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 #include "config.h"
  39
  40 #include <math.h>
  41
  42 #include "../nb_kernel.h"
  43 #include "gromacs/legacyheaders/types/simple.h"
  44 #include "gromacs/math/vec.h"
  45 #include "gromacs/legacyheaders/nrnb.h"
  46
  47 #include "gromacs/simd/math_x86_avx_128_fma_single.h"
  48 #include "kernelutil_x86_avx_128_fma_single.h"
  49
  50 /*
  51  * Gromacs nonbonded kernel:   nb_kernel_ElecEw_VdwLJEw_GeomW4P1_VF_avx_128_fma_single
  52  * Electrostatics interaction: Ewald
  53  * VdW interaction:            LJEwald
  54  * Geometry:                   Water4-Particle
  55  * Calculate force/pot:        PotentialAndForce
  56  */
  57 void
  58 nb_kernel_ElecEw_VdwLJEw_GeomW4P1_VF_avx_128_fma_single
  59                     (t_nblist                    * gmx_restrict       nlist,
  60                      rvec                        * gmx_restrict          xx,
  61                      rvec                        * gmx_restrict          ff,
  62                      t_forcerec                  * gmx_restrict          fr,
  63                      t_mdatoms                   * gmx_restrict     mdatoms,
  64                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
  65                      t_nrnb                      * gmx_restrict        nrnb)
  66 {
  67     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
  68      * just 0 for non-waters.
  69      * Suffixes A,B,C,D refer to j loop unrolling done with AVX_128, e.g. for the four different
  70      * jnr indices corresponding to data put in the four positions in the SIMD register.
  71      */
  72     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
  73     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
  74     int              jnrA,jnrB,jnrC,jnrD;
  75     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
  76     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
  77     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
  78     real             rcutoff_scalar;
  79     real             *shiftvec,*fshift,*x,*f;
  80     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
  81     real             scratch[4*DIM];
  82     __m128           fscal,rcutoff,rcutoff2,jidxall;
  83     int              vdwioffset0;
  84     __m128           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
  85     int              vdwioffset1;
  86     __m128           ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
  87     int              vdwioffset2;
  88     __m128           ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
  89     int              vdwioffset3;
  90     __m128           ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
  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           dx30,dy30,dz30,rsq30,rinv30,rinvsq30,r30,qq30,c6_30,c12_30;
  97     __m128           velec,felec,velecsum,facel,crf,krf,krf2;
  98     real             *charge;
  99     int              nvdwtype;
 100     __m128           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 101     int              *vdwtype;
 102     real             *vdwparam;
 103     __m128           one_sixth   = _mm_set1_ps(1.0/6.0);
 104     __m128           one_twelfth = _mm_set1_ps(1.0/12.0);
 105     __m128           c6grid_00;
 106     __m128           c6grid_10;
 107     __m128           c6grid_20;
 108     __m128           c6grid_30;
 109     real             *vdwgridparam;
 110     __m128           ewclj,ewclj2,ewclj6,ewcljrsq,poly,exponent,f6A,f6B,sh_lj_ewald;
 111     __m128           one_half = _mm_set1_ps(0.5);
 112     __m128           minus_one = _mm_set1_ps(-1.0);
 113     __m128i          ewitab;
 114     __m128           ewtabscale,eweps,twoeweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
 115     __m128           beta,beta2,beta3,zeta2,pmecorrF,pmecorrV,rinv3;
 116     real             *ewtab;
 117     __m128           dummy_mask,cutoff_mask;
 118     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
 119     __m128           one     = _mm_set1_ps(1.0);
 120     __m128           two     = _mm_set1_ps(2.0);
 121     x                = xx[0];
 122     f                = ff[0];
 123
 124     nri              = nlist->nri;
 125     iinr             = nlist->iinr;
 126     jindex           = nlist->jindex;
 127     jjnr             = nlist->jjnr;
 128     shiftidx         = nlist->shift;
 129     gid              = nlist->gid;
 130     shiftvec         = fr->shift_vec[0];
 131     fshift           = fr->fshift[0];
 132     facel            = _mm_set1_ps(fr->epsfac);
 133     charge           = mdatoms->chargeA;
 134     nvdwtype         = fr->ntype;
 135     vdwparam         = fr->nbfp;
 136     vdwtype          = mdatoms->typeA;
 137     vdwgridparam     = fr->ljpme_c6grid;
 138     sh_lj_ewald      = _mm_set1_ps(fr->ic->sh_lj_ewald);
 139     ewclj            = _mm_set1_ps(fr->ewaldcoeff_lj);
 140     ewclj2           = _mm_mul_ps(minus_one,_mm_mul_ps(ewclj,ewclj));
 141
 142     sh_ewald         = _mm_set1_ps(fr->ic->sh_ewald);
 143     beta             = _mm_set1_ps(fr->ic->ewaldcoeff_q);
 144     beta2            = _mm_mul_ps(beta,beta);
 145     beta3            = _mm_mul_ps(beta,beta2);
 146     ewtab            = fr->ic->tabq_coul_FDV0;
 147     ewtabscale       = _mm_set1_ps(fr->ic->tabq_scale);
 148     ewtabhalfspace   = _mm_set1_ps(0.5/fr->ic->tabq_scale);
 149
 150     /* Setup water-specific parameters */
 151     inr              = nlist->iinr[0];
 152     iq1              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+1]));
 153     iq2              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+2]));
 154     iq3              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+3]));
 155     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 156
 157     /* Avoid stupid compiler warnings */
 158     jnrA = jnrB = jnrC = jnrD = 0;
 159     j_coord_offsetA = 0;
 160     j_coord_offsetB = 0;
 161     j_coord_offsetC = 0;
 162     j_coord_offsetD = 0;
 163
 164     outeriter        = 0;
 165     inneriter        = 0;
 166
 167     for(iidx=0;iidx<4*DIM;iidx++)
 168     {
 169         scratch[iidx] = 0.0;
 170     }
 171
 172     /* Start outer loop over neighborlists */
 173     for(iidx=0; iidx<nri; iidx++)
 174     {
 175         /* Load shift vector for this list */
 176         i_shift_offset   = DIM*shiftidx[iidx];
 177
 178         /* Load limits for loop over neighbors */
 179         j_index_start    = jindex[iidx];
 180         j_index_end      = jindex[iidx+1];
 181
 182         /* Get outer coordinate index */
 183         inr              = iinr[iidx];
 184         i_coord_offset   = DIM*inr;
 185
 186         /* Load i particle coords and add shift vector */
 187         gmx_mm_load_shift_and_4rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
 188                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
 189
 190         fix0             = _mm_setzero_ps();
 191         fiy0             = _mm_setzero_ps();
 192         fiz0             = _mm_setzero_ps();
 193         fix1             = _mm_setzero_ps();
 194         fiy1             = _mm_setzero_ps();
 195         fiz1             = _mm_setzero_ps();
 196         fix2             = _mm_setzero_ps();
 197         fiy2             = _mm_setzero_ps();
 198         fiz2             = _mm_setzero_ps();
 199         fix3             = _mm_setzero_ps();
 200         fiy3             = _mm_setzero_ps();
 201         fiz3             = _mm_setzero_ps();
 202
 203         /* Reset potential sums */
 204         velecsum         = _mm_setzero_ps();
 205         vvdwsum          = _mm_setzero_ps();
 206
