src/gromacs/gmxlib/nonbonded/nb_kernel_sse4_1_double/nb_kernel_ElecEw_VdwLJ_GeomW3P1_sse4_1_double.c

   1 /*
   2  * This file is part of the GROMACS molecular simulation package.
   3  *
   4  * Copyright (c) 2012,2013,2014, by the GROMACS development team, led by
   5  * Mark Abraham, David van der Spoel, Berk Hess, and Erik Lindahl,
   6  * and including many others, as listed in the AUTHORS file in the
   7  * top-level source directory and at http://www.gromacs.org.
   8  *
   9  * GROMACS is free software; you can redistribute it and/or
  10  * modify it under the terms of the GNU Lesser General Public License
  11  * as published by the Free Software Foundation; either version 2.1
  12  * of the License, or (at your option) any later version.
  13  *
  14  * GROMACS is distributed in the hope that it will be useful,
  15  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  16  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  17  * Lesser General Public License for more details.
  18  *
  19  * You should have received a copy of the GNU Lesser General Public
  20  * License along with GROMACS; if not, see
  21  * http://www.gnu.org/licenses, or write to the Free Software Foundation,
  22  * Inc., 51 Franklin Street, Fifth Floor, Boston, MA  02110-1301  USA.
  23  *
  24  * If you want to redistribute modifications to GROMACS, please
  25  * consider that scientific software is very special. Version
  26  * control is crucial - bugs must be traceable. We will be happy to
  27  * consider code for inclusion in the official distribution, but
  28  * derived work must not be called official GROMACS. Details are found
  29  * in the README & COPYING files - if they are missing, get the
  30  * official version at http://www.gromacs.org.
  31  *
  32  * To help us fund GROMACS development, we humbly ask that you cite
  33  * the research papers on the package. Check out http://www.gromacs.org.
  34  */
  35 /*
  36  * Note: this file was generated by the GROMACS sse4_1_double kernel generator.
  37  */
  38 #include "config.h"
  39
  40 #include <math.h>
  41
  42 #include "../nb_kernel.h"
  43 #include "gromacs/legacyheaders/types/simple.h"
  44 #include "gromacs/math/vec.h"
  45 #include "gromacs/legacyheaders/nrnb.h"
  46
  47 #include "gromacs/simd/math_x86_sse4_1_double.h"
  48 #include "kernelutil_x86_sse4_1_double.h"
  49
  50 /*
  51  * Gromacs nonbonded kernel:   nb_kernel_ElecEw_VdwLJ_GeomW3P1_VF_sse4_1_double
  52  * Electrostatics interaction: Ewald
  53  * VdW interaction:            LennardJones
  54  * Geometry:                   Water3-Particle
  55  * Calculate force/pot:        PotentialAndForce
  56  */
  57 void
  58 nb_kernel_ElecEw_VdwLJ_GeomW3P1_VF_sse4_1_double
  59                     (t_nblist                    * gmx_restrict       nlist,
  60                      rvec                        * gmx_restrict          xx,
  61                      rvec                        * gmx_restrict          ff,
  62                      t_forcerec                  * gmx_restrict          fr,
  63                      t_mdatoms                   * gmx_restrict     mdatoms,
  64                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
  65                      t_nrnb                      * gmx_restrict        nrnb)
  66 {
  67     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
  68      * just 0 for non-waters.
  69      * Suffixes A,B refer to j loop unrolling done with SSE double precision, e.g. for the two different
  70      * jnr indices corresponding to data put in the four positions in the SIMD register.
  71      */
  72     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
  73     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
  74     int              jnrA,jnrB;
  75     int              j_coord_offsetA,j_coord_offsetB;
  76     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
  77     real             rcutoff_scalar;
  78     real             *shiftvec,*fshift,*x,*f;
  79     __m128d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
  80     int              vdwioffset0;
  81     __m128d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
  82     int              vdwioffset1;
  83     __m128d          ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
  84     int              vdwioffset2;
  85     __m128d          ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
  86     int              vdwjidx0A,vdwjidx0B;
  87     __m128d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  88     __m128d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  89     __m128d          dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
  90     __m128d          dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
  91     __m128d          velec,felec,velecsum,facel,crf,krf,krf2;
  92     real             *charge;
  93     int              nvdwtype;
  94     __m128d          rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
  95     int              *vdwtype;
  96     real             *vdwparam;
  97     __m128d          one_sixth   = _mm_set1_pd(1.0/6.0);
  98     __m128d          one_twelfth = _mm_set1_pd(1.0/12.0);
  99     __m128i          ewitab;
 100     __m128d          ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
 101     real             *ewtab;
 102     __m128d          dummy_mask,cutoff_mask;
 103     __m128d          signbit   = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
 104     __m128d          one     = _mm_set1_pd(1.0);
 105     __m128d          two     = _mm_set1_pd(2.0);
 106     x                = xx[0];
 107     f                = ff[0];
 108
 109     nri              = nlist->nri;
 110     iinr             = nlist->iinr;
 111     jindex           = nlist->jindex;
 112     jjnr             = nlist->jjnr;
 113     shiftidx         = nlist->shift;
 114     gid              = nlist->gid;
 115     shiftvec         = fr->shift_vec[0];
 116     fshift           = fr->fshift[0];
 117     facel            = _mm_set1_pd(fr->epsfac);
 118     charge           = mdatoms->chargeA;
 119     nvdwtype         = fr->ntype;
 120     vdwparam         = fr->nbfp;
 121     vdwtype          = mdatoms->typeA;
 122
 123     sh_ewald         = _mm_set1_pd(fr->ic->sh_ewald);
 124     ewtab            = fr->ic->tabq_coul_FDV0;
