src/gromacs/gmxlib/nonbonded/nb_kernel_avx_256_single/nb_kernel_ElecCoul_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_ElecCoul_VdwNone_GeomP1P1_VF_avx_256_single
  52  * Electrostatics interaction: Coulomb
  53  * VdW interaction:            None
  54  * Geometry:                   Particle-Particle
  55  * Calculate force/pot:        PotentialAndForce
  56  */
  57 void
  58 nb_kernel_ElecCoul_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     __m256           dummy_mask,cutoff_mask;
  94     __m256           signbit = _mm256_castsi256_ps( _mm256_set1_epi32(0x80000000) );
  95     __m256           one     = _mm256_set1_ps(1.0);
  96     __m256           two     = _mm256_set1_ps(2.0);
  97     x                = xx[0];
  98     f                = ff[0];
  99
 100     nri              = nlist->nri;
 101     iinr             = nlist->iinr;
 102     jindex           = nlist->jindex;
 103     jjnr             = nlist->jjnr;
 104     shiftidx         = nlist->shift;
 105     gid              = nlist->gid;
 106     shiftvec         = fr->shift_vec[0];
 107     fshift           = fr->fshift[0];
 108     facel            = _mm256_set1_ps(fr->epsfac);
 109     charge           = mdatoms->chargeA;
 110
 111     /* Avoid stupid compiler warnings */
 112     jnrA = jnrB = jnrC = jnrD = jnrE = jnrF = jnrG = jnrH = 0;
 113     j_coord_offsetA = 0;
 114     j_coord_offsetB = 0;
 115     j_coord_offsetC = 0;
 116     j_coord_offsetD = 0;
 117     j_coord_offsetE = 0;
 118     j_coord_offsetF = 0;
 119     j_coord_offsetG = 0;
 120     j_coord_offsetH = 0;
 121
 122     outeriter        = 0;
 123     inneriter        = 0;
 124
 125     for(iidx=0;iidx<4*DIM;iidx++)
 126     {
 127         scratch[iidx] = 0.0;
 128     }
 129
 130     /* Start outer loop over neighborlists */
 131     for(iidx=0; iidx<nri; iidx++)
 132     {
 133         /* Load shift vector for this list */
 134         i_shift_offset   = DIM*shiftidx[iidx];
 135
 136         /* Load limits for loop over neighbors */
 137         j_index_start    = jindex[iidx];
 138         j_index_end      = jindex[iidx+1];
 139
 140         /* Get outer coordinate index */
 141         inr              = iinr[iidx];
 142         i_coord_offset   = DIM*inr;
 143
 144         /* Load i particle coords and add shift vector */
 145         gmx_mm256_load_shift_and_1rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 146
 147         fix0             = _mm256_setzero_ps();
 148         fiy0             = _mm256_setzero_ps();
 149         fiz0             = _mm256_setzero_ps();
 150
 151         /* Load parameters for i particles */
 152         iq0              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+0]));
 153
 154         /* Reset potential sums */
 155         velecsum         = _mm256_setzero_ps();
 156
 157         /* Start inner kernel loop */
 158         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+7]>=0; jidx+=8)
 159         {
 160
 161             /* Get j neighbor index, and coordinate index */
 162             jnrA             = jjnr[jidx];
 163             jnrB             = jjnr[jidx+1];
 164             jnrC             = jjnr[jidx+2];
 165             jnrD             = jjnr[jidx+3];
 166             jnrE             = jjnr[jidx+4];
 167             jnrF             = jjnr[jidx+5];
 168             jnrG             = jjnr[jidx+6];
 169             jnrH             = jjnr[jidx+7];
 170             j_coord_offsetA  = DIM*jnrA;
 171             j_coord_offsetB  = DIM*jnrB;
 172             j_coord_offsetC  = DIM*jnrC;
 173             j_coord_offsetD  = DIM*jnrD;
 174             j_coord_offsetE  = DIM*jnrE;
 175             j_coord_offsetF  = DIM*jnrF;
 176             j_coord_offsetG  = DIM*jnrG;
 177             j_coord_offsetH  = DIM*jnrH;
 178
