src/gromacs/gmxlib/nonbonded/nb_kernel_avx_128_fma_double/nb_kernel_ElecEwSw_VdwNone_GeomP1P1_avx_128_fma_double.cpp

   1 /*
   2  * This file is part of the GROMACS molecular simulation package.
   3  *
   4  * Copyright (c) 2012,2013,2014,2015,2017,2018, by the GROMACS development team, led by
   5  * Mark Abraham, David van der Spoel, Berk Hess, and Erik Lindahl,
   6  * and including many others, as listed in the AUTHORS file in the
   7  * top-level source directory and at http://www.gromacs.org.
   8  *
   9  * GROMACS is free software; you can redistribute it and/or
  10  * modify it under the terms of the GNU Lesser General Public License
  11  * as published by the Free Software Foundation; either version 2.1
  12  * of the License, or (at your option) any later version.
  13  *
  14  * GROMACS is distributed in the hope that it will be useful,
  15  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  16  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  17  * Lesser General Public License for more details.
  18  *
  19  * You should have received a copy of the GNU Lesser General Public
  20  * License along with GROMACS; if not, see
  21  * http://www.gnu.org/licenses, or write to the Free Software Foundation,
  22  * Inc., 51 Franklin Street, Fifth Floor, Boston, MA  02110-1301  USA.
  23  *
  24  * If you want to redistribute modifications to GROMACS, please
  25  * consider that scientific software is very special. Version
  26  * control is crucial - bugs must be traceable. We will be happy to
  27  * consider code for inclusion in the official distribution, but
  28  * derived work must not be called official GROMACS. Details are found
  29  * in the README & COPYING files - if they are missing, get the
  30  * official version at http://www.gromacs.org.
  31  *
  32  * To help us fund GROMACS development, we humbly ask that you cite
  33  * the research papers on the package. Check out http://www.gromacs.org.
  34  */
  35 /*
  36  * Note: this file was generated by the GROMACS avx_128_fma_double kernel generator.
  37  */
  38 #include "gmxpre.h"
  39
  40 #include "config.h"
  41
  42 #include <math.h>
  43
  44 #include "../nb_kernel.h"
  45 #include "gromacs/gmxlib/nrnb.h"
  46
  47 #include "kernelutil_x86_avx_128_fma_double.h"
  48
  49 /*
  50  * Gromacs nonbonded kernel:   nb_kernel_ElecEwSw_VdwNone_GeomP1P1_VF_avx_128_fma_double
  51  * Electrostatics interaction: Ewald
  52  * VdW interaction:            None
  53  * Geometry:                   Particle-Particle
  54  * Calculate force/pot:        PotentialAndForce
  55  */
  56 void
  57 nb_kernel_ElecEwSw_VdwNone_GeomP1P1_VF_avx_128_fma_double
  58                     (t_nblist                    * gmx_restrict       nlist,
  59                      rvec                        * gmx_restrict          xx,
  60                      rvec                        * gmx_restrict          ff,
  61                      struct t_forcerec           * gmx_restrict          fr,
  62                      t_mdatoms                   * gmx_restrict     mdatoms,
  63                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
  64                      t_nrnb                      * gmx_restrict        nrnb)
  65 {
  66     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
  67      * just 0 for non-waters.
  68      * Suffixes A,B refer to j loop unrolling done with SSE double precision, e.g. for the two different
  69      * jnr indices corresponding to data put in the four positions in the SIMD register.
