src/gromacs/gmxlib/nonbonded/nb_kernel_sse2_double/nb_kernel_ElecEwSw_VdwNone_GeomP1P1_sse2_double.c

   1 /*
   2  * This file is part of the GROMACS molecular simulation package.
   3  *
   4  * Copyright (c) 2012,2013,2014, 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 sse2_double kernel generator.
  37  */
  38 #include "config.h"
  39
  40 #include <math.h>
  41
  42 #include "../nb_kernel.h"
  43 #include "gromacs/legacyheaders/types/simple.h"
  44 #include "gromacs/math/vec.h"
  45 #include "gromacs/legacyheaders/nrnb.h"
  46
  47 #include "gromacs/simd/math_x86_sse2_double.h"
  48 #include "kernelutil_x86_sse2_double.h"
  49
  50 /*
  51  * Gromacs nonbonded kernel:   nb_kernel_ElecEwSw_VdwNone_GeomP1P1_VF_sse2_double
  52  * Electrostatics interaction: Ewald
  53  * VdW interaction:            None
  54  * Geometry:                   Particle-Particle
  55  * Calculate force/pot:        PotentialAndForce
  56  */
  57 void
  58 nb_kernel_ElecEwSw_VdwNone_GeomP1P1_VF_sse2_double
  59                     (t_nblist                    * gmx_restrict       nlist,
  60                      rvec                        * gmx_restrict          xx,
  61                      rvec                        * gmx_restrict          ff,
  62                      t_forcerec                  * gmx_restrict          fr,
  63                      t_mdatoms                   * gmx_restrict     mdatoms,
  64                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
  65                      t_nrnb                      * gmx_restrict        nrnb)
  66 {
  67     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
  68      * just 0 for non-waters.
  69      * Suffixes A,B refer to j loop unrolling done with SSE double precision, e.g. for the two different
  70      * jnr indices corresponding to data put in the four positions in the SIMD register.
  71      */
  72     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
  73     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
  74     int              jnrA,jnrB;
  75     int              j_coord_offsetA,j_coord_offsetB;
  76     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
  77     real             rcutoff_scalar;
  78     real             *shiftvec,*fshift,*x,*f;
  79     __m128d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
  80     int              vdwioffset0;
  81     __m128d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
  82     int              vdwjidx0A,vdwjidx0B;
  83     __m128d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  84     __m128d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  85     __m128d          velec,felec,velecsum,facel,crf,krf,krf2;
  86     real             *charge;
  87     __m128i          ewitab;
  88     __m128d          ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
  89     real             *ewtab;
  90     __m128d          rswitch,swV3,swV4,swV5,swF2,swF3,swF4,d,d2,sw,dsw;
  91     real             rswitch_scalar,d_scalar;
  92     __m128d          dummy_mask,cutoff_mask;
  93     __m128d          signbit   = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
  94     __m128d          one     = _mm_set1_pd(1.0);
  95     __m128d          two     = _mm_set1_pd(2.0);
  96     x                = xx[0];
  97     f                = ff[0];
  98
  99     nri              = nlist->nri;
 100     iinr             = nlist->iinr;
 101     jindex           = nlist->jindex;
 102     jjnr             = nlist->jjnr;
 103     shiftidx         = nlist->shift;
 104     gid              = nlist->gid;
 105     shiftvec         = fr->shift_vec[0];
 106     fshift           = fr->fshift[0];
 107     facel            = _mm_set1_pd(fr->epsfac);
 108     charge           = mdatoms->chargeA;
 109
 110     sh_ewald         = _mm_set1_pd(fr->ic->sh_ewald);
 111     ewtab            = fr->ic->tabq_coul_FDV0;
 112     ewtabscale       = _mm_set1_pd(fr->ic->tabq_scale);
 113     ewtabhalfspace   = _mm_set1_pd(0.5/fr->ic->tabq_scale);
 114
 115     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 116     rcutoff_scalar   = fr->rcoulomb;
 117     rcutoff          = _mm_set1_pd(rcutoff_scalar);
 118     rcutoff2         = _mm_mul_pd(rcutoff,rcutoff);
 119
 120     rswitch_scalar   = fr->rcoulomb_switch;
 121     rswitch          = _mm_set1_pd(rswitch_scalar);
 122     /* Setup switch parameters */
