src/gromacs/gmxlib/nonbonded/nb_kernel_avx_128_fma_single/nb_kernel_ElecEw_VdwLJEw_GeomW4W4_avx_128_fma_single.c

   1 /*
   2  * This file is part of the GROMACS molecular simulation package.
   3  *
   4  * Copyright (c) 2012,2013,2014,2015,2017, 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_GeomW4W4_VF_avx_128_fma_single
  51  * Electrostatics interaction: Ewald
  52  * VdW interaction:            LJEwald
  53  * Geometry:                   Water4-Water4
  54  * Calculate force/pot:        PotentialAndForce
  55  */
  56 void
  57 nb_kernel_ElecEw_VdwLJEw_GeomW4W4_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              vdwioffset3;
  89     __m128           ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
  90     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
  91     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  92     int              vdwjidx1A,vdwjidx1B,vdwjidx1C,vdwjidx1D;
  93     __m128           jx1,jy1,jz1,fjx1,fjy1,fjz1,jq1,isaj1;
  94     int              vdwjidx2A,vdwjidx2B,vdwjidx2C,vdwjidx2D;
  95     __m128           jx2,jy2,jz2,fjx2,fjy2,fjz2,jq2,isaj2;
  96     int              vdwjidx3A,vdwjidx3B,vdwjidx3C,vdwjidx3D;
  97     __m128           jx3,jy3,jz3,fjx3,fjy3,fjz3,jq3,isaj3;
  98     __m128           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  99     __m128           dx11,dy11,dz11,rsq11,rinv11,rinvsq11,r11,qq11,c6_11,c12_11;
 100     __m128           dx12,dy12,dz12,rsq12,rinv12,rinvsq12,r12,qq12,c6_12,c12_12;
 101     __m128           dx13,dy13,dz13,rsq13,rinv13,rinvsq13,r13,qq13,c6_13,c12_13;
 102     __m128           dx21,dy21,dz21,rsq21,rinv21,rinvsq21,r21,qq21,c6_21,c12_21;
 103     __m128           dx22,dy22,dz22,rsq22,rinv22,rinvsq22,r22,qq22,c6_22,c12_22;
 104     __m128           dx23,dy23,dz23,rsq23,rinv23,rinvsq23,r23,qq23,c6_23,c12_23;
 105     __m128           dx31,dy31,dz31,rsq31,rinv31,rinvsq31,r31,qq31,c6_31,c12_31;
 106     __m128           dx32,dy32,dz32,rsq32,rinv32,rinvsq32,r32,qq32,c6_32,c12_32;
 107     __m128           dx33,dy33,dz33,rsq33,rinv33,rinvsq33,r33,qq33,c6_33,c12_33;
 108     __m128           velec,felec,velecsum,facel,crf,krf,krf2;
 109     real             *charge;
 110     int              nvdwtype;
 111     __m128           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 112     int              *vdwtype;
 113     real             *vdwparam;
 114     __m128           one_sixth   = _mm_set1_ps(1.0/6.0);
 115     __m128           one_twelfth = _mm_set1_ps(1.0/12.0);
 116     __m128           c6grid_00;
 117     __m128           c6grid_11;
 118     __m128           c6grid_12;
 119     __m128           c6grid_13;
 120     __m128           c6grid_21;
 121     __m128           c6grid_22;
 122     __m128           c6grid_23;
 123     __m128           c6grid_31;
 124     __m128           c6grid_32;
 125     __m128           c6grid_33;
 126     real             *vdwgridparam;
 127     __m128           ewclj,ewclj2,ewclj6,ewcljrsq,poly,exponent,f6A,f6B,sh_lj_ewald;
 128     __m128           one_half = _mm_set1_ps(0.5);
 129     __m128           minus_one = _mm_set1_ps(-1.0);
 130     __m128i          ewitab;
 131     __m128           ewtabscale,eweps,twoeweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
 132     __m128           beta,beta2,beta3,zeta2,pmecorrF,pmecorrV,rinv3;
 133     real             *ewtab;
 134     __m128           dummy_mask,cutoff_mask;
 135     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
 136     __m128           one     = _mm_set1_ps(1.0);
 137     __m128           two     = _mm_set1_ps(2.0);
 138     x                = xx[0];
 139     f                = ff[0];
 140
 141     nri              = nlist->nri;
 142     iinr             = nlist->iinr;
 143     jindex           = nlist->jindex;
 144     jjnr             = nlist->jjnr;
 145     shiftidx         = nlist->shift;
 146     gid              = nlist->gid;
 147     shiftvec         = fr->shift_vec[0];
 148     fshift           = fr->fshift[0];
 149     facel            = _mm_set1_ps(fr->ic->epsfac);
 150     charge           = mdatoms->chargeA;
 151     nvdwtype         = fr->ntype;
 152     vdwparam         = fr->nbfp;
 153     vdwtype          = mdatoms->typeA;
 154     vdwgridparam     = fr->ljpme_c6grid;
 155     sh_lj_ewald      = _mm_set1_ps(fr->ic->sh_lj_ewald);
 156     ewclj            = _mm_set1_ps(fr->ic->ewaldcoeff_lj);
 157     ewclj2           = _mm_mul_ps(minus_one,_mm_mul_ps(ewclj,ewclj));
 158
 159     sh_ewald         = _mm_set1_ps(fr->ic->sh_ewald);
 160     beta             = _mm_set1_ps(fr->ic->ewaldcoeff_q);
 161     beta2            = _mm_mul_ps(beta,beta);
 162     beta3            = _mm_mul_ps(beta,beta2);
 163     ewtab            = fr->ic->tabq_coul_FDV0;
 164     ewtabscale       = _mm_set1_ps(fr->ic->tabq_scale);
 165     ewtabhalfspace   = _mm_set1_ps(0.5/fr->ic->tabq_scale);
 166
 167     /* Setup water-specific parameters */
 168     inr              = nlist->iinr[0];
 169     iq1              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+1]));
 170     iq2              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+2]));
 171     iq3              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+3]));
 172     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 173
 174     jq1              = _mm_set1_ps(charge[inr+1]);
 175     jq2              = _mm_set1_ps(charge[inr+2]);
 176     jq3              = _mm_set1_ps(charge[inr+3]);
 177     vdwjidx0A        = 2*vdwtype[inr+0];
 178     c6_00            = _mm_set1_ps(vdwparam[vdwioffset0+vdwjidx0A]);
 179     c12_00           = _mm_set1_ps(vdwparam[vdwioffset0+vdwjidx0A+1]);
 180     c6grid_00        = _mm_set1_ps(vdwgridparam[vdwioffset0+vdwjidx0A]);
 181     qq11             = _mm_mul_ps(iq1,jq1);
 182     qq12             = _mm_mul_ps(iq1,jq2);
 183     qq13             = _mm_mul_ps(iq1,jq3);
 184     qq21             = _mm_mul_ps(iq2,jq1);
 185     qq22             = _mm_mul_ps(iq2,jq2);
 186     qq23             = _mm_mul_ps(iq2,jq3);
 187     qq31             = _mm_mul_ps(iq3,jq1);
 188     qq32             = _mm_mul_ps(iq3,jq2);
 189     qq33             = _mm_mul_ps(iq3,jq3);
 190
 191     /* Avoid stupid compiler warnings */
 192     jnrA = jnrB = jnrC = jnrD = 0;
 193     j_coord_offsetA = 0;
 194     j_coord_offsetB = 0;
 195     j_coord_offsetC = 0;
 196     j_coord_offsetD = 0;
 197
 198     outeriter        = 0;
 199     inneriter        = 0;
 200
 201     for(iidx=0;iidx<4*DIM;iidx++)
 202     {
 203         scratch[iidx] = 0.0;
 204     }
 205
 206     /* Start outer loop over neighborlists */
 207     for(iidx=0; iidx<nri; iidx++)
 208     {
 209         /* Load shift vector for this list */
 210         i_shift_offset   = DIM*shiftidx[iidx];
 211
 212         /* Load limits for loop over neighbors */
 213         j_index_start    = jindex[iidx];
 214         j_index_end      = jindex[iidx+1];
 215
 216         /* Get outer coordinate index */
 217         inr              = iinr[iidx];
 218         i_coord_offset   = DIM*inr;
 219
 220         /* Load i particle coords and add shift vector */
 221         gmx_mm_load_shift_and_4rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
 222                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
 223
 224         fix0             = _mm_setzero_ps();
 225         fiy0             = _mm_setzero_ps();
 226         fiz0             = _mm_setzero_ps();
 227         fix1             = _mm_setzero_ps();
 228         fiy1             = _mm_setzero_ps();
 229         fiz1             = _mm_setzero_ps();
 230         fix2             = _mm_setzero_ps();
 231         fiy2             = _mm_setzero_ps();
 232         fiz2             = _mm_setzero_ps();
 233         fix3             = _mm_setzero_ps();
 234         fiy3             = _mm_setzero_ps();
 235         fiz3             = _mm_setzero_ps();
 236
 237         /* Reset potential sums */
 238         velecsum         = _mm_setzero_ps();
 239         vvdwsum          = _mm_setzero_ps();
 240
 241         /* Start inner kernel loop */
 242         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 243         {
 244
 245             /* Get j neighbor index, and coordinate index */
 246             jnrA             = jjnr[jidx];
 247             jnrB             = jjnr[jidx+1];
 248             jnrC             = jjnr[jidx+2];
 249             jnrD             = jjnr[jidx+3];
 250             j_coord_offsetA  = DIM*jnrA;
 251             j_coord_offsetB  = DIM*jnrB;
 252             j_coord_offsetC  = DIM*jnrC;
 253             j_coord_offsetD  = DIM*jnrD;
 254
 255             /* load j atom coordinates */
 256             gmx_mm_load_4rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 257                                               x+j_coord_offsetC,x+j_coord_offsetD,
 258                                               &jx0,&jy0,&jz0,&jx1,&jy1,&jz1,&jx2,
 259                                               &jy2,&jz2,&jx3,&jy3,&jz3);
 260
 261             /* Calculate displacement vector */
 262             dx00             = _mm_sub_ps(ix0,jx0);
 263             dy00             = _mm_sub_ps(iy0,jy0);
 264             dz00             = _mm_sub_ps(iz0,jz0);
 265             dx11             = _mm_sub_ps(ix1,jx1);
 266             dy11             = _mm_sub_ps(iy1,jy1);
 267             dz11             = _mm_sub_ps(iz1,jz1);
 268             dx12             = _mm_sub_ps(ix1,jx2);
 269             dy12             = _mm_sub_ps(iy1,jy2);
 270             dz12             = _mm_sub_ps(iz1,jz2);
 271             dx13             = _mm_sub_ps(ix1,jx3);
 272             dy13             = _mm_sub_ps(iy1,jy3);
 273             dz13             = _mm_sub_ps(iz1,jz3);
 274             dx21             = _mm_sub_ps(ix2,jx1);
 275             dy21             = _mm_sub_ps(iy2,jy1);
 276             dz21             = _mm_sub_ps(iz2,jz1);
 277             dx22             = _mm_sub_ps(ix2,jx2);
 278             dy22             = _mm_sub_ps(iy2,jy2);
 279             dz22             = _mm_sub_ps(iz2,jz2);
 280             dx23             = _mm_sub_ps(ix2,jx3);
 281             dy23             = _mm_sub_ps(iy2,jy3);
 282             dz23             = _mm_sub_ps(iz2,jz3);
 283             dx31             = _mm_sub_ps(ix3,jx1);
 284             dy31             = _mm_sub_ps(iy3,jy1);
 285             dz31             = _mm_sub_ps(iz3,jz1);
 286             dx32             = _mm_sub_ps(ix3,jx2);
 287             dy32             = _mm_sub_ps(iy3,jy2);
 288             dz32             = _mm_sub_ps(iz3,jz2);
 289             dx33             = _mm_sub_ps(ix3,jx3);
 290             dy33             = _mm_sub_ps(iy3,jy3);
 291             dz33             = _mm_sub_ps(iz3,jz3);
 292
 293             /* Calculate squared distance and things based on it */
 294             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 295             rsq11            = gmx_mm_calc_rsq_ps(dx11,dy11,dz11);
 296             rsq12            = gmx_mm_calc_rsq_ps(dx12,dy12,dz12);
 297             rsq13            = gmx_mm_calc_rsq_ps(dx13,dy13,dz13);
 298             rsq21            = gmx_mm_calc_rsq_ps(dx21,dy21,dz21);
 299             rsq22            = gmx_mm_calc_rsq_ps(dx22,dy22,dz22);
 300             rsq23            = gmx_mm_calc_rsq_ps(dx23,dy23,dz23);
 301             rsq31            = gmx_mm_calc_rsq_ps(dx31,dy31,dz31);
 302             rsq32            = gmx_mm_calc_rsq_ps(dx32,dy32,dz32);
 303             rsq33            = gmx_mm_calc_rsq_ps(dx33,dy33,dz33);
 304
 305             rinv00           = avx128fma_invsqrt_f(rsq00);
 306             rinv11           = avx128fma_invsqrt_f(rsq11);
 307             rinv12           = avx128fma_invsqrt_f(rsq12);
 308             rinv13           = avx128fma_invsqrt_f(rsq13);
 309             rinv21           = avx128fma_invsqrt_f(rsq21);
 310             rinv22           = avx128fma_invsqrt_f(rsq22);
 311             rinv23           = avx128fma_invsqrt_f(rsq23);
 312             rinv31           = avx128fma_invsqrt_f(rsq31);
 313             rinv32           = avx128fma_invsqrt_f(rsq32);
 314             rinv33           = avx128fma_invsqrt_f(rsq33);
 315
 316             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 317             rinvsq11         = _mm_mul_ps(rinv11,rinv11);
 318             rinvsq12         = _mm_mul_ps(rinv12,rinv12);
 319             rinvsq13         = _mm_mul_ps(rinv13,rinv13);
 320             rinvsq21         = _mm_mul_ps(rinv21,rinv21);
 321             rinvsq22         = _mm_mul_ps(rinv22,rinv22);
 322             rinvsq23         = _mm_mul_ps(rinv23,rinv23);
 323             rinvsq31         = _mm_mul_ps(rinv31,rinv31);
 324             rinvsq32         = _mm_mul_ps(rinv32,rinv32);
