src/gromacs/gmxlib/nonbonded/nb_kernel_avx_128_fma_single/nb_kernel_ElecEw_VdwLJEw_GeomW4P1_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 "gromacs/math/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_GeomW4P1_VF_avx_128_fma_single
  54  * Electrostatics interaction: Ewald
  55  * VdW interaction:            LJEwald
  56  * Geometry:                   Water4-Particle
  57  * Calculate force/pot:        PotentialAndForce
  58  */
  59 void
  60 nb_kernel_ElecEw_VdwLJEw_GeomW4P1_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              vdwioffset3;
  92     __m128           ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
  93     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
  94     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  95     __m128           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  96     __m128           dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
  97     __m128           dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
  98     __m128           dx30,dy30,dz30,rsq30,rinv30,rinvsq30,r30,qq30,c6_30,c12_30;
  99     __m128           velec,felec,velecsum,facel,crf,krf,krf2;
 100     real             *charge;
 101     int              nvdwtype;
 102     __m128           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 103     int              *vdwtype;
 104     real             *vdwparam;
 105     __m128           one_sixth   = _mm_set1_ps(1.0/6.0);
 106     __m128           one_twelfth = _mm_set1_ps(1.0/12.0);
 107     __m128           c6grid_00;
 108     __m128           c6grid_10;
 109     __m128           c6grid_20;
 110     __m128           c6grid_30;
 111     real             *vdwgridparam;
 112     __m128           ewclj,ewclj2,ewclj6,ewcljrsq,poly,exponent,f6A,f6B,sh_lj_ewald;
 113     __m128           one_half = _mm_set1_ps(0.5);
 114     __m128           minus_one = _mm_set1_ps(-1.0);
 115     __m128i          ewitab;
 116     __m128           ewtabscale,eweps,twoeweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
 117     __m128           beta,beta2,beta3,zeta2,pmecorrF,pmecorrV,rinv3;
 118     real             *ewtab;
 119     __m128           dummy_mask,cutoff_mask;
 120     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
 121     __m128           one     = _mm_set1_ps(1.0);
 122     __m128           two     = _mm_set1_ps(2.0);
 123     x                = xx[0];
 124     f                = ff[0];
 125
 126     nri              = nlist->nri;
 127     iinr             = nlist->iinr;
 128     jindex           = nlist->jindex;
 129     jjnr             = nlist->jjnr;
 130     shiftidx         = nlist->shift;
 131     gid              = nlist->gid;
 132     shiftvec         = fr->shift_vec[0];
 133     fshift           = fr->fshift[0];
 134     facel            = _mm_set1_ps(fr->epsfac);
 135     charge           = mdatoms->chargeA;
 136     nvdwtype         = fr->ntype;
 137     vdwparam         = fr->nbfp;
 138     vdwtype          = mdatoms->typeA;
 139     vdwgridparam     = fr->ljpme_c6grid;
 140     sh_lj_ewald      = _mm_set1_ps(fr->ic->sh_lj_ewald);
 141     ewclj            = _mm_set1_ps(fr->ewaldcoeff_lj);
 142     ewclj2           = _mm_mul_ps(minus_one,_mm_mul_ps(ewclj,ewclj));
 143
 144     sh_ewald         = _mm_set1_ps(fr->ic->sh_ewald);
 145     beta             = _mm_set1_ps(fr->ic->ewaldcoeff_q);
 146     beta2            = _mm_mul_ps(beta,beta);
 147     beta3            = _mm_mul_ps(beta,beta2);
 148     ewtab            = fr->ic->tabq_coul_FDV0;
 149     ewtabscale       = _mm_set1_ps(fr->ic->tabq_scale);
 150     ewtabhalfspace   = _mm_set1_ps(0.5/fr->ic->tabq_scale);
 151
 152     /* Setup water-specific parameters */
 153     inr              = nlist->iinr[0];
 154     iq1              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+1]));
 155     iq2              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+2]));
 156     iq3              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+3]));
 157     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 158
 159     /* Avoid stupid compiler warnings */
 160     jnrA = jnrB = jnrC = jnrD = 0;
 161     j_coord_offsetA = 0;
 162     j_coord_offsetB = 0;
 163     j_coord_offsetC = 0;
 164     j_coord_offsetD = 0;
 165
 166     outeriter        = 0;
 167     inneriter        = 0;
 168
 169     for(iidx=0;iidx<4*DIM;iidx++)
 170     {
 171         scratch[iidx] = 0.0;
 172     }
 173
 174     /* Start outer loop over neighborlists */
 175     for(iidx=0; iidx<nri; iidx++)
 176     {
 177         /* Load shift vector for this list */
 178         i_shift_offset   = DIM*shiftidx[iidx];
 179
 180         /* Load limits for loop over neighbors */
 181         j_index_start    = jindex[iidx];
 182         j_index_end      = jindex[iidx+1];
 183
 184         /* Get outer coordinate index */
 185         inr              = iinr[iidx];
 186         i_coord_offset   = DIM*inr;
 187
 188         /* Load i particle coords and add shift vector */
 189         gmx_mm_load_shift_and_4rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
 190                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
 191
 192         fix0             = _mm_setzero_ps();
 193         fiy0             = _mm_setzero_ps();
 194         fiz0             = _mm_setzero_ps();
 195         fix1             = _mm_setzero_ps();
 196         fiy1             = _mm_setzero_ps();
 197         fiz1             = _mm_setzero_ps();
 198         fix2             = _mm_setzero_ps();
 199         fiy2             = _mm_setzero_ps();
 200         fiz2             = _mm_setzero_ps();
 201         fix3             = _mm_setzero_ps();
 202         fiy3             = _mm_setzero_ps();
 203         fiz3             = _mm_setzero_ps();
 204
 205         /* Reset potential sums */
 206         velecsum         = _mm_setzero_ps();
 207         vvdwsum          = _mm_setzero_ps();
 208
 209         /* Start inner kernel loop */
 210         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 211         {
 212
 213             /* Get j neighbor index, and coordinate index */
 214             jnrA             = jjnr[jidx];
 215             jnrB             = jjnr[jidx+1];
 216             jnrC             = jjnr[jidx+2];
 217             jnrD             = jjnr[jidx+3];
 218             j_coord_offsetA  = DIM*jnrA;
 219             j_coord_offsetB  = DIM*jnrB;
 220             j_coord_offsetC  = DIM*jnrC;
 221             j_coord_offsetD  = DIM*jnrD;
 222
 223             /* load j atom coordinates */
 224             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 225                                               x+j_coord_offsetC,x+j_coord_offsetD,
 226                                               &jx0,&jy0,&jz0);
 227
 228             /* Calculate displacement vector */
 229             dx00             = _mm_sub_ps(ix0,jx0);
 230             dy00             = _mm_sub_ps(iy0,jy0);
 231             dz00             = _mm_sub_ps(iz0,jz0);
 232             dx10             = _mm_sub_ps(ix1,jx0);
 233             dy10             = _mm_sub_ps(iy1,jy0);
 234             dz10             = _mm_sub_ps(iz1,jz0);
 235             dx20             = _mm_sub_ps(ix2,jx0);
 236             dy20             = _mm_sub_ps(iy2,jy0);
 237             dz20             = _mm_sub_ps(iz2,jz0);
 238             dx30             = _mm_sub_ps(ix3,jx0);
 239             dy30             = _mm_sub_ps(iy3,jy0);
 240             dz30             = _mm_sub_ps(iz3,jz0);
 241
 242             /* Calculate squared distance and things based on it */
 243             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 244             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 245             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 246             rsq30            = gmx_mm_calc_rsq_ps(dx30,dy30,dz30);
 247
 248             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 249             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 250             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 251             rinv30           = gmx_mm_invsqrt_ps(rsq30);
 252
