src/gmxlib/nonbonded/nb_kernel_avx_256_single/nb_kernel_ElecEw_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_ElecEw_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_ElecEw_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     /* Avoid stupid compiler warnings */
 112     jnrA = jnrB = jnrC = jnrD = jnrE = jnrF = jnrG = jnrH = 0;
 113     j_coord_offsetA = 0;
 114     j_coord_offsetB = 0;
 115     j_coord_offsetC = 0;
 116     j_coord_offsetD = 0;
 117     j_coord_offsetE = 0;
 118     j_coord_offsetF = 0;
 119     j_coord_offsetG = 0;
 120     j_coord_offsetH = 0;
 121
 122     outeriter        = 0;
 123     inneriter        = 0;
 124
 125     for(iidx=0;iidx<4*DIM;iidx++)
 126     {
 127         scratch[iidx] = 0.0;
 128     }
 129
 130     /* Start outer loop over neighborlists */
 131     for(iidx=0; iidx<nri; iidx++)
 132     {
 133         /* Load shift vector for this list */
 134         i_shift_offset   = DIM*shiftidx[iidx];
 135
 136         /* Load limits for loop over neighbors */
 137         j_index_start    = jindex[iidx];
 138         j_index_end      = jindex[iidx+1];
 139
 140         /* Get outer coordinate index */
 141         inr              = iinr[iidx];
 142         i_coord_offset   = DIM*inr;
 143
 144         /* Load i particle coords and add shift vector */
 145         gmx_mm256_load_shift_and_1rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 146
 147         fix0             = _mm256_setzero_ps();
 148         fiy0             = _mm256_setzero_ps();
 149         fiz0             = _mm256_setzero_ps();
 150
 151         /* Load parameters for i particles */
 152         iq0              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+0]));
 153
 154         /* Reset potential sums */
 155         velecsum         = _mm256_setzero_ps();
 156
 157         /* Start inner kernel loop */
 158         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+7]>=0; jidx+=8)
 159         {
 160
 161             /* Get j neighbor index, and coordinate index */
 162             jnrA             = jjnr[jidx];
 163             jnrB             = jjnr[jidx+1];
 164             jnrC             = jjnr[jidx+2];
 165             jnrD             = jjnr[jidx+3];
 166             jnrE             = jjnr[jidx+4];
 167             jnrF             = jjnr[jidx+5];
 168             jnrG             = jjnr[jidx+6];
 169             jnrH             = jjnr[jidx+7];
 170             j_coord_offsetA  = DIM*jnrA;
 171             j_coord_offsetB  = DIM*jnrB;
 172             j_coord_offsetC  = DIM*jnrC;
 173             j_coord_offsetD  = DIM*jnrD;
 174             j_coord_offsetE  = DIM*jnrE;
 175             j_coord_offsetF  = DIM*jnrF;
 176             j_coord_offsetG  = DIM*jnrG;
 177             j_coord_offsetH  = DIM*jnrH;
 178
 179             /* load j atom coordinates */
 180             gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 181                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 182                                                  x+j_coord_offsetE,x+j_coord_offsetF,
 183                                                  x+j_coord_offsetG,x+j_coord_offsetH,
 184                                                  &jx0,&jy0,&jz0);
 185
 186             /* Calculate displacement vector */
 187             dx00             = _mm256_sub_ps(ix0,jx0);
 188             dy00             = _mm256_sub_ps(iy0,jy0);
 189             dz00             = _mm256_sub_ps(iz0,jz0);
 190
 191             /* Calculate squared distance and things based on it */
 192             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
 193
 194             rinv00           = gmx_mm256_invsqrt_ps(rsq00);
 195
 196             rinvsq00         = _mm256_mul_ps(rinv00,rinv00);
 197
 198             /* Load parameters for j particles */
 199             jq0              = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 200                                                                  charge+jnrC+0,charge+jnrD+0,
 201                                                                  charge+jnrE+0,charge+jnrF+0,
 202                                                                  charge+jnrG+0,charge+jnrH+0);
 203
 204             /**************************
 205              * CALCULATE INTERACTIONS *
 206              **************************/
 207
 208             r00              = _mm256_mul_ps(rsq00,rinv00);
 209
 210             /* Compute parameters for interactions between i and j atoms */
 211             qq00             = _mm256_mul_ps(iq0,jq0);
 212
 213             /* EWALD ELECTROSTATICS */
 214
 215             /* Analytical PME correction */
 216             zeta2            = _mm256_mul_ps(beta2,rsq00);
 217             rinv3            = _mm256_mul_ps(rinvsq00,rinv00);
 218             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
