src/gromacs/gmxlib/nonbonded/nb_kernel_avx_128_fma_double/nb_kernel_ElecEwSh_VdwLJSh_GeomP1P1_avx_128_fma_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 avx_128_fma_double kernel generator.
  37  */
  38 #include "gmxpre.h"
  39
  40 #include "config.h"
  41
  42 #include <math.h>
  43
  44 #include "../nb_kernel.h"
  45 #include "gromacs/legacyheaders/types/simple.h"
  46 #include "gromacs/math/vec.h"
  47 #include "gromacs/legacyheaders/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_ElecEwSh_VdwLJSh_GeomP1P1_VF_avx_128_fma_double
  54  * Electrostatics interaction: Ewald
  55  * VdW interaction:            LennardJones
  56  * Geometry:                   Particle-Particle
  57  * Calculate force/pot:        PotentialAndForce
  58  */
  59 void
  60 nb_kernel_ElecEwSh_VdwLJSh_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     int              nvdwtype;
  90     __m128d          rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
  91     int              *vdwtype;
  92     real             *vdwparam;
  93     __m128d          one_sixth   = _mm_set1_pd(1.0/6.0);
  94     __m128d          one_twelfth = _mm_set1_pd(1.0/12.0);
  95     __m128i          ewitab;
  96     __m128d          ewtabscale,eweps,twoeweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
  97     real             *ewtab;
  98     __m128d          dummy_mask,cutoff_mask;
  99     __m128d          signbit   = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
 100     __m128d          one     = _mm_set1_pd(1.0);
 101     __m128d          two     = _mm_set1_pd(2.0);
 102     x                = xx[0];
 103     f                = ff[0];
 104
 105     nri              = nlist->nri;
 106     iinr             = nlist->iinr;
 107     jindex           = nlist->jindex;
 108     jjnr             = nlist->jjnr;
 109     shiftidx         = nlist->shift;
 110     gid              = nlist->gid;
 111     shiftvec         = fr->shift_vec[0];
 112     fshift           = fr->fshift[0];
 113     facel            = _mm_set1_pd(fr->epsfac);
 114     charge           = mdatoms->chargeA;
 115     nvdwtype         = fr->ntype;
 116     vdwparam         = fr->nbfp;
 117     vdwtype          = mdatoms->typeA;
 118
 119     sh_ewald         = _mm_set1_pd(fr->ic->sh_ewald);
 120     ewtab            = fr->ic->tabq_coul_FDV0;
 121     ewtabscale       = _mm_set1_pd(fr->ic->tabq_scale);
 122     ewtabhalfspace   = _mm_set1_pd(0.5/fr->ic->tabq_scale);
 123
 124     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 125     rcutoff_scalar   = fr->rcoulomb;
 126     rcutoff          = _mm_set1_pd(rcutoff_scalar);
 127     rcutoff2         = _mm_mul_pd(rcutoff,rcutoff);
 128
 129     sh_vdw_invrcut6  = _mm_set1_pd(fr->ic->sh_invrc6);
 130     rvdw             = _mm_set1_pd(fr->rvdw);
 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         vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 164
 165         /* Reset potential sums */
 166         velecsum         = _mm_setzero_pd();
 167         vvdwsum          = _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             vdwjidx0A        = 2*vdwtype[jnrA+0];
 198             vdwjidx0B        = 2*vdwtype[jnrB+0];
 199
 200             /**************************
 201              * CALCULATE INTERACTIONS *
 202              **************************/
 203
 204             if (gmx_mm_any_lt(rsq00,rcutoff2))
 205             {
 206
 207             r00              = _mm_mul_pd(rsq00,rinv00);
 208
 209             /* Compute parameters for interactions between i and j atoms */
 210             qq00             = _mm_mul_pd(iq0,jq0);
 211             gmx_mm_load_2pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,
 212                                          vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
 213
 214             /* EWALD ELECTROSTATICS */
 215
 216             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 217             ewrt             = _mm_mul_pd(r00,ewtabscale);
 218             ewitab           = _mm_cvttpd_epi32(ewrt);
 219 #ifdef __XOP__
 220             eweps            = _mm_frcz_pd(ewrt);
 221 #else
 222             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 223 #endif
