src/gromacs/gmxlib/nonbonded/nb_kernel_sse2_double/nb_kernel_ElecEw_VdwLJEw_GeomP1P1_sse2_double.cpp

   1 /*
   2  * This file is part of the GROMACS molecular simulation package.
   3  *
   4  * Copyright (c) 2012,2013,2014,2015,2017,2018, by the GROMACS development team, led by
   5  * Mark Abraham, David van der Spoel, Berk Hess, and Erik Lindahl,
   6  * and including many others, as listed in the AUTHORS file in the
   7  * top-level source directory and at http://www.gromacs.org.
   8  *
   9  * GROMACS is free software; you can redistribute it and/or
  10  * modify it under the terms of the GNU Lesser General Public License
  11  * as published by the Free Software Foundation; either version 2.1
  12  * of the License, or (at your option) any later version.
  13  *
  14  * GROMACS is distributed in the hope that it will be useful,
  15  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  16  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  17  * Lesser General Public License for more details.
  18  *
  19  * You should have received a copy of the GNU Lesser General Public
  20  * License along with GROMACS; if not, see
  21  * http://www.gnu.org/licenses, or write to the Free Software Foundation,
  22  * Inc., 51 Franklin Street, Fifth Floor, Boston, MA  02110-1301  USA.
  23  *
  24  * If you want to redistribute modifications to GROMACS, please
  25  * consider that scientific software is very special. Version
  26  * control is crucial - bugs must be traceable. We will be happy to
  27  * consider code for inclusion in the official distribution, but
  28  * derived work must not be called official GROMACS. Details are found
  29  * in the README & COPYING files - if they are missing, get the
  30  * official version at http://www.gromacs.org.
  31  *
  32  * To help us fund GROMACS development, we humbly ask that you cite
  33  * the research papers on the package. Check out http://www.gromacs.org.
  34  */
  35 /*
  36  * Note: this file was generated by the GROMACS sse2_double kernel generator.
  37  */
  38 #include "gmxpre.h"
  39
  40 #include "config.h"
  41
  42 #include <math.h>
  43
  44 #include "../nb_kernel.h"
  45 #include "gromacs/gmxlib/nrnb.h"
  46
  47 #include "kernelutil_x86_sse2_double.h"
  48
  49 /*
  50  * Gromacs nonbonded kernel:   nb_kernel_ElecEw_VdwLJEw_GeomP1P1_VF_sse2_double
  51  * Electrostatics interaction: Ewald
  52  * VdW interaction:            LJEwald
  53  * Geometry:                   Particle-Particle
  54  * Calculate force/pot:        PotentialAndForce
  55  */
  56 void
  57 nb_kernel_ElecEw_VdwLJEw_GeomP1P1_VF_sse2_double
  58                     (t_nblist                    * gmx_restrict       nlist,
  59                      rvec                        * gmx_restrict          xx,
  60                      rvec                        * gmx_restrict          ff,
  61                      struct t_forcerec           * gmx_restrict          fr,
  62                      t_mdatoms                   * gmx_restrict     mdatoms,
  63                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
  64                      t_nrnb                      * gmx_restrict        nrnb)
  65 {
  66     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
  67      * just 0 for non-waters.
  68      * Suffixes A,B refer to j loop unrolling done with SSE double precision, e.g. for the two different
  69      * jnr indices corresponding to data put in the four positions in the SIMD register.
