src/gromacs/gmxlib/nonbonded/nb_kernel_sse2_double/nb_kernel_ElecEw_VdwLJEw_GeomW3P1_sse2_double.c

   1 /*
   2  * This file is part of the GROMACS molecular simulation package.
   3  *
   4  * Copyright (c) 2012,2013,2014, by the GROMACS development team, led by
   5  * Mark Abraham, David van der Spoel, Berk Hess, and Erik Lindahl,
   6  * and including many others, as listed in the AUTHORS file in the
   7  * top-level source directory and at http://www.gromacs.org.
   8  *
   9  * GROMACS is free software; you can redistribute it and/or
  10  * modify it under the terms of the GNU Lesser General Public License
  11  * as published by the Free Software Foundation; either version 2.1
  12  * of the License, or (at your option) any later version.
  13  *
  14  * GROMACS is distributed in the hope that it will be useful,
  15  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  16  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  17  * Lesser General Public License for more details.
  18  *
  19  * You should have received a copy of the GNU Lesser General Public
  20  * License along with GROMACS; if not, see
  21  * http://www.gnu.org/licenses, or write to the Free Software Foundation,
  22  * Inc., 51 Franklin Street, Fifth Floor, Boston, MA  02110-1301  USA.
  23  *
  24  * If you want to redistribute modifications to GROMACS, please
  25  * consider that scientific software is very special. Version
  26  * control is crucial - bugs must be traceable. We will be happy to
  27  * consider code for inclusion in the official distribution, but
  28  * derived work must not be called official GROMACS. Details are found
  29  * in the README & COPYING files - if they are missing, get the
  30  * official version at http://www.gromacs.org.
  31  *
  32  * To help us fund GROMACS development, we humbly ask that you cite
  33  * the research papers on the package. Check out http://www.gromacs.org.
  34  */
  35 /*
  36  * Note: this file was generated by the GROMACS sse2_double kernel generator.
  37  */
  38 #include "config.h"
  39
  40 #include <math.h>
  41
  42 #include "../nb_kernel.h"
  43 #include "gromacs/legacyheaders/types/simple.h"
  44 #include "gromacs/math/vec.h"
  45 #include "gromacs/legacyheaders/nrnb.h"
  46
  47 #include "gromacs/simd/math_x86_sse2_double.h"
  48 #include "kernelutil_x86_sse2_double.h"
  49
  50 /*
  51  * Gromacs nonbonded kernel:   nb_kernel_ElecEw_VdwLJEw_GeomW3P1_VF_sse2_double
  52  * Electrostatics interaction: Ewald
  53  * VdW interaction:            LJEwald
  54  * Geometry:                   Water3-Particle
  55  * Calculate force/pot:        PotentialAndForce
  56  */
  57 void
  58 nb_kernel_ElecEw_VdwLJEw_GeomW3P1_VF_sse2_double
  59                     (t_nblist                    * gmx_restrict       nlist,
  60                      rvec                        * gmx_restrict          xx,
  61                      rvec                        * gmx_restrict          ff,
  62                      t_forcerec                  * gmx_restrict          fr,
  63                      t_mdatoms                   * gmx_restrict     mdatoms,
  64                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
  65                      t_nrnb                      * gmx_restrict        nrnb)
  66 {
  67     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
  68      * just 0 for non-waters.
  69      * Suffixes A,B refer to j loop unrolling done with SSE double precision, e.g. for the two different
  70      * jnr indices corresponding to data put in the four positions in the SIMD register.
  71      */
  72     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
  73     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
  74     int              jnrA,jnrB;
  75     int              j_coord_offsetA,j_coord_offsetB;
  76     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
  77     real             rcutoff_scalar;
  78     real             *shiftvec,*fshift,*x,*f;
  79     __m128d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
  80     int              vdwioffset0;
  81     __m128d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
  82     int              vdwioffset1;
  83     __m128d          ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
  84     int              vdwioffset2;
  85     __m128d          ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
  86     int              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     __m128d           c6grid_00;
 100     __m128d           c6grid_10;
 101     __m128d           c6grid_20;
 102     __m128d           ewclj,ewclj2,ewclj6,ewcljrsq,poly,exponent,f6A,f6B,sh_lj_ewald;
 103     real             *vdwgridparam;
 104     __m128d           one_half = _mm_set1_pd(0.5);
 105     __m128d           minus_one = _mm_set1_pd(-1.0);
 106     __m128i          ewitab;
 107     __m128d          ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
 108     real             *ewtab;
 109     __m128d          dummy_mask,cutoff_mask;
 110     __m128d          signbit   = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
 111     __m128d          one     = _mm_set1_pd(1.0);
 112     __m128d          two     = _mm_set1_pd(2.0);
 113     x                = xx[0];
 114     f                = ff[0];
 115
 116     nri              = nlist->nri;
 117     iinr             = nlist->iinr;
 118     jindex           = nlist->jindex;
 119     jjnr             = nlist->jjnr;
 120     shiftidx         = nlist->shift;
 121     gid              = nlist->gid;
 122     shiftvec         = fr->shift_vec[0];
 123     fshift           = fr->fshift[0];
 124     facel            = _mm_set1_pd(fr->epsfac);
 125     charge           = mdatoms->chargeA;
 126     nvdwtype         = fr->ntype;
 127     vdwparam         = fr->nbfp;
 128     vdwtype          = mdatoms->typeA;
 129     vdwgridparam     = fr->ljpme_c6grid;
 130     sh_lj_ewald      = _mm_set1_pd(fr->ic->sh_lj_ewald);
 131     ewclj            = _mm_set1_pd(fr->ewaldcoeff_lj);
 132     ewclj2           = _mm_mul_pd(minus_one,_mm_mul_pd(ewclj,ewclj));
 133
 134     sh_ewald         = _mm_set1_pd(fr->ic->sh_ewald);
 135     ewtab            = fr->ic->tabq_coul_FDV0;
 136     ewtabscale       = _mm_set1_pd(fr->ic->tabq_scale);
 137     ewtabhalfspace   = _mm_set1_pd(0.5/fr->ic->tabq_scale);
 138
 139     /* Setup water-specific parameters */
 140     inr              = nlist->iinr[0];
 141     iq0              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+0]));
 142     iq1              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+1]));
 143     iq2              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+2]));
 144     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 145
 146     /* Avoid stupid compiler warnings */
 147     jnrA = jnrB = 0;
 148     j_coord_offsetA = 0;
 149     j_coord_offsetB = 0;
 150
 151     outeriter        = 0;
 152     inneriter        = 0;
 153
 154     /* Start outer loop over neighborlists */
 155     for(iidx=0; iidx<nri; iidx++)
 156     {
 157         /* Load shift vector for this list */
 158         i_shift_offset   = DIM*shiftidx[iidx];