 224             twoeweps         = _mm_add_pd(eweps,eweps);
 225             ewitab           = _mm_slli_epi32(ewitab,2);
 226             ewtabF           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 227             ewtabD           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
 228             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 229             ewtabV           = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
 230             ewtabFn          = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,1) +2);
 231             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 232             felec            = _mm_macc_pd(eweps,ewtabD,ewtabF);
 233             velec            = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
 234             velec            = _mm_mul_pd(qq00,_mm_sub_pd(_mm_sub_pd(rinv00,sh_ewald),velec));
 235             felec            = _mm_mul_pd(_mm_mul_pd(qq00,rinv00),_mm_sub_pd(rinvsq00,felec));
 236
 237             /* LENNARD-JONES DISPERSION/REPULSION */
 238
 239             rinvsix          = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 240             vvdw6            = _mm_mul_pd(c6_00,rinvsix);
 241             vvdw12           = _mm_mul_pd(c12_00,_mm_mul_pd(rinvsix,rinvsix));
 242             vvdw             = _mm_msub_pd(_mm_nmacc_pd(c12_00,_mm_mul_pd(sh_vdw_invrcut6,sh_vdw_invrcut6),vvdw12),one_twelfth,
 243                                            _mm_mul_pd(_mm_nmacc_pd( c6_00,sh_vdw_invrcut6,vvdw6),one_sixth));
 244             fvdw             = _mm_mul_pd(_mm_sub_pd(vvdw12,vvdw6),rinvsq00);
 245
 246             cutoff_mask      = _mm_cmplt_pd(rsq00,rcutoff2);
 247
 248             /* Update potential sum for this i atom from the interaction with this j atom. */
 249             velec            = _mm_and_pd(velec,cutoff_mask);
 250             velecsum         = _mm_add_pd(velecsum,velec);
 251             vvdw             = _mm_and_pd(vvdw,cutoff_mask);
 252             vvdwsum          = _mm_add_pd(vvdwsum,vvdw);
 253
 254             fscal            = _mm_add_pd(felec,fvdw);
 255
 256             fscal            = _mm_and_pd(fscal,cutoff_mask);
 257
 258             /* Update vectorial force */
 259             fix0             = _mm_macc_pd(dx00,fscal,fix0);
 260             fiy0             = _mm_macc_pd(dy00,fscal,fiy0);
 261             fiz0             = _mm_macc_pd(dz00,fscal,fiz0);
 262
 263             gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,
 264                                                    _mm_mul_pd(dx00,fscal),
 265                                                    _mm_mul_pd(dy00,fscal),
 266                                                    _mm_mul_pd(dz00,fscal));
 267
 268             }
 269
 270             /* Inner loop uses 67 flops */
 271         }
 272
 273         if(jidx<j_index_end)
 274         {
 275
 276             jnrA             = jjnr[jidx];
 277             j_coord_offsetA  = DIM*jnrA;
 278
 279             /* load j atom coordinates */
 280             gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
 281                                               &jx0,&jy0,&jz0);
 282
 283             /* Calculate displacement vector */
 284             dx00             = _mm_sub_pd(ix0,jx0);
 285             dy00             = _mm_sub_pd(iy0,jy0);
 286             dz00             = _mm_sub_pd(iz0,jz0);
 287
 288             /* Calculate squared distance and things based on it */
 289             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 290
 291             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 292
 293             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 294
 295             /* Load parameters for j particles */
 296             jq0              = _mm_load_sd(charge+jnrA+0);
 297             vdwjidx0A        = 2*vdwtype[jnrA+0];
 298
 299             /**************************
 300              * CALCULATE INTERACTIONS *
 301              **************************/
 302
 303             if (gmx_mm_any_lt(rsq00,rcutoff2))
 304             {
 305
 306             r00              = _mm_mul_pd(rsq00,rinv00);
 307
 308             /* Compute parameters for interactions between i and j atoms */
