src/gromacs/gmxlib/nonbonded/nb_kernel_avx_128_fma_single/nb_kernel_ElecEw_VdwLJEw_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_VdwLJEw_GeomW3W3_VF_avx_128_fma_single
  51  * Electrostatics interaction: Ewald
  52  * VdW interaction:            LJEwald
  53  * Geometry:                   Water3-Water3
  54  * Calculate force/pot:        PotentialAndForce
  55  */
  56 void
  57 nb_kernel_ElecEw_VdwLJEw_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     __m128           c6grid_00;
 112     __m128           c6grid_01;
 113     __m128           c6grid_02;
 114     __m128           c6grid_10;
 115     __m128           c6grid_11;
 116     __m128           c6grid_12;
 117     __m128           c6grid_20;
 118     __m128           c6grid_21;
 119     __m128           c6grid_22;
 120     real             *vdwgridparam;
 121     __m128           ewclj,ewclj2,ewclj6,ewcljrsq,poly,exponent,f6A,f6B,sh_lj_ewald;
 122     __m128           one_half = _mm_set1_ps(0.5);
 123     __m128           minus_one = _mm_set1_ps(-1.0);
 124     __m128i          ewitab;
 125     __m128           ewtabscale,eweps,twoeweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
 126     __m128           beta,beta2,beta3,zeta2,pmecorrF,pmecorrV,rinv3;
 127     real             *ewtab;
 128     __m128           dummy_mask,cutoff_mask;
 129     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
 130     __m128           one     = _mm_set1_ps(1.0);
 131     __m128           two     = _mm_set1_ps(2.0);
 132     x                = xx[0];
 133     f                = ff[0];
 134
 135     nri              = nlist->nri;
 136     iinr             = nlist->iinr;
 137     jindex           = nlist->jindex;
 138     jjnr             = nlist->jjnr;
 139     shiftidx         = nlist->shift;
 140     gid              = nlist->gid;
 141     shiftvec         = fr->shift_vec[0];
 142     fshift           = fr->fshift[0];
 143     facel            = _mm_set1_ps(fr->ic->epsfac);
 144     charge           = mdatoms->chargeA;
 145     nvdwtype         = fr->ntype;
 146     vdwparam         = fr->nbfp;
 147     vdwtype          = mdatoms->typeA;
 148     vdwgridparam     = fr->ljpme_c6grid;
 149     sh_lj_ewald      = _mm_set1_ps(fr->ic->sh_lj_ewald);
 150     ewclj            = _mm_set1_ps(fr->ic->ewaldcoeff_lj);
 151     ewclj2           = _mm_mul_ps(minus_one,_mm_mul_ps(ewclj,ewclj));
 152
 153     sh_ewald         = _mm_set1_ps(fr->ic->sh_ewald);
 154     beta             = _mm_set1_ps(fr->ic->ewaldcoeff_q);
 155     beta2            = _mm_mul_ps(beta,beta);
 156     beta3            = _mm_mul_ps(beta,beta2);
 157     ewtab            = fr->ic->tabq_coul_FDV0;
 158     ewtabscale       = _mm_set1_ps(fr->ic->tabq_scale);
 159     ewtabhalfspace   = _mm_set1_ps(0.5/fr->ic->tabq_scale);
 160
 161     /* Setup water-specific parameters */
 162     inr              = nlist->iinr[0];
 163     iq0              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+0]));
 164     iq1              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+1]));
 165     iq2              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+2]));
 166     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 167
 168     jq0              = _mm_set1_ps(charge[inr+0]);
 169     jq1              = _mm_set1_ps(charge[inr+1]);
 170     jq2              = _mm_set1_ps(charge[inr+2]);
 171     vdwjidx0A        = 2*vdwtype[inr+0];
 172     qq00             = _mm_mul_ps(iq0,jq0);
 173     c6_00            = _mm_set1_ps(vdwparam[vdwioffset0+vdwjidx0A]);
 174     c12_00           = _mm_set1_ps(vdwparam[vdwioffset0+vdwjidx0A+1]);
 175     c6grid_00        = _mm_set1_ps(vdwgridparam[vdwioffset0+vdwjidx0A]);
 176     qq01             = _mm_mul_ps(iq0,jq1);
 177     qq02             = _mm_mul_ps(iq0,jq2);
 178     qq10             = _mm_mul_ps(iq1,jq0);
 179     qq11             = _mm_mul_ps(iq1,jq1);
 180     qq12             = _mm_mul_ps(iq1,jq2);
 181     qq20             = _mm_mul_ps(iq2,jq0);
 182     qq21             = _mm_mul_ps(iq2,jq1);
 183     qq22             = _mm_mul_ps(iq2,jq2);
 184
 185     /* Avoid stupid compiler warnings */
 186     jnrA = jnrB = jnrC = jnrD = 0;
 187     j_coord_offsetA = 0;
 188     j_coord_offsetB = 0;
 189     j_coord_offsetC = 0;
 190     j_coord_offsetD = 0;
 191
 192     outeriter        = 0;
 193     inneriter        = 0;
 194
 195     for(iidx=0;iidx<4*DIM;iidx++)
 196     {
 197         scratch[iidx] = 0.0;
 198     }
 199
 200     /* Start outer loop over neighborlists */
 201     for(iidx=0; iidx<nri; iidx++)
 202     {
 203         /* Load shift vector for this list */
 204         i_shift_offset   = DIM*shiftidx[iidx];
 205
 206         /* Load limits for loop over neighbors */
 207         j_index_start    = jindex[iidx];
 208         j_index_end      = jindex[iidx+1];
 209
 210         /* Get outer coordinate index */
 211         inr              = iinr[iidx];
 212         i_coord_offset   = DIM*inr;
 213
 214         /* Load i particle coords and add shift vector */
 215         gmx_mm_load_shift_and_3rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
 216                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 217
 218         fix0             = _mm_setzero_ps();
 219         fiy0             = _mm_setzero_ps();
 220         fiz0             = _mm_setzero_ps();
 221         fix1             = _mm_setzero_ps();
 222         fiy1             = _mm_setzero_ps();
 223         fiz1             = _mm_setzero_ps();
 224         fix2             = _mm_setzero_ps();
 225         fiy2             = _mm_setzero_ps();
 226         fiz2             = _mm_setzero_ps();
 227
 228         /* Reset potential sums */
 229         velecsum         = _mm_setzero_ps();
 230         vvdwsum          = _mm_setzero_ps();
 231
 232         /* Start inner kernel loop */
 233         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 234         {
 235
 236             /* Get j neighbor index, and coordinate index */
 237             jnrA             = jjnr[jidx];
 238             jnrB             = jjnr[jidx+1];
 239             jnrC             = jjnr[jidx+2];
 240             jnrD             = jjnr[jidx+3];
 241             j_coord_offsetA  = DIM*jnrA;
 242             j_coord_offsetB  = DIM*jnrB;
 243             j_coord_offsetC  = DIM*jnrC;
 244             j_coord_offsetD  = DIM*jnrD;
 245
 246             /* load j atom coordinates */
 247             gmx_mm_load_3rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 248                                               x+j_coord_offsetC,x+j_coord_offsetD,
 249                                               &jx0,&jy0,&jz0,&jx1,&jy1,&jz1,&jx2,&jy2,&jz2);
 250
 251             /* Calculate displacement vector */
 252             dx00             = _mm_sub_ps(ix0,jx0);
 253             dy00             = _mm_sub_ps(iy0,jy0);
 254             dz00             = _mm_sub_ps(iz0,jz0);
 255             dx01             = _mm_sub_ps(ix0,jx1);
 256             dy01             = _mm_sub_ps(iy0,jy1);
 257             dz01             = _mm_sub_ps(iz0,jz1);
 258             dx02             = _mm_sub_ps(ix0,jx2);
 259             dy02             = _mm_sub_ps(iy0,jy2);
 260             dz02             = _mm_sub_ps(iz0,jz2);
 261             dx10             = _mm_sub_ps(ix1,jx0);
 262             dy10             = _mm_sub_ps(iy1,jy0);
 263             dz10             = _mm_sub_ps(iz1,jz0);
 264             dx11             = _mm_sub_ps(ix1,jx1);
 265             dy11             = _mm_sub_ps(iy1,jy1);
 266             dz11             = _mm_sub_ps(iz1,jz1);
 267             dx12             = _mm_sub_ps(ix1,jx2);
 268             dy12             = _mm_sub_ps(iy1,jy2);
 269             dz12             = _mm_sub_ps(iz1,jz2);
 270             dx20             = _mm_sub_ps(ix2,jx0);
 271             dy20             = _mm_sub_ps(iy2,jy0);
 272             dz20             = _mm_sub_ps(iz2,jz0);
 273             dx21             = _mm_sub_ps(ix2,jx1);
 274             dy21             = _mm_sub_ps(iy2,jy1);
 275             dz21             = _mm_sub_ps(iz2,jz1);
 276             dx22             = _mm_sub_ps(ix2,jx2);
 277             dy22             = _mm_sub_ps(iy2,jy2);
 278             dz22             = _mm_sub_ps(iz2,jz2);
 279
 280             /* Calculate squared distance and things based on it */
 281             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 282             rsq01            = gmx_mm_calc_rsq_ps(dx01,dy01,dz01);
 283             rsq02            = gmx_mm_calc_rsq_ps(dx02,dy02,dz02);
 284             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 285             rsq11            = gmx_mm_calc_rsq_ps(dx11,dy11,dz11);
 286             rsq12            = gmx_mm_calc_rsq_ps(dx12,dy12,dz12);
 287             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 288             rsq21            = gmx_mm_calc_rsq_ps(dx21,dy21,dz21);
 289             rsq22            = gmx_mm_calc_rsq_ps(dx22,dy22,dz22);
 290
 291             rinv00           = avx128fma_invsqrt_f(rsq00);
 292             rinv01           = avx128fma_invsqrt_f(rsq01);
 293             rinv02           = avx128fma_invsqrt_f(rsq02);
 294             rinv10           = avx128fma_invsqrt_f(rsq10);
 295             rinv11           = avx128fma_invsqrt_f(rsq11);
 296             rinv12           = avx128fma_invsqrt_f(rsq12);
 297             rinv20           = avx128fma_invsqrt_f(rsq20);
 298             rinv21           = avx128fma_invsqrt_f(rsq21);
 299             rinv22           = avx128fma_invsqrt_f(rsq22);
 300
 301             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 302             rinvsq01         = _mm_mul_ps(rinv01,rinv01);
 303             rinvsq02         = _mm_mul_ps(rinv02,rinv02);
 304             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 305             rinvsq11         = _mm_mul_ps(rinv11,rinv11);
 306             rinvsq12         = _mm_mul_ps(rinv12,rinv12);
 307             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 308             rinvsq21         = _mm_mul_ps(rinv21,rinv21);
 309             rinvsq22         = _mm_mul_ps(rinv22,rinv22);
 310
 311             fjx0             = _mm_setzero_ps();
 312             fjy0             = _mm_setzero_ps();
 313             fjz0             = _mm_setzero_ps();
 314             fjx1             = _mm_setzero_ps();
 315             fjy1             = _mm_setzero_ps();
 316             fjz1             = _mm_setzero_ps();
 317             fjx2             = _mm_setzero_ps();
 318             fjy2             = _mm_setzero_ps();
 319             fjz2             = _mm_setzero_ps();
 320
 321             /**************************
 322              * CALCULATE INTERACTIONS *
 323              **************************/
 324
 325             r00              = _mm_mul_ps(rsq00,rinv00);
 326
 327             /* EWALD ELECTROSTATICS */
 328
 329             /* Analytical PME correction */
 330             zeta2            = _mm_mul_ps(beta2,rsq00);
 331             rinv3            = _mm_mul_ps(rinvsq00,rinv00);
 332             pmecorrF         = avx128fma_pmecorrF_f(zeta2);
 333             felec            = _mm_macc_ps(pmecorrF,beta3,rinv3);
 334             felec            = _mm_mul_ps(qq00,felec);
 335             pmecorrV         = avx128fma_pmecorrV_f(zeta2);
 336             velec            = _mm_nmacc_ps(pmecorrV,beta,rinv00);
 337             velec            = _mm_mul_ps(qq00,velec);
 338
 339             /* Analytical LJ-PME */
 340             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 341             ewcljrsq         = _mm_mul_ps(ewclj2,rsq00);
 342             ewclj6           = _mm_mul_ps(ewclj2,_mm_mul_ps(ewclj2,ewclj2));
 343             exponent         = avx128fma_exp_f(ewcljrsq);
 344             /* poly = exp(-(beta*r)^2) * (1 + (beta*r)^2 + (beta*r)^4 /2) */
 345             poly             = _mm_mul_ps(exponent,_mm_macc_ps(_mm_mul_ps(ewcljrsq,ewcljrsq),one_half,_mm_sub_ps(one,ewcljrsq)));
 346             /* vvdw6 = [C6 - C6grid * (1-poly)]/r6 */
 347             vvdw6            = _mm_mul_ps(_mm_macc_ps(-c6grid_00,_mm_sub_ps(one,poly),c6_00),rinvsix);
 348             vvdw12           = _mm_mul_ps(c12_00,_mm_mul_ps(rinvsix,rinvsix));
 349             vvdw             = _mm_msub_ps(vvdw12,one_twelfth,_mm_mul_ps(vvdw6,one_sixth));
 350             /* fvdw = vvdw12/r - (vvdw6/r + (C6grid * exponent * beta^6)/r) */
 351             fvdw             = _mm_mul_ps(_mm_add_ps(vvdw12,_mm_msub_ps(_mm_mul_ps(c6grid_00,one_sixth),_mm_mul_ps(exponent,ewclj6),vvdw6)),rinvsq00);
