src/gromacs/gmxlib/nonbonded/nb_kernel_avx_256_double/nb_kernel_ElecCoul_VdwNone_GeomW3W3_avx_256_double.c

   1 /*
   2  * This file is part of the GROMACS molecular simulation package.
   3  *
   4  * Copyright (c) 2012,2013,2014,2015,2017, by the GROMACS development team, led by
   5  * Mark Abraham, David van der Spoel, Berk Hess, and Erik Lindahl,
   6  * and including many others, as listed in the AUTHORS file in the
   7  * top-level source directory and at http://www.gromacs.org.
   8  *
   9  * GROMACS is free software; you can redistribute it and/or
  10  * modify it under the terms of the GNU Lesser General Public License
  11  * as published by the Free Software Foundation; either version 2.1
  12  * of the License, or (at your option) any later version.
  13  *
  14  * GROMACS is distributed in the hope that it will be useful,
  15  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  16  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  17  * Lesser General Public License for more details.
  18  *
  19  * You should have received a copy of the GNU Lesser General Public
  20  * License along with GROMACS; if not, see
  21  * http://www.gnu.org/licenses, or write to the Free Software Foundation,
  22  * Inc., 51 Franklin Street, Fifth Floor, Boston, MA  02110-1301  USA.
  23  *
  24  * If you want to redistribute modifications to GROMACS, please
  25  * consider that scientific software is very special. Version
  26  * control is crucial - bugs must be traceable. We will be happy to
  27  * consider code for inclusion in the official distribution, but
  28  * derived work must not be called official GROMACS. Details are found
  29  * in the README & COPYING files - if they are missing, get the
  30  * official version at http://www.gromacs.org.
  31  *
  32  * To help us fund GROMACS development, we humbly ask that you cite
  33  * the research papers on the package. Check out http://www.gromacs.org.
  34  */
  35 /*
  36  * Note: this file was generated by the GROMACS avx_256_double kernel generator.
  37  */
  38 #include "gmxpre.h"
  39
  40 #include "config.h"
  41
  42 #include <math.h>
  43
  44 #include "../nb_kernel.h"
  45 #include "gromacs/gmxlib/nrnb.h"
  46
  47 #include "kernelutil_x86_avx_256_double.h"
  48
  49 /*
  50  * Gromacs nonbonded kernel:   nb_kernel_ElecCoul_VdwNone_GeomW3W3_VF_avx_256_double
  51  * Electrostatics interaction: Coulomb
  52  * VdW interaction:            None
  53  * Geometry:                   Water3-Water3
  54  * Calculate force/pot:        PotentialAndForce
  55  */
  56 void
  57 nb_kernel_ElecCoul_VdwNone_GeomW3W3_VF_avx_256_double
  58                     (t_nblist                    * gmx_restrict       nlist,
  59                      rvec                        * gmx_restrict          xx,
  60                      rvec                        * gmx_restrict          ff,
  61                      struct t_forcerec           * gmx_restrict          fr,
  62                      t_mdatoms                   * gmx_restrict     mdatoms,
  63                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
  64                      t_nrnb                      * gmx_restrict        nrnb)
  65 {
  66     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
  67      * just 0 for non-waters.
  68      * Suffixes A,B,C,D refer to j loop unrolling done with AVX, e.g. for the four different
  69      * jnr indices corresponding to data put in the four positions in the SIMD register.
  70      */
  71     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
  72     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
  73     int              jnrA,jnrB,jnrC,jnrD;
  74     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
  75     int              jnrlistE,jnrlistF,jnrlistG,jnrlistH;
  76     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
  77     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
  78     real             rcutoff_scalar;
  79     real             *shiftvec,*fshift,*x,*f;
  80     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
  81     real             scratch[4*DIM];
  82     __m256d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
  83     real *           vdwioffsetptr0;
  84     __m256d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
  85     real *           vdwioffsetptr1;
  86     __m256d          ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
  87     real *           vdwioffsetptr2;
  88     __m256d          ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
  89     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
  90     __m256d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  91     int              vdwjidx1A,vdwjidx1B,vdwjidx1C,vdwjidx1D;
  92     __m256d          jx1,jy1,jz1,fjx1,fjy1,fjz1,jq1,isaj1;
  93     int              vdwjidx2A,vdwjidx2B,vdwjidx2C,vdwjidx2D;
  94     __m256d          jx2,jy2,jz2,fjx2,fjy2,fjz2,jq2,isaj2;
  95     __m256d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  96     __m256d          dx01,dy01,dz01,rsq01,rinv01,rinvsq01,r01,qq01,c6_01,c12_01;
  97     __m256d          dx02,dy02,dz02,rsq02,rinv02,rinvsq02,r02,qq02,c6_02,c12_02;
  98     __m256d          dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
  99     __m256d          dx11,dy11,dz11,rsq11,rinv11,rinvsq11,r11,qq11,c6_11,c12_11;
 100     __m256d          dx12,dy12,dz12,rsq12,rinv12,rinvsq12,r12,qq12,c6_12,c12_12;
 101     __m256d          dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
 102     __m256d          dx21,dy21,dz21,rsq21,rinv21,rinvsq21,r21,qq21,c6_21,c12_21;
 103     __m256d          dx22,dy22,dz22,rsq22,rinv22,rinvsq22,r22,qq22,c6_22,c12_22;
 104     __m256d          velec,felec,velecsum,facel,crf,krf,krf2;
 105     real             *charge;
 106     __m256d          dummy_mask,cutoff_mask;
 107     __m128           tmpmask0,tmpmask1;
 108     __m256d          signbit = _mm256_castsi256_pd( _mm256_set1_epi32(0x80000000) );
 109     __m256d          one     = _mm256_set1_pd(1.0);
 110     __m256d          two     = _mm256_set1_pd(2.0);
 111     x                = xx[0];
 112     f                = ff[0];
 113
 114     nri              = nlist->nri;
 115     iinr             = nlist->iinr;
 116     jindex           = nlist->jindex;
 117     jjnr             = nlist->jjnr;
 118     shiftidx         = nlist->shift;
 119     gid              = nlist->gid;
 120     shiftvec         = fr->shift_vec[0];
 121     fshift           = fr->fshift[0];
 122     facel            = _mm256_set1_pd(fr->ic->epsfac);
 123     charge           = mdatoms->chargeA;
 124
 125     /* Setup water-specific parameters */
 126     inr              = nlist->iinr[0];
 127     iq0              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+0]));
 128     iq1              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+1]));
 129     iq2              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+2]));
 130
 131     jq0              = _mm256_set1_pd(charge[inr+0]);
 132     jq1              = _mm256_set1_pd(charge[inr+1]);
 133     jq2              = _mm256_set1_pd(charge[inr+2]);
 134     qq00             = _mm256_mul_pd(iq0,jq0);
 135     qq01             = _mm256_mul_pd(iq0,jq1);
 136     qq02             = _mm256_mul_pd(iq0,jq2);
 137     qq10             = _mm256_mul_pd(iq1,jq0);
 138     qq11             = _mm256_mul_pd(iq1,jq1);
 139     qq12             = _mm256_mul_pd(iq1,jq2);
 140     qq20             = _mm256_mul_pd(iq2,jq0);
 141     qq21             = _mm256_mul_pd(iq2,jq1);
 142     qq22             = _mm256_mul_pd(iq2,jq2);
 143
 144     /* Avoid stupid compiler warnings */
 145     jnrA = jnrB = jnrC = jnrD = 0;
 146     j_coord_offsetA = 0;
 147     j_coord_offsetB = 0;
 148     j_coord_offsetC = 0;
 149     j_coord_offsetD = 0;
 150
 151     outeriter        = 0;
 152     inneriter        = 0;
 153
 154     for(iidx=0;iidx<4*DIM;iidx++)
 155     {
 156         scratch[iidx] = 0.0;
 157     }
 158
 159     /* Start outer loop over neighborlists */
 160     for(iidx=0; iidx<nri; iidx++)
 161     {
 162         /* Load shift vector for this list */
 163         i_shift_offset   = DIM*shiftidx[iidx];
 164
 165         /* Load limits for loop over neighbors */
 166         j_index_start    = jindex[iidx];
 167         j_index_end      = jindex[iidx+1];
 168
 169         /* Get outer coordinate index */
 170         inr              = iinr[iidx];
 171         i_coord_offset   = DIM*inr;
 172
 173         /* Load i particle coords and add shift vector */
 174         gmx_mm256_load_shift_and_3rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
 175                                                     &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 176
 177         fix0             = _mm256_setzero_pd();
 178         fiy0             = _mm256_setzero_pd();
 179         fiz0             = _mm256_setzero_pd();
 180         fix1             = _mm256_setzero_pd();
 181         fiy1             = _mm256_setzero_pd();
 182         fiz1             = _mm256_setzero_pd();
 183         fix2             = _mm256_setzero_pd();
 184         fiy2             = _mm256_setzero_pd();
 185         fiz2             = _mm256_setzero_pd();
 186
 187         /* Reset potential sums */
 188         velecsum         = _mm256_setzero_pd();
 189
 190         /* Start inner kernel loop */
 191         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 192         {
 193
 194             /* Get j neighbor index, and coordinate index */
 195             jnrA             = jjnr[jidx];
 196             jnrB             = jjnr[jidx+1];
 197             jnrC             = jjnr[jidx+2];
 198             jnrD             = jjnr[jidx+3];
 199             j_coord_offsetA  = DIM*jnrA;
 200             j_coord_offsetB  = DIM*jnrB;
 201             j_coord_offsetC  = DIM*jnrC;
 202             j_coord_offsetD  = DIM*jnrD;
 203
 204             /* load j atom coordinates */
 205             gmx_mm256_load_3rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 206                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 207                                               &jx0,&jy0,&jz0,&jx1,&jy1,&jz1,&jx2,&jy2,&jz2);
 208
 209             /* Calculate displacement vector */
 210             dx00             = _mm256_sub_pd(ix0,jx0);
 211             dy00             = _mm256_sub_pd(iy0,jy0);
 212             dz00             = _mm256_sub_pd(iz0,jz0);
 213             dx01             = _mm256_sub_pd(ix0,jx1);
