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) 2012,2013,2014, 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_VdwLJ_GeomW3P1_VF_avx_128_fma_single
  52  * Electrostatics interaction: Ewald
  53  * VdW interaction:            LennardJones
  54  * Geometry:                   Water3-Particle
  55  * Calculate force/pot:        PotentialAndForce
  56  */
  57 void
  58 nb_kernel_ElecEw_VdwLJ_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     __m128i          ewitab;
 103     __m128           ewtabscale,eweps,twoeweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
 104     __m128           beta,beta2,beta3,zeta2,pmecorrF,pmecorrV,rinv3;
 105     real             *ewtab;
 106     __m128           dummy_mask,cutoff_mask;
 107     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
 108     __m128           one     = _mm_set1_ps(1.0);
 109     __m128           two     = _mm_set1_ps(2.0);
 110     x                = xx[0];
 111     f                = ff[0];
 112
 113     nri              = nlist->nri;
 114     iinr             = nlist->iinr;
 115     jindex           = nlist->jindex;
 116     jjnr             = nlist->jjnr;
 117     shiftidx         = nlist->shift;
 118     gid              = nlist->gid;
 119     shiftvec         = fr->shift_vec[0];
 120     fshift           = fr->fshift[0];
 121     facel            = _mm_set1_ps(fr->epsfac);
 122     charge           = mdatoms->chargeA;
 123     nvdwtype         = fr->ntype;
 124     vdwparam         = fr->nbfp;
 125     vdwtype          = mdatoms->typeA;
 126
 127     sh_ewald         = _mm_set1_ps(fr->ic->sh_ewald);
 128     beta             = _mm_set1_ps(fr->ic->ewaldcoeff_q);
 129     beta2            = _mm_mul_ps(beta,beta);
 130     beta3            = _mm_mul_ps(beta,beta2);
 131     ewtab            = fr->ic->tabq_coul_FDV0;
 132     ewtabscale       = _mm_set1_ps(fr->ic->tabq_scale);
 133     ewtabhalfspace   = _mm_set1_ps(0.5/fr->ic->tabq_scale);
 134
 135     /* Setup water-specific parameters */
 136     inr              = nlist->iinr[0];
 137     iq0              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+0]));
 138     iq1              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+1]));
 139     iq2              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+2]));
 140     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 141
 142     /* Avoid stupid compiler warnings */
 143     jnrA = jnrB = jnrC = jnrD = 0;
 144     j_coord_offsetA = 0;
 145     j_coord_offsetB = 0;
 146     j_coord_offsetC = 0;
 147     j_coord_offsetD = 0;
 148
 149     outeriter        = 0;
 150     inneriter        = 0;
 151
 152     for(iidx=0;iidx<4*DIM;iidx++)
 153     {
 154         scratch[iidx] = 0.0;
 155     }
 156
 157     /* Start outer loop over neighborlists */
 158     for(iidx=0; iidx<nri; iidx++)
 159     {
 160         /* Load shift vector for this list */
 161         i_shift_offset   = DIM*shiftidx[iidx];
 162
 163         /* Load limits for loop over neighbors */
 164         j_index_start    = jindex[iidx];
 165         j_index_end      = jindex[iidx+1];
 166
 167         /* Get outer coordinate index */
 168         inr              = iinr[iidx];
 169         i_coord_offset   = DIM*inr;
 170
 171         /* Load i particle coords and add shift vector */
 172         gmx_mm_load_shift_and_3rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
 173                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 174
 175         fix0             = _mm_setzero_ps();
 176         fiy0             = _mm_setzero_ps();
 177         fiz0             = _mm_setzero_ps();
 178         fix1             = _mm_setzero_ps();
 179         fiy1             = _mm_setzero_ps();
 180         fiz1             = _mm_setzero_ps();
 181         fix2             = _mm_setzero_ps();
 182         fiy2             = _mm_setzero_ps();
 183         fiz2             = _mm_setzero_ps();
 184
 185         /* Reset potential sums */
 186         velecsum         = _mm_setzero_ps();
 187         vvdwsum          = _mm_setzero_ps();
 188
 189         /* Start inner kernel loop */
 190         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 191         {
 192
 193             /* Get j neighbor index, and coordinate index */
 194             jnrA             = jjnr[jidx];
 195             jnrB             = jjnr[jidx+1];
 196             jnrC             = jjnr[jidx+2];
 197             jnrD             = jjnr[jidx+3];
 198             j_coord_offsetA  = DIM*jnrA;
 199             j_coord_offsetB  = DIM*jnrB;
 200             j_coord_offsetC  = DIM*jnrC;
 201             j_coord_offsetD  = DIM*jnrD;
 202
 203             /* load j atom coordinates */
 204             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 205                                               x+j_coord_offsetC,x+j_coord_offsetD,
 206                                               &jx0,&jy0,&jz0);
 207
 208             /* Calculate displacement vector */
 209             dx00             = _mm_sub_ps(ix0,jx0);
 210             dy00             = _mm_sub_ps(iy0,jy0);
 211             dz00             = _mm_sub_ps(iz0,jz0);
 212             dx10             = _mm_sub_ps(ix1,jx0);
 213             dy10             = _mm_sub_ps(iy1,jy0);
 214             dz10             = _mm_sub_ps(iz1,jz0);
 215             dx20             = _mm_sub_ps(ix2,jx0);
 216             dy20             = _mm_sub_ps(iy2,jy0);
 217             dz20             = _mm_sub_ps(iz2,jz0);
 218
 219             /* Calculate squared distance and things based on it */
 220             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 221             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 222             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 223
 224             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 225             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 226             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 227
