src/gmxlib/nonbonded/nb_kernel_avx_256_single/nb_kernel_ElecEw_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_ElecEw_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_ElecEw_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     /* Avoid stupid compiler warnings */
 124     jnrA = jnrB = jnrC = jnrD = jnrE = jnrF = jnrG = jnrH = 0;
 125     j_coord_offsetA = 0;
 126     j_coord_offsetB = 0;
 127     j_coord_offsetC = 0;
 128     j_coord_offsetD = 0;
 129     j_coord_offsetE = 0;
 130     j_coord_offsetF = 0;
 131     j_coord_offsetG = 0;
 132     j_coord_offsetH = 0;
 133
 134     outeriter        = 0;
 135     inneriter        = 0;
 136
 137     for(iidx=0;iidx<4*DIM;iidx++)
 138     {
 139         scratch[iidx] = 0.0;
 140     }
 141
 142     /* Start outer loop over neighborlists */
 143     for(iidx=0; iidx<nri; iidx++)
 144     {
 145         /* Load shift vector for this list */
 146         i_shift_offset   = DIM*shiftidx[iidx];
 147
 148         /* Load limits for loop over neighbors */
 149         j_index_start    = jindex[iidx];
 150         j_index_end      = jindex[iidx+1];
 151
 152         /* Get outer coordinate index */
 153         inr              = iinr[iidx];
 154         i_coord_offset   = DIM*inr;
 155
 156         /* Load i particle coords and add shift vector */
 157         gmx_mm256_load_shift_and_3rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
 158                                                     &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 159
 160         fix0             = _mm256_setzero_ps();
 161         fiy0             = _mm256_setzero_ps();
 162         fiz0             = _mm256_setzero_ps();
 163         fix1             = _mm256_setzero_ps();
 164         fiy1             = _mm256_setzero_ps();
 165         fiz1             = _mm256_setzero_ps();
 166         fix2             = _mm256_setzero_ps();
 167         fiy2             = _mm256_setzero_ps();
 168         fiz2             = _mm256_setzero_ps();
 169
 170         /* Reset potential sums */
 171         velecsum         = _mm256_setzero_ps();
 172
 173         /* Start inner kernel loop */
 174         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+7]>=0; jidx+=8)
 175         {
 176
 177             /* Get j neighbor index, and coordinate index */
 178             jnrA             = jjnr[jidx];
 179             jnrB             = jjnr[jidx+1];
 180             jnrC             = jjnr[jidx+2];
 181             jnrD             = jjnr[jidx+3];
 182             jnrE             = jjnr[jidx+4];
 183             jnrF             = jjnr[jidx+5];
 184             jnrG             = jjnr[jidx+6];
 185             jnrH             = jjnr[jidx+7];
 186             j_coord_offsetA  = DIM*jnrA;
 187             j_coord_offsetB  = DIM*jnrB;
 188             j_coord_offsetC  = DIM*jnrC;
 189             j_coord_offsetD  = DIM*jnrD;
 190             j_coord_offsetE  = DIM*jnrE;
 191             j_coord_offsetF  = DIM*jnrF;
 192             j_coord_offsetG  = DIM*jnrG;
 193             j_coord_offsetH  = DIM*jnrH;
 194
 195             /* load j atom coordinates */
 196             gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 197                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 198                                                  x+j_coord_offsetE,x+j_coord_offsetF,
 199                                                  x+j_coord_offsetG,x+j_coord_offsetH,
 200                                                  &jx0,&jy0,&jz0);
 201
 202             /* Calculate displacement vector */
 203             dx00             = _mm256_sub_ps(ix0,jx0);
 204             dy00             = _mm256_sub_ps(iy0,jy0);
 205             dz00             = _mm256_sub_ps(iz0,jz0);
 206             dx10             = _mm256_sub_ps(ix1,jx0);
 207             dy10             = _mm256_sub_ps(iy1,jy0);
 208             dz10             = _mm256_sub_ps(iz1,jz0);
 209             dx20             = _mm256_sub_ps(ix2,jx0);
 210             dy20             = _mm256_sub_ps(iy2,jy0);
 211             dz20             = _mm256_sub_ps(iz2,jz0);
 212
 213             /* Calculate squared distance and things based on it */
 214             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
 215             rsq10            = gmx_mm256_calc_rsq_ps(dx10,dy10,dz10);
 216             rsq20            = gmx_mm256_calc_rsq_ps(dx20,dy20,dz20);
 217
 218             rinv00           = gmx_mm256_invsqrt_ps(rsq00);
 219             rinv10           = gmx_mm256_invsqrt_ps(rsq10);
 220             rinv20           = gmx_mm256_invsqrt_ps(rsq20);
 221
 222             rinvsq00         = _mm256_mul_ps(rinv00,rinv00);
 223             rinvsq10         = _mm256_mul_ps(rinv10,rinv10);
 224             rinvsq20         = _mm256_mul_ps(rinv20,rinv20);
 225
 226             /* Load parameters for j particles */
 227             jq0              = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 228                                                                  charge+jnrC+0,charge+jnrD+0,
 229                                                                  charge+jnrE+0,charge+jnrF+0,
 230                                                                  charge+jnrG+0,charge+jnrH+0);
 231
 232             fjx0             = _mm256_setzero_ps();
 233             fjy0             = _mm256_setzero_ps();
 234             fjz0             = _mm256_setzero_ps();
 235
 236             /**************************
 237              * CALCULATE INTERACTIONS *
 238              **************************/
 239
 240             r00              = _mm256_mul_ps(rsq00,rinv00);
 241
 242             /* Compute parameters for interactions between i and j atoms */
 243             qq00             = _mm256_mul_ps(iq0,jq0);
 244
 245             /* EWALD ELECTROSTATICS */
 246
 247             /* Analytical PME correction */
 248             zeta2            = _mm256_mul_ps(beta2,rsq00);
 249             rinv3            = _mm256_mul_ps(rinvsq00,rinv00);
 250             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
 251             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
 252             felec            = _mm256_mul_ps(qq00,felec);