 214             dy01             = _mm256_sub_pd(iy0,jy1);
 215             dz01             = _mm256_sub_pd(iz0,jz1);
 216             dx02             = _mm256_sub_pd(ix0,jx2);
 217             dy02             = _mm256_sub_pd(iy0,jy2);
 218             dz02             = _mm256_sub_pd(iz0,jz2);
 219             dx10             = _mm256_sub_pd(ix1,jx0);
 220             dy10             = _mm256_sub_pd(iy1,jy0);
 221             dz10             = _mm256_sub_pd(iz1,jz0);
 222             dx11             = _mm256_sub_pd(ix1,jx1);
 223             dy11             = _mm256_sub_pd(iy1,jy1);
 224             dz11             = _mm256_sub_pd(iz1,jz1);
 225             dx12             = _mm256_sub_pd(ix1,jx2);
 226             dy12             = _mm256_sub_pd(iy1,jy2);
 227             dz12             = _mm256_sub_pd(iz1,jz2);
 228             dx20             = _mm256_sub_pd(ix2,jx0);
 229             dy20             = _mm256_sub_pd(iy2,jy0);
 230             dz20             = _mm256_sub_pd(iz2,jz0);
 231             dx21             = _mm256_sub_pd(ix2,jx1);
 232             dy21             = _mm256_sub_pd(iy2,jy1);
 233             dz21             = _mm256_sub_pd(iz2,jz1);
 234             dx22             = _mm256_sub_pd(ix2,jx2);
 235             dy22             = _mm256_sub_pd(iy2,jy2);
 236             dz22             = _mm256_sub_pd(iz2,jz2);
 237
 238             /* Calculate squared distance and things based on it */
 239             rsq00            = gmx_mm256_calc_rsq_pd(dx00,dy00,dz00);
 240             rsq01            = gmx_mm256_calc_rsq_pd(dx01,dy01,dz01);
 241             rsq02            = gmx_mm256_calc_rsq_pd(dx02,dy02,dz02);
 242             rsq10            = gmx_mm256_calc_rsq_pd(dx10,dy10,dz10);
 243             rsq11            = gmx_mm256_calc_rsq_pd(dx11,dy11,dz11);
 244             rsq12            = gmx_mm256_calc_rsq_pd(dx12,dy12,dz12);
 245             rsq20            = gmx_mm256_calc_rsq_pd(dx20,dy20,dz20);
 246             rsq21            = gmx_mm256_calc_rsq_pd(dx21,dy21,dz21);
 247             rsq22            = gmx_mm256_calc_rsq_pd(dx22,dy22,dz22);
 248
 249             rinv00           = avx256_invsqrt_d(rsq00);
 250             rinv01           = avx256_invsqrt_d(rsq01);
 251             rinv02           = avx256_invsqrt_d(rsq02);
 252             rinv10           = avx256_invsqrt_d(rsq10);
 253             rinv11           = avx256_invsqrt_d(rsq11);
 254             rinv12           = avx256_invsqrt_d(rsq12);
 255             rinv20           = avx256_invsqrt_d(rsq20);
 256             rinv21           = avx256_invsqrt_d(rsq21);
 257             rinv22           = avx256_invsqrt_d(rsq22);
 258
 259             rinvsq00         = _mm256_mul_pd(rinv00,rinv00);
 260             rinvsq01         = _mm256_mul_pd(rinv01,rinv01);
 261             rinvsq02         = _mm256_mul_pd(rinv02,rinv02);
 262             rinvsq10         = _mm256_mul_pd(rinv10,rinv10);
 263             rinvsq11         = _mm256_mul_pd(rinv11,rinv11);
 264             rinvsq12         = _mm256_mul_pd(rinv12,rinv12);
 265             rinvsq20         = _mm256_mul_pd(rinv20,rinv20);
 266             rinvsq21         = _mm256_mul_pd(rinv21,rinv21);
 267             rinvsq22         = _mm256_mul_pd(rinv22,rinv22);
 268
 269             fjx0             = _mm256_setzero_pd();
 270             fjy0             = _mm256_setzero_pd();
 271             fjz0             = _mm256_setzero_pd();
 272             fjx1             = _mm256_setzero_pd();
 273             fjy1             = _mm256_setzero_pd();
 274             fjz1             = _mm256_setzero_pd();
 275             fjx2             = _mm256_setzero_pd();
 276             fjy2             = _mm256_setzero_pd();
 277             fjz2             = _mm256_setzero_pd();
 278
 279             /**************************
 280              * CALCULATE INTERACTIONS *
 281              **************************/
 282
 283             /* COULOMB ELECTROSTATICS */
 284             velec            = _mm256_mul_pd(qq00,rinv00);
 285             felec            = _mm256_mul_pd(velec,rinvsq00);
 286
 287             /* Update potential sum for this i atom from the interaction with this j atom. */
 288             velecsum         = _mm256_add_pd(velecsum,velec);
 289
 290             fscal            = felec;
 291
 292             /* Calculate temporary vectorial force */
 293             tx               = _mm256_mul_pd(fscal,dx00);
 294             ty               = _mm256_mul_pd(fscal,dy00);
 295             tz               = _mm256_mul_pd(fscal,dz00);
 296
 297             /* Update vectorial force */
 298             fix0             = _mm256_add_pd(fix0,tx);
 299             fiy0             = _mm256_add_pd(fiy0,ty);
 300             fiz0             = _mm256_add_pd(fiz0,tz);
 301
 302             fjx0             = _mm256_add_pd(fjx0,tx);
 303             fjy0             = _mm256_add_pd(fjy0,ty);
 304             fjz0             = _mm256_add_pd(fjz0,tz);
 305
 306             /**************************
 307              * CALCULATE INTERACTIONS *
 308              **************************/
 309
 310             /* COULOMB ELECTROSTATICS */
 311             velec            = _mm256_mul_pd(qq01,rinv01);
 312             felec            = _mm256_mul_pd(velec,rinvsq01);
 313
 314             /* Update potential sum for this i atom from the interaction with this j atom. */
 315             velecsum         = _mm256_add_pd(velecsum,velec);
 316
 317             fscal            = felec;
 318
 319             /* Calculate temporary vectorial force */
 320             tx               = _mm256_mul_pd(fscal,dx01);
 321             ty               = _mm256_mul_pd(fscal,dy01);
 322             tz               = _mm256_mul_pd(fscal,dz01);
 323
 324             /* Update vectorial force */
 325             fix0             = _mm256_add_pd(fix0,tx);
 326             fiy0             = _mm256_add_pd(fiy0,ty);
 327             fiz0             = _mm256_add_pd(fiz0,tz);
 328
 329             fjx1             = _mm256_add_pd(fjx1,tx);
 330             fjy1             = _mm256_add_pd(fjy1,ty);
 331             fjz1             = _mm256_add_pd(fjz1,tz);
 332
 333             /**************************
 334              * CALCULATE INTERACTIONS *
 335              **************************/
 336
 337             /* COULOMB ELECTROSTATICS */
 338             velec            = _mm256_mul_pd(qq02,rinv02);
 339             felec            = _mm256_mul_pd(velec,rinvsq02);
 340
 341             /* Update potential sum for this i atom from the interaction with this j atom. */
 342             velecsum         = _mm256_add_pd(velecsum,velec);
 343
 344             fscal            = felec;
 345
 346             /* Calculate temporary vectorial force */
 347             tx               = _mm256_mul_pd(fscal,dx02);
 348             ty               = _mm256_mul_pd(fscal,dy02);
 349             tz               = _mm256_mul_pd(fscal,dz02);
 350
 351             /* Update vectorial force */
 352             fix0             = _mm256_add_pd(fix0,tx);
 353             fiy0             = _mm256_add_pd(fiy0,ty);
 354             fiz0             = _mm256_add_pd(fiz0,tz);
 355
 356             fjx2             = _mm256_add_pd(fjx2,tx);
 357             fjy2             = _mm256_add_pd(fjy2,ty);
 358             fjz2             = _mm256_add_pd(fjz2,tz);
 359
 360             /**************************
 361              * CALCULATE INTERACTIONS *
 362              **************************/
 363
 364             /* COULOMB ELECTROSTATICS */
 365             velec            = _mm256_mul_pd(qq10,rinv10);
 366             felec            = _mm256_mul_pd(velec,rinvsq10);
 367
 368             /* Update potential sum for this i atom from the interaction with this j atom. */
 369             velecsum         = _mm256_add_pd(velecsum,velec);
 370
 371             fscal            = felec;
 372
 373             /* Calculate temporary vectorial force */
 374             tx               = _mm256_mul_pd(fscal,dx10);
 375             ty               = _mm256_mul_pd(fscal,dy10);
 376             tz               = _mm256_mul_pd(fscal,dz10);
 377
 378             /* Update vectorial force */
 379             fix1             = _mm256_add_pd(fix1,tx);
 380             fiy1             = _mm256_add_pd(fiy1,ty);
 381             fiz1             = _mm256_add_pd(fiz1,tz);
 382
 383             fjx0             = _mm256_add_pd(fjx0,tx);
 384             fjy0             = _mm256_add_pd(fjy0,ty);
 385             fjz0             = _mm256_add_pd(fjz0,tz);
 386
 387             /**************************
 388              * CALCULATE INTERACTIONS *
 389              **************************/
 390
 391             /* COULOMB ELECTROSTATICS */
 392             velec            = _mm256_mul_pd(qq11,rinv11);
 393             felec            = _mm256_mul_pd(velec,rinvsq11);
 394
 395             /* Update potential sum for this i atom from the interaction with this j atom. */
 396             velecsum         = _mm256_add_pd(velecsum,velec);
 397
 398             fscal            = felec;
 399
 400             /* Calculate temporary vectorial force */
 401             tx               = _mm256_mul_pd(fscal,dx11);
 402             ty               = _mm256_mul_pd(fscal,dy11);
 403             tz               = _mm256_mul_pd(fscal,dz11);
 404
 405             /* Update vectorial force */
 406             fix1             = _mm256_add_pd(fix1,tx);
 407             fiy1             = _mm256_add_pd(fiy1,ty);
 408             fiz1             = _mm256_add_pd(fiz1,tz);
 409
 410             fjx1             = _mm256_add_pd(fjx1,tx);
 411             fjy1             = _mm256_add_pd(fjy1,ty);
 412             fjz1             = _mm256_add_pd(fjz1,tz);
 413
 414             /**************************
 415              * CALCULATE INTERACTIONS *
 416              **************************/
 417
 418             /* COULOMB ELECTROSTATICS */
 419             velec            = _mm256_mul_pd(qq12,rinv12);
 420             felec            = _mm256_mul_pd(velec,rinvsq12);
 421
 422             /* Update potential sum for this i atom from the interaction with this j atom. */
 423             velecsum         = _mm256_add_pd(velecsum,velec);
 424
 425             fscal            = felec;
 426
 427             /* Calculate temporary vectorial force */
 428             tx               = _mm256_mul_pd(fscal,dx12);