 228             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 229             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 230             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 231
 232             /* Load parameters for j particles */
 233             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 234                                                               charge+jnrC+0,charge+jnrD+0);
 235             vdwjidx0A        = 2*vdwtype[jnrA+0];
 236             vdwjidx0B        = 2*vdwtype[jnrB+0];
 237             vdwjidx0C        = 2*vdwtype[jnrC+0];
 238             vdwjidx0D        = 2*vdwtype[jnrD+0];
 239
 240             fjx0             = _mm_setzero_ps();
 241             fjy0             = _mm_setzero_ps();
 242             fjz0             = _mm_setzero_ps();
 243
 244             /**************************
 245              * CALCULATE INTERACTIONS *
 246              **************************/
 247
 248             r00              = _mm_mul_ps(rsq00,rinv00);
 249
 250             /* Compute parameters for interactions between i and j atoms */
 251             qq00             = _mm_mul_ps(iq0,jq0);
 252             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 253                                          vdwparam+vdwioffset0+vdwjidx0B,
 254                                          vdwparam+vdwioffset0+vdwjidx0C,
 255                                          vdwparam+vdwioffset0+vdwjidx0D,
 256                                          &c6_00,&c12_00);
 257
 258             /* EWALD ELECTROSTATICS */
 259
 260             /* Analytical PME correction */
 261             zeta2            = _mm_mul_ps(beta2,rsq00);
 262             rinv3            = _mm_mul_ps(rinvsq00,rinv00);
 263             pmecorrF         = gmx_mm_pmecorrF_ps(zeta2);
 264             felec            = _mm_macc_ps(pmecorrF,beta3,rinv3);
 265             felec            = _mm_mul_ps(qq00,felec);
 266             pmecorrV         = gmx_mm_pmecorrV_ps(zeta2);
 267             velec            = _mm_nmacc_ps(pmecorrV,beta,rinv00);
 268             velec            = _mm_mul_ps(qq00,velec);
 269
 270             /* LENNARD-JONES DISPERSION/REPULSION */
 271
 272             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 273             vvdw6            = _mm_mul_ps(c6_00,rinvsix);
 274             vvdw12           = _mm_mul_ps(c12_00,_mm_mul_ps(rinvsix,rinvsix));
 275             vvdw             = _mm_msub_ps(vvdw12,one_twelfth,_mm_mul_ps(vvdw6,one_sixth));
 276             fvdw             = _mm_mul_ps(_mm_sub_ps(vvdw12,vvdw6),rinvsq00);
 277
 278             /* Update potential sum for this i atom from the interaction with this j atom. */
 279             velecsum         = _mm_add_ps(velecsum,velec);
 280             vvdwsum          = _mm_add_ps(vvdwsum,vvdw);
 281
 282             fscal            = _mm_add_ps(felec,fvdw);
 283
 284              /* Update vectorial force */
 285             fix0             = _mm_macc_ps(dx00,fscal,fix0);
 286             fiy0             = _mm_macc_ps(dy00,fscal,fiy0);
 287             fiz0             = _mm_macc_ps(dz00,fscal,fiz0);
 288
 289             fjx0             = _mm_macc_ps(dx00,fscal,fjx0);
 290             fjy0             = _mm_macc_ps(dy00,fscal,fjy0);
 291             fjz0             = _mm_macc_ps(dz00,fscal,fjz0);
 292
 293             /**************************
 294              * CALCULATE INTERACTIONS *
 295              **************************/
 296
 297             r10              = _mm_mul_ps(rsq10,rinv10);
 298
 299             /* Compute parameters for interactions between i and j atoms */
 300             qq10             = _mm_mul_ps(iq1,jq0);
 301
 302             /* EWALD ELECTROSTATICS */
 303
 304             /* Analytical PME correction */
 305             zeta2            = _mm_mul_ps(beta2,rsq10);
 306             rinv3            = _mm_mul_ps(rinvsq10,rinv10);
 307             pmecorrF         = gmx_mm_pmecorrF_ps(zeta2);
 308             felec            = _mm_macc_ps(pmecorrF,beta3,rinv3);
 309             felec            = _mm_mul_ps(qq10,felec);
 310             pmecorrV         = gmx_mm_pmecorrV_ps(zeta2);
 311             velec            = _mm_nmacc_ps(pmecorrV,beta,rinv10);
 312             velec            = _mm_mul_ps(qq10,velec);
 313
 314             /* Update potential sum for this i atom from the interaction with this j atom. */
 315             velecsum         = _mm_add_ps(velecsum,velec);
 316
 317             fscal            = felec;
 318
 319              /* Update vectorial force */
 320             fix1             = _mm_macc_ps(dx10,fscal,fix1);
 321             fiy1             = _mm_macc_ps(dy10,fscal,fiy1);
 322             fiz1             = _mm_macc_ps(dz10,fscal,fiz1);
 323
 324             fjx0             = _mm_macc_ps(dx10,fscal,fjx0);
 325             fjy0             = _mm_macc_ps(dy10,fscal,fjy0);
 326             fjz0             = _mm_macc_ps(dz10,fscal,fjz0);
 327
 328             /**************************
 329              * CALCULATE INTERACTIONS *
 330              **************************/
 331
 332             r20              = _mm_mul_ps(rsq20,rinv20);
 333
 334             /* Compute parameters for interactions between i and j atoms */
 335             qq20             = _mm_mul_ps(iq2,jq0);
 336
 337             /* EWALD ELECTROSTATICS */
 338
 339             /* Analytical PME correction */
 340             zeta2            = _mm_mul_ps(beta2,rsq20);
 341             rinv3            = _mm_mul_ps(rinvsq20,rinv20);
 342             pmecorrF         = gmx_mm_pmecorrF_ps(zeta2);
 343             felec            = _mm_macc_ps(pmecorrF,beta3,rinv3);
 344             felec            = _mm_mul_ps(qq20,felec);
 345             pmecorrV         = gmx_mm_pmecorrV_ps(zeta2);
