src/gromacs/gmxlib/nonbonded/nb_kernel_avx_256_single/nb_kernel_ElecEw_VdwNone_GeomP1P1_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_GeomP1P1_VF_avx_256_single
  54  * Electrostatics interaction: Ewald
  55  * VdW interaction:            None
  56  * Geometry:                   Particle-Particle
  57  * Calculate force/pot:        PotentialAndForce
  58  */
  59 void
  60 nb_kernel_ElecEw_VdwNone_GeomP1P1_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     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D,vdwjidx0E,vdwjidx0F,vdwjidx0G,vdwjidx0H;
  91     __m256           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  92     __m256           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  93     __m256           velec,felec,velecsum,facel,crf,krf,krf2;
  94     real             *charge;
  95     __m256i          ewitab;
  96     __m128i          ewitab_lo,ewitab_hi;
  97     __m256           ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
  98     __m256           beta,beta2,beta3,zeta2,pmecorrF,pmecorrV,rinv3;
  99     real             *ewtab;
 100     __m256           dummy_mask,cutoff_mask;
 101     __m256           signbit = _mm256_castsi256_ps( _mm256_set1_epi32(0x80000000) );
 102     __m256           one     = _mm256_set1_ps(1.0);
 103     __m256           two     = _mm256_set1_ps(2.0);
 104     x                = xx[0];
 105     f                = ff[0];
 106
 107     nri              = nlist->nri;
 108     iinr             = nlist->iinr;
 109     jindex           = nlist->jindex;
 110     jjnr             = nlist->jjnr;
 111     shiftidx         = nlist->shift;
 112     gid              = nlist->gid;
 113     shiftvec         = fr->shift_vec[0];
 114     fshift           = fr->fshift[0];
 115     facel            = _mm256_set1_ps(fr->epsfac);
 116     charge           = mdatoms->chargeA;
 117
 118     sh_ewald         = _mm256_set1_ps(fr->ic->sh_ewald);
 119     beta             = _mm256_set1_ps(fr->ic->ewaldcoeff_q);
 120     beta2            = _mm256_mul_ps(beta,beta);
 121     beta3            = _mm256_mul_ps(beta,beta2);
 122
 123     ewtab            = fr->ic->tabq_coul_FDV0;
 124     ewtabscale       = _mm256_set1_ps(fr->ic->tabq_scale);
 125     ewtabhalfspace   = _mm256_set1_ps(0.5/fr->ic->tabq_scale);
 126
 127     /* Avoid stupid compiler warnings */
 128     jnrA = jnrB = jnrC = jnrD = jnrE = jnrF = jnrG = jnrH = 0;
 129     j_coord_offsetA = 0;
 130     j_coord_offsetB = 0;
 131     j_coord_offsetC = 0;
 132     j_coord_offsetD = 0;
 133     j_coord_offsetE = 0;
 134     j_coord_offsetF = 0;
 135     j_coord_offsetG = 0;
 136     j_coord_offsetH = 0;
 137
 138     outeriter        = 0;
 139     inneriter        = 0;
 140
 141     for(iidx=0;iidx<4*DIM;iidx++)
 142     {
 143         scratch[iidx] = 0.0;
 144     }
 145
 146     /* Start outer loop over neighborlists */
 147     for(iidx=0; iidx<nri; iidx++)
 148     {
 149         /* Load shift vector for this list */
 150         i_shift_offset   = DIM*shiftidx[iidx];
 151
 152         /* Load limits for loop over neighbors */
 153         j_index_start    = jindex[iidx];
 154         j_index_end      = jindex[iidx+1];
 155
 156         /* Get outer coordinate index */
 157         inr              = iinr[iidx];
 158         i_coord_offset   = DIM*inr;
 159
 160         /* Load i particle coords and add shift vector */
 161         gmx_mm256_load_shift_and_1rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 162
 163         fix0             = _mm256_setzero_ps();
 164         fiy0             = _mm256_setzero_ps();
 165         fiz0             = _mm256_setzero_ps();
 166
 167         /* Load parameters for i particles */
 168         iq0              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+0]));
 169
 170         /* Reset potential sums */
 171         velecsum         = _mm256_setzero_ps();
 172
 173         /* Start inner kernel loop */
 174         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+7]>=0; jidx+=8)
 175         {
 176
 177             /* Get j neighbor index, and coordinate index */
 178             jnrA             = jjnr[jidx];
 179             jnrB             = jjnr[jidx+1];
 180             jnrC             = jjnr[jidx+2];
 181             jnrD             = jjnr[jidx+3];
 182             jnrE             = jjnr[jidx+4];
 183             jnrF             = jjnr[jidx+5];
 184             jnrG             = jjnr[jidx+6];
 185             jnrH             = jjnr[jidx+7];
 186             j_coord_offsetA  = DIM*jnrA;
 187             j_coord_offsetB  = DIM*jnrB;
 188             j_coord_offsetC  = DIM*jnrC;
