src/gromacs/gmxlib/nonbonded/nb_kernel_avx_256_single/nb_kernel_ElecEw_VdwNone_GeomW4P1_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 "types/simple.h"
  44 #include "gromacs/math/vec.h"
  45 #include "nrnb.h"
  46
  47 #include "gromacs/simd/math_x86_avx_256_single.h"
  48 #include "kernelutil_x86_avx_256_single.h"
  49
  50 /*
  51  * Gromacs nonbonded kernel:   nb_kernel_ElecEw_VdwNone_GeomW4P1_VF_avx_256_single
  52  * Electrostatics interaction: Ewald
  53  * VdW interaction:            None
  54  * Geometry:                   Water4-Particle
  55  * Calculate force/pot:        PotentialAndForce
  56  */
  57 void
  58 nb_kernel_ElecEw_VdwNone_GeomW4P1_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 *           vdwioffsetptr1;
  87     __m256           ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
  88     real *           vdwioffsetptr2;
  89     __m256           ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
  90     real *           vdwioffsetptr3;
  91     __m256           ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
  92     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D,vdwjidx0E,vdwjidx0F,vdwjidx0G,vdwjidx0H;
  93     __m256           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  94     __m256           dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
  95     __m256           dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
  96     __m256           dx30,dy30,dz30,rsq30,rinv30,rinvsq30,r30,qq30,c6_30,c12_30;
  97     __m256           velec,felec,velecsum,facel,crf,krf,krf2;
  98     real             *charge;
  99     __m256i          ewitab;
 100     __m128i          ewitab_lo,ewitab_hi;
 101     __m256           ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
 102     __m256           beta,beta2,beta3,zeta2,pmecorrF,pmecorrV,rinv3;
 103     real             *ewtab;
 104     __m256           dummy_mask,cutoff_mask;
 105     __m256           signbit = _mm256_castsi256_ps( _mm256_set1_epi32(0x80000000) );
 106     __m256           one     = _mm256_set1_ps(1.0);
 107     __m256           two     = _mm256_set1_ps(2.0);
 108     x                = xx[0];
 109     f                = ff[0];
 110
 111     nri              = nlist->nri;
 112     iinr             = nlist->iinr;
 113     jindex           = nlist->jindex;
 114     jjnr             = nlist->jjnr;
 115     shiftidx         = nlist->shift;
 116     gid              = nlist->gid;
 117     shiftvec         = fr->shift_vec[0];
 118     fshift           = fr->fshift[0];
 119     facel            = _mm256_set1_ps(fr->epsfac);
 120     charge           = mdatoms->chargeA;
 121
 122     sh_ewald         = _mm256_set1_ps(fr->ic->sh_ewald);
 123     beta             = _mm256_set1_ps(fr->ic->ewaldcoeff_q);
 124     beta2            = _mm256_mul_ps(beta,beta);
 125     beta3            = _mm256_mul_ps(beta,beta2);
 126
 127     ewtab            = fr->ic->tabq_coul_FDV0;
 128     ewtabscale       = _mm256_set1_ps(fr->ic->tabq_scale);
 129     ewtabhalfspace   = _mm256_set1_ps(0.5/fr->ic->tabq_scale);
 130
 131     /* Setup water-specific parameters */
 132     inr              = nlist->iinr[0];
 133     iq1              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+1]));
 134     iq2              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+2]));
 135     iq3              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+3]));
 136
 137     /* Avoid stupid compiler warnings */
 138     jnrA = jnrB = jnrC = jnrD = jnrE = jnrF = jnrG = jnrH = 0;
 139     j_coord_offsetA = 0;
 140     j_coord_offsetB = 0;
 141     j_coord_offsetC = 0;
 142     j_coord_offsetD = 0;
 143     j_coord_offsetE = 0;
 144     j_coord_offsetF = 0;
 145     j_coord_offsetG = 0;
 146     j_coord_offsetH = 0;
 147
 148     outeriter        = 0;
 149     inneriter        = 0;
 150
 151     for(iidx=0;iidx<4*DIM;iidx++)
 152     {
 153         scratch[iidx] = 0.0;
 154     }
 155
 156     /* Start outer loop over neighborlists */
 157     for(iidx=0; iidx<nri; iidx++)
 158     {
 159         /* Load shift vector for this list */
 160         i_shift_offset   = DIM*shiftidx[iidx];
 161
 162         /* Load limits for loop over neighbors */
 163         j_index_start    = jindex[iidx];
 164         j_index_end      = jindex[iidx+1];
 165
 166         /* Get outer coordinate index */
 167         inr              = iinr[iidx];
 168         i_coord_offset   = DIM*inr;
 169
 170         /* Load i particle coords and add shift vector */
 171         gmx_mm256_load_shift_and_3rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset+DIM,
 172                                                     &ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
 173
 174         fix1             = _mm256_setzero_ps();
 175         fiy1             = _mm256_setzero_ps();
 176         fiz1             = _mm256_setzero_ps();
 177         fix2             = _mm256_setzero_ps();
 178         fiy2             = _mm256_setzero_ps();
 179         fiz2             = _mm256_setzero_ps();
 180         fix3             = _mm256_setzero_ps();
 181         fiy3             = _mm256_setzero_ps();
 182         fiz3             = _mm256_setzero_ps();
 183
 184         /* Reset potential sums */
 185         velecsum         = _mm256_setzero_ps();
 186
 187         /* Start inner kernel loop */
 188         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+7]>=0; jidx+=8)
 189         {
 190
 191             /* Get j neighbor index, and coordinate index */
 192             jnrA             = jjnr[jidx];
 193             jnrB             = jjnr[jidx+1];
 194             jnrC             = jjnr[jidx+2];
 195             jnrD             = jjnr[jidx+3];
 196             jnrE             = jjnr[jidx+4];
 197             jnrF             = jjnr[jidx+5];
 198             jnrG             = jjnr[jidx+6];
 199             jnrH             = jjnr[jidx+7];
 200             j_coord_offsetA  = DIM*jnrA;
 201             j_coord_offsetB  = DIM*jnrB;
 202             j_coord_offsetC  = DIM*jnrC;
 203             j_coord_offsetD  = DIM*jnrD;
 204             j_coord_offsetE  = DIM*jnrE;
 205             j_coord_offsetF  = DIM*jnrF;
 206             j_coord_offsetG  = DIM*jnrG;
 207             j_coord_offsetH  = DIM*jnrH;
 208
 209             /* load j atom coordinates */
 210             gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 211                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 212                                                  x+j_coord_offsetE,x+j_coord_offsetF,
 213                                                  x+j_coord_offsetG,x+j_coord_offsetH,
