src/gmxlib/nonbonded/nb_kernel_avx_256_single/nb_kernel_ElecEwSh_VdwNone_GeomW3P1_avx_256_single.c

   1 /*
   2  * Note: this file was generated by the Gromacs avx_256_single kernel generator.
   3  *
   4  *                This source code is part of
   5  *
   6  *                 G   R   O   M   A   C   S
   7  *
   8  * Copyright (c) 2001-2012, The GROMACS Development Team
   9  *
  10  * Gromacs is a library for molecular simulation and trajectory analysis,
  11  * written by Erik Lindahl, David van der Spoel, Berk Hess, and others - for
  12  * a full list of developers and information, check out http://www.gromacs.org
  13  *
  14  * This program is free software; you can redistribute it and/or modify it under
  15  * the terms of the GNU Lesser General Public License as published by the Free
  16  * Software Foundation; either version 2 of the License, or (at your option) any
  17  * later version.
  18  *
  19  * To help fund GROMACS development, we humbly ask that you cite
  20  * the papers people have written on it - you can find them on the website.
  21  */
  22 #ifdef HAVE_CONFIG_H
  23 #include <config.h>
  24 #endif
  25
  26 #include <math.h>
  27
  28 #include "../nb_kernel.h"
  29 #include "types/simple.h"
  30 #include "vec.h"
  31 #include "nrnb.h"
  32
  33 #include "gmx_math_x86_avx_256_single.h"
  34 #include "kernelutil_x86_avx_256_single.h"
  35
  36 /*
  37  * Gromacs nonbonded kernel:   nb_kernel_ElecEwSh_VdwNone_GeomW3P1_VF_avx_256_single
  38  * Electrostatics interaction: Ewald
  39  * VdW interaction:            None
  40  * Geometry:                   Water3-Particle
  41  * Calculate force/pot:        PotentialAndForce
  42  */
  43 void
  44 nb_kernel_ElecEwSh_VdwNone_GeomW3P1_VF_avx_256_single
  45                     (t_nblist * gmx_restrict                nlist,
  46                      rvec * gmx_restrict                    xx,
  47                      rvec * gmx_restrict                    ff,
  48                      t_forcerec * gmx_restrict              fr,
  49                      t_mdatoms * gmx_restrict               mdatoms,
  50                      nb_kernel_data_t * gmx_restrict        kernel_data,
  51                      t_nrnb * gmx_restrict                  nrnb)
  52 {
  53     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
  54      * just 0 for non-waters.
  55      * Suffixes A,B,C,D,E,F,G,H refer to j loop unrolling done with AVX, e.g. for the eight different
  56      * jnr indices corresponding to data put in the four positions in the SIMD register.
  57      */
  58     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
  59     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
  60     int              jnrA,jnrB,jnrC,jnrD;
  61     int              jnrE,jnrF,jnrG,jnrH;
  62     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
  63     int              jnrlistE,jnrlistF,jnrlistG,jnrlistH;
  64     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
  65     int              j_coord_offsetE,j_coord_offsetF,j_coord_offsetG,j_coord_offsetH;
  66     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
  67     real             rcutoff_scalar;
  68     real             *shiftvec,*fshift,*x,*f;
  69     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD,*fjptrE,*fjptrF,*fjptrG,*fjptrH;
  70     real             scratch[4*DIM];
  71     __m256           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
  72     real *           vdwioffsetptr0;
  73     __m256           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
  74     real *           vdwioffsetptr1;
  75     __m256           ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
  76     real *           vdwioffsetptr2;
  77     __m256           ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
  78     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D,vdwjidx0E,vdwjidx0F,vdwjidx0G,vdwjidx0H;
  79     __m256           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  80     __m256           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  81     __m256           dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
  82     __m256           dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
  83     __m256           velec,felec,velecsum,facel,crf,krf,krf2;
  84     real             *charge;
  85     __m256i          ewitab;
  86     __m128i          ewitab_lo,ewitab_hi;
  87     __m256           ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
  88     __m256           beta,beta2,beta3,zeta2,pmecorrF,pmecorrV,rinv3;
  89     real             *ewtab;
  90     __m256           dummy_mask,cutoff_mask;
  91     __m256           signbit = _mm256_castsi256_ps( _mm256_set1_epi32(0x80000000) );
  92     __m256           one     = _mm256_set1_ps(1.0);
  93     __m256           two     = _mm256_set1_ps(2.0);
  94     x                = xx[0];
  95     f                = ff[0];
  96
  97     nri              = nlist->nri;
  98     iinr             = nlist->iinr;
  99     jindex           = nlist->jindex;
 100     jjnr             = nlist->jjnr;
 101     shiftidx         = nlist->shift;
 102     gid              = nlist->gid;
 103     shiftvec         = fr->shift_vec[0];
 104     fshift           = fr->fshift[0];
 105     facel            = _mm256_set1_ps(fr->epsfac);
 106     charge           = mdatoms->chargeA;
 107
 108     sh_ewald         = _mm256_set1_ps(fr->ic->sh_ewald);
 109     beta             = _mm256_set1_ps(fr->ic->ewaldcoeff);
 110     beta2            = _mm256_mul_ps(beta,beta);
 111     beta3            = _mm256_mul_ps(beta,beta2);
 112
 113     ewtab            = fr->ic->tabq_coul_FDV0;
 114     ewtabscale       = _mm256_set1_ps(fr->ic->tabq_scale);
 115     ewtabhalfspace   = _mm256_set1_ps(0.5/fr->ic->tabq_scale);
 116
 117     /* Setup water-specific parameters */
 118     inr              = nlist->iinr[0];
 119     iq0              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+0]));
 120     iq1              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+1]));
 121     iq2              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+2]));
 122
 123     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 124     rcutoff_scalar   = fr->rcoulomb;
 125     rcutoff          = _mm256_set1_ps(rcutoff_scalar);
 126     rcutoff2         = _mm256_mul_ps(rcutoff,rcutoff);
 127
 128     /* Avoid stupid compiler warnings */
 129     jnrA = jnrB = jnrC = jnrD = jnrE = jnrF = jnrG = jnrH = 0;
 130     j_coord_offsetA = 0;
 131     j_coord_offsetB = 0;
