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 "config.h"
  39
  40 #include <math.h>
  41
  42 #include "../nb_kernel.h"
  43 #include "gromacs/legacyheaders/types/simple.h"
  44 #include "gromacs/math/vec.h"
  45 #include "gromacs/legacyheaders/nrnb.h"
  46
  47 #include "gromacs/simd/math_x86_avx_256_single.h"
  48 #include "kernelutil_x86_avx_256_single.h"
  49
  50 /*
  51  * Gromacs nonbonded kernel:   nb_kernel_ElecEw_VdwNone_GeomP1P1_VF_avx_256_single
  52  * Electrostatics interaction: Ewald
  53  * VdW interaction:            None
  54  * Geometry:                   Particle-Particle
  55  * Calculate force/pot:        PotentialAndForce
  56  */
  57 void
  58 nb_kernel_ElecEw_VdwNone_GeomP1P1_VF_avx_256_single
  59                     (t_nblist                    * gmx_restrict       nlist,
  60                      rvec                        * gmx_restrict          xx,
  61                      rvec                        * gmx_restrict          ff,
  62                      t_forcerec                  * gmx_restrict          fr,
  63                      t_mdatoms                   * gmx_restrict     mdatoms,
  64                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
  65                      t_nrnb                      * gmx_restrict        nrnb)
  66 {
  67     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
  68      * just 0 for non-waters.
  69      * Suffixes A,B,C,D,E,F,G,H refer to j loop unrolling done with AVX, e.g. for the eight different
  70      * jnr indices corresponding to data put in the four positions in the SIMD register.
  71      */
  72     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
  73     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
  74     int              jnrA,jnrB,jnrC,jnrD;
  75     int              jnrE,jnrF,jnrG,jnrH;
  76     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
  77     int              jnrlistE,jnrlistF,jnrlistG,jnrlistH;
  78     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
  79     int              j_coord_offsetE,j_coord_offsetF,j_coord_offsetG,j_coord_offsetH;
  80     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
  81     real             rcutoff_scalar;
  82     real             *shiftvec,*fshift,*x,*f;
  83     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD,*fjptrE,*fjptrF,*fjptrG,*fjptrH;
  84     real             scratch[4*DIM];
  85     __m256           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
  86     real *           vdwioffsetptr0;
  87     __m256           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
  88     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D,vdwjidx0E,vdwjidx0F,vdwjidx0G,vdwjidx0H;
  89     __m256           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  90     __m256           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  91     __m256           velec,felec,velecsum,facel,crf,krf,krf2;
  92     real             *charge;
  93     __m256i          ewitab;
  94     __m128i          ewitab_lo,ewitab_hi;
  95     __m256           ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
  96     __m256           beta,beta2,beta3,zeta2,pmecorrF,pmecorrV,rinv3;
  97     real             *ewtab;
  98     __m256           dummy_mask,cutoff_mask;
  99     __m256           signbit = _mm256_castsi256_ps( _mm256_set1_epi32(0x80000000) );
 100     __m256           one     = _mm256_set1_ps(1.0);
 101     __m256           two     = _mm256_set1_ps(2.0);
 102     x                = xx[0];
 103     f                = ff[0];
 104
 105     nri              = nlist->nri;
 106     iinr             = nlist->iinr;
 107     jindex           = nlist->jindex;
 108     jjnr             = nlist->jjnr;
 109     shiftidx         = nlist->shift;
 110     gid              = nlist->gid;
 111     shiftvec         = fr->shift_vec[0];
 112     fshift           = fr->fshift[0];
 113     facel            = _mm256_set1_ps(fr->epsfac);
 114     charge           = mdatoms->chargeA;
 115
 116     sh_ewald         = _mm256_set1_ps(fr->ic->sh_ewald);
 117     beta             = _mm256_set1_ps(fr->ic->ewaldcoeff_q);
 118     beta2            = _mm256_mul_ps(beta,beta);
 119     beta3            = _mm256_mul_ps(beta,beta2);
 120
 121     ewtab            = fr->ic->tabq_coul_FDV0;
 122     ewtabscale       = _mm256_set1_ps(fr->ic->tabq_scale);
 123     ewtabhalfspace   = _mm256_set1_ps(0.5/fr->ic->tabq_scale);
 124
 125     /* Avoid stupid compiler warnings */
 126     jnrA = jnrB = jnrC = jnrD = jnrE = jnrF = jnrG = jnrH = 0;
 127     j_coord_offsetA = 0;
 128     j_coord_offsetB = 0;
 129     j_coord_offsetC = 0;
 130     j_coord_offsetD = 0;
 131     j_coord_offsetE = 0;
 132     j_coord_offsetF = 0;