 214                                                  &jx0,&jy0,&jz0);
 215
 216             /* Calculate displacement vector */
 217             dx10             = _mm256_sub_ps(ix1,jx0);
 218             dy10             = _mm256_sub_ps(iy1,jy0);
 219             dz10             = _mm256_sub_ps(iz1,jz0);
 220             dx20             = _mm256_sub_ps(ix2,jx0);
 221             dy20             = _mm256_sub_ps(iy2,jy0);
 222             dz20             = _mm256_sub_ps(iz2,jz0);
 223             dx30             = _mm256_sub_ps(ix3,jx0);
 224             dy30             = _mm256_sub_ps(iy3,jy0);
 225             dz30             = _mm256_sub_ps(iz3,jz0);
 226
 227             /* Calculate squared distance and things based on it */
 228             rsq10            = gmx_mm256_calc_rsq_ps(dx10,dy10,dz10);
 229             rsq20            = gmx_mm256_calc_rsq_ps(dx20,dy20,dz20);
 230             rsq30            = gmx_mm256_calc_rsq_ps(dx30,dy30,dz30);
 231
 232             rinv10           = gmx_mm256_invsqrt_ps(rsq10);
 233             rinv20           = gmx_mm256_invsqrt_ps(rsq20);
 234             rinv30           = gmx_mm256_invsqrt_ps(rsq30);
 235
 236             rinvsq10         = _mm256_mul_ps(rinv10,rinv10);
 237             rinvsq20         = _mm256_mul_ps(rinv20,rinv20);
 238             rinvsq30         = _mm256_mul_ps(rinv30,rinv30);
 239
 240             /* Load parameters for j particles */
 241             jq0              = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 242                                                                  charge+jnrC+0,charge+jnrD+0,
 243                                                                  charge+jnrE+0,charge+jnrF+0,
 244                                                                  charge+jnrG+0,charge+jnrH+0);
 245
 246             fjx0             = _mm256_setzero_ps();
 247             fjy0             = _mm256_setzero_ps();
 248             fjz0             = _mm256_setzero_ps();
 249
 250             /**************************
 251              * CALCULATE INTERACTIONS *
 252              **************************/
 253
 254             r10              = _mm256_mul_ps(rsq10,rinv10);
 255
 256             /* Compute parameters for interactions between i and j atoms */
 257             qq10             = _mm256_mul_ps(iq1,jq0);
 258
 259             /* EWALD ELECTROSTATICS */
 260
 261             /* Analytical PME correction */
 262             zeta2            = _mm256_mul_ps(beta2,rsq10);
 263             rinv3            = _mm256_mul_ps(rinvsq10,rinv10);
 264             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
 265             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
 266             felec            = _mm256_mul_ps(qq10,felec);
 267             pmecorrV         = gmx_mm256_pmecorrV_ps(zeta2);
 268             pmecorrV         = _mm256_mul_ps(pmecorrV,beta);
 269             velec            = _mm256_sub_ps(rinv10,pmecorrV);
 270             velec            = _mm256_mul_ps(qq10,velec);
 271
 272             /* Update potential sum for this i atom from the interaction with this j atom. */
 273             velecsum         = _mm256_add_ps(velecsum,velec);
 274
 275             fscal            = felec;
 276
 277             /* Calculate temporary vectorial force */
 278             tx               = _mm256_mul_ps(fscal,dx10);
 279             ty               = _mm256_mul_ps(fscal,dy10);
 280             tz               = _mm256_mul_ps(fscal,dz10);
 281
 282             /* Update vectorial force */
 283             fix1             = _mm256_add_ps(fix1,tx);
 284             fiy1             = _mm256_add_ps(fiy1,ty);
 285             fiz1             = _mm256_add_ps(fiz1,tz);
 286
 287             fjx0             = _mm256_add_ps(fjx0,tx);
 288             fjy0             = _mm256_add_ps(fjy0,ty);
 289             fjz0             = _mm256_add_ps(fjz0,tz);
 290
 291             /**************************
 292              * CALCULATE INTERACTIONS *
 293              **************************/
 294
 295             r20              = _mm256_mul_ps(rsq20,rinv20);
 296
 297             /* Compute parameters for interactions between i and j atoms */
 298             qq20             = _mm256_mul_ps(iq2,jq0);
 299
 300             /* EWALD ELECTROSTATICS */
 301
 302             /* Analytical PME correction */
 303             zeta2            = _mm256_mul_ps(beta2,rsq20);
 304             rinv3            = _mm256_mul_ps(rinvsq20,rinv20);
 305             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
 306             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
 307             felec            = _mm256_mul_ps(qq20,felec);
 308             pmecorrV         = gmx_mm256_pmecorrV_ps(zeta2);
 309             pmecorrV         = _mm256_mul_ps(pmecorrV,beta);
 310             velec            = _mm256_sub_ps(rinv20,pmecorrV);
 311             velec            = _mm256_mul_ps(qq20,velec);
 312
 313             /* Update potential sum for this i atom from the interaction with this j atom. */
 314             velecsum         = _mm256_add_ps(velecsum,velec);
 315
 316             fscal            = felec;
 317
 318             /* Calculate temporary vectorial force */
 319             tx               = _mm256_mul_ps(fscal,dx20);
 320             ty               = _mm256_mul_ps(fscal,dy20);
 321             tz               = _mm256_mul_ps(fscal,dz20);
 322
 323             /* Update vectorial force */
 324             fix2             = _mm256_add_ps(fix2,tx);
 325             fiy2             = _mm256_add_ps(fiy2,ty);
 326             fiz2             = _mm256_add_ps(fiz2,tz);
 327
 328             fjx0             = _mm256_add_ps(fjx0,tx);
 329             fjy0             = _mm256_add_ps(fjy0,ty);
 330             fjz0             = _mm256_add_ps(fjz0,tz);
 331
 332             /**************************
 333              * CALCULATE INTERACTIONS *
 334              **************************/
 335
 336             r30              = _mm256_mul_ps(rsq30,rinv30);
 337
 338             /* Compute parameters for interactions between i and j atoms */
 339             qq30             = _mm256_mul_ps(iq3,jq0);
 340
 341             /* EWALD ELECTROSTATICS */
 342
 343             /* Analytical PME correction */
 344             zeta2            = _mm256_mul_ps(beta2,rsq30);
 345             rinv3            = _mm256_mul_ps(rinvsq30,rinv30);
 346             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
 347             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
 348             felec            = _mm256_mul_ps(qq30,felec);
