src/gromacs/gmxlib/nonbonded/nb_kernel_avx_128_fma_single/nb_kernel_ElecEw_VdwLJEw_GeomW3P1_avx_128_fma_single.c

   1 /*
   2  * This file is part of the GROMACS molecular simulation package.
   3  *
   4  * Copyright (c) 2012,2013,2014, by the GROMACS development team, led by
   5  * Mark Abraham, David van der Spoel, Berk Hess, and Erik Lindahl,
   6  * and including many others, as listed in the AUTHORS file in the
   7  * top-level source directory and at http://www.gromacs.org.
   8  *
   9  * GROMACS is free software; you can redistribute it and/or
  10  * modify it under the terms of the GNU Lesser General Public License
  11  * as published by the Free Software Foundation; either version 2.1
  12  * of the License, or (at your option) any later version.
  13  *
  14  * GROMACS is distributed in the hope that it will be useful,
  15  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  16  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  17  * Lesser General Public License for more details.
  18  *
  19  * You should have received a copy of the GNU Lesser General Public
  20  * License along with GROMACS; if not, see
  21  * http://www.gnu.org/licenses, or write to the Free Software Foundation,
  22  * Inc., 51 Franklin Street, Fifth Floor, Boston, MA  02110-1301  USA.
  23  *
  24  * If you want to redistribute modifications to GROMACS, please
  25  * consider that scientific software is very special. Version
  26  * control is crucial - bugs must be traceable. We will be happy to
  27  * consider code for inclusion in the official distribution, but
  28  * derived work must not be called official GROMACS. Details are found
  29  * in the README & COPYING files - if they are missing, get the
  30  * official version at http://www.gromacs.org.
  31  *
  32  * To help us fund GROMACS development, we humbly ask that you cite
  33  * the research papers on the package. Check out http://www.gromacs.org.
  34  */
  35 /*
  36  * Note: this file was generated by the GROMACS avx_128_fma_single kernel generator.
  37  */
  38 #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_GeomW3P1_VF_avx_128_fma_single
  52  * Electrostatics interaction: Ewald
  53  * VdW interaction:            LJEwald
  54  * Geometry:                   Water3-Particle
  55  * Calculate force/pot:        PotentialAndForce
  56  */
  57 void
  58 nb_kernel_ElecEw_VdwLJEw_GeomW3P1_VF_avx_128_fma_single
  59                     (t_nblist                    * gmx_restrict       nlist,
  60                      rvec                        * gmx_restrict          xx,
  61                      rvec                        * gmx_restrict          ff,
  62                      t_forcerec                  * gmx_restrict          fr,
  63                      t_mdatoms                   * gmx_restrict     mdatoms,
  64                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
  65                      t_nrnb                      * gmx_restrict        nrnb)
  66 {
  67     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
  68      * just 0 for non-waters.
  69      * Suffixes A,B,C,D refer to j loop unrolling done with AVX_128, e.g. for the four different
  70      * jnr indices corresponding to data put in the four positions in the SIMD register.
  71      */
  72     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
  73     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
  74     int              jnrA,jnrB,jnrC,jnrD;
  75     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
  76     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
  77     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
  78     real             rcutoff_scalar;
  79     real             *shiftvec,*fshift,*x,*f;
  80     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
  81     real             scratch[4*DIM];
  82     __m128           fscal,rcutoff,rcutoff2,jidxall;
  83     int              vdwioffset0;
  84     __m128           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
  85     int              vdwioffset1;
  86     __m128           ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
  87     int              vdwioffset2;
  88     __m128           ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
  89     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
  90     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  91     __m128           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  92     __m128           dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
  93     __m128           dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
  94     __m128           velec,felec,velecsum,facel,crf,krf,krf2;
  95     real             *charge;
  96     int              nvdwtype;
  97     __m128           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
  98     int              *vdwtype;
  99     real             *vdwparam;
 100     __m128           one_sixth   = _mm_set1_ps(1.0/6.0);
 101     __m128           one_twelfth = _mm_set1_ps(1.0/12.0);
 102     __m128           c6grid_00;
 103     __m128           c6grid_10;
 104     __m128           c6grid_20;
 105     real             *vdwgridparam;
 106     __m128           ewclj,ewclj2,ewclj6,ewcljrsq,poly,exponent,f6A,f6B,sh_lj_ewald;
 107     __m128           one_half = _mm_set1_ps(0.5);
 108     __m128           minus_one = _mm_set1_ps(-1.0);
 109     __m128i          ewitab;
 110     __m128           ewtabscale,eweps,twoeweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
 111     __m128           beta,beta2,beta3,zeta2,pmecorrF,pmecorrV,rinv3;
 112     real             *ewtab;
 113     __m128           dummy_mask,cutoff_mask;
 114     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
 115     __m128           one     = _mm_set1_ps(1.0);
 116     __m128           two     = _mm_set1_ps(2.0);
 117     x                = xx[0];
 118     f                = ff[0];
 119
 120     nri              = nlist->nri;
 121     iinr             = nlist->iinr;
 122     jindex           = nlist->jindex;
 123     jjnr             = nlist->jjnr;
 124     shiftidx         = nlist->shift;
 125     gid              = nlist->gid;
 126     shiftvec         = fr->shift_vec[0];
 127     fshift           = fr->fshift[0];
 128     facel            = _mm_set1_ps(fr->epsfac);
 129     charge           = mdatoms->chargeA;
 130     nvdwtype         = fr->ntype;
 131     vdwparam         = fr->nbfp;
 132     vdwtype          = mdatoms->typeA;
 133     vdwgridparam     = fr->ljpme_c6grid;
 134     sh_lj_ewald      = _mm_set1_ps(fr->ic->sh_lj_ewald);
 135     ewclj            = _mm_set1_ps(fr->ewaldcoeff_lj);
 136     ewclj2           = _mm_mul_ps(minus_one,_mm_mul_ps(ewclj,ewclj));
 137
 138     sh_ewald         = _mm_set1_ps(fr->ic->sh_ewald);
 139     beta             = _mm_set1_ps(fr->ic->ewaldcoeff_q);
 140     beta2            = _mm_mul_ps(beta,beta);
 141     beta3            = _mm_mul_ps(beta,beta2);
 142     ewtab            = fr->ic->tabq_coul_FDV0;
 143     ewtabscale       = _mm_set1_ps(fr->ic->tabq_scale);
