src/gromacs/gmxlib/nonbonded/nb_kernel_avx_256_single/nb_kernel_ElecEw_VdwNone_GeomW3W3_avx_256_single.c

   1 /*
   2  * This file is part of the GROMACS molecular simulation package.
   3  *
   4  * Copyright (c) 2012,2013,2014, by the GROMACS development team, led by
   5  * Mark Abraham, David van der Spoel, Berk Hess, and Erik Lindahl,
   6  * and including many others, as listed in the AUTHORS file in the
   7  * top-level source directory and at http://www.gromacs.org.
   8  *
   9  * GROMACS is free software; you can redistribute it and/or
  10  * modify it under the terms of the GNU Lesser General Public License
  11  * as published by the Free Software Foundation; either version 2.1
  12  * of the License, or (at your option) any later version.
  13  *
  14  * GROMACS is distributed in the hope that it will be useful,
  15  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  16  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  17  * Lesser General Public License for more details.
  18  *
  19  * You should have received a copy of the GNU Lesser General Public
  20  * License along with GROMACS; if not, see
  21  * http://www.gnu.org/licenses, or write to the Free Software Foundation,
  22  * Inc., 51 Franklin Street, Fifth Floor, Boston, MA  02110-1301  USA.
  23  *
  24  * If you want to redistribute modifications to GROMACS, please
  25  * consider that scientific software is very special. Version
  26  * control is crucial - bugs must be traceable. We will be happy to
  27  * consider code for inclusion in the official distribution, but
  28  * derived work must not be called official GROMACS. Details are found
  29  * in the README & COPYING files - if they are missing, get the
  30  * official version at http://www.gromacs.org.
  31  *
  32  * To help us fund GROMACS development, we humbly ask that you cite
  33  * the research papers on the package. Check out http://www.gromacs.org.
  34  */
  35 /*
  36  * Note: this file was generated by the GROMACS avx_256_single kernel generator.
  37  */
  38 #include "config.h"
  39
  40 #include <math.h>
  41
  42 #include "../nb_kernel.h"
  43 #include "types/simple.h"
  44 #include "gromacs/math/vec.h"
  45 #include "nrnb.h"
  46
  47 #include "gromacs/simd/math_x86_avx_256_single.h"
  48 #include "kernelutil_x86_avx_256_single.h"
  49
  50 /*
  51  * Gromacs nonbonded kernel:   nb_kernel_ElecEw_VdwNone_GeomW3W3_VF_avx_256_single
  52  * Electrostatics interaction: Ewald
  53  * VdW interaction:            None
  54  * Geometry:                   Water3-Water3
  55  * Calculate force/pot:        PotentialAndForce
  56  */
  57 void
  58 nb_kernel_ElecEw_VdwNone_GeomW3W3_VF_avx_256_single
  59                     (t_nblist                    * gmx_restrict       nlist,
  60                      rvec                        * gmx_restrict          xx,
  61                      rvec                        * gmx_restrict          ff,
  62                      t_forcerec                  * gmx_restrict          fr,
  63                      t_mdatoms                   * gmx_restrict     mdatoms,
  64                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
  65                      t_nrnb                      * gmx_restrict        nrnb)
  66 {
  67     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
  68      * just 0 for non-waters.
  69      * Suffixes A,B,C,D,E,F,G,H refer to j loop unrolling done with AVX, e.g. for the eight different
  70      * jnr indices corresponding to data put in the four positions in the SIMD register.
  71      */
  72     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
  73     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
  74     int              jnrA,jnrB,jnrC,jnrD;
  75     int              jnrE,jnrF,jnrG,jnrH;
  76     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
  77     int              jnrlistE,jnrlistF,jnrlistG,jnrlistH;
  78     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
  79     int              j_coord_offsetE,j_coord_offsetF,j_coord_offsetG,j_coord_offsetH;
  80     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
  81     real             rcutoff_scalar;
  82     real             *shiftvec,*fshift,*x,*f;
  83     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD,*fjptrE,*fjptrF,*fjptrG,*fjptrH;
  84     real             scratch[4*DIM];
  85     __m256           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
  86     real *           vdwioffsetptr0;
  87     __m256           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
  88     real *           vdwioffsetptr1;
  89     __m256           ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
  90     real *           vdwioffsetptr2;
  91     __m256           ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
  92     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D,vdwjidx0E,vdwjidx0F,vdwjidx0G,vdwjidx0H;
  93     __m256           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  94     int              vdwjidx1A,vdwjidx1B,vdwjidx1C,vdwjidx1D,vdwjidx1E,vdwjidx1F,vdwjidx1G,vdwjidx1H;
  95     __m256           jx1,jy1,jz1,fjx1,fjy1,fjz1,jq1,isaj1;
  96     int              vdwjidx2A,vdwjidx2B,vdwjidx2C,vdwjidx2D,vdwjidx2E,vdwjidx2F,vdwjidx2G,vdwjidx2H;
  97     __m256           jx2,jy2,jz2,fjx2,fjy2,fjz2,jq2,isaj2;
  98     __m256           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  99     __m256           dx01,dy01,dz01,rsq01,rinv01,rinvsq01,r01,qq01,c6_01,c12_01;
 100     __m256           dx02,dy02,dz02,rsq02,rinv02,rinvsq02,r02,qq02,c6_02,c12_02;
 101     __m256           dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
 102     __m256           dx11,dy11,dz11,rsq11,rinv11,rinvsq11,r11,qq11,c6_11,c12_11;
 103     __m256           dx12,dy12,dz12,rsq12,rinv12,rinvsq12,r12,qq12,c6_12,c12_12;
 104     __m256           dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
 105     __m256           dx21,dy21,dz21,rsq21,rinv21,rinvsq21,r21,qq21,c6_21,c12_21;
 106     __m256           dx22,dy22,dz22,rsq22,rinv22,rinvsq22,r22,qq22,c6_22,c12_22;
 107     __m256           velec,felec,velecsum,facel,crf,krf,krf2;
 108     real             *charge;
 109     __m256i          ewitab;
 110     __m128i          ewitab_lo,ewitab_hi;
 111     __m256           ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
 112     __m256           beta,beta2,beta3,zeta2,pmecorrF,pmecorrV,rinv3;
 113     real             *ewtab;
 114     __m256           dummy_mask,cutoff_mask;
 115     __m256           signbit = _mm256_castsi256_ps( _mm256_set1_epi32(0x80000000) );
 116     __m256           one     = _mm256_set1_ps(1.0);
 117     __m256           two     = _mm256_set1_ps(2.0);
 118     x                = xx[0];
 119     f                = ff[0];
 120
 121     nri              = nlist->nri;
 122     iinr             = nlist->iinr;
 123     jindex           = nlist->jindex;
 124     jjnr             = nlist->jjnr;
 125     shiftidx         = nlist->shift;
 126     gid              = nlist->gid;
 127     shiftvec         = fr->shift_vec[0];
 128     fshift           = fr->fshift[0];
 129     facel            = _mm256_set1_ps(fr->epsfac);
 130     charge           = mdatoms->chargeA;
 131
 132     sh_ewald         = _mm256_set1_ps(fr->ic->sh_ewald);
 133     beta             = _mm256_set1_ps(fr->ic->ewaldcoeff_q);
 134     beta2            = _mm256_mul_ps(beta,beta);
 135     beta3            = _mm256_mul_ps(beta,beta2);
 136
 137     ewtab            = fr->ic->tabq_coul_FDV0;
 138     ewtabscale       = _mm256_set1_ps(fr->ic->tabq_scale);
 139     ewtabhalfspace   = _mm256_set1_ps(0.5/fr->ic->tabq_scale);
 140
 141     /* Setup water-specific parameters */
 142     inr              = nlist->iinr[0];
 143     iq0              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+0]));
 144     iq1              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+1]));
 145     iq2              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+2]));
 146
 147     jq0              = _mm256_set1_ps(charge[inr+0]);
 148     jq1              = _mm256_set1_ps(charge[inr+1]);
 149     jq2              = _mm256_set1_ps(charge[inr+2]);
 150     qq00             = _mm256_mul_ps(iq0,jq0);
 151     qq01             = _mm256_mul_ps(iq0,jq1);
 152     qq02             = _mm256_mul_ps(iq0,jq2);
 153     qq10             = _mm256_mul_ps(iq1,jq0);
 154     qq11             = _mm256_mul_ps(iq1,jq1);
 155     qq12             = _mm256_mul_ps(iq1,jq2);
 156     qq20             = _mm256_mul_ps(iq2,jq0);
 157     qq21             = _mm256_mul_ps(iq2,jq1);
 158     qq22             = _mm256_mul_ps(iq2,jq2);
 159
 160     /* Avoid stupid compiler warnings */
 161     jnrA = jnrB = jnrC = jnrD = jnrE = jnrF = jnrG = jnrH = 0;
 162     j_coord_offsetA = 0;
 163     j_coord_offsetB = 0;
 164     j_coord_offsetC = 0;
 165     j_coord_offsetD = 0;
 166     j_coord_offsetE = 0;
 167     j_coord_offsetF = 0;
 168     j_coord_offsetG = 0;
 169     j_coord_offsetH = 0;
 170
 171     outeriter        = 0;
 172     inneriter        = 0;
 173
 174     for(iidx=0;iidx<4*DIM;iidx++)
 175     {
 176         scratch[iidx] = 0.0;
 177     }
 178
 179     /* Start outer loop over neighborlists */
 180     for(iidx=0; iidx<nri; iidx++)
 181     {
 182         /* Load shift vector for this list */
 183         i_shift_offset   = DIM*shiftidx[iidx];
 184
 185         /* Load limits for loop over neighbors */
 186         j_index_start    = jindex[iidx];
 187         j_index_end      = jindex[iidx+1];
 188
 189         /* Get outer coordinate index */
 190         inr              = iinr[iidx];
 191         i_coord_offset   = DIM*inr;
 192
 193         /* Load i particle coords and add shift vector */
 194         gmx_mm256_load_shift_and_3rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
 195                                                     &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 196
 197         fix0             = _mm256_setzero_ps();
 198         fiy0             = _mm256_setzero_ps();
 199         fiz0             = _mm256_setzero_ps();
 200         fix1             = _mm256_setzero_ps();
 201         fiy1             = _mm256_setzero_ps();
 202         fiz1             = _mm256_setzero_ps();
 203         fix2             = _mm256_setzero_ps();
 204         fiy2             = _mm256_setzero_ps();
 205         fiz2             = _mm256_setzero_ps();
 206
 207         /* Reset potential sums */
 208         velecsum         = _mm256_setzero_ps();
 209
 210         /* Start inner kernel loop */
 211         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+7]>=0; jidx+=8)
 212         {
 213
 214             /* Get j neighbor index, and coordinate index */
 215             jnrA             = jjnr[jidx];
 216             jnrB             = jjnr[jidx+1];
 217             jnrC             = jjnr[jidx+2];
 218             jnrD             = jjnr[jidx+3];
 219             jnrE             = jjnr[jidx+4];
 220             jnrF             = jjnr[jidx+5];
 221             jnrG             = jjnr[jidx+6];
 222             jnrH             = jjnr[jidx+7];
 223             j_coord_offsetA  = DIM*jnrA;
 224             j_coord_offsetB  = DIM*jnrB;
 225             j_coord_offsetC  = DIM*jnrC;
 226             j_coord_offsetD  = DIM*jnrD;
 227             j_coord_offsetE  = DIM*jnrE;
 228             j_coord_offsetF  = DIM*jnrF;
 229             j_coord_offsetG  = DIM*jnrG;
 230             j_coord_offsetH  = DIM*jnrH;
 231
 232             /* load j atom coordinates */
 233             gmx_mm256_load_3rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 234                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 235                                                  x+j_coord_offsetE,x+j_coord_offsetF,
 236                                                  x+j_coord_offsetG,x+j_coord_offsetH,
 237                                               &jx0,&jy0,&jz0,&jx1,&jy1,&jz1,&jx2,&jy2,&jz2);
 238
 239             /* Calculate displacement vector */
 240             dx00             = _mm256_sub_ps(ix0,jx0);
 241             dy00             = _mm256_sub_ps(iy0,jy0);
 242             dz00             = _mm256_sub_ps(iz0,jz0);
 243             dx01             = _mm256_sub_ps(ix0,jx1);
 244             dy01             = _mm256_sub_ps(iy0,jy1);
 245             dz01             = _mm256_sub_ps(iz0,jz1);
 246             dx02             = _mm256_sub_ps(ix0,jx2);
 247             dy02             = _mm256_sub_ps(iy0,jy2);
 248             dz02             = _mm256_sub_ps(iz0,jz2);
 249             dx10             = _mm256_sub_ps(ix1,jx0);
 250             dy10             = _mm256_sub_ps(iy1,jy0);
 251             dz10             = _mm256_sub_ps(iz1,jz0);
 252             dx11             = _mm256_sub_ps(ix1,jx1);
 253             dy11             = _mm256_sub_ps(iy1,jy1);
 254             dz11             = _mm256_sub_ps(iz1,jz1);
 255             dx12             = _mm256_sub_ps(ix1,jx2);
 256             dy12             = _mm256_sub_ps(iy1,jy2);
 257             dz12             = _mm256_sub_ps(iz1,jz2);
 258             dx20             = _mm256_sub_ps(ix2,jx0);
 259             dy20             = _mm256_sub_ps(iy2,jy0);
 260             dz20             = _mm256_sub_ps(iz2,jz0);
 261             dx21             = _mm256_sub_ps(ix2,jx1);
 262             dy21             = _mm256_sub_ps(iy2,jy1);
 263             dz21             = _mm256_sub_ps(iz2,jz1);