 123     d_scalar         = rcutoff_scalar-rswitch_scalar;
 124     d                = _mm_set1_pd(d_scalar);
 125     swV3             = _mm_set1_pd(-10.0/(d_scalar*d_scalar*d_scalar));
 126     swV4             = _mm_set1_pd( 15.0/(d_scalar*d_scalar*d_scalar*d_scalar));
 127     swV5             = _mm_set1_pd( -6.0/(d_scalar*d_scalar*d_scalar*d_scalar*d_scalar));
 128     swF2             = _mm_set1_pd(-30.0/(d_scalar*d_scalar*d_scalar));
 129     swF3             = _mm_set1_pd( 60.0/(d_scalar*d_scalar*d_scalar*d_scalar));
 130     swF4             = _mm_set1_pd(-30.0/(d_scalar*d_scalar*d_scalar*d_scalar*d_scalar));
 131
 132     /* Avoid stupid compiler warnings */
 133     jnrA = jnrB = 0;
 134     j_coord_offsetA = 0;
 135     j_coord_offsetB = 0;
 136
 137     outeriter        = 0;
 138     inneriter        = 0;
 139
 140     /* Start outer loop over neighborlists */
 141     for(iidx=0; iidx<nri; iidx++)
 142     {
 143         /* Load shift vector for this list */
 144         i_shift_offset   = DIM*shiftidx[iidx];
 145
 146         /* Load limits for loop over neighbors */
 147         j_index_start    = jindex[iidx];
 148         j_index_end      = jindex[iidx+1];
 149
 150         /* Get outer coordinate index */
 151         inr              = iinr[iidx];
 152         i_coord_offset   = DIM*inr;
 153
 154         /* Load i particle coords and add shift vector */
 155         gmx_mm_load_shift_and_1rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 156
 157         fix0             = _mm_setzero_pd();
 158         fiy0             = _mm_setzero_pd();
 159         fiz0             = _mm_setzero_pd();
 160
 161         /* Load parameters for i particles */
 162         iq0              = _mm_mul_pd(facel,_mm_load1_pd(charge+inr+0));
 163
 164         /* Reset potential sums */
 165         velecsum         = _mm_setzero_pd();
 166
 167         /* Start inner kernel loop */
 168         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 169         {
 170
 171             /* Get j neighbor index, and coordinate index */
 172             jnrA             = jjnr[jidx];
 173             jnrB             = jjnr[jidx+1];
 174             j_coord_offsetA  = DIM*jnrA;
 175             j_coord_offsetB  = DIM*jnrB;
 176
 177             /* load j atom coordinates */
 178             gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 179                                               &jx0,&jy0,&jz0);
 180
 181             /* Calculate displacement vector */
 182             dx00             = _mm_sub_pd(ix0,jx0);
 183             dy00             = _mm_sub_pd(iy0,jy0);
 184             dz00             = _mm_sub_pd(iz0,jz0);
 185
 186             /* Calculate squared distance and things based on it */
 187             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 188
 189             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 190
 191             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 192
 193             /* Load parameters for j particles */
 194             jq0              = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
 195
 196             /**************************
 197              * CALCULATE INTERACTIONS *
 198              **************************/
 199
 200             if (gmx_mm_any_lt(rsq00,rcutoff2))
 201             {
 202
 203             r00              = _mm_mul_pd(rsq00,rinv00);
 204
 205             /* Compute parameters for interactions between i and j atoms */
 206             qq00             = _mm_mul_pd(iq0,jq0);
 207
 208             /* EWALD ELECTROSTATICS */
 209
 210             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 211             ewrt             = _mm_mul_pd(r00,ewtabscale);
 212             ewitab           = _mm_cvttpd_epi32(ewrt);
 213             eweps            = _mm_sub_pd(ewrt,_mm_cvtepi32_pd(ewitab));
 214             ewitab           = _mm_slli_epi32(ewitab,2);
 215             ewtabF           = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 216             ewtabD           = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,1) );
 217             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 218             ewtabV           = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,0) +2);
 219             ewtabFn          = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,1) +2);
