src/gmxlib/nonbonded/nb_kernel_avx_256_single/nb_kernel_ElecEwSh_VdwNone_GeomP1P1_avx_256_single.c

   1 /*
   2  * Note: this file was generated by the Gromacs avx_256_single kernel generator.
   3  *
   4  *                This source code is part of
   5  *
   6  *                 G   R   O   M   A   C   S
   7  *
   8  * Copyright (c) 2001-2012, The GROMACS Development Team
   9  *
  10  * Gromacs is a library for molecular simulation and trajectory analysis,
  11  * written by Erik Lindahl, David van der Spoel, Berk Hess, and others - for
  12  * a full list of developers and information, check out http://www.gromacs.org
  13  *
  14  * This program is free software; you can redistribute it and/or modify it under
  15  * the terms of the GNU Lesser General Public License as published by the Free
  16  * Software Foundation; either version 2 of the License, or (at your option) any
  17  * later version.
  18  *
  19  * To help fund GROMACS development, we humbly ask that you cite
  20  * the papers people have written on it - you can find them on the website.
  21  */
  22 #ifdef HAVE_CONFIG_H
  23 #include <config.h>
  24 #endif
  25
  26 #include <math.h>
  27
  28 #include "../nb_kernel.h"
  29 #include "types/simple.h"
  30 #include "vec.h"
  31 #include "nrnb.h"
  32
  33 #include "gmx_math_x86_avx_256_single.h"
  34 #include "kernelutil_x86_avx_256_single.h"
  35
  36 /*
  37  * Gromacs nonbonded kernel:   nb_kernel_ElecEwSh_VdwNone_GeomP1P1_VF_avx_256_single
  38  * Electrostatics interaction: Ewald
  39  * VdW interaction:            None
  40  * Geometry:                   Particle-Particle
  41  * Calculate force/pot:        PotentialAndForce
  42  */
  43 void
  44 nb_kernel_ElecEwSh_VdwNone_GeomP1P1_VF_avx_256_single
  45                     (t_nblist * gmx_restrict                nlist,
  46                      rvec * gmx_restrict                    xx,
  47                      rvec * gmx_restrict                    ff,
  48                      t_forcerec * gmx_restrict              fr,
  49                      t_mdatoms * gmx_restrict               mdatoms,
  50                      nb_kernel_data_t * gmx_restrict        kernel_data,
  51                      t_nrnb * gmx_restrict                  nrnb)
  52 {
  53     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
  54      * just 0 for non-waters.
  55      * Suffixes A,B,C,D,E,F,G,H refer to j loop unrolling done with AVX, e.g. for the eight different
  56      * jnr indices corresponding to data put in the four positions in the SIMD register.
  57      */
  58     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
  59     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
  60     int              jnrA,jnrB,jnrC,jnrD;
  61     int              jnrE,jnrF,jnrG,jnrH;
  62     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
  63     int              jnrlistE,jnrlistF,jnrlistG,jnrlistH;
  64     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
  65     int              j_coord_offsetE,j_coord_offsetF,j_coord_offsetG,j_coord_offsetH;
  66     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
  67     real             rcutoff_scalar;
  68     real             *shiftvec,*fshift,*x,*f;
  69     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD,*fjptrE,*fjptrF,*fjptrG,*fjptrH;
  70     real             scratch[4*DIM];
  71     __m256           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
  72     real *           vdwioffsetptr0;
  73     __m256           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
  74     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D,vdwjidx0E,vdwjidx0F,vdwjidx0G,vdwjidx0H;
  75     __m256           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  76     __m256           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  77     __m256           velec,felec,velecsum,facel,crf,krf,krf2;
  78     real             *charge;
  79     __m256i          ewitab;
  80     __m128i          ewitab_lo,ewitab_hi;
  81     __m256           ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
  82     __m256           beta,beta2,beta3,zeta2,pmecorrF,pmecorrV,rinv3;
  83     real             *ewtab;
  84     __m256           dummy_mask,cutoff_mask;
  85     __m256           signbit = _mm256_castsi256_ps( _mm256_set1_epi32(0x80000000) );