 179             /* load j atom coordinates */
 180             gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 181                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 182                                                  x+j_coord_offsetE,x+j_coord_offsetF,
 183                                                  x+j_coord_offsetG,x+j_coord_offsetH,
 184                                                  &jx0,&jy0,&jz0);
 185
 186             /* Calculate displacement vector */
 187             dx00             = _mm256_sub_ps(ix0,jx0);
 188             dy00             = _mm256_sub_ps(iy0,jy0);
 189             dz00             = _mm256_sub_ps(iz0,jz0);
 190
 191             /* Calculate squared distance and things based on it */
 192             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
 193
 194             rinv00           = gmx_mm256_invsqrt_ps(rsq00);
 195
 196             rinvsq00         = _mm256_mul_ps(rinv00,rinv00);
 197
 198             /* Load parameters for j particles */
 199             jq0              = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 200                                                                  charge+jnrC+0,charge+jnrD+0,
 201                                                                  charge+jnrE+0,charge+jnrF+0,
 202                                                                  charge+jnrG+0,charge+jnrH+0);
 203
 204             /**************************
 205              * CALCULATE INTERACTIONS *
 206              **************************/
 207
 208             /* Compute parameters for interactions between i and j atoms */
 209             qq00             = _mm256_mul_ps(iq0,jq0);
 210
 211             /* COULOMB ELECTROSTATICS */
 212             velec            = _mm256_mul_ps(qq00,rinv00);
 213             felec            = _mm256_mul_ps(velec,rinvsq00);
 214
 215             /* Update potential sum for this i atom from the interaction with this j atom. */
 216             velecsum         = _mm256_add_ps(velecsum,velec);
 217
 218             fscal            = felec;
 219
 220             /* Calculate temporary vectorial force */
 221             tx               = _mm256_mul_ps(fscal,dx00);
 222             ty               = _mm256_mul_ps(fscal,dy00);
 223             tz               = _mm256_mul_ps(fscal,dz00);
 224
 225             /* Update vectorial force */
 226             fix0             = _mm256_add_ps(fix0,tx);
 227             fiy0             = _mm256_add_ps(fiy0,ty);
 228             fiz0             = _mm256_add_ps(fiz0,tz);
 229
 230             fjptrA             = f+j_coord_offsetA;
 231             fjptrB             = f+j_coord_offsetB;
 232             fjptrC             = f+j_coord_offsetC;
 233             fjptrD             = f+j_coord_offsetD;
 234             fjptrE             = f+j_coord_offsetE;
 235             fjptrF             = f+j_coord_offsetF;
 236             fjptrG             = f+j_coord_offsetG;
 237             fjptrH             = f+j_coord_offsetH;
 238             gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,tx,ty,tz);
 239
 240             /* Inner loop uses 27 flops */
 241         }
 242
 243         if(jidx<j_index_end)
 244         {
 245
 246             /* Get j neighbor index, and coordinate index */
 247             jnrlistA         = jjnr[jidx];
 248             jnrlistB         = jjnr[jidx+1];
 249             jnrlistC         = jjnr[jidx+2];
 250             jnrlistD         = jjnr[jidx+3];
 251             jnrlistE         = jjnr[jidx+4];
 252             jnrlistF         = jjnr[jidx+5];
 253             jnrlistG         = jjnr[jidx+6];
 254             jnrlistH         = jjnr[jidx+7];
 255             /* Sign of each element will be negative for non-real atoms.
