src/gromacs/gmxlib/nonbonded/nb_kernel_avx_128_fma_single/nb_kernel_ElecEwSh_VdwLJEwSh_GeomW3P1_avx_128_fma_single.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_single 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_avx_128_fma_single.h"
  48 #include "kernelutil_x86_avx_128_fma_single.h"
  49
  50 /*
  51  * Gromacs nonbonded kernel:   nb_kernel_ElecEwSh_VdwLJEwSh_GeomW3P1_VF_avx_128_fma_single
  52  * Electrostatics interaction: Ewald
  53  * VdW interaction:            LJEwald
  54  * Geometry:                   Water3-Particle
  55  * Calculate force/pot:        PotentialAndForce
  56  */
  57 void
  58 nb_kernel_ElecEwSh_VdwLJEwSh_GeomW3P1_VF_avx_128_fma_single
  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,C,D refer to j loop unrolling done with AVX_128, e.g. for the four 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,jnrC,jnrD;
  75     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
  76     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
  77     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
  78     real             rcutoff_scalar;
  79     real             *shiftvec,*fshift,*x,*f;
  80     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
  81     real             scratch[4*DIM];
  82     __m128           fscal,rcutoff,rcutoff2,jidxall;
  83     int              vdwioffset0;
  84     __m128           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
  85     int              vdwioffset1;
  86     __m128           ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
  87     int              vdwioffset2;
  88     __m128           ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
  89     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
  90     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  91     __m128           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  92     __m128           dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
  93     __m128           dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
  94     __m128           velec,felec,velecsum,facel,crf,krf,krf2;
  95     real             *charge;
  96     int              nvdwtype;
  97     __m128           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
  98     int              *vdwtype;
  99     real             *vdwparam;
 100     __m128           one_sixth   = _mm_set1_ps(1.0/6.0);
 101     __m128           one_twelfth = _mm_set1_ps(1.0/12.0);
 102     __m128           c6grid_00;
 103     __m128           c6grid_10;
 104     __m128           c6grid_20;
 105     real             *vdwgridparam;
 106     __m128           ewclj,ewclj2,ewclj6,ewcljrsq,poly,exponent,f6A,f6B,sh_lj_ewald;
 107     __m128           one_half = _mm_set1_ps(0.5);
 108     __m128           minus_one = _mm_set1_ps(-1.0);
 109     __m128i          ewitab;
 110     __m128           ewtabscale,eweps,twoeweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
 111     __m128           beta,beta2,beta3,zeta2,pmecorrF,pmecorrV,rinv3;
 112     real             *ewtab;
 113     __m128           dummy_mask,cutoff_mask;
 114     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
 115     __m128           one     = _mm_set1_ps(1.0);
 116     __m128           two     = _mm_set1_ps(2.0);
 117     x                = xx[0];
 118     f                = ff[0];
 119
 120     nri              = nlist->nri;
 121     iinr             = nlist->iinr;
 122     jindex           = nlist->jindex;
 123     jjnr             = nlist->jjnr;
 124     shiftidx         = nlist->shift;
 125     gid              = nlist->gid;
 126     shiftvec         = fr->shift_vec[0];
 127     fshift           = fr->fshift[0];
 128     facel            = _mm_set1_ps(fr->epsfac);
 129     charge           = mdatoms->chargeA;
 130     nvdwtype         = fr->ntype;
 131     vdwparam         = fr->nbfp;
 132     vdwtype          = mdatoms->typeA;
 133     vdwgridparam     = fr->ljpme_c6grid;
 134     sh_lj_ewald      = _mm_set1_ps(fr->ic->sh_lj_ewald);
 135     ewclj            = _mm_set1_ps(fr->ewaldcoeff_lj);
 136     ewclj2           = _mm_mul_ps(minus_one,_mm_mul_ps(ewclj,ewclj));
 137
 138     sh_ewald         = _mm_set1_ps(fr->ic->sh_ewald);
 139     beta             = _mm_set1_ps(fr->ic->ewaldcoeff_q);
 140     beta2            = _mm_mul_ps(beta,beta);
 141     beta3            = _mm_mul_ps(beta,beta2);
 142     ewtab            = fr->ic->tabq_coul_FDV0;
 143     ewtabscale       = _mm_set1_ps(fr->ic->tabq_scale);
 144     ewtabhalfspace   = _mm_set1_ps(0.5/fr->ic->tabq_scale);
 145
 146     /* Setup water-specific parameters */
 147     inr              = nlist->iinr[0];
 148     iq0              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+0]));
 149     iq1              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+1]));
 150     iq2              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+2]));
 151     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 152
 153     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 154     rcutoff_scalar   = fr->rcoulomb;
 155     rcutoff          = _mm_set1_ps(rcutoff_scalar);
 156     rcutoff2         = _mm_mul_ps(rcutoff,rcutoff);
 157
 158     sh_vdw_invrcut6  = _mm_set1_ps(fr->ic->sh_invrc6);
 159     rvdw             = _mm_set1_ps(fr->rvdw);
 160
 161     /* Avoid stupid compiler warnings */
 162     jnrA = jnrB = jnrC = jnrD = 0;
 163     j_coord_offsetA = 0;
 164     j_coord_offsetB = 0;
 165     j_coord_offsetC = 0;
 166     j_coord_offsetD = 0;
 167
 168     outeriter        = 0;
 169     inneriter        = 0;
 170
 171     for(iidx=0;iidx<4*DIM;iidx++)
 172     {
 173         scratch[iidx] = 0.0;
 174     }
 175
 176     /* Start outer loop over neighborlists */
 177     for(iidx=0; iidx<nri; iidx++)
 178     {
 179         /* Load shift vector for this list */
 180         i_shift_offset   = DIM*shiftidx[iidx];
 181
 182         /* Load limits for loop over neighbors */
 183         j_index_start    = jindex[iidx];
 184         j_index_end      = jindex[iidx+1];
 185
 186         /* Get outer coordinate index */
 187         inr              = iinr[iidx];
 188         i_coord_offset   = DIM*inr;
 189
 190         /* Load i particle coords and add shift vector */
 191         gmx_mm_load_shift_and_3rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
 192                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 193
 194         fix0             = _mm_setzero_ps();
 195         fiy0             = _mm_setzero_ps();
 196         fiz0             = _mm_setzero_ps();
 197         fix1             = _mm_setzero_ps();
 198         fiy1             = _mm_setzero_ps();
 199         fiz1             = _mm_setzero_ps();
 200         fix2             = _mm_setzero_ps();
 201         fiy2             = _mm_setzero_ps();
 202         fiz2             = _mm_setzero_ps();
 203
 204         /* Reset potential sums */
 205         velecsum         = _mm_setzero_ps();
 206         vvdwsum          = _mm_setzero_ps();
 207
 208         /* Start inner kernel loop */
 209         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 210         {
 211
 212             /* Get j neighbor index, and coordinate index */
 213             jnrA             = jjnr[jidx];
 214             jnrB             = jjnr[jidx+1];
 215             jnrC             = jjnr[jidx+2];
 216             jnrD             = jjnr[jidx+3];
 217             j_coord_offsetA  = DIM*jnrA;
 218             j_coord_offsetB  = DIM*jnrB;
