src/gromacs/gmxlib/nonbonded/nb_kernel_avx_128_fma_single/nb_kernel_ElecEw_VdwNone_GeomW4P1_avx_128_fma_single.c

   1 /*
   2  * This file is part of the GROMACS molecular simulation package.
   3  *
   4  * Copyright (c) 2012,2013,2014, by the GROMACS development team, led by
   5  * Mark Abraham, David van der Spoel, Berk Hess, and Erik Lindahl,
   6  * and including many others, as listed in the AUTHORS file in the
   7  * top-level source directory and at http://www.gromacs.org.
   8  *
   9  * GROMACS is free software; you can redistribute it and/or
  10  * modify it under the terms of the GNU Lesser General Public License
  11  * as published by the Free Software Foundation; either version 2.1
  12  * of the License, or (at your option) any later version.
  13  *
  14  * GROMACS is distributed in the hope that it will be useful,
  15  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  16  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  17  * Lesser General Public License for more details.
  18  *
  19  * You should have received a copy of the GNU Lesser General Public
  20  * License along with GROMACS; if not, see
  21  * http://www.gnu.org/licenses, or write to the Free Software Foundation,
  22  * Inc., 51 Franklin Street, Fifth Floor, Boston, MA  02110-1301  USA.
  23  *
  24  * If you want to redistribute modifications to GROMACS, please
  25  * consider that scientific software is very special. Version
  26  * control is crucial - bugs must be traceable. We will be happy to
  27  * consider code for inclusion in the official distribution, but
  28  * derived work must not be called official GROMACS. Details are found
  29  * in the README & COPYING files - if they are missing, get the
  30  * official version at http://www.gromacs.org.
  31  *
  32  * To help us fund GROMACS development, we humbly ask that you cite
  33  * the research papers on the package. Check out http://www.gromacs.org.
  34  */
  35 /*
  36  * Note: this file was generated by the GROMACS avx_128_fma_single kernel generator.
  37  */
  38 #include "config.h"
  39
  40 #include <math.h>
  41
  42 #include "../nb_kernel.h"
  43 #include "gromacs/legacyheaders/types/simple.h"
  44 #include "gromacs/math/vec.h"
  45 #include "gromacs/legacyheaders/nrnb.h"
  46
  47 #include "gromacs/simd/math_x86_avx_128_fma_single.h"
  48 #include "kernelutil_x86_avx_128_fma_single.h"
  49
  50 /*
  51  * Gromacs nonbonded kernel:   nb_kernel_ElecEw_VdwNone_GeomW4P1_VF_avx_128_fma_single
  52  * Electrostatics interaction: Ewald
  53  * VdW interaction:            None
  54  * Geometry:                   Water4-Particle
  55  * Calculate force/pot:        PotentialAndForce
  56  */
  57 void
  58 nb_kernel_ElecEw_VdwNone_GeomW4P1_VF_avx_128_fma_single
  59                     (t_nblist                    * gmx_restrict       nlist,
  60                      rvec                        * gmx_restrict          xx,
  61                      rvec                        * gmx_restrict          ff,
  62                      t_forcerec                  * gmx_restrict          fr,
  63                      t_mdatoms                   * gmx_restrict     mdatoms,
  64                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
  65                      t_nrnb                      * gmx_restrict        nrnb)
  66 {
  67     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
  68      * just 0 for non-waters.
  69      * Suffixes A,B,C,D refer to j loop unrolling done with AVX_128, e.g. for the four different
  70      * jnr indices corresponding to data put in the four positions in the SIMD register.
  71      */
  72     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
  73     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
  74     int              jnrA,jnrB,jnrC,jnrD;
  75     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
  76     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
  77     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
  78     real             rcutoff_scalar;
  79     real             *shiftvec,*fshift,*x,*f;
  80     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
  81     real             scratch[4*DIM];
  82     __m128           fscal,rcutoff,rcutoff2,jidxall;
  83     int              vdwioffset1;
  84     __m128           ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
  85     int              vdwioffset2;
  86     __m128           ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
  87     int              vdwioffset3;
  88     __m128           ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
  89     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
  90     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  91     __m128           dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
  92     __m128           dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
  93     __m128           dx30,dy30,dz30,rsq30,rinv30,rinvsq30,r30,qq30,c6_30,c12_30;
  94     __m128           velec,felec,velecsum,facel,crf,krf,krf2;
  95     real             *charge;
  96     __m128i          ewitab;
  97     __m128           ewtabscale,eweps,twoeweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
  98     __m128           beta,beta2,beta3,zeta2,pmecorrF,pmecorrV,rinv3;
  99     real             *ewtab;
 100     __m128           dummy_mask,cutoff_mask;
 101     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
 102     __m128           one     = _mm_set1_ps(1.0);
 103     __m128           two     = _mm_set1_ps(2.0);