 256              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 257              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 258              */
 259             dummy_mask = gmx_mm256_set_m128(gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx+4)),_mm_setzero_si128())),
 260                                             gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128())));
 261
 262             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 263             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 264             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 265             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 266             jnrE       = (jnrlistE>=0) ? jnrlistE : 0;
 267             jnrF       = (jnrlistF>=0) ? jnrlistF : 0;
 268             jnrG       = (jnrlistG>=0) ? jnrlistG : 0;
 269             jnrH       = (jnrlistH>=0) ? jnrlistH : 0;
 270             j_coord_offsetA  = DIM*jnrA;
 271             j_coord_offsetB  = DIM*jnrB;
 272             j_coord_offsetC  = DIM*jnrC;
 273             j_coord_offsetD  = DIM*jnrD;
 274             j_coord_offsetE  = DIM*jnrE;
 275             j_coord_offsetF  = DIM*jnrF;
 276             j_coord_offsetG  = DIM*jnrG;
 277             j_coord_offsetH  = DIM*jnrH;
 278
 279             /* load j atom coordinates */
 280             gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 281                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 282                                                  x+j_coord_offsetE,x+j_coord_offsetF,
 283                                                  x+j_coord_offsetG,x+j_coord_offsetH,
 284                                                  &jx0,&jy0,&jz0);
 285
 286             /* Calculate displacement vector */
 287             dx00             = _mm256_sub_ps(ix0,jx0);
 288             dy00             = _mm256_sub_ps(iy0,jy0);
 289             dz00             = _mm256_sub_ps(iz0,jz0);
 290
 291             /* Calculate squared distance and things based on it */
 292             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
 293
 294             rinv00           = gmx_mm256_invsqrt_ps(rsq00);
 295
 296             rinvsq00         = _mm256_mul_ps(rinv00,rinv00);
 297
 298             /* Load parameters for j particles */
 299             jq0              = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 300                                                                  charge+jnrC+0,charge+jnrD+0,
 301                                                                  charge+jnrE+0,charge+jnrF+0,
 302                                                                  charge+jnrG+0,charge+jnrH+0);
 303
 304             /**************************
 305              * CALCULATE INTERACTIONS *
 306              **************************/
 307
 308             /* Compute parameters for interactions between i and j atoms */
 309             qq00             = _mm256_mul_ps(iq0,jq0);
 310
 311             /* COULOMB ELECTROSTATICS */
 312             velec            = _mm256_mul_ps(qq00,rinv00);
 313             felec            = _mm256_mul_ps(velec,rinvsq00);
 314
 315             /* Update potential sum for this i atom from the interaction with this j atom. */
 316             velec            = _mm256_andnot_ps(dummy_mask,velec);
 317             velecsum         = _mm256_add_ps(velecsum,velec);
 318
 319             fscal            = felec;
 320
 321             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
 322
 323             /* Calculate temporary vectorial force */
 324             tx               = _mm256_mul_ps(fscal,dx00);
 325             ty               = _mm256_mul_ps(fscal,dy00);
 326             tz               = _mm256_mul_ps(fscal,dz00);
 327
 328             /* Update vectorial force */
 329             fix0             = _mm256_add_ps(fix0,tx);
 330             fiy0             = _mm256_add_ps(fiy0,ty);
 331             fiz0             = _mm256_add_ps(fiz0,tz);
 332
 333             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 334             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 335             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 336             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 337             fjptrE             = (jnrlistE>=0) ? f+j_coord_offsetE : scratch;
 338             fjptrF             = (jnrlistF>=0) ? f+j_coord_offsetF : scratch;
 339             fjptrG             = (jnrlistG>=0) ? f+j_coord_offsetG : scratch;
 340             fjptrH             = (jnrlistH>=0) ? f+j_coord_offsetH : scratch;
 341             gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,tx,ty,tz);
 342
 343             /* Inner loop uses 27 flops */
 344         }
 345
 346         /* End of innermost loop */
 347
 348         gmx_mm256_update_iforce_1atom_swizzle_ps(fix0,fiy0,fiz0,
 349                                                  f+i_coord_offset,fshift+i_shift_offset);
 350
 351         ggid                        = gid[iidx];
 352         /* Update potential energies */
 353         gmx_mm256_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid);
 354
 355         /* Increment number of inner iterations */
 356         inneriter                  += j_index_end - j_index_start;
 357
 358         /* Outer loop uses 8 flops */
 359     }
 360
 361     /* Increment number of outer iterations */
 362     outeriter        += nri;
 363
 364     /* Update outer/inner flops */
 365
 366     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VF,outeriter*8 + inneriter*27);
 367 }
 368 /*
 369  * Gromacs nonbonded kernel:   nb_kernel_ElecCoul_VdwNone_GeomP1P1_F_avx_256_single
 370  * Electrostatics interaction: Coulomb
 371  * VdW interaction:            None
 372  * Geometry:                   Particle-Particle
 373  * Calculate force/pot:        Force
 374  */
 375 void
 376 nb_kernel_ElecCoul_VdwNone_GeomP1P1_F_avx_256_single
 377                     (t_nblist                    * gmx_restrict       nlist,
 378                      rvec                        * gmx_restrict          xx,
 379                      rvec                        * gmx_restrict          ff,
 380                      t_forcerec                  * gmx_restrict          fr,
 381                      t_mdatoms                   * gmx_restrict     mdatoms,
 382                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 383                      t_nrnb                      * gmx_restrict        nrnb)
 384 {
 385     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 386      * just 0 for non-waters.
