src/gromacs/gmxlib/nonbonded/nb_kernel_avx_256_single/nb_kernel_ElecEwSh_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_ElecEwSh_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_ElecEwSh_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     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 128     rcutoff_scalar   = fr->rcoulomb;
 129     rcutoff          = _mm256_set1_ps(rcutoff_scalar);
 130     rcutoff2         = _mm256_mul_ps(rcutoff,rcutoff);
 131
 132     /* Avoid stupid compiler warnings */
 133     jnrA = jnrB = jnrC = jnrD = jnrE = jnrF = jnrG = jnrH = 0;
 134     j_coord_offsetA = 0;
 135     j_coord_offsetB = 0;
 136     j_coord_offsetC = 0;
 137     j_coord_offsetD = 0;
 138     j_coord_offsetE = 0;
 139     j_coord_offsetF = 0;
 140     j_coord_offsetG = 0;
 141     j_coord_offsetH = 0;
 142
 143     outeriter        = 0;
 144     inneriter        = 0;
 145
 146     for(iidx=0;iidx<4*DIM;iidx++)
 147     {
 148         scratch[iidx] = 0.0;
 149     }
 150
 151     /* Start outer loop over neighborlists */
 152     for(iidx=0; iidx<nri; iidx++)
 153     {
 154         /* Load shift vector for this list */
 155         i_shift_offset   = DIM*shiftidx[iidx];
 156
 157         /* Load limits for loop over neighbors */
 158         j_index_start    = jindex[iidx];
 159         j_index_end      = jindex[iidx+1];
 160
 161         /* Get outer coordinate index */
 162         inr              = iinr[iidx];
 163         i_coord_offset   = DIM*inr;
 164
 165         /* Load i particle coords and add shift vector */
 166         gmx_mm256_load_shift_and_1rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 167
 168         fix0             = _mm256_setzero_ps();
 169         fiy0             = _mm256_setzero_ps();
 170         fiz0             = _mm256_setzero_ps();
 171
 172         /* Load parameters for i particles */
 173         iq0              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+0]));
 174
 175         /* Reset potential sums */
 176         velecsum         = _mm256_setzero_ps();
 177
 178         /* Start inner kernel loop */
 179         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+7]>=0; jidx+=8)
 180         {
 181
 182             /* Get j neighbor index, and coordinate index */
 183             jnrA             = jjnr[jidx];
 184             jnrB             = jjnr[jidx+1];
 185             jnrC             = jjnr[jidx+2];
 186             jnrD             = jjnr[jidx+3];
 187             jnrE             = jjnr[jidx+4];
 188             jnrF             = jjnr[jidx+5];
 189             jnrG             = jjnr[jidx+6];
 190             jnrH             = jjnr[jidx+7];
 191             j_coord_offsetA  = DIM*jnrA;
 192             j_coord_offsetB  = DIM*jnrB;
 193             j_coord_offsetC  = DIM*jnrC;
 194             j_coord_offsetD  = DIM*jnrD;
 195             j_coord_offsetE  = DIM*jnrE;
 196             j_coord_offsetF  = DIM*jnrF;
 197             j_coord_offsetG  = DIM*jnrG;
 198             j_coord_offsetH  = DIM*jnrH;
 199
 200             /* load j atom coordinates */
 201             gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 202                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 203                                                  x+j_coord_offsetE,x+j_coord_offsetF,
 204                                                  x+j_coord_offsetG,x+j_coord_offsetH,
 205                                                  &jx0,&jy0,&jz0);
 206
 207             /* Calculate displacement vector */
 208             dx00             = _mm256_sub_ps(ix0,jx0);
 209             dy00             = _mm256_sub_ps(iy0,jy0);
 210             dz00             = _mm256_sub_ps(iz0,jz0);
 211
 212             /* Calculate squared distance and things based on it */
 213             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
 214
