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

   1 /*
   2  * This file is part of the GROMACS molecular simulation package.
   3  *
   4  * Copyright (c) 2012,2013,2014, by the GROMACS development team, led by
   5  * Mark Abraham, David van der Spoel, Berk Hess, and Erik Lindahl,
   6  * and including many others, as listed in the AUTHORS file in the
   7  * top-level source directory and at http://www.gromacs.org.
   8  *
   9  * GROMACS is free software; you can redistribute it and/or
  10  * modify it under the terms of the GNU Lesser General Public License
  11  * as published by the Free Software Foundation; either version 2.1
  12  * of the License, or (at your option) any later version.
  13  *
  14  * GROMACS is distributed in the hope that it will be useful,
  15  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  16  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  17  * Lesser General Public License for more details.
  18  *
  19  * You should have received a copy of the GNU Lesser General Public
  20  * License along with GROMACS; if not, see
  21  * http://www.gnu.org/licenses, or write to the Free Software Foundation,
  22  * Inc., 51 Franklin Street, Fifth Floor, Boston, MA  02110-1301  USA.
  23  *
  24  * If you want to redistribute modifications to GROMACS, please
  25  * consider that scientific software is very special. Version
  26  * control is crucial - bugs must be traceable. We will be happy to
  27  * consider code for inclusion in the official distribution, but
  28  * derived work must not be called official GROMACS. Details are found
  29  * in the README & COPYING files - if they are missing, get the
  30  * official version at http://www.gromacs.org.
  31  *
  32  * To help us fund GROMACS development, we humbly ask that you cite
  33  * the research papers on the package. Check out http://www.gromacs.org.
  34  */
  35 /*
  36  * Note: this file was generated by the GROMACS sse2_double kernel generator.
  37  */
  38 #include "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_sse2_double.h"
  50 #include "kernelutil_x86_sse2_double.h"
  51
  52 /*
  53  * Gromacs nonbonded kernel:   nb_kernel_ElecEw_VdwLJEw_GeomP1P1_VF_sse2_double
  54  * Electrostatics interaction: Ewald
  55  * VdW interaction:            LJEwald
  56  * Geometry:                   Particle-Particle
  57  * Calculate force/pot:        PotentialAndForce
  58  */
  59 void
  60 nb_kernel_ElecEw_VdwLJEw_GeomP1P1_VF_sse2_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     __m128d           c6grid_00;
  96     __m128d           ewclj,ewclj2,ewclj6,ewcljrsq,poly,exponent,f6A,f6B,sh_lj_ewald;
  97     real             *vdwgridparam;
  98     __m128d           one_half = _mm_set1_pd(0.5);
  99     __m128d           minus_one = _mm_set1_pd(-1.0);
 100     __m128i          ewitab;
 101     __m128d          ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
 102     real             *ewtab;
 103     __m128d          dummy_mask,cutoff_mask;
 104     __m128d          signbit   = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
 105     __m128d          one     = _mm_set1_pd(1.0);
 106     __m128d          two     = _mm_set1_pd(2.0);
 107     x                = xx[0];
 108     f                = ff[0];
 109
 110     nri              = nlist->nri;
 111     iinr             = nlist->iinr;
 112     jindex           = nlist->jindex;
 113     jjnr             = nlist->jjnr;
 114     shiftidx         = nlist->shift;
 115     gid              = nlist->gid;
 116     shiftvec         = fr->shift_vec[0];
 117     fshift           = fr->fshift[0];
 118     facel            = _mm_set1_pd(fr->epsfac);
 119     charge           = mdatoms->chargeA;
 120     nvdwtype         = fr->ntype;
 121     vdwparam         = fr->nbfp;
 122     vdwtype          = mdatoms->typeA;
 123     vdwgridparam     = fr->ljpme_c6grid;
 124     sh_lj_ewald      = _mm_set1_pd(fr->ic->sh_lj_ewald);
 125     ewclj            = _mm_set1_pd(fr->ewaldcoeff_lj);
 126     ewclj2           = _mm_mul_pd(minus_one,_mm_mul_pd(ewclj,ewclj));
 127
 128     sh_ewald         = _mm_set1_pd(fr->ic->sh_ewald);
 129     ewtab            = fr->ic->tabq_coul_FDV0;
 130     ewtabscale       = _mm_set1_pd(fr->ic->tabq_scale);
 131     ewtabhalfspace   = _mm_set1_pd(0.5/fr->ic->tabq_scale);
 132
 133     /* Avoid stupid compiler warnings */
 134     jnrA = jnrB = 0;
 135     j_coord_offsetA = 0;