 220             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 221             felec            = _mm_add_pd(ewtabF,_mm_mul_pd(eweps,ewtabD));
 222             velec            = _mm_sub_pd(ewtabV,_mm_mul_pd(_mm_mul_pd(ewtabhalfspace,eweps),_mm_add_pd(ewtabF,felec)));
 223             velec            = _mm_mul_pd(qq00,_mm_sub_pd(rinv00,velec));
 224             felec            = _mm_mul_pd(_mm_mul_pd(qq00,rinv00),_mm_sub_pd(rinvsq00,felec));
 225
 226             d                = _mm_sub_pd(r00,rswitch);
 227             d                = _mm_max_pd(d,_mm_setzero_pd());
 228             d2               = _mm_mul_pd(d,d);
 229             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)))))));
 230
 231             dsw              = _mm_mul_pd(d2,_mm_add_pd(swF2,_mm_mul_pd(d,_mm_add_pd(swF3,_mm_mul_pd(d,swF4)))));
 232
 233             /* Evaluate switch function */
 234             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
 235             felec            = _mm_sub_pd( _mm_mul_pd(felec,sw) , _mm_mul_pd(rinv00,_mm_mul_pd(velec,dsw)) );
 236             velec            = _mm_mul_pd(velec,sw);
 237             cutoff_mask      = _mm_cmplt_pd(rsq00,rcutoff2);
 238
 239             /* Update potential sum for this i atom from the interaction with this j atom. */
 240             velec            = _mm_and_pd(velec,cutoff_mask);
 241             velecsum         = _mm_add_pd(velecsum,velec);
 242
 243             fscal            = felec;
 244
 245             fscal            = _mm_and_pd(fscal,cutoff_mask);
 246
 247             /* Calculate temporary vectorial force */
 248             tx               = _mm_mul_pd(fscal,dx00);
 249             ty               = _mm_mul_pd(fscal,dy00);
 250             tz               = _mm_mul_pd(fscal,dz00);
 251
 252             /* Update vectorial force */
 253             fix0             = _mm_add_pd(fix0,tx);
 254             fiy0             = _mm_add_pd(fiy0,ty);
 255             fiz0             = _mm_add_pd(fiz0,tz);
 256
 257             gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,tx,ty,tz);
 258
 259             }
 260
 261             /* Inner loop uses 65 flops */
 262         }
 263
 264         if(jidx<j_index_end)
 265         {
 266
 267             jnrA             = jjnr[jidx];
 268             j_coord_offsetA  = DIM*jnrA;
 269
 270             /* load j atom coordinates */
 271             gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
 272                                               &jx0,&jy0,&jz0);
 273
 274             /* Calculate displacement vector */
 275             dx00             = _mm_sub_pd(ix0,jx0);
 276             dy00             = _mm_sub_pd(iy0,jy0);
 277             dz00             = _mm_sub_pd(iz0,jz0);
 278
 279             /* Calculate squared distance and things based on it */
 280             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 281
 282             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 283
 284             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 285
 286             /* Load parameters for j particles */
 287             jq0              = _mm_load_sd(charge+jnrA+0);
 288
 289             /**************************
 290              * CALCULATE INTERACTIONS *
 291              **************************/
 292
 293             if (gmx_mm_any_lt(rsq00,rcutoff2))
 294             {
 295
 296             r00              = _mm_mul_pd(rsq00,rinv00);
 297
 298             /* Compute parameters for interactions between i and j atoms */
 299             qq00             = _mm_mul_pd(iq0,jq0);
 300
 301             /* EWALD ELECTROSTATICS */
 302
 303             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 304             ewrt             = _mm_mul_pd(r00,ewtabscale);
 305             ewitab           = _mm_cvttpd_epi32(ewrt);
 306             eweps            = _mm_sub_pd(ewrt,_mm_cvtepi32_pd(ewitab));
 307             ewitab           = _mm_slli_epi32(ewitab,2);
 308             ewtabF           = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 309             ewtabD           = _mm_setzero_pd();
 310             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 311             ewtabV           = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,0) +2);
 312             ewtabFn          = _mm_setzero_pd();