  86     __m256           one     = _mm256_set1_ps(1.0);
  87     __m256           two     = _mm256_set1_ps(2.0);
  88     x                = xx[0];
  89     f                = ff[0];
  90
  91     nri              = nlist->nri;
  92     iinr             = nlist->iinr;
  93     jindex           = nlist->jindex;
  94     jjnr             = nlist->jjnr;
  95     shiftidx         = nlist->shift;
  96     gid              = nlist->gid;
  97     shiftvec         = fr->shift_vec[0];
  98     fshift           = fr->fshift[0];
  99     facel            = _mm256_set1_ps(fr->epsfac);
 100     charge           = mdatoms->chargeA;
 101
 102     sh_ewald         = _mm256_set1_ps(fr->ic->sh_ewald);
 103     beta             = _mm256_set1_ps(fr->ic->ewaldcoeff);
 104     beta2            = _mm256_mul_ps(beta,beta);
 105     beta3            = _mm256_mul_ps(beta,beta2);
 106
 107     ewtab            = fr->ic->tabq_coul_FDV0;
 108     ewtabscale       = _mm256_set1_ps(fr->ic->tabq_scale);
 109     ewtabhalfspace   = _mm256_set1_ps(0.5/fr->ic->tabq_scale);
 110
 111     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 112     rcutoff_scalar   = fr->rcoulomb;
 113     rcutoff          = _mm256_set1_ps(rcutoff_scalar);
 114     rcutoff2         = _mm256_mul_ps(rcutoff,rcutoff);
 115
 116     /* Avoid stupid compiler warnings */
 117     jnrA = jnrB = jnrC = jnrD = jnrE = jnrF = jnrG = jnrH = 0;
 118     j_coord_offsetA = 0;
 119     j_coord_offsetB = 0;
 120     j_coord_offsetC = 0;
 121     j_coord_offsetD = 0;
 122     j_coord_offsetE = 0;
 123     j_coord_offsetF = 0;
 124     j_coord_offsetG = 0;
 125     j_coord_offsetH = 0;
 126
 127     outeriter        = 0;
 128     inneriter        = 0;
 129
 130     for(iidx=0;iidx<4*DIM;iidx++)
 131     {
 132         scratch[iidx] = 0.0;
 133     }
 134
 135     /* Start outer loop over neighborlists */
 136     for(iidx=0; iidx<nri; iidx++)
 137     {
 138         /* Load shift vector for this list */
 139         i_shift_offset   = DIM*shiftidx[iidx];
 140
 141         /* Load limits for loop over neighbors */
 142         j_index_start    = jindex[iidx];
 143         j_index_end      = jindex[iidx+1];
 144
 145         /* Get outer coordinate index */
 146         inr              = iinr[iidx];
 147         i_coord_offset   = DIM*inr;
 148
 149         /* Load i particle coords and add shift vector */
 150         gmx_mm256_load_shift_and_1rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 151
 152         fix0             = _mm256_setzero_ps();
 153         fiy0             = _mm256_setzero_ps();
 154         fiz0             = _mm256_setzero_ps();
 155
 156         /* Load parameters for i particles */
 157         iq0              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+0]));
 158
 159         /* Reset potential sums */
 160         velecsum         = _mm256_setzero_ps();
 161
 162         /* Start inner kernel loop */
 163         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+7]>=0; jidx+=8)
 164         {
 165
 166             /* Get j neighbor index, and coordinate index */
 167             jnrA             = jjnr[jidx];
 168             jnrB             = jjnr[jidx+1];
 169             jnrC             = jjnr[jidx+2];
 170             jnrD             = jjnr[jidx+3];
 171             jnrE             = jjnr[jidx+4];
 172             jnrF             = jjnr[jidx+5];
 173             jnrG             = jjnr[jidx+6];
 174             jnrH             = jjnr[jidx+7];
 175             j_coord_offsetA  = DIM*jnrA;
 176             j_coord_offsetB  = DIM*jnrB;
 177             j_coord_offsetC  = DIM*jnrC;
 178             j_coord_offsetD  = DIM*jnrD;
 179             j_coord_offsetE  = DIM*jnrE;
 180             j_coord_offsetF  = DIM*jnrF;
 181             j_coord_offsetG  = DIM*jnrG;
 182             j_coord_offsetH  = DIM*jnrH;
 183
 184             /* load j atom coordinates */
 185             gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 186                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 187                                                  x+j_coord_offsetE,x+j_coord_offsetF,
 188                                                  x+j_coord_offsetG,x+j_coord_offsetH,
 189                                                  &jx0,&jy0,&jz0);