 352
 353             /* Update potential sum for this i atom from the interaction with this j atom. */
 354             velecsum         = _mm_add_ps(velecsum,velec);
 355             vvdwsum          = _mm_add_ps(vvdwsum,vvdw);
 356
 357             fscal            = _mm_add_ps(felec,fvdw);
 358
 359              /* Update vectorial force */
 360             fix0             = _mm_macc_ps(dx00,fscal,fix0);
 361             fiy0             = _mm_macc_ps(dy00,fscal,fiy0);
 362             fiz0             = _mm_macc_ps(dz00,fscal,fiz0);
 363
 364             fjx0             = _mm_macc_ps(dx00,fscal,fjx0);
 365             fjy0             = _mm_macc_ps(dy00,fscal,fjy0);
 366             fjz0             = _mm_macc_ps(dz00,fscal,fjz0);
 367
 368             /**************************
 369              * CALCULATE INTERACTIONS *
 370              **************************/
 371
 372             r01              = _mm_mul_ps(rsq01,rinv01);
 373
 374             /* EWALD ELECTROSTATICS */
 375
 376             /* Analytical PME correction */
 377             zeta2            = _mm_mul_ps(beta2,rsq01);
 378             rinv3            = _mm_mul_ps(rinvsq01,rinv01);
 379             pmecorrF         = avx128fma_pmecorrF_f(zeta2);
 380             felec            = _mm_macc_ps(pmecorrF,beta3,rinv3);
 381             felec            = _mm_mul_ps(qq01,felec);
 382             pmecorrV         = avx128fma_pmecorrV_f(zeta2);
 383             velec            = _mm_nmacc_ps(pmecorrV,beta,rinv01);
 384             velec            = _mm_mul_ps(qq01,velec);
 385
 386             /* Update potential sum for this i atom from the interaction with this j atom. */
 387             velecsum         = _mm_add_ps(velecsum,velec);
 388
 389             fscal            = felec;
 390
 391              /* Update vectorial force */
 392             fix0             = _mm_macc_ps(dx01,fscal,fix0);
 393             fiy0             = _mm_macc_ps(dy01,fscal,fiy0);
 394             fiz0             = _mm_macc_ps(dz01,fscal,fiz0);
 395
 396             fjx1             = _mm_macc_ps(dx01,fscal,fjx1);
 397             fjy1             = _mm_macc_ps(dy01,fscal,fjy1);
 398             fjz1             = _mm_macc_ps(dz01,fscal,fjz1);
 399
 400             /**************************
 401              * CALCULATE INTERACTIONS *
 402              **************************/
 403
 404             r02              = _mm_mul_ps(rsq02,rinv02);
 405
 406             /* EWALD ELECTROSTATICS */
 407
 408             /* Analytical PME correction */
 409             zeta2            = _mm_mul_ps(beta2,rsq02);
 410             rinv3            = _mm_mul_ps(rinvsq02,rinv02);
 411             pmecorrF         = avx128fma_pmecorrF_f(zeta2);
 412             felec            = _mm_macc_ps(pmecorrF,beta3,rinv3);
 413             felec            = _mm_mul_ps(qq02,felec);
 414             pmecorrV         = avx128fma_pmecorrV_f(zeta2);
 415             velec            = _mm_nmacc_ps(pmecorrV,beta,rinv02);
 416             velec            = _mm_mul_ps(qq02,velec);
 417
 418             /* Update potential sum for this i atom from the interaction with this j atom. */
 419             velecsum         = _mm_add_ps(velecsum,velec);
 420
 421             fscal            = felec;
 422
 423              /* Update vectorial force */
 424             fix0             = _mm_macc_ps(dx02,fscal,fix0);
 425             fiy0             = _mm_macc_ps(dy02,fscal,fiy0);
 426             fiz0             = _mm_macc_ps(dz02,fscal,fiz0);
 427
 428             fjx2             = _mm_macc_ps(dx02,fscal,fjx2);
 429             fjy2             = _mm_macc_ps(dy02,fscal,fjy2);
 430             fjz2             = _mm_macc_ps(dz02,fscal,fjz2);
 431
 432             /**************************
 433              * CALCULATE INTERACTIONS *
 434              **************************/
 435
 436             r10              = _mm_mul_ps(rsq10,rinv10);
 437
 438             /* EWALD ELECTROSTATICS */
 439
 440             /* Analytical PME correction */
 441             zeta2            = _mm_mul_ps(beta2,rsq10);
 442             rinv3            = _mm_mul_ps(rinvsq10,rinv10);
 443             pmecorrF         = avx128fma_pmecorrF_f(zeta2);
 444             felec            = _mm_macc_ps(pmecorrF,beta3,rinv3);
 445             felec            = _mm_mul_ps(qq10,felec);
 446             pmecorrV         = avx128fma_pmecorrV_f(zeta2);
 447             velec            = _mm_nmacc_ps(pmecorrV,beta,rinv10);
 448             velec            = _mm_mul_ps(qq10,velec);
 449
 450             /* Update potential sum for this i atom from the interaction with this j atom. */
 451             velecsum         = _mm_add_ps(velecsum,velec);
 452
 453             fscal            = felec;
 454
 455              /* Update vectorial force */
 456             fix1             = _mm_macc_ps(dx10,fscal,fix1);
 457             fiy1             = _mm_macc_ps(dy10,fscal,fiy1);
 458             fiz1             = _mm_macc_ps(dz10,fscal,fiz1);
 459
 460             fjx0             = _mm_macc_ps(dx10,fscal,fjx0);
 461             fjy0             = _mm_macc_ps(dy10,fscal,fjy0);
 462             fjz0             = _mm_macc_ps(dz10,fscal,fjz0);
 463
 464             /**************************
 465              * CALCULATE INTERACTIONS *
 466              **************************/
 467
 468             r11              = _mm_mul_ps(rsq11,rinv11);
 469
 470             /* EWALD ELECTROSTATICS */
 471
 472             /* Analytical PME correction */
 473             zeta2            = _mm_mul_ps(beta2,rsq11);
 474             rinv3            = _mm_mul_ps(rinvsq11,rinv11);
 475             pmecorrF         = avx128fma_pmecorrF_f(zeta2);
 476             felec            = _mm_macc_ps(pmecorrF,beta3,rinv3);
 477             felec            = _mm_mul_ps(qq11,felec);
 478             pmecorrV         = avx128fma_pmecorrV_f(zeta2);
 479             velec            = _mm_nmacc_ps(pmecorrV,beta,rinv11);
 480             velec            = _mm_mul_ps(qq11,velec);
 481
 482             /* Update potential sum for this i atom from the interaction with this j atom. */
 483             velecsum         = _mm_add_ps(velecsum,velec);
 484
 485             fscal            = felec;
 486
 487              /* Update vectorial force */
 488             fix1             = _mm_macc_ps(dx11,fscal,fix1);
 489             fiy1             = _mm_macc_ps(dy11,fscal,fiy1);
 490             fiz1             = _mm_macc_ps(dz11,fscal,fiz1);
 491
 492             fjx1             = _mm_macc_ps(dx11,fscal,fjx1);
 493             fjy1             = _mm_macc_ps(dy11,fscal,fjy1);
 494             fjz1             = _mm_macc_ps(dz11,fscal,fjz1);
 495
 496             /**************************
 497              * CALCULATE INTERACTIONS *
 498              **************************/
 499
 500             r12              = _mm_mul_ps(rsq12,rinv12);
 501
 502             /* EWALD ELECTROSTATICS */
 503
 504             /* Analytical PME correction */
 505             zeta2            = _mm_mul_ps(beta2,rsq12);
 506             rinv3            = _mm_mul_ps(rinvsq12,rinv12);
 507             pmecorrF         = avx128fma_pmecorrF_f(zeta2);
 508             felec            = _mm_macc_ps(pmecorrF,beta3,rinv3);
 509             felec            = _mm_mul_ps(qq12,felec);
 510             pmecorrV         = avx128fma_pmecorrV_f(zeta2);
 511             velec            = _mm_nmacc_ps(pmecorrV,beta,rinv12);
 512             velec            = _mm_mul_ps(qq12,velec);
 513
 514             /* Update potential sum for this i atom from the interaction with this j atom. */
 515             velecsum         = _mm_add_ps(velecsum,velec);
 516
 517             fscal            = felec;
 518
 519              /* Update vectorial force */
 520             fix1             = _mm_macc_ps(dx12,fscal,fix1);
 521             fiy1             = _mm_macc_ps(dy12,fscal,fiy1);
 522             fiz1             = _mm_macc_ps(dz12,fscal,fiz1);
 523
 524             fjx2             = _mm_macc_ps(dx12,fscal,fjx2);
 525             fjy2             = _mm_macc_ps(dy12,fscal,fjy2);
 526             fjz2             = _mm_macc_ps(dz12,fscal,fjz2);
 527
 528             /**************************
 529              * CALCULATE INTERACTIONS *
 530              **************************/
 531
 532             r20              = _mm_mul_ps(rsq20,rinv20);
 533
 534             /* EWALD ELECTROSTATICS */
 535
 536             /* Analytical PME correction */
 537             zeta2            = _mm_mul_ps(beta2,rsq20);
 538             rinv3            = _mm_mul_ps(rinvsq20,rinv20);
 539             pmecorrF         = avx128fma_pmecorrF_f(zeta2);
 540             felec            = _mm_macc_ps(pmecorrF,beta3,rinv3);
 541             felec            = _mm_mul_ps(qq20,felec);
 542             pmecorrV         = avx128fma_pmecorrV_f(zeta2);
 543             velec            = _mm_nmacc_ps(pmecorrV,beta,rinv20);
 544             velec            = _mm_mul_ps(qq20,velec);
 545
 546             /* Update potential sum for this i atom from the interaction with this j atom. */
 547             velecsum         = _mm_add_ps(velecsum,velec);
 548
 549             fscal            = felec;
 550
 551              /* Update vectorial force */
 552             fix2             = _mm_macc_ps(dx20,fscal,fix2);
 553             fiy2             = _mm_macc_ps(dy20,fscal,fiy2);
 554             fiz2             = _mm_macc_ps(dz20,fscal,fiz2);
 555
 556             fjx0             = _mm_macc_ps(dx20,fscal,fjx0);
 557             fjy0             = _mm_macc_ps(dy20,fscal,fjy0);
 558             fjz0             = _mm_macc_ps(dz20,fscal,fjz0);
 559
 560             /**************************
 561              * CALCULATE INTERACTIONS *
 562              **************************/
 563
 564             r21              = _mm_mul_ps(rsq21,rinv21);
 565
 566             /* EWALD ELECTROSTATICS */
 567
 568             /* Analytical PME correction */
 569             zeta2            = _mm_mul_ps(beta2,rsq21);
 570             rinv3            = _mm_mul_ps(rinvsq21,rinv21);
 571             pmecorrF         = avx128fma_pmecorrF_f(zeta2);
 572             felec            = _mm_macc_ps(pmecorrF,beta3,rinv3);
 573             felec            = _mm_mul_ps(qq21,felec);
 574             pmecorrV         = avx128fma_pmecorrV_f(zeta2);
 575             velec            = _mm_nmacc_ps(pmecorrV,beta,rinv21);
 576             velec            = _mm_mul_ps(qq21,velec);
 577
 578             /* Update potential sum for this i atom from the interaction with this j atom. */
 579             velecsum         = _mm_add_ps(velecsum,velec);
 580
 581             fscal            = felec;
 582
 583              /* Update vectorial force */
 584             fix2             = _mm_macc_ps(dx21,fscal,fix2);
 585             fiy2             = _mm_macc_ps(dy21,fscal,fiy2);
 586             fiz2             = _mm_macc_ps(dz21,fscal,fiz2);
 587
 588             fjx1             = _mm_macc_ps(dx21,fscal,fjx1);
 589             fjy1             = _mm_macc_ps(dy21,fscal,fjy1);
 590             fjz1             = _mm_macc_ps(dz21,fscal,fjz1);
 591
 592             /**************************
 593              * CALCULATE INTERACTIONS *
 594              **************************/
 595
 596             r22              = _mm_mul_ps(rsq22,rinv22);
 597
 598             /* EWALD ELECTROSTATICS */
 599
 600             /* Analytical PME correction */
 601             zeta2            = _mm_mul_ps(beta2,rsq22);
 602             rinv3            = _mm_mul_ps(rinvsq22,rinv22);
 603             pmecorrF         = avx128fma_pmecorrF_f(zeta2);
 604             felec            = _mm_macc_ps(pmecorrF,beta3,rinv3);
 605             felec            = _mm_mul_ps(qq22,felec);
 606             pmecorrV         = avx128fma_pmecorrV_f(zeta2);
 607             velec            = _mm_nmacc_ps(pmecorrV,beta,rinv22);
 608             velec            = _mm_mul_ps(qq22,velec);
 609
 610             /* Update potential sum for this i atom from the interaction with this j atom. */
 611             velecsum         = _mm_add_ps(velecsum,velec);
 612
 613             fscal            = felec;
 614
 615              /* Update vectorial force */
 616             fix2             = _mm_macc_ps(dx22,fscal,fix2);
 617             fiy2             = _mm_macc_ps(dy22,fscal,fiy2);
 618             fiz2             = _mm_macc_ps(dz22,fscal,fiz2);
 619
 620             fjx2             = _mm_macc_ps(dx22,fscal,fjx2);
