src/gromacs/gmxlib/nonbonded/nb_kernel_avx_128_fma_single/nb_kernel_ElecEw_VdwLJ_GeomW3P1_avx_128_fma_single.c

   1 /*
   2  * This file is part of the GROMACS molecular simulation package.
   3  *
   4  * Copyright (c) 2012,2013, by the GROMACS development team, led by
   5  * Mark Abraham, David van der Spoel, Berk Hess, and Erik Lindahl,
   6  * and including many others, as listed in the AUTHORS file in the
   7  * top-level source directory and at http://www.gromacs.org.
   8  *
   9  * GROMACS is free software; you can redistribute it and/or
  10  * modify it under the terms of the GNU Lesser General Public License
  11  * as published by the Free Software Foundation; either version 2.1
  12  * of the License, or (at your option) any later version.
  13  *
  14  * GROMACS is distributed in the hope that it will be useful,
  15  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  16  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  17  * Lesser General Public License for more details.
  18  *
  19  * You should have received a copy of the GNU Lesser General Public
  20  * License along with GROMACS; if not, see
  21  * http://www.gnu.org/licenses, or write to the Free Software Foundation,
  22  * Inc., 51 Franklin Street, Fifth Floor, Boston, MA  02110-1301  USA.
  23  *
  24  * If you want to redistribute modifications to GROMACS, please
  25  * consider that scientific software is very special. Version
  26  * control is crucial - bugs must be traceable. We will be happy to
  27  * consider code for inclusion in the official distribution, but
  28  * derived work must not be called official GROMACS. Details are found
  29  * in the README & COPYING files - if they are missing, get the
  30  * official version at http://www.gromacs.org.
  31  *
  32  * To help us fund GROMACS development, we humbly ask that you cite
  33  * the research papers on the package. Check out http://www.gromacs.org.
  34  */
  35 /*
  36  * Note: this file was generated by the GROMACS avx_128_fma_single kernel generator.
  37  */
  38 #ifdef HAVE_CONFIG_H
  39 #include <config.h>
  40 #endif
  41
  42 #include <math.h>
  43
  44 #include "../nb_kernel.h"
  45 #include "types/simple.h"
  46 #include "vec.h"
  47 #include "nrnb.h"
  48
  49 #include "gromacs/simd/math_x86_avx_128_fma_single.h"
  50 #include "kernelutil_x86_avx_128_fma_single.h"
  51
  52 /*
  53  * Gromacs nonbonded kernel:   nb_kernel_ElecEw_VdwLJ_GeomW3P1_VF_avx_128_fma_single
  54  * Electrostatics interaction: Ewald
  55  * VdW interaction:            LennardJones
  56  * Geometry:                   Water3-Particle
  57  * Calculate force/pot:        PotentialAndForce
  58  */
  59 void
  60 nb_kernel_ElecEw_VdwLJ_GeomW3P1_VF_avx_128_fma_single
  61                     (t_nblist                    * gmx_restrict       nlist,
  62                      rvec                        * gmx_restrict          xx,
  63                      rvec                        * gmx_restrict          ff,
  64                      t_forcerec                  * gmx_restrict          fr,
  65                      t_mdatoms                   * gmx_restrict     mdatoms,
  66                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
  67                      t_nrnb                      * gmx_restrict        nrnb)
  68 {
  69     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
  70      * just 0 for non-waters.
  71      * Suffixes A,B,C,D refer to j loop unrolling done with AVX_128, e.g. for the four different
  72      * jnr indices corresponding to data put in the four positions in the SIMD register.
  73      */
  74     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
  75     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
  76     int              jnrA,jnrB,jnrC,jnrD;
  77     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
  78     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
  79     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
  80     real             rcutoff_scalar;
  81     real             *shiftvec,*fshift,*x,*f;
  82     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
  83     real             scratch[4*DIM];
  84     __m128           fscal,rcutoff,rcutoff2,jidxall;
  85     int              vdwioffset0;
  86     __m128           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
  87     int              vdwioffset1;
  88     __m128           ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
  89     int              vdwioffset2;
  90     __m128           ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
  91     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
  92     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  93     __m128           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  94     __m128           dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
  95     __m128           dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
  96     __m128           velec,felec,velecsum,facel,crf,krf,krf2;
  97     real             *charge;
  98     int              nvdwtype;
  99     __m128           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 100     int              *vdwtype;
 101     real             *vdwparam;
 102     __m128           one_sixth   = _mm_set1_ps(1.0/6.0);
 103     __m128           one_twelfth = _mm_set1_ps(1.0/12.0);
 104     __m128i          ewitab;
 105     __m128           ewtabscale,eweps,twoeweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
 106     __m128           beta,beta2,beta3,zeta2,pmecorrF,pmecorrV,rinv3;
 107     real             *ewtab;
 108     __m128           dummy_mask,cutoff_mask;
 109     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
 110     __m128           one     = _mm_set1_ps(1.0);
 111     __m128           two     = _mm_set1_ps(2.0);
 112     x                = xx[0];
 113     f                = ff[0];
 114
 115     nri              = nlist->nri;
 116     iinr             = nlist->iinr;
 117     jindex           = nlist->jindex;
 118     jjnr             = nlist->jjnr;
 119     shiftidx         = nlist->shift;