 136     j_coord_offsetB = 0;
 137
 138     outeriter        = 0;
 139     inneriter        = 0;
 140
 141     /* Start outer loop over neighborlists */
 142     for(iidx=0; iidx<nri; iidx++)
 143     {
 144         /* Load shift vector for this list */
 145         i_shift_offset   = DIM*shiftidx[iidx];
 146
 147         /* Load limits for loop over neighbors */
 148         j_index_start    = jindex[iidx];
 149         j_index_end      = jindex[iidx+1];
 150
 151         /* Get outer coordinate index */
 152         inr              = iinr[iidx];
 153         i_coord_offset   = DIM*inr;
 154
 155         /* Load i particle coords and add shift vector */
 156         gmx_mm_load_shift_and_1rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 157
 158         fix0             = _mm_setzero_pd();
 159         fiy0             = _mm_setzero_pd();
 160         fiz0             = _mm_setzero_pd();
 161
 162         /* Load parameters for i particles */
 163         iq0              = _mm_mul_pd(facel,_mm_load1_pd(charge+inr+0));
 164         vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 165
 166         /* Reset potential sums */
 167         velecsum         = _mm_setzero_pd();
 168         vvdwsum          = _mm_setzero_pd();
 169
 170         /* Start inner kernel loop */
 171         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 172         {
 173
 174             /* Get j neighbor index, and coordinate index */
 175             jnrA             = jjnr[jidx];
 176             jnrB             = jjnr[jidx+1];
 177             j_coord_offsetA  = DIM*jnrA;
 178             j_coord_offsetB  = DIM*jnrB;
 179
 180             /* load j atom coordinates */
 181             gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 182                                               &jx0,&jy0,&jz0);
 183
 184             /* Calculate displacement vector */
 185             dx00             = _mm_sub_pd(ix0,jx0);
 186             dy00             = _mm_sub_pd(iy0,jy0);
 187             dz00             = _mm_sub_pd(iz0,jz0);
 188
 189             /* Calculate squared distance and things based on it */
 190             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 191
 192             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 193
 194             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 195
 196             /* Load parameters for j particles */
 197             jq0              = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
 198             vdwjidx0A        = 2*vdwtype[jnrA+0];
 199             vdwjidx0B        = 2*vdwtype[jnrB+0];
 200
 201             /**************************
 202              * CALCULATE INTERACTIONS *
 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             gmx_mm_load_2pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,
 210                                          vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
 211
 212             c6grid_00       = gmx_mm_load_2real_swizzle_pd(vdwgridparam+vdwioffset0+vdwjidx0A,
 213                                                                vdwgridparam+vdwioffset0+vdwjidx0B);
 214
 215             /* EWALD ELECTROSTATICS */
 216
 217             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 218             ewrt             = _mm_mul_pd(r00,ewtabscale);
 219             ewitab           = _mm_cvttpd_epi32(ewrt);
 220             eweps            = _mm_sub_pd(ewrt,_mm_cvtepi32_pd(ewitab));
 221             ewitab           = _mm_slli_epi32(ewitab,2);
 222             ewtabF           = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 223             ewtabD           = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,1) );
 224             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 225             ewtabV           = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,0) +2);
 226             ewtabFn          = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,1) +2);
 227             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 228             felec            = _mm_add_pd(ewtabF,_mm_mul_pd(eweps,ewtabD));
 229             velec            = _mm_sub_pd(ewtabV,_mm_mul_pd(_mm_mul_pd(ewtabhalfspace,eweps),_mm_add_pd(ewtabF,felec)));
 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             /* Analytical LJ-PME */