 346             velec            = _mm_nmacc_ps(pmecorrV,beta,rinv20);
 347             velec            = _mm_mul_ps(qq20,velec);
 348
 349             /* Update potential sum for this i atom from the interaction with this j atom. */
 350             velecsum         = _mm_add_ps(velecsum,velec);
 351
 352             fscal            = felec;
 353
 354              /* Update vectorial force */
 355             fix2             = _mm_macc_ps(dx20,fscal,fix2);
 356             fiy2             = _mm_macc_ps(dy20,fscal,fiy2);
 357             fiz2             = _mm_macc_ps(dz20,fscal,fiz2);
 358
 359             fjx0             = _mm_macc_ps(dx20,fscal,fjx0);
 360             fjy0             = _mm_macc_ps(dy20,fscal,fjy0);
 361             fjz0             = _mm_macc_ps(dz20,fscal,fjz0);
 362
 363             fjptrA             = f+j_coord_offsetA;
 364             fjptrB             = f+j_coord_offsetB;
 365             fjptrC             = f+j_coord_offsetC;
 366             fjptrD             = f+j_coord_offsetD;
 367
 368             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 369
 370             /* Inner loop uses 99 flops */
 371         }
 372
 373         if(jidx<j_index_end)
 374         {
 375
 376             /* Get j neighbor index, and coordinate index */
 377             jnrlistA         = jjnr[jidx];
 378             jnrlistB         = jjnr[jidx+1];
 379             jnrlistC         = jjnr[jidx+2];
 380             jnrlistD         = jjnr[jidx+3];
 381             /* Sign of each element will be negative for non-real atoms.
 382              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 383              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 384              */
 385             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 386             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 387             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 388             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 389             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 390             j_coord_offsetA  = DIM*jnrA;
 391             j_coord_offsetB  = DIM*jnrB;
 392             j_coord_offsetC  = DIM*jnrC;
 393             j_coord_offsetD  = DIM*jnrD;
 394
 395             /* load j atom coordinates */
 396             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 397                                               x+j_coord_offsetC,x+j_coord_offsetD,
 398                                               &jx0,&jy0,&jz0);
 399
 400             /* Calculate displacement vector */
 401             dx00             = _mm_sub_ps(ix0,jx0);
 402             dy00             = _mm_sub_ps(iy0,jy0);
 403             dz00             = _mm_sub_ps(iz0,jz0);
 404             dx10             = _mm_sub_ps(ix1,jx0);
 405             dy10             = _mm_sub_ps(iy1,jy0);
 406             dz10             = _mm_sub_ps(iz1,jz0);
 407             dx20             = _mm_sub_ps(ix2,jx0);
 408             dy20             = _mm_sub_ps(iy2,jy0);
 409             dz20             = _mm_sub_ps(iz2,jz0);
 410
 411             /* Calculate squared distance and things based on it */
 412             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 413             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 414             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 415
 416             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 417             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 418             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 419
 420             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 421             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 422             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 423
 424             /* Load parameters for j particles */
 425             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 426                                                               charge+jnrC+0,charge+jnrD+0);
 427             vdwjidx0A        = 2*vdwtype[jnrA+0];
 428             vdwjidx0B        = 2*vdwtype[jnrB+0];
 429             vdwjidx0C        = 2*vdwtype[jnrC+0];
 430             vdwjidx0D        = 2*vdwtype[jnrD+0];
 431
 432             fjx0             = _mm_setzero_ps();
 433             fjy0             = _mm_setzero_ps();
 434             fjz0             = _mm_setzero_ps();
 435
 436             /**************************
 437              * CALCULATE INTERACTIONS *
 438              **************************/
 439
 440             r00              = _mm_mul_ps(rsq00,rinv00);
 441             r00              = _mm_andnot_ps(dummy_mask,r00);
 442
 443             /* Compute parameters for interactions between i and j atoms */
 444             qq00             = _mm_mul_ps(iq0,jq0);
 445             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 446                                          vdwparam+vdwioffset0+vdwjidx0B,
 447                                          vdwparam+vdwioffset0+vdwjidx0C,
 448                                          vdwparam+vdwioffset0+vdwjidx0D,
 449                                          &c6_00,&c12_00);
 450
 451             /* EWALD ELECTROSTATICS */
 452
 453             /* Analytical PME correction */
 454             zeta2            = _mm_mul_ps(beta2,rsq00);
 455             rinv3            = _mm_mul_ps(rinvsq00,rinv00);
 456             pmecorrF         = gmx_mm_pmecorrF_ps(zeta2);
 457             felec            = _mm_macc_ps(pmecorrF,beta3,rinv3);
 458             felec            = _mm_mul_ps(qq00,felec);
 459             pmecorrV         = gmx_mm_pmecorrV_ps(zeta2);
 460             velec            = _mm_nmacc_ps(pmecorrV,beta,rinv00);
 461             velec            = _mm_mul_ps(qq00,velec);
 462
 463             /* LENNARD-JONES DISPERSION/REPULSION */
 464