 215             rinv00           = gmx_mm256_invsqrt_ps(rsq00);
 216
 217             rinvsq00         = _mm256_mul_ps(rinv00,rinv00);
 218
 219             /* Load parameters for j particles */
 220             jq0              = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 221                                                                  charge+jnrC+0,charge+jnrD+0,
 222                                                                  charge+jnrE+0,charge+jnrF+0,
 223                                                                  charge+jnrG+0,charge+jnrH+0);
 224
 225             /**************************
 226              * CALCULATE INTERACTIONS *
 227              **************************/
 228
 229             if (gmx_mm256_any_lt(rsq00,rcutoff2))
 230             {
 231
 232             r00              = _mm256_mul_ps(rsq00,rinv00);
 233
 234             /* Compute parameters for interactions between i and j atoms */
 235             qq00             = _mm256_mul_ps(iq0,jq0);
 236
 237             /* EWALD ELECTROSTATICS */
 238
 239             /* Analytical PME correction */
 240             zeta2            = _mm256_mul_ps(beta2,rsq00);
 241             rinv3            = _mm256_mul_ps(rinvsq00,rinv00);
 242             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
 243             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
 244             felec            = _mm256_mul_ps(qq00,felec);
 245             pmecorrV         = gmx_mm256_pmecorrV_ps(zeta2);
 246             pmecorrV         = _mm256_mul_ps(pmecorrV,beta);
 247             velec            = _mm256_sub_ps(_mm256_sub_ps(rinv00,sh_ewald),pmecorrV);
 248             velec            = _mm256_mul_ps(qq00,velec);
 249
 250             cutoff_mask      = _mm256_cmp_ps(rsq00,rcutoff2,_CMP_LT_OQ);
 251
 252             /* Update potential sum for this i atom from the interaction with this j atom. */
 253             velec            = _mm256_and_ps(velec,cutoff_mask);
 254             velecsum         = _mm256_add_ps(velecsum,velec);
 255
 256             fscal            = felec;
 257
 258             fscal            = _mm256_and_ps(fscal,cutoff_mask);
 259
 260             /* Calculate temporary vectorial force */
 261             tx               = _mm256_mul_ps(fscal,dx00);
 262             ty               = _mm256_mul_ps(fscal,dy00);
 263             tz               = _mm256_mul_ps(fscal,dz00);
 264
 265             /* Update vectorial force */
 266             fix0             = _mm256_add_ps(fix0,tx);
 267             fiy0             = _mm256_add_ps(fiy0,ty);
 268             fiz0             = _mm256_add_ps(fiz0,tz);
 269
 270             fjptrA             = f+j_coord_offsetA;
 271             fjptrB             = f+j_coord_offsetB;
 272             fjptrC             = f+j_coord_offsetC;
 273             fjptrD             = f+j_coord_offsetD;
 274             fjptrE             = f+j_coord_offsetE;
 275             fjptrF             = f+j_coord_offsetF;
 276             fjptrG             = f+j_coord_offsetG;
 277             fjptrH             = f+j_coord_offsetH;
 278             gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,tx,ty,tz);
 279
 280             }
 281
 282             /* Inner loop uses 109 flops */
 283         }
 284
 285         if(jidx<j_index_end)
 286         {
 287
 288             /* Get j neighbor index, and coordinate index */
 289             jnrlistA         = jjnr[jidx];
 290             jnrlistB         = jjnr[jidx+1];
 291             jnrlistC         = jjnr[jidx+2];
 292             jnrlistD         = jjnr[jidx+3];
 293             jnrlistE         = jjnr[jidx+4];
 294             jnrlistF         = jjnr[jidx+5];
 295             jnrlistG         = jjnr[jidx+6];
 296             jnrlistH         = jjnr[jidx+7];
 297             /* Sign of each element will be negative for non-real atoms.