 190
 191             /* Calculate displacement vector */
 192             dx00             = _mm256_sub_ps(ix0,jx0);
 193             dy00             = _mm256_sub_ps(iy0,jy0);
 194             dz00             = _mm256_sub_ps(iz0,jz0);
 195
 196             /* Calculate squared distance and things based on it */
 197             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
 198
 199             rinv00           = gmx_mm256_invsqrt_ps(rsq00);
 200
 201             rinvsq00         = _mm256_mul_ps(rinv00,rinv00);
 202
 203             /* Load parameters for j particles */
 204             jq0              = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 205                                                                  charge+jnrC+0,charge+jnrD+0,
 206                                                                  charge+jnrE+0,charge+jnrF+0,
 207                                                                  charge+jnrG+0,charge+jnrH+0);
 208
 209             /**************************
 210              * CALCULATE INTERACTIONS *
 211              **************************/
 212
 213             if (gmx_mm256_any_lt(rsq00,rcutoff2))
 214             {
 215
 216             r00              = _mm256_mul_ps(rsq00,rinv00);
 217
 218             /* Compute parameters for interactions between i and j atoms */
 219             qq00             = _mm256_mul_ps(iq0,jq0);
 220
 221             /* EWALD ELECTROSTATICS */
 222
 223             /* Analytical PME correction */
 224             zeta2            = _mm256_mul_ps(beta2,rsq00);
 225             rinv3            = _mm256_mul_ps(rinvsq00,rinv00);
 226             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
 227             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
 228             felec            = _mm256_mul_ps(qq00,felec);
 229             pmecorrV         = gmx_mm256_pmecorrV_ps(zeta2);
 230             pmecorrV         = _mm256_mul_ps(pmecorrV,beta);
 231             velec            = _mm256_sub_ps(_mm256_sub_ps(rinv00,sh_ewald),pmecorrV);
 232             velec            = _mm256_mul_ps(qq00,velec);
 233
 234             cutoff_mask      = _mm256_cmp_ps(rsq00,rcutoff2,_CMP_LT_OQ);
 235
 236             /* Update potential sum for this i atom from the interaction with this j atom. */
 237             velec            = _mm256_and_ps(velec,cutoff_mask);
 238             velecsum         = _mm256_add_ps(velecsum,velec);
 239
 240             fscal            = felec;
 241
 242             fscal            = _mm256_and_ps(fscal,cutoff_mask);
 243
 244             /* Calculate temporary vectorial force */
 245             tx               = _mm256_mul_ps(fscal,dx00);
 246             ty               = _mm256_mul_ps(fscal,dy00);
 247             tz               = _mm256_mul_ps(fscal,dz00);
 248
 249             /* Update vectorial force */
 250             fix0             = _mm256_add_ps(fix0,tx);
 251             fiy0             = _mm256_add_ps(fiy0,ty);
 252             fiz0             = _mm256_add_ps(fiz0,tz);
 253
 254             fjptrA             = f+j_coord_offsetA;
 255             fjptrB             = f+j_coord_offsetB;
 256             fjptrC             = f+j_coord_offsetC;
 257             fjptrD             = f+j_coord_offsetD;
 258             fjptrE             = f+j_coord_offsetE;
 259             fjptrF             = f+j_coord_offsetF;
 260             fjptrG             = f+j_coord_offsetG;
 261             fjptrH             = f+j_coord_offsetH;
 262             gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,tx,ty,tz);
 263
 264             }
 265
 266             /* Inner loop uses 109 flops */
 267         }
 268
 269         if(jidx<j_index_end)
 270         {
 271
 272             /* Get j neighbor index, and coordinate index */
 273             jnrlistA         = jjnr[jidx];
 274             jnrlistB         = jjnr[jidx+1];
 275             jnrlistC         = jjnr[jidx+2];
 276             jnrlistD         = jjnr[jidx+3];
 277             jnrlistE         = jjnr[jidx+4];
 278             jnrlistF         = jjnr[jidx+5];
 279             jnrlistG         = jjnr[jidx+6];
 280             jnrlistH         = jjnr[jidx+7];
 281             /* Sign of each element will be negative for non-real atoms.