 133     j_coord_offsetG = 0;
 134     j_coord_offsetH = 0;
 135
 136     outeriter        = 0;
 137     inneriter        = 0;
 138
 139     for(iidx=0;iidx<4*DIM;iidx++)
 140     {
 141         scratch[iidx] = 0.0;
 142     }
 143
 144     /* Start outer loop over neighborlists */
 145     for(iidx=0; iidx<nri; iidx++)
 146     {
 147         /* Load shift vector for this list */
 148         i_shift_offset   = DIM*shiftidx[iidx];
 149
 150         /* Load limits for loop over neighbors */
 151         j_index_start    = jindex[iidx];
 152         j_index_end      = jindex[iidx+1];
 153
 154         /* Get outer coordinate index */
 155         inr              = iinr[iidx];
 156         i_coord_offset   = DIM*inr;
 157
 158         /* Load i particle coords and add shift vector */
 159         gmx_mm256_load_shift_and_1rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 160
 161         fix0             = _mm256_setzero_ps();
 162         fiy0             = _mm256_setzero_ps();
 163         fiz0             = _mm256_setzero_ps();
 164
 165         /* Load parameters for i particles */
 166         iq0              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+0]));
 167
 168         /* Reset potential sums */
 169         velecsum         = _mm256_setzero_ps();
 170
 171         /* Start inner kernel loop */
 172         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+7]>=0; jidx+=8)
 173         {
 174
 175             /* Get j neighbor index, and coordinate index */
 176             jnrA             = jjnr[jidx];
 177             jnrB             = jjnr[jidx+1];
 178             jnrC             = jjnr[jidx+2];
 179             jnrD             = jjnr[jidx+3];
 180             jnrE             = jjnr[jidx+4];
 181             jnrF             = jjnr[jidx+5];
 182             jnrG             = jjnr[jidx+6];
 183             jnrH             = jjnr[jidx+7];
 184             j_coord_offsetA  = DIM*jnrA;
 185             j_coord_offsetB  = DIM*jnrB;
 186             j_coord_offsetC  = DIM*jnrC;
 187             j_coord_offsetD  = DIM*jnrD;
 188             j_coord_offsetE  = DIM*jnrE;
 189             j_coord_offsetF  = DIM*jnrF;
 190             j_coord_offsetG  = DIM*jnrG;
 191             j_coord_offsetH  = DIM*jnrH;
 192
 193             /* load j atom coordinates */
 194             gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 195                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 196                                                  x+j_coord_offsetE,x+j_coord_offsetF,
 197                                                  x+j_coord_offsetG,x+j_coord_offsetH,
 198                                                  &jx0,&jy0,&jz0);
 199
 200             /* Calculate displacement vector */
 201             dx00             = _mm256_sub_ps(ix0,jx0);
 202             dy00             = _mm256_sub_ps(iy0,jy0);
 203             dz00             = _mm256_sub_ps(iz0,jz0);
 204
 205             /* Calculate squared distance and things based on it */
 206             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
 207
 208             rinv00           = gmx_mm256_invsqrt_ps(rsq00);
 209
 210             rinvsq00         = _mm256_mul_ps(rinv00,rinv00);
 211
 212             /* Load parameters for j particles */
 213             jq0              = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 214                                                                  charge+jnrC+0,charge+jnrD+0,
 215                                                                  charge+jnrE+0,charge+jnrF+0,
 216                                                                  charge+jnrG+0,charge+jnrH+0);
 217
 218             /**************************
 219              * CALCULATE INTERACTIONS *
 220              **************************/
 221
 222             r00              = _mm256_mul_ps(rsq00,rinv00);
 223
 224             /* Compute parameters for interactions between i and j atoms */
 225             qq00             = _mm256_mul_ps(iq0,jq0);
 226
 227             /* EWALD ELECTROSTATICS */
 228
 229             /* Analytical PME correction */
 230             zeta2            = _mm256_mul_ps(beta2,rsq00);
 231             rinv3            = _mm256_mul_ps(rinvsq00,rinv00);
 232             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
 233             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
 234             felec            = _mm256_mul_ps(qq00,felec);
 235             pmecorrV         = gmx_mm256_pmecorrV_ps(zeta2);
 236             pmecorrV         = _mm256_mul_ps(pmecorrV,beta);
 237             velec            = _mm256_sub_ps(rinv00,pmecorrV);