 125     ewtabscale       = _mm_set1_pd(fr->ic->tabq_scale);
 126     ewtabhalfspace   = _mm_set1_pd(0.5/fr->ic->tabq_scale);
 127
 128     /* Setup water-specific parameters */
 129     inr              = nlist->iinr[0];
 130     iq0              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+0]));
 131     iq1              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+1]));
 132     iq2              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+2]));
 133     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 134
 135     /* Avoid stupid compiler warnings */
 136     jnrA = jnrB = 0;
 137     j_coord_offsetA = 0;
 138     j_coord_offsetB = 0;
 139
 140     outeriter        = 0;
 141     inneriter        = 0;
 142
 143     /* Start outer loop over neighborlists */
 144     for(iidx=0; iidx<nri; iidx++)
 145     {
 146         /* Load shift vector for this list */
 147         i_shift_offset   = DIM*shiftidx[iidx];
 148
 149         /* Load limits for loop over neighbors */
 150         j_index_start    = jindex[iidx];
 151         j_index_end      = jindex[iidx+1];
 152
 153         /* Get outer coordinate index */
 154         inr              = iinr[iidx];
 155         i_coord_offset   = DIM*inr;
 156
 157         /* Load i particle coords and add shift vector */
 158         gmx_mm_load_shift_and_3rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
 159                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 160
 161         fix0             = _mm_setzero_pd();
 162         fiy0             = _mm_setzero_pd();
 163         fiz0             = _mm_setzero_pd();
 164         fix1             = _mm_setzero_pd();
 165         fiy1             = _mm_setzero_pd();
 166         fiz1             = _mm_setzero_pd();
 167         fix2             = _mm_setzero_pd();
 168         fiy2             = _mm_setzero_pd();
 169         fiz2             = _mm_setzero_pd();
 170
 171         /* Reset potential sums */
 172         velecsum         = _mm_setzero_pd();
 173         vvdwsum          = _mm_setzero_pd();
 174
 175         /* Start inner kernel loop */
 176         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 177         {
 178
 179             /* Get j neighbor index, and coordinate index */
 180             jnrA             = jjnr[jidx];
 181             jnrB             = jjnr[jidx+1];
 182             j_coord_offsetA  = DIM*jnrA;
 183             j_coord_offsetB  = DIM*jnrB;
 184
 185             /* load j atom coordinates */
 186             gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 187                                               &jx0,&jy0,&jz0);
 188
 189             /* Calculate displacement vector */
 190             dx00             = _mm_sub_pd(ix0,jx0);
 191             dy00             = _mm_sub_pd(iy0,jy0);
 192             dz00             = _mm_sub_pd(iz0,jz0);
 193             dx10             = _mm_sub_pd(ix1,jx0);
 194             dy10             = _mm_sub_pd(iy1,jy0);
 195             dz10             = _mm_sub_pd(iz1,jz0);
 196             dx20             = _mm_sub_pd(ix2,jx0);
 197             dy20             = _mm_sub_pd(iy2,jy0);
 198             dz20             = _mm_sub_pd(iz2,jz0);
 199
 200             /* Calculate squared distance and things based on it */
 201             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 202             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
 203             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
 204
 205             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 206             rinv10           = gmx_mm_invsqrt_pd(rsq10);
 207             rinv20           = gmx_mm_invsqrt_pd(rsq20);
 208
 209             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 210             rinvsq10         = _mm_mul_pd(rinv10,rinv10);
 211             rinvsq20         = _mm_mul_pd(rinv20,rinv20);
 212
 213             /* Load parameters for j particles */
 214             jq0              = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
 215             vdwjidx0A        = 2*vdwtype[jnrA+0];
 216             vdwjidx0B        = 2*vdwtype[jnrB+0];
 217
 218             fjx0             = _mm_setzero_pd();
 219             fjy0             = _mm_setzero_pd();
 220             fjz0             = _mm_setzero_pd();
 221
 222             /**************************
 223              * CALCULATE INTERACTIONS *
 224              **************************/
 225
 226             r00              = _mm_mul_pd(rsq00,rinv00);
 227
 228             /* Compute parameters for interactions between i and j atoms */
 229             qq00             = _mm_mul_pd(iq0,jq0);
 230             gmx_mm_load_2pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,
 231                                          vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
 232
 233             /* EWALD ELECTROSTATICS */
 234
 235             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 236             ewrt             = _mm_mul_pd(r00,ewtabscale);
 237             ewitab           = _mm_cvttpd_epi32(ewrt);
 238             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 239             ewitab           = _mm_slli_epi32(ewitab,2);
 240             ewtabF           = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 241             ewtabD           = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,1) );
 242             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 243             ewtabV           = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,0) +2);
 244             ewtabFn          = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,1) +2);
 245             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 246             felec            = _mm_add_pd(ewtabF,_mm_mul_pd(eweps,ewtabD));
 247             velec            = _mm_sub_pd(ewtabV,_mm_mul_pd(_mm_mul_pd(ewtabhalfspace,eweps),_mm_add_pd(ewtabF,felec)));
 248             velec            = _mm_mul_pd(qq00,_mm_sub_pd(rinv00,velec));
 249             felec            = _mm_mul_pd(_mm_mul_pd(qq00,rinv00),_mm_sub_pd(rinvsq00,felec));
 250
 251             /* LENNARD-JONES DISPERSION/REPULSION */
 252
 253             rinvsix          = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 254             vvdw6            = _mm_mul_pd(c6_00,rinvsix);