 387      * Suffixes A,B,C,D,E,F,G,H refer to j loop unrolling done with AVX, e.g. for the eight different
 388      * jnr indices corresponding to data put in the four positions in the SIMD register.
 389      */
 390     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 391     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 392     int              jnrA,jnrB,jnrC,jnrD;
 393     int              jnrE,jnrF,jnrG,jnrH;
 394     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
 395     int              jnrlistE,jnrlistF,jnrlistG,jnrlistH;
 396     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 397     int              j_coord_offsetE,j_coord_offsetF,j_coord_offsetG,j_coord_offsetH;
 398     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 399     real             rcutoff_scalar;
 400     real             *shiftvec,*fshift,*x,*f;
 401     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD,*fjptrE,*fjptrF,*fjptrG,*fjptrH;
 402     real             scratch[4*DIM];
 403     __m256           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 404     real *           vdwioffsetptr0;
 405     __m256           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 406     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D,vdwjidx0E,vdwjidx0F,vdwjidx0G,vdwjidx0H;
 407     __m256           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 408     __m256           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 409     __m256           velec,felec,velecsum,facel,crf,krf,krf2;
 410     real             *charge;
 411     __m256           dummy_mask,cutoff_mask;
 412     __m256           signbit = _mm256_castsi256_ps( _mm256_set1_epi32(0x80000000) );
 413     __m256           one     = _mm256_set1_ps(1.0);
 414     __m256           two     = _mm256_set1_ps(2.0);
 415     x                = xx[0];
 416     f                = ff[0];
 417
 418     nri              = nlist->nri;
 419     iinr             = nlist->iinr;
 420     jindex           = nlist->jindex;
 421     jjnr             = nlist->jjnr;
 422     shiftidx         = nlist->shift;
 423     gid              = nlist->gid;
 424     shiftvec         = fr->shift_vec[0];
 425     fshift           = fr->fshift[0];
 426     facel            = _mm256_set1_ps(fr->epsfac);
 427     charge           = mdatoms->chargeA;
 428
 429     /* Avoid stupid compiler warnings */
 430     jnrA = jnrB = jnrC = jnrD = jnrE = jnrF = jnrG = jnrH = 0;
 431     j_coord_offsetA = 0;
 432     j_coord_offsetB = 0;
 433     j_coord_offsetC = 0;
 434     j_coord_offsetD = 0;
 435     j_coord_offsetE = 0;
 436     j_coord_offsetF = 0;
 437     j_coord_offsetG = 0;
 438     j_coord_offsetH = 0;
 439
 440     outeriter        = 0;
 441     inneriter        = 0;
 442
 443     for(iidx=0;iidx<4*DIM;iidx++)
 444     {
 445         scratch[iidx] = 0.0;
 446     }
 447
 448     /* Start outer loop over neighborlists */
 449     for(iidx=0; iidx<nri; iidx++)
 450     {
 451         /* Load shift vector for this list */
 452         i_shift_offset   = DIM*shiftidx[iidx];
 453
 454         /* Load limits for loop over neighbors */
 455         j_index_start    = jindex[iidx];
 456         j_index_end      = jindex[iidx+1];
 457
 458         /* Get outer coordinate index */
 459         inr              = iinr[iidx];
 460         i_coord_offset   = DIM*inr;
 461
 462         /* Load i particle coords and add shift vector */
 463         gmx_mm256_load_shift_and_1rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 464
 465         fix0             = _mm256_setzero_ps();
 466         fiy0             = _mm256_setzero_ps();
 467         fiz0             = _mm256_setzero_ps();
 468
 469         /* Load parameters for i particles */
 470         iq0              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+0]));
 471
 472         /* Start inner kernel loop */
 473         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+7]>=0; jidx+=8)
 474         {
 475
 476             /* Get j neighbor index, and coordinate index */
 477             jnrA             = jjnr[jidx];
 478             jnrB             = jjnr[jidx+1];
 479             jnrC             = jjnr[jidx+2];
 480             jnrD             = jjnr[jidx+3];