 621             fjy2             = _mm_macc_ps(dy22,fscal,fjy2);
 622             fjz2             = _mm_macc_ps(dz22,fscal,fjz2);
 623
 624             fjptrA             = f+j_coord_offsetA;
 625             fjptrB             = f+j_coord_offsetB;
 626             fjptrC             = f+j_coord_offsetC;
 627             fjptrD             = f+j_coord_offsetD;
 628
 629             gmx_mm_decrement_3rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,
 630                                                    fjx0,fjy0,fjz0,fjx1,fjy1,fjz1,fjx2,fjy2,fjz2);
 631
 632             /* Inner loop uses 285 flops */
 633         }
 634
 635         if(jidx<j_index_end)
 636         {
 637
 638             /* Get j neighbor index, and coordinate index */
 639             jnrlistA         = jjnr[jidx];
 640             jnrlistB         = jjnr[jidx+1];
 641             jnrlistC         = jjnr[jidx+2];
 642             jnrlistD         = jjnr[jidx+3];
 643             /* Sign of each element will be negative for non-real atoms.
 644              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 645              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 646              */
 647             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 648             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 649             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 650             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 651             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 652             j_coord_offsetA  = DIM*jnrA;
 653             j_coord_offsetB  = DIM*jnrB;
 654             j_coord_offsetC  = DIM*jnrC;
 655             j_coord_offsetD  = DIM*jnrD;
 656
 657             /* load j atom coordinates */
 658             gmx_mm_load_3rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 659                                               x+j_coord_offsetC,x+j_coord_offsetD,
 660                                               &jx0,&jy0,&jz0,&jx1,&jy1,&jz1,&jx2,&jy2,&jz2);
 661
 662             /* Calculate displacement vector */
 663             dx00             = _mm_sub_ps(ix0,jx0);
 664             dy00             = _mm_sub_ps(iy0,jy0);
 665             dz00             = _mm_sub_ps(iz0,jz0);
 666             dx01             = _mm_sub_ps(ix0,jx1);
 667             dy01             = _mm_sub_ps(iy0,jy1);
 668             dz01             = _mm_sub_ps(iz0,jz1);
 669             dx02             = _mm_sub_ps(ix0,jx2);
 670             dy02             = _mm_sub_ps(iy0,jy2);
 671             dz02             = _mm_sub_ps(iz0,jz2);
 672             dx10             = _mm_sub_ps(ix1,jx0);
 673             dy10             = _mm_sub_ps(iy1,jy0);
 674             dz10             = _mm_sub_ps(iz1,jz0);
 675             dx11             = _mm_sub_ps(ix1,jx1);
 676             dy11             = _mm_sub_ps(iy1,jy1);
 677             dz11             = _mm_sub_ps(iz1,jz1);
 678             dx12             = _mm_sub_ps(ix1,jx2);
 679             dy12             = _mm_sub_ps(iy1,jy2);
 680             dz12             = _mm_sub_ps(iz1,jz2);
 681             dx20             = _mm_sub_ps(ix2,jx0);
 682             dy20             = _mm_sub_ps(iy2,jy0);
 683             dz20             = _mm_sub_ps(iz2,jz0);
 684             dx21             = _mm_sub_ps(ix2,jx1);
 685             dy21             = _mm_sub_ps(iy2,jy1);
 686             dz21             = _mm_sub_ps(iz2,jz1);
 687             dx22             = _mm_sub_ps(ix2,jx2);
 688             dy22             = _mm_sub_ps(iy2,jy2);
 689             dz22             = _mm_sub_ps(iz2,jz2);
 690
 691             /* Calculate squared distance and things based on it */
 692             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 693             rsq01            = gmx_mm_calc_rsq_ps(dx01,dy01,dz01);
 694             rsq02            = gmx_mm_calc_rsq_ps(dx02,dy02,dz02);
 695             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 696             rsq11            = gmx_mm_calc_rsq_ps(dx11,dy11,dz11);
 697             rsq12            = gmx_mm_calc_rsq_ps(dx12,dy12,dz12);
 698             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 699             rsq21            = gmx_mm_calc_rsq_ps(dx21,dy21,dz21);
 700             rsq22            = gmx_mm_calc_rsq_ps(dx22,dy22,dz22);
 701
 702             rinv00           = avx128fma_invsqrt_f(rsq00);
 703             rinv01           = avx128fma_invsqrt_f(rsq01);
 704             rinv02           = avx128fma_invsqrt_f(rsq02);
 705             rinv10           = avx128fma_invsqrt_f(rsq10);
 706             rinv11           = avx128fma_invsqrt_f(rsq11);
 707             rinv12           = avx128fma_invsqrt_f(rsq12);
 708             rinv20           = avx128fma_invsqrt_f(rsq20);
 709             rinv21           = avx128fma_invsqrt_f(rsq21);
 710             rinv22           = avx128fma_invsqrt_f(rsq22);
 711
 712             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 713             rinvsq01         = _mm_mul_ps(rinv01,rinv01);
 714             rinvsq02         = _mm_mul_ps(rinv02,rinv02);
 715             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 716             rinvsq11         = _mm_mul_ps(rinv11,rinv11);
 717             rinvsq12         = _mm_mul_ps(rinv12,rinv12);
 718             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 719             rinvsq21         = _mm_mul_ps(rinv21,rinv21);
 720             rinvsq22         = _mm_mul_ps(rinv22,rinv22);
 721
 722             fjx0             = _mm_setzero_ps();
 723             fjy0             = _mm_setzero_ps();
 724             fjz0             = _mm_setzero_ps();
 725             fjx1             = _mm_setzero_ps();
 726             fjy1             = _mm_setzero_ps();
 727             fjz1             = _mm_setzero_ps();
 728             fjx2             = _mm_setzero_ps();
 729             fjy2             = _mm_setzero_ps();
 730             fjz2             = _mm_setzero_ps();
 731
 732             /**************************
 733              * CALCULATE INTERACTIONS *
 734              **************************/
 735
 736             r00              = _mm_mul_ps(rsq00,rinv00);
 737             r00              = _mm_andnot_ps(dummy_mask,r00);
 738
 739             /* EWALD ELECTROSTATICS */
 740
 741             /* Analytical PME correction */
 742             zeta2            = _mm_mul_ps(beta2,rsq00);
 743             rinv3            = _mm_mul_ps(rinvsq00,rinv00);
 744             pmecorrF         = avx128fma_pmecorrF_f(zeta2);
 745             felec            = _mm_macc_ps(pmecorrF,beta3,rinv3);
 746             felec            = _mm_mul_ps(qq00,felec);
 747             pmecorrV         = avx128fma_pmecorrV_f(zeta2);
 748             velec            = _mm_nmacc_ps(pmecorrV,beta,rinv00);
 749             velec            = _mm_mul_ps(qq00,velec);
 750
 751             /* Analytical LJ-PME */
 752             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 753             ewcljrsq         = _mm_mul_ps(ewclj2,rsq00);
 754             ewclj6           = _mm_mul_ps(ewclj2,_mm_mul_ps(ewclj2,ewclj2));
 755             exponent         = avx128fma_exp_f(ewcljrsq);
 756             /* poly = exp(-(beta*r)^2) * (1 + (beta*r)^2 + (beta*r)^4 /2) */
 757             poly             = _mm_mul_ps(exponent,_mm_macc_ps(_mm_mul_ps(ewcljrsq,ewcljrsq),one_half,_mm_sub_ps(one,ewcljrsq)));
 758             /* vvdw6 = [C6 - C6grid * (1-poly)]/r6 */
 759             vvdw6            = _mm_mul_ps(_mm_macc_ps(-c6grid_00,_mm_sub_ps(one,poly),c6_00),rinvsix);
 760             vvdw12           = _mm_mul_ps(c12_00,_mm_mul_ps(rinvsix,rinvsix));
 761             vvdw             = _mm_msub_ps(vvdw12,one_twelfth,_mm_mul_ps(vvdw6,one_sixth));
 762             /* fvdw = vvdw12/r - (vvdw6/r + (C6grid * exponent * beta^6)/r) */
 763             fvdw             = _mm_mul_ps(_mm_add_ps(vvdw12,_mm_msub_ps(_mm_mul_ps(c6grid_00,one_sixth),_mm_mul_ps(exponent,ewclj6),vvdw6)),rinvsq00);
 764
 765             /* Update potential sum for this i atom from the interaction with this j atom. */
 766             velec            = _mm_andnot_ps(dummy_mask,velec);
 767             velecsum         = _mm_add_ps(velecsum,velec);
 768             vvdw             = _mm_andnot_ps(dummy_mask,vvdw);
 769             vvdwsum          = _mm_add_ps(vvdwsum,vvdw);
 770
 771             fscal            = _mm_add_ps(felec,fvdw);
 772
 773             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 774
 775              /* Update vectorial force */
 776             fix0             = _mm_macc_ps(dx00,fscal,fix0);
 777             fiy0             = _mm_macc_ps(dy00,fscal,fiy0);
 778             fiz0             = _mm_macc_ps(dz00,fscal,fiz0);
 779
 780             fjx0             = _mm_macc_ps(dx00,fscal,fjx0);
 781             fjy0             = _mm_macc_ps(dy00,fscal,fjy0);
 782             fjz0             = _mm_macc_ps(dz00,fscal,fjz0);
 783
 784             /**************************
 785              * CALCULATE INTERACTIONS *
 786              **************************/
 787
 788             r01              = _mm_mul_ps(rsq01,rinv01);
 789             r01              = _mm_andnot_ps(dummy_mask,r01);
 790
 791             /* EWALD ELECTROSTATICS */
 792
 793             /* Analytical PME correction */
 794             zeta2            = _mm_mul_ps(beta2,rsq01);
 795             rinv3            = _mm_mul_ps(rinvsq01,rinv01);
 796             pmecorrF         = avx128fma_pmecorrF_f(zeta2);
 797             felec            = _mm_macc_ps(pmecorrF,beta3,rinv3);
 798             felec            = _mm_mul_ps(qq01,felec);
 799             pmecorrV         = avx128fma_pmecorrV_f(zeta2);
 800             velec            = _mm_nmacc_ps(pmecorrV,beta,rinv01);
 801             velec            = _mm_mul_ps(qq01,velec);
 802
 803             /* Update potential sum for this i atom from the interaction with this j atom. */
 804             velec            = _mm_andnot_ps(dummy_mask,velec);
 805             velecsum         = _mm_add_ps(velecsum,velec);
 806
 807             fscal            = felec;
 808
 809             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 810
 811              /* Update vectorial force */
 812             fix0             = _mm_macc_ps(dx01,fscal,fix0);
 813             fiy0             = _mm_macc_ps(dy01,fscal,fiy0);
 814             fiz0             = _mm_macc_ps(dz01,fscal,fiz0);
 815
 816             fjx1             = _mm_macc_ps(dx01,fscal,fjx1);
 817             fjy1             = _mm_macc_ps(dy01,fscal,fjy1);
 818             fjz1             = _mm_macc_ps(dz01,fscal,fjz1);
 819
 820             /**************************
 821              * CALCULATE INTERACTIONS *
 822              **************************/
 823
 824             r02              = _mm_mul_ps(rsq02,rinv02);
 825             r02              = _mm_andnot_ps(dummy_mask,r02);
 826
 827             /* EWALD ELECTROSTATICS */
 828
 829             /* Analytical PME correction */
 830             zeta2            = _mm_mul_ps(beta2,rsq02);
 831             rinv3            = _mm_mul_ps(rinvsq02,rinv02);
 832             pmecorrF         = avx128fma_pmecorrF_f(zeta2);
 833             felec            = _mm_macc_ps(pmecorrF,beta3,rinv3);
 834             felec            = _mm_mul_ps(qq02,felec);
 835             pmecorrV         = avx128fma_pmecorrV_f(zeta2);
 836             velec            = _mm_nmacc_ps(pmecorrV,beta,rinv02);
 837             velec            = _mm_mul_ps(qq02,velec);
 838
 839             /* Update potential sum for this i atom from the interaction with this j atom. */
 840             velec            = _mm_andnot_ps(dummy_mask,velec);
 841             velecsum         = _mm_add_ps(velecsum,velec);
 842
 843             fscal            = felec;
 844
 845             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 846
 847              /* Update vectorial force */
 848             fix0             = _mm_macc_ps(dx02,fscal,fix0);
 849             fiy0             = _mm_macc_ps(dy02,fscal,fiy0);
 850             fiz0             = _mm_macc_ps(dz02,fscal,fiz0);
 851
 852             fjx2             = _mm_macc_ps(dx02,fscal,fjx2);
 853             fjy2             = _mm_macc_ps(dy02,fscal,fjy2);
 854             fjz2             = _mm_macc_ps(dz02,fscal,fjz2);
 855
 856             /**************************
 857              * CALCULATE INTERACTIONS *
 858              **************************/
 859
