src/gmxlib/nonbonded/nb_kernel_avx_256_double/nb_kernel_ElecEw_VdwNone_GeomW3P1_avx_256_double.c

   1 /*
   2  * Note: this file was generated by the Gromacs avx_256_double 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_double.h"
  34 #include "kernelutil_x86_avx_256_double.h"
  35
  36 /*
  37  * Gromacs nonbonded kernel:   nb_kernel_ElecEw_VdwNone_GeomW3P1_VF_avx_256_double
  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_double
  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 refer to j loop unrolling done with AVX, e.g. for the four 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              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
  62     int              jnrlistE,jnrlistF,jnrlistG,jnrlistH;
  63     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
  64     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
  65     real             rcutoff_scalar;
  66     real             *shiftvec,*fshift,*x,*f;
  67     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
  68     real             scratch[4*DIM];
  69     __m256d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
  70     real *           vdwioffsetptr0;
  71     __m256d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
  72     real *           vdwioffsetptr1;
  73     __m256d          ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
  74     real *           vdwioffsetptr2;
  75     __m256d          ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
  76     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
  77     __m256d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  78     __m256d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  79     __m256d          dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
  80     __m256d          dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
  81     __m256d          velec,felec,velecsum,facel,crf,krf,krf2;
  82     real             *charge;
  83     __m128i          ewitab;
  84     __m256d          ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
  85     __m256d          beta,beta2,beta3,zeta2,pmecorrF,pmecorrV,rinv3;
  86     real             *ewtab;
  87     __m256d          dummy_mask,cutoff_mask;
  88     __m128           tmpmask0,tmpmask1;
  89     __m256d          signbit = _mm256_castsi256_pd( _mm256_set1_epi32(0x80000000) );
  90     __m256d          one     = _mm256_set1_pd(1.0);
  91     __m256d          two     = _mm256_set1_pd(2.0);
  92     x                = xx[0];
  93     f                = ff[0];
  94
  95     nri              = nlist->nri;
  96     iinr             = nlist->iinr;
  97     jindex           = nlist->jindex;
  98     jjnr             = nlist->jjnr;
  99     shiftidx         = nlist->shift;
 100     gid              = nlist->gid;
 101     shiftvec         = fr->shift_vec[0];
 102     fshift           = fr->fshift[0];
 103     facel            = _mm256_set1_pd(fr->epsfac);
 104     charge           = mdatoms->chargeA;
 105
 106     sh_ewald         = _mm256_set1_pd(fr->ic->sh_ewald);
 107     beta             = _mm256_set1_pd(fr->ic->ewaldcoeff);
 108     beta2            = _mm256_mul_pd(beta,beta);
 109     beta3            = _mm256_mul_pd(beta,beta2);
 110
 111     ewtab            = fr->ic->tabq_coul_FDV0;
 112     ewtabscale       = _mm256_set1_pd(fr->ic->tabq_scale);
 113     ewtabhalfspace   = _mm256_set1_pd(0.5/fr->ic->tabq_scale);
 114
 115     /* Setup water-specific parameters */
 116     inr              = nlist->iinr[0];
 117     iq0              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+0]));
 118     iq1              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+1]));
 119     iq2              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+2]));
 120
 121     /* Avoid stupid compiler warnings */
 122     jnrA = jnrB = jnrC = jnrD = 0;
 123     j_coord_offsetA = 0;
 124     j_coord_offsetB = 0;
 125     j_coord_offsetC = 0;
 126     j_coord_offsetD = 0;
 127
 128     outeriter        = 0;
 129     inneriter        = 0;
 130
 131     for(iidx=0;iidx<4*DIM;iidx++)
 132     {
 133         scratch[iidx] = 0.0;
 134     }
 135
 136     /* Start outer loop over neighborlists */
 137     for(iidx=0; iidx<nri; iidx++)
 138     {
 139         /* Load shift vector for this list */
 140         i_shift_offset   = DIM*shiftidx[iidx];
 141
 142         /* Load limits for loop over neighbors */
 143         j_index_start    = jindex[iidx];
 144         j_index_end      = jindex[iidx+1];
 145
 146         /* Get outer coordinate index */
 147         inr              = iinr[iidx];
 148         i_coord_offset   = DIM*inr;
 149
 150         /* Load i particle coords and add shift vector */
 151         gmx_mm256_load_shift_and_3rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
 152                                                     &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 153
 154         fix0             = _mm256_setzero_pd();
 155         fiy0             = _mm256_setzero_pd();
 156         fiz0             = _mm256_setzero_pd();
 157         fix1             = _mm256_setzero_pd();
 158         fiy1             = _mm256_setzero_pd();
 159         fiz1             = _mm256_setzero_pd();
 160         fix2             = _mm256_setzero_pd();
 161         fiy2             = _mm256_setzero_pd();
 162         fiz2             = _mm256_setzero_pd();
 163
 164         /* Reset potential sums */
 165         velecsum         = _mm256_setzero_pd();
 166
 167         /* Start inner kernel loop */
 168         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 169         {
 170
 171             /* Get j neighbor index, and coordinate index */
 172             jnrA             = jjnr[jidx];
 173             jnrB             = jjnr[jidx+1];
 174             jnrC             = jjnr[jidx+2];
 175             jnrD             = jjnr[jidx+3];
 176             j_coord_offsetA  = DIM*jnrA;
 177             j_coord_offsetB  = DIM*jnrB;
 178             j_coord_offsetC  = DIM*jnrC;
 179             j_coord_offsetD  = DIM*jnrD;
 180
 181             /* load j atom coordinates */
 182             gmx_mm256_load_1rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 183                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 184                                                  &jx0,&jy0,&jz0);
 185
 186             /* Calculate displacement vector */
 187             dx00             = _mm256_sub_pd(ix0,jx0);
 188             dy00             = _mm256_sub_pd(iy0,jy0);
 189             dz00             = _mm256_sub_pd(iz0,jz0);
 190             dx10             = _mm256_sub_pd(ix1,jx0);
 191             dy10             = _mm256_sub_pd(iy1,jy0);
 192             dz10             = _mm256_sub_pd(iz1,jz0);
 193             dx20             = _mm256_sub_pd(ix2,jx0);
