src/gromacs/gmxlib/nonbonded/nb_kernel_sse4_1_double/nb_kernel_ElecEw_VdwLJ_GeomW4P1_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 "gmxpre.h"
  39
  40 #include "config.h"
  41
  42 #include <math.h>
  43
  44 #include "../nb_kernel.h"
  45 #include "gromacs/legacyheaders/types/simple.h"
  46 #include "gromacs/math/vec.h"
  47 #include "gromacs/legacyheaders/nrnb.h"
  48
  49 #include "gromacs/simd/math_x86_sse4_1_double.h"
  50 #include "kernelutil_x86_sse4_1_double.h"
  51
  52 /*
  53  * Gromacs nonbonded kernel:   nb_kernel_ElecEw_VdwLJ_GeomW4P1_VF_sse4_1_double
  54  * Electrostatics interaction: Ewald
  55  * VdW interaction:            LennardJones
  56  * Geometry:                   Water4-Particle
  57  * Calculate force/pot:        PotentialAndForce
  58  */
  59 void
  60 nb_kernel_ElecEw_VdwLJ_GeomW4P1_VF_sse4_1_double
  61                     (t_nblist                    * gmx_restrict       nlist,
  62                      rvec                        * gmx_restrict          xx,
  63                      rvec                        * gmx_restrict          ff,
  64                      t_forcerec                  * gmx_restrict          fr,
  65                      t_mdatoms                   * gmx_restrict     mdatoms,
  66                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
  67                      t_nrnb                      * gmx_restrict        nrnb)
  68 {
  69     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
  70      * just 0 for non-waters.
  71      * Suffixes A,B refer to j loop unrolling done with SSE double precision, e.g. for the two different
  72      * jnr indices corresponding to data put in the four positions in the SIMD register.
  73      */
  74     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
  75     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
  76     int              jnrA,jnrB;
  77     int              j_coord_offsetA,j_coord_offsetB;
  78     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
  79     real             rcutoff_scalar;
  80     real             *shiftvec,*fshift,*x,*f;
  81     __m128d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
  82     int              vdwioffset0;
  83     __m128d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
  84     int              vdwioffset1;
  85     __m128d          ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
  86     int              vdwioffset2;
  87     __m128d          ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
  88     int              vdwioffset3;
  89     __m128d          ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
  90     int              vdwjidx0A,vdwjidx0B;
  91     __m128d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  92     __m128d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  93     __m128d          dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
  94     __m128d          dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
  95     __m128d          dx30,dy30,dz30,rsq30,rinv30,rinvsq30,r30,qq30,c6_30,c12_30;
  96     __m128d          velec,felec,velecsum,facel,crf,krf,krf2;
  97     real             *charge;
  98     int              nvdwtype;
  99     __m128d          rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 100     int              *vdwtype;
 101     real             *vdwparam;
 102     __m128d          one_sixth   = _mm_set1_pd(1.0/6.0);
 103     __m128d          one_twelfth = _mm_set1_pd(1.0/12.0);
 104     __m128i          ewitab;
 105     __m128d          ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
 106     real             *ewtab;
 107     __m128d          dummy_mask,cutoff_mask;
 108     __m128d          signbit   = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
 109     __m128d          one     = _mm_set1_pd(1.0);
 110     __m128d          two     = _mm_set1_pd(2.0);
 111     x                = xx[0];
 112     f                = ff[0];
 113
 114     nri              = nlist->nri;
 115     iinr             = nlist->iinr;
 116     jindex           = nlist->jindex;
 117     jjnr             = nlist->jjnr;
 118     shiftidx         = nlist->shift;
 119     gid              = nlist->gid;
 120     shiftvec         = fr->shift_vec[0];
 121     fshift           = fr->fshift[0];
 122     facel            = _mm_set1_pd(fr->epsfac);
 123     charge           = mdatoms->chargeA;
 124     nvdwtype         = fr->ntype;
 125     vdwparam         = fr->nbfp;
 126     vdwtype          = mdatoms->typeA;
 127
 128     sh_ewald         = _mm_set1_pd(fr->ic->sh_ewald);
 129     ewtab            = fr->ic->tabq_coul_FDV0;
 130     ewtabscale       = _mm_set1_pd(fr->ic->tabq_scale);
 131     ewtabhalfspace   = _mm_set1_pd(0.5/fr->ic->tabq_scale);
 132
 133     /* Setup water-specific parameters */
 134     inr              = nlist->iinr[0];
 135     iq1              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+1]));
 136     iq2              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+2]));
 137     iq3              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+3]));
 138     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 139
 140     /* Avoid stupid compiler warnings */
 141     jnrA = jnrB = 0;
 142     j_coord_offsetA = 0;
 143     j_coord_offsetB = 0;
 144
 145     outeriter        = 0;
 146     inneriter        = 0;
 147
 148     /* Start outer loop over neighborlists */
 149     for(iidx=0; iidx<nri; iidx++)
 150     {
 151         /* Load shift vector for this list */
 152         i_shift_offset   = DIM*shiftidx[iidx];
 153
 154         /* Load limits for loop over neighbors */
 155         j_index_start    = jindex[iidx];
 156         j_index_end      = jindex[iidx+1];
 157
 158         /* Get outer coordinate index */
 159         inr              = iinr[iidx];
 160         i_coord_offset   = DIM*inr;
 161
 162         /* Load i particle coords and add shift vector */
 163         gmx_mm_load_shift_and_4rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
 164                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
 165
 166         fix0             = _mm_setzero_pd();
 167         fiy0             = _mm_setzero_pd();
 168         fiz0             = _mm_setzero_pd();
 169         fix1             = _mm_setzero_pd();
 170         fiy1             = _mm_setzero_pd();
 171         fiz1             = _mm_setzero_pd();
 172         fix2             = _mm_setzero_pd();
 173         fiy2             = _mm_setzero_pd();
 174         fiz2             = _mm_setzero_pd();
 175         fix3             = _mm_setzero_pd();
 176         fiy3             = _mm_setzero_pd();
 177         fiz3             = _mm_setzero_pd();
 178
 179         /* Reset potential sums */
 180         velecsum         = _mm_setzero_pd();
 181         vvdwsum          = _mm_setzero_pd();
 182
 183         /* Start inner kernel loop */
 184         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 185         {
 186
 187             /* Get j neighbor index, and coordinate index */
 188             jnrA             = jjnr[jidx];