 159
 160         /* Load limits for loop over neighbors */
 161         j_index_start    = jindex[iidx];
 162         j_index_end      = jindex[iidx+1];
 163
 164         /* Get outer coordinate index */
 165         inr              = iinr[iidx];
 166         i_coord_offset   = DIM*inr;
 167
 168         /* Load i particle coords and add shift vector */
 169         gmx_mm_load_shift_and_3rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
 170                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 171
 172         fix0             = _mm_setzero_pd();
 173         fiy0             = _mm_setzero_pd();
 174         fiz0             = _mm_setzero_pd();
 175         fix1             = _mm_setzero_pd();
 176         fiy1             = _mm_setzero_pd();
 177         fiz1             = _mm_setzero_pd();
 178         fix2             = _mm_setzero_pd();
 179         fiy2             = _mm_setzero_pd();
 180         fiz2             = _mm_setzero_pd();
 181
 182         /* Reset potential sums */
 183         velecsum         = _mm_setzero_pd();
 184         vvdwsum          = _mm_setzero_pd();
 185
 186         /* Start inner kernel loop */
 187         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 188         {
 189
 190             /* Get j neighbor index, and coordinate index */
 191             jnrA             = jjnr[jidx];
 192             jnrB             = jjnr[jidx+1];
 193             j_coord_offsetA  = DIM*jnrA;
 194             j_coord_offsetB  = DIM*jnrB;
 195
 196             /* load j atom coordinates */
 197             gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 198                                               &jx0,&jy0,&jz0);
 199
 200             /* Calculate displacement vector */
 201             dx00             = _mm_sub_pd(ix0,jx0);
 202             dy00             = _mm_sub_pd(iy0,jy0);
 203             dz00             = _mm_sub_pd(iz0,jz0);
 204             dx10             = _mm_sub_pd(ix1,jx0);
 205             dy10             = _mm_sub_pd(iy1,jy0);
 206             dz10             = _mm_sub_pd(iz1,jz0);
 207             dx20             = _mm_sub_pd(ix2,jx0);
 208             dy20             = _mm_sub_pd(iy2,jy0);
 209             dz20             = _mm_sub_pd(iz2,jz0);
 210
 211             /* Calculate squared distance and things based on it */
 212             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 213             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
 214             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
 215
 216             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 217             rinv10           = gmx_mm_invsqrt_pd(rsq10);
 218             rinv20           = gmx_mm_invsqrt_pd(rsq20);
 219
 220             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 221             rinvsq10         = _mm_mul_pd(rinv10,rinv10);
 222             rinvsq20         = _mm_mul_pd(rinv20,rinv20);
 223
 224             /* Load parameters for j particles */
 225             jq0              = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
 226             vdwjidx0A        = 2*vdwtype[jnrA+0];
 227             vdwjidx0B        = 2*vdwtype[jnrB+0];
 228
 229             fjx0             = _mm_setzero_pd();
 230             fjy0             = _mm_setzero_pd();
 231             fjz0             = _mm_setzero_pd();
 232
 233             /**************************
 234              * CALCULATE INTERACTIONS *
 235              **************************/
 236
 237             r00              = _mm_mul_pd(rsq00,rinv00);
 238
 239             /* Compute parameters for interactions between i and j atoms */
 240             qq00             = _mm_mul_pd(iq0,jq0);
 241             gmx_mm_load_2pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,
 242                                          vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
 243
 244             c6grid_00       = gmx_mm_load_2real_swizzle_pd(vdwgridparam+vdwioffset0+vdwjidx0A,
 245                                                                vdwgridparam+vdwioffset0+vdwjidx0B);
 246
 247             /* EWALD ELECTROSTATICS */
 248
 249             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 250             ewrt             = _mm_mul_pd(r00,ewtabscale);
 251             ewitab           = _mm_cvttpd_epi32(ewrt);
 252             eweps            = _mm_sub_pd(ewrt,_mm_cvtepi32_pd(ewitab));
 253             ewitab           = _mm_slli_epi32(ewitab,2);
 254             ewtabF           = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 255             ewtabD           = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,1) );
 256             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 257             ewtabV           = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,0) +2);
 258             ewtabFn          = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,1) +2);
 259             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 260             felec            = _mm_add_pd(ewtabF,_mm_mul_pd(eweps,ewtabD));
 261             velec            = _mm_sub_pd(ewtabV,_mm_mul_pd(_mm_mul_pd(ewtabhalfspace,eweps),_mm_add_pd(ewtabF,felec)));
 262             velec            = _mm_mul_pd(qq00,_mm_sub_pd(rinv00,velec));
 263             felec            = _mm_mul_pd(_mm_mul_pd(qq00,rinv00),_mm_sub_pd(rinvsq00,felec));
 264
 265             /* Analytical LJ-PME */
 266             rinvsix          = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 267             ewcljrsq         = _mm_mul_pd(ewclj2,rsq00);
 268             ewclj6           = _mm_mul_pd(ewclj2,_mm_mul_pd(ewclj2,ewclj2));
 269             exponent         = gmx_simd_exp_d(ewcljrsq);
 270             /* poly = exp(-(beta*r)^2) * (1 + (beta*r)^2 + (beta*r)^4 /2) */
 271             poly             = _mm_mul_pd(exponent,_mm_add_pd(_mm_sub_pd(one,ewcljrsq),_mm_mul_pd(_mm_mul_pd(ewcljrsq,ewcljrsq),one_half)));
 272             /* vvdw6 = [C6 - C6grid * (1-poly)]/r6 */
 273             vvdw6            = _mm_mul_pd(_mm_sub_pd(c6_00,_mm_mul_pd(c6grid_00,_mm_sub_pd(one,poly))),rinvsix);
 274             vvdw12           = _mm_mul_pd(c12_00,_mm_mul_pd(rinvsix,rinvsix));
 275             vvdw             = _mm_sub_pd(_mm_mul_pd(vvdw12,one_twelfth),_mm_mul_pd(vvdw6,one_sixth));
 276             /* fvdw = vvdw12/r - (vvdw6/r + (C6grid * exponent * beta^6)/r) */
 277             fvdw             = _mm_mul_pd(_mm_sub_pd(vvdw12,_mm_sub_pd(vvdw6,_mm_mul_pd(_mm_mul_pd(c6grid_00,one_sixth),_mm_mul_pd(exponent,ewclj6)))),rinvsq00);
 278
 279             /* Update potential sum for this i atom from the interaction with this j atom. */
 280             velecsum         = _mm_add_pd(velecsum,velec);
 281             vvdwsum          = _mm_add_pd(vvdwsum,vvdw);
 282
 283             fscal            = _mm_add_pd(felec,fvdw);
 284
 285             /* Calculate temporary vectorial force */
 286             tx               = _mm_mul_pd(fscal,dx00);
 287             ty               = _mm_mul_pd(fscal,dy00);
 288             tz               = _mm_mul_pd(fscal,dz00);
