src/gromacs/gmxlib/nonbonded/nb_kernel_sse2_double/nb_kernel_ElecEw_VdwLJ_GeomW4P1_sse2_double.cpp

   1 /*
   2  * This file is part of the GROMACS molecular simulation package.
   3  *
   4  * Copyright (c) 2012,2013,2014,2015,2017,2018, by the GROMACS development team, led by
   5  * Mark Abraham, David van der Spoel, Berk Hess, and Erik Lindahl,
   6  * and including many others, as listed in the AUTHORS file in the
   7  * top-level source directory and at http://www.gromacs.org.
   8  *
   9  * GROMACS is free software; you can redistribute it and/or
  10  * modify it under the terms of the GNU Lesser General Public License
  11  * as published by the Free Software Foundation; either version 2.1
  12  * of the License, or (at your option) any later version.
  13  *
  14  * GROMACS is distributed in the hope that it will be useful,
  15  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  16  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  17  * Lesser General Public License for more details.
  18  *
  19  * You should have received a copy of the GNU Lesser General Public
  20  * License along with GROMACS; if not, see
  21  * http://www.gnu.org/licenses, or write to the Free Software Foundation,
  22  * Inc., 51 Franklin Street, Fifth Floor, Boston, MA  02110-1301  USA.
  23  *
  24  * If you want to redistribute modifications to GROMACS, please
  25  * consider that scientific software is very special. Version
  26  * control is crucial - bugs must be traceable. We will be happy to
  27  * consider code for inclusion in the official distribution, but
  28  * derived work must not be called official GROMACS. Details are found
  29  * in the README & COPYING files - if they are missing, get the
  30  * official version at http://www.gromacs.org.
  31  *
  32  * To help us fund GROMACS development, we humbly ask that you cite
  33  * the research papers on the package. Check out http://www.gromacs.org.
  34  */
  35 /*
  36  * Note: this file was generated by the GROMACS sse2_double kernel generator.
  37  */
  38 #include "gmxpre.h"
  39
  40 #include "config.h"
  41
  42 #include <math.h>
  43
  44 #include "../nb_kernel.h"
  45 #include "gromacs/gmxlib/nrnb.h"
  46
  47 #include "kernelutil_x86_sse2_double.h"
  48
  49 /*
  50  * Gromacs nonbonded kernel:   nb_kernel_ElecEw_VdwLJ_GeomW4P1_VF_sse2_double
  51  * Electrostatics interaction: Ewald
  52  * VdW interaction:            LennardJones
  53  * Geometry:                   Water4-Particle
  54  * Calculate force/pot:        PotentialAndForce
  55  */
  56 void
  57 nb_kernel_ElecEw_VdwLJ_GeomW4P1_VF_sse2_double
  58                     (t_nblist                    * gmx_restrict       nlist,
  59                      rvec                        * gmx_restrict          xx,
  60                      rvec                        * gmx_restrict          ff,
  61                      struct t_forcerec           * gmx_restrict          fr,
  62                      t_mdatoms                   * gmx_restrict     mdatoms,
  63                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
  64                      t_nrnb                      * gmx_restrict        nrnb)
  65 {
  66     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
  67      * just 0 for non-waters.
  68      * Suffixes A,B refer to j loop unrolling done with SSE double precision, e.g. for the two different
  69      * jnr indices corresponding to data put in the four positions in the SIMD register.
  70      */
  71     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
  72     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
  73     int              jnrA,jnrB;
  74     int              j_coord_offsetA,j_coord_offsetB;
  75     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
  76     real             rcutoff_scalar;
  77     real             *shiftvec,*fshift,*x,*f;
  78     __m128d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
  79     int              vdwioffset0;
  80     __m128d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
  81     int              vdwioffset1;
  82     __m128d          ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
  83     int              vdwioffset2;
  84     __m128d          ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
  85     int              vdwioffset3;
  86     __m128d          ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
  87     int              vdwjidx0A,vdwjidx0B;
  88     __m128d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  89     __m128d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  90     __m128d          dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
  91     __m128d          dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
  92     __m128d          dx30,dy30,dz30,rsq30,rinv30,rinvsq30,r30,qq30,c6_30,c12_30;
  93     __m128d          velec,felec,velecsum,facel,crf,krf,krf2;
  94     real             *charge;
  95     int              nvdwtype;
  96     __m128d          rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
  97     int              *vdwtype;
  98     real             *vdwparam;
  99     __m128d          one_sixth   = _mm_set1_pd(1.0/6.0);
 100     __m128d          one_twelfth = _mm_set1_pd(1.0/12.0);
 101     __m128i          ewitab;
 102     __m128d          ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
 103     real             *ewtab;
 104     __m128d          dummy_mask,cutoff_mask;
 105     __m128d          signbit   = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
 106     __m128d          one     = _mm_set1_pd(1.0);
 107     __m128d          two     = _mm_set1_pd(2.0);
 108     x                = xx[0];
 109     f                = ff[0];
 110
 111     nri              = nlist->nri;
 112     iinr             = nlist->iinr;
 113     jindex           = nlist->jindex;
 114     jjnr             = nlist->jjnr;
 115     shiftidx         = nlist->shift;
 116     gid              = nlist->gid;
 117     shiftvec         = fr->shift_vec[0];
 118     fshift           = fr->fshift[0];
 119     facel            = _mm_set1_pd(fr->ic->epsfac);
 120     charge           = mdatoms->chargeA;
 121     nvdwtype         = fr->ntype;
 122     vdwparam         = fr->nbfp;
 123     vdwtype          = mdatoms->typeA;
 124
 125     sh_ewald         = _mm_set1_pd(fr->ic->sh_ewald);
 126     ewtab            = fr->ic->tabq_coul_FDV0;
 127     ewtabscale       = _mm_set1_pd(fr->ic->tabq_scale);
 128     ewtabhalfspace   = _mm_set1_pd(0.5/fr->ic->tabq_scale);
 129
 130     /* Setup water-specific parameters */
 131     inr              = nlist->iinr[0];
 132     iq1              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+1]));
 133     iq2              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+2]));
 134     iq3              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+3]));
 135     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 136
 137     /* Avoid stupid compiler warnings */
 138     jnrA = jnrB = 0;
 139     j_coord_offsetA = 0;
 140     j_coord_offsetB = 0;
 141
 142     outeriter        = 0;
 143     inneriter        = 0;
 144
 145     /* Start outer loop over neighborlists */
 146     for(iidx=0; iidx<nri; iidx++)
 147     {
 148         /* Load shift vector for this list */
 149         i_shift_offset   = DIM*shiftidx[iidx];