  70      */
  71     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
  72     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
  73     int              jnrA,jnrB;
  74     int              j_coord_offsetA,j_coord_offsetB;
  75     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
  76     real             rcutoff_scalar;
  77     real             *shiftvec,*fshift,*x,*f;
  78     __m128d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
  79     int              vdwioffset0;
  80     __m128d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
  81     int              vdwjidx0A,vdwjidx0B;
  82     __m128d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  83     __m128d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  84     __m128d          velec,felec,velecsum,facel,crf,krf,krf2;
  85     real             *charge;
  86     __m128i          ewitab;
  87     __m128d          ewtabscale,eweps,twoeweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
  88     real             *ewtab;
  89     __m128d          rswitch,swV3,swV4,swV5,swF2,swF3,swF4,d,d2,sw,dsw;
  90     real             rswitch_scalar,d_scalar;
  91     __m128d          dummy_mask,cutoff_mask;
  92     __m128d          signbit   = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
  93     __m128d          one     = _mm_set1_pd(1.0);
  94     __m128d          two     = _mm_set1_pd(2.0);
  95     x                = xx[0];
  96     f                = ff[0];
  97
  98     nri              = nlist->nri;
  99     iinr             = nlist->iinr;
 100     jindex           = nlist->jindex;
 101     jjnr             = nlist->jjnr;
 102     shiftidx         = nlist->shift;
 103     gid              = nlist->gid;
 104     shiftvec         = fr->shift_vec[0];
 105     fshift           = fr->fshift[0];
 106     facel            = _mm_set1_pd(fr->ic->epsfac);
 107     charge           = mdatoms->chargeA;
 108
 109     sh_ewald         = _mm_set1_pd(fr->ic->sh_ewald);
 110     ewtab            = fr->ic->tabq_coul_FDV0;
 111     ewtabscale       = _mm_set1_pd(fr->ic->tabq_scale);
 112     ewtabhalfspace   = _mm_set1_pd(0.5/fr->ic->tabq_scale);
 113
 114     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 115     rcutoff_scalar   = fr->ic->rcoulomb;
 116     rcutoff          = _mm_set1_pd(rcutoff_scalar);
 117     rcutoff2         = _mm_mul_pd(rcutoff,rcutoff);
 118
 119     rswitch_scalar   = fr->ic->rcoulomb_switch;
 120     rswitch          = _mm_set1_pd(rswitch_scalar);
 121     /* Setup switch parameters */
 122     d_scalar         = rcutoff_scalar-rswitch_scalar;
 123     d                = _mm_set1_pd(d_scalar);
 124     swV3             = _mm_set1_pd(-10.0/(d_scalar*d_scalar*d_scalar));
 125     swV4             = _mm_set1_pd( 15.0/(d_scalar*d_scalar*d_scalar*d_scalar));
 126     swV5             = _mm_set1_pd( -6.0/(d_scalar*d_scalar*d_scalar*d_scalar*d_scalar));
 127     swF2             = _mm_set1_pd(-30.0/(d_scalar*d_scalar*d_scalar));
 128     swF3             = _mm_set1_pd( 60.0/(d_scalar*d_scalar*d_scalar*d_scalar));
 129     swF4             = _mm_set1_pd(-30.0/(d_scalar*d_scalar*d_scalar*d_scalar*d_scalar));
 130
 131     /* Avoid stupid compiler warnings */
 132     jnrA = jnrB = 0;
 133     j_coord_offsetA = 0;
 134     j_coord_offsetB = 0;
 135
 136     outeriter        = 0;
 137     inneriter        = 0;
 138
 139     /* Start outer loop over neighborlists */
 140     for(iidx=0; iidx<nri; iidx++)
 141     {
 142         /* Load shift vector for this list */
 143         i_shift_offset   = DIM*shiftidx[iidx];
 144
 145         /* Load limits for loop over neighbors */
 146         j_index_start    = jindex[iidx];
 147         j_index_end      = jindex[iidx+1];
 148
 149         /* Get outer coordinate index */
 150         inr              = iinr[iidx];
 151         i_coord_offset   = DIM*inr;
 152
 153         /* Load i particle coords and add shift vector */
 154         gmx_mm_load_shift_and_1rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 155
 156         fix0             = _mm_setzero_pd();
 157         fiy0             = _mm_setzero_pd();
 158         fiz0             = _mm_setzero_pd();
 159
 160         /* Load parameters for i particles */
