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 "gmxpre.h"
  39
  40 #include "config.h"
  41
  42 #include <math.h>
  43
  44 #include "../nb_kernel.h"
  45 #include "gromacs/legacyheaders/types/simple.h"
  46 #include "gromacs/math/vec.h"
  47 #include "gromacs/legacyheaders/nrnb.h"
  48
  49 #include "gromacs/simd/math_x86_avx_256_single.h"
  50 #include "kernelutil_x86_avx_256_single.h"
  51
  52 /*
  53  * Gromacs nonbonded kernel:   nb_kernel_ElecEw_VdwNone_GeomW3W3_VF_avx_256_single
  54  * Electrostatics interaction: Ewald
  55  * VdW interaction:            None
  56  * Geometry:                   Water3-Water3
  57  * Calculate force/pot:        PotentialAndForce
  58  */
  59 void
  60 nb_kernel_ElecEw_VdwNone_GeomW3W3_VF_avx_256_single
  61                     (t_nblist                    * gmx_restrict       nlist,
  62                      rvec                        * gmx_restrict          xx,
  63                      rvec                        * gmx_restrict          ff,
  64                      t_forcerec                  * gmx_restrict          fr,
  65                      t_mdatoms                   * gmx_restrict     mdatoms,
  66                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
  67                      t_nrnb                      * gmx_restrict        nrnb)
  68 {
  69     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
  70      * just 0 for non-waters.
  71      * Suffixes A,B,C,D,E,F,G,H refer to j loop unrolling done with AVX, e.g. for the eight different
  72      * jnr indices corresponding to data put in the four positions in the SIMD register.
  73      */
  74     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
  75     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
  76     int              jnrA,jnrB,jnrC,jnrD;
  77     int              jnrE,jnrF,jnrG,jnrH;
  78     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
  79     int              jnrlistE,jnrlistF,jnrlistG,jnrlistH;
  80     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
  81     int              j_coord_offsetE,j_coord_offsetF,j_coord_offsetG,j_coord_offsetH;
  82     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
  83     real             rcutoff_scalar;
  84     real             *shiftvec,*fshift,*x,*f;
  85     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD,*fjptrE,*fjptrF,*fjptrG,*fjptrH;
  86     real             scratch[4*DIM];
  87     __m256           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
  88     real *           vdwioffsetptr0;
  89     __m256           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
  90     real *           vdwioffsetptr1;
  91     __m256           ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
  92     real *           vdwioffsetptr2;
  93     __m256           ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
  94     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D,vdwjidx0E,vdwjidx0F,vdwjidx0G,vdwjidx0H;
  95     __m256           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  96     int              vdwjidx1A,vdwjidx1B,vdwjidx1C,vdwjidx1D,vdwjidx1E,vdwjidx1F,vdwjidx1G,vdwjidx1H;
  97     __m256           jx1,jy1,jz1,fjx1,fjy1,fjz1,jq1,isaj1;
  98     int              vdwjidx2A,vdwjidx2B,vdwjidx2C,vdwjidx2D,vdwjidx2E,vdwjidx2F,vdwjidx2G,vdwjidx2H;
  99     __m256           jx2,jy2,jz2,fjx2,fjy2,fjz2,jq2,isaj2;
 100     __m256           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 101     __m256           dx01,dy01,dz01,rsq01,rinv01,rinvsq01,r01,qq01,c6_01,c12_01;
 102     __m256           dx02,dy02,dz02,rsq02,rinv02,rinvsq02,r02,qq02,c6_02,c12_02;
 103     __m256           dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
 104     __m256           dx11,dy11,dz11,rsq11,rinv11,rinvsq11,r11,qq11,c6_11,c12_11;
 105     __m256           dx12,dy12,dz12,rsq12,rinv12,rinvsq12,r12,qq12,c6_12,c12_12;
 106     __m256           dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
 107     __m256           dx21,dy21,dz21,rsq21,rinv21,rinvsq21,r21,qq21,c6_21,c12_21;
 108     __m256           dx22,dy22,dz22,rsq22,rinv22,rinvsq22,r22,qq22,c6_22,c12_22;
 109     __m256           velec,felec,velecsum,facel,crf,krf,krf2;
 110     real             *charge;
 111     __m256i          ewitab;
 112     __m128i          ewitab_lo,ewitab_hi;
 113     __m256           ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
 114     __m256           beta,beta2,beta3,zeta2,pmecorrF,pmecorrV,rinv3;
 115     real             *ewtab;
 116     __m256           dummy_mask,cutoff_mask;
 117     __m256           signbit = _mm256_castsi256_ps( _mm256_set1_epi32(0x80000000) );
 118     __m256           one     = _mm256_set1_ps(1.0);
 119     __m256           two     = _mm256_set1_ps(2.0);
 120     x                = xx[0];
 121     f                = ff[0];
 122
 123     nri              = nlist->nri;
 124     iinr             = nlist->iinr;
 125     jindex           = nlist->jindex;
 126     jjnr             = nlist->jjnr;
 127     shiftidx         = nlist->shift;
 128     gid              = nlist->gid;
 129     shiftvec         = fr->shift_vec[0];
 130     fshift           = fr->fshift[0];
 131     facel            = _mm256_set1_ps(fr->epsfac);
 132     charge           = mdatoms->chargeA;
 133
 134     sh_ewald         = _mm256_set1_ps(fr->ic->sh_ewald);
 135     beta             = _mm256_set1_ps(fr->ic->ewaldcoeff_q);
 136     beta2            = _mm256_mul_ps(beta,beta);
 137     beta3            = _mm256_mul_ps(beta,beta2);
 138
 139     ewtab            = fr->ic->tabq_coul_FDV0;
 140     ewtabscale       = _mm256_set1_ps(fr->ic->tabq_scale);
 141     ewtabhalfspace   = _mm256_set1_ps(0.5/fr->ic->tabq_scale);
 142
 143     /* Setup water-specific parameters */
 144     inr              = nlist->iinr[0];
 145     iq0              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+0]));
 146     iq1              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+1]));
 147     iq2              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+2]));
 148
 149     jq0              = _mm256_set1_ps(charge[inr+0]);
 150     jq1              = _mm256_set1_ps(charge[inr+1]);
 151     jq2              = _mm256_set1_ps(charge[inr+2]);
 152     qq00             = _mm256_mul_ps(iq0,jq0);
 153     qq01             = _mm256_mul_ps(iq0,jq1);
 154     qq02             = _mm256_mul_ps(iq0,jq2);
 155     qq10             = _mm256_mul_ps(iq1,jq0);
 156     qq11             = _mm256_mul_ps(iq1,jq1);
 157     qq12             = _mm256_mul_ps(iq1,jq2);
 158     qq20             = _mm256_mul_ps(iq2,jq0);
 159     qq21             = _mm256_mul_ps(iq2,jq1);
 160     qq22             = _mm256_mul_ps(iq2,jq2);
 161
 162     /* Avoid stupid compiler warnings */
 163     jnrA = jnrB = jnrC = jnrD = jnrE = jnrF = jnrG = jnrH = 0;
 164     j_coord_offsetA = 0;
 165     j_coord_offsetB = 0;
 166     j_coord_offsetC = 0;
 167     j_coord_offsetD = 0;
 168     j_coord_offsetE = 0;
 169     j_coord_offsetF = 0;
 170     j_coord_offsetG = 0;
 171     j_coord_offsetH = 0;
 172
 173     outeriter        = 0;
 174     inneriter        = 0;
 175
 176     for(iidx=0;iidx<4*DIM;iidx++)
 177     {
 178         scratch[iidx] = 0.0;
 179     }
 180
 181     /* Start outer loop over neighborlists */
 182     for(iidx=0; iidx<nri; iidx++)
 183     {
 184         /* Load shift vector for this list */
 185         i_shift_offset   = DIM*shiftidx[iidx];
 186
 187         /* Load limits for loop over neighbors */
 188         j_index_start    = jindex[iidx];
 189         j_index_end      = jindex[iidx+1];
 190
 191         /* Get outer coordinate index */
 192         inr              = iinr[iidx];
 193         i_coord_offset   = DIM*inr;
 194
 195         /* Load i particle coords and add shift vector */
 196         gmx_mm256_load_shift_and_3rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
 197                                                     &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 198
 199         fix0             = _mm256_setzero_ps();
 200         fiy0             = _mm256_setzero_ps();
 201         fiz0             = _mm256_setzero_ps();
 202         fix1             = _mm256_setzero_ps();
 203         fiy1             = _mm256_setzero_ps();
 204         fiz1             = _mm256_setzero_ps();
 205         fix2             = _mm256_setzero_ps();
 206         fiy2             = _mm256_setzero_ps();
 207         fiz2             = _mm256_setzero_ps();
 208
 209         /* Reset potential sums */
 210         velecsum         = _mm256_setzero_ps();
 211
 212         /* Start inner kernel loop */
 213         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+7]>=0; jidx+=8)
 214         {
 215
 216             /* Get j neighbor index, and coordinate index */
 217             jnrA             = jjnr[jidx];
 218             jnrB             = jjnr[jidx+1];
 219             jnrC             = jjnr[jidx+2];
 220             jnrD             = jjnr[jidx+3];
 221             jnrE             = jjnr[jidx+4];
 222             jnrF             = jjnr[jidx+5];
 223             jnrG             = jjnr[jidx+6];
 224             jnrH             = jjnr[jidx+7];
 225             j_coord_offsetA  = DIM*jnrA;
 226             j_coord_offsetB  = DIM*jnrB;
 227             j_coord_offsetC  = DIM*jnrC;
 228             j_coord_offsetD  = DIM*jnrD;
 229             j_coord_offsetE  = DIM*jnrE;
 230             j_coord_offsetF  = DIM*jnrF;
 231             j_coord_offsetG  = DIM*jnrG;
 232             j_coord_offsetH  = DIM*jnrH;
 233
 234             /* load j atom coordinates */
 235             gmx_mm256_load_3rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 236                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 237                                                  x+j_coord_offsetE,x+j_coord_offsetF,
 238                                                  x+j_coord_offsetG,x+j_coord_offsetH,
 239                                               &jx0,&jy0,&jz0,&jx1,&jy1,&jz1,&jx2,&jy2,&jz2);
 240
 241             /* Calculate displacement vector */
 242             dx00             = _mm256_sub_ps(ix0,jx0);
 243             dy00             = _mm256_sub_ps(iy0,jy0);
 244             dz00             = _mm256_sub_ps(iz0,jz0);
 245             dx01             = _mm256_sub_ps(ix0,jx1);
 246             dy01             = _mm256_sub_ps(iy0,jy1);
 247             dz01             = _mm256_sub_ps(iz0,jz1);
 248             dx02             = _mm256_sub_ps(ix0,jx2);
 249             dy02             = _mm256_sub_ps(iy0,jy2);
 250             dz02             = _mm256_sub_ps(iz0,jz2);
 251             dx10             = _mm256_sub_ps(ix1,jx0);
 252             dy10             = _mm256_sub_ps(iy1,jy0);
 253             dz10             = _mm256_sub_ps(iz1,jz0);
 254             dx11             = _mm256_sub_ps(ix1,jx1);
 255             dy11             = _mm256_sub_ps(iy1,jy1);
 256             dz11             = _mm256_sub_ps(iz1,jz1);
 257             dx12             = _mm256_sub_ps(ix1,jx2);
 258             dy12             = _mm256_sub_ps(iy1,jy2);
 259             dz12             = _mm256_sub_ps(iz1,jz2);
 260             dx20             = _mm256_sub_ps(ix2,jx0);
 261             dy20             = _mm256_sub_ps(iy2,jy0);
 262             dz20             = _mm256_sub_ps(iz2,jz0);
 263             dx21             = _mm256_sub_ps(ix2,jx1);
 264             dy21             = _mm256_sub_ps(iy2,jy1);
 265             dz21             = _mm256_sub_ps(iz2,jz1);
 266             dx22             = _mm256_sub_ps(ix2,jx2);
 267             dy22             = _mm256_sub_ps(iy2,jy2);
 268             dz22             = _mm256_sub_ps(iz2,jz2);
 269
 270             /* Calculate squared distance and things based on it */
 271             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
 272             rsq01            = gmx_mm256_calc_rsq_ps(dx01,dy01,dz01);
 273             rsq02            = gmx_mm256_calc_rsq_ps(dx02,dy02,dz02);
 274             rsq10            = gmx_mm256_calc_rsq_ps(dx10,dy10,dz10);
 275             rsq11            = gmx_mm256_calc_rsq_ps(dx11,dy11,dz11);
 276             rsq12            = gmx_mm256_calc_rsq_ps(dx12,dy12,dz12);
 277             rsq20            = gmx_mm256_calc_rsq_ps(dx20,dy20,dz20);
 278             rsq21            = gmx_mm256_calc_rsq_ps(dx21,dy21,dz21);
 279             rsq22            = gmx_mm256_calc_rsq_ps(dx22,dy22,dz22);
 280
 281             rinv00           = gmx_mm256_invsqrt_ps(rsq00);
 282             rinv01           = gmx_mm256_invsqrt_ps(rsq01);
 283             rinv02           = gmx_mm256_invsqrt_ps(rsq02);
 284             rinv10           = gmx_mm256_invsqrt_ps(rsq10);
 285             rinv11           = gmx_mm256_invsqrt_ps(rsq11);
 286             rinv12           = gmx_mm256_invsqrt_ps(rsq12);
 287             rinv20           = gmx_mm256_invsqrt_ps(rsq20);
 288             rinv21           = gmx_mm256_invsqrt_ps(rsq21);
 289             rinv22           = gmx_mm256_invsqrt_ps(rsq22);
 290
 291             rinvsq00         = _mm256_mul_ps(rinv00,rinv00);
 292             rinvsq01         = _mm256_mul_ps(rinv01,rinv01);