 349             pmecorrV         = gmx_mm256_pmecorrV_ps(zeta2);
 350             pmecorrV         = _mm256_mul_ps(pmecorrV,beta);
 351             velec            = _mm256_sub_ps(rinv30,pmecorrV);
 352             velec            = _mm256_mul_ps(qq30,velec);
 353
 354             /* Update potential sum for this i atom from the interaction with this j atom. */
 355             velecsum         = _mm256_add_ps(velecsum,velec);
 356
 357             fscal            = felec;
 358
 359             /* Calculate temporary vectorial force */
 360             tx               = _mm256_mul_ps(fscal,dx30);
 361             ty               = _mm256_mul_ps(fscal,dy30);
 362             tz               = _mm256_mul_ps(fscal,dz30);
 363
 364             /* Update vectorial force */
 365             fix3             = _mm256_add_ps(fix3,tx);
 366             fiy3             = _mm256_add_ps(fiy3,ty);
 367             fiz3             = _mm256_add_ps(fiz3,tz);
 368
 369             fjx0             = _mm256_add_ps(fjx0,tx);
 370             fjy0             = _mm256_add_ps(fjy0,ty);
 371             fjz0             = _mm256_add_ps(fjz0,tz);
 372
 373             fjptrA             = f+j_coord_offsetA;
 374             fjptrB             = f+j_coord_offsetB;
 375             fjptrC             = f+j_coord_offsetC;
 376             fjptrD             = f+j_coord_offsetD;
 377             fjptrE             = f+j_coord_offsetE;
 378             fjptrF             = f+j_coord_offsetF;
 379             fjptrG             = f+j_coord_offsetG;
 380             fjptrH             = f+j_coord_offsetH;
 381
 382             gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,fjx0,fjy0,fjz0);
 383
 384             /* Inner loop uses 255 flops */
 385         }
 386
 387         if(jidx<j_index_end)
 388         {
 389
 390             /* Get j neighbor index, and coordinate index */
 391             jnrlistA         = jjnr[jidx];
 392             jnrlistB         = jjnr[jidx+1];
 393             jnrlistC         = jjnr[jidx+2];
 394             jnrlistD         = jjnr[jidx+3];
 395             jnrlistE         = jjnr[jidx+4];
 396             jnrlistF         = jjnr[jidx+5];
 397             jnrlistG         = jjnr[jidx+6];
 398             jnrlistH         = jjnr[jidx+7];
 399             /* Sign of each element will be negative for non-real atoms.
 400              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 401              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 402              */
 403             dummy_mask = gmx_mm256_set_m128(gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx+4)),_mm_setzero_si128())),
 404                                             gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128())));
 405
 406             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 407             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 408             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 409             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 410             jnrE       = (jnrlistE>=0) ? jnrlistE : 0;
 411             jnrF       = (jnrlistF>=0) ? jnrlistF : 0;
 412             jnrG       = (jnrlistG>=0) ? jnrlistG : 0;
 413             jnrH       = (jnrlistH>=0) ? jnrlistH : 0;
 414             j_coord_offsetA  = DIM*jnrA;
 415             j_coord_offsetB  = DIM*jnrB;
 416             j_coord_offsetC  = DIM*jnrC;
 417             j_coord_offsetD  = DIM*jnrD;
 418             j_coord_offsetE  = DIM*jnrE;
 419             j_coord_offsetF  = DIM*jnrF;
 420             j_coord_offsetG  = DIM*jnrG;
 421             j_coord_offsetH  = DIM*jnrH;
 422
 423             /* load j atom coordinates */
 424             gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 425                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 426                                                  x+j_coord_offsetE,x+j_coord_offsetF,
 427                                                  x+j_coord_offsetG,x+j_coord_offsetH,
 428                                                  &jx0,&jy0,&jz0);
 429
 430             /* Calculate displacement vector */
 431             dx10             = _mm256_sub_ps(ix1,jx0);
 432             dy10             = _mm256_sub_ps(iy1,jy0);
 433             dz10             = _mm256_sub_ps(iz1,jz0);
 434             dx20             = _mm256_sub_ps(ix2,jx0);
 435             dy20             = _mm256_sub_ps(iy2,jy0);
 436             dz20             = _mm256_sub_ps(iz2,jz0);
 437             dx30             = _mm256_sub_ps(ix3,jx0);
 438             dy30             = _mm256_sub_ps(iy3,jy0);
 439             dz30             = _mm256_sub_ps(iz3,jz0);
 440
 441             /* Calculate squared distance and things based on it */
 442             rsq10            = gmx_mm256_calc_rsq_ps(dx10,dy10,dz10);
 443             rsq20            = gmx_mm256_calc_rsq_ps(dx20,dy20,dz20);
 444             rsq30            = gmx_mm256_calc_rsq_ps(dx30,dy30,dz30);
 445
 446             rinv10           = gmx_mm256_invsqrt_ps(rsq10);
 447             rinv20           = gmx_mm256_invsqrt_ps(rsq20);
 448             rinv30           = gmx_mm256_invsqrt_ps(rsq30);
 449
 450             rinvsq10         = _mm256_mul_ps(rinv10,rinv10);
 451             rinvsq20         = _mm256_mul_ps(rinv20,rinv20);
 452             rinvsq30         = _mm256_mul_ps(rinv30,rinv30);
 453
 454             /* Load parameters for j particles */
 455             jq0              = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 456                                                                  charge+jnrC+0,charge+jnrD+0,
 457                                                                  charge+jnrE+0,charge+jnrF+0,
 458                                                                  charge+jnrG+0,charge+jnrH+0);
 459
 460             fjx0             = _mm256_setzero_ps();
 461             fjy0             = _mm256_setzero_ps();
 462             fjz0             = _mm256_setzero_ps();
 463
 464             /**************************
 465              * CALCULATE INTERACTIONS *
 466              **************************/
 467
 468             r10              = _mm256_mul_ps(rsq10,rinv10);
 469             r10              = _mm256_andnot_ps(dummy_mask,r10);
 470
 471             /* Compute parameters for interactions between i and j atoms */
 472             qq10             = _mm256_mul_ps(iq1,jq0);
 473
 474             /* EWALD ELECTROSTATICS */
 475
 476             /* Analytical PME correction */
 477             zeta2            = _mm256_mul_ps(beta2,rsq10);