 298              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 299              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 300              */
 301             dummy_mask = gmx_mm256_set_m128(gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx+4)),_mm_setzero_si128())),
 302                                             gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128())));
 303
 304             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 305             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 306             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 307             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 308             jnrE       = (jnrlistE>=0) ? jnrlistE : 0;
 309             jnrF       = (jnrlistF>=0) ? jnrlistF : 0;
 310             jnrG       = (jnrlistG>=0) ? jnrlistG : 0;
 311             jnrH       = (jnrlistH>=0) ? jnrlistH : 0;
 312             j_coord_offsetA  = DIM*jnrA;
 313             j_coord_offsetB  = DIM*jnrB;
 314             j_coord_offsetC  = DIM*jnrC;
 315             j_coord_offsetD  = DIM*jnrD;
 316             j_coord_offsetE  = DIM*jnrE;
 317             j_coord_offsetF  = DIM*jnrF;
 318             j_coord_offsetG  = DIM*jnrG;
 319             j_coord_offsetH  = DIM*jnrH;
 320
 321             /* load j atom coordinates */
 322             gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 323                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 324                                                  x+j_coord_offsetE,x+j_coord_offsetF,
 325                                                  x+j_coord_offsetG,x+j_coord_offsetH,
 326                                                  &jx0,&jy0,&jz0);
 327
 328             /* Calculate displacement vector */
 329             dx00             = _mm256_sub_ps(ix0,jx0);
 330             dy00             = _mm256_sub_ps(iy0,jy0);
 331             dz00             = _mm256_sub_ps(iz0,jz0);
 332
 333             /* Calculate squared distance and things based on it */
 334             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
 335
 336             rinv00           = gmx_mm256_invsqrt_ps(rsq00);
 337
 338             rinvsq00         = _mm256_mul_ps(rinv00,rinv00);
 339
 340             /* Load parameters for j particles */
 341             jq0              = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 342                                                                  charge+jnrC+0,charge+jnrD+0,
 343                                                                  charge+jnrE+0,charge+jnrF+0,
 344                                                                  charge+jnrG+0,charge+jnrH+0);
 345
 346             /**************************
 347              * CALCULATE INTERACTIONS *
 348              **************************/
 349
 350             if (gmx_mm256_any_lt(rsq00,rcutoff2))
 351             {
 352
 353             r00              = _mm256_mul_ps(rsq00,rinv00);
 354             r00              = _mm256_andnot_ps(dummy_mask,r00);
 355
 356             /* Compute parameters for interactions between i and j atoms */
 357             qq00             = _mm256_mul_ps(iq0,jq0);
 358
 359             /* EWALD ELECTROSTATICS */
 360
 361             /* Analytical PME correction */
 362             zeta2            = _mm256_mul_ps(beta2,rsq00);
 363             rinv3            = _mm256_mul_ps(rinvsq00,rinv00);
 364             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
 365             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
 366             felec            = _mm256_mul_ps(qq00,felec);
 367             pmecorrV         = gmx_mm256_pmecorrV_ps(zeta2);
 368             pmecorrV         = _mm256_mul_ps(pmecorrV,beta);
 369             velec            = _mm256_sub_ps(_mm256_sub_ps(rinv00,sh_ewald),pmecorrV);
 370             velec            = _mm256_mul_ps(qq00,velec);
 371
 372             cutoff_mask      = _mm256_cmp_ps(rsq00,rcutoff2,_CMP_LT_OQ);
 373
 374             /* Update potential sum for this i atom from the interaction with this j atom. */
 375             velec            = _mm256_and_ps(velec,cutoff_mask);
 376             velec            = _mm256_andnot_ps(dummy_mask,velec);
 377             velecsum         = _mm256_add_ps(velecsum,velec);
 378
 379             fscal            = felec;
 380