 104     x                = xx[0];
 105     f                = ff[0];
 106
 107     nri              = nlist->nri;
 108     iinr             = nlist->iinr;
 109     jindex           = nlist->jindex;
 110     jjnr             = nlist->jjnr;
 111     shiftidx         = nlist->shift;
 112     gid              = nlist->gid;
 113     shiftvec         = fr->shift_vec[0];
 114     fshift           = fr->fshift[0];
 115     facel            = _mm_set1_ps(fr->epsfac);
 116     charge           = mdatoms->chargeA;
 117
 118     sh_ewald         = _mm_set1_ps(fr->ic->sh_ewald);
 119     beta             = _mm_set1_ps(fr->ic->ewaldcoeff_q);
 120     beta2            = _mm_mul_ps(beta,beta);
 121     beta3            = _mm_mul_ps(beta,beta2);
 122     ewtab            = fr->ic->tabq_coul_FDV0;
 123     ewtabscale       = _mm_set1_ps(fr->ic->tabq_scale);
 124     ewtabhalfspace   = _mm_set1_ps(0.5/fr->ic->tabq_scale);
 125
 126     /* Setup water-specific parameters */
 127     inr              = nlist->iinr[0];
 128     iq1              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+1]));
 129     iq2              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+2]));
 130     iq3              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+3]));
 131
 132     /* Avoid stupid compiler warnings */
 133     jnrA = jnrB = jnrC = jnrD = 0;
 134     j_coord_offsetA = 0;
 135     j_coord_offsetB = 0;
 136     j_coord_offsetC = 0;
 137     j_coord_offsetD = 0;
 138
 139     outeriter        = 0;
 140     inneriter        = 0;
 141
 142     for(iidx=0;iidx<4*DIM;iidx++)
 143     {
 144         scratch[iidx] = 0.0;
 145     }
 146
 147     /* Start outer loop over neighborlists */
 148     for(iidx=0; iidx<nri; iidx++)
 149     {
 150         /* Load shift vector for this list */
 151         i_shift_offset   = DIM*shiftidx[iidx];
 152
 153         /* Load limits for loop over neighbors */
 154         j_index_start    = jindex[iidx];
 155         j_index_end      = jindex[iidx+1];
 156
 157         /* Get outer coordinate index */
 158         inr              = iinr[iidx];
 159         i_coord_offset   = DIM*inr;
 160
 161         /* Load i particle coords and add shift vector */
 162         gmx_mm_load_shift_and_3rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset+DIM,
 163                                                  &ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
 164
 165         fix1             = _mm_setzero_ps();
 166         fiy1             = _mm_setzero_ps();
 167         fiz1             = _mm_setzero_ps();
 168         fix2             = _mm_setzero_ps();
 169         fiy2             = _mm_setzero_ps();
 170         fiz2             = _mm_setzero_ps();
 171         fix3             = _mm_setzero_ps();
 172         fiy3             = _mm_setzero_ps();
 173         fiz3             = _mm_setzero_ps();
 174
 175         /* Reset potential sums */
 176         velecsum         = _mm_setzero_ps();
 177
 178         /* Start inner kernel loop */
 179         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 180         {
 181
 182             /* Get j neighbor index, and coordinate index */
 183             jnrA             = jjnr[jidx];
 184             jnrB             = jjnr[jidx+1];
 185             jnrC             = jjnr[jidx+2];
 186             jnrD             = jjnr[jidx+3];
 187             j_coord_offsetA  = DIM*jnrA;
 188             j_coord_offsetB  = DIM*jnrB;
 189             j_coord_offsetC  = DIM*jnrC;
 190             j_coord_offsetD  = DIM*jnrD;
 191
 192             /* load j atom coordinates */
 193             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 194                                               x+j_coord_offsetC,x+j_coord_offsetD,
 195                                               &jx0,&jy0,&jz0);
 196
 197             /* Calculate displacement vector */
 198             dx10             = _mm_sub_ps(ix1,jx0);
 199             dy10             = _mm_sub_ps(iy1,jy0);
 200             dz10             = _mm_sub_ps(iz1,jz0);
 201             dx20             = _mm_sub_ps(ix2,jx0);
 202             dy20             = _mm_sub_ps(iy2,jy0);
 203             dz20             = _mm_sub_ps(iz2,jz0);
 204             dx30             = _mm_sub_ps(ix3,jx0);
 205             dy30             = _mm_sub_ps(iy3,jy0);
 206             dz30             = _mm_sub_ps(iz3,jz0);
 207
 208             /* Calculate squared distance and things based on it */
 209             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 210             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 211             rsq30            = gmx_mm_calc_rsq_ps(dx30,dy30,dz30);
 212
 213             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 214             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 215             rinv30           = gmx_mm_invsqrt_ps(rsq30);
 216
 217             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 218             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 219             rinvsq30         = _mm_mul_ps(rinv30,rinv30);
 220
 221             /* Load parameters for j particles */
 222             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 223                                                               charge+jnrC+0,charge+jnrD+0);
 224
 225             fjx0             = _mm_setzero_ps();
 226             fjy0             = _mm_setzero_ps();
 227             fjz0             = _mm_setzero_ps();
 228
 229             /**************************
 230              * CALCULATE INTERACTIONS *
