src/gromacs/gmxlib/nonbonded/nb_kernel_sse2_double/nb_kernel_ElecEwSh_VdwLJEwSh_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 #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_sse2_double.h"
  50 #include "kernelutil_x86_sse2_double.h"
  51
  52 /*
  53  * Gromacs nonbonded kernel:   nb_kernel_ElecEwSh_VdwLJEwSh_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_ElecEwSh_VdwLJEwSh_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     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 134     rcutoff_scalar   = fr->rcoulomb;
 135     rcutoff          = _mm_set1_pd(rcutoff_scalar);
 136     rcutoff2         = _mm_mul_pd(rcutoff,rcutoff);
 137
 138     sh_vdw_invrcut6  = _mm_set1_pd(fr->ic->sh_invrc6);
 139     rvdw             = _mm_set1_pd(fr->rvdw);
 140
 141     /* Avoid stupid compiler warnings */
 142     jnrA = jnrB = 0;
 143     j_coord_offsetA = 0;
 144     j_coord_offsetB = 0;
 145
 146     outeriter        = 0;
 147     inneriter        = 0;
 148
 149     /* Start outer loop over neighborlists */
 150     for(iidx=0; iidx<nri; iidx++)
 151     {
 152         /* Load shift vector for this list */
 153         i_shift_offset   = DIM*shiftidx[iidx];
 154
 155         /* Load limits for loop over neighbors */
 156         j_index_start    = jindex[iidx];
 157         j_index_end      = jindex[iidx+1];
 158
 159         /* Get outer coordinate index */
 160         inr              = iinr[iidx];
 161         i_coord_offset   = DIM*inr;
 162
 163         /* Load i particle coords and add shift vector */
 164         gmx_mm_load_shift_and_1rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 165
 166         fix0             = _mm_setzero_pd();
 167         fiy0             = _mm_setzero_pd();
 168         fiz0             = _mm_setzero_pd();
 169
 170         /* Load parameters for i particles */
 171         iq0              = _mm_mul_pd(facel,_mm_load1_pd(charge+inr+0));
 172         vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 173
 174         /* Reset potential sums */
 175         velecsum         = _mm_setzero_pd();
 176         vvdwsum          = _mm_setzero_pd();
 177
 178         /* Start inner kernel loop */
 179         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 180         {
 181
 182             /* Get j neighbor index, and coordinate index */
 183             jnrA             = jjnr[jidx];
 184             jnrB             = jjnr[jidx+1];
 185             j_coord_offsetA  = DIM*jnrA;
 186             j_coord_offsetB  = DIM*jnrB;
 187
 188             /* load j atom coordinates */
 189             gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 190                                               &jx0,&jy0,&jz0);
 191
 192             /* Calculate displacement vector */
 193             dx00             = _mm_sub_pd(ix0,jx0);
 194             dy00             = _mm_sub_pd(iy0,jy0);
 195             dz00             = _mm_sub_pd(iz0,jz0);
 196
 197             /* Calculate squared distance and things based on it */
 198             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 199
 200             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 201
 202             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 203
 204             /* Load parameters for j particles */
 205             jq0              = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
 206             vdwjidx0A        = 2*vdwtype[jnrA+0];
 207             vdwjidx0B        = 2*vdwtype[jnrB+0];
 208
 209             /**************************
 210              * CALCULATE INTERACTIONS *
 211              **************************/
 212
 213             if (gmx_mm_any_lt(rsq00,rcutoff2))
 214             {
 215
 216             r00              = _mm_mul_pd(rsq00,rinv00);
 217
 218             /* Compute parameters for interactions between i and j atoms */
 219             qq00             = _mm_mul_pd(iq0,jq0);
 220             gmx_mm_load_2pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,