 120     gid              = nlist->gid;
 121     shiftvec         = fr->shift_vec[0];
 122     fshift           = fr->fshift[0];
 123     facel            = _mm_set1_ps(fr->epsfac);
 124     charge           = mdatoms->chargeA;
 125     nvdwtype         = fr->ntype;
 126     vdwparam         = fr->nbfp;
 127     vdwtype          = mdatoms->typeA;
 128
 129     sh_ewald         = _mm_set1_ps(fr->ic->sh_ewald);
 130     beta             = _mm_set1_ps(fr->ic->ewaldcoeff);
 131     beta2            = _mm_mul_ps(beta,beta);
 132     beta3            = _mm_mul_ps(beta,beta2);
 133     ewtab            = fr->ic->tabq_coul_FDV0;
 134     ewtabscale       = _mm_set1_ps(fr->ic->tabq_scale);
 135     ewtabhalfspace   = _mm_set1_ps(0.5/fr->ic->tabq_scale);
 136
 137     /* Setup water-specific parameters */
 138     inr              = nlist->iinr[0];
 139     iq0              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+0]));
 140     iq1              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+1]));
 141     iq2              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+2]));
 142     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 143
 144     /* Avoid stupid compiler warnings */
 145     jnrA = jnrB = jnrC = jnrD = 0;
 146     j_coord_offsetA = 0;
 147     j_coord_offsetB = 0;
 148     j_coord_offsetC = 0;
 149     j_coord_offsetD = 0;
 150
 151     outeriter        = 0;
 152     inneriter        = 0;
 153
 154     for(iidx=0;iidx<4*DIM;iidx++)
 155     {
 156         scratch[iidx] = 0.0;
 157     }
 158
 159     /* Start outer loop over neighborlists */
 160     for(iidx=0; iidx<nri; iidx++)
 161     {
 162         /* Load shift vector for this list */
 163         i_shift_offset   = DIM*shiftidx[iidx];
 164
 165         /* Load limits for loop over neighbors */
 166         j_index_start    = jindex[iidx];
 167         j_index_end      = jindex[iidx+1];
 168
 169         /* Get outer coordinate index */
 170         inr              = iinr[iidx];
 171         i_coord_offset   = DIM*inr;
 172
 173         /* Load i particle coords and add shift vector */
 174         gmx_mm_load_shift_and_3rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
 175                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 176
 177         fix0             = _mm_setzero_ps();
 178         fiy0             = _mm_setzero_ps();
 179         fiz0             = _mm_setzero_ps();
 180         fix1             = _mm_setzero_ps();
 181         fiy1             = _mm_setzero_ps();
 182         fiz1             = _mm_setzero_ps();
 183         fix2             = _mm_setzero_ps();
 184         fiy2             = _mm_setzero_ps();
 185         fiz2             = _mm_setzero_ps();
 186
 187         /* Reset potential sums */
 188         velecsum         = _mm_setzero_ps();
 189         vvdwsum          = _mm_setzero_ps();
 190
 191         /* Start inner kernel loop */
 192         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 193         {
 194
 195             /* Get j neighbor index, and coordinate index */
 196             jnrA             = jjnr[jidx];
 197             jnrB             = jjnr[jidx+1];
 198             jnrC             = jjnr[jidx+2];
 199             jnrD             = jjnr[jidx+3];
 200             j_coord_offsetA  = DIM*jnrA;
 201             j_coord_offsetB  = DIM*jnrB;
 202             j_coord_offsetC  = DIM*jnrC;
 203             j_coord_offsetD  = DIM*jnrD;
 204
 205             /* load j atom coordinates */
 206             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 207                                               x+j_coord_offsetC,x+j_coord_offsetD,
 208                                               &jx0,&jy0,&jz0);
 209
 210             /* Calculate displacement vector */
 211             dx00             = _mm_sub_ps(ix0,jx0);
 212             dy00             = _mm_sub_ps(iy0,jy0);
 213             dz00             = _mm_sub_ps(iz0,jz0);
 214             dx10             = _mm_sub_ps(ix1,jx0);
 215             dy10             = _mm_sub_ps(iy1,jy0);
 216             dz10             = _mm_sub_ps(iz1,jz0);
 217             dx20             = _mm_sub_ps(ix2,jx0);
 218             dy20             = _mm_sub_ps(iy2,jy0);
 219             dz20             = _mm_sub_ps(iz2,jz0);
 220
 221             /* Calculate squared distance and things based on it */
 222             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 223             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 224             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 225
 226             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 227             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 228             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 229
 230             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 231             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 232             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 233
 234             /* Load parameters for j particles */
 235             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 236                                                               charge+jnrC+0,charge+jnrD+0);
 237             vdwjidx0A        = 2*vdwtype[jnrA+0];
 238             vdwjidx0B        = 2*vdwtype[jnrB+0];
 239             vdwjidx0C        = 2*vdwtype[jnrC+0];
 240             vdwjidx0D        = 2*vdwtype[jnrD+0];
 241
 242             fjx0             = _mm_setzero_ps();
 243             fjy0             = _mm_setzero_ps();
 244             fjz0             = _mm_setzero_ps();
 245
 246             /**************************
 247              * CALCULATE INTERACTIONS *
 248              **************************/
 249
 250             r00              = _mm_mul_ps(rsq00,rinv00);
 251
 252             /* Compute parameters for interactions between i and j atoms */
 253             qq00             = _mm_mul_ps(iq0,jq0);
 254             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 255                                          vdwparam+vdwioffset0+vdwjidx0B,