 313             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 314             felec            = _mm_add_pd(ewtabF,_mm_mul_pd(eweps,ewtabD));
 315             velec            = _mm_sub_pd(ewtabV,_mm_mul_pd(_mm_mul_pd(ewtabhalfspace,eweps),_mm_add_pd(ewtabF,felec)));
 316             velec            = _mm_mul_pd(qq00,_mm_sub_pd(rinv00,velec));
 317             felec            = _mm_mul_pd(_mm_mul_pd(qq00,rinv00),_mm_sub_pd(rinvsq00,felec));
 318
 319             d                = _mm_sub_pd(r00,rswitch);
 320             d                = _mm_max_pd(d,_mm_setzero_pd());
 321             d2               = _mm_mul_pd(d,d);
 322             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)))))));
 323
 324             dsw              = _mm_mul_pd(d2,_mm_add_pd(swF2,_mm_mul_pd(d,_mm_add_pd(swF3,_mm_mul_pd(d,swF4)))));
 325
 326             /* Evaluate switch function */
 327             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
 328             felec            = _mm_sub_pd( _mm_mul_pd(felec,sw) , _mm_mul_pd(rinv00,_mm_mul_pd(velec,dsw)) );
 329             velec            = _mm_mul_pd(velec,sw);
 330             cutoff_mask      = _mm_cmplt_pd(rsq00,rcutoff2);
 331
 332             /* Update potential sum for this i atom from the interaction with this j atom. */
 333             velec            = _mm_and_pd(velec,cutoff_mask);
 334             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 335             velecsum         = _mm_add_pd(velecsum,velec);
 336
 337             fscal            = felec;
 338
 339             fscal            = _mm_and_pd(fscal,cutoff_mask);
 340
 341             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 342
 343             /* Calculate temporary vectorial force */
 344             tx               = _mm_mul_pd(fscal,dx00);
 345             ty               = _mm_mul_pd(fscal,dy00);
 346             tz               = _mm_mul_pd(fscal,dz00);
 347
 348             /* Update vectorial force */
 349             fix0             = _mm_add_pd(fix0,tx);
 350             fiy0             = _mm_add_pd(fiy0,ty);
 351             fiz0             = _mm_add_pd(fiz0,tz);
 352
 353             gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,tx,ty,tz);
 354
 355             }
 356
 357             /* Inner loop uses 65 flops */
 358         }
 359
 360         /* End of innermost loop */
 361
 362         gmx_mm_update_iforce_1atom_swizzle_pd(fix0,fiy0,fiz0,
 363                                               f+i_coord_offset,fshift+i_shift_offset);
 364
 365         ggid                        = gid[iidx];
 366         /* Update potential energies */
 367         gmx_mm_update_1pot_pd(velecsum,kernel_data->energygrp_elec+ggid);
 368
 369         /* Increment number of inner iterations */
 370         inneriter                  += j_index_end - j_index_start;
 371
 372         /* Outer loop uses 8 flops */
 373     }
 374
 375     /* Increment number of outer iterations */
 376     outeriter        += nri;
 377
 378     /* Update outer/inner flops */
 379
 380     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VF,outeriter*8 + inneriter*65);
 381 }
 382 /*
 383  * Gromacs nonbonded kernel:   nb_kernel_ElecEwSw_VdwNone_GeomP1P1_F_sse2_double
 384  * Electrostatics interaction: Ewald
 385  * VdW interaction:            None
 386  * Geometry:                   Particle-Particle
 387  * Calculate force/pot:        Force
 388  */
 389 void
 390 nb_kernel_ElecEwSw_VdwNone_GeomP1P1_F_sse2_double
 391                     (t_nblist                    * gmx_restrict       nlist,
 392                      rvec                        * gmx_restrict          xx,
 393                      rvec                        * gmx_restrict          ff,
 394                      t_forcerec                  * gmx_restrict          fr,
 395                      t_mdatoms                   * gmx_restrict     mdatoms,
 396                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 397                      t_nrnb                      * gmx_restrict        nrnb)
 398 {
 399     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 400      * just 0 for non-waters.
 401      * Suffixes A,B refer to j loop unrolling done with SSE double precision, e.g. for the two different
 402      * jnr indices corresponding to data put in the four positions in the SIMD register.