 221                                          vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
 222
 223             c6grid_00       = gmx_mm_load_2real_swizzle_pd(vdwgridparam+vdwioffset0+vdwjidx0A,
 224                                                                vdwgridparam+vdwioffset0+vdwjidx0B);
 225
 226             /* EWALD ELECTROSTATICS */
 227
 228             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 229             ewrt             = _mm_mul_pd(r00,ewtabscale);
 230             ewitab           = _mm_cvttpd_epi32(ewrt);
 231             eweps            = _mm_sub_pd(ewrt,_mm_cvtepi32_pd(ewitab));
 232             ewitab           = _mm_slli_epi32(ewitab,2);
 233             ewtabF           = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 234             ewtabD           = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,1) );
 235             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 236             ewtabV           = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,0) +2);
 237             ewtabFn          = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,1) +2);
 238             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 239             felec            = _mm_add_pd(ewtabF,_mm_mul_pd(eweps,ewtabD));
 240             velec            = _mm_sub_pd(ewtabV,_mm_mul_pd(_mm_mul_pd(ewtabhalfspace,eweps),_mm_add_pd(ewtabF,felec)));
 241             velec            = _mm_mul_pd(qq00,_mm_sub_pd(_mm_sub_pd(rinv00,sh_ewald),velec));
 242             felec            = _mm_mul_pd(_mm_mul_pd(qq00,rinv00),_mm_sub_pd(rinvsq00,felec));
 243
 244             /* Analytical LJ-PME */
 245             rinvsix          = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 246             ewcljrsq         = _mm_mul_pd(ewclj2,rsq00);
 247             ewclj6           = _mm_mul_pd(ewclj2,_mm_mul_pd(ewclj2,ewclj2));
 248             exponent         = gmx_simd_exp_d(ewcljrsq);
 249             /* poly = exp(-(beta*r)^2) * (1 + (beta*r)^2 + (beta*r)^4 /2) */
 250             poly             = _mm_mul_pd(exponent,_mm_add_pd(_mm_sub_pd(one,ewcljrsq),_mm_mul_pd(_mm_mul_pd(ewcljrsq,ewcljrsq),one_half)));
 251             /* vvdw6 = [C6 - C6grid * (1-poly)]/r6 */
 252             vvdw6            = _mm_mul_pd(_mm_sub_pd(c6_00,_mm_mul_pd(c6grid_00,_mm_sub_pd(one,poly))),rinvsix);
 253             vvdw12           = _mm_mul_pd(c12_00,_mm_mul_pd(rinvsix,rinvsix));
 254             vvdw             = _mm_sub_pd(_mm_mul_pd( _mm_sub_pd(vvdw12 , _mm_mul_pd(c12_00,_mm_mul_pd(sh_vdw_invrcut6,sh_vdw_invrcut6))),one_twelfth),
 255                                _mm_mul_pd( _mm_sub_pd(vvdw6,_mm_add_pd(_mm_mul_pd(c6_00,sh_vdw_invrcut6),_mm_mul_pd(c6grid_00,sh_lj_ewald))),one_sixth));
 256             /* fvdw = vvdw12/r - (vvdw6/r + (C6grid * exponent * beta^6)/r) */
 257             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);
 258
 259             cutoff_mask      = _mm_cmplt_pd(rsq00,rcutoff2);
 260
 261             /* Update potential sum for this i atom from the interaction with this j atom. */
 262             velec            = _mm_and_pd(velec,cutoff_mask);
 263             velecsum         = _mm_add_pd(velecsum,velec);
 264             vvdw             = _mm_and_pd(vvdw,cutoff_mask);
 265             vvdwsum          = _mm_add_pd(vvdwsum,vvdw);
 266
 267             fscal            = _mm_add_pd(felec,fvdw);
 268
 269             fscal            = _mm_and_pd(fscal,cutoff_mask);
 270
 271             /* Calculate temporary vectorial force */
 272             tx               = _mm_mul_pd(fscal,dx00);
 273             ty               = _mm_mul_pd(fscal,dy00);
 274             tz               = _mm_mul_pd(fscal,dz00);
 275
 276             /* Update vectorial force */
 277             fix0             = _mm_add_pd(fix0,tx);
 278             fiy0             = _mm_add_pd(fiy0,ty);
 279             fiz0             = _mm_add_pd(fiz0,tz);