 219             j_coord_offsetC  = DIM*jnrC;
 220             j_coord_offsetD  = DIM*jnrD;
 221
 222             /* load j atom coordinates */
 223             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 224                                               x+j_coord_offsetC,x+j_coord_offsetD,
 225                                               &jx0,&jy0,&jz0);
 226
 227             /* Calculate displacement vector */
 228             dx00             = _mm_sub_ps(ix0,jx0);
 229             dy00             = _mm_sub_ps(iy0,jy0);
 230             dz00             = _mm_sub_ps(iz0,jz0);
 231             dx10             = _mm_sub_ps(ix1,jx0);
 232             dy10             = _mm_sub_ps(iy1,jy0);
 233             dz10             = _mm_sub_ps(iz1,jz0);
 234             dx20             = _mm_sub_ps(ix2,jx0);
 235             dy20             = _mm_sub_ps(iy2,jy0);
 236             dz20             = _mm_sub_ps(iz2,jz0);
 237
 238             /* Calculate squared distance and things based on it */
 239             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 240             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 241             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 242
 243             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 244             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 245             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 246
 247             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 248             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 249             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 250
 251             /* Load parameters for j particles */
 252             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 253                                                               charge+jnrC+0,charge+jnrD+0);
 254             vdwjidx0A        = 2*vdwtype[jnrA+0];
 255             vdwjidx0B        = 2*vdwtype[jnrB+0];
 256             vdwjidx0C        = 2*vdwtype[jnrC+0];
 257             vdwjidx0D        = 2*vdwtype[jnrD+0];
 258
 259             fjx0             = _mm_setzero_ps();
 260             fjy0             = _mm_setzero_ps();
 261             fjz0             = _mm_setzero_ps();
 262
 263             /**************************
 264              * CALCULATE INTERACTIONS *
 265              **************************/
 266
 267             if (gmx_mm_any_lt(rsq00,rcutoff2))
 268             {
 269
 270             r00              = _mm_mul_ps(rsq00,rinv00);
 271
 272             /* Compute parameters for interactions between i and j atoms */
 273             qq00             = _mm_mul_ps(iq0,jq0);
 274             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 275                                          vdwparam+vdwioffset0+vdwjidx0B,
 276                                          vdwparam+vdwioffset0+vdwjidx0C,
 277                                          vdwparam+vdwioffset0+vdwjidx0D,
 278                                          &c6_00,&c12_00);
 279
 280             c6grid_00       = gmx_mm_load_4real_swizzle_ps(vdwgridparam+vdwioffset0+vdwjidx0A,
 281                                                                vdwgridparam+vdwioffset0+vdwjidx0B,
 282                                                                vdwgridparam+vdwioffset0+vdwjidx0C,
 283                                                                vdwgridparam+vdwioffset0+vdwjidx0D);
 284
 285             /* EWALD ELECTROSTATICS */
 286
 287             /* Analytical PME correction */
 288             zeta2            = _mm_mul_ps(beta2,rsq00);
 289             rinv3            = _mm_mul_ps(rinvsq00,rinv00);
 290             pmecorrF         = gmx_mm_pmecorrF_ps(zeta2);
 291             felec            = _mm_macc_ps(pmecorrF,beta3,rinv3);
 292             felec            = _mm_mul_ps(qq00,felec);
 293             pmecorrV         = gmx_mm_pmecorrV_ps(zeta2);
 294             velec            = _mm_nmacc_ps(pmecorrV,beta,_mm_sub_ps(rinv00,sh_ewald));
 295             velec            = _mm_mul_ps(qq00,velec);
 296
 297             /* Analytical LJ-PME */
 298             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 299             ewcljrsq         = _mm_mul_ps(ewclj2,rsq00);
 300             ewclj6           = _mm_mul_ps(ewclj2,_mm_mul_ps(ewclj2,ewclj2));
 301             exponent         = gmx_simd_exp_r(ewcljrsq);
 302             /* poly = exp(-(beta*r)^2) * (1 + (beta*r)^2 + (beta*r)^4 /2) */
 303             poly             = _mm_mul_ps(exponent,_mm_macc_ps(_mm_mul_ps(ewcljrsq,ewcljrsq),one_half,_mm_sub_ps(one,ewcljrsq)));
 304             /* vvdw6 = [C6 - C6grid * (1-poly)]/r6 */
 305             vvdw6            = _mm_mul_ps(_mm_macc_ps(-c6grid_00,_mm_sub_ps(one,poly),c6_00),rinvsix);
 306             vvdw12           = _mm_mul_ps(c12_00,_mm_mul_ps(rinvsix,rinvsix));
 307             vvdw             = _mm_msub_ps(_mm_nmacc_ps(c12_00,_mm_mul_ps(sh_vdw_invrcut6,sh_vdw_invrcut6),vvdw12),one_twelfth,
 308                                           _mm_mul_ps(_mm_sub_ps(vvdw6,_mm_macc_ps(c6grid_00,sh_lj_ewald,_mm_mul_ps(c6_00,sh_vdw_invrcut6))),one_sixth));
 309             /* fvdw = vvdw12/r - (vvdw6/r + (C6grid * exponent * beta^6)/r) */
 310             fvdw             = _mm_mul_ps(_mm_add_ps(vvdw12,_mm_msub_ps(_mm_mul_ps(c6grid_00,one_sixth),_mm_mul_ps(exponent,ewclj6),vvdw6)),rinvsq00);
 311
 312             cutoff_mask      = _mm_cmplt_ps(rsq00,rcutoff2);
 313
 314             /* Update potential sum for this i atom from the interaction with this j atom. */
 315             velec            = _mm_and_ps(velec,cutoff_mask);
 316             velecsum         = _mm_add_ps(velecsum,velec);
 317             vvdw             = _mm_and_ps(vvdw,cutoff_mask);
 318             vvdwsum          = _mm_add_ps(vvdwsum,vvdw);
 319
 320             fscal            = _mm_add_ps(felec,fvdw);
 321
 322             fscal            = _mm_and_ps(fscal,cutoff_mask);
 323
 324              /* Update vectorial force */
 325             fix0             = _mm_macc_ps(dx00,fscal,fix0);
 326             fiy0             = _mm_macc_ps(dy00,fscal,fiy0);
 327             fiz0             = _mm_macc_ps(dz00,fscal,fiz0);
 328
 329             fjx0             = _mm_macc_ps(dx00,fscal,fjx0);
 330             fjy0             = _mm_macc_ps(dy00,fscal,fjy0);
 331             fjz0             = _mm_macc_ps(dz00,fscal,fjz0);
 332
 333             }
 334
 335             /**************************
 336              * CALCULATE INTERACTIONS *
 337              **************************/
 338
 339             if (gmx_mm_any_lt(rsq10,rcutoff2))
 340             {
 341
 342             r10              = _mm_mul_ps(rsq10,rinv10);
 343
 344             /* Compute parameters for interactions between i and j atoms */
 345             qq10             = _mm_mul_ps(iq1,jq0);
 346
 347             /* EWALD ELECTROSTATICS */
 348
 349             /* Analytical PME correction */
 350             zeta2            = _mm_mul_ps(beta2,rsq10);
 351             rinv3            = _mm_mul_ps(rinvsq10,rinv10);
 352             pmecorrF         = gmx_mm_pmecorrF_ps(zeta2);
 353             felec            = _mm_macc_ps(pmecorrF,beta3,rinv3);
 354             felec            = _mm_mul_ps(qq10,felec);
 355             pmecorrV         = gmx_mm_pmecorrV_ps(zeta2);
 356             velec            = _mm_nmacc_ps(pmecorrV,beta,_mm_sub_ps(rinv10,sh_ewald));