 234             rinvsix          = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 235             ewcljrsq         = _mm_mul_pd(ewclj2,rsq00);
 236             ewclj6           = _mm_mul_pd(ewclj2,_mm_mul_pd(ewclj2,ewclj2));
 237             exponent         = gmx_simd_exp_d(ewcljrsq);
 238             /* poly = exp(-(beta*r)^2) * (1 + (beta*r)^2 + (beta*r)^4 /2) */
 239             poly             = _mm_mul_pd(exponent,_mm_add_pd(_mm_sub_pd(one,ewcljrsq),_mm_mul_pd(_mm_mul_pd(ewcljrsq,ewcljrsq),one_half)));
 240             /* vvdw6 = [C6 - C6grid * (1-poly)]/r6 */
 241             vvdw6            = _mm_mul_pd(_mm_sub_pd(c6_00,_mm_mul_pd(c6grid_00,_mm_sub_pd(one,poly))),rinvsix);
 242             vvdw12           = _mm_mul_pd(c12_00,_mm_mul_pd(rinvsix,rinvsix));
 243             vvdw             = _mm_sub_pd(_mm_mul_pd(vvdw12,one_twelfth),_mm_mul_pd(vvdw6,one_sixth));
 244             /* fvdw = vvdw12/r - (vvdw6/r + (C6grid * exponent * beta^6)/r) */
 245             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);
 246
 247             /* Update potential sum for this i atom from the interaction with this j atom. */
 248             velecsum         = _mm_add_pd(velecsum,velec);
 249             vvdwsum          = _mm_add_pd(vvdwsum,vvdw);
 250
 251             fscal            = _mm_add_pd(felec,fvdw);
 252
 253             /* Calculate temporary vectorial force */
 254             tx               = _mm_mul_pd(fscal,dx00);
 255             ty               = _mm_mul_pd(fscal,dy00);
 256             tz               = _mm_mul_pd(fscal,dz00);
 257
 258             /* Update vectorial force */
 259             fix0             = _mm_add_pd(fix0,tx);
 260             fiy0             = _mm_add_pd(fiy0,ty);
 261             fiz0             = _mm_add_pd(fiz0,tz);
 262
 263             gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,tx,ty,tz);
 264
 265             /* Inner loop uses 69 flops */
 266         }
 267
 268         if(jidx<j_index_end)
 269         {
 270
 271             jnrA             = jjnr[jidx];
 272             j_coord_offsetA  = DIM*jnrA;
 273
 274             /* load j atom coordinates */
 275             gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
 276                                               &jx0,&jy0,&jz0);
 277
 278             /* Calculate displacement vector */
 279             dx00             = _mm_sub_pd(ix0,jx0);
 280             dy00             = _mm_sub_pd(iy0,jy0);
 281             dz00             = _mm_sub_pd(iz0,jz0);
 282
 283             /* Calculate squared distance and things based on it */
 284             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 285
 286             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 287
 288             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 289
 290             /* Load parameters for j particles */
 291             jq0              = _mm_load_sd(charge+jnrA+0);
 292             vdwjidx0A        = 2*vdwtype[jnrA+0];
 293
 294             /**************************
 295              * CALCULATE INTERACTIONS *
 296              **************************/
 297
 298             r00              = _mm_mul_pd(rsq00,rinv00);
 299
 300             /* Compute parameters for interactions between i and j atoms */
 301             qq00             = _mm_mul_pd(iq0,jq0);
 302             gmx_mm_load_1pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,&c6_00,&c12_00);
 303
 304             c6grid_00       = gmx_mm_load_1real_pd(vdwgridparam+vdwioffset0+vdwjidx0A);
 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             eweps            = _mm_sub_pd(ewrt,_mm_cvtepi32_pd(ewitab));
 312             ewitab           = _mm_slli_epi32(ewitab,2);
 313             ewtabF           = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 314             ewtabD           = _mm_setzero_pd();
 315             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 316             ewtabV           = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,0) +2);
 317             ewtabFn          = _mm_setzero_pd();
 318             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 319             felec            = _mm_add_pd(ewtabF,_mm_mul_pd(eweps,ewtabD));
 320             velec            = _mm_sub_pd(ewtabV,_mm_mul_pd(_mm_mul_pd(ewtabhalfspace,eweps),_mm_add_pd(ewtabF,felec)));
 321             velec            = _mm_mul_pd(qq00,_mm_sub_pd(rinv00,velec));