 161         iq0              = _mm_mul_pd(facel,_mm_load1_pd(charge+inr+0));
 162
 163         /* Reset potential sums */
 164         velecsum         = _mm_setzero_pd();
 165
 166         /* Start inner kernel loop */
 167         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 168         {
 169
 170             /* Get j neighbor index, and coordinate index */
 171             jnrA             = jjnr[jidx];
 172             jnrB             = jjnr[jidx+1];
 173             j_coord_offsetA  = DIM*jnrA;
 174             j_coord_offsetB  = DIM*jnrB;
 175
 176             /* load j atom coordinates */
 177             gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 178                                               &jx0,&jy0,&jz0);
 179
 180             /* Calculate displacement vector */
 181             dx00             = _mm_sub_pd(ix0,jx0);
 182             dy00             = _mm_sub_pd(iy0,jy0);
 183             dz00             = _mm_sub_pd(iz0,jz0);
 184
 185             /* Calculate squared distance and things based on it */
 186             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 187
 188             rinv00           = avx128fma_invsqrt_d(rsq00);
 189
 190             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 191
 192             /* Load parameters for j particles */
 193             jq0              = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
 194
 195             /**************************
 196              * CALCULATE INTERACTIONS *
 197              **************************/
 198
 199             if (gmx_mm_any_lt(rsq00,rcutoff2))
 200             {
 201
 202             r00              = _mm_mul_pd(rsq00,rinv00);
 203
 204             /* Compute parameters for interactions between i and j atoms */
 205             qq00             = _mm_mul_pd(iq0,jq0);
 206
 207             /* EWALD ELECTROSTATICS */
 208
 209             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 210             ewrt             = _mm_mul_pd(r00,ewtabscale);
 211             ewitab           = _mm_cvttpd_epi32(ewrt);
 212 #ifdef __XOP__
 213             eweps            = _mm_frcz_pd(ewrt);
 214 #else
 215             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 216 #endif
 217             twoeweps         = _mm_add_pd(eweps,eweps);
 218             ewitab           = _mm_slli_epi32(ewitab,2);
 219             ewtabF           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 220             ewtabD           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
 221             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 222             ewtabV           = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
 223             ewtabFn          = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,1) +2);
 224             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 225             felec            = _mm_macc_pd(eweps,ewtabD,ewtabF);
 226             velec            = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
 227             velec            = _mm_mul_pd(qq00,_mm_sub_pd(rinv00,velec));
 228             felec            = _mm_mul_pd(_mm_mul_pd(qq00,rinv00),_mm_sub_pd(rinvsq00,felec));
 229
 230             d                = _mm_sub_pd(r00,rswitch);
 231             d                = _mm_max_pd(d,_mm_setzero_pd());
 232             d2               = _mm_mul_pd(d,d);
 233             sw               = _mm_add_pd(one,_mm_mul_pd(d2,_mm_mul_pd(d,_mm_macc_pd(d,_mm_macc_pd(d,swV5,swV4),swV3))));
 234
 235             dsw              = _mm_mul_pd(d2,_mm_macc_pd(d,_mm_macc_pd(d,swF4,swF3),swF2));
 236
 237             /* Evaluate switch function */
 238             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
 239             felec            = _mm_msub_pd( felec,sw , _mm_mul_pd(rinv00,_mm_mul_pd(velec,dsw)) );
 240             velec            = _mm_mul_pd(velec,sw);
 241             cutoff_mask      = _mm_cmplt_pd(rsq00,rcutoff2);
 242
 243             /* Update potential sum for this i atom from the interaction with this j atom. */
 244             velec            = _mm_and_pd(velec,cutoff_mask);
 245             velecsum         = _mm_add_pd(velecsum,velec);
 246
