src/gromacs/gmxlib/nonbonded/nb_kernel_sse2_double/nb_kernel_ElecEwSh_VdwLJSh_GeomW4P1_sse2_double.c

   1 /*
   2  * This file is part of the GROMACS molecular simulation package.
   3  *
   4  * Copyright (c) 2012,2013,2014, by the GROMACS development team, led by
   5  * Mark Abraham, David van der Spoel, Berk Hess, and Erik Lindahl,
   6  * and including many others, as listed in the AUTHORS file in the
   7  * top-level source directory and at http://www.gromacs.org.
   8  *
   9  * GROMACS is free software; you can redistribute it and/or
  10  * modify it under the terms of the GNU Lesser General Public License
  11  * as published by the Free Software Foundation; either version 2.1
  12  * of the License, or (at your option) any later version.
  13  *
  14  * GROMACS is distributed in the hope that it will be useful,
  15  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  16  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  17  * Lesser General Public License for more details.
  18  *
  19  * You should have received a copy of the GNU Lesser General Public
  20  * License along with GROMACS; if not, see
  21  * http://www.gnu.org/licenses, or write to the Free Software Foundation,
  22  * Inc., 51 Franklin Street, Fifth Floor, Boston, MA  02110-1301  USA.
  23  *
  24  * If you want to redistribute modifications to GROMACS, please
  25  * consider that scientific software is very special. Version
  26  * control is crucial - bugs must be traceable. We will be happy to
  27  * consider code for inclusion in the official distribution, but
  28  * derived work must not be called official GROMACS. Details are found
  29  * in the README & COPYING files - if they are missing, get the
  30  * official version at http://www.gromacs.org.
  31  *
  32  * To help us fund GROMACS development, we humbly ask that you cite
  33  * the research papers on the package. Check out http://www.gromacs.org.
  34  */
  35 /*
  36  * Note: this file was generated by the GROMACS sse2_double kernel generator.
  37  */
  38 #include "config.h"
  39
  40 #include <math.h>
  41
  42 #include "../nb_kernel.h"
  43 #include "gromacs/legacyheaders/types/simple.h"
  44 #include "gromacs/math/vec.h"
  45 #include "gromacs/legacyheaders/nrnb.h"
  46
  47 #include "gromacs/simd/math_x86_sse2_double.h"
  48 #include "kernelutil_x86_sse2_double.h"
  49
  50 /*
  51  * Gromacs nonbonded kernel:   nb_kernel_ElecEwSh_VdwLJSh_GeomW4P1_VF_sse2_double
  52  * Electrostatics interaction: Ewald
  53  * VdW interaction:            LennardJones
  54  * Geometry:                   Water4-Particle
  55  * Calculate force/pot:        PotentialAndForce
  56  */
  57 void
  58 nb_kernel_ElecEwSh_VdwLJSh_GeomW4P1_VF_sse2_double
  59                     (t_nblist                    * gmx_restrict       nlist,
  60                      rvec                        * gmx_restrict          xx,
  61                      rvec                        * gmx_restrict          ff,
  62                      t_forcerec                  * gmx_restrict          fr,
  63                      t_mdatoms                   * gmx_restrict     mdatoms,
  64                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
  65                      t_nrnb                      * gmx_restrict        nrnb)
  66 {
  67     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
  68      * just 0 for non-waters.
  69      * Suffixes A,B refer to j loop unrolling done with SSE double precision, e.g. for the two different
  70      * jnr indices corresponding to data put in the four positions in the SIMD register.
  71      */
  72     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
  73     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
  74     int              jnrA,jnrB;
  75     int              j_coord_offsetA,j_coord_offsetB;
  76     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
  77     real             rcutoff_scalar;
  78     real             *shiftvec,*fshift,*x,*f;
  79     __m128d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
  80     int              vdwioffset0;
  81     __m128d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
  82     int              vdwioffset1;
  83     __m128d          ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
  84     int              vdwioffset2;
  85     __m128d          ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
  86     int              vdwioffset3;
  87     __m128d          ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
  88     int              vdwjidx0A,vdwjidx0B;
  89     __m128d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  90     __m128d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  91     __m128d          dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
  92     __m128d          dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
  93     __m128d          dx30,dy30,dz30,rsq30,rinv30,rinvsq30,r30,qq30,c6_30,c12_30;
  94     __m128d          velec,felec,velecsum,facel,crf,krf,krf2;
  95     real             *charge;
  96     int              nvdwtype;
  97     __m128d          rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
  98     int              *vdwtype;
  99     real             *vdwparam;
 100     __m128d          one_sixth   = _mm_set1_pd(1.0/6.0);
 101     __m128d          one_twelfth = _mm_set1_pd(1.0/12.0);
 102     __m128i          ewitab;
 103     __m128d          ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
 104     real             *ewtab;
 105     __m128d          dummy_mask,cutoff_mask;
 106     __m128d          signbit   = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
 107     __m128d          one     = _mm_set1_pd(1.0);
 108     __m128d          two     = _mm_set1_pd(2.0);
 109     x                = xx[0];
 110     f                = ff[0];
 111
 112     nri              = nlist->nri;
 113     iinr             = nlist->iinr;
 114     jindex           = nlist->jindex;
 115     jjnr             = nlist->jjnr;
 116     shiftidx         = nlist->shift;
 117     gid              = nlist->gid;
 118     shiftvec         = fr->shift_vec[0];
 119     fshift           = fr->fshift[0];
 120     facel            = _mm_set1_pd(fr->epsfac);
 121     charge           = mdatoms->chargeA;
 122     nvdwtype         = fr->ntype;
 123     vdwparam         = fr->nbfp;
 124     vdwtype          = mdatoms->typeA;
 125
 126     sh_ewald         = _mm_set1_pd(fr->ic->sh_ewald);
 127     ewtab            = fr->ic->tabq_coul_FDV0;
 128     ewtabscale       = _mm_set1_pd(fr->ic->tabq_scale);
 129     ewtabhalfspace   = _mm_set1_pd(0.5/fr->ic->tabq_scale);
 130
 131     /* Setup water-specific parameters */
 132     inr              = nlist->iinr[0];
 133     iq1              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+1]));
 134     iq2              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+2]));
 135     iq3              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+3]));
 136     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 137
 138     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 139     rcutoff_scalar   = fr->rcoulomb;
 140     rcutoff          = _mm_set1_pd(rcutoff_scalar);
 141     rcutoff2         = _mm_mul_pd(rcutoff,rcutoff);
 142
 143     sh_vdw_invrcut6  = _mm_set1_pd(fr->ic->sh_invrc6);
 144     rvdw             = _mm_set1_pd(fr->rvdw);
 145
 146     /* Avoid stupid compiler warnings */
 147     jnrA = jnrB = 0;
 148     j_coord_offsetA = 0;
 149     j_coord_offsetB = 0;
 150
 151     outeriter        = 0;
 152     inneriter        = 0;
 153
 154     /* Start outer loop over neighborlists */
 155     for(iidx=0; iidx<nri; iidx++)
 156     {
 157         /* Load shift vector for this list */
 158         i_shift_offset   = DIM*shiftidx[iidx];
 159
 160         /* Load limits for loop over neighbors */
 161         j_index_start    = jindex[iidx];
 162         j_index_end      = jindex[iidx+1];
 163
 164         /* Get outer coordinate index */
 165         inr              = iinr[iidx];