 478             rinv3            = _mm256_mul_ps(rinvsq10,rinv10);
 479             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
 480             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
 481             felec            = _mm256_mul_ps(qq10,felec);
 482             pmecorrV         = gmx_mm256_pmecorrV_ps(zeta2);
 483             pmecorrV         = _mm256_mul_ps(pmecorrV,beta);
 484             velec            = _mm256_sub_ps(rinv10,pmecorrV);
 485             velec            = _mm256_mul_ps(qq10,velec);
 486
 487             /* Update potential sum for this i atom from the interaction with this j atom. */
 488             velec            = _mm256_andnot_ps(dummy_mask,velec);
 489             velecsum         = _mm256_add_ps(velecsum,velec);
 490
 491             fscal            = felec;
 492
 493             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
 494
 495             /* Calculate temporary vectorial force */
 496             tx               = _mm256_mul_ps(fscal,dx10);
 497             ty               = _mm256_mul_ps(fscal,dy10);
 498             tz               = _mm256_mul_ps(fscal,dz10);
 499
 500             /* Update vectorial force */
 501             fix1             = _mm256_add_ps(fix1,tx);
 502             fiy1             = _mm256_add_ps(fiy1,ty);
 503             fiz1             = _mm256_add_ps(fiz1,tz);
 504
 505             fjx0             = _mm256_add_ps(fjx0,tx);
 506             fjy0             = _mm256_add_ps(fjy0,ty);
 507             fjz0             = _mm256_add_ps(fjz0,tz);
 508
 509             /**************************
 510              * CALCULATE INTERACTIONS *
 511              **************************/
 512
 513             r20              = _mm256_mul_ps(rsq20,rinv20);
 514             r20              = _mm256_andnot_ps(dummy_mask,r20);
 515
 516             /* Compute parameters for interactions between i and j atoms */
 517             qq20             = _mm256_mul_ps(iq2,jq0);
 518
 519             /* EWALD ELECTROSTATICS */
 520
 521             /* Analytical PME correction */
 522             zeta2            = _mm256_mul_ps(beta2,rsq20);
 523             rinv3            = _mm256_mul_ps(rinvsq20,rinv20);
 524             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
 525             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
 526             felec            = _mm256_mul_ps(qq20,felec);
 527             pmecorrV         = gmx_mm256_pmecorrV_ps(zeta2);
 528             pmecorrV         = _mm256_mul_ps(pmecorrV,beta);
 529             velec            = _mm256_sub_ps(rinv20,pmecorrV);
 530             velec            = _mm256_mul_ps(qq20,velec);
 531
 532             /* Update potential sum for this i atom from the interaction with this j atom. */
 533             velec            = _mm256_andnot_ps(dummy_mask,velec);
 534             velecsum         = _mm256_add_ps(velecsum,velec);
 535
 536             fscal            = felec;
 537
 538             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
 539
 540             /* Calculate temporary vectorial force */
 541             tx               = _mm256_mul_ps(fscal,dx20);
 542             ty               = _mm256_mul_ps(fscal,dy20);
 543             tz               = _mm256_mul_ps(fscal,dz20);
 544
 545             /* Update vectorial force */
 546             fix2             = _mm256_add_ps(fix2,tx);
 547             fiy2             = _mm256_add_ps(fiy2,ty);
 548             fiz2             = _mm256_add_ps(fiz2,tz);
 549
 550             fjx0             = _mm256_add_ps(fjx0,tx);
 551             fjy0             = _mm256_add_ps(fjy0,ty);
 552             fjz0             = _mm256_add_ps(fjz0,tz);
 553
 554             /**************************
 555              * CALCULATE INTERACTIONS *
 556              **************************/
 557
 558             r30              = _mm256_mul_ps(rsq30,rinv30);
 559             r30              = _mm256_andnot_ps(dummy_mask,r30);
 560
 561             /* Compute parameters for interactions between i and j atoms */
 562             qq30             = _mm256_mul_ps(iq3,jq0);
 563
 564             /* EWALD ELECTROSTATICS */
 565
 566             /* Analytical PME correction */
 567             zeta2            = _mm256_mul_ps(beta2,rsq30);
 568             rinv3            = _mm256_mul_ps(rinvsq30,rinv30);
 569             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
 570             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
 571             felec            = _mm256_mul_ps(qq30,felec);
 572             pmecorrV         = gmx_mm256_pmecorrV_ps(zeta2);
 573             pmecorrV         = _mm256_mul_ps(pmecorrV,beta);
 574             velec            = _mm256_sub_ps(rinv30,pmecorrV);
 575             velec            = _mm256_mul_ps(qq30,velec);
 576
 577             /* Update potential sum for this i atom from the interaction with this j atom. */
 578             velec            = _mm256_andnot_ps(dummy_mask,velec);
 579             velecsum         = _mm256_add_ps(velecsum,velec);
 580
 581             fscal            = felec;
 582
 583             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
 584
 585             /* Calculate temporary vectorial force */
 586             tx               = _mm256_mul_ps(fscal,dx30);
 587             ty               = _mm256_mul_ps(fscal,dy30);
 588             tz               = _mm256_mul_ps(fscal,dz30);
 589
 590             /* Update vectorial force */
 591             fix3             = _mm256_add_ps(fix3,tx);
 592             fiy3             = _mm256_add_ps(fiy3,ty);
 593             fiz3             = _mm256_add_ps(fiz3,tz);
 594
 595             fjx0             = _mm256_add_ps(fjx0,tx);
 596             fjy0             = _mm256_add_ps(fjy0,ty);
 597             fjz0             = _mm256_add_ps(fjz0,tz);
 598
 599             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 600             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 601             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 602             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 603             fjptrE             = (jnrlistE>=0) ? f+j_coord_offsetE : scratch;
 604             fjptrF             = (jnrlistF>=0) ? f+j_coord_offsetF : scratch;
 605             fjptrG             = (jnrlistG>=0) ? f+j_coord_offsetG : scratch;
 606             fjptrH             = (jnrlistH>=0) ? f+j_coord_offsetH : scratch;
 607
 608             gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,fjx0,fjy0,fjz0);
 609
 610             /* Inner loop uses 258 flops */
 611         }
 612
 613         /* End of innermost loop */