 325             rinvsq33         = _mm_mul_ps(rinv33,rinv33);
 326
 327             fjx0             = _mm_setzero_ps();
 328             fjy0             = _mm_setzero_ps();
 329             fjz0             = _mm_setzero_ps();
 330             fjx1             = _mm_setzero_ps();
 331             fjy1             = _mm_setzero_ps();
 332             fjz1             = _mm_setzero_ps();
 333             fjx2             = _mm_setzero_ps();
 334             fjy2             = _mm_setzero_ps();
 335             fjz2             = _mm_setzero_ps();
 336             fjx3             = _mm_setzero_ps();
 337             fjy3             = _mm_setzero_ps();
 338             fjz3             = _mm_setzero_ps();
 339
 340             /**************************
 341              * CALCULATE INTERACTIONS *
 342              **************************/
 343
 344             r00              = _mm_mul_ps(rsq00,rinv00);
 345
 346             /* Analytical LJ-PME */
 347             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 348             ewcljrsq         = _mm_mul_ps(ewclj2,rsq00);
 349             ewclj6           = _mm_mul_ps(ewclj2,_mm_mul_ps(ewclj2,ewclj2));
 350             exponent         = avx128fma_exp_f(ewcljrsq);
 351             /* poly = exp(-(beta*r)^2) * (1 + (beta*r)^2 + (beta*r)^4 /2) */
 352             poly             = _mm_mul_ps(exponent,_mm_macc_ps(_mm_mul_ps(ewcljrsq,ewcljrsq),one_half,_mm_sub_ps(one,ewcljrsq)));
 353             /* vvdw6 = [C6 - C6grid * (1-poly)]/r6 */
 354             vvdw6            = _mm_mul_ps(_mm_macc_ps(-c6grid_00,_mm_sub_ps(one,poly),c6_00),rinvsix);
 355             vvdw12           = _mm_mul_ps(c12_00,_mm_mul_ps(rinvsix,rinvsix));
 356             vvdw             = _mm_msub_ps(vvdw12,one_twelfth,_mm_mul_ps(vvdw6,one_sixth));
 357             /* fvdw = vvdw12/r - (vvdw6/r + (C6grid * exponent * beta^6)/r) */
 358             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);
 359
 360             /* Update potential sum for this i atom from the interaction with this j atom. */
 361             vvdwsum          = _mm_add_ps(vvdwsum,vvdw);
 362
 363             fscal            = fvdw;
 364
 365              /* Update vectorial force */
 366             fix0             = _mm_macc_ps(dx00,fscal,fix0);
 367             fiy0             = _mm_macc_ps(dy00,fscal,fiy0);
 368             fiz0             = _mm_macc_ps(dz00,fscal,fiz0);
 369
 370             fjx0             = _mm_macc_ps(dx00,fscal,fjx0);
 371             fjy0             = _mm_macc_ps(dy00,fscal,fjy0);
 372             fjz0             = _mm_macc_ps(dz00,fscal,fjz0);
 373
 374             /**************************
 375              * CALCULATE INTERACTIONS *
 376              **************************/
 377
 378             r11              = _mm_mul_ps(rsq11,rinv11);
 379
 380             /* EWALD ELECTROSTATICS */
 381
 382             /* Analytical PME correction */
 383             zeta2            = _mm_mul_ps(beta2,rsq11);
 384             rinv3            = _mm_mul_ps(rinvsq11,rinv11);
 385             pmecorrF         = avx128fma_pmecorrF_f(zeta2);
 386             felec            = _mm_macc_ps(pmecorrF,beta3,rinv3);
 387             felec            = _mm_mul_ps(qq11,felec);
 388             pmecorrV         = avx128fma_pmecorrV_f(zeta2);
 389             velec            = _mm_nmacc_ps(pmecorrV,beta,rinv11);
 390             velec            = _mm_mul_ps(qq11,velec);
 391
 392             /* Update potential sum for this i atom from the interaction with this j atom. */
 393             velecsum         = _mm_add_ps(velecsum,velec);
 394
 395             fscal            = felec;
 396
 397              /* Update vectorial force */
 398             fix1             = _mm_macc_ps(dx11,fscal,fix1);
 399             fiy1             = _mm_macc_ps(dy11,fscal,fiy1);
 400             fiz1             = _mm_macc_ps(dz11,fscal,fiz1);
 401
 402             fjx1             = _mm_macc_ps(dx11,fscal,fjx1);
 403             fjy1             = _mm_macc_ps(dy11,fscal,fjy1);
 404             fjz1             = _mm_macc_ps(dz11,fscal,fjz1);
 405
 406             /**************************
 407              * CALCULATE INTERACTIONS *
 408              **************************/
 409
 410             r12              = _mm_mul_ps(rsq12,rinv12);
 411
 412             /* EWALD ELECTROSTATICS */
 413
 414             /* Analytical PME correction */
 415             zeta2            = _mm_mul_ps(beta2,rsq12);
 416             rinv3            = _mm_mul_ps(rinvsq12,rinv12);
 417             pmecorrF         = avx128fma_pmecorrF_f(zeta2);
 418             felec            = _mm_macc_ps(pmecorrF,beta3,rinv3);
 419             felec            = _mm_mul_ps(qq12,felec);
 420             pmecorrV         = avx128fma_pmecorrV_f(zeta2);
 421             velec            = _mm_nmacc_ps(pmecorrV,beta,rinv12);
 422             velec            = _mm_mul_ps(qq12,velec);
 423
 424             /* Update potential sum for this i atom from the interaction with this j atom. */
 425             velecsum         = _mm_add_ps(velecsum,velec);
 426
 427             fscal            = felec;
 428
 429              /* Update vectorial force */
 430             fix1             = _mm_macc_ps(dx12,fscal,fix1);
 431             fiy1             = _mm_macc_ps(dy12,fscal,fiy1);
 432             fiz1             = _mm_macc_ps(dz12,fscal,fiz1);
 433
 434             fjx2             = _mm_macc_ps(dx12,fscal,fjx2);
 435             fjy2             = _mm_macc_ps(dy12,fscal,fjy2);
 436             fjz2             = _mm_macc_ps(dz12,fscal,fjz2);
 437
 438             /**************************
 439              * CALCULATE INTERACTIONS *
 440              **************************/
 441
 442             r13              = _mm_mul_ps(rsq13,rinv13);
 443
 444             /* EWALD ELECTROSTATICS */
 445
 446             /* Analytical PME correction */
 447             zeta2            = _mm_mul_ps(beta2,rsq13);
 448             rinv3            = _mm_mul_ps(rinvsq13,rinv13);
 449             pmecorrF         = avx128fma_pmecorrF_f(zeta2);
 450             felec            = _mm_macc_ps(pmecorrF,beta3,rinv3);
 451             felec            = _mm_mul_ps(qq13,felec);
 452             pmecorrV         = avx128fma_pmecorrV_f(zeta2);
 453             velec            = _mm_nmacc_ps(pmecorrV,beta,rinv13);
 454             velec            = _mm_mul_ps(qq13,velec);
 455
 456             /* Update potential sum for this i atom from the interaction with this j atom. */
 457             velecsum         = _mm_add_ps(velecsum,velec);
 458
 459             fscal            = felec;
 460
 461              /* Update vectorial force */
 462             fix1             = _mm_macc_ps(dx13,fscal,fix1);
 463             fiy1             = _mm_macc_ps(dy13,fscal,fiy1);
 464             fiz1             = _mm_macc_ps(dz13,fscal,fiz1);
 465
 466             fjx3             = _mm_macc_ps(dx13,fscal,fjx3);
 467             fjy3             = _mm_macc_ps(dy13,fscal,fjy3);
 468             fjz3             = _mm_macc_ps(dz13,fscal,fjz3);
 469
 470             /**************************
 471              * CALCULATE INTERACTIONS *
 472              **************************/
 473
 474             r21              = _mm_mul_ps(rsq21,rinv21);
 475
 476             /* EWALD ELECTROSTATICS */
 477
 478             /* Analytical PME correction */
 479             zeta2            = _mm_mul_ps(beta2,rsq21);
 480             rinv3            = _mm_mul_ps(rinvsq21,rinv21);
 481             pmecorrF         = avx128fma_pmecorrF_f(zeta2);
 482             felec            = _mm_macc_ps(pmecorrF,beta3,rinv3);
 483             felec            = _mm_mul_ps(qq21,felec);
 484             pmecorrV         = avx128fma_pmecorrV_f(zeta2);
 485             velec            = _mm_nmacc_ps(pmecorrV,beta,rinv21);
 486             velec            = _mm_mul_ps(qq21,velec);
 487
 488             /* Update potential sum for this i atom from the interaction with this j atom. */
 489             velecsum         = _mm_add_ps(velecsum,velec);
 490
 491             fscal            = felec;
 492
 493              /* Update vectorial force */
 494             fix2             = _mm_macc_ps(dx21,fscal,fix2);
 495             fiy2             = _mm_macc_ps(dy21,fscal,fiy2);
 496             fiz2             = _mm_macc_ps(dz21,fscal,fiz2);
 497
 498             fjx1             = _mm_macc_ps(dx21,fscal,fjx1);
 499             fjy1             = _mm_macc_ps(dy21,fscal,fjy1);
 500             fjz1             = _mm_macc_ps(dz21,fscal,fjz1);
 501
 502             /**************************
 503              * CALCULATE INTERACTIONS *
 504              **************************/
 505
 506             r22              = _mm_mul_ps(rsq22,rinv22);
 507
 508             /* EWALD ELECTROSTATICS */
 509
 510             /* Analytical PME correction */
 511             zeta2            = _mm_mul_ps(beta2,rsq22);
 512             rinv3            = _mm_mul_ps(rinvsq22,rinv22);
 513             pmecorrF         = avx128fma_pmecorrF_f(zeta2);
 514             felec            = _mm_macc_ps(pmecorrF,beta3,rinv3);
 515             felec            = _mm_mul_ps(qq22,felec);
 516             pmecorrV         = avx128fma_pmecorrV_f(zeta2);
 517             velec            = _mm_nmacc_ps(pmecorrV,beta,rinv22);
 518             velec            = _mm_mul_ps(qq22,velec);
 519
 520             /* Update potential sum for this i atom from the interaction with this j atom. */
 521             velecsum         = _mm_add_ps(velecsum,velec);
 522
 523             fscal            = felec;
 524
 525              /* Update vectorial force */
 526             fix2             = _mm_macc_ps(dx22,fscal,fix2);
 527             fiy2             = _mm_macc_ps(dy22,fscal,fiy2);
 528             fiz2             = _mm_macc_ps(dz22,fscal,fiz2);
 529
 530             fjx2             = _mm_macc_ps(dx22,fscal,fjx2);
 531             fjy2             = _mm_macc_ps(dy22,fscal,fjy2);
 532             fjz2             = _mm_macc_ps(dz22,fscal,fjz2);
 533
 534             /**************************
 535              * CALCULATE INTERACTIONS *
 536              **************************/
 537
 538             r23              = _mm_mul_ps(rsq23,rinv23);
 539
 540             /* EWALD ELECTROSTATICS */
 541
 542             /* Analytical PME correction */
 543             zeta2            = _mm_mul_ps(beta2,rsq23);
 544             rinv3            = _mm_mul_ps(rinvsq23,rinv23);
 545             pmecorrF         = avx128fma_pmecorrF_f(zeta2);
 546             felec            = _mm_macc_ps(pmecorrF,beta3,rinv3);
 547             felec            = _mm_mul_ps(qq23,felec);
 548             pmecorrV         = avx128fma_pmecorrV_f(zeta2);
 549             velec            = _mm_nmacc_ps(pmecorrV,beta,rinv23);
 550             velec            = _mm_mul_ps(qq23,velec);
 551
 552             /* Update potential sum for this i atom from the interaction with this j atom. */
 553             velecsum         = _mm_add_ps(velecsum,velec);
 554
 555             fscal            = felec;
 556
 557              /* Update vectorial force */
 558             fix2             = _mm_macc_ps(dx23,fscal,fix2);
 559             fiy2             = _mm_macc_ps(dy23,fscal,fiy2);
 560             fiz2             = _mm_macc_ps(dz23,fscal,fiz2);
 561
 562             fjx3             = _mm_macc_ps(dx23,fscal,fjx3);
 563             fjy3             = _mm_macc_ps(dy23,fscal,fjy3);
 564             fjz3             = _mm_macc_ps(dz23,fscal,fjz3);
 565
 566             /**************************
 567              * CALCULATE INTERACTIONS *
 568              **************************/
 569
 570             r31              = _mm_mul_ps(rsq31,rinv31);
 571
 572             /* EWALD ELECTROSTATICS */
 573
 574             /* Analytical PME correction */
 575             zeta2            = _mm_mul_ps(beta2,rsq31);
 576             rinv3            = _mm_mul_ps(rinvsq31,rinv31);
 577             pmecorrF         = avx128fma_pmecorrF_f(zeta2);
 578             felec            = _mm_macc_ps(pmecorrF,beta3,rinv3);
 579             felec            = _mm_mul_ps(qq31,felec);
 580             pmecorrV         = avx128fma_pmecorrV_f(zeta2);
 581             velec            = _mm_nmacc_ps(pmecorrV,beta,rinv31);
 582             velec            = _mm_mul_ps(qq31,velec);
 583
 584             /* Update potential sum for this i atom from the interaction with this j atom. */
 585             velecsum         = _mm_add_ps(velecsum,velec);
 586
 587             fscal            = felec;
 588
 589              /* Update vectorial force */
 590             fix3             = _mm_macc_ps(dx31,fscal,fix3);
 591             fiy3             = _mm_macc_ps(dy31,fscal,fiy3);
