src/gromacs/gmxlib/nonbonded/nb_kernel_sse4_1_double/nb_kernel_ElecEwSh_VdwLJEwSh_GeomP1P1_sse4_1_double.c

   1 /*
   2  * This file is part of the GROMACS molecular simulation package.
   3  *
   4  * Copyright (c) 2012,2013,2014,2015,2017, 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 #include "gmxpre.h"
  39
  40 #include "config.h"
  41
  42 #include <math.h>
  43
  44 #include "../nb_kernel.h"
  45 #include "gromacs/gmxlib/nrnb.h"
  46
  47 #include "kernelutil_x86_sse4_1_double.h"
  48
  49 /*
  50  * Gromacs nonbonded kernel:   nb_kernel_ElecEwSh_VdwLJEwSh_GeomP1P1_VF_sse4_1_double
  51  * Electrostatics interaction: Ewald
  52  * VdW interaction:            LJEwald
  53  * Geometry:                   Particle-Particle
  54  * Calculate force/pot:        PotentialAndForce
  55  */
  56 void
  57 nb_kernel_ElecEwSh_VdwLJEwSh_GeomP1P1_VF_sse4_1_double
  58                     (t_nblist                    * gmx_restrict       nlist,
  59                      rvec                        * gmx_restrict          xx,
  60                      rvec                        * gmx_restrict          ff,
  61                      struct t_forcerec           * gmx_restrict          fr,
  62                      t_mdatoms                   * gmx_restrict     mdatoms,
  63                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
  64                      t_nrnb                      * gmx_restrict        nrnb)
  65 {
  66     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
  67      * just 0 for non-waters.
  68      * Suffixes A,B refer to j loop unrolling done with SSE double precision, e.g. for the two different
  69      * jnr indices corresponding to data put in the four positions in the SIMD register.
  70      */
  71     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
  72     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
  73     int              jnrA,jnrB;
  74     int              j_coord_offsetA,j_coord_offsetB;
  75     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
  76     real             rcutoff_scalar;
  77     real             *shiftvec,*fshift,*x,*f;
  78     __m128d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
  79     int              vdwioffset0;
  80     __m128d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
  81     int              vdwjidx0A,vdwjidx0B;
  82     __m128d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  83     __m128d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  84     __m128d          velec,felec,velecsum,facel,crf,krf,krf2;
  85     real             *charge;
  86     int              nvdwtype;
  87     __m128d          rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
  88     int              *vdwtype;
  89     real             *vdwparam;
  90     __m128d          one_sixth   = _mm_set1_pd(1.0/6.0);
  91     __m128d          one_twelfth = _mm_set1_pd(1.0/12.0);
  92     __m128d           c6grid_00;
  93     __m128d           ewclj,ewclj2,ewclj6,ewcljrsq,poly,exponent,f6A,f6B,sh_lj_ewald;
  94     real             *vdwgridparam;
  95     __m128d           one_half = _mm_set1_pd(0.5);
  96     __m128d           minus_one = _mm_set1_pd(-1.0);
  97     __m128i          ewitab;
  98     __m128d          ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
  99     real             *ewtab;
 100     __m128d          dummy_mask,cutoff_mask;
 101     __m128d          signbit   = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
 102     __m128d          one     = _mm_set1_pd(1.0);
 103     __m128d          two     = _mm_set1_pd(2.0);
 104     x                = xx[0];
 105     f                = ff[0];
 106
 107     nri              = nlist->nri;
 108     iinr             = nlist->iinr;
 109     jindex           = nlist->jindex;
 110     jjnr             = nlist->jjnr;
 111     shiftidx         = nlist->shift;
 112     gid              = nlist->gid;
 113     shiftvec         = fr->shift_vec[0];
 114     fshift           = fr->fshift[0];
 115     facel            = _mm_set1_pd(fr->ic->epsfac);
 116     charge           = mdatoms->chargeA;
 117     nvdwtype         = fr->ntype;
 118     vdwparam         = fr->nbfp;
 119     vdwtype          = mdatoms->typeA;