 860             r10              = _mm_mul_ps(rsq10,rinv10);
 861             r10              = _mm_andnot_ps(dummy_mask,r10);
 862
 863             /* EWALD ELECTROSTATICS */
 864
 865             /* Analytical PME correction */
 866             zeta2            = _mm_mul_ps(beta2,rsq10);
 867             rinv3            = _mm_mul_ps(rinvsq10,rinv10);
 868             pmecorrF         = avx128fma_pmecorrF_f(zeta2);
 869             felec            = _mm_macc_ps(pmecorrF,beta3,rinv3);
 870             felec            = _mm_mul_ps(qq10,felec);
 871             pmecorrV         = avx128fma_pmecorrV_f(zeta2);
 872             velec            = _mm_nmacc_ps(pmecorrV,beta,rinv10);
 873             velec            = _mm_mul_ps(qq10,velec);
 874
 875             /* Update potential sum for this i atom from the interaction with this j atom. */
 876             velec            = _mm_andnot_ps(dummy_mask,velec);
 877             velecsum         = _mm_add_ps(velecsum,velec);
 878
 879             fscal            = felec;
 880
 881             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 882
 883              /* Update vectorial force */
 884             fix1             = _mm_macc_ps(dx10,fscal,fix1);
 885             fiy1             = _mm_macc_ps(dy10,fscal,fiy1);
 886             fiz1             = _mm_macc_ps(dz10,fscal,fiz1);
 887
 888             fjx0             = _mm_macc_ps(dx10,fscal,fjx0);
 889             fjy0             = _mm_macc_ps(dy10,fscal,fjy0);
 890             fjz0             = _mm_macc_ps(dz10,fscal,fjz0);
 891
 892             /**************************
 893              * CALCULATE INTERACTIONS *
 894              **************************/
 895
 896             r11              = _mm_mul_ps(rsq11,rinv11);
 897             r11              = _mm_andnot_ps(dummy_mask,r11);
 898
 899             /* EWALD ELECTROSTATICS */
 900
 901             /* Analytical PME correction */
 902             zeta2            = _mm_mul_ps(beta2,rsq11);
 903             rinv3            = _mm_mul_ps(rinvsq11,rinv11);
 904             pmecorrF         = avx128fma_pmecorrF_f(zeta2);
 905             felec            = _mm_macc_ps(pmecorrF,beta3,rinv3);
 906             felec            = _mm_mul_ps(qq11,felec);
 907             pmecorrV         = avx128fma_pmecorrV_f(zeta2);
 908             velec            = _mm_nmacc_ps(pmecorrV,beta,rinv11);
 909             velec            = _mm_mul_ps(qq11,velec);
 910
 911             /* Update potential sum for this i atom from the interaction with this j atom. */
 912             velec            = _mm_andnot_ps(dummy_mask,velec);
 913             velecsum         = _mm_add_ps(velecsum,velec);
 914
 915             fscal            = felec;
 916
 917             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 918
 919              /* Update vectorial force */
 920             fix1             = _mm_macc_ps(dx11,fscal,fix1);
 921             fiy1             = _mm_macc_ps(dy11,fscal,fiy1);
 922             fiz1             = _mm_macc_ps(dz11,fscal,fiz1);
 923
 924             fjx1             = _mm_macc_ps(dx11,fscal,fjx1);
 925             fjy1             = _mm_macc_ps(dy11,fscal,fjy1);
 926             fjz1             = _mm_macc_ps(dz11,fscal,fjz1);
 927
 928             /**************************
 929              * CALCULATE INTERACTIONS *
 930              **************************/
 931
 932             r12              = _mm_mul_ps(rsq12,rinv12);
 933             r12              = _mm_andnot_ps(dummy_mask,r12);
 934
 935             /* EWALD ELECTROSTATICS */
 936
 937             /* Analytical PME correction */
 938             zeta2            = _mm_mul_ps(beta2,rsq12);
 939             rinv3            = _mm_mul_ps(rinvsq12,rinv12);
 940             pmecorrF         = avx128fma_pmecorrF_f(zeta2);
 941             felec            = _mm_macc_ps(pmecorrF,beta3,rinv3);
 942             felec            = _mm_mul_ps(qq12,felec);
 943             pmecorrV         = avx128fma_pmecorrV_f(zeta2);
 944             velec            = _mm_nmacc_ps(pmecorrV,beta,rinv12);
 945             velec            = _mm_mul_ps(qq12,velec);
 946
 947             /* Update potential sum for this i atom from the interaction with this j atom. */
 948             velec            = _mm_andnot_ps(dummy_mask,velec);
 949             velecsum         = _mm_add_ps(velecsum,velec);
 950
 951             fscal            = felec;
 952
 953             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 954
 955              /* Update vectorial force */
 956             fix1             = _mm_macc_ps(dx12,fscal,fix1);
 957             fiy1             = _mm_macc_ps(dy12,fscal,fiy1);
 958             fiz1             = _mm_macc_ps(dz12,fscal,fiz1);
 959
 960             fjx2             = _mm_macc_ps(dx12,fscal,fjx2);
 961             fjy2             = _mm_macc_ps(dy12,fscal,fjy2);
 962             fjz2             = _mm_macc_ps(dz12,fscal,fjz2);
 963
 964             /**************************
 965              * CALCULATE INTERACTIONS *
 966              **************************/
 967
 968             r20              = _mm_mul_ps(rsq20,rinv20);
 969             r20              = _mm_andnot_ps(dummy_mask,r20);
 970
 971             /* EWALD ELECTROSTATICS */
 972
 973             /* Analytical PME correction */
 974             zeta2            = _mm_mul_ps(beta2,rsq20);
 975             rinv3            = _mm_mul_ps(rinvsq20,rinv20);
 976             pmecorrF         = avx128fma_pmecorrF_f(zeta2);
 977             felec            = _mm_macc_ps(pmecorrF,beta3,rinv3);
 978             felec            = _mm_mul_ps(qq20,felec);
 979             pmecorrV         = avx128fma_pmecorrV_f(zeta2);
 980             velec            = _mm_nmacc_ps(pmecorrV,beta,rinv20);
 981             velec            = _mm_mul_ps(qq20,velec);
 982
 983             /* Update potential sum for this i atom from the interaction with this j atom. */
 984             velec            = _mm_andnot_ps(dummy_mask,velec);
 985             velecsum         = _mm_add_ps(velecsum,velec);
 986
 987             fscal            = felec;
 988
 989             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 990
 991              /* Update vectorial force */
 992             fix2             = _mm_macc_ps(dx20,fscal,fix2);
 993             fiy2             = _mm_macc_ps(dy20,fscal,fiy2);
 994             fiz2             = _mm_macc_ps(dz20,fscal,fiz2);
 995
 996             fjx0             = _mm_macc_ps(dx20,fscal,fjx0);
 997             fjy0             = _mm_macc_ps(dy20,fscal,fjy0);
 998             fjz0             = _mm_macc_ps(dz20,fscal,fjz0);
 999
1000             /**************************
1001              * CALCULATE INTERACTIONS *
1002              **************************/
1003
1004             r21              = _mm_mul_ps(rsq21,rinv21);
1005             r21              = _mm_andnot_ps(dummy_mask,r21);
1006
1007             /* EWALD ELECTROSTATICS */
1008
1009             /* Analytical PME correction */
1010             zeta2            = _mm_mul_ps(beta2,rsq21);
1011             rinv3            = _mm_mul_ps(rinvsq21,rinv21);
1012             pmecorrF         = avx128fma_pmecorrF_f(zeta2);
1013             felec            = _mm_macc_ps(pmecorrF,beta3,rinv3);
1014             felec            = _mm_mul_ps(qq21,felec);
1015             pmecorrV         = avx128fma_pmecorrV_f(zeta2);
1016             velec            = _mm_nmacc_ps(pmecorrV,beta,rinv21);
1017             velec            = _mm_mul_ps(qq21,velec);
1018
1019             /* Update potential sum for this i atom from the interaction with this j atom. */
1020             velec            = _mm_andnot_ps(dummy_mask,velec);
1021             velecsum         = _mm_add_ps(velecsum,velec);
1022
1023             fscal            = felec;
1024
1025             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1026
1027              /* Update vectorial force */
1028             fix2             = _mm_macc_ps(dx21,fscal,fix2);
1029             fiy2             = _mm_macc_ps(dy21,fscal,fiy2);
1030             fiz2             = _mm_macc_ps(dz21,fscal,fiz2);
1031
1032             fjx1             = _mm_macc_ps(dx21,fscal,fjx1);
1033             fjy1             = _mm_macc_ps(dy21,fscal,fjy1);
1034             fjz1             = _mm_macc_ps(dz21,fscal,fjz1);
1035
1036             /**************************
1037              * CALCULATE INTERACTIONS *
1038              **************************/
1039
1040             r22              = _mm_mul_ps(rsq22,rinv22);
1041             r22              = _mm_andnot_ps(dummy_mask,r22);
1042
1043             /* EWALD ELECTROSTATICS */
1044
1045             /* Analytical PME correction */
1046             zeta2            = _mm_mul_ps(beta2,rsq22);
1047             rinv3            = _mm_mul_ps(rinvsq22,rinv22);
1048             pmecorrF         = avx128fma_pmecorrF_f(zeta2);
1049             felec            = _mm_macc_ps(pmecorrF,beta3,rinv3);
1050             felec            = _mm_mul_ps(qq22,felec);
1051             pmecorrV         = avx128fma_pmecorrV_f(zeta2);
1052             velec            = _mm_nmacc_ps(pmecorrV,beta,rinv22);
1053             velec            = _mm_mul_ps(qq22,velec);
1054
1055             /* Update potential sum for this i atom from the interaction with this j atom. */
1056             velec            = _mm_andnot_ps(dummy_mask,velec);
1057             velecsum         = _mm_add_ps(velecsum,velec);
1058
1059             fscal            = felec;
1060
1061             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1062
1063              /* Update vectorial force */
1064             fix2             = _mm_macc_ps(dx22,fscal,fix2);
1065             fiy2             = _mm_macc_ps(dy22,fscal,fiy2);
1066             fiz2             = _mm_macc_ps(dz22,fscal,fiz2);
1067
1068             fjx2             = _mm_macc_ps(dx22,fscal,fjx2);
1069             fjy2             = _mm_macc_ps(dy22,fscal,fjy2);
1070             fjz2             = _mm_macc_ps(dz22,fscal,fjz2);
1071
1072             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
1073             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
1074             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
1075             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
1076
1077             gmx_mm_decrement_3rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,
1078                                                    fjx0,fjy0,fjz0,fjx1,fjy1,fjz1,fjx2,fjy2,fjz2);
1079
1080             /* Inner loop uses 294 flops */
1081         }
1082
1083         /* End of innermost loop */
1084
1085         gmx_mm_update_iforce_3atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
1086                                               f+i_coord_offset,fshift+i_shift_offset);
1087
1088         ggid                        = gid[iidx];
1089         /* Update potential energies */
1090         gmx_mm_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid);
1091         gmx_mm_update_1pot_ps(vvdwsum,kernel_data->energygrp_vdw+ggid);
1092
1093         /* Increment number of inner iterations */
1094         inneriter                  += j_index_end - j_index_start;
1095
1096         /* Outer loop uses 20 flops */
1097     }
1098
1099     /* Increment number of outer iterations */
1100     outeriter        += nri;
1101
1102     /* Update outer/inner flops */
1103
1104     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3W3_VF,outeriter*20 + inneriter*294);
1105 }
1106 /*
1107  * Gromacs nonbonded kernel:   nb_kernel_ElecEw_VdwLJEw_GeomW3W3_F_avx_128_fma_single
1108  * Electrostatics interaction: Ewald
1109  * VdW interaction:            LJEwald
1110  * Geometry:                   Water3-Water3
1111  * Calculate force/pot:        Force
1112  */
1113 void
1114 nb_kernel_ElecEw_VdwLJEw_GeomW3W3_F_avx_128_fma_single
1115                     (t_nblist                    * gmx_restrict       nlist,
1116                      rvec                        * gmx_restrict          xx,
1117                      rvec                        * gmx_restrict          ff,
1118                      struct t_forcerec           * gmx_restrict          fr,
1119                      t_mdatoms                   * gmx_restrict     mdatoms,
1120                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
1121                      t_nrnb                      * gmx_restrict        nrnb)
1122 {
1123     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
1124      * just 0 for non-waters.
