src/gromacs/gmxlib/nonbonded/nb_kernel_sse4_1_double/nb_kernel_ElecEwSw_VdwNone_GeomP1P1_sse4_1_double.c

   1 /*
   2  * This file is part of the GROMACS molecular simulation package.
   3  *
   4  * Copyright (c) 2012,2013, 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 sse4_1_double kernel generator.
  37  */
  38 #ifdef HAVE_CONFIG_H
  39 #include <config.h>
  40 #endif
  41
  42 #include <math.h>
  43
  44 #include "../nb_kernel.h"
  45 #include "types/simple.h"
  46 #include "vec.h"
  47 #include "nrnb.h"
  48
  49 #include "gromacs/simd/math_x86_sse4_1_double.h"
  50 #include "kernelutil_x86_sse4_1_double.h"
  51
  52 /*
  53  * Gromacs nonbonded kernel:   nb_kernel_ElecEwSw_VdwNone_GeomP1P1_VF_sse4_1_double
  54  * Electrostatics interaction: Ewald
  55  * VdW interaction:            None
  56  * Geometry:                   Particle-Particle
  57  * Calculate force/pot:        PotentialAndForce
  58  */
  59 void
  60 nb_kernel_ElecEwSw_VdwNone_GeomP1P1_VF_sse4_1_double
  61                     (t_nblist                    * gmx_restrict       nlist,
  62                      rvec                        * gmx_restrict          xx,
  63                      rvec                        * gmx_restrict          ff,
  64                      t_forcerec                  * gmx_restrict          fr,
  65                      t_mdatoms                   * gmx_restrict     mdatoms,
  66                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
  67                      t_nrnb                      * gmx_restrict        nrnb)
  68 {
  69     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
  70      * just 0 for non-waters.
  71      * Suffixes A,B refer to j loop unrolling done with SSE double precision, e.g. for the two different
  72      * jnr indices corresponding to data put in the four positions in the SIMD register.
  73      */
  74     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
  75     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
  76     int              jnrA,jnrB;
  77     int              j_coord_offsetA,j_coord_offsetB;
  78     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
  79     real             rcutoff_scalar;
  80     real             *shiftvec,*fshift,*x,*f;
  81     __m128d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
  82     int              vdwioffset0;
  83     __m128d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
  84     int              vdwjidx0A,vdwjidx0B;
  85     __m128d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  86     __m128d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  87     __m128d          velec,felec,velecsum,facel,crf,krf,krf2;
  88     real             *charge;
  89     __m128i          ewitab;
  90     __m128d          ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
  91     real             *ewtab;
  92     __m128d          rswitch,swV3,swV4,swV5,swF2,swF3,swF4,d,d2,sw,dsw;
  93     real             rswitch_scalar,d_scalar;
  94     __m128d          dummy_mask,cutoff_mask;
  95     __m128d          signbit   = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
  96     __m128d          one     = _mm_set1_pd(1.0);
  97     __m128d          two     = _mm_set1_pd(2.0);
  98     x                = xx[0];
  99     f                = ff[0];
 100
 101     nri              = nlist->nri;
 102     iinr             = nlist->iinr;
 103     jindex           = nlist->jindex;
 104     jjnr             = nlist->jjnr;
 105     shiftidx         = nlist->shift;
 106     gid              = nlist->gid;
 107     shiftvec         = fr->shift_vec[0];
 108     fshift           = fr->fshift[0];
 109     facel            = _mm_set1_pd(fr->epsfac);
 110     charge           = mdatoms->chargeA;
 111
 112     sh_ewald         = _mm_set1_pd(fr->ic->sh_ewald);
 113     ewtab            = fr->ic->tabq_coul_FDV0;
 114     ewtabscale       = _mm_set1_pd(fr->ic->tabq_scale);
 115     ewtabhalfspace   = _mm_set1_pd(0.5/fr->ic->tabq_scale);
 116
