src/gromacs/gmxlib/nonbonded/nb_kernel_avx_128_fma_double/nb_kernel_ElecEwSw_VdwNone_GeomP1P1_avx_128_fma_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 avx_128_fma_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_avx_128_fma_double.h"
  50 #include "kernelutil_x86_avx_128_fma_double.h"
  51
  52 /*
  53  * Gromacs nonbonded kernel:   nb_kernel_ElecEwSw_VdwNone_GeomP1P1_VF_avx_128_fma_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_avx_128_fma_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,twoeweps,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 #ifdef __XOP__
 216             eweps            = _mm_frcz_pd(ewrt);
 217 #else
 218             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 219 #endif
 220             twoeweps         = _mm_add_pd(eweps,eweps);
 221             ewitab           = _mm_slli_epi32(ewitab,2);
 222             ewtabF           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 223             ewtabD           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
 224             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 225             ewtabV           = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
 226             ewtabFn          = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,1) +2);
 227             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 228             felec            = _mm_macc_pd(eweps,ewtabD,ewtabF);
 229             velec            = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
 230             velec            = _mm_mul_pd(qq00,_mm_sub_pd(rinv00,velec));
 231             felec            = _mm_mul_pd(_mm_mul_pd(qq00,rinv00),_mm_sub_pd(rinvsq00,felec));
 232
 233             d                = _mm_sub_pd(r00,rswitch);
 234             d                = _mm_max_pd(d,_mm_setzero_pd());
 235             d2               = _mm_mul_pd(d,d);
 236             sw               = _mm_add_pd(one,_mm_mul_pd(d2,_mm_mul_pd(d,_mm_macc_pd(d,_mm_macc_pd(d,swV5,swV4),swV3))));
 237
 238             dsw              = _mm_mul_pd(d2,_mm_macc_pd(d,_mm_macc_pd(d,swF4,swF3),swF2));
 239
 240             /* Evaluate switch function */
 241             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
 242             felec            = _mm_msub_pd( felec,sw , _mm_mul_pd(rinv00,_mm_mul_pd(velec,dsw)) );
 243             velec            = _mm_mul_pd(velec,sw);
 244             cutoff_mask      = _mm_cmplt_pd(rsq00,rcutoff2);
 245
 246             /* Update potential sum for this i atom from the interaction with this j atom. */
 247             velec            = _mm_and_pd(velec,cutoff_mask);
 248             velecsum         = _mm_add_pd(velecsum,velec);
 249
 250             fscal            = felec;
 251
 252             fscal            = _mm_and_pd(fscal,cutoff_mask);
 253
 254             /* Update vectorial force */
 255             fix0             = _mm_macc_pd(dx00,fscal,fix0);
 256             fiy0             = _mm_macc_pd(dy00,fscal,fiy0);
 257             fiz0             = _mm_macc_pd(dz00,fscal,fiz0);
 258
 259             gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,
 260                                                    _mm_mul_pd(dx00,fscal),
 261                                                    _mm_mul_pd(dy00,fscal),
 262                                                    _mm_mul_pd(dz00,fscal));
 263
 264             }
 265
 266             /* Inner loop uses 68 flops */
 267         }
 268
 269         if(jidx<j_index_end)
 270         {
 271
 272             jnrA             = jjnr[jidx];
 273             j_coord_offsetA  = DIM*jnrA;
 274
 275             /* load j atom coordinates */
 276             gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
 277                                               &jx0,&jy0,&jz0);
 278
 279             /* Calculate displacement vector */
 280             dx00             = _mm_sub_pd(ix0,jx0);
 281             dy00             = _mm_sub_pd(iy0,jy0);
 282             dz00             = _mm_sub_pd(iz0,jz0);
 283
 284             /* Calculate squared distance and things based on it */
 285             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 286
 287             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 288
 289             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 290
 291             /* Load parameters for j particles */
 292             jq0              = _mm_load_sd(charge+jnrA+0);
 293
 294             /**************************
 295              * CALCULATE INTERACTIONS *
 296              **************************/
 297
 298             if (gmx_mm_any_lt(rsq00,rcutoff2))
 299             {
 300
 301             r00              = _mm_mul_pd(rsq00,rinv00);
 302
 303             /* Compute parameters for interactions between i and j atoms */
 304             qq00             = _mm_mul_pd(iq0,jq0);
 305
 306             /* EWALD ELECTROSTATICS */
 307
