src/gromacs/gmxlib/nonbonded/nb_kernel_avx_128_fma_single/nb_kernel_ElecEw_VdwLJEw_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 #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_single.h"
  50 #include "kernelutil_x86_avx_128_fma_single.h"
  51
  52 /*
  53  * Gromacs nonbonded kernel:   nb_kernel_ElecEw_VdwLJEw_GeomW3P1_VF_avx_128_fma_single
  54  * Electrostatics interaction: Ewald
  55  * VdW interaction:            LJEwald
  56  * Geometry:                   Water3-Particle
  57  * Calculate force/pot:        PotentialAndForce
  58  */
  59 void
  60 nb_kernel_ElecEw_VdwLJEw_GeomW3P1_VF_avx_128_fma_single
  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,C,D refer to j loop unrolling done with AVX_128, e.g. for the four 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,jnrC,jnrD;
  77     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
  78     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
  79     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
  80     real             rcutoff_scalar;
  81     real             *shiftvec,*fshift,*x,*f;
  82     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
  83     real             scratch[4*DIM];
  84     __m128           fscal,rcutoff,rcutoff2,jidxall;
  85     int              vdwioffset0;
  86     __m128           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
  87     int              vdwioffset1;
  88     __m128           ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
  89     int              vdwioffset2;
  90     __m128           ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
  91     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
  92     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  93     __m128           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  94     __m128           dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
  95     __m128           dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
  96     __m128           velec,felec,velecsum,facel,crf,krf,krf2;
  97     real             *charge;
  98     int              nvdwtype;
  99     __m128           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 100     int              *vdwtype;
 101     real             *vdwparam;
 102     __m128           one_sixth   = _mm_set1_ps(1.0/6.0);
 103     __m128           one_twelfth = _mm_set1_ps(1.0/12.0);
 104     __m128           c6grid_00;
 105     __m128           c6grid_10;
 106     __m128           c6grid_20;
 107     real             *vdwgridparam;
 108     __m128           ewclj,ewclj2,ewclj6,ewcljrsq,poly,exponent,f6A,f6B,sh_lj_ewald;
 109     __m128           one_half = _mm_set1_ps(0.5);
 110     __m128           minus_one = _mm_set1_ps(-1.0);
 111     __m128i          ewitab;
 112     __m128           ewtabscale,eweps,twoeweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
 113     __m128           beta,beta2,beta3,zeta2,pmecorrF,pmecorrV,rinv3;
 114     real             *ewtab;
 115     __m128           dummy_mask,cutoff_mask;
 116     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
 117     __m128           one     = _mm_set1_ps(1.0);
 118     __m128           two     = _mm_set1_ps(2.0);
 119     x                = xx[0];
 120     f                = ff[0];
 121
 122     nri              = nlist->nri;
 123     iinr             = nlist->iinr;
 124     jindex           = nlist->jindex;
 125     jjnr             = nlist->jjnr;
 126     shiftidx         = nlist->shift;
 127     gid              = nlist->gid;
 128     shiftvec         = fr->shift_vec[0];
 129     fshift           = fr->fshift[0];
 130     facel            = _mm_set1_ps(fr->epsfac);
 131     charge           = mdatoms->chargeA;
 132     nvdwtype         = fr->ntype;
 133     vdwparam         = fr->nbfp;
 134     vdwtype          = mdatoms->typeA;
 135     vdwgridparam     = fr->ljpme_c6grid;
 136     sh_lj_ewald      = _mm_set1_ps(fr->ic->sh_lj_ewald);
 137     ewclj            = _mm_set1_ps(fr->ewaldcoeff_lj);
 138     ewclj2           = _mm_mul_ps(minus_one,_mm_mul_ps(ewclj,ewclj));
 139
 140     sh_ewald         = _mm_set1_ps(fr->ic->sh_ewald);
 141     beta             = _mm_set1_ps(fr->ic->ewaldcoeff_q);
 142     beta2            = _mm_mul_ps(beta,beta);
 143     beta3            = _mm_mul_ps(beta,beta2);
 144     ewtab            = fr->ic->tabq_coul_FDV0;
 145     ewtabscale       = _mm_set1_ps(fr->ic->tabq_scale);
 146     ewtabhalfspace   = _mm_set1_ps(0.5/fr->ic->tabq_scale);
 147
 148     /* Setup water-specific parameters */
 149     inr              = nlist->iinr[0];
 150     iq0              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+0]));
 151     iq1              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+1]));
 152     iq2              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+2]));
 153     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 154
 155     /* Avoid stupid compiler warnings */
 156     jnrA = jnrB = jnrC = jnrD = 0;
 157     j_coord_offsetA = 0;
 158     j_coord_offsetB = 0;
 159     j_coord_offsetC = 0;
 160     j_coord_offsetD = 0;
 161
 162     outeriter        = 0;
 163     inneriter        = 0;
 164
 165     for(iidx=0;iidx<4*DIM;iidx++)
 166     {
 167         scratch[iidx] = 0.0;
 168     }
 169
 170     /* Start outer loop over neighborlists */
 171     for(iidx=0; iidx<nri; iidx++)
 172     {
 173         /* Load shift vector for this list */
 174         i_shift_offset   = DIM*shiftidx[iidx];
 175
 176         /* Load limits for loop over neighbors */
 177         j_index_start    = jindex[iidx];
 178         j_index_end      = jindex[iidx+1];
 179
 180         /* Get outer coordinate index */
 181         inr              = iinr[iidx];
 182         i_coord_offset   = DIM*inr;
 183
 184         /* Load i particle coords and add shift vector */
 185         gmx_mm_load_shift_and_3rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
 186                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 187
 188         fix0             = _mm_setzero_ps();
 189         fiy0             = _mm_setzero_ps();
 190         fiz0             = _mm_setzero_ps();
 191         fix1             = _mm_setzero_ps();
 192         fiy1             = _mm_setzero_ps();
 193         fiz1             = _mm_setzero_ps();
 194         fix2             = _mm_setzero_ps();
 195         fiy2             = _mm_setzero_ps();
 196         fiz2             = _mm_setzero_ps();
 197
 198         /* Reset potential sums */
 199         velecsum         = _mm_setzero_ps();
 200         vvdwsum          = _mm_setzero_ps();
 201
 202         /* Start inner kernel loop */
 203         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 204         {
 205
 206             /* Get j neighbor index, and coordinate index */
 207             jnrA             = jjnr[jidx];