 465             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 466             vvdw6            = _mm_mul_ps(c6_00,rinvsix);
 467             vvdw12           = _mm_mul_ps(c12_00,_mm_mul_ps(rinvsix,rinvsix));
 468             vvdw             = _mm_msub_ps(vvdw12,one_twelfth,_mm_mul_ps(vvdw6,one_sixth));
 469             fvdw             = _mm_mul_ps(_mm_sub_ps(vvdw12,vvdw6),rinvsq00);
 470
 471             /* Update potential sum for this i atom from the interaction with this j atom. */
 472             velec            = _mm_andnot_ps(dummy_mask,velec);
 473             velecsum         = _mm_add_ps(velecsum,velec);
 474             vvdw             = _mm_andnot_ps(dummy_mask,vvdw);
 475             vvdwsum          = _mm_add_ps(vvdwsum,vvdw);
 476
 477             fscal            = _mm_add_ps(felec,fvdw);
 478
 479             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 480
 481              /* Update vectorial force */
 482             fix0             = _mm_macc_ps(dx00,fscal,fix0);
 483             fiy0             = _mm_macc_ps(dy00,fscal,fiy0);
 484             fiz0             = _mm_macc_ps(dz00,fscal,fiz0);
 485
 486             fjx0             = _mm_macc_ps(dx00,fscal,fjx0);
 487             fjy0             = _mm_macc_ps(dy00,fscal,fjy0);
 488             fjz0             = _mm_macc_ps(dz00,fscal,fjz0);
 489
 490             /**************************
 491              * CALCULATE INTERACTIONS *
 492              **************************/
 493
 494             r10              = _mm_mul_ps(rsq10,rinv10);
 495             r10              = _mm_andnot_ps(dummy_mask,r10);
 496
 497             /* Compute parameters for interactions between i and j atoms */
 498             qq10             = _mm_mul_ps(iq1,jq0);
 499
 500             /* EWALD ELECTROSTATICS */
 501
 502             /* Analytical PME correction */
 503             zeta2            = _mm_mul_ps(beta2,rsq10);
 504             rinv3            = _mm_mul_ps(rinvsq10,rinv10);
 505             pmecorrF         = gmx_mm_pmecorrF_ps(zeta2);
 506             felec            = _mm_macc_ps(pmecorrF,beta3,rinv3);
 507             felec            = _mm_mul_ps(qq10,felec);
 508             pmecorrV         = gmx_mm_pmecorrV_ps(zeta2);
 509             velec            = _mm_nmacc_ps(pmecorrV,beta,rinv10);
 510             velec            = _mm_mul_ps(qq10,velec);
 511
 512             /* Update potential sum for this i atom from the interaction with this j atom. */
 513             velec            = _mm_andnot_ps(dummy_mask,velec);
 514             velecsum         = _mm_add_ps(velecsum,velec);
 515
 516             fscal            = felec;
 517
 518             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 519
 520              /* Update vectorial force */
 521             fix1             = _mm_macc_ps(dx10,fscal,fix1);
 522             fiy1             = _mm_macc_ps(dy10,fscal,fiy1);
 523             fiz1             = _mm_macc_ps(dz10,fscal,fiz1);
 524
 525             fjx0             = _mm_macc_ps(dx10,fscal,fjx0);
 526             fjy0             = _mm_macc_ps(dy10,fscal,fjy0);
 527             fjz0             = _mm_macc_ps(dz10,fscal,fjz0);
 528
 529             /**************************
 530              * CALCULATE INTERACTIONS *
 531              **************************/
 532
 533             r20              = _mm_mul_ps(rsq20,rinv20);
 534             r20              = _mm_andnot_ps(dummy_mask,r20);
 535
 536             /* Compute parameters for interactions between i and j atoms */
 537             qq20             = _mm_mul_ps(iq2,jq0);
 538
 539             /* EWALD ELECTROSTATICS */
 540
 541             /* Analytical PME correction */
 542             zeta2            = _mm_mul_ps(beta2,rsq20);
 543             rinv3            = _mm_mul_ps(rinvsq20,rinv20);
 544             pmecorrF         = gmx_mm_pmecorrF_ps(zeta2);
 545             felec            = _mm_macc_ps(pmecorrF,beta3,rinv3);
 546             felec            = _mm_mul_ps(qq20,felec);
 547             pmecorrV         = gmx_mm_pmecorrV_ps(zeta2);
 548             velec            = _mm_nmacc_ps(pmecorrV,beta,rinv20);
 549             velec            = _mm_mul_ps(qq20,velec);
 550
 551             /* Update potential sum for this i atom from the interaction with this j atom. */
 552             velec            = _mm_andnot_ps(dummy_mask,velec);
 553             velecsum         = _mm_add_ps(velecsum,velec);
 554
 555             fscal            = felec;
 556
 557             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 558
 559              /* Update vectorial force */
 560             fix2             = _mm_macc_ps(dx20,fscal,fix2);
 561             fiy2             = _mm_macc_ps(dy20,fscal,fiy2);
 562             fiz2             = _mm_macc_ps(dz20,fscal,fiz2);
 563
 564             fjx0             = _mm_macc_ps(dx20,fscal,fjx0);
 565             fjy0             = _mm_macc_ps(dy20,fscal,fjy0);
 566             fjz0             = _mm_macc_ps(dz20,fscal,fjz0);
 567
 568             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 569             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 570             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 571             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 572
 573             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 574
 575             /* Inner loop uses 102 flops */
 576         }
 577
 578         /* End of innermost loop */
 579
 580         gmx_mm_update_iforce_3atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
 581                                               f+i_coord_offset,fshift+i_shift_offset);
 582
 583         ggid                        = gid[iidx];
 584         /* Update potential energies */
 585         gmx_mm_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid);
 586         gmx_mm_update_1pot_ps(vvdwsum,kernel_data->energygrp_vdw+ggid);
