src/gromacs/gmxlib/nonbonded/nb_kernel_avx_256_double/nb_kernel_ElecEwSh_VdwNone_GeomW4P1_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_ElecEwSh_VdwNone_GeomW4P1_VF_avx_256_double
  38  * Electrostatics interaction: Ewald
  39  * VdW interaction:            None
  40  * Geometry:                   Water4-Particle
  41  * Calculate force/pot:        PotentialAndForce
  42  */
  43 void
  44 nb_kernel_ElecEwSh_VdwNone_GeomW4P1_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 *           vdwioffsetptr1;
  71     __m256d          ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
  72     real *           vdwioffsetptr2;
  73     __m256d          ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
  74     real *           vdwioffsetptr3;
  75     __m256d          ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
  76     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
  77     __m256d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  78     __m256d          dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
  79     __m256d          dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
  80     __m256d          dx30,dy30,dz30,rsq30,rinv30,rinvsq30,r30,qq30,c6_30,c12_30;
  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     iq1              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+1]));
 118     iq2              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+2]));
 119     iq3              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+3]));
 120
 121     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 122     rcutoff_scalar   = fr->rcoulomb;
 123     rcutoff          = _mm256_set1_pd(rcutoff_scalar);
 124     rcutoff2         = _mm256_mul_pd(rcutoff,rcutoff);
 125
 126     /* Avoid stupid compiler warnings */
 127     jnrA = jnrB = jnrC = jnrD = 0;
 128     j_coord_offsetA = 0;
 129     j_coord_offsetB = 0;
 130     j_coord_offsetC = 0;
 131     j_coord_offsetD = 0;
 132
 133     outeriter        = 0;
 134     inneriter        = 0;
 135
 136     for(iidx=0;iidx<4*DIM;iidx++)
 137     {
 138         scratch[iidx] = 0.0;
 139     }
 140
 141     /* Start outer loop over neighborlists */
 142     for(iidx=0; iidx<nri; iidx++)
 143     {
 144         /* Load shift vector for this list */
 145         i_shift_offset   = DIM*shiftidx[iidx];
 146
 147         /* Load limits for loop over neighbors */
 148         j_index_start    = jindex[iidx];
 149         j_index_end      = jindex[iidx+1];
 150
 151         /* Get outer coordinate index */
 152         inr              = iinr[iidx];
 153         i_coord_offset   = DIM*inr;
 154
 155         /* Load i particle coords and add shift vector */
 156         gmx_mm256_load_shift_and_3rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset+DIM,
 157                                                     &ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
 158
 159         fix1             = _mm256_setzero_pd();
 160         fiy1             = _mm256_setzero_pd();
 161         fiz1             = _mm256_setzero_pd();
 162         fix2             = _mm256_setzero_pd();
 163         fiy2             = _mm256_setzero_pd();
 164         fiz2             = _mm256_setzero_pd();
 165         fix3             = _mm256_setzero_pd();
 166         fiy3             = _mm256_setzero_pd();
 167         fiz3             = _mm256_setzero_pd();
 168
 169         /* Reset potential sums */
 170         velecsum         = _mm256_setzero_pd();
 171
 172         /* Start inner kernel loop */
 173         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 174         {
 175
 176             /* Get j neighbor index, and coordinate index */
 177             jnrA             = jjnr[jidx];
 178             jnrB             = jjnr[jidx+1];
 179             jnrC             = jjnr[jidx+2];
 180             jnrD             = jjnr[jidx+3];
 181             j_coord_offsetA  = DIM*jnrA;
 182             j_coord_offsetB  = DIM*jnrB;
 183             j_coord_offsetC  = DIM*jnrC;
 184             j_coord_offsetD  = DIM*jnrD;
 185
 186             /* load j atom coordinates */
 187             gmx_mm256_load_1rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 188                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 189                                                  &jx0,&jy0,&jz0);
 190
 191             /* Calculate displacement vector */
 192             dx10             = _mm256_sub_pd(ix1,jx0);
 193             dy10             = _mm256_sub_pd(iy1,jy0);
 194             dz10             = _mm256_sub_pd(iz1,jz0);
 195             dx20             = _mm256_sub_pd(ix2,jx0);
 196             dy20             = _mm256_sub_pd(iy2,jy0);
 197             dz20             = _mm256_sub_pd(iz2,jz0);
 198             dx30             = _mm256_sub_pd(ix3,jx0);
 199             dy30             = _mm256_sub_pd(iy3,jy0);
 200             dz30             = _mm256_sub_pd(iz3,jz0);
 201
 202             /* Calculate squared distance and things based on it */
 203             rsq10            = gmx_mm256_calc_rsq_pd(dx10,dy10,dz10);
 204             rsq20            = gmx_mm256_calc_rsq_pd(dx20,dy20,dz20);
 205             rsq30            = gmx_mm256_calc_rsq_pd(dx30,dy30,dz30);
 206
 207             rinv10           = gmx_mm256_invsqrt_pd(rsq10);
 208             rinv20           = gmx_mm256_invsqrt_pd(rsq20);
 209             rinv30           = gmx_mm256_invsqrt_pd(rsq30);
 210
 211             rinvsq10         = _mm256_mul_pd(rinv10,rinv10);
 212             rinvsq20         = _mm256_mul_pd(rinv20,rinv20);
 213             rinvsq30         = _mm256_mul_pd(rinv30,rinv30);
 214
 215             /* Load parameters for j particles */
 216             jq0              = gmx_mm256_load_4real_swizzle_pd(charge+jnrA+0,charge+jnrB+0,
 217                                                                  charge+jnrC+0,charge+jnrD+0);
 218
 219             fjx0             = _mm256_setzero_pd();
 220             fjy0             = _mm256_setzero_pd();
 221             fjz0             = _mm256_setzero_pd();
 222
 223             /**************************
 224              * CALCULATE INTERACTIONS *
 225              **************************/
 226
 227             if (gmx_mm256_any_lt(rsq10,rcutoff2))
 228             {
 229
 230             r10              = _mm256_mul_pd(rsq10,rinv10);
 231
 232             /* Compute parameters for interactions between i and j atoms */
 233             qq10             = _mm256_mul_pd(iq1,jq0);
 234
 235             /* EWALD ELECTROSTATICS */
 236
 237             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 238             ewrt             = _mm256_mul_pd(r10,ewtabscale);
 239             ewitab           = _mm256_cvttpd_epi32(ewrt);
 240             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
 241             ewitab           = _mm_slli_epi32(ewitab,2);
 242             ewtabF           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 243             ewtabD           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
 244             ewtabV           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,2) );
 245             ewtabFn          = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,3) );
 246             GMX_MM256_FULLTRANSPOSE4_PD(ewtabF,ewtabD,ewtabV,ewtabFn);
 247             felec            = _mm256_add_pd(ewtabF,_mm256_mul_pd(eweps,ewtabD));
 248             velec            = _mm256_sub_pd(ewtabV,_mm256_mul_pd(_mm256_mul_pd(ewtabhalfspace,eweps),_mm256_add_pd(ewtabF,felec)));
 249             velec            = _mm256_mul_pd(qq10,_mm256_sub_pd(_mm256_sub_pd(rinv10,sh_ewald),velec));