 253             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 254             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 255             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 256             rinvsq30         = _mm_mul_ps(rinv30,rinv30);
 257
 258             /* Load parameters for j particles */
 259             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 260                                                               charge+jnrC+0,charge+jnrD+0);
 261             vdwjidx0A        = 2*vdwtype[jnrA+0];
 262             vdwjidx0B        = 2*vdwtype[jnrB+0];
 263             vdwjidx0C        = 2*vdwtype[jnrC+0];
 264             vdwjidx0D        = 2*vdwtype[jnrD+0];
 265
 266             fjx0             = _mm_setzero_ps();
 267             fjy0             = _mm_setzero_ps();
 268             fjz0             = _mm_setzero_ps();
 269
 270             /**************************
 271              * CALCULATE INTERACTIONS *
 272              **************************/
 273
 274             r00              = _mm_mul_ps(rsq00,rinv00);
 275
 276             /* Compute parameters for interactions between i and j atoms */
 277             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 278                                          vdwparam+vdwioffset0+vdwjidx0B,
 279                                          vdwparam+vdwioffset0+vdwjidx0C,
 280                                          vdwparam+vdwioffset0+vdwjidx0D,
 281                                          &c6_00,&c12_00);
 282
 283             c6grid_00       = gmx_mm_load_4real_swizzle_ps(vdwgridparam+vdwioffset0+vdwjidx0A,
 284                                                                vdwgridparam+vdwioffset0+vdwjidx0B,
 285                                                                vdwgridparam+vdwioffset0+vdwjidx0C,
 286                                                                vdwgridparam+vdwioffset0+vdwjidx0D);
 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             vvdwsum          = _mm_add_ps(vvdwsum,vvdw);
 304
 305             fscal            = fvdw;
 306
 307              /* Update vectorial force */
 308             fix0             = _mm_macc_ps(dx00,fscal,fix0);
 309             fiy0             = _mm_macc_ps(dy00,fscal,fiy0);
 310             fiz0             = _mm_macc_ps(dz00,fscal,fiz0);
 311
 312             fjx0             = _mm_macc_ps(dx00,fscal,fjx0);
 313             fjy0             = _mm_macc_ps(dy00,fscal,fjy0);
 314             fjz0             = _mm_macc_ps(dz00,fscal,fjz0);
 315
 316             /**************************
 317              * CALCULATE INTERACTIONS *
 318              **************************/
 319
 320             r10              = _mm_mul_ps(rsq10,rinv10);
 321
 322             /* Compute parameters for interactions between i and j atoms */
 323             qq10             = _mm_mul_ps(iq1,jq0);
 324
 325             /* EWALD ELECTROSTATICS */
 326
 327             /* Analytical PME correction */
 328             zeta2            = _mm_mul_ps(beta2,rsq10);
 329             rinv3            = _mm_mul_ps(rinvsq10,rinv10);
 330             pmecorrF         = gmx_mm_pmecorrF_ps(zeta2);
 331             felec            = _mm_macc_ps(pmecorrF,beta3,rinv3);
 332             felec            = _mm_mul_ps(qq10,felec);
 333             pmecorrV         = gmx_mm_pmecorrV_ps(zeta2);
 334             velec            = _mm_nmacc_ps(pmecorrV,beta,rinv10);
 335             velec            = _mm_mul_ps(qq10,velec);
 336
 337             /* Update potential sum for this i atom from the interaction with this j atom. */
 338             velecsum         = _mm_add_ps(velecsum,velec);
 339
 340             fscal            = felec;
 341
 342              /* Update vectorial force */
 343             fix1             = _mm_macc_ps(dx10,fscal,fix1);
 344             fiy1             = _mm_macc_ps(dy10,fscal,fiy1);
 345             fiz1             = _mm_macc_ps(dz10,fscal,fiz1);
 346
 347             fjx0             = _mm_macc_ps(dx10,fscal,fjx0);
 348             fjy0             = _mm_macc_ps(dy10,fscal,fjy0);
 349             fjz0             = _mm_macc_ps(dz10,fscal,fjz0);
 350
 351             /**************************
 352              * CALCULATE INTERACTIONS *
 353              **************************/
 354
 355             r20              = _mm_mul_ps(rsq20,rinv20);
 356
 357             /* Compute parameters for interactions between i and j atoms */
 358             qq20             = _mm_mul_ps(iq2,jq0);
 359
 360             /* EWALD ELECTROSTATICS */
 361
 362             /* Analytical PME correction */
 363             zeta2            = _mm_mul_ps(beta2,rsq20);
 364             rinv3            = _mm_mul_ps(rinvsq20,rinv20);
 365             pmecorrF         = gmx_mm_pmecorrF_ps(zeta2);
 366             felec            = _mm_macc_ps(pmecorrF,beta3,rinv3);
 367             felec            = _mm_mul_ps(qq20,felec);
 368             pmecorrV         = gmx_mm_pmecorrV_ps(zeta2);
 369             velec            = _mm_nmacc_ps(pmecorrV,beta,rinv20);
 370             velec            = _mm_mul_ps(qq20,velec);
 371
 372             /* Update potential sum for this i atom from the interaction with this j atom. */
 373             velecsum         = _mm_add_ps(velecsum,velec);
 374
 375             fscal            = felec;
 376
 377              /* Update vectorial force */
 378             fix2             = _mm_macc_ps(dx20,fscal,fix2);
 379             fiy2             = _mm_macc_ps(dy20,fscal,fiy2);
 380             fiz2             = _mm_macc_ps(dz20,fscal,fiz2);
 381
 382             fjx0             = _mm_macc_ps(dx20,fscal,fjx0);
 383             fjy0             = _mm_macc_ps(dy20,fscal,fjy0);
 384             fjz0             = _mm_macc_ps(dz20,fscal,fjz0);
 385
 386             /**************************
 387              * CALCULATE INTERACTIONS *
 388              **************************/
 389
 390             r30              = _mm_mul_ps(rsq30,rinv30);
 391
 392             /* Compute parameters for interactions between i and j atoms */
 393             qq30             = _mm_mul_ps(iq3,jq0);
 394
 395             /* EWALD ELECTROSTATICS */
 396
 397             /* Analytical PME correction */
 398             zeta2            = _mm_mul_ps(beta2,rsq30);
 399             rinv3            = _mm_mul_ps(rinvsq30,rinv30);
 400             pmecorrF         = gmx_mm_pmecorrF_ps(zeta2);
 401             felec            = _mm_macc_ps(pmecorrF,beta3,rinv3);
 402             felec            = _mm_mul_ps(qq30,felec);
 403             pmecorrV         = gmx_mm_pmecorrV_ps(zeta2);
 404             velec            = _mm_nmacc_ps(pmecorrV,beta,rinv30);
 405             velec            = _mm_mul_ps(qq30,velec);
 406
 407             /* Update potential sum for this i atom from the interaction with this j atom. */
 408             velecsum         = _mm_add_ps(velecsum,velec);
 409
 410             fscal            = felec;
 411
 412              /* Update vectorial force */
 413             fix3             = _mm_macc_ps(dx30,fscal,fix3);
 414             fiy3             = _mm_macc_ps(dy30,fscal,fiy3);
 415             fiz3             = _mm_macc_ps(dz30,fscal,fiz3);
 416
 417             fjx0             = _mm_macc_ps(dx30,fscal,fjx0);
 418             fjy0             = _mm_macc_ps(dy30,fscal,fjy0);
 419             fjz0             = _mm_macc_ps(dz30,fscal,fjz0);
 420
 421             fjptrA             = f+j_coord_offsetA;
 422             fjptrB             = f+j_coord_offsetB;
 423             fjptrC             = f+j_coord_offsetC;
 424             fjptrD             = f+j_coord_offsetD;
 425
 426             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 427
 428             /* Inner loop uses 137 flops */
 429         }
 430
 431         if(jidx<j_index_end)
 432         {
 433
 434             /* Get j neighbor index, and coordinate index */
 435             jnrlistA         = jjnr[jidx];
 436             jnrlistB         = jjnr[jidx+1];
 437             jnrlistC         = jjnr[jidx+2];
 438             jnrlistD         = jjnr[jidx+3];
 439             /* Sign of each element will be negative for non-real atoms.