 144     ewtabhalfspace   = _mm_set1_ps(0.5/fr->ic->tabq_scale);
 145
 146     /* Setup water-specific parameters */
 147     inr              = nlist->iinr[0];
 148     iq0              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+0]));
 149     iq1              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+1]));
 150     iq2              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+2]));
 151     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 152
 153     /* Avoid stupid compiler warnings */
 154     jnrA = jnrB = jnrC = jnrD = 0;
 155     j_coord_offsetA = 0;
 156     j_coord_offsetB = 0;
 157     j_coord_offsetC = 0;
 158     j_coord_offsetD = 0;
 159
 160     outeriter        = 0;
 161     inneriter        = 0;
 162
 163     for(iidx=0;iidx<4*DIM;iidx++)
 164     {
 165         scratch[iidx] = 0.0;
 166     }
 167
 168     /* Start outer loop over neighborlists */
 169     for(iidx=0; iidx<nri; iidx++)
 170     {
 171         /* Load shift vector for this list */
 172         i_shift_offset   = DIM*shiftidx[iidx];
 173
 174         /* Load limits for loop over neighbors */
 175         j_index_start    = jindex[iidx];
 176         j_index_end      = jindex[iidx+1];
 177
 178         /* Get outer coordinate index */
 179         inr              = iinr[iidx];
 180         i_coord_offset   = DIM*inr;
 181
 182         /* Load i particle coords and add shift vector */
 183         gmx_mm_load_shift_and_3rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
 184                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 185
 186         fix0             = _mm_setzero_ps();
 187         fiy0             = _mm_setzero_ps();
 188         fiz0             = _mm_setzero_ps();
 189         fix1             = _mm_setzero_ps();
 190         fiy1             = _mm_setzero_ps();
 191         fiz1             = _mm_setzero_ps();
 192         fix2             = _mm_setzero_ps();
 193         fiy2             = _mm_setzero_ps();
 194         fiz2             = _mm_setzero_ps();
 195
 196         /* Reset potential sums */
 197         velecsum         = _mm_setzero_ps();
 198         vvdwsum          = _mm_setzero_ps();
 199
 200         /* Start inner kernel loop */
 201         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 202         {
 203
 204             /* Get j neighbor index, and coordinate index */
 205             jnrA             = jjnr[jidx];
 206             jnrB             = jjnr[jidx+1];
 207             jnrC             = jjnr[jidx+2];
 208             jnrD             = jjnr[jidx+3];
 209             j_coord_offsetA  = DIM*jnrA;
 210             j_coord_offsetB  = DIM*jnrB;
 211             j_coord_offsetC  = DIM*jnrC;
 212             j_coord_offsetD  = DIM*jnrD;
 213
 214             /* load j atom coordinates */
 215             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 216                                               x+j_coord_offsetC,x+j_coord_offsetD,
 217                                               &jx0,&jy0,&jz0);
 218
 219             /* Calculate displacement vector */
 220             dx00             = _mm_sub_ps(ix0,jx0);
 221             dy00             = _mm_sub_ps(iy0,jy0);
 222             dz00             = _mm_sub_ps(iz0,jz0);
 223             dx10             = _mm_sub_ps(ix1,jx0);
 224             dy10             = _mm_sub_ps(iy1,jy0);
 225             dz10             = _mm_sub_ps(iz1,jz0);
 226             dx20             = _mm_sub_ps(ix2,jx0);
 227             dy20             = _mm_sub_ps(iy2,jy0);
 228             dz20             = _mm_sub_ps(iz2,jz0);
 229
 230             /* Calculate squared distance and things based on it */
 231             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 232             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 233             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 234
 235             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 236             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 237             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 238
 239             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 240             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 241             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 242
 243             /* Load parameters for j particles */
 244             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 245                                                               charge+jnrC+0,charge+jnrD+0);
 246             vdwjidx0A        = 2*vdwtype[jnrA+0];
 247             vdwjidx0B        = 2*vdwtype[jnrB+0];
 248             vdwjidx0C        = 2*vdwtype[jnrC+0];
 249             vdwjidx0D        = 2*vdwtype[jnrD+0];
 250
 251             fjx0             = _mm_setzero_ps();
 252             fjy0             = _mm_setzero_ps();
 253             fjz0             = _mm_setzero_ps();
 254
 255             /**************************
 256              * CALCULATE INTERACTIONS *
 257              **************************/
 258
 259             r00              = _mm_mul_ps(rsq00,rinv00);
 260
 261             /* Compute parameters for interactions between i and j atoms */
 262             qq00             = _mm_mul_ps(iq0,jq0);
 263             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 264                                          vdwparam+vdwioffset0+vdwjidx0B,
 265                                          vdwparam+vdwioffset0+vdwjidx0C,
 266                                          vdwparam+vdwioffset0+vdwjidx0D,
 267                                          &c6_00,&c12_00);
 268
 269             c6grid_00       = gmx_mm_load_4real_swizzle_ps(vdwgridparam+vdwioffset0+vdwjidx0A,
 270                                                                vdwgridparam+vdwioffset0+vdwjidx0B,
 271                                                                vdwgridparam+vdwioffset0+vdwjidx0C,
 272                                                                vdwgridparam+vdwioffset0+vdwjidx0D);
 273
 274             /* EWALD ELECTROSTATICS */
 275
 276             /* Analytical PME correction */
 277             zeta2            = _mm_mul_ps(beta2,rsq00);
 278             rinv3            = _mm_mul_ps(rinvsq00,rinv00);
 279             pmecorrF         = gmx_mm_pmecorrF_ps(zeta2);
 280             felec            = _mm_macc_ps(pmecorrF,beta3,rinv3);
 281             felec            = _mm_mul_ps(qq00,felec);
 282             pmecorrV         = gmx_mm_pmecorrV_ps(zeta2);
 283             velec            = _mm_nmacc_ps(pmecorrV,beta,rinv00);
 284             velec            = _mm_mul_ps(qq00,velec);
 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             velecsum         = _mm_add_ps(velecsum,velec);
 302             vvdwsum          = _mm_add_ps(vvdwsum,vvdw);
 303
 304             fscal            = _mm_add_ps(felec,fvdw);
 305
 306              /* Update vectorial force */
 307             fix0             = _mm_macc_ps(dx00,fscal,fix0);