 189             j_coord_offsetD  = DIM*jnrD;
 190             j_coord_offsetE  = DIM*jnrE;
 191             j_coord_offsetF  = DIM*jnrF;
 192             j_coord_offsetG  = DIM*jnrG;
 193             j_coord_offsetH  = DIM*jnrH;
 194
 195             /* load j atom coordinates */
 196             gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 197                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 198                                                  x+j_coord_offsetE,x+j_coord_offsetF,
 199                                                  x+j_coord_offsetG,x+j_coord_offsetH,
 200                                                  &jx0,&jy0,&jz0);
 201
 202             /* Calculate displacement vector */
 203             dx00             = _mm256_sub_ps(ix0,jx0);
 204             dy00             = _mm256_sub_ps(iy0,jy0);
 205             dz00             = _mm256_sub_ps(iz0,jz0);
 206
 207             /* Calculate squared distance and things based on it */
 208             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
 209
 210             rinv00           = gmx_mm256_invsqrt_ps(rsq00);
 211
 212             rinvsq00         = _mm256_mul_ps(rinv00,rinv00);
 213
 214             /* Load parameters for j particles */
 215             jq0              = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 216                                                                  charge+jnrC+0,charge+jnrD+0,
 217                                                                  charge+jnrE+0,charge+jnrF+0,
 218                                                                  charge+jnrG+0,charge+jnrH+0);
 219
 220             /**************************
 221              * CALCULATE INTERACTIONS *
 222              **************************/
 223
 224             r00              = _mm256_mul_ps(rsq00,rinv00);
 225
 226             /* Compute parameters for interactions between i and j atoms */
 227             qq00             = _mm256_mul_ps(iq0,jq0);
 228
 229             /* EWALD ELECTROSTATICS */
 230
 231             /* Analytical PME correction */
 232             zeta2            = _mm256_mul_ps(beta2,rsq00);
 233             rinv3            = _mm256_mul_ps(rinvsq00,rinv00);
 234             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
 235             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
 236             felec            = _mm256_mul_ps(qq00,felec);
 237             pmecorrV         = gmx_mm256_pmecorrV_ps(zeta2);
 238             pmecorrV         = _mm256_mul_ps(pmecorrV,beta);
 239             velec            = _mm256_sub_ps(rinv00,pmecorrV);
 240             velec            = _mm256_mul_ps(qq00,velec);
 241
 242             /* Update potential sum for this i atom from the interaction with this j atom. */
 243             velecsum         = _mm256_add_ps(velecsum,velec);
 244
 245             fscal            = felec;
 246
 247             /* Calculate temporary vectorial force */
 248             tx               = _mm256_mul_ps(fscal,dx00);
 249             ty               = _mm256_mul_ps(fscal,dy00);
 250             tz               = _mm256_mul_ps(fscal,dz00);
 251
 252             /* Update vectorial force */
 253             fix0             = _mm256_add_ps(fix0,tx);
 254             fiy0             = _mm256_add_ps(fiy0,ty);
 255             fiz0             = _mm256_add_ps(fiz0,tz);
 256
 257             fjptrA             = f+j_coord_offsetA;
 258             fjptrB             = f+j_coord_offsetB;
 259             fjptrC             = f+j_coord_offsetC;
 260             fjptrD             = f+j_coord_offsetD;
 261             fjptrE             = f+j_coord_offsetE;
 262             fjptrF             = f+j_coord_offsetF;
 263             fjptrG             = f+j_coord_offsetG;
 264             fjptrH             = f+j_coord_offsetH;
 265             gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,tx,ty,tz);
 266
 267             /* Inner loop uses 84 flops */
 268         }
 269
 270         if(jidx<j_index_end)
 271         {
 272
 273             /* Get j neighbor index, and coordinate index */
 274             jnrlistA         = jjnr[jidx];
 275             jnrlistB         = jjnr[jidx+1];
 276             jnrlistC         = jjnr[jidx+2];
 277             jnrlistD         = jjnr[jidx+3];
 278             jnrlistE         = jjnr[jidx+4];
 279             jnrlistF         = jjnr[jidx+5];
 280             jnrlistG         = jjnr[jidx+6];
 281             jnrlistH         = jjnr[jidx+7];
 282             /* Sign of each element will be negative for non-real atoms.