 253             pmecorrV         = gmx_mm256_pmecorrV_ps(zeta2);
 254             pmecorrV         = _mm256_mul_ps(pmecorrV,beta);
 255             velec            = _mm256_sub_ps(rinv00,pmecorrV);
 256             velec            = _mm256_mul_ps(qq00,velec);
 257
 258             /* Update potential sum for this i atom from the interaction with this j atom. */
 259             velecsum         = _mm256_add_ps(velecsum,velec);
 260
 261             fscal            = felec;
 262
 263             /* Calculate temporary vectorial force */
 264             tx               = _mm256_mul_ps(fscal,dx00);
 265             ty               = _mm256_mul_ps(fscal,dy00);
 266             tz               = _mm256_mul_ps(fscal,dz00);
 267
 268             /* Update vectorial force */
 269             fix0             = _mm256_add_ps(fix0,tx);
 270             fiy0             = _mm256_add_ps(fiy0,ty);
 271             fiz0             = _mm256_add_ps(fiz0,tz);
 272
 273             fjx0             = _mm256_add_ps(fjx0,tx);
 274             fjy0             = _mm256_add_ps(fjy0,ty);
 275             fjz0             = _mm256_add_ps(fjz0,tz);
 276
 277             /**************************
 278              * CALCULATE INTERACTIONS *
 279              **************************/
 280
 281             r10              = _mm256_mul_ps(rsq10,rinv10);
 282
 283             /* Compute parameters for interactions between i and j atoms */
 284             qq10             = _mm256_mul_ps(iq1,jq0);
 285
 286             /* EWALD ELECTROSTATICS */
 287
 288             /* Analytical PME correction */
 289             zeta2            = _mm256_mul_ps(beta2,rsq10);
 290             rinv3            = _mm256_mul_ps(rinvsq10,rinv10);
 291             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
 292             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
 293             felec            = _mm256_mul_ps(qq10,felec);
 294             pmecorrV         = gmx_mm256_pmecorrV_ps(zeta2);
 295             pmecorrV         = _mm256_mul_ps(pmecorrV,beta);
 296             velec            = _mm256_sub_ps(rinv10,pmecorrV);
 297             velec            = _mm256_mul_ps(qq10,velec);
 298
 299             /* Update potential sum for this i atom from the interaction with this j atom. */
 300             velecsum         = _mm256_add_ps(velecsum,velec);
 301
 302             fscal            = felec;
 303
 304             /* Calculate temporary vectorial force */
 305             tx               = _mm256_mul_ps(fscal,dx10);
 306             ty               = _mm256_mul_ps(fscal,dy10);
 307             tz               = _mm256_mul_ps(fscal,dz10);
 308
 309             /* Update vectorial force */
 310             fix1             = _mm256_add_ps(fix1,tx);
 311             fiy1             = _mm256_add_ps(fiy1,ty);
 312             fiz1             = _mm256_add_ps(fiz1,tz);
 313
 314             fjx0             = _mm256_add_ps(fjx0,tx);
 315             fjy0             = _mm256_add_ps(fjy0,ty);
 316             fjz0             = _mm256_add_ps(fjz0,tz);
 317
 318             /**************************
 319              * CALCULATE INTERACTIONS *
 320              **************************/
 321
 322             r20              = _mm256_mul_ps(rsq20,rinv20);
 323
 324             /* Compute parameters for interactions between i and j atoms */
 325             qq20             = _mm256_mul_ps(iq2,jq0);
 326
 327             /* EWALD ELECTROSTATICS */
 328
 329             /* Analytical PME correction */
 330             zeta2            = _mm256_mul_ps(beta2,rsq20);
 331             rinv3            = _mm256_mul_ps(rinvsq20,rinv20);
 332             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
 333             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
 334             felec            = _mm256_mul_ps(qq20,felec);
 335             pmecorrV         = gmx_mm256_pmecorrV_ps(zeta2);
 336             pmecorrV         = _mm256_mul_ps(pmecorrV,beta);
 337             velec            = _mm256_sub_ps(rinv20,pmecorrV);
 338             velec            = _mm256_mul_ps(qq20,velec);
 339
 340             /* Update potential sum for this i atom from the interaction with this j atom. */
 341             velecsum         = _mm256_add_ps(velecsum,velec);
 342
 343             fscal            = felec;
 344
 345             /* Calculate temporary vectorial force */
 346             tx               = _mm256_mul_ps(fscal,dx20);
 347             ty               = _mm256_mul_ps(fscal,dy20);
 348             tz               = _mm256_mul_ps(fscal,dz20);
 349
 350             /* Update vectorial force */
 351             fix2             = _mm256_add_ps(fix2,tx);
 352             fiy2             = _mm256_add_ps(fiy2,ty);
 353             fiz2             = _mm256_add_ps(fiz2,tz);
 354
 355             fjx0             = _mm256_add_ps(fjx0,tx);
 356             fjy0             = _mm256_add_ps(fjy0,ty);
 357             fjz0             = _mm256_add_ps(fjz0,tz);
 358
 359             fjptrA             = f+j_coord_offsetA;
 360             fjptrB             = f+j_coord_offsetB;
 361             fjptrC             = f+j_coord_offsetC;
 362             fjptrD             = f+j_coord_offsetD;
 363             fjptrE             = f+j_coord_offsetE;
 364             fjptrF             = f+j_coord_offsetF;
 365             fjptrG             = f+j_coord_offsetG;
 366             fjptrH             = f+j_coord_offsetH;
 367
 368             gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,fjx0,fjy0,fjz0);
 369
 370             /* Inner loop uses 255 flops */
 371         }
 372
 373         if(jidx<j_index_end)
 374         {
 375
 376             /* Get j neighbor index, and coordinate index */
 377             jnrlistA         = jjnr[jidx];
 378             jnrlistB         = jjnr[jidx+1];
 379             jnrlistC         = jjnr[jidx+2];
 380             jnrlistD         = jjnr[jidx+3];
 381             jnrlistE         = jjnr[jidx+4];
 382             jnrlistF         = jjnr[jidx+5];
 383             jnrlistG         = jjnr[jidx+6];
 384             jnrlistH         = jjnr[jidx+7];
 385             /* Sign of each element will be negative for non-real atoms.