 189             jnrB             = jjnr[jidx+1];
 190             j_coord_offsetA  = DIM*jnrA;
 191             j_coord_offsetB  = DIM*jnrB;
 192
 193             /* load j atom coordinates */
 194             gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 195                                               &jx0,&jy0,&jz0);
 196
 197             /* Calculate displacement vector */
 198             dx00             = _mm_sub_pd(ix0,jx0);
 199             dy00             = _mm_sub_pd(iy0,jy0);
 200             dz00             = _mm_sub_pd(iz0,jz0);
 201             dx10             = _mm_sub_pd(ix1,jx0);
 202             dy10             = _mm_sub_pd(iy1,jy0);
 203             dz10             = _mm_sub_pd(iz1,jz0);
 204             dx20             = _mm_sub_pd(ix2,jx0);
 205             dy20             = _mm_sub_pd(iy2,jy0);
 206             dz20             = _mm_sub_pd(iz2,jz0);
 207             dx30             = _mm_sub_pd(ix3,jx0);
 208             dy30             = _mm_sub_pd(iy3,jy0);
 209             dz30             = _mm_sub_pd(iz3,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             rsq30            = gmx_mm_calc_rsq_pd(dx30,dy30,dz30);
 216
 217             rinv10           = gmx_mm_invsqrt_pd(rsq10);
 218             rinv20           = gmx_mm_invsqrt_pd(rsq20);
 219             rinv30           = gmx_mm_invsqrt_pd(rsq30);
 220
 221             rinvsq00         = gmx_mm_inv_pd(rsq00);
 222             rinvsq10         = _mm_mul_pd(rinv10,rinv10);
 223             rinvsq20         = _mm_mul_pd(rinv20,rinv20);
 224             rinvsq30         = _mm_mul_pd(rinv30,rinv30);
 225
 226             /* Load parameters for j particles */
 227             jq0              = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
 228             vdwjidx0A        = 2*vdwtype[jnrA+0];
 229             vdwjidx0B        = 2*vdwtype[jnrB+0];
 230
 231             fjx0             = _mm_setzero_pd();
 232             fjy0             = _mm_setzero_pd();
 233             fjz0             = _mm_setzero_pd();
 234
 235             /**************************
 236              * CALCULATE INTERACTIONS *
 237              **************************/
 238
 239             /* Compute parameters for interactions between i and j atoms */
 240             gmx_mm_load_2pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,
 241                                          vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
 242
 243             /* LENNARD-JONES DISPERSION/REPULSION */
 244
 245             rinvsix          = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 246             vvdw6            = _mm_mul_pd(c6_00,rinvsix);
 247             vvdw12           = _mm_mul_pd(c12_00,_mm_mul_pd(rinvsix,rinvsix));
 248             vvdw             = _mm_sub_pd( _mm_mul_pd(vvdw12,one_twelfth) , _mm_mul_pd(vvdw6,one_sixth) );
 249             fvdw             = _mm_mul_pd(_mm_sub_pd(vvdw12,vvdw6),rinvsq00);
 250
 251             /* Update potential sum for this i atom from the interaction with this j atom. */
 252             vvdwsum          = _mm_add_pd(vvdwsum,vvdw);
 253
 254             fscal            = fvdw;
 255
 256             /* Calculate temporary vectorial force */
 257             tx               = _mm_mul_pd(fscal,dx00);
 258             ty               = _mm_mul_pd(fscal,dy00);
 259             tz               = _mm_mul_pd(fscal,dz00);
 260
 261             /* Update vectorial force */
 262             fix0             = _mm_add_pd(fix0,tx);
 263             fiy0             = _mm_add_pd(fiy0,ty);
 264             fiz0             = _mm_add_pd(fiz0,tz);
 265
 266             fjx0             = _mm_add_pd(fjx0,tx);
 267             fjy0             = _mm_add_pd(fjy0,ty);
 268             fjz0             = _mm_add_pd(fjz0,tz);
 269
 270             /**************************
 271              * CALCULATE INTERACTIONS *
 272              **************************/
 273
 274             r10              = _mm_mul_pd(rsq10,rinv10);
 275
 276             /* Compute parameters for interactions between i and j atoms */
 277             qq10             = _mm_mul_pd(iq1,jq0);
 278
 279             /* EWALD ELECTROSTATICS */
 280
 281             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 282             ewrt             = _mm_mul_pd(r10,ewtabscale);
 283             ewitab           = _mm_cvttpd_epi32(ewrt);
 284             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 285             ewitab           = _mm_slli_epi32(ewitab,2);
 286             ewtabF           = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 287             ewtabD           = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,1) );
 288             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 289             ewtabV           = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,0) +2);
 290             ewtabFn          = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,1) +2);
 291             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 292             felec            = _mm_add_pd(ewtabF,_mm_mul_pd(eweps,ewtabD));
 293             velec            = _mm_sub_pd(ewtabV,_mm_mul_pd(_mm_mul_pd(ewtabhalfspace,eweps),_mm_add_pd(ewtabF,felec)));
 294             velec            = _mm_mul_pd(qq10,_mm_sub_pd(rinv10,velec));
 295             felec            = _mm_mul_pd(_mm_mul_pd(qq10,rinv10),_mm_sub_pd(rinvsq10,felec));
 296
 297             /* Update potential sum for this i atom from the interaction with this j atom. */
 298             velecsum         = _mm_add_pd(velecsum,velec);
 299
 300             fscal            = felec;
 301
 302             /* Calculate temporary vectorial force */
 303             tx               = _mm_mul_pd(fscal,dx10);
 304             ty               = _mm_mul_pd(fscal,dy10);
 305             tz               = _mm_mul_pd(fscal,dz10);
 306
 307             /* Update vectorial force */
 308             fix1             = _mm_add_pd(fix1,tx);
 309             fiy1             = _mm_add_pd(fiy1,ty);
 310             fiz1             = _mm_add_pd(fiz1,tz);
 311
 312             fjx0             = _mm_add_pd(fjx0,tx);
 313             fjy0             = _mm_add_pd(fjy0,ty);
 314             fjz0             = _mm_add_pd(fjz0,tz);
 315
 316             /**************************
 317              * CALCULATE INTERACTIONS *
 318              **************************/
 319
 320             r20              = _mm_mul_pd(rsq20,rinv20);
 321
 322             /* Compute parameters for interactions between i and j atoms */
 323             qq20             = _mm_mul_pd(iq2,jq0);
 324
 325             /* EWALD ELECTROSTATICS */
 326
 327             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 328             ewrt             = _mm_mul_pd(r20,ewtabscale);
 329             ewitab           = _mm_cvttpd_epi32(ewrt);
 330             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 331             ewitab           = _mm_slli_epi32(ewitab,2);