 207         /* Start inner kernel loop */
 208         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 209         {
 210
 211             /* Get j neighbor index, and coordinate index */
 212             jnrA             = jjnr[jidx];
 213             jnrB             = jjnr[jidx+1];
 214             jnrC             = jjnr[jidx+2];
 215             jnrD             = jjnr[jidx+3];
 216             j_coord_offsetA  = DIM*jnrA;
 217             j_coord_offsetB  = DIM*jnrB;
 218             j_coord_offsetC  = DIM*jnrC;
 219             j_coord_offsetD  = DIM*jnrD;
 220
 221             /* load j atom coordinates */
 222             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 223                                               x+j_coord_offsetC,x+j_coord_offsetD,
 224                                               &jx0,&jy0,&jz0);
 225
 226             /* Calculate displacement vector */
 227             dx00             = _mm_sub_ps(ix0,jx0);
 228             dy00             = _mm_sub_ps(iy0,jy0);
 229             dz00             = _mm_sub_ps(iz0,jz0);
 230             dx10             = _mm_sub_ps(ix1,jx0);
 231             dy10             = _mm_sub_ps(iy1,jy0);
 232             dz10             = _mm_sub_ps(iz1,jz0);
 233             dx20             = _mm_sub_ps(ix2,jx0);
 234             dy20             = _mm_sub_ps(iy2,jy0);
 235             dz20             = _mm_sub_ps(iz2,jz0);
 236             dx30             = _mm_sub_ps(ix3,jx0);
 237             dy30             = _mm_sub_ps(iy3,jy0);
 238             dz30             = _mm_sub_ps(iz3,jz0);
 239
 240             /* Calculate squared distance and things based on it */
 241             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 242             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 243             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 244             rsq30            = gmx_mm_calc_rsq_ps(dx30,dy30,dz30);
 245
 246             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 247             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 248             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 249             rinv30           = gmx_mm_invsqrt_ps(rsq30);
 250
 251             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 252             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 253             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 254             rinvsq30         = _mm_mul_ps(rinv30,rinv30);
 255
 256             /* Load parameters for j particles */
 257             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 258                                                               charge+jnrC+0,charge+jnrD+0);
 259             vdwjidx0A        = 2*vdwtype[jnrA+0];
 260             vdwjidx0B        = 2*vdwtype[jnrB+0];
 261             vdwjidx0C        = 2*vdwtype[jnrC+0];
 262             vdwjidx0D        = 2*vdwtype[jnrD+0];
 263
 264             fjx0             = _mm_setzero_ps();
 265             fjy0             = _mm_setzero_ps();
 266             fjz0             = _mm_setzero_ps();
 267
 268             /**************************
 269              * CALCULATE INTERACTIONS *
 270              **************************/
 271
 272             r00              = _mm_mul_ps(rsq00,rinv00);
 273
 274             /* Compute parameters for interactions between i and j atoms */
 275             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 276                                          vdwparam+vdwioffset0+vdwjidx0B,
 277                                          vdwparam+vdwioffset0+vdwjidx0C,
 278                                          vdwparam+vdwioffset0+vdwjidx0D,
 279                                          &c6_00,&c12_00);
 280
 281             c6grid_00       = gmx_mm_load_4real_swizzle_ps(vdwgridparam+vdwioffset0+vdwjidx0A,
 282                                                                vdwgridparam+vdwioffset0+vdwjidx0B,
 283                                                                vdwgridparam+vdwioffset0+vdwjidx0C,
 284                                                                vdwgridparam+vdwioffset0+vdwjidx0D);
 285
 286             /* Analytical LJ-PME */
 287             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 288             ewcljrsq         = _mm_mul_ps(ewclj2,rsq00);
 289             ewclj6           = _mm_mul_ps(ewclj2,_mm_mul_ps(ewclj2,ewclj2));
 290             exponent         = gmx_simd_exp_r(ewcljrsq);
 291             /* poly = exp(-(beta*r)^2) * (1 + (beta*r)^2 + (beta*r)^4 /2) */
 292             poly             = _mm_mul_ps(exponent,_mm_macc_ps(_mm_mul_ps(ewcljrsq,ewcljrsq),one_half,_mm_sub_ps(one,ewcljrsq)));
 293             /* vvdw6 = [C6 - C6grid * (1-poly)]/r6 */
 294             vvdw6            = _mm_mul_ps(_mm_macc_ps(-c6grid_00,_mm_sub_ps(one,poly),c6_00),rinvsix);
 295             vvdw12           = _mm_mul_ps(c12_00,_mm_mul_ps(rinvsix,rinvsix));
 296             vvdw             = _mm_msub_ps(vvdw12,one_twelfth,_mm_mul_ps(vvdw6,one_sixth));
 297             /* fvdw = vvdw12/r - (vvdw6/r + (C6grid * exponent * beta^6)/r) */
 298             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);
 299
 300             /* Update potential sum for this i atom from the interaction with this j atom. */
 301             vvdwsum          = _mm_add_ps(vvdwsum,vvdw);
 302
 303             fscal            = fvdw;
 304
 305              /* Update vectorial force */
 306             fix0             = _mm_macc_ps(dx00,fscal,fix0);
 307             fiy0             = _mm_macc_ps(dy00,fscal,fiy0);
 308             fiz0             = _mm_macc_ps(dz00,fscal,fiz0);
 309
 310             fjx0             = _mm_macc_ps(dx00,fscal,fjx0);
 311             fjy0             = _mm_macc_ps(dy00,fscal,fjy0);
 312             fjz0             = _mm_macc_ps(dz00,fscal,fjz0);
 313
 314             /**************************
 315              * CALCULATE INTERACTIONS *
 316              **************************/
 317
 318             r10              = _mm_mul_ps(rsq10,rinv10);
 319
 320             /* Compute parameters for interactions between i and j atoms */
 321             qq10             = _mm_mul_ps(iq1,jq0);
 322
 323             /* EWALD ELECTROSTATICS */
 324
 325             /* Analytical PME correction */
 326             zeta2            = _mm_mul_ps(beta2,rsq10);
 327             rinv3            = _mm_mul_ps(rinvsq10,rinv10);
 328             pmecorrF         = gmx_mm_pmecorrF_ps(zeta2);
 329             felec            = _mm_macc_ps(pmecorrF,beta3,rinv3);
 330             felec            = _mm_mul_ps(qq10,felec);
 331             pmecorrV         = gmx_mm_pmecorrV_ps(zeta2);
 332             velec            = _mm_nmacc_ps(pmecorrV,beta,rinv10);
 333             velec            = _mm_mul_ps(qq10,velec);
 334
 335             /* Update potential sum for this i atom from the interaction with this j atom. */
 336             velecsum         = _mm_add_ps(velecsum,velec);
 337
 338             fscal            = felec;
 339
 340              /* Update vectorial force */
 341             fix1             = _mm_macc_ps(dx10,fscal,fix1);
 342             fiy1             = _mm_macc_ps(dy10,fscal,fiy1);
 343             fiz1             = _mm_macc_ps(dz10,fscal,fiz1);
 344
 345             fjx0             = _mm_macc_ps(dx10,fscal,fjx0);
 346             fjy0             = _mm_macc_ps(dy10,fscal,fjy0);
 347             fjz0             = _mm_macc_ps(dz10,fscal,fjz0);
 348
 349             /**************************
 350              * CALCULATE INTERACTIONS *
 351              **************************/
 352
