src/gromacs/gmxlib/nonbonded/nb_kernel_sse4_1_double/nb_kernel_ElecEw_VdwLJEw_GeomP1P1_sse4_1_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 sse4_1_double kernel generator.
  37  */
  38 #ifdef HAVE_CONFIG_H
  39 #include <config.h>
  40 #endif
  41
  42 #include <math.h>
  43
  44 #include "../nb_kernel.h"
  45 #include "types/simple.h"
  46 #include "vec.h"
  47 #include "nrnb.h"
  48
  49 #include "gromacs/simd/math_x86_sse4_1_double.h"
  50 #include "kernelutil_x86_sse4_1_double.h"
  51
  52 /*
  53  * Gromacs nonbonded kernel:   nb_kernel_ElecEw_VdwLJEw_GeomP1P1_VF_sse4_1_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_sse4_1_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             c6grid_00       = gmx_mm_load_2real_swizzle_pd(vdwgridparam+vdwioffset0+vdwjidx0A,
 212                                                                vdwgridparam+vdwioffset0+vdwjidx0B);
 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             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 220             ewitab           = _mm_slli_epi32(ewitab,2);
 221             ewtabF           = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 222             ewtabD           = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,1) );
 223             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 224             ewtabV           = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,0) +2);
 225             ewtabFn          = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,1) +2);
 226             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 227             felec            = _mm_add_pd(ewtabF,_mm_mul_pd(eweps,ewtabD));
 228             velec            = _mm_sub_pd(ewtabV,_mm_mul_pd(_mm_mul_pd(ewtabhalfspace,eweps),_mm_add_pd(ewtabF,felec)));
 229             velec            = _mm_mul_pd(qq00,_mm_sub_pd(rinv00,velec));
 230             felec            = _mm_mul_pd(_mm_mul_pd(qq00,rinv00),_mm_sub_pd(rinvsq00,felec));
 231
 232             /* Analytical LJ-PME */
 233             rinvsix          = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 234             ewcljrsq         = _mm_mul_pd(ewclj2,rsq00);
 235             ewclj6           = _mm_mul_pd(ewclj2,_mm_mul_pd(ewclj2,ewclj2));
 236             exponent         = gmx_simd_exp_d(ewcljrsq);
 237             /* poly = exp(-(beta*r)^2) * (1 + (beta*r)^2 + (beta*r)^4 /2) */
 238             poly             = _mm_mul_pd(exponent,_mm_add_pd(_mm_sub_pd(one,ewcljrsq),_mm_mul_pd(_mm_mul_pd(ewcljrsq,ewcljrsq),one_half)));
 239             /* vvdw6 = [C6 - C6grid * (1-poly)]/r6 */
 240             vvdw6            = _mm_mul_pd(_mm_sub_pd(c6_00,_mm_mul_pd(c6grid_00,_mm_sub_pd(one,poly))),rinvsix);
 241             vvdw12           = _mm_mul_pd(c12_00,_mm_mul_pd(rinvsix,rinvsix));
 242             vvdw             = _mm_sub_pd(_mm_mul_pd(vvdw12,one_twelfth),_mm_mul_pd(vvdw6,one_sixth));
 243             /* fvdw = vvdw12/r - (vvdw6/r + (C6grid * exponent * beta^6)/r) */
 244             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);
 245
 246             /* Update potential sum for this i atom from the interaction with this j atom. */
 247             velecsum         = _mm_add_pd(velecsum,velec);
 248             vvdwsum          = _mm_add_pd(vvdwsum,vvdw);
 249
 250             fscal            = _mm_add_pd(felec,fvdw);
 251
 252             /* Calculate temporary vectorial force */
 253             tx               = _mm_mul_pd(fscal,dx00);
 254             ty               = _mm_mul_pd(fscal,dy00);
 255             tz               = _mm_mul_pd(fscal,dz00);
 256
 257             /* Update vectorial force */
 258             fix0             = _mm_add_pd(fix0,tx);