 332             ewtabF           = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 333             ewtabD           = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,1) );
 334             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 335             ewtabV           = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,0) +2);
 336             ewtabFn          = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,1) +2);
 337             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 338             felec            = _mm_add_pd(ewtabF,_mm_mul_pd(eweps,ewtabD));
 339             velec            = _mm_sub_pd(ewtabV,_mm_mul_pd(_mm_mul_pd(ewtabhalfspace,eweps),_mm_add_pd(ewtabF,felec)));
 340             velec            = _mm_mul_pd(qq20,_mm_sub_pd(rinv20,velec));
 341             felec            = _mm_mul_pd(_mm_mul_pd(qq20,rinv20),_mm_sub_pd(rinvsq20,felec));
 342
 343             /* Update potential sum for this i atom from the interaction with this j atom. */
 344             velecsum         = _mm_add_pd(velecsum,velec);
 345
 346             fscal            = felec;
 347
 348             /* Calculate temporary vectorial force */
 349             tx               = _mm_mul_pd(fscal,dx20);
 350             ty               = _mm_mul_pd(fscal,dy20);
 351             tz               = _mm_mul_pd(fscal,dz20);
 352
 353             /* Update vectorial force */
 354             fix2             = _mm_add_pd(fix2,tx);
 355             fiy2             = _mm_add_pd(fiy2,ty);
 356             fiz2             = _mm_add_pd(fiz2,tz);
 357
 358             fjx0             = _mm_add_pd(fjx0,tx);
 359             fjy0             = _mm_add_pd(fjy0,ty);
 360             fjz0             = _mm_add_pd(fjz0,tz);
 361
 362             /**************************
 363              * CALCULATE INTERACTIONS *
 364              **************************/
 365
 366             r30              = _mm_mul_pd(rsq30,rinv30);
 367
 368             /* Compute parameters for interactions between i and j atoms */
 369             qq30             = _mm_mul_pd(iq3,jq0);
 370
 371             /* EWALD ELECTROSTATICS */
 372
 373             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 374             ewrt             = _mm_mul_pd(r30,ewtabscale);
 375             ewitab           = _mm_cvttpd_epi32(ewrt);
 376             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 377             ewitab           = _mm_slli_epi32(ewitab,2);
 378             ewtabF           = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 379             ewtabD           = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,1) );
 380             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 381             ewtabV           = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,0) +2);
 382             ewtabFn          = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,1) +2);
 383             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 384             felec            = _mm_add_pd(ewtabF,_mm_mul_pd(eweps,ewtabD));
 385             velec            = _mm_sub_pd(ewtabV,_mm_mul_pd(_mm_mul_pd(ewtabhalfspace,eweps),_mm_add_pd(ewtabF,felec)));
 386             velec            = _mm_mul_pd(qq30,_mm_sub_pd(rinv30,velec));
 387             felec            = _mm_mul_pd(_mm_mul_pd(qq30,rinv30),_mm_sub_pd(rinvsq30,felec));
 388
 389             /* Update potential sum for this i atom from the interaction with this j atom. */
 390             velecsum         = _mm_add_pd(velecsum,velec);
 391
 392             fscal            = felec;
 393
 394             /* Calculate temporary vectorial force */
 395             tx               = _mm_mul_pd(fscal,dx30);
 396             ty               = _mm_mul_pd(fscal,dy30);
 397             tz               = _mm_mul_pd(fscal,dz30);
 398
 399             /* Update vectorial force */
 400             fix3             = _mm_add_pd(fix3,tx);
 401             fiy3             = _mm_add_pd(fiy3,ty);
 402             fiz3             = _mm_add_pd(fiz3,tz);
 403
 404             fjx0             = _mm_add_pd(fjx0,tx);
 405             fjy0             = _mm_add_pd(fjy0,ty);
 406             fjz0             = _mm_add_pd(fjz0,tz);
 407
 408             gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,fjx0,fjy0,fjz0);
 409
 410             /* Inner loop uses 158 flops */
 411         }
 412
 413         if(jidx<j_index_end)
 414         {
 415
 416             jnrA             = jjnr[jidx];
 417             j_coord_offsetA  = DIM*jnrA;
 418
 419             /* load j atom coordinates */
 420             gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
 421                                               &jx0,&jy0,&jz0);
 422
 423             /* Calculate displacement vector */
 424             dx00             = _mm_sub_pd(ix0,jx0);
 425             dy00             = _mm_sub_pd(iy0,jy0);
 426             dz00             = _mm_sub_pd(iz0,jz0);
 427             dx10             = _mm_sub_pd(ix1,jx0);
 428             dy10             = _mm_sub_pd(iy1,jy0);
 429             dz10             = _mm_sub_pd(iz1,jz0);
 430             dx20             = _mm_sub_pd(ix2,jx0);
 431             dy20             = _mm_sub_pd(iy2,jy0);
 432             dz20             = _mm_sub_pd(iz2,jz0);
 433             dx30             = _mm_sub_pd(ix3,jx0);
 434             dy30             = _mm_sub_pd(iy3,jy0);
 435             dz30             = _mm_sub_pd(iz3,jz0);
 436
 437             /* Calculate squared distance and things based on it */
 438             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 439             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
 440             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
 441             rsq30            = gmx_mm_calc_rsq_pd(dx30,dy30,dz30);
 442
 443             rinv10           = gmx_mm_invsqrt_pd(rsq10);
 444             rinv20           = gmx_mm_invsqrt_pd(rsq20);
 445             rinv30           = gmx_mm_invsqrt_pd(rsq30);
 446
 447             rinvsq00         = gmx_mm_inv_pd(rsq00);
 448             rinvsq10         = _mm_mul_pd(rinv10,rinv10);
 449             rinvsq20         = _mm_mul_pd(rinv20,rinv20);
 450             rinvsq30         = _mm_mul_pd(rinv30,rinv30);
 451
 452             /* Load parameters for j particles */
 453             jq0              = _mm_load_sd(charge+jnrA+0);
 454             vdwjidx0A        = 2*vdwtype[jnrA+0];
 455
 456             fjx0             = _mm_setzero_pd();
 457             fjy0             = _mm_setzero_pd();
 458             fjz0             = _mm_setzero_pd();
 459
 460             /**************************
 461              * CALCULATE INTERACTIONS *
 462              **************************/
 463
 464             /* Compute parameters for interactions between i and j atoms */
 465             gmx_mm_load_1pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,&c6_00,&c12_00);
 466
 467             /* LENNARD-JONES DISPERSION/REPULSION */
 468
 469             rinvsix          = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 470             vvdw6            = _mm_mul_pd(c6_00,rinvsix);