1125      * Suffixes A,B,C,D refer to j loop unrolling done with AVX_128, e.g. for the four different
1126      * jnr indices corresponding to data put in the four positions in the SIMD register.
1127      */
1128     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
1129     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
1130     int              jnrA,jnrB,jnrC,jnrD;
1131     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
1132     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
1133     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
1134     real             rcutoff_scalar;
1135     real             *shiftvec,*fshift,*x,*f;
1136     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
1137     real             scratch[4*DIM];
1138     __m128           fscal,rcutoff,rcutoff2,jidxall;
1139     int              vdwioffset0;
1140     __m128           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
1141     int              vdwioffset1;
1142     __m128           ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
1143     int              vdwioffset2;
1144     __m128           ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
1145     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
1146     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
1147     int              vdwjidx1A,vdwjidx1B,vdwjidx1C,vdwjidx1D;
1148     __m128           jx1,jy1,jz1,fjx1,fjy1,fjz1,jq1,isaj1;
1149     int              vdwjidx2A,vdwjidx2B,vdwjidx2C,vdwjidx2D;
1150     __m128           jx2,jy2,jz2,fjx2,fjy2,fjz2,jq2,isaj2;
1151     __m128           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
1152     __m128           dx01,dy01,dz01,rsq01,rinv01,rinvsq01,r01,qq01,c6_01,c12_01;
1153     __m128           dx02,dy02,dz02,rsq02,rinv02,rinvsq02,r02,qq02,c6_02,c12_02;
1154     __m128           dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
1155     __m128           dx11,dy11,dz11,rsq11,rinv11,rinvsq11,r11,qq11,c6_11,c12_11;
1156     __m128           dx12,dy12,dz12,rsq12,rinv12,rinvsq12,r12,qq12,c6_12,c12_12;
1157     __m128           dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
1158     __m128           dx21,dy21,dz21,rsq21,rinv21,rinvsq21,r21,qq21,c6_21,c12_21;
1159     __m128           dx22,dy22,dz22,rsq22,rinv22,rinvsq22,r22,qq22,c6_22,c12_22;
1160     __m128           velec,felec,velecsum,facel,crf,krf,krf2;
1161     real             *charge;
1162     int              nvdwtype;
1163     __m128           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
1164     int              *vdwtype;
1165     real             *vdwparam;
1166     __m128           one_sixth   = _mm_set1_ps(1.0/6.0);
1167     __m128           one_twelfth = _mm_set1_ps(1.0/12.0);
1168     __m128           c6grid_00;
1169     __m128           c6grid_01;
1170     __m128           c6grid_02;
1171     __m128           c6grid_10;
1172     __m128           c6grid_11;
1173     __m128           c6grid_12;
1174     __m128           c6grid_20;
1175     __m128           c6grid_21;
1176     __m128           c6grid_22;
1177     real             *vdwgridparam;
1178     __m128           ewclj,ewclj2,ewclj6,ewcljrsq,poly,exponent,f6A,f6B,sh_lj_ewald;
1179     __m128           one_half = _mm_set1_ps(0.5);
1180     __m128           minus_one = _mm_set1_ps(-1.0);
1181     __m128i          ewitab;
1182     __m128           ewtabscale,eweps,twoeweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
1183     __m128           beta,beta2,beta3,zeta2,pmecorrF,pmecorrV,rinv3;
1184     real             *ewtab;
1185     __m128           dummy_mask,cutoff_mask;
1186     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
1187     __m128           one     = _mm_set1_ps(1.0);
1188     __m128           two     = _mm_set1_ps(2.0);
1189     x                = xx[0];
1190     f                = ff[0];
1191
1192     nri              = nlist->nri;
1193     iinr             = nlist->iinr;
1194     jindex           = nlist->jindex;
1195     jjnr             = nlist->jjnr;
1196     shiftidx         = nlist->shift;
1197     gid              = nlist->gid;
1198     shiftvec         = fr->shift_vec[0];
1199     fshift           = fr->fshift[0];
1200     facel            = _mm_set1_ps(fr->ic->epsfac);
1201     charge           = mdatoms->chargeA;
1202     nvdwtype         = fr->ntype;
1203     vdwparam         = fr->nbfp;
1204     vdwtype          = mdatoms->typeA;
1205     vdwgridparam     = fr->ljpme_c6grid;
1206     sh_lj_ewald      = _mm_set1_ps(fr->ic->sh_lj_ewald);
1207     ewclj            = _mm_set1_ps(fr->ic->ewaldcoeff_lj);
1208     ewclj2           = _mm_mul_ps(minus_one,_mm_mul_ps(ewclj,ewclj));
1209
1210     sh_ewald         = _mm_set1_ps(fr->ic->sh_ewald);
1211     beta             = _mm_set1_ps(fr->ic->ewaldcoeff_q);
1212     beta2            = _mm_mul_ps(beta,beta);
1213     beta3            = _mm_mul_ps(beta,beta2);
1214     ewtab            = fr->ic->tabq_coul_F;
1215     ewtabscale       = _mm_set1_ps(fr->ic->tabq_scale);
1216     ewtabhalfspace   = _mm_set1_ps(0.5/fr->ic->tabq_scale);
1217
1218     /* Setup water-specific parameters */
1219     inr              = nlist->iinr[0];
1220     iq0              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+0]));
1221     iq1              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+1]));
1222     iq2              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+2]));
1223     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
1224
1225     jq0              = _mm_set1_ps(charge[inr+0]);
1226     jq1              = _mm_set1_ps(charge[inr+1]);
1227     jq2              = _mm_set1_ps(charge[inr+2]);
1228     vdwjidx0A        = 2*vdwtype[inr+0];
1229     qq00             = _mm_mul_ps(iq0,jq0);
1230     c6_00            = _mm_set1_ps(vdwparam[vdwioffset0+vdwjidx0A]);
1231     c12_00           = _mm_set1_ps(vdwparam[vdwioffset0+vdwjidx0A+1]);
1232     c6grid_00        = _mm_set1_ps(vdwgridparam[vdwioffset0+vdwjidx0A]);
1233     qq01             = _mm_mul_ps(iq0,jq1);
1234     qq02             = _mm_mul_ps(iq0,jq2);
1235     qq10             = _mm_mul_ps(iq1,jq0);
1236     qq11             = _mm_mul_ps(iq1,jq1);
1237     qq12             = _mm_mul_ps(iq1,jq2);
1238     qq20             = _mm_mul_ps(iq2,jq0);
1239     qq21             = _mm_mul_ps(iq2,jq1);
1240     qq22             = _mm_mul_ps(iq2,jq2);
1241
1242     /* Avoid stupid compiler warnings */
1243     jnrA = jnrB = jnrC = jnrD = 0;
1244     j_coord_offsetA = 0;
1245     j_coord_offsetB = 0;
1246     j_coord_offsetC = 0;
1247     j_coord_offsetD = 0;
1248
1249     outeriter        = 0;
1250     inneriter        = 0;
1251
1252     for(iidx=0;iidx<4*DIM;iidx++)
1253     {
1254         scratch[iidx] = 0.0;
1255     }
1256
1257     /* Start outer loop over neighborlists */
1258     for(iidx=0; iidx<nri; iidx++)
1259     {
1260         /* Load shift vector for this list */
1261         i_shift_offset   = DIM*shiftidx[iidx];
1262
1263         /* Load limits for loop over neighbors */
1264         j_index_start    = jindex[iidx];
1265         j_index_end      = jindex[iidx+1];
1266
1267         /* Get outer coordinate index */
1268         inr              = iinr[iidx];
1269         i_coord_offset   = DIM*inr;
1270
1271         /* Load i particle coords and add shift vector */
1272         gmx_mm_load_shift_and_3rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
1273                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
1274
1275         fix0             = _mm_setzero_ps();
1276         fiy0             = _mm_setzero_ps();
1277         fiz0             = _mm_setzero_ps();
1278         fix1             = _mm_setzero_ps();
1279         fiy1             = _mm_setzero_ps();
1280         fiz1             = _mm_setzero_ps();
1281         fix2             = _mm_setzero_ps();
1282         fiy2             = _mm_setzero_ps();
1283         fiz2             = _mm_setzero_ps();
1284
1285         /* Start inner kernel loop */
1286         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
1287         {
1288
1289             /* Get j neighbor index, and coordinate index */
1290             jnrA             = jjnr[jidx];
1291             jnrB             = jjnr[jidx+1];
1292             jnrC             = jjnr[jidx+2];
1293             jnrD             = jjnr[jidx+3];
1294             j_coord_offsetA  = DIM*jnrA;
1295             j_coord_offsetB  = DIM*jnrB;
1296             j_coord_offsetC  = DIM*jnrC;
1297             j_coord_offsetD  = DIM*jnrD;
1298
1299             /* load j atom coordinates */
1300             gmx_mm_load_3rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
1301                                               x+j_coord_offsetC,x+j_coord_offsetD,
1302                                               &jx0,&jy0,&jz0,&jx1,&jy1,&jz1,&jx2,&jy2,&jz2);
1303
1304             /* Calculate displacement vector */
1305             dx00             = _mm_sub_ps(ix0,jx0);
1306             dy00             = _mm_sub_ps(iy0,jy0);
1307             dz00             = _mm_sub_ps(iz0,jz0);
1308             dx01             = _mm_sub_ps(ix0,jx1);
1309             dy01             = _mm_sub_ps(iy0,jy1);
1310             dz01             = _mm_sub_ps(iz0,jz1);
1311             dx02             = _mm_sub_ps(ix0,jx2);
1312             dy02             = _mm_sub_ps(iy0,jy2);
1313             dz02             = _mm_sub_ps(iz0,jz2);
1314             dx10             = _mm_sub_ps(ix1,jx0);
1315             dy10             = _mm_sub_ps(iy1,jy0);
1316             dz10             = _mm_sub_ps(iz1,jz0);
1317             dx11             = _mm_sub_ps(ix1,jx1);
1318             dy11             = _mm_sub_ps(iy1,jy1);
1319             dz11             = _mm_sub_ps(iz1,jz1);
1320             dx12             = _mm_sub_ps(ix1,jx2);
1321             dy12             = _mm_sub_ps(iy1,jy2);
1322             dz12             = _mm_sub_ps(iz1,jz2);
1323             dx20             = _mm_sub_ps(ix2,jx0);
1324             dy20             = _mm_sub_ps(iy2,jy0);
1325             dz20             = _mm_sub_ps(iz2,jz0);
1326             dx21             = _mm_sub_ps(ix2,jx1);
1327             dy21             = _mm_sub_ps(iy2,jy1);
1328             dz21             = _mm_sub_ps(iz2,jz1);
1329             dx22             = _mm_sub_ps(ix2,jx2);
1330             dy22             = _mm_sub_ps(iy2,jy2);
1331             dz22             = _mm_sub_ps(iz2,jz2);
1332
1333             /* Calculate squared distance and things based on it */
1334             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
1335             rsq01            = gmx_mm_calc_rsq_ps(dx01,dy01,dz01);
1336             rsq02            = gmx_mm_calc_rsq_ps(dx02,dy02,dz02);
1337             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
1338             rsq11            = gmx_mm_calc_rsq_ps(dx11,dy11,dz11);
1339             rsq12            = gmx_mm_calc_rsq_ps(dx12,dy12,dz12);