 132     j_coord_offsetC = 0;
 133     j_coord_offsetD = 0;
 134     j_coord_offsetE = 0;
 135     j_coord_offsetF = 0;
 136     j_coord_offsetG = 0;
 137     j_coord_offsetH = 0;
 138
 139     outeriter        = 0;
 140     inneriter        = 0;
 141
 142     for(iidx=0;iidx<4*DIM;iidx++)
 143     {
 144         scratch[iidx] = 0.0;
 145     }
 146
 147     /* Start outer loop over neighborlists */
 148     for(iidx=0; iidx<nri; iidx++)
 149     {
 150         /* Load shift vector for this list */
 151         i_shift_offset   = DIM*shiftidx[iidx];
 152
 153         /* Load limits for loop over neighbors */
 154         j_index_start    = jindex[iidx];
 155         j_index_end      = jindex[iidx+1];
 156
 157         /* Get outer coordinate index */
 158         inr              = iinr[iidx];
 159         i_coord_offset   = DIM*inr;
 160
 161         /* Load i particle coords and add shift vector */
 162         gmx_mm256_load_shift_and_3rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
 163                                                     &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 164
 165         fix0             = _mm256_setzero_ps();
 166         fiy0             = _mm256_setzero_ps();
 167         fiz0             = _mm256_setzero_ps();
 168         fix1             = _mm256_setzero_ps();
 169         fiy1             = _mm256_setzero_ps();
 170         fiz1             = _mm256_setzero_ps();
 171         fix2             = _mm256_setzero_ps();
 172         fiy2             = _mm256_setzero_ps();
 173         fiz2             = _mm256_setzero_ps();
 174
 175         /* Reset potential sums */
 176         velecsum         = _mm256_setzero_ps();
 177
 178         /* Start inner kernel loop */
 179         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+7]>=0; jidx+=8)
 180         {
 181
 182             /* Get j neighbor index, and coordinate index */
 183             jnrA             = jjnr[jidx];
 184             jnrB             = jjnr[jidx+1];
 185             jnrC             = jjnr[jidx+2];
 186             jnrD             = jjnr[jidx+3];
 187             jnrE             = jjnr[jidx+4];
 188             jnrF             = jjnr[jidx+5];
 189             jnrG             = jjnr[jidx+6];
 190             jnrH             = jjnr[jidx+7];
 191             j_coord_offsetA  = DIM*jnrA;
 192             j_coord_offsetB  = DIM*jnrB;
 193             j_coord_offsetC  = DIM*jnrC;
 194             j_coord_offsetD  = DIM*jnrD;
 195             j_coord_offsetE  = DIM*jnrE;
 196             j_coord_offsetF  = DIM*jnrF;
 197             j_coord_offsetG  = DIM*jnrG;
 198             j_coord_offsetH  = DIM*jnrH;
 199
 200             /* load j atom coordinates */
 201             gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 202                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 203                                                  x+j_coord_offsetE,x+j_coord_offsetF,
 204                                                  x+j_coord_offsetG,x+j_coord_offsetH,
 205                                                  &jx0,&jy0,&jz0);
 206
 207             /* Calculate displacement vector */
 208             dx00             = _mm256_sub_ps(ix0,jx0);
 209             dy00             = _mm256_sub_ps(iy0,jy0);
 210             dz00             = _mm256_sub_ps(iz0,jz0);
 211             dx10             = _mm256_sub_ps(ix1,jx0);
 212             dy10             = _mm256_sub_ps(iy1,jy0);
 213             dz10             = _mm256_sub_ps(iz1,jz0);
 214             dx20             = _mm256_sub_ps(ix2,jx0);
 215             dy20             = _mm256_sub_ps(iy2,jy0);
 216             dz20             = _mm256_sub_ps(iz2,jz0);
 217
 218             /* Calculate squared distance and things based on it */
 219             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
 220             rsq10            = gmx_mm256_calc_rsq_ps(dx10,dy10,dz10);
 221             rsq20            = gmx_mm256_calc_rsq_ps(dx20,dy20,dz20);
 222
 223             rinv00           = gmx_mm256_invsqrt_ps(rsq00);
 224             rinv10           = gmx_mm256_invsqrt_ps(rsq10);
 225             rinv20           = gmx_mm256_invsqrt_ps(rsq20);
 226
 227             rinvsq00         = _mm256_mul_ps(rinv00,rinv00);
 228             rinvsq10         = _mm256_mul_ps(rinv10,rinv10);
 229             rinvsq20         = _mm256_mul_ps(rinv20,rinv20);
 230
 231             /* Load parameters for j particles */
 232             jq0              = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 233                                                                  charge+jnrC+0,charge+jnrD+0,
 234                                                                  charge+jnrE+0,charge+jnrF+0,
 235                                                                  charge+jnrG+0,charge+jnrH+0);
 236
 237             fjx0             = _mm256_setzero_ps();
 238             fjy0             = _mm256_setzero_ps();
 239             fjz0             = _mm256_setzero_ps();
 240
 241             /**************************
 242              * CALCULATE INTERACTIONS *
 243              **************************/
 244
 245             if (gmx_mm256_any_lt(rsq00,rcutoff2))
 246             {
 247
 248             r00              = _mm256_mul_ps(rsq00,rinv00);
 249
 250             /* Compute parameters for interactions between i and j atoms */
 251             qq00             = _mm256_mul_ps(iq0,jq0);
 252
 253             /* EWALD ELECTROSTATICS */
 254
 255             /* Analytical PME correction */
 256             zeta2            = _mm256_mul_ps(beta2,rsq00);
 257             rinv3            = _mm256_mul_ps(rinvsq00,rinv00);
 258             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
 259             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
 260             felec            = _mm256_mul_ps(qq00,felec);
 261             pmecorrV         = gmx_mm256_pmecorrV_ps(zeta2);
 262             pmecorrV         = _mm256_mul_ps(pmecorrV,beta);
 263             velec            = _mm256_sub_ps(_mm256_sub_ps(rinv00,sh_ewald),pmecorrV);
 264             velec            = _mm256_mul_ps(qq00,velec);
 265
 266             cutoff_mask      = _mm256_cmp_ps(rsq00,rcutoff2,_CMP_LT_OQ);
 267
 268             /* Update potential sum for this i atom from the interaction with this j atom. */
 269             velec            = _mm256_and_ps(velec,cutoff_mask);
 270             velecsum         = _mm256_add_ps(velecsum,velec);
 271
 272             fscal            = felec;
 273