 117     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 118     rcutoff_scalar   = fr->rcoulomb;
 119     rcutoff          = _mm_set1_pd(rcutoff_scalar);
 120     rcutoff2         = _mm_mul_pd(rcutoff,rcutoff);
 121
 122     rswitch_scalar   = fr->rcoulomb_switch;
 123     rswitch          = _mm_set1_pd(rswitch_scalar);
 124     /* Setup switch parameters */
 125     d_scalar         = rcutoff_scalar-rswitch_scalar;
 126     d                = _mm_set1_pd(d_scalar);
 127     swV3             = _mm_set1_pd(-10.0/(d_scalar*d_scalar*d_scalar));
 128     swV4             = _mm_set1_pd( 15.0/(d_scalar*d_scalar*d_scalar*d_scalar));
 129     swV5             = _mm_set1_pd( -6.0/(d_scalar*d_scalar*d_scalar*d_scalar*d_scalar));
 130     swF2             = _mm_set1_pd(-30.0/(d_scalar*d_scalar*d_scalar));
 131     swF3             = _mm_set1_pd( 60.0/(d_scalar*d_scalar*d_scalar*d_scalar));
 132     swF4             = _mm_set1_pd(-30.0/(d_scalar*d_scalar*d_scalar*d_scalar*d_scalar));
 133
 134     /* Avoid stupid compiler warnings */
 135     jnrA = jnrB = 0;
 136     j_coord_offsetA = 0;
 137     j_coord_offsetB = 0;
 138
 139     outeriter        = 0;
 140     inneriter        = 0;
 141
 142     /* Start outer loop over neighborlists */
 143     for(iidx=0; iidx<nri; iidx++)
 144     {
 145         /* Load shift vector for this list */
 146         i_shift_offset   = DIM*shiftidx[iidx];
 147
 148         /* Load limits for loop over neighbors */
 149         j_index_start    = jindex[iidx];
 150         j_index_end      = jindex[iidx+1];
 151
 152         /* Get outer coordinate index */
 153         inr              = iinr[iidx];
 154         i_coord_offset   = DIM*inr;
 155
 156         /* Load i particle coords and add shift vector */
 157         gmx_mm_load_shift_and_1rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 158
 159         fix0             = _mm_setzero_pd();
 160         fiy0             = _mm_setzero_pd();
 161         fiz0             = _mm_setzero_pd();
 162
 163         /* Load parameters for i particles */
 164         iq0              = _mm_mul_pd(facel,_mm_load1_pd(charge+inr+0));
 165
 166         /* Reset potential sums */
 167         velecsum         = _mm_setzero_pd();
 168
 169         /* Start inner kernel loop */
 170         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 171         {
 172
 173             /* Get j neighbor index, and coordinate index */
 174             jnrA             = jjnr[jidx];
 175             jnrB             = jjnr[jidx+1];
 176             j_coord_offsetA  = DIM*jnrA;
 177             j_coord_offsetB  = DIM*jnrB;
 178
 179             /* load j atom coordinates */
 180             gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 181                                               &jx0,&jy0,&jz0);
 182
 183             /* Calculate displacement vector */
 184             dx00             = _mm_sub_pd(ix0,jx0);
 185             dy00             = _mm_sub_pd(iy0,jy0);
 186             dz00             = _mm_sub_pd(iz0,jz0);
 187
 188             /* Calculate squared distance and things based on it */
 189             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 190
 191             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 192
 193             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 194
 195             /* Load parameters for j particles */
 196             jq0              = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
 197
 198             /**************************
 199              * CALCULATE INTERACTIONS *
 200              **************************/
 201
 202             if (gmx_mm_any_lt(rsq00,rcutoff2))
 203             {
 204
 205             r00              = _mm_mul_pd(rsq00,rinv00);
 206
 207             /* Compute parameters for interactions between i and j atoms */
 208             qq00             = _mm_mul_pd(iq0,jq0);
 209
 210             /* EWALD ELECTROSTATICS */
 211
 212             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 213             ewrt             = _mm_mul_pd(r00,ewtabscale);