 231              **************************/
 232
 233             r10              = _mm_mul_ps(rsq10,rinv10);
 234
 235             /* Compute parameters for interactions between i and j atoms */
 236             qq10             = _mm_mul_ps(iq1,jq0);
 237
 238             /* EWALD ELECTROSTATICS */
 239
 240             /* Analytical PME correction */
 241             zeta2            = _mm_mul_ps(beta2,rsq10);
 242             rinv3            = _mm_mul_ps(rinvsq10,rinv10);
 243             pmecorrF         = gmx_mm_pmecorrF_ps(zeta2);
 244             felec            = _mm_macc_ps(pmecorrF,beta3,rinv3);
 245             felec            = _mm_mul_ps(qq10,felec);
 246             pmecorrV         = gmx_mm_pmecorrV_ps(zeta2);
 247             velec            = _mm_nmacc_ps(pmecorrV,beta,rinv10);
 248             velec            = _mm_mul_ps(qq10,velec);
 249
 250             /* Update potential sum for this i atom from the interaction with this j atom. */
 251             velecsum         = _mm_add_ps(velecsum,velec);
 252
 253             fscal            = felec;
 254
 255              /* Update vectorial force */
 256             fix1             = _mm_macc_ps(dx10,fscal,fix1);
 257             fiy1             = _mm_macc_ps(dy10,fscal,fiy1);
 258             fiz1             = _mm_macc_ps(dz10,fscal,fiz1);
 259
 260             fjx0             = _mm_macc_ps(dx10,fscal,fjx0);
 261             fjy0             = _mm_macc_ps(dy10,fscal,fjy0);
 262             fjz0             = _mm_macc_ps(dz10,fscal,fjz0);
 263
 264             /**************************
 265              * CALCULATE INTERACTIONS *
 266              **************************/
 267
 268             r20              = _mm_mul_ps(rsq20,rinv20);
 269
 270             /* Compute parameters for interactions between i and j atoms */
 271             qq20             = _mm_mul_ps(iq2,jq0);
 272
 273             /* EWALD ELECTROSTATICS */
 274
 275             /* Analytical PME correction */
 276             zeta2            = _mm_mul_ps(beta2,rsq20);
 277             rinv3            = _mm_mul_ps(rinvsq20,rinv20);
 278             pmecorrF         = gmx_mm_pmecorrF_ps(zeta2);
 279             felec            = _mm_macc_ps(pmecorrF,beta3,rinv3);
 280             felec            = _mm_mul_ps(qq20,felec);
 281             pmecorrV         = gmx_mm_pmecorrV_ps(zeta2);
 282             velec            = _mm_nmacc_ps(pmecorrV,beta,rinv20);
 283             velec            = _mm_mul_ps(qq20,velec);
 284
 285             /* Update potential sum for this i atom from the interaction with this j atom. */
 286             velecsum         = _mm_add_ps(velecsum,velec);
 287
 288             fscal            = felec;
 289
 290              /* Update vectorial force */
 291             fix2             = _mm_macc_ps(dx20,fscal,fix2);
 292             fiy2             = _mm_macc_ps(dy20,fscal,fiy2);
 293             fiz2             = _mm_macc_ps(dz20,fscal,fiz2);
 294
 295             fjx0             = _mm_macc_ps(dx20,fscal,fjx0);
 296             fjy0             = _mm_macc_ps(dy20,fscal,fjy0);
 297             fjz0             = _mm_macc_ps(dz20,fscal,fjz0);
 298
 299             /**************************
 300              * CALCULATE INTERACTIONS *
 301              **************************/
 302
 303             r30              = _mm_mul_ps(rsq30,rinv30);
 304
 305             /* Compute parameters for interactions between i and j atoms */
 306             qq30             = _mm_mul_ps(iq3,jq0);
 307
 308             /* EWALD ELECTROSTATICS */
 309
 310             /* Analytical PME correction */
 311             zeta2            = _mm_mul_ps(beta2,rsq30);
 312             rinv3            = _mm_mul_ps(rinvsq30,rinv30);
 313             pmecorrF         = gmx_mm_pmecorrF_ps(zeta2);
 314             felec            = _mm_macc_ps(pmecorrF,beta3,rinv3);
 315             felec            = _mm_mul_ps(qq30,felec);
 316             pmecorrV         = gmx_mm_pmecorrV_ps(zeta2);
 317             velec            = _mm_nmacc_ps(pmecorrV,beta,rinv30);
 318             velec            = _mm_mul_ps(qq30,velec);
 319
 320             /* Update potential sum for this i atom from the interaction with this j atom. */
 321             velecsum         = _mm_add_ps(velecsum,velec);
 322
 323             fscal            = felec;
 324
 325              /* Update vectorial force */
 326             fix3             = _mm_macc_ps(dx30,fscal,fix3);
 327             fiy3             = _mm_macc_ps(dy30,fscal,fiy3);
 328             fiz3             = _mm_macc_ps(dz30,fscal,fiz3);
 329
 330             fjx0             = _mm_macc_ps(dx30,fscal,fjx0);
 331             fjy0             = _mm_macc_ps(dy30,fscal,fjy0);
 332             fjz0             = _mm_macc_ps(dz30,fscal,fjz0);
 333
 334             fjptrA             = f+j_coord_offsetA;
 335             fjptrB             = f+j_coord_offsetB;
 336             fjptrC             = f+j_coord_offsetC;
 337             fjptrD             = f+j_coord_offsetD;
 338
 339             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 340
 341             /* Inner loop uses 87 flops */
 342         }
 343
 344         if(jidx<j_index_end)
 345         {
 346
 347             /* Get j neighbor index, and coordinate index */
 348             jnrlistA         = jjnr[jidx];
 349             jnrlistB         = jjnr[jidx+1];
 350             jnrlistC         = jjnr[jidx+2];
 351             jnrlistD         = jjnr[jidx+3];
 352             /* Sign of each element will be negative for non-real atoms.