 429             ty               = _mm256_mul_pd(fscal,dy12);
 430             tz               = _mm256_mul_pd(fscal,dz12);
 431
 432             /* Update vectorial force */
 433             fix1             = _mm256_add_pd(fix1,tx);
 434             fiy1             = _mm256_add_pd(fiy1,ty);
 435             fiz1             = _mm256_add_pd(fiz1,tz);
 436
 437             fjx2             = _mm256_add_pd(fjx2,tx);
 438             fjy2             = _mm256_add_pd(fjy2,ty);
 439             fjz2             = _mm256_add_pd(fjz2,tz);
 440
 441             /**************************
 442              * CALCULATE INTERACTIONS *
 443              **************************/
 444
 445             /* COULOMB ELECTROSTATICS */
 446             velec            = _mm256_mul_pd(qq20,rinv20);
 447             felec            = _mm256_mul_pd(velec,rinvsq20);
 448
 449             /* Update potential sum for this i atom from the interaction with this j atom. */
 450             velecsum         = _mm256_add_pd(velecsum,velec);
 451
 452             fscal            = felec;
 453
 454             /* Calculate temporary vectorial force */
 455             tx               = _mm256_mul_pd(fscal,dx20);
 456             ty               = _mm256_mul_pd(fscal,dy20);
 457             tz               = _mm256_mul_pd(fscal,dz20);
 458
 459             /* Update vectorial force */
 460             fix2             = _mm256_add_pd(fix2,tx);
 461             fiy2             = _mm256_add_pd(fiy2,ty);
 462             fiz2             = _mm256_add_pd(fiz2,tz);
 463
 464             fjx0             = _mm256_add_pd(fjx0,tx);
 465             fjy0             = _mm256_add_pd(fjy0,ty);
 466             fjz0             = _mm256_add_pd(fjz0,tz);
 467
 468             /**************************
 469              * CALCULATE INTERACTIONS *
 470              **************************/
 471
 472             /* COULOMB ELECTROSTATICS */
 473             velec            = _mm256_mul_pd(qq21,rinv21);
 474             felec            = _mm256_mul_pd(velec,rinvsq21);
 475
 476             /* Update potential sum for this i atom from the interaction with this j atom. */
 477             velecsum         = _mm256_add_pd(velecsum,velec);
 478
 479             fscal            = felec;
 480
 481             /* Calculate temporary vectorial force */
 482             tx               = _mm256_mul_pd(fscal,dx21);
 483             ty               = _mm256_mul_pd(fscal,dy21);
 484             tz               = _mm256_mul_pd(fscal,dz21);
 485
 486             /* Update vectorial force */
 487             fix2             = _mm256_add_pd(fix2,tx);
 488             fiy2             = _mm256_add_pd(fiy2,ty);
 489             fiz2             = _mm256_add_pd(fiz2,tz);
 490
 491             fjx1             = _mm256_add_pd(fjx1,tx);
 492             fjy1             = _mm256_add_pd(fjy1,ty);
 493             fjz1             = _mm256_add_pd(fjz1,tz);
 494
 495             /**************************
 496              * CALCULATE INTERACTIONS *
 497              **************************/
 498
 499             /* COULOMB ELECTROSTATICS */
 500             velec            = _mm256_mul_pd(qq22,rinv22);
 501             felec            = _mm256_mul_pd(velec,rinvsq22);
 502
 503             /* Update potential sum for this i atom from the interaction with this j atom. */
 504             velecsum         = _mm256_add_pd(velecsum,velec);
 505
 506             fscal            = felec;
 507
 508             /* Calculate temporary vectorial force */
 509             tx               = _mm256_mul_pd(fscal,dx22);
 510             ty               = _mm256_mul_pd(fscal,dy22);
 511             tz               = _mm256_mul_pd(fscal,dz22);
 512
 513             /* Update vectorial force */
 514             fix2             = _mm256_add_pd(fix2,tx);
 515             fiy2             = _mm256_add_pd(fiy2,ty);
 516             fiz2             = _mm256_add_pd(fiz2,tz);
 517
 518             fjx2             = _mm256_add_pd(fjx2,tx);
 519             fjy2             = _mm256_add_pd(fjy2,ty);
 520             fjz2             = _mm256_add_pd(fjz2,tz);
 521
 522             fjptrA             = f+j_coord_offsetA;
 523             fjptrB             = f+j_coord_offsetB;
 524             fjptrC             = f+j_coord_offsetC;
 525             fjptrD             = f+j_coord_offsetD;
 526
 527             gmx_mm256_decrement_3rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,
 528                                                       fjx0,fjy0,fjz0,fjx1,fjy1,fjz1,fjx2,fjy2,fjz2);
 529
 530             /* Inner loop uses 243 flops */
 531         }
 532
 533         if(jidx<j_index_end)
 534         {
 535
 536             /* Get j neighbor index, and coordinate index */
 537             jnrlistA         = jjnr[jidx];
 538             jnrlistB         = jjnr[jidx+1];
 539             jnrlistC         = jjnr[jidx+2];
 540             jnrlistD         = jjnr[jidx+3];
 541             /* Sign of each element will be negative for non-real atoms.
 542              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 543              * so use it as val = _mm_andnot_pd(mask,val) to clear dummy entries.
 544              */
 545             tmpmask0 = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 546
 547             tmpmask1 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(3,3,2,2));
 548             tmpmask0 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(1,1,0,0));
 549             dummy_mask = _mm256_castps_pd(gmx_mm256_set_m128(tmpmask1,tmpmask0));
 550
 551             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 552             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 553             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 554             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 555             j_coord_offsetA  = DIM*jnrA;
 556             j_coord_offsetB  = DIM*jnrB;
 557             j_coord_offsetC  = DIM*jnrC;
 558             j_coord_offsetD  = DIM*jnrD;
 559
 560             /* load j atom coordinates */
 561             gmx_mm256_load_3rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 562                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 563                                               &jx0,&jy0,&jz0,&jx1,&jy1,&jz1,&jx2,&jy2,&jz2);
 564
 565             /* Calculate displacement vector */
 566             dx00             = _mm256_sub_pd(ix0,jx0);
 567             dy00             = _mm256_sub_pd(iy0,jy0);
 568             dz00             = _mm256_sub_pd(iz0,jz0);
 569             dx01             = _mm256_sub_pd(ix0,jx1);
 570             dy01             = _mm256_sub_pd(iy0,jy1);
 571             dz01             = _mm256_sub_pd(iz0,jz1);
 572             dx02             = _mm256_sub_pd(ix0,jx2);
 573             dy02             = _mm256_sub_pd(iy0,jy2);
 574             dz02             = _mm256_sub_pd(iz0,jz2);
 575             dx10             = _mm256_sub_pd(ix1,jx0);
 576             dy10             = _mm256_sub_pd(iy1,jy0);
 577             dz10             = _mm256_sub_pd(iz1,jz0);
 578             dx11             = _mm256_sub_pd(ix1,jx1);
 579             dy11             = _mm256_sub_pd(iy1,jy1);
 580             dz11             = _mm256_sub_pd(iz1,jz1);
 581             dx12             = _mm256_sub_pd(ix1,jx2);
 582             dy12             = _mm256_sub_pd(iy1,jy2);
 583             dz12             = _mm256_sub_pd(iz1,jz2);
 584             dx20             = _mm256_sub_pd(ix2,jx0);
 585             dy20             = _mm256_sub_pd(iy2,jy0);
 586             dz20             = _mm256_sub_pd(iz2,jz0);
 587             dx21             = _mm256_sub_pd(ix2,jx1);
 588             dy21             = _mm256_sub_pd(iy2,jy1);
 589             dz21             = _mm256_sub_pd(iz2,jz1);
 590             dx22             = _mm256_sub_pd(ix2,jx2);
 591             dy22             = _mm256_sub_pd(iy2,jy2);
 592             dz22             = _mm256_sub_pd(iz2,jz2);
 593
 594             /* Calculate squared distance and things based on it */
 595             rsq00            = gmx_mm256_calc_rsq_pd(dx00,dy00,dz00);
 596             rsq01            = gmx_mm256_calc_rsq_pd(dx01,dy01,dz01);
 597             rsq02            = gmx_mm256_calc_rsq_pd(dx02,dy02,dz02);
 598             rsq10            = gmx_mm256_calc_rsq_pd(dx10,dy10,dz10);
 599             rsq11            = gmx_mm256_calc_rsq_pd(dx11,dy11,dz11);
 600             rsq12            = gmx_mm256_calc_rsq_pd(dx12,dy12,dz12);
 601             rsq20            = gmx_mm256_calc_rsq_pd(dx20,dy20,dz20);
 602             rsq21            = gmx_mm256_calc_rsq_pd(dx21,dy21,dz21);
 603             rsq22            = gmx_mm256_calc_rsq_pd(dx22,dy22,dz22);
 604
 605             rinv00           = avx256_invsqrt_d(rsq00);
 606             rinv01           = avx256_invsqrt_d(rsq01);
 607             rinv02           = avx256_invsqrt_d(rsq02);
 608             rinv10           = avx256_invsqrt_d(rsq10);
 609             rinv11           = avx256_invsqrt_d(rsq11);
 610             rinv12           = avx256_invsqrt_d(rsq12);
 611             rinv20           = avx256_invsqrt_d(rsq20);
 612             rinv21           = avx256_invsqrt_d(rsq21);
 613             rinv22           = avx256_invsqrt_d(rsq22);
 614
 615             rinvsq00         = _mm256_mul_pd(rinv00,rinv00);
 616             rinvsq01         = _mm256_mul_pd(rinv01,rinv01);
 617             rinvsq02         = _mm256_mul_pd(rinv02,rinv02);
 618             rinvsq10         = _mm256_mul_pd(rinv10,rinv10);
 619             rinvsq11         = _mm256_mul_pd(rinv11,rinv11);
 620             rinvsq12         = _mm256_mul_pd(rinv12,rinv12);
 621             rinvsq20         = _mm256_mul_pd(rinv20,rinv20);
 622             rinvsq21         = _mm256_mul_pd(rinv21,rinv21);
 623             rinvsq22         = _mm256_mul_pd(rinv22,rinv22);
 624
 625             fjx0             = _mm256_setzero_pd();
 626             fjy0             = _mm256_setzero_pd();
 627             fjz0             = _mm256_setzero_pd();
 628             fjx1             = _mm256_setzero_pd();
 629             fjy1             = _mm256_setzero_pd();
 630             fjz1             = _mm256_setzero_pd();
 631             fjx2             = _mm256_setzero_pd();
 632             fjy2             = _mm256_setzero_pd();
 633             fjz2             = _mm256_setzero_pd();