 280
 281             gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,tx,ty,tz);
 282
 283             }
 284
 285             /* Inner loop uses 82 flops */
 286         }
 287
 288         if(jidx<j_index_end)
 289         {
 290
 291             jnrA             = jjnr[jidx];
 292             j_coord_offsetA  = DIM*jnrA;
 293
 294             /* load j atom coordinates */
 295             gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
 296                                               &jx0,&jy0,&jz0);
 297
 298             /* Calculate displacement vector */
 299             dx00             = _mm_sub_pd(ix0,jx0);
 300             dy00             = _mm_sub_pd(iy0,jy0);
 301             dz00             = _mm_sub_pd(iz0,jz0);
 302
 303             /* Calculate squared distance and things based on it */
 304             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 305
 306             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 307
 308             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 309
 310             /* Load parameters for j particles */
 311             jq0              = _mm_load_sd(charge+jnrA+0);
 312             vdwjidx0A        = 2*vdwtype[jnrA+0];
 313
 314             /**************************
 315              * CALCULATE INTERACTIONS *
 316              **************************/
 317
 318             if (gmx_mm_any_lt(rsq00,rcutoff2))
 319             {
 320
 321             r00              = _mm_mul_pd(rsq00,rinv00);
 322
 323             /* Compute parameters for interactions between i and j atoms */
 324             qq00             = _mm_mul_pd(iq0,jq0);
 325             gmx_mm_load_1pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,&c6_00,&c12_00);
 326
 327             c6grid_00       = gmx_mm_load_1real_pd(vdwgridparam+vdwioffset0+vdwjidx0A);
 328
 329             /* EWALD ELECTROSTATICS */
 330
 331             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 332             ewrt             = _mm_mul_pd(r00,ewtabscale);
 333             ewitab           = _mm_cvttpd_epi32(ewrt);
 334             eweps            = _mm_sub_pd(ewrt,_mm_cvtepi32_pd(ewitab));
 335             ewitab           = _mm_slli_epi32(ewitab,2);
 336             ewtabF           = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 337             ewtabD           = _mm_setzero_pd();
 338             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 339             ewtabV           = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,0) +2);
 340             ewtabFn          = _mm_setzero_pd();
 341             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 342             felec            = _mm_add_pd(ewtabF,_mm_mul_pd(eweps,ewtabD));
 343             velec            = _mm_sub_pd(ewtabV,_mm_mul_pd(_mm_mul_pd(ewtabhalfspace,eweps),_mm_add_pd(ewtabF,felec)));
 344             velec            = _mm_mul_pd(qq00,_mm_sub_pd(_mm_sub_pd(rinv00,sh_ewald),velec));
 345             felec            = _mm_mul_pd(_mm_mul_pd(qq00,rinv00),_mm_sub_pd(rinvsq00,felec));
 346
 347             /* Analytical LJ-PME */
 348             rinvsix          = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 349             ewcljrsq         = _mm_mul_pd(ewclj2,rsq00);
 350             ewclj6           = _mm_mul_pd(ewclj2,_mm_mul_pd(ewclj2,ewclj2));
 351             exponent         = gmx_simd_exp_d(ewcljrsq);
 352             /* poly = exp(-(beta*r)^2) * (1 + (beta*r)^2 + (beta*r)^4 /2) */
 353             poly             = _mm_mul_pd(exponent,_mm_add_pd(_mm_sub_pd(one,ewcljrsq),_mm_mul_pd(_mm_mul_pd(ewcljrsq,ewcljrsq),one_half)));
 354             /* vvdw6 = [C6 - C6grid * (1-poly)]/r6 */
 355             vvdw6            = _mm_mul_pd(_mm_sub_pd(c6_00,_mm_mul_pd(c6grid_00,_mm_sub_pd(one,poly))),rinvsix);
 356             vvdw12           = _mm_mul_pd(c12_00,_mm_mul_pd(rinvsix,rinvsix));