 353              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 354              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 355              */
 356             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 357             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 358             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 359             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 360             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 361             j_coord_offsetA  = DIM*jnrA;
 362             j_coord_offsetB  = DIM*jnrB;
 363             j_coord_offsetC  = DIM*jnrC;
 364             j_coord_offsetD  = DIM*jnrD;
 365
 366             /* load j atom coordinates */
 367             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 368                                               x+j_coord_offsetC,x+j_coord_offsetD,
 369                                               &jx0,&jy0,&jz0);
 370
 371             /* Calculate displacement vector */
 372             dx10             = _mm_sub_ps(ix1,jx0);
 373             dy10             = _mm_sub_ps(iy1,jy0);
 374             dz10             = _mm_sub_ps(iz1,jz0);
 375             dx20             = _mm_sub_ps(ix2,jx0);
 376             dy20             = _mm_sub_ps(iy2,jy0);
 377             dz20             = _mm_sub_ps(iz2,jz0);
 378             dx30             = _mm_sub_ps(ix3,jx0);
 379             dy30             = _mm_sub_ps(iy3,jy0);
 380             dz30             = _mm_sub_ps(iz3,jz0);
 381
 382             /* Calculate squared distance and things based on it */
 383             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 384             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 385             rsq30            = gmx_mm_calc_rsq_ps(dx30,dy30,dz30);
 386
 387             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 388             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 389             rinv30           = gmx_mm_invsqrt_ps(rsq30);
 390
 391             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 392             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 393             rinvsq30         = _mm_mul_ps(rinv30,rinv30);
 394
 395             /* Load parameters for j particles */
 396             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 397                                                               charge+jnrC+0,charge+jnrD+0);
 398
 399             fjx0             = _mm_setzero_ps();
 400             fjy0             = _mm_setzero_ps();
 401             fjz0             = _mm_setzero_ps();
 402
 403             /**************************
 404              * CALCULATE INTERACTIONS *
 405              **************************/
 406
 407             r10              = _mm_mul_ps(rsq10,rinv10);
 408             r10              = _mm_andnot_ps(dummy_mask,r10);
 409
 410             /* Compute parameters for interactions between i and j atoms */
 411             qq10             = _mm_mul_ps(iq1,jq0);
 412
 413             /* EWALD ELECTROSTATICS */
 414
 415             /* Analytical PME correction */
 416             zeta2            = _mm_mul_ps(beta2,rsq10);
 417             rinv3            = _mm_mul_ps(rinvsq10,rinv10);
 418             pmecorrF         = gmx_mm_pmecorrF_ps(zeta2);
 419             felec            = _mm_macc_ps(pmecorrF,beta3,rinv3);
 420             felec            = _mm_mul_ps(qq10,felec);
 421             pmecorrV         = gmx_mm_pmecorrV_ps(zeta2);
 422             velec            = _mm_nmacc_ps(pmecorrV,beta,rinv10);
 423             velec            = _mm_mul_ps(qq10,velec);
 424
 425             /* Update potential sum for this i atom from the interaction with this j atom. */
 426             velec            = _mm_andnot_ps(dummy_mask,velec);
 427             velecsum         = _mm_add_ps(velecsum,velec);
 428
 429             fscal            = felec;
 430
 431             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 432
 433              /* Update vectorial force */
 434             fix1             = _mm_macc_ps(dx10,fscal,fix1);
 435             fiy1             = _mm_macc_ps(dy10,fscal,fiy1);
 436             fiz1             = _mm_macc_ps(dz10,fscal,fiz1);
 437
 438             fjx0             = _mm_macc_ps(dx10,fscal,fjx0);
 439             fjy0             = _mm_macc_ps(dy10,fscal,fjy0);
 440             fjz0             = _mm_macc_ps(dz10,fscal,fjz0);
 441
 442             /**************************
 443              * CALCULATE INTERACTIONS *
 444              **************************/
 445
 446             r20              = _mm_mul_ps(rsq20,rinv20);
 447             r20              = _mm_andnot_ps(dummy_mask,r20);
 448
 449             /* Compute parameters for interactions between i and j atoms */
 450             qq20             = _mm_mul_ps(iq2,jq0);
 451
 452             /* EWALD ELECTROSTATICS */
 453
 454             /* Analytical PME correction */
 455             zeta2            = _mm_mul_ps(beta2,rsq20);
 456             rinv3            = _mm_mul_ps(rinvsq20,rinv20);
 457             pmecorrF         = gmx_mm_pmecorrF_ps(zeta2);
 458             felec            = _mm_macc_ps(pmecorrF,beta3,rinv3);
 459             felec            = _mm_mul_ps(qq20,felec);
 460             pmecorrV         = gmx_mm_pmecorrV_ps(zeta2);
 461             velec            = _mm_nmacc_ps(pmecorrV,beta,rinv20);
 462             velec            = _mm_mul_ps(qq20,velec);
 463
 464             /* Update potential sum for this i atom from the interaction with this j atom. */