 634
 635             /**************************
 636              * CALCULATE INTERACTIONS *
 637              **************************/
 638
 639             /* COULOMB ELECTROSTATICS */
 640             velec            = _mm256_mul_pd(qq00,rinv00);
 641             felec            = _mm256_mul_pd(velec,rinvsq00);
 642
 643             /* Update potential sum for this i atom from the interaction with this j atom. */
 644             velec            = _mm256_andnot_pd(dummy_mask,velec);
 645             velecsum         = _mm256_add_pd(velecsum,velec);
 646
 647             fscal            = felec;
 648
 649             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
 650
 651             /* Calculate temporary vectorial force */
 652             tx               = _mm256_mul_pd(fscal,dx00);
 653             ty               = _mm256_mul_pd(fscal,dy00);
 654             tz               = _mm256_mul_pd(fscal,dz00);
 655
 656             /* Update vectorial force */
 657             fix0             = _mm256_add_pd(fix0,tx);
 658             fiy0             = _mm256_add_pd(fiy0,ty);
 659             fiz0             = _mm256_add_pd(fiz0,tz);
 660
 661             fjx0             = _mm256_add_pd(fjx0,tx);
 662             fjy0             = _mm256_add_pd(fjy0,ty);
 663             fjz0             = _mm256_add_pd(fjz0,tz);
 664
 665             /**************************
 666              * CALCULATE INTERACTIONS *
 667              **************************/
 668
 669             /* COULOMB ELECTROSTATICS */
 670             velec            = _mm256_mul_pd(qq01,rinv01);
 671             felec            = _mm256_mul_pd(velec,rinvsq01);
 672
 673             /* Update potential sum for this i atom from the interaction with this j atom. */
 674             velec            = _mm256_andnot_pd(dummy_mask,velec);
 675             velecsum         = _mm256_add_pd(velecsum,velec);
 676
 677             fscal            = felec;
 678
 679             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
 680
 681             /* Calculate temporary vectorial force */
 682             tx               = _mm256_mul_pd(fscal,dx01);
 683             ty               = _mm256_mul_pd(fscal,dy01);
 684             tz               = _mm256_mul_pd(fscal,dz01);
 685
 686             /* Update vectorial force */
 687             fix0             = _mm256_add_pd(fix0,tx);
 688             fiy0             = _mm256_add_pd(fiy0,ty);
 689             fiz0             = _mm256_add_pd(fiz0,tz);
 690
 691             fjx1             = _mm256_add_pd(fjx1,tx);
 692             fjy1             = _mm256_add_pd(fjy1,ty);
 693             fjz1             = _mm256_add_pd(fjz1,tz);
 694
 695             /**************************
 696              * CALCULATE INTERACTIONS *
 697              **************************/
 698
 699             /* COULOMB ELECTROSTATICS */
 700             velec            = _mm256_mul_pd(qq02,rinv02);
 701             felec            = _mm256_mul_pd(velec,rinvsq02);
 702
 703             /* Update potential sum for this i atom from the interaction with this j atom. */
 704             velec            = _mm256_andnot_pd(dummy_mask,velec);
 705             velecsum         = _mm256_add_pd(velecsum,velec);
 706
 707             fscal            = felec;
 708
 709             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
 710
 711             /* Calculate temporary vectorial force */
 712             tx               = _mm256_mul_pd(fscal,dx02);
 713             ty               = _mm256_mul_pd(fscal,dy02);
 714             tz               = _mm256_mul_pd(fscal,dz02);
 715
 716             /* Update vectorial force */
 717             fix0             = _mm256_add_pd(fix0,tx);
 718             fiy0             = _mm256_add_pd(fiy0,ty);
 719             fiz0             = _mm256_add_pd(fiz0,tz);
 720
 721             fjx2             = _mm256_add_pd(fjx2,tx);
 722             fjy2             = _mm256_add_pd(fjy2,ty);
 723             fjz2             = _mm256_add_pd(fjz2,tz);
 724
 725             /**************************
 726              * CALCULATE INTERACTIONS *
 727              **************************/
 728
 729             /* COULOMB ELECTROSTATICS */
 730             velec            = _mm256_mul_pd(qq10,rinv10);
 731             felec            = _mm256_mul_pd(velec,rinvsq10);
 732
 733             /* Update potential sum for this i atom from the interaction with this j atom. */
 734             velec            = _mm256_andnot_pd(dummy_mask,velec);
 735             velecsum         = _mm256_add_pd(velecsum,velec);
 736
 737             fscal            = felec;
 738
 739             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
 740
 741             /* Calculate temporary vectorial force */
 742             tx               = _mm256_mul_pd(fscal,dx10);
 743             ty               = _mm256_mul_pd(fscal,dy10);
 744             tz               = _mm256_mul_pd(fscal,dz10);
 745
 746             /* Update vectorial force */
 747             fix1             = _mm256_add_pd(fix1,tx);
 748             fiy1             = _mm256_add_pd(fiy1,ty);
 749             fiz1             = _mm256_add_pd(fiz1,tz);
 750
 751             fjx0             = _mm256_add_pd(fjx0,tx);
 752             fjy0             = _mm256_add_pd(fjy0,ty);
 753             fjz0             = _mm256_add_pd(fjz0,tz);
 754
 755             /**************************
 756              * CALCULATE INTERACTIONS *
 757              **************************/
 758
 759             /* COULOMB ELECTROSTATICS */
 760             velec            = _mm256_mul_pd(qq11,rinv11);
 761             felec            = _mm256_mul_pd(velec,rinvsq11);
 762
 763             /* Update potential sum for this i atom from the interaction with this j atom. */
 764             velec            = _mm256_andnot_pd(dummy_mask,velec);
 765             velecsum         = _mm256_add_pd(velecsum,velec);
 766
 767             fscal            = felec;
 768
 769             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
 770
 771             /* Calculate temporary vectorial force */
 772             tx               = _mm256_mul_pd(fscal,dx11);
 773             ty               = _mm256_mul_pd(fscal,dy11);
 774             tz               = _mm256_mul_pd(fscal,dz11);
 775
 776             /* Update vectorial force */
 777             fix1             = _mm256_add_pd(fix1,tx);
 778             fiy1             = _mm256_add_pd(fiy1,ty);
 779             fiz1             = _mm256_add_pd(fiz1,tz);
 780
 781             fjx1             = _mm256_add_pd(fjx1,tx);
 782             fjy1             = _mm256_add_pd(fjy1,ty);
 783             fjz1             = _mm256_add_pd(fjz1,tz);
 784
 785             /**************************
 786              * CALCULATE INTERACTIONS *
 787              **************************/
 788
 789             /* COULOMB ELECTROSTATICS */
 790             velec            = _mm256_mul_pd(qq12,rinv12);
 791             felec            = _mm256_mul_pd(velec,rinvsq12);
 792
 793             /* Update potential sum for this i atom from the interaction with this j atom. */
 794             velec            = _mm256_andnot_pd(dummy_mask,velec);
 795             velecsum         = _mm256_add_pd(velecsum,velec);
 796
 797             fscal            = felec;
 798
 799             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
 800
 801             /* Calculate temporary vectorial force */
 802             tx               = _mm256_mul_pd(fscal,dx12);
 803             ty               = _mm256_mul_pd(fscal,dy12);
 804             tz               = _mm256_mul_pd(fscal,dz12);
 805
 806             /* Update vectorial force */
 807             fix1             = _mm256_add_pd(fix1,tx);
 808             fiy1             = _mm256_add_pd(fiy1,ty);
 809             fiz1             = _mm256_add_pd(fiz1,tz);
 810
 811             fjx2             = _mm256_add_pd(fjx2,tx);
 812             fjy2             = _mm256_add_pd(fjy2,ty);
 813             fjz2             = _mm256_add_pd(fjz2,tz);
 814
 815             /**************************
 816              * CALCULATE INTERACTIONS *
 817              **************************/
 818
 819             /* COULOMB ELECTROSTATICS */
 820             velec            = _mm256_mul_pd(qq20,rinv20);
 821             felec            = _mm256_mul_pd(velec,rinvsq20);
 822
 823             /* Update potential sum for this i atom from the interaction with this j atom. */
 824             velec            = _mm256_andnot_pd(dummy_mask,velec);
 825             velecsum         = _mm256_add_pd(velecsum,velec);
 826
 827             fscal            = felec;
 828
 829             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
 830
 831             /* Calculate temporary vectorial force */
 832             tx               = _mm256_mul_pd(fscal,dx20);
 833             ty               = _mm256_mul_pd(fscal,dy20);
 834             tz               = _mm256_mul_pd(fscal,dz20);
 835
 836             /* Update vectorial force */
 837             fix2             = _mm256_add_pd(fix2,tx);
 838             fiy2             = _mm256_add_pd(fiy2,ty);
 839             fiz2             = _mm256_add_pd(fiz2,tz);
 840
 841             fjx0             = _mm256_add_pd(fjx0,tx);
 842             fjy0             = _mm256_add_pd(fjy0,ty);
 843             fjz0             = _mm256_add_pd(fjz0,tz);
 844
 845             /**************************
 846              * CALCULATE INTERACTIONS *
 847              **************************/
 848
 849             /* COULOMB ELECTROSTATICS */
 850             velec            = _mm256_mul_pd(qq21,rinv21);
 851             felec            = _mm256_mul_pd(velec,rinvsq21);
 852
 853             /* Update potential sum for this i atom from the interaction with this j atom. */
 854             velec            = _mm256_andnot_pd(dummy_mask,velec);
 855             velecsum         = _mm256_add_pd(velecsum,velec);
 856
 857             fscal            = felec;
 858
 859             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
 860
 861             /* Calculate temporary vectorial force */
 862             tx               = _mm256_mul_pd(fscal,dx21);
 863             ty               = _mm256_mul_pd(fscal,dy21);
 864             tz               = _mm256_mul_pd(fscal,dz21);
 865
 866             /* Update vectorial force */