  70      */
  71     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
  72     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
  73     int              jnrA,jnrB;
  74     int              j_coord_offsetA,j_coord_offsetB;
  75     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
  76     real             rcutoff_scalar;
  77     real             *shiftvec,*fshift,*x,*f;
  78     __m128d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
  79     int              vdwioffset0;
  80     __m128d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
  81     int              vdwjidx0A,vdwjidx0B;
  82     __m128d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  83     __m128d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  84     __m128d          velec,felec,velecsum,facel,crf,krf,krf2;
  85     real             *charge;
  86     int              nvdwtype;
  87     __m128d          rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
  88     int              *vdwtype;
  89     real             *vdwparam;
  90     __m128d          one_sixth   = _mm_set1_pd(1.0/6.0);
  91     __m128d          one_twelfth = _mm_set1_pd(1.0/12.0);
  92     __m128d           c6grid_00;
  93     __m128d           ewclj,ewclj2,ewclj6,ewcljrsq,poly,exponent,f6A,f6B,sh_lj_ewald;
  94     real             *vdwgridparam;
  95     __m128d           one_half = _mm_set1_pd(0.5);
  96     __m128d           minus_one = _mm_set1_pd(-1.0);
  97     __m128i          ewitab;
  98     __m128d          ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
  99     real             *ewtab;
 100     __m128d          dummy_mask,cutoff_mask;
 101     __m128d          signbit   = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
 102     __m128d          one     = _mm_set1_pd(1.0);
 103     __m128d          two     = _mm_set1_pd(2.0);
 104     x                = xx[0];
 105     f                = ff[0];
 106
 107     nri              = nlist->nri;
 108     iinr             = nlist->iinr;
 109     jindex           = nlist->jindex;
 110     jjnr             = nlist->jjnr;
 111     shiftidx         = nlist->shift;
 112     gid              = nlist->gid;
 113     shiftvec         = fr->shift_vec[0];
 114     fshift           = fr->fshift[0];
 115     facel            = _mm_set1_pd(fr->ic->epsfac);
 116     charge           = mdatoms->chargeA;
 117     nvdwtype         = fr->ntype;
 118     vdwparam         = fr->nbfp;
 119     vdwtype          = mdatoms->typeA;
 120     vdwgridparam     = fr->ljpme_c6grid;
 121     sh_lj_ewald      = _mm_set1_pd(fr->ic->sh_lj_ewald);
 122     ewclj            = _mm_set1_pd(fr->ic->ewaldcoeff_lj);
 123     ewclj2           = _mm_mul_pd(minus_one,_mm_mul_pd(ewclj,ewclj));
 124
 125     sh_ewald         = _mm_set1_pd(fr->ic->sh_ewald);
 126     ewtab            = fr->ic->tabq_coul_FDV0;
 127     ewtabscale       = _mm_set1_pd(fr->ic->tabq_scale);
 128     ewtabhalfspace   = _mm_set1_pd(0.5/fr->ic->tabq_scale);
 129
 130     /* Avoid stupid compiler warnings */
 131     jnrA = jnrB = 0;
 132     j_coord_offsetA = 0;
 133     j_coord_offsetB = 0;
 134
 135     outeriter        = 0;
 136     inneriter        = 0;
 137
 138     /* Start outer loop over neighborlists */
 139     for(iidx=0; iidx<nri; iidx++)
 140     {
 141         /* Load shift vector for this list */
 142         i_shift_offset   = DIM*shiftidx[iidx];
 143
 144         /* Load limits for loop over neighbors */
 145         j_index_start    = jindex[iidx];
 146         j_index_end      = jindex[iidx+1];
 147
 148         /* Get outer coordinate index */
 149         inr              = iinr[iidx];
 150         i_coord_offset   = DIM*inr;
 151
 152         /* Load i particle coords and add shift vector */
 153         gmx_mm_load_shift_and_1rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 154
 155         fix0             = _mm_setzero_pd();