 322             felec            = _mm_mul_pd(_mm_mul_pd(qq00,rinv00),_mm_sub_pd(rinvsq00,felec));
 323
 324             /* Analytical LJ-PME */
 325             rinvsix          = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 326             ewcljrsq         = _mm_mul_pd(ewclj2,rsq00);
 327             ewclj6           = _mm_mul_pd(ewclj2,_mm_mul_pd(ewclj2,ewclj2));
 328             exponent         = gmx_simd_exp_d(ewcljrsq);
 329             /* poly = exp(-(beta*r)^2) * (1 + (beta*r)^2 + (beta*r)^4 /2) */
 330             poly             = _mm_mul_pd(exponent,_mm_add_pd(_mm_sub_pd(one,ewcljrsq),_mm_mul_pd(_mm_mul_pd(ewcljrsq,ewcljrsq),one_half)));
 331             /* vvdw6 = [C6 - C6grid * (1-poly)]/r6 */
 332             vvdw6            = _mm_mul_pd(_mm_sub_pd(c6_00,_mm_mul_pd(c6grid_00,_mm_sub_pd(one,poly))),rinvsix);
 333             vvdw12           = _mm_mul_pd(c12_00,_mm_mul_pd(rinvsix,rinvsix));
 334             vvdw             = _mm_sub_pd(_mm_mul_pd(vvdw12,one_twelfth),_mm_mul_pd(vvdw6,one_sixth));
 335             /* fvdw = vvdw12/r - (vvdw6/r + (C6grid * exponent * beta^6)/r) */
 336             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);
 337
 338             /* Update potential sum for this i atom from the interaction with this j atom. */
 339             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 340             velecsum         = _mm_add_pd(velecsum,velec);
 341             vvdw             = _mm_unpacklo_pd(vvdw,_mm_setzero_pd());
 342             vvdwsum          = _mm_add_pd(vvdwsum,vvdw);
 343
 344             fscal            = _mm_add_pd(felec,fvdw);
 345
 346             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 347
 348             /* Calculate temporary vectorial force */
 349             tx               = _mm_mul_pd(fscal,dx00);
 350             ty               = _mm_mul_pd(fscal,dy00);
 351             tz               = _mm_mul_pd(fscal,dz00);
 352
 353             /* Update vectorial force */
 354             fix0             = _mm_add_pd(fix0,tx);
 355             fiy0             = _mm_add_pd(fiy0,ty);
 356             fiz0             = _mm_add_pd(fiz0,tz);
 357
 358             gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,tx,ty,tz);
 359
 360             /* Inner loop uses 69 flops */
 361         }
 362
 363         /* End of innermost loop */
 364
 365         gmx_mm_update_iforce_1atom_swizzle_pd(fix0,fiy0,fiz0,
 366                                               f+i_coord_offset,fshift+i_shift_offset);
 367
 368         ggid                        = gid[iidx];
 369         /* Update potential energies */
 370         gmx_mm_update_1pot_pd(velecsum,kernel_data->energygrp_elec+ggid);
 371         gmx_mm_update_1pot_pd(vvdwsum,kernel_data->energygrp_vdw+ggid);
 372
 373         /* Increment number of inner iterations */
 374         inneriter                  += j_index_end - j_index_start;
 375
 376         /* Outer loop uses 9 flops */
 377     }
 378
 379     /* Increment number of outer iterations */
 380     outeriter        += nri;
 381
 382     /* Update outer/inner flops */
 383
 384     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_VF,outeriter*9 + inneriter*69);
 385 }
 386 /*
 387  * Gromacs nonbonded kernel:   nb_kernel_ElecEw_VdwLJEw_GeomP1P1_F_sse2_double
 388  * Electrostatics interaction: Ewald
 389  * VdW interaction:            LJEwald
 390  * Geometry:                   Particle-Particle
 391  * Calculate force/pot:        Force
 392  */
 393 void
 394 nb_kernel_ElecEw_VdwLJEw_GeomP1P1_F_sse2_double
 395                     (t_nblist                    * gmx_restrict       nlist,
 396                      rvec                        * gmx_restrict          xx,
 397                      rvec                        * gmx_restrict          ff,
 398                      t_forcerec                  * gmx_restrict          fr,
 399                      t_mdatoms                   * gmx_restrict     mdatoms,
 400                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 401                      t_nrnb                      * gmx_restrict        nrnb)
 402 {
 403     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 404      * just 0 for non-waters.
 405      * Suffixes A,B refer to j loop unrolling done with SSE double precision, e.g. for the two different
 406      * jnr indices corresponding to data put in the four positions in the SIMD register.
