src/gromacs/gmxlib/nonbonded/nb_kernel_avx_128_fma_double/nb_kernel_ElecEwSh_VdwLJEwSh_GeomW3P1_avx_128_fma_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 avx_128_fma_double kernel generator.
  37  */
  38 #ifdef HAVE_CONFIG_H
  39 #include <config.h>
  40 #endif
  41
  42 #include <math.h>
  43
  44 #include "../nb_kernel.h"
  45 #include "types/simple.h"
  46 #include "vec.h"
  47 #include "nrnb.h"
  48
  49 #include "gromacs/simd/math_x86_avx_128_fma_double.h"
  50 #include "kernelutil_x86_avx_128_fma_double.h"
  51
  52 /*
  53  * Gromacs nonbonded kernel:   nb_kernel_ElecEwSh_VdwLJEwSh_GeomW3P1_VF_avx_128_fma_double
  54  * Electrostatics interaction: Ewald
  55  * VdW interaction:            LJEwald
  56  * Geometry:                   Water3-Particle
  57  * Calculate force/pot:        PotentialAndForce
  58  */
  59 void
  60 nb_kernel_ElecEwSh_VdwLJEwSh_GeomW3P1_VF_avx_128_fma_double
  61                     (t_nblist                    * gmx_restrict       nlist,
  62                      rvec                        * gmx_restrict          xx,
  63                      rvec                        * gmx_restrict          ff,
  64                      t_forcerec                  * gmx_restrict          fr,
  65                      t_mdatoms                   * gmx_restrict     mdatoms,
  66                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
  67                      t_nrnb                      * gmx_restrict        nrnb)
  68 {
  69     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
  70      * just 0 for non-waters.
  71      * Suffixes A,B refer to j loop unrolling done with SSE double precision, e.g. for the two different
  72      * jnr indices corresponding to data put in the four positions in the SIMD register.
  73      */
  74     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
  75     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
  76     int              jnrA,jnrB;
  77     int              j_coord_offsetA,j_coord_offsetB;
  78     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
  79     real             rcutoff_scalar;
  80     real             *shiftvec,*fshift,*x,*f;
  81     __m128d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
  82     int              vdwioffset0;
  83     __m128d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
  84     int              vdwioffset1;
  85     __m128d          ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
  86     int              vdwioffset2;
  87     __m128d          ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
  88     int              vdwjidx0A,vdwjidx0B;
  89     __m128d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  90     __m128d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  91     __m128d          dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
  92     __m128d          dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
  93     __m128d          velec,felec,velecsum,facel,crf,krf,krf2;
  94     real             *charge;
  95     int              nvdwtype;
  96     __m128d          rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
  97     int              *vdwtype;
  98     real             *vdwparam;
  99     __m128d          one_sixth   = _mm_set1_pd(1.0/6.0);
 100     __m128d          one_twelfth = _mm_set1_pd(1.0/12.0);
 101     __m128d           c6grid_00;
 102     __m128d           c6grid_10;
 103     __m128d           c6grid_20;
 104     real             *vdwgridparam;
 105     __m128d           ewclj,ewclj2,ewclj6,ewcljrsq,poly,exponent,f6A,f6B,sh_lj_ewald;
 106     __m128d           one_half  = _mm_set1_pd(0.5);
 107     __m128d           minus_one = _mm_set1_pd(-1.0);
 108     __m128i          ewitab;
 109     __m128d          ewtabscale,eweps,twoeweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
 110     real             *ewtab;
 111     __m128d          dummy_mask,cutoff_mask;
 112     __m128d          signbit   = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
 113     __m128d          one     = _mm_set1_pd(1.0);
 114     __m128d          two     = _mm_set1_pd(2.0);
 115     x                = xx[0];
 116     f                = ff[0];
 117
 118     nri              = nlist->nri;
 119     iinr             = nlist->iinr;
 120     jindex           = nlist->jindex;
 121     jjnr             = nlist->jjnr;
 122     shiftidx         = nlist->shift;
 123     gid              = nlist->gid;
 124     shiftvec         = fr->shift_vec[0];
 125     fshift           = fr->fshift[0];
 126     facel            = _mm_set1_pd(fr->epsfac);
 127     charge           = mdatoms->chargeA;
 128     nvdwtype         = fr->ntype;
 129     vdwparam         = fr->nbfp;
 130     vdwtype          = mdatoms->typeA;
 131     vdwgridparam     = fr->ljpme_c6grid;
 132     sh_lj_ewald      = _mm_set1_pd(fr->ic->sh_lj_ewald);
 133     ewclj            = _mm_set1_pd(fr->ewaldcoeff_lj);
 134     ewclj2           = _mm_mul_pd(minus_one,_mm_mul_pd(ewclj,ewclj));
 135
 136     sh_ewald         = _mm_set1_pd(fr->ic->sh_ewald);
 137     ewtab            = fr->ic->tabq_coul_FDV0;
 138     ewtabscale       = _mm_set1_pd(fr->ic->tabq_scale);
 139     ewtabhalfspace   = _mm_set1_pd(0.5/fr->ic->tabq_scale);
 140
 141     /* Setup water-specific parameters */
 142     inr              = nlist->iinr[0];
 143     iq0              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+0]));
 144     iq1              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+1]));
 145     iq2              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+2]));
 146     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 147
 148     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 149     rcutoff_scalar   = fr->rcoulomb;
 150     rcutoff          = _mm_set1_pd(rcutoff_scalar);
 151     rcutoff2         = _mm_mul_pd(rcutoff,rcutoff);
 152
 153     sh_vdw_invrcut6  = _mm_set1_pd(fr->ic->sh_invrc6);
 154     rvdw             = _mm_set1_pd(fr->rvdw);
 155
 156     /* Avoid stupid compiler warnings */
 157     jnrA = jnrB = 0;
 158     j_coord_offsetA = 0;
 159     j_coord_offsetB = 0;
 160
 161     outeriter        = 0;
 162     inneriter        = 0;
 163
 164     /* Start outer loop over neighborlists */
 165     for(iidx=0; iidx<nri; iidx++)
 166     {
 167         /* Load shift vector for this list */
 168         i_shift_offset   = DIM*shiftidx[iidx];
 169
 170         /* Load limits for loop over neighbors */
 171         j_index_start    = jindex[iidx];
 172         j_index_end      = jindex[iidx+1];
 173
 174         /* Get outer coordinate index */
 175         inr              = iinr[iidx];
 176         i_coord_offset   = DIM*inr;
 177
 178         /* Load i particle coords and add shift vector */
 179         gmx_mm_load_shift_and_3rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
 180                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 181
 182         fix0             = _mm_setzero_pd();
 183         fiy0             = _mm_setzero_pd();
 184         fiz0             = _mm_setzero_pd();
 185         fix1             = _mm_setzero_pd();
 186         fiy1             = _mm_setzero_pd();
 187         fiz1             = _mm_setzero_pd();
 188         fix2             = _mm_setzero_pd();
 189         fiy2             = _mm_setzero_pd();
 190         fiz2             = _mm_setzero_pd();
 191
 192         /* Reset potential sums */
 193         velecsum         = _mm_setzero_pd();