 353             r20              = _mm_mul_ps(rsq20,rinv20);
 354
 355             /* Compute parameters for interactions between i and j atoms */
 356             qq20             = _mm_mul_ps(iq2,jq0);
 357
 358             /* EWALD ELECTROSTATICS */
 359
 360             /* Analytical PME correction */
 361             zeta2            = _mm_mul_ps(beta2,rsq20);
 362             rinv3            = _mm_mul_ps(rinvsq20,rinv20);
 363             pmecorrF         = gmx_mm_pmecorrF_ps(zeta2);
 364             felec            = _mm_macc_ps(pmecorrF,beta3,rinv3);
 365             felec            = _mm_mul_ps(qq20,felec);
 366             pmecorrV         = gmx_mm_pmecorrV_ps(zeta2);
 367             velec            = _mm_nmacc_ps(pmecorrV,beta,rinv20);
 368             velec            = _mm_mul_ps(qq20,velec);
 369
 370             /* Update potential sum for this i atom from the interaction with this j atom. */
 371             velecsum         = _mm_add_ps(velecsum,velec);
 372
 373             fscal            = felec;
 374
 375              /* Update vectorial force */
 376             fix2             = _mm_macc_ps(dx20,fscal,fix2);
 377             fiy2             = _mm_macc_ps(dy20,fscal,fiy2);
 378             fiz2             = _mm_macc_ps(dz20,fscal,fiz2);
 379
 380             fjx0             = _mm_macc_ps(dx20,fscal,fjx0);
 381             fjy0             = _mm_macc_ps(dy20,fscal,fjy0);
 382             fjz0             = _mm_macc_ps(dz20,fscal,fjz0);
 383
 384             /**************************
 385              * CALCULATE INTERACTIONS *
 386              **************************/
 387
 388             r30              = _mm_mul_ps(rsq30,rinv30);
 389
 390             /* Compute parameters for interactions between i and j atoms */
 391             qq30             = _mm_mul_ps(iq3,jq0);
 392
 393             /* EWALD ELECTROSTATICS */
 394
 395             /* Analytical PME correction */
 396             zeta2            = _mm_mul_ps(beta2,rsq30);
 397             rinv3            = _mm_mul_ps(rinvsq30,rinv30);
 398             pmecorrF         = gmx_mm_pmecorrF_ps(zeta2);
 399             felec            = _mm_macc_ps(pmecorrF,beta3,rinv3);
 400             felec            = _mm_mul_ps(qq30,felec);
 401             pmecorrV         = gmx_mm_pmecorrV_ps(zeta2);
 402             velec            = _mm_nmacc_ps(pmecorrV,beta,rinv30);
 403             velec            = _mm_mul_ps(qq30,velec);
 404
 405             /* Update potential sum for this i atom from the interaction with this j atom. */
 406             velecsum         = _mm_add_ps(velecsum,velec);
 407
 408             fscal            = felec;
 409
 410              /* Update vectorial force */
 411             fix3             = _mm_macc_ps(dx30,fscal,fix3);
 412             fiy3             = _mm_macc_ps(dy30,fscal,fiy3);
 413             fiz3             = _mm_macc_ps(dz30,fscal,fiz3);
 414
 415             fjx0             = _mm_macc_ps(dx30,fscal,fjx0);
 416             fjy0             = _mm_macc_ps(dy30,fscal,fjy0);
 417             fjz0             = _mm_macc_ps(dz30,fscal,fjz0);
 418
 419             fjptrA             = f+j_coord_offsetA;
 420             fjptrB             = f+j_coord_offsetB;
 421             fjptrC             = f+j_coord_offsetC;
 422             fjptrD             = f+j_coord_offsetD;
 423
 424             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 425
 426             /* Inner loop uses 137 flops */
 427         }
 428
 429         if(jidx<j_index_end)
 430         {
 431
 432             /* Get j neighbor index, and coordinate index */
 433             jnrlistA         = jjnr[jidx];
 434             jnrlistB         = jjnr[jidx+1];
 435             jnrlistC         = jjnr[jidx+2];
 436             jnrlistD         = jjnr[jidx+3];
 437             /* Sign of each element will be negative for non-real atoms.
 438              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 439              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 440              */
 441             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 442             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 443             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 444             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 445             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 446             j_coord_offsetA  = DIM*jnrA;
 447             j_coord_offsetB  = DIM*jnrB;
 448             j_coord_offsetC  = DIM*jnrC;
 449             j_coord_offsetD  = DIM*jnrD;
 450
 451             /* load j atom coordinates */
 452             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 453                                               x+j_coord_offsetC,x+j_coord_offsetD,
 454                                               &jx0,&jy0,&jz0);
 455
 456             /* Calculate displacement vector */
 457             dx00             = _mm_sub_ps(ix0,jx0);
 458             dy00             = _mm_sub_ps(iy0,jy0);
 459             dz00             = _mm_sub_ps(iz0,jz0);
 460             dx10             = _mm_sub_ps(ix1,jx0);
 461             dy10             = _mm_sub_ps(iy1,jy0);
 462             dz10             = _mm_sub_ps(iz1,jz0);
 463             dx20             = _mm_sub_ps(ix2,jx0);
 464             dy20             = _mm_sub_ps(iy2,jy0);
 465             dz20             = _mm_sub_ps(iz2,jz0);
 466             dx30             = _mm_sub_ps(ix3,jx0);
 467             dy30             = _mm_sub_ps(iy3,jy0);
 468             dz30             = _mm_sub_ps(iz3,jz0);
 469
 470             /* Calculate squared distance and things based on it */
 471             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 472             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 473             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 474             rsq30            = gmx_mm_calc_rsq_ps(dx30,dy30,dz30);
 475
 476             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 477             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 478             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 479             rinv30           = gmx_mm_invsqrt_ps(rsq30);
 480
 481             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 482             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 483             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 484             rinvsq30         = _mm_mul_ps(rinv30,rinv30);
 485
 486             /* Load parameters for j particles */
 487             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 488                                                               charge+jnrC+0,charge+jnrD+0);
 489             vdwjidx0A        = 2*vdwtype[jnrA+0];
 490             vdwjidx0B        = 2*vdwtype[jnrB+0];
 491             vdwjidx0C        = 2*vdwtype[jnrC+0];
 492             vdwjidx0D        = 2*vdwtype[jnrD+0];
 493
 494             fjx0             = _mm_setzero_ps();
 495             fjy0             = _mm_setzero_ps();
 496             fjz0             = _mm_setzero_ps();
 497
 498             /**************************
 499              * CALCULATE INTERACTIONS *
 500              **************************/
 501
 502             r00              = _mm_mul_ps(rsq00,rinv00);
 503             r00              = _mm_andnot_ps(dummy_mask,r00);
 504
 505             /* Compute parameters for interactions between i and j atoms */
 506             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 507                                          vdwparam+vdwioffset0+vdwjidx0B,
 508                                          vdwparam+vdwioffset0+vdwjidx0C,