 283              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 284              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 285              */
 286             dummy_mask = gmx_mm256_set_m128(gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx+4)),_mm_setzero_si128())),
 287                                             gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128())));
 288
 289             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 290             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 291             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 292             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 293             jnrE       = (jnrlistE>=0) ? jnrlistE : 0;
 294             jnrF       = (jnrlistF>=0) ? jnrlistF : 0;
 295             jnrG       = (jnrlistG>=0) ? jnrlistG : 0;
 296             jnrH       = (jnrlistH>=0) ? jnrlistH : 0;
 297             j_coord_offsetA  = DIM*jnrA;
 298             j_coord_offsetB  = DIM*jnrB;
 299             j_coord_offsetC  = DIM*jnrC;
 300             j_coord_offsetD  = DIM*jnrD;
 301             j_coord_offsetE  = DIM*jnrE;
 302             j_coord_offsetF  = DIM*jnrF;
 303             j_coord_offsetG  = DIM*jnrG;
 304             j_coord_offsetH  = DIM*jnrH;
 305
 306             /* load j atom coordinates */
 307             gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 308                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 309                                                  x+j_coord_offsetE,x+j_coord_offsetF,
 310                                                  x+j_coord_offsetG,x+j_coord_offsetH,
 311                                                  &jx0,&jy0,&jz0);
 312
 313             /* Calculate displacement vector */
 314             dx00             = _mm256_sub_ps(ix0,jx0);
 315             dy00             = _mm256_sub_ps(iy0,jy0);
 316             dz00             = _mm256_sub_ps(iz0,jz0);
 317
 318             /* Calculate squared distance and things based on it */
 319             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
 320
 321             rinv00           = gmx_mm256_invsqrt_ps(rsq00);
 322
 323             rinvsq00         = _mm256_mul_ps(rinv00,rinv00);
 324
 325             /* Load parameters for j particles */
 326             jq0              = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 327                                                                  charge+jnrC+0,charge+jnrD+0,
 328                                                                  charge+jnrE+0,charge+jnrF+0,
 329                                                                  charge+jnrG+0,charge+jnrH+0);
 330
 331             /**************************
 332              * CALCULATE INTERACTIONS *
 333              **************************/
 334
 335             r00              = _mm256_mul_ps(rsq00,rinv00);
 336             r00              = _mm256_andnot_ps(dummy_mask,r00);
 337
 338             /* Compute parameters for interactions between i and j atoms */
 339             qq00             = _mm256_mul_ps(iq0,jq0);
 340
 341             /* EWALD ELECTROSTATICS */
 342
 343             /* Analytical PME correction */
 344             zeta2            = _mm256_mul_ps(beta2,rsq00);
 345             rinv3            = _mm256_mul_ps(rinvsq00,rinv00);
 346             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
 347             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
 348             felec            = _mm256_mul_ps(qq00,felec);
 349             pmecorrV         = gmx_mm256_pmecorrV_ps(zeta2);
 350             pmecorrV         = _mm256_mul_ps(pmecorrV,beta);
 351             velec            = _mm256_sub_ps(rinv00,pmecorrV);