 194         vvdwsum          = _mm_setzero_pd();
 195
 196         /* Start inner kernel loop */
 197         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 198         {
 199
 200             /* Get j neighbor index, and coordinate index */
 201             jnrA             = jjnr[jidx];
 202             jnrB             = jjnr[jidx+1];
 203             j_coord_offsetA  = DIM*jnrA;
 204             j_coord_offsetB  = DIM*jnrB;
 205
 206             /* load j atom coordinates */
 207             gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 208                                               &jx0,&jy0,&jz0);
 209
 210             /* Calculate displacement vector */
 211             dx00             = _mm_sub_pd(ix0,jx0);
 212             dy00             = _mm_sub_pd(iy0,jy0);
 213             dz00             = _mm_sub_pd(iz0,jz0);
 214             dx10             = _mm_sub_pd(ix1,jx0);
 215             dy10             = _mm_sub_pd(iy1,jy0);
 216             dz10             = _mm_sub_pd(iz1,jz0);
 217             dx20             = _mm_sub_pd(ix2,jx0);
 218             dy20             = _mm_sub_pd(iy2,jy0);
 219             dz20             = _mm_sub_pd(iz2,jz0);
 220
 221             /* Calculate squared distance and things based on it */
 222             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 223             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
 224             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
 225
 226             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 227             rinv10           = gmx_mm_invsqrt_pd(rsq10);
 228             rinv20           = gmx_mm_invsqrt_pd(rsq20);
 229
 230             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 231             rinvsq10         = _mm_mul_pd(rinv10,rinv10);
 232             rinvsq20         = _mm_mul_pd(rinv20,rinv20);
 233
 234             /* Load parameters for j particles */
 235             jq0              = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
 236             vdwjidx0A        = 2*vdwtype[jnrA+0];
 237             vdwjidx0B        = 2*vdwtype[jnrB+0];
 238
 239             fjx0             = _mm_setzero_pd();
 240             fjy0             = _mm_setzero_pd();
 241             fjz0             = _mm_setzero_pd();
 242
 243             /**************************
 244              * CALCULATE INTERACTIONS *
 245              **************************/
 246
 247             if (gmx_mm_any_lt(rsq00,rcutoff2))
 248             {
 249
 250             r00              = _mm_mul_pd(rsq00,rinv00);
 251
 252             /* Compute parameters for interactions between i and j atoms */
 253             qq00             = _mm_mul_pd(iq0,jq0);
 254             gmx_mm_load_2pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,
 255                                          vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
 256             c6grid_00       = gmx_mm_load_2real_swizzle_pd(vdwgridparam+vdwioffset0+vdwjidx0A,
 257                                                                vdwgridparam+vdwioffset0+vdwjidx0B);
 258
 259             /* EWALD ELECTROSTATICS */
 260
 261             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 262             ewrt             = _mm_mul_pd(r00,ewtabscale);
 263             ewitab           = _mm_cvttpd_epi32(ewrt);
 264 #ifdef __XOP__
 265             eweps            = _mm_frcz_pd(ewrt);
 266 #else
 267             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 268 #endif
 269             twoeweps         = _mm_add_pd(eweps,eweps);
 270             ewitab           = _mm_slli_epi32(ewitab,2);
 271             ewtabF           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 272             ewtabD           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
 273             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 274             ewtabV           = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
 275             ewtabFn          = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,1) +2);
 276             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 277             felec            = _mm_macc_pd(eweps,ewtabD,ewtabF);
 278             velec            = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
 279             velec            = _mm_mul_pd(qq00,_mm_sub_pd(_mm_sub_pd(rinv00,sh_ewald),velec));
 280             felec            = _mm_mul_pd(_mm_mul_pd(qq00,rinv00),_mm_sub_pd(rinvsq00,felec));
 281
 282             /* Analytical LJ-PME */
 283             rinvsix          = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 284             ewcljrsq         = _mm_mul_pd(ewclj2,rsq00);
 285             ewclj6           = _mm_mul_pd(ewclj2,_mm_mul_pd(ewclj2,ewclj2));
 286             exponent         = gmx_simd_exp_d(ewcljrsq);
 287             /* poly = exp(-(beta*r)^2) * (1 + (beta*r)^2 + (beta*r)^4 /2) */
 288             poly             = _mm_mul_pd(exponent,_mm_macc_pd(_mm_mul_pd(ewcljrsq,ewcljrsq),one_half,_mm_sub_pd(one,ewcljrsq)));
 289             /* vvdw6 = [C6 - C6grid * (1-poly)]/r6 */
 290             vvdw6            = _mm_mul_pd(_mm_macc_pd(-c6grid_00,_mm_sub_pd(one,poly),c6_00),rinvsix);
 291             vvdw12           = _mm_mul_pd(c12_00,_mm_mul_pd(rinvsix,rinvsix));
 292             vvdw             = _mm_msub_pd(_mm_nmacc_pd(c12_00,_mm_mul_pd(sh_vdw_invrcut6,sh_vdw_invrcut6),vvdw12),one_twelfth,
 293                                _mm_mul_pd(_mm_sub_pd(vvdw6,_mm_macc_pd(c6grid_00,sh_lj_ewald,_mm_mul_pd(c6_00,sh_vdw_invrcut6))),one_sixth));
 294             /* fvdw = vvdw12/r - (vvdw6/r + (C6grid * exponent * beta^6)/r) */
 295             fvdw             = _mm_mul_pd(_mm_add_pd(vvdw12,_mm_msub_pd(_mm_mul_pd(c6grid_00,one_sixth),_mm_mul_pd(exponent,ewclj6),vvdw6)),rinvsq00);
 296
 297             cutoff_mask      = _mm_cmplt_pd(rsq00,rcutoff2);
 298
 299             /* Update potential sum for this i atom from the interaction with this j atom. */
 300             velec            = _mm_and_pd(velec,cutoff_mask);
 301             velecsum         = _mm_add_pd(velecsum,velec);
 302             vvdw             = _mm_and_pd(vvdw,cutoff_mask);
 303             vvdwsum          = _mm_add_pd(vvdwsum,vvdw);
 304
 305             fscal            = _mm_add_pd(felec,fvdw);
 306
 307             fscal            = _mm_and_pd(fscal,cutoff_mask);
 308
 309             /* Update vectorial force */
 310             fix0             = _mm_macc_pd(dx00,fscal,fix0);
 311             fiy0             = _mm_macc_pd(dy00,fscal,fiy0);
 312             fiz0             = _mm_macc_pd(dz00,fscal,fiz0);
 313
 314             fjx0             = _mm_macc_pd(dx00,fscal,fjx0);
 315             fjy0             = _mm_macc_pd(dy00,fscal,fjy0);
 316             fjz0             = _mm_macc_pd(dz00,fscal,fjz0);
 317
 318             }
 319
 320             /**************************
 321              * CALCULATE INTERACTIONS *
 322              **************************/
 323
 324             if (gmx_mm_any_lt(rsq10,rcutoff2))
 325             {
 326
 327             r10              = _mm_mul_pd(rsq10,rinv10);
 328
 329             /* Compute parameters for interactions between i and j atoms */
 330             qq10             = _mm_mul_pd(iq1,jq0);
 331
 332             /* EWALD ELECTROSTATICS */
 333