 289
 290             /* Update vectorial force */
 291             fix0             = _mm_add_pd(fix0,tx);
 292             fiy0             = _mm_add_pd(fiy0,ty);
 293             fiz0             = _mm_add_pd(fiz0,tz);
 294
 295             fjx0             = _mm_add_pd(fjx0,tx);
 296             fjy0             = _mm_add_pd(fjy0,ty);
 297             fjz0             = _mm_add_pd(fjz0,tz);
 298
 299             /**************************
 300              * CALCULATE INTERACTIONS *
 301              **************************/
 302
 303             r10              = _mm_mul_pd(rsq10,rinv10);
 304
 305             /* Compute parameters for interactions between i and j atoms */
 306             qq10             = _mm_mul_pd(iq1,jq0);
 307
 308             /* EWALD ELECTROSTATICS */
 309
 310             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 311             ewrt             = _mm_mul_pd(r10,ewtabscale);
 312             ewitab           = _mm_cvttpd_epi32(ewrt);
 313             eweps            = _mm_sub_pd(ewrt,_mm_cvtepi32_pd(ewitab));
 314             ewitab           = _mm_slli_epi32(ewitab,2);
 315             ewtabF           = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 316             ewtabD           = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,1) );
 317             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 318             ewtabV           = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,0) +2);
 319             ewtabFn          = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,1) +2);
 320             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 321             felec            = _mm_add_pd(ewtabF,_mm_mul_pd(eweps,ewtabD));
 322             velec            = _mm_sub_pd(ewtabV,_mm_mul_pd(_mm_mul_pd(ewtabhalfspace,eweps),_mm_add_pd(ewtabF,felec)));
 323             velec            = _mm_mul_pd(qq10,_mm_sub_pd(rinv10,velec));
 324             felec            = _mm_mul_pd(_mm_mul_pd(qq10,rinv10),_mm_sub_pd(rinvsq10,felec));
 325
 326             /* Update potential sum for this i atom from the interaction with this j atom. */
 327             velecsum         = _mm_add_pd(velecsum,velec);
 328
 329             fscal            = felec;
 330
 331             /* Calculate temporary vectorial force */
 332             tx               = _mm_mul_pd(fscal,dx10);
 333             ty               = _mm_mul_pd(fscal,dy10);
 334             tz               = _mm_mul_pd(fscal,dz10);
 335
 336             /* Update vectorial force */
 337             fix1             = _mm_add_pd(fix1,tx);
 338             fiy1             = _mm_add_pd(fiy1,ty);
 339             fiz1             = _mm_add_pd(fiz1,tz);
 340
 341             fjx0             = _mm_add_pd(fjx0,tx);
 342             fjy0             = _mm_add_pd(fjy0,ty);
 343             fjz0             = _mm_add_pd(fjz0,tz);
 344
 345             /**************************
 346              * CALCULATE INTERACTIONS *
 347              **************************/
 348
 349             r20              = _mm_mul_pd(rsq20,rinv20);
 350
 351             /* Compute parameters for interactions between i and j atoms */
 352             qq20             = _mm_mul_pd(iq2,jq0);
 353
 354             /* EWALD ELECTROSTATICS */
 355
 356             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 357             ewrt             = _mm_mul_pd(r20,ewtabscale);
 358             ewitab           = _mm_cvttpd_epi32(ewrt);
 359             eweps            = _mm_sub_pd(ewrt,_mm_cvtepi32_pd(ewitab));
 360             ewitab           = _mm_slli_epi32(ewitab,2);
 361             ewtabF           = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 362             ewtabD           = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,1) );
 363             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 364             ewtabV           = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,0) +2);
 365             ewtabFn          = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,1) +2);
 366             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 367             felec            = _mm_add_pd(ewtabF,_mm_mul_pd(eweps,ewtabD));
 368             velec            = _mm_sub_pd(ewtabV,_mm_mul_pd(_mm_mul_pd(ewtabhalfspace,eweps),_mm_add_pd(ewtabF,felec)));
 369             velec            = _mm_mul_pd(qq20,_mm_sub_pd(rinv20,velec));
 370             felec            = _mm_mul_pd(_mm_mul_pd(qq20,rinv20),_mm_sub_pd(rinvsq20,felec));
 371
 372             /* Update potential sum for this i atom from the interaction with this j atom. */
 373             velecsum         = _mm_add_pd(velecsum,velec);
 374
 375             fscal            = felec;
 376
 377             /* Calculate temporary vectorial force */
 378             tx               = _mm_mul_pd(fscal,dx20);
 379             ty               = _mm_mul_pd(fscal,dy20);
 380             tz               = _mm_mul_pd(fscal,dz20);
 381
 382             /* Update vectorial force */
 383             fix2             = _mm_add_pd(fix2,tx);
 384             fiy2             = _mm_add_pd(fiy2,ty);
 385             fiz2             = _mm_add_pd(fiz2,tz);
 386
 387             fjx0             = _mm_add_pd(fjx0,tx);
 388             fjy0             = _mm_add_pd(fjy0,ty);
 389             fjz0             = _mm_add_pd(fjz0,tz);
 390
 391             gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,fjx0,fjy0,fjz0);
 392
 393             /* Inner loop uses 154 flops */
 394         }
 395
 396         if(jidx<j_index_end)
 397         {
 398
 399             jnrA             = jjnr[jidx];
 400             j_coord_offsetA  = DIM*jnrA;
 401
 402             /* load j atom coordinates */
 403             gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
 404                                               &jx0,&jy0,&jz0);
 405
 406             /* Calculate displacement vector */
 407             dx00             = _mm_sub_pd(ix0,jx0);
 408             dy00             = _mm_sub_pd(iy0,jy0);
 409             dz00             = _mm_sub_pd(iz0,jz0);
 410             dx10             = _mm_sub_pd(ix1,jx0);
 411             dy10             = _mm_sub_pd(iy1,jy0);
 412             dz10             = _mm_sub_pd(iz1,jz0);
 413             dx20             = _mm_sub_pd(ix2,jx0);
 414             dy20             = _mm_sub_pd(iy2,jy0);
 415             dz20             = _mm_sub_pd(iz2,jz0);
 416
 417             /* Calculate squared distance and things based on it */
 418             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 419             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
 420             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
 421
 422             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 423             rinv10           = gmx_mm_invsqrt_pd(rsq10);
 424             rinv20           = gmx_mm_invsqrt_pd(rsq20);
 425
 426             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 427             rinvsq10         = _mm_mul_pd(rinv10,rinv10);
 428             rinvsq20         = _mm_mul_pd(rinv20,rinv20);