 282              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 283              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 284              */
 285             dummy_mask = gmx_mm256_set_m128(gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx+4)),_mm_setzero_si128())),
 286                                             gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128())));
 287
 288             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 289             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 290             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 291             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 292             jnrE       = (jnrlistE>=0) ? jnrlistE : 0;
 293             jnrF       = (jnrlistF>=0) ? jnrlistF : 0;
 294             jnrG       = (jnrlistG>=0) ? jnrlistG : 0;
 295             jnrH       = (jnrlistH>=0) ? jnrlistH : 0;
 296             j_coord_offsetA  = DIM*jnrA;
 297             j_coord_offsetB  = DIM*jnrB;
 298             j_coord_offsetC  = DIM*jnrC;
 299             j_coord_offsetD  = DIM*jnrD;
 300             j_coord_offsetE  = DIM*jnrE;
 301             j_coord_offsetF  = DIM*jnrF;
 302             j_coord_offsetG  = DIM*jnrG;
 303             j_coord_offsetH  = DIM*jnrH;
 304
 305             /* load j atom coordinates */
 306             gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 307                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 308                                                  x+j_coord_offsetE,x+j_coord_offsetF,
 309                                                  x+j_coord_offsetG,x+j_coord_offsetH,
 310                                                  &jx0,&jy0,&jz0);
 311
 312             /* Calculate displacement vector */
 313             dx00             = _mm256_sub_ps(ix0,jx0);
 314             dy00             = _mm256_sub_ps(iy0,jy0);
 315             dz00             = _mm256_sub_ps(iz0,jz0);
 316
 317             /* Calculate squared distance and things based on it */
 318             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
 319
 320             rinv00           = gmx_mm256_invsqrt_ps(rsq00);
 321
 322             rinvsq00         = _mm256_mul_ps(rinv00,rinv00);
 323
 324             /* Load parameters for j particles */
 325             jq0              = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 326                                                                  charge+jnrC+0,charge+jnrD+0,
 327                                                                  charge+jnrE+0,charge+jnrF+0,
 328                                                                  charge+jnrG+0,charge+jnrH+0);
 329
 330             /**************************
 331              * CALCULATE INTERACTIONS *
 332              **************************/
 333
 334             if (gmx_mm256_any_lt(rsq00,rcutoff2))
 335             {
 336
 337             r00              = _mm256_mul_ps(rsq00,rinv00);
 338             r00              = _mm256_andnot_ps(dummy_mask,r00);
 339
 340             /* Compute parameters for interactions between i and j atoms */
 341             qq00             = _mm256_mul_ps(iq0,jq0);
 342
 343             /* EWALD ELECTROSTATICS */
 344
 345             /* Analytical PME correction */
 346             zeta2            = _mm256_mul_ps(beta2,rsq00);
 347             rinv3            = _mm256_mul_ps(rinvsq00,rinv00);
 348             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
 349             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
 350             felec            = _mm256_mul_ps(qq00,felec);
 351             pmecorrV         = gmx_mm256_pmecorrV_ps(zeta2);
 352             pmecorrV         = _mm256_mul_ps(pmecorrV,beta);
 353             velec            = _mm256_sub_ps(_mm256_sub_ps(rinv00,sh_ewald),pmecorrV);
 354             velec            = _mm256_mul_ps(qq00,velec);
 355
 356             cutoff_mask      = _mm256_cmp_ps(rsq00,rcutoff2,_CMP_LT_OQ);
 357
 358             /* Update potential sum for this i atom from the interaction with this j atom. */
 359             velec            = _mm256_and_ps(velec,cutoff_mask);
 360             velec            = _mm256_andnot_ps(dummy_mask,velec);
 361             velecsum         = _mm256_add_ps(velecsum,velec);
 362
 363             fscal            = felec;