 334             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 335             ewrt             = _mm_mul_pd(r10,ewtabscale);
 336             ewitab           = _mm_cvttpd_epi32(ewrt);
 337 #ifdef __XOP__
 338             eweps            = _mm_frcz_pd(ewrt);
 339 #else
 340             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 341 #endif
 342             twoeweps         = _mm_add_pd(eweps,eweps);
 343             ewitab           = _mm_slli_epi32(ewitab,2);
 344             ewtabF           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 345             ewtabD           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
 346             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 347             ewtabV           = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
 348             ewtabFn          = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,1) +2);
 349             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 350             felec            = _mm_macc_pd(eweps,ewtabD,ewtabF);
 351             velec            = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
 352             velec            = _mm_mul_pd(qq10,_mm_sub_pd(_mm_sub_pd(rinv10,sh_ewald),velec));
 353             felec            = _mm_mul_pd(_mm_mul_pd(qq10,rinv10),_mm_sub_pd(rinvsq10,felec));
 354
 355             cutoff_mask      = _mm_cmplt_pd(rsq10,rcutoff2);
 356
 357             /* Update potential sum for this i atom from the interaction with this j atom. */
 358             velec            = _mm_and_pd(velec,cutoff_mask);
 359             velecsum         = _mm_add_pd(velecsum,velec);
 360
 361             fscal            = felec;
 362
 363             fscal            = _mm_and_pd(fscal,cutoff_mask);
 364
 365             /* Update vectorial force */
 366             fix1             = _mm_macc_pd(dx10,fscal,fix1);
 367             fiy1             = _mm_macc_pd(dy10,fscal,fiy1);
 368             fiz1             = _mm_macc_pd(dz10,fscal,fiz1);
 369
 370             fjx0             = _mm_macc_pd(dx10,fscal,fjx0);
 371             fjy0             = _mm_macc_pd(dy10,fscal,fjy0);
 372             fjz0             = _mm_macc_pd(dz10,fscal,fjz0);
 373
 374             }
 375
 376             /**************************
 377              * CALCULATE INTERACTIONS *
 378              **************************/
 379
 380             if (gmx_mm_any_lt(rsq20,rcutoff2))
 381             {
 382
 383             r20              = _mm_mul_pd(rsq20,rinv20);
 384
 385             /* Compute parameters for interactions between i and j atoms */
 386             qq20             = _mm_mul_pd(iq2,jq0);
 387
 388             /* EWALD ELECTROSTATICS */
 389
 390             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 391             ewrt             = _mm_mul_pd(r20,ewtabscale);
 392             ewitab           = _mm_cvttpd_epi32(ewrt);
 393 #ifdef __XOP__
 394             eweps            = _mm_frcz_pd(ewrt);
 395 #else
 396             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 397 #endif
 398             twoeweps         = _mm_add_pd(eweps,eweps);
 399             ewitab           = _mm_slli_epi32(ewitab,2);
 400             ewtabF           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 401             ewtabD           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
 402             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 403             ewtabV           = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
 404             ewtabFn          = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,1) +2);
 405             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 406             felec            = _mm_macc_pd(eweps,ewtabD,ewtabF);
 407             velec            = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
 408             velec            = _mm_mul_pd(qq20,_mm_sub_pd(_mm_sub_pd(rinv20,sh_ewald),velec));
 409             felec            = _mm_mul_pd(_mm_mul_pd(qq20,rinv20),_mm_sub_pd(rinvsq20,felec));
 410
 411             cutoff_mask      = _mm_cmplt_pd(rsq20,rcutoff2);
 412
 413             /* Update potential sum for this i atom from the interaction with this j atom. */
 414             velec            = _mm_and_pd(velec,cutoff_mask);
 415             velecsum         = _mm_add_pd(velecsum,velec);
 416
 417             fscal            = felec;
 418
 419             fscal            = _mm_and_pd(fscal,cutoff_mask);
 420
 421             /* Update vectorial force */
 422             fix2             = _mm_macc_pd(dx20,fscal,fix2);
 423             fiy2             = _mm_macc_pd(dy20,fscal,fiy2);
 424             fiz2             = _mm_macc_pd(dz20,fscal,fiz2);
 425
 426             fjx0             = _mm_macc_pd(dx20,fscal,fjx0);
 427             fjy0             = _mm_macc_pd(dy20,fscal,fjy0);
 428             fjz0             = _mm_macc_pd(dz20,fscal,fjz0);
 429
 430             }
 431
 432             gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,fjx0,fjy0,fjz0);
 433
 434             /* Inner loop uses 179 flops */
 435         }
 436
 437         if(jidx<j_index_end)
 438         {
 439
 440             jnrA             = jjnr[jidx];
 441             j_coord_offsetA  = DIM*jnrA;
 442
 443             /* load j atom coordinates */
 444             gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
 445                                               &jx0,&jy0,&jz0);
 446
 447             /* Calculate displacement vector */
 448             dx00             = _mm_sub_pd(ix0,jx0);
 449             dy00             = _mm_sub_pd(iy0,jy0);
 450             dz00             = _mm_sub_pd(iz0,jz0);
 451             dx10             = _mm_sub_pd(ix1,jx0);
 452             dy10             = _mm_sub_pd(iy1,jy0);
 453             dz10             = _mm_sub_pd(iz1,jz0);
 454             dx20             = _mm_sub_pd(ix2,jx0);
 455             dy20             = _mm_sub_pd(iy2,jy0);
 456             dz20             = _mm_sub_pd(iz2,jz0);
 457
 458             /* Calculate squared distance and things based on it */
 459             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 460             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
 461             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
 462
 463             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 464             rinv10           = gmx_mm_invsqrt_pd(rsq10);
 465             rinv20           = gmx_mm_invsqrt_pd(rsq20);
 466
 467             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 468             rinvsq10         = _mm_mul_pd(rinv10,rinv10);
 469             rinvsq20         = _mm_mul_pd(rinv20,rinv20);
 470
 471             /* Load parameters for j particles */
 472             jq0              = _mm_load_sd(charge+jnrA+0);
 473             vdwjidx0A        = 2*vdwtype[jnrA+0];
 474
 475             fjx0             = _mm_setzero_pd();
 476             fjy0             = _mm_setzero_pd();
 477             fjz0             = _mm_setzero_pd();
 478
 479             /**************************
 480              * CALCULATE INTERACTIONS *
 481              **************************/
 482
 483             if (gmx_mm_any_lt(rsq00,rcutoff2))
 484             {
 485
 486             r00              = _mm_mul_pd(rsq00,rinv00);
 487
 488             /* Compute parameters for interactions between i and j atoms */
 489             qq00             = _mm_mul_pd(iq0,jq0);