 429
 430             /* Load parameters for j particles */
 431             jq0              = _mm_load_sd(charge+jnrA+0);
 432             vdwjidx0A        = 2*vdwtype[jnrA+0];
 433
 434             fjx0             = _mm_setzero_pd();
 435             fjy0             = _mm_setzero_pd();
 436             fjz0             = _mm_setzero_pd();
 437
 438             /**************************
 439              * CALCULATE INTERACTIONS *
 440              **************************/
 441
 442             r00              = _mm_mul_pd(rsq00,rinv00);
 443
 444             /* Compute parameters for interactions between i and j atoms */
 445             qq00             = _mm_mul_pd(iq0,jq0);
 446             gmx_mm_load_1pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,&c6_00,&c12_00);
 447
 448             c6grid_00       = gmx_mm_load_1real_pd(vdwgridparam+vdwioffset0+vdwjidx0A);
 449
 450             /* EWALD ELECTROSTATICS */
 451
 452             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 453             ewrt             = _mm_mul_pd(r00,ewtabscale);
 454             ewitab           = _mm_cvttpd_epi32(ewrt);
 455             eweps            = _mm_sub_pd(ewrt,_mm_cvtepi32_pd(ewitab));
 456             ewitab           = _mm_slli_epi32(ewitab,2);
 457             ewtabF           = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 458             ewtabD           = _mm_setzero_pd();
 459             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 460             ewtabV           = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,0) +2);
 461             ewtabFn          = _mm_setzero_pd();
 462             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 463             felec            = _mm_add_pd(ewtabF,_mm_mul_pd(eweps,ewtabD));
 464             velec            = _mm_sub_pd(ewtabV,_mm_mul_pd(_mm_mul_pd(ewtabhalfspace,eweps),_mm_add_pd(ewtabF,felec)));
 465             velec            = _mm_mul_pd(qq00,_mm_sub_pd(rinv00,velec));
 466             felec            = _mm_mul_pd(_mm_mul_pd(qq00,rinv00),_mm_sub_pd(rinvsq00,felec));
 467
 468             /* Analytical LJ-PME */
 469             rinvsix          = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 470             ewcljrsq         = _mm_mul_pd(ewclj2,rsq00);
 471             ewclj6           = _mm_mul_pd(ewclj2,_mm_mul_pd(ewclj2,ewclj2));
 472             exponent         = gmx_simd_exp_d(ewcljrsq);
 473             /* poly = exp(-(beta*r)^2) * (1 + (beta*r)^2 + (beta*r)^4 /2) */
 474             poly             = _mm_mul_pd(exponent,_mm_add_pd(_mm_sub_pd(one,ewcljrsq),_mm_mul_pd(_mm_mul_pd(ewcljrsq,ewcljrsq),one_half)));
 475             /* vvdw6 = [C6 - C6grid * (1-poly)]/r6 */
 476             vvdw6            = _mm_mul_pd(_mm_sub_pd(c6_00,_mm_mul_pd(c6grid_00,_mm_sub_pd(one,poly))),rinvsix);
 477             vvdw12           = _mm_mul_pd(c12_00,_mm_mul_pd(rinvsix,rinvsix));
 478             vvdw             = _mm_sub_pd(_mm_mul_pd(vvdw12,one_twelfth),_mm_mul_pd(vvdw6,one_sixth));
 479             /* fvdw = vvdw12/r - (vvdw6/r + (C6grid * exponent * beta^6)/r) */
 480             fvdw             = _mm_mul_pd(_mm_sub_pd(vvdw12,_mm_sub_pd(vvdw6,_mm_mul_pd(_mm_mul_pd(c6grid_00,one_sixth),_mm_mul_pd(exponent,ewclj6)))),rinvsq00);
 481
 482             /* Update potential sum for this i atom from the interaction with this j atom. */
 483             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 484             velecsum         = _mm_add_pd(velecsum,velec);
 485             vvdw             = _mm_unpacklo_pd(vvdw,_mm_setzero_pd());
 486             vvdwsum          = _mm_add_pd(vvdwsum,vvdw);
 487
 488             fscal            = _mm_add_pd(felec,fvdw);
 489
 490             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 491
 492             /* Calculate temporary vectorial force */
 493             tx               = _mm_mul_pd(fscal,dx00);
 494             ty               = _mm_mul_pd(fscal,dy00);
 495             tz               = _mm_mul_pd(fscal,dz00);
 496
 497             /* Update vectorial force */
 498             fix0             = _mm_add_pd(fix0,tx);
 499             fiy0             = _mm_add_pd(fiy0,ty);
 500             fiz0             = _mm_add_pd(fiz0,tz);
 501
 502             fjx0             = _mm_add_pd(fjx0,tx);
 503             fjy0             = _mm_add_pd(fjy0,ty);
 504             fjz0             = _mm_add_pd(fjz0,tz);
 505
 506             /**************************
 507              * CALCULATE INTERACTIONS *
 508              **************************/
 509
 510             r10              = _mm_mul_pd(rsq10,rinv10);
 511
 512             /* Compute parameters for interactions between i and j atoms */
 513             qq10             = _mm_mul_pd(iq1,jq0);
 514
 515             /* EWALD ELECTROSTATICS */
 516
 517             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 518             ewrt             = _mm_mul_pd(r10,ewtabscale);
 519             ewitab           = _mm_cvttpd_epi32(ewrt);
 520             eweps            = _mm_sub_pd(ewrt,_mm_cvtepi32_pd(ewitab));
 521             ewitab           = _mm_slli_epi32(ewitab,2);
 522             ewtabF           = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 523             ewtabD           = _mm_setzero_pd();
 524             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 525             ewtabV           = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,0) +2);
 526             ewtabFn          = _mm_setzero_pd();
 527             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 528             felec            = _mm_add_pd(ewtabF,_mm_mul_pd(eweps,ewtabD));
 529             velec            = _mm_sub_pd(ewtabV,_mm_mul_pd(_mm_mul_pd(ewtabhalfspace,eweps),_mm_add_pd(ewtabF,felec)));
 530             velec            = _mm_mul_pd(qq10,_mm_sub_pd(rinv10,velec));
 531             felec            = _mm_mul_pd(_mm_mul_pd(qq10,rinv10),_mm_sub_pd(rinvsq10,felec));
 532
 533             /* Update potential sum for this i atom from the interaction with this j atom. */
 534             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 535             velecsum         = _mm_add_pd(velecsum,velec);
 536
 537             fscal            = felec;
 538
 539             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 540
 541             /* Calculate temporary vectorial force */
 542             tx               = _mm_mul_pd(fscal,dx10);
 543             ty               = _mm_mul_pd(fscal,dy10);
 544             tz               = _mm_mul_pd(fscal,dz10);
 545
 546             /* Update vectorial force */
 547             fix1             = _mm_add_pd(fix1,tx);
 548             fiy1             = _mm_add_pd(fiy1,ty);
 549             fiz1             = _mm_add_pd(fiz1,tz);
 550
 551             fjx0             = _mm_add_pd(fjx0,tx);
 552             fjy0             = _mm_add_pd(fjy0,ty);
 553             fjz0             = _mm_add_pd(fjz0,tz);
 554
 555             /**************************
 556              * CALCULATE INTERACTIONS *
 557              **************************/
 558