 120     vdwgridparam     = fr->ljpme_c6grid;
 121     sh_lj_ewald      = _mm_set1_pd(fr->ic->sh_lj_ewald);
 122     ewclj            = _mm_set1_pd(fr->ic->ewaldcoeff_lj);
 123     ewclj2           = _mm_mul_pd(minus_one,_mm_mul_pd(ewclj,ewclj));
 124
 125     sh_ewald         = _mm_set1_pd(fr->ic->sh_ewald);
 126     ewtab            = fr->ic->tabq_coul_FDV0;
 127     ewtabscale       = _mm_set1_pd(fr->ic->tabq_scale);
 128     ewtabhalfspace   = _mm_set1_pd(0.5/fr->ic->tabq_scale);
 129
 130     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 131     rcutoff_scalar   = fr->ic->rcoulomb;
 132     rcutoff          = _mm_set1_pd(rcutoff_scalar);
 133     rcutoff2         = _mm_mul_pd(rcutoff,rcutoff);
 134
 135     sh_vdw_invrcut6  = _mm_set1_pd(fr->ic->sh_invrc6);
 136     rvdw             = _mm_set1_pd(fr->ic->rvdw);
 137
 138     /* Avoid stupid compiler warnings */
 139     jnrA = jnrB = 0;
 140     j_coord_offsetA = 0;
 141     j_coord_offsetB = 0;
 142
 143     outeriter        = 0;
 144     inneriter        = 0;
 145
 146     /* Start outer loop over neighborlists */
 147     for(iidx=0; iidx<nri; iidx++)
 148     {
 149         /* Load shift vector for this list */
 150         i_shift_offset   = DIM*shiftidx[iidx];
 151
 152         /* Load limits for loop over neighbors */
 153         j_index_start    = jindex[iidx];
 154         j_index_end      = jindex[iidx+1];
 155
 156         /* Get outer coordinate index */
 157         inr              = iinr[iidx];
 158         i_coord_offset   = DIM*inr;
 159
 160         /* Load i particle coords and add shift vector */
 161         gmx_mm_load_shift_and_1rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 162
 163         fix0             = _mm_setzero_pd();
 164         fiy0             = _mm_setzero_pd();
 165         fiz0             = _mm_setzero_pd();
 166
 167         /* Load parameters for i particles */
 168         iq0              = _mm_mul_pd(facel,_mm_load1_pd(charge+inr+0));
 169         vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 170
 171         /* Reset potential sums */
 172         velecsum         = _mm_setzero_pd();
 173         vvdwsum          = _mm_setzero_pd();
 174
 175         /* Start inner kernel loop */
 176         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 177         {
 178
 179             /* Get j neighbor index, and coordinate index */
 180             jnrA             = jjnr[jidx];
 181             jnrB             = jjnr[jidx+1];
 182             j_coord_offsetA  = DIM*jnrA;
 183             j_coord_offsetB  = DIM*jnrB;
 184
 185             /* load j atom coordinates */
 186             gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 187                                               &jx0,&jy0,&jz0);
 188
 189             /* Calculate displacement vector */
 190             dx00             = _mm_sub_pd(ix0,jx0);
 191             dy00             = _mm_sub_pd(iy0,jy0);
 192             dz00             = _mm_sub_pd(iz0,jz0);
 193
 194             /* Calculate squared distance and things based on it */
 195             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 196
 197             rinv00           = sse41_invsqrt_d(rsq00);
 198
 199             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 200
 201             /* Load parameters for j particles */
 202             jq0              = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
 203             vdwjidx0A        = 2*vdwtype[jnrA+0];
 204             vdwjidx0B        = 2*vdwtype[jnrB+0];
 205
 206             /**************************
 207              * CALCULATE INTERACTIONS *
 208              **************************/
 209
 210             if (gmx_mm_any_lt(rsq00,rcutoff2))
 211             {
 212
 213             r00              = _mm_mul_pd(rsq00,rinv00);
 214
 215             /* Compute parameters for interactions between i and j atoms */
 216             qq00             = _mm_mul_pd(iq0,jq0);
 217             gmx_mm_load_2pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,
 218                                          vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
 219             c6grid_00       = gmx_mm_load_2real_swizzle_pd(vdwgridparam+vdwioffset0+vdwjidx0A,