 471             vvdw12           = _mm_mul_pd(c12_00,_mm_mul_pd(rinvsix,rinvsix));
 472             vvdw             = _mm_sub_pd( _mm_mul_pd(vvdw12,one_twelfth) , _mm_mul_pd(vvdw6,one_sixth) );
 473             fvdw             = _mm_mul_pd(_mm_sub_pd(vvdw12,vvdw6),rinvsq00);
 474
 475             /* Update potential sum for this i atom from the interaction with this j atom. */
 476             vvdw             = _mm_unpacklo_pd(vvdw,_mm_setzero_pd());
 477             vvdwsum          = _mm_add_pd(vvdwsum,vvdw);
 478
 479             fscal            = fvdw;
 480
 481             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 482
 483             /* Calculate temporary vectorial force */
 484             tx               = _mm_mul_pd(fscal,dx00);
 485             ty               = _mm_mul_pd(fscal,dy00);
 486             tz               = _mm_mul_pd(fscal,dz00);
 487
 488             /* Update vectorial force */
 489             fix0             = _mm_add_pd(fix0,tx);
 490             fiy0             = _mm_add_pd(fiy0,ty);
 491             fiz0             = _mm_add_pd(fiz0,tz);
 492
 493             fjx0             = _mm_add_pd(fjx0,tx);
 494             fjy0             = _mm_add_pd(fjy0,ty);
 495             fjz0             = _mm_add_pd(fjz0,tz);
 496
 497             /**************************
 498              * CALCULATE INTERACTIONS *
 499              **************************/
 500
 501             r10              = _mm_mul_pd(rsq10,rinv10);
 502
 503             /* Compute parameters for interactions between i and j atoms */
 504             qq10             = _mm_mul_pd(iq1,jq0);
 505
 506             /* EWALD ELECTROSTATICS */
 507
 508             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 509             ewrt             = _mm_mul_pd(r10,ewtabscale);
 510             ewitab           = _mm_cvttpd_epi32(ewrt);
 511             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 512             ewitab           = _mm_slli_epi32(ewitab,2);
 513             ewtabF           = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 514             ewtabD           = _mm_setzero_pd();
 515             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 516             ewtabV           = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,0) +2);
 517             ewtabFn          = _mm_setzero_pd();
 518             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 519             felec            = _mm_add_pd(ewtabF,_mm_mul_pd(eweps,ewtabD));
 520             velec            = _mm_sub_pd(ewtabV,_mm_mul_pd(_mm_mul_pd(ewtabhalfspace,eweps),_mm_add_pd(ewtabF,felec)));
 521             velec            = _mm_mul_pd(qq10,_mm_sub_pd(rinv10,velec));
 522             felec            = _mm_mul_pd(_mm_mul_pd(qq10,rinv10),_mm_sub_pd(rinvsq10,felec));
 523
 524             /* Update potential sum for this i atom from the interaction with this j atom. */
 525             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 526             velecsum         = _mm_add_pd(velecsum,velec);
 527
 528             fscal            = felec;
 529
 530             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 531
 532             /* Calculate temporary vectorial force */
 533             tx               = _mm_mul_pd(fscal,dx10);
 534             ty               = _mm_mul_pd(fscal,dy10);
 535             tz               = _mm_mul_pd(fscal,dz10);
 536
 537             /* Update vectorial force */
 538             fix1             = _mm_add_pd(fix1,tx);
 539             fiy1             = _mm_add_pd(fiy1,ty);
 540             fiz1             = _mm_add_pd(fiz1,tz);
 541
 542             fjx0             = _mm_add_pd(fjx0,tx);
 543             fjy0             = _mm_add_pd(fjy0,ty);
 544             fjz0             = _mm_add_pd(fjz0,tz);
 545
 546             /**************************
 547              * CALCULATE INTERACTIONS *
 548              **************************/
 549
 550             r20              = _mm_mul_pd(rsq20,rinv20);
 551
 552             /* Compute parameters for interactions between i and j atoms */
 553             qq20             = _mm_mul_pd(iq2,jq0);
 554
 555             /* EWALD ELECTROSTATICS */
 556
 557             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 558             ewrt             = _mm_mul_pd(r20,ewtabscale);
 559             ewitab           = _mm_cvttpd_epi32(ewrt);
 560             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 561             ewitab           = _mm_slli_epi32(ewitab,2);
 562             ewtabF           = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 563             ewtabD           = _mm_setzero_pd();
 564             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 565             ewtabV           = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,0) +2);
 566             ewtabFn          = _mm_setzero_pd();
 567             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 568             felec            = _mm_add_pd(ewtabF,_mm_mul_pd(eweps,ewtabD));
 569             velec            = _mm_sub_pd(ewtabV,_mm_mul_pd(_mm_mul_pd(ewtabhalfspace,eweps),_mm_add_pd(ewtabF,felec)));
 570             velec            = _mm_mul_pd(qq20,_mm_sub_pd(rinv20,velec));
 571             felec            = _mm_mul_pd(_mm_mul_pd(qq20,rinv20),_mm_sub_pd(rinvsq20,felec));
 572
 573             /* Update potential sum for this i atom from the interaction with this j atom. */
 574             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 575             velecsum         = _mm_add_pd(velecsum,velec);
 576
 577             fscal            = felec;
 578
 579             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 580
 581             /* Calculate temporary vectorial force */
 582             tx               = _mm_mul_pd(fscal,dx20);
 583             ty               = _mm_mul_pd(fscal,dy20);
 584             tz               = _mm_mul_pd(fscal,dz20);
 585
 586             /* Update vectorial force */
 587             fix2             = _mm_add_pd(fix2,tx);
 588             fiy2             = _mm_add_pd(fiy2,ty);
 589             fiz2             = _mm_add_pd(fiz2,tz);
 590
 591             fjx0             = _mm_add_pd(fjx0,tx);
 592             fjy0             = _mm_add_pd(fjy0,ty);
 593             fjz0             = _mm_add_pd(fjz0,tz);
 594
 595             /**************************
 596              * CALCULATE INTERACTIONS *
 597              **************************/
 598
 599             r30              = _mm_mul_pd(rsq30,rinv30);
 600
 601             /* Compute parameters for interactions between i and j atoms */
 602             qq30             = _mm_mul_pd(iq3,jq0);
 603
 604             /* EWALD ELECTROSTATICS */
 605
 606             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 607             ewrt             = _mm_mul_pd(r30,ewtabscale);
 608             ewitab           = _mm_cvttpd_epi32(ewrt);