 509                                          vdwparam+vdwioffset0+vdwjidx0D,
 510                                          &c6_00,&c12_00);
 511
 512             c6grid_00       = gmx_mm_load_4real_swizzle_ps(vdwgridparam+vdwioffset0+vdwjidx0A,
 513                                                                vdwgridparam+vdwioffset0+vdwjidx0B,
 514                                                                vdwgridparam+vdwioffset0+vdwjidx0C,
 515                                                                vdwgridparam+vdwioffset0+vdwjidx0D);
 516
 517             /* Analytical LJ-PME */
 518             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 519             ewcljrsq         = _mm_mul_ps(ewclj2,rsq00);
 520             ewclj6           = _mm_mul_ps(ewclj2,_mm_mul_ps(ewclj2,ewclj2));
 521             exponent         = gmx_simd_exp_r(ewcljrsq);
 522             /* poly = exp(-(beta*r)^2) * (1 + (beta*r)^2 + (beta*r)^4 /2) */
 523             poly             = _mm_mul_ps(exponent,_mm_macc_ps(_mm_mul_ps(ewcljrsq,ewcljrsq),one_half,_mm_sub_ps(one,ewcljrsq)));
 524             /* vvdw6 = [C6 - C6grid * (1-poly)]/r6 */
 525             vvdw6            = _mm_mul_ps(_mm_macc_ps(-c6grid_00,_mm_sub_ps(one,poly),c6_00),rinvsix);
 526             vvdw12           = _mm_mul_ps(c12_00,_mm_mul_ps(rinvsix,rinvsix));
 527             vvdw             = _mm_msub_ps(vvdw12,one_twelfth,_mm_mul_ps(vvdw6,one_sixth));
 528             /* fvdw = vvdw12/r - (vvdw6/r + (C6grid * exponent * beta^6)/r) */
 529             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);
 530
 531             /* Update potential sum for this i atom from the interaction with this j atom. */
 532             vvdw             = _mm_andnot_ps(dummy_mask,vvdw);
 533             vvdwsum          = _mm_add_ps(vvdwsum,vvdw);
 534
 535             fscal            = fvdw;
 536
 537             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 538
 539              /* Update vectorial force */
 540             fix0             = _mm_macc_ps(dx00,fscal,fix0);
 541             fiy0             = _mm_macc_ps(dy00,fscal,fiy0);
 542             fiz0             = _mm_macc_ps(dz00,fscal,fiz0);
 543
 544             fjx0             = _mm_macc_ps(dx00,fscal,fjx0);
 545             fjy0             = _mm_macc_ps(dy00,fscal,fjy0);
 546             fjz0             = _mm_macc_ps(dz00,fscal,fjz0);
 547
 548             /**************************
 549              * CALCULATE INTERACTIONS *
 550              **************************/
 551
 552             r10              = _mm_mul_ps(rsq10,rinv10);
 553             r10              = _mm_andnot_ps(dummy_mask,r10);
 554
 555             /* Compute parameters for interactions between i and j atoms */
 556             qq10             = _mm_mul_ps(iq1,jq0);
 557
 558             /* EWALD ELECTROSTATICS */
 559
 560             /* Analytical PME correction */
 561             zeta2            = _mm_mul_ps(beta2,rsq10);
 562             rinv3            = _mm_mul_ps(rinvsq10,rinv10);
 563             pmecorrF         = gmx_mm_pmecorrF_ps(zeta2);
 564             felec            = _mm_macc_ps(pmecorrF,beta3,rinv3);
 565             felec            = _mm_mul_ps(qq10,felec);
 566             pmecorrV         = gmx_mm_pmecorrV_ps(zeta2);
 567             velec            = _mm_nmacc_ps(pmecorrV,beta,rinv10);
 568             velec            = _mm_mul_ps(qq10,velec);
 569
 570             /* Update potential sum for this i atom from the interaction with this j atom. */
 571             velec            = _mm_andnot_ps(dummy_mask,velec);
 572             velecsum         = _mm_add_ps(velecsum,velec);
 573
 574             fscal            = felec;
 575
 576             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 577
 578              /* Update vectorial force */
 579             fix1             = _mm_macc_ps(dx10,fscal,fix1);
 580             fiy1             = _mm_macc_ps(dy10,fscal,fiy1);
 581             fiz1             = _mm_macc_ps(dz10,fscal,fiz1);
 582
 583             fjx0             = _mm_macc_ps(dx10,fscal,fjx0);
 584             fjy0             = _mm_macc_ps(dy10,fscal,fjy0);
 585             fjz0             = _mm_macc_ps(dz10,fscal,fjz0);
 586
 587             /**************************
 588              * CALCULATE INTERACTIONS *
 589              **************************/
 590
 591             r20              = _mm_mul_ps(rsq20,rinv20);
 592             r20              = _mm_andnot_ps(dummy_mask,r20);
 593
 594             /* Compute parameters for interactions between i and j atoms */
 595             qq20             = _mm_mul_ps(iq2,jq0);
 596
 597             /* EWALD ELECTROSTATICS */
 598
 599             /* Analytical PME correction */
 600             zeta2            = _mm_mul_ps(beta2,rsq20);
 601             rinv3            = _mm_mul_ps(rinvsq20,rinv20);
 602             pmecorrF         = gmx_mm_pmecorrF_ps(zeta2);
 603             felec            = _mm_macc_ps(pmecorrF,beta3,rinv3);
 604             felec            = _mm_mul_ps(qq20,felec);
 605             pmecorrV         = gmx_mm_pmecorrV_ps(zeta2);
 606             velec            = _mm_nmacc_ps(pmecorrV,beta,rinv20);
 607             velec            = _mm_mul_ps(qq20,velec);
 608
 609             /* Update potential sum for this i atom from the interaction with this j atom. */
 610             velec            = _mm_andnot_ps(dummy_mask,velec);
 611             velecsum         = _mm_add_ps(velecsum,velec);
 612
 613             fscal            = felec;
 614
 615             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 616
 617              /* Update vectorial force */
 618             fix2             = _mm_macc_ps(dx20,fscal,fix2);
 619             fiy2             = _mm_macc_ps(dy20,fscal,fiy2);
 620             fiz2             = _mm_macc_ps(dz20,fscal,fiz2);
 621
 622             fjx0             = _mm_macc_ps(dx20,fscal,fjx0);
 623             fjy0             = _mm_macc_ps(dy20,fscal,fjy0);
 624             fjz0             = _mm_macc_ps(dz20,fscal,fjz0);
 625
 626             /**************************
 627              * CALCULATE INTERACTIONS *
 628              **************************/
 629
 630             r30              = _mm_mul_ps(rsq30,rinv30);
 631             r30              = _mm_andnot_ps(dummy_mask,r30);
 632
 633             /* Compute parameters for interactions between i and j atoms */
 634             qq30             = _mm_mul_ps(iq3,jq0);
 635
 636             /* EWALD ELECTROSTATICS */
 637
 638             /* Analytical PME correction */
 639             zeta2            = _mm_mul_ps(beta2,rsq30);
 640             rinv3            = _mm_mul_ps(rinvsq30,rinv30);
 641             pmecorrF         = gmx_mm_pmecorrF_ps(zeta2);
 642             felec            = _mm_macc_ps(pmecorrF,beta3,rinv3);
 643             felec            = _mm_mul_ps(qq30,felec);
 644             pmecorrV         = gmx_mm_pmecorrV_ps(zeta2);
 645             velec            = _mm_nmacc_ps(pmecorrV,beta,rinv30);
 646             velec            = _mm_mul_ps(qq30,velec);
 647
 648             /* Update potential sum for this i atom from the interaction with this j atom. */
 649             velec            = _mm_andnot_ps(dummy_mask,velec);
 650             velecsum         = _mm_add_ps(velecsum,velec);
 651
 652             fscal            = felec;
 653
 654             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 655
 656              /* Update vectorial force */
 657             fix3             = _mm_macc_ps(dx30,fscal,fix3);
 658             fiy3             = _mm_macc_ps(dy30,fscal,fiy3);