 293             rinvsq02         = _mm256_mul_ps(rinv02,rinv02);
 294             rinvsq10         = _mm256_mul_ps(rinv10,rinv10);
 295             rinvsq11         = _mm256_mul_ps(rinv11,rinv11);
 296             rinvsq12         = _mm256_mul_ps(rinv12,rinv12);
 297             rinvsq20         = _mm256_mul_ps(rinv20,rinv20);
 298             rinvsq21         = _mm256_mul_ps(rinv21,rinv21);
 299             rinvsq22         = _mm256_mul_ps(rinv22,rinv22);
 300
 301             fjx0             = _mm256_setzero_ps();
 302             fjy0             = _mm256_setzero_ps();
 303             fjz0             = _mm256_setzero_ps();
 304             fjx1             = _mm256_setzero_ps();
 305             fjy1             = _mm256_setzero_ps();
 306             fjz1             = _mm256_setzero_ps();
 307             fjx2             = _mm256_setzero_ps();
 308             fjy2             = _mm256_setzero_ps();
 309             fjz2             = _mm256_setzero_ps();
 310
 311             /**************************
 312              * CALCULATE INTERACTIONS *
 313              **************************/
 314
 315             r00              = _mm256_mul_ps(rsq00,rinv00);
 316
 317             /* EWALD ELECTROSTATICS */
 318
 319             /* Analytical PME correction */
 320             zeta2            = _mm256_mul_ps(beta2,rsq00);
 321             rinv3            = _mm256_mul_ps(rinvsq00,rinv00);
 322             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
 323             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
 324             felec            = _mm256_mul_ps(qq00,felec);
 325             pmecorrV         = gmx_mm256_pmecorrV_ps(zeta2);
 326             pmecorrV         = _mm256_mul_ps(pmecorrV,beta);
 327             velec            = _mm256_sub_ps(rinv00,pmecorrV);
 328             velec            = _mm256_mul_ps(qq00,velec);
 329
 330             /* Update potential sum for this i atom from the interaction with this j atom. */
 331             velecsum         = _mm256_add_ps(velecsum,velec);
 332
 333             fscal            = felec;
 334
 335             /* Calculate temporary vectorial force */
 336             tx               = _mm256_mul_ps(fscal,dx00);
 337             ty               = _mm256_mul_ps(fscal,dy00);
 338             tz               = _mm256_mul_ps(fscal,dz00);
 339
 340             /* Update vectorial force */
 341             fix0             = _mm256_add_ps(fix0,tx);
 342             fiy0             = _mm256_add_ps(fiy0,ty);
 343             fiz0             = _mm256_add_ps(fiz0,tz);
 344
 345             fjx0             = _mm256_add_ps(fjx0,tx);
 346             fjy0             = _mm256_add_ps(fjy0,ty);
 347             fjz0             = _mm256_add_ps(fjz0,tz);
 348
 349             /**************************
 350              * CALCULATE INTERACTIONS *
 351              **************************/
 352
 353             r01              = _mm256_mul_ps(rsq01,rinv01);
 354
 355             /* EWALD ELECTROSTATICS */
 356
 357             /* Analytical PME correction */
 358             zeta2            = _mm256_mul_ps(beta2,rsq01);
 359             rinv3            = _mm256_mul_ps(rinvsq01,rinv01);
 360             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
 361             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
 362             felec            = _mm256_mul_ps(qq01,felec);
 363             pmecorrV         = gmx_mm256_pmecorrV_ps(zeta2);
 364             pmecorrV         = _mm256_mul_ps(pmecorrV,beta);
 365             velec            = _mm256_sub_ps(rinv01,pmecorrV);
 366             velec            = _mm256_mul_ps(qq01,velec);
 367
 368             /* Update potential sum for this i atom from the interaction with this j atom. */
 369             velecsum         = _mm256_add_ps(velecsum,velec);
 370
 371             fscal            = felec;
 372
 373             /* Calculate temporary vectorial force */
 374             tx               = _mm256_mul_ps(fscal,dx01);
 375             ty               = _mm256_mul_ps(fscal,dy01);
 376             tz               = _mm256_mul_ps(fscal,dz01);
 377
 378             /* Update vectorial force */
 379             fix0             = _mm256_add_ps(fix0,tx);
 380             fiy0             = _mm256_add_ps(fiy0,ty);
 381             fiz0             = _mm256_add_ps(fiz0,tz);
 382
 383             fjx1             = _mm256_add_ps(fjx1,tx);
 384             fjy1             = _mm256_add_ps(fjy1,ty);
 385             fjz1             = _mm256_add_ps(fjz1,tz);
 386
 387             /**************************
 388              * CALCULATE INTERACTIONS *
 389              **************************/
 390
 391             r02              = _mm256_mul_ps(rsq02,rinv02);
 392
 393             /* EWALD ELECTROSTATICS */
 394
 395             /* Analytical PME correction */
 396             zeta2            = _mm256_mul_ps(beta2,rsq02);
 397             rinv3            = _mm256_mul_ps(rinvsq02,rinv02);
 398             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
 399             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
 400             felec            = _mm256_mul_ps(qq02,felec);
 401             pmecorrV         = gmx_mm256_pmecorrV_ps(zeta2);
 402             pmecorrV         = _mm256_mul_ps(pmecorrV,beta);
 403             velec            = _mm256_sub_ps(rinv02,pmecorrV);
 404             velec            = _mm256_mul_ps(qq02,velec);
 405
 406             /* Update potential sum for this i atom from the interaction with this j atom. */
 407             velecsum         = _mm256_add_ps(velecsum,velec);
 408
 409             fscal            = felec;
 410
 411             /* Calculate temporary vectorial force */
 412             tx               = _mm256_mul_ps(fscal,dx02);
 413             ty               = _mm256_mul_ps(fscal,dy02);
 414             tz               = _mm256_mul_ps(fscal,dz02);
 415
 416             /* Update vectorial force */
 417             fix0             = _mm256_add_ps(fix0,tx);
 418             fiy0             = _mm256_add_ps(fiy0,ty);
 419             fiz0             = _mm256_add_ps(fiz0,tz);
 420
 421             fjx2             = _mm256_add_ps(fjx2,tx);
 422             fjy2             = _mm256_add_ps(fjy2,ty);
 423             fjz2             = _mm256_add_ps(fjz2,tz);
 424
 425             /**************************
 426              * CALCULATE INTERACTIONS *
 427              **************************/
 428
 429             r10              = _mm256_mul_ps(rsq10,rinv10);
 430
 431             /* EWALD ELECTROSTATICS */
 432
 433             /* Analytical PME correction */
 434             zeta2            = _mm256_mul_ps(beta2,rsq10);
 435             rinv3            = _mm256_mul_ps(rinvsq10,rinv10);
 436             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
 437             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
 438             felec            = _mm256_mul_ps(qq10,felec);
 439             pmecorrV         = gmx_mm256_pmecorrV_ps(zeta2);
 440             pmecorrV         = _mm256_mul_ps(pmecorrV,beta);
 441             velec            = _mm256_sub_ps(rinv10,pmecorrV);
 442             velec            = _mm256_mul_ps(qq10,velec);
 443
 444             /* Update potential sum for this i atom from the interaction with this j atom. */
 445             velecsum         = _mm256_add_ps(velecsum,velec);
 446
 447             fscal            = felec;
 448
 449             /* Calculate temporary vectorial force */
 450             tx               = _mm256_mul_ps(fscal,dx10);
 451             ty               = _mm256_mul_ps(fscal,dy10);
 452             tz               = _mm256_mul_ps(fscal,dz10);
 453
 454             /* Update vectorial force */
 455             fix1             = _mm256_add_ps(fix1,tx);
 456             fiy1             = _mm256_add_ps(fiy1,ty);
 457             fiz1             = _mm256_add_ps(fiz1,tz);
 458
 459             fjx0             = _mm256_add_ps(fjx0,tx);
 460             fjy0             = _mm256_add_ps(fjy0,ty);
 461             fjz0             = _mm256_add_ps(fjz0,tz);
 462
 463             /**************************
 464              * CALCULATE INTERACTIONS *
 465              **************************/
 466
 467             r11              = _mm256_mul_ps(rsq11,rinv11);
 468
 469             /* EWALD ELECTROSTATICS */
 470
 471             /* Analytical PME correction */
 472             zeta2            = _mm256_mul_ps(beta2,rsq11);
 473             rinv3            = _mm256_mul_ps(rinvsq11,rinv11);
 474             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
 475             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
 476             felec            = _mm256_mul_ps(qq11,felec);
 477             pmecorrV         = gmx_mm256_pmecorrV_ps(zeta2);
 478             pmecorrV         = _mm256_mul_ps(pmecorrV,beta);
 479             velec            = _mm256_sub_ps(rinv11,pmecorrV);
 480             velec            = _mm256_mul_ps(qq11,velec);
 481
 482             /* Update potential sum for this i atom from the interaction with this j atom. */
 483             velecsum         = _mm256_add_ps(velecsum,velec);
 484
 485             fscal            = felec;
 486
 487             /* Calculate temporary vectorial force */
 488             tx               = _mm256_mul_ps(fscal,dx11);
 489             ty               = _mm256_mul_ps(fscal,dy11);
 490             tz               = _mm256_mul_ps(fscal,dz11);
 491
 492             /* Update vectorial force */
 493             fix1             = _mm256_add_ps(fix1,tx);
 494             fiy1             = _mm256_add_ps(fiy1,ty);
 495             fiz1             = _mm256_add_ps(fiz1,tz);
 496
 497             fjx1             = _mm256_add_ps(fjx1,tx);
 498             fjy1             = _mm256_add_ps(fjy1,ty);
 499             fjz1             = _mm256_add_ps(fjz1,tz);
 500
 501             /**************************
 502              * CALCULATE INTERACTIONS *
 503              **************************/
 504
 505             r12              = _mm256_mul_ps(rsq12,rinv12);
 506
 507             /* EWALD ELECTROSTATICS */
 508
 509             /* Analytical PME correction */
 510             zeta2            = _mm256_mul_ps(beta2,rsq12);
 511             rinv3            = _mm256_mul_ps(rinvsq12,rinv12);
 512             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
 513             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
 514             felec            = _mm256_mul_ps(qq12,felec);
 515             pmecorrV         = gmx_mm256_pmecorrV_ps(zeta2);
 516             pmecorrV         = _mm256_mul_ps(pmecorrV,beta);
 517             velec            = _mm256_sub_ps(rinv12,pmecorrV);
 518             velec            = _mm256_mul_ps(qq12,velec);
 519
 520             /* Update potential sum for this i atom from the interaction with this j atom. */
 521             velecsum         = _mm256_add_ps(velecsum,velec);
 522
 523             fscal            = felec;
 524
 525             /* Calculate temporary vectorial force */
 526             tx               = _mm256_mul_ps(fscal,dx12);
 527             ty               = _mm256_mul_ps(fscal,dy12);
 528             tz               = _mm256_mul_ps(fscal,dz12);
 529
 530             /* Update vectorial force */
 531             fix1             = _mm256_add_ps(fix1,tx);
 532             fiy1             = _mm256_add_ps(fiy1,ty);
 533             fiz1             = _mm256_add_ps(fiz1,tz);
 534
 535             fjx2             = _mm256_add_ps(fjx2,tx);
 536             fjy2             = _mm256_add_ps(fjy2,ty);
 537             fjz2             = _mm256_add_ps(fjz2,tz);
 538
 539             /**************************
 540              * CALCULATE INTERACTIONS *
 541              **************************/
 542
 543             r20              = _mm256_mul_ps(rsq20,rinv20);
 544
 545             /* EWALD ELECTROSTATICS */
 546
 547             /* Analytical PME correction */
 548             zeta2            = _mm256_mul_ps(beta2,rsq20);
 549             rinv3            = _mm256_mul_ps(rinvsq20,rinv20);
 550             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
 551             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
 552             felec            = _mm256_mul_ps(qq20,felec);
 553             pmecorrV         = gmx_mm256_pmecorrV_ps(zeta2);
 554             pmecorrV         = _mm256_mul_ps(pmecorrV,beta);
 555             velec            = _mm256_sub_ps(rinv20,pmecorrV);
 556             velec            = _mm256_mul_ps(qq20,velec);
 557
 558             /* Update potential sum for this i atom from the interaction with this j atom. */
 559             velecsum         = _mm256_add_ps(velecsum,velec);
 560
 561             fscal            = felec;
 562
 563             /* Calculate temporary vectorial force */
 564             tx               = _mm256_mul_ps(fscal,dx20);
 565             ty               = _mm256_mul_ps(fscal,dy20);
 566             tz               = _mm256_mul_ps(fscal,dz20);
 567
 568             /* Update vectorial force */
 569             fix2             = _mm256_add_ps(fix2,tx);