 308             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 309             ewrt             = _mm_mul_pd(r00,ewtabscale);
 310             ewitab           = _mm_cvttpd_epi32(ewrt);
 311 #ifdef __XOP__
 312             eweps            = _mm_frcz_pd(ewrt);
 313 #else
 314             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 315 #endif
 316             twoeweps         = _mm_add_pd(eweps,eweps);
 317             ewitab           = _mm_slli_epi32(ewitab,2);
 318             ewtabF           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 319             ewtabD           = _mm_setzero_pd();
 320             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 321             ewtabV           = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
 322             ewtabFn          = _mm_setzero_pd();
 323             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 324             felec            = _mm_macc_pd(eweps,ewtabD,ewtabF);
 325             velec            = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
 326             velec            = _mm_mul_pd(qq00,_mm_sub_pd(rinv00,velec));
 327             felec            = _mm_mul_pd(_mm_mul_pd(qq00,rinv00),_mm_sub_pd(rinvsq00,felec));
 328
 329             d                = _mm_sub_pd(r00,rswitch);
 330             d                = _mm_max_pd(d,_mm_setzero_pd());
 331             d2               = _mm_mul_pd(d,d);
 332             sw               = _mm_add_pd(one,_mm_mul_pd(d2,_mm_mul_pd(d,_mm_macc_pd(d,_mm_macc_pd(d,swV5,swV4),swV3))));
 333
 334             dsw              = _mm_mul_pd(d2,_mm_macc_pd(d,_mm_macc_pd(d,swF4,swF3),swF2));
 335
 336             /* Evaluate switch function */
 337             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
 338             felec            = _mm_msub_pd( felec,sw , _mm_mul_pd(rinv00,_mm_mul_pd(velec,dsw)) );
 339             velec            = _mm_mul_pd(velec,sw);
 340             cutoff_mask      = _mm_cmplt_pd(rsq00,rcutoff2);
 341
 342             /* Update potential sum for this i atom from the interaction with this j atom. */
 343             velec            = _mm_and_pd(velec,cutoff_mask);
 344             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 345             velecsum         = _mm_add_pd(velecsum,velec);
 346
 347             fscal            = felec;
 348
 349             fscal            = _mm_and_pd(fscal,cutoff_mask);
 350
 351             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 352
 353             /* Update vectorial force */
 354             fix0             = _mm_macc_pd(dx00,fscal,fix0);
 355             fiy0             = _mm_macc_pd(dy00,fscal,fiy0);
 356             fiz0             = _mm_macc_pd(dz00,fscal,fiz0);
 357
 358             gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,
 359                                                    _mm_mul_pd(dx00,fscal),
 360                                                    _mm_mul_pd(dy00,fscal),
 361                                                    _mm_mul_pd(dz00,fscal));
 362
 363             }
 364
 365             /* Inner loop uses 68 flops */
 366         }
 367
 368         /* End of innermost loop */
 369
 370         gmx_mm_update_iforce_1atom_swizzle_pd(fix0,fiy0,fiz0,
 371                                               f+i_coord_offset,fshift+i_shift_offset);
 372
 373         ggid                        = gid[iidx];
 374         /* Update potential energies */
 375         gmx_mm_update_1pot_pd(velecsum,kernel_data->energygrp_elec+ggid);
 376
 377         /* Increment number of inner iterations */
 378         inneriter                  += j_index_end - j_index_start;
 379
 380         /* Outer loop uses 8 flops */
 381     }
 382
 383     /* Increment number of outer iterations */
 384     outeriter        += nri;
 385
 386     /* Update outer/inner flops */
 387
 388     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VF,outeriter*8 + inneriter*68);
 389 }
 390 /*
 391  * Gromacs nonbonded kernel:   nb_kernel_ElecEwSw_VdwNone_GeomP1P1_F_avx_128_fma_double
 392  * Electrostatics interaction: Ewald
 393  * VdW interaction:            None
 394  * Geometry:                   Particle-Particle
 395  * Calculate force/pot:        Force
 396  */
 397 void
 398 nb_kernel_ElecEwSw_VdwNone_GeomP1P1_F_avx_128_fma_double
 399                     (t_nblist                    * gmx_restrict       nlist,
 400                      rvec                        * gmx_restrict          xx,
 401                      rvec                        * gmx_restrict          ff,
 402                      t_forcerec                  * gmx_restrict          fr,
 403                      t_mdatoms                   * gmx_restrict     mdatoms,
 404                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 405                      t_nrnb                      * gmx_restrict        nrnb)
 406 {
 407     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 408      * just 0 for non-waters.