 194             dy20             = _mm256_sub_pd(iy2,jy0);
 195             dz20             = _mm256_sub_pd(iz2,jz0);
 196
 197             /* Calculate squared distance and things based on it */
 198             rsq00            = gmx_mm256_calc_rsq_pd(dx00,dy00,dz00);
 199             rsq10            = gmx_mm256_calc_rsq_pd(dx10,dy10,dz10);
 200             rsq20            = gmx_mm256_calc_rsq_pd(dx20,dy20,dz20);
 201
 202             rinv00           = gmx_mm256_invsqrt_pd(rsq00);
 203             rinv10           = gmx_mm256_invsqrt_pd(rsq10);
 204             rinv20           = gmx_mm256_invsqrt_pd(rsq20);
 205
 206             rinvsq00         = _mm256_mul_pd(rinv00,rinv00);
 207             rinvsq10         = _mm256_mul_pd(rinv10,rinv10);
 208             rinvsq20         = _mm256_mul_pd(rinv20,rinv20);
 209
 210             /* Load parameters for j particles */
 211             jq0              = gmx_mm256_load_4real_swizzle_pd(charge+jnrA+0,charge+jnrB+0,
 212                                                                  charge+jnrC+0,charge+jnrD+0);
 213
 214             fjx0             = _mm256_setzero_pd();
 215             fjy0             = _mm256_setzero_pd();
 216             fjz0             = _mm256_setzero_pd();
 217
 218             /**************************
 219              * CALCULATE INTERACTIONS *
 220              **************************/
 221
 222             r00              = _mm256_mul_pd(rsq00,rinv00);
 223
 224             /* Compute parameters for interactions between i and j atoms */
 225             qq00             = _mm256_mul_pd(iq0,jq0);
 226
 227             /* EWALD ELECTROSTATICS */
 228
 229             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 230             ewrt             = _mm256_mul_pd(r00,ewtabscale);
 231             ewitab           = _mm256_cvttpd_epi32(ewrt);
 232             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
 233             ewitab           = _mm_slli_epi32(ewitab,2);
 234             ewtabF           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 235             ewtabD           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
 236             ewtabV           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,2) );
 237             ewtabFn          = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,3) );
 238             GMX_MM256_FULLTRANSPOSE4_PD(ewtabF,ewtabD,ewtabV,ewtabFn);
 239             felec            = _mm256_add_pd(ewtabF,_mm256_mul_pd(eweps,ewtabD));
 240             velec            = _mm256_sub_pd(ewtabV,_mm256_mul_pd(_mm256_mul_pd(ewtabhalfspace,eweps),_mm256_add_pd(ewtabF,felec)));
 241             velec            = _mm256_mul_pd(qq00,_mm256_sub_pd(rinv00,velec));
 242             felec            = _mm256_mul_pd(_mm256_mul_pd(qq00,rinv00),_mm256_sub_pd(rinvsq00,felec));
 243
 244             /* Update potential sum for this i atom from the interaction with this j atom. */
 245             velecsum         = _mm256_add_pd(velecsum,velec);
 246
 247             fscal            = felec;
 248
 249             /* Calculate temporary vectorial force */
 250             tx               = _mm256_mul_pd(fscal,dx00);
 251             ty               = _mm256_mul_pd(fscal,dy00);
 252             tz               = _mm256_mul_pd(fscal,dz00);
 253
 254             /* Update vectorial force */
 255             fix0             = _mm256_add_pd(fix0,tx);
 256             fiy0             = _mm256_add_pd(fiy0,ty);
 257             fiz0             = _mm256_add_pd(fiz0,tz);
 258
 259             fjx0             = _mm256_add_pd(fjx0,tx);
 260             fjy0             = _mm256_add_pd(fjy0,ty);
 261             fjz0             = _mm256_add_pd(fjz0,tz);
 262
 263             /**************************
 264              * CALCULATE INTERACTIONS *
 265              **************************/
 266
 267             r10              = _mm256_mul_pd(rsq10,rinv10);
 268
 269             /* Compute parameters for interactions between i and j atoms */
 270             qq10             = _mm256_mul_pd(iq1,jq0);
 271
 272             /* EWALD ELECTROSTATICS */
 273
 274             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 275             ewrt             = _mm256_mul_pd(r10,ewtabscale);
 276             ewitab           = _mm256_cvttpd_epi32(ewrt);
 277             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
 278             ewitab           = _mm_slli_epi32(ewitab,2);
 279             ewtabF           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 280             ewtabD           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
 281             ewtabV           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,2) );
 282             ewtabFn          = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,3) );
 283             GMX_MM256_FULLTRANSPOSE4_PD(ewtabF,ewtabD,ewtabV,ewtabFn);
 284             felec            = _mm256_add_pd(ewtabF,_mm256_mul_pd(eweps,ewtabD));
 285             velec            = _mm256_sub_pd(ewtabV,_mm256_mul_pd(_mm256_mul_pd(ewtabhalfspace,eweps),_mm256_add_pd(ewtabF,felec)));
 286             velec            = _mm256_mul_pd(qq10,_mm256_sub_pd(rinv10,velec));
 287             felec            = _mm256_mul_pd(_mm256_mul_pd(qq10,rinv10),_mm256_sub_pd(rinvsq10,felec));
 288
 289             /* Update potential sum for this i atom from the interaction with this j atom. */
 290             velecsum         = _mm256_add_pd(velecsum,velec);
 291
 292             fscal            = felec;
 293
 294             /* Calculate temporary vectorial force */
 295             tx               = _mm256_mul_pd(fscal,dx10);
 296             ty               = _mm256_mul_pd(fscal,dy10);
 297             tz               = _mm256_mul_pd(fscal,dz10);
 298
 299             /* Update vectorial force */
 300             fix1             = _mm256_add_pd(fix1,tx);
 301             fiy1             = _mm256_add_pd(fiy1,ty);
 302             fiz1             = _mm256_add_pd(fiz1,tz);
 303
 304             fjx0             = _mm256_add_pd(fjx0,tx);
 305             fjy0             = _mm256_add_pd(fjy0,ty);
 306             fjz0             = _mm256_add_pd(fjz0,tz);
 307
 308             /**************************
 309              * CALCULATE INTERACTIONS *
 310              **************************/
 311
 312             r20              = _mm256_mul_pd(rsq20,rinv20);
 313
 314             /* Compute parameters for interactions between i and j atoms */
 315             qq20             = _mm256_mul_pd(iq2,jq0);
 316
 317             /* EWALD ELECTROSTATICS */
 318
 319             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 320             ewrt             = _mm256_mul_pd(r20,ewtabscale);