 570             fiy2             = _mm256_add_ps(fiy2,ty);
 571             fiz2             = _mm256_add_ps(fiz2,tz);
 572
 573             fjx0             = _mm256_add_ps(fjx0,tx);
 574             fjy0             = _mm256_add_ps(fjy0,ty);
 575             fjz0             = _mm256_add_ps(fjz0,tz);
 576
 577             /**************************
 578              * CALCULATE INTERACTIONS *
 579              **************************/
 580
 581             r21              = _mm256_mul_ps(rsq21,rinv21);
 582
 583             /* EWALD ELECTROSTATICS */
 584
 585             /* Analytical PME correction */
 586             zeta2            = _mm256_mul_ps(beta2,rsq21);
 587             rinv3            = _mm256_mul_ps(rinvsq21,rinv21);
 588             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
 589             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
 590             felec            = _mm256_mul_ps(qq21,felec);
 591             pmecorrV         = gmx_mm256_pmecorrV_ps(zeta2);
 592             pmecorrV         = _mm256_mul_ps(pmecorrV,beta);
 593             velec            = _mm256_sub_ps(rinv21,pmecorrV);
 594             velec            = _mm256_mul_ps(qq21,velec);
 595
 596             /* Update potential sum for this i atom from the interaction with this j atom. */
 597             velecsum         = _mm256_add_ps(velecsum,velec);
 598
 599             fscal            = felec;
 600
 601             /* Calculate temporary vectorial force */
 602             tx               = _mm256_mul_ps(fscal,dx21);
 603             ty               = _mm256_mul_ps(fscal,dy21);
 604             tz               = _mm256_mul_ps(fscal,dz21);
 605
 606             /* Update vectorial force */
 607             fix2             = _mm256_add_ps(fix2,tx);
 608             fiy2             = _mm256_add_ps(fiy2,ty);
 609             fiz2             = _mm256_add_ps(fiz2,tz);
 610
 611             fjx1             = _mm256_add_ps(fjx1,tx);
 612             fjy1             = _mm256_add_ps(fjy1,ty);
 613             fjz1             = _mm256_add_ps(fjz1,tz);
 614
 615             /**************************
 616              * CALCULATE INTERACTIONS *
 617              **************************/
 618
 619             r22              = _mm256_mul_ps(rsq22,rinv22);
 620
 621             /* EWALD ELECTROSTATICS */
 622
 623             /* Analytical PME correction */
 624             zeta2            = _mm256_mul_ps(beta2,rsq22);
 625             rinv3            = _mm256_mul_ps(rinvsq22,rinv22);
 626             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
 627             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
 628             felec            = _mm256_mul_ps(qq22,felec);
 629             pmecorrV         = gmx_mm256_pmecorrV_ps(zeta2);
 630             pmecorrV         = _mm256_mul_ps(pmecorrV,beta);
 631             velec            = _mm256_sub_ps(rinv22,pmecorrV);
 632             velec            = _mm256_mul_ps(qq22,velec);
 633
 634             /* Update potential sum for this i atom from the interaction with this j atom. */
 635             velecsum         = _mm256_add_ps(velecsum,velec);
 636
 637             fscal            = felec;
 638
 639             /* Calculate temporary vectorial force */
 640             tx               = _mm256_mul_ps(fscal,dx22);
 641             ty               = _mm256_mul_ps(fscal,dy22);
 642             tz               = _mm256_mul_ps(fscal,dz22);
 643
 644             /* Update vectorial force */
 645             fix2             = _mm256_add_ps(fix2,tx);
 646             fiy2             = _mm256_add_ps(fiy2,ty);
 647             fiz2             = _mm256_add_ps(fiz2,tz);
 648
 649             fjx2             = _mm256_add_ps(fjx2,tx);
 650             fjy2             = _mm256_add_ps(fjy2,ty);
 651             fjz2             = _mm256_add_ps(fjz2,tz);
 652
 653             fjptrA             = f+j_coord_offsetA;
 654             fjptrB             = f+j_coord_offsetB;
 655             fjptrC             = f+j_coord_offsetC;
 656             fjptrD             = f+j_coord_offsetD;
 657             fjptrE             = f+j_coord_offsetE;
 658             fjptrF             = f+j_coord_offsetF;
 659             fjptrG             = f+j_coord_offsetG;
 660             fjptrH             = f+j_coord_offsetH;
 661
 662             gmx_mm256_decrement_3rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,
 663                                                       fjx0,fjy0,fjz0,fjx1,fjy1,fjz1,fjx2,fjy2,fjz2);
 664
 665             /* Inner loop uses 756 flops */
 666         }
 667
 668         if(jidx<j_index_end)
 669         {
 670
 671             /* Get j neighbor index, and coordinate index */
 672             jnrlistA         = jjnr[jidx];
 673             jnrlistB         = jjnr[jidx+1];
 674             jnrlistC         = jjnr[jidx+2];
 675             jnrlistD         = jjnr[jidx+3];
 676             jnrlistE         = jjnr[jidx+4];
 677             jnrlistF         = jjnr[jidx+5];
 678             jnrlistG         = jjnr[jidx+6];
 679             jnrlistH         = jjnr[jidx+7];
 680             /* Sign of each element will be negative for non-real atoms.
 681              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 682              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 683              */
 684             dummy_mask = gmx_mm256_set_m128(gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx+4)),_mm_setzero_si128())),
 685                                             gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128())));
 686
 687             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 688             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 689             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 690             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 691             jnrE       = (jnrlistE>=0) ? jnrlistE : 0;
 692             jnrF       = (jnrlistF>=0) ? jnrlistF : 0;
 693             jnrG       = (jnrlistG>=0) ? jnrlistG : 0;
 694             jnrH       = (jnrlistH>=0) ? jnrlistH : 0;
 695             j_coord_offsetA  = DIM*jnrA;
 696             j_coord_offsetB  = DIM*jnrB;
 697             j_coord_offsetC  = DIM*jnrC;
 698             j_coord_offsetD  = DIM*jnrD;
 699             j_coord_offsetE  = DIM*jnrE;
 700             j_coord_offsetF  = DIM*jnrF;
 701             j_coord_offsetG  = DIM*jnrG;
 702             j_coord_offsetH  = DIM*jnrH;
 703
 704             /* load j atom coordinates */
 705             gmx_mm256_load_3rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 706                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 707                                                  x+j_coord_offsetE,x+j_coord_offsetF,
 708                                                  x+j_coord_offsetG,x+j_coord_offsetH,
 709                                               &jx0,&jy0,&jz0,&jx1,&jy1,&jz1,&jx2,&jy2,&jz2);
 710
 711             /* Calculate displacement vector */
 712             dx00             = _mm256_sub_ps(ix0,jx0);
 713             dy00             = _mm256_sub_ps(iy0,jy0);
 714             dz00             = _mm256_sub_ps(iz0,jz0);
 715             dx01             = _mm256_sub_ps(ix0,jx1);
 716             dy01             = _mm256_sub_ps(iy0,jy1);
 717             dz01             = _mm256_sub_ps(iz0,jz1);
 718             dx02             = _mm256_sub_ps(ix0,jx2);
 719             dy02             = _mm256_sub_ps(iy0,jy2);
 720             dz02             = _mm256_sub_ps(iz0,jz2);
 721             dx10             = _mm256_sub_ps(ix1,jx0);
 722             dy10             = _mm256_sub_ps(iy1,jy0);
 723             dz10             = _mm256_sub_ps(iz1,jz0);
 724             dx11             = _mm256_sub_ps(ix1,jx1);
 725             dy11             = _mm256_sub_ps(iy1,jy1);
 726             dz11             = _mm256_sub_ps(iz1,jz1);
 727             dx12             = _mm256_sub_ps(ix1,jx2);
 728             dy12             = _mm256_sub_ps(iy1,jy2);
 729             dz12             = _mm256_sub_ps(iz1,jz2);
 730             dx20             = _mm256_sub_ps(ix2,jx0);
 731             dy20             = _mm256_sub_ps(iy2,jy0);
 732             dz20             = _mm256_sub_ps(iz2,jz0);
 733             dx21             = _mm256_sub_ps(ix2,jx1);
 734             dy21             = _mm256_sub_ps(iy2,jy1);
 735             dz21             = _mm256_sub_ps(iz2,jz1);
 736             dx22             = _mm256_sub_ps(ix2,jx2);
 737             dy22             = _mm256_sub_ps(iy2,jy2);
 738             dz22             = _mm256_sub_ps(iz2,jz2);
 739
 740             /* Calculate squared distance and things based on it */
 741             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
 742             rsq01            = gmx_mm256_calc_rsq_ps(dx01,dy01,dz01);
 743             rsq02            = gmx_mm256_calc_rsq_ps(dx02,dy02,dz02);
 744             rsq10            = gmx_mm256_calc_rsq_ps(dx10,dy10,dz10);
 745             rsq11            = gmx_mm256_calc_rsq_ps(dx11,dy11,dz11);
 746             rsq12            = gmx_mm256_calc_rsq_ps(dx12,dy12,dz12);
 747             rsq20            = gmx_mm256_calc_rsq_ps(dx20,dy20,dz20);
 748             rsq21            = gmx_mm256_calc_rsq_ps(dx21,dy21,dz21);
 749             rsq22            = gmx_mm256_calc_rsq_ps(dx22,dy22,dz22);
 750
 751             rinv00           = gmx_mm256_invsqrt_ps(rsq00);
 752             rinv01           = gmx_mm256_invsqrt_ps(rsq01);
 753             rinv02           = gmx_mm256_invsqrt_ps(rsq02);
 754             rinv10           = gmx_mm256_invsqrt_ps(rsq10);
 755             rinv11           = gmx_mm256_invsqrt_ps(rsq11);
 756             rinv12           = gmx_mm256_invsqrt_ps(rsq12);
 757             rinv20           = gmx_mm256_invsqrt_ps(rsq20);
 758             rinv21           = gmx_mm256_invsqrt_ps(rsq21);
 759             rinv22           = gmx_mm256_invsqrt_ps(rsq22);
 760
 761             rinvsq00         = _mm256_mul_ps(rinv00,rinv00);
 762             rinvsq01         = _mm256_mul_ps(rinv01,rinv01);
 763             rinvsq02         = _mm256_mul_ps(rinv02,rinv02);
 764             rinvsq10         = _mm256_mul_ps(rinv10,rinv10);
 765             rinvsq11         = _mm256_mul_ps(rinv11,rinv11);
 766             rinvsq12         = _mm256_mul_ps(rinv12,rinv12);
 767             rinvsq20         = _mm256_mul_ps(rinv20,rinv20);
 768             rinvsq21         = _mm256_mul_ps(rinv21,rinv21);
 769             rinvsq22         = _mm256_mul_ps(rinv22,rinv22);
 770
 771             fjx0             = _mm256_setzero_ps();
 772             fjy0             = _mm256_setzero_ps();
 773             fjz0             = _mm256_setzero_ps();
 774             fjx1             = _mm256_setzero_ps();
 775             fjy1             = _mm256_setzero_ps();
 776             fjz1             = _mm256_setzero_ps();
 777             fjx2             = _mm256_setzero_ps();
 778             fjy2             = _mm256_setzero_ps();
 779             fjz2             = _mm256_setzero_ps();
 780
 781             /**************************
 782              * CALCULATE INTERACTIONS *
 783              **************************/
 784
 785             r00              = _mm256_mul_ps(rsq00,rinv00);
 786             r00              = _mm256_andnot_ps(dummy_mask,r00);
 787
 788             /* EWALD ELECTROSTATICS */
 789
 790             /* Analytical PME correction */
 791             zeta2            = _mm256_mul_ps(beta2,rsq00);
 792             rinv3            = _mm256_mul_ps(rinvsq00,rinv00);
 793             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
 794             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
 795             felec            = _mm256_mul_ps(qq00,felec);
 796             pmecorrV         = gmx_mm256_pmecorrV_ps(zeta2);
 797             pmecorrV         = _mm256_mul_ps(pmecorrV,beta);
 798             velec            = _mm256_sub_ps(rinv00,pmecorrV);
 799             velec            = _mm256_mul_ps(qq00,velec);
 800
 801             /* Update potential sum for this i atom from the interaction with this j atom. */
 802             velec            = _mm256_andnot_ps(dummy_mask,velec);
 803             velecsum         = _mm256_add_ps(velecsum,velec);
 804
 805             fscal            = felec;
 806
 807             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
 808
 809             /* Calculate temporary vectorial force */
 810             tx               = _mm256_mul_ps(fscal,dx00);
 811             ty               = _mm256_mul_ps(fscal,dy00);
 812             tz               = _mm256_mul_ps(fscal,dz00);
 813
 814             /* Update vectorial force */
 815             fix0             = _mm256_add_ps(fix0,tx);
 816             fiy0             = _mm256_add_ps(fiy0,ty);
 817             fiz0             = _mm256_add_ps(fiz0,tz);
 818
 819             fjx0             = _mm256_add_ps(fjx0,tx);
 820             fjy0             = _mm256_add_ps(fjy0,ty);
 821             fjz0             = _mm256_add_ps(fjz0,tz);
 822
 823             /**************************
 824              * CALCULATE INTERACTIONS *