 308             fiy0             = _mm_macc_ps(dy00,fscal,fiy0);
 309             fiz0             = _mm_macc_ps(dz00,fscal,fiz0);
 310
 311             fjx0             = _mm_macc_ps(dx00,fscal,fjx0);
 312             fjy0             = _mm_macc_ps(dy00,fscal,fjy0);
 313             fjz0             = _mm_macc_ps(dz00,fscal,fjz0);
 314
 315             /**************************
 316              * CALCULATE INTERACTIONS *
 317              **************************/
 318
 319             r10              = _mm_mul_ps(rsq10,rinv10);
 320
 321             /* Compute parameters for interactions between i and j atoms */
 322             qq10             = _mm_mul_ps(iq1,jq0);
 323
 324             /* EWALD ELECTROSTATICS */
 325
 326             /* Analytical PME correction */
 327             zeta2            = _mm_mul_ps(beta2,rsq10);
 328             rinv3            = _mm_mul_ps(rinvsq10,rinv10);
 329             pmecorrF         = gmx_mm_pmecorrF_ps(zeta2);
 330             felec            = _mm_macc_ps(pmecorrF,beta3,rinv3);
 331             felec            = _mm_mul_ps(qq10,felec);
 332             pmecorrV         = gmx_mm_pmecorrV_ps(zeta2);
 333             velec            = _mm_nmacc_ps(pmecorrV,beta,rinv10);
 334             velec            = _mm_mul_ps(qq10,velec);
 335
 336             /* Update potential sum for this i atom from the interaction with this j atom. */
 337             velecsum         = _mm_add_ps(velecsum,velec);
 338
 339             fscal            = felec;
 340
 341              /* Update vectorial force */
 342             fix1             = _mm_macc_ps(dx10,fscal,fix1);
 343             fiy1             = _mm_macc_ps(dy10,fscal,fiy1);
 344             fiz1             = _mm_macc_ps(dz10,fscal,fiz1);
 345
 346             fjx0             = _mm_macc_ps(dx10,fscal,fjx0);
 347             fjy0             = _mm_macc_ps(dy10,fscal,fjy0);
 348             fjz0             = _mm_macc_ps(dz10,fscal,fjz0);
 349
 350             /**************************
 351              * CALCULATE INTERACTIONS *
 352              **************************/
 353
 354             r20              = _mm_mul_ps(rsq20,rinv20);
 355
 356             /* Compute parameters for interactions between i and j atoms */
 357             qq20             = _mm_mul_ps(iq2,jq0);
 358
 359             /* EWALD ELECTROSTATICS */
 360
 361             /* Analytical PME correction */
 362             zeta2            = _mm_mul_ps(beta2,rsq20);
 363             rinv3            = _mm_mul_ps(rinvsq20,rinv20);
 364             pmecorrF         = gmx_mm_pmecorrF_ps(zeta2);
 365             felec            = _mm_macc_ps(pmecorrF,beta3,rinv3);
 366             felec            = _mm_mul_ps(qq20,felec);
 367             pmecorrV         = gmx_mm_pmecorrV_ps(zeta2);
 368             velec            = _mm_nmacc_ps(pmecorrV,beta,rinv20);
 369             velec            = _mm_mul_ps(qq20,velec);
 370
 371             /* Update potential sum for this i atom from the interaction with this j atom. */
 372             velecsum         = _mm_add_ps(velecsum,velec);
 373
 374             fscal            = felec;
 375
 376              /* Update vectorial force */
 377             fix2             = _mm_macc_ps(dx20,fscal,fix2);
 378             fiy2             = _mm_macc_ps(dy20,fscal,fiy2);
 379             fiz2             = _mm_macc_ps(dz20,fscal,fiz2);
 380
 381             fjx0             = _mm_macc_ps(dx20,fscal,fjx0);
 382             fjy0             = _mm_macc_ps(dy20,fscal,fjy0);
 383             fjz0             = _mm_macc_ps(dz20,fscal,fjz0);
 384
 385             fjptrA             = f+j_coord_offsetA;
 386             fjptrB             = f+j_coord_offsetB;
 387             fjptrC             = f+j_coord_offsetC;
 388             fjptrD             = f+j_coord_offsetD;
 389
 390             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 391
 392             /* Inner loop uses 111 flops */
 393         }
 394
 395         if(jidx<j_index_end)
 396         {
 397
 398             /* Get j neighbor index, and coordinate index */
 399             jnrlistA         = jjnr[jidx];
 400             jnrlistB         = jjnr[jidx+1];
 401             jnrlistC         = jjnr[jidx+2];
 402             jnrlistD         = jjnr[jidx+3];
 403             /* Sign of each element will be negative for non-real atoms.
 404              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 405              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 406              */
 407             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 408             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 409             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 410             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 411             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 412             j_coord_offsetA  = DIM*jnrA;
 413             j_coord_offsetB  = DIM*jnrB;
 414             j_coord_offsetC  = DIM*jnrC;
 415             j_coord_offsetD  = DIM*jnrD;
 416
 417             /* load j atom coordinates */
 418             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 419                                               x+j_coord_offsetC,x+j_coord_offsetD,
 420                                               &jx0,&jy0,&jz0);
 421
 422             /* Calculate displacement vector */
 423             dx00             = _mm_sub_ps(ix0,jx0);
 424             dy00             = _mm_sub_ps(iy0,jy0);
 425             dz00             = _mm_sub_ps(iz0,jz0);
 426             dx10             = _mm_sub_ps(ix1,jx0);
 427             dy10             = _mm_sub_ps(iy1,jy0);
 428             dz10             = _mm_sub_ps(iz1,jz0);
 429             dx20             = _mm_sub_ps(ix2,jx0);
 430             dy20             = _mm_sub_ps(iy2,jy0);
 431             dz20             = _mm_sub_ps(iz2,jz0);
 432
 433             /* Calculate squared distance and things based on it */
 434             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 435             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 436             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 437
 438             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 439             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 440             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 441
 442             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 443             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 444             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 445
 446             /* Load parameters for j particles */
 447             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 448                                                               charge+jnrC+0,charge+jnrD+0);
 449             vdwjidx0A        = 2*vdwtype[jnrA+0];
 450             vdwjidx0B        = 2*vdwtype[jnrB+0];
 451             vdwjidx0C        = 2*vdwtype[jnrC+0];
 452             vdwjidx0D        = 2*vdwtype[jnrD+0];