 208             jnrB             = jjnr[jidx+1];
 209             jnrC             = jjnr[jidx+2];
 210             jnrD             = jjnr[jidx+3];
 211             j_coord_offsetA  = DIM*jnrA;
 212             j_coord_offsetB  = DIM*jnrB;
 213             j_coord_offsetC  = DIM*jnrC;
 214             j_coord_offsetD  = DIM*jnrD;
 215
 216             /* load j atom coordinates */
 217             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 218                                               x+j_coord_offsetC,x+j_coord_offsetD,
 219                                               &jx0,&jy0,&jz0);
 220
 221             /* Calculate displacement vector */
 222             dx00             = _mm_sub_ps(ix0,jx0);
 223             dy00             = _mm_sub_ps(iy0,jy0);
 224             dz00             = _mm_sub_ps(iz0,jz0);
 225             dx10             = _mm_sub_ps(ix1,jx0);
 226             dy10             = _mm_sub_ps(iy1,jy0);
 227             dz10             = _mm_sub_ps(iz1,jz0);
 228             dx20             = _mm_sub_ps(ix2,jx0);
 229             dy20             = _mm_sub_ps(iy2,jy0);
 230             dz20             = _mm_sub_ps(iz2,jz0);
 231
 232             /* Calculate squared distance and things based on it */
 233             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 234             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 235             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 236
 237             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 238             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 239             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 240
 241             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 242             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 243             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 244
 245             /* Load parameters for j particles */
 246             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 247                                                               charge+jnrC+0,charge+jnrD+0);
 248             vdwjidx0A        = 2*vdwtype[jnrA+0];
 249             vdwjidx0B        = 2*vdwtype[jnrB+0];
 250             vdwjidx0C        = 2*vdwtype[jnrC+0];
 251             vdwjidx0D        = 2*vdwtype[jnrD+0];
 252
 253             fjx0             = _mm_setzero_ps();
 254             fjy0             = _mm_setzero_ps();
 255             fjz0             = _mm_setzero_ps();
 256
 257             /**************************
 258              * CALCULATE INTERACTIONS *
 259              **************************/
 260
 261             r00              = _mm_mul_ps(rsq00,rinv00);
 262
 263             /* Compute parameters for interactions between i and j atoms */
 264             qq00             = _mm_mul_ps(iq0,jq0);
 265             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 266                                          vdwparam+vdwioffset0+vdwjidx0B,
 267                                          vdwparam+vdwioffset0+vdwjidx0C,
 268                                          vdwparam+vdwioffset0+vdwjidx0D,
 269                                          &c6_00,&c12_00);
 270
 271             c6grid_00       = gmx_mm_load_4real_swizzle_ps(vdwgridparam+vdwioffset0+vdwjidx0A,
 272                                                                vdwgridparam+vdwioffset0+vdwjidx0B,
 273                                                                vdwgridparam+vdwioffset0+vdwjidx0C,
 274                                                                vdwgridparam+vdwioffset0+vdwjidx0D);
 275
 276             /* EWALD ELECTROSTATICS */
 277
 278             /* Analytical PME correction */
 279             zeta2            = _mm_mul_ps(beta2,rsq00);
 280             rinv3            = _mm_mul_ps(rinvsq00,rinv00);
 281             pmecorrF         = gmx_mm_pmecorrF_ps(zeta2);
 282             felec            = _mm_macc_ps(pmecorrF,beta3,rinv3);
 283             felec            = _mm_mul_ps(qq00,felec);
 284             pmecorrV         = gmx_mm_pmecorrV_ps(zeta2);
 285             velec            = _mm_nmacc_ps(pmecorrV,beta,rinv00);
 286             velec            = _mm_mul_ps(qq00,velec);
 287
 288             /* Analytical LJ-PME */
 289             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 290             ewcljrsq         = _mm_mul_ps(ewclj2,rsq00);
 291             ewclj6           = _mm_mul_ps(ewclj2,_mm_mul_ps(ewclj2,ewclj2));
 292             exponent         = gmx_simd_exp_r(ewcljrsq);
 293             /* poly = exp(-(beta*r)^2) * (1 + (beta*r)^2 + (beta*r)^4 /2) */
 294             poly             = _mm_mul_ps(exponent,_mm_macc_ps(_mm_mul_ps(ewcljrsq,ewcljrsq),one_half,_mm_sub_ps(one,ewcljrsq)));
 295             /* vvdw6 = [C6 - C6grid * (1-poly)]/r6 */
 296             vvdw6            = _mm_mul_ps(_mm_macc_ps(-c6grid_00,_mm_sub_ps(one,poly),c6_00),rinvsix);
 297             vvdw12           = _mm_mul_ps(c12_00,_mm_mul_ps(rinvsix,rinvsix));
 298             vvdw             = _mm_msub_ps(vvdw12,one_twelfth,_mm_mul_ps(vvdw6,one_sixth));
 299             /* fvdw = vvdw12/r - (vvdw6/r + (C6grid * exponent * beta^6)/r) */
 300             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);
 301
 302             /* Update potential sum for this i atom from the interaction with this j atom. */
 303             velecsum         = _mm_add_ps(velecsum,velec);
 304             vvdwsum          = _mm_add_ps(vvdwsum,vvdw);
 305
 306             fscal            = _mm_add_ps(felec,fvdw);
 307
 308              /* Update vectorial force */
 309             fix0             = _mm_macc_ps(dx00,fscal,fix0);
 310             fiy0             = _mm_macc_ps(dy00,fscal,fiy0);
 311             fiz0             = _mm_macc_ps(dz00,fscal,fiz0);
 312
 313             fjx0             = _mm_macc_ps(dx00,fscal,fjx0);
 314             fjy0             = _mm_macc_ps(dy00,fscal,fjy0);
 315             fjz0             = _mm_macc_ps(dz00,fscal,fjz0);
 316
 317             /**************************
 318              * CALCULATE INTERACTIONS *
 319              **************************/
 320
 321             r10              = _mm_mul_ps(rsq10,rinv10);
 322
 323             /* Compute parameters for interactions between i and j atoms */
 324             qq10             = _mm_mul_ps(iq1,jq0);
 325
 326             /* EWALD ELECTROSTATICS */
 327
 328             /* Analytical PME correction */
 329             zeta2            = _mm_mul_ps(beta2,rsq10);
 330             rinv3            = _mm_mul_ps(rinvsq10,rinv10);
 331             pmecorrF         = gmx_mm_pmecorrF_ps(zeta2);
 332             felec            = _mm_macc_ps(pmecorrF,beta3,rinv3);
 333             felec            = _mm_mul_ps(qq10,felec);
 334             pmecorrV         = gmx_mm_pmecorrV_ps(zeta2);