 166         i_coord_offset   = DIM*inr;
 167
 168         /* Load i particle coords and add shift vector */
 169         gmx_mm_load_shift_and_4rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
 170                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
 171
 172         fix0             = _mm_setzero_pd();
 173         fiy0             = _mm_setzero_pd();
 174         fiz0             = _mm_setzero_pd();
 175         fix1             = _mm_setzero_pd();
 176         fiy1             = _mm_setzero_pd();
 177         fiz1             = _mm_setzero_pd();
 178         fix2             = _mm_setzero_pd();
 179         fiy2             = _mm_setzero_pd();
 180         fiz2             = _mm_setzero_pd();
 181         fix3             = _mm_setzero_pd();
 182         fiy3             = _mm_setzero_pd();
 183         fiz3             = _mm_setzero_pd();
 184
 185         /* Reset potential sums */
 186         velecsum         = _mm_setzero_pd();
 187         vvdwsum          = _mm_setzero_pd();
 188
 189         /* Start inner kernel loop */
 190         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 191         {
 192
 193             /* Get j neighbor index, and coordinate index */
 194             jnrA             = jjnr[jidx];
 195             jnrB             = jjnr[jidx+1];
 196             j_coord_offsetA  = DIM*jnrA;
 197             j_coord_offsetB  = DIM*jnrB;
 198
 199             /* load j atom coordinates */
 200             gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 201                                               &jx0,&jy0,&jz0);
 202
 203             /* Calculate displacement vector */
 204             dx00             = _mm_sub_pd(ix0,jx0);
 205             dy00             = _mm_sub_pd(iy0,jy0);
 206             dz00             = _mm_sub_pd(iz0,jz0);
 207             dx10             = _mm_sub_pd(ix1,jx0);
 208             dy10             = _mm_sub_pd(iy1,jy0);
 209             dz10             = _mm_sub_pd(iz1,jz0);
 210             dx20             = _mm_sub_pd(ix2,jx0);
 211             dy20             = _mm_sub_pd(iy2,jy0);
 212             dz20             = _mm_sub_pd(iz2,jz0);
 213             dx30             = _mm_sub_pd(ix3,jx0);
 214             dy30             = _mm_sub_pd(iy3,jy0);
 215             dz30             = _mm_sub_pd(iz3,jz0);
 216
 217             /* Calculate squared distance and things based on it */
 218             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 219             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
 220             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
 221             rsq30            = gmx_mm_calc_rsq_pd(dx30,dy30,dz30);
 222
 223             rinv10           = gmx_mm_invsqrt_pd(rsq10);
 224             rinv20           = gmx_mm_invsqrt_pd(rsq20);
 225             rinv30           = gmx_mm_invsqrt_pd(rsq30);
 226
 227             rinvsq00         = gmx_mm_inv_pd(rsq00);
 228             rinvsq10         = _mm_mul_pd(rinv10,rinv10);
 229             rinvsq20         = _mm_mul_pd(rinv20,rinv20);
 230             rinvsq30         = _mm_mul_pd(rinv30,rinv30);
 231
 232             /* Load parameters for j particles */
 233             jq0              = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
 234             vdwjidx0A        = 2*vdwtype[jnrA+0];
 235             vdwjidx0B        = 2*vdwtype[jnrB+0];
 236
 237             fjx0             = _mm_setzero_pd();
 238             fjy0             = _mm_setzero_pd();
 239             fjz0             = _mm_setzero_pd();
 240
 241             /**************************
 242              * CALCULATE INTERACTIONS *
 243              **************************/
 244
 245             if (gmx_mm_any_lt(rsq00,rcutoff2))
 246             {
 247
 248             /* Compute parameters for interactions between i and j atoms */
 249             gmx_mm_load_2pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,
 250                                          vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
 251
 252             /* LENNARD-JONES DISPERSION/REPULSION */
 253
 254             rinvsix          = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 255             vvdw6            = _mm_mul_pd(c6_00,rinvsix);
 256             vvdw12           = _mm_mul_pd(c12_00,_mm_mul_pd(rinvsix,rinvsix));
 257             vvdw             = _mm_sub_pd(_mm_mul_pd( _mm_sub_pd(vvdw12 , _mm_mul_pd(c12_00,_mm_mul_pd(sh_vdw_invrcut6,sh_vdw_invrcut6))), one_twelfth) ,
 258                                           _mm_mul_pd( _mm_sub_pd(vvdw6,_mm_mul_pd(c6_00,sh_vdw_invrcut6)),one_sixth));
 259             fvdw             = _mm_mul_pd(_mm_sub_pd(vvdw12,vvdw6),rinvsq00);
 260
 261             cutoff_mask      = _mm_cmplt_pd(rsq00,rcutoff2);
 262
 263             /* Update potential sum for this i atom from the interaction with this j atom. */
 264             vvdw             = _mm_and_pd(vvdw,cutoff_mask);
 265             vvdwsum          = _mm_add_pd(vvdwsum,vvdw);
 266
 267             fscal            = fvdw;
 268
 269             fscal            = _mm_and_pd(fscal,cutoff_mask);
 270
 271             /* Calculate temporary vectorial force */
 272             tx               = _mm_mul_pd(fscal,dx00);
 273             ty               = _mm_mul_pd(fscal,dy00);
 274             tz               = _mm_mul_pd(fscal,dz00);
 275
 276             /* Update vectorial force */
 277             fix0             = _mm_add_pd(fix0,tx);
 278             fiy0             = _mm_add_pd(fiy0,ty);
 279             fiz0             = _mm_add_pd(fiz0,tz);
 280
 281             fjx0             = _mm_add_pd(fjx0,tx);
 282             fjy0             = _mm_add_pd(fjy0,ty);
 283             fjz0             = _mm_add_pd(fjz0,tz);
 284
 285             }
 286
 287             /**************************
 288              * CALCULATE INTERACTIONS *
 289              **************************/
 290
 291             if (gmx_mm_any_lt(rsq10,rcutoff2))
 292             {
 293
 294             r10              = _mm_mul_pd(rsq10,rinv10);
 295
 296             /* Compute parameters for interactions between i and j atoms */
 297             qq10             = _mm_mul_pd(iq1,jq0);
 298
 299             /* EWALD ELECTROSTATICS */
 300
 301             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 302             ewrt             = _mm_mul_pd(r10,ewtabscale);
 303             ewitab           = _mm_cvttpd_epi32(ewrt);
 304             eweps            = _mm_sub_pd(ewrt,_mm_cvtepi32_pd(ewitab));
 305             ewitab           = _mm_slli_epi32(ewitab,2);
 306             ewtabF           = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 307             ewtabD           = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,1) );
 308             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 309             ewtabV           = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,0) +2);
 310             ewtabFn          = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,1) +2);
 311             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 312             felec            = _mm_add_pd(ewtabF,_mm_mul_pd(eweps,ewtabD));
 313             velec            = _mm_sub_pd(ewtabV,_mm_mul_pd(_mm_mul_pd(ewtabhalfspace,eweps),_mm_add_pd(ewtabF,felec)));
 314             velec            = _mm_mul_pd(qq10,_mm_sub_pd(_mm_sub_pd(rinv10,sh_ewald),velec));
 315             felec            = _mm_mul_pd(_mm_mul_pd(qq10,rinv10),_mm_sub_pd(rinvsq10,felec));
 316
 317             cutoff_mask      = _mm_cmplt_pd(rsq10,rcutoff2);
 318
 319             /* Update potential sum for this i atom from the interaction with this j atom. */
 320             velec            = _mm_and_pd(velec,cutoff_mask);
 321             velecsum         = _mm_add_pd(velecsum,velec);
 322
 323             fscal            = felec;
 324
 325             fscal            = _mm_and_pd(fscal,cutoff_mask);
 326
 327             /* Calculate temporary vectorial force */