 357             vvdw             = _mm_sub_pd(_mm_mul_pd( _mm_sub_pd(vvdw12 , _mm_mul_pd(c12_00,_mm_mul_pd(sh_vdw_invrcut6,sh_vdw_invrcut6))),one_twelfth),
 358                                _mm_mul_pd( _mm_sub_pd(vvdw6,_mm_add_pd(_mm_mul_pd(c6_00,sh_vdw_invrcut6),_mm_mul_pd(c6grid_00,sh_lj_ewald))),one_sixth));
 359             /* fvdw = vvdw12/r - (vvdw6/r + (C6grid * exponent * beta^6)/r) */
 360             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);
 361
 362             cutoff_mask      = _mm_cmplt_pd(rsq00,rcutoff2);
 363
 364             /* Update potential sum for this i atom from the interaction with this j atom. */
 365             velec            = _mm_and_pd(velec,cutoff_mask);
 366             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 367             velecsum         = _mm_add_pd(velecsum,velec);
 368             vvdw             = _mm_and_pd(vvdw,cutoff_mask);
 369             vvdw             = _mm_unpacklo_pd(vvdw,_mm_setzero_pd());
 370             vvdwsum          = _mm_add_pd(vvdwsum,vvdw);
 371
 372             fscal            = _mm_add_pd(felec,fvdw);
 373
 374             fscal            = _mm_and_pd(fscal,cutoff_mask);
 375
 376             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 377
 378             /* Calculate temporary vectorial force */
 379             tx               = _mm_mul_pd(fscal,dx00);
 380             ty               = _mm_mul_pd(fscal,dy00);
 381             tz               = _mm_mul_pd(fscal,dz00);
 382
 383             /* Update vectorial force */
 384             fix0             = _mm_add_pd(fix0,tx);
 385             fiy0             = _mm_add_pd(fiy0,ty);
 386             fiz0             = _mm_add_pd(fiz0,tz);
 387
 388             gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,tx,ty,tz);
 389
 390             }
 391
 392             /* Inner loop uses 82 flops */
 393         }
 394
 395         /* End of innermost loop */
 396
 397         gmx_mm_update_iforce_1atom_swizzle_pd(fix0,fiy0,fiz0,
 398                                               f+i_coord_offset,fshift+i_shift_offset);
 399
 400         ggid                        = gid[iidx];
 401         /* Update potential energies */
 402         gmx_mm_update_1pot_pd(velecsum,kernel_data->energygrp_elec+ggid);
 403         gmx_mm_update_1pot_pd(vvdwsum,kernel_data->energygrp_vdw+ggid);
 404
 405         /* Increment number of inner iterations */
 406         inneriter                  += j_index_end - j_index_start;
 407
 408         /* Outer loop uses 9 flops */
 409     }
 410
 411     /* Increment number of outer iterations */
 412     outeriter        += nri;
 413
 414     /* Update outer/inner flops */
 415
 416     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_VF,outeriter*9 + inneriter*82);
 417 }
 418 /*
 419  * Gromacs nonbonded kernel:   nb_kernel_ElecEwSh_VdwLJEwSh_GeomP1P1_F_sse2_double
 420  * Electrostatics interaction: Ewald
 421  * VdW interaction:            LJEwald
 422  * Geometry:                   Particle-Particle
 423  * Calculate force/pot:        Force
 424  */
 425 void
 426 nb_kernel_ElecEwSh_VdwLJEwSh_GeomP1P1_F_sse2_double
 427                     (t_nblist                    * gmx_restrict       nlist,
 428                      rvec                        * gmx_restrict          xx,
 429                      rvec                        * gmx_restrict          ff,
 430                      t_forcerec                  * gmx_restrict          fr,
 431                      t_mdatoms                   * gmx_restrict     mdatoms,
 432                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 433                      t_nrnb                      * gmx_restrict        nrnb)
 434 {
 435     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 436      * just 0 for non-waters.
 437      * Suffixes A,B refer to j loop unrolling done with SSE double precision, e.g. for the two different
 438      * jnr indices corresponding to data put in the four positions in the SIMD register.