 309             qq00             = _mm_mul_pd(iq0,jq0);
 310             gmx_mm_load_1pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,&c6_00,&c12_00);
 311
 312             /* EWALD ELECTROSTATICS */
 313
 314             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 315             ewrt             = _mm_mul_pd(r00,ewtabscale);
 316             ewitab           = _mm_cvttpd_epi32(ewrt);
 317 #ifdef __XOP__
 318             eweps            = _mm_frcz_pd(ewrt);
 319 #else
 320             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 321 #endif
 322             twoeweps         = _mm_add_pd(eweps,eweps);
 323             ewitab           = _mm_slli_epi32(ewitab,2);
 324             ewtabF           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 325             ewtabD           = _mm_setzero_pd();
 326             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 327             ewtabV           = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
 328             ewtabFn          = _mm_setzero_pd();
 329             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 330             felec            = _mm_macc_pd(eweps,ewtabD,ewtabF);
 331             velec            = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
 332             velec            = _mm_mul_pd(qq00,_mm_sub_pd(_mm_sub_pd(rinv00,sh_ewald),velec));
 333             felec            = _mm_mul_pd(_mm_mul_pd(qq00,rinv00),_mm_sub_pd(rinvsq00,felec));
 334
 335             /* LENNARD-JONES DISPERSION/REPULSION */
 336
 337             rinvsix          = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 338             vvdw6            = _mm_mul_pd(c6_00,rinvsix);
 339             vvdw12           = _mm_mul_pd(c12_00,_mm_mul_pd(rinvsix,rinvsix));
 340             vvdw             = _mm_msub_pd(_mm_nmacc_pd(c12_00,_mm_mul_pd(sh_vdw_invrcut6,sh_vdw_invrcut6),vvdw12),one_twelfth,
 341                                            _mm_mul_pd(_mm_nmacc_pd( c6_00,sh_vdw_invrcut6,vvdw6),one_sixth));
 342             fvdw             = _mm_mul_pd(_mm_sub_pd(vvdw12,vvdw6),rinvsq00);
 343
 344             cutoff_mask      = _mm_cmplt_pd(rsq00,rcutoff2);
 345
 346             /* Update potential sum for this i atom from the interaction with this j atom. */
 347             velec            = _mm_and_pd(velec,cutoff_mask);
 348             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 349             velecsum         = _mm_add_pd(velecsum,velec);
 350             vvdw             = _mm_and_pd(vvdw,cutoff_mask);
 351             vvdw             = _mm_unpacklo_pd(vvdw,_mm_setzero_pd());
 352             vvdwsum          = _mm_add_pd(vvdwsum,vvdw);
 353
 354             fscal            = _mm_add_pd(felec,fvdw);
 355
 356             fscal            = _mm_and_pd(fscal,cutoff_mask);
 357
 358             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 359
 360             /* Update vectorial force */
 361             fix0             = _mm_macc_pd(dx00,fscal,fix0);
 362             fiy0             = _mm_macc_pd(dy00,fscal,fiy0);
 363             fiz0             = _mm_macc_pd(dz00,fscal,fiz0);
 364
 365             gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,
 366                                                    _mm_mul_pd(dx00,fscal),
 367                                                    _mm_mul_pd(dy00,fscal),
 368                                                    _mm_mul_pd(dz00,fscal));
 369
 370             }
 371
 372             /* Inner loop uses 67 flops */
 373         }
 374
 375         /* End of innermost loop */
 376
 377         gmx_mm_update_iforce_1atom_swizzle_pd(fix0,fiy0,fiz0,
 378                                               f+i_coord_offset,fshift+i_shift_offset);
 379
 380         ggid                        = gid[iidx];
 381         /* Update potential energies */
 382         gmx_mm_update_1pot_pd(velecsum,kernel_data->energygrp_elec+ggid);
 383         gmx_mm_update_1pot_pd(vvdwsum,kernel_data->energygrp_vdw+ggid);
 384
 385         /* Increment number of inner iterations */
 386         inneriter                  += j_index_end - j_index_start;
 387
 388         /* Outer loop uses 9 flops */
 389     }
 390