 256                                          vdwparam+vdwioffset0+vdwjidx0C,
 257                                          vdwparam+vdwioffset0+vdwjidx0D,
 258                                          &c6_00,&c12_00);
 259
 260             /* EWALD ELECTROSTATICS */
 261
 262             /* Analytical PME correction */
 263             zeta2            = _mm_mul_ps(beta2,rsq00);
 264             rinv3            = _mm_mul_ps(rinvsq00,rinv00);
 265             pmecorrF         = gmx_mm_pmecorrF_ps(zeta2);
 266             felec            = _mm_macc_ps(pmecorrF,beta3,rinv3);
 267             felec            = _mm_mul_ps(qq00,felec);
 268             pmecorrV         = gmx_mm_pmecorrV_ps(zeta2);
 269             velec            = _mm_nmacc_ps(pmecorrV,beta,rinv00);
 270             velec            = _mm_mul_ps(qq00,velec);
 271
 272             /* LENNARD-JONES DISPERSION/REPULSION */
 273
 274             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 275             vvdw6            = _mm_mul_ps(c6_00,rinvsix);
 276             vvdw12           = _mm_mul_ps(c12_00,_mm_mul_ps(rinvsix,rinvsix));
 277             vvdw             = _mm_msub_ps(vvdw12,one_twelfth,_mm_mul_ps(vvdw6,one_sixth));
 278             fvdw             = _mm_mul_ps(_mm_sub_ps(vvdw12,vvdw6),rinvsq00);
 279
 280             /* Update potential sum for this i atom from the interaction with this j atom. */
 281             velecsum         = _mm_add_ps(velecsum,velec);
 282             vvdwsum          = _mm_add_ps(vvdwsum,vvdw);
 283
 284             fscal            = _mm_add_ps(felec,fvdw);
 285
 286              /* Update vectorial force */
 287             fix0             = _mm_macc_ps(dx00,fscal,fix0);
 288             fiy0             = _mm_macc_ps(dy00,fscal,fiy0);
 289             fiz0             = _mm_macc_ps(dz00,fscal,fiz0);
 290
 291             fjx0             = _mm_macc_ps(dx00,fscal,fjx0);
 292             fjy0             = _mm_macc_ps(dy00,fscal,fjy0);
 293             fjz0             = _mm_macc_ps(dz00,fscal,fjz0);
 294
 295             /**************************
 296              * CALCULATE INTERACTIONS *
 297              **************************/
 298
 299             r10              = _mm_mul_ps(rsq10,rinv10);
 300
 301             /* Compute parameters for interactions between i and j atoms */
 302             qq10             = _mm_mul_ps(iq1,jq0);
 303
 304             /* EWALD ELECTROSTATICS */
 305
 306             /* Analytical PME correction */
 307             zeta2            = _mm_mul_ps(beta2,rsq10);
 308             rinv3            = _mm_mul_ps(rinvsq10,rinv10);
 309             pmecorrF         = gmx_mm_pmecorrF_ps(zeta2);
 310             felec            = _mm_macc_ps(pmecorrF,beta3,rinv3);
 311             felec            = _mm_mul_ps(qq10,felec);
 312             pmecorrV         = gmx_mm_pmecorrV_ps(zeta2);
 313             velec            = _mm_nmacc_ps(pmecorrV,beta,rinv10);
 314             velec            = _mm_mul_ps(qq10,velec);
 315
 316             /* Update potential sum for this i atom from the interaction with this j atom. */
 317             velecsum         = _mm_add_ps(velecsum,velec);
 318
 319             fscal            = felec;
 320
 321              /* Update vectorial force */
 322             fix1             = _mm_macc_ps(dx10,fscal,fix1);
 323             fiy1             = _mm_macc_ps(dy10,fscal,fiy1);
 324             fiz1             = _mm_macc_ps(dz10,fscal,fiz1);
 325
 326             fjx0             = _mm_macc_ps(dx10,fscal,fjx0);
 327             fjy0             = _mm_macc_ps(dy10,fscal,fjy0);
 328             fjz0             = _mm_macc_ps(dz10,fscal,fjz0);
 329
 330             /**************************
 331              * CALCULATE INTERACTIONS *
 332              **************************/
 333
 334             r20              = _mm_mul_ps(rsq20,rinv20);
 335
 336             /* Compute parameters for interactions between i and j atoms */
 337             qq20             = _mm_mul_ps(iq2,jq0);
 338
 339             /* EWALD ELECTROSTATICS */
 340
 341             /* Analytical PME correction */
 342             zeta2            = _mm_mul_ps(beta2,rsq20);
 343             rinv3            = _mm_mul_ps(rinvsq20,rinv20);
 344             pmecorrF         = gmx_mm_pmecorrF_ps(zeta2);
 345             felec            = _mm_macc_ps(pmecorrF,beta3,rinv3);
 346             felec            = _mm_mul_ps(qq20,felec);
 347             pmecorrV         = gmx_mm_pmecorrV_ps(zeta2);
 348             velec            = _mm_nmacc_ps(pmecorrV,beta,rinv20);
 349             velec            = _mm_mul_ps(qq20,velec);
 350
 351             /* Update potential sum for this i atom from the interaction with this j atom. */
 352             velecsum         = _mm_add_ps(velecsum,velec);
 353
 354             fscal            = felec;
 355
 356              /* Update vectorial force */
 357             fix2             = _mm_macc_ps(dx20,fscal,fix2);
 358             fiy2             = _mm_macc_ps(dy20,fscal,fiy2);
 359             fiz2             = _mm_macc_ps(dz20,fscal,fiz2);
 360
 361             fjx0             = _mm_macc_ps(dx20,fscal,fjx0);
 362             fjy0             = _mm_macc_ps(dy20,fscal,fjy0);
 363             fjz0             = _mm_macc_ps(dz20,fscal,fjz0);
 364
 365             fjptrA             = f+j_coord_offsetA;
 366             fjptrB             = f+j_coord_offsetB;
 367             fjptrC             = f+j_coord_offsetC;
 368             fjptrD             = f+j_coord_offsetD;
 369
 370             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 371
 372             /* Inner loop uses 99 flops */
 373         }
 374
 375         if(jidx<j_index_end)
 376         {
 377
 378             /* Get j neighbor index, and coordinate index */
 379             jnrlistA         = jjnr[jidx];
 380             jnrlistB         = jjnr[jidx+1];
 381             jnrlistC         = jjnr[jidx+2];
 382             jnrlistD         = jjnr[jidx+3];
 383             /* Sign of each element will be negative for non-real atoms.