 259             fiy0             = _mm_add_pd(fiy0,ty);
 260             fiz0             = _mm_add_pd(fiz0,tz);
 261
 262             gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,tx,ty,tz);
 263
 264             /* Inner loop uses 69 flops */
 265         }
 266
 267         if(jidx<j_index_end)
 268         {
 269
 270             jnrA             = jjnr[jidx];
 271             j_coord_offsetA  = DIM*jnrA;
 272
 273             /* load j atom coordinates */
 274             gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
 275                                               &jx0,&jy0,&jz0);
 276
 277             /* Calculate displacement vector */
 278             dx00             = _mm_sub_pd(ix0,jx0);
 279             dy00             = _mm_sub_pd(iy0,jy0);
 280             dz00             = _mm_sub_pd(iz0,jz0);
 281
 282             /* Calculate squared distance and things based on it */
 283             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 284
 285             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 286
 287             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 288
 289             /* Load parameters for j particles */
 290             jq0              = _mm_load_sd(charge+jnrA+0);
 291             vdwjidx0A        = 2*vdwtype[jnrA+0];
 292
 293             /**************************
 294              * CALCULATE INTERACTIONS *
 295              **************************/
 296
 297             r00              = _mm_mul_pd(rsq00,rinv00);
 298
 299             /* Compute parameters for interactions between i and j atoms */
 300             qq00             = _mm_mul_pd(iq0,jq0);
 301             gmx_mm_load_1pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,&c6_00,&c12_00);
 302
 303             c6grid_00       = gmx_mm_load_1real_pd(vdwgridparam+vdwioffset0+vdwjidx0A);
 304
 305             /* EWALD ELECTROSTATICS */
 306
 307             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 308             ewrt             = _mm_mul_pd(r00,ewtabscale);
 309             ewitab           = _mm_cvttpd_epi32(ewrt);
 310             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 311             ewitab           = _mm_slli_epi32(ewitab,2);
 312             ewtabF           = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 313             ewtabD           = _mm_setzero_pd();
 314             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 315             ewtabV           = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,0) +2);
 316             ewtabFn          = _mm_setzero_pd();
 317             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 318             felec            = _mm_add_pd(ewtabF,_mm_mul_pd(eweps,ewtabD));
 319             velec            = _mm_sub_pd(ewtabV,_mm_mul_pd(_mm_mul_pd(ewtabhalfspace,eweps),_mm_add_pd(ewtabF,felec)));
 320             velec            = _mm_mul_pd(qq00,_mm_sub_pd(rinv00,velec));
 321             felec            = _mm_mul_pd(_mm_mul_pd(qq00,rinv00),_mm_sub_pd(rinvsq00,felec));
 322
 323             /* Analytical LJ-PME */
 324             rinvsix          = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 325             ewcljrsq         = _mm_mul_pd(ewclj2,rsq00);
 326             ewclj6           = _mm_mul_pd(ewclj2,_mm_mul_pd(ewclj2,ewclj2));
 327             exponent         = gmx_simd_exp_d(ewcljrsq);
 328             /* poly = exp(-(beta*r)^2) * (1 + (beta*r)^2 + (beta*r)^4 /2) */
 329             poly             = _mm_mul_pd(exponent,_mm_add_pd(_mm_sub_pd(one,ewcljrsq),_mm_mul_pd(_mm_mul_pd(ewcljrsq,ewcljrsq),one_half)));
 330             /* vvdw6 = [C6 - C6grid * (1-poly)]/r6 */