 391     /* Increment number of outer iterations */
 392     outeriter        += nri;
 393
 394     /* Update outer/inner flops */
 395
 396     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_VF,outeriter*9 + inneriter*67);
 397 }
 398 /*
 399  * Gromacs nonbonded kernel:   nb_kernel_ElecEwSh_VdwLJSh_GeomP1P1_F_avx_128_fma_double
 400  * Electrostatics interaction: Ewald
 401  * VdW interaction:            LennardJones
 402  * Geometry:                   Particle-Particle
 403  * Calculate force/pot:        Force
 404  */
 405 void
 406 nb_kernel_ElecEwSh_VdwLJSh_GeomP1P1_F_avx_128_fma_double
 407                     (t_nblist                    * gmx_restrict       nlist,
 408                      rvec                        * gmx_restrict          xx,
 409                      rvec                        * gmx_restrict          ff,
 410                      t_forcerec                  * gmx_restrict          fr,
 411                      t_mdatoms                   * gmx_restrict     mdatoms,
 412                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 413                      t_nrnb                      * gmx_restrict        nrnb)
 414 {
 415     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 416      * just 0 for non-waters.
 417      * Suffixes A,B refer to j loop unrolling done with SSE double precision, e.g. for the two different
 418      * jnr indices corresponding to data put in the four positions in the SIMD register.
 419      */
 420     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 421     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 422     int              jnrA,jnrB;
 423     int              j_coord_offsetA,j_coord_offsetB;
 424     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 425     real             rcutoff_scalar;
 426     real             *shiftvec,*fshift,*x,*f;
 427     __m128d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 428     int              vdwioffset0;
 429     __m128d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 430     int              vdwjidx0A,vdwjidx0B;
 431     __m128d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 432     __m128d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 433     __m128d          velec,felec,velecsum,facel,crf,krf,krf2;
 434     real             *charge;
 435     int              nvdwtype;
 436     __m128d          rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 437     int              *vdwtype;
 438     real             *vdwparam;
 439     __m128d          one_sixth   = _mm_set1_pd(1.0/6.0);
 440     __m128d          one_twelfth = _mm_set1_pd(1.0/12.0);
 441     __m128i          ewitab;
 442     __m128d          ewtabscale,eweps,twoeweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
 443     real             *ewtab;
 444     __m128d          dummy_mask,cutoff_mask;
 445     __m128d          signbit   = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
 446     __m128d          one     = _mm_set1_pd(1.0);
 447     __m128d          two     = _mm_set1_pd(2.0);
 448     x                = xx[0];
 449     f                = ff[0];
 450
 451     nri              = nlist->nri;
 452     iinr             = nlist->iinr;
 453     jindex           = nlist->jindex;
 454     jjnr             = nlist->jjnr;
 455     shiftidx         = nlist->shift;
 456     gid              = nlist->gid;
 457     shiftvec         = fr->shift_vec[0];
 458     fshift           = fr->fshift[0];
 459     facel            = _mm_set1_pd(fr->epsfac);
 460     charge           = mdatoms->chargeA;
 461     nvdwtype         = fr->ntype;
 462     vdwparam         = fr->nbfp;
 463     vdwtype          = mdatoms->typeA;
 464
 465     sh_ewald         = _mm_set1_pd(fr->ic->sh_ewald);
 466     ewtab            = fr->ic->tabq_coul_F;
 467     ewtabscale       = _mm_set1_pd(fr->ic->tabq_scale);
 468     ewtabhalfspace   = _mm_set1_pd(0.5/fr->ic->tabq_scale);
 469
 470     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 471     rcutoff_scalar   = fr->rcoulomb;
 472     rcutoff          = _mm_set1_pd(rcutoff_scalar);
 473     rcutoff2         = _mm_mul_pd(rcutoff,rcutoff);
 474
 475     sh_vdw_invrcut6  = _mm_set1_pd(fr->ic->sh_invrc6);
 476     rvdw             = _mm_set1_pd(fr->rvdw);