 614
 615         gmx_mm256_update_iforce_3atom_swizzle_ps(fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
 616                                                  f+i_coord_offset+DIM,fshift+i_shift_offset);
 617
 618         ggid                        = gid[iidx];
 619         /* Update potential energies */
 620         gmx_mm256_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid);
 621
 622         /* Increment number of inner iterations */
 623         inneriter                  += j_index_end - j_index_start;
 624
 625         /* Outer loop uses 19 flops */
 626     }
 627
 628     /* Increment number of outer iterations */
 629     outeriter        += nri;
 630
 631     /* Update outer/inner flops */
 632
 633     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_W4_VF,outeriter*19 + inneriter*258);
 634 }
 635 /*
 636  * Gromacs nonbonded kernel:   nb_kernel_ElecEw_VdwNone_GeomW4P1_F_avx_256_single
 637  * Electrostatics interaction: Ewald
 638  * VdW interaction:            None
 639  * Geometry:                   Water4-Particle
 640  * Calculate force/pot:        Force
 641  */
 642 void
 643 nb_kernel_ElecEw_VdwNone_GeomW4P1_F_avx_256_single
 644                     (t_nblist                    * gmx_restrict       nlist,
 645                      rvec                        * gmx_restrict          xx,
 646                      rvec                        * gmx_restrict          ff,
 647                      t_forcerec                  * gmx_restrict          fr,
 648                      t_mdatoms                   * gmx_restrict     mdatoms,
 649                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 650                      t_nrnb                      * gmx_restrict        nrnb)
 651 {
 652     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 653      * just 0 for non-waters.
 654      * Suffixes A,B,C,D,E,F,G,H refer to j loop unrolling done with AVX, e.g. for the eight different
 655      * jnr indices corresponding to data put in the four positions in the SIMD register.
 656      */
 657     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 658     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 659     int              jnrA,jnrB,jnrC,jnrD;
 660     int              jnrE,jnrF,jnrG,jnrH;
 661     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
 662     int              jnrlistE,jnrlistF,jnrlistG,jnrlistH;
 663     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 664     int              j_coord_offsetE,j_coord_offsetF,j_coord_offsetG,j_coord_offsetH;
 665     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 666     real             rcutoff_scalar;
 667     real             *shiftvec,*fshift,*x,*f;
 668     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD,*fjptrE,*fjptrF,*fjptrG,*fjptrH;
 669     real             scratch[4*DIM];
 670     __m256           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 671     real *           vdwioffsetptr1;
 672     __m256           ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
 673     real *           vdwioffsetptr2;
 674     __m256           ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
 675     real *           vdwioffsetptr3;
 676     __m256           ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
 677     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D,vdwjidx0E,vdwjidx0F,vdwjidx0G,vdwjidx0H;
 678     __m256           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 679     __m256           dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
 680     __m256           dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
 681     __m256           dx30,dy30,dz30,rsq30,rinv30,rinvsq30,r30,qq30,c6_30,c12_30;
 682     __m256           velec,felec,velecsum,facel,crf,krf,krf2;
 683     real             *charge;
 684     __m256i          ewitab;
 685     __m128i          ewitab_lo,ewitab_hi;
 686     __m256           ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
 687     __m256           beta,beta2,beta3,zeta2,pmecorrF,pmecorrV,rinv3;
 688     real             *ewtab;
 689     __m256           dummy_mask,cutoff_mask;
 690     __m256           signbit = _mm256_castsi256_ps( _mm256_set1_epi32(0x80000000) );
 691     __m256           one     = _mm256_set1_ps(1.0);
 692     __m256           two     = _mm256_set1_ps(2.0);
 693     x                = xx[0];
 694     f                = ff[0];
 695
 696     nri              = nlist->nri;
 697     iinr             = nlist->iinr;
 698     jindex           = nlist->jindex;
 699     jjnr             = nlist->jjnr;
 700     shiftidx         = nlist->shift;
 701     gid              = nlist->gid;
 702     shiftvec         = fr->shift_vec[0];
 703     fshift           = fr->fshift[0];
 704     facel            = _mm256_set1_ps(fr->epsfac);
 705     charge           = mdatoms->chargeA;
 706
 707     sh_ewald         = _mm256_set1_ps(fr->ic->sh_ewald);
 708     beta             = _mm256_set1_ps(fr->ic->ewaldcoeff_q);
 709     beta2            = _mm256_mul_ps(beta,beta);
 710     beta3            = _mm256_mul_ps(beta,beta2);
 711
 712     ewtab            = fr->ic->tabq_coul_F;
 713     ewtabscale       = _mm256_set1_ps(fr->ic->tabq_scale);
 714     ewtabhalfspace   = _mm256_set1_ps(0.5/fr->ic->tabq_scale);
 715
 716     /* Setup water-specific parameters */
 717     inr              = nlist->iinr[0];
 718     iq1              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+1]));
 719     iq2              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+2]));
 720     iq3              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+3]));
 721
 722     /* Avoid stupid compiler warnings */
 723     jnrA = jnrB = jnrC = jnrD = jnrE = jnrF = jnrG = jnrH = 0;
 724     j_coord_offsetA = 0;
 725     j_coord_offsetB = 0;
 726     j_coord_offsetC = 0;
 727     j_coord_offsetD = 0;
 728     j_coord_offsetE = 0;
 729     j_coord_offsetF = 0;
 730     j_coord_offsetG = 0;
 731     j_coord_offsetH = 0;
 732
 733     outeriter        = 0;
 734     inneriter        = 0;
 735
 736     for(iidx=0;iidx<4*DIM;iidx++)
 737     {
 738         scratch[iidx] = 0.0;
 739     }
 740
 741     /* Start outer loop over neighborlists */
 742     for(iidx=0; iidx<nri; iidx++)
 743     {
 744         /* Load shift vector for this list */
 745         i_shift_offset   = DIM*shiftidx[iidx];
 746
 747         /* Load limits for loop over neighbors */
 748         j_index_start    = jindex[iidx];
 749         j_index_end      = jindex[iidx+1];
 750
 751         /* Get outer coordinate index */
 752         inr              = iinr[iidx];
 753         i_coord_offset   = DIM*inr;