 247             fscal            = felec;
 248
 249             fscal            = _mm_and_pd(fscal,cutoff_mask);
 250
 251             /* Update vectorial force */
 252             fix0             = _mm_macc_pd(dx00,fscal,fix0);
 253             fiy0             = _mm_macc_pd(dy00,fscal,fiy0);
 254             fiz0             = _mm_macc_pd(dz00,fscal,fiz0);
 255
 256             gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,
 257                                                    _mm_mul_pd(dx00,fscal),
 258                                                    _mm_mul_pd(dy00,fscal),
 259                                                    _mm_mul_pd(dz00,fscal));
 260
 261             }
 262
 263             /* Inner loop uses 68 flops */
 264         }
 265
 266         if(jidx<j_index_end)
 267         {
 268
 269             jnrA             = jjnr[jidx];
 270             j_coord_offsetA  = DIM*jnrA;
 271
 272             /* load j atom coordinates */
 273             gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
 274                                               &jx0,&jy0,&jz0);
 275
 276             /* Calculate displacement vector */
 277             dx00             = _mm_sub_pd(ix0,jx0);
 278             dy00             = _mm_sub_pd(iy0,jy0);
 279             dz00             = _mm_sub_pd(iz0,jz0);
 280
 281             /* Calculate squared distance and things based on it */
 282             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 283
 284             rinv00           = avx128fma_invsqrt_d(rsq00);
 285
 286             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 287
 288             /* Load parameters for j particles */
 289             jq0              = _mm_load_sd(charge+jnrA+0);
 290
 291             /**************************
 292              * CALCULATE INTERACTIONS *
 293              **************************/
 294
 295             if (gmx_mm_any_lt(rsq00,rcutoff2))
 296             {
 297
 298             r00              = _mm_mul_pd(rsq00,rinv00);
 299
 300             /* Compute parameters for interactions between i and j atoms */
 301             qq00             = _mm_mul_pd(iq0,jq0);
 302
 303             /* EWALD ELECTROSTATICS */
 304
 305             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 306             ewrt             = _mm_mul_pd(r00,ewtabscale);
 307             ewitab           = _mm_cvttpd_epi32(ewrt);
 308 #ifdef __XOP__
 309             eweps            = _mm_frcz_pd(ewrt);
 310 #else
 311             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 312 #endif
 313             twoeweps         = _mm_add_pd(eweps,eweps);
 314             ewitab           = _mm_slli_epi32(ewitab,2);
 315             ewtabF           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 316             ewtabD           = _mm_setzero_pd();
 317             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 318             ewtabV           = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
 319             ewtabFn          = _mm_setzero_pd();
 320             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 321             felec            = _mm_macc_pd(eweps,ewtabD,ewtabF);
 322             velec            = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
 323             velec            = _mm_mul_pd(qq00,_mm_sub_pd(rinv00,velec));
 324             felec            = _mm_mul_pd(_mm_mul_pd(qq00,rinv00),_mm_sub_pd(rinvsq00,felec));
 325
 326             d                = _mm_sub_pd(r00,rswitch);
 327             d                = _mm_max_pd(d,_mm_setzero_pd());
 328             d2               = _mm_mul_pd(d,d);
 329             sw               = _mm_add_pd(one,_mm_mul_pd(d2,_mm_mul_pd(d,_mm_macc_pd(d,_mm_macc_pd(d,swV5,swV4),swV3))));
 330
 331             dsw              = _mm_mul_pd(d2,_mm_macc_pd(d,_mm_macc_pd(d,swF4,swF3),swF2));
 332
 333             /* Evaluate switch function */
 334             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
 335             felec            = _mm_msub_pd( felec,sw , _mm_mul_pd(rinv00,_mm_mul_pd(velec,dsw)) );
 336             velec            = _mm_mul_pd(velec,sw);
 337             cutoff_mask      = _mm_cmplt_pd(rsq00,rcutoff2);
 338