 440              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 441              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 442              */
 443             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 444             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 445             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 446             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 447             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 448             j_coord_offsetA  = DIM*jnrA;
 449             j_coord_offsetB  = DIM*jnrB;
 450             j_coord_offsetC  = DIM*jnrC;
 451             j_coord_offsetD  = DIM*jnrD;
 452
 453             /* load j atom coordinates */
 454             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 455                                               x+j_coord_offsetC,x+j_coord_offsetD,
 456                                               &jx0,&jy0,&jz0);
 457
 458             /* Calculate displacement vector */
 459             dx00             = _mm_sub_ps(ix0,jx0);
 460             dy00             = _mm_sub_ps(iy0,jy0);
 461             dz00             = _mm_sub_ps(iz0,jz0);
 462             dx10             = _mm_sub_ps(ix1,jx0);
 463             dy10             = _mm_sub_ps(iy1,jy0);
 464             dz10             = _mm_sub_ps(iz1,jz0);
 465             dx20             = _mm_sub_ps(ix2,jx0);
 466             dy20             = _mm_sub_ps(iy2,jy0);
 467             dz20             = _mm_sub_ps(iz2,jz0);
 468             dx30             = _mm_sub_ps(ix3,jx0);
 469             dy30             = _mm_sub_ps(iy3,jy0);
 470             dz30             = _mm_sub_ps(iz3,jz0);
 471
 472             /* Calculate squared distance and things based on it */
 473             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 474             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 475             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 476             rsq30            = gmx_mm_calc_rsq_ps(dx30,dy30,dz30);
 477
 478             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 479             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 480             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 481             rinv30           = gmx_mm_invsqrt_ps(rsq30);
 482
 483             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 484             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 485             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 486             rinvsq30         = _mm_mul_ps(rinv30,rinv30);
 487
 488             /* Load parameters for j particles */
 489             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 490                                                               charge+jnrC+0,charge+jnrD+0);
 491             vdwjidx0A        = 2*vdwtype[jnrA+0];
 492             vdwjidx0B        = 2*vdwtype[jnrB+0];
 493             vdwjidx0C        = 2*vdwtype[jnrC+0];
 494             vdwjidx0D        = 2*vdwtype[jnrD+0];
 495
 496             fjx0             = _mm_setzero_ps();
 497             fjy0             = _mm_setzero_ps();
 498             fjz0             = _mm_setzero_ps();
 499
 500             /**************************
 501              * CALCULATE INTERACTIONS *
 502              **************************/
 503
 504             r00              = _mm_mul_ps(rsq00,rinv00);
 505             r00              = _mm_andnot_ps(dummy_mask,r00);
 506
 507             /* Compute parameters for interactions between i and j atoms */
 508             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 509                                          vdwparam+vdwioffset0+vdwjidx0B,
 510                                          vdwparam+vdwioffset0+vdwjidx0C,
 511                                          vdwparam+vdwioffset0+vdwjidx0D,
 512                                          &c6_00,&c12_00);
 513
 514             c6grid_00       = gmx_mm_load_4real_swizzle_ps(vdwgridparam+vdwioffset0+vdwjidx0A,
 515                                                                vdwgridparam+vdwioffset0+vdwjidx0B,
 516                                                                vdwgridparam+vdwioffset0+vdwjidx0C,
 517                                                                vdwgridparam+vdwioffset0+vdwjidx0D);
 518
 519             /* Analytical LJ-PME */
 520             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 521             ewcljrsq         = _mm_mul_ps(ewclj2,rsq00);
 522             ewclj6           = _mm_mul_ps(ewclj2,_mm_mul_ps(ewclj2,ewclj2));
 523             exponent         = gmx_simd_exp_r(ewcljrsq);
 524             /* poly = exp(-(beta*r)^2) * (1 + (beta*r)^2 + (beta*r)^4 /2) */
 525             poly             = _mm_mul_ps(exponent,_mm_macc_ps(_mm_mul_ps(ewcljrsq,ewcljrsq),one_half,_mm_sub_ps(one,ewcljrsq)));
 526             /* vvdw6 = [C6 - C6grid * (1-poly)]/r6 */
 527             vvdw6            = _mm_mul_ps(_mm_macc_ps(-c6grid_00,_mm_sub_ps(one,poly),c6_00),rinvsix);
 528             vvdw12           = _mm_mul_ps(c12_00,_mm_mul_ps(rinvsix,rinvsix));
 529             vvdw             = _mm_msub_ps(vvdw12,one_twelfth,_mm_mul_ps(vvdw6,one_sixth));
 530             /* fvdw = vvdw12/r - (vvdw6/r + (C6grid * exponent * beta^6)/r) */
 531             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);
 532
 533             /* Update potential sum for this i atom from the interaction with this j atom. */
 534             vvdw             = _mm_andnot_ps(dummy_mask,vvdw);
 535             vvdwsum          = _mm_add_ps(vvdwsum,vvdw);
 536
 537             fscal            = fvdw;
 538
 539             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 540
 541              /* Update vectorial force */
 542             fix0             = _mm_macc_ps(dx00,fscal,fix0);
 543             fiy0             = _mm_macc_ps(dy00,fscal,fiy0);
 544             fiz0             = _mm_macc_ps(dz00,fscal,fiz0);
 545
 546             fjx0             = _mm_macc_ps(dx00,fscal,fjx0);
 547             fjy0             = _mm_macc_ps(dy00,fscal,fjy0);
 548             fjz0             = _mm_macc_ps(dz00,fscal,fjz0);
 549
 550             /**************************
 551              * CALCULATE INTERACTIONS *
 552              **************************/
 553
 554             r10              = _mm_mul_ps(rsq10,rinv10);
 555             r10              = _mm_andnot_ps(dummy_mask,r10);
 556
 557             /* Compute parameters for interactions between i and j atoms */
 558             qq10             = _mm_mul_ps(iq1,jq0);
 559
 560             /* EWALD ELECTROSTATICS */
 561
 562             /* Analytical PME correction */
 563             zeta2            = _mm_mul_ps(beta2,rsq10);
 564             rinv3            = _mm_mul_ps(rinvsq10,rinv10);
 565             pmecorrF         = gmx_mm_pmecorrF_ps(zeta2);
 566             felec            = _mm_macc_ps(pmecorrF,beta3,rinv3);
 567             felec            = _mm_mul_ps(qq10,felec);