 609             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 610             ewitab           = _mm_slli_epi32(ewitab,2);
 611             ewtabF           = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 612             ewtabD           = _mm_setzero_pd();
 613             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 614             ewtabV           = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,0) +2);
 615             ewtabFn          = _mm_setzero_pd();
 616             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 617             felec            = _mm_add_pd(ewtabF,_mm_mul_pd(eweps,ewtabD));
 618             velec            = _mm_sub_pd(ewtabV,_mm_mul_pd(_mm_mul_pd(ewtabhalfspace,eweps),_mm_add_pd(ewtabF,felec)));
 619             velec            = _mm_mul_pd(qq30,_mm_sub_pd(rinv30,velec));
 620             felec            = _mm_mul_pd(_mm_mul_pd(qq30,rinv30),_mm_sub_pd(rinvsq30,felec));
 621
 622             /* Update potential sum for this i atom from the interaction with this j atom. */
 623             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 624             velecsum         = _mm_add_pd(velecsum,velec);
 625
 626             fscal            = felec;
 627
 628             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 629
 630             /* Calculate temporary vectorial force */
 631             tx               = _mm_mul_pd(fscal,dx30);
 632             ty               = _mm_mul_pd(fscal,dy30);
 633             tz               = _mm_mul_pd(fscal,dz30);
 634
 635             /* Update vectorial force */
 636             fix3             = _mm_add_pd(fix3,tx);
 637             fiy3             = _mm_add_pd(fiy3,ty);
 638             fiz3             = _mm_add_pd(fiz3,tz);
 639
 640             fjx0             = _mm_add_pd(fjx0,tx);
 641             fjy0             = _mm_add_pd(fjy0,ty);
 642             fjz0             = _mm_add_pd(fjz0,tz);
 643
 644             gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,fjx0,fjy0,fjz0);
 645
 646             /* Inner loop uses 158 flops */
 647         }
 648
 649         /* End of innermost loop */
 650
 651         gmx_mm_update_iforce_4atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
 652                                               f+i_coord_offset,fshift+i_shift_offset);
 653
 654         ggid                        = gid[iidx];
 655         /* Update potential energies */
 656         gmx_mm_update_1pot_pd(velecsum,kernel_data->energygrp_elec+ggid);
 657         gmx_mm_update_1pot_pd(vvdwsum,kernel_data->energygrp_vdw+ggid);
 658
 659         /* Increment number of inner iterations */
 660         inneriter                  += j_index_end - j_index_start;
 661
 662         /* Outer loop uses 26 flops */
 663     }
 664
 665     /* Increment number of outer iterations */
 666     outeriter        += nri;
 667
 668     /* Update outer/inner flops */
 669
 670     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W4_VF,outeriter*26 + inneriter*158);
 671 }
 672 /*
 673  * Gromacs nonbonded kernel:   nb_kernel_ElecEw_VdwLJ_GeomW4P1_F_sse4_1_double
 674  * Electrostatics interaction: Ewald
 675  * VdW interaction:            LennardJones
 676  * Geometry:                   Water4-Particle
 677  * Calculate force/pot:        Force
 678  */
 679 void
 680 nb_kernel_ElecEw_VdwLJ_GeomW4P1_F_sse4_1_double
 681                     (t_nblist                    * gmx_restrict       nlist,
 682                      rvec                        * gmx_restrict          xx,
 683                      rvec                        * gmx_restrict          ff,
 684                      t_forcerec                  * gmx_restrict          fr,
 685                      t_mdatoms                   * gmx_restrict     mdatoms,
 686                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 687                      t_nrnb                      * gmx_restrict        nrnb)
 688 {
 689     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 690      * just 0 for non-waters.
 691      * Suffixes A,B refer to j loop unrolling done with SSE double precision, e.g. for the two different
 692      * jnr indices corresponding to data put in the four positions in the SIMD register.
 693      */
 694     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 695     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 696     int              jnrA,jnrB;
 697     int              j_coord_offsetA,j_coord_offsetB;
 698     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 699     real             rcutoff_scalar;
 700     real             *shiftvec,*fshift,*x,*f;
 701     __m128d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 702     int              vdwioffset0;
 703     __m128d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 704     int              vdwioffset1;
 705     __m128d          ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
 706     int              vdwioffset2;
 707     __m128d          ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
 708     int              vdwioffset3;
 709     __m128d          ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
 710     int              vdwjidx0A,vdwjidx0B;
 711     __m128d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 712     __m128d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 713     __m128d          dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
 714     __m128d          dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
 715     __m128d          dx30,dy30,dz30,rsq30,rinv30,rinvsq30,r30,qq30,c6_30,c12_30;
 716     __m128d          velec,felec,velecsum,facel,crf,krf,krf2;
 717     real             *charge;
 718     int              nvdwtype;
 719     __m128d          rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 720     int              *vdwtype;
 721     real             *vdwparam;
 722     __m128d          one_sixth   = _mm_set1_pd(1.0/6.0);
 723     __m128d          one_twelfth = _mm_set1_pd(1.0/12.0);
 724     __m128i          ewitab;
 725     __m128d          ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
 726     real             *ewtab;
 727     __m128d          dummy_mask,cutoff_mask;
 728     __m128d          signbit   = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
 729     __m128d          one     = _mm_set1_pd(1.0);
 730     __m128d          two     = _mm_set1_pd(2.0);
 731     x                = xx[0];
 732     f                = ff[0];
 733
 734     nri              = nlist->nri;
 735     iinr             = nlist->iinr;
 736     jindex           = nlist->jindex;
 737     jjnr             = nlist->jjnr;
 738     shiftidx         = nlist->shift;
 739     gid              = nlist->gid;
 740     shiftvec         = fr->shift_vec[0];
 741     fshift           = fr->fshift[0];
 742     facel            = _mm_set1_pd(fr->epsfac);
 743     charge           = mdatoms->chargeA;
 744     nvdwtype         = fr->ntype;
 745     vdwparam         = fr->nbfp;
 746     vdwtype          = mdatoms->typeA;
 747
 748     sh_ewald         = _mm_set1_pd(fr->ic->sh_ewald);