 481             jnrE             = jjnr[jidx+4];
 482             jnrF             = jjnr[jidx+5];
 483             jnrG             = jjnr[jidx+6];
 484             jnrH             = jjnr[jidx+7];
 485             j_coord_offsetA  = DIM*jnrA;
 486             j_coord_offsetB  = DIM*jnrB;
 487             j_coord_offsetC  = DIM*jnrC;
 488             j_coord_offsetD  = DIM*jnrD;
 489             j_coord_offsetE  = DIM*jnrE;
 490             j_coord_offsetF  = DIM*jnrF;
 491             j_coord_offsetG  = DIM*jnrG;
 492             j_coord_offsetH  = DIM*jnrH;
 493
 494             /* load j atom coordinates */
 495             gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 496                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 497                                                  x+j_coord_offsetE,x+j_coord_offsetF,
 498                                                  x+j_coord_offsetG,x+j_coord_offsetH,
 499                                                  &jx0,&jy0,&jz0);
 500
 501             /* Calculate displacement vector */
 502             dx00             = _mm256_sub_ps(ix0,jx0);
 503             dy00             = _mm256_sub_ps(iy0,jy0);
 504             dz00             = _mm256_sub_ps(iz0,jz0);
 505
 506             /* Calculate squared distance and things based on it */
 507             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
 508
 509             rinv00           = gmx_mm256_invsqrt_ps(rsq00);
 510
 511             rinvsq00         = _mm256_mul_ps(rinv00,rinv00);
 512
 513             /* Load parameters for j particles */
 514             jq0              = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 515                                                                  charge+jnrC+0,charge+jnrD+0,
 516                                                                  charge+jnrE+0,charge+jnrF+0,
 517                                                                  charge+jnrG+0,charge+jnrH+0);
 518
 519             /**************************
 520              * CALCULATE INTERACTIONS *
 521              **************************/
 522
 523             /* Compute parameters for interactions between i and j atoms */
 524             qq00             = _mm256_mul_ps(iq0,jq0);
 525
 526             /* COULOMB ELECTROSTATICS */
 527             velec            = _mm256_mul_ps(qq00,rinv00);
 528             felec            = _mm256_mul_ps(velec,rinvsq00);
 529
 530             fscal            = felec;
 531
 532             /* Calculate temporary vectorial force */
 533             tx               = _mm256_mul_ps(fscal,dx00);
 534             ty               = _mm256_mul_ps(fscal,dy00);
 535             tz               = _mm256_mul_ps(fscal,dz00);
 536
 537             /* Update vectorial force */
 538             fix0             = _mm256_add_ps(fix0,tx);
 539             fiy0             = _mm256_add_ps(fiy0,ty);
 540             fiz0             = _mm256_add_ps(fiz0,tz);
 541
 542             fjptrA             = f+j_coord_offsetA;
 543             fjptrB             = f+j_coord_offsetB;
 544             fjptrC             = f+j_coord_offsetC;
 545             fjptrD             = f+j_coord_offsetD;
 546             fjptrE             = f+j_coord_offsetE;
 547             fjptrF             = f+j_coord_offsetF;
 548             fjptrG             = f+j_coord_offsetG;
 549             fjptrH             = f+j_coord_offsetH;
 550             gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,tx,ty,tz);
 551
 552             /* Inner loop uses 26 flops */
 553         }
 554
 555         if(jidx<j_index_end)
 556         {
 557
 558             /* Get j neighbor index, and coordinate index */
 559             jnrlistA         = jjnr[jidx];
 560             jnrlistB         = jjnr[jidx+1];
 561             jnrlistC         = jjnr[jidx+2];
 562             jnrlistD         = jjnr[jidx+3];
 563             jnrlistE         = jjnr[jidx+4];
 564             jnrlistF         = jjnr[jidx+5];
 565             jnrlistG         = jjnr[jidx+6];
 566             jnrlistH         = jjnr[jidx+7];
 567             /* Sign of each element will be negative for non-real atoms.