 384              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 385              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 386              */
 387             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 388             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 389             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 390             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 391             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 392             j_coord_offsetA  = DIM*jnrA;
 393             j_coord_offsetB  = DIM*jnrB;
 394             j_coord_offsetC  = DIM*jnrC;
 395             j_coord_offsetD  = DIM*jnrD;
 396
 397             /* load j atom coordinates */
 398             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 399                                               x+j_coord_offsetC,x+j_coord_offsetD,
 400                                               &jx0,&jy0,&jz0);
 401
 402             /* Calculate displacement vector */
 403             dx00             = _mm_sub_ps(ix0,jx0);
 404             dy00             = _mm_sub_ps(iy0,jy0);
 405             dz00             = _mm_sub_ps(iz0,jz0);
 406             dx10             = _mm_sub_ps(ix1,jx0);
 407             dy10             = _mm_sub_ps(iy1,jy0);
 408             dz10             = _mm_sub_ps(iz1,jz0);
 409             dx20             = _mm_sub_ps(ix2,jx0);
 410             dy20             = _mm_sub_ps(iy2,jy0);
 411             dz20             = _mm_sub_ps(iz2,jz0);
 412
 413             /* Calculate squared distance and things based on it */
 414             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 415             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 416             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 417
 418             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 419             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 420             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 421
 422             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 423             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 424             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 425
 426             /* Load parameters for j particles */
 427             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 428                                                               charge+jnrC+0,charge+jnrD+0);
 429             vdwjidx0A        = 2*vdwtype[jnrA+0];
 430             vdwjidx0B        = 2*vdwtype[jnrB+0];
 431             vdwjidx0C        = 2*vdwtype[jnrC+0];
 432             vdwjidx0D        = 2*vdwtype[jnrD+0];
 433
 434             fjx0             = _mm_setzero_ps();
 435             fjy0             = _mm_setzero_ps();
 436             fjz0             = _mm_setzero_ps();
 437
 438             /**************************
 439              * CALCULATE INTERACTIONS *
 440              **************************/
 441
 442             r00              = _mm_mul_ps(rsq00,rinv00);
 443             r00              = _mm_andnot_ps(dummy_mask,r00);
 444
 445             /* Compute parameters for interactions between i and j atoms */
 446             qq00             = _mm_mul_ps(iq0,jq0);
 447             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 448                                          vdwparam+vdwioffset0+vdwjidx0B,
 449                                          vdwparam+vdwioffset0+vdwjidx0C,
 450                                          vdwparam+vdwioffset0+vdwjidx0D,
 451                                          &c6_00,&c12_00);
 452
 453             /* EWALD ELECTROSTATICS */
 454
 455             /* Analytical PME correction */
 456             zeta2            = _mm_mul_ps(beta2,rsq00);
 457             rinv3            = _mm_mul_ps(rinvsq00,rinv00);
 458             pmecorrF         = gmx_mm_pmecorrF_ps(zeta2);
 459             felec            = _mm_macc_ps(pmecorrF,beta3,rinv3);
 460             felec            = _mm_mul_ps(qq00,felec);
 461             pmecorrV         = gmx_mm_pmecorrV_ps(zeta2);
 462             velec            = _mm_nmacc_ps(pmecorrV,beta,rinv00);
 463             velec            = _mm_mul_ps(qq00,velec);
 464
 465             /* LENNARD-JONES DISPERSION/REPULSION */
 466
 467             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 468             vvdw6            = _mm_mul_ps(c6_00,rinvsix);
 469             vvdw12           = _mm_mul_ps(c12_00,_mm_mul_ps(rinvsix,rinvsix));
 470             vvdw             = _mm_msub_ps(vvdw12,one_twelfth,_mm_mul_ps(vvdw6,one_sixth));
 471             fvdw             = _mm_mul_ps(_mm_sub_ps(vvdw12,vvdw6),rinvsq00);
 472
 473             /* Update potential sum for this i atom from the interaction with this j atom. */
 474             velec            = _mm_andnot_ps(dummy_mask,velec);
 475             velecsum         = _mm_add_ps(velecsum,velec);
 476             vvdw             = _mm_andnot_ps(dummy_mask,vvdw);
 477             vvdwsum          = _mm_add_ps(vvdwsum,vvdw);
 478
 479             fscal            = _mm_add_ps(felec,fvdw);
 480
 481             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 482
 483              /* Update vectorial force */
 484             fix0             = _mm_macc_ps(dx00,fscal,fix0);
 485             fiy0             = _mm_macc_ps(dy00,fscal,fiy0);
 486             fiz0             = _mm_macc_ps(dz00,fscal,fiz0);
 487
 488             fjx0             = _mm_macc_ps(dx00,fscal,fjx0);
 489             fjy0             = _mm_macc_ps(dy00,fscal,fjy0);
 490             fjz0             = _mm_macc_ps(dz00,fscal,fjz0);
 491
 492             /**************************
 493              * CALCULATE INTERACTIONS *
 494              **************************/
 495
 496             r10              = _mm_mul_ps(rsq10,rinv10);
 497             r10              = _mm_andnot_ps(dummy_mask,r10);
 498
 499             /* Compute parameters for interactions between i and j atoms */
 500             qq10             = _mm_mul_ps(iq1,jq0);
 501
 502             /* EWALD ELECTROSTATICS */
 503
 504             /* Analytical PME correction */
 505             zeta2            = _mm_mul_ps(beta2,rsq10);