 364
 365             fscal            = _mm256_and_ps(fscal,cutoff_mask);
 366
 367             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
 368
 369             /* Calculate temporary vectorial force */
 370             tx               = _mm256_mul_ps(fscal,dx00);
 371             ty               = _mm256_mul_ps(fscal,dy00);
 372             tz               = _mm256_mul_ps(fscal,dz00);
 373
 374             /* Update vectorial force */
 375             fix0             = _mm256_add_ps(fix0,tx);
 376             fiy0             = _mm256_add_ps(fiy0,ty);
 377             fiz0             = _mm256_add_ps(fiz0,tz);
 378
 379             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 380             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 381             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 382             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 383             fjptrE             = (jnrlistE>=0) ? f+j_coord_offsetE : scratch;
 384             fjptrF             = (jnrlistF>=0) ? f+j_coord_offsetF : scratch;
 385             fjptrG             = (jnrlistG>=0) ? f+j_coord_offsetG : scratch;
 386             fjptrH             = (jnrlistH>=0) ? f+j_coord_offsetH : scratch;
 387             gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,tx,ty,tz);
 388
 389             }
 390
 391             /* Inner loop uses 110 flops */
 392         }
 393
 394         /* End of innermost loop */
 395
 396         gmx_mm256_update_iforce_1atom_swizzle_ps(fix0,fiy0,fiz0,
 397                                                  f+i_coord_offset,fshift+i_shift_offset);
 398
 399         ggid                        = gid[iidx];
 400         /* Update potential energies */
 401         gmx_mm256_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid);
 402
 403         /* Increment number of inner iterations */
 404         inneriter                  += j_index_end - j_index_start;
 405
 406         /* Outer loop uses 8 flops */
 407     }
 408
 409     /* Increment number of outer iterations */
 410     outeriter        += nri;
 411
 412     /* Update outer/inner flops */
 413
 414     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VF,outeriter*8 + inneriter*110);
 415 }
 416 /*
 417  * Gromacs nonbonded kernel:   nb_kernel_ElecEwSh_VdwNone_GeomP1P1_F_avx_256_single
 418  * Electrostatics interaction: Ewald
 419  * VdW interaction:            None
 420  * Geometry:                   Particle-Particle
 421  * Calculate force/pot:        Force
 422  */
 423 void
 424 nb_kernel_ElecEwSh_VdwNone_GeomP1P1_F_avx_256_single
 425                     (t_nblist * gmx_restrict                nlist,
 426                      rvec * gmx_restrict                    xx,
 427                      rvec * gmx_restrict                    ff,
 428                      t_forcerec * gmx_restrict              fr,
 429                      t_mdatoms * gmx_restrict               mdatoms,
 430                      nb_kernel_data_t * gmx_restrict        kernel_data,
 431                      t_nrnb * gmx_restrict                  nrnb)
 432 {
 433     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 434      * just 0 for non-waters.
 435      * Suffixes A,B,C,D,E,F,G,H refer to j loop unrolling done with AVX, e.g. for the eight different
 436      * jnr indices corresponding to data put in the four positions in the SIMD register.
 437      */
 438     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 439     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 440     int              jnrA,jnrB,jnrC,jnrD;
 441     int              jnrE,jnrF,jnrG,jnrH;
 442     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
 443     int              jnrlistE,jnrlistF,jnrlistG,jnrlistH;
 444     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 445     int              j_coord_offsetE,j_coord_offsetF,j_coord_offsetG,j_coord_offsetH;
 446     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 447     real             rcutoff_scalar;
 448     real             *shiftvec,*fshift,*x,*f;
 449     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD,*fjptrE,*fjptrF,*fjptrG,*fjptrH;
 450     real             scratch[4*DIM];
 451     __m256           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 452     real *           vdwioffsetptr0;