 150
 151         /* Load limits for loop over neighbors */
 152         j_index_start    = jindex[iidx];
 153         j_index_end      = jindex[iidx+1];
 154
 155         /* Get outer coordinate index */
 156         inr              = iinr[iidx];
 157         i_coord_offset   = DIM*inr;
 158
 159         /* Load i particle coords and add shift vector */
 160         gmx_mm_load_shift_and_4rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
 161                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
 162
 163         fix0             = _mm_setzero_pd();
 164         fiy0             = _mm_setzero_pd();
 165         fiz0             = _mm_setzero_pd();
 166         fix1             = _mm_setzero_pd();
 167         fiy1             = _mm_setzero_pd();
 168         fiz1             = _mm_setzero_pd();
 169         fix2             = _mm_setzero_pd();
 170         fiy2             = _mm_setzero_pd();
 171         fiz2             = _mm_setzero_pd();
 172         fix3             = _mm_setzero_pd();
 173         fiy3             = _mm_setzero_pd();
 174         fiz3             = _mm_setzero_pd();
 175
 176         /* Reset potential sums */
 177         velecsum         = _mm_setzero_pd();
 178         vvdwsum          = _mm_setzero_pd();
 179
 180         /* Start inner kernel loop */
 181         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 182         {
 183
 184             /* Get j neighbor index, and coordinate index */
 185             jnrA             = jjnr[jidx];
 186             jnrB             = jjnr[jidx+1];
 187             j_coord_offsetA  = DIM*jnrA;
 188             j_coord_offsetB  = DIM*jnrB;
 189
 190             /* load j atom coordinates */
 191             gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 192                                               &jx0,&jy0,&jz0);
 193
 194             /* Calculate displacement vector */
 195             dx00             = _mm_sub_pd(ix0,jx0);
 196             dy00             = _mm_sub_pd(iy0,jy0);
 197             dz00             = _mm_sub_pd(iz0,jz0);
 198             dx10             = _mm_sub_pd(ix1,jx0);
 199             dy10             = _mm_sub_pd(iy1,jy0);
 200             dz10             = _mm_sub_pd(iz1,jz0);
 201             dx20             = _mm_sub_pd(ix2,jx0);
 202             dy20             = _mm_sub_pd(iy2,jy0);
 203             dz20             = _mm_sub_pd(iz2,jz0);
 204             dx30             = _mm_sub_pd(ix3,jx0);
 205             dy30             = _mm_sub_pd(iy3,jy0);
 206             dz30             = _mm_sub_pd(iz3,jz0);
 207
 208             /* Calculate squared distance and things based on it */
 209             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 210             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
 211             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
 212             rsq30            = gmx_mm_calc_rsq_pd(dx30,dy30,dz30);
 213
 214             rinv10           = sse2_invsqrt_d(rsq10);
 215             rinv20           = sse2_invsqrt_d(rsq20);
 216             rinv30           = sse2_invsqrt_d(rsq30);
 217
 218             rinvsq00         = sse2_inv_d(rsq00);
 219             rinvsq10         = _mm_mul_pd(rinv10,rinv10);
 220             rinvsq20         = _mm_mul_pd(rinv20,rinv20);
 221             rinvsq30         = _mm_mul_pd(rinv30,rinv30);
 222
 223             /* Load parameters for j particles */
 224             jq0              = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
 225             vdwjidx0A        = 2*vdwtype[jnrA+0];
 226             vdwjidx0B        = 2*vdwtype[jnrB+0];
 227
 228             fjx0             = _mm_setzero_pd();
 229             fjy0             = _mm_setzero_pd();
 230             fjz0             = _mm_setzero_pd();
 231
 232             /**************************
 233              * CALCULATE INTERACTIONS *
 234              **************************/
 235
 236             /* Compute parameters for interactions between i and j atoms */
 237             gmx_mm_load_2pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,
 238                                          vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
 239
 240             /* LENNARD-JONES DISPERSION/REPULSION */
 241
 242             rinvsix          = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 243             vvdw6            = _mm_mul_pd(c6_00,rinvsix);
 244             vvdw12           = _mm_mul_pd(c12_00,_mm_mul_pd(rinvsix,rinvsix));
 245             vvdw             = _mm_sub_pd( _mm_mul_pd(vvdw12,one_twelfth) , _mm_mul_pd(vvdw6,one_sixth) );
 246             fvdw             = _mm_mul_pd(_mm_sub_pd(vvdw12,vvdw6),rinvsq00);
 247
 248             /* Update potential sum for this i atom from the interaction with this j atom. */
 249             vvdwsum          = _mm_add_pd(vvdwsum,vvdw);
 250
 251             fscal            = fvdw;
 252
 253             /* Calculate temporary vectorial force */
 254             tx               = _mm_mul_pd(fscal,dx00);
 255             ty               = _mm_mul_pd(fscal,dy00);
 256             tz               = _mm_mul_pd(fscal,dz00);
 257
 258             /* Update vectorial force */
 259             fix0             = _mm_add_pd(fix0,tx);
 260             fiy0             = _mm_add_pd(fiy0,ty);
 261             fiz0             = _mm_add_pd(fiz0,tz);
 262
 263             fjx0             = _mm_add_pd(fjx0,tx);
 264             fjy0             = _mm_add_pd(fjy0,ty);
 265             fjz0             = _mm_add_pd(fjz0,tz);
 266
 267             /**************************
 268              * CALCULATE INTERACTIONS *
 269              **************************/
 270
 271             r10              = _mm_mul_pd(rsq10,rinv10);
 272
 273             /* Compute parameters for interactions between i and j atoms */
 274             qq10             = _mm_mul_pd(iq1,jq0);
 275
 276             /* EWALD ELECTROSTATICS */
 277
 278             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 279             ewrt             = _mm_mul_pd(r10,ewtabscale);
 280             ewitab           = _mm_cvttpd_epi32(ewrt);
 281             eweps            = _mm_sub_pd(ewrt,_mm_cvtepi32_pd(ewitab));
 282             ewitab           = _mm_slli_epi32(ewitab,2);
 283             ewtabF           = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 284             ewtabD           = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,1) );
 285             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 286             ewtabV           = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,0) +2);
 287             ewtabFn          = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,1) +2);
 288             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 289             felec            = _mm_add_pd(ewtabF,_mm_mul_pd(eweps,ewtabD));
 290             velec            = _mm_sub_pd(ewtabV,_mm_mul_pd(_mm_mul_pd(ewtabhalfspace,eweps),_mm_add_pd(ewtabF,felec)));
 291             velec            = _mm_mul_pd(qq10,_mm_sub_pd(rinv10,velec));
 292             felec            = _mm_mul_pd(_mm_mul_pd(qq10,rinv10),_mm_sub_pd(rinvsq10,felec));
 293
 294             /* Update potential sum for this i atom from the interaction with this j atom. */