 453
 454             fjx0             = _mm_setzero_ps();
 455             fjy0             = _mm_setzero_ps();
 456             fjz0             = _mm_setzero_ps();
 457
 458             /**************************
 459              * CALCULATE INTERACTIONS *
 460              **************************/
 461
 462             r00              = _mm_mul_ps(rsq00,rinv00);
 463             r00              = _mm_andnot_ps(dummy_mask,r00);
 464
 465             /* Compute parameters for interactions between i and j atoms */
 466             qq00             = _mm_mul_ps(iq0,jq0);
 467             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 468                                          vdwparam+vdwioffset0+vdwjidx0B,
 469                                          vdwparam+vdwioffset0+vdwjidx0C,
 470                                          vdwparam+vdwioffset0+vdwjidx0D,
 471                                          &c6_00,&c12_00);
 472
 473             c6grid_00       = gmx_mm_load_4real_swizzle_ps(vdwgridparam+vdwioffset0+vdwjidx0A,
 474                                                                vdwgridparam+vdwioffset0+vdwjidx0B,
 475                                                                vdwgridparam+vdwioffset0+vdwjidx0C,
 476                                                                vdwgridparam+vdwioffset0+vdwjidx0D);
 477
 478             /* EWALD ELECTROSTATICS */
 479
 480             /* Analytical PME correction */
 481             zeta2            = _mm_mul_ps(beta2,rsq00);
 482             rinv3            = _mm_mul_ps(rinvsq00,rinv00);
 483             pmecorrF         = gmx_mm_pmecorrF_ps(zeta2);
 484             felec            = _mm_macc_ps(pmecorrF,beta3,rinv3);
 485             felec            = _mm_mul_ps(qq00,felec);
 486             pmecorrV         = gmx_mm_pmecorrV_ps(zeta2);
 487             velec            = _mm_nmacc_ps(pmecorrV,beta,rinv00);
 488             velec            = _mm_mul_ps(qq00,velec);
 489
 490             /* Analytical LJ-PME */
 491             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 492             ewcljrsq         = _mm_mul_ps(ewclj2,rsq00);
 493             ewclj6           = _mm_mul_ps(ewclj2,_mm_mul_ps(ewclj2,ewclj2));
 494             exponent         = gmx_simd_exp_r(ewcljrsq);
 495             /* poly = exp(-(beta*r)^2) * (1 + (beta*r)^2 + (beta*r)^4 /2) */
 496             poly             = _mm_mul_ps(exponent,_mm_macc_ps(_mm_mul_ps(ewcljrsq,ewcljrsq),one_half,_mm_sub_ps(one,ewcljrsq)));
 497             /* vvdw6 = [C6 - C6grid * (1-poly)]/r6 */
 498             vvdw6            = _mm_mul_ps(_mm_macc_ps(-c6grid_00,_mm_sub_ps(one,poly),c6_00),rinvsix);
 499             vvdw12           = _mm_mul_ps(c12_00,_mm_mul_ps(rinvsix,rinvsix));
 500             vvdw             = _mm_msub_ps(vvdw12,one_twelfth,_mm_mul_ps(vvdw6,one_sixth));
 501             /* fvdw = vvdw12/r - (vvdw6/r + (C6grid * exponent * beta^6)/r) */
 502             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);
 503
 504             /* Update potential sum for this i atom from the interaction with this j atom. */
 505             velec            = _mm_andnot_ps(dummy_mask,velec);
 506             velecsum         = _mm_add_ps(velecsum,velec);
 507             vvdw             = _mm_andnot_ps(dummy_mask,vvdw);
 508             vvdwsum          = _mm_add_ps(vvdwsum,vvdw);
 509
 510             fscal            = _mm_add_ps(felec,fvdw);
 511
 512             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 513
 514              /* Update vectorial force */
 515             fix0             = _mm_macc_ps(dx00,fscal,fix0);
 516             fiy0             = _mm_macc_ps(dy00,fscal,fiy0);
 517             fiz0             = _mm_macc_ps(dz00,fscal,fiz0);
 518
 519             fjx0             = _mm_macc_ps(dx00,fscal,fjx0);
 520             fjy0             = _mm_macc_ps(dy00,fscal,fjy0);
 521             fjz0             = _mm_macc_ps(dz00,fscal,fjz0);
 522
 523             /**************************
 524              * CALCULATE INTERACTIONS *
 525              **************************/
 526
 527             r10              = _mm_mul_ps(rsq10,rinv10);
 528             r10              = _mm_andnot_ps(dummy_mask,r10);
 529
 530             /* Compute parameters for interactions between i and j atoms */
 531             qq10             = _mm_mul_ps(iq1,jq0);
 532
 533             /* EWALD ELECTROSTATICS */
 534
 535             /* Analytical PME correction */
 536             zeta2            = _mm_mul_ps(beta2,rsq10);
 537             rinv3            = _mm_mul_ps(rinvsq10,rinv10);
 538             pmecorrF         = gmx_mm_pmecorrF_ps(zeta2);
 539             felec            = _mm_macc_ps(pmecorrF,beta3,rinv3);
 540             felec            = _mm_mul_ps(qq10,felec);
 541             pmecorrV         = gmx_mm_pmecorrV_ps(zeta2);
 542             velec            = _mm_nmacc_ps(pmecorrV,beta,rinv10);
 543             velec            = _mm_mul_ps(qq10,velec);
 544
 545             /* Update potential sum for this i atom from the interaction with this j atom. */
 546             velec            = _mm_andnot_ps(dummy_mask,velec);
 547             velecsum         = _mm_add_ps(velecsum,velec);
 548
 549             fscal            = felec;
 550
 551             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 552
 553              /* Update vectorial force */
 554             fix1             = _mm_macc_ps(dx10,fscal,fix1);
 555             fiy1             = _mm_macc_ps(dy10,fscal,fiy1);
 556             fiz1             = _mm_macc_ps(dz10,fscal,fiz1);
 557
 558             fjx0             = _mm_macc_ps(dx10,fscal,fjx0);
 559             fjy0             = _mm_macc_ps(dy10,fscal,fjy0);
 560             fjz0             = _mm_macc_ps(dz10,fscal,fjz0);
 561
 562             /**************************
 563              * CALCULATE INTERACTIONS *
 564              **************************/
 565
 566             r20              = _mm_mul_ps(rsq20,rinv20);
 567             r20              = _mm_andnot_ps(dummy_mask,r20);
 568
 569             /* Compute parameters for interactions between i and j atoms */
 570             qq20             = _mm_mul_ps(iq2,jq0);
 571
 572             /* EWALD ELECTROSTATICS */
 573
 574             /* Analytical PME correction */
 575             zeta2            = _mm_mul_ps(beta2,rsq20);
 576             rinv3            = _mm_mul_ps(rinvsq20,rinv20);
 577             pmecorrF         = gmx_mm_pmecorrF_ps(zeta2);
 578             felec            = _mm_macc_ps(pmecorrF,beta3,rinv3);
 579             felec            = _mm_mul_ps(qq20,felec);
 580             pmecorrV         = gmx_mm_pmecorrV_ps(zeta2);
 581             velec            = _mm_nmacc_ps(pmecorrV,beta,rinv20);
 582             velec            = _mm_mul_ps(qq20,velec);
 583
 584             /* Update potential sum for this i atom from the interaction with this j atom. */
 585             velec            = _mm_andnot_ps(dummy_mask,velec);