 506             rinv3            = _mm_mul_ps(rinvsq10,rinv10);
 507             pmecorrF         = gmx_mm_pmecorrF_ps(zeta2);
 508             felec            = _mm_macc_ps(pmecorrF,beta3,rinv3);
 509             felec            = _mm_mul_ps(qq10,felec);
 510             pmecorrV         = gmx_mm_pmecorrV_ps(zeta2);
 511             velec            = _mm_nmacc_ps(pmecorrV,beta,rinv10);
 512             velec            = _mm_mul_ps(qq10,velec);
 513
 514             /* Update potential sum for this i atom from the interaction with this j atom. */
 515             velec            = _mm_andnot_ps(dummy_mask,velec);
 516             velecsum         = _mm_add_ps(velecsum,velec);
 517
 518             fscal            = felec;
 519
 520             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 521
 522              /* Update vectorial force */
 523             fix1             = _mm_macc_ps(dx10,fscal,fix1);
 524             fiy1             = _mm_macc_ps(dy10,fscal,fiy1);
 525             fiz1             = _mm_macc_ps(dz10,fscal,fiz1);
 526
 527             fjx0             = _mm_macc_ps(dx10,fscal,fjx0);
 528             fjy0             = _mm_macc_ps(dy10,fscal,fjy0);
 529             fjz0             = _mm_macc_ps(dz10,fscal,fjz0);
 530
 531             /**************************
 532              * CALCULATE INTERACTIONS *
 533              **************************/
 534
 535             r20              = _mm_mul_ps(rsq20,rinv20);
 536             r20              = _mm_andnot_ps(dummy_mask,r20);
 537
 538             /* Compute parameters for interactions between i and j atoms */
 539             qq20             = _mm_mul_ps(iq2,jq0);
 540
 541             /* EWALD ELECTROSTATICS */
 542
 543             /* Analytical PME correction */
 544             zeta2            = _mm_mul_ps(beta2,rsq20);
 545             rinv3            = _mm_mul_ps(rinvsq20,rinv20);
 546             pmecorrF         = gmx_mm_pmecorrF_ps(zeta2);
 547             felec            = _mm_macc_ps(pmecorrF,beta3,rinv3);
 548             felec            = _mm_mul_ps(qq20,felec);
 549             pmecorrV         = gmx_mm_pmecorrV_ps(zeta2);
 550             velec            = _mm_nmacc_ps(pmecorrV,beta,rinv20);
 551             velec            = _mm_mul_ps(qq20,velec);
 552
 553             /* Update potential sum for this i atom from the interaction with this j atom. */
 554             velec            = _mm_andnot_ps(dummy_mask,velec);
 555             velecsum         = _mm_add_ps(velecsum,velec);
 556
 557             fscal            = felec;
 558
 559             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 560
 561              /* Update vectorial force */
 562             fix2             = _mm_macc_ps(dx20,fscal,fix2);
 563             fiy2             = _mm_macc_ps(dy20,fscal,fiy2);
 564             fiz2             = _mm_macc_ps(dz20,fscal,fiz2);
 565
 566             fjx0             = _mm_macc_ps(dx20,fscal,fjx0);
 567             fjy0             = _mm_macc_ps(dy20,fscal,fjy0);
 568             fjz0             = _mm_macc_ps(dz20,fscal,fjz0);
 569
 570             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 571             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 572             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 573             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 574
 575             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 576
 577             /* Inner loop uses 102 flops */
 578         }
 579
 580         /* End of innermost loop */
 581
 582         gmx_mm_update_iforce_3atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
 583                                               f+i_coord_offset,fshift+i_shift_offset);
 584
 585         ggid                        = gid[iidx];
 586         /* Update potential energies */
 587         gmx_mm_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid);
 588         gmx_mm_update_1pot_ps(vvdwsum,kernel_data->energygrp_vdw+ggid);
 589
 590         /* Increment number of inner iterations */
 591         inneriter                  += j_index_end - j_index_start;
 592
 593         /* Outer loop uses 20 flops */
 594     }
 595
 596     /* Increment number of outer iterations */
 597     outeriter        += nri;
 598
 599     /* Update outer/inner flops */
 600
 601     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3_VF,outeriter*20 + inneriter*102);
 602 }
 603 /*
 604  * Gromacs nonbonded kernel:   nb_kernel_ElecEw_VdwLJ_GeomW3P1_F_avx_128_fma_single
 605  * Electrostatics interaction: Ewald
 606  * VdW interaction:            LennardJones
 607  * Geometry:                   Water3-Particle
 608  * Calculate force/pot:        Force
 609  */
 610 void
 611 nb_kernel_ElecEw_VdwLJ_GeomW3P1_F_avx_128_fma_single
 612                     (t_nblist                    * gmx_restrict       nlist,
 613                      rvec                        * gmx_restrict          xx,
 614                      rvec                        * gmx_restrict          ff,
 615                      t_forcerec                  * gmx_restrict          fr,
 616                      t_mdatoms                   * gmx_restrict     mdatoms,
 617                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 618                      t_nrnb                      * gmx_restrict        nrnb)
 619 {
 620     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 621      * just 0 for non-waters.
 622      * Suffixes A,B,C,D refer to j loop unrolling done with AVX_128, e.g. for the four different
 623      * jnr indices corresponding to data put in the four positions in the SIMD register.
 624      */
 625     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 626     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 627     int              jnrA,jnrB,jnrC,jnrD;
 628     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
 629     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 630     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 631     real             rcutoff_scalar;
 632     real             *shiftvec,*fshift,*x,*f;
 633     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
 634     real             scratch[4*DIM];
 635     __m128           fscal,rcutoff,rcutoff2,jidxall;