 592             fiz3             = _mm_macc_ps(dz31,fscal,fiz3);
 593
 594             fjx1             = _mm_macc_ps(dx31,fscal,fjx1);
 595             fjy1             = _mm_macc_ps(dy31,fscal,fjy1);
 596             fjz1             = _mm_macc_ps(dz31,fscal,fjz1);
 597
 598             /**************************
 599              * CALCULATE INTERACTIONS *
 600              **************************/
 601
 602             r32              = _mm_mul_ps(rsq32,rinv32);
 603
 604             /* EWALD ELECTROSTATICS */
 605
 606             /* Analytical PME correction */
 607             zeta2            = _mm_mul_ps(beta2,rsq32);
 608             rinv3            = _mm_mul_ps(rinvsq32,rinv32);
 609             pmecorrF         = avx128fma_pmecorrF_f(zeta2);
 610             felec            = _mm_macc_ps(pmecorrF,beta3,rinv3);
 611             felec            = _mm_mul_ps(qq32,felec);
 612             pmecorrV         = avx128fma_pmecorrV_f(zeta2);
 613             velec            = _mm_nmacc_ps(pmecorrV,beta,rinv32);
 614             velec            = _mm_mul_ps(qq32,velec);
 615
 616             /* Update potential sum for this i atom from the interaction with this j atom. */
 617             velecsum         = _mm_add_ps(velecsum,velec);
 618
 619             fscal            = felec;
 620
 621              /* Update vectorial force */
 622             fix3             = _mm_macc_ps(dx32,fscal,fix3);
 623             fiy3             = _mm_macc_ps(dy32,fscal,fiy3);
 624             fiz3             = _mm_macc_ps(dz32,fscal,fiz3);
 625
 626             fjx2             = _mm_macc_ps(dx32,fscal,fjx2);
 627             fjy2             = _mm_macc_ps(dy32,fscal,fjy2);
 628             fjz2             = _mm_macc_ps(dz32,fscal,fjz2);
 629
 630             /**************************
 631              * CALCULATE INTERACTIONS *
 632              **************************/
 633
 634             r33              = _mm_mul_ps(rsq33,rinv33);
 635
 636             /* EWALD ELECTROSTATICS */
 637
 638             /* Analytical PME correction */
 639             zeta2            = _mm_mul_ps(beta2,rsq33);
 640             rinv3            = _mm_mul_ps(rinvsq33,rinv33);
 641             pmecorrF         = avx128fma_pmecorrF_f(zeta2);
 642             felec            = _mm_macc_ps(pmecorrF,beta3,rinv3);
 643             felec            = _mm_mul_ps(qq33,felec);
 644             pmecorrV         = avx128fma_pmecorrV_f(zeta2);
 645             velec            = _mm_nmacc_ps(pmecorrV,beta,rinv33);
 646             velec            = _mm_mul_ps(qq33,velec);
 647
 648             /* Update potential sum for this i atom from the interaction with this j atom. */
 649             velecsum         = _mm_add_ps(velecsum,velec);
 650
 651             fscal            = felec;
 652
 653              /* Update vectorial force */
 654             fix3             = _mm_macc_ps(dx33,fscal,fix3);
 655             fiy3             = _mm_macc_ps(dy33,fscal,fiy3);
 656             fiz3             = _mm_macc_ps(dz33,fscal,fiz3);
 657
 658             fjx3             = _mm_macc_ps(dx33,fscal,fjx3);
 659             fjy3             = _mm_macc_ps(dy33,fscal,fjy3);
 660             fjz3             = _mm_macc_ps(dz33,fscal,fjz3);
 661
 662             fjptrA             = f+j_coord_offsetA;
 663             fjptrB             = f+j_coord_offsetB;
 664             fjptrC             = f+j_coord_offsetC;
 665             fjptrD             = f+j_coord_offsetD;
 666
 667             gmx_mm_decrement_4rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,
 668                                                    fjx0,fjy0,fjz0,fjx1,fjy1,fjz1,
 669                                                    fjx2,fjy2,fjz2,fjx3,fjy3,fjz3);
 670
 671             /* Inner loop uses 314 flops */
 672         }
 673
 674         if(jidx<j_index_end)
 675         {
 676
 677             /* Get j neighbor index, and coordinate index */
 678             jnrlistA         = jjnr[jidx];
 679             jnrlistB         = jjnr[jidx+1];
 680             jnrlistC         = jjnr[jidx+2];
 681             jnrlistD         = jjnr[jidx+3];
 682             /* Sign of each element will be negative for non-real atoms.
 683              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 684              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 685              */
 686             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 687             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 688             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 689             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 690             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 691             j_coord_offsetA  = DIM*jnrA;
 692             j_coord_offsetB  = DIM*jnrB;
 693             j_coord_offsetC  = DIM*jnrC;
 694             j_coord_offsetD  = DIM*jnrD;
 695
 696             /* load j atom coordinates */
 697             gmx_mm_load_4rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 698                                               x+j_coord_offsetC,x+j_coord_offsetD,
 699                                               &jx0,&jy0,&jz0,&jx1,&jy1,&jz1,&jx2,
 700                                               &jy2,&jz2,&jx3,&jy3,&jz3);
 701
 702             /* Calculate displacement vector */
 703             dx00             = _mm_sub_ps(ix0,jx0);
 704             dy00             = _mm_sub_ps(iy0,jy0);
 705             dz00             = _mm_sub_ps(iz0,jz0);
 706             dx11             = _mm_sub_ps(ix1,jx1);
 707             dy11             = _mm_sub_ps(iy1,jy1);
 708             dz11             = _mm_sub_ps(iz1,jz1);
 709             dx12             = _mm_sub_ps(ix1,jx2);
 710             dy12             = _mm_sub_ps(iy1,jy2);
 711             dz12             = _mm_sub_ps(iz1,jz2);
 712             dx13             = _mm_sub_ps(ix1,jx3);
 713             dy13             = _mm_sub_ps(iy1,jy3);
 714             dz13             = _mm_sub_ps(iz1,jz3);
 715             dx21             = _mm_sub_ps(ix2,jx1);
 716             dy21             = _mm_sub_ps(iy2,jy1);
 717             dz21             = _mm_sub_ps(iz2,jz1);
 718             dx22             = _mm_sub_ps(ix2,jx2);
 719             dy22             = _mm_sub_ps(iy2,jy2);
 720             dz22             = _mm_sub_ps(iz2,jz2);
 721             dx23             = _mm_sub_ps(ix2,jx3);
 722             dy23             = _mm_sub_ps(iy2,jy3);
 723             dz23             = _mm_sub_ps(iz2,jz3);
 724             dx31             = _mm_sub_ps(ix3,jx1);
 725             dy31             = _mm_sub_ps(iy3,jy1);
 726             dz31             = _mm_sub_ps(iz3,jz1);
 727             dx32             = _mm_sub_ps(ix3,jx2);
 728             dy32             = _mm_sub_ps(iy3,jy2);
 729             dz32             = _mm_sub_ps(iz3,jz2);
 730             dx33             = _mm_sub_ps(ix3,jx3);
 731             dy33             = _mm_sub_ps(iy3,jy3);
 732             dz33             = _mm_sub_ps(iz3,jz3);
 733
 734             /* Calculate squared distance and things based on it */
 735             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 736             rsq11            = gmx_mm_calc_rsq_ps(dx11,dy11,dz11);
 737             rsq12            = gmx_mm_calc_rsq_ps(dx12,dy12,dz12);
 738             rsq13            = gmx_mm_calc_rsq_ps(dx13,dy13,dz13);
 739             rsq21            = gmx_mm_calc_rsq_ps(dx21,dy21,dz21);
 740             rsq22            = gmx_mm_calc_rsq_ps(dx22,dy22,dz22);
 741             rsq23            = gmx_mm_calc_rsq_ps(dx23,dy23,dz23);
 742             rsq31            = gmx_mm_calc_rsq_ps(dx31,dy31,dz31);
 743             rsq32            = gmx_mm_calc_rsq_ps(dx32,dy32,dz32);
 744             rsq33            = gmx_mm_calc_rsq_ps(dx33,dy33,dz33);
 745
 746             rinv00           = avx128fma_invsqrt_f(rsq00);
 747             rinv11           = avx128fma_invsqrt_f(rsq11);
 748             rinv12           = avx128fma_invsqrt_f(rsq12);
 749             rinv13           = avx128fma_invsqrt_f(rsq13);
 750             rinv21           = avx128fma_invsqrt_f(rsq21);
 751             rinv22           = avx128fma_invsqrt_f(rsq22);
 752             rinv23           = avx128fma_invsqrt_f(rsq23);
 753             rinv31           = avx128fma_invsqrt_f(rsq31);
 754             rinv32           = avx128fma_invsqrt_f(rsq32);
 755             rinv33           = avx128fma_invsqrt_f(rsq33);
 756
 757             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 758             rinvsq11         = _mm_mul_ps(rinv11,rinv11);
 759             rinvsq12         = _mm_mul_ps(rinv12,rinv12);
 760             rinvsq13         = _mm_mul_ps(rinv13,rinv13);
 761             rinvsq21         = _mm_mul_ps(rinv21,rinv21);
 762             rinvsq22         = _mm_mul_ps(rinv22,rinv22);
 763             rinvsq23         = _mm_mul_ps(rinv23,rinv23);
 764             rinvsq31         = _mm_mul_ps(rinv31,rinv31);
 765             rinvsq32         = _mm_mul_ps(rinv32,rinv32);
 766             rinvsq33         = _mm_mul_ps(rinv33,rinv33);
 767
 768             fjx0             = _mm_setzero_ps();
 769             fjy0             = _mm_setzero_ps();
 770             fjz0             = _mm_setzero_ps();
 771             fjx1             = _mm_setzero_ps();
 772             fjy1             = _mm_setzero_ps();
 773             fjz1             = _mm_setzero_ps();
 774             fjx2             = _mm_setzero_ps();
 775             fjy2             = _mm_setzero_ps();
 776             fjz2             = _mm_setzero_ps();
 777             fjx3             = _mm_setzero_ps();
 778             fjy3             = _mm_setzero_ps();
 779             fjz3             = _mm_setzero_ps();
 780
 781             /**************************
 782              * CALCULATE INTERACTIONS *
 783              **************************/
 784
 785             r00              = _mm_mul_ps(rsq00,rinv00);
 786             r00              = _mm_andnot_ps(dummy_mask,r00);
 787
 788             /* Analytical LJ-PME */
 789             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 790             ewcljrsq         = _mm_mul_ps(ewclj2,rsq00);
 791             ewclj6           = _mm_mul_ps(ewclj2,_mm_mul_ps(ewclj2,ewclj2));
 792             exponent         = avx128fma_exp_f(ewcljrsq);
 793             /* poly = exp(-(beta*r)^2) * (1 + (beta*r)^2 + (beta*r)^4 /2) */
 794             poly             = _mm_mul_ps(exponent,_mm_macc_ps(_mm_mul_ps(ewcljrsq,ewcljrsq),one_half,_mm_sub_ps(one,ewcljrsq)));
 795             /* vvdw6 = [C6 - C6grid * (1-poly)]/r6 */
 796             vvdw6            = _mm_mul_ps(_mm_macc_ps(-c6grid_00,_mm_sub_ps(one,poly),c6_00),rinvsix);
 797             vvdw12           = _mm_mul_ps(c12_00,_mm_mul_ps(rinvsix,rinvsix));
 798             vvdw             = _mm_msub_ps(vvdw12,one_twelfth,_mm_mul_ps(vvdw6,one_sixth));
 799             /* fvdw = vvdw12/r - (vvdw6/r + (C6grid * exponent * beta^6)/r) */
 800             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);
 801
 802             /* Update potential sum for this i atom from the interaction with this j atom. */
 803             vvdw             = _mm_andnot_ps(dummy_mask,vvdw);
 804             vvdwsum          = _mm_add_ps(vvdwsum,vvdw);
 805
 806             fscal            = fvdw;
 807
 808             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 809
 810              /* Update vectorial force */
 811             fix0             = _mm_macc_ps(dx00,fscal,fix0);
 812             fiy0             = _mm_macc_ps(dy00,fscal,fiy0);
 813             fiz0             = _mm_macc_ps(dz00,fscal,fiz0);
 814
 815             fjx0             = _mm_macc_ps(dx00,fscal,fjx0);
 816             fjy0             = _mm_macc_ps(dy00,fscal,fjy0);
 817             fjz0             = _mm_macc_ps(dz00,fscal,fjz0);
 818
 819             /**************************
 820              * CALCULATE INTERACTIONS *
 821              **************************/
 822
 823             r11              = _mm_mul_ps(rsq11,rinv11);
 824             r11              = _mm_andnot_ps(dummy_mask,r11);
 825
 826             /* EWALD ELECTROSTATICS */
 827
 828             /* Analytical PME correction */
 829             zeta2            = _mm_mul_ps(beta2,rsq11);
 830             rinv3            = _mm_mul_ps(rinvsq11,rinv11);
 831             pmecorrF         = avx128fma_pmecorrF_f(zeta2);
 832             felec            = _mm_macc_ps(pmecorrF,beta3,rinv3);
 833             felec            = _mm_mul_ps(qq11,felec);
 834             pmecorrV         = avx128fma_pmecorrV_f(zeta2);
 835             velec            = _mm_nmacc_ps(pmecorrV,beta,rinv11);
 836             velec            = _mm_mul_ps(qq11,velec);
 837