 156         fiy0             = _mm_setzero_pd();
 157         fiz0             = _mm_setzero_pd();
 158
 159         /* Load parameters for i particles */
 160         iq0              = _mm_mul_pd(facel,_mm_load1_pd(charge+inr+0));
 161         vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 162
 163         /* Reset potential sums */
 164         velecsum         = _mm_setzero_pd();
 165         vvdwsum          = _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           = sse2_invsqrt_d(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             vdwjidx0A        = 2*vdwtype[jnrA+0];
 196             vdwjidx0B        = 2*vdwtype[jnrB+0];
 197
 198             /**************************
 199              * CALCULATE INTERACTIONS *
 200              **************************/
 201
 202             r00              = _mm_mul_pd(rsq00,rinv00);
 203
 204             /* Compute parameters for interactions between i and j atoms */
 205             qq00             = _mm_mul_pd(iq0,jq0);
 206             gmx_mm_load_2pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,
 207                                          vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
 208
 209             c6grid_00       = gmx_mm_load_2real_swizzle_pd(vdwgridparam+vdwioffset0+vdwjidx0A,
 210                                                                vdwgridparam+vdwioffset0+vdwjidx0B);
 211
 212             /* EWALD ELECTROSTATICS */
 213
 214             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 215             ewrt             = _mm_mul_pd(r00,ewtabscale);
 216             ewitab           = _mm_cvttpd_epi32(ewrt);
 217             eweps            = _mm_sub_pd(ewrt,_mm_cvtepi32_pd(ewitab));
 218             ewitab           = _mm_slli_epi32(ewitab,2);
 219             ewtabF           = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 220             ewtabD           = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,1) );
 221             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 222             ewtabV           = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,0) +2);
 223             ewtabFn          = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,1) +2);
 224             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 225             felec            = _mm_add_pd(ewtabF,_mm_mul_pd(eweps,ewtabD));
 226             velec            = _mm_sub_pd(ewtabV,_mm_mul_pd(_mm_mul_pd(ewtabhalfspace,eweps),_mm_add_pd(ewtabF,felec)));
 227             velec            = _mm_mul_pd(qq00,_mm_sub_pd(rinv00,velec));
 228             felec            = _mm_mul_pd(_mm_mul_pd(qq00,rinv00),_mm_sub_pd(rinvsq00,felec));
 229
 230             /* Analytical LJ-PME */
 231             rinvsix          = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 232             ewcljrsq         = _mm_mul_pd(ewclj2,rsq00);
 233             ewclj6           = _mm_mul_pd(ewclj2,_mm_mul_pd(ewclj2,ewclj2));
 234             exponent         = sse2_exp_d(ewcljrsq);
 235             /* poly = exp(-(beta*r)^2) * (1 + (beta*r)^2 + (beta*r)^4 /2) */
 236             poly             = _mm_mul_pd(exponent,_mm_add_pd(_mm_sub_pd(one,ewcljrsq),_mm_mul_pd(_mm_mul_pd(ewcljrsq,ewcljrsq),one_half)));
 237             /* vvdw6 = [C6 - C6grid * (1-poly)]/r6 */
 238             vvdw6            = _mm_mul_pd(_mm_sub_pd(c6_00,_mm_mul_pd(c6grid_00,_mm_sub_pd(one,poly))),rinvsix);
 239             vvdw12           = _mm_mul_pd(c12_00,_mm_mul_pd(rinvsix,rinvsix));
 240             vvdw             = _mm_sub_pd(_mm_mul_pd(vvdw12,one_twelfth),_mm_mul_pd(vvdw6,one_sixth));
 241             /* fvdw = vvdw12/r - (vvdw6/r + (C6grid * exponent * beta^6)/r) */
 242             fvdw             = _mm_mul_pd(_mm_sub_pd(vvdw12,_mm_sub_pd(vvdw6,_mm_mul_pd(_mm_mul_pd(c6grid_00,one_sixth),_mm_mul_pd(exponent,ewclj6)))),rinvsq00);
 243