 214             ewitab           = _mm_cvttpd_epi32(ewrt);
 215             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 216             ewitab           = _mm_slli_epi32(ewitab,2);
 217             ewtabF           = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 218             ewtabD           = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,1) );
 219             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 220             ewtabV           = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,0) +2);
 221             ewtabFn          = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,1) +2);
 222             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 223             felec            = _mm_add_pd(ewtabF,_mm_mul_pd(eweps,ewtabD));
 224             velec            = _mm_sub_pd(ewtabV,_mm_mul_pd(_mm_mul_pd(ewtabhalfspace,eweps),_mm_add_pd(ewtabF,felec)));
 225             velec            = _mm_mul_pd(qq00,_mm_sub_pd(rinv00,velec));
 226             felec            = _mm_mul_pd(_mm_mul_pd(qq00,rinv00),_mm_sub_pd(rinvsq00,felec));
 227
 228             d                = _mm_sub_pd(r00,rswitch);
 229             d                = _mm_max_pd(d,_mm_setzero_pd());
 230             d2               = _mm_mul_pd(d,d);
 231             sw               = _mm_add_pd(one,_mm_mul_pd(d2,_mm_mul_pd(d,_mm_add_pd(swV3,_mm_mul_pd(d,_mm_add_pd(swV4,_mm_mul_pd(d,swV5)))))));
 232
 233             dsw              = _mm_mul_pd(d2,_mm_add_pd(swF2,_mm_mul_pd(d,_mm_add_pd(swF3,_mm_mul_pd(d,swF4)))));
 234
 235             /* Evaluate switch function */
 236             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
 237             felec            = _mm_sub_pd( _mm_mul_pd(felec,sw) , _mm_mul_pd(rinv00,_mm_mul_pd(velec,dsw)) );
 238             velec            = _mm_mul_pd(velec,sw);
 239             cutoff_mask      = _mm_cmplt_pd(rsq00,rcutoff2);
 240
 241             /* Update potential sum for this i atom from the interaction with this j atom. */
 242             velec            = _mm_and_pd(velec,cutoff_mask);
 243             velecsum         = _mm_add_pd(velecsum,velec);
 244
 245             fscal            = felec;
 246
 247             fscal            = _mm_and_pd(fscal,cutoff_mask);
 248
 249             /* Calculate temporary vectorial force */
 250             tx               = _mm_mul_pd(fscal,dx00);
 251             ty               = _mm_mul_pd(fscal,dy00);
 252             tz               = _mm_mul_pd(fscal,dz00);
 253
 254             /* Update vectorial force */
 255             fix0             = _mm_add_pd(fix0,tx);
 256             fiy0             = _mm_add_pd(fiy0,ty);
 257             fiz0             = _mm_add_pd(fiz0,tz);
 258
 259             gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,tx,ty,tz);
 260
 261             }
 262
 263             /* Inner loop uses 65 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           = gmx_mm_invsqrt_pd(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             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 309             ewitab           = _mm_slli_epi32(ewitab,2);
 310             ewtabF           = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 311             ewtabD           = _mm_setzero_pd();
 312             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 313             ewtabV           = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,0) +2);
 314             ewtabFn          = _mm_setzero_pd();
 315             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 316             felec            = _mm_add_pd(ewtabF,_mm_mul_pd(eweps,ewtabD));
 317             velec            = _mm_sub_pd(ewtabV,_mm_mul_pd(_mm_mul_pd(ewtabhalfspace,eweps),_mm_add_pd(ewtabF,felec)));
 318             velec            = _mm_mul_pd(qq00,_mm_sub_pd(rinv00,velec));
 319             felec            = _mm_mul_pd(_mm_mul_pd(qq00,rinv00),_mm_sub_pd(rinvsq00,felec));