 386              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 387              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 388              */
 389             dummy_mask = gmx_mm256_set_m128(gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx+4)),_mm_setzero_si128())),
 390                                             gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128())));
 391
 392             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 393             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 394             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 395             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 396             jnrE       = (jnrlistE>=0) ? jnrlistE : 0;
 397             jnrF       = (jnrlistF>=0) ? jnrlistF : 0;
 398             jnrG       = (jnrlistG>=0) ? jnrlistG : 0;
 399             jnrH       = (jnrlistH>=0) ? jnrlistH : 0;
 400             j_coord_offsetA  = DIM*jnrA;
 401             j_coord_offsetB  = DIM*jnrB;
 402             j_coord_offsetC  = DIM*jnrC;
 403             j_coord_offsetD  = DIM*jnrD;
 404             j_coord_offsetE  = DIM*jnrE;
 405             j_coord_offsetF  = DIM*jnrF;
 406             j_coord_offsetG  = DIM*jnrG;
 407             j_coord_offsetH  = DIM*jnrH;
 408
 409             /* load j atom coordinates */
 410             gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 411                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 412                                                  x+j_coord_offsetE,x+j_coord_offsetF,
 413                                                  x+j_coord_offsetG,x+j_coord_offsetH,
 414                                                  &jx0,&jy0,&jz0);
 415
 416             /* Calculate displacement vector */
 417             dx00             = _mm256_sub_ps(ix0,jx0);
 418             dy00             = _mm256_sub_ps(iy0,jy0);
 419             dz00             = _mm256_sub_ps(iz0,jz0);
 420             dx10             = _mm256_sub_ps(ix1,jx0);
 421             dy10             = _mm256_sub_ps(iy1,jy0);
 422             dz10             = _mm256_sub_ps(iz1,jz0);
 423             dx20             = _mm256_sub_ps(ix2,jx0);
 424             dy20             = _mm256_sub_ps(iy2,jy0);
 425             dz20             = _mm256_sub_ps(iz2,jz0);
 426
 427             /* Calculate squared distance and things based on it */
 428             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
 429             rsq10            = gmx_mm256_calc_rsq_ps(dx10,dy10,dz10);
 430             rsq20            = gmx_mm256_calc_rsq_ps(dx20,dy20,dz20);
 431
 432             rinv00           = gmx_mm256_invsqrt_ps(rsq00);
 433             rinv10           = gmx_mm256_invsqrt_ps(rsq10);
 434             rinv20           = gmx_mm256_invsqrt_ps(rsq20);
 435
 436             rinvsq00         = _mm256_mul_ps(rinv00,rinv00);
 437             rinvsq10         = _mm256_mul_ps(rinv10,rinv10);
 438             rinvsq20         = _mm256_mul_ps(rinv20,rinv20);
 439
 440             /* Load parameters for j particles */
 441             jq0              = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 442                                                                  charge+jnrC+0,charge+jnrD+0,
 443                                                                  charge+jnrE+0,charge+jnrF+0,
 444                                                                  charge+jnrG+0,charge+jnrH+0);
 445
 446             fjx0             = _mm256_setzero_ps();
 447             fjy0             = _mm256_setzero_ps();
 448             fjz0             = _mm256_setzero_ps();
 449
 450             /**************************
 451              * CALCULATE INTERACTIONS *
 452              **************************/
 453
 454             r00              = _mm256_mul_ps(rsq00,rinv00);
 455             r00              = _mm256_andnot_ps(dummy_mask,r00);
 456
 457             /* Compute parameters for interactions between i and j atoms */
 458             qq00             = _mm256_mul_ps(iq0,jq0);
 459
 460             /* EWALD ELECTROSTATICS */
 461
 462             /* Analytical PME correction */
 463             zeta2            = _mm256_mul_ps(beta2,rsq00);
 464             rinv3            = _mm256_mul_ps(rinvsq00,rinv00);
 465             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
 466             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
 467             felec            = _mm256_mul_ps(qq00,felec);
 468             pmecorrV         = gmx_mm256_pmecorrV_ps(zeta2);
 469             pmecorrV         = _mm256_mul_ps(pmecorrV,beta);
 470             velec            = _mm256_sub_ps(rinv00,pmecorrV);
 471             velec            = _mm256_mul_ps(qq00,velec);
 472
 473             /* Update potential sum for this i atom from the interaction with this j atom. */
 474             velec            = _mm256_andnot_ps(dummy_mask,velec);
 475             velecsum         = _mm256_add_ps(velecsum,velec);
 476
 477             fscal            = felec;
 478
 479             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
 480
 481             /* Calculate temporary vectorial force */
 482             tx               = _mm256_mul_ps(fscal,dx00);
 483             ty               = _mm256_mul_ps(fscal,dy00);
 484             tz               = _mm256_mul_ps(fscal,dz00);
 485
 486             /* Update vectorial force */
 487             fix0             = _mm256_add_ps(fix0,tx);
 488             fiy0             = _mm256_add_ps(fiy0,ty);
 489             fiz0             = _mm256_add_ps(fiz0,tz);
 490
 491             fjx0             = _mm256_add_ps(fjx0,tx);
 492             fjy0             = _mm256_add_ps(fjy0,ty);
 493             fjz0             = _mm256_add_ps(fjz0,tz);
 494
 495             /**************************
 496              * CALCULATE INTERACTIONS *
 497              **************************/
 498
 499             r10              = _mm256_mul_ps(rsq10,rinv10);
 500             r10              = _mm256_andnot_ps(dummy_mask,r10);
 501
 502             /* Compute parameters for interactions between i and j atoms */
 503             qq10             = _mm256_mul_ps(iq1,jq0);
 504
 505             /* EWALD ELECTROSTATICS */
 506
 507             /* Analytical PME correction */