 381             fscal            = _mm256_and_ps(fscal,cutoff_mask);
 382
 383             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
 384
 385             /* Calculate temporary vectorial force */
 386             tx               = _mm256_mul_ps(fscal,dx00);
 387             ty               = _mm256_mul_ps(fscal,dy00);
 388             tz               = _mm256_mul_ps(fscal,dz00);
 389
 390             /* Update vectorial force */
 391             fix0             = _mm256_add_ps(fix0,tx);
 392             fiy0             = _mm256_add_ps(fiy0,ty);
 393             fiz0             = _mm256_add_ps(fiz0,tz);
 394
 395             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 396             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 397             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 398             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 399             fjptrE             = (jnrlistE>=0) ? f+j_coord_offsetE : scratch;
 400             fjptrF             = (jnrlistF>=0) ? f+j_coord_offsetF : scratch;
 401             fjptrG             = (jnrlistG>=0) ? f+j_coord_offsetG : scratch;
 402             fjptrH             = (jnrlistH>=0) ? f+j_coord_offsetH : scratch;
 403             gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,tx,ty,tz);
 404
 405             }
 406
 407             /* Inner loop uses 110 flops */
 408         }
 409
 410         /* End of innermost loop */
 411
 412         gmx_mm256_update_iforce_1atom_swizzle_ps(fix0,fiy0,fiz0,
 413                                                  f+i_coord_offset,fshift+i_shift_offset);
 414
 415         ggid                        = gid[iidx];
 416         /* Update potential energies */
 417         gmx_mm256_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid);
 418
 419         /* Increment number of inner iterations */
 420         inneriter                  += j_index_end - j_index_start;
 421
 422         /* Outer loop uses 8 flops */
 423     }
 424
 425     /* Increment number of outer iterations */
 426     outeriter        += nri;
 427
 428     /* Update outer/inner flops */
 429
 430     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VF,outeriter*8 + inneriter*110);
 431 }
 432 /*
 433  * Gromacs nonbonded kernel:   nb_kernel_ElecEwSh_VdwNone_GeomP1P1_F_avx_256_single
 434  * Electrostatics interaction: Ewald
 435  * VdW interaction:            None
 436  * Geometry:                   Particle-Particle
 437  * Calculate force/pot:        Force
 438  */
 439 void
 440 nb_kernel_ElecEwSh_VdwNone_GeomP1P1_F_avx_256_single
 441                     (t_nblist                    * gmx_restrict       nlist,
 442                      rvec                        * gmx_restrict          xx,
 443                      rvec                        * gmx_restrict          ff,
 444                      t_forcerec                  * gmx_restrict          fr,
 445                      t_mdatoms                   * gmx_restrict     mdatoms,
 446                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 447                      t_nrnb                      * gmx_restrict        nrnb)
 448 {
 449     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 450      * just 0 for non-waters.
 451      * Suffixes A,B,C,D,E,F,G,H refer to j loop unrolling done with AVX, e.g. for the eight different
 452      * jnr indices corresponding to data put in the four positions in the SIMD register.
 453      */
 454     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 455     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 456     int              jnrA,jnrB,jnrC,jnrD;
 457     int              jnrE,jnrF,jnrG,jnrH;
 458     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
 459     int              jnrlistE,jnrlistF,jnrlistG,jnrlistH;
 460     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 461     int              j_coord_offsetE,j_coord_offsetF,j_coord_offsetG,j_coord_offsetH;
 462     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 463     real             rcutoff_scalar;
 464     real             *shiftvec,*fshift,*x,*f;
 465     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD,*fjptrE,*fjptrF,*fjptrG,*fjptrH;
 466     real             scratch[4*DIM];
 467     __m256           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 468     real *           vdwioffsetptr0;