 477
 478     /* Avoid stupid compiler warnings */
 479     jnrA = jnrB = 0;
 480     j_coord_offsetA = 0;
 481     j_coord_offsetB = 0;
 482
 483     outeriter        = 0;
 484     inneriter        = 0;
 485
 486     /* Start outer loop over neighborlists */
 487     for(iidx=0; iidx<nri; iidx++)
 488     {
 489         /* Load shift vector for this list */
 490         i_shift_offset   = DIM*shiftidx[iidx];
 491
 492         /* Load limits for loop over neighbors */
 493         j_index_start    = jindex[iidx];
 494         j_index_end      = jindex[iidx+1];
 495
 496         /* Get outer coordinate index */
 497         inr              = iinr[iidx];
 498         i_coord_offset   = DIM*inr;
 499
 500         /* Load i particle coords and add shift vector */
 501         gmx_mm_load_shift_and_1rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 502
 503         fix0             = _mm_setzero_pd();
 504         fiy0             = _mm_setzero_pd();
 505         fiz0             = _mm_setzero_pd();
 506
 507         /* Load parameters for i particles */
 508         iq0              = _mm_mul_pd(facel,_mm_load1_pd(charge+inr+0));
 509         vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 510
 511         /* Start inner kernel loop */
 512         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 513         {
 514
 515             /* Get j neighbor index, and coordinate index */
 516             jnrA             = jjnr[jidx];
 517             jnrB             = jjnr[jidx+1];
 518             j_coord_offsetA  = DIM*jnrA;
 519             j_coord_offsetB  = DIM*jnrB;
 520
 521             /* load j atom coordinates */
 522             gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 523                                               &jx0,&jy0,&jz0);
 524
 525             /* Calculate displacement vector */
 526             dx00             = _mm_sub_pd(ix0,jx0);
 527             dy00             = _mm_sub_pd(iy0,jy0);
 528             dz00             = _mm_sub_pd(iz0,jz0);
 529
 530             /* Calculate squared distance and things based on it */
 531             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 532
 533             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 534
 535             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 536
 537             /* Load parameters for j particles */
 538             jq0              = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
 539             vdwjidx0A        = 2*vdwtype[jnrA+0];
 540             vdwjidx0B        = 2*vdwtype[jnrB+0];
 541
 542             /**************************
 543              * CALCULATE INTERACTIONS *
 544              **************************/
 545
 546             if (gmx_mm_any_lt(rsq00,rcutoff2))
 547             {
 548
 549             r00              = _mm_mul_pd(rsq00,rinv00);
 550
 551             /* Compute parameters for interactions between i and j atoms */
 552             qq00             = _mm_mul_pd(iq0,jq0);
 553             gmx_mm_load_2pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,
 554                                          vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
 555
 556             /* EWALD ELECTROSTATICS */
 557
 558             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 559             ewrt             = _mm_mul_pd(r00,ewtabscale);
 560             ewitab           = _mm_cvttpd_epi32(ewrt);
 561 #ifdef __XOP__
 562             eweps            = _mm_frcz_pd(ewrt);
 563 #else
 564             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 565 #endif
 566             twoeweps         = _mm_add_pd(eweps,eweps);
 567             gmx_mm_load_2pair_swizzle_pd(ewtab+_mm_extract_epi32(ewitab,0),ewtab+_mm_extract_epi32(ewitab,1),
 568                                          &ewtabF,&ewtabFn);
 569             felec            = _mm_macc_pd(eweps,ewtabFn,_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF));
 570             felec            = _mm_mul_pd(_mm_mul_pd(qq00,rinv00),_mm_sub_pd(rinvsq00,felec));
 571
 572             /* LENNARD-JONES DISPERSION/REPULSION */
 573