 490             gmx_mm_load_1pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,&c6_00,&c12_00);
 491             c6grid_00       = gmx_mm_load_1real_pd(vdwgridparam+vdwioffset0+vdwjidx0A);
 492
 493             /* EWALD ELECTROSTATICS */
 494
 495             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 496             ewrt             = _mm_mul_pd(r00,ewtabscale);
 497             ewitab           = _mm_cvttpd_epi32(ewrt);
 498 #ifdef __XOP__
 499             eweps            = _mm_frcz_pd(ewrt);
 500 #else
 501             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 502 #endif
 503             twoeweps         = _mm_add_pd(eweps,eweps);
 504             ewitab           = _mm_slli_epi32(ewitab,2);
 505             ewtabF           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 506             ewtabD           = _mm_setzero_pd();
 507             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 508             ewtabV           = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
 509             ewtabFn          = _mm_setzero_pd();
 510             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 511             felec            = _mm_macc_pd(eweps,ewtabD,ewtabF);
 512             velec            = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
 513             velec            = _mm_mul_pd(qq00,_mm_sub_pd(_mm_sub_pd(rinv00,sh_ewald),velec));
 514             felec            = _mm_mul_pd(_mm_mul_pd(qq00,rinv00),_mm_sub_pd(rinvsq00,felec));
 515
 516             /* Analytical LJ-PME */
 517             rinvsix          = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 518             ewcljrsq         = _mm_mul_pd(ewclj2,rsq00);
 519             ewclj6           = _mm_mul_pd(ewclj2,_mm_mul_pd(ewclj2,ewclj2));
 520             exponent         = gmx_simd_exp_d(ewcljrsq);
 521             /* poly = exp(-(beta*r)^2) * (1 + (beta*r)^2 + (beta*r)^4 /2) */
 522             poly             = _mm_mul_pd(exponent,_mm_macc_pd(_mm_mul_pd(ewcljrsq,ewcljrsq),one_half,_mm_sub_pd(one,ewcljrsq)));
 523             /* vvdw6 = [C6 - C6grid * (1-poly)]/r6 */
 524             vvdw6            = _mm_mul_pd(_mm_macc_pd(-c6grid_00,_mm_sub_pd(one,poly),c6_00),rinvsix);
 525             vvdw12           = _mm_mul_pd(c12_00,_mm_mul_pd(rinvsix,rinvsix));
 526             vvdw             = _mm_msub_pd(_mm_nmacc_pd(c12_00,_mm_mul_pd(sh_vdw_invrcut6,sh_vdw_invrcut6),vvdw12),one_twelfth,
 527                                _mm_mul_pd(_mm_sub_pd(vvdw6,_mm_macc_pd(c6grid_00,sh_lj_ewald,_mm_mul_pd(c6_00,sh_vdw_invrcut6))),one_sixth));
 528             /* fvdw = vvdw12/r - (vvdw6/r + (C6grid * exponent * beta^6)/r) */
 529             fvdw             = _mm_mul_pd(_mm_add_pd(vvdw12,_mm_msub_pd(_mm_mul_pd(c6grid_00,one_sixth),_mm_mul_pd(exponent,ewclj6),vvdw6)),rinvsq00);
 530
 531             cutoff_mask      = _mm_cmplt_pd(rsq00,rcutoff2);
 532
 533             /* Update potential sum for this i atom from the interaction with this j atom. */
 534             velec            = _mm_and_pd(velec,cutoff_mask);
 535             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 536             velecsum         = _mm_add_pd(velecsum,velec);
 537             vvdw             = _mm_and_pd(vvdw,cutoff_mask);
 538             vvdw             = _mm_unpacklo_pd(vvdw,_mm_setzero_pd());
 539             vvdwsum          = _mm_add_pd(vvdwsum,vvdw);
 540
 541             fscal            = _mm_add_pd(felec,fvdw);
 542
 543             fscal            = _mm_and_pd(fscal,cutoff_mask);
 544
 545             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 546
 547             /* Update vectorial force */
 548             fix0             = _mm_macc_pd(dx00,fscal,fix0);
 549             fiy0             = _mm_macc_pd(dy00,fscal,fiy0);
 550             fiz0             = _mm_macc_pd(dz00,fscal,fiz0);
 551
 552             fjx0             = _mm_macc_pd(dx00,fscal,fjx0);
 553             fjy0             = _mm_macc_pd(dy00,fscal,fjy0);
 554             fjz0             = _mm_macc_pd(dz00,fscal,fjz0);
 555
 556             }
 557
 558             /**************************
 559              * CALCULATE INTERACTIONS *
 560              **************************/
 561
 562             if (gmx_mm_any_lt(rsq10,rcutoff2))
 563             {
 564
 565             r10              = _mm_mul_pd(rsq10,rinv10);
 566
 567             /* Compute parameters for interactions between i and j atoms */
 568             qq10             = _mm_mul_pd(iq1,jq0);
 569
 570             /* EWALD ELECTROSTATICS */
 571
 572             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 573             ewrt             = _mm_mul_pd(r10,ewtabscale);
 574             ewitab           = _mm_cvttpd_epi32(ewrt);
 575 #ifdef __XOP__
 576             eweps            = _mm_frcz_pd(ewrt);
 577 #else
 578             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 579 #endif
 580             twoeweps         = _mm_add_pd(eweps,eweps);
 581             ewitab           = _mm_slli_epi32(ewitab,2);
 582             ewtabF           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 583             ewtabD           = _mm_setzero_pd();
 584             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 585             ewtabV           = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
 586             ewtabFn          = _mm_setzero_pd();
 587             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 588             felec            = _mm_macc_pd(eweps,ewtabD,ewtabF);
 589             velec            = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
 590             velec            = _mm_mul_pd(qq10,_mm_sub_pd(_mm_sub_pd(rinv10,sh_ewald),velec));
 591             felec            = _mm_mul_pd(_mm_mul_pd(qq10,rinv10),_mm_sub_pd(rinvsq10,felec));
 592
 593             cutoff_mask      = _mm_cmplt_pd(rsq10,rcutoff2);
 594
 595             /* Update potential sum for this i atom from the interaction with this j atom. */
 596             velec            = _mm_and_pd(velec,cutoff_mask);
 597             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 598             velecsum         = _mm_add_pd(velecsum,velec);
 599
 600             fscal            = felec;
 601
 602             fscal            = _mm_and_pd(fscal,cutoff_mask);
 603
 604             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 605
 606             /* Update vectorial force */
 607             fix1             = _mm_macc_pd(dx10,fscal,fix1);
 608             fiy1             = _mm_macc_pd(dy10,fscal,fiy1);
 609             fiz1             = _mm_macc_pd(dz10,fscal,fiz1);
 610
 611             fjx0             = _mm_macc_pd(dx10,fscal,fjx0);
 612             fjy0             = _mm_macc_pd(dy10,fscal,fjy0);
 613             fjz0             = _mm_macc_pd(dz10,fscal,fjz0);
 614
 615             }
 616
 617             /**************************
 618              * CALCULATE INTERACTIONS *
 619              **************************/
 620
 621             if (gmx_mm_any_lt(rsq20,rcutoff2))
 622             {
 623
 624             r20              = _mm_mul_pd(rsq20,rinv20);
 625
 626             /* Compute parameters for interactions between i and j atoms */
 627             qq20             = _mm_mul_pd(iq2,jq0);
 628
 629             /* EWALD ELECTROSTATICS */
 630
 631             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 632             ewrt             = _mm_mul_pd(r20,ewtabscale);