 465             velec            = _mm_andnot_ps(dummy_mask,velec);
 466             velecsum         = _mm_add_ps(velecsum,velec);
 467
 468             fscal            = felec;
 469
 470             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 471
 472              /* Update vectorial force */
 473             fix2             = _mm_macc_ps(dx20,fscal,fix2);
 474             fiy2             = _mm_macc_ps(dy20,fscal,fiy2);
 475             fiz2             = _mm_macc_ps(dz20,fscal,fiz2);
 476
 477             fjx0             = _mm_macc_ps(dx20,fscal,fjx0);
 478             fjy0             = _mm_macc_ps(dy20,fscal,fjy0);
 479             fjz0             = _mm_macc_ps(dz20,fscal,fjz0);
 480
 481             /**************************
 482              * CALCULATE INTERACTIONS *
 483              **************************/
 484
 485             r30              = _mm_mul_ps(rsq30,rinv30);
 486             r30              = _mm_andnot_ps(dummy_mask,r30);
 487
 488             /* Compute parameters for interactions between i and j atoms */
 489             qq30             = _mm_mul_ps(iq3,jq0);
 490
 491             /* EWALD ELECTROSTATICS */
 492
 493             /* Analytical PME correction */
 494             zeta2            = _mm_mul_ps(beta2,rsq30);
 495             rinv3            = _mm_mul_ps(rinvsq30,rinv30);
 496             pmecorrF         = gmx_mm_pmecorrF_ps(zeta2);
 497             felec            = _mm_macc_ps(pmecorrF,beta3,rinv3);
 498             felec            = _mm_mul_ps(qq30,felec);
 499             pmecorrV         = gmx_mm_pmecorrV_ps(zeta2);
 500             velec            = _mm_nmacc_ps(pmecorrV,beta,rinv30);
 501             velec            = _mm_mul_ps(qq30,velec);
 502
 503             /* Update potential sum for this i atom from the interaction with this j atom. */
 504             velec            = _mm_andnot_ps(dummy_mask,velec);
 505             velecsum         = _mm_add_ps(velecsum,velec);
 506
 507             fscal            = felec;
 508
 509             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 510
 511              /* Update vectorial force */
 512             fix3             = _mm_macc_ps(dx30,fscal,fix3);
 513             fiy3             = _mm_macc_ps(dy30,fscal,fiy3);
 514             fiz3             = _mm_macc_ps(dz30,fscal,fiz3);
 515
 516             fjx0             = _mm_macc_ps(dx30,fscal,fjx0);
 517             fjy0             = _mm_macc_ps(dy30,fscal,fjy0);
 518             fjz0             = _mm_macc_ps(dz30,fscal,fjz0);
 519
 520             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 521             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 522             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 523             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 524
 525             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 526
 527             /* Inner loop uses 90 flops */
 528         }
 529
 530         /* End of innermost loop */
 531
 532         gmx_mm_update_iforce_3atom_swizzle_ps(fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
 533                                               f+i_coord_offset+DIM,fshift+i_shift_offset);
 534
 535         ggid                        = gid[iidx];
 536         /* Update potential energies */
 537         gmx_mm_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid);
 538
 539         /* Increment number of inner iterations */
 540         inneriter                  += j_index_end - j_index_start;
 541
 542         /* Outer loop uses 19 flops */
 543     }
 544
 545     /* Increment number of outer iterations */
 546     outeriter        += nri;
 547
 548     /* Update outer/inner flops */
 549
 550     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_W4_VF,outeriter*19 + inneriter*90);
 551 }
 552 /*
 553  * Gromacs nonbonded kernel:   nb_kernel_ElecEw_VdwNone_GeomW4P1_F_avx_128_fma_single
 554  * Electrostatics interaction: Ewald
 555  * VdW interaction:            None
 556  * Geometry:                   Water4-Particle
 557  * Calculate force/pot:        Force
 558  */
 559 void
 560 nb_kernel_ElecEw_VdwNone_GeomW4P1_F_avx_128_fma_single
 561                     (t_nblist                    * gmx_restrict       nlist,
 562                      rvec                        * gmx_restrict          xx,
 563                      rvec                        * gmx_restrict          ff,
 564                      t_forcerec                  * gmx_restrict          fr,
 565                      t_mdatoms                   * gmx_restrict     mdatoms,
 566                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 567                      t_nrnb                      * gmx_restrict        nrnb)
 568 {
 569     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 570      * just 0 for non-waters.
 571      * Suffixes A,B,C,D refer to j loop unrolling done with AVX_128, e.g. for the four different
 572      * jnr indices corresponding to data put in the four positions in the SIMD register.
 573      */
 574     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 575     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 576     int              jnrA,jnrB,jnrC,jnrD;
 577     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
 578     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 579     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 580     real             rcutoff_scalar;
 581     real             *shiftvec,*fshift,*x,*f;
 582     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
 583     real             scratch[4*DIM];
 584     __m128           fscal,rcutoff,rcutoff2,jidxall;
