src/gromacs/gmxlib/nonbonded/nb_kernel_avx_128_fma_single/nb_kernel_ElecEw_VdwLJ_GeomW3W3_avx_128_fma_single.cpp

   1 /*
   2  * This file is part of the GROMACS molecular simulation package.
   3  *
   4  * Copyright (c) 2012,2013,2014,2015,2017,2018, 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 "gmxpre.h"
  39
  40 #include "config.h"
  41
  42 #include <math.h>
  43
  44 #include "../nb_kernel.h"
  45 #include "gromacs/gmxlib/nrnb.h"
  46
  47 #include "kernelutil_x86_avx_128_fma_single.h"
  48
  49 /*
  50  * Gromacs nonbonded kernel:   nb_kernel_ElecEw_VdwLJ_GeomW3W3_VF_avx_128_fma_single
  51  * Electrostatics interaction: Ewald
  52  * VdW interaction:            LennardJones
  53  * Geometry:                   Water3-Water3
  54  * Calculate force/pot:        PotentialAndForce
  55  */
  56 void
  57 nb_kernel_ElecEw_VdwLJ_GeomW3W3_VF_avx_128_fma_single
  58                     (t_nblist                    * gmx_restrict       nlist,
  59                      rvec                        * gmx_restrict          xx,
  60                      rvec                        * gmx_restrict          ff,
  61                      struct t_forcerec           * gmx_restrict          fr,
  62                      t_mdatoms                   * gmx_restrict     mdatoms,
  63                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
  64                      t_nrnb                      * gmx_restrict        nrnb)
  65 {
  66     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
  67      * just 0 for non-waters.
  68      * Suffixes A,B,C,D refer to j loop unrolling done with AVX_128, e.g. for the four different
  69      * jnr indices corresponding to data put in the four positions in the SIMD register.
  70      */
  71     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
  72     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
  73     int              jnrA,jnrB,jnrC,jnrD;
  74     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
  75     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
  76     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
  77     real             rcutoff_scalar;
  78     real             *shiftvec,*fshift,*x,*f;
  79     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
  80     real             scratch[4*DIM];
  81     __m128           fscal,rcutoff,rcutoff2,jidxall;
  82     int              vdwioffset0;
  83     __m128           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
  84     int              vdwioffset1;
  85     __m128           ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
  86     int              vdwioffset2;
  87     __m128           ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
  88     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
  89     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  90     int              vdwjidx1A,vdwjidx1B,vdwjidx1C,vdwjidx1D;
  91     __m128           jx1,jy1,jz1,fjx1,fjy1,fjz1,jq1,isaj1;
  92     int              vdwjidx2A,vdwjidx2B,vdwjidx2C,vdwjidx2D;
  93     __m128           jx2,jy2,jz2,fjx2,fjy2,fjz2,jq2,isaj2;
  94     __m128           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  95     __m128           dx01,dy01,dz01,rsq01,rinv01,rinvsq01,r01,qq01,c6_01,c12_01;
  96     __m128           dx02,dy02,dz02,rsq02,rinv02,rinvsq02,r02,qq02,c6_02,c12_02;
  97     __m128           dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
  98     __m128           dx11,dy11,dz11,rsq11,rinv11,rinvsq11,r11,qq11,c6_11,c12_11;
  99     __m128           dx12,dy12,dz12,rsq12,rinv12,rinvsq12,r12,qq12,c6_12,c12_12;
 100     __m128           dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
 101     __m128           dx21,dy21,dz21,rsq21,rinv21,rinvsq21,r21,qq21,c6_21,c12_21;
 102     __m128           dx22,dy22,dz22,rsq22,rinv22,rinvsq22,r22,qq22,c6_22,c12_22;
 103     __m128           velec,felec,velecsum,facel,crf,krf,krf2;
 104     real             *charge;
 105     int              nvdwtype;
 106     __m128           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 107     int              *vdwtype;
 108     real             *vdwparam;
 109     __m128           one_sixth   = _mm_set1_ps(1.0/6.0);
 110     __m128           one_twelfth = _mm_set1_ps(1.0/12.0);
 111     __m128i          ewitab;
 112     __m128           ewtabscale,eweps,twoeweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
 113     __m128           beta,beta2,beta3,zeta2,pmecorrF,pmecorrV,rinv3;
 114     real             *ewtab;
 115     __m128           dummy_mask,cutoff_mask;
 116     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
 117     __m128           one     = _mm_set1_ps(1.0);
 118     __m128           two     = _mm_set1_ps(2.0);
 119     x                = xx[0];
 120     f                = ff[0];
 121
 122     nri              = nlist->nri;
 123     iinr             = nlist->iinr;
 124     jindex           = nlist->jindex;
 125     jjnr             = nlist->jjnr;
 126     shiftidx         = nlist->shift;
 127     gid              = nlist->gid;
 128     shiftvec         = fr->shift_vec[0];
 129     fshift           = fr->fshift[0];
 130     facel            = _mm_set1_ps(fr->ic->epsfac);
 131     charge           = mdatoms->chargeA;
 132     nvdwtype         = fr->ntype;
 133     vdwparam         = fr->nbfp;
 134     vdwtype          = mdatoms->typeA;
 135
 136     sh_ewald         = _mm_set1_ps(fr->ic->sh_ewald);
 137     beta             = _mm_set1_ps(fr->ic->ewaldcoeff_q);
 138     beta2            = _mm_mul_ps(beta,beta);
 139     beta3            = _mm_mul_ps(beta,beta2);
 140     ewtab            = fr->ic->tabq_coul_FDV0;
 141     ewtabscale       = _mm_set1_ps(fr->ic->tabq_scale);
 142     ewtabhalfspace   = _mm_set1_ps(0.5/fr->ic->tabq_scale);
 143
 144     /* Setup water-specific parameters */
 145     inr              = nlist->iinr[0];
 146     iq0              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+0]));
 147     iq1              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+1]));
 148     iq2              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+2]));
 149     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 150
 151     jq0              = _mm_set1_ps(charge[inr+0]);
 152     jq1              = _mm_set1_ps(charge[inr+1]);
 153     jq2              = _mm_set1_ps(charge[inr+2]);
 154     vdwjidx0A        = 2*vdwtype[inr+0];
 155     qq00             = _mm_mul_ps(iq0,jq0);
 156     c6_00            = _mm_set1_ps(vdwparam[vdwioffset0+vdwjidx0A]);
 157     c12_00           = _mm_set1_ps(vdwparam[vdwioffset0+vdwjidx0A+1]);
 158     qq01             = _mm_mul_ps(iq0,jq1);
 159     qq02             = _mm_mul_ps(iq0,jq2);
 160     qq10             = _mm_mul_ps(iq1,jq0);
 161     qq11             = _mm_mul_ps(iq1,jq1);
 162     qq12             = _mm_mul_ps(iq1,jq2);
 163     qq20             = _mm_mul_ps(iq2,jq0);
 164     qq21             = _mm_mul_ps(iq2,jq1);
 165     qq22             = _mm_mul_ps(iq2,jq2);
 166
 167     /* Avoid stupid compiler warnings */
 168     jnrA = jnrB = jnrC = jnrD = 0;
 169     j_coord_offsetA = 0;
 170     j_coord_offsetB = 0;
 171     j_coord_offsetC = 0;
 172     j_coord_offsetD = 0;
 173
 174     outeriter        = 0;
 175     inneriter        = 0;
 176
 177     for(iidx=0;iidx<4*DIM;iidx++)
 178     {
 179         scratch[iidx] = 0.0;
 180     }
 181
 182     /* Start outer loop over neighborlists */
 183     for(iidx=0; iidx<nri; iidx++)
 184     {
 185         /* Load shift vector for this list */
 186         i_shift_offset   = DIM*shiftidx[iidx];
 187
 188         /* Load limits for loop over neighbors */
 189         j_index_start    = jindex[iidx];
 190         j_index_end      = jindex[iidx+1];
 191
 192         /* Get outer coordinate index */
 193         inr              = iinr[iidx];
 194         i_coord_offset   = DIM*inr;
 195
 196         /* Load i particle coords and add shift vector */
 197         gmx_mm_load_shift_and_3rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
 198                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 199
 200         fix0             = _mm_setzero_ps();
 201         fiy0             = _mm_setzero_ps();
 202         fiz0             = _mm_setzero_ps();
 203         fix1             = _mm_setzero_ps();
 204         fiy1             = _mm_setzero_ps();
 205         fiz1             = _mm_setzero_ps();
 206         fix2             = _mm_setzero_ps();
 207         fiy2             = _mm_setzero_ps();
 208         fiz2             = _mm_setzero_ps();
 209
 210         /* Reset potential sums */
 211         velecsum         = _mm_setzero_ps();
 212         vvdwsum          = _mm_setzero_ps();
 213
 214         /* Start inner kernel loop */
 215         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 216         {
 217
 218             /* Get j neighbor index, and coordinate index */
 219             jnrA             = jjnr[jidx];
 220             jnrB             = jjnr[jidx+1];
 221             jnrC             = jjnr[jidx+2];
 222             jnrD             = jjnr[jidx+3];
 223             j_coord_offsetA  = DIM*jnrA;
 224             j_coord_offsetB  = DIM*jnrB;
 225             j_coord_offsetC  = DIM*jnrC;
 226             j_coord_offsetD  = DIM*jnrD;
 227
 228             /* load j atom coordinates */
 229             gmx_mm_load_3rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 230                                               x+j_coord_offsetC,x+j_coord_offsetD,
 231                                               &jx0,&jy0,&jz0,&jx1,&jy1,&jz1,&jx2,&jy2,&jz2);
 232
 233             /* Calculate displacement vector */
 234             dx00             = _mm_sub_ps(ix0,jx0);
 235             dy00             = _mm_sub_ps(iy0,jy0);
 236             dz00             = _mm_sub_ps(iz0,jz0);
 237             dx01             = _mm_sub_ps(ix0,jx1);
 238             dy01             = _mm_sub_ps(iy0,jy1);
 239             dz01             = _mm_sub_ps(iz0,jz1);
 240             dx02             = _mm_sub_ps(ix0,jx2);
 241             dy02             = _mm_sub_ps(iy0,jy2);
 242             dz02             = _mm_sub_ps(iz0,jz2);
 243             dx10             = _mm_sub_ps(ix1,jx0);
 244             dy10             = _mm_sub_ps(iy1,jy0);
 245             dz10             = _mm_sub_ps(iz1,jz0);
 246             dx11             = _mm_sub_ps(ix1,jx1);
 247             dy11             = _mm_sub_ps(iy1,jy1);
 248             dz11             = _mm_sub_ps(iz1,jz1);
 249             dx12             = _mm_sub_ps(ix1,jx2);
 250             dy12             = _mm_sub_ps(iy1,jy2);
 251             dz12             = _mm_sub_ps(iz1,jz2);
 252             dx20             = _mm_sub_ps(ix2,jx0);
 253             dy20             = _mm_sub_ps(iy2,jy0);
 254             dz20             = _mm_sub_ps(iz2,jz0);
 255             dx21             = _mm_sub_ps(ix2,jx1);
 256             dy21             = _mm_sub_ps(iy2,jy1);
 257             dz21             = _mm_sub_ps(iz2,jz1);
 258             dx22             = _mm_sub_ps(ix2,jx2);
 259             dy22             = _mm_sub_ps(iy2,jy2);
 260             dz22             = _mm_sub_ps(iz2,jz2);
 261
 262             /* Calculate squared distance and things based on it */
 263             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 264             rsq01            = gmx_mm_calc_rsq_ps(dx01,dy01,dz01);
 265             rsq02            = gmx_mm_calc_rsq_ps(dx02,dy02,dz02);
 266             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 267             rsq11            = gmx_mm_calc_rsq_ps(dx11,dy11,dz11);