1340             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
1341             rsq21            = gmx_mm_calc_rsq_ps(dx21,dy21,dz21);
1342             rsq22            = gmx_mm_calc_rsq_ps(dx22,dy22,dz22);
1343
1344             rinv00           = avx128fma_invsqrt_f(rsq00);
1345             rinv01           = avx128fma_invsqrt_f(rsq01);
1346             rinv02           = avx128fma_invsqrt_f(rsq02);
1347             rinv10           = avx128fma_invsqrt_f(rsq10);
1348             rinv11           = avx128fma_invsqrt_f(rsq11);
1349             rinv12           = avx128fma_invsqrt_f(rsq12);
1350             rinv20           = avx128fma_invsqrt_f(rsq20);
1351             rinv21           = avx128fma_invsqrt_f(rsq21);
1352             rinv22           = avx128fma_invsqrt_f(rsq22);
1353
1354             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
1355             rinvsq01         = _mm_mul_ps(rinv01,rinv01);
1356             rinvsq02         = _mm_mul_ps(rinv02,rinv02);
1357             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
1358             rinvsq11         = _mm_mul_ps(rinv11,rinv11);
1359             rinvsq12         = _mm_mul_ps(rinv12,rinv12);
1360             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
1361             rinvsq21         = _mm_mul_ps(rinv21,rinv21);
1362             rinvsq22         = _mm_mul_ps(rinv22,rinv22);
1363
1364             fjx0             = _mm_setzero_ps();
1365             fjy0             = _mm_setzero_ps();
1366             fjz0             = _mm_setzero_ps();
1367             fjx1             = _mm_setzero_ps();
1368             fjy1             = _mm_setzero_ps();
1369             fjz1             = _mm_setzero_ps();
1370             fjx2             = _mm_setzero_ps();
1371             fjy2             = _mm_setzero_ps();
1372             fjz2             = _mm_setzero_ps();
1373
1374             /**************************
1375              * CALCULATE INTERACTIONS *
1376              **************************/
1377
1378             r00              = _mm_mul_ps(rsq00,rinv00);
1379
1380             /* EWALD ELECTROSTATICS */
1381
1382             /* Analytical PME correction */
1383             zeta2            = _mm_mul_ps(beta2,rsq00);
1384             rinv3            = _mm_mul_ps(rinvsq00,rinv00);
1385             pmecorrF         = avx128fma_pmecorrF_f(zeta2);
1386             felec            = _mm_macc_ps(pmecorrF,beta3,rinv3);
1387             felec            = _mm_mul_ps(qq00,felec);
1388
1389             /* Analytical LJ-PME */
1390             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
1391             ewcljrsq         = _mm_mul_ps(ewclj2,rsq00);
1392             ewclj6           = _mm_mul_ps(ewclj2,_mm_mul_ps(ewclj2,ewclj2));
1393             exponent         = avx128fma_exp_f(ewcljrsq);
1394             /* poly = exp(-(beta*r)^2) * (1 + (beta*r)^2 + (beta*r)^4 /2) */
1395             poly             = _mm_mul_ps(exponent,_mm_macc_ps(_mm_mul_ps(ewcljrsq,ewcljrsq),one_half,_mm_sub_ps(one,ewcljrsq)));
1396             /* f6A = 6 * C6grid * (1 - poly) */
1397             f6A              = _mm_mul_ps(c6grid_00,_mm_sub_ps(one,poly));
1398             /* f6B = C6grid * exponent * beta^6 */
1399             f6B              = _mm_mul_ps(_mm_mul_ps(c6grid_00,one_sixth),_mm_mul_ps(exponent,ewclj6));
1400             /* fvdw = 12*C12/r13 - ((6*C6 - f6A)/r6 + f6B)/r */
1401             fvdw              = _mm_mul_ps(_mm_macc_ps(_mm_msub_ps(c12_00,rinvsix,_mm_sub_ps(c6_00,f6A)),rinvsix,f6B),rinvsq00);
1402
1403             fscal            = _mm_add_ps(felec,fvdw);
1404
1405              /* Update vectorial force */
1406             fix0             = _mm_macc_ps(dx00,fscal,fix0);
1407             fiy0             = _mm_macc_ps(dy00,fscal,fiy0);
1408             fiz0             = _mm_macc_ps(dz00,fscal,fiz0);
1409
1410             fjx0             = _mm_macc_ps(dx00,fscal,fjx0);
1411             fjy0             = _mm_macc_ps(dy00,fscal,fjy0);
1412             fjz0             = _mm_macc_ps(dz00,fscal,fjz0);
1413
1414             /**************************
1415              * CALCULATE INTERACTIONS *
1416              **************************/
1417
1418             r01              = _mm_mul_ps(rsq01,rinv01);
1419
1420             /* EWALD ELECTROSTATICS */
1421
1422             /* Analytical PME correction */
1423             zeta2            = _mm_mul_ps(beta2,rsq01);
1424             rinv3            = _mm_mul_ps(rinvsq01,rinv01);
1425             pmecorrF         = avx128fma_pmecorrF_f(zeta2);
1426             felec            = _mm_macc_ps(pmecorrF,beta3,rinv3);
1427             felec            = _mm_mul_ps(qq01,felec);
1428
1429             fscal            = felec;
1430
1431              /* Update vectorial force */
1432             fix0             = _mm_macc_ps(dx01,fscal,fix0);
1433             fiy0             = _mm_macc_ps(dy01,fscal,fiy0);
1434             fiz0             = _mm_macc_ps(dz01,fscal,fiz0);
1435
1436             fjx1             = _mm_macc_ps(dx01,fscal,fjx1);
1437             fjy1             = _mm_macc_ps(dy01,fscal,fjy1);
1438             fjz1             = _mm_macc_ps(dz01,fscal,fjz1);
1439
1440             /**************************
1441              * CALCULATE INTERACTIONS *
1442              **************************/
1443
1444             r02              = _mm_mul_ps(rsq02,rinv02);
1445
1446             /* EWALD ELECTROSTATICS */
1447
1448             /* Analytical PME correction */
1449             zeta2            = _mm_mul_ps(beta2,rsq02);
1450             rinv3            = _mm_mul_ps(rinvsq02,rinv02);
1451             pmecorrF         = avx128fma_pmecorrF_f(zeta2);
1452             felec            = _mm_macc_ps(pmecorrF,beta3,rinv3);
1453             felec            = _mm_mul_ps(qq02,felec);
1454
1455             fscal            = felec;
1456
1457              /* Update vectorial force */
1458             fix0             = _mm_macc_ps(dx02,fscal,fix0);
1459             fiy0             = _mm_macc_ps(dy02,fscal,fiy0);
1460             fiz0             = _mm_macc_ps(dz02,fscal,fiz0);
1461
1462             fjx2             = _mm_macc_ps(dx02,fscal,fjx2);
1463             fjy2             = _mm_macc_ps(dy02,fscal,fjy2);
1464             fjz2             = _mm_macc_ps(dz02,fscal,fjz2);
1465
1466             /**************************
1467              * CALCULATE INTERACTIONS *
1468              **************************/
1469
1470             r10              = _mm_mul_ps(rsq10,rinv10);
1471
1472             /* EWALD ELECTROSTATICS */
1473
1474             /* Analytical PME correction */
1475             zeta2            = _mm_mul_ps(beta2,rsq10);
1476             rinv3            = _mm_mul_ps(rinvsq10,rinv10);
1477             pmecorrF         = avx128fma_pmecorrF_f(zeta2);
1478             felec            = _mm_macc_ps(pmecorrF,beta3,rinv3);
1479             felec            = _mm_mul_ps(qq10,felec);
1480
1481             fscal            = felec;
1482
1483              /* Update vectorial force */
1484             fix1             = _mm_macc_ps(dx10,fscal,fix1);
1485             fiy1             = _mm_macc_ps(dy10,fscal,fiy1);
1486             fiz1             = _mm_macc_ps(dz10,fscal,fiz1);
1487
1488             fjx0             = _mm_macc_ps(dx10,fscal,fjx0);
1489             fjy0             = _mm_macc_ps(dy10,fscal,fjy0);
1490             fjz0             = _mm_macc_ps(dz10,fscal,fjz0);
1491
1492             /**************************
1493              * CALCULATE INTERACTIONS *
1494              **************************/
1495
1496             r11              = _mm_mul_ps(rsq11,rinv11);
1497
1498             /* EWALD ELECTROSTATICS */
1499
1500             /* Analytical PME correction */
1501             zeta2            = _mm_mul_ps(beta2,rsq11);
1502             rinv3            = _mm_mul_ps(rinvsq11,rinv11);
1503             pmecorrF         = avx128fma_pmecorrF_f(zeta2);
1504             felec            = _mm_macc_ps(pmecorrF,beta3,rinv3);
1505             felec            = _mm_mul_ps(qq11,felec);
1506
1507             fscal            = felec;
1508
1509              /* Update vectorial force */
1510             fix1             = _mm_macc_ps(dx11,fscal,fix1);
1511             fiy1             = _mm_macc_ps(dy11,fscal,fiy1);
1512             fiz1             = _mm_macc_ps(dz11,fscal,fiz1);
1513
1514             fjx1             = _mm_macc_ps(dx11,fscal,fjx1);
1515             fjy1             = _mm_macc_ps(dy11,fscal,fjy1);
1516             fjz1             = _mm_macc_ps(dz11,fscal,fjz1);
1517
1518             /**************************
1519              * CALCULATE INTERACTIONS *
1520              **************************/
1521
1522             r12              = _mm_mul_ps(rsq12,rinv12);
1523
1524             /* EWALD ELECTROSTATICS */
1525
1526             /* Analytical PME correction */
1527             zeta2            = _mm_mul_ps(beta2,rsq12);
1528             rinv3            = _mm_mul_ps(rinvsq12,rinv12);
1529             pmecorrF         = avx128fma_pmecorrF_f(zeta2);
1530             felec            = _mm_macc_ps(pmecorrF,beta3,rinv3);
1531             felec            = _mm_mul_ps(qq12,felec);
1532
1533             fscal            = felec;
1534
1535              /* Update vectorial force */
1536             fix1             = _mm_macc_ps(dx12,fscal,fix1);
1537             fiy1             = _mm_macc_ps(dy12,fscal,fiy1);
1538             fiz1             = _mm_macc_ps(dz12,fscal,fiz1);
1539
1540             fjx2             = _mm_macc_ps(dx12,fscal,fjx2);
1541             fjy2             = _mm_macc_ps(dy12,fscal,fjy2);
1542             fjz2             = _mm_macc_ps(dz12,fscal,fjz2);
1543
1544             /**************************
1545              * CALCULATE INTERACTIONS *
1546              **************************/
1547
1548             r20              = _mm_mul_ps(rsq20,rinv20);
1549
1550             /* EWALD ELECTROSTATICS */
1551
1552             /* Analytical PME correction */
1553             zeta2            = _mm_mul_ps(beta2,rsq20);
1554             rinv3            = _mm_mul_ps(rinvsq20,rinv20);
1555             pmecorrF         = avx128fma_pmecorrF_f(zeta2);
1556             felec            = _mm_macc_ps(pmecorrF,beta3,rinv3);
1557             felec            = _mm_mul_ps(qq20,felec);
1558
1559             fscal            = felec;
1560
1561              /* Update vectorial force */
1562             fix2             = _mm_macc_ps(dx20,fscal,fix2);
1563             fiy2             = _mm_macc_ps(dy20,fscal,fiy2);
1564             fiz2             = _mm_macc_ps(dz20,fscal,fiz2);
1565
1566             fjx0             = _mm_macc_ps(dx20,fscal,fjx0);
1567             fjy0             = _mm_macc_ps(dy20,fscal,fjy0);
1568             fjz0             = _mm_macc_ps(dz20,fscal,fjz0);
1569
1570             /**************************