 407      */
 408     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 409     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 410     int              jnrA,jnrB;
 411     int              j_coord_offsetA,j_coord_offsetB;
 412     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 413     real             rcutoff_scalar;
 414     real             *shiftvec,*fshift,*x,*f;
 415     __m128d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 416     int              vdwioffset0;
 417     __m128d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 418     int              vdwjidx0A,vdwjidx0B;
 419     __m128d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 420     __m128d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 421     __m128d          velec,felec,velecsum,facel,crf,krf,krf2;
 422     real             *charge;
 423     int              nvdwtype;
 424     __m128d          rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 425     int              *vdwtype;
 426     real             *vdwparam;
 427     __m128d          one_sixth   = _mm_set1_pd(1.0/6.0);
 428     __m128d          one_twelfth = _mm_set1_pd(1.0/12.0);
 429     __m128d           c6grid_00;
 430     __m128d           ewclj,ewclj2,ewclj6,ewcljrsq,poly,exponent,f6A,f6B,sh_lj_ewald;
 431     real             *vdwgridparam;
 432     __m128d           one_half = _mm_set1_pd(0.5);
 433     __m128d           minus_one = _mm_set1_pd(-1.0);
 434     __m128i          ewitab;
 435     __m128d          ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
 436     real             *ewtab;
 437     __m128d          dummy_mask,cutoff_mask;
 438     __m128d          signbit   = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
 439     __m128d          one     = _mm_set1_pd(1.0);
 440     __m128d          two     = _mm_set1_pd(2.0);
 441     x                = xx[0];
 442     f                = ff[0];
 443
 444     nri              = nlist->nri;
 445     iinr             = nlist->iinr;
 446     jindex           = nlist->jindex;
 447     jjnr             = nlist->jjnr;
 448     shiftidx         = nlist->shift;
 449     gid              = nlist->gid;
 450     shiftvec         = fr->shift_vec[0];
 451     fshift           = fr->fshift[0];
 452     facel            = _mm_set1_pd(fr->epsfac);
 453     charge           = mdatoms->chargeA;
 454     nvdwtype         = fr->ntype;
 455     vdwparam         = fr->nbfp;
 456     vdwtype          = mdatoms->typeA;
 457     vdwgridparam     = fr->ljpme_c6grid;
 458     sh_lj_ewald      = _mm_set1_pd(fr->ic->sh_lj_ewald);
 459     ewclj            = _mm_set1_pd(fr->ewaldcoeff_lj);
 460     ewclj2           = _mm_mul_pd(minus_one,_mm_mul_pd(ewclj,ewclj));
 461
 462     sh_ewald         = _mm_set1_pd(fr->ic->sh_ewald);
 463     ewtab            = fr->ic->tabq_coul_F;
 464     ewtabscale       = _mm_set1_pd(fr->ic->tabq_scale);
 465     ewtabhalfspace   = _mm_set1_pd(0.5/fr->ic->tabq_scale);
 466
 467     /* Avoid stupid compiler warnings */
 468     jnrA = jnrB = 0;
 469     j_coord_offsetA = 0;
 470     j_coord_offsetB = 0;
 471
 472     outeriter        = 0;
 473     inneriter        = 0;
 474
 475     /* Start outer loop over neighborlists */
 476     for(iidx=0; iidx<nri; iidx++)
 477     {
 478         /* Load shift vector for this list */
 479         i_shift_offset   = DIM*shiftidx[iidx];
 480
 481         /* Load limits for loop over neighbors */
 482         j_index_start    = jindex[iidx];
 483         j_index_end      = jindex[iidx+1];
 484
 485         /* Get outer coordinate index */
 486         inr              = iinr[iidx];
 487         i_coord_offset   = DIM*inr;
 488
 489         /* Load i particle coords and add shift vector */
 490         gmx_mm_load_shift_and_1rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 491
 492         fix0             = _mm_setzero_pd();
 493         fiy0             = _mm_setzero_pd();
 494         fiz0             = _mm_setzero_pd();
 495
 496         /* Load parameters for i particles */
 497         iq0              = _mm_mul_pd(facel,_mm_load1_pd(charge+inr+0));
 498         vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 499
 500         /* Start inner kernel loop */
 501         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 502         {
 503
 504             /* Get j neighbor index, and coordinate index */
 505             jnrA             = jjnr[jidx];