 636     int              vdwioffset0;
 637     __m128           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 638     int              vdwioffset1;
 639     __m128           ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
 640     int              vdwioffset2;
 641     __m128           ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
 642     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
 643     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 644     __m128           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 645     __m128           dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
 646     __m128           dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
 647     __m128           velec,felec,velecsum,facel,crf,krf,krf2;
 648     real             *charge;
 649     int              nvdwtype;
 650     __m128           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 651     int              *vdwtype;
 652     real             *vdwparam;
 653     __m128           one_sixth   = _mm_set1_ps(1.0/6.0);
 654     __m128           one_twelfth = _mm_set1_ps(1.0/12.0);
 655     __m128i          ewitab;
 656     __m128           ewtabscale,eweps,twoeweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
 657     __m128           beta,beta2,beta3,zeta2,pmecorrF,pmecorrV,rinv3;
 658     real             *ewtab;
 659     __m128           dummy_mask,cutoff_mask;
 660     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
 661     __m128           one     = _mm_set1_ps(1.0);
 662     __m128           two     = _mm_set1_ps(2.0);
 663     x                = xx[0];
 664     f                = ff[0];
 665
 666     nri              = nlist->nri;
 667     iinr             = nlist->iinr;
 668     jindex           = nlist->jindex;
 669     jjnr             = nlist->jjnr;
 670     shiftidx         = nlist->shift;
 671     gid              = nlist->gid;
 672     shiftvec         = fr->shift_vec[0];
 673     fshift           = fr->fshift[0];
 674     facel            = _mm_set1_ps(fr->epsfac);
 675     charge           = mdatoms->chargeA;
 676     nvdwtype         = fr->ntype;
 677     vdwparam         = fr->nbfp;
 678     vdwtype          = mdatoms->typeA;
 679
 680     sh_ewald         = _mm_set1_ps(fr->ic->sh_ewald);
 681     beta             = _mm_set1_ps(fr->ic->ewaldcoeff);
 682     beta2            = _mm_mul_ps(beta,beta);
 683     beta3            = _mm_mul_ps(beta,beta2);
 684     ewtab            = fr->ic->tabq_coul_F;
 685     ewtabscale       = _mm_set1_ps(fr->ic->tabq_scale);
 686     ewtabhalfspace   = _mm_set1_ps(0.5/fr->ic->tabq_scale);
 687
 688     /* Setup water-specific parameters */
 689     inr              = nlist->iinr[0];
 690     iq0              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+0]));
 691     iq1              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+1]));
 692     iq2              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+2]));
 693     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 694
 695     /* Avoid stupid compiler warnings */
 696     jnrA = jnrB = jnrC = jnrD = 0;
 697     j_coord_offsetA = 0;
 698     j_coord_offsetB = 0;
 699     j_coord_offsetC = 0;
 700     j_coord_offsetD = 0;
 701
 702     outeriter        = 0;
 703     inneriter        = 0;
 704
 705     for(iidx=0;iidx<4*DIM;iidx++)
 706     {
 707         scratch[iidx] = 0.0;
 708     }
 709
 710     /* Start outer loop over neighborlists */
 711     for(iidx=0; iidx<nri; iidx++)
 712     {
 713         /* Load shift vector for this list */
 714         i_shift_offset   = DIM*shiftidx[iidx];
 715
 716         /* Load limits for loop over neighbors */
 717         j_index_start    = jindex[iidx];
 718         j_index_end      = jindex[iidx+1];
 719
 720         /* Get outer coordinate index */
 721         inr              = iinr[iidx];
 722         i_coord_offset   = DIM*inr;
 723
 724         /* Load i particle coords and add shift vector */
 725         gmx_mm_load_shift_and_3rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
 726                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 727
 728         fix0             = _mm_setzero_ps();
 729         fiy0             = _mm_setzero_ps();
 730         fiz0             = _mm_setzero_ps();
 731         fix1             = _mm_setzero_ps();
 732         fiy1             = _mm_setzero_ps();
 733         fiz1             = _mm_setzero_ps();
 734         fix2             = _mm_setzero_ps();
 735         fiy2             = _mm_setzero_ps();
 736         fiz2             = _mm_setzero_ps();
 737
 738         /* Start inner kernel loop */
 739         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 740         {
 741
 742             /* Get j neighbor index, and coordinate index */
 743             jnrA             = jjnr[jidx];
 744             jnrB             = jjnr[jidx+1];
 745             jnrC             = jjnr[jidx+2];
 746             jnrD             = jjnr[jidx+3];
 747             j_coord_offsetA  = DIM*jnrA;
 748             j_coord_offsetB  = DIM*jnrB;
 749             j_coord_offsetC  = DIM*jnrC;
 750             j_coord_offsetD  = DIM*jnrD;
 751
 752             /* load j atom coordinates */
 753             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 754                                               x+j_coord_offsetC,x+j_coord_offsetD,
 755                                               &jx0,&jy0,&jz0);
 756
 757             /* Calculate displacement vector */
 758             dx00             = _mm_sub_ps(ix0,jx0);
 759             dy00             = _mm_sub_ps(iy0,jy0);
 760             dz00             = _mm_sub_ps(iz0,jz0);
 761             dx10             = _mm_sub_ps(ix1,jx0);
 762             dy10             = _mm_sub_ps(iy1,jy0);
 763             dz10             = _mm_sub_ps(iz1,jz0);
 764             dx20             = _mm_sub_ps(ix2,jx0);
 765             dy20             = _mm_sub_ps(iy2,jy0);
 766             dz20             = _mm_sub_ps(iz2,jz0);
 767
 768             /* Calculate squared distance and things based on it */
 769             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 770             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 771             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 772