 268             rsq12            = gmx_mm_calc_rsq_ps(dx12,dy12,dz12);
 269             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 270             rsq21            = gmx_mm_calc_rsq_ps(dx21,dy21,dz21);
 271             rsq22            = gmx_mm_calc_rsq_ps(dx22,dy22,dz22);
 272
 273             rinv00           = avx128fma_invsqrt_f(rsq00);
 274             rinv01           = avx128fma_invsqrt_f(rsq01);
 275             rinv02           = avx128fma_invsqrt_f(rsq02);
 276             rinv10           = avx128fma_invsqrt_f(rsq10);
 277             rinv11           = avx128fma_invsqrt_f(rsq11);
 278             rinv12           = avx128fma_invsqrt_f(rsq12);
 279             rinv20           = avx128fma_invsqrt_f(rsq20);
 280             rinv21           = avx128fma_invsqrt_f(rsq21);
 281             rinv22           = avx128fma_invsqrt_f(rsq22);
 282
 283             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 284             rinvsq01         = _mm_mul_ps(rinv01,rinv01);
 285             rinvsq02         = _mm_mul_ps(rinv02,rinv02);
 286             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 287             rinvsq11         = _mm_mul_ps(rinv11,rinv11);
 288             rinvsq12         = _mm_mul_ps(rinv12,rinv12);
 289             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 290             rinvsq21         = _mm_mul_ps(rinv21,rinv21);
 291             rinvsq22         = _mm_mul_ps(rinv22,rinv22);
 292
 293             fjx0             = _mm_setzero_ps();
 294             fjy0             = _mm_setzero_ps();
 295             fjz0             = _mm_setzero_ps();
 296             fjx1             = _mm_setzero_ps();
 297             fjy1             = _mm_setzero_ps();
 298             fjz1             = _mm_setzero_ps();
 299             fjx2             = _mm_setzero_ps();
 300             fjy2             = _mm_setzero_ps();
 301             fjz2             = _mm_setzero_ps();
 302
 303             /**************************
 304              * CALCULATE INTERACTIONS *
 305              **************************/
 306
 307             r00              = _mm_mul_ps(rsq00,rinv00);
 308
 309             /* EWALD ELECTROSTATICS */
 310
 311             /* Analytical PME correction */
 312             zeta2            = _mm_mul_ps(beta2,rsq00);
 313             rinv3            = _mm_mul_ps(rinvsq00,rinv00);
 314             pmecorrF         = avx128fma_pmecorrF_f(zeta2);
 315             felec            = _mm_macc_ps(pmecorrF,beta3,rinv3);
 316             felec            = _mm_mul_ps(qq00,felec);
 317             pmecorrV         = avx128fma_pmecorrV_f(zeta2);
 318             velec            = _mm_nmacc_ps(pmecorrV,beta,rinv00);
 319             velec            = _mm_mul_ps(qq00,velec);
 320
 321             /* LENNARD-JONES DISPERSION/REPULSION */
 322
 323             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 324             vvdw6            = _mm_mul_ps(c6_00,rinvsix);
 325             vvdw12           = _mm_mul_ps(c12_00,_mm_mul_ps(rinvsix,rinvsix));
 326             vvdw             = _mm_msub_ps(vvdw12,one_twelfth,_mm_mul_ps(vvdw6,one_sixth));
 327             fvdw             = _mm_mul_ps(_mm_sub_ps(vvdw12,vvdw6),rinvsq00);
 328
 329             /* Update potential sum for this i atom from the interaction with this j atom. */
 330             velecsum         = _mm_add_ps(velecsum,velec);
 331             vvdwsum          = _mm_add_ps(vvdwsum,vvdw);
 332
 333             fscal            = _mm_add_ps(felec,fvdw);
 334
 335              /* Update vectorial force */
 336             fix0             = _mm_macc_ps(dx00,fscal,fix0);
 337             fiy0             = _mm_macc_ps(dy00,fscal,fiy0);
 338             fiz0             = _mm_macc_ps(dz00,fscal,fiz0);
 339
 340             fjx0             = _mm_macc_ps(dx00,fscal,fjx0);
 341             fjy0             = _mm_macc_ps(dy00,fscal,fjy0);
 342             fjz0             = _mm_macc_ps(dz00,fscal,fjz0);
 343
 344             /**************************
 345              * CALCULATE INTERACTIONS *
 346              **************************/
 347
 348             r01              = _mm_mul_ps(rsq01,rinv01);
 349
 350             /* EWALD ELECTROSTATICS */
 351
 352             /* Analytical PME correction */
 353             zeta2            = _mm_mul_ps(beta2,rsq01);
 354             rinv3            = _mm_mul_ps(rinvsq01,rinv01);
 355             pmecorrF         = avx128fma_pmecorrF_f(zeta2);
 356             felec            = _mm_macc_ps(pmecorrF,beta3,rinv3);
 357             felec            = _mm_mul_ps(qq01,felec);
 358             pmecorrV         = avx128fma_pmecorrV_f(zeta2);
 359             velec            = _mm_nmacc_ps(pmecorrV,beta,rinv01);
 360             velec            = _mm_mul_ps(qq01,velec);
 361
 362             /* Update potential sum for this i atom from the interaction with this j atom. */
 363             velecsum         = _mm_add_ps(velecsum,velec);
 364
 365             fscal            = felec;
 366
 367              /* Update vectorial force */
 368             fix0             = _mm_macc_ps(dx01,fscal,fix0);
 369             fiy0             = _mm_macc_ps(dy01,fscal,fiy0);
 370             fiz0             = _mm_macc_ps(dz01,fscal,fiz0);
 371
 372             fjx1             = _mm_macc_ps(dx01,fscal,fjx1);
 373             fjy1             = _mm_macc_ps(dy01,fscal,fjy1);
 374             fjz1             = _mm_macc_ps(dz01,fscal,fjz1);
 375
 376             /**************************
 377              * CALCULATE INTERACTIONS *
 378              **************************/
 379
 380             r02              = _mm_mul_ps(rsq02,rinv02);
 381
 382             /* EWALD ELECTROSTATICS */
 383
 384             /* Analytical PME correction */
 385             zeta2            = _mm_mul_ps(beta2,rsq02);
 386             rinv3            = _mm_mul_ps(rinvsq02,rinv02);
 387             pmecorrF         = avx128fma_pmecorrF_f(zeta2);
 388             felec            = _mm_macc_ps(pmecorrF,beta3,rinv3);
 389             felec            = _mm_mul_ps(qq02,felec);
 390             pmecorrV         = avx128fma_pmecorrV_f(zeta2);
 391             velec            = _mm_nmacc_ps(pmecorrV,beta,rinv02);
 392             velec            = _mm_mul_ps(qq02,velec);
 393
 394             /* Update potential sum for this i atom from the interaction with this j atom. */
 395             velecsum         = _mm_add_ps(velecsum,velec);
 396
 397             fscal            = felec;
 398
 399              /* Update vectorial force */
 400             fix0             = _mm_macc_ps(dx02,fscal,fix0);
 401             fiy0             = _mm_macc_ps(dy02,fscal,fiy0);
 402             fiz0             = _mm_macc_ps(dz02,fscal,fiz0);
 403
 404             fjx2             = _mm_macc_ps(dx02,fscal,fjx2);
 405             fjy2             = _mm_macc_ps(dy02,fscal,fjy2);
 406             fjz2             = _mm_macc_ps(dz02,fscal,fjz2);
 407
 408             /**************************
 409              * CALCULATE INTERACTIONS *
 410              **************************/
 411
 412             r10              = _mm_mul_ps(rsq10,rinv10);
 413
 414             /* EWALD ELECTROSTATICS */
 415
 416             /* Analytical PME correction */
 417             zeta2            = _mm_mul_ps(beta2,rsq10);
 418             rinv3            = _mm_mul_ps(rinvsq10,rinv10);
 419             pmecorrF         = avx128fma_pmecorrF_f(zeta2);
 420             felec            = _mm_macc_ps(pmecorrF,beta3,rinv3);
 421             felec            = _mm_mul_ps(qq10,felec);
 422             pmecorrV         = avx128fma_pmecorrV_f(zeta2);
 423             velec            = _mm_nmacc_ps(pmecorrV,beta,rinv10);
 424             velec            = _mm_mul_ps(qq10,velec);
 425
 426             /* Update potential sum for this i atom from the interaction with this j atom. */
 427             velecsum         = _mm_add_ps(velecsum,velec);
 428
 429             fscal            = felec;
 430
 431              /* Update vectorial force */
 432             fix1             = _mm_macc_ps(dx10,fscal,fix1);
 433             fiy1             = _mm_macc_ps(dy10,fscal,fiy1);
 434             fiz1             = _mm_macc_ps(dz10,fscal,fiz1);
 435
 436             fjx0             = _mm_macc_ps(dx10,fscal,fjx0);
 437             fjy0             = _mm_macc_ps(dy10,fscal,fjy0);
 438             fjz0             = _mm_macc_ps(dz10,fscal,fjz0);
 439
 440             /**************************
 441              * CALCULATE INTERACTIONS *
 442              **************************/
 443
 444             r11              = _mm_mul_ps(rsq11,rinv11);
 445
 446             /* EWALD ELECTROSTATICS */
 447
 448             /* Analytical PME correction */
 449             zeta2            = _mm_mul_ps(beta2,rsq11);
 450             rinv3            = _mm_mul_ps(rinvsq11,rinv11);
 451             pmecorrF         = avx128fma_pmecorrF_f(zeta2);
 452             felec            = _mm_macc_ps(pmecorrF,beta3,rinv3);
 453             felec            = _mm_mul_ps(qq11,felec);
 454             pmecorrV         = avx128fma_pmecorrV_f(zeta2);
 455             velec            = _mm_nmacc_ps(pmecorrV,beta,rinv11);
 456             velec            = _mm_mul_ps(qq11,velec);
 457
 458             /* Update potential sum for this i atom from the interaction with this j atom. */
 459             velecsum         = _mm_add_ps(velecsum,velec);
 460
 461             fscal            = felec;
 462
 463              /* Update vectorial force */
 464             fix1             = _mm_macc_ps(dx11,fscal,fix1);
 465             fiy1             = _mm_macc_ps(dy11,fscal,fiy1);
 466             fiz1             = _mm_macc_ps(dz11,fscal,fiz1);
 467
 468             fjx1             = _mm_macc_ps(dx11,fscal,fjx1);
 469             fjy1             = _mm_macc_ps(dy11,fscal,fjy1);
 470             fjz1             = _mm_macc_ps(dz11,fscal,fjz1);
 471
 472             /**************************
 473              * CALCULATE INTERACTIONS *
 474              **************************/
 475
 476             r12              = _mm_mul_ps(rsq12,rinv12);
 477
 478             /* EWALD ELECTROSTATICS */
 479
 480             /* Analytical PME correction */
 481             zeta2            = _mm_mul_ps(beta2,rsq12);
 482             rinv3            = _mm_mul_ps(rinvsq12,rinv12);
 483             pmecorrF         = avx128fma_pmecorrF_f(zeta2);
 484             felec            = _mm_macc_ps(pmecorrF,beta3,rinv3);
 485             felec            = _mm_mul_ps(qq12,felec);
 486             pmecorrV         = avx128fma_pmecorrV_f(zeta2);
 487             velec            = _mm_nmacc_ps(pmecorrV,beta,rinv12);
 488             velec            = _mm_mul_ps(qq12,velec);
 489
 490             /* Update potential sum for this i atom from the interaction with this j atom. */
 491             velecsum         = _mm_add_ps(velecsum,velec);
 492
 493             fscal            = felec;
 494
 495              /* Update vectorial force */
 496             fix1             = _mm_macc_ps(dx12,fscal,fix1);
 497             fiy1             = _mm_macc_ps(dy12,fscal,fiy1);
 498             fiz1             = _mm_macc_ps(dz12,fscal,fiz1);
 499
 500             fjx2             = _mm_macc_ps(dx12,fscal,fjx2);
 501             fjy2             = _mm_macc_ps(dy12,fscal,fjy2);
 502             fjz2             = _mm_macc_ps(dz12,fscal,fjz2);
 503
 504             /**************************
 505              * CALCULATE INTERACTIONS *
 506              **************************/
 507
 508             r20              = _mm_mul_ps(rsq20,rinv20);