 506             jnrB             = jjnr[jidx+1];
 507             j_coord_offsetA  = DIM*jnrA;
 508             j_coord_offsetB  = DIM*jnrB;
 509
 510             /* load j atom coordinates */
 511             gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 512                                               &jx0,&jy0,&jz0);
 513
 514             /* Calculate displacement vector */
 515             dx00             = _mm_sub_pd(ix0,jx0);
 516             dy00             = _mm_sub_pd(iy0,jy0);
 517             dz00             = _mm_sub_pd(iz0,jz0);
 518
 519             /* Calculate squared distance and things based on it */
 520             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 521
 522             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 523
 524             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 525
 526             /* Load parameters for j particles */
 527             jq0              = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
 528             vdwjidx0A        = 2*vdwtype[jnrA+0];
 529             vdwjidx0B        = 2*vdwtype[jnrB+0];
 530
 531             /**************************
 532              * CALCULATE INTERACTIONS *
 533              **************************/
 534
 535             r00              = _mm_mul_pd(rsq00,rinv00);
 536
 537             /* Compute parameters for interactions between i and j atoms */
 538             qq00             = _mm_mul_pd(iq0,jq0);
 539             gmx_mm_load_2pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,
 540                                          vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
 541
 542             c6grid_00       = gmx_mm_load_2real_swizzle_pd(vdwgridparam+vdwioffset0+vdwjidx0A,
 543                                                                vdwgridparam+vdwioffset0+vdwjidx0B);
 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             eweps            = _mm_sub_pd(ewrt,_mm_cvtepi32_pd(ewitab));
 551             gmx_mm_load_2pair_swizzle_pd(ewtab+gmx_mm_extract_epi32(ewitab,0),ewtab+gmx_mm_extract_epi32(ewitab,1),
 552                                          &ewtabF,&ewtabFn);
 553             felec            = _mm_add_pd(_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF),_mm_mul_pd(eweps,ewtabFn));
 554             felec            = _mm_mul_pd(_mm_mul_pd(qq00,rinv00),_mm_sub_pd(rinvsq00,felec));
 555
 556             /* Analytical LJ-PME */
 557             rinvsix          = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 558             ewcljrsq         = _mm_mul_pd(ewclj2,rsq00);
 559             ewclj6           = _mm_mul_pd(ewclj2,_mm_mul_pd(ewclj2,ewclj2));
 560             exponent         = gmx_simd_exp_d(ewcljrsq);
 561             /* poly = exp(-(beta*r)^2) * (1 + (beta*r)^2 + (beta*r)^4 /2) */
 562             poly             = _mm_mul_pd(exponent,_mm_add_pd(_mm_sub_pd(one,ewcljrsq),_mm_mul_pd(_mm_mul_pd(ewcljrsq,ewcljrsq),one_half)));
 563             /* f6A = 6 * C6grid * (1 - poly) */
 564             f6A              = _mm_mul_pd(c6grid_00,_mm_sub_pd(one,poly));
 565             /* f6B = C6grid * exponent * beta^6 */
 566             f6B              = _mm_mul_pd(_mm_mul_pd(c6grid_00,one_sixth),_mm_mul_pd(exponent,ewclj6));
 567             /* fvdw = 12*C12/r13 - ((6*C6 - f6A)/r6 + f6B)/r */
 568             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);
 569
 570             fscal            = _mm_add_pd(felec,fvdw);
 571
 572             /* Calculate temporary vectorial force */
 573             tx               = _mm_mul_pd(fscal,dx00);
 574             ty               = _mm_mul_pd(fscal,dy00);
 575             tz               = _mm_mul_pd(fscal,dz00);
 576
 577             /* Update vectorial force */
 578             fix0             = _mm_add_pd(fix0,tx);
 579             fiy0             = _mm_add_pd(fiy0,ty);
 580             fiz0             = _mm_add_pd(fiz0,tz);
 581
 582             gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,tx,ty,tz);
 583
 584             /* Inner loop uses 59 flops */
 585         }
 586
 587         if(jidx<j_index_end)
 588         {
 589
 590             jnrA             = jjnr[jidx];
 591             j_coord_offsetA  = DIM*jnrA;
 592
 593             /* load j atom coordinates */
 594             gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