 508             zeta2            = _mm256_mul_ps(beta2,rsq10);
 509             rinv3            = _mm256_mul_ps(rinvsq10,rinv10);
 510             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
 511             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
 512             felec            = _mm256_mul_ps(qq10,felec);
 513             pmecorrV         = gmx_mm256_pmecorrV_ps(zeta2);
 514             pmecorrV         = _mm256_mul_ps(pmecorrV,beta);
 515             velec            = _mm256_sub_ps(rinv10,pmecorrV);
 516             velec            = _mm256_mul_ps(qq10,velec);
 517
 518             /* Update potential sum for this i atom from the interaction with this j atom. */
 519             velec            = _mm256_andnot_ps(dummy_mask,velec);
 520             velecsum         = _mm256_add_ps(velecsum,velec);
 521
 522             fscal            = felec;
 523
 524             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
 525
 526             /* Calculate temporary vectorial force */
 527             tx               = _mm256_mul_ps(fscal,dx10);
 528             ty               = _mm256_mul_ps(fscal,dy10);
 529             tz               = _mm256_mul_ps(fscal,dz10);
 530
 531             /* Update vectorial force */
 532             fix1             = _mm256_add_ps(fix1,tx);
 533             fiy1             = _mm256_add_ps(fiy1,ty);
 534             fiz1             = _mm256_add_ps(fiz1,tz);
 535
 536             fjx0             = _mm256_add_ps(fjx0,tx);
 537             fjy0             = _mm256_add_ps(fjy0,ty);
 538             fjz0             = _mm256_add_ps(fjz0,tz);
 539
 540             /**************************
 541              * CALCULATE INTERACTIONS *
 542              **************************/
 543
 544             r20              = _mm256_mul_ps(rsq20,rinv20);
 545             r20              = _mm256_andnot_ps(dummy_mask,r20);
 546
 547             /* Compute parameters for interactions between i and j atoms */
 548             qq20             = _mm256_mul_ps(iq2,jq0);
 549
 550             /* EWALD ELECTROSTATICS */
 551
 552             /* Analytical PME correction */
 553             zeta2            = _mm256_mul_ps(beta2,rsq20);
 554             rinv3            = _mm256_mul_ps(rinvsq20,rinv20);
 555             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
 556             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
 557             felec            = _mm256_mul_ps(qq20,felec);
 558             pmecorrV         = gmx_mm256_pmecorrV_ps(zeta2);
 559             pmecorrV         = _mm256_mul_ps(pmecorrV,beta);
 560             velec            = _mm256_sub_ps(rinv20,pmecorrV);
 561             velec            = _mm256_mul_ps(qq20,velec);
 562
 563             /* Update potential sum for this i atom from the interaction with this j atom. */
 564             velec            = _mm256_andnot_ps(dummy_mask,velec);
 565             velecsum         = _mm256_add_ps(velecsum,velec);
 566
 567             fscal            = felec;
 568
 569             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
 570
 571             /* Calculate temporary vectorial force */
 572             tx               = _mm256_mul_ps(fscal,dx20);
 573             ty               = _mm256_mul_ps(fscal,dy20);
 574             tz               = _mm256_mul_ps(fscal,dz20);
 575
 576             /* Update vectorial force */
 577             fix2             = _mm256_add_ps(fix2,tx);
 578             fiy2             = _mm256_add_ps(fiy2,ty);
 579             fiz2             = _mm256_add_ps(fiz2,tz);
 580
 581             fjx0             = _mm256_add_ps(fjx0,tx);
 582             fjy0             = _mm256_add_ps(fjy0,ty);
 583             fjz0             = _mm256_add_ps(fjz0,tz);
 584
 585             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 586             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 587             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 588             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 589             fjptrE             = (jnrlistE>=0) ? f+j_coord_offsetE : scratch;
 590             fjptrF             = (jnrlistF>=0) ? f+j_coord_offsetF : scratch;
 591             fjptrG             = (jnrlistG>=0) ? f+j_coord_offsetG : scratch;
 592             fjptrH             = (jnrlistH>=0) ? f+j_coord_offsetH : scratch;
 593
 594             gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,fjx0,fjy0,fjz0);
 595
 596             /* Inner loop uses 258 flops */
 597         }
 598
 599         /* End of innermost loop */
 600
 601         gmx_mm256_update_iforce_3atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
 602                                                  f+i_coord_offset,fshift+i_shift_offset);
 603
 604         ggid                        = gid[iidx];
 605         /* Update potential energies */
 606         gmx_mm256_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid);
 607
 608         /* Increment number of inner iterations */
 609         inneriter                  += j_index_end - j_index_start;
 610
 611         /* Outer loop uses 19 flops */
 612     }
 613
 614     /* Increment number of outer iterations */
 615     outeriter        += nri;
 616
 617     /* Update outer/inner flops */
 618
 619     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_W3_VF,outeriter*19 + inneriter*258);
 620 }
 621 /*
 622  * Gromacs nonbonded kernel:   nb_kernel_ElecEw_VdwNone_GeomW3P1_F_avx_256_single
 623  * Electrostatics interaction: Ewald
 624  * VdW interaction:            None
 625  * Geometry:                   Water3-Particle
 626  * Calculate force/pot:        Force
 627  */
 628 void
 629 nb_kernel_ElecEw_VdwNone_GeomW3P1_F_avx_256_single
 630                     (t_nblist * gmx_restrict                nlist,
 631                      rvec * gmx_restrict                    xx,
 632                      rvec * gmx_restrict                    ff,
 633                      t_forcerec * gmx_restrict              fr,
 634                      t_mdatoms * gmx_restrict               mdatoms,
 635                      nb_kernel_data_t * gmx_restrict        kernel_data,
 636                      t_nrnb * gmx_restrict                  nrnb)
 637 {
 638     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 639      * just 0 for non-waters.