 453     __m256           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 454     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D,vdwjidx0E,vdwjidx0F,vdwjidx0G,vdwjidx0H;
 455     __m256           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 456     __m256           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 457     __m256           velec,felec,velecsum,facel,crf,krf,krf2;
 458     real             *charge;
 459     __m256i          ewitab;
 460     __m128i          ewitab_lo,ewitab_hi;
 461     __m256           ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
 462     __m256           beta,beta2,beta3,zeta2,pmecorrF,pmecorrV,rinv3;
 463     real             *ewtab;
 464     __m256           dummy_mask,cutoff_mask;
 465     __m256           signbit = _mm256_castsi256_ps( _mm256_set1_epi32(0x80000000) );
 466     __m256           one     = _mm256_set1_ps(1.0);
 467     __m256           two     = _mm256_set1_ps(2.0);
 468     x                = xx[0];
 469     f                = ff[0];
 470
 471     nri              = nlist->nri;
 472     iinr             = nlist->iinr;
 473     jindex           = nlist->jindex;
 474     jjnr             = nlist->jjnr;
 475     shiftidx         = nlist->shift;
 476     gid              = nlist->gid;
 477     shiftvec         = fr->shift_vec[0];
 478     fshift           = fr->fshift[0];
 479     facel            = _mm256_set1_ps(fr->epsfac);
 480     charge           = mdatoms->chargeA;
 481
 482     sh_ewald         = _mm256_set1_ps(fr->ic->sh_ewald);
 483     beta             = _mm256_set1_ps(fr->ic->ewaldcoeff);
 484     beta2            = _mm256_mul_ps(beta,beta);
 485     beta3            = _mm256_mul_ps(beta,beta2);
 486
 487     ewtab            = fr->ic->tabq_coul_F;
 488     ewtabscale       = _mm256_set1_ps(fr->ic->tabq_scale);
 489     ewtabhalfspace   = _mm256_set1_ps(0.5/fr->ic->tabq_scale);
 490
 491     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 492     rcutoff_scalar   = fr->rcoulomb;
 493     rcutoff          = _mm256_set1_ps(rcutoff_scalar);
 494     rcutoff2         = _mm256_mul_ps(rcutoff,rcutoff);
 495
 496     /* Avoid stupid compiler warnings */
 497     jnrA = jnrB = jnrC = jnrD = jnrE = jnrF = jnrG = jnrH = 0;
 498     j_coord_offsetA = 0;
 499     j_coord_offsetB = 0;
 500     j_coord_offsetC = 0;
 501     j_coord_offsetD = 0;
 502     j_coord_offsetE = 0;
 503     j_coord_offsetF = 0;
 504     j_coord_offsetG = 0;
 505     j_coord_offsetH = 0;
 506
 507     outeriter        = 0;
 508     inneriter        = 0;
 509
 510     for(iidx=0;iidx<4*DIM;iidx++)
 511     {
 512         scratch[iidx] = 0.0;
 513     }
 514
 515     /* Start outer loop over neighborlists */
 516     for(iidx=0; iidx<nri; iidx++)
 517     {
 518         /* Load shift vector for this list */
 519         i_shift_offset   = DIM*shiftidx[iidx];
 520
 521         /* Load limits for loop over neighbors */
 522         j_index_start    = jindex[iidx];
 523         j_index_end      = jindex[iidx+1];
 524
 525         /* Get outer coordinate index */
 526         inr              = iinr[iidx];
 527         i_coord_offset   = DIM*inr;
 528
 529         /* Load i particle coords and add shift vector */
 530         gmx_mm256_load_shift_and_1rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 531
 532         fix0             = _mm256_setzero_ps();
 533         fiy0             = _mm256_setzero_ps();
 534         fiz0             = _mm256_setzero_ps();
 535
 536         /* Load parameters for i particles */
 537         iq0              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+0]));
 538
 539         /* Start inner kernel loop */
 540         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+7]>=0; jidx+=8)
 541         {
 542
 543             /* Get j neighbor index, and coordinate index */
 544             jnrA             = jjnr[jidx];
 545             jnrB             = jjnr[jidx+1];
 546             jnrC             = jjnr[jidx+2];
 547             jnrD             = jjnr[jidx+3];
 548             jnrE             = jjnr[jidx+4];
 549             jnrF             = jjnr[jidx+5];
 550             jnrG             = jjnr[jidx+6];
 551             jnrH             = jjnr[jidx+7];
 552             j_coord_offsetA  = DIM*jnrA;
 553             j_coord_offsetB  = DIM*jnrB;
 554             j_coord_offsetC  = DIM*jnrC;
 555             j_coord_offsetD  = DIM*jnrD;