 321             ewitab           = _mm256_cvttpd_epi32(ewrt);
 322             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
 323             ewitab           = _mm_slli_epi32(ewitab,2);
 324             ewtabF           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 325             ewtabD           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
 326             ewtabV           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,2) );
 327             ewtabFn          = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,3) );
 328             GMX_MM256_FULLTRANSPOSE4_PD(ewtabF,ewtabD,ewtabV,ewtabFn);
 329             felec            = _mm256_add_pd(ewtabF,_mm256_mul_pd(eweps,ewtabD));
 330             velec            = _mm256_sub_pd(ewtabV,_mm256_mul_pd(_mm256_mul_pd(ewtabhalfspace,eweps),_mm256_add_pd(ewtabF,felec)));
 331             velec            = _mm256_mul_pd(qq20,_mm256_sub_pd(rinv20,velec));
 332             felec            = _mm256_mul_pd(_mm256_mul_pd(qq20,rinv20),_mm256_sub_pd(rinvsq20,felec));
 333
 334             /* Update potential sum for this i atom from the interaction with this j atom. */
 335             velecsum         = _mm256_add_pd(velecsum,velec);
 336
 337             fscal            = felec;
 338
 339             /* Calculate temporary vectorial force */
 340             tx               = _mm256_mul_pd(fscal,dx20);
 341             ty               = _mm256_mul_pd(fscal,dy20);
 342             tz               = _mm256_mul_pd(fscal,dz20);
 343
 344             /* Update vectorial force */
 345             fix2             = _mm256_add_pd(fix2,tx);
 346             fiy2             = _mm256_add_pd(fiy2,ty);
 347             fiz2             = _mm256_add_pd(fiz2,tz);
 348
 349             fjx0             = _mm256_add_pd(fjx0,tx);
 350             fjy0             = _mm256_add_pd(fjy0,ty);
 351             fjz0             = _mm256_add_pd(fjz0,tz);
 352
 353             fjptrA             = f+j_coord_offsetA;
 354             fjptrB             = f+j_coord_offsetB;
 355             fjptrC             = f+j_coord_offsetC;
 356             fjptrD             = f+j_coord_offsetD;
 357
 358             gmx_mm256_decrement_1rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 359
 360             /* Inner loop uses 126 flops */
 361         }
 362
 363         if(jidx<j_index_end)
 364         {
 365
 366             /* Get j neighbor index, and coordinate index */
 367             jnrlistA         = jjnr[jidx];
 368             jnrlistB         = jjnr[jidx+1];
 369             jnrlistC         = jjnr[jidx+2];
 370             jnrlistD         = jjnr[jidx+3];
 371             /* Sign of each element will be negative for non-real atoms.
 372              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 373              * so use it as val = _mm_andnot_pd(mask,val) to clear dummy entries.
 374              */
 375             tmpmask0 = gmx_mm_castsi128_pd(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 376
 377             tmpmask1 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(3,3,2,2));
 378             tmpmask0 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(1,1,0,0));
 379             dummy_mask = _mm256_castps_pd(gmx_mm256_set_m128(tmpmask1,tmpmask0));
 380
 381             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 382             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 383             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 384             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 385             j_coord_offsetA  = DIM*jnrA;
 386             j_coord_offsetB  = DIM*jnrB;
 387             j_coord_offsetC  = DIM*jnrC;
 388             j_coord_offsetD  = DIM*jnrD;
 389
 390             /* load j atom coordinates */
 391             gmx_mm256_load_1rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 392                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 393                                                  &jx0,&jy0,&jz0);
 394
 395             /* Calculate displacement vector */
 396             dx00             = _mm256_sub_pd(ix0,jx0);
 397             dy00             = _mm256_sub_pd(iy0,jy0);
 398             dz00             = _mm256_sub_pd(iz0,jz0);
 399             dx10             = _mm256_sub_pd(ix1,jx0);
 400             dy10             = _mm256_sub_pd(iy1,jy0);
 401             dz10             = _mm256_sub_pd(iz1,jz0);
 402             dx20             = _mm256_sub_pd(ix2,jx0);
 403             dy20             = _mm256_sub_pd(iy2,jy0);
 404             dz20             = _mm256_sub_pd(iz2,jz0);
 405
 406             /* Calculate squared distance and things based on it */
 407             rsq00            = gmx_mm256_calc_rsq_pd(dx00,dy00,dz00);
 408             rsq10            = gmx_mm256_calc_rsq_pd(dx10,dy10,dz10);
 409             rsq20            = gmx_mm256_calc_rsq_pd(dx20,dy20,dz20);
 410
 411             rinv00           = gmx_mm256_invsqrt_pd(rsq00);
 412             rinv10           = gmx_mm256_invsqrt_pd(rsq10);
 413             rinv20           = gmx_mm256_invsqrt_pd(rsq20);
 414
 415             rinvsq00         = _mm256_mul_pd(rinv00,rinv00);
 416             rinvsq10         = _mm256_mul_pd(rinv10,rinv10);
 417             rinvsq20         = _mm256_mul_pd(rinv20,rinv20);
 418
 419             /* Load parameters for j particles */
 420             jq0              = gmx_mm256_load_4real_swizzle_pd(charge+jnrA+0,charge+jnrB+0,
 421                                                                  charge+jnrC+0,charge+jnrD+0);
 422
 423             fjx0             = _mm256_setzero_pd();
 424             fjy0             = _mm256_setzero_pd();
 425             fjz0             = _mm256_setzero_pd();
 426
 427             /**************************
 428              * CALCULATE INTERACTIONS *
 429              **************************/
 430
 431             r00              = _mm256_mul_pd(rsq00,rinv00);
 432             r00              = _mm256_andnot_pd(dummy_mask,r00);
 433
 434             /* Compute parameters for interactions between i and j atoms */
 435             qq00             = _mm256_mul_pd(iq0,jq0);
 436
 437             /* EWALD ELECTROSTATICS */
 438
 439             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 440             ewrt             = _mm256_mul_pd(r00,ewtabscale);
 441             ewitab           = _mm256_cvttpd_epi32(ewrt);
 442             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
 443             ewitab           = _mm_slli_epi32(ewitab,2);
 444             ewtabF           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 445             ewtabD           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
 446             ewtabV           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,2) );