 220                                                                vdwgridparam+vdwioffset0+vdwjidx0B);
 221
 222             /* EWALD ELECTROSTATICS */
 223
 224             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 225             ewrt             = _mm_mul_pd(r00,ewtabscale);
 226             ewitab           = _mm_cvttpd_epi32(ewrt);
 227             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 228             ewitab           = _mm_slli_epi32(ewitab,2);
 229             ewtabF           = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 230             ewtabD           = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,1) );
 231             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 232             ewtabV           = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,0) +2);
 233             ewtabFn          = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,1) +2);
 234             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 235             felec            = _mm_add_pd(ewtabF,_mm_mul_pd(eweps,ewtabD));
 236             velec            = _mm_sub_pd(ewtabV,_mm_mul_pd(_mm_mul_pd(ewtabhalfspace,eweps),_mm_add_pd(ewtabF,felec)));
 237             velec            = _mm_mul_pd(qq00,_mm_sub_pd(_mm_sub_pd(rinv00,sh_ewald),velec));
 238             felec            = _mm_mul_pd(_mm_mul_pd(qq00,rinv00),_mm_sub_pd(rinvsq00,felec));
 239
 240             /* Analytical LJ-PME */
 241             rinvsix          = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 242             ewcljrsq         = _mm_mul_pd(ewclj2,rsq00);
 243             ewclj6           = _mm_mul_pd(ewclj2,_mm_mul_pd(ewclj2,ewclj2));
 244             exponent         = sse41_exp_d(ewcljrsq);
 245             /* poly = exp(-(beta*r)^2) * (1 + (beta*r)^2 + (beta*r)^4 /2) */
 246             poly             = _mm_mul_pd(exponent,_mm_add_pd(_mm_sub_pd(one,ewcljrsq),_mm_mul_pd(_mm_mul_pd(ewcljrsq,ewcljrsq),one_half)));
 247             /* vvdw6 = [C6 - C6grid * (1-poly)]/r6 */
 248             vvdw6            = _mm_mul_pd(_mm_sub_pd(c6_00,_mm_mul_pd(c6grid_00,_mm_sub_pd(one,poly))),rinvsix);
 249             vvdw12           = _mm_mul_pd(c12_00,_mm_mul_pd(rinvsix,rinvsix));
 250             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),
 251                                _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));
 252             /* fvdw = vvdw12/r - (vvdw6/r + (C6grid * exponent * beta^6)/r) */
 253             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);
 254
 255             cutoff_mask      = _mm_cmplt_pd(rsq00,rcutoff2);
 256
 257             /* Update potential sum for this i atom from the interaction with this j atom. */
 258             velec            = _mm_and_pd(velec,cutoff_mask);
 259             velecsum         = _mm_add_pd(velecsum,velec);
 260             vvdw             = _mm_and_pd(vvdw,cutoff_mask);
 261             vvdwsum          = _mm_add_pd(vvdwsum,vvdw);
 262
 263             fscal            = _mm_add_pd(felec,fvdw);
 264
 265             fscal            = _mm_and_pd(fscal,cutoff_mask);
 266
 267             /* Calculate temporary vectorial force */
 268             tx               = _mm_mul_pd(fscal,dx00);
 269             ty               = _mm_mul_pd(fscal,dy00);
 270             tz               = _mm_mul_pd(fscal,dz00);
 271
 272             /* Update vectorial force */
 273             fix0             = _mm_add_pd(fix0,tx);
 274             fiy0             = _mm_add_pd(fiy0,ty);
 275             fiz0             = _mm_add_pd(fiz0,tz);
 276
 277             gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,tx,ty,tz);
 278
 279             }
 280
 281             /* Inner loop uses 81 flops */
 282         }
 283
 284         if(jidx<j_index_end)
 285         {
 286
 287             jnrA             = jjnr[jidx];
 288             j_coord_offsetA  = DIM*jnrA;
 289
 290             /* load j atom coordinates */
 291             gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
 292                                               &jx0,&jy0,&jz0);
 293
 294             /* Calculate displacement vector */
 295             dx00             = _mm_sub_pd(ix0,jx0);
 296             dy00             = _mm_sub_pd(iy0,jy0);