 509
 510             /* EWALD ELECTROSTATICS */
 511
 512             /* Analytical PME correction */
 513             zeta2            = _mm_mul_ps(beta2,rsq20);
 514             rinv3            = _mm_mul_ps(rinvsq20,rinv20);
 515             pmecorrF         = avx128fma_pmecorrF_f(zeta2);
 516             felec            = _mm_macc_ps(pmecorrF,beta3,rinv3);
 517             felec            = _mm_mul_ps(qq20,felec);
 518             pmecorrV         = avx128fma_pmecorrV_f(zeta2);
 519             velec            = _mm_nmacc_ps(pmecorrV,beta,rinv20);
 520             velec            = _mm_mul_ps(qq20,velec);
 521
 522             /* Update potential sum for this i atom from the interaction with this j atom. */
 523             velecsum         = _mm_add_ps(velecsum,velec);
 524
 525             fscal            = felec;
 526
 527              /* Update vectorial force */
 528             fix2             = _mm_macc_ps(dx20,fscal,fix2);
 529             fiy2             = _mm_macc_ps(dy20,fscal,fiy2);
 530             fiz2             = _mm_macc_ps(dz20,fscal,fiz2);
 531
 532             fjx0             = _mm_macc_ps(dx20,fscal,fjx0);
 533             fjy0             = _mm_macc_ps(dy20,fscal,fjy0);
 534             fjz0             = _mm_macc_ps(dz20,fscal,fjz0);
 535
 536             /**************************
 537              * CALCULATE INTERACTIONS *
 538              **************************/
 539
 540             r21              = _mm_mul_ps(rsq21,rinv21);
 541
 542             /* EWALD ELECTROSTATICS */
 543
 544             /* Analytical PME correction */
 545             zeta2            = _mm_mul_ps(beta2,rsq21);
 546             rinv3            = _mm_mul_ps(rinvsq21,rinv21);
 547             pmecorrF         = avx128fma_pmecorrF_f(zeta2);
 548             felec            = _mm_macc_ps(pmecorrF,beta3,rinv3);
 549             felec            = _mm_mul_ps(qq21,felec);
 550             pmecorrV         = avx128fma_pmecorrV_f(zeta2);
 551             velec            = _mm_nmacc_ps(pmecorrV,beta,rinv21);
 552             velec            = _mm_mul_ps(qq21,velec);
 553
 554             /* Update potential sum for this i atom from the interaction with this j atom. */
 555             velecsum         = _mm_add_ps(velecsum,velec);
 556
 557             fscal            = felec;
 558
 559              /* Update vectorial force */
 560             fix2             = _mm_macc_ps(dx21,fscal,fix2);
 561             fiy2             = _mm_macc_ps(dy21,fscal,fiy2);
 562             fiz2             = _mm_macc_ps(dz21,fscal,fiz2);
 563
 564             fjx1             = _mm_macc_ps(dx21,fscal,fjx1);
 565             fjy1             = _mm_macc_ps(dy21,fscal,fjy1);
 566             fjz1             = _mm_macc_ps(dz21,fscal,fjz1);
 567
 568             /**************************
 569              * CALCULATE INTERACTIONS *
 570              **************************/
 571
 572             r22              = _mm_mul_ps(rsq22,rinv22);
 573
 574             /* EWALD ELECTROSTATICS */
 575
 576             /* Analytical PME correction */
 577             zeta2            = _mm_mul_ps(beta2,rsq22);
 578             rinv3            = _mm_mul_ps(rinvsq22,rinv22);
 579             pmecorrF         = avx128fma_pmecorrF_f(zeta2);
 580             felec            = _mm_macc_ps(pmecorrF,beta3,rinv3);
 581             felec            = _mm_mul_ps(qq22,felec);
 582             pmecorrV         = avx128fma_pmecorrV_f(zeta2);
 583             velec            = _mm_nmacc_ps(pmecorrV,beta,rinv22);
 584             velec            = _mm_mul_ps(qq22,velec);
 585
 586             /* Update potential sum for this i atom from the interaction with this j atom. */
 587             velecsum         = _mm_add_ps(velecsum,velec);
 588
 589             fscal            = felec;
 590
 591              /* Update vectorial force */
 592             fix2             = _mm_macc_ps(dx22,fscal,fix2);
 593             fiy2             = _mm_macc_ps(dy22,fscal,fiy2);
 594             fiz2             = _mm_macc_ps(dz22,fscal,fiz2);
 595
 596             fjx2             = _mm_macc_ps(dx22,fscal,fjx2);
 597             fjy2             = _mm_macc_ps(dy22,fscal,fjy2);
 598             fjz2             = _mm_macc_ps(dz22,fscal,fjz2);
 599
 600             fjptrA             = f+j_coord_offsetA;
 601             fjptrB             = f+j_coord_offsetB;
 602             fjptrC             = f+j_coord_offsetC;
 603             fjptrD             = f+j_coord_offsetD;
 604
 605             gmx_mm_decrement_3rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,
 606                                                    fjx0,fjy0,fjz0,fjx1,fjy1,fjz1,fjx2,fjy2,fjz2);
 607
 608             /* Inner loop uses 273 flops */
 609         }
 610
 611         if(jidx<j_index_end)
 612         {
 613
 614             /* Get j neighbor index, and coordinate index */
 615             jnrlistA         = jjnr[jidx];
 616             jnrlistB         = jjnr[jidx+1];
 617             jnrlistC         = jjnr[jidx+2];
 618             jnrlistD         = jjnr[jidx+3];
 619             /* Sign of each element will be negative for non-real atoms.
 620              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 621              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 622              */
 623             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 624             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 625             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 626             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 627             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 628             j_coord_offsetA  = DIM*jnrA;
 629             j_coord_offsetB  = DIM*jnrB;
 630             j_coord_offsetC  = DIM*jnrC;
 631             j_coord_offsetD  = DIM*jnrD;
 632
 633             /* load j atom coordinates */
 634             gmx_mm_load_3rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 635                                               x+j_coord_offsetC,x+j_coord_offsetD,
 636                                               &jx0,&jy0,&jz0,&jx1,&jy1,&jz1,&jx2,&jy2,&jz2);
 637
 638             /* Calculate displacement vector */
 639             dx00             = _mm_sub_ps(ix0,jx0);
 640             dy00             = _mm_sub_ps(iy0,jy0);
 641             dz00             = _mm_sub_ps(iz0,jz0);
 642             dx01             = _mm_sub_ps(ix0,jx1);
 643             dy01             = _mm_sub_ps(iy0,jy1);
 644             dz01             = _mm_sub_ps(iz0,jz1);
 645             dx02             = _mm_sub_ps(ix0,jx2);
 646             dy02             = _mm_sub_ps(iy0,jy2);
 647             dz02             = _mm_sub_ps(iz0,jz2);
 648             dx10             = _mm_sub_ps(ix1,jx0);
 649             dy10             = _mm_sub_ps(iy1,jy0);
 650             dz10             = _mm_sub_ps(iz1,jz0);
 651             dx11             = _mm_sub_ps(ix1,jx1);
 652             dy11             = _mm_sub_ps(iy1,jy1);
 653             dz11             = _mm_sub_ps(iz1,jz1);
 654             dx12             = _mm_sub_ps(ix1,jx2);
 655             dy12             = _mm_sub_ps(iy1,jy2);
 656             dz12             = _mm_sub_ps(iz1,jz2);
 657             dx20             = _mm_sub_ps(ix2,jx0);
 658             dy20             = _mm_sub_ps(iy2,jy0);
 659             dz20             = _mm_sub_ps(iz2,jz0);
 660             dx21             = _mm_sub_ps(ix2,jx1);
 661             dy21             = _mm_sub_ps(iy2,jy1);
 662             dz21             = _mm_sub_ps(iz2,jz1);
 663             dx22             = _mm_sub_ps(ix2,jx2);
 664             dy22             = _mm_sub_ps(iy2,jy2);
 665             dz22             = _mm_sub_ps(iz2,jz2);
 666
 667             /* Calculate squared distance and things based on it */
 668             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 669             rsq01            = gmx_mm_calc_rsq_ps(dx01,dy01,dz01);
 670             rsq02            = gmx_mm_calc_rsq_ps(dx02,dy02,dz02);
 671             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 672             rsq11            = gmx_mm_calc_rsq_ps(dx11,dy11,dz11);
 673             rsq12            = gmx_mm_calc_rsq_ps(dx12,dy12,dz12);
 674             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 675             rsq21            = gmx_mm_calc_rsq_ps(dx21,dy21,dz21);
 676             rsq22            = gmx_mm_calc_rsq_ps(dx22,dy22,dz22);
 677
 678             rinv00           = avx128fma_invsqrt_f(rsq00);
 679             rinv01           = avx128fma_invsqrt_f(rsq01);
 680             rinv02           = avx128fma_invsqrt_f(rsq02);
 681             rinv10           = avx128fma_invsqrt_f(rsq10);
 682             rinv11           = avx128fma_invsqrt_f(rsq11);
 683             rinv12           = avx128fma_invsqrt_f(rsq12);
 684             rinv20           = avx128fma_invsqrt_f(rsq20);
 685             rinv21           = avx128fma_invsqrt_f(rsq21);
 686             rinv22           = avx128fma_invsqrt_f(rsq22);
 687
 688             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 689             rinvsq01         = _mm_mul_ps(rinv01,rinv01);
 690             rinvsq02         = _mm_mul_ps(rinv02,rinv02);
 691             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 692             rinvsq11         = _mm_mul_ps(rinv11,rinv11);
 693             rinvsq12         = _mm_mul_ps(rinv12,rinv12);
 694             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 695             rinvsq21         = _mm_mul_ps(rinv21,rinv21);
 696             rinvsq22         = _mm_mul_ps(rinv22,rinv22);
 697
 698             fjx0             = _mm_setzero_ps();
 699             fjy0             = _mm_setzero_ps();
 700             fjz0             = _mm_setzero_ps();
 701             fjx1             = _mm_setzero_ps();
 702             fjy1             = _mm_setzero_ps();
 703             fjz1             = _mm_setzero_ps();
 704             fjx2             = _mm_setzero_ps();
 705             fjy2             = _mm_setzero_ps();
 706             fjz2             = _mm_setzero_ps();
 707
 708             /**************************
 709              * CALCULATE INTERACTIONS *
 710              **************************/
 711
 712             r00              = _mm_mul_ps(rsq00,rinv00);
 713             r00              = _mm_andnot_ps(dummy_mask,r00);
 714
 715             /* EWALD ELECTROSTATICS */
 716
 717             /* Analytical PME correction */
 718             zeta2            = _mm_mul_ps(beta2,rsq00);
 719             rinv3            = _mm_mul_ps(rinvsq00,rinv00);
 720             pmecorrF         = avx128fma_pmecorrF_f(zeta2);
 721             felec            = _mm_macc_ps(pmecorrF,beta3,rinv3);
 722             felec            = _mm_mul_ps(qq00,felec);
 723             pmecorrV         = avx128fma_pmecorrV_f(zeta2);
 724             velec            = _mm_nmacc_ps(pmecorrV,beta,rinv00);
 725             velec            = _mm_mul_ps(qq00,velec);
 726
 727             /* LENNARD-JONES DISPERSION/REPULSION */
 728
 729             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 730             vvdw6            = _mm_mul_ps(c6_00,rinvsix);
 731             vvdw12           = _mm_mul_ps(c12_00,_mm_mul_ps(rinvsix,rinvsix));
 732             vvdw             = _mm_msub_ps(vvdw12,one_twelfth,_mm_mul_ps(vvdw6,one_sixth));
 733             fvdw             = _mm_mul_ps(_mm_sub_ps(vvdw12,vvdw6),rinvsq00);
 734
 735             /* Update potential sum for this i atom from the interaction with this j atom. */