 339             /* Update potential sum for this i atom from the interaction with this j atom. */
 340             velec            = _mm_and_pd(velec,cutoff_mask);
 341             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 342             velecsum         = _mm_add_pd(velecsum,velec);
 343
 344             fscal            = felec;
 345
 346             fscal            = _mm_and_pd(fscal,cutoff_mask);
 347
 348             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 349
 350             /* Update vectorial force */
 351             fix0             = _mm_macc_pd(dx00,fscal,fix0);
 352             fiy0             = _mm_macc_pd(dy00,fscal,fiy0);
 353             fiz0             = _mm_macc_pd(dz00,fscal,fiz0);
 354
 355             gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,
 356                                                    _mm_mul_pd(dx00,fscal),
 357                                                    _mm_mul_pd(dy00,fscal),
 358                                                    _mm_mul_pd(dz00,fscal));
 359
 360             }
 361
 362             /* Inner loop uses 68 flops */
 363         }
 364
 365         /* End of innermost loop */
 366
 367         gmx_mm_update_iforce_1atom_swizzle_pd(fix0,fiy0,fiz0,
 368                                               f+i_coord_offset,fshift+i_shift_offset);
 369
 370         ggid                        = gid[iidx];
 371         /* Update potential energies */
 372         gmx_mm_update_1pot_pd(velecsum,kernel_data->energygrp_elec+ggid);
 373
 374         /* Increment number of inner iterations */
 375         inneriter                  += j_index_end - j_index_start;
 376
 377         /* Outer loop uses 8 flops */
 378     }
 379
 380     /* Increment number of outer iterations */
 381     outeriter        += nri;
 382
 383     /* Update outer/inner flops */
 384
 385     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VF,outeriter*8 + inneriter*68);
 386 }
 387 /*
 388  * Gromacs nonbonded kernel:   nb_kernel_ElecEwSw_VdwNone_GeomP1P1_F_avx_128_fma_double
 389  * Electrostatics interaction: Ewald
 390  * VdW interaction:            None
 391  * Geometry:                   Particle-Particle
 392  * Calculate force/pot:        Force
 393  */
 394 void
 395 nb_kernel_ElecEwSw_VdwNone_GeomP1P1_F_avx_128_fma_double
 396                     (t_nblist                    * gmx_restrict       nlist,
 397                      rvec                        * gmx_restrict          xx,
 398                      rvec                        * gmx_restrict          ff,
 399                      struct t_forcerec           * gmx_restrict          fr,
 400                      t_mdatoms                   * gmx_restrict     mdatoms,
 401                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 402                      t_nrnb                      * gmx_restrict        nrnb)
 403 {
 404     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 405      * just 0 for non-waters.
 406      * Suffixes A,B refer to j loop unrolling done with SSE double precision, e.g. for the two different
 407      * jnr indices corresponding to data put in the four positions in the SIMD register.
 408      */
 409     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 410     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 411     int              jnrA,jnrB;
 412     int              j_coord_offsetA,j_coord_offsetB;
 413     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 414     real             rcutoff_scalar;
 415     real             *shiftvec,*fshift,*x,*f;
 416     __m128d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 417     int              vdwioffset0;
 418     __m128d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 419     int              vdwjidx0A,vdwjidx0B;
 420     __m128d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 421     __m128d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 422     __m128d          velec,felec,velecsum,facel,crf,krf,krf2;
 423     real             *charge;
 424     __m128i          ewitab;
 425     __m128d          ewtabscale,eweps,twoeweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
 426     real             *ewtab;