 297             dz00             = _mm_sub_pd(iz0,jz0);
 298
 299             /* Calculate squared distance and things based on it */
 300             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 301
 302             rinv00           = sse41_invsqrt_d(rsq00);
 303
 304             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 305
 306             /* Load parameters for j particles */
 307             jq0              = _mm_load_sd(charge+jnrA+0);
 308             vdwjidx0A        = 2*vdwtype[jnrA+0];
 309
 310             /**************************
 311              * CALCULATE INTERACTIONS *
 312              **************************/
 313
 314             if (gmx_mm_any_lt(rsq00,rcutoff2))
 315             {
 316
 317             r00              = _mm_mul_pd(rsq00,rinv00);
 318
 319             /* Compute parameters for interactions between i and j atoms */
 320             qq00             = _mm_mul_pd(iq0,jq0);
 321             gmx_mm_load_1pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,&c6_00,&c12_00);
 322
 323             c6grid_00       = gmx_mm_load_1real_pd(vdwgridparam+vdwioffset0+vdwjidx0A);
 324
 325             /* EWALD ELECTROSTATICS */
 326
 327             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 328             ewrt             = _mm_mul_pd(r00,ewtabscale);
 329             ewitab           = _mm_cvttpd_epi32(ewrt);
 330             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 331             ewitab           = _mm_slli_epi32(ewitab,2);
 332             ewtabF           = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 333             ewtabD           = _mm_setzero_pd();
 334             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 335             ewtabV           = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,0) +2);
 336             ewtabFn          = _mm_setzero_pd();
 337             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 338             felec            = _mm_add_pd(ewtabF,_mm_mul_pd(eweps,ewtabD));
 339             velec            = _mm_sub_pd(ewtabV,_mm_mul_pd(_mm_mul_pd(ewtabhalfspace,eweps),_mm_add_pd(ewtabF,felec)));
 340             velec            = _mm_mul_pd(qq00,_mm_sub_pd(_mm_sub_pd(rinv00,sh_ewald),velec));
 341             felec            = _mm_mul_pd(_mm_mul_pd(qq00,rinv00),_mm_sub_pd(rinvsq00,felec));
 342
 343             /* Analytical LJ-PME */
 344             rinvsix          = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 345             ewcljrsq         = _mm_mul_pd(ewclj2,rsq00);
 346             ewclj6           = _mm_mul_pd(ewclj2,_mm_mul_pd(ewclj2,ewclj2));
 347             exponent         = sse41_exp_d(ewcljrsq);
 348             /* poly = exp(-(beta*r)^2) * (1 + (beta*r)^2 + (beta*r)^4 /2) */
 349             poly             = _mm_mul_pd(exponent,_mm_add_pd(_mm_sub_pd(one,ewcljrsq),_mm_mul_pd(_mm_mul_pd(ewcljrsq,ewcljrsq),one_half)));
 350             /* vvdw6 = [C6 - C6grid * (1-poly)]/r6 */
 351             vvdw6            = _mm_mul_pd(_mm_sub_pd(c6_00,_mm_mul_pd(c6grid_00,_mm_sub_pd(one,poly))),rinvsix);
 352             vvdw12           = _mm_mul_pd(c12_00,_mm_mul_pd(rinvsix,rinvsix));
 353             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),
 354                                _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));
 355             /* fvdw = vvdw12/r - (vvdw6/r + (C6grid * exponent * beta^6)/r) */
 356             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);
 357
 358             cutoff_mask      = _mm_cmplt_pd(rsq00,rcutoff2);
 359
 360             /* Update potential sum for this i atom from the interaction with this j atom. */
 361             velec            = _mm_and_pd(velec,cutoff_mask);
 362             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 363             velecsum         = _mm_add_pd(velecsum,velec);
 364             vvdw             = _mm_and_pd(vvdw,cutoff_mask);
 365             vvdw             = _mm_unpacklo_pd(vvdw,_mm_setzero_pd());
 366             vvdwsum          = _mm_add_pd(vvdwsum,vvdw);
 367
 368             fscal            = _mm_add_pd(felec,fvdw);
 369
 370             fscal            = _mm_and_pd(fscal,cutoff_mask);
 371