 238             velec            = _mm256_mul_ps(qq00,velec);
 239
 240             /* Update potential sum for this i atom from the interaction with this j atom. */
 241             velecsum         = _mm256_add_ps(velecsum,velec);
 242
 243             fscal            = felec;
 244
 245             /* Calculate temporary vectorial force */
 246             tx               = _mm256_mul_ps(fscal,dx00);
 247             ty               = _mm256_mul_ps(fscal,dy00);
 248             tz               = _mm256_mul_ps(fscal,dz00);
 249
 250             /* Update vectorial force */
 251             fix0             = _mm256_add_ps(fix0,tx);
 252             fiy0             = _mm256_add_ps(fiy0,ty);
 253             fiz0             = _mm256_add_ps(fiz0,tz);
 254
 255             fjptrA             = f+j_coord_offsetA;
 256             fjptrB             = f+j_coord_offsetB;
 257             fjptrC             = f+j_coord_offsetC;
 258             fjptrD             = f+j_coord_offsetD;
 259             fjptrE             = f+j_coord_offsetE;
 260             fjptrF             = f+j_coord_offsetF;
 261             fjptrG             = f+j_coord_offsetG;
 262             fjptrH             = f+j_coord_offsetH;
 263             gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,tx,ty,tz);
 264
 265             /* Inner loop uses 84 flops */
 266         }
 267
 268         if(jidx<j_index_end)
 269         {
 270
 271             /* Get j neighbor index, and coordinate index */
 272             jnrlistA         = jjnr[jidx];
 273             jnrlistB         = jjnr[jidx+1];
 274             jnrlistC         = jjnr[jidx+2];
 275             jnrlistD         = jjnr[jidx+3];
 276             jnrlistE         = jjnr[jidx+4];
 277             jnrlistF         = jjnr[jidx+5];
 278             jnrlistG         = jjnr[jidx+6];
 279             jnrlistH         = jjnr[jidx+7];
 280             /* Sign of each element will be negative for non-real atoms.
 281              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 282              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 283              */
 284             dummy_mask = gmx_mm256_set_m128(gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx+4)),_mm_setzero_si128())),
 285                                             gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128())));
 286
 287             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 288             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 289             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 290             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 291             jnrE       = (jnrlistE>=0) ? jnrlistE : 0;
 292             jnrF       = (jnrlistF>=0) ? jnrlistF : 0;
 293             jnrG       = (jnrlistG>=0) ? jnrlistG : 0;
 294             jnrH       = (jnrlistH>=0) ? jnrlistH : 0;
 295             j_coord_offsetA  = DIM*jnrA;
 296             j_coord_offsetB  = DIM*jnrB;
 297             j_coord_offsetC  = DIM*jnrC;
 298             j_coord_offsetD  = DIM*jnrD;
 299             j_coord_offsetE  = DIM*jnrE;
 300             j_coord_offsetF  = DIM*jnrF;
 301             j_coord_offsetG  = DIM*jnrG;
 302             j_coord_offsetH  = DIM*jnrH;
 303
 304             /* load j atom coordinates */
 305             gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 306                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 307                                                  x+j_coord_offsetE,x+j_coord_offsetF,
 308                                                  x+j_coord_offsetG,x+j_coord_offsetH,
 309                                                  &jx0,&jy0,&jz0);
 310
 311             /* Calculate displacement vector */
 312             dx00             = _mm256_sub_ps(ix0,jx0);
 313             dy00             = _mm256_sub_ps(iy0,jy0);
 314             dz00             = _mm256_sub_ps(iz0,jz0);
 315
 316             /* Calculate squared distance and things based on it */
 317             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
 318
 319             rinv00           = gmx_mm256_invsqrt_ps(rsq00);
 320
 321             rinvsq00         = _mm256_mul_ps(rinv00,rinv00);
 322
 323             /* Load parameters for j particles */
 324             jq0              = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 325                                                                  charge+jnrC+0,charge+jnrD+0,
 326                                                                  charge+jnrE+0,charge+jnrF+0,
 327                                                                  charge+jnrG+0,charge+jnrH+0);
 328
 329             /**************************