 264             dx22             = _mm256_sub_ps(ix2,jx2);
 265             dy22             = _mm256_sub_ps(iy2,jy2);
 266             dz22             = _mm256_sub_ps(iz2,jz2);
 267
 268             /* Calculate squared distance and things based on it */
 269             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
 270             rsq01            = gmx_mm256_calc_rsq_ps(dx01,dy01,dz01);
 271             rsq02            = gmx_mm256_calc_rsq_ps(dx02,dy02,dz02);
 272             rsq10            = gmx_mm256_calc_rsq_ps(dx10,dy10,dz10);
 273             rsq11            = gmx_mm256_calc_rsq_ps(dx11,dy11,dz11);
 274             rsq12            = gmx_mm256_calc_rsq_ps(dx12,dy12,dz12);
 275             rsq20            = gmx_mm256_calc_rsq_ps(dx20,dy20,dz20);
 276             rsq21            = gmx_mm256_calc_rsq_ps(dx21,dy21,dz21);
 277             rsq22            = gmx_mm256_calc_rsq_ps(dx22,dy22,dz22);
 278
 279             rinv00           = gmx_mm256_invsqrt_ps(rsq00);
 280             rinv01           = gmx_mm256_invsqrt_ps(rsq01);
 281             rinv02           = gmx_mm256_invsqrt_ps(rsq02);
 282             rinv10           = gmx_mm256_invsqrt_ps(rsq10);
 283             rinv11           = gmx_mm256_invsqrt_ps(rsq11);
 284             rinv12           = gmx_mm256_invsqrt_ps(rsq12);
 285             rinv20           = gmx_mm256_invsqrt_ps(rsq20);
 286             rinv21           = gmx_mm256_invsqrt_ps(rsq21);
 287             rinv22           = gmx_mm256_invsqrt_ps(rsq22);
 288
 289             rinvsq00         = _mm256_mul_ps(rinv00,rinv00);
 290             rinvsq01         = _mm256_mul_ps(rinv01,rinv01);
 291             rinvsq02         = _mm256_mul_ps(rinv02,rinv02);
 292             rinvsq10         = _mm256_mul_ps(rinv10,rinv10);
 293             rinvsq11         = _mm256_mul_ps(rinv11,rinv11);
 294             rinvsq12         = _mm256_mul_ps(rinv12,rinv12);
 295             rinvsq20         = _mm256_mul_ps(rinv20,rinv20);
 296             rinvsq21         = _mm256_mul_ps(rinv21,rinv21);
 297             rinvsq22         = _mm256_mul_ps(rinv22,rinv22);
 298
 299             fjx0             = _mm256_setzero_ps();
 300             fjy0             = _mm256_setzero_ps();
 301             fjz0             = _mm256_setzero_ps();
 302             fjx1             = _mm256_setzero_ps();
 303             fjy1             = _mm256_setzero_ps();
 304             fjz1             = _mm256_setzero_ps();
 305             fjx2             = _mm256_setzero_ps();
 306             fjy2             = _mm256_setzero_ps();
 307             fjz2             = _mm256_setzero_ps();
 308
 309             /**************************
 310              * CALCULATE INTERACTIONS *
 311              **************************/
 312
 313             r00              = _mm256_mul_ps(rsq00,rinv00);
 314
 315             /* EWALD ELECTROSTATICS */
 316
 317             /* Analytical PME correction */
 318             zeta2            = _mm256_mul_ps(beta2,rsq00);
 319             rinv3            = _mm256_mul_ps(rinvsq00,rinv00);
 320             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
 321             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
 322             felec            = _mm256_mul_ps(qq00,felec);
 323             pmecorrV         = gmx_mm256_pmecorrV_ps(zeta2);
 324             pmecorrV         = _mm256_mul_ps(pmecorrV,beta);
 325             velec            = _mm256_sub_ps(rinv00,pmecorrV);
 326             velec            = _mm256_mul_ps(qq00,velec);
 327
 328             /* Update potential sum for this i atom from the interaction with this j atom. */
 329             velecsum         = _mm256_add_ps(velecsum,velec);
 330
 331             fscal            = felec;
 332
 333             /* Calculate temporary vectorial force */
 334             tx               = _mm256_mul_ps(fscal,dx00);
 335             ty               = _mm256_mul_ps(fscal,dy00);
 336             tz               = _mm256_mul_ps(fscal,dz00);
 337
 338             /* Update vectorial force */
 339             fix0             = _mm256_add_ps(fix0,tx);
 340             fiy0             = _mm256_add_ps(fiy0,ty);
 341             fiz0             = _mm256_add_ps(fiz0,tz);
 342
 343             fjx0             = _mm256_add_ps(fjx0,tx);
 344             fjy0             = _mm256_add_ps(fjy0,ty);
 345             fjz0             = _mm256_add_ps(fjz0,tz);
 346
 347             /**************************
 348              * CALCULATE INTERACTIONS *
 349              **************************/
 350
 351             r01              = _mm256_mul_ps(rsq01,rinv01);
 352
 353             /* EWALD ELECTROSTATICS */
 354
 355             /* Analytical PME correction */
 356             zeta2            = _mm256_mul_ps(beta2,rsq01);
 357             rinv3            = _mm256_mul_ps(rinvsq01,rinv01);
 358             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
 359             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
 360             felec            = _mm256_mul_ps(qq01,felec);
 361             pmecorrV         = gmx_mm256_pmecorrV_ps(zeta2);
 362             pmecorrV         = _mm256_mul_ps(pmecorrV,beta);
 363             velec            = _mm256_sub_ps(rinv01,pmecorrV);
 364             velec            = _mm256_mul_ps(qq01,velec);
 365
 366             /* Update potential sum for this i atom from the interaction with this j atom. */
 367             velecsum         = _mm256_add_ps(velecsum,velec);
 368
 369             fscal            = felec;
 370
 371             /* Calculate temporary vectorial force */
 372             tx               = _mm256_mul_ps(fscal,dx01);
 373             ty               = _mm256_mul_ps(fscal,dy01);
 374             tz               = _mm256_mul_ps(fscal,dz01);
 375
 376             /* Update vectorial force */
 377             fix0             = _mm256_add_ps(fix0,tx);
 378             fiy0             = _mm256_add_ps(fiy0,ty);
 379             fiz0             = _mm256_add_ps(fiz0,tz);
 380
 381             fjx1             = _mm256_add_ps(fjx1,tx);
 382             fjy1             = _mm256_add_ps(fjy1,ty);
 383             fjz1             = _mm256_add_ps(fjz1,tz);
 384
 385             /**************************
 386              * CALCULATE INTERACTIONS *
 387              **************************/
 388
 389             r02              = _mm256_mul_ps(rsq02,rinv02);
 390
 391             /* EWALD ELECTROSTATICS */
 392
 393             /* Analytical PME correction */
 394             zeta2            = _mm256_mul_ps(beta2,rsq02);
 395             rinv3            = _mm256_mul_ps(rinvsq02,rinv02);
 396             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
 397             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
 398             felec            = _mm256_mul_ps(qq02,felec);
 399             pmecorrV         = gmx_mm256_pmecorrV_ps(zeta2);
 400             pmecorrV         = _mm256_mul_ps(pmecorrV,beta);
 401             velec            = _mm256_sub_ps(rinv02,pmecorrV);
 402             velec            = _mm256_mul_ps(qq02,velec);
 403
 404             /* Update potential sum for this i atom from the interaction with this j atom. */
 405             velecsum         = _mm256_add_ps(velecsum,velec);
 406
 407             fscal            = felec;
 408
 409             /* Calculate temporary vectorial force */
 410             tx               = _mm256_mul_ps(fscal,dx02);
 411             ty               = _mm256_mul_ps(fscal,dy02);
 412             tz               = _mm256_mul_ps(fscal,dz02);
 413
 414             /* Update vectorial force */
 415             fix0             = _mm256_add_ps(fix0,tx);
 416             fiy0             = _mm256_add_ps(fiy0,ty);
 417             fiz0             = _mm256_add_ps(fiz0,tz);
 418
 419             fjx2             = _mm256_add_ps(fjx2,tx);
 420             fjy2             = _mm256_add_ps(fjy2,ty);
 421             fjz2             = _mm256_add_ps(fjz2,tz);
 422
 423             /**************************
 424              * CALCULATE INTERACTIONS *
 425              **************************/
 426
 427             r10              = _mm256_mul_ps(rsq10,rinv10);
 428
 429             /* EWALD ELECTROSTATICS */
 430
 431             /* Analytical PME correction */
 432             zeta2            = _mm256_mul_ps(beta2,rsq10);
 433             rinv3            = _mm256_mul_ps(rinvsq10,rinv10);
 434             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
 435             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
 436             felec            = _mm256_mul_ps(qq10,felec);
 437             pmecorrV         = gmx_mm256_pmecorrV_ps(zeta2);
 438             pmecorrV         = _mm256_mul_ps(pmecorrV,beta);
 439             velec            = _mm256_sub_ps(rinv10,pmecorrV);
 440             velec            = _mm256_mul_ps(qq10,velec);
 441
 442             /* Update potential sum for this i atom from the interaction with this j atom. */
 443             velecsum         = _mm256_add_ps(velecsum,velec);
 444
 445             fscal            = felec;
 446
 447             /* Calculate temporary vectorial force */
 448             tx               = _mm256_mul_ps(fscal,dx10);
 449             ty               = _mm256_mul_ps(fscal,dy10);
 450             tz               = _mm256_mul_ps(fscal,dz10);
 451
 452             /* Update vectorial force */
 453             fix1             = _mm256_add_ps(fix1,tx);
 454             fiy1             = _mm256_add_ps(fiy1,ty);
 455             fiz1             = _mm256_add_ps(fiz1,tz);
 456
 457             fjx0             = _mm256_add_ps(fjx0,tx);
 458             fjy0             = _mm256_add_ps(fjy0,ty);
 459             fjz0             = _mm256_add_ps(fjz0,tz);
 460
 461             /**************************
 462              * CALCULATE INTERACTIONS *
 463              **************************/
 464
 465             r11              = _mm256_mul_ps(rsq11,rinv11);
 466
 467             /* EWALD ELECTROSTATICS */
 468
 469             /* Analytical PME correction */
 470             zeta2            = _mm256_mul_ps(beta2,rsq11);
 471             rinv3            = _mm256_mul_ps(rinvsq11,rinv11);
 472             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
 473             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
 474             felec            = _mm256_mul_ps(qq11,felec);
 475             pmecorrV         = gmx_mm256_pmecorrV_ps(zeta2);
 476             pmecorrV         = _mm256_mul_ps(pmecorrV,beta);
 477             velec            = _mm256_sub_ps(rinv11,pmecorrV);
 478             velec            = _mm256_mul_ps(qq11,velec);
 479
 480             /* Update potential sum for this i atom from the interaction with this j atom. */
 481             velecsum         = _mm256_add_ps(velecsum,velec);
 482
 483             fscal            = felec;
 484
 485             /* Calculate temporary vectorial force */
 486             tx               = _mm256_mul_ps(fscal,dx11);
 487             ty               = _mm256_mul_ps(fscal,dy11);
 488             tz               = _mm256_mul_ps(fscal,dz11);
 489
 490             /* Update vectorial force */
 491             fix1             = _mm256_add_ps(fix1,tx);
 492             fiy1             = _mm256_add_ps(fiy1,ty);
 493             fiz1             = _mm256_add_ps(fiz1,tz);
 494
 495             fjx1             = _mm256_add_ps(fjx1,tx);
 496             fjy1             = _mm256_add_ps(fjy1,ty);
 497             fjz1             = _mm256_add_ps(fjz1,tz);
 498
 499             /**************************
 500              * CALCULATE INTERACTIONS *
 501              **************************/
 502
 503             r12              = _mm256_mul_ps(rsq12,rinv12);
 504
 505             /* EWALD ELECTROSTATICS */
 506
 507             /* Analytical PME correction */
 508             zeta2            = _mm256_mul_ps(beta2,rsq12);
 509             rinv3            = _mm256_mul_ps(rinvsq12,rinv12);
 510             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
 511             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
 512             felec            = _mm256_mul_ps(qq12,felec);
 513             pmecorrV         = gmx_mm256_pmecorrV_ps(zeta2);
 514             pmecorrV         = _mm256_mul_ps(pmecorrV,beta);
 515             velec            = _mm256_sub_ps(rinv12,pmecorrV);
 516             velec            = _mm256_mul_ps(qq12,velec);
 517
 518             /* Update potential sum for this i atom from the interaction with this j atom. */
 519             velecsum         = _mm256_add_ps(velecsum,velec);
 520
 521             fscal            = felec;
 522
 523             /* Calculate temporary vectorial force */
 524             tx               = _mm256_mul_ps(fscal,dx12);
 525             ty               = _mm256_mul_ps(fscal,dy12);
 526             tz               = _mm256_mul_ps(fscal,dz12);
 527
 528             /* Update vectorial force */
 529             fix1             = _mm256_add_ps(fix1,tx);
 530             fiy1             = _mm256_add_ps(fiy1,ty);
 531             fiz1             = _mm256_add_ps(fiz1,tz);
 532
 533             fjx2             = _mm256_add_ps(fjx2,tx);
 534             fjy2             = _mm256_add_ps(fjy2,ty);
 535             fjz2             = _mm256_add_ps(fjz2,tz);
 536
 537             /**************************