 352             velec            = _mm256_mul_ps(qq00,velec);
 353
 354             /* Update potential sum for this i atom from the interaction with this j atom. */
 355             velec            = _mm256_andnot_ps(dummy_mask,velec);
 356             velecsum         = _mm256_add_ps(velecsum,velec);
 357
 358             fscal            = felec;
 359
 360             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
 361
 362             /* Calculate temporary vectorial force */
 363             tx               = _mm256_mul_ps(fscal,dx00);
 364             ty               = _mm256_mul_ps(fscal,dy00);
 365             tz               = _mm256_mul_ps(fscal,dz00);
 366
 367             /* Update vectorial force */
 368             fix0             = _mm256_add_ps(fix0,tx);
 369             fiy0             = _mm256_add_ps(fiy0,ty);
 370             fiz0             = _mm256_add_ps(fiz0,tz);
 371
 372             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 373             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 374             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 375             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 376             fjptrE             = (jnrlistE>=0) ? f+j_coord_offsetE : scratch;
 377             fjptrF             = (jnrlistF>=0) ? f+j_coord_offsetF : scratch;
 378             fjptrG             = (jnrlistG>=0) ? f+j_coord_offsetG : scratch;
 379             fjptrH             = (jnrlistH>=0) ? f+j_coord_offsetH : scratch;
 380             gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,tx,ty,tz);
 381
 382             /* Inner loop uses 85 flops */
 383         }
 384
 385         /* End of innermost loop */
 386
 387         gmx_mm256_update_iforce_1atom_swizzle_ps(fix0,fiy0,fiz0,
 388                                                  f+i_coord_offset,fshift+i_shift_offset);
 389
 390         ggid                        = gid[iidx];
 391         /* Update potential energies */
 392         gmx_mm256_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid);
 393
 394         /* Increment number of inner iterations */
 395         inneriter                  += j_index_end - j_index_start;
 396
 397         /* Outer loop uses 8 flops */
 398     }
 399
 400     /* Increment number of outer iterations */
 401     outeriter        += nri;
 402
 403     /* Update outer/inner flops */
 404
 405     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VF,outeriter*8 + inneriter*85);
 406 }
 407 /*
 408  * Gromacs nonbonded kernel:   nb_kernel_ElecEw_VdwNone_GeomP1P1_F_avx_256_single
 409  * Electrostatics interaction: Ewald
 410  * VdW interaction:            None
 411  * Geometry:                   Particle-Particle
 412  * Calculate force/pot:        Force
 413  */
 414 void
 415 nb_kernel_ElecEw_VdwNone_GeomP1P1_F_avx_256_single
 416                     (t_nblist                    * gmx_restrict       nlist,
 417                      rvec                        * gmx_restrict          xx,
 418                      rvec                        * gmx_restrict          ff,
 419                      t_forcerec                  * gmx_restrict          fr,
 420                      t_mdatoms                   * gmx_restrict     mdatoms,
 421                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 422                      t_nrnb                      * gmx_restrict        nrnb)
 423 {
 424     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 425      * just 0 for non-waters.
 426      * Suffixes A,B,C,D,E,F,G,H refer to j loop unrolling done with AVX, e.g. for the eight different
 427      * jnr indices corresponding to data put in the four positions in the SIMD register.