 328             tx               = _mm_mul_pd(fscal,dx10);
 329             ty               = _mm_mul_pd(fscal,dy10);
 330             tz               = _mm_mul_pd(fscal,dz10);
 331
 332             /* Update vectorial force */
 333             fix1             = _mm_add_pd(fix1,tx);
 334             fiy1             = _mm_add_pd(fiy1,ty);
 335             fiz1             = _mm_add_pd(fiz1,tz);
 336
 337             fjx0             = _mm_add_pd(fjx0,tx);
 338             fjy0             = _mm_add_pd(fjy0,ty);
 339             fjz0             = _mm_add_pd(fjz0,tz);
 340
 341             }
 342
 343             /**************************
 344              * CALCULATE INTERACTIONS *
 345              **************************/
 346
 347             if (gmx_mm_any_lt(rsq20,rcutoff2))
 348             {
 349
 350             r20              = _mm_mul_pd(rsq20,rinv20);
 351
 352             /* Compute parameters for interactions between i and j atoms */
 353             qq20             = _mm_mul_pd(iq2,jq0);
 354
 355             /* EWALD ELECTROSTATICS */
 356
 357             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 358             ewrt             = _mm_mul_pd(r20,ewtabscale);
 359             ewitab           = _mm_cvttpd_epi32(ewrt);
 360             eweps            = _mm_sub_pd(ewrt,_mm_cvtepi32_pd(ewitab));
 361             ewitab           = _mm_slli_epi32(ewitab,2);
 362             ewtabF           = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 363             ewtabD           = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,1) );
 364             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 365             ewtabV           = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,0) +2);
 366             ewtabFn          = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,1) +2);
 367             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 368             felec            = _mm_add_pd(ewtabF,_mm_mul_pd(eweps,ewtabD));
 369             velec            = _mm_sub_pd(ewtabV,_mm_mul_pd(_mm_mul_pd(ewtabhalfspace,eweps),_mm_add_pd(ewtabF,felec)));
 370             velec            = _mm_mul_pd(qq20,_mm_sub_pd(_mm_sub_pd(rinv20,sh_ewald),velec));
 371             felec            = _mm_mul_pd(_mm_mul_pd(qq20,rinv20),_mm_sub_pd(rinvsq20,felec));
 372
 373             cutoff_mask      = _mm_cmplt_pd(rsq20,rcutoff2);
 374
 375             /* Update potential sum for this i atom from the interaction with this j atom. */
 376             velec            = _mm_and_pd(velec,cutoff_mask);
 377             velecsum         = _mm_add_pd(velecsum,velec);
 378
 379             fscal            = felec;
 380
 381             fscal            = _mm_and_pd(fscal,cutoff_mask);
 382
 383             /* Calculate temporary vectorial force */
 384             tx               = _mm_mul_pd(fscal,dx20);
 385             ty               = _mm_mul_pd(fscal,dy20);
 386             tz               = _mm_mul_pd(fscal,dz20);
 387
 388             /* Update vectorial force */
 389             fix2             = _mm_add_pd(fix2,tx);
 390             fiy2             = _mm_add_pd(fiy2,ty);
 391             fiz2             = _mm_add_pd(fiz2,tz);
 392
 393             fjx0             = _mm_add_pd(fjx0,tx);
 394             fjy0             = _mm_add_pd(fjy0,ty);
 395             fjz0             = _mm_add_pd(fjz0,tz);
 396
 397             }
 398
 399             /**************************
 400              * CALCULATE INTERACTIONS *
 401              **************************/
 402
 403             if (gmx_mm_any_lt(rsq30,rcutoff2))
 404             {
 405
 406             r30              = _mm_mul_pd(rsq30,rinv30);
 407
 408             /* Compute parameters for interactions between i and j atoms */
 409             qq30             = _mm_mul_pd(iq3,jq0);
 410
 411             /* EWALD ELECTROSTATICS */
 412
 413             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 414             ewrt             = _mm_mul_pd(r30,ewtabscale);
 415             ewitab           = _mm_cvttpd_epi32(ewrt);
 416             eweps            = _mm_sub_pd(ewrt,_mm_cvtepi32_pd(ewitab));
 417             ewitab           = _mm_slli_epi32(ewitab,2);
 418             ewtabF           = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 419             ewtabD           = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,1) );
 420             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 421             ewtabV           = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,0) +2);
 422             ewtabFn          = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,1) +2);
 423             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 424             felec            = _mm_add_pd(ewtabF,_mm_mul_pd(eweps,ewtabD));
 425             velec            = _mm_sub_pd(ewtabV,_mm_mul_pd(_mm_mul_pd(ewtabhalfspace,eweps),_mm_add_pd(ewtabF,felec)));
 426             velec            = _mm_mul_pd(qq30,_mm_sub_pd(_mm_sub_pd(rinv30,sh_ewald),velec));
 427             felec            = _mm_mul_pd(_mm_mul_pd(qq30,rinv30),_mm_sub_pd(rinvsq30,felec));
 428
 429             cutoff_mask      = _mm_cmplt_pd(rsq30,rcutoff2);
 430
 431             /* Update potential sum for this i atom from the interaction with this j atom. */
 432             velec            = _mm_and_pd(velec,cutoff_mask);
 433             velecsum         = _mm_add_pd(velecsum,velec);
 434
 435             fscal            = felec;
 436
 437             fscal            = _mm_and_pd(fscal,cutoff_mask);
 438
 439             /* Calculate temporary vectorial force */
 440             tx               = _mm_mul_pd(fscal,dx30);
 441             ty               = _mm_mul_pd(fscal,dy30);
 442             tz               = _mm_mul_pd(fscal,dz30);
 443
 444             /* Update vectorial force */
 445             fix3             = _mm_add_pd(fix3,tx);
 446             fiy3             = _mm_add_pd(fiy3,ty);
 447             fiz3             = _mm_add_pd(fiz3,tz);
 448
 449             fjx0             = _mm_add_pd(fjx0,tx);
 450             fjy0             = _mm_add_pd(fjy0,ty);
 451             fjz0             = _mm_add_pd(fjz0,tz);
 452
 453             }
 454
 455             gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,fjx0,fjy0,fjz0);
 456
 457             /* Inner loop uses 182 flops */
 458         }
 459
 460         if(jidx<j_index_end)
 461         {
 462
 463             jnrA             = jjnr[jidx];
 464             j_coord_offsetA  = DIM*jnrA;
 465
 466             /* load j atom coordinates */
 467             gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
 468                                               &jx0,&jy0,&jz0);
 469
 470             /* Calculate displacement vector */
 471             dx00             = _mm_sub_pd(ix0,jx0);
 472             dy00             = _mm_sub_pd(iy0,jy0);
 473             dz00             = _mm_sub_pd(iz0,jz0);
 474             dx10             = _mm_sub_pd(ix1,jx0);
 475             dy10             = _mm_sub_pd(iy1,jy0);
 476             dz10             = _mm_sub_pd(iz1,jz0);
 477             dx20             = _mm_sub_pd(ix2,jx0);
 478             dy20             = _mm_sub_pd(iy2,jy0);
 479             dz20             = _mm_sub_pd(iz2,jz0);
 480             dx30             = _mm_sub_pd(ix3,jx0);
 481             dy30             = _mm_sub_pd(iy3,jy0);
 482             dz30             = _mm_sub_pd(iz3,jz0);
 483
 484             /* Calculate squared distance and things based on it */
 485             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 486             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
 487             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
 488             rsq30            = gmx_mm_calc_rsq_pd(dx30,dy30,dz30);
 489
 490             rinv10           = gmx_mm_invsqrt_pd(rsq10);
 491             rinv20           = gmx_mm_invsqrt_pd(rsq20);
 492             rinv30           = gmx_mm_invsqrt_pd(rsq30);
 493