 749     ewtab            = fr->ic->tabq_coul_F;
 750     ewtabscale       = _mm_set1_pd(fr->ic->tabq_scale);
 751     ewtabhalfspace   = _mm_set1_pd(0.5/fr->ic->tabq_scale);
 752
 753     /* Setup water-specific parameters */
 754     inr              = nlist->iinr[0];
 755     iq1              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+1]));
 756     iq2              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+2]));
 757     iq3              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+3]));
 758     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 759
 760     /* Avoid stupid compiler warnings */
 761     jnrA = jnrB = 0;
 762     j_coord_offsetA = 0;
 763     j_coord_offsetB = 0;
 764
 765     outeriter        = 0;
 766     inneriter        = 0;
 767
 768     /* Start outer loop over neighborlists */
 769     for(iidx=0; iidx<nri; iidx++)
 770     {
 771         /* Load shift vector for this list */
 772         i_shift_offset   = DIM*shiftidx[iidx];
 773
 774         /* Load limits for loop over neighbors */
 775         j_index_start    = jindex[iidx];
 776         j_index_end      = jindex[iidx+1];
 777
 778         /* Get outer coordinate index */
 779         inr              = iinr[iidx];
 780         i_coord_offset   = DIM*inr;
 781
 782         /* Load i particle coords and add shift vector */
 783         gmx_mm_load_shift_and_4rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
 784                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
 785
 786         fix0             = _mm_setzero_pd();
 787         fiy0             = _mm_setzero_pd();
 788         fiz0             = _mm_setzero_pd();
 789         fix1             = _mm_setzero_pd();
 790         fiy1             = _mm_setzero_pd();
 791         fiz1             = _mm_setzero_pd();
 792         fix2             = _mm_setzero_pd();
 793         fiy2             = _mm_setzero_pd();
 794         fiz2             = _mm_setzero_pd();
 795         fix3             = _mm_setzero_pd();
 796         fiy3             = _mm_setzero_pd();
 797         fiz3             = _mm_setzero_pd();
 798
 799         /* Start inner kernel loop */
 800         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 801         {
 802
 803             /* Get j neighbor index, and coordinate index */
 804             jnrA             = jjnr[jidx];
 805             jnrB             = jjnr[jidx+1];
 806             j_coord_offsetA  = DIM*jnrA;
 807             j_coord_offsetB  = DIM*jnrB;
 808
 809             /* load j atom coordinates */
 810             gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 811                                               &jx0,&jy0,&jz0);
 812
 813             /* Calculate displacement vector */
 814             dx00             = _mm_sub_pd(ix0,jx0);
 815             dy00             = _mm_sub_pd(iy0,jy0);
 816             dz00             = _mm_sub_pd(iz0,jz0);
 817             dx10             = _mm_sub_pd(ix1,jx0);
 818             dy10             = _mm_sub_pd(iy1,jy0);
 819             dz10             = _mm_sub_pd(iz1,jz0);
 820             dx20             = _mm_sub_pd(ix2,jx0);
 821             dy20             = _mm_sub_pd(iy2,jy0);
 822             dz20             = _mm_sub_pd(iz2,jz0);
 823             dx30             = _mm_sub_pd(ix3,jx0);
 824             dy30             = _mm_sub_pd(iy3,jy0);
 825             dz30             = _mm_sub_pd(iz3,jz0);
 826
 827             /* Calculate squared distance and things based on it */
 828             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 829             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
 830             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
 831             rsq30            = gmx_mm_calc_rsq_pd(dx30,dy30,dz30);
 832
 833             rinv10           = gmx_mm_invsqrt_pd(rsq10);
 834             rinv20           = gmx_mm_invsqrt_pd(rsq20);
 835             rinv30           = gmx_mm_invsqrt_pd(rsq30);
 836
 837             rinvsq00         = gmx_mm_inv_pd(rsq00);
 838             rinvsq10         = _mm_mul_pd(rinv10,rinv10);
 839             rinvsq20         = _mm_mul_pd(rinv20,rinv20);
 840             rinvsq30         = _mm_mul_pd(rinv30,rinv30);
 841
 842             /* Load parameters for j particles */
 843             jq0              = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
 844             vdwjidx0A        = 2*vdwtype[jnrA+0];
 845             vdwjidx0B        = 2*vdwtype[jnrB+0];
 846
 847             fjx0             = _mm_setzero_pd();
 848             fjy0             = _mm_setzero_pd();
 849             fjz0             = _mm_setzero_pd();
 850
 851             /**************************
 852              * CALCULATE INTERACTIONS *
 853              **************************/
 854
 855             /* Compute parameters for interactions between i and j atoms */
 856             gmx_mm_load_2pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,
 857                                          vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
 858
 859             /* LENNARD-JONES DISPERSION/REPULSION */
 860
 861             rinvsix          = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 862             fvdw             = _mm_mul_pd(_mm_sub_pd(_mm_mul_pd(c12_00,rinvsix),c6_00),_mm_mul_pd(rinvsix,rinvsq00));
 863
 864             fscal            = fvdw;
 865
 866             /* Calculate temporary vectorial force */
 867             tx               = _mm_mul_pd(fscal,dx00);
 868             ty               = _mm_mul_pd(fscal,dy00);
 869             tz               = _mm_mul_pd(fscal,dz00);
 870
 871             /* Update vectorial force */
 872             fix0             = _mm_add_pd(fix0,tx);
 873             fiy0             = _mm_add_pd(fiy0,ty);
 874             fiz0             = _mm_add_pd(fiz0,tz);
 875
 876             fjx0             = _mm_add_pd(fjx0,tx);
 877             fjy0             = _mm_add_pd(fjy0,ty);
 878             fjz0             = _mm_add_pd(fjz0,tz);
 879
 880             /**************************
 881              * CALCULATE INTERACTIONS *
 882              **************************/
 883
 884             r10              = _mm_mul_pd(rsq10,rinv10);
 885
 886             /* Compute parameters for interactions between i and j atoms */
 887             qq10             = _mm_mul_pd(iq1,jq0);
 888
 889             /* EWALD ELECTROSTATICS */
 890
 891             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 892             ewrt             = _mm_mul_pd(r10,ewtabscale);
 893             ewitab           = _mm_cvttpd_epi32(ewrt);
 894             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 895             gmx_mm_load_2pair_swizzle_pd(ewtab+gmx_mm_extract_epi32(ewitab,0),ewtab+gmx_mm_extract_epi32(ewitab,1),
 896                                          &ewtabF,&ewtabFn);
 897             felec            = _mm_add_pd(_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF),_mm_mul_pd(eweps,ewtabFn));
 898             felec            = _mm_mul_pd(_mm_mul_pd(qq10,rinv10),_mm_sub_pd(rinvsq10,felec));