 736             velec            = _mm_andnot_ps(dummy_mask,velec);
 737             velecsum         = _mm_add_ps(velecsum,velec);
 738             vvdw             = _mm_andnot_ps(dummy_mask,vvdw);
 739             vvdwsum          = _mm_add_ps(vvdwsum,vvdw);
 740
 741             fscal            = _mm_add_ps(felec,fvdw);
 742
 743             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 744
 745              /* Update vectorial force */
 746             fix0             = _mm_macc_ps(dx00,fscal,fix0);
 747             fiy0             = _mm_macc_ps(dy00,fscal,fiy0);
 748             fiz0             = _mm_macc_ps(dz00,fscal,fiz0);
 749
 750             fjx0             = _mm_macc_ps(dx00,fscal,fjx0);
 751             fjy0             = _mm_macc_ps(dy00,fscal,fjy0);
 752             fjz0             = _mm_macc_ps(dz00,fscal,fjz0);
 753
 754             /**************************
 755              * CALCULATE INTERACTIONS *
 756              **************************/
 757
 758             r01              = _mm_mul_ps(rsq01,rinv01);
 759             r01              = _mm_andnot_ps(dummy_mask,r01);
 760
 761             /* EWALD ELECTROSTATICS */
 762
 763             /* Analytical PME correction */
 764             zeta2            = _mm_mul_ps(beta2,rsq01);
 765             rinv3            = _mm_mul_ps(rinvsq01,rinv01);
 766             pmecorrF         = avx128fma_pmecorrF_f(zeta2);
 767             felec            = _mm_macc_ps(pmecorrF,beta3,rinv3);
 768             felec            = _mm_mul_ps(qq01,felec);
 769             pmecorrV         = avx128fma_pmecorrV_f(zeta2);
 770             velec            = _mm_nmacc_ps(pmecorrV,beta,rinv01);
 771             velec            = _mm_mul_ps(qq01,velec);
 772
 773             /* Update potential sum for this i atom from the interaction with this j atom. */
 774             velec            = _mm_andnot_ps(dummy_mask,velec);
 775             velecsum         = _mm_add_ps(velecsum,velec);
 776
 777             fscal            = felec;
 778
 779             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 780
 781              /* Update vectorial force */
 782             fix0             = _mm_macc_ps(dx01,fscal,fix0);
 783             fiy0             = _mm_macc_ps(dy01,fscal,fiy0);
 784             fiz0             = _mm_macc_ps(dz01,fscal,fiz0);
 785
 786             fjx1             = _mm_macc_ps(dx01,fscal,fjx1);
 787             fjy1             = _mm_macc_ps(dy01,fscal,fjy1);
 788             fjz1             = _mm_macc_ps(dz01,fscal,fjz1);
 789
 790             /**************************
 791              * CALCULATE INTERACTIONS *
 792              **************************/
 793
 794             r02              = _mm_mul_ps(rsq02,rinv02);
 795             r02              = _mm_andnot_ps(dummy_mask,r02);
 796
 797             /* EWALD ELECTROSTATICS */
 798
 799             /* Analytical PME correction */
 800             zeta2            = _mm_mul_ps(beta2,rsq02);
 801             rinv3            = _mm_mul_ps(rinvsq02,rinv02);
 802             pmecorrF         = avx128fma_pmecorrF_f(zeta2);
 803             felec            = _mm_macc_ps(pmecorrF,beta3,rinv3);
 804             felec            = _mm_mul_ps(qq02,felec);
 805             pmecorrV         = avx128fma_pmecorrV_f(zeta2);
 806             velec            = _mm_nmacc_ps(pmecorrV,beta,rinv02);
 807             velec            = _mm_mul_ps(qq02,velec);
 808
 809             /* Update potential sum for this i atom from the interaction with this j atom. */
 810             velec            = _mm_andnot_ps(dummy_mask,velec);
 811             velecsum         = _mm_add_ps(velecsum,velec);
 812
 813             fscal            = felec;
 814
 815             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 816
 817              /* Update vectorial force */
 818             fix0             = _mm_macc_ps(dx02,fscal,fix0);
 819             fiy0             = _mm_macc_ps(dy02,fscal,fiy0);
 820             fiz0             = _mm_macc_ps(dz02,fscal,fiz0);
 821
 822             fjx2             = _mm_macc_ps(dx02,fscal,fjx2);
 823             fjy2             = _mm_macc_ps(dy02,fscal,fjy2);
 824             fjz2             = _mm_macc_ps(dz02,fscal,fjz2);
 825
 826             /**************************
 827              * CALCULATE INTERACTIONS *
 828              **************************/
 829
 830             r10              = _mm_mul_ps(rsq10,rinv10);
 831             r10              = _mm_andnot_ps(dummy_mask,r10);
 832
 833             /* EWALD ELECTROSTATICS */
 834
 835             /* Analytical PME correction */
 836             zeta2            = _mm_mul_ps(beta2,rsq10);
 837             rinv3            = _mm_mul_ps(rinvsq10,rinv10);
 838             pmecorrF         = avx128fma_pmecorrF_f(zeta2);
 839             felec            = _mm_macc_ps(pmecorrF,beta3,rinv3);
 840             felec            = _mm_mul_ps(qq10,felec);
 841             pmecorrV         = avx128fma_pmecorrV_f(zeta2);
 842             velec            = _mm_nmacc_ps(pmecorrV,beta,rinv10);
 843             velec            = _mm_mul_ps(qq10,velec);
 844
 845             /* Update potential sum for this i atom from the interaction with this j atom. */
 846             velec            = _mm_andnot_ps(dummy_mask,velec);
 847             velecsum         = _mm_add_ps(velecsum,velec);
 848
 849             fscal            = felec;
 850
 851             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 852
 853              /* Update vectorial force */
 854             fix1             = _mm_macc_ps(dx10,fscal,fix1);
 855             fiy1             = _mm_macc_ps(dy10,fscal,fiy1);
 856             fiz1             = _mm_macc_ps(dz10,fscal,fiz1);
 857
 858             fjx0             = _mm_macc_ps(dx10,fscal,fjx0);
 859             fjy0             = _mm_macc_ps(dy10,fscal,fjy0);
 860             fjz0             = _mm_macc_ps(dz10,fscal,fjz0);
 861
 862             /**************************
 863              * CALCULATE INTERACTIONS *
 864              **************************/
 865
 866             r11              = _mm_mul_ps(rsq11,rinv11);
 867             r11              = _mm_andnot_ps(dummy_mask,r11);
 868
 869             /* EWALD ELECTROSTATICS */
 870
 871             /* Analytical PME correction */
 872             zeta2            = _mm_mul_ps(beta2,rsq11);
 873             rinv3            = _mm_mul_ps(rinvsq11,rinv11);
 874             pmecorrF         = avx128fma_pmecorrF_f(zeta2);
 875             felec            = _mm_macc_ps(pmecorrF,beta3,rinv3);
 876             felec            = _mm_mul_ps(qq11,felec);
 877             pmecorrV         = avx128fma_pmecorrV_f(zeta2);
 878             velec            = _mm_nmacc_ps(pmecorrV,beta,rinv11);
 879             velec            = _mm_mul_ps(qq11,velec);
 880
 881             /* Update potential sum for this i atom from the interaction with this j atom. */
 882             velec            = _mm_andnot_ps(dummy_mask,velec);
 883             velecsum         = _mm_add_ps(velecsum,velec);
 884
 885             fscal            = felec;
 886
 887             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 888
 889              /* Update vectorial force */
 890             fix1             = _mm_macc_ps(dx11,fscal,fix1);
 891             fiy1             = _mm_macc_ps(dy11,fscal,fiy1);
 892             fiz1             = _mm_macc_ps(dz11,fscal,fiz1);
 893
 894             fjx1             = _mm_macc_ps(dx11,fscal,fjx1);
 895             fjy1             = _mm_macc_ps(dy11,fscal,fjy1);
 896             fjz1             = _mm_macc_ps(dz11,fscal,fjz1);
 897
 898             /**************************
 899              * CALCULATE INTERACTIONS *
 900              **************************/
 901
 902             r12              = _mm_mul_ps(rsq12,rinv12);
 903             r12              = _mm_andnot_ps(dummy_mask,r12);
 904
 905             /* EWALD ELECTROSTATICS */
 906
 907             /* Analytical PME correction */
 908             zeta2            = _mm_mul_ps(beta2,rsq12);
 909             rinv3            = _mm_mul_ps(rinvsq12,rinv12);
 910             pmecorrF         = avx128fma_pmecorrF_f(zeta2);
 911             felec            = _mm_macc_ps(pmecorrF,beta3,rinv3);
 912             felec            = _mm_mul_ps(qq12,felec);
 913             pmecorrV         = avx128fma_pmecorrV_f(zeta2);
 914             velec            = _mm_nmacc_ps(pmecorrV,beta,rinv12);
 915             velec            = _mm_mul_ps(qq12,velec);
 916
 917             /* Update potential sum for this i atom from the interaction with this j atom. */
 918             velec            = _mm_andnot_ps(dummy_mask,velec);
 919             velecsum         = _mm_add_ps(velecsum,velec);
 920
 921             fscal            = felec;
 922
 923             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 924
 925              /* Update vectorial force */
 926             fix1             = _mm_macc_ps(dx12,fscal,fix1);
 927             fiy1             = _mm_macc_ps(dy12,fscal,fiy1);
 928             fiz1             = _mm_macc_ps(dz12,fscal,fiz1);
 929
 930             fjx2             = _mm_macc_ps(dx12,fscal,fjx2);
 931             fjy2             = _mm_macc_ps(dy12,fscal,fjy2);
 932             fjz2             = _mm_macc_ps(dz12,fscal,fjz2);
 933
 934             /**************************
 935              * CALCULATE INTERACTIONS *
 936              **************************/
 937
 938             r20              = _mm_mul_ps(rsq20,rinv20);
 939             r20              = _mm_andnot_ps(dummy_mask,r20);
 940
 941             /* EWALD ELECTROSTATICS */
 942
 943             /* Analytical PME correction */
 944             zeta2            = _mm_mul_ps(beta2,rsq20);
 945             rinv3            = _mm_mul_ps(rinvsq20,rinv20);
 946             pmecorrF         = avx128fma_pmecorrF_f(zeta2);
 947             felec            = _mm_macc_ps(pmecorrF,beta3,rinv3);
 948             felec            = _mm_mul_ps(qq20,felec);
 949             pmecorrV         = avx128fma_pmecorrV_f(zeta2);
 950             velec            = _mm_nmacc_ps(pmecorrV,beta,rinv20);
 951             velec            = _mm_mul_ps(qq20,velec);
 952
 953             /* Update potential sum for this i atom from the interaction with this j atom. */
 954             velec            = _mm_andnot_ps(dummy_mask,velec);
 955             velecsum         = _mm_add_ps(velecsum,velec);
 956
 957             fscal            = felec;
 958
 959             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 960
 961              /* Update vectorial force */
 962             fix2             = _mm_macc_ps(dx20,fscal,fix2);
 963             fiy2             = _mm_macc_ps(dy20,fscal,fiy2);
 964             fiz2             = _mm_macc_ps(dz20,fscal,fiz2);
 965
 966             fjx0             = _mm_macc_ps(dx20,fscal,fjx0);
 967             fjy0             = _mm_macc_ps(dy20,fscal,fjy0);
 968             fjz0             = _mm_macc_ps(dz20,fscal,fjz0);
 969
 970             /**************************
 971              * CALCULATE INTERACTIONS *
 972              **************************/
 973
 974             r21              = _mm_mul_ps(rsq21,rinv21);
 975             r21              = _mm_andnot_ps(dummy_mask,r21);
 976
 977             /* EWALD ELECTROSTATICS */
 978
 979             /* Analytical PME correction */
 980             zeta2            = _mm_mul_ps(beta2,rsq21);
 981             rinv3            = _mm_mul_ps(rinvsq21,rinv21);
 982             pmecorrF         = avx128fma_pmecorrF_f(zeta2);