 494             rinvsq00         = gmx_mm_inv_pd(rsq00);
 495             rinvsq10         = _mm_mul_pd(rinv10,rinv10);
 496             rinvsq20         = _mm_mul_pd(rinv20,rinv20);
 497             rinvsq30         = _mm_mul_pd(rinv30,rinv30);
 498
 499             /* Load parameters for j particles */
 500             jq0              = _mm_load_sd(charge+jnrA+0);
 501             vdwjidx0A        = 2*vdwtype[jnrA+0];
 502
 503             fjx0             = _mm_setzero_pd();
 504             fjy0             = _mm_setzero_pd();
 505             fjz0             = _mm_setzero_pd();
 506
 507             /**************************
 508              * CALCULATE INTERACTIONS *
 509              **************************/
 510
 511             if (gmx_mm_any_lt(rsq00,rcutoff2))
 512             {
 513
 514             /* Compute parameters for interactions between i and j atoms */
 515             gmx_mm_load_1pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,&c6_00,&c12_00);
 516
 517             /* LENNARD-JONES DISPERSION/REPULSION */
 518
 519             rinvsix          = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 520             vvdw6            = _mm_mul_pd(c6_00,rinvsix);
 521             vvdw12           = _mm_mul_pd(c12_00,_mm_mul_pd(rinvsix,rinvsix));
 522             vvdw             = _mm_sub_pd(_mm_mul_pd( _mm_sub_pd(vvdw12 , _mm_mul_pd(c12_00,_mm_mul_pd(sh_vdw_invrcut6,sh_vdw_invrcut6))), one_twelfth) ,
 523                                           _mm_mul_pd( _mm_sub_pd(vvdw6,_mm_mul_pd(c6_00,sh_vdw_invrcut6)),one_sixth));
 524             fvdw             = _mm_mul_pd(_mm_sub_pd(vvdw12,vvdw6),rinvsq00);
 525
 526             cutoff_mask      = _mm_cmplt_pd(rsq00,rcutoff2);
 527
 528             /* Update potential sum for this i atom from the interaction with this j atom. */
 529             vvdw             = _mm_and_pd(vvdw,cutoff_mask);
 530             vvdw             = _mm_unpacklo_pd(vvdw,_mm_setzero_pd());
 531             vvdwsum          = _mm_add_pd(vvdwsum,vvdw);
 532
 533             fscal            = fvdw;
 534
 535             fscal            = _mm_and_pd(fscal,cutoff_mask);
 536
 537             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 538
 539             /* Calculate temporary vectorial force */
 540             tx               = _mm_mul_pd(fscal,dx00);
 541             ty               = _mm_mul_pd(fscal,dy00);
 542             tz               = _mm_mul_pd(fscal,dz00);
 543
 544             /* Update vectorial force */
 545             fix0             = _mm_add_pd(fix0,tx);
 546             fiy0             = _mm_add_pd(fiy0,ty);
 547             fiz0             = _mm_add_pd(fiz0,tz);
 548
 549             fjx0             = _mm_add_pd(fjx0,tx);
 550             fjy0             = _mm_add_pd(fjy0,ty);
 551             fjz0             = _mm_add_pd(fjz0,tz);
 552
 553             }
 554
 555             /**************************
 556              * CALCULATE INTERACTIONS *
 557              **************************/
 558
 559             if (gmx_mm_any_lt(rsq10,rcutoff2))
 560             {
 561
 562             r10              = _mm_mul_pd(rsq10,rinv10);
 563
 564             /* Compute parameters for interactions between i and j atoms */
 565             qq10             = _mm_mul_pd(iq1,jq0);
 566
 567             /* EWALD ELECTROSTATICS */
 568
 569             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 570             ewrt             = _mm_mul_pd(r10,ewtabscale);
 571             ewitab           = _mm_cvttpd_epi32(ewrt);
 572             eweps            = _mm_sub_pd(ewrt,_mm_cvtepi32_pd(ewitab));
 573             ewitab           = _mm_slli_epi32(ewitab,2);
 574             ewtabF           = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 575             ewtabD           = _mm_setzero_pd();
 576             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 577             ewtabV           = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,0) +2);
 578             ewtabFn          = _mm_setzero_pd();
 579             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 580             felec            = _mm_add_pd(ewtabF,_mm_mul_pd(eweps,ewtabD));
 581             velec            = _mm_sub_pd(ewtabV,_mm_mul_pd(_mm_mul_pd(ewtabhalfspace,eweps),_mm_add_pd(ewtabF,felec)));
 582             velec            = _mm_mul_pd(qq10,_mm_sub_pd(_mm_sub_pd(rinv10,sh_ewald),velec));
 583             felec            = _mm_mul_pd(_mm_mul_pd(qq10,rinv10),_mm_sub_pd(rinvsq10,felec));
 584
 585             cutoff_mask      = _mm_cmplt_pd(rsq10,rcutoff2);
 586
 587             /* Update potential sum for this i atom from the interaction with this j atom. */
 588             velec            = _mm_and_pd(velec,cutoff_mask);
 589             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 590             velecsum         = _mm_add_pd(velecsum,velec);
 591
 592             fscal            = felec;
 593
 594             fscal            = _mm_and_pd(fscal,cutoff_mask);
 595
 596             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 597
 598             /* Calculate temporary vectorial force */
 599             tx               = _mm_mul_pd(fscal,dx10);
 600             ty               = _mm_mul_pd(fscal,dy10);
 601             tz               = _mm_mul_pd(fscal,dz10);
 602
 603             /* Update vectorial force */
 604             fix1             = _mm_add_pd(fix1,tx);
 605             fiy1             = _mm_add_pd(fiy1,ty);
 606             fiz1             = _mm_add_pd(fiz1,tz);
 607
 608             fjx0             = _mm_add_pd(fjx0,tx);
 609             fjy0             = _mm_add_pd(fjy0,ty);
 610             fjz0             = _mm_add_pd(fjz0,tz);
 611
 612             }
 613
 614             /**************************
 615              * CALCULATE INTERACTIONS *
 616              **************************/
 617
 618             if (gmx_mm_any_lt(rsq20,rcutoff2))
 619             {
 620
 621             r20              = _mm_mul_pd(rsq20,rinv20);
 622
 623             /* Compute parameters for interactions between i and j atoms */
 624             qq20             = _mm_mul_pd(iq2,jq0);
 625
 626             /* EWALD ELECTROSTATICS */
 627
 628             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 629             ewrt             = _mm_mul_pd(r20,ewtabscale);
 630             ewitab           = _mm_cvttpd_epi32(ewrt);
 631             eweps            = _mm_sub_pd(ewrt,_mm_cvtepi32_pd(ewitab));
 632             ewitab           = _mm_slli_epi32(ewitab,2);
 633             ewtabF           = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 634             ewtabD           = _mm_setzero_pd();
 635             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 636             ewtabV           = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,0) +2);
 637             ewtabFn          = _mm_setzero_pd();
 638             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 639             felec            = _mm_add_pd(ewtabF,_mm_mul_pd(eweps,ewtabD));
 640             velec            = _mm_sub_pd(ewtabV,_mm_mul_pd(_mm_mul_pd(ewtabhalfspace,eweps),_mm_add_pd(ewtabF,felec)));
 641             velec            = _mm_mul_pd(qq20,_mm_sub_pd(_mm_sub_pd(rinv20,sh_ewald),velec));
 642             felec            = _mm_mul_pd(_mm_mul_pd(qq20,rinv20),_mm_sub_pd(rinvsq20,felec));
 643
 644             cutoff_mask      = _mm_cmplt_pd(rsq20,rcutoff2);
 645
 646             /* Update potential sum for this i atom from the interaction with this j atom. */
 647             velec            = _mm_and_pd(velec,cutoff_mask);
 648             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 649             velecsum         = _mm_add_pd(velecsum,velec);
 650
 651             fscal            = felec;
 652
 653             fscal            = _mm_and_pd(fscal,cutoff_mask);
 654
 655             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 656
 657             /* Calculate temporary vectorial force */