 331             vvdw6            = _mm_mul_pd(_mm_sub_pd(c6_00,_mm_mul_pd(c6grid_00,_mm_sub_pd(one,poly))),rinvsix);
 332             vvdw12           = _mm_mul_pd(c12_00,_mm_mul_pd(rinvsix,rinvsix));
 333             vvdw             = _mm_sub_pd(_mm_mul_pd(vvdw12,one_twelfth),_mm_mul_pd(vvdw6,one_sixth));
 334             /* fvdw = vvdw12/r - (vvdw6/r + (C6grid * exponent * beta^6)/r) */
 335             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);
 336
 337             /* Update potential sum for this i atom from the interaction with this j atom. */
 338             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 339             velecsum         = _mm_add_pd(velecsum,velec);
 340             vvdw             = _mm_unpacklo_pd(vvdw,_mm_setzero_pd());
 341             vvdwsum          = _mm_add_pd(vvdwsum,vvdw);
 342
 343             fscal            = _mm_add_pd(felec,fvdw);
 344
 345             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 346
 347             /* Calculate temporary vectorial force */
 348             tx               = _mm_mul_pd(fscal,dx00);
 349             ty               = _mm_mul_pd(fscal,dy00);
 350             tz               = _mm_mul_pd(fscal,dz00);
 351
 352             /* Update vectorial force */
 353             fix0             = _mm_add_pd(fix0,tx);
 354             fiy0             = _mm_add_pd(fiy0,ty);
 355             fiz0             = _mm_add_pd(fiz0,tz);
 356
 357             gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,tx,ty,tz);
 358
 359             /* Inner loop uses 69 flops */
 360         }
 361
 362         /* End of innermost loop */
 363
 364         gmx_mm_update_iforce_1atom_swizzle_pd(fix0,fiy0,fiz0,
 365                                               f+i_coord_offset,fshift+i_shift_offset);
 366
 367         ggid                        = gid[iidx];
 368         /* Update potential energies */
 369         gmx_mm_update_1pot_pd(velecsum,kernel_data->energygrp_elec+ggid);
 370         gmx_mm_update_1pot_pd(vvdwsum,kernel_data->energygrp_vdw+ggid);
 371
 372         /* Increment number of inner iterations */
 373         inneriter                  += j_index_end - j_index_start;
 374
 375         /* Outer loop uses 9 flops */
 376     }
 377
 378     /* Increment number of outer iterations */
 379     outeriter        += nri;
 380
 381     /* Update outer/inner flops */
 382
 383     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_VF,outeriter*9 + inneriter*69);
 384 }
 385 /*
 386  * Gromacs nonbonded kernel:   nb_kernel_ElecEw_VdwLJEw_GeomP1P1_F_sse4_1_double
 387  * Electrostatics interaction: Ewald
 388  * VdW interaction:            LJEwald
 389  * Geometry:                   Particle-Particle
 390  * Calculate force/pot:        Force
 391  */
 392 void
 393 nb_kernel_ElecEw_VdwLJEw_GeomP1P1_F_sse4_1_double
 394                     (t_nblist                    * gmx_restrict       nlist,
 395                      rvec                        * gmx_restrict          xx,
 396                      rvec                        * gmx_restrict          ff,
 397                      t_forcerec                  * gmx_restrict          fr,
 398                      t_mdatoms                   * gmx_restrict     mdatoms,
 399                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 400                      t_nrnb                      * gmx_restrict        nrnb)
 401 {
 402     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 403      * just 0 for non-waters.
 404      * Suffixes A,B refer to j loop unrolling done with SSE double precision, e.g. for the two different
 405      * jnr indices corresponding to data put in the four positions in the SIMD register.