 633             ewitab           = _mm_cvttpd_epi32(ewrt);
 634 #ifdef __XOP__
 635             eweps            = _mm_frcz_pd(ewrt);
 636 #else
 637             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 638 #endif
 639             twoeweps         = _mm_add_pd(eweps,eweps);
 640             ewitab           = _mm_slli_epi32(ewitab,2);
 641             ewtabF           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 642             ewtabD           = _mm_setzero_pd();
 643             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 644             ewtabV           = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
 645             ewtabFn          = _mm_setzero_pd();
 646             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 647             felec            = _mm_macc_pd(eweps,ewtabD,ewtabF);
 648             velec            = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
 649             velec            = _mm_mul_pd(qq20,_mm_sub_pd(_mm_sub_pd(rinv20,sh_ewald),velec));
 650             felec            = _mm_mul_pd(_mm_mul_pd(qq20,rinv20),_mm_sub_pd(rinvsq20,felec));
 651
 652             cutoff_mask      = _mm_cmplt_pd(rsq20,rcutoff2);
 653
 654             /* Update potential sum for this i atom from the interaction with this j atom. */
 655             velec            = _mm_and_pd(velec,cutoff_mask);
 656             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 657             velecsum         = _mm_add_pd(velecsum,velec);
 658
 659             fscal            = felec;
 660
 661             fscal            = _mm_and_pd(fscal,cutoff_mask);
 662
 663             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 664
 665             /* Update vectorial force */
 666             fix2             = _mm_macc_pd(dx20,fscal,fix2);
 667             fiy2             = _mm_macc_pd(dy20,fscal,fiy2);
 668             fiz2             = _mm_macc_pd(dz20,fscal,fiz2);
 669
 670             fjx0             = _mm_macc_pd(dx20,fscal,fjx0);
 671             fjy0             = _mm_macc_pd(dy20,fscal,fjy0);
 672             fjz0             = _mm_macc_pd(dz20,fscal,fjz0);
 673
 674             }
 675
 676             gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,fjx0,fjy0,fjz0);
 677
 678             /* Inner loop uses 179 flops */
 679         }
 680
 681         /* End of innermost loop */
 682
 683         gmx_mm_update_iforce_3atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
 684                                               f+i_coord_offset,fshift+i_shift_offset);
 685
 686         ggid                        = gid[iidx];
 687         /* Update potential energies */
 688         gmx_mm_update_1pot_pd(velecsum,kernel_data->energygrp_elec+ggid);
 689         gmx_mm_update_1pot_pd(vvdwsum,kernel_data->energygrp_vdw+ggid);
 690
 691         /* Increment number of inner iterations */
 692         inneriter                  += j_index_end - j_index_start;
 693
 694         /* Outer loop uses 20 flops */
 695     }
 696
 697     /* Increment number of outer iterations */
 698     outeriter        += nri;
 699
 700     /* Update outer/inner flops */
 701
 702     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3_VF,outeriter*20 + inneriter*179);
 703 }
 704 /*
 705  * Gromacs nonbonded kernel:   nb_kernel_ElecEwSh_VdwLJEwSh_GeomW3P1_F_avx_128_fma_double
 706  * Electrostatics interaction: Ewald
 707  * VdW interaction:            LJEwald
 708  * Geometry:                   Water3-Particle
 709  * Calculate force/pot:        Force
 710  */
 711 void
 712 nb_kernel_ElecEwSh_VdwLJEwSh_GeomW3P1_F_avx_128_fma_double
 713                     (t_nblist                    * gmx_restrict       nlist,
 714                      rvec                        * gmx_restrict          xx,
 715                      rvec                        * gmx_restrict          ff,
 716                      t_forcerec                  * gmx_restrict          fr,
 717                      t_mdatoms                   * gmx_restrict     mdatoms,
 718                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 719                      t_nrnb                      * gmx_restrict        nrnb)
 720 {
 721     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 722      * just 0 for non-waters.
 723      * Suffixes A,B refer to j loop unrolling done with SSE double precision, e.g. for the two different
 724      * jnr indices corresponding to data put in the four positions in the SIMD register.
 725      */
 726     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 727     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 728     int              jnrA,jnrB;
 729     int              j_coord_offsetA,j_coord_offsetB;
 730     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 731     real             rcutoff_scalar;
 732     real             *shiftvec,*fshift,*x,*f;
 733     __m128d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 734     int              vdwioffset0;
 735     __m128d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 736     int              vdwioffset1;
 737     __m128d          ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
 738     int              vdwioffset2;
 739     __m128d          ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
 740     int              vdwjidx0A,vdwjidx0B;
 741     __m128d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 742     __m128d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 743     __m128d          dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
 744     __m128d          dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
 745     __m128d          velec,felec,velecsum,facel,crf,krf,krf2;
 746     real             *charge;
 747     int              nvdwtype;
 748     __m128d          rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 749     int              *vdwtype;
 750     real             *vdwparam;
 751     __m128d          one_sixth   = _mm_set1_pd(1.0/6.0);
 752     __m128d          one_twelfth = _mm_set1_pd(1.0/12.0);
 753     __m128d           c6grid_00;
 754     __m128d           c6grid_10;
 755     __m128d           c6grid_20;
 756     real             *vdwgridparam;
 757     __m128d           ewclj,ewclj2,ewclj6,ewcljrsq,poly,exponent,f6A,f6B,sh_lj_ewald;
 758     __m128d           one_half  = _mm_set1_pd(0.5);
 759     __m128d           minus_one = _mm_set1_pd(-1.0);
 760     __m128i          ewitab;
 761     __m128d          ewtabscale,eweps,twoeweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
 762     real             *ewtab;
 763     __m128d          dummy_mask,cutoff_mask;
 764     __m128d          signbit   = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
 765     __m128d          one     = _mm_set1_pd(1.0);
 766     __m128d          two     = _mm_set1_pd(2.0);
 767     x                = xx[0];
 768     f                = ff[0];
 769
 770     nri              = nlist->nri;
 771     iinr             = nlist->iinr;
 772     jindex           = nlist->jindex;
 773     jjnr             = nlist->jjnr;
 774     shiftidx         = nlist->shift;
 775     gid              = nlist->gid;
 776     shiftvec         = fr->shift_vec[0];
 777     fshift           = fr->fshift[0];
 778     facel            = _mm_set1_pd(fr->epsfac);
 779     charge           = mdatoms->chargeA;
 780     nvdwtype         = fr->ntype;
 781     vdwparam         = fr->nbfp;