 295             velecsum         = _mm_add_pd(velecsum,velec);
 296
 297             fscal            = felec;
 298
 299             /* Calculate temporary vectorial force */
 300             tx               = _mm_mul_pd(fscal,dx10);
 301             ty               = _mm_mul_pd(fscal,dy10);
 302             tz               = _mm_mul_pd(fscal,dz10);
 303
 304             /* Update vectorial force */
 305             fix1             = _mm_add_pd(fix1,tx);
 306             fiy1             = _mm_add_pd(fiy1,ty);
 307             fiz1             = _mm_add_pd(fiz1,tz);
 308
 309             fjx0             = _mm_add_pd(fjx0,tx);
 310             fjy0             = _mm_add_pd(fjy0,ty);
 311             fjz0             = _mm_add_pd(fjz0,tz);
 312
 313             /**************************
 314              * CALCULATE INTERACTIONS *
 315              **************************/
 316
 317             r20              = _mm_mul_pd(rsq20,rinv20);
 318
 319             /* Compute parameters for interactions between i and j atoms */
 320             qq20             = _mm_mul_pd(iq2,jq0);
 321
 322             /* EWALD ELECTROSTATICS */
 323
 324             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 325             ewrt             = _mm_mul_pd(r20,ewtabscale);
 326             ewitab           = _mm_cvttpd_epi32(ewrt);
 327             eweps            = _mm_sub_pd(ewrt,_mm_cvtepi32_pd(ewitab));
 328             ewitab           = _mm_slli_epi32(ewitab,2);
 329             ewtabF           = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 330             ewtabD           = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,1) );
 331             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 332             ewtabV           = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,0) +2);
 333             ewtabFn          = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,1) +2);
 334             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 335             felec            = _mm_add_pd(ewtabF,_mm_mul_pd(eweps,ewtabD));
 336             velec            = _mm_sub_pd(ewtabV,_mm_mul_pd(_mm_mul_pd(ewtabhalfspace,eweps),_mm_add_pd(ewtabF,felec)));
 337             velec            = _mm_mul_pd(qq20,_mm_sub_pd(rinv20,velec));
 338             felec            = _mm_mul_pd(_mm_mul_pd(qq20,rinv20),_mm_sub_pd(rinvsq20,felec));
 339
 340             /* Update potential sum for this i atom from the interaction with this j atom. */
 341             velecsum         = _mm_add_pd(velecsum,velec);
 342
 343             fscal            = felec;
 344
 345             /* Calculate temporary vectorial force */
 346             tx               = _mm_mul_pd(fscal,dx20);
 347             ty               = _mm_mul_pd(fscal,dy20);
 348             tz               = _mm_mul_pd(fscal,dz20);
 349
 350             /* Update vectorial force */
 351             fix2             = _mm_add_pd(fix2,tx);
 352             fiy2             = _mm_add_pd(fiy2,ty);
 353             fiz2             = _mm_add_pd(fiz2,tz);
 354
 355             fjx0             = _mm_add_pd(fjx0,tx);
 356             fjy0             = _mm_add_pd(fjy0,ty);
 357             fjz0             = _mm_add_pd(fjz0,tz);
 358
 359             /**************************
 360              * CALCULATE INTERACTIONS *
 361              **************************/
 362
 363             r30              = _mm_mul_pd(rsq30,rinv30);
 364
 365             /* Compute parameters for interactions between i and j atoms */
 366             qq30             = _mm_mul_pd(iq3,jq0);
 367
 368             /* EWALD ELECTROSTATICS */
 369
 370             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 371             ewrt             = _mm_mul_pd(r30,ewtabscale);
 372             ewitab           = _mm_cvttpd_epi32(ewrt);
 373             eweps            = _mm_sub_pd(ewrt,_mm_cvtepi32_pd(ewitab));
 374             ewitab           = _mm_slli_epi32(ewitab,2);
 375             ewtabF           = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 376             ewtabD           = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,1) );
 377             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 378             ewtabV           = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,0) +2);
 379             ewtabFn          = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,1) +2);
 380             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 381             felec            = _mm_add_pd(ewtabF,_mm_mul_pd(eweps,ewtabD));
 382             velec            = _mm_sub_pd(ewtabV,_mm_mul_pd(_mm_mul_pd(ewtabhalfspace,eweps),_mm_add_pd(ewtabF,felec)));
 383             velec            = _mm_mul_pd(qq30,_mm_sub_pd(rinv30,velec));
 384             felec            = _mm_mul_pd(_mm_mul_pd(qq30,rinv30),_mm_sub_pd(rinvsq30,felec));
 385
 386             /* Update potential sum for this i atom from the interaction with this j atom. */
 387             velecsum         = _mm_add_pd(velecsum,velec);
 388
 389             fscal            = felec;
 390
 391             /* Calculate temporary vectorial force */
 392             tx               = _mm_mul_pd(fscal,dx30);
 393             ty               = _mm_mul_pd(fscal,dy30);
 394             tz               = _mm_mul_pd(fscal,dz30);
 395
 396             /* Update vectorial force */
 397             fix3             = _mm_add_pd(fix3,tx);
 398             fiy3             = _mm_add_pd(fiy3,ty);
 399             fiz3             = _mm_add_pd(fiz3,tz);
 400
 401             fjx0             = _mm_add_pd(fjx0,tx);
 402             fjy0             = _mm_add_pd(fjy0,ty);
 403             fjz0             = _mm_add_pd(fjz0,tz);
 404
 405             gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,fjx0,fjy0,fjz0);
 406
 407             /* Inner loop uses 158 flops */
 408         }
 409
 410         if(jidx<j_index_end)
 411         {
 412
 413             jnrA             = jjnr[jidx];
 414             j_coord_offsetA  = DIM*jnrA;
 415
 416             /* load j atom coordinates */
 417             gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
 418                                               &jx0,&jy0,&jz0);
 419
 420             /* Calculate displacement vector */
 421             dx00             = _mm_sub_pd(ix0,jx0);
 422             dy00             = _mm_sub_pd(iy0,jy0);
 423             dz00             = _mm_sub_pd(iz0,jz0);
 424             dx10             = _mm_sub_pd(ix1,jx0);
 425             dy10             = _mm_sub_pd(iy1,jy0);
 426             dz10             = _mm_sub_pd(iz1,jz0);
 427             dx20             = _mm_sub_pd(ix2,jx0);
 428             dy20             = _mm_sub_pd(iy2,jy0);
 429             dz20             = _mm_sub_pd(iz2,jz0);
 430             dx30             = _mm_sub_pd(ix3,jx0);
 431             dy30             = _mm_sub_pd(iy3,jy0);
 432             dz30             = _mm_sub_pd(iz3,jz0);
 433
 434             /* Calculate squared distance and things based on it */
 435             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 436             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
 437             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
 438             rsq30            = gmx_mm_calc_rsq_pd(dx30,dy30,dz30);
 439