 357             velec            = _mm_mul_ps(qq10,velec);
 358
 359             cutoff_mask      = _mm_cmplt_ps(rsq10,rcutoff2);
 360
 361             /* Update potential sum for this i atom from the interaction with this j atom. */
 362             velec            = _mm_and_ps(velec,cutoff_mask);
 363             velecsum         = _mm_add_ps(velecsum,velec);
 364
 365             fscal            = felec;
 366
 367             fscal            = _mm_and_ps(fscal,cutoff_mask);
 368
 369              /* Update vectorial force */
 370             fix1             = _mm_macc_ps(dx10,fscal,fix1);
 371             fiy1             = _mm_macc_ps(dy10,fscal,fiy1);
 372             fiz1             = _mm_macc_ps(dz10,fscal,fiz1);
 373
 374             fjx0             = _mm_macc_ps(dx10,fscal,fjx0);
 375             fjy0             = _mm_macc_ps(dy10,fscal,fjy0);
 376             fjz0             = _mm_macc_ps(dz10,fscal,fjz0);
 377
 378             }
 379
 380             /**************************
 381              * CALCULATE INTERACTIONS *
 382              **************************/
 383
 384             if (gmx_mm_any_lt(rsq20,rcutoff2))
 385             {
 386
 387             r20              = _mm_mul_ps(rsq20,rinv20);
 388
 389             /* Compute parameters for interactions between i and j atoms */
 390             qq20             = _mm_mul_ps(iq2,jq0);
 391
 392             /* EWALD ELECTROSTATICS */
 393
 394             /* Analytical PME correction */
 395             zeta2            = _mm_mul_ps(beta2,rsq20);
 396             rinv3            = _mm_mul_ps(rinvsq20,rinv20);
 397             pmecorrF         = gmx_mm_pmecorrF_ps(zeta2);
 398             felec            = _mm_macc_ps(pmecorrF,beta3,rinv3);
 399             felec            = _mm_mul_ps(qq20,felec);
 400             pmecorrV         = gmx_mm_pmecorrV_ps(zeta2);
 401             velec            = _mm_nmacc_ps(pmecorrV,beta,_mm_sub_ps(rinv20,sh_ewald));
 402             velec            = _mm_mul_ps(qq20,velec);
 403
 404             cutoff_mask      = _mm_cmplt_ps(rsq20,rcutoff2);
 405
 406             /* Update potential sum for this i atom from the interaction with this j atom. */
 407             velec            = _mm_and_ps(velec,cutoff_mask);
 408             velecsum         = _mm_add_ps(velecsum,velec);
 409
 410             fscal            = felec;
 411
 412             fscal            = _mm_and_ps(fscal,cutoff_mask);
 413
 414              /* Update vectorial force */
 415             fix2             = _mm_macc_ps(dx20,fscal,fix2);
 416             fiy2             = _mm_macc_ps(dy20,fscal,fiy2);
 417             fiz2             = _mm_macc_ps(dz20,fscal,fiz2);
 418
 419             fjx0             = _mm_macc_ps(dx20,fscal,fjx0);
 420             fjy0             = _mm_macc_ps(dy20,fscal,fjy0);
 421             fjz0             = _mm_macc_ps(dz20,fscal,fjz0);
 422
 423             }
 424
 425             fjptrA             = f+j_coord_offsetA;
 426             fjptrB             = f+j_coord_offsetB;
 427             fjptrC             = f+j_coord_offsetC;
 428             fjptrD             = f+j_coord_offsetD;
 429
 430             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 431
 432             /* Inner loop uses 129 flops */
 433         }
 434
 435         if(jidx<j_index_end)
 436         {
 437
 438             /* Get j neighbor index, and coordinate index */
 439             jnrlistA         = jjnr[jidx];
 440             jnrlistB         = jjnr[jidx+1];
 441             jnrlistC         = jjnr[jidx+2];
 442             jnrlistD         = jjnr[jidx+3];
 443             /* Sign of each element will be negative for non-real atoms.
 444              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 445              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 446              */
 447             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 448             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 449             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 450             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 451             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 452             j_coord_offsetA  = DIM*jnrA;
 453             j_coord_offsetB  = DIM*jnrB;
 454             j_coord_offsetC  = DIM*jnrC;
 455             j_coord_offsetD  = DIM*jnrD;
 456
 457             /* load j atom coordinates */
 458             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 459                                               x+j_coord_offsetC,x+j_coord_offsetD,
 460                                               &jx0,&jy0,&jz0);
 461
 462             /* Calculate displacement vector */
 463             dx00             = _mm_sub_ps(ix0,jx0);
 464             dy00             = _mm_sub_ps(iy0,jy0);
 465             dz00             = _mm_sub_ps(iz0,jz0);
 466             dx10             = _mm_sub_ps(ix1,jx0);
 467             dy10             = _mm_sub_ps(iy1,jy0);
 468             dz10             = _mm_sub_ps(iz1,jz0);
 469             dx20             = _mm_sub_ps(ix2,jx0);
 470             dy20             = _mm_sub_ps(iy2,jy0);
 471             dz20             = _mm_sub_ps(iz2,jz0);
 472
 473             /* Calculate squared distance and things based on it */
 474             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 475             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 476             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 477
 478             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 479             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 480             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 481
 482             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 483             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 484             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 485
 486             /* Load parameters for j particles */
 487             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 488                                                               charge+jnrC+0,charge+jnrD+0);
 489             vdwjidx0A        = 2*vdwtype[jnrA+0];
 490             vdwjidx0B        = 2*vdwtype[jnrB+0];
 491             vdwjidx0C        = 2*vdwtype[jnrC+0];
 492             vdwjidx0D        = 2*vdwtype[jnrD+0];
 493
 494             fjx0             = _mm_setzero_ps();
 495             fjy0             = _mm_setzero_ps();
 496             fjz0             = _mm_setzero_ps();
 497
 498             /**************************
 499              * CALCULATE INTERACTIONS *
 500              **************************/
 501
 502             if (gmx_mm_any_lt(rsq00,rcutoff2))
 503             {
 504
 505             r00              = _mm_mul_ps(rsq00,rinv00);
 506             r00              = _mm_andnot_ps(dummy_mask,r00);
 507
 508             /* Compute parameters for interactions between i and j atoms */
 509             qq00             = _mm_mul_ps(iq0,jq0);
 510             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 511                                          vdwparam+vdwioffset0+vdwjidx0B,
 512                                          vdwparam+vdwioffset0+vdwjidx0C,
 513                                          vdwparam+vdwioffset0+vdwjidx0D,
 514                                          &c6_00,&c12_00);
 515