 469     __m256           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 470     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D,vdwjidx0E,vdwjidx0F,vdwjidx0G,vdwjidx0H;
 471     __m256           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 472     __m256           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 473     __m256           velec,felec,velecsum,facel,crf,krf,krf2;
 474     real             *charge;
 475     __m256i          ewitab;
 476     __m128i          ewitab_lo,ewitab_hi;
 477     __m256           ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
 478     __m256           beta,beta2,beta3,zeta2,pmecorrF,pmecorrV,rinv3;
 479     real             *ewtab;
 480     __m256           dummy_mask,cutoff_mask;
 481     __m256           signbit = _mm256_castsi256_ps( _mm256_set1_epi32(0x80000000) );
 482     __m256           one     = _mm256_set1_ps(1.0);
 483     __m256           two     = _mm256_set1_ps(2.0);
 484     x                = xx[0];
 485     f                = ff[0];
 486
 487     nri              = nlist->nri;
 488     iinr             = nlist->iinr;
 489     jindex           = nlist->jindex;
 490     jjnr             = nlist->jjnr;
 491     shiftidx         = nlist->shift;
 492     gid              = nlist->gid;
 493     shiftvec         = fr->shift_vec[0];
 494     fshift           = fr->fshift[0];
 495     facel            = _mm256_set1_ps(fr->epsfac);
 496     charge           = mdatoms->chargeA;
 497
 498     sh_ewald         = _mm256_set1_ps(fr->ic->sh_ewald);
 499     beta             = _mm256_set1_ps(fr->ic->ewaldcoeff_q);
 500     beta2            = _mm256_mul_ps(beta,beta);
 501     beta3            = _mm256_mul_ps(beta,beta2);
 502
 503     ewtab            = fr->ic->tabq_coul_F;
 504     ewtabscale       = _mm256_set1_ps(fr->ic->tabq_scale);
 505     ewtabhalfspace   = _mm256_set1_ps(0.5/fr->ic->tabq_scale);
 506
 507     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 508     rcutoff_scalar   = fr->rcoulomb;
 509     rcutoff          = _mm256_set1_ps(rcutoff_scalar);
 510     rcutoff2         = _mm256_mul_ps(rcutoff,rcutoff);
 511
 512     /* Avoid stupid compiler warnings */
 513     jnrA = jnrB = jnrC = jnrD = jnrE = jnrF = jnrG = jnrH = 0;
 514     j_coord_offsetA = 0;
 515     j_coord_offsetB = 0;
 516     j_coord_offsetC = 0;
 517     j_coord_offsetD = 0;
 518     j_coord_offsetE = 0;
 519     j_coord_offsetF = 0;
 520     j_coord_offsetG = 0;
 521     j_coord_offsetH = 0;
 522
 523     outeriter        = 0;
 524     inneriter        = 0;
 525
 526     for(iidx=0;iidx<4*DIM;iidx++)
 527     {
 528         scratch[iidx] = 0.0;
 529     }
 530
 531     /* Start outer loop over neighborlists */
 532     for(iidx=0; iidx<nri; iidx++)
 533     {
 534         /* Load shift vector for this list */
 535         i_shift_offset   = DIM*shiftidx[iidx];
 536
 537         /* Load limits for loop over neighbors */
 538         j_index_start    = jindex[iidx];
 539         j_index_end      = jindex[iidx+1];
 540
 541         /* Get outer coordinate index */
 542         inr              = iinr[iidx];
 543         i_coord_offset   = DIM*inr;
 544
 545         /* Load i particle coords and add shift vector */
 546         gmx_mm256_load_shift_and_1rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 547
 548         fix0             = _mm256_setzero_ps();
 549         fiy0             = _mm256_setzero_ps();
 550         fiz0             = _mm256_setzero_ps();
 551
 552         /* Load parameters for i particles */
 553         iq0              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+0]));
 554
 555         /* Start inner kernel loop */
 556         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+7]>=0; jidx+=8)
 557         {
 558
 559             /* Get j neighbor index, and coordinate index */
 560             jnrA             = jjnr[jidx];
 561             jnrB             = jjnr[jidx+1];
 562             jnrC             = jjnr[jidx+2];
 563             jnrD             = jjnr[jidx+3];
 564             jnrE             = jjnr[jidx+4];
 565             jnrF             = jjnr[jidx+5];
 566             jnrG             = jjnr[jidx+6];
 567             jnrH             = jjnr[jidx+7];
 568             j_coord_offsetA  = DIM*jnrA;
 569             j_coord_offsetB  = DIM*jnrB;
 570             j_coord_offsetC  = DIM*jnrC;
 571             j_coord_offsetD  = DIM*jnrD;