 440             rinv10           = sse2_invsqrt_d(rsq10);
 441             rinv20           = sse2_invsqrt_d(rsq20);
 442             rinv30           = sse2_invsqrt_d(rsq30);
 443
 444             rinvsq00         = sse2_inv_d(rsq00);
 445             rinvsq10         = _mm_mul_pd(rinv10,rinv10);
 446             rinvsq20         = _mm_mul_pd(rinv20,rinv20);
 447             rinvsq30         = _mm_mul_pd(rinv30,rinv30);
 448
 449             /* Load parameters for j particles */
 450             jq0              = _mm_load_sd(charge+jnrA+0);
 451             vdwjidx0A        = 2*vdwtype[jnrA+0];
 452
 453             fjx0             = _mm_setzero_pd();
 454             fjy0             = _mm_setzero_pd();
 455             fjz0             = _mm_setzero_pd();
 456
 457             /**************************
 458              * CALCULATE INTERACTIONS *
 459              **************************/
 460
 461             /* Compute parameters for interactions between i and j atoms */
 462             gmx_mm_load_1pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,&c6_00,&c12_00);
 463
 464             /* LENNARD-JONES DISPERSION/REPULSION */
 465
 466             rinvsix          = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 467             vvdw6            = _mm_mul_pd(c6_00,rinvsix);
 468             vvdw12           = _mm_mul_pd(c12_00,_mm_mul_pd(rinvsix,rinvsix));
 469             vvdw             = _mm_sub_pd( _mm_mul_pd(vvdw12,one_twelfth) , _mm_mul_pd(vvdw6,one_sixth) );
 470             fvdw             = _mm_mul_pd(_mm_sub_pd(vvdw12,vvdw6),rinvsq00);
 471
 472             /* Update potential sum for this i atom from the interaction with this j atom. */
 473             vvdw             = _mm_unpacklo_pd(vvdw,_mm_setzero_pd());
 474             vvdwsum          = _mm_add_pd(vvdwsum,vvdw);
 475
 476             fscal            = fvdw;
 477
 478             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 479
 480             /* Calculate temporary vectorial force */
 481             tx               = _mm_mul_pd(fscal,dx00);
 482             ty               = _mm_mul_pd(fscal,dy00);
 483             tz               = _mm_mul_pd(fscal,dz00);
 484
 485             /* Update vectorial force */
 486             fix0             = _mm_add_pd(fix0,tx);
 487             fiy0             = _mm_add_pd(fiy0,ty);
 488             fiz0             = _mm_add_pd(fiz0,tz);
 489
 490             fjx0             = _mm_add_pd(fjx0,tx);
 491             fjy0             = _mm_add_pd(fjy0,ty);
 492             fjz0             = _mm_add_pd(fjz0,tz);
 493
 494             /**************************
 495              * CALCULATE INTERACTIONS *
 496              **************************/
 497
 498             r10              = _mm_mul_pd(rsq10,rinv10);
 499
 500             /* Compute parameters for interactions between i and j atoms */
 501             qq10             = _mm_mul_pd(iq1,jq0);
 502
 503             /* EWALD ELECTROSTATICS */
 504
 505             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 506             ewrt             = _mm_mul_pd(r10,ewtabscale);
 507             ewitab           = _mm_cvttpd_epi32(ewrt);
 508             eweps            = _mm_sub_pd(ewrt,_mm_cvtepi32_pd(ewitab));
 509             ewitab           = _mm_slli_epi32(ewitab,2);
 510             ewtabF           = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 511             ewtabD           = _mm_setzero_pd();
 512             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 513             ewtabV           = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,0) +2);
 514             ewtabFn          = _mm_setzero_pd();
 515             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 516             felec            = _mm_add_pd(ewtabF,_mm_mul_pd(eweps,ewtabD));
 517             velec            = _mm_sub_pd(ewtabV,_mm_mul_pd(_mm_mul_pd(ewtabhalfspace,eweps),_mm_add_pd(ewtabF,felec)));
 518             velec            = _mm_mul_pd(qq10,_mm_sub_pd(rinv10,velec));
 519             felec            = _mm_mul_pd(_mm_mul_pd(qq10,rinv10),_mm_sub_pd(rinvsq10,felec));
 520
 521             /* Update potential sum for this i atom from the interaction with this j atom. */
 522             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 523             velecsum         = _mm_add_pd(velecsum,velec);
 524
 525             fscal            = felec;
 526
 527             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 528
 529             /* Calculate temporary vectorial force */
 530             tx               = _mm_mul_pd(fscal,dx10);
 531             ty               = _mm_mul_pd(fscal,dy10);
 532             tz               = _mm_mul_pd(fscal,dz10);
 533
 534             /* Update vectorial force */
 535             fix1             = _mm_add_pd(fix1,tx);
 536             fiy1             = _mm_add_pd(fiy1,ty);
 537             fiz1             = _mm_add_pd(fiz1,tz);
 538
 539             fjx0             = _mm_add_pd(fjx0,tx);
 540             fjy0             = _mm_add_pd(fjy0,ty);
 541             fjz0             = _mm_add_pd(fjz0,tz);
 542
 543             /**************************
 544              * CALCULATE INTERACTIONS *
 545              **************************/
 546
 547             r20              = _mm_mul_pd(rsq20,rinv20);
 548
 549             /* Compute parameters for interactions between i and j atoms */
 550             qq20             = _mm_mul_pd(iq2,jq0);
 551
 552             /* EWALD ELECTROSTATICS */
 553
 554             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 555             ewrt             = _mm_mul_pd(r20,ewtabscale);
 556             ewitab           = _mm_cvttpd_epi32(ewrt);
 557             eweps            = _mm_sub_pd(ewrt,_mm_cvtepi32_pd(ewitab));
 558             ewitab           = _mm_slli_epi32(ewitab,2);
 559             ewtabF           = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 560             ewtabD           = _mm_setzero_pd();
 561             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 562             ewtabV           = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,0) +2);
 563             ewtabFn          = _mm_setzero_pd();
 564             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 565             felec            = _mm_add_pd(ewtabF,_mm_mul_pd(eweps,ewtabD));
 566             velec            = _mm_sub_pd(ewtabV,_mm_mul_pd(_mm_mul_pd(ewtabhalfspace,eweps),_mm_add_pd(ewtabF,felec)));
 567             velec            = _mm_mul_pd(qq20,_mm_sub_pd(rinv20,velec));
 568             felec            = _mm_mul_pd(_mm_mul_pd(qq20,rinv20),_mm_sub_pd(rinvsq20,felec));
 569
 570             /* Update potential sum for this i atom from the interaction with this j atom. */
 571             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 572             velecsum         = _mm_add_pd(velecsum,velec);
 573
 574             fscal            = felec;
 575
 576             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 577
 578             /* Calculate temporary vectorial force */
 579             tx               = _mm_mul_pd(fscal,dx20);
 580             ty               = _mm_mul_pd(fscal,dy20);
 581             tz               = _mm_mul_pd(fscal,dz20);
 582
 583             /* Update vectorial force */