 586             velecsum         = _mm_add_ps(velecsum,velec);
 587
 588             fscal            = felec;
 589
 590             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 591
 592              /* Update vectorial force */
 593             fix2             = _mm_macc_ps(dx20,fscal,fix2);
 594             fiy2             = _mm_macc_ps(dy20,fscal,fiy2);
 595             fiz2             = _mm_macc_ps(dz20,fscal,fiz2);
 596
 597             fjx0             = _mm_macc_ps(dx20,fscal,fjx0);
 598             fjy0             = _mm_macc_ps(dy20,fscal,fjy0);
 599             fjz0             = _mm_macc_ps(dz20,fscal,fjz0);
 600
 601             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 602             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 603             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 604             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 605
 606             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 607
 608             /* Inner loop uses 114 flops */
 609         }
 610
 611         /* End of innermost loop */
 612
 613         gmx_mm_update_iforce_3atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
 614                                               f+i_coord_offset,fshift+i_shift_offset);
 615
 616         ggid                        = gid[iidx];
 617         /* Update potential energies */
 618         gmx_mm_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid);
 619         gmx_mm_update_1pot_ps(vvdwsum,kernel_data->energygrp_vdw+ggid);
 620
 621         /* Increment number of inner iterations */
 622         inneriter                  += j_index_end - j_index_start;
 623
 624         /* Outer loop uses 20 flops */
 625     }
 626
 627     /* Increment number of outer iterations */
 628     outeriter        += nri;
 629
 630     /* Update outer/inner flops */
 631
 632     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3_VF,outeriter*20 + inneriter*114);
 633 }
 634 /*
 635  * Gromacs nonbonded kernel:   nb_kernel_ElecEw_VdwLJEw_GeomW3P1_F_avx_128_fma_single
 636  * Electrostatics interaction: Ewald
 637  * VdW interaction:            LJEwald
 638  * Geometry:                   Water3-Particle
 639  * Calculate force/pot:        Force
 640  */
 641 void
 642 nb_kernel_ElecEw_VdwLJEw_GeomW3P1_F_avx_128_fma_single
 643                     (t_nblist                    * gmx_restrict       nlist,
 644                      rvec                        * gmx_restrict          xx,
 645                      rvec                        * gmx_restrict          ff,
 646                      t_forcerec                  * gmx_restrict          fr,
 647                      t_mdatoms                   * gmx_restrict     mdatoms,
 648                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 649                      t_nrnb                      * gmx_restrict        nrnb)
 650 {
 651     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 652      * just 0 for non-waters.
 653      * Suffixes A,B,C,D refer to j loop unrolling done with AVX_128, e.g. for the four different
 654      * jnr indices corresponding to data put in the four positions in the SIMD register.
 655      */
 656     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 657     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 658     int              jnrA,jnrB,jnrC,jnrD;
 659     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
 660     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 661     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 662     real             rcutoff_scalar;
 663     real             *shiftvec,*fshift,*x,*f;
 664     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
 665     real             scratch[4*DIM];
 666     __m128           fscal,rcutoff,rcutoff2,jidxall;
 667     int              vdwioffset0;
 668     __m128           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 669     int              vdwioffset1;
 670     __m128           ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
 671     int              vdwioffset2;
 672     __m128           ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
 673     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
 674     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 675     __m128           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 676     __m128           dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
 677     __m128           dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
 678     __m128           velec,felec,velecsum,facel,crf,krf,krf2;
 679     real             *charge;
 680     int              nvdwtype;
 681     __m128           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 682     int              *vdwtype;
 683     real             *vdwparam;
 684     __m128           one_sixth   = _mm_set1_ps(1.0/6.0);
 685     __m128           one_twelfth = _mm_set1_ps(1.0/12.0);
 686     __m128           c6grid_00;
 687     __m128           c6grid_10;
 688     __m128           c6grid_20;
 689     real             *vdwgridparam;
 690     __m128           ewclj,ewclj2,ewclj6,ewcljrsq,poly,exponent,f6A,f6B,sh_lj_ewald;
 691     __m128           one_half = _mm_set1_ps(0.5);
 692     __m128           minus_one = _mm_set1_ps(-1.0);
 693     __m128i          ewitab;
 694     __m128           ewtabscale,eweps,twoeweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
 695     __m128           beta,beta2,beta3,zeta2,pmecorrF,pmecorrV,rinv3;
 696     real             *ewtab;
 697     __m128           dummy_mask,cutoff_mask;
 698     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
 699     __m128           one     = _mm_set1_ps(1.0);
 700     __m128           two     = _mm_set1_ps(2.0);
 701     x                = xx[0];
 702     f                = ff[0];
 703
 704     nri              = nlist->nri;
 705     iinr             = nlist->iinr;
 706     jindex           = nlist->jindex;
 707     jjnr             = nlist->jjnr;
 708     shiftidx         = nlist->shift;
 709     gid              = nlist->gid;
 710     shiftvec         = fr->shift_vec[0];
 711     fshift           = fr->fshift[0];
 712     facel            = _mm_set1_ps(fr->epsfac);
 713     charge           = mdatoms->chargeA;
 714     nvdwtype         = fr->ntype;
 715     vdwparam         = fr->nbfp;
 716     vdwtype          = mdatoms->typeA;
 717     vdwgridparam     = fr->ljpme_c6grid;
 718     sh_lj_ewald      = _mm_set1_ps(fr->ic->sh_lj_ewald);
 719     ewclj            = _mm_set1_ps(fr->ewaldcoeff_lj);
 720     ewclj2           = _mm_mul_ps(minus_one,_mm_mul_ps(ewclj,ewclj));
 721
 722     sh_ewald         = _mm_set1_ps(fr->ic->sh_ewald);
 723     beta             = _mm_set1_ps(fr->ic->ewaldcoeff_q);
 724     beta2            = _mm_mul_ps(beta,beta);
 725     beta3            = _mm_mul_ps(beta,beta2);
 726     ewtab            = fr->ic->tabq_coul_F;
 727     ewtabscale       = _mm_set1_ps(fr->ic->tabq_scale);
 728     ewtabhalfspace   = _mm_set1_ps(0.5/fr->ic->tabq_scale);