 572             j_coord_offsetE  = DIM*jnrE;
 573             j_coord_offsetF  = DIM*jnrF;
 574             j_coord_offsetG  = DIM*jnrG;
 575             j_coord_offsetH  = DIM*jnrH;
 576
 577             /* load j atom coordinates */
 578             gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 579                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 580                                                  x+j_coord_offsetE,x+j_coord_offsetF,
 581                                                  x+j_coord_offsetG,x+j_coord_offsetH,
 582                                                  &jx0,&jy0,&jz0);
 583
 584             /* Calculate displacement vector */
 585             dx00             = _mm256_sub_ps(ix0,jx0);
 586             dy00             = _mm256_sub_ps(iy0,jy0);
 587             dz00             = _mm256_sub_ps(iz0,jz0);
 588
 589             /* Calculate squared distance and things based on it */
 590             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
 591
 592             rinv00           = gmx_mm256_invsqrt_ps(rsq00);
 593
 594             rinvsq00         = _mm256_mul_ps(rinv00,rinv00);
 595
 596             /* Load parameters for j particles */
 597             jq0              = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 598                                                                  charge+jnrC+0,charge+jnrD+0,
 599                                                                  charge+jnrE+0,charge+jnrF+0,
 600                                                                  charge+jnrG+0,charge+jnrH+0);
 601
 602             /**************************
 603              * CALCULATE INTERACTIONS *
 604              **************************/
 605
 606             if (gmx_mm256_any_lt(rsq00,rcutoff2))
 607             {
 608
 609             r00              = _mm256_mul_ps(rsq00,rinv00);
 610
 611             /* Compute parameters for interactions between i and j atoms */
 612             qq00             = _mm256_mul_ps(iq0,jq0);
 613
 614             /* EWALD ELECTROSTATICS */
 615
 616             /* Analytical PME correction */
 617             zeta2            = _mm256_mul_ps(beta2,rsq00);
 618             rinv3            = _mm256_mul_ps(rinvsq00,rinv00);
 619             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
 620             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
 621             felec            = _mm256_mul_ps(qq00,felec);
 622
 623             cutoff_mask      = _mm256_cmp_ps(rsq00,rcutoff2,_CMP_LT_OQ);
 624
 625             fscal            = felec;
 626
 627             fscal            = _mm256_and_ps(fscal,cutoff_mask);
 628
 629             /* Calculate temporary vectorial force */
 630             tx               = _mm256_mul_ps(fscal,dx00);
 631             ty               = _mm256_mul_ps(fscal,dy00);
 632             tz               = _mm256_mul_ps(fscal,dz00);
 633
 634             /* Update vectorial force */
 635             fix0             = _mm256_add_ps(fix0,tx);
 636             fiy0             = _mm256_add_ps(fiy0,ty);
 637             fiz0             = _mm256_add_ps(fiz0,tz);
 638
 639             fjptrA             = f+j_coord_offsetA;
 640             fjptrB             = f+j_coord_offsetB;
 641             fjptrC             = f+j_coord_offsetC;
 642             fjptrD             = f+j_coord_offsetD;
 643             fjptrE             = f+j_coord_offsetE;
 644             fjptrF             = f+j_coord_offsetF;
 645             fjptrG             = f+j_coord_offsetG;
 646             fjptrH             = f+j_coord_offsetH;
 647             gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,tx,ty,tz);
 648
 649             }
 650
 651             /* Inner loop uses 59 flops */
 652         }
 653
 654         if(jidx<j_index_end)
 655         {
 656
 657             /* Get j neighbor index, and coordinate index */
 658             jnrlistA         = jjnr[jidx];
 659             jnrlistB         = jjnr[jidx+1];
 660             jnrlistC         = jjnr[jidx+2];
 661             jnrlistD         = jjnr[jidx+3];
 662             jnrlistE         = jjnr[jidx+4];
 663             jnrlistF         = jjnr[jidx+5];
 664             jnrlistG         = jjnr[jidx+6];
 665             jnrlistH         = jjnr[jidx+7];
 666             /* Sign of each element will be negative for non-real atoms.