 584             fix2             = _mm_add_pd(fix2,tx);
 585             fiy2             = _mm_add_pd(fiy2,ty);
 586             fiz2             = _mm_add_pd(fiz2,tz);
 587
 588             fjx0             = _mm_add_pd(fjx0,tx);
 589             fjy0             = _mm_add_pd(fjy0,ty);
 590             fjz0             = _mm_add_pd(fjz0,tz);
 591
 592             /**************************
 593              * CALCULATE INTERACTIONS *
 594              **************************/
 595
 596             r30              = _mm_mul_pd(rsq30,rinv30);
 597
 598             /* Compute parameters for interactions between i and j atoms */
 599             qq30             = _mm_mul_pd(iq3,jq0);
 600
 601             /* EWALD ELECTROSTATICS */
 602
 603             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 604             ewrt             = _mm_mul_pd(r30,ewtabscale);
 605             ewitab           = _mm_cvttpd_epi32(ewrt);
 606             eweps            = _mm_sub_pd(ewrt,_mm_cvtepi32_pd(ewitab));
 607             ewitab           = _mm_slli_epi32(ewitab,2);
 608             ewtabF           = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 609             ewtabD           = _mm_setzero_pd();
 610             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 611             ewtabV           = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,0) +2);
 612             ewtabFn          = _mm_setzero_pd();
 613             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 614             felec            = _mm_add_pd(ewtabF,_mm_mul_pd(eweps,ewtabD));
 615             velec            = _mm_sub_pd(ewtabV,_mm_mul_pd(_mm_mul_pd(ewtabhalfspace,eweps),_mm_add_pd(ewtabF,felec)));
 616             velec            = _mm_mul_pd(qq30,_mm_sub_pd(rinv30,velec));
 617             felec            = _mm_mul_pd(_mm_mul_pd(qq30,rinv30),_mm_sub_pd(rinvsq30,felec));
 618
 619             /* Update potential sum for this i atom from the interaction with this j atom. */
 620             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 621             velecsum         = _mm_add_pd(velecsum,velec);
 622
 623             fscal            = felec;
 624
 625             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 626
 627             /* Calculate temporary vectorial force */
 628             tx               = _mm_mul_pd(fscal,dx30);
 629             ty               = _mm_mul_pd(fscal,dy30);
 630             tz               = _mm_mul_pd(fscal,dz30);
 631
 632             /* Update vectorial force */
 633             fix3             = _mm_add_pd(fix3,tx);
 634             fiy3             = _mm_add_pd(fiy3,ty);
 635             fiz3             = _mm_add_pd(fiz3,tz);
 636
 637             fjx0             = _mm_add_pd(fjx0,tx);
 638             fjy0             = _mm_add_pd(fjy0,ty);
 639             fjz0             = _mm_add_pd(fjz0,tz);
 640
 641             gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,fjx0,fjy0,fjz0);
 642
 643             /* Inner loop uses 158 flops */
 644         }
 645
 646         /* End of innermost loop */
 647
 648         gmx_mm_update_iforce_4atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
 649                                               f+i_coord_offset,fshift+i_shift_offset);
 650
 651         ggid                        = gid[iidx];
 652         /* Update potential energies */
 653         gmx_mm_update_1pot_pd(velecsum,kernel_data->energygrp_elec+ggid);
 654         gmx_mm_update_1pot_pd(vvdwsum,kernel_data->energygrp_vdw+ggid);
 655
 656         /* Increment number of inner iterations */
 657         inneriter                  += j_index_end - j_index_start;
 658
 659         /* Outer loop uses 26 flops */
 660     }
 661
 662     /* Increment number of outer iterations */
 663     outeriter        += nri;
 664
 665     /* Update outer/inner flops */
 666
 667     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W4_VF,outeriter*26 + inneriter*158);
 668 }
 669 /*
 670  * Gromacs nonbonded kernel:   nb_kernel_ElecEw_VdwLJ_GeomW4P1_F_sse2_double
 671  * Electrostatics interaction: Ewald
 672  * VdW interaction:            LennardJones
 673  * Geometry:                   Water4-Particle
 674  * Calculate force/pot:        Force
 675  */
 676 void
 677 nb_kernel_ElecEw_VdwLJ_GeomW4P1_F_sse2_double
 678                     (t_nblist                    * gmx_restrict       nlist,
 679                      rvec                        * gmx_restrict          xx,
 680                      rvec                        * gmx_restrict          ff,
 681                      struct t_forcerec           * gmx_restrict          fr,
 682                      t_mdatoms                   * gmx_restrict     mdatoms,
 683                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 684                      t_nrnb                      * gmx_restrict        nrnb)
 685 {
 686     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 687      * just 0 for non-waters.
 688      * Suffixes A,B refer to j loop unrolling done with SSE double precision, e.g. for the two different
 689      * jnr indices corresponding to data put in the four positions in the SIMD register.
 690      */
 691     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 692     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 693     int              jnrA,jnrB;
 694     int              j_coord_offsetA,j_coord_offsetB;
 695     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 696     real             rcutoff_scalar;
 697     real             *shiftvec,*fshift,*x,*f;
 698     __m128d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 699     int              vdwioffset0;
 700     __m128d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 701     int              vdwioffset1;
 702     __m128d          ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
 703     int              vdwioffset2;
 704     __m128d          ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
 705     int              vdwioffset3;
 706     __m128d          ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
 707     int              vdwjidx0A,vdwjidx0B;
 708     __m128d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 709     __m128d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 710     __m128d          dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
 711     __m128d          dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
 712     __m128d          dx30,dy30,dz30,rsq30,rinv30,rinvsq30,r30,qq30,c6_30,c12_30;
 713     __m128d          velec,felec,velecsum,facel,crf,krf,krf2;
 714     real             *charge;
 715     int              nvdwtype;
 716     __m128d          rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 717     int              *vdwtype;
 718     real             *vdwparam;
 719     __m128d          one_sixth   = _mm_set1_pd(1.0/6.0);
 720     __m128d          one_twelfth = _mm_set1_pd(1.0/12.0);
 721     __m128i          ewitab;
 722     __m128d          ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
 723     real             *ewtab;
 724     __m128d          dummy_mask,cutoff_mask;
 725     __m128d          signbit   = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
 726     __m128d          one     = _mm_set1_pd(1.0);