 330              * CALCULATE INTERACTIONS *
 331              **************************/
 332
 333             r00              = _mm256_mul_ps(rsq00,rinv00);
 334             r00              = _mm256_andnot_ps(dummy_mask,r00);
 335
 336             /* Compute parameters for interactions between i and j atoms */
 337             qq00             = _mm256_mul_ps(iq0,jq0);
 338
 339             /* EWALD ELECTROSTATICS */
 340
 341             /* Analytical PME correction */
 342             zeta2            = _mm256_mul_ps(beta2,rsq00);
 343             rinv3            = _mm256_mul_ps(rinvsq00,rinv00);
 344             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
 345             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
 346             felec            = _mm256_mul_ps(qq00,felec);
 347             pmecorrV         = gmx_mm256_pmecorrV_ps(zeta2);
 348             pmecorrV         = _mm256_mul_ps(pmecorrV,beta);
 349             velec            = _mm256_sub_ps(rinv00,pmecorrV);
 350             velec            = _mm256_mul_ps(qq00,velec);
 351
 352             /* Update potential sum for this i atom from the interaction with this j atom. */
 353             velec            = _mm256_andnot_ps(dummy_mask,velec);
 354             velecsum         = _mm256_add_ps(velecsum,velec);
 355
 356             fscal            = felec;
 357
 358             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
 359
 360             /* Calculate temporary vectorial force */
 361             tx               = _mm256_mul_ps(fscal,dx00);
 362             ty               = _mm256_mul_ps(fscal,dy00);
 363             tz               = _mm256_mul_ps(fscal,dz00);
 364
 365             /* Update vectorial force */
 366             fix0             = _mm256_add_ps(fix0,tx);
 367             fiy0             = _mm256_add_ps(fiy0,ty);
 368             fiz0             = _mm256_add_ps(fiz0,tz);
 369
 370             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 371             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 372             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 373             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 374             fjptrE             = (jnrlistE>=0) ? f+j_coord_offsetE : scratch;
 375             fjptrF             = (jnrlistF>=0) ? f+j_coord_offsetF : scratch;
 376             fjptrG             = (jnrlistG>=0) ? f+j_coord_offsetG : scratch;
 377             fjptrH             = (jnrlistH>=0) ? f+j_coord_offsetH : scratch;
 378             gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,tx,ty,tz);
 379
 380             /* Inner loop uses 85 flops */
 381         }
 382
 383         /* End of innermost loop */
 384
 385         gmx_mm256_update_iforce_1atom_swizzle_ps(fix0,fiy0,fiz0,
 386                                                  f+i_coord_offset,fshift+i_shift_offset);
 387
 388         ggid                        = gid[iidx];
 389         /* Update potential energies */
 390         gmx_mm256_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid);
 391
 392         /* Increment number of inner iterations */
 393         inneriter                  += j_index_end - j_index_start;
 394
 395         /* Outer loop uses 8 flops */
 396     }
 397
 398     /* Increment number of outer iterations */
 399     outeriter        += nri;
 400
 401     /* Update outer/inner flops */
 402
 403     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VF,outeriter*8 + inneriter*85);
 404 }
 405 /*
 406  * Gromacs nonbonded kernel:   nb_kernel_ElecEw_VdwNone_GeomP1P1_F_avx_256_single
 407  * Electrostatics interaction: Ewald
 408  * VdW interaction:            None
 409  * Geometry:                   Particle-Particle
 410  * Calculate force/pot:        Force
 411  */
 412 void
 413 nb_kernel_ElecEw_VdwNone_GeomP1P1_F_avx_256_single
 414                     (t_nblist                    * gmx_restrict       nlist,
 415                      rvec                        * gmx_restrict          xx,
 416                      rvec                        * gmx_restrict          ff,
 417                      t_forcerec                  * gmx_restrict          fr,
 418                      t_mdatoms                   * gmx_restrict     mdatoms,
 419                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 420                      t_nrnb                      * gmx_restrict        nrnb)
 421 {
 422     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 423      * just 0 for non-waters.
 424      * Suffixes A,B,C,D,E,F,G,H refer to j loop unrolling done with AVX, e.g. for the eight different
 425      * jnr indices corresponding to data put in the four positions in the SIMD register.