 427     __m128d          rswitch,swV3,swV4,swV5,swF2,swF3,swF4,d,d2,sw,dsw;
 428     real             rswitch_scalar,d_scalar;
 429     __m128d          dummy_mask,cutoff_mask;
 430     __m128d          signbit   = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
 431     __m128d          one     = _mm_set1_pd(1.0);
 432     __m128d          two     = _mm_set1_pd(2.0);
 433     x                = xx[0];
 434     f                = ff[0];
 435
 436     nri              = nlist->nri;
 437     iinr             = nlist->iinr;
 438     jindex           = nlist->jindex;
 439     jjnr             = nlist->jjnr;
 440     shiftidx         = nlist->shift;
 441     gid              = nlist->gid;
 442     shiftvec         = fr->shift_vec[0];
 443     fshift           = fr->fshift[0];
 444     facel            = _mm_set1_pd(fr->ic->epsfac);
 445     charge           = mdatoms->chargeA;
 446
 447     sh_ewald         = _mm_set1_pd(fr->ic->sh_ewald);
 448     ewtab            = fr->ic->tabq_coul_FDV0;
 449     ewtabscale       = _mm_set1_pd(fr->ic->tabq_scale);
 450     ewtabhalfspace   = _mm_set1_pd(0.5/fr->ic->tabq_scale);
 451
 452     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 453     rcutoff_scalar   = fr->ic->rcoulomb;
 454     rcutoff          = _mm_set1_pd(rcutoff_scalar);
 455     rcutoff2         = _mm_mul_pd(rcutoff,rcutoff);
 456
 457     rswitch_scalar   = fr->ic->rcoulomb_switch;
 458     rswitch          = _mm_set1_pd(rswitch_scalar);
 459     /* Setup switch parameters */
 460     d_scalar         = rcutoff_scalar-rswitch_scalar;
 461     d                = _mm_set1_pd(d_scalar);
 462     swV3             = _mm_set1_pd(-10.0/(d_scalar*d_scalar*d_scalar));
 463     swV4             = _mm_set1_pd( 15.0/(d_scalar*d_scalar*d_scalar*d_scalar));
 464     swV5             = _mm_set1_pd( -6.0/(d_scalar*d_scalar*d_scalar*d_scalar*d_scalar));
 465     swF2             = _mm_set1_pd(-30.0/(d_scalar*d_scalar*d_scalar));
 466     swF3             = _mm_set1_pd( 60.0/(d_scalar*d_scalar*d_scalar*d_scalar));
 467     swF4             = _mm_set1_pd(-30.0/(d_scalar*d_scalar*d_scalar*d_scalar*d_scalar));
 468
 469     /* Avoid stupid compiler warnings */
 470     jnrA = jnrB = 0;
 471     j_coord_offsetA = 0;
 472     j_coord_offsetB = 0;
 473
 474     outeriter        = 0;
 475     inneriter        = 0;
 476
 477     /* Start outer loop over neighborlists */
 478     for(iidx=0; iidx<nri; iidx++)
 479     {
 480         /* Load shift vector for this list */
 481         i_shift_offset   = DIM*shiftidx[iidx];
 482
 483         /* Load limits for loop over neighbors */
 484         j_index_start    = jindex[iidx];
 485         j_index_end      = jindex[iidx+1];
 486
 487         /* Get outer coordinate index */
 488         inr              = iinr[iidx];
 489         i_coord_offset   = DIM*inr;
 490
 491         /* Load i particle coords and add shift vector */
 492         gmx_mm_load_shift_and_1rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 493
 494         fix0             = _mm_setzero_pd();
 495         fiy0             = _mm_setzero_pd();
 496         fiz0             = _mm_setzero_pd();
 497
 498         /* Load parameters for i particles */
 499         iq0              = _mm_mul_pd(facel,_mm_load1_pd(charge+inr+0));
 500
 501         /* Start inner kernel loop */
 502         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 503         {
 504
 505             /* Get j neighbor index, and coordinate index */
 506             jnrA             = jjnr[jidx];
 507             jnrB             = jjnr[jidx+1];
 508             j_coord_offsetA  = DIM*jnrA;
 509             j_coord_offsetB  = DIM*jnrB;
 510
 511             /* load j atom coordinates */
 512             gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 513                                               &jx0,&jy0,&jz0);
 514
 515             /* Calculate displacement vector */
 516             dx00             = _mm_sub_pd(ix0,jx0);
 517             dy00             = _mm_sub_pd(iy0,jy0);
 518             dz00             = _mm_sub_pd(iz0,jz0);
 519
 520             /* Calculate squared distance and things based on it */
 521             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 522