 640      * Suffixes A,B,C,D,E,F,G,H refer to j loop unrolling done with AVX, e.g. for the eight different
 641      * jnr indices corresponding to data put in the four positions in the SIMD register.
 642      */
 643     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 644     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 645     int              jnrA,jnrB,jnrC,jnrD;
 646     int              jnrE,jnrF,jnrG,jnrH;
 647     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
 648     int              jnrlistE,jnrlistF,jnrlistG,jnrlistH;
 649     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 650     int              j_coord_offsetE,j_coord_offsetF,j_coord_offsetG,j_coord_offsetH;
 651     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 652     real             rcutoff_scalar;
 653     real             *shiftvec,*fshift,*x,*f;
 654     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD,*fjptrE,*fjptrF,*fjptrG,*fjptrH;
 655     real             scratch[4*DIM];
 656     __m256           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 657     real *           vdwioffsetptr0;
 658     __m256           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 659     real *           vdwioffsetptr1;
 660     __m256           ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
 661     real *           vdwioffsetptr2;
 662     __m256           ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
 663     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D,vdwjidx0E,vdwjidx0F,vdwjidx0G,vdwjidx0H;
 664     __m256           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 665     __m256           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 666     __m256           dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
 667     __m256           dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
 668     __m256           velec,felec,velecsum,facel,crf,krf,krf2;
 669     real             *charge;
 670     __m256i          ewitab;
 671     __m128i          ewitab_lo,ewitab_hi;
 672     __m256           ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
 673     __m256           beta,beta2,beta3,zeta2,pmecorrF,pmecorrV,rinv3;
 674     real             *ewtab;
 675     __m256           dummy_mask,cutoff_mask;
 676     __m256           signbit = _mm256_castsi256_ps( _mm256_set1_epi32(0x80000000) );
 677     __m256           one     = _mm256_set1_ps(1.0);
 678     __m256           two     = _mm256_set1_ps(2.0);
 679     x                = xx[0];
 680     f                = ff[0];
 681
 682     nri              = nlist->nri;
 683     iinr             = nlist->iinr;
 684     jindex           = nlist->jindex;
 685     jjnr             = nlist->jjnr;
 686     shiftidx         = nlist->shift;
 687     gid              = nlist->gid;
 688     shiftvec         = fr->shift_vec[0];
 689     fshift           = fr->fshift[0];
 690     facel            = _mm256_set1_ps(fr->epsfac);
 691     charge           = mdatoms->chargeA;
 692
 693     sh_ewald         = _mm256_set1_ps(fr->ic->sh_ewald);
 694     beta             = _mm256_set1_ps(fr->ic->ewaldcoeff);
 695     beta2            = _mm256_mul_ps(beta,beta);
 696     beta3            = _mm256_mul_ps(beta,beta2);
 697
 698     ewtab            = fr->ic->tabq_coul_F;
 699     ewtabscale       = _mm256_set1_ps(fr->ic->tabq_scale);
 700     ewtabhalfspace   = _mm256_set1_ps(0.5/fr->ic->tabq_scale);
 701
 702     /* Setup water-specific parameters */
 703     inr              = nlist->iinr[0];
 704     iq0              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+0]));
 705     iq1              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+1]));
 706     iq2              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+2]));
 707
 708     /* Avoid stupid compiler warnings */
 709     jnrA = jnrB = jnrC = jnrD = jnrE = jnrF = jnrG = jnrH = 0;
 710     j_coord_offsetA = 0;
 711     j_coord_offsetB = 0;
 712     j_coord_offsetC = 0;
 713     j_coord_offsetD = 0;
 714     j_coord_offsetE = 0;
 715     j_coord_offsetF = 0;
 716     j_coord_offsetG = 0;
 717     j_coord_offsetH = 0;
 718
 719     outeriter        = 0;
 720     inneriter        = 0;
 721
 722     for(iidx=0;iidx<4*DIM;iidx++)
 723     {
 724         scratch[iidx] = 0.0;
 725     }
 726
 727     /* Start outer loop over neighborlists */
 728     for(iidx=0; iidx<nri; iidx++)
 729     {
 730         /* Load shift vector for this list */
 731         i_shift_offset   = DIM*shiftidx[iidx];
 732
 733         /* Load limits for loop over neighbors */
 734         j_index_start    = jindex[iidx];
 735         j_index_end      = jindex[iidx+1];
 736
 737         /* Get outer coordinate index */
 738         inr              = iinr[iidx];
 739         i_coord_offset   = DIM*inr;
 740
 741         /* Load i particle coords and add shift vector */
 742         gmx_mm256_load_shift_and_3rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
 743                                                     &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 744
 745         fix0             = _mm256_setzero_ps();
 746         fiy0             = _mm256_setzero_ps();
 747         fiz0             = _mm256_setzero_ps();
 748         fix1             = _mm256_setzero_ps();
 749         fiy1             = _mm256_setzero_ps();
 750         fiz1             = _mm256_setzero_ps();
 751         fix2             = _mm256_setzero_ps();
 752         fiy2             = _mm256_setzero_ps();
 753         fiz2             = _mm256_setzero_ps();
 754
 755         /* Start inner kernel loop */
 756         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+7]>=0; jidx+=8)
 757         {
 758
 759             /* Get j neighbor index, and coordinate index */
 760             jnrA             = jjnr[jidx];
 761             jnrB             = jjnr[jidx+1];
 762             jnrC             = jjnr[jidx+2];
 763             jnrD             = jjnr[jidx+3];
 764             jnrE             = jjnr[jidx+4];
 765             jnrF             = jjnr[jidx+5];
 766             jnrG             = jjnr[jidx+6];
 767             jnrH             = jjnr[jidx+7];