 244             /* Update potential sum for this i atom from the interaction with this j atom. */
 245             velecsum         = _mm_add_pd(velecsum,velec);
 246             vvdwsum          = _mm_add_pd(vvdwsum,vvdw);
 247
 248             fscal            = _mm_add_pd(felec,fvdw);
 249
 250             /* Calculate temporary vectorial force */
 251             tx               = _mm_mul_pd(fscal,dx00);
 252             ty               = _mm_mul_pd(fscal,dy00);
 253             tz               = _mm_mul_pd(fscal,dz00);
 254
 255             /* Update vectorial force */
 256             fix0             = _mm_add_pd(fix0,tx);
 257             fiy0             = _mm_add_pd(fiy0,ty);
 258             fiz0             = _mm_add_pd(fiz0,tz);
 259
 260             gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,tx,ty,tz);
 261
 262             /* Inner loop uses 69 flops */
 263         }
 264
 265         if(jidx<j_index_end)
 266         {
 267
 268             jnrA             = jjnr[jidx];
 269             j_coord_offsetA  = DIM*jnrA;
 270
 271             /* load j atom coordinates */
 272             gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
 273                                               &jx0,&jy0,&jz0);
 274
 275             /* Calculate displacement vector */
 276             dx00             = _mm_sub_pd(ix0,jx0);
 277             dy00             = _mm_sub_pd(iy0,jy0);
 278             dz00             = _mm_sub_pd(iz0,jz0);
 279
 280             /* Calculate squared distance and things based on it */
 281             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 282
 283             rinv00           = sse2_invsqrt_d(rsq00);
 284
 285             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 286
 287             /* Load parameters for j particles */
 288             jq0              = _mm_load_sd(charge+jnrA+0);
 289             vdwjidx0A        = 2*vdwtype[jnrA+0];
 290
 291             /**************************
 292              * CALCULATE INTERACTIONS *
 293              **************************/
 294
 295             r00              = _mm_mul_pd(rsq00,rinv00);
 296
 297             /* Compute parameters for interactions between i and j atoms */
 298             qq00             = _mm_mul_pd(iq0,jq0);
 299             gmx_mm_load_1pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,&c6_00,&c12_00);
 300
 301             c6grid_00       = gmx_mm_load_1real_pd(vdwgridparam+vdwioffset0+vdwjidx0A);
 302
 303             /* EWALD ELECTROSTATICS */
 304
 305             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 306             ewrt             = _mm_mul_pd(r00,ewtabscale);
 307             ewitab           = _mm_cvttpd_epi32(ewrt);
 308             eweps            = _mm_sub_pd(ewrt,_mm_cvtepi32_pd(ewitab));
 309             ewitab           = _mm_slli_epi32(ewitab,2);
 310             ewtabF           = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 311             ewtabD           = _mm_setzero_pd();
 312             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 313             ewtabV           = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,0) +2);
 314             ewtabFn          = _mm_setzero_pd();
 315             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 316             felec            = _mm_add_pd(ewtabF,_mm_mul_pd(eweps,ewtabD));
 317             velec            = _mm_sub_pd(ewtabV,_mm_mul_pd(_mm_mul_pd(ewtabhalfspace,eweps),_mm_add_pd(ewtabF,felec)));
 318             velec            = _mm_mul_pd(qq00,_mm_sub_pd(rinv00,velec));
 319             felec            = _mm_mul_pd(_mm_mul_pd(qq00,rinv00),_mm_sub_pd(rinvsq00,felec));
 320
 321             /* Analytical LJ-PME */
 322             rinvsix          = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 323             ewcljrsq         = _mm_mul_pd(ewclj2,rsq00);