 754
 755         /* Load i particle coords and add shift vector */
 756         gmx_mm256_load_shift_and_3rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset+DIM,
 757                                                     &ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
 758
 759         fix1             = _mm256_setzero_ps();
 760         fiy1             = _mm256_setzero_ps();
 761         fiz1             = _mm256_setzero_ps();
 762         fix2             = _mm256_setzero_ps();
 763         fiy2             = _mm256_setzero_ps();
 764         fiz2             = _mm256_setzero_ps();
 765         fix3             = _mm256_setzero_ps();
 766         fiy3             = _mm256_setzero_ps();
 767         fiz3             = _mm256_setzero_ps();
 768
 769         /* Start inner kernel loop */
 770         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+7]>=0; jidx+=8)
 771         {
 772
 773             /* Get j neighbor index, and coordinate index */
 774             jnrA             = jjnr[jidx];
 775             jnrB             = jjnr[jidx+1];
 776             jnrC             = jjnr[jidx+2];
 777             jnrD             = jjnr[jidx+3];
 778             jnrE             = jjnr[jidx+4];
 779             jnrF             = jjnr[jidx+5];
 780             jnrG             = jjnr[jidx+6];
 781             jnrH             = jjnr[jidx+7];
 782             j_coord_offsetA  = DIM*jnrA;
 783             j_coord_offsetB  = DIM*jnrB;
 784             j_coord_offsetC  = DIM*jnrC;
 785             j_coord_offsetD  = DIM*jnrD;
 786             j_coord_offsetE  = DIM*jnrE;
 787             j_coord_offsetF  = DIM*jnrF;
 788             j_coord_offsetG  = DIM*jnrG;
 789             j_coord_offsetH  = DIM*jnrH;
 790
 791             /* load j atom coordinates */
 792             gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 793                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 794                                                  x+j_coord_offsetE,x+j_coord_offsetF,
 795                                                  x+j_coord_offsetG,x+j_coord_offsetH,
 796                                                  &jx0,&jy0,&jz0);
 797
 798             /* Calculate displacement vector */
 799             dx10             = _mm256_sub_ps(ix1,jx0);
 800             dy10             = _mm256_sub_ps(iy1,jy0);
 801             dz10             = _mm256_sub_ps(iz1,jz0);
 802             dx20             = _mm256_sub_ps(ix2,jx0);
 803             dy20             = _mm256_sub_ps(iy2,jy0);
 804             dz20             = _mm256_sub_ps(iz2,jz0);
 805             dx30             = _mm256_sub_ps(ix3,jx0);
 806             dy30             = _mm256_sub_ps(iy3,jy0);
 807             dz30             = _mm256_sub_ps(iz3,jz0);
 808
 809             /* Calculate squared distance and things based on it */
 810             rsq10            = gmx_mm256_calc_rsq_ps(dx10,dy10,dz10);
 811             rsq20            = gmx_mm256_calc_rsq_ps(dx20,dy20,dz20);
 812             rsq30            = gmx_mm256_calc_rsq_ps(dx30,dy30,dz30);
 813
 814             rinv10           = gmx_mm256_invsqrt_ps(rsq10);
 815             rinv20           = gmx_mm256_invsqrt_ps(rsq20);
 816             rinv30           = gmx_mm256_invsqrt_ps(rsq30);
 817
 818             rinvsq10         = _mm256_mul_ps(rinv10,rinv10);
 819             rinvsq20         = _mm256_mul_ps(rinv20,rinv20);
 820             rinvsq30         = _mm256_mul_ps(rinv30,rinv30);
 821
 822             /* Load parameters for j particles */
 823             jq0              = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 824                                                                  charge+jnrC+0,charge+jnrD+0,
 825                                                                  charge+jnrE+0,charge+jnrF+0,
 826                                                                  charge+jnrG+0,charge+jnrH+0);
 827
 828             fjx0             = _mm256_setzero_ps();
 829             fjy0             = _mm256_setzero_ps();
 830             fjz0             = _mm256_setzero_ps();
 831
 832             /**************************
 833              * CALCULATE INTERACTIONS *
 834              **************************/
 835
 836             r10              = _mm256_mul_ps(rsq10,rinv10);
 837
 838             /* Compute parameters for interactions between i and j atoms */
 839             qq10             = _mm256_mul_ps(iq1,jq0);
 840
 841             /* EWALD ELECTROSTATICS */
 842
 843             /* Analytical PME correction */
 844             zeta2            = _mm256_mul_ps(beta2,rsq10);
 845             rinv3            = _mm256_mul_ps(rinvsq10,rinv10);
 846             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
 847             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
 848             felec            = _mm256_mul_ps(qq10,felec);
 849
 850             fscal            = felec;
 851
 852             /* Calculate temporary vectorial force */
 853             tx               = _mm256_mul_ps(fscal,dx10);
 854             ty               = _mm256_mul_ps(fscal,dy10);
 855             tz               = _mm256_mul_ps(fscal,dz10);
 856
 857             /* Update vectorial force */
 858             fix1             = _mm256_add_ps(fix1,tx);
 859             fiy1             = _mm256_add_ps(fiy1,ty);
 860             fiz1             = _mm256_add_ps(fiz1,tz);
 861
 862             fjx0             = _mm256_add_ps(fjx0,tx);
 863             fjy0             = _mm256_add_ps(fjy0,ty);
 864             fjz0             = _mm256_add_ps(fjz0,tz);
 865
 866             /**************************
 867              * CALCULATE INTERACTIONS *
 868              **************************/
 869
 870             r20              = _mm256_mul_ps(rsq20,rinv20);
 871
 872             /* Compute parameters for interactions between i and j atoms */
 873             qq20             = _mm256_mul_ps(iq2,jq0);
 874
 875             /* EWALD ELECTROSTATICS */
 876
 877             /* Analytical PME correction */
 878             zeta2            = _mm256_mul_ps(beta2,rsq20);
 879             rinv3            = _mm256_mul_ps(rinvsq20,rinv20);
 880             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
 881             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
 882             felec            = _mm256_mul_ps(qq20,felec);
 883
 884             fscal            = felec;
 885
 886             /* Calculate temporary vectorial force */
 887             tx               = _mm256_mul_ps(fscal,dx20);
 888             ty               = _mm256_mul_ps(fscal,dy20);