 559             r20              = _mm_mul_pd(rsq20,rinv20);
 560
 561             /* Compute parameters for interactions between i and j atoms */
 562             qq20             = _mm_mul_pd(iq2,jq0);
 563
 564             /* EWALD ELECTROSTATICS */
 565
 566             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 567             ewrt             = _mm_mul_pd(r20,ewtabscale);
 568             ewitab           = _mm_cvttpd_epi32(ewrt);
 569             eweps            = _mm_sub_pd(ewrt,_mm_cvtepi32_pd(ewitab));
 570             ewitab           = _mm_slli_epi32(ewitab,2);
 571             ewtabF           = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 572             ewtabD           = _mm_setzero_pd();
 573             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 574             ewtabV           = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,0) +2);
 575             ewtabFn          = _mm_setzero_pd();
 576             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 577             felec            = _mm_add_pd(ewtabF,_mm_mul_pd(eweps,ewtabD));
 578             velec            = _mm_sub_pd(ewtabV,_mm_mul_pd(_mm_mul_pd(ewtabhalfspace,eweps),_mm_add_pd(ewtabF,felec)));
 579             velec            = _mm_mul_pd(qq20,_mm_sub_pd(rinv20,velec));
 580             felec            = _mm_mul_pd(_mm_mul_pd(qq20,rinv20),_mm_sub_pd(rinvsq20,felec));
 581
 582             /* Update potential sum for this i atom from the interaction with this j atom. */
 583             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 584             velecsum         = _mm_add_pd(velecsum,velec);
 585
 586             fscal            = felec;
 587
 588             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 589
 590             /* Calculate temporary vectorial force */
 591             tx               = _mm_mul_pd(fscal,dx20);
 592             ty               = _mm_mul_pd(fscal,dy20);
 593             tz               = _mm_mul_pd(fscal,dz20);
 594
 595             /* Update vectorial force */
 596             fix2             = _mm_add_pd(fix2,tx);
 597             fiy2             = _mm_add_pd(fiy2,ty);
 598             fiz2             = _mm_add_pd(fiz2,tz);
 599
 600             fjx0             = _mm_add_pd(fjx0,tx);
 601             fjy0             = _mm_add_pd(fjy0,ty);
 602             fjz0             = _mm_add_pd(fjz0,tz);
 603
 604             gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,fjx0,fjy0,fjz0);
 605
 606             /* Inner loop uses 154 flops */
 607         }
 608
 609         /* End of innermost loop */
 610
 611         gmx_mm_update_iforce_3atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
 612                                               f+i_coord_offset,fshift+i_shift_offset);
 613
 614         ggid                        = gid[iidx];
 615         /* Update potential energies */
 616         gmx_mm_update_1pot_pd(velecsum,kernel_data->energygrp_elec+ggid);
 617         gmx_mm_update_1pot_pd(vvdwsum,kernel_data->energygrp_vdw+ggid);
 618
 619         /* Increment number of inner iterations */
 620         inneriter                  += j_index_end - j_index_start;
 621
 622         /* Outer loop uses 20 flops */
 623     }
 624
 625     /* Increment number of outer iterations */
 626     outeriter        += nri;
 627
 628     /* Update outer/inner flops */
 629
 630     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3_VF,outeriter*20 + inneriter*154);
 631 }
 632 /*
 633  * Gromacs nonbonded kernel:   nb_kernel_ElecEw_VdwLJEw_GeomW3P1_F_sse2_double
 634  * Electrostatics interaction: Ewald
 635  * VdW interaction:            LJEwald
 636  * Geometry:                   Water3-Particle
 637  * Calculate force/pot:        Force
 638  */
 639 void
 640 nb_kernel_ElecEw_VdwLJEw_GeomW3P1_F_sse2_double
 641                     (t_nblist                    * gmx_restrict       nlist,
 642                      rvec                        * gmx_restrict          xx,
 643                      rvec                        * gmx_restrict          ff,
 644                      t_forcerec                  * gmx_restrict          fr,
 645                      t_mdatoms                   * gmx_restrict     mdatoms,
 646                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 647                      t_nrnb                      * gmx_restrict        nrnb)
 648 {
 649     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 650      * just 0 for non-waters.
 651      * Suffixes A,B refer to j loop unrolling done with SSE double precision, e.g. for the two different
 652      * jnr indices corresponding to data put in the four positions in the SIMD register.
 653      */
 654     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 655     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 656     int              jnrA,jnrB;
 657     int              j_coord_offsetA,j_coord_offsetB;
 658     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 659     real             rcutoff_scalar;
 660     real             *shiftvec,*fshift,*x,*f;
 661     __m128d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 662     int              vdwioffset0;
 663     __m128d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 664     int              vdwioffset1;
 665     __m128d          ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
 666     int              vdwioffset2;
 667     __m128d          ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
 668     int              vdwjidx0A,vdwjidx0B;
 669     __m128d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 670     __m128d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 671     __m128d          dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
 672     __m128d          dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
 673     __m128d          velec,felec,velecsum,facel,crf,krf,krf2;
 674     real             *charge;
 675     int              nvdwtype;
 676     __m128d          rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 677     int              *vdwtype;
 678     real             *vdwparam;
 679     __m128d          one_sixth   = _mm_set1_pd(1.0/6.0);
 680     __m128d          one_twelfth = _mm_set1_pd(1.0/12.0);
 681     __m128d           c6grid_00;
 682     __m128d           c6grid_10;
 683     __m128d           c6grid_20;
 684     __m128d           ewclj,ewclj2,ewclj6,ewcljrsq,poly,exponent,f6A,f6B,sh_lj_ewald;
 685     real             *vdwgridparam;
 686     __m128d           one_half = _mm_set1_pd(0.5);
 687     __m128d           minus_one = _mm_set1_pd(-1.0);
 688     __m128i          ewitab;
 689     __m128d          ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
 690     real             *ewtab;
 691     __m128d          dummy_mask,cutoff_mask;
 692     __m128d          signbit   = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
 693     __m128d          one     = _mm_set1_pd(1.0);
 694     __m128d          two     = _mm_set1_pd(2.0);
 695     x                = xx[0];
 696     f                = ff[0];
 697
 698     nri              = nlist->nri;