 335             velec            = _mm_nmacc_ps(pmecorrV,beta,rinv10);
 336             velec            = _mm_mul_ps(qq10,velec);
 337
 338             /* Update potential sum for this i atom from the interaction with this j atom. */
 339             velecsum         = _mm_add_ps(velecsum,velec);
 340
 341             fscal            = felec;
 342
 343              /* Update vectorial force */
 344             fix1             = _mm_macc_ps(dx10,fscal,fix1);
 345             fiy1             = _mm_macc_ps(dy10,fscal,fiy1);
 346             fiz1             = _mm_macc_ps(dz10,fscal,fiz1);
 347
 348             fjx0             = _mm_macc_ps(dx10,fscal,fjx0);
 349             fjy0             = _mm_macc_ps(dy10,fscal,fjy0);
 350             fjz0             = _mm_macc_ps(dz10,fscal,fjz0);
 351
 352             /**************************
 353              * CALCULATE INTERACTIONS *
 354              **************************/
 355
 356             r20              = _mm_mul_ps(rsq20,rinv20);
 357
 358             /* Compute parameters for interactions between i and j atoms */
 359             qq20             = _mm_mul_ps(iq2,jq0);
 360
 361             /* EWALD ELECTROSTATICS */
 362
 363             /* Analytical PME correction */
 364             zeta2            = _mm_mul_ps(beta2,rsq20);
 365             rinv3            = _mm_mul_ps(rinvsq20,rinv20);
 366             pmecorrF         = gmx_mm_pmecorrF_ps(zeta2);
 367             felec            = _mm_macc_ps(pmecorrF,beta3,rinv3);
 368             felec            = _mm_mul_ps(qq20,felec);
 369             pmecorrV         = gmx_mm_pmecorrV_ps(zeta2);
 370             velec            = _mm_nmacc_ps(pmecorrV,beta,rinv20);
 371             velec            = _mm_mul_ps(qq20,velec);
 372
 373             /* Update potential sum for this i atom from the interaction with this j atom. */
 374             velecsum         = _mm_add_ps(velecsum,velec);
 375
 376             fscal            = felec;
 377
 378              /* Update vectorial force */
 379             fix2             = _mm_macc_ps(dx20,fscal,fix2);
 380             fiy2             = _mm_macc_ps(dy20,fscal,fiy2);
 381             fiz2             = _mm_macc_ps(dz20,fscal,fiz2);
 382
 383             fjx0             = _mm_macc_ps(dx20,fscal,fjx0);
 384             fjy0             = _mm_macc_ps(dy20,fscal,fjy0);
 385             fjz0             = _mm_macc_ps(dz20,fscal,fjz0);
 386
 387             fjptrA             = f+j_coord_offsetA;
 388             fjptrB             = f+j_coord_offsetB;
 389             fjptrC             = f+j_coord_offsetC;
 390             fjptrD             = f+j_coord_offsetD;
 391
 392             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 393
 394             /* Inner loop uses 111 flops */
 395         }
 396
 397         if(jidx<j_index_end)
 398         {
 399
 400             /* Get j neighbor index, and coordinate index */
 401             jnrlistA         = jjnr[jidx];
 402             jnrlistB         = jjnr[jidx+1];
 403             jnrlistC         = jjnr[jidx+2];
 404             jnrlistD         = jjnr[jidx+3];
 405             /* Sign of each element will be negative for non-real atoms.
 406              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 407              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 408              */
 409             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 410             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 411             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 412             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 413             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 414             j_coord_offsetA  = DIM*jnrA;
 415             j_coord_offsetB  = DIM*jnrB;
 416             j_coord_offsetC  = DIM*jnrC;
 417             j_coord_offsetD  = DIM*jnrD;
 418
 419             /* load j atom coordinates */
 420             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 421                                               x+j_coord_offsetC,x+j_coord_offsetD,
 422                                               &jx0,&jy0,&jz0);
 423
 424             /* Calculate displacement vector */
 425             dx00             = _mm_sub_ps(ix0,jx0);
 426             dy00             = _mm_sub_ps(iy0,jy0);
 427             dz00             = _mm_sub_ps(iz0,jz0);
 428             dx10             = _mm_sub_ps(ix1,jx0);
 429             dy10             = _mm_sub_ps(iy1,jy0);
 430             dz10             = _mm_sub_ps(iz1,jz0);
 431             dx20             = _mm_sub_ps(ix2,jx0);
 432             dy20             = _mm_sub_ps(iy2,jy0);
 433             dz20             = _mm_sub_ps(iz2,jz0);
 434
 435             /* Calculate squared distance and things based on it */
 436             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 437             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 438             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 439
 440             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 441             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 442             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 443
 444             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 445             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 446             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 447
 448             /* Load parameters for j particles */
 449             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 450                                                               charge+jnrC+0,charge+jnrD+0);
 451             vdwjidx0A        = 2*vdwtype[jnrA+0];
 452             vdwjidx0B        = 2*vdwtype[jnrB+0];
 453             vdwjidx0C        = 2*vdwtype[jnrC+0];
 454             vdwjidx0D        = 2*vdwtype[jnrD+0];
 455
 456             fjx0             = _mm_setzero_ps();
 457             fjy0             = _mm_setzero_ps();
 458             fjz0             = _mm_setzero_ps();
 459
 460             /**************************
 461              * CALCULATE INTERACTIONS *
 462              **************************/
 463
 464             r00              = _mm_mul_ps(rsq00,rinv00);
 465             r00              = _mm_andnot_ps(dummy_mask,r00);
 466
 467             /* Compute parameters for interactions between i and j atoms */
 468             qq00             = _mm_mul_ps(iq0,jq0);
 469             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 470                                          vdwparam+vdwioffset0+vdwjidx0B,
 471                                          vdwparam+vdwioffset0+vdwjidx0C,
 472                                          vdwparam+vdwioffset0+vdwjidx0D,
 473                                          &c6_00,&c12_00);
 474