 439      */
 440     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 441     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 442     int              jnrA,jnrB;
 443     int              j_coord_offsetA,j_coord_offsetB;
 444     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 445     real             rcutoff_scalar;
 446     real             *shiftvec,*fshift,*x,*f;
 447     __m128d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 448     int              vdwioffset0;
 449     __m128d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 450     int              vdwjidx0A,vdwjidx0B;
 451     __m128d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 452     __m128d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 453     __m128d          velec,felec,velecsum,facel,crf,krf,krf2;
 454     real             *charge;
 455     int              nvdwtype;
 456     __m128d          rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 457     int              *vdwtype;
 458     real             *vdwparam;
 459     __m128d          one_sixth   = _mm_set1_pd(1.0/6.0);
 460     __m128d          one_twelfth = _mm_set1_pd(1.0/12.0);
 461     __m128d           c6grid_00;
 462     __m128d           ewclj,ewclj2,ewclj6,ewcljrsq,poly,exponent,f6A,f6B,sh_lj_ewald;
 463     real             *vdwgridparam;
 464     __m128d           one_half = _mm_set1_pd(0.5);
 465     __m128d           minus_one = _mm_set1_pd(-1.0);
 466     __m128i          ewitab;
 467     __m128d          ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
 468     real             *ewtab;
 469     __m128d          dummy_mask,cutoff_mask;
 470     __m128d          signbit   = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
 471     __m128d          one     = _mm_set1_pd(1.0);
 472     __m128d          two     = _mm_set1_pd(2.0);
 473     x                = xx[0];
 474     f                = ff[0];
 475
 476     nri              = nlist->nri;
 477     iinr             = nlist->iinr;
 478     jindex           = nlist->jindex;
 479     jjnr             = nlist->jjnr;
 480     shiftidx         = nlist->shift;
 481     gid              = nlist->gid;
 482     shiftvec         = fr->shift_vec[0];
 483     fshift           = fr->fshift[0];
 484     facel            = _mm_set1_pd(fr->epsfac);
 485     charge           = mdatoms->chargeA;
 486     nvdwtype         = fr->ntype;
 487     vdwparam         = fr->nbfp;
 488     vdwtype          = mdatoms->typeA;
 489     vdwgridparam     = fr->ljpme_c6grid;
 490     sh_lj_ewald      = _mm_set1_pd(fr->ic->sh_lj_ewald);
 491     ewclj            = _mm_set1_pd(fr->ewaldcoeff_lj);
 492     ewclj2           = _mm_mul_pd(minus_one,_mm_mul_pd(ewclj,ewclj));
 493
 494     sh_ewald         = _mm_set1_pd(fr->ic->sh_ewald);
 495     ewtab            = fr->ic->tabq_coul_F;
 496     ewtabscale       = _mm_set1_pd(fr->ic->tabq_scale);
 497     ewtabhalfspace   = _mm_set1_pd(0.5/fr->ic->tabq_scale);
 498
 499     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 500     rcutoff_scalar   = fr->rcoulomb;
 501     rcutoff          = _mm_set1_pd(rcutoff_scalar);
 502     rcutoff2         = _mm_mul_pd(rcutoff,rcutoff);
 503
 504     sh_vdw_invrcut6  = _mm_set1_pd(fr->ic->sh_invrc6);
 505     rvdw             = _mm_set1_pd(fr->rvdw);
 506
 507     /* Avoid stupid compiler warnings */
 508     jnrA = jnrB = 0;
 509     j_coord_offsetA = 0;
 510     j_coord_offsetB = 0;
 511
 512     outeriter        = 0;
 513     inneriter        = 0;
 514
 515     /* Start outer loop over neighborlists */
 516     for(iidx=0; iidx<nri; iidx++)
 517     {
 518         /* Load shift vector for this list */
 519         i_shift_offset   = DIM*shiftidx[iidx];
 520
 521         /* Load limits for loop over neighbors */
 522         j_index_start    = jindex[iidx];
 523         j_index_end      = jindex[iidx+1];
 524
 525         /* Get outer coordinate index */
 526         inr              = iinr[iidx];
 527         i_coord_offset   = DIM*inr;
 528
 529         /* Load i particle coords and add shift vector */
 530         gmx_mm_load_shift_and_1rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 531
 532         fix0             = _mm_setzero_pd();
 533         fiy0             = _mm_setzero_pd();
 534         fiz0             = _mm_setzero_pd();
 535
 536         /* Load parameters for i particles */
 537         iq0              = _mm_mul_pd(facel,_mm_load1_pd(charge+inr+0));
 538         vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 539
 540         /* Start inner kernel loop */
 541         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 542         {
 543
 544             /* Get j neighbor index, and coordinate index */
 545             jnrA             = jjnr[jidx];