 403      */
 404     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 405     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 406     int              jnrA,jnrB;
 407     int              j_coord_offsetA,j_coord_offsetB;
 408     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 409     real             rcutoff_scalar;
 410     real             *shiftvec,*fshift,*x,*f;
 411     __m128d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 412     int              vdwioffset0;
 413     __m128d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 414     int              vdwjidx0A,vdwjidx0B;
 415     __m128d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 416     __m128d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 417     __m128d          velec,felec,velecsum,facel,crf,krf,krf2;
 418     real             *charge;
 419     __m128i          ewitab;
 420     __m128d          ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
 421     real             *ewtab;
 422     __m128d          rswitch,swV3,swV4,swV5,swF2,swF3,swF4,d,d2,sw,dsw;
 423     real             rswitch_scalar,d_scalar;
 424     __m128d          dummy_mask,cutoff_mask;
 425     __m128d          signbit   = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
 426     __m128d          one     = _mm_set1_pd(1.0);
 427     __m128d          two     = _mm_set1_pd(2.0);
 428     x                = xx[0];
 429     f                = ff[0];
 430
 431     nri              = nlist->nri;
 432     iinr             = nlist->iinr;
 433     jindex           = nlist->jindex;
 434     jjnr             = nlist->jjnr;
 435     shiftidx         = nlist->shift;
 436     gid              = nlist->gid;
 437     shiftvec         = fr->shift_vec[0];
 438     fshift           = fr->fshift[0];
 439     facel            = _mm_set1_pd(fr->epsfac);
 440     charge           = mdatoms->chargeA;
 441
 442     sh_ewald         = _mm_set1_pd(fr->ic->sh_ewald);
 443     ewtab            = fr->ic->tabq_coul_FDV0;
 444     ewtabscale       = _mm_set1_pd(fr->ic->tabq_scale);
 445     ewtabhalfspace   = _mm_set1_pd(0.5/fr->ic->tabq_scale);
 446
 447     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 448     rcutoff_scalar   = fr->rcoulomb;
 449     rcutoff          = _mm_set1_pd(rcutoff_scalar);
 450     rcutoff2         = _mm_mul_pd(rcutoff,rcutoff);
 451
 452     rswitch_scalar   = fr->rcoulomb_switch;
 453     rswitch          = _mm_set1_pd(rswitch_scalar);
 454     /* Setup switch parameters */
 455     d_scalar         = rcutoff_scalar-rswitch_scalar;
 456     d                = _mm_set1_pd(d_scalar);
 457     swV3             = _mm_set1_pd(-10.0/(d_scalar*d_scalar*d_scalar));
 458     swV4             = _mm_set1_pd( 15.0/(d_scalar*d_scalar*d_scalar*d_scalar));
 459     swV5             = _mm_set1_pd( -6.0/(d_scalar*d_scalar*d_scalar*d_scalar*d_scalar));
 460     swF2             = _mm_set1_pd(-30.0/(d_scalar*d_scalar*d_scalar));
 461     swF3             = _mm_set1_pd( 60.0/(d_scalar*d_scalar*d_scalar*d_scalar));
 462     swF4             = _mm_set1_pd(-30.0/(d_scalar*d_scalar*d_scalar*d_scalar*d_scalar));
 463
 464     /* Avoid stupid compiler warnings */
 465     jnrA = jnrB = 0;
 466     j_coord_offsetA = 0;
 467     j_coord_offsetB = 0;
 468
 469     outeriter        = 0;
 470     inneriter        = 0;
 471
 472     /* Start outer loop over neighborlists */
 473     for(iidx=0; iidx<nri; iidx++)
 474     {
 475         /* Load shift vector for this list */
 476         i_shift_offset   = DIM*shiftidx[iidx];
 477
 478         /* Load limits for loop over neighbors */
 479         j_index_start    = jindex[iidx];
 480         j_index_end      = jindex[iidx+1];
 481
 482         /* Get outer coordinate index */
 483         inr              = iinr[iidx];
 484         i_coord_offset   = DIM*inr;
 485
 486         /* Load i particle coords and add shift vector */
 487         gmx_mm_load_shift_and_1rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 488
 489         fix0             = _mm_setzero_pd();
 490         fiy0             = _mm_setzero_pd();
 491         fiz0             = _mm_setzero_pd();
 492
 493         /* Load parameters for i particles */
 494         iq0              = _mm_mul_pd(facel,_mm_load1_pd(charge+inr+0));
 495
 496         /* Start inner kernel loop */
 497         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 498         {
 499
 500             /* Get j neighbor index, and coordinate index */
 501             jnrA             = jjnr[jidx];