 409      * Suffixes A,B refer to j loop unrolling done with SSE double precision, e.g. for the two different
 410      * jnr indices corresponding to data put in the four positions in the SIMD register.
 411      */
 412     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 413     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 414     int              jnrA,jnrB;
 415     int              j_coord_offsetA,j_coord_offsetB;
 416     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 417     real             rcutoff_scalar;
 418     real             *shiftvec,*fshift,*x,*f;
 419     __m128d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 420     int              vdwioffset0;
 421     __m128d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 422     int              vdwjidx0A,vdwjidx0B;
 423     __m128d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 424     __m128d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 425     __m128d          velec,felec,velecsum,facel,crf,krf,krf2;
 426     real             *charge;
 427     __m128i          ewitab;
 428     __m128d          ewtabscale,eweps,twoeweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
 429     real             *ewtab;
 430     __m128d          rswitch,swV3,swV4,swV5,swF2,swF3,swF4,d,d2,sw,dsw;
 431     real             rswitch_scalar,d_scalar;
 432     __m128d          dummy_mask,cutoff_mask;
 433     __m128d          signbit   = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
 434     __m128d          one     = _mm_set1_pd(1.0);
 435     __m128d          two     = _mm_set1_pd(2.0);
 436     x                = xx[0];
 437     f                = ff[0];
 438
 439     nri              = nlist->nri;
 440     iinr             = nlist->iinr;
 441     jindex           = nlist->jindex;
 442     jjnr             = nlist->jjnr;
 443     shiftidx         = nlist->shift;
 444     gid              = nlist->gid;
 445     shiftvec         = fr->shift_vec[0];
 446     fshift           = fr->fshift[0];
 447     facel            = _mm_set1_pd(fr->epsfac);
 448     charge           = mdatoms->chargeA;
 449
 450     sh_ewald         = _mm_set1_pd(fr->ic->sh_ewald);
 451     ewtab            = fr->ic->tabq_coul_FDV0;
 452     ewtabscale       = _mm_set1_pd(fr->ic->tabq_scale);
 453     ewtabhalfspace   = _mm_set1_pd(0.5/fr->ic->tabq_scale);
 454
 455     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 456     rcutoff_scalar   = fr->rcoulomb;
 457     rcutoff          = _mm_set1_pd(rcutoff_scalar);
 458     rcutoff2         = _mm_mul_pd(rcutoff,rcutoff);
 459
 460     rswitch_scalar   = fr->rcoulomb_switch;
 461     rswitch          = _mm_set1_pd(rswitch_scalar);
 462     /* Setup switch parameters */
 463     d_scalar         = rcutoff_scalar-rswitch_scalar;
 464     d                = _mm_set1_pd(d_scalar);
 465     swV3             = _mm_set1_pd(-10.0/(d_scalar*d_scalar*d_scalar));
 466     swV4             = _mm_set1_pd( 15.0/(d_scalar*d_scalar*d_scalar*d_scalar));
 467     swV5             = _mm_set1_pd( -6.0/(d_scalar*d_scalar*d_scalar*d_scalar*d_scalar));
 468     swF2             = _mm_set1_pd(-30.0/(d_scalar*d_scalar*d_scalar));
 469     swF3             = _mm_set1_pd( 60.0/(d_scalar*d_scalar*d_scalar*d_scalar));
 470     swF4             = _mm_set1_pd(-30.0/(d_scalar*d_scalar*d_scalar*d_scalar*d_scalar));
 471
 472     /* Avoid stupid compiler warnings */
 473     jnrA = jnrB = 0;
 474     j_coord_offsetA = 0;
 475     j_coord_offsetB = 0;
 476
 477     outeriter        = 0;
 478     inneriter        = 0;
 479
 480     /* Start outer loop over neighborlists */
 481     for(iidx=0; iidx<nri; iidx++)
 482     {
 483         /* Load shift vector for this list */
 484         i_shift_offset   = DIM*shiftidx[iidx];
 485
 486         /* Load limits for loop over neighbors */
 487         j_index_start    = jindex[iidx];
 488         j_index_end      = jindex[iidx+1];
 489
 490         /* Get outer coordinate index */
 491         inr              = iinr[iidx];
 492         i_coord_offset   = DIM*inr;
 493
 494         /* Load i particle coords and add shift vector */
 495         gmx_mm_load_shift_and_1rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 496
 497         fix0             = _mm_setzero_pd();
 498         fiy0             = _mm_setzero_pd();
 499         fiz0             = _mm_setzero_pd();
 500
 501         /* Load parameters for i particles */
 502         iq0              = _mm_mul_pd(facel,_mm_load1_pd(charge+inr+0));
 503
 504         /* Start inner kernel loop */
 505         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 506         {
 507