 516             c6grid_00       = gmx_mm_load_4real_swizzle_ps(vdwgridparam+vdwioffset0+vdwjidx0A,
 517                                                                vdwgridparam+vdwioffset0+vdwjidx0B,
 518                                                                vdwgridparam+vdwioffset0+vdwjidx0C,
 519                                                                vdwgridparam+vdwioffset0+vdwjidx0D);
 520
 521             /* EWALD ELECTROSTATICS */
 522
 523             /* Analytical PME correction */
 524             zeta2            = _mm_mul_ps(beta2,rsq00);
 525             rinv3            = _mm_mul_ps(rinvsq00,rinv00);
 526             pmecorrF         = gmx_mm_pmecorrF_ps(zeta2);
 527             felec            = _mm_macc_ps(pmecorrF,beta3,rinv3);
 528             felec            = _mm_mul_ps(qq00,felec);
 529             pmecorrV         = gmx_mm_pmecorrV_ps(zeta2);
 530             velec            = _mm_nmacc_ps(pmecorrV,beta,_mm_sub_ps(rinv00,sh_ewald));
 531             velec            = _mm_mul_ps(qq00,velec);
 532
 533             /* Analytical LJ-PME */
 534             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 535             ewcljrsq         = _mm_mul_ps(ewclj2,rsq00);
 536             ewclj6           = _mm_mul_ps(ewclj2,_mm_mul_ps(ewclj2,ewclj2));
 537             exponent         = gmx_simd_exp_r(ewcljrsq);
 538             /* poly = exp(-(beta*r)^2) * (1 + (beta*r)^2 + (beta*r)^4 /2) */
 539             poly             = _mm_mul_ps(exponent,_mm_macc_ps(_mm_mul_ps(ewcljrsq,ewcljrsq),one_half,_mm_sub_ps(one,ewcljrsq)));
 540             /* vvdw6 = [C6 - C6grid * (1-poly)]/r6 */
 541             vvdw6            = _mm_mul_ps(_mm_macc_ps(-c6grid_00,_mm_sub_ps(one,poly),c6_00),rinvsix);
 542             vvdw12           = _mm_mul_ps(c12_00,_mm_mul_ps(rinvsix,rinvsix));
 543             vvdw             = _mm_msub_ps(_mm_nmacc_ps(c12_00,_mm_mul_ps(sh_vdw_invrcut6,sh_vdw_invrcut6),vvdw12),one_twelfth,
 544                                           _mm_mul_ps(_mm_sub_ps(vvdw6,_mm_macc_ps(c6grid_00,sh_lj_ewald,_mm_mul_ps(c6_00,sh_vdw_invrcut6))),one_sixth));
 545             /* fvdw = vvdw12/r - (vvdw6/r + (C6grid * exponent * beta^6)/r) */
 546             fvdw             = _mm_mul_ps(_mm_add_ps(vvdw12,_mm_msub_ps(_mm_mul_ps(c6grid_00,one_sixth),_mm_mul_ps(exponent,ewclj6),vvdw6)),rinvsq00);
 547
 548             cutoff_mask      = _mm_cmplt_ps(rsq00,rcutoff2);
 549
 550             /* Update potential sum for this i atom from the interaction with this j atom. */
 551             velec            = _mm_and_ps(velec,cutoff_mask);
 552             velec            = _mm_andnot_ps(dummy_mask,velec);
 553             velecsum         = _mm_add_ps(velecsum,velec);
 554             vvdw             = _mm_and_ps(vvdw,cutoff_mask);
 555             vvdw             = _mm_andnot_ps(dummy_mask,vvdw);
 556             vvdwsum          = _mm_add_ps(vvdwsum,vvdw);
 557
 558             fscal            = _mm_add_ps(felec,fvdw);
 559
 560             fscal            = _mm_and_ps(fscal,cutoff_mask);
 561
 562             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 563
 564              /* Update vectorial force */
 565             fix0             = _mm_macc_ps(dx00,fscal,fix0);
 566             fiy0             = _mm_macc_ps(dy00,fscal,fiy0);
 567             fiz0             = _mm_macc_ps(dz00,fscal,fiz0);
 568
 569             fjx0             = _mm_macc_ps(dx00,fscal,fjx0);
 570             fjy0             = _mm_macc_ps(dy00,fscal,fjy0);
 571             fjz0             = _mm_macc_ps(dz00,fscal,fjz0);
 572
 573             }
 574
 575             /**************************
 576              * CALCULATE INTERACTIONS *
 577              **************************/
 578
 579             if (gmx_mm_any_lt(rsq10,rcutoff2))
 580             {
 581
 582             r10              = _mm_mul_ps(rsq10,rinv10);
 583             r10              = _mm_andnot_ps(dummy_mask,r10);
 584
 585             /* Compute parameters for interactions between i and j atoms */
 586             qq10             = _mm_mul_ps(iq1,jq0);
 587
 588             /* EWALD ELECTROSTATICS */
 589
 590             /* Analytical PME correction */
 591             zeta2            = _mm_mul_ps(beta2,rsq10);
 592             rinv3            = _mm_mul_ps(rinvsq10,rinv10);
 593             pmecorrF         = gmx_mm_pmecorrF_ps(zeta2);
 594             felec            = _mm_macc_ps(pmecorrF,beta3,rinv3);
 595             felec            = _mm_mul_ps(qq10,felec);
 596             pmecorrV         = gmx_mm_pmecorrV_ps(zeta2);
 597             velec            = _mm_nmacc_ps(pmecorrV,beta,_mm_sub_ps(rinv10,sh_ewald));
 598             velec            = _mm_mul_ps(qq10,velec);
 599
 600             cutoff_mask      = _mm_cmplt_ps(rsq10,rcutoff2);
 601
 602             /* Update potential sum for this i atom from the interaction with this j atom. */
 603             velec            = _mm_and_ps(velec,cutoff_mask);
 604             velec            = _mm_andnot_ps(dummy_mask,velec);
 605             velecsum         = _mm_add_ps(velecsum,velec);
 606
 607             fscal            = felec;
 608
 609             fscal            = _mm_and_ps(fscal,cutoff_mask);
 610
 611             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 612
 613              /* Update vectorial force */
 614             fix1             = _mm_macc_ps(dx10,fscal,fix1);
 615             fiy1             = _mm_macc_ps(dy10,fscal,fiy1);
 616             fiz1             = _mm_macc_ps(dz10,fscal,fiz1);
 617
 618             fjx0             = _mm_macc_ps(dx10,fscal,fjx0);
 619             fjy0             = _mm_macc_ps(dy10,fscal,fjy0);
 620             fjz0             = _mm_macc_ps(dz10,fscal,fjz0);
 621
 622             }
 623
 624             /**************************
 625              * CALCULATE INTERACTIONS *
 626              **************************/
 627
 628             if (gmx_mm_any_lt(rsq20,rcutoff2))
 629             {
 630
 631             r20              = _mm_mul_ps(rsq20,rinv20);
 632             r20              = _mm_andnot_ps(dummy_mask,r20);
 633
 634             /* Compute parameters for interactions between i and j atoms */
 635             qq20             = _mm_mul_ps(iq2,jq0);
 636
 637             /* EWALD ELECTROSTATICS */
 638
 639             /* Analytical PME correction */
 640             zeta2            = _mm_mul_ps(beta2,rsq20);
 641             rinv3            = _mm_mul_ps(rinvsq20,rinv20);
 642             pmecorrF         = gmx_mm_pmecorrF_ps(zeta2);
 643             felec            = _mm_macc_ps(pmecorrF,beta3,rinv3);
 644             felec            = _mm_mul_ps(qq20,felec);
 645             pmecorrV         = gmx_mm_pmecorrV_ps(zeta2);
 646             velec            = _mm_nmacc_ps(pmecorrV,beta,_mm_sub_ps(rinv20,sh_ewald));
 647             velec            = _mm_mul_ps(qq20,velec);
 648
 649             cutoff_mask      = _mm_cmplt_ps(rsq20,rcutoff2);
 650
 651             /* Update potential sum for this i atom from the interaction with this j atom. */
 652             velec            = _mm_and_ps(velec,cutoff_mask);
 653             velec            = _mm_andnot_ps(dummy_mask,velec);
 654             velecsum         = _mm_add_ps(velecsum,velec);
 655
 656             fscal            = felec;
 657
 658             fscal            = _mm_and_ps(fscal,cutoff_mask);
 659
 660             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 661