 324             ewclj6           = _mm_mul_pd(ewclj2,_mm_mul_pd(ewclj2,ewclj2));
 325             exponent         = sse2_exp_d(ewcljrsq);
 326             /* poly = exp(-(beta*r)^2) * (1 + (beta*r)^2 + (beta*r)^4 /2) */
 327             poly             = _mm_mul_pd(exponent,_mm_add_pd(_mm_sub_pd(one,ewcljrsq),_mm_mul_pd(_mm_mul_pd(ewcljrsq,ewcljrsq),one_half)));
 328             /* vvdw6 = [C6 - C6grid * (1-poly)]/r6 */
 329             vvdw6            = _mm_mul_pd(_mm_sub_pd(c6_00,_mm_mul_pd(c6grid_00,_mm_sub_pd(one,poly))),rinvsix);
 330             vvdw12           = _mm_mul_pd(c12_00,_mm_mul_pd(rinvsix,rinvsix));
 331             vvdw             = _mm_sub_pd(_mm_mul_pd(vvdw12,one_twelfth),_mm_mul_pd(vvdw6,one_sixth));
 332             /* fvdw = vvdw12/r - (vvdw6/r + (C6grid * exponent * beta^6)/r) */
 333             fvdw             = _mm_mul_pd(_mm_sub_pd(vvdw12,_mm_sub_pd(vvdw6,_mm_mul_pd(_mm_mul_pd(c6grid_00,one_sixth),_mm_mul_pd(exponent,ewclj6)))),rinvsq00);
 334
 335             /* Update potential sum for this i atom from the interaction with this j atom. */
 336             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 337             velecsum         = _mm_add_pd(velecsum,velec);
 338             vvdw             = _mm_unpacklo_pd(vvdw,_mm_setzero_pd());
 339             vvdwsum          = _mm_add_pd(vvdwsum,vvdw);
 340
 341             fscal            = _mm_add_pd(felec,fvdw);
 342
 343             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 344
 345             /* Calculate temporary vectorial force */
 346             tx               = _mm_mul_pd(fscal,dx00);
 347             ty               = _mm_mul_pd(fscal,dy00);
 348             tz               = _mm_mul_pd(fscal,dz00);
 349
 350             /* Update vectorial force */
 351             fix0             = _mm_add_pd(fix0,tx);
 352             fiy0             = _mm_add_pd(fiy0,ty);
 353             fiz0             = _mm_add_pd(fiz0,tz);
 354
 355             gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,tx,ty,tz);
 356
 357             /* Inner loop uses 69 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         gmx_mm_update_1pot_pd(vvdwsum,kernel_data->energygrp_vdw+ggid);
 369
 370         /* Increment number of inner iterations */
 371         inneriter                  += j_index_end - j_index_start;
 372
 373         /* Outer loop uses 9 flops */
 374     }
 375
 376     /* Increment number of outer iterations */
 377     outeriter        += nri;
 378
 379     /* Update outer/inner flops */
 380
 381     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_VF,outeriter*9 + inneriter*69);
 382 }
 383 /*
 384  * Gromacs nonbonded kernel:   nb_kernel_ElecEw_VdwLJEw_GeomP1P1_F_sse2_double
 385  * Electrostatics interaction: Ewald
 386  * VdW interaction:            LJEwald
 387  * Geometry:                   Particle-Particle
 388  * Calculate force/pot:        Force
 389  */
 390 void
 391 nb_kernel_ElecEw_VdwLJEw_GeomP1P1_F_sse2_double
 392                     (t_nblist                    * gmx_restrict       nlist,
 393                      rvec                        * gmx_restrict          xx,
 394                      rvec                        * gmx_restrict          ff,
 395                      struct t_forcerec           * gmx_restrict          fr,
 396                      t_mdatoms                   * gmx_restrict     mdatoms,
 397                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 398                      t_nrnb                      * gmx_restrict        nrnb)
 399 {
 400     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 401      * just 0 for non-waters.
 402      * Suffixes A,B refer to j loop unrolling done with SSE double precision, e.g. for the two different
 403      * jnr indices corresponding to data put in the four positions in the SIMD register.