 447             ewtabFn          = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,3) );
 448             GMX_MM256_FULLTRANSPOSE4_PD(ewtabF,ewtabD,ewtabV,ewtabFn);
 449             felec            = _mm256_add_pd(ewtabF,_mm256_mul_pd(eweps,ewtabD));
 450             velec            = _mm256_sub_pd(ewtabV,_mm256_mul_pd(_mm256_mul_pd(ewtabhalfspace,eweps),_mm256_add_pd(ewtabF,felec)));
 451             velec            = _mm256_mul_pd(qq00,_mm256_sub_pd(rinv00,velec));
 452             felec            = _mm256_mul_pd(_mm256_mul_pd(qq00,rinv00),_mm256_sub_pd(rinvsq00,felec));
 453
 454             /* Update potential sum for this i atom from the interaction with this j atom. */
 455             velec            = _mm256_andnot_pd(dummy_mask,velec);
 456             velecsum         = _mm256_add_pd(velecsum,velec);
 457
 458             fscal            = felec;
 459
 460             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
 461
 462             /* Calculate temporary vectorial force */
 463             tx               = _mm256_mul_pd(fscal,dx00);
 464             ty               = _mm256_mul_pd(fscal,dy00);
 465             tz               = _mm256_mul_pd(fscal,dz00);
 466
 467             /* Update vectorial force */
 468             fix0             = _mm256_add_pd(fix0,tx);
 469             fiy0             = _mm256_add_pd(fiy0,ty);
 470             fiz0             = _mm256_add_pd(fiz0,tz);
 471
 472             fjx0             = _mm256_add_pd(fjx0,tx);
 473             fjy0             = _mm256_add_pd(fjy0,ty);
 474             fjz0             = _mm256_add_pd(fjz0,tz);
 475
 476             /**************************
 477              * CALCULATE INTERACTIONS *
 478              **************************/
 479
 480             r10              = _mm256_mul_pd(rsq10,rinv10);
 481             r10              = _mm256_andnot_pd(dummy_mask,r10);
 482
 483             /* Compute parameters for interactions between i and j atoms */
 484             qq10             = _mm256_mul_pd(iq1,jq0);
 485
 486             /* EWALD ELECTROSTATICS */
 487
 488             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 489             ewrt             = _mm256_mul_pd(r10,ewtabscale);
 490             ewitab           = _mm256_cvttpd_epi32(ewrt);
 491             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
 492             ewitab           = _mm_slli_epi32(ewitab,2);
 493             ewtabF           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 494             ewtabD           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
 495             ewtabV           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,2) );
 496             ewtabFn          = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,3) );
 497             GMX_MM256_FULLTRANSPOSE4_PD(ewtabF,ewtabD,ewtabV,ewtabFn);
 498             felec            = _mm256_add_pd(ewtabF,_mm256_mul_pd(eweps,ewtabD));
 499             velec            = _mm256_sub_pd(ewtabV,_mm256_mul_pd(_mm256_mul_pd(ewtabhalfspace,eweps),_mm256_add_pd(ewtabF,felec)));
 500             velec            = _mm256_mul_pd(qq10,_mm256_sub_pd(rinv10,velec));
 501             felec            = _mm256_mul_pd(_mm256_mul_pd(qq10,rinv10),_mm256_sub_pd(rinvsq10,felec));
 502
 503             /* Update potential sum for this i atom from the interaction with this j atom. */
 504             velec            = _mm256_andnot_pd(dummy_mask,velec);
 505             velecsum         = _mm256_add_pd(velecsum,velec);
 506
 507             fscal            = felec;
 508
 509             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
 510
 511             /* Calculate temporary vectorial force */
 512             tx               = _mm256_mul_pd(fscal,dx10);
 513             ty               = _mm256_mul_pd(fscal,dy10);
 514             tz               = _mm256_mul_pd(fscal,dz10);
 515
 516             /* Update vectorial force */
 517             fix1             = _mm256_add_pd(fix1,tx);
 518             fiy1             = _mm256_add_pd(fiy1,ty);
 519             fiz1             = _mm256_add_pd(fiz1,tz);
 520
 521             fjx0             = _mm256_add_pd(fjx0,tx);
 522             fjy0             = _mm256_add_pd(fjy0,ty);
 523             fjz0             = _mm256_add_pd(fjz0,tz);
 524
 525             /**************************
 526              * CALCULATE INTERACTIONS *
 527              **************************/
 528
 529             r20              = _mm256_mul_pd(rsq20,rinv20);
 530             r20              = _mm256_andnot_pd(dummy_mask,r20);
 531
 532             /* Compute parameters for interactions between i and j atoms */
 533             qq20             = _mm256_mul_pd(iq2,jq0);
 534
 535             /* EWALD ELECTROSTATICS */
 536
 537             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 538             ewrt             = _mm256_mul_pd(r20,ewtabscale);
 539             ewitab           = _mm256_cvttpd_epi32(ewrt);
 540             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
 541             ewitab           = _mm_slli_epi32(ewitab,2);
 542             ewtabF           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 543             ewtabD           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
 544             ewtabV           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,2) );
 545             ewtabFn          = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,3) );
 546             GMX_MM256_FULLTRANSPOSE4_PD(ewtabF,ewtabD,ewtabV,ewtabFn);
 547             felec            = _mm256_add_pd(ewtabF,_mm256_mul_pd(eweps,ewtabD));
 548             velec            = _mm256_sub_pd(ewtabV,_mm256_mul_pd(_mm256_mul_pd(ewtabhalfspace,eweps),_mm256_add_pd(ewtabF,felec)));
 549             velec            = _mm256_mul_pd(qq20,_mm256_sub_pd(rinv20,velec));
 550             felec            = _mm256_mul_pd(_mm256_mul_pd(qq20,rinv20),_mm256_sub_pd(rinvsq20,felec));
 551
 552             /* Update potential sum for this i atom from the interaction with this j atom. */
 553             velec            = _mm256_andnot_pd(dummy_mask,velec);
 554             velecsum         = _mm256_add_pd(velecsum,velec);
 555
 556             fscal            = felec;
 557
 558             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
 559
 560             /* Calculate temporary vectorial force */
 561             tx               = _mm256_mul_pd(fscal,dx20);
 562             ty               = _mm256_mul_pd(fscal,dy20);
 563             tz               = _mm256_mul_pd(fscal,dz20);
 564
 565             /* Update vectorial force */
 566             fix2             = _mm256_add_pd(fix2,tx);
 567             fiy2             = _mm256_add_pd(fiy2,ty);
 568             fiz2             = _mm256_add_pd(fiz2,tz);