 568             pmecorrV         = gmx_mm_pmecorrV_ps(zeta2);
 569             velec            = _mm_nmacc_ps(pmecorrV,beta,rinv10);
 570             velec            = _mm_mul_ps(qq10,velec);
 571
 572             /* Update potential sum for this i atom from the interaction with this j atom. */
 573             velec            = _mm_andnot_ps(dummy_mask,velec);
 574             velecsum         = _mm_add_ps(velecsum,velec);
 575
 576             fscal            = felec;
 577
 578             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 579
 580              /* Update vectorial force */
 581             fix1             = _mm_macc_ps(dx10,fscal,fix1);
 582             fiy1             = _mm_macc_ps(dy10,fscal,fiy1);
 583             fiz1             = _mm_macc_ps(dz10,fscal,fiz1);
 584
 585             fjx0             = _mm_macc_ps(dx10,fscal,fjx0);
 586             fjy0             = _mm_macc_ps(dy10,fscal,fjy0);
 587             fjz0             = _mm_macc_ps(dz10,fscal,fjz0);
 588
 589             /**************************
 590              * CALCULATE INTERACTIONS *
 591              **************************/
 592
 593             r20              = _mm_mul_ps(rsq20,rinv20);
 594             r20              = _mm_andnot_ps(dummy_mask,r20);
 595
 596             /* Compute parameters for interactions between i and j atoms */
 597             qq20             = _mm_mul_ps(iq2,jq0);
 598
 599             /* EWALD ELECTROSTATICS */
 600
 601             /* Analytical PME correction */
 602             zeta2            = _mm_mul_ps(beta2,rsq20);
 603             rinv3            = _mm_mul_ps(rinvsq20,rinv20);
 604             pmecorrF         = gmx_mm_pmecorrF_ps(zeta2);
 605             felec            = _mm_macc_ps(pmecorrF,beta3,rinv3);
 606             felec            = _mm_mul_ps(qq20,felec);
 607             pmecorrV         = gmx_mm_pmecorrV_ps(zeta2);
 608             velec            = _mm_nmacc_ps(pmecorrV,beta,rinv20);
 609             velec            = _mm_mul_ps(qq20,velec);
 610
 611             /* Update potential sum for this i atom from the interaction with this j atom. */
 612             velec            = _mm_andnot_ps(dummy_mask,velec);
 613             velecsum         = _mm_add_ps(velecsum,velec);
 614
 615             fscal            = felec;
 616
 617             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 618
 619              /* Update vectorial force */
 620             fix2             = _mm_macc_ps(dx20,fscal,fix2);
 621             fiy2             = _mm_macc_ps(dy20,fscal,fiy2);
 622             fiz2             = _mm_macc_ps(dz20,fscal,fiz2);
 623
 624             fjx0             = _mm_macc_ps(dx20,fscal,fjx0);
 625             fjy0             = _mm_macc_ps(dy20,fscal,fjy0);
 626             fjz0             = _mm_macc_ps(dz20,fscal,fjz0);
 627
 628             /**************************
 629              * CALCULATE INTERACTIONS *
 630              **************************/
 631
 632             r30              = _mm_mul_ps(rsq30,rinv30);
 633             r30              = _mm_andnot_ps(dummy_mask,r30);
 634
 635             /* Compute parameters for interactions between i and j atoms */
 636             qq30             = _mm_mul_ps(iq3,jq0);
 637
 638             /* EWALD ELECTROSTATICS */
 639
 640             /* Analytical PME correction */
 641             zeta2            = _mm_mul_ps(beta2,rsq30);
 642             rinv3            = _mm_mul_ps(rinvsq30,rinv30);
 643             pmecorrF         = gmx_mm_pmecorrF_ps(zeta2);
 644             felec            = _mm_macc_ps(pmecorrF,beta3,rinv3);
 645             felec            = _mm_mul_ps(qq30,felec);
 646             pmecorrV         = gmx_mm_pmecorrV_ps(zeta2);
 647             velec            = _mm_nmacc_ps(pmecorrV,beta,rinv30);
 648             velec            = _mm_mul_ps(qq30,velec);
 649
 650             /* Update potential sum for this i atom from the interaction with this j atom. */
 651             velec            = _mm_andnot_ps(dummy_mask,velec);
 652             velecsum         = _mm_add_ps(velecsum,velec);
 653
 654             fscal            = felec;
 655
 656             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 657
 658              /* Update vectorial force */
 659             fix3             = _mm_macc_ps(dx30,fscal,fix3);
 660             fiy3             = _mm_macc_ps(dy30,fscal,fiy3);
 661             fiz3             = _mm_macc_ps(dz30,fscal,fiz3);
 662
 663             fjx0             = _mm_macc_ps(dx30,fscal,fjx0);
 664             fjy0             = _mm_macc_ps(dy30,fscal,fjy0);
 665             fjz0             = _mm_macc_ps(dz30,fscal,fjz0);
 666
 667             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 668             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 669             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 670             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 671
 672             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 673
 674             /* Inner loop uses 141 flops */
 675         }
 676
 677         /* End of innermost loop */
 678
 679         gmx_mm_update_iforce_4atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
 680                                               f+i_coord_offset,fshift+i_shift_offset);
 681
 682         ggid                        = gid[iidx];
 683         /* Update potential energies */
 684         gmx_mm_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid);
 685         gmx_mm_update_1pot_ps(vvdwsum,kernel_data->energygrp_vdw+ggid);
 686
 687         /* Increment number of inner iterations */
 688         inneriter                  += j_index_end - j_index_start;
 689
 690         /* Outer loop uses 26 flops */
 691     }
 692
 693     /* Increment number of outer iterations */
 694     outeriter        += nri;
 695
 696     /* Update outer/inner flops */
 697
 698     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W4_VF,outeriter*26 + inneriter*141);
 699 }
 700 /*
 701  * Gromacs nonbonded kernel:   nb_kernel_ElecEw_VdwLJEw_GeomW4P1_F_avx_128_fma_single
 702  * Electrostatics interaction: Ewald
 703  * VdW interaction:            LJEwald
 704  * Geometry:                   Water4-Particle
 705  * Calculate force/pot:        Force
 706  */
 707 void
 708 nb_kernel_ElecEw_VdwLJEw_GeomW4P1_F_avx_128_fma_single
 709                     (t_nblist                    * gmx_restrict       nlist,
 710                      rvec                        * gmx_restrict          xx,
 711                      rvec                        * gmx_restrict          ff,
 712                      t_forcerec                  * gmx_restrict          fr,
 713                      t_mdatoms                   * gmx_restrict     mdatoms,
 714                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 715                      t_nrnb                      * gmx_restrict        nrnb)
 716 {
 717     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 718      * just 0 for non-waters.