 729
 730     /* Setup water-specific parameters */
 731     inr              = nlist->iinr[0];
 732     iq0              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+0]));
 733     iq1              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+1]));
 734     iq2              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+2]));
 735     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 736
 737     /* Avoid stupid compiler warnings */
 738     jnrA = jnrB = jnrC = jnrD = 0;
 739     j_coord_offsetA = 0;
 740     j_coord_offsetB = 0;
 741     j_coord_offsetC = 0;
 742     j_coord_offsetD = 0;
 743
 744     outeriter        = 0;
 745     inneriter        = 0;
 746
 747     for(iidx=0;iidx<4*DIM;iidx++)
 748     {
 749         scratch[iidx] = 0.0;
 750     }
 751
 752     /* Start outer loop over neighborlists */
 753     for(iidx=0; iidx<nri; iidx++)
 754     {
 755         /* Load shift vector for this list */
 756         i_shift_offset   = DIM*shiftidx[iidx];
 757
 758         /* Load limits for loop over neighbors */
 759         j_index_start    = jindex[iidx];
 760         j_index_end      = jindex[iidx+1];
 761
 762         /* Get outer coordinate index */
 763         inr              = iinr[iidx];
 764         i_coord_offset   = DIM*inr;
 765
 766         /* Load i particle coords and add shift vector */
 767         gmx_mm_load_shift_and_3rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
 768                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 769
 770         fix0             = _mm_setzero_ps();
 771         fiy0             = _mm_setzero_ps();
 772         fiz0             = _mm_setzero_ps();
 773         fix1             = _mm_setzero_ps();
 774         fiy1             = _mm_setzero_ps();
 775         fiz1             = _mm_setzero_ps();
 776         fix2             = _mm_setzero_ps();
 777         fiy2             = _mm_setzero_ps();
 778         fiz2             = _mm_setzero_ps();
 779
 780         /* Start inner kernel loop */
 781         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 782         {
 783
 784             /* Get j neighbor index, and coordinate index */
 785             jnrA             = jjnr[jidx];
 786             jnrB             = jjnr[jidx+1];
 787             jnrC             = jjnr[jidx+2];
 788             jnrD             = jjnr[jidx+3];
 789             j_coord_offsetA  = DIM*jnrA;
 790             j_coord_offsetB  = DIM*jnrB;
 791             j_coord_offsetC  = DIM*jnrC;
 792             j_coord_offsetD  = DIM*jnrD;
 793
 794             /* load j atom coordinates */
 795             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 796                                               x+j_coord_offsetC,x+j_coord_offsetD,
 797                                               &jx0,&jy0,&jz0);
 798
 799             /* Calculate displacement vector */
 800             dx00             = _mm_sub_ps(ix0,jx0);
 801             dy00             = _mm_sub_ps(iy0,jy0);
 802             dz00             = _mm_sub_ps(iz0,jz0);
 803             dx10             = _mm_sub_ps(ix1,jx0);
 804             dy10             = _mm_sub_ps(iy1,jy0);
 805             dz10             = _mm_sub_ps(iz1,jz0);
 806             dx20             = _mm_sub_ps(ix2,jx0);
 807             dy20             = _mm_sub_ps(iy2,jy0);
 808             dz20             = _mm_sub_ps(iz2,jz0);
 809
 810             /* Calculate squared distance and things based on it */
 811             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 812             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 813             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 814
 815             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 816             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 817             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 818
 819             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 820             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 821             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 822
 823             /* Load parameters for j particles */
 824             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 825                                                               charge+jnrC+0,charge+jnrD+0);
 826             vdwjidx0A        = 2*vdwtype[jnrA+0];
 827             vdwjidx0B        = 2*vdwtype[jnrB+0];
 828             vdwjidx0C        = 2*vdwtype[jnrC+0];
 829             vdwjidx0D        = 2*vdwtype[jnrD+0];
 830
 831             fjx0             = _mm_setzero_ps();
 832             fjy0             = _mm_setzero_ps();
 833             fjz0             = _mm_setzero_ps();
 834
 835             /**************************
 836              * CALCULATE INTERACTIONS *
 837              **************************/
 838
 839             r00              = _mm_mul_ps(rsq00,rinv00);
 840
 841             /* Compute parameters for interactions between i and j atoms */
 842             qq00             = _mm_mul_ps(iq0,jq0);
 843             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 844                                          vdwparam+vdwioffset0+vdwjidx0B,
 845                                          vdwparam+vdwioffset0+vdwjidx0C,
 846                                          vdwparam+vdwioffset0+vdwjidx0D,
 847                                          &c6_00,&c12_00);
 848
 849             c6grid_00       = gmx_mm_load_4real_swizzle_ps(vdwgridparam+vdwioffset0+vdwjidx0A,
 850                                                                vdwgridparam+vdwioffset0+vdwjidx0B,
 851                                                                vdwgridparam+vdwioffset0+vdwjidx0C,
 852                                                                vdwgridparam+vdwioffset0+vdwjidx0D);
 853
 854             /* EWALD ELECTROSTATICS */
 855
 856             /* Analytical PME correction */
 857             zeta2            = _mm_mul_ps(beta2,rsq00);
 858             rinv3            = _mm_mul_ps(rinvsq00,rinv00);
 859             pmecorrF         = gmx_mm_pmecorrF_ps(zeta2);
 860             felec            = _mm_macc_ps(pmecorrF,beta3,rinv3);
 861             felec            = _mm_mul_ps(qq00,felec);
 862
 863             /* Analytical LJ-PME */
 864             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 865             ewcljrsq         = _mm_mul_ps(ewclj2,rsq00);
 866             ewclj6           = _mm_mul_ps(ewclj2,_mm_mul_ps(ewclj2,ewclj2));
 867             exponent         = gmx_simd_exp_r(ewcljrsq);
 868             /* poly = exp(-(beta*r)^2) * (1 + (beta*r)^2 + (beta*r)^4 /2) */
 869             poly             = _mm_mul_ps(exponent,_mm_macc_ps(_mm_mul_ps(ewcljrsq,ewcljrsq),one_half,_mm_sub_ps(one,ewcljrsq)));
 870             /* f6A = 6 * C6grid * (1 - poly) */
 871             f6A              = _mm_mul_ps(c6grid_00,_mm_sub_ps(one,poly));
 872             /* f6B = C6grid * exponent * beta^6 */