 475             c6grid_00       = gmx_mm_load_4real_swizzle_ps(vdwgridparam+vdwioffset0+vdwjidx0A,
 476                                                                vdwgridparam+vdwioffset0+vdwjidx0B,
 477                                                                vdwgridparam+vdwioffset0+vdwjidx0C,
 478                                                                vdwgridparam+vdwioffset0+vdwjidx0D);
 479
 480             /* EWALD ELECTROSTATICS */
 481
 482             /* Analytical PME correction */
 483             zeta2            = _mm_mul_ps(beta2,rsq00);
 484             rinv3            = _mm_mul_ps(rinvsq00,rinv00);
 485             pmecorrF         = gmx_mm_pmecorrF_ps(zeta2);
 486             felec            = _mm_macc_ps(pmecorrF,beta3,rinv3);
 487             felec            = _mm_mul_ps(qq00,felec);
 488             pmecorrV         = gmx_mm_pmecorrV_ps(zeta2);
 489             velec            = _mm_nmacc_ps(pmecorrV,beta,rinv00);
 490             velec            = _mm_mul_ps(qq00,velec);
 491
 492             /* Analytical LJ-PME */
 493             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 494             ewcljrsq         = _mm_mul_ps(ewclj2,rsq00);
 495             ewclj6           = _mm_mul_ps(ewclj2,_mm_mul_ps(ewclj2,ewclj2));
 496             exponent         = gmx_simd_exp_r(ewcljrsq);
 497             /* poly = exp(-(beta*r)^2) * (1 + (beta*r)^2 + (beta*r)^4 /2) */
 498             poly             = _mm_mul_ps(exponent,_mm_macc_ps(_mm_mul_ps(ewcljrsq,ewcljrsq),one_half,_mm_sub_ps(one,ewcljrsq)));
 499             /* vvdw6 = [C6 - C6grid * (1-poly)]/r6 */
 500             vvdw6            = _mm_mul_ps(_mm_macc_ps(-c6grid_00,_mm_sub_ps(one,poly),c6_00),rinvsix);
 501             vvdw12           = _mm_mul_ps(c12_00,_mm_mul_ps(rinvsix,rinvsix));
 502             vvdw             = _mm_msub_ps(vvdw12,one_twelfth,_mm_mul_ps(vvdw6,one_sixth));
 503             /* fvdw = vvdw12/r - (vvdw6/r + (C6grid * exponent * beta^6)/r) */
 504             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);
 505
 506             /* Update potential sum for this i atom from the interaction with this j atom. */
 507             velec            = _mm_andnot_ps(dummy_mask,velec);
 508             velecsum         = _mm_add_ps(velecsum,velec);
 509             vvdw             = _mm_andnot_ps(dummy_mask,vvdw);
 510             vvdwsum          = _mm_add_ps(vvdwsum,vvdw);
 511
 512             fscal            = _mm_add_ps(felec,fvdw);
 513
 514             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 515
 516              /* Update vectorial force */
 517             fix0             = _mm_macc_ps(dx00,fscal,fix0);
 518             fiy0             = _mm_macc_ps(dy00,fscal,fiy0);
 519             fiz0             = _mm_macc_ps(dz00,fscal,fiz0);
 520
 521             fjx0             = _mm_macc_ps(dx00,fscal,fjx0);
 522             fjy0             = _mm_macc_ps(dy00,fscal,fjy0);
 523             fjz0             = _mm_macc_ps(dz00,fscal,fjz0);
 524
 525             /**************************
 526              * CALCULATE INTERACTIONS *
 527              **************************/
 528
 529             r10              = _mm_mul_ps(rsq10,rinv10);
 530             r10              = _mm_andnot_ps(dummy_mask,r10);
 531
 532             /* Compute parameters for interactions between i and j atoms */
 533             qq10             = _mm_mul_ps(iq1,jq0);
 534
 535             /* EWALD ELECTROSTATICS */
 536
 537             /* Analytical PME correction */
 538             zeta2            = _mm_mul_ps(beta2,rsq10);
 539             rinv3            = _mm_mul_ps(rinvsq10,rinv10);
 540             pmecorrF         = gmx_mm_pmecorrF_ps(zeta2);
 541             felec            = _mm_macc_ps(pmecorrF,beta3,rinv3);
 542             felec            = _mm_mul_ps(qq10,felec);
 543             pmecorrV         = gmx_mm_pmecorrV_ps(zeta2);
 544             velec            = _mm_nmacc_ps(pmecorrV,beta,rinv10);
 545             velec            = _mm_mul_ps(qq10,velec);
 546
 547             /* Update potential sum for this i atom from the interaction with this j atom. */
 548             velec            = _mm_andnot_ps(dummy_mask,velec);
 549             velecsum         = _mm_add_ps(velecsum,velec);
 550
 551             fscal            = felec;
 552
 553             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 554
 555              /* Update vectorial force */
 556             fix1             = _mm_macc_ps(dx10,fscal,fix1);
 557             fiy1             = _mm_macc_ps(dy10,fscal,fiy1);
 558             fiz1             = _mm_macc_ps(dz10,fscal,fiz1);
 559
 560             fjx0             = _mm_macc_ps(dx10,fscal,fjx0);
 561             fjy0             = _mm_macc_ps(dy10,fscal,fjy0);
 562             fjz0             = _mm_macc_ps(dz10,fscal,fjz0);
 563
 564             /**************************
 565              * CALCULATE INTERACTIONS *
 566              **************************/
 567
 568             r20              = _mm_mul_ps(rsq20,rinv20);
 569             r20              = _mm_andnot_ps(dummy_mask,r20);
 570
 571             /* Compute parameters for interactions between i and j atoms */
 572             qq20             = _mm_mul_ps(iq2,jq0);
 573
 574             /* EWALD ELECTROSTATICS */
 575
 576             /* Analytical PME correction */
 577             zeta2            = _mm_mul_ps(beta2,rsq20);
 578             rinv3            = _mm_mul_ps(rinvsq20,rinv20);
 579             pmecorrF         = gmx_mm_pmecorrF_ps(zeta2);
 580             felec            = _mm_macc_ps(pmecorrF,beta3,rinv3);
 581             felec            = _mm_mul_ps(qq20,felec);
 582             pmecorrV         = gmx_mm_pmecorrV_ps(zeta2);
 583             velec            = _mm_nmacc_ps(pmecorrV,beta,rinv20);
 584             velec            = _mm_mul_ps(qq20,velec);
 585
 586             /* Update potential sum for this i atom from the interaction with this j atom. */
 587             velec            = _mm_andnot_ps(dummy_mask,velec);
 588             velecsum         = _mm_add_ps(velecsum,velec);
 589
 590             fscal            = felec;
 591
 592             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 593
 594              /* Update vectorial force */
 595             fix2             = _mm_macc_ps(dx20,fscal,fix2);
 596             fiy2             = _mm_macc_ps(dy20,fscal,fiy2);
 597             fiz2             = _mm_macc_ps(dz20,fscal,fiz2);
 598
 599             fjx0             = _mm_macc_ps(dx20,fscal,fjx0);
 600             fjy0             = _mm_macc_ps(dy20,fscal,fjy0);
 601             fjz0             = _mm_macc_ps(dz20,fscal,fjz0);
 602
 603             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 604             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 605             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 606             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 607