 727     __m128d          two     = _mm_set1_pd(2.0);
 728     x                = xx[0];
 729     f                = ff[0];
 730
 731     nri              = nlist->nri;
 732     iinr             = nlist->iinr;
 733     jindex           = nlist->jindex;
 734     jjnr             = nlist->jjnr;
 735     shiftidx         = nlist->shift;
 736     gid              = nlist->gid;
 737     shiftvec         = fr->shift_vec[0];
 738     fshift           = fr->fshift[0];
 739     facel            = _mm_set1_pd(fr->ic->epsfac);
 740     charge           = mdatoms->chargeA;
 741     nvdwtype         = fr->ntype;
 742     vdwparam         = fr->nbfp;
 743     vdwtype          = mdatoms->typeA;
 744
 745     sh_ewald         = _mm_set1_pd(fr->ic->sh_ewald);
 746     ewtab            = fr->ic->tabq_coul_F;
 747     ewtabscale       = _mm_set1_pd(fr->ic->tabq_scale);
 748     ewtabhalfspace   = _mm_set1_pd(0.5/fr->ic->tabq_scale);
 749
 750     /* Setup water-specific parameters */
 751     inr              = nlist->iinr[0];
 752     iq1              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+1]));
 753     iq2              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+2]));
 754     iq3              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+3]));
 755     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 756
 757     /* Avoid stupid compiler warnings */
 758     jnrA = jnrB = 0;
 759     j_coord_offsetA = 0;
 760     j_coord_offsetB = 0;
 761
 762     outeriter        = 0;
 763     inneriter        = 0;
 764
 765     /* Start outer loop over neighborlists */
 766     for(iidx=0; iidx<nri; iidx++)
 767     {
 768         /* Load shift vector for this list */
 769         i_shift_offset   = DIM*shiftidx[iidx];
 770
 771         /* Load limits for loop over neighbors */
 772         j_index_start    = jindex[iidx];
 773         j_index_end      = jindex[iidx+1];
 774
 775         /* Get outer coordinate index */
 776         inr              = iinr[iidx];
 777         i_coord_offset   = DIM*inr;
 778
 779         /* Load i particle coords and add shift vector */
 780         gmx_mm_load_shift_and_4rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
 781                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
 782
 783         fix0             = _mm_setzero_pd();
 784         fiy0             = _mm_setzero_pd();
 785         fiz0             = _mm_setzero_pd();
 786         fix1             = _mm_setzero_pd();
 787         fiy1             = _mm_setzero_pd();
 788         fiz1             = _mm_setzero_pd();
 789         fix2             = _mm_setzero_pd();
 790         fiy2             = _mm_setzero_pd();
 791         fiz2             = _mm_setzero_pd();
 792         fix3             = _mm_setzero_pd();
 793         fiy3             = _mm_setzero_pd();
 794         fiz3             = _mm_setzero_pd();
 795
 796         /* Start inner kernel loop */
 797         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 798         {
 799
 800             /* Get j neighbor index, and coordinate index */
 801             jnrA             = jjnr[jidx];
 802             jnrB             = jjnr[jidx+1];
 803             j_coord_offsetA  = DIM*jnrA;
 804             j_coord_offsetB  = DIM*jnrB;
 805
 806             /* load j atom coordinates */
 807             gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 808                                               &jx0,&jy0,&jz0);
 809
 810             /* Calculate displacement vector */
 811             dx00             = _mm_sub_pd(ix0,jx0);
 812             dy00             = _mm_sub_pd(iy0,jy0);
 813             dz00             = _mm_sub_pd(iz0,jz0);
 814             dx10             = _mm_sub_pd(ix1,jx0);
 815             dy10             = _mm_sub_pd(iy1,jy0);
 816             dz10             = _mm_sub_pd(iz1,jz0);
 817             dx20             = _mm_sub_pd(ix2,jx0);
 818             dy20             = _mm_sub_pd(iy2,jy0);
 819             dz20             = _mm_sub_pd(iz2,jz0);
 820             dx30             = _mm_sub_pd(ix3,jx0);
 821             dy30             = _mm_sub_pd(iy3,jy0);
 822             dz30             = _mm_sub_pd(iz3,jz0);
 823
 824             /* Calculate squared distance and things based on it */
 825             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 826             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
 827             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
 828             rsq30            = gmx_mm_calc_rsq_pd(dx30,dy30,dz30);
 829
 830             rinv10           = sse2_invsqrt_d(rsq10);
 831             rinv20           = sse2_invsqrt_d(rsq20);
 832             rinv30           = sse2_invsqrt_d(rsq30);
 833
 834             rinvsq00         = sse2_inv_d(rsq00);
 835             rinvsq10         = _mm_mul_pd(rinv10,rinv10);
 836             rinvsq20         = _mm_mul_pd(rinv20,rinv20);
 837             rinvsq30         = _mm_mul_pd(rinv30,rinv30);
 838
 839             /* Load parameters for j particles */
 840             jq0              = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
 841             vdwjidx0A        = 2*vdwtype[jnrA+0];
 842             vdwjidx0B        = 2*vdwtype[jnrB+0];
 843
 844             fjx0             = _mm_setzero_pd();
 845             fjy0             = _mm_setzero_pd();
 846             fjz0             = _mm_setzero_pd();
 847
 848             /**************************
 849              * CALCULATE INTERACTIONS *
 850              **************************/
 851
 852             /* Compute parameters for interactions between i and j atoms */
 853             gmx_mm_load_2pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,
 854                                          vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
 855
 856             /* LENNARD-JONES DISPERSION/REPULSION */
 857
 858             rinvsix          = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 859             fvdw             = _mm_mul_pd(_mm_sub_pd(_mm_mul_pd(c12_00,rinvsix),c6_00),_mm_mul_pd(rinvsix,rinvsq00));
 860
 861             fscal            = fvdw;
 862
 863             /* Calculate temporary vectorial force */
 864             tx               = _mm_mul_pd(fscal,dx00);
 865             ty               = _mm_mul_pd(fscal,dy00);
 866             tz               = _mm_mul_pd(fscal,dz00);
 867
 868             /* Update vectorial force */
 869             fix0             = _mm_add_pd(fix0,tx);
 870             fiy0             = _mm_add_pd(fiy0,ty);
 871             fiz0             = _mm_add_pd(fiz0,tz);
 872
 873             fjx0             = _mm_add_pd(fjx0,tx);
 874             fjy0             = _mm_add_pd(fjy0,ty);
 875             fjz0             = _mm_add_pd(fjz0,tz);
 876
 877             /**************************
 878              * CALCULATE INTERACTIONS *
 879              **************************/
 880
 881             r10              = _mm_mul_pd(rsq10,rinv10);
 882
 883             /* Compute parameters for interactions between i and j atoms */
 884             qq10             = _mm_mul_pd(iq1,jq0);
 885
 886             /* EWALD ELECTROSTATICS */
 887
 888             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 889             ewrt             = _mm_mul_pd(r10,ewtabscale);