 838             /* Update potential sum for this i atom from the interaction with this j atom. */
 839             velec            = _mm_andnot_ps(dummy_mask,velec);
 840             velecsum         = _mm_add_ps(velecsum,velec);
 841
 842             fscal            = felec;
 843
 844             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 845
 846              /* Update vectorial force */
 847             fix1             = _mm_macc_ps(dx11,fscal,fix1);
 848             fiy1             = _mm_macc_ps(dy11,fscal,fiy1);
 849             fiz1             = _mm_macc_ps(dz11,fscal,fiz1);
 850
 851             fjx1             = _mm_macc_ps(dx11,fscal,fjx1);
 852             fjy1             = _mm_macc_ps(dy11,fscal,fjy1);
 853             fjz1             = _mm_macc_ps(dz11,fscal,fjz1);
 854
 855             /**************************
 856              * CALCULATE INTERACTIONS *
 857              **************************/
 858
 859             r12              = _mm_mul_ps(rsq12,rinv12);
 860             r12              = _mm_andnot_ps(dummy_mask,r12);
 861
 862             /* EWALD ELECTROSTATICS */
 863
 864             /* Analytical PME correction */
 865             zeta2            = _mm_mul_ps(beta2,rsq12);
 866             rinv3            = _mm_mul_ps(rinvsq12,rinv12);
 867             pmecorrF         = avx128fma_pmecorrF_f(zeta2);
 868             felec            = _mm_macc_ps(pmecorrF,beta3,rinv3);
 869             felec            = _mm_mul_ps(qq12,felec);
 870             pmecorrV         = avx128fma_pmecorrV_f(zeta2);
 871             velec            = _mm_nmacc_ps(pmecorrV,beta,rinv12);
 872             velec            = _mm_mul_ps(qq12,velec);
 873
 874             /* Update potential sum for this i atom from the interaction with this j atom. */
 875             velec            = _mm_andnot_ps(dummy_mask,velec);
 876             velecsum         = _mm_add_ps(velecsum,velec);
 877
 878             fscal            = felec;
 879
 880             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 881
 882              /* Update vectorial force */
 883             fix1             = _mm_macc_ps(dx12,fscal,fix1);
 884             fiy1             = _mm_macc_ps(dy12,fscal,fiy1);
 885             fiz1             = _mm_macc_ps(dz12,fscal,fiz1);
 886
 887             fjx2             = _mm_macc_ps(dx12,fscal,fjx2);
 888             fjy2             = _mm_macc_ps(dy12,fscal,fjy2);
 889             fjz2             = _mm_macc_ps(dz12,fscal,fjz2);
 890
 891             /**************************
 892              * CALCULATE INTERACTIONS *
 893              **************************/
 894
 895             r13              = _mm_mul_ps(rsq13,rinv13);
 896             r13              = _mm_andnot_ps(dummy_mask,r13);
 897
 898             /* EWALD ELECTROSTATICS */
 899
 900             /* Analytical PME correction */
 901             zeta2            = _mm_mul_ps(beta2,rsq13);
 902             rinv3            = _mm_mul_ps(rinvsq13,rinv13);
 903             pmecorrF         = avx128fma_pmecorrF_f(zeta2);
 904             felec            = _mm_macc_ps(pmecorrF,beta3,rinv3);
 905             felec            = _mm_mul_ps(qq13,felec);
 906             pmecorrV         = avx128fma_pmecorrV_f(zeta2);
 907             velec            = _mm_nmacc_ps(pmecorrV,beta,rinv13);
 908             velec            = _mm_mul_ps(qq13,velec);
 909
 910             /* Update potential sum for this i atom from the interaction with this j atom. */
 911             velec            = _mm_andnot_ps(dummy_mask,velec);
 912             velecsum         = _mm_add_ps(velecsum,velec);
 913
 914             fscal            = felec;
 915
 916             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 917
 918              /* Update vectorial force */
 919             fix1             = _mm_macc_ps(dx13,fscal,fix1);
 920             fiy1             = _mm_macc_ps(dy13,fscal,fiy1);
 921             fiz1             = _mm_macc_ps(dz13,fscal,fiz1);
 922
 923             fjx3             = _mm_macc_ps(dx13,fscal,fjx3);
 924             fjy3             = _mm_macc_ps(dy13,fscal,fjy3);
 925             fjz3             = _mm_macc_ps(dz13,fscal,fjz3);
 926
 927             /**************************
 928              * CALCULATE INTERACTIONS *
 929              **************************/
 930
 931             r21              = _mm_mul_ps(rsq21,rinv21);
 932             r21              = _mm_andnot_ps(dummy_mask,r21);
 933
 934             /* EWALD ELECTROSTATICS */
 935
 936             /* Analytical PME correction */
 937             zeta2            = _mm_mul_ps(beta2,rsq21);
 938             rinv3            = _mm_mul_ps(rinvsq21,rinv21);
 939             pmecorrF         = avx128fma_pmecorrF_f(zeta2);
 940             felec            = _mm_macc_ps(pmecorrF,beta3,rinv3);
 941             felec            = _mm_mul_ps(qq21,felec);
 942             pmecorrV         = avx128fma_pmecorrV_f(zeta2);
 943             velec            = _mm_nmacc_ps(pmecorrV,beta,rinv21);
 944             velec            = _mm_mul_ps(qq21,velec);
 945
 946             /* Update potential sum for this i atom from the interaction with this j atom. */
 947             velec            = _mm_andnot_ps(dummy_mask,velec);
 948             velecsum         = _mm_add_ps(velecsum,velec);
 949
 950             fscal            = felec;
 951
 952             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 953
 954              /* Update vectorial force */
 955             fix2             = _mm_macc_ps(dx21,fscal,fix2);
 956             fiy2             = _mm_macc_ps(dy21,fscal,fiy2);
 957             fiz2             = _mm_macc_ps(dz21,fscal,fiz2);
 958
 959             fjx1             = _mm_macc_ps(dx21,fscal,fjx1);
 960             fjy1             = _mm_macc_ps(dy21,fscal,fjy1);
 961             fjz1             = _mm_macc_ps(dz21,fscal,fjz1);
 962
 963             /**************************
 964              * CALCULATE INTERACTIONS *
 965              **************************/
 966
 967             r22              = _mm_mul_ps(rsq22,rinv22);
 968             r22              = _mm_andnot_ps(dummy_mask,r22);
 969
 970             /* EWALD ELECTROSTATICS */
 971
 972             /* Analytical PME correction */
 973             zeta2            = _mm_mul_ps(beta2,rsq22);
 974             rinv3            = _mm_mul_ps(rinvsq22,rinv22);
 975             pmecorrF         = avx128fma_pmecorrF_f(zeta2);
 976             felec            = _mm_macc_ps(pmecorrF,beta3,rinv3);
 977             felec            = _mm_mul_ps(qq22,felec);
 978             pmecorrV         = avx128fma_pmecorrV_f(zeta2);
 979             velec            = _mm_nmacc_ps(pmecorrV,beta,rinv22);
 980             velec            = _mm_mul_ps(qq22,velec);
 981
 982             /* Update potential sum for this i atom from the interaction with this j atom. */
 983             velec            = _mm_andnot_ps(dummy_mask,velec);
 984             velecsum         = _mm_add_ps(velecsum,velec);
 985
 986             fscal            = felec;
 987
 988             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 989
 990              /* Update vectorial force */
 991             fix2             = _mm_macc_ps(dx22,fscal,fix2);
 992             fiy2             = _mm_macc_ps(dy22,fscal,fiy2);
 993             fiz2             = _mm_macc_ps(dz22,fscal,fiz2);
 994
 995             fjx2             = _mm_macc_ps(dx22,fscal,fjx2);
 996             fjy2             = _mm_macc_ps(dy22,fscal,fjy2);
 997             fjz2             = _mm_macc_ps(dz22,fscal,fjz2);
 998
 999             /**************************
1000              * CALCULATE INTERACTIONS *
1001              **************************/
1002
1003             r23              = _mm_mul_ps(rsq23,rinv23);
1004             r23              = _mm_andnot_ps(dummy_mask,r23);
1005
1006             /* EWALD ELECTROSTATICS */
1007
1008             /* Analytical PME correction */
1009             zeta2            = _mm_mul_ps(beta2,rsq23);
1010             rinv3            = _mm_mul_ps(rinvsq23,rinv23);
1011             pmecorrF         = avx128fma_pmecorrF_f(zeta2);
1012             felec            = _mm_macc_ps(pmecorrF,beta3,rinv3);
1013             felec            = _mm_mul_ps(qq23,felec);
1014             pmecorrV         = avx128fma_pmecorrV_f(zeta2);
1015             velec            = _mm_nmacc_ps(pmecorrV,beta,rinv23);
1016             velec            = _mm_mul_ps(qq23,velec);
1017
1018             /* Update potential sum for this i atom from the interaction with this j atom. */
1019             velec            = _mm_andnot_ps(dummy_mask,velec);
1020             velecsum         = _mm_add_ps(velecsum,velec);
1021
1022             fscal            = felec;
1023
1024             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1025
1026              /* Update vectorial force */
1027             fix2             = _mm_macc_ps(dx23,fscal,fix2);
1028             fiy2             = _mm_macc_ps(dy23,fscal,fiy2);
1029             fiz2             = _mm_macc_ps(dz23,fscal,fiz2);
1030
1031             fjx3             = _mm_macc_ps(dx23,fscal,fjx3);
1032             fjy3             = _mm_macc_ps(dy23,fscal,fjy3);
1033             fjz3             = _mm_macc_ps(dz23,fscal,fjz3);
1034
1035             /**************************
1036              * CALCULATE INTERACTIONS *
1037              **************************/
1038
1039             r31              = _mm_mul_ps(rsq31,rinv31);
1040             r31              = _mm_andnot_ps(dummy_mask,r31);
1041
1042             /* EWALD ELECTROSTATICS */
1043
1044             /* Analytical PME correction */
1045             zeta2            = _mm_mul_ps(beta2,rsq31);
1046             rinv3            = _mm_mul_ps(rinvsq31,rinv31);
1047             pmecorrF         = avx128fma_pmecorrF_f(zeta2);
1048             felec            = _mm_macc_ps(pmecorrF,beta3,rinv3);
1049             felec            = _mm_mul_ps(qq31,felec);
1050             pmecorrV         = avx128fma_pmecorrV_f(zeta2);
1051             velec            = _mm_nmacc_ps(pmecorrV,beta,rinv31);
1052             velec            = _mm_mul_ps(qq31,velec);
1053
1054             /* Update potential sum for this i atom from the interaction with this j atom. */
1055             velec            = _mm_andnot_ps(dummy_mask,velec);
1056             velecsum         = _mm_add_ps(velecsum,velec);
1057
1058             fscal            = felec;
1059
1060             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1061
1062              /* Update vectorial force */
1063             fix3             = _mm_macc_ps(dx31,fscal,fix3);
1064             fiy3             = _mm_macc_ps(dy31,fscal,fiy3);
1065             fiz3             = _mm_macc_ps(dz31,fscal,fiz3);
1066
1067             fjx1             = _mm_macc_ps(dx31,fscal,fjx1);
1068             fjy1             = _mm_macc_ps(dy31,fscal,fjy1);
1069             fjz1             = _mm_macc_ps(dz31,fscal,fjz1);
1070
1071             /**************************
1072              * CALCULATE INTERACTIONS *
1073              **************************/
1074
1075             r32              = _mm_mul_ps(rsq32,rinv32);
1076             r32              = _mm_andnot_ps(dummy_mask,r32);
1077
1078             /* EWALD ELECTROSTATICS */
1079
1080             /* Analytical PME correction */
1081             zeta2            = _mm_mul_ps(beta2,rsq32);
1082             rinv3            = _mm_mul_ps(rinvsq32,rinv32);
1083             pmecorrF         = avx128fma_pmecorrF_f(zeta2);
1084             felec            = _mm_macc_ps(pmecorrF,beta3,rinv3);
1085             felec            = _mm_mul_ps(qq32,felec);
1086             pmecorrV         = avx128fma_pmecorrV_f(zeta2);
1087             velec            = _mm_nmacc_ps(pmecorrV,beta,rinv32);
1088             velec            = _mm_mul_ps(qq32,velec);
1089
1090             /* Update potential sum for this i atom from the interaction with this j atom. */
1091             velec            = _mm_andnot_ps(dummy_mask,velec);
1092             velecsum         = _mm_add_ps(velecsum,velec);
1093
1094             fscal            = felec;
1095
1096             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1097
1098              /* Update vectorial force */
1099             fix3             = _mm_macc_ps(dx32,fscal,fix3);
1100             fiy3             = _mm_macc_ps(dy32,fscal,fiy3);
1101             fiz3             = _mm_macc_ps(dz32,fscal,fiz3);
1102
1103             fjx2             = _mm_macc_ps(dx32,fscal,fjx2);
1104             fjy2             = _mm_macc_ps(dy32,fscal,fjy2);
1105             fjz2             = _mm_macc_ps(dz32,fscal,fjz2);
1106
1107             /**************************
