              * CALCULATE INTERACTIONS *
 986              **************************/
 987
 988             r00              = _mm_mul_pd(rsq00,rinv00);
 989
 990             /* Compute parameters for interactions between i and j atoms */
 991             qq00             = _mm_mul_pd(iq0,jq0);
 992             gmx_mm_load_1pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,&c6_00,&c12_00);
 993
 994             c6grid_00       = gmx_mm_load_1real_pd(vdwgridparam+vdwioffset0+vdwjidx0A);
 995
 996             /* EWALD ELECTROSTATICS */
 997
 998             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 999             ewrt             = _mm_mul_pd(r00,ewtabscale);
1000             ewitab           = _mm_cvttpd_epi32(ewrt);
1001             eweps            = _mm_sub_pd(ewrt,_mm_cvtepi32_pd(ewitab));
1002             gmx_mm_load_1pair_swizzle_pd(ewtab+gmx_mm_extract_epi32(ewitab,0),&ewtabF,&ewtabFn);
1003             felec            = _mm_add_pd(_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF),_mm_mul_pd(eweps,ewtabFn));
1004             felec            = _mm_mul_pd(_mm_mul_pd(qq00,rinv00),_mm_sub_pd(rinvsq00,felec));
1005
1006             /* Analytical LJ-PME */
1007             rinvsix          = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
1008             ewcljrsq         = _mm_mul_pd(ewclj2,rsq00);
1009             ewclj6           = _mm_mul_pd(ewclj2,_mm_mul_pd(ewclj2,ewclj2));
1010             exponent         = sse2_exp_d(ewcljrsq);
1011             /* poly = exp(-(beta*r)^2) * (1 + (beta*r)^2 + (beta*r)^4 /2) */
1012             poly             = _mm_mul_pd(exponent,_mm_add_pd(_mm_sub_pd(one,ewcljrsq),_mm_mul_pd(_mm_mul_pd(ewcljrsq,ewcljrsq),one_half)));
1013             /* f6A = 6 * C6grid * (1 - poly) */
1014             f6A              = _mm_mul_pd(c6grid_00,_mm_sub_pd(one,poly));
1015             /* f6B = C6grid * exponent * beta^6 */
1016             f6B              = _mm_mul_pd(_mm_mul_pd(c6grid_00,one_sixth),_mm_mul_pd(exponent,ewclj6));
1017             /* fvdw = 12*C12/r13 - ((6*C6 - f6A)/r6 + f6B)/r */
1018             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);
1019
1020             fscal            = _mm_add_pd(felec,fvdw);
1021
1022             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1023
1024             /* Calculate temporary vectorial force */
1025             tx               = _mm_mul_pd(fscal,dx00);
1026             ty               = _mm_mul_pd(fscal,dy00);
1027             tz               = _mm_mul_pd(fscal,dz00);
1028
1029             /* Update vectorial force */
1030             fix0             = _mm_add_pd(fix0,tx);
1031             fiy0             = _mm_add_pd(fiy0,ty);
1032             fiz0             = _mm_add_pd(fiz0,tz);
1033
1034             fjx0             = _mm_add_pd(fjx0,tx);
1035             fjy0             = _mm_add_pd(fjy0,ty);
1036             fjz0             = _mm_add_pd(fjz0,tz);
1037
1038             /**************************
1039              * CALCULATE INTERACTIONS *
1040              **************************/
1041
1042             r10              = _mm_mul_pd(rsq10,rinv10);
1043
1044             /* Compute parameters for interactions between i and j atoms */
1045             qq10             = _mm_mul_pd(iq1,jq0);
1046
1047             /* EWALD ELECTROSTATICS */
1048
1049             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1050             ewrt             = _mm_mul_pd(r10,ewtabscale);
1051             ewitab           = _mm_cvttpd_epi32(ewrt);
1052             eweps            = _mm_sub_pd(ewrt,_mm_cvtepi32_pd(ewitab));
1053             gmx_mm_load_1pair_swizzle_pd(ewtab+gmx_mm_extract_epi32(ewitab,0),&ewtabF,&ewtabFn);
1054             felec            = _mm_add_pd(_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF),_mm_mul_pd(eweps,ewtabFn));
1055             felec            = _mm_mul_pd(_mm_mul_pd(qq10,rinv10),_mm_sub_pd(rinvsq10,felec));
1056
1057             fscal            = felec;
1058
1059             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1060
1061             /* Calculate temporary vectorial force */
1062             tx               = _mm_mul_pd(fscal,dx10);
1063             ty               = _mm_mul_pd(fscal,dy10);
1064             tz               = _mm_mul_pd(fscal,dz10);
1065
1066             /* Update vectorial force */
1067             fix1             = _mm_add_pd(fix1,tx);
1068             fiy1             = _mm_add_pd(fiy1,ty);
1069             fiz1             = _mm_add_pd(fiz1,tz);
1070
1071             fjx0             = _mm_add_pd(fjx0,tx);
1072             fjy0             = _mm_add_pd(fjy0,ty);
1073             fjz0             = _mm_add_pd(fjz0,tz);
1074
1075             /**************************
1076              * CALCULATE INTERACTIONS *
1077              **************************/
1078
1079             r20              = _mm_mul_pd(rsq20,rinv20);
1080
1081             /* Compute parameters for interactions between i and j atoms */
1082             qq20             = _mm_mul_pd(iq2,jq0);
1083
1084             /* EWALD ELECTROSTATICS */
1085
1086             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1087             ewrt             = _mm_mul_pd(r20,ewtabscale);
1088             ewitab           = _mm_cvttpd_epi32(ewrt);
1089             eweps            = _mm_sub_pd(ewrt,_mm_cvtepi32_pd(ewitab));
1090             gmx_mm_load_1pair_swizzle_pd(ewtab+gmx_mm_extract_epi32(ewitab,0),&ewtabF,&ewtabFn);
1091             felec            = _mm_add_pd(_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF),_mm_mul_pd(eweps,ewtabFn));
1092             felec            = _mm_mul_pd(_mm_mul_pd(qq20,rinv20),_mm_sub_pd(rinvsq20,felec));
1093
1094             fscal            = felec;
1095
1096             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1097
1098             /* Calculate temporary vectorial force */
1099             tx               = _mm_mul_pd(fscal,dx20);
1100             ty               = _mm_mul_pd(fscal,dy20);
1101             tz               = _mm_mul_pd(fscal,dz20);
1102
1103             /* Update vectorial force */
1104             fix2             = _mm_add_pd(fix2,tx);
1105             fiy2             = _mm_add_pd(fiy2,ty);
1106             fiz2             = _mm_add_pd(fiz2,tz);
1107
1108             fjx0             = _mm_add_pd(fjx0,tx);
1109             fjy0             = _mm_add_pd(fjy0,ty);
1110             fjz0             = _mm_add_pd(fjz0,tz);
1111
1112             gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,fjx0,fjy0,fjz0);
1113
1114             /* Inner loop uses 134 flops */
1115         }
1116
1117         /* End of innermost loop */
1118
1119         gmx_mm_update_iforce_3atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
1120                                               f+i_coord_offset,fshift+i_shift_offset);
1121
1122         /* Increment number of inner iterations */
1123         inneriter                  += j_index_end - j_index_start;
1124
1125         /* Outer loop uses 18 flops */
1126     }
1127
1128     /* Increment number of outer iterations */
1129     outeriter        += nri;
1130
1131     /* Update outer/inner flops */
1132
1133     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3_F,outeriter*18 + inneriter*134);
1134 }