 274             fscal            = _mm256_and_ps(fscal,cutoff_mask);
 275
 276             /* Calculate temporary vectorial force */
 277             tx               = _mm256_mul_ps(fscal,dx00);
 278             ty               = _mm256_mul_ps(fscal,dy00);
 279             tz               = _mm256_mul_ps(fscal,dz00);
 280
 281             /* Update vectorial force */
 282             fix0             = _mm256_add_ps(fix0,tx);
 283             fiy0             = _mm256_add_ps(fiy0,ty);
 284             fiz0             = _mm256_add_ps(fiz0,tz);
 285
 286             fjx0             = _mm256_add_ps(fjx0,tx);
 287             fjy0             = _mm256_add_ps(fjy0,ty);
 288             fjz0             = _mm256_add_ps(fjz0,tz);
 289
 290             }
 291
 292             /**************************
 293              * CALCULATE INTERACTIONS *
 294              **************************/
 295
 296             if (gmx_mm256_any_lt(rsq10,rcutoff2))
 297             {
 298
 299             r10              = _mm256_mul_ps(rsq10,rinv10);
 300
 301             /* Compute parameters for interactions between i and j atoms */
 302             qq10             = _mm256_mul_ps(iq1,jq0);
 303
 304             /* EWALD ELECTROSTATICS */
 305
 306             /* Analytical PME correction */
 307             zeta2            = _mm256_mul_ps(beta2,rsq10);
 308             rinv3            = _mm256_mul_ps(rinvsq10,rinv10);
 309             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
 310             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
 311             felec            = _mm256_mul_ps(qq10,felec);
 312             pmecorrV         = gmx_mm256_pmecorrV_ps(zeta2);
 313             pmecorrV         = _mm256_mul_ps(pmecorrV,beta);
 314             velec            = _mm256_sub_ps(_mm256_sub_ps(rinv10,sh_ewald),pmecorrV);
 315             velec            = _mm256_mul_ps(qq10,velec);
 316
 317             cutoff_mask      = _mm256_cmp_ps(rsq10,rcutoff2,_CMP_LT_OQ);
 318
 319             /* Update potential sum for this i atom from the interaction with this j atom. */
 320             velec            = _mm256_and_ps(velec,cutoff_mask);
 321             velecsum         = _mm256_add_ps(velecsum,velec);
 322
 323             fscal            = felec;
 324
 325             fscal            = _mm256_and_ps(fscal,cutoff_mask);
 326
 327             /* Calculate temporary vectorial force */
 328             tx               = _mm256_mul_ps(fscal,dx10);
 329             ty               = _mm256_mul_ps(fscal,dy10);
 330             tz               = _mm256_mul_ps(fscal,dz10);
 331
 332             /* Update vectorial force */
 333             fix1             = _mm256_add_ps(fix1,tx);
 334             fiy1             = _mm256_add_ps(fiy1,ty);
 335             fiz1             = _mm256_add_ps(fiz1,tz);
 336
 337             fjx0             = _mm256_add_ps(fjx0,tx);
 338             fjy0             = _mm256_add_ps(fjy0,ty);
 339             fjz0             = _mm256_add_ps(fjz0,tz);
 340
 341             }
 342
 343             /**************************
 344              * CALCULATE INTERACTIONS *
 345              **************************/
 346
 347             if (gmx_mm256_any_lt(rsq20,rcutoff2))
 348             {
 349
 350             r20              = _mm256_mul_ps(rsq20,rinv20);
 351
 352             /* Compute parameters for interactions between i and j atoms */
 353             qq20             = _mm256_mul_ps(iq2,jq0);
 354
 355             /* EWALD ELECTROSTATICS */
 356
 357             /* Analytical PME correction */
 358             zeta2            = _mm256_mul_ps(beta2,rsq20);
 359             rinv3            = _mm256_mul_ps(rinvsq20,rinv20);
 360             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
 361             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
 362             felec            = _mm256_mul_ps(qq20,felec);
 363             pmecorrV         = gmx_mm256_pmecorrV_ps(zeta2);
 364             pmecorrV         = _mm256_mul_ps(pmecorrV,beta);
 365             velec            = _mm256_sub_ps(_mm256_sub_ps(rinv20,sh_ewald),pmecorrV);
 366             velec            = _mm256_mul_ps(qq20,velec);
 367
 368             cutoff_mask      = _mm256_cmp_ps(rsq20,rcutoff2,_CMP_LT_OQ);
 369
 370             /* Update potential sum for this i atom from the interaction with this j atom. */
 371             velec            = _mm256_and_ps(velec,cutoff_mask);
 372             velecsum         = _mm256_add_ps(velecsum,velec);
 373
 374             fscal            = felec;
 375
 376             fscal            = _mm256_and_ps(fscal,cutoff_mask);
 377
 378             /* Calculate temporary vectorial force */
 379             tx               = _mm256_mul_ps(fscal,dx20);
 380             ty               = _mm256_mul_ps(fscal,dy20);
 381             tz               = _mm256_mul_ps(fscal,dz20);
 382
 383             /* Update vectorial force */
 384             fix2             = _mm256_add_ps(fix2,tx);
 385             fiy2             = _mm256_add_ps(fiy2,ty);
 386             fiz2             = _mm256_add_ps(fiz2,tz);
 387
 388             fjx0             = _mm256_add_ps(fjx0,tx);
 389             fjy0             = _mm256_add_ps(fjy0,ty);
 390             fjz0             = _mm256_add_ps(fjz0,tz);
 391
 392             }
 393
 394             fjptrA             = f+j_coord_offsetA;
 395             fjptrB             = f+j_coord_offsetB;
 396             fjptrC             = f+j_coord_offsetC;
 397             fjptrD             = f+j_coord_offsetD;
 398             fjptrE             = f+j_coord_offsetE;
 399             fjptrF             = f+j_coord_offsetF;
 400             fjptrG             = f+j_coord_offsetG;
 401             fjptrH             = f+j_coord_offsetH;
 402
 403             gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,fjx0,fjy0,fjz0);
 404
 405             /* Inner loop uses 330 flops */
 406         }
 407
 408         if(jidx<j_index_end)
 409         {
 410
 411             /* Get j neighbor index, and coordinate index */
 412             jnrlistA         = jjnr[jidx];
 413             jnrlistB         = jjnr[jidx+1];
 414             jnrlistC         = jjnr[jidx+2];
 415             jnrlistD         = jjnr[jidx+3];
 416             jnrlistE         = jjnr[jidx+4];
 417             jnrlistF         = jjnr[jidx+5];
 418             jnrlistG         = jjnr[jidx+6];
 419             jnrlistH         = jjnr[jidx+7];
 420             /* Sign of each element will be negative for non-real atoms.