 372             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 373
 374             /* Calculate temporary vectorial force */
 375             tx               = _mm_mul_pd(fscal,dx00);
 376             ty               = _mm_mul_pd(fscal,dy00);
 377             tz               = _mm_mul_pd(fscal,dz00);
 378
 379             /* Update vectorial force */
 380             fix0             = _mm_add_pd(fix0,tx);
 381             fiy0             = _mm_add_pd(fiy0,ty);
 382             fiz0             = _mm_add_pd(fiz0,tz);
 383
 384             gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,tx,ty,tz);
 385
 386             }
 387
 388             /* Inner loop uses 81 flops */
 389         }
 390
 391         /* End of innermost loop */
 392
 393         gmx_mm_update_iforce_1atom_swizzle_pd(fix0,fiy0,fiz0,
 394                                               f+i_coord_offset,fshift+i_shift_offset);
 395
 396         ggid                        = gid[iidx];
 397         /* Update potential energies */
 398         gmx_mm_update_1pot_pd(velecsum,kernel_data->energygrp_elec+ggid);
 399         gmx_mm_update_1pot_pd(vvdwsum,kernel_data->energygrp_vdw+ggid);
 400
 401         /* Increment number of inner iterations */
 402         inneriter                  += j_index_end - j_index_start;
 403
 404         /* Outer loop uses 9 flops */
 405     }
 406
 407     /* Increment number of outer iterations */
 408     outeriter        += nri;
 409
 410     /* Update outer/inner flops */
 411
 412     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_VF,outeriter*9 + inneriter*81);
 413 }
 414 /*
 415  * Gromacs nonbonded kernel:   nb_kernel_ElecEwSh_VdwLJEwSh_GeomP1P1_F_sse4_1_double
 416  * Electrostatics interaction: Ewald
 417  * VdW interaction:            LJEwald
 418  * Geometry:                   Particle-Particle
 419  * Calculate force/pot:        Force
 420  */
 421 void
 422 nb_kernel_ElecEwSh_VdwLJEwSh_GeomP1P1_F_sse4_1_double
 423                     (t_nblist                    * gmx_restrict       nlist,
 424                      rvec                        * gmx_restrict          xx,
 425                      rvec                        * gmx_restrict          ff,
 426                      struct t_forcerec           * gmx_restrict          fr,
 427                      t_mdatoms                   * gmx_restrict     mdatoms,
 428                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 429                      t_nrnb                      * gmx_restrict        nrnb)
 430 {
 431     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 432      * just 0 for non-waters.
 433      * Suffixes A,B refer to j loop unrolling done with SSE double precision, e.g. for the two different
 434      * jnr indices corresponding to data put in the four positions in the SIMD register.
 435      */
 436     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 437     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 438     int              jnrA,jnrB;
 439     int              j_coord_offsetA,j_coord_offsetB;
 440     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 441     real             rcutoff_scalar;
 442     real             *shiftvec,*fshift,*x,*f;
 443     __m128d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 444     int              vdwioffset0;
 445     __m128d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 446     int              vdwjidx0A,vdwjidx0B;
 447     __m128d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 448     __m128d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 449     __m128d          velec,felec,velecsum,facel,crf,krf,krf2;
 450     real             *charge;
 451     int              nvdwtype;
 452     __m128d          rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 453     int              *vdwtype;
 454     real             *vdwparam;
 455     __m128d          one_sixth   = _mm_set1_pd(1.0/6.0);
 456     __m128d          one_twelfth = _mm_set1_pd(1.0/12.0);
 457     __m128d           c6grid_00;
 458     __m128d           ewclj,ewclj2,ewclj6,ewcljrsq,poly,exponent,f6A,f6B,sh_lj_ewald;
 459     real             *vdwgridparam;
 460     __m128d           one_half = _mm_set1_pd(0.5);