 250             felec            = _mm256_mul_pd(_mm256_mul_pd(qq10,rinv10),_mm256_sub_pd(rinvsq10,felec));
 251
 252             cutoff_mask      = _mm256_cmp_pd(rsq10,rcutoff2,_CMP_LT_OQ);
 253
 254             /* Update potential sum for this i atom from the interaction with this j atom. */
 255             velec            = _mm256_and_pd(velec,cutoff_mask);
 256             velecsum         = _mm256_add_pd(velecsum,velec);
 257
 258             fscal            = felec;
 259
 260             fscal            = _mm256_and_pd(fscal,cutoff_mask);
 261
 262             /* Calculate temporary vectorial force */
 263             tx               = _mm256_mul_pd(fscal,dx10);
 264             ty               = _mm256_mul_pd(fscal,dy10);
 265             tz               = _mm256_mul_pd(fscal,dz10);
 266
 267             /* Update vectorial force */
 268             fix1             = _mm256_add_pd(fix1,tx);
 269             fiy1             = _mm256_add_pd(fiy1,ty);
 270             fiz1             = _mm256_add_pd(fiz1,tz);
 271
 272             fjx0             = _mm256_add_pd(fjx0,tx);
 273             fjy0             = _mm256_add_pd(fjy0,ty);
 274             fjz0             = _mm256_add_pd(fjz0,tz);
 275
 276             }
 277
 278             /**************************
 279              * CALCULATE INTERACTIONS *
 280              **************************/
 281
 282             if (gmx_mm256_any_lt(rsq20,rcutoff2))
 283             {
 284
 285             r20              = _mm256_mul_pd(rsq20,rinv20);
 286
 287             /* Compute parameters for interactions between i and j atoms */
 288             qq20             = _mm256_mul_pd(iq2,jq0);
 289
 290             /* EWALD ELECTROSTATICS */
 291
 292             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 293             ewrt             = _mm256_mul_pd(r20,ewtabscale);
 294             ewitab           = _mm256_cvttpd_epi32(ewrt);
 295             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
 296             ewitab           = _mm_slli_epi32(ewitab,2);
 297             ewtabF           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 298             ewtabD           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
 299             ewtabV           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,2) );
 300             ewtabFn          = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,3) );
 301             GMX_MM256_FULLTRANSPOSE4_PD(ewtabF,ewtabD,ewtabV,ewtabFn);
 302             felec            = _mm256_add_pd(ewtabF,_mm256_mul_pd(eweps,ewtabD));
 303             velec            = _mm256_sub_pd(ewtabV,_mm256_mul_pd(_mm256_mul_pd(ewtabhalfspace,eweps),_mm256_add_pd(ewtabF,felec)));
 304             velec            = _mm256_mul_pd(qq20,_mm256_sub_pd(_mm256_sub_pd(rinv20,sh_ewald),velec));
 305             felec            = _mm256_mul_pd(_mm256_mul_pd(qq20,rinv20),_mm256_sub_pd(rinvsq20,felec));
 306
 307             cutoff_mask      = _mm256_cmp_pd(rsq20,rcutoff2,_CMP_LT_OQ);
 308
 309             /* Update potential sum for this i atom from the interaction with this j atom. */
 310             velec            = _mm256_and_pd(velec,cutoff_mask);
 311             velecsum         = _mm256_add_pd(velecsum,velec);
 312
 313             fscal            = felec;
 314
 315             fscal            = _mm256_and_pd(fscal,cutoff_mask);
 316
 317             /* Calculate temporary vectorial force */
 318             tx               = _mm256_mul_pd(fscal,dx20);
 319             ty               = _mm256_mul_pd(fscal,dy20);
 320             tz               = _mm256_mul_pd(fscal,dz20);
 321
 322             /* Update vectorial force */
 323             fix2             = _mm256_add_pd(fix2,tx);
 324             fiy2             = _mm256_add_pd(fiy2,ty);
 325             fiz2             = _mm256_add_pd(fiz2,tz);
 326
 327             fjx0             = _mm256_add_pd(fjx0,tx);
 328             fjy0             = _mm256_add_pd(fjy0,ty);
 329             fjz0             = _mm256_add_pd(fjz0,tz);
 330
 331             }
 332
 333             /**************************
 334              * CALCULATE INTERACTIONS *
 335              **************************/
 336
 337             if (gmx_mm256_any_lt(rsq30,rcutoff2))
 338             {
 339
 340             r30              = _mm256_mul_pd(rsq30,rinv30);
 341
 342             /* Compute parameters for interactions between i and j atoms */
 343             qq30             = _mm256_mul_pd(iq3,jq0);
 344
 345             /* EWALD ELECTROSTATICS */
 346
 347             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 348             ewrt             = _mm256_mul_pd(r30,ewtabscale);
 349             ewitab           = _mm256_cvttpd_epi32(ewrt);
 350             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
 351             ewitab           = _mm_slli_epi32(ewitab,2);
 352             ewtabF           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 353             ewtabD           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
 354             ewtabV           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,2) );
 355             ewtabFn          = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,3) );
 356             GMX_MM256_FULLTRANSPOSE4_PD(ewtabF,ewtabD,ewtabV,ewtabFn);
 357             felec            = _mm256_add_pd(ewtabF,_mm256_mul_pd(eweps,ewtabD));
 358             velec            = _mm256_sub_pd(ewtabV,_mm256_mul_pd(_mm256_mul_pd(ewtabhalfspace,eweps),_mm256_add_pd(ewtabF,felec)));
 359             velec            = _mm256_mul_pd(qq30,_mm256_sub_pd(_mm256_sub_pd(rinv30,sh_ewald),velec));
 360             felec            = _mm256_mul_pd(_mm256_mul_pd(qq30,rinv30),_mm256_sub_pd(rinvsq30,felec));
 361
 362             cutoff_mask      = _mm256_cmp_pd(rsq30,rcutoff2,_CMP_LT_OQ);
 363
 364             /* Update potential sum for this i atom from the interaction with this j atom. */
 365             velec            = _mm256_and_pd(velec,cutoff_mask);
 366             velecsum         = _mm256_add_pd(velecsum,velec);
 367
 368             fscal            = felec;
 369
 370             fscal            = _mm256_and_pd(fscal,cutoff_mask);
 371
 372             /* Calculate temporary vectorial force */
 373             tx               = _mm256_mul_pd(fscal,dx30);
 374             ty               = _mm256_mul_pd(fscal,dy30);
 375             tz               = _mm256_mul_pd(fscal,dz30);
 376
 377             /* Update vectorial force */
 378             fix3             = _mm256_add_pd(fix3,tx);
 379             fiy3             = _mm256_add_pd(fiy3,ty);
 380             fiz3             = _mm256_add_pd(fiz3,tz);
 381
 382             fjx0             = _mm256_add_pd(fjx0,tx);
 383             fjy0             = _mm256_add_pd(fjy0,ty);
 384             fjz0             = _mm256_add_pd(fjz0,tz);
 385
 386             }
 387
 388             fjptrA             = f+j_coord_offsetA;
 389             fjptrB             = f+j_coord_offsetB;
 390             fjptrC             = f+j_coord_offsetC;
 391             fjptrD             = f+j_coord_offsetD;
 392
 393             gmx_mm256_decrement_1rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 394
 395             /* Inner loop uses 141 flops */
 396         }
 397
 398         if(jidx<j_index_end)
 399         {
 400
 401             /* Get j neighbor index, and coordinate index */
 402             jnrlistA         = jjnr[jidx];
 403             jnrlistB         = jjnr[jidx+1];
 404             jnrlistC         = jjnr[jidx+2];
 405             jnrlistD         = jjnr[jidx+3];
 406             /* Sign of each element will be negative for non-real atoms.