 899
 900             fscal            = felec;
 901
 902             /* Calculate temporary vectorial force */
 903             tx               = _mm_mul_pd(fscal,dx10);
 904             ty               = _mm_mul_pd(fscal,dy10);
 905             tz               = _mm_mul_pd(fscal,dz10);
 906
 907             /* Update vectorial force */
 908             fix1             = _mm_add_pd(fix1,tx);
 909             fiy1             = _mm_add_pd(fiy1,ty);
 910             fiz1             = _mm_add_pd(fiz1,tz);
 911
 912             fjx0             = _mm_add_pd(fjx0,tx);
 913             fjy0             = _mm_add_pd(fjy0,ty);
 914             fjz0             = _mm_add_pd(fjz0,tz);
 915
 916             /**************************
 917              * CALCULATE INTERACTIONS *
 918              **************************/
 919
 920             r20              = _mm_mul_pd(rsq20,rinv20);
 921
 922             /* Compute parameters for interactions between i and j atoms */
 923             qq20             = _mm_mul_pd(iq2,jq0);
 924
 925             /* EWALD ELECTROSTATICS */
 926
 927             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 928             ewrt             = _mm_mul_pd(r20,ewtabscale);
 929             ewitab           = _mm_cvttpd_epi32(ewrt);
 930             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 931             gmx_mm_load_2pair_swizzle_pd(ewtab+gmx_mm_extract_epi32(ewitab,0),ewtab+gmx_mm_extract_epi32(ewitab,1),
 932                                          &ewtabF,&ewtabFn);
 933             felec            = _mm_add_pd(_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF),_mm_mul_pd(eweps,ewtabFn));
 934             felec            = _mm_mul_pd(_mm_mul_pd(qq20,rinv20),_mm_sub_pd(rinvsq20,felec));
 935
 936             fscal            = felec;
 937
 938             /* Calculate temporary vectorial force */
 939             tx               = _mm_mul_pd(fscal,dx20);
 940             ty               = _mm_mul_pd(fscal,dy20);
 941             tz               = _mm_mul_pd(fscal,dz20);
 942
 943             /* Update vectorial force */
 944             fix2             = _mm_add_pd(fix2,tx);
 945             fiy2             = _mm_add_pd(fiy2,ty);
 946             fiz2             = _mm_add_pd(fiz2,tz);
 947
 948             fjx0             = _mm_add_pd(fjx0,tx);
 949             fjy0             = _mm_add_pd(fjy0,ty);
 950             fjz0             = _mm_add_pd(fjz0,tz);
 951
 952             /**************************
 953              * CALCULATE INTERACTIONS *
 954              **************************/
 955
 956             r30              = _mm_mul_pd(rsq30,rinv30);
 957
 958             /* Compute parameters for interactions between i and j atoms */
 959             qq30             = _mm_mul_pd(iq3,jq0);
 960
 961             /* EWALD ELECTROSTATICS */
 962
 963             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 964             ewrt             = _mm_mul_pd(r30,ewtabscale);
 965             ewitab           = _mm_cvttpd_epi32(ewrt);
 966             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 967             gmx_mm_load_2pair_swizzle_pd(ewtab+gmx_mm_extract_epi32(ewitab,0),ewtab+gmx_mm_extract_epi32(ewitab,1),
 968                                          &ewtabF,&ewtabFn);
 969             felec            = _mm_add_pd(_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF),_mm_mul_pd(eweps,ewtabFn));
 970             felec            = _mm_mul_pd(_mm_mul_pd(qq30,rinv30),_mm_sub_pd(rinvsq30,felec));
 971
 972             fscal            = felec;
 973
 974             /* Calculate temporary vectorial force */
 975             tx               = _mm_mul_pd(fscal,dx30);
 976             ty               = _mm_mul_pd(fscal,dy30);
 977             tz               = _mm_mul_pd(fscal,dz30);
 978
 979             /* Update vectorial force */
 980             fix3             = _mm_add_pd(fix3,tx);
 981             fiy3             = _mm_add_pd(fiy3,ty);
 982             fiz3             = _mm_add_pd(fiz3,tz);
 983
 984             fjx0             = _mm_add_pd(fjx0,tx);
 985             fjy0             = _mm_add_pd(fjy0,ty);
 986             fjz0             = _mm_add_pd(fjz0,tz);
 987
 988             gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,fjx0,fjy0,fjz0);
 989
 990             /* Inner loop uses 138 flops */
 991         }
 992
 993         if(jidx<j_index_end)
 994         {
 995
 996             jnrA             = jjnr[jidx];
 997             j_coord_offsetA  = DIM*jnrA;
 998
 999             /* load j atom coordinates */
1000             gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
1001                                               &jx0,&jy0,&jz0);
1002
1003             /* Calculate displacement vector */
1004             dx00             = _mm_sub_pd(ix0,jx0);
1005             dy00             = _mm_sub_pd(iy0,jy0);
1006             dz00             = _mm_sub_pd(iz0,jz0);
1007             dx10             = _mm_sub_pd(ix1,jx0);
1008             dy10             = _mm_sub_pd(iy1,jy0);
1009             dz10             = _mm_sub_pd(iz1,jz0);
1010             dx20             = _mm_sub_pd(ix2,jx0);
1011             dy20             = _mm_sub_pd(iy2,jy0);
1012             dz20             = _mm_sub_pd(iz2,jz0);
1013             dx30             = _mm_sub_pd(ix3,jx0);
1014             dy30             = _mm_sub_pd(iy3,jy0);
1015             dz30             = _mm_sub_pd(iz3,jz0);
1016
1017             /* Calculate squared distance and things based on it */
1018             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
1019             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
1020             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
1021             rsq30            = gmx_mm_calc_rsq_pd(dx30,dy30,dz30);
1022
1023             rinv10           = gmx_mm_invsqrt_pd(rsq10);
1024             rinv20           = gmx_mm_invsqrt_pd(rsq20);
1025             rinv30           = gmx_mm_invsqrt_pd(rsq30);
1026
1027             rinvsq00         = gmx_mm_inv_pd(rsq00);
1028             rinvsq10         = _mm_mul_pd(rinv10,rinv10);
1029             rinvsq20         = _mm_mul_pd(rinv20,rinv20);
1030             rinvsq30         = _mm_mul_pd(rinv30,rinv30);
1031
1032             /* Load parameters for j particles */
1033             jq0              = _mm_load_sd(charge+jnrA+0);
1034             vdwjidx0A        = 2*vdwtype[jnrA+0];
1035
1036             fjx0             = _mm_setzero_pd();
1037             fjy0             = _mm_setzero_pd();
1038             fjz0             = _mm_setzero_pd();
1039
1040             /**************************
1041              * CALCULATE INTERACTIONS *
1042              **************************/
1043
1044             /* Compute parameters for interactions between i and j atoms */
1045             gmx_mm_load_1pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,&c6_00,&c12_00);
1046
1047             /* LENNARD-JONES DISPERSION/REPULSION */
1048