 421              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 422              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 423              */
 424             dummy_mask = gmx_mm256_set_m128(gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx+4)),_mm_setzero_si128())),
 425                                             gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128())));
 426
 427             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 428             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 429             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 430             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 431             jnrE       = (jnrlistE>=0) ? jnrlistE : 0;
 432             jnrF       = (jnrlistF>=0) ? jnrlistF : 0;
 433             jnrG       = (jnrlistG>=0) ? jnrlistG : 0;
 434             jnrH       = (jnrlistH>=0) ? jnrlistH : 0;
 435             j_coord_offsetA  = DIM*jnrA;
 436             j_coord_offsetB  = DIM*jnrB;
 437             j_coord_offsetC  = DIM*jnrC;
 438             j_coord_offsetD  = DIM*jnrD;
 439             j_coord_offsetE  = DIM*jnrE;
 440             j_coord_offsetF  = DIM*jnrF;
 441             j_coord_offsetG  = DIM*jnrG;
 442             j_coord_offsetH  = DIM*jnrH;
 443
 444             /* load j atom coordinates */
 445             gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 446                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 447                                                  x+j_coord_offsetE,x+j_coord_offsetF,
 448                                                  x+j_coord_offsetG,x+j_coord_offsetH,
 449                                                  &jx0,&jy0,&jz0);
 450
 451             /* Calculate displacement vector */
 452             dx00             = _mm256_sub_ps(ix0,jx0);
 453             dy00             = _mm256_sub_ps(iy0,jy0);
 454             dz00             = _mm256_sub_ps(iz0,jz0);
 455             dx10             = _mm256_sub_ps(ix1,jx0);
 456             dy10             = _mm256_sub_ps(iy1,jy0);
 457             dz10             = _mm256_sub_ps(iz1,jz0);
 458             dx20             = _mm256_sub_ps(ix2,jx0);
 459             dy20             = _mm256_sub_ps(iy2,jy0);
 460             dz20             = _mm256_sub_ps(iz2,jz0);
 461
 462             /* Calculate squared distance and things based on it */
 463             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
 464             rsq10            = gmx_mm256_calc_rsq_ps(dx10,dy10,dz10);
 465             rsq20            = gmx_mm256_calc_rsq_ps(dx20,dy20,dz20);
 466
 467             rinv00           = gmx_mm256_invsqrt_ps(rsq00);
 468             rinv10           = gmx_mm256_invsqrt_ps(rsq10);
 469             rinv20           = gmx_mm256_invsqrt_ps(rsq20);
 470
 471             rinvsq00         = _mm256_mul_ps(rinv00,rinv00);
 472             rinvsq10         = _mm256_mul_ps(rinv10,rinv10);
 473             rinvsq20         = _mm256_mul_ps(rinv20,rinv20);
 474
 475             /* Load parameters for j particles */
 476             jq0              = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 477                                                                  charge+jnrC+0,charge+jnrD+0,
 478                                                                  charge+jnrE+0,charge+jnrF+0,
 479                                                                  charge+jnrG+0,charge+jnrH+0);
 480
 481             fjx0             = _mm256_setzero_ps();
 482             fjy0             = _mm256_setzero_ps();
 483             fjz0             = _mm256_setzero_ps();
 484
 485             /**************************
 486              * CALCULATE INTERACTIONS *
 487              **************************/
 488
 489             if (gmx_mm256_any_lt(rsq00,rcutoff2))
 490             {
 491
 492             r00              = _mm256_mul_ps(rsq00,rinv00);
 493             r00              = _mm256_andnot_ps(dummy_mask,r00);
 494
 495             /* Compute parameters for interactions between i and j atoms */
 496             qq00             = _mm256_mul_ps(iq0,jq0);
 497
 498             /* EWALD ELECTROSTATICS */
 499
 500             /* Analytical PME correction */
 501             zeta2            = _mm256_mul_ps(beta2,rsq00);
 502             rinv3            = _mm256_mul_ps(rinvsq00,rinv00);
 503             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
 504             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
 505             felec            = _mm256_mul_ps(qq00,felec);
 506             pmecorrV         = gmx_mm256_pmecorrV_ps(zeta2);
 507             pmecorrV         = _mm256_mul_ps(pmecorrV,beta);
 508             velec            = _mm256_sub_ps(_mm256_sub_ps(rinv00,sh_ewald),pmecorrV);
 509             velec            = _mm256_mul_ps(qq00,velec);
 510
 511             cutoff_mask      = _mm256_cmp_ps(rsq00,rcutoff2,_CMP_LT_OQ);
 512
 513             /* Update potential sum for this i atom from the interaction with this j atom. */
 514             velec            = _mm256_and_ps(velec,cutoff_mask);
 515             velec            = _mm256_andnot_ps(dummy_mask,velec);
 516             velecsum         = _mm256_add_ps(velecsum,velec);
 517
 518             fscal            = felec;
 519
 520             fscal            = _mm256_and_ps(fscal,cutoff_mask);
 521
 522             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
 523
 524             /* Calculate temporary vectorial force */
 525             tx               = _mm256_mul_ps(fscal,dx00);
 526             ty               = _mm256_mul_ps(fscal,dy00);
 527             tz               = _mm256_mul_ps(fscal,dz00);
 528
 529             /* Update vectorial force */
 530             fix0             = _mm256_add_ps(fix0,tx);
 531             fiy0             = _mm256_add_ps(fiy0,ty);
 532             fiz0             = _mm256_add_ps(fiz0,tz);
 533
 534             fjx0             = _mm256_add_ps(fjx0,tx);
 535             fjy0             = _mm256_add_ps(fjy0,ty);
 536             fjz0             = _mm256_add_ps(fjz0,tz);
 537
 538             }
 539
 540             /**************************
 541              * CALCULATE INTERACTIONS *
 542              **************************/
 543
 544             if (gmx_mm256_any_lt(rsq10,rcutoff2))
 545             {
 546
 547             r10              = _mm256_mul_ps(rsq10,rinv10);
 548             r10              = _mm256_andnot_ps(dummy_mask,r10);
 549
 550             /* Compute parameters for interactions between i and j atoms */
 551             qq10             = _mm256_mul_ps(iq1,jq0);
 552
 553             /* EWALD ELECTROSTATICS */
 554
 555             /* Analytical PME correction */
 556             zeta2            = _mm256_mul_ps(beta2,rsq10);
 557             rinv3            = _mm256_mul_ps(rinvsq10,rinv10);
 558             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
 559             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
 560             felec            = _mm256_mul_ps(qq10,felec);
 561             pmecorrV         = gmx_mm256_pmecorrV_ps(zeta2);
 562             pmecorrV         = _mm256_mul_ps(pmecorrV,beta);
 563             velec            = _mm256_sub_ps(_mm256_sub_ps(rinv10,sh_ewald),pmecorrV);
 564             velec            = _mm256_mul_ps(qq10,velec);
 565
 566             cutoff_mask      = _mm256_cmp_ps(rsq10,rcutoff2,_CMP_LT_OQ);
 567
 568             /* Update potential sum for this i atom from the interaction with this j atom. */
 569             velec            = _mm256_and_ps(velec,cutoff_mask);
 570             velec            = _mm256_andnot_ps(dummy_mask,velec);
 571             velecsum         = _mm256_add_ps(velecsum,velec);
 572
 573             fscal            = felec;
 574
 575             fscal            = _mm256_and_ps(fscal,cutoff_mask);
 576
 577             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
 578
 579             /* Calculate temporary vectorial force */