 719      * Suffixes A,B,C,D refer to j loop unrolling done with AVX_128, e.g. for the four different
 720      * jnr indices corresponding to data put in the four positions in the SIMD register.
 721      */
 722     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 723     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 724     int              jnrA,jnrB,jnrC,jnrD;
 725     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
 726     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 727     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 728     real             rcutoff_scalar;
 729     real             *shiftvec,*fshift,*x,*f;
 730     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
 731     real             scratch[4*DIM];
 732     __m128           fscal,rcutoff,rcutoff2,jidxall;
 733     int              vdwioffset0;
 734     __m128           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 735     int              vdwioffset1;
 736     __m128           ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
 737     int              vdwioffset2;
 738     __m128           ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
 739     int              vdwioffset3;
 740     __m128           ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
 741     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
 742     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 743     __m128           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 744     __m128           dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
 745     __m128           dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
 746     __m128           dx30,dy30,dz30,rsq30,rinv30,rinvsq30,r30,qq30,c6_30,c12_30;
 747     __m128           velec,felec,velecsum,facel,crf,krf,krf2;
 748     real             *charge;
 749     int              nvdwtype;
 750     __m128           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 751     int              *vdwtype;
 752     real             *vdwparam;
 753     __m128           one_sixth   = _mm_set1_ps(1.0/6.0);
 754     __m128           one_twelfth = _mm_set1_ps(1.0/12.0);
 755     __m128           c6grid_00;
 756     __m128           c6grid_10;
 757     __m128           c6grid_20;
 758     __m128           c6grid_30;
 759     real             *vdwgridparam;
 760     __m128           ewclj,ewclj2,ewclj6,ewcljrsq,poly,exponent,f6A,f6B,sh_lj_ewald;
 761     __m128           one_half = _mm_set1_ps(0.5);
 762     __m128           minus_one = _mm_set1_ps(-1.0);
 763     __m128i          ewitab;
 764     __m128           ewtabscale,eweps,twoeweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
 765     __m128           beta,beta2,beta3,zeta2,pmecorrF,pmecorrV,rinv3;
 766     real             *ewtab;
 767     __m128           dummy_mask,cutoff_mask;
 768     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
 769     __m128           one     = _mm_set1_ps(1.0);
 770     __m128           two     = _mm_set1_ps(2.0);
 771     x                = xx[0];
 772     f                = ff[0];
 773
 774     nri              = nlist->nri;
 775     iinr             = nlist->iinr;
 776     jindex           = nlist->jindex;
 777     jjnr             = nlist->jjnr;
 778     shiftidx         = nlist->shift;
 779     gid              = nlist->gid;
 780     shiftvec         = fr->shift_vec[0];
 781     fshift           = fr->fshift[0];
 782     facel            = _mm_set1_ps(fr->epsfac);
 783     charge           = mdatoms->chargeA;
 784     nvdwtype         = fr->ntype;
 785     vdwparam         = fr->nbfp;
 786     vdwtype          = mdatoms->typeA;
 787     vdwgridparam     = fr->ljpme_c6grid;
 788     sh_lj_ewald      = _mm_set1_ps(fr->ic->sh_lj_ewald);
 789     ewclj            = _mm_set1_ps(fr->ewaldcoeff_lj);
 790     ewclj2           = _mm_mul_ps(minus_one,_mm_mul_ps(ewclj,ewclj));
 791
 792     sh_ewald         = _mm_set1_ps(fr->ic->sh_ewald);
 793     beta             = _mm_set1_ps(fr->ic->ewaldcoeff_q);
 794     beta2            = _mm_mul_ps(beta,beta);
 795     beta3            = _mm_mul_ps(beta,beta2);
 796     ewtab            = fr->ic->tabq_coul_F;
 797     ewtabscale       = _mm_set1_ps(fr->ic->tabq_scale);
 798     ewtabhalfspace   = _mm_set1_ps(0.5/fr->ic->tabq_scale);
 799
 800     /* Setup water-specific parameters */
 801     inr              = nlist->iinr[0];
 802     iq1              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+1]));
 803     iq2              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+2]));
 804     iq3              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+3]));
 805     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 806
 807     /* Avoid stupid compiler warnings */
 808     jnrA = jnrB = jnrC = jnrD = 0;
 809     j_coord_offsetA = 0;
 810     j_coord_offsetB = 0;
 811     j_coord_offsetC = 0;
 812     j_coord_offsetD = 0;
 813
 814     outeriter        = 0;
 815     inneriter        = 0;
 816
 817     for(iidx=0;iidx<4*DIM;iidx++)
 818     {
 819         scratch[iidx] = 0.0;
 820     }
 821
 822     /* Start outer loop over neighborlists */
 823     for(iidx=0; iidx<nri; iidx++)
 824     {
 825         /* Load shift vector for this list */
 826         i_shift_offset   = DIM*shiftidx[iidx];
 827
 828         /* Load limits for loop over neighbors */
 829         j_index_start    = jindex[iidx];
 830         j_index_end      = jindex[iidx+1];
 831
 832         /* Get outer coordinate index */
 833         inr              = iinr[iidx];
 834         i_coord_offset   = DIM*inr;
 835
 836         /* Load i particle coords and add shift vector */
 837         gmx_mm_load_shift_and_4rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
 838                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
 839
 840         fix0             = _mm_setzero_ps();
 841         fiy0             = _mm_setzero_ps();
 842         fiz0             = _mm_setzero_ps();
 843         fix1             = _mm_setzero_ps();
 844         fiy1             = _mm_setzero_ps();
 845         fiz1             = _mm_setzero_ps();
 846         fix2             = _mm_setzero_ps();
 847         fiy2             = _mm_setzero_ps();
 848         fiz2             = _mm_setzero_ps();
 849         fix3             = _mm_setzero_ps();
 850         fiy3             = _mm_setzero_ps();
 851         fiz3             = _mm_setzero_ps();
 852
 853         /* Start inner kernel loop */
 854         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 855         {
 856
 857             /* Get j neighbor index, and coordinate index */
 858             jnrA             = jjnr[jidx];
 859             jnrB             = jjnr[jidx+1];
 860             jnrC             = jjnr[jidx+2];
 861             jnrD             = jjnr[jidx+3];
 862             j_coord_offsetA  = DIM*jnrA;
 863             j_coord_offsetB  = DIM*jnrB;
 864             j_coord_offsetC  = DIM*jnrC;
 865             j_coord_offsetD  = DIM*jnrD;
 866