 406      */
 407     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 408     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 409     int              jnrA,jnrB;
 410     int              j_coord_offsetA,j_coord_offsetB;
 411     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 412     real             rcutoff_scalar;
 413     real             *shiftvec,*fshift,*x,*f;
 414     __m128d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 415     int              vdwioffset0;
 416     __m128d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 417     int              vdwjidx0A,vdwjidx0B;
 418     __m128d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 419     __m128d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 420     __m128d          velec,felec,velecsum,facel,crf,krf,krf2;
 421     real             *charge;
 422     int              nvdwtype;
 423     __m128d          rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 424     int              *vdwtype;
 425     real             *vdwparam;
 426     __m128d          one_sixth   = _mm_set1_pd(1.0/6.0);
 427     __m128d          one_twelfth = _mm_set1_pd(1.0/12.0);
 428     __m128d           c6grid_00;
 429     __m128d           ewclj,ewclj2,ewclj6,ewcljrsq,poly,exponent,f6A,f6B,sh_lj_ewald;
 430     real             *vdwgridparam;
 431     __m128d           one_half = _mm_set1_pd(0.5);
 432     __m128d           minus_one = _mm_set1_pd(-1.0);
 433     __m128i          ewitab;
 434     __m128d          ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
 435     real             *ewtab;
 436     __m128d          dummy_mask,cutoff_mask;
 437     __m128d          signbit   = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
 438     __m128d          one     = _mm_set1_pd(1.0);
 439     __m128d          two     = _mm_set1_pd(2.0);
 440     x                = xx[0];
 441     f                = ff[0];
 442
 443     nri              = nlist->nri;
 444     iinr             = nlist->iinr;
 445     jindex           = nlist->jindex;
 446     jjnr             = nlist->jjnr;
 447     shiftidx         = nlist->shift;
 448     gid              = nlist->gid;
 449     shiftvec         = fr->shift_vec[0];
 450     fshift           = fr->fshift[0];
 451     facel            = _mm_set1_pd(fr->epsfac);
 452     charge           = mdatoms->chargeA;
 453     nvdwtype         = fr->ntype;
 454     vdwparam         = fr->nbfp;
 455     vdwtype          = mdatoms->typeA;
 456     vdwgridparam     = fr->ljpme_c6grid;
 457     sh_lj_ewald      = _mm_set1_pd(fr->ic->sh_lj_ewald);
 458     ewclj            = _mm_set1_pd(fr->ewaldcoeff_lj);
 459     ewclj2           = _mm_mul_pd(minus_one,_mm_mul_pd(ewclj,ewclj));
 460
 461     sh_ewald         = _mm_set1_pd(fr->ic->sh_ewald);
 462     ewtab            = fr->ic->tabq_coul_F;
 463     ewtabscale       = _mm_set1_pd(fr->ic->tabq_scale);
 464     ewtabhalfspace   = _mm_set1_pd(0.5/fr->ic->tabq_scale);
 465
 466     /* Avoid stupid compiler warnings */
 467     jnrA = jnrB = 0;
 468     j_coord_offsetA = 0;
 469     j_coord_offsetB = 0;
 470
 471     outeriter        = 0;
 472     inneriter        = 0;
 473
 474     /* Start outer loop over neighborlists */
 475     for(iidx=0; iidx<nri; iidx++)
 476     {
 477         /* Load shift vector for this list */
 478         i_shift_offset   = DIM*shiftidx[iidx];
 479
 480         /* Load limits for loop over neighbors */
 481         j_index_start    = jindex[iidx];
 482         j_index_end      = jindex[iidx+1];
 483
 484         /* Get outer coordinate index */
 485         inr              = iinr[iidx];
 486         i_coord_offset   = DIM*inr;
 487
 488         /* Load i particle coords and add shift vector */
 489         gmx_mm_load_shift_and_1rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 490
 491         fix0             = _mm_setzero_pd();
 492         fiy0             = _mm_setzero_pd();
 493         fiz0             = _mm_setzero_pd();
 494
 495         /* Load parameters for i particles */
 496         iq0              = _mm_mul_pd(facel,_mm_load1_pd(charge+inr+0));
 497         vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 498
 499         /* Start inner kernel loop */
 500         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 501         {
 502
 503             /* Get j neighbor index, and coordinate index */
 504             jnrA             = jjnr[jidx];
 505             jnrB             = jjnr[jidx+1];