 523             rinv00           = avx128fma_invsqrt_d(rsq00);
 524
 525             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 526
 527             /* Load parameters for j particles */
 528             jq0              = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
 529
 530             /**************************
 531              * CALCULATE INTERACTIONS *
 532              **************************/
 533
 534             if (gmx_mm_any_lt(rsq00,rcutoff2))
 535             {
 536
 537             r00              = _mm_mul_pd(rsq00,rinv00);
 538
 539             /* Compute parameters for interactions between i and j atoms */
 540             qq00             = _mm_mul_pd(iq0,jq0);
 541
 542             /* EWALD ELECTROSTATICS */
 543
 544             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 545             ewrt             = _mm_mul_pd(r00,ewtabscale);
 546             ewitab           = _mm_cvttpd_epi32(ewrt);
 547 #ifdef __XOP__
 548             eweps            = _mm_frcz_pd(ewrt);
 549 #else
 550             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 551 #endif
 552             twoeweps         = _mm_add_pd(eweps,eweps);
 553             ewitab           = _mm_slli_epi32(ewitab,2);
 554             ewtabF           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 555             ewtabD           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
 556             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 557             ewtabV           = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
 558             ewtabFn          = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,1) +2);
 559             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 560             felec            = _mm_macc_pd(eweps,ewtabD,ewtabF);
 561             velec            = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
 562             velec            = _mm_mul_pd(qq00,_mm_sub_pd(rinv00,velec));
 563             felec            = _mm_mul_pd(_mm_mul_pd(qq00,rinv00),_mm_sub_pd(rinvsq00,felec));
 564
 565             d                = _mm_sub_pd(r00,rswitch);
 566             d                = _mm_max_pd(d,_mm_setzero_pd());
 567             d2               = _mm_mul_pd(d,d);
 568             sw               = _mm_add_pd(one,_mm_mul_pd(d2,_mm_mul_pd(d,_mm_macc_pd(d,_mm_macc_pd(d,swV5,swV4),swV3))));
 569
 570             dsw              = _mm_mul_pd(d2,_mm_macc_pd(d,_mm_macc_pd(d,swF4,swF3),swF2));
 571
 572             /* Evaluate switch function */
 573             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
 574             felec            = _mm_msub_pd( felec,sw , _mm_mul_pd(rinv00,_mm_mul_pd(velec,dsw)) );
 575             cutoff_mask      = _mm_cmplt_pd(rsq00,rcutoff2);
 576
 577             fscal            = felec;
 578
 579             fscal            = _mm_and_pd(fscal,cutoff_mask);
 580
 581             /* Update vectorial force */
 582             fix0             = _mm_macc_pd(dx00,fscal,fix0);
 583             fiy0             = _mm_macc_pd(dy00,fscal,fiy0);
 584             fiz0             = _mm_macc_pd(dz00,fscal,fiz0);
 585
 586             gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,
 587                                                    _mm_mul_pd(dx00,fscal),
 588                                                    _mm_mul_pd(dy00,fscal),
 589                                                    _mm_mul_pd(dz00,fscal));
 590
 591             }
 592
 593             /* Inner loop uses 65 flops */
 594         }
 595
 596         if(jidx<j_index_end)
 597         {
 598
 599             jnrA             = jjnr[jidx];
 600             j_coord_offsetA  = DIM*jnrA;
 601
 602             /* load j atom coordinates */
 603             gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
 604                                               &jx0,&jy0,&jz0);
 605
 606             /* Calculate displacement vector */
 607             dx00             = _mm_sub_pd(ix0,jx0);
 608             dy00             = _mm_sub_pd(iy0,jy0);
 609             dz00             = _mm_sub_pd(iz0,jz0);
 610
 611             /* Calculate squared distance and things based on it */