 662              /* Update vectorial force */
 663             fix2             = _mm_macc_ps(dx20,fscal,fix2);
 664             fiy2             = _mm_macc_ps(dy20,fscal,fiy2);
 665             fiz2             = _mm_macc_ps(dz20,fscal,fiz2);
 666
 667             fjx0             = _mm_macc_ps(dx20,fscal,fjx0);
 668             fjy0             = _mm_macc_ps(dy20,fscal,fjy0);
 669             fjz0             = _mm_macc_ps(dz20,fscal,fjz0);
 670
 671             }
 672
 673             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 674             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 675             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 676             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 677
 678             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 679
 680             /* Inner loop uses 132 flops */
 681         }
 682
 683         /* End of innermost loop */
 684
 685         gmx_mm_update_iforce_3atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
 686                                               f+i_coord_offset,fshift+i_shift_offset);
 687
 688         ggid                        = gid[iidx];
 689         /* Update potential energies */
 690         gmx_mm_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid);
 691         gmx_mm_update_1pot_ps(vvdwsum,kernel_data->energygrp_vdw+ggid);
 692
 693         /* Increment number of inner iterations */
 694         inneriter                  += j_index_end - j_index_start;
 695
 696         /* Outer loop uses 20 flops */
 697     }
 698
 699     /* Increment number of outer iterations */
 700     outeriter        += nri;
 701
 702     /* Update outer/inner flops */
 703
 704     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3_VF,outeriter*20 + inneriter*132);
 705 }
 706 /*
 707  * Gromacs nonbonded kernel:   nb_kernel_ElecEwSh_VdwLJEwSh_GeomW3P1_F_avx_128_fma_single
 708  * Electrostatics interaction: Ewald
 709  * VdW interaction:            LJEwald
 710  * Geometry:                   Water3-Particle
 711  * Calculate force/pot:        Force
 712  */
 713 void
 714 nb_kernel_ElecEwSh_VdwLJEwSh_GeomW3P1_F_avx_128_fma_single
 715                     (t_nblist                    * gmx_restrict       nlist,
 716                      rvec                        * gmx_restrict          xx,
 717                      rvec                        * gmx_restrict          ff,
 718                      t_forcerec                  * gmx_restrict          fr,
 719                      t_mdatoms                   * gmx_restrict     mdatoms,
 720                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 721                      t_nrnb                      * gmx_restrict        nrnb)
 722 {
 723     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 724      * just 0 for non-waters.
 725      * Suffixes A,B,C,D refer to j loop unrolling done with AVX_128, e.g. for the four different
 726      * jnr indices corresponding to data put in the four positions in the SIMD register.
 727      */
 728     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 729     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 730     int              jnrA,jnrB,jnrC,jnrD;
 731     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
 732     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 733     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 734     real             rcutoff_scalar;
 735     real             *shiftvec,*fshift,*x,*f;
 736     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
 737     real             scratch[4*DIM];
 738     __m128           fscal,rcutoff,rcutoff2,jidxall;
 739     int              vdwioffset0;
 740     __m128           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 741     int              vdwioffset1;
 742     __m128           ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
 743     int              vdwioffset2;
 744     __m128           ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
 745     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
 746     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 747     __m128           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 748     __m128           dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
 749     __m128           dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
 750     __m128           velec,felec,velecsum,facel,crf,krf,krf2;
 751     real             *charge;
 752     int              nvdwtype;
 753     __m128           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 754     int              *vdwtype;
 755     real             *vdwparam;
 756     __m128           one_sixth   = _mm_set1_ps(1.0/6.0);
 757     __m128           one_twelfth = _mm_set1_ps(1.0/12.0);
 758     __m128           c6grid_00;
 759     __m128           c6grid_10;
 760     __m128           c6grid_20;
 761     real             *vdwgridparam;
 762     __m128           ewclj,ewclj2,ewclj6,ewcljrsq,poly,exponent,f6A,f6B,sh_lj_ewald;
 763     __m128           one_half = _mm_set1_ps(0.5);
 764     __m128           minus_one = _mm_set1_ps(-1.0);
 765     __m128i          ewitab;
 766     __m128           ewtabscale,eweps,twoeweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
 767     __m128           beta,beta2,beta3,zeta2,pmecorrF,pmecorrV,rinv3;
 768     real             *ewtab;
 769     __m128           dummy_mask,cutoff_mask;
 770     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
 771     __m128           one     = _mm_set1_ps(1.0);
 772     __m128           two     = _mm_set1_ps(2.0);
 773     x                = xx[0];
 774     f                = ff[0];
 775
 776     nri              = nlist->nri;
 777     iinr             = nlist->iinr;
 778     jindex           = nlist->jindex;
 779     jjnr             = nlist->jjnr;
 780     shiftidx         = nlist->shift;
 781     gid              = nlist->gid;
 782     shiftvec         = fr->shift_vec[0];
 783     fshift           = fr->fshift[0];
 784     facel            = _mm_set1_ps(fr->epsfac);
 785     charge           = mdatoms->chargeA;
 786     nvdwtype         = fr->ntype;
 787     vdwparam         = fr->nbfp;
 788     vdwtype          = mdatoms->typeA;
 789     vdwgridparam     = fr->ljpme_c6grid;
 790     sh_lj_ewald      = _mm_set1_ps(fr->ic->sh_lj_ewald);
 791     ewclj            = _mm_set1_ps(fr->ewaldcoeff_lj);
 792     ewclj2           = _mm_mul_ps(minus_one,_mm_mul_ps(ewclj,ewclj));
 793
 794     sh_ewald         = _mm_set1_ps(fr->ic->sh_ewald);
 795     beta             = _mm_set1_ps(fr->ic->ewaldcoeff_q);
 796     beta2            = _mm_mul_ps(beta,beta);
 797     beta3            = _mm_mul_ps(beta,beta2);
 798     ewtab            = fr->ic->tabq_coul_F;
 799     ewtabscale       = _mm_set1_ps(fr->ic->tabq_scale);
 800     ewtabhalfspace   = _mm_set1_ps(0.5/fr->ic->tabq_scale);
 801
 802     /* Setup water-specific parameters */
 803     inr              = nlist->iinr[0];
 804     iq0              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+0]));
 805     iq1              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+1]));
 806     iq2              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+2]));
 807     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 808
 809     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 810     rcutoff_scalar   = fr->rcoulomb;
 811     rcutoff          = _mm_set1_ps(rcutoff_scalar);