 255             vvdw12           = _mm_mul_pd(c12_00,_mm_mul_pd(rinvsix,rinvsix));
 256             vvdw             = _mm_sub_pd( _mm_mul_pd(vvdw12,one_twelfth) , _mm_mul_pd(vvdw6,one_sixth) );
 257             fvdw             = _mm_mul_pd(_mm_sub_pd(vvdw12,vvdw6),rinvsq00);
 258
 259             /* Update potential sum for this i atom from the interaction with this j atom. */
 260             velecsum         = _mm_add_pd(velecsum,velec);
 261             vvdwsum          = _mm_add_pd(vvdwsum,vvdw);
 262
 263             fscal            = _mm_add_pd(felec,fvdw);
 264
 265             /* Calculate temporary vectorial force */
 266             tx               = _mm_mul_pd(fscal,dx00);
 267             ty               = _mm_mul_pd(fscal,dy00);
 268             tz               = _mm_mul_pd(fscal,dz00);
 269
 270             /* Update vectorial force */
 271             fix0             = _mm_add_pd(fix0,tx);
 272             fiy0             = _mm_add_pd(fiy0,ty);
 273             fiz0             = _mm_add_pd(fiz0,tz);
 274
 275             fjx0             = _mm_add_pd(fjx0,tx);
 276             fjy0             = _mm_add_pd(fjy0,ty);
 277             fjz0             = _mm_add_pd(fjz0,tz);
 278
 279             /**************************
 280              * CALCULATE INTERACTIONS *
 281              **************************/
 282
 283             r10              = _mm_mul_pd(rsq10,rinv10);
 284
 285             /* Compute parameters for interactions between i and j atoms */
 286             qq10             = _mm_mul_pd(iq1,jq0);
 287
 288             /* EWALD ELECTROSTATICS */
 289
 290             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 291             ewrt             = _mm_mul_pd(r10,ewtabscale);
 292             ewitab           = _mm_cvttpd_epi32(ewrt);
 293             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 294             ewitab           = _mm_slli_epi32(ewitab,2);
 295             ewtabF           = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 296             ewtabD           = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,1) );
 297             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 298             ewtabV           = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,0) +2);
 299             ewtabFn          = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,1) +2);
 300             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 301             felec            = _mm_add_pd(ewtabF,_mm_mul_pd(eweps,ewtabD));
 302             velec            = _mm_sub_pd(ewtabV,_mm_mul_pd(_mm_mul_pd(ewtabhalfspace,eweps),_mm_add_pd(ewtabF,felec)));
 303             velec            = _mm_mul_pd(qq10,_mm_sub_pd(rinv10,velec));
 304             felec            = _mm_mul_pd(_mm_mul_pd(qq10,rinv10),_mm_sub_pd(rinvsq10,felec));
 305
 306             /* Update potential sum for this i atom from the interaction with this j atom. */
 307             velecsum         = _mm_add_pd(velecsum,velec);
 308
 309             fscal            = felec;
 310
 311             /* Calculate temporary vectorial force */
 312             tx               = _mm_mul_pd(fscal,dx10);
 313             ty               = _mm_mul_pd(fscal,dy10);
 314             tz               = _mm_mul_pd(fscal,dz10);
 315
 316             /* Update vectorial force */
 317             fix1             = _mm_add_pd(fix1,tx);
 318             fiy1             = _mm_add_pd(fiy1,ty);
 319             fiz1             = _mm_add_pd(fiz1,tz);
 320
 321             fjx0             = _mm_add_pd(fjx0,tx);
 322             fjy0             = _mm_add_pd(fjy0,ty);
 323             fjz0             = _mm_add_pd(fjz0,tz);
 324
 325             /**************************
 326              * CALCULATE INTERACTIONS *
 327              **************************/
 328
 329             r20              = _mm_mul_pd(rsq20,rinv20);
 330
 331             /* Compute parameters for interactions between i and j atoms */
 332             qq20             = _mm_mul_pd(iq2,jq0);
 333
 334             /* EWALD ELECTROSTATICS */
 335
 336             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 337             ewrt             = _mm_mul_pd(r20,ewtabscale);
 338             ewitab           = _mm_cvttpd_epi32(ewrt);
 339             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 340             ewitab           = _mm_slli_epi32(ewitab,2);
 341             ewtabF           = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 342             ewtabD           = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,1) );
 343             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 344             ewtabV           = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,0) +2);
 345             ewtabFn          = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,1) +2);
 346             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 347             felec            = _mm_add_pd(ewtabF,_mm_mul_pd(eweps,ewtabD));
 348             velec            = _mm_sub_pd(ewtabV,_mm_mul_pd(_mm_mul_pd(ewtabhalfspace,eweps),_mm_add_pd(ewtabF,felec)));
 349             velec            = _mm_mul_pd(qq20,_mm_sub_pd(rinv20,velec));
 350             felec            = _mm_mul_pd(_mm_mul_pd(qq20,rinv20),_mm_sub_pd(rinvsq20,felec));
 351
 352             /* Update potential sum for this i atom from the interaction with this j atom. */
 353             velecsum         = _mm_add_pd(velecsum,velec);
 354
 355             fscal            = felec;
 356
 357             /* Calculate temporary vectorial force */
 358             tx               = _mm_mul_pd(fscal,dx20);
 359             ty               = _mm_mul_pd(fscal,dy20);
 360             tz               = _mm_mul_pd(fscal,dz20);
 361
 362             /* Update vectorial force */
 363             fix2             = _mm_add_pd(fix2,tx);
 364             fiy2             = _mm_add_pd(fiy2,ty);
 365             fiz2             = _mm_add_pd(fiz2,tz);
 366
 367             fjx0             = _mm_add_pd(fjx0,tx);
 368             fjy0             = _mm_add_pd(fjy0,ty);
 369             fjz0             = _mm_add_pd(fjz0,tz);
 370
 371             gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,fjx0,fjy0,fjz0);
 372
 373             /* Inner loop uses 138 flops */