1108              * CALCULATE INTERACTIONS *
1109              **************************/
1110
1111             r33              = _mm_mul_ps(rsq33,rinv33);
1112             r33              = _mm_andnot_ps(dummy_mask,r33);
1113
1114             /* EWALD ELECTROSTATICS */
1115
1116             /* Analytical PME correction */
1117             zeta2            = _mm_mul_ps(beta2,rsq33);
1118             rinv3            = _mm_mul_ps(rinvsq33,rinv33);
1119             pmecorrF         = avx128fma_pmecorrF_f(zeta2);
1120             felec            = _mm_macc_ps(pmecorrF,beta3,rinv3);
1121             felec            = _mm_mul_ps(qq33,felec);
1122             pmecorrV         = avx128fma_pmecorrV_f(zeta2);
1123             velec            = _mm_nmacc_ps(pmecorrV,beta,rinv33);
1124             velec            = _mm_mul_ps(qq33,velec);
1125
1126             /* Update potential sum for this i atom from the interaction with this j atom. */
1127             velec            = _mm_andnot_ps(dummy_mask,velec);
1128             velecsum         = _mm_add_ps(velecsum,velec);
1129
1130             fscal            = felec;
1131
1132             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1133
1134              /* Update vectorial force */
1135             fix3             = _mm_macc_ps(dx33,fscal,fix3);
1136             fiy3             = _mm_macc_ps(dy33,fscal,fiy3);
1137             fiz3             = _mm_macc_ps(dz33,fscal,fiz3);
1138
1139             fjx3             = _mm_macc_ps(dx33,fscal,fjx3);
1140             fjy3             = _mm_macc_ps(dy33,fscal,fjy3);
1141             fjz3             = _mm_macc_ps(dz33,fscal,fjz3);
1142
1143             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
1144             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
1145             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
1146             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
1147
1148             gmx_mm_decrement_4rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,
1149                                                    fjx0,fjy0,fjz0,fjx1,fjy1,fjz1,
1150                                                    fjx2,fjy2,fjz2,fjx3,fjy3,fjz3);
1151
1152             /* Inner loop uses 324 flops */
1153         }
1154
1155         /* End of innermost loop */
1156
1157         gmx_mm_update_iforce_4atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
1158                                               f+i_coord_offset,fshift+i_shift_offset);
1159
1160         ggid                        = gid[iidx];
1161         /* Update potential energies */
1162         gmx_mm_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid);
1163         gmx_mm_update_1pot_ps(vvdwsum,kernel_data->energygrp_vdw+ggid);
1164
1165         /* Increment number of inner iterations */
1166         inneriter                  += j_index_end - j_index_start;
1167
1168         /* Outer loop uses 26 flops */
1169     }
1170
1171     /* Increment number of outer iterations */
1172     outeriter        += nri;
1173
1174     /* Update outer/inner flops */
1175
1176     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W4W4_VF,outeriter*26 + inneriter*324);
1177 }
1178 /*
1179  * Gromacs nonbonded kernel:   nb_kernel_ElecEw_VdwLJEw_GeomW4W4_F_avx_128_fma_single
1180  * Electrostatics interaction: Ewald
1181  * VdW interaction:            LJEwald
1182  * Geometry:                   Water4-Water4
1183  * Calculate force/pot:        Force
1184  */
1185 void
1186 nb_kernel_ElecEw_VdwLJEw_GeomW4W4_F_avx_128_fma_single
1187                     (t_nblist                    * gmx_restrict       nlist,
1188                      rvec                        * gmx_restrict          xx,
1189                      rvec                        * gmx_restrict          ff,
1190                      struct t_forcerec           * gmx_restrict          fr,
1191                      t_mdatoms                   * gmx_restrict     mdatoms,
1192                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
1193                      t_nrnb                      * gmx_restrict        nrnb)
1194 {
1195     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
1196      * just 0 for non-waters.
1197      * Suffixes A,B,C,D refer to j loop unrolling done with AVX_128, e.g. for the four different
1198      * jnr indices corresponding to data put in the four positions in the SIMD register.
1199      */
1200     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
1201     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
1202     int              jnrA,jnrB,jnrC,jnrD;
1203     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
1204     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
1205     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
1206     real             rcutoff_scalar;
1207     real             *shiftvec,*fshift,*x,*f;
1208     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
1209     real             scratch[4*DIM];
1210     __m128           fscal,rcutoff,rcutoff2,jidxall;
1211     int              vdwioffset0;
1212     __m128           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
1213     int              vdwioffset1;
1214     __m128           ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
1215     int              vdwioffset2;
1216     __m128           ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
1217     int              vdwioffset3;
1218     __m128           ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
1219     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
1220     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
1221     int              vdwjidx1A,vdwjidx1B,vdwjidx1C,vdwjidx1D;
1222     __m128           jx1,jy1,jz1,fjx1,fjy1,fjz1,jq1,isaj1;
1223     int              vdwjidx2A,vdwjidx2B,vdwjidx2C,vdwjidx2D;
1224     __m128           jx2,jy2,jz2,fjx2,fjy2,fjz2,jq2,isaj2;
1225     int              vdwjidx3A,vdwjidx3B,vdwjidx3C,vdwjidx3D;
1226     __m128           jx3,jy3,jz3,fjx3,fjy3,fjz3,jq3,isaj3;
1227     __m128           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
1228     __m128           dx11,dy11,dz11,rsq11,rinv11,rinvsq11,r11,qq11,c6_11,c12_11;
1229     __m128           dx12,dy12,dz12,rsq12,rinv12,rinvsq12,r12,qq12,c6_12,c12_12;
1230     __m128           dx13,dy13,dz13,rsq13,rinv13,rinvsq13,r13,qq13,c6_13,c12_13;
1231     __m128           dx21,dy21,dz21,rsq21,rinv21,rinvsq21,r21,qq21,c6_21,c12_21;
1232     __m128           dx22,dy22,dz22,rsq22,rinv22,rinvsq22,r22,qq22,c6_22,c12_22;
1233     __m128           dx23,dy23,dz23,rsq23,rinv23,rinvsq23,r23,qq23,c6_23,c12_23;
1234     __m128           dx31,dy31,dz31,rsq31,rinv31,rinvsq31,r31,qq31,c6_31,c12_31;
1235     __m128           dx32,dy32,dz32,rsq32,rinv32,rinvsq32,r32,qq32,c6_32,c12_32;
1236     __m128           dx33,dy33,dz33,rsq33,rinv33,rinvsq33,r33,qq33,c6_33,c12_33;
1237     __m128           velec,felec,velecsum,facel,crf,krf,krf2;
1238     real             *charge;
1239     int              nvdwtype;
1240     __m128           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
1241     int              *vdwtype;
1242     real             *vdwparam;
1243     __m128           one_sixth   = _mm_set1_ps(1.0/6.0);
1244     __m128           one_twelfth = _mm_set1_ps(1.0/12.0);
1245     __m128           c6grid_00;
1246     __m128           c6grid_11;
1247     __m128           c6grid_12;
1248     __m128           c6grid_13;
1249     __m128           c6grid_21;
1250     __m128           c6grid_22;
1251     __m128           c6grid_23;
1252     __m128           c6grid_31;
1253     __m128           c6grid_32;
1254     __m128           c6grid_33;
1255     real             *vdwgridparam;
1256     __m128           ewclj,ewclj2,ewclj6,ewcljrsq,poly,exponent,f6A,f6B,sh_lj_ewald;
1257     __m128           one_half = _mm_set1_ps(0.5);
1258     __m128           minus_one = _mm_set1_ps(-1.0);
1259     __m128i          ewitab;
1260     __m128           ewtabscale,eweps,twoeweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
1261     __m128           beta,beta2,beta3,zeta2,pmecorrF,pmecorrV,rinv3;
1262     real             *ewtab;
1263     __m128           dummy_mask,cutoff_mask;
1264     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
1265     __m128           one     = _mm_set1_ps(1.0);
1266     __m128           two     = _mm_set1_ps(2.0);
1267     x                = xx[0];
1268     f                = ff[0];
1269
1270     nri              = nlist->nri;
1271     iinr             = nlist->iinr;
1272     jindex           = nlist->jindex;
1273     jjnr             = nlist->jjnr;
1274     shiftidx         = nlist->shift;
1275     gid              = nlist->gid;
1276     shiftvec         = fr->shift_vec[0];
1277     fshift           = fr->fshift[0];
1278     facel            = _mm_set1_ps(fr->ic->epsfac);
1279     charge           = mdatoms->chargeA;
1280     nvdwtype         = fr->ntype;
1281     vdwparam         = fr->nbfp;
1282     vdwtype          = mdatoms->typeA;
1283     vdwgridparam     = fr->ljpme_c6grid;
1284     sh_lj_ewald      = _mm_set1_ps(fr->ic->sh_lj_ewald);
1285     ewclj            = _mm_set1_ps(fr->ic->ewaldcoeff_lj);
1286     ewclj2           = _mm_mul_ps(minus_one,_mm_mul_ps(ewclj,ewclj));
1287
1288     sh_ewald         = _mm_set1_ps(fr->ic->sh_ewald);
1289     beta             = _mm_set1_ps(fr->ic->ewaldcoeff_q);
1290     beta2            = _mm_mul_ps(beta,beta);
1291     beta3            = _mm_mul_ps(beta,beta2);
1292     ewtab            = fr->ic->tabq_coul_F;
1293     ewtabscale       = _mm_set1_ps(fr->ic->tabq_scale);
1294     ewtabhalfspace   = _mm_set1_ps(0.5/fr->ic->tabq_scale);
1295
1296     /* Setup water-specific parameters */
1297     inr              = nlist->iinr[0];
1298     iq1              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+1]));
1299     iq2              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+2]));
1300     iq3              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+3]));
1301     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
1302
1303     jq1              = _mm_set1_ps(charge[inr+1]);
1304     jq2              = _mm_set1_ps(charge[inr+2]);
1305     jq3              = _mm_set1_ps(charge[inr+3]);
1306     vdwjidx0A        = 2*vdwtype[inr+0];
1307     c6_00            = _mm_set1_ps(vdwparam[vdwioffset0+vdwjidx0A]);
1308     c12_00           = _mm_set1_ps(vdwparam[vdwioffset0+vdwjidx0A+1]);
1309     c6grid_00        = _mm_set1_ps(vdwgridparam[vdwioffset0+vdwjidx0A]);
1310     qq11             = _mm_mul_ps(iq1,jq1);
1311     qq12             = _mm_mul_ps(iq1,jq2);
1312     qq13             = _mm_mul_ps(iq1,jq3);
1313     qq21             = _mm_mul_ps(iq2,jq1);
1314     qq22             = _mm_mul_ps(iq2,jq2);
1315     qq23             = _mm_mul_ps(iq2,jq3);
1316     qq31             = _mm_mul_ps(iq3,jq1);
1317     qq32             = _mm_mul_ps(iq3,jq2);
1318     qq33             = _mm_mul_ps(iq3,jq3);
1319
1320     /* Avoid stupid compiler warnings */
1321     jnrA = jnrB = jnrC = jnrD = 0;
1322     j_coord_offsetA = 0;
1323     j_coord_offsetB = 0;
1324     j_coord_offsetC = 0;
1325     j_coord_offsetD = 0;
1326
1327     outeriter        = 0;
1328     inneriter        = 0;
1329
1330     for(iidx=0;iidx<4*DIM;iidx++)
1331     {
1332         scratch[iidx] = 0.0;
1333     }
1334
1335     /* Start outer loop over neighborlists */
1336     for(iidx=0; iidx<nri; iidx++)
1337     {
1338         /* Load shift vector for this list */
1339         i_shift_offset   = DIM*shiftidx[iidx];
1340
1341         /* Load limits for loop over neighbors */
1342         j_index_start    = jindex[iidx];
1343         j_index_end      = jindex[iidx+1];
1344
1345         /* Get outer coordinate index */
1346         inr              = iinr[iidx];
1347         i_coord_offset   = DIM*inr;
1348
1349         /* Load i particle coords and add shift vector */
1350         gmx_mm_load_shift_and_4rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
1351                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
1352
1353         fix0             = _mm_setzero_ps();
1354         fiy0             = _mm_setzero_ps();
1355         fiz0             = _mm_setzero_ps();
1356         fix1             = _mm_setzero_ps();
1357         fiy1             = _mm_setzero_ps();
1358         fiz1             = _mm_setzero_ps();
1359         fix2             = _mm_setzero_ps();
1360         fiy2             = _mm_setzero_ps();
1361         fiz2             = _mm_setzero_ps();
1362         fix3             = _mm_setzero_ps();
1363         fiy3             = _mm_setzero_ps();
1364         fiz3             = _mm_setzero_ps();
1365
1366         /* Start inner kernel loop */
1367         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
1368         {
1369