 812     rcutoff2         = _mm_mul_ps(rcutoff,rcutoff);
 813
 814     sh_vdw_invrcut6  = _mm_set1_ps(fr->ic->sh_invrc6);
 815     rvdw             = _mm_set1_ps(fr->rvdw);
 816
 817     /* Avoid stupid compiler warnings */
 818     jnrA = jnrB = jnrC = jnrD = 0;
 819     j_coord_offsetA = 0;
 820     j_coord_offsetB = 0;
 821     j_coord_offsetC = 0;
 822     j_coord_offsetD = 0;
 823
 824     outeriter        = 0;
 825     inneriter        = 0;
 826
 827     for(iidx=0;iidx<4*DIM;iidx++)
 828     {
 829         scratch[iidx] = 0.0;
 830     }
 831
 832     /* Start outer loop over neighborlists */
 833     for(iidx=0; iidx<nri; iidx++)
 834     {
 835         /* Load shift vector for this list */
 836         i_shift_offset   = DIM*shiftidx[iidx];
 837
 838         /* Load limits for loop over neighbors */
 839         j_index_start    = jindex[iidx];
 840         j_index_end      = jindex[iidx+1];
 841
 842         /* Get outer coordinate index */
 843         inr              = iinr[iidx];
 844         i_coord_offset   = DIM*inr;
 845
 846         /* Load i particle coords and add shift vector */
 847         gmx_mm_load_shift_and_3rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
 848                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 849
 850         fix0             = _mm_setzero_ps();
 851         fiy0             = _mm_setzero_ps();
 852         fiz0             = _mm_setzero_ps();
 853         fix1             = _mm_setzero_ps();
 854         fiy1             = _mm_setzero_ps();
 855         fiz1             = _mm_setzero_ps();
 856         fix2             = _mm_setzero_ps();
 857         fiy2             = _mm_setzero_ps();
 858         fiz2             = _mm_setzero_ps();
 859
 860         /* Start inner kernel loop */
 861         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 862         {
 863
 864             /* Get j neighbor index, and coordinate index */
 865             jnrA             = jjnr[jidx];
 866             jnrB             = jjnr[jidx+1];
 867             jnrC             = jjnr[jidx+2];
 868             jnrD             = jjnr[jidx+3];
 869             j_coord_offsetA  = DIM*jnrA;
 870             j_coord_offsetB  = DIM*jnrB;
 871             j_coord_offsetC  = DIM*jnrC;
 872             j_coord_offsetD  = DIM*jnrD;
 873
 874             /* load j atom coordinates */
 875             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 876                                               x+j_coord_offsetC,x+j_coord_offsetD,
 877                                               &jx0,&jy0,&jz0);
 878
 879             /* Calculate displacement vector */
 880             dx00             = _mm_sub_ps(ix0,jx0);
 881             dy00             = _mm_sub_ps(iy0,jy0);
 882             dz00             = _mm_sub_ps(iz0,jz0);
 883             dx10             = _mm_sub_ps(ix1,jx0);
 884             dy10             = _mm_sub_ps(iy1,jy0);
 885             dz10             = _mm_sub_ps(iz1,jz0);
 886             dx20             = _mm_sub_ps(ix2,jx0);
 887             dy20             = _mm_sub_ps(iy2,jy0);
 888             dz20             = _mm_sub_ps(iz2,jz0);
 889
 890             /* Calculate squared distance and things based on it */
 891             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 892             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 893             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 894
 895             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 896             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 897             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 898
 899             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 900             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 901             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 902
 903             /* Load parameters for j particles */
 904             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 905                                                               charge+jnrC+0,charge+jnrD+0);
 906             vdwjidx0A        = 2*vdwtype[jnrA+0];
 907             vdwjidx0B        = 2*vdwtype[jnrB+0];
 908             vdwjidx0C        = 2*vdwtype[jnrC+0];
 909             vdwjidx0D        = 2*vdwtype[jnrD+0];
 910
 911             fjx0             = _mm_setzero_ps();
 912             fjy0             = _mm_setzero_ps();
 913             fjz0             = _mm_setzero_ps();
 914
 915             /**************************
 916              * CALCULATE INTERACTIONS *
 917              **************************/
 918
 919             if (gmx_mm_any_lt(rsq00,rcutoff2))
 920             {
 921
 922             r00              = _mm_mul_ps(rsq00,rinv00);
 923
 924             /* Compute parameters for interactions between i and j atoms */
 925             qq00             = _mm_mul_ps(iq0,jq0);
 926             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 927                                          vdwparam+vdwioffset0+vdwjidx0B,
 928                                          vdwparam+vdwioffset0+vdwjidx0C,
 929                                          vdwparam+vdwioffset0+vdwjidx0D,
 930                                          &c6_00,&c12_00);
 931
 932             c6grid_00       = gmx_mm_load_4real_swizzle_ps(vdwgridparam+vdwioffset0+vdwjidx0A,
 933                                                                vdwgridparam+vdwioffset0+vdwjidx0B,
 934                                                                vdwgridparam+vdwioffset0+vdwjidx0C,
 935                                                                vdwgridparam+vdwioffset0+vdwjidx0D);
 936
 937             /* EWALD ELECTROSTATICS */
 938
 939             /* Analytical PME correction */
 940             zeta2            = _mm_mul_ps(beta2,rsq00);
 941             rinv3            = _mm_mul_ps(rinvsq00,rinv00);
 942             pmecorrF         = gmx_mm_pmecorrF_ps(zeta2);
 943             felec            = _mm_macc_ps(pmecorrF,beta3,rinv3);
 944             felec            = _mm_mul_ps(qq00,felec);
 945
 946             /* Analytical LJ-PME */
 947             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 948             ewcljrsq         = _mm_mul_ps(ewclj2,rsq00);
 949             ewclj6           = _mm_mul_ps(ewclj2,_mm_mul_ps(ewclj2,ewclj2));
 950             exponent         = gmx_simd_exp_r(ewcljrsq);
 951             /* poly = exp(-(beta*r)^2) * (1 + (beta*r)^2 + (beta*r)^4 /2) */
 952             poly             = _mm_mul_ps(exponent,_mm_macc_ps(_mm_mul_ps(ewcljrsq,ewcljrsq),one_half,_mm_sub_ps(one,ewcljrsq)));
 953             /* f6A = 6 * C6grid * (1 - poly) */
 954             f6A              = _mm_mul_ps(c6grid_00,_mm_sub_ps(one,poly));
 955             /* f6B = C6grid * exponent * beta^6 */
 956             f6B              = _mm_mul_ps(_mm_mul_ps(c6grid_00,one_sixth),_mm_mul_ps(exponent,ewclj6));
 957             /* fvdw = 12*C12/r13 - ((6*C6 - f6A)/r6 + f6B)/r */
 958             fvdw              = _mm_mul_ps(_mm_macc_ps(_mm_msub_ps(c12_00,rinvsix,_mm_sub_ps(c6_00,f6A)),rinvsix,f6B),rinvsq00);
 959
 960             cutoff_mask      = _mm_cmplt_ps(rsq00,rcutoff2);
 961
 962             fscal            = _mm_add_ps(felec,fvdw);
 963
 964             fscal            = _mm_and_ps(fscal,cutoff_mask);
 965
 966              /* Update vectorial force */
 967             fix0             = _mm_macc_ps(dx00,fscal,fix0);