 612             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 613
 614             rinv00           = avx128fma_invsqrt_d(rsq00);
 615
 616             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 617
 618             /* Load parameters for j particles */
 619             jq0              = _mm_load_sd(charge+jnrA+0);
 620
 621             /**************************
 622              * CALCULATE INTERACTIONS *
 623              **************************/
 624
 625             if (gmx_mm_any_lt(rsq00,rcutoff2))
 626             {
 627
 628             r00              = _mm_mul_pd(rsq00,rinv00);
 629
 630             /* Compute parameters for interactions between i and j atoms */
 631             qq00             = _mm_mul_pd(iq0,jq0);
 632
 633             /* EWALD ELECTROSTATICS */
 634
 635             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 636             ewrt             = _mm_mul_pd(r00,ewtabscale);
 637             ewitab           = _mm_cvttpd_epi32(ewrt);
 638 #ifdef __XOP__
 639             eweps            = _mm_frcz_pd(ewrt);
 640 #else
 641             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 642 #endif
 643             twoeweps         = _mm_add_pd(eweps,eweps);
 644             ewitab           = _mm_slli_epi32(ewitab,2);
 645             ewtabF           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 646             ewtabD           = _mm_setzero_pd();
 647             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 648             ewtabV           = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
 649             ewtabFn          = _mm_setzero_pd();
 650             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 651             felec            = _mm_macc_pd(eweps,ewtabD,ewtabF);
 652             velec            = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
 653             velec            = _mm_mul_pd(qq00,_mm_sub_pd(rinv00,velec));
 654             felec            = _mm_mul_pd(_mm_mul_pd(qq00,rinv00),_mm_sub_pd(rinvsq00,felec));
 655
 656             d                = _mm_sub_pd(r00,rswitch);
 657             d                = _mm_max_pd(d,_mm_setzero_pd());
 658             d2               = _mm_mul_pd(d,d);
 659             sw               = _mm_add_pd(one,_mm_mul_pd(d2,_mm_mul_pd(d,_mm_macc_pd(d,_mm_macc_pd(d,swV5,swV4),swV3))));
 660
 661             dsw              = _mm_mul_pd(d2,_mm_macc_pd(d,_mm_macc_pd(d,swF4,swF3),swF2));
 662
 663             /* Evaluate switch function */
 664             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
 665             felec            = _mm_msub_pd( felec,sw , _mm_mul_pd(rinv00,_mm_mul_pd(velec,dsw)) );
 666             cutoff_mask      = _mm_cmplt_pd(rsq00,rcutoff2);
 667
 668             fscal            = felec;
 669
 670             fscal            = _mm_and_pd(fscal,cutoff_mask);
 671
 672             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 673
 674             /* Update vectorial force */
 675             fix0             = _mm_macc_pd(dx00,fscal,fix0);
 676             fiy0             = _mm_macc_pd(dy00,fscal,fiy0);
 677             fiz0             = _mm_macc_pd(dz00,fscal,fiz0);
 678
 679             gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,
 680                                                    _mm_mul_pd(dx00,fscal),
 681                                                    _mm_mul_pd(dy00,fscal),
 682                                                    _mm_mul_pd(dz00,fscal));
 683
 684             }
 685
 686             /* Inner loop uses 65 flops */
 687         }
 688
 689         /* End of innermost loop */
 690
 691         gmx_mm_update_iforce_1atom_swizzle_pd(fix0,fiy0,fiz0,
 692                                               f+i_coord_offset,fshift+i_shift_offset);
 693
 694         /* Increment number of inner iterations */
 695         inneriter                  += j_index_end - j_index_start;
 696
 697         /* Outer loop uses 7 flops */
 698     }
 699
 700     /* Increment number of outer iterations */
 701     outeriter        += nri;
 702
 703     /* Update outer/inner flops */
 704
 705     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_F,outeriter*7 + inneriter*65);
 706 }