 699     iinr             = nlist->iinr;
 700     jindex           = nlist->jindex;
 701     jjnr             = nlist->jjnr;
 702     shiftidx         = nlist->shift;
 703     gid              = nlist->gid;
 704     shiftvec         = fr->shift_vec[0];
 705     fshift           = fr->fshift[0];
 706     facel            = _mm_set1_pd(fr->epsfac);
 707     charge           = mdatoms->chargeA;
 708     nvdwtype         = fr->ntype;
 709     vdwparam         = fr->nbfp;
 710     vdwtype          = mdatoms->typeA;
 711     vdwgridparam     = fr->ljpme_c6grid;
 712     sh_lj_ewald      = _mm_set1_pd(fr->ic->sh_lj_ewald);
 713     ewclj            = _mm_set1_pd(fr->ewaldcoeff_lj);
 714     ewclj2           = _mm_mul_pd(minus_one,_mm_mul_pd(ewclj,ewclj));
 715
 716     sh_ewald         = _mm_set1_pd(fr->ic->sh_ewald);
 717     ewtab            = fr->ic->tabq_coul_F;
 718     ewtabscale       = _mm_set1_pd(fr->ic->tabq_scale);
 719     ewtabhalfspace   = _mm_set1_pd(0.5/fr->ic->tabq_scale);
 720
 721     /* Setup water-specific parameters */
 722     inr              = nlist->iinr[0];
 723     iq0              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+0]));
 724     iq1              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+1]));
 725     iq2              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+2]));
 726     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 727
 728     /* Avoid stupid compiler warnings */
 729     jnrA = jnrB = 0;
 730     j_coord_offsetA = 0;
 731     j_coord_offsetB = 0;
 732
 733     outeriter        = 0;
 734     inneriter        = 0;
 735
 736     /* Start outer loop over neighborlists */
 737     for(iidx=0; iidx<nri; iidx++)
 738     {
 739         /* Load shift vector for this list */
 740         i_shift_offset   = DIM*shiftidx[iidx];
 741
 742         /* Load limits for loop over neighbors */
 743         j_index_start    = jindex[iidx];
 744         j_index_end      = jindex[iidx+1];
 745
 746         /* Get outer coordinate index */
 747         inr              = iinr[iidx];
 748         i_coord_offset   = DIM*inr;
 749
 750         /* Load i particle coords and add shift vector */
 751         gmx_mm_load_shift_and_3rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
 752                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 753
 754         fix0             = _mm_setzero_pd();
 755         fiy0             = _mm_setzero_pd();
 756         fiz0             = _mm_setzero_pd();
 757         fix1             = _mm_setzero_pd();
 758         fiy1             = _mm_setzero_pd();
 759         fiz1             = _mm_setzero_pd();
 760         fix2             = _mm_setzero_pd();
 761         fiy2             = _mm_setzero_pd();
 762         fiz2             = _mm_setzero_pd();
 763
 764         /* Start inner kernel loop */
 765         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 766         {
 767
 768             /* Get j neighbor index, and coordinate index */
 769             jnrA             = jjnr[jidx];
 770             jnrB             = jjnr[jidx+1];
 771             j_coord_offsetA  = DIM*jnrA;
 772             j_coord_offsetB  = DIM*jnrB;
 773
 774             /* load j atom coordinates */
 775             gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 776                                               &jx0,&jy0,&jz0);
 777
 778             /* Calculate displacement vector */
 779             dx00             = _mm_sub_pd(ix0,jx0);
 780             dy00             = _mm_sub_pd(iy0,jy0);
 781             dz00             = _mm_sub_pd(iz0,jz0);
 782             dx10             = _mm_sub_pd(ix1,jx0);
 783             dy10             = _mm_sub_pd(iy1,jy0);
 784             dz10             = _mm_sub_pd(iz1,jz0);
 785             dx20             = _mm_sub_pd(ix2,jx0);
 786             dy20             = _mm_sub_pd(iy2,jy0);
 787             dz20             = _mm_sub_pd(iz2,jz0);
 788
 789             /* Calculate squared distance and things based on it */
 790             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 791             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
 792             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
 793
 794             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 795             rinv10           = gmx_mm_invsqrt_pd(rsq10);
 796             rinv20           = gmx_mm_invsqrt_pd(rsq20);
 797
 798             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 799             rinvsq10         = _mm_mul_pd(rinv10,rinv10);
 800             rinvsq20         = _mm_mul_pd(rinv20,rinv20);
 801
 802             /* Load parameters for j particles */
 803             jq0              = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
 804             vdwjidx0A        = 2*vdwtype[jnrA+0];
 805             vdwjidx0B        = 2*vdwtype[jnrB+0];
 806
 807             fjx0             = _mm_setzero_pd();
 808             fjy0             = _mm_setzero_pd();
 809             fjz0             = _mm_setzero_pd();
 810
 811             /**************************
 812              * CALCULATE INTERACTIONS *
 813              **************************/
 814
 815             r00              = _mm_mul_pd(rsq00,rinv00);
 816
 817             /* Compute parameters for interactions between i and j atoms */
 818             qq00             = _mm_mul_pd(iq0,jq0);
 819             gmx_mm_load_2pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,
 820                                          vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
 821
 822             c6grid_00       = gmx_mm_load_2real_swizzle_pd(vdwgridparam+vdwioffset0+vdwjidx0A,
 823                                                                vdwgridparam+vdwioffset0+vdwjidx0B);
 824
 825             /* EWALD ELECTROSTATICS */
 826
 827             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 828             ewrt             = _mm_mul_pd(r00,ewtabscale);
 829             ewitab           = _mm_cvttpd_epi32(ewrt);
 830             eweps            = _mm_sub_pd(ewrt,_mm_cvtepi32_pd(ewitab));
 831             gmx_mm_load_2pair_swizzle_pd(ewtab+gmx_mm_extract_epi32(ewitab,0),ewtab+gmx_mm_extract_epi32(ewitab,1),
 832                                          &ewtabF,&ewtabFn);
 833             felec            = _mm_add_pd(_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF),_mm_mul_pd(eweps,ewtabFn));
 834             felec            = _mm_mul_pd(_mm_mul_pd(qq00,rinv00),_mm_sub_pd(rinvsq00,felec));
 835
 836             /* Analytical LJ-PME */
 837             rinvsix          = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 838             ewcljrsq         = _mm_mul_pd(ewclj2,rsq00);
 839             ewclj6           = _mm_mul_pd(ewclj2,_mm_mul_pd(ewclj2,ewclj2));
 840             exponent         = gmx_simd_exp_d(ewcljrsq);
 841             /* poly = exp(-(beta*r)^2) * (1 + (beta*r)^2 + (beta*r)^4 /2) */
 842             poly             = _mm_mul_pd(exponent,_mm_add_pd(_mm_sub_pd(one,ewcljrsq),_mm_mul_pd(_mm_mul_pd(ewcljrsq,ewcljrsq),one_half)));
 843             /* f6A = 6 * C6grid * (1 - poly) */