 407              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 408              * so use it as val = _mm_andnot_pd(mask,val) to clear dummy entries.
 409              */
 410             tmpmask0 = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 411
 412             tmpmask1 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(3,3,2,2));
 413             tmpmask0 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(1,1,0,0));
 414             dummy_mask = _mm256_castps_pd(gmx_mm256_set_m128(tmpmask1,tmpmask0));
 415
 416             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 417             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 418             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 419             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 420             j_coord_offsetA  = DIM*jnrA;
 421             j_coord_offsetB  = DIM*jnrB;
 422             j_coord_offsetC  = DIM*jnrC;
 423             j_coord_offsetD  = DIM*jnrD;
 424
 425             /* load j atom coordinates */
 426             gmx_mm256_load_1rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 427                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 428                                                  &jx0,&jy0,&jz0);
 429
 430             /* Calculate displacement vector */
 431             dx10             = _mm256_sub_pd(ix1,jx0);
 432             dy10             = _mm256_sub_pd(iy1,jy0);
 433             dz10             = _mm256_sub_pd(iz1,jz0);
 434             dx20             = _mm256_sub_pd(ix2,jx0);
 435             dy20             = _mm256_sub_pd(iy2,jy0);
 436             dz20             = _mm256_sub_pd(iz2,jz0);
 437             dx30             = _mm256_sub_pd(ix3,jx0);
 438             dy30             = _mm256_sub_pd(iy3,jy0);
 439             dz30             = _mm256_sub_pd(iz3,jz0);
 440
 441             /* Calculate squared distance and things based on it */
 442             rsq10            = gmx_mm256_calc_rsq_pd(dx10,dy10,dz10);
 443             rsq20            = gmx_mm256_calc_rsq_pd(dx20,dy20,dz20);
 444             rsq30            = gmx_mm256_calc_rsq_pd(dx30,dy30,dz30);
 445
 446             rinv10           = gmx_mm256_invsqrt_pd(rsq10);
 447             rinv20           = gmx_mm256_invsqrt_pd(rsq20);
 448             rinv30           = gmx_mm256_invsqrt_pd(rsq30);
 449
 450             rinvsq10         = _mm256_mul_pd(rinv10,rinv10);
 451             rinvsq20         = _mm256_mul_pd(rinv20,rinv20);
 452             rinvsq30         = _mm256_mul_pd(rinv30,rinv30);
 453
 454             /* Load parameters for j particles */
 455             jq0              = gmx_mm256_load_4real_swizzle_pd(charge+jnrA+0,charge+jnrB+0,
 456                                                                  charge+jnrC+0,charge+jnrD+0);
 457
 458             fjx0             = _mm256_setzero_pd();
 459             fjy0             = _mm256_setzero_pd();
 460             fjz0             = _mm256_setzero_pd();
 461
 462             /**************************
 463              * CALCULATE INTERACTIONS *
 464              **************************/
 465
 466             if (gmx_mm256_any_lt(rsq10,rcutoff2))
 467             {
 468
 469             r10              = _mm256_mul_pd(rsq10,rinv10);
 470             r10              = _mm256_andnot_pd(dummy_mask,r10);
 471
 472             /* Compute parameters for interactions between i and j atoms */
 473             qq10             = _mm256_mul_pd(iq1,jq0);
 474
 475             /* EWALD ELECTROSTATICS */
 476
 477             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 478             ewrt             = _mm256_mul_pd(r10,ewtabscale);
 479             ewitab           = _mm256_cvttpd_epi32(ewrt);
 480             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
 481             ewitab           = _mm_slli_epi32(ewitab,2);
 482             ewtabF           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 483             ewtabD           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
 484             ewtabV           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,2) );
 485             ewtabFn          = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,3) );
 486             GMX_MM256_FULLTRANSPOSE4_PD(ewtabF,ewtabD,ewtabV,ewtabFn);
 487             felec            = _mm256_add_pd(ewtabF,_mm256_mul_pd(eweps,ewtabD));
 488             velec            = _mm256_sub_pd(ewtabV,_mm256_mul_pd(_mm256_mul_pd(ewtabhalfspace,eweps),_mm256_add_pd(ewtabF,felec)));
 489             velec            = _mm256_mul_pd(qq10,_mm256_sub_pd(_mm256_sub_pd(rinv10,sh_ewald),velec));
 490             felec            = _mm256_mul_pd(_mm256_mul_pd(qq10,rinv10),_mm256_sub_pd(rinvsq10,felec));
 491
 492             cutoff_mask      = _mm256_cmp_pd(rsq10,rcutoff2,_CMP_LT_OQ);
 493
 494             /* Update potential sum for this i atom from the interaction with this j atom. */
 495             velec            = _mm256_and_pd(velec,cutoff_mask);
 496             velec            = _mm256_andnot_pd(dummy_mask,velec);
 497             velecsum         = _mm256_add_pd(velecsum,velec);
 498
 499             fscal            = felec;
 500
 501             fscal            = _mm256_and_pd(fscal,cutoff_mask);
 502
 503             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
 504
 505             /* Calculate temporary vectorial force */
 506             tx               = _mm256_mul_pd(fscal,dx10);
 507             ty               = _mm256_mul_pd(fscal,dy10);
 508             tz               = _mm256_mul_pd(fscal,dz10);
 509
 510             /* Update vectorial force */
 511             fix1             = _mm256_add_pd(fix1,tx);
 512             fiy1             = _mm256_add_pd(fiy1,ty);
 513             fiz1             = _mm256_add_pd(fiz1,tz);
 514
 515             fjx0             = _mm256_add_pd(fjx0,tx);
 516             fjy0             = _mm256_add_pd(fjy0,ty);
 517             fjz0             = _mm256_add_pd(fjz0,tz);
 518
 519             }
 520
 521             /**************************
 522              * CALCULATE INTERACTIONS *
 523              **************************/
 524
 525             if (gmx_mm256_any_lt(rsq20,rcutoff2))
 526             {
 527
 528             r20              = _mm256_mul_pd(rsq20,rinv20);
 529             r20              = _mm256_andnot_pd(dummy_mask,r20);
 530
 531             /* Compute parameters for interactions between i and j atoms */
 532             qq20             = _mm256_mul_pd(iq2,jq0);
 533
 534             /* EWALD ELECTROSTATICS */
 535
 536             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 537             ewrt             = _mm256_mul_pd(r20,ewtabscale);
 538             ewitab           = _mm256_cvttpd_epi32(ewrt);
 539             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
 540             ewitab           = _mm_slli_epi32(ewitab,2);
 541             ewtabF           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 542             ewtabD           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
 543             ewtabV           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,2) );