 608             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 609
 610             /* Inner loop uses 114 flops */
 611         }
 612
 613         /* End of innermost loop */
 614
 615         gmx_mm_update_iforce_3atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
 616                                               f+i_coord_offset,fshift+i_shift_offset);
 617
 618         ggid                        = gid[iidx];
 619         /* Update potential energies */
 620         gmx_mm_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid);
 621         gmx_mm_update_1pot_ps(vvdwsum,kernel_data->energygrp_vdw+ggid);
 622
 623         /* Increment number of inner iterations */
 624         inneriter                  += j_index_end - j_index_start;
 625
 626         /* Outer loop uses 20 flops */
 627     }
 628
 629     /* Increment number of outer iterations */
 630     outeriter        += nri;
 631
 632     /* Update outer/inner flops */
 633
 634     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3_VF,outeriter*20 + inneriter*114);
 635 }
 636 /*
 637  * Gromacs nonbonded kernel:   nb_kernel_ElecEw_VdwLJEw_GeomW3P1_F_avx_128_fma_single
 638  * Electrostatics interaction: Ewald
 639  * VdW interaction:            LJEwald
 640  * Geometry:                   Water3-Particle
 641  * Calculate force/pot:        Force
 642  */
 643 void
 644 nb_kernel_ElecEw_VdwLJEw_GeomW3P1_F_avx_128_fma_single
 645                     (t_nblist                    * gmx_restrict       nlist,
 646                      rvec                        * gmx_restrict          xx,
 647                      rvec                        * gmx_restrict          ff,
 648                      t_forcerec                  * gmx_restrict          fr,
 649                      t_mdatoms                   * gmx_restrict     mdatoms,
 650                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 651                      t_nrnb                      * gmx_restrict        nrnb)
 652 {
 653     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 654      * just 0 for non-waters.
 655      * Suffixes A,B,C,D refer to j loop unrolling done with AVX_128, e.g. for the four different
 656      * jnr indices corresponding to data put in the four positions in the SIMD register.
 657      */
 658     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 659     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 660     int              jnrA,jnrB,jnrC,jnrD;
 661     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
 662     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 663     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 664     real             rcutoff_scalar;
 665     real             *shiftvec,*fshift,*x,*f;
 666     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
 667     real             scratch[4*DIM];
 668     __m128           fscal,rcutoff,rcutoff2,jidxall;
 669     int              vdwioffset0;
 670     __m128           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 671     int              vdwioffset1;
 672     __m128           ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
 673     int              vdwioffset2;
 674     __m128           ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
 675     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
 676     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 677     __m128           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 678     __m128           dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
 679     __m128           dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
 680     __m128           velec,felec,velecsum,facel,crf,krf,krf2;
 681     real             *charge;
 682     int              nvdwtype;
 683     __m128           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 684     int              *vdwtype;
 685     real             *vdwparam;
 686     __m128           one_sixth   = _mm_set1_ps(1.0/6.0);
 687     __m128           one_twelfth = _mm_set1_ps(1.0/12.0);
 688     __m128           c6grid_00;
 689     __m128           c6grid_10;
 690     __m128           c6grid_20;
 691     real             *vdwgridparam;
 692     __m128           ewclj,ewclj2,ewclj6,ewcljrsq,poly,exponent,f6A,f6B,sh_lj_ewald;
 693     __m128           one_half = _mm_set1_ps(0.5);
 694     __m128           minus_one = _mm_set1_ps(-1.0);
 695     __m128i          ewitab;
 696     __m128           ewtabscale,eweps,twoeweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
 697     __m128           beta,beta2,beta3,zeta2,pmecorrF,pmecorrV,rinv3;
 698     real             *ewtab;
 699     __m128           dummy_mask,cutoff_mask;
 700     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
 701     __m128           one     = _mm_set1_ps(1.0);
 702     __m128           two     = _mm_set1_ps(2.0);
 703     x                = xx[0];
 704     f                = ff[0];
 705
 706     nri              = nlist->nri;
 707     iinr             = nlist->iinr;
 708     jindex           = nlist->jindex;
 709     jjnr             = nlist->jjnr;
 710     shiftidx         = nlist->shift;
 711     gid              = nlist->gid;
 712     shiftvec         = fr->shift_vec[0];
 713     fshift           = fr->fshift[0];
 714     facel            = _mm_set1_ps(fr->epsfac);
 715     charge           = mdatoms->chargeA;
 716     nvdwtype         = fr->ntype;
 717     vdwparam         = fr->nbfp;
 718     vdwtype          = mdatoms->typeA;
 719     vdwgridparam     = fr->ljpme_c6grid;
 720     sh_lj_ewald      = _mm_set1_ps(fr->ic->sh_lj_ewald);
 721     ewclj            = _mm_set1_ps(fr->ewaldcoeff_lj);
 722     ewclj2           = _mm_mul_ps(minus_one,_mm_mul_ps(ewclj,ewclj));
 723
 724     sh_ewald         = _mm_set1_ps(fr->ic->sh_ewald);
 725     beta             = _mm_set1_ps(fr->ic->ewaldcoeff_q);
 726     beta2            = _mm_mul_ps(beta,beta);
 727     beta3            = _mm_mul_ps(beta,beta2);
 728     ewtab            = fr->ic->tabq_coul_F;
 729     ewtabscale       = _mm_set1_ps(fr->ic->tabq_scale);
 730     ewtabhalfspace   = _mm_set1_ps(0.5/fr->ic->tabq_scale);
 731
 732     /* Setup water-specific parameters */
 733     inr              = nlist->iinr[0];
 734     iq0              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+0]));
 735     iq1              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+1]));
 736     iq2              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+2]));
 737     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 738
 739     /* Avoid stupid compiler warnings */
 740     jnrA = jnrB = jnrC = jnrD = 0;
 741     j_coord_offsetA = 0;
 742     j_coord_offsetB = 0;
 743     j_coord_offsetC = 0;