 538              * CALCULATE INTERACTIONS *
 539              **************************/
 540
 541             r20              = _mm256_mul_ps(rsq20,rinv20);
 542
 543             /* EWALD ELECTROSTATICS */
 544
 545             /* Analytical PME correction */
 546             zeta2            = _mm256_mul_ps(beta2,rsq20);
 547             rinv3            = _mm256_mul_ps(rinvsq20,rinv20);
 548             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
 549             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
 550             felec            = _mm256_mul_ps(qq20,felec);
 551             pmecorrV         = gmx_mm256_pmecorrV_ps(zeta2);
 552             pmecorrV         = _mm256_mul_ps(pmecorrV,beta);
 553             velec            = _mm256_sub_ps(rinv20,pmecorrV);
 554             velec            = _mm256_mul_ps(qq20,velec);
 555
 556             /* Update potential sum for this i atom from the interaction with this j atom. */
 557             velecsum         = _mm256_add_ps(velecsum,velec);
 558
 559             fscal            = felec;
 560
 561             /* Calculate temporary vectorial force */
 562             tx               = _mm256_mul_ps(fscal,dx20);
 563             ty               = _mm256_mul_ps(fscal,dy20);
 564             tz               = _mm256_mul_ps(fscal,dz20);
 565
 566             /* Update vectorial force */
 567             fix2             = _mm256_add_ps(fix2,tx);
 568             fiy2             = _mm256_add_ps(fiy2,ty);
 569             fiz2             = _mm256_add_ps(fiz2,tz);
 570
 571             fjx0             = _mm256_add_ps(fjx0,tx);
 572             fjy0             = _mm256_add_ps(fjy0,ty);
 573             fjz0             = _mm256_add_ps(fjz0,tz);
 574
 575             /**************************
 576              * CALCULATE INTERACTIONS *
 577              **************************/
 578
 579             r21              = _mm256_mul_ps(rsq21,rinv21);
 580
 581             /* EWALD ELECTROSTATICS */
 582
 583             /* Analytical PME correction */
 584             zeta2            = _mm256_mul_ps(beta2,rsq21);
 585             rinv3            = _mm256_mul_ps(rinvsq21,rinv21);
 586             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
 587             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
 588             felec            = _mm256_mul_ps(qq21,felec);
 589             pmecorrV         = gmx_mm256_pmecorrV_ps(zeta2);
 590             pmecorrV         = _mm256_mul_ps(pmecorrV,beta);
 591             velec            = _mm256_sub_ps(rinv21,pmecorrV);
 592             velec            = _mm256_mul_ps(qq21,velec);
 593
 594             /* Update potential sum for this i atom from the interaction with this j atom. */
 595             velecsum         = _mm256_add_ps(velecsum,velec);
 596
 597             fscal            = felec;
 598
 599             /* Calculate temporary vectorial force */
 600             tx               = _mm256_mul_ps(fscal,dx21);
 601             ty               = _mm256_mul_ps(fscal,dy21);
 602             tz               = _mm256_mul_ps(fscal,dz21);
 603
 604             /* Update vectorial force */
 605             fix2             = _mm256_add_ps(fix2,tx);
 606             fiy2             = _mm256_add_ps(fiy2,ty);
 607             fiz2             = _mm256_add_ps(fiz2,tz);
 608
 609             fjx1             = _mm256_add_ps(fjx1,tx);
 610             fjy1             = _mm256_add_ps(fjy1,ty);
 611             fjz1             = _mm256_add_ps(fjz1,tz);
 612
 613             /**************************
 614              * CALCULATE INTERACTIONS *
 615              **************************/
 616
 617             r22              = _mm256_mul_ps(rsq22,rinv22);
 618
 619             /* EWALD ELECTROSTATICS */
 620
 621             /* Analytical PME correction */
 622             zeta2            = _mm256_mul_ps(beta2,rsq22);
 623             rinv3            = _mm256_mul_ps(rinvsq22,rinv22);
 624             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
 625             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
 626             felec            = _mm256_mul_ps(qq22,felec);
 627             pmecorrV         = gmx_mm256_pmecorrV_ps(zeta2);
 628             pmecorrV         = _mm256_mul_ps(pmecorrV,beta);
 629             velec            = _mm256_sub_ps(rinv22,pmecorrV);
 630             velec            = _mm256_mul_ps(qq22,velec);
 631
 632             /* Update potential sum for this i atom from the interaction with this j atom. */
 633             velecsum         = _mm256_add_ps(velecsum,velec);
 634
 635             fscal            = felec;
 636
 637             /* Calculate temporary vectorial force */
 638             tx               = _mm256_mul_ps(fscal,dx22);
 639             ty               = _mm256_mul_ps(fscal,dy22);
 640             tz               = _mm256_mul_ps(fscal,dz22);
 641
 642             /* Update vectorial force */
 643             fix2             = _mm256_add_ps(fix2,tx);
 644             fiy2             = _mm256_add_ps(fiy2,ty);
 645             fiz2             = _mm256_add_ps(fiz2,tz);
 646
 647             fjx2             = _mm256_add_ps(fjx2,tx);
 648             fjy2             = _mm256_add_ps(fjy2,ty);
 649             fjz2             = _mm256_add_ps(fjz2,tz);
 650
 651             fjptrA             = f+j_coord_offsetA;
 652             fjptrB             = f+j_coord_offsetB;
 653             fjptrC             = f+j_coord_offsetC;
 654             fjptrD             = f+j_coord_offsetD;
 655             fjptrE             = f+j_coord_offsetE;
 656             fjptrF             = f+j_coord_offsetF;
 657             fjptrG             = f+j_coord_offsetG;
 658             fjptrH             = f+j_coord_offsetH;
 659
 660             gmx_mm256_decrement_3rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,
 661                                                       fjx0,fjy0,fjz0,fjx1,fjy1,fjz1,fjx2,fjy2,fjz2);
 662
 663             /* Inner loop uses 756 flops */
 664         }
 665
 666         if(jidx<j_index_end)
 667         {
 668
 669             /* Get j neighbor index, and coordinate index */
 670             jnrlistA         = jjnr[jidx];
 671             jnrlistB         = jjnr[jidx+1];
 672             jnrlistC         = jjnr[jidx+2];
 673             jnrlistD         = jjnr[jidx+3];
 674             jnrlistE         = jjnr[jidx+4];
 675             jnrlistF         = jjnr[jidx+5];
 676             jnrlistG         = jjnr[jidx+6];
 677             jnrlistH         = jjnr[jidx+7];
 678             /* Sign of each element will be negative for non-real atoms.
 679              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 680              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 681              */
 682             dummy_mask = gmx_mm256_set_m128(gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx+4)),_mm_setzero_si128())),
 683                                             gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128())));
 684
 685             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 686             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 687             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 688             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 689             jnrE       = (jnrlistE>=0) ? jnrlistE : 0;
 690             jnrF       = (jnrlistF>=0) ? jnrlistF : 0;
 691             jnrG       = (jnrlistG>=0) ? jnrlistG : 0;
 692             jnrH       = (jnrlistH>=0) ? jnrlistH : 0;
 693             j_coord_offsetA  = DIM*jnrA;
 694             j_coord_offsetB  = DIM*jnrB;
 695             j_coord_offsetC  = DIM*jnrC;
 696             j_coord_offsetD  = DIM*jnrD;
 697             j_coord_offsetE  = DIM*jnrE;
 698             j_coord_offsetF  = DIM*jnrF;
 699             j_coord_offsetG  = DIM*jnrG;
 700             j_coord_offsetH  = DIM*jnrH;
 701
 702             /* load j atom coordinates */
 703             gmx_mm256_load_3rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 704                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 705                                                  x+j_coord_offsetE,x+j_coord_offsetF,
 706                                                  x+j_coord_offsetG,x+j_coord_offsetH,
 707                                               &jx0,&jy0,&jz0,&jx1,&jy1,&jz1,&jx2,&jy2,&jz2);
 708
 709             /* Calculate displacement vector */
 710             dx00             = _mm256_sub_ps(ix0,jx0);
 711             dy00             = _mm256_sub_ps(iy0,jy0);
 712             dz00             = _mm256_sub_ps(iz0,jz0);
 713             dx01             = _mm256_sub_ps(ix0,jx1);
 714             dy01             = _mm256_sub_ps(iy0,jy1);
 715             dz01             = _mm256_sub_ps(iz0,jz1);
 716             dx02             = _mm256_sub_ps(ix0,jx2);
 717             dy02             = _mm256_sub_ps(iy0,jy2);
 718             dz02             = _mm256_sub_ps(iz0,jz2);
 719             dx10             = _mm256_sub_ps(ix1,jx0);
 720             dy10             = _mm256_sub_ps(iy1,jy0);
 721             dz10             = _mm256_sub_ps(iz1,jz0);
 722             dx11             = _mm256_sub_ps(ix1,jx1);
 723             dy11             = _mm256_sub_ps(iy1,jy1);
 724             dz11             = _mm256_sub_ps(iz1,jz1);
 725             dx12             = _mm256_sub_ps(ix1,jx2);
 726             dy12             = _mm256_sub_ps(iy1,jy2);
 727             dz12             = _mm256_sub_ps(iz1,jz2);
 728             dx20             = _mm256_sub_ps(ix2,jx0);
 729             dy20             = _mm256_sub_ps(iy2,jy0);
 730             dz20             = _mm256_sub_ps(iz2,jz0);
 731             dx21             = _mm256_sub_ps(ix2,jx1);
 732             dy21             = _mm256_sub_ps(iy2,jy1);
 733             dz21             = _mm256_sub_ps(iz2,jz1);
 734             dx22             = _mm256_sub_ps(ix2,jx2);
 735             dy22             = _mm256_sub_ps(iy2,jy2);
 736             dz22             = _mm256_sub_ps(iz2,jz2);
 737
 738             /* Calculate squared distance and things based on it */
 739             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
 740             rsq01            = gmx_mm256_calc_rsq_ps(dx01,dy01,dz01);
 741             rsq02            = gmx_mm256_calc_rsq_ps(dx02,dy02,dz02);
 742             rsq10            = gmx_mm256_calc_rsq_ps(dx10,dy10,dz10);
 743             rsq11            = gmx_mm256_calc_rsq_ps(dx11,dy11,dz11);
 744             rsq12            = gmx_mm256_calc_rsq_ps(dx12,dy12,dz12);
 745             rsq20            = gmx_mm256_calc_rsq_ps(dx20,dy20,dz20);
 746             rsq21            = gmx_mm256_calc_rsq_ps(dx21,dy21,dz21);
 747             rsq22            = gmx_mm256_calc_rsq_ps(dx22,dy22,dz22);
 748
 749             rinv00           = gmx_mm256_invsqrt_ps(rsq00);
 750             rinv01           = gmx_mm256_invsqrt_ps(rsq01);
 751             rinv02           = gmx_mm256_invsqrt_ps(rsq02);
 752             rinv10           = gmx_mm256_invsqrt_ps(rsq10);
 753             rinv11           = gmx_mm256_invsqrt_ps(rsq11);
 754             rinv12           = gmx_mm256_invsqrt_ps(rsq12);
 755             rinv20           = gmx_mm256_invsqrt_ps(rsq20);
 756             rinv21           = gmx_mm256_invsqrt_ps(rsq21);
 757             rinv22           = gmx_mm256_invsqrt_ps(rsq22);
 758
 759             rinvsq00         = _mm256_mul_ps(rinv00,rinv00);
 760             rinvsq01         = _mm256_mul_ps(rinv01,rinv01);
 761             rinvsq02         = _mm256_mul_ps(rinv02,rinv02);
 762             rinvsq10         = _mm256_mul_ps(rinv10,rinv10);
 763             rinvsq11         = _mm256_mul_ps(rinv11,rinv11);
 764             rinvsq12         = _mm256_mul_ps(rinv12,rinv12);
 765             rinvsq20         = _mm256_mul_ps(rinv20,rinv20);
 766             rinvsq21         = _mm256_mul_ps(rinv21,rinv21);
 767             rinvsq22         = _mm256_mul_ps(rinv22,rinv22);
 768
 769             fjx0             = _mm256_setzero_ps();
 770             fjy0             = _mm256_setzero_ps();
 771             fjz0             = _mm256_setzero_ps();
 772             fjx1             = _mm256_setzero_ps();
 773             fjy1             = _mm256_setzero_ps();
 774             fjz1             = _mm256_setzero_ps();
 775             fjx2             = _mm256_setzero_ps();
 776             fjy2             = _mm256_setzero_ps();
 777             fjz2             = _mm256_setzero_ps();
 778
 779             /**************************
 780              * CALCULATE INTERACTIONS *
 781              **************************/
 782
 783             r00              = _mm256_mul_ps(rsq00,rinv00);
 784             r00              = _mm256_andnot_ps(dummy_mask,r00);
 785
 786             /* EWALD ELECTROSTATICS */
 787
 788             /* Analytical PME correction */
 789             zeta2            = _mm256_mul_ps(beta2,rsq00);
 790             rinv3            = _mm256_mul_ps(rinvsq00,rinv00);
 791             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
 792             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
 793             felec            = _mm256_mul_ps(qq00,felec);
 794             pmecorrV         = gmx_mm256_pmecorrV_ps(zeta2);
 795             pmecorrV         = _mm256_mul_ps(pmecorrV,beta);
 796             velec            = _mm256_sub_ps(rinv00,pmecorrV);