 889             tz               = _mm256_mul_ps(fscal,dz20);
 890
 891             /* Update vectorial force */
 892             fix2             = _mm256_add_ps(fix2,tx);
 893             fiy2             = _mm256_add_ps(fiy2,ty);
 894             fiz2             = _mm256_add_ps(fiz2,tz);
 895
 896             fjx0             = _mm256_add_ps(fjx0,tx);
 897             fjy0             = _mm256_add_ps(fjy0,ty);
 898             fjz0             = _mm256_add_ps(fjz0,tz);
 899
 900             /**************************
 901              * CALCULATE INTERACTIONS *
 902              **************************/
 903
 904             r30              = _mm256_mul_ps(rsq30,rinv30);
 905
 906             /* Compute parameters for interactions between i and j atoms */
 907             qq30             = _mm256_mul_ps(iq3,jq0);
 908
 909             /* EWALD ELECTROSTATICS */
 910
 911             /* Analytical PME correction */
 912             zeta2            = _mm256_mul_ps(beta2,rsq30);
 913             rinv3            = _mm256_mul_ps(rinvsq30,rinv30);
 914             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
 915             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
 916             felec            = _mm256_mul_ps(qq30,felec);
 917
 918             fscal            = felec;
 919
 920             /* Calculate temporary vectorial force */
 921             tx               = _mm256_mul_ps(fscal,dx30);
 922             ty               = _mm256_mul_ps(fscal,dy30);
 923             tz               = _mm256_mul_ps(fscal,dz30);
 924
 925             /* Update vectorial force */
 926             fix3             = _mm256_add_ps(fix3,tx);
 927             fiy3             = _mm256_add_ps(fiy3,ty);
 928             fiz3             = _mm256_add_ps(fiz3,tz);
 929
 930             fjx0             = _mm256_add_ps(fjx0,tx);
 931             fjy0             = _mm256_add_ps(fjy0,ty);
 932             fjz0             = _mm256_add_ps(fjz0,tz);
 933
 934             fjptrA             = f+j_coord_offsetA;
 935             fjptrB             = f+j_coord_offsetB;
 936             fjptrC             = f+j_coord_offsetC;
 937             fjptrD             = f+j_coord_offsetD;
 938             fjptrE             = f+j_coord_offsetE;
 939             fjptrF             = f+j_coord_offsetF;
 940             fjptrG             = f+j_coord_offsetG;
 941             fjptrH             = f+j_coord_offsetH;
 942
 943             gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,fjx0,fjy0,fjz0);
 944
 945             /* Inner loop uses 171 flops */
 946         }
 947
 948         if(jidx<j_index_end)
 949         {
 950
 951             /* Get j neighbor index, and coordinate index */
 952             jnrlistA         = jjnr[jidx];
 953             jnrlistB         = jjnr[jidx+1];
 954             jnrlistC         = jjnr[jidx+2];
 955             jnrlistD         = jjnr[jidx+3];
 956             jnrlistE         = jjnr[jidx+4];
 957             jnrlistF         = jjnr[jidx+5];
 958             jnrlistG         = jjnr[jidx+6];
 959             jnrlistH         = jjnr[jidx+7];
 960             /* Sign of each element will be negative for non-real atoms.
 961              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 962              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 963              */
 964             dummy_mask = gmx_mm256_set_m128(gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx+4)),_mm_setzero_si128())),
 965                                             gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128())));
 966
 967             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 968             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 969             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 970             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 971             jnrE       = (jnrlistE>=0) ? jnrlistE : 0;
 972             jnrF       = (jnrlistF>=0) ? jnrlistF : 0;
 973             jnrG       = (jnrlistG>=0) ? jnrlistG : 0;
 974             jnrH       = (jnrlistH>=0) ? jnrlistH : 0;
 975             j_coord_offsetA  = DIM*jnrA;
 976             j_coord_offsetB  = DIM*jnrB;
 977             j_coord_offsetC  = DIM*jnrC;
 978             j_coord_offsetD  = DIM*jnrD;
 979             j_coord_offsetE  = DIM*jnrE;
 980             j_coord_offsetF  = DIM*jnrF;
 981             j_coord_offsetG  = DIM*jnrG;
 982             j_coord_offsetH  = DIM*jnrH;
 983
 984             /* load j atom coordinates */
 985             gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 986                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 987                                                  x+j_coord_offsetE,x+j_coord_offsetF,
 988                                                  x+j_coord_offsetG,x+j_coord_offsetH,
 989                                                  &jx0,&jy0,&jz0);
 990
 991             /* Calculate displacement vector */
 992             dx10             = _mm256_sub_ps(ix1,jx0);
 993             dy10             = _mm256_sub_ps(iy1,jy0);
 994             dz10             = _mm256_sub_ps(iz1,jz0);
 995             dx20             = _mm256_sub_ps(ix2,jx0);
 996             dy20             = _mm256_sub_ps(iy2,jy0);
 997             dz20             = _mm256_sub_ps(iz2,jz0);
 998             dx30             = _mm256_sub_ps(ix3,jx0);
 999             dy30             = _mm256_sub_ps(iy3,jy0);
1000             dz30             = _mm256_sub_ps(iz3,jz0);
1001
1002             /* Calculate squared distance and things based on it */
1003             rsq10            = gmx_mm256_calc_rsq_ps(dx10,dy10,dz10);
1004             rsq20            = gmx_mm256_calc_rsq_ps(dx20,dy20,dz20);
1005             rsq30            = gmx_mm256_calc_rsq_ps(dx30,dy30,dz30);
1006
1007             rinv10           = gmx_mm256_invsqrt_ps(rsq10);
1008             rinv20           = gmx_mm256_invsqrt_ps(rsq20);
1009             rinv30           = gmx_mm256_invsqrt_ps(rsq30);
1010
1011             rinvsq10         = _mm256_mul_ps(rinv10,rinv10);
1012             rinvsq20         = _mm256_mul_ps(rinv20,rinv20);
1013             rinvsq30         = _mm256_mul_ps(rinv30,rinv30);
1014
1015             /* Load parameters for j particles */
1016             jq0              = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
1017                                                                  charge+jnrC+0,charge+jnrD+0,
1018                                                                  charge+jnrE+0,charge+jnrF+0,
1019                                                                  charge+jnrG+0,charge+jnrH+0);