 374         }
 375
 376         if(jidx<j_index_end)
 377         {
 378
 379             jnrA             = jjnr[jidx];
 380             j_coord_offsetA  = DIM*jnrA;
 381
 382             /* load j atom coordinates */
 383             gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
 384                                               &jx0,&jy0,&jz0);
 385
 386             /* Calculate displacement vector */
 387             dx00             = _mm_sub_pd(ix0,jx0);
 388             dy00             = _mm_sub_pd(iy0,jy0);
 389             dz00             = _mm_sub_pd(iz0,jz0);
 390             dx10             = _mm_sub_pd(ix1,jx0);
 391             dy10             = _mm_sub_pd(iy1,jy0);
 392             dz10             = _mm_sub_pd(iz1,jz0);
 393             dx20             = _mm_sub_pd(ix2,jx0);
 394             dy20             = _mm_sub_pd(iy2,jy0);
 395             dz20             = _mm_sub_pd(iz2,jz0);
 396
 397             /* Calculate squared distance and things based on it */
 398             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 399             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
 400             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
 401
 402             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 403             rinv10           = gmx_mm_invsqrt_pd(rsq10);
 404             rinv20           = gmx_mm_invsqrt_pd(rsq20);
 405
 406             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 407             rinvsq10         = _mm_mul_pd(rinv10,rinv10);
 408             rinvsq20         = _mm_mul_pd(rinv20,rinv20);
 409
 410             /* Load parameters for j particles */
 411             jq0              = _mm_load_sd(charge+jnrA+0);
 412             vdwjidx0A        = 2*vdwtype[jnrA+0];
 413
 414             fjx0             = _mm_setzero_pd();
 415             fjy0             = _mm_setzero_pd();
 416             fjz0             = _mm_setzero_pd();
 417
 418             /**************************
 419              * CALCULATE INTERACTIONS *
 420              **************************/
 421
 422             r00              = _mm_mul_pd(rsq00,rinv00);
 423
 424             /* Compute parameters for interactions between i and j atoms */
 425             qq00             = _mm_mul_pd(iq0,jq0);
 426             gmx_mm_load_1pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,&c6_00,&c12_00);
 427
 428             /* EWALD ELECTROSTATICS */
 429
 430             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 431             ewrt             = _mm_mul_pd(r00,ewtabscale);
 432             ewitab           = _mm_cvttpd_epi32(ewrt);
 433             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 434             ewitab           = _mm_slli_epi32(ewitab,2);
 435             ewtabF           = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 436             ewtabD           = _mm_setzero_pd();
 437             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 438             ewtabV           = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,0) +2);
 439             ewtabFn          = _mm_setzero_pd();
 440             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 441             felec            = _mm_add_pd(ewtabF,_mm_mul_pd(eweps,ewtabD));
 442             velec            = _mm_sub_pd(ewtabV,_mm_mul_pd(_mm_mul_pd(ewtabhalfspace,eweps),_mm_add_pd(ewtabF,felec)));
 443             velec            = _mm_mul_pd(qq00,_mm_sub_pd(rinv00,velec));
 444             felec            = _mm_mul_pd(_mm_mul_pd(qq00,rinv00),_mm_sub_pd(rinvsq00,felec));
 445
 446             /* LENNARD-JONES DISPERSION/REPULSION */
 447
 448             rinvsix          = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 449             vvdw6            = _mm_mul_pd(c6_00,rinvsix);
 450             vvdw12           = _mm_mul_pd(c12_00,_mm_mul_pd(rinvsix,rinvsix));
 451             vvdw             = _mm_sub_pd( _mm_mul_pd(vvdw12,one_twelfth) , _mm_mul_pd(vvdw6,one_sixth) );
 452             fvdw             = _mm_mul_pd(_mm_sub_pd(vvdw12,vvdw6),rinvsq00);
 453
 454             /* Update potential sum for this i atom from the interaction with this j atom. */
 455             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 456             velecsum         = _mm_add_pd(velecsum,velec);
 457             vvdw             = _mm_unpacklo_pd(vvdw,_mm_setzero_pd());
 458             vvdwsum          = _mm_add_pd(vvdwsum,vvdw);
 459
 460             fscal            = _mm_add_pd(felec,fvdw);
 461
 462             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 463
 464             /* Calculate temporary vectorial force */
 465             tx               = _mm_mul_pd(fscal,dx00);
 466             ty               = _mm_mul_pd(fscal,dy00);
 467             tz               = _mm_mul_pd(fscal,dz00);
 468
 469             /* Update vectorial force */
 470             fix0             = _mm_add_pd(fix0,tx);
 471             fiy0             = _mm_add_pd(fiy0,ty);
 472             fiz0             = _mm_add_pd(fiz0,tz);
 473
 474             fjx0             = _mm_add_pd(fjx0,tx);
 475             fjy0             = _mm_add_pd(fjy0,ty);
 476             fjz0             = _mm_add_pd(fjz0,tz);
 477
 478             /**************************
 479              * CALCULATE INTERACTIONS *
 480              **************************/
 481
 482             r10              = _mm_mul_pd(rsq10,rinv10);
 483
 484             /* Compute parameters for interactions between i and j atoms */
 485             qq10             = _mm_mul_pd(iq1,jq0);
 486
 487             /* EWALD ELECTROSTATICS */
 488
 489             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 490             ewrt             = _mm_mul_pd(r10,ewtabscale);
 491             ewitab           = _mm_cvttpd_epi32(ewrt);
 492             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 493             ewitab           = _mm_slli_epi32(ewitab,2);
 494             ewtabF           = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 495             ewtabD           = _mm_setzero_pd();
 496             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 497             ewtabV           = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,0) +2);
 498             ewtabFn          = _mm_setzero_pd();