 797             velec            = _mm256_mul_ps(qq00,velec);
 798
 799             /* Update potential sum for this i atom from the interaction with this j atom. */
 800             velec            = _mm256_andnot_ps(dummy_mask,velec);
 801             velecsum         = _mm256_add_ps(velecsum,velec);
 802
 803             fscal            = felec;
 804
 805             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
 806
 807             /* Calculate temporary vectorial force */
 808             tx               = _mm256_mul_ps(fscal,dx00);
 809             ty               = _mm256_mul_ps(fscal,dy00);
 810             tz               = _mm256_mul_ps(fscal,dz00);
 811
 812             /* Update vectorial force */
 813             fix0             = _mm256_add_ps(fix0,tx);
 814             fiy0             = _mm256_add_ps(fiy0,ty);
 815             fiz0             = _mm256_add_ps(fiz0,tz);
 816
 817             fjx0             = _mm256_add_ps(fjx0,tx);
 818             fjy0             = _mm256_add_ps(fjy0,ty);
 819             fjz0             = _mm256_add_ps(fjz0,tz);
 820
 821             /**************************
 822              * CALCULATE INTERACTIONS *
 823              **************************/
 824
 825             r01              = _mm256_mul_ps(rsq01,rinv01);
 826             r01              = _mm256_andnot_ps(dummy_mask,r01);
 827
 828             /* EWALD ELECTROSTATICS */
 829
 830             /* Analytical PME correction */
 831             zeta2            = _mm256_mul_ps(beta2,rsq01);
 832             rinv3            = _mm256_mul_ps(rinvsq01,rinv01);
 833             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
 834             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
 835             felec            = _mm256_mul_ps(qq01,felec);
 836             pmecorrV         = gmx_mm256_pmecorrV_ps(zeta2);
 837             pmecorrV         = _mm256_mul_ps(pmecorrV,beta);
 838             velec            = _mm256_sub_ps(rinv01,pmecorrV);
 839             velec            = _mm256_mul_ps(qq01,velec);
 840
 841             /* Update potential sum for this i atom from the interaction with this j atom. */
 842             velec            = _mm256_andnot_ps(dummy_mask,velec);
 843             velecsum         = _mm256_add_ps(velecsum,velec);
 844
 845             fscal            = felec;
 846
 847             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
 848
 849             /* Calculate temporary vectorial force */
 850             tx               = _mm256_mul_ps(fscal,dx01);
 851             ty               = _mm256_mul_ps(fscal,dy01);
 852             tz               = _mm256_mul_ps(fscal,dz01);
 853
 854             /* Update vectorial force */
 855             fix0             = _mm256_add_ps(fix0,tx);
 856             fiy0             = _mm256_add_ps(fiy0,ty);
 857             fiz0             = _mm256_add_ps(fiz0,tz);
 858
 859             fjx1             = _mm256_add_ps(fjx1,tx);
 860             fjy1             = _mm256_add_ps(fjy1,ty);
 861             fjz1             = _mm256_add_ps(fjz1,tz);
 862
 863             /**************************
 864              * CALCULATE INTERACTIONS *
 865              **************************/
 866
 867             r02              = _mm256_mul_ps(rsq02,rinv02);
 868             r02              = _mm256_andnot_ps(dummy_mask,r02);
 869
 870             /* EWALD ELECTROSTATICS */
 871
 872             /* Analytical PME correction */
 873             zeta2            = _mm256_mul_ps(beta2,rsq02);
 874             rinv3            = _mm256_mul_ps(rinvsq02,rinv02);
 875             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
 876             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
 877             felec            = _mm256_mul_ps(qq02,felec);
 878             pmecorrV         = gmx_mm256_pmecorrV_ps(zeta2);
 879             pmecorrV         = _mm256_mul_ps(pmecorrV,beta);
 880             velec            = _mm256_sub_ps(rinv02,pmecorrV);
 881             velec            = _mm256_mul_ps(qq02,velec);
 882
 883             /* Update potential sum for this i atom from the interaction with this j atom. */
 884             velec            = _mm256_andnot_ps(dummy_mask,velec);
 885             velecsum         = _mm256_add_ps(velecsum,velec);
 886
 887             fscal            = felec;
 888
 889             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
 890
 891             /* Calculate temporary vectorial force */
 892             tx               = _mm256_mul_ps(fscal,dx02);
 893             ty               = _mm256_mul_ps(fscal,dy02);
 894             tz               = _mm256_mul_ps(fscal,dz02);
 895
 896             /* Update vectorial force */
 897             fix0             = _mm256_add_ps(fix0,tx);
 898             fiy0             = _mm256_add_ps(fiy0,ty);
 899             fiz0             = _mm256_add_ps(fiz0,tz);
 900
 901             fjx2             = _mm256_add_ps(fjx2,tx);
 902             fjy2             = _mm256_add_ps(fjy2,ty);
 903             fjz2             = _mm256_add_ps(fjz2,tz);
 904
 905             /**************************
 906              * CALCULATE INTERACTIONS *
 907              **************************/
 908
 909             r10              = _mm256_mul_ps(rsq10,rinv10);
 910             r10              = _mm256_andnot_ps(dummy_mask,r10);
 911
 912             /* EWALD ELECTROSTATICS */
 913
 914             /* Analytical PME correction */
 915             zeta2            = _mm256_mul_ps(beta2,rsq10);
 916             rinv3            = _mm256_mul_ps(rinvsq10,rinv10);
 917             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
 918             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
 919             felec            = _mm256_mul_ps(qq10,felec);
 920             pmecorrV         = gmx_mm256_pmecorrV_ps(zeta2);
 921             pmecorrV         = _mm256_mul_ps(pmecorrV,beta);
 922             velec            = _mm256_sub_ps(rinv10,pmecorrV);
 923             velec            = _mm256_mul_ps(qq10,velec);
 924
 925             /* Update potential sum for this i atom from the interaction with this j atom. */
 926             velec            = _mm256_andnot_ps(dummy_mask,velec);
 927             velecsum         = _mm256_add_ps(velecsum,velec);
 928
 929             fscal            = felec;
 930
 931             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
 932
 933             /* Calculate temporary vectorial force */
 934             tx               = _mm256_mul_ps(fscal,dx10);
 935             ty               = _mm256_mul_ps(fscal,dy10);
 936             tz               = _mm256_mul_ps(fscal,dz10);
 937
 938             /* Update vectorial force */
 939             fix1             = _mm256_add_ps(fix1,tx);
 940             fiy1             = _mm256_add_ps(fiy1,ty);
 941             fiz1             = _mm256_add_ps(fiz1,tz);
 942
 943             fjx0             = _mm256_add_ps(fjx0,tx);
 944             fjy0             = _mm256_add_ps(fjy0,ty);
 945             fjz0             = _mm256_add_ps(fjz0,tz);
 946
 947             /**************************
 948              * CALCULATE INTERACTIONS *
 949              **************************/
 950
 951             r11              = _mm256_mul_ps(rsq11,rinv11);
 952             r11              = _mm256_andnot_ps(dummy_mask,r11);
 953
 954             /* EWALD ELECTROSTATICS */
 955
 956             /* Analytical PME correction */
 957             zeta2            = _mm256_mul_ps(beta2,rsq11);
 958             rinv3            = _mm256_mul_ps(rinvsq11,rinv11);
 959             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
 960             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
 961             felec            = _mm256_mul_ps(qq11,felec);
 962             pmecorrV         = gmx_mm256_pmecorrV_ps(zeta2);
 963             pmecorrV         = _mm256_mul_ps(pmecorrV,beta);
 964             velec            = _mm256_sub_ps(rinv11,pmecorrV);
 965             velec            = _mm256_mul_ps(qq11,velec);
 966
 967             /* Update potential sum for this i atom from the interaction with this j atom. */
 968             velec            = _mm256_andnot_ps(dummy_mask,velec);
 969             velecsum         = _mm256_add_ps(velecsum,velec);
 970
 971             fscal            = felec;
 972
 973             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
 974
 975             /* Calculate temporary vectorial force */
 976             tx               = _mm256_mul_ps(fscal,dx11);
 977             ty               = _mm256_mul_ps(fscal,dy11);
 978             tz               = _mm256_mul_ps(fscal,dz11);
 979
 980             /* Update vectorial force */
 981             fix1             = _mm256_add_ps(fix1,tx);
 982             fiy1             = _mm256_add_ps(fiy1,ty);
 983             fiz1             = _mm256_add_ps(fiz1,tz);
 984
 985             fjx1             = _mm256_add_ps(fjx1,tx);
 986             fjy1             = _mm256_add_ps(fjy1,ty);
 987             fjz1             = _mm256_add_ps(fjz1,tz);
 988
 989             /**************************
 990              * CALCULATE INTERACTIONS *
 991              **************************/
 992
 993             r12              = _mm256_mul_ps(rsq12,rinv12);
 994             r12              = _mm256_andnot_ps(dummy_mask,r12);
 995
 996             /* EWALD ELECTROSTATICS */
 997
 998             /* Analytical PME correction */
 999             zeta2            = _mm256_mul_ps(beta2,rsq12);
1000             rinv3            = _mm256_mul_ps(rinvsq12,rinv12);
1001             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
1002             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
1003             felec            = _mm256_mul_ps(qq12,felec);
1004             pmecorrV         = gmx_mm256_pmecorrV_ps(zeta2);
1005             pmecorrV         = _mm256_mul_ps(pmecorrV,beta);
1006             velec            = _mm256_sub_ps(rinv12,pmecorrV);
1007             velec            = _mm256_mul_ps(qq12,velec);
1008
1009             /* Update potential sum for this i atom from the interaction with this j atom. */
1010             velec            = _mm256_andnot_ps(dummy_mask,velec);
1011             velecsum         = _mm256_add_ps(velecsum,velec);
1012
1013             fscal            = felec;
1014
1015             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
1016
1017             /* Calculate temporary vectorial force */
1018             tx               = _mm256_mul_ps(fscal,dx12);
1019             ty               = _mm256_mul_ps(fscal,dy12);
1020             tz               = _mm256_mul_ps(fscal,dz12);
1021
1022             /* Update vectorial force */
1023             fix1             = _mm256_add_ps(fix1,tx);
1024             fiy1             = _mm256_add_ps(fiy1,ty);
1025             fiz1             = _mm256_add_ps(fiz1,tz);
1026
1027             fjx2             = _mm256_add_ps(fjx2,tx);
1028             fjy2             = _mm256_add_ps(fjy2,ty);
1029             fjz2             = _mm256_add_ps(fjz2,tz);
1030
1031             /**************************
1032              * CALCULATE INTERACTIONS *
1033              **************************/
1034
1035             r20              = _mm256_mul_ps(rsq20,rinv20);
1036             r20              = _mm256_andnot_ps(dummy_mask,r20);
1037
1038             /* EWALD ELECTROSTATICS */
1039
1040             /* Analytical PME correction */
1041             zeta2            = _mm256_mul_ps(beta2,rsq20);
1042             rinv3            = _mm256_mul_ps(rinvsq20,rinv20);
1043             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
1044             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
1045             felec            = _mm256_mul_ps(qq20,felec);
1046             pmecorrV         = gmx_mm256_pmecorrV_ps(zeta2);
1047             pmecorrV         = _mm256_mul_ps(pmecorrV,beta);
1048             velec            = _mm256_sub_ps(rinv20,pmecorrV);
1049             velec            = _mm256_mul_ps(qq20,velec);
1050
1051             /* Update potential sum for this i atom from the interaction with this j atom. */
1052             velec            = _mm256_andnot_ps(dummy_mask,velec);
1053             velecsum         = _mm256_add_ps(velecsum,velec);
1054
1055             fscal            = felec;
1056
1057             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
1058
1059             /* Calculate temporary vectorial force */
1060             tx               = _mm256_mul_ps(fscal,dx20);
1061             ty               = _mm256_mul_ps(fscal,dy20);
1062             tz               = _mm256_mul_ps(fscal,dz20);
1063
1064             /* Update vectorial force */
1065             fix2             = _mm256_add_ps(fix2,tx);
1066             fiy2             = _mm256_add_ps(fiy2,ty);
1067             fiz2             = _mm256_add_ps(fiz2,tz);
1068
1069             fjx0             = _mm256_add_ps(fjx0,tx);
1070             fjy0             = _mm256_add_ps(fjy0,ty);
1071             fjz0             = _mm256_add_ps(fjz0,tz);
1072
1073             /**************************
1074              * CALCULATE INTERACTIONS *
1075              **************************/
1076
1077             r21              = _mm256_mul_ps(rsq21,rinv21);
1078             r21              = _mm256_andnot_ps(dummy_mask,r21);