 773             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 774             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 775             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 776
 777             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 778             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 779             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 780
 781             /* Load parameters for j particles */
 782             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 783                                                               charge+jnrC+0,charge+jnrD+0);
 784             vdwjidx0A        = 2*vdwtype[jnrA+0];
 785             vdwjidx0B        = 2*vdwtype[jnrB+0];
 786             vdwjidx0C        = 2*vdwtype[jnrC+0];
 787             vdwjidx0D        = 2*vdwtype[jnrD+0];
 788
 789             fjx0             = _mm_setzero_ps();
 790             fjy0             = _mm_setzero_ps();
 791             fjz0             = _mm_setzero_ps();
 792
 793             /**************************
 794              * CALCULATE INTERACTIONS *
 795              **************************/
 796
 797             r00              = _mm_mul_ps(rsq00,rinv00);
 798
 799             /* Compute parameters for interactions between i and j atoms */
 800             qq00             = _mm_mul_ps(iq0,jq0);
 801             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 802                                          vdwparam+vdwioffset0+vdwjidx0B,
 803                                          vdwparam+vdwioffset0+vdwjidx0C,
 804                                          vdwparam+vdwioffset0+vdwjidx0D,
 805                                          &c6_00,&c12_00);
 806
 807             /* EWALD ELECTROSTATICS */
 808
 809             /* Analytical PME correction */
 810             zeta2            = _mm_mul_ps(beta2,rsq00);
 811             rinv3            = _mm_mul_ps(rinvsq00,rinv00);
 812             pmecorrF         = gmx_mm_pmecorrF_ps(zeta2);
 813             felec            = _mm_macc_ps(pmecorrF,beta3,rinv3);
 814             felec            = _mm_mul_ps(qq00,felec);
 815
 816             /* LENNARD-JONES DISPERSION/REPULSION */
 817
 818             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 819             fvdw             = _mm_mul_ps(_mm_msub_ps(c12_00,rinvsix,c6_00),_mm_mul_ps(rinvsix,rinvsq00));
 820
 821             fscal            = _mm_add_ps(felec,fvdw);
 822
 823              /* Update vectorial force */
 824             fix0             = _mm_macc_ps(dx00,fscal,fix0);
 825             fiy0             = _mm_macc_ps(dy00,fscal,fiy0);
 826             fiz0             = _mm_macc_ps(dz00,fscal,fiz0);
 827
 828             fjx0             = _mm_macc_ps(dx00,fscal,fjx0);
 829             fjy0             = _mm_macc_ps(dy00,fscal,fjy0);
 830             fjz0             = _mm_macc_ps(dz00,fscal,fjz0);
 831
 832             /**************************
 833              * CALCULATE INTERACTIONS *
 834              **************************/
 835
 836             r10              = _mm_mul_ps(rsq10,rinv10);
 837
 838             /* Compute parameters for interactions between i and j atoms */
 839             qq10             = _mm_mul_ps(iq1,jq0);
 840
 841             /* EWALD ELECTROSTATICS */
 842
 843             /* Analytical PME correction */
 844             zeta2            = _mm_mul_ps(beta2,rsq10);
 845             rinv3            = _mm_mul_ps(rinvsq10,rinv10);
 846             pmecorrF         = gmx_mm_pmecorrF_ps(zeta2);
 847             felec            = _mm_macc_ps(pmecorrF,beta3,rinv3);
 848             felec            = _mm_mul_ps(qq10,felec);
 849
 850             fscal            = felec;
 851
 852              /* Update vectorial force */
 853             fix1             = _mm_macc_ps(dx10,fscal,fix1);
 854             fiy1             = _mm_macc_ps(dy10,fscal,fiy1);
 855             fiz1             = _mm_macc_ps(dz10,fscal,fiz1);
 856
 857             fjx0             = _mm_macc_ps(dx10,fscal,fjx0);
 858             fjy0             = _mm_macc_ps(dy10,fscal,fjy0);
 859             fjz0             = _mm_macc_ps(dz10,fscal,fjz0);
 860
 861             /**************************
 862              * CALCULATE INTERACTIONS *
 863              **************************/
 864
 865             r20              = _mm_mul_ps(rsq20,rinv20);
 866
 867             /* Compute parameters for interactions between i and j atoms */
 868             qq20             = _mm_mul_ps(iq2,jq0);
 869
 870             /* EWALD ELECTROSTATICS */
 871
 872             /* Analytical PME correction */
 873             zeta2            = _mm_mul_ps(beta2,rsq20);
 874             rinv3            = _mm_mul_ps(rinvsq20,rinv20);
 875             pmecorrF         = gmx_mm_pmecorrF_ps(zeta2);
 876             felec            = _mm_macc_ps(pmecorrF,beta3,rinv3);
 877             felec            = _mm_mul_ps(qq20,felec);
 878
 879             fscal            = felec;
 880
 881              /* Update vectorial force */
 882             fix2             = _mm_macc_ps(dx20,fscal,fix2);
 883             fiy2             = _mm_macc_ps(dy20,fscal,fiy2);
 884             fiz2             = _mm_macc_ps(dz20,fscal,fiz2);
 885
 886             fjx0             = _mm_macc_ps(dx20,fscal,fjx0);
 887             fjy0             = _mm_macc_ps(dy20,fscal,fjy0);
 888             fjz0             = _mm_macc_ps(dz20,fscal,fjz0);
 889
 890             fjptrA             = f+j_coord_offsetA;
 891             fjptrB             = f+j_coord_offsetB;
 892             fjptrC             = f+j_coord_offsetC;
 893             fjptrD             = f+j_coord_offsetD;
 894
 895             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 896
 897             /* Inner loop uses 91 flops */
 898         }
 899
 900         if(jidx<j_index_end)
 901         {
 902
 903             /* Get j neighbor index, and coordinate index */
 904             jnrlistA         = jjnr[jidx];
 905             jnrlistB         = jjnr[jidx+1];
 906             jnrlistC         = jjnr[jidx+2];
 907             jnrlistD         = jjnr[jidx+3];
 908             /* Sign of each element will be negative for non-real atoms.