 585     int              vdwioffset1;
 586     __m128           ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
 587     int              vdwioffset2;
 588     __m128           ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
 589     int              vdwioffset3;
 590     __m128           ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
 591     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
 592     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 593     __m128           dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
 594     __m128           dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
 595     __m128           dx30,dy30,dz30,rsq30,rinv30,rinvsq30,r30,qq30,c6_30,c12_30;
 596     __m128           velec,felec,velecsum,facel,crf,krf,krf2;
 597     real             *charge;
 598     __m128i          ewitab;
 599     __m128           ewtabscale,eweps,twoeweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
 600     __m128           beta,beta2,beta3,zeta2,pmecorrF,pmecorrV,rinv3;
 601     real             *ewtab;
 602     __m128           dummy_mask,cutoff_mask;
 603     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
 604     __m128           one     = _mm_set1_ps(1.0);
 605     __m128           two     = _mm_set1_ps(2.0);
 606     x                = xx[0];
 607     f                = ff[0];
 608
 609     nri              = nlist->nri;
 610     iinr             = nlist->iinr;
 611     jindex           = nlist->jindex;
 612     jjnr             = nlist->jjnr;
 613     shiftidx         = nlist->shift;
 614     gid              = nlist->gid;
 615     shiftvec         = fr->shift_vec[0];
 616     fshift           = fr->fshift[0];
 617     facel            = _mm_set1_ps(fr->epsfac);
 618     charge           = mdatoms->chargeA;
 619
 620     sh_ewald         = _mm_set1_ps(fr->ic->sh_ewald);
 621     beta             = _mm_set1_ps(fr->ic->ewaldcoeff_q);
 622     beta2            = _mm_mul_ps(beta,beta);
 623     beta3            = _mm_mul_ps(beta,beta2);
 624     ewtab            = fr->ic->tabq_coul_F;
 625     ewtabscale       = _mm_set1_ps(fr->ic->tabq_scale);
 626     ewtabhalfspace   = _mm_set1_ps(0.5/fr->ic->tabq_scale);
 627
 628     /* Setup water-specific parameters */
 629     inr              = nlist->iinr[0];
 630     iq1              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+1]));
 631     iq2              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+2]));
 632     iq3              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+3]));
 633
 634     /* Avoid stupid compiler warnings */
 635     jnrA = jnrB = jnrC = jnrD = 0;
 636     j_coord_offsetA = 0;
 637     j_coord_offsetB = 0;
 638     j_coord_offsetC = 0;
 639     j_coord_offsetD = 0;
 640
 641     outeriter        = 0;
 642     inneriter        = 0;
 643
 644     for(iidx=0;iidx<4*DIM;iidx++)
 645     {
 646         scratch[iidx] = 0.0;
 647     }
 648
 649     /* Start outer loop over neighborlists */
 650     for(iidx=0; iidx<nri; iidx++)
 651     {
 652         /* Load shift vector for this list */
 653         i_shift_offset   = DIM*shiftidx[iidx];
 654
 655         /* Load limits for loop over neighbors */
 656         j_index_start    = jindex[iidx];
 657         j_index_end      = jindex[iidx+1];
 658
 659         /* Get outer coordinate index */
 660         inr              = iinr[iidx];
 661         i_coord_offset   = DIM*inr;
 662
 663         /* Load i particle coords and add shift vector */
 664         gmx_mm_load_shift_and_3rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset+DIM,
 665                                                  &ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
 666
 667         fix1             = _mm_setzero_ps();
 668         fiy1             = _mm_setzero_ps();
 669         fiz1             = _mm_setzero_ps();
 670         fix2             = _mm_setzero_ps();
 671         fiy2             = _mm_setzero_ps();
 672         fiz2             = _mm_setzero_ps();
 673         fix3             = _mm_setzero_ps();
 674         fiy3             = _mm_setzero_ps();
 675         fiz3             = _mm_setzero_ps();
 676
 677         /* Start inner kernel loop */
 678         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 679         {
 680
 681             /* Get j neighbor index, and coordinate index */
 682             jnrA             = jjnr[jidx];
 683             jnrB             = jjnr[jidx+1];
 684             jnrC             = jjnr[jidx+2];
 685             jnrD             = jjnr[jidx+3];
 686             j_coord_offsetA  = DIM*jnrA;
 687             j_coord_offsetB  = DIM*jnrB;
 688             j_coord_offsetC  = DIM*jnrC;
 689             j_coord_offsetD  = DIM*jnrD;
 690
 691             /* load j atom coordinates */
 692             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 693                                               x+j_coord_offsetC,x+j_coord_offsetD,
 694                                               &jx0,&jy0,&jz0);
 695
 696             /* Calculate displacement vector */
 697             dx10             = _mm_sub_ps(ix1,jx0);
 698             dy10             = _mm_sub_ps(iy1,jy0);
 699             dz10             = _mm_sub_ps(iz1,jz0);
 700             dx20             = _mm_sub_ps(ix2,jx0);
 701             dy20             = _mm_sub_ps(iy2,jy0);
 702             dz20             = _mm_sub_ps(iz2,jz0);
 703             dx30             = _mm_sub_ps(ix3,jx0);
 704             dy30             = _mm_sub_ps(iy3,jy0);
 705             dz30             = _mm_sub_ps(iz3,jz0);
 706