1370             /* Get j neighbor index, and coordinate index */
1371             jnrA             = jjnr[jidx];
1372             jnrB             = jjnr[jidx+1];
1373             jnrC             = jjnr[jidx+2];
1374             jnrD             = jjnr[jidx+3];
1375             j_coord_offsetA  = DIM*jnrA;
1376             j_coord_offsetB  = DIM*jnrB;
1377             j_coord_offsetC  = DIM*jnrC;
1378             j_coord_offsetD  = DIM*jnrD;
1379
1380             /* load j atom coordinates */
1381             gmx_mm_load_4rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
1382                                               x+j_coord_offsetC,x+j_coord_offsetD,
1383                                               &jx0,&jy0,&jz0,&jx1,&jy1,&jz1,&jx2,
1384                                               &jy2,&jz2,&jx3,&jy3,&jz3);
1385
1386             /* Calculate displacement vector */
1387             dx00             = _mm_sub_ps(ix0,jx0);
1388             dy00             = _mm_sub_ps(iy0,jy0);
1389             dz00             = _mm_sub_ps(iz0,jz0);
1390             dx11             = _mm_sub_ps(ix1,jx1);
1391             dy11             = _mm_sub_ps(iy1,jy1);
1392             dz11             = _mm_sub_ps(iz1,jz1);
1393             dx12             = _mm_sub_ps(ix1,jx2);
1394             dy12             = _mm_sub_ps(iy1,jy2);
1395             dz12             = _mm_sub_ps(iz1,jz2);
1396             dx13             = _mm_sub_ps(ix1,jx3);
1397             dy13             = _mm_sub_ps(iy1,jy3);
1398             dz13             = _mm_sub_ps(iz1,jz3);
1399             dx21             = _mm_sub_ps(ix2,jx1);
1400             dy21             = _mm_sub_ps(iy2,jy1);
1401             dz21             = _mm_sub_ps(iz2,jz1);
1402             dx22             = _mm_sub_ps(ix2,jx2);
1403             dy22             = _mm_sub_ps(iy2,jy2);
1404             dz22             = _mm_sub_ps(iz2,jz2);
1405             dx23             = _mm_sub_ps(ix2,jx3);
1406             dy23             = _mm_sub_ps(iy2,jy3);
1407             dz23             = _mm_sub_ps(iz2,jz3);
1408             dx31             = _mm_sub_ps(ix3,jx1);
1409             dy31             = _mm_sub_ps(iy3,jy1);
1410             dz31             = _mm_sub_ps(iz3,jz1);
1411             dx32             = _mm_sub_ps(ix3,jx2);
1412             dy32             = _mm_sub_ps(iy3,jy2);
1413             dz32             = _mm_sub_ps(iz3,jz2);
1414             dx33             = _mm_sub_ps(ix3,jx3);
1415             dy33             = _mm_sub_ps(iy3,jy3);
1416             dz33             = _mm_sub_ps(iz3,jz3);
1417
1418             /* Calculate squared distance and things based on it */
1419             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
1420             rsq11            = gmx_mm_calc_rsq_ps(dx11,dy11,dz11);
1421             rsq12            = gmx_mm_calc_rsq_ps(dx12,dy12,dz12);
1422             rsq13            = gmx_mm_calc_rsq_ps(dx13,dy13,dz13);
1423             rsq21            = gmx_mm_calc_rsq_ps(dx21,dy21,dz21);
1424             rsq22            = gmx_mm_calc_rsq_ps(dx22,dy22,dz22);
1425             rsq23            = gmx_mm_calc_rsq_ps(dx23,dy23,dz23);
1426             rsq31            = gmx_mm_calc_rsq_ps(dx31,dy31,dz31);
1427             rsq32            = gmx_mm_calc_rsq_ps(dx32,dy32,dz32);
1428             rsq33            = gmx_mm_calc_rsq_ps(dx33,dy33,dz33);
1429
1430             rinv00           = avx128fma_invsqrt_f(rsq00);
1431             rinv11           = avx128fma_invsqrt_f(rsq11);
1432             rinv12           = avx128fma_invsqrt_f(rsq12);
1433             rinv13           = avx128fma_invsqrt_f(rsq13);
1434             rinv21           = avx128fma_invsqrt_f(rsq21);
1435             rinv22           = avx128fma_invsqrt_f(rsq22);
1436             rinv23           = avx128fma_invsqrt_f(rsq23);
1437             rinv31           = avx128fma_invsqrt_f(rsq31);
1438             rinv32           = avx128fma_invsqrt_f(rsq32);
1439             rinv33           = avx128fma_invsqrt_f(rsq33);
1440
1441             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
1442             rinvsq11         = _mm_mul_ps(rinv11,rinv11);
1443             rinvsq12         = _mm_mul_ps(rinv12,rinv12);
1444             rinvsq13         = _mm_mul_ps(rinv13,rinv13);
1445             rinvsq21         = _mm_mul_ps(rinv21,rinv21);
1446             rinvsq22         = _mm_mul_ps(rinv22,rinv22);
1447             rinvsq23         = _mm_mul_ps(rinv23,rinv23);
1448             rinvsq31         = _mm_mul_ps(rinv31,rinv31);
1449             rinvsq32         = _mm_mul_ps(rinv32,rinv32);
1450             rinvsq33         = _mm_mul_ps(rinv33,rinv33);
1451
1452             fjx0             = _mm_setzero_ps();
1453             fjy0             = _mm_setzero_ps();
1454             fjz0             = _mm_setzero_ps();
1455             fjx1             = _mm_setzero_ps();
1456             fjy1             = _mm_setzero_ps();
1457             fjz1             = _mm_setzero_ps();
1458             fjx2             = _mm_setzero_ps();
1459             fjy2             = _mm_setzero_ps();
1460             fjz2             = _mm_setzero_ps();
1461             fjx3             = _mm_setzero_ps();
1462             fjy3             = _mm_setzero_ps();
1463             fjz3             = _mm_setzero_ps();
1464
1465             /**************************
1466              * CALCULATE INTERACTIONS *
1467              **************************/
1468
1469             r00              = _mm_mul_ps(rsq00,rinv00);
1470
1471             /* Analytical LJ-PME */
1472             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
1473             ewcljrsq         = _mm_mul_ps(ewclj2,rsq00);
1474             ewclj6           = _mm_mul_ps(ewclj2,_mm_mul_ps(ewclj2,ewclj2));
1475             exponent         = avx128fma_exp_f(ewcljrsq);
1476             /* poly = exp(-(beta*r)^2) * (1 + (beta*r)^2 + (beta*r)^4 /2) */
1477             poly             = _mm_mul_ps(exponent,_mm_macc_ps(_mm_mul_ps(ewcljrsq,ewcljrsq),one_half,_mm_sub_ps(one,ewcljrsq)));
1478             /* f6A = 6 * C6grid * (1 - poly) */
1479             f6A              = _mm_mul_ps(c6grid_00,_mm_sub_ps(one,poly));
1480             /* f6B = C6grid * exponent * beta^6 */
1481             f6B              = _mm_mul_ps(_mm_mul_ps(c6grid_00,one_sixth),_mm_mul_ps(exponent,ewclj6));
1482             /* fvdw = 12*C12/r13 - ((6*C6 - f6A)/r6 + f6B)/r */
1483             fvdw              = _mm_mul_ps(_mm_macc_ps(_mm_msub_ps(c12_00,rinvsix,_mm_sub_ps(c6_00,f6A)),rinvsix,f6B),rinvsq00);
1484
1485             fscal            = fvdw;
1486
1487              /* Update vectorial force */
1488             fix0             = _mm_macc_ps(dx00,fscal,fix0);
1489             fiy0             = _mm_macc_ps(dy00,fscal,fiy0);
1490             fiz0             = _mm_macc_ps(dz00,fscal,fiz0);
1491
1492             fjx0             = _mm_macc_ps(dx00,fscal,fjx0);
1493             fjy0             = _mm_macc_ps(dy00,fscal,fjy0);
1494             fjz0             = _mm_macc_ps(dz00,fscal,fjz0);
1495
1496             /**************************
1497              * CALCULATE INTERACTIONS *
1498              **************************/
1499
1500             r11              = _mm_mul_ps(rsq11,rinv11);
1501
1502             /* EWALD ELECTROSTATICS */
1503
1504             /* Analytical PME correction */
1505             zeta2            = _mm_mul_ps(beta2,rsq11);
1506             rinv3            = _mm_mul_ps(rinvsq11,rinv11);
1507             pmecorrF         = avx128fma_pmecorrF_f(zeta2);
1508             felec            = _mm_macc_ps(pmecorrF,beta3,rinv3);
1509             felec            = _mm_mul_ps(qq11,felec);
1510
1511             fscal            = felec;
1512
1513              /* Update vectorial force */
1514             fix1             = _mm_macc_ps(dx11,fscal,fix1);
1515             fiy1             = _mm_macc_ps(dy11,fscal,fiy1);
1516             fiz1             = _mm_macc_ps(dz11,fscal,fiz1);
1517
1518             fjx1             = _mm_macc_ps(dx11,fscal,fjx1);
1519             fjy1             = _mm_macc_ps(dy11,fscal,fjy1);
1520             fjz1             = _mm_macc_ps(dz11,fscal,fjz1);
1521
1522             /**************************
1523              * CALCULATE INTERACTIONS *
1524              **************************/
1525
1526             r12              = _mm_mul_ps(rsq12,rinv12);
1527
1528             /* EWALD ELECTROSTATICS */
1529
1530             /* Analytical PME correction */
1531             zeta2            = _mm_mul_ps(beta2,rsq12);
1532             rinv3            = _mm_mul_ps(rinvsq12,rinv12);
1533             pmecorrF         = avx128fma_pmecorrF_f(zeta2);
1534             felec            = _mm_macc_ps(pmecorrF,beta3,rinv3);
1535             felec            = _mm_mul_ps(qq12,felec);
1536
1537             fscal            = felec;
1538
1539              /* Update vectorial force */
1540             fix1             = _mm_macc_ps(dx12,fscal,fix1);
1541             fiy1             = _mm_macc_ps(dy12,fscal,fiy1);
1542             fiz1             = _mm_macc_ps(dz12,fscal,fiz1);
1543
1544             fjx2             = _mm_macc_ps(dx12,fscal,fjx2);
1545             fjy2             = _mm_macc_ps(dy12,fscal,fjy2);
1546             fjz2             = _mm_macc_ps(dz12,fscal,fjz2);
1547
1548             /**************************
1549              * CALCULATE INTERACTIONS *
1550              **************************/
1551
1552             r13              = _mm_mul_ps(rsq13,rinv13);
1553
1554             /* EWALD ELECTROSTATICS */
1555
1556             /* Analytical PME correction */
1557             zeta2            = _mm_mul_ps(beta2,rsq13);
1558             rinv3            = _mm_mul_ps(rinvsq13,rinv13);
1559             pmecorrF         = avx128fma_pmecorrF_f(zeta2);
1560             felec            = _mm_macc_ps(pmecorrF,beta3,rinv3);
1561             felec            = _mm_mul_ps(qq13,felec);
1562
1563             fscal            = felec;
1564
1565              /* Update vectorial force */
1566             fix1             = _mm_macc_ps(dx13,fscal,fix1);
1567             fiy1             = _mm_macc_ps(dy13,fscal,fiy1);
1568             fiz1             = _mm_macc_ps(dz13,fscal,fiz1);
1569
1570             fjx3             = _mm_macc_ps(dx13,fscal,fjx3);
1571             fjy3             = _mm_macc_ps(dy13,fscal,fjy3);
1572             fjz3             = _mm_macc_ps(dz13,fscal,fjz3);
1573
1574             /**************************
1575              * CALCULATE INTERACTIONS *
1576              **************************/
1577
1578             r21              = _mm_mul_ps(rsq21,rinv21);
1579
1580             /* EWALD ELECTROSTATICS */
1581
1582             /* Analytical PME correction */
1583             zeta2            = _mm_mul_ps(beta2,rsq21);
1584             rinv3            = _mm_mul_ps(rinvsq21,rinv21);
1585             pmecorrF         = avx128fma_pmecorrF_f(zeta2);
1586             felec            = _mm_macc_ps(pmecorrF,beta3,rinv3);
1587             felec            = _mm_mul_ps(qq21,felec);
1588
1589             fscal            = felec;
1590
1591              /* Update vectorial force */
1592             fix2             = _mm_macc_ps(dx21,fscal,fix2);
1593             fiy2             = _mm_macc_ps(dy21,fscal,fiy2);
1594             fiz2             = _mm_macc_ps(dz21,fscal,fiz2);
1595
1596             fjx1             = _mm_macc_ps(dx21,fscal,fjx1);
1597             fjy1             = _mm_macc_ps(dy21,fscal,fjy1);
1598             fjz1             = _mm_macc_ps(dz21,fscal,fjz1);
1599
1600             /**************************
1601              * CALCULATE INTERACTIONS *
1602              **************************/
1603
1604             r22              = _mm_mul_ps(rsq22,rinv22);
1605
1606             /* EWALD ELECTROSTATICS */
1607
1608             /* Analytical PME correction */
1609             zeta2            = _mm_mul_ps(beta2,rsq22);
1610             rinv3            = _mm_mul_ps(rinvsq22,rinv22);
1611             pmecorrF         = avx128fma_pmecorrF_f(zeta2);
1612             felec            = _mm_macc_ps(pmecorrF,beta3,rinv3);
1613             felec            = _mm_mul_ps(qq22,felec);
1614
1615             fscal            = felec;
1616
1617              /* Update vectorial force */
1618             fix2             = _mm_macc_ps(dx22,fscal,fix2);
1619             fiy2             = _mm_macc_ps(dy22,fscal,fiy2);
1620             fiz2             = _mm_macc_ps(dz22,fscal,fiz2);
1621
1622             fjx2             = _mm_macc_ps(dx22,fscal,fjx2);
1623             fjy2             = _mm_macc_ps(dy22,fscal,fjy2);
1624             fjz2             = _mm_macc_ps(dz22,fscal,fjz2);
1625
1626             /**************************
1627              * CALCULATE INTERACTIONS *