 426      */
 427     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 428     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 429     int              jnrA,jnrB,jnrC,jnrD;
 430     int              jnrE,jnrF,jnrG,jnrH;
 431     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
 432     int              jnrlistE,jnrlistF,jnrlistG,jnrlistH;
 433     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 434     int              j_coord_offsetE,j_coord_offsetF,j_coord_offsetG,j_coord_offsetH;
 435     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 436     real             rcutoff_scalar;
 437     real             *shiftvec,*fshift,*x,*f;
 438     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD,*fjptrE,*fjptrF,*fjptrG,*fjptrH;
 439     real             scratch[4*DIM];
 440     __m256           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 441     real *           vdwioffsetptr0;
 442     __m256           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 443     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D,vdwjidx0E,vdwjidx0F,vdwjidx0G,vdwjidx0H;
 444     __m256           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 445     __m256           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 446     __m256           velec,felec,velecsum,facel,crf,krf,krf2;
 447     real             *charge;
 448     __m256i          ewitab;
 449     __m128i          ewitab_lo,ewitab_hi;
 450     __m256           ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
 451     __m256           beta,beta2,beta3,zeta2,pmecorrF,pmecorrV,rinv3;
 452     real             *ewtab;
 453     __m256           dummy_mask,cutoff_mask;
 454     __m256           signbit = _mm256_castsi256_ps( _mm256_set1_epi32(0x80000000) );
 455     __m256           one     = _mm256_set1_ps(1.0);
 456     __m256           two     = _mm256_set1_ps(2.0);
 457     x                = xx[0];
 458     f                = ff[0];
 459
 460     nri              = nlist->nri;
 461     iinr             = nlist->iinr;
 462     jindex           = nlist->jindex;
 463     jjnr             = nlist->jjnr;
 464     shiftidx         = nlist->shift;
 465     gid              = nlist->gid;
 466     shiftvec         = fr->shift_vec[0];
 467     fshift           = fr->fshift[0];
 468     facel            = _mm256_set1_ps(fr->epsfac);
 469     charge           = mdatoms->chargeA;
 470
 471     sh_ewald         = _mm256_set1_ps(fr->ic->sh_ewald);
 472     beta             = _mm256_set1_ps(fr->ic->ewaldcoeff_q);
 473     beta2            = _mm256_mul_ps(beta,beta);
 474     beta3            = _mm256_mul_ps(beta,beta2);
 475
 476     ewtab            = fr->ic->tabq_coul_F;
 477     ewtabscale       = _mm256_set1_ps(fr->ic->tabq_scale);
 478     ewtabhalfspace   = _mm256_set1_ps(0.5/fr->ic->tabq_scale);
 479
 480     /* Avoid stupid compiler warnings */
 481     jnrA = jnrB = jnrC = jnrD = jnrE = jnrF = jnrG = jnrH = 0;
 482     j_coord_offsetA = 0;
 483     j_coord_offsetB = 0;
 484     j_coord_offsetC = 0;
 485     j_coord_offsetD = 0;
 486     j_coord_offsetE = 0;
 487     j_coord_offsetF = 0;
 488     j_coord_offsetG = 0;
 489     j_coord_offsetH = 0;
 490
 491     outeriter        = 0;
 492     inneriter        = 0;
 493
 494     for(iidx=0;iidx<4*DIM;iidx++)
 495     {
 496         scratch[iidx] = 0.0;
 497     }
 498
 499     /* Start outer loop over neighborlists */
 500     for(iidx=0; iidx<nri; iidx++)
 501     {
 502         /* Load shift vector for this list */
 503         i_shift_offset   = DIM*shiftidx[iidx];
 504
 505         /* Load limits for loop over neighbors */
 506         j_index_start    = jindex[iidx];
 507         j_index_end      = jindex[iidx+1];
 508
 509         /* Get outer coordinate index */
 510         inr              = iinr[iidx];
 511         i_coord_offset   = DIM*inr;
 512
 513         /* Load i particle coords and add shift vector */
 514         gmx_mm256_load_shift_and_1rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 515
 516         fix0             = _mm256_setzero_ps();
 517         fiy0             = _mm256_setzero_ps();
 518         fiz0             = _mm256_setzero_ps();
 519
 520         /* Load parameters for i particles */
 521         iq0              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+0]));
 522
 523         /* Start inner kernel loop */
 524         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+7]>=0; jidx+=8)
 525         {
 526
 527             /* Get j neighbor index, and coordinate index */
 528             jnrA             = jjnr[jidx];
 529             jnrB             = jjnr[jidx+1];