 890             ewitab           = _mm_cvttpd_epi32(ewrt);
 891             eweps            = _mm_sub_pd(ewrt,_mm_cvtepi32_pd(ewitab));
 892             gmx_mm_load_2pair_swizzle_pd(ewtab+gmx_mm_extract_epi32(ewitab,0),ewtab+gmx_mm_extract_epi32(ewitab,1),
 893                                          &ewtabF,&ewtabFn);
 894             felec            = _mm_add_pd(_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF),_mm_mul_pd(eweps,ewtabFn));
 895             felec            = _mm_mul_pd(_mm_mul_pd(qq10,rinv10),_mm_sub_pd(rinvsq10,felec));
 896
 897             fscal            = felec;
 898
 899             /* Calculate temporary vectorial force */
 900             tx               = _mm_mul_pd(fscal,dx10);
 901             ty               = _mm_mul_pd(fscal,dy10);
 902             tz               = _mm_mul_pd(fscal,dz10);
 903
 904             /* Update vectorial force */
 905             fix1             = _mm_add_pd(fix1,tx);
 906             fiy1             = _mm_add_pd(fiy1,ty);
 907             fiz1             = _mm_add_pd(fiz1,tz);
 908
 909             fjx0             = _mm_add_pd(fjx0,tx);
 910             fjy0             = _mm_add_pd(fjy0,ty);
 911             fjz0             = _mm_add_pd(fjz0,tz);
 912
 913             /**************************
 914              * CALCULATE INTERACTIONS *
 915              **************************/
 916
 917             r20              = _mm_mul_pd(rsq20,rinv20);
 918
 919             /* Compute parameters for interactions between i and j atoms */
 920             qq20             = _mm_mul_pd(iq2,jq0);
 921
 922             /* EWALD ELECTROSTATICS */
 923
 924             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 925             ewrt             = _mm_mul_pd(r20,ewtabscale);
 926             ewitab           = _mm_cvttpd_epi32(ewrt);
 927             eweps            = _mm_sub_pd(ewrt,_mm_cvtepi32_pd(ewitab));
 928             gmx_mm_load_2pair_swizzle_pd(ewtab+gmx_mm_extract_epi32(ewitab,0),ewtab+gmx_mm_extract_epi32(ewitab,1),
 929                                          &ewtabF,&ewtabFn);
 930             felec            = _mm_add_pd(_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF),_mm_mul_pd(eweps,ewtabFn));
 931             felec            = _mm_mul_pd(_mm_mul_pd(qq20,rinv20),_mm_sub_pd(rinvsq20,felec));
 932
 933             fscal            = felec;
 934
 935             /* Calculate temporary vectorial force */
 936             tx               = _mm_mul_pd(fscal,dx20);
 937             ty               = _mm_mul_pd(fscal,dy20);
 938             tz               = _mm_mul_pd(fscal,dz20);
 939
 940             /* Update vectorial force */
 941             fix2             = _mm_add_pd(fix2,tx);
 942             fiy2             = _mm_add_pd(fiy2,ty);
 943             fiz2             = _mm_add_pd(fiz2,tz);
 944
 945             fjx0             = _mm_add_pd(fjx0,tx);
 946             fjy0             = _mm_add_pd(fjy0,ty);
 947             fjz0             = _mm_add_pd(fjz0,tz);
 948
 949             /**************************
 950              * CALCULATE INTERACTIONS *
 951              **************************/
 952
 953             r30              = _mm_mul_pd(rsq30,rinv30);
 954
 955             /* Compute parameters for interactions between i and j atoms */
 956             qq30             = _mm_mul_pd(iq3,jq0);
 957
 958             /* EWALD ELECTROSTATICS */
 959
 960             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 961             ewrt             = _mm_mul_pd(r30,ewtabscale);
 962             ewitab           = _mm_cvttpd_epi32(ewrt);
 963             eweps            = _mm_sub_pd(ewrt,_mm_cvtepi32_pd(ewitab));
 964             gmx_mm_load_2pair_swizzle_pd(ewtab+gmx_mm_extract_epi32(ewitab,0),ewtab+gmx_mm_extract_epi32(ewitab,1),
 965                                          &ewtabF,&ewtabFn);
 966             felec            = _mm_add_pd(_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF),_mm_mul_pd(eweps,ewtabFn));
 967             felec            = _mm_mul_pd(_mm_mul_pd(qq30,rinv30),_mm_sub_pd(rinvsq30,felec));
 968
 969             fscal            = felec;
 970
 971             /* Calculate temporary vectorial force */
 972             tx               = _mm_mul_pd(fscal,dx30);
 973             ty               = _mm_mul_pd(fscal,dy30);
 974             tz               = _mm_mul_pd(fscal,dz30);
 975
 976             /* Update vectorial force */
 977             fix3             = _mm_add_pd(fix3,tx);
 978             fiy3             = _mm_add_pd(fiy3,ty);
 979             fiz3             = _mm_add_pd(fiz3,tz);
 980
 981             fjx0             = _mm_add_pd(fjx0,tx);
 982             fjy0             = _mm_add_pd(fjy0,ty);
 983             fjz0             = _mm_add_pd(fjz0,tz);
 984
 985             gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,fjx0,fjy0,fjz0);
 986
 987             /* Inner loop uses 138 flops */
 988         }
 989
 990         if(jidx<j_index_end)
 991         {
 992
 993             jnrA             = jjnr[jidx];
 994             j_coord_offsetA  = DIM*jnrA;
 995
 996             /* load j atom coordinates */
 997             gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
 998                                               &jx0,&jy0,&jz0);
 999
1000             /* Calculate displacement vector */
1001             dx00             = _mm_sub_pd(ix0,jx0);
1002             dy00             = _mm_sub_pd(iy0,jy0);
1003             dz00             = _mm_sub_pd(iz0,jz0);
1004             dx10             = _mm_sub_pd(ix1,jx0);
1005             dy10             = _mm_sub_pd(iy1,jy0);
1006             dz10             = _mm_sub_pd(iz1,jz0);
1007             dx20             = _mm_sub_pd(ix2,jx0);
1008             dy20             = _mm_sub_pd(iy2,jy0);
1009             dz20             = _mm_sub_pd(iz2,jz0);
1010             dx30             = _mm_sub_pd(ix3,jx0);
1011             dy30             = _mm_sub_pd(iy3,jy0);
1012             dz30             = _mm_sub_pd(iz3,jz0);
1013
1014             /* Calculate squared distance and things based on it */
1015             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
1016             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
1017             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
1018             rsq30            = gmx_mm_calc_rsq_pd(dx30,dy30,dz30);
1019
1020             rinv10           = sse2_invsqrt_d(rsq10);
1021             rinv20           = sse2_invsqrt_d(rsq20);
1022             rinv30           = sse2_invsqrt_d(rsq30);
1023
1024             rinvsq00         = sse2_inv_d(rsq00);
1025             rinvsq10         = _mm_mul_pd(rinv10,rinv10);
1026             rinvsq20         = _mm_mul_pd(rinv20,rinv20);
1027             rinvsq30         = _mm_mul_pd(rinv30,rinv30);
1028
1029             /* Load parameters for j particles */
1030             jq0              = _mm_load_sd(charge+jnrA+0);
1031             vdwjidx0A        = 2*vdwtype[jnrA+0];
1032
1033             fjx0             = _mm_setzero_pd();
1034             fjy0             = _mm_setzero_pd();
1035             fjz0             = _mm_setzero_pd();
1036
1037             /**************************
1038              * CALCULATE INTERACTIONS *
1039              **************************/
1040