 587
 588         /* Increment number of inner iterations */
 589         inneriter                  += j_index_end - j_index_start;
 590
 591         /* Outer loop uses 20 flops */
 592     }
 593
 594     /* Increment number of outer iterations */
 595     outeriter        += nri;
 596
 597     /* Update outer/inner flops */
 598
 599     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3_VF,outeriter*20 + inneriter*102);
 600 }
 601 /*
 602  * Gromacs nonbonded kernel:   nb_kernel_ElecEw_VdwLJ_GeomW3P1_F_avx_128_fma_single
 603  * Electrostatics interaction: Ewald
 604  * VdW interaction:            LennardJones
 605  * Geometry:                   Water3-Particle
 606  * Calculate force/pot:        Force
 607  */
 608 void
 609 nb_kernel_ElecEw_VdwLJ_GeomW3P1_F_avx_128_fma_single
 610                     (t_nblist                    * gmx_restrict       nlist,
 611                      rvec                        * gmx_restrict          xx,
 612                      rvec                        * gmx_restrict          ff,
 613                      t_forcerec                  * gmx_restrict          fr,
 614                      t_mdatoms                   * gmx_restrict     mdatoms,
 615                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 616                      t_nrnb                      * gmx_restrict        nrnb)
 617 {
 618     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 619      * just 0 for non-waters.
 620      * Suffixes A,B,C,D refer to j loop unrolling done with AVX_128, e.g. for the four different
 621      * jnr indices corresponding to data put in the four positions in the SIMD register.
 622      */
 623     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 624     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 625     int              jnrA,jnrB,jnrC,jnrD;
 626     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
 627     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 628     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 629     real             rcutoff_scalar;
 630     real             *shiftvec,*fshift,*x,*f;
 631     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
 632     real             scratch[4*DIM];
 633     __m128           fscal,rcutoff,rcutoff2,jidxall;
 634     int              vdwioffset0;
 635     __m128           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 636     int              vdwioffset1;
 637     __m128           ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
 638     int              vdwioffset2;
 639     __m128           ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
 640     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
 641     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 642     __m128           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 643     __m128           dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
 644     __m128           dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
 645     __m128           velec,felec,velecsum,facel,crf,krf,krf2;
 646     real             *charge;
 647     int              nvdwtype;
 648     __m128           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 649     int              *vdwtype;
 650     real             *vdwparam;
 651     __m128           one_sixth   = _mm_set1_ps(1.0/6.0);
 652     __m128           one_twelfth = _mm_set1_ps(1.0/12.0);
 653     __m128i          ewitab;
 654     __m128           ewtabscale,eweps,twoeweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
 655     __m128           beta,beta2,beta3,zeta2,pmecorrF,pmecorrV,rinv3;
 656     real             *ewtab;
 657     __m128           dummy_mask,cutoff_mask;
 658     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
 659     __m128           one     = _mm_set1_ps(1.0);
 660     __m128           two     = _mm_set1_ps(2.0);
 661     x                = xx[0];
 662     f                = ff[0];
 663
 664     nri              = nlist->nri;
 665     iinr             = nlist->iinr;
 666     jindex           = nlist->jindex;
 667     jjnr             = nlist->jjnr;
 668     shiftidx         = nlist->shift;
 669     gid              = nlist->gid;
 670     shiftvec         = fr->shift_vec[0];
 671     fshift           = fr->fshift[0];
 672     facel            = _mm_set1_ps(fr->epsfac);
 673     charge           = mdatoms->chargeA;
 674     nvdwtype         = fr->ntype;
 675     vdwparam         = fr->nbfp;
 676     vdwtype          = mdatoms->typeA;
 677
 678     sh_ewald         = _mm_set1_ps(fr->ic->sh_ewald);
 679     beta             = _mm_set1_ps(fr->ic->ewaldcoeff_q);
 680     beta2            = _mm_mul_ps(beta,beta);
 681     beta3            = _mm_mul_ps(beta,beta2);
 682     ewtab            = fr->ic->tabq_coul_F;
 683     ewtabscale       = _mm_set1_ps(fr->ic->tabq_scale);
 684     ewtabhalfspace   = _mm_set1_ps(0.5/fr->ic->tabq_scale);
 685
 686     /* Setup water-specific parameters */
 687     inr              = nlist->iinr[0];
 688     iq0              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+0]));
 689     iq1              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+1]));
 690     iq2              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+2]));
 691     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 692
 693     /* Avoid stupid compiler warnings */
 694     jnrA = jnrB = jnrC = jnrD = 0;
 695     j_coord_offsetA = 0;
 696     j_coord_offsetB = 0;
 697     j_coord_offsetC = 0;
 698     j_coord_offsetD = 0;
 699
 700     outeriter        = 0;
 701     inneriter        = 0;
 702
 703     for(iidx=0;iidx<4*DIM;iidx++)
 704     {
 705         scratch[iidx] = 0.0;
 706     }
 707
 708     /* Start outer loop over neighborlists */
 709     for(iidx=0; iidx<nri; iidx++)
 710     {
 711         /* Load shift vector for this list */
 712         i_shift_offset   = DIM*shiftidx[iidx];
 713