 461     __m128d           minus_one = _mm_set1_pd(-1.0);
 462     __m128i          ewitab;
 463     __m128d          ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
 464     real             *ewtab;
 465     __m128d          dummy_mask,cutoff_mask;
 466     __m128d          signbit   = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
 467     __m128d          one     = _mm_set1_pd(1.0);
 468     __m128d          two     = _mm_set1_pd(2.0);
 469     x                = xx[0];
 470     f                = ff[0];
 471
 472     nri              = nlist->nri;
 473     iinr             = nlist->iinr;
 474     jindex           = nlist->jindex;
 475     jjnr             = nlist->jjnr;
 476     shiftidx         = nlist->shift;
 477     gid              = nlist->gid;
 478     shiftvec         = fr->shift_vec[0];
 479     fshift           = fr->fshift[0];
 480     facel            = _mm_set1_pd(fr->ic->epsfac);
 481     charge           = mdatoms->chargeA;
 482     nvdwtype         = fr->ntype;
 483     vdwparam         = fr->nbfp;
 484     vdwtype          = mdatoms->typeA;
 485     vdwgridparam     = fr->ljpme_c6grid;
 486     sh_lj_ewald      = _mm_set1_pd(fr->ic->sh_lj_ewald);
 487     ewclj            = _mm_set1_pd(fr->ic->ewaldcoeff_lj);
 488     ewclj2           = _mm_mul_pd(minus_one,_mm_mul_pd(ewclj,ewclj));
 489
 490     sh_ewald         = _mm_set1_pd(fr->ic->sh_ewald);
 491     ewtab            = fr->ic->tabq_coul_F;
 492     ewtabscale       = _mm_set1_pd(fr->ic->tabq_scale);
 493     ewtabhalfspace   = _mm_set1_pd(0.5/fr->ic->tabq_scale);
 494
 495     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 496     rcutoff_scalar   = fr->ic->rcoulomb;
 497     rcutoff          = _mm_set1_pd(rcutoff_scalar);
 498     rcutoff2         = _mm_mul_pd(rcutoff,rcutoff);
 499
 500     sh_vdw_invrcut6  = _mm_set1_pd(fr->ic->sh_invrc6);
 501     rvdw             = _mm_set1_pd(fr->ic->rvdw);
 502
 503     /* Avoid stupid compiler warnings */
 504     jnrA = jnrB = 0;
 505     j_coord_offsetA = 0;
 506     j_coord_offsetB = 0;
 507
 508     outeriter        = 0;
 509     inneriter        = 0;
 510
 511     /* Start outer loop over neighborlists */
 512     for(iidx=0; iidx<nri; iidx++)
 513     {
 514         /* Load shift vector for this list */
 515         i_shift_offset   = DIM*shiftidx[iidx];
 516
 517         /* Load limits for loop over neighbors */
 518         j_index_start    = jindex[iidx];
 519         j_index_end      = jindex[iidx+1];
 520
 521         /* Get outer coordinate index */
 522         inr              = iinr[iidx];
 523         i_coord_offset   = DIM*inr;
 524
 525         /* Load i particle coords and add shift vector */
 526         gmx_mm_load_shift_and_1rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 527
 528         fix0             = _mm_setzero_pd();
 529         fiy0             = _mm_setzero_pd();
 530         fiz0             = _mm_setzero_pd();
 531
 532         /* Load parameters for i particles */
 533         iq0              = _mm_mul_pd(facel,_mm_load1_pd(charge+inr+0));
 534         vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 535
 536         /* Start inner kernel loop */
 537         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 538         {
 539
 540             /* Get j neighbor index, and coordinate index */
 541             jnrA             = jjnr[jidx];
 542             jnrB             = jjnr[jidx+1];
 543             j_coord_offsetA  = DIM*jnrA;
 544             j_coord_offsetB  = DIM*jnrB;
 545
 546             /* load j atom coordinates */
 547             gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 548                                               &jx0,&jy0,&jz0);
 549
 550             /* Calculate displacement vector */
 551             dx00             = _mm_sub_pd(ix0,jx0);
 552             dy00             = _mm_sub_pd(iy0,jy0);
 553             dz00             = _mm_sub_pd(iz0,jz0);
 554
 555             /* Calculate squared distance and things based on it */
 556             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 557
 558             rinv00           = sse41_invsqrt_d(rsq00);
 559
 560             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 561
 562             /* Load parameters for j particles */