 508             /* Get j neighbor index, and coordinate index */
 509             jnrA             = jjnr[jidx];
 510             jnrB             = jjnr[jidx+1];
 511             j_coord_offsetA  = DIM*jnrA;
 512             j_coord_offsetB  = DIM*jnrB;
 513
 514             /* load j atom coordinates */
 515             gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 516                                               &jx0,&jy0,&jz0);
 517
 518             /* Calculate displacement vector */
 519             dx00             = _mm_sub_pd(ix0,jx0);
 520             dy00             = _mm_sub_pd(iy0,jy0);
 521             dz00             = _mm_sub_pd(iz0,jz0);
 522
 523             /* Calculate squared distance and things based on it */
 524             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 525
 526             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 527
 528             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 529
 530             /* Load parameters for j particles */
 531             jq0              = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
 532
 533             /**************************
 534              * CALCULATE INTERACTIONS *
 535              **************************/
 536
 537             if (gmx_mm_any_lt(rsq00,rcutoff2))
 538             {
 539
 540             r00              = _mm_mul_pd(rsq00,rinv00);
 541
 542             /* Compute parameters for interactions between i and j atoms */
 543             qq00             = _mm_mul_pd(iq0,jq0);
 544
 545             /* EWALD ELECTROSTATICS */
 546
 547             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 548             ewrt             = _mm_mul_pd(r00,ewtabscale);
 549             ewitab           = _mm_cvttpd_epi32(ewrt);
 550 #ifdef __XOP__
 551             eweps            = _mm_frcz_pd(ewrt);
 552 #else
 553             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 554 #endif
 555             twoeweps         = _mm_add_pd(eweps,eweps);
 556             ewitab           = _mm_slli_epi32(ewitab,2);
 557             ewtabF           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 558             ewtabD           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
 559             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 560             ewtabV           = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
 561             ewtabFn          = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,1) +2);
 562             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 563             felec            = _mm_macc_pd(eweps,ewtabD,ewtabF);
 564             velec            = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
 565             velec            = _mm_mul_pd(qq00,_mm_sub_pd(rinv00,velec));
 566             felec            = _mm_mul_pd(_mm_mul_pd(qq00,rinv00),_mm_sub_pd(rinvsq00,felec));
 567
 568             d                = _mm_sub_pd(r00,rswitch);
 569             d                = _mm_max_pd(d,_mm_setzero_pd());
 570             d2               = _mm_mul_pd(d,d);
 571             sw               = _mm_add_pd(one,_mm_mul_pd(d2,_mm_mul_pd(d,_mm_macc_pd(d,_mm_macc_pd(d,swV5,swV4),swV3))));
 572
 573             dsw              = _mm_mul_pd(d2,_mm_macc_pd(d,_mm_macc_pd(d,swF4,swF3),swF2));
 574
 575             /* Evaluate switch function */
 576             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
 577             felec            = _mm_msub_pd( felec,sw , _mm_mul_pd(rinv00,_mm_mul_pd(velec,dsw)) );
 578             cutoff_mask      = _mm_cmplt_pd(rsq00,rcutoff2);
 579
 580             fscal            = felec;
 581
 582             fscal            = _mm_and_pd(fscal,cutoff_mask);
 583
 584             /* Update vectorial force */
 585             fix0             = _mm_macc_pd(dx00,fscal,fix0);
 586             fiy0             = _mm_macc_pd(dy00,fscal,fiy0);
 587             fiz0             = _mm_macc_pd(dz00,fscal,fiz0);
 588
 589             gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,
 590                                                    _mm_mul_pd(dx00,fscal),
 591                                                    _mm_mul_pd(dy00,fscal),
 592                                                    _mm_mul_pd(dz00,fscal));
 593
 594             }
 595
 596             /* Inner loop uses 65 flops */
 597         }
 598
 599         if(jidx<j_index_end)
 600         {
 601
 602             jnrA             = jjnr[jidx];
 603             j_coord_offsetA  = DIM*jnrA;
 604
 605             /* load j atom coordinates */
 606             gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