 867             fix2             = _mm256_add_pd(fix2,tx);
 868             fiy2             = _mm256_add_pd(fiy2,ty);
 869             fiz2             = _mm256_add_pd(fiz2,tz);
 870
 871             fjx1             = _mm256_add_pd(fjx1,tx);
 872             fjy1             = _mm256_add_pd(fjy1,ty);
 873             fjz1             = _mm256_add_pd(fjz1,tz);
 874
 875             /**************************
 876              * CALCULATE INTERACTIONS *
 877              **************************/
 878
 879             /* COULOMB ELECTROSTATICS */
 880             velec            = _mm256_mul_pd(qq22,rinv22);
 881             felec            = _mm256_mul_pd(velec,rinvsq22);
 882
 883             /* Update potential sum for this i atom from the interaction with this j atom. */
 884             velec            = _mm256_andnot_pd(dummy_mask,velec);
 885             velecsum         = _mm256_add_pd(velecsum,velec);
 886
 887             fscal            = felec;
 888
 889             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
 890
 891             /* Calculate temporary vectorial force */
 892             tx               = _mm256_mul_pd(fscal,dx22);
 893             ty               = _mm256_mul_pd(fscal,dy22);
 894             tz               = _mm256_mul_pd(fscal,dz22);
 895
 896             /* Update vectorial force */
 897             fix2             = _mm256_add_pd(fix2,tx);
 898             fiy2             = _mm256_add_pd(fiy2,ty);
 899             fiz2             = _mm256_add_pd(fiz2,tz);
 900
 901             fjx2             = _mm256_add_pd(fjx2,tx);
 902             fjy2             = _mm256_add_pd(fjy2,ty);
 903             fjz2             = _mm256_add_pd(fjz2,tz);
 904
 905             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 906             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 907             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 908             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 909
 910             gmx_mm256_decrement_3rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,
 911                                                       fjx0,fjy0,fjz0,fjx1,fjy1,fjz1,fjx2,fjy2,fjz2);
 912
 913             /* Inner loop uses 243 flops */
 914         }
 915
 916         /* End of innermost loop */
 917
 918         gmx_mm256_update_iforce_3atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
 919                                                  f+i_coord_offset,fshift+i_shift_offset);
 920
 921         ggid                        = gid[iidx];
 922         /* Update potential energies */
 923         gmx_mm256_update_1pot_pd(velecsum,kernel_data->energygrp_elec+ggid);
 924
 925         /* Increment number of inner iterations */
 926         inneriter                  += j_index_end - j_index_start;
 927
 928         /* Outer loop uses 19 flops */
 929     }
 930
 931     /* Increment number of outer iterations */
 932     outeriter        += nri;
 933
 934     /* Update outer/inner flops */
 935
 936     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_W3W3_VF,outeriter*19 + inneriter*243);
 937 }
 938 /*
 939  * Gromacs nonbonded kernel:   nb_kernel_ElecCoul_VdwNone_GeomW3W3_F_avx_256_double
 940  * Electrostatics interaction: Coulomb
 941  * VdW interaction:            None
 942  * Geometry:                   Water3-Water3
 943  * Calculate force/pot:        Force
 944  */
 945 void
 946 nb_kernel_ElecCoul_VdwNone_GeomW3W3_F_avx_256_double
 947                     (t_nblist                    * gmx_restrict       nlist,
 948                      rvec                        * gmx_restrict          xx,
 949                      rvec                        * gmx_restrict          ff,
 950                      struct t_forcerec           * gmx_restrict          fr,
 951                      t_mdatoms                   * gmx_restrict     mdatoms,
 952                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 953                      t_nrnb                      * gmx_restrict        nrnb)
 954 {
 955     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 956      * just 0 for non-waters.
 957      * Suffixes A,B,C,D refer to j loop unrolling done with AVX, e.g. for the four different
 958      * jnr indices corresponding to data put in the four positions in the SIMD register.
 959      */
 960     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 961     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 962     int              jnrA,jnrB,jnrC,jnrD;
 963     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
 964     int              jnrlistE,jnrlistF,jnrlistG,jnrlistH;
 965     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 966     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 967     real             rcutoff_scalar;
 968     real             *shiftvec,*fshift,*x,*f;
 969     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
 970     real             scratch[4*DIM];
 971     __m256d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 972     real *           vdwioffsetptr0;
 973     __m256d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 974     real *           vdwioffsetptr1;
 975     __m256d          ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
 976     real *           vdwioffsetptr2;
 977     __m256d          ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
 978     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
 979     __m256d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 980     int              vdwjidx1A,vdwjidx1B,vdwjidx1C,vdwjidx1D;
 981     __m256d          jx1,jy1,jz1,fjx1,fjy1,fjz1,jq1,isaj1;
 982     int              vdwjidx2A,vdwjidx2B,vdwjidx2C,vdwjidx2D;
 983     __m256d          jx2,jy2,jz2,fjx2,fjy2,fjz2,jq2,isaj2;
 984     __m256d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 985     __m256d          dx01,dy01,dz01,rsq01,rinv01,rinvsq01,r01,qq01,c6_01,c12_01;
 986     __m256d          dx02,dy02,dz02,rsq02,rinv02,rinvsq02,r02,qq02,c6_02,c12_02;
 987     __m256d          dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
 988     __m256d          dx11,dy11,dz11,rsq11,rinv11,rinvsq11,r11,qq11,c6_11,c12_11;
 989     __m256d          dx12,dy12,dz12,rsq12,rinv12,rinvsq12,r12,qq12,c6_12,c12_12;
 990     __m256d          dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
 991     __m256d          dx21,dy21,dz21,rsq21,rinv21,rinvsq21,r21,qq21,c6_21,c12_21;
 992     __m256d          dx22,dy22,dz22,rsq22,rinv22,rinvsq22,r22,qq22,c6_22,c12_22;
 993     __m256d          velec,felec,velecsum,facel,crf,krf,krf2;
 994     real             *charge;
 995     __m256d          dummy_mask,cutoff_mask;
 996     __m128           tmpmask0,tmpmask1;
 997     __m256d          signbit = _mm256_castsi256_pd( _mm256_set1_epi32(0x80000000) );
 998     __m256d          one     = _mm256_set1_pd(1.0);
 999     __m256d          two     = _mm256_set1_pd(2.0);
1000     x                = xx[0];
1001     f                = ff[0];
1002
1003     nri              = nlist->nri;
1004     iinr             = nlist->iinr;
1005     jindex           = nlist->jindex;
1006     jjnr             = nlist->jjnr;
1007     shiftidx         = nlist->shift;
1008     gid              = nlist->gid;
1009     shiftvec         = fr->shift_vec[0];
1010     fshift           = fr->fshift[0];
1011     facel            = _mm256_set1_pd(fr->ic->epsfac);
1012     charge           = mdatoms->chargeA;
1013
1014     /* Setup water-specific parameters */
1015     inr              = nlist->iinr[0];
1016     iq0              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+0]));
1017     iq1              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+1]));
1018     iq2              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+2]));
1019
1020     jq0              = _mm256_set1_pd(charge[inr+0]);
1021     jq1              = _mm256_set1_pd(charge[inr+1]);
1022     jq2              = _mm256_set1_pd(charge[inr+2]);
1023     qq00             = _mm256_mul_pd(iq0,jq0);
1024     qq01             = _mm256_mul_pd(iq0,jq1);
1025     qq02             = _mm256_mul_pd(iq0,jq2);
1026     qq10             = _mm256_mul_pd(iq1,jq0);
1027     qq11             = _mm256_mul_pd(iq1,jq1);
1028     qq12             = _mm256_mul_pd(iq1,jq2);
1029     qq20             = _mm256_mul_pd(iq2,jq0);
1030     qq21             = _mm256_mul_pd(iq2,jq1);
1031     qq22             = _mm256_mul_pd(iq2,jq2);
1032
1033     /* Avoid stupid compiler warnings */
1034     jnrA = jnrB = jnrC = jnrD = 0;
1035     j_coord_offsetA = 0;
1036     j_coord_offsetB = 0;
1037     j_coord_offsetC = 0;
1038     j_coord_offsetD = 0;
1039
1040     outeriter        = 0;
1041     inneriter        = 0;
1042
1043     for(iidx=0;iidx<4*DIM;iidx++)
1044     {
1045         scratch[iidx] = 0.0;
1046     }
1047
1048     /* Start outer loop over neighborlists */
1049     for(iidx=0; iidx<nri; iidx++)
1050     {
1051         /* Load shift vector for this list */
1052         i_shift_offset   = DIM*shiftidx[iidx];
1053
1054         /* Load limits for loop over neighbors */
1055         j_index_start    = jindex[iidx];
1056         j_index_end      = jindex[iidx+1];
1057
1058         /* Get outer coordinate index */
1059         inr              = iinr[iidx];
1060         i_coord_offset   = DIM*inr;
1061
1062         /* Load i particle coords and add shift vector */
1063         gmx_mm256_load_shift_and_3rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
1064                                                     &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
1065
1066         fix0             = _mm256_setzero_pd();
1067         fiy0             = _mm256_setzero_pd();
1068         fiz0             = _mm256_setzero_pd();
1069         fix1             = _mm256_setzero_pd();
1070         fiy1             = _mm256_setzero_pd();
1071         fiz1             = _mm256_setzero_pd();
1072         fix2             = _mm256_setzero_pd();
1073         fiy2             = _mm256_setzero_pd();
1074         fiz2             = _mm256_setzero_pd();
1075
1076         /* Start inner kernel loop */
1077         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
1078         {
1079
1080             /* Get j neighbor index, and coordinate index */
1081             jnrA             = jjnr[jidx];
1082             jnrB             = jjnr[jidx+1];
1083             jnrC             = jjnr[jidx+2];
1084             jnrD             = jjnr[jidx+3];
1085             j_coord_offsetA  = DIM*jnrA;