1571              * CALCULATE INTERACTIONS *
1572              **************************/
1573
1574             r21              = _mm_mul_ps(rsq21,rinv21);
1575
1576             /* EWALD ELECTROSTATICS */
1577
1578             /* Analytical PME correction */
1579             zeta2            = _mm_mul_ps(beta2,rsq21);
1580             rinv3            = _mm_mul_ps(rinvsq21,rinv21);
1581             pmecorrF         = avx128fma_pmecorrF_f(zeta2);
1582             felec            = _mm_macc_ps(pmecorrF,beta3,rinv3);
1583             felec            = _mm_mul_ps(qq21,felec);
1584
1585             fscal            = felec;
1586
1587              /* Update vectorial force */
1588             fix2             = _mm_macc_ps(dx21,fscal,fix2);
1589             fiy2             = _mm_macc_ps(dy21,fscal,fiy2);
1590             fiz2             = _mm_macc_ps(dz21,fscal,fiz2);
1591
1592             fjx1             = _mm_macc_ps(dx21,fscal,fjx1);
1593             fjy1             = _mm_macc_ps(dy21,fscal,fjy1);
1594             fjz1             = _mm_macc_ps(dz21,fscal,fjz1);
1595
1596             /**************************
1597              * CALCULATE INTERACTIONS *
1598              **************************/
1599
1600             r22              = _mm_mul_ps(rsq22,rinv22);
1601
1602             /* EWALD ELECTROSTATICS */
1603
1604             /* Analytical PME correction */
1605             zeta2            = _mm_mul_ps(beta2,rsq22);
1606             rinv3            = _mm_mul_ps(rinvsq22,rinv22);
1607             pmecorrF         = avx128fma_pmecorrF_f(zeta2);
1608             felec            = _mm_macc_ps(pmecorrF,beta3,rinv3);
1609             felec            = _mm_mul_ps(qq22,felec);
1610
1611             fscal            = felec;
1612
1613              /* Update vectorial force */
1614             fix2             = _mm_macc_ps(dx22,fscal,fix2);
1615             fiy2             = _mm_macc_ps(dy22,fscal,fiy2);
1616             fiz2             = _mm_macc_ps(dz22,fscal,fiz2);
1617
1618             fjx2             = _mm_macc_ps(dx22,fscal,fjx2);
1619             fjy2             = _mm_macc_ps(dy22,fscal,fjy2);
1620             fjz2             = _mm_macc_ps(dz22,fscal,fjz2);
1621
1622             fjptrA             = f+j_coord_offsetA;
1623             fjptrB             = f+j_coord_offsetB;
1624             fjptrC             = f+j_coord_offsetC;
1625             fjptrD             = f+j_coord_offsetD;
1626
1627             gmx_mm_decrement_3rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,
1628                                                    fjx0,fjy0,fjz0,fjx1,fjy1,fjz1,fjx2,fjy2,fjz2);
1629
1630             /* Inner loop uses 273 flops */
1631         }
1632
1633         if(jidx<j_index_end)
1634         {
1635
1636             /* Get j neighbor index, and coordinate index */
1637             jnrlistA         = jjnr[jidx];
1638             jnrlistB         = jjnr[jidx+1];
1639             jnrlistC         = jjnr[jidx+2];
1640             jnrlistD         = jjnr[jidx+3];
1641             /* Sign of each element will be negative for non-real atoms.
1642              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
1643              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
1644              */
1645             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
1646             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
1647             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
1648             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
1649             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
1650             j_coord_offsetA  = DIM*jnrA;
1651             j_coord_offsetB  = DIM*jnrB;
1652             j_coord_offsetC  = DIM*jnrC;
1653             j_coord_offsetD  = DIM*jnrD;
1654
1655             /* load j atom coordinates */
1656             gmx_mm_load_3rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
1657                                               x+j_coord_offsetC,x+j_coord_offsetD,
1658                                               &jx0,&jy0,&jz0,&jx1,&jy1,&jz1,&jx2,&jy2,&jz2);
1659
1660             /* Calculate displacement vector */
1661             dx00             = _mm_sub_ps(ix0,jx0);
1662             dy00             = _mm_sub_ps(iy0,jy0);
1663             dz00             = _mm_sub_ps(iz0,jz0);
1664             dx01             = _mm_sub_ps(ix0,jx1);
1665             dy01             = _mm_sub_ps(iy0,jy1);
1666             dz01             = _mm_sub_ps(iz0,jz1);
1667             dx02             = _mm_sub_ps(ix0,jx2);
1668             dy02             = _mm_sub_ps(iy0,jy2);
1669             dz02             = _mm_sub_ps(iz0,jz2);
1670             dx10             = _mm_sub_ps(ix1,jx0);
1671             dy10             = _mm_sub_ps(iy1,jy0);
1672             dz10             = _mm_sub_ps(iz1,jz0);
1673             dx11             = _mm_sub_ps(ix1,jx1);
1674             dy11             = _mm_sub_ps(iy1,jy1);
1675             dz11             = _mm_sub_ps(iz1,jz1);
1676             dx12             = _mm_sub_ps(ix1,jx2);
1677             dy12             = _mm_sub_ps(iy1,jy2);
1678             dz12             = _mm_sub_ps(iz1,jz2);
1679             dx20             = _mm_sub_ps(ix2,jx0);
1680             dy20             = _mm_sub_ps(iy2,jy0);
1681             dz20             = _mm_sub_ps(iz2,jz0);
1682             dx21             = _mm_sub_ps(ix2,jx1);
1683             dy21             = _mm_sub_ps(iy2,jy1);
1684             dz21             = _mm_sub_ps(iz2,jz1);
1685             dx22             = _mm_sub_ps(ix2,jx2);
1686             dy22             = _mm_sub_ps(iy2,jy2);
1687             dz22             = _mm_sub_ps(iz2,jz2);
1688
1689             /* Calculate squared distance and things based on it */
1690             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
1691             rsq01            = gmx_mm_calc_rsq_ps(dx01,dy01,dz01);
1692             rsq02            = gmx_mm_calc_rsq_ps(dx02,dy02,dz02);
1693             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
1694             rsq11            = gmx_mm_calc_rsq_ps(dx11,dy11,dz11);
1695             rsq12            = gmx_mm_calc_rsq_ps(dx12,dy12,dz12);
1696             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
1697             rsq21            = gmx_mm_calc_rsq_ps(dx21,dy21,dz21);
1698             rsq22            = gmx_mm_calc_rsq_ps(dx22,dy22,dz22);
1699
1700             rinv00           = avx128fma_invsqrt_f(rsq00);
1701             rinv01           = avx128fma_invsqrt_f(rsq01);
1702             rinv02           = avx128fma_invsqrt_f(rsq02);
1703             rinv10           = avx128fma_invsqrt_f(rsq10);
1704             rinv11           = avx128fma_invsqrt_f(rsq11);
1705             rinv12           = avx128fma_invsqrt_f(rsq12);
1706             rinv20           = avx128fma_invsqrt_f(rsq20);
1707             rinv21           = avx128fma_invsqrt_f(rsq21);
1708             rinv22           = avx128fma_invsqrt_f(rsq22);
1709
1710             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
1711             rinvsq01         = _mm_mul_ps(rinv01,rinv01);
1712             rinvsq02         = _mm_mul_ps(rinv02,rinv02);
1713             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
1714             rinvsq11         = _mm_mul_ps(rinv11,rinv11);
1715             rinvsq12         = _mm_mul_ps(rinv12,rinv12);
1716             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
1717             rinvsq21         = _mm_mul_ps(rinv21,rinv21);
1718             rinvsq22         = _mm_mul_ps(rinv22,rinv22);
1719
1720             fjx0             = _mm_setzero_ps();
1721             fjy0             = _mm_setzero_ps();
1722             fjz0             = _mm_setzero_ps();
1723             fjx1             = _mm_setzero_ps();
1724             fjy1             = _mm_setzero_ps();
1725             fjz1             = _mm_setzero_ps();
1726             fjx2             = _mm_setzero_ps();
1727             fjy2             = _mm_setzero_ps();
1728             fjz2             = _mm_setzero_ps();
1729
1730             /**************************
1731              * CALCULATE INTERACTIONS *
1732              **************************/
1733
1734             r00              = _mm_mul_ps(rsq00,rinv00);
1735             r00              = _mm_andnot_ps(dummy_mask,r00);
1736
1737             /* EWALD ELECTROSTATICS */
1738
1739             /* Analytical PME correction */
1740             zeta2            = _mm_mul_ps(beta2,rsq00);
1741             rinv3            = _mm_mul_ps(rinvsq00,rinv00);
1742             pmecorrF         = avx128fma_pmecorrF_f(zeta2);
1743             felec            = _mm_macc_ps(pmecorrF,beta3,rinv3);
1744             felec            = _mm_mul_ps(qq00,felec);
1745
1746             /* Analytical LJ-PME */
1747             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
1748             ewcljrsq         = _mm_mul_ps(ewclj2,rsq00);
1749             ewclj6           = _mm_mul_ps(ewclj2,_mm_mul_ps(ewclj2,ewclj2));
1750             exponent         = avx128fma_exp_f(ewcljrsq);
1751             /* poly = exp(-(beta*r)^2) * (1 + (beta*r)^2 + (beta*r)^4 /2) */
1752             poly             = _mm_mul_ps(exponent,_mm_macc_ps(_mm_mul_ps(ewcljrsq,ewcljrsq),one_half,_mm_sub_ps(one,ewcljrsq)));
1753             /* f6A = 6 * C6grid * (1 - poly) */
1754             f6A              = _mm_mul_ps(c6grid_00,_mm_sub_ps(one,poly));
1755             /* f6B = C6grid * exponent * beta^6 */
1756             f6B              = _mm_mul_ps(_mm_mul_ps(c6grid_00,one_sixth),_mm_mul_ps(exponent,ewclj6));
1757             /* fvdw = 12*C12/r13 - ((6*C6 - f6A)/r6 + f6B)/r */
1758             fvdw              = _mm_mul_ps(_mm_macc_ps(_mm_msub_ps(c12_00,rinvsix,_mm_sub_ps(c6_00,f6A)),rinvsix,f6B),rinvsq00);
1759
1760             fscal            = _mm_add_ps(felec,fvdw);
1761
1762             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1763
1764              /* Update vectorial force */
1765             fix0             = _mm_macc_ps(dx00,fscal,fix0);
1766             fiy0             = _mm_macc_ps(dy00,fscal,fiy0);
1767             fiz0             = _mm_macc_ps(dz00,fscal,fiz0);
1768
1769             fjx0             = _mm_macc_ps(dx00,fscal,fjx0);
1770             fjy0             = _mm_macc_ps(dy00,fscal,fjy0);
1771             fjz0             = _mm_macc_ps(dz00,fscal,fjz0);
1772
1773             /**************************
1774              * CALCULATE INTERACTIONS *
1775              **************************/
1776
1777             r01              = _mm_mul_ps(rsq01,rinv01);
1778             r01              = _mm_andnot_ps(dummy_mask,r01);
1779
1780             /* EWALD ELECTROSTATICS */
1781
1782             /* Analytical PME correction */
1783             zeta2            = _mm_mul_ps(beta2,rsq01);
1784             rinv3            = _mm_mul_ps(rinvsq01,rinv01);