 580             tx               = _mm256_mul_ps(fscal,dx10);
 581             ty               = _mm256_mul_ps(fscal,dy10);
 582             tz               = _mm256_mul_ps(fscal,dz10);
 583
 584             /* Update vectorial force */
 585             fix1             = _mm256_add_ps(fix1,tx);
 586             fiy1             = _mm256_add_ps(fiy1,ty);
 587             fiz1             = _mm256_add_ps(fiz1,tz);
 588
 589             fjx0             = _mm256_add_ps(fjx0,tx);
 590             fjy0             = _mm256_add_ps(fjy0,ty);
 591             fjz0             = _mm256_add_ps(fjz0,tz);
 592
 593             }
 594
 595             /**************************
 596              * CALCULATE INTERACTIONS *
 597              **************************/
 598
 599             if (gmx_mm256_any_lt(rsq20,rcutoff2))
 600             {
 601
 602             r20              = _mm256_mul_ps(rsq20,rinv20);
 603             r20              = _mm256_andnot_ps(dummy_mask,r20);
 604
 605             /* Compute parameters for interactions between i and j atoms */
 606             qq20             = _mm256_mul_ps(iq2,jq0);
 607
 608             /* EWALD ELECTROSTATICS */
 609
 610             /* Analytical PME correction */
 611             zeta2            = _mm256_mul_ps(beta2,rsq20);
 612             rinv3            = _mm256_mul_ps(rinvsq20,rinv20);
 613             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
 614             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
 615             felec            = _mm256_mul_ps(qq20,felec);
 616             pmecorrV         = gmx_mm256_pmecorrV_ps(zeta2);
 617             pmecorrV         = _mm256_mul_ps(pmecorrV,beta);
 618             velec            = _mm256_sub_ps(_mm256_sub_ps(rinv20,sh_ewald),pmecorrV);
 619             velec            = _mm256_mul_ps(qq20,velec);
 620
 621             cutoff_mask      = _mm256_cmp_ps(rsq20,rcutoff2,_CMP_LT_OQ);
 622
 623             /* Update potential sum for this i atom from the interaction with this j atom. */
 624             velec            = _mm256_and_ps(velec,cutoff_mask);
 625             velec            = _mm256_andnot_ps(dummy_mask,velec);
 626             velecsum         = _mm256_add_ps(velecsum,velec);
 627
 628             fscal            = felec;
 629
 630             fscal            = _mm256_and_ps(fscal,cutoff_mask);
 631
 632             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
 633
 634             /* Calculate temporary vectorial force */
 635             tx               = _mm256_mul_ps(fscal,dx20);
 636             ty               = _mm256_mul_ps(fscal,dy20);
 637             tz               = _mm256_mul_ps(fscal,dz20);
 638
 639             /* Update vectorial force */
 640             fix2             = _mm256_add_ps(fix2,tx);
 641             fiy2             = _mm256_add_ps(fiy2,ty);
 642             fiz2             = _mm256_add_ps(fiz2,tz);
 643
 644             fjx0             = _mm256_add_ps(fjx0,tx);
 645             fjy0             = _mm256_add_ps(fjy0,ty);
 646             fjz0             = _mm256_add_ps(fjz0,tz);
 647
 648             }
 649
 650             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 651             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 652             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 653             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 654             fjptrE             = (jnrlistE>=0) ? f+j_coord_offsetE : scratch;
 655             fjptrF             = (jnrlistF>=0) ? f+j_coord_offsetF : scratch;
 656             fjptrG             = (jnrlistG>=0) ? f+j_coord_offsetG : scratch;
 657             fjptrH             = (jnrlistH>=0) ? f+j_coord_offsetH : scratch;
 658
 659             gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,fjx0,fjy0,fjz0);
 660
 661             /* Inner loop uses 333 flops */
 662         }
 663
 664         /* End of innermost loop */
 665
 666         gmx_mm256_update_iforce_3atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
 667                                                  f+i_coord_offset,fshift+i_shift_offset);
 668
 669         ggid                        = gid[iidx];
 670         /* Update potential energies */
 671         gmx_mm256_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid);
 672
 673         /* Increment number of inner iterations */
 674         inneriter                  += j_index_end - j_index_start;
 675
 676         /* Outer loop uses 19 flops */
 677     }
 678
 679     /* Increment number of outer iterations */
 680     outeriter        += nri;
 681
 682     /* Update outer/inner flops */
 683
 684     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_W3_VF,outeriter*19 + inneriter*333);
 685 }
 686 /*
 687  * Gromacs nonbonded kernel:   nb_kernel_ElecEwSh_VdwNone_GeomW3P1_F_avx_256_single
 688  * Electrostatics interaction: Ewald
 689  * VdW interaction:            None
 690  * Geometry:                   Water3-Particle
 691  * Calculate force/pot:        Force
 692  */
 693 void
 694 nb_kernel_ElecEwSh_VdwNone_GeomW3P1_F_avx_256_single
 695                     (t_nblist * gmx_restrict                nlist,
 696                      rvec * gmx_restrict                    xx,
 697                      rvec * gmx_restrict                    ff,
 698                      t_forcerec * gmx_restrict              fr,
 699                      t_mdatoms * gmx_restrict               mdatoms,
 700                      nb_kernel_data_t * gmx_restrict        kernel_data,
 701                      t_nrnb * gmx_restrict                  nrnb)
 702 {
 703     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 704      * just 0 for non-waters.
 705      * Suffixes A,B,C,D,E,F,G,H refer to j loop unrolling done with AVX, e.g. for the eight different
 706      * jnr indices corresponding to data put in the four positions in the SIMD register.
 707      */
 708     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 709     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 710     int              jnrA,jnrB,jnrC,jnrD;
 711     int              jnrE,jnrF,jnrG,jnrH;
 712     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
 713     int              jnrlistE,jnrlistF,jnrlistG,jnrlistH;
 714     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 715     int              j_coord_offsetE,j_coord_offsetF,j_coord_offsetG,j_coord_offsetH;
 716     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 717     real             rcutoff_scalar;
 718     real             *shiftvec,*fshift,*x,*f;
 719     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD,*fjptrE,*fjptrF,*fjptrG,*fjptrH;
 720     real             scratch[4*DIM];
 721     __m256           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 722     real *           vdwioffsetptr0;
 723     __m256           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 724     real *           vdwioffsetptr1;
 725     __m256           ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
 726     real *           vdwioffsetptr2;
 727     __m256           ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
 728     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D,vdwjidx0E,vdwjidx0F,vdwjidx0G,vdwjidx0H;
 729     __m256           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 730     __m256           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 731     __m256           dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
 732     __m256           dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
 733     __m256           velec,felec,velecsum,facel,crf,krf,krf2;
 734     real             *charge;
 735     __m256i          ewitab;
 736     __m128i          ewitab_lo,ewitab_hi;
 737     __m256           ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
 738     __m256           beta,beta2,beta3,zeta2,pmecorrF,pmecorrV,rinv3;
 739     real             *ewtab;
 740     __m256           dummy_mask,cutoff_mask;
 741     __m256           signbit = _mm256_castsi256_ps( _mm256_set1_epi32(0x80000000) );