 873             f6B              = _mm_mul_ps(_mm_mul_ps(c6grid_00,one_sixth),_mm_mul_ps(exponent,ewclj6));
 874             /* fvdw = 12*C12/r13 - ((6*C6 - f6A)/r6 + f6B)/r */
 875             fvdw              = _mm_mul_ps(_mm_macc_ps(_mm_msub_ps(c12_00,rinvsix,_mm_sub_ps(c6_00,f6A)),rinvsix,f6B),rinvsq00);
 876
 877             fscal            = _mm_add_ps(felec,fvdw);
 878
 879              /* Update vectorial force */
 880             fix0             = _mm_macc_ps(dx00,fscal,fix0);
 881             fiy0             = _mm_macc_ps(dy00,fscal,fiy0);
 882             fiz0             = _mm_macc_ps(dz00,fscal,fiz0);
 883
 884             fjx0             = _mm_macc_ps(dx00,fscal,fjx0);
 885             fjy0             = _mm_macc_ps(dy00,fscal,fjy0);
 886             fjz0             = _mm_macc_ps(dz00,fscal,fjz0);
 887
 888             /**************************
 889              * CALCULATE INTERACTIONS *
 890              **************************/
 891
 892             r10              = _mm_mul_ps(rsq10,rinv10);
 893
 894             /* Compute parameters for interactions between i and j atoms */
 895             qq10             = _mm_mul_ps(iq1,jq0);
 896
 897             /* EWALD ELECTROSTATICS */
 898
 899             /* Analytical PME correction */
 900             zeta2            = _mm_mul_ps(beta2,rsq10);
 901             rinv3            = _mm_mul_ps(rinvsq10,rinv10);
 902             pmecorrF         = gmx_mm_pmecorrF_ps(zeta2);
 903             felec            = _mm_macc_ps(pmecorrF,beta3,rinv3);
 904             felec            = _mm_mul_ps(qq10,felec);
 905
 906             fscal            = felec;
 907
 908              /* Update vectorial force */
 909             fix1             = _mm_macc_ps(dx10,fscal,fix1);
 910             fiy1             = _mm_macc_ps(dy10,fscal,fiy1);
 911             fiz1             = _mm_macc_ps(dz10,fscal,fiz1);
 912
 913             fjx0             = _mm_macc_ps(dx10,fscal,fjx0);
 914             fjy0             = _mm_macc_ps(dy10,fscal,fjy0);
 915             fjz0             = _mm_macc_ps(dz10,fscal,fjz0);
 916
 917             /**************************
 918              * CALCULATE INTERACTIONS *
 919              **************************/
 920
 921             r20              = _mm_mul_ps(rsq20,rinv20);
 922
 923             /* Compute parameters for interactions between i and j atoms */
 924             qq20             = _mm_mul_ps(iq2,jq0);
 925
 926             /* EWALD ELECTROSTATICS */
 927
 928             /* Analytical PME correction */
 929             zeta2            = _mm_mul_ps(beta2,rsq20);
 930             rinv3            = _mm_mul_ps(rinvsq20,rinv20);
 931             pmecorrF         = gmx_mm_pmecorrF_ps(zeta2);
 932             felec            = _mm_macc_ps(pmecorrF,beta3,rinv3);
 933             felec            = _mm_mul_ps(qq20,felec);
 934
 935             fscal            = felec;
 936
 937              /* Update vectorial force */
 938             fix2             = _mm_macc_ps(dx20,fscal,fix2);
 939             fiy2             = _mm_macc_ps(dy20,fscal,fiy2);
 940             fiz2             = _mm_macc_ps(dz20,fscal,fiz2);
 941
 942             fjx0             = _mm_macc_ps(dx20,fscal,fjx0);
 943             fjy0             = _mm_macc_ps(dy20,fscal,fjy0);
 944             fjz0             = _mm_macc_ps(dz20,fscal,fjz0);
 945
 946             fjptrA             = f+j_coord_offsetA;
 947             fjptrB             = f+j_coord_offsetB;
 948             fjptrC             = f+j_coord_offsetC;
 949             fjptrD             = f+j_coord_offsetD;
 950
 951             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 952
 953             /* Inner loop uses 105 flops */
 954         }
 955
 956         if(jidx<j_index_end)
 957         {
 958
 959             /* Get j neighbor index, and coordinate index */
 960             jnrlistA         = jjnr[jidx];
 961             jnrlistB         = jjnr[jidx+1];
 962             jnrlistC         = jjnr[jidx+2];
 963             jnrlistD         = jjnr[jidx+3];
 964             /* Sign of each element will be negative for non-real atoms.
 965              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 966              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 967              */
 968             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 969             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 970             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 971             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 972             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 973             j_coord_offsetA  = DIM*jnrA;
 974             j_coord_offsetB  = DIM*jnrB;
 975             j_coord_offsetC  = DIM*jnrC;
 976             j_coord_offsetD  = DIM*jnrD;
 977
 978             /* load j atom coordinates */
 979             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 980                                               x+j_coord_offsetC,x+j_coord_offsetD,
 981                                               &jx0,&jy0,&jz0);
 982
 983             /* Calculate displacement vector */
 984             dx00             = _mm_sub_ps(ix0,jx0);
 985             dy00             = _mm_sub_ps(iy0,jy0);
 986             dz00             = _mm_sub_ps(iz0,jz0);
 987             dx10             = _mm_sub_ps(ix1,jx0);
 988             dy10             = _mm_sub_ps(iy1,jy0);
 989             dz10             = _mm_sub_ps(iz1,jz0);
 990             dx20             = _mm_sub_ps(ix2,jx0);
 991             dy20             = _mm_sub_ps(iy2,jy0);
 992             dz20             = _mm_sub_ps(iz2,jz0);
 993
 994             /* Calculate squared distance and things based on it */
 995             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 996             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 997             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 998
 999             rinv00           = gmx_mm_invsqrt_ps(rsq00);
1000             rinv10           = gmx_mm_invsqrt_ps(rsq10);
1001             rinv20           = gmx_mm_invsqrt_ps(rsq20);
1002
1003             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
1004             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
1005             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
1006
1007             /* Load parameters for j particles */
1008             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
1009                                                               charge+jnrC+0,charge+jnrD+0);
1010             vdwjidx0A        = 2*vdwtype[jnrA+0];
1011             vdwjidx0B        = 2*vdwtype[jnrB+0];
1012             vdwjidx0C        = 2*vdwtype[jnrC+0];
1013             vdwjidx0D        = 2*vdwtype[jnrD+0];
1014
1015             fjx0             = _mm_setzero_ps();
1016             fjy0             = _mm_setzero_ps();
1017             fjz0             = _mm_setzero_ps();
1018
1019             /**************************
1020              * CALCULATE INTERACTIONS *
1021              **************************/