 404      */
 405     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 406     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 407     int              jnrA,jnrB;
 408     int              j_coord_offsetA,j_coord_offsetB;
 409     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 410     real             rcutoff_scalar;
 411     real             *shiftvec,*fshift,*x,*f;
 412     __m128d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 413     int              vdwioffset0;
 414     __m128d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 415     int              vdwjidx0A,vdwjidx0B;
 416     __m128d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 417     __m128d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 418     __m128d          velec,felec,velecsum,facel,crf,krf,krf2;
 419     real             *charge;
 420     int              nvdwtype;
 421     __m128d          rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 422     int              *vdwtype;
 423     real             *vdwparam;
 424     __m128d          one_sixth   = _mm_set1_pd(1.0/6.0);
 425     __m128d          one_twelfth = _mm_set1_pd(1.0/12.0);
 426     __m128d           c6grid_00;
 427     __m128d           ewclj,ewclj2,ewclj6,ewcljrsq,poly,exponent,f6A,f6B,sh_lj_ewald;
 428     real             *vdwgridparam;
 429     __m128d           one_half = _mm_set1_pd(0.5);
 430     __m128d           minus_one = _mm_set1_pd(-1.0);
 431     __m128i          ewitab;
 432     __m128d          ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
 433     real             *ewtab;
 434     __m128d          dummy_mask,cutoff_mask;
 435     __m128d          signbit   = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
 436     __m128d          one     = _mm_set1_pd(1.0);
 437     __m128d          two     = _mm_set1_pd(2.0);
 438     x                = xx[0];
 439     f                = ff[0];
 440
 441     nri              = nlist->nri;
 442     iinr             = nlist->iinr;
 443     jindex           = nlist->jindex;
 444     jjnr             = nlist->jjnr;
 445     shiftidx         = nlist->shift;
 446     gid              = nlist->gid;
 447     shiftvec         = fr->shift_vec[0];
 448     fshift           = fr->fshift[0];
 449     facel            = _mm_set1_pd(fr->ic->epsfac);
 450     charge           = mdatoms->chargeA;
 451     nvdwtype         = fr->ntype;
 452     vdwparam         = fr->nbfp;
 453     vdwtype          = mdatoms->typeA;
 454     vdwgridparam     = fr->ljpme_c6grid;
 455     sh_lj_ewald      = _mm_set1_pd(fr->ic->sh_lj_ewald);
 456     ewclj            = _mm_set1_pd(fr->ic->ewaldcoeff_lj);
 457     ewclj2           = _mm_mul_pd(minus_one,_mm_mul_pd(ewclj,ewclj));
 458
 459     sh_ewald         = _mm_set1_pd(fr->ic->sh_ewald);
 460     ewtab            = fr->ic->tabq_coul_F;
 461     ewtabscale       = _mm_set1_pd(fr->ic->tabq_scale);
 462     ewtabhalfspace   = _mm_set1_pd(0.5/fr->ic->tabq_scale);
 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         vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 496
 497         /* Start inner kernel loop */
 498         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 499         {
 500
 501             /* Get j neighbor index, and coordinate index */
 502             jnrA             = jjnr[jidx];
 503             jnrB             = jjnr[jidx+1];
 504             j_coord_offsetA  = DIM*jnrA;
 505             j_coord_offsetB  = DIM*jnrB;
 506
 507             /* load j atom coordinates */
 508             gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 509                                               &jx0,&jy0,&jz0);
 510
 511             /* Calculate displacement vector */
 512             dx00             = _mm_sub_pd(ix0,jx0);
 513             dy00             = _mm_sub_pd(iy0,jy0);
 514             dz00             = _mm_sub_pd(iz0,jz0);
 515
 516             /* Calculate squared distance and things based on it */
 517             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 518
 519             rinv00           = sse2_invsqrt_d(rsq00);