 569
 570             fjx0             = _mm256_add_pd(fjx0,tx);
 571             fjy0             = _mm256_add_pd(fjy0,ty);
 572             fjz0             = _mm256_add_pd(fjz0,tz);
 573
 574             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 575             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 576             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 577             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 578
 579             gmx_mm256_decrement_1rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 580
 581             /* Inner loop uses 129 flops */
 582         }
 583
 584         /* End of innermost loop */
 585
 586         gmx_mm256_update_iforce_3atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
 587                                                  f+i_coord_offset,fshift+i_shift_offset);
 588
 589         ggid                        = gid[iidx];
 590         /* Update potential energies */
 591         gmx_mm256_update_1pot_pd(velecsum,kernel_data->energygrp_elec+ggid);
 592
 593         /* Increment number of inner iterations */
 594         inneriter                  += j_index_end - j_index_start;
 595
 596         /* Outer loop uses 19 flops */
 597     }
 598
 599     /* Increment number of outer iterations */
 600     outeriter        += nri;
 601
 602     /* Update outer/inner flops */
 603
 604     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_W3_VF,outeriter*19 + inneriter*129);
 605 }
 606 /*
 607  * Gromacs nonbonded kernel:   nb_kernel_ElecEw_VdwNone_GeomW3P1_F_avx_256_double
 608  * Electrostatics interaction: Ewald
 609  * VdW interaction:            None
 610  * Geometry:                   Water3-Particle
 611  * Calculate force/pot:        Force
 612  */
 613 void
 614 nb_kernel_ElecEw_VdwNone_GeomW3P1_F_avx_256_double
 615                     (t_nblist * gmx_restrict                nlist,
 616                      rvec * gmx_restrict                    xx,
 617                      rvec * gmx_restrict                    ff,
 618                      t_forcerec * gmx_restrict              fr,
 619                      t_mdatoms * gmx_restrict               mdatoms,
 620                      nb_kernel_data_t * gmx_restrict        kernel_data,
 621                      t_nrnb * gmx_restrict                  nrnb)
 622 {
 623     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 624      * just 0 for non-waters.
 625      * Suffixes A,B,C,D refer to j loop unrolling done with AVX, e.g. for the four different
 626      * jnr indices corresponding to data put in the four positions in the SIMD register.
 627      */
 628     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 629     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 630     int              jnrA,jnrB,jnrC,jnrD;
 631     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
 632     int              jnrlistE,jnrlistF,jnrlistG,jnrlistH;
 633     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 634     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 635     real             rcutoff_scalar;
 636     real             *shiftvec,*fshift,*x,*f;
 637     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
 638     real             scratch[4*DIM];
 639     __m256d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 640     real *           vdwioffsetptr0;
 641     __m256d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 642     real *           vdwioffsetptr1;
 643     __m256d          ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
 644     real *           vdwioffsetptr2;
 645     __m256d          ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
 646     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
 647     __m256d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 648     __m256d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 649     __m256d          dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
 650     __m256d          dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
 651     __m256d          velec,felec,velecsum,facel,crf,krf,krf2;
 652     real             *charge;
 653     __m128i          ewitab;
 654     __m256d          ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
 655     __m256d          beta,beta2,beta3,zeta2,pmecorrF,pmecorrV,rinv3;
 656     real             *ewtab;
 657     __m256d          dummy_mask,cutoff_mask;
 658     __m128           tmpmask0,tmpmask1;
 659     __m256d          signbit = _mm256_castsi256_pd( _mm256_set1_epi32(0x80000000) );
 660     __m256d          one     = _mm256_set1_pd(1.0);
 661     __m256d          two     = _mm256_set1_pd(2.0);
 662     x                = xx[0];
 663     f                = ff[0];
 664
 665     nri              = nlist->nri;
 666     iinr             = nlist->iinr;
 667     jindex           = nlist->jindex;
 668     jjnr             = nlist->jjnr;
 669     shiftidx         = nlist->shift;
 670     gid              = nlist->gid;
 671     shiftvec         = fr->shift_vec[0];
 672     fshift           = fr->fshift[0];
 673     facel            = _mm256_set1_pd(fr->epsfac);
 674     charge           = mdatoms->chargeA;
 675
 676     sh_ewald         = _mm256_set1_pd(fr->ic->sh_ewald);
 677     beta             = _mm256_set1_pd(fr->ic->ewaldcoeff);
 678     beta2            = _mm256_mul_pd(beta,beta);
 679     beta3            = _mm256_mul_pd(beta,beta2);
 680
 681     ewtab            = fr->ic->tabq_coul_F;
 682     ewtabscale       = _mm256_set1_pd(fr->ic->tabq_scale);
 683     ewtabhalfspace   = _mm256_set1_pd(0.5/fr->ic->tabq_scale);
 684
 685     /* Setup water-specific parameters */
 686     inr              = nlist->iinr[0];
 687     iq0              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+0]));
 688     iq1              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+1]));
 689     iq2              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+2]));
 690
 691     /* Avoid stupid compiler warnings */
 692     jnrA = jnrB = jnrC = jnrD = 0;
 693     j_coord_offsetA = 0;
 694     j_coord_offsetB = 0;
 695     j_coord_offsetC = 0;
 696     j_coord_offsetD = 0;
 697
 698     outeriter        = 0;
 699     inneriter        = 0;
 700
 701     for(iidx=0;iidx<4*DIM;iidx++)
 702     {
 703         scratch[iidx] = 0.0;
 704     }
 705
 706     /* Start outer loop over neighborlists */
 707     for(iidx=0; iidx<nri; iidx++)
 708     {
 709         /* Load shift vector for this list */
 710         i_shift_offset   = DIM*shiftidx[iidx];
 711
 712         /* Load limits for loop over neighbors */
 713         j_index_start    = jindex[iidx];
 714         j_index_end      = jindex[iidx+1];
 715
 716         /* Get outer coordinate index */
 717         inr              = iinr[iidx];