 714         /* Load limits for loop over neighbors */
 715         j_index_start    = jindex[iidx];
 716         j_index_end      = jindex[iidx+1];
 717
 718         /* Get outer coordinate index */
 719         inr              = iinr[iidx];
 720         i_coord_offset   = DIM*inr;
 721
 722         /* Load i particle coords and add shift vector */
 723         gmx_mm_load_shift_and_3rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
 724                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 725
 726         fix0             = _mm_setzero_ps();
 727         fiy0             = _mm_setzero_ps();
 728         fiz0             = _mm_setzero_ps();
 729         fix1             = _mm_setzero_ps();
 730         fiy1             = _mm_setzero_ps();
 731         fiz1             = _mm_setzero_ps();
 732         fix2             = _mm_setzero_ps();
 733         fiy2             = _mm_setzero_ps();
 734         fiz2             = _mm_setzero_ps();
 735
 736         /* Start inner kernel loop */
 737         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 738         {
 739
 740             /* Get j neighbor index, and coordinate index */
 741             jnrA             = jjnr[jidx];
 742             jnrB             = jjnr[jidx+1];
 743             jnrC             = jjnr[jidx+2];
 744             jnrD             = jjnr[jidx+3];
 745             j_coord_offsetA  = DIM*jnrA;
 746             j_coord_offsetB  = DIM*jnrB;
 747             j_coord_offsetC  = DIM*jnrC;
 748             j_coord_offsetD  = DIM*jnrD;
 749
 750             /* load j atom coordinates */
 751             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 752                                               x+j_coord_offsetC,x+j_coord_offsetD,
 753                                               &jx0,&jy0,&jz0);
 754
 755             /* Calculate displacement vector */
 756             dx00             = _mm_sub_ps(ix0,jx0);
 757             dy00             = _mm_sub_ps(iy0,jy0);
 758             dz00             = _mm_sub_ps(iz0,jz0);
 759             dx10             = _mm_sub_ps(ix1,jx0);
 760             dy10             = _mm_sub_ps(iy1,jy0);
 761             dz10             = _mm_sub_ps(iz1,jz0);
 762             dx20             = _mm_sub_ps(ix2,jx0);
 763             dy20             = _mm_sub_ps(iy2,jy0);
 764             dz20             = _mm_sub_ps(iz2,jz0);
 765
 766             /* Calculate squared distance and things based on it */
 767             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 768             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 769             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 770
 771             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 772             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 773             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 774
 775             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 776             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 777             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 778
 779             /* Load parameters for j particles */
 780             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 781                                                               charge+jnrC+0,charge+jnrD+0);
 782             vdwjidx0A        = 2*vdwtype[jnrA+0];
 783             vdwjidx0B        = 2*vdwtype[jnrB+0];
 784             vdwjidx0C        = 2*vdwtype[jnrC+0];
 785             vdwjidx0D        = 2*vdwtype[jnrD+0];
 786
 787             fjx0             = _mm_setzero_ps();
 788             fjy0             = _mm_setzero_ps();
 789             fjz0             = _mm_setzero_ps();
 790
 791             /**************************
 792              * CALCULATE INTERACTIONS *
 793              **************************/
 794
 795             r00              = _mm_mul_ps(rsq00,rinv00);
 796
 797             /* Compute parameters for interactions between i and j atoms */
 798             qq00             = _mm_mul_ps(iq0,jq0);
 799             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 800                                          vdwparam+vdwioffset0+vdwjidx0B,
 801                                          vdwparam+vdwioffset0+vdwjidx0C,
 802                                          vdwparam+vdwioffset0+vdwjidx0D,
 803                                          &c6_00,&c12_00);
 804
 805             /* EWALD ELECTROSTATICS */
 806
 807             /* Analytical PME correction */
 808             zeta2            = _mm_mul_ps(beta2,rsq00);
 809             rinv3            = _mm_mul_ps(rinvsq00,rinv00);
 810             pmecorrF         = gmx_mm_pmecorrF_ps(zeta2);
 811             felec            = _mm_macc_ps(pmecorrF,beta3,rinv3);
 812             felec            = _mm_mul_ps(qq00,felec);
 813
 814             /* LENNARD-JONES DISPERSION/REPULSION */
 815
 816             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 817             fvdw             = _mm_mul_ps(_mm_msub_ps(c12_00,rinvsix,c6_00),_mm_mul_ps(rinvsix,rinvsq00));
 818
 819             fscal            = _mm_add_ps(felec,fvdw);
 820
 821              /* Update vectorial force */
 822             fix0             = _mm_macc_ps(dx00,fscal,fix0);
 823             fiy0             = _mm_macc_ps(dy00,fscal,fiy0);
 824             fiz0             = _mm_macc_ps(dz00,fscal,fiz0);
 825
 826             fjx0             = _mm_macc_ps(dx00,fscal,fjx0);
 827             fjy0             = _mm_macc_ps(dy00,fscal,fjy0);
 828             fjz0             = _mm_macc_ps(dz00,fscal,fjz0);
 829
 830             /**************************
 831              * CALCULATE INTERACTIONS *
 832              **************************/
 833
 834             r10              = _mm_mul_ps(rsq10,rinv10);
 835
 836             /* Compute parameters for interactions between i and j atoms */