 909              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 910              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 911              */
 912             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 913             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 914             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 915             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 916             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 917             j_coord_offsetA  = DIM*jnrA;
 918             j_coord_offsetB  = DIM*jnrB;
 919             j_coord_offsetC  = DIM*jnrC;
 920             j_coord_offsetD  = DIM*jnrD;
 921
 922             /* load j atom coordinates */
 923             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 924                                               x+j_coord_offsetC,x+j_coord_offsetD,
 925                                               &jx0,&jy0,&jz0);
 926
 927             /* Calculate displacement vector */
 928             dx00             = _mm_sub_ps(ix0,jx0);
 929             dy00             = _mm_sub_ps(iy0,jy0);
 930             dz00             = _mm_sub_ps(iz0,jz0);
 931             dx10             = _mm_sub_ps(ix1,jx0);
 932             dy10             = _mm_sub_ps(iy1,jy0);
 933             dz10             = _mm_sub_ps(iz1,jz0);
 934             dx20             = _mm_sub_ps(ix2,jx0);
 935             dy20             = _mm_sub_ps(iy2,jy0);
 936             dz20             = _mm_sub_ps(iz2,jz0);
 937
 938             /* Calculate squared distance and things based on it */
 939             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 940             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 941             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 942
 943             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 944             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 945             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 946
 947             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 948             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 949             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 950
 951             /* Load parameters for j particles */
 952             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 953                                                               charge+jnrC+0,charge+jnrD+0);
 954             vdwjidx0A        = 2*vdwtype[jnrA+0];
 955             vdwjidx0B        = 2*vdwtype[jnrB+0];
 956             vdwjidx0C        = 2*vdwtype[jnrC+0];
 957             vdwjidx0D        = 2*vdwtype[jnrD+0];
 958
 959             fjx0             = _mm_setzero_ps();
 960             fjy0             = _mm_setzero_ps();
 961             fjz0             = _mm_setzero_ps();
 962
 963             /**************************
 964              * CALCULATE INTERACTIONS *
 965              **************************/
 966
 967             r00              = _mm_mul_ps(rsq00,rinv00);
 968             r00              = _mm_andnot_ps(dummy_mask,r00);
 969
 970             /* Compute parameters for interactions between i and j atoms */
 971             qq00             = _mm_mul_ps(iq0,jq0);
 972             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 973                                          vdwparam+vdwioffset0+vdwjidx0B,
 974                                          vdwparam+vdwioffset0+vdwjidx0C,
 975                                          vdwparam+vdwioffset0+vdwjidx0D,
 976                                          &c6_00,&c12_00);
 977
 978             /* EWALD ELECTROSTATICS */
 979
 980             /* Analytical PME correction */
 981             zeta2            = _mm_mul_ps(beta2,rsq00);
 982             rinv3            = _mm_mul_ps(rinvsq00,rinv00);
 983             pmecorrF         = gmx_mm_pmecorrF_ps(zeta2);
 984             felec            = _mm_macc_ps(pmecorrF,beta3,rinv3);
 985             felec            = _mm_mul_ps(qq00,felec);
 986
 987             /* LENNARD-JONES DISPERSION/REPULSION */
 988
 989             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 990             fvdw             = _mm_mul_ps(_mm_msub_ps(c12_00,rinvsix,c6_00),_mm_mul_ps(rinvsix,rinvsq00));
 991
 992             fscal            = _mm_add_ps(felec,fvdw);
 993
 994             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 995
 996              /* Update vectorial force */
 997             fix0             = _mm_macc_ps(dx00,fscal,fix0);
 998             fiy0             = _mm_macc_ps(dy00,fscal,fiy0);
 999             fiz0             = _mm_macc_ps(dz00,fscal,fiz0);
1000
1001             fjx0             = _mm_macc_ps(dx00,fscal,fjx0);
1002             fjy0             = _mm_macc_ps(dy00,fscal,fjy0);
1003             fjz0             = _mm_macc_ps(dz00,fscal,fjz0);
1004
1005             /**************************
1006              * CALCULATE INTERACTIONS *
1007              **************************/
1008
1009             r10              = _mm_mul_ps(rsq10,rinv10);
1010             r10              = _mm_andnot_ps(dummy_mask,r10);
1011
1012             /* Compute parameters for interactions between i and j atoms */
1013             qq10             = _mm_mul_ps(iq1,jq0);
1014
1015             /* EWALD ELECTROSTATICS */
1016
1017             /* Analytical PME correction */
1018             zeta2            = _mm_mul_ps(beta2,rsq10);
1019             rinv3            = _mm_mul_ps(rinvsq10,rinv10);
1020             pmecorrF         = gmx_mm_pmecorrF_ps(zeta2);
1021             felec            = _mm_macc_ps(pmecorrF,beta3,rinv3);
1022             felec            = _mm_mul_ps(qq10,felec);
1023
1024             fscal            = felec;
1025
1026             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1027
1028              /* Update vectorial force */
1029             fix1             = _mm_macc_ps(dx10,fscal,fix1);
1030             fiy1             = _mm_macc_ps(dy10,fscal,fiy1);
1031             fiz1             = _mm_macc_ps(dz10,fscal,fiz1);
1032
1033             fjx0             = _mm_macc_ps(dx10,fscal,fjx0);
1034             fjy0             = _mm_macc_ps(dy10,fscal,fjy0);
1035             fjz0             = _mm_macc_ps(dz10,fscal,fjz0);
1036
1037             /**************************
1038              * CALCULATE INTERACTIONS *
1039              **************************/
1040
1041             r20              = _mm_mul_ps(rsq20,rinv20);
1042             r20              = _mm_andnot_ps(dummy_mask,r20);
1043
1044             /* Compute parameters for interactions between i and j atoms */
1045             qq20             = _mm_mul_ps(iq2,jq0);
1046
1047             /* EWALD ELECTROSTATICS */
1048
1049             /* Analytical PME correction */
1050             zeta2            = _mm_mul_ps(beta2,rsq20);
1051             rinv3            = _mm_mul_ps(rinvsq20,rinv20);
1052             pmecorrF         = gmx_mm_pmecorrF_ps(zeta2);
1053             felec            = _mm_macc_ps(pmecorrF,beta3,rinv3);
1054             felec            = _mm_mul_ps(qq20,felec);
1055
1056             fscal            = felec;
1057
1058             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1059
1060              /* Update vectorial force */
1061             fix2             = _mm_macc_ps(dx20,fscal,fix2);
1062             fiy2             = _mm_macc_ps(dy20,fscal,fiy2);
1063             fiz2             = _mm_macc_ps(dz20,fscal,fiz2);
1064
1065             fjx0             = _mm_macc_ps(dx20,fscal,fjx0);
1066             fjy0             = _mm_macc_ps(dy20,fscal,fjy0);
1067             fjz0             = _mm_macc_ps(dz20,fscal,fjz0);
1068
1069             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
1070             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
1071             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
1072             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
1073
1074             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
1075
1076             /* Inner loop uses 94 flops */
1077         }
1078
1079         /* End of innermost loop */
1080
1081         gmx_mm_update_iforce_3atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
1082                                               f+i_coord_offset,fshift+i_shift_offset);
1083
1084         /* Increment number of inner iterations */
1085         inneriter                  += j_index_end - j_index_start;
1086
1087         /* Outer loop uses 18 flops */
1088     }
1089
1090     /* Increment number of outer iterations */
1091     outeriter        += nri;
1092
1093     /* Update outer/inner flops */
1094
1095     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3_F,outeriter*18 + inneriter*94);
1096 }