 742     __m256           one     = _mm256_set1_ps(1.0);
 743     __m256           two     = _mm256_set1_ps(2.0);
 744     x                = xx[0];
 745     f                = ff[0];
 746
 747     nri              = nlist->nri;
 748     iinr             = nlist->iinr;
 749     jindex           = nlist->jindex;
 750     jjnr             = nlist->jjnr;
 751     shiftidx         = nlist->shift;
 752     gid              = nlist->gid;
 753     shiftvec         = fr->shift_vec[0];
 754     fshift           = fr->fshift[0];
 755     facel            = _mm256_set1_ps(fr->epsfac);
 756     charge           = mdatoms->chargeA;
 757
 758     sh_ewald         = _mm256_set1_ps(fr->ic->sh_ewald);
 759     beta             = _mm256_set1_ps(fr->ic->ewaldcoeff);
 760     beta2            = _mm256_mul_ps(beta,beta);
 761     beta3            = _mm256_mul_ps(beta,beta2);
 762
 763     ewtab            = fr->ic->tabq_coul_F;
 764     ewtabscale       = _mm256_set1_ps(fr->ic->tabq_scale);
 765     ewtabhalfspace   = _mm256_set1_ps(0.5/fr->ic->tabq_scale);
 766
 767     /* Setup water-specific parameters */
 768     inr              = nlist->iinr[0];
 769     iq0              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+0]));
 770     iq1              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+1]));
 771     iq2              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+2]));
 772
 773     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 774     rcutoff_scalar   = fr->rcoulomb;
 775     rcutoff          = _mm256_set1_ps(rcutoff_scalar);
 776     rcutoff2         = _mm256_mul_ps(rcutoff,rcutoff);
 777
 778     /* Avoid stupid compiler warnings */
 779     jnrA = jnrB = jnrC = jnrD = jnrE = jnrF = jnrG = jnrH = 0;
 780     j_coord_offsetA = 0;
 781     j_coord_offsetB = 0;
 782     j_coord_offsetC = 0;
 783     j_coord_offsetD = 0;
 784     j_coord_offsetE = 0;
 785     j_coord_offsetF = 0;
 786     j_coord_offsetG = 0;
 787     j_coord_offsetH = 0;
 788
 789     outeriter        = 0;
 790     inneriter        = 0;
 791
 792     for(iidx=0;iidx<4*DIM;iidx++)
 793     {
 794         scratch[iidx] = 0.0;
 795     }
 796
 797     /* Start outer loop over neighborlists */
 798     for(iidx=0; iidx<nri; iidx++)
 799     {
 800         /* Load shift vector for this list */
 801         i_shift_offset   = DIM*shiftidx[iidx];
 802
 803         /* Load limits for loop over neighbors */
 804         j_index_start    = jindex[iidx];
 805         j_index_end      = jindex[iidx+1];
 806
 807         /* Get outer coordinate index */
 808         inr              = iinr[iidx];
 809         i_coord_offset   = DIM*inr;
 810
 811         /* Load i particle coords and add shift vector */
 812         gmx_mm256_load_shift_and_3rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
 813                                                     &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 814
 815         fix0             = _mm256_setzero_ps();
 816         fiy0             = _mm256_setzero_ps();
 817         fiz0             = _mm256_setzero_ps();
 818         fix1             = _mm256_setzero_ps();
 819         fiy1             = _mm256_setzero_ps();
 820         fiz1             = _mm256_setzero_ps();
 821         fix2             = _mm256_setzero_ps();
 822         fiy2             = _mm256_setzero_ps();
 823         fiz2             = _mm256_setzero_ps();
 824
 825         /* Start inner kernel loop */
 826         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+7]>=0; jidx+=8)
 827         {
 828
 829             /* Get j neighbor index, and coordinate index */
 830             jnrA             = jjnr[jidx];
 831             jnrB             = jjnr[jidx+1];
 832             jnrC             = jjnr[jidx+2];
 833             jnrD             = jjnr[jidx+3];
 834             jnrE             = jjnr[jidx+4];
 835             jnrF             = jjnr[jidx+5];
 836             jnrG             = jjnr[jidx+6];
 837             jnrH             = jjnr[jidx+7];
 838             j_coord_offsetA  = DIM*jnrA;
 839             j_coord_offsetB  = DIM*jnrB;
 840             j_coord_offsetC  = DIM*jnrC;
 841             j_coord_offsetD  = DIM*jnrD;
 842             j_coord_offsetE  = DIM*jnrE;
 843             j_coord_offsetF  = DIM*jnrF;
 844             j_coord_offsetG  = DIM*jnrG;
 845             j_coord_offsetH  = DIM*jnrH;
 846
 847             /* load j atom coordinates */
 848             gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 849                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 850                                                  x+j_coord_offsetE,x+j_coord_offsetF,
 851                                                  x+j_coord_offsetG,x+j_coord_offsetH,
 852                                                  &jx0,&jy0,&jz0);
 853
 854             /* Calculate displacement vector */
 855             dx00             = _mm256_sub_ps(ix0,jx0);
 856             dy00             = _mm256_sub_ps(iy0,jy0);
 857             dz00             = _mm256_sub_ps(iz0,jz0);
 858             dx10             = _mm256_sub_ps(ix1,jx0);
 859             dy10             = _mm256_sub_ps(iy1,jy0);
 860             dz10             = _mm256_sub_ps(iz1,jz0);
 861             dx20             = _mm256_sub_ps(ix2,jx0);
 862             dy20             = _mm256_sub_ps(iy2,jy0);
 863             dz20             = _mm256_sub_ps(iz2,jz0);
 864
 865             /* Calculate squared distance and things based on it */
 866             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
 867             rsq10            = gmx_mm256_calc_rsq_ps(dx10,dy10,dz10);
 868             rsq20            = gmx_mm256_calc_rsq_ps(dx20,dy20,dz20);
 869
 870             rinv00           = gmx_mm256_invsqrt_ps(rsq00);
 871             rinv10           = gmx_mm256_invsqrt_ps(rsq10);
 872             rinv20           = gmx_mm256_invsqrt_ps(rsq20);
 873
 874             rinvsq00         = _mm256_mul_ps(rinv00,rinv00);
 875             rinvsq10         = _mm256_mul_ps(rinv10,rinv10);
 876             rinvsq20         = _mm256_mul_ps(rinv20,rinv20);
 877
 878             /* Load parameters for j particles */
 879             jq0              = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 880                                                                  charge+jnrC+0,charge+jnrD+0,
 881                                                                  charge+jnrE+0,charge+jnrF+0,
 882                                                                  charge+jnrG+0,charge+jnrH+0);
 883
 884             fjx0             = _mm256_setzero_ps();
 885             fjy0             = _mm256_setzero_ps();
 886             fjz0             = _mm256_setzero_ps();
 887
 888             /**************************
 889              * CALCULATE INTERACTIONS *
 890              **************************/
 891
 892             if (gmx_mm256_any_lt(rsq00,rcutoff2))
 893             {
 894
 895             r00              = _mm256_mul_ps(rsq00,rinv00);
 896
 897             /* Compute parameters for interactions between i and j atoms */
 898             qq00             = _mm256_mul_ps(iq0,jq0);
 899
 900             /* EWALD ELECTROSTATICS */
 901
 902             /* Analytical PME correction */
 903             zeta2            = _mm256_mul_ps(beta2,rsq00);
 904             rinv3            = _mm256_mul_ps(rinvsq00,rinv00);
 905             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
 906             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
 907             felec            = _mm256_mul_ps(qq00,felec);
 908
 909             cutoff_mask      = _mm256_cmp_ps(rsq00,rcutoff2,_CMP_LT_OQ);
 910
 911             fscal            = felec;
 912
 913             fscal            = _mm256_and_ps(fscal,cutoff_mask);
 914
 915             /* Calculate temporary vectorial force */
 916             tx               = _mm256_mul_ps(fscal,dx00);
 917             ty               = _mm256_mul_ps(fscal,dy00);