 983             felec            = _mm_macc_ps(pmecorrF,beta3,rinv3);
 984             felec            = _mm_mul_ps(qq21,felec);
 985             pmecorrV         = avx128fma_pmecorrV_f(zeta2);
 986             velec            = _mm_nmacc_ps(pmecorrV,beta,rinv21);
 987             velec            = _mm_mul_ps(qq21,velec);
 988
 989             /* Update potential sum for this i atom from the interaction with this j atom. */
 990             velec            = _mm_andnot_ps(dummy_mask,velec);
 991             velecsum         = _mm_add_ps(velecsum,velec);
 992
 993             fscal            = felec;
 994
 995             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 996
 997              /* Update vectorial force */
 998             fix2             = _mm_macc_ps(dx21,fscal,fix2);
 999             fiy2             = _mm_macc_ps(dy21,fscal,fiy2);
1000             fiz2             = _mm_macc_ps(dz21,fscal,fiz2);
1001
1002             fjx1             = _mm_macc_ps(dx21,fscal,fjx1);
1003             fjy1             = _mm_macc_ps(dy21,fscal,fjy1);
1004             fjz1             = _mm_macc_ps(dz21,fscal,fjz1);
1005
1006             /**************************
1007              * CALCULATE INTERACTIONS *
1008              **************************/
1009
1010             r22              = _mm_mul_ps(rsq22,rinv22);
1011             r22              = _mm_andnot_ps(dummy_mask,r22);
1012
1013             /* EWALD ELECTROSTATICS */
1014
1015             /* Analytical PME correction */
1016             zeta2            = _mm_mul_ps(beta2,rsq22);
1017             rinv3            = _mm_mul_ps(rinvsq22,rinv22);
1018             pmecorrF         = avx128fma_pmecorrF_f(zeta2);
1019             felec            = _mm_macc_ps(pmecorrF,beta3,rinv3);
1020             felec            = _mm_mul_ps(qq22,felec);
1021             pmecorrV         = avx128fma_pmecorrV_f(zeta2);
1022             velec            = _mm_nmacc_ps(pmecorrV,beta,rinv22);
1023             velec            = _mm_mul_ps(qq22,velec);
1024
1025             /* Update potential sum for this i atom from the interaction with this j atom. */
1026             velec            = _mm_andnot_ps(dummy_mask,velec);
1027             velecsum         = _mm_add_ps(velecsum,velec);
1028
1029             fscal            = felec;
1030
1031             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1032
1033              /* Update vectorial force */
1034             fix2             = _mm_macc_ps(dx22,fscal,fix2);
1035             fiy2             = _mm_macc_ps(dy22,fscal,fiy2);
1036             fiz2             = _mm_macc_ps(dz22,fscal,fiz2);
1037
1038             fjx2             = _mm_macc_ps(dx22,fscal,fjx2);
1039             fjy2             = _mm_macc_ps(dy22,fscal,fjy2);
1040             fjz2             = _mm_macc_ps(dz22,fscal,fjz2);
1041
1042             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
1043             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
1044             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
1045             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
1046
1047             gmx_mm_decrement_3rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,
1048                                                    fjx0,fjy0,fjz0,fjx1,fjy1,fjz1,fjx2,fjy2,fjz2);
1049
1050             /* Inner loop uses 282 flops */
1051         }
1052
1053         /* End of innermost loop */
1054
1055         gmx_mm_update_iforce_3atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
1056                                               f+i_coord_offset,fshift+i_shift_offset);
1057
1058         ggid                        = gid[iidx];
1059         /* Update potential energies */
1060         gmx_mm_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid);
1061         gmx_mm_update_1pot_ps(vvdwsum,kernel_data->energygrp_vdw+ggid);
1062
1063         /* Increment number of inner iterations */
1064         inneriter                  += j_index_end - j_index_start;
1065
1066         /* Outer loop uses 20 flops */
1067     }
1068
1069     /* Increment number of outer iterations */
1070     outeriter        += nri;
1071
1072     /* Update outer/inner flops */
1073
1074     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3W3_VF,outeriter*20 + inneriter*282);
1075 }
1076 /*
1077  * Gromacs nonbonded kernel:   nb_kernel_ElecEw_VdwLJ_GeomW3W3_F_avx_128_fma_single
1078  * Electrostatics interaction: Ewald
1079  * VdW interaction:            LennardJones
1080  * Geometry:                   Water3-Water3
1081  * Calculate force/pot:        Force
1082  */
1083 void
1084 nb_kernel_ElecEw_VdwLJ_GeomW3W3_F_avx_128_fma_single
1085                     (t_nblist                    * gmx_restrict       nlist,
1086                      rvec                        * gmx_restrict          xx,
1087                      rvec                        * gmx_restrict          ff,
1088                      struct t_forcerec           * gmx_restrict          fr,
1089                      t_mdatoms                   * gmx_restrict     mdatoms,
1090                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
1091                      t_nrnb                      * gmx_restrict        nrnb)
1092 {
1093     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
1094      * just 0 for non-waters.
1095      * Suffixes A,B,C,D refer to j loop unrolling done with AVX_128, e.g. for the four different
1096      * jnr indices corresponding to data put in the four positions in the SIMD register.
1097      */
1098     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
1099     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
1100     int              jnrA,jnrB,jnrC,jnrD;
1101     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
1102     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
1103     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
1104     real             rcutoff_scalar;
1105     real             *shiftvec,*fshift,*x,*f;
1106     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
1107     real             scratch[4*DIM];
1108     __m128           fscal,rcutoff,rcutoff2,jidxall;
1109     int              vdwioffset0;
1110     __m128           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
1111     int              vdwioffset1;
1112     __m128           ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
1113     int              vdwioffset2;
1114     __m128           ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
1115     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
1116     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
1117     int              vdwjidx1A,vdwjidx1B,vdwjidx1C,vdwjidx1D;
1118     __m128           jx1,jy1,jz1,fjx1,fjy1,fjz1,jq1,isaj1;
1119     int              vdwjidx2A,vdwjidx2B,vdwjidx2C,vdwjidx2D;
1120     __m128           jx2,jy2,jz2,fjx2,fjy2,fjz2,jq2,isaj2;
1121     __m128           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
1122     __m128           dx01,dy01,dz01,rsq01,rinv01,rinvsq01,r01,qq01,c6_01,c12_01;
1123     __m128           dx02,dy02,dz02,rsq02,rinv02,rinvsq02,r02,qq02,c6_02,c12_02;
1124     __m128           dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
1125     __m128           dx11,dy11,dz11,rsq11,rinv11,rinvsq11,r11,qq11,c6_11,c12_11;
1126     __m128           dx12,dy12,dz12,rsq12,rinv12,rinvsq12,r12,qq12,c6_12,c12_12;
1127     __m128           dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
1128     __m128           dx21,dy21,dz21,rsq21,rinv21,rinvsq21,r21,qq21,c6_21,c12_21;
1129     __m128           dx22,dy22,dz22,rsq22,rinv22,rinvsq22,r22,qq22,c6_22,c12_22;
1130     __m128           velec,felec,velecsum,facel,crf,krf,krf2;
1131     real             *charge;
1132     int              nvdwtype;
1133     __m128           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
1134     int              *vdwtype;
1135     real             *vdwparam;
1136     __m128           one_sixth   = _mm_set1_ps(1.0/6.0);
1137     __m128           one_twelfth = _mm_set1_ps(1.0/12.0);
1138     __m128i          ewitab;
1139     __m128           ewtabscale,eweps,twoeweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
1140     __m128           beta,beta2,beta3,zeta2,pmecorrF,pmecorrV,rinv3;
1141     real             *ewtab;
1142     __m128           dummy_mask,cutoff_mask;
1143     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
1144     __m128           one     = _mm_set1_ps(1.0);
1145     __m128           two     = _mm_set1_ps(2.0);
1146     x                = xx[0];
1147     f                = ff[0];
1148
1149     nri              = nlist->nri;
1150     iinr             = nlist->iinr;
1151     jindex           = nlist->jindex;
1152     jjnr             = nlist->jjnr;
1153     shiftidx         = nlist->shift;
1154     gid              = nlist->gid;
1155     shiftvec         = fr->shift_vec[0];
1156     fshift           = fr->fshift[0];
1157     facel            = _mm_set1_ps(fr->ic->epsfac);
1158     charge           = mdatoms->chargeA;
1159     nvdwtype         = fr->ntype;
1160     vdwparam         = fr->nbfp;
1161     vdwtype          = mdatoms->typeA;
1162
1163     sh_ewald         = _mm_set1_ps(fr->ic->sh_ewald);
1164     beta             = _mm_set1_ps(fr->ic->ewaldcoeff_q);
1165     beta2            = _mm_mul_ps(beta,beta);
1166     beta3            = _mm_mul_ps(beta,beta2);
1167     ewtab            = fr->ic->tabq_coul_F;
1168     ewtabscale       = _mm_set1_ps(fr->ic->tabq_scale);
1169     ewtabhalfspace   = _mm_set1_ps(0.5/fr->ic->tabq_scale);
1170
1171     /* Setup water-specific parameters */
1172     inr              = nlist->iinr[0];
1173     iq0              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+0]));
1174     iq1              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+1]));
1175     iq2              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+2]));
1176     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
1177
1178     jq0              = _mm_set1_ps(charge[inr+0]);
1179     jq1              = _mm_set1_ps(charge[inr+1]);
1180     jq2              = _mm_set1_ps(charge[inr+2]);
1181     vdwjidx0A        = 2*vdwtype[inr+0];
1182     qq00             = _mm_mul_ps(iq0,jq0);
1183     c6_00            = _mm_set1_ps(vdwparam[vdwioffset0+vdwjidx0A]);
1184     c12_00           = _mm_set1_ps(vdwparam[vdwioffset0+vdwjidx0A+1]);
1185     qq01             = _mm_mul_ps(iq0,jq1);
1186     qq02             = _mm_mul_ps(iq0,jq2);
1187     qq10             = _mm_mul_ps(iq1,jq0);
1188     qq11             = _mm_mul_ps(iq1,jq1);
1189     qq12             = _mm_mul_ps(iq1,jq2);
1190     qq20             = _mm_mul_ps(iq2,jq0);
1191     qq21             = _mm_mul_ps(iq2,jq1);
1192     qq22             = _mm_mul_ps(iq2,jq2);
1193
1194     /* Avoid stupid compiler warnings */
1195     jnrA = jnrB = jnrC = jnrD = 0;
1196     j_coord_offsetA = 0;
1197     j_coord_offsetB = 0;
1198     j_coord_offsetC = 0;
1199     j_coord_offsetD = 0;
1200
1201     outeriter        = 0;
1202     inneriter        = 0;
1203
1204     for(iidx=0;iidx<4*DIM;iidx++)
1205     {
1206         scratch[iidx] = 0.0;
1207     }
1208
1209     /* Start outer loop over neighborlists */
1210     for(iidx=0; iidx<nri; iidx++)
1211     {
1212         /* Load shift vector for this list */
1213         i_shift_offset   = DIM*shiftidx[iidx];
1214
1215         /* Load limits for loop over neighbors */
1216         j_index_start    = jindex[iidx];
1217         j_index_end      = jindex[iidx+1];
1218
1219         /* Get outer coordinate index */