1079
1080             /* EWALD ELECTROSTATICS */
1081
1082             /* Analytical PME correction */
1083             zeta2            = _mm256_mul_ps(beta2,rsq21);
1084             rinv3            = _mm256_mul_ps(rinvsq21,rinv21);
1085             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
1086             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
1087             felec            = _mm256_mul_ps(qq21,felec);
1088             pmecorrV         = gmx_mm256_pmecorrV_ps(zeta2);
1089             pmecorrV         = _mm256_mul_ps(pmecorrV,beta);
1090             velec            = _mm256_sub_ps(rinv21,pmecorrV);
1091             velec            = _mm256_mul_ps(qq21,velec);
1092
1093             /* Update potential sum for this i atom from the interaction with this j atom. */
1094             velec            = _mm256_andnot_ps(dummy_mask,velec);
1095             velecsum         = _mm256_add_ps(velecsum,velec);
1096
1097             fscal            = felec;
1098
1099             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
1100
1101             /* Calculate temporary vectorial force */
1102             tx               = _mm256_mul_ps(fscal,dx21);
1103             ty               = _mm256_mul_ps(fscal,dy21);
1104             tz               = _mm256_mul_ps(fscal,dz21);
1105
1106             /* Update vectorial force */
1107             fix2             = _mm256_add_ps(fix2,tx);
1108             fiy2             = _mm256_add_ps(fiy2,ty);
1109             fiz2             = _mm256_add_ps(fiz2,tz);
1110
1111             fjx1             = _mm256_add_ps(fjx1,tx);
1112             fjy1             = _mm256_add_ps(fjy1,ty);
1113             fjz1             = _mm256_add_ps(fjz1,tz);
1114
1115             /**************************
1116              * CALCULATE INTERACTIONS *
1117              **************************/
1118
1119             r22              = _mm256_mul_ps(rsq22,rinv22);
1120             r22              = _mm256_andnot_ps(dummy_mask,r22);
1121
1122             /* EWALD ELECTROSTATICS */
1123
1124             /* Analytical PME correction */
1125             zeta2            = _mm256_mul_ps(beta2,rsq22);
1126             rinv3            = _mm256_mul_ps(rinvsq22,rinv22);
1127             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
1128             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
1129             felec            = _mm256_mul_ps(qq22,felec);
1130             pmecorrV         = gmx_mm256_pmecorrV_ps(zeta2);
1131             pmecorrV         = _mm256_mul_ps(pmecorrV,beta);
1132             velec            = _mm256_sub_ps(rinv22,pmecorrV);
1133             velec            = _mm256_mul_ps(qq22,velec);
1134
1135             /* Update potential sum for this i atom from the interaction with this j atom. */
1136             velec            = _mm256_andnot_ps(dummy_mask,velec);
1137             velecsum         = _mm256_add_ps(velecsum,velec);
1138
1139             fscal            = felec;
1140
1141             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
1142
1143             /* Calculate temporary vectorial force */
1144             tx               = _mm256_mul_ps(fscal,dx22);
1145             ty               = _mm256_mul_ps(fscal,dy22);
1146             tz               = _mm256_mul_ps(fscal,dz22);
1147
1148             /* Update vectorial force */
1149             fix2             = _mm256_add_ps(fix2,tx);
1150             fiy2             = _mm256_add_ps(fiy2,ty);
1151             fiz2             = _mm256_add_ps(fiz2,tz);
1152
1153             fjx2             = _mm256_add_ps(fjx2,tx);
1154             fjy2             = _mm256_add_ps(fjy2,ty);
1155             fjz2             = _mm256_add_ps(fjz2,tz);
1156
1157             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
1158             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
1159             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
1160             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
1161             fjptrE             = (jnrlistE>=0) ? f+j_coord_offsetE : scratch;
1162             fjptrF             = (jnrlistF>=0) ? f+j_coord_offsetF : scratch;
1163             fjptrG             = (jnrlistG>=0) ? f+j_coord_offsetG : scratch;
1164             fjptrH             = (jnrlistH>=0) ? f+j_coord_offsetH : scratch;
1165
1166             gmx_mm256_decrement_3rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,
1167                                                       fjx0,fjy0,fjz0,fjx1,fjy1,fjz1,fjx2,fjy2,fjz2);
1168
1169             /* Inner loop uses 765 flops */
1170         }
1171
1172         /* End of innermost loop */
1173
1174         gmx_mm256_update_iforce_3atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
1175                                                  f+i_coord_offset,fshift+i_shift_offset);
1176
1177         ggid                        = gid[iidx];
1178         /* Update potential energies */
1179         gmx_mm256_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid);
1180
1181         /* Increment number of inner iterations */
1182         inneriter                  += j_index_end - j_index_start;
1183
1184         /* Outer loop uses 19 flops */
1185     }
1186
1187     /* Increment number of outer iterations */
1188     outeriter        += nri;
1189
1190     /* Update outer/inner flops */
1191
1192     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_W3W3_VF,outeriter*19 + inneriter*765);
1193 }
1194 /*
1195  * Gromacs nonbonded kernel:   nb_kernel_ElecEw_VdwNone_GeomW3W3_F_avx_256_single
1196  * Electrostatics interaction: Ewald
1197  * VdW interaction:            None
1198  * Geometry:                   Water3-Water3
1199  * Calculate force/pot:        Force
1200  */
1201 void
1202 nb_kernel_ElecEw_VdwNone_GeomW3W3_F_avx_256_single
1203                     (t_nblist                    * gmx_restrict       nlist,
1204                      rvec                        * gmx_restrict          xx,
1205                      rvec                        * gmx_restrict          ff,
1206                      t_forcerec                  * gmx_restrict          fr,
1207                      t_mdatoms                   * gmx_restrict     mdatoms,
1208                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
1209                      t_nrnb                      * gmx_restrict        nrnb)
1210 {
1211     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
1212      * just 0 for non-waters.
1213      * Suffixes A,B,C,D,E,F,G,H refer to j loop unrolling done with AVX, e.g. for the eight different
1214      * jnr indices corresponding to data put in the four positions in the SIMD register.
1215      */
1216     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
1217     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
1218     int              jnrA,jnrB,jnrC,jnrD;
1219     int              jnrE,jnrF,jnrG,jnrH;
1220     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
1221     int              jnrlistE,jnrlistF,jnrlistG,jnrlistH;
1222     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
1223     int              j_coord_offsetE,j_coord_offsetF,j_coord_offsetG,j_coord_offsetH;
1224     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
1225     real             rcutoff_scalar;
1226     real             *shiftvec,*fshift,*x,*f;
1227     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD,*fjptrE,*fjptrF,*fjptrG,*fjptrH;
1228     real             scratch[4*DIM];
1229     __m256           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
1230     real *           vdwioffsetptr0;
1231     __m256           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
1232     real *           vdwioffsetptr1;
1233     __m256           ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
1234     real *           vdwioffsetptr2;
1235     __m256           ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
1236     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D,vdwjidx0E,vdwjidx0F,vdwjidx0G,vdwjidx0H;
1237     __m256           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
1238     int              vdwjidx1A,vdwjidx1B,vdwjidx1C,vdwjidx1D,vdwjidx1E,vdwjidx1F,vdwjidx1G,vdwjidx1H;
1239     __m256           jx1,jy1,jz1,fjx1,fjy1,fjz1,jq1,isaj1;
1240     int              vdwjidx2A,vdwjidx2B,vdwjidx2C,vdwjidx2D,vdwjidx2E,vdwjidx2F,vdwjidx2G,vdwjidx2H;
1241     __m256           jx2,jy2,jz2,fjx2,fjy2,fjz2,jq2,isaj2;
1242     __m256           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
1243     __m256           dx01,dy01,dz01,rsq01,rinv01,rinvsq01,r01,qq01,c6_01,c12_01;
1244     __m256           dx02,dy02,dz02,rsq02,rinv02,rinvsq02,r02,qq02,c6_02,c12_02;
1245     __m256           dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
1246     __m256           dx11,dy11,dz11,rsq11,rinv11,rinvsq11,r11,qq11,c6_11,c12_11;
1247     __m256           dx12,dy12,dz12,rsq12,rinv12,rinvsq12,r12,qq12,c6_12,c12_12;
1248     __m256           dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
1249     __m256           dx21,dy21,dz21,rsq21,rinv21,rinvsq21,r21,qq21,c6_21,c12_21;
1250     __m256           dx22,dy22,dz22,rsq22,rinv22,rinvsq22,r22,qq22,c6_22,c12_22;
1251     __m256           velec,felec,velecsum,facel,crf,krf,krf2;
1252     real             *charge;
1253     __m256i          ewitab;
1254     __m128i          ewitab_lo,ewitab_hi;
1255     __m256           ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
1256     __m256           beta,beta2,beta3,zeta2,pmecorrF,pmecorrV,rinv3;
1257     real             *ewtab;
1258     __m256           dummy_mask,cutoff_mask;
1259     __m256           signbit = _mm256_castsi256_ps( _mm256_set1_epi32(0x80000000) );
1260     __m256           one     = _mm256_set1_ps(1.0);
1261     __m256           two     = _mm256_set1_ps(2.0);
1262     x                = xx[0];
1263     f                = ff[0];
1264
1265     nri              = nlist->nri;
1266     iinr             = nlist->iinr;
1267     jindex           = nlist->jindex;
1268     jjnr             = nlist->jjnr;
1269     shiftidx         = nlist->shift;
1270     gid              = nlist->gid;
1271     shiftvec         = fr->shift_vec[0];
1272     fshift           = fr->fshift[0];
1273     facel            = _mm256_set1_ps(fr->epsfac);
1274     charge           = mdatoms->chargeA;
1275
1276     sh_ewald         = _mm256_set1_ps(fr->ic->sh_ewald);
1277     beta             = _mm256_set1_ps(fr->ic->ewaldcoeff_q);
1278     beta2            = _mm256_mul_ps(beta,beta);
1279     beta3            = _mm256_mul_ps(beta,beta2);
1280
1281     ewtab            = fr->ic->tabq_coul_F;
1282     ewtabscale       = _mm256_set1_ps(fr->ic->tabq_scale);
1283     ewtabhalfspace   = _mm256_set1_ps(0.5/fr->ic->tabq_scale);
1284
1285     /* Setup water-specific parameters */
1286     inr              = nlist->iinr[0];
1287     iq0              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+0]));
1288     iq1              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+1]));
1289     iq2              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+2]));
1290
1291     jq0              = _mm256_set1_ps(charge[inr+0]);
1292     jq1              = _mm256_set1_ps(charge[inr+1]);
1293     jq2              = _mm256_set1_ps(charge[inr+2]);
1294     qq00             = _mm256_mul_ps(iq0,jq0);
1295     qq01             = _mm256_mul_ps(iq0,jq1);
1296     qq02             = _mm256_mul_ps(iq0,jq2);
1297     qq10             = _mm256_mul_ps(iq1,jq0);
1298     qq11             = _mm256_mul_ps(iq1,jq1);
1299     qq12             = _mm256_mul_ps(iq1,jq2);
1300     qq20             = _mm256_mul_ps(iq2,jq0);
1301     qq21             = _mm256_mul_ps(iq2,jq1);
1302     qq22             = _mm256_mul_ps(iq2,jq2);
1303
1304     /* Avoid stupid compiler warnings */
1305     jnrA = jnrB = jnrC = jnrD = jnrE = jnrF = jnrG = jnrH = 0;
1306     j_coord_offsetA = 0;
1307     j_coord_offsetB = 0;
1308     j_coord_offsetC = 0;
1309     j_coord_offsetD = 0;
1310     j_coord_offsetE = 0;
1311     j_coord_offsetF = 0;
1312     j_coord_offsetG = 0;
1313     j_coord_offsetH = 0;
1314
1315     outeriter        = 0;
1316     inneriter        = 0;
1317
1318     for(iidx=0;iidx<4*DIM;iidx++)
1319     {
1320         scratch[iidx] = 0.0;
1321     }
1322
1323     /* Start outer loop over neighborlists */
1324     for(iidx=0; iidx<nri; iidx++)
1325     {
1326         /* Load shift vector for this list */
1327         i_shift_offset   = DIM*shiftidx[iidx];
1328
1329         /* Load limits for loop over neighbors */
1330         j_index_start    = jindex[iidx];
1331         j_index_end      = jindex[iidx+1];
1332
1333         /* Get outer coordinate index */
1334         inr              = iinr[iidx];
1335         i_coord_offset   = DIM*inr;
1336
1337         /* Load i particle coords and add shift vector */
1338         gmx_mm256_load_shift_and_3rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
1339                                                     &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
1340
1341         fix0             = _mm256_setzero_ps();
1342         fiy0             = _mm256_setzero_ps();
1343         fiz0             = _mm256_setzero_ps();
1344         fix1             = _mm256_setzero_ps();
1345         fiy1             = _mm256_setzero_ps();
1346         fiz1             = _mm256_setzero_ps();
1347         fix2             = _mm256_setzero_ps();