 968             fiy0             = _mm_macc_ps(dy00,fscal,fiy0);
 969             fiz0             = _mm_macc_ps(dz00,fscal,fiz0);
 970
 971             fjx0             = _mm_macc_ps(dx00,fscal,fjx0);
 972             fjy0             = _mm_macc_ps(dy00,fscal,fjy0);
 973             fjz0             = _mm_macc_ps(dz00,fscal,fjz0);
 974
 975             }
 976
 977             /**************************
 978              * CALCULATE INTERACTIONS *
 979              **************************/
 980
 981             if (gmx_mm_any_lt(rsq10,rcutoff2))
 982             {
 983
 984             r10              = _mm_mul_ps(rsq10,rinv10);
 985
 986             /* Compute parameters for interactions between i and j atoms */
 987             qq10             = _mm_mul_ps(iq1,jq0);
 988
 989             /* EWALD ELECTROSTATICS */
 990
 991             /* Analytical PME correction */
 992             zeta2            = _mm_mul_ps(beta2,rsq10);
 993             rinv3            = _mm_mul_ps(rinvsq10,rinv10);
 994             pmecorrF         = gmx_mm_pmecorrF_ps(zeta2);
 995             felec            = _mm_macc_ps(pmecorrF,beta3,rinv3);
 996             felec            = _mm_mul_ps(qq10,felec);
 997
 998             cutoff_mask      = _mm_cmplt_ps(rsq10,rcutoff2);
 999
1000             fscal            = felec;
1001
1002             fscal            = _mm_and_ps(fscal,cutoff_mask);
1003
1004              /* Update vectorial force */
1005             fix1             = _mm_macc_ps(dx10,fscal,fix1);
1006             fiy1             = _mm_macc_ps(dy10,fscal,fiy1);
1007             fiz1             = _mm_macc_ps(dz10,fscal,fiz1);
1008
1009             fjx0             = _mm_macc_ps(dx10,fscal,fjx0);
1010             fjy0             = _mm_macc_ps(dy10,fscal,fjy0);
1011             fjz0             = _mm_macc_ps(dz10,fscal,fjz0);
1012
1013             }
1014
1015             /**************************
1016              * CALCULATE INTERACTIONS *
1017              **************************/
1018
1019             if (gmx_mm_any_lt(rsq20,rcutoff2))
1020             {
1021
1022             r20              = _mm_mul_ps(rsq20,rinv20);
1023
1024             /* Compute parameters for interactions between i and j atoms */
1025             qq20             = _mm_mul_ps(iq2,jq0);
1026
1027             /* EWALD ELECTROSTATICS */
1028
1029             /* Analytical PME correction */
1030             zeta2            = _mm_mul_ps(beta2,rsq20);
1031             rinv3            = _mm_mul_ps(rinvsq20,rinv20);
1032             pmecorrF         = gmx_mm_pmecorrF_ps(zeta2);
1033             felec            = _mm_macc_ps(pmecorrF,beta3,rinv3);
1034             felec            = _mm_mul_ps(qq20,felec);
1035
1036             cutoff_mask      = _mm_cmplt_ps(rsq20,rcutoff2);
1037
1038             fscal            = felec;
1039
1040             fscal            = _mm_and_ps(fscal,cutoff_mask);
1041
1042              /* Update vectorial force */
1043             fix2             = _mm_macc_ps(dx20,fscal,fix2);
1044             fiy2             = _mm_macc_ps(dy20,fscal,fiy2);
1045             fiz2             = _mm_macc_ps(dz20,fscal,fiz2);
1046
1047             fjx0             = _mm_macc_ps(dx20,fscal,fjx0);
1048             fjy0             = _mm_macc_ps(dy20,fscal,fjy0);
1049             fjz0             = _mm_macc_ps(dz20,fscal,fjz0);
1050
1051             }
1052
1053             fjptrA             = f+j_coord_offsetA;
1054             fjptrB             = f+j_coord_offsetB;
1055             fjptrC             = f+j_coord_offsetC;
1056             fjptrD             = f+j_coord_offsetD;
1057
1058             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
1059
1060             /* Inner loop uses 114 flops */
1061         }
1062
1063         if(jidx<j_index_end)
1064         {
1065
1066             /* Get j neighbor index, and coordinate index */
1067             jnrlistA         = jjnr[jidx];
1068             jnrlistB         = jjnr[jidx+1];
1069             jnrlistC         = jjnr[jidx+2];
1070             jnrlistD         = jjnr[jidx+3];
1071             /* Sign of each element will be negative for non-real atoms.
1072              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
1073              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
1074              */
1075             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
1076             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
1077             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
1078             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
1079             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
1080             j_coord_offsetA  = DIM*jnrA;
1081             j_coord_offsetB  = DIM*jnrB;
1082             j_coord_offsetC  = DIM*jnrC;
1083             j_coord_offsetD  = DIM*jnrD;
1084
1085             /* load j atom coordinates */
1086             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
1087                                               x+j_coord_offsetC,x+j_coord_offsetD,
1088                                               &jx0,&jy0,&jz0);
1089
1090             /* Calculate displacement vector */
1091             dx00             = _mm_sub_ps(ix0,jx0);
1092             dy00             = _mm_sub_ps(iy0,jy0);
1093             dz00             = _mm_sub_ps(iz0,jz0);
1094             dx10             = _mm_sub_ps(ix1,jx0);
1095             dy10             = _mm_sub_ps(iy1,jy0);
1096             dz10             = _mm_sub_ps(iz1,jz0);
1097             dx20             = _mm_sub_ps(ix2,jx0);
1098             dy20             = _mm_sub_ps(iy2,jy0);
1099             dz20             = _mm_sub_ps(iz2,jz0);
1100
1101             /* Calculate squared distance and things based on it */
1102             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
1103             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
1104             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
1105
1106             rinv00           = gmx_mm_invsqrt_ps(rsq00);
1107             rinv10           = gmx_mm_invsqrt_ps(rsq10);
1108             rinv20           = gmx_mm_invsqrt_ps(rsq20);
1109
1110             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
1111             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
1112             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
1113
1114             /* Load parameters for j particles */
1115             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
1116                                                               charge+jnrC+0,charge+jnrD+0);
1117             vdwjidx0A        = 2*vdwtype[jnrA+0];
1118             vdwjidx0B        = 2*vdwtype[jnrB+0];
1119             vdwjidx0C        = 2*vdwtype[jnrC+0];
1120             vdwjidx0D        = 2*vdwtype[jnrD+0];
1121
1122             fjx0             = _mm_setzero_ps();
1123             fjy0             = _mm_setzero_ps();
1124             fjz0             = _mm_setzero_ps();
1125
1126             /**************************
1127              * CALCULATE INTERACTIONS *
1128              **************************/
1129
1130             if (gmx_mm_any_lt(rsq00,rcutoff2))
1131             {
1132
1133             r00              = _mm_mul_ps(rsq00,rinv00);
1134             r00              = _mm_andnot_ps(dummy_mask,r00);
1135
1136             /* Compute parameters for interactions between i and j atoms */