 544             ewtabFn          = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,3) );
 545             GMX_MM256_FULLTRANSPOSE4_PD(ewtabF,ewtabD,ewtabV,ewtabFn);
 546             felec            = _mm256_add_pd(ewtabF,_mm256_mul_pd(eweps,ewtabD));
 547             velec            = _mm256_sub_pd(ewtabV,_mm256_mul_pd(_mm256_mul_pd(ewtabhalfspace,eweps),_mm256_add_pd(ewtabF,felec)));
 548             velec            = _mm256_mul_pd(qq20,_mm256_sub_pd(_mm256_sub_pd(rinv20,sh_ewald),velec));
 549             felec            = _mm256_mul_pd(_mm256_mul_pd(qq20,rinv20),_mm256_sub_pd(rinvsq20,felec));
 550
 551             cutoff_mask      = _mm256_cmp_pd(rsq20,rcutoff2,_CMP_LT_OQ);
 552
 553             /* Update potential sum for this i atom from the interaction with this j atom. */
 554             velec            = _mm256_and_pd(velec,cutoff_mask);
 555             velec            = _mm256_andnot_pd(dummy_mask,velec);
 556             velecsum         = _mm256_add_pd(velecsum,velec);
 557
 558             fscal            = felec;
 559
 560             fscal            = _mm256_and_pd(fscal,cutoff_mask);
 561
 562             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
 563
 564             /* Calculate temporary vectorial force */
 565             tx               = _mm256_mul_pd(fscal,dx20);
 566             ty               = _mm256_mul_pd(fscal,dy20);
 567             tz               = _mm256_mul_pd(fscal,dz20);
 568
 569             /* Update vectorial force */
 570             fix2             = _mm256_add_pd(fix2,tx);
 571             fiy2             = _mm256_add_pd(fiy2,ty);
 572             fiz2             = _mm256_add_pd(fiz2,tz);
 573
 574             fjx0             = _mm256_add_pd(fjx0,tx);
 575             fjy0             = _mm256_add_pd(fjy0,ty);
 576             fjz0             = _mm256_add_pd(fjz0,tz);
 577
 578             }
 579
 580             /**************************
 581              * CALCULATE INTERACTIONS *
 582              **************************/
 583
 584             if (gmx_mm256_any_lt(rsq30,rcutoff2))
 585             {
 586
 587             r30              = _mm256_mul_pd(rsq30,rinv30);
 588             r30              = _mm256_andnot_pd(dummy_mask,r30);
 589
 590             /* Compute parameters for interactions between i and j atoms */
 591             qq30             = _mm256_mul_pd(iq3,jq0);
 592
 593             /* EWALD ELECTROSTATICS */
 594
 595             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 596             ewrt             = _mm256_mul_pd(r30,ewtabscale);
 597             ewitab           = _mm256_cvttpd_epi32(ewrt);
 598             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
 599             ewitab           = _mm_slli_epi32(ewitab,2);
 600             ewtabF           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 601             ewtabD           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
 602             ewtabV           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,2) );
 603             ewtabFn          = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,3) );
 604             GMX_MM256_FULLTRANSPOSE4_PD(ewtabF,ewtabD,ewtabV,ewtabFn);
 605             felec            = _mm256_add_pd(ewtabF,_mm256_mul_pd(eweps,ewtabD));
 606             velec            = _mm256_sub_pd(ewtabV,_mm256_mul_pd(_mm256_mul_pd(ewtabhalfspace,eweps),_mm256_add_pd(ewtabF,felec)));
 607             velec            = _mm256_mul_pd(qq30,_mm256_sub_pd(_mm256_sub_pd(rinv30,sh_ewald),velec));
 608             felec            = _mm256_mul_pd(_mm256_mul_pd(qq30,rinv30),_mm256_sub_pd(rinvsq30,felec));
 609
 610             cutoff_mask      = _mm256_cmp_pd(rsq30,rcutoff2,_CMP_LT_OQ);
 611
 612             /* Update potential sum for this i atom from the interaction with this j atom. */
 613             velec            = _mm256_and_pd(velec,cutoff_mask);
 614             velec            = _mm256_andnot_pd(dummy_mask,velec);
 615             velecsum         = _mm256_add_pd(velecsum,velec);
 616
 617             fscal            = felec;
 618
 619             fscal            = _mm256_and_pd(fscal,cutoff_mask);
 620
 621             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
 622
 623             /* Calculate temporary vectorial force */
 624             tx               = _mm256_mul_pd(fscal,dx30);
 625             ty               = _mm256_mul_pd(fscal,dy30);
 626             tz               = _mm256_mul_pd(fscal,dz30);
 627
 628             /* Update vectorial force */
 629             fix3             = _mm256_add_pd(fix3,tx);
 630             fiy3             = _mm256_add_pd(fiy3,ty);
 631             fiz3             = _mm256_add_pd(fiz3,tz);
 632
 633             fjx0             = _mm256_add_pd(fjx0,tx);
 634             fjy0             = _mm256_add_pd(fjy0,ty);
 635             fjz0             = _mm256_add_pd(fjz0,tz);
 636
 637             }
 638
 639             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 640             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 641             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 642             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 643
 644             gmx_mm256_decrement_1rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 645
 646             /* Inner loop uses 144 flops */
 647         }
 648
 649         /* End of innermost loop */
 650
 651         gmx_mm256_update_iforce_3atom_swizzle_pd(fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
 652                                                  f+i_coord_offset+DIM,fshift+i_shift_offset);
 653
 654         ggid                        = gid[iidx];
 655         /* Update potential energies */
 656         gmx_mm256_update_1pot_pd(velecsum,kernel_data->energygrp_elec+ggid);
 657
 658         /* Increment number of inner iterations */
 659         inneriter                  += j_index_end - j_index_start;
 660
 661         /* Outer loop uses 19 flops */
 662     }
 663
 664     /* Increment number of outer iterations */
 665     outeriter        += nri;
 666
 667     /* Update outer/inner flops */
 668
 669     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_W4_VF,outeriter*19 + inneriter*144);
 670 }
 671 /*
 672  * Gromacs nonbonded kernel:   nb_kernel_ElecEwSh_VdwNone_GeomW4P1_F_avx_256_double
 673  * Electrostatics interaction: Ewald
 674  * VdW interaction:            None
 675  * Geometry:                   Water4-Particle
 676  * Calculate force/pot:        Force
 677  */
 678 void
 679 nb_kernel_ElecEwSh_VdwNone_GeomW4P1_F_avx_256_double
 680                     (t_nblist * gmx_restrict                nlist,
 681                      rvec * gmx_restrict                    xx,
 682                      rvec * gmx_restrict                    ff,
 683                      t_forcerec * gmx_restrict              fr,
 684                      t_mdatoms * gmx_restrict               mdatoms,
 685                      nb_kernel_data_t * gmx_restrict        kernel_data,
 686                      t_nrnb * gmx_restrict                  nrnb)
 687 {
 688     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 689      * just 0 for non-waters.