 867             /* load j atom coordinates */
 868             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 869                                               x+j_coord_offsetC,x+j_coord_offsetD,
 870                                               &jx0,&jy0,&jz0);
 871
 872             /* Calculate displacement vector */
 873             dx00             = _mm_sub_ps(ix0,jx0);
 874             dy00             = _mm_sub_ps(iy0,jy0);
 875             dz00             = _mm_sub_ps(iz0,jz0);
 876             dx10             = _mm_sub_ps(ix1,jx0);
 877             dy10             = _mm_sub_ps(iy1,jy0);
 878             dz10             = _mm_sub_ps(iz1,jz0);
 879             dx20             = _mm_sub_ps(ix2,jx0);
 880             dy20             = _mm_sub_ps(iy2,jy0);
 881             dz20             = _mm_sub_ps(iz2,jz0);
 882             dx30             = _mm_sub_ps(ix3,jx0);
 883             dy30             = _mm_sub_ps(iy3,jy0);
 884             dz30             = _mm_sub_ps(iz3,jz0);
 885
 886             /* Calculate squared distance and things based on it */
 887             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 888             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 889             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 890             rsq30            = gmx_mm_calc_rsq_ps(dx30,dy30,dz30);
 891
 892             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 893             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 894             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 895             rinv30           = gmx_mm_invsqrt_ps(rsq30);
 896
 897             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 898             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 899             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 900             rinvsq30         = _mm_mul_ps(rinv30,rinv30);
 901
 902             /* Load parameters for j particles */
 903             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 904                                                               charge+jnrC+0,charge+jnrD+0);
 905             vdwjidx0A        = 2*vdwtype[jnrA+0];
 906             vdwjidx0B        = 2*vdwtype[jnrB+0];
 907             vdwjidx0C        = 2*vdwtype[jnrC+0];
 908             vdwjidx0D        = 2*vdwtype[jnrD+0];
 909
 910             fjx0             = _mm_setzero_ps();
 911             fjy0             = _mm_setzero_ps();
 912             fjz0             = _mm_setzero_ps();
 913
 914             /**************************
 915              * CALCULATE INTERACTIONS *
 916              **************************/
 917
 918             r00              = _mm_mul_ps(rsq00,rinv00);
 919
 920             /* Compute parameters for interactions between i and j atoms */
 921             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 922                                          vdwparam+vdwioffset0+vdwjidx0B,
 923                                          vdwparam+vdwioffset0+vdwjidx0C,
 924                                          vdwparam+vdwioffset0+vdwjidx0D,
 925                                          &c6_00,&c12_00);
 926
 927             c6grid_00       = gmx_mm_load_4real_swizzle_ps(vdwgridparam+vdwioffset0+vdwjidx0A,
 928                                                                vdwgridparam+vdwioffset0+vdwjidx0B,
 929                                                                vdwgridparam+vdwioffset0+vdwjidx0C,
 930                                                                vdwgridparam+vdwioffset0+vdwjidx0D);
 931
 932             /* Analytical LJ-PME */
 933             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 934             ewcljrsq         = _mm_mul_ps(ewclj2,rsq00);
 935             ewclj6           = _mm_mul_ps(ewclj2,_mm_mul_ps(ewclj2,ewclj2));
 936             exponent         = gmx_simd_exp_r(ewcljrsq);
 937             /* poly = exp(-(beta*r)^2) * (1 + (beta*r)^2 + (beta*r)^4 /2) */
 938             poly             = _mm_mul_ps(exponent,_mm_macc_ps(_mm_mul_ps(ewcljrsq,ewcljrsq),one_half,_mm_sub_ps(one,ewcljrsq)));
 939             /* f6A = 6 * C6grid * (1 - poly) */
 940             f6A              = _mm_mul_ps(c6grid_00,_mm_sub_ps(one,poly));
 941             /* f6B = C6grid * exponent * beta^6 */
 942             f6B              = _mm_mul_ps(_mm_mul_ps(c6grid_00,one_sixth),_mm_mul_ps(exponent,ewclj6));
 943             /* fvdw = 12*C12/r13 - ((6*C6 - f6A)/r6 + f6B)/r */
 944             fvdw              = _mm_mul_ps(_mm_macc_ps(_mm_msub_ps(c12_00,rinvsix,_mm_sub_ps(c6_00,f6A)),rinvsix,f6B),rinvsq00);
 945
 946             fscal            = fvdw;
 947
 948              /* Update vectorial force */
 949             fix0             = _mm_macc_ps(dx00,fscal,fix0);
 950             fiy0             = _mm_macc_ps(dy00,fscal,fiy0);
 951             fiz0             = _mm_macc_ps(dz00,fscal,fiz0);
 952
 953             fjx0             = _mm_macc_ps(dx00,fscal,fjx0);
 954             fjy0             = _mm_macc_ps(dy00,fscal,fjy0);
 955             fjz0             = _mm_macc_ps(dz00,fscal,fjz0);
 956
 957             /**************************
 958              * CALCULATE INTERACTIONS *
 959              **************************/
 960
 961             r10              = _mm_mul_ps(rsq10,rinv10);
 962
 963             /* Compute parameters for interactions between i and j atoms */
 964             qq10             = _mm_mul_ps(iq1,jq0);
 965
 966             /* EWALD ELECTROSTATICS */
 967
 968             /* Analytical PME correction */
 969             zeta2            = _mm_mul_ps(beta2,rsq10);
 970             rinv3            = _mm_mul_ps(rinvsq10,rinv10);
 971             pmecorrF         = gmx_mm_pmecorrF_ps(zeta2);
 972             felec            = _mm_macc_ps(pmecorrF,beta3,rinv3);
 973             felec            = _mm_mul_ps(qq10,felec);
 974
 975             fscal            = felec;
 976
 977              /* Update vectorial force */
 978             fix1             = _mm_macc_ps(dx10,fscal,fix1);
 979             fiy1             = _mm_macc_ps(dy10,fscal,fiy1);
 980             fiz1             = _mm_macc_ps(dz10,fscal,fiz1);
 981
 982             fjx0             = _mm_macc_ps(dx10,fscal,fjx0);
 983             fjy0             = _mm_macc_ps(dy10,fscal,fjy0);
 984             fjz0             = _mm_macc_ps(dz10,fscal,fjz0);
 985
 986             /**************************
 987              * CALCULATE INTERACTIONS *
 988              **************************/
 989
 990             r20              = _mm_mul_ps(rsq20,rinv20);
 991
 992             /* Compute parameters for interactions between i and j atoms */
 993             qq20             = _mm_mul_ps(iq2,jq0);
 994
 995             /* EWALD ELECTROSTATICS */
 996
 997             /* Analytical PME correction */
 998             zeta2            = _mm_mul_ps(beta2,rsq20);
 999             rinv3            = _mm_mul_ps(rinvsq20,rinv20);
1000             pmecorrF         = gmx_mm_pmecorrF_ps(zeta2);