 574             rinvsix          = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 575             fvdw             = _mm_mul_pd(_mm_msub_pd(c12_00,rinvsix,c6_00),_mm_mul_pd(rinvsix,rinvsq00));
 576
 577             cutoff_mask      = _mm_cmplt_pd(rsq00,rcutoff2);
 578
 579             fscal            = _mm_add_pd(felec,fvdw);
 580
 581             fscal            = _mm_and_pd(fscal,cutoff_mask);
 582
 583             /* Update vectorial force */
 584             fix0             = _mm_macc_pd(dx00,fscal,fix0);
 585             fiy0             = _mm_macc_pd(dy00,fscal,fiy0);
 586             fiz0             = _mm_macc_pd(dz00,fscal,fiz0);
 587
 588             gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,
 589                                                    _mm_mul_pd(dx00,fscal),
 590                                                    _mm_mul_pd(dy00,fscal),
 591                                                    _mm_mul_pd(dz00,fscal));
 592
 593             }
 594
 595             /* Inner loop uses 49 flops */
 596         }
 597
 598         if(jidx<j_index_end)
 599         {
 600
 601             jnrA             = jjnr[jidx];
 602             j_coord_offsetA  = DIM*jnrA;
 603
 604             /* load j atom coordinates */
 605             gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
 606                                               &jx0,&jy0,&jz0);
 607
 608             /* Calculate displacement vector */
 609             dx00             = _mm_sub_pd(ix0,jx0);
 610             dy00             = _mm_sub_pd(iy0,jy0);
 611             dz00             = _mm_sub_pd(iz0,jz0);
 612
 613             /* Calculate squared distance and things based on it */
 614             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 615
 616             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 617
 618             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 619
 620             /* Load parameters for j particles */
 621             jq0              = _mm_load_sd(charge+jnrA+0);
 622             vdwjidx0A        = 2*vdwtype[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             gmx_mm_load_1pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,&c6_00,&c12_00);
 636
 637             /* EWALD ELECTROSTATICS */
 638
 639             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 640             ewrt             = _mm_mul_pd(r00,ewtabscale);
 641             ewitab           = _mm_cvttpd_epi32(ewrt);
 642 #ifdef __XOP__
 643             eweps            = _mm_frcz_pd(ewrt);
 644 #else
 645             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 646 #endif
 647             twoeweps         = _mm_add_pd(eweps,eweps);
 648             gmx_mm_load_1pair_swizzle_pd(ewtab+_mm_extract_epi32(ewitab,0),&ewtabF,&ewtabFn);
 649             felec            = _mm_macc_pd(eweps,ewtabFn,_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF));
 650             felec            = _mm_mul_pd(_mm_mul_pd(qq00,rinv00),_mm_sub_pd(rinvsq00,felec));
 651
 652             /* LENNARD-JONES DISPERSION/REPULSION */
 653
 654             rinvsix          = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 655             fvdw             = _mm_mul_pd(_mm_msub_pd(c12_00,rinvsix,c6_00),_mm_mul_pd(rinvsix,rinvsq00));
 656
 657             cutoff_mask      = _mm_cmplt_pd(rsq00,rcutoff2);
 658
 659             fscal            = _mm_add_pd(felec,fvdw);
 660
 661             fscal            = _mm_and_pd(fscal,cutoff_mask);
 662
 663             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 664
 665             /* Update vectorial force */
 666             fix0             = _mm_macc_pd(dx00,fscal,fix0);
 667             fiy0             = _mm_macc_pd(dy00,fscal,fiy0);
 668             fiz0             = _mm_macc_pd(dz00,fscal,fiz0);
 669
 670             gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,
 671                                                    _mm_mul_pd(dx00,fscal),
 672                                                    _mm_mul_pd(dy00,fscal),
 673                                                    _mm_mul_pd(dz00,fscal));
 674
 675             }
 676
 677             /* Inner loop uses 49 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_VDW_F,outeriter*7 + inneriter*49);
 697 }