 502             jnrB             = jjnr[jidx+1];
 503             j_coord_offsetA  = DIM*jnrA;
 504             j_coord_offsetB  = DIM*jnrB;
 505
 506             /* load j atom coordinates */
 507             gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 508                                               &jx0,&jy0,&jz0);
 509
 510             /* Calculate displacement vector */
 511             dx00             = _mm_sub_pd(ix0,jx0);
 512             dy00             = _mm_sub_pd(iy0,jy0);
 513             dz00             = _mm_sub_pd(iz0,jz0);
 514
 515             /* Calculate squared distance and things based on it */
 516             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 517
 518             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 519
 520             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 521
 522             /* Load parameters for j particles */
 523             jq0              = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
 524
 525             /**************************
 526              * CALCULATE INTERACTIONS *
 527              **************************/
 528
 529             if (gmx_mm_any_lt(rsq00,rcutoff2))
 530             {
 531
 532             r00              = _mm_mul_pd(rsq00,rinv00);
 533
 534             /* Compute parameters for interactions between i and j atoms */
 535             qq00             = _mm_mul_pd(iq0,jq0);
 536
 537             /* EWALD ELECTROSTATICS */
 538
 539             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 540             ewrt             = _mm_mul_pd(r00,ewtabscale);
 541             ewitab           = _mm_cvttpd_epi32(ewrt);
 542             eweps            = _mm_sub_pd(ewrt,_mm_cvtepi32_pd(ewitab));
 543             ewitab           = _mm_slli_epi32(ewitab,2);
 544             ewtabF           = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 545             ewtabD           = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,1) );
 546             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 547             ewtabV           = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,0) +2);
 548             ewtabFn          = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,1) +2);
 549             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 550             felec            = _mm_add_pd(ewtabF,_mm_mul_pd(eweps,ewtabD));
 551             velec            = _mm_sub_pd(ewtabV,_mm_mul_pd(_mm_mul_pd(ewtabhalfspace,eweps),_mm_add_pd(ewtabF,felec)));
 552             velec            = _mm_mul_pd(qq00,_mm_sub_pd(rinv00,velec));
 553             felec            = _mm_mul_pd(_mm_mul_pd(qq00,rinv00),_mm_sub_pd(rinvsq00,felec));
 554
 555             d                = _mm_sub_pd(r00,rswitch);
 556             d                = _mm_max_pd(d,_mm_setzero_pd());
 557             d2               = _mm_mul_pd(d,d);
 558             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)))))));
 559
 560             dsw              = _mm_mul_pd(d2,_mm_add_pd(swF2,_mm_mul_pd(d,_mm_add_pd(swF3,_mm_mul_pd(d,swF4)))));
 561
 562             /* Evaluate switch function */
 563             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
 564             felec            = _mm_sub_pd( _mm_mul_pd(felec,sw) , _mm_mul_pd(rinv00,_mm_mul_pd(velec,dsw)) );
 565             cutoff_mask      = _mm_cmplt_pd(rsq00,rcutoff2);
 566
 567             fscal            = felec;
 568
 569             fscal            = _mm_and_pd(fscal,cutoff_mask);
 570
 571             /* Calculate temporary vectorial force */
 572             tx               = _mm_mul_pd(fscal,dx00);
 573             ty               = _mm_mul_pd(fscal,dy00);
 574             tz               = _mm_mul_pd(fscal,dz00);
 575
 576             /* Update vectorial force */
 577             fix0             = _mm_add_pd(fix0,tx);
 578             fiy0             = _mm_add_pd(fiy0,ty);
 579             fiz0             = _mm_add_pd(fiz0,tz);
 580
 581             gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,tx,ty,tz);
 582
 583             }
 584
 585             /* Inner loop uses 62 flops */
 586         }
 587
 588         if(jidx<j_index_end)
 589         {
 590
 591             jnrA             = jjnr[jidx];
 592             j_coord_offsetA  = DIM*jnrA;
 593
 594             /* load j atom coordinates */
 595             gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