1785             pmecorrF         = avx128fma_pmecorrF_f(zeta2);
1786             felec            = _mm_macc_ps(pmecorrF,beta3,rinv3);
1787             felec            = _mm_mul_ps(qq01,felec);
1788
1789             fscal            = felec;
1790
1791             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1792
1793              /* Update vectorial force */
1794             fix0             = _mm_macc_ps(dx01,fscal,fix0);
1795             fiy0             = _mm_macc_ps(dy01,fscal,fiy0);
1796             fiz0             = _mm_macc_ps(dz01,fscal,fiz0);
1797
1798             fjx1             = _mm_macc_ps(dx01,fscal,fjx1);
1799             fjy1             = _mm_macc_ps(dy01,fscal,fjy1);
1800             fjz1             = _mm_macc_ps(dz01,fscal,fjz1);
1801
1802             /**************************
1803              * CALCULATE INTERACTIONS *
1804              **************************/
1805
1806             r02              = _mm_mul_ps(rsq02,rinv02);
1807             r02              = _mm_andnot_ps(dummy_mask,r02);
1808
1809             /* EWALD ELECTROSTATICS */
1810
1811             /* Analytical PME correction */
1812             zeta2            = _mm_mul_ps(beta2,rsq02);
1813             rinv3            = _mm_mul_ps(rinvsq02,rinv02);
1814             pmecorrF         = avx128fma_pmecorrF_f(zeta2);
1815             felec            = _mm_macc_ps(pmecorrF,beta3,rinv3);
1816             felec            = _mm_mul_ps(qq02,felec);
1817
1818             fscal            = felec;
1819
1820             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1821
1822              /* Update vectorial force */
1823             fix0             = _mm_macc_ps(dx02,fscal,fix0);
1824             fiy0             = _mm_macc_ps(dy02,fscal,fiy0);
1825             fiz0             = _mm_macc_ps(dz02,fscal,fiz0);
1826
1827             fjx2             = _mm_macc_ps(dx02,fscal,fjx2);
1828             fjy2             = _mm_macc_ps(dy02,fscal,fjy2);
1829             fjz2             = _mm_macc_ps(dz02,fscal,fjz2);
1830
1831             /**************************
1832              * CALCULATE INTERACTIONS *
1833              **************************/
1834
1835             r10              = _mm_mul_ps(rsq10,rinv10);
1836             r10              = _mm_andnot_ps(dummy_mask,r10);
1837
1838             /* EWALD ELECTROSTATICS */
1839
1840             /* Analytical PME correction */
1841             zeta2            = _mm_mul_ps(beta2,rsq10);
1842             rinv3            = _mm_mul_ps(rinvsq10,rinv10);
1843             pmecorrF         = avx128fma_pmecorrF_f(zeta2);
1844             felec            = _mm_macc_ps(pmecorrF,beta3,rinv3);
1845             felec            = _mm_mul_ps(qq10,felec);
1846
1847             fscal            = felec;
1848
1849             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1850
1851              /* Update vectorial force */
1852             fix1             = _mm_macc_ps(dx10,fscal,fix1);
1853             fiy1             = _mm_macc_ps(dy10,fscal,fiy1);
1854             fiz1             = _mm_macc_ps(dz10,fscal,fiz1);
1855
1856             fjx0             = _mm_macc_ps(dx10,fscal,fjx0);
1857             fjy0             = _mm_macc_ps(dy10,fscal,fjy0);
1858             fjz0             = _mm_macc_ps(dz10,fscal,fjz0);
1859
1860             /**************************
1861              * CALCULATE INTERACTIONS *
1862              **************************/
1863
1864             r11              = _mm_mul_ps(rsq11,rinv11);
1865             r11              = _mm_andnot_ps(dummy_mask,r11);
1866
1867             /* EWALD ELECTROSTATICS */
1868
1869             /* Analytical PME correction */
1870             zeta2            = _mm_mul_ps(beta2,rsq11);
1871             rinv3            = _mm_mul_ps(rinvsq11,rinv11);
1872             pmecorrF         = avx128fma_pmecorrF_f(zeta2);
1873             felec            = _mm_macc_ps(pmecorrF,beta3,rinv3);
1874             felec            = _mm_mul_ps(qq11,felec);
1875
1876             fscal            = felec;
1877
1878             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1879
1880              /* Update vectorial force */
1881             fix1             = _mm_macc_ps(dx11,fscal,fix1);
1882             fiy1             = _mm_macc_ps(dy11,fscal,fiy1);
1883             fiz1             = _mm_macc_ps(dz11,fscal,fiz1);
1884
1885             fjx1             = _mm_macc_ps(dx11,fscal,fjx1);
1886             fjy1             = _mm_macc_ps(dy11,fscal,fjy1);
1887             fjz1             = _mm_macc_ps(dz11,fscal,fjz1);
1888
1889             /**************************
1890              * CALCULATE INTERACTIONS *
1891              **************************/
1892
1893             r12              = _mm_mul_ps(rsq12,rinv12);
1894             r12              = _mm_andnot_ps(dummy_mask,r12);
1895
1896             /* EWALD ELECTROSTATICS */
1897
1898             /* Analytical PME correction */
1899             zeta2            = _mm_mul_ps(beta2,rsq12);
1900             rinv3            = _mm_mul_ps(rinvsq12,rinv12);
1901             pmecorrF         = avx128fma_pmecorrF_f(zeta2);
1902             felec            = _mm_macc_ps(pmecorrF,beta3,rinv3);
1903             felec            = _mm_mul_ps(qq12,felec);
1904
1905             fscal            = felec;
1906
1907             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1908
1909              /* Update vectorial force */
1910             fix1             = _mm_macc_ps(dx12,fscal,fix1);
1911             fiy1             = _mm_macc_ps(dy12,fscal,fiy1);
1912             fiz1             = _mm_macc_ps(dz12,fscal,fiz1);
1913
1914             fjx2             = _mm_macc_ps(dx12,fscal,fjx2);
1915             fjy2             = _mm_macc_ps(dy12,fscal,fjy2);
1916             fjz2             = _mm_macc_ps(dz12,fscal,fjz2);
1917
1918             /**************************
1919              * CALCULATE INTERACTIONS *
1920              **************************/
1921
1922             r20              = _mm_mul_ps(rsq20,rinv20);
1923             r20              = _mm_andnot_ps(dummy_mask,r20);
1924
1925             /* EWALD ELECTROSTATICS */
1926
1927             /* Analytical PME correction */
1928             zeta2            = _mm_mul_ps(beta2,rsq20);
1929             rinv3            = _mm_mul_ps(rinvsq20,rinv20);
1930             pmecorrF         = avx128fma_pmecorrF_f(zeta2);
1931             felec            = _mm_macc_ps(pmecorrF,beta3,rinv3);
1932             felec            = _mm_mul_ps(qq20,felec);
1933
1934             fscal            = felec;
1935
1936             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1937
1938              /* Update vectorial force */
1939             fix2             = _mm_macc_ps(dx20,fscal,fix2);
1940             fiy2             = _mm_macc_ps(dy20,fscal,fiy2);
1941             fiz2             = _mm_macc_ps(dz20,fscal,fiz2);
1942
1943             fjx0             = _mm_macc_ps(dx20,fscal,fjx0);
1944             fjy0             = _mm_macc_ps(dy20,fscal,fjy0);
1945             fjz0             = _mm_macc_ps(dz20,fscal,fjz0);
1946
1947             /**************************
1948              * CALCULATE INTERACTIONS *
1949              **************************/
1950
1951             r21              = _mm_mul_ps(rsq21,rinv21);
1952             r21              = _mm_andnot_ps(dummy_mask,r21);
1953
1954             /* EWALD ELECTROSTATICS */
1955
1956             /* Analytical PME correction */
1957             zeta2            = _mm_mul_ps(beta2,rsq21);
1958             rinv3            = _mm_mul_ps(rinvsq21,rinv21);
1959             pmecorrF         = avx128fma_pmecorrF_f(zeta2);
1960             felec            = _mm_macc_ps(pmecorrF,beta3,rinv3);
1961             felec            = _mm_mul_ps(qq21,felec);
1962
1963             fscal            = felec;
1964
1965             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1966
1967              /* Update vectorial force */
1968             fix2             = _mm_macc_ps(dx21,fscal,fix2);
1969             fiy2             = _mm_macc_ps(dy21,fscal,fiy2);
1970             fiz2             = _mm_macc_ps(dz21,fscal,fiz2);
1971
1972             fjx1             = _mm_macc_ps(dx21,fscal,fjx1);
1973             fjy1             = _mm_macc_ps(dy21,fscal,fjy1);
1974             fjz1             = _mm_macc_ps(dz21,fscal,fjz1);
1975
1976             /**************************
1977              * CALCULATE INTERACTIONS *
1978              **************************/
1979
1980             r22              = _mm_mul_ps(rsq22,rinv22);
1981             r22              = _mm_andnot_ps(dummy_mask,r22);
1982
1983             /* EWALD ELECTROSTATICS */
1984
1985             /* Analytical PME correction */
1986             zeta2            = _mm_mul_ps(beta2,rsq22);
1987             rinv3            = _mm_mul_ps(rinvsq22,rinv22);
1988             pmecorrF         = avx128fma_pmecorrF_f(zeta2);
1989             felec            = _mm_macc_ps(pmecorrF,beta3,rinv3);
1990             felec            = _mm_mul_ps(qq22,felec);
1991
1992             fscal            = felec;
1993
1994             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1995
1996              /* Update vectorial force */
1997             fix2             = _mm_macc_ps(dx22,fscal,fix2);
1998             fiy2             = _mm_macc_ps(dy22,fscal,fiy2);
1999             fiz2             = _mm_macc_ps(dz22,fscal,fiz2);
2000
2001             fjx2             = _mm_macc_ps(dx22,fscal,fjx2);
2002             fjy2             = _mm_macc_ps(dy22,fscal,fjy2);
2003             fjz2             = _mm_macc_ps(dz22,fscal,fjz2);
2004
2005             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
2006             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
2007             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
2008             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
2009
2010             gmx_mm_decrement_3rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,
2011                                                    fjx0,fjy0,fjz0,fjx1,fjy1,fjz1,fjx2,fjy2,fjz2);
2012
2013             /* Inner loop uses 282 flops */
2014         }
2015
2016         /* End of innermost loop */
2017
2018         gmx_mm_update_iforce_3atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
2019                                               f+i_coord_offset,fshift+i_shift_offset);
2020
2021         /* Increment number of inner iterations */
2022         inneriter                  += j_index_end - j_index_start;
2023
2024         /* Outer loop uses 18 flops */
2025     }
2026
2027     /* Increment number of outer iterations */
2028     outeriter        += nri;
2029
2030     /* Update outer/inner flops */
2031
2032     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3W3_F,outeriter*18 + inneriter*282);
2033 }