 690      * Suffixes A,B,C,D refer to j loop unrolling done with AVX, e.g. for the four different
 691      * jnr indices corresponding to data put in the four positions in the SIMD register.
 692      */
 693     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 694     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 695     int              jnrA,jnrB,jnrC,jnrD;
 696     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
 697     int              jnrlistE,jnrlistF,jnrlistG,jnrlistH;
 698     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 699     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 700     real             rcutoff_scalar;
 701     real             *shiftvec,*fshift,*x,*f;
 702     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
 703     real             scratch[4*DIM];
 704     __m256d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 705     real *           vdwioffsetptr1;
 706     __m256d          ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
 707     real *           vdwioffsetptr2;
 708     __m256d          ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
 709     real *           vdwioffsetptr3;
 710     __m256d          ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
 711     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
 712     __m256d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 713     __m256d          dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
 714     __m256d          dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
 715     __m256d          dx30,dy30,dz30,rsq30,rinv30,rinvsq30,r30,qq30,c6_30,c12_30;
 716     __m256d          velec,felec,velecsum,facel,crf,krf,krf2;
 717     real             *charge;
 718     __m128i          ewitab;
 719     __m256d          ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
 720     __m256d          beta,beta2,beta3,zeta2,pmecorrF,pmecorrV,rinv3;
 721     real             *ewtab;
 722     __m256d          dummy_mask,cutoff_mask;
 723     __m128           tmpmask0,tmpmask1;
 724     __m256d          signbit = _mm256_castsi256_pd( _mm256_set1_epi32(0x80000000) );
 725     __m256d          one     = _mm256_set1_pd(1.0);
 726     __m256d          two     = _mm256_set1_pd(2.0);
 727     x                = xx[0];
 728     f                = ff[0];
 729
 730     nri              = nlist->nri;
 731     iinr             = nlist->iinr;
 732     jindex           = nlist->jindex;
 733     jjnr             = nlist->jjnr;
 734     shiftidx         = nlist->shift;
 735     gid              = nlist->gid;
 736     shiftvec         = fr->shift_vec[0];
 737     fshift           = fr->fshift[0];
 738     facel            = _mm256_set1_pd(fr->epsfac);
 739     charge           = mdatoms->chargeA;
 740
 741     sh_ewald         = _mm256_set1_pd(fr->ic->sh_ewald);
 742     beta             = _mm256_set1_pd(fr->ic->ewaldcoeff);
 743     beta2            = _mm256_mul_pd(beta,beta);
 744     beta3            = _mm256_mul_pd(beta,beta2);
 745
 746     ewtab            = fr->ic->tabq_coul_F;
 747     ewtabscale       = _mm256_set1_pd(fr->ic->tabq_scale);
 748     ewtabhalfspace   = _mm256_set1_pd(0.5/fr->ic->tabq_scale);
 749
 750     /* Setup water-specific parameters */
 751     inr              = nlist->iinr[0];
 752     iq1              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+1]));
 753     iq2              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+2]));
 754     iq3              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+3]));
 755
 756     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 757     rcutoff_scalar   = fr->rcoulomb;
 758     rcutoff          = _mm256_set1_pd(rcutoff_scalar);
 759     rcutoff2         = _mm256_mul_pd(rcutoff,rcutoff);
 760
 761     /* Avoid stupid compiler warnings */
 762     jnrA = jnrB = jnrC = jnrD = 0;
 763     j_coord_offsetA = 0;
 764     j_coord_offsetB = 0;
 765     j_coord_offsetC = 0;
 766     j_coord_offsetD = 0;
 767
 768     outeriter        = 0;
 769     inneriter        = 0;
 770
 771     for(iidx=0;iidx<4*DIM;iidx++)
 772     {
 773         scratch[iidx] = 0.0;
 774     }
 775
 776     /* Start outer loop over neighborlists */
 777     for(iidx=0; iidx<nri; iidx++)
 778     {
 779         /* Load shift vector for this list */
 780         i_shift_offset   = DIM*shiftidx[iidx];
 781
 782         /* Load limits for loop over neighbors */
 783         j_index_start    = jindex[iidx];
 784         j_index_end      = jindex[iidx+1];
 785
 786         /* Get outer coordinate index */
 787         inr              = iinr[iidx];
 788         i_coord_offset   = DIM*inr;
 789
 790         /* Load i particle coords and add shift vector */
 791         gmx_mm256_load_shift_and_3rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset+DIM,
 792                                                     &ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
 793
 794         fix1             = _mm256_setzero_pd();
 795         fiy1             = _mm256_setzero_pd();
 796         fiz1             = _mm256_setzero_pd();
 797         fix2             = _mm256_setzero_pd();
 798         fiy2             = _mm256_setzero_pd();
 799         fiz2             = _mm256_setzero_pd();
 800         fix3             = _mm256_setzero_pd();
 801         fiy3             = _mm256_setzero_pd();
 802         fiz3             = _mm256_setzero_pd();
 803
 804         /* Start inner kernel loop */
 805         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 806         {
 807
 808             /* Get j neighbor index, and coordinate index */
 809             jnrA             = jjnr[jidx];
 810             jnrB             = jjnr[jidx+1];
 811             jnrC             = jjnr[jidx+2];
 812             jnrD             = jjnr[jidx+3];
 813             j_coord_offsetA  = DIM*jnrA;
 814             j_coord_offsetB  = DIM*jnrB;
 815             j_coord_offsetC  = DIM*jnrC;
 816             j_coord_offsetD  = DIM*jnrD;
 817
 818             /* load j atom coordinates */
 819             gmx_mm256_load_1rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 820                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 821                                                  &jx0,&jy0,&jz0);
 822
 823             /* Calculate displacement vector */
 824             dx10             = _mm256_sub_pd(ix1,jx0);
 825             dy10             = _mm256_sub_pd(iy1,jy0);
 826             dz10             = _mm256_sub_pd(iz1,jz0);
 827             dx20             = _mm256_sub_pd(ix2,jx0);
 828             dy20             = _mm256_sub_pd(iy2,jy0);
 829             dz20             = _mm256_sub_pd(iz2,jz0);