 320
 321             d                = _mm_sub_pd(r00,rswitch);
 322             d                = _mm_max_pd(d,_mm_setzero_pd());
 323             d2               = _mm_mul_pd(d,d);
 324             sw               = _mm_add_pd(one,_mm_mul_pd(d2,_mm_mul_pd(d,_mm_add_pd(swV3,_mm_mul_pd(d,_mm_add_pd(swV4,_mm_mul_pd(d,swV5)))))));
 325
 326             dsw              = _mm_mul_pd(d2,_mm_add_pd(swF2,_mm_mul_pd(d,_mm_add_pd(swF3,_mm_mul_pd(d,swF4)))));
 327
 328             /* Evaluate switch function */
 329             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
 330             felec            = _mm_sub_pd( _mm_mul_pd(felec,sw) , _mm_mul_pd(rinv00,_mm_mul_pd(velec,dsw)) );
 331             velec            = _mm_mul_pd(velec,sw);
 332             cutoff_mask      = _mm_cmplt_pd(rsq00,rcutoff2);
 333
 334             /* Update potential sum for this i atom from the interaction with this j atom. */
 335             velec            = _mm_and_pd(velec,cutoff_mask);
 336             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 337             velecsum         = _mm_add_pd(velecsum,velec);
 338
 339             fscal            = felec;
 340
 341             fscal            = _mm_and_pd(fscal,cutoff_mask);
 342
 343             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 344
 345             /* Calculate temporary vectorial force */
 346             tx               = _mm_mul_pd(fscal,dx00);
 347             ty               = _mm_mul_pd(fscal,dy00);
 348             tz               = _mm_mul_pd(fscal,dz00);
 349
 350             /* Update vectorial force */
 351             fix0             = _mm_add_pd(fix0,tx);
 352             fiy0             = _mm_add_pd(fiy0,ty);
 353             fiz0             = _mm_add_pd(fiz0,tz);
 354
 355             gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,tx,ty,tz);
 356
 357             }
 358
 359             /* Inner loop uses 65 flops */
 360         }
 361
 362         /* End of innermost loop */
 363
 364         gmx_mm_update_iforce_1atom_swizzle_pd(fix0,fiy0,fiz0,
 365                                               f+i_coord_offset,fshift+i_shift_offset);
 366
 367         ggid                        = gid[iidx];
 368         /* Update potential energies */
 369         gmx_mm_update_1pot_pd(velecsum,kernel_data->energygrp_elec+ggid);
 370
 371         /* Increment number of inner iterations */
 372         inneriter                  += j_index_end - j_index_start;
 373
 374         /* Outer loop uses 8 flops */
 375     }
 376
 377     /* Increment number of outer iterations */
 378     outeriter        += nri;
 379
 380     /* Update outer/inner flops */
 381
 382     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VF,outeriter*8 + inneriter*65);
 383 }
 384 /*
 385  * Gromacs nonbonded kernel:   nb_kernel_ElecEwSw_VdwNone_GeomP1P1_F_sse4_1_double
 386  * Electrostatics interaction: Ewald
 387  * VdW interaction:            None
 388  * Geometry:                   Particle-Particle
 389  * Calculate force/pot:        Force
 390  */
 391 void
 392 nb_kernel_ElecEwSw_VdwNone_GeomP1P1_F_sse4_1_double
 393                     (t_nblist                    * gmx_restrict       nlist,
 394                      rvec                        * gmx_restrict          xx,
 395                      rvec                        * gmx_restrict          ff,
 396                      t_forcerec                  * gmx_restrict          fr,
 397                      t_mdatoms                   * gmx_restrict     mdatoms,
 398                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 399                      t_nrnb                      * gmx_restrict        nrnb)
 400 {
 401     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 402      * just 0 for non-waters.
 403      * Suffixes A,B refer to j loop unrolling done with SSE double precision, e.g. for the two different
 404      * jnr indices corresponding to data put in the four positions in the SIMD register.