 707             /* Calculate squared distance and things based on it */
 708             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 709             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 710             rsq30            = gmx_mm_calc_rsq_ps(dx30,dy30,dz30);
 711
 712             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 713             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 714             rinv30           = gmx_mm_invsqrt_ps(rsq30);
 715
 716             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 717             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 718             rinvsq30         = _mm_mul_ps(rinv30,rinv30);
 719
 720             /* Load parameters for j particles */
 721             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 722                                                               charge+jnrC+0,charge+jnrD+0);
 723
 724             fjx0             = _mm_setzero_ps();
 725             fjy0             = _mm_setzero_ps();
 726             fjz0             = _mm_setzero_ps();
 727
 728             /**************************
 729              * CALCULATE INTERACTIONS *
 730              **************************/
 731
 732             r10              = _mm_mul_ps(rsq10,rinv10);
 733
 734             /* Compute parameters for interactions between i and j atoms */
 735             qq10             = _mm_mul_ps(iq1,jq0);
 736
 737             /* EWALD ELECTROSTATICS */
 738
 739             /* Analytical PME correction */
 740             zeta2            = _mm_mul_ps(beta2,rsq10);
 741             rinv3            = _mm_mul_ps(rinvsq10,rinv10);
 742             pmecorrF         = gmx_mm_pmecorrF_ps(zeta2);
 743             felec            = _mm_macc_ps(pmecorrF,beta3,rinv3);
 744             felec            = _mm_mul_ps(qq10,felec);
 745
 746             fscal            = felec;
 747
 748              /* Update vectorial force */
 749             fix1             = _mm_macc_ps(dx10,fscal,fix1);
 750             fiy1             = _mm_macc_ps(dy10,fscal,fiy1);
 751             fiz1             = _mm_macc_ps(dz10,fscal,fiz1);
 752
 753             fjx0             = _mm_macc_ps(dx10,fscal,fjx0);
 754             fjy0             = _mm_macc_ps(dy10,fscal,fjy0);
 755             fjz0             = _mm_macc_ps(dz10,fscal,fjz0);
 756
 757             /**************************
 758              * CALCULATE INTERACTIONS *
 759              **************************/
 760
 761             r20              = _mm_mul_ps(rsq20,rinv20);
 762
 763             /* Compute parameters for interactions between i and j atoms */
 764             qq20             = _mm_mul_ps(iq2,jq0);
 765
 766             /* EWALD ELECTROSTATICS */
 767
 768             /* Analytical PME correction */
 769             zeta2            = _mm_mul_ps(beta2,rsq20);
 770             rinv3            = _mm_mul_ps(rinvsq20,rinv20);
 771             pmecorrF         = gmx_mm_pmecorrF_ps(zeta2);
 772             felec            = _mm_macc_ps(pmecorrF,beta3,rinv3);
 773             felec            = _mm_mul_ps(qq20,felec);
 774
 775             fscal            = felec;
 776
 777              /* Update vectorial force */
 778             fix2             = _mm_macc_ps(dx20,fscal,fix2);
 779             fiy2             = _mm_macc_ps(dy20,fscal,fiy2);
 780             fiz2             = _mm_macc_ps(dz20,fscal,fiz2);
 781
 782             fjx0             = _mm_macc_ps(dx20,fscal,fjx0);
 783             fjy0             = _mm_macc_ps(dy20,fscal,fjy0);
 784             fjz0             = _mm_macc_ps(dz20,fscal,fjz0);
 785
 786             /**************************
 787              * CALCULATE INTERACTIONS *
 788              **************************/
 789
 790             r30              = _mm_mul_ps(rsq30,rinv30);
 791
 792             /* Compute parameters for interactions between i and j atoms */
 793             qq30             = _mm_mul_ps(iq3,jq0);
 794
 795             /* EWALD ELECTROSTATICS */
 796
 797             /* Analytical PME correction */
 798             zeta2            = _mm_mul_ps(beta2,rsq30);
 799             rinv3            = _mm_mul_ps(rinvsq30,rinv30);
 800             pmecorrF         = gmx_mm_pmecorrF_ps(zeta2);
 801             felec            = _mm_macc_ps(pmecorrF,beta3,rinv3);
 802             felec            = _mm_mul_ps(qq30,felec);
 803
 804             fscal            = felec;
 805
 806              /* Update vectorial force */
 807             fix3             = _mm_macc_ps(dx30,fscal,fix3);
 808             fiy3             = _mm_macc_ps(dy30,fscal,fiy3);
 809             fiz3             = _mm_macc_ps(dz30,fscal,fiz3);
 810
 811             fjx0             = _mm_macc_ps(dx30,fscal,fjx0);
 812             fjy0             = _mm_macc_ps(dy30,fscal,fjy0);
 813             fjz0             = _mm_macc_ps(dz30,fscal,fjz0);
 814
 815             fjptrA             = f+j_coord_offsetA;
 816             fjptrB             = f+j_coord_offsetB;
 817             fjptrC             = f+j_coord_offsetC;
 818             fjptrD             = f+j_coord_offsetD;
 819
 820             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 821
 822             /* Inner loop uses 84 flops */
 823         }
 824
 825         if(jidx<j_index_end)
 826         {
 827
 828             /* Get j neighbor index, and coordinate index */
 829             jnrlistA         = jjnr[jidx];
 830             jnrlistB         = jjnr[jidx+1];
 831             jnrlistC         = jjnr[jidx+2];
 832             jnrlistD         = jjnr[jidx+3];
 833             /* Sign of each element will be negative for non-real atoms.