 782     vdwtype          = mdatoms->typeA;
 783     vdwgridparam     = fr->ljpme_c6grid;
 784     sh_lj_ewald      = _mm_set1_pd(fr->ic->sh_lj_ewald);
 785     ewclj            = _mm_set1_pd(fr->ewaldcoeff_lj);
 786     ewclj2           = _mm_mul_pd(minus_one,_mm_mul_pd(ewclj,ewclj));
 787
 788     sh_ewald         = _mm_set1_pd(fr->ic->sh_ewald);
 789     ewtab            = fr->ic->tabq_coul_F;
 790     ewtabscale       = _mm_set1_pd(fr->ic->tabq_scale);
 791     ewtabhalfspace   = _mm_set1_pd(0.5/fr->ic->tabq_scale);
 792
 793     /* Setup water-specific parameters */
 794     inr              = nlist->iinr[0];
 795     iq0              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+0]));
 796     iq1              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+1]));
 797     iq2              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+2]));
 798     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 799
 800     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 801     rcutoff_scalar   = fr->rcoulomb;
 802     rcutoff          = _mm_set1_pd(rcutoff_scalar);
 803     rcutoff2         = _mm_mul_pd(rcutoff,rcutoff);
 804
 805     sh_vdw_invrcut6  = _mm_set1_pd(fr->ic->sh_invrc6);
 806     rvdw             = _mm_set1_pd(fr->rvdw);
 807
 808     /* Avoid stupid compiler warnings */
 809     jnrA = jnrB = 0;
 810     j_coord_offsetA = 0;
 811     j_coord_offsetB = 0;
 812
 813     outeriter        = 0;
 814     inneriter        = 0;
 815
 816     /* Start outer loop over neighborlists */
 817     for(iidx=0; iidx<nri; iidx++)
 818     {
 819         /* Load shift vector for this list */
 820         i_shift_offset   = DIM*shiftidx[iidx];
 821
 822         /* Load limits for loop over neighbors */
 823         j_index_start    = jindex[iidx];
 824         j_index_end      = jindex[iidx+1];
 825
 826         /* Get outer coordinate index */
 827         inr              = iinr[iidx];
 828         i_coord_offset   = DIM*inr;
 829
 830         /* Load i particle coords and add shift vector */
 831         gmx_mm_load_shift_and_3rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
 832                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 833
 834         fix0             = _mm_setzero_pd();
 835         fiy0             = _mm_setzero_pd();
 836         fiz0             = _mm_setzero_pd();
 837         fix1             = _mm_setzero_pd();
 838         fiy1             = _mm_setzero_pd();
 839         fiz1             = _mm_setzero_pd();
 840         fix2             = _mm_setzero_pd();
 841         fiy2             = _mm_setzero_pd();
 842         fiz2             = _mm_setzero_pd();
 843
 844         /* Start inner kernel loop */
 845         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 846         {
 847
 848             /* Get j neighbor index, and coordinate index */
 849             jnrA             = jjnr[jidx];
 850             jnrB             = jjnr[jidx+1];
 851             j_coord_offsetA  = DIM*jnrA;
 852             j_coord_offsetB  = DIM*jnrB;
 853
 854             /* load j atom coordinates */
 855             gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 856                                               &jx0,&jy0,&jz0);
 857
 858             /* Calculate displacement vector */
 859             dx00             = _mm_sub_pd(ix0,jx0);
 860             dy00             = _mm_sub_pd(iy0,jy0);
 861             dz00             = _mm_sub_pd(iz0,jz0);
 862             dx10             = _mm_sub_pd(ix1,jx0);
 863             dy10             = _mm_sub_pd(iy1,jy0);
 864             dz10             = _mm_sub_pd(iz1,jz0);
 865             dx20             = _mm_sub_pd(ix2,jx0);
 866             dy20             = _mm_sub_pd(iy2,jy0);
 867             dz20             = _mm_sub_pd(iz2,jz0);
 868
 869             /* Calculate squared distance and things based on it */
 870             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 871             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
 872             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
 873
 874             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 875             rinv10           = gmx_mm_invsqrt_pd(rsq10);
 876             rinv20           = gmx_mm_invsqrt_pd(rsq20);
 877
 878             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 879             rinvsq10         = _mm_mul_pd(rinv10,rinv10);
 880             rinvsq20         = _mm_mul_pd(rinv20,rinv20);
 881
 882             /* Load parameters for j particles */
 883             jq0              = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
 884             vdwjidx0A        = 2*vdwtype[jnrA+0];
 885             vdwjidx0B        = 2*vdwtype[jnrB+0];
 886
 887             fjx0             = _mm_setzero_pd();
 888             fjy0             = _mm_setzero_pd();
 889             fjz0             = _mm_setzero_pd();
 890
 891             /**************************
 892              * CALCULATE INTERACTIONS *
 893              **************************/
 894
 895             if (gmx_mm_any_lt(rsq00,rcutoff2))
 896             {
 897
 898             r00              = _mm_mul_pd(rsq00,rinv00);
 899
 900             /* Compute parameters for interactions between i and j atoms */
 901             qq00             = _mm_mul_pd(iq0,jq0);
 902             gmx_mm_load_2pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,
 903                                          vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
 904             c6grid_00       = gmx_mm_load_2real_swizzle_pd(vdwgridparam+vdwioffset0+vdwjidx0A,
 905                                                                vdwgridparam+vdwioffset0+vdwjidx0B);
 906
 907             /* EWALD ELECTROSTATICS */
 908
 909             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 910             ewrt             = _mm_mul_pd(r00,ewtabscale);
 911             ewitab           = _mm_cvttpd_epi32(ewrt);
 912 #ifdef __XOP__
 913             eweps            = _mm_frcz_pd(ewrt);
 914 #else
 915             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 916 #endif
 917             twoeweps         = _mm_add_pd(eweps,eweps);
 918             gmx_mm_load_2pair_swizzle_pd(ewtab+_mm_extract_epi32(ewitab,0),ewtab+_mm_extract_epi32(ewitab,1),
 919                                          &ewtabF,&ewtabFn);
 920             felec            = _mm_macc_pd(eweps,ewtabFn,_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF));
 921             felec            = _mm_mul_pd(_mm_mul_pd(qq00,rinv00),_mm_sub_pd(rinvsq00,felec));
 922
 923             /* Analytical LJ-PME */
 924             rinvsix          = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 925             ewcljrsq         = _mm_mul_pd(ewclj2,rsq00);
 926             ewclj6           = _mm_mul_pd(ewclj2,_mm_mul_pd(ewclj2,ewclj2));
 927             exponent         = gmx_simd_exp_d(ewcljrsq);
 928             /* poly = exp(-(beta*r)^2) * (1 + (beta*r)^2 + (beta*r)^4 /2) */
 929             poly             = _mm_mul_pd(exponent,_mm_macc_pd(_mm_mul_pd(ewcljrsq,ewcljrsq),one_half,_mm_sub_pd(one,ewcljrsq)));
 930             /* f6A = 6 * C6grid * (1 - poly) */
 931             f6A              = _mm_mul_pd(c6grid_00,_mm_sub_pd(one,poly));
 932             /* f6B = C6grid * exponent * beta^6 */
 933             f6B              = _mm_mul_pd(_mm_mul_pd(c6grid_00,one_sixth),_mm_mul_pd(exponent,ewclj6));