 768             j_coord_offsetA  = DIM*jnrA;
 769             j_coord_offsetB  = DIM*jnrB;
 770             j_coord_offsetC  = DIM*jnrC;
 771             j_coord_offsetD  = DIM*jnrD;
 772             j_coord_offsetE  = DIM*jnrE;
 773             j_coord_offsetF  = DIM*jnrF;
 774             j_coord_offsetG  = DIM*jnrG;
 775             j_coord_offsetH  = DIM*jnrH;
 776
 777             /* load j atom coordinates */
 778             gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 779                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 780                                                  x+j_coord_offsetE,x+j_coord_offsetF,
 781                                                  x+j_coord_offsetG,x+j_coord_offsetH,
 782                                                  &jx0,&jy0,&jz0);
 783
 784             /* Calculate displacement vector */
 785             dx00             = _mm256_sub_ps(ix0,jx0);
 786             dy00             = _mm256_sub_ps(iy0,jy0);
 787             dz00             = _mm256_sub_ps(iz0,jz0);
 788             dx10             = _mm256_sub_ps(ix1,jx0);
 789             dy10             = _mm256_sub_ps(iy1,jy0);
 790             dz10             = _mm256_sub_ps(iz1,jz0);
 791             dx20             = _mm256_sub_ps(ix2,jx0);
 792             dy20             = _mm256_sub_ps(iy2,jy0);
 793             dz20             = _mm256_sub_ps(iz2,jz0);
 794
 795             /* Calculate squared distance and things based on it */
 796             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
 797             rsq10            = gmx_mm256_calc_rsq_ps(dx10,dy10,dz10);
 798             rsq20            = gmx_mm256_calc_rsq_ps(dx20,dy20,dz20);
 799
 800             rinv00           = gmx_mm256_invsqrt_ps(rsq00);
 801             rinv10           = gmx_mm256_invsqrt_ps(rsq10);
 802             rinv20           = gmx_mm256_invsqrt_ps(rsq20);
 803
 804             rinvsq00         = _mm256_mul_ps(rinv00,rinv00);
 805             rinvsq10         = _mm256_mul_ps(rinv10,rinv10);
 806             rinvsq20         = _mm256_mul_ps(rinv20,rinv20);
 807
 808             /* Load parameters for j particles */
 809             jq0              = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 810                                                                  charge+jnrC+0,charge+jnrD+0,
 811                                                                  charge+jnrE+0,charge+jnrF+0,
 812                                                                  charge+jnrG+0,charge+jnrH+0);
 813
 814             fjx0             = _mm256_setzero_ps();
 815             fjy0             = _mm256_setzero_ps();
 816             fjz0             = _mm256_setzero_ps();
 817
 818             /**************************
 819              * CALCULATE INTERACTIONS *
 820              **************************/
 821
 822             r00              = _mm256_mul_ps(rsq00,rinv00);
 823
 824             /* Compute parameters for interactions between i and j atoms */
 825             qq00             = _mm256_mul_ps(iq0,jq0);
 826
 827             /* EWALD ELECTROSTATICS */
 828
 829             /* Analytical PME correction */
 830             zeta2            = _mm256_mul_ps(beta2,rsq00);
 831             rinv3            = _mm256_mul_ps(rinvsq00,rinv00);
 832             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
 833             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
 834             felec            = _mm256_mul_ps(qq00,felec);
 835
 836             fscal            = felec;
 837
 838             /* Calculate temporary vectorial force */
 839             tx               = _mm256_mul_ps(fscal,dx00);
 840             ty               = _mm256_mul_ps(fscal,dy00);
 841             tz               = _mm256_mul_ps(fscal,dz00);
 842
 843             /* Update vectorial force */
 844             fix0             = _mm256_add_ps(fix0,tx);
 845             fiy0             = _mm256_add_ps(fiy0,ty);
 846             fiz0             = _mm256_add_ps(fiz0,tz);
 847
 848             fjx0             = _mm256_add_ps(fjx0,tx);
 849             fjy0             = _mm256_add_ps(fjy0,ty);
 850             fjz0             = _mm256_add_ps(fjz0,tz);
 851
 852             /**************************
 853              * CALCULATE INTERACTIONS *
 854              **************************/
 855
 856             r10              = _mm256_mul_ps(rsq10,rinv10);
 857
 858             /* Compute parameters for interactions between i and j atoms */
 859             qq10             = _mm256_mul_ps(iq1,jq0);
 860
 861             /* EWALD ELECTROSTATICS */
 862
 863             /* Analytical PME correction */
 864             zeta2            = _mm256_mul_ps(beta2,rsq10);
 865             rinv3            = _mm256_mul_ps(rinvsq10,rinv10);
 866             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
 867             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
 868             felec            = _mm256_mul_ps(qq10,felec);
 869
 870             fscal            = felec;
 871
 872             /* Calculate temporary vectorial force */
 873             tx               = _mm256_mul_ps(fscal,dx10);
 874             ty               = _mm256_mul_ps(fscal,dy10);
 875             tz               = _mm256_mul_ps(fscal,dz10);
 876
 877             /* Update vectorial force */
 878             fix1             = _mm256_add_ps(fix1,tx);
 879             fiy1             = _mm256_add_ps(fiy1,ty);
 880             fiz1             = _mm256_add_ps(fiz1,tz);
 881
 882             fjx0             = _mm256_add_ps(fjx0,tx);
 883             fjy0             = _mm256_add_ps(fjy0,ty);
 884             fjz0             = _mm256_add_ps(fjz0,tz);
 885
 886             /**************************
 887              * CALCULATE INTERACTIONS *
 888              **************************/
 889
 890             r20              = _mm256_mul_ps(rsq20,rinv20);
 891
 892             /* Compute parameters for interactions between i and j atoms */
 893             qq20             = _mm256_mul_ps(iq2,jq0);
 894
 895             /* EWALD ELECTROSTATICS */
 896
 897             /* Analytical PME correction */