 428      */
 429     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 430     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 431     int              jnrA,jnrB,jnrC,jnrD;
 432     int              jnrE,jnrF,jnrG,jnrH;
 433     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
 434     int              jnrlistE,jnrlistF,jnrlistG,jnrlistH;
 435     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 436     int              j_coord_offsetE,j_coord_offsetF,j_coord_offsetG,j_coord_offsetH;
 437     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 438     real             rcutoff_scalar;
 439     real             *shiftvec,*fshift,*x,*f;
 440     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD,*fjptrE,*fjptrF,*fjptrG,*fjptrH;
 441     real             scratch[4*DIM];
 442     __m256           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 443     real *           vdwioffsetptr0;
 444     __m256           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 445     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D,vdwjidx0E,vdwjidx0F,vdwjidx0G,vdwjidx0H;
 446     __m256           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 447     __m256           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 448     __m256           velec,felec,velecsum,facel,crf,krf,krf2;
 449     real             *charge;
 450     __m256i          ewitab;
 451     __m128i          ewitab_lo,ewitab_hi;
 452     __m256           ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
 453     __m256           beta,beta2,beta3,zeta2,pmecorrF,pmecorrV,rinv3;
 454     real             *ewtab;
 455     __m256           dummy_mask,cutoff_mask;
 456     __m256           signbit = _mm256_castsi256_ps( _mm256_set1_epi32(0x80000000) );
 457     __m256           one     = _mm256_set1_ps(1.0);
 458     __m256           two     = _mm256_set1_ps(2.0);
 459     x                = xx[0];
 460     f                = ff[0];
 461
 462     nri              = nlist->nri;
 463     iinr             = nlist->iinr;
 464     jindex           = nlist->jindex;
 465     jjnr             = nlist->jjnr;
 466     shiftidx         = nlist->shift;
 467     gid              = nlist->gid;
 468     shiftvec         = fr->shift_vec[0];
 469     fshift           = fr->fshift[0];
 470     facel            = _mm256_set1_ps(fr->epsfac);
 471     charge           = mdatoms->chargeA;
 472
 473     sh_ewald         = _mm256_set1_ps(fr->ic->sh_ewald);
 474     beta             = _mm256_set1_ps(fr->ic->ewaldcoeff_q);
 475     beta2            = _mm256_mul_ps(beta,beta);
 476     beta3            = _mm256_mul_ps(beta,beta2);
 477
 478     ewtab            = fr->ic->tabq_coul_F;
 479     ewtabscale       = _mm256_set1_ps(fr->ic->tabq_scale);
 480     ewtabhalfspace   = _mm256_set1_ps(0.5/fr->ic->tabq_scale);
 481
 482     /* Avoid stupid compiler warnings */
 483     jnrA = jnrB = jnrC = jnrD = jnrE = jnrF = jnrG = jnrH = 0;
 484     j_coord_offsetA = 0;
 485     j_coord_offsetB = 0;
 486     j_coord_offsetC = 0;
 487     j_coord_offsetD = 0;
 488     j_coord_offsetE = 0;
 489     j_coord_offsetF = 0;
 490     j_coord_offsetG = 0;
 491     j_coord_offsetH = 0;
 492
 493     outeriter        = 0;
 494     inneriter        = 0;
 495
 496     for(iidx=0;iidx<4*DIM;iidx++)
 497     {
 498         scratch[iidx] = 0.0;
 499     }
 500
 501     /* Start outer loop over neighborlists */
 502     for(iidx=0; iidx<nri; iidx++)
 503     {
 504         /* Load shift vector for this list */
 505         i_shift_offset   = DIM*shiftidx[iidx];
 506
 507         /* Load limits for loop over neighbors */
 508         j_index_start    = jindex[iidx];
 509         j_index_end      = jindex[iidx+1];
 510
 511         /* Get outer coordinate index */
 512         inr              = iinr[iidx];
 513         i_coord_offset   = DIM*inr;
 514
 515         /* Load i particle coords and add shift vector */
 516         gmx_mm256_load_shift_and_1rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 517
 518         fix0             = _mm256_setzero_ps();
 519         fiy0             = _mm256_setzero_ps();
 520         fiz0             = _mm256_setzero_ps();
 521
 522         /* Load parameters for i particles */
 523         iq0              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+0]));
 524
 525         /* Start inner kernel loop */
 526         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+7]>=0; jidx+=8)
 527         {
 528
 529             /* Get j neighbor index, and coordinate index */
 530             jnrA             = jjnr[jidx];
 531             jnrB             = jjnr[jidx+1];