1220         inr              = iinr[iidx];
1221         i_coord_offset   = DIM*inr;
1222
1223         /* Load i particle coords and add shift vector */
1224         gmx_mm_load_shift_and_3rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
1225                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
1226
1227         fix0             = _mm_setzero_ps();
1228         fiy0             = _mm_setzero_ps();
1229         fiz0             = _mm_setzero_ps();
1230         fix1             = _mm_setzero_ps();
1231         fiy1             = _mm_setzero_ps();
1232         fiz1             = _mm_setzero_ps();
1233         fix2             = _mm_setzero_ps();
1234         fiy2             = _mm_setzero_ps();
1235         fiz2             = _mm_setzero_ps();
1236
1237         /* Start inner kernel loop */
1238         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
1239         {
1240
1241             /* Get j neighbor index, and coordinate index */
1242             jnrA             = jjnr[jidx];
1243             jnrB             = jjnr[jidx+1];
1244             jnrC             = jjnr[jidx+2];
1245             jnrD             = jjnr[jidx+3];
1246             j_coord_offsetA  = DIM*jnrA;
1247             j_coord_offsetB  = DIM*jnrB;
1248             j_coord_offsetC  = DIM*jnrC;
1249             j_coord_offsetD  = DIM*jnrD;
1250
1251             /* load j atom coordinates */
1252             gmx_mm_load_3rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
1253                                               x+j_coord_offsetC,x+j_coord_offsetD,
1254                                               &jx0,&jy0,&jz0,&jx1,&jy1,&jz1,&jx2,&jy2,&jz2);
1255
1256             /* Calculate displacement vector */
1257             dx00             = _mm_sub_ps(ix0,jx0);
1258             dy00             = _mm_sub_ps(iy0,jy0);
1259             dz00             = _mm_sub_ps(iz0,jz0);
1260             dx01             = _mm_sub_ps(ix0,jx1);
1261             dy01             = _mm_sub_ps(iy0,jy1);
1262             dz01             = _mm_sub_ps(iz0,jz1);
1263             dx02             = _mm_sub_ps(ix0,jx2);
1264             dy02             = _mm_sub_ps(iy0,jy2);
1265             dz02             = _mm_sub_ps(iz0,jz2);
1266             dx10             = _mm_sub_ps(ix1,jx0);
1267             dy10             = _mm_sub_ps(iy1,jy0);
1268             dz10             = _mm_sub_ps(iz1,jz0);
1269             dx11             = _mm_sub_ps(ix1,jx1);
1270             dy11             = _mm_sub_ps(iy1,jy1);
1271             dz11             = _mm_sub_ps(iz1,jz1);
1272             dx12             = _mm_sub_ps(ix1,jx2);
1273             dy12             = _mm_sub_ps(iy1,jy2);
1274             dz12             = _mm_sub_ps(iz1,jz2);
1275             dx20             = _mm_sub_ps(ix2,jx0);
1276             dy20             = _mm_sub_ps(iy2,jy0);
1277             dz20             = _mm_sub_ps(iz2,jz0);
1278             dx21             = _mm_sub_ps(ix2,jx1);
1279             dy21             = _mm_sub_ps(iy2,jy1);
1280             dz21             = _mm_sub_ps(iz2,jz1);
1281             dx22             = _mm_sub_ps(ix2,jx2);
1282             dy22             = _mm_sub_ps(iy2,jy2);
1283             dz22             = _mm_sub_ps(iz2,jz2);
1284
1285             /* Calculate squared distance and things based on it */
1286             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
1287             rsq01            = gmx_mm_calc_rsq_ps(dx01,dy01,dz01);
1288             rsq02            = gmx_mm_calc_rsq_ps(dx02,dy02,dz02);
1289             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
1290             rsq11            = gmx_mm_calc_rsq_ps(dx11,dy11,dz11);
1291             rsq12            = gmx_mm_calc_rsq_ps(dx12,dy12,dz12);
1292             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
1293             rsq21            = gmx_mm_calc_rsq_ps(dx21,dy21,dz21);
1294             rsq22            = gmx_mm_calc_rsq_ps(dx22,dy22,dz22);
1295
1296             rinv00           = avx128fma_invsqrt_f(rsq00);
1297             rinv01           = avx128fma_invsqrt_f(rsq01);
1298             rinv02           = avx128fma_invsqrt_f(rsq02);
1299             rinv10           = avx128fma_invsqrt_f(rsq10);
1300             rinv11           = avx128fma_invsqrt_f(rsq11);
1301             rinv12           = avx128fma_invsqrt_f(rsq12);
1302             rinv20           = avx128fma_invsqrt_f(rsq20);
1303             rinv21           = avx128fma_invsqrt_f(rsq21);
1304             rinv22           = avx128fma_invsqrt_f(rsq22);
1305
1306             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
1307             rinvsq01         = _mm_mul_ps(rinv01,rinv01);
1308             rinvsq02         = _mm_mul_ps(rinv02,rinv02);
1309             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
1310             rinvsq11         = _mm_mul_ps(rinv11,rinv11);
1311             rinvsq12         = _mm_mul_ps(rinv12,rinv12);
1312             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
1313             rinvsq21         = _mm_mul_ps(rinv21,rinv21);
1314             rinvsq22         = _mm_mul_ps(rinv22,rinv22);
1315
1316             fjx0             = _mm_setzero_ps();
1317             fjy0             = _mm_setzero_ps();
1318             fjz0             = _mm_setzero_ps();
1319             fjx1             = _mm_setzero_ps();
1320             fjy1             = _mm_setzero_ps();
1321             fjz1             = _mm_setzero_ps();
1322             fjx2             = _mm_setzero_ps();
1323             fjy2             = _mm_setzero_ps();
1324             fjz2             = _mm_setzero_ps();
1325
1326             /**************************
1327              * CALCULATE INTERACTIONS *
1328              **************************/
1329
1330             r00              = _mm_mul_ps(rsq00,rinv00);
1331
1332             /* EWALD ELECTROSTATICS */
1333
1334             /* Analytical PME correction */
1335             zeta2            = _mm_mul_ps(beta2,rsq00);
1336             rinv3            = _mm_mul_ps(rinvsq00,rinv00);
1337             pmecorrF         = avx128fma_pmecorrF_f(zeta2);
1338             felec            = _mm_macc_ps(pmecorrF,beta3,rinv3);
1339             felec            = _mm_mul_ps(qq00,felec);
1340
1341             /* LENNARD-JONES DISPERSION/REPULSION */
1342
1343             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
1344             fvdw             = _mm_mul_ps(_mm_msub_ps(c12_00,rinvsix,c6_00),_mm_mul_ps(rinvsix,rinvsq00));
1345
1346             fscal            = _mm_add_ps(felec,fvdw);
1347
1348              /* Update vectorial force */
1349             fix0             = _mm_macc_ps(dx00,fscal,fix0);
1350             fiy0             = _mm_macc_ps(dy00,fscal,fiy0);
1351             fiz0             = _mm_macc_ps(dz00,fscal,fiz0);
1352
1353             fjx0             = _mm_macc_ps(dx00,fscal,fjx0);
1354             fjy0             = _mm_macc_ps(dy00,fscal,fjy0);
1355             fjz0             = _mm_macc_ps(dz00,fscal,fjz0);
1356
1357             /**************************
1358              * CALCULATE INTERACTIONS *
1359              **************************/
1360
1361             r01              = _mm_mul_ps(rsq01,rinv01);
1362
1363             /* EWALD ELECTROSTATICS */
1364
1365             /* Analytical PME correction */
1366             zeta2            = _mm_mul_ps(beta2,rsq01);
1367             rinv3            = _mm_mul_ps(rinvsq01,rinv01);
1368             pmecorrF         = avx128fma_pmecorrF_f(zeta2);
1369             felec            = _mm_macc_ps(pmecorrF,beta3,rinv3);
1370             felec            = _mm_mul_ps(qq01,felec);
1371
1372             fscal            = felec;
1373
1374              /* Update vectorial force */
1375             fix0             = _mm_macc_ps(dx01,fscal,fix0);
1376             fiy0             = _mm_macc_ps(dy01,fscal,fiy0);
1377             fiz0             = _mm_macc_ps(dz01,fscal,fiz0);
1378
1379             fjx1             = _mm_macc_ps(dx01,fscal,fjx1);
1380             fjy1             = _mm_macc_ps(dy01,fscal,fjy1);
1381             fjz1             = _mm_macc_ps(dz01,fscal,fjz1);
1382
1383             /**************************
1384              * CALCULATE INTERACTIONS *
1385              **************************/
1386
1387             r02              = _mm_mul_ps(rsq02,rinv02);
1388
1389             /* EWALD ELECTROSTATICS */
1390
1391             /* Analytical PME correction */
1392             zeta2            = _mm_mul_ps(beta2,rsq02);
1393             rinv3            = _mm_mul_ps(rinvsq02,rinv02);
1394             pmecorrF         = avx128fma_pmecorrF_f(zeta2);
1395             felec            = _mm_macc_ps(pmecorrF,beta3,rinv3);
1396             felec            = _mm_mul_ps(qq02,felec);
1397
1398             fscal            = felec;
1399
1400              /* Update vectorial force */
1401             fix0             = _mm_macc_ps(dx02,fscal,fix0);
1402             fiy0             = _mm_macc_ps(dy02,fscal,fiy0);
1403             fiz0             = _mm_macc_ps(dz02,fscal,fiz0);
1404
1405             fjx2             = _mm_macc_ps(dx02,fscal,fjx2);
1406             fjy2             = _mm_macc_ps(dy02,fscal,fjy2);
1407             fjz2             = _mm_macc_ps(dz02,fscal,fjz2);
1408
1409             /**************************
1410              * CALCULATE INTERACTIONS *
1411              **************************/
1412
1413             r10              = _mm_mul_ps(rsq10,rinv10);
1414
1415             /* EWALD ELECTROSTATICS */
1416
1417             /* Analytical PME correction */
1418             zeta2            = _mm_mul_ps(beta2,rsq10);
1419             rinv3            = _mm_mul_ps(rinvsq10,rinv10);
1420             pmecorrF         = avx128fma_pmecorrF_f(zeta2);
1421             felec            = _mm_macc_ps(pmecorrF,beta3,rinv3);
1422             felec            = _mm_mul_ps(qq10,felec);
1423
1424             fscal            = felec;
1425
1426              /* Update vectorial force */
1427             fix1             = _mm_macc_ps(dx10,fscal,fix1);
1428             fiy1             = _mm_macc_ps(dy10,fscal,fiy1);
1429             fiz1             = _mm_macc_ps(dz10,fscal,fiz1);
1430
1431             fjx0             = _mm_macc_ps(dx10,fscal,fjx0);
1432             fjy0             = _mm_macc_ps(dy10,fscal,fjy0);
1433             fjz0             = _mm_macc_ps(dz10,fscal,fjz0);
1434
1435             /**************************
1436              * CALCULATE INTERACTIONS *
1437              **************************/
1438
1439             r11              = _mm_mul_ps(rsq11,rinv11);
1440
1441             /* EWALD ELECTROSTATICS */
1442
1443             /* Analytical PME correction */
1444             zeta2            = _mm_mul_ps(beta2,rsq11);
1445             rinv3            = _mm_mul_ps(rinvsq11,rinv11);
1446             pmecorrF         = avx128fma_pmecorrF_f(zeta2);
1447             felec            = _mm_macc_ps(pmecorrF,beta3,rinv3);
1448             felec            = _mm_mul_ps(qq11,felec);
1449
1450             fscal            = felec;
1451
1452              /* Update vectorial force */
1453             fix1             = _mm_macc_ps(dx11,fscal,fix1);
1454             fiy1             = _mm_macc_ps(dy11,fscal,fiy1);
1455             fiz1             = _mm_macc_ps(dz11,fscal,fiz1);
1456
1457             fjx1             = _mm_macc_ps(dx11,fscal,fjx1);
1458             fjy1             = _mm_macc_ps(dy11,fscal,fjy1);
1459             fjz1             = _mm_macc_ps(dz11,fscal,fjz1);
1460
1461             /**************************
1462              * CALCULATE INTERACTIONS *