 898             zeta2            = _mm256_mul_ps(beta2,rsq20);
 899             rinv3            = _mm256_mul_ps(rinvsq20,rinv20);
 900             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
 901             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
 902             felec            = _mm256_mul_ps(qq20,felec);
 903
 904             fscal            = felec;
 905
 906             /* Calculate temporary vectorial force */
 907             tx               = _mm256_mul_ps(fscal,dx20);
 908             ty               = _mm256_mul_ps(fscal,dy20);
 909             tz               = _mm256_mul_ps(fscal,dz20);
 910
 911             /* Update vectorial force */
 912             fix2             = _mm256_add_ps(fix2,tx);
 913             fiy2             = _mm256_add_ps(fiy2,ty);
 914             fiz2             = _mm256_add_ps(fiz2,tz);
 915
 916             fjx0             = _mm256_add_ps(fjx0,tx);
 917             fjy0             = _mm256_add_ps(fjy0,ty);
 918             fjz0             = _mm256_add_ps(fjz0,tz);
 919
 920             fjptrA             = f+j_coord_offsetA;
 921             fjptrB             = f+j_coord_offsetB;
 922             fjptrC             = f+j_coord_offsetC;
 923             fjptrD             = f+j_coord_offsetD;
 924             fjptrE             = f+j_coord_offsetE;
 925             fjptrF             = f+j_coord_offsetF;
 926             fjptrG             = f+j_coord_offsetG;
 927             fjptrH             = f+j_coord_offsetH;
 928
 929             gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,fjx0,fjy0,fjz0);
 930
 931             /* Inner loop uses 171 flops */
 932         }
 933
 934         if(jidx<j_index_end)
 935         {
 936
 937             /* Get j neighbor index, and coordinate index */
 938             jnrlistA         = jjnr[jidx];
 939             jnrlistB         = jjnr[jidx+1];
 940             jnrlistC         = jjnr[jidx+2];
 941             jnrlistD         = jjnr[jidx+3];
 942             jnrlistE         = jjnr[jidx+4];
 943             jnrlistF         = jjnr[jidx+5];
 944             jnrlistG         = jjnr[jidx+6];
 945             jnrlistH         = jjnr[jidx+7];
 946             /* Sign of each element will be negative for non-real atoms.
 947              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 948              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 949              */
 950             dummy_mask = gmx_mm256_set_m128(gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx+4)),_mm_setzero_si128())),
 951                                             gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128())));
 952
 953             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 954             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 955             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 956             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 957             jnrE       = (jnrlistE>=0) ? jnrlistE : 0;
 958             jnrF       = (jnrlistF>=0) ? jnrlistF : 0;
 959             jnrG       = (jnrlistG>=0) ? jnrlistG : 0;
 960             jnrH       = (jnrlistH>=0) ? jnrlistH : 0;
 961             j_coord_offsetA  = DIM*jnrA;
 962             j_coord_offsetB  = DIM*jnrB;
 963             j_coord_offsetC  = DIM*jnrC;
 964             j_coord_offsetD  = DIM*jnrD;
 965             j_coord_offsetE  = DIM*jnrE;
 966             j_coord_offsetF  = DIM*jnrF;
 967             j_coord_offsetG  = DIM*jnrG;
 968             j_coord_offsetH  = DIM*jnrH;
 969
 970             /* load j atom coordinates */
 971             gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 972                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 973                                                  x+j_coord_offsetE,x+j_coord_offsetF,
 974                                                  x+j_coord_offsetG,x+j_coord_offsetH,
 975                                                  &jx0,&jy0,&jz0);
 976
 977             /* Calculate displacement vector */
 978             dx00             = _mm256_sub_ps(ix0,jx0);
 979             dy00             = _mm256_sub_ps(iy0,jy0);
 980             dz00             = _mm256_sub_ps(iz0,jz0);
 981             dx10             = _mm256_sub_ps(ix1,jx0);
 982             dy10             = _mm256_sub_ps(iy1,jy0);
 983             dz10             = _mm256_sub_ps(iz1,jz0);
 984             dx20             = _mm256_sub_ps(ix2,jx0);
 985             dy20             = _mm256_sub_ps(iy2,jy0);
 986             dz20             = _mm256_sub_ps(iz2,jz0);
 987
 988             /* Calculate squared distance and things based on it */
 989             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
 990             rsq10            = gmx_mm256_calc_rsq_ps(dx10,dy10,dz10);
 991             rsq20            = gmx_mm256_calc_rsq_ps(dx20,dy20,dz20);
 992
 993             rinv00           = gmx_mm256_invsqrt_ps(rsq00);
 994             rinv10           = gmx_mm256_invsqrt_ps(rsq10);
 995             rinv20           = gmx_mm256_invsqrt_ps(rsq20);
 996
 997             rinvsq00         = _mm256_mul_ps(rinv00,rinv00);
 998             rinvsq10         = _mm256_mul_ps(rinv10,rinv10);
 999             rinvsq20         = _mm256_mul_ps(rinv20,rinv20);
1000
1001             /* Load parameters for j particles */
1002             jq0              = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
1003                                                                  charge+jnrC+0,charge+jnrD+0,
1004                                                                  charge+jnrE+0,charge+jnrF+0,
1005                                                                  charge+jnrG+0,charge+jnrH+0);
1006
1007             fjx0             = _mm256_setzero_ps();
1008             fjy0             = _mm256_setzero_ps();
1009             fjz0             = _mm256_setzero_ps();
1010
1011             /**************************
1012              * CALCULATE INTERACTIONS *
1013              **************************/
1014
1015             r00              = _mm256_mul_ps(rsq00,rinv00);