 532             jnrC             = jjnr[jidx+2];
 533             jnrD             = jjnr[jidx+3];
 534             jnrE             = jjnr[jidx+4];
 535             jnrF             = jjnr[jidx+5];
 536             jnrG             = jjnr[jidx+6];
 537             jnrH             = jjnr[jidx+7];
 538             j_coord_offsetA  = DIM*jnrA;
 539             j_coord_offsetB  = DIM*jnrB;
 540             j_coord_offsetC  = DIM*jnrC;
 541             j_coord_offsetD  = DIM*jnrD;
 542             j_coord_offsetE  = DIM*jnrE;
 543             j_coord_offsetF  = DIM*jnrF;
 544             j_coord_offsetG  = DIM*jnrG;
 545             j_coord_offsetH  = DIM*jnrH;
 546
 547             /* load j atom coordinates */
 548             gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 549                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 550                                                  x+j_coord_offsetE,x+j_coord_offsetF,
 551                                                  x+j_coord_offsetG,x+j_coord_offsetH,
 552                                                  &jx0,&jy0,&jz0);
 553
 554             /* Calculate displacement vector */
 555             dx00             = _mm256_sub_ps(ix0,jx0);
 556             dy00             = _mm256_sub_ps(iy0,jy0);
 557             dz00             = _mm256_sub_ps(iz0,jz0);
 558
 559             /* Calculate squared distance and things based on it */
 560             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
 561
 562             rinv00           = gmx_mm256_invsqrt_ps(rsq00);
 563
 564             rinvsq00         = _mm256_mul_ps(rinv00,rinv00);
 565
 566             /* Load parameters for j particles */
 567             jq0              = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 568                                                                  charge+jnrC+0,charge+jnrD+0,
 569                                                                  charge+jnrE+0,charge+jnrF+0,
 570                                                                  charge+jnrG+0,charge+jnrH+0);
 571
 572             /**************************
 573              * CALCULATE INTERACTIONS *
 574              **************************/
 575
 576             r00              = _mm256_mul_ps(rsq00,rinv00);
 577
 578             /* Compute parameters for interactions between i and j atoms */
 579             qq00             = _mm256_mul_ps(iq0,jq0);
 580
 581             /* EWALD ELECTROSTATICS */
 582
 583             /* Analytical PME correction */
 584             zeta2            = _mm256_mul_ps(beta2,rsq00);
 585             rinv3            = _mm256_mul_ps(rinvsq00,rinv00);
 586             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
 587             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
 588             felec            = _mm256_mul_ps(qq00,felec);
 589
 590             fscal            = felec;
 591
 592             /* Calculate temporary vectorial force */
 593             tx               = _mm256_mul_ps(fscal,dx00);
 594             ty               = _mm256_mul_ps(fscal,dy00);
 595             tz               = _mm256_mul_ps(fscal,dz00);
 596
 597             /* Update vectorial force */
 598             fix0             = _mm256_add_ps(fix0,tx);
 599             fiy0             = _mm256_add_ps(fiy0,ty);
 600             fiz0             = _mm256_add_ps(fiz0,tz);
 601
 602             fjptrA             = f+j_coord_offsetA;
 603             fjptrB             = f+j_coord_offsetB;
 604             fjptrC             = f+j_coord_offsetC;
 605             fjptrD             = f+j_coord_offsetD;
 606             fjptrE             = f+j_coord_offsetE;
 607             fjptrF             = f+j_coord_offsetF;
 608             fjptrG             = f+j_coord_offsetG;
 609             fjptrH             = f+j_coord_offsetH;
 610             gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,tx,ty,tz);
 611
 612             /* Inner loop uses 56 flops */
 613         }
 614
 615         if(jidx<j_index_end)
 616         {
 617
 618             /* Get j neighbor index, and coordinate index */
 619             jnrlistA         = jjnr[jidx];
 620             jnrlistB         = jjnr[jidx+1];
 621             jnrlistC         = jjnr[jidx+2];
 622             jnrlistD         = jjnr[jidx+3];
 623             jnrlistE         = jjnr[jidx+4];
 624             jnrlistF         = jjnr[jidx+5];
 625             jnrlistG         = jjnr[jidx+6];
 626             jnrlistH         = jjnr[jidx+7];
 627             /* Sign of each element will be negative for non-real atoms.