 837             qq10             = _mm_mul_ps(iq1,jq0);
 838
 839             /* EWALD ELECTROSTATICS */
 840
 841             /* Analytical PME correction */
 842             zeta2            = _mm_mul_ps(beta2,rsq10);
 843             rinv3            = _mm_mul_ps(rinvsq10,rinv10);
 844             pmecorrF         = gmx_mm_pmecorrF_ps(zeta2);
 845             felec            = _mm_macc_ps(pmecorrF,beta3,rinv3);
 846             felec            = _mm_mul_ps(qq10,felec);
 847
 848             fscal            = felec;
 849
 850              /* Update vectorial force */
 851             fix1             = _mm_macc_ps(dx10,fscal,fix1);
 852             fiy1             = _mm_macc_ps(dy10,fscal,fiy1);
 853             fiz1             = _mm_macc_ps(dz10,fscal,fiz1);
 854
 855             fjx0             = _mm_macc_ps(dx10,fscal,fjx0);
 856             fjy0             = _mm_macc_ps(dy10,fscal,fjy0);
 857             fjz0             = _mm_macc_ps(dz10,fscal,fjz0);
 858
 859             /**************************
 860              * CALCULATE INTERACTIONS *
 861              **************************/
 862
 863             r20              = _mm_mul_ps(rsq20,rinv20);
 864
 865             /* Compute parameters for interactions between i and j atoms */
 866             qq20             = _mm_mul_ps(iq2,jq0);
 867
 868             /* EWALD ELECTROSTATICS */
 869
 870             /* Analytical PME correction */
 871             zeta2            = _mm_mul_ps(beta2,rsq20);
 872             rinv3            = _mm_mul_ps(rinvsq20,rinv20);
 873             pmecorrF         = gmx_mm_pmecorrF_ps(zeta2);
 874             felec            = _mm_macc_ps(pmecorrF,beta3,rinv3);
 875             felec            = _mm_mul_ps(qq20,felec);
 876
 877             fscal            = felec;
 878
 879              /* Update vectorial force */
 880             fix2             = _mm_macc_ps(dx20,fscal,fix2);
 881             fiy2             = _mm_macc_ps(dy20,fscal,fiy2);
 882             fiz2             = _mm_macc_ps(dz20,fscal,fiz2);
 883
 884             fjx0             = _mm_macc_ps(dx20,fscal,fjx0);
 885             fjy0             = _mm_macc_ps(dy20,fscal,fjy0);
 886             fjz0             = _mm_macc_ps(dz20,fscal,fjz0);
 887
 888             fjptrA             = f+j_coord_offsetA;
 889             fjptrB             = f+j_coord_offsetB;
 890             fjptrC             = f+j_coord_offsetC;
 891             fjptrD             = f+j_coord_offsetD;
 892
 893             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 894
 895             /* Inner loop uses 91 flops */
 896         }
 897
 898         if(jidx<j_index_end)
 899         {
 900
 901             /* Get j neighbor index, and coordinate index */
 902             jnrlistA         = jjnr[jidx];
 903             jnrlistB         = jjnr[jidx+1];
 904             jnrlistC         = jjnr[jidx+2];
 905             jnrlistD         = jjnr[jidx+3];
 906             /* Sign of each element will be negative for non-real atoms.
 907              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 908              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 909              */
 910             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 911             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 912             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 913             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 914             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 915             j_coord_offsetA  = DIM*jnrA;
 916             j_coord_offsetB  = DIM*jnrB;
 917             j_coord_offsetC  = DIM*jnrC;
 918             j_coord_offsetD  = DIM*jnrD;
 919
 920             /* load j atom coordinates */
 921             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 922                                               x+j_coord_offsetC,x+j_coord_offsetD,
 923                                               &jx0,&jy0,&jz0);
 924
 925             /* Calculate displacement vector */
 926             dx00             = _mm_sub_ps(ix0,jx0);
 927             dy00             = _mm_sub_ps(iy0,jy0);
 928             dz00             = _mm_sub_ps(iz0,jz0);
 929             dx10             = _mm_sub_ps(ix1,jx0);
 930             dy10             = _mm_sub_ps(iy1,jy0);
 931             dz10             = _mm_sub_ps(iz1,jz0);
 932             dx20             = _mm_sub_ps(ix2,jx0);
 933             dy20             = _mm_sub_ps(iy2,jy0);
 934             dz20             = _mm_sub_ps(iz2,jz0);
 935
 936             /* Calculate squared distance and things based on it */
 937             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 938             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 939             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 940
 941             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 942             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 943             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 944
 945             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 946             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 947             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 948
 949             /* Load parameters for j particles */
 950             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 951                                                               charge+jnrC+0,charge+jnrD+0);
 952             vdwjidx0A        = 2*vdwtype[jnrA+0];
 953             vdwjidx0B        = 2*vdwtype[jnrB+0];
 954             vdwjidx0C        = 2*vdwtype[jnrC+0];
 955             vdwjidx0D        = 2*vdwtype[jnrD+0];
 956
 957             fjx0             = _mm_setzero_ps();
 958             fjy0             = _mm_setzero_ps();
 959             fjz0             = _mm_setzero_ps();
 960
 961             /**************************