1348         fiy2             = _mm256_setzero_ps();
1349         fiz2             = _mm256_setzero_ps();
1350
1351         /* Start inner kernel loop */
1352         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+7]>=0; jidx+=8)
1353         {
1354
1355             /* Get j neighbor index, and coordinate index */
1356             jnrA             = jjnr[jidx];
1357             jnrB             = jjnr[jidx+1];
1358             jnrC             = jjnr[jidx+2];
1359             jnrD             = jjnr[jidx+3];
1360             jnrE             = jjnr[jidx+4];
1361             jnrF             = jjnr[jidx+5];
1362             jnrG             = jjnr[jidx+6];
1363             jnrH             = jjnr[jidx+7];
1364             j_coord_offsetA  = DIM*jnrA;
1365             j_coord_offsetB  = DIM*jnrB;
1366             j_coord_offsetC  = DIM*jnrC;
1367             j_coord_offsetD  = DIM*jnrD;
1368             j_coord_offsetE  = DIM*jnrE;
1369             j_coord_offsetF  = DIM*jnrF;
1370             j_coord_offsetG  = DIM*jnrG;
1371             j_coord_offsetH  = DIM*jnrH;
1372
1373             /* load j atom coordinates */
1374             gmx_mm256_load_3rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
1375                                                  x+j_coord_offsetC,x+j_coord_offsetD,
1376                                                  x+j_coord_offsetE,x+j_coord_offsetF,
1377                                                  x+j_coord_offsetG,x+j_coord_offsetH,
1378                                               &jx0,&jy0,&jz0,&jx1,&jy1,&jz1,&jx2,&jy2,&jz2);
1379
1380             /* Calculate displacement vector */
1381             dx00             = _mm256_sub_ps(ix0,jx0);
1382             dy00             = _mm256_sub_ps(iy0,jy0);
1383             dz00             = _mm256_sub_ps(iz0,jz0);
1384             dx01             = _mm256_sub_ps(ix0,jx1);
1385             dy01             = _mm256_sub_ps(iy0,jy1);
1386             dz01             = _mm256_sub_ps(iz0,jz1);
1387             dx02             = _mm256_sub_ps(ix0,jx2);
1388             dy02             = _mm256_sub_ps(iy0,jy2);
1389             dz02             = _mm256_sub_ps(iz0,jz2);
1390             dx10             = _mm256_sub_ps(ix1,jx0);
1391             dy10             = _mm256_sub_ps(iy1,jy0);
1392             dz10             = _mm256_sub_ps(iz1,jz0);
1393             dx11             = _mm256_sub_ps(ix1,jx1);
1394             dy11             = _mm256_sub_ps(iy1,jy1);
1395             dz11             = _mm256_sub_ps(iz1,jz1);
1396             dx12             = _mm256_sub_ps(ix1,jx2);
1397             dy12             = _mm256_sub_ps(iy1,jy2);
1398             dz12             = _mm256_sub_ps(iz1,jz2);
1399             dx20             = _mm256_sub_ps(ix2,jx0);
1400             dy20             = _mm256_sub_ps(iy2,jy0);
1401             dz20             = _mm256_sub_ps(iz2,jz0);
1402             dx21             = _mm256_sub_ps(ix2,jx1);
1403             dy21             = _mm256_sub_ps(iy2,jy1);
1404             dz21             = _mm256_sub_ps(iz2,jz1);
1405             dx22             = _mm256_sub_ps(ix2,jx2);
1406             dy22             = _mm256_sub_ps(iy2,jy2);
1407             dz22             = _mm256_sub_ps(iz2,jz2);
1408
1409             /* Calculate squared distance and things based on it */
1410             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
1411             rsq01            = gmx_mm256_calc_rsq_ps(dx01,dy01,dz01);
1412             rsq02            = gmx_mm256_calc_rsq_ps(dx02,dy02,dz02);
1413             rsq10            = gmx_mm256_calc_rsq_ps(dx10,dy10,dz10);
1414             rsq11            = gmx_mm256_calc_rsq_ps(dx11,dy11,dz11);
1415             rsq12            = gmx_mm256_calc_rsq_ps(dx12,dy12,dz12);
1416             rsq20            = gmx_mm256_calc_rsq_ps(dx20,dy20,dz20);
1417             rsq21            = gmx_mm256_calc_rsq_ps(dx21,dy21,dz21);
1418             rsq22            = gmx_mm256_calc_rsq_ps(dx22,dy22,dz22);
1419
1420             rinv00           = gmx_mm256_invsqrt_ps(rsq00);
1421             rinv01           = gmx_mm256_invsqrt_ps(rsq01);
1422             rinv02           = gmx_mm256_invsqrt_ps(rsq02);
1423             rinv10           = gmx_mm256_invsqrt_ps(rsq10);
1424             rinv11           = gmx_mm256_invsqrt_ps(rsq11);
1425             rinv12           = gmx_mm256_invsqrt_ps(rsq12);
1426             rinv20           = gmx_mm256_invsqrt_ps(rsq20);
1427             rinv21           = gmx_mm256_invsqrt_ps(rsq21);
1428             rinv22           = gmx_mm256_invsqrt_ps(rsq22);
1429
1430             rinvsq00         = _mm256_mul_ps(rinv00,rinv00);
1431             rinvsq01         = _mm256_mul_ps(rinv01,rinv01);
1432             rinvsq02         = _mm256_mul_ps(rinv02,rinv02);
1433             rinvsq10         = _mm256_mul_ps(rinv10,rinv10);
1434             rinvsq11         = _mm256_mul_ps(rinv11,rinv11);
1435             rinvsq12         = _mm256_mul_ps(rinv12,rinv12);
1436             rinvsq20         = _mm256_mul_ps(rinv20,rinv20);
1437             rinvsq21         = _mm256_mul_ps(rinv21,rinv21);
1438             rinvsq22         = _mm256_mul_ps(rinv22,rinv22);
1439
1440             fjx0             = _mm256_setzero_ps();
1441             fjy0             = _mm256_setzero_ps();
1442             fjz0             = _mm256_setzero_ps();
1443             fjx1             = _mm256_setzero_ps();
1444             fjy1             = _mm256_setzero_ps();
1445             fjz1             = _mm256_setzero_ps();
1446             fjx2             = _mm256_setzero_ps();
1447             fjy2             = _mm256_setzero_ps();
1448             fjz2             = _mm256_setzero_ps();
1449
1450             /**************************
1451              * CALCULATE INTERACTIONS *
1452              **************************/
1453
1454             r00              = _mm256_mul_ps(rsq00,rinv00);
1455
1456             /* EWALD ELECTROSTATICS */
1457
1458             /* Analytical PME correction */
1459             zeta2            = _mm256_mul_ps(beta2,rsq00);
1460             rinv3            = _mm256_mul_ps(rinvsq00,rinv00);
1461             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
1462             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
1463             felec            = _mm256_mul_ps(qq00,felec);
1464
1465             fscal            = felec;
1466
1467             /* Calculate temporary vectorial force */
1468             tx               = _mm256_mul_ps(fscal,dx00);
1469             ty               = _mm256_mul_ps(fscal,dy00);
1470             tz               = _mm256_mul_ps(fscal,dz00);
1471
1472             /* Update vectorial force */
1473             fix0             = _mm256_add_ps(fix0,tx);
1474             fiy0             = _mm256_add_ps(fiy0,ty);
1475             fiz0             = _mm256_add_ps(fiz0,tz);
1476
1477             fjx0             = _mm256_add_ps(fjx0,tx);
1478             fjy0             = _mm256_add_ps(fjy0,ty);
1479             fjz0             = _mm256_add_ps(fjz0,tz);
1480
1481             /**************************
1482              * CALCULATE INTERACTIONS *
1483              **************************/
1484
1485             r01              = _mm256_mul_ps(rsq01,rinv01);
1486
1487             /* EWALD ELECTROSTATICS */
1488
1489             /* Analytical PME correction */
1490             zeta2            = _mm256_mul_ps(beta2,rsq01);
1491             rinv3            = _mm256_mul_ps(rinvsq01,rinv01);
1492             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
1493             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
1494             felec            = _mm256_mul_ps(qq01,felec);
1495
1496             fscal            = felec;
1497
1498             /* Calculate temporary vectorial force */
1499             tx               = _mm256_mul_ps(fscal,dx01);
1500             ty               = _mm256_mul_ps(fscal,dy01);
1501             tz               = _mm256_mul_ps(fscal,dz01);
1502
1503             /* Update vectorial force */
1504             fix0             = _mm256_add_ps(fix0,tx);
1505             fiy0             = _mm256_add_ps(fiy0,ty);
1506             fiz0             = _mm256_add_ps(fiz0,tz);
1507
1508             fjx1             = _mm256_add_ps(fjx1,tx);
1509             fjy1             = _mm256_add_ps(fjy1,ty);
1510             fjz1             = _mm256_add_ps(fjz1,tz);
1511
1512             /**************************
1513              * CALCULATE INTERACTIONS *
1514              **************************/
1515
1516             r02              = _mm256_mul_ps(rsq02,rinv02);
1517
1518             /* EWALD ELECTROSTATICS */
1519
1520             /* Analytical PME correction */
1521             zeta2            = _mm256_mul_ps(beta2,rsq02);
1522             rinv3            = _mm256_mul_ps(rinvsq02,rinv02);
1523             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
1524             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
1525             felec            = _mm256_mul_ps(qq02,felec);
1526
1527             fscal            = felec;
1528
1529             /* Calculate temporary vectorial force */
1530             tx               = _mm256_mul_ps(fscal,dx02);
1531             ty               = _mm256_mul_ps(fscal,dy02);
1532             tz               = _mm256_mul_ps(fscal,dz02);
1533
1534             /* Update vectorial force */
1535             fix0             = _mm256_add_ps(fix0,tx);
1536             fiy0             = _mm256_add_ps(fiy0,ty);
1537             fiz0             = _mm256_add_ps(fiz0,tz);
1538
1539             fjx2             = _mm256_add_ps(fjx2,tx);
1540             fjy2             = _mm256_add_ps(fjy2,ty);
1541             fjz2             = _mm256_add_ps(fjz2,tz);
1542
1543             /**************************
1544              * CALCULATE INTERACTIONS *
1545              **************************/
1546
1547             r10              = _mm256_mul_ps(rsq10,rinv10);
1548
1549             /* EWALD ELECTROSTATICS */
1550
1551             /* Analytical PME correction */
1552             zeta2            = _mm256_mul_ps(beta2,rsq10);
1553             rinv3            = _mm256_mul_ps(rinvsq10,rinv10);
1554             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
1555             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
1556             felec            = _mm256_mul_ps(qq10,felec);
1557
1558             fscal            = felec;
1559
1560             /* Calculate temporary vectorial force */
1561             tx               = _mm256_mul_ps(fscal,dx10);
1562             ty               = _mm256_mul_ps(fscal,dy10);
1563             tz               = _mm256_mul_ps(fscal,dz10);
1564
1565             /* Update vectorial force */
1566             fix1             = _mm256_add_ps(fix1,tx);
1567             fiy1             = _mm256_add_ps(fiy1,ty);
1568             fiz1             = _mm256_add_ps(fiz1,tz);
1569
1570             fjx0             = _mm256_add_ps(fjx0,tx);
1571             fjy0             = _mm256_add_ps(fjy0,ty);
1572             fjz0             = _mm256_add_ps(fjz0,tz);
1573
1574             /**************************
1575              * CALCULATE INTERACTIONS *
1576              **************************/
1577
1578             r11              = _mm256_mul_ps(rsq11,rinv11);
1579
1580             /* EWALD ELECTROSTATICS */
1581
1582             /* Analytical PME correction */
1583             zeta2            = _mm256_mul_ps(beta2,rsq11);
1584             rinv3            = _mm256_mul_ps(rinvsq11,rinv11);
1585             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
1586             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
1587             felec            = _mm256_mul_ps(qq11,felec);
1588
1589             fscal            = felec;
1590
1591             /* Calculate temporary vectorial force */
1592             tx               = _mm256_mul_ps(fscal,dx11);
1593             ty               = _mm256_mul_ps(fscal,dy11);
1594             tz               = _mm256_mul_ps(fscal,dz11);
1595
1596             /* Update vectorial force */
1597             fix1             = _mm256_add_ps(fix1,tx);
1598             fiy1             = _mm256_add_ps(fiy1,ty);
1599             fiz1             = _mm256_add_ps(fiz1,tz);
1600
1601             fjx1             = _mm256_add_ps(fjx1,tx);
1602             fjy1             = _mm256_add_ps(fjy1,ty);
1603             fjz1             = _mm256_add_ps(fjz1,tz);
1604
1605             /**************************
1606              * CALCULATE INTERACTIONS *
1607              **************************/
1608
1609             r12              = _mm256_mul_ps(rsq12,rinv12);
1610
1611             /* EWALD ELECTROSTATICS */
1612
1613             /* Analytical PME correction */
1614             zeta2            = _mm256_mul_ps(beta2,rsq12);
1615             rinv3            = _mm256_mul_ps(rinvsq12,rinv12);
1616             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
1617             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
1618             felec            = _mm256_mul_ps(qq12,felec);
1619
1620             fscal            = felec;
1621
1622             /* Calculate temporary vectorial force */
1623             tx               = _mm256_mul_ps(fscal,dx12);
1624             ty               = _mm256_mul_ps(fscal,dy12);
1625             tz               = _mm256_mul_ps(fscal,dz12);