 219             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
 220             felec            = _mm256_mul_ps(qq00,felec);
 221             pmecorrV         = gmx_mm256_pmecorrV_ps(zeta2);
 222             pmecorrV         = _mm256_mul_ps(pmecorrV,beta);
 223             velec            = _mm256_sub_ps(rinv00,pmecorrV);
 224             velec            = _mm256_mul_ps(qq00,velec);
 225
 226             /* Update potential sum for this i atom from the interaction with this j atom. */
 227             velecsum         = _mm256_add_ps(velecsum,velec);
 228
 229             fscal            = felec;
 230
 231             /* Calculate temporary vectorial force */
 232             tx               = _mm256_mul_ps(fscal,dx00);
 233             ty               = _mm256_mul_ps(fscal,dy00);
 234             tz               = _mm256_mul_ps(fscal,dz00);
 235
 236             /* Update vectorial force */
 237             fix0             = _mm256_add_ps(fix0,tx);
 238             fiy0             = _mm256_add_ps(fiy0,ty);
 239             fiz0             = _mm256_add_ps(fiz0,tz);
 240
 241             fjptrA             = f+j_coord_offsetA;
 242             fjptrB             = f+j_coord_offsetB;
 243             fjptrC             = f+j_coord_offsetC;
 244             fjptrD             = f+j_coord_offsetD;
 245             fjptrE             = f+j_coord_offsetE;
 246             fjptrF             = f+j_coord_offsetF;
 247             fjptrG             = f+j_coord_offsetG;
 248             fjptrH             = f+j_coord_offsetH;
 249             gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,tx,ty,tz);
 250
 251             /* Inner loop uses 84 flops */
 252         }
 253
 254         if(jidx<j_index_end)
 255         {
 256
 257             /* Get j neighbor index, and coordinate index */
 258             jnrlistA         = jjnr[jidx];
 259             jnrlistB         = jjnr[jidx+1];
 260             jnrlistC         = jjnr[jidx+2];
 261             jnrlistD         = jjnr[jidx+3];
 262             jnrlistE         = jjnr[jidx+4];
 263             jnrlistF         = jjnr[jidx+5];
 264             jnrlistG         = jjnr[jidx+6];
 265             jnrlistH         = jjnr[jidx+7];
 266             /* Sign of each element will be negative for non-real atoms.
 267              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 268              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 269              */
 270             dummy_mask = gmx_mm256_set_m128(gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx+4)),_mm_setzero_si128())),
 271                                             gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128())));
 272
 273             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 274             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 275             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 276             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 277             jnrE       = (jnrlistE>=0) ? jnrlistE : 0;
 278             jnrF       = (jnrlistF>=0) ? jnrlistF : 0;
 279             jnrG       = (jnrlistG>=0) ? jnrlistG : 0;
 280             jnrH       = (jnrlistH>=0) ? jnrlistH : 0;
 281             j_coord_offsetA  = DIM*jnrA;
 282             j_coord_offsetB  = DIM*jnrB;
 283             j_coord_offsetC  = DIM*jnrC;
 284             j_coord_offsetD  = DIM*jnrD;
 285             j_coord_offsetE  = DIM*jnrE;
 286             j_coord_offsetF  = DIM*jnrF;
 287             j_coord_offsetG  = DIM*jnrG;
 288             j_coord_offsetH  = DIM*jnrH;
 289
 290             /* load j atom coordinates */
 291             gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 292                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 293                                                  x+j_coord_offsetE,x+j_coord_offsetF,
 294                                                  x+j_coord_offsetG,x+j_coord_offsetH,
 295                                                  &jx0,&jy0,&jz0);
 296
 297             /* Calculate displacement vector */
 298             dx00             = _mm256_sub_ps(ix0,jx0);
 299             dy00             = _mm256_sub_ps(iy0,jy0);
 300             dz00             = _mm256_sub_ps(iz0,jz0);
 301
 302             /* Calculate squared distance and things based on it */
 303             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
 304
 305             rinv00           = gmx_mm256_invsqrt_ps(rsq00);
 306
 307             rinvsq00         = _mm256_mul_ps(rinv00,rinv00);
 308
 309             /* Load parameters for j particles */
 310             jq0              = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 311                                                                  charge+jnrC+0,charge+jnrD+0,