 596                                               &jx0,&jy0,&jz0);
 597
 598             /* Calculate displacement vector */
 599             dx00             = _mm_sub_pd(ix0,jx0);
 600             dy00             = _mm_sub_pd(iy0,jy0);
 601             dz00             = _mm_sub_pd(iz0,jz0);
 602
 603             /* Calculate squared distance and things based on it */
 604             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 605
 606             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 607
 608             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 609
 610             /* Load parameters for j particles */
 611             jq0              = _mm_load_sd(charge+jnrA+0);
 612
 613             /**************************
 614              * CALCULATE INTERACTIONS *
 615              **************************/
 616
 617             if (gmx_mm_any_lt(rsq00,rcutoff2))
 618             {
 619
 620             r00              = _mm_mul_pd(rsq00,rinv00);
 621
 622             /* Compute parameters for interactions between i and j atoms */
 623             qq00             = _mm_mul_pd(iq0,jq0);
 624
 625             /* EWALD ELECTROSTATICS */
 626
 627             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 628             ewrt             = _mm_mul_pd(r00,ewtabscale);
 629             ewitab           = _mm_cvttpd_epi32(ewrt);
 630             eweps            = _mm_sub_pd(ewrt,_mm_cvtepi32_pd(ewitab));
 631             ewitab           = _mm_slli_epi32(ewitab,2);
 632             ewtabF           = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 633             ewtabD           = _mm_setzero_pd();
 634             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 635             ewtabV           = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,0) +2);
 636             ewtabFn          = _mm_setzero_pd();
 637             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 638             felec            = _mm_add_pd(ewtabF,_mm_mul_pd(eweps,ewtabD));
 639             velec            = _mm_sub_pd(ewtabV,_mm_mul_pd(_mm_mul_pd(ewtabhalfspace,eweps),_mm_add_pd(ewtabF,felec)));
 640             velec            = _mm_mul_pd(qq00,_mm_sub_pd(rinv00,velec));
 641             felec            = _mm_mul_pd(_mm_mul_pd(qq00,rinv00),_mm_sub_pd(rinvsq00,felec));
 642
 643             d                = _mm_sub_pd(r00,rswitch);
 644             d                = _mm_max_pd(d,_mm_setzero_pd());
 645             d2               = _mm_mul_pd(d,d);
 646             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)))))));
 647
 648             dsw              = _mm_mul_pd(d2,_mm_add_pd(swF2,_mm_mul_pd(d,_mm_add_pd(swF3,_mm_mul_pd(d,swF4)))));
 649
 650             /* Evaluate switch function */
 651             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
 652             felec            = _mm_sub_pd( _mm_mul_pd(felec,sw) , _mm_mul_pd(rinv00,_mm_mul_pd(velec,dsw)) );
 653             cutoff_mask      = _mm_cmplt_pd(rsq00,rcutoff2);
 654
 655             fscal            = felec;
 656
 657             fscal            = _mm_and_pd(fscal,cutoff_mask);
 658
 659             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 660
 661             /* Calculate temporary vectorial force */
 662             tx               = _mm_mul_pd(fscal,dx00);
 663             ty               = _mm_mul_pd(fscal,dy00);
 664             tz               = _mm_mul_pd(fscal,dz00);
 665
 666             /* Update vectorial force */
 667             fix0             = _mm_add_pd(fix0,tx);
 668             fiy0             = _mm_add_pd(fiy0,ty);
 669             fiz0             = _mm_add_pd(fiz0,tz);
 670
 671             gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,tx,ty,tz);
 672
 673             }
 674
 675             /* Inner loop uses 62 flops */
 676         }
 677
 678         /* End of innermost loop */
 679
 680         gmx_mm_update_iforce_1atom_swizzle_pd(fix0,fiy0,fiz0,
 681                                               f+i_coord_offset,fshift+i_shift_offset);
 682
 683         /* Increment number of inner iterations */
 684         inneriter                  += j_index_end - j_index_start;
 685
 686         /* Outer loop uses 7 flops */
 687     }
 688
 689     /* Increment number of outer iterations */
 690     outeriter        += nri;
 691
 692     /* Update outer/inner flops */
 693
 694     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_F,outeriter*7 + inneriter*62);
 695 }