1086             j_coord_offsetB  = DIM*jnrB;
1087             j_coord_offsetC  = DIM*jnrC;
1088             j_coord_offsetD  = DIM*jnrD;
1089
1090             /* load j atom coordinates */
1091             gmx_mm256_load_3rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
1092                                                  x+j_coord_offsetC,x+j_coord_offsetD,
1093                                               &jx0,&jy0,&jz0,&jx1,&jy1,&jz1,&jx2,&jy2,&jz2);
1094
1095             /* Calculate displacement vector */
1096             dx00             = _mm256_sub_pd(ix0,jx0);
1097             dy00             = _mm256_sub_pd(iy0,jy0);
1098             dz00             = _mm256_sub_pd(iz0,jz0);
1099             dx01             = _mm256_sub_pd(ix0,jx1);
1100             dy01             = _mm256_sub_pd(iy0,jy1);
1101             dz01             = _mm256_sub_pd(iz0,jz1);
1102             dx02             = _mm256_sub_pd(ix0,jx2);
1103             dy02             = _mm256_sub_pd(iy0,jy2);
1104             dz02             = _mm256_sub_pd(iz0,jz2);
1105             dx10             = _mm256_sub_pd(ix1,jx0);
1106             dy10             = _mm256_sub_pd(iy1,jy0);
1107             dz10             = _mm256_sub_pd(iz1,jz0);
1108             dx11             = _mm256_sub_pd(ix1,jx1);
1109             dy11             = _mm256_sub_pd(iy1,jy1);
1110             dz11             = _mm256_sub_pd(iz1,jz1);
1111             dx12             = _mm256_sub_pd(ix1,jx2);
1112             dy12             = _mm256_sub_pd(iy1,jy2);
1113             dz12             = _mm256_sub_pd(iz1,jz2);
1114             dx20             = _mm256_sub_pd(ix2,jx0);
1115             dy20             = _mm256_sub_pd(iy2,jy0);
1116             dz20             = _mm256_sub_pd(iz2,jz0);
1117             dx21             = _mm256_sub_pd(ix2,jx1);
1118             dy21             = _mm256_sub_pd(iy2,jy1);
1119             dz21             = _mm256_sub_pd(iz2,jz1);
1120             dx22             = _mm256_sub_pd(ix2,jx2);
1121             dy22             = _mm256_sub_pd(iy2,jy2);
1122             dz22             = _mm256_sub_pd(iz2,jz2);
1123
1124             /* Calculate squared distance and things based on it */
1125             rsq00            = gmx_mm256_calc_rsq_pd(dx00,dy00,dz00);
1126             rsq01            = gmx_mm256_calc_rsq_pd(dx01,dy01,dz01);
1127             rsq02            = gmx_mm256_calc_rsq_pd(dx02,dy02,dz02);
1128             rsq10            = gmx_mm256_calc_rsq_pd(dx10,dy10,dz10);
1129             rsq11            = gmx_mm256_calc_rsq_pd(dx11,dy11,dz11);
1130             rsq12            = gmx_mm256_calc_rsq_pd(dx12,dy12,dz12);
1131             rsq20            = gmx_mm256_calc_rsq_pd(dx20,dy20,dz20);
1132             rsq21            = gmx_mm256_calc_rsq_pd(dx21,dy21,dz21);
1133             rsq22            = gmx_mm256_calc_rsq_pd(dx22,dy22,dz22);
1134
1135             rinv00           = avx256_invsqrt_d(rsq00);
1136             rinv01           = avx256_invsqrt_d(rsq01);
1137             rinv02           = avx256_invsqrt_d(rsq02);
1138             rinv10           = avx256_invsqrt_d(rsq10);
1139             rinv11           = avx256_invsqrt_d(rsq11);
1140             rinv12           = avx256_invsqrt_d(rsq12);
1141             rinv20           = avx256_invsqrt_d(rsq20);
1142             rinv21           = avx256_invsqrt_d(rsq21);
1143             rinv22           = avx256_invsqrt_d(rsq22);
1144
1145             rinvsq00         = _mm256_mul_pd(rinv00,rinv00);
1146             rinvsq01         = _mm256_mul_pd(rinv01,rinv01);
1147             rinvsq02         = _mm256_mul_pd(rinv02,rinv02);
1148             rinvsq10         = _mm256_mul_pd(rinv10,rinv10);
1149             rinvsq11         = _mm256_mul_pd(rinv11,rinv11);
1150             rinvsq12         = _mm256_mul_pd(rinv12,rinv12);
1151             rinvsq20         = _mm256_mul_pd(rinv20,rinv20);
1152             rinvsq21         = _mm256_mul_pd(rinv21,rinv21);
1153             rinvsq22         = _mm256_mul_pd(rinv22,rinv22);
1154
1155             fjx0             = _mm256_setzero_pd();
1156             fjy0             = _mm256_setzero_pd();
1157             fjz0             = _mm256_setzero_pd();
1158             fjx1             = _mm256_setzero_pd();
1159             fjy1             = _mm256_setzero_pd();
1160             fjz1             = _mm256_setzero_pd();
1161             fjx2             = _mm256_setzero_pd();
1162             fjy2             = _mm256_setzero_pd();
1163             fjz2             = _mm256_setzero_pd();
1164
1165             /**************************
1166              * CALCULATE INTERACTIONS *
1167              **************************/
1168
1169             /* COULOMB ELECTROSTATICS */
1170             velec            = _mm256_mul_pd(qq00,rinv00);
1171             felec            = _mm256_mul_pd(velec,rinvsq00);
1172
1173             fscal            = felec;
1174
1175             /* Calculate temporary vectorial force */
1176             tx               = _mm256_mul_pd(fscal,dx00);
1177             ty               = _mm256_mul_pd(fscal,dy00);
1178             tz               = _mm256_mul_pd(fscal,dz00);
1179
1180             /* Update vectorial force */
1181             fix0             = _mm256_add_pd(fix0,tx);
1182             fiy0             = _mm256_add_pd(fiy0,ty);
1183             fiz0             = _mm256_add_pd(fiz0,tz);
1184
1185             fjx0             = _mm256_add_pd(fjx0,tx);
1186             fjy0             = _mm256_add_pd(fjy0,ty);
1187             fjz0             = _mm256_add_pd(fjz0,tz);
1188
1189             /**************************
1190              * CALCULATE INTERACTIONS *
1191              **************************/
1192
1193             /* COULOMB ELECTROSTATICS */
1194             velec            = _mm256_mul_pd(qq01,rinv01);
1195             felec            = _mm256_mul_pd(velec,rinvsq01);
1196
1197             fscal            = felec;
1198
1199             /* Calculate temporary vectorial force */
1200             tx               = _mm256_mul_pd(fscal,dx01);
1201             ty               = _mm256_mul_pd(fscal,dy01);
1202             tz               = _mm256_mul_pd(fscal,dz01);
1203
1204             /* Update vectorial force */
1205             fix0             = _mm256_add_pd(fix0,tx);
1206             fiy0             = _mm256_add_pd(fiy0,ty);
1207             fiz0             = _mm256_add_pd(fiz0,tz);
1208
1209             fjx1             = _mm256_add_pd(fjx1,tx);
1210             fjy1             = _mm256_add_pd(fjy1,ty);
1211             fjz1             = _mm256_add_pd(fjz1,tz);
1212
1213             /**************************
1214              * CALCULATE INTERACTIONS *
1215              **************************/
1216
1217             /* COULOMB ELECTROSTATICS */
1218             velec            = _mm256_mul_pd(qq02,rinv02);
1219             felec            = _mm256_mul_pd(velec,rinvsq02);
1220
1221             fscal            = felec;
1222
1223             /* Calculate temporary vectorial force */
1224             tx               = _mm256_mul_pd(fscal,dx02);
1225             ty               = _mm256_mul_pd(fscal,dy02);
1226             tz               = _mm256_mul_pd(fscal,dz02);
1227
1228             /* Update vectorial force */
1229             fix0             = _mm256_add_pd(fix0,tx);
1230             fiy0             = _mm256_add_pd(fiy0,ty);
1231             fiz0             = _mm256_add_pd(fiz0,tz);
1232
1233             fjx2             = _mm256_add_pd(fjx2,tx);
1234             fjy2             = _mm256_add_pd(fjy2,ty);
1235             fjz2             = _mm256_add_pd(fjz2,tz);
1236
1237             /**************************
1238              * CALCULATE INTERACTIONS *
1239              **************************/
1240
1241             /* COULOMB ELECTROSTATICS */
1242             velec            = _mm256_mul_pd(qq10,rinv10);
1243             felec            = _mm256_mul_pd(velec,rinvsq10);
1244
1245             fscal            = felec;
1246
1247             /* Calculate temporary vectorial force */
1248             tx               = _mm256_mul_pd(fscal,dx10);
1249             ty               = _mm256_mul_pd(fscal,dy10);
1250             tz               = _mm256_mul_pd(fscal,dz10);
1251
1252             /* Update vectorial force */
1253             fix1             = _mm256_add_pd(fix1,tx);
1254             fiy1             = _mm256_add_pd(fiy1,ty);
1255             fiz1             = _mm256_add_pd(fiz1,tz);
1256
1257             fjx0             = _mm256_add_pd(fjx0,tx);
1258             fjy0             = _mm256_add_pd(fjy0,ty);
1259             fjz0             = _mm256_add_pd(fjz0,tz);
1260
1261             /**************************
1262              * CALCULATE INTERACTIONS *
1263              **************************/
1264
1265             /* COULOMB ELECTROSTATICS */
1266             velec            = _mm256_mul_pd(qq11,rinv11);
1267             felec            = _mm256_mul_pd(velec,rinvsq11);
1268
1269             fscal            = felec;
1270
1271             /* Calculate temporary vectorial force */
1272             tx               = _mm256_mul_pd(fscal,dx11);
1273             ty               = _mm256_mul_pd(fscal,dy11);
1274             tz               = _mm256_mul_pd(fscal,dz11);
1275
1276             /* Update vectorial force */
1277             fix1             = _mm256_add_pd(fix1,tx);
1278             fiy1             = _mm256_add_pd(fiy1,ty);
1279             fiz1             = _mm256_add_pd(fiz1,tz);
1280
1281             fjx1             = _mm256_add_pd(fjx1,tx);
1282             fjy1             = _mm256_add_pd(fjy1,ty);
1283             fjz1             = _mm256_add_pd(fjz1,tz);
1284
1285             /**************************
1286              * CALCULATE INTERACTIONS *
1287              **************************/
1288
1289             /* COULOMB ELECTROSTATICS */
1290             velec            = _mm256_mul_pd(qq12,rinv12);
1291             felec            = _mm256_mul_pd(velec,rinvsq12);
1292
1293             fscal            = felec;
1294
1295             /* Calculate temporary vectorial force */
1296             tx               = _mm256_mul_pd(fscal,dx12);
1297             ty               = _mm256_mul_pd(fscal,dy12);
1298             tz               = _mm256_mul_pd(fscal,dz12);
1299
1300             /* Update vectorial force */
1301             fix1             = _mm256_add_pd(fix1,tx);