1016             r00              = _mm256_andnot_ps(dummy_mask,r00);
1017
1018             /* Compute parameters for interactions between i and j atoms */
1019             qq00             = _mm256_mul_ps(iq0,jq0);
1020
1021             /* EWALD ELECTROSTATICS */
1022
1023             /* Analytical PME correction */
1024             zeta2            = _mm256_mul_ps(beta2,rsq00);
1025             rinv3            = _mm256_mul_ps(rinvsq00,rinv00);
1026             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
1027             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
1028             felec            = _mm256_mul_ps(qq00,felec);
1029
1030             fscal            = felec;
1031
1032             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
1033
1034             /* Calculate temporary vectorial force */
1035             tx               = _mm256_mul_ps(fscal,dx00);
1036             ty               = _mm256_mul_ps(fscal,dy00);
1037             tz               = _mm256_mul_ps(fscal,dz00);
1038
1039             /* Update vectorial force */
1040             fix0             = _mm256_add_ps(fix0,tx);
1041             fiy0             = _mm256_add_ps(fiy0,ty);
1042             fiz0             = _mm256_add_ps(fiz0,tz);
1043
1044             fjx0             = _mm256_add_ps(fjx0,tx);
1045             fjy0             = _mm256_add_ps(fjy0,ty);
1046             fjz0             = _mm256_add_ps(fjz0,tz);
1047
1048             /**************************
1049              * CALCULATE INTERACTIONS *
1050              **************************/
1051
1052             r10              = _mm256_mul_ps(rsq10,rinv10);
1053             r10              = _mm256_andnot_ps(dummy_mask,r10);
1054
1055             /* Compute parameters for interactions between i and j atoms */
1056             qq10             = _mm256_mul_ps(iq1,jq0);
1057
1058             /* EWALD ELECTROSTATICS */
1059
1060             /* Analytical PME correction */
1061             zeta2            = _mm256_mul_ps(beta2,rsq10);
1062             rinv3            = _mm256_mul_ps(rinvsq10,rinv10);
1063             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
1064             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
1065             felec            = _mm256_mul_ps(qq10,felec);
1066
1067             fscal            = felec;
1068
1069             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
1070
1071             /* Calculate temporary vectorial force */
1072             tx               = _mm256_mul_ps(fscal,dx10);
1073             ty               = _mm256_mul_ps(fscal,dy10);
1074             tz               = _mm256_mul_ps(fscal,dz10);
1075
1076             /* Update vectorial force */
1077             fix1             = _mm256_add_ps(fix1,tx);
1078             fiy1             = _mm256_add_ps(fiy1,ty);
1079             fiz1             = _mm256_add_ps(fiz1,tz);
1080
1081             fjx0             = _mm256_add_ps(fjx0,tx);
1082             fjy0             = _mm256_add_ps(fjy0,ty);
1083             fjz0             = _mm256_add_ps(fjz0,tz);
1084
1085             /**************************
1086              * CALCULATE INTERACTIONS *
1087              **************************/
1088
1089             r20              = _mm256_mul_ps(rsq20,rinv20);
1090             r20              = _mm256_andnot_ps(dummy_mask,r20);
1091
1092             /* Compute parameters for interactions between i and j atoms */
1093             qq20             = _mm256_mul_ps(iq2,jq0);
1094
1095             /* EWALD ELECTROSTATICS */
1096
1097             /* Analytical PME correction */
1098             zeta2            = _mm256_mul_ps(beta2,rsq20);
1099             rinv3            = _mm256_mul_ps(rinvsq20,rinv20);
1100             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
1101             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
1102             felec            = _mm256_mul_ps(qq20,felec);
1103
1104             fscal            = felec;
1105
1106             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
1107
1108             /* Calculate temporary vectorial force */
1109             tx               = _mm256_mul_ps(fscal,dx20);
1110             ty               = _mm256_mul_ps(fscal,dy20);
1111             tz               = _mm256_mul_ps(fscal,dz20);
1112
1113             /* Update vectorial force */
1114             fix2             = _mm256_add_ps(fix2,tx);
1115             fiy2             = _mm256_add_ps(fiy2,ty);
1116             fiz2             = _mm256_add_ps(fiz2,tz);
1117
1118             fjx0             = _mm256_add_ps(fjx0,tx);
1119             fjy0             = _mm256_add_ps(fjy0,ty);
1120             fjz0             = _mm256_add_ps(fjz0,tz);
1121
1122             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
1123             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
1124             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
1125             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
1126             fjptrE             = (jnrlistE>=0) ? f+j_coord_offsetE : scratch;
1127             fjptrF             = (jnrlistF>=0) ? f+j_coord_offsetF : scratch;
1128             fjptrG             = (jnrlistG>=0) ? f+j_coord_offsetG : scratch;
1129             fjptrH             = (jnrlistH>=0) ? f+j_coord_offsetH : scratch;
1130
1131             gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,fjx0,fjy0,fjz0);
1132
1133             /* Inner loop uses 174 flops */
1134         }
1135
1136         /* End of innermost loop */
1137
1138         gmx_mm256_update_iforce_3atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
1139                                                  f+i_coord_offset,fshift+i_shift_offset);
1140
1141         /* Increment number of inner iterations */
1142         inneriter                  += j_index_end - j_index_start;
1143
1144         /* Outer loop uses 18 flops */
1145     }
1146
1147     /* Increment number of outer iterations */
1148     outeriter        += nri;
1149
1150     /* Update outer/inner flops */
1151
1152     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_W3_F,outeriter*18 + inneriter*174);
1153 }