 825              **************************/
 826
 827             r01              = _mm256_mul_ps(rsq01,rinv01);
 828             r01              = _mm256_andnot_ps(dummy_mask,r01);
 829
 830             /* EWALD ELECTROSTATICS */
 831
 832             /* Analytical PME correction */
 833             zeta2            = _mm256_mul_ps(beta2,rsq01);
 834             rinv3            = _mm256_mul_ps(rinvsq01,rinv01);
 835             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
 836             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
 837             felec            = _mm256_mul_ps(qq01,felec);
 838             pmecorrV         = gmx_mm256_pmecorrV_ps(zeta2);
 839             pmecorrV         = _mm256_mul_ps(pmecorrV,beta);
 840             velec            = _mm256_sub_ps(rinv01,pmecorrV);
 841             velec            = _mm256_mul_ps(qq01,velec);
 842
 843             /* Update potential sum for this i atom from the interaction with this j atom. */
 844             velec            = _mm256_andnot_ps(dummy_mask,velec);
 845             velecsum         = _mm256_add_ps(velecsum,velec);
 846
 847             fscal            = felec;
 848
 849             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
 850
 851             /* Calculate temporary vectorial force */
 852             tx               = _mm256_mul_ps(fscal,dx01);
 853             ty               = _mm256_mul_ps(fscal,dy01);
 854             tz               = _mm256_mul_ps(fscal,dz01);
 855
 856             /* Update vectorial force */
 857             fix0             = _mm256_add_ps(fix0,tx);
 858             fiy0             = _mm256_add_ps(fiy0,ty);
 859             fiz0             = _mm256_add_ps(fiz0,tz);
 860
 861             fjx1             = _mm256_add_ps(fjx1,tx);
 862             fjy1             = _mm256_add_ps(fjy1,ty);
 863             fjz1             = _mm256_add_ps(fjz1,tz);
 864
 865             /**************************
 866              * CALCULATE INTERACTIONS *
 867              **************************/
 868
 869             r02              = _mm256_mul_ps(rsq02,rinv02);
 870             r02              = _mm256_andnot_ps(dummy_mask,r02);
 871
 872             /* EWALD ELECTROSTATICS */
 873
 874             /* Analytical PME correction */
 875             zeta2            = _mm256_mul_ps(beta2,rsq02);
 876             rinv3            = _mm256_mul_ps(rinvsq02,rinv02);
 877             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
 878             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
 879             felec            = _mm256_mul_ps(qq02,felec);
 880             pmecorrV         = gmx_mm256_pmecorrV_ps(zeta2);
 881             pmecorrV         = _mm256_mul_ps(pmecorrV,beta);
 882             velec            = _mm256_sub_ps(rinv02,pmecorrV);
 883             velec            = _mm256_mul_ps(qq02,velec);
 884
 885             /* Update potential sum for this i atom from the interaction with this j atom. */
 886             velec            = _mm256_andnot_ps(dummy_mask,velec);
 887             velecsum         = _mm256_add_ps(velecsum,velec);
 888
 889             fscal            = felec;
 890
 891             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
 892
 893             /* Calculate temporary vectorial force */
 894             tx               = _mm256_mul_ps(fscal,dx02);
 895             ty               = _mm256_mul_ps(fscal,dy02);
 896             tz               = _mm256_mul_ps(fscal,dz02);
 897
 898             /* Update vectorial force */
 899             fix0             = _mm256_add_ps(fix0,tx);
 900             fiy0             = _mm256_add_ps(fiy0,ty);
 901             fiz0             = _mm256_add_ps(fiz0,tz);
 902
 903             fjx2             = _mm256_add_ps(fjx2,tx);
 904             fjy2             = _mm256_add_ps(fjy2,ty);
 905             fjz2             = _mm256_add_ps(fjz2,tz);
 906
 907             /**************************
 908              * CALCULATE INTERACTIONS *
 909              **************************/
 910
 911             r10              = _mm256_mul_ps(rsq10,rinv10);
 912             r10              = _mm256_andnot_ps(dummy_mask,r10);
 913
 914             /* EWALD ELECTROSTATICS */
 915
 916             /* Analytical PME correction */
 917             zeta2            = _mm256_mul_ps(beta2,rsq10);
 918             rinv3            = _mm256_mul_ps(rinvsq10,rinv10);
 919             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
 920             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
 921             felec            = _mm256_mul_ps(qq10,felec);
 922             pmecorrV         = gmx_mm256_pmecorrV_ps(zeta2);
 923             pmecorrV         = _mm256_mul_ps(pmecorrV,beta);
 924             velec            = _mm256_sub_ps(rinv10,pmecorrV);
 925             velec            = _mm256_mul_ps(qq10,velec);
 926
 927             /* Update potential sum for this i atom from the interaction with this j atom. */
 928             velec            = _mm256_andnot_ps(dummy_mask,velec);
 929             velecsum         = _mm256_add_ps(velecsum,velec);
 930
 931             fscal            = felec;
 932
 933             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
 934
 935             /* Calculate temporary vectorial force */
 936             tx               = _mm256_mul_ps(fscal,dx10);
 937             ty               = _mm256_mul_ps(fscal,dy10);
 938             tz               = _mm256_mul_ps(fscal,dz10);
 939
 940             /* Update vectorial force */
 941             fix1             = _mm256_add_ps(fix1,tx);
 942             fiy1             = _mm256_add_ps(fiy1,ty);
 943             fiz1             = _mm256_add_ps(fiz1,tz);
 944
 945             fjx0             = _mm256_add_ps(fjx0,tx);
 946             fjy0             = _mm256_add_ps(fjy0,ty);
 947             fjz0             = _mm256_add_ps(fjz0,tz);
 948
 949             /**************************
 950              * CALCULATE INTERACTIONS *
 951              **************************/
 952
 953             r11              = _mm256_mul_ps(rsq11,rinv11);
 954             r11              = _mm256_andnot_ps(dummy_mask,r11);
 955
 956             /* EWALD ELECTROSTATICS */
 957
 958             /* Analytical PME correction */
 959             zeta2            = _mm256_mul_ps(beta2,rsq11);
 960             rinv3            = _mm256_mul_ps(rinvsq11,rinv11);
 961             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
 962             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
 963             felec            = _mm256_mul_ps(qq11,felec);
 964             pmecorrV         = gmx_mm256_pmecorrV_ps(zeta2);
 965             pmecorrV         = _mm256_mul_ps(pmecorrV,beta);
 966             velec            = _mm256_sub_ps(rinv11,pmecorrV);
 967             velec            = _mm256_mul_ps(qq11,velec);
 968
 969             /* Update potential sum for this i atom from the interaction with this j atom. */
 970             velec            = _mm256_andnot_ps(dummy_mask,velec);
 971             velecsum         = _mm256_add_ps(velecsum,velec);
 972
 973             fscal            = felec;
 974
 975             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
 976
 977             /* Calculate temporary vectorial force */
 978             tx               = _mm256_mul_ps(fscal,dx11);
 979             ty               = _mm256_mul_ps(fscal,dy11);
 980             tz               = _mm256_mul_ps(fscal,dz11);
 981
 982             /* Update vectorial force */
 983             fix1             = _mm256_add_ps(fix1,tx);
 984             fiy1             = _mm256_add_ps(fiy1,ty);
 985             fiz1             = _mm256_add_ps(fiz1,tz);
 986
 987             fjx1             = _mm256_add_ps(fjx1,tx);
 988             fjy1             = _mm256_add_ps(fjy1,ty);
 989             fjz1             = _mm256_add_ps(fjz1,tz);
 990
 991             /**************************
 992              * CALCULATE INTERACTIONS *
 993              **************************/
 994
 995             r12              = _mm256_mul_ps(rsq12,rinv12);
 996             r12              = _mm256_andnot_ps(dummy_mask,r12);
 997
 998             /* EWALD ELECTROSTATICS */
 999
1000             /* Analytical PME correction */
1001             zeta2            = _mm256_mul_ps(beta2,rsq12);
1002             rinv3            = _mm256_mul_ps(rinvsq12,rinv12);
1003             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
1004             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
1005             felec            = _mm256_mul_ps(qq12,felec);
1006             pmecorrV         = gmx_mm256_pmecorrV_ps(zeta2);
1007             pmecorrV         = _mm256_mul_ps(pmecorrV,beta);
1008             velec            = _mm256_sub_ps(rinv12,pmecorrV);
1009             velec            = _mm256_mul_ps(qq12,velec);
1010
1011             /* Update potential sum for this i atom from the interaction with this j atom. */
1012             velec            = _mm256_andnot_ps(dummy_mask,velec);
1013             velecsum         = _mm256_add_ps(velecsum,velec);
1014
1015             fscal            = felec;
1016
1017             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
1018
1019             /* Calculate temporary vectorial force */
1020             tx               = _mm256_mul_ps(fscal,dx12);
1021             ty               = _mm256_mul_ps(fscal,dy12);
1022             tz               = _mm256_mul_ps(fscal,dz12);
1023
1024             /* Update vectorial force */
1025             fix1             = _mm256_add_ps(fix1,tx);
1026             fiy1             = _mm256_add_ps(fiy1,ty);
1027             fiz1             = _mm256_add_ps(fiz1,tz);
1028
1029             fjx2             = _mm256_add_ps(fjx2,tx);
1030             fjy2             = _mm256_add_ps(fjy2,ty);
1031             fjz2             = _mm256_add_ps(fjz2,tz);
1032
1033             /**************************
1034              * CALCULATE INTERACTIONS *
1035              **************************/
1036
1037             r20              = _mm256_mul_ps(rsq20,rinv20);
1038             r20              = _mm256_andnot_ps(dummy_mask,r20);
1039
1040             /* EWALD ELECTROSTATICS */
1041
1042             /* Analytical PME correction */
1043             zeta2            = _mm256_mul_ps(beta2,rsq20);
1044             rinv3            = _mm256_mul_ps(rinvsq20,rinv20);
1045             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
1046             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
1047             felec            = _mm256_mul_ps(qq20,felec);
1048             pmecorrV         = gmx_mm256_pmecorrV_ps(zeta2);
1049             pmecorrV         = _mm256_mul_ps(pmecorrV,beta);
1050             velec            = _mm256_sub_ps(rinv20,pmecorrV);
1051             velec            = _mm256_mul_ps(qq20,velec);
1052
1053             /* Update potential sum for this i atom from the interaction with this j atom. */
1054             velec            = _mm256_andnot_ps(dummy_mask,velec);
1055             velecsum         = _mm256_add_ps(velecsum,velec);
1056
1057             fscal            = felec;
1058
1059             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
1060
1061             /* Calculate temporary vectorial force */
1062             tx               = _mm256_mul_ps(fscal,dx20);
1063             ty               = _mm256_mul_ps(fscal,dy20);
1064             tz               = _mm256_mul_ps(fscal,dz20);
1065
1066             /* Update vectorial force */
1067             fix2             = _mm256_add_ps(fix2,tx);
1068             fiy2             = _mm256_add_ps(fiy2,ty);
1069             fiz2             = _mm256_add_ps(fiz2,tz);
1070
1071             fjx0             = _mm256_add_ps(fjx0,tx);
1072             fjy0             = _mm256_add_ps(fjy0,ty);
1073             fjz0             = _mm256_add_ps(fjz0,tz);
1074
1075             /**************************
1076              * CALCULATE INTERACTIONS *
1077              **************************/
1078
1079             r21              = _mm256_mul_ps(rsq21,rinv21);
1080             r21              = _mm256_andnot_ps(dummy_mask,r21);
1081
1082             /* EWALD ELECTROSTATICS */
1083
1084             /* Analytical PME correction */
1085             zeta2            = _mm256_mul_ps(beta2,rsq21);
1086             rinv3            = _mm256_mul_ps(rinvsq21,rinv21);
1087             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
1088             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
1089             felec            = _mm256_mul_ps(qq21,felec);
1090             pmecorrV         = gmx_mm256_pmecorrV_ps(zeta2);
1091             pmecorrV         = _mm256_mul_ps(pmecorrV,beta);
1092             velec            = _mm256_sub_ps(rinv21,pmecorrV);
1093             velec            = _mm256_mul_ps(qq21,velec);
1094
1095             /* Update potential sum for this i atom from the interaction with this j atom. */
1096             velec            = _mm256_andnot_ps(dummy_mask,velec);
1097             velecsum         = _mm256_add_ps(velecsum,velec);
1098