 659             fiz3             = _mm_macc_ps(dz30,fscal,fiz3);
 660
 661             fjx0             = _mm_macc_ps(dx30,fscal,fjx0);
 662             fjy0             = _mm_macc_ps(dy30,fscal,fjy0);
 663             fjz0             = _mm_macc_ps(dz30,fscal,fjz0);
 664
 665             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 666             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 667             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 668             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 669
 670             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 671
 672             /* Inner loop uses 141 flops */
 673         }
 674
 675         /* End of innermost loop */
 676
 677         gmx_mm_update_iforce_4atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
 678                                               f+i_coord_offset,fshift+i_shift_offset);
 679
 680         ggid                        = gid[iidx];
 681         /* Update potential energies */
 682         gmx_mm_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid);
 683         gmx_mm_update_1pot_ps(vvdwsum,kernel_data->energygrp_vdw+ggid);
 684
 685         /* Increment number of inner iterations */
 686         inneriter                  += j_index_end - j_index_start;
 687
 688         /* Outer loop uses 26 flops */
 689     }
 690
 691     /* Increment number of outer iterations */
 692     outeriter        += nri;
 693
 694     /* Update outer/inner flops */
 695
 696     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W4_VF,outeriter*26 + inneriter*141);
 697 }
 698 /*
 699  * Gromacs nonbonded kernel:   nb_kernel_ElecEw_VdwLJEw_GeomW4P1_F_avx_128_fma_single
 700  * Electrostatics interaction: Ewald
 701  * VdW interaction:            LJEwald
 702  * Geometry:                   Water4-Particle
 703  * Calculate force/pot:        Force
 704  */
 705 void
 706 nb_kernel_ElecEw_VdwLJEw_GeomW4P1_F_avx_128_fma_single
 707                     (t_nblist                    * gmx_restrict       nlist,
 708                      rvec                        * gmx_restrict          xx,
 709                      rvec                        * gmx_restrict          ff,
 710                      t_forcerec                  * gmx_restrict          fr,
 711                      t_mdatoms                   * gmx_restrict     mdatoms,
 712                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 713                      t_nrnb                      * gmx_restrict        nrnb)
 714 {
 715     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 716      * just 0 for non-waters.
 717      * Suffixes A,B,C,D refer to j loop unrolling done with AVX_128, e.g. for the four different
 718      * jnr indices corresponding to data put in the four positions in the SIMD register.
 719      */
 720     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 721     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 722     int              jnrA,jnrB,jnrC,jnrD;
 723     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
 724     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 725     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 726     real             rcutoff_scalar;
 727     real             *shiftvec,*fshift,*x,*f;
 728     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
 729     real             scratch[4*DIM];
 730     __m128           fscal,rcutoff,rcutoff2,jidxall;
 731     int              vdwioffset0;
 732     __m128           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 733     int              vdwioffset1;
 734     __m128           ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
 735     int              vdwioffset2;
 736     __m128           ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
 737     int              vdwioffset3;
 738     __m128           ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
 739     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
 740     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 741     __m128           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 742     __m128           dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
 743     __m128           dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
 744     __m128           dx30,dy30,dz30,rsq30,rinv30,rinvsq30,r30,qq30,c6_30,c12_30;
 745     __m128           velec,felec,velecsum,facel,crf,krf,krf2;
 746     real             *charge;
 747     int              nvdwtype;
 748     __m128           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 749     int              *vdwtype;
 750     real             *vdwparam;
 751     __m128           one_sixth   = _mm_set1_ps(1.0/6.0);
 752     __m128           one_twelfth = _mm_set1_ps(1.0/12.0);
 753     __m128           c6grid_00;
 754     __m128           c6grid_10;
 755     __m128           c6grid_20;
 756     __m128           c6grid_30;
 757     real             *vdwgridparam;
 758     __m128           ewclj,ewclj2,ewclj6,ewcljrsq,poly,exponent,f6A,f6B,sh_lj_ewald;
 759     __m128           one_half = _mm_set1_ps(0.5);
 760     __m128           minus_one = _mm_set1_ps(-1.0);
 761     __m128i          ewitab;
 762     __m128           ewtabscale,eweps,twoeweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
 763     __m128           beta,beta2,beta3,zeta2,pmecorrF,pmecorrV,rinv3;
 764     real             *ewtab;
 765     __m128           dummy_mask,cutoff_mask;
 766     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
 767     __m128           one     = _mm_set1_ps(1.0);
 768     __m128           two     = _mm_set1_ps(2.0);
 769     x                = xx[0];
 770     f                = ff[0];
 771
 772     nri              = nlist->nri;
 773     iinr             = nlist->iinr;
 774     jindex           = nlist->jindex;
 775     jjnr             = nlist->jjnr;
 776     shiftidx         = nlist->shift;
 777     gid              = nlist->gid;
 778     shiftvec         = fr->shift_vec[0];
 779     fshift           = fr->fshift[0];
 780     facel            = _mm_set1_ps(fr->epsfac);
 781     charge           = mdatoms->chargeA;
 782     nvdwtype         = fr->ntype;
 783     vdwparam         = fr->nbfp;
 784     vdwtype          = mdatoms->typeA;
 785     vdwgridparam     = fr->ljpme_c6grid;
 786     sh_lj_ewald      = _mm_set1_ps(fr->ic->sh_lj_ewald);
 787     ewclj            = _mm_set1_ps(fr->ewaldcoeff_lj);
 788     ewclj2           = _mm_mul_ps(minus_one,_mm_mul_ps(ewclj,ewclj));
 789
 790     sh_ewald         = _mm_set1_ps(fr->ic->sh_ewald);
 791     beta             = _mm_set1_ps(fr->ic->ewaldcoeff_q);
 792     beta2            = _mm_mul_ps(beta,beta);
 793     beta3            = _mm_mul_ps(beta,beta2);
 794     ewtab            = fr->ic->tabq_coul_F;
 795     ewtabscale       = _mm_set1_ps(fr->ic->tabq_scale);
 796     ewtabhalfspace   = _mm_set1_ps(0.5/fr->ic->tabq_scale);
 797
 798     /* Setup water-specific parameters */
 799     inr              = nlist->iinr[0];
 800     iq1              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+1]));
 801     iq2              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+2]));
 802     iq3              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+3]));
 803     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 804
 805     /* Avoid stupid compiler warnings */
 806     jnrA = jnrB = jnrC = jnrD = 0;
 807     j_coord_offsetA = 0;
 808     j_coord_offsetB = 0;
 809     j_coord_offsetC = 0;
 810     j_coord_offsetD = 0;
 811
 812     outeriter        = 0;
 813     inneriter        = 0;
 814
 815     for(iidx=0;iidx<4*DIM;iidx++)