1626
1627             /* Update vectorial force */
1628             fix1             = _mm256_add_ps(fix1,tx);
1629             fiy1             = _mm256_add_ps(fiy1,ty);
1630             fiz1             = _mm256_add_ps(fiz1,tz);
1631
1632             fjx2             = _mm256_add_ps(fjx2,tx);
1633             fjy2             = _mm256_add_ps(fjy2,ty);
1634             fjz2             = _mm256_add_ps(fjz2,tz);
1635
1636             /**************************
1637              * CALCULATE INTERACTIONS *
1638              **************************/
1639
1640             r20              = _mm256_mul_ps(rsq20,rinv20);
1641
1642             /* EWALD ELECTROSTATICS */
1643
1644             /* Analytical PME correction */
1645             zeta2            = _mm256_mul_ps(beta2,rsq20);
1646             rinv3            = _mm256_mul_ps(rinvsq20,rinv20);
1647             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
1648             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
1649             felec            = _mm256_mul_ps(qq20,felec);
1650
1651             fscal            = felec;
1652
1653             /* Calculate temporary vectorial force */
1654             tx               = _mm256_mul_ps(fscal,dx20);
1655             ty               = _mm256_mul_ps(fscal,dy20);
1656             tz               = _mm256_mul_ps(fscal,dz20);
1657
1658             /* Update vectorial force */
1659             fix2             = _mm256_add_ps(fix2,tx);
1660             fiy2             = _mm256_add_ps(fiy2,ty);
1661             fiz2             = _mm256_add_ps(fiz2,tz);
1662
1663             fjx0             = _mm256_add_ps(fjx0,tx);
1664             fjy0             = _mm256_add_ps(fjy0,ty);
1665             fjz0             = _mm256_add_ps(fjz0,tz);
1666
1667             /**************************
1668              * CALCULATE INTERACTIONS *
1669              **************************/
1670
1671             r21              = _mm256_mul_ps(rsq21,rinv21);
1672
1673             /* EWALD ELECTROSTATICS */
1674
1675             /* Analytical PME correction */
1676             zeta2            = _mm256_mul_ps(beta2,rsq21);
1677             rinv3            = _mm256_mul_ps(rinvsq21,rinv21);
1678             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
1679             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
1680             felec            = _mm256_mul_ps(qq21,felec);
1681
1682             fscal            = felec;
1683
1684             /* Calculate temporary vectorial force */
1685             tx               = _mm256_mul_ps(fscal,dx21);
1686             ty               = _mm256_mul_ps(fscal,dy21);
1687             tz               = _mm256_mul_ps(fscal,dz21);
1688
1689             /* Update vectorial force */
1690             fix2             = _mm256_add_ps(fix2,tx);
1691             fiy2             = _mm256_add_ps(fiy2,ty);
1692             fiz2             = _mm256_add_ps(fiz2,tz);
1693
1694             fjx1             = _mm256_add_ps(fjx1,tx);
1695             fjy1             = _mm256_add_ps(fjy1,ty);
1696             fjz1             = _mm256_add_ps(fjz1,tz);
1697
1698             /**************************
1699              * CALCULATE INTERACTIONS *
1700              **************************/
1701
1702             r22              = _mm256_mul_ps(rsq22,rinv22);
1703
1704             /* EWALD ELECTROSTATICS */
1705
1706             /* Analytical PME correction */
1707             zeta2            = _mm256_mul_ps(beta2,rsq22);
1708             rinv3            = _mm256_mul_ps(rinvsq22,rinv22);
1709             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
1710             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
1711             felec            = _mm256_mul_ps(qq22,felec);
1712
1713             fscal            = felec;
1714
1715             /* Calculate temporary vectorial force */
1716             tx               = _mm256_mul_ps(fscal,dx22);
1717             ty               = _mm256_mul_ps(fscal,dy22);
1718             tz               = _mm256_mul_ps(fscal,dz22);
1719
1720             /* Update vectorial force */
1721             fix2             = _mm256_add_ps(fix2,tx);
1722             fiy2             = _mm256_add_ps(fiy2,ty);
1723             fiz2             = _mm256_add_ps(fiz2,tz);
1724
1725             fjx2             = _mm256_add_ps(fjx2,tx);
1726             fjy2             = _mm256_add_ps(fjy2,ty);
1727             fjz2             = _mm256_add_ps(fjz2,tz);
1728
1729             fjptrA             = f+j_coord_offsetA;
1730             fjptrB             = f+j_coord_offsetB;
1731             fjptrC             = f+j_coord_offsetC;
1732             fjptrD             = f+j_coord_offsetD;
1733             fjptrE             = f+j_coord_offsetE;
1734             fjptrF             = f+j_coord_offsetF;
1735             fjptrG             = f+j_coord_offsetG;
1736             fjptrH             = f+j_coord_offsetH;
1737
1738             gmx_mm256_decrement_3rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,
1739                                                       fjx0,fjy0,fjz0,fjx1,fjy1,fjz1,fjx2,fjy2,fjz2);
1740
1741             /* Inner loop uses 504 flops */
1742         }
1743
1744         if(jidx<j_index_end)
1745         {
1746
1747             /* Get j neighbor index, and coordinate index */
1748             jnrlistA         = jjnr[jidx];
1749             jnrlistB         = jjnr[jidx+1];
1750             jnrlistC         = jjnr[jidx+2];
1751             jnrlistD         = jjnr[jidx+3];
1752             jnrlistE         = jjnr[jidx+4];
1753             jnrlistF         = jjnr[jidx+5];
1754             jnrlistG         = jjnr[jidx+6];
1755             jnrlistH         = jjnr[jidx+7];
1756             /* Sign of each element will be negative for non-real atoms.
1757              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
1758              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
1759              */
1760             dummy_mask = gmx_mm256_set_m128(gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx+4)),_mm_setzero_si128())),
1761                                             gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128())));
1762
1763             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
1764             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
1765             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
1766             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
1767             jnrE       = (jnrlistE>=0) ? jnrlistE : 0;
1768             jnrF       = (jnrlistF>=0) ? jnrlistF : 0;
1769             jnrG       = (jnrlistG>=0) ? jnrlistG : 0;
1770             jnrH       = (jnrlistH>=0) ? jnrlistH : 0;
1771             j_coord_offsetA  = DIM*jnrA;
1772             j_coord_offsetB  = DIM*jnrB;
1773             j_coord_offsetC  = DIM*jnrC;
1774             j_coord_offsetD  = DIM*jnrD;
1775             j_coord_offsetE  = DIM*jnrE;
1776             j_coord_offsetF  = DIM*jnrF;
1777             j_coord_offsetG  = DIM*jnrG;
1778             j_coord_offsetH  = DIM*jnrH;
1779
1780             /* load j atom coordinates */
1781             gmx_mm256_load_3rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
1782                                                  x+j_coord_offsetC,x+j_coord_offsetD,
1783                                                  x+j_coord_offsetE,x+j_coord_offsetF,
1784                                                  x+j_coord_offsetG,x+j_coord_offsetH,
1785                                               &jx0,&jy0,&jz0,&jx1,&jy1,&jz1,&jx2,&jy2,&jz2);
1786
1787             /* Calculate displacement vector */
1788             dx00             = _mm256_sub_ps(ix0,jx0);
1789             dy00             = _mm256_sub_ps(iy0,jy0);
1790             dz00             = _mm256_sub_ps(iz0,jz0);
1791             dx01             = _mm256_sub_ps(ix0,jx1);
1792             dy01             = _mm256_sub_ps(iy0,jy1);
1793             dz01             = _mm256_sub_ps(iz0,jz1);
1794             dx02             = _mm256_sub_ps(ix0,jx2);
1795             dy02             = _mm256_sub_ps(iy0,jy2);
1796             dz02             = _mm256_sub_ps(iz0,jz2);
1797             dx10             = _mm256_sub_ps(ix1,jx0);
1798             dy10             = _mm256_sub_ps(iy1,jy0);
1799             dz10             = _mm256_sub_ps(iz1,jz0);
1800             dx11             = _mm256_sub_ps(ix1,jx1);
1801             dy11             = _mm256_sub_ps(iy1,jy1);
1802             dz11             = _mm256_sub_ps(iz1,jz1);
1803             dx12             = _mm256_sub_ps(ix1,jx2);
1804             dy12             = _mm256_sub_ps(iy1,jy2);
1805             dz12             = _mm256_sub_ps(iz1,jz2);
1806             dx20             = _mm256_sub_ps(ix2,jx0);
1807             dy20             = _mm256_sub_ps(iy2,jy0);
1808             dz20             = _mm256_sub_ps(iz2,jz0);
1809             dx21             = _mm256_sub_ps(ix2,jx1);
1810             dy21             = _mm256_sub_ps(iy2,jy1);
1811             dz21             = _mm256_sub_ps(iz2,jz1);
1812             dx22             = _mm256_sub_ps(ix2,jx2);
1813             dy22             = _mm256_sub_ps(iy2,jy2);
1814             dz22             = _mm256_sub_ps(iz2,jz2);
1815
1816             /* Calculate squared distance and things based on it */
1817             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
1818             rsq01            = gmx_mm256_calc_rsq_ps(dx01,dy01,dz01);
1819             rsq02            = gmx_mm256_calc_rsq_ps(dx02,dy02,dz02);
1820             rsq10            = gmx_mm256_calc_rsq_ps(dx10,dy10,dz10);
1821             rsq11            = gmx_mm256_calc_rsq_ps(dx11,dy11,dz11);
1822             rsq12            = gmx_mm256_calc_rsq_ps(dx12,dy12,dz12);
1823             rsq20            = gmx_mm256_calc_rsq_ps(dx20,dy20,dz20);
1824             rsq21            = gmx_mm256_calc_rsq_ps(dx21,dy21,dz21);
1825             rsq22            = gmx_mm256_calc_rsq_ps(dx22,dy22,dz22);
1826
1827             rinv00           = gmx_mm256_invsqrt_ps(rsq00);
1828             rinv01           = gmx_mm256_invsqrt_ps(rsq01);
1829             rinv02           = gmx_mm256_invsqrt_ps(rsq02);
1830             rinv10           = gmx_mm256_invsqrt_ps(rsq10);
1831             rinv11           = gmx_mm256_invsqrt_ps(rsq11);
1832             rinv12           = gmx_mm256_invsqrt_ps(rsq12);
1833             rinv20           = gmx_mm256_invsqrt_ps(rsq20);
1834             rinv21           = gmx_mm256_invsqrt_ps(rsq21);
1835             rinv22           = gmx_mm256_invsqrt_ps(rsq22);
1836
1837             rinvsq00         = _mm256_mul_ps(rinv00,rinv00);
1838             rinvsq01         = _mm256_mul_ps(rinv01,rinv01);
1839             rinvsq02         = _mm256_mul_ps(rinv02,rinv02);
1840             rinvsq10         = _mm256_mul_ps(rinv10,rinv10);
1841             rinvsq11         = _mm256_mul_ps(rinv11,rinv11);
1842             rinvsq12         = _mm256_mul_ps(rinv12,rinv12);
1843             rinvsq20         = _mm256_mul_ps(rinv20,rinv20);
1844             rinvsq21         = _mm256_mul_ps(rinv21,rinv21);
1845             rinvsq22         = _mm256_mul_ps(rinv22,rinv22);
1846
1847             fjx0             = _mm256_setzero_ps();
1848             fjy0             = _mm256_setzero_ps();
1849             fjz0             = _mm256_setzero_ps();
1850             fjx1             = _mm256_setzero_ps();
1851             fjy1             = _mm256_setzero_ps();
1852             fjz1             = _mm256_setzero_ps();
1853             fjx2             = _mm256_setzero_ps();
1854             fjy2             = _mm256_setzero_ps();
1855             fjz2             = _mm256_setzero_ps();
1856
1857             /**************************
1858              * CALCULATE INTERACTIONS *
1859              **************************/
1860
1861             r00              = _mm256_mul_ps(rsq00,rinv00);
1862             r00              = _mm256_andnot_ps(dummy_mask,r00);
1863
1864             /* EWALD ELECTROSTATICS */
1865
1866             /* Analytical PME correction */
1867             zeta2            = _mm256_mul_ps(beta2,rsq00);
1868             rinv3            = _mm256_mul_ps(rinvsq00,rinv00);
1869             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
1870             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
1871             felec            = _mm256_mul_ps(qq00,felec);
1872
1873             fscal            = felec;
1874
1875             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
1876
1877             /* Calculate temporary vectorial force */
1878             tx               = _mm256_mul_ps(fscal,dx00);
1879             ty               = _mm256_mul_ps(fscal,dy00);
1880             tz               = _mm256_mul_ps(fscal,dz00);
1881
1882             /* Update vectorial force */
1883             fix0             = _mm256_add_ps(fix0,tx);
1884             fiy0             = _mm256_add_ps(fiy0,ty);
1885             fiz0             = _mm256_add_ps(fiz0,tz);
1886
1887             fjx0             = _mm256_add_ps(fjx0,tx);
1888             fjy0             = _mm256_add_ps(fjy0,ty);
1889             fjz0             = _mm256_add_ps(fjz0,tz);
1890
1891             /**************************
1892              * CALCULATE INTERACTIONS *
1893              **************************/
1894