 718         i_coord_offset   = DIM*inr;
 719
 720         /* Load i particle coords and add shift vector */
 721         gmx_mm256_load_shift_and_3rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
 722                                                     &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 723
 724         fix0             = _mm256_setzero_pd();
 725         fiy0             = _mm256_setzero_pd();
 726         fiz0             = _mm256_setzero_pd();
 727         fix1             = _mm256_setzero_pd();
 728         fiy1             = _mm256_setzero_pd();
 729         fiz1             = _mm256_setzero_pd();
 730         fix2             = _mm256_setzero_pd();
 731         fiy2             = _mm256_setzero_pd();
 732         fiz2             = _mm256_setzero_pd();
 733
 734         /* Start inner kernel loop */
 735         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 736         {
 737
 738             /* Get j neighbor index, and coordinate index */
 739             jnrA             = jjnr[jidx];
 740             jnrB             = jjnr[jidx+1];
 741             jnrC             = jjnr[jidx+2];
 742             jnrD             = jjnr[jidx+3];
 743             j_coord_offsetA  = DIM*jnrA;
 744             j_coord_offsetB  = DIM*jnrB;
 745             j_coord_offsetC  = DIM*jnrC;
 746             j_coord_offsetD  = DIM*jnrD;
 747
 748             /* load j atom coordinates */
 749             gmx_mm256_load_1rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 750                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 751                                                  &jx0,&jy0,&jz0);
 752
 753             /* Calculate displacement vector */
 754             dx00             = _mm256_sub_pd(ix0,jx0);
 755             dy00             = _mm256_sub_pd(iy0,jy0);
 756             dz00             = _mm256_sub_pd(iz0,jz0);
 757             dx10             = _mm256_sub_pd(ix1,jx0);
 758             dy10             = _mm256_sub_pd(iy1,jy0);
 759             dz10             = _mm256_sub_pd(iz1,jz0);
 760             dx20             = _mm256_sub_pd(ix2,jx0);
 761             dy20             = _mm256_sub_pd(iy2,jy0);
 762             dz20             = _mm256_sub_pd(iz2,jz0);
 763
 764             /* Calculate squared distance and things based on it */
 765             rsq00            = gmx_mm256_calc_rsq_pd(dx00,dy00,dz00);
 766             rsq10            = gmx_mm256_calc_rsq_pd(dx10,dy10,dz10);
 767             rsq20            = gmx_mm256_calc_rsq_pd(dx20,dy20,dz20);
 768
 769             rinv00           = gmx_mm256_invsqrt_pd(rsq00);
 770             rinv10           = gmx_mm256_invsqrt_pd(rsq10);
 771             rinv20           = gmx_mm256_invsqrt_pd(rsq20);
 772
 773             rinvsq00         = _mm256_mul_pd(rinv00,rinv00);
 774             rinvsq10         = _mm256_mul_pd(rinv10,rinv10);
 775             rinvsq20         = _mm256_mul_pd(rinv20,rinv20);
 776
 777             /* Load parameters for j particles */
 778             jq0              = gmx_mm256_load_4real_swizzle_pd(charge+jnrA+0,charge+jnrB+0,
 779                                                                  charge+jnrC+0,charge+jnrD+0);
 780
 781             fjx0             = _mm256_setzero_pd();
 782             fjy0             = _mm256_setzero_pd();
 783             fjz0             = _mm256_setzero_pd();
 784
 785             /**************************
 786              * CALCULATE INTERACTIONS *
 787              **************************/
 788
 789             r00              = _mm256_mul_pd(rsq00,rinv00);
 790
 791             /* Compute parameters for interactions between i and j atoms */
 792             qq00             = _mm256_mul_pd(iq0,jq0);
 793
 794             /* EWALD ELECTROSTATICS */
 795
 796             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 797             ewrt             = _mm256_mul_pd(r00,ewtabscale);
 798             ewitab           = _mm256_cvttpd_epi32(ewrt);
 799             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
 800             gmx_mm256_load_4pair_swizzle_pd(ewtab + _mm_extract_epi32(ewitab,0),ewtab + _mm_extract_epi32(ewitab,1),
 801                                             ewtab + _mm_extract_epi32(ewitab,2),ewtab + _mm_extract_epi32(ewitab,3),
 802                                             &ewtabF,&ewtabFn);
 803             felec            = _mm256_add_pd(_mm256_mul_pd( _mm256_sub_pd(one,eweps),ewtabF),_mm256_mul_pd(eweps,ewtabFn));
 804             felec            = _mm256_mul_pd(_mm256_mul_pd(qq00,rinv00),_mm256_sub_pd(rinvsq00,felec));
 805
 806             fscal            = felec;
 807
 808             /* Calculate temporary vectorial force */
 809             tx               = _mm256_mul_pd(fscal,dx00);
 810             ty               = _mm256_mul_pd(fscal,dy00);
 811             tz               = _mm256_mul_pd(fscal,dz00);
 812
 813             /* Update vectorial force */
 814             fix0             = _mm256_add_pd(fix0,tx);
 815             fiy0             = _mm256_add_pd(fiy0,ty);
 816             fiz0             = _mm256_add_pd(fiz0,tz);
 817
 818             fjx0             = _mm256_add_pd(fjx0,tx);
 819             fjy0             = _mm256_add_pd(fjy0,ty);
 820             fjz0             = _mm256_add_pd(fjz0,tz);
 821
 822             /**************************
 823              * CALCULATE INTERACTIONS *
 824              **************************/
 825
 826             r10              = _mm256_mul_pd(rsq10,rinv10);
 827
 828             /* Compute parameters for interactions between i and j atoms */
 829             qq10             = _mm256_mul_pd(iq1,jq0);
 830
 831             /* EWALD ELECTROSTATICS */
 832
 833             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 834             ewrt             = _mm256_mul_pd(r10,ewtabscale);
 835             ewitab           = _mm256_cvttpd_epi32(ewrt);
 836             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
 837             gmx_mm256_load_4pair_swizzle_pd(ewtab + _mm_extract_epi32(ewitab,0),ewtab + _mm_extract_epi32(ewitab,1),
 838                                             ewtab + _mm_extract_epi32(ewitab,2),ewtab + _mm_extract_epi32(ewitab,3),
 839                                             &ewtabF,&ewtabFn);
 840             felec            = _mm256_add_pd(_mm256_mul_pd( _mm256_sub_pd(one,eweps),ewtabF),_mm256_mul_pd(eweps,ewtabFn));
 841             felec            = _mm256_mul_pd(_mm256_mul_pd(qq10,rinv10),_mm256_sub_pd(rinvsq10,felec));
 842
 843             fscal            = felec;
 844
 845             /* Calculate temporary vectorial force */
 846             tx               = _mm256_mul_pd(fscal,dx10);
 847             ty               = _mm256_mul_pd(fscal,dy10);