 830             dx30             = _mm256_sub_pd(ix3,jx0);
 831             dy30             = _mm256_sub_pd(iy3,jy0);
 832             dz30             = _mm256_sub_pd(iz3,jz0);
 833
 834             /* Calculate squared distance and things based on it */
 835             rsq10            = gmx_mm256_calc_rsq_pd(dx10,dy10,dz10);
 836             rsq20            = gmx_mm256_calc_rsq_pd(dx20,dy20,dz20);
 837             rsq30            = gmx_mm256_calc_rsq_pd(dx30,dy30,dz30);
 838
 839             rinv10           = gmx_mm256_invsqrt_pd(rsq10);
 840             rinv20           = gmx_mm256_invsqrt_pd(rsq20);
 841             rinv30           = gmx_mm256_invsqrt_pd(rsq30);
 842
 843             rinvsq10         = _mm256_mul_pd(rinv10,rinv10);
 844             rinvsq20         = _mm256_mul_pd(rinv20,rinv20);
 845             rinvsq30         = _mm256_mul_pd(rinv30,rinv30);
 846
 847             /* Load parameters for j particles */
 848             jq0              = gmx_mm256_load_4real_swizzle_pd(charge+jnrA+0,charge+jnrB+0,
 849                                                                  charge+jnrC+0,charge+jnrD+0);
 850
 851             fjx0             = _mm256_setzero_pd();
 852             fjy0             = _mm256_setzero_pd();
 853             fjz0             = _mm256_setzero_pd();
 854
 855             /**************************
 856              * CALCULATE INTERACTIONS *
 857              **************************/
 858
 859             if (gmx_mm256_any_lt(rsq10,rcutoff2))
 860             {
 861
 862             r10              = _mm256_mul_pd(rsq10,rinv10);
 863
 864             /* Compute parameters for interactions between i and j atoms */
 865             qq10             = _mm256_mul_pd(iq1,jq0);
 866
 867             /* EWALD ELECTROSTATICS */
 868
 869             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 870             ewrt             = _mm256_mul_pd(r10,ewtabscale);
 871             ewitab           = _mm256_cvttpd_epi32(ewrt);
 872             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
 873             gmx_mm256_load_4pair_swizzle_pd(ewtab + _mm_extract_epi32(ewitab,0),ewtab + _mm_extract_epi32(ewitab,1),
 874                                             ewtab + _mm_extract_epi32(ewitab,2),ewtab + _mm_extract_epi32(ewitab,3),
 875                                             &ewtabF,&ewtabFn);
 876             felec            = _mm256_add_pd(_mm256_mul_pd( _mm256_sub_pd(one,eweps),ewtabF),_mm256_mul_pd(eweps,ewtabFn));
 877             felec            = _mm256_mul_pd(_mm256_mul_pd(qq10,rinv10),_mm256_sub_pd(rinvsq10,felec));
 878
 879             cutoff_mask      = _mm256_cmp_pd(rsq10,rcutoff2,_CMP_LT_OQ);
 880
 881             fscal            = felec;
 882
 883             fscal            = _mm256_and_pd(fscal,cutoff_mask);
 884
 885             /* Calculate temporary vectorial force */
 886             tx               = _mm256_mul_pd(fscal,dx10);
 887             ty               = _mm256_mul_pd(fscal,dy10);
 888             tz               = _mm256_mul_pd(fscal,dz10);
 889
 890             /* Update vectorial force */
 891             fix1             = _mm256_add_pd(fix1,tx);
 892             fiy1             = _mm256_add_pd(fiy1,ty);
 893             fiz1             = _mm256_add_pd(fiz1,tz);
 894
 895             fjx0             = _mm256_add_pd(fjx0,tx);
 896             fjy0             = _mm256_add_pd(fjy0,ty);
 897             fjz0             = _mm256_add_pd(fjz0,tz);
 898
 899             }
 900
 901             /**************************
 902              * CALCULATE INTERACTIONS *
 903              **************************/
 904
 905             if (gmx_mm256_any_lt(rsq20,rcutoff2))
 906             {
 907
 908             r20              = _mm256_mul_pd(rsq20,rinv20);
 909
 910             /* Compute parameters for interactions between i and j atoms */
 911             qq20             = _mm256_mul_pd(iq2,jq0);
 912
 913             /* EWALD ELECTROSTATICS */
 914
 915             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 916             ewrt             = _mm256_mul_pd(r20,ewtabscale);
 917             ewitab           = _mm256_cvttpd_epi32(ewrt);
 918             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
 919             gmx_mm256_load_4pair_swizzle_pd(ewtab + _mm_extract_epi32(ewitab,0),ewtab + _mm_extract_epi32(ewitab,1),
 920                                             ewtab + _mm_extract_epi32(ewitab,2),ewtab + _mm_extract_epi32(ewitab,3),
 921                                             &ewtabF,&ewtabFn);
 922             felec            = _mm256_add_pd(_mm256_mul_pd( _mm256_sub_pd(one,eweps),ewtabF),_mm256_mul_pd(eweps,ewtabFn));
 923             felec            = _mm256_mul_pd(_mm256_mul_pd(qq20,rinv20),_mm256_sub_pd(rinvsq20,felec));
 924
 925             cutoff_mask      = _mm256_cmp_pd(rsq20,rcutoff2,_CMP_LT_OQ);
 926
 927             fscal            = felec;
 928
 929             fscal            = _mm256_and_pd(fscal,cutoff_mask);
 930
 931             /* Calculate temporary vectorial force */
 932             tx               = _mm256_mul_pd(fscal,dx20);
 933             ty               = _mm256_mul_pd(fscal,dy20);
 934             tz               = _mm256_mul_pd(fscal,dz20);
 935
 936             /* Update vectorial force */
 937             fix2             = _mm256_add_pd(fix2,tx);
 938             fiy2             = _mm256_add_pd(fiy2,ty);
 939             fiz2             = _mm256_add_pd(fiz2,tz);
 940
 941             fjx0             = _mm256_add_pd(fjx0,tx);
 942             fjy0             = _mm256_add_pd(fjy0,ty);
 943             fjz0             = _mm256_add_pd(fjz0,tz);
 944
 945             }
 946
 947             /**************************
 948              * CALCULATE INTERACTIONS *
 949              **************************/
 950
 951             if (gmx_mm256_any_lt(rsq30,rcutoff2))
 952             {
 953
 954             r30              = _mm256_mul_pd(rsq30,rinv30);
 955
 956             /* Compute parameters for interactions between i and j atoms */
 957             qq30             = _mm256_mul_pd(iq3,jq0);
 958
 959             /* EWALD ELECTROSTATICS */
 960
 961             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 962             ewrt             = _mm256_mul_pd(r30,ewtabscale);
 963             ewitab           = _mm256_cvttpd_epi32(ewrt);
 964             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
 965             gmx_mm256_load_4pair_swizzle_pd(ewtab + _mm_extract_epi32(ewitab,0),ewtab + _mm_extract_epi32(ewitab,1),