 628              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 629              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 630              */
 631             dummy_mask = gmx_mm256_set_m128(gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx+4)),_mm_setzero_si128())),
 632                                             gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128())));
 633
 634             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 635             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 636             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 637             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 638             jnrE       = (jnrlistE>=0) ? jnrlistE : 0;
 639             jnrF       = (jnrlistF>=0) ? jnrlistF : 0;
 640             jnrG       = (jnrlistG>=0) ? jnrlistG : 0;
 641             jnrH       = (jnrlistH>=0) ? jnrlistH : 0;
 642             j_coord_offsetA  = DIM*jnrA;
 643             j_coord_offsetB  = DIM*jnrB;
 644             j_coord_offsetC  = DIM*jnrC;
 645             j_coord_offsetD  = DIM*jnrD;
 646             j_coord_offsetE  = DIM*jnrE;
 647             j_coord_offsetF  = DIM*jnrF;
 648             j_coord_offsetG  = DIM*jnrG;
 649             j_coord_offsetH  = DIM*jnrH;
 650
 651             /* load j atom coordinates */
 652             gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 653                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 654                                                  x+j_coord_offsetE,x+j_coord_offsetF,
 655                                                  x+j_coord_offsetG,x+j_coord_offsetH,
 656                                                  &jx0,&jy0,&jz0);
 657
 658             /* Calculate displacement vector */
 659             dx00             = _mm256_sub_ps(ix0,jx0);
 660             dy00             = _mm256_sub_ps(iy0,jy0);
 661             dz00             = _mm256_sub_ps(iz0,jz0);
 662
 663             /* Calculate squared distance and things based on it */
 664             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
 665
 666             rinv00           = gmx_mm256_invsqrt_ps(rsq00);
 667
 668             rinvsq00         = _mm256_mul_ps(rinv00,rinv00);
 669
 670             /* Load parameters for j particles */
 671             jq0              = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 672                                                                  charge+jnrC+0,charge+jnrD+0,
 673                                                                  charge+jnrE+0,charge+jnrF+0,
 674                                                                  charge+jnrG+0,charge+jnrH+0);
 675
 676             /**************************
 677              * CALCULATE INTERACTIONS *
 678              **************************/
 679
 680             r00              = _mm256_mul_ps(rsq00,rinv00);
 681             r00              = _mm256_andnot_ps(dummy_mask,r00);
 682
 683             /* Compute parameters for interactions between i and j atoms */
 684             qq00             = _mm256_mul_ps(iq0,jq0);
 685
 686             /* EWALD ELECTROSTATICS */
 687
 688             /* Analytical PME correction */
 689             zeta2            = _mm256_mul_ps(beta2,rsq00);
 690             rinv3            = _mm256_mul_ps(rinvsq00,rinv00);
 691             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
 692             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
 693             felec            = _mm256_mul_ps(qq00,felec);
 694
 695             fscal            = felec;
 696
 697             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
 698
 699             /* Calculate temporary vectorial force */
 700             tx               = _mm256_mul_ps(fscal,dx00);
 701             ty               = _mm256_mul_ps(fscal,dy00);
 702             tz               = _mm256_mul_ps(fscal,dz00);
 703
 704             /* Update vectorial force */
 705             fix0             = _mm256_add_ps(fix0,tx);
 706             fiy0             = _mm256_add_ps(fiy0,ty);
 707             fiz0             = _mm256_add_ps(fiz0,tz);
 708
 709             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 710             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 711             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 712             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 713             fjptrE             = (jnrlistE>=0) ? f+j_coord_offsetE : scratch;
 714             fjptrF             = (jnrlistF>=0) ? f+j_coord_offsetF : scratch;
 715             fjptrG             = (jnrlistG>=0) ? f+j_coord_offsetG : scratch;
 716             fjptrH             = (jnrlistH>=0) ? f+j_coord_offsetH : scratch;
 717             gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,tx,ty,tz);
 718
 719             /* Inner loop uses 57 flops */
 720         }
 721
 722         /* End of innermost loop */
 723
 724         gmx_mm256_update_iforce_1atom_swizzle_ps(fix0,fiy0,fiz0,
 725                                                  f+i_coord_offset,fshift+i_shift_offset);
 726
 727         /* Increment number of inner iterations */
 728         inneriter                  += j_index_end - j_index_start;
 729
 730         /* Outer loop uses 7 flops */
 731     }
 732
 733     /* Increment number of outer iterations */
 734     outeriter        += nri;
 735
 736     /* Update outer/inner flops */
 737
 738     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_F,outeriter*7 + inneriter*57);
 739 }