 918             tz               = _mm256_mul_ps(fscal,dz00);
 919
 920             /* Update vectorial force */
 921             fix0             = _mm256_add_ps(fix0,tx);
 922             fiy0             = _mm256_add_ps(fiy0,ty);
 923             fiz0             = _mm256_add_ps(fiz0,tz);
 924
 925             fjx0             = _mm256_add_ps(fjx0,tx);
 926             fjy0             = _mm256_add_ps(fjy0,ty);
 927             fjz0             = _mm256_add_ps(fjz0,tz);
 928
 929             }
 930
 931             /**************************
 932              * CALCULATE INTERACTIONS *
 933              **************************/
 934
 935             if (gmx_mm256_any_lt(rsq10,rcutoff2))
 936             {
 937
 938             r10              = _mm256_mul_ps(rsq10,rinv10);
 939
 940             /* Compute parameters for interactions between i and j atoms */
 941             qq10             = _mm256_mul_ps(iq1,jq0);
 942
 943             /* EWALD ELECTROSTATICS */
 944
 945             /* Analytical PME correction */
 946             zeta2            = _mm256_mul_ps(beta2,rsq10);
 947             rinv3            = _mm256_mul_ps(rinvsq10,rinv10);
 948             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
 949             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
 950             felec            = _mm256_mul_ps(qq10,felec);
 951
 952             cutoff_mask      = _mm256_cmp_ps(rsq10,rcutoff2,_CMP_LT_OQ);
 953
 954             fscal            = felec;
 955
 956             fscal            = _mm256_and_ps(fscal,cutoff_mask);
 957
 958             /* Calculate temporary vectorial force */
 959             tx               = _mm256_mul_ps(fscal,dx10);
 960             ty               = _mm256_mul_ps(fscal,dy10);
 961             tz               = _mm256_mul_ps(fscal,dz10);
 962
 963             /* Update vectorial force */
 964             fix1             = _mm256_add_ps(fix1,tx);
 965             fiy1             = _mm256_add_ps(fiy1,ty);
 966             fiz1             = _mm256_add_ps(fiz1,tz);
 967
 968             fjx0             = _mm256_add_ps(fjx0,tx);
 969             fjy0             = _mm256_add_ps(fjy0,ty);
 970             fjz0             = _mm256_add_ps(fjz0,tz);
 971
 972             }
 973
 974             /**************************
 975              * CALCULATE INTERACTIONS *
 976              **************************/
 977
 978             if (gmx_mm256_any_lt(rsq20,rcutoff2))
 979             {
 980
 981             r20              = _mm256_mul_ps(rsq20,rinv20);
 982
 983             /* Compute parameters for interactions between i and j atoms */
 984             qq20             = _mm256_mul_ps(iq2,jq0);
 985
 986             /* EWALD ELECTROSTATICS */
 987
 988             /* Analytical PME correction */
 989             zeta2            = _mm256_mul_ps(beta2,rsq20);
 990             rinv3            = _mm256_mul_ps(rinvsq20,rinv20);
 991             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
 992             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
 993             felec            = _mm256_mul_ps(qq20,felec);
 994
 995             cutoff_mask      = _mm256_cmp_ps(rsq20,rcutoff2,_CMP_LT_OQ);
 996
 997             fscal            = felec;
 998
 999             fscal            = _mm256_and_ps(fscal,cutoff_mask);
1000
1001             /* Calculate temporary vectorial force */
1002             tx               = _mm256_mul_ps(fscal,dx20);
1003             ty               = _mm256_mul_ps(fscal,dy20);
1004             tz               = _mm256_mul_ps(fscal,dz20);
1005
1006             /* Update vectorial force */
1007             fix2             = _mm256_add_ps(fix2,tx);
1008             fiy2             = _mm256_add_ps(fiy2,ty);
1009             fiz2             = _mm256_add_ps(fiz2,tz);
1010
1011             fjx0             = _mm256_add_ps(fjx0,tx);
1012             fjy0             = _mm256_add_ps(fjy0,ty);
1013             fjz0             = _mm256_add_ps(fjz0,tz);
1014
1015             }
1016
1017             fjptrA             = f+j_coord_offsetA;
1018             fjptrB             = f+j_coord_offsetB;
1019             fjptrC             = f+j_coord_offsetC;
1020             fjptrD             = f+j_coord_offsetD;
1021             fjptrE             = f+j_coord_offsetE;
1022             fjptrF             = f+j_coord_offsetF;
1023             fjptrG             = f+j_coord_offsetG;
1024             fjptrH             = f+j_coord_offsetH;
1025
1026             gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,fjx0,fjy0,fjz0);
1027
1028             /* Inner loop uses 180 flops */
1029         }
1030
1031         if(jidx<j_index_end)
1032         {
1033
1034             /* Get j neighbor index, and coordinate index */
1035             jnrlistA         = jjnr[jidx];
1036             jnrlistB         = jjnr[jidx+1];
1037             jnrlistC         = jjnr[jidx+2];
1038             jnrlistD         = jjnr[jidx+3];
1039             jnrlistE         = jjnr[jidx+4];
1040             jnrlistF         = jjnr[jidx+5];
1041             jnrlistG         = jjnr[jidx+6];
1042             jnrlistH         = jjnr[jidx+7];
1043             /* Sign of each element will be negative for non-real atoms.
1044              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
1045              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
1046              */
1047             dummy_mask = gmx_mm256_set_m128(gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx+4)),_mm_setzero_si128())),
1048                                             gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128())));
1049
1050             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
1051             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
1052             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
1053             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
1054             jnrE       = (jnrlistE>=0) ? jnrlistE : 0;
1055             jnrF       = (jnrlistF>=0) ? jnrlistF : 0;
1056             jnrG       = (jnrlistG>=0) ? jnrlistG : 0;
1057             jnrH       = (jnrlistH>=0) ? jnrlistH : 0;
1058             j_coord_offsetA  = DIM*jnrA;
1059             j_coord_offsetB  = DIM*jnrB;
1060             j_coord_offsetC  = DIM*jnrC;
1061             j_coord_offsetD  = DIM*jnrD;
1062             j_coord_offsetE  = DIM*jnrE;
1063             j_coord_offsetF  = DIM*jnrF;
1064             j_coord_offsetG  = DIM*jnrG;
1065             j_coord_offsetH  = DIM*jnrH;
1066
1067             /* load j atom coordinates */
1068             gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
1069                                                  x+j_coord_offsetC,x+j_coord_offsetD,
1070                                                  x+j_coord_offsetE,x+j_coord_offsetF,
1071                                                  x+j_coord_offsetG,x+j_coord_offsetH,
1072                                                  &jx0,&jy0,&jz0);
1073
1074             /* Calculate displacement vector */
1075             dx00             = _mm256_sub_ps(ix0,jx0);
1076             dy00             = _mm256_sub_ps(iy0,jy0);
1077             dz00             = _mm256_sub_ps(iz0,jz0);
1078             dx10             = _mm256_sub_ps(ix1,jx0);
1079             dy10             = _mm256_sub_ps(iy1,jy0);
1080             dz10             = _mm256_sub_ps(iz1,jz0);
1081             dx20             = _mm256_sub_ps(ix2,jx0);
1082             dy20             = _mm256_sub_ps(iy2,jy0);
1083             dz20             = _mm256_sub_ps(iz2,jz0);
1084
1085             /* Calculate squared distance and things based on it */
1086             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
1087             rsq10            = gmx_mm256_calc_rsq_ps(dx10,dy10,dz10);
1088             rsq20            = gmx_mm256_calc_rsq_ps(dx20,dy20,dz20);
1089
1090             rinv00           = gmx_mm256_invsqrt_ps(rsq00);
1091             rinv10           = gmx_mm256_invsqrt_ps(rsq10);