 499             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 500             felec            = _mm_add_pd(ewtabF,_mm_mul_pd(eweps,ewtabD));
 501             velec            = _mm_sub_pd(ewtabV,_mm_mul_pd(_mm_mul_pd(ewtabhalfspace,eweps),_mm_add_pd(ewtabF,felec)));
 502             velec            = _mm_mul_pd(qq10,_mm_sub_pd(rinv10,velec));
 503             felec            = _mm_mul_pd(_mm_mul_pd(qq10,rinv10),_mm_sub_pd(rinvsq10,felec));
 504
 505             /* Update potential sum for this i atom from the interaction with this j atom. */
 506             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 507             velecsum         = _mm_add_pd(velecsum,velec);
 508
 509             fscal            = felec;
 510
 511             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 512
 513             /* Calculate temporary vectorial force */
 514             tx               = _mm_mul_pd(fscal,dx10);
 515             ty               = _mm_mul_pd(fscal,dy10);
 516             tz               = _mm_mul_pd(fscal,dz10);
 517
 518             /* Update vectorial force */
 519             fix1             = _mm_add_pd(fix1,tx);
 520             fiy1             = _mm_add_pd(fiy1,ty);
 521             fiz1             = _mm_add_pd(fiz1,tz);
 522
 523             fjx0             = _mm_add_pd(fjx0,tx);
 524             fjy0             = _mm_add_pd(fjy0,ty);
 525             fjz0             = _mm_add_pd(fjz0,tz);
 526
 527             /**************************
 528              * CALCULATE INTERACTIONS *
 529              **************************/
 530
 531             r20              = _mm_mul_pd(rsq20,rinv20);
 532
 533             /* Compute parameters for interactions between i and j atoms */
 534             qq20             = _mm_mul_pd(iq2,jq0);
 535
 536             /* EWALD ELECTROSTATICS */
 537
 538             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 539             ewrt             = _mm_mul_pd(r20,ewtabscale);
 540             ewitab           = _mm_cvttpd_epi32(ewrt);
 541             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 542             ewitab           = _mm_slli_epi32(ewitab,2);
 543             ewtabF           = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 544             ewtabD           = _mm_setzero_pd();
 545             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 546             ewtabV           = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,0) +2);
 547             ewtabFn          = _mm_setzero_pd();
 548             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 549             felec            = _mm_add_pd(ewtabF,_mm_mul_pd(eweps,ewtabD));
 550             velec            = _mm_sub_pd(ewtabV,_mm_mul_pd(_mm_mul_pd(ewtabhalfspace,eweps),_mm_add_pd(ewtabF,felec)));
 551             velec            = _mm_mul_pd(qq20,_mm_sub_pd(rinv20,velec));
 552             felec            = _mm_mul_pd(_mm_mul_pd(qq20,rinv20),_mm_sub_pd(rinvsq20,felec));
 553
 554             /* Update potential sum for this i atom from the interaction with this j atom. */
 555             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 556             velecsum         = _mm_add_pd(velecsum,velec);
 557
 558             fscal            = felec;
 559
 560             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 561
 562             /* Calculate temporary vectorial force */
 563             tx               = _mm_mul_pd(fscal,dx20);
 564             ty               = _mm_mul_pd(fscal,dy20);
 565             tz               = _mm_mul_pd(fscal,dz20);
 566
 567             /* Update vectorial force */
 568             fix2             = _mm_add_pd(fix2,tx);
 569             fiy2             = _mm_add_pd(fiy2,ty);
 570             fiz2             = _mm_add_pd(fiz2,tz);
 571
 572             fjx0             = _mm_add_pd(fjx0,tx);
 573             fjy0             = _mm_add_pd(fjy0,ty);
 574             fjz0             = _mm_add_pd(fjz0,tz);
 575
 576             gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,fjx0,fjy0,fjz0);
 577
 578             /* Inner loop uses 138 flops */
 579         }
 580
 581         /* End of innermost loop */
 582
 583         gmx_mm_update_iforce_3atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
 584                                               f+i_coord_offset,fshift+i_shift_offset);
 585
 586         ggid                        = gid[iidx];
 587         /* Update potential energies */
 588         gmx_mm_update_1pot_pd(velecsum,kernel_data->energygrp_elec+ggid);
 589         gmx_mm_update_1pot_pd(vvdwsum,kernel_data->energygrp_vdw+ggid);
 590
 591         /* Increment number of inner iterations */
 592         inneriter                  += j_index_end - j_index_start;
 593
 594         /* Outer loop uses 20 flops */
 595     }
 596
 597     /* Increment number of outer iterations */
 598     outeriter        += nri;
 599
 600     /* Update outer/inner flops */
 601
 602     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3_VF,outeriter*20 + inneriter*138);
 603 }
 604 /*
 605  * Gromacs nonbonded kernel:   nb_kernel_ElecEw_VdwLJ_GeomW3P1_F_sse4_1_double
 606  * Electrostatics interaction: Ewald
 607  * VdW interaction:            LennardJones
 608  * Geometry:                   Water3-Particle
 609  * Calculate force/pot:        Force
 610  */
 611 void
 612 nb_kernel_ElecEw_VdwLJ_GeomW3P1_F_sse4_1_double
 613                     (t_nblist                    * gmx_restrict       nlist,
 614                      rvec                        * gmx_restrict          xx,
 615                      rvec                        * gmx_restrict          ff,
 616                      t_forcerec                  * gmx_restrict          fr,
 617                      t_mdatoms                   * gmx_restrict     mdatoms,
 618                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 619                      t_nrnb                      * gmx_restrict        nrnb)
 620 {
 621     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 622      * just 0 for non-waters.
 623      * Suffixes A,B refer to j loop unrolling done with SSE double precision, e.g. for the two different
 624      * jnr indices corresponding to data put in the four positions in the SIMD register.
