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