 506             j_coord_offsetA  = DIM*jnrA;
 507             j_coord_offsetB  = DIM*jnrB;
 508
 509             /* load j atom coordinates */
 510             gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 511                                               &jx0,&jy0,&jz0);
 512
 513             /* Calculate displacement vector */
 514             dx00             = _mm_sub_pd(ix0,jx0);
 515             dy00             = _mm_sub_pd(iy0,jy0);
 516             dz00             = _mm_sub_pd(iz0,jz0);
 517
 518             /* Calculate squared distance and things based on it */
 519             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 520
 521             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 522
 523             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 524
 525             /* Load parameters for j particles */
 526             jq0              = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
 527             vdwjidx0A        = 2*vdwtype[jnrA+0];
 528             vdwjidx0B        = 2*vdwtype[jnrB+0];
 529
 530             /**************************
 531              * CALCULATE INTERACTIONS *
 532              **************************/
 533
 534             r00              = _mm_mul_pd(rsq00,rinv00);
 535
 536             /* Compute parameters for interactions between i and j atoms */
 537             qq00             = _mm_mul_pd(iq0,jq0);
 538             gmx_mm_load_2pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,
 539                                          vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
 540             c6grid_00       = gmx_mm_load_2real_swizzle_pd(vdwgridparam+vdwioffset0+vdwjidx0A,
 541                                                                vdwgridparam+vdwioffset0+vdwjidx0B);
 542
 543             /* EWALD ELECTROSTATICS */
 544
 545             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 546             ewrt             = _mm_mul_pd(r00,ewtabscale);
 547             ewitab           = _mm_cvttpd_epi32(ewrt);
 548             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 549             gmx_mm_load_2pair_swizzle_pd(ewtab+gmx_mm_extract_epi32(ewitab,0),ewtab+gmx_mm_extract_epi32(ewitab,1),
 550                                          &ewtabF,&ewtabFn);
 551             felec            = _mm_add_pd(_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF),_mm_mul_pd(eweps,ewtabFn));
 552             felec            = _mm_mul_pd(_mm_mul_pd(qq00,rinv00),_mm_sub_pd(rinvsq00,felec));
 553
 554             /* Analytical LJ-PME */
 555             rinvsix          = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 556             ewcljrsq         = _mm_mul_pd(ewclj2,rsq00);
 557             ewclj6           = _mm_mul_pd(ewclj2,_mm_mul_pd(ewclj2,ewclj2));
 558             exponent         = gmx_simd_exp_d(ewcljrsq);
 559             /* poly = exp(-(beta*r)^2) * (1 + (beta*r)^2 + (beta*r)^4 /2) */
 560             poly             = _mm_mul_pd(exponent,_mm_add_pd(_mm_sub_pd(one,ewcljrsq),_mm_mul_pd(_mm_mul_pd(ewcljrsq,ewcljrsq),one_half)));
 561             /* f6A = 6 * C6grid * (1 - poly) */
 562             f6A              = _mm_mul_pd(c6grid_00,_mm_sub_pd(one,poly));
 563             /* f6B = C6grid * exponent * beta^6 */
 564             f6B              = _mm_mul_pd(_mm_mul_pd(c6grid_00,one_sixth),_mm_mul_pd(exponent,ewclj6));
 565             /* fvdw = 12*C12/r13 - ((6*C6 - f6A)/r6 + f6B)/r */
 566             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);
 567
 568             fscal            = _mm_add_pd(felec,fvdw);
 569
 570             /* Calculate temporary vectorial force */
 571             tx               = _mm_mul_pd(fscal,dx00);
 572             ty               = _mm_mul_pd(fscal,dy00);
 573             tz               = _mm_mul_pd(fscal,dz00);
 574
 575             /* Update vectorial force */
 576             fix0             = _mm_add_pd(fix0,tx);
 577             fiy0             = _mm_add_pd(fiy0,ty);
 578             fiz0             = _mm_add_pd(fiz0,tz);
 579
 580             gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,tx,ty,tz);
 581
 582             /* Inner loop uses 59 flops */
 583         }
 584
 585         if(jidx<j_index_end)
 586         {
 587
 588             jnrA             = jjnr[jidx];
 589             j_coord_offsetA  = DIM*jnrA;
 590
 591             /* load j atom coordinates */