1099             fscal            = felec;
1100
1101             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
1102
1103             /* Calculate temporary vectorial force */
1104             tx               = _mm256_mul_ps(fscal,dx21);
1105             ty               = _mm256_mul_ps(fscal,dy21);
1106             tz               = _mm256_mul_ps(fscal,dz21);
1107
1108             /* Update vectorial force */
1109             fix2             = _mm256_add_ps(fix2,tx);
1110             fiy2             = _mm256_add_ps(fiy2,ty);
1111             fiz2             = _mm256_add_ps(fiz2,tz);
1112
1113             fjx1             = _mm256_add_ps(fjx1,tx);
1114             fjy1             = _mm256_add_ps(fjy1,ty);
1115             fjz1             = _mm256_add_ps(fjz1,tz);
1116
1117             /**************************
1118              * CALCULATE INTERACTIONS *
1119              **************************/
1120
1121             r22              = _mm256_mul_ps(rsq22,rinv22);
1122             r22              = _mm256_andnot_ps(dummy_mask,r22);
1123
1124             /* EWALD ELECTROSTATICS */
1125
1126             /* Analytical PME correction */
1127             zeta2            = _mm256_mul_ps(beta2,rsq22);
1128             rinv3            = _mm256_mul_ps(rinvsq22,rinv22);
1129             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
1130             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
1131             felec            = _mm256_mul_ps(qq22,felec);
1132             pmecorrV         = gmx_mm256_pmecorrV_ps(zeta2);
1133             pmecorrV         = _mm256_mul_ps(pmecorrV,beta);
1134             velec            = _mm256_sub_ps(rinv22,pmecorrV);
1135             velec            = _mm256_mul_ps(qq22,velec);
1136
1137             /* Update potential sum for this i atom from the interaction with this j atom. */
1138             velec            = _mm256_andnot_ps(dummy_mask,velec);
1139             velecsum         = _mm256_add_ps(velecsum,velec);
1140
1141             fscal            = felec;
1142
1143             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
1144
1145             /* Calculate temporary vectorial force */
1146             tx               = _mm256_mul_ps(fscal,dx22);
1147             ty               = _mm256_mul_ps(fscal,dy22);
1148             tz               = _mm256_mul_ps(fscal,dz22);
1149
1150             /* Update vectorial force */
1151             fix2             = _mm256_add_ps(fix2,tx);
1152             fiy2             = _mm256_add_ps(fiy2,ty);
1153             fiz2             = _mm256_add_ps(fiz2,tz);
1154
1155             fjx2             = _mm256_add_ps(fjx2,tx);
1156             fjy2             = _mm256_add_ps(fjy2,ty);
1157             fjz2             = _mm256_add_ps(fjz2,tz);
1158
1159             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
1160             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
1161             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
1162             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
1163             fjptrE             = (jnrlistE>=0) ? f+j_coord_offsetE : scratch;
1164             fjptrF             = (jnrlistF>=0) ? f+j_coord_offsetF : scratch;
1165             fjptrG             = (jnrlistG>=0) ? f+j_coord_offsetG : scratch;
1166             fjptrH             = (jnrlistH>=0) ? f+j_coord_offsetH : scratch;
1167
1168             gmx_mm256_decrement_3rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,
1169                                                       fjx0,fjy0,fjz0,fjx1,fjy1,fjz1,fjx2,fjy2,fjz2);
1170
1171             /* Inner loop uses 765 flops */
1172         }
1173
1174         /* End of innermost loop */
1175
1176         gmx_mm256_update_iforce_3atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
1177                                                  f+i_coord_offset,fshift+i_shift_offset);
1178
1179         ggid                        = gid[iidx];
1180         /* Update potential energies */
1181         gmx_mm256_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid);
1182
1183         /* Increment number of inner iterations */
1184         inneriter                  += j_index_end - j_index_start;
1185
1186         /* Outer loop uses 19 flops */
1187     }
1188
1189     /* Increment number of outer iterations */
1190     outeriter        += nri;
1191
1192     /* Update outer/inner flops */
1193
1194     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_W3W3_VF,outeriter*19 + inneriter*765);
1195 }
1196 /*
1197  * Gromacs nonbonded kernel:   nb_kernel_ElecEw_VdwNone_GeomW3W3_F_avx_256_single
1198  * Electrostatics interaction: Ewald
1199  * VdW interaction:            None
1200  * Geometry:                   Water3-Water3
1201  * Calculate force/pot:        Force
1202  */
1203 void
1204 nb_kernel_ElecEw_VdwNone_GeomW3W3_F_avx_256_single
1205                     (t_nblist                    * gmx_restrict       nlist,
1206                      rvec                        * gmx_restrict          xx,
1207                      rvec                        * gmx_restrict          ff,
1208                      t_forcerec                  * gmx_restrict          fr,
1209                      t_mdatoms                   * gmx_restrict     mdatoms,
1210                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
1211                      t_nrnb                      * gmx_restrict        nrnb)
1212 {
1213     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
1214      * just 0 for non-waters.
1215      * Suffixes A,B,C,D,E,F,G,H refer to j loop unrolling done with AVX, e.g. for the eight different
1216      * jnr indices corresponding to data put in the four positions in the SIMD register.
1217      */
1218     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
1219     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
1220     int              jnrA,jnrB,jnrC,jnrD;
1221     int              jnrE,jnrF,jnrG,jnrH;
1222     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
1223     int              jnrlistE,jnrlistF,jnrlistG,jnrlistH;
1224     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
1225     int              j_coord_offsetE,j_coord_offsetF,j_coord_offsetG,j_coord_offsetH;
1226     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
1227     real             rcutoff_scalar;
1228     real             *shiftvec,*fshift,*x,*f;
1229     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD,*fjptrE,*fjptrF,*fjptrG,*fjptrH;
1230     real             scratch[4*DIM];
1231     __m256           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
1232     real *           vdwioffsetptr0;
1233     __m256           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
1234     real *           vdwioffsetptr1;
1235     __m256           ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
1236     real *           vdwioffsetptr2;
1237     __m256           ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
1238     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D,vdwjidx0E,vdwjidx0F,vdwjidx0G,vdwjidx0H;
1239     __m256           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
1240     int              vdwjidx1A,vdwjidx1B,vdwjidx1C,vdwjidx1D,vdwjidx1E,vdwjidx1F,vdwjidx1G,vdwjidx1H;
1241     __m256           jx1,jy1,jz1,fjx1,fjy1,fjz1,jq1,isaj1;
1242     int              vdwjidx2A,vdwjidx2B,vdwjidx2C,vdwjidx2D,vdwjidx2E,vdwjidx2F,vdwjidx2G,vdwjidx2H;
1243     __m256           jx2,jy2,jz2,fjx2,fjy2,fjz2,jq2,isaj2;
1244     __m256           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
1245     __m256           dx01,dy01,dz01,rsq01,rinv01,rinvsq01,r01,qq01,c6_01,c12_01;
1246     __m256           dx02,dy02,dz02,rsq02,rinv02,rinvsq02,r02,qq02,c6_02,c12_02;
1247     __m256           dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
1248     __m256           dx11,dy11,dz11,rsq11,rinv11,rinvsq11,r11,qq11,c6_11,c12_11;
1249     __m256           dx12,dy12,dz12,rsq12,rinv12,rinvsq12,r12,qq12,c6_12,c12_12;
1250     __m256           dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
1251     __m256           dx21,dy21,dz21,rsq21,rinv21,rinvsq21,r21,qq21,c6_21,c12_21;
1252     __m256           dx22,dy22,dz22,rsq22,rinv22,rinvsq22,r22,qq22,c6_22,c12_22;
1253     __m256           velec,felec,velecsum,facel,crf,krf,krf2;
1254     real             *charge;
1255     __m256i          ewitab;
1256     __m128i          ewitab_lo,ewitab_hi;
1257     __m256           ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
1258     __m256           beta,beta2,beta3,zeta2,pmecorrF,pmecorrV,rinv3;
1259     real             *ewtab;
1260     __m256           dummy_mask,cutoff_mask;
1261     __m256           signbit = _mm256_castsi256_ps( _mm256_set1_epi32(0x80000000) );
1262     __m256           one     = _mm256_set1_ps(1.0);
1263     __m256           two     = _mm256_set1_ps(2.0);
1264     x                = xx[0];
1265     f                = ff[0];
1266
1267     nri              = nlist->nri;
1268     iinr             = nlist->iinr;
1269     jindex           = nlist->jindex;
1270     jjnr             = nlist->jjnr;
1271     shiftidx         = nlist->shift;
1272     gid              = nlist->gid;
1273     shiftvec         = fr->shift_vec[0];
1274     fshift           = fr->fshift[0];
1275     facel            = _mm256_set1_ps(fr->epsfac);
1276     charge           = mdatoms->chargeA;
1277
1278     sh_ewald         = _mm256_set1_ps(fr->ic->sh_ewald);
1279     beta             = _mm256_set1_ps(fr->ic->ewaldcoeff_q);
1280     beta2            = _mm256_mul_ps(beta,beta);
1281     beta3            = _mm256_mul_ps(beta,beta2);
1282
1283     ewtab            = fr->ic->tabq_coul_F;
1284     ewtabscale       = _mm256_set1_ps(fr->ic->tabq_scale);
1285     ewtabhalfspace   = _mm256_set1_ps(0.5/fr->ic->tabq_scale);
1286
1287     /* Setup water-specific parameters */
1288     inr              = nlist->iinr[0];
1289     iq0              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+0]));
1290     iq1              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+1]));
1291     iq2              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+2]));
1292
1293     jq0              = _mm256_set1_ps(charge[inr+0]);
1294     jq1              = _mm256_set1_ps(charge[inr+1]);
1295     jq2              = _mm256_set1_ps(charge[inr+2]);
1296     qq00             = _mm256_mul_ps(iq0,jq0);
1297     qq01             = _mm256_mul_ps(iq0,jq1);
1298     qq02             = _mm256_mul_ps(iq0,jq2);
1299     qq10             = _mm256_mul_ps(iq1,jq0);
1300     qq11             = _mm256_mul_ps(iq1,jq1);
1301     qq12             = _mm256_mul_ps(iq1,jq2);
1302     qq20             = _mm256_mul_ps(iq2,jq0);
1303     qq21             = _mm256_mul_ps(iq2,jq1);
1304     qq22             = _mm256_mul_ps(iq2,jq2);
1305
1306     /* Avoid stupid compiler warnings */
1307     jnrA = jnrB = jnrC = jnrD = jnrE = jnrF = jnrG = jnrH = 0;
1308     j_coord_offsetA = 0;
1309     j_coord_offsetB = 0;
1310     j_coord_offsetC = 0;
1311     j_coord_offsetD = 0;
1312     j_coord_offsetE = 0;
1313     j_coord_offsetF = 0;
1314     j_coord_offsetG = 0;
1315     j_coord_offsetH = 0;
1316
1317     outeriter        = 0;
1318     inneriter        = 0;
1319
1320     for(iidx=0;iidx<4*DIM;iidx++)
1321     {
1322         scratch[iidx] = 0.0;
1323     }
1324
1325     /* Start outer loop over neighborlists */
1326     for(iidx=0; iidx<nri; iidx++)
1327     {
1328         /* Load shift vector for this list */
1329         i_shift_offset   = DIM*shiftidx[iidx];
1330
1331         /* Load limits for loop over neighbors */
1332         j_index_start    = jindex[iidx];
1333         j_index_end      = jindex[iidx+1];
1334
1335         /* Get outer coordinate index */
1336         inr              = iinr[iidx];
1337         i_coord_offset   = DIM*inr;
1338
1339         /* Load i particle coords and add shift vector */
1340         gmx_mm256_load_shift_and_3rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
1341                                                     &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
1342
1343         fix0             = _mm256_setzero_ps();
1344         fiy0             = _mm256_setzero_ps();
1345         fiz0             = _mm256_setzero_ps();
1346         fix1             = _mm256_setzero_ps();
1347         fiy1             = _mm256_setzero_ps();
1348         fiz1             = _mm256_setzero_ps();
1349         fix2             = _mm256_setzero_ps();
1350         fiy2             = _mm256_setzero_ps();
1351         fiz2             = _mm256_setzero_ps();
1352
1353         /* Start inner kernel loop */
1354         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+7]>=0; jidx+=8)
1355         {
1356
1357             /* Get j neighbor index, and coordinate index */
1358             jnrA             = jjnr[jidx];
1359             jnrB             = jjnr[jidx+1];
1360             jnrC             = jjnr[jidx+2];
1361             jnrD             = jjnr[jidx+3];
1362             jnrE             = jjnr[jidx+4];
1363             jnrF             = jjnr[jidx+5];
1364             jnrG             = jjnr[jidx+6];
1365             jnrH             = jjnr[jidx+7];