 405      */
 406     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 407     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 408     int              jnrA,jnrB;
 409     int              j_coord_offsetA,j_coord_offsetB;
 410     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 411     real             rcutoff_scalar;
 412     real             *shiftvec,*fshift,*x,*f;
 413     __m128d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 414     int              vdwioffset0;
 415     __m128d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 416     int              vdwjidx0A,vdwjidx0B;
 417     __m128d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 418     __m128d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 419     __m128d          velec,felec,velecsum,facel,crf,krf,krf2;
 420     real             *charge;
 421     __m128i          ewitab;
 422     __m128d          ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
 423     real             *ewtab;
 424     __m128d          rswitch,swV3,swV4,swV5,swF2,swF3,swF4,d,d2,sw,dsw;
 425     real             rswitch_scalar,d_scalar;
 426     __m128d          dummy_mask,cutoff_mask;
 427     __m128d          signbit   = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
 428     __m128d          one     = _mm_set1_pd(1.0);
 429     __m128d          two     = _mm_set1_pd(2.0);
 430     x                = xx[0];
 431     f                = ff[0];
 432
 433     nri              = nlist->nri;
 434     iinr             = nlist->iinr;
 435     jindex           = nlist->jindex;
 436     jjnr             = nlist->jjnr;
 437     shiftidx         = nlist->shift;
 438     gid              = nlist->gid;
 439     shiftvec         = fr->shift_vec[0];
 440     fshift           = fr->fshift[0];
 441     facel            = _mm_set1_pd(fr->epsfac);
 442     charge           = mdatoms->chargeA;
 443
 444     sh_ewald         = _mm_set1_pd(fr->ic->sh_ewald);
 445     ewtab            = fr->ic->tabq_coul_FDV0;
 446     ewtabscale       = _mm_set1_pd(fr->ic->tabq_scale);
 447     ewtabhalfspace   = _mm_set1_pd(0.5/fr->ic->tabq_scale);
 448
 449     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 450     rcutoff_scalar   = fr->rcoulomb;
 451     rcutoff          = _mm_set1_pd(rcutoff_scalar);
 452     rcutoff2         = _mm_mul_pd(rcutoff,rcutoff);
 453
 454     rswitch_scalar   = fr->rcoulomb_switch;
 455     rswitch          = _mm_set1_pd(rswitch_scalar);
 456     /* Setup switch parameters */
 457     d_scalar         = rcutoff_scalar-rswitch_scalar;
 458     d                = _mm_set1_pd(d_scalar);
 459     swV3             = _mm_set1_pd(-10.0/(d_scalar*d_scalar*d_scalar));
 460     swV4             = _mm_set1_pd( 15.0/(d_scalar*d_scalar*d_scalar*d_scalar));
 461     swV5             = _mm_set1_pd( -6.0/(d_scalar*d_scalar*d_scalar*d_scalar*d_scalar));
 462     swF2             = _mm_set1_pd(-30.0/(d_scalar*d_scalar*d_scalar));
 463     swF3             = _mm_set1_pd( 60.0/(d_scalar*d_scalar*d_scalar*d_scalar));
 464     swF4             = _mm_set1_pd(-30.0/(d_scalar*d_scalar*d_scalar*d_scalar*d_scalar));
 465
 466     /* Avoid stupid compiler warnings */
 467     jnrA = jnrB = 0;
 468     j_coord_offsetA = 0;
 469     j_coord_offsetB = 0;
 470
 471     outeriter        = 0;
 472     inneriter        = 0;
 473
 474     /* Start outer loop over neighborlists */
 475     for(iidx=0; iidx<nri; iidx++)
 476     {
 477         /* Load shift vector for this list */
 478         i_shift_offset   = DIM*shiftidx[iidx];
 479
 480         /* Load limits for loop over neighbors */
 481         j_index_start    = jindex[iidx];
 482         j_index_end      = jindex[iidx+1];
 483
 484         /* Get outer coordinate index */
 485         inr              = iinr[iidx];
 486         i_coord_offset   = DIM*inr;
 487
 488         /* Load i particle coords and add shift vector */
 489         gmx_mm_load_shift_and_1rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 490
 491         fix0             = _mm_setzero_pd();
 492         fiy0             = _mm_setzero_pd();
 493         fiz0             = _mm_setzero_pd();
 494
 495         /* Load parameters for i particles */
 496         iq0              = _mm_mul_pd(facel,_mm_load1_pd(charge+inr+0));
 497
 498         /* Start inner kernel loop */
 499         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 500         {
 501
 502             /* Get j neighbor index, and coordinate index */
 503             jnrA             = jjnr[jidx];