 312                                                                  charge+jnrE+0,charge+jnrF+0,
 313                                                                  charge+jnrG+0,charge+jnrH+0);
 314
 315             /**************************
 316              * CALCULATE INTERACTIONS *
 317              **************************/
 318
 319             r00              = _mm256_mul_ps(rsq00,rinv00);
 320             r00              = _mm256_andnot_ps(dummy_mask,r00);
 321
 322             /* Compute parameters for interactions between i and j atoms */
 323             qq00             = _mm256_mul_ps(iq0,jq0);
 324
 325             /* EWALD ELECTROSTATICS */
 326
 327             /* Analytical PME correction */
 328             zeta2            = _mm256_mul_ps(beta2,rsq00);
 329             rinv3            = _mm256_mul_ps(rinvsq00,rinv00);
 330             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
 331             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
 332             felec            = _mm256_mul_ps(qq00,felec);
 333             pmecorrV         = gmx_mm256_pmecorrV_ps(zeta2);
 334             pmecorrV         = _mm256_mul_ps(pmecorrV,beta);
 335             velec            = _mm256_sub_ps(rinv00,pmecorrV);
 336             velec            = _mm256_mul_ps(qq00,velec);
 337
 338             /* Update potential sum for this i atom from the interaction with this j atom. */
 339             velec            = _mm256_andnot_ps(dummy_mask,velec);
 340             velecsum         = _mm256_add_ps(velecsum,velec);
 341
 342             fscal            = felec;
 343
 344             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
 345
 346             /* Calculate temporary vectorial force */
 347             tx               = _mm256_mul_ps(fscal,dx00);
 348             ty               = _mm256_mul_ps(fscal,dy00);
 349             tz               = _mm256_mul_ps(fscal,dz00);
 350
 351             /* Update vectorial force */
 352             fix0             = _mm256_add_ps(fix0,tx);
 353             fiy0             = _mm256_add_ps(fiy0,ty);
 354             fiz0             = _mm256_add_ps(fiz0,tz);
 355
 356             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 357             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 358             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 359             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 360             fjptrE             = (jnrlistE>=0) ? f+j_coord_offsetE : scratch;
 361             fjptrF             = (jnrlistF>=0) ? f+j_coord_offsetF : scratch;
 362             fjptrG             = (jnrlistG>=0) ? f+j_coord_offsetG : scratch;
 363             fjptrH             = (jnrlistH>=0) ? f+j_coord_offsetH : scratch;
 364             gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,tx,ty,tz);
 365
 366             /* Inner loop uses 85 flops */
 367         }
 368
 369         /* End of innermost loop */
 370
 371         gmx_mm256_update_iforce_1atom_swizzle_ps(fix0,fiy0,fiz0,
 372                                                  f+i_coord_offset,fshift+i_shift_offset);
 373
 374         ggid                        = gid[iidx];
 375         /* Update potential energies */
 376         gmx_mm256_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid);
 377
 378         /* Increment number of inner iterations */
 379         inneriter                  += j_index_end - j_index_start;
 380
 381         /* Outer loop uses 8 flops */
 382     }
 383
 384     /* Increment number of outer iterations */
 385     outeriter        += nri;
 386
 387     /* Update outer/inner flops */
 388
 389     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VF,outeriter*8 + inneriter*85);
 390 }
 391 /*
 392  * Gromacs nonbonded kernel:   nb_kernel_ElecEw_VdwNone_GeomP1P1_F_avx_256_single
 393  * Electrostatics interaction: Ewald
 394  * VdW interaction:            None
 395  * Geometry:                   Particle-Particle
 396  * Calculate force/pot:        Force
 397  */
 398 void
 399 nb_kernel_ElecEw_VdwNone_GeomP1P1_F_avx_256_single
 400                     (t_nblist * gmx_restrict                nlist,
 401                      rvec * gmx_restrict                    xx,
 402                      rvec * gmx_restrict                    ff,
 403                      t_forcerec * gmx_restrict              fr,
 404                      t_mdatoms * gmx_restrict               mdatoms,
 405                      nb_kernel_data_t * gmx_restrict        kernel_data,
 406                      t_nrnb * gmx_restrict                  nrnb)
 407 {
 408     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 409      * just 0 for non-waters.
 410      * Suffixes A,B,C,D,E,F,G,H refer to j loop unrolling done with AVX, e.g. for the eight different
 411      * jnr indices corresponding to data put in the four positions in the SIMD register.