 962              * CALCULATE INTERACTIONS *
 963              **************************/
 964
 965             r00              = _mm_mul_ps(rsq00,rinv00);
 966             r00              = _mm_andnot_ps(dummy_mask,r00);
 967
 968             /* Compute parameters for interactions between i and j atoms */
 969             qq00             = _mm_mul_ps(iq0,jq0);
 970             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 971                                          vdwparam+vdwioffset0+vdwjidx0B,
 972                                          vdwparam+vdwioffset0+vdwjidx0C,
 973                                          vdwparam+vdwioffset0+vdwjidx0D,
 974                                          &c6_00,&c12_00);
 975
 976             /* EWALD ELECTROSTATICS */
 977
 978             /* Analytical PME correction */
 979             zeta2            = _mm_mul_ps(beta2,rsq00);
 980             rinv3            = _mm_mul_ps(rinvsq00,rinv00);
 981             pmecorrF         = gmx_mm_pmecorrF_ps(zeta2);
 982             felec            = _mm_macc_ps(pmecorrF,beta3,rinv3);
 983             felec            = _mm_mul_ps(qq00,felec);
 984
 985             /* LENNARD-JONES DISPERSION/REPULSION */
 986
 987             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 988             fvdw             = _mm_mul_ps(_mm_msub_ps(c12_00,rinvsix,c6_00),_mm_mul_ps(rinvsix,rinvsq00));
 989
 990             fscal            = _mm_add_ps(felec,fvdw);
 991
 992             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 993
 994              /* Update vectorial force */
 995             fix0             = _mm_macc_ps(dx00,fscal,fix0);
 996             fiy0             = _mm_macc_ps(dy00,fscal,fiy0);
 997             fiz0             = _mm_macc_ps(dz00,fscal,fiz0);
 998
 999             fjx0             = _mm_macc_ps(dx00,fscal,fjx0);
1000             fjy0             = _mm_macc_ps(dy00,fscal,fjy0);
1001             fjz0             = _mm_macc_ps(dz00,fscal,fjz0);
1002
1003             /**************************
1004              * CALCULATE INTERACTIONS *
1005              **************************/
1006
1007             r10              = _mm_mul_ps(rsq10,rinv10);
1008             r10              = _mm_andnot_ps(dummy_mask,r10);
1009
1010             /* Compute parameters for interactions between i and j atoms */
1011             qq10             = _mm_mul_ps(iq1,jq0);
1012
1013             /* EWALD ELECTROSTATICS */
1014
1015             /* Analytical PME correction */
1016             zeta2            = _mm_mul_ps(beta2,rsq10);
1017             rinv3            = _mm_mul_ps(rinvsq10,rinv10);
1018             pmecorrF         = gmx_mm_pmecorrF_ps(zeta2);
1019             felec            = _mm_macc_ps(pmecorrF,beta3,rinv3);
1020             felec            = _mm_mul_ps(qq10,felec);
1021
1022             fscal            = felec;
1023
1024             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1025
1026              /* Update vectorial force */
1027             fix1             = _mm_macc_ps(dx10,fscal,fix1);
1028             fiy1             = _mm_macc_ps(dy10,fscal,fiy1);
1029             fiz1             = _mm_macc_ps(dz10,fscal,fiz1);
1030
1031             fjx0             = _mm_macc_ps(dx10,fscal,fjx0);
1032             fjy0             = _mm_macc_ps(dy10,fscal,fjy0);
1033             fjz0             = _mm_macc_ps(dz10,fscal,fjz0);
1034
1035             /**************************
1036              * CALCULATE INTERACTIONS *
1037              **************************/
1038
1039             r20              = _mm_mul_ps(rsq20,rinv20);
1040             r20              = _mm_andnot_ps(dummy_mask,r20);
1041
1042             /* Compute parameters for interactions between i and j atoms */
1043             qq20             = _mm_mul_ps(iq2,jq0);
1044
1045             /* EWALD ELECTROSTATICS */
1046
1047             /* Analytical PME correction */
1048             zeta2            = _mm_mul_ps(beta2,rsq20);
1049             rinv3            = _mm_mul_ps(rinvsq20,rinv20);
1050             pmecorrF         = gmx_mm_pmecorrF_ps(zeta2);
1051             felec            = _mm_macc_ps(pmecorrF,beta3,rinv3);
1052             felec            = _mm_mul_ps(qq20,felec);
1053
1054             fscal            = felec;
1055
1056             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1057
1058              /* Update vectorial force */
1059             fix2             = _mm_macc_ps(dx20,fscal,fix2);
1060             fiy2             = _mm_macc_ps(dy20,fscal,fiy2);
1061             fiz2             = _mm_macc_ps(dz20,fscal,fiz2);
1062
1063             fjx0             = _mm_macc_ps(dx20,fscal,fjx0);
1064             fjy0             = _mm_macc_ps(dy20,fscal,fjy0);
1065             fjz0             = _mm_macc_ps(dz20,fscal,fjz0);
1066
1067             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
1068             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
1069             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
1070             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
1071
1072             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
1073
1074             /* Inner loop uses 94 flops */
1075         }
1076
1077         /* End of innermost loop */
1078
1079         gmx_mm_update_iforce_3atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
1080                                               f+i_coord_offset,fshift+i_shift_offset);
1081
1082         /* Increment number of inner iterations */
1083         inneriter                  += j_index_end - j_index_start;
1084
1085         /* Outer loop uses 18 flops */
1086     }
1087
1088     /* Increment number of outer iterations */
1089     outeriter        += nri;
1090
1091     /* Update outer/inner flops */
1092
1093     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3_F,outeriter*18 + inneriter*94);
1094 }