 625      */
 626     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 627     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 628     int              jnrA,jnrB;
 629     int              j_coord_offsetA,j_coord_offsetB;
 630     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 631     real             rcutoff_scalar;
 632     real             *shiftvec,*fshift,*x,*f;
 633     __m128d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 634     int              vdwioffset0;
 635     __m128d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 636     int              vdwioffset1;
 637     __m128d          ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
 638     int              vdwioffset2;
 639     __m128d          ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
 640     int              vdwjidx0A,vdwjidx0B;
 641     __m128d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 642     __m128d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 643     __m128d          dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
 644     __m128d          dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
 645     __m128d          velec,felec,velecsum,facel,crf,krf,krf2;
 646     real             *charge;
 647     int              nvdwtype;
 648     __m128d          rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 649     int              *vdwtype;
 650     real             *vdwparam;
 651     __m128d          one_sixth   = _mm_set1_pd(1.0/6.0);
 652     __m128d          one_twelfth = _mm_set1_pd(1.0/12.0);
 653     __m128i          ewitab;
 654     __m128d          ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
 655     real             *ewtab;
 656     __m128d          dummy_mask,cutoff_mask;
 657     __m128d          signbit   = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
 658     __m128d          one     = _mm_set1_pd(1.0);
 659     __m128d          two     = _mm_set1_pd(2.0);
 660     x                = xx[0];
 661     f                = ff[0];
 662
 663     nri              = nlist->nri;
 664     iinr             = nlist->iinr;
 665     jindex           = nlist->jindex;
 666     jjnr             = nlist->jjnr;
 667     shiftidx         = nlist->shift;
 668     gid              = nlist->gid;
 669     shiftvec         = fr->shift_vec[0];
 670     fshift           = fr->fshift[0];
 671     facel            = _mm_set1_pd(fr->epsfac);
 672     charge           = mdatoms->chargeA;
 673     nvdwtype         = fr->ntype;
 674     vdwparam         = fr->nbfp;
 675     vdwtype          = mdatoms->typeA;
 676
 677     sh_ewald         = _mm_set1_pd(fr->ic->sh_ewald);
 678     ewtab            = fr->ic->tabq_coul_F;
 679     ewtabscale       = _mm_set1_pd(fr->ic->tabq_scale);
 680     ewtabhalfspace   = _mm_set1_pd(0.5/fr->ic->tabq_scale);
 681
 682     /* Setup water-specific parameters */
 683     inr              = nlist->iinr[0];
 684     iq0              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+0]));
 685     iq1              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+1]));
 686     iq2              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+2]));
 687     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 688
 689     /* Avoid stupid compiler warnings */
 690     jnrA = jnrB = 0;
 691     j_coord_offsetA = 0;
 692     j_coord_offsetB = 0;
 693
 694     outeriter        = 0;
 695     inneriter        = 0;
 696
 697     /* Start outer loop over neighborlists */
 698     for(iidx=0; iidx<nri; iidx++)
 699     {
 700         /* Load shift vector for this list */
 701         i_shift_offset   = DIM*shiftidx[iidx];
 702
 703         /* Load limits for loop over neighbors */
 704         j_index_start    = jindex[iidx];
 705         j_index_end      = jindex[iidx+1];
 706
 707         /* Get outer coordinate index */
 708         inr              = iinr[iidx];
 709         i_coord_offset   = DIM*inr;
 710
 711         /* Load i particle coords and add shift vector */
 712         gmx_mm_load_shift_and_3rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
 713                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 714
 715         fix0             = _mm_setzero_pd();
 716         fiy0             = _mm_setzero_pd();
 717         fiz0             = _mm_setzero_pd();
 718         fix1             = _mm_setzero_pd();
 719         fiy1             = _mm_setzero_pd();
 720         fiz1             = _mm_setzero_pd();
 721         fix2             = _mm_setzero_pd();
 722         fiy2             = _mm_setzero_pd();
 723         fiz2             = _mm_setzero_pd();
 724
 725         /* Start inner kernel loop */
 726         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 727         {
 728
 729             /* Get j neighbor index, and coordinate index */
 730             jnrA             = jjnr[jidx];
 731             jnrB             = jjnr[jidx+1];
 732             j_coord_offsetA  = DIM*jnrA;
 733             j_coord_offsetB  = DIM*jnrB;
 734
 735             /* load j atom coordinates */
 736             gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 737                                               &jx0,&jy0,&jz0);
 738
 739             /* Calculate displacement vector */
 740             dx00             = _mm_sub_pd(ix0,jx0);
 741             dy00             = _mm_sub_pd(iy0,jy0);
 742             dz00             = _mm_sub_pd(iz0,jz0);
 743             dx10             = _mm_sub_pd(ix1,jx0);
 744             dy10             = _mm_sub_pd(iy1,jy0);
 745             dz10             = _mm_sub_pd(iz1,jz0);
 746             dx20             = _mm_sub_pd(ix2,jx0);
 747             dy20             = _mm_sub_pd(iy2,jy0);
 748             dz20             = _mm_sub_pd(iz2,jz0);
 749
 750             /* Calculate squared distance and things based on it */
 751             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 752             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
 753             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
 754
 755             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 756             rinv10           = gmx_mm_invsqrt_pd(rsq10);
 757             rinv20           = gmx_mm_invsqrt_pd(rsq20);
 758
 759             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 760             rinvsq10         = _mm_mul_pd(rinv10,rinv10);
 761             rinvsq20         = _mm_mul_pd(rinv20,rinv20);
 762
 763             /* Load parameters for j particles */
 764             jq0              = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
 765             vdwjidx0A        = 2*vdwtype[jnrA+0];
 766             vdwjidx0B        = 2*vdwtype[jnrB+0];
 767
 768             fjx0             = _mm_setzero_pd();
 769             fjy0             = _mm_setzero_pd();
 770             fjz0             = _mm_setzero_pd();
 771
 772             /**************************
 773              * CALCULATE INTERACTIONS *
 774              **************************/
 775
 776             r00              = _mm_mul_pd(rsq00,rinv00);
 777