 412      */
 413     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 414     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 415     int              jnrA,jnrB,jnrC,jnrD;
 416     int              jnrE,jnrF,jnrG,jnrH;
 417     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
 418     int              jnrlistE,jnrlistF,jnrlistG,jnrlistH;
 419     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 420     int              j_coord_offsetE,j_coord_offsetF,j_coord_offsetG,j_coord_offsetH;
 421     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 422     real             rcutoff_scalar;
 423     real             *shiftvec,*fshift,*x,*f;
 424     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD,*fjptrE,*fjptrF,*fjptrG,*fjptrH;
 425     real             scratch[4*DIM];
 426     __m256           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 427     real *           vdwioffsetptr0;
 428     __m256           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 429     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D,vdwjidx0E,vdwjidx0F,vdwjidx0G,vdwjidx0H;
 430     __m256           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 431     __m256           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 432     __m256           velec,felec,velecsum,facel,crf,krf,krf2;
 433     real             *charge;
 434     __m256i          ewitab;
 435     __m128i          ewitab_lo,ewitab_hi;
 436     __m256           ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
 437     __m256           beta,beta2,beta3,zeta2,pmecorrF,pmecorrV,rinv3;
 438     real             *ewtab;
 439     __m256           dummy_mask,cutoff_mask;
 440     __m256           signbit = _mm256_castsi256_ps( _mm256_set1_epi32(0x80000000) );
 441     __m256           one     = _mm256_set1_ps(1.0);
 442     __m256           two     = _mm256_set1_ps(2.0);
 443     x                = xx[0];
 444     f                = ff[0];
 445
 446     nri              = nlist->nri;
 447     iinr             = nlist->iinr;
 448     jindex           = nlist->jindex;
 449     jjnr             = nlist->jjnr;
 450     shiftidx         = nlist->shift;
 451     gid              = nlist->gid;
 452     shiftvec         = fr->shift_vec[0];
 453     fshift           = fr->fshift[0];
 454     facel            = _mm256_set1_ps(fr->epsfac);
 455     charge           = mdatoms->chargeA;
 456
 457     sh_ewald         = _mm256_set1_ps(fr->ic->sh_ewald);
 458     beta             = _mm256_set1_ps(fr->ic->ewaldcoeff);
 459     beta2            = _mm256_mul_ps(beta,beta);
 460     beta3            = _mm256_mul_ps(beta,beta2);
 461
 462     ewtab            = fr->ic->tabq_coul_F;
 463     ewtabscale       = _mm256_set1_ps(fr->ic->tabq_scale);
 464     ewtabhalfspace   = _mm256_set1_ps(0.5/fr->ic->tabq_scale);
 465
 466     /* Avoid stupid compiler warnings */
 467     jnrA = jnrB = jnrC = jnrD = jnrE = jnrF = jnrG = jnrH = 0;
 468     j_coord_offsetA = 0;
 469     j_coord_offsetB = 0;
 470     j_coord_offsetC = 0;
 471     j_coord_offsetD = 0;
 472     j_coord_offsetE = 0;
 473     j_coord_offsetF = 0;
 474     j_coord_offsetG = 0;
 475     j_coord_offsetH = 0;
 476
 477     outeriter        = 0;
 478     inneriter        = 0;
 479
 480     for(iidx=0;iidx<4*DIM;iidx++)
 481     {
 482         scratch[iidx] = 0.0;
 483     }
 484
 485     /* Start outer loop over neighborlists */
 486     for(iidx=0; iidx<nri; iidx++)
 487     {
 488         /* Load shift vector for this list */
 489         i_shift_offset   = DIM*shiftidx[iidx];
 490
 491         /* Load limits for loop over neighbors */
 492         j_index_start    = jindex[iidx];
 493         j_index_end      = jindex[iidx+1];
 494
 495         /* Get outer coordinate index */
 496         inr              = iinr[iidx];
 497         i_coord_offset   = DIM*inr;
 498
 499         /* Load i particle coords and add shift vector */
 500         gmx_mm256_load_shift_and_1rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 501
 502         fix0             = _mm256_setzero_ps();
 503         fiy0             = _mm256_setzero_ps();
 504         fiz0             = _mm256_setzero_ps();
 505
 506         /* Load parameters for i particles */
 507         iq0              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+0]));
 508
 509         /* Start inner kernel loop */
 510         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+7]>=0; jidx+=8)
 511         {
 512
 513             /* Get j neighbor index, and coordinate index */
 514             jnrA             = jjnr[jidx];
 515             jnrB             = jjnr[jidx+1];