 744     j_coord_offsetD = 0;
 745
 746     outeriter        = 0;
 747     inneriter        = 0;
 748
 749     for(iidx=0;iidx<4*DIM;iidx++)
 750     {
 751         scratch[iidx] = 0.0;
 752     }
 753
 754     /* Start outer loop over neighborlists */
 755     for(iidx=0; iidx<nri; iidx++)
 756     {
 757         /* Load shift vector for this list */
 758         i_shift_offset   = DIM*shiftidx[iidx];
 759
 760         /* Load limits for loop over neighbors */
 761         j_index_start    = jindex[iidx];
 762         j_index_end      = jindex[iidx+1];
 763
 764         /* Get outer coordinate index */
 765         inr              = iinr[iidx];
 766         i_coord_offset   = DIM*inr;
 767
 768         /* Load i particle coords and add shift vector */
 769         gmx_mm_load_shift_and_3rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
 770                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 771
 772         fix0             = _mm_setzero_ps();
 773         fiy0             = _mm_setzero_ps();
 774         fiz0             = _mm_setzero_ps();
 775         fix1             = _mm_setzero_ps();
 776         fiy1             = _mm_setzero_ps();
 777         fiz1             = _mm_setzero_ps();
 778         fix2             = _mm_setzero_ps();
 779         fiy2             = _mm_setzero_ps();
 780         fiz2             = _mm_setzero_ps();
 781
 782         /* Start inner kernel loop */
 783         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 784         {
 785
 786             /* Get j neighbor index, and coordinate index */
 787             jnrA             = jjnr[jidx];
 788             jnrB             = jjnr[jidx+1];
 789             jnrC             = jjnr[jidx+2];
 790             jnrD             = jjnr[jidx+3];
 791             j_coord_offsetA  = DIM*jnrA;
 792             j_coord_offsetB  = DIM*jnrB;
 793             j_coord_offsetC  = DIM*jnrC;
 794             j_coord_offsetD  = DIM*jnrD;
 795
 796             /* load j atom coordinates */
 797             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 798                                               x+j_coord_offsetC,x+j_coord_offsetD,
 799                                               &jx0,&jy0,&jz0);
 800
 801             /* Calculate displacement vector */
 802             dx00             = _mm_sub_ps(ix0,jx0);
 803             dy00             = _mm_sub_ps(iy0,jy0);
 804             dz00             = _mm_sub_ps(iz0,jz0);
 805             dx10             = _mm_sub_ps(ix1,jx0);
 806             dy10             = _mm_sub_ps(iy1,jy0);
 807             dz10             = _mm_sub_ps(iz1,jz0);
 808             dx20             = _mm_sub_ps(ix2,jx0);
 809             dy20             = _mm_sub_ps(iy2,jy0);
 810             dz20             = _mm_sub_ps(iz2,jz0);
 811
 812             /* Calculate squared distance and things based on it */
 813             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 814             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 815             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 816
 817             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 818             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 819             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 820
 821             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 822             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 823             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 824
 825             /* Load parameters for j particles */
 826             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 827                                                               charge+jnrC+0,charge+jnrD+0);
 828             vdwjidx0A        = 2*vdwtype[jnrA+0];
 829             vdwjidx0B        = 2*vdwtype[jnrB+0];
 830             vdwjidx0C        = 2*vdwtype[jnrC+0];
 831             vdwjidx0D        = 2*vdwtype[jnrD+0];
 832
 833             fjx0             = _mm_setzero_ps();
 834             fjy0             = _mm_setzero_ps();
 835             fjz0             = _mm_setzero_ps();
 836
 837             /**************************
 838              * CALCULATE INTERACTIONS *
 839              **************************/
 840
 841             r00              = _mm_mul_ps(rsq00,rinv00);
 842
 843             /* Compute parameters for interactions between i and j atoms */
 844             qq00             = _mm_mul_ps(iq0,jq0);
 845             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 846                                          vdwparam+vdwioffset0+vdwjidx0B,
 847                                          vdwparam+vdwioffset0+vdwjidx0C,
 848                                          vdwparam+vdwioffset0+vdwjidx0D,
 849                                          &c6_00,&c12_00);
 850
 851             c6grid_00       = gmx_mm_load_4real_swizzle_ps(vdwgridparam+vdwioffset0+vdwjidx0A,
 852                                                                vdwgridparam+vdwioffset0+vdwjidx0B,
 853                                                                vdwgridparam+vdwioffset0+vdwjidx0C,
 854                                                                vdwgridparam+vdwioffset0+vdwjidx0D);
 855
 856             /* EWALD ELECTROSTATICS */
 857
 858             /* Analytical PME correction */
 859             zeta2            = _mm_mul_ps(beta2,rsq00);
 860             rinv3            = _mm_mul_ps(rinvsq00,rinv00);
 861             pmecorrF         = gmx_mm_pmecorrF_ps(zeta2);
 862             felec            = _mm_macc_ps(pmecorrF,beta3,rinv3);
 863             felec            = _mm_mul_ps(qq00,felec);
 864
 865             /* Analytical LJ-PME */
 866             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 867             ewcljrsq         = _mm_mul_ps(ewclj2,rsq00);
 868             ewclj6           = _mm_mul_ps(ewclj2,_mm_mul_ps(ewclj2,ewclj2));
 869             exponent         = gmx_simd_exp_r(ewcljrsq);
 870             /* poly = exp(-(beta*r)^2) * (1 + (beta*r)^2 + (beta*r)^4 /2) */
 871             poly             = _mm_mul_ps(exponent,_mm_macc_ps(_mm_mul_ps(ewcljrsq,ewcljrsq),one_half,_mm_sub_ps(one,ewcljrsq)));
 872             /* f6A = 6 * C6grid * (1 - poly) */
 873             f6A              = _mm_mul_ps(c6grid_00,_mm_sub_ps(one,poly));
 874             /* f6B = C6grid * exponent * beta^6 */
 875             f6B              = _mm_mul_ps(_mm_mul_ps(c6grid_00,one_sixth),_mm_mul_ps(exponent,ewclj6));
 876             /* fvdw = 12*C12/r13 - ((6*C6 - f6A)/r6 + f6B)/r */
 877             fvdw              = _mm_mul_ps(_mm_macc_ps(_mm_msub_ps(c12_00,rinvsix,_mm_sub_ps(c6_00,f6A)),rinvsix,f6B),rinvsq00);
 878
 879             fscal            = _mm_add_ps(felec,fvdw);
 880
 881              /* Update vectorial force */