 658             tx               = _mm_mul_pd(fscal,dx20);
 659             ty               = _mm_mul_pd(fscal,dy20);
 660             tz               = _mm_mul_pd(fscal,dz20);
 661
 662             /* Update vectorial force */
 663             fix2             = _mm_add_pd(fix2,tx);
 664             fiy2             = _mm_add_pd(fiy2,ty);
 665             fiz2             = _mm_add_pd(fiz2,tz);
 666
 667             fjx0             = _mm_add_pd(fjx0,tx);
 668             fjy0             = _mm_add_pd(fjy0,ty);
 669             fjz0             = _mm_add_pd(fjz0,tz);
 670
 671             }
 672
 673             /**************************
 674              * CALCULATE INTERACTIONS *
 675              **************************/
 676
 677             if (gmx_mm_any_lt(rsq30,rcutoff2))
 678             {
 679
 680             r30              = _mm_mul_pd(rsq30,rinv30);
 681
 682             /* Compute parameters for interactions between i and j atoms */
 683             qq30             = _mm_mul_pd(iq3,jq0);
 684
 685             /* EWALD ELECTROSTATICS */
 686
 687             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 688             ewrt             = _mm_mul_pd(r30,ewtabscale);
 689             ewitab           = _mm_cvttpd_epi32(ewrt);
 690             eweps            = _mm_sub_pd(ewrt,_mm_cvtepi32_pd(ewitab));
 691             ewitab           = _mm_slli_epi32(ewitab,2);
 692             ewtabF           = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 693             ewtabD           = _mm_setzero_pd();
 694             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 695             ewtabV           = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,0) +2);
 696             ewtabFn          = _mm_setzero_pd();
 697             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 698             felec            = _mm_add_pd(ewtabF,_mm_mul_pd(eweps,ewtabD));
 699             velec            = _mm_sub_pd(ewtabV,_mm_mul_pd(_mm_mul_pd(ewtabhalfspace,eweps),_mm_add_pd(ewtabF,felec)));
 700             velec            = _mm_mul_pd(qq30,_mm_sub_pd(_mm_sub_pd(rinv30,sh_ewald),velec));
 701             felec            = _mm_mul_pd(_mm_mul_pd(qq30,rinv30),_mm_sub_pd(rinvsq30,felec));
 702
 703             cutoff_mask      = _mm_cmplt_pd(rsq30,rcutoff2);
 704
 705             /* Update potential sum for this i atom from the interaction with this j atom. */
 706             velec            = _mm_and_pd(velec,cutoff_mask);
 707             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 708             velecsum         = _mm_add_pd(velecsum,velec);
 709
 710             fscal            = felec;
 711
 712             fscal            = _mm_and_pd(fscal,cutoff_mask);
 713
 714             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 715
 716             /* Calculate temporary vectorial force */
 717             tx               = _mm_mul_pd(fscal,dx30);
 718             ty               = _mm_mul_pd(fscal,dy30);
 719             tz               = _mm_mul_pd(fscal,dz30);
 720
 721             /* Update vectorial force */
 722             fix3             = _mm_add_pd(fix3,tx);
 723             fiy3             = _mm_add_pd(fiy3,ty);
 724             fiz3             = _mm_add_pd(fiz3,tz);
 725
 726             fjx0             = _mm_add_pd(fjx0,tx);
 727             fjy0             = _mm_add_pd(fjy0,ty);
 728             fjz0             = _mm_add_pd(fjz0,tz);
 729
 730             }
 731
 732             gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,fjx0,fjy0,fjz0);
 733
 734             /* Inner loop uses 182 flops */
 735         }
 736
 737         /* End of innermost loop */
 738
 739         gmx_mm_update_iforce_4atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
 740                                               f+i_coord_offset,fshift+i_shift_offset);
 741
 742         ggid                        = gid[iidx];
 743         /* Update potential energies */
 744         gmx_mm_update_1pot_pd(velecsum,kernel_data->energygrp_elec+ggid);
 745         gmx_mm_update_1pot_pd(vvdwsum,kernel_data->energygrp_vdw+ggid);
 746
 747         /* Increment number of inner iterations */
 748         inneriter                  += j_index_end - j_index_start;
 749
 750         /* Outer loop uses 26 flops */
 751     }
 752
 753     /* Increment number of outer iterations */
 754     outeriter        += nri;
 755
 756     /* Update outer/inner flops */
 757
 758     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W4_VF,outeriter*26 + inneriter*182);
 759 }
 760 /*
 761  * Gromacs nonbonded kernel:   nb_kernel_ElecEwSh_VdwLJSh_GeomW4P1_F_sse2_double
 762  * Electrostatics interaction: Ewald
 763  * VdW interaction:            LennardJones
 764  * Geometry:                   Water4-Particle
 765  * Calculate force/pot:        Force
 766  */
 767 void
 768 nb_kernel_ElecEwSh_VdwLJSh_GeomW4P1_F_sse2_double
 769                     (t_nblist                    * gmx_restrict       nlist,
 770                      rvec                        * gmx_restrict          xx,
 771                      rvec                        * gmx_restrict          ff,
 772                      t_forcerec                  * gmx_restrict          fr,
 773                      t_mdatoms                   * gmx_restrict     mdatoms,
 774                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 775                      t_nrnb                      * gmx_restrict        nrnb)
 776 {
 777     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 778      * just 0 for non-waters.
 779      * Suffixes A,B refer to j loop unrolling done with SSE double precision, e.g. for the two different
 780      * jnr indices corresponding to data put in the four positions in the SIMD register.
 781      */
 782     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 783     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 784     int              jnrA,jnrB;
 785     int              j_coord_offsetA,j_coord_offsetB;
 786     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 787     real             rcutoff_scalar;
 788     real             *shiftvec,*fshift,*x,*f;
 789     __m128d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 790     int              vdwioffset0;
 791     __m128d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 792     int              vdwioffset1;
 793     __m128d          ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
 794     int              vdwioffset2;
 795     __m128d          ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
 796     int              vdwioffset3;
 797     __m128d          ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
 798     int              vdwjidx0A,vdwjidx0B;
 799     __m128d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 800     __m128d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 801     __m128d          dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
 802     __m128d          dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
 803     __m128d          dx30,dy30,dz30,rsq30,rinv30,rinvsq30,r30,qq30,c6_30,c12_30;
 804     __m128d          velec,felec,velecsum,facel,crf,krf,krf2;
 805     real             *charge;
 806     int              nvdwtype;
 807     __m128d          rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 808     int              *vdwtype;
 809     real             *vdwparam;
 810     __m128d          one_sixth   = _mm_set1_pd(1.0/6.0);
 811     __m128d          one_twelfth = _mm_set1_pd(1.0/12.0);
 812     __m128i          ewitab;
 813     __m128d          ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
 814     real             *ewtab;
 815     __m128d          dummy_mask,cutoff_mask;
 816     __m128d          signbit   = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
 817     __m128d          one     = _mm_set1_pd(1.0);
 818     __m128d          two     = _mm_set1_pd(2.0);
 819     x                = xx[0];
 820     f                = ff[0];
 821
 822     nri              = nlist->nri;