 516             jnrC             = jjnr[jidx+2];
 517             jnrD             = jjnr[jidx+3];
 518             jnrE             = jjnr[jidx+4];
 519             jnrF             = jjnr[jidx+5];
 520             jnrG             = jjnr[jidx+6];
 521             jnrH             = jjnr[jidx+7];
 522             j_coord_offsetA  = DIM*jnrA;
 523             j_coord_offsetB  = DIM*jnrB;
 524             j_coord_offsetC  = DIM*jnrC;
 525             j_coord_offsetD  = DIM*jnrD;
 526             j_coord_offsetE  = DIM*jnrE;
 527             j_coord_offsetF  = DIM*jnrF;
 528             j_coord_offsetG  = DIM*jnrG;
 529             j_coord_offsetH  = DIM*jnrH;
 530
 531             /* load j atom coordinates */
 532             gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 533                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 534                                                  x+j_coord_offsetE,x+j_coord_offsetF,
 535                                                  x+j_coord_offsetG,x+j_coord_offsetH,
 536                                                  &jx0,&jy0,&jz0);
 537
 538             /* Calculate displacement vector */
 539             dx00             = _mm256_sub_ps(ix0,jx0);
 540             dy00             = _mm256_sub_ps(iy0,jy0);
 541             dz00             = _mm256_sub_ps(iz0,jz0);
 542
 543             /* Calculate squared distance and things based on it */
 544             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
 545
 546             rinv00           = gmx_mm256_invsqrt_ps(rsq00);
 547
 548             rinvsq00         = _mm256_mul_ps(rinv00,rinv00);
 549
 550             /* Load parameters for j particles */
 551             jq0              = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 552                                                                  charge+jnrC+0,charge+jnrD+0,
 553                                                                  charge+jnrE+0,charge+jnrF+0,
 554                                                                  charge+jnrG+0,charge+jnrH+0);
 555
 556             /**************************
 557              * CALCULATE INTERACTIONS *
 558              **************************/
 559
 560             r00              = _mm256_mul_ps(rsq00,rinv00);
 561
 562             /* Compute parameters for interactions between i and j atoms */
 563             qq00             = _mm256_mul_ps(iq0,jq0);
 564
 565             /* EWALD ELECTROSTATICS */
 566
 567             /* Analytical PME correction */
 568             zeta2            = _mm256_mul_ps(beta2,rsq00);
 569             rinv3            = _mm256_mul_ps(rinvsq00,rinv00);
 570             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
 571             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
 572             felec            = _mm256_mul_ps(qq00,felec);
 573
 574             fscal            = felec;
 575
 576             /* Calculate temporary vectorial force */
 577             tx               = _mm256_mul_ps(fscal,dx00);
 578             ty               = _mm256_mul_ps(fscal,dy00);
 579             tz               = _mm256_mul_ps(fscal,dz00);
 580
 581             /* Update vectorial force */
 582             fix0             = _mm256_add_ps(fix0,tx);
 583             fiy0             = _mm256_add_ps(fiy0,ty);
 584             fiz0             = _mm256_add_ps(fiz0,tz);
 585
 586             fjptrA             = f+j_coord_offsetA;
 587             fjptrB             = f+j_coord_offsetB;
 588             fjptrC             = f+j_coord_offsetC;
 589             fjptrD             = f+j_coord_offsetD;
 590             fjptrE             = f+j_coord_offsetE;
 591             fjptrF             = f+j_coord_offsetF;
 592             fjptrG             = f+j_coord_offsetG;
 593             fjptrH             = f+j_coord_offsetH;
 594             gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,tx,ty,tz);
 595
 596             /* Inner loop uses 56 flops */
 597         }
 598
 599         if(jidx<j_index_end)
 600         {
 601
 602             /* Get j neighbor index, and coordinate index */
 603             jnrlistA         = jjnr[jidx];
 604             jnrlistB         = jjnr[jidx+1];
 605             jnrlistC         = jjnr[jidx+2];
 606             jnrlistD         = jjnr[jidx+3];
 607             jnrlistE         = jjnr[jidx+4];
 608             jnrlistF         = jjnr[jidx+5];
 609             jnrlistG         = jjnr[jidx+6];
 610             jnrlistH         = jjnr[jidx+7];
 611             /* Sign of each element will be negative for non-real atoms.