 778             /* Compute parameters for interactions between i and j atoms */
 779             qq00             = _mm_mul_pd(iq0,jq0);
 780             gmx_mm_load_2pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,
 781                                          vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
 782
 783             /* EWALD ELECTROSTATICS */
 784
 785             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 786             ewrt             = _mm_mul_pd(r00,ewtabscale);
 787             ewitab           = _mm_cvttpd_epi32(ewrt);
 788             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 789             gmx_mm_load_2pair_swizzle_pd(ewtab+gmx_mm_extract_epi32(ewitab,0),ewtab+gmx_mm_extract_epi32(ewitab,1),
 790                                          &ewtabF,&ewtabFn);
 791             felec            = _mm_add_pd(_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF),_mm_mul_pd(eweps,ewtabFn));
 792             felec            = _mm_mul_pd(_mm_mul_pd(qq00,rinv00),_mm_sub_pd(rinvsq00,felec));
 793
 794             /* LENNARD-JONES DISPERSION/REPULSION */
 795
 796             rinvsix          = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 797             fvdw             = _mm_mul_pd(_mm_sub_pd(_mm_mul_pd(c12_00,rinvsix),c6_00),_mm_mul_pd(rinvsix,rinvsq00));
 798
 799             fscal            = _mm_add_pd(felec,fvdw);
 800
 801             /* Calculate temporary vectorial force */
 802             tx               = _mm_mul_pd(fscal,dx00);
 803             ty               = _mm_mul_pd(fscal,dy00);
 804             tz               = _mm_mul_pd(fscal,dz00);
 805
 806             /* Update vectorial force */
 807             fix0             = _mm_add_pd(fix0,tx);
 808             fiy0             = _mm_add_pd(fiy0,ty);
 809             fiz0             = _mm_add_pd(fiz0,tz);
 810
 811             fjx0             = _mm_add_pd(fjx0,tx);
 812             fjy0             = _mm_add_pd(fjy0,ty);
 813             fjz0             = _mm_add_pd(fjz0,tz);
 814
 815             /**************************
 816              * CALCULATE INTERACTIONS *
 817              **************************/
 818
 819             r10              = _mm_mul_pd(rsq10,rinv10);
 820
 821             /* Compute parameters for interactions between i and j atoms */
 822             qq10             = _mm_mul_pd(iq1,jq0);
 823
 824             /* EWALD ELECTROSTATICS */
 825
 826             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 827             ewrt             = _mm_mul_pd(r10,ewtabscale);
 828             ewitab           = _mm_cvttpd_epi32(ewrt);
 829             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 830             gmx_mm_load_2pair_swizzle_pd(ewtab+gmx_mm_extract_epi32(ewitab,0),ewtab+gmx_mm_extract_epi32(ewitab,1),
 831                                          &ewtabF,&ewtabFn);
 832             felec            = _mm_add_pd(_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF),_mm_mul_pd(eweps,ewtabFn));
 833             felec            = _mm_mul_pd(_mm_mul_pd(qq10,rinv10),_mm_sub_pd(rinvsq10,felec));
 834
 835             fscal            = felec;
 836
 837             /* Calculate temporary vectorial force */
 838             tx               = _mm_mul_pd(fscal,dx10);
 839             ty               = _mm_mul_pd(fscal,dy10);
 840             tz               = _mm_mul_pd(fscal,dz10);
 841
 842             /* Update vectorial force */
 843             fix1             = _mm_add_pd(fix1,tx);
 844             fiy1             = _mm_add_pd(fiy1,ty);
 845             fiz1             = _mm_add_pd(fiz1,tz);
 846
 847             fjx0             = _mm_add_pd(fjx0,tx);
 848             fjy0             = _mm_add_pd(fjy0,ty);
 849             fjz0             = _mm_add_pd(fjz0,tz);
 850
 851             /**************************
 852              * CALCULATE INTERACTIONS *
 853              **************************/
 854
 855             r20              = _mm_mul_pd(rsq20,rinv20);
 856
 857             /* Compute parameters for interactions between i and j atoms */
 858             qq20             = _mm_mul_pd(iq2,jq0);
 859
 860             /* EWALD ELECTROSTATICS */
 861
 862             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 863             ewrt             = _mm_mul_pd(r20,ewtabscale);
 864             ewitab           = _mm_cvttpd_epi32(ewrt);
 865             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 866             gmx_mm_load_2pair_swizzle_pd(ewtab+gmx_mm_extract_epi32(ewitab,0),ewtab+gmx_mm_extract_epi32(ewitab,1),
 867                                          &ewtabF,&ewtabFn);
 868             felec            = _mm_add_pd(_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF),_mm_mul_pd(eweps,ewtabFn));
 869             felec            = _mm_mul_pd(_mm_mul_pd(qq20,rinv20),_mm_sub_pd(rinvsq20,felec));
 870
 871             fscal            = felec;
 872
 873             /* Calculate temporary vectorial force */
 874             tx               = _mm_mul_pd(fscal,dx20);
 875             ty               = _mm_mul_pd(fscal,dy20);
 876             tz               = _mm_mul_pd(fscal,dz20);
 877
 878             /* Update vectorial force */
 879             fix2             = _mm_add_pd(fix2,tx);
 880             fiy2             = _mm_add_pd(fiy2,ty);
 881             fiz2             = _mm_add_pd(fiz2,tz);
 882
 883             fjx0             = _mm_add_pd(fjx0,tx);
 884             fjy0             = _mm_add_pd(fjy0,ty);
 885             fjz0             = _mm_add_pd(fjz0,tz);
 886
 887             gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,fjx0,fjy0,fjz0);
 888
 889             /* Inner loop uses 118 flops */
 890         }
 891
 892         if(jidx<j_index_end)
 893         {
 894
 895             jnrA             = jjnr[jidx];
 896             j_coord_offsetA  = DIM*jnrA;
 897
 898             /* load j atom coordinates */
 899             gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
 900                                               &jx0,&jy0,&jz0);
 901
 902             /* Calculate displacement vector */
 903             dx00             = _mm_sub_pd(ix0,jx0);
 904             dy00             = _mm_sub_pd(iy0,jy0);
 905             dz00             = _mm_sub_pd(iz0,jz0);
 906             dx10             = _mm_sub_pd(ix1,jx0);
 907             dy10             = _mm_sub_pd(iy1,jy0);
 908             dz10             = _mm_sub_pd(iz1,jz0);
 909             dx20             = _mm_sub_pd(ix2,jx0);
 910             dy20             = _mm_sub_pd(iy2,jy0);
 911             dz20             = _mm_sub_pd(iz2,jz0);
 912
 913             /* Calculate squared distance and things based on it */
 914             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 915             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
 916             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
 917
 918             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 919             rinv10           = gmx_mm_invsqrt_pd(rsq10);