1022
1023             r00              = _mm_mul_ps(rsq00,rinv00);
1024             r00              = _mm_andnot_ps(dummy_mask,r00);
1025
1026             /* Compute parameters for interactions between i and j atoms */
1027             qq00             = _mm_mul_ps(iq0,jq0);
1028             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
1029                                          vdwparam+vdwioffset0+vdwjidx0B,
1030                                          vdwparam+vdwioffset0+vdwjidx0C,
1031                                          vdwparam+vdwioffset0+vdwjidx0D,
1032                                          &c6_00,&c12_00);
1033
1034             c6grid_00       = gmx_mm_load_4real_swizzle_ps(vdwgridparam+vdwioffset0+vdwjidx0A,
1035                                                                vdwgridparam+vdwioffset0+vdwjidx0B,
1036                                                                vdwgridparam+vdwioffset0+vdwjidx0C,
1037                                                                vdwgridparam+vdwioffset0+vdwjidx0D);
1038
1039             /* EWALD ELECTROSTATICS */
1040
1041             /* Analytical PME correction */
1042             zeta2            = _mm_mul_ps(beta2,rsq00);
1043             rinv3            = _mm_mul_ps(rinvsq00,rinv00);
1044             pmecorrF         = gmx_mm_pmecorrF_ps(zeta2);
1045             felec            = _mm_macc_ps(pmecorrF,beta3,rinv3);
1046             felec            = _mm_mul_ps(qq00,felec);
1047
1048             /* Analytical LJ-PME */
1049             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
1050             ewcljrsq         = _mm_mul_ps(ewclj2,rsq00);
1051             ewclj6           = _mm_mul_ps(ewclj2,_mm_mul_ps(ewclj2,ewclj2));
1052             exponent         = gmx_simd_exp_r(ewcljrsq);
1053             /* poly = exp(-(beta*r)^2) * (1 + (beta*r)^2 + (beta*r)^4 /2) */
1054             poly             = _mm_mul_ps(exponent,_mm_macc_ps(_mm_mul_ps(ewcljrsq,ewcljrsq),one_half,_mm_sub_ps(one,ewcljrsq)));
1055             /* f6A = 6 * C6grid * (1 - poly) */
1056             f6A              = _mm_mul_ps(c6grid_00,_mm_sub_ps(one,poly));
1057             /* f6B = C6grid * exponent * beta^6 */
1058             f6B              = _mm_mul_ps(_mm_mul_ps(c6grid_00,one_sixth),_mm_mul_ps(exponent,ewclj6));
1059             /* fvdw = 12*C12/r13 - ((6*C6 - f6A)/r6 + f6B)/r */
1060             fvdw              = _mm_mul_ps(_mm_macc_ps(_mm_msub_ps(c12_00,rinvsix,_mm_sub_ps(c6_00,f6A)),rinvsix,f6B),rinvsq00);
1061
1062             fscal            = _mm_add_ps(felec,fvdw);
1063
1064             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1065
1066              /* Update vectorial force */
1067             fix0             = _mm_macc_ps(dx00,fscal,fix0);
1068             fiy0             = _mm_macc_ps(dy00,fscal,fiy0);
1069             fiz0             = _mm_macc_ps(dz00,fscal,fiz0);
1070
1071             fjx0             = _mm_macc_ps(dx00,fscal,fjx0);
1072             fjy0             = _mm_macc_ps(dy00,fscal,fjy0);
1073             fjz0             = _mm_macc_ps(dz00,fscal,fjz0);
1074
1075             /**************************
1076              * CALCULATE INTERACTIONS *
1077              **************************/
1078
1079             r10              = _mm_mul_ps(rsq10,rinv10);
1080             r10              = _mm_andnot_ps(dummy_mask,r10);
1081
1082             /* Compute parameters for interactions between i and j atoms */
1083             qq10             = _mm_mul_ps(iq1,jq0);
1084
1085             /* EWALD ELECTROSTATICS */
1086
1087             /* Analytical PME correction */
1088             zeta2            = _mm_mul_ps(beta2,rsq10);
1089             rinv3            = _mm_mul_ps(rinvsq10,rinv10);
1090             pmecorrF         = gmx_mm_pmecorrF_ps(zeta2);
1091             felec            = _mm_macc_ps(pmecorrF,beta3,rinv3);
1092             felec            = _mm_mul_ps(qq10,felec);
1093
1094             fscal            = felec;
1095
1096             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1097
1098              /* Update vectorial force */
1099             fix1             = _mm_macc_ps(dx10,fscal,fix1);
1100             fiy1             = _mm_macc_ps(dy10,fscal,fiy1);
1101             fiz1             = _mm_macc_ps(dz10,fscal,fiz1);
1102
1103             fjx0             = _mm_macc_ps(dx10,fscal,fjx0);
1104             fjy0             = _mm_macc_ps(dy10,fscal,fjy0);
1105             fjz0             = _mm_macc_ps(dz10,fscal,fjz0);
1106
1107             /**************************
1108              * CALCULATE INTERACTIONS *
1109              **************************/
1110
1111             r20              = _mm_mul_ps(rsq20,rinv20);
1112             r20              = _mm_andnot_ps(dummy_mask,r20);
1113
1114             /* Compute parameters for interactions between i and j atoms */
1115             qq20             = _mm_mul_ps(iq2,jq0);
1116
1117             /* EWALD ELECTROSTATICS */
1118
1119             /* Analytical PME correction */
1120             zeta2            = _mm_mul_ps(beta2,rsq20);
1121             rinv3            = _mm_mul_ps(rinvsq20,rinv20);
1122             pmecorrF         = gmx_mm_pmecorrF_ps(zeta2);
1123             felec            = _mm_macc_ps(pmecorrF,beta3,rinv3);
1124             felec            = _mm_mul_ps(qq20,felec);
1125
1126             fscal            = felec;
1127
1128             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1129
1130              /* Update vectorial force */
1131             fix2             = _mm_macc_ps(dx20,fscal,fix2);
1132             fiy2             = _mm_macc_ps(dy20,fscal,fiy2);
1133             fiz2             = _mm_macc_ps(dz20,fscal,fiz2);
1134
1135             fjx0             = _mm_macc_ps(dx20,fscal,fjx0);
1136             fjy0             = _mm_macc_ps(dy20,fscal,fjy0);
1137             fjz0             = _mm_macc_ps(dz20,fscal,fjz0);
1138
1139             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
1140             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
1141             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
1142             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
1143
1144             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
1145
1146             /* Inner loop uses 108 flops */
1147         }
1148
1149         /* End of innermost loop */
1150
1151         gmx_mm_update_iforce_3atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
1152                                               f+i_coord_offset,fshift+i_shift_offset);
1153
1154         /* Increment number of inner iterations */
1155         inneriter                  += j_index_end - j_index_start;
1156
1157         /* Outer loop uses 18 flops */
1158     }
1159
1160     /* Increment number of outer iterations */
1161     outeriter        += nri;
1162
1163     /* Update outer/inner flops */
1164
1165     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3_F,outeriter*18 + inneriter*108);
1166 }