1001             felec            = _mm_macc_ps(pmecorrF,beta3,rinv3);
1002             felec            = _mm_mul_ps(qq20,felec);
1003
1004             fscal            = felec;
1005
1006              /* Update vectorial force */
1007             fix2             = _mm_macc_ps(dx20,fscal,fix2);
1008             fiy2             = _mm_macc_ps(dy20,fscal,fiy2);
1009             fiz2             = _mm_macc_ps(dz20,fscal,fiz2);
1010
1011             fjx0             = _mm_macc_ps(dx20,fscal,fjx0);
1012             fjy0             = _mm_macc_ps(dy20,fscal,fjy0);
1013             fjz0             = _mm_macc_ps(dz20,fscal,fjz0);
1014
1015             /**************************
1016              * CALCULATE INTERACTIONS *
1017              **************************/
1018
1019             r30              = _mm_mul_ps(rsq30,rinv30);
1020
1021             /* Compute parameters for interactions between i and j atoms */
1022             qq30             = _mm_mul_ps(iq3,jq0);
1023
1024             /* EWALD ELECTROSTATICS */
1025
1026             /* Analytical PME correction */
1027             zeta2            = _mm_mul_ps(beta2,rsq30);
1028             rinv3            = _mm_mul_ps(rinvsq30,rinv30);
1029             pmecorrF         = gmx_mm_pmecorrF_ps(zeta2);
1030             felec            = _mm_macc_ps(pmecorrF,beta3,rinv3);
1031             felec            = _mm_mul_ps(qq30,felec);
1032
1033             fscal            = felec;
1034
1035              /* Update vectorial force */
1036             fix3             = _mm_macc_ps(dx30,fscal,fix3);
1037             fiy3             = _mm_macc_ps(dy30,fscal,fiy3);
1038             fiz3             = _mm_macc_ps(dz30,fscal,fiz3);
1039
1040             fjx0             = _mm_macc_ps(dx30,fscal,fjx0);
1041             fjy0             = _mm_macc_ps(dy30,fscal,fjy0);
1042             fjz0             = _mm_macc_ps(dz30,fscal,fjz0);
1043
1044             fjptrA             = f+j_coord_offsetA;
1045             fjptrB             = f+j_coord_offsetB;
1046             fjptrC             = f+j_coord_offsetC;
1047             fjptrD             = f+j_coord_offsetD;
1048
1049             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
1050
1051             /* Inner loop uses 131 flops */
1052         }
1053
1054         if(jidx<j_index_end)
1055         {
1056
1057             /* Get j neighbor index, and coordinate index */
1058             jnrlistA         = jjnr[jidx];
1059             jnrlistB         = jjnr[jidx+1];
1060             jnrlistC         = jjnr[jidx+2];
1061             jnrlistD         = jjnr[jidx+3];
1062             /* Sign of each element will be negative for non-real atoms.
1063              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
1064              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
1065              */
1066             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
1067             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
1068             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
1069             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
1070             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
1071             j_coord_offsetA  = DIM*jnrA;
1072             j_coord_offsetB  = DIM*jnrB;
1073             j_coord_offsetC  = DIM*jnrC;
1074             j_coord_offsetD  = DIM*jnrD;
1075
1076             /* load j atom coordinates */
1077             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
1078                                               x+j_coord_offsetC,x+j_coord_offsetD,
1079                                               &jx0,&jy0,&jz0);
1080
1081             /* Calculate displacement vector */
1082             dx00             = _mm_sub_ps(ix0,jx0);
1083             dy00             = _mm_sub_ps(iy0,jy0);
1084             dz00             = _mm_sub_ps(iz0,jz0);
1085             dx10             = _mm_sub_ps(ix1,jx0);
1086             dy10             = _mm_sub_ps(iy1,jy0);
1087             dz10             = _mm_sub_ps(iz1,jz0);
1088             dx20             = _mm_sub_ps(ix2,jx0);
1089             dy20             = _mm_sub_ps(iy2,jy0);
1090             dz20             = _mm_sub_ps(iz2,jz0);
1091             dx30             = _mm_sub_ps(ix3,jx0);
1092             dy30             = _mm_sub_ps(iy3,jy0);
1093             dz30             = _mm_sub_ps(iz3,jz0);
1094
1095             /* Calculate squared distance and things based on it */
1096             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
1097             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
1098             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
1099             rsq30            = gmx_mm_calc_rsq_ps(dx30,dy30,dz30);
1100
1101             rinv00           = gmx_mm_invsqrt_ps(rsq00);
1102             rinv10           = gmx_mm_invsqrt_ps(rsq10);
1103             rinv20           = gmx_mm_invsqrt_ps(rsq20);
1104             rinv30           = gmx_mm_invsqrt_ps(rsq30);
1105
1106             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
1107             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
1108             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
1109             rinvsq30         = _mm_mul_ps(rinv30,rinv30);
1110
1111             /* Load parameters for j particles */
1112             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
1113                                                               charge+jnrC+0,charge+jnrD+0);
1114             vdwjidx0A        = 2*vdwtype[jnrA+0];
1115             vdwjidx0B        = 2*vdwtype[jnrB+0];
1116             vdwjidx0C        = 2*vdwtype[jnrC+0];
1117             vdwjidx0D        = 2*vdwtype[jnrD+0];
1118
1119             fjx0             = _mm_setzero_ps();
1120             fjy0             = _mm_setzero_ps();
1121             fjz0             = _mm_setzero_ps();
1122
1123             /**************************
1124              * CALCULATE INTERACTIONS *
1125              **************************/
1126
1127             r00              = _mm_mul_ps(rsq00,rinv00);
1128             r00              = _mm_andnot_ps(dummy_mask,r00);
1129
1130             /* Compute parameters for interactions between i and j atoms */
1131             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
1132                                          vdwparam+vdwioffset0+vdwjidx0B,
1133                                          vdwparam+vdwioffset0+vdwjidx0C,
1134                                          vdwparam+vdwioffset0+vdwjidx0D,
1135                                          &c6_00,&c12_00);
1136
1137             c6grid_00       = gmx_mm_load_4real_swizzle_ps(vdwgridparam+vdwioffset0+vdwjidx0A,
1138                                                                vdwgridparam+vdwioffset0+vdwjidx0B,
1139                                                                vdwgridparam+vdwioffset0+vdwjidx0C,