 816     {
 817         scratch[iidx] = 0.0;
 818     }
 819
 820     /* Start outer loop over neighborlists */
 821     for(iidx=0; iidx<nri; iidx++)
 822     {
 823         /* Load shift vector for this list */
 824         i_shift_offset   = DIM*shiftidx[iidx];
 825
 826         /* Load limits for loop over neighbors */
 827         j_index_start    = jindex[iidx];
 828         j_index_end      = jindex[iidx+1];
 829
 830         /* Get outer coordinate index */
 831         inr              = iinr[iidx];
 832         i_coord_offset   = DIM*inr;
 833
 834         /* Load i particle coords and add shift vector */
 835         gmx_mm_load_shift_and_4rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
 836                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
 837
 838         fix0             = _mm_setzero_ps();
 839         fiy0             = _mm_setzero_ps();
 840         fiz0             = _mm_setzero_ps();
 841         fix1             = _mm_setzero_ps();
 842         fiy1             = _mm_setzero_ps();
 843         fiz1             = _mm_setzero_ps();
 844         fix2             = _mm_setzero_ps();
 845         fiy2             = _mm_setzero_ps();
 846         fiz2             = _mm_setzero_ps();
 847         fix3             = _mm_setzero_ps();
 848         fiy3             = _mm_setzero_ps();
 849         fiz3             = _mm_setzero_ps();
 850
 851         /* Start inner kernel loop */
 852         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 853         {
 854
 855             /* Get j neighbor index, and coordinate index */
 856             jnrA             = jjnr[jidx];
 857             jnrB             = jjnr[jidx+1];
 858             jnrC             = jjnr[jidx+2];
 859             jnrD             = jjnr[jidx+3];
 860             j_coord_offsetA  = DIM*jnrA;
 861             j_coord_offsetB  = DIM*jnrB;
 862             j_coord_offsetC  = DIM*jnrC;
 863             j_coord_offsetD  = DIM*jnrD;
 864
 865             /* load j atom coordinates */
 866             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 867                                               x+j_coord_offsetC,x+j_coord_offsetD,
 868                                               &jx0,&jy0,&jz0);
 869
 870             /* Calculate displacement vector */
 871             dx00             = _mm_sub_ps(ix0,jx0);
 872             dy00             = _mm_sub_ps(iy0,jy0);
 873             dz00             = _mm_sub_ps(iz0,jz0);
 874             dx10             = _mm_sub_ps(ix1,jx0);
 875             dy10             = _mm_sub_ps(iy1,jy0);
 876             dz10             = _mm_sub_ps(iz1,jz0);
 877             dx20             = _mm_sub_ps(ix2,jx0);
 878             dy20             = _mm_sub_ps(iy2,jy0);
 879             dz20             = _mm_sub_ps(iz2,jz0);
 880             dx30             = _mm_sub_ps(ix3,jx0);
 881             dy30             = _mm_sub_ps(iy3,jy0);
 882             dz30             = _mm_sub_ps(iz3,jz0);
 883
 884             /* Calculate squared distance and things based on it */
 885             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 886             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 887             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 888             rsq30            = gmx_mm_calc_rsq_ps(dx30,dy30,dz30);
 889
 890             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 891             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 892             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 893             rinv30           = gmx_mm_invsqrt_ps(rsq30);
 894
 895             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 896             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 897             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 898             rinvsq30         = _mm_mul_ps(rinv30,rinv30);
 899
 900             /* Load parameters for j particles */
 901             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 902                                                               charge+jnrC+0,charge+jnrD+0);
 903             vdwjidx0A        = 2*vdwtype[jnrA+0];
 904             vdwjidx0B        = 2*vdwtype[jnrB+0];
 905             vdwjidx0C        = 2*vdwtype[jnrC+0];
 906             vdwjidx0D        = 2*vdwtype[jnrD+0];
 907
 908             fjx0             = _mm_setzero_ps();
 909             fjy0             = _mm_setzero_ps();
 910             fjz0             = _mm_setzero_ps();
 911
 912             /**************************
 913              * CALCULATE INTERACTIONS *
 914              **************************/
 915
 916             r00              = _mm_mul_ps(rsq00,rinv00);
 917
 918             /* Compute parameters for interactions between i and j atoms */
 919             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 920                                          vdwparam+vdwioffset0+vdwjidx0B,
 921                                          vdwparam+vdwioffset0+vdwjidx0C,
 922                                          vdwparam+vdwioffset0+vdwjidx0D,
 923                                          &c6_00,&c12_00);
 924
 925             c6grid_00       = gmx_mm_load_4real_swizzle_ps(vdwgridparam+vdwioffset0+vdwjidx0A,
 926                                                                vdwgridparam+vdwioffset0+vdwjidx0B,
 927                                                                vdwgridparam+vdwioffset0+vdwjidx0C,
 928                                                                vdwgridparam+vdwioffset0+vdwjidx0D);
 929
 930             /* Analytical LJ-PME */
 931             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 932             ewcljrsq         = _mm_mul_ps(ewclj2,rsq00);
 933             ewclj6           = _mm_mul_ps(ewclj2,_mm_mul_ps(ewclj2,ewclj2));
 934             exponent         = gmx_simd_exp_r(ewcljrsq);
 935             /* poly = exp(-(beta*r)^2) * (1 + (beta*r)^2 + (beta*r)^4 /2) */
 936             poly             = _mm_mul_ps(exponent,_mm_macc_ps(_mm_mul_ps(ewcljrsq,ewcljrsq),one_half,_mm_sub_ps(one,ewcljrsq)));
 937             /* f6A = 6 * C6grid * (1 - poly) */
 938             f6A              = _mm_mul_ps(c6grid_00,_mm_sub_ps(one,poly));
 939             /* f6B = C6grid * exponent * beta^6 */
 940             f6B              = _mm_mul_ps(_mm_mul_ps(c6grid_00,one_sixth),_mm_mul_ps(exponent,ewclj6));
 941             /* fvdw = 12*C12/r13 - ((6*C6 - f6A)/r6 + f6B)/r */
 942             fvdw              = _mm_mul_ps(_mm_macc_ps(_mm_msub_ps(c12_00,rinvsix,_mm_sub_ps(c6_00,f6A)),rinvsix,f6B),rinvsq00);
 943
 944             fscal            = fvdw;
 945
 946              /* Update vectorial force */
 947             fix0             = _mm_macc_ps(dx00,fscal,fix0);
 948             fiy0             = _mm_macc_ps(dy00,fscal,fiy0);
 949             fiz0             = _mm_macc_ps(dz00,fscal,fiz0);
 950
 951             fjx0             = _mm_macc_ps(dx00,fscal,fjx0);
 952             fjy0             = _mm_macc_ps(dy00,fscal,fjy0);
 953             fjz0             = _mm_macc_ps(dz00,fscal,fjz0);
 954
 955             /**************************
 956              * CALCULATE INTERACTIONS *
 957              **************************/
 958
 959             r10              = _mm_mul_ps(rsq10,rinv10);
 960
 961             /* Compute parameters for interactions between i and j atoms */
 962             qq10             = _mm_mul_ps(iq1,jq0);
 963
 964             /* EWALD ELECTROSTATICS */
 965
 966             /* Analytical PME correction */
 967             zeta2            = _mm_mul_ps(beta2,rsq10);
 968             rinv3            = _mm_mul_ps(rinvsq10,rinv10);