1463              **************************/
1464
1465             r12              = _mm_mul_ps(rsq12,rinv12);
1466
1467             /* EWALD ELECTROSTATICS */
1468
1469             /* Analytical PME correction */
1470             zeta2            = _mm_mul_ps(beta2,rsq12);
1471             rinv3            = _mm_mul_ps(rinvsq12,rinv12);
1472             pmecorrF         = avx128fma_pmecorrF_f(zeta2);
1473             felec            = _mm_macc_ps(pmecorrF,beta3,rinv3);
1474             felec            = _mm_mul_ps(qq12,felec);
1475
1476             fscal            = felec;
1477
1478              /* Update vectorial force */
1479             fix1             = _mm_macc_ps(dx12,fscal,fix1);
1480             fiy1             = _mm_macc_ps(dy12,fscal,fiy1);
1481             fiz1             = _mm_macc_ps(dz12,fscal,fiz1);
1482
1483             fjx2             = _mm_macc_ps(dx12,fscal,fjx2);
1484             fjy2             = _mm_macc_ps(dy12,fscal,fjy2);
1485             fjz2             = _mm_macc_ps(dz12,fscal,fjz2);
1486
1487             /**************************
1488              * CALCULATE INTERACTIONS *
1489              **************************/
1490
1491             r20              = _mm_mul_ps(rsq20,rinv20);
1492
1493             /* EWALD ELECTROSTATICS */
1494
1495             /* Analytical PME correction */
1496             zeta2            = _mm_mul_ps(beta2,rsq20);
1497             rinv3            = _mm_mul_ps(rinvsq20,rinv20);
1498             pmecorrF         = avx128fma_pmecorrF_f(zeta2);
1499             felec            = _mm_macc_ps(pmecorrF,beta3,rinv3);
1500             felec            = _mm_mul_ps(qq20,felec);
1501
1502             fscal            = felec;
1503
1504              /* Update vectorial force */
1505             fix2             = _mm_macc_ps(dx20,fscal,fix2);
1506             fiy2             = _mm_macc_ps(dy20,fscal,fiy2);
1507             fiz2             = _mm_macc_ps(dz20,fscal,fiz2);
1508
1509             fjx0             = _mm_macc_ps(dx20,fscal,fjx0);
1510             fjy0             = _mm_macc_ps(dy20,fscal,fjy0);
1511             fjz0             = _mm_macc_ps(dz20,fscal,fjz0);
1512
1513             /**************************
1514              * CALCULATE INTERACTIONS *
1515              **************************/
1516
1517             r21              = _mm_mul_ps(rsq21,rinv21);
1518
1519             /* EWALD ELECTROSTATICS */
1520
1521             /* Analytical PME correction */
1522             zeta2            = _mm_mul_ps(beta2,rsq21);
1523             rinv3            = _mm_mul_ps(rinvsq21,rinv21);
1524             pmecorrF         = avx128fma_pmecorrF_f(zeta2);
1525             felec            = _mm_macc_ps(pmecorrF,beta3,rinv3);
1526             felec            = _mm_mul_ps(qq21,felec);
1527
1528             fscal            = felec;
1529
1530              /* Update vectorial force */
1531             fix2             = _mm_macc_ps(dx21,fscal,fix2);
1532             fiy2             = _mm_macc_ps(dy21,fscal,fiy2);
1533             fiz2             = _mm_macc_ps(dz21,fscal,fiz2);
1534
1535             fjx1             = _mm_macc_ps(dx21,fscal,fjx1);
1536             fjy1             = _mm_macc_ps(dy21,fscal,fjy1);
1537             fjz1             = _mm_macc_ps(dz21,fscal,fjz1);
1538
1539             /**************************
1540              * CALCULATE INTERACTIONS *
1541              **************************/
1542
1543             r22              = _mm_mul_ps(rsq22,rinv22);
1544
1545             /* EWALD ELECTROSTATICS */
1546
1547             /* Analytical PME correction */
1548             zeta2            = _mm_mul_ps(beta2,rsq22);
1549             rinv3            = _mm_mul_ps(rinvsq22,rinv22);
1550             pmecorrF         = avx128fma_pmecorrF_f(zeta2);
1551             felec            = _mm_macc_ps(pmecorrF,beta3,rinv3);
1552             felec            = _mm_mul_ps(qq22,felec);
1553
1554             fscal            = felec;
1555
1556              /* Update vectorial force */
1557             fix2             = _mm_macc_ps(dx22,fscal,fix2);
1558             fiy2             = _mm_macc_ps(dy22,fscal,fiy2);
1559             fiz2             = _mm_macc_ps(dz22,fscal,fiz2);
1560
1561             fjx2             = _mm_macc_ps(dx22,fscal,fjx2);
1562             fjy2             = _mm_macc_ps(dy22,fscal,fjy2);
1563             fjz2             = _mm_macc_ps(dz22,fscal,fjz2);
1564
1565             fjptrA             = f+j_coord_offsetA;
1566             fjptrB             = f+j_coord_offsetB;
1567             fjptrC             = f+j_coord_offsetC;
1568             fjptrD             = f+j_coord_offsetD;
1569
1570             gmx_mm_decrement_3rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,
1571                                                    fjx0,fjy0,fjz0,fjx1,fjy1,fjz1,fjx2,fjy2,fjz2);
1572
1573             /* Inner loop uses 259 flops */
1574         }
1575
1576         if(jidx<j_index_end)
1577         {
1578
1579             /* Get j neighbor index, and coordinate index */
1580             jnrlistA         = jjnr[jidx];
1581             jnrlistB         = jjnr[jidx+1];
1582             jnrlistC         = jjnr[jidx+2];
1583             jnrlistD         = jjnr[jidx+3];
1584             /* Sign of each element will be negative for non-real atoms.
1585              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
1586              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
1587              */
1588             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
1589             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
1590             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
1591             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
1592             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
1593             j_coord_offsetA  = DIM*jnrA;
1594             j_coord_offsetB  = DIM*jnrB;
1595             j_coord_offsetC  = DIM*jnrC;
1596             j_coord_offsetD  = DIM*jnrD;
1597
1598             /* load j atom coordinates */
1599             gmx_mm_load_3rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
1600                                               x+j_coord_offsetC,x+j_coord_offsetD,
1601                                               &jx0,&jy0,&jz0,&jx1,&jy1,&jz1,&jx2,&jy2,&jz2);
1602
1603             /* Calculate displacement vector */
1604             dx00             = _mm_sub_ps(ix0,jx0);
1605             dy00             = _mm_sub_ps(iy0,jy0);
1606             dz00             = _mm_sub_ps(iz0,jz0);
1607             dx01             = _mm_sub_ps(ix0,jx1);
1608             dy01             = _mm_sub_ps(iy0,jy1);
1609             dz01             = _mm_sub_ps(iz0,jz1);
1610             dx02             = _mm_sub_ps(ix0,jx2);
1611             dy02             = _mm_sub_ps(iy0,jy2);
1612             dz02             = _mm_sub_ps(iz0,jz2);
1613             dx10             = _mm_sub_ps(ix1,jx0);
1614             dy10             = _mm_sub_ps(iy1,jy0);
1615             dz10             = _mm_sub_ps(iz1,jz0);
1616             dx11             = _mm_sub_ps(ix1,jx1);
1617             dy11             = _mm_sub_ps(iy1,jy1);
1618             dz11             = _mm_sub_ps(iz1,jz1);
1619             dx12             = _mm_sub_ps(ix1,jx2);
1620             dy12             = _mm_sub_ps(iy1,jy2);
1621             dz12             = _mm_sub_ps(iz1,jz2);
1622             dx20             = _mm_sub_ps(ix2,jx0);
1623             dy20             = _mm_sub_ps(iy2,jy0);
1624             dz20             = _mm_sub_ps(iz2,jz0);
1625             dx21             = _mm_sub_ps(ix2,jx1);
1626             dy21             = _mm_sub_ps(iy2,jy1);
1627             dz21             = _mm_sub_ps(iz2,jz1);
1628             dx22             = _mm_sub_ps(ix2,jx2);
1629             dy22             = _mm_sub_ps(iy2,jy2);
1630             dz22             = _mm_sub_ps(iz2,jz2);
1631
1632             /* Calculate squared distance and things based on it */
1633             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
1634             rsq01            = gmx_mm_calc_rsq_ps(dx01,dy01,dz01);
1635             rsq02            = gmx_mm_calc_rsq_ps(dx02,dy02,dz02);
1636             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
1637             rsq11            = gmx_mm_calc_rsq_ps(dx11,dy11,dz11);
1638             rsq12            = gmx_mm_calc_rsq_ps(dx12,dy12,dz12);
1639             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
1640             rsq21            = gmx_mm_calc_rsq_ps(dx21,dy21,dz21);
1641             rsq22            = gmx_mm_calc_rsq_ps(dx22,dy22,dz22);
1642
1643             rinv00           = avx128fma_invsqrt_f(rsq00);
1644             rinv01           = avx128fma_invsqrt_f(rsq01);
1645             rinv02           = avx128fma_invsqrt_f(rsq02);
1646             rinv10           = avx128fma_invsqrt_f(rsq10);
1647             rinv11           = avx128fma_invsqrt_f(rsq11);
1648             rinv12           = avx128fma_invsqrt_f(rsq12);
1649             rinv20           = avx128fma_invsqrt_f(rsq20);
1650             rinv21           = avx128fma_invsqrt_f(rsq21);
1651             rinv22           = avx128fma_invsqrt_f(rsq22);
1652
1653             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
1654             rinvsq01         = _mm_mul_ps(rinv01,rinv01);
1655             rinvsq02         = _mm_mul_ps(rinv02,rinv02);
1656             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
1657             rinvsq11         = _mm_mul_ps(rinv11,rinv11);
1658             rinvsq12         = _mm_mul_ps(rinv12,rinv12);
1659             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
1660             rinvsq21         = _mm_mul_ps(rinv21,rinv21);
1661             rinvsq22         = _mm_mul_ps(rinv22,rinv22);
1662
1663             fjx0             = _mm_setzero_ps();
1664             fjy0             = _mm_setzero_ps();
1665             fjz0             = _mm_setzero_ps();
1666             fjx1             = _mm_setzero_ps();
1667             fjy1             = _mm_setzero_ps();
1668             fjz1             = _mm_setzero_ps();
1669             fjx2             = _mm_setzero_ps();
1670             fjy2             = _mm_setzero_ps();
1671             fjz2             = _mm_setzero_ps();
1672
1673             /**************************
1674              * CALCULATE INTERACTIONS *
1675              **************************/
1676
1677             r00              = _mm_mul_ps(rsq00,rinv00);
1678             r00              = _mm_andnot_ps(dummy_mask,r00);
1679
1680             /* EWALD ELECTROSTATICS */
1681
1682             /* Analytical PME correction */
1683             zeta2            = _mm_mul_ps(beta2,rsq00);
1684             rinv3            = _mm_mul_ps(rinvsq00,rinv00);
1685             pmecorrF         = avx128fma_pmecorrF_f(zeta2);
1686             felec            = _mm_macc_ps(pmecorrF,beta3,rinv3);
1687             felec            = _mm_mul_ps(qq00,felec);
1688
1689             /* LENNARD-JONES DISPERSION/REPULSION */
1690
1691             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
1692             fvdw             = _mm_mul_ps(_mm_msub_ps(c12_00,rinvsix,c6_00),_mm_mul_ps(rinvsix,rinvsq00));
1693
1694             fscal            = _mm_add_ps(felec,fvdw);
1695
1696             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1697
1698              /* Update vectorial force */
1699             fix0             = _mm_macc_ps(dx00,fscal,fix0);
1700             fiy0             = _mm_macc_ps(dy00,fscal,fiy0);