 520
 521             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 522
 523             /* Load parameters for j particles */
 524             jq0              = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
 525             vdwjidx0A        = 2*vdwtype[jnrA+0];
 526             vdwjidx0B        = 2*vdwtype[jnrB+0];
 527
 528             /**************************
 529              * CALCULATE INTERACTIONS *
 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             gmx_mm_load_2pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,
 537                                          vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
 538
 539             c6grid_00       = gmx_mm_load_2real_swizzle_pd(vdwgridparam+vdwioffset0+vdwjidx0A,
 540                                                                vdwgridparam+vdwioffset0+vdwjidx0B);
 541
 542             /* EWALD ELECTROSTATICS */
 543
 544             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 545             ewrt             = _mm_mul_pd(r00,ewtabscale);
 546             ewitab           = _mm_cvttpd_epi32(ewrt);
 547             eweps            = _mm_sub_pd(ewrt,_mm_cvtepi32_pd(ewitab));
 548             gmx_mm_load_2pair_swizzle_pd(ewtab+gmx_mm_extract_epi32(ewitab,0),ewtab+gmx_mm_extract_epi32(ewitab,1),
 549                                          &ewtabF,&ewtabFn);
 550             felec            = _mm_add_pd(_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF),_mm_mul_pd(eweps,ewtabFn));
 551             felec            = _mm_mul_pd(_mm_mul_pd(qq00,rinv00),_mm_sub_pd(rinvsq00,felec));
 552
 553             /* Analytical LJ-PME */
 554             rinvsix          = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 555             ewcljrsq         = _mm_mul_pd(ewclj2,rsq00);
 556             ewclj6           = _mm_mul_pd(ewclj2,_mm_mul_pd(ewclj2,ewclj2));
 557             exponent         = sse2_exp_d(ewcljrsq);
 558             /* poly = exp(-(beta*r)^2) * (1 + (beta*r)^2 + (beta*r)^4 /2) */
 559             poly             = _mm_mul_pd(exponent,_mm_add_pd(_mm_sub_pd(one,ewcljrsq),_mm_mul_pd(_mm_mul_pd(ewcljrsq,ewcljrsq),one_half)));
 560             /* f6A = 6 * C6grid * (1 - poly) */
 561             f6A              = _mm_mul_pd(c6grid_00,_mm_sub_pd(one,poly));
 562             /* f6B = C6grid * exponent * beta^6 */
 563             f6B              = _mm_mul_pd(_mm_mul_pd(c6grid_00,one_sixth),_mm_mul_pd(exponent,ewclj6));
 564             /* fvdw = 12*C12/r13 - ((6*C6 - f6A)/r6 + f6B)/r */
 565             fvdw              = _mm_mul_pd(_mm_add_pd(_mm_mul_pd(_mm_sub_pd(_mm_mul_pd(c12_00,rinvsix),_mm_sub_pd(c6_00,f6A)),rinvsix),f6B),rinvsq00);
 566
 567             fscal            = _mm_add_pd(felec,fvdw);
 568
 569             /* Calculate temporary vectorial force */
 570             tx               = _mm_mul_pd(fscal,dx00);
 571             ty               = _mm_mul_pd(fscal,dy00);
 572             tz               = _mm_mul_pd(fscal,dz00);
 573
 574             /* Update vectorial force */
 575             fix0             = _mm_add_pd(fix0,tx);
 576             fiy0             = _mm_add_pd(fiy0,ty);
 577             fiz0             = _mm_add_pd(fiz0,tz);
 578
 579             gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,tx,ty,tz);
 580
 581             /* Inner loop uses 59 flops */
 582         }
 583
 584         if(jidx<j_index_end)
 585         {
 586
 587             jnrA             = jjnr[jidx];
 588             j_coord_offsetA  = DIM*jnrA;
 589
 590             /* load j atom coordinates */
 591             gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
 592                                               &jx0,&jy0,&jz0);
 593
 594             /* Calculate displacement vector */
 595             dx00             = _mm_sub_pd(ix0,jx0);
 596             dy00             = _mm_sub_pd(iy0,jy0);
 597             dz00             = _mm_sub_pd(iz0,jz0);
 598
 599             /* Calculate squared distance and things based on it */