1366             j_coord_offsetA  = DIM*jnrA;
1367             j_coord_offsetB  = DIM*jnrB;
1368             j_coord_offsetC  = DIM*jnrC;
1369             j_coord_offsetD  = DIM*jnrD;
1370             j_coord_offsetE  = DIM*jnrE;
1371             j_coord_offsetF  = DIM*jnrF;
1372             j_coord_offsetG  = DIM*jnrG;
1373             j_coord_offsetH  = DIM*jnrH;
1374
1375             /* load j atom coordinates */
1376             gmx_mm256_load_3rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
1377                                                  x+j_coord_offsetC,x+j_coord_offsetD,
1378                                                  x+j_coord_offsetE,x+j_coord_offsetF,
1379                                                  x+j_coord_offsetG,x+j_coord_offsetH,
1380                                               &jx0,&jy0,&jz0,&jx1,&jy1,&jz1,&jx2,&jy2,&jz2);
1381
1382             /* Calculate displacement vector */
1383             dx00             = _mm256_sub_ps(ix0,jx0);
1384             dy00             = _mm256_sub_ps(iy0,jy0);
1385             dz00             = _mm256_sub_ps(iz0,jz0);
1386             dx01             = _mm256_sub_ps(ix0,jx1);
1387             dy01             = _mm256_sub_ps(iy0,jy1);
1388             dz01             = _mm256_sub_ps(iz0,jz1);
1389             dx02             = _mm256_sub_ps(ix0,jx2);
1390             dy02             = _mm256_sub_ps(iy0,jy2);
1391             dz02             = _mm256_sub_ps(iz0,jz2);
1392             dx10             = _mm256_sub_ps(ix1,jx0);
1393             dy10             = _mm256_sub_ps(iy1,jy0);
1394             dz10             = _mm256_sub_ps(iz1,jz0);
1395             dx11             = _mm256_sub_ps(ix1,jx1);
1396             dy11             = _mm256_sub_ps(iy1,jy1);
1397             dz11             = _mm256_sub_ps(iz1,jz1);
1398             dx12             = _mm256_sub_ps(ix1,jx2);
1399             dy12             = _mm256_sub_ps(iy1,jy2);
1400             dz12             = _mm256_sub_ps(iz1,jz2);
1401             dx20             = _mm256_sub_ps(ix2,jx0);
1402             dy20             = _mm256_sub_ps(iy2,jy0);
1403             dz20             = _mm256_sub_ps(iz2,jz0);
1404             dx21             = _mm256_sub_ps(ix2,jx1);
1405             dy21             = _mm256_sub_ps(iy2,jy1);
1406             dz21             = _mm256_sub_ps(iz2,jz1);
1407             dx22             = _mm256_sub_ps(ix2,jx2);
1408             dy22             = _mm256_sub_ps(iy2,jy2);
1409             dz22             = _mm256_sub_ps(iz2,jz2);
1410
1411             /* Calculate squared distance and things based on it */
1412             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
1413             rsq01            = gmx_mm256_calc_rsq_ps(dx01,dy01,dz01);
1414             rsq02            = gmx_mm256_calc_rsq_ps(dx02,dy02,dz02);
1415             rsq10            = gmx_mm256_calc_rsq_ps(dx10,dy10,dz10);
1416             rsq11            = gmx_mm256_calc_rsq_ps(dx11,dy11,dz11);
1417             rsq12            = gmx_mm256_calc_rsq_ps(dx12,dy12,dz12);
1418             rsq20            = gmx_mm256_calc_rsq_ps(dx20,dy20,dz20);
1419             rsq21            = gmx_mm256_calc_rsq_ps(dx21,dy21,dz21);
1420             rsq22            = gmx_mm256_calc_rsq_ps(dx22,dy22,dz22);
1421
1422             rinv00           = gmx_mm256_invsqrt_ps(rsq00);
1423             rinv01           = gmx_mm256_invsqrt_ps(rsq01);
1424             rinv02           = gmx_mm256_invsqrt_ps(rsq02);
1425             rinv10           = gmx_mm256_invsqrt_ps(rsq10);
1426             rinv11           = gmx_mm256_invsqrt_ps(rsq11);
1427             rinv12           = gmx_mm256_invsqrt_ps(rsq12);
1428             rinv20           = gmx_mm256_invsqrt_ps(rsq20);
1429             rinv21           = gmx_mm256_invsqrt_ps(rsq21);
1430             rinv22           = gmx_mm256_invsqrt_ps(rsq22);
1431
1432             rinvsq00         = _mm256_mul_ps(rinv00,rinv00);
1433             rinvsq01         = _mm256_mul_ps(rinv01,rinv01);
1434             rinvsq02         = _mm256_mul_ps(rinv02,rinv02);
1435             rinvsq10         = _mm256_mul_ps(rinv10,rinv10);
1436             rinvsq11         = _mm256_mul_ps(rinv11,rinv11);
1437             rinvsq12         = _mm256_mul_ps(rinv12,rinv12);
1438             rinvsq20         = _mm256_mul_ps(rinv20,rinv20);
1439             rinvsq21         = _mm256_mul_ps(rinv21,rinv21);
1440             rinvsq22         = _mm256_mul_ps(rinv22,rinv22);
1441
1442             fjx0             = _mm256_setzero_ps();
1443             fjy0             = _mm256_setzero_ps();
1444             fjz0             = _mm256_setzero_ps();
1445             fjx1             = _mm256_setzero_ps();
1446             fjy1             = _mm256_setzero_ps();
1447             fjz1             = _mm256_setzero_ps();
1448             fjx2             = _mm256_setzero_ps();
1449             fjy2             = _mm256_setzero_ps();
1450             fjz2             = _mm256_setzero_ps();
1451
1452             /**************************
1453              * CALCULATE INTERACTIONS *
1454              **************************/
1455
1456             r00              = _mm256_mul_ps(rsq00,rinv00);
1457
1458             /* EWALD ELECTROSTATICS */
1459
1460             /* Analytical PME correction */
1461             zeta2            = _mm256_mul_ps(beta2,rsq00);
1462             rinv3            = _mm256_mul_ps(rinvsq00,rinv00);
1463             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
1464             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
1465             felec            = _mm256_mul_ps(qq00,felec);
1466
1467             fscal            = felec;
1468
1469             /* Calculate temporary vectorial force */
1470             tx               = _mm256_mul_ps(fscal,dx00);
1471             ty               = _mm256_mul_ps(fscal,dy00);
1472             tz               = _mm256_mul_ps(fscal,dz00);
1473
1474             /* Update vectorial force */
1475             fix0             = _mm256_add_ps(fix0,tx);
1476             fiy0             = _mm256_add_ps(fiy0,ty);
1477             fiz0             = _mm256_add_ps(fiz0,tz);
1478
1479             fjx0             = _mm256_add_ps(fjx0,tx);
1480             fjy0             = _mm256_add_ps(fjy0,ty);
1481             fjz0             = _mm256_add_ps(fjz0,tz);
1482
1483             /**************************
1484              * CALCULATE INTERACTIONS *
1485              **************************/
1486
1487             r01              = _mm256_mul_ps(rsq01,rinv01);
1488
1489             /* EWALD ELECTROSTATICS */
1490
1491             /* Analytical PME correction */
1492             zeta2            = _mm256_mul_ps(beta2,rsq01);
1493             rinv3            = _mm256_mul_ps(rinvsq01,rinv01);
1494             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
1495             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
1496             felec            = _mm256_mul_ps(qq01,felec);
1497
1498             fscal            = felec;
1499
1500             /* Calculate temporary vectorial force */
1501             tx               = _mm256_mul_ps(fscal,dx01);
1502             ty               = _mm256_mul_ps(fscal,dy01);
1503             tz               = _mm256_mul_ps(fscal,dz01);
1504
1505             /* Update vectorial force */
1506             fix0             = _mm256_add_ps(fix0,tx);
1507             fiy0             = _mm256_add_ps(fiy0,ty);
1508             fiz0             = _mm256_add_ps(fiz0,tz);
1509
1510             fjx1             = _mm256_add_ps(fjx1,tx);
1511             fjy1             = _mm256_add_ps(fjy1,ty);
1512             fjz1             = _mm256_add_ps(fjz1,tz);
1513
1514             /**************************
1515              * CALCULATE INTERACTIONS *
1516              **************************/
1517
1518             r02              = _mm256_mul_ps(rsq02,rinv02);
1519
1520             /* EWALD ELECTROSTATICS */
1521
1522             /* Analytical PME correction */
1523             zeta2            = _mm256_mul_ps(beta2,rsq02);
1524             rinv3            = _mm256_mul_ps(rinvsq02,rinv02);
1525             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
1526             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
1527             felec            = _mm256_mul_ps(qq02,felec);
1528
1529             fscal            = felec;
1530
1531             /* Calculate temporary vectorial force */
1532             tx               = _mm256_mul_ps(fscal,dx02);
1533             ty               = _mm256_mul_ps(fscal,dy02);
1534             tz               = _mm256_mul_ps(fscal,dz02);
1535
1536             /* Update vectorial force */
1537             fix0             = _mm256_add_ps(fix0,tx);
1538             fiy0             = _mm256_add_ps(fiy0,ty);
1539             fiz0             = _mm256_add_ps(fiz0,tz);
1540
1541             fjx2             = _mm256_add_ps(fjx2,tx);
1542             fjy2             = _mm256_add_ps(fjy2,ty);
1543             fjz2             = _mm256_add_ps(fjz2,tz);
1544
1545             /**************************
1546              * CALCULATE INTERACTIONS *
1547              **************************/
1548
1549             r10              = _mm256_mul_ps(rsq10,rinv10);
1550
1551             /* EWALD ELECTROSTATICS */
1552
1553             /* Analytical PME correction */
1554             zeta2            = _mm256_mul_ps(beta2,rsq10);
1555             rinv3            = _mm256_mul_ps(rinvsq10,rinv10);
1556             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
1557             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
1558             felec            = _mm256_mul_ps(qq10,felec);
1559
1560             fscal            = felec;
1561
1562             /* Calculate temporary vectorial force */
1563             tx               = _mm256_mul_ps(fscal,dx10);
1564             ty               = _mm256_mul_ps(fscal,dy10);
1565             tz               = _mm256_mul_ps(fscal,dz10);
1566
1567             /* Update vectorial force */
1568             fix1             = _mm256_add_ps(fix1,tx);
1569             fiy1             = _mm256_add_ps(fiy1,ty);
1570             fiz1             = _mm256_add_ps(fiz1,tz);
1571
1572             fjx0             = _mm256_add_ps(fjx0,tx);
1573             fjy0             = _mm256_add_ps(fjy0,ty);
1574             fjz0             = _mm256_add_ps(fjz0,tz);
1575
1576             /**************************
1577              * CALCULATE INTERACTIONS *
1578              **************************/
1579
1580             r11              = _mm256_mul_ps(rsq11,rinv11);
1581
1582             /* EWALD ELECTROSTATICS */
1583
1584             /* Analytical PME correction */
1585             zeta2            = _mm256_mul_ps(beta2,rsq11);
1586             rinv3            = _mm256_mul_ps(rinvsq11,rinv11);
1587             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
1588             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
1589             felec            = _mm256_mul_ps(qq11,felec);
1590
1591             fscal            = felec;
1592
1593             /* Calculate temporary vectorial force */
1594             tx               = _mm256_mul_ps(fscal,dx11);
1595             ty               = _mm256_mul_ps(fscal,dy11);
1596             tz               = _mm256_mul_ps(fscal,dz11);
1597
1598             /* Update vectorial force */
1599             fix1             = _mm256_add_ps(fix1,tx);
1600             fiy1             = _mm256_add_ps(fiy1,ty);
1601             fiz1             = _mm256_add_ps(fiz1,tz);
1602
1603             fjx1             = _mm256_add_ps(fjx1,tx);
1604             fjy1             = _mm256_add_ps(fjy1,ty);
1605             fjz1             = _mm256_add_ps(fjz1,tz);
1606
1607             /**************************
1608              * CALCULATE INTERACTIONS *
1609              **************************/
1610
1611             r12              = _mm256_mul_ps(rsq12,rinv12);
1612
1613             /* EWALD ELECTROSTATICS */
1614
1615             /* Analytical PME correction */
1616             zeta2            = _mm256_mul_ps(beta2,rsq12);
1617             rinv3            = _mm256_mul_ps(rinvsq12,rinv12);
1618             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
1619             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
1620             felec            = _mm256_mul_ps(qq12,felec);
1621
1622             fscal            = felec;
1623
1624             /* Calculate temporary vectorial force */
1625             tx               = _mm256_mul_ps(fscal,dx12);
1626             ty               = _mm256_mul_ps(fscal,dy12);
1627             tz               = _mm256_mul_ps(fscal,dz12);
1628
1629             /* Update vectorial force */
1630             fix1             = _mm256_add_ps(fix1,tx);
1631             fiy1             = _mm256_add_ps(fiy1,ty);
1632             fiz1             = _mm256_add_ps(fiz1,tz);
1633
1634             fjx2             = _mm256_add_ps(fjx2,tx);
1635             fjy2             = _mm256_add_ps(fjy2,ty);
1636             fjz2             = _mm256_add_ps(fjz2,tz);
1637
1638             /**************************