1049             rinvsix          = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
1050             fvdw             = _mm_mul_pd(_mm_sub_pd(_mm_mul_pd(c12_00,rinvsix),c6_00),_mm_mul_pd(rinvsix,rinvsq00));
1051
1052             fscal            = fvdw;
1053
1054             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1055
1056             /* Calculate temporary vectorial force */
1057             tx               = _mm_mul_pd(fscal,dx00);
1058             ty               = _mm_mul_pd(fscal,dy00);
1059             tz               = _mm_mul_pd(fscal,dz00);
1060
1061             /* Update vectorial force */
1062             fix0             = _mm_add_pd(fix0,tx);
1063             fiy0             = _mm_add_pd(fiy0,ty);
1064             fiz0             = _mm_add_pd(fiz0,tz);
1065
1066             fjx0             = _mm_add_pd(fjx0,tx);
1067             fjy0             = _mm_add_pd(fjy0,ty);
1068             fjz0             = _mm_add_pd(fjz0,tz);
1069
1070             /**************************
1071              * CALCULATE INTERACTIONS *
1072              **************************/
1073
1074             r10              = _mm_mul_pd(rsq10,rinv10);
1075
1076             /* Compute parameters for interactions between i and j atoms */
1077             qq10             = _mm_mul_pd(iq1,jq0);
1078
1079             /* EWALD ELECTROSTATICS */
1080
1081             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1082             ewrt             = _mm_mul_pd(r10,ewtabscale);
1083             ewitab           = _mm_cvttpd_epi32(ewrt);
1084             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
1085             gmx_mm_load_1pair_swizzle_pd(ewtab+gmx_mm_extract_epi32(ewitab,0),&ewtabF,&ewtabFn);
1086             felec            = _mm_add_pd(_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF),_mm_mul_pd(eweps,ewtabFn));
1087             felec            = _mm_mul_pd(_mm_mul_pd(qq10,rinv10),_mm_sub_pd(rinvsq10,felec));
1088
1089             fscal            = felec;
1090
1091             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1092
1093             /* Calculate temporary vectorial force */
1094             tx               = _mm_mul_pd(fscal,dx10);
1095             ty               = _mm_mul_pd(fscal,dy10);
1096             tz               = _mm_mul_pd(fscal,dz10);
1097
1098             /* Update vectorial force */
1099             fix1             = _mm_add_pd(fix1,tx);
1100             fiy1             = _mm_add_pd(fiy1,ty);
1101             fiz1             = _mm_add_pd(fiz1,tz);
1102
1103             fjx0             = _mm_add_pd(fjx0,tx);
1104             fjy0             = _mm_add_pd(fjy0,ty);
1105             fjz0             = _mm_add_pd(fjz0,tz);
1106
1107             /**************************
1108              * CALCULATE INTERACTIONS *
1109              **************************/
1110
1111             r20              = _mm_mul_pd(rsq20,rinv20);
1112
1113             /* Compute parameters for interactions between i and j atoms */
1114             qq20             = _mm_mul_pd(iq2,jq0);
1115
1116             /* EWALD ELECTROSTATICS */
1117
1118             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1119             ewrt             = _mm_mul_pd(r20,ewtabscale);
1120             ewitab           = _mm_cvttpd_epi32(ewrt);
1121             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
1122             gmx_mm_load_1pair_swizzle_pd(ewtab+gmx_mm_extract_epi32(ewitab,0),&ewtabF,&ewtabFn);
1123             felec            = _mm_add_pd(_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF),_mm_mul_pd(eweps,ewtabFn));
1124             felec            = _mm_mul_pd(_mm_mul_pd(qq20,rinv20),_mm_sub_pd(rinvsq20,felec));
1125
1126             fscal            = felec;
1127
1128             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1129
1130             /* Calculate temporary vectorial force */
1131             tx               = _mm_mul_pd(fscal,dx20);
1132             ty               = _mm_mul_pd(fscal,dy20);
1133             tz               = _mm_mul_pd(fscal,dz20);
1134
1135             /* Update vectorial force */
1136             fix2             = _mm_add_pd(fix2,tx);
1137             fiy2             = _mm_add_pd(fiy2,ty);
1138             fiz2             = _mm_add_pd(fiz2,tz);
1139
1140             fjx0             = _mm_add_pd(fjx0,tx);
1141             fjy0             = _mm_add_pd(fjy0,ty);
1142             fjz0             = _mm_add_pd(fjz0,tz);
1143
1144             /**************************
1145              * CALCULATE INTERACTIONS *
1146              **************************/
1147
1148             r30              = _mm_mul_pd(rsq30,rinv30);
1149
1150             /* Compute parameters for interactions between i and j atoms */
1151             qq30             = _mm_mul_pd(iq3,jq0);
1152
1153             /* EWALD ELECTROSTATICS */
1154
1155             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1156             ewrt             = _mm_mul_pd(r30,ewtabscale);
1157             ewitab           = _mm_cvttpd_epi32(ewrt);
1158             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
1159             gmx_mm_load_1pair_swizzle_pd(ewtab+gmx_mm_extract_epi32(ewitab,0),&ewtabF,&ewtabFn);
1160             felec            = _mm_add_pd(_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF),_mm_mul_pd(eweps,ewtabFn));
1161             felec            = _mm_mul_pd(_mm_mul_pd(qq30,rinv30),_mm_sub_pd(rinvsq30,felec));
1162
1163             fscal            = felec;
1164
1165             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1166
1167             /* Calculate temporary vectorial force */
1168             tx               = _mm_mul_pd(fscal,dx30);
1169             ty               = _mm_mul_pd(fscal,dy30);
1170             tz               = _mm_mul_pd(fscal,dz30);
1171
1172             /* Update vectorial force */
1173             fix3             = _mm_add_pd(fix3,tx);
1174             fiy3             = _mm_add_pd(fiy3,ty);
1175             fiz3             = _mm_add_pd(fiz3,tz);
1176
1177             fjx0             = _mm_add_pd(fjx0,tx);
1178             fjy0             = _mm_add_pd(fjy0,ty);
1179             fjz0             = _mm_add_pd(fjz0,tz);
1180
1181             gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,fjx0,fjy0,fjz0);
1182
1183             /* Inner loop uses 138 flops */
1184         }
1185
1186         /* End of innermost loop */
1187
1188         gmx_mm_update_iforce_4atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
1189                                               f+i_coord_offset,fshift+i_shift_offset);
1190
1191         /* Increment number of inner iterations */
1192         inneriter                  += j_index_end - j_index_start;
1193
1194         /* Outer loop uses 24 flops */
1195     }
1196
1197     /* Increment number of outer iterations */
1198     outeriter        += nri;
1199
1200     /* Update outer/inner flops */
1201
1202     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W4_F,outeriter*24 + inneriter*138);
1203 }