 563             jq0              = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
 564             vdwjidx0A        = 2*vdwtype[jnrA+0];
 565             vdwjidx0B        = 2*vdwtype[jnrB+0];
 566
 567             /**************************
 568              * CALCULATE INTERACTIONS *
 569              **************************/
 570
 571             if (gmx_mm_any_lt(rsq00,rcutoff2))
 572             {
 573
 574             r00              = _mm_mul_pd(rsq00,rinv00);
 575
 576             /* Compute parameters for interactions between i and j atoms */
 577             qq00             = _mm_mul_pd(iq0,jq0);
 578             gmx_mm_load_2pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,
 579                                          vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
 580             c6grid_00       = gmx_mm_load_2real_swizzle_pd(vdwgridparam+vdwioffset0+vdwjidx0A,
 581                                                                vdwgridparam+vdwioffset0+vdwjidx0B);
 582
 583             /* EWALD ELECTROSTATICS */
 584
 585             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 586             ewrt             = _mm_mul_pd(r00,ewtabscale);
 587             ewitab           = _mm_cvttpd_epi32(ewrt);
 588             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 589             gmx_mm_load_2pair_swizzle_pd(ewtab+gmx_mm_extract_epi32(ewitab,0),ewtab+gmx_mm_extract_epi32(ewitab,1),
 590                                          &ewtabF,&ewtabFn);
 591             felec            = _mm_add_pd(_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF),_mm_mul_pd(eweps,ewtabFn));
 592             felec            = _mm_mul_pd(_mm_mul_pd(qq00,rinv00),_mm_sub_pd(rinvsq00,felec));
 593
 594             /* Analytical LJ-PME */
 595             rinvsix          = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 596             ewcljrsq         = _mm_mul_pd(ewclj2,rsq00);
 597             ewclj6           = _mm_mul_pd(ewclj2,_mm_mul_pd(ewclj2,ewclj2));
 598             exponent         = sse41_exp_d(ewcljrsq);
 599             /* poly = exp(-(beta*r)^2) * (1 + (beta*r)^2 + (beta*r)^4 /2) */
 600             poly             = _mm_mul_pd(exponent,_mm_add_pd(_mm_sub_pd(one,ewcljrsq),_mm_mul_pd(_mm_mul_pd(ewcljrsq,ewcljrsq),one_half)));
 601             /* f6A = 6 * C6grid * (1 - poly) */
 602             f6A              = _mm_mul_pd(c6grid_00,_mm_sub_pd(one,poly));
 603             /* f6B = C6grid * exponent * beta^6 */
 604             f6B              = _mm_mul_pd(_mm_mul_pd(c6grid_00,one_sixth),_mm_mul_pd(exponent,ewclj6));
 605             /* fvdw = 12*C12/r13 - ((6*C6 - f6A)/r6 + f6B)/r */
 606             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);
 607
 608             cutoff_mask      = _mm_cmplt_pd(rsq00,rcutoff2);
 609
 610             fscal            = _mm_add_pd(felec,fvdw);
 611
 612             fscal            = _mm_and_pd(fscal,cutoff_mask);
 613
 614             /* Calculate temporary vectorial force */
 615             tx               = _mm_mul_pd(fscal,dx00);
 616             ty               = _mm_mul_pd(fscal,dy00);
 617             tz               = _mm_mul_pd(fscal,dz00);
 618
 619             /* Update vectorial force */
 620             fix0             = _mm_add_pd(fix0,tx);
 621             fiy0             = _mm_add_pd(fiy0,ty);
 622             fiz0             = _mm_add_pd(fiz0,tz);
 623
 624             gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,tx,ty,tz);
 625
 626             }
 627
 628             /* Inner loop uses 62 flops */
 629         }
 630
 631         if(jidx<j_index_end)
 632         {
 633
 634             jnrA             = jjnr[jidx];
 635             j_coord_offsetA  = DIM*jnrA;
 636
 637             /* load j atom coordinates */
 638             gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
 639                                               &jx0,&jy0,&jz0);
 640
 641             /* Calculate displacement vector */
 642             dx00             = _mm_sub_pd(ix0,jx0);
 643             dy00             = _mm_sub_pd(iy0,jy0);
 644             dz00             = _mm_sub_pd(iz0,jz0);
 645
 646             /* Calculate squared distance and things based on it */