 844             f6A              = _mm_mul_pd(c6grid_00,_mm_sub_pd(one,poly));
 845             /* f6B = C6grid * exponent * beta^6 */
 846             f6B              = _mm_mul_pd(_mm_mul_pd(c6grid_00,one_sixth),_mm_mul_pd(exponent,ewclj6));
 847             /* fvdw = 12*C12/r13 - ((6*C6 - f6A)/r6 + f6B)/r */
 848             fvdw              = _mm_mul_pd(_mm_add_pd(_mm_mul_pd(_mm_sub_pd(_mm_mul_pd(c12_00,rinvsix),_mm_sub_pd(c6_00,f6A)),rinvsix),f6B),rinvsq00);
 849
 850             fscal            = _mm_add_pd(felec,fvdw);
 851
 852             /* Calculate temporary vectorial force */
 853             tx               = _mm_mul_pd(fscal,dx00);
 854             ty               = _mm_mul_pd(fscal,dy00);
 855             tz               = _mm_mul_pd(fscal,dz00);
 856
 857             /* Update vectorial force */
 858             fix0             = _mm_add_pd(fix0,tx);
 859             fiy0             = _mm_add_pd(fiy0,ty);
 860             fiz0             = _mm_add_pd(fiz0,tz);
 861
 862             fjx0             = _mm_add_pd(fjx0,tx);
 863             fjy0             = _mm_add_pd(fjy0,ty);
 864             fjz0             = _mm_add_pd(fjz0,tz);
 865
 866             /**************************
 867              * CALCULATE INTERACTIONS *
 868              **************************/
 869
 870             r10              = _mm_mul_pd(rsq10,rinv10);
 871
 872             /* Compute parameters for interactions between i and j atoms */
 873             qq10             = _mm_mul_pd(iq1,jq0);
 874
 875             /* EWALD ELECTROSTATICS */
 876
 877             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 878             ewrt             = _mm_mul_pd(r10,ewtabscale);
 879             ewitab           = _mm_cvttpd_epi32(ewrt);
 880             eweps            = _mm_sub_pd(ewrt,_mm_cvtepi32_pd(ewitab));
 881             gmx_mm_load_2pair_swizzle_pd(ewtab+gmx_mm_extract_epi32(ewitab,0),ewtab+gmx_mm_extract_epi32(ewitab,1),
 882                                          &ewtabF,&ewtabFn);
 883             felec            = _mm_add_pd(_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF),_mm_mul_pd(eweps,ewtabFn));
 884             felec            = _mm_mul_pd(_mm_mul_pd(qq10,rinv10),_mm_sub_pd(rinvsq10,felec));
 885
 886             fscal            = felec;
 887
 888             /* Calculate temporary vectorial force */
 889             tx               = _mm_mul_pd(fscal,dx10);
 890             ty               = _mm_mul_pd(fscal,dy10);
 891             tz               = _mm_mul_pd(fscal,dz10);
 892
 893             /* Update vectorial force */
 894             fix1             = _mm_add_pd(fix1,tx);
 895             fiy1             = _mm_add_pd(fiy1,ty);
 896             fiz1             = _mm_add_pd(fiz1,tz);
 897
 898             fjx0             = _mm_add_pd(fjx0,tx);
 899             fjy0             = _mm_add_pd(fjy0,ty);
 900             fjz0             = _mm_add_pd(fjz0,tz);
 901
 902             /**************************
 903              * CALCULATE INTERACTIONS *
 904              **************************/
 905
 906             r20              = _mm_mul_pd(rsq20,rinv20);
 907
 908             /* Compute parameters for interactions between i and j atoms */
 909             qq20             = _mm_mul_pd(iq2,jq0);
 910
 911             /* EWALD ELECTROSTATICS */
 912
 913             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 914             ewrt             = _mm_mul_pd(r20,ewtabscale);
 915             ewitab           = _mm_cvttpd_epi32(ewrt);
 916             eweps            = _mm_sub_pd(ewrt,_mm_cvtepi32_pd(ewitab));
 917             gmx_mm_load_2pair_swizzle_pd(ewtab+gmx_mm_extract_epi32(ewitab,0),ewtab+gmx_mm_extract_epi32(ewitab,1),
 918                                          &ewtabF,&ewtabFn);
 919             felec            = _mm_add_pd(_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF),_mm_mul_pd(eweps,ewtabFn));
 920             felec            = _mm_mul_pd(_mm_mul_pd(qq20,rinv20),_mm_sub_pd(rinvsq20,felec));
 921
 922             fscal            = felec;
 923
 924             /* Calculate temporary vectorial force */
 925             tx               = _mm_mul_pd(fscal,dx20);
 926             ty               = _mm_mul_pd(fscal,dy20);
 927             tz               = _mm_mul_pd(fscal,dz20);
 928
 929             /* Update vectorial force */
 930             fix2             = _mm_add_pd(fix2,tx);
 931             fiy2             = _mm_add_pd(fiy2,ty);
 932             fiz2             = _mm_add_pd(fiz2,tz);
 933
 934             fjx0             = _mm_add_pd(fjx0,tx);
 935             fjy0             = _mm_add_pd(fjy0,ty);
 936             fjz0             = _mm_add_pd(fjz0,tz);
 937
 938             gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,fjx0,fjy0,fjz0);
 939
 940             /* Inner loop uses 134 flops */
 941         }
 942
 943         if(jidx<j_index_end)
 944         {
 945
 946             jnrA             = jjnr[jidx];
 947             j_coord_offsetA  = DIM*jnrA;
 948
 949             /* load j atom coordinates */
 950             gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
 951                                               &jx0,&jy0,&jz0);
 952
 953             /* Calculate displacement vector */
 954             dx00             = _mm_sub_pd(ix0,jx0);
 955             dy00             = _mm_sub_pd(iy0,jy0);
 956             dz00             = _mm_sub_pd(iz0,jz0);
 957             dx10             = _mm_sub_pd(ix1,jx0);
 958             dy10             = _mm_sub_pd(iy1,jy0);
 959             dz10             = _mm_sub_pd(iz1,jz0);
 960             dx20             = _mm_sub_pd(ix2,jx0);
 961             dy20             = _mm_sub_pd(iy2,jy0);
 962             dz20             = _mm_sub_pd(iz2,jz0);
 963
 964             /* Calculate squared distance and things based on it */
 965             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 966             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
 967             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
 968
 969             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 970             rinv10           = gmx_mm_invsqrt_pd(rsq10);
 971             rinv20           = gmx_mm_invsqrt_pd(rsq20);
 972
 973             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 974             rinvsq10         = _mm_mul_pd(rinv10,rinv10);
 975             rinvsq20         = _mm_mul_pd(rinv20,rinv20);
 976
 977             /* Load parameters for j particles */
 978             jq0              = _mm_load_sd(charge+jnrA+0);
 979             vdwjidx0A        = 2*vdwtype[jnrA+0];
 980
 981             fjx0             = _mm_setzero_pd();
 982             fjy0             = _mm_setzero_pd();
 983             fjz0             = _mm_setzero_pd();
 984
 985             /**************************
 986              * CALCULATE INTERACTIONS *
 987              **************************/
 988
 989             r00              = _mm_mul_pd(rsq00,rinv00);
 990
 991             /* Compute parameters for interactions between i and j atoms */