 848             tz               = _mm256_mul_pd(fscal,dz10);
 849
 850             /* Update vectorial force */
 851             fix1             = _mm256_add_pd(fix1,tx);
 852             fiy1             = _mm256_add_pd(fiy1,ty);
 853             fiz1             = _mm256_add_pd(fiz1,tz);
 854
 855             fjx0             = _mm256_add_pd(fjx0,tx);
 856             fjy0             = _mm256_add_pd(fjy0,ty);
 857             fjz0             = _mm256_add_pd(fjz0,tz);
 858
 859             /**************************
 860              * CALCULATE INTERACTIONS *
 861              **************************/
 862
 863             r20              = _mm256_mul_pd(rsq20,rinv20);
 864
 865             /* Compute parameters for interactions between i and j atoms */
 866             qq20             = _mm256_mul_pd(iq2,jq0);
 867
 868             /* EWALD ELECTROSTATICS */
 869
 870             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 871             ewrt             = _mm256_mul_pd(r20,ewtabscale);
 872             ewitab           = _mm256_cvttpd_epi32(ewrt);
 873             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
 874             gmx_mm256_load_4pair_swizzle_pd(ewtab + _mm_extract_epi32(ewitab,0),ewtab + _mm_extract_epi32(ewitab,1),
 875                                             ewtab + _mm_extract_epi32(ewitab,2),ewtab + _mm_extract_epi32(ewitab,3),
 876                                             &ewtabF,&ewtabFn);
 877             felec            = _mm256_add_pd(_mm256_mul_pd( _mm256_sub_pd(one,eweps),ewtabF),_mm256_mul_pd(eweps,ewtabFn));
 878             felec            = _mm256_mul_pd(_mm256_mul_pd(qq20,rinv20),_mm256_sub_pd(rinvsq20,felec));
 879
 880             fscal            = felec;
 881
 882             /* Calculate temporary vectorial force */
 883             tx               = _mm256_mul_pd(fscal,dx20);
 884             ty               = _mm256_mul_pd(fscal,dy20);
 885             tz               = _mm256_mul_pd(fscal,dz20);
 886
 887             /* Update vectorial force */
 888             fix2             = _mm256_add_pd(fix2,tx);
 889             fiy2             = _mm256_add_pd(fiy2,ty);
 890             fiz2             = _mm256_add_pd(fiz2,tz);
 891
 892             fjx0             = _mm256_add_pd(fjx0,tx);
 893             fjy0             = _mm256_add_pd(fjy0,ty);
 894             fjz0             = _mm256_add_pd(fjz0,tz);
 895
 896             fjptrA             = f+j_coord_offsetA;
 897             fjptrB             = f+j_coord_offsetB;
 898             fjptrC             = f+j_coord_offsetC;
 899             fjptrD             = f+j_coord_offsetD;
 900
 901             gmx_mm256_decrement_1rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 902
 903             /* Inner loop uses 111 flops */
 904         }
 905
 906         if(jidx<j_index_end)
 907         {
 908
 909             /* Get j neighbor index, and coordinate index */
 910             jnrlistA         = jjnr[jidx];
 911             jnrlistB         = jjnr[jidx+1];
 912             jnrlistC         = jjnr[jidx+2];
 913             jnrlistD         = jjnr[jidx+3];
 914             /* Sign of each element will be negative for non-real atoms.
 915              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 916              * so use it as val = _mm_andnot_pd(mask,val) to clear dummy entries.
 917              */
 918             tmpmask0 = gmx_mm_castsi128_pd(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 919
 920             tmpmask1 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(3,3,2,2));
 921             tmpmask0 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(1,1,0,0));
 922             dummy_mask = _mm256_castps_pd(gmx_mm256_set_m128(tmpmask1,tmpmask0));
 923
 924             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 925             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 926             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 927             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 928             j_coord_offsetA  = DIM*jnrA;
 929             j_coord_offsetB  = DIM*jnrB;
 930             j_coord_offsetC  = DIM*jnrC;
 931             j_coord_offsetD  = DIM*jnrD;
 932
 933             /* load j atom coordinates */
 934             gmx_mm256_load_1rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 935                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 936                                                  &jx0,&jy0,&jz0);
 937
 938             /* Calculate displacement vector */
 939             dx00             = _mm256_sub_pd(ix0,jx0);
 940             dy00             = _mm256_sub_pd(iy0,jy0);
 941             dz00             = _mm256_sub_pd(iz0,jz0);
 942             dx10             = _mm256_sub_pd(ix1,jx0);
 943             dy10             = _mm256_sub_pd(iy1,jy0);
 944             dz10             = _mm256_sub_pd(iz1,jz0);
 945             dx20             = _mm256_sub_pd(ix2,jx0);
 946             dy20             = _mm256_sub_pd(iy2,jy0);
 947             dz20             = _mm256_sub_pd(iz2,jz0);
 948
 949             /* Calculate squared distance and things based on it */
 950             rsq00            = gmx_mm256_calc_rsq_pd(dx00,dy00,dz00);
 951             rsq10            = gmx_mm256_calc_rsq_pd(dx10,dy10,dz10);
 952             rsq20            = gmx_mm256_calc_rsq_pd(dx20,dy20,dz20);
 953
 954             rinv00           = gmx_mm256_invsqrt_pd(rsq00);
 955             rinv10           = gmx_mm256_invsqrt_pd(rsq10);
 956             rinv20           = gmx_mm256_invsqrt_pd(rsq20);
 957
 958             rinvsq00         = _mm256_mul_pd(rinv00,rinv00);
 959             rinvsq10         = _mm256_mul_pd(rinv10,rinv10);
 960             rinvsq20         = _mm256_mul_pd(rinv20,rinv20);
 961
 962             /* Load parameters for j particles */
 963             jq0              = gmx_mm256_load_4real_swizzle_pd(charge+jnrA+0,charge+jnrB+0,
 964                                                                  charge+jnrC+0,charge+jnrD+0);
 965
 966             fjx0             = _mm256_setzero_pd();
 967             fjy0             = _mm256_setzero_pd();
 968             fjz0             = _mm256_setzero_pd();
 969
 970             /**************************
 971              * CALCULATE INTERACTIONS *
 972              **************************/
 973
 974             r00              = _mm256_mul_pd(rsq00,rinv00);
 975             r00              = _mm256_andnot_pd(dummy_mask,r00);
 976
 977             /* Compute parameters for interactions between i and j atoms */
 978             qq00             = _mm256_mul_pd(iq0,jq0);
 979
 980             /* EWALD ELECTROSTATICS */
 981
 982             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 983             ewrt             = _mm256_mul_pd(r00,ewtabscale);