1302             fiy1             = _mm256_add_pd(fiy1,ty);
1303             fiz1             = _mm256_add_pd(fiz1,tz);
1304
1305             fjx2             = _mm256_add_pd(fjx2,tx);
1306             fjy2             = _mm256_add_pd(fjy2,ty);
1307             fjz2             = _mm256_add_pd(fjz2,tz);
1308
1309             /**************************
1310              * CALCULATE INTERACTIONS *
1311              **************************/
1312
1313             /* COULOMB ELECTROSTATICS */
1314             velec            = _mm256_mul_pd(qq20,rinv20);
1315             felec            = _mm256_mul_pd(velec,rinvsq20);
1316
1317             fscal            = felec;
1318
1319             /* Calculate temporary vectorial force */
1320             tx               = _mm256_mul_pd(fscal,dx20);
1321             ty               = _mm256_mul_pd(fscal,dy20);
1322             tz               = _mm256_mul_pd(fscal,dz20);
1323
1324             /* Update vectorial force */
1325             fix2             = _mm256_add_pd(fix2,tx);
1326             fiy2             = _mm256_add_pd(fiy2,ty);
1327             fiz2             = _mm256_add_pd(fiz2,tz);
1328
1329             fjx0             = _mm256_add_pd(fjx0,tx);
1330             fjy0             = _mm256_add_pd(fjy0,ty);
1331             fjz0             = _mm256_add_pd(fjz0,tz);
1332
1333             /**************************
1334              * CALCULATE INTERACTIONS *
1335              **************************/
1336
1337             /* COULOMB ELECTROSTATICS */
1338             velec            = _mm256_mul_pd(qq21,rinv21);
1339             felec            = _mm256_mul_pd(velec,rinvsq21);
1340
1341             fscal            = felec;
1342
1343             /* Calculate temporary vectorial force */
1344             tx               = _mm256_mul_pd(fscal,dx21);
1345             ty               = _mm256_mul_pd(fscal,dy21);
1346             tz               = _mm256_mul_pd(fscal,dz21);
1347
1348             /* Update vectorial force */
1349             fix2             = _mm256_add_pd(fix2,tx);
1350             fiy2             = _mm256_add_pd(fiy2,ty);
1351             fiz2             = _mm256_add_pd(fiz2,tz);
1352
1353             fjx1             = _mm256_add_pd(fjx1,tx);
1354             fjy1             = _mm256_add_pd(fjy1,ty);
1355             fjz1             = _mm256_add_pd(fjz1,tz);
1356
1357             /**************************
1358              * CALCULATE INTERACTIONS *
1359              **************************/
1360
1361             /* COULOMB ELECTROSTATICS */
1362             velec            = _mm256_mul_pd(qq22,rinv22);
1363             felec            = _mm256_mul_pd(velec,rinvsq22);
1364
1365             fscal            = felec;
1366
1367             /* Calculate temporary vectorial force */
1368             tx               = _mm256_mul_pd(fscal,dx22);
1369             ty               = _mm256_mul_pd(fscal,dy22);
1370             tz               = _mm256_mul_pd(fscal,dz22);
1371
1372             /* Update vectorial force */
1373             fix2             = _mm256_add_pd(fix2,tx);
1374             fiy2             = _mm256_add_pd(fiy2,ty);
1375             fiz2             = _mm256_add_pd(fiz2,tz);
1376
1377             fjx2             = _mm256_add_pd(fjx2,tx);
1378             fjy2             = _mm256_add_pd(fjy2,ty);
1379             fjz2             = _mm256_add_pd(fjz2,tz);
1380
1381             fjptrA             = f+j_coord_offsetA;
1382             fjptrB             = f+j_coord_offsetB;
1383             fjptrC             = f+j_coord_offsetC;
1384             fjptrD             = f+j_coord_offsetD;
1385
1386             gmx_mm256_decrement_3rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,
1387                                                       fjx0,fjy0,fjz0,fjx1,fjy1,fjz1,fjx2,fjy2,fjz2);
1388
1389             /* Inner loop uses 234 flops */
1390         }
1391
1392         if(jidx<j_index_end)
1393         {
1394
1395             /* Get j neighbor index, and coordinate index */
1396             jnrlistA         = jjnr[jidx];
1397             jnrlistB         = jjnr[jidx+1];
1398             jnrlistC         = jjnr[jidx+2];
1399             jnrlistD         = jjnr[jidx+3];
1400             /* Sign of each element will be negative for non-real atoms.
1401              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
1402              * so use it as val = _mm_andnot_pd(mask,val) to clear dummy entries.
1403              */
1404             tmpmask0 = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
1405
1406             tmpmask1 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(3,3,2,2));
1407             tmpmask0 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(1,1,0,0));
1408             dummy_mask = _mm256_castps_pd(gmx_mm256_set_m128(tmpmask1,tmpmask0));
1409
1410             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
1411             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
1412             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
1413             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
1414             j_coord_offsetA  = DIM*jnrA;
1415             j_coord_offsetB  = DIM*jnrB;
1416             j_coord_offsetC  = DIM*jnrC;
1417             j_coord_offsetD  = DIM*jnrD;
1418
1419             /* load j atom coordinates */
1420             gmx_mm256_load_3rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
1421                                                  x+j_coord_offsetC,x+j_coord_offsetD,
1422                                               &jx0,&jy0,&jz0,&jx1,&jy1,&jz1,&jx2,&jy2,&jz2);
1423
1424             /* Calculate displacement vector */
1425             dx00             = _mm256_sub_pd(ix0,jx0);
1426             dy00             = _mm256_sub_pd(iy0,jy0);
1427             dz00             = _mm256_sub_pd(iz0,jz0);
1428             dx01             = _mm256_sub_pd(ix0,jx1);
1429             dy01             = _mm256_sub_pd(iy0,jy1);
1430             dz01             = _mm256_sub_pd(iz0,jz1);
1431             dx02             = _mm256_sub_pd(ix0,jx2);
1432             dy02             = _mm256_sub_pd(iy0,jy2);
1433             dz02             = _mm256_sub_pd(iz0,jz2);
1434             dx10             = _mm256_sub_pd(ix1,jx0);
1435             dy10             = _mm256_sub_pd(iy1,jy0);
1436             dz10             = _mm256_sub_pd(iz1,jz0);
1437             dx11             = _mm256_sub_pd(ix1,jx1);
1438             dy11             = _mm256_sub_pd(iy1,jy1);
1439             dz11             = _mm256_sub_pd(iz1,jz1);
1440             dx12             = _mm256_sub_pd(ix1,jx2);
1441             dy12             = _mm256_sub_pd(iy1,jy2);
1442             dz12             = _mm256_sub_pd(iz1,jz2);
1443             dx20             = _mm256_sub_pd(ix2,jx0);
1444             dy20             = _mm256_sub_pd(iy2,jy0);
1445             dz20             = _mm256_sub_pd(iz2,jz0);
1446             dx21             = _mm256_sub_pd(ix2,jx1);
1447             dy21             = _mm256_sub_pd(iy2,jy1);
1448             dz21             = _mm256_sub_pd(iz2,jz1);
1449             dx22             = _mm256_sub_pd(ix2,jx2);
1450             dy22             = _mm256_sub_pd(iy2,jy2);
1451             dz22             = _mm256_sub_pd(iz2,jz2);
1452
1453             /* Calculate squared distance and things based on it */
1454             rsq00            = gmx_mm256_calc_rsq_pd(dx00,dy00,dz00);
1455             rsq01            = gmx_mm256_calc_rsq_pd(dx01,dy01,dz01);
1456             rsq02            = gmx_mm256_calc_rsq_pd(dx02,dy02,dz02);
1457             rsq10            = gmx_mm256_calc_rsq_pd(dx10,dy10,dz10);
1458             rsq11            = gmx_mm256_calc_rsq_pd(dx11,dy11,dz11);
1459             rsq12            = gmx_mm256_calc_rsq_pd(dx12,dy12,dz12);
1460             rsq20            = gmx_mm256_calc_rsq_pd(dx20,dy20,dz20);
1461             rsq21            = gmx_mm256_calc_rsq_pd(dx21,dy21,dz21);
1462             rsq22            = gmx_mm256_calc_rsq_pd(dx22,dy22,dz22);
1463
1464             rinv00           = avx256_invsqrt_d(rsq00);
1465             rinv01           = avx256_invsqrt_d(rsq01);
1466             rinv02           = avx256_invsqrt_d(rsq02);
1467             rinv10           = avx256_invsqrt_d(rsq10);
1468             rinv11           = avx256_invsqrt_d(rsq11);
1469             rinv12           = avx256_invsqrt_d(rsq12);
1470             rinv20           = avx256_invsqrt_d(rsq20);
1471             rinv21           = avx256_invsqrt_d(rsq21);
1472             rinv22           = avx256_invsqrt_d(rsq22);
1473
1474             rinvsq00         = _mm256_mul_pd(rinv00,rinv00);
1475             rinvsq01         = _mm256_mul_pd(rinv01,rinv01);
1476             rinvsq02         = _mm256_mul_pd(rinv02,rinv02);
1477             rinvsq10         = _mm256_mul_pd(rinv10,rinv10);
1478             rinvsq11         = _mm256_mul_pd(rinv11,rinv11);
1479             rinvsq12         = _mm256_mul_pd(rinv12,rinv12);
1480             rinvsq20         = _mm256_mul_pd(rinv20,rinv20);
1481             rinvsq21         = _mm256_mul_pd(rinv21,rinv21);
1482             rinvsq22         = _mm256_mul_pd(rinv22,rinv22);
1483
1484             fjx0             = _mm256_setzero_pd();
1485             fjy0             = _mm256_setzero_pd();
1486             fjz0             = _mm256_setzero_pd();
1487             fjx1             = _mm256_setzero_pd();
1488             fjy1             = _mm256_setzero_pd();
1489             fjz1             = _mm256_setzero_pd();
1490             fjx2             = _mm256_setzero_pd();
1491             fjy2             = _mm256_setzero_pd();
1492             fjz2             = _mm256_setzero_pd();
1493
1494             /**************************
1495              * CALCULATE INTERACTIONS *
1496              **************************/
1497
1498             /* COULOMB ELECTROSTATICS */
1499             velec            = _mm256_mul_pd(qq00,rinv00);
1500             felec            = _mm256_mul_pd(velec,rinvsq00);
1501
1502             fscal            = felec;
1503
1504             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
1505
1506             /* Calculate temporary vectorial force */
1507             tx               = _mm256_mul_pd(fscal,dx00);
1508             ty               = _mm256_mul_pd(fscal,dy00);
1509             tz               = _mm256_mul_pd(fscal,dz00);
1510
1511             /* Update vectorial force */