 882             fix0             = _mm_macc_ps(dx00,fscal,fix0);
 883             fiy0             = _mm_macc_ps(dy00,fscal,fiy0);
 884             fiz0             = _mm_macc_ps(dz00,fscal,fiz0);
 885
 886             fjx0             = _mm_macc_ps(dx00,fscal,fjx0);
 887             fjy0             = _mm_macc_ps(dy00,fscal,fjy0);
 888             fjz0             = _mm_macc_ps(dz00,fscal,fjz0);
 889
 890             /**************************
 891              * CALCULATE INTERACTIONS *
 892              **************************/
 893
 894             r10              = _mm_mul_ps(rsq10,rinv10);
 895
 896             /* Compute parameters for interactions between i and j atoms */
 897             qq10             = _mm_mul_ps(iq1,jq0);
 898
 899             /* EWALD ELECTROSTATICS */
 900
 901             /* Analytical PME correction */
 902             zeta2            = _mm_mul_ps(beta2,rsq10);
 903             rinv3            = _mm_mul_ps(rinvsq10,rinv10);
 904             pmecorrF         = gmx_mm_pmecorrF_ps(zeta2);
 905             felec            = _mm_macc_ps(pmecorrF,beta3,rinv3);
 906             felec            = _mm_mul_ps(qq10,felec);
 907
 908             fscal            = felec;
 909
 910              /* Update vectorial force */
 911             fix1             = _mm_macc_ps(dx10,fscal,fix1);
 912             fiy1             = _mm_macc_ps(dy10,fscal,fiy1);
 913             fiz1             = _mm_macc_ps(dz10,fscal,fiz1);
 914
 915             fjx0             = _mm_macc_ps(dx10,fscal,fjx0);
 916             fjy0             = _mm_macc_ps(dy10,fscal,fjy0);
 917             fjz0             = _mm_macc_ps(dz10,fscal,fjz0);
 918
 919             /**************************
 920              * CALCULATE INTERACTIONS *
 921              **************************/
 922
 923             r20              = _mm_mul_ps(rsq20,rinv20);
 924
 925             /* Compute parameters for interactions between i and j atoms */
 926             qq20             = _mm_mul_ps(iq2,jq0);
 927
 928             /* EWALD ELECTROSTATICS */
 929
 930             /* Analytical PME correction */
 931             zeta2            = _mm_mul_ps(beta2,rsq20);
 932             rinv3            = _mm_mul_ps(rinvsq20,rinv20);
 933             pmecorrF         = gmx_mm_pmecorrF_ps(zeta2);
 934             felec            = _mm_macc_ps(pmecorrF,beta3,rinv3);
 935             felec            = _mm_mul_ps(qq20,felec);
 936
 937             fscal            = felec;
 938
 939              /* Update vectorial force */
 940             fix2             = _mm_macc_ps(dx20,fscal,fix2);
 941             fiy2             = _mm_macc_ps(dy20,fscal,fiy2);
 942             fiz2             = _mm_macc_ps(dz20,fscal,fiz2);
 943
 944             fjx0             = _mm_macc_ps(dx20,fscal,fjx0);
 945             fjy0             = _mm_macc_ps(dy20,fscal,fjy0);
 946             fjz0             = _mm_macc_ps(dz20,fscal,fjz0);
 947
 948             fjptrA             = f+j_coord_offsetA;
 949             fjptrB             = f+j_coord_offsetB;
 950             fjptrC             = f+j_coord_offsetC;
 951             fjptrD             = f+j_coord_offsetD;
 952
 953             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 954
 955             /* Inner loop uses 105 flops */
 956         }
 957
 958         if(jidx<j_index_end)
 959         {
 960
 961             /* Get j neighbor index, and coordinate index */
 962             jnrlistA         = jjnr[jidx];
 963             jnrlistB         = jjnr[jidx+1];
 964             jnrlistC         = jjnr[jidx+2];
 965             jnrlistD         = jjnr[jidx+3];
 966             /* Sign of each element will be negative for non-real atoms.
 967              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 968              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 969              */
 970             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 971             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 972             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 973             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 974             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 975             j_coord_offsetA  = DIM*jnrA;
 976             j_coord_offsetB  = DIM*jnrB;
 977             j_coord_offsetC  = DIM*jnrC;
 978             j_coord_offsetD  = DIM*jnrD;
 979
 980             /* load j atom coordinates */
 981             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 982                                               x+j_coord_offsetC,x+j_coord_offsetD,
 983                                               &jx0,&jy0,&jz0);
 984
 985             /* Calculate displacement vector */
 986             dx00             = _mm_sub_ps(ix0,jx0);
 987             dy00             = _mm_sub_ps(iy0,jy0);
 988             dz00             = _mm_sub_ps(iz0,jz0);
 989             dx10             = _mm_sub_ps(ix1,jx0);
 990             dy10             = _mm_sub_ps(iy1,jy0);
 991             dz10             = _mm_sub_ps(iz1,jz0);
 992             dx20             = _mm_sub_ps(ix2,jx0);
 993             dy20             = _mm_sub_ps(iy2,jy0);
 994             dz20             = _mm_sub_ps(iz2,jz0);
 995
 996             /* Calculate squared distance and things based on it */
 997             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 998             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 999             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
1000
1001             rinv00           = gmx_mm_invsqrt_ps(rsq00);
1002             rinv10           = gmx_mm_invsqrt_ps(rsq10);
1003             rinv20           = gmx_mm_invsqrt_ps(rsq20);
1004
1005             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
1006             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
1007             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
1008
1009             /* Load parameters for j particles */
1010             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
1011                                                               charge+jnrC+0,charge+jnrD+0);
1012             vdwjidx0A        = 2*vdwtype[jnrA+0];
1013             vdwjidx0B        = 2*vdwtype[jnrB+0];
1014             vdwjidx0C        = 2*vdwtype[jnrC+0];
1015             vdwjidx0D        = 2*vdwtype[jnrD+0];
1016
1017             fjx0             = _mm_setzero_ps();
1018             fjy0             = _mm_setzero_ps();
1019             fjz0             = _mm_setzero_ps();
1020
1021             /**************************
1022              * CALCULATE INTERACTIONS *
1023              **************************/
1024
1025             r00              = _mm_mul_ps(rsq00,rinv00);
1026             r00              = _mm_andnot_ps(dummy_mask,r00);
1027