 556             j_coord_offsetE  = DIM*jnrE;
 557             j_coord_offsetF  = DIM*jnrF;
 558             j_coord_offsetG  = DIM*jnrG;
 559             j_coord_offsetH  = DIM*jnrH;
 560
 561             /* load j atom coordinates */
 562             gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 563                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 564                                                  x+j_coord_offsetE,x+j_coord_offsetF,
 565                                                  x+j_coord_offsetG,x+j_coord_offsetH,
 566                                                  &jx0,&jy0,&jz0);
 567
 568             /* Calculate displacement vector */
 569             dx00             = _mm256_sub_ps(ix0,jx0);
 570             dy00             = _mm256_sub_ps(iy0,jy0);
 571             dz00             = _mm256_sub_ps(iz0,jz0);
 572
 573             /* Calculate squared distance and things based on it */
 574             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
 575
 576             rinv00           = gmx_mm256_invsqrt_ps(rsq00);
 577
 578             rinvsq00         = _mm256_mul_ps(rinv00,rinv00);
 579
 580             /* Load parameters for j particles */
 581             jq0              = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 582                                                                  charge+jnrC+0,charge+jnrD+0,
 583                                                                  charge+jnrE+0,charge+jnrF+0,
 584                                                                  charge+jnrG+0,charge+jnrH+0);
 585
 586             /**************************
 587              * CALCULATE INTERACTIONS *
 588              **************************/
 589
 590             if (gmx_mm256_any_lt(rsq00,rcutoff2))
 591             {
 592
 593             r00              = _mm256_mul_ps(rsq00,rinv00);
 594
 595             /* Compute parameters for interactions between i and j atoms */
 596             qq00             = _mm256_mul_ps(iq0,jq0);
 597
 598             /* EWALD ELECTROSTATICS */
 599
 600             /* Analytical PME correction */
 601             zeta2            = _mm256_mul_ps(beta2,rsq00);
 602             rinv3            = _mm256_mul_ps(rinvsq00,rinv00);
 603             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
 604             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
 605             felec            = _mm256_mul_ps(qq00,felec);
 606
 607             cutoff_mask      = _mm256_cmp_ps(rsq00,rcutoff2,_CMP_LT_OQ);
 608
 609             fscal            = felec;
 610
 611             fscal            = _mm256_and_ps(fscal,cutoff_mask);
 612
 613             /* Calculate temporary vectorial force */
 614             tx               = _mm256_mul_ps(fscal,dx00);
 615             ty               = _mm256_mul_ps(fscal,dy00);
 616             tz               = _mm256_mul_ps(fscal,dz00);
 617
 618             /* Update vectorial force */
 619             fix0             = _mm256_add_ps(fix0,tx);
 620             fiy0             = _mm256_add_ps(fiy0,ty);
 621             fiz0             = _mm256_add_ps(fiz0,tz);
 622
 623             fjptrA             = f+j_coord_offsetA;
 624             fjptrB             = f+j_coord_offsetB;
 625             fjptrC             = f+j_coord_offsetC;
 626             fjptrD             = f+j_coord_offsetD;
 627             fjptrE             = f+j_coord_offsetE;
 628             fjptrF             = f+j_coord_offsetF;
 629             fjptrG             = f+j_coord_offsetG;
 630             fjptrH             = f+j_coord_offsetH;
 631             gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,tx,ty,tz);
 632
 633             }
 634
 635             /* Inner loop uses 59 flops */
 636         }
 637
 638         if(jidx<j_index_end)
 639         {
 640
 641             /* Get j neighbor index, and coordinate index */
 642             jnrlistA         = jjnr[jidx];
 643             jnrlistB         = jjnr[jidx+1];
 644             jnrlistC         = jjnr[jidx+2];
 645             jnrlistD         = jjnr[jidx+3];
 646             jnrlistE         = jjnr[jidx+4];
 647             jnrlistF         = jjnr[jidx+5];
 648             jnrlistG         = jjnr[jidx+6];
 649             jnrlistH         = jjnr[jidx+7];
 650             /* Sign of each element will be negative for non-real atoms.