 592             gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
 593                                               &jx0,&jy0,&jz0);
 594
 595             /* Calculate displacement vector */
 596             dx00             = _mm_sub_pd(ix0,jx0);
 597             dy00             = _mm_sub_pd(iy0,jy0);
 598             dz00             = _mm_sub_pd(iz0,jz0);
 599
 600             /* Calculate squared distance and things based on it */
 601             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 602
 603             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 604
 605             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 606
 607             /* Load parameters for j particles */
 608             jq0              = _mm_load_sd(charge+jnrA+0);
 609             vdwjidx0A        = 2*vdwtype[jnrA+0];
 610
 611             /**************************
 612              * CALCULATE INTERACTIONS *
 613              **************************/
 614
 615             r00              = _mm_mul_pd(rsq00,rinv00);
 616
 617             /* Compute parameters for interactions between i and j atoms */
 618             qq00             = _mm_mul_pd(iq0,jq0);
 619             gmx_mm_load_1pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,&c6_00,&c12_00);
 620
 621             c6grid_00       = gmx_mm_load_1real_pd(vdwgridparam+vdwioffset0+vdwjidx0A);
 622
 623             /* EWALD ELECTROSTATICS */
 624
 625             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 626             ewrt             = _mm_mul_pd(r00,ewtabscale);
 627             ewitab           = _mm_cvttpd_epi32(ewrt);
 628             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 629             gmx_mm_load_1pair_swizzle_pd(ewtab+gmx_mm_extract_epi32(ewitab,0),&ewtabF,&ewtabFn);
 630             felec            = _mm_add_pd(_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF),_mm_mul_pd(eweps,ewtabFn));
 631             felec            = _mm_mul_pd(_mm_mul_pd(qq00,rinv00),_mm_sub_pd(rinvsq00,felec));
 632
 633             /* Analytical LJ-PME */
 634             rinvsix          = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 635             ewcljrsq         = _mm_mul_pd(ewclj2,rsq00);
 636             ewclj6           = _mm_mul_pd(ewclj2,_mm_mul_pd(ewclj2,ewclj2));
 637             exponent         = gmx_simd_exp_d(ewcljrsq);
 638             /* poly = exp(-(beta*r)^2) * (1 + (beta*r)^2 + (beta*r)^4 /2) */
 639             poly             = _mm_mul_pd(exponent,_mm_add_pd(_mm_sub_pd(one,ewcljrsq),_mm_mul_pd(_mm_mul_pd(ewcljrsq,ewcljrsq),one_half)));
 640             /* f6A = 6 * C6grid * (1 - poly) */
 641             f6A              = _mm_mul_pd(c6grid_00,_mm_sub_pd(one,poly));
 642             /* f6B = C6grid * exponent * beta^6 */
 643             f6B              = _mm_mul_pd(_mm_mul_pd(c6grid_00,one_sixth),_mm_mul_pd(exponent,ewclj6));
 644             /* fvdw = 12*C12/r13 - ((6*C6 - f6A)/r6 + f6B)/r */
 645             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);
 646
 647             fscal            = _mm_add_pd(felec,fvdw);
 648
 649             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 650
 651             /* Calculate temporary vectorial force */
 652             tx               = _mm_mul_pd(fscal,dx00);
 653             ty               = _mm_mul_pd(fscal,dy00);
 654             tz               = _mm_mul_pd(fscal,dz00);
 655
 656             /* Update vectorial force */
 657             fix0             = _mm_add_pd(fix0,tx);
 658             fiy0             = _mm_add_pd(fiy0,ty);
 659             fiz0             = _mm_add_pd(fiz0,tz);
 660
 661             gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,tx,ty,tz);
 662
 663             /* Inner loop uses 59 flops */
 664         }
 665
 666         /* End of innermost loop */
 667
 668         gmx_mm_update_iforce_1atom_swizzle_pd(fix0,fiy0,fiz0,
 669                                               f+i_coord_offset,fshift+i_shift_offset);
 670
 671         /* Increment number of inner iterations */
 672         inneriter                  += j_index_end - j_index_start;
 673
 674         /* Outer loop uses 7 flops */
 675     }
 676
 677     /* Increment number of outer iterations */
 678     outeriter        += nri;
 679
 680     /* Update outer/inner flops */
 681
 682     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_F,outeriter*7 + inneriter*59);
 683 }