 934             /* fvdw = 12*C12/r13 - ((6*C6 - f6A)/r6 + f6B)/r */
 935             fvdw              = _mm_mul_pd(_mm_macc_pd(_mm_msub_pd(c12_00,rinvsix,_mm_sub_pd(c6_00,f6A)),rinvsix,f6B),rinvsq00);
 936
 937             cutoff_mask      = _mm_cmplt_pd(rsq00,rcutoff2);
 938
 939             fscal            = _mm_add_pd(felec,fvdw);
 940
 941             fscal            = _mm_and_pd(fscal,cutoff_mask);
 942
 943             /* Update vectorial force */
 944             fix0             = _mm_macc_pd(dx00,fscal,fix0);
 945             fiy0             = _mm_macc_pd(dy00,fscal,fiy0);
 946             fiz0             = _mm_macc_pd(dz00,fscal,fiz0);
 947
 948             fjx0             = _mm_macc_pd(dx00,fscal,fjx0);
 949             fjy0             = _mm_macc_pd(dy00,fscal,fjy0);
 950             fjz0             = _mm_macc_pd(dz00,fscal,fjz0);
 951
 952             }
 953
 954             /**************************
 955              * CALCULATE INTERACTIONS *
 956              **************************/
 957
 958             if (gmx_mm_any_lt(rsq10,rcutoff2))
 959             {
 960
 961             r10              = _mm_mul_pd(rsq10,rinv10);
 962
 963             /* Compute parameters for interactions between i and j atoms */
 964             qq10             = _mm_mul_pd(iq1,jq0);
 965
 966             /* EWALD ELECTROSTATICS */
 967
 968             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 969             ewrt             = _mm_mul_pd(r10,ewtabscale);
 970             ewitab           = _mm_cvttpd_epi32(ewrt);
 971 #ifdef __XOP__
 972             eweps            = _mm_frcz_pd(ewrt);
 973 #else
 974             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 975 #endif
 976             twoeweps         = _mm_add_pd(eweps,eweps);
 977             gmx_mm_load_2pair_swizzle_pd(ewtab+_mm_extract_epi32(ewitab,0),ewtab+_mm_extract_epi32(ewitab,1),
 978                                          &ewtabF,&ewtabFn);
 979             felec            = _mm_macc_pd(eweps,ewtabFn,_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF));
 980             felec            = _mm_mul_pd(_mm_mul_pd(qq10,rinv10),_mm_sub_pd(rinvsq10,felec));
 981
 982             cutoff_mask      = _mm_cmplt_pd(rsq10,rcutoff2);
 983
 984             fscal            = felec;
 985
 986             fscal            = _mm_and_pd(fscal,cutoff_mask);
 987
 988             /* Update vectorial force */
 989             fix1             = _mm_macc_pd(dx10,fscal,fix1);
 990             fiy1             = _mm_macc_pd(dy10,fscal,fiy1);
 991             fiz1             = _mm_macc_pd(dz10,fscal,fiz1);
 992
 993             fjx0             = _mm_macc_pd(dx10,fscal,fjx0);
 994             fjy0             = _mm_macc_pd(dy10,fscal,fjy0);
 995             fjz0             = _mm_macc_pd(dz10,fscal,fjz0);
 996
 997             }
 998
 999             /**************************
1000              * CALCULATE INTERACTIONS *
1001              **************************/
1002
1003             if (gmx_mm_any_lt(rsq20,rcutoff2))
1004             {
1005
1006             r20              = _mm_mul_pd(rsq20,rinv20);
1007
1008             /* Compute parameters for interactions between i and j atoms */
1009             qq20             = _mm_mul_pd(iq2,jq0);
1010
1011             /* EWALD ELECTROSTATICS */
1012
1013             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1014             ewrt             = _mm_mul_pd(r20,ewtabscale);
1015             ewitab           = _mm_cvttpd_epi32(ewrt);
1016 #ifdef __XOP__
1017             eweps            = _mm_frcz_pd(ewrt);
1018 #else
1019             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
1020 #endif
1021             twoeweps         = _mm_add_pd(eweps,eweps);
1022             gmx_mm_load_2pair_swizzle_pd(ewtab+_mm_extract_epi32(ewitab,0),ewtab+_mm_extract_epi32(ewitab,1),
1023                                          &ewtabF,&ewtabFn);
1024             felec            = _mm_macc_pd(eweps,ewtabFn,_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF));
1025             felec            = _mm_mul_pd(_mm_mul_pd(qq20,rinv20),_mm_sub_pd(rinvsq20,felec));
1026
1027             cutoff_mask      = _mm_cmplt_pd(rsq20,rcutoff2);
1028
1029             fscal            = felec;
1030
1031             fscal            = _mm_and_pd(fscal,cutoff_mask);
1032
1033             /* Update vectorial force */
1034             fix2             = _mm_macc_pd(dx20,fscal,fix2);
1035             fiy2             = _mm_macc_pd(dy20,fscal,fiy2);
1036             fiz2             = _mm_macc_pd(dz20,fscal,fiz2);
1037
1038             fjx0             = _mm_macc_pd(dx20,fscal,fjx0);
1039             fjy0             = _mm_macc_pd(dy20,fscal,fjy0);
1040             fjz0             = _mm_macc_pd(dz20,fscal,fjz0);
1041
1042             }
1043
1044             gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,fjx0,fjy0,fjz0);
1045
1046             /* Inner loop uses 150 flops */
1047         }
1048
1049         if(jidx<j_index_end)
1050         {
1051
1052             jnrA             = jjnr[jidx];
1053             j_coord_offsetA  = DIM*jnrA;
1054
1055             /* load j atom coordinates */
1056             gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
1057                                               &jx0,&jy0,&jz0);
1058
1059             /* Calculate displacement vector */
1060             dx00             = _mm_sub_pd(ix0,jx0);
1061             dy00             = _mm_sub_pd(iy0,jy0);
1062             dz00             = _mm_sub_pd(iz0,jz0);
1063             dx10             = _mm_sub_pd(ix1,jx0);
1064             dy10             = _mm_sub_pd(iy1,jy0);
1065             dz10             = _mm_sub_pd(iz1,jz0);
1066             dx20             = _mm_sub_pd(ix2,jx0);
1067             dy20             = _mm_sub_pd(iy2,jy0);
1068             dz20             = _mm_sub_pd(iz2,jz0);
1069
1070             /* Calculate squared distance and things based on it */
1071             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
1072             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
1073             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
1074
1075             rinv00           = gmx_mm_invsqrt_pd(rsq00);
1076             rinv10           = gmx_mm_invsqrt_pd(rsq10);
1077             rinv20           = gmx_mm_invsqrt_pd(rsq20);
1078
1079             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
1080             rinvsq10         = _mm_mul_pd(rinv10,rinv10);
1081             rinvsq20         = _mm_mul_pd(rinv20,rinv20);
1082
1083             /* Load parameters for j particles */
1084             jq0              = _mm_load_sd(charge+jnrA+0);
1085             vdwjidx0A        = 2*vdwtype[jnrA+0];
1086
1087             fjx0             = _mm_setzero_pd();
1088             fjy0             = _mm_setzero_pd();
1089             fjz0             = _mm_setzero_pd();
1090
1091             /**************************
1092              * CALCULATE INTERACTIONS *
1093              **************************/
1094
1095             if (gmx_mm_any_lt(rsq00,rcutoff2))
1096             {
1097
1098             r00              = _mm_mul_pd(rsq00,rinv00);
1099
1100             /* Compute parameters for interactions between i and j atoms */
1101             qq00             = _mm_mul_pd(iq0,jq0);
1102             gmx_mm_load_1pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,&c6_00,&c12_00);