 647             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 648
 649             rinv00           = sse41_invsqrt_d(rsq00);
 650
 651             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 652
 653             /* Load parameters for j particles */
 654             jq0              = _mm_load_sd(charge+jnrA+0);
 655             vdwjidx0A        = 2*vdwtype[jnrA+0];
 656
 657             /**************************
 658              * CALCULATE INTERACTIONS *
 659              **************************/
 660
 661             if (gmx_mm_any_lt(rsq00,rcutoff2))
 662             {
 663
 664             r00              = _mm_mul_pd(rsq00,rinv00);
 665
 666             /* Compute parameters for interactions between i and j atoms */
 667             qq00             = _mm_mul_pd(iq0,jq0);
 668             gmx_mm_load_1pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,&c6_00,&c12_00);
 669
 670             c6grid_00       = gmx_mm_load_1real_pd(vdwgridparam+vdwioffset0+vdwjidx0A);
 671
 672             /* EWALD ELECTROSTATICS */
 673
 674             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 675             ewrt             = _mm_mul_pd(r00,ewtabscale);
 676             ewitab           = _mm_cvttpd_epi32(ewrt);
 677             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 678             gmx_mm_load_1pair_swizzle_pd(ewtab+gmx_mm_extract_epi32(ewitab,0),&ewtabF,&ewtabFn);
 679             felec            = _mm_add_pd(_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF),_mm_mul_pd(eweps,ewtabFn));
 680             felec            = _mm_mul_pd(_mm_mul_pd(qq00,rinv00),_mm_sub_pd(rinvsq00,felec));
 681
 682             /* Analytical LJ-PME */
 683             rinvsix          = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 684             ewcljrsq         = _mm_mul_pd(ewclj2,rsq00);
 685             ewclj6           = _mm_mul_pd(ewclj2,_mm_mul_pd(ewclj2,ewclj2));
 686             exponent         = sse41_exp_d(ewcljrsq);
 687             /* poly = exp(-(beta*r)^2) * (1 + (beta*r)^2 + (beta*r)^4 /2) */
 688             poly             = _mm_mul_pd(exponent,_mm_add_pd(_mm_sub_pd(one,ewcljrsq),_mm_mul_pd(_mm_mul_pd(ewcljrsq,ewcljrsq),one_half)));
 689             /* f6A = 6 * C6grid * (1 - poly) */
 690             f6A              = _mm_mul_pd(c6grid_00,_mm_sub_pd(one,poly));
 691             /* f6B = C6grid * exponent * beta^6 */
 692             f6B              = _mm_mul_pd(_mm_mul_pd(c6grid_00,one_sixth),_mm_mul_pd(exponent,ewclj6));
 693             /* fvdw = 12*C12/r13 - ((6*C6 - f6A)/r6 + f6B)/r */
 694             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);
 695
 696             cutoff_mask      = _mm_cmplt_pd(rsq00,rcutoff2);
 697
 698             fscal            = _mm_add_pd(felec,fvdw);
 699
 700             fscal            = _mm_and_pd(fscal,cutoff_mask);
 701
 702             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 703
 704             /* Calculate temporary vectorial force */
 705             tx               = _mm_mul_pd(fscal,dx00);
 706             ty               = _mm_mul_pd(fscal,dy00);
 707             tz               = _mm_mul_pd(fscal,dz00);
 708
 709             /* Update vectorial force */
 710             fix0             = _mm_add_pd(fix0,tx);
 711             fiy0             = _mm_add_pd(fiy0,ty);
 712             fiz0             = _mm_add_pd(fiz0,tz);
 713
 714             gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,tx,ty,tz);
 715
 716             }
 717
 718             /* Inner loop uses 62 flops */
 719         }
 720
 721         /* End of innermost loop */
 722
 723         gmx_mm_update_iforce_1atom_swizzle_pd(fix0,fiy0,fiz0,
 724                                               f+i_coord_offset,fshift+i_shift_offset);
 725
 726         /* Increment number of inner iterations */
 727         inneriter                  += j_index_end - j_index_start;
 728
 729         /* Outer loop uses 7 flops */
 730     }
 731
 732     /* Increment number of outer iterations */
 733     outeriter        += nri;
 734
 735     /* Update outer/inner flops */
 736
 737     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_F,outeriter*7 + inneriter*62);
 738 }