 546             jnrB             = jjnr[jidx+1];
 547             j_coord_offsetA  = DIM*jnrA;
 548             j_coord_offsetB  = DIM*jnrB;
 549
 550             /* load j atom coordinates */
 551             gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 552                                               &jx0,&jy0,&jz0);
 553
 554             /* Calculate displacement vector */
 555             dx00             = _mm_sub_pd(ix0,jx0);
 556             dy00             = _mm_sub_pd(iy0,jy0);
 557             dz00             = _mm_sub_pd(iz0,jz0);
 558
 559             /* Calculate squared distance and things based on it */
 560             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 561
 562             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 563
 564             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 565
 566             /* Load parameters for j particles */
 567             jq0              = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
 568             vdwjidx0A        = 2*vdwtype[jnrA+0];
 569             vdwjidx0B        = 2*vdwtype[jnrB+0];
 570
 571             /**************************
 572              * CALCULATE INTERACTIONS *
 573              **************************/
 574
 575             if (gmx_mm_any_lt(rsq00,rcutoff2))
 576             {
 577
 578             r00              = _mm_mul_pd(rsq00,rinv00);
 579
 580             /* Compute parameters for interactions between i and j atoms */
 581             qq00             = _mm_mul_pd(iq0,jq0);
 582             gmx_mm_load_2pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,
 583                                          vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
 584
 585             c6grid_00       = gmx_mm_load_2real_swizzle_pd(vdwgridparam+vdwioffset0+vdwjidx0A,
 586                                                                vdwgridparam+vdwioffset0+vdwjidx0B);
 587
 588             /* EWALD ELECTROSTATICS */
 589
 590             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 591             ewrt             = _mm_mul_pd(r00,ewtabscale);
 592             ewitab           = _mm_cvttpd_epi32(ewrt);
 593             eweps            = _mm_sub_pd(ewrt,_mm_cvtepi32_pd(ewitab));
 594             gmx_mm_load_2pair_swizzle_pd(ewtab+gmx_mm_extract_epi32(ewitab,0),ewtab+gmx_mm_extract_epi32(ewitab,1),
 595                                          &ewtabF,&ewtabFn);
 596             felec            = _mm_add_pd(_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF),_mm_mul_pd(eweps,ewtabFn));
 597             felec            = _mm_mul_pd(_mm_mul_pd(qq00,rinv00),_mm_sub_pd(rinvsq00,felec));
 598
 599             /* Analytical LJ-PME */
 600             rinvsix          = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 601             ewcljrsq         = _mm_mul_pd(ewclj2,rsq00);
 602             ewclj6           = _mm_mul_pd(ewclj2,_mm_mul_pd(ewclj2,ewclj2));
 603             exponent         = gmx_simd_exp_d(ewcljrsq);
 604             /* poly = exp(-(beta*r)^2) * (1 + (beta*r)^2 + (beta*r)^4 /2) */
 605             poly             = _mm_mul_pd(exponent,_mm_add_pd(_mm_sub_pd(one,ewcljrsq),_mm_mul_pd(_mm_mul_pd(ewcljrsq,ewcljrsq),one_half)));
 606             /* f6A = 6 * C6grid * (1 - poly) */
 607             f6A              = _mm_mul_pd(c6grid_00,_mm_sub_pd(one,poly));
 608             /* f6B = C6grid * exponent * beta^6 */
 609             f6B              = _mm_mul_pd(_mm_mul_pd(c6grid_00,one_sixth),_mm_mul_pd(exponent,ewclj6));
 610             /* fvdw = 12*C12/r13 - ((6*C6 - f6A)/r6 + f6B)/r */
 611             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);
 612
 613             cutoff_mask      = _mm_cmplt_pd(rsq00,rcutoff2);
 614
 615             fscal            = _mm_add_pd(felec,fvdw);
 616
 617             fscal            = _mm_and_pd(fscal,cutoff_mask);
 618
 619             /* Calculate temporary vectorial force */
 620             tx               = _mm_mul_pd(fscal,dx00);
 621             ty               = _mm_mul_pd(fscal,dy00);
 622             tz               = _mm_mul_pd(fscal,dz00);
 623
 624             /* Update vectorial force */
 625             fix0             = _mm_add_pd(fix0,tx);
 626             fiy0             = _mm_add_pd(fiy0,ty);
 627             fiz0             = _mm_add_pd(fiz0,tz);
 628
 629             gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,tx,ty,tz);
 630
 631             }
 632
 633             /* Inner loop uses 62 flops */
 634         }
 635
 636         if(jidx<j_index_end)
 637         {
 638
 639             jnrA             = jjnr[jidx];
 640             j_coord_offsetA  = DIM*jnrA;
 641