1103             c6grid_00       = gmx_mm_load_1real_pd(vdwgridparam+vdwioffset0+vdwjidx0A);
1104
1105             /* EWALD ELECTROSTATICS */
1106
1107             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1108             ewrt             = _mm_mul_pd(r00,ewtabscale);
1109             ewitab           = _mm_cvttpd_epi32(ewrt);
1110 #ifdef __XOP__
1111             eweps            = _mm_frcz_pd(ewrt);
1112 #else
1113             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
1114 #endif
1115             twoeweps         = _mm_add_pd(eweps,eweps);
1116             gmx_mm_load_1pair_swizzle_pd(ewtab+_mm_extract_epi32(ewitab,0),&ewtabF,&ewtabFn);
1117             felec            = _mm_macc_pd(eweps,ewtabFn,_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF));
1118             felec            = _mm_mul_pd(_mm_mul_pd(qq00,rinv00),_mm_sub_pd(rinvsq00,felec));
1119
1120             /* Analytical LJ-PME */
1121             rinvsix          = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
1122             ewcljrsq         = _mm_mul_pd(ewclj2,rsq00);
1123             ewclj6           = _mm_mul_pd(ewclj2,_mm_mul_pd(ewclj2,ewclj2));
1124             exponent         = gmx_simd_exp_d(ewcljrsq);
1125             /* poly = exp(-(beta*r)^2) * (1 + (beta*r)^2 + (beta*r)^4 /2) */
1126             poly             = _mm_mul_pd(exponent,_mm_macc_pd(_mm_mul_pd(ewcljrsq,ewcljrsq),one_half,_mm_sub_pd(one,ewcljrsq)));
1127             /* f6A = 6 * C6grid * (1 - poly) */
1128             f6A              = _mm_mul_pd(c6grid_00,_mm_sub_pd(one,poly));
1129             /* f6B = C6grid * exponent * beta^6 */
1130             f6B              = _mm_mul_pd(_mm_mul_pd(c6grid_00,one_sixth),_mm_mul_pd(exponent,ewclj6));
1131             /* fvdw = 12*C12/r13 - ((6*C6 - f6A)/r6 + f6B)/r */
1132             fvdw              = _mm_mul_pd(_mm_macc_pd(_mm_msub_pd(c12_00,rinvsix,_mm_sub_pd(c6_00,f6A)),rinvsix,f6B),rinvsq00);
1133
1134             cutoff_mask      = _mm_cmplt_pd(rsq00,rcutoff2);
1135
1136             fscal            = _mm_add_pd(felec,fvdw);
1137
1138             fscal            = _mm_and_pd(fscal,cutoff_mask);
1139
1140             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1141
1142             /* Update vectorial force */
1143             fix0             = _mm_macc_pd(dx00,fscal,fix0);
1144             fiy0             = _mm_macc_pd(dy00,fscal,fiy0);
1145             fiz0             = _mm_macc_pd(dz00,fscal,fiz0);
1146
1147             fjx0             = _mm_macc_pd(dx00,fscal,fjx0);
1148             fjy0             = _mm_macc_pd(dy00,fscal,fjy0);
1149             fjz0             = _mm_macc_pd(dz00,fscal,fjz0);
1150
1151             }
1152
1153             /**************************
1154              * CALCULATE INTERACTIONS *
1155              **************************/
1156
1157             if (gmx_mm_any_lt(rsq10,rcutoff2))
1158             {
1159
1160             r10              = _mm_mul_pd(rsq10,rinv10);
1161
1162             /* Compute parameters for interactions between i and j atoms */
1163             qq10             = _mm_mul_pd(iq1,jq0);
1164
1165             /* EWALD ELECTROSTATICS */
1166
1167             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1168             ewrt             = _mm_mul_pd(r10,ewtabscale);
1169             ewitab           = _mm_cvttpd_epi32(ewrt);
1170 #ifdef __XOP__
1171             eweps            = _mm_frcz_pd(ewrt);
1172 #else
1173             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
1174 #endif
1175             twoeweps         = _mm_add_pd(eweps,eweps);
1176             gmx_mm_load_1pair_swizzle_pd(ewtab+_mm_extract_epi32(ewitab,0),&ewtabF,&ewtabFn);
1177             felec            = _mm_macc_pd(eweps,ewtabFn,_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF));
1178             felec            = _mm_mul_pd(_mm_mul_pd(qq10,rinv10),_mm_sub_pd(rinvsq10,felec));
1179
1180             cutoff_mask      = _mm_cmplt_pd(rsq10,rcutoff2);
1181
1182             fscal            = felec;
1183
1184             fscal            = _mm_and_pd(fscal,cutoff_mask);
1185
1186             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1187
1188             /* Update vectorial force */
1189             fix1             = _mm_macc_pd(dx10,fscal,fix1);
1190             fiy1             = _mm_macc_pd(dy10,fscal,fiy1);
1191             fiz1             = _mm_macc_pd(dz10,fscal,fiz1);
1192
1193             fjx0             = _mm_macc_pd(dx10,fscal,fjx0);
1194             fjy0             = _mm_macc_pd(dy10,fscal,fjy0);
1195             fjz0             = _mm_macc_pd(dz10,fscal,fjz0);
1196
1197             }
1198
1199             /**************************
1200              * CALCULATE INTERACTIONS *
1201              **************************/
1202
1203             if (gmx_mm_any_lt(rsq20,rcutoff2))
1204             {
1205
1206             r20              = _mm_mul_pd(rsq20,rinv20);
1207
1208             /* Compute parameters for interactions between i and j atoms */
1209             qq20             = _mm_mul_pd(iq2,jq0);
1210
1211             /* EWALD ELECTROSTATICS */
1212
1213             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1214             ewrt             = _mm_mul_pd(r20,ewtabscale);
1215             ewitab           = _mm_cvttpd_epi32(ewrt);
1216 #ifdef __XOP__
1217             eweps            = _mm_frcz_pd(ewrt);
1218 #else
1219             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
1220 #endif
1221             twoeweps         = _mm_add_pd(eweps,eweps);
1222             gmx_mm_load_1pair_swizzle_pd(ewtab+_mm_extract_epi32(ewitab,0),&ewtabF,&ewtabFn);
1223             felec            = _mm_macc_pd(eweps,ewtabFn,_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF));
1224             felec            = _mm_mul_pd(_mm_mul_pd(qq20,rinv20),_mm_sub_pd(rinvsq20,felec));
1225
1226             cutoff_mask      = _mm_cmplt_pd(rsq20,rcutoff2);
1227
1228             fscal            = felec;
1229
1230             fscal            = _mm_and_pd(fscal,cutoff_mask);
1231
1232             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1233
1234             /* Update vectorial force */
1235             fix2             = _mm_macc_pd(dx20,fscal,fix2);
1236             fiy2             = _mm_macc_pd(dy20,fscal,fiy2);
1237             fiz2             = _mm_macc_pd(dz20,fscal,fiz2);
1238
1239             fjx0             = _mm_macc_pd(dx20,fscal,fjx0);
1240             fjy0             = _mm_macc_pd(dy20,fscal,fjy0);
1241             fjz0             = _mm_macc_pd(dz20,fscal,fjz0);
1242
1243             }
1244
1245             gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,fjx0,fjy0,fjz0);
1246
1247             /* Inner loop uses 150 flops */
1248         }
1249
1250         /* End of innermost loop */
1251
1252         gmx_mm_update_iforce_3atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
1253                                               f+i_coord_offset,fshift+i_shift_offset);
1254
1255         /* Increment number of inner iterations */
1256         inneriter                  += j_index_end - j_index_start;
1257
1258         /* Outer loop uses 18 flops */
1259     }
1260
1261     /* Increment number of outer iterations */
1262     outeriter        += nri;
1263
1264     /* Update outer/inner flops */
1265
1266     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3_F,outeriter*18 + inneriter*150);
1267 }