 607                                               &jx0,&jy0,&jz0);
 608
 609             /* Calculate displacement vector */
 610             dx00             = _mm_sub_pd(ix0,jx0);
 611             dy00             = _mm_sub_pd(iy0,jy0);
 612             dz00             = _mm_sub_pd(iz0,jz0);
 613
 614             /* Calculate squared distance and things based on it */
 615             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 616
 617             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 618
 619             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 620
 621             /* Load parameters for j particles */
 622             jq0              = _mm_load_sd(charge+jnrA+0);
 623
 624             /**************************
 625              * CALCULATE INTERACTIONS *
 626              **************************/
 627
 628             if (gmx_mm_any_lt(rsq00,rcutoff2))
 629             {
 630
 631             r00              = _mm_mul_pd(rsq00,rinv00);
 632
 633             /* Compute parameters for interactions between i and j atoms */
 634             qq00             = _mm_mul_pd(iq0,jq0);
 635
 636             /* EWALD ELECTROSTATICS */
 637
 638             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 639             ewrt             = _mm_mul_pd(r00,ewtabscale);
 640             ewitab           = _mm_cvttpd_epi32(ewrt);
 641 #ifdef __XOP__
 642             eweps            = _mm_frcz_pd(ewrt);
 643 #else
 644             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 645 #endif
 646             twoeweps         = _mm_add_pd(eweps,eweps);
 647             ewitab           = _mm_slli_epi32(ewitab,2);
 648             ewtabF           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 649             ewtabD           = _mm_setzero_pd();
 650             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 651             ewtabV           = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
 652             ewtabFn          = _mm_setzero_pd();
 653             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 654             felec            = _mm_macc_pd(eweps,ewtabD,ewtabF);
 655             velec            = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
 656             velec            = _mm_mul_pd(qq00,_mm_sub_pd(rinv00,velec));
 657             felec            = _mm_mul_pd(_mm_mul_pd(qq00,rinv00),_mm_sub_pd(rinvsq00,felec));
 658
 659             d                = _mm_sub_pd(r00,rswitch);
 660             d                = _mm_max_pd(d,_mm_setzero_pd());
 661             d2               = _mm_mul_pd(d,d);
 662             sw               = _mm_add_pd(one,_mm_mul_pd(d2,_mm_mul_pd(d,_mm_macc_pd(d,_mm_macc_pd(d,swV5,swV4),swV3))));
 663
 664             dsw              = _mm_mul_pd(d2,_mm_macc_pd(d,_mm_macc_pd(d,swF4,swF3),swF2));
 665
 666             /* Evaluate switch function */
 667             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
 668             felec            = _mm_msub_pd( felec,sw , _mm_mul_pd(rinv00,_mm_mul_pd(velec,dsw)) );
 669             cutoff_mask      = _mm_cmplt_pd(rsq00,rcutoff2);
 670
 671             fscal            = felec;
 672
 673             fscal            = _mm_and_pd(fscal,cutoff_mask);
 674
 675             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 676
 677             /* Update vectorial force */
 678             fix0             = _mm_macc_pd(dx00,fscal,fix0);
 679             fiy0             = _mm_macc_pd(dy00,fscal,fiy0);
 680             fiz0             = _mm_macc_pd(dz00,fscal,fiz0);
 681
 682             gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,
 683                                                    _mm_mul_pd(dx00,fscal),
 684                                                    _mm_mul_pd(dy00,fscal),
 685                                                    _mm_mul_pd(dz00,fscal));
 686
 687             }
 688
 689             /* Inner loop uses 65 flops */
 690         }
 691
 692         /* End of innermost loop */
 693
 694         gmx_mm_update_iforce_1atom_swizzle_pd(fix0,fiy0,fiz0,
 695                                               f+i_coord_offset,fshift+i_shift_offset);
 696
 697         /* Increment number of inner iterations */
 698         inneriter                  += j_index_end - j_index_start;
 699
 700         /* Outer loop uses 7 flops */
 701     }
 702
 703     /* Increment number of outer iterations */
 704     outeriter        += nri;
 705
 706     /* Update outer/inner flops */
 707
 708     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_F,outeriter*7 + inneriter*65);
 709 }