 920             rinv20           = gmx_mm_invsqrt_pd(rsq20);
 921
 922             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 923             rinvsq10         = _mm_mul_pd(rinv10,rinv10);
 924             rinvsq20         = _mm_mul_pd(rinv20,rinv20);
 925
 926             /* Load parameters for j particles */
 927             jq0              = _mm_load_sd(charge+jnrA+0);
 928             vdwjidx0A        = 2*vdwtype[jnrA+0];
 929
 930             fjx0             = _mm_setzero_pd();
 931             fjy0             = _mm_setzero_pd();
 932             fjz0             = _mm_setzero_pd();
 933
 934             /**************************
 935              * CALCULATE INTERACTIONS *
 936              **************************/
 937
 938             r00              = _mm_mul_pd(rsq00,rinv00);
 939
 940             /* Compute parameters for interactions between i and j atoms */
 941             qq00             = _mm_mul_pd(iq0,jq0);
 942             gmx_mm_load_1pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,&c6_00,&c12_00);
 943
 944             /* EWALD ELECTROSTATICS */
 945
 946             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 947             ewrt             = _mm_mul_pd(r00,ewtabscale);
 948             ewitab           = _mm_cvttpd_epi32(ewrt);
 949             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 950             gmx_mm_load_1pair_swizzle_pd(ewtab+gmx_mm_extract_epi32(ewitab,0),&ewtabF,&ewtabFn);
 951             felec            = _mm_add_pd(_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF),_mm_mul_pd(eweps,ewtabFn));
 952             felec            = _mm_mul_pd(_mm_mul_pd(qq00,rinv00),_mm_sub_pd(rinvsq00,felec));
 953
 954             /* LENNARD-JONES DISPERSION/REPULSION */
 955
 956             rinvsix          = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 957             fvdw             = _mm_mul_pd(_mm_sub_pd(_mm_mul_pd(c12_00,rinvsix),c6_00),_mm_mul_pd(rinvsix,rinvsq00));
 958
 959             fscal            = _mm_add_pd(felec,fvdw);
 960
 961             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 962
 963             /* Calculate temporary vectorial force */
 964             tx               = _mm_mul_pd(fscal,dx00);
 965             ty               = _mm_mul_pd(fscal,dy00);
 966             tz               = _mm_mul_pd(fscal,dz00);
 967
 968             /* Update vectorial force */
 969             fix0             = _mm_add_pd(fix0,tx);
 970             fiy0             = _mm_add_pd(fiy0,ty);
 971             fiz0             = _mm_add_pd(fiz0,tz);
 972
 973             fjx0             = _mm_add_pd(fjx0,tx);
 974             fjy0             = _mm_add_pd(fjy0,ty);
 975             fjz0             = _mm_add_pd(fjz0,tz);
 976
 977             /**************************
 978              * CALCULATE INTERACTIONS *
 979              **************************/
 980
 981             r10              = _mm_mul_pd(rsq10,rinv10);
 982
 983             /* Compute parameters for interactions between i and j atoms */
 984             qq10             = _mm_mul_pd(iq1,jq0);
 985
 986             /* EWALD ELECTROSTATICS */
 987
 988             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 989             ewrt             = _mm_mul_pd(r10,ewtabscale);
 990             ewitab           = _mm_cvttpd_epi32(ewrt);
 991             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 992             gmx_mm_load_1pair_swizzle_pd(ewtab+gmx_mm_extract_epi32(ewitab,0),&ewtabF,&ewtabFn);
 993             felec            = _mm_add_pd(_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF),_mm_mul_pd(eweps,ewtabFn));
 994             felec            = _mm_mul_pd(_mm_mul_pd(qq10,rinv10),_mm_sub_pd(rinvsq10,felec));
 995
 996             fscal            = felec;
 997
 998             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 999
1000             /* Calculate temporary vectorial force */
1001             tx               = _mm_mul_pd(fscal,dx10);
1002             ty               = _mm_mul_pd(fscal,dy10);
1003             tz               = _mm_mul_pd(fscal,dz10);
1004
1005             /* Update vectorial force */
1006             fix1             = _mm_add_pd(fix1,tx);
1007             fiy1             = _mm_add_pd(fiy1,ty);
1008             fiz1             = _mm_add_pd(fiz1,tz);
1009
1010             fjx0             = _mm_add_pd(fjx0,tx);
1011             fjy0             = _mm_add_pd(fjy0,ty);
1012             fjz0             = _mm_add_pd(fjz0,tz);
1013
1014             /**************************
1015              * CALCULATE INTERACTIONS *
1016              **************************/
1017
1018             r20              = _mm_mul_pd(rsq20,rinv20);
1019
1020             /* Compute parameters for interactions between i and j atoms */
1021             qq20             = _mm_mul_pd(iq2,jq0);
1022
1023             /* EWALD ELECTROSTATICS */
1024
1025             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1026             ewrt             = _mm_mul_pd(r20,ewtabscale);
1027             ewitab           = _mm_cvttpd_epi32(ewrt);
1028             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
1029             gmx_mm_load_1pair_swizzle_pd(ewtab+gmx_mm_extract_epi32(ewitab,0),&ewtabF,&ewtabFn);
1030             felec            = _mm_add_pd(_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF),_mm_mul_pd(eweps,ewtabFn));
1031             felec            = _mm_mul_pd(_mm_mul_pd(qq20,rinv20),_mm_sub_pd(rinvsq20,felec));
1032
1033             fscal            = felec;
1034
1035             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1036
1037             /* Calculate temporary vectorial force */
1038             tx               = _mm_mul_pd(fscal,dx20);
1039             ty               = _mm_mul_pd(fscal,dy20);
1040             tz               = _mm_mul_pd(fscal,dz20);
1041
1042             /* Update vectorial force */
1043             fix2             = _mm_add_pd(fix2,tx);
1044             fiy2             = _mm_add_pd(fiy2,ty);
1045             fiz2             = _mm_add_pd(fiz2,tz);
1046
1047             fjx0             = _mm_add_pd(fjx0,tx);
1048             fjy0             = _mm_add_pd(fjy0,ty);
1049             fjz0             = _mm_add_pd(fjz0,tz);
1050
1051             gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,fjx0,fjy0,fjz0);
1052
1053             /* Inner loop uses 118 flops */
1054         }
1055
1056         /* End of innermost loop */
1057
1058         gmx_mm_update_iforce_3atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
1059                                               f+i_coord_offset,fshift+i_shift_offset);
1060
1061         /* Increment number of inner iterations */
1062         inneriter                  += j_index_end - j_index_start;
1063
1064         /* Outer loop uses 18 flops */
1065     }
1066
1067     /* Increment number of outer iterations */
1068     outeriter        += nri;
1069
1070     /* Update outer/inner flops */
1071
1072     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3_F,outeriter*18 + inneriter*118);
1073 }