1639              * CALCULATE INTERACTIONS *
1640              **************************/
1641
1642             r20              = _mm256_mul_ps(rsq20,rinv20);
1643
1644             /* EWALD ELECTROSTATICS */
1645
1646             /* Analytical PME correction */
1647             zeta2            = _mm256_mul_ps(beta2,rsq20);
1648             rinv3            = _mm256_mul_ps(rinvsq20,rinv20);
1649             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
1650             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
1651             felec            = _mm256_mul_ps(qq20,felec);
1652
1653             fscal            = felec;
1654
1655             /* Calculate temporary vectorial force */
1656             tx               = _mm256_mul_ps(fscal,dx20);
1657             ty               = _mm256_mul_ps(fscal,dy20);
1658             tz               = _mm256_mul_ps(fscal,dz20);
1659
1660             /* Update vectorial force */
1661             fix2             = _mm256_add_ps(fix2,tx);
1662             fiy2             = _mm256_add_ps(fiy2,ty);
1663             fiz2             = _mm256_add_ps(fiz2,tz);
1664
1665             fjx0             = _mm256_add_ps(fjx0,tx);
1666             fjy0             = _mm256_add_ps(fjy0,ty);
1667             fjz0             = _mm256_add_ps(fjz0,tz);
1668
1669             /**************************
1670              * CALCULATE INTERACTIONS *
1671              **************************/
1672
1673             r21              = _mm256_mul_ps(rsq21,rinv21);
1674
1675             /* EWALD ELECTROSTATICS */
1676
1677             /* Analytical PME correction */
1678             zeta2            = _mm256_mul_ps(beta2,rsq21);
1679             rinv3            = _mm256_mul_ps(rinvsq21,rinv21);
1680             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
1681             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
1682             felec            = _mm256_mul_ps(qq21,felec);
1683
1684             fscal            = felec;
1685
1686             /* Calculate temporary vectorial force */
1687             tx               = _mm256_mul_ps(fscal,dx21);
1688             ty               = _mm256_mul_ps(fscal,dy21);
1689             tz               = _mm256_mul_ps(fscal,dz21);
1690
1691             /* Update vectorial force */
1692             fix2             = _mm256_add_ps(fix2,tx);
1693             fiy2             = _mm256_add_ps(fiy2,ty);
1694             fiz2             = _mm256_add_ps(fiz2,tz);
1695
1696             fjx1             = _mm256_add_ps(fjx1,tx);
1697             fjy1             = _mm256_add_ps(fjy1,ty);
1698             fjz1             = _mm256_add_ps(fjz1,tz);
1699
1700             /**************************
1701              * CALCULATE INTERACTIONS *
1702              **************************/
1703
1704             r22              = _mm256_mul_ps(rsq22,rinv22);
1705
1706             /* EWALD ELECTROSTATICS */
1707
1708             /* Analytical PME correction */
1709             zeta2            = _mm256_mul_ps(beta2,rsq22);
1710             rinv3            = _mm256_mul_ps(rinvsq22,rinv22);
1711             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
1712             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
1713             felec            = _mm256_mul_ps(qq22,felec);
1714
1715             fscal            = felec;
1716
1717             /* Calculate temporary vectorial force */
1718             tx               = _mm256_mul_ps(fscal,dx22);
1719             ty               = _mm256_mul_ps(fscal,dy22);
1720             tz               = _mm256_mul_ps(fscal,dz22);
1721
1722             /* Update vectorial force */
1723             fix2             = _mm256_add_ps(fix2,tx);
1724             fiy2             = _mm256_add_ps(fiy2,ty);
1725             fiz2             = _mm256_add_ps(fiz2,tz);
1726
1727             fjx2             = _mm256_add_ps(fjx2,tx);
1728             fjy2             = _mm256_add_ps(fjy2,ty);
1729             fjz2             = _mm256_add_ps(fjz2,tz);
1730
1731             fjptrA             = f+j_coord_offsetA;
1732             fjptrB             = f+j_coord_offsetB;
1733             fjptrC             = f+j_coord_offsetC;
1734             fjptrD             = f+j_coord_offsetD;
1735             fjptrE             = f+j_coord_offsetE;
1736             fjptrF             = f+j_coord_offsetF;
1737             fjptrG             = f+j_coord_offsetG;
1738             fjptrH             = f+j_coord_offsetH;
1739
1740             gmx_mm256_decrement_3rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,
1741                                                       fjx0,fjy0,fjz0,fjx1,fjy1,fjz1,fjx2,fjy2,fjz2);
1742
1743             /* Inner loop uses 504 flops */
1744         }
1745
1746         if(jidx<j_index_end)
1747         {
1748
1749             /* Get j neighbor index, and coordinate index */
1750             jnrlistA         = jjnr[jidx];
1751             jnrlistB         = jjnr[jidx+1];
1752             jnrlistC         = jjnr[jidx+2];
1753             jnrlistD         = jjnr[jidx+3];
1754             jnrlistE         = jjnr[jidx+4];
1755             jnrlistF         = jjnr[jidx+5];
1756             jnrlistG         = jjnr[jidx+6];
1757             jnrlistH         = jjnr[jidx+7];
1758             /* Sign of each element will be negative for non-real atoms.
1759              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
1760              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
1761              */
1762             dummy_mask = gmx_mm256_set_m128(gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx+4)),_mm_setzero_si128())),
1763                                             gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128())));
1764
1765             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
1766             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
1767             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
1768             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
1769             jnrE       = (jnrlistE>=0) ? jnrlistE : 0;
1770             jnrF       = (jnrlistF>=0) ? jnrlistF : 0;
1771             jnrG       = (jnrlistG>=0) ? jnrlistG : 0;
1772             jnrH       = (jnrlistH>=0) ? jnrlistH : 0;
1773             j_coord_offsetA  = DIM*jnrA;
1774             j_coord_offsetB  = DIM*jnrB;
1775             j_coord_offsetC  = DIM*jnrC;
1776             j_coord_offsetD  = DIM*jnrD;
1777             j_coord_offsetE  = DIM*jnrE;
1778             j_coord_offsetF  = DIM*jnrF;
1779             j_coord_offsetG  = DIM*jnrG;
1780             j_coord_offsetH  = DIM*jnrH;
1781
1782             /* load j atom coordinates */
1783             gmx_mm256_load_3rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
1784                                                  x+j_coord_offsetC,x+j_coord_offsetD,
1785                                                  x+j_coord_offsetE,x+j_coord_offsetF,
1786                                                  x+j_coord_offsetG,x+j_coord_offsetH,
1787                                               &jx0,&jy0,&jz0,&jx1,&jy1,&jz1,&jx2,&jy2,&jz2);
1788
1789             /* Calculate displacement vector */
1790             dx00             = _mm256_sub_ps(ix0,jx0);
1791             dy00             = _mm256_sub_ps(iy0,jy0);
1792             dz00             = _mm256_sub_ps(iz0,jz0);
1793             dx01             = _mm256_sub_ps(ix0,jx1);
1794             dy01             = _mm256_sub_ps(iy0,jy1);
1795             dz01             = _mm256_sub_ps(iz0,jz1);
1796             dx02             = _mm256_sub_ps(ix0,jx2);
1797             dy02             = _mm256_sub_ps(iy0,jy2);
1798             dz02             = _mm256_sub_ps(iz0,jz2);
1799             dx10             = _mm256_sub_ps(ix1,jx0);
1800             dy10             = _mm256_sub_ps(iy1,jy0);
1801             dz10             = _mm256_sub_ps(iz1,jz0);
1802             dx11             = _mm256_sub_ps(ix1,jx1);
1803             dy11             = _mm256_sub_ps(iy1,jy1);
1804             dz11             = _mm256_sub_ps(iz1,jz1);
1805             dx12             = _mm256_sub_ps(ix1,jx2);
1806             dy12             = _mm256_sub_ps(iy1,jy2);
1807             dz12             = _mm256_sub_ps(iz1,jz2);
1808             dx20             = _mm256_sub_ps(ix2,jx0);
1809             dy20             = _mm256_sub_ps(iy2,jy0);
1810             dz20             = _mm256_sub_ps(iz2,jz0);
1811             dx21             = _mm256_sub_ps(ix2,jx1);
1812             dy21             = _mm256_sub_ps(iy2,jy1);
1813             dz21             = _mm256_sub_ps(iz2,jz1);
1814             dx22             = _mm256_sub_ps(ix2,jx2);
1815             dy22             = _mm256_sub_ps(iy2,jy2);
1816             dz22             = _mm256_sub_ps(iz2,jz2);
1817
1818             /* Calculate squared distance and things based on it */
1819             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
1820             rsq01            = gmx_mm256_calc_rsq_ps(dx01,dy01,dz01);
1821             rsq02            = gmx_mm256_calc_rsq_ps(dx02,dy02,dz02);
1822             rsq10            = gmx_mm256_calc_rsq_ps(dx10,dy10,dz10);
1823             rsq11            = gmx_mm256_calc_rsq_ps(dx11,dy11,dz11);
1824             rsq12            = gmx_mm256_calc_rsq_ps(dx12,dy12,dz12);
1825             rsq20            = gmx_mm256_calc_rsq_ps(dx20,dy20,dz20);
1826             rsq21            = gmx_mm256_calc_rsq_ps(dx21,dy21,dz21);
1827             rsq22            = gmx_mm256_calc_rsq_ps(dx22,dy22,dz22);
1828
1829             rinv00           = gmx_mm256_invsqrt_ps(rsq00);
1830             rinv01           = gmx_mm256_invsqrt_ps(rsq01);
1831             rinv02           = gmx_mm256_invsqrt_ps(rsq02);
1832             rinv10           = gmx_mm256_invsqrt_ps(rsq10);
1833             rinv11           = gmx_mm256_invsqrt_ps(rsq11);
1834             rinv12           = gmx_mm256_invsqrt_ps(rsq12);
1835             rinv20           = gmx_mm256_invsqrt_ps(rsq20);
1836             rinv21           = gmx_mm256_invsqrt_ps(rsq21);
1837             rinv22           = gmx_mm256_invsqrt_ps(rsq22);
1838
1839             rinvsq00         = _mm256_mul_ps(rinv00,rinv00);
1840             rinvsq01         = _mm256_mul_ps(rinv01,rinv01);
1841             rinvsq02         = _mm256_mul_ps(rinv02,rinv02);
1842             rinvsq10         = _mm256_mul_ps(rinv10,rinv10);
1843             rinvsq11         = _mm256_mul_ps(rinv11,rinv11);
1844             rinvsq12         = _mm256_mul_ps(rinv12,rinv12);
1845             rinvsq20         = _mm256_mul_ps(rinv20,rinv20);
1846             rinvsq21         = _mm256_mul_ps(rinv21,rinv21);
1847             rinvsq22         = _mm256_mul_ps(rinv22,rinv22);
1848
1849             fjx0             = _mm256_setzero_ps();
1850             fjy0             = _mm256_setzero_ps();
1851             fjz0             = _mm256_setzero_ps();
1852             fjx1             = _mm256_setzero_ps();
1853             fjy1             = _mm256_setzero_ps();
1854             fjz1             = _mm256_setzero_ps();
1855             fjx2             = _mm256_setzero_ps();
1856             fjy2             = _mm256_setzero_ps();
1857             fjz2             = _mm256_setzero_ps();
1858
1859             /**************************
1860              * CALCULATE INTERACTIONS *
1861              **************************/
1862
1863             r00              = _mm256_mul_ps(rsq00,rinv00);
1864             r00              = _mm256_andnot_ps(dummy_mask,r00);
1865
1866             /* EWALD ELECTROSTATICS */
1867
1868             /* Analytical PME correction */
1869             zeta2            = _mm256_mul_ps(beta2,rsq00);
1870             rinv3            = _mm256_mul_ps(rinvsq00,rinv00);
1871             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
1872             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
1873             felec            = _mm256_mul_ps(qq00,felec);
1874
1875             fscal            = felec;
1876
1877             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
1878
1879             /* Calculate temporary vectorial force */
1880             tx               = _mm256_mul_ps(fscal,dx00);
1881             ty               = _mm256_mul_ps(fscal,dy00);
1882             tz               = _mm256_mul_ps(fscal,dz00);
1883
1884             /* Update vectorial force */
1885             fix0             = _mm256_add_ps(fix0,tx);
1886             fiy0             = _mm256_add_ps(fiy0,ty);
1887             fiz0             = _mm256_add_ps(fiz0,tz);
1888
1889             fjx0             = _mm256_add_ps(fjx0,tx);
1890             fjy0             = _mm256_add_ps(fjy0,ty);
1891             fjz0             = _mm256_add_ps(fjz0,tz);
1892
1893             /**************************
1894              * CALCULATE INTERACTIONS *
1895              **************************/
1896
1897             r01              = _mm256_mul_ps(rsq01,rinv01);
1898             r01              = _mm256_andnot_ps(dummy_mask,r01);
1899
1900             /* EWALD ELECTROSTATICS */
1901
1902             /* Analytical PME correction */