 642             /* load j atom coordinates */
 643             gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
 644                                               &jx0,&jy0,&jz0);
 645
 646             /* Calculate displacement vector */
 647             dx00             = _mm_sub_pd(ix0,jx0);
 648             dy00             = _mm_sub_pd(iy0,jy0);
 649             dz00             = _mm_sub_pd(iz0,jz0);
 650
 651             /* Calculate squared distance and things based on it */
 652             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 653
 654             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 655
 656             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 657
 658             /* Load parameters for j particles */
 659             jq0              = _mm_load_sd(charge+jnrA+0);
 660             vdwjidx0A        = 2*vdwtype[jnrA+0];
 661
 662             /**************************
 663              * CALCULATE INTERACTIONS *
 664              **************************/
 665
 666             if (gmx_mm_any_lt(rsq00,rcutoff2))
 667             {
 668
 669             r00              = _mm_mul_pd(rsq00,rinv00);
 670
 671             /* Compute parameters for interactions between i and j atoms */
 672             qq00             = _mm_mul_pd(iq0,jq0);
 673             gmx_mm_load_1pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,&c6_00,&c12_00);
 674
 675             c6grid_00       = gmx_mm_load_1real_pd(vdwgridparam+vdwioffset0+vdwjidx0A);
 676
 677             /* EWALD ELECTROSTATICS */
 678
 679             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 680             ewrt             = _mm_mul_pd(r00,ewtabscale);
 681             ewitab           = _mm_cvttpd_epi32(ewrt);
 682             eweps            = _mm_sub_pd(ewrt,_mm_cvtepi32_pd(ewitab));
 683             gmx_mm_load_1pair_swizzle_pd(ewtab+gmx_mm_extract_epi32(ewitab,0),&ewtabF,&ewtabFn);
 684             felec            = _mm_add_pd(_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF),_mm_mul_pd(eweps,ewtabFn));
 685             felec            = _mm_mul_pd(_mm_mul_pd(qq00,rinv00),_mm_sub_pd(rinvsq00,felec));
 686
 687             /* Analytical LJ-PME */
 688             rinvsix          = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 689             ewcljrsq         = _mm_mul_pd(ewclj2,rsq00);
 690             ewclj6           = _mm_mul_pd(ewclj2,_mm_mul_pd(ewclj2,ewclj2));
 691             exponent         = gmx_simd_exp_d(ewcljrsq);
 692             /* poly = exp(-(beta*r)^2) * (1 + (beta*r)^2 + (beta*r)^4 /2) */
 693             poly             = _mm_mul_pd(exponent,_mm_add_pd(_mm_sub_pd(one,ewcljrsq),_mm_mul_pd(_mm_mul_pd(ewcljrsq,ewcljrsq),one_half)));
 694             /* f6A = 6 * C6grid * (1 - poly) */
 695             f6A              = _mm_mul_pd(c6grid_00,_mm_sub_pd(one,poly));
 696             /* f6B = C6grid * exponent * beta^6 */
 697             f6B              = _mm_mul_pd(_mm_mul_pd(c6grid_00,one_sixth),_mm_mul_pd(exponent,ewclj6));
 698             /* fvdw = 12*C12/r13 - ((6*C6 - f6A)/r6 + f6B)/r */
 699             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);
 700
 701             cutoff_mask      = _mm_cmplt_pd(rsq00,rcutoff2);
 702
 703             fscal            = _mm_add_pd(felec,fvdw);
 704
 705             fscal            = _mm_and_pd(fscal,cutoff_mask);
 706
 707             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 708
 709             /* Calculate temporary vectorial force */
 710             tx               = _mm_mul_pd(fscal,dx00);
 711             ty               = _mm_mul_pd(fscal,dy00);
 712             tz               = _mm_mul_pd(fscal,dz00);
 713
 714             /* Update vectorial force */
 715             fix0             = _mm_add_pd(fix0,tx);
 716             fiy0             = _mm_add_pd(fiy0,ty);
 717             fiz0             = _mm_add_pd(fiz0,tz);
 718
 719             gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,tx,ty,tz);
 720
 721             }
 722
 723             /* Inner loop uses 62 flops */
 724         }
 725
 726         /* End of innermost loop */
 727
 728         gmx_mm_update_iforce_1atom_swizzle_pd(fix0,fiy0,fiz0,
 729                                               f+i_coord_offset,fshift+i_shift_offset);
 730
 731         /* Increment number of inner iterations */
 732         inneriter                  += j_index_end - j_index_start;
 733
 734         /* Outer loop uses 7 flops */
 735     }
 736
 737     /* Increment number of outer iterations */
 738     outeriter        += nri;
 739
 740     /* Update outer/inner flops */
 741
 742     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_F,outeriter*7 + inneriter*62);
 743 }