 966                                             ewtab + _mm_extract_epi32(ewitab,2),ewtab + _mm_extract_epi32(ewitab,3),
 967                                             &ewtabF,&ewtabFn);
 968             felec            = _mm256_add_pd(_mm256_mul_pd( _mm256_sub_pd(one,eweps),ewtabF),_mm256_mul_pd(eweps,ewtabFn));
 969             felec            = _mm256_mul_pd(_mm256_mul_pd(qq30,rinv30),_mm256_sub_pd(rinvsq30,felec));
 970
 971             cutoff_mask      = _mm256_cmp_pd(rsq30,rcutoff2,_CMP_LT_OQ);
 972
 973             fscal            = felec;
 974
 975             fscal            = _mm256_and_pd(fscal,cutoff_mask);
 976
 977             /* Calculate temporary vectorial force */
 978             tx               = _mm256_mul_pd(fscal,dx30);
 979             ty               = _mm256_mul_pd(fscal,dy30);
 980             tz               = _mm256_mul_pd(fscal,dz30);
 981
 982             /* Update vectorial force */
 983             fix3             = _mm256_add_pd(fix3,tx);
 984             fiy3             = _mm256_add_pd(fiy3,ty);
 985             fiz3             = _mm256_add_pd(fiz3,tz);
 986
 987             fjx0             = _mm256_add_pd(fjx0,tx);
 988             fjy0             = _mm256_add_pd(fjy0,ty);
 989             fjz0             = _mm256_add_pd(fjz0,tz);
 990
 991             }
 992
 993             fjptrA             = f+j_coord_offsetA;
 994             fjptrB             = f+j_coord_offsetB;
 995             fjptrC             = f+j_coord_offsetC;
 996             fjptrD             = f+j_coord_offsetD;
 997
 998             gmx_mm256_decrement_1rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 999
1000             /* Inner loop uses 120 flops */
1001         }
1002
1003         if(jidx<j_index_end)
1004         {
1005
1006             /* Get j neighbor index, and coordinate index */
1007             jnrlistA         = jjnr[jidx];
1008             jnrlistB         = jjnr[jidx+1];
1009             jnrlistC         = jjnr[jidx+2];
1010             jnrlistD         = jjnr[jidx+3];
1011             /* Sign of each element will be negative for non-real atoms.
1012              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
1013              * so use it as val = _mm_andnot_pd(mask,val) to clear dummy entries.
1014              */
1015             tmpmask0 = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
1016
1017             tmpmask1 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(3,3,2,2));
1018             tmpmask0 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(1,1,0,0));
1019             dummy_mask = _mm256_castps_pd(gmx_mm256_set_m128(tmpmask1,tmpmask0));
1020
1021             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
1022             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
1023             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
1024             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
1025             j_coord_offsetA  = DIM*jnrA;
1026             j_coord_offsetB  = DIM*jnrB;
1027             j_coord_offsetC  = DIM*jnrC;
1028             j_coord_offsetD  = DIM*jnrD;
1029
1030             /* load j atom coordinates */
1031             gmx_mm256_load_1rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
1032                                                  x+j_coord_offsetC,x+j_coord_offsetD,
1033                                                  &jx0,&jy0,&jz0);
1034
1035             /* Calculate displacement vector */
1036             dx10             = _mm256_sub_pd(ix1,jx0);
1037             dy10             = _mm256_sub_pd(iy1,jy0);
1038             dz10             = _mm256_sub_pd(iz1,jz0);
1039             dx20             = _mm256_sub_pd(ix2,jx0);
1040             dy20             = _mm256_sub_pd(iy2,jy0);
1041             dz20             = _mm256_sub_pd(iz2,jz0);
1042             dx30             = _mm256_sub_pd(ix3,jx0);
1043             dy30             = _mm256_sub_pd(iy3,jy0);
1044             dz30             = _mm256_sub_pd(iz3,jz0);
1045
1046             /* Calculate squared distance and things based on it */
1047             rsq10            = gmx_mm256_calc_rsq_pd(dx10,dy10,dz10);
1048             rsq20            = gmx_mm256_calc_rsq_pd(dx20,dy20,dz20);
1049             rsq30            = gmx_mm256_calc_rsq_pd(dx30,dy30,dz30);
1050
1051             rinv10           = gmx_mm256_invsqrt_pd(rsq10);
1052             rinv20           = gmx_mm256_invsqrt_pd(rsq20);
1053             rinv30           = gmx_mm256_invsqrt_pd(rsq30);
1054
1055             rinvsq10         = _mm256_mul_pd(rinv10,rinv10);
1056             rinvsq20         = _mm256_mul_pd(rinv20,rinv20);
1057             rinvsq30         = _mm256_mul_pd(rinv30,rinv30);
1058
1059             /* Load parameters for j particles */
1060             jq0              = gmx_mm256_load_4real_swizzle_pd(charge+jnrA+0,charge+jnrB+0,
1061                                                                  charge+jnrC+0,charge+jnrD+0);
1062
1063             fjx0             = _mm256_setzero_pd();
1064             fjy0             = _mm256_setzero_pd();
1065             fjz0             = _mm256_setzero_pd();
1066
1067             /**************************
1068              * CALCULATE INTERACTIONS *
1069              **************************/
1070
1071             if (gmx_mm256_any_lt(rsq10,rcutoff2))
1072             {
1073
1074             r10              = _mm256_mul_pd(rsq10,rinv10);
1075             r10              = _mm256_andnot_pd(dummy_mask,r10);
1076
1077             /* Compute parameters for interactions between i and j atoms */
1078             qq10             = _mm256_mul_pd(iq1,jq0);
1079
1080             /* EWALD ELECTROSTATICS */
1081
1082             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1083             ewrt             = _mm256_mul_pd(r10,ewtabscale);
1084             ewitab           = _mm256_cvttpd_epi32(ewrt);
1085             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
1086             gmx_mm256_load_4pair_swizzle_pd(ewtab + _mm_extract_epi32(ewitab,0),ewtab + _mm_extract_epi32(ewitab,1),
1087                                             ewtab + _mm_extract_epi32(ewitab,2),ewtab + _mm_extract_epi32(ewitab,3),
1088                                             &ewtabF,&ewtabFn);
1089             felec            = _mm256_add_pd(_mm256_mul_pd( _mm256_sub_pd(one,eweps),ewtabF),_mm256_mul_pd(eweps,ewtabFn));
1090             felec            = _mm256_mul_pd(_mm256_mul_pd(qq10,rinv10),_mm256_sub_pd(rinvsq10,felec));
1091
1092             cutoff_mask      = _mm256_cmp_pd(rsq10,rcutoff2,_CMP_LT_OQ);
1093
1094             fscal            = felec;
1095