 600             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 601
 602             rinv00           = sse2_invsqrt_d(rsq00);
 603
 604             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 605
 606             /* Load parameters for j particles */
 607             jq0              = _mm_load_sd(charge+jnrA+0);
 608             vdwjidx0A        = 2*vdwtype[jnrA+0];
 609
 610             /**************************
 611              * CALCULATE INTERACTIONS *
 612              **************************/
 613
 614             r00              = _mm_mul_pd(rsq00,rinv00);
 615
 616             /* Compute parameters for interactions between i and j atoms */
 617             qq00             = _mm_mul_pd(iq0,jq0);
 618             gmx_mm_load_1pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,&c6_00,&c12_00);
 619
 620             c6grid_00       = gmx_mm_load_1real_pd(vdwgridparam+vdwioffset0+vdwjidx0A);
 621
 622             /* EWALD ELECTROSTATICS */
 623
 624             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 625             ewrt             = _mm_mul_pd(r00,ewtabscale);
 626             ewitab           = _mm_cvttpd_epi32(ewrt);
 627             eweps            = _mm_sub_pd(ewrt,_mm_cvtepi32_pd(ewitab));
 628             gmx_mm_load_1pair_swizzle_pd(ewtab+gmx_mm_extract_epi32(ewitab,0),&ewtabF,&ewtabFn);
 629             felec            = _mm_add_pd(_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF),_mm_mul_pd(eweps,ewtabFn));
 630             felec            = _mm_mul_pd(_mm_mul_pd(qq00,rinv00),_mm_sub_pd(rinvsq00,felec));
 631
 632             /* Analytical LJ-PME */
 633             rinvsix          = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 634             ewcljrsq         = _mm_mul_pd(ewclj2,rsq00);
 635             ewclj6           = _mm_mul_pd(ewclj2,_mm_mul_pd(ewclj2,ewclj2));
 636             exponent         = sse2_exp_d(ewcljrsq);
 637             /* poly = exp(-(beta*r)^2) * (1 + (beta*r)^2 + (beta*r)^4 /2) */
 638             poly             = _mm_mul_pd(exponent,_mm_add_pd(_mm_sub_pd(one,ewcljrsq),_mm_mul_pd(_mm_mul_pd(ewcljrsq,ewcljrsq),one_half)));
 639             /* f6A = 6 * C6grid * (1 - poly) */
 640             f6A              = _mm_mul_pd(c6grid_00,_mm_sub_pd(one,poly));
 641             /* f6B = C6grid * exponent * beta^6 */
 642             f6B              = _mm_mul_pd(_mm_mul_pd(c6grid_00,one_sixth),_mm_mul_pd(exponent,ewclj6));
 643             /* fvdw = 12*C12/r13 - ((6*C6 - f6A)/r6 + f6B)/r */
 644             fvdw              = _mm_mul_pd(_mm_add_pd(_mm_mul_pd(_mm_sub_pd(_mm_mul_pd(c12_00,rinvsix),_mm_sub_pd(c6_00,f6A)),rinvsix),f6B),rinvsq00);
 645
 646             fscal            = _mm_add_pd(felec,fvdw);
 647
 648             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 649
 650             /* Calculate temporary vectorial force */
 651             tx               = _mm_mul_pd(fscal,dx00);
 652             ty               = _mm_mul_pd(fscal,dy00);
 653             tz               = _mm_mul_pd(fscal,dz00);
 654
 655             /* Update vectorial force */
 656             fix0             = _mm_add_pd(fix0,tx);
 657             fiy0             = _mm_add_pd(fiy0,ty);
 658             fiz0             = _mm_add_pd(fiz0,tz);
 659
 660             gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,tx,ty,tz);
 661
 662             /* Inner loop uses 59 flops */
 663         }
 664
 665         /* End of innermost loop */
 666
 667         gmx_mm_update_iforce_1atom_swizzle_pd(fix0,fiy0,fiz0,
 668                                               f+i_coord_offset,fshift+i_shift_offset);
 669
 670         /* Increment number of inner iterations */
 671         inneriter                  += j_index_end - j_index_start;
 672
 673         /* Outer loop uses 7 flops */
 674     }
 675
 676     /* Increment number of outer iterations */
 677     outeriter        += nri;
 678
 679     /* Update outer/inner flops */
 680
 681     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_F,outeriter*7 + inneriter*59);
 682 }