 992             qq00             = _mm_mul_pd(iq0,jq0);
 993             gmx_mm_load_1pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,&c6_00,&c12_00);
 994
 995             c6grid_00       = gmx_mm_load_1real_pd(vdwgridparam+vdwioffset0+vdwjidx0A);
 996
 997             /* EWALD ELECTROSTATICS */
 998
 999             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1000             ewrt             = _mm_mul_pd(r00,ewtabscale);
1001             ewitab           = _mm_cvttpd_epi32(ewrt);
1002             eweps            = _mm_sub_pd(ewrt,_mm_cvtepi32_pd(ewitab));
1003             gmx_mm_load_1pair_swizzle_pd(ewtab+gmx_mm_extract_epi32(ewitab,0),&ewtabF,&ewtabFn);
1004             felec            = _mm_add_pd(_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF),_mm_mul_pd(eweps,ewtabFn));
1005             felec            = _mm_mul_pd(_mm_mul_pd(qq00,rinv00),_mm_sub_pd(rinvsq00,felec));
1006
1007             /* Analytical LJ-PME */
1008             rinvsix          = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
1009             ewcljrsq         = _mm_mul_pd(ewclj2,rsq00);
1010             ewclj6           = _mm_mul_pd(ewclj2,_mm_mul_pd(ewclj2,ewclj2));
1011             exponent         = gmx_simd_exp_d(ewcljrsq);
1012             /* poly = exp(-(beta*r)^2) * (1 + (beta*r)^2 + (beta*r)^4 /2) */
1013             poly             = _mm_mul_pd(exponent,_mm_add_pd(_mm_sub_pd(one,ewcljrsq),_mm_mul_pd(_mm_mul_pd(ewcljrsq,ewcljrsq),one_half)));
1014             /* f6A = 6 * C6grid * (1 - poly) */
1015             f6A              = _mm_mul_pd(c6grid_00,_mm_sub_pd(one,poly));
1016             /* f6B = C6grid * exponent * beta^6 */
1017             f6B              = _mm_mul_pd(_mm_mul_pd(c6grid_00,one_sixth),_mm_mul_pd(exponent,ewclj6));
1018             /* fvdw = 12*C12/r13 - ((6*C6 - f6A)/r6 + f6B)/r */
1019             fvdw              = _mm_mul_pd(_mm_add_pd(_mm_mul_pd(_mm_sub_pd(_mm_mul_pd(c12_00,rinvsix),_mm_sub_pd(c6_00,f6A)),rinvsix),f6B),rinvsq00);
1020
1021             fscal            = _mm_add_pd(felec,fvdw);
1022
1023             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1024
1025             /* Calculate temporary vectorial force */
1026             tx               = _mm_mul_pd(fscal,dx00);
1027             ty               = _mm_mul_pd(fscal,dy00);
1028             tz               = _mm_mul_pd(fscal,dz00);
1029
1030             /* Update vectorial force */
1031             fix0             = _mm_add_pd(fix0,tx);
1032             fiy0             = _mm_add_pd(fiy0,ty);
1033             fiz0             = _mm_add_pd(fiz0,tz);
1034
1035             fjx0             = _mm_add_pd(fjx0,tx);
1036             fjy0             = _mm_add_pd(fjy0,ty);
1037             fjz0             = _mm_add_pd(fjz0,tz);
1038
1039             /**************************
1040              * CALCULATE INTERACTIONS *
1041              **************************/
1042
1043             r10              = _mm_mul_pd(rsq10,rinv10);
1044
1045             /* Compute parameters for interactions between i and j atoms */
1046             qq10             = _mm_mul_pd(iq1,jq0);
1047
1048             /* EWALD ELECTROSTATICS */
1049
1050             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1051             ewrt             = _mm_mul_pd(r10,ewtabscale);
1052             ewitab           = _mm_cvttpd_epi32(ewrt);
1053             eweps            = _mm_sub_pd(ewrt,_mm_cvtepi32_pd(ewitab));
1054             gmx_mm_load_1pair_swizzle_pd(ewtab+gmx_mm_extract_epi32(ewitab,0),&ewtabF,&ewtabFn);
1055             felec            = _mm_add_pd(_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF),_mm_mul_pd(eweps,ewtabFn));
1056             felec            = _mm_mul_pd(_mm_mul_pd(qq10,rinv10),_mm_sub_pd(rinvsq10,felec));
1057
1058             fscal            = felec;
1059
1060             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1061
1062             /* Calculate temporary vectorial force */
1063             tx               = _mm_mul_pd(fscal,dx10);
1064             ty               = _mm_mul_pd(fscal,dy10);
1065             tz               = _mm_mul_pd(fscal,dz10);
1066
1067             /* Update vectorial force */
1068             fix1             = _mm_add_pd(fix1,tx);
1069             fiy1             = _mm_add_pd(fiy1,ty);
1070             fiz1             = _mm_add_pd(fiz1,tz);
1071
1072             fjx0             = _mm_add_pd(fjx0,tx);
1073             fjy0             = _mm_add_pd(fjy0,ty);
1074             fjz0             = _mm_add_pd(fjz0,tz);
1075
1076             /**************************
1077              * CALCULATE INTERACTIONS *
1078              **************************/
1079
1080             r20              = _mm_mul_pd(rsq20,rinv20);
1081
1082             /* Compute parameters for interactions between i and j atoms */
1083             qq20             = _mm_mul_pd(iq2,jq0);
1084
1085             /* EWALD ELECTROSTATICS */
1086
1087             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1088             ewrt             = _mm_mul_pd(r20,ewtabscale);
1089             ewitab           = _mm_cvttpd_epi32(ewrt);
1090             eweps            = _mm_sub_pd(ewrt,_mm_cvtepi32_pd(ewitab));
1091             gmx_mm_load_1pair_swizzle_pd(ewtab+gmx_mm_extract_epi32(ewitab,0),&ewtabF,&ewtabFn);
1092             felec            = _mm_add_pd(_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF),_mm_mul_pd(eweps,ewtabFn));
1093             felec            = _mm_mul_pd(_mm_mul_pd(qq20,rinv20),_mm_sub_pd(rinvsq20,felec));
1094
1095             fscal            = felec;
1096
1097             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1098
1099             /* Calculate temporary vectorial force */
1100             tx               = _mm_mul_pd(fscal,dx20);
1101             ty               = _mm_mul_pd(fscal,dy20);
1102             tz               = _mm_mul_pd(fscal,dz20);
1103
1104             /* Update vectorial force */
1105             fix2             = _mm_add_pd(fix2,tx);
1106             fiy2             = _mm_add_pd(fiy2,ty);
1107             fiz2             = _mm_add_pd(fiz2,tz);
1108
1109             fjx0             = _mm_add_pd(fjx0,tx);
1110             fjy0             = _mm_add_pd(fjy0,ty);
1111             fjz0             = _mm_add_pd(fjz0,tz);
1112
1113             gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,fjx0,fjy0,fjz0);
1114
1115             /* Inner loop uses 134 flops */
1116         }
1117
1118         /* End of innermost loop */
1119
1120         gmx_mm_update_iforce_3atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
1121                                               f+i_coord_offset,fshift+i_shift_offset);
1122
1123         /* Increment number of inner iterations */
1124         inneriter                  += j_index_end - j_index_start;
1125
1126         /* Outer loop uses 18 flops */
1127     }
1128
1129     /* Increment number of outer iterations */
1130     outeriter        += nri;
1131
1132     /* Update outer/inner flops */
1133
1134     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3_F,outeriter*18 + inneriter*134);
1135 }