 651              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 652              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 653              */
 654             dummy_mask = gmx_mm256_set_m128(gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx+4)),_mm_setzero_si128())),
 655                                             gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128())));
 656
 657             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 658             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 659             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 660             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 661             jnrE       = (jnrlistE>=0) ? jnrlistE : 0;
 662             jnrF       = (jnrlistF>=0) ? jnrlistF : 0;
 663             jnrG       = (jnrlistG>=0) ? jnrlistG : 0;
 664             jnrH       = (jnrlistH>=0) ? jnrlistH : 0;
 665             j_coord_offsetA  = DIM*jnrA;
 666             j_coord_offsetB  = DIM*jnrB;
 667             j_coord_offsetC  = DIM*jnrC;
 668             j_coord_offsetD  = DIM*jnrD;
 669             j_coord_offsetE  = DIM*jnrE;
 670             j_coord_offsetF  = DIM*jnrF;
 671             j_coord_offsetG  = DIM*jnrG;
 672             j_coord_offsetH  = DIM*jnrH;
 673
 674             /* load j atom coordinates */
 675             gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 676                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 677                                                  x+j_coord_offsetE,x+j_coord_offsetF,
 678                                                  x+j_coord_offsetG,x+j_coord_offsetH,
 679                                                  &jx0,&jy0,&jz0);
 680
 681             /* Calculate displacement vector */
 682             dx00             = _mm256_sub_ps(ix0,jx0);
 683             dy00             = _mm256_sub_ps(iy0,jy0);
 684             dz00             = _mm256_sub_ps(iz0,jz0);
 685
 686             /* Calculate squared distance and things based on it */
 687             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
 688
 689             rinv00           = gmx_mm256_invsqrt_ps(rsq00);
 690
 691             rinvsq00         = _mm256_mul_ps(rinv00,rinv00);
 692
 693             /* Load parameters for j particles */
 694             jq0              = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 695                                                                  charge+jnrC+0,charge+jnrD+0,
 696                                                                  charge+jnrE+0,charge+jnrF+0,
 697                                                                  charge+jnrG+0,charge+jnrH+0);
 698
 699             /**************************
 700              * CALCULATE INTERACTIONS *
 701              **************************/
 702
 703             if (gmx_mm256_any_lt(rsq00,rcutoff2))
 704             {
 705
 706             r00              = _mm256_mul_ps(rsq00,rinv00);
 707             r00              = _mm256_andnot_ps(dummy_mask,r00);
 708
 709             /* Compute parameters for interactions between i and j atoms */
 710             qq00             = _mm256_mul_ps(iq0,jq0);
 711
 712             /* EWALD ELECTROSTATICS */
 713
 714             /* Analytical PME correction */
 715             zeta2            = _mm256_mul_ps(beta2,rsq00);
 716             rinv3            = _mm256_mul_ps(rinvsq00,rinv00);
 717             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
 718             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
 719             felec            = _mm256_mul_ps(qq00,felec);
 720
 721             cutoff_mask      = _mm256_cmp_ps(rsq00,rcutoff2,_CMP_LT_OQ);
 722
 723             fscal            = felec;
 724
 725             fscal            = _mm256_and_ps(fscal,cutoff_mask);
 726
 727             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
 728
 729             /* Calculate temporary vectorial force */
 730             tx               = _mm256_mul_ps(fscal,dx00);
 731             ty               = _mm256_mul_ps(fscal,dy00);
 732             tz               = _mm256_mul_ps(fscal,dz00);
 733
 734             /* Update vectorial force */
 735             fix0             = _mm256_add_ps(fix0,tx);
 736             fiy0             = _mm256_add_ps(fiy0,ty);
 737             fiz0             = _mm256_add_ps(fiz0,tz);
 738
 739             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 740             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 741             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 742             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 743             fjptrE             = (jnrlistE>=0) ? f+j_coord_offsetE : scratch;
 744             fjptrF             = (jnrlistF>=0) ? f+j_coord_offsetF : scratch;
 745             fjptrG             = (jnrlistG>=0) ? f+j_coord_offsetG : scratch;
 746             fjptrH             = (jnrlistH>=0) ? f+j_coord_offsetH : scratch;
 747             gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,tx,ty,tz);
 748
 749             }
 750
 751             /* Inner loop uses 60 flops */
 752         }
 753
 754         /* End of innermost loop */
 755
 756         gmx_mm256_update_iforce_1atom_swizzle_ps(fix0,fiy0,fiz0,
 757                                                  f+i_coord_offset,fshift+i_shift_offset);
 758
 759         /* Increment number of inner iterations */
 760         inneriter                  += j_index_end - j_index_start;
 761
 762         /* Outer loop uses 7 flops */
 763     }
 764
 765     /* Increment number of outer iterations */
 766     outeriter        += nri;
 767
 768     /* Update outer/inner flops */
 769
 770     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_F,outeriter*7 + inneriter*60);
 771 }