1701             fiz0             = _mm_macc_ps(dz00,fscal,fiz0);
1702
1703             fjx0             = _mm_macc_ps(dx00,fscal,fjx0);
1704             fjy0             = _mm_macc_ps(dy00,fscal,fjy0);
1705             fjz0             = _mm_macc_ps(dz00,fscal,fjz0);
1706
1707             /**************************
1708              * CALCULATE INTERACTIONS *
1709              **************************/
1710
1711             r01              = _mm_mul_ps(rsq01,rinv01);
1712             r01              = _mm_andnot_ps(dummy_mask,r01);
1713
1714             /* EWALD ELECTROSTATICS */
1715
1716             /* Analytical PME correction */
1717             zeta2            = _mm_mul_ps(beta2,rsq01);
1718             rinv3            = _mm_mul_ps(rinvsq01,rinv01);
1719             pmecorrF         = avx128fma_pmecorrF_f(zeta2);
1720             felec            = _mm_macc_ps(pmecorrF,beta3,rinv3);
1721             felec            = _mm_mul_ps(qq01,felec);
1722
1723             fscal            = felec;
1724
1725             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1726
1727              /* Update vectorial force */
1728             fix0             = _mm_macc_ps(dx01,fscal,fix0);
1729             fiy0             = _mm_macc_ps(dy01,fscal,fiy0);
1730             fiz0             = _mm_macc_ps(dz01,fscal,fiz0);
1731
1732             fjx1             = _mm_macc_ps(dx01,fscal,fjx1);
1733             fjy1             = _mm_macc_ps(dy01,fscal,fjy1);
1734             fjz1             = _mm_macc_ps(dz01,fscal,fjz1);
1735
1736             /**************************
1737              * CALCULATE INTERACTIONS *
1738              **************************/
1739
1740             r02              = _mm_mul_ps(rsq02,rinv02);
1741             r02              = _mm_andnot_ps(dummy_mask,r02);
1742
1743             /* EWALD ELECTROSTATICS */
1744
1745             /* Analytical PME correction */
1746             zeta2            = _mm_mul_ps(beta2,rsq02);
1747             rinv3            = _mm_mul_ps(rinvsq02,rinv02);
1748             pmecorrF         = avx128fma_pmecorrF_f(zeta2);
1749             felec            = _mm_macc_ps(pmecorrF,beta3,rinv3);
1750             felec            = _mm_mul_ps(qq02,felec);
1751
1752             fscal            = felec;
1753
1754             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1755
1756              /* Update vectorial force */
1757             fix0             = _mm_macc_ps(dx02,fscal,fix0);
1758             fiy0             = _mm_macc_ps(dy02,fscal,fiy0);
1759             fiz0             = _mm_macc_ps(dz02,fscal,fiz0);
1760
1761             fjx2             = _mm_macc_ps(dx02,fscal,fjx2);
1762             fjy2             = _mm_macc_ps(dy02,fscal,fjy2);
1763             fjz2             = _mm_macc_ps(dz02,fscal,fjz2);
1764
1765             /**************************
1766              * CALCULATE INTERACTIONS *
1767              **************************/
1768
1769             r10              = _mm_mul_ps(rsq10,rinv10);
1770             r10              = _mm_andnot_ps(dummy_mask,r10);
1771
1772             /* EWALD ELECTROSTATICS */
1773
1774             /* Analytical PME correction */
1775             zeta2            = _mm_mul_ps(beta2,rsq10);
1776             rinv3            = _mm_mul_ps(rinvsq10,rinv10);
1777             pmecorrF         = avx128fma_pmecorrF_f(zeta2);
1778             felec            = _mm_macc_ps(pmecorrF,beta3,rinv3);
1779             felec            = _mm_mul_ps(qq10,felec);
1780
1781             fscal            = felec;
1782
1783             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1784
1785              /* Update vectorial force */
1786             fix1             = _mm_macc_ps(dx10,fscal,fix1);
1787             fiy1             = _mm_macc_ps(dy10,fscal,fiy1);
1788             fiz1             = _mm_macc_ps(dz10,fscal,fiz1);
1789
1790             fjx0             = _mm_macc_ps(dx10,fscal,fjx0);
1791             fjy0             = _mm_macc_ps(dy10,fscal,fjy0);
1792             fjz0             = _mm_macc_ps(dz10,fscal,fjz0);
1793
1794             /**************************
1795              * CALCULATE INTERACTIONS *
1796              **************************/
1797
1798             r11              = _mm_mul_ps(rsq11,rinv11);
1799             r11              = _mm_andnot_ps(dummy_mask,r11);
1800
1801             /* EWALD ELECTROSTATICS */
1802
1803             /* Analytical PME correction */
1804             zeta2            = _mm_mul_ps(beta2,rsq11);
1805             rinv3            = _mm_mul_ps(rinvsq11,rinv11);
1806             pmecorrF         = avx128fma_pmecorrF_f(zeta2);
1807             felec            = _mm_macc_ps(pmecorrF,beta3,rinv3);
1808             felec            = _mm_mul_ps(qq11,felec);
1809
1810             fscal            = felec;
1811
1812             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1813
1814              /* Update vectorial force */
1815             fix1             = _mm_macc_ps(dx11,fscal,fix1);
1816             fiy1             = _mm_macc_ps(dy11,fscal,fiy1);
1817             fiz1             = _mm_macc_ps(dz11,fscal,fiz1);
1818
1819             fjx1             = _mm_macc_ps(dx11,fscal,fjx1);
1820             fjy1             = _mm_macc_ps(dy11,fscal,fjy1);
1821             fjz1             = _mm_macc_ps(dz11,fscal,fjz1);
1822
1823             /**************************
1824              * CALCULATE INTERACTIONS *
1825              **************************/
1826
1827             r12              = _mm_mul_ps(rsq12,rinv12);
1828             r12              = _mm_andnot_ps(dummy_mask,r12);
1829
1830             /* EWALD ELECTROSTATICS */
1831
1832             /* Analytical PME correction */
1833             zeta2            = _mm_mul_ps(beta2,rsq12);
1834             rinv3            = _mm_mul_ps(rinvsq12,rinv12);
1835             pmecorrF         = avx128fma_pmecorrF_f(zeta2);
1836             felec            = _mm_macc_ps(pmecorrF,beta3,rinv3);
1837             felec            = _mm_mul_ps(qq12,felec);
1838
1839             fscal            = felec;
1840
1841             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1842
1843              /* Update vectorial force */
1844             fix1             = _mm_macc_ps(dx12,fscal,fix1);
1845             fiy1             = _mm_macc_ps(dy12,fscal,fiy1);
1846             fiz1             = _mm_macc_ps(dz12,fscal,fiz1);
1847
1848             fjx2             = _mm_macc_ps(dx12,fscal,fjx2);
1849             fjy2             = _mm_macc_ps(dy12,fscal,fjy2);
1850             fjz2             = _mm_macc_ps(dz12,fscal,fjz2);
1851
1852             /**************************
1853              * CALCULATE INTERACTIONS *
1854              **************************/
1855
1856             r20              = _mm_mul_ps(rsq20,rinv20);
1857             r20              = _mm_andnot_ps(dummy_mask,r20);
1858
1859             /* EWALD ELECTROSTATICS */
1860
1861             /* Analytical PME correction */
1862             zeta2            = _mm_mul_ps(beta2,rsq20);
1863             rinv3            = _mm_mul_ps(rinvsq20,rinv20);
1864             pmecorrF         = avx128fma_pmecorrF_f(zeta2);
1865             felec            = _mm_macc_ps(pmecorrF,beta3,rinv3);
1866             felec            = _mm_mul_ps(qq20,felec);
1867
1868             fscal            = felec;
1869
1870             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1871
1872              /* Update vectorial force */
1873             fix2             = _mm_macc_ps(dx20,fscal,fix2);
1874             fiy2             = _mm_macc_ps(dy20,fscal,fiy2);
1875             fiz2             = _mm_macc_ps(dz20,fscal,fiz2);
1876
1877             fjx0             = _mm_macc_ps(dx20,fscal,fjx0);
1878             fjy0             = _mm_macc_ps(dy20,fscal,fjy0);
1879             fjz0             = _mm_macc_ps(dz20,fscal,fjz0);
1880
1881             /**************************
1882              * CALCULATE INTERACTIONS *
1883              **************************/
1884
1885             r21              = _mm_mul_ps(rsq21,rinv21);
1886             r21              = _mm_andnot_ps(dummy_mask,r21);
1887
1888             /* EWALD ELECTROSTATICS */
1889
1890             /* Analytical PME correction */
1891             zeta2            = _mm_mul_ps(beta2,rsq21);
1892             rinv3            = _mm_mul_ps(rinvsq21,rinv21);
1893             pmecorrF         = avx128fma_pmecorrF_f(zeta2);
1894             felec            = _mm_macc_ps(pmecorrF,beta3,rinv3);
1895             felec            = _mm_mul_ps(qq21,felec);
1896
1897             fscal            = felec;
1898
1899             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1900
1901              /* Update vectorial force */
1902             fix2             = _mm_macc_ps(dx21,fscal,fix2);
1903             fiy2             = _mm_macc_ps(dy21,fscal,fiy2);
1904             fiz2             = _mm_macc_ps(dz21,fscal,fiz2);
1905
1906             fjx1             = _mm_macc_ps(dx21,fscal,fjx1);
1907             fjy1             = _mm_macc_ps(dy21,fscal,fjy1);
1908             fjz1             = _mm_macc_ps(dz21,fscal,fjz1);
1909
1910             /**************************
1911              * CALCULATE INTERACTIONS *
1912              **************************/
1913
1914             r22              = _mm_mul_ps(rsq22,rinv22);
1915             r22              = _mm_andnot_ps(dummy_mask,r22);
1916
1917             /* EWALD ELECTROSTATICS */
1918
1919             /* Analytical PME correction */
1920             zeta2            = _mm_mul_ps(beta2,rsq22);
1921             rinv3            = _mm_mul_ps(rinvsq22,rinv22);
1922             pmecorrF         = avx128fma_pmecorrF_f(zeta2);
1923             felec            = _mm_macc_ps(pmecorrF,beta3,rinv3);
1924             felec            = _mm_mul_ps(qq22,felec);
1925
1926             fscal            = felec;
1927
1928             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1929
1930              /* Update vectorial force */
1931             fix2             = _mm_macc_ps(dx22,fscal,fix2);
1932             fiy2             = _mm_macc_ps(dy22,fscal,fiy2);
1933             fiz2             = _mm_macc_ps(dz22,fscal,fiz2);
1934
1935             fjx2             = _mm_macc_ps(dx22,fscal,fjx2);
1936             fjy2             = _mm_macc_ps(dy22,fscal,fjy2);
1937             fjz2             = _mm_macc_ps(dz22,fscal,fjz2);
1938
1939             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
1940             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
1941             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
1942             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
1943
1944             gmx_mm_decrement_3rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,
1945                                                    fjx0,fjy0,fjz0,fjx1,fjy1,fjz1,fjx2,fjy2,fjz2);
1946
1947             /* Inner loop uses 268 flops */
1948         }
1949
1950         /* End of innermost loop */
1951
1952         gmx_mm_update_iforce_3atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
1953                                               f+i_coord_offset,fshift+i_shift_offset);
1954
1955         /* Increment number of inner iterations */
1956         inneriter                  += j_index_end - j_index_start;
1957
1958         /* Outer loop uses 18 flops */
1959     }
1960
1961     /* Increment number of outer iterations */
1962     outeriter        += nri;
1963
1964     /* Update outer/inner flops */
1965
1966     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3W3_F,outeriter*18 + inneriter*268);
1967 }