1020
1021             fjx0             = _mm256_setzero_ps();
1022             fjy0             = _mm256_setzero_ps();
1023             fjz0             = _mm256_setzero_ps();
1024
1025             /**************************
1026              * CALCULATE INTERACTIONS *
1027              **************************/
1028
1029             r10              = _mm256_mul_ps(rsq10,rinv10);
1030             r10              = _mm256_andnot_ps(dummy_mask,r10);
1031
1032             /* Compute parameters for interactions between i and j atoms */
1033             qq10             = _mm256_mul_ps(iq1,jq0);
1034
1035             /* EWALD ELECTROSTATICS */
1036
1037             /* Analytical PME correction */
1038             zeta2            = _mm256_mul_ps(beta2,rsq10);
1039             rinv3            = _mm256_mul_ps(rinvsq10,rinv10);
1040             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
1041             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
1042             felec            = _mm256_mul_ps(qq10,felec);
1043
1044             fscal            = felec;
1045
1046             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
1047
1048             /* Calculate temporary vectorial force */
1049             tx               = _mm256_mul_ps(fscal,dx10);
1050             ty               = _mm256_mul_ps(fscal,dy10);
1051             tz               = _mm256_mul_ps(fscal,dz10);
1052
1053             /* Update vectorial force */
1054             fix1             = _mm256_add_ps(fix1,tx);
1055             fiy1             = _mm256_add_ps(fiy1,ty);
1056             fiz1             = _mm256_add_ps(fiz1,tz);
1057
1058             fjx0             = _mm256_add_ps(fjx0,tx);
1059             fjy0             = _mm256_add_ps(fjy0,ty);
1060             fjz0             = _mm256_add_ps(fjz0,tz);
1061
1062             /**************************
1063              * CALCULATE INTERACTIONS *
1064              **************************/
1065
1066             r20              = _mm256_mul_ps(rsq20,rinv20);
1067             r20              = _mm256_andnot_ps(dummy_mask,r20);
1068
1069             /* Compute parameters for interactions between i and j atoms */
1070             qq20             = _mm256_mul_ps(iq2,jq0);
1071
1072             /* EWALD ELECTROSTATICS */
1073
1074             /* Analytical PME correction */
1075             zeta2            = _mm256_mul_ps(beta2,rsq20);
1076             rinv3            = _mm256_mul_ps(rinvsq20,rinv20);
1077             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
1078             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
1079             felec            = _mm256_mul_ps(qq20,felec);
1080
1081             fscal            = felec;
1082
1083             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
1084
1085             /* Calculate temporary vectorial force */
1086             tx               = _mm256_mul_ps(fscal,dx20);
1087             ty               = _mm256_mul_ps(fscal,dy20);
1088             tz               = _mm256_mul_ps(fscal,dz20);
1089
1090             /* Update vectorial force */
1091             fix2             = _mm256_add_ps(fix2,tx);
1092             fiy2             = _mm256_add_ps(fiy2,ty);
1093             fiz2             = _mm256_add_ps(fiz2,tz);
1094
1095             fjx0             = _mm256_add_ps(fjx0,tx);
1096             fjy0             = _mm256_add_ps(fjy0,ty);
1097             fjz0             = _mm256_add_ps(fjz0,tz);
1098
1099             /**************************
1100              * CALCULATE INTERACTIONS *
1101              **************************/
1102
1103             r30              = _mm256_mul_ps(rsq30,rinv30);
1104             r30              = _mm256_andnot_ps(dummy_mask,r30);
1105
1106             /* Compute parameters for interactions between i and j atoms */
1107             qq30             = _mm256_mul_ps(iq3,jq0);
1108
1109             /* EWALD ELECTROSTATICS */
1110
1111             /* Analytical PME correction */
1112             zeta2            = _mm256_mul_ps(beta2,rsq30);
1113             rinv3            = _mm256_mul_ps(rinvsq30,rinv30);
1114             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
1115             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
1116             felec            = _mm256_mul_ps(qq30,felec);
1117
1118             fscal            = felec;
1119
1120             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
1121
1122             /* Calculate temporary vectorial force */
1123             tx               = _mm256_mul_ps(fscal,dx30);
1124             ty               = _mm256_mul_ps(fscal,dy30);
1125             tz               = _mm256_mul_ps(fscal,dz30);
1126
1127             /* Update vectorial force */
1128             fix3             = _mm256_add_ps(fix3,tx);
1129             fiy3             = _mm256_add_ps(fiy3,ty);
1130             fiz3             = _mm256_add_ps(fiz3,tz);
1131
1132             fjx0             = _mm256_add_ps(fjx0,tx);
1133             fjy0             = _mm256_add_ps(fjy0,ty);
1134             fjz0             = _mm256_add_ps(fjz0,tz);
1135
1136             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
1137             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
1138             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
1139             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
1140             fjptrE             = (jnrlistE>=0) ? f+j_coord_offsetE : scratch;
1141             fjptrF             = (jnrlistF>=0) ? f+j_coord_offsetF : scratch;
1142             fjptrG             = (jnrlistG>=0) ? f+j_coord_offsetG : scratch;
1143             fjptrH             = (jnrlistH>=0) ? f+j_coord_offsetH : scratch;
1144
1145             gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,fjx0,fjy0,fjz0);
1146
1147             /* Inner loop uses 174 flops */
1148         }
1149
1150         /* End of innermost loop */
1151
1152         gmx_mm256_update_iforce_3atom_swizzle_ps(fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
1153                                                  f+i_coord_offset+DIM,fshift+i_shift_offset);
1154
1155         /* Increment number of inner iterations */
1156         inneriter                  += j_index_end - j_index_start;
1157
1158         /* Outer loop uses 18 flops */
1159     }
1160
1161     /* Increment number of outer iterations */
1162     outeriter        += nri;
1163
1164     /* Update outer/inner flops */
1165
1166     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_W4_F,outeriter*18 + inneriter*174);
1167 }