1512             fix0             = _mm256_add_pd(fix0,tx);
1513             fiy0             = _mm256_add_pd(fiy0,ty);
1514             fiz0             = _mm256_add_pd(fiz0,tz);
1515
1516             fjx0             = _mm256_add_pd(fjx0,tx);
1517             fjy0             = _mm256_add_pd(fjy0,ty);
1518             fjz0             = _mm256_add_pd(fjz0,tz);
1519
1520             /**************************
1521              * CALCULATE INTERACTIONS *
1522              **************************/
1523
1524             /* COULOMB ELECTROSTATICS */
1525             velec            = _mm256_mul_pd(qq01,rinv01);
1526             felec            = _mm256_mul_pd(velec,rinvsq01);
1527
1528             fscal            = felec;
1529
1530             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
1531
1532             /* Calculate temporary vectorial force */
1533             tx               = _mm256_mul_pd(fscal,dx01);
1534             ty               = _mm256_mul_pd(fscal,dy01);
1535             tz               = _mm256_mul_pd(fscal,dz01);
1536
1537             /* Update vectorial force */
1538             fix0             = _mm256_add_pd(fix0,tx);
1539             fiy0             = _mm256_add_pd(fiy0,ty);
1540             fiz0             = _mm256_add_pd(fiz0,tz);
1541
1542             fjx1             = _mm256_add_pd(fjx1,tx);
1543             fjy1             = _mm256_add_pd(fjy1,ty);
1544             fjz1             = _mm256_add_pd(fjz1,tz);
1545
1546             /**************************
1547              * CALCULATE INTERACTIONS *
1548              **************************/
1549
1550             /* COULOMB ELECTROSTATICS */
1551             velec            = _mm256_mul_pd(qq02,rinv02);
1552             felec            = _mm256_mul_pd(velec,rinvsq02);
1553
1554             fscal            = felec;
1555
1556             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
1557
1558             /* Calculate temporary vectorial force */
1559             tx               = _mm256_mul_pd(fscal,dx02);
1560             ty               = _mm256_mul_pd(fscal,dy02);
1561             tz               = _mm256_mul_pd(fscal,dz02);
1562
1563             /* Update vectorial force */
1564             fix0             = _mm256_add_pd(fix0,tx);
1565             fiy0             = _mm256_add_pd(fiy0,ty);
1566             fiz0             = _mm256_add_pd(fiz0,tz);
1567
1568             fjx2             = _mm256_add_pd(fjx2,tx);
1569             fjy2             = _mm256_add_pd(fjy2,ty);
1570             fjz2             = _mm256_add_pd(fjz2,tz);
1571
1572             /**************************
1573              * CALCULATE INTERACTIONS *
1574              **************************/
1575
1576             /* COULOMB ELECTROSTATICS */
1577             velec            = _mm256_mul_pd(qq10,rinv10);
1578             felec            = _mm256_mul_pd(velec,rinvsq10);
1579
1580             fscal            = felec;
1581
1582             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
1583
1584             /* Calculate temporary vectorial force */
1585             tx               = _mm256_mul_pd(fscal,dx10);
1586             ty               = _mm256_mul_pd(fscal,dy10);
1587             tz               = _mm256_mul_pd(fscal,dz10);
1588
1589             /* Update vectorial force */
1590             fix1             = _mm256_add_pd(fix1,tx);
1591             fiy1             = _mm256_add_pd(fiy1,ty);
1592             fiz1             = _mm256_add_pd(fiz1,tz);
1593
1594             fjx0             = _mm256_add_pd(fjx0,tx);
1595             fjy0             = _mm256_add_pd(fjy0,ty);
1596             fjz0             = _mm256_add_pd(fjz0,tz);
1597
1598             /**************************
1599              * CALCULATE INTERACTIONS *
1600              **************************/
1601
1602             /* COULOMB ELECTROSTATICS */
1603             velec            = _mm256_mul_pd(qq11,rinv11);
1604             felec            = _mm256_mul_pd(velec,rinvsq11);
1605
1606             fscal            = felec;
1607
1608             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
1609
1610             /* Calculate temporary vectorial force */
1611             tx               = _mm256_mul_pd(fscal,dx11);
1612             ty               = _mm256_mul_pd(fscal,dy11);
1613             tz               = _mm256_mul_pd(fscal,dz11);
1614
1615             /* Update vectorial force */
1616             fix1             = _mm256_add_pd(fix1,tx);
1617             fiy1             = _mm256_add_pd(fiy1,ty);
1618             fiz1             = _mm256_add_pd(fiz1,tz);
1619
1620             fjx1             = _mm256_add_pd(fjx1,tx);
1621             fjy1             = _mm256_add_pd(fjy1,ty);
1622             fjz1             = _mm256_add_pd(fjz1,tz);
1623
1624             /**************************
1625              * CALCULATE INTERACTIONS *
1626              **************************/
1627
1628             /* COULOMB ELECTROSTATICS */
1629             velec            = _mm256_mul_pd(qq12,rinv12);
1630             felec            = _mm256_mul_pd(velec,rinvsq12);
1631
1632             fscal            = felec;
1633
1634             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
1635
1636             /* Calculate temporary vectorial force */
1637             tx               = _mm256_mul_pd(fscal,dx12);
1638             ty               = _mm256_mul_pd(fscal,dy12);
1639             tz               = _mm256_mul_pd(fscal,dz12);
1640
1641             /* Update vectorial force */
1642             fix1             = _mm256_add_pd(fix1,tx);
1643             fiy1             = _mm256_add_pd(fiy1,ty);
1644             fiz1             = _mm256_add_pd(fiz1,tz);
1645
1646             fjx2             = _mm256_add_pd(fjx2,tx);
1647             fjy2             = _mm256_add_pd(fjy2,ty);
1648             fjz2             = _mm256_add_pd(fjz2,tz);
1649
1650             /**************************
1651              * CALCULATE INTERACTIONS *
1652              **************************/
1653
1654             /* COULOMB ELECTROSTATICS */
1655             velec            = _mm256_mul_pd(qq20,rinv20);
1656             felec            = _mm256_mul_pd(velec,rinvsq20);
1657
1658             fscal            = felec;
1659
1660             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
1661
1662             /* Calculate temporary vectorial force */
1663             tx               = _mm256_mul_pd(fscal,dx20);
1664             ty               = _mm256_mul_pd(fscal,dy20);
1665             tz               = _mm256_mul_pd(fscal,dz20);
1666
1667             /* Update vectorial force */
1668             fix2             = _mm256_add_pd(fix2,tx);
1669             fiy2             = _mm256_add_pd(fiy2,ty);
1670             fiz2             = _mm256_add_pd(fiz2,tz);
1671
1672             fjx0             = _mm256_add_pd(fjx0,tx);
1673             fjy0             = _mm256_add_pd(fjy0,ty);
1674             fjz0             = _mm256_add_pd(fjz0,tz);
1675
1676             /**************************
1677              * CALCULATE INTERACTIONS *
1678              **************************/
1679
1680             /* COULOMB ELECTROSTATICS */
1681             velec            = _mm256_mul_pd(qq21,rinv21);
1682             felec            = _mm256_mul_pd(velec,rinvsq21);
1683
1684             fscal            = felec;
1685
1686             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
1687
1688             /* Calculate temporary vectorial force */
1689             tx               = _mm256_mul_pd(fscal,dx21);
1690             ty               = _mm256_mul_pd(fscal,dy21);
1691             tz               = _mm256_mul_pd(fscal,dz21);
1692
1693             /* Update vectorial force */
1694             fix2             = _mm256_add_pd(fix2,tx);
1695             fiy2             = _mm256_add_pd(fiy2,ty);
1696             fiz2             = _mm256_add_pd(fiz2,tz);
1697
1698             fjx1             = _mm256_add_pd(fjx1,tx);
1699             fjy1             = _mm256_add_pd(fjy1,ty);
1700             fjz1             = _mm256_add_pd(fjz1,tz);
1701
1702             /**************************
1703              * CALCULATE INTERACTIONS *
1704              **************************/
1705
1706             /* COULOMB ELECTROSTATICS */
1707             velec            = _mm256_mul_pd(qq22,rinv22);
1708             felec            = _mm256_mul_pd(velec,rinvsq22);
1709
1710             fscal            = felec;
1711
1712             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
1713
1714             /* Calculate temporary vectorial force */
1715             tx               = _mm256_mul_pd(fscal,dx22);
1716             ty               = _mm256_mul_pd(fscal,dy22);
1717             tz               = _mm256_mul_pd(fscal,dz22);
1718
1719             /* Update vectorial force */
1720             fix2             = _mm256_add_pd(fix2,tx);
1721             fiy2             = _mm256_add_pd(fiy2,ty);
1722             fiz2             = _mm256_add_pd(fiz2,tz);
1723
1724             fjx2             = _mm256_add_pd(fjx2,tx);
1725             fjy2             = _mm256_add_pd(fjy2,ty);
1726             fjz2             = _mm256_add_pd(fjz2,tz);
1727
1728             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
1729             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
1730             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
1731             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
1732
1733             gmx_mm256_decrement_3rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,
1734                                                       fjx0,fjy0,fjz0,fjx1,fjy1,fjz1,fjx2,fjy2,fjz2);
1735
1736             /* Inner loop uses 234 flops */
1737         }
1738
1739         /* End of innermost loop */
1740
1741         gmx_mm256_update_iforce_3atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
1742                                                  f+i_coord_offset,fshift+i_shift_offset);
1743
1744         /* Increment number of inner iterations */
1745         inneriter                  += j_index_end - j_index_start;
1746
1747         /* Outer loop uses 18 flops */
1748     }
1749
1750     /* Increment number of outer iterations */
1751     outeriter        += nri;
1752
1753     /* Update outer/inner flops */
1754
1755     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_W3W3_F,outeriter*18 + inneriter*234);
1756 }