 969             pmecorrF         = gmx_mm_pmecorrF_ps(zeta2);
 970             felec            = _mm_macc_ps(pmecorrF,beta3,rinv3);
 971             felec            = _mm_mul_ps(qq10,felec);
 972
 973             fscal            = felec;
 974
 975              /* Update vectorial force */
 976             fix1             = _mm_macc_ps(dx10,fscal,fix1);
 977             fiy1             = _mm_macc_ps(dy10,fscal,fiy1);
 978             fiz1             = _mm_macc_ps(dz10,fscal,fiz1);
 979
 980             fjx0             = _mm_macc_ps(dx10,fscal,fjx0);
 981             fjy0             = _mm_macc_ps(dy10,fscal,fjy0);
 982             fjz0             = _mm_macc_ps(dz10,fscal,fjz0);
 983
 984             /**************************
 985              * CALCULATE INTERACTIONS *
 986              **************************/
 987
 988             r20              = _mm_mul_ps(rsq20,rinv20);
 989
 990             /* Compute parameters for interactions between i and j atoms */
 991             qq20             = _mm_mul_ps(iq2,jq0);
 992
 993             /* EWALD ELECTROSTATICS */
 994
 995             /* Analytical PME correction */
 996             zeta2            = _mm_mul_ps(beta2,rsq20);
 997             rinv3            = _mm_mul_ps(rinvsq20,rinv20);
 998             pmecorrF         = gmx_mm_pmecorrF_ps(zeta2);
 999             felec            = _mm_macc_ps(pmecorrF,beta3,rinv3);
1000             felec            = _mm_mul_ps(qq20,felec);
1001
1002             fscal            = felec;
1003
1004              /* Update vectorial force */
1005             fix2             = _mm_macc_ps(dx20,fscal,fix2);
1006             fiy2             = _mm_macc_ps(dy20,fscal,fiy2);
1007             fiz2             = _mm_macc_ps(dz20,fscal,fiz2);
1008
1009             fjx0             = _mm_macc_ps(dx20,fscal,fjx0);
1010             fjy0             = _mm_macc_ps(dy20,fscal,fjy0);
1011             fjz0             = _mm_macc_ps(dz20,fscal,fjz0);
1012
1013             /**************************
1014              * CALCULATE INTERACTIONS *
1015              **************************/
1016
1017             r30              = _mm_mul_ps(rsq30,rinv30);
1018
1019             /* Compute parameters for interactions between i and j atoms */
1020             qq30             = _mm_mul_ps(iq3,jq0);
1021
1022             /* EWALD ELECTROSTATICS */
1023
1024             /* Analytical PME correction */
1025             zeta2            = _mm_mul_ps(beta2,rsq30);
1026             rinv3            = _mm_mul_ps(rinvsq30,rinv30);
1027             pmecorrF         = gmx_mm_pmecorrF_ps(zeta2);
1028             felec            = _mm_macc_ps(pmecorrF,beta3,rinv3);
1029             felec            = _mm_mul_ps(qq30,felec);
1030
1031             fscal            = felec;
1032
1033              /* Update vectorial force */
1034             fix3             = _mm_macc_ps(dx30,fscal,fix3);
1035             fiy3             = _mm_macc_ps(dy30,fscal,fiy3);
1036             fiz3             = _mm_macc_ps(dz30,fscal,fiz3);
1037
1038             fjx0             = _mm_macc_ps(dx30,fscal,fjx0);
1039             fjy0             = _mm_macc_ps(dy30,fscal,fjy0);
1040             fjz0             = _mm_macc_ps(dz30,fscal,fjz0);
1041
1042             fjptrA             = f+j_coord_offsetA;
1043             fjptrB             = f+j_coord_offsetB;
1044             fjptrC             = f+j_coord_offsetC;
1045             fjptrD             = f+j_coord_offsetD;
1046
1047             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
1048
1049             /* Inner loop uses 131 flops */
1050         }
1051
1052         if(jidx<j_index_end)
1053         {
1054
1055             /* Get j neighbor index, and coordinate index */
1056             jnrlistA         = jjnr[jidx];
1057             jnrlistB         = jjnr[jidx+1];
1058             jnrlistC         = jjnr[jidx+2];
1059             jnrlistD         = jjnr[jidx+3];
1060             /* Sign of each element will be negative for non-real atoms.
1061              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
1062              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
1063              */
1064             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
1065             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
1066             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
1067             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
1068             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
1069             j_coord_offsetA  = DIM*jnrA;
1070             j_coord_offsetB  = DIM*jnrB;
1071             j_coord_offsetC  = DIM*jnrC;
1072             j_coord_offsetD  = DIM*jnrD;
1073
1074             /* load j atom coordinates */
1075             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
1076                                               x+j_coord_offsetC,x+j_coord_offsetD,
1077                                               &jx0,&jy0,&jz0);
1078
1079             /* Calculate displacement vector */
1080             dx00             = _mm_sub_ps(ix0,jx0);
1081             dy00             = _mm_sub_ps(iy0,jy0);
1082             dz00             = _mm_sub_ps(iz0,jz0);
1083             dx10             = _mm_sub_ps(ix1,jx0);
1084             dy10             = _mm_sub_ps(iy1,jy0);
1085             dz10             = _mm_sub_ps(iz1,jz0);
1086             dx20             = _mm_sub_ps(ix2,jx0);
1087             dy20             = _mm_sub_ps(iy2,jy0);
1088             dz20             = _mm_sub_ps(iz2,jz0);
1089             dx30             = _mm_sub_ps(ix3,jx0);
1090             dy30             = _mm_sub_ps(iy3,jy0);
1091             dz30             = _mm_sub_ps(iz3,jz0);
1092
1093             /* Calculate squared distance and things based on it */
1094             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
1095             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
1096             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
1097             rsq30            = gmx_mm_calc_rsq_ps(dx30,dy30,dz30);
1098
1099             rinv00           = gmx_mm_invsqrt_ps(rsq00);
1100             rinv10           = gmx_mm_invsqrt_ps(rsq10);
1101             rinv20           = gmx_mm_invsqrt_ps(rsq20);
1102             rinv30           = gmx_mm_invsqrt_ps(rsq30);
1103
1104             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
1105             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
1106             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
1107             rinvsq30         = _mm_mul_ps(rinv30,rinv30);
1108
1109             /* Load parameters for j particles */
1110             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
1111                                                               charge+jnrC+0,charge+jnrD+0);
1112             vdwjidx0A        = 2*vdwtype[jnrA+0];
1113             vdwjidx0B        = 2*vdwtype[jnrB+0];
1114             vdwjidx0C        = 2*vdwtype[jnrC+0];
1115             vdwjidx0D        = 2*vdwtype[jnrD+0];
1116
1117             fjx0             = _mm_setzero_ps();
1118             fjy0             = _mm_setzero_ps();
1119             fjz0             = _mm_setzero_ps();
1120
1121             /**************************
1122              * CALCULATE INTERACTIONS *
1123              **************************/
1124
1125             r00              = _mm_mul_ps(rsq00,rinv00);
1126             r00              = _mm_andnot_ps(dummy_mask,r00);
1127
1128             /* Compute parameters for interactions between i and j atoms */
1129             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