1092             rinv20           = gmx_mm256_invsqrt_ps(rsq20);
1093
1094             rinvsq00         = _mm256_mul_ps(rinv00,rinv00);
1095             rinvsq10         = _mm256_mul_ps(rinv10,rinv10);
1096             rinvsq20         = _mm256_mul_ps(rinv20,rinv20);
1097
1098             /* Load parameters for j particles */
1099             jq0              = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
1100                                                                  charge+jnrC+0,charge+jnrD+0,
1101                                                                  charge+jnrE+0,charge+jnrF+0,
1102                                                                  charge+jnrG+0,charge+jnrH+0);
1103
1104             fjx0             = _mm256_setzero_ps();
1105             fjy0             = _mm256_setzero_ps();
1106             fjz0             = _mm256_setzero_ps();
1107
1108             /**************************
1109              * CALCULATE INTERACTIONS *
1110              **************************/
1111
1112             if (gmx_mm256_any_lt(rsq00,rcutoff2))
1113             {
1114
1115             r00              = _mm256_mul_ps(rsq00,rinv00);
1116             r00              = _mm256_andnot_ps(dummy_mask,r00);
1117
1118             /* Compute parameters for interactions between i and j atoms */
1119             qq00             = _mm256_mul_ps(iq0,jq0);
1120
1121             /* EWALD ELECTROSTATICS */
1122
1123             /* Analytical PME correction */
1124             zeta2            = _mm256_mul_ps(beta2,rsq00);
1125             rinv3            = _mm256_mul_ps(rinvsq00,rinv00);
1126             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
1127             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
1128             felec            = _mm256_mul_ps(qq00,felec);
1129
1130             cutoff_mask      = _mm256_cmp_ps(rsq00,rcutoff2,_CMP_LT_OQ);
1131
1132             fscal            = felec;
1133
1134             fscal            = _mm256_and_ps(fscal,cutoff_mask);
1135
1136             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
1137
1138             /* Calculate temporary vectorial force */
1139             tx               = _mm256_mul_ps(fscal,dx00);
1140             ty               = _mm256_mul_ps(fscal,dy00);
1141             tz               = _mm256_mul_ps(fscal,dz00);
1142
1143             /* Update vectorial force */
1144             fix0             = _mm256_add_ps(fix0,tx);
1145             fiy0             = _mm256_add_ps(fiy0,ty);
1146             fiz0             = _mm256_add_ps(fiz0,tz);
1147
1148             fjx0             = _mm256_add_ps(fjx0,tx);
1149             fjy0             = _mm256_add_ps(fjy0,ty);
1150             fjz0             = _mm256_add_ps(fjz0,tz);
1151
1152             }
1153
1154             /**************************
1155              * CALCULATE INTERACTIONS *
1156              **************************/
1157
1158             if (gmx_mm256_any_lt(rsq10,rcutoff2))
1159             {
1160
1161             r10              = _mm256_mul_ps(rsq10,rinv10);
1162             r10              = _mm256_andnot_ps(dummy_mask,r10);
1163
1164             /* Compute parameters for interactions between i and j atoms */
1165             qq10             = _mm256_mul_ps(iq1,jq0);
1166
1167             /* EWALD ELECTROSTATICS */
1168
1169             /* Analytical PME correction */
1170             zeta2            = _mm256_mul_ps(beta2,rsq10);
1171             rinv3            = _mm256_mul_ps(rinvsq10,rinv10);
1172             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
1173             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
1174             felec            = _mm256_mul_ps(qq10,felec);
1175
1176             cutoff_mask      = _mm256_cmp_ps(rsq10,rcutoff2,_CMP_LT_OQ);
1177
1178             fscal            = felec;
1179
1180             fscal            = _mm256_and_ps(fscal,cutoff_mask);
1181
1182             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
1183
1184             /* Calculate temporary vectorial force */
1185             tx               = _mm256_mul_ps(fscal,dx10);
1186             ty               = _mm256_mul_ps(fscal,dy10);
1187             tz               = _mm256_mul_ps(fscal,dz10);
1188
1189             /* Update vectorial force */
1190             fix1             = _mm256_add_ps(fix1,tx);
1191             fiy1             = _mm256_add_ps(fiy1,ty);
1192             fiz1             = _mm256_add_ps(fiz1,tz);
1193
1194             fjx0             = _mm256_add_ps(fjx0,tx);
1195             fjy0             = _mm256_add_ps(fjy0,ty);
1196             fjz0             = _mm256_add_ps(fjz0,tz);
1197
1198             }
1199
1200             /**************************
1201              * CALCULATE INTERACTIONS *
1202              **************************/
1203
1204             if (gmx_mm256_any_lt(rsq20,rcutoff2))
1205             {
1206
1207             r20              = _mm256_mul_ps(rsq20,rinv20);
1208             r20              = _mm256_andnot_ps(dummy_mask,r20);
1209
1210             /* Compute parameters for interactions between i and j atoms */
1211             qq20             = _mm256_mul_ps(iq2,jq0);
1212
1213             /* EWALD ELECTROSTATICS */
1214
1215             /* Analytical PME correction */
1216             zeta2            = _mm256_mul_ps(beta2,rsq20);
1217             rinv3            = _mm256_mul_ps(rinvsq20,rinv20);
1218             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
1219             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
1220             felec            = _mm256_mul_ps(qq20,felec);
1221
1222             cutoff_mask      = _mm256_cmp_ps(rsq20,rcutoff2,_CMP_LT_OQ);
1223
1224             fscal            = felec;
1225
1226             fscal            = _mm256_and_ps(fscal,cutoff_mask);
1227
1228             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
1229
1230             /* Calculate temporary vectorial force */
1231             tx               = _mm256_mul_ps(fscal,dx20);
1232             ty               = _mm256_mul_ps(fscal,dy20);
1233             tz               = _mm256_mul_ps(fscal,dz20);
1234
1235             /* Update vectorial force */
1236             fix2             = _mm256_add_ps(fix2,tx);
1237             fiy2             = _mm256_add_ps(fiy2,ty);
1238             fiz2             = _mm256_add_ps(fiz2,tz);
1239
1240             fjx0             = _mm256_add_ps(fjx0,tx);
1241             fjy0             = _mm256_add_ps(fjy0,ty);
1242             fjz0             = _mm256_add_ps(fjz0,tz);
1243
1244             }
1245
1246             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
1247             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
1248             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
1249             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
1250             fjptrE             = (jnrlistE>=0) ? f+j_coord_offsetE : scratch;
1251             fjptrF             = (jnrlistF>=0) ? f+j_coord_offsetF : scratch;
1252             fjptrG             = (jnrlistG>=0) ? f+j_coord_offsetG : scratch;
1253             fjptrH             = (jnrlistH>=0) ? f+j_coord_offsetH : scratch;
1254
1255             gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,fjx0,fjy0,fjz0);
1256
1257             /* Inner loop uses 183 flops */
1258         }
1259
1260         /* End of innermost loop */
1261
1262         gmx_mm256_update_iforce_3atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
1263                                                  f+i_coord_offset,fshift+i_shift_offset);
1264
1265         /* Increment number of inner iterations */
1266         inneriter                  += j_index_end - j_index_start;
1267
1268         /* Outer loop uses 18 flops */
1269     }
1270
1271     /* Increment number of outer iterations */
1272     outeriter        += nri;
1273
1274     /* Update outer/inner flops */
1275
1276     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_W3_F,outeriter*18 + inneriter*183);
1277 }