 984             ewitab           = _mm256_cvttpd_epi32(ewrt);
 985             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
 986             gmx_mm256_load_4pair_swizzle_pd(ewtab + _mm_extract_epi32(ewitab,0),ewtab + _mm_extract_epi32(ewitab,1),
 987                                             ewtab + _mm_extract_epi32(ewitab,2),ewtab + _mm_extract_epi32(ewitab,3),
 988                                             &ewtabF,&ewtabFn);
 989             felec            = _mm256_add_pd(_mm256_mul_pd( _mm256_sub_pd(one,eweps),ewtabF),_mm256_mul_pd(eweps,ewtabFn));
 990             felec            = _mm256_mul_pd(_mm256_mul_pd(qq00,rinv00),_mm256_sub_pd(rinvsq00,felec));
 991
 992             fscal            = felec;
 993
 994             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
 995
 996             /* Calculate temporary vectorial force */
 997             tx               = _mm256_mul_pd(fscal,dx00);
 998             ty               = _mm256_mul_pd(fscal,dy00);
 999             tz               = _mm256_mul_pd(fscal,dz00);
1000
1001             /* Update vectorial force */
1002             fix0             = _mm256_add_pd(fix0,tx);
1003             fiy0             = _mm256_add_pd(fiy0,ty);
1004             fiz0             = _mm256_add_pd(fiz0,tz);
1005
1006             fjx0             = _mm256_add_pd(fjx0,tx);
1007             fjy0             = _mm256_add_pd(fjy0,ty);
1008             fjz0             = _mm256_add_pd(fjz0,tz);
1009
1010             /**************************
1011              * CALCULATE INTERACTIONS *
1012              **************************/
1013
1014             r10              = _mm256_mul_pd(rsq10,rinv10);
1015             r10              = _mm256_andnot_pd(dummy_mask,r10);
1016
1017             /* Compute parameters for interactions between i and j atoms */
1018             qq10             = _mm256_mul_pd(iq1,jq0);
1019
1020             /* EWALD ELECTROSTATICS */
1021
1022             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1023             ewrt             = _mm256_mul_pd(r10,ewtabscale);
1024             ewitab           = _mm256_cvttpd_epi32(ewrt);
1025             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
1026             gmx_mm256_load_4pair_swizzle_pd(ewtab + _mm_extract_epi32(ewitab,0),ewtab + _mm_extract_epi32(ewitab,1),
1027                                             ewtab + _mm_extract_epi32(ewitab,2),ewtab + _mm_extract_epi32(ewitab,3),
1028                                             &ewtabF,&ewtabFn);
1029             felec            = _mm256_add_pd(_mm256_mul_pd( _mm256_sub_pd(one,eweps),ewtabF),_mm256_mul_pd(eweps,ewtabFn));
1030             felec            = _mm256_mul_pd(_mm256_mul_pd(qq10,rinv10),_mm256_sub_pd(rinvsq10,felec));
1031
1032             fscal            = felec;
1033
1034             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
1035
1036             /* Calculate temporary vectorial force */
1037             tx               = _mm256_mul_pd(fscal,dx10);
1038             ty               = _mm256_mul_pd(fscal,dy10);
1039             tz               = _mm256_mul_pd(fscal,dz10);
1040
1041             /* Update vectorial force */
1042             fix1             = _mm256_add_pd(fix1,tx);
1043             fiy1             = _mm256_add_pd(fiy1,ty);
1044             fiz1             = _mm256_add_pd(fiz1,tz);
1045
1046             fjx0             = _mm256_add_pd(fjx0,tx);
1047             fjy0             = _mm256_add_pd(fjy0,ty);
1048             fjz0             = _mm256_add_pd(fjz0,tz);
1049
1050             /**************************
1051              * CALCULATE INTERACTIONS *
1052              **************************/
1053
1054             r20              = _mm256_mul_pd(rsq20,rinv20);
1055             r20              = _mm256_andnot_pd(dummy_mask,r20);
1056
1057             /* Compute parameters for interactions between i and j atoms */
1058             qq20             = _mm256_mul_pd(iq2,jq0);
1059
1060             /* EWALD ELECTROSTATICS */
1061
1062             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1063             ewrt             = _mm256_mul_pd(r20,ewtabscale);
1064             ewitab           = _mm256_cvttpd_epi32(ewrt);
1065             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
1066             gmx_mm256_load_4pair_swizzle_pd(ewtab + _mm_extract_epi32(ewitab,0),ewtab + _mm_extract_epi32(ewitab,1),
1067                                             ewtab + _mm_extract_epi32(ewitab,2),ewtab + _mm_extract_epi32(ewitab,3),
1068                                             &ewtabF,&ewtabFn);
1069             felec            = _mm256_add_pd(_mm256_mul_pd( _mm256_sub_pd(one,eweps),ewtabF),_mm256_mul_pd(eweps,ewtabFn));
1070             felec            = _mm256_mul_pd(_mm256_mul_pd(qq20,rinv20),_mm256_sub_pd(rinvsq20,felec));
1071
1072             fscal            = felec;
1073
1074             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
1075
1076             /* Calculate temporary vectorial force */
1077             tx               = _mm256_mul_pd(fscal,dx20);
1078             ty               = _mm256_mul_pd(fscal,dy20);
1079             tz               = _mm256_mul_pd(fscal,dz20);
1080
1081             /* Update vectorial force */
1082             fix2             = _mm256_add_pd(fix2,tx);
1083             fiy2             = _mm256_add_pd(fiy2,ty);
1084             fiz2             = _mm256_add_pd(fiz2,tz);
1085
1086             fjx0             = _mm256_add_pd(fjx0,tx);
1087             fjy0             = _mm256_add_pd(fjy0,ty);
1088             fjz0             = _mm256_add_pd(fjz0,tz);
1089
1090             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
1091             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
1092             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
1093             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
1094
1095             gmx_mm256_decrement_1rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
1096
1097             /* Inner loop uses 114 flops */
1098         }
1099
1100         /* End of innermost loop */
1101
1102         gmx_mm256_update_iforce_3atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
1103                                                  f+i_coord_offset,fshift+i_shift_offset);
1104
1105         /* Increment number of inner iterations */
1106         inneriter                  += j_index_end - j_index_start;
1107
1108         /* Outer loop uses 18 flops */
1109     }
1110
1111     /* Increment number of outer iterations */
1112     outeriter        += nri;
1113
1114     /* Update outer/inner flops */
1115
1116     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_W3_F,outeriter*18 + inneriter*114);
1117 }