 612              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 613              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 614              */
 615             dummy_mask = gmx_mm256_set_m128(gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx+4)),_mm_setzero_si128())),
 616                                             gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128())));
 617
 618             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 619             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 620             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 621             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 622             jnrE       = (jnrlistE>=0) ? jnrlistE : 0;
 623             jnrF       = (jnrlistF>=0) ? jnrlistF : 0;
 624             jnrG       = (jnrlistG>=0) ? jnrlistG : 0;
 625             jnrH       = (jnrlistH>=0) ? jnrlistH : 0;
 626             j_coord_offsetA  = DIM*jnrA;
 627             j_coord_offsetB  = DIM*jnrB;
 628             j_coord_offsetC  = DIM*jnrC;
 629             j_coord_offsetD  = DIM*jnrD;
 630             j_coord_offsetE  = DIM*jnrE;
 631             j_coord_offsetF  = DIM*jnrF;
 632             j_coord_offsetG  = DIM*jnrG;
 633             j_coord_offsetH  = DIM*jnrH;
 634
 635             /* load j atom coordinates */
 636             gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 637                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 638                                                  x+j_coord_offsetE,x+j_coord_offsetF,
 639                                                  x+j_coord_offsetG,x+j_coord_offsetH,
 640                                                  &jx0,&jy0,&jz0);
 641
 642             /* Calculate displacement vector */
 643             dx00             = _mm256_sub_ps(ix0,jx0);
 644             dy00             = _mm256_sub_ps(iy0,jy0);
 645             dz00             = _mm256_sub_ps(iz0,jz0);
 646
 647             /* Calculate squared distance and things based on it */
 648             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
 649
 650             rinv00           = gmx_mm256_invsqrt_ps(rsq00);
 651
 652             rinvsq00         = _mm256_mul_ps(rinv00,rinv00);
 653
 654             /* Load parameters for j particles */
 655             jq0              = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 656                                                                  charge+jnrC+0,charge+jnrD+0,
 657                                                                  charge+jnrE+0,charge+jnrF+0,
 658                                                                  charge+jnrG+0,charge+jnrH+0);
 659
 660             /**************************
 661              * CALCULATE INTERACTIONS *
 662              **************************/
 663
 664             r00              = _mm256_mul_ps(rsq00,rinv00);
 665             r00              = _mm256_andnot_ps(dummy_mask,r00);
 666
 667             /* Compute parameters for interactions between i and j atoms */
 668             qq00             = _mm256_mul_ps(iq0,jq0);
 669
 670             /* EWALD ELECTROSTATICS */
 671
 672             /* Analytical PME correction */
 673             zeta2            = _mm256_mul_ps(beta2,rsq00);
 674             rinv3            = _mm256_mul_ps(rinvsq00,rinv00);
 675             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
 676             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
 677             felec            = _mm256_mul_ps(qq00,felec);
 678
 679             fscal            = felec;
 680
 681             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
 682
 683             /* Calculate temporary vectorial force */
 684             tx               = _mm256_mul_ps(fscal,dx00);
 685             ty               = _mm256_mul_ps(fscal,dy00);
 686             tz               = _mm256_mul_ps(fscal,dz00);
 687
 688             /* Update vectorial force */
 689             fix0             = _mm256_add_ps(fix0,tx);
 690             fiy0             = _mm256_add_ps(fiy0,ty);
 691             fiz0             = _mm256_add_ps(fiz0,tz);
 692
 693             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 694             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 695             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 696             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 697             fjptrE             = (jnrlistE>=0) ? f+j_coord_offsetE : scratch;
 698             fjptrF             = (jnrlistF>=0) ? f+j_coord_offsetF : scratch;
 699             fjptrG             = (jnrlistG>=0) ? f+j_coord_offsetG : scratch;
 700             fjptrH             = (jnrlistH>=0) ? f+j_coord_offsetH : scratch;
 701             gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,tx,ty,tz);
 702
 703             /* Inner loop uses 57 flops */
 704         }
 705
 706         /* End of innermost loop */
 707
 708         gmx_mm256_update_iforce_1atom_swizzle_ps(fix0,fiy0,fiz0,
 709                                                  f+i_coord_offset,fshift+i_shift_offset);
 710
 711         /* Increment number of inner iterations */
 712         inneriter                  += j_index_end - j_index_start;
 713
 714         /* Outer loop uses 7 flops */
 715     }
 716
 717     /* Increment number of outer iterations */
 718     outeriter        += nri;
 719
 720     /* Update outer/inner flops */
 721
 722     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_F,outeriter*7 + inneriter*57);
 723 }