 530             jnrC             = jjnr[jidx+2];
 531             jnrD             = jjnr[jidx+3];
 532             jnrE             = jjnr[jidx+4];
 533             jnrF             = jjnr[jidx+5];
 534             jnrG             = jjnr[jidx+6];
 535             jnrH             = jjnr[jidx+7];
 536             j_coord_offsetA  = DIM*jnrA;
 537             j_coord_offsetB  = DIM*jnrB;
 538             j_coord_offsetC  = DIM*jnrC;
 539             j_coord_offsetD  = DIM*jnrD;
 540             j_coord_offsetE  = DIM*jnrE;
 541             j_coord_offsetF  = DIM*jnrF;
 542             j_coord_offsetG  = DIM*jnrG;
 543             j_coord_offsetH  = DIM*jnrH;
 544
 545             /* load j atom coordinates */
 546             gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 547                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 548                                                  x+j_coord_offsetE,x+j_coord_offsetF,
 549                                                  x+j_coord_offsetG,x+j_coord_offsetH,
 550                                                  &jx0,&jy0,&jz0);
 551
 552             /* Calculate displacement vector */
 553             dx00             = _mm256_sub_ps(ix0,jx0);
 554             dy00             = _mm256_sub_ps(iy0,jy0);
 555             dz00             = _mm256_sub_ps(iz0,jz0);
 556
 557             /* Calculate squared distance and things based on it */
 558             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
 559
 560             rinv00           = gmx_mm256_invsqrt_ps(rsq00);
 561
 562             rinvsq00         = _mm256_mul_ps(rinv00,rinv00);
 563
 564             /* Load parameters for j particles */
 565             jq0              = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 566                                                                  charge+jnrC+0,charge+jnrD+0,
 567                                                                  charge+jnrE+0,charge+jnrF+0,
 568                                                                  charge+jnrG+0,charge+jnrH+0);
 569
 570             /**************************
 571              * CALCULATE INTERACTIONS *
 572              **************************/
 573
 574             r00              = _mm256_mul_ps(rsq00,rinv00);
 575
 576             /* Compute parameters for interactions between i and j atoms */
 577             qq00             = _mm256_mul_ps(iq0,jq0);
 578
 579             /* EWALD ELECTROSTATICS */
 580
 581             /* Analytical PME correction */
 582             zeta2            = _mm256_mul_ps(beta2,rsq00);
 583             rinv3            = _mm256_mul_ps(rinvsq00,rinv00);
 584             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
 585             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
 586             felec            = _mm256_mul_ps(qq00,felec);
 587
 588             fscal            = felec;
 589
 590             /* Calculate temporary vectorial force */
 591             tx               = _mm256_mul_ps(fscal,dx00);
 592             ty               = _mm256_mul_ps(fscal,dy00);
 593             tz               = _mm256_mul_ps(fscal,dz00);
 594
 595             /* Update vectorial force */
 596             fix0             = _mm256_add_ps(fix0,tx);
 597             fiy0             = _mm256_add_ps(fiy0,ty);
 598             fiz0             = _mm256_add_ps(fiz0,tz);
 599
 600             fjptrA             = f+j_coord_offsetA;
 601             fjptrB             = f+j_coord_offsetB;
 602             fjptrC             = f+j_coord_offsetC;
 603             fjptrD             = f+j_coord_offsetD;
 604             fjptrE             = f+j_coord_offsetE;
 605             fjptrF             = f+j_coord_offsetF;
 606             fjptrG             = f+j_coord_offsetG;
 607             fjptrH             = f+j_coord_offsetH;
 608             gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,tx,ty,tz);
 609
 610             /* Inner loop uses 56 flops */
 611         }
 612
 613         if(jidx<j_index_end)
 614         {
 615
 616             /* Get j neighbor index, and coordinate index */
 617             jnrlistA         = jjnr[jidx];
 618             jnrlistB         = jjnr[jidx+1];
 619             jnrlistC         = jjnr[jidx+2];
 620             jnrlistD         = jjnr[jidx+3];
 621             jnrlistE         = jjnr[jidx+4];
 622             jnrlistF         = jjnr[jidx+5];
 623             jnrlistG         = jjnr[jidx+6];
 624             jnrlistH         = jjnr[jidx+7];
 625             /* Sign of each element will be negative for non-real atoms.