1130                                          vdwparam+vdwioffset0+vdwjidx0B,
1131                                          vdwparam+vdwioffset0+vdwjidx0C,
1132                                          vdwparam+vdwioffset0+vdwjidx0D,
1133                                          &c6_00,&c12_00);
1134
1135             c6grid_00       = gmx_mm_load_4real_swizzle_ps(vdwgridparam+vdwioffset0+vdwjidx0A,
1136                                                                vdwgridparam+vdwioffset0+vdwjidx0B,
1137                                                                vdwgridparam+vdwioffset0+vdwjidx0C,
1138                                                                vdwgridparam+vdwioffset0+vdwjidx0D);
1139
1140             /* Analytical LJ-PME */
1141             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
1142             ewcljrsq         = _mm_mul_ps(ewclj2,rsq00);
1143             ewclj6           = _mm_mul_ps(ewclj2,_mm_mul_ps(ewclj2,ewclj2));
1144             exponent         = gmx_simd_exp_r(ewcljrsq);
1145             /* poly = exp(-(beta*r)^2) * (1 + (beta*r)^2 + (beta*r)^4 /2) */
1146             poly             = _mm_mul_ps(exponent,_mm_macc_ps(_mm_mul_ps(ewcljrsq,ewcljrsq),one_half,_mm_sub_ps(one,ewcljrsq)));
1147             /* f6A = 6 * C6grid * (1 - poly) */
1148             f6A              = _mm_mul_ps(c6grid_00,_mm_sub_ps(one,poly));
1149             /* f6B = C6grid * exponent * beta^6 */
1150             f6B              = _mm_mul_ps(_mm_mul_ps(c6grid_00,one_sixth),_mm_mul_ps(exponent,ewclj6));
1151             /* fvdw = 12*C12/r13 - ((6*C6 - f6A)/r6 + f6B)/r */
1152             fvdw              = _mm_mul_ps(_mm_macc_ps(_mm_msub_ps(c12_00,rinvsix,_mm_sub_ps(c6_00,f6A)),rinvsix,f6B),rinvsq00);
1153
1154             fscal            = fvdw;
1155
1156             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1157
1158              /* Update vectorial force */
1159             fix0             = _mm_macc_ps(dx00,fscal,fix0);
1160             fiy0             = _mm_macc_ps(dy00,fscal,fiy0);
1161             fiz0             = _mm_macc_ps(dz00,fscal,fiz0);
1162
1163             fjx0             = _mm_macc_ps(dx00,fscal,fjx0);
1164             fjy0             = _mm_macc_ps(dy00,fscal,fjy0);
1165             fjz0             = _mm_macc_ps(dz00,fscal,fjz0);
1166
1167             /**************************
1168              * CALCULATE INTERACTIONS *
1169              **************************/
1170
1171             r10              = _mm_mul_ps(rsq10,rinv10);
1172             r10              = _mm_andnot_ps(dummy_mask,r10);
1173
1174             /* Compute parameters for interactions between i and j atoms */
1175             qq10             = _mm_mul_ps(iq1,jq0);
1176
1177             /* EWALD ELECTROSTATICS */
1178
1179             /* Analytical PME correction */
1180             zeta2            = _mm_mul_ps(beta2,rsq10);
1181             rinv3            = _mm_mul_ps(rinvsq10,rinv10);
1182             pmecorrF         = gmx_mm_pmecorrF_ps(zeta2);
1183             felec            = _mm_macc_ps(pmecorrF,beta3,rinv3);
1184             felec            = _mm_mul_ps(qq10,felec);
1185
1186             fscal            = felec;
1187
1188             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1189
1190              /* Update vectorial force */
1191             fix1             = _mm_macc_ps(dx10,fscal,fix1);
1192             fiy1             = _mm_macc_ps(dy10,fscal,fiy1);
1193             fiz1             = _mm_macc_ps(dz10,fscal,fiz1);
1194
1195             fjx0             = _mm_macc_ps(dx10,fscal,fjx0);
1196             fjy0             = _mm_macc_ps(dy10,fscal,fjy0);
1197             fjz0             = _mm_macc_ps(dz10,fscal,fjz0);
1198
1199             /**************************
1200              * CALCULATE INTERACTIONS *
1201              **************************/
1202
1203             r20              = _mm_mul_ps(rsq20,rinv20);
1204             r20              = _mm_andnot_ps(dummy_mask,r20);
1205
1206             /* Compute parameters for interactions between i and j atoms */
1207             qq20             = _mm_mul_ps(iq2,jq0);
1208
1209             /* EWALD ELECTROSTATICS */
1210
1211             /* Analytical PME correction */
1212             zeta2            = _mm_mul_ps(beta2,rsq20);
1213             rinv3            = _mm_mul_ps(rinvsq20,rinv20);
1214             pmecorrF         = gmx_mm_pmecorrF_ps(zeta2);
1215             felec            = _mm_macc_ps(pmecorrF,beta3,rinv3);
1216             felec            = _mm_mul_ps(qq20,felec);
1217
1218             fscal            = felec;
1219
1220             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1221
1222              /* Update vectorial force */
1223             fix2             = _mm_macc_ps(dx20,fscal,fix2);
1224             fiy2             = _mm_macc_ps(dy20,fscal,fiy2);
1225             fiz2             = _mm_macc_ps(dz20,fscal,fiz2);
1226
1227             fjx0             = _mm_macc_ps(dx20,fscal,fjx0);
1228             fjy0             = _mm_macc_ps(dy20,fscal,fjy0);
1229             fjz0             = _mm_macc_ps(dz20,fscal,fjz0);
1230
1231             /**************************
1232              * CALCULATE INTERACTIONS *
1233              **************************/
1234
1235             r30              = _mm_mul_ps(rsq30,rinv30);
1236             r30              = _mm_andnot_ps(dummy_mask,r30);
1237
1238             /* Compute parameters for interactions between i and j atoms */
1239             qq30             = _mm_mul_ps(iq3,jq0);
1240
1241             /* EWALD ELECTROSTATICS */
1242
1243             /* Analytical PME correction */
1244             zeta2            = _mm_mul_ps(beta2,rsq30);
1245             rinv3            = _mm_mul_ps(rinvsq30,rinv30);
1246             pmecorrF         = gmx_mm_pmecorrF_ps(zeta2);
1247             felec            = _mm_macc_ps(pmecorrF,beta3,rinv3);
1248             felec            = _mm_mul_ps(qq30,felec);
1249
1250             fscal            = felec;
1251
1252             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1253
1254              /* Update vectorial force */
1255             fix3             = _mm_macc_ps(dx30,fscal,fix3);
1256             fiy3             = _mm_macc_ps(dy30,fscal,fiy3);
1257             fiz3             = _mm_macc_ps(dz30,fscal,fiz3);
1258
1259             fjx0             = _mm_macc_ps(dx30,fscal,fjx0);
1260             fjy0             = _mm_macc_ps(dy30,fscal,fjy0);
1261             fjz0             = _mm_macc_ps(dz30,fscal,fjz0);
1262
1263             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
1264             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
1265             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
1266             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
1267
1268             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
1269
1270             /* Inner loop uses 135 flops */
1271         }
1272
1273         /* End of innermost loop */
1274
1275         gmx_mm_update_iforce_4atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
1276                                               f+i_coord_offset,fshift+i_shift_offset);
1277
1278         /* Increment number of inner iterations */
1279         inneriter                  += j_index_end - j_index_start;
1280
1281         /* Outer loop uses 24 flops */
1282     }
1283
1284     /* Increment number of outer iterations */
1285     outeriter        += nri;
1286
1287     /* Update outer/inner flops */
1288
1289     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W4_F,outeriter*24 + inneriter*135);
1290 }