1628              **************************/
1629
1630             r23              = _mm_mul_ps(rsq23,rinv23);
1631
1632             /* EWALD ELECTROSTATICS */
1633
1634             /* Analytical PME correction */
1635             zeta2            = _mm_mul_ps(beta2,rsq23);
1636             rinv3            = _mm_mul_ps(rinvsq23,rinv23);
1637             pmecorrF         = avx128fma_pmecorrF_f(zeta2);
1638             felec            = _mm_macc_ps(pmecorrF,beta3,rinv3);
1639             felec            = _mm_mul_ps(qq23,felec);
1640
1641             fscal            = felec;
1642
1643              /* Update vectorial force */
1644             fix2             = _mm_macc_ps(dx23,fscal,fix2);
1645             fiy2             = _mm_macc_ps(dy23,fscal,fiy2);
1646             fiz2             = _mm_macc_ps(dz23,fscal,fiz2);
1647
1648             fjx3             = _mm_macc_ps(dx23,fscal,fjx3);
1649             fjy3             = _mm_macc_ps(dy23,fscal,fjy3);
1650             fjz3             = _mm_macc_ps(dz23,fscal,fjz3);
1651
1652             /**************************
1653              * CALCULATE INTERACTIONS *
1654              **************************/
1655
1656             r31              = _mm_mul_ps(rsq31,rinv31);
1657
1658             /* EWALD ELECTROSTATICS */
1659
1660             /* Analytical PME correction */
1661             zeta2            = _mm_mul_ps(beta2,rsq31);
1662             rinv3            = _mm_mul_ps(rinvsq31,rinv31);
1663             pmecorrF         = avx128fma_pmecorrF_f(zeta2);
1664             felec            = _mm_macc_ps(pmecorrF,beta3,rinv3);
1665             felec            = _mm_mul_ps(qq31,felec);
1666
1667             fscal            = felec;
1668
1669              /* Update vectorial force */
1670             fix3             = _mm_macc_ps(dx31,fscal,fix3);
1671             fiy3             = _mm_macc_ps(dy31,fscal,fiy3);
1672             fiz3             = _mm_macc_ps(dz31,fscal,fiz3);
1673
1674             fjx1             = _mm_macc_ps(dx31,fscal,fjx1);
1675             fjy1             = _mm_macc_ps(dy31,fscal,fjy1);
1676             fjz1             = _mm_macc_ps(dz31,fscal,fjz1);
1677
1678             /**************************
1679              * CALCULATE INTERACTIONS *
1680              **************************/
1681
1682             r32              = _mm_mul_ps(rsq32,rinv32);
1683
1684             /* EWALD ELECTROSTATICS */
1685
1686             /* Analytical PME correction */
1687             zeta2            = _mm_mul_ps(beta2,rsq32);
1688             rinv3            = _mm_mul_ps(rinvsq32,rinv32);
1689             pmecorrF         = avx128fma_pmecorrF_f(zeta2);
1690             felec            = _mm_macc_ps(pmecorrF,beta3,rinv3);
1691             felec            = _mm_mul_ps(qq32,felec);
1692
1693             fscal            = felec;
1694
1695              /* Update vectorial force */
1696             fix3             = _mm_macc_ps(dx32,fscal,fix3);
1697             fiy3             = _mm_macc_ps(dy32,fscal,fiy3);
1698             fiz3             = _mm_macc_ps(dz32,fscal,fiz3);
1699
1700             fjx2             = _mm_macc_ps(dx32,fscal,fjx2);
1701             fjy2             = _mm_macc_ps(dy32,fscal,fjy2);
1702             fjz2             = _mm_macc_ps(dz32,fscal,fjz2);
1703
1704             /**************************
1705              * CALCULATE INTERACTIONS *
1706              **************************/
1707
1708             r33              = _mm_mul_ps(rsq33,rinv33);
1709
1710             /* EWALD ELECTROSTATICS */
1711
1712             /* Analytical PME correction */
1713             zeta2            = _mm_mul_ps(beta2,rsq33);
1714             rinv3            = _mm_mul_ps(rinvsq33,rinv33);
1715             pmecorrF         = avx128fma_pmecorrF_f(zeta2);
1716             felec            = _mm_macc_ps(pmecorrF,beta3,rinv3);
1717             felec            = _mm_mul_ps(qq33,felec);
1718
1719             fscal            = felec;
1720
1721              /* Update vectorial force */
1722             fix3             = _mm_macc_ps(dx33,fscal,fix3);
1723             fiy3             = _mm_macc_ps(dy33,fscal,fiy3);
1724             fiz3             = _mm_macc_ps(dz33,fscal,fiz3);
1725
1726             fjx3             = _mm_macc_ps(dx33,fscal,fjx3);
1727             fjy3             = _mm_macc_ps(dy33,fscal,fjy3);
1728             fjz3             = _mm_macc_ps(dz33,fscal,fjz3);
1729
1730             fjptrA             = f+j_coord_offsetA;
1731             fjptrB             = f+j_coord_offsetB;
1732             fjptrC             = f+j_coord_offsetC;
1733             fjptrD             = f+j_coord_offsetD;
1734
1735             gmx_mm_decrement_4rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,
1736                                                    fjx0,fjy0,fjz0,fjx1,fjy1,fjz1,
1737                                                    fjx2,fjy2,fjz2,fjx3,fjy3,fjz3);
1738
1739             /* Inner loop uses 302 flops */
1740         }
1741
1742         if(jidx<j_index_end)
1743         {
1744
1745             /* Get j neighbor index, and coordinate index */
1746             jnrlistA         = jjnr[jidx];
1747             jnrlistB         = jjnr[jidx+1];
1748             jnrlistC         = jjnr[jidx+2];
1749             jnrlistD         = jjnr[jidx+3];
1750             /* Sign of each element will be negative for non-real atoms.
1751              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
1752              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
1753              */
1754             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
1755             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
1756             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
1757             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
1758             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
1759             j_coord_offsetA  = DIM*jnrA;
1760             j_coord_offsetB  = DIM*jnrB;
1761             j_coord_offsetC  = DIM*jnrC;
1762             j_coord_offsetD  = DIM*jnrD;
1763
1764             /* load j atom coordinates */
1765             gmx_mm_load_4rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
1766                                               x+j_coord_offsetC,x+j_coord_offsetD,
1767                                               &jx0,&jy0,&jz0,&jx1,&jy1,&jz1,&jx2,
1768                                               &jy2,&jz2,&jx3,&jy3,&jz3);
1769
1770             /* Calculate displacement vector */
1771             dx00             = _mm_sub_ps(ix0,jx0);
1772             dy00             = _mm_sub_ps(iy0,jy0);
1773             dz00             = _mm_sub_ps(iz0,jz0);
1774             dx11             = _mm_sub_ps(ix1,jx1);
1775             dy11             = _mm_sub_ps(iy1,jy1);
1776             dz11             = _mm_sub_ps(iz1,jz1);
1777             dx12             = _mm_sub_ps(ix1,jx2);
1778             dy12             = _mm_sub_ps(iy1,jy2);
1779             dz12             = _mm_sub_ps(iz1,jz2);
1780             dx13             = _mm_sub_ps(ix1,jx3);
1781             dy13             = _mm_sub_ps(iy1,jy3);
1782             dz13             = _mm_sub_ps(iz1,jz3);
1783             dx21             = _mm_sub_ps(ix2,jx1);
1784             dy21             = _mm_sub_ps(iy2,jy1);
1785             dz21             = _mm_sub_ps(iz2,jz1);
1786             dx22             = _mm_sub_ps(ix2,jx2);
1787             dy22             = _mm_sub_ps(iy2,jy2);
1788             dz22             = _mm_sub_ps(iz2,jz2);
1789             dx23             = _mm_sub_ps(ix2,jx3);
1790             dy23             = _mm_sub_ps(iy2,jy3);
1791             dz23             = _mm_sub_ps(iz2,jz3);
1792             dx31             = _mm_sub_ps(ix3,jx1);
1793             dy31             = _mm_sub_ps(iy3,jy1);
1794             dz31             = _mm_sub_ps(iz3,jz1);
1795             dx32             = _mm_sub_ps(ix3,jx2);
1796             dy32             = _mm_sub_ps(iy3,jy2);
1797             dz32             = _mm_sub_ps(iz3,jz2);
1798             dx33             = _mm_sub_ps(ix3,jx3);
1799             dy33             = _mm_sub_ps(iy3,jy3);
1800             dz33             = _mm_sub_ps(iz3,jz3);
1801
1802             /* Calculate squared distance and things based on it */
1803             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
1804             rsq11            = gmx_mm_calc_rsq_ps(dx11,dy11,dz11);
1805             rsq12            = gmx_mm_calc_rsq_ps(dx12,dy12,dz12);
1806             rsq13            = gmx_mm_calc_rsq_ps(dx13,dy13,dz13);
1807             rsq21            = gmx_mm_calc_rsq_ps(dx21,dy21,dz21);
1808             rsq22            = gmx_mm_calc_rsq_ps(dx22,dy22,dz22);
1809             rsq23            = gmx_mm_calc_rsq_ps(dx23,dy23,dz23);
1810             rsq31            = gmx_mm_calc_rsq_ps(dx31,dy31,dz31);
1811             rsq32            = gmx_mm_calc_rsq_ps(dx32,dy32,dz32);
1812             rsq33            = gmx_mm_calc_rsq_ps(dx33,dy33,dz33);
1813
1814             rinv00           = avx128fma_invsqrt_f(rsq00);
1815             rinv11           = avx128fma_invsqrt_f(rsq11);
1816             rinv12           = avx128fma_invsqrt_f(rsq12);
1817             rinv13           = avx128fma_invsqrt_f(rsq13);
1818             rinv21           = avx128fma_invsqrt_f(rsq21);
1819             rinv22           = avx128fma_invsqrt_f(rsq22);
1820             rinv23           = avx128fma_invsqrt_f(rsq23);
1821             rinv31           = avx128fma_invsqrt_f(rsq31);
1822             rinv32           = avx128fma_invsqrt_f(rsq32);
1823             rinv33           = avx128fma_invsqrt_f(rsq33);
1824
1825             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
1826             rinvsq11         = _mm_mul_ps(rinv11,rinv11);
1827             rinvsq12         = _mm_mul_ps(rinv12,rinv12);
1828             rinvsq13         = _mm_mul_ps(rinv13,rinv13);
1829             rinvsq21         = _mm_mul_ps(rinv21,rinv21);
1830             rinvsq22         = _mm_mul_ps(rinv22,rinv22);
1831             rinvsq23         = _mm_mul_ps(rinv23,rinv23);
1832             rinvsq31         = _mm_mul_ps(rinv31,rinv31);
1833             rinvsq32         = _mm_mul_ps(rinv32,rinv32);
1834             rinvsq33         = _mm_mul_ps(rinv33,rinv33);
1835
1836             fjx0             = _mm_setzero_ps();
1837             fjy0             = _mm_setzero_ps();
1838             fjz0             = _mm_setzero_ps();
1839             fjx1             = _mm_setzero_ps();
1840             fjy1             = _mm_setzero_ps();
1841             fjz1             = _mm_setzero_ps();
1842             fjx2             = _mm_setzero_ps();
1843             fjy2             = _mm_setzero_ps();
1844             fjz2             = _mm_setzero_ps();
1845             fjx3             = _mm_setzero_ps();
1846             fjy3             = _mm_setzero_ps();
1847             fjz3             = _mm_setzero_ps();
1848
1849             /**************************
1850              * CALCULATE INTERACTIONS *
1851              **************************/
1852
1853             r00              = _mm_mul_ps(rsq00,rinv00);
1854             r00              = _mm_andnot_ps(dummy_mask,r00);
1855
1856             /* Analytical LJ-PME */
1857             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
1858             ewcljrsq         = _mm_mul_ps(ewclj2,rsq00);
1859             ewclj6           = _mm_mul_ps(ewclj2,_mm_mul_ps(ewclj2,ewclj2));
1860             exponent         = avx128fma_exp_f(ewcljrsq);
1861             /* poly = exp(-(beta*r)^2) * (1 + (beta*r)^2 + (beta*r)^4 /2) */
1862             poly             = _mm_mul_ps(exponent,_mm_macc_ps(_mm_mul_ps(ewcljrsq,ewcljrsq),one_half,_mm_sub_ps(one,ewcljrsq)));
1863             /* f6A = 6 * C6grid * (1 - poly) */
1864             f6A              = _mm_mul_ps(c6grid_00,_mm_sub_ps(one,poly));
1865             /* f6B = C6grid * exponent * beta^6 */
1866             f6B              = _mm_mul_ps(_mm_mul_ps(c6grid_00,one_sixth),_mm_mul_ps(exponent,ewclj6));
1867             /* fvdw = 12*C12/r13 - ((6*C6 - f6A)/r6 + f6B)/r */
1868             fvdw              = _mm_mul_ps(_mm_macc_ps(_mm_msub_ps(c12_00,rinvsix,_mm_sub_ps(c6_00,f6A)),rinvsix,f6B),rinvsq00);
1869
1870             fscal            = fvdw;
1871
1872             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1873
1874              /* Update vectorial force */
1875             fix0             = _mm_macc_ps(dx00,fscal,fix0);
1876             fiy0             = _mm_macc_ps(dy00,fscal,fiy0);
1877             fiz0             = _mm_macc_ps(dz00,fscal,fiz0);
1878
1879             fjx0             = _mm_macc_ps(dx00,fscal,fjx0);
1880             fjy0             = _mm_macc_ps(dy00,fscal,fjy0);