 626              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 627              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 628              */
 629             dummy_mask = gmx_mm256_set_m128(gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx+4)),_mm_setzero_si128())),
 630                                             gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128())));
 631
 632             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 633             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 634             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 635             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 636             jnrE       = (jnrlistE>=0) ? jnrlistE : 0;
 637             jnrF       = (jnrlistF>=0) ? jnrlistF : 0;
 638             jnrG       = (jnrlistG>=0) ? jnrlistG : 0;
 639             jnrH       = (jnrlistH>=0) ? jnrlistH : 0;
 640             j_coord_offsetA  = DIM*jnrA;
 641             j_coord_offsetB  = DIM*jnrB;
 642             j_coord_offsetC  = DIM*jnrC;
 643             j_coord_offsetD  = DIM*jnrD;
 644             j_coord_offsetE  = DIM*jnrE;
 645             j_coord_offsetF  = DIM*jnrF;
 646             j_coord_offsetG  = DIM*jnrG;
 647             j_coord_offsetH  = DIM*jnrH;
 648
 649             /* load j atom coordinates */
 650             gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 651                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 652                                                  x+j_coord_offsetE,x+j_coord_offsetF,
 653                                                  x+j_coord_offsetG,x+j_coord_offsetH,
 654                                                  &jx0,&jy0,&jz0);
 655
 656             /* Calculate displacement vector */
 657             dx00             = _mm256_sub_ps(ix0,jx0);
 658             dy00             = _mm256_sub_ps(iy0,jy0);
 659             dz00             = _mm256_sub_ps(iz0,jz0);
 660
 661             /* Calculate squared distance and things based on it */
 662             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
 663
 664             rinv00           = gmx_mm256_invsqrt_ps(rsq00);
 665
 666             rinvsq00         = _mm256_mul_ps(rinv00,rinv00);
 667
 668             /* Load parameters for j particles */
 669             jq0              = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 670                                                                  charge+jnrC+0,charge+jnrD+0,
 671                                                                  charge+jnrE+0,charge+jnrF+0,
 672                                                                  charge+jnrG+0,charge+jnrH+0);
 673
 674             /**************************
 675              * CALCULATE INTERACTIONS *
 676              **************************/
 677
 678             r00              = _mm256_mul_ps(rsq00,rinv00);
 679             r00              = _mm256_andnot_ps(dummy_mask,r00);
 680
 681             /* Compute parameters for interactions between i and j atoms */
 682             qq00             = _mm256_mul_ps(iq0,jq0);
 683
 684             /* EWALD ELECTROSTATICS */
 685
 686             /* Analytical PME correction */
 687             zeta2            = _mm256_mul_ps(beta2,rsq00);
 688             rinv3            = _mm256_mul_ps(rinvsq00,rinv00);
 689             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
 690             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
 691             felec            = _mm256_mul_ps(qq00,felec);
 692
 693             fscal            = felec;
 694
 695             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
 696
 697             /* Calculate temporary vectorial force */
 698             tx               = _mm256_mul_ps(fscal,dx00);
 699             ty               = _mm256_mul_ps(fscal,dy00);
 700             tz               = _mm256_mul_ps(fscal,dz00);
 701
 702             /* Update vectorial force */
 703             fix0             = _mm256_add_ps(fix0,tx);
 704             fiy0             = _mm256_add_ps(fiy0,ty);
 705             fiz0             = _mm256_add_ps(fiz0,tz);
 706
 707             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 708             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 709             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 710             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 711             fjptrE             = (jnrlistE>=0) ? f+j_coord_offsetE : scratch;
 712             fjptrF             = (jnrlistF>=0) ? f+j_coord_offsetF : scratch;
 713             fjptrG             = (jnrlistG>=0) ? f+j_coord_offsetG : scratch;
 714             fjptrH             = (jnrlistH>=0) ? f+j_coord_offsetH : scratch;
 715             gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,tx,ty,tz);
 716
 717             /* Inner loop uses 57 flops */
 718         }
 719
 720         /* End of innermost loop */
 721
 722         gmx_mm256_update_iforce_1atom_swizzle_ps(fix0,fiy0,fiz0,
 723                                                  f+i_coord_offset,fshift+i_shift_offset);
 724
 725         /* Increment number of inner iterations */
 726         inneriter                  += j_index_end - j_index_start;
 727
 728         /* Outer loop uses 7 flops */
 729     }
 730
 731     /* Increment number of outer iterations */
 732     outeriter        += nri;
 733
 734     /* Update outer/inner flops */
 735
 736     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_F,outeriter*7 + inneriter*57);
 737 }