1041             /* Compute parameters for interactions between i and j atoms */
1042             gmx_mm_load_1pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,&c6_00,&c12_00);
1043
1044             /* LENNARD-JONES DISPERSION/REPULSION */
1045
1046             rinvsix          = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
1047             fvdw             = _mm_mul_pd(_mm_sub_pd(_mm_mul_pd(c12_00,rinvsix),c6_00),_mm_mul_pd(rinvsix,rinvsq00));
1048
1049             fscal            = fvdw;
1050
1051             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1052
1053             /* Calculate temporary vectorial force */
1054             tx               = _mm_mul_pd(fscal,dx00);
1055             ty               = _mm_mul_pd(fscal,dy00);
1056             tz               = _mm_mul_pd(fscal,dz00);
1057
1058             /* Update vectorial force */
1059             fix0             = _mm_add_pd(fix0,tx);
1060             fiy0             = _mm_add_pd(fiy0,ty);
1061             fiz0             = _mm_add_pd(fiz0,tz);
1062
1063             fjx0             = _mm_add_pd(fjx0,tx);
1064             fjy0             = _mm_add_pd(fjy0,ty);
1065             fjz0             = _mm_add_pd(fjz0,tz);
1066
1067             /**************************
1068              * CALCULATE INTERACTIONS *
1069              **************************/
1070
1071             r10              = _mm_mul_pd(rsq10,rinv10);
1072
1073             /* Compute parameters for interactions between i and j atoms */
1074             qq10             = _mm_mul_pd(iq1,jq0);
1075
1076             /* EWALD ELECTROSTATICS */
1077
1078             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1079             ewrt             = _mm_mul_pd(r10,ewtabscale);
1080             ewitab           = _mm_cvttpd_epi32(ewrt);
1081             eweps            = _mm_sub_pd(ewrt,_mm_cvtepi32_pd(ewitab));
1082             gmx_mm_load_1pair_swizzle_pd(ewtab+gmx_mm_extract_epi32(ewitab,0),&ewtabF,&ewtabFn);
1083             felec            = _mm_add_pd(_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF),_mm_mul_pd(eweps,ewtabFn));
1084             felec            = _mm_mul_pd(_mm_mul_pd(qq10,rinv10),_mm_sub_pd(rinvsq10,felec));
1085
1086             fscal            = felec;
1087
1088             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1089
1090             /* Calculate temporary vectorial force */
1091             tx               = _mm_mul_pd(fscal,dx10);
1092             ty               = _mm_mul_pd(fscal,dy10);
1093             tz               = _mm_mul_pd(fscal,dz10);
1094
1095             /* Update vectorial force */
1096             fix1             = _mm_add_pd(fix1,tx);
1097             fiy1             = _mm_add_pd(fiy1,ty);
1098             fiz1             = _mm_add_pd(fiz1,tz);
1099
1100             fjx0             = _mm_add_pd(fjx0,tx);
1101             fjy0             = _mm_add_pd(fjy0,ty);
1102             fjz0             = _mm_add_pd(fjz0,tz);
1103
1104             /**************************
1105              * CALCULATE INTERACTIONS *
1106              **************************/
1107
1108             r20              = _mm_mul_pd(rsq20,rinv20);
1109
1110             /* Compute parameters for interactions between i and j atoms */
1111             qq20             = _mm_mul_pd(iq2,jq0);
1112
1113             /* EWALD ELECTROSTATICS */
1114
1115             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1116             ewrt             = _mm_mul_pd(r20,ewtabscale);
1117             ewitab           = _mm_cvttpd_epi32(ewrt);
1118             eweps            = _mm_sub_pd(ewrt,_mm_cvtepi32_pd(ewitab));
1119             gmx_mm_load_1pair_swizzle_pd(ewtab+gmx_mm_extract_epi32(ewitab,0),&ewtabF,&ewtabFn);
1120             felec            = _mm_add_pd(_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF),_mm_mul_pd(eweps,ewtabFn));
1121             felec            = _mm_mul_pd(_mm_mul_pd(qq20,rinv20),_mm_sub_pd(rinvsq20,felec));
1122
1123             fscal            = felec;
1124
1125             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1126
1127             /* Calculate temporary vectorial force */
1128             tx               = _mm_mul_pd(fscal,dx20);
1129             ty               = _mm_mul_pd(fscal,dy20);
1130             tz               = _mm_mul_pd(fscal,dz20);
1131
1132             /* Update vectorial force */
1133             fix2             = _mm_add_pd(fix2,tx);
1134             fiy2             = _mm_add_pd(fiy2,ty);
1135             fiz2             = _mm_add_pd(fiz2,tz);
1136
1137             fjx0             = _mm_add_pd(fjx0,tx);
1138             fjy0             = _mm_add_pd(fjy0,ty);
1139             fjz0             = _mm_add_pd(fjz0,tz);
1140
1141             /**************************
1142              * CALCULATE INTERACTIONS *
1143              **************************/
1144
1145             r30              = _mm_mul_pd(rsq30,rinv30);
1146
1147             /* Compute parameters for interactions between i and j atoms */
1148             qq30             = _mm_mul_pd(iq3,jq0);
1149
1150             /* EWALD ELECTROSTATICS */
1151
1152             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1153             ewrt             = _mm_mul_pd(r30,ewtabscale);
1154             ewitab           = _mm_cvttpd_epi32(ewrt);
1155             eweps            = _mm_sub_pd(ewrt,_mm_cvtepi32_pd(ewitab));
1156             gmx_mm_load_1pair_swizzle_pd(ewtab+gmx_mm_extract_epi32(ewitab,0),&ewtabF,&ewtabFn);
1157             felec            = _mm_add_pd(_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF),_mm_mul_pd(eweps,ewtabFn));
1158             felec            = _mm_mul_pd(_mm_mul_pd(qq30,rinv30),_mm_sub_pd(rinvsq30,felec));
1159
1160             fscal            = felec;
1161
1162             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1163
1164             /* Calculate temporary vectorial force */
1165             tx               = _mm_mul_pd(fscal,dx30);
1166             ty               = _mm_mul_pd(fscal,dy30);
1167             tz               = _mm_mul_pd(fscal,dz30);
1168
1169             /* Update vectorial force */
1170             fix3             = _mm_add_pd(fix3,tx);
1171             fiy3             = _mm_add_pd(fiy3,ty);
1172             fiz3             = _mm_add_pd(fiz3,tz);
1173
1174             fjx0             = _mm_add_pd(fjx0,tx);
1175             fjy0             = _mm_add_pd(fjy0,ty);
1176             fjz0             = _mm_add_pd(fjz0,tz);
1177
1178             gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,fjx0,fjy0,fjz0);
1179
1180             /* Inner loop uses 138 flops */
1181         }
1182
1183         /* End of innermost loop */
1184
1185         gmx_mm_update_iforce_4atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
1186                                               f+i_coord_offset,fshift+i_shift_offset);
1187
1188         /* Increment number of inner iterations */
1189         inneriter                  += j_index_end - j_index_start;
1190
1191         /* Outer loop uses 24 flops */
1192     }
1193
1194     /* Increment number of outer iterations */
1195     outeriter        += nri;
1196
1197     /* Update outer/inner flops */
1198
1199     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W4_F,outeriter*24 + inneriter*138);
1200 }