 504             jnrB             = jjnr[jidx+1];
 505             j_coord_offsetA  = DIM*jnrA;
 506             j_coord_offsetB  = DIM*jnrB;
 507
 508             /* load j atom coordinates */
 509             gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 510                                               &jx0,&jy0,&jz0);
 511
 512             /* Calculate displacement vector */
 513             dx00             = _mm_sub_pd(ix0,jx0);
 514             dy00             = _mm_sub_pd(iy0,jy0);
 515             dz00             = _mm_sub_pd(iz0,jz0);
 516
 517             /* Calculate squared distance and things based on it */
 518             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 519
 520             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 521
 522             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 523
 524             /* Load parameters for j particles */
 525             jq0              = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
 526
 527             /**************************
 528              * CALCULATE INTERACTIONS *
 529              **************************/
 530
 531             if (gmx_mm_any_lt(rsq00,rcutoff2))
 532             {
 533
 534             r00              = _mm_mul_pd(rsq00,rinv00);
 535
 536             /* Compute parameters for interactions between i and j atoms */
 537             qq00             = _mm_mul_pd(iq0,jq0);
 538
 539             /* EWALD ELECTROSTATICS */
 540
 541             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 542             ewrt             = _mm_mul_pd(r00,ewtabscale);
 543             ewitab           = _mm_cvttpd_epi32(ewrt);
 544             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 545             ewitab           = _mm_slli_epi32(ewitab,2);
 546             ewtabF           = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 547             ewtabD           = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,1) );
 548             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 549             ewtabV           = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,0) +2);
 550             ewtabFn          = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,1) +2);
 551             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 552             felec            = _mm_add_pd(ewtabF,_mm_mul_pd(eweps,ewtabD));
 553             velec            = _mm_sub_pd(ewtabV,_mm_mul_pd(_mm_mul_pd(ewtabhalfspace,eweps),_mm_add_pd(ewtabF,felec)));
 554             velec            = _mm_mul_pd(qq00,_mm_sub_pd(rinv00,velec));
 555             felec            = _mm_mul_pd(_mm_mul_pd(qq00,rinv00),_mm_sub_pd(rinvsq00,felec));
 556
 557             d                = _mm_sub_pd(r00,rswitch);
 558             d                = _mm_max_pd(d,_mm_setzero_pd());
 559             d2               = _mm_mul_pd(d,d);
 560             sw               = _mm_add_pd(one,_mm_mul_pd(d2,_mm_mul_pd(d,_mm_add_pd(swV3,_mm_mul_pd(d,_mm_add_pd(swV4,_mm_mul_pd(d,swV5)))))));
 561
 562             dsw              = _mm_mul_pd(d2,_mm_add_pd(swF2,_mm_mul_pd(d,_mm_add_pd(swF3,_mm_mul_pd(d,swF4)))));
 563
 564             /* Evaluate switch function */
 565             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
 566             felec            = _mm_sub_pd( _mm_mul_pd(felec,sw) , _mm_mul_pd(rinv00,_mm_mul_pd(velec,dsw)) );
 567             cutoff_mask      = _mm_cmplt_pd(rsq00,rcutoff2);
 568
 569             fscal            = felec;
 570
 571             fscal            = _mm_and_pd(fscal,cutoff_mask);
 572
 573             /* Calculate temporary vectorial force */
 574             tx               = _mm_mul_pd(fscal,dx00);
 575             ty               = _mm_mul_pd(fscal,dy00);
 576             tz               = _mm_mul_pd(fscal,dz00);
 577
 578             /* Update vectorial force */
 579             fix0             = _mm_add_pd(fix0,tx);
 580             fiy0             = _mm_add_pd(fiy0,ty);
 581             fiz0             = _mm_add_pd(fiz0,tz);
 582
 583             gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,tx,ty,tz);
 584
 585             }
 586
 587             /* Inner loop uses 62 flops */
 588         }
 589
 590         if(jidx<j_index_end)
 591         {
 592
 593             jnrA             = jjnr[jidx];
 594             j_coord_offsetA  = DIM*jnrA;
 595
 596             /* load j atom coordinates */
 597             gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
 598                                               &jx0,&jy0,&jz0);