 823     iinr             = nlist->iinr;
 824     jindex           = nlist->jindex;
 825     jjnr             = nlist->jjnr;
 826     shiftidx         = nlist->shift;
 827     gid              = nlist->gid;
 828     shiftvec         = fr->shift_vec[0];
 829     fshift           = fr->fshift[0];
 830     facel            = _mm_set1_pd(fr->epsfac);
 831     charge           = mdatoms->chargeA;
 832     nvdwtype         = fr->ntype;
 833     vdwparam         = fr->nbfp;
 834     vdwtype          = mdatoms->typeA;
 835
 836     sh_ewald         = _mm_set1_pd(fr->ic->sh_ewald);
 837     ewtab            = fr->ic->tabq_coul_F;
 838     ewtabscale       = _mm_set1_pd(fr->ic->tabq_scale);
 839     ewtabhalfspace   = _mm_set1_pd(0.5/fr->ic->tabq_scale);
 840
 841     /* Setup water-specific parameters */
 842     inr              = nlist->iinr[0];
 843     iq1              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+1]));
 844     iq2              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+2]));
 845     iq3              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+3]));
 846     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 847
 848     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 849     rcutoff_scalar   = fr->rcoulomb;
 850     rcutoff          = _mm_set1_pd(rcutoff_scalar);
 851     rcutoff2         = _mm_mul_pd(rcutoff,rcutoff);
 852
 853     sh_vdw_invrcut6  = _mm_set1_pd(fr->ic->sh_invrc6);
 854     rvdw             = _mm_set1_pd(fr->rvdw);
 855
 856     /* Avoid stupid compiler warnings */
 857     jnrA = jnrB = 0;
 858     j_coord_offsetA = 0;
 859     j_coord_offsetB = 0;
 860
 861     outeriter        = 0;
 862     inneriter        = 0;
 863
 864     /* Start outer loop over neighborlists */
 865     for(iidx=0; iidx<nri; iidx++)
 866     {
 867         /* Load shift vector for this list */
 868         i_shift_offset   = DIM*shiftidx[iidx];
 869
 870         /* Load limits for loop over neighbors */
 871         j_index_start    = jindex[iidx];
 872         j_index_end      = jindex[iidx+1];
 873
 874         /* Get outer coordinate index */
 875         inr              = iinr[iidx];
 876         i_coord_offset   = DIM*inr;
 877
 878         /* Load i particle coords and add shift vector */
 879         gmx_mm_load_shift_and_4rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
 880                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
 881
 882         fix0             = _mm_setzero_pd();
 883         fiy0             = _mm_setzero_pd();
 884         fiz0             = _mm_setzero_pd();
 885         fix1             = _mm_setzero_pd();
 886         fiy1             = _mm_setzero_pd();
 887         fiz1             = _mm_setzero_pd();
 888         fix2             = _mm_setzero_pd();
 889         fiy2             = _mm_setzero_pd();
 890         fiz2             = _mm_setzero_pd();
 891         fix3             = _mm_setzero_pd();
 892         fiy3             = _mm_setzero_pd();
 893         fiz3             = _mm_setzero_pd();
 894
 895         /* Start inner kernel loop */
 896         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 897         {
 898
 899             /* Get j neighbor index, and coordinate index */
 900             jnrA             = jjnr[jidx];
 901             jnrB             = jjnr[jidx+1];
 902             j_coord_offsetA  = DIM*jnrA;
 903             j_coord_offsetB  = DIM*jnrB;
 904
 905             /* load j atom coordinates */
 906             gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 907                                               &jx0,&jy0,&jz0);
 908
 909             /* Calculate displacement vector */
 910             dx00             = _mm_sub_pd(ix0,jx0);
 911             dy00             = _mm_sub_pd(iy0,jy0);
 912             dz00             = _mm_sub_pd(iz0,jz0);
 913             dx10             = _mm_sub_pd(ix1,jx0);
 914             dy10             = _mm_sub_pd(iy1,jy0);
 915             dz10             = _mm_sub_pd(iz1,jz0);
 916             dx20             = _mm_sub_pd(ix2,jx0);
 917             dy20             = _mm_sub_pd(iy2,jy0);
 918             dz20             = _mm_sub_pd(iz2,jz0);
 919             dx30             = _mm_sub_pd(ix3,jx0);
 920             dy30             = _mm_sub_pd(iy3,jy0);
 921             dz30             = _mm_sub_pd(iz3,jz0);
 922
 923             /* Calculate squared distance and things based on it */
 924             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 925             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
 926             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
 927             rsq30            = gmx_mm_calc_rsq_pd(dx30,dy30,dz30);
 928
 929             rinv10           = gmx_mm_invsqrt_pd(rsq10);
 930             rinv20           = gmx_mm_invsqrt_pd(rsq20);
 931             rinv30           = gmx_mm_invsqrt_pd(rsq30);
 932
 933             rinvsq00         = gmx_mm_inv_pd(rsq00);
 934             rinvsq10         = _mm_mul_pd(rinv10,rinv10);
 935             rinvsq20         = _mm_mul_pd(rinv20,rinv20);
 936             rinvsq30         = _mm_mul_pd(rinv30,rinv30);
 937
 938             /* Load parameters for j particles */
 939             jq0              = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
 940             vdwjidx0A        = 2*vdwtype[jnrA+0];
 941             vdwjidx0B        = 2*vdwtype[jnrB+0];
 942
 943             fjx0             = _mm_setzero_pd();
 944             fjy0             = _mm_setzero_pd();
 945             fjz0             = _mm_setzero_pd();
 946
 947             /**************************
 948              * CALCULATE INTERACTIONS *
 949              **************************/
 950
 951             if (gmx_mm_any_lt(rsq00,rcutoff2))
 952             {
 953
 954             /* Compute parameters for interactions between i and j atoms */
 955             gmx_mm_load_2pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,
 956                                          vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
 957
 958             /* LENNARD-JONES DISPERSION/REPULSION */
 959
 960             rinvsix          = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 961             fvdw             = _mm_mul_pd(_mm_sub_pd(_mm_mul_pd(c12_00,rinvsix),c6_00),_mm_mul_pd(rinvsix,rinvsq00));
 962
 963             cutoff_mask      = _mm_cmplt_pd(rsq00,rcutoff2);
 964
 965             fscal            = fvdw;
 966
 967             fscal            = _mm_and_pd(fscal,cutoff_mask);
 968
 969             /* Calculate temporary vectorial force */
 970             tx               = _mm_mul_pd(fscal,dx00);
 971             ty               = _mm_mul_pd(fscal,dy00);
 972             tz               = _mm_mul_pd(fscal,dz00);
 973
 974             /* Update vectorial force */
 975             fix0             = _mm_add_pd(fix0,tx);
 976             fiy0             = _mm_add_pd(fiy0,ty);
 977             fiz0             = _mm_add_pd(fiz0,tz);
 978
 979             fjx0             = _mm_add_pd(fjx0,tx);
 980             fjy0             = _mm_add_pd(fjy0,ty);
 981             fjz0             = _mm_add_pd(fjz0,tz);
 982
 983             }
 984
 985             /**************************
 986              * CALCULATE INTERACTIONS *
 987              **************************/
 988
 989             if (gmx_mm_any_lt(rsq10,rcutoff2))
 990             {
 991
 992             r10              = _mm_mul_pd(rsq10,rinv10);
 993
 994             /* Compute parameters for interactions between i and j atoms */
 995             qq10             = _mm_mul_pd(iq1,jq0);
 996
 997             /* EWALD ELECTROSTATICS */
 998
 999             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1000             ewrt             = _mm_mul_pd(r10,ewtabscale);
1001             ewitab           = _mm_cvttpd_epi32(ewrt);
1002             eweps            = _mm_sub_pd(ewrt,_mm_cvtepi32_pd(ewitab));
1003             gmx_mm_load_2pair_swizzle_pd(ewtab+gmx_mm_extract_epi32(ewitab,0),ewtab+gmx_mm_extract_epi32(ewitab,1),