 568              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 569              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 570              */
 571             dummy_mask = gmx_mm256_set_m128(gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx+4)),_mm_setzero_si128())),
 572                                             gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128())));
 573
 574             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 575             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 576             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 577             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 578             jnrE       = (jnrlistE>=0) ? jnrlistE : 0;
 579             jnrF       = (jnrlistF>=0) ? jnrlistF : 0;
 580             jnrG       = (jnrlistG>=0) ? jnrlistG : 0;
 581             jnrH       = (jnrlistH>=0) ? jnrlistH : 0;
 582             j_coord_offsetA  = DIM*jnrA;
 583             j_coord_offsetB  = DIM*jnrB;
 584             j_coord_offsetC  = DIM*jnrC;
 585             j_coord_offsetD  = DIM*jnrD;
 586             j_coord_offsetE  = DIM*jnrE;
 587             j_coord_offsetF  = DIM*jnrF;
 588             j_coord_offsetG  = DIM*jnrG;
 589             j_coord_offsetH  = DIM*jnrH;
 590
 591             /* load j atom coordinates */
 592             gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 593                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 594                                                  x+j_coord_offsetE,x+j_coord_offsetF,
 595                                                  x+j_coord_offsetG,x+j_coord_offsetH,
 596                                                  &jx0,&jy0,&jz0);
 597
 598             /* Calculate displacement vector */
 599             dx00             = _mm256_sub_ps(ix0,jx0);
 600             dy00             = _mm256_sub_ps(iy0,jy0);
 601             dz00             = _mm256_sub_ps(iz0,jz0);
 602
 603             /* Calculate squared distance and things based on it */
 604             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
 605
 606             rinv00           = gmx_mm256_invsqrt_ps(rsq00);
 607
 608             rinvsq00         = _mm256_mul_ps(rinv00,rinv00);
 609
 610             /* Load parameters for j particles */
 611             jq0              = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 612                                                                  charge+jnrC+0,charge+jnrD+0,
 613                                                                  charge+jnrE+0,charge+jnrF+0,
 614                                                                  charge+jnrG+0,charge+jnrH+0);
 615
 616             /**************************
 617              * CALCULATE INTERACTIONS *
 618              **************************/
 619
 620             /* Compute parameters for interactions between i and j atoms */
 621             qq00             = _mm256_mul_ps(iq0,jq0);
 622
 623             /* COULOMB ELECTROSTATICS */
 624             velec            = _mm256_mul_ps(qq00,rinv00);
 625             felec            = _mm256_mul_ps(velec,rinvsq00);
 626
 627             fscal            = felec;
 628
 629             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
 630
 631             /* Calculate temporary vectorial force */
 632             tx               = _mm256_mul_ps(fscal,dx00);
 633             ty               = _mm256_mul_ps(fscal,dy00);
 634             tz               = _mm256_mul_ps(fscal,dz00);
 635
 636             /* Update vectorial force */
 637             fix0             = _mm256_add_ps(fix0,tx);
 638             fiy0             = _mm256_add_ps(fiy0,ty);
 639             fiz0             = _mm256_add_ps(fiz0,tz);
 640
 641             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 642             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 643             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 644             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 645             fjptrE             = (jnrlistE>=0) ? f+j_coord_offsetE : scratch;
 646             fjptrF             = (jnrlistF>=0) ? f+j_coord_offsetF : scratch;
 647             fjptrG             = (jnrlistG>=0) ? f+j_coord_offsetG : scratch;
 648             fjptrH             = (jnrlistH>=0) ? f+j_coord_offsetH : scratch;
 649             gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,tx,ty,tz);
 650
 651             /* Inner loop uses 26 flops */
 652         }
 653
 654         /* End of innermost loop */
 655
 656         gmx_mm256_update_iforce_1atom_swizzle_ps(fix0,fiy0,fiz0,
 657                                                  f+i_coord_offset,fshift+i_shift_offset);
 658
 659         /* Increment number of inner iterations */
 660         inneriter                  += j_index_end - j_index_start;
 661
 662         /* Outer loop uses 7 flops */
 663     }
 664
 665     /* Increment number of outer iterations */
 666     outeriter        += nri;
 667
 668     /* Update outer/inner flops */
 669
 670     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_F,outeriter*7 + inneriter*26);
 671 }