1096             fscal            = _mm256_and_pd(fscal,cutoff_mask);
1097
1098             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
1099
1100             /* Calculate temporary vectorial force */
1101             tx               = _mm256_mul_pd(fscal,dx10);
1102             ty               = _mm256_mul_pd(fscal,dy10);
1103             tz               = _mm256_mul_pd(fscal,dz10);
1104
1105             /* Update vectorial force */
1106             fix1             = _mm256_add_pd(fix1,tx);
1107             fiy1             = _mm256_add_pd(fiy1,ty);
1108             fiz1             = _mm256_add_pd(fiz1,tz);
1109
1110             fjx0             = _mm256_add_pd(fjx0,tx);
1111             fjy0             = _mm256_add_pd(fjy0,ty);
1112             fjz0             = _mm256_add_pd(fjz0,tz);
1113
1114             }
1115
1116             /**************************
1117              * CALCULATE INTERACTIONS *
1118              **************************/
1119
1120             if (gmx_mm256_any_lt(rsq20,rcutoff2))
1121             {
1122
1123             r20              = _mm256_mul_pd(rsq20,rinv20);
1124             r20              = _mm256_andnot_pd(dummy_mask,r20);
1125
1126             /* Compute parameters for interactions between i and j atoms */
1127             qq20             = _mm256_mul_pd(iq2,jq0);
1128
1129             /* EWALD ELECTROSTATICS */
1130
1131             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1132             ewrt             = _mm256_mul_pd(r20,ewtabscale);
1133             ewitab           = _mm256_cvttpd_epi32(ewrt);
1134             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
1135             gmx_mm256_load_4pair_swizzle_pd(ewtab + _mm_extract_epi32(ewitab,0),ewtab + _mm_extract_epi32(ewitab,1),
1136                                             ewtab + _mm_extract_epi32(ewitab,2),ewtab + _mm_extract_epi32(ewitab,3),
1137                                             &ewtabF,&ewtabFn);
1138             felec            = _mm256_add_pd(_mm256_mul_pd( _mm256_sub_pd(one,eweps),ewtabF),_mm256_mul_pd(eweps,ewtabFn));
1139             felec            = _mm256_mul_pd(_mm256_mul_pd(qq20,rinv20),_mm256_sub_pd(rinvsq20,felec));
1140
1141             cutoff_mask      = _mm256_cmp_pd(rsq20,rcutoff2,_CMP_LT_OQ);
1142
1143             fscal            = felec;
1144
1145             fscal            = _mm256_and_pd(fscal,cutoff_mask);
1146
1147             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
1148
1149             /* Calculate temporary vectorial force */
1150             tx               = _mm256_mul_pd(fscal,dx20);
1151             ty               = _mm256_mul_pd(fscal,dy20);
1152             tz               = _mm256_mul_pd(fscal,dz20);
1153
1154             /* Update vectorial force */
1155             fix2             = _mm256_add_pd(fix2,tx);
1156             fiy2             = _mm256_add_pd(fiy2,ty);
1157             fiz2             = _mm256_add_pd(fiz2,tz);
1158
1159             fjx0             = _mm256_add_pd(fjx0,tx);
1160             fjy0             = _mm256_add_pd(fjy0,ty);
1161             fjz0             = _mm256_add_pd(fjz0,tz);
1162
1163             }
1164
1165             /**************************
1166              * CALCULATE INTERACTIONS *
1167              **************************/
1168
1169             if (gmx_mm256_any_lt(rsq30,rcutoff2))
1170             {
1171
1172             r30              = _mm256_mul_pd(rsq30,rinv30);
1173             r30              = _mm256_andnot_pd(dummy_mask,r30);
1174
1175             /* Compute parameters for interactions between i and j atoms */
1176             qq30             = _mm256_mul_pd(iq3,jq0);
1177
1178             /* EWALD ELECTROSTATICS */
1179
1180             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1181             ewrt             = _mm256_mul_pd(r30,ewtabscale);
1182             ewitab           = _mm256_cvttpd_epi32(ewrt);
1183             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
1184             gmx_mm256_load_4pair_swizzle_pd(ewtab + _mm_extract_epi32(ewitab,0),ewtab + _mm_extract_epi32(ewitab,1),
1185                                             ewtab + _mm_extract_epi32(ewitab,2),ewtab + _mm_extract_epi32(ewitab,3),
1186                                             &ewtabF,&ewtabFn);
1187             felec            = _mm256_add_pd(_mm256_mul_pd( _mm256_sub_pd(one,eweps),ewtabF),_mm256_mul_pd(eweps,ewtabFn));
1188             felec            = _mm256_mul_pd(_mm256_mul_pd(qq30,rinv30),_mm256_sub_pd(rinvsq30,felec));
1189
1190             cutoff_mask      = _mm256_cmp_pd(rsq30,rcutoff2,_CMP_LT_OQ);
1191
1192             fscal            = felec;
1193
1194             fscal            = _mm256_and_pd(fscal,cutoff_mask);
1195
1196             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
1197
1198             /* Calculate temporary vectorial force */
1199             tx               = _mm256_mul_pd(fscal,dx30);
1200             ty               = _mm256_mul_pd(fscal,dy30);
1201             tz               = _mm256_mul_pd(fscal,dz30);
1202
1203             /* Update vectorial force */
1204             fix3             = _mm256_add_pd(fix3,tx);
1205             fiy3             = _mm256_add_pd(fiy3,ty);
1206             fiz3             = _mm256_add_pd(fiz3,tz);
1207
1208             fjx0             = _mm256_add_pd(fjx0,tx);
1209             fjy0             = _mm256_add_pd(fjy0,ty);
1210             fjz0             = _mm256_add_pd(fjz0,tz);
1211
1212             }
1213
1214             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
1215             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
1216             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
1217             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
1218
1219             gmx_mm256_decrement_1rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
1220
1221             /* Inner loop uses 123 flops */
1222         }
1223
1224         /* End of innermost loop */
1225
1226         gmx_mm256_update_iforce_3atom_swizzle_pd(fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
1227                                                  f+i_coord_offset+DIM,fshift+i_shift_offset);
1228
1229         /* Increment number of inner iterations */
1230         inneriter                  += j_index_end - j_index_start;
1231
1232         /* Outer loop uses 18 flops */
1233     }
1234
1235     /* Increment number of outer iterations */
1236     outeriter        += nri;
1237
1238     /* Update outer/inner flops */
1239
1240     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_W4_F,outeriter*18 + inneriter*123);
1241 }