1004                                          &ewtabF,&ewtabFn);
1005             felec            = _mm_add_pd(_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF),_mm_mul_pd(eweps,ewtabFn));
1006             felec            = _mm_mul_pd(_mm_mul_pd(qq10,rinv10),_mm_sub_pd(rinvsq10,felec));
1007
1008             cutoff_mask      = _mm_cmplt_pd(rsq10,rcutoff2);
1009
1010             fscal            = felec;
1011
1012             fscal            = _mm_and_pd(fscal,cutoff_mask);
1013
1014             /* Calculate temporary vectorial force */
1015             tx               = _mm_mul_pd(fscal,dx10);
1016             ty               = _mm_mul_pd(fscal,dy10);
1017             tz               = _mm_mul_pd(fscal,dz10);
1018
1019             /* Update vectorial force */
1020             fix1             = _mm_add_pd(fix1,tx);
1021             fiy1             = _mm_add_pd(fiy1,ty);
1022             fiz1             = _mm_add_pd(fiz1,tz);
1023
1024             fjx0             = _mm_add_pd(fjx0,tx);
1025             fjy0             = _mm_add_pd(fjy0,ty);
1026             fjz0             = _mm_add_pd(fjz0,tz);
1027
1028             }
1029
1030             /**************************
1031              * CALCULATE INTERACTIONS *
1032              **************************/
1033
1034             if (gmx_mm_any_lt(rsq20,rcutoff2))
1035             {
1036
1037             r20              = _mm_mul_pd(rsq20,rinv20);
1038
1039             /* Compute parameters for interactions between i and j atoms */
1040             qq20             = _mm_mul_pd(iq2,jq0);
1041
1042             /* EWALD ELECTROSTATICS */
1043
1044             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1045             ewrt             = _mm_mul_pd(r20,ewtabscale);
1046             ewitab           = _mm_cvttpd_epi32(ewrt);
1047             eweps            = _mm_sub_pd(ewrt,_mm_cvtepi32_pd(ewitab));
1048             gmx_mm_load_2pair_swizzle_pd(ewtab+gmx_mm_extract_epi32(ewitab,0),ewtab+gmx_mm_extract_epi32(ewitab,1),
1049                                          &ewtabF,&ewtabFn);
1050             felec            = _mm_add_pd(_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF),_mm_mul_pd(eweps,ewtabFn));
1051             felec            = _mm_mul_pd(_mm_mul_pd(qq20,rinv20),_mm_sub_pd(rinvsq20,felec));
1052
1053             cutoff_mask      = _mm_cmplt_pd(rsq20,rcutoff2);
1054
1055             fscal            = felec;
1056
1057             fscal            = _mm_and_pd(fscal,cutoff_mask);
1058
1059             /* Calculate temporary vectorial force */
1060             tx               = _mm_mul_pd(fscal,dx20);
1061             ty               = _mm_mul_pd(fscal,dy20);
1062             tz               = _mm_mul_pd(fscal,dz20);
1063
1064             /* Update vectorial force */
1065             fix2             = _mm_add_pd(fix2,tx);
1066             fiy2             = _mm_add_pd(fiy2,ty);
1067             fiz2             = _mm_add_pd(fiz2,tz);
1068
1069             fjx0             = _mm_add_pd(fjx0,tx);
1070             fjy0             = _mm_add_pd(fjy0,ty);
1071             fjz0             = _mm_add_pd(fjz0,tz);
1072
1073             }
1074
1075             /**************************
1076              * CALCULATE INTERACTIONS *
1077              **************************/
1078
1079             if (gmx_mm_any_lt(rsq30,rcutoff2))
1080             {
1081
1082             r30              = _mm_mul_pd(rsq30,rinv30);
1083
1084             /* Compute parameters for interactions between i and j atoms */
1085             qq30             = _mm_mul_pd(iq3,jq0);
1086
1087             /* EWALD ELECTROSTATICS */
1088
1089             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1090             ewrt             = _mm_mul_pd(r30,ewtabscale);
1091             ewitab           = _mm_cvttpd_epi32(ewrt);
1092             eweps            = _mm_sub_pd(ewrt,_mm_cvtepi32_pd(ewitab));
1093             gmx_mm_load_2pair_swizzle_pd(ewtab+gmx_mm_extract_epi32(ewitab,0),ewtab+gmx_mm_extract_epi32(ewitab,1),
1094                                          &ewtabF,&ewtabFn);
1095             felec            = _mm_add_pd(_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF),_mm_mul_pd(eweps,ewtabFn));
1096             felec            = _mm_mul_pd(_mm_mul_pd(qq30,rinv30),_mm_sub_pd(rinvsq30,felec));
1097
1098             cutoff_mask      = _mm_cmplt_pd(rsq30,rcutoff2);
1099
1100             fscal            = felec;
1101
1102             fscal            = _mm_and_pd(fscal,cutoff_mask);
1103
1104             /* Calculate temporary vectorial force */
1105             tx               = _mm_mul_pd(fscal,dx30);
1106             ty               = _mm_mul_pd(fscal,dy30);
1107             tz               = _mm_mul_pd(fscal,dz30);
1108
1109             /* Update vectorial force */
1110             fix3             = _mm_add_pd(fix3,tx);
1111             fiy3             = _mm_add_pd(fiy3,ty);
1112             fiz3             = _mm_add_pd(fiz3,tz);
1113
1114             fjx0             = _mm_add_pd(fjx0,tx);
1115             fjy0             = _mm_add_pd(fjy0,ty);
1116             fjz0             = _mm_add_pd(fjz0,tz);
1117
1118             }
1119
1120             gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,fjx0,fjy0,fjz0);
1121
1122             /* Inner loop uses 150 flops */
1123         }
1124
1125         if(jidx<j_index_end)
1126         {
1127
1128             jnrA             = jjnr[jidx];
1129             j_coord_offsetA  = DIM*jnrA;
1130
1131             /* load j atom coordinates */
1132             gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
1133                                               &jx0,&jy0,&jz0);
1134
1135             /* Calculate displacement vector */
1136             dx00             = _mm_sub_pd(ix0,jx0);
1137             dy00             = _mm_sub_pd(iy0,jy0);
1138             dz00             = _mm_sub_pd(iz0,jz0);
1139             dx10             = _mm_sub_pd(ix1,jx0);
1140             dy10             = _mm_sub_pd(iy1,jy0);
1141             dz10             = _mm_sub_pd(iz1,jz0);
1142             dx20             = _mm_sub_pd(ix2,jx0);
1143             dy20             = _mm_sub_pd(iy2,jy0);
1144             dz20             = _mm_sub_pd(iz2,jz0);
1145             dx30             = _mm_sub_pd(ix3,jx0);
1146             dy30             = _mm_sub_pd(iy3,jy0);
1147             dz30             = _mm_sub_pd(iz3,jz0);
1148
1149             /* Calculate squared distance and things based on it */
1150             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
1151             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
1152             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
1153             rsq30            = gmx_mm_calc_rsq_pd(dx30,dy30,dz30);
1154
1155             rinv10           = gmx_mm_invsqrt_pd(rsq10);
1156             rinv20           = gmx_mm_invsqrt_pd(rsq20);
1157             rinv30           = gmx_mm_invsqrt_pd(rsq30);
1158
1159             rinvsq00         = gmx_mm_inv_pd(rsq00);
1160             rinvsq10         = _mm_mul_pd(rinv10,rinv10);
1161             rinvsq20         = _mm_mul_pd(rinv20,rinv20);
1162             rinvsq30         = _mm_mul_pd(rinv30,rinv30);
1163
1164             /* Load parameters for j particles */
1165             jq0              = _mm_load_sd(charge+jnrA+0);
1166             vdwjidx0A        = 2*vdwtype[jnrA+0];
1167
1168             fjx0             = _mm_setzero_pd();
1169             fjy0             = _mm_setzero_pd();
1170             fjz0             = _mm_setzero_pd();
1171
1172             /**************************
1173              * CALCULATE INTERACTIONS *
1174              **************************/
1175
1176             if (gmx_mm_any_lt(rsq00,rcutoff2))
1177             {
1178
1179             /* Compute parameters for interactions between i and j atoms */
1180             gmx_mm_load_1pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,&c6_00,&c12_00);
1181