 595                                               &jx0,&jy0,&jz0);
 596
 597             /* Calculate displacement vector */
 598             dx00             = _mm_sub_pd(ix0,jx0);
 599             dy00             = _mm_sub_pd(iy0,jy0);
 600             dz00             = _mm_sub_pd(iz0,jz0);
 601
 602             /* Calculate squared distance and things based on it */
 603             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 604
 605             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 606
 607             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 608
 609             /* Load parameters for j particles */
 610             jq0              = _mm_load_sd(charge+jnrA+0);
 611             vdwjidx0A        = 2*vdwtype[jnrA+0];
 612
 613             /**************************
 614              * CALCULATE INTERACTIONS *
 615              **************************/
 616
 617             r00              = _mm_mul_pd(rsq00,rinv00);
 618
 619             /* Compute parameters for interactions between i and j atoms */
 620             qq00             = _mm_mul_pd(iq0,jq0);
 621             gmx_mm_load_1pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,&c6_00,&c12_00);
 622
 623             c6grid_00       = gmx_mm_load_1real_pd(vdwgridparam+vdwioffset0+vdwjidx0A);
 624
 625             /* EWALD ELECTROSTATICS */
 626
 627             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 628             ewrt             = _mm_mul_pd(r00,ewtabscale);
 629             ewitab           = _mm_cvttpd_epi32(ewrt);
 630             eweps            = _mm_sub_pd(ewrt,_mm_cvtepi32_pd(ewitab));
 631             gmx_mm_load_1pair_swizzle_pd(ewtab+gmx_mm_extract_epi32(ewitab,0),&ewtabF,&ewtabFn);
 632             felec            = _mm_add_pd(_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF),_mm_mul_pd(eweps,ewtabFn));
 633             felec            = _mm_mul_pd(_mm_mul_pd(qq00,rinv00),_mm_sub_pd(rinvsq00,felec));
 634
 635             /* Analytical LJ-PME */
 636             rinvsix          = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 637             ewcljrsq         = _mm_mul_pd(ewclj2,rsq00);
 638             ewclj6           = _mm_mul_pd(ewclj2,_mm_mul_pd(ewclj2,ewclj2));
 639             exponent         = gmx_simd_exp_d(ewcljrsq);
 640             /* poly = exp(-(beta*r)^2) * (1 + (beta*r)^2 + (beta*r)^4 /2) */
 641             poly             = _mm_mul_pd(exponent,_mm_add_pd(_mm_sub_pd(one,ewcljrsq),_mm_mul_pd(_mm_mul_pd(ewcljrsq,ewcljrsq),one_half)));
 642             /* f6A = 6 * C6grid * (1 - poly) */
 643             f6A              = _mm_mul_pd(c6grid_00,_mm_sub_pd(one,poly));
 644             /* f6B = C6grid * exponent * beta^6 */
 645             f6B              = _mm_mul_pd(_mm_mul_pd(c6grid_00,one_sixth),_mm_mul_pd(exponent,ewclj6));
 646             /* fvdw = 12*C12/r13 - ((6*C6 - f6A)/r6 + f6B)/r */
 647             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);
 648
 649             fscal            = _mm_add_pd(felec,fvdw);
 650
 651             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 652
 653             /* Calculate temporary vectorial force */
 654             tx               = _mm_mul_pd(fscal,dx00);
 655             ty               = _mm_mul_pd(fscal,dy00);
 656             tz               = _mm_mul_pd(fscal,dz00);
 657
 658             /* Update vectorial force */
 659             fix0             = _mm_add_pd(fix0,tx);
 660             fiy0             = _mm_add_pd(fiy0,ty);
 661             fiz0             = _mm_add_pd(fiz0,tz);
 662
 663             gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,tx,ty,tz);
 664
 665             /* Inner loop uses 59 flops */
 666         }
 667
 668         /* End of innermost loop */
 669
 670         gmx_mm_update_iforce_1atom_swizzle_pd(fix0,fiy0,fiz0,
 671                                               f+i_coord_offset,fshift+i_shift_offset);
 672
 673         /* Increment number of inner iterations */
 674         inneriter                  += j_index_end - j_index_start;
 675
 676         /* Outer loop uses 7 flops */
 677     }
 678
 679     /* Increment number of outer iterations */
 680     outeriter        += nri;
 681
 682     /* Update outer/inner flops */
 683
 684     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_F,outeriter*7 + inneriter*59);
 685 }