1140                                                                vdwgridparam+vdwioffset0+vdwjidx0D);
1141
1142             /* Analytical LJ-PME */
1143             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
1144             ewcljrsq         = _mm_mul_ps(ewclj2,rsq00);
1145             ewclj6           = _mm_mul_ps(ewclj2,_mm_mul_ps(ewclj2,ewclj2));
1146             exponent         = gmx_simd_exp_r(ewcljrsq);
1147             /* poly = exp(-(beta*r)^2) * (1 + (beta*r)^2 + (beta*r)^4 /2) */
1148             poly             = _mm_mul_ps(exponent,_mm_macc_ps(_mm_mul_ps(ewcljrsq,ewcljrsq),one_half,_mm_sub_ps(one,ewcljrsq)));
1149             /* f6A = 6 * C6grid * (1 - poly) */
1150             f6A              = _mm_mul_ps(c6grid_00,_mm_sub_ps(one,poly));
1151             /* f6B = C6grid * exponent * beta^6 */
1152             f6B              = _mm_mul_ps(_mm_mul_ps(c6grid_00,one_sixth),_mm_mul_ps(exponent,ewclj6));
1153             /* fvdw = 12*C12/r13 - ((6*C6 - f6A)/r6 + f6B)/r */
1154             fvdw              = _mm_mul_ps(_mm_macc_ps(_mm_msub_ps(c12_00,rinvsix,_mm_sub_ps(c6_00,f6A)),rinvsix,f6B),rinvsq00);
1155
1156             fscal            = fvdw;
1157
1158             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1159
1160              /* Update vectorial force */
1161             fix0             = _mm_macc_ps(dx00,fscal,fix0);
1162             fiy0             = _mm_macc_ps(dy00,fscal,fiy0);
1163             fiz0             = _mm_macc_ps(dz00,fscal,fiz0);
1164
1165             fjx0             = _mm_macc_ps(dx00,fscal,fjx0);
1166             fjy0             = _mm_macc_ps(dy00,fscal,fjy0);
1167             fjz0             = _mm_macc_ps(dz00,fscal,fjz0);
1168
1169             /**************************
1170              * CALCULATE INTERACTIONS *
1171              **************************/
1172
1173             r10              = _mm_mul_ps(rsq10,rinv10);
1174             r10              = _mm_andnot_ps(dummy_mask,r10);
1175
1176             /* Compute parameters for interactions between i and j atoms */
1177             qq10             = _mm_mul_ps(iq1,jq0);
1178
1179             /* EWALD ELECTROSTATICS */
1180
1181             /* Analytical PME correction */
1182             zeta2            = _mm_mul_ps(beta2,rsq10);
1183             rinv3            = _mm_mul_ps(rinvsq10,rinv10);
1184             pmecorrF         = gmx_mm_pmecorrF_ps(zeta2);
1185             felec            = _mm_macc_ps(pmecorrF,beta3,rinv3);
1186             felec            = _mm_mul_ps(qq10,felec);
1187
1188             fscal            = felec;
1189
1190             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1191
1192              /* Update vectorial force */
1193             fix1             = _mm_macc_ps(dx10,fscal,fix1);
1194             fiy1             = _mm_macc_ps(dy10,fscal,fiy1);
1195             fiz1             = _mm_macc_ps(dz10,fscal,fiz1);
1196
1197             fjx0             = _mm_macc_ps(dx10,fscal,fjx0);
1198             fjy0             = _mm_macc_ps(dy10,fscal,fjy0);
1199             fjz0             = _mm_macc_ps(dz10,fscal,fjz0);
1200
1201             /**************************
1202              * CALCULATE INTERACTIONS *
1203              **************************/
1204
1205             r20              = _mm_mul_ps(rsq20,rinv20);
1206             r20              = _mm_andnot_ps(dummy_mask,r20);
1207
1208             /* Compute parameters for interactions between i and j atoms */
1209             qq20             = _mm_mul_ps(iq2,jq0);
1210
1211             /* EWALD ELECTROSTATICS */
1212
1213             /* Analytical PME correction */
1214             zeta2            = _mm_mul_ps(beta2,rsq20);
1215             rinv3            = _mm_mul_ps(rinvsq20,rinv20);
1216             pmecorrF         = gmx_mm_pmecorrF_ps(zeta2);
1217             felec            = _mm_macc_ps(pmecorrF,beta3,rinv3);
1218             felec            = _mm_mul_ps(qq20,felec);
1219
1220             fscal            = felec;
1221
1222             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1223
1224              /* Update vectorial force */
1225             fix2             = _mm_macc_ps(dx20,fscal,fix2);
1226             fiy2             = _mm_macc_ps(dy20,fscal,fiy2);
1227             fiz2             = _mm_macc_ps(dz20,fscal,fiz2);
1228
1229             fjx0             = _mm_macc_ps(dx20,fscal,fjx0);
1230             fjy0             = _mm_macc_ps(dy20,fscal,fjy0);
1231             fjz0             = _mm_macc_ps(dz20,fscal,fjz0);
1232
1233             /**************************
1234              * CALCULATE INTERACTIONS *
1235              **************************/
1236
1237             r30              = _mm_mul_ps(rsq30,rinv30);
1238             r30              = _mm_andnot_ps(dummy_mask,r30);
1239
1240             /* Compute parameters for interactions between i and j atoms */
1241             qq30             = _mm_mul_ps(iq3,jq0);
1242
1243             /* EWALD ELECTROSTATICS */
1244
1245             /* Analytical PME correction */
1246             zeta2            = _mm_mul_ps(beta2,rsq30);
1247             rinv3            = _mm_mul_ps(rinvsq30,rinv30);
1248             pmecorrF         = gmx_mm_pmecorrF_ps(zeta2);
1249             felec            = _mm_macc_ps(pmecorrF,beta3,rinv3);
1250             felec            = _mm_mul_ps(qq30,felec);
1251
1252             fscal            = felec;
1253
1254             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1255
1256              /* Update vectorial force */
1257             fix3             = _mm_macc_ps(dx30,fscal,fix3);
1258             fiy3             = _mm_macc_ps(dy30,fscal,fiy3);
1259             fiz3             = _mm_macc_ps(dz30,fscal,fiz3);
1260
1261             fjx0             = _mm_macc_ps(dx30,fscal,fjx0);
1262             fjy0             = _mm_macc_ps(dy30,fscal,fjy0);
1263             fjz0             = _mm_macc_ps(dz30,fscal,fjz0);
1264
1265             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
1266             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
1267             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
1268             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
1269
1270             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
1271
1272             /* Inner loop uses 135 flops */
1273         }
1274
1275         /* End of innermost loop */
1276
1277         gmx_mm_update_iforce_4atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
1278                                               f+i_coord_offset,fshift+i_shift_offset);
1279
1280         /* Increment number of inner iterations */
1281         inneriter                  += j_index_end - j_index_start;
1282
1283         /* Outer loop uses 24 flops */
1284     }
1285
1286     /* Increment number of outer iterations */
1287     outeriter        += nri;
1288
1289     /* Update outer/inner flops */
1290
1291     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W4_F,outeriter*24 + inneriter*135);
1292 }