 834              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 835              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 836              */
 837             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 838             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 839             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 840             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 841             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 842             j_coord_offsetA  = DIM*jnrA;
 843             j_coord_offsetB  = DIM*jnrB;
 844             j_coord_offsetC  = DIM*jnrC;
 845             j_coord_offsetD  = DIM*jnrD;
 846
 847             /* load j atom coordinates */
 848             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 849                                               x+j_coord_offsetC,x+j_coord_offsetD,
 850                                               &jx0,&jy0,&jz0);
 851
 852             /* Calculate displacement vector */
 853             dx10             = _mm_sub_ps(ix1,jx0);
 854             dy10             = _mm_sub_ps(iy1,jy0);
 855             dz10             = _mm_sub_ps(iz1,jz0);
 856             dx20             = _mm_sub_ps(ix2,jx0);
 857             dy20             = _mm_sub_ps(iy2,jy0);
 858             dz20             = _mm_sub_ps(iz2,jz0);
 859             dx30             = _mm_sub_ps(ix3,jx0);
 860             dy30             = _mm_sub_ps(iy3,jy0);
 861             dz30             = _mm_sub_ps(iz3,jz0);
 862
 863             /* Calculate squared distance and things based on it */
 864             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 865             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 866             rsq30            = gmx_mm_calc_rsq_ps(dx30,dy30,dz30);
 867
 868             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 869             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 870             rinv30           = gmx_mm_invsqrt_ps(rsq30);
 871
 872             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 873             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 874             rinvsq30         = _mm_mul_ps(rinv30,rinv30);
 875
 876             /* Load parameters for j particles */
 877             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 878                                                               charge+jnrC+0,charge+jnrD+0);
 879
 880             fjx0             = _mm_setzero_ps();
 881             fjy0             = _mm_setzero_ps();
 882             fjz0             = _mm_setzero_ps();
 883
 884             /**************************
 885              * CALCULATE INTERACTIONS *
 886              **************************/
 887
 888             r10              = _mm_mul_ps(rsq10,rinv10);
 889             r10              = _mm_andnot_ps(dummy_mask,r10);
 890
 891             /* Compute parameters for interactions between i and j atoms */
 892             qq10             = _mm_mul_ps(iq1,jq0);
 893
 894             /* EWALD ELECTROSTATICS */
 895
 896             /* Analytical PME correction */
 897             zeta2            = _mm_mul_ps(beta2,rsq10);
 898             rinv3            = _mm_mul_ps(rinvsq10,rinv10);
 899             pmecorrF         = gmx_mm_pmecorrF_ps(zeta2);
 900             felec            = _mm_macc_ps(pmecorrF,beta3,rinv3);
 901             felec            = _mm_mul_ps(qq10,felec);
 902
 903             fscal            = felec;
 904
 905             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 906
 907              /* Update vectorial force */
 908             fix1             = _mm_macc_ps(dx10,fscal,fix1);
 909             fiy1             = _mm_macc_ps(dy10,fscal,fiy1);
 910             fiz1             = _mm_macc_ps(dz10,fscal,fiz1);
 911
 912             fjx0             = _mm_macc_ps(dx10,fscal,fjx0);
 913             fjy0             = _mm_macc_ps(dy10,fscal,fjy0);
 914             fjz0             = _mm_macc_ps(dz10,fscal,fjz0);
 915
 916             /**************************
 917              * CALCULATE INTERACTIONS *
 918              **************************/
 919
 920             r20              = _mm_mul_ps(rsq20,rinv20);
 921             r20              = _mm_andnot_ps(dummy_mask,r20);
 922
 923             /* Compute parameters for interactions between i and j atoms */
 924             qq20             = _mm_mul_ps(iq2,jq0);
 925
 926             /* EWALD ELECTROSTATICS */
 927
 928             /* Analytical PME correction */
 929             zeta2            = _mm_mul_ps(beta2,rsq20);
 930             rinv3            = _mm_mul_ps(rinvsq20,rinv20);
 931             pmecorrF         = gmx_mm_pmecorrF_ps(zeta2);
 932             felec            = _mm_macc_ps(pmecorrF,beta3,rinv3);
 933             felec            = _mm_mul_ps(qq20,felec);
 934
 935             fscal            = felec;
 936
 937             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 938
 939              /* Update vectorial force */
 940             fix2             = _mm_macc_ps(dx20,fscal,fix2);
 941             fiy2             = _mm_macc_ps(dy20,fscal,fiy2);
 942             fiz2             = _mm_macc_ps(dz20,fscal,fiz2);
 943
 944             fjx0             = _mm_macc_ps(dx20,fscal,fjx0);
 945             fjy0             = _mm_macc_ps(dy20,fscal,fjy0);
 946             fjz0             = _mm_macc_ps(dz20,fscal,fjz0);
 947
 948             /**************************
 949              * CALCULATE INTERACTIONS *
 950              **************************/
 951
 952             r30              = _mm_mul_ps(rsq30,rinv30);
 953             r30              = _mm_andnot_ps(dummy_mask,r30);
 954
 955             /* Compute parameters for interactions between i and j atoms */
 956             qq30             = _mm_mul_ps(iq3,jq0);
 957
 958             /* EWALD ELECTROSTATICS */
 959
 960             /* Analytical PME correction */
 961             zeta2            = _mm_mul_ps(beta2,rsq30);
 962             rinv3            = _mm_mul_ps(rinvsq30,rinv30);
 963             pmecorrF         = gmx_mm_pmecorrF_ps(zeta2);
 964             felec            = _mm_macc_ps(pmecorrF,beta3,rinv3);
 965             felec            = _mm_mul_ps(qq30,felec);
 966
 967             fscal            = felec;
 968
 969             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 970
 971              /* Update vectorial force */
 972             fix3             = _mm_macc_ps(dx30,fscal,fix3);
 973             fiy3             = _mm_macc_ps(dy30,fscal,fiy3);
 974             fiz3             = _mm_macc_ps(dz30,fscal,fiz3);
 975
 976             fjx0             = _mm_macc_ps(dx30,fscal,fjx0);
 977             fjy0             = _mm_macc_ps(dy30,fscal,fjy0);
 978             fjz0             = _mm_macc_ps(dz30,fscal,fjz0);
 979
 980             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 981             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 982             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 983             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 984
 985             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 986
 987             /* Inner loop uses 87 flops */
 988         }
 989
 990         /* End of innermost loop */
 991
 992         gmx_mm_update_iforce_3atom_swizzle_ps(fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
 993                                               f+i_coord_offset+DIM,fshift+i_shift_offset);
 994
 995         /* Increment number of inner iterations */
 996         inneriter                  += j_index_end - j_index_start;
 997
 998         /* Outer loop uses 18 flops */
 999     }
1000
1001     /* Increment number of outer iterations */
1002     outeriter        += nri;
1003
1004     /* Update outer/inner flops */
1005
1006     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_W4_F,outeriter*18 + inneriter*87);
1007 }