1028             /* Compute parameters for interactions between i and j atoms */
1029             qq00             = _mm_mul_ps(iq0,jq0);
1030             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
1031                                          vdwparam+vdwioffset0+vdwjidx0B,
1032                                          vdwparam+vdwioffset0+vdwjidx0C,
1033                                          vdwparam+vdwioffset0+vdwjidx0D,
1034                                          &c6_00,&c12_00);
1035
1036             c6grid_00       = gmx_mm_load_4real_swizzle_ps(vdwgridparam+vdwioffset0+vdwjidx0A,
1037                                                                vdwgridparam+vdwioffset0+vdwjidx0B,
1038                                                                vdwgridparam+vdwioffset0+vdwjidx0C,
1039                                                                vdwgridparam+vdwioffset0+vdwjidx0D);
1040
1041             /* EWALD ELECTROSTATICS */
1042
1043             /* Analytical PME correction */
1044             zeta2            = _mm_mul_ps(beta2,rsq00);
1045             rinv3            = _mm_mul_ps(rinvsq00,rinv00);
1046             pmecorrF         = gmx_mm_pmecorrF_ps(zeta2);
1047             felec            = _mm_macc_ps(pmecorrF,beta3,rinv3);
1048             felec            = _mm_mul_ps(qq00,felec);
1049
1050             /* Analytical LJ-PME */
1051             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
1052             ewcljrsq         = _mm_mul_ps(ewclj2,rsq00);
1053             ewclj6           = _mm_mul_ps(ewclj2,_mm_mul_ps(ewclj2,ewclj2));
1054             exponent         = gmx_simd_exp_r(ewcljrsq);
1055             /* poly = exp(-(beta*r)^2) * (1 + (beta*r)^2 + (beta*r)^4 /2) */
1056             poly             = _mm_mul_ps(exponent,_mm_macc_ps(_mm_mul_ps(ewcljrsq,ewcljrsq),one_half,_mm_sub_ps(one,ewcljrsq)));
1057             /* f6A = 6 * C6grid * (1 - poly) */
1058             f6A              = _mm_mul_ps(c6grid_00,_mm_sub_ps(one,poly));
1059             /* f6B = C6grid * exponent * beta^6 */
1060             f6B              = _mm_mul_ps(_mm_mul_ps(c6grid_00,one_sixth),_mm_mul_ps(exponent,ewclj6));
1061             /* fvdw = 12*C12/r13 - ((6*C6 - f6A)/r6 + f6B)/r */
1062             fvdw              = _mm_mul_ps(_mm_macc_ps(_mm_msub_ps(c12_00,rinvsix,_mm_sub_ps(c6_00,f6A)),rinvsix,f6B),rinvsq00);
1063
1064             fscal            = _mm_add_ps(felec,fvdw);
1065
1066             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1067
1068              /* Update vectorial force */
1069             fix0             = _mm_macc_ps(dx00,fscal,fix0);
1070             fiy0             = _mm_macc_ps(dy00,fscal,fiy0);
1071             fiz0             = _mm_macc_ps(dz00,fscal,fiz0);
1072
1073             fjx0             = _mm_macc_ps(dx00,fscal,fjx0);
1074             fjy0             = _mm_macc_ps(dy00,fscal,fjy0);
1075             fjz0             = _mm_macc_ps(dz00,fscal,fjz0);
1076
1077             /**************************
1078              * CALCULATE INTERACTIONS *
1079              **************************/
1080
1081             r10              = _mm_mul_ps(rsq10,rinv10);
1082             r10              = _mm_andnot_ps(dummy_mask,r10);
1083
1084             /* Compute parameters for interactions between i and j atoms */
1085             qq10             = _mm_mul_ps(iq1,jq0);
1086
1087             /* EWALD ELECTROSTATICS */
1088
1089             /* Analytical PME correction */
1090             zeta2            = _mm_mul_ps(beta2,rsq10);
1091             rinv3            = _mm_mul_ps(rinvsq10,rinv10);
1092             pmecorrF         = gmx_mm_pmecorrF_ps(zeta2);
1093             felec            = _mm_macc_ps(pmecorrF,beta3,rinv3);
1094             felec            = _mm_mul_ps(qq10,felec);
1095
1096             fscal            = felec;
1097
1098             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1099
1100              /* Update vectorial force */
1101             fix1             = _mm_macc_ps(dx10,fscal,fix1);
1102             fiy1             = _mm_macc_ps(dy10,fscal,fiy1);
1103             fiz1             = _mm_macc_ps(dz10,fscal,fiz1);
1104
1105             fjx0             = _mm_macc_ps(dx10,fscal,fjx0);
1106             fjy0             = _mm_macc_ps(dy10,fscal,fjy0);
1107             fjz0             = _mm_macc_ps(dz10,fscal,fjz0);
1108
1109             /**************************
1110              * CALCULATE INTERACTIONS *
1111              **************************/
1112
1113             r20              = _mm_mul_ps(rsq20,rinv20);
1114             r20              = _mm_andnot_ps(dummy_mask,r20);
1115
1116             /* Compute parameters for interactions between i and j atoms */
1117             qq20             = _mm_mul_ps(iq2,jq0);
1118
1119             /* EWALD ELECTROSTATICS */
1120
1121             /* Analytical PME correction */
1122             zeta2            = _mm_mul_ps(beta2,rsq20);
1123             rinv3            = _mm_mul_ps(rinvsq20,rinv20);
1124             pmecorrF         = gmx_mm_pmecorrF_ps(zeta2);
1125             felec            = _mm_macc_ps(pmecorrF,beta3,rinv3);
1126             felec            = _mm_mul_ps(qq20,felec);
1127
1128             fscal            = felec;
1129
1130             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1131
1132              /* Update vectorial force */
1133             fix2             = _mm_macc_ps(dx20,fscal,fix2);
1134             fiy2             = _mm_macc_ps(dy20,fscal,fiy2);
1135             fiz2             = _mm_macc_ps(dz20,fscal,fiz2);
1136
1137             fjx0             = _mm_macc_ps(dx20,fscal,fjx0);
1138             fjy0             = _mm_macc_ps(dy20,fscal,fjy0);
1139             fjz0             = _mm_macc_ps(dz20,fscal,fjz0);
1140
1141             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
1142             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
1143             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
1144             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
1145
1146             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
1147
1148             /* Inner loop uses 108 flops */
1149         }
1150
1151         /* End of innermost loop */
1152
1153         gmx_mm_update_iforce_3atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
1154                                               f+i_coord_offset,fshift+i_shift_offset);
1155
1156         /* Increment number of inner iterations */
1157         inneriter                  += j_index_end - j_index_start;
1158
1159         /* Outer loop uses 18 flops */
1160     }
1161
1162     /* Increment number of outer iterations */
1163     outeriter        += nri;
1164
1165     /* Update outer/inner flops */
1166
1167     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3_F,outeriter*18 + inneriter*108);
1168 }