1895             r01              = _mm256_mul_ps(rsq01,rinv01);
1896             r01              = _mm256_andnot_ps(dummy_mask,r01);
1897
1898             /* EWALD ELECTROSTATICS */
1899
1900             /* Analytical PME correction */
1901             zeta2            = _mm256_mul_ps(beta2,rsq01);
1902             rinv3            = _mm256_mul_ps(rinvsq01,rinv01);
1903             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
1904             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
1905             felec            = _mm256_mul_ps(qq01,felec);
1906
1907             fscal            = felec;
1908
1909             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
1910
1911             /* Calculate temporary vectorial force */
1912             tx               = _mm256_mul_ps(fscal,dx01);
1913             ty               = _mm256_mul_ps(fscal,dy01);
1914             tz               = _mm256_mul_ps(fscal,dz01);
1915
1916             /* Update vectorial force */
1917             fix0             = _mm256_add_ps(fix0,tx);
1918             fiy0             = _mm256_add_ps(fiy0,ty);
1919             fiz0             = _mm256_add_ps(fiz0,tz);
1920
1921             fjx1             = _mm256_add_ps(fjx1,tx);
1922             fjy1             = _mm256_add_ps(fjy1,ty);
1923             fjz1             = _mm256_add_ps(fjz1,tz);
1924
1925             /**************************
1926              * CALCULATE INTERACTIONS *
1927              **************************/
1928
1929             r02              = _mm256_mul_ps(rsq02,rinv02);
1930             r02              = _mm256_andnot_ps(dummy_mask,r02);
1931
1932             /* EWALD ELECTROSTATICS */
1933
1934             /* Analytical PME correction */
1935             zeta2            = _mm256_mul_ps(beta2,rsq02);
1936             rinv3            = _mm256_mul_ps(rinvsq02,rinv02);
1937             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
1938             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
1939             felec            = _mm256_mul_ps(qq02,felec);
1940
1941             fscal            = felec;
1942
1943             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
1944
1945             /* Calculate temporary vectorial force */
1946             tx               = _mm256_mul_ps(fscal,dx02);
1947             ty               = _mm256_mul_ps(fscal,dy02);
1948             tz               = _mm256_mul_ps(fscal,dz02);
1949
1950             /* Update vectorial force */
1951             fix0             = _mm256_add_ps(fix0,tx);
1952             fiy0             = _mm256_add_ps(fiy0,ty);
1953             fiz0             = _mm256_add_ps(fiz0,tz);
1954
1955             fjx2             = _mm256_add_ps(fjx2,tx);
1956             fjy2             = _mm256_add_ps(fjy2,ty);
1957             fjz2             = _mm256_add_ps(fjz2,tz);
1958
1959             /**************************
1960              * CALCULATE INTERACTIONS *
1961              **************************/
1962
1963             r10              = _mm256_mul_ps(rsq10,rinv10);
1964             r10              = _mm256_andnot_ps(dummy_mask,r10);
1965
1966             /* EWALD ELECTROSTATICS */
1967
1968             /* Analytical PME correction */
1969             zeta2            = _mm256_mul_ps(beta2,rsq10);
1970             rinv3            = _mm256_mul_ps(rinvsq10,rinv10);
1971             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
1972             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
1973             felec            = _mm256_mul_ps(qq10,felec);
1974
1975             fscal            = felec;
1976
1977             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
1978
1979             /* Calculate temporary vectorial force */
1980             tx               = _mm256_mul_ps(fscal,dx10);
1981             ty               = _mm256_mul_ps(fscal,dy10);
1982             tz               = _mm256_mul_ps(fscal,dz10);
1983
1984             /* Update vectorial force */
1985             fix1             = _mm256_add_ps(fix1,tx);
1986             fiy1             = _mm256_add_ps(fiy1,ty);
1987             fiz1             = _mm256_add_ps(fiz1,tz);
1988
1989             fjx0             = _mm256_add_ps(fjx0,tx);
1990             fjy0             = _mm256_add_ps(fjy0,ty);
1991             fjz0             = _mm256_add_ps(fjz0,tz);
1992
1993             /**************************
1994              * CALCULATE INTERACTIONS *
1995              **************************/
1996
1997             r11              = _mm256_mul_ps(rsq11,rinv11);
1998             r11              = _mm256_andnot_ps(dummy_mask,r11);
1999
2000             /* EWALD ELECTROSTATICS */
2001
2002             /* Analytical PME correction */
2003             zeta2            = _mm256_mul_ps(beta2,rsq11);
2004             rinv3            = _mm256_mul_ps(rinvsq11,rinv11);
2005             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
2006             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
2007             felec            = _mm256_mul_ps(qq11,felec);
2008
2009             fscal            = felec;
2010
2011             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
2012
2013             /* Calculate temporary vectorial force */
2014             tx               = _mm256_mul_ps(fscal,dx11);
2015             ty               = _mm256_mul_ps(fscal,dy11);
2016             tz               = _mm256_mul_ps(fscal,dz11);
2017
2018             /* Update vectorial force */
2019             fix1             = _mm256_add_ps(fix1,tx);
2020             fiy1             = _mm256_add_ps(fiy1,ty);
2021             fiz1             = _mm256_add_ps(fiz1,tz);
2022
2023             fjx1             = _mm256_add_ps(fjx1,tx);
2024             fjy1             = _mm256_add_ps(fjy1,ty);
2025             fjz1             = _mm256_add_ps(fjz1,tz);
2026
2027             /**************************
2028              * CALCULATE INTERACTIONS *
2029              **************************/
2030
2031             r12              = _mm256_mul_ps(rsq12,rinv12);
2032             r12              = _mm256_andnot_ps(dummy_mask,r12);
2033
2034             /* EWALD ELECTROSTATICS */
2035
2036             /* Analytical PME correction */
2037             zeta2            = _mm256_mul_ps(beta2,rsq12);
2038             rinv3            = _mm256_mul_ps(rinvsq12,rinv12);
2039             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
2040             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
2041             felec            = _mm256_mul_ps(qq12,felec);
2042
2043             fscal            = felec;
2044
2045             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
2046
2047             /* Calculate temporary vectorial force */
2048             tx               = _mm256_mul_ps(fscal,dx12);
2049             ty               = _mm256_mul_ps(fscal,dy12);
2050             tz               = _mm256_mul_ps(fscal,dz12);
2051
2052             /* Update vectorial force */
2053             fix1             = _mm256_add_ps(fix1,tx);
2054             fiy1             = _mm256_add_ps(fiy1,ty);
2055             fiz1             = _mm256_add_ps(fiz1,tz);
2056
2057             fjx2             = _mm256_add_ps(fjx2,tx);
2058             fjy2             = _mm256_add_ps(fjy2,ty);
2059             fjz2             = _mm256_add_ps(fjz2,tz);
2060
2061             /**************************
2062              * CALCULATE INTERACTIONS *
2063              **************************/
2064
2065             r20              = _mm256_mul_ps(rsq20,rinv20);
2066             r20              = _mm256_andnot_ps(dummy_mask,r20);
2067
2068             /* EWALD ELECTROSTATICS */
2069
2070             /* Analytical PME correction */
2071             zeta2            = _mm256_mul_ps(beta2,rsq20);
2072             rinv3            = _mm256_mul_ps(rinvsq20,rinv20);
2073             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
2074             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
2075             felec            = _mm256_mul_ps(qq20,felec);
2076
2077             fscal            = felec;
2078
2079             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
2080
2081             /* Calculate temporary vectorial force */
2082             tx               = _mm256_mul_ps(fscal,dx20);
2083             ty               = _mm256_mul_ps(fscal,dy20);
2084             tz               = _mm256_mul_ps(fscal,dz20);
2085
2086             /* Update vectorial force */
2087             fix2             = _mm256_add_ps(fix2,tx);
2088             fiy2             = _mm256_add_ps(fiy2,ty);
2089             fiz2             = _mm256_add_ps(fiz2,tz);
2090
2091             fjx0             = _mm256_add_ps(fjx0,tx);
2092             fjy0             = _mm256_add_ps(fjy0,ty);
2093             fjz0             = _mm256_add_ps(fjz0,tz);
2094
2095             /**************************
2096              * CALCULATE INTERACTIONS *
2097              **************************/
2098
2099             r21              = _mm256_mul_ps(rsq21,rinv21);
2100             r21              = _mm256_andnot_ps(dummy_mask,r21);
2101
2102             /* EWALD ELECTROSTATICS */
2103
2104             /* Analytical PME correction */
2105             zeta2            = _mm256_mul_ps(beta2,rsq21);
2106             rinv3            = _mm256_mul_ps(rinvsq21,rinv21);
2107             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
2108             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
2109             felec            = _mm256_mul_ps(qq21,felec);
2110
2111             fscal            = felec;
2112
2113             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
2114
2115             /* Calculate temporary vectorial force */
2116             tx               = _mm256_mul_ps(fscal,dx21);
2117             ty               = _mm256_mul_ps(fscal,dy21);
2118             tz               = _mm256_mul_ps(fscal,dz21);
2119
2120             /* Update vectorial force */
2121             fix2             = _mm256_add_ps(fix2,tx);
2122             fiy2             = _mm256_add_ps(fiy2,ty);
2123             fiz2             = _mm256_add_ps(fiz2,tz);
2124
2125             fjx1             = _mm256_add_ps(fjx1,tx);
2126             fjy1             = _mm256_add_ps(fjy1,ty);
2127             fjz1             = _mm256_add_ps(fjz1,tz);
2128
2129             /**************************
2130              * CALCULATE INTERACTIONS *
2131              **************************/
2132
2133             r22              = _mm256_mul_ps(rsq22,rinv22);
2134             r22              = _mm256_andnot_ps(dummy_mask,r22);
2135
2136             /* EWALD ELECTROSTATICS */
2137
2138             /* Analytical PME correction */
2139             zeta2            = _mm256_mul_ps(beta2,rsq22);
2140             rinv3            = _mm256_mul_ps(rinvsq22,rinv22);
2141             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
2142             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
2143             felec            = _mm256_mul_ps(qq22,felec);
2144
2145             fscal            = felec;
2146
2147             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
2148
2149             /* Calculate temporary vectorial force */
2150             tx               = _mm256_mul_ps(fscal,dx22);
2151             ty               = _mm256_mul_ps(fscal,dy22);
2152             tz               = _mm256_mul_ps(fscal,dz22);
2153
2154             /* Update vectorial force */
2155             fix2             = _mm256_add_ps(fix2,tx);
2156             fiy2             = _mm256_add_ps(fiy2,ty);
2157             fiz2             = _mm256_add_ps(fiz2,tz);
2158
2159             fjx2             = _mm256_add_ps(fjx2,tx);
2160             fjy2             = _mm256_add_ps(fjy2,ty);
2161             fjz2             = _mm256_add_ps(fjz2,tz);
2162
2163             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
2164             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
2165             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
2166             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
2167             fjptrE             = (jnrlistE>=0) ? f+j_coord_offsetE : scratch;
2168             fjptrF             = (jnrlistF>=0) ? f+j_coord_offsetF : scratch;
2169             fjptrG             = (jnrlistG>=0) ? f+j_coord_offsetG : scratch;
2170             fjptrH             = (jnrlistH>=0) ? f+j_coord_offsetH : scratch;
2171
2172             gmx_mm256_decrement_3rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,
2173                                                       fjx0,fjy0,fjz0,fjx1,fjy1,fjz1,fjx2,fjy2,fjz2);
2174
2175             /* Inner loop uses 513 flops */
2176         }
2177
2178         /* End of innermost loop */
2179
2180         gmx_mm256_update_iforce_3atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
2181                                                  f+i_coord_offset,fshift+i_shift_offset);
2182
2183         /* Increment number of inner iterations */
2184         inneriter                  += j_index_end - j_index_start;
2185
2186         /* Outer loop uses 18 flops */
2187     }
2188
2189     /* Increment number of outer iterations */
2190     outeriter        += nri;
2191
2192     /* Update outer/inner flops */
2193
2194     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_W3W3_F,outeriter*18 + inneriter*513);
2195 }