1137             qq00             = _mm_mul_ps(iq0,jq0);
1138             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
1139                                          vdwparam+vdwioffset0+vdwjidx0B,
1140                                          vdwparam+vdwioffset0+vdwjidx0C,
1141                                          vdwparam+vdwioffset0+vdwjidx0D,
1142                                          &c6_00,&c12_00);
1143
1144             c6grid_00       = gmx_mm_load_4real_swizzle_ps(vdwgridparam+vdwioffset0+vdwjidx0A,
1145                                                                vdwgridparam+vdwioffset0+vdwjidx0B,
1146                                                                vdwgridparam+vdwioffset0+vdwjidx0C,
1147                                                                vdwgridparam+vdwioffset0+vdwjidx0D);
1148
1149             /* EWALD ELECTROSTATICS */
1150
1151             /* Analytical PME correction */
1152             zeta2            = _mm_mul_ps(beta2,rsq00);
1153             rinv3            = _mm_mul_ps(rinvsq00,rinv00);
1154             pmecorrF         = gmx_mm_pmecorrF_ps(zeta2);
1155             felec            = _mm_macc_ps(pmecorrF,beta3,rinv3);
1156             felec            = _mm_mul_ps(qq00,felec);
1157
1158             /* Analytical LJ-PME */
1159             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
1160             ewcljrsq         = _mm_mul_ps(ewclj2,rsq00);
1161             ewclj6           = _mm_mul_ps(ewclj2,_mm_mul_ps(ewclj2,ewclj2));
1162             exponent         = gmx_simd_exp_r(ewcljrsq);
1163             /* poly = exp(-(beta*r)^2) * (1 + (beta*r)^2 + (beta*r)^4 /2) */
1164             poly             = _mm_mul_ps(exponent,_mm_macc_ps(_mm_mul_ps(ewcljrsq,ewcljrsq),one_half,_mm_sub_ps(one,ewcljrsq)));
1165             /* f6A = 6 * C6grid * (1 - poly) */
1166             f6A              = _mm_mul_ps(c6grid_00,_mm_sub_ps(one,poly));
1167             /* f6B = C6grid * exponent * beta^6 */
1168             f6B              = _mm_mul_ps(_mm_mul_ps(c6grid_00,one_sixth),_mm_mul_ps(exponent,ewclj6));
1169             /* fvdw = 12*C12/r13 - ((6*C6 - f6A)/r6 + f6B)/r */
1170             fvdw              = _mm_mul_ps(_mm_macc_ps(_mm_msub_ps(c12_00,rinvsix,_mm_sub_ps(c6_00,f6A)),rinvsix,f6B),rinvsq00);
1171
1172             cutoff_mask      = _mm_cmplt_ps(rsq00,rcutoff2);
1173
1174             fscal            = _mm_add_ps(felec,fvdw);
1175
1176             fscal            = _mm_and_ps(fscal,cutoff_mask);
1177
1178             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1179
1180              /* Update vectorial force */
1181             fix0             = _mm_macc_ps(dx00,fscal,fix0);
1182             fiy0             = _mm_macc_ps(dy00,fscal,fiy0);
1183             fiz0             = _mm_macc_ps(dz00,fscal,fiz0);
1184
1185             fjx0             = _mm_macc_ps(dx00,fscal,fjx0);
1186             fjy0             = _mm_macc_ps(dy00,fscal,fjy0);
1187             fjz0             = _mm_macc_ps(dz00,fscal,fjz0);
1188
1189             }
1190
1191             /**************************
1192              * CALCULATE INTERACTIONS *
1193              **************************/
1194
1195             if (gmx_mm_any_lt(rsq10,rcutoff2))
1196             {
1197
1198             r10              = _mm_mul_ps(rsq10,rinv10);
1199             r10              = _mm_andnot_ps(dummy_mask,r10);
1200
1201             /* Compute parameters for interactions between i and j atoms */
1202             qq10             = _mm_mul_ps(iq1,jq0);
1203
1204             /* EWALD ELECTROSTATICS */
1205
1206             /* Analytical PME correction */
1207             zeta2            = _mm_mul_ps(beta2,rsq10);
1208             rinv3            = _mm_mul_ps(rinvsq10,rinv10);
1209             pmecorrF         = gmx_mm_pmecorrF_ps(zeta2);
1210             felec            = _mm_macc_ps(pmecorrF,beta3,rinv3);
1211             felec            = _mm_mul_ps(qq10,felec);
1212
1213             cutoff_mask      = _mm_cmplt_ps(rsq10,rcutoff2);
1214
1215             fscal            = felec;
1216
1217             fscal            = _mm_and_ps(fscal,cutoff_mask);
1218
1219             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1220
1221              /* Update vectorial force */
1222             fix1             = _mm_macc_ps(dx10,fscal,fix1);
1223             fiy1             = _mm_macc_ps(dy10,fscal,fiy1);
1224             fiz1             = _mm_macc_ps(dz10,fscal,fiz1);
1225
1226             fjx0             = _mm_macc_ps(dx10,fscal,fjx0);
1227             fjy0             = _mm_macc_ps(dy10,fscal,fjy0);
1228             fjz0             = _mm_macc_ps(dz10,fscal,fjz0);
1229
1230             }
1231
1232             /**************************
1233              * CALCULATE INTERACTIONS *
1234              **************************/
1235
1236             if (gmx_mm_any_lt(rsq20,rcutoff2))
1237             {
1238
1239             r20              = _mm_mul_ps(rsq20,rinv20);
1240             r20              = _mm_andnot_ps(dummy_mask,r20);
1241
1242             /* Compute parameters for interactions between i and j atoms */
1243             qq20             = _mm_mul_ps(iq2,jq0);
1244
1245             /* EWALD ELECTROSTATICS */
1246
1247             /* Analytical PME correction */
1248             zeta2            = _mm_mul_ps(beta2,rsq20);
1249             rinv3            = _mm_mul_ps(rinvsq20,rinv20);
1250             pmecorrF         = gmx_mm_pmecorrF_ps(zeta2);
1251             felec            = _mm_macc_ps(pmecorrF,beta3,rinv3);
1252             felec            = _mm_mul_ps(qq20,felec);
1253
1254             cutoff_mask      = _mm_cmplt_ps(rsq20,rcutoff2);
1255
1256             fscal            = felec;
1257
1258             fscal            = _mm_and_ps(fscal,cutoff_mask);
1259
1260             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1261
1262              /* Update vectorial force */
1263             fix2             = _mm_macc_ps(dx20,fscal,fix2);
1264             fiy2             = _mm_macc_ps(dy20,fscal,fiy2);
1265             fiz2             = _mm_macc_ps(dz20,fscal,fiz2);
1266
1267             fjx0             = _mm_macc_ps(dx20,fscal,fjx0);
1268             fjy0             = _mm_macc_ps(dy20,fscal,fjy0);
1269             fjz0             = _mm_macc_ps(dz20,fscal,fjz0);
1270
1271             }
1272
1273             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
1274             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
1275             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
1276             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
1277
1278             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
1279
1280             /* Inner loop uses 117 flops */
1281         }
1282
1283         /* End of innermost loop */
1284
1285         gmx_mm_update_iforce_3atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
1286                                               f+i_coord_offset,fshift+i_shift_offset);
1287
1288         /* Increment number of inner iterations */
1289         inneriter                  += j_index_end - j_index_start;
1290
1291         /* Outer loop uses 18 flops */
1292     }
1293
1294     /* Increment number of outer iterations */
1295     outeriter        += nri;
1296
1297     /* Update outer/inner flops */
1298
1299     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3_F,outeriter*18 + inneriter*117);
1300 }