1881             fjz0             = _mm_macc_ps(dz00,fscal,fjz0);
1882
1883             /**************************
1884              * CALCULATE INTERACTIONS *
1885              **************************/
1886
1887             r11              = _mm_mul_ps(rsq11,rinv11);
1888             r11              = _mm_andnot_ps(dummy_mask,r11);
1889
1890             /* EWALD ELECTROSTATICS */
1891
1892             /* Analytical PME correction */
1893             zeta2            = _mm_mul_ps(beta2,rsq11);
1894             rinv3            = _mm_mul_ps(rinvsq11,rinv11);
1895             pmecorrF         = avx128fma_pmecorrF_f(zeta2);
1896             felec            = _mm_macc_ps(pmecorrF,beta3,rinv3);
1897             felec            = _mm_mul_ps(qq11,felec);
1898
1899             fscal            = felec;
1900
1901             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1902
1903              /* Update vectorial force */
1904             fix1             = _mm_macc_ps(dx11,fscal,fix1);
1905             fiy1             = _mm_macc_ps(dy11,fscal,fiy1);
1906             fiz1             = _mm_macc_ps(dz11,fscal,fiz1);
1907
1908             fjx1             = _mm_macc_ps(dx11,fscal,fjx1);
1909             fjy1             = _mm_macc_ps(dy11,fscal,fjy1);
1910             fjz1             = _mm_macc_ps(dz11,fscal,fjz1);
1911
1912             /**************************
1913              * CALCULATE INTERACTIONS *
1914              **************************/
1915
1916             r12              = _mm_mul_ps(rsq12,rinv12);
1917             r12              = _mm_andnot_ps(dummy_mask,r12);
1918
1919             /* EWALD ELECTROSTATICS */
1920
1921             /* Analytical PME correction */
1922             zeta2            = _mm_mul_ps(beta2,rsq12);
1923             rinv3            = _mm_mul_ps(rinvsq12,rinv12);
1924             pmecorrF         = avx128fma_pmecorrF_f(zeta2);
1925             felec            = _mm_macc_ps(pmecorrF,beta3,rinv3);
1926             felec            = _mm_mul_ps(qq12,felec);
1927
1928             fscal            = felec;
1929
1930             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1931
1932              /* Update vectorial force */
1933             fix1             = _mm_macc_ps(dx12,fscal,fix1);
1934             fiy1             = _mm_macc_ps(dy12,fscal,fiy1);
1935             fiz1             = _mm_macc_ps(dz12,fscal,fiz1);
1936
1937             fjx2             = _mm_macc_ps(dx12,fscal,fjx2);
1938             fjy2             = _mm_macc_ps(dy12,fscal,fjy2);
1939             fjz2             = _mm_macc_ps(dz12,fscal,fjz2);
1940
1941             /**************************
1942              * CALCULATE INTERACTIONS *
1943              **************************/
1944
1945             r13              = _mm_mul_ps(rsq13,rinv13);
1946             r13              = _mm_andnot_ps(dummy_mask,r13);
1947
1948             /* EWALD ELECTROSTATICS */
1949
1950             /* Analytical PME correction */
1951             zeta2            = _mm_mul_ps(beta2,rsq13);
1952             rinv3            = _mm_mul_ps(rinvsq13,rinv13);
1953             pmecorrF         = avx128fma_pmecorrF_f(zeta2);
1954             felec            = _mm_macc_ps(pmecorrF,beta3,rinv3);
1955             felec            = _mm_mul_ps(qq13,felec);
1956
1957             fscal            = felec;
1958
1959             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1960
1961              /* Update vectorial force */
1962             fix1             = _mm_macc_ps(dx13,fscal,fix1);
1963             fiy1             = _mm_macc_ps(dy13,fscal,fiy1);
1964             fiz1             = _mm_macc_ps(dz13,fscal,fiz1);
1965
1966             fjx3             = _mm_macc_ps(dx13,fscal,fjx3);
1967             fjy3             = _mm_macc_ps(dy13,fscal,fjy3);
1968             fjz3             = _mm_macc_ps(dz13,fscal,fjz3);
1969
1970             /**************************
1971              * CALCULATE INTERACTIONS *
1972              **************************/
1973
1974             r21              = _mm_mul_ps(rsq21,rinv21);
1975             r21              = _mm_andnot_ps(dummy_mask,r21);
1976
1977             /* EWALD ELECTROSTATICS */
1978
1979             /* Analytical PME correction */
1980             zeta2            = _mm_mul_ps(beta2,rsq21);
1981             rinv3            = _mm_mul_ps(rinvsq21,rinv21);
1982             pmecorrF         = avx128fma_pmecorrF_f(zeta2);
1983             felec            = _mm_macc_ps(pmecorrF,beta3,rinv3);
1984             felec            = _mm_mul_ps(qq21,felec);
1985
1986             fscal            = felec;
1987
1988             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1989
1990              /* Update vectorial force */
1991             fix2             = _mm_macc_ps(dx21,fscal,fix2);
1992             fiy2             = _mm_macc_ps(dy21,fscal,fiy2);
1993             fiz2             = _mm_macc_ps(dz21,fscal,fiz2);
1994
1995             fjx1             = _mm_macc_ps(dx21,fscal,fjx1);
1996             fjy1             = _mm_macc_ps(dy21,fscal,fjy1);
1997             fjz1             = _mm_macc_ps(dz21,fscal,fjz1);
1998
1999             /**************************
2000              * CALCULATE INTERACTIONS *
2001              **************************/
2002
2003             r22              = _mm_mul_ps(rsq22,rinv22);
2004             r22              = _mm_andnot_ps(dummy_mask,r22);
2005
2006             /* EWALD ELECTROSTATICS */
2007
2008             /* Analytical PME correction */
2009             zeta2            = _mm_mul_ps(beta2,rsq22);
2010             rinv3            = _mm_mul_ps(rinvsq22,rinv22);
2011             pmecorrF         = avx128fma_pmecorrF_f(zeta2);
2012             felec            = _mm_macc_ps(pmecorrF,beta3,rinv3);
2013             felec            = _mm_mul_ps(qq22,felec);
2014
2015             fscal            = felec;
2016
2017             fscal            = _mm_andnot_ps(dummy_mask,fscal);
2018
2019              /* Update vectorial force */
2020             fix2             = _mm_macc_ps(dx22,fscal,fix2);
2021             fiy2             = _mm_macc_ps(dy22,fscal,fiy2);
2022             fiz2             = _mm_macc_ps(dz22,fscal,fiz2);
2023
2024             fjx2             = _mm_macc_ps(dx22,fscal,fjx2);
2025             fjy2             = _mm_macc_ps(dy22,fscal,fjy2);
2026             fjz2             = _mm_macc_ps(dz22,fscal,fjz2);
2027
2028             /**************************
2029              * CALCULATE INTERACTIONS *
2030              **************************/
2031
2032             r23              = _mm_mul_ps(rsq23,rinv23);
2033             r23              = _mm_andnot_ps(dummy_mask,r23);
2034
2035             /* EWALD ELECTROSTATICS */
2036
2037             /* Analytical PME correction */
2038             zeta2            = _mm_mul_ps(beta2,rsq23);
2039             rinv3            = _mm_mul_ps(rinvsq23,rinv23);
2040             pmecorrF         = avx128fma_pmecorrF_f(zeta2);
2041             felec            = _mm_macc_ps(pmecorrF,beta3,rinv3);
2042             felec            = _mm_mul_ps(qq23,felec);
2043
2044             fscal            = felec;
2045
2046             fscal            = _mm_andnot_ps(dummy_mask,fscal);
2047
2048              /* Update vectorial force */
2049             fix2             = _mm_macc_ps(dx23,fscal,fix2);
2050             fiy2             = _mm_macc_ps(dy23,fscal,fiy2);
2051             fiz2             = _mm_macc_ps(dz23,fscal,fiz2);
2052
2053             fjx3             = _mm_macc_ps(dx23,fscal,fjx3);
2054             fjy3             = _mm_macc_ps(dy23,fscal,fjy3);
2055             fjz3             = _mm_macc_ps(dz23,fscal,fjz3);
2056
2057             /**************************
2058              * CALCULATE INTERACTIONS *
2059              **************************/
2060
2061             r31              = _mm_mul_ps(rsq31,rinv31);
2062             r31              = _mm_andnot_ps(dummy_mask,r31);
2063
2064             /* EWALD ELECTROSTATICS */
2065
2066             /* Analytical PME correction */
2067             zeta2            = _mm_mul_ps(beta2,rsq31);
2068             rinv3            = _mm_mul_ps(rinvsq31,rinv31);
2069             pmecorrF         = avx128fma_pmecorrF_f(zeta2);
2070             felec            = _mm_macc_ps(pmecorrF,beta3,rinv3);
2071             felec            = _mm_mul_ps(qq31,felec);
2072
2073             fscal            = felec;
2074
2075             fscal            = _mm_andnot_ps(dummy_mask,fscal);
2076
2077              /* Update vectorial force */
2078             fix3             = _mm_macc_ps(dx31,fscal,fix3);
2079             fiy3             = _mm_macc_ps(dy31,fscal,fiy3);
2080             fiz3             = _mm_macc_ps(dz31,fscal,fiz3);
2081
2082             fjx1             = _mm_macc_ps(dx31,fscal,fjx1);
2083             fjy1             = _mm_macc_ps(dy31,fscal,fjy1);
2084             fjz1             = _mm_macc_ps(dz31,fscal,fjz1);
2085
2086             /**************************
2087              * CALCULATE INTERACTIONS *
2088              **************************/
2089
2090             r32              = _mm_mul_ps(rsq32,rinv32);
2091             r32              = _mm_andnot_ps(dummy_mask,r32);
2092
2093             /* EWALD ELECTROSTATICS */
2094
2095             /* Analytical PME correction */
2096             zeta2            = _mm_mul_ps(beta2,rsq32);
2097             rinv3            = _mm_mul_ps(rinvsq32,rinv32);
2098             pmecorrF         = avx128fma_pmecorrF_f(zeta2);
2099             felec            = _mm_macc_ps(pmecorrF,beta3,rinv3);
2100             felec            = _mm_mul_ps(qq32,felec);
2101
2102             fscal            = felec;
2103
2104             fscal            = _mm_andnot_ps(dummy_mask,fscal);
2105
2106              /* Update vectorial force */
2107             fix3             = _mm_macc_ps(dx32,fscal,fix3);
2108             fiy3             = _mm_macc_ps(dy32,fscal,fiy3);
2109             fiz3             = _mm_macc_ps(dz32,fscal,fiz3);
2110
2111             fjx2             = _mm_macc_ps(dx32,fscal,fjx2);
2112             fjy2             = _mm_macc_ps(dy32,fscal,fjy2);
2113             fjz2             = _mm_macc_ps(dz32,fscal,fjz2);
2114
2115             /**************************
2116              * CALCULATE INTERACTIONS *
2117              **************************/
2118
2119             r33              = _mm_mul_ps(rsq33,rinv33);
2120             r33              = _mm_andnot_ps(dummy_mask,r33);
2121
2122             /* EWALD ELECTROSTATICS */
2123
2124             /* Analytical PME correction */
2125             zeta2            = _mm_mul_ps(beta2,rsq33);
2126             rinv3            = _mm_mul_ps(rinvsq33,rinv33);
2127             pmecorrF         = avx128fma_pmecorrF_f(zeta2);
2128             felec            = _mm_macc_ps(pmecorrF,beta3,rinv3);
2129             felec            = _mm_mul_ps(qq33,felec);
2130
2131             fscal            = felec;
2132
2133             fscal            = _mm_andnot_ps(dummy_mask,fscal);
2134
2135              /* Update vectorial force */
2136             fix3             = _mm_macc_ps(dx33,fscal,fix3);
2137             fiy3             = _mm_macc_ps(dy33,fscal,fiy3);
2138             fiz3             = _mm_macc_ps(dz33,fscal,fiz3);
2139
2140             fjx3             = _mm_macc_ps(dx33,fscal,fjx3);
2141             fjy3             = _mm_macc_ps(dy33,fscal,fjy3);
2142             fjz3             = _mm_macc_ps(dz33,fscal,fjz3);
2143
2144             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
2145             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
2146             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
2147             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
2148
2149             gmx_mm_decrement_4rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,
2150                                                    fjx0,fjy0,fjz0,fjx1,fjy1,fjz1,
2151                                                    fjx2,fjy2,fjz2,fjx3,fjy3,fjz3);
2152
2153             /* Inner loop uses 312 flops */
2154         }
2155
2156         /* End of innermost loop */
2157
2158         gmx_mm_update_iforce_4atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
2159                                               f+i_coord_offset,fshift+i_shift_offset);
2160
2161         /* Increment number of inner iterations */
2162         inneriter                  += j_index_end - j_index_start;
2163
2164         /* Outer loop uses 24 flops */
2165     }
2166
2167     /* Increment number of outer iterations */
2168     outeriter        += nri;
2169
2170     /* Update outer/inner flops */
2171
2172     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W4W4_F,outeriter*24 + inneriter*312);
2173 }