 667              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 668              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 669              */
 670             dummy_mask = gmx_mm256_set_m128(gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx+4)),_mm_setzero_si128())),
 671                                             gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128())));
 672
 673             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 674             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 675             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 676             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 677             jnrE       = (jnrlistE>=0) ? jnrlistE : 0;
 678             jnrF       = (jnrlistF>=0) ? jnrlistF : 0;
 679             jnrG       = (jnrlistG>=0) ? jnrlistG : 0;
 680             jnrH       = (jnrlistH>=0) ? jnrlistH : 0;
 681             j_coord_offsetA  = DIM*jnrA;
 682             j_coord_offsetB  = DIM*jnrB;
 683             j_coord_offsetC  = DIM*jnrC;
 684             j_coord_offsetD  = DIM*jnrD;
 685             j_coord_offsetE  = DIM*jnrE;
 686             j_coord_offsetF  = DIM*jnrF;
 687             j_coord_offsetG  = DIM*jnrG;
 688             j_coord_offsetH  = DIM*jnrH;
 689
 690             /* load j atom coordinates */
 691             gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 692                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 693                                                  x+j_coord_offsetE,x+j_coord_offsetF,
 694                                                  x+j_coord_offsetG,x+j_coord_offsetH,
 695                                                  &jx0,&jy0,&jz0);
 696
 697             /* Calculate displacement vector */
 698             dx00             = _mm256_sub_ps(ix0,jx0);
 699             dy00             = _mm256_sub_ps(iy0,jy0);
 700             dz00             = _mm256_sub_ps(iz0,jz0);
 701
 702             /* Calculate squared distance and things based on it */
 703             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
 704
 705             rinv00           = gmx_mm256_invsqrt_ps(rsq00);
 706
 707             rinvsq00         = _mm256_mul_ps(rinv00,rinv00);
 708
 709             /* Load parameters for j particles */
 710             jq0              = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 711                                                                  charge+jnrC+0,charge+jnrD+0,
 712                                                                  charge+jnrE+0,charge+jnrF+0,
 713                                                                  charge+jnrG+0,charge+jnrH+0);
 714
 715             /**************************
 716              * CALCULATE INTERACTIONS *
 717              **************************/
 718
 719             if (gmx_mm256_any_lt(rsq00,rcutoff2))
 720             {
 721
 722             r00              = _mm256_mul_ps(rsq00,rinv00);
 723             r00              = _mm256_andnot_ps(dummy_mask,r00);
 724
 725             /* Compute parameters for interactions between i and j atoms */
 726             qq00             = _mm256_mul_ps(iq0,jq0);
 727
 728             /* EWALD ELECTROSTATICS */
 729
 730             /* Analytical PME correction */
 731             zeta2            = _mm256_mul_ps(beta2,rsq00);
 732             rinv3            = _mm256_mul_ps(rinvsq00,rinv00);
 733             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
 734             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
 735             felec            = _mm256_mul_ps(qq00,felec);
 736
 737             cutoff_mask      = _mm256_cmp_ps(rsq00,rcutoff2,_CMP_LT_OQ);
 738
 739             fscal            = felec;
 740
 741             fscal            = _mm256_and_ps(fscal,cutoff_mask);
 742
 743             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
 744
 745             /* Calculate temporary vectorial force */
 746             tx               = _mm256_mul_ps(fscal,dx00);
 747             ty               = _mm256_mul_ps(fscal,dy00);
 748             tz               = _mm256_mul_ps(fscal,dz00);
 749
 750             /* Update vectorial force */
 751             fix0             = _mm256_add_ps(fix0,tx);
 752             fiy0             = _mm256_add_ps(fiy0,ty);
 753             fiz0             = _mm256_add_ps(fiz0,tz);
 754
 755             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 756             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 757             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 758             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 759             fjptrE             = (jnrlistE>=0) ? f+j_coord_offsetE : scratch;
 760             fjptrF             = (jnrlistF>=0) ? f+j_coord_offsetF : scratch;
 761             fjptrG             = (jnrlistG>=0) ? f+j_coord_offsetG : scratch;
 762             fjptrH             = (jnrlistH>=0) ? f+j_coord_offsetH : scratch;
 763             gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,tx,ty,tz);
 764
 765             }
 766
 767             /* Inner loop uses 60 flops */
 768         }
 769
 770         /* End of innermost loop */
 771
 772         gmx_mm256_update_iforce_1atom_swizzle_ps(fix0,fiy0,fiz0,
 773                                                  f+i_coord_offset,fshift+i_shift_offset);
 774
 775         /* Increment number of inner iterations */
 776         inneriter                  += j_index_end - j_index_start;
 777
 778         /* Outer loop uses 7 flops */
 779     }
 780
 781     /* Increment number of outer iterations */
 782     outeriter        += nri;
 783
 784     /* Update outer/inner flops */
 785
 786     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_F,outeriter*7 + inneriter*60);
 787 }