 599
 600             /* Calculate displacement vector */
 601             dx00             = _mm_sub_pd(ix0,jx0);
 602             dy00             = _mm_sub_pd(iy0,jy0);
 603             dz00             = _mm_sub_pd(iz0,jz0);
 604
 605             /* Calculate squared distance and things based on it */
 606             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 607
 608             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 609
 610             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 611
 612             /* Load parameters for j particles */
 613             jq0              = _mm_load_sd(charge+jnrA+0);
 614
 615             /**************************
 616              * CALCULATE INTERACTIONS *
 617              **************************/
 618
 619             if (gmx_mm_any_lt(rsq00,rcutoff2))
 620             {
 621
 622             r00              = _mm_mul_pd(rsq00,rinv00);
 623
 624             /* Compute parameters for interactions between i and j atoms */
 625             qq00             = _mm_mul_pd(iq0,jq0);
 626
 627             /* EWALD ELECTROSTATICS */
 628
 629             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 630             ewrt             = _mm_mul_pd(r00,ewtabscale);
 631             ewitab           = _mm_cvttpd_epi32(ewrt);
 632             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 633             ewitab           = _mm_slli_epi32(ewitab,2);
 634             ewtabF           = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 635             ewtabD           = _mm_setzero_pd();
 636             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 637             ewtabV           = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,0) +2);
 638             ewtabFn          = _mm_setzero_pd();
 639             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 640             felec            = _mm_add_pd(ewtabF,_mm_mul_pd(eweps,ewtabD));
 641             velec            = _mm_sub_pd(ewtabV,_mm_mul_pd(_mm_mul_pd(ewtabhalfspace,eweps),_mm_add_pd(ewtabF,felec)));
 642             velec            = _mm_mul_pd(qq00,_mm_sub_pd(rinv00,velec));
 643             felec            = _mm_mul_pd(_mm_mul_pd(qq00,rinv00),_mm_sub_pd(rinvsq00,felec));
 644
 645             d                = _mm_sub_pd(r00,rswitch);
 646             d                = _mm_max_pd(d,_mm_setzero_pd());
 647             d2               = _mm_mul_pd(d,d);
 648             sw               = _mm_add_pd(one,_mm_mul_pd(d2,_mm_mul_pd(d,_mm_add_pd(swV3,_mm_mul_pd(d,_mm_add_pd(swV4,_mm_mul_pd(d,swV5)))))));
 649
 650             dsw              = _mm_mul_pd(d2,_mm_add_pd(swF2,_mm_mul_pd(d,_mm_add_pd(swF3,_mm_mul_pd(d,swF4)))));
 651
 652             /* Evaluate switch function */
 653             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
 654             felec            = _mm_sub_pd( _mm_mul_pd(felec,sw) , _mm_mul_pd(rinv00,_mm_mul_pd(velec,dsw)) );
 655             cutoff_mask      = _mm_cmplt_pd(rsq00,rcutoff2);
 656
 657             fscal            = felec;
 658
 659             fscal            = _mm_and_pd(fscal,cutoff_mask);
 660
 661             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 662
 663             /* Calculate temporary vectorial force */
 664             tx               = _mm_mul_pd(fscal,dx00);
 665             ty               = _mm_mul_pd(fscal,dy00);
 666             tz               = _mm_mul_pd(fscal,dz00);
 667
 668             /* Update vectorial force */
 669             fix0             = _mm_add_pd(fix0,tx);
 670             fiy0             = _mm_add_pd(fiy0,ty);
 671             fiz0             = _mm_add_pd(fiz0,tz);
 672
 673             gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,tx,ty,tz);
 674
 675             }
 676
 677             /* Inner loop uses 62 flops */
 678         }
 679
 680         /* End of innermost loop */
 681
 682         gmx_mm_update_iforce_1atom_swizzle_pd(fix0,fiy0,fiz0,
 683                                               f+i_coord_offset,fshift+i_shift_offset);
 684
 685         /* Increment number of inner iterations */
 686         inneriter                  += j_index_end - j_index_start;
 687
 688         /* Outer loop uses 7 flops */
 689     }
 690
 691     /* Increment number of outer iterations */
 692     outeriter        += nri;
 693
 694     /* Update outer/inner flops */
 695
 696     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_F,outeriter*7 + inneriter*62);
 697 }