1182             /* LENNARD-JONES DISPERSION/REPULSION */
1183
1184             rinvsix          = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
1185             fvdw             = _mm_mul_pd(_mm_sub_pd(_mm_mul_pd(c12_00,rinvsix),c6_00),_mm_mul_pd(rinvsix,rinvsq00));
1186
1187             cutoff_mask      = _mm_cmplt_pd(rsq00,rcutoff2);
1188
1189             fscal            = fvdw;
1190
1191             fscal            = _mm_and_pd(fscal,cutoff_mask);
1192
1193             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1194
1195             /* Calculate temporary vectorial force */
1196             tx               = _mm_mul_pd(fscal,dx00);
1197             ty               = _mm_mul_pd(fscal,dy00);
1198             tz               = _mm_mul_pd(fscal,dz00);
1199
1200             /* Update vectorial force */
1201             fix0             = _mm_add_pd(fix0,tx);
1202             fiy0             = _mm_add_pd(fiy0,ty);
1203             fiz0             = _mm_add_pd(fiz0,tz);
1204
1205             fjx0             = _mm_add_pd(fjx0,tx);
1206             fjy0             = _mm_add_pd(fjy0,ty);
1207             fjz0             = _mm_add_pd(fjz0,tz);
1208
1209             }
1210
1211             /**************************
1212              * CALCULATE INTERACTIONS *
1213              **************************/
1214
1215             if (gmx_mm_any_lt(rsq10,rcutoff2))
1216             {
1217
1218             r10              = _mm_mul_pd(rsq10,rinv10);
1219
1220             /* Compute parameters for interactions between i and j atoms */
1221             qq10             = _mm_mul_pd(iq1,jq0);
1222
1223             /* EWALD ELECTROSTATICS */
1224
1225             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1226             ewrt             = _mm_mul_pd(r10,ewtabscale);
1227             ewitab           = _mm_cvttpd_epi32(ewrt);
1228             eweps            = _mm_sub_pd(ewrt,_mm_cvtepi32_pd(ewitab));
1229             gmx_mm_load_1pair_swizzle_pd(ewtab+gmx_mm_extract_epi32(ewitab,0),&ewtabF,&ewtabFn);
1230             felec            = _mm_add_pd(_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF),_mm_mul_pd(eweps,ewtabFn));
1231             felec            = _mm_mul_pd(_mm_mul_pd(qq10,rinv10),_mm_sub_pd(rinvsq10,felec));
1232
1233             cutoff_mask      = _mm_cmplt_pd(rsq10,rcutoff2);
1234
1235             fscal            = felec;
1236
1237             fscal            = _mm_and_pd(fscal,cutoff_mask);
1238
1239             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1240
1241             /* Calculate temporary vectorial force */
1242             tx               = _mm_mul_pd(fscal,dx10);
1243             ty               = _mm_mul_pd(fscal,dy10);
1244             tz               = _mm_mul_pd(fscal,dz10);
1245
1246             /* Update vectorial force */
1247             fix1             = _mm_add_pd(fix1,tx);
1248             fiy1             = _mm_add_pd(fiy1,ty);
1249             fiz1             = _mm_add_pd(fiz1,tz);
1250
1251             fjx0             = _mm_add_pd(fjx0,tx);
1252             fjy0             = _mm_add_pd(fjy0,ty);
1253             fjz0             = _mm_add_pd(fjz0,tz);
1254
1255             }
1256
1257             /**************************
1258              * CALCULATE INTERACTIONS *
1259              **************************/
1260
1261             if (gmx_mm_any_lt(rsq20,rcutoff2))
1262             {
1263
1264             r20              = _mm_mul_pd(rsq20,rinv20);
1265
1266             /* Compute parameters for interactions between i and j atoms */
1267             qq20             = _mm_mul_pd(iq2,jq0);
1268
1269             /* EWALD ELECTROSTATICS */
1270
1271             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1272             ewrt             = _mm_mul_pd(r20,ewtabscale);
1273             ewitab           = _mm_cvttpd_epi32(ewrt);
1274             eweps            = _mm_sub_pd(ewrt,_mm_cvtepi32_pd(ewitab));
1275             gmx_mm_load_1pair_swizzle_pd(ewtab+gmx_mm_extract_epi32(ewitab,0),&ewtabF,&ewtabFn);
1276             felec            = _mm_add_pd(_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF),_mm_mul_pd(eweps,ewtabFn));
1277             felec            = _mm_mul_pd(_mm_mul_pd(qq20,rinv20),_mm_sub_pd(rinvsq20,felec));
1278
1279             cutoff_mask      = _mm_cmplt_pd(rsq20,rcutoff2);
1280
1281             fscal            = felec;
1282
1283             fscal            = _mm_and_pd(fscal,cutoff_mask);
1284
1285             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1286
1287             /* Calculate temporary vectorial force */
1288             tx               = _mm_mul_pd(fscal,dx20);
1289             ty               = _mm_mul_pd(fscal,dy20);
1290             tz               = _mm_mul_pd(fscal,dz20);
1291
1292             /* Update vectorial force */
1293             fix2             = _mm_add_pd(fix2,tx);
1294             fiy2             = _mm_add_pd(fiy2,ty);
1295             fiz2             = _mm_add_pd(fiz2,tz);
1296
1297             fjx0             = _mm_add_pd(fjx0,tx);
1298             fjy0             = _mm_add_pd(fjy0,ty);
1299             fjz0             = _mm_add_pd(fjz0,tz);
1300
1301             }
1302
1303             /**************************
1304              * CALCULATE INTERACTIONS *
1305              **************************/
1306
1307             if (gmx_mm_any_lt(rsq30,rcutoff2))
1308             {
1309
1310             r30              = _mm_mul_pd(rsq30,rinv30);
1311
1312             /* Compute parameters for interactions between i and j atoms */
1313             qq30             = _mm_mul_pd(iq3,jq0);
1314
1315             /* EWALD ELECTROSTATICS */
1316
1317             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1318             ewrt             = _mm_mul_pd(r30,ewtabscale);
1319             ewitab           = _mm_cvttpd_epi32(ewrt);
1320             eweps            = _mm_sub_pd(ewrt,_mm_cvtepi32_pd(ewitab));
1321             gmx_mm_load_1pair_swizzle_pd(ewtab+gmx_mm_extract_epi32(ewitab,0),&ewtabF,&ewtabFn);
1322             felec            = _mm_add_pd(_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF),_mm_mul_pd(eweps,ewtabFn));
1323             felec            = _mm_mul_pd(_mm_mul_pd(qq30,rinv30),_mm_sub_pd(rinvsq30,felec));
1324
1325             cutoff_mask      = _mm_cmplt_pd(rsq30,rcutoff2);
1326
1327             fscal            = felec;
1328
1329             fscal            = _mm_and_pd(fscal,cutoff_mask);
1330
1331             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1332
1333             /* Calculate temporary vectorial force */
1334             tx               = _mm_mul_pd(fscal,dx30);
1335             ty               = _mm_mul_pd(fscal,dy30);
1336             tz               = _mm_mul_pd(fscal,dz30);
1337
1338             /* Update vectorial force */
1339             fix3             = _mm_add_pd(fix3,tx);
1340             fiy3             = _mm_add_pd(fiy3,ty);
1341             fiz3             = _mm_add_pd(fiz3,tz);
1342
1343             fjx0             = _mm_add_pd(fjx0,tx);
1344             fjy0             = _mm_add_pd(fjy0,ty);
1345             fjz0             = _mm_add_pd(fjz0,tz);
1346
1347             }
1348
1349             gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,fjx0,fjy0,fjz0);
1350
1351             /* Inner loop uses 150 flops */
1352         }
1353
1354         /* End of innermost loop */
1355
1356         gmx_mm_update_iforce_4atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
1357                                               f+i_coord_offset,fshift+i_shift_offset);
1358
1359         /* Increment number of inner iterations */
1360         inneriter                  += j_index_end - j_index_start;
1361
1362         /* Outer loop uses 24 flops */
1363     }
1364
1365     /* Increment number of outer iterations */
1366     outeriter        += nri;
1367
1368     /* Update outer/inner flops */
1369
1370     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W4_F,outeriter*24 + inneriter*150);
1371 }