1903             zeta2            = _mm256_mul_ps(beta2,rsq01);
1904             rinv3            = _mm256_mul_ps(rinvsq01,rinv01);
1905             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
1906             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
1907             felec            = _mm256_mul_ps(qq01,felec);
1908
1909             fscal            = felec;
1910
1911             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
1912
1913             /* Calculate temporary vectorial force */
1914             tx               = _mm256_mul_ps(fscal,dx01);
1915             ty               = _mm256_mul_ps(fscal,dy01);
1916             tz               = _mm256_mul_ps(fscal,dz01);
1917
1918             /* Update vectorial force */
1919             fix0             = _mm256_add_ps(fix0,tx);
1920             fiy0             = _mm256_add_ps(fiy0,ty);
1921             fiz0             = _mm256_add_ps(fiz0,tz);
1922
1923             fjx1             = _mm256_add_ps(fjx1,tx);
1924             fjy1             = _mm256_add_ps(fjy1,ty);
1925             fjz1             = _mm256_add_ps(fjz1,tz);
1926
1927             /**************************
1928              * CALCULATE INTERACTIONS *
1929              **************************/
1930
1931             r02              = _mm256_mul_ps(rsq02,rinv02);
1932             r02              = _mm256_andnot_ps(dummy_mask,r02);
1933
1934             /* EWALD ELECTROSTATICS */
1935
1936             /* Analytical PME correction */
1937             zeta2            = _mm256_mul_ps(beta2,rsq02);
1938             rinv3            = _mm256_mul_ps(rinvsq02,rinv02);
1939             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
1940             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
1941             felec            = _mm256_mul_ps(qq02,felec);
1942
1943             fscal            = felec;
1944
1945             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
1946
1947             /* Calculate temporary vectorial force */
1948             tx               = _mm256_mul_ps(fscal,dx02);
1949             ty               = _mm256_mul_ps(fscal,dy02);
1950             tz               = _mm256_mul_ps(fscal,dz02);
1951
1952             /* Update vectorial force */
1953             fix0             = _mm256_add_ps(fix0,tx);
1954             fiy0             = _mm256_add_ps(fiy0,ty);
1955             fiz0             = _mm256_add_ps(fiz0,tz);
1956
1957             fjx2             = _mm256_add_ps(fjx2,tx);
1958             fjy2             = _mm256_add_ps(fjy2,ty);
1959             fjz2             = _mm256_add_ps(fjz2,tz);
1960
1961             /**************************
1962              * CALCULATE INTERACTIONS *
1963              **************************/
1964
1965             r10              = _mm256_mul_ps(rsq10,rinv10);
1966             r10              = _mm256_andnot_ps(dummy_mask,r10);
1967
1968             /* EWALD ELECTROSTATICS */
1969
1970             /* Analytical PME correction */
1971             zeta2            = _mm256_mul_ps(beta2,rsq10);
1972             rinv3            = _mm256_mul_ps(rinvsq10,rinv10);
1973             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
1974             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
1975             felec            = _mm256_mul_ps(qq10,felec);
1976
1977             fscal            = felec;
1978
1979             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
1980
1981             /* Calculate temporary vectorial force */
1982             tx               = _mm256_mul_ps(fscal,dx10);
1983             ty               = _mm256_mul_ps(fscal,dy10);
1984             tz               = _mm256_mul_ps(fscal,dz10);
1985
1986             /* Update vectorial force */
1987             fix1             = _mm256_add_ps(fix1,tx);
1988             fiy1             = _mm256_add_ps(fiy1,ty);
1989             fiz1             = _mm256_add_ps(fiz1,tz);
1990
1991             fjx0             = _mm256_add_ps(fjx0,tx);
1992             fjy0             = _mm256_add_ps(fjy0,ty);
1993             fjz0             = _mm256_add_ps(fjz0,tz);
1994
1995             /**************************
1996              * CALCULATE INTERACTIONS *
1997              **************************/
1998
1999             r11              = _mm256_mul_ps(rsq11,rinv11);
2000             r11              = _mm256_andnot_ps(dummy_mask,r11);
2001
2002             /* EWALD ELECTROSTATICS */
2003
2004             /* Analytical PME correction */
2005             zeta2            = _mm256_mul_ps(beta2,rsq11);
2006             rinv3            = _mm256_mul_ps(rinvsq11,rinv11);
2007             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
2008             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
2009             felec            = _mm256_mul_ps(qq11,felec);
2010
2011             fscal            = felec;
2012
2013             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
2014
2015             /* Calculate temporary vectorial force */
2016             tx               = _mm256_mul_ps(fscal,dx11);
2017             ty               = _mm256_mul_ps(fscal,dy11);
2018             tz               = _mm256_mul_ps(fscal,dz11);
2019
2020             /* Update vectorial force */
2021             fix1             = _mm256_add_ps(fix1,tx);
2022             fiy1             = _mm256_add_ps(fiy1,ty);
2023             fiz1             = _mm256_add_ps(fiz1,tz);
2024
2025             fjx1             = _mm256_add_ps(fjx1,tx);
2026             fjy1             = _mm256_add_ps(fjy1,ty);
2027             fjz1             = _mm256_add_ps(fjz1,tz);
2028
2029             /**************************
2030              * CALCULATE INTERACTIONS *
2031              **************************/
2032
2033             r12              = _mm256_mul_ps(rsq12,rinv12);
2034             r12              = _mm256_andnot_ps(dummy_mask,r12);
2035
2036             /* EWALD ELECTROSTATICS */
2037
2038             /* Analytical PME correction */
2039             zeta2            = _mm256_mul_ps(beta2,rsq12);
2040             rinv3            = _mm256_mul_ps(rinvsq12,rinv12);
2041             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
2042             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
2043             felec            = _mm256_mul_ps(qq12,felec);
2044
2045             fscal            = felec;
2046
2047             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
2048
2049             /* Calculate temporary vectorial force */
2050             tx               = _mm256_mul_ps(fscal,dx12);
2051             ty               = _mm256_mul_ps(fscal,dy12);
2052             tz               = _mm256_mul_ps(fscal,dz12);
2053
2054             /* Update vectorial force */
2055             fix1             = _mm256_add_ps(fix1,tx);
2056             fiy1             = _mm256_add_ps(fiy1,ty);
2057             fiz1             = _mm256_add_ps(fiz1,tz);
2058
2059             fjx2             = _mm256_add_ps(fjx2,tx);
2060             fjy2             = _mm256_add_ps(fjy2,ty);
2061             fjz2             = _mm256_add_ps(fjz2,tz);
2062
2063             /**************************
2064              * CALCULATE INTERACTIONS *
2065              **************************/
2066
2067             r20              = _mm256_mul_ps(rsq20,rinv20);
2068             r20              = _mm256_andnot_ps(dummy_mask,r20);
2069
2070             /* EWALD ELECTROSTATICS */
2071
2072             /* Analytical PME correction */
2073             zeta2            = _mm256_mul_ps(beta2,rsq20);
2074             rinv3            = _mm256_mul_ps(rinvsq20,rinv20);
2075             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
2076             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
2077             felec            = _mm256_mul_ps(qq20,felec);
2078
2079             fscal            = felec;
2080
2081             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
2082
2083             /* Calculate temporary vectorial force */
2084             tx               = _mm256_mul_ps(fscal,dx20);
2085             ty               = _mm256_mul_ps(fscal,dy20);
2086             tz               = _mm256_mul_ps(fscal,dz20);
2087
2088             /* Update vectorial force */
2089             fix2             = _mm256_add_ps(fix2,tx);
2090             fiy2             = _mm256_add_ps(fiy2,ty);
2091             fiz2             = _mm256_add_ps(fiz2,tz);
2092
2093             fjx0             = _mm256_add_ps(fjx0,tx);
2094             fjy0             = _mm256_add_ps(fjy0,ty);
2095             fjz0             = _mm256_add_ps(fjz0,tz);
2096
2097             /**************************
2098              * CALCULATE INTERACTIONS *
2099              **************************/
2100
2101             r21              = _mm256_mul_ps(rsq21,rinv21);
2102             r21              = _mm256_andnot_ps(dummy_mask,r21);
2103
2104             /* EWALD ELECTROSTATICS */
2105
2106             /* Analytical PME correction */
2107             zeta2            = _mm256_mul_ps(beta2,rsq21);
2108             rinv3            = _mm256_mul_ps(rinvsq21,rinv21);
2109             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
2110             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
2111             felec            = _mm256_mul_ps(qq21,felec);
2112
2113             fscal            = felec;
2114
2115             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
2116
2117             /* Calculate temporary vectorial force */
2118             tx               = _mm256_mul_ps(fscal,dx21);
2119             ty               = _mm256_mul_ps(fscal,dy21);
2120             tz               = _mm256_mul_ps(fscal,dz21);
2121
2122             /* Update vectorial force */
2123             fix2             = _mm256_add_ps(fix2,tx);
2124             fiy2             = _mm256_add_ps(fiy2,ty);
2125             fiz2             = _mm256_add_ps(fiz2,tz);
2126
2127             fjx1             = _mm256_add_ps(fjx1,tx);
2128             fjy1             = _mm256_add_ps(fjy1,ty);
2129             fjz1             = _mm256_add_ps(fjz1,tz);
2130
2131             /**************************
2132              * CALCULATE INTERACTIONS *
2133              **************************/
2134
2135             r22              = _mm256_mul_ps(rsq22,rinv22);
2136             r22              = _mm256_andnot_ps(dummy_mask,r22);
2137
2138             /* EWALD ELECTROSTATICS */
2139
2140             /* Analytical PME correction */
2141             zeta2            = _mm256_mul_ps(beta2,rsq22);
2142             rinv3            = _mm256_mul_ps(rinvsq22,rinv22);
2143             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
2144             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
2145             felec            = _mm256_mul_ps(qq22,felec);
2146
2147             fscal            = felec;
2148
2149             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
2150
2151             /* Calculate temporary vectorial force */
2152             tx               = _mm256_mul_ps(fscal,dx22);
2153             ty               = _mm256_mul_ps(fscal,dy22);
2154             tz               = _mm256_mul_ps(fscal,dz22);
2155
2156             /* Update vectorial force */
2157             fix2             = _mm256_add_ps(fix2,tx);
2158             fiy2             = _mm256_add_ps(fiy2,ty);
2159             fiz2             = _mm256_add_ps(fiz2,tz);
2160
2161             fjx2             = _mm256_add_ps(fjx2,tx);
2162             fjy2             = _mm256_add_ps(fjy2,ty);
2163             fjz2             = _mm256_add_ps(fjz2,tz);
2164
2165             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
2166             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
2167             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
2168             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
2169             fjptrE             = (jnrlistE>=0) ? f+j_coord_offsetE : scratch;
2170             fjptrF             = (jnrlistF>=0) ? f+j_coord_offsetF : scratch;
2171             fjptrG             = (jnrlistG>=0) ? f+j_coord_offsetG : scratch;
2172             fjptrH             = (jnrlistH>=0) ? f+j_coord_offsetH : scratch;
2173
2174             gmx_mm256_decrement_3rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,
2175                                                       fjx0,fjy0,fjz0,fjx1,fjy1,fjz1,fjx2,fjy2,fjz2);
2176
2177             /* Inner loop uses 513 flops */
2178         }
2179
2180         /* End of innermost loop */
2181
2182         gmx_mm256_update_iforce_3atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
2183                                                  f+i_coord_offset,fshift+i_shift_offset);
2184
2185         /* Increment number of inner iterations */
2186         inneriter                  += j_index_end - j_index_start;
2187
2188         /* Outer loop uses 18 flops */
2189     }
2190
2191     /* Increment number of outer iterations */
2192     outeriter        += nri;
2193
2194     /* Update outer